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

\input texinfo @c * texinfo *
@c % $Id: fftw.texi,v 1.256 2003/01/15 21:09:33 stevenj Exp $
@c %**start of header
@setfilename fftw.info
@settitle FFTW
@c %**end of header
@dircategory Science
@direntry
* FFTW: (fftw). FFTW User's Manual
@end direntry
@include version.texi
@setchapternewpage odd
@c define constant index (ct)
@defcodeindex ct
@syncodeindex ct fn
@syncodeindex vr fn
@syncodeindex pg fn
@syncodeindex tp fn
@c define foreign function index (ff)
@defcodeindex ff
@syncodeindex ff cp
@c define foreign constant index (fc)
@defcodeindex fc
@syncodeindex fc cp
@c define foreign program index (fp)
@defcodeindex fp
@syncodeindex fp cp
@ifinfo
This is the FFTW User's manual.
Copyright @copyright{} 19971999 Massachusetts Institute of Technology
Permission is granted to make and distribute verbatim copies of this
manual provided the copyright notice and this permission notice are
preserved on all copies.
Permission is granted to copy and distribute modified versions of this
manual under the conditions for verbatim copying, provided that the
entire resulting derived work is distributed under the terms of a
permission notice identical to this one.
Permission is granted to copy and distribute translations of this manual
into another language, under the above conditions for modified versions,
except that this permission notice may be stated in a translation
approved by the Free Software Foundation.
@end ifinfo
@titlepage
@sp 10
@comment The title is printed in a large font.
@title FFTW User's Manual
@subtitle For version @value{VERSION}, @value{UPDATED}
@author Matteo Frigo
@author Steven G. Johnson
@c The following two commands start the copyright page.
@page
@vskip 0pt plus 1filll
Copyright @copyright{} 19971999 Massachusetts Institute of Technology.
Permission is granted to make and distribute verbatim copies of this
manual provided the copyright notice and this permission notice are
preserved on all copies.
Permission is granted to copy and distribute modified versions of this
manual under the conditions for verbatim copying, provided that the
entire resulting derived work is distributed under the terms of a
permission notice identical to this one.
Permission is granted to copy and distribute translations of this manual
into another language, under the above conditions for modified versions,
except that this permission notice may be stated in a translation
approved by the Free Software Foundation.
@end titlepage
@node Top, Introduction, (dir), (dir)
@ifinfo
@top FFTW User Manual
Welcome to FFTW, the Fastest Fourier Transform in the West. FFTW is a
collection of fast C routines to compute the discrete Fourier transform.
This manual documents FFTW version @value{VERSION}.
@end ifinfo
@menu
* Introduction::
* Tutorial::
* FFTW Reference::
* Parallel FFTW::
* Calling FFTW from Fortran::
* Installation and Customization::
* Acknowledgments::
* License and Copyright::
* Concept Index::
* Library Index::
@detailmenu  The Detailed Node Listing 
Tutorial
* Complex Onedimensional Transforms Tutorial::
* Complex Multidimensional Transforms Tutorial::
* Real Onedimensional Transforms Tutorial::
* Real Multidimensional Transforms Tutorial::
* Multidimensional Array Format::
* Words of Wisdom::
Multidimensional Array Format
* Rowmajor Format::
* Columnmajor Format::
* Static Arrays in C::
* Dynamic Arrays in C::
* Dynamic Arrays in CThe Wrong Way::
Words of Wisdom
* Caveats in Using Wisdom:: What you should worry about in using wisdom
* Importing and Exporting Wisdom:: I/O of wisdom to disk and other media
FFTW Reference
* Data Types:: real, complex, and halfcomplex numbers
* Onedimensional Transforms Reference::
* Multidimensional Transforms Reference::
* Real Onedimensional Transforms Reference::
* Real Multidimensional Transforms Reference::
* Wisdom Reference::
* Memory Allocator Reference::
* Thread safety::
Onedimensional Transforms Reference
* fftw_create_plan:: Plan Creation
* Discussion on Specific Plans::
* fftw:: Plan Execution
* fftw_destroy_plan:: Plan Destruction
* What FFTW Really Computes:: Definition of the DFT.
Multidimensional Transforms Reference
* fftwnd_create_plan:: Plan Creation
* fftwnd:: Plan Execution
* fftwnd_destroy_plan:: Plan Destruction
* What FFTWND Really Computes::
Real Onedimensional Transforms Reference
* rfftw_create_plan:: Plan Creation
* rfftw:: Plan Execution
* rfftw_destroy_plan:: Plan Destruction
* What RFFTW Really Computes::
Real Multidimensional Transforms Reference
* rfftwnd_create_plan:: Plan Creation
* rfftwnd:: Plan Execution
* Array Dimensions for Real Multidimensional Transforms::
* Strides in Inplace RFFTWND::
* rfftwnd_destroy_plan:: Plan Destruction
* What RFFTWND Really Computes::
Wisdom Reference
* fftw_export_wisdom::
* fftw_import_wisdom::
* fftw_forget_wisdom::
Parallel FFTW
* Multithreaded FFTW::
* MPI FFTW::
Multithreaded FFTW
* Installation and Supported Hardware/Software::
* Usage of Multithreaded FFTW::
* How Many Threads to Use?::
* Using Multithreaded FFTW in a Multithreaded Program::
* Tips for Optimal Threading::
MPI FFTW
* MPI FFTW Installation::
* Usage of MPI FFTW for Complex Multidimensional Transforms::
* MPI Data Layout::
* Usage of MPI FFTW for Real Multidimensional Transforms::
* Usage of MPI FFTW for Complex Onedimensional Transforms::
* MPI Tips::
Calling FFTW from Fortran
* Wrapper Routines::
* FFTW Constants in Fortran::
* Fortran Examples::
Installation and Customization
* Installation on Unix::
* Installation on nonUnix Systems::
* Installing FFTW in both single and double precision::
* gcc and Pentium hacks::
* Customizing the timer::
* Generating your own code::
@end detailmenu
@end menu
@c ************************************************************
@node Introduction, Tutorial, Top, Top
@chapter Introduction
This manual documents version @value{VERSION} of FFTW, the @emph{Fastest
Fourier Transform in the West}. FFTW is a comprehensive collection of
fast C routines for computing the discrete Fourier transform (DFT) in
one or more dimensions, of both real and complex data, and of arbitrary
input size. FFTW also includes parallel transforms for both shared and
distributedmemory systems. We assume herein that the reader is already
familiar with the properties and uses of the DFT that are relevant to
her application. Otherwise, see e.g. @cite{The Fast Fourier Transform}
by E. O. Brigham (PrenticeHall, Englewood Cliffs, NJ, 1974).
@uref{http://www.fftw.org, Our web page} also has links to
FFTrelated information online.
@cindex FFTW
FFTW is usually faster (and sometimes much faster) than all other
freelyavailable Fourier transform programs found on the Net. For
transforms whose size is a power of two, it compares favorably with the
FFT codes in Sun's Performance Library and IBM's ESSL library, which are
targeted at specific machines. Moreover, FFTW's performance is
@emph{portable}. Indeed, FFTW is unique in that it automatically adapts
itself to your machine, your cache, the size of your memory, the number
of registers, and all the other factors that normally make it impossible
to optimize a program for more than one machine. An extensive
comparison of FFTW's performance with that of other Fourier transform
codes has been made. The results are available on the Web at
@uref{http://theory.lcs.mit.edu/~benchfft, the benchFFT home page}.
@cindex benchmark
@fpindex benchfft
In order to use FFTW effectively, you need to understand one basic
concept of FFTW's internal structure. FFTW does not used a fixed
algorithm for computing the transform, but it can adapt the DFT
algorithm to details of the underlying hardware in order to achieve best
performance. Hence, the computation of the transform is split into two
phases. First, FFTW's @dfn{planner} is called, which ``learns'' the
@cindex plan
fastest way to compute the transform on your machine. The planner
@cindex planner
produces a data structure called a @dfn{plan} that contains this
information. Subsequently, the plan is passed to FFTW's @dfn{executor},
@cindex executor
along with an array of input data. The executor computes the actual
transform, as dictated by the plan. The plan can be reused as many
times as needed. In typical highperformance applications, many
transforms of the same size are computed, and consequently a
relativelyexpensive initialization of this sort is acceptable. On the
other hand, if you need a single transform of a given size, the onetime
cost of the planner becomes significant. For this case, FFTW provides
fast planners based on heuristics or on previously computed plans.
The pattern of planning/execution applies to all four operation modes of
FFTW, that is, @w{I) onedimensional} complex transforms (FFTW), @w{II)
multidimensional} complex transforms (FFTWND), @w{III) onedimensional}
transforms of real data (RFFTW), @w{IV) multidimensional} transforms of
real data (RFFTWND). Each mode comes with its own planner and executor.
Besides the automatic performance adaptation performed by the planner,
it is also possible for advanced users to customize FFTW for their
special needs. As distributed, FFTW works most efficiently for arrays
whose size can be factored into small primes (@math{2}, @math{3},
@math{5}, and @math{7}), and uses a slower generalpurpose routine for
other factors. FFTW, however, comes with a code generator that can
produce fast C programs for any particular array size you may care
about.
@cindex code generator
For example, if you need transforms of size
@ifinfo
@math{513 = 19 x 3^3},
@end ifinfo
@tex
$513 = 19 \cdot 3^3$,
@end tex
@ifhtml
513 = 19*3<sup>3</sup>,
@end ifhtml
you can customize FFTW to support the factor @math{19} efficiently.
FFTW can exploit multiple processors if you have them. FFTW comes with
a sharedmemory implementation on top of POSIX (and similar) threads, as
well as a distributedmemory implementation based on MPI.
@cindex parallel transform
@cindex threads
@cindex MPI
We also provide an experimental parallel implementation written in Cilk,
@emph{the superior programming tool of choice for discriminating
hackers} (Olin Shivers). (See @uref{http://supertech.lcs.mit.edu/cilk,
the Cilk home page}.)
@cindex Cilk
For more information regarding FFTW, see the paper, ``The Fastest
Fourier Transform in the West,'' by M. Frigo and S. G. Johnson, which is
the technical report MITLCSTR728 (Sep. '97). See also, ``FFTW: An
Adaptive Software Architecture for the FFT,'' by M. Frigo and
S. G. Johnson, which appeared in the 23rd International Conference on
Acoustics, Speech, and Signal Processing (@cite{Proc. ICASSP 1998}
@b{3}, p. 1381). The code generator is described in the paper ``A Fast
Fourier Transform Compiler'',
@cindex compiler
by M. Frigo, to appear in the @cite{Proceedings of the 1999 ACM SIGPLAN
Conference on Programming Language Design and Implementation (PLDI),
Atlanta, Georgia, May 1999}. These papers, along with the latest
version of FFTW, the FAQ, benchmarks, and other links, are available at
@uref{http://www.fftw.org, the FFTW home page}. The current
version of FFTW incorporates many good ideas from the past thirty years
of FFT literature. In one way or another, FFTW uses the CooleyTukey
algorithm, the Prime Factor algorithm, Rader's algorithm for prime
sizes, and the splitradix algorithm (with a variation due to Dan
Bernstein). Our code generator also produces new algorithms that we do
not yet completely understand.
@cindex algorithm
The reader is referred to the cited papers for the appropriate
references.
The rest of this manual is organized as follows. We first discuss the
sequential (oneprocessor) implementation. We start by describing the
basic features of FFTW in @ref{Tutorial}. This discussion includes the
storage scheme of multidimensional arrays (@ref{Multidimensional Array
Format}) and FFTW's mechanisms for storing plans on disk (@ref{Words of
Wisdom}). Next, @ref{FFTW Reference} provides comprehensive
documentation of all FFTW's features. Parallel transforms are discussed
in their own chapter @ref{Parallel FFTW}. Fortran programmers can also
use FFTW, as described in @ref{Calling FFTW from Fortran}.
@ref{Installation and Customization} explains how to install FFTW in
your computer system and how to adapt FFTW to your needs. License and
copyright information is given in @ref{License and Copyright}. Finally,
we thank all the people who helped us in @ref{Acknowledgments}.
@c ************************************************************
@node Tutorial, FFTW Reference, Introduction, Top
@chapter Tutorial
@cindex Tutorial
This chapter describes the basic usage of FFTW, i.e., how to compute the
Fourier transform of a single array. This chapter tells the truth, but
not the @emph{whole} truth. Specifically, FFTW implements additional
routines and flags, providing extra functionality, that are not
documented here. @xref{FFTW Reference}, for more complete information.
(Note that you need to compile and install FFTW before you can use it in
a program. @xref{Installation and Customization}, for the details of
the installation.)
Here, we assume a default installation of FFTW. In some installations
(particulary from binary packages), the FFTW header files and libraries
are prefixed with @samp{@code{d}} or @samp{@code{s}} to indicate
versions in double or single precision, respectively. The usage of FFTW
in that case is the same, except that @code{#include} directives and
link commands must use the appropriate prefix. @xref{Installing FFTW in
both single and double precision}, for more information.
This tutorial chapter is structured as follows. @ref{Complex
Onedimensional Transforms Tutorial} describes the basic usage of the
onedimensional transform of complex data. @ref{Complex
Multidimensional Transforms Tutorial} describes the basic usage of the
multidimensional transform of complex data. @ref{Real Onedimensional
Transforms Tutorial} describes the onedimensional transform of real
data and its inverse. Finally, @ref{Real Multidimensional Transforms
Tutorial} describes the multidimensional transform of real data and its
inverse. We recommend that you read these sections in the order that
they are presented. We then discuss two topics in detail. In
@ref{Multidimensional Array Format}, we discuss the various
alternatives for storing multidimensional arrays in memory. @ref{Words
of Wisdom} shows how you can save FFTW's plans for future use.
@menu
* Complex Onedimensional Transforms Tutorial::
* Complex Multidimensional Transforms Tutorial::
* Real Onedimensional Transforms Tutorial::
* Real Multidimensional Transforms Tutorial::
* Multidimensional Array Format::
* Words of Wisdom::
@end menu
@node Complex Onedimensional Transforms Tutorial, Complex Multidimensional Transforms Tutorial, Tutorial, Tutorial
@section Complex Onedimensional Transforms Tutorial
@cindex complex onedimensional transform
@cindex complex transform
The basic usage of FFTW is simple. A typical call to FFTW looks like:
@example
#include <fftw.h>
...
@{
fftw_complex in[N], out[N];
fftw_plan p;
...
p = fftw_create_plan(N, FFTW_FORWARD, FFTW_ESTIMATE);
...
fftw_one(p, in, out);
...
fftw_destroy_plan(p);
@}
@end example
The first thing we do is to create a @dfn{plan}, which is an object
@cindex plan
that contains all the data that FFTW needs to compute the FFT, using the
following function:
@example
fftw_plan fftw_create_plan(int n, fftw_direction dir, int flags);
@end example
@findex fftw_create_plan
@findex fftw_direction
@tindex fftw_plan
The first argument, @code{n}, is the size of the transform you are
trying to compute. The size @code{n} can be any positive integer, but
sizes that are products of small factors are transformed most
efficiently. The second argument, @code{dir}, can be either
@code{FFTW_FORWARD} or @code{FFTW_BACKWARD}, and indicates the direction
of the transform you
@ctindex FFTW_FORWARD
@ctindex FFTW_BACKWARD
are interested in. Alternatively, you can use the sign of the exponent
in the transform, @math{1} or @math{+1}, which corresponds to
@code{FFTW_FORWARD} or @code{FFTW_BACKWARD} respectively. The
@code{flags} argument is either @code{FFTW_MEASURE} or
@cindex flags
@code{FFTW_ESTIMATE}. @code{FFTW_MEASURE} means that FFTW actually runs
@ctindex FFTW_MEASURE
and measures the execution time of several FFTs in order to find the
best way to compute the transform of size @code{n}. This may take some
time, depending on your installation and on the precision of the timer
in your machine. @code{FFTW_ESTIMATE}, on the contrary, does not run
any computation, and just builds a
@ctindex FFTW_ESTIMATE
reasonable plan, which may be suboptimal. In other words, if your
program performs many transforms of the same size and initialization
time is not important, use @code{FFTW_MEASURE}; otherwise use the
estimate. (A compromise between these two extremes exists. @xref{Words
of Wisdom}.)
Once the plan has been created, you can use it as many times as you like
for transforms on arrays of the same size. When you are done with the
plan, you deallocate it by calling @code{fftw_destroy_plan(plan)}.
@findex fftw_destroy_plan
The transform itself is computed by passing the plan along with the
input and output arrays to @code{fftw_one}:
@example
void fftw_one(fftw_plan plan, fftw_complex *in, fftw_complex *out);
@end example
@findex fftw_one
Note that the transform is out of place: @code{in} and @code{out} must
point to distinct arrays. It operates on data of type
@code{fftw_complex}, a data structure with real (@code{in[i].re}) and
imaginary (@code{in[i].im}) floatingpoint components. The @code{in}
and @code{out} arrays should have the length specified when the plan was
created. An alternative function, @code{fftw}, allows you to
efficiently perform multiple and/or strided transforms (@pxref{FFTW
Reference}).
@tindex fftw_complex
The DFT results are stored inorder in the array @code{out}, with the
zerofrequency (DC) component in @code{out[0]}.
@cindex frequency
The array @code{in} is not modified. Users should note that FFTW
computes an unnormalized DFT, the sign of whose exponent is given by the
@code{dir} parameter of @code{fftw_create_plan}. Thus, computing a
forward followed by a backward transform (or vice versa) results in the
original array scaled by @code{n}. @xref{What FFTW Really Computes},
for the definition of DFT.
@cindex normalization
A program using FFTW should be linked with @code{lfftw lm} on Unix
systems, or with the FFTW and standard math libraries in general.
@cindex linking on Unix
@node Complex Multidimensional Transforms Tutorial, Real Onedimensional Transforms Tutorial, Complex Onedimensional Transforms Tutorial, Tutorial
@section Complex Multidimensional Transforms Tutorial
@cindex complex multidimensional transform
@cindex multidimensional transform
FFTW can also compute transforms of any number of dimensions
(@dfn{rank}). The syntax is similar to that for the onedimensional
@cindex rank
transforms, with @samp{fftw_} replaced by @samp{fftwnd_} (which stands
for ``@code{fftw} in @code{N} dimensions'').
As before, we @code{#include <fftw.h>} and create a plan for the
transforms, this time of type @code{fftwnd_plan}:
@example
fftwnd_plan fftwnd_create_plan(int rank, const int *n,
fftw_direction dir, int flags);
@end example
@tindex fftwnd_plan
@tindex fftw_direction
@findex fftwnd_create_plan
@code{rank} is the dimensionality of the array, and can be any
nonnegative integer. The next argument, @code{n}, is a pointer to an
integer array of length @code{rank} containing the (positive) sizes of
each dimension of the array. (Note that the array will be stored in
rowmajor order. @xref{Multidimensional Array Format}, for information
on rowmajor order.) The last two parameters are the same as in
@code{fftw_create_plan}. We now, however, have an additional possible
flag, @code{FFTW_IN_PLACE}, since @code{fftwnd} supports true inplace
@cindex flags
@ctindex FFTW_IN_PLACE
@findex fftwnd
transforms. Multiple flags are combined using a bitwise @dfn{or}
(@samp{}). (An @dfn{inplace} transform is one in which the output
data overwrite the input data. It thus requires half as much memory
asand is often faster thanits opposite, an @dfn{outofplace}
transform.)
@cindex inplace transform
@cindex outofplace transform
For two and threedimensional transforms, FFTWND provides alternative
routines that accept the sizes of each dimension directly, rather than
indirectly through a rank and an array of sizes. These are otherwise
identical to @code{fftwnd_create_plan}, and are sometimes more
convenient:
@example
fftwnd_plan fftw2d_create_plan(int nx, int ny,
fftw_direction dir, int flags);
fftwnd_plan fftw3d_create_plan(int nx, int ny, int nz,
fftw_direction dir, int flags);
@end example
@findex fftw2d_create_plan
@findex fftw3d_create_plan
Once the plan has been created, you can use it any number of times for
transforms of the same size. When you do not need a plan anymore, you
can deallocate the plan by calling @code{fftwnd_destroy_plan(plan)}.
@findex fftwnd_destroy_plan
Given a plan, you can compute the transform of an array of data by
calling:
@example
void fftwnd_one(fftwnd_plan plan, fftw_complex *in, fftw_complex *out);
@end example
@findex fftwnd_one
Here, @code{in} and @code{out} point to multidimensional arrays in
rowmajor order, of the size specified when the plan was created. In
the case of an inplace transform, the @code{out} parameter is ignored
and the output data are stored in the @code{in} array. The results are
stored inorder, unnormalized, with the zerofrequency component in
@code{out[0]}.
@cindex frequency
A forward followed by a backward transform (or viceversa) yields the
original data multiplied by the size of the array (i.e. the product of
the dimensions). @xref{What FFTWND Really Computes}, for a discussion
of what FFTWND computes.
@cindex normalization
For example, code to perform an inplace FFT of a threedimensional
array might look like:
@example
#include <fftw.h>
...
@{
fftw_complex in[L][M][N];
fftwnd_plan p;
...
p = fftw3d_create_plan(L, M, N, FFTW_FORWARD,
FFTW_MEASURE  FFTW_IN_PLACE);
...
fftwnd_one(p, &in[0][0][0], NULL);
...
fftwnd_destroy_plan(p);
@}
@end example
Note that @code{in} is a staticallydeclared array, which is
automatically in rowmajor order, but we must take the address of the
first element in order to fit the type expected by @code{fftwnd_one}.
(@xref{Multidimensional Array Format}.)
