1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646
|
/**************************** vectori256.h *******************************
* Author: Agner Fog
* Date created: 2012-05-30
* Last modified: 2017-02-19
* Version: 1.27
* Project: vector classes
* Description:
* Header file defining integer vector classes as interface to intrinsic
* functions in x86 microprocessors with AVX2 and later instruction sets.
*
* Instructions:
* Use Gnu, Intel or Microsoft C++ compiler. Compile for the desired
* instruction set, which must be at least AVX2.
*
* The following vector classes are defined here:
* Vec256b Vector of 256 1-bit unsigned integers or Booleans
* Vec32c Vector of 32 8-bit signed integers
* Vec32uc Vector of 32 8-bit unsigned integers
* Vec32cb Vector of 32 Booleans for use with Vec32c and Vec32uc
* Vec16s Vector of 16 16-bit signed integers
* Vec16us Vector of 16 16-bit unsigned integers
* Vec16sb Vector of 16 Booleans for use with Vec16s and Vec16us
* Vec8i Vector of 8 32-bit signed integers
* Vec8ui Vector of 8 32-bit unsigned integers
* Vec8ib Vector of 8 Booleans for use with Vec8i and Vec8ui
* Vec4q Vector of 4 64-bit signed integers
* Vec4uq Vector of 4 64-bit unsigned integers
* Vec4qb Vector of 4 Booleans for use with Vec4q and Vec4uq
*
* Each vector object is represented internally in the CPU as a 256-bit register.
* This header file defines operators and functions for these vectors.
*
* For example:
* Vec8i a(1,2,3,4,5,6,7,8), b(9,10,11,12,13,14,15,16), c;
* c = a + b; // now c contains (10,12,14,16,18,20,22,24)
*
* For detailed instructions, see VectorClass.pdf
*
* (c) Copyright 2012-2017 GNU General Public License http://www.gnu.org/licenses
*****************************************************************************/
// check combination of header files
#if defined (VECTORI256_H)
#if VECTORI256_H != 2
#error Two different versions of vectori256.h included
#endif
#else
#define VECTORI256_H 2
#ifdef VECTORF256_H
#error Please put header file vectori256.h before vectorf256.h
#endif
#if INSTRSET < 8 // AVX2 required
#error Wrong instruction set for vectori256.h, AVX2 required or use vectori256e.h
#endif
#include "vectori128.h"
#ifdef VCL_NAMESPACE
namespace VCL_NAMESPACE {
#endif
/*****************************************************************************
*
* Join two 128-bit vectors
*
*****************************************************************************/
#define set_m128ir(lo,hi) _mm256_inserti128_si256(_mm256_castsi128_si256(lo),(hi),1)
/*****************************************************************************
*
* Vector of 256 1-bit unsigned integers or Booleans
*
*****************************************************************************/
class Vec256b {
protected:
__m256i ymm; // Integer vector
public:
// Default constructor:
Vec256b() {
}
// Constructor to broadcast the same value into all elements
// Removed because of undesired implicit conversions
//Vec256b(int i) {
// ymm = _mm256_set1_epi32(-(i & 1));}
// Constructor to build from two Vec128b:
Vec256b(Vec128b const & a0, Vec128b const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec256b(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec256b & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Type cast operator to convert to __m256i used in intrinsics
operator __m256i() const {
return ymm;
}
// Member function to load from array (unaligned)
Vec256b & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
// You may use load_a instead of load if you are certain that p points to an address
// divisible by 32, but there is hardly any speed advantage of load_a on modern processors
Vec256b & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Member function to store into array (unaligned)
void store(void * p) const {
_mm256_storeu_si256((__m256i*)p, ymm);
}
// Member function to store into array, aligned by 32
// You may use store_a instead of store if you are certain that p points to an address
// divisible by 32, but there is hardly any speed advantage of load_a on modern processors
void store_a(void * p) const {
_mm256_store_si256((__m256i*)p, ymm);
}
// Member function to change a single bit
// Note: This function is inefficient. Use load function if changing more than one bit
Vec256b const & set_bit(uint32_t index, int value) {
static uint64_t m[8] = {0,0,0,0,1,0,0,0};
int wi = (index >> 6) & 3; // qword index
int bi = index & 0x3F; // bit index within qword w
__m256i mask = Vec256b().load(m+4-wi); // 1 in qword number wi
mask = _mm256_sll_epi64(mask,_mm_cvtsi32_si128(bi)); // mask with bit number b set
if (value & 1) {
ymm = _mm256_or_si256(mask,ymm);
}
else {
ymm = _mm256_andnot_si256(mask,ymm);
}
return *this;
}
// Member function to get a single bit
// Note: This function is inefficient. Use store function if reading more than one bit
int get_bit(uint32_t index) const {
union {
__m256i x;
uint8_t i[32];
} u;
u.x = ymm;
int wi = (index >> 3) & 0x1F; // byte index
int bi = index & 7; // bit index within byte w
return (u.i[wi] >> bi) & 1;
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
bool operator [] (uint32_t index) const {
return get_bit(index) != 0;
}
// Member functions to split into two Vec128b:
Vec128b get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec128b get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
static int size() {
return 256;
}
};
// Define operators for this class
// vector operator & : bitwise and
static inline Vec256b operator & (Vec256b const & a, Vec256b const & b) {
return _mm256_and_si256(a, b);
}
static inline Vec256b operator && (Vec256b const & a, Vec256b const & b) {
return a & b;
}
// vector operator | : bitwise or
static inline Vec256b operator | (Vec256b const & a, Vec256b const & b) {
return _mm256_or_si256(a, b);
}
static inline Vec256b operator || (Vec256b const & a, Vec256b const & b) {
return a | b;
}
// vector operator ^ : bitwise xor
static inline Vec256b operator ^ (Vec256b const & a, Vec256b const & b) {
return _mm256_xor_si256(a, b);
}
// vector operator ~ : bitwise not
static inline Vec256b operator ~ (Vec256b const & a) {
return _mm256_xor_si256(a, _mm256_set1_epi32(-1));
}
// vector operator &= : bitwise and
static inline Vec256b & operator &= (Vec256b & a, Vec256b const & b) {
a = a & b;
return a;
}
// vector operator |= : bitwise or
static inline Vec256b & operator |= (Vec256b & a, Vec256b const & b) {
a = a | b;
return a;
}
// vector operator ^= : bitwise xor
static inline Vec256b & operator ^= (Vec256b & a, Vec256b const & b) {
a = a ^ b;
return a;
}
// Define functions for this class
// function andnot: a & ~ b
static inline Vec256b andnot (Vec256b const & a, Vec256b const & b) {
return _mm256_andnot_si256(b, a);
}
/*****************************************************************************
*
* Generate compile-time constant vector
*
*****************************************************************************/
// Generate a constant vector of 8 integers stored in memory.
// Can be converted to any integer vector type
template <int32_t i0, int32_t i1, int32_t i2, int32_t i3, int32_t i4, int32_t i5, int32_t i6, int32_t i7>
static inline __m256i constant8i() {
static const union {
int32_t i[8];
__m256i ymm;
} u = {{i0,i1,i2,i3,i4,i5,i6,i7}};
return u.ymm;
}
template <uint32_t i0, uint32_t i1, uint32_t i2, uint32_t i3, uint32_t i4, uint32_t i5, uint32_t i6, uint32_t i7>
static inline __m256i constant8ui() {
return constant8i<int32_t(i0), int32_t(i1), int32_t(i2), int32_t(i3), int32_t(i4), int32_t(i5), int32_t(i6), int32_t(i7)>();
}
/*****************************************************************************
*
* selectb function
*
*****************************************************************************/
// Select between two sources, byte by byte. Used in various functions and operators
// Corresponds to this pseudocode:
// for (int i = 0; i < 32; i++) result[i] = s[i] ? a[i] : b[i];
// Each byte in s must be either 0 (false) or 0xFF (true). No other values are allowed.
// Only bit 7 in each byte of s is checked,
static inline __m256i selectb (__m256i const & s, __m256i const & a, __m256i const & b) {
return _mm256_blendv_epi8 (b, a, s);
}
/*****************************************************************************
*
* Horizontal Boolean functions
*
*****************************************************************************/
// horizontal_and. Returns true if all bits are 1
static inline bool horizontal_and (Vec256b const & a) {
return _mm256_testc_si256(a,constant8i<-1,-1,-1,-1,-1,-1,-1,-1>()) != 0;
}
// horizontal_or. Returns true if at least one bit is 1
static inline bool horizontal_or (Vec256b const & a) {
return ! _mm256_testz_si256(a,a);
}
/*****************************************************************************
*
* Vector of 32 8-bit signed integers
*
*****************************************************************************/
class Vec32c : public Vec256b {
public:
// Default constructor:
Vec32c(){
}
// Constructor to broadcast the same value into all elements:
Vec32c(int i) {
ymm = _mm256_set1_epi8((char)i);
}
// Constructor to build from all elements:
Vec32c(int8_t i0, int8_t i1, int8_t i2, int8_t i3, int8_t i4, int8_t i5, int8_t i6, int8_t i7,
int8_t i8, int8_t i9, int8_t i10, int8_t i11, int8_t i12, int8_t i13, int8_t i14, int8_t i15,
int8_t i16, int8_t i17, int8_t i18, int8_t i19, int8_t i20, int8_t i21, int8_t i22, int8_t i23,
int8_t i24, int8_t i25, int8_t i26, int8_t i27, int8_t i28, int8_t i29, int8_t i30, int8_t i31) {
ymm = _mm256_setr_epi8(i0, i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, i14, i15,
i16, i17, i18, i19, i20, i21, i22, i23, i24, i25, i26, i27, i28, i29, i30, i31);
}
// Constructor to build from two Vec16c:
Vec32c(Vec16c const & a0, Vec16c const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec32c(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec32c & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Type cast operator to convert to __m256i used in intrinsics
operator __m256i() const {
return ymm;
}
// Member function to load from array (unaligned)
Vec32c & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec32c & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Partial load. Load n elements and set the rest to 0
Vec32c & load_partial(int n, void const * p) {
if (n <= 0) {
*this = 0;
}
else if (n <= 16) {
*this = Vec32c(Vec16c().load_partial(n, p), 0);
}
else if (n < 32) {
*this = Vec32c(Vec16c().load(p), Vec16c().load_partial(n-16, (char const*)p+16));
}
else {
load(p);
}
return *this;
}
// Partial store. Store n elements
void store_partial(int n, void * p) const {
if (n <= 0) {
return;
}
else if (n <= 16) {
get_low().store_partial(n, p);
}
else if (n < 32) {
get_low().store(p);
get_high().store_partial(n-16, (char*)p+16);
}
else {
store(p);
}
}
// cut off vector to n elements. The last 32-n elements are set to zero
Vec32c & cutoff(int n) {
if (uint32_t(n) >= 32) return *this;
static const union {
int32_t i[16];
char c[64];
} mask = {{-1,-1,-1,-1,-1,-1,-1,-1,0,0,0,0,0,0,0,0}};
*this &= Vec32c().load(mask.c+32-n);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec32c const & insert(uint32_t index, int8_t value) {
static const int8_t maskl[64] = {0,0,0,0, 0,0,0,0, 0,0,0,0 ,0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0,
-1,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0 ,0,0,0,0, 0,0,0,0, 0,0,0,0};
__m256i broad = _mm256_set1_epi8(value); // broadcast value into all elements
__m256i mask = _mm256_loadu_si256((__m256i const*)(maskl+32-(index & 0x1F))); // mask with FF at index position
ymm = selectb(mask,broad,ymm);
return *this;
}
// Member function extract a single element from vector
int8_t extract(uint32_t index) const {
int8_t x[32];
store(x);
return x[index & 0x1F];
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
int8_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec16c:
Vec16c get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec16c get_high() const {
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
return _mm256_extractf128_si256(ymm,1); // workaround bug in MS compiler VS 11
#else
return _mm256_extracti128_si256(ymm,1);
#endif
}
static int size() {
return 32;
}
};
/*****************************************************************************
*
* Vec32cb: Vector of 32 Booleans for use with Vec32c and Vec32uc
*
*****************************************************************************/
class Vec32cb : public Vec32c {
public:
// Default constructor:
Vec32cb(){
}
// Constructor to build from all elements:
Vec32cb(bool x0, bool x1, bool x2, bool x3, bool x4, bool x5, bool x6, bool x7,
bool x8, bool x9, bool x10, bool x11, bool x12, bool x13, bool x14, bool x15,
bool x16, bool x17, bool x18, bool x19, bool x20, bool x21, bool x22, bool x23,
bool x24, bool x25, bool x26, bool x27, bool x28, bool x29, bool x30, bool x31) :
Vec32c(-int8_t(x0), -int8_t(x1), -int8_t(x2), -int8_t(x3), -int8_t(x4), -int8_t(x5), -int8_t(x6), -int8_t(x7),
-int8_t(x8), -int8_t(x9), -int8_t(x10), -int8_t(x11), -int8_t(x12), -int8_t(x13), -int8_t(x14), -int8_t(x15),
-int8_t(x16), -int8_t(x17), -int8_t(x18), -int8_t(x19), -int8_t(x20), -int8_t(x21), -int8_t(x22), -int8_t(x23),
-int8_t(x24), -int8_t(x25), -int8_t(x26), -int8_t(x27), -int8_t(x28), -int8_t(x29), -int8_t(x30), -int8_t(x31))
{}
// Constructor to convert from type __m256i used in intrinsics:
Vec32cb(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec32cb & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Constructor to broadcast scalar value:
Vec32cb(bool b) : Vec32c(-int8_t(b)) {
}
// Assignment operator to broadcast scalar value:
Vec32cb & operator = (bool b) {
*this = Vec32cb(b);
return *this;
}
private: // Prevent constructing from int, etc.
Vec32cb(int b);
Vec32cb & operator = (int x);
public:
// Member functions to split into two Vec16c:
Vec16cb get_low() const {
return Vec16cb(Vec32c::get_low());
}
Vec16cb get_high() const {
return Vec16cb(Vec32c::get_high());
}
Vec32cb & insert (int index, bool a) {
Vec32c::insert(index, -(int)a);
return *this;
}
// Member function extract a single element from vector
bool extract(uint32_t index) const {
return Vec32c::extract(index) != 0;
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
bool operator [] (uint32_t index) const {
return extract(index);
}
};
/*****************************************************************************
*
* Define operators for Vec32cb
*
*****************************************************************************/
// vector operator & : bitwise and
static inline Vec32cb operator & (Vec32cb const & a, Vec32cb const & b) {
return Vec32cb(Vec256b(a) & Vec256b(b));
}
static inline Vec32cb operator && (Vec32cb const & a, Vec32cb const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec32cb & operator &= (Vec32cb & a, Vec32cb const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec32cb operator | (Vec32cb const & a, Vec32cb const & b) {
return Vec32cb(Vec256b(a) | Vec256b(b));
}
static inline Vec32cb operator || (Vec32cb const & a, Vec32cb const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec32cb & operator |= (Vec32cb & a, Vec32cb const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec32cb operator ^ (Vec32cb const & a, Vec32cb const & b) {
return Vec32cb(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec32cb & operator ^= (Vec32cb & a, Vec32cb const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec32cb operator ~ (Vec32cb const & a) {
return Vec32cb( ~ Vec256b(a));
}
// vector operator ! : element not
static inline Vec32cb operator ! (Vec32cb const & a) {
return ~ a;
}
// vector function andnot
static inline Vec32cb andnot (Vec32cb const & a, Vec32cb const & b) {
return Vec32cb(andnot(Vec256b(a), Vec256b(b)));
}
/*****************************************************************************
*
* Operators for Vec32c
*
*****************************************************************************/
// vector operator + : add element by element
static inline Vec32c operator + (Vec32c const & a, Vec32c const & b) {
return _mm256_add_epi8(a, b);
}
// vector operator += : add
static inline Vec32c & operator += (Vec32c & a, Vec32c const & b) {
a = a + b;
return a;
}
// postfix operator ++
static inline Vec32c operator ++ (Vec32c & a, int) {
Vec32c a0 = a;
a = a + 1;
return a0;
}
// prefix operator ++
static inline Vec32c & operator ++ (Vec32c & a) {
a = a + 1;
return a;
}
// vector operator - : subtract element by element
static inline Vec32c operator - (Vec32c const & a, Vec32c const & b) {
return _mm256_sub_epi8(a, b);
}
// vector operator - : unary minus
static inline Vec32c operator - (Vec32c const & a) {
return _mm256_sub_epi8(_mm256_setzero_si256(), a);
}
// vector operator -= : add
static inline Vec32c & operator -= (Vec32c & a, Vec32c const & b) {
a = a - b;
return a;
}
// postfix operator --
static inline Vec32c operator -- (Vec32c & a, int) {
Vec32c a0 = a;
a = a - 1;
return a0;
}
// prefix operator --
static inline Vec32c & operator -- (Vec32c & a) {
a = a - 1;
return a;
}
// vector operator * : multiply element by element
static inline Vec32c operator * (Vec32c const & a, Vec32c const & b) {
// There is no 8-bit multiply in SSE2. Split into two 16-bit multiplies
__m256i aodd = _mm256_srli_epi16(a,8); // odd numbered elements of a
__m256i bodd = _mm256_srli_epi16(b,8); // odd numbered elements of b
__m256i muleven = _mm256_mullo_epi16(a,b); // product of even numbered elements
__m256i mulodd = _mm256_mullo_epi16(aodd,bodd); // product of odd numbered elements
mulodd = _mm256_slli_epi16(mulodd,8); // put odd numbered elements back in place
__m256i mask = _mm256_set1_epi32(0x00FF00FF); // mask for even positions
__m256i product = selectb(mask,muleven,mulodd); // interleave even and odd
return product;
}
// vector operator *= : multiply
static inline Vec32c & operator *= (Vec32c & a, Vec32c const & b) {
a = a * b;
return a;
}
// vector operator << : shift left all elements
static inline Vec32c operator << (Vec32c const & a, int b) {
uint32_t mask = (uint32_t)0xFF >> (uint32_t)b; // mask to remove bits that are shifted out
__m256i am = _mm256_and_si256(a,_mm256_set1_epi8((char)mask));// remove bits that will overflow
__m256i res = _mm256_sll_epi16(am,_mm_cvtsi32_si128(b)); // 16-bit shifts
return res;
}
// vector operator <<= : shift left
static inline Vec32c & operator <<= (Vec32c & a, int b) {
a = a << b;
return a;
}
// vector operator >> : shift right arithmetic all elements
static inline Vec32c operator >> (Vec32c const & a, int b) {
__m256i aeven = _mm256_slli_epi16(a,8); // even numbered elements of a. get sign bit in position
aeven = _mm256_sra_epi16(aeven,_mm_cvtsi32_si128(b+8)); // shift arithmetic, back to position
__m256i aodd = _mm256_sra_epi16(a,_mm_cvtsi32_si128(b)); // shift odd numbered elements arithmetic
__m256i mask = _mm256_set1_epi32(0x00FF00FF); // mask for even positions
__m256i res = selectb(mask,aeven,aodd); // interleave even and odd
return res;
}
// vector operator >>= : shift right artihmetic
static inline Vec32c & operator >>= (Vec32c & a, int b) {
a = a >> b;
return a;
}
// vector operator == : returns true for elements for which a == b
static inline Vec32cb operator == (Vec32c const & a, Vec32c const & b) {
return _mm256_cmpeq_epi8(a,b);
}
// vector operator != : returns true for elements for which a != b
static inline Vec32cb operator != (Vec32c const & a, Vec32c const & b) {
return Vec32cb(Vec32c(~(a == b)));
}
// vector operator > : returns true for elements for which a > b (signed)
static inline Vec32cb operator > (Vec32c const & a, Vec32c const & b) {
return _mm256_cmpgt_epi8(a,b);
}
// vector operator < : returns true for elements for which a < b (signed)
static inline Vec32cb operator < (Vec32c const & a, Vec32c const & b) {
return b > a;
}
// vector operator >= : returns true for elements for which a >= b (signed)
static inline Vec32cb operator >= (Vec32c const & a, Vec32c const & b) {
return Vec32cb(Vec32c(~(b > a)));
}
// vector operator <= : returns true for elements for which a <= b (signed)
static inline Vec32cb operator <= (Vec32c const & a, Vec32c const & b) {
return b >= a;
}
// vector operator & : bitwise and
static inline Vec32c operator & (Vec32c const & a, Vec32c const & b) {
return Vec32c(Vec256b(a) & Vec256b(b));
}
static inline Vec32c operator && (Vec32c const & a, Vec32c const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec32c & operator &= (Vec32c & a, Vec32c const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec32c operator | (Vec32c const & a, Vec32c const & b) {
return Vec32c(Vec256b(a) | Vec256b(b));
}
static inline Vec32c operator || (Vec32c const & a, Vec32c const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec32c & operator |= (Vec32c & a, Vec32c const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec32c operator ^ (Vec32c const & a, Vec32c const & b) {
return Vec32c(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec32c & operator ^= (Vec32c & a, Vec32c const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec32c operator ~ (Vec32c const & a) {
return Vec32c( ~ Vec256b(a));
}
// vector operator ! : logical not, returns true for elements == 0
static inline Vec32cb operator ! (Vec32c const & a) {
return _mm256_cmpeq_epi8(a,_mm256_setzero_si256());
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 16; i++) result[i] = s[i] ? a[i] : b[i];
// Each byte in s must be either 0 (false) or -1 (true). No other values are allowed.
static inline Vec32c select (Vec32cb const & s, Vec32c const & a, Vec32c const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec32c if_add (Vec32cb const & f, Vec32c const & a, Vec32c const & b) {
return a + (Vec32c(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
static inline int32_t horizontal_add (Vec32c const & a) {
__m256i sum1 = _mm256_sad_epu8(a,_mm256_setzero_si256());
__m256i sum2 = _mm256_shuffle_epi32(sum1,2);
__m256i sum3 = _mm256_add_epi16(sum1,sum2);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum4 = _mm256_extractf128_si256(sum3,1); // bug in MS VS 11
#else
__m128i sum4 = _mm256_extracti128_si256(sum3,1);
#endif
__m128i sum5 = _mm_add_epi16(_mm256_castsi256_si128(sum3),sum4);
int8_t sum6 = (int8_t)_mm_cvtsi128_si32(sum5); // truncate to 8 bits
return sum6; // sign extend to 32 bits
}
// Horizontal add extended: Calculates the sum of all vector elements.
// Each element is sign-extended before addition to avoid overflow
static inline int32_t horizontal_add_x (Vec32c const & a) {
__m256i aeven = _mm256_slli_epi16(a,8); // even numbered elements of a. get sign bit in position
aeven = _mm256_srai_epi16(aeven,8); // sign extend even numbered elements
__m256i aodd = _mm256_srai_epi16(a,8); // sign extend odd numbered elements
__m256i sum1 = _mm256_add_epi16(aeven,aodd); // add even and odd elements
__m256i sum2 = _mm256_hadd_epi16(sum1,sum1); // horizontally add 2x8 elements in 3 steps
__m256i sum3 = _mm256_hadd_epi16(sum2,sum2);
__m256i sum4 = _mm256_hadd_epi16(sum3,sum3);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum5 = _mm256_extractf128_si256(sum4,1); // bug in MS VS 11
#else
__m128i sum5 = _mm256_extracti128_si256(sum4,1); // get high sum
#endif
__m128i sum6 = _mm_add_epi16(_mm256_castsi256_si128(sum4),sum5);// add high and low sum
int16_t sum7 = (int16_t)_mm_cvtsi128_si32(sum6); // 16 bit sum
return sum7; // sign extend to 32 bits
}
// function add_saturated: add element by element, signed with saturation
static inline Vec32c add_saturated(Vec32c const & a, Vec32c const & b) {
return _mm256_adds_epi8(a, b);
}
// function sub_saturated: subtract element by element, signed with saturation
static inline Vec32c sub_saturated(Vec32c const & a, Vec32c const & b) {
return _mm256_subs_epi8(a, b);
}
// function max: a > b ? a : b
static inline Vec32c max(Vec32c const & a, Vec32c const & b) {
return _mm256_max_epi8(a,b);
}
// function min: a < b ? a : b
static inline Vec32c min(Vec32c const & a, Vec32c const & b) {
return _mm256_min_epi8(a,b);
}
// function abs: a >= 0 ? a : -a
static inline Vec32c abs(Vec32c const & a) {
return _mm256_sign_epi8(a,a);
}
// function abs_saturated: same as abs, saturate if overflow
static inline Vec32c abs_saturated(Vec32c const & a) {
__m256i absa = abs(a); // abs(a)
__m256i overfl = _mm256_cmpgt_epi8(_mm256_setzero_si256(),absa); // 0 > a
return _mm256_add_epi8(absa,overfl); // subtract 1 if 0x80
}
// function rotate_left all elements
// Use negative count to rotate right
static inline Vec32c rotate_left(Vec32c const & a, int b) {
__m128i bb = _mm_cvtsi32_si128(b & 7); // b modulo 8
__m128i mbb = _mm_cvtsi32_si128((8-b) & 7); // 8-b modulo 8
__m256i maskeven = _mm256_set1_epi32(0x00FF00FF); // mask for even numbered bytes
__m256i even = _mm256_and_si256(a,maskeven); // even numbered bytes of a
__m256i odd = _mm256_andnot_si256(maskeven,a); // odd numbered bytes of a
__m256i evenleft = _mm256_sll_epi16(even,bb); // even bytes of a << b
__m256i oddleft = _mm256_sll_epi16(odd,bb); // odd bytes of a << b
__m256i evenright = _mm256_srl_epi16(even,mbb); // even bytes of a >> 8-b
__m256i oddright = _mm256_srl_epi16(odd,mbb); // odd bytes of a >> 8-b
__m256i evenrot = _mm256_or_si256(evenleft,evenright); // even bytes of a rotated
__m256i oddrot = _mm256_or_si256(oddleft,oddright); // odd bytes of a rotated
__m256i allrot = selectb(maskeven,evenrot,oddrot); // all bytes rotated
return allrot;
}
/*****************************************************************************
*
* Vector of 16 8-bit unsigned integers
*
*****************************************************************************/
class Vec32uc : public Vec32c {
public:
// Default constructor:
Vec32uc(){
}
// Constructor to broadcast the same value into all elements:
Vec32uc(uint32_t i) {
ymm = _mm256_set1_epi8((char)i);
}
// Constructor to build from all elements:
Vec32uc(uint8_t i0, uint8_t i1, uint8_t i2, uint8_t i3, uint8_t i4, uint8_t i5, uint8_t i6, uint8_t i7,
uint8_t i8, uint8_t i9, uint8_t i10, uint8_t i11, uint8_t i12, uint8_t i13, uint8_t i14, uint8_t i15,
uint8_t i16, uint8_t i17, uint8_t i18, uint8_t i19, uint8_t i20, uint8_t i21, uint8_t i22, uint8_t i23,
uint8_t i24, uint8_t i25, uint8_t i26, uint8_t i27, uint8_t i28, uint8_t i29, uint8_t i30, uint8_t i31) {
ymm = _mm256_setr_epi8(i0, i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, i14, i15,
i16, i17, i18, i19, i20, i21, i22, i23, i24, i25, i26, i27, i28, i29, i30, i31);
}
// Constructor to build from two Vec16uc:
Vec32uc(Vec16uc const & a0, Vec16uc const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec32uc(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec32uc & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Member function to load from array (unaligned)
Vec32uc & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec32uc & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec32uc const & insert(uint32_t index, uint8_t value) {
Vec32c::insert(index, value);
return *this;
}
// Member function extract a single element from vector
uint8_t extract(uint32_t index) const {
return Vec32c::extract(index);
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
uint8_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec16uc:
Vec16uc get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec16uc get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
};
// Define operators for this class
// vector operator + : add
static inline Vec32uc operator + (Vec32uc const & a, Vec32uc const & b) {
return Vec32uc (Vec32c(a) + Vec32c(b));
}
// vector operator - : subtract
static inline Vec32uc operator - (Vec32uc const & a, Vec32uc const & b) {
return Vec32uc (Vec32c(a) - Vec32c(b));
}
// vector operator * : multiply
static inline Vec32uc operator * (Vec32uc const & a, Vec32uc const & b) {
return Vec32uc (Vec32c(a) * Vec32c(b));
}
// vector operator << : shift left all elements
static inline Vec32uc operator << (Vec32uc const & a, uint32_t b) {
uint32_t mask = (uint32_t)0xFF >> (uint32_t)b; // mask to remove bits that are shifted out
__m256i am = _mm256_and_si256(a,_mm256_set1_epi8((char)mask));// remove bits that will overflow
__m256i res = _mm256_sll_epi16(am,_mm_cvtsi32_si128(b)); // 16-bit shifts
return res;
}
// vector operator << : shift left all elements
static inline Vec32uc operator << (Vec32uc const & a, int32_t b) {
return a << (uint32_t)b;
}
// vector operator >> : shift right logical all elements
static inline Vec32uc operator >> (Vec32uc const & a, uint32_t b) {
uint32_t mask = (uint32_t)0xFF << (uint32_t)b; // mask to remove bits that are shifted out
__m256i am = _mm256_and_si256(a,_mm256_set1_epi8((char)mask));// remove bits that will overflow
__m256i res = _mm256_srl_epi16(am,_mm_cvtsi32_si128(b)); // 16-bit shifts
return res;
}
// vector operator >> : shift right logical all elements
static inline Vec32uc operator >> (Vec32uc const & a, int32_t b) {
return a >> (uint32_t)b;
}
// vector operator >>= : shift right artihmetic
static inline Vec32uc & operator >>= (Vec32uc & a, uint32_t b) {
a = a >> b;
return a;
}
// vector operator >= : returns true for elements for which a >= b (unsigned)
static inline Vec32cb operator >= (Vec32uc const & a, Vec32uc const & b) {
return _mm256_cmpeq_epi8(_mm256_max_epu8(a,b), a); // a == max(a,b)
}
// vector operator <= : returns true for elements for which a <= b (unsigned)
static inline Vec32cb operator <= (Vec32uc const & a, Vec32uc const & b) {
return b >= a;
}
// vector operator > : returns true for elements for which a > b (unsigned)
static inline Vec32cb operator > (Vec32uc const & a, Vec32uc const & b) {
return Vec32cb(Vec32c(~(b >= a)));
}
// vector operator < : returns true for elements for which a < b (unsigned)
static inline Vec32cb operator < (Vec32uc const & a, Vec32uc const & b) {
return b > a;
}
// vector operator & : bitwise and
static inline Vec32uc operator & (Vec32uc const & a, Vec32uc const & b) {
return Vec32uc(Vec256b(a) & Vec256b(b));
}
static inline Vec32uc operator && (Vec32uc const & a, Vec32uc const & b) {
return a & b;
}
// vector operator | : bitwise or
static inline Vec32uc operator | (Vec32uc const & a, Vec32uc const & b) {
return Vec32uc(Vec256b(a) | Vec256b(b));
}
static inline Vec32uc operator || (Vec32uc const & a, Vec32uc const & b) {
return a | b;
}
// vector operator ^ : bitwise xor
static inline Vec32uc operator ^ (Vec32uc const & a, Vec32uc const & b) {
return Vec32uc(Vec256b(a) ^ Vec256b(b));
}
// vector operator ~ : bitwise not
static inline Vec32uc operator ~ (Vec32uc const & a) {
return Vec32uc( ~ Vec256b(a));
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 32; i++) result[i] = s[i] ? a[i] : b[i];
// Each byte in s must be either 0 (false) or -1 (true). No other values are allowed.
// (s is signed)
static inline Vec32uc select (Vec32cb const & s, Vec32uc const & a, Vec32uc const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec32uc if_add (Vec32cb const & f, Vec32uc const & a, Vec32uc const & b) {
return a + (Vec32uc(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
// (Note: horizontal_add_x(Vec32uc) is slightly faster)
static inline uint32_t horizontal_add (Vec32uc const & a) {
__m256i sum1 = _mm256_sad_epu8(a,_mm256_setzero_si256());
__m256i sum2 = _mm256_shuffle_epi32(sum1,2);
__m256i sum3 = _mm256_add_epi16(sum1,sum2);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum4 = _mm256_extractf128_si256(sum3,1); // bug in MS compiler VS 11
#else
__m128i sum4 = _mm256_extracti128_si256(sum3,1);
#endif
__m128i sum5 = _mm_add_epi16(_mm256_castsi256_si128(sum3),sum4);
uint8_t sum6 = (uint8_t)_mm_cvtsi128_si32(sum5); // truncate to 8 bits
return sum6; // zero extend to 32 bits
}
// Horizontal add extended: Calculates the sum of all vector elements.
