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
// Copyright (c) 2016 The vulkano developers
// Licensed under the Apache License, Version 2.0
// <LICENSE-APACHE or
// https://www.apache.org/licenses/LICENSE-2.0> or the MIT
// license <LICENSE-MIT or https://opensource.org/licenses/MIT>,
// at your option. All files in the project carrying such
// notice may not be copied, modified, or distributed except
// according to those terms.
//! Suballocators are used to divide a *region* into smaller *suballocations*.
//!
//! See also [the parent module] for details about memory allocation in Vulkan.
//!
//! [the parent module]: super
use self::host::SlotId;
use super::{array_vec::ArrayVec, AllocationCreateInfo, AllocationCreationError};
use crate::{
device::{Device, DeviceOwned},
image::ImageTiling,
memory::DeviceMemory,
DeviceSize, OomError, VulkanError, VulkanObject,
};
use crossbeam_queue::ArrayQueue;
use parking_lot::Mutex;
use std::{
cell::Cell,
error::Error,
ffi::c_void,
fmt::{self, Display},
mem::{self, ManuallyDrop, MaybeUninit},
num::NonZeroU64,
ops::Range,
ptr::{self, NonNull},
slice,
sync::{
atomic::{AtomicU64, Ordering},
Arc,
},
};
/// Memory allocations are portions of memory that are reserved for a specific resource or purpose.
///
/// There's a few ways you can obtain a `MemoryAlloc` in Vulkano. Most commonly you will probably
/// want to use a [memory allocator]. If you already have a [`DeviceMemory`] block on hand that you
/// would like to turn into an allocation, you can use [the constructor]. Lastly, you can use a
/// [suballocator] if you want to create multiple smaller allocations out of a bigger one.
///
/// [memory allocator]: super::MemoryAllocator
/// [the constructor]: Self::new
/// [suballocator]: Suballocator
#[derive(Debug)]
pub struct MemoryAlloc {
offset: DeviceSize,
size: DeviceSize,
// Needed when binding resources to the allocation in order to avoid aliasing memory.
allocation_type: AllocationType,
// Mapped pointer to the start of the allocation or `None` is the memory is not host-visible.
mapped_ptr: Option<NonNull<c_void>>,
// Used by the suballocators to align allocations to the non-coherent atom size when the memory
// type is host-visible but not host-coherent. This will be `None` for any other memory type.
atom_size: Option<NonZeroU64>,
// Used in the `Drop` impl to free the allocation if required.
parent: AllocParent,
}
#[derive(Debug)]
enum AllocParent {
FreeList {
allocator: Arc<FreeListAllocator>,
id: SlotId,
},
Buddy {
allocator: Arc<BuddyAllocator>,
order: usize,
offset: DeviceSize,
},
Pool {
allocator: Arc<PoolAllocatorInner>,
index: DeviceSize,
},
Bump(Arc<BumpAllocator>),
Root(Arc<DeviceMemory>),
Dedicated(DeviceMemory),
}
// It is safe to share `mapped_ptr` between threads because the user would have to use unsafe code
// themself to get UB in the first place.
unsafe impl Send for MemoryAlloc {}
unsafe impl Sync for MemoryAlloc {}
impl MemoryAlloc {
/// Creates a new `MemoryAlloc`.
///
/// The memory is mapped automatically if it's host-visible.
#[inline]
pub fn new(device_memory: DeviceMemory) -> Result<Self, AllocationCreationError> {
let device = device_memory.device();
let physical_device = device.physical_device();
let memory_type_index = device_memory.memory_type_index();
let property_flags = &physical_device.memory_properties().memory_types
[memory_type_index as usize]
.property_flags;
let mapped_ptr = if property_flags.host_visible {
let fns = device.fns();
let mut output = MaybeUninit::uninit();
// This is always valid because we are mapping the whole range.
unsafe {
(fns.v1_0.map_memory)(
device.handle(),
device_memory.handle(),
0,
ash::vk::WHOLE_SIZE,
ash::vk::MemoryMapFlags::empty(),
output.as_mut_ptr(),
)
.result()
.map_err(VulkanError::from)?;
NonNull::new(output.assume_init())
}
} else {
None
};
let atom_size = (property_flags.host_visible && !property_flags.host_coherent)
.then_some(physical_device.properties().non_coherent_atom_size)
.and_then(NonZeroU64::new);
Ok(MemoryAlloc {
offset: 0,
size: device_memory.allocation_size(),
allocation_type: AllocationType::Unknown,
mapped_ptr,
atom_size,
parent: if device_memory.is_dedicated() {
AllocParent::Dedicated(device_memory)
} else {
AllocParent::Root(Arc::new(device_memory))
},
})
}
/// Returns the offset of the allocation within the [`DeviceMemory`] block.
#[inline]
pub fn offset(&self) -> DeviceSize {
self.offset
}
/// Returns the size of the allocation.
#[inline]
pub fn size(&self) -> DeviceSize {
self.size
}
/// Returns the type of resources that can be bound to this allocation.
#[inline]
pub fn allocation_type(&self) -> AllocationType {
self.allocation_type
}
/// Returns the mapped pointer to the start of the allocation if the memory is host-visible,
/// otherwise returns [`None`].
#[inline]
pub fn mapped_ptr(&self) -> Option<NonNull<c_void>> {
self.mapped_ptr
}
/// Returns a mapped slice to the data within the allocation if the memory is host-visible,
/// otherwise returns [`None`].
///
/// # Safety
///
/// - While the returned slice exists, there must be no operations pending or executing in a
/// GPU queue that write to the same memory.
#[inline]
pub unsafe fn mapped_slice(&self) -> Option<&[u8]> {
self.mapped_ptr
.map(|ptr| slice::from_raw_parts(ptr.as_ptr().cast(), self.size as usize))
}
/// Returns a mapped mutable slice to the data within the allocation if the memory is
/// host-visible, otherwise returns [`None`].
///
/// # Safety
///
/// - While the returned slice exists, there must be no operations pending or executing in a
/// GPU queue that access the same memory.
#[inline]
pub unsafe fn mapped_slice_mut(&mut self) -> Option<&mut [u8]> {
self.mapped_ptr
.map(|ptr| slice::from_raw_parts_mut(ptr.as_ptr().cast(), self.size as usize))
}
pub(crate) unsafe fn write(&self, range: Range<DeviceSize>) -> Option<&mut [u8]> {
debug_assert!(!range.is_empty() && range.end <= self.size);
self.mapped_ptr.map(|ptr| {
slice::from_raw_parts_mut(
ptr.as_ptr().add(range.start as usize).cast(),
(range.end - range.start) as usize,
)
})
}
/// Invalidates the host (CPU) cache for a range of the allocation.
///
/// You must call this method before the memory is read by the host, if the device previously
/// wrote to the memory. It has no effect if the memory is not mapped or if the memory is
/// [host-coherent].
///
/// `range` is specified in bytes relative to the start of the allocation. The start and end of
/// `range` must be a multiple of the [`non_coherent_atom_size`] device property, but
/// `range.end` can also equal to `self.size()`.
///
/// # Safety
///
/// - If there are memory writes by the GPU that have not been propagated into the CPU cache,
/// then there must not be any references in Rust code to the specified `range` of the memory.
///
/// # Panics
///
/// - Panics if `range` is empty.
/// - Panics if `range.end` exceeds `self.size`.
/// - Panics if `range.start` or `range.end` are not a multiple of the `non_coherent_atom_size`.
///
/// [host-coherent]: super::MemoryPropertyFlags::host_coherent
/// [`non_coherent_atom_size`]: crate::device::Properties::non_coherent_atom_size
#[inline]
pub unsafe fn invalidate_range(&self, range: Range<DeviceSize>) -> Result<(), OomError> {
// VUID-VkMappedMemoryRange-memory-00684
if let Some(atom_size) = self.atom_size {
let range = self.create_memory_range(range, atom_size.get());
let device = self.device();
let fns = device.fns();
(fns.v1_0.invalidate_mapped_memory_ranges)(device.handle(), 1, &range)
.result()
.map_err(VulkanError::from)?;
} else {
// FIXME:
// self.debug_validate_memory_range(&range);
}
Ok(())
}
/// Flushes the host (CPU) cache for a range of the allocation.
///
/// You must call this method after writing to the memory from the host, if the device is going
/// to read the memory. It has no effect if the memory is not mapped or if the memory is
/// [host-coherent].
///
/// `range` is specified in bytes relative to the start of the allocation. The start and end of
/// `range` must be a multiple of the [`non_coherent_atom_size`] device property, but
/// `range.end` can also equal to `self.size()`.
///
/// # Safety
///
/// - There must be no operations pending or executing in a GPU queue that access the specified
/// `range` of the memory.
///
/// # Panics
///
/// - Panics if `range` is empty.
/// - Panics if `range.end` exceeds `self.size`.
/// - Panics if `range.start` or `range.end` are not a multiple of the `non_coherent_atom_size`.
///
/// [host-coherent]: super::MemoryPropertyFlags::host_coherent
/// [`non_coherent_atom_size`]: crate::device::Properties::non_coherent_atom_size
#[inline]
pub unsafe fn flush_range(&self, range: Range<DeviceSize>) -> Result<(), OomError> {
// VUID-VkMappedMemoryRange-memory-00684
if let Some(atom_size) = self.atom_size {
let range = self.create_memory_range(range, atom_size.get());
let device = self.device();
let fns = device.fns();
(fns.v1_0.flush_mapped_memory_ranges)(device.handle(), 1, &range)
.result()
.map_err(VulkanError::from)?;
} else {
// FIXME:
// self.debug_validate_memory_range(&range);
}
Ok(())
}
fn create_memory_range(
&self,
range: Range<DeviceSize>,
atom_size: DeviceSize,
) -> ash::vk::MappedMemoryRange {
assert!(!range.is_empty() && range.end <= self.size);
// VUID-VkMappedMemoryRange-size-00685
// Guaranteed because we always map the entire `DeviceMemory`.
// VUID-VkMappedMemoryRange-offset-00687
// VUID-VkMappedMemoryRange-size-01390
assert!(
range.start % atom_size == 0 && (range.end % atom_size == 0 || range.end == self.size)
);
// VUID-VkMappedMemoryRange-offset-00687
// Guaranteed as long as `range.start` is aligned because the suballocators always align
// `self.offset` to the non-coherent atom size for non-coherent host-visible memory.
let offset = self.offset + range.start;
let mut size = range.end - range.start;
let device_memory = self.device_memory();
// VUID-VkMappedMemoryRange-size-01390
if offset + size < device_memory.allocation_size() {
// We align the size in case `range.end == self.size`. We can do this without aliasing
// other allocations because the suballocators ensure that all allocations are aligned
// to the atom size for non-coherent host-visible memory.
size = align_up(size, atom_size);
}
ash::vk::MappedMemoryRange {
memory: device_memory.handle(),
offset,
size,
..Default::default()
}
}
/// This exists because even if no cache control is required, the parameters should still be
/// valid, otherwise you might have bugs in your code forever just because your memory happens
/// to be host-coherent.
#[allow(dead_code)]
fn debug_validate_memory_range(&self, range: &Range<DeviceSize>) {
debug_assert!(!range.is_empty() && range.end <= self.size);
debug_assert!({
let atom_size = self
.device()
.physical_device()
.properties()
.non_coherent_atom_size;
range.start % atom_size == 0 && (range.end % atom_size == 0 || range.end == self.size)
});
}
pub(crate) fn atom_size(&self) -> Option<NonZeroU64> {
self.atom_size
}
/// Returns the underlying block of [`DeviceMemory`].
#[inline]
pub fn device_memory(&self) -> &DeviceMemory {
match &self.parent {
AllocParent::FreeList { allocator, .. } => &allocator.device_memory,
AllocParent::Buddy { allocator, .. } => &allocator.device_memory,
AllocParent::Pool { allocator, .. } => &allocator.device_memory,
AllocParent::Bump(allocator) => &allocator.device_memory,
AllocParent::Root(device_memory) => device_memory,
AllocParent::Dedicated(device_memory) => device_memory,
}
}
/// Returns the parent allocation if this allocation is a [suballocation], otherwise returns
/// [`None`].
///
/// [suballocation]: Suballocator
#[inline]
pub fn parent_allocation(&self) -> Option<&Self> {
match &self.parent {
AllocParent::FreeList { allocator, .. } => Some(&allocator.region),
AllocParent::Buddy { allocator, .. } => Some(&allocator.region),
AllocParent::Pool { allocator, .. } => Some(&allocator.region),
AllocParent::Bump(allocator) => Some(&allocator.region),
AllocParent::Root(_) => None,
AllocParent::Dedicated(_) => None,
}
}
/// Returns `true` if this allocation is the root of the [memory hierarchy].