@node Real Onedimensional Transforms Tutorial, Real Multidimensional Transforms Tutorial, Complex Multidimensional Transforms Tutorial, Tutorial
@section Real Onedimensional Transforms Tutorial
@cindex real transform
@cindex complex to real transform
@cindex RFFTW
If the input data are purely real, you can save roughly a factor of two
in both time and storage by using the @dfn{rfftw} transforms, which are
FFTs specialized for real data. The output of a such a transform is a
@dfn{halfcomplex} array, which consists of only half of the complex DFT
amplitudes (since the negativefrequency amplitudes for real data are
the complex conjugate of the positivefrequency amplitudes).
@cindex halfcomplex array
In exchange for these speed and space advantages, the user sacrifices
some of the simplicity of FFTW's complex transforms. First of all, to
allow maximum performance, the output format of the onedimensional real
transforms is different from that used by the multidimensional
transforms. Second, the inverse transform (halfcomplex to real) has the
sideeffect of destroying its input array. Neither of these
inconveniences should pose a serious problem for users, but it is
important to be aware of them. (Both the inconvenient output format
and the sideeffect of the inverse transform can be ameliorated for
onedimensional transforms, at the expense of some performance, by using
instead the multidimensional transform routines with a rank of one.)
The computation of the plan is similar to that for the complex
transforms. First, you @code{#include <rfftw.h>}. Then, you create a
plan (of type @code{rfftw_plan}) by calling:
@example
rfftw_plan rfftw_create_plan(int n, fftw_direction dir, int flags);
@end example
@tindex rfftw_plan
@tindex fftw_direction
@findex rfftw_create_plan
@code{n} is the length of the @emph{real} array in the transform (even
for halfcomplextoreal transforms), and can be any positive integer
(although sizes with small factors are transformed more efficiently).
@code{dir} is either @code{FFTW_REAL_TO_COMPLEX} or
@code{FFTW_COMPLEX_TO_REAL}.
@ctindex FFTW_REAL_TO_COMPLEX
@ctindex FFTW_COMPLEX_TO_REAL
The @code{flags} parameter is the same as in @code{fftw_create_plan}.
Once created, a plan can be used for any number of transforms, and is
deallocated when you are done with it by calling
@code{rfftw_destroy_plan(plan)}.
@findex rfftw_destroy_plan
Given a plan, a realtocomplex or complextoreal transform is computed
by calling:
@example
void rfftw_one(rfftw_plan plan, fftw_real *in, fftw_real *out);
@end example
@findex rfftw_one
(Note that @code{fftw_real} is an alias for the floatingpoint type for
which FFTW was compiled.) Depending upon the direction of the plan,
either the input or the output array is halfcomplex, and is stored in
the following format:
@cindex halfcomplex array
@tex
$$
r_0, r_1, r_2, \ldots, r_{n/2}, i_{(n+1)/21}, \ldots, i_2, i_1
$$
@end tex
@ifinfo
r0, r1, r2, r(n/2), i((n+1)/21), ..., i2, i1
@end ifinfo
@ifhtml
<p align=center>
r<sub>0</sub>, r<sub>1</sub>, r<sub>2</sub>, ..., r<sub>n/2</sub>, i<sub>(n+1)/21</sub>, ..., i<sub>2</sub>, i<sub>1</sub>
</p>
@end ifhtml
Here,
@ifinfo
rk
@end ifinfo
@tex
$r_k$
@end tex
@ifhtml
r<sub>k</sub>
@end ifhtml
is the real part of the @math{k}th output, and
@ifinfo
ik
@end ifinfo
@tex
$i_k$
@end tex
@ifhtml
i<sub>k</sub>
@end ifhtml
is the imaginary part. (We follow here the C convention that integer
division is rounded down, e.g. @math{7 / 2 = 3}.) For a halfcomplex
array @code{hc[]}, the @math{k}th component has its real part in
@code{hc[k]} and its imaginary part in @code{hc[nk]}, with the
exception of @code{k} @code{==} @code{0} or @code{n/2} (the latter only
if n is even)in these two cases, the imaginary part is zero due to
symmetries of the realcomplex transform, and is not stored. Thus, the
transform of @code{n} real values is a halfcomplex array of length
@code{n}, and vice versa. @footnote{The output for the
multidimensional rfftw is a moreconventional array of
@code{fftw_complex} values, but the format here permitted us greater
efficiency in one dimension.} This is actually only half of the DFT
spectrum of the data. Although the other half can be obtained by
complex conjugation, it is not required by many applications such as
convolution and filtering.
Like the complex transforms, the RFFTW transforms are unnormalized, so a
forward followed by a backward transform (or viceversa) yields the
original data scaled by the length of the array, @code{n}.
@cindex normalization
Let us reiterate here our warning that an @code{FFTW_COMPLEX_TO_REAL}
transform has the sideeffect of destroying its (halfcomplex) input.
The @code{FFTW_REAL_TO_COMPLEX} transform, however, leaves its (real)
input untouched, just as you would hope.
As an example, here is an outline of how you might use RFFTW to compute
the power spectrum of a real array (i.e. the squares of the absolute
values of the DFT amplitudes):
@cindex power spectrum
@example
#include <rfftw.h>
...
@{
fftw_real in[N], out[N], power_spectrum[N/2+1];
rfftw_plan p;
int k;
...
p = rfftw_create_plan(N, FFTW_REAL_TO_COMPLEX, FFTW_ESTIMATE);
...
rfftw_one(p, in, out);
power_spectrum[0] = out[0]*out[0]; /* DC component */
for (k = 1; k < (N+1)/2; ++k) /* (k < N/2 rounded up) */
power_spectrum[k] = out[k]*out[k] + out[Nk]*out[Nk];
if (N % 2 == 0) /* N is even */
power_spectrum[N/2] = out[N/2]*out[N/2]; /* Nyquist freq. */
...
rfftw_destroy_plan(p);
@}
@end example
Programs using RFFTW should link with @code{lrfftw lfftw lm} on Unix,
or with the FFTW, RFFTW, and math libraries in general.
@cindex linking on Unix
@node Real Multidimensional Transforms Tutorial, Multidimensional Array Format, Real Onedimensional Transforms Tutorial, Tutorial
@section Real Multidimensional Transforms Tutorial
@cindex real multidimensional transform
FFTW includes multidimensional transforms for real data of any rank.
As with the onedimensional real transforms, they save roughly a factor
of two in time and storage over complex transforms of the same size.
Also as in one dimension, these gains come at the expense of some
increase in complexitythe output format is different from the
onedimensional RFFTW (and is more similar to that of the complex FFTW)
and the inverse (complex to real) transforms have the sideeffect of
overwriting their input data.
To use the real multidimensional transforms, you first @code{#include
<rfftw.h>} and then create a plan for the size and direction of
transform that you are interested in:
@example
rfftwnd_plan rfftwnd_create_plan(int rank, const int *n,
fftw_direction dir, int flags);
@end example
@tindex rfftwnd_plan
@findex rfftwnd_create_plan
The first two parameters describe the size of the real data (not the
halfcomplex data, which will have different dimensions). The last two
parameters are the same as those for @code{rfftw_create_plan}. Just as
for fftwnd, there are two alternate versions of this routine,
@code{rfftw2d_create_plan} and @code{rfftw3d_create_plan}, that are
sometimes more convenient for two and threedimensional transforms.
@findex rfftw2d_create_plan
@findex rfftw3d_create_plan
Also as in fftwnd, rfftwnd supports true inplace transforms, specified
by including @code{FFTW_IN_PLACE} in the flags.
Once created, a plan can be used for any number of transforms, and is
deallocated by calling @code{rfftwnd_destroy_plan(plan)}.
Given a plan, the transform is computed by calling one of the following
two routines:
@example
void rfftwnd_one_real_to_complex(rfftwnd_plan plan,
fftw_real *in, fftw_complex *out);
void rfftwnd_one_complex_to_real(rfftwnd_plan plan,
fftw_complex *in, fftw_real *out);
@end example
@findex rfftwnd_one_real_to_complex
@findex rfftwnd_one_complex_to_real
As is clear from their names and parameter types, the former function is
for @code{FFTW_REAL_TO_COMPLEX} transforms and the latter is for
@code{FFTW_COMPLEX_TO_REAL} transforms. (We could have used only a
single routine, since the direction of the transform is encoded in the
plan, but we wanted to correctly express the datatypes of the
parameters.) The latter routine, as we discuss elsewhere, has the
sideeffect of overwriting its input (except when the rank of the array
is one). In both cases, the @code{out} parameter is ignored for
inplace transforms.
The format of the complex arrays deserves careful attention.
@cindex rfftwnd array format
Suppose that the real data has dimensions
@tex
$n_1 \times n_2 \times \cdots \times n_d$
@end tex
@ifinfo
n1 x n2 x ... x nd
@end ifinfo
@ifhtml
n<sub>1</sub> x n<sub>2</sub> x ... x n<sub>d</sub>
@end ifhtml
(in rowmajor order). Then, after a realtocomplex transform, the
output is an
@tex
$n_1 \times n_2 \times \cdots \times (n_d/2+1)$
@end tex
@ifinfo
n1 x n2 x ... x (nd/2+1)
@end ifinfo
@ifhtml
n<sub>1</sub> x n<sub>2</sub> x ... x (n<sub>d</sub>/2+1)
@end ifhtml
array of @code{fftw_complex} values in rowmajor order, corresponding to
slightly over half of the output of the corresponding complex transform.
(Note that the division is rounded down.) The ordering of the data is
otherwise exactly the same as in the complex case. (In principle, the
output could be exactly half the size of the complex transform output,
but in more than one dimension this requires too complicated a format to
be practical.) Note that, unlike the onedimensional RFFTW, the real
and imaginary parts of the DFT amplitudes are here stored together in
the natural way.
Since the complex data is slightly larger than the real data, some
complications arise for inplace transforms. In this case, the final
dimension of the real data must be padded with extra values to
accommodate the size of the complex datatwo extra if the last
dimension is even and one if it is odd.
@cindex padding
That is, the last dimension of the real data must physically contain
@tex
$2 (n_d/2+1)$
@end tex
@ifinfo
2 * (nd/2+1)
@end ifinfo
@ifhtml
2 * (n<sub>d</sub>/2+1)
@end ifhtml
@code{fftw_real} values (exactly enough to hold the complex data).
This physical array size does not, however, change the @emph{logical}
array sizeonly
@tex
$n_d$
@end tex
@ifinfo
nd
@end ifinfo
@ifhtml
n<sub>d</sub>
@end ifhtml
values are actually stored in the last dimension, and
@tex
$n_d$
@end tex
@ifinfo
nd
@end ifinfo
@ifhtml
n<sub>d</sub>
@end ifhtml
is the last dimension passed to @code{rfftwnd_create_plan}.
For example, consider the transform of a twodimensional real array of
size @code{nx} by @code{ny}. The output of the @code{rfftwnd} transform
is a twodimensional complex array of size @code{nx} by @code{ny/2+1},
where the @code{y} dimension has been cut nearly in half because of
redundancies in the output. Because @code{fftw_complex} is twice the
size of @code{fftw_real}, the output array is slightly bigger than the
input array. Thus, if we want to compute the transform in place, we
must @emph{pad} the input array so that it is of size @code{nx} by
@code{2*(ny/2+1)}. If @code{ny} is even, then there are two padding
elements at the end of each row (which need not be initialized, as they
are only used for output).
@ifhtml
The following illustration depicts the input and output arrays just
described, for both the outofplace and inplace transforms (with the
arrows indicating consecutive memory locations):
<p align=center><img src="rfftwnd.gif" width=389 height=583>
@end ifhtml
@tex
Figure 1 depicts the input and output arrays just
described, for both the outofplace and inplace transforms (with the
arrows indicating consecutive memory locations).
{
\pageinsert
\vfill
\vskip405pt
\hskip40pt
\special{psfile="rfftwnd.eps"
}
\vskip 24pt
Figure 1: Illustration of the data layout for real to complex
transforms.
\vfill
\endinsert}
@end tex
The RFFTWND transforms are unnormalized, so a forward followed by a
backward transform will result in the original data scaled by the number
of real data elementsthat is, the product of the (logical) dimensions
of the real data.
@cindex normalization
Below, we illustrate the use of RFFTWND by showing how you might use it
to compute the (cyclic) convolution of twodimensional real arrays
@code{a} and @code{b} (using the identity that a convolution corresponds
to a pointwise product of the Fourier transforms). For variety,
inplace transforms are used for the forward FFTs and an outofplace
transform is used for the inverse transform.
@cindex convolution
@cindex cyclic convolution
@example
#include <rfftw.h>
...
@{
fftw_real a[M][2*(N/2+1)], b[M][2*(N/2+1)], c[M][N];
fftw_complex *A, *B, C[M][N/2+1];
rfftwnd_plan p, pinv;
fftw_real scale = 1.0 / (M * N);
int i, j;
...
p = rfftw2d_create_plan(M, N, FFTW_REAL_TO_COMPLEX,
FFTW_ESTIMATE  FFTW_IN_PLACE);
pinv = rfftw2d_create_plan(M, N, FFTW_COMPLEX_TO_REAL,
FFTW_ESTIMATE);
/* aliases for accessing complex transform outputs: */
A = (fftw_complex*) &a[0][0];
B = (fftw_complex*) &b[0][0];
...
for (i = 0; i < M; ++i)
for (j = 0; j < N; ++j) @{
a[i][j] = ... ;
b[i][j] = ... ;
@}
...
rfftwnd_one_real_to_complex(p, &a[0][0], NULL);
rfftwnd_one_real_to_complex(p, &b[0][0], NULL);
for (i = 0; i < M; ++i)
for (j = 0; j < N/2+1; ++j) @{
int ij = i*(N/2+1) + j;
C[i][j].re = (A[ij].re * B[ij].re
 A[ij].im * B[ij].im) * scale;
C[i][j].im = (A[ij].re * B[ij].im
+ A[ij].im * B[ij].re) * scale;
@}
/* inverse transform to get c, the convolution of a and b;
this has the side effect of overwriting C */
rfftwnd_one_complex_to_real(pinv, &C[0][0], &c[0][0]);
...
rfftwnd_destroy_plan(p);
rfftwnd_destroy_plan(pinv);
@}
@end example
We access the complex outputs of the inplace transforms by casting
each real array to a @code{fftw_complex} pointer. Because this is a
``flat'' pointer, we have to compute the rowmajor index @code{ij}
explicitly in the convolution product loop.
@cindex rowmajor
In order to normalize the convolution, we must multiply by a scale
factorwe can do so either before or after the inverse transform, and
choose the former because it obviates the necessity of an additional
loop.
@cindex normalization
Notice the limits of the loops and the dimensions of the various arrays.
As with the onedimensional RFFTW, an outofplace
@code{FFTW_COMPLEX_TO_REAL} transform has the sideeffect of overwriting
its input array. (The realtocomplex transform, on the other hand,
leaves its input array untouched.) If you use RFFTWND for a rankone
transform, however, this sideeffect does not occur. Because of this
fact (and the simpler output format), users may find the RFFTWND
interface more convenient than RFFTW for onedimensional transforms.
However, RFFTWND in one dimension is slightly slower than RFFTW because
RFFTWND uses an extra buffer array internally.
@c 
@node Multidimensional Array Format, Words of Wisdom, Real Multidimensional Transforms Tutorial, Tutorial
@section Multidimensional Array Format
This section describes the format in which multidimensional arrays are
stored. We felt that a detailed discussion of this topic was necessary,
since it is often a source of confusion among users and several
different formats are common. Although the comments below refer to
@code{fftwnd}, they are also applicable to the @code{rfftwnd} routines.
@menu
* Rowmajor Format::
* Columnmajor Format::
* Static Arrays in C::
* Dynamic Arrays in C::
* Dynamic Arrays in CThe Wrong Way::
@end menu
@node Rowmajor Format, Columnmajor Format, Multidimensional Array Format, Multidimensional Array Format
@subsection Rowmajor Format
@cindex rowmajor
The multidimensional arrays passed to @code{fftwnd} are expected to be
stored as a single contiguous block in @dfn{rowmajor} order (sometimes
called ``C order''). Basically, this means that as you step through
adjacent memory locations, the first dimension's index varies most
slowly and the last dimension's index varies most quickly.
To be more explicit, let us consider an array of rank @math{d} whose
dimensions are
@tex
$n_1 \times n_2 \times n_3 \times \cdots \times n_d$.
@end tex
@ifinfo
n1 x n2 x n3 x ... x nd.
@end ifinfo
@ifhtml
n<sub>1</sub> x n<sub>2</sub> x n<sub>3</sub> x ... x n<sub>d</sub>.
@end ifhtml
Now, we specify a location in the array by a sequence of (zerobased) indices,
one for each dimension:
@tex
$(i_1, i_2, i_3, \ldots, i_d)$.
@end tex
@ifinfo
(i1, i2, ..., id).
@end ifinfo
@ifhtml
(i<sub>1</sub>, i<sub>2</sub>, i<sub>3</sub>,..., i<sub>d</sub>).
@end ifhtml
If the array is stored in rowmajor
order, then this element is located at the position
@tex
$i_d + n_d (i_{d1} + n_{d1} (\ldots + n_2 i_1))$.
@end tex
@ifinfo
id + nd * (id1 + nd1 * (... + n2 * i1)).
@end ifinfo
@ifhtml
i<sub>d</sub> + n<sub>d</sub> * (i<sub>d1</sub> + n<sub>d1</sub> * (... + n<sub>2</sub> * i<sub>1</sub>)).
@end ifhtml
Note that each element of the array must be of type @code{fftw_complex};
i.e. a (real, imaginary) pair of (doubleprecision) numbers. Note also
that, in @code{fftwnd}, the expression above is multiplied by the stride
to get the actual array indexthis is useful in situations where each
element of the multidimensional array is actually a data structure or
another array, and you just want to transform a single field. In most
cases, however, you use a stride of 1.
@cindex stride
@node Columnmajor Format, Static Arrays in C, Rowmajor Format, Multidimensional Array Format
@subsection Columnmajor Format
@cindex columnmajor
Readers from the Fortran world are used to arrays stored in
@dfn{columnmajor} order (sometimes called ``Fortran order''). This is
essentially the exact opposite of rowmajor order in that, here, the
@emph{first} dimension's index varies most quickly.
If you have an array stored in columnmajor order and wish to transform
it using @code{fftwnd}, it is quite easy to do. When creating the plan,
simply pass the dimensions of the array to @code{fftwnd_create_plan} in
@emph{reverse order}. For example, if your array is a rank three
@code{N x M x L} matrix in columnmajor order, you should pass the
dimensions of the array as if it were an @code{L x M x N} matrix (which
it is, from the perspective of @code{fftwnd}). This is done for you
automatically by the FFTW Fortran wrapper routines (@pxref{Calling FFTW
from Fortran}).
@cindex Fortrancallable wrappers
@node Static Arrays in C, Dynamic Arrays in C, Columnmajor Format, Multidimensional Array Format
@subsection Static Arrays in C
@cindex C multidimensional arrays
Multidimensional arrays declared statically (that is, at compile time,
not necessarily with the @code{static} keyword) in C are @emph{already}
in rowmajor order. You don't have to do anything special to transform
them. (@xref{Complex Multidimensional Transforms Tutorial}, for an
example of this sort of code.)
@node Dynamic Arrays in C, Dynamic Arrays in CThe Wrong Way, Static Arrays in C, Multidimensional Array Format
@subsection Dynamic Arrays in C
Often, especially for large arrays, it is desirable to allocate the
arrays dynamically, at runtime. This isn't too hard to do, although it
is not as straightforward for multidimensional arrays as it is for
onedimensional arrays.
Creating the array is simple: using a dynamicallocation routine like
@code{malloc}, allocate an array big enough to store N @code{fftw_complex}
values, where N is the product of the sizes of the array dimensions
(i.e. the total number of complex values in the array). For example,
here is code to allocate a 5x12x27 rank 3 array:
@ffindex malloc
@example
fftw_complex *an_array;
an_array = (fftw_complex *) malloc(5 * 12 * 27 * sizeof(fftw_complex));
@end example
Accessing the array elements, however, is more trickyyou can't simply
use multiple applications of the @samp{[]} operator like you could for
static arrays. Instead, you have to explicitly compute the offset into
the array using the formula given earlier for rowmajor arrays. For
example, to reference the @math{(i,j,k)}th element of the array
allocated above, you would use the expression @code{an_array[k + 27 * (j
+ 12 * i)]}.
This pain can be alleviated somewhat by defining appropriate macros, or,
in C++, creating a class and overloading the @samp{()} operator.
@node Dynamic Arrays in CThe Wrong Way, , Dynamic Arrays in C, Multidimensional Array Format
@subsection Dynamic Arrays in CThe Wrong Way
A different method for allocating multidimensional arrays in C is often
suggested that is incompatible with @code{fftwnd}: @emph{using it will
cause FFTW to die a painful death}. We discuss the technique here,
however, because it is so commonly known and used. This method is to
create arrays of pointers of arrays of pointers of @dots{}etcetera. For
example, the analogue in this method to the example above is:
@example
int i,j;
fftw_complex ***a_bad_array; /* another way to make a 5x12x27 array */
a_bad_array = (fftw_complex ***) malloc(5 * sizeof(fftw_complex **));
for (i = 0; i < 5; ++i) @{
a_bad_array[i] =
(fftw_complex **) malloc(12 * sizeof(fftw_complex *));
for (j = 0; j < 12; ++j)
a_bad_array[i][j] =
(fftw_complex *) malloc(27 * sizeof(fftw_complex));
@}
@end example
As you can see, this sort of array is inconvenient to allocate (and
deallocate). On the other hand, it has the advantage that the
@math{(i,j,k)}th element can be referenced simply by
@code{a_bad_array[i][j][k]}.