// Each element is zero-extended before addition to avoid overflow
static inline uint32_t horizontal_add_x (Vec32uc const & a) {
__m256i sum1 = _mm256_sad_epu8(a,_mm256_setzero_si256());
__m256i sum2 = _mm256_shuffle_epi32(sum1,2);
__m256i sum3 = _mm256_add_epi16(sum1,sum2);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum4 = _mm256_extractf128_si256(sum3,1); // bug in MS compiler VS 11
#else
__m128i sum4 = _mm256_extracti128_si256(sum3,1);
#endif
__m128i sum5 = _mm_add_epi16(_mm256_castsi256_si128(sum3),sum4);
return _mm_cvtsi128_si32(sum5);
}
// function add_saturated: add element by element, unsigned with saturation
static inline Vec32uc add_saturated(Vec32uc const & a, Vec32uc const & b) {
return _mm256_adds_epu8(a, b);
}
// function sub_saturated: subtract element by element, unsigned with saturation
static inline Vec32uc sub_saturated(Vec32uc const & a, Vec32uc const & b) {
return _mm256_subs_epu8(a, b);
}
// function max: a > b ? a : b
static inline Vec32uc max(Vec32uc const & a, Vec32uc const & b) {
return _mm256_max_epu8(a,b);
}
// function min: a < b ? a : b
static inline Vec32uc min(Vec32uc const & a, Vec32uc const & b) {
return _mm256_min_epu8(a,b);
}
/*****************************************************************************
*
* Vector of 16 16-bit signed integers
*
*****************************************************************************/
class Vec16s : public Vec256b {
public:
// Default constructor:
Vec16s() {
}
// Constructor to broadcast the same value into all elements:
Vec16s(int i) {
ymm = _mm256_set1_epi16((int16_t)i);
}
// Constructor to build from all elements:
Vec16s(int16_t i0, int16_t i1, int16_t i2, int16_t i3, int16_t i4, int16_t i5, int16_t i6, int16_t i7,
int16_t i8, int16_t i9, int16_t i10, int16_t i11, int16_t i12, int16_t i13, int16_t i14, int16_t i15) {
ymm = _mm256_setr_epi16(i0, i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, i14, i15 );
}
// Constructor to build from two Vec8s:
Vec16s(Vec8s const & a0, Vec8s const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec16s(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec16s & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Type cast operator to convert to __m256i used in intrinsics
operator __m256i() const {
return ymm;
}
// Member function to load from array (unaligned)
Vec16s & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec16s & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Partial load. Load n elements and set the rest to 0
Vec16s & load_partial(int n, void const * p) {
if (n <= 0) {
*this = 0;
}
else if (n <= 8) {
*this = Vec16s(Vec8s().load_partial(n, p), 0);
}
else if (n < 16) {
*this = Vec16s(Vec8s().load(p), Vec8s().load_partial(n-8, (int16_t const*)p+8));
}
else {
load(p);
}
return *this;
}
// Partial store. Store n elements
void store_partial(int n, void * p) const {
if (n <= 0) {
return;
}
else if (n <= 8) {
get_low().store_partial(n, p);
}
else if (n < 16) {
get_low().store(p);
get_high().store_partial(n-8, (int16_t*)p+8);
}
else {
store(p);
}
}
// cut off vector to n elements. The last 16-n elements are set to zero
Vec16s & cutoff(int n) {
*this = Vec32c(*this).cutoff(n * 2);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec16s const & insert(uint32_t index, int16_t value) {
static const int16_t m[32] = {0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0, -1,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0};
__m256i mask = Vec256b().load(m + 16 - (index & 0x0F));
__m256i broad = _mm256_set1_epi16(value);
ymm = selectb(mask, broad, ymm);
return *this;
}
// Member function extract a single element from vector
int16_t extract(uint32_t index) const {
int16_t x[16];
store(x);
return x[index & 0x0F];
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
int16_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec8s:
Vec8s get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec8s get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
static int size() {
return 16;
}
};
/*****************************************************************************
*
* Vec16sb: Vector of 16 Booleans for use with Vec16s and Vec16us
*
*****************************************************************************/
class Vec16sb : public Vec16s {
public:
// Default constructor:
Vec16sb() {
}
// Constructor to build from all elements:
Vec16sb(bool x0, bool x1, bool x2, bool x3, bool x4, bool x5, bool x6, bool x7,
bool x8, bool x9, bool x10, bool x11, bool x12, bool x13, bool x14, bool x15) :
Vec16s(-int16_t(x0), -int16_t(x1), -int16_t(x2), -int16_t(x3), -int16_t(x4), -int16_t(x5), -int16_t(x6), -int16_t(x7),
-int16_t(x8), -int16_t(x9), -int16_t(x10), -int16_t(x11), -int16_t(x12), -int16_t(x13), -int16_t(x14), -int16_t(x15))
{}
// Constructor to convert from type __m256i used in intrinsics:
Vec16sb(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec16sb & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Constructor to broadcast scalar value:
Vec16sb(bool b) : Vec16s(-int16_t(b)) {
}
// Assignment operator to broadcast scalar value:
Vec16sb & operator = (bool b) {
*this = Vec16sb(b);
return *this;
}
private: // Prevent constructing from int, etc.
Vec16sb(int b);
Vec16sb & operator = (int x);
public:
Vec8sb get_low() const {
return Vec8sb(Vec16s::get_low());
}
Vec8sb get_high() const {
return Vec8sb(Vec16s::get_high());
}
Vec16sb & insert (int index, bool a) {
Vec16s::insert(index, -(int)a);
return *this;
}
// Member function extract a single element from vector
bool extract(uint32_t index) const {
return Vec16s::extract(index) != 0;
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
bool operator [] (uint32_t index) const {
return extract(index);
}
};
/*****************************************************************************
*
* Define operators for Vec16sb
*
*****************************************************************************/
// vector operator & : bitwise and
static inline Vec16sb operator & (Vec16sb const & a, Vec16sb const & b) {
return Vec16sb(Vec256b(a) & Vec256b(b));
}
static inline Vec16sb operator && (Vec16sb const & a, Vec16sb const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec16sb & operator &= (Vec16sb & a, Vec16sb const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec16sb operator | (Vec16sb const & a, Vec16sb const & b) {
return Vec16sb(Vec256b(a) | Vec256b(b));
}
static inline Vec16sb operator || (Vec16sb const & a, Vec16sb const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec16sb & operator |= (Vec16sb & a, Vec16sb const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec16sb operator ^ (Vec16sb const & a, Vec16sb const & b) {
return Vec16sb(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec16sb & operator ^= (Vec16sb & a, Vec16sb const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec16sb operator ~ (Vec16sb const & a) {
return Vec16sb( ~ Vec256b(a));
}
// vector operator ! : element not
static inline Vec16sb operator ! (Vec16sb const & a) {
return ~ a;
}
// vector function andnot
static inline Vec16sb andnot (Vec16sb const & a, Vec16sb const & b) {
return Vec16sb(andnot(Vec256b(a), Vec256b(b)));
}
/*****************************************************************************
*
* Operators for Vec16s
*
*****************************************************************************/
// vector operator + : add element by element
static inline Vec16s operator + (Vec16s const & a, Vec16s const & b) {
return _mm256_add_epi16(a, b);
}
// vector operator += : add
static inline Vec16s & operator += (Vec16s & a, Vec16s const & b) {
a = a + b;
return a;
}
// postfix operator ++
static inline Vec16s operator ++ (Vec16s & a, int) {
Vec16s a0 = a;
a = a + 1;
return a0;
}
// prefix operator ++
static inline Vec16s & operator ++ (Vec16s & a) {
a = a + 1;
return a;
}
// vector operator - : subtract element by element
static inline Vec16s operator - (Vec16s const & a, Vec16s const & b) {
return _mm256_sub_epi16(a, b);
}
// vector operator - : unary minus
static inline Vec16s operator - (Vec16s const & a) {
return _mm256_sub_epi16(_mm256_setzero_si256(), a);
}
// vector operator -= : subtract
static inline Vec16s & operator -= (Vec16s & a, Vec16s const & b) {
a = a - b;
return a;
}
// postfix operator --
static inline Vec16s operator -- (Vec16s & a, int) {
Vec16s a0 = a;
a = a - 1;
return a0;
}
// prefix operator --
static inline Vec16s & operator -- (Vec16s & a) {
a = a - 1;
return a;
}
// vector operator * : multiply element by element
static inline Vec16s operator * (Vec16s const & a, Vec16s const & b) {
return _mm256_mullo_epi16(a, b);
}
// vector operator *= : multiply
static inline Vec16s & operator *= (Vec16s & a, Vec16s const & b) {
a = a * b;
return a;
}
// vector operator / : divide all elements by same integer
// See bottom of file
// vector operator << : shift left
static inline Vec16s operator << (Vec16s const & a, int b) {
return _mm256_sll_epi16(a,_mm_cvtsi32_si128(b));
}
// vector operator <<= : shift left
static inline Vec16s & operator <<= (Vec16s & a, int b) {
a = a << b;
return a;
}
// vector operator >> : shift right arithmetic
static inline Vec16s operator >> (Vec16s const & a, int b) {
return _mm256_sra_epi16(a,_mm_cvtsi32_si128(b));
}
// vector operator >>= : shift right arithmetic
static inline Vec16s & operator >>= (Vec16s & a, int b) {
a = a >> b;
return a;
}
// vector operator == : returns true for elements for which a == b
static inline Vec16sb operator == (Vec16s const & a, Vec16s const & b) {
return _mm256_cmpeq_epi16(a, b);
}
// vector operator != : returns true for elements for which a != b
static inline Vec16sb operator != (Vec16s const & a, Vec16s const & b) {
return Vec16sb(Vec16s(~(a == b)));
}
// vector operator > : returns true for elements for which a > b
static inline Vec16sb operator > (Vec16s const & a, Vec16s const & b) {
return _mm256_cmpgt_epi16(a, b);
}
// vector operator < : returns true for elements for which a < b
static inline Vec16sb operator < (Vec16s const & a, Vec16s const & b) {
return b > a;
}
// vector operator >= : returns true for elements for which a >= b (signed)
static inline Vec16sb operator >= (Vec16s const & a, Vec16s const & b) {
return Vec16sb(Vec16s(~(b > a)));
}
// vector operator <= : returns true for elements for which a <= b (signed)
static inline Vec16sb operator <= (Vec16s const & a, Vec16s const & b) {
return b >= a;
}
// vector operator & : bitwise and
static inline Vec16s operator & (Vec16s const & a, Vec16s const & b) {
return Vec16s(Vec256b(a) & Vec256b(b));
}
static inline Vec16s operator && (Vec16s const & a, Vec16s const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec16s & operator &= (Vec16s & a, Vec16s const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec16s operator | (Vec16s const & a, Vec16s const & b) {
return Vec16s(Vec256b(a) | Vec256b(b));
}
static inline Vec16s operator || (Vec16s const & a, Vec16s const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec16s & operator |= (Vec16s & a, Vec16s const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec16s operator ^ (Vec16s const & a, Vec16s const & b) {
return Vec16s(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec16s & operator ^= (Vec16s & a, Vec16s const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec16s operator ~ (Vec16s const & a) {
return Vec16s( ~ Vec256b(a));
}
// vector operator ! : logical not, returns true for elements == 0
static inline Vec16sb operator ! (Vec16s const & a) {
return _mm256_cmpeq_epi16(a,_mm256_setzero_si256());
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 16; i++) result[i] = s[i] ? a[i] : b[i];
// Each byte in s must be either 0 (false) or -1 (true). No other values are allowed.
// (s is signed)
static inline Vec16s select (Vec16sb const & s, Vec16s const & a, Vec16s const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec16s if_add (Vec16sb const & f, Vec16s const & a, Vec16s const & b) {
return a + (Vec16s(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
static inline int32_t horizontal_add (Vec16s const & a) {
__m256i sum1 = _mm256_hadd_epi16(a,a); // horizontally add 2x8 elements in 3 steps
__m256i sum2 = _mm256_hadd_epi16(sum1,sum1);
__m256i sum3 = _mm256_hadd_epi16(sum2,sum2);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum4 = _mm256_extractf128_si256(sum3,1); // bug in MS compiler VS 11
#else
__m128i sum4 = _mm256_extracti128_si256(sum3,1); // get high part
#endif
__m128i sum5 = _mm_add_epi16(_mm256_castsi256_si128(sum3),sum4); // add low and high parts
int16_t sum6 = (int16_t)_mm_cvtsi128_si32(sum5); // truncate to 16 bits
return sum6; // sign extend to 32 bits
}
// Horizontal add extended: Calculates the sum of all vector elements.
// Elements are sign extended before adding to avoid overflow
static inline int32_t horizontal_add_x (Vec16s const & a) {
__m256i aeven = _mm256_slli_epi32(a,16); // even numbered elements of a. get sign bit in position
aeven = _mm256_srai_epi32(aeven,16); // sign extend even numbered elements
__m256i aodd = _mm256_srai_epi32(a,16); // sign extend odd numbered elements
__m256i sum1 = _mm256_add_epi32(aeven,aodd); // add even and odd elements
__m256i sum2 = _mm256_hadd_epi32(sum1,sum1); // horizontally add 2x4 elements in 2 steps
__m256i sum3 = _mm256_hadd_epi32(sum2,sum2);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum4 = _mm256_extractf128_si256(sum3,1); // bug in MS compiler VS 11
#else
__m128i sum4 = _mm256_extracti128_si256(sum3,1);
#endif
__m128i sum5 = _mm_add_epi32(_mm256_castsi256_si128(sum3),sum4);
return _mm_cvtsi128_si32(sum5);
}
// function add_saturated: add element by element, signed with saturation
static inline Vec16s add_saturated(Vec16s const & a, Vec16s const & b) {
return _mm256_adds_epi16(a, b);
}
// function sub_saturated: subtract element by element, signed with saturation
static inline Vec16s sub_saturated(Vec16s const & a, Vec16s const & b) {
return _mm256_subs_epi16(a, b);
}
// function max: a > b ? a : b
static inline Vec16s max(Vec16s const & a, Vec16s const & b) {
return _mm256_max_epi16(a,b);
}
// function min: a < b ? a : b
static inline Vec16s min(Vec16s const & a, Vec16s const & b) {
return _mm256_min_epi16(a,b);
}
// function abs: a >= 0 ? a : -a
static inline Vec16s abs(Vec16s const & a) {
return _mm256_sign_epi16(a,a);
}
// function abs_saturated: same as abs, saturate if overflow
static inline Vec16s abs_saturated(Vec16s const & a) {
__m256i absa = abs(a); // abs(a)
__m256i overfl = _mm256_srai_epi16(absa,15); // sign
return _mm256_add_epi16(absa,overfl); // subtract 1 if 0x8000
}
// function rotate_left all elements
// Use negative count to rotate right
static inline Vec16s rotate_left(Vec16s const & a, int b) {
__m256i left = _mm256_sll_epi16(a,_mm_cvtsi32_si128(b & 0x0F)); // a << b
__m256i right = _mm256_srl_epi16(a,_mm_cvtsi32_si128((16-b) & 0x0F)); // a >> (16 - b)
__m256i rot = _mm256_or_si256(left,right); // or
return rot;
}
/*****************************************************************************
*
* Vector of 16 16-bit unsigned integers
*
*****************************************************************************/
class Vec16us : public Vec16s {
public:
// Default constructor:
Vec16us(){
}
// Constructor to broadcast the same value into all elements:
Vec16us(uint32_t i) {
ymm = _mm256_set1_epi16((int16_t)i);
}
// Constructor to build from all elements:
Vec16us(uint16_t i0, uint16_t i1, uint16_t i2, uint16_t i3, uint16_t i4, uint16_t i5, uint16_t i6, uint16_t i7,
uint16_t i8, uint16_t i9, uint16_t i10, uint16_t i11, uint16_t i12, uint16_t i13, uint16_t i14, uint16_t i15) {
ymm = _mm256_setr_epi16(i0, i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, i14, i15 );
}
// Constructor to build from two Vec8us:
Vec16us(Vec8us const & a0, Vec8us const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec16us(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec16us & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Member function to load from array (unaligned)
Vec16us & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec16us & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec16us const & insert(uint32_t index, uint16_t value) {
Vec16s::insert(index, value);
return *this;
}
// Member function extract a single element from vector
uint16_t extract(uint32_t index) const {
return Vec16s::extract(index);
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
uint16_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec8us:
Vec8us get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec8us get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
};
// Define operators for this class
// vector operator + : add
static inline Vec16us operator + (Vec16us const & a, Vec16us const & b) {
return Vec16us (Vec16s(a) + Vec16s(b));
}
// vector operator - : subtract
static inline Vec16us operator - (Vec16us const & a, Vec16us const & b) {
return Vec16us (Vec16s(a) - Vec16s(b));
}
// vector operator * : multiply
static inline Vec16us operator * (Vec16us const & a, Vec16us const & b) {
return Vec16us (Vec16s(a) * Vec16s(b));
}
// vector operator / : divide
// See bottom of file
// vector operator >> : shift right logical all elements
static inline Vec16us operator >> (Vec16us const & a, uint32_t b) {
return _mm256_srl_epi16(a,_mm_cvtsi32_si128(b));
}
// vector operator >> : shift right logical all elements
static inline Vec16us operator >> (Vec16us const & a, int32_t b) {
return a >> (uint32_t)b;
}
// vector operator >>= : shift right artihmetic
static inline Vec16us & operator >>= (Vec16us & a, uint32_t b) {
a = a >> b;
return a;
}
// vector operator << : shift left all elements
static inline Vec16us operator << (Vec16us const & a, uint32_t b) {
return _mm256_sll_epi16(a,_mm_cvtsi32_si128(b));
}
// vector operator << : shift left all elements
static inline Vec16us operator << (Vec16us const & a, int32_t b) {
return a << (uint32_t)b;
}
// vector operator >= : returns true for elements for which a >= b (unsigned)
static inline Vec16sb operator >= (Vec16us const & a, Vec16us const & b) {
__m256i max_ab = _mm256_max_epu16(a,b); // max(a,b), unsigned
return _mm256_cmpeq_epi16(a,max_ab); // a == max(a,b)
}
// vector operator <= : returns true for elements for which a <= b (unsigned)
static inline Vec16sb operator <= (Vec16us const & a, Vec16us const & b) {
return b >= a;
}
// vector operator > : returns true for elements for which a > b (unsigned)
static inline Vec16sb operator > (Vec16us const & a, Vec16us const & b) {
return Vec16sb(Vec16s(~(b >= a)));
}
// vector operator < : returns true for elements for which a < b (unsigned)
static inline Vec16sb operator < (Vec16us const & a, Vec16us const & b) {
return b > a;
}
// vector operator & : bitwise and
static inline Vec16us operator & (Vec16us const & a, Vec16us const & b) {
return Vec16us(Vec256b(a) & Vec256b(b));
}
static inline Vec16us operator && (Vec16us const & a, Vec16us const & b) {
return a & b;
}
// vector operator | : bitwise or
static inline Vec16us operator | (Vec16us const & a, Vec16us const & b) {
return Vec16us(Vec256b(a) | Vec256b(b));
}
static inline Vec16us operator || (Vec16us const & a, Vec16us const & b) {
return a | b;
}
// vector operator ^ : bitwise xor
static inline Vec16us operator ^ (Vec16us const & a, Vec16us const & b) {
return Vec16us(Vec256b(a) ^ Vec256b(b));
}
// vector operator ~ : bitwise not
static inline Vec16us operator ~ (Vec16us const & a) {
return Vec16us( ~ Vec256b(a));
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 8; i++) result[i] = s[i] ? a[i] : b[i];
// Each word in s must be either 0 (false) or -1 (true). No other values are allowed.
// (s is signed)
static inline Vec16us select (Vec16sb const & s, Vec16us const & a, Vec16us const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec16us if_add (Vec16sb const & f, Vec16us const & a, Vec16us const & b) {
return a + (Vec16us(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
static inline uint32_t horizontal_add (Vec16us const & a) {
__m256i sum1 = _mm256_hadd_epi16(a,a); // horizontally add 2x8 elements in 3 steps
__m256i sum2 = _mm256_hadd_epi16(sum1,sum1);
__m256i sum3 = _mm256_hadd_epi16(sum2,sum2);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum4 = _mm256_extractf128_si256(sum3,1); // bug in MS compiler VS 11
#else
__m128i sum4 = _mm256_extracti128_si256(sum3,1); // get high part
#endif
__m128i sum5 = _mm_add_epi32(_mm256_castsi256_si128(sum3),sum4); // add low and high parts
return _mm_cvtsi128_si32(sum5);
}
// Horizontal add extended: Calculates the sum of all vector elements.
// Each element is zero-extended before addition to avoid overflow
static inline uint32_t horizontal_add_x (Vec16us const & a) {
__m256i mask = _mm256_set1_epi32(0x0000FFFF); // mask for even positions
__m256i aeven = _mm256_and_si256(a,mask); // even numbered elements of a
__m256i aodd = _mm256_srli_epi32(a,16); // zero extend odd numbered elements
__m256i sum1 = _mm256_add_epi32(aeven,aodd); // add even and odd elements
__m256i sum2 = _mm256_hadd_epi32(sum1,sum1); // horizontally add 2x4 elements in 2 steps
__m256i sum3 = _mm256_hadd_epi32(sum2,sum2);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum4 = _mm256_extractf128_si256(sum3,1); // bug in MS compiler VS 11
#else
__m128i sum4 = _mm256_extracti128_si256(sum3,1); // get high part
#endif
__m128i sum5 = _mm_add_epi32(_mm256_castsi256_si128(sum3),sum4); // add low and high parts
return _mm_cvtsi128_si32(sum5);
}
// function add_saturated: add element by element, unsigned with saturation
static inline Vec16us add_saturated(Vec16us const & a, Vec16us const & b) {
return _mm256_adds_epu16(a, b);
}
// function sub_saturated: subtract element by element, unsigned with saturation
static inline Vec16us sub_saturated(Vec16us const & a, Vec16us const & b) {
return _mm256_subs_epu16(a, b);
}
// function max: a > b ? a : b
static inline Vec16us max(Vec16us const & a, Vec16us const & b) {
return _mm256_max_epu16(a,b);
}
// function min: a < b ? a : b
static inline Vec16us min(Vec16us const & a, Vec16us const & b) {
return _mm256_min_epu16(a,b);
}
/*****************************************************************************
*
* Vector of 8 32-bit signed integers
*
*****************************************************************************/
class Vec8i : public Vec256b {
public:
// Default constructor:
Vec8i() {
}
// Constructor to broadcast the same value into all elements:
Vec8i(int i) {
ymm = _mm256_set1_epi32(i);
}
// Constructor to build from all elements:
Vec8i(int32_t i0, int32_t i1, int32_t i2, int32_t i3, int32_t i4, int32_t i5, int32_t i6, int32_t i7) {
ymm = _mm256_setr_epi32(i0, i1, i2, i3, i4, i5, i6, i7);
}
// Constructor to build from two Vec4i:
Vec8i(Vec4i const & a0, Vec4i const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec8i(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec8i & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Type cast operator to convert to __m256i used in intrinsics
operator __m256i() const {
return ymm;
}
// Member function to load from array (unaligned)
Vec8i & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec8i & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Partial load. Load n elements and set the rest to 0
Vec8i & load_partial(int n, void const * p) {
if (n <= 0) {
*this = 0;
}
else if (n <= 4) {
*this = Vec8i(Vec4i().load_partial(n, p), 0);
}
else if (n < 8) {
*this = Vec8i(Vec4i().load(p), Vec4i().load_partial(n-4, (int32_t const*)p+4));
}
else {
load(p);
}
return *this;
}
// Partial store. Store n elements
void store_partial(int n, void * p) const {
if (n <= 0) {
return;
}
else if (n <= 4) {
get_low().store_partial(n, p);
}
else if (n < 8) {
get_low().store(p);
get_high().store_partial(n-4, (int32_t*)p+4);
}
else {
store(p);
}
}
// cut off vector to n elements. The last 8-n elements are set to zero
Vec8i & cutoff(int n) {
*this = Vec32c(*this).cutoff(n * 4);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec8i const & insert(uint32_t index, int32_t value) {
static const int32_t maskl[16] = {0,0,0,0,0,0,0,0, -1,0,0,0,0,0,0,0};
__m256i broad = _mm256_set1_epi32(value); // broadcast value into all elements
__m256i mask = Vec256b().load(maskl + 8 - (index & 7)); // mask with FFFFFFFF at index position
ymm = selectb (mask, broad, ymm);
return *this;
}
// Member function extract a single element from vector
int32_t extract(uint32_t index) const {
int32_t x[8];
store(x);
return x[index & 7];
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
int32_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec4i:
Vec4i get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec4i get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
static int size() {
return 8;
}
};
/*****************************************************************************
*
* Vec8ib: Vector of 8 Booleans for use with Vec8i and Vec8ui
*
*****************************************************************************/
class Vec8ib : public Vec8i {
public:
// Default constructor:
Vec8ib() {
}
// Constructor to build from all elements:
Vec8ib(bool x0, bool x1, bool x2, bool x3, bool x4, bool x5, bool x6, bool x7) :
Vec8i(-int32_t(x0), -int32_t(x1), -int32_t(x2), -int32_t(x3), -int32_t(x4), -int32_t(x5), -int32_t(x6), -int32_t(x7))
{}
// Constructor to convert from type __m256i used in intrinsics:
Vec8ib(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec8ib & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Constructor to broadcast scalar value:
Vec8ib(bool b) : Vec8i(-int32_t(b)) {
}
// Assignment operator to broadcast scalar value:
Vec8ib & operator = (bool b) {
*this = Vec8ib(b);
return *this;
}
private: // Prevent constructing from int, etc.
Vec8ib(int b);
Vec8ib & operator = (int x);
public:
Vec4ib get_low() const {
return Vec4ib(Vec8i::get_low());
}
Vec4ib get_high() const {
return Vec4ib(Vec8i::get_high());
}
Vec8ib & insert (int index, bool a) {
Vec8i::insert(index, -(int)a);
return *this;
}
// Member function extract a single element from vector
bool extract(uint32_t index) const {
return Vec8i::extract(index) != 0;
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
bool operator [] (uint32_t index) const {
return extract(index);
}
};
/*****************************************************************************
*
* Define operators for Vec8ib
*
*****************************************************************************/
// vector operator & : bitwise and
static inline Vec8ib operator & (Vec8ib const & a, Vec8ib const & b) {
return Vec8ib(Vec256b(a) & Vec256b(b));
}
static inline Vec8ib operator && (Vec8ib const & a, Vec8ib const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec8ib & operator &= (Vec8ib & a, Vec8ib const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec8ib operator | (Vec8ib const & a, Vec8ib const & b) {
return Vec8ib(Vec256b(a) | Vec256b(b));
}
static inline Vec8ib operator || (Vec8ib const & a, Vec8ib const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec8ib & operator |= (Vec8ib & a, Vec8ib const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec8ib operator ^ (Vec8ib const & a, Vec8ib const & b) {
return Vec8ib(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec8ib & operator ^= (Vec8ib & a, Vec8ib const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec8ib operator ~ (Vec8ib const & a) {
return Vec8ib( ~ Vec256b(a));
}
// vector operator ! : element not
static inline Vec8ib operator ! (Vec8ib const & a) {
return ~ a;
}
// vector function andnot
static inline Vec8ib andnot (Vec8ib const & a, Vec8ib const & b) {
return Vec8ib(andnot(Vec256b(a), Vec256b(b)));
}
/*****************************************************************************
*
* Operators for Vec8i
*
*****************************************************************************/
// vector operator + : add element by element
static inline Vec8i operator + (Vec8i const & a, Vec8i const & b) {
return _mm256_add_epi32(a, b);
}
// vector operator += : add
static inline Vec8i & operator += (Vec8i & a, Vec8i const & b) {
a = a + b;
return a;
}
// postfix operator ++
static inline Vec8i operator ++ (Vec8i & a, int) {
Vec8i a0 = a;
a = a + 1;
return a0;
}
// prefix operator ++
static inline Vec8i & operator ++ (Vec8i & a) {
a = a + 1;
return a;
}
// vector operator - : subtract element by element
static inline Vec8i operator - (Vec8i const & a, Vec8i const & b) {
return _mm256_sub_epi32(a, b);
}
// vector operator - : unary minus
static inline Vec8i operator - (Vec8i const & a) {
return _mm256_sub_epi32(_mm256_setzero_si256(), a);
}
// vector operator -= : subtract
static inline Vec8i & operator -= (Vec8i & a, Vec8i const & b) {
a = a - b;
return a;
}
// postfix operator --
static inline Vec8i operator -- (Vec8i & a, int) {
Vec8i a0 = a;
a = a - 1;
return a0;
}
// prefix operator --
static inline Vec8i & operator -- (Vec8i & a) {
a = a - 1;
return a;
}
// vector operator * : multiply element by element
static inline Vec8i operator * (Vec8i const & a, Vec8i const & b) {
return _mm256_mullo_epi32(a, b);
}
// vector operator *= : multiply
static inline Vec8i & operator *= (Vec8i & a, Vec8i const & b) {
a = a * b;
return a;
}
// vector operator / : divide all elements by same integer
// See bottom of file
// vector operator << : shift left
static inline Vec8i operator << (Vec8i const & a, int32_t b) {
return _mm256_sll_epi32(a, _mm_cvtsi32_si128(b));
}
// vector operator <<= : shift left
static inline Vec8i & operator <<= (Vec8i & a, int32_t b) {
a = a << b;
return a;
}
// vector operator >> : shift right arithmetic
static inline Vec8i operator >> (Vec8i const & a, int32_t b) {
return _mm256_sra_epi32(a, _mm_cvtsi32_si128(b));
}
// vector operator >>= : shift right arithmetic
static inline Vec8i & operator >>= (Vec8i & a, int32_t b) {
a = a >> b;
return a;
}
// vector operator == : returns true for elements for which a == b
static inline Vec8ib operator == (Vec8i const & a, Vec8i const & b) {
return _mm256_cmpeq_epi32(a, b);
}
// vector operator != : returns true for elements for which a != b
static inline Vec8ib operator != (Vec8i const & a, Vec8i const & b) {
return Vec8ib(Vec8i(~(a == b)));
}
// vector operator > : returns true for elements for which a > b
static inline Vec8ib operator > (Vec8i const & a, Vec8i const & b) {
return _mm256_cmpgt_epi32(a, b);
}
// vector operator < : returns true for elements for which a < b
static inline Vec8ib operator < (Vec8i const & a, Vec8i const & b) {
return b > a;
}
// vector operator >= : returns true for elements for which a >= b (signed)
static inline Vec8ib operator >= (Vec8i const & a, Vec8i const & b) {
return Vec8ib(Vec8i(~(b > a)));
}
// vector operator <= : returns true for elements for which a <= b (signed)
static inline Vec8ib operator <= (Vec8i const & a, Vec8i const & b) {
return b >= a;
}
// vector operator & : bitwise and
static inline Vec8i operator & (Vec8i const & a, Vec8i const & b) {
return Vec8i(Vec256b(a) & Vec256b(b));
}
static inline Vec8i operator && (Vec8i const & a, Vec8i const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec8i & operator &= (Vec8i & a, Vec8i const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec8i operator | (Vec8i const & a, Vec8i const & b) {
return Vec8i(Vec256b(a) | Vec256b(b));
}
static inline Vec8i operator || (Vec8i const & a, Vec8i const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec8i & operator |= (Vec8i & a, Vec8i const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec8i operator ^ (Vec8i const & a, Vec8i const & b) {
return Vec8i(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec8i & operator ^= (Vec8i & a, Vec8i const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec8i operator ~ (Vec8i const & a) {
return Vec8i( ~ Vec256b(a));
}
// vector operator ! : returns true for elements == 0
static inline Vec8ib operator ! (Vec8i const & a) {
return _mm256_cmpeq_epi32(a, _mm256_setzero_si256());
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 8; i++) result[i] = s[i] ? a[i] : b[i];
// Each byte in s must be either 0 (false) or -1 (true). No other values are allowed.