///
/// [memory hierarchy]: Suballocator#memory-hierarchies
#[inline]
pub fn is_root(&self) -> bool {
matches!(&self.parent, AllocParent::Root(_))
}
/// Returns `true` if this allocation is a [dedicated allocation].
///
/// [dedicated allocation]: crate::memory::MemoryAllocateInfo#structfield.dedicated_allocation
#[inline]
pub fn is_dedicated(&self) -> bool {
matches!(&self.parent, AllocParent::Dedicated(_))
}
/// Returns the underlying block of [`DeviceMemory`] if this allocation [is the root
/// allocation] and is not [aliased], otherwise returns the allocation back wrapped in [`Err`].
///
/// [is the root allocation]: Self::is_root
/// [aliased]: Self::alias
#[inline]
pub fn try_unwrap(self) -> Result<DeviceMemory, Self> {
let this = ManuallyDrop::new(self);
// SAFETY: This is safe because even if a panic happens, `self.parent` can not be
// double-freed since `self` was wrapped in `ManuallyDrop`. If we fail to unwrap the
// `DeviceMemory`, the copy of `self.parent` is forgotten and only then is the
// `ManuallyDrop` wrapper removed from `self`.
match unsafe { ptr::read(&this.parent) } {
AllocParent::Root(device_memory) => {
Arc::try_unwrap(device_memory).map_err(|device_memory| {
mem::forget(device_memory);
ManuallyDrop::into_inner(this)
})
}
parent => {
mem::forget(parent);
Err(ManuallyDrop::into_inner(this))
}
}
}
/// Duplicates the allocation, creating aliased memory. Returns [`None`] if the allocation [is
/// a dedicated allocation].
///
/// You might consider using this method if you want to optimize memory usage by aliasing
/// render targets for example, in which case you will have to double and triple check that the
/// memory is not used concurrently unless it only involves reading. You are highly discouraged
/// from doing this unless you have a reason to.
///
/// # Safety
///
/// - You must ensure memory accesses are synchronized yourself.
///
/// [memory hierarchy]: Suballocator#memory-hierarchies
/// [is a dedicated allocation]: Self::is_dedicated
#[inline]
pub unsafe fn alias(&self) -> Option<Self> {
self.root().map(|device_memory| MemoryAlloc {
parent: AllocParent::Root(device_memory.clone()),
..*self
})
}
fn root(&self) -> Option<&Arc<DeviceMemory>> {
match &self.parent {
AllocParent::FreeList { allocator, .. } => Some(&allocator.device_memory),
AllocParent::Buddy { allocator, .. } => Some(&allocator.device_memory),
AllocParent::Pool { allocator, .. } => Some(&allocator.device_memory),
AllocParent::Bump(allocator) => Some(&allocator.device_memory),
AllocParent::Root(device_memory) => Some(device_memory),
AllocParent::Dedicated(_) => None,
}
}
/// Increases the offset of the allocation by the specified `amount` and shrinks its size by
/// the same amount.
///
/// # Panics
///
/// - Panics if the `amount` exceeds the size of the allocation.
#[inline]
pub fn shift(&mut self, amount: DeviceSize) {
assert!(amount <= self.size);
self.offset += amount;
self.size -= amount;
}
/// Shrinks the size of the allocation to the specified `new_size`.
///
/// # Panics
///
/// - Panics if the `new_size` exceeds the current size of the allocation.
#[inline]
pub fn shrink(&mut self, new_size: DeviceSize) {
assert!(new_size <= self.size);
self.size = new_size;
}
/// Sets the offset of the allocation without checking for memory aliasing.
///
/// See also [`shift`], which moves the offset safely.
///
/// # Safety
///
/// - You must ensure that the allocation doesn't alias any other allocations within the
/// [`DeviceMemory`] block, and if it does, then you must ensure memory accesses are
/// synchronized yourself.
/// - You must ensure the allocation still fits inside the `DeviceMemory` block.
///
/// [`shift`]: Self::shift
#[inline]
pub unsafe fn set_offset(&mut self, new_offset: DeviceSize) {
self.offset = new_offset;
}
/// Sets the size of the allocation without checking for memory aliasing.
///
/// See also [`shrink`], which sets the size safely.
///
/// # Safety
///
/// - You must ensure that the allocation doesn't alias any other allocations within the
/// [`DeviceMemory`] block, and if it does, then you must ensure memory accesses are
/// synchronized yourself.
/// - You must ensure the allocation still fits inside the `DeviceMemory` block.
///
/// [`shrink`]: Self::shrink
#[inline]
pub unsafe fn set_size(&mut self, new_size: DeviceSize) {
self.size = new_size;
}
/// Sets the allocation type.
///
/// This might cause memory aliasing due to [buffer-image granularity] conflicts if the
/// allocation type is [`Linear`] or [`NonLinear`] and is changed to a different one.
///
/// # Safety
///
/// - You must ensure that the allocation doesn't alias any other allocations within the
/// [`DeviceMemory`] block, and if it does, then you must ensure memory accesses are
/// synchronized yourself.
///
/// [buffer-image granularity]: super#buffer-image-granularity
/// [`Linear`]: AllocationType::Linear
/// [`NonLinear`]: AllocationType::NonLinear
#[inline]
pub unsafe fn set_allocation_type(&mut self, new_type: AllocationType) {
self.allocation_type = new_type;
}
}
impl Drop for MemoryAlloc {
#[inline]
fn drop(&mut self) {
match &self.parent {
AllocParent::FreeList { allocator, id } => {
allocator.free(*id);
}
AllocParent::Buddy {
allocator,
order,
offset,
} => {
allocator.free(*order, *offset);
}
AllocParent::Pool { allocator, index } => {
allocator.free(*index);
}
// The bump allocator can't free individually, but we need to keep a reference to it so
// it don't get reset or dropped while in use.
AllocParent::Bump(_) => {}
// A root allocation frees itself once all references to the `DeviceMemory` are dropped.
AllocParent::Root(_) => {}
// Dedicated allocations free themselves when the `DeviceMemory` is dropped.
AllocParent::Dedicated(_) => {}
}
}
}
unsafe impl DeviceOwned for MemoryAlloc {
#[inline]
fn device(&self) -> &Arc<Device> {
self.device_memory().device()
}
}
/// Suballocators are used to divide a *region* into smaller *suballocations*.
///
/// # Regions
///
/// As the name implies, a region is a contiguous portion of memory. It may be the whole dedicated
/// block of [`DeviceMemory`], or only a part of it. Regions are just [allocations] like any other,
/// but we use this term to refer specifically to an allocation that is to be suballocated. Every
/// suballocator is created with a region to work with.
///
/// # Free-lists
///
/// A free-list, also kind of predictably, refers to a list of (sub)allocations within a region
/// that are currently free. Every (sub)allocator that can free allocations dynamically (in any
/// order) needs to keep a free-list of some sort. This list is then consulted when new allocations
/// are made, and can be used to coalesce neighboring allocations that are free into bigger ones.
///
/// # Memory hierarchies
///
/// Different applications have wildly different allocation needs, and there's no way to cover them
/// all with a single type of allocator. Furthermore, different allocators have different
/// trade-offs and are best suited to specific tasks. To account for all possible use-cases,
/// Vulkano offers the ability to create *memory hierarchies*. We refer to the [`DeviceMemory`] as
/// the root of any such hierarchy, even though technically the driver has levels that are further
/// up, because those `DeviceMemory` blocks need to be allocated from physical memory [pages]
/// themselves, but since those levels are not accessible to us we don't need to consider them. You
/// can create any number of levels/branches from there, bounded only by the amount of available
/// memory within a `DeviceMemory` block. You can suballocate the root into regions, which are then
/// suballocated into further regions and so on, creating hierarchies of arbitrary height.
///
/// As an added bonus, memory hierarchies lend themselves perfectly to the concept of composability
/// we all love so much, making them a natural fit for Rust. For one, a region can be allocated any
/// way, and fed into any suballocator. Also, once you are done with a branch of a hierarchy,
/// meaning there are no more suballocations in use within the region of that branch, and you would
/// like to reuse the region, you can do so safely! All suballocators have a `try_into_region`
/// method for this purpose. This means that you can replace one suballocator with another without
/// consulting any of the higher levels in the hierarchy.
///
/// # Examples
///
/// Allocating a region to suballocatate:
///
/// ```
/// use vulkano::memory::{DeviceMemory, MemoryAllocateInfo, MemoryType};
/// use vulkano::memory::allocator::MemoryAlloc;
/// # let device: std::sync::Arc<vulkano::device::Device> = return;
///
/// // First you need to find a suitable memory type.
/// let memory_type_index = device
/// .physical_device()
/// .memory_properties()
/// .memory_types
/// .iter()
/// .enumerate()
/// // In a real-world scenario, you would probably want to rank the memory types based on your
/// // requirements, instead of picking the first one that satisfies them. Also, you have to
/// // take the requirements of the resources you want to allocate memory for into consideration.
/// .find_map(|(index, MemoryType { property_flags, .. })| {
/// property_flags.device_local.then_some(index)
/// })
/// .unwrap() as u32;
///
/// let region = MemoryAlloc::new(
/// DeviceMemory::allocate(
/// device.clone(),
/// MemoryAllocateInfo {
/// allocation_size: 64 * 1024 * 1024,
/// memory_type_index,
/// ..Default::default()
/// },
/// )
/// .unwrap(),
/// )
/// .unwrap();
///
/// // You can now feed `region` into any suballocator.
/// ```
///
/// # Implementing the trait
///
/// Please don't.
///
/// [allocations]: MemoryAlloc
/// [pages]: super#pages
pub unsafe trait Suballocator: DeviceOwned {
/// Whether this allocator needs to block or not.
///
/// This is used by the [`GenericMemoryAllocator`] to specialize the allocation strategy to the
/// suballocator at compile time.
///
/// [`GenericMemoryAllocator`]: super::GenericMemoryAllocator
const IS_BLOCKING: bool;
/// Whether the allocator needs [`cleanup`] to be called before memory can be released.
///
/// This is used by the [`GenericMemoryAllocator`] to specialize the allocation strategy to the
/// suballocator at compile time.
///
/// [`cleanup`]: Self::cleanup
/// [`GenericMemoryAllocator`]: super::GenericMemoryAllocator
const NEEDS_CLEANUP: bool;
/// Creates a new suballocator for the given [region].
///
/// [region]: Self#regions
fn new(region: MemoryAlloc) -> Self
where
Self: Sized;
/// Creates a new suballocation within the [region].
///
/// [region]: Self#regions
fn allocate(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError>;
/// Creates a new suballocation within the [region] without checking the parameters.
///
/// # Safety
///
/// - `create_info.size` must not be zero.
/// - `create_info.alignment` must not be zero.
/// - `create_info.alignment` must be a power of two.
///
/// [region]: Self#regions
#[cfg_attr(not(feature = "document_unchecked"), doc(hidden))]
unsafe fn allocate_unchecked(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError>;
/// Returns a reference to the underlying [region].
///
/// [region]: Self#regions
fn region(&self) -> &MemoryAlloc;
/// Returns the underlying [region] if there are no other strong references to the allocator,
/// otherwise hands you back the allocator wrapped in [`Err`]. Allocations made with the
/// allocator count as references for as long as they are alive.
///
/// [region]: Self#regions
fn try_into_region(self) -> Result<MemoryAlloc, Self>
where
Self: Sized;
/// Returns the total amount of free space that is left in the [region].
///
/// [region]: Self#regions
fn free_size(&self) -> DeviceSize;
/// Tries to free some space, if applicable.
fn cleanup(&mut self);
}
/// Parameters to create a new [allocation] using a [suballocator].
///
/// [allocation]: MemoryAlloc
/// [suballocator]: Suballocator
#[derive(Clone, Debug)]
pub struct SuballocationCreateInfo {
/// Size of the allocation in bytes.
///
/// The default value is `0`, which must be overridden.
pub size: DeviceSize,
/// [Alignment] of the allocation in bytes. Must be a power of 2.
///
/// The default value is `0`, which must be overridden.
///
/// [Alignment]: super#alignment
pub alignment: DeviceSize,
/// Type of resources that can be bound to the allocation.
///
/// The default value is [`AllocationType::Unknown`].
pub allocation_type: AllocationType,
pub _ne: crate::NonExhaustive,
}
impl Default for SuballocationCreateInfo {
#[inline]
fn default() -> Self {
SuballocationCreateInfo {
size: 0,
alignment: 0,
allocation_type: AllocationType::Unknown,
_ne: crate::NonExhaustive(()),
}
}
}
impl From<AllocationCreateInfo<'_>> for SuballocationCreateInfo {
#[inline]
fn from(create_info: AllocationCreateInfo<'_>) -> Self {
SuballocationCreateInfo {
size: create_info.requirements.size,
alignment: create_info.requirements.alignment,
allocation_type: create_info.allocation_type,
_ne: crate::NonExhaustive(()),
}
}
}
impl SuballocationCreateInfo {
pub(super) fn validate(&self) {
assert!(self.size > 0);
assert!(self.alignment > 0);
assert!(self.alignment.is_power_of_two());
}
}
/// Tells the [suballocator] what type of resource will be bound to the allocation, so that it can
/// optimize memory usage while still respecting the [buffer-image granularity].