If you like this technique and want to maximize convenience in accessing
the array, but still want to pass the array to FFTW, you can use a
hybrid method. Allocate the array as one contiguous block, but also
declare an array of arrays of pointers that point to appropriate places
in the block. That sort of trick is beyond the scope of this
documentation; for more information on multidimensional arrays in C,
see the @code{comp.lang.c}
@uref{http://www.eskimo.com/~scs/Cfaq/s6.html, FAQ}.
@c 
@node Words of Wisdom, , Multidimensional Array Format, Tutorial
@section Words of Wisdom
@cindex wisdom
@cindex saving plans to disk
FFTW implements a method for saving plans to disk and restoring them.
In fact, what FFTW does is more general than just saving and loading
plans. The mechanism is called @dfn{@code{wisdom}}. Here, we describe
this feature at a high level. @xref{FFTW Reference}, for a less casual
(but more complete) discussion of how to use @code{wisdom} in FFTW.
Plans created with the @code{FFTW_MEASURE} option produce nearoptimal
FFT performance, but it can take a long time to compute a plan because
FFTW must actually measure the runtime of many possible plans and select
the best one. This is designed for the situations where so many
transforms of the same size must be computed that the startup time is
irrelevant. For short initialization times but slightly slower
transforms, we have provided @code{FFTW_ESTIMATE}. The @code{wisdom}
mechanism is a way to get the best of both worlds. There are, however,
certain caveats that the user must be aware of in using @code{wisdom}.
For this reason, @code{wisdom} is an optional feature which is not
enabled by default.
At its simplest, @code{wisdom} provides a way of saving plans to disk so
that they can be reused in other program runs. You create a plan with
the flags @code{FFTW_MEASURE} and @code{FFTW_USE_WISDOM}, and then save
the @code{wisdom} using @code{fftw_export_wisdom}:
@ctindex FFTW_USE_WISDOM
@example
plan = fftw_create_plan(..., ...  FFTW_MEASURE  FFTW_USE_WISDOM);
fftw_export_wisdom(...);
@end example
@findex fftw_export_wisdom
The next time you run the program, you can restore the @code{wisdom}
with @code{fftw_import_wisdom}, and then recreate the plan using the
same flags as before. This time, however, the same optimal plan will be
created very quickly without measurements. (FFTW still needs some time
to compute trigonometric tables, however.) The basic outline is:
@example
fftw_import_wisdom(...);
plan = fftw_create_plan(..., ...  FFTW_USE_WISDOM);
@end example
@findex fftw_import_wisdom
Wisdom is more than mere rote memorization, however. FFTW's
@code{wisdom} encompasses all of the knowledge and measurements that
were used to create the plan for a given size. Therefore, existing
@code{wisdom} is also applied to the creation of other plans of
different sizes.
Whenever a plan is created with the @code{FFTW_MEASURE} and
@code{FFTW_USE_WISDOM} flags, @code{wisdom} is generated. Thereafter,
plans for any transform with a similar factorization will be computed
more quickly, so long as they use the @code{FFTW_USE_WISDOM} flag. In
fact, for transforms with the same factors and of equal or lesser size,
no measurements at all need to be made and an optimal plan can be
created with negligible delay!
For example, suppose that you create a plan for
@tex
$N = 2^{16}$.
@end tex
@ifinfo
N = 2^16.
@end ifinfo
@ifhtml
N = 2<sup>16</sup>.
@end ifhtml
Then, for any equal or smaller power of two, FFTW can create a
plan (with the same direction and flags) quickly, using the
precomputed @code{wisdom}. Even for larger powers of two, or sizes that
are a power of two times some other prime factors, plans will be
computed more quickly than they would otherwise (although some
measurements still have to be made).
The @code{wisdom} is cumulative, and is stored in a global, private data
structure managed internally by FFTW. The storage space required is
minimal, proportional to the logarithm of the sizes the @code{wisdom} was
generated from. The @code{wisdom} can be forgotten (and its associated
memory freed) by a call to @code{fftw_forget_wisdom()}; otherwise, it is
remembered until the program terminates. It can also be exported to a
file, a string, or any other medium using @code{fftw_export_wisdom} and
restored during a subsequent execution of the program (or a different
program) using @code{fftw_import_wisdom} (these functions are described
below).
Because @code{wisdom} is incorporated into FFTW at a very low level, the
same @code{wisdom} can be used for onedimensional transforms,
multidimensional transforms, and even the parallel extensions to FFTW.
Just include @code{FFTW_USE_WISDOM} in the flags for whatever plans you
create (i.e., always plan wisely).
Plans created with the @code{FFTW_ESTIMATE} plan can use @code{wisdom},
but cannot generate it; only @code{FFTW_MEASURE} plans actually produce
@code{wisdom}. Also, plans can only use @code{wisdom} generated from
plans created with the same direction and flags. For example, a size
@code{42} @code{FFTW_BACKWARD} transform will not use @code{wisdom}
produced by a size @code{42} @code{FFTW_FORWARD} transform. The only
exception to this rule is that @code{FFTW_ESTIMATE} plans can use
@code{wisdom} from @code{FFTW_MEASURE} plans.
@menu
* Caveats in Using Wisdom:: What you should worry about in using wisdom
* Importing and Exporting Wisdom:: I/O of wisdom to disk and other media
@end menu
@node Caveats in Using Wisdom, Importing and Exporting Wisdom, Words of Wisdom, Words of Wisdom
@subsection Caveats in Using Wisdom
@cindex wisdom, problems with
@quotation
@ifhtml
<i>
@end ifhtml
For in much wisdom is much grief, and he that increaseth knowledge
increaseth sorrow.
@ifhtml
</i>
@end ifhtml
[Ecclesiastes 1:18]
@cindex Ecclesiastes
@end quotation
There are pitfalls to using @code{wisdom}, in that it can negate FFTW's
ability to adapt to changing hardware and other conditions. For example,
it would be perfectly possible to export @code{wisdom} from a program
running on one processor and import it into a program running on another
processor. Doing so, however, would mean that the second program would
use plans optimized for the first processor, instead of the one it is
running on.
It should be safe to reuse @code{wisdom} as long as the hardware and
program binaries remain unchanged. (Actually, the optimal plan may
change even between runs of the same binary on identical hardware, due
to differences in the virtual memory environment, etcetera. Users
seriously interested in performance should worry about this problem,
too.) It is likely that, if the same @code{wisdom} is used for two
different program binaries, even running on the same machine, the plans
may be suboptimal because of differing code alignments. It is
therefore wise to recreate @code{wisdom} every time an application is
recompiled. The more the underlying hardware and software changes
between the creation of @code{wisdom} and its use, the greater grows the
risk of suboptimal plans.
@node Importing and Exporting Wisdom, , Caveats in Using Wisdom, Words of Wisdom
@subsection Importing and Exporting Wisdom
@cindex wisdom, import and export
@example
void fftw_export_wisdom_to_file(FILE *output_file);
fftw_status fftw_import_wisdom_from_file(FILE *input_file);
@end example
@findex fftw_export_wisdom_to_file
@findex fftw_import_wisdom_from_file
@code{fftw_export_wisdom_to_file} writes the @code{wisdom} to
@code{output_file}, which must be a file open for
writing. @code{fftw_import_wisdom_from_file} reads the @code{wisdom}
from @code{input_file}, which must be a file open for reading, and
returns @code{FFTW_SUCCESS} if successful and @code{FFTW_FAILURE}
otherwise. In both cases, the file is left open and must be closed by
the caller. It is perfectly fine if other data lie before or after the
@code{wisdom} in the file, as long as the file is positioned at the
beginning of the @code{wisdom} data before import.
@example
char *fftw_export_wisdom_to_string(void);
fftw_status fftw_import_wisdom_from_string(const char *input_string)
@end example
@findex fftw_export_wisdom_to_string
@findex fftw_import_wisdom_from_string
@code{fftw_export_wisdom_to_string} allocates a string, exports the
@code{wisdom} to it in @code{NULL}terminated format, and returns a
pointer to the string. If there is an error in allocating or writing
the data, it returns @code{NULL}. The caller is responsible for
deallocating the string (with @code{fftw_free}) when she is done with
it. @code{fftw_import_wisdom_from_string} imports the @code{wisdom} from
@code{input_string}, returning @code{FFTW_SUCCESS} if successful and
@code{FFTW_FAILURE} otherwise.
Exporting @code{wisdom} does not affect the store of @code{wisdom}. Imported
@code{wisdom} supplements the current store rather than replacing it
(except when there is conflicting @code{wisdom}, in which case the older
@code{wisdom} is discarded). The format of the exported @code{wisdom} is
``nerdreadable'' LISPlike ASCII text; we will not document it here
except to note that it is insensitive to white space (interested users
can contact us for more details).
@cindex LISP
@cindex nerdreadable text
@xref{FFTW Reference}, for more information, and for a description of
how you can implement @code{wisdom} import/export for other media
besides files and strings.
The following is a brief example in which the @code{wisdom} is read from
a file, a plan is created (possibly generating more @code{wisdom}), and
then the @code{wisdom} is exported to a string and printed to
@code{stdout}.
@example
@{
fftw_plan plan;
char *wisdom_string;
FILE *input_file;
/* open file to read wisdom from */
input_file = fopen("sample.wisdom", "r");
if (FFTW_FAILURE == fftw_import_wisdom_from_file(input_file))
printf("Error reading wisdom!\n");
fclose(input_file); /* be sure to close the file! */
/* create a plan for N=64, possibly creating and/or using wisdom */
plan = fftw_create_plan(64,FFTW_FORWARD,
FFTW_MEASURE  FFTW_USE_WISDOM);
/* ... do some computations with the plan ... */
/* always destroy plans when you are done */
fftw_destroy_plan(plan);
/* write the wisdom to a string */
wisdom_string = fftw_export_wisdom_to_string();
if (wisdom_string != NULL) @{
printf("Accumulated wisdom: %s\n",wisdom_string);
/* Just for fun, destroy and restore the wisdom */
fftw_forget_wisdom(); /* all gone! */
fftw_import_wisdom_from_string(wisdom_string);
/* wisdom is back! */
fftw_free(wisdom_string); /* deallocate it since we're done */
@}
@}
@end example
@c ************************************************************
@node FFTW Reference, Parallel FFTW, Tutorial, Top
@chapter FFTW Reference
This chapter provides a complete reference for all sequential (i.e.,
oneprocessor) FFTW functions. We first define the data types upon
which FFTW operates, that is, real, complex, and ``halfcomplex'' numbers
(@pxref{Data Types}). Then, in four sections, we explain the FFTW
program interface for complex onedimensional transforms
(@pxref{Onedimensional Transforms Reference}), complex
multidimensional transforms (@pxref{Multidimensional Transforms
Reference}), and real onedimensional transforms (@pxref{Real
Onedimensional Transforms Reference}), real multidimensional
transforms (@pxref{Real Multidimensional Transforms Reference}).
@ref{Wisdom Reference} describes the @code{wisdom} mechanism for
exporting and importing plans. Finally, @ref{Memory Allocator
Reference} describes how to change FFTW's default memory allocator.
For parallel transforms, @xref{Parallel FFTW}.
@menu
* Data Types:: real, complex, and halfcomplex numbers
* Onedimensional Transforms Reference::
* Multidimensional Transforms Reference::
* Real Onedimensional Transforms Reference::
* Real Multidimensional Transforms Reference::
* Wisdom Reference::
* Memory Allocator Reference::
* Thread safety::
@end menu
@c 
@node Data Types, Onedimensional Transforms Reference, FFTW Reference, FFTW Reference
@section Data Types
@cindex real number
@cindex complex number
@cindex halfcomplex array
The routines in the FFTW package use three main kinds of data types.
@dfn{Real} and @dfn{complex} numbers should be already known to the
reader. We also use the term @dfn{halfcomplex} to describe complex
arrays in a special packed format used by the onedimensional real
transforms (taking advantage of the @dfn{hermitian} symmetry that arises
in those cases).
By including @code{<fftw.h>} or @code{<rfftw.h>}, you will have access
to the following definitions:
@example
typedef double fftw_real;
typedef struct @{
fftw_real re, im;
@} fftw_complex;
#define c_re(c) ((c).re)
#define c_im(c) ((c).im)
@end example
@tindex fftw_real
@tindex fftw_complex
All FFTW operations are performed on the @code{fftw_real} and
@code{fftw_complex} data types. For @code{fftw_complex} numbers, the
two macros @code{c_re} and @code{c_im} retrieve, respectively, the real
and imaginary parts of the number.
A @dfn{real array} is an array of real numbers. A @dfn{complex array}
is an array of complex numbers. A onedimensional array @math{X} of
@math{n} complex numbers is @dfn{hermitian} if the following property
holds:
@tex
for all $0 \leq i < n$, we have $X_i = X^{*}_{ni}$, where
$x^*$ denotes the complex conjugate of $x$.
@end tex
@ifinfo
for all @math{0 <= i < n}, we have @math{X[i] = conj(X[ni])}.
@end ifinfo
@ifhtml
for all 0 <= i < n, we have X<sub>i</sub> = conj(X<sub>ni</sub>)}.
@end ifhtml
Hermitian arrays are relevant to FFTW because the Fourier transform of a
real array is hermitian.
Because of its symmetry, a hermitian array can be stored in half the
space of a complex array of the same size. FFTW's onedimensional real
transforms store hermitian arrays as @dfn{halfcomplex} arrays. A
halfcomplex array of size @math{n} is
@cindex hermitian array
a onedimensional array of @math{n} @code{fftw_real} numbers. A
hermitian array @math{X} in stored into a halfcomplex array @math{Y} as
follows.
@tex
For all integers $i$ such that $0 \leq i \leq n / 2$, we have $Y_i :=
\hbox{Re}(X_i)$. For all integers $i$ such that $0 < i < n / 2$,
we have $Y_{n  i} := \hbox{Im}(X_i)$.
@end tex
@ifinfo
For all integers @math{i} such that @math{0 <= i <= n / 2}, we have
@math{Y[i] = Re(X[i])}. For all integers @math{i} such that @math{0 <
i < n / 2}, we have @math{Y[ni] = Im(X[i])}.
@end ifinfo
@ifhtml
For all integers i such that 0 <= i <= n / 2, we have
Y<sub>i</sub> = Re(X<sub>i</sub>). For all integers i such that 0
< i < n / 2, we have Y<sub>ni</sub> = Im(X<sub>i</sub>).
@end ifhtml
We now illustrate halfcomplex storage for @math{n = 4} and @math{n = 5},
since the scheme depends on the parity of @math{n}. Let @math{n = 4}.
In this case, we have
@tex
$Y_0 := \hbox{Re}(X_0)$, $Y_1 := \hbox{Re}(X_1)$,
$Y_2 := \hbox{Re}(X_2)$, and $Y_3 := \hbox{Im}(X_1)$.
@end tex
@ifinfo
@math{Y[0] = Re(X[0])}, @math{Y[1] = Re(X[1])},
@math{Y[2] = Re(X[2])}, and @math{Y[3] = Im(X[1])}.
@end ifinfo
@ifhtml
Y<sub>0</sub> = Re(X<sub>0</sub>), Y<sub>1</sub> = Re(X<sub>1</sub>),
Y<sub>2</sub> = Re(X<sub>2</sub>), and Y<sub>3</sub> = Im(X<sub>1</sub>).
@end ifhtml
Let now @math{n = 5}. In this case, we have
@tex
$Y_0 := \hbox{Re}(X_0)$, $Y_1 := \hbox{Re}(X_1)$,
$Y_2 := \hbox{Re}(X_2)$, $Y_3 := \hbox{Im}(X_2)$, and
$Y_4 := \hbox{Im}(X_1)$.
@end tex
@ifinfo
@math{Y[0] = Re(X[0])}, @math{Y[1] = Re(X[1])},
@math{Y[2] = Re(X[2])}, @math{Y[3] = Im(X[2])}, and
@math{Y[4] = Im(X[1])}.
@end ifinfo
@ifhtml
Y<sub>0</sub> = Re(X<sub>0</sub>), Y<sub>1</sub> = Re(X<sub>1</sub>),
Y<sub>2</sub> = Re(X<sub>2</sub>), Y<sub>3</sub> = Im(X<sub>2</sub>),
and Y<sub>4</sub> = Im(X<sub>1</sub>).
@end ifhtml
@cindex floatingpoint precision
By default, the type @code{fftw_real} equals the C type @code{double}.
To work in single precision rather than double precision, @code{#define}
the symbol @code{FFTW_ENABLE_FLOAT} in @code{fftw.h} and then recompile
the library. On Unix systems, you can instead use @code{configure
enablefloat} at installation time (@pxref{Installation and
Customization}).
@fpindex configure
@ctindex FFTW_ENABLE_FLOAT
In version 1 of FFTW, the data types were called @code{FFTW_REAL} and
@code{FFTW_COMPLEX}. We changed the capitalization for consistency with
the rest of FFTW's conventions. The old names are still supported, but
their use is deprecated.
@tindex FFTW_REAL
@tindex FFTW_COMPLEX
@c 
@node Onedimensional Transforms Reference, Multidimensional Transforms Reference, Data Types, FFTW Reference
@section Onedimensional Transforms Reference
The onedimensional complex routines are generally prefixed with
@code{fftw_}. Programs using FFTW should be linked with @code{lfftw
lm} on Unix systems, or with the FFTW and standard math libraries in
general.
@menu
* fftw_create_plan:: Plan Creation
* Discussion on Specific Plans::
* fftw:: Plan Execution
* fftw_destroy_plan:: Plan Destruction
* What FFTW Really Computes:: Definition of the DFT.
@end menu
@node fftw_create_plan, Discussion on Specific Plans, Onedimensional Transforms Reference, Onedimensional Transforms Reference
@subsection Plan Creation for Onedimensional Transforms
@example
#include <fftw.h>
fftw_plan fftw_create_plan(int n, fftw_direction dir,
int flags);
fftw_plan fftw_create_plan_specific(int n, fftw_direction dir,
int flags,
fftw_complex *in, int istride,
fftw_complex *out, int ostride);
@end example
@tindex fftw_plan
@tindex fftw_direction
@findex fftw_create_plan
@findex fftw_create_plan_specific
The function @code{fftw_create_plan} creates a plan, which is
a data structure containing all the information that @code{fftw}
needs in order to compute the 1D Fourier transform. You can
create as many plans as you need, but only one plan for a given
array size is required (a plan can be reused many times).
@code{fftw_create_plan} returns a valid plan, or @code{NULL}
if, for some reason, the plan can't be created. In the
default installation, this cannot happen, but it is possible
to configure FFTW in such a way that some input sizes are
forbidden, and FFTW cannot create a plan.
The @code{fftw_create_plan_specific} variant takes as additional
arguments specific input/output arrays and their strides. For the last
four arguments, you should pass the arrays and strides that you will
eventually be passing to @code{fftw}. The resulting plans will be
optimized for those arrays and strides, although they may be used on
other arrays as well. Note: the contents of the in and out arrays are
@emph{destroyed} by the specific planner (the initial contents are
ignored, so the arrays need not have been initialized).
@subsubheading Arguments
@itemize @bullet
@item
@code{n} is the size of the transform. It can be
any positive integer.
@itemize @minus
@item
FFTW is best at handling sizes of the form
@ifinfo
@math{2^a 3^b 5^c 7^d 11^e 13^f},
@end ifinfo
@tex
$2^a 3^b 5^c 7^d 11^e 13^f$,
@end tex
@ifhtml
2<SUP>a</SUP> 3<SUP>b</SUP> 5<SUP>c</SUP> 7<SUP>d</SUP>
11<SUP>e</SUP> 13<SUP>f</SUP>,
@end ifhtml
where @math{e+f} is either @math{0} or
@math{1}, and the other exponents are arbitrary. Other sizes are
computed by means of a slow, generalpurpose routine (which nevertheless
retains
@tex
$O(n \log n)$
@end tex
@ifinfo
O(n lg n)
@end ifinfo
@ifhtml
O(n lg n)
@end ifhtml
performance, even for prime sizes). (It is
possible to customize FFTW for different array sizes.
@xref{Installation and Customization}, for more information.) Transforms
whose sizes are powers of @math{2} are especially fast.
@end itemize
@item
@code{dir} is the sign of the exponent in the formula that
defines the Fourier transform. It can be @math{1} or @math{+1}.
The aliases @code{FFTW_FORWARD} and @code{FFTW_BACKWARD}
are provided, where @code{FFTW_FORWARD} stands for @math{1}.
@item
@cindex flags
@code{flags} is a boolean OR (@samp{}) of zero or more of the following:
@itemize @minus
@item
@code{FFTW_MEASURE}: this flag tells FFTW to find the optimal plan by
actually @emph{computing} several FFTs and measuring their
execution time. Depending on the installation, this can take some
time. @footnote{The basic problem is the resolution of the clock:
FFTW needs to run for a certain time for the clock to be reliable.}
@item
@code{FFTW_ESTIMATE}: do not run any FFT and provide a ``reasonable''
plan (for a RISC processor with many registers). If neither
@code{FFTW_ESTIMATE} nor @code{FFTW_MEASURE} is provided, the default is
@code{FFTW_ESTIMATE}.
@item
@code{FFTW_OUT_OF_PLACE}: produce a plan assuming that the input and
output arrays will be distinct (this is the default).
@ctindex FFTW_OUT_OF_PLACE
@item
@cindex inplace transform
@code{FFTW_IN_PLACE}: produce a plan assuming that you want the output
in the input array. The algorithm used is not necessarily in place:
FFTW is able to compute true inplace transforms only for small values
of @code{n}. If FFTW is not able to compute the transform inplace, it
will allocate a temporary array (unless you provide one yourself),
compute the transform out of place, and copy the result back.
@emph{Warning: This option changes the meaning of some parameters of
@code{fftw}} (@pxref{fftw,,Computing the Onedimensional Transform}).