// (s is signed)
static inline Vec8i select (Vec8ib const & s, Vec8i const & a, Vec8i const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec8i if_add (Vec8ib const & f, Vec8i const & a, Vec8i const & b) {
return a + (Vec8i(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
static inline int32_t horizontal_add (Vec8i const & a) {
__m256i sum1 = _mm256_hadd_epi32(a,a); // horizontally add 2x4 elements in 2 steps
__m256i sum2 = _mm256_hadd_epi32(sum1,sum1);
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum3 = _mm256_extractf128_si256(sum2,1); // bug in MS VS 11
#else
__m128i sum3 = _mm256_extracti128_si256(sum2,1); // get high part
#endif
__m128i sum4 = _mm_add_epi32(_mm256_castsi256_si128(sum2),sum3); // add low and high parts
return _mm_cvtsi128_si32(sum4);
}
// Horizontal add extended: Calculates the sum of all vector elements.
// Elements are sign extended before adding to avoid overflow
// static inline int64_t horizontal_add_x (Vec8i const & a); // defined below
// function add_saturated: add element by element, signed with saturation
static inline Vec8i add_saturated(Vec8i const & a, Vec8i const & b) {
__m256i sum = _mm256_add_epi32(a, b); // a + b
__m256i axb = _mm256_xor_si256(a, b); // check if a and b have different sign
__m256i axs = _mm256_xor_si256(a, sum); // check if a and sum have different sign
__m256i overf1 = _mm256_andnot_si256(axb,axs); // check if sum has wrong sign
__m256i overf2 = _mm256_srai_epi32(overf1,31); // -1 if overflow
__m256i asign = _mm256_srli_epi32(a,31); // 1 if a < 0
__m256i sat1 = _mm256_srli_epi32(overf2,1); // 7FFFFFFF if overflow
__m256i sat2 = _mm256_add_epi32(sat1,asign); // 7FFFFFFF if positive overflow 80000000 if negative overflow
return selectb(overf2,sat2,sum); // sum if not overflow, else sat2
}
// function sub_saturated: subtract element by element, signed with saturation
static inline Vec8i sub_saturated(Vec8i const & a, Vec8i const & b) {
__m256i diff = _mm256_sub_epi32(a, b); // a + b
__m256i axb = _mm256_xor_si256(a, b); // check if a and b have different sign
__m256i axs = _mm256_xor_si256(a, diff); // check if a and sum have different sign
__m256i overf1 = _mm256_and_si256(axb,axs); // check if sum has wrong sign
__m256i overf2 = _mm256_srai_epi32(overf1,31); // -1 if overflow
__m256i asign = _mm256_srli_epi32(a,31); // 1 if a < 0
__m256i sat1 = _mm256_srli_epi32(overf2,1); // 7FFFFFFF if overflow
__m256i sat2 = _mm256_add_epi32(sat1,asign); // 7FFFFFFF if positive overflow 80000000 if negative overflow
return selectb(overf2,sat2,diff); // diff if not overflow, else sat2
}
// function max: a > b ? a : b
static inline Vec8i max(Vec8i const & a, Vec8i const & b) {
return _mm256_max_epi32(a,b);
}
// function min: a < b ? a : b
static inline Vec8i min(Vec8i const & a, Vec8i const & b) {
return _mm256_min_epi32(a,b);
}
// function abs: a >= 0 ? a : -a
static inline Vec8i abs(Vec8i const & a) {
return _mm256_sign_epi32(a,a);
}
// function abs_saturated: same as abs, saturate if overflow
static inline Vec8i abs_saturated(Vec8i const & a) {
__m256i absa = abs(a); // abs(a)
__m256i overfl = _mm256_srai_epi32(absa,31); // sign
return _mm256_add_epi32(absa,overfl); // subtract 1 if 0x80000000
}
// function rotate_left all elements
// Use negative count to rotate right
static inline Vec8i rotate_left(Vec8i const & a, int b) {
#ifdef __AVX512VL__
return _mm256_rolv_epi32(a, _mm256_set1_epi32(b));
#else
__m256i left = _mm256_sll_epi32(a,_mm_cvtsi32_si128(b & 0x1F)); // a << b
__m256i right = _mm256_srl_epi32(a,_mm_cvtsi32_si128((32-b) & 0x1F)); // a >> (32 - b)
__m256i rot = _mm256_or_si256(left,right); // or
return rot;
#endif
}
/*****************************************************************************
*
* Vector of 8 32-bit unsigned integers
*
*****************************************************************************/
class Vec8ui : public Vec8i {
public:
// Default constructor:
Vec8ui() {
}
// Constructor to broadcast the same value into all elements:
Vec8ui(uint32_t i) {
ymm = _mm256_set1_epi32(i);
}
// Constructor to build from all elements:
Vec8ui(uint32_t i0, uint32_t i1, uint32_t i2, uint32_t i3, uint32_t i4, uint32_t i5, uint32_t i6, uint32_t i7) {
ymm = _mm256_setr_epi32(i0, i1, i2, i3, i4, i5, i6, i7);
}
// Constructor to build from two Vec4ui:
Vec8ui(Vec4ui const & a0, Vec4ui const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec8ui(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec8ui & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Member function to load from array (unaligned)
Vec8ui & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec8ui & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec8ui const & insert(uint32_t index, uint32_t value) {
Vec8i::insert(index, value);
return *this;
}
// Member function extract a single element from vector
uint32_t extract(uint32_t index) const {
return Vec8i::extract(index);
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
uint32_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec4ui:
Vec4ui get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec4ui get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
};
// Define operators for this class
// vector operator + : add
static inline Vec8ui operator + (Vec8ui const & a, Vec8ui const & b) {
return Vec8ui (Vec8i(a) + Vec8i(b));
}
// vector operator - : subtract
static inline Vec8ui operator - (Vec8ui const & a, Vec8ui const & b) {
return Vec8ui (Vec8i(a) - Vec8i(b));
}
// vector operator * : multiply
static inline Vec8ui operator * (Vec8ui const & a, Vec8ui const & b) {
return Vec8ui (Vec8i(a) * Vec8i(b));
}
// vector operator / : divide
// See bottom of file
// vector operator >> : shift right logical all elements
static inline Vec8ui operator >> (Vec8ui const & a, uint32_t b) {
return _mm256_srl_epi32(a,_mm_cvtsi32_si128(b));
}
// vector operator >> : shift right logical all elements
static inline Vec8ui operator >> (Vec8ui const & a, int32_t b) {
return a >> (uint32_t)b;
}
// vector operator >>= : shift right logical
static inline Vec8ui & operator >>= (Vec8ui & a, uint32_t b) {
a = a >> b;
return a;
}
// vector operator << : shift left all elements
static inline Vec8ui operator << (Vec8ui const & a, uint32_t b) {
return Vec8ui ((Vec8i)a << (int32_t)b);
}
// vector operator << : shift left all elements
static inline Vec8ui operator << (Vec8ui const & a, int32_t b) {
return Vec8ui ((Vec8i)a << (int32_t)b);
}
// vector operator > : returns true for elements for which a > b (unsigned)
static inline Vec8ib operator > (Vec8ui const & a, Vec8ui const & b) {
__m256i signbit = _mm256_set1_epi32(0x80000000);
__m256i a1 = _mm256_xor_si256(a,signbit);
__m256i b1 = _mm256_xor_si256(b,signbit);
return _mm256_cmpgt_epi32(a1,b1); // signed compare
}
// vector operator < : returns true for elements for which a < b (unsigned)
static inline Vec8ib operator < (Vec8ui const & a, Vec8ui const & b) {
return b > a;
}
// vector operator >= : returns true for elements for which a >= b (unsigned)
static inline Vec8ib operator >= (Vec8ui const & a, Vec8ui const & b) {
__m256i max_ab = _mm256_max_epu32(a,b); // max(a,b), unsigned
return _mm256_cmpeq_epi32(a,max_ab); // a == max(a,b)
}
// vector operator <= : returns true for elements for which a <= b (unsigned)
static inline Vec8ib operator <= (Vec8ui const & a, Vec8ui const & b) {
return b >= a;
}
// vector operator & : bitwise and
static inline Vec8ui operator & (Vec8ui const & a, Vec8ui const & b) {
return Vec8ui(Vec256b(a) & Vec256b(b));
}
static inline Vec8ui operator && (Vec8ui const & a, Vec8ui const & b) {
return a & b;
}
// vector operator | : bitwise or
static inline Vec8ui operator | (Vec8ui const & a, Vec8ui const & b) {
return Vec8ui(Vec256b(a) | Vec256b(b));
}
static inline Vec8ui operator || (Vec8ui const & a, Vec8ui const & b) {
return a | b;
}
// vector operator ^ : bitwise xor
static inline Vec8ui operator ^ (Vec8ui const & a, Vec8ui const & b) {
return Vec8ui(Vec256b(a) ^ Vec256b(b));
}
// vector operator ~ : bitwise not
static inline Vec8ui operator ~ (Vec8ui const & a) {
return Vec8ui( ~ Vec256b(a));
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 16; i++) result[i] = s[i] ? a[i] : b[i];
// Each word in s must be either 0 (false) or -1 (true). No other values are allowed.
// (s is signed)
static inline Vec8ui select (Vec8ib const & s, Vec8ui const & a, Vec8ui const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec8ui if_add (Vec8ib const & f, Vec8ui const & a, Vec8ui const & b) {
return a + (Vec8ui(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
static inline uint32_t horizontal_add (Vec8ui const & a) {
return horizontal_add((Vec8i)a);
}
// Horizontal add extended: Calculates the sum of all vector elements.
// Elements are zero extended before adding to avoid overflow
// static inline uint64_t horizontal_add_x (Vec8ui const & a); // defined later
// function add_saturated: add element by element, unsigned with saturation
static inline Vec8ui add_saturated(Vec8ui const & a, Vec8ui const & b) {
Vec8ui sum = a + b;
Vec8ui aorb = Vec8ui(a | b);
Vec8ui overflow = Vec8ui(sum < aorb); // overflow if a + b < (a | b)
return Vec8ui (sum | overflow); // return 0xFFFFFFFF if overflow
}
// function sub_saturated: subtract element by element, unsigned with saturation
static inline Vec8ui sub_saturated(Vec8ui const & a, Vec8ui const & b) {
Vec8ui diff = a - b;
Vec8ui underflow = Vec8ui(diff > a); // underflow if a - b > a
return _mm256_andnot_si256(underflow,diff); // return 0 if underflow
}
// function max: a > b ? a : b
static inline Vec8ui max(Vec8ui const & a, Vec8ui const & b) {
return _mm256_max_epu32(a,b);
}
// function min: a < b ? a : b
static inline Vec8ui min(Vec8ui const & a, Vec8ui const & b) {
return _mm256_min_epu32(a,b);
}
/*****************************************************************************
*
* Vector of 4 64-bit signed integers
*
*****************************************************************************/
class Vec4q : public Vec256b {
public:
// Default constructor:
Vec4q() {
}
// Constructor to broadcast the same value into all elements:
Vec4q(int64_t i) {
#if defined (_MSC_VER) && _MSC_VER < 1900 && ! defined (__x86_64__) && ! defined(__INTEL_COMPILER)
// MS compiler cannot use _mm256_set1_epi64x in 32 bit mode, and
// cannot put 64-bit values into xmm register without using
// mmx registers, and it makes no emms
union {
int64_t q[4];
int32_t r[8];
} u;
u.q[0] = u.q[1] = u.q[2] = u.q[3] = i;
ymm = _mm256_setr_epi32(u.r[0], u.r[1], u.r[2], u.r[3], u.r[4], u.r[5], u.r[6], u.r[7]);
#else
ymm = _mm256_set1_epi64x(i);
#endif
}
// Constructor to build from all elements:
Vec4q(int64_t i0, int64_t i1, int64_t i2, int64_t i3) {
#if defined (_MSC_VER) && _MSC_VER < 1900 && ! defined (__x86_64__) && ! defined(__INTEL_COMPILER)
// MS compiler cannot put 64-bit values into xmm register without using
// mmx registers, and it makes no emms
union {
int64_t q[4];
int32_t r[8];
} u;
u.q[0] = i0; u.q[1] = i1; u.q[2] = i2; u.q[3] = i3;
ymm = _mm256_setr_epi32(u.r[0], u.r[1], u.r[2], u.r[3], u.r[4], u.r[5], u.r[6], u.r[7]);
#else
ymm = _mm256_setr_epi64x(i0, i1, i2, i3);
#endif
}
// Constructor to build from two Vec2q:
Vec4q(Vec2q const & a0, Vec2q const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec4q(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec4q & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Type cast operator to convert to __m256i used in intrinsics
operator __m256i() const {
return ymm;
}
// Member function to load from array (unaligned)
Vec4q & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec4q & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Partial load. Load n elements and set the rest to 0
Vec4q & load_partial(int n, void const * p) {
if (n <= 0) {
*this = 0;
}
else if (n <= 2) {
*this = Vec4q(Vec2q().load_partial(n, p), 0);
}
else if (n < 4) {
*this = Vec4q(Vec2q().load(p), Vec2q().load_partial(n-2, (int64_t const*)p+2));
}
else {
load(p);
}
return *this;
}
// Partial store. Store n elements
void store_partial(int n, void * p) const {
if (n <= 0) {
return;
}
else if (n <= 2) {
get_low().store_partial(n, p);
}
else if (n < 4) {
get_low().store(p);
get_high().store_partial(n-2, (int64_t*)p+2);
}
else {
store(p);
}
}
// cut off vector to n elements. The last 8-n elements are set to zero
Vec4q & cutoff(int n) {
*this = Vec32c(*this).cutoff(n * 8);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec4q const & insert(uint32_t index, int64_t value) {
Vec4q x(value);
switch (index) {
case 0:
ymm = _mm256_blend_epi32(ymm,x,0x03); break;
case 1:
ymm = _mm256_blend_epi32(ymm,x,0x0C); break;
case 2:
ymm = _mm256_blend_epi32(ymm,x,0x30); break;
case 3:
ymm = _mm256_blend_epi32(ymm,x,0xC0); break;
}
return *this;
}
// Member function extract a single element from vector
int64_t extract(uint32_t index) const {
int64_t x[4];
store(x);
return x[index & 3];
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
int64_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec2q:
Vec2q get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec2q get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
static int size() {
return 4;
}
};
/*****************************************************************************
*
* Vec4qb: Vector of 4 Booleans for use with Vec4q and Vec4uq
*
*****************************************************************************/
class Vec4qb : public Vec4q {
public:
// Default constructor:
Vec4qb() {
}
// Constructor to build from all elements:
Vec4qb(bool x0, bool x1, bool x2, bool x3) :
Vec4q(-int64_t(x0), -int64_t(x1), -int64_t(x2), -int64_t(x3)) {
}
// Constructor to convert from type __m256i used in intrinsics:
Vec4qb(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec4qb & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Constructor to broadcast scalar value:
Vec4qb(bool b) : Vec4q(-int64_t(b)) {
}
// Assignment operator to broadcast scalar value:
Vec4qb & operator = (bool b) {
*this = Vec4qb(b);
return *this;
}
private: // Prevent constructing from int, etc.
Vec4qb(int b);
Vec4qb & operator = (int x);
public:
// Member functions to split into two Vec2qb:
Vec2qb get_low() const {
return Vec2qb(Vec4q::get_low());
}
Vec2qb get_high() const {
return Vec2qb(Vec4q::get_high());
}
Vec4qb & insert (int index, bool a) {
Vec4q::insert(index, -(int64_t)a);
return *this;
}
// Member function extract a single element from vector
bool extract(uint32_t index) const {
return Vec4q::extract(index) != 0;
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
bool operator [] (uint32_t index) const {
return extract(index);
}
};
/*****************************************************************************
*
* Define operators for Vec4qb
*
*****************************************************************************/
// vector operator & : bitwise and
static inline Vec4qb operator & (Vec4qb const & a, Vec4qb const & b) {
return Vec4qb(Vec256b(a) & Vec256b(b));
}
static inline Vec4qb operator && (Vec4qb const & a, Vec4qb const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec4qb & operator &= (Vec4qb & a, Vec4qb const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec4qb operator | (Vec4qb const & a, Vec4qb const & b) {
return Vec4qb(Vec256b(a) | Vec256b(b));
}
static inline Vec4qb operator || (Vec4qb const & a, Vec4qb const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec4qb & operator |= (Vec4qb & a, Vec4qb const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec4qb operator ^ (Vec4qb const & a, Vec4qb const & b) {
return Vec4qb(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec4qb & operator ^= (Vec4qb & a, Vec4qb const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec4qb operator ~ (Vec4qb const & a) {
return Vec4qb( ~ Vec256b(a));
}
// vector operator ! : element not
static inline Vec4qb operator ! (Vec4qb const & a) {
return ~ a;
}
// vector function andnot
static inline Vec4qb andnot (Vec4qb const & a, Vec4qb const & b) {
return Vec4qb(andnot(Vec256b(a), Vec256b(b)));
}
/*****************************************************************************
*
* Operators for Vec4q
*
*****************************************************************************/
// vector operator + : add element by element
static inline Vec4q operator + (Vec4q const & a, Vec4q const & b) {
return _mm256_add_epi64(a, b);
}
// vector operator += : add
static inline Vec4q & operator += (Vec4q & a, Vec4q const & b) {
a = a + b;
return a;
}
// postfix operator ++
static inline Vec4q operator ++ (Vec4q & a, int) {
Vec4q a0 = a;
a = a + 1;
return a0;
}
// prefix operator ++
static inline Vec4q & operator ++ (Vec4q & a) {
a = a + 1;
return a;
}
// vector operator - : subtract element by element
static inline Vec4q operator - (Vec4q const & a, Vec4q const & b) {
return _mm256_sub_epi64(a, b);
}
// vector operator - : unary minus
static inline Vec4q operator - (Vec4q const & a) {
return _mm256_sub_epi64(_mm256_setzero_si256(), a);
}
// vector operator -= : subtract
static inline Vec4q & operator -= (Vec4q & a, Vec4q const & b) {
a = a - b;
return a;
}
// postfix operator --
static inline Vec4q operator -- (Vec4q & a, int) {
Vec4q a0 = a;
a = a - 1;
return a0;
}
// prefix operator --
static inline Vec4q & operator -- (Vec4q & a) {
a = a - 1;
return a;
}
// vector operator * : multiply element by element
static inline Vec4q operator * (Vec4q const & a, Vec4q const & b) {
#if defined (__AVX512DQ__) && defined (__AVX512VL__)
return _mm256_mullo_epi64(a, b);
#else
// instruction does not exist. Split into 32-bit multiplies
__m256i bswap = _mm256_shuffle_epi32(b,0xB1); // swap H<->L
__m256i prodlh = _mm256_mullo_epi32(a,bswap); // 32 bit L*H products
__m256i zero = _mm256_setzero_si256(); // 0
__m256i prodlh2 = _mm256_hadd_epi32(prodlh,zero); // a0Lb0H+a0Hb0L,a1Lb1H+a1Hb1L,0,0
__m256i prodlh3 = _mm256_shuffle_epi32(prodlh2,0x73); // 0, a0Lb0H+a0Hb0L, 0, a1Lb1H+a1Hb1L
__m256i prodll = _mm256_mul_epu32(a,b); // a0Lb0L,a1Lb1L, 64 bit unsigned products
__m256i prod = _mm256_add_epi64(prodll,prodlh3); // a0Lb0L+(a0Lb0H+a0Hb0L)<<32, a1Lb1L+(a1Lb1H+a1Hb1L)<<32
return prod;
#endif
}
// vector operator *= : multiply
static inline Vec4q & operator *= (Vec4q & a, Vec4q const & b) {
a = a * b;
return a;
}
// vector operator << : shift left
static inline Vec4q operator << (Vec4q const & a, int32_t b) {
return _mm256_sll_epi64(a, _mm_cvtsi32_si128(b));
}
// vector operator <<= : shift left
static inline Vec4q & operator <<= (Vec4q & a, int32_t b) {
a = a << b;
return a;
}
// vector operator >> : shift right arithmetic
static inline Vec4q operator >> (Vec4q const & a, int32_t b) {
// instruction does not exist. Split into 32-bit shifts
if (b <= 32) {
__m128i bb = _mm_cvtsi32_si128(b); // b
__m256i sra = _mm256_sra_epi32(a,bb); // a >> b signed dwords
__m256i srl = _mm256_srl_epi64(a,bb); // a >> b unsigned qwords
__m256i mask = constant8i<0,-1,0,-1,0,-1,0,-1>(); // mask for signed high part
return selectb(mask, sra, srl);
}
else { // b > 32
__m128i bm32 = _mm_cvtsi32_si128(b-32); // b - 32
__m256i sign = _mm256_srai_epi32(a,31); // sign of a
__m256i sra2 = _mm256_sra_epi32(a,bm32); // a >> (b-32) signed dwords
__m256i sra3 = _mm256_srli_epi64(sra2,32); // a >> (b-32) >> 32 (second shift unsigned qword)
__m256i mask = constant8i<0,-1,0,-1,0,-1,0,-1>(); // mask for high part containing only sign
return selectb(mask, sign ,sra3);
}
}
// vector operator >>= : shift right arithmetic
static inline Vec4q & operator >>= (Vec4q & a, int32_t b) {
a = a >> b;
return a;
}
// vector operator == : returns true for elements for which a == b
static inline Vec4qb operator == (Vec4q const & a, Vec4q const & b) {
return _mm256_cmpeq_epi64(a, b);
}
// vector operator != : returns true for elements for which a != b
static inline Vec4qb operator != (Vec4q const & a, Vec4q const & b) {
return Vec4qb(Vec4q(~(a == b)));
}
// vector operator < : returns true for elements for which a < b
static inline Vec4qb operator < (Vec4q const & a, Vec4q const & b) {
return _mm256_cmpgt_epi64(b, a);
}
// vector operator > : returns true for elements for which a > b
static inline Vec4qb operator > (Vec4q const & a, Vec4q const & b) {
return b < a;
}
// vector operator >= : returns true for elements for which a >= b (signed)
static inline Vec4qb operator >= (Vec4q const & a, Vec4q const & b) {
return Vec4qb(Vec4q(~(a < b)));
}
// vector operator <= : returns true for elements for which a <= b (signed)
static inline Vec4qb operator <= (Vec4q const & a, Vec4q const & b) {
return b >= a;
}
// vector operator & : bitwise and
static inline Vec4q operator & (Vec4q const & a, Vec4q const & b) {
return Vec4q(Vec256b(a) & Vec256b(b));
}
static inline Vec4q operator && (Vec4q const & a, Vec4q const & b) {
return a & b;
}
// vector operator &= : bitwise and
static inline Vec4q & operator &= (Vec4q & a, Vec4q const & b) {
a = a & b;
return a;
}
// vector operator | : bitwise or
static inline Vec4q operator | (Vec4q const & a, Vec4q const & b) {
return Vec4q(Vec256b(a) | Vec256b(b));
}
static inline Vec4q operator || (Vec4q const & a, Vec4q const & b) {
return a | b;
}
// vector operator |= : bitwise or
static inline Vec4q & operator |= (Vec4q & a, Vec4q const & b) {
a = a | b;
return a;
}
// vector operator ^ : bitwise xor
static inline Vec4q operator ^ (Vec4q const & a, Vec4q const & b) {
return Vec4q(Vec256b(a) ^ Vec256b(b));
}
// vector operator ^= : bitwise xor
static inline Vec4q & operator ^= (Vec4q & a, Vec4q const & b) {
a = a ^ b;
return a;
}
// vector operator ~ : bitwise not
static inline Vec4q operator ~ (Vec4q const & a) {
return Vec4q( ~ Vec256b(a));
}
// vector operator ! : logical not, returns true for elements == 0
static inline Vec4qb operator ! (Vec4q const & a) {
return a == Vec4q(_mm256_setzero_si256());
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 4; i++) result[i] = s[i] ? a[i] : b[i];
// Each byte in s must be either 0 (false) or -1 (true). No other values are allowed.
// (s is signed)
static inline Vec4q select (Vec4qb const & s, Vec4q const & a, Vec4q const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec4q if_add (Vec4qb const & f, Vec4q const & a, Vec4q const & b) {
return a + (Vec4q(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
static inline int64_t horizontal_add (Vec4q const & a) {
__m256i sum1 = _mm256_shuffle_epi32(a,0x0E); // high element
__m256i sum2 = _mm256_add_epi64(a,sum1); // sum
#if defined (_MSC_VER) && _MSC_VER <= 1700 && ! defined(__INTEL_COMPILER)
__m128i sum3 = _mm256_extractf128_si256(sum2, 1); // bug in MS compiler VS 11
#else
__m128i sum3 = _mm256_extracti128_si256(sum2, 1); // get high part
#endif
__m128i sum4 = _mm_add_epi64(_mm256_castsi256_si128(sum2),sum3); // add low and high parts
#if defined(__x86_64__)
return _mm_cvtsi128_si64(sum4); // 64 bit mode
#else
union {
__m128i x; // silly definition of _mm256_storel_epi64 requires __m256i
uint64_t i;
} u;
_mm_storel_epi64(&u.x,sum4);
return u.i;
#endif
}
// function max: a > b ? a : b
static inline Vec4q max(Vec4q const & a, Vec4q const & b) {
return select(a > b, a, b);
}
// function min: a < b ? a : b
static inline Vec4q min(Vec4q const & a, Vec4q const & b) {
return select(a < b, a, b);
}
// function abs: a >= 0 ? a : -a
static inline Vec4q abs(Vec4q const & a) {
__m256i sign = _mm256_cmpgt_epi64(_mm256_setzero_si256(), a);// 0 > a
__m256i inv = _mm256_xor_si256(a, sign); // invert bits if negative
return _mm256_sub_epi64(inv, sign); // add 1
}
// function abs_saturated: same as abs, saturate if overflow
static inline Vec4q abs_saturated(Vec4q const & a) {
__m256i absa = abs(a); // abs(a)
__m256i overfl = _mm256_cmpgt_epi64(_mm256_setzero_si256(), absa); // 0 > a
return _mm256_add_epi64(absa, overfl); // subtract 1 if 0x8000000000000000
}
// function rotate_left all elements
// Use negative count to rotate right
static inline Vec4q rotate_left(Vec4q const & a, int b) {
#ifdef __AVX512VL__
return _mm256_rolv_epi64(a, _mm256_set1_epi64x(int64_t(b)));
#else
__m256i left = _mm256_sll_epi64(a,_mm_cvtsi32_si128(b & 0x3F)); // a << b
__m256i right = _mm256_srl_epi64(a,_mm_cvtsi32_si128((64-b) & 0x3F)); // a >> (64 - b)
__m256i rot = _mm256_or_si256(left, right); // or
return rot;
#endif
}
/*****************************************************************************
*
* Vector of 4 64-bit unsigned integers
*
*****************************************************************************/
class Vec4uq : public Vec4q {
public:
// Default constructor:
Vec4uq() {
}
// Constructor to broadcast the same value into all elements:
Vec4uq(uint64_t i) {
ymm = Vec4q(i);
}
// Constructor to build from all elements:
Vec4uq(uint64_t i0, uint64_t i1, uint64_t i2, uint64_t i3) {
ymm = Vec4q(i0, i1, i2, i3);
}
// Constructor to build from two Vec2uq:
Vec4uq(Vec2uq const & a0, Vec2uq const & a1) {
ymm = set_m128ir(a0, a1);
}
// Constructor to convert from type __m256i used in intrinsics:
Vec4uq(__m256i const & x) {
ymm = x;
}
// Assignment operator to convert from type __m256i used in intrinsics:
Vec4uq & operator = (__m256i const & x) {
ymm = x;
return *this;
}
// Member function to load from array (unaligned)
Vec4uq & load(void const * p) {
ymm = _mm256_loadu_si256((__m256i const*)p);
return *this;
}
// Member function to load from array, aligned by 32
Vec4uq & load_a(void const * p) {
ymm = _mm256_load_si256((__m256i const*)p);
return *this;
}
// Member function to change a single element in vector
// Note: This function is inefficient. Use load function if changing more than one element
Vec4uq const & insert(uint32_t index, uint64_t value) {
Vec4q::insert(index, value);
return *this;
}
// Member function extract a single element from vector
uint64_t extract(uint32_t index) const {
return Vec4q::extract(index);
}
// Extract a single element. Use store function if extracting more than one element.
// Operator [] can only read an element, not write.
uint64_t operator [] (uint32_t index) const {
return extract(index);
}
// Member functions to split into two Vec2uq:
Vec2uq get_low() const {
return _mm256_castsi256_si128(ymm);
}
Vec2uq get_high() const {
return _mm256_extractf128_si256(ymm,1);
}
};
// Define operators for this class
// vector operator + : add
static inline Vec4uq operator + (Vec4uq const & a, Vec4uq const & b) {
return Vec4uq (Vec4q(a) + Vec4q(b));
}
// vector operator - : subtract
static inline Vec4uq operator - (Vec4uq const & a, Vec4uq const & b) {
return Vec4uq (Vec4q(a) - Vec4q(b));
}
// vector operator * : multiply element by element
static inline Vec4uq operator * (Vec4uq const & a, Vec4uq const & b) {
return Vec4uq (Vec4q(a) * Vec4q(b));
}
// vector operator >> : shift right logical all elements
static inline Vec4uq operator >> (Vec4uq const & a, uint32_t b) {
return _mm256_srl_epi64(a,_mm_cvtsi32_si128(b));
}
// vector operator >> : shift right logical all elements
static inline Vec4uq operator >> (Vec4uq const & a, int32_t b) {
return a >> (uint32_t)b;
}
// vector operator >>= : shift right artihmetic
static inline Vec4uq & operator >>= (Vec4uq & a, uint32_t b) {
a = a >> b;
return a;
}
// vector operator << : shift left all elements
static inline Vec4uq operator << (Vec4uq const & a, uint32_t b) {
return Vec4uq ((Vec4q)a << (int32_t)b);
}
// vector operator << : shift left all elements
static inline Vec4uq operator << (Vec4uq const & a, int32_t b) {
return Vec4uq ((Vec4q)a << b);
}
// vector operator > : returns true for elements for which a > b (unsigned)
static inline Vec4qb operator > (Vec4uq const & a, Vec4uq const & b) {
//#if defined ( __XOP__ ) // AMD XOP instruction set
__m256i sign64 = Vec4uq(0x8000000000000000);
__m256i aflip = _mm256_xor_si256(a, sign64);
__m256i bflip = _mm256_xor_si256(b, sign64);
Vec4q cmp = _mm256_cmpgt_epi64(aflip,bflip);
return Vec4qb(cmp);
}
// vector operator < : returns true for elements for which a < b (unsigned)
static inline Vec4qb operator < (Vec4uq const & a, Vec4uq const & b) {
return b > a;
}
// vector operator >= : returns true for elements for which a >= b (unsigned)
static inline Vec4qb operator >= (Vec4uq const & a, Vec4uq const & b) {
return Vec4qb(Vec4q(~(b > a)));
}
// vector operator <= : returns true for elements for which a <= b (unsigned)
static inline Vec4qb operator <= (Vec4uq const & a, Vec4uq const & b) {
return b >= a;
}
// vector operator & : bitwise and
static inline Vec4uq operator & (Vec4uq const & a, Vec4uq const & b) {
return Vec4uq(Vec256b(a) & Vec256b(b));
}
static inline Vec4uq operator && (Vec4uq const & a, Vec4uq const & b) {
return a & b;
}
// vector operator | : bitwise or
static inline Vec4uq operator | (Vec4uq const & a, Vec4uq const & b) {
return Vec4uq(Vec256b(a) | Vec256b(b));
}
static inline Vec4uq operator || (Vec4uq const & a, Vec4uq const & b) {
return a | b;
}
// vector operator ^ : bitwise xor
static inline Vec4uq operator ^ (Vec4uq const & a, Vec4uq const & b) {
return Vec4uq(Vec256b(a) ^ Vec256b(b));
}
// Functions for this class
// Select between two operands. Corresponds to this pseudocode:
// for (int i = 0; i < 4; i++) result[i] = s[i] ? a[i] : b[i];
// Each word in s must be either 0 (false) or -1 (true). No other values are allowed.