///
/// [suballocator]: Suballocator
/// [buffer-image granularity]: super#buffer-image-granularity
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
pub enum AllocationType {
/// The type of resource is unknown, it might be either linear or non-linear. What this means is
/// that allocations created with this type must always be aligned to the buffer-image
/// granularity.
Unknown = 0,
/// The resource is linear, e.g. buffers, linear images. A linear allocation following another
/// linear allocation never needs to be aligned to the buffer-image granularity.
Linear = 1,
/// The resource is non-linear, e.g. optimal images. A non-linear allocation following another
/// non-linear allocation never needs to be aligned to the buffer-image granularity.
NonLinear = 2,
}
impl From<ImageTiling> for AllocationType {
#[inline]
fn from(tiling: ImageTiling) -> Self {
match tiling {
ImageTiling::Optimal => AllocationType::NonLinear,
ImageTiling::Linear => AllocationType::Linear,
}
}
}
/// Error that can be returned when using a [suballocator].
///
/// [suballocator]: Suballocator
#[derive(Clone, Debug, PartialEq, Eq)]
pub enum SuballocationCreationError {
/// There is no more space available in the region.
OutOfRegionMemory,
/// The region has enough free space to satisfy the request but is too fragmented.
FragmentedRegion,
/// The allocation was larger than the allocator's block size, meaning that this error would
/// arise with the parameters no matter the state the allocator was in.
///
/// This can be used to let the [`GenericMemoryAllocator`] know that allocating a new block of
/// [`DeviceMemory`] and trying to suballocate it with the same parameters would not solve the
/// issue.
///
/// [`GenericMemoryAllocator`]: super::GenericMemoryAllocator
BlockSizeExceeded,
}
impl Error for SuballocationCreationError {}
impl Display for SuballocationCreationError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
write!(
f,
"{}",
match self {
Self::OutOfRegionMemory => "out of region memory",
Self::FragmentedRegion => "the region is too fragmented",
Self::BlockSizeExceeded =>
"the allocation size was greater than the suballocator's block size",
}
)
}
}
/// A [suballocator] that uses the most generic [free-list].
///
/// The strength of this allocator is that it can create and free allocations completely
/// dynamically, which means they can be any size and created/freed in any order. The downside is
/// that this always leads to horrific [external fragmentation] the more such dynamic allocations
/// are made. Therefore, this allocator is best suited for long-lived allocations. If you need
/// to create allocations of various sizes, but can't afford this fragmentation, then the
/// [`BuddyAllocator`] is your best buddy. If you need to create allocations which share a similar
/// size, consider the [`PoolAllocator`]. Lastly, if you need to allocate very often, then
/// [`BumpAllocator`] is best suited.
///
/// See also [the `Suballocator` implementation].
///
/// # Algorithm
///
/// The free-list stores suballocations which can have any offset and size. When an allocation
/// request is made, the list is searched using the best-fit strategy, meaning that the smallest
/// suballocation that fits the request is chosen. If required, the chosen suballocation is trimmed
/// at the ends and the ends are returned to the free-list. As such, no [internal fragmentation]
/// occurs. The front might need to be trimmed because of [alignment requirements] and the end
/// because of a larger than required size. When an allocation is freed, the allocator checks if
/// the adjacent suballocations are free, and if so it coalesces them into a bigger one before
/// putting it in the free-list.
///
/// # Efficiency
///
/// The allocator is synchronized internally with a lock, which is held only for a very short
/// period each time an allocation is created and freed. The free-list is sorted by size, which
/// means that when allocating, finding a best-fit is always possible in *O*(log(*n*)) time in the
/// worst case. When freeing, the coalescing requires us to remove the adjacent free suballocations
/// from the free-list which is *O*(log(*n*)), and insert the possibly coalesced suballocation into
/// the free-list which has the same time complexity, so in total freeing is *O*(log(*n*)).
///
/// There is one notable edge-case: after the allocator finds a best-fit, it is possible that it
/// needs to align the suballocation's offset to a higher value, after which the requested size
/// might no longer fit. In such a case, the next free suballocation in sorted order is tried until
/// a fit is successful. If this issue is encountered with all candidates, then the time complexity
/// would be *O*(*n*). However, this scenario is extremely unlikely which is why we are not
/// considering it in the above analysis. Additionally, if your free-list is filled with
/// allocations that all have the same size then that seems pretty sus. Sounds like you're in dire
/// need of a `PoolAllocator`.
///
/// # Examples
///
/// Most commonly you will not want to use this suballocator directly but rather use it within
/// [`GenericMemoryAllocator`], having one global [`StandardMemoryAllocator`] for most if not all
/// of your allocation needs.
///
/// Basic usage as a global allocator for long-lived resources:
///
/// ```
/// use vulkano::format::Format;
/// use vulkano::image::{ImageDimensions, ImmutableImage};
/// use vulkano::memory::allocator::StandardMemoryAllocator;
/// # let device: std::sync::Arc<vulkano::device::Device> = return;
///
/// let memory_allocator = StandardMemoryAllocator::new_default(device.clone());
///
/// # fn read_textures() -> Vec<Vec<u8>> { Vec::new() }
/// # let mut command_buffer_builder: vulkano::command_buffer::AutoCommandBufferBuilder<vulkano::command_buffer::PrimaryAutoCommandBuffer<vulkano::command_buffer::allocator::StandardCommandBufferAlloc>> = return;
/// // Allocate some resources.
/// let textures_data: Vec<Vec<u8>> = read_textures();
/// let textures = textures_data.into_iter().map(|data| {
/// ImmutableImage::from_iter(
/// &memory_allocator,
/// data,
/// ImageDimensions::Dim2d {
/// width: 1024,
/// height: 1024,
/// array_layers: 1,
/// },
/// 1.into(),
/// Format::R8G8B8A8_UNORM,
/// &mut command_buffer_builder,
/// )
/// .unwrap()
/// });
/// ```
///
/// For use in allocating buffers for [`CpuBufferPool`]:
///
/// ```
/// use std::sync::Arc;
/// use vulkano::buffer::CpuBufferPool;
/// use vulkano::memory::allocator::StandardMemoryAllocator;
/// # let device: std::sync::Arc<vulkano::device::Device> = return;
///
/// // We need to wrap the allocator in an `Arc` so that we can share ownership of it.
/// let memory_allocator = Arc::new(StandardMemoryAllocator::new_default(device.clone()));
/// let buffer_pool = CpuBufferPool::<u32>::upload(memory_allocator.clone());
///
/// // You can continue using `memory_allocator` for other things.
/// ```
///
/// Sometimes, it is neccessary to suballocate an allocation. If you don't want to allocate new
/// [`DeviceMemory`] blocks to suballocate, perhaps because of concerns of memory wastage or
/// allocation efficiency, you can use your existing global `StandardMemoryAllocator` to allocate
/// regions for your suballocation needs:
///
/// ```
/// use vulkano::memory::allocator::{MemoryAllocator, StandardMemoryAllocator, SuballocationCreateInfo};
///
/// # let device: std::sync::Arc<vulkano::device::Device> = return;
/// let memory_allocator = StandardMemoryAllocator::new_default(device.clone());
///
/// # let memory_type_index = 0;
/// let region = memory_allocator.allocate_from_type(
/// // When choosing the index, you have to make sure that the memory type is allowed for the
/// // type of resource that you want to bind the suballocations to.
/// memory_type_index,
/// SuballocationCreateInfo {
/// // This will be the size of your region.
/// size: 16 * 1024 * 1024,
/// // It generally does not matter what the alignment is, because you're going to
/// // suballocate the allocation anyway, and not bind it directly.
/// alignment: 1,
/// ..Default::default()
/// },
/// )
/// .unwrap();
///
/// // You can now feed the `region` into any suballocator.
/// ```
///
/// [suballocator]: Suballocator
/// [free-list]: Suballocator#free-lists
/// [external fragmentation]: super#external-fragmentation
/// [the `Suballocator` implementation]: Suballocator#impl-Suballocator-for-Arc<FreeListAllocator>
/// [internal fragmentation]: super#internal-fragmentation
/// [alignment requirements]: super#alignment
/// [`GenericMemoryAllocator`]: super::GenericMemoryAllocator
/// [`StandardMemoryAllocator`]: super::StandardMemoryAllocator
/// [`CpuBufferPool`]: crate::buffer::CpuBufferPool
#[derive(Debug)]
pub struct FreeListAllocator {
region: MemoryAlloc,
device_memory: Arc<DeviceMemory>,
buffer_image_granularity: DeviceSize,
atom_size: DeviceSize,
// Total memory remaining in the region.
free_size: AtomicU64,
state: Mutex<FreeListAllocatorState>,
}
impl FreeListAllocator {
/// Creates a new `FreeListAllocator` for the given [region].
///
/// # Panics
///
/// - Panics if `region.allocation_type` is not [`AllocationType::Unknown`]. This is done to
/// avoid checking for a special case of [buffer-image granularity] conflict.
/// - Panics if `region` is a [dedicated allocation].
///
/// [region]: Suballocator#regions
/// [buffer-image granularity]: super#buffer-image-granularity
/// [dedicated allocation]: MemoryAlloc::is_dedicated
#[inline]
pub fn new(region: MemoryAlloc) -> Arc<Self> {
// NOTE(Marc): This number was pulled straight out of my a-
const AVERAGE_ALLOCATION_SIZE: DeviceSize = 64 * 1024;
assert!(region.allocation_type == AllocationType::Unknown);
let device_memory = region
.root()
.expect("dedicated allocations can't be suballocated")
.clone();
let buffer_image_granularity = device_memory
.device()
.physical_device()
.properties()
.buffer_image_granularity;
let atom_size = region.atom_size.map(NonZeroU64::get).unwrap_or(1);
let free_size = AtomicU64::new(region.size);
let capacity = (region.size / AVERAGE_ALLOCATION_SIZE) as usize;
let mut nodes = host::PoolAllocator::new(capacity + 64);
let mut free_list = Vec::with_capacity(capacity / 16 + 16);
let root_id = nodes.allocate(SuballocationListNode {
prev: None,
next: None,
offset: region.offset,
size: region.size,
ty: SuballocationType::Free,
});
free_list.push(root_id);
let state = Mutex::new(FreeListAllocatorState { nodes, free_list });
Arc::new(FreeListAllocator {
region,
device_memory,
buffer_image_granularity,
atom_size,
free_size,
state,
})
}
fn free(&self, id: SlotId) {
let mut state = self.state.lock();
self.free_size
.fetch_add(state.nodes.get(id).size, Ordering::Release);
state.nodes.get_mut(id).ty = SuballocationType::Free;
state.coalesce(id);
state.free(id);
}
}
unsafe impl Suballocator for Arc<FreeListAllocator> {
const IS_BLOCKING: bool = true;
const NEEDS_CLEANUP: bool = false;
#[inline]
fn new(region: MemoryAlloc) -> Self {
FreeListAllocator::new(region)
}
/// Creates a new suballocation within the [region].
///
/// # Panics
///
/// - Panics if `create_info.size` is zero.
/// - Panics if `create_info.alignment` is zero.
/// - Panics if `create_info.alignment` is not a power of two.
///
/// # Errors
///
/// - Returns [`OutOfRegionMemory`] if there are no free suballocations large enough so satisfy
/// the request.
/// - Returns [`FragmentedRegion`] if a suballocation large enough to satisfy the request could
/// have been formed, but wasn't because of [external fragmentation].
///
/// [region]: Suballocator#regions
/// [`allocate`]: Suballocator::allocate
/// [`OutOfRegionMemory`]: SuballocationCreationError::OutOfRegionMemory
/// [`FragmentedRegion`]: SuballocationCreationError::FragmentedRegion
/// [external fragmentation]: super#external-fragmentation
#[inline]
fn allocate(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
create_info.validate();
unsafe { self.allocate_unchecked(create_info) }
}
#[inline]
unsafe fn allocate_unchecked(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
fn has_granularity_conflict(prev_ty: SuballocationType, ty: AllocationType) -> bool {
if prev_ty == SuballocationType::Free {
false
} else if prev_ty == SuballocationType::Unknown {
true
} else {
prev_ty != ty.into()
}
}
let SuballocationCreateInfo {
size,
alignment,
allocation_type,
_ne: _,
} = create_info;
let alignment = DeviceSize::max(alignment, self.atom_size);
let mut state = self.state.lock();
match state.free_list.last() {
Some(&last) if state.nodes.get(last).size >= size => {
let index = match state
.free_list
.binary_search_by_key(&size, |&x| state.nodes.get(x).size)
{
// Exact fit.