The inplace option is mainly provided for people who want to write
their own inplace multidimensional Fourier transform, using FFTW as a
base. For example, consider a threedimensional @code{n * n * n}
transform. An outofplace algorithm will need another array (which may
be huge). However, FFTW can compute the inplace transform along
each dimension using only a temporary array of size @code{n}.
Moreover, if FFTW happens to be able to compute the transform truly
inplace, no temporary array and no copying are needed. As distributed,
FFTW `knows' how to compute inplace transforms of size 1, 2, 3, 4, 5, 6,
7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 32 and 64.
The default mode of operation is @code{FFTW_OUT_OF_PLACE}.
@item
@cindex wisdom
@code{FFTW_USE_WISDOM}: use any @code{wisdom} that is available to help
in the creation of the plan. (@xref{Words of Wisdom}.)
This can greatly speed the creation of plans, especially with the
@code{FFTW_MEASURE} option. @code{FFTW_ESTIMATE} plans can also take
advantage of @code{wisdom} to produce a more optimal plan (based on past
measurements) than the estimation heuristic would normally
generate. When the @code{FFTW_MEASURE} option is used, new @code{wisdom}
will also be generated if the current transform size is not completely
understood by existing @code{wisdom}.
@end itemize
@item
@code{in}, @code{out}, @code{istride}, @code{ostride} (only for
@code{fftw_create_plan_specific}): see corresponding arguments in the
description of @code{fftw}. (@xref{fftw,,Computing the Onedimensional
Transform}.) In particular, the @code{out} and @code{ostride}
parameters have the same special meaning for @code{FFTW_IN_PLACE}
transforms as they have for @code{fftw}.
@end itemize
@node Discussion on Specific Plans, fftw, fftw_create_plan, Onedimensional Transforms Reference
@subsection Discussion on Specific Plans
@cindex specific planner
We recommend the use of the specific planners, even in cases where you
will be transforming arrays different from those passed to the specific
planners, as they confer the following advantages:
@itemize @bullet
@item
The resulting plans will be optimized for your specific arrays and
strides. This may or may not make a significant difference, but it
certainly doesn't hurt. (The ordinary planner does its planning based
upon a strideone temporary array that it allocates.)
@item
Less intermediate storage is required during the planning process. (The
ordinary planner uses O(@code{N}) temporary storage, where @code{N} is
the maximum dimension, while it is creating the plan.)
@item
For multidimensional transforms, new parameters become accessible for
optimization by the planner. (Since multidimensional arrays can be
very large, we don't dare to allocate one in the ordinary planner for
experimentation. This prevents us from doing certain optimizations
that can yield dramatic improvements in some cases.)
@end itemize
On the other hand, note that @emph{the specific planner destroys the
contents of the @code{in} and @code{out} arrays}.
@node fftw, fftw_destroy_plan, Discussion on Specific Plans, Onedimensional Transforms Reference
@subsection Computing the Onedimensional Transform
@example
#include <fftw.h>
void fftw(fftw_plan plan, int howmany,
fftw_complex *in, int istride, int idist,
fftw_complex *out, int ostride, int odist);
void fftw_one(fftw_plan plan, fftw_complex *in,
fftw_complex *out);
@end example
@findex fftw
@findex fftw_one
The function @code{fftw} computes the onedimensional Fourier transform,
using a plan created by @code{fftw_create_plan} (@xref{fftw_create_plan,
, Plan Creation for Onedimensional Transforms}.) The function
@code{fftw_one} provides a simplified interface for the common case of
single input array of stride 1.
@cindex stride
@subsubheading Arguments
@itemize @bullet
@item
@code{plan} is the plan created by @code{fftw_create_plan}
(@pxref{fftw_create_plan,,Plan Creation for Onedimensional Transforms}).
@item
@code{howmany} is the number of transforms @code{fftw} will compute.
It is faster to tell FFTW to compute many transforms, instead of
simply calling @code{fftw} many times.
@item
@code{in}, @code{istride} and @code{idist} describe the input array(s).
There are @code{howmany} input arrays; the first one is pointed to by
@code{in}, the second one is pointed to by @code{in + idist}, and so on,
up to @code{in + (howmany  1) * idist}. Each input array consists of
complex numbers (@pxref{Data Types}), which are not necessarily
contiguous in memory. Specifically, @code{in[0]} is the first element
of the first array, @code{in[istride]} is the second element of the
first array, and so on. In general, the @code{i}th element of the
@code{j}th input array will be in position @code{in[i * istride + j *
idist]}.
@item
@code{out}, @code{ostride} and @code{odist} describe the output
array(s). The format is the same as for the input array.
@itemize @minus
@item @emph{Inplace transforms}:
@cindex inplace transform
If the @code{plan} specifies an inplace transform, @code{ostride} and
@code{odist} are always ignored. If @code{out} is @code{NULL},
@code{out} is ignored, too. Otherwise, @code{out} is interpreted as a
pointer to an array of @code{n} complex numbers, that FFTW will use as
temporary space to perform the inplace computation. @code{out} is used
as scratch space and its contents destroyed. In this case, @code{out}
must be an ordinary array whose elements are contiguous in memory (no
striding).
@end itemize
@end itemize
The function @code{fftw_one} transforms a single, contiguous input array
to a contiguous output array. By definition, the call
@example
fftw_one(plan, in, out)
@end example
is equivalent to
@example
fftw(plan, 1, in, 1, 0, out, 1, 0)
@end example
@node fftw_destroy_plan, What FFTW Really Computes, fftw, Onedimensional Transforms Reference
@subsection Destroying a Onedimensional Plan
@example
#include <fftw.h>
void fftw_destroy_plan(fftw_plan plan);
@end example
@tindex fftw_destroy_plan
The function @code{fftw_destroy_plan} frees the plan @code{plan} and
releases all the memory associated with it. After destruction, a plan
is no longer valid.
@node What FFTW Really Computes, , fftw_destroy_plan, Onedimensional Transforms Reference
@subsection What FFTW Really Computes
@cindex Discrete Fourier Transform
In this section, we define precisely what FFTW computes. Please be
warned that different authors and software packages might employ
different conventions than FFTW does.
The forward transform of a complex array @math{X} of size
@math{n} computes an array @math{Y}, where
@tex
$$
Y_i = \sum_{j = 0}^{n  1} X_j e^{2\pi i j \sqrt{1}/n} \ .
$$
@end tex
@ifinfo
@center Y[i] = sum for j = 0 to (n  1) of X[j] * exp(2 pi i j sqrt(1)/n) .
@end ifinfo
@ifhtml
<center><IMG SRC="equation1.gif" ALIGN="top"></center>
@end ifhtml
The backward transform computes
@tex
$$
Y_i = \sum_{j = 0}^{n  1} X_j e^{2\pi i j \sqrt{1}/n} \ .
$$
@end tex
@ifinfo
@center Y[i] = sum for j = 0 to (n  1) of X[j] * exp(2 pi i j sqrt(1)/n) .
@end ifinfo
@ifhtml
<center><IMG SRC="equation2.gif" ALIGN="top"></center>
@end ifhtml
@cindex normalization
FFTW computes an unnormalized transform, that is, the equation
@math{IFFT(FFT(X)) = n X} holds. In other words, applying the forward
and then the backward transform will multiply the input by @math{n}.
@cindex frequency
An @code{FFTW_FORWARD} transform corresponds to a sign of @math{1} in
the exponent of the DFT. Note also that we use the standard
``inorder'' output orderingthe @math{k}th output corresponds to the
frequency @math{k/n} (or @math{k/T}, where @math{T} is your total
sampling period). For those who like to think in terms of positive and
negative frequencies, this means that the positive frequencies are
stored in the first half of the output and the negative frequencies are
stored in backwards order in the second half of the output. (The
frequency @math{k/n} is the same as the frequency @math{(nk)/n}.)
@c 
@node Multidimensional Transforms Reference, Real Onedimensional Transforms Reference, Onedimensional Transforms Reference, FFTW Reference
@section Multidimensional Transforms Reference
@cindex complex multidimensional transform
@cindex multidimensional transform
The multidimensional complex routines are generally prefixed with
@code{fftwnd_}. Programs using FFTWND should be linked with @code{lfftw
lm} on Unix systems, or with the FFTW and standard math libraries in
general.
@cindex FFTWND
@menu
* fftwnd_create_plan:: Plan Creation
* fftwnd:: Plan Execution
* fftwnd_destroy_plan:: Plan Destruction
* What FFTWND Really Computes::
@end menu
@node fftwnd_create_plan, fftwnd, Multidimensional Transforms Reference, Multidimensional Transforms Reference
@subsection Plan Creation for Multidimensional Transforms
@example
#include <fftw.h>
fftwnd_plan fftwnd_create_plan(int rank, const int *n,
fftw_direction dir, int flags);
fftwnd_plan fftw2d_create_plan(int nx, int ny,
fftw_direction dir, int flags);
fftwnd_plan fftw3d_create_plan(int nx, int ny, int nz,
fftw_direction dir, int flags);
fftwnd_plan fftwnd_create_plan_specific(int rank, const int *n,
fftw_direction dir,
int flags,
fftw_complex *in, int istride,
fftw_complex *out, int ostride);
fftwnd_plan fftw2d_create_plan_specific(int nx, int ny,
fftw_direction dir,
int flags,
fftw_complex *in, int istride,
fftw_complex *out, int ostride);
fftwnd_plan fftw3d_create_plan_specific(int nx, int ny, int nz,
fftw_direction dir, int flags,
fftw_complex *in, int istride,
fftw_complex *out, int ostride);
@end example
@tindex fftwnd_plan
@tindex fftw_direction
@findex fftwnd_create_plan
@findex fftw2d_create_plan
@findex fftw3d_create_plan
@findex fftwnd_create_plan_specific
@findex fftw2d_create_plan_specific
@findex fftw3d_create_plan_specific
The function @code{fftwnd_create_plan} creates a plan, which is a data
structure containing all the information that @code{fftwnd} needs in
order to compute a multidimensional Fourier transform. You can create
as many plans as you need, but only one plan for a given array size is
required (a plan can be reused many times). The functions
@code{fftw2d_create_plan} and @code{fftw3d_create_plan} are optional,
alternative interfaces to @code{fftwnd_create_plan} for two and three
dimensions, respectively.
@code{fftwnd_create_plan} returns a valid plan, or @code{NULL} if, for
some reason, the plan can't be created. This can happen if memory runs
out or if the arguments are invalid in some way (e.g. if @code{rank} <
0).
The @code{create_plan_specific} variants take as additional arguments
specific input/output arrays and their strides. For the last four
arguments, you should pass the arrays and strides that you will
eventually be passing to @code{fftwnd}. The resulting plans will be
optimized for those arrays and strides, although they may be used on
other arrays as well. Note: the contents of the in and out arrays are
@emph{destroyed} by the specific planner (the initial contents are
ignored, so the arrays need not have been initialized).
@xref{Discussion on Specific Plans}, for a discussion on specific plans.
@subsubheading Arguments
@itemize @bullet
@item
@code{rank} is the dimensionality of the arrays to be transformed. It
can be any nonnegative integer.
@item
@code{n} is a pointer to an array of @code{rank} integers, giving the
size of each dimension of the arrays to be transformed. These sizes,
which must be positive integers, correspond to the dimensions of
@cindex rowmajor
rowmajor arraysi.e. @code{n[0]} is the size of the dimension whose
indices vary most slowly, and so on. (@xref{Multidimensional Array
Format}, for more information on rowmajor storage.)
@xref{fftw_create_plan,,Plan Creation for Onedimensional Transforms},
for more information regarding optimal array sizes.
@item
@code{nx} and @code{ny} in @code{fftw2d_create_plan} are positive
integers specifying the dimensions of the rank 2 array to be
transformed. i.e. they specify that the transform will operate on
@code{nx x ny} arrays in rowmajor order, where @code{nx} is the number
of rows and @code{ny} is the number of columns.
@item
@code{nx}, @code{ny} and @code{nz} in @code{fftw3d_create_plan} are
positive integers specifying the dimensions of the rank 3 array to be
transformed. i.e. they specify that the transform will operate on
@code{nx x ny x nz} arrays in rowmajor order.
@item
@code{dir} is the sign of the exponent in the formula that defines the
Fourier transform. It can be @math{1} or @math{+1}. The aliases
@code{FFTW_FORWARD} and @code{FFTW_BACKWARD} are provided, where
@code{FFTW_FORWARD} stands for @math{1}.
@item
@cindex flags
@code{flags} is a boolean OR (@samp{}) of zero or more of the following:
@itemize @minus
@item
@code{FFTW_MEASURE}: this flag tells FFTW to find the optimal plan by
actually @emph{computing} several FFTs and measuring their execution
time.
@item
@code{FFTW_ESTIMATE}: do not run any FFT and provide a ``reasonable''
plan (for a RISC processor with many registers). If neither
@code{FFTW_ESTIMATE} nor @code{FFTW_MEASURE} is provided, the default is
@code{FFTW_ESTIMATE}.
@item
@code{FFTW_OUT_OF_PLACE}: produce a plan assuming that the input
and output arrays will be distinct (this is the default).
@item
@code{FFTW_IN_PLACE}: produce a plan assuming that you want to perform
the transform inplace. (Unlike the onedimensional transform, this
``really'' @footnote{@code{fftwnd} actually may use some temporary
storage (hidden in the plan), but this storage space is only the size of
the largest dimension of the array, rather than being as big as the
entire array. (Unless you use @code{fftwnd} to perform onedimensional
transforms, in which case the temporary storage required for inplace
transforms @emph{is} as big as the entire array.)} performs the
transform inplace.) Note that, if you want to perform inplace
transforms, you @emph{must} use a plan created with this option.
The default mode of operation is @code{FFTW_OUT_OF_PLACE}.
@item
@cindex wisdom
@code{FFTW_USE_WISDOM}: use any @code{wisdom} that is available to help
in the creation of the plan. (@xref{Words of Wisdom}.) This can greatly
speed the creation of plans, especially with the @code{FFTW_MEASURE}
option. @code{FFTW_ESTIMATE} plans can also take advantage of
@code{wisdom} to produce a more optimal plan (based on past
measurements) than the estimation heuristic would normally
generate. When the @code{FFTW_MEASURE} option is used, new @code{wisdom}
will also be generated if the current transform size is not completely
understood by existing @code{wisdom}. Note that the same @code{wisdom}
is shared between onedimensional and multidimensional transforms.
@end itemize
@item
@code{in}, @code{out}, @code{istride}, @code{ostride} (only for the
@code{_create_plan_specific} variants): see corresponding arguments in
the description of @code{fftwnd}. (@xref{fftwnd,,Computing the
Multidimensional Transform}.)
@end itemize
@node fftwnd, fftwnd_destroy_plan, fftwnd_create_plan, Multidimensional Transforms Reference
@subsection Computing the Multidimensional Transform
@example
#include <fftw.h>
void fftwnd(fftwnd_plan plan, int howmany,
fftw_complex *in, int istride, int idist,
fftw_complex *out, int ostride, int odist);
void fftwnd_one(fftwnd_plan p, fftw_complex *in,
fftw_complex *out);
@end example
@findex fftwnd
@findex fftwnd_one
The function @code{fftwnd} computes one or more multidimensional
Fourier Transforms, using a plan created by @code{fftwnd_create_plan}
(@pxref{fftwnd_create_plan,,Plan Creation for Multidimensional
Transforms}). (Note that the plan determines the rank and dimensions of
the array to be transformed.) The function @code{fftwnd_one} provides a
simplified interface for the common case of single input array of stride
1.
@cindex stride
@subsubheading Arguments
@itemize @bullet
@item
@code{plan} is the plan created by @code{fftwnd_create_plan}.
(@pxref{fftwnd_create_plan,,Plan Creation for Multidimensional
Transforms}). In the case of two and threedimensional transforms, it
could also have been created by @code{fftw2d_create_plan} or
@code{fftw3d_create_plan}, respectively.
@item
@code{howmany} is the number of multidimensional transforms
@code{fftwnd} will compute.
@item
@code{in}, @code{istride} and @code{idist} describe the input array(s).
There are @code{howmany} multidimensional input arrays; the first one
is pointed to by @code{in}, the second one is pointed to by @code{in +
idist}, and so on, up to @code{in + (howmany  1) * idist}. Each
multidimensional input array consists of complex numbers (@pxref{Data
Types}), stored in rowmajor format (@pxref{Multidimensional Array
Format}), which are not necessarily contiguous in memory. Specifically,
@code{in[0]} is the first element of the first array, @code{in[istride]}
is the second element of the first array, and so on. In general, the
@code{i}th element of the @code{j}th input array will be in position
@code{in[i * istride + j * idist]}. Note that, here, @code{i} refers to
an index into the rowmajor format for the multidimensional array,
rather than an index in any particular dimension.
@itemize @minus
@item @emph{Inplace transforms}:
@cindex inplace transform
For plans created with the @code{FFTW_IN_PLACE} option, the transform is
computed inplacethe output is returned in the @code{in} array, using
the same strides, etcetera, as were used in the input.
@end itemize
@item
@code{out}, @code{ostride} and @code{odist} describe the output array(s).
The format is the same as for the input array.
@itemize @minus
@item @emph{Inplace transforms}:
These parameters are ignored for plans created with the
@code{FFTW_IN_PLACE} option.
@end itemize
@end itemize
The function @code{fftwnd_one} transforms a single, contiguous input
array to a contiguous output array. By definition, the call
@example
fftwnd_one(plan, in, out)
@end example
is equivalent to
@example
fftwnd(plan, 1, in, 1, 0, out, 1, 0)
@end example
@node fftwnd_destroy_plan, What FFTWND Really Computes, fftwnd, Multidimensional Transforms Reference
@subsection Destroying a Multidimensional Plan
@example
#include <fftw.h>
void fftwnd_destroy_plan(fftwnd_plan plan);
@end example
@findex fftwnd_destroy_plan
The function @code{fftwnd_destroy_plan} frees the plan @code{plan}
and releases all the memory associated with it. After destruction,
a plan is no longer valid.
@node What FFTWND Really Computes, , fftwnd_destroy_plan, Multidimensional Transforms Reference
@subsection What FFTWND Really Computes
@cindex Discrete Fourier Transform
The conventions that we follow for the multidimensional transform are
analogous to those for the onedimensional transform. In particular, the
forward transform has a negative sign in the exponent and neither the
forward nor the backward transforms will perform any normalization.
Computing the backward transform of the forward transform will multiply
the array by the product of its dimensions. The output is inorder, and
the zeroth element of the output is the amplitude of the zero frequency
component.
@tex
The exact mathematical definition of our multidimensional transform
follows. Let $X$ be a $d$dimensional complex array whose elements are
$X[j_1, j_2, \ldots, j_d]$, where $0 \leq j_s < n_s$ for all~$s \in \{
1, 2, \ldots, d \}$. Let also $\omega_s = e^{2\pi \sqrt{1}/n_s}$, for
all ~$s \in \{ 1, 2, \ldots, d \}$.
The forward transform computes a complex array~$Y$, whose
structure is the same as that of~$X$, defined by
$$
Y[i_1, i_2, \ldots, i_d] =
\sum_{j_1 = 0}^{n_1  1}
\sum_{j_2 = 0}^{n_2  1}
\cdots
\sum_{j_d = 0}^{n_d  1}
X[j_1, j_2, \ldots, j_d]
\omega_1^{i_1 j_1}
\omega_2^{i_2 j_2}
\cdots
\omega_d^{i_d j_d} \ .
$$
The backward transform computes
$$
Y[i_1, i_2, \ldots, i_d] =
\sum_{j_1 = 0}^{n_1  1}
\sum_{j_2 = 0}^{n_2  1}
\cdots
\sum_{j_d = 0}^{n_d  1}
X[j_1, j_2, \ldots, j_d]
\omega_1^{i_1 j_1}
\omega_2^{i_2 j_2}
\cdots
\omega_d^{i_d j_d} \ .
$$
Computing the forward transform followed by the backward transform
will multiply the array by $\prod_{s=1}^{d} n_d$.
@end tex
@ifinfo
The @TeX{} version of this manual contains the exact definition of the
@math{n}dimensional transform FFTW uses. It is not possible to
display the definition on a ASCII terminal properly.
@end ifinfo
@ifhtml
The Gods forbade using HTML to display mathematical formulas. Please
see the TeX or Postscript version of this manual for the proper
definition of the ndimensional Fourier transform that FFTW
uses. For completeness, we include a bitmap of the TeX output below:
<P><center><IMG SRC="equation3.gif" ALIGN="top"></center>
@end ifhtml
@c 
@node Real Onedimensional Transforms Reference, Real Multidimensional Transforms Reference, Multidimensional Transforms Reference, FFTW Reference
@section Real Onedimensional Transforms Reference
The onedimensional real routines are generally prefixed with
@code{rfftw_}. @footnote{The etymologicallycorrect spelling would be
@code{frftw_}, but it is hard to remember.} Programs using RFFTW
should be linked with @code{lrfftw lfftw lm} on Unix systems, or with
the RFFTW, the FFTW, and the standard math libraries in general.
@cindex RFFTW
@cindex real transform
@cindex complex to real transform
@menu
* rfftw_create_plan:: Plan Creation
* rfftw:: Plan Execution
* rfftw_destroy_plan:: Plan Destruction
* What RFFTW Really Computes::
@end menu
@node rfftw_create_plan, rfftw, Real Onedimensional Transforms Reference, Real Onedimensional Transforms Reference
@subsection Plan Creation for Real Onedimensional Transforms
@example
#include <rfftw.h>
rfftw_plan rfftw_create_plan(int n, fftw_direction dir, int flags);
rfftw_plan rfftw_create_plan_specific(int n, fftw_direction dir,
int flags, fftw_real *in, int istride,
fftw_real *out, int ostride);
@end example
@tindex rfftw_plan
@findex rfftw_create_plan
@findex rfftw_create_plan_specific
The function @code{rfftw_create_plan} creates a plan, which is a data
structure containing all the information that @code{rfftw} needs in
order to compute the 1D real Fourier transform. You can create as many
plans as you need, but only one plan for a given array size is required
(a plan can be reused many times).