// (s is signed)
static inline Vec4uq select (Vec4qb const & s, Vec4uq const & a, Vec4uq const & b) {
return selectb(s,a,b);
}
// Conditional add: For all vector elements i: result[i] = f[i] ? (a[i] + b[i]) : a[i]
static inline Vec4uq if_add (Vec4qb const & f, Vec4uq const & a, Vec4uq const & b) {
return a + (Vec4uq(f) & b);
}
// Horizontal add: Calculates the sum of all vector elements.
// Overflow will wrap around
static inline uint64_t horizontal_add (Vec4uq const & a) {
return horizontal_add((Vec4q)a);
}
// Horizontal add extended: Calculates the sum of all vector elements.
// Elements are sing/zero extended before adding to avoid overflow
static inline int64_t horizontal_add_x (Vec8i const & a) {
__m256i signs = _mm256_srai_epi32(a,31); // sign of all elements
Vec4q a01 = _mm256_unpacklo_epi32(a,signs); // sign-extended a0, a1, a4, a5
Vec4q a23 = _mm256_unpackhi_epi32(a,signs); // sign-extended a2, a3, a6, a7
return horizontal_add(a01 + a23);
}
static inline uint64_t horizontal_add_x (Vec8ui const & a) {
__m256i zero = _mm256_setzero_si256(); // 0
__m256i a01 = _mm256_unpacklo_epi32(a,zero); // zero-extended a0, a1
__m256i a23 = _mm256_unpackhi_epi32(a,zero); // zero-extended a2, a3
return horizontal_add(Vec4q(a01) + Vec4q(a23));
}
// function max: a > b ? a : b
static inline Vec4uq max(Vec4uq const & a, Vec4uq const & b) {
return Vec4uq(select(a > b, a, b));
}
// function min: a < b ? a : b
static inline Vec4uq min(Vec4uq const & a, Vec4uq const & b) {
return Vec4uq(select(a > b, b, a));
}
/*****************************************************************************
*
* Vector permute functions
*
******************************************************************************
*
* These permute functions can reorder the elements of a vector and optionally
* set some elements to zero.
*
* The indexes are inserted as template parameters in <>. These indexes must be
* constants. Each template parameter is an index to the element you want to select.
* An index of -1 will generate zero. An index of -256 means don't care.
*
* Example:
* Vec8i a(10,11,12,13,14,15,16,17); // a is (10,11,12,13,14,15,16,17)
* Vec8i b;
* b = permute8i<0,2,7,7,-1,-1,1,1>(a); // b is (10,12,17,17, 0, 0,11,11)
*
* A lot of the code here is metaprogramming aiming to find the instructions
* that best fit the template parameters and instruction set. The metacode
* will be reduced out to leave only a few vector instructions in release
* mode with optimization on.
*****************************************************************************/
// Permute vector of 4 64-bit integers.
// Index -1 gives 0, index -256 means don't care.
template <int i0, int i1, int i2, int i3 >
static inline Vec4q permute4q(Vec4q const & a) {
// Combine indexes into a single bitfield, with 8 bits for each
const int m1 = (i0 & 3) | (i1 & 3) << 8 | (i2 & 3) << 16 | (i3 & 3) << 24;
// Mask to zero out negative indexes
const int mz = (i0<0 ? 0 : 0xFF) | (i1<0 ? 0 : 0xFF) << 8 | (i2<0 ? 0 : 0xFF) << 16 | (i3<0 ? 0 : 0xFF) << 24;
// zeroing needed
const bool dozero = ((i0|i1|i2|i3) & 0x80) != 0;
if (((m1 ^ 0x03020100) & mz) == 0) {
// no shuffling
if (dozero) {
// zero some elements
const __m256i maskz = constant8i <
i0 < 0 ? 0 : -1, i0 < 0 ? 0 : -1, i1 < 0 ? 0 : -1, i1 < 0 ? 0 : -1,
i2 < 0 ? 0 : -1, i2 < 0 ? 0 : -1, i3 < 0 ? 0 : -1, i3 < 0 ? 0 : -1 > ();
return _mm256_and_si256(a, maskz);
}
return a; // do nothing
}
if (((m1 ^ 0x02020000) & 0x02020202 & mz) == 0) {
// no exchange of data between low and high half
if (((m1 ^ (m1 >> 16)) & 0x0101 & mz & (mz >> 16)) == 0 && !dozero) {
// same pattern in low and high half. use VPSHUFD
const int sd = (((i0>=0)?(i0&1):(i2&1)) * 10 + 4) | (((i1>=0)?(i1&1):(i3&1)) * 10 + 4) << 4;
return _mm256_shuffle_epi32(a, sd);
}
// use VPSHUFB
const __m256i mm = constant8i <
i0 < 0 ? -1 : (i0 & 1) * 0x08080808 + 0x03020100,
i0 < 0 ? -1 : (i0 & 1) * 0x08080808 + 0x07060504,
i1 < 0 ? -1 : (i1 & 1) * 0x08080808 + 0x03020100,
i1 < 0 ? -1 : (i1 & 1) * 0x08080808 + 0x07060504,
i2 < 0 ? -1 : (i2 & 1) * 0x08080808 + 0x03020100,
i2 < 0 ? -1 : (i2 & 1) * 0x08080808 + 0x07060504,
i3 < 0 ? -1 : (i3 & 1) * 0x08080808 + 0x03020100,
i3 < 0 ? -1 : (i3 & 1) * 0x08080808 + 0x07060504 > ();
return _mm256_shuffle_epi8(a, mm);
}
// general case. Use VPERMQ
const int ms = (i0 & 3) | (i1 & 3) << 2 | (i2 & 3) << 4 | (i3 & 3) << 6;
__m256i t1 = _mm256_permute4x64_epi64(a, ms);
if (dozero) {
// zero some elements
const __m256i maskz = constant8i <
i0 < 0 ? 0 : -1, i0 < 0 ? 0 : -1, i1 < 0 ? 0 : -1, i1 < 0 ? 0 : -1,
i2 < 0 ? 0 : -1, i2 < 0 ? 0 : -1, i3 < 0 ? 0 : -1, i3 < 0 ? 0 : -1 > ();
return _mm256_and_si256(t1, maskz);
}
return t1;
}
template <int i0, int i1, int i2, int i3>
static inline Vec4uq permute4uq(Vec4uq const & a) {
return Vec4uq (permute4q<i0,i1,i2,i3> (a));
}
// Permute vector of 8 32-bit integers.
// Index -1 gives 0, index -256 means don't care.
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7 >
static inline Vec8i permute8i(Vec8i const & a) {
// Combine indexes into a single bitfield, with 4 bits for each
const int m1 = (i0&7) | (i1&7)<<4 | (i2&7)<<8 | (i3&7)<<12
| (i4&7)<<16 | (i5&7)<<20 | (i6&7)<<24 | (i7&7)<<28;
// Mask to zero out negative indexes
const int mz = (i0<0?0:0xF) | (i1<0?0:0xF)<<4 | (i2<0?0:0xF)<<8 | (i3<0?0:0xF)<<12
| (i4<0?0:0xF)<<16 | (i5<0?0:0xF)<<20 | (i6<0?0:0xF)<<24 | (i7<0?0:0xF)<<28;
// zeroing needed
const bool dozero = ((i0|i1|i2|i3|i4|i5|i6|i7) & 0x80) != 0;
__m256i t1, mask;
if (((m1 ^ 0x76543210) & mz) == 0) {
// no shuffling
if (dozero) {
// zero some elements
mask = constant8i <
i0 < 0 ? 0 : -1, i1 < 0 ? 0 : -1, i2 < 0 ? 0 : -1, i3 < 0 ? 0 : -1,
i4 < 0 ? 0 : -1, i5 < 0 ? 0 : -1, i6 < 0 ? 0 : -1, i7 < 0 ? 0 : -1 > ();
return _mm256_and_si256(a, mask);
}
return a; // do nothing
}
// Check if we can use 64-bit permute. Even numbered indexes must be even and odd numbered
// indexes must be equal to the preceding index + 1, except for negative indexes.
if (((m1 ^ 0x10101010) & 0x11111111 & mz) == 0 && ((m1 ^ m1 >> 4) & 0x0E0E0E0E & mz & mz >> 4) == 0) {
const bool partialzero = int((i0^i1)|(i2^i3)|(i4^i5)|(i6^i7)) < 0; // part of a 64-bit block is zeroed
const int blank1 = partialzero ? -0x100 : -1; // ignore or zero
const int n0 = i0 > 0 ? i0 /2 : i1 > 0 ? i1 /2 : blank1; // indexes for 64 bit blend
const int n1 = i2 > 0 ? i2 /2 : i3 > 0 ? i3 /2 : blank1;
const int n2 = i4 > 0 ? i4 /2 : i5 > 0 ? i5 /2 : blank1;
const int n3 = i6 > 0 ? i6 /2 : i7 > 0 ? i7 /2 : blank1;
// do 64-bit permute
t1 = permute4q<n0,n1,n2,n3> (Vec4q(a));
if (blank1 == -1 || !dozero) {
return t1;
}
// need more zeroing
mask = constant8i <
i0 < 0 ? 0 : -1, i1 < 0 ? 0 : -1, i2 < 0 ? 0 : -1, i3 < 0 ? 0 : -1,
i4 < 0 ? 0 : -1, i5 < 0 ? 0 : -1, i6 < 0 ? 0 : -1, i7 < 0 ? 0 : -1 > ();
return _mm256_and_si256(t1, mask);
}
if (((m1 ^ 0x44440000) & 0x44444444 & mz) == 0) {
// no exchange of data between low and high half
if (((m1 ^ (m1 >> 16)) & 0x3333 & mz & (mz >> 16)) == 0 && !dozero) {
// same pattern in low and high half. use VPSHUFD
const int sd = ((i0>=0)?(i0&3):(i4&3)) | ((i1>=0)?(i1&3):(i5&3)) << 2 |
((i2>=0)?(i2&3):(i6&3)) << 4 | ((i3>=0)?(i3&3):(i7&3)) << 6;
return _mm256_shuffle_epi32(a, sd);
}
// use VPSHUFB
mask = constant8i <
i0 < 0 ? -1 : (i0 & 3) * 0x04040404 + 0x03020100,
i1 < 0 ? -1 : (i1 & 3) * 0x04040404 + 0x03020100,
i2 < 0 ? -1 : (i2 & 3) * 0x04040404 + 0x03020100,
i3 < 0 ? -1 : (i3 & 3) * 0x04040404 + 0x03020100,
i4 < 0 ? -1 : (i4 & 3) * 0x04040404 + 0x03020100,
i5 < 0 ? -1 : (i5 & 3) * 0x04040404 + 0x03020100,
i6 < 0 ? -1 : (i6 & 3) * 0x04040404 + 0x03020100,
i7 < 0 ? -1 : (i7 & 3) * 0x04040404 + 0x03020100 > ();
return _mm256_shuffle_epi8(a, mask);
}
// general case. Use VPERMD
mask = constant8i <
i0 < 0 ? -1 : (i0 & 7), i1 < 0 ? -1 : (i1 & 7),
i2 < 0 ? -1 : (i2 & 7), i3 < 0 ? -1 : (i3 & 7),
i4 < 0 ? -1 : (i4 & 7), i5 < 0 ? -1 : (i5 & 7),
i6 < 0 ? -1 : (i6 & 7), i7 < 0 ? -1 : (i7 & 7) > ();
#if defined (_MSC_VER) && _MSC_VER < 1700 && ! defined(__INTEL_COMPILER)
// bug in MS VS 11 beta: operands in wrong order. fixed in v. 11.0
t1 = _mm256_permutevar8x32_epi32(mask, a); // ms
#elif defined (GCC_VERSION) && GCC_VERSION <= 40700 && !defined(__INTEL_COMPILER) && !defined(__clang__)
// Gcc 4.7.0 also has operands in wrong order. fixed in version 4.7.1
t1 = _mm256_permutevar8x32_epi32(mask, a); // GCC
#else
t1 = _mm256_permutevar8x32_epi32(a, mask); // no-bug version
#endif
if (dozero) {
// zero some elements
mask = constant8i <
i0 < 0 ? 0 : -1, i1 < 0 ? 0 : -1, i2 < 0 ? 0 : -1, i3 < 0 ? 0 : -1,
i4 < 0 ? 0 : -1, i5 < 0 ? 0 : -1, i6 < 0 ? 0 : -1, i7 < 0 ? 0 : -1 > ();
return _mm256_and_si256(t1, mask);
}
return t1;
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7 >
static inline Vec8ui permute8ui(Vec8ui const & a) {
return Vec8ui (permute8i<i0,i1,i2,i3,i4,i5,i6,i7> (a));
}
// Permute vector of 16 16-bit integers.
// Index -1 gives 0, index -256 means don't care.
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15 >
static inline Vec16s permute16s(Vec16s const & a) {
// Combine indexes 0 - 7 into a single bitfield, with 4 bits for each
const int mlo = (i0&0xF) | (i1&0xF)<<4 | (i2&0xF)<<8 | (i3&0xF)<<12
| (i4&0xF)<<16 | (i5&0xF)<<20 | (i6&0xF)<<24 | (i7&0xF)<<28;
// Combine indexes 8 - 15 into a single bitfield, with 4 bits for each
const int mhi = (i8&0xF) | (i9&0xF)<<4 | (i10&0xF)<<8 | (i11&0xF)<<12
| (i12&0xF)<<16 | (i13&0xF)<<20 | (i14&0xF)<<24 | (i15&0xF)<<28;
// Mask to zero out negative indexes 0 - 7
const int zlo = (i0<0?0:0xF) | (i1<0?0:0xF)<<4 | (i2<0?0:0xF)<<8 | (i3<0?0:0xF)<<12
| (i4<0?0:0xF)<<16 | (i5<0?0:0xF)<<20 | (i6<0?0:0xF)<<24 | (i7<0?0:0xF)<<28;
// Mask to zero out negative indexes 8 - 15
const int zhi = (i8<0?0:0xF) | (i9<0?0:0xF)<<4 | (i10<0?0:0xF)<<8 | (i11<0?0:0xF)<<12
| (i12<0?0:0xF)<<16 | (i13<0?0:0xF)<<20 | (i14<0?0:0xF)<<24 | (i15<0?0:0xF)<<28;
// zeroing needed
const bool dozero = ((i0|i1|i2|i3|i4|i5|i6|i7|i8|i9|i10|i11|i12|i13|i14|i15) & 0x80) != 0;
__m256i t1, mask;
// special case: all zero
if (zlo == 0 && zhi == 0) {
return _mm256_setzero_si256();
}
// special case: rotate 128 bits
if (i0>=0 && i0 < 16 && i1 ==((i0+1)&7) && i2 ==((i0+2)&7) && i3 ==((i0+3)&7) && i4 ==((i0+4)&7) && i5 ==((i0+5)&7) && i6 ==((i0+6)&7) && i7 ==((i0+7)&7)
&& i8 ==i0 +8 && i9 ==i1 +8 && i10==i2 +8 && i11==i3 +8 && i12==i4 +8 && i13==i5 +8 && i14==i6 +8 && i15==i7 +8 ) {
return _mm256_alignr_epi8(a, a, (i0 & 7) * 2);
}
// special case: rotate 256 bits
if (i0>=0 && i0 < 16 && i1 ==((i0+1 )&15) && i2 ==((i0+2 )&15) && i3 ==((i0+3 )&15) && i4 ==((i0+4 )&15) && i5 ==((i0+5 )&15) && i6 ==((i0+6 )&15) && i7 ==((i0+7 )&15)
&& i8 ==((i0+8 )&15) && i9 ==((i0+9 )&15) && i10==((i0+10)&15) && i11==((i0+11)&15) && i12==((i0+12)&15) && i13==((i0+13)&15) && i14==((i0+14)&15) && i15==((i0+15)&15)) {
t1 = _mm256_permute4x64_epi64(a, 0x4E);
return _mm256_alignr_epi8(a, t1, (i0 & 7) * 2);
}
// special case: no exchange of data between 64-bit sections, and same pattern in low and high 128 bits:
// can use VPSHUFLW or VPSHUFHW
if (((mlo ^ 0x44440000) & 0xCCCCCCCC & zlo) == 0 && ((mhi ^ 0xCCCC8888) & 0xCCCCCCCC & zhi) == 0
&& ((mlo ^ mhi) & 0x33333333 & zlo & zhi) == 0) {
const int slo = (i0 >= 0 ? (i0&3) : i8 >= 0 ? (i8&3) : 0) | (i1 >= 0 ? (i1&3) : i9 >= 0 ? (i9&3) : 1) << 2
| (i2 >= 0 ? (i2&3) : i10 >= 0 ? (i10&3) : 2) << 4 | (i3 >= 0 ? (i3&3) : i11 >= 0 ? (i11&3) : 3) << 6;
const int shi = (i4 >= 0 ? (i4&3) : i12 >= 0 ? (i12&3) : 0) | (i5 >= 0 ? (i5&3) : i13 >= 0 ? (i13&3) : 1) << 2
| (i6 >= 0 ? (i6&3) : i14 >= 0 ? (i14&3) : 2) << 4 | (i7 >= 0 ? (i7&3) : i15 >= 0 ? (i15&3) : 3) << 6;
if (shi == 0xE4 && slo == 0xE4) { // no permute
if (dozero) {
// zero some elements
const __m256i maskz = constant8i<
int((i0 <0?0:0xFFFF) | (i1 <0?0:0xFFFF0000)),
int((i2 <0?0:0xFFFF) | (i3 <0?0:0xFFFF0000)),
int((i4 <0?0:0xFFFF) | (i5 <0?0:0xFFFF0000)),
int((i6 <0?0:0xFFFF) | (i7 <0?0:0xFFFF0000)),
int((i8 <0?0:0xFFFF) | (i9 <0?0:0xFFFF0000)),
int((i10<0?0:0xFFFF) | (i11<0?0:0xFFFF0000)),
int((i12<0?0:0xFFFF) | (i13<0?0:0xFFFF0000)),
int((i14<0?0:0xFFFF) | (i15<0?0:0xFFFF0000)) > ();
return _mm256_and_si256(a, maskz);
}
return a; // do nothing
}
if (shi == 0xE4 && !dozero) {
return _mm256_shufflelo_epi16(a, slo); // low permute only
}
if (slo == 0xE4 && !dozero) {
return _mm256_shufflehi_epi16(a, shi); // high permute only
}
}
// Check if we can use 32-bit permute. Even numbered indexes must be even and odd numbered
// indexes must be equal to the preceding index + 1, except for negative indexes.
if (((mlo ^ 0x10101010) & 0x11111111 & zlo) == 0 && ((mlo ^ mlo >> 4) & 0x0E0E0E0E & zlo & zlo >> 4) == 0 &&
((mhi ^ 0x10101010) & 0x11111111 & zhi) == 0 && ((mhi ^ mhi >> 4) & 0x0E0E0E0E & zhi & zhi >> 4) == 0 ) {
const bool partialzero = int((i0^i1)|(i2^i3)|(i4^i5)|(i6^i7)|(i8^i9)|(i10^i11)|(i12^i13)|(i14^i15)) < 0; // part of a 32-bit block is zeroed
const int blank1 = partialzero ? -0x100 : -1; // ignore or zero
const int n0 = i0 > 0 ? i0 /2 : i1 > 0 ? i1 /2 : blank1; // indexes for 64 bit blend
const int n1 = i2 > 0 ? i2 /2 : i3 > 0 ? i3 /2 : blank1;
const int n2 = i4 > 0 ? i4 /2 : i5 > 0 ? i5 /2 : blank1;
const int n3 = i6 > 0 ? i6 /2 : i7 > 0 ? i7 /2 : blank1;
const int n4 = i8 > 0 ? i8 /2 : i9 > 0 ? i9 /2 : blank1;
const int n5 = i10> 0 ? i10/2 : i11> 0 ? i11/2 : blank1;
const int n6 = i12> 0 ? i12/2 : i13> 0 ? i13/2 : blank1;
const int n7 = i14> 0 ? i14/2 : i15> 0 ? i15/2 : blank1;
// do 32-bit permute
t1 = permute8i<n0,n1,n2,n3,n4,n5,n6,n7> (Vec8i(a));
if (blank1 == -1 || !dozero) {
return t1;
}
// need more zeroing
mask = constant8i<
int((i0 <0?0:0xFFFF) | (i1 <0?0:0xFFFF0000)),
int((i2 <0?0:0xFFFF) | (i3 <0?0:0xFFFF0000)),
int((i4 <0?0:0xFFFF) | (i5 <0?0:0xFFFF0000)),
int((i6 <0?0:0xFFFF) | (i7 <0?0:0xFFFF0000)),
int((i8 <0?0:0xFFFF) | (i9 <0?0:0xFFFF0000)),
int((i10<0?0:0xFFFF) | (i11<0?0:0xFFFF0000)),
int((i12<0?0:0xFFFF) | (i13<0?0:0xFFFF0000)),
int((i14<0?0:0xFFFF) | (i15<0?0:0xFFFF0000)) > ();
return _mm256_and_si256(t1, mask);
}
// special case: all elements from same half
if ((mlo & 0x88888888 & zlo) == 0 && ((mhi ^ 0x88888888) & 0x88888888 & zhi) == 0) {
mask = constant8i<
(i0 < 0 ? 0xFFFF : (i0 & 7) * 0x202 + 0x100) | (i1 < 0 ? 0xFFFF : (i1 & 7) * 0x202 + 0x100) << 16,
(i2 < 0 ? 0xFFFF : (i2 & 7) * 0x202 + 0x100) | (i3 < 0 ? 0xFFFF : (i3 & 7) * 0x202 + 0x100) << 16,
(i4 < 0 ? 0xFFFF : (i4 & 7) * 0x202 + 0x100) | (i5 < 0 ? 0xFFFF : (i5 & 7) * 0x202 + 0x100) << 16,
(i6 < 0 ? 0xFFFF : (i6 & 7) * 0x202 + 0x100) | (i7 < 0 ? 0xFFFF : (i7 & 7) * 0x202 + 0x100) << 16,
(i8 < 0 ? 0xFFFF : (i8 & 7) * 0x202 + 0x100) | (i9 < 0 ? 0xFFFF : (i9 & 7) * 0x202 + 0x100) << 16,
(i10 < 0 ? 0xFFFF : (i10 & 7) * 0x202 + 0x100) | (i11 < 0 ? 0xFFFF : (i11 & 7) * 0x202 + 0x100) << 16,
(i12 < 0 ? 0xFFFF : (i12 & 7) * 0x202 + 0x100) | (i13 < 0 ? 0xFFFF : (i13 & 7) * 0x202 + 0x100) << 16,
(i14 < 0 ? 0xFFFF : (i14 & 7) * 0x202 + 0x100) | (i15 < 0 ? 0xFFFF : (i15 & 7) * 0x202 + 0x100) << 16 > ();
return _mm256_shuffle_epi8(a, mask);
}
// special case: all elements from low half
if ((mlo & 0x88888888 & zlo) == 0 && (mhi & 0x88888888 & zhi) == 0) {
mask = constant8i<
(i0 < 0 ? 0xFFFF : (i0 & 7) * 0x202 + 0x100) | (i1 < 0 ? 0xFFFF : (i1 & 7) * 0x202 + 0x100) << 16,
(i2 < 0 ? 0xFFFF : (i2 & 7) * 0x202 + 0x100) | (i3 < 0 ? 0xFFFF : (i3 & 7) * 0x202 + 0x100) << 16,
(i4 < 0 ? 0xFFFF : (i4 & 7) * 0x202 + 0x100) | (i5 < 0 ? 0xFFFF : (i5 & 7) * 0x202 + 0x100) << 16,
(i6 < 0 ? 0xFFFF : (i6 & 7) * 0x202 + 0x100) | (i7 < 0 ? 0xFFFF : (i7 & 7) * 0x202 + 0x100) << 16,
(i8 < 0 ? 0xFFFF : (i8 & 7) * 0x202 + 0x100) | (i9 < 0 ? 0xFFFF : (i9 & 7) * 0x202 + 0x100) << 16,
(i10 < 0 ? 0xFFFF : (i10 & 7) * 0x202 + 0x100) | (i11 < 0 ? 0xFFFF : (i11 & 7) * 0x202 + 0x100) << 16,
(i12 < 0 ? 0xFFFF : (i12 & 7) * 0x202 + 0x100) | (i13 < 0 ? 0xFFFF : (i13 & 7) * 0x202 + 0x100) << 16,
(i14 < 0 ? 0xFFFF : (i14 & 7) * 0x202 + 0x100) | (i15 < 0 ? 0xFFFF : (i15 & 7) * 0x202 + 0x100) << 16 > ();
t1 = _mm256_inserti128_si256(a, _mm256_castsi256_si128(a), 1); // low, low
return _mm256_shuffle_epi8(t1, mask);
}
// special case: all elements from high half
if (((mlo ^ 0x88888888) & 0x88888888 & zlo) == 0 && ((mhi ^ 0x88888888) & 0x88888888 & zhi) == 0) {
mask = constant8i<
(i0 < 0 ? 0xFFFF : (i0 & 7) * 0x202 + 0x100) | (i1 < 0 ? 0xFFFF : (i1 & 7) * 0x202 + 0x100) << 16,
(i2 < 0 ? 0xFFFF : (i2 & 7) * 0x202 + 0x100) | (i3 < 0 ? 0xFFFF : (i3 & 7) * 0x202 + 0x100) << 16,
(i4 < 0 ? 0xFFFF : (i4 & 7) * 0x202 + 0x100) | (i5 < 0 ? 0xFFFF : (i5 & 7) * 0x202 + 0x100) << 16,
(i6 < 0 ? 0xFFFF : (i6 & 7) * 0x202 + 0x100) | (i7 < 0 ? 0xFFFF : (i7 & 7) * 0x202 + 0x100) << 16,
(i8 < 0 ? 0xFFFF : (i8 & 7) * 0x202 + 0x100) | (i9 < 0 ? 0xFFFF : (i9 & 7) * 0x202 + 0x100) << 16,
(i10 < 0 ? 0xFFFF : (i10 & 7) * 0x202 + 0x100) | (i11 < 0 ? 0xFFFF : (i11 & 7) * 0x202 + 0x100) << 16,
(i12 < 0 ? 0xFFFF : (i12 & 7) * 0x202 + 0x100) | (i13 < 0 ? 0xFFFF : (i13 & 7) * 0x202 + 0x100) << 16,
(i14 < 0 ? 0xFFFF : (i14 & 7) * 0x202 + 0x100) | (i15 < 0 ? 0xFFFF : (i15 & 7) * 0x202 + 0x100) << 16 > ();
t1 = _mm256_permute4x64_epi64(a, 0xEE); // high, high
return _mm256_shuffle_epi8(t1, mask);
}
// special case: all elements from opposite half
if (((mlo ^ 0x88888888) & 0x88888888 & zlo) == 0 && (mhi & 0x88888888 & zhi) == 0) {
mask = constant8i<
(i0 < 0 ? 0xFFFF : (i0 & 7) * 0x202 + 0x100) | (i1 < 0 ? 0xFFFF : (i1 & 7) * 0x202 + 0x100) << 16,
(i2 < 0 ? 0xFFFF : (i2 & 7) * 0x202 + 0x100) | (i3 < 0 ? 0xFFFF : (i3 & 7) * 0x202 + 0x100) << 16,
(i4 < 0 ? 0xFFFF : (i4 & 7) * 0x202 + 0x100) | (i5 < 0 ? 0xFFFF : (i5 & 7) * 0x202 + 0x100) << 16,
(i6 < 0 ? 0xFFFF : (i6 & 7) * 0x202 + 0x100) | (i7 < 0 ? 0xFFFF : (i7 & 7) * 0x202 + 0x100) << 16,
(i8 < 0 ? 0xFFFF : (i8 & 7) * 0x202 + 0x100) | (i9 < 0 ? 0xFFFF : (i9 & 7) * 0x202 + 0x100) << 16,
(i10 < 0 ? 0xFFFF : (i10 & 7) * 0x202 + 0x100) | (i11 < 0 ? 0xFFFF : (i11 & 7) * 0x202 + 0x100) << 16,
(i12 < 0 ? 0xFFFF : (i12 & 7) * 0x202 + 0x100) | (i13 < 0 ? 0xFFFF : (i13 & 7) * 0x202 + 0x100) << 16,
(i14 < 0 ? 0xFFFF : (i14 & 7) * 0x202 + 0x100) | (i15 < 0 ? 0xFFFF : (i15 & 7) * 0x202 + 0x100) << 16 > ();
t1 = _mm256_permute4x64_epi64(a, 0x4E); // high, low
return _mm256_shuffle_epi8(t1, mask);
}
// general case: elements from both halves
const __m256i mmsame = constant8i<