Ok(index) => index,
// Next-best fit. Note that `index == free_list.len()` can not be because we
// checked that the free-list contains a suballocation that is big enough.
Err(index) => index,
};
for &id in &state.free_list[index..] {
let suballoc = state.nodes.get(id);
let mut offset = align_up(suballoc.offset, alignment);
if let Some(prev_id) = suballoc.prev {
let prev = state.nodes.get(prev_id);
if are_blocks_on_same_page(
prev.offset,
prev.size,
offset,
self.buffer_image_granularity,
) && has_granularity_conflict(prev.ty, allocation_type)
{
offset = align_up(offset, self.buffer_image_granularity);
}
}
if offset + size <= suballoc.offset + suballoc.size {
state.allocate(id);
state.split(id, offset, size);
state.nodes.get_mut(id).ty = allocation_type.into();
self.free_size.fetch_sub(size, Ordering::Release);
return Ok(MemoryAlloc {
offset,
size,
allocation_type,
mapped_ptr: self.region.mapped_ptr.and_then(|ptr| {
NonNull::new(
ptr.as_ptr().add((offset - self.region.offset) as usize),
)
}),
atom_size: self.region.atom_size,
parent: AllocParent::FreeList {
allocator: self.clone(),
id,
},
});
}
}
// There is not enough space due to alignment requirements.
Err(SuballocationCreationError::OutOfRegionMemory)
}
// There would be enough space if the region wasn't so fragmented. :(
Some(_) if self.free_size() >= size => {
Err(SuballocationCreationError::FragmentedRegion)
}
// There is not enough space.
Some(_) => Err(SuballocationCreationError::OutOfRegionMemory),
// There is no space at all.
None => Err(SuballocationCreationError::OutOfRegionMemory),
}
}
#[inline]
fn region(&self) -> &MemoryAlloc {
&self.region
}
#[inline]
fn try_into_region(self) -> Result<MemoryAlloc, Self> {
Arc::try_unwrap(self).map(|allocator| allocator.region)
}
#[inline]
fn free_size(&self) -> DeviceSize {
self.free_size.load(Ordering::Acquire)
}
#[inline]
fn cleanup(&mut self) {}
}
unsafe impl DeviceOwned for FreeListAllocator {
#[inline]
fn device(&self) -> &Arc<Device> {
self.device_memory.device()
}
}
#[derive(Debug)]
struct FreeListAllocatorState {
nodes: host::PoolAllocator<SuballocationListNode>,
// Free suballocations sorted by size in ascending order. This means we can always find a
// best-fit in *O*(log(*n*)) time in the worst case, and iterating in order is very efficient.
free_list: Vec<SlotId>,
}
#[derive(Clone, Copy, Debug)]
struct SuballocationListNode {
prev: Option<SlotId>,
next: Option<SlotId>,
offset: DeviceSize,
size: DeviceSize,
ty: SuballocationType,
}
/// Tells us if a suballocation is free, and if not, whether it is linear or not. This is needed in
/// order to be able to respect the buffer-image granularity.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum SuballocationType {
Unknown,
Linear,
NonLinear,
Free,
}
impl From<AllocationType> for SuballocationType {
fn from(ty: AllocationType) -> Self {
match ty {
AllocationType::Unknown => SuballocationType::Unknown,
AllocationType::Linear => SuballocationType::Linear,
AllocationType::NonLinear => SuballocationType::NonLinear,
}
}
}
impl FreeListAllocatorState {
/// Removes the target suballocation from the free-list. The free-list must contain it.
fn allocate(&mut self, node_id: SlotId) {
debug_assert!(self.free_list.contains(&node_id));
let node = self.nodes.get(node_id);
match self
.free_list
.binary_search_by_key(&node.size, |&x| self.nodes.get(x).size)
{
Ok(index) => {
// If there are multiple free suballocations with the same size, the search might
// have returned any one, so we need to find the one corresponding to the target ID.
if self.free_list[index] == node_id {
self.free_list.remove(index);
return;
}
// Check all previous indices that point to suballocations with the same size.
{
let mut index = index;
loop {
index = index.wrapping_sub(1);
if let Some(&id) = self.free_list.get(index) {
if id == node_id {
self.free_list.remove(index);
return;
}
if self.nodes.get(id).size != node.size {
break;
}
} else {
break;
}
}
}
// Check all next indices that point to suballocations with the same size.
{
let mut index = index;
loop {
index += 1;
if let Some(&id) = self.free_list.get(index) {
if id == node_id {
self.free_list.remove(index);
return;
}
if self.nodes.get(id).size != node.size {
break;
}
} else {
break;
}
}
}
unreachable!();
}
Err(_) => unreachable!(),
}
}
/// Fits a suballocation inside the target one, splitting the target at the ends if required.
fn split(&mut self, node_id: SlotId, offset: DeviceSize, size: DeviceSize) {
let node = self.nodes.get(node_id);
debug_assert!(node.ty == SuballocationType::Free);
debug_assert!(offset >= node.offset);
debug_assert!(offset + size <= node.offset + node.size);
let padding_front = offset - node.offset;
let padding_back = node.offset + node.size - offset - size;
if padding_front > 0 {
let padding = SuballocationListNode {
prev: node.prev,
next: Some(node_id),
offset: node.offset,
size: padding_front,
ty: SuballocationType::Free,
};
let padding_id = self.nodes.allocate(padding);
if let Some(prev_id) = padding.prev {
self.nodes.get_mut(prev_id).next = Some(padding_id);
}
let node = self.nodes.get_mut(node_id);
node.prev = Some(padding_id);
node.offset = offset;
node.size -= padding.size;
self.free(padding_id);
}
if padding_back > 0 {
let padding = SuballocationListNode {
prev: Some(node_id),
next: node.next,
offset: offset + size,
size: padding_back,
ty: SuballocationType::Free,
};
let padding_id = self.nodes.allocate(padding);
if let Some(next_id) = padding.next {
self.nodes.get_mut(next_id).prev = Some(padding_id);
}
let node = self.nodes.get_mut(node_id);
node.next = Some(padding_id);
node.size -= padding.size;
self.free(padding_id);
}
}
/// Inserts the target suballocation into the free-list. The free-list must not contain it
/// already.
fn free(&mut self, node_id: SlotId) {
debug_assert!(!self.free_list.contains(&node_id));
let node = self.nodes.get(node_id);
let (Ok(index) | Err(index)) = self
.free_list
.binary_search_by_key(&node.size, |&x| self.nodes.get(x).size);
self.free_list.insert(index, node_id);
}
/// Coalesces the target (free) suballocation with adjacent ones that are also free.
fn coalesce(&mut self, node_id: SlotId) {
let node = self.nodes.get(node_id);
debug_assert!(node.ty == SuballocationType::Free);
if let Some(prev_id) = node.prev {
let prev = self.nodes.get(prev_id);
if prev.ty == SuballocationType::Free {
self.allocate(prev_id);
self.nodes.free(prev_id);
let node = self.nodes.get_mut(node_id);
node.prev = prev.prev;
node.offset = prev.offset;
node.size += prev.size; // nom nom nom
if let Some(prev_id) = node.prev {
self.nodes.get_mut(prev_id).next = Some(node_id);
}
}
}
if let Some(next_id) = node.next {
let next = self.nodes.get(next_id);
if next.ty == SuballocationType::Free {
self.allocate(next_id);
self.nodes.free(next_id);
let node = self.nodes.get_mut(node_id);
node.next = next.next;
node.size += next.size;
if let Some(next_id) = node.next {
self.nodes.get_mut(next_id).prev = Some(node_id);
}
}
}
}
}
/// A [suballocator] whose structure forms a binary tree of power-of-two-sized suballocations.
///
/// That is, all allocation sizes are rounded up to the next power of two. This helps reduce
/// [external fragmentation] by a lot, at the expense of possibly severe [internal fragmentation]
/// if you're not careful. For example, if you needed an allocation size of 64MiB, you would be
/// wasting no memory. But with an allocation size of 70MiB, you would use a whole 128MiB instead,
/// wasting 45% of the memory. Use this algorithm if you need to create and free a lot of
/// allocations, which would cause too much external fragmentation when using
/// [`FreeListAllocator`]. However, if the sizes of your allocations are more or less the same,
/// then the [`PoolAllocator`] would be a better choice and would eliminate external fragmentation
/// completely.
///
/// See also [the `Suballocator` implementation].
///
/// # Algorithm
///
/// Say you have a [region] of size 256MiB, and you want to allocate 14MiB. Assuming there are no
/// existing allocations, the `BuddyAllocator` would split the 256MiB root *node* into two 128MiB
/// nodes. These two nodes are called *buddies*. The allocator would then proceed to split the left
/// node recursively until it wouldn't be able to fit the allocation anymore. In this example, that
/// would happen after 4 splits and end up with a node size of 16MiB. Since the allocation
/// requested was 14MiB, 2MiB would become internal fragmentation and be unusable for the lifetime
/// of the allocation. When an allocation is freed, this process is done backwards, checking if the
/// buddy of each node on the way up is free and if so they are coalesced.
///
/// Each possible node size has an *order*, with the smallest node size being of order 0 and the
/// largest of the highest order. With this notion, node sizes are proportional to 2<sup>*n*</sup>
/// where *n* is the order. The highest order is determined from the size of the region and a
/// constant minimum node size, which we chose to be 16B: log(*region size* / 16) or
/// equiavalently log(*region size*) - 4 (assuming
/// *region size* ≥ 16).
///
/// It's safe to say that this algorithm works best if you have some level of control over your
/// allocation sizes, so that you don't end up allocating twice as much memory. An example of this
/// would be when you need to allocate regions for other allocators, such as the `PoolAllocator` or
/// the [`BumpAllocator`].
///
/// # Efficiency
///
/// The allocator is synchronized internally with a lock, which is held only for a very short
/// period each time an allocation is created and freed. The time complexity of both allocation and
/// freeing is *O*(*m*) in the worst case where *m* is the highest order, which equates to *O*(log
/// (*n*)) where *n* is the size of the region.
///
/// # Examples
///
/// Basic usage together with [`GenericMemoryAllocator`], to allocate resources that have a
/// moderately low life span (for example if you have a lot of images, each of which needs to be
/// resized every now and then):
///
/// ```
/// use std::sync::Arc;
/// use vulkano::memory::allocator::{
/// BuddyAllocator, GenericMemoryAllocator, GenericMemoryAllocatorCreateInfo,
/// };
///
/// # let device: std::sync::Arc<vulkano::device::Device> = return;
/// let memory_allocator = GenericMemoryAllocator::<Arc<BuddyAllocator>>::new(
/// device.clone(),
/// GenericMemoryAllocatorCreateInfo {
/// // Your block sizes must be powers of two, because `BuddyAllocator` only accepts
/// // power-of-two-sized regions.
/// block_sizes: &[(0, 64 * 1024 * 1024)],
/// ..Default::default()
/// },
/// )
/// .unwrap();
///
/// // Now you can use `memory_allocator` to allocate whatever it is you need.
/// ```
///
/// [suballocator]: Suballocator
/// [internal fragmentation]: super#internal-fragmentation
/// [external fragmentation]: super#external-fragmentation
/// [the `Suballocator` implementation]: Suballocator#impl-Suballocator-for-Arc<BuddyAllocator>
/// [region]: Suballocator#regions
/// [`GenericMemoryAllocator`]: super::GenericMemoryAllocator
#[derive(Debug)]
pub struct BuddyAllocator {
region: MemoryAlloc,
device_memory: Arc<DeviceMemory>,
buffer_image_granularity: DeviceSize,
atom_size: DeviceSize,
// Total memory remaining in the region.
free_size: AtomicU64,
state: Mutex<BuddyAllocatorState>,
}
impl BuddyAllocator {
const MIN_NODE_SIZE: DeviceSize = 16;
/// Arbitrary maximum number of orders, used to avoid a 2D `Vec`. Together with a minimum node
/// size of 16, this is enough for a 64GiB region.
const MAX_ORDERS: usize = 32;
/// Creates a new `BuddyAllocator` for the given [region].
///
/// # Panics
///
/// - Panics if `region.allocation_type` is not [`AllocationType::Unknown`]. This is done to
/// avoid checking for a special case of [buffer-image granularity] conflict.
/// - Panics if `region.size` is not a power of two.
/// - Panics if `region.size` is not in the range \[16B, 64GiB\].
/// - Panics if `region` is a [dedicated allocation].