@code{rfftw_create_plan} returns a valid plan, or @code{NULL} if, for
some reason, the plan can't be created. In the default installation,
this cannot happen, but it is possible to configure RFFTW in such a way
that some input sizes are forbidden, and RFFTW cannot create a plan.
The @code{rfftw_create_plan_specific} variant takes as additional
arguments specific input/output arrays and their strides. For the last
four arguments, you should pass the arrays and strides that you will
eventually be passing to @code{rfftw}. The resulting plans will be
optimized for those arrays and strides, although they may be used on
other arrays as well. Note: the contents of the in and out arrays are
@emph{destroyed} by the specific planner (the initial contents are
ignored, so the arrays need not have been initialized).
@xref{Discussion on Specific Plans}, for a discussion on specific plans.
@subsubheading Arguments
@itemize @bullet
@item
@code{n} is the size of the transform. It can be
any positive integer.
@itemize @minus
@item
RFFTW is best at handling sizes of the form
@ifinfo
@math{2^a 3^b 5^c 7^d 11^e 13^f},
@end ifinfo
@tex
$2^a 3^b 5^c 7^d 11^e 13^f$,
@end tex
@ifhtml
2<SUP>a</SUP> 3<SUP>b</SUP> 5<SUP>c</SUP> 7<SUP>d</SUP>
11<SUP>e</SUP> 13<SUP>f</SUP>,
@end ifhtml
where @math{e+f} is either @math{0} or
@math{1}, and the other exponents are arbitrary. Other sizes are
computed by means of a slow, generalpurpose routine (reducing to
@ifinfo
@math{O(n^2)}
@end ifinfo
@tex
$O(n^2)$
@end tex
@ifhtml
O(n<sup>2</sup>)
@end ifhtml
performance for prime sizes). (It is possible to customize RFFTW for
different array sizes. @xref{Installation and Customization}, for more
information.) Transforms whose sizes are powers of @math{2} are
especially fast. If you have large prime factors, it may be faster to
switch over to the complex FFTW routines, which have
@iftex
@tex
$O(n \log n)$
@end tex
@end iftex
@ifinfo
O(n lg n)
@end ifinfo
@ifhtml
O(n lg n)
@end ifhtml
performance even for prime sizes (we don't know of a similar algorithm
specialized for real data, unfortunately).
@end itemize
@item
@code{dir} is the direction of the desired transform, either
@code{FFTW_REAL_TO_COMPLEX} or @code{FFTW_COMPLEX_TO_REAL},
corresponding to @code{FFTW_FORWARD} or @code{FFTW_BACKWARD},
respectively.
@ctindex FFTW_REAL_TO_COMPLEX
@ctindex FFTW_COMPLEX_TO_REAL
@item
@cindex flags
@code{flags} is a boolean OR (@samp{}) of zero or more of the following:
@itemize @minus
@item
@code{FFTW_MEASURE}: this flag tells RFFTW to find the optimal plan by
actually @emph{computing} several FFTs and measuring their
execution time. Depending on the installation, this can take some
time.
@item
@code{FFTW_ESTIMATE}: do not run any FFT and provide a ``reasonable''
plan (for a RISC processor with many registers). If neither
@code{FFTW_ESTIMATE} nor @code{FFTW_MEASURE} is provided, the default is
@code{FFTW_ESTIMATE}.
@item
@code{FFTW_OUT_OF_PLACE}: produce a plan assuming that the input
and output arrays will be distinct (this is the default).
@item
@code{FFTW_IN_PLACE}: produce a plan assuming that you want the output
in the input array. The algorithm used is not necessarily in place:
RFFTW is able to compute true inplace transforms only for small values
of @code{n}. If RFFTW is not able to compute the transform inplace, it
will allocate a temporary array (unless you provide one yourself),
compute the transform out of place, and copy the result back.
@emph{Warning: This option changes the meaning of some parameters of
@code{rfftw}} (@pxref{rfftw,,Computing the Real Onedimensional Transform}).
The default mode of operation is @code{FFTW_OUT_OF_PLACE}.
@item
@code{FFTW_USE_WISDOM}: use any @code{wisdom} that is available to help
in the creation of the plan. (@xref{Words of Wisdom}.)
This can greatly speed the creation of plans, especially with the
@code{FFTW_MEASURE} option. @code{FFTW_ESTIMATE} plans can also take
advantage of @code{wisdom} to produce a more optimal plan (based on past
measurements) than the estimation heuristic would normally
generate. When the @code{FFTW_MEASURE} option is used, new @code{wisdom}
will also be generated if the current transform size is not completely
understood by existing @code{wisdom}.
@end itemize
@item
@code{in}, @code{out}, @code{istride}, @code{ostride} (only for
@code{rfftw_create_plan_specific}): see corresponding arguments in the
description of @code{rfftw}. (@xref{rfftw,,Computing the Real
Onedimensional Transform}.) In particular, the @code{out} and
@code{ostride} parameters have the same special meaning for
@code{FFTW_IN_PLACE} transforms as they have for @code{rfftw}.
@end itemize
@node rfftw, rfftw_destroy_plan, rfftw_create_plan, Real Onedimensional Transforms Reference
@subsection Computing the Real Onedimensional Transform
@example
#include <rfftw.h>
void rfftw(rfftw_plan plan, int howmany,
fftw_real *in, int istride, int idist,
fftw_real *out, int ostride, int odist);
void rfftw_one(rfftw_plan plan, fftw_real *in, fftw_real *out);
@end example
@findex rfftw
@findex rfftw_one
The function @code{rfftw} computes the Real Onedimensional Fourier
Transform, using a plan created by @code{rfftw_create_plan}
(@pxref{rfftw_create_plan,,Plan Creation for Real Onedimensional
Transforms}). The function @code{rfftw_one} provides a simplified
interface for the common case of single input array of stride 1.
@cindex stride
@emph{Important:} When invoked for an outofplace,
@code{FFTW_COMPLEX_TO_REAL} transform, the input array is overwritten
with scratch values by these routines. The input array is not modified
for @code{FFTW_REAL_TO_COMPLEX} transforms.
@subsubheading Arguments
@itemize @bullet
@item
@code{plan} is the plan created by @code{rfftw_create_plan}
(@pxref{rfftw_create_plan,,Plan Creation for Real Onedimensional
Transforms}).
@item
@code{howmany} is the number of transforms @code{rfftw} will compute.
It is faster to tell RFFTW to compute many transforms, instead of
simply calling @code{rfftw} many times.
@item
@code{in}, @code{istride} and @code{idist} describe the input array(s).
There are two cases. If the @code{plan} defines a
@code{FFTW_REAL_TO_COMPLEX} transform, @code{in} is a real array.
Otherwise, for @code{FFTW_COMPLEX_TO_REAL} transforms, @code{in} is a
halfcomplex array @emph{whose contents will be destroyed}.
@item
@code{out}, @code{ostride} and @code{odist} describe the output
array(s), and have the same meaning as the corresponding parameters for
the input array.
@itemize @minus
@item @emph{Inplace transforms}:
If the @code{plan} specifies an inplace transform, @code{ostride} and
@code{odist} are always ignored. If @code{out} is @code{NULL},
@code{out} is ignored, too. Otherwise, @code{out} is interpreted as a
pointer to an array of @code{n} complex numbers, that FFTW will use as
temporary space to perform the inplace computation. @code{out} is used
as scratch space and its contents destroyed. In this case, @code{out}
must be an ordinary array whose elements are contiguous in memory (no
striding).
@end itemize
@end itemize
The function @code{rfftw_one} transforms a single, contiguous input array
to a contiguous output array. By definition, the call
@example
rfftw_one(plan, in, out)
@end example
is equivalent to
@example
rfftw(plan, 1, in, 1, 0, out, 1, 0)
@end example
@node rfftw_destroy_plan, What RFFTW Really Computes, rfftw, Real Onedimensional Transforms Reference
@subsection Destroying a Real Onedimensional Plan
@example
#include <rfftw.h>
void rfftw_destroy_plan(rfftw_plan plan);
@end example
@findex rfftw_destroy_plan
The function @code{rfftw_destroy_plan} frees the plan @code{plan} and
releases all the memory associated with it. After destruction, a plan
is no longer valid.
@node What RFFTW Really Computes, , rfftw_destroy_plan, Real Onedimensional Transforms Reference
@subsection What RFFTW Really Computes
@cindex Discrete Fourier Transform
In this section, we define precisely what RFFTW computes.
The real to complex (@code{FFTW_REAL_TO_COMPLEX}) transform of a real
array @math{X} of size @math{n} computes an hermitian array @math{Y},
where
@tex
$$
Y_i = \sum_{j = 0}^{n  1} X_j e^{2\pi i j \sqrt{1}/n}
$$
@end tex
@ifinfo
@center Y[i] = sum for j = 0 to (n  1) of X[j] * exp(2 pi i j sqrt(1)/n)
@end ifinfo
@ifhtml
<center><IMG SRC="equation1.gif" ALIGN="top"></center>
@end ifhtml
(That @math{Y} is a hermitian array is not intended to be obvious,
although the proof is easy.) The hermitian array @math{Y} is stored in
halfcomplex order (@pxref{Data Types}). Currently, RFFTW provides no
way to compute a real to complex transform with a positive sign in the
exponent.
The complex to real (@code{FFTW_COMPLEX_TO_REAL}) transform of a hermitian
array @math{X} of size @math{n} computes a real array @math{Y}, where
@tex
$$
Y_i = \sum_{j = 0}^{n  1} X_j e^{2\pi i j \sqrt{1}/n}
$$
@end tex
@ifinfo
@center Y[i] = sum for j = 0 to (n  1) of X[j] * exp(2 pi i j sqrt(1)/n)
@end ifinfo
@ifhtml
<center><IMG SRC="equation2.gif" ALIGN="top"></center>
@end ifhtml
(That @math{Y} is a real array is not intended to be obvious, although
the proof is easy.) The hermitian input array @math{X} is stored in
halfcomplex order (@pxref{Data Types}). Currently, RFFTW provides no
way to compute a complex to real transform with a negative sign in the
exponent.
@cindex normalization
Like FFTW, RFFTW computes an unnormalized transform. In other words,
applying the real to complex (forward) and then the complex to real
(backward) transform will multiply the input by @math{n}.
@c 
@node Real Multidimensional Transforms Reference, Wisdom Reference, Real Onedimensional Transforms Reference, FFTW Reference
@section Real Multidimensional Transforms Reference
@cindex real multidimensional transform
@cindex multidimensional transform
The multidimensional real routines are generally prefixed with
@code{rfftwnd_}. Programs using RFFTWND should be linked with
@code{lrfftw lfftw lm} on Unix systems, or with the FFTW, RFFTW, and
standard math libraries in general.
@cindex RFFTWND
@menu
* rfftwnd_create_plan:: Plan Creation
* rfftwnd:: Plan Execution
* Array Dimensions for Real Multidimensional Transforms::
* Strides in Inplace RFFTWND::
* rfftwnd_destroy_plan:: Plan Destruction
* What RFFTWND Really Computes::
@end menu
@node rfftwnd_create_plan, rfftwnd, Real Multidimensional Transforms Reference, Real Multidimensional Transforms Reference
@subsection Plan Creation for Real Multidimensional Transforms
@example
#include <rfftw.h>
rfftwnd_plan rfftwnd_create_plan(int rank, const int *n,
fftw_direction dir, int flags);
rfftwnd_plan rfftw2d_create_plan(int nx, int ny,
fftw_direction dir, int flags);
rfftwnd_plan rfftw3d_create_plan(int nx, int ny, int nz,
fftw_direction dir, int flags);
@end example
@tindex rfftwnd_plan
@tindex fftw_direction
@findex rfftwnd_create_plan
@findex rfftw2d_create_plan
@findex rfftw3d_create_plan
The function @code{rfftwnd_create_plan} creates a plan, which is a data
structure containing all the information that @code{rfftwnd} needs in
order to compute a multidimensional real Fourier transform. You can
create as many plans as you need, but only one plan for a given array
size is required (a plan can be reused many times). The functions
@code{rfftw2d_create_plan} and @code{rfftw3d_create_plan} are optional,
alternative interfaces to @code{rfftwnd_create_plan} for two and three
dimensions, respectively.
@code{rfftwnd_create_plan} returns a valid plan, or @code{NULL} if, for
some reason, the plan can't be created. This can happen if the
arguments are invalid in some way (e.g. if @code{rank} < 0).
@subsubheading Arguments
@itemize @bullet
@item
@code{rank} is the dimensionality of the arrays to be transformed. It
can be any nonnegative integer.
@item
@code{n} is a pointer to an array of @code{rank} integers, giving the
size of each dimension of the arrays to be transformed. Note that these
are always the dimensions of the @emph{real} arrays; the complex arrays
have different dimensions (@pxref{Array Dimensions for Real
Multidimensional Transforms}). These sizes, which must be positive
integers, correspond to the dimensions of rowmajor
arraysi.e. @code{n[0]} is the size of the dimension whose indices
vary most slowly, and so on. (@xref{Multidimensional Array Format}, for
more information.)
@itemize @minus
@item
@xref{rfftw_create_plan,,Plan Creation for Real Onedimensional Transforms},
for more information regarding optimal array sizes.
@end itemize
@item
@code{nx} and @code{ny} in @code{rfftw2d_create_plan} are positive
integers specifying the dimensions of the rank 2 array to be
transformed. i.e. they specify that the transform will operate on
@code{nx x ny} arrays in rowmajor order, where @code{nx} is the number
of rows and @code{ny} is the number of columns.
@item
@code{nx}, @code{ny} and @code{nz} in @code{rfftw3d_create_plan} are
positive integers specifying the dimensions of the rank 3 array to be
transformed. i.e. they specify that the transform will operate on
@code{nx x ny x nz} arrays in rowmajor order.
@item
@code{dir} is the direction of the desired transform, either
@code{FFTW_REAL_TO_COMPLEX} or @code{FFTW_COMPLEX_TO_REAL},
corresponding to @code{FFTW_FORWARD} or @code{FFTW_BACKWARD},
respectively.
@item
@cindex flags
@code{flags} is a boolean OR (@samp{}) of zero or more of the following:
@itemize @minus
@item
@code{FFTW_MEASURE}: this flag tells FFTW to find the optimal plan by
actually @emph{computing} several FFTs and measuring their execution
time.
@item
@code{FFTW_ESTIMATE}: do not run any FFT and provide a ``reasonable''
plan (for a RISC processor with many registers). If neither
@code{FFTW_ESTIMATE} nor @code{FFTW_MEASURE} is provided, the default is
@code{FFTW_ESTIMATE}.
@item
@code{FFTW_OUT_OF_PLACE}: produce a plan assuming that the input
and output arrays will be distinct (this is the default).
@item
@cindex inplace transform
@code{FFTW_IN_PLACE}: produce a plan assuming that you want to perform
the transform inplace. (Unlike the onedimensional transform, this
``really'' performs the transform inplace.) Note that, if you want to
perform inplace transforms, you @emph{must} use a plan created with
this option. The use of this option has important implications for the
size of the input/output array (@pxref{rfftwnd,,Computing the Real
Multidimensional Transform}).
The default mode of operation is @code{FFTW_OUT_OF_PLACE}.
@item
@cindex wisdom
@code{FFTW_USE_WISDOM}: use any @code{wisdom} that is available to help
in the creation of the plan. (@xref{Words of Wisdom}.) This can greatly
speed the creation of plans, especially with the @code{FFTW_MEASURE}
option. @code{FFTW_ESTIMATE} plans can also take advantage of
@code{wisdom} to produce a more optimal plan (based on past
measurements) than the estimation heuristic would normally
generate. When the @code{FFTW_MEASURE} option is used, new @code{wisdom}
will also be generated if the current transform size is not completely
understood by existing @code{wisdom}. Note that the same @code{wisdom}
is shared between onedimensional and multidimensional transforms.
@end itemize
@end itemize
@node rfftwnd, Array Dimensions for Real Multidimensional Transforms, rfftwnd_create_plan, Real Multidimensional Transforms Reference
@subsection Computing the Real Multidimensional Transform
@example
#include <rfftw.h>
void rfftwnd_real_to_complex(rfftwnd_plan plan, int howmany,
fftw_real *in, int istride, int idist,
fftw_complex *out, int ostride, int odist);
void rfftwnd_complex_to_real(rfftwnd_plan plan, int howmany,
fftw_complex *in, int istride, int idist,
fftw_real *out, int ostride, int odist);
void rfftwnd_one_real_to_complex(rfftwnd_plan p, fftw_real *in,
fftw_complex *out);
void rfftwnd_one_complex_to_real(rfftwnd_plan p, fftw_complex *in,
fftw_real *out);
@end example
@findex rfftwnd_real_to_complex
@findex rfftwnd_complex_to_real
@findex rfftwnd_one_real_to_complex
@findex rfftwnd_one_complex_to_real
These functions compute the real multidimensional Fourier Transform,
using a plan created by @code{rfftwnd_create_plan}
(@pxref{rfftwnd_create_plan,,Plan Creation for Real Multidimensional
Transforms}). (Note that the plan determines the rank and dimensions of
the array to be transformed.) The @samp{@code{rfftwnd_one_}} functions
provide a simplified interface for the common case of single input array
of stride 1. Unlike other transform routines in FFTW, we here use
separate functions for the two directions of the transform in order to
correctly express the datatypes of the parameters.
@emph{Important:} When invoked for an outofplace,
@code{FFTW_COMPLEX_TO_REAL} transform with @code{rank > 1}, the input
array is overwritten with scratch values by these routines. The input
array is not modified for @code{FFTW_REAL_TO_COMPLEX} transforms or for
@code{FFTW_COMPLEX_TO_REAL} with @code{rank == 1}.
@subsubheading Arguments
@itemize @bullet
@item
@code{plan} is the plan created by @code{rfftwnd_create_plan}.
(@pxref{rfftwnd_create_plan,,Plan Creation for Real Multidimensional
Transforms}). In the case of two and threedimensional transforms, it
could also have been created by @code{rfftw2d_create_plan} or
@code{rfftw3d_create_plan}, respectively.
@code{FFTW_REAL_TO_COMPLEX} plans must be used with the
@samp{@code{real_to_complex}} functions, and @code{FFTW_COMPLEX_TO_REAL}
plans must be used with the @samp{@code{complex_to_real}} functions. It
is an error to mismatch the plan direction and the transform function.
@item
@code{howmany} is the number of transforms to be computed.
@item
@cindex stride
@code{in}, @code{istride} and @code{idist} describe the input array(s).
There are @code{howmany} input arrays; the first one is pointed to by
@code{in}, the second one is pointed to by @code{in + idist}, and so on,
up to @code{in + (howmany  1) * idist}. Each input array is stored in
rowmajor format (@pxref{Multidimensional Array Format}), and is not
necessarily contiguous in memory. Specifically, @code{in[0]} is the
first element of the first array, @code{in[istride]} is the second
element of the first array, and so on. In general, the @code{i}th
element of the @code{j}th input array will be in position @code{in[i *
istride + j * idist]}. Note that, here, @code{i} refers to an index into
the rowmajor format for the multidimensional array, rather than an
index in any particular dimension.
The dimensions of the arrays are different for real and complex data,
and are discussed in more detail below (@pxref{Array Dimensions for Real
Multidimensional Transforms}).
@itemize @minus
@item @emph{Inplace transforms}:
For plans created with the @code{FFTW_IN_PLACE} option, the transform is
computed inplacethe output is returned in the @code{in} array. The
meaning of the @code{stride} and @code{dist} parameters in this case is
subtle and is discussed below (@pxref{Strides in Inplace RFFTWND}).
@end itemize
@item
@code{out}, @code{ostride} and @code{odist} describe the output
array(s). The format is the same as that for the input array. See
below for a discussion of the dimensions of the output array for real
and complex data.
@itemize @minus
@item @emph{Inplace transforms}:
These parameters are ignored for plans created with the
@code{FFTW_IN_PLACE} option.
@end itemize
@end itemize
The function @code{rfftwnd_one} transforms a single, contiguous input
array to a contiguous output array. By definition, the call
@example
rfftwnd_one_...(plan, in, out)
@end example
is equivalent to
@example
rfftwnd_...(plan, 1, in, 1, 0, out, 1, 0)
@end example
@node Array Dimensions for Real Multidimensional Transforms, Strides in Inplace RFFTWND, rfftwnd, Real Multidimensional Transforms Reference
@subsection Array Dimensions for Real Multidimensional Transforms
@cindex rfftwnd array format
The output of a multidimensional transform of real data contains
symmetries that, in principle, make half of the outputs redundant
(@pxref{What RFFTWND Really Computes}). In practice, it is not
possible to entirely realize these savings in an efficient and
understandable format. Instead, the output of the rfftwnd transforms is
@emph{slightly} over half of the output of the corresponding complex
transform. We do not ``pack'' the data in any way, but store it as an
ordinary array of @code{fftw_complex} values. In fact, this data is
simply a subsection of what would be the array in the corresponding
complex transform.
Specifically, for a real transform of dimensions
@tex
$n_1 \times n_2 \times \cdots \times n_d$,
@end tex
@ifinfo
n1 x n2 x ... x nd,
@end ifinfo
@ifhtml
n<sub>1</sub> x n<sub>2</sub> x ... x n<sub>d</sub>,
@end ifhtml
the complex data is an
@tex
$n_1 \times n_2 \times \cdots \times (n_d/2+1)$
@end tex
@ifinfo
n1 x n2 x ... x (nd/2+1)
@end ifinfo
@ifhtml
n<sub>1</sub> x n<sub>2</sub> x ... x (n<sub>d</sub>/2+1)
@end ifhtml
array of @code{fftw_complex} values in rowmajor order (with the
division rounded down). That is, we only store the lower half (plus one
element) of the last dimension of the data from the ordinary complex
transform. (We could have instead taken half of any other dimension,
but implementation turns out to be simpler if the last, contiguous,
dimension is used.)