((i0 ^8) < 8 ? 0xFFFF : (i0 & 7) * 0x202 + 0x100) | ((i1 ^8) < 8 ? 0xFFFF : (i1 & 7) * 0x202 + 0x100) << 16,
((i2 ^8) < 8 ? 0xFFFF : (i2 & 7) * 0x202 + 0x100) | ((i3 ^8) < 8 ? 0xFFFF : (i3 & 7) * 0x202 + 0x100) << 16,
((i4 ^8) < 8 ? 0xFFFF : (i4 & 7) * 0x202 + 0x100) | ((i5 ^8) < 8 ? 0xFFFF : (i5 & 7) * 0x202 + 0x100) << 16,
((i6 ^8) < 8 ? 0xFFFF : (i6 & 7) * 0x202 + 0x100) | ((i7 ^8) < 8 ? 0xFFFF : (i7 & 7) * 0x202 + 0x100) << 16,
(i8 < 8 ? 0xFFFF : (i8 & 7) * 0x202 + 0x100) | (i9 < 8 ? 0xFFFF : (i9 & 7) * 0x202 + 0x100) << 16,
(i10 < 8 ? 0xFFFF : (i10 & 7) * 0x202 + 0x100) | (i11 < 8 ? 0xFFFF : (i11 & 7) * 0x202 + 0x100) << 16,
(i12 < 8 ? 0xFFFF : (i12 & 7) * 0x202 + 0x100) | (i13 < 8 ? 0xFFFF : (i13 & 7) * 0x202 + 0x100) << 16,
(i14 < 8 ? 0xFFFF : (i14 & 7) * 0x202 + 0x100) | (i15 < 8 ? 0xFFFF : (i15 & 7) * 0x202 + 0x100) << 16 > ();
const __m256i mmopposite = constant8i<
(i0 < 8 ? 0xFFFF : (i0 & 7) * 0x202 + 0x100) | (i1 < 8 ? 0xFFFF : (i1 & 7) * 0x202 + 0x100) << 16,
(i2 < 8 ? 0xFFFF : (i2 & 7) * 0x202 + 0x100) | (i3 < 8 ? 0xFFFF : (i3 & 7) * 0x202 + 0x100) << 16,
(i4 < 8 ? 0xFFFF : (i4 & 7) * 0x202 + 0x100) | (i5 < 8 ? 0xFFFF : (i5 & 7) * 0x202 + 0x100) << 16,
(i6 < 8 ? 0xFFFF : (i6 & 7) * 0x202 + 0x100) | (i7 < 8 ? 0xFFFF : (i7 & 7) * 0x202 + 0x100) << 16,
((i8 ^8) < 8 ? 0xFFFF : (i8 & 7) * 0x202 + 0x100) | ((i9 ^8) < 8 ? 0xFFFF : (i9 & 7) * 0x202 + 0x100) << 16,
((i10^8) < 8 ? 0xFFFF : (i10 & 7) * 0x202 + 0x100) | ((i11^8) < 8 ? 0xFFFF : (i11 & 7) * 0x202 + 0x100) << 16,
((i12^8) < 8 ? 0xFFFF : (i12 & 7) * 0x202 + 0x100) | ((i13^8) < 8 ? 0xFFFF : (i13 & 7) * 0x202 + 0x100) << 16,
((i14^8) < 8 ? 0xFFFF : (i14 & 7) * 0x202 + 0x100) | ((i15^8) < 8 ? 0xFFFF : (i15 & 7) * 0x202 + 0x100) << 16 > ();
__m256i topp = _mm256_permute4x64_epi64(a, 0x4E); // high, low
__m256i r1 = _mm256_shuffle_epi8(topp, mmopposite);
__m256i r2 = _mm256_shuffle_epi8(a, mmsame);
return _mm256_or_si256(r1, r2);
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15 >
static inline Vec16us permute16us(Vec16us const & a) {
return Vec16us (permute16s<i0,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15> (a));
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15,
int i16, int i17, int i18, int i19, int i20, int i21, int i22, int i23,
int i24, int i25, int i26, int i27, int i28, int i29, int i30, int i31 >
static inline Vec32c permute32c(Vec32c const & a) {
// collect bit 4 of each index
const int m1 =
(i0 &16)>>4 | (i1 &16)>>3 | (i2 &16)>>2 | (i3 &16)>>1 | (i4 &16) | (i5 &16)<<1 | (i6 &16)<<2 | (i7 &16)<<3 |
(i8 &16)<<4 | (i9 &16)<<5 | (i10&16)<<6 | (i11&16)<<7 | (i12&16)<<8 | (i13&16)<<9 | (i14&16)<<10 | (i15&16)<<11 |
(i16&16)<<12 | (i17&16)<<13 | (i18&16)<<14 | (i19&16)<<15 | (i20&16)<<16 | (i21&16)<<17 | (i22&16)<<18 | (i23&16)<<19 |
(i24&16)<<20 | (i25&16)<<21 | (i26&16)<<22 | (i27&16)<<23 | (i28&16)<<24 | (i29&16)<<25 | (i30&16)<<26 | (i31&16)<<27 ;
// check which elements to set to zero
const int mz = ~ (
(i0 <0) | (i1 <0)<<1 | (i2 <0)<<2 | (i3 <0)<<3 | (i4 <0)<<4 | (i5 <0)<<5 | (i6 <0)<<6 | (i7 <0)<<7 |
(i8 <0)<<8 | (i9 <0)<<9 | (i10<0)<<10 | (i11<0)<<11 | (i12<0)<<12 | (i13<0)<<13 | (i14<0)<<14 | (i15<0)<<15 |
(i16<0)<<16 | (i17<0)<<17 | (i18<0)<<18 | (i19<0)<<19 | (i20<0)<<20 | (i21<0)<<21 | (i22<0)<<22 | (i23<0)<<23 |
(i24<0)<<24 | (i25<0)<<25 | (i26<0)<<26 | (i27<0)<<27 | (i28<0)<<28 | (i29<0)<<29 | (i30<0)<<30 | (i31<0)<<31 );
// Combine indexes 0-7, 8-15, 16-23, 24-31 into a bitfields, with 8 bits for each
const uint64_t g0 = (i0 &0x1F)|(i1 &0x1F)<<8|(i2 &0x1F)<<16|(i3 &0x1F)<<24|(i4 &0x1FLL)<<32|(i5 &0x1FLL)<<40|(i6 &0x1FLL)<<48|(i7 &0x1FLL)<<56;
const uint64_t g1 = (i8 &0x1F)|(i9 &0x1F)<<8|(i10&0x1F)<<16|(i11&0x1F)<<24|(i12&0x1FLL)<<32|(i13&0x1FLL)<<40|(i14&0x1FLL)<<48|(i15&0x1FLL)<<56;
const uint64_t g2 = (i16&0x1F)|(i17&0x1F)<<8|(i18&0x1F)<<16|(i19&0x1F)<<24|(i20&0x1FLL)<<32|(i21&0x1FLL)<<40|(i22&0x1FLL)<<48|(i23&0x1FLL)<<56;
const uint64_t g3 = (i24&0x1F)|(i25&0x1F)<<8|(i26&0x1F)<<16|(i27&0x1F)<<24|(i28&0x1FLL)<<32|(i29&0x1FLL)<<40|(i30&0x1FLL)<<48|(i31&0x1FLL)<<56;
// Masks to zero out negative indexes
const uint64_t z0 = (i0 <0?0:0xFF)|(i1 <0?0:0xFF)<<8|(i2 <0?0:0xFF)<<16|(i3 <0?0:0xFF)<<24|(i4 <0?0:0xFFLL)<<32|(i5 <0?0:0xFFLL)<<40|(i6 <0?0:0xFFLL)<<48|(i7 <0?0:0xFFLL)<<56;
const uint64_t z1 = (i8 <0?0:0xFF)|(i9 <0?0:0xFF)<<8|(i10<0?0:0xFF)<<16|(i11<0?0:0xFF)<<24|(i12<0?0:0xFFLL)<<32|(i13<0?0:0xFFLL)<<40|(i14<0?0:0xFFLL)<<48|(i15<0?0:0xFFLL)<<56;
const uint64_t z2 = (i16<0?0:0xFF)|(i17<0?0:0xFF)<<8|(i18<0?0:0xFF)<<16|(i19<0?0:0xFF)<<24|(i20<0?0:0xFFLL)<<32|(i21<0?0:0xFFLL)<<40|(i22<0?0:0xFFLL)<<48|(i23<0?0:0xFFLL)<<56;
const uint64_t z3 = (i24<0?0:0xFF)|(i25<0?0:0xFF)<<8|(i26<0?0:0xFF)<<16|(i27<0?0:0xFF)<<24|(i28<0?0:0xFFLL)<<32|(i29<0?0:0xFFLL)<<40|(i30<0?0:0xFFLL)<<48|(i31<0?0:0xFFLL)<<56;
// zeroing needed
const bool dozero = ((i0|i1|i2|i3|i4|i5|i6|i7|i8|i9|i10|i11|i12|i13|i14|i15|i16|i17|i18|i19|i20|i21|i22|i23|i24|i25|i26|i27|i28|i29|i30|i31) & 0x80) != 0;
__m256i t1, mask;
// special case: all zero
if (mz == 0) return _mm256_setzero_si256();
// special case: no permute
if ((i0 <0||i0 == 0) && (i1 <0||i1 == 1) && (i2 <0||i2 == 2) && (i3 <0||i3 == 3) && (i4 <0||i4 == 4) && (i5 <0||i5 == 5) && (i6 <0||i6 == 6) && (i7 <0||i7 == 7) &&
(i8 <0||i8 == 8) && (i9 <0||i9 == 9) && (i10<0||i10==10) && (i11<0||i11==11) && (i12<0||i12==12) && (i13<0||i13==13) && (i14<0||i14==14) && (i15<0||i15==15) &&
(i16<0||i16==16) && (i17<0||i17==17) && (i18<0||i18==18) && (i19<0||i19==19) && (i20<0||i20==20) && (i21<0||i21==21) && (i22<0||i22==22) && (i23<0||i23==23) &&
(i24<0||i24==24) && (i25<0||i25==25) && (i26<0||i26==26) && (i27<0||i27==27) && (i28<0||i28==28) && (i29<0||i29==29) && (i30<0||i30==30) && (i31<0||i31==31)) {
if (dozero) {
// zero some elements
mask = constant8i <
int((i0 <0?0:0xFF) | (i1 <0?0:0xFF00) | (i2 <0?0:0xFF0000) | (i3 <0?0:0xFF000000)),
int((i4 <0?0:0xFF) | (i5 <0?0:0xFF00) | (i6 <0?0:0xFF0000) | (i7 <0?0:0xFF000000)),
int((i8 <0?0:0xFF) | (i9 <0?0:0xFF00) | (i10<0?0:0xFF0000) | (i11<0?0:0xFF000000)),
int((i12<0?0:0xFF) | (i13<0?0:0xFF00) | (i14<0?0:0xFF0000) | (i15<0?0:0xFF000000)),
int((i16<0?0:0xFF) | (i17<0?0:0xFF00) | (i18<0?0:0xFF0000) | (i19<0?0:0xFF000000)),
int((i20<0?0:0xFF) | (i21<0?0:0xFF00) | (i22<0?0:0xFF0000) | (i23<0?0:0xFF000000)),
int((i24<0?0:0xFF) | (i25<0?0:0xFF00) | (i26<0?0:0xFF0000) | (i27<0?0:0xFF000000)),
int((i28<0?0:0xFF) | (i29<0?0:0xFF00) | (i30<0?0:0xFF0000) | (i31<0?0:0xFF000000)) > ();
return _mm256_and_si256(a, mask);
}
return a; // do nothing
}
// special case: rotate 128 bits
if (i0>=0 && i0 < 32 && i1 ==((i0+1 )&15) && i2 ==((i0+2 )&15) && i3 ==((i0+3 )&15) && i4 ==((i0+4 )&15) && i5 ==((i0+5 )&15) && i6 ==((i0+6 )&15) && i7 ==((i0+7 )&15)
&& i8 ==((i0+8 )&15) && i9 ==((i0+9 )&15) && i10==((i0+10)&15) && i11==((i0+11)&15) && i12==((i0+12)&15) && i13==((i0+13)&15) && i14==((i0+14)&15) && i15==((i0+15)&15)
&& i16==i0 +16 && i17==i1 +16 && i18==i2 +16 && i19==i3 +16 && i20==i4 +16 && i21==i5 +16 && i22==i6 +16 && i23==i7 +16
&& i24==i8 +16 && i25==i9 +16 && i26==i10+16 && i27==i11+16 && i28==i12+16 && i29==i13+16 && i30==i14+16 && i31==i15+16 ) {
return _mm256_alignr_epi8(a, a, i0 & 15);
}
// special case: rotate 256 bits
if (i0>=0 && i0 < 32 && i1 ==((i0+1 )&31) && i2 ==((i0+2 )&31) && i3 ==((i0+3 )&31) && i4 ==((i0+4 )&31) && i5 ==((i0+5 )&31) && i6 ==((i0+6 )&31) && i7 ==((i0+7 )&31)
&& i8 ==((i0+8 )&31) && i9 ==((i0+9 )&31) && i10==((i0+10)&31) && i11==((i0+11)&31) && i12==((i0+12)&31) && i13==((i0+13)&31) && i14==((i0+14)&31) && i15==((i0+15)&31)
&& i16==((i0+16)&31) && i17==((i0+17)&31) && i18==((i0+18)&31) && i19==((i0+19)&31) && i20==((i0+20)&31) && i21==((i0+21)&31) && i22==((i0+22)&31) && i23==((i0+23)&31)
&& i24==((i0+24)&31) && i25==((i0+25)&31) && i26==((i0+26)&31) && i27==((i0+27)&31) && i28==((i0+28)&31) && i29==((i0+29)&31) && i30==((i0+30)&31) && i31==((i0+31)&31)) {
t1 = _mm256_permute4x64_epi64(a, 0x4E);
return _mm256_alignr_epi8(a, t1, i0 & 15);
}
// Check if we can use 16-bit permute. Even numbered indexes must be even and odd numbered
// indexes must be equal to the preceding index + 1, except for negative indexes.
if (((g0 ^ 0x0100010001000100) & 0x0101010101010101 & z0) == 0 && ((g0 ^ g0 >> 8) & 0x00FE00FE00FE00FE & z0 & z0 >> 8) == 0 &&
((g1 ^ 0x0100010001000100) & 0x0101010101010101 & z1) == 0 && ((g1 ^ g1 >> 8) & 0x00FE00FE00FE00FE & z1 & z1 >> 8) == 0 &&
((g2 ^ 0x0100010001000100) & 0x0101010101010101 & z2) == 0 && ((g2 ^ g2 >> 8) & 0x00FE00FE00FE00FE & z2 & z2 >> 8) == 0 &&
((g3 ^ 0x0100010001000100) & 0x0101010101010101 & z3) == 0 && ((g3 ^ g3 >> 8) & 0x00FE00FE00FE00FE & z3 & z3 >> 8) == 0 ) {
const bool partialzero = int((i0^i1)|(i2^i3)|(i4^i5)|(i6^i7)|(i8^i9)|(i10^i11)|(i12^i13)|(i14^i15)
|(i16^i17)|(i18^i19)|(i20^i21)|(i22^i23)|(i24^i25)|(i26^i27)|(i28^i29)|(i30^i31)) < 0; // part of a 16-bit block is zeroed
const int blank1 = partialzero ? -0x100 : -1; // ignore or zero
const int n0 = i0 > 0 ? i0 /2 : i1 > 0 ? i1 /2 : blank1; // indexes for 64 bit blend
const int n1 = i2 > 0 ? i2 /2 : i3 > 0 ? i3 /2 : blank1;
const int n2 = i4 > 0 ? i4 /2 : i5 > 0 ? i5 /2 : blank1;
const int n3 = i6 > 0 ? i6 /2 : i7 > 0 ? i7 /2 : blank1;
const int n4 = i8 > 0 ? i8 /2 : i9 > 0 ? i9 /2 : blank1;
const int n5 = i10> 0 ? i10/2 : i11> 0 ? i11/2 : blank1;
const int n6 = i12> 0 ? i12/2 : i13> 0 ? i13/2 : blank1;
const int n7 = i14> 0 ? i14/2 : i15> 0 ? i15/2 : blank1;
const int n8 = i16> 0 ? i16/2 : i17> 0 ? i17/2 : blank1;
const int n9 = i18> 0 ? i18/2 : i19> 0 ? i19/2 : blank1;
const int n10= i20> 0 ? i20/2 : i21> 0 ? i21/2 : blank1;
const int n11= i22> 0 ? i22/2 : i23> 0 ? i23/2 : blank1;
const int n12= i24> 0 ? i24/2 : i25> 0 ? i25/2 : blank1;
const int n13= i26> 0 ? i26/2 : i27> 0 ? i27/2 : blank1;
const int n14= i28> 0 ? i28/2 : i29> 0 ? i29/2 : blank1;
const int n15= i30> 0 ? i30/2 : i31> 0 ? i31/2 : blank1;
// do 16-bit permute
t1 = permute16s<n0,n1,n2,n3,n4,n5,n6,n7,n8,n9,n10,n11,n12,n13,n14,n15> (Vec16s(a));
if (blank1 == -1 || !dozero) {
return t1;
}
// need more zeroing
mask = constant8i <
int((i0 <0?0:0xFF) | (i1 <0?0:0xFF00) | (i2 <0?0:0xFF0000) | (i3 <0?0:0xFF000000)),
int((i4 <0?0:0xFF) | (i5 <0?0:0xFF00) | (i6 <0?0:0xFF0000) | (i7 <0?0:0xFF000000)),
int((i8 <0?0:0xFF) | (i9 <0?0:0xFF00) | (i10<0?0:0xFF0000) | (i11<0?0:0xFF000000)),
int((i12<0?0:0xFF) | (i13<0?0:0xFF00) | (i14<0?0:0xFF0000) | (i15<0?0:0xFF000000)),
int((i16<0?0:0xFF) | (i17<0?0:0xFF00) | (i18<0?0:0xFF0000) | (i19<0?0:0xFF000000)),
int((i20<0?0:0xFF) | (i21<0?0:0xFF00) | (i22<0?0:0xFF0000) | (i23<0?0:0xFF000000)),
int((i24<0?0:0xFF) | (i25<0?0:0xFF00) | (i26<0?0:0xFF0000) | (i27<0?0:0xFF000000)),
int((i28<0?0:0xFF) | (i29<0?0:0xFF00) | (i30<0?0:0xFF0000) | (i31<0?0:0xFF000000)) > ();
return _mm256_and_si256(a, mask);
}
// special case: all elements from same half
if (((m1 ^ 0xFFFF0000) & mz) == 0) {
mask = constant8i <
(i0 & 0xFF) | (i1 & 0xFF) << 8 | (i2 & 0xFF) << 16 | (i3 & 0xFF) << 24,
(i4 & 0xFF) | (i5 & 0xFF) << 8 | (i6 & 0xFF) << 16 | (i7 & 0xFF) << 24,
(i8 & 0xFF) | (i9 & 0xFF) << 8 | (i10 & 0xFF) << 16 | (i11 & 0xFF) << 24,
(i12 & 0xFF) | (i13 & 0xFF) << 8 | (i14 & 0xFF) << 16 | (i15 & 0xFF) << 24,
(i16 & 0xEF) | (i17 & 0xEF) << 8 | (i18 & 0xEF) << 16 | (i19 & 0xEF) << 24,
(i20 & 0xEF) | (i21 & 0xEF) << 8 | (i22 & 0xEF) << 16 | (i23 & 0xEF) << 24,
(i24 & 0xEF) | (i25 & 0xEF) << 8 | (i26 & 0xEF) << 16 | (i27 & 0xEF) << 24,
(i28 & 0xEF) | (i29 & 0xEF) << 8 | (i30 & 0xEF) << 16 | (i31 & 0xEF) << 24 > ();
return _mm256_shuffle_epi8(a, mask);
}
// special case: all elements from low half
if ((m1 & mz) == 0) {
mask = constant8i <
(i0 & 0xFF) | (i1 & 0xFF) << 8 | (i2 & 0xFF) << 16 | (i3 & 0xFF) << 24,
(i4 & 0xFF) | (i5 & 0xFF) << 8 | (i6 & 0xFF) << 16 | (i7 & 0xFF) << 24,
(i8 & 0xFF) | (i9 & 0xFF) << 8 | (i10 & 0xFF) << 16 | (i11 & 0xFF) << 24,
(i12 & 0xFF) | (i13 & 0xFF) << 8 | (i14 & 0xFF) << 16 | (i15 & 0xFF) << 24,
(i16 & 0xFF) | (i17 & 0xFF) << 8 | (i18 & 0xFF) << 16 | (i19 & 0xFF) << 24,
(i20 & 0xFF) | (i21 & 0xFF) << 8 | (i22 & 0xFF) << 16 | (i23 & 0xFF) << 24,
(i24 & 0xFF) | (i25 & 0xFF) << 8 | (i26 & 0xFF) << 16 | (i27 & 0xFF) << 24,
(i28 & 0xFF) | (i29 & 0xFF) << 8 | (i30 & 0xFF) << 16 | (i31 & 0xFF) << 24 > ();
t1 = _mm256_inserti128_si256(a, _mm256_castsi256_si128(a), 1); // low, low
return _mm256_shuffle_epi8(t1, mask);
}
// special case: all elements from high half
if (((m1 ^ 0xFFFFFFFF) & mz) == 0) {
mask = constant8i <
(i0 & 0xEF) | (i1 & 0xEF) << 8 | (i2 & 0xEF) << 16 | (i3 & 0xEF) << 24,
(i4 & 0xEF) | (i5 & 0xEF) << 8 | (i6 & 0xEF) << 16 | (i7 & 0xEF) << 24,
(i8 & 0xEF) | (i9 & 0xEF) << 8 | (i10 & 0xEF) << 16 | (i11 & 0xEF) << 24,
(i12 & 0xEF) | (i13 & 0xEF) << 8 | (i14 & 0xEF) << 16 | (i15 & 0xEF) << 24,
(i16 & 0xEF) | (i17 & 0xEF) << 8 | (i18 & 0xEF) << 16 | (i19 & 0xEF) << 24,
(i20 & 0xEF) | (i21 & 0xEF) << 8 | (i22 & 0xEF) << 16 | (i23 & 0xEF) << 24,
(i24 & 0xEF) | (i25 & 0xEF) << 8 | (i26 & 0xEF) << 16 | (i27 & 0xEF) << 24,
(i28 & 0xEF) | (i29 & 0xEF) << 8 | (i30 & 0xEF) << 16 | (i31 & 0xEF) << 24 > ();
t1 = _mm256_permute4x64_epi64(a, 0xEE); // high, high
return _mm256_shuffle_epi8(t1, mask);
}
// special case: all elements from opposite half
if (((m1 ^ 0x0000FFFF) & mz) == 0) {
mask = constant8i<
(i0 & 0xEF) | (i1 & 0xEF) << 8 | (i2 & 0xEF) << 16 | (i3 & 0xEF) << 24,
(i4 & 0xEF) | (i5 & 0xEF) << 8 | (i6 & 0xEF) << 16 | (i7 & 0xEF) << 24,
(i8 & 0xEF) | (i9 & 0xEF) << 8 | (i10 & 0xEF) << 16 | (i11 & 0xEF) << 24,
(i12 & 0xEF) | (i13 & 0xEF) << 8 | (i14 & 0xEF) << 16 | (i15 & 0xEF) << 24,
(i16 & 0xFF) | (i17 & 0xFF) << 8 | (i18 & 0xFF) << 16 | (i19 & 0xFF) << 24,
(i20 & 0xFF) | (i21 & 0xFF) << 8 | (i22 & 0xFF) << 16 | (i23 & 0xFF) << 24,
(i24 & 0xFF) | (i25 & 0xFF) << 8 | (i26 & 0xFF) << 16 | (i27 & 0xFF) << 24,
(i28 & 0xFF) | (i29 & 0xFF) << 8 | (i30 & 0xFF) << 16 | (i31 & 0xFF) << 24 > ();
t1 = _mm256_permute4x64_epi64(a, 0x4E); // high, low
return _mm256_shuffle_epi8(t1, mask);
}
// general case: elements from both halves
const __m256i mmsame = constant8i <
((i0 &0xF0)?0xFF:(i0 &15)) | ((i1 &0xF0)?0xFF:(i1 &15)) << 8 | ((i2 &0xF0)?0xFF:(i2 &15)) << 16 | ((i3 &0xF0)?0xFF:(i3 &15)) << 24,
((i4 &0xF0)?0xFF:(i4 &15)) | ((i5 &0xF0)?0xFF:(i5 &15)) << 8 | ((i6 &0xF0)?0xFF:(i6 &15)) << 16 | ((i7 &0xF0)?0xFF:(i7 &15)) << 24,
((i8 &0xF0)?0xFF:(i8 &15)) | ((i9 &0xF0)?0xFF:(i9 &15)) << 8 | ((i10&0xF0)?0xFF:(i10&15)) << 16 | ((i11&0xF0)?0xFF:(i11&15)) << 24,
((i12&0xF0)?0xFF:(i12&15)) | ((i13&0xF0)?0xFF:(i13&15)) << 8 | ((i14&0xF0)?0xFF:(i14&15)) << 16 | ((i15&0xF0)?0xFF:(i15&15)) << 24,
((i16&0xF0)!=0x10?0xFF:(i16&15)) | ((i17&0xF0)!=0x10?0xFF:(i17&15)) << 8 | ((i18&0xF0)!=0x10?0xFF:(i18&15)) << 16 | ((i19&0xF0)!=0x10?0xFF:(i19&15)) << 24,
((i20&0xF0)!=0x10?0xFF:(i20&15)) | ((i21&0xF0)!=0x10?0xFF:(i21&15)) << 8 | ((i22&0xF0)!=0x10?0xFF:(i22&15)) << 16 | ((i23&0xF0)!=0x10?0xFF:(i23&15)) << 24,
((i24&0xF0)!=0x10?0xFF:(i24&15)) | ((i25&0xF0)!=0x10?0xFF:(i25&15)) << 8 | ((i26&0xF0)!=0x10?0xFF:(i26&15)) << 16 | ((i27&0xF0)!=0x10?0xFF:(i27&15)) << 24,
((i28&0xF0)!=0x10?0xFF:(i28&15)) | ((i29&0xF0)!=0x10?0xFF:(i29&15)) << 8 | ((i30&0xF0)!=0x10?0xFF:(i30&15)) << 16 | ((i31&0xF0)!=0x10?0xFF:(i31&15)) << 24 > ();
const __m256i mmopposite = constant8i <
((i0 &0xF0)!=0x10?0xFF:(i0 &15)) | ((i1 &0xF0)!=0x10?0xFF:(i1 &15)) << 8 | ((i2 &0xF0)!=0x10?0xFF:(i2 &15)) << 16 | ((i3 &0xF0)!=0x10?0xFF:(i3 &15)) << 24,
((i4 &0xF0)!=0x10?0xFF:(i4 &15)) | ((i5 &0xF0)!=0x10?0xFF:(i5 &15)) << 8 | ((i6 &0xF0)!=0x10?0xFF:(i6 &15)) << 16 | ((i7 &0xF0)!=0x10?0xFF:(i7 &15)) << 24,
((i8 &0xF0)!=0x10?0xFF:(i8 &15)) | ((i9 &0xF0)!=0x10?0xFF:(i9 &15)) << 8 | ((i10&0xF0)!=0x10?0xFF:(i10&15)) << 16 | ((i11&0xF0)!=0x10?0xFF:(i11&15)) << 24,
((i12&0xF0)!=0x10?0xFF:(i12&15)) | ((i13&0xF0)!=0x10?0xFF:(i13&15)) << 8 | ((i14&0xF0)!=0x10?0xFF:(i14&15)) << 16 | ((i15&0xF0)!=0x10?0xFF:(i15&15)) << 24,
((i16&0xF0)?0xFF:(i16&15)) | ((i17&0xF0)?0xFF:(i17&15)) << 8 | ((i18&0xF0)?0xFF:(i18&15)) << 16 | ((i19&0xF0)?0xFF:(i19&15)) << 24,
((i20&0xF0)?0xFF:(i20&15)) | ((i21&0xF0)?0xFF:(i21&15)) << 8 | ((i22&0xF0)?0xFF:(i22&15)) << 16 | ((i23&0xF0)?0xFF:(i23&15)) << 24,
((i24&0xF0)?0xFF:(i24&15)) | ((i25&0xF0)?0xFF:(i25&15)) << 8 | ((i26&0xF0)?0xFF:(i26&15)) << 16 | ((i27&0xF0)?0xFF:(i27&15)) << 24,
((i28&0xF0)?0xFF:(i28&15)) | ((i29&0xF0)?0xFF:(i29&15)) << 8 | ((i30&0xF0)?0xFF:(i30&15)) << 16 | ((i31&0xF0)?0xFF:(i31&15)) << 24 > ();
__m256i topp = _mm256_permute4x64_epi64(a, 0x4E); // high, low
__m256i r1 = _mm256_shuffle_epi8(topp, mmopposite);
__m256i r2 = _mm256_shuffle_epi8(a, mmsame);
return _mm256_or_si256(r1, r2);
}
template <
int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15,
int i16, int i17, int i18, int i19, int i20, int i21, int i22, int i23,
int i24, int i25, int i26, int i27, int i28, int i29, int i30, int i31 >
static inline Vec32uc permute32uc(Vec32uc const & a) {
return Vec32uc (permute32c<i0,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,
i16,i17,i18,i19,i20,i21,i22,i23,i24,i25,i26,i27,i28,i29,i30,i31> (a));
}
/*****************************************************************************
*
* Vector blend functions
*
******************************************************************************
*
* These blend functions can mix elements from two different vectors and
* optionally set some elements to zero.
*
* The indexes are inserted as template parameters in <>. These indexes must be
* constants. Each template parameter is an index to the element you want to
* select, where higher indexes indicate an element from the second source
* vector. For example, if each vector has 8 elements, then indexes 0 - 7
* will select an element from the first vector and indexes 8 - 15 will select
* an element from the second vector. A negative index will generate zero.
*
* Example:
* Vec8i a(100,101,102,103,104,105,106,107); // a is (100, 101, 102, 103, 104, 105, 106, 107)
* Vec8i b(200,201,202,203,204,205,206,207); // b is (200, 201, 202, 203, 204, 205, 206, 207)
* Vec8i c;
* c = blend8i<1,0,9,8,7,-1,15,15> (a,b); // c is (101, 100, 201, 200, 107, 0, 207, 207)
*
* A lot of the code here is metaprogramming aiming to find the instructions
* that best fit the template parameters and instruction set. The metacode
* will be reduced out to leave only a few vector instructions in release
* mode with optimization on.