///
/// [region]: Suballocator#regions
/// [buffer-image granularity]: super#buffer-image-granularity
/// [dedicated allocation]: MemoryAlloc::is_dedicated
#[inline]
pub fn new(region: MemoryAlloc) -> Arc<Self> {
const EMPTY_FREE_LIST: Vec<DeviceSize> = Vec::new();
let max_order = (region.size / Self::MIN_NODE_SIZE).trailing_zeros() as usize;
assert!(region.allocation_type == AllocationType::Unknown);
assert!(region.size.is_power_of_two());
assert!(region.size >= Self::MIN_NODE_SIZE && max_order < Self::MAX_ORDERS);
let device_memory = region
.root()
.expect("dedicated allocations can't be suballocated")
.clone();
let buffer_image_granularity = device_memory
.device()
.physical_device()
.properties()
.buffer_image_granularity;
let atom_size = region.atom_size.map(NonZeroU64::get).unwrap_or(1);
let free_size = AtomicU64::new(region.size);
let mut free_list = ArrayVec::new(max_order + 1, [EMPTY_FREE_LIST; Self::MAX_ORDERS]);
// The root node has the lowest offset and highest order, so it's the whole region.
free_list[max_order].push(region.offset);
let state = Mutex::new(BuddyAllocatorState { free_list });
Arc::new(BuddyAllocator {
region,
device_memory,
buffer_image_granularity,
atom_size,
free_size,
state,
})
}
fn free(&self, min_order: usize, mut offset: DeviceSize) {
let mut state = self.state.lock();
// Try to coalesce nodes while incrementing the order.
for (order, free_list) in state.free_list.iter_mut().enumerate().skip(min_order) {
let size = Self::MIN_NODE_SIZE << order;
let buddy_offset = ((offset - self.region.offset) ^ size) + self.region.offset;
match free_list.binary_search(&buddy_offset) {
// If the buddy is in the free-list, we can coalesce.
Ok(index) => {
free_list.remove(index);
offset = DeviceSize::min(offset, buddy_offset);
}
// Otherwise free the node.
Err(_) => {
let (Ok(index) | Err(index)) = free_list.binary_search(&offset);
free_list.insert(index, offset);
self.free_size
.fetch_add(Self::MIN_NODE_SIZE << min_order, Ordering::Release);
break;
}
}
}
}
}
unsafe impl Suballocator for Arc<BuddyAllocator> {
const IS_BLOCKING: bool = true;
const NEEDS_CLEANUP: bool = false;
#[inline]
fn new(region: MemoryAlloc) -> Self {
BuddyAllocator::new(region)
}
/// Creates a new suballocation within the [region].
///
/// # Panics
///
/// - Panics if `create_info.size` is zero.
/// - Panics if `create_info.alignment` is zero.
/// - Panics if `create_info.alignment` is not a power of two.
///
/// # Errors
///
/// - Returns [`OutOfRegionMemory`] if there are no free nodes large enough so satisfy the
/// request.
/// - Returns [`FragmentedRegion`] if a node large enough to satisfy the request could have
/// been formed, but wasn't because of [external fragmentation].
///
/// [region]: Suballocator#regions
/// [`allocate`]: Suballocator::allocate
/// [`OutOfRegionMemory`]: SuballocationCreationError::OutOfRegionMemory
/// [`FragmentedRegion`]: SuballocationCreationError::FragmentedRegion
/// [external fragmentation]: super#external-fragmentation
#[inline]
fn allocate(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
create_info.validate();
unsafe { self.allocate_unchecked(create_info) }
}
#[inline]
unsafe fn allocate_unchecked(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
/// Returns the largest power of two smaller or equal to the input.
fn prev_power_of_two(val: DeviceSize) -> DeviceSize {
const MAX_POWER_OF_TWO: DeviceSize = 1 << (DeviceSize::BITS - 1);
MAX_POWER_OF_TWO
.checked_shr(val.leading_zeros())
.unwrap_or(0)
}
let SuballocationCreateInfo {
mut size,
mut alignment,
allocation_type,
_ne: _,
} = create_info;
if allocation_type == AllocationType::Unknown
|| allocation_type == AllocationType::NonLinear
{
size = align_up(size, self.buffer_image_granularity);
alignment = DeviceSize::max(alignment, self.buffer_image_granularity);
}
let size = DeviceSize::max(size, BuddyAllocator::MIN_NODE_SIZE).next_power_of_two();
let alignment = DeviceSize::max(alignment, self.atom_size);
let min_order = (size / BuddyAllocator::MIN_NODE_SIZE).trailing_zeros() as usize;
let mut state = self.state.lock();
// Start searching at the lowest possible order going up.
for (order, free_list) in state.free_list.iter_mut().enumerate().skip(min_order) {
for (index, &offset) in free_list.iter().enumerate() {
if offset % alignment == 0 {
free_list.remove(index);
// Go in the opposite direction, splitting nodes from higher orders. The lowest
// order doesn't need any splitting.
for (order, free_list) in state
.free_list
.iter_mut()
.enumerate()
.skip(min_order)
.take(order - min_order)
.rev()
{
let size = BuddyAllocator::MIN_NODE_SIZE << order;
let right_child = offset + size;
// Insert the right child in sorted order.
let (Ok(index) | Err(index)) = free_list.binary_search(&right_child);
free_list.insert(index, right_child);
// Repeat splitting for the left child if required in the next loop turn.
}
self.free_size.fetch_sub(size, Ordering::Release);
return Ok(MemoryAlloc {
offset,
size: create_info.size,
allocation_type,
mapped_ptr: self.region.mapped_ptr.and_then(|ptr| {
NonNull::new(ptr.as_ptr().add((offset - self.region.offset) as usize))
}),
atom_size: self.region.atom_size,
parent: AllocParent::Buddy {
allocator: self.clone(),
order: min_order,
offset, // The offset in the alloc itself can change.
},
});
}
}
}
if prev_power_of_two(self.free_size()) >= create_info.size {
// A node large enough could be formed if the region wasn't so fragmented.
Err(SuballocationCreationError::FragmentedRegion)
} else {
Err(SuballocationCreationError::OutOfRegionMemory)
}
}
#[inline]
fn region(&self) -> &MemoryAlloc {
&self.region
}
#[inline]
fn try_into_region(self) -> Result<MemoryAlloc, Self> {
Arc::try_unwrap(self).map(|allocator| allocator.region)
}
/// Returns the total amount of free space left in the [region] that is available to the
/// allocator, which means that [internal fragmentation] is excluded.
///
/// [region]: Suballocator#regions
/// [internal fragmentation]: super#internal-fragmentation
#[inline]
fn free_size(&self) -> DeviceSize {
self.free_size.load(Ordering::Acquire)
}
#[inline]
fn cleanup(&mut self) {}
}
unsafe impl DeviceOwned for BuddyAllocator {
#[inline]
fn device(&self) -> &Arc<Device> {
self.device_memory.device()
}
}
#[derive(Debug)]
struct BuddyAllocatorState {
// Every order has its own free-list for convenience, so that we don't have to traverse a tree.
// Each free-list is sorted by offset because we want to find the first-fit as this strategy
// minimizes external fragmentation.
free_list: ArrayVec<Vec<DeviceSize>, { BuddyAllocator::MAX_ORDERS }>,
}
/// A [suballocator] using a pool of fixed-size blocks as a [free-list].
///
/// Since the size of the blocks is fixed, you can not create allocations bigger than that. You can
/// create smaller ones, though, which leads to more and more [internal fragmentation] the smaller
/// the allocations get. This is generally a good trade-off, as internal fragmentation is nowhere
/// near as hard to deal with as [external fragmentation].
///
/// See also [the `Suballocator` implementation].
///
/// # Algorithm
///
/// The free-list contains indices of blocks in the region that are available, so allocation
/// consists merely of popping an index from the free-list. The same goes for freeing, all that is
/// required is to push the index of the block into the free-list. Note that this is only possible
/// because the blocks have a fixed size. Due to this one fact, the free-list doesn't need to be
/// sorted or traversed. As long as there is a free block, it will do, no matter which block it is.
///
/// Since the `PoolAllocator` doesn't keep a list of suballocations that are currently in use,
/// resolving [buffer-image granularity] conflicts on a case-by-case basis is not possible.
/// Therefore, it is an all or nothing situation:
///
/// - you use the allocator for only one type of allocation, [`Linear`] or [`NonLinear`], or
/// - you allow both but align the blocks to the granularity so that no conflics can happen.
///
/// The way this is done is that every suballocation inherits the allocation type of the region.
/// The latter is done by using a region whose allocation type is [`Unknown`]. You are discouraged
/// from using this type if you can avoid it.
///
/// The block size can end up bigger than specified if the allocator is created with a region whose
/// allocation type is `Unknown`. In that case all blocks are aligned to the buffer-image
/// granularity, which may or may not cause signifficant memory usage increase. Say for example
/// your driver reports a granularity of 4KiB. If you need a block size of 8KiB, you would waste no
/// memory. On the other hand, if you needed a block size of 6KiB, you would be wasting 25% of the
/// memory. In such a scenario you are highly encouraged to use a different allocation type.
///
/// The reverse is also true: with an allocation type other than `Unknown`, not all memory within a
/// block may be usable depending on the requested [suballocation]. For instance, with a block size
/// of 1152B (9 * 128B) and a suballocation with `alignment: 256`, a block at an odd index could
/// not utilize its first 128B, reducing its effective size to 1024B. This is usually only relevant
/// with small block sizes, as [alignment requirements] are usually rather small, but it completely
/// depends on the resource and driver.
///
/// In summary, the block size you choose has a signifficant impact on internal fragmentation due
/// to the two reasons described above. You need to choose your block size carefully, *especially*
/// if you require small allocations. Some rough guidelines:
///
/// - Always [align] your blocks to a sufficiently large power of 2. This does **not** mean your
/// block size must be a power of two. For example with a block size of 3KiB, your blocks would
/// be aligned to 1KiB.
/// - Prefer not using the allocation type `Unknown`. You can always create as many
/// `PoolAllocator`s as you like for different allocation types and sizes, and they can all work
/// within the same memory block. You should be safe from fragmentation if your blocks are
/// aligned to 1KiB.
/// - If you must use the allocation type `Unknown`, then you should be safe from fragmentation on
/// pretty much any driver if your blocks are aligned to 64KiB. Keep in mind that this might
/// change any time as new devices appear or new drivers come out. Always look at the properties
/// of the devices you want to support before relying on any such data.
///
/// # Efficiency
///
/// In theory, a pool allocator is the ideal one because it causes no external fragmentation, and
/// both allocation and freeing is *O*(1). It also never needs to lock and hence also lends itself
/// perfectly to concurrency. But of course, there is the trade-off that block sizes are not
/// dynamic.
///
/// As you can imagine, the `PoolAllocator` is the perfect fit if you know the sizes of the
/// allocations you will be making, and they are more or less in the same size class. But this
/// allocation algorithm really shines when combined with others, as most do. For one, nothing is
/// stopping you from having multiple `PoolAllocator`s for many different size classes. You could
/// consider a pool of pools, by layering `PoolAllocator` with itself, but this would have the
/// downside that the regions of the pools for all size classes would have to match. Usually this
/// is not desired. If you want pools for different size classes to all have about the same number
/// of blocks, or you even know that some size classes require more or less blocks (because of how
/// many resources you will be allocating for each), then you need an allocator that can allocate
/// regions of different sizes. You can use the [`FreeListAllocator`] for this, if external
/// fragmentation is not an issue, otherwise you might consider using the [`BuddyAllocator`]. On
/// the other hand, you might also want to consider having a `PoolAllocator` at the top of a
/// [hierarchy]. Again, this allocator never needs to lock making it *the* perfect fit for a global
/// concurrent allocator, which hands out large regions which can then be suballocated locally on a
/// thread, by the [`BumpAllocator`] for example.
///
/// # Examples
///
/// Basic usage together with [`GenericMemoryAllocator`]:
///
/// ```
/// use std::sync::Arc;
/// use vulkano::memory::allocator::{
/// GenericMemoryAllocator, GenericMemoryAllocatorCreateInfo, PoolAllocator,
/// };
///
/// # let device: std::sync::Arc<vulkano::device::Device> = return;
/// let memory_allocator = GenericMemoryAllocator::<Arc<PoolAllocator<{ 64 * 1024 }>>>::new(
/// device.clone(),
/// GenericMemoryAllocatorCreateInfo {
/// block_sizes: &[(0, 64 * 1024 * 1024)],
/// ..Default::default()
/// },
/// )
/// .unwrap();
///
/// // Now you can use `memory_allocator` to allocate whatever it is you need.