@cindex inplace transform
@cindex padding
Since the complex data is slightly larger than the real data, some
complications arise for inplace transforms. In this case, the final
dimension of the real data must be padded with extra values to
accommodate the size of the complex datatwo extra if the last
dimension is even and one if it is odd. That is, the last dimension of
the real data must physically contain
@tex
$2 (n_d/2+1)$
@end tex
@ifinfo
2 * (nd/2+1)
@end ifinfo
@ifhtml
2 * (n<sub>d</sub>/2+1)
@end ifhtml
@code{fftw_real} values (exactly enough to hold the complex data).
This physical array size does not, however, change the @emph{logical}
array sizeonly
@tex
$n_d$
@end tex
@ifinfo
nd
@end ifinfo
@ifhtml
n<sub>d</sub>
@end ifhtml
values are actually stored in the last dimension, and
@tex
$n_d$
@end tex
@ifinfo
nd
@end ifinfo
@ifhtml
n<sub>d</sub>
@end ifhtml
is the last dimension passed to @code{rfftwnd_create_plan}.
@node Strides in Inplace RFFTWND, rfftwnd_destroy_plan, Array Dimensions for Real Multidimensional Transforms, Real Multidimensional Transforms Reference
@subsection Strides in Inplace RFFTWND
@cindex rfftwnd array format
@cindex stride
The fact that the input and output datatypes are different for rfftwnd
complicates the meaning of the @code{stride} and @code{dist} parameters
of inplace transformsare they in units of @code{fftw_real} or
@code{fftw_complex} elements? When reading the input, they are
interpreted in units of the datatype of the input data. When writing
the output, the @code{istride} and @code{idist} are translated to the
output datatype's ``units'' in one of two ways, corresponding to the two
most common situations in which @code{stride} and @code{dist} parameters
are useful. Below, we refer to these ``translated'' parameters as
@code{ostride_t} and @code{odist_t}. (Note that these are computed
internally by rfftwnd; the actual @code{ostride} and @code{odist}
parameters are ignored for inplace transforms.)
First, there is the case where you are transforming a number of
contiguous arrays located one after another in memory. In this
situation, @code{istride} is @code{1} and @code{idist} is the product of
the physical dimensions of the array. @code{ostride_t} and
@code{odist_t} are then chosen so that the output arrays are contiguous
and lie on top of the input arrays. @code{ostride_t} is therefore
@code{1}. For a realtocomplex transform, @code{odist_t} is
@code{idist/2}; for a complextoreal transform, @code{odist_t} is
@code{idist*2}.
The second case is when you have an array in which each element has
@code{nc} components (e.g. a structure with @code{nc} numeric fields),
and you want to transform all of the components at once. Here,
@code{istride} is @code{nc} and @code{idist} is @code{1}. For this
case, it is natural to want the output to also have @code{nc}
consecutive components, now of the output data type; this is exactly
what rfftwnd does. Specifically, it uses an @code{ostride_t} equal to
@code{istride}, and an @code{odist_t} of @code{1}. (Astute readers will
realize that some extra buffer space is required in order to perform
such a transform; this is handled automatically by rfftwnd.)
The general rule is as follows. @code{ostride_t} equals @code{istride}.
If @code{idist} is @code{1} and @code{idist} is less than
@code{istride}, then @code{odist_t} is @code{1}. Otherwise, for a
realtocomplex transform @code{odist_t} is @code{idist/2} and for a
complextoreal transform @code{odist_t} is @code{idist*2}.
@node rfftwnd_destroy_plan, What RFFTWND Really Computes, Strides in Inplace RFFTWND, Real Multidimensional Transforms Reference
@subsection Destroying a Multidimensional Plan
@example
#include <rfftw.h>
void rfftwnd_destroy_plan(rfftwnd_plan plan);
@end example
@findex rfftwnd_destroy_plan
The function @code{rfftwnd_destroy_plan} frees the plan @code{plan}
and releases all the memory associated with it. After destruction,
a plan is no longer valid.
@node What RFFTWND Really Computes, , rfftwnd_destroy_plan, Real Multidimensional Transforms Reference
@subsection What RFFTWND Really Computes
@cindex Discrete Fourier Transform
The conventions that we follow for the real multidimensional transform
are analogous to those for the complex multidimensional transform. In
particular, the forward transform has a negative sign in the exponent
and neither the forward nor the backward transforms will perform any
normalization. Computing the backward transform of the forward
transform will multiply the array by the product of its dimensions (that
is, the logical dimensions of the real data). The forward transform is
realtocomplex and the backward transform is complextoreal.
@cindex Discrete Fourier Transform
@cindex hermitian array
@tex
The exact mathematical definition of our real multidimensional
transform follows.
@noindent@emph{Real to complex (forward) transform.}
Let $X$ be a $d$dimensional real array whose elements are $X[j_1, j_2,
\ldots, j_d]$, where $0 \leq j_s < n_s$ for all~$s \in \{ 1, 2, \ldots,
d \}$. Let also $\omega_s = e^{2\pi \sqrt{1}/n_s}$, for all ~$s \in \{
1, 2, \ldots, d \}$.
The real to complex transform computes a complex array~$Y$, whose
structure is the same as that of~$X$, defined by
$$
Y[i_1, i_2, \ldots, i_d] =
\sum_{j_1 = 0}^{n_1  1}
\sum_{j_2 = 0}^{n_2  1}
\cdots
\sum_{j_d = 0}^{n_d  1}
X[j_1, j_2, \ldots, j_d]
\omega_1^{i_1 j_1}
\omega_2^{i_2 j_2}
\cdots
\omega_d^{i_d j_d} \ .
$$
The output array $Y$ enjoys a multidimensional hermitian symmetry, that
is, the identity $Y[i_1, i_2, \ldots, i_d] = Y[n_1i_1, n_2i_2, \ldots,
n_d  i_d]^{*}$ holds for all $0 \leq i_s < n_s$. Because of this
symmetry, $Y$ is stored in the peculiar way described in @ref{Array
Dimensions for Real Multidimensional Transforms}.
@cindex hermitian array
@noindent@emph{Complex to real (backward) transform.} Let $X$ be a
$d$dimensional complex array whose elements are $X[j_1, j_2, \ldots,
j_d]$, where $0 \leq j_s < n_s$ for all~$s \in \{ 1, 2, \ldots, d \}$.
The array $X$ must be hermitian, that is, the identity $X[j_1, j_2,
\ldots, j_d] = X[n_1j_1, n_2j_2, \ldots, n_d  j_d]^{*}$ must hold for all
$0 \leq j_s < n_s$. Moreover, $X$ must be stored in memory in the
peculiar way described in @ref{Array Dimensions for Real
Multidimensional Transforms}.
Let $\omega_s = e^{2\pi \sqrt{1}/n_s}$, for all ~$s \in \{ 1, 2,
\ldots, d \}$. The complex to real transform computes a real array~$Y$, whose
structure is the same as that of~$X$, defined by
$$
Y[i_1, i_2, \ldots, i_d] =
\sum_{j_1 = 0}^{n_1  1}
\sum_{j_2 = 0}^{n_2  1}
\cdots
\sum_{j_d = 0}^{n_d  1}
X[j_1, j_2, \ldots, j_d]
\omega_1^{i_1 j_1}
\omega_2^{i_2 j_2}
\cdots
\omega_d^{i_d j_d} \ .
$$
(That $Y$ is real is not meant to be obvious, although the proof is
easy.)
Computing the forward transform followed by the backward transform
will multiply the array by $\prod_{s=1}^{d} n_d$.
@end tex
@ifinfo
The @TeX{} version of this manual contains the exact definition of the
@math{n}dimensional transform RFFTWND uses. It is not possible to
display the definition on a ASCII terminal properly.
@end ifinfo
@ifhtml
The Gods forbade using HTML to display mathematical formulas. Please
see the TeX or Postscript version of this manual for the proper
definition of the ndimensional real Fourier transform that RFFTW
uses. For completeness, we include a bitmap of the TeX output below:
<P><center><IMG SRC="equation4.gif" ALIGN="top"></center>
@end ifhtml
@c 
@node Wisdom Reference, Memory Allocator Reference, Real Multidimensional Transforms Reference, FFTW Reference
@section Wisdom Reference
@menu
* fftw_export_wisdom::
* fftw_import_wisdom::
* fftw_forget_wisdom::
@end menu
@cindex wisdom
@node fftw_export_wisdom, fftw_import_wisdom, Wisdom Reference, Wisdom Reference
@subsection Exporting Wisdom
@example
#include <fftw.h>
void fftw_export_wisdom(void (*emitter)(char c, void *), void *data);
void fftw_export_wisdom_to_file(FILE *output_file);
char *fftw_export_wisdom_to_string(void);
@end example
@findex fftw_export_wisdom
@findex fftw_export_wisdom_to_file
@findex fftw_export_wisdom_to_string
These functions allow you to export all currently accumulated
@code{wisdom} in a form from which it can be later imported and
restored, even during a separate run of the program. (@xref{Words of Wisdom}.) The current store of @code{wisdom} is not
affected by calling any of these routines.
@code{fftw_export_wisdom} exports the @code{wisdom} to any output
medium, as specified by the callback function
@code{emitter}. @code{emitter} is a @code{putc}like function that
writes the character @code{c} to some output; its second parameter is
the @code{data} pointer passed to @code{fftw_export_wisdom}. For
convenience, the following two ``wrapper'' routines are provided:
@code{fftw_export_wisdom_to_file} writes the @code{wisdom} to the
current position in @code{output_file}, which should be open with write
permission. Upon exit, the file remains open and is positioned at the
end of the @code{wisdom} data.
@code{fftw_export_wisdom_to_string} returns a pointer to a
@code{NULL}terminated string holding the @code{wisdom} data. This
string is dynamically allocated, and it is the responsibility of the
caller to deallocate it with @code{fftw_free} when it is no longer
needed.
All of these routines export the wisdom in the same format, which we
will not document here except to say that it is LISPlike ASCII text
that is insensitive to white space.
@node fftw_import_wisdom, fftw_forget_wisdom, fftw_export_wisdom, Wisdom Reference
@subsection Importing Wisdom
@example
#include <fftw.h>
fftw_status fftw_import_wisdom(int (*get_input)(void *), void *data);
fftw_status fftw_import_wisdom_from_file(FILE *input_file);
fftw_status fftw_import_wisdom_from_string(const char *input_string);
@end example
@findex fftw_import_wisdom
@findex fftw_import_wisdom_from_file
@findex fftw_import_wisdom_from_string
These functions import @code{wisdom} into a program from data stored by
the @code{fftw_export_wisdom} functions above. (@xref{Words of Wisdom}.)
The imported @code{wisdom} supplements rather than replaces any
@code{wisdom} already accumulated by the running program (except when
there is conflicting @code{wisdom}, in which case the existing wisdom is
replaced).
@code{fftw_import_wisdom} imports @code{wisdom} from any input medium,
as specified by the callback function @code{get_input}. @code{get_input}
is a @code{getc}like function that returns the next character in the
input; its parameter is the @code{data} pointer passed to
@code{fftw_import_wisdom}. If the end of the input data is reached
(which should never happen for valid data), it may return either
@code{NULL} (ASCII 0) or @code{EOF} (as defined in @code{<stdio.h>}).
For convenience, the following two ``wrapper'' routines are provided:
@code{fftw_import_wisdom_from_file} reads @code{wisdom} from the
current position in @code{input_file}, which should be open with read
permission. Upon exit, the file remains open and is positioned at the
end of the @code{wisdom} data.
@code{fftw_import_wisdom_from_string} reads @code{wisdom} from the
@code{NULL}terminated string @code{input_string}.
The return value of these routines is @code{FFTW_SUCCESS} if the wisdom
was read successfully, and @code{FFTW_FAILURE} otherwise. Note that, in
all of these functions, any data in the input stream past the end of the
@code{wisdom} data is simply ignored (it is not even read if the
@code{wisdom} data is wellformed).
@node fftw_forget_wisdom, , fftw_import_wisdom, Wisdom Reference
@subsection Forgetting Wisdom
@example
#include <fftw.h>
void fftw_forget_wisdom(void);
@end example
@findex fftw_forget_wisdom
Calling @code{fftw_forget_wisdom} causes all accumulated @code{wisdom}
to be discarded and its associated memory to be freed. (New
@code{wisdom} can still be gathered subsequently, however.)
@c 
@node Memory Allocator Reference, Thread safety, Wisdom Reference, FFTW Reference
@section Memory Allocator Reference
@example
#include <fftw.h>
void *(*fftw_malloc_hook) (size_t n);
void (*fftw_free_hook) (void *p);
@end example
@vindex fftw_malloc_hook
@findex fftw_malloc
@ffindex malloc
@vindex fftw_free_hook
Whenever it has to allocate and release memory, FFTW ordinarily calls
@code{malloc} and @code{free}.
If @code{malloc} fails, FFTW prints an error message and exits. This
behavior may be undesirable in some applications. Also, special
memoryhandling functions may be necessary in certain
environments. Consequently, FFTW provides means by which you can install
your own memory allocator and take whatever errorcorrecting action you
find appropriate. The variables @code{fftw_malloc_hook} and
@code{fftw_free_hook} are pointers to functions, and they are normally
@code{NULL}. If you set those variables to point to other functions,
then FFTW will use your routines instead of @code{malloc} and
@code{free}. @code{fftw_malloc_hook} must point to a @code{malloc}like
function, and @code{fftw_free_hook} must point to a @code{free}like
function.
@c 
@node Thread safety, , Memory Allocator Reference, FFTW Reference
@section Thread safety
@cindex threads
@cindex thread safety
Users writing multithreaded programs must concern themselves with the
@dfn{thread safety} of the libraries they usethat is, whether it is
safe to call routines in parallel from multiple threads. FFTW can be
used in such an environment, but some care must be taken because certain
parts of FFTW use private global variables to share data between calls.
In particular, the plancreation functions share trigonometric tables
and accumulated @code{wisdom}. (Users should note that these comments
only apply to programs using sharedmemory threads. Parallelism using
MPI or forked processes involves a separate addressspace and global
variables for each process, and is not susceptible to problems of this
sort.)
The central restriction of FFTW is that it is not safe to create
multiple plans in parallel. You must either create all of your plans
from a single thread, or instead use a semaphore, mutex, or other
mechanism to ensure that different threads don't attempt to create plans
at the same time. The same restriction also holds for destruction of
plans and importing/forgetting @code{wisdom}. Once created, a plan may
safely be used in any thread.
The actual transform routines in FFTW (@code{fftw_one}, etcetera) are
reentrant and threadsafe, so it is fine to call them simultaneously
from multiple threads. Another question arises, howeveris it safe to
use the @emph{same plan} for multiple transforms in parallel? (It would
be unsafe if, for example, the plan were modified in some way by the
transform.) We address this question by defining an additional planner
flag, @code{FFTW_THREADSAFE}.
@ctindex FFTW_THREADSAFE
When included in the flags for any of the plancreation routines,
@code{FFTW_THREADSAFE} guarantees that the resulting plan will be
readonly and safe to use in parallel by multiple threads.
@c ************************************************************
@node Parallel FFTW, Calling FFTW from Fortran, FFTW Reference, Top
@chapter Parallel FFTW
@cindex parallel transform
In this chapter we discuss the use of FFTW in a parallel environment,
documenting the different parallel libraries that we have provided.
(Users calling FFTW from a multithreaded program should also consult
@ref{Thread safety}.) The FFTW package currently contains three parallel
transform implementations that leverage the uniprocessor FFTW code:
@itemize @bullet
@item
@cindex threads
The first set of routines utilizes sharedmemory threads for parallel
one and multidimensional transforms of both real and complex data.
Any program using FFTW can be trivially modified to use the
multithreaded routines. This code can use any common threads
implementation, including POSIX threads. (POSIX threads are available
on most Unix variants, including Linux.) These routines are located in
the @code{threads} directory, and are documented in @ref{Multithreaded
FFTW}.
@item
@cindex MPI
@cindex distributed memory
The @code{mpi} directory contains multidimensional transforms
of real and complex data for parallel machines supporting MPI. It also
includes parallel onedimensional transforms for complex data. The main
feature of this code is that it supports distributedmemory transforms,
so it runs on everything from workstation clusters to massivelyparallel
supercomputers. More information on MPI can be found at the
@uref{http://www.mcs.anl.gov/mpi, MPI home page}. The FFTW MPI routines
are documented in @ref{MPI FFTW}.
@item
@cindex Cilk
We also have an experimental parallel implementation written in Cilk, a
Clike parallel language developed at MIT and currently available for
several SMP platforms. For more information on Cilk see
@uref{http://supertech.lcs.mit.edu/cilk, the Cilk home page}. The FFTW
Cilk code can be found in the @code{cilk} directory, with parallelized
one and multidimensional transforms of complex data. The Cilk FFTW
routines are documented in @code{cilk/README}.
@end itemize
@menu
* Multithreaded FFTW::
* MPI FFTW::
@end menu
@c 
@node Multithreaded FFTW, MPI FFTW, Parallel FFTW, Parallel FFTW
@section Multithreaded FFTW
@cindex threads
In this section we document the parallel FFTW routines for sharedmemory
threads on SMP hardware. These routines, which support parallel one
and multidimensional transforms of both real and complex data, are the
easiest way to take advantage of multiple processors with FFTW. They
work just like the corresponding uniprocessor transform routines, except
that they take the number of parallel threads to use as an extra
parameter. Any program that uses the uniprocessor FFTW can be trivially
modified to use the multithreaded FFTW.
@menu
* Installation and Supported Hardware/Software::
* Usage of Multithreaded FFTW::
* How Many Threads to Use?::
* Using Multithreaded FFTW in a Multithreaded Program::
* Tips for Optimal Threading::
@end menu
@c 
@node Installation and Supported Hardware/Software, Usage of Multithreaded FFTW, Multithreaded FFTW, Multithreaded FFTW
@subsection Installation and Supported Hardware/Software
All of the FFTW threads code is located in the @code{threads}
subdirectory of the FFTW package. On Unix systems, the FFTW threads
libraries and header files can be automatically configured, compiled,
and installed along with the uniprocessor FFTW libraries simply by
including @code{enablethreads} in the flags to the @code{configure}
script (@pxref{Installation on Unix}). (Note also that the threads
routines, when enabled, are automatically tested by the @samp{@code{make
check}} selftests.)
@fpindex configure
The threads routines require your operating system to have some sort of
sharedmemory threads support. Specifically, the FFTW threads package
works with POSIX threads (available on most Unix variants, including
Linux), Solaris threads, @uref{http://www.be.com,BeOS} threads (tested
on BeOS DR8.2), Mach C threads (reported to work by users), and Win32
threads (reported to work by users). (There is also untested code to
use MacOS MP threads.) We also support using
@uref{http://www.openmp.org,OpenMP} or SGI MP compiler directives to
launch threads, enabled by using @code{withopenmp} or
@code{withsgimp} in addition to @code{enablethreads}. This is
especially useful if you are employing that sort of directive in your
own code, in order to minimize conflicts. If you have a sharedmemory
machine that uses a different threads API, it should be a simple matter
of programming to include support for it; see the file
@code{fftw_threadsint.h} for more detail.
SMP hardware is not required, although of course you need multiple
processors to get any benefit from the multithreaded transforms.
@c 
@node Usage of Multithreaded FFTW, How Many Threads to Use?, Installation and Supported Hardware/Software, Multithreaded FFTW
@subsection Usage of Multithreaded FFTW
Here, it is assumed that the reader is already familiar with the usage
of the uniprocessor FFTW routines, described elsewhere in this manual.
We only describe what one has to change in order to use the
multithreaded routines.
First, instead of including @code{<fftw.h>} or @code{<rfftw.h>}, you
should include the files @code{<fftw_threads.h>} or
@code{<rfftw_threads.h>}, respectively.
Second, before calling any FFTW routines, you should call the function:
@example
int fftw_threads_init(void);
@end example
@findex fftw_threads_init
This function, which should only be called once (probably in your
@code{main()} function), performs any onetime initialization required
to use threads on your system. It returns zero if successful, and a
nonzero value if there was an error (in which case, something is
seriously wrong and you should probably exit the program).
Third, when you want to actually compute the transform, you should use
one of the following transform routines instead of the ordinary FFTW
functions:
@example
fftw_threads(nthreads, plan, howmany, in, istride,
idist, out, ostride, odist);
@findex fftw_threads
fftw_threads_one(nthreads, plan, in, out);
@findex fftw_threads_one
fftwnd_threads(nthreads, plan, howmany, in, istride,
idist, out, ostride, odist);
@findex fftwnd_threads
fftwnd_threads_one(nthreads, plan, in, out);
@findex fftwnd_threads_one
rfftw_threads(nthreads, plan, howmany, in, istride,
idist, out, ostride, odist);
@findex rfftw_threads
rfftw_threads_one(nthreads, plan, in, out);
@findex rfftw_threads_one
rfftwnd_threads_real_to_complex(nthreads, plan, howmany, in,
istride, idist, out, ostride, odist);
@findex rfftwnd_threads_real_to_complex
rfftwnd_threads_one_real_to_complex(nthreads, plan, in, out);
@findex rfftwnd_threads_one_real_to_complex
rfftwnd_threads_complex_to_real(nthreads, plan, howmany, in,
istride, idist, out, ostride, odist);
@findex rfftwnd_threads_complex_to_real
rfftwnd_threads_one_real_to_complex(nthreads, plan, in, out);
@findex rfftwnd_threads_one_real_to_complex
rfftwnd_threads_one_complex_to_real(nthreads, plan, in, out);
@findex rfftwnd_threads_one_complex_to_real
@end example
All of these routines take exactly the same arguments and have exactly
the same effects as their uniprocessor counterparts (i.e. without the
@samp{@code{_threads}}) @emph{except} that they take one extra
parameter, @code{nthreads} (of type @code{int}), before the normal
parameters.@footnote{There is one exception: when performing
onedimensional inplace transforms, the @code{out} parameter is always
ignored by the multithreaded routines, instead of being used as a
workspace if it is non@code{NULL} as in the uniprocessor routines. The
multithreaded routines always allocate their own workspace (the size of
which depends upon the number of threads).} The @code{nthreads}
parameter specifies the number of threads of execution to use when
performing the transform (actually, the maximum number of threads).