*****************************************************************************/
template <int i0, int i1, int i2, int i3>
static inline Vec4q blend4q(Vec4q const & a, Vec4q const & b) {
// Combine indexes into a single bitfield, with 8 bits for each
const int m1 = (i0 & 7) | (i1 & 7) << 8 | (i2 & 7) << 16 | (i3 & 7) << 24;
// Mask to zero out negative indexes
const int mz = (i0<0 ? 0 : 0xFF) | (i1<0 ? 0 : 0xFF) << 8 | (i2<0 ? 0 : 0xFF) << 16 | (i3<0 ? 0 : 0xFF) << 24;
// zeroing needed. An index of -0x100 means don't care
const bool dozero = ((i0|i1|i2|i3) & 0x80) != 0;
__m256i t1, mask;
// special case: 128 bit blend/permute
if (((m1 ^ 0x01000100) & 0x01010101 & mz) == 0 && (((m1 + 0x00010001) ^ (m1 >> 8)) & 0x00FF00FF & mz & mz >> 8) == 0) {
{
const int j0 = i0 >= 0 ? i0 / 2 : i1 >= 0 ? i1 / 2 : 4; // index for low 128 bits
const int j1 = i2 >= 0 ? i2 / 2 : i3 >= 0 ? i3 / 2 : 4; // index for high 128 bits
const bool partialzero = int((i0 ^ i1) | (i2 ^ i3)) < 0; // part of a 128-bit block is zeroed
switch (j0 | j1 << 4) {
case 0x00:
t1 = _mm256_inserti128_si256(a, _mm256_castsi256_si128(a), 1); break;
case 0x02:
t1 = _mm256_inserti128_si256(b, _mm256_castsi256_si128(a), 1); break;
case 0x04:
if (dozero && !partialzero) return _mm256_inserti128_si256(_mm256_setzero_si256(), _mm256_castsi256_si128(a), 1);
t1 = _mm256_inserti128_si256(a, _mm256_castsi256_si128(a), 1); break;
case 0x12:
t1 = _mm256_inserti128_si256(a, _mm256_castsi256_si128(b), 0); break;
case 0x14:
if (dozero && !partialzero) return _mm256_inserti128_si256(a,_mm_setzero_si128(), 0);
t1 = a; break;
case 0x01: case 0x10: case 0x11: // all from a
return permute4q <i0, i1, i2, i3> (a);
case 0x20:
t1 = _mm256_inserti128_si256(a, _mm256_castsi256_si128(b), 1); break;
case 0x22:
t1 = _mm256_inserti128_si256(b, _mm256_castsi256_si128(b), 1); break;
case 0x24:
if (dozero && !partialzero) return _mm256_inserti128_si256(_mm256_setzero_si256(), _mm256_castsi256_si128(b), 1);
t1 = _mm256_inserti128_si256(b, _mm256_castsi256_si128(b), 1); break;
case 0x30:
t1 = _mm256_inserti128_si256(b, _mm256_castsi256_si128(a), 0); break;
case 0x34:
if (dozero && !partialzero) return _mm256_inserti128_si256(b,_mm_setzero_si128(), 0);
t1 = b; break;
case 0x23: case 0x32: case 0x33: // all from b
return permute4q <i0^4, i1^4, i2^4, i3^4> (b);
case 0x40:
if (dozero && !partialzero) return _mm256_castsi128_si256(_mm_and_si128(_mm256_castsi256_si128(a),_mm256_castsi256_si128(a)));
t1 = a; break;
case 0x42:
if (dozero && !partialzero) return _mm256_castsi128_si256(_mm_and_si128(_mm256_castsi256_si128(b),_mm256_castsi256_si128(b)));
t1 = b; break;
case 0x44:
return _mm256_setzero_si256();
default:
t1 = _mm256_permute2x128_si256(a, b, (j0&0x0F) | (j1&0x0F) << 4);
}
}
RETURNORZERO:
if (dozero) {
// zero some elements
const __m256i maskz = constant8i <
i0 < 0 ? 0 : -1, i0 < 0 ? 0 : -1, i1 < 0 ? 0 : -1, i1 < 0 ? 0 : -1,
i2 < 0 ? 0 : -1, i2 < 0 ? 0 : -1, i3 < 0 ? 0 : -1, i3 < 0 ? 0 : -1 > ();
return _mm256_and_si256(t1, maskz);
}
return t1;
}
// special case: all from a
if ((m1 & 0x04040404 & mz) == 0) {
return permute4q <i0, i1, i2, i3> (a);
}
// special case: all from b
if ((~m1 & 0x04040404 & mz) == 0) {
return permute4q <i0^4, i1^4, i2^4, i3^4> (b);
}
// special case: blend without permute
if (((m1 ^ 0x03020100) & 0xFBFBFBFB & mz) == 0) {
mask = constant8i <
(i0 & 4) ? -1 : 0, (i0 & 4) ? -1 : 0, (i1 & 4) ? -1 : 0, (i1 & 4) ? -1 : 0,
(i2 & 4) ? -1 : 0, (i2 & 4) ? -1 : 0, (i3 & 4) ? -1 : 0, (i3 & 4) ? -1 : 0 > ();
t1 = _mm256_blendv_epi8(a, b, mask); // blend
goto RETURNORZERO;
}
// special case: shift left
if (i0 > 0 && i0 < 4 && mz == -1 && (m1 ^ ((i0 & 3) * 0x01010101 + 0x03020100)) == 0) {
t1 = _mm256_permute2x128_si256(a, b, 0x21);
if (i0 < 2) return _mm256_alignr_epi8(t1, a, (i0 & 1) * 8);
else return _mm256_alignr_epi8(b, t1, (i0 & 1) * 8);
}
// special case: shift right
if (i0 > 4 && i0 < 8 && mz == -1 && (m1 ^ 0x04040404 ^ ((i0 & 3) * 0x01010101 + 0x03020100)) == 0) {
t1 = _mm256_permute2x128_si256(b, a, 0x21);
if (i0 < 6) return _mm256_alignr_epi8(t1, b, (i0 & 1) * 8);
else return _mm256_alignr_epi8(a, t1, (i0 & 1) * 8);
}
// special case: unpack low
if (((m1 ^ 0x06020400) & mz) == 0) {
t1 = _mm256_unpacklo_epi64(a, b);
goto RETURNORZERO;
}
// special case: unpack low
if (((m1 ^ 0x02060004) & mz) == 0) {
t1 = _mm256_unpacklo_epi64(b, a);
goto RETURNORZERO;
}
// special case: unpack high
if (((m1 ^ 0x07030501) & mz) == 0) {
t1 = _mm256_unpackhi_epi64(a, b);
goto RETURNORZERO;
}
// special case: unpack high
if (((m1 ^ 0x03070105) & mz) == 0) {
t1 = _mm256_unpackhi_epi64(b, a);
goto RETURNORZERO;
}
// general case: permute and blend and possibly zero
const int blank = dozero ? -1 : -0x100; // ignore or zero
// permute and blend
__m256i ta = permute4q <
(i0 & 4) ? blank : i0, (i1 & 4) ? blank : i1, (i2 & 4) ? blank : i2, (i3 & 4) ? blank : i3 > (a);
__m256i tb = permute4q <
((i0^4) & 4) ? blank : i0^4, ((i1^4) & 4) ? blank : i1^4, ((i2^4) & 4) ? blank : i2^4, ((i3^4) & 4) ? blank : i3^4 > (b);
if (blank == -1) {
// we have zeroed, need only to OR
return _mm256_or_si256(ta, tb);
}
// no zeroing, need to blend
mask = constant8i <
(i0 & 4) ? -1 : 0, (i0 & 4) ? -1 : 0, (i1 & 4) ? -1 : 0, (i1 & 4) ? -1 : 0,
(i2 & 4) ? -1 : 0, (i2 & 4) ? -1 : 0, (i3 & 4) ? -1 : 0, (i3 & 4) ? -1 : 0 > ();
return _mm256_blendv_epi8(ta, tb, mask); // blend
}
template <int i0, int i1, int i2, int i3>
static inline Vec4uq blend4uq(Vec4uq const & a, Vec4uq const & b) {
return Vec4uq( blend4q<i0,i1,i2,i3> (a,b));
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7>
static inline Vec8i blend8i(Vec8i const & a, Vec8i const & b) {
const int ior = i0 | i1 | i2 | i3 | i4 | i5 | i6 | i7; // OR indexes
// is zeroing needed
const bool do_zero = ior < 0 && (ior & 0x80); // at least one index is negative, and not -0x100
// Combine all the indexes into a single bitfield, with 4 bits for each
const int m1 = (i0&0xF) | (i1&0xF)<<4 | (i2&0xF)<<8 | (i3&0xF)<<12 | (i4&0xF)<<16 | (i5&0xF)<<20 | (i6&0xF)<<24 | (i7&0xF)<<28;
// Mask to zero out negative indexes
const int mz = (i0<0?0:0xF) | (i1<0?0:0xF)<<4 | (i2<0?0:0xF)<<8 | (i3<0?0:0xF)<<12 | (i4<0?0:0xF)<<16 | (i5<0?0:0xF)<<20 | (i6<0?0:0xF)<<24 | (i7<0?0:0xF)<<28;
__m256i t1, mask;
if (mz == 0) return _mm256_setzero_si256(); // all zero
// special case: 64 bit blend/permute
if (((m1 ^ 0x10101010) & 0x11111111 & mz) == 0 && ((m1 ^ (m1 >> 4)) & 0x0E0E0E0E & mz & mz >> 4) == 0) {
// check if part of a 64-bit block is zeroed
const bool partialzero = int((i0^i1) | (i2^i3) | (i4^i5) | (i6^i7)) < 0;
const int blank1 = partialzero ? -0x100 : -1; // ignore if zeroing later anyway
// indexes for 64 bit blend
const int j0 = i0 >= 0 ? i0 / 2 : i1 >= 0 ? i1 / 2 : blank1;
const int j1 = i2 >= 0 ? i2 / 2 : i3 >= 0 ? i3 / 2 : blank1;
const int j2 = i4 >= 0 ? i4 / 2 : i5 >= 0 ? i5 / 2 : blank1;
const int j3 = i6 >= 0 ? i6 / 2 : i7 >= 0 ? i7 / 2 : blank1;
// 64-bit blend and permute
t1 = blend4q<j0,j1,j2,j3>(Vec4q(a), Vec4q(b));
if (partialzero && do_zero) {
// zero some elements
mask = constant8i< i0 < 0 ? 0 : -1, i1 < 0 ? 0 : -1, i2 < 0 ? 0 : -1, i3 < 0 ? 0 : -1,
i4 < 0 ? 0 : -1, i5 < 0 ? 0 : -1, i6 < 0 ? 0 : -1, i7 < 0 ? 0 : -1 > ();
return _mm256_and_si256(t1, mask);
}
return t1;
}
if ((m1 & 0x88888888 & mz) == 0) {
// all from a
return permute8i<i0, i1, i2, i3, i4, i5, i6, i7> (a);
}
if (((m1 ^ 0x88888888) & 0x88888888 & mz) == 0) {
// all from b
return permute8i<i0&~8, i1&~8, i2&~8, i3&~8, i4&~8, i5&~8, i6&~8, i7&~8> (b);
}
if ((((m1 & 0x77777777) ^ 0x76543210) & mz) == 0) {
// blend and zero, no permute
mask = constant8i<(i0&8)?0:-1, (i1&8)?0:-1, (i2&8)?0:-1, (i3&8)?0:-1, (i4&8)?0:-1, (i5&8)?0:-1, (i6&8)?0:-1, (i7&8)?0:-1> ();
t1 = select(mask, a, b);
if (!do_zero) return t1;
// zero some elements
mask = constant8i< (i0<0&&(i0&8)) ? 0 : -1, (i1<0&&(i1&8)) ? 0 : -1, (i2<0&&(i2&8)) ? 0 : -1, (i3<0&&(i3&8)) ? 0 : -1,
(i4<0&&(i4&8)) ? 0 : -1, (i5<0&&(i5&8)) ? 0 : -1, (i6<0&&(i6&8)) ? 0 : -1, (i7<0&&(i7&8)) ? 0 : -1 > ();
return _mm256_and_si256(t1, mask);
}
// special case: shift left
if (i0 > 0 && i0 < 8 && mz == -1 && (m1 ^ ((i0 & 7) * 0x11111111u + 0x76543210u)) == 0) {
t1 = _mm256_permute2x128_si256(a, b, 0x21);
if (i0 < 4) return _mm256_alignr_epi8(t1, a, (i0 & 3) * 4);
else return _mm256_alignr_epi8(b, t1, (i0 & 3) * 4);
}
// special case: shift right
if (i0 > 8 && i0 < 16 && mz == -1 && (m1 ^ 0x88888888 ^ ((i0 & 7) * 0x11111111u + 0x76543210u)) == 0) {
t1 = _mm256_permute2x128_si256(b, a, 0x21);
if (i0 < 12) return _mm256_alignr_epi8(t1, b, (i0 & 3) * 4);
else return _mm256_alignr_epi8(a, t1, (i0 & 3) * 4);
}
// general case: permute and blend and possible zero
const int blank = do_zero ? -1 : -0x100; // ignore or zero
Vec8i ta = permute8i <
(uint32_t)i0 < 8 ? i0 : blank,
(uint32_t)i1 < 8 ? i1 : blank,
(uint32_t)i2 < 8 ? i2 : blank,
(uint32_t)i3 < 8 ? i3 : blank,
(uint32_t)i4 < 8 ? i4 : blank,
(uint32_t)i5 < 8 ? i5 : blank,
(uint32_t)i6 < 8 ? i6 : blank,
(uint32_t)i7 < 8 ? i7 : blank > (a);
Vec8i tb = permute8i <
(uint32_t)(i0^8) < 8 ? (i0^8) : blank,
(uint32_t)(i1^8) < 8 ? (i1^8) : blank,
(uint32_t)(i2^8) < 8 ? (i2^8) : blank,
(uint32_t)(i3^8) < 8 ? (i3^8) : blank,
(uint32_t)(i4^8) < 8 ? (i4^8) : blank,
(uint32_t)(i5^8) < 8 ? (i5^8) : blank,
(uint32_t)(i6^8) < 8 ? (i6^8) : blank,
(uint32_t)(i7^8) < 8 ? (i7^8) : blank > (b);
if (blank == -1) {
return _mm256_or_si256(ta, tb);
}
// no zeroing, need to blend
const int maskb = ((i0 >> 3) & 1) | ((i1 >> 2) & 2) | ((i2 >> 1) & 4) | (i3 & 8) |
((i4 << 1) & 0x10) | ((i5 << 2) & 0x20) | ((i6 << 3) & 0x40) | ((i7 << 4) & 0x80);
return _mm256_blend_epi32(ta, tb, maskb); // blend
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7>
static inline Vec8ui blend8ui(Vec8ui const & a, Vec8ui const & b) {
return Vec8ui( blend8i<i0,i1,i2,i3,i4,i5,i6,i7> (a,b));
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15 >
static inline Vec16s blend16s(Vec16s const & a, Vec16s const & b) {
// collect bit 4 of each index
const int m1 =
(i0 &16)>>4 | (i1 &16)>>3 | (i2 &16)>>2 | (i3 &16)>>1 | (i4 &16) | (i5 &16)<<1 | (i6 &16)<<2 | (i7 &16)<<3 |
(i8 &16)<<4 | (i9 &16)<<5 | (i10&16)<<6 | (i11&16)<<7 | (i12&16)<<8 | (i13&16)<<9 | (i14&16)<<10 | (i15&16)<<11 ;
// check which elements to set to zero
const int mz = 0x0000FFFF ^ (
(i0 <0) | (i1 <0)<<1 | (i2 <0)<<2 | (i3 <0)<<3 | (i4 <0)<<4 | (i5 <0)<<5 | (i6 <0)<<6 | (i7 <0)<<7 |
(i8 <0)<<8 | (i9 <0)<<9 | (i10<0)<<10 | (i11<0)<<11 | (i12<0)<<12 | (i13<0)<<13 | (i14<0)<<14 | (i15<0)<<15 );
__m256i t1, mask;
// special case: all zero
if (mz == 0) return _mm256_setzero_si256();
// special case: all from a
if ((m1 & mz) == 0) {
return permute16s<i0,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15> (a);
}
// special case: all from b
if (((m1 ^ 0xFFFF) & mz) == 0) {
return permute16s<i0^16,i1^16,i2^16,i3^16,i4^16,i5^16,i6^16,i7^16,i8^16,i9^16,i10^16,i11^16,i12^16,i13^16,i14^16,i15^16 > (b);
}
// special case: blend without permute
if ((i0 <0||(i0 &15)== 0) && (i1 <0||(i1 &15)== 1) && (i2 <0||(i2 &15)== 2) && (i3 <0||(i3 &15)== 3) &&
(i4 <0||(i4 &15)== 4) && (i5 <0||(i5 &15)== 5) && (i6 <0||(i6 &15)== 6) && (i7 <0||(i7 &15)== 7) &&
(i8 <0||(i8 &15)== 8) && (i9 <0||(i9 &15)== 9) && (i10<0||(i10&15)==10) && (i11<0||(i11&15)==11) &&
(i12<0||(i12&15)==12) && (i13<0||(i13&15)==13) && (i14<0||(i14&15)==14) && (i15<0||(i15&15)==15)) {
mask = constant8i <
int(((i0 & 16) ? 0xFFFF : 0) | ((i1 & 16) ? 0xFFFF0000 : 0)),
int(((i2 & 16) ? 0xFFFF : 0) | ((i3 & 16) ? 0xFFFF0000 : 0)),
int(((i4 & 16) ? 0xFFFF : 0) | ((i5 & 16) ? 0xFFFF0000 : 0)),
int(((i6 & 16) ? 0xFFFF : 0) | ((i7 & 16) ? 0xFFFF0000 : 0)),
int(((i8 & 16) ? 0xFFFF : 0) | ((i9 & 16) ? 0xFFFF0000 : 0)),
int(((i10& 16) ? 0xFFFF : 0) | ((i11& 16) ? 0xFFFF0000 : 0)),
int(((i12& 16) ? 0xFFFF : 0) | ((i13& 16) ? 0xFFFF0000 : 0)),
int(((i14& 16) ? 0xFFFF : 0) | ((i15& 16) ? 0xFFFF0000 : 0)) > ();
t1 = _mm256_blendv_epi8(a, b, mask); // blend
if (mz != 0xFFFF) {
// zero some elements
mask = constant8i <
int((i0 < 0 ? 0 : 0xFFFF) | (i1 < 0 ? 0 : 0xFFFF0000)),
int((i2 < 0 ? 0 : 0xFFFF) | (i3 < 0 ? 0 : 0xFFFF0000)),
int((i4 < 0 ? 0 : 0xFFFF) | (i5 < 0 ? 0 : 0xFFFF0000)),
int((i6 < 0 ? 0 : 0xFFFF) | (i7 < 0 ? 0 : 0xFFFF0000)),
int((i8 < 0 ? 0 : 0xFFFF) | (i9 < 0 ? 0 : 0xFFFF0000)),
int((i10 < 0 ? 0 : 0xFFFF) | (i11 < 0 ? 0 : 0xFFFF0000)),
int((i12 < 0 ? 0 : 0xFFFF) | (i13 < 0 ? 0 : 0xFFFF0000)),
int((i14 < 0 ? 0 : 0xFFFF) | (i15 < 0 ? 0 : 0xFFFF0000)) > ();
return _mm256_and_si256(t1, mask);
}
return t1;
}
// special case: shift left
const int slb = i0 > 0 ? i0 : i15 - 15;
if (slb > 0 && slb < 16
&& (i0==slb+ 0||i0<0) && (i1==slb+ 1||i1<0) && (i2 ==slb+ 2||i2 <0) && (i3 ==slb+ 3||i3 <0) && (i4 ==slb+ 4||i4 <0) && (i5 ==slb+ 5||i5 <0) && (i6 ==slb+ 6||i6 <0) && (i7 ==slb+ 7||i7 <0)
&& (i8==slb+ 8||i8<0) && (i9==slb+ 9||i9<0) && (i10==slb+10||i10<0) && (i11==slb+11||i11<0) && (i12==slb+12||i12<0) && (i13==slb+13||i13<0) && (i14==slb+14||i14<0) && (i15==slb+15||i15<0)) {
t1 = _mm256_permute2x128_si256(a, b, 0x21);
if (slb < 8) t1 = _mm256_alignr_epi8(t1, a, (slb & 7) * 2);
else t1 = _mm256_alignr_epi8(b, t1, (slb & 7) * 2);
if (mz != 0xFFFF) {
// zero some elements
mask = constant8i <
int((i0 < 0 ? 0 : 0xFFFF) | (i1 < 0 ? 0 : 0xFFFF0000)),
int((i2 < 0 ? 0 : 0xFFFF) | (i3 < 0 ? 0 : 0xFFFF0000)),
int((i4 < 0 ? 0 : 0xFFFF) | (i5 < 0 ? 0 : 0xFFFF0000)),
int((i6 < 0 ? 0 : 0xFFFF) | (i7 < 0 ? 0 : 0xFFFF0000)),
int((i8 < 0 ? 0 : 0xFFFF) | (i9 < 0 ? 0 : 0xFFFF0000)),
int((i10 < 0 ? 0 : 0xFFFF) | (i11 < 0 ? 0 : 0xFFFF0000)),
int((i12 < 0 ? 0 : 0xFFFF) | (i13 < 0 ? 0 : 0xFFFF0000)),
int((i14 < 0 ? 0 : 0xFFFF) | (i15 < 0 ? 0 : 0xFFFF0000)) > ();
return _mm256_and_si256(t1, mask);
}
return t1;
}
// special case: shift right
const int srb = i0 > 0 ? (i0^16) : (i15^16) - 15;
if (srb > 0 && srb < 16
&& ((i0 ^16)==srb+ 0||i0 <0) && ((i1 ^16)==srb+ 1||i1 <0) && ((i2 ^16)==srb+ 2||i2 <0) && ((i3 ^16)==srb+ 3||i3 <0) && ((i4 ^16)==srb+ 4||i4 <0) && ((i5 ^16)==srb+ 5||i5 <0) && ((i6 ^16)==srb+ 6||i6 <0) && ((i7 ^16)==srb+ 7||i7 <0)
&& ((i8 ^16)==srb+ 8||i8 <0) && ((i9 ^16)==srb+ 9||i9 <0) && ((i10^16)==srb+10||i10<0) && ((i11^16)==srb+11||i11<0) && ((i12^16)==srb+12||i12<0) && ((i13^16)==srb+13||i13<0) && ((i14^16)==srb+14||i14<0) && ((i15^16)==srb+15||i15<0)) {
t1 = _mm256_permute2x128_si256(b, a, 0x21);
if (srb < 8) t1 = _mm256_alignr_epi8(t1, b, (srb & 7) * 2);
else t1 = _mm256_alignr_epi8(a, t1, (srb & 7) * 2);
if (mz != 0xFFFF) {
// zero some elements
mask = constant8i <
int((i0 < 0 ? 0 : 0xFFFF) | (i1 < 0 ? 0 : 0xFFFF0000)),
int((i2 < 0 ? 0 : 0xFFFF) | (i3 < 0 ? 0 : 0xFFFF0000)),
int((i4 < 0 ? 0 : 0xFFFF) | (i5 < 0 ? 0 : 0xFFFF0000)),
int((i6 < 0 ? 0 : 0xFFFF) | (i7 < 0 ? 0 : 0xFFFF0000)),
int((i8 < 0 ? 0 : 0xFFFF) | (i9 < 0 ? 0 : 0xFFFF0000)),
int((i10 < 0 ? 0 : 0xFFFF) | (i11 < 0 ? 0 : 0xFFFF0000)),
int((i12 < 0 ? 0 : 0xFFFF) | (i13 < 0 ? 0 : 0xFFFF0000)),
int((i14 < 0 ? 0 : 0xFFFF) | (i15 < 0 ? 0 : 0xFFFF0000)) > ();
return _mm256_and_si256(t1, mask);
}
return t1;
}
// general case: permute and blend and possibly zero
const int blank = (mz == 0xFFFF) ? -0x100 : -1; // ignore or zero
// permute and blend
__m256i ta = permute16s <
(i0 &16)?blank:i0 , (i1 &16)?blank:i1 , (i2 &16)?blank:i2 , (i3 &16)?blank:i3 ,
(i4 &16)?blank:i4 , (i5 &16)?blank:i5 , (i6 &16)?blank:i6 , (i7 &16)?blank:i7 ,
(i8 &16)?blank:i8 , (i9 &16)?blank:i9 , (i10&16)?blank:i10, (i11&16)?blank:i11,
(i12&16)?blank:i12, (i13&16)?blank:i13, (i14&16)?blank:i14, (i15&16)?blank:i15 > (a);
__m256i tb = permute16s <
((i0 ^16)&16)?blank:i0 ^16, ((i1 ^16)&16)?blank:i1 ^16, ((i2 ^16)&16)?blank:i2 ^16, ((i3 ^16)&16)?blank:i3 ^16,
((i4 ^16)&16)?blank:i4 ^16, ((i5 ^16)&16)?blank:i5 ^16, ((i6 ^16)&16)?blank:i6 ^16, ((i7 ^16)&16)?blank:i7 ^16,
((i8 ^16)&16)?blank:i8 ^16, ((i9 ^16)&16)?blank:i9 ^16, ((i10^16)&16)?blank:i10^16, ((i11^16)&16)?blank:i11^16,
((i12^16)&16)?blank:i12^16, ((i13^16)&16)?blank:i13^16, ((i14^16)&16)?blank:i14^16, ((i15^16)&16)?blank:i15^16 > (b);
if (blank == -1) {
// we have zeroed, need only to OR
return _mm256_or_si256(ta, tb);
}
// no zeroing, need to blend
mask = constant8i <
int(((i0 & 16) ? 0xFFFF : 0) | ((i1 & 16) ? 0xFFFF0000 : 0)),
int(((i2 & 16) ? 0xFFFF : 0) | ((i3 & 16) ? 0xFFFF0000 : 0)),
int(((i4 & 16) ? 0xFFFF : 0) | ((i5 & 16) ? 0xFFFF0000 : 0)),
int(((i6 & 16) ? 0xFFFF : 0) | ((i7 & 16) ? 0xFFFF0000 : 0)),
int(((i8 & 16) ? 0xFFFF : 0) | ((i9 & 16) ? 0xFFFF0000 : 0)),
int(((i10& 16) ? 0xFFFF : 0) | ((i11& 16) ? 0xFFFF0000 : 0)),
int(((i12& 16) ? 0xFFFF : 0) | ((i13& 16) ? 0xFFFF0000 : 0)),
int(((i14& 16) ? 0xFFFF : 0) | ((i15& 16) ? 0xFFFF0000 : 0)) > ();
return _mm256_blendv_epi8(ta, tb, mask); // blend
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15 >
static inline Vec16us blend16us(Vec16us const & a, Vec16us const & b) {
return Vec16us( blend16s<i0,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15> (a,b));
}
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15,
int i16, int i17, int i18, int i19, int i20, int i21, int i22, int i23,
int i24, int i25, int i26, int i27, int i28, int i29, int i30, int i31 >
static inline Vec32c blend32c(Vec32c const & a, Vec32c const & b) {
// collect bit 5 of each index
const int m1 =
(i0 &32)>>5 | (i1 &32)>>4 | (i2 &32)>>3 | (i3 &32)>>2 | (i4 &32)>>1 | (i5 &32) | (i6 &32)<<1 | (i7 &32)<<2 |
(i8 &32)<<3 | (i9 &32)<<4 | (i10&32)<<5 | (i11&32)<<6 | (i12&32)<<7 | (i13&32)<<8 | (i14&32)<<9 | (i15&32)<<10 |
(i16&32)<<11 | (i17&32)<<12 | (i18&32)<<13 | (i19&32)<<14 | (i20&32)<<15 | (i21&32)<<16 | (i22&32)<<17 | (i23&32)<<18 |
(i24&32)<<19 | (i25&32)<<20 | (i26&32)<<21 | (i27&32)<<22 | (i28&32)<<23 | (i29&32)<<24 | (i30&32)<<25 | (i31&32)<<26 ;
// check which elements to set to zero
const int mz = ~ (
(i0 <0) | (i1 <0)<<1 | (i2 <0)<<2 | (i3 <0)<<3 | (i4 <0)<<4 | (i5 <0)<<5 | (i6 <0)<<6 | (i7 <0)<<7 |
(i8 <0)<<8 | (i9 <0)<<9 | (i10<0)<<10 | (i11<0)<<11 | (i12<0)<<12 | (i13<0)<<13 | (i14<0)<<14 | (i15<0)<<15 |
(i16<0)<<16 | (i17<0)<<17 | (i18<0)<<18 | (i19<0)<<19 | (i20<0)<<20 | (i21<0)<<21 | (i22<0)<<22 | (i23<0)<<23 |
(i24<0)<<24 | (i25<0)<<25 | (i26<0)<<26 | (i27<0)<<27 | (i28<0)<<28 | (i29<0)<<29 | (i30<0)<<30 | (i31<0)<<31 );
__m256i t1, mask;
// special case: all zero
if (mz == 0) return _mm256_setzero_si256();
// special case: all from a
if ((m1 & mz) == 0) {
return permute32c<i0,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,
i16,i17,i18,i19,i20,i21,i22,i23,i24,i25,i26,i27,i28,i29,i30,i31> (a);
}
// special case: all from b
if ((~m1 & mz) == 0) {
return permute32c<i0^32,i1^32,i2^32,i3^32,i4^32,i5^32,i6^32,i7^32,i8^32,i9^32,i10^32,i11^32,i12^32,i13^32,i14^32,i15^32,
i16^32,i17^32,i18^32,i19^32,i20^32,i21^32,i22^32,i23^32,i24^32,i25^32,i26^32,i27^32,i28^32,i29^32,i30^32,i31^32> (b);
}
// special case: blend without permute
if ((i0 <0||(i0 &31)== 0) && (i1 <0||(i1 &31)== 1) && (i2 <0||(i2 &31)== 2) && (i3 <0||(i3 &31)== 3) &&
(i4 <0||(i4 &31)== 4) && (i5 <0||(i5 &31)== 5) && (i6 <0||(i6 &31)== 6) && (i7 <0||(i7 &31)== 7) &&
(i8 <0||(i8 &31)== 8) && (i9 <0||(i9 &31)== 9) && (i10<0||(i10&31)==10) && (i11<0||(i11&31)==11) &&
(i12<0||(i12&31)==12) && (i13<0||(i13&31)==13) && (i14<0||(i14&31)==14) && (i15<0||(i15&31)==15) &&
(i16<0||(i16&31)==16) && (i17<0||(i17&31)==17) && (i18<0||(i18&31)==18) && (i19<0||(i19&31)==19) &&
(i20<0||(i20&31)==20) && (i21<0||(i21&31)==21) && (i22<0||(i22&31)==22) && (i23<0||(i23&31)==23) &&
(i24<0||(i24&31)==24) && (i25<0||(i25&31)==25) && (i26<0||(i26&31)==26) && (i27<0||(i27&31)==27) &&
(i28<0||(i28&31)==28) && (i29<0||(i29&31)==29) && (i30<0||(i30&31)==30) && (i31<0||(i31&31)==31) ) {
mask = constant8i <
int(((i0 <<2)&0x80) | ((i1 <<10)&0x8000) | ((i2 <<18)&0x800000) | (uint32_t(i3 <<26)&0x80000000)) ,
int(((i4 <<2)&0x80) | ((i5 <<10)&0x8000) | ((i6 <<18)&0x800000) | (uint32_t(i7 <<26)&0x80000000)) ,
int(((i8 <<2)&0x80) | ((i9 <<10)&0x8000) | ((i10<<18)&0x800000) | (uint32_t(i11<<26)&0x80000000)) ,
int(((i12<<2)&0x80) | ((i13<<10)&0x8000) | ((i14<<18)&0x800000) | (uint32_t(i15<<26)&0x80000000)) ,
int(((i16<<2)&0x80) | ((i17<<10)&0x8000) | ((i18<<18)&0x800000) | (uint32_t(i19<<26)&0x80000000)) ,
int(((i20<<2)&0x80) | ((i21<<10)&0x8000) | ((i22<<18)&0x800000) | (uint32_t(i23<<26)&0x80000000)) ,
int(((i24<<2)&0x80) | ((i25<<10)&0x8000) | ((i26<<18)&0x800000) | (uint32_t(i27<<26)&0x80000000)) ,
int(((i28<<2)&0x80) | ((i29<<10)&0x8000) | ((i30<<18)&0x800000) | (uint32_t(i31<<26)&0x80000000)) > ();
t1 = _mm256_blendv_epi8(a, b, mask); // blend
if (mz != -1) {
// zero some elements
const __m256i maskz = constant8i <
int((i0 <0?0:0xFF) | (i1 <0?0:0xFF00) | (i2 <0?0:0xFF0000) | (i3 <0?0:0xFF000000)),
int((i4 <0?0:0xFF) | (i5 <0?0:0xFF00) | (i6 <0?0:0xFF0000) | (i7 <0?0:0xFF000000)),
int((i8 <0?0:0xFF) | (i9 <0?0:0xFF00) | (i10<0?0:0xFF0000) | (i11<0?0:0xFF000000)),
int((i12<0?0:0xFF) | (i13<0?0:0xFF00) | (i14<0?0:0xFF0000) | (i15<0?0:0xFF000000)),
int((i16<0?0:0xFF) | (i17<0?0:0xFF00) | (i18<0?0:0xFF0000) | (i19<0?0:0xFF000000)),
int((i20<0?0:0xFF) | (i21<0?0:0xFF00) | (i22<0?0:0xFF0000) | (i23<0?0:0xFF000000)),
int((i24<0?0:0xFF) | (i25<0?0:0xFF00) | (i26<0?0:0xFF0000) | (i27<0?0:0xFF000000)),
int((i28<0?0:0xFF) | (i29<0?0:0xFF00) | (i30<0?0:0xFF0000) | (i31<0?0:0xFF000000)) > ();
return _mm256_and_si256(t1, maskz);
}
return t1;
}
// special case: shift left
const int slb = i0 > 0 ? i0 : i31 - 31;
if (slb > 0 && slb < 32
&& (i0 ==slb+ 0||i0 <0) && (i1 ==slb+ 1||i1 <0) && (i2 ==slb+ 2||i2 <0) && (i3 ==slb+ 3||i3 <0)
&& (i4 ==slb+ 4||i4 <0) && (i5 ==slb+ 5||i5 <0) && (i6 ==slb+ 6||i6 <0) && (i7 ==slb+ 7||i7 <0)
&& (i8 ==slb+ 8||i8 <0) && (i9 ==slb+ 9||i9 <0) && (i10==slb+10||i10<0) && (i11==slb+11||i11<0)
&& (i12==slb+12||i12<0) && (i13==slb+13||i13<0) && (i14==slb+14||i14<0) && (i15==slb+15||i15<0)
&& (i16==slb+16||i16<0) && (i17==slb+17||i17<0) && (i18==slb+18||i18<0) && (i19==slb+19||i19<0)
&& (i20==slb+20||i20<0) && (i21==slb+21||i21<0) && (i22==slb+22||i22<0) && (i23==slb+23||i23<0)
&& (i24==slb+24||i24<0) && (i25==slb+25||i25<0) && (i26==slb+26||i26<0) && (i27==slb+27||i27<0)
&& (i28==slb+28||i28<0) && (i29==slb+29||i29<0) && (i30==slb+30||i30<0) && (i31==slb+31||i31<0)) {
t1 = _mm256_permute2x128_si256(a, b, 0x21);
if (slb < 16) t1 = _mm256_alignr_epi8(t1, a, slb & 15);
else t1 = _mm256_alignr_epi8(b, t1, slb & 15);
if (mz != -1) {
// zero some elements
const __m256i maskz = constant8i <
int((i0 <0?0:0xFF) | (i1 <0?0:0xFF00) | (i2 <0?0:0xFF0000) | (i3 <0?0:0xFF000000)),
int((i4 <0?0:0xFF) | (i5 <0?0:0xFF00) | (i6 <0?0:0xFF0000) | (i7 <0?0:0xFF000000)),
int((i8 <0?0:0xFF) | (i9 <0?0:0xFF00) | (i10<0?0:0xFF0000) | (i11<0?0:0xFF000000)),
int((i12<0?0:0xFF) | (i13<0?0:0xFF00) | (i14<0?0:0xFF0000) | (i15<0?0:0xFF000000)),
int((i16<0?0:0xFF) | (i17<0?0:0xFF00) | (i18<0?0:0xFF0000) | (i19<0?0:0xFF000000)),
int((i20<0?0:0xFF) | (i21<0?0:0xFF00) | (i22<0?0:0xFF0000) | (i23<0?0:0xFF000000)),
int((i24<0?0:0xFF) | (i25<0?0:0xFF00) | (i26<0?0:0xFF0000) | (i27<0?0:0xFF000000)),
int((i28<0?0:0xFF) | (i29<0?0:0xFF00) | (i30<0?0:0xFF0000) | (i31<0?0:0xFF000000)) > ();
return _mm256_and_si256(t1, maskz);
}
return t1;
}
// special case: shift right
const int srb = i0 > 0 ? (i0^32) : (i31^32) - 31;
if (srb > 0 && srb < 32
&& ((i0 ^32)==srb+ 0||i0 <0) && ((i1 ^32)==srb+ 1||i1 <0) && ((i2 ^32)==srb+ 2||i2 <0) && ((i3 ^32)==srb+ 3||i3 <0)
&& ((i4 ^32)==srb+ 4||i4 <0) && ((i5 ^32)==srb+ 5||i5 <0) && ((i6 ^32)==srb+ 6||i6 <0) && ((i7 ^32)==srb+ 7||i7 <0)
&& ((i8 ^32)==srb+ 8||i8 <0) && ((i9 ^32)==srb+ 9||i9 <0) && ((i10^32)==srb+10||i10<0) && ((i11^32)==srb+11||i11<0)
&& ((i12^32)==srb+12||i12<0) && ((i13^32)==srb+13||i13<0) && ((i14^32)==srb+14||i14<0) && ((i15^32)==srb+15||i15<0)
&& ((i16^32)==srb+16||i16<0) && ((i17^32)==srb+17||i17<0) && ((i18^32)==srb+18||i18<0) && ((i19^32)==srb+19||i19<0)
&& ((i20^32)==srb+20||i20<0) && ((i21^32)==srb+21||i21<0) && ((i22^32)==srb+22||i22<0) && ((i23^32)==srb+23||i23<0)
&& ((i24^32)==srb+24||i24<0) && ((i25^32)==srb+25||i25<0) && ((i26^32)==srb+26||i26<0) && ((i27^32)==srb+27||i27<0)
&& ((i28^32)==srb+28||i28<0) && ((i29^32)==srb+29||i29<0) && ((i30^32)==srb+30||i30<0) && ((i31^32)==srb+31||i31<0)) {
t1 = _mm256_permute2x128_si256(b, a, 0x21);
if (srb < 16) t1 = _mm256_alignr_epi8(t1, b, srb & 15);
else t1 = _mm256_alignr_epi8(a, t1, srb & 15);
if (mz != -1) {
// zero some elements
const __m256i maskz = constant8i <
int((i0 <0?0:0xFF) | (i1 <0?0:0xFF00) | (i2 <0?0:0xFF0000) | (i3 <0?0:0xFF000000)),
int((i4 <0?0:0xFF) | (i5 <0?0:0xFF00) | (i6 <0?0:0xFF0000) | (i7 <0?0:0xFF000000)),
int((i8 <0?0:0xFF) | (i9 <0?0:0xFF00) | (i10<0?0:0xFF0000) | (i11<0?0:0xFF000000)),
int((i12<0?0:0xFF) | (i13<0?0:0xFF00) | (i14<0?0:0xFF0000) | (i15<0?0:0xFF000000)),
int((i16<0?0:0xFF) | (i17<0?0:0xFF00) | (i18<0?0:0xFF0000) | (i19<0?0:0xFF000000)),
int((i20<0?0:0xFF) | (i21<0?0:0xFF00) | (i22<0?0:0xFF0000) | (i23<0?0:0xFF000000)),
int((i24<0?0:0xFF) | (i25<0?0:0xFF00) | (i26<0?0:0xFF0000) | (i27<0?0:0xFF000000)),
int((i28<0?0:0xFF) | (i29<0?0:0xFF00) | (i30<0?0:0xFF0000) | (i31<0?0:0xFF000000)) > ();
return _mm256_and_si256(t1, maskz);
}
return t1;
}
// general case: permute and blend and possible zero
const int blank = (mz == -1) ? -0x100 : -1; // ignore or zero
// permute and blend
__m256i ta = permute32c <
(i0 &32)?blank:i0 , (i1 &32)?blank:i1 , (i2 &32)?blank:i2 , (i3 &32)?blank:i3 ,
(i4 &32)?blank:i4 , (i5 &32)?blank:i5 , (i6 &32)?blank:i6 , (i7 &32)?blank:i7 ,
(i8 &32)?blank:i8 , (i9 &32)?blank:i9 , (i10&32)?blank:i10, (i11&32)?blank:i11,
(i12&32)?blank:i12, (i13&32)?blank:i13, (i14&32)?blank:i14, (i15&32)?blank:i15,
(i16&32)?blank:i16, (i17&32)?blank:i17, (i18&32)?blank:i18, (i19&32)?blank:i19,
(i20&32)?blank:i20, (i21&32)?blank:i21, (i22&32)?blank:i22, (i23&32)?blank:i23,
(i24&32)?blank:i24, (i25&32)?blank:i25, (i26&32)?blank:i26, (i27&32)?blank:i27,
(i28&32)?blank:i28, (i29&32)?blank:i29, (i30&32)?blank:i30, (i31&32)?blank:i31 > (a);
__m256i tb = permute32c <
((i0 ^32)&32)?blank:i0 ^32, ((i1 ^32)&32)?blank:i1 ^32, ((i2 ^32)&32)?blank:i2 ^32, ((i3 ^32)&32)?blank:i3 ^32,
((i4 ^32)&32)?blank:i4 ^32, ((i5 ^32)&32)?blank:i5 ^32, ((i6 ^32)&32)?blank:i6 ^32, ((i7 ^32)&32)?blank:i7 ^32,
((i8 ^32)&32)?blank:i8 ^32, ((i9 ^32)&32)?blank:i9 ^32, ((i10^32)&32)?blank:i10^32, ((i11^32)&32)?blank:i11^32,
((i12^32)&32)?blank:i12^32, ((i13^32)&32)?blank:i13^32, ((i14^32)&32)?blank:i14^32, ((i15^32)&32)?blank:i15^32,
((i16^32)&32)?blank:i16^32, ((i17^32)&32)?blank:i17^32, ((i18^32)&32)?blank:i18^32, ((i19^32)&32)?blank:i19^32,
((i20^32)&32)?blank:i20^32, ((i21^32)&32)?blank:i21^32, ((i22^32)&32)?blank:i22^32, ((i23^32)&32)?blank:i23^32,
((i24^32)&32)?blank:i24^32, ((i25^32)&32)?blank:i25^32, ((i26^32)&32)?blank:i26^32, ((i27^32)&32)?blank:i27^32,
((i28^32)&32)?blank:i28^32, ((i29^32)&32)?blank:i29^32, ((i30^32)&32)?blank:i30^32, ((i31^32)&32)?blank:i31^32 > (b);
if (blank == -1) {
// we have zeroed, need only to OR
return _mm256_or_si256(ta, tb);
}
// no zeroing, need to blend
mask = constant8i <
int(((i0 <<2)&0x80) | ((i1 <<10)&0x8000) | ((i2 <<18)&0x800000) | (uint32_t(i3 <<26)&0x80000000)) ,
int(((i4 <<2)&0x80) | ((i5 <<10)&0x8000) | ((i6 <<18)&0x800000) | (uint32_t(i7 <<26)&0x80000000)) ,
int(((i8 <<2)&0x80) | ((i9 <<10)&0x8000) | ((i10<<18)&0x800000) | (uint32_t(i11<<26)&0x80000000)) ,
int(((i12<<2)&0x80) | ((i13<<10)&0x8000) | ((i14<<18)&0x800000) | (uint32_t(i15<<26)&0x80000000)) ,
int(((i16<<2)&0x80) | ((i17<<10)&0x8000) | ((i18<<18)&0x800000) | (uint32_t(i19<<26)&0x80000000)) ,
int(((i20<<2)&0x80) | ((i21<<10)&0x8000) | ((i22<<18)&0x800000) | (uint32_t(i23<<26)&0x80000000)) ,
int(((i24<<2)&0x80) | ((i25<<10)&0x8000) | ((i26<<18)&0x800000) | (uint32_t(i27<<26)&0x80000000)) ,
int(((i28<<2)&0x80) | ((i29<<10)&0x8000) | ((i30<<18)&0x800000) | (uint32_t(i31<<26)&0x80000000)) > ();
return _mm256_blendv_epi8(ta, tb, mask); // blend
}
template <
int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7,
int i8, int i9, int i10, int i11, int i12, int i13, int i14, int i15,
int i16, int i17, int i18, int i19, int i20, int i21, int i22, int i23,
int i24, int i25, int i26, int i27, int i28, int i29, int i30, int i31 >
static inline Vec32uc blend32uc(Vec32uc const & a, Vec32uc const & b) {
return Vec32uc (blend32c<i0,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,
i16,i17,i18,i19,i20,i21,i22,i23,i24,i25,i26,i27,i28,i29,i30,i31> (a, b));
}
/*****************************************************************************
*
* Vector lookup functions
*
******************************************************************************
*
* These functions use vector elements as indexes into a table.