/// ```
///
/// [suballocator]: Suballocator
/// [free-list]: Suballocator#free-lists
/// [internal fragmentation]: super#internal-fragmentation
/// [external fragmentation]: super#external-fragmentation
/// [the `Suballocator` implementation]: Suballocator#impl-Suballocator-for-Arc<PoolAllocator<BLOCK_SIZE>>
/// [region]: Suballocator#regions
/// [buffer-image granularity]: super#buffer-image-granularity
/// [`Linear`]: AllocationType::Linear
/// [`NonLinear`]: AllocationType::NonLinear
/// [`Unknown`]: AllocationType::Unknown
/// [suballocation]: SuballocationCreateInfo
/// [alignment requirements]: super#memory-requirements
/// [align]: super#alignment
/// [hierarchy]: Suballocator#memory-hierarchies
/// [`GenericMemoryAllocator`]: super::GenericMemoryAllocator
#[derive(Debug)]
#[repr(transparent)]
pub struct PoolAllocator<const BLOCK_SIZE: DeviceSize> {
inner: PoolAllocatorInner,
}
impl<const BLOCK_SIZE: DeviceSize> PoolAllocator<BLOCK_SIZE> {
/// Creates a new `PoolAllocator` for the given [region].
///
/// # Panics
///
/// - Panics if `region.size < BLOCK_SIZE`.
/// - Panics if `region` is a [dedicated allocation].
///
/// [region]: Suballocator#regions
/// [dedicated allocation]: MemoryAlloc::is_dedicated
#[inline]
pub fn new(
region: MemoryAlloc,
#[cfg(test)] buffer_image_granularity: DeviceSize,
) -> Arc<Self> {
Arc::new(PoolAllocator {
inner: PoolAllocatorInner::new(
region,
BLOCK_SIZE,
#[cfg(test)]
buffer_image_granularity,
),
})
}
/// Size of a block. Can be bigger than `BLOCK_SIZE` due to alignment requirements.
#[inline]
pub fn block_size(&self) -> DeviceSize {
self.inner.block_size
}
/// Total number of blocks available to the allocator. This is always equal to
/// `self.region().size() / self.block_size()`.
#[inline]
pub fn block_count(&self) -> usize {
self.inner.free_list.capacity()
}
/// Number of free blocks.
#[inline]
pub fn free_count(&self) -> usize {
self.inner.free_list.len()
}
}
unsafe impl<const BLOCK_SIZE: DeviceSize> Suballocator for Arc<PoolAllocator<BLOCK_SIZE>> {
const IS_BLOCKING: bool = false;
const NEEDS_CLEANUP: bool = false;
#[inline]
fn new(region: MemoryAlloc) -> Self {
PoolAllocator::new(
region,
#[cfg(test)]
1,
)
}
/// Creates a new suballocation within the [region].
///
/// > **Note**: `create_info.allocation_type` is silently ignored because all suballocations
/// > inherit the allocation type from the region.
///
/// # Panics
///
/// - Panics if `create_info.size` is zero.
/// - Panics if `create_info.alignment` is zero.
/// - Panics if `create_info.alignment` is not a power of two.
///
/// # Errors
///
/// - Returns [`OutOfRegionMemory`] if the [free-list] is empty.
/// - Returns [`OutOfRegionMemory`] if the allocation can't fit inside a block. Only the first
/// block in the free-list is tried, which means that if one block isn't usable due to
/// [internal fragmentation] but a different one would be, you still get this error. See the
/// [type-level documentation] for details on how to properly configure your allocator.
/// - Returns [`BlockSizeExceeded`] if `create_info.size` exceeds `BLOCK_SIZE`.
///
/// [region]: Suballocator#regions
/// [`allocate`]: Suballocator::allocate
/// [`OutOfRegionMemory`]: SuballocationCreationError::OutOfRegionMemory
/// [free-list]: Suballocator#free-lists
/// [internal fragmentation]: super#internal-fragmentation
/// [type-level documentation]: PoolAllocator
/// [`BlockSizeExceeded`]: SuballocationCreationError::BlockSizeExceeded
#[inline]
fn allocate(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
create_info.validate();
unsafe { self.allocate_unchecked(create_info) }
}
#[inline]
unsafe fn allocate_unchecked(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
// SAFETY: `PoolAllocator<BLOCK_SIZE>` and `PoolAllocatorInner` have the same layout.
//
// This is not quite optimal, because we are always cloning the `Arc` even if allocation
// fails, in which case the `Arc` gets cloned and dropped for no reason. Unfortunately,
// there is currently no way to turn `&Arc<T>` into `&Arc<U>` that is sound.
Arc::from_raw(Arc::into_raw(self.clone()).cast::<PoolAllocatorInner>())
.allocate_unchecked(create_info)
}
#[inline]
fn region(&self) -> &MemoryAlloc {
&self.inner.region
}
#[inline]
fn try_into_region(self) -> Result<MemoryAlloc, Self> {
Arc::try_unwrap(self).map(|allocator| allocator.inner.region)
}
#[inline]
fn free_size(&self) -> DeviceSize {
self.free_count() as DeviceSize * self.block_size()
}
#[inline]
fn cleanup(&mut self) {}
}
unsafe impl<const BLOCK_SIZE: DeviceSize> DeviceOwned for PoolAllocator<BLOCK_SIZE> {
#[inline]
fn device(&self) -> &Arc<Device> {
self.inner.device_memory.device()
}
}
#[derive(Debug)]
struct PoolAllocatorInner {
region: MemoryAlloc,
device_memory: Arc<DeviceMemory>,
atom_size: DeviceSize,
block_size: DeviceSize,
// Unsorted list of free block indices.
free_list: ArrayQueue<DeviceSize>,
}
impl PoolAllocatorInner {
fn new(
region: MemoryAlloc,
mut block_size: DeviceSize,
#[cfg(test)] buffer_image_granularity: DeviceSize,
) -> Self {
let device_memory = region
.root()
.expect("dedicated allocations can't be suballocated")
.clone();
#[cfg(not(test))]
let buffer_image_granularity = device_memory
.device()
.physical_device()
.properties()
.buffer_image_granularity;
let atom_size = region.atom_size.map(NonZeroU64::get).unwrap_or(1);
if region.allocation_type == AllocationType::Unknown {
block_size = align_up(block_size, buffer_image_granularity);
}
let block_count = region.size / block_size;
let free_list = ArrayQueue::new(block_count as usize);
for i in 0..block_count {
free_list.push(i).unwrap();
}
PoolAllocatorInner {
region,
device_memory,
atom_size,
block_size,
free_list,
}
}
unsafe fn allocate_unchecked(
self: Arc<Self>,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
let SuballocationCreateInfo {
size,
alignment,
allocation_type: _,
_ne: _,
} = create_info;
let alignment = DeviceSize::max(alignment, self.atom_size);
let index = self
.free_list
.pop()
.ok_or(SuballocationCreationError::OutOfRegionMemory)?;
let unaligned_offset = self.region.offset + index * self.block_size;
let offset = align_up(unaligned_offset, alignment);
if offset + size > unaligned_offset + self.block_size {
self.free_list.push(index).unwrap();
return if size > self.block_size {
Err(SuballocationCreationError::BlockSizeExceeded)
} else {
// There is not enough space due to alignment requirements.
Err(SuballocationCreationError::OutOfRegionMemory)
};
}
Ok(MemoryAlloc {
offset,
size,
allocation_type: self.region.allocation_type,
mapped_ptr: self.region.mapped_ptr.and_then(|ptr| {
NonNull::new(ptr.as_ptr().add((offset - self.region.offset) as usize))
}),
atom_size: self.region.atom_size,
parent: AllocParent::Pool {
allocator: self,
index,
},
})
}
fn free(&self, index: DeviceSize) {
self.free_list.push(index).unwrap();
}
}
/// A [suballocator] which can allocate dynamically, but can only free all allocations at once.
///
/// With bump allocation, the used up space increases linearly as allocations are made and
/// allocations can never be freed individually, which is why this algorithm is also called *linear
/// allocation*. It is also known as *arena allocation*.
///
/// `BumpAllocator`s are best suited for very short-lived (say a few frames at best) resources that
/// need to be allocated often (say each frame), to really take advantage of the performance gains.
/// For creating long-lived allocations, [`FreeListAllocator`] is best suited. The way you would
/// typically use this allocator is to have one for each frame in flight. At the start of a frame,
/// you reset it and allocate your resources with it. You write to the resources, render with them,
/// and drop them at the end of the frame.
///
/// See also [the `Suballocator` implementation].
///
/// # Algorithm
///
/// What happens is that every time you make an allocation, you receive one with an offset
/// corresponding to the *free start* within the [region], and then the free start is *bumped*, so
/// that following allocations wouldn't alias it. As you can imagine, this is **extremely fast**,
/// because it doesn't need to keep a [free-list]. It only needs to do a few additions and
/// comparisons. But beware, **fast is about all this is**. It is horribly memory inefficient when
/// used wrong, and is very susceptible to [memory leaks].
///
/// Once you know that you are done with the allocations, meaning you know they have all been
/// dropped, you can safely reset the allocator using the [`try_reset`] method as long as the
/// allocator is not shared between threads. It is hard to safely reset a bump allocator that is
/// used concurrently. In such a scenario it's best not to reset it at all and instead drop it once
/// it reaches the end of the [region], freeing the region to a higher level in the [hierarchy]
/// once all threads have dropped their reference to the allocator. This is one of the reasons you
/// are generally advised to use one `BumpAllocator` per thread if you can.
///
/// # Efficiency
///
/// Allocation is *O*(1), and so is resetting the allocator (freeing all allocations). Allocation
/// is always lock-free, and most of the time even wait-free. The only case in which it is not
/// wait-free is if a lot of allocations are made concurrently, which results in CPU-level
/// contention. Therefore, if you for example need to allocate a lot of buffers each frame from
/// multiple threads, you might get better performance by using one `BumpAllocator` per thread.
///
/// The reason synchronization can be avoided entirely is that the created allocations can be
/// dropped without needing to talk back to the allocator to free anything. The other allocation
/// algorithms all have a free-list which needs to be modified once an allocation is dropped. Since
/// Vulkano's buffers and images are `Sync`, that means that even if the allocator only allocates
/// from one thread, it can still be used to free from multiple threads.
///
/// [suballocator]: Suballocator
/// [the `Suballocator` implementation]: Suballocator#impl-Suballocator-for-Arc<BumpAllocator>
/// [region]: Suballocator#regions
/// [free-list]: Suballocator#free-lists
/// [memory leaks]: super#leakage
/// [`try_reset`]: Self::try_reset
/// [hierarchy]: Suballocator#memory-hierarchies
#[derive(Debug)]
pub struct BumpAllocator {
region: MemoryAlloc,
device_memory: Arc<DeviceMemory>,
buffer_image_granularity: DeviceSize,
atom_size: DeviceSize,
// Encodes the previous allocation type in the 2 least signifficant bits and the free start in
// the rest.
state: AtomicU64,
}
impl BumpAllocator {
/// Creates a new `BumpAllocator` for the given [region].
///
/// # Panics
///
/// - Panics if `region` is a [dedicated allocation].
///
/// [region]: Suballocator#regions
/// [dedicated allocation]: MemoryAlloc::is_dedicated
#[inline]
pub fn new(region: MemoryAlloc) -> Arc<Self> {
let device_memory = region
.root()
.expect("dedicated allocations can't be suballocated")
.clone();
let buffer_image_granularity = device_memory
.device()
.physical_device()
.properties()
.buffer_image_granularity;
let atom_size = region.atom_size.map(NonZeroU64::get).unwrap_or(1);
let state = AtomicU64::new(region.allocation_type as u64);
Arc::new(BumpAllocator {
region,
device_memory,
buffer_image_granularity,
atom_size,
state,
})
}
/// Resets the free start back to the beginning of the [region] if there are no other strong
/// references to the allocator.
///
/// [region]: Suballocator#regions
#[inline]
pub fn try_reset(self: &mut Arc<Self>) -> Result<(), BumpAllocatorResetError> {
Arc::get_mut(self)
.map(|allocator| {
*allocator.state.get_mut() = allocator.region.allocation_type as u64;
})
.ok_or(BumpAllocatorResetError)
}
/// Resets the free-start to the beginning of the [region] without checking if there are other
/// strong references to the allocator.
///
/// This could be useful if you cloned the [`Arc`] yourself, and can guarantee that no
/// allocations currently hold a reference to it.
///
/// As a safe alternative, you can let the `Arc` do all the work. Simply drop it once it
/// reaches the end of the region. After all threads do that, the region will be freed to the
/// next level up the [hierarchy]. If you only use the allocator on one thread and need shared
/// ownership, you can use `Rc<RefCell<Arc<BumpAllocator>>>` together with [`try_reset`] for a
/// safe alternative as well.
///
/// # Safety
///
/// - All allocations made with the allocator must have been dropped.