@cindex number of threads
For example, to parallelize a single onedimensional transform of
complex data, instead of calling the uniprocessor @code{fftw_one(plan,
in, out)}, you would call @code{fftw_threads_one(nthreads, plan, in,
out)}. Passing an @code{nthreads} of @code{1} means to use only one
thread (the main thread), and is equivalent to calling the uniprocessor
routine. Passing an @code{nthreads} of @code{2} means that the
transform is potentially parallelized over two threads (and two
processors, if you have them), and so on.
These are the only changes you need to make to your source code. Calls
to all other FFTW routines (plan creation, destruction, wisdom,
etcetera) are not parallelized and remain the same. (The same plans and
wisdom are used by both uniprocessor and multithreaded transforms.)
Your arrays are allocated and formatted in the same way, and so on.
Programs using the parallel complex transforms should be linked with
@code{lfftw_threads lfftw lm} on Unix. Programs using the parallel
real transforms should be linked with @code{lrfftw_threads
lfftw_threads lrfftw lfftw lm}. You will also need to link with
whatever library is responsible for threads on your system
(e.g. @code{lpthread} on Linux).
@cindex linking on Unix
@c 
@node How Many Threads to Use?, Using Multithreaded FFTW in a Multithreaded Program, Usage of Multithreaded FFTW, Multithreaded FFTW
@subsection How Many Threads to Use?
@cindex number of threads
There is a fair amount of overhead involved in spawning and synchronizing
threads, so the optimal number of threads to use depends upon the size
of the transform as well as on the number of processors you have.
As a general rule, you don't want to use more threads than you have
processors. (Using more threads will work, but there will be extra
overhead with no benefit.) In fact, if the problem size is too small,
you may want to use fewer threads than you have processors.
You will have to experiment with your system to see what level of
parallelization is best for your problem size. Useful tools to help you
do this are the test programs that are automatically compiled along with
the threads libraries, @code{fftw_threads_test} and
@code{rfftw_threads_test} (in the @code{threads} subdirectory). These
@pindex fftw_threads_test
@pindex rfftw_threads_test
take the same arguments as the other FFTW test programs (see
@code{tests/README}), except that they also take the number of threads
to use as a first argument, and report the parallel speedup in speed
tests. For example,
@example
fftw_threads_test 2 s 128x128
@end example
will benchmark complex 128x128 transforms using two threads and report
the speedup relative to the uniprocessor transform.
@cindex benchmark
For instance, on a 4processor 200MHz Pentium Pro system running Linux
2.2.0, we found that the "crossover" point at which 2 threads became
beneficial for complex transforms was about 4k points, while 4 threads
became beneficial at 8k points.
@c 
@node Using Multithreaded FFTW in a Multithreaded Program, Tips for Optimal Threading, How Many Threads to Use?, Multithreaded FFTW
@subsection Using Multithreaded FFTW in a Multithreaded Program
@cindex thread safety
It is perfectly possible to use the multithreaded FFTW routines from a
multithreaded program (e.g. have multiple threads computing
multithreaded transforms simultaneously). If you have the processors,
more power to you! However, the same restrictions apply as for the
uniprocessor FFTW routines (@pxref{Thread safety}). In particular, you
should recall that you may not create or destroy plans in parallel.
@c 
@node Tips for Optimal Threading, , Using Multithreaded FFTW in a Multithreaded Program, Multithreaded FFTW
@subsection Tips for Optimal Threading
Not all transforms are equally wellparallelized by the multithreaded
FFTW routines. (This is merely a consequence of laziness on the part of
the implementors, and is not inherent to the algorithms employed.)
Mainly, the limitations are in the parallel onedimensional transforms.
The things to avoid if you want optimal parallelization are as follows:
@subsection Parallelization deficiencies in onedimensional transforms
@itemize @bullet
@item
Large prime factors can sometimes parallelize poorly. Of course, you
should avoid these anyway if you want high performance.
@item
@cindex inplace transform
Single inplace transforms don't parallelize completely. (Multiple
inplace transforms, i.e. @code{howmany > 1}, are fine.) Again, you
should avoid these in any case if you want high performance, as they
require transforming to a scratch array and copying back.
@item
Single realcomplex (@code{rfftw}) transforms don't parallelize
completely. This is unfortunate, but parallelizing this correctly would
have involved a lot of extra code (and a much larger library). You
still get some benefit from additional processors, but if you have a
very large number of processors you will probably be better off using
the parallel complex (@code{fftw}) transforms. Note that
multidimensional real transforms or multiple onedimensional real
transforms are fine.
@end itemize
@c 
@node MPI FFTW, , Multithreaded FFTW, Parallel FFTW
@section MPI FFTW
@cindex MPI
This section describes the MPI FFTW routines for distributedmemory (and
sharedmemory) machines supporting MPI (Message Passing Interface). The
MPI routines are significantly different from the ordinary FFTW because
the transform data here are @emph{distributed} over multiple processes,
so that each process gets only a portion of the array.
@cindex distributed memory
Currently, multidimensional transforms of both real and complex data,
as well as onedimensional transforms of complex data, are supported.
@menu
* MPI FFTW Installation::
* Usage of MPI FFTW for Complex Multidimensional Transforms::
* MPI Data Layout::
* Usage of MPI FFTW for Real Multidimensional Transforms::
* Usage of MPI FFTW for Complex Onedimensional Transforms::
* MPI Tips::
@end menu
@c 
@node MPI FFTW Installation, Usage of MPI FFTW for Complex Multidimensional Transforms, MPI FFTW, MPI FFTW
@subsection MPI FFTW Installation
The FFTW MPI library code is all located in the @code{mpi} subdirectoy
of the FFTW package (along with source code for test programs). On Unix
systems, the FFTW MPI libraries and header files can be automatically
configured, compiled, and installed along with the uniprocessor FFTW
libraries simply by including @code{enablempi} in the flags to the
@code{configure} script (@pxref{Installation on Unix}).
@fpindex configure
The only requirement of the FFTW MPI code is that you have the standard
MPI 1.1 (or later) libraries and header files installed on your system.
A free implementation of MPI is available from
@uref{http://wwwunix.mcs.anl.gov/mpi/mpich/,the MPICH home page}.
Previous versions of the FFTW MPI routines have had an unfortunate
tendency to expose bugs in MPI implementations. The current version has
been largely rewritten, and hopefully avoids some of the problems. If
you run into difficulties, try passing the optional workspace to
@code{(r)fftwnd_mpi} (see below), as this allows us to use the standard
(and hopefully welltested) @code{MPI_Alltoall} primitive for
@ffindex MPI_Alltoall
communications. Please let us know (@email{fftw@@fftw.org})
how things work out.
@pindex fftw_mpi_test
@pindex rfftw_mpi_test
Several test programs are included in the @code{mpi} directory. The
ones most useful to you are probably the @code{fftw_mpi_test} and
@code{rfftw_mpi_test} programs, which are run just like an ordinary MPI
program and accept the same parameters as the other FFTW test programs
(c.f. @code{tests/README}). For example, @code{mpirun @i{...params...}
fftw_mpi_test r 0} will run nonterminating complextransform
correctness tests of random dimensions. They can also do performance
benchmarks.
@c 
@node Usage of MPI FFTW for Complex Multidimensional Transforms, MPI Data Layout, MPI FFTW Installation, MPI FFTW
@subsection Usage of MPI FFTW for Complex Multidimensional Transforms
Usage of the MPI FFTW routines is similar to that of the uniprocessor
FFTW. We assume that the reader already understands the usage of the
uniprocessor FFTW routines, described elsewhere in this manual. Some
familiarity with MPI is also helpful.
A typical program performing a complex twodimensional MPI transform
might look something like:
@example
#include <fftw_mpi.h>
int main(int argc, char **argv)
@{
const int NX = ..., NY = ...;
fftwnd_mpi_plan plan;
fftw_complex *data;
MPI_Init(&argc,&argv);
plan = fftw2d_mpi_create_plan(MPI_COMM_WORLD,
NX, NY,
FFTW_FORWARD, FFTW_ESTIMATE);
...allocate and initialize data...
fftwnd_mpi(p, 1, data, NULL, FFTW_NORMAL_ORDER);
...
fftwnd_mpi_destroy_plan(plan);
MPI_Finalize();
@}
@end example
The calls to @code{MPI_Init} and @code{MPI_Finalize} are required in all
@ffindex MPI_Init
@ffindex MPI_Finalize
MPI programs; see the @uref{http://www.mcs.anl.gov/mpi/,MPI home page}
for more information. Note that all of your processes run the program
in parallel, as a group; there is no explicit launching of
threads/processes in an MPI program.
@cindex plan
As in the ordinary FFTW, the first thing we do is to create a plan (of
type @code{fftwnd_mpi_plan}), using:
@example
fftwnd_mpi_plan fftw2d_mpi_create_plan(MPI_Comm comm,
int nx, int ny,
fftw_direction dir, int flags);
@end example
@findex fftw2d_mpi_create_plan
@tindex fftwnd_mpi_plan
Except for the first argument, the parameters are identical to those of
@code{fftw2d_create_plan}. (There are also analogous
@code{fftwnd_mpi_create_plan} and @code{fftw3d_mpi_create_plan}
functions. Transforms of any rank greater than one are supported.)
@findex fftwnd_mpi_create_plan
@findex fftw3d_mpi_create_plan
The first argument is an MPI @dfn{communicator}, which specifies the
group of processes that are to be involved in the transform; the
standard constant @code{MPI_COMM_WORLD} indicates all available
processes.
@fcindex MPI_COMM_WORLD
Next, one has to allocate and initialize the data. This is somewhat
tricky, because the transform data is distributed across the processes
involved in the transform. It is discussed in detail by the next
section (@pxref{MPI Data Layout}).
The actual computation of the transform is performed by the function
@code{fftwnd_mpi}, which differs somewhat from its uniprocessor
equivalent and is described by:
@example
void fftwnd_mpi(fftwnd_mpi_plan p,
int n_fields,
fftw_complex *local_data, fftw_complex *work,
fftwnd_mpi_output_order output_order);
@end example
@findex fftwnd_mpi
There are several things to notice here:
@itemize @bullet
@item
@cindex inplace transform
First of all, all @code{fftw_mpi} transforms are inplace: the output is
in the @code{local_data} parameter, and there is no need to specify
@code{FFTW_IN_PLACE} in the plan flags.
@item
@cindex n_fields
@cindex stride
The MPI transforms also only support a limited subset of the
@code{howmany}/@code{stride}/@code{dist} functionality of the
uniprocessor routines: the @code{n_fields} parameter is equivalent to
@code{howmany=n_fields}, @code{stride=n_fields}, and @code{dist=1}.
(Conceptually, the @code{n_fields} parameter allows you to transform an
array of contiguous vectors, each with length @code{n_fields}.)
@code{n_fields} is @code{1} if you are only transforming a single,
ordinary array.
@item
The @code{work} parameter is an optional workspace. If it is not
@code{NULL}, it should be exactly the same size as the @code{local_data}
array. If it is provided, FFTW is able to use the builtin
@code{MPI_Alltoall} primitive for (often) greater efficiency at the
@ffindex MPI_Alltoall
expense of extra storage space.
@item
Finally, the last parameter specifies whether the output data has the
same ordering as the input data (@code{FFTW_NORMAL_ORDER}), or if it is
transposed (@code{FFTW_TRANSPOSED_ORDER}). Leaving the data transposed
@ctindex FFTW_NORMAL_ORDER
@ctindex FFTW_TRANSPOSED_ORDER
results in significant performance improvements due to a saved
communication step (needed to untranspose the data). Specifically, the
first two dimensions of the array are transposed, as is described in
more detail by the next section.
@end itemize
@cindex normalization
The output of @code{fftwnd_mpi} is identical to that of the
corresponding uniprocessor transform. In particular, you should recall
our conventions for normalization and the sign of the transform
exponent.
The same plan can be used to compute many transforms of the same size.
After you are done with it, you should deallocate it by calling
@code{fftwnd_mpi_destroy_plan}.
@findex fftwnd_mpi_destroy_plan
@cindex blocking
@ffindex MPI_Barrier
@b{Important:} The FFTW MPI routines must be called in the same order by
all processes involved in the transform. You should assume that they
all are blocking, as if each contained a call to @code{MPI_Barrier}.
Programs using the FFTW MPI routines should be linked with
@code{lfftw_mpi lfftw lm} on Unix, in addition to whatever libraries
are required for MPI.
@cindex linking on Unix
@c 
@node MPI Data Layout, Usage of MPI FFTW for Real Multidimensional Transforms, Usage of MPI FFTW for Complex Multidimensional Transforms, MPI FFTW
@subsection MPI Data Layout
@cindex distributed memory
@cindex distributed array format
The transform data used by the MPI FFTW routines is @dfn{distributed}: a
distinct portion of it resides with each process involved in the
transform. This allows the transform to be parallelized, for example,
over a cluster of workstations, each with its own separate memory, so
that you can take advantage of the total memory of all the processors
you are parallelizing over.
In particular, the array is divided according to the rows (first
dimension) of the data: each process gets a subset of the rows of the
@cindex slab decomposition
data. (This is sometimes called a ``slab decomposition.'') One
consequence of this is that you can't take advantage of more processors
than you have rows (e.g. @code{64x64x64} matrix can at most use 64
processors). This isn't usually much of a limitation, however, as each
processor needs a fair amount of data in order for the
parallelcomputation benefits to outweight the communications costs.
Below, the first dimension of the data will be referred to as
@samp{@code{x}} and the second dimension as @samp{@code{y}}.
FFTW supplies a routine to tell you exactly how much data resides on the
current process:
@example
void fftwnd_mpi_local_sizes(fftwnd_mpi_plan p,
int *local_nx,
int *local_x_start,
int *local_ny_after_transpose,
int *local_y_start_after_transpose,
int *total_local_size);
@end example
@findex fftwnd_mpi_local_sizes
Given a plan @code{p}, the other parameters of this routine are set to
values describing the required data layout, described below.
@code{total_local_size} is the number of @code{fftw_complex} elements
that you must allocate for your local data (and workspace, if you
choose). (This value should, of course, be multiplied by
@code{n_fields} if that parameter to @code{fftwnd_mpi} is not @code{1}.)
The data on the current process has @code{local_nx} rows, starting at
row @code{local_x_start}. If @code{fftwnd_mpi} is called with
@code{FFTW_TRANSPOSED_ORDER} output, then @code{y} will be the first
dimension of the output, and the local @code{y} extent will be given by
@code{local_ny_after_transpose} and
@code{local_y_start_after_transpose}. Otherwise, the output has the
same dimensions and layout as the input.
For instance, suppose you want to transform threedimensional data of
size @code{nx x ny x nz}. Then, the current process will store a subset
of this data, of size @code{local_nx x ny x nz}, where the @code{x}
indices correspond to the range @code{local_x_start} to
@code{local_x_start+local_nx1} in the ``real'' (i.e. logical) array.
If @code{fftwnd_mpi} is called with @code{FFTW_TRANSPOSED_ORDER} output,
@ctindex FFTW_TRANSPOSED_ORDER
then the result will be a @code{ny x nx x nz} array, of which a
@code{local_ny_after_transpose x nx x nz} subset is stored on the
current process (corresponding to @code{y} values starting at
@code{local_y_start_after_transpose}).
The following is an example of allocating such a threedimensional array
array (@code{local_data}) before the transform and initializing it to
some function @code{f(x,y,z)}:
@example
fftwnd_mpi_local_sizes(plan, &local_nx, &local_x_start,
&local_ny_after_transpose,
&local_y_start_after_transpose,
&total_local_size);
local_data = (fftw_complex*) malloc(sizeof(fftw_complex) *
total_local_size);
for (x = 0; x < local_nx; ++x)
for (y = 0; y < ny; ++y)
for (z = 0; z < nz; ++z)
local_data[(x*ny + y)*nz + z]
= f(x + local_x_start, y, z);
@end example
Some important things to remember:
@itemize @bullet
@item
Although the local data is of dimensions @code{local_nx x ny x nz} in
the above example, do @emph{not} allocate the array to be of size
@code{local_nx*ny*nz}. Use @code{total_local_size} instead.
@item
The amount of data on each process will not necessarily be the same; in
fact, @code{local_nx} may even be zero for some processes. (For
example, suppose you are doing a @code{6x6} transform on four
processors. There is no way to effectively use the fourth processor in
a slab decomposition, so we leave it empty. Proof left as an exercise
for the reader.)
@item
@cindex rowmajor
All arrays are, of course, in rowmajor order (@pxref{Multidimensional
Array Format}).
@item
If you want to compute the inverse transform of the output of
@code{fftwnd_mpi}, the dimensions of the inverse transform are given by
the dimensions of the output of the forward transform. For example, if
you are using @code{FFTW_TRANSPOSED_ORDER} output in the above example,
then the inverse plan should be created with dimensions @code{ny x nx x
nz}.
@item
The data layout only depends upon the dimensions of the array, not on
the plan, so you are guaranteed that different plans for the same size
(or inverse plans) will use the same (consistent) data layouts.
@end itemize
@c 
@node Usage of MPI FFTW for Real Multidimensional Transforms, Usage of MPI FFTW for Complex Onedimensional Transforms, MPI Data Layout, MPI FFTW
@subsection Usage of MPI FFTW for Real Multidimensional Transforms
MPI transforms specialized for real data are also available, similiar to
the uniprocessor @code{rfftwnd} transforms. Just as in the uniprocessor
case, the realdata MPI functions gain roughly a factor of two in speed
(and save a factor of two in space) at the expense of more complicated
data formats in the calling program. Before reading this section, you
should definitely understand how to call the uniprocessor @code{rfftwnd}
functions and also the complex MPI FFTW functions.
The following is an example of a program using @code{rfftwnd_mpi}. It
computes the size @code{nx x ny x nz} transform of a real function
@code{f(x,y,z)}, multiplies the imaginary part by @code{2} for fun, then
computes the inverse transform. (We'll also use
@code{FFTW_TRANSPOSED_ORDER} output for the transform, and additionally
supply the optional workspace parameter to @code{rfftwnd_mpi}, just to
add a little spice.)
@example
#include <rfftw_mpi.h>
int main(int argc, char **argv)
@{
const int nx = ..., ny = ..., nz = ...;
int local_nx, local_x_start, local_ny_after_transpose,
local_y_start_after_transpose, total_local_size;
int x, y, z;
rfftwnd_mpi_plan plan, iplan;
fftw_real *data, *work;
fftw_complex *cdata;
MPI_Init(&argc,&argv);
/* create the forward and backward plans: */
plan = rfftw3d_mpi_create_plan(MPI_COMM_WORLD,
nx, ny, nz,
FFTW_REAL_TO_COMPLEX,
FFTW_ESTIMATE);
@findex rfftw3d_mpi_create_plan
iplan = rfftw3d_mpi_create_plan(MPI_COMM_WORLD,
/* dim.'s of REAL data > */ nx, ny, nz,
FFTW_COMPLEX_TO_REAL,
FFTW_ESTIMATE);
rfftwnd_mpi_local_sizes(plan, &local_nx, &local_x_start,
&local_ny_after_transpose,
&local_y_start_after_transpose,
&total_local_size);
@findex rfftwnd_mpi_local_sizes
data = (fftw_real*) malloc(sizeof(fftw_real) * total_local_size);
/* workspace is the same size as the data: */
work = (fftw_real*) malloc(sizeof(fftw_real) * total_local_size);
/* initialize data to f(x,y,z): */
for (x = 0; x < local_nx; ++x)
for (y = 0; y < ny; ++y)
for (z = 0; z < nz; ++z)
data[(x*ny + y) * (2*(nz/2+1)) + z]
= f(x + local_x_start, y, z);
/* Now, compute the forward transform: */
rfftwnd_mpi(plan, 1, data, work, FFTW_TRANSPOSED_ORDER);
@findex rfftwnd_mpi
/* the data is now complex, so typecast a pointer: */
cdata = (fftw_complex*) data;
/* multiply imaginary part by 2, for fun:
(note that the data is transposed) */
for (y = 0; y < local_ny_after_transpose; ++y)
for (x = 0; x < nx; ++x)
for (z = 0; z < (nz/2+1); ++z)
cdata[(y*nx + x) * (nz/2+1) + z].im
*= 2.0;
/* Finally, compute the inverse transform; the result
is transposed back to the original data layout: */
rfftwnd_mpi(iplan, 1, data, work, FFTW_TRANSPOSED_ORDER);
free(data);
free(work);
rfftwnd_mpi_destroy_plan(plan);
@findex rfftwnd_mpi_destroy_plan
rfftwnd_mpi_destroy_plan(iplan);
MPI_Finalize();
@}
@end example
There's a lot of stuff in this example, but it's all just what you would
have guessed, right? We replaced all the @code{fftwnd_mpi*} functions
by @code{rfftwnd_mpi*}, but otherwise the parameters were pretty much
the same. The data layout distributed among the processes just like for
the complex transforms (@pxref{MPI Data Layout}), but in addition the
final dimension is padded just like it is for the uniprocessor inplace
real transforms (@pxref{Array Dimensions for Real Multidimensional
Transforms}).