* The table is given as one or more vectors or as an array.
*
* This can be used for several purposes:
* - table lookup
* - permute or blend with variable indexes
* - blend from more than two sources
* - gather non-contiguous data
*
* An index out of range may produce any value - the actual value produced is
* implementation dependent and may be different for different instruction
* sets. An index out of range does not produce an error message or exception.
*
* Example:
* Vec8i a(2,0,0,6,4,3,5,0); // index a is ( 2, 0, 0, 6, 4, 3, 5, 0)
* Vec8i b(100,101,102,103,104,105,106,107); // table b is (100, 101, 102, 103, 104, 105, 106, 107)
* Vec8i c;
* c = lookup8 (a,b); // c is (102, 100, 100, 106, 104, 103, 105, 100)
*
*****************************************************************************/
static inline Vec32c lookup32(Vec32c const & index, Vec32c const & table) {
#ifdef __XOP__ // AMD XOP instruction set. Use VPPERM
Vec16c t0 = _mm_perm_epi8(table.get_low(), table.get_high(), index.get_low());
Vec16c t1 = _mm_perm_epi8(table.get_low(), table.get_high(), index.get_high());
return Vec32c(t0, t1);
#else
Vec32c f0 = constant8i<0,0,0,0,0x10101010,0x10101010,0x10101010,0x10101010>();
Vec32c f1 = constant8i<0x10101010,0x10101010,0x10101010,0x10101010,0,0,0,0>();
Vec32c tablef = _mm256_permute4x64_epi64(table, 0x4E); // low and high parts swapped
Vec32c r0 = _mm256_shuffle_epi8(table, (index ^ f0) + 0x70);
Vec32c r1 = _mm256_shuffle_epi8(tablef, (index ^ f1) + 0x70);
return r0 | r1;
#endif
}
template <int n>
static inline Vec32c lookup(Vec32uc const & index, void const * table) {
if (n <= 0) return 0;
if (n <= 16) {
Vec16c tt = Vec16c().load(table);
Vec16c r0 = lookup16(index.get_low(), tt);
Vec16c r1 = lookup16(index.get_high(), tt);
return Vec32c(r0, r1);
}
if (n <= 32) return lookup32(index, Vec32c().load(table));
// n > 32. Limit index
Vec32uc index1;
if ((n & (n-1)) == 0) {
// n is a power of 2, make index modulo n
index1 = Vec32uc(index) & uint8_t(n-1);
}
else {
// n is not a power of 2, limit to n-1
index1 = min(Vec32uc(index), uint8_t(n-1));
}
Vec8ui mask0 = Vec8ui(0x000000FF); // mask 8 bits
Vec32c t0 = _mm256_i32gather_epi32((const int *)table, __m256i(mask0 & Vec8ui(index1)), 1); // positions 0, 4, 8, ...
Vec32c t1 = _mm256_i32gather_epi32((const int *)table, __m256i(mask0 & _mm256_srli_epi32(index1, 8)), 1); // positions 1, 5, 9, ...
Vec32c t2 = _mm256_i32gather_epi32((const int *)table, __m256i(mask0 & _mm256_srli_epi32(index1,16)), 1); // positions 2, 6, 10, ...
Vec32c t3 = _mm256_i32gather_epi32((const int *)table, _mm256_srli_epi32(index1,24), 1); // positions 3, 7, 11, ...
t0 = t0 & mask0;
t1 = _mm256_slli_epi32(t1 & mask0, 8);
t2 = _mm256_slli_epi32(t2 & mask0, 16);
t3 = _mm256_slli_epi32(t3, 24);
return (t0 | t3) | (t1 | t2);
}
template <int n>
static inline Vec32c lookup(Vec32c const & index, void const * table) {
return lookup<n>(Vec32uc(index), table);
}
static inline Vec16s lookup16(Vec16s const & index, Vec16s const & table) {
return Vec16s(lookup32(Vec32c(index * 0x202 + 0x100), Vec32c(table)));
}
template <int n>
static inline Vec16s lookup(Vec16s const & index, void const * table) {
if (n <= 0) return 0;
if (n <= 8) {
Vec8s table1 = Vec8s().load(table);
return Vec16s(
lookup8 (index.get_low(), table1),
lookup8 (index.get_high(), table1));
}
if (n <= 16) return lookup16(index, Vec16s().load(table));
// n > 16. Limit index
Vec16us index1;
if ((n & (n-1)) == 0) {
// n is a power of 2, make index modulo n
index1 = Vec16us(index) & (n-1);
}
else {
// n is not a power of 2, limit to n-1
index1 = min(Vec16us(index), n-1);
}
Vec16s t1 = _mm256_i32gather_epi32((const int *)table, __m256i(Vec8ui(index1) & 0x0000FFFF), 2); // even positions
Vec16s t2 = _mm256_i32gather_epi32((const int *)table, _mm256_srli_epi32(index1, 16) , 2); // odd positions
return blend16s<0,16,2,18,4,20,6,22,8,24,10,26,12,28,14,30>(t1, t2);
}
static inline Vec8i lookup8(Vec8i const & index, Vec8i const & table) {
return _mm256_permutevar8x32_epi32(table, index);
}
template <int n>
static inline Vec8i lookup(Vec8i const & index, void const * table) {
if (n <= 0) return 0;
if (n <= 8) {
Vec8i table1 = Vec8i().load(table);
return lookup8(index, table1);
}
if (n <= 16) {
Vec8i table1 = Vec8i().load(table);
Vec8i table2 = Vec8i().load((int32_t const*)table + 8);
Vec8i y1 = lookup8(index, table1);
Vec8i y2 = lookup8(index, table2);
Vec8ib s = index > 7;
return select(s, y2, y1);
}
// n > 16. Limit index
Vec8ui index1;
if ((n & (n-1)) == 0) {
// n is a power of 2, make index modulo n
index1 = Vec8ui(index) & (n-1);
}
else {
// n is not a power of 2, limit to n-1
index1 = min(Vec8ui(index), n-1);
}
return _mm256_i32gather_epi32((const int *)table, index1, 4);
}
static inline Vec4q lookup4(Vec4q const & index, Vec4q const & table) {
return Vec4q(lookup8(Vec8i(index * 0x200000002ll + 0x100000000ll), Vec8i(table)));
}
template <int n>
static inline Vec4q lookup(Vec4q const & index, int64_t const * table) {
if (n <= 0) return 0;
// n > 0. Limit index
Vec4uq index1;
if ((n & (n-1)) == 0) {
// n is a power of 2, make index modulo n
index1 = Vec4uq(index) & (n-1);
}
else {
// n is not a power of 2, limit to n-1.
// There is no 64-bit min instruction, but we can use the 32-bit unsigned min,
// since n is a 32-bit integer
index1 = Vec4uq(min(Vec8ui(index), constant8i<n-1, 0, n-1, 0, n-1, 0, n-1, 0>()));
}
// old compilers can't agree how to define a 64 bit integer. Intel and MS use __int64, gcc use long long
#if defined (__clang__) && CLANG_VERSION < 30400
// clang 3.3 uses const int * in accordance with official Intel doc., which is wrong. will be fixed
return _mm256_i64gather_epi64((const int *)table, index1, 8);
#elif defined (_MSC_VER) && _MSC_VER < 1700 && ! defined(__INTEL_COMPILER)
// Old MS and Intel use non-standard type __int64
return _mm256_i64gather_epi64((const int64_t *)table, index1, 8);
#else
// Gnu, Clang 3.4, MS 11.0
return _mm256_i64gather_epi64((const long long *)table, index1, 8);
#endif
}
/*****************************************************************************
*
* Other permutations with variable indexes
*
*****************************************************************************/
// Function shift_bytes_up: shift whole vector left by b bytes.
// You may use a permute function instead if b is a compile-time constant
static inline Vec32c shift_bytes_up(Vec32c const & a, int b) {
if (b < 16) {
return Vec32c(shift_bytes_up(a.get_low(),b), shift_bytes_up(a.get_high(),b) | shift_bytes_down(a.get_low(),16-b));
}
else {
return Vec32c(Vec16c(0), shift_bytes_up(a.get_high(),b-16));
}
}
// Function shift_bytes_down: shift whole vector right by b bytes
// You may use a permute function instead if b is a compile-time constant
static inline Vec32c shift_bytes_down(Vec32c const & a, int b) {
if (b < 16) {
return Vec32c(shift_bytes_down(a.get_low(),b) | shift_bytes_up(a.get_high(),16-b), shift_bytes_down(a.get_high(),b));
}
else {
return Vec32c(shift_bytes_down(a.get_high(),b-16), Vec16c(0));
}
}
/*****************************************************************************
*
* Gather functions with fixed indexes
*
*****************************************************************************/
// Load elements from array a with indices i0, i1, i2, i3, i4, i5, i6, i7
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7>
static inline Vec8i gather8i(void const * a) {
Static_error_check<(i0|i1|i2|i3|i4|i5|i6|i7)>=0> Negative_array_index; // Error message if index is negative
const int i01min = i0 < i1 ? i0 : i1;
const int i23min = i2 < i3 ? i2 : i3;
const int i45min = i4 < i5 ? i4 : i5;
const int i67min = i6 < i7 ? i6 : i7;
const int i0123min = i01min < i23min ? i01min : i23min;
const int i4567min = i45min < i67min ? i45min : i67min;
const int imin = i0123min < i4567min ? i0123min : i4567min;
const int i01max = i0 > i1 ? i0 : i1;
const int i23max = i2 > i3 ? i2 : i3;
const int i45max = i4 > i5 ? i4 : i5;
const int i67max = i6 > i7 ? i6 : i7;
const int i0123max = i01max > i23max ? i01max : i23max;
const int i4567max = i45max > i67max ? i45max : i67max;
const int imax = i0123max > i4567max ? i0123max : i4567max;
if (imax - imin <= 7) {
// load one contiguous block and permute
if (imax > 7) {
// make sure we don't read past the end of the array
Vec8i b = Vec8i().load((int32_t const *)a + imax-7);
return permute8i<i0-imax+7, i1-imax+7, i2-imax+7, i3-imax+7, i4-imax+7, i5-imax+7, i6-imax+7, i7-imax+7>(b);
}
else {
Vec8i b = Vec8i().load((int32_t const *)a + imin);
return permute8i<i0-imin, i1-imin, i2-imin, i3-imin, i4-imin, i5-imin, i6-imin, i7-imin>(b);
}
}
if ((i0<imin+8 || i0>imax-8) && (i1<imin+8 || i1>imax-8) && (i2<imin+8 || i2>imax-8) && (i3<imin+8 || i3>imax-8)
&& (i4<imin+8 || i4>imax-8) && (i5<imin+8 || i5>imax-8) && (i6<imin+8 || i6>imax-8) && (i7<imin+8 || i7>imax-8)) {
// load two contiguous blocks and blend
Vec8i b = Vec8i().load((int32_t const *)a + imin);
Vec8i c = Vec8i().load((int32_t const *)a + imax-7);
const int j0 = i0<imin+8 ? i0-imin : 15-imax+i0;
const int j1 = i1<imin+8 ? i1-imin : 15-imax+i1;
const int j2 = i2<imin+8 ? i2-imin : 15-imax+i2;
const int j3 = i3<imin+8 ? i3-imin : 15-imax+i3;
const int j4 = i4<imin+8 ? i4-imin : 15-imax+i4;
const int j5 = i5<imin+8 ? i5-imin : 15-imax+i5;
const int j6 = i6<imin+8 ? i6-imin : 15-imax+i6;
const int j7 = i7<imin+8 ? i7-imin : 15-imax+i7;
return blend8i<j0, j1, j2, j3, j4, j5, j6, j7>(b, c);
}
// use AVX2 gather
return _mm256_i32gather_epi32((const int *)a, Vec8i(i0,i1,i2,i3,i4,i5,i6,i7), 4);
}
template <int i0, int i1, int i2, int i3>
static inline Vec4q gather4q(void const * a) {
Static_error_check<(i0|i1|i2|i3)>=0> Negative_array_index; // Error message if index is negative
const int i01min = i0 < i1 ? i0 : i1;
const int i23min = i2 < i3 ? i2 : i3;
const int imin = i01min < i23min ? i01min : i23min;
const int i01max = i0 > i1 ? i0 : i1;
const int i23max = i2 > i3 ? i2 : i3;
const int imax = i01max > i23max ? i01max : i23max;
if (imax - imin <= 3) {
// load one contiguous block and permute
if (imax > 3) {
// make sure we don't read past the end of the array
Vec4q b = Vec4q().load((int64_t const *)a + imax-3);
return permute4q<i0-imax+3, i1-imax+3, i2-imax+3, i3-imax+3>(b);
}
else {
Vec4q b = Vec4q().load((int64_t const *)a + imin);
return permute4q<i0-imin, i1-imin, i2-imin, i3-imin>(b);
}
}
if ((i0<imin+4 || i0>imax-4) && (i1<imin+4 || i1>imax-4) && (i2<imin+4 || i2>imax-4) && (i3<imin+4 || i3>imax-4)) {
// load two contiguous blocks and blend
Vec4q b = Vec4q().load((int64_t const *)a + imin);
Vec4q c = Vec4q().load((int64_t const *)a + imax-3);
const int j0 = i0<imin+4 ? i0-imin : 7-imax+i0;
const int j1 = i1<imin+4 ? i1-imin : 7-imax+i1;
const int j2 = i2<imin+4 ? i2-imin : 7-imax+i2;
const int j3 = i3<imin+4 ? i3-imin : 7-imax+i3;
return blend4q<j0, j1, j2, j3>(b, c);
}
// use AVX2 gather
// old compilers can't agree how to define a 64 bit integer. Intel and MS use __int64, gcc use long long
#if defined (__clang__) && CLANG_VERSION < 30400
// clang 3.3 uses const int * in accordance with official Intel doc., which is wrong. will be fixed
return _mm256_i32gather_epi64((const int *)a, Vec4i(i0,i1,i2,i3), 8);
#elif defined (_MSC_VER) && _MSC_VER < 1700 && ! defined(__INTEL_COMPILER)
// Old MS and Intel use non-standard type __int64
return _mm256_i32gather_epi64((const int64_t *)a, Vec4i(i0,i1,i2,i3), 8);
#else
// Gnu, Clang 3.4, MS 11.0
return _mm256_i32gather_epi64((const long long *)a, Vec4i(i0,i1,i2,i3), 8);
#endif
}
/*****************************************************************************
*
* Vector scatter functions
*
******************************************************************************
*
* These functions write the elements of a vector to arbitrary positions in an
* array in memory. Each vector element is written to an array position
* determined by an index. An element is not written if the corresponding
* index is out of range.
* The indexes can be specified as constant template parameters or as an
* integer vector.
*
* The scatter functions are useful if the data are distributed in a sparce
* manner into the array. If the array is dense then it is more efficient
* to permute the data into the right positions and then write the whole
* permuted vector into the array.