///
/// [region]: Suballocator#regions
/// [hierarchy]: Suballocator#memory-hierarchies
/// [`try_reset`]: Self::try_reset
#[inline]
pub unsafe fn reset_unchecked(&self) {
self.state
.store(self.region.allocation_type as u64, Ordering::Release);
}
}
unsafe impl Suballocator for Arc<BumpAllocator> {
const IS_BLOCKING: bool = false;
const NEEDS_CLEANUP: bool = true;
#[inline]
fn new(region: MemoryAlloc) -> Self {
BumpAllocator::new(region)
}
/// Creates a new suballocation within the [region].
///
/// # Panics
///
/// - Panics if `create_info.size` is zero.
/// - Panics if `create_info.alignment` is zero.
/// - Panics if `create_info.alignment` is not a power of two.
///
/// # Errors
///
/// - Returns [`OutOfRegionMemory`] if the requested allocation can't fit in the free space
/// remaining in the region.
///
/// [region]: Suballocator#regions
/// [`allocate`]: Suballocator::allocate
/// [`OutOfRegionMemory`]: SuballocationCreationError::OutOfRegionMemory
#[inline]
fn allocate(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
create_info.validate();
unsafe { self.allocate_unchecked(create_info) }
}
#[inline]
unsafe fn allocate_unchecked(
&self,
create_info: SuballocationCreateInfo,
) -> Result<MemoryAlloc, SuballocationCreationError> {
const SPIN_LIMIT: u32 = 6;
// NOTE(Marc): The following code is a minimal version `Backoff` taken from
// crossbeam_utils v0.8.11, because we didn't want to add a dependency for a couple lines
// that are used in one place only.
/// Original documentation:
/// https://docs.rs/crossbeam-utils/0.8.11/crossbeam_utils/struct.Backoff.html
struct Backoff {
step: Cell<u32>,
}
impl Backoff {
fn new() -> Self {
Backoff { step: Cell::new(0) }
}
fn spin(&self) {
for _ in 0..1 << self.step.get().min(SPIN_LIMIT) {
core::hint::spin_loop();
}
if self.step.get() <= SPIN_LIMIT {
self.step.set(self.step.get() + 1);
}
}
}
fn has_granularity_conflict(prev_ty: AllocationType, ty: AllocationType) -> bool {
prev_ty == AllocationType::Unknown || prev_ty != ty
}
let SuballocationCreateInfo {
size,
alignment,
allocation_type,
_ne: _,
} = create_info;
let alignment = DeviceSize::max(alignment, self.atom_size);
let backoff = Backoff::new();
let mut state = self.state.load(Ordering::Relaxed);
loop {
let free_start = state >> 2;
let prev_alloc_type = match state & 0b11 {
0 => AllocationType::Unknown,
1 => AllocationType::Linear,
2 => AllocationType::NonLinear,
_ => unreachable!(),
};
let prev_end = self.region.offset + free_start;
let mut offset = align_up(prev_end, alignment);
if prev_end > 0
&& are_blocks_on_same_page(prev_end, 0, offset, self.buffer_image_granularity)
&& has_granularity_conflict(prev_alloc_type, allocation_type)
{
offset = align_up(offset, self.buffer_image_granularity);
}
let free_start = offset - self.region.offset + size;
if free_start > self.region.size {
return Err(SuballocationCreationError::OutOfRegionMemory);
}
let new_state = free_start << 2 | allocation_type as u64;
match self.state.compare_exchange_weak(
state,
new_state,
Ordering::Release,
Ordering::Relaxed,
) {
Ok(_) => {
return Ok(MemoryAlloc {
offset,
size,
allocation_type,
mapped_ptr: self.region.mapped_ptr.and_then(|ptr| {
NonNull::new(ptr.as_ptr().add((offset - self.region.offset) as usize))
}),
atom_size: self.region.atom_size,
parent: AllocParent::Bump(self.clone()),
});
}
Err(new_state) => {
state = new_state;
backoff.spin();
}
}
}
}
#[inline]
fn region(&self) -> &MemoryAlloc {
&self.region
}
#[inline]
fn try_into_region(self) -> Result<MemoryAlloc, Self> {
Arc::try_unwrap(self).map(|allocator| allocator.region)
}
#[inline]
fn free_size(&self) -> DeviceSize {
self.region.size - (self.state.load(Ordering::Acquire) >> 2)
}
#[inline]
fn cleanup(&mut self) {
let _ = self.try_reset();
}
}
unsafe impl DeviceOwned for BumpAllocator {
#[inline]
fn device(&self) -> &Arc<Device> {
self.device_memory.device()
}
}
/// Error that can be returned when resetting the [`BumpAllocator`].
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct BumpAllocatorResetError;
impl Error for BumpAllocatorResetError {}
impl Display for BumpAllocatorResetError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.write_str("the allocator is still in use")
}
}
pub(crate) fn align_up(val: DeviceSize, alignment: DeviceSize) -> DeviceSize {
align_down(val + alignment - 1, alignment)
}
fn align_down(val: DeviceSize, alignment: DeviceSize) -> DeviceSize {
debug_assert!(alignment.is_power_of_two());
val & !(alignment - 1)
}
/// Checks if resouces A and B share a page.
///
/// > **Note**: Assumes `a_offset + a_size > 0` and `a_offset + a_size <= b_offset`.
fn are_blocks_on_same_page(
a_offset: DeviceSize,
a_size: DeviceSize,
b_offset: DeviceSize,
page_size: DeviceSize,
) -> bool {
debug_assert!(a_offset + a_size > 0);
debug_assert!(a_offset + a_size <= b_offset);
let a_end = a_offset + a_size - 1;
let a_end_page = align_down(a_end, page_size);
let b_start_page = align_down(b_offset, page_size);
a_end_page == b_start_page
}
/// Allocators for memory on the host, used to speed up the allocators for the device.
mod host {
use std::num::NonZeroUsize;
/// Allocates objects from a pool on the host, which has the following benefits:
///
/// - Allocation is much faster because there is no need to consult the global allocator or even
/// worse, the operating system, each time a small object needs to be created.
/// - Freeing is extremely fast, because the whole pool can be dropped at once. This is
/// particularily useful for linked structures, whose nodes need to be freed one-by-one by
/// traversing the whole structure otherwise.
/// - Cache locality is somewhat improved for linked structures with few nodes.
///
/// The allocator doesn't hand out pointers but rather IDs that are relative to the pool. This
/// simplifies the logic because the pool can easily be moved and hence also resized, but the
/// downside is that the whole pool and possibly also the free-list must be copied when it runs
/// out of memory. It is therefore best to start out with a safely large capacity.
#[derive(Debug)]
pub(super) struct PoolAllocator<T> {
pool: Vec<T>,
// LIFO list of free allocations, which means that newly freed allocations are always
// reused first before bumping the free start.
free_list: Vec<SlotId>,
}
impl<T> PoolAllocator<T> {
pub fn new(capacity: usize) -> Self {
debug_assert!(capacity > 0);
let mut pool = Vec::new();
let mut free_list = Vec::new();
pool.reserve_exact(capacity);
free_list.reserve_exact(capacity);
// All IDs are free at the start.
for index in (1..=capacity).rev() {
free_list.push(SlotId(NonZeroUsize::new(index).unwrap()));
}
PoolAllocator { pool, free_list }
}
/// Allocates a slot and initializes it with the provided value. Returns the ID of the slot.
pub fn allocate(&mut self, val: T) -> SlotId {
let id = self.free_list.pop().unwrap_or_else(|| {
// The free-list is empty, we need another pool.
let new_len = self.pool.len() * 3 / 2;
let additional = new_len - self.pool.len();
self.pool.reserve_exact(additional);
self.free_list.reserve_exact(additional);
// Add the new IDs to the free-list.
let len = self.pool.len();
let cap = self.pool.capacity();
for id in (len + 2..=cap).rev() {
// SAFETY: The `new_unchecked` is safe because:
// - `id` is bound to the range [len + 2, cap].
// - There is no way to add 2 to an unsigned integer (`len`) such that the
// result is 0, except for an overflow, which is why rustc can't optimize this
// out (unlike in the above loop where the range has a constant start).
// - `Vec::reserve_exact` panics if the new capacity exceeds `isize::MAX` bytes,
// so the length of the pool can not be `usize::MAX - 1`.
let id = SlotId(unsafe { NonZeroUsize::new_unchecked(id) });
self.free_list.push(id);
}
// Smallest free ID.
SlotId(NonZeroUsize::new(len + 1).unwrap())
});
if let Some(x) = self.pool.get_mut(id.0.get() - 1) {
// We're reusing a slot, initialize it with the new value.
*x = val;
} else {
// We're using a fresh slot. We always put IDs in order into the free-list, so the
// next free ID must be for the slot right after the end of the occupied slots.
debug_assert!(id.0.get() - 1 == self.pool.len());
self.pool.push(val);
}
id
}
/// Returns the slot with the given ID to the allocator to be reused. The [`SlotId`] should
/// not be used again afterward.
pub fn free(&mut self, id: SlotId) {
debug_assert!(!self.free_list.contains(&id));
self.free_list.push(id);
}
/// Returns a mutable reference to the slot with the given ID.
pub fn get_mut(&mut self, id: SlotId) -> &mut T {
debug_assert!(!self.free_list.contains(&id));
// SAFETY: This is safe because:
// - The only way to obtain a `SlotId` is through `Self::allocate`.
// - `Self::allocate` returns `SlotId`s in the range [1, self.pool.len()].
// - `self.pool` only grows and never shrinks.
unsafe { self.pool.get_unchecked_mut(id.0.get() - 1) }
}
}
impl<T: Copy> PoolAllocator<T> {
/// Returns a copy of the slot with the given ID.
pub fn get(&self, id: SlotId) -> T {
debug_assert!(!self.free_list.contains(&id));
// SAFETY: Same as the `get_unchecked_mut` above.
*unsafe { self.pool.get_unchecked(id.0.get() - 1) }
}
}
/// ID of a slot in the pool of the `host::PoolAllocator`. This is used to limit the visibility
/// of the actual ID to this `host` module, making it easier to reason about unsafe code.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub(super) struct SlotId(NonZeroUsize);
}
#[cfg(test)]
mod tests {
use super::*;
use std::thread;
const DUMMY_INFO: SuballocationCreateInfo = SuballocationCreateInfo {
size: 1,
alignment: 1,
allocation_type: AllocationType::Unknown,
_ne: crate::NonExhaustive(()),
};
const DUMMY_INFO_LINEAR: SuballocationCreateInfo = SuballocationCreateInfo {
allocation_type: AllocationType::Linear,
..DUMMY_INFO
};
#[test]
fn free_list_allocator_capacity() {
const THREADS: DeviceSize = 12;
const ALLOCATIONS_PER_THREAD: DeviceSize = 100;
const ALLOCATION_STEP: DeviceSize = 117;
const REGION_SIZE: DeviceSize =
(ALLOCATION_STEP * (THREADS + 1) * THREADS / 2) * ALLOCATIONS_PER_THREAD;
let allocator = dummy_allocator!(FreeListAllocator, REGION_SIZE);
let allocs = ArrayQueue::new((ALLOCATIONS_PER_THREAD * THREADS) as usize);
// Using threads to randomize allocation order.
thread::scope(|scope| {
for i in 1..=THREADS {
let (allocator, allocs) = (&allocator, &allocs);
scope.spawn(move || {
let size = i * ALLOCATION_STEP;
for _ in 0..ALLOCATIONS_PER_THREAD {
allocs
.push(
allocator
.allocate(SuballocationCreateInfo { size, ..DUMMY_INFO })
.unwrap(),
)
.unwrap();
}
});
}
});
assert!(allocator.allocate(DUMMY_INFO).is_err());
assert!(allocator.free_size() == 0);
drop(allocs);
assert!(allocator.free_size() == REGION_SIZE);
assert!(allocator
.allocate(SuballocationCreateInfo {
size: REGION_SIZE,
..DUMMY_INFO
})
.is_ok());
}
#[test]
fn free_list_allocator_respects_alignment() {
const INFO: SuballocationCreateInfo = SuballocationCreateInfo {
alignment: 256,
..DUMMY_INFO
};
const REGION_SIZE: DeviceSize = 10 * INFO.alignment;
let allocator = dummy_allocator!(FreeListAllocator, REGION_SIZE);
let mut allocs = Vec::with_capacity(10);
for _ in 0..10 {
allocs.push(allocator.allocate(INFO).unwrap());
}
assert!(allocator.allocate(INFO).is_err());
assert!(allocator.free_size() == REGION_SIZE - 10);
}
#[test]
fn free_list_allocator_respects_granularity() {
const GRANULARITY: DeviceSize = 16;
const REGION_SIZE: DeviceSize = 2 * GRANULARITY;
let allocator = dummy_allocator!(FreeListAllocator, REGION_SIZE, GRANULARITY);
let mut linear_allocs = Vec::with_capacity(GRANULARITY as usize);
let mut non_linear_allocs = Vec::with_capacity(GRANULARITY as usize);
for i in 0..REGION_SIZE {
if i % 2 == 0 {
linear_allocs.push(
allocator
.allocate(SuballocationCreateInfo {
allocation_type: AllocationType::Linear,
..DUMMY_INFO
})
.unwrap(),
);
} else {
non_linear_allocs.push(
allocator
.allocate(SuballocationCreateInfo {
allocation_type: AllocationType::NonLinear,
..DUMMY_INFO
})
.unwrap(),
);
}
}
assert!(allocator.allocate(DUMMY_INFO_LINEAR).is_err());
assert!(allocator.free_size() == 0);
drop(linear_allocs);
assert!(allocator
.allocate(SuballocationCreateInfo {
size: GRANULARITY,
..DUMMY_INFO
})
.is_ok());
let _alloc = allocator.allocate(DUMMY_INFO).unwrap();
assert!(allocator.allocate(DUMMY_INFO).is_err());
assert!(allocator.allocate(DUMMY_INFO_LINEAR).is_err());
}
#[test]
fn pool_allocator_capacity() {
const BLOCK_SIZE: DeviceSize = 1024;
fn dummy_allocator(
device: Arc<Device>,
allocation_size: DeviceSize,
) -> Arc<PoolAllocator<BLOCK_SIZE>> {
let device_memory = DeviceMemory::allocate(
device,
MemoryAllocateInfo {
allocation_size,
memory_type_index: 0,
..Default::default()
},
)
.unwrap();
PoolAllocator::new(MemoryAlloc::new(device_memory).unwrap(), 1)
}
let (device, _) = gfx_dev_and_queue!();
assert_should_panic!({ dummy_allocator(device.clone(), BLOCK_SIZE - 1) });
let allocator = dummy_allocator(device.clone(), 2 * BLOCK_SIZE - 1);
{
let alloc = allocator.allocate(DUMMY_INFO).unwrap();
assert!(allocator.allocate(DUMMY_INFO).is_err());
drop(alloc);
let _alloc = allocator.allocate(DUMMY_INFO).unwrap();
}
let allocator = dummy_allocator(device, 2 * BLOCK_SIZE);
{
let alloc1 = allocator.allocate(DUMMY_INFO).unwrap();
let alloc2 = allocator.allocate(DUMMY_INFO).unwrap();
assert!(allocator.allocate(DUMMY_INFO).is_err());
drop(alloc1);
let alloc1 = allocator.allocate(DUMMY_INFO).unwrap();
assert!(allocator.allocate(DUMMY_INFO).is_err());
drop(alloc1);
drop(alloc2);
let _alloc1 = allocator.allocate(DUMMY_INFO).unwrap();
let _alloc2 = allocator.allocate(DUMMY_INFO).unwrap();
}
}
#[test]
fn pool_allocator_respects_alignment() {
const BLOCK_SIZE: DeviceSize = 1024 + 128;
const INFO_A: SuballocationCreateInfo = SuballocationCreateInfo {
size: BLOCK_SIZE,
alignment: 256,
..DUMMY_INFO
};
const INFO_B: SuballocationCreateInfo = SuballocationCreateInfo {
size: 1024,
..INFO_A
};
let allocator = {
let (device, _) = gfx_dev_and_queue!();
let device_memory = DeviceMemory::allocate(
device,
MemoryAllocateInfo {
allocation_size: 10 * BLOCK_SIZE,
memory_type_index: 0,
..Default::default()
},
)
.unwrap();
PoolAllocator::<BLOCK_SIZE>::new(MemoryAlloc::new(device_memory).unwrap(), 1)
};
// This uses the fact that block indices are inserted into the free-list in order, so
// the first allocation succeeds because the block has an even index, while the second
// has an odd index.
allocator.allocate(INFO_A).unwrap();
assert!(allocator.allocate(INFO_A).is_err());
allocator.allocate(INFO_A).unwrap();
assert!(allocator.allocate(INFO_A).is_err());
for _ in 0..10 {
allocator.allocate(INFO_B).unwrap();
}
}
#[test]
fn pool_allocator_respects_granularity() {
const BLOCK_SIZE: DeviceSize = 128;
fn dummy_allocator(
device: Arc<Device>,
allocation_type: AllocationType,
) -> Arc<PoolAllocator<BLOCK_SIZE>> {
let device_memory = DeviceMemory::allocate(
device,
MemoryAllocateInfo {
allocation_size: 1024,
memory_type_index: 0,
..Default::default()
},
)
.unwrap();
let mut region = MemoryAlloc::new(device_memory).unwrap();
unsafe { region.set_allocation_type(allocation_type) };
PoolAllocator::new(region, 256)
}
let (device, _) = gfx_dev_and_queue!();
let allocator = dummy_allocator(device.clone(), AllocationType::Unknown);
assert!(allocator.block_count() == 4);
let allocator = dummy_allocator(device.clone(), AllocationType::Linear);
assert!(allocator.block_count() == 8);
let allocator = dummy_allocator(device, AllocationType::NonLinear);
assert!(allocator.block_count() == 8);
}
#[test]
fn buddy_allocator_capacity() {
const MAX_ORDER: usize = 10;
const REGION_SIZE: DeviceSize = BuddyAllocator::MIN_NODE_SIZE << MAX_ORDER;
let allocator = dummy_allocator!(BuddyAllocator, REGION_SIZE);
let mut allocs = Vec::with_capacity(1 << MAX_ORDER);
for order in 0..=MAX_ORDER {
let size = BuddyAllocator::MIN_NODE_SIZE << order;
for _ in 0..1 << (MAX_ORDER - order) {
allocs.push(
allocator
.allocate(SuballocationCreateInfo { size, ..DUMMY_INFO })
.unwrap(),
);
}
assert!(allocator.allocate(DUMMY_INFO).is_err());
assert!(allocator.free_size() == 0);
allocs.clear();
}
let mut orders = (0..MAX_ORDER).collect::<Vec<_>>();
for mid in 0..MAX_ORDER {
orders.rotate_left(mid);
for &order in &orders {
let size = BuddyAllocator::MIN_NODE_SIZE << order;
allocs.push(
allocator
.allocate(SuballocationCreateInfo { size, ..DUMMY_INFO })
.unwrap(),
);
}
let _alloc = allocator.allocate(DUMMY_INFO).unwrap();
assert!(allocator.allocate(DUMMY_INFO).is_err());
assert!(allocator.free_size() == 0);
allocs.clear();
}
}
#[test]
fn buddy_allocator_respects_alignment() {
const REGION_SIZE: DeviceSize = 4096;
let allocator = dummy_allocator!(BuddyAllocator, REGION_SIZE);
{
const INFO: SuballocationCreateInfo = SuballocationCreateInfo {
alignment: 4096,
..DUMMY_INFO
};
let _alloc = allocator.allocate(INFO).unwrap();
assert!(allocator.allocate(INFO).is_err());
assert!(allocator.free_size() == REGION_SIZE - BuddyAllocator::MIN_NODE_SIZE);
}
{
const INFO_A: SuballocationCreateInfo = SuballocationCreateInfo {
alignment: 256,
..DUMMY_INFO
};
const ALLOCATIONS_A: DeviceSize = REGION_SIZE / INFO_A.alignment;
const INFO_B: SuballocationCreateInfo = SuballocationCreateInfo {
alignment: 16,
..DUMMY_INFO
};
const ALLOCATIONS_B: DeviceSize = REGION_SIZE / INFO_B.alignment - ALLOCATIONS_A;
let mut allocs =
Vec::with_capacity((REGION_SIZE / BuddyAllocator::MIN_NODE_SIZE) as usize);
for _ in 0..ALLOCATIONS_A {
allocs.push(allocator.allocate(INFO_A).unwrap());
}
assert!(allocator.allocate(INFO_A).is_err());
assert!(
allocator.free_size()
== REGION_SIZE - ALLOCATIONS_A * BuddyAllocator::MIN_NODE_SIZE
);
for _ in 0..ALLOCATIONS_B {
allocs.push(allocator.allocate(INFO_B).unwrap());
}
assert!(allocator.allocate(DUMMY_INFO).is_err());
assert!(allocator.free_size() == 0);
}
}
#[test]
fn buddy_allocator_respects_granularity() {
const GRANULARITY: DeviceSize = 256;
const REGION_SIZE: DeviceSize = 2 * GRANULARITY;
let allocator = dummy_allocator!(BuddyAllocator, REGION_SIZE, GRANULARITY);
{
const ALLOCATIONS: DeviceSize = REGION_SIZE / BuddyAllocator::MIN_NODE_SIZE;
let mut allocs = Vec::with_capacity(ALLOCATIONS as usize);
for _ in 0..ALLOCATIONS {
allocs.push(allocator.allocate(DUMMY_INFO_LINEAR).unwrap());
}
assert!(allocator.allocate(DUMMY_INFO_LINEAR).is_err());
assert!(allocator.free_size() == 0);
}
{
let _alloc1 = allocator.allocate(DUMMY_INFO).unwrap();
let _alloc2 = allocator.allocate(DUMMY_INFO).unwrap();
assert!(allocator.allocate(DUMMY_INFO).is_err());
assert!(allocator.free_size() == 0);
}
}
#[test]
fn bump_allocator_respects_alignment() {
const INFO: SuballocationCreateInfo = SuballocationCreateInfo {
alignment: 16,
..DUMMY_INFO
};
let allocator = dummy_allocator!(BumpAllocator, INFO.alignment * 10);
for _ in 0..10 {
allocator.allocate(INFO).unwrap();
}
assert!(allocator.allocate(INFO).is_err());
for _ in 0..INFO.alignment - 1 {
allocator.allocate(DUMMY_INFO).unwrap();
}
assert!(allocator.allocate(INFO).is_err());
assert!(allocator.free_size() == 0);
}
#[test]
fn bump_allocator_respects_granularity() {
const ALLOCATIONS: DeviceSize = 10;
const GRANULARITY: DeviceSize = 1024;
let mut allocator = dummy_allocator!(BumpAllocator, GRANULARITY * ALLOCATIONS, GRANULARITY);
for i in 0..ALLOCATIONS {
for _ in 0..GRANULARITY {
allocator
.allocate(SuballocationCreateInfo {
allocation_type: if i % 2 == 0 {
AllocationType::NonLinear
} else {
AllocationType::Linear
},
..DUMMY_INFO
})
.unwrap();
}
}
assert!(allocator.allocate(DUMMY_INFO_LINEAR).is_err());
assert!(allocator.free_size() == 0);
allocator.try_reset().unwrap();
for i in 0..ALLOCATIONS {
allocator
.allocate(SuballocationCreateInfo {
allocation_type: if i % 2 == 0 {
AllocationType::Linear
} else {
AllocationType::NonLinear
},
..DUMMY_INFO
})
.unwrap();
}
assert!(allocator.allocate(DUMMY_INFO_LINEAR).is_err());
assert!(allocator.free_size() == GRANULARITY - 1);
}
#[test]
fn bump_allocator_syncness() {
const THREADS: DeviceSize = 12;
const ALLOCATIONS_PER_THREAD: DeviceSize = 100_000;
const ALLOCATION_STEP: DeviceSize = 117;
const REGION_SIZE: DeviceSize =
(ALLOCATION_STEP * (THREADS + 1) * THREADS / 2) * ALLOCATIONS_PER_THREAD;
let mut allocator = dummy_allocator!(BumpAllocator, REGION_SIZE);
thread::scope(|scope| {
for i in 1..=THREADS {
let allocator = &allocator;
scope.spawn(move || {
let size = i * ALLOCATION_STEP;
for _ in 0..ALLOCATIONS_PER_THREAD {
allocator
.allocate(SuballocationCreateInfo { size, ..DUMMY_INFO })
.unwrap();
}
});
}
});
assert!(allocator.allocate(DUMMY_INFO).is_err());
assert!(allocator.free_size() == 0);
allocator.try_reset().unwrap();
assert!(allocator.free_size() == REGION_SIZE);
}
macro_rules! dummy_allocator {
($type:ty, $size:expr) => {
dummy_allocator!($type, $size, 1)
};
($type:ty, $size:expr, $granularity:expr) => {
dummy_allocator!($type, $size, $granularity, AllocationType::Unknown)
};
($type:ty, $size:expr, $granularity:expr, $allocation_type:expr) => {{
let (device, _) = gfx_dev_and_queue!();
let device_memory = DeviceMemory::allocate(
device,
MemoryAllocateInfo {
allocation_size: $size,
memory_type_index: 0,
..Default::default()
},
)
.unwrap();
let mut allocator = <$type>::new(MemoryAlloc::new(device_memory).unwrap());
Arc::get_mut(&mut allocator)
.unwrap()
.buffer_image_granularity = $granularity;
allocator
}};
}
use crate::memory::MemoryAllocateInfo;
pub(self) use dummy_allocator;
}