*
* Example:
* Vec8q a(10,11,12,13,14,15,16,17);
* int64_t b[16] = {0};
* scatter<0,2,14,10,1,-1,5,9>(a,b);
* // Now, b = {10,14,11,0,0,16,0,0,0,17,13,0,0,0,12,0}
*
*****************************************************************************/
template <int i0, int i1, int i2, int i3, int i4, int i5, int i6, int i7>
static inline void scatter(Vec8i const & data, void * array) {
#if defined (__AVX512VL__)
__m256i indx = constant8i<i0,i1,i2,i3,i4,i5,i6,i7>();
__mmask16 mask = uint16_t(i0>=0 | (i1>=0)<<1 | (i2>=0)<<2 | (i3>=0)<<3| (i4>=0)<<4| (i5>=0)<<5| (i6>=0)<<6| (i7>=0)<<7);
_mm256_mask_i32scatter_epi32((int*)array, mask, indx, data, 4);
#elif defined (__AVX512F__)
__m512i indx = _mm512_castsi256_si512(constant8i<i0,i1,i2,i3,i4,i5,i6,i7>());
__mmask16 mask = uint16_t(i0>=0 | (i1>=0)<<1 | (i2>=0)<<2 | (i3>=0)<<3| (i4>=0)<<4| (i5>=0)<<5| (i6>=0)<<6| (i7>=0)<<7);
_mm512_mask_i32scatter_epi32((int*)array, mask, indx, _mm512_castsi256_si512(data), 4);
#else
int32_t* arr = (int32_t*)array;
const int index[8] = {i0,i1,i2,i3,i4,i5,i6,i7};
for (int i = 0; i < 8; i++) {
if (index[i] >= 0) arr[index[i]] = data[i];
}
#endif
}
template <int i0, int i1, int i2, int i3>
static inline void scatter(Vec4q const & data, void * array) {
#if defined (__AVX512VL__)
__m128i indx = constant4i<i0,i1,i2,i3>();
__mmask16 mask = uint16_t(i0>=0 | (i1>=0)<<1 | (i2>=0)<<2 | (i3>=0)<<3);
_mm256_mask_i32scatter_epi64((long long *)array, mask, indx, data, 8);
#elif defined (__AVX512F__)
__m256i indx = _mm256_castsi128_si256(constant4i<i0,i1,i2,i3>());
__mmask16 mask = uint16_t(i0>=0 | (i1>=0)<<1 | (i2>=0)<<2 | (i3>=0)<<3);
_mm512_mask_i32scatter_epi64((long long*)array, mask, indx, _mm512_castsi256_si512(data), 8);
#else
int64_t* arr = (int64_t*)array;
const int index[4] = {i0,i1,i2,i3};
for (int i = 0; i < 4; i++) {
if (index[i] >= 0) arr[index[i]] = data[i];
}
#endif
}
static inline void scatter(Vec8i const & index, uint32_t limit, Vec8i const & data, void * array) {
#if defined (__AVX512VL__)
__mmask16 mask = _mm256_cmplt_epu32_mask(index, Vec8ui(limit));
_mm256_mask_i32scatter_epi32((int*)array, mask, index, data, 4);
#elif defined (__AVX512F__)
// 16 bit mask. upper 8 bits are (0<0) = false
__mmask16 mask = _mm512_cmplt_epu32_mask(_mm512_castsi256_si512(index), _mm512_castsi256_si512(Vec8ui(limit)));
_mm512_mask_i32scatter_epi32((int*)array, mask, _mm512_castsi256_si512(index), _mm512_castsi256_si512(data), 4);
#else
int32_t* arr = (int32_t*)array;
for (int i = 0; i < 8; i++) {
if (uint32_t(index[i]) < limit) arr[index[i]] = data[i];
}
#endif
}
static inline void scatter(Vec4q const & index, uint32_t limit, Vec4q const & data, void * array) {
#if defined (__AVX512VL__)
__mmask16 mask = _mm256_cmplt_epu64_mask(index, Vec4uq(uint64_t(limit)));
_mm256_mask_i64scatter_epi64((long long*)array, mask, index, data, 8);
#elif defined (__AVX512F__)
// 16 bit mask. upper 8 bits are (0<0) = false
__mmask16 mask = _mm512_cmplt_epu64_mask(_mm512_castsi256_si512(index), _mm512_castsi256_si512(Vec4uq(uint64_t(limit))));
_mm512_mask_i64scatter_epi64((long long*)array, mask, _mm512_castsi256_si512(index), _mm512_castsi256_si512(data), 8);
#else
int64_t* arr = (int64_t*)array;
for (int i = 0; i < 4; i++) {
if (uint64_t(index[i]) < uint64_t(limit)) arr[index[i]] = data[i];
}
#endif
}
static inline void scatter(Vec4i const & index, uint32_t limit, Vec4q const & data, void * array) {
#if defined (__AVX512VL__)
__mmask16 mask = _mm_cmplt_epu32_mask(index, Vec4ui(limit));
_mm256_mask_i32scatter_epi64((long long*)array, mask, index, data, 8);
#elif defined (__AVX512F__)
// 16 bit mask. upper 8 bits are (0<0) = false
__mmask16 mask = _mm512_cmplt_epu32_mask(_mm512_castsi128_si512(index), _mm512_castsi128_si512(Vec4ui(limit)));
_mm512_mask_i32scatter_epi64((long long*)array, mask, _mm256_castsi128_si256(index), _mm512_castsi256_si512(data), 8);
#else
int64_t* arr = (int64_t*)array;
for (int i = 0; i < 4; i++) {
if (uint32_t(index[i]) < limit) arr[index[i]] = data[i];
}
#endif
}
/*****************************************************************************
*
* Functions for conversion between integer sizes
*
*****************************************************************************/
// Extend 8-bit integers to 16-bit integers, signed and unsigned
// Function extend_low : extends the low 16 elements to 16 bits with sign extension
static inline Vec16s extend_low (Vec32c const & a) {
__m256i a2 = permute4q<0,-256,1,-256>(Vec4q(a)); // get bits 64-127 to position 128-191
__m256i sign = _mm256_cmpgt_epi8(_mm256_setzero_si256(),a2); // 0 > a2
return _mm256_unpacklo_epi8(a2, sign); // interleave with sign extensions
}
// Function extend_high : extends the high 16 elements to 16 bits with sign extension
static inline Vec16s extend_high (Vec32c const & a) {
__m256i a2 = permute4q<-256,2,-256,3>(Vec4q(a)); // get bits 128-191 to position 64-127
__m256i sign = _mm256_cmpgt_epi8(_mm256_setzero_si256(),a2); // 0 > a2
return _mm256_unpackhi_epi8(a2, sign); // interleave with sign extensions
}
// Function extend_low : extends the low 16 elements to 16 bits with zero extension
static inline Vec16us extend_low (Vec32uc const & a) {
__m256i a2 = permute4q<0,-256,1,-256>(Vec4q(a)); // get bits 64-127 to position 128-191
return _mm256_unpacklo_epi8(a2, _mm256_setzero_si256()); // interleave with zero extensions
}
// Function extend_high : extends the high 19 elements to 16 bits with zero extension
static inline Vec16us extend_high (Vec32uc const & a) {
__m256i a2 = permute4q<-256,2,-256,3>(Vec4q(a)); // get bits 128-191 to position 64-127
return _mm256_unpackhi_epi8(a2, _mm256_setzero_si256()); // interleave with zero extensions
}
// Extend 16-bit integers to 32-bit integers, signed and unsigned
// Function extend_low : extends the low 8 elements to 32 bits with sign extension
static inline Vec8i extend_low (Vec16s const & a) {
__m256i a2 = permute4q<0,-256,1,-256>(Vec4q(a)); // get bits 64-127 to position 128-191
__m256i sign = _mm256_srai_epi16(a2, 15); // sign bit
return _mm256_unpacklo_epi16(a2 ,sign); // interleave with sign extensions
}
// Function extend_high : extends the high 8 elements to 32 bits with sign extension
static inline Vec8i extend_high (Vec16s const & a) {
__m256i a2 = permute4q<-256,2,-256,3>(Vec4q(a)); // get bits 128-191 to position 64-127
__m256i sign = _mm256_srai_epi16(a2, 15); // sign bit
return _mm256_unpackhi_epi16(a2, sign); // interleave with sign extensions
}
// Function extend_low : extends the low 8 elements to 32 bits with zero extension
static inline Vec8ui extend_low (Vec16us const & a) {
__m256i a2 = permute4q<0,-256,1,-256>(Vec4q(a)); // get bits 64-127 to position 128-191
return _mm256_unpacklo_epi16(a2, _mm256_setzero_si256()); // interleave with zero extensions
}
// Function extend_high : extends the high 8 elements to 32 bits with zero extension
static inline Vec8ui extend_high (Vec16us const & a) {
__m256i a2 = permute4q<-256,2,-256,3>(Vec4q(a)); // get bits 128-191 to position 64-127
return _mm256_unpackhi_epi16(a2, _mm256_setzero_si256()); // interleave with zero extensions
}
// Extend 32-bit integers to 64-bit integers, signed and unsigned
// Function extend_low : extends the low 4 elements to 64 bits with sign extension
static inline Vec4q extend_low (Vec8i const & a) {
__m256i a2 = permute4q<0,-256,1,-256>(Vec4q(a)); // get bits 64-127 to position 128-191
__m256i sign = _mm256_srai_epi32(a2, 31); // sign bit
return _mm256_unpacklo_epi32(a2, sign); // interleave with sign extensions
}
// Function extend_high : extends the high 4 elements to 64 bits with sign extension
static inline Vec4q extend_high (Vec8i const & a) {
__m256i a2 = permute4q<-256,2,-256,3>(Vec4q(a)); // get bits 128-191 to position 64-127
__m256i sign = _mm256_srai_epi32(a2, 31); // sign bit
return _mm256_unpackhi_epi32(a2, sign); // interleave with sign extensions
}
// Function extend_low : extends the low 4 elements to 64 bits with zero extension
static inline Vec4uq extend_low (Vec8ui const & a) {
__m256i a2 = permute4q<0,-256,1,-256>(Vec4q(a)); // get bits 64-127 to position 128-191
return _mm256_unpacklo_epi32(a2, _mm256_setzero_si256()); // interleave with zero extensions
}
// Function extend_high : extends the high 4 elements to 64 bits with zero extension
static inline Vec4uq extend_high (Vec8ui const & a) {
__m256i a2 = permute4q<-256,2,-256,3>(Vec4q(a)); // get bits 128-191 to position 64-127
return _mm256_unpackhi_epi32(a2, _mm256_setzero_si256()); // interleave with zero extensions
}
// Compress 16-bit integers to 8-bit integers, signed and unsigned, with and without saturation
// Function compress : packs two vectors of 16-bit integers into one vector of 8-bit integers
// Overflow wraps around
static inline Vec32c compress (Vec16s const & low, Vec16s const & high) {
__m256i mask = _mm256_set1_epi32(0x00FF00FF); // mask for low bytes
__m256i lowm = _mm256_and_si256(low, mask); // bytes of low
__m256i highm = _mm256_and_si256(high, mask); // bytes of high
__m256i pk = _mm256_packus_epi16(lowm, highm); // unsigned pack
return _mm256_permute4x64_epi64(pk, 0xD8); // put in right place
}
// Function compress : packs two vectors of 16-bit integers into one vector of 8-bit integers
// Signed, with saturation
static inline Vec32c compress_saturated (Vec16s const & low, Vec16s const & high) {
__m256i pk = _mm256_packs_epi16(low,high); // packed with signed saturation
return _mm256_permute4x64_epi64(pk, 0xD8); // put in right place
}
// Function compress : packs two vectors of 16-bit integers to one vector of 8-bit integers
// Unsigned, overflow wraps around
static inline Vec32uc compress (Vec16us const & low, Vec16us const & high) {
return Vec32uc (compress((Vec16s)low, (Vec16s)high));
}
// Function compress : packs two vectors of 16-bit integers into one vector of 8-bit integers
// Unsigned, with saturation
static inline Vec32uc compress_saturated (Vec16us const & low, Vec16us const & high) {
__m256i maxval = _mm256_set1_epi32(0x00FF00FF); // maximum value
__m256i minval = _mm256_setzero_si256(); // minimum value = 0
__m256i low1 = _mm256_min_epu16(low,maxval); // upper limit
__m256i high1 = _mm256_min_epu16(high,maxval); // upper limit
__m256i low2 = _mm256_max_epu16(low1,minval); // lower limit
__m256i high2 = _mm256_max_epu16(high1,minval); // lower limit
__m256i pk = _mm256_packus_epi16(low2,high2); // this instruction saturates from signed 32 bit to unsigned 16 bit
return _mm256_permute4x64_epi64(pk, 0xD8); // put in right place
}
// Compress 32-bit integers to 16-bit integers, signed and unsigned, with and without saturation
// Function compress : packs two vectors of 32-bit integers into one vector of 16-bit integers
// Overflow wraps around
static inline Vec16s compress (Vec8i const & low, Vec8i const & high) {
__m256i mask = _mm256_set1_epi32(0x0000FFFF); // mask for low words
__m256i lowm = _mm256_and_si256(low,mask); // bytes of low
__m256i highm = _mm256_and_si256(high,mask); // bytes of high
__m256i pk = _mm256_packus_epi32(lowm,highm); // unsigned pack
return _mm256_permute4x64_epi64(pk, 0xD8); // put in right place
}
// Function compress : packs two vectors of 32-bit integers into one vector of 16-bit integers
// Signed with saturation
static inline Vec16s compress_saturated (Vec8i const & low, Vec8i const & high) {
__m256i pk = _mm256_packs_epi32(low,high); // pack with signed saturation
return _mm256_permute4x64_epi64(pk, 0xD8); // put in right place
}
// Function compress : packs two vectors of 32-bit integers into one vector of 16-bit integers
// Overflow wraps around
static inline Vec16us compress (Vec8ui const & low, Vec8ui const & high) {
return Vec16us (compress((Vec8i)low, (Vec8i)high));
}
// Function compress : packs two vectors of 32-bit integers into one vector of 16-bit integers
// Unsigned, with saturation
static inline Vec16us compress_saturated (Vec8ui const & low, Vec8ui const & high) {
__m256i maxval = _mm256_set1_epi32(0x0000FFFF); // maximum value
__m256i minval = _mm256_setzero_si256(); // minimum value = 0
__m256i low1 = _mm256_min_epu32(low,maxval); // upper limit
__m256i high1 = _mm256_min_epu32(high,maxval); // upper limit
__m256i low2 = _mm256_max_epu32(low1,minval); // lower limit
__m256i high2 = _mm256_max_epu32(high1,minval); // lower limit
__m256i pk = _mm256_packus_epi32(low2,high2); // this instruction saturates from signed 32 bit to unsigned 16 bit
return _mm256_permute4x64_epi64(pk, 0xD8); // put in right place
}
// Compress 64-bit integers to 32-bit integers, signed and unsigned, with and without saturation
// Function compress : packs two vectors of 64-bit integers into one vector of 32-bit integers
// Overflow wraps around
static inline Vec8i compress (Vec4q const & low, Vec4q const & high) {
__m256i low2 = _mm256_shuffle_epi32(low,0xD8); // low dwords of low to pos. 0 and 32
__m256i high2 = _mm256_shuffle_epi32(high,0xD8); // low dwords of high to pos. 0 and 32
__m256i pk = _mm256_unpacklo_epi64(low2,high2); // interleave
return _mm256_permute4x64_epi64(pk, 0xD8); // put in right place
}
// Function compress : packs two vectors of 64-bit integers into one vector of 32-bit integers
// Signed, with saturation
static inline Vec8i compress_saturated (Vec4q const & a, Vec4q const & b) {
Vec4q maxval = constant8ui<0x7FFFFFFF,0,0x7FFFFFFF,0,0x7FFFFFFF,0,0x7FFFFFFF,0>();
Vec4q minval = constant8ui<0x80000000,0xFFFFFFFF,0x80000000,0xFFFFFFFF,0x80000000,0xFFFFFFFF,0x80000000,0xFFFFFFFF>();
Vec4q a1 = min(a,maxval);
Vec4q b1 = min(b,maxval);
Vec4q a2 = max(a1,minval);
Vec4q b2 = max(b1,minval);
return compress(a2,b2);
}
// Function compress : packs two vectors of 32-bit integers into one vector of 16-bit integers
// Overflow wraps around
static inline Vec8ui compress (Vec4uq const & low, Vec4uq const & high) {
return Vec8ui (compress((Vec4q)low, (Vec4q)high));
}
// Function compress : packs two vectors of 64-bit integers into one vector of 32-bit integers
// Unsigned, with saturation
static inline Vec8ui compress_saturated (Vec4uq const & low, Vec4uq const & high) {
__m256i zero = _mm256_setzero_si256(); // 0
__m256i lowzero = _mm256_cmpeq_epi32(low,zero); // for each dword is zero
__m256i highzero = _mm256_cmpeq_epi32(high,zero); // for each dword is zero
__m256i mone = _mm256_set1_epi32(-1); // FFFFFFFF
__m256i lownz = _mm256_xor_si256(lowzero,mone); // for each dword is nonzero
__m256i highnz = _mm256_xor_si256(highzero,mone); // for each dword is nonzero
__m256i lownz2 = _mm256_srli_epi64(lownz,32); // shift down to low dword
__m256i highnz2 = _mm256_srli_epi64(highnz,32); // shift down to low dword
__m256i lowsatur = _mm256_or_si256(low,lownz2); // low, saturated
__m256i hisatur = _mm256_or_si256(high,highnz2); // high, saturated
return Vec8ui (compress(Vec4q(lowsatur), Vec4q(hisatur)));
}
/*****************************************************************************
*
* Integer division operators
*
* Please see the file vectori128.h for explanation.
*
*****************************************************************************/
// vector operator / : divide each element by divisor
// vector of 8 32-bit signed integers
static inline Vec8i operator / (Vec8i const & a, Divisor_i const & d) {
__m256i m = _mm256_broadcastq_epi64(d.getm()); // broadcast multiplier
__m256i sgn = _mm256_broadcastq_epi64(d.getsign()); // broadcast sign of d
__m256i t1 = _mm256_mul_epi32(a,m); // 32x32->64 bit signed multiplication of even elements of a
__m256i t2 = _mm256_srli_epi64(t1,32); // high dword of even numbered results
__m256i t3 = _mm256_srli_epi64(a,32); // get odd elements of a into position for multiplication
__m256i t4 = _mm256_mul_epi32(t3,m); // 32x32->64 bit signed multiplication of odd elements
__m256i t5 = constant8i<0,-1,0,-1,0,-1,0,-1> (); // mask for odd elements
__m256i t7 = _mm256_blendv_epi8(t2,t4,t5); // blend two results
__m256i t8 = _mm256_add_epi32(t7,a); // add
__m256i t9 = _mm256_sra_epi32(t8,d.gets1()); // shift right artihmetic
__m256i t10 = _mm256_srai_epi32(a,31); // sign of a
__m256i t11 = _mm256_sub_epi32(t10,sgn); // sign of a - sign of d
__m256i t12 = _mm256_sub_epi32(t9,t11); // + 1 if a < 0, -1 if d < 0
return _mm256_xor_si256(t12,sgn); // change sign if divisor negative
}
// vector of 8 32-bit unsigned integers
static inline Vec8ui operator / (Vec8ui const & a, Divisor_ui const & d) {
__m256i m = _mm256_broadcastq_epi64(d.getm()); // broadcast multiplier
__m256i t1 = _mm256_mul_epu32(a,m); // 32x32->64 bit unsigned multiplication of even elements of a
__m256i t2 = _mm256_srli_epi64(t1,32); // high dword of even numbered results
__m256i t3 = _mm256_srli_epi64(a,32); // get odd elements of a into position for multiplication
__m256i t4 = _mm256_mul_epu32(t3,m); // 32x32->64 bit unsigned multiplication of odd elements
__m256i t5 = constant8i<0,-1,0,-1,0,-1,0,-1> (); // mask for odd elements
__m256i t7 = _mm256_blendv_epi8(t2,t4,t5); // blend two results
__m256i t8 = _mm256_sub_epi32(a,t7); // subtract
__m256i t9 = _mm256_srl_epi32(t8,d.gets1()); // shift right logical
__m256i t10 = _mm256_add_epi32(t7,t9); // add
return _mm256_srl_epi32(t10,d.gets2()); // shift right logical
}
// vector of 16 16-bit signed integers
static inline Vec16s operator / (Vec16s const & a, Divisor_s const & d) {
__m256i m = _mm256_broadcastq_epi64(d.getm()); // broadcast multiplier
__m256i sgn = _mm256_broadcastq_epi64(d.getsign()); // broadcast sign of d
__m256i t1 = _mm256_mulhi_epi16(a, m); // multiply high signed words
__m256i t2 = _mm256_add_epi16(t1,a); // + a
__m256i t3 = _mm256_sra_epi16(t2,d.gets1()); // shift right artihmetic
__m256i t4 = _mm256_srai_epi16(a,15); // sign of a
__m256i t5 = _mm256_sub_epi16(t4,sgn); // sign of a - sign of d
__m256i t6 = _mm256_sub_epi16(t3,t5); // + 1 if a < 0, -1 if d < 0
return _mm256_xor_si256(t6,sgn); // change sign if divisor negative
}
// vector of 16 16-bit unsigned integers
static inline Vec16us operator / (Vec16us const & a, Divisor_us const & d) {
__m256i m = _mm256_broadcastq_epi64(d.getm()); // broadcast multiplier
__m256i t1 = _mm256_mulhi_epu16(a, m); // multiply high signed words
__m256i t2 = _mm256_sub_epi16(a,t1); // subtract
__m256i t3 = _mm256_srl_epi16(t2,d.gets1()); // shift right logical
__m256i t4 = _mm256_add_epi16(t1,t3); // add
return _mm256_srl_epi16(t4,d.gets2()); // shift right logical
}
// vector of 32 8-bit signed integers
static inline Vec32c operator / (Vec32c const & a, Divisor_s const & d) {
// expand into two Vec16s
Vec16s low = extend_low(a) / d;
Vec16s high = extend_high(a) / d;
return compress(low,high);
}
// vector of 32 8-bit unsigned integers
static inline Vec32uc operator / (Vec32uc const & a, Divisor_us const & d) {
// expand into two Vec16s
Vec16us low = extend_low(a) / d;
Vec16us high = extend_high(a) / d;
return compress(low,high);
}
// vector operator /= : divide
static inline Vec8i & operator /= (Vec8i & a, Divisor_i const & d) {
a = a / d;
return a;
}
// vector operator /= : divide
static inline Vec8ui & operator /= (Vec8ui & a, Divisor_ui const & d) {
a = a / d;
return a;
}
// vector operator /= : divide
static inline Vec16s & operator /= (Vec16s & a, Divisor_s const & d) {
a = a / d;
return a;
}
// vector operator /= : divide
static inline Vec16us & operator /= (Vec16us & a, Divisor_us const & d) {
a = a / d;
return a;
}
// vector operator /= : divide
static inline Vec32c & operator /= (Vec32c & a, Divisor_s const & d) {
a = a / d;
return a;
}
// vector operator /= : divide
static inline Vec32uc & operator /= (Vec32uc & a, Divisor_us const & d) {
a = a / d;
return a;
}
/*****************************************************************************
*
* Integer division 2: divisor is a compile-time constant
*
*****************************************************************************/
// Divide Vec8i by compile-time constant
template <int32_t d>
static inline Vec8i divide_by_i(Vec8i const & x) {
Static_error_check<(d!=0)> Dividing_by_zero; // Error message if dividing by zero
if (d == 1) return x;
if (d == -1) return -x;
if (uint32_t(d) == 0x80000000u) return Vec8i(x == Vec8i(0x80000000)) & 1; // prevent overflow when changing sign
const uint32_t d1 = d > 0 ? uint32_t(d) : -uint32_t(d); // compile-time abs(d). (force GCC compiler to treat d as 32 bits, not 64 bits)
if ((d1 & (d1-1)) == 0) {
// d1 is a power of 2. use shift
const int k = bit_scan_reverse_const(d1);
__m256i sign;
if (k > 1) sign = _mm256_srai_epi32(x, k-1); else sign = x; // k copies of sign bit
__m256i bias = _mm256_srli_epi32(sign, 32-k); // bias = x >= 0 ? 0 : k-1
__m256i xpbias = _mm256_add_epi32 (x, bias); // x + bias
__m256i q = _mm256_srai_epi32(xpbias, k); // (x + bias) >> k
if (d > 0) return q; // d > 0: return q
return _mm256_sub_epi32(_mm256_setzero_si256(), q); // d < 0: return -q
}
// general case
const int32_t sh = bit_scan_reverse_const(uint32_t(d1)-1); // ceil(log2(d1)) - 1. (d1 < 2 handled by power of 2 case)
const int32_t mult = int(1 + (uint64_t(1) << (32+sh)) / uint32_t(d1) - (int64_t(1) << 32)); // multiplier
const Divisor_i div(mult, sh, d < 0 ? -1 : 0);
return x / div;
}
// define Vec8i a / const_int(d)
template <int32_t d>
static inline Vec8i operator / (Vec8i const & a, Const_int_t<d>) {
return divide_by_i<d>(a);
}
// define Vec8i a / const_uint(d)
template <uint32_t d>
static inline Vec8i operator / (Vec8i const & a, Const_uint_t<d>) {
Static_error_check< (d<0x80000000u) > Error_overflow_dividing_signed_by_unsigned; // Error: dividing signed by overflowing unsigned
return divide_by_i<int32_t(d)>(a); // signed divide
}
// vector operator /= : divide
template <int32_t d>
static inline Vec8i & operator /= (Vec8i & a, Const_int_t<d> b) {
a = a / b;
return a;
}
// vector operator /= : divide
template <uint32_t d>
static inline Vec8i & operator /= (Vec8i & a, Const_uint_t<d> b) {
a = a / b;
return a;
}
// Divide Vec8ui by compile-time constant
template <uint32_t d>
static inline Vec8ui divide_by_ui(Vec8ui const & x) {
Static_error_check<(d!=0)> Dividing_by_zero; // Error message if dividing by zero
if (d == 1) return x; // divide by 1
const int b = bit_scan_reverse_const(d); // floor(log2(d))
if ((uint32_t(d) & (uint32_t(d)-1)) == 0) {
// d is a power of 2. use shift
return _mm256_srli_epi32(x, b); // x >> b
}
// general case (d > 2)
uint32_t mult = uint32_t((uint64_t(1) << (b+32)) / d); // multiplier = 2^(32+b) / d
const uint64_t rem = (uint64_t(1) << (b+32)) - uint64_t(d)*mult; // remainder 2^(32+b) % d
const bool round_down = (2*rem < d); // check if fraction is less than 0.5
if (!round_down) {
mult = mult + 1; // round up mult
}
// do 32*32->64 bit unsigned multiplication and get high part of result
const __m256i multv = _mm256_set_epi32(0,mult,0,mult,0,mult,0,mult);// zero-extend mult and broadcast
__m256i t1 = _mm256_mul_epu32(x,multv); // 32x32->64 bit unsigned multiplication of x[0] and x[2]
if (round_down) {
t1 = _mm256_add_epi64(t1,multv); // compensate for rounding error. (x+1)*m replaced by x*m+m to avoid overflow
}
__m256i t2 = _mm256_srli_epi64(t1,32); // high dword of result 0 and 2
__m256i t3 = _mm256_srli_epi64(x,32); // get x[1] and x[3] into position for multiplication
__m256i t4 = _mm256_mul_epu32(t3,multv); // 32x32->64 bit unsigned multiplication of x[1] and x[3]
if (round_down) {
t4 = _mm256_add_epi64(t4,multv); // compensate for rounding error. (x+1)*m replaced by x*m+m to avoid overflow
}
__m256i t5 = _mm256_set_epi32(-1,0,-1,0,-1,0,-1,0); // mask of dword 1 and 3
__m256i t7 = _mm256_blendv_epi8(t2,t4,t5); // blend two results
Vec8ui q = _mm256_srli_epi32(t7, b); // shift right by b
return q; // no overflow possible
}
// define Vec8ui a / const_uint(d)
template <uint32_t d>
static inline Vec8ui operator / (Vec8ui const & a, Const_uint_t<d>) {
return divide_by_ui<d>(a);
}
// define Vec8ui a / const_int(d)
template <int32_t d>
static inline Vec8ui operator / (Vec8ui const & a, Const_int_t<d>) {
Static_error_check< (d>=0) > Error_dividing_unsigned_by_negative;// Error: dividing unsigned by negative is ambiguous
return divide_by_ui<d>(a); // unsigned divide
}
// vector operator /= : divide
template <uint32_t d>
static inline Vec8ui & operator /= (Vec8ui & a, Const_uint_t<d> b) {
a = a / b;
return a;
}
// vector operator /= : divide
template <int32_t d>
static inline Vec8ui & operator /= (Vec8ui & a, Const_int_t<d> b) {
a = a / b;
return a;
}
// Divide Vec16s by compile-time constant
template <int d>
static inline Vec16s divide_by_i(Vec16s const & x) {
const int16_t d0 = int16_t(d); // truncate d to 16 bits
Static_error_check<(d0 != 0)> Dividing_by_zero; // Error message if dividing by zero
if (d0 == 1) return x; // divide by 1
if (d0 == -1) return -x; // divide by -1
if (uint16_t(d0) == 0x8000u) return Vec16s(x == Vec16s(0x8000)) & 1;// prevent overflow when changing sign
const uint16_t d1 = d0 > 0 ? d0 : -d0; // compile-time abs(d0)
if ((d1 & (d1-1)) == 0) {
// d is a power of 2. use shift
const int k = bit_scan_reverse_const(uint32_t(d1));
__m256i sign;
if (k > 1) sign = _mm256_srai_epi16(x, k-1); else sign = x; // k copies of sign bit
__m256i bias = _mm256_srli_epi16(sign, 16-k); // bias = x >= 0 ? 0 : k-1
__m256i xpbias = _mm256_add_epi16 (x, bias); // x + bias
__m256i q = _mm256_srai_epi16(xpbias, k); // (x + bias) >> k
if (d0 > 0) return q; // d0 > 0: return q
return _mm256_sub_epi16(_mm256_setzero_si256(), q); // d0 < 0: return -q
}
// general case
const int L = bit_scan_reverse_const(uint16_t(d1-1)) + 1; // ceil(log2(d)). (d < 2 handled above)
const int16_t mult = int16_t(1 + (1u << (15+L)) / uint32_t(d1) - 0x10000);// multiplier
const int shift1 = L - 1;
const Divisor_s div(mult, shift1, d0 > 0 ? 0 : -1);
return x / div;
}
// define Vec16s a / const_int(d)
template <int d>
static inline Vec16s operator / (Vec16s const & a, Const_int_t<d>) {
return divide_by_i<d>(a);
}
// define Vec16s a / const_uint(d)
template <uint32_t d>
static inline Vec16s operator / (Vec16s const & a, Const_uint_t<d>) {
Static_error_check< (d<0x8000u) > Error_overflow_dividing_signed_by_unsigned; // Error: dividing signed by overflowing unsigned
return divide_by_i<int(d)>(a); // signed divide
}
// vector operator /= : divide
template <int32_t d>
static inline Vec16s & operator /= (Vec16s & a, Const_int_t<d> b) {
a = a / b;
return a;
}
// vector operator /= : divide
template <uint32_t d>
static inline Vec16s & operator /= (Vec16s & a, Const_uint_t<d> b) {
a = a / b;
return a;
}
// Divide Vec16us by compile-time constant
template <uint32_t d>
static inline Vec16us divide_by_ui(Vec16us const & x) {
const uint16_t d0 = uint16_t(d); // truncate d to 16 bits
Static_error_check<(d0 != 0)> Dividing_by_zero; // Error message if dividing by zero
if (d0 == 1) return x; // divide by 1
const int b = bit_scan_reverse_const(d0); // floor(log2(d))
if ((d0 & (d0-1)) == 0) {
// d is a power of 2. use shift
return _mm256_srli_epi16(x, b); // x >> b
}
// general case (d > 2)
uint16_t mult = uint16_t((uint32_t(1) << (b+16)) / d0); // multiplier = 2^(32+b) / d
const uint32_t rem = (uint32_t(1) << (b+16)) - uint32_t(d0)*mult;// remainder 2^(32+b) % d
const bool round_down = (2*rem < d0); // check if fraction is less than 0.5
Vec16us x1 = x;
if (round_down) {
x1 = x1 + 1; // round down mult and compensate by adding 1 to x
}
else {
mult = mult + 1; // round up mult. no compensation needed
}
const __m256i multv = _mm256_set1_epi16(mult); // broadcast mult
__m256i xm = _mm256_mulhi_epu16(x1, multv); // high part of 16x16->32 bit unsigned multiplication
Vec16us q = _mm256_srli_epi16(xm, b); // shift right by b
if (round_down) {
Vec16sb overfl = (x1 == Vec16us(_mm256_setzero_si256())); // check for overflow of x+1
return select(overfl, Vec16us(mult >> b), q); // deal with overflow (rarely needed)
}
else {
return q; // no overflow possible
}
}
// define Vec16us a / const_uint(d)
template <uint32_t d>
static inline Vec16us operator / (Vec16us const & a, Const_uint_t<d>) {
return divide_by_ui<d>(a);
}
// define Vec16us a / const_int(d)
template <int d>
static inline Vec16us operator / (Vec16us const & a, Const_int_t<d>) {
Static_error_check< (d>=0) > Error_dividing_unsigned_by_negative;// Error: dividing unsigned by negative is ambiguous
return divide_by_ui<d>(a); // unsigned divide
}
// vector operator /= : divide
template <uint32_t d>
static inline Vec16us & operator /= (Vec16us & a, Const_uint_t<d> b) {
a = a / b;
return a;
}
// vector operator /= : divide
template <int32_t d>
static inline Vec16us & operator /= (Vec16us & a, Const_int_t<d> b) {
a = a / b;
return a;
}
// define Vec32c a / const_int(d)
template <int d>
static inline Vec32c operator / (Vec32c const & a, Const_int_t<d>) {
// expand into two Vec16s
Vec16s low = extend_low(a) / Const_int_t<d>();
Vec16s high = extend_high(a) / Const_int_t<d>();
return compress(low,high);
}
// define Vec32c a / const_uint(d)
template <uint32_t d>
static inline Vec32c operator / (Vec32c const & a, Const_uint_t<d>) {
Static_error_check< (uint8_t(d)<0x80u) > Error_overflow_dividing_signed_by_unsigned; // Error: dividing signed by overflowing unsigned
return a / Const_int_t<d>(); // signed divide
}
// vector operator /= : divide
template <int32_t d>
static inline Vec32c & operator /= (Vec32c & a, Const_int_t<d> b) {
a = a / b;
return a;
}
// vector operator /= : divide
template <uint32_t d>
static inline Vec32c & operator /= (Vec32c & a, Const_uint_t<d> b) {
a = a / b;
return a;
}
// define Vec32uc a / const_uint(d)
template <uint32_t d>
static inline Vec32uc operator / (Vec32uc const & a, Const_uint_t<d>) {
// expand into two Vec16us
Vec16us low = extend_low(a) / Const_uint_t<d>();
Vec16us high = extend_high(a) / Const_uint_t<d>();
return compress(low,high);
}
// define Vec32uc a / const_int(d)
template <int d>
static inline Vec32uc operator / (Vec32uc const & a, Const_int_t<d>) {
Static_error_check< (int8_t(d)>=0) > Error_dividing_unsigned_by_negative;// Error: dividing unsigned by negative is ambiguous
return a / Const_uint_t<d>(); // unsigned divide
}
// vector operator /= : divide
template <uint32_t d>
static inline Vec32uc & operator /= (Vec32uc & a, Const_uint_t<d> b) {
a = a / b;
return a;
}
// vector operator /= : divide
template <int32_t d>
static inline Vec32uc & operator /= (Vec32uc & a, Const_int_t<d> b) {
a = a / b;
return a;
}
/*****************************************************************************
*
* Horizontal scan functions
*
*****************************************************************************/
// Get index to the first element that is true. Return -1 if all are false
static inline int horizontal_find_first(Vec32cb const & x) {
uint32_t a = _mm256_movemask_epi8(x);
if (a == 0) return -1;
int32_t b = bit_scan_forward(a);
return b;
}
static inline int horizontal_find_first(Vec16sb const & x) {
return horizontal_find_first(Vec32cb(x)) >> 1;
}
static inline int horizontal_find_first(Vec8ib const & x) {
return horizontal_find_first(Vec32cb(x)) >> 2;
}
static inline int horizontal_find_first(Vec4qb const & x) {
return horizontal_find_first(Vec32cb(x)) >> 3;
}
// Count the number of elements that are true
static inline uint32_t horizontal_count(Vec32cb const & x) {
uint32_t a = _mm256_movemask_epi8(x);
return vml_popcnt(a);
}
static inline uint32_t horizontal_count(Vec16sb const & x) {
return horizontal_count(Vec32cb(x)) >> 1;
}
static inline uint32_t horizontal_count(Vec8ib const & x) {
return horizontal_count(Vec32cb(x)) >> 2;
}
static inline uint32_t horizontal_count(Vec4qb const & x) {
return horizontal_count(Vec32cb(x)) >> 3;
}
/*****************************************************************************
*
* Boolean <-> bitfield conversion functions
*
*****************************************************************************/
// to_bits: convert boolean vector to integer bitfield
static inline uint32_t to_bits(Vec32cb const & x) {
return (uint32_t)_mm256_movemask_epi8(x);
}
// to_Vec16c: convert integer bitfield to boolean vector
static inline Vec32cb to_Vec32cb(uint32_t x) {
return Vec32cb(Vec32c(to_Vec16cb(uint16_t(x)), to_Vec16cb(uint16_t(x>>16))));
}
// to_bits: convert boolean vector to integer bitfield
static inline uint16_t to_bits(Vec16sb const & x) {
__m128i a = _mm_packs_epi16(x.get_low(), x.get_high()); // 16-bit words to bytes
return (uint16_t)_mm_movemask_epi8(a);
}
// to_Vec16sb: convert integer bitfield to boolean vector
static inline Vec16sb to_Vec16sb(uint16_t x) {
return Vec16sb(Vec16s(to_Vec8sb(uint8_t(x)), to_Vec8sb(uint8_t(x>>8))));
}
#if INSTRSET < 9 || MAX_VECTOR_SIZE < 512
// These functions are defined in Vectori512.h if AVX512 instruction set is used
// to_bits: convert boolean vector to integer bitfield
static inline uint8_t to_bits(Vec8ib const & x) {
__m128i a = _mm_packs_epi32(x.get_low(), x.get_high()); // 32-bit dwords to 16-bit words
__m128i b = _mm_packs_epi16(a, a); // 16-bit words to bytes
return (uint8_t)_mm_movemask_epi8(b);
}
// to_Vec8ib: convert integer bitfield to boolean vector
static inline Vec8ib to_Vec8ib(uint8_t x) {
return Vec8ib(Vec8i(to_Vec4ib(x), to_Vec4ib(x>>4)));
}
// to_bits: convert boolean vector to integer bitfield
static inline uint8_t to_bits(Vec4qb const & x) {
uint32_t a = _mm256_movemask_epi8(x);
return ((a & 1) | ((a >> 7) & 2)) | (((a >> 14) & 4) | ((a >> 21) & 8));
}
// to_Vec4qb: convert integer bitfield to boolean vector
static inline Vec4qb to_Vec4qb(uint8_t x) {
return Vec4qb(Vec4q(-(x&1), -((x>>1)&1), -((x>>2)&1), -((x>>3)&1)));
}
#else // function prototypes here only
// to_bits: convert boolean vector to integer bitfield
static inline uint8_t to_bits(Vec8ib x);
// to_Vec8ib: convert integer bitfield to boolean vector
static inline Vec8ib to_Vec8ib(uint8_t x);
// to_bits: convert boolean vector to integer bitfield
static inline uint8_t to_bits(Vec4qb x);
// to_Vec4qb: convert integer bitfield to boolean vector
static inline Vec4qb to_Vec4qb(uint8_t x);
#endif // INSTRSET < 9 || MAX_VECTOR_SIZE < 512
#ifdef VCL_NAMESPACE
}
#endif
#endif // VECTORI256_H
|