@cindex padding
In particular, the @code{z} dimension of the real input data is padded
to a size @code{2*(nz/2+1)}, and after the transform it contains
@code{nz/2+1} complex values.
@cindex distributed array format
@cindex rfftwnd array format
Some other important things to know about the real MPI transforms:
@itemize @bullet
@item
As for the uniprocessor @code{rfftwnd_create_plan}, the dimensions
passed for the @code{FFTW_COMPLEX_TO_REAL} plan are those of the
@emph{real} data. In particular, even when @code{FFTW_TRANSPOSED_ORDER}
@ctindex FFTW_COMPLEX_TO_REAL
@ctindex FFTW_TRANSPOSED_ORDER
is used as in this case, the dimensions are those of the (untransposed)
real output, not the (transposed) complex input. (For the complex MPI
transforms, on the other hand, the dimensions are always those of the
input array.)
@item
The output ordering of the transform (@code{FFTW_TRANSPOSED_ORDER} or
@code{FFTW_TRANSPOSED_ORDER}) @emph{must} be the same for both forward
and backward transforms. (This is not required in the complex case.)
@item
@code{total_local_size} is the required size in @code{fftw_real} values,
not @code{fftw_complex} values as it is for the complex transforms.
@item
@code{local_ny_after_transpose} and @code{local_y_start_after_transpose}
describe the portion of the array after the transform; that is, they are
indices in the complex array for an @code{FFTW_REAL_TO_COMPLEX} transform
and in the real array for an @code{FFTW_COMPLEX_TO_REAL} transform.
@item
@code{rfftwnd_mpi} always expects @code{fftw_real*} array arguments, but
of course these pointers can refer to either real or complex arrays,
depending upon which side of the transform you are on. Just as for
inplace uniprocessor real transforms (and also in the example above),
this is most easily handled by typecasting to a complex pointer when
handling the complex data.
@item
As with the complex transforms, there are also
@code{rfftwnd_create_plan} and @code{rfftw2d_create_plan} functions, and
any rank greater than one is supported.
@findex rfftwnd_create_plan
@findex rfftw2d_create_plan
@end itemize
Programs using the MPI FFTW real transforms should link with
@code{lrfftw_mpi lfftw_mpi lrfftw lfftw lm} on Unix.
@cindex linking on Unix
@c 
@node Usage of MPI FFTW for Complex Onedimensional Transforms, MPI Tips, Usage of MPI FFTW for Real Multidimensional Transforms, MPI FFTW
@subsection Usage of MPI FFTW for Complex Onedimensional Transforms
The MPI FFTW also includes routines for parallel onedimensional
transforms of complex data (only). Although the speedup is generally
worse than it is for the multidimensional routines,@footnote{The 1D
transforms require much more communication. All the communication in
our FFT routines takes the form of an alltoall communication: the
multidimensional transforms require two alltoall communications (or
one, if you use @code{FFTW_TRANSPOSED_ORDER}), while the onedimensional
transforms require @emph{three} (or two, if you use scrambled input or
output).} these distributedmemory onedimensional transforms are
especially useful for performing onedimensional transforms that don't
fit into the memory of a single machine.
The usage of these routines is straightforward, and is similar to that
of the multidimensional MPI transform functions. You first include the
header @code{<fftw_mpi.h>} and then create a plan by calling:
@example
fftw_mpi_plan fftw_mpi_create_plan(MPI_Comm comm, int n,
fftw_direction dir, int flags);
@end example
@findex fftw_mpi_create_plan
@tindex fftw_mpi_plan
The last three arguments are the same as for @code{fftw_create_plan}
(except that all MPI transforms are automatically @code{FFTW_IN_PLACE}).
The first argument specifies the group of processes you are using, and
is usually @code{MPI_COMM_WORLD} (all processes).
@fcindex MPI_COMM_WORLD
A plan can be used for many transforms of the same size, and is
destroyed when you are done with it by calling
@code{fftw_mpi_destroy_plan(plan)}.
@findex fftw_mpi_destroy_plan
If you don't care about the ordering of the input or output data of the
transform, you can include @code{FFTW_SCRAMBLED_INPUT} and/or
@code{FFTW_SCRAMBLED_OUTPUT} in the @code{flags}.
@ctindex FFTW_SCRAMBLED_INPUT
@ctindex FFTW_SCRAMBLED_OUTPUT
@cindex flags
These save some communications at the expense of having the input and/or
output reordered in an undocumented way. For example, if you are
performing an FFTbased convolution, you might use
@code{FFTW_SCRAMBLED_OUTPUT} for the forward transform and
@code{FFTW_SCRAMBLED_INPUT} for the inverse transform.
The transform itself is computed by:
@example
void fftw_mpi(fftw_mpi_plan p, int n_fields,
fftw_complex *local_data, fftw_complex *work);
@end example
@findex fftw_mpi
@cindex n_fields
@code{n_fields}, as in @code{fftwnd_mpi}, is equivalent to
@code{howmany=n_fields}, @code{stride=n_fields}, and @code{dist=1}, and
should be @code{1} when you are computing the transform of a single
array. @code{local_data} contains the portion of the array local to the
current process, described below. @code{work} is either @code{NULL} or
an array exactly the same size as @code{local_data}; in the latter case,
FFTW can use the @code{MPI_Alltoall} communications primitive which is
(usually) faster at the expense of extra storage. Upon return,
@code{local_data} contains the portion of the output local to the
current process (see below).
@ffindex MPI_Alltoall
@cindex distributed array format
To find out what portion of the array is stored local to the current
process, you call the following routine:
@example
void fftw_mpi_local_sizes(fftw_mpi_plan p,
int *local_n, int *local_start,
int *local_n_after_transform,
int *local_start_after_transform,
int *total_local_size);
@end example
@findex fftw_mpi_local_sizes
@code{total_local_size} is the number of @code{fftw_complex} elements
you should actually allocate for @code{local_data} (and @code{work}).
@code{local_n} and @code{local_start} indicate that the current process
stores @code{local_n} elements corresponding to the indices
@code{local_start} to @code{local_start+local_n1} in the ``real''
array. @emph{After the transform, the process may store a different
portion of the array.} The portion of the data stored on the process
after the transform is given by @code{local_n_after_transform} and
@code{local_start_after_transform}. This data is exactly the same as a
contiguous segment of the corresponding uniprocessor transform output
(i.e. an inorder sequence of sequential frequency bins).
Note that, if you compute both a forward and a backward transform of the
same size, the local sizes are guaranteed to be consistent. That is,
the local size after the forward transform will be the same as the local
size before the backward transform, and vice versa.
Programs using the FFTW MPI routines should be linked with
@code{lfftw_mpi lfftw lm} on Unix, in addition to whatever libraries
are required for MPI.
@cindex linking on Unix
@c 
@node MPI Tips, , Usage of MPI FFTW for Complex Onedimensional Transforms, MPI FFTW
@subsection MPI Tips
There are several things you should consider in order to get the best
performance out of the MPI FFTW routines.
@cindex loadbalancing
First, if possible, the first and second dimensions of your data should
be divisible by the number of processes you are using. (If only one can
be divisible, then you should choose the first dimension.) This allows
the computational load to be spread evenly among the processes, and also
reduces the communications complexity and overhead. In the
onedimensional transform case, the size of the transform should ideally
be divisible by the @emph{square} of the number of processors.
@ctindex FFTW_TRANSPOSED_ORDER
Second, you should consider using the @code{FFTW_TRANSPOSED_ORDER}
output format if it is not too burdensome. The speed gains from
communications savings are usually substantial.
Third, you should consider allocating a workspace for
@code{(r)fftw(nd)_mpi}, as this can often
(but not always) improve performance (at the cost of extra storage).
Fourth, you should experiment with the best number of processors to use
for your problem. (There comes a point of diminishing returns, when the
communications costs outweigh the computational benefits.@footnote{An
FFT is particularly hard on communications systems, as it requires an
@dfn{alltoall} communication, which is more or less the worst possible
case.}) The @code{fftw_mpi_test} program can output helpful performance
benchmarks.
@pindex fftw_mpi_test
@cindex benchmark
It accepts the same parameters as the uniprocessor test programs
(c.f. @code{tests/README}) and is run like an ordinary MPI program. For
example, @code{mpirun np 4 fftw_mpi_test s 128x128x128} will benchmark
a @code{128x128x128} transform on four processors, reporting timings and
parallel speedups for all variants of @code{fftwnd_mpi} (transposed,
with workspace, etcetera). (Note also that there is the
@code{rfftw_mpi_test} program for the real transforms.)
@pindex rfftw_mpi_test
@c ************************************************************
@node Calling FFTW from Fortran, Installation and Customization, Parallel FFTW, Top
@chapter Calling FFTW from Fortran
@cindex Fortrancallable wrappers
The standard FFTW libraries include special wrapper functions that allow
Fortran programs to call FFTW subroutines. This chapter describes how
those functions may be employed to use FFTW from Fortran. We assume
here that the reader is already familiar with the usage of FFTW in C, as
described elsewhere in this manual.
In general, it is not possible to call C functions directly from
Fortran, due to Fortran's inability to pass arguments by value and also
because Fortran compilers typically expect identifiers to be mangled
@cindex compiler
somehow for linking. However, if C functions are written in a special
way, they @emph{are} callable from Fortran, and we have employed this
technique to create Fortrancallable ``wrapper'' functions around the
main FFTW routines. These wrapper functions are included in the FFTW
libraries by default, unless a Fortran compiler isn't found on your
system or @code{disablefortran} is included in the @code{configure}
flags.
As a result, calling FFTW from Fortran requires little more than
appending @samp{@code{_f77}} to the function names and then linking
normally with the FFTW libraries. There are a few wrinkles, however, as
we shall discuss below.
@menu
* Wrapper Routines::
* FFTW Constants in Fortran::
* Fortran Examples::
@end menu
@c 
@node Wrapper Routines, FFTW Constants in Fortran, Calling FFTW from Fortran, Calling FFTW from Fortran
@section Wrapper Routines
All of the uniprocessor and multithreaded transform routines have
Fortrancallable wrappers, except for the wisdom import/export functions
(since it is not possible to exchange string and file arguments portably
with Fortran) and the specific planner routines (@pxref{Discussion on
Specific Plans}). The name of the wrapper routine is the same as that
of the corresponding C routine, but with
@code{fftw/fftwnd/rfftw/rfftwnd} replaced by
@code{fftw_f77/fftwnd_f77/rfftw_f77/rfftwnd_f77}. For example, in
Fortran, instead of calling @code{fftw_one} you would call
@code{fftw_f77_one}.@footnote{Technically, Fortran 77 identifiers are
not allowed to have more than 6 characters, nor may they contain
underscores. Any compiler that enforces this limitation doesn't deserve
to link to FFTW.}
@findex fftw_f77_one
For the most part, all of the arguments to the functions are the same,
with the following exceptions:
@itemize @bullet
@item
@code{plan} variables (what would be of type @code{fftw_plan},
@code{rfftwnd_plan}, etcetera, in C), must be declared as a type that is
the same size as a pointer (address) on your machine. (Fortran has no
generic pointer type.) The Fortran @code{integer} type is usually the
same size as a pointer, but you need to be wary (especially on 64bit
machines). (You could also use @code{integer*4} on a 32bit machine and
@code{integer*8} on a 64bit machine.) Ugh. (@code{g77} has a special
type, @code{integer(kind=7)}, that is defined to be the same size as a
pointer.)
@item
Any function that returns a value (e.g. @code{fftw_create_plan}) is
converted into a subroutine. The return value is converted into an
additional (first) parameter of the wrapper subroutine. (The reason for
this is that some Fortran implementations seem to have trouble with C
function return values.)
@item
@cindex inplace transform
When performing onedimensional @code{FFTW_IN_PLACE} transforms, you
don't have the option of passing @code{NULL} for the @code{out} argument
(since there is no way to pass @code{NULL} from Fortran). Therefore,
when performing such transforms, you @emph{must} allocate and pass a
contiguous scratch array of the same size as the transform. Note that
for inplace multidimensional (@code{(r)fftwnd}) transforms, the
@code{out} argument is ignored, so you can pass anything for that
parameter.
@item
@cindex columnmajor
The wrapper routines expect multidimensional arrays to be in
columnmajor order, which is the ordinary format of Fortran arrays.
They do this transparently and costlessly simply by reversing the order
of the dimensions passed to FFTW, but this has one important consequence
for multidimensional realcomplex transforms, discussed below.
@end itemize
@cindex floatingpoint precision
In general, you should take care to use Fortran data types that
correspond to (i.e. are the same size as) the C types used by FFTW. If
your C and Fortran compilers are made by the same vendor, the
correspondence is usually straightforward (i.e. @code{integer}
corresponds to @code{int}, @code{real} corresponds to @code{float},
etcetera). Such simple correspondences are assumed in the examples
below. The examples also assume that FFTW was compiled in
double precision (the default).
@c 
@node FFTW Constants in Fortran, Fortran Examples, Wrapper Routines, Calling FFTW from Fortran
@section FFTW Constants in Fortran
When creating plans in FFTW, a number of constants are used to specify
options, such as @code{FFTW_FORWARD} or @code{FFTW_USE_WISDOM}. The
same constants must be used with the wrapper routines, but of course the
C header files where the constants are defined can't be incorporated
directly into Fortran code.
Instead, we have placed Fortran equivalents of the FFTW constant
definitions in the file @code{fortran/fftw_f77.i} of the FFTW package.
If your Fortran compiler supports a preprocessor, you can use that to
incorporate this file into your code whenever you need to call FFTW.
Otherwise, you will have to paste the constant definitions in directly.
They are:
@example
integer FFTW_FORWARD,FFTW_BACKWARD
parameter (FFTW_FORWARD=1,FFTW_BACKWARD=1)
integer FFTW_REAL_TO_COMPLEX,FFTW_COMPLEX_TO_REAL
parameter (FFTW_REAL_TO_COMPLEX=1,FFTW_COMPLEX_TO_REAL=1)
integer FFTW_ESTIMATE,FFTW_MEASURE
parameter (FFTW_ESTIMATE=0,FFTW_MEASURE=1)
integer FFTW_OUT_OF_PLACE,FFTW_IN_PLACE,FFTW_USE_WISDOM
parameter (FFTW_OUT_OF_PLACE=0)
parameter (FFTW_IN_PLACE=8,FFTW_USE_WISDOM=16)
integer FFTW_THREADSAFE
parameter (FFTW_THREADSAFE=128)
@end example
@cindex flags
In C, you combine different flags (like @code{FFTW_USE_WISDOM} and
@code{FFTW_MEASURE}) using the @samp{@code{}} operator; in Fortran you
should just use @samp{@code{+}}.
@c 
@node Fortran Examples, , FFTW Constants in Fortran, Calling FFTW from Fortran
@section Fortran Examples
In C you might have something like the following to transform a
onedimensional complex array:
@example
fftw_complex in[N], *out[N];
fftw_plan plan;
plan = fftw_create_plan(N,FFTW_FORWARD,FFTW_ESTIMATE);
fftw_one(plan,in,out);
fftw_destroy_plan(plan);
@end example
In Fortran, you use the following to accomplish the same thing:
@example
double complex in, out
dimension in(N), out(N)
integer plan
call fftw_f77_create_plan(plan,N,FFTW_FORWARD,FFTW_ESTIMATE)
call fftw_f77_one(plan,in,out)
call fftw_f77_destroy_plan(plan)
@end example
@findex fftw_f77_create_plan
@findex fftw_f77_one
@findex fftw_f77_destroy_plan
Notice how all routines are called as Fortran subroutines, and the plan
is returned via the first argument to @code{fftw_f77_create_plan}.
@emph{Important:} these examples assume that @code{integer} is the same
size as a pointer, and may need modification on a 64bit machine.
@xref{Wrapper Routines}, above. To do the same thing, but using 8
threads in parallel (@pxref{Multithreaded FFTW}), you would simply
replace the call to @code{fftw_f77_one} with:
@example
call fftw_f77_threads_one(8,plan,in,out)
@end example
@findex fftw_f77_threads_one
To transform a threedimensional array inplace with C, you might do:
@example
fftw_complex arr[L][M][N];
fftwnd_plan plan;
int n[3] = @{L,M,N@};
plan = fftwnd_create_plan(3,n,FFTW_FORWARD,
FFTW_ESTIMATE  FFTW_IN_PLACE);
fftwnd_one(plan, arr, 0);
fftwnd_destroy_plan(plan);
@end example
In Fortran, you would use this instead:
@example
double complex arr
dimension arr(L,M,N)
integer n
dimension n(3)
integer plan
n(1) = L
n(2) = M
n(3) = N
call fftwnd_f77_create_plan(plan,3,n,FFTW_FORWARD,
+ FFTW_ESTIMATE + FFTW_IN_PLACE)
call fftwnd_f77_one(plan, arr, 0)
call fftwnd_f77_destroy_plan(plan)
@end example
@findex fftwnd_f77_create_plan
@findex fftwnd_f77_one
@findex fftwnd_f77_destroy_plan
Instead of calling @code{fftwnd_f77_create_plan(plan,3,n,...)}, we could
also have called @code{fftw3d_f77_create_plan(plan,L,M,N,...)}.
@findex fftw3d_f77_create_plan
Note that we pass the array dimensions in the "natural" order; also
note that the last argument to @code{fftwnd_f77} is ignored since the
transform is @code{FFTW_IN_PLACE}.
To transform a onedimensional real array in Fortran, you might do:
@example
double precision in, out
dimension in(N), out(N)
integer plan
call rfftw_f77_create_plan(plan,N,FFTW_REAL_TO_COMPLEX,
+ FFTW_ESTIMATE)
call rfftw_f77_one(plan,in,out)
call rfftw_f77_destroy_plan(plan)
@end example
@findex rfftw_f77_create_plan
@findex rfftw_f77_one
@findex rfftw_f77_destroy_plan
To transform a twodimensional real array, out of place, you might use
the following:
@example
double precision in
double complex out
dimension in(M,N), out(M/2 + 1, N)
integer plan
call rfftw2d_f77_create_plan(plan,M,N,FFTW_REAL_TO_COMPLEX,
+ FFTW_ESTIMATE)
call rfftwnd_f77_one_real_to_complex(plan, in, out)
call rfftwnd_f77_destroy_plan(plan)
@end example
@findex rfftw2d_f77_create_plan
@findex rfftwnd_f77_one_real_to_complex
@findex rfftwnd_f77_destroy_plan
@b{Important:} Notice that it is the @emph{first} dimension of the
complex output array that is cut in half in Fortran, rather than the
last dimension as in C. This is a consequence of the wrapper routines
reversing the order of the array dimensions passed to FFTW so that the
Fortran program can use its ordinary columnmajor order.
@cindex columnmajor
@cindex rfftwnd array format
@c ************************************************************
@node Installation and Customization, Acknowledgments, Calling FFTW from Fortran, Top
@chapter Installation and Customization
This chapter describes the installation and customization of FFTW, the
latest version of which may be downloaded from
@uref{http://www.fftw.org, the FFTW home page}.
As distributed, FFTW makes very few assumptions about your system. All
you need is an ANSI C compiler (@code{gcc} is fine, although
vendorprovided compilers often produce faster code).
@cindex compiler
However, installation of FFTW is somewhat simpler if you have a Unix or
a GNU system, such as Linux. In this chapter, we first describe the
installation of FFTW on Unix and nonUnix systems. We then describe how
you can customize FFTW to achieve better performance. Specifically, you
can I) enable @code{gcc}/x86specific hacks that improve performance on
Pentia and PentiumPro's; II) adapt FFTW to use the highresolution clock
of your machine, if any; III) produce code (@emph{codelets}) to support
fast transforms of sizes that are not supported efficiently by the
standard FFTW distribution.
@cindex installation
@menu
* Installation on Unix::
* Installation on nonUnix Systems::
* Installing FFTW in both single and double precision::
* gcc and Pentium hacks::
* Customizing the timer::
* Generating your own code::
@end menu
@node Installation on Unix, Installation on nonUnix Systems, Installation and Customization, Installation and Customization
@section Installation on Unix
FFTW comes with a @code{configure} program in the GNU style.
Installation can be as simple as:
@fpindex configure
@example
./configure
make
make install
@end example
This will build the uniprocessor complex and real transform libraries
along with the test programs. We strongly recommend that you use GNU
@code{make} if it is available; on some systems it is called
@code{gmake}. The ``@code{make install}'' command installs the fftw and
rfftw libraries in standard places, and typically requires root
privileges (unless you specify a different install directory with the
@code{prefix} flag to @code{configure}). You can also type
``@code{make check}'' to put the FFTW test programs through their paces.
If you have problems during configuration or compilation, you may want
to run ``@code{make distclean}'' before trying again; this ensures that
you don't have any stale files left over from previous compilation
attempts.
The @code{configure} script knows good @code{CFLAGS} (C compiler flags)
@cindex compiler flags
for a few systems. If your system is not known, the @code{configure}
script will print out a warning. @footnote{Each version of @code{cc}
seems to have its own magic incantation to get the fastest code most of
the timeyou'd think that people would have agreed upon some
convention, e.g. "@code{Omax}", by now.} In this case, you can compile
FFTW with the command
@example
make CFLAGS="<write your CFLAGS here>"
@end example
If you do find an optimal set of @code{CFLAGS} for your system, please
let us know what they are (along with the output of @code{config.guess})
so that we can include them in future releases.
The @code{configure} program supports all the standard flags defined by
the GNU Coding Standards; see the @code{INSTALL} file in FFTW or
@uref{http://www.gnu.org/prep/standards_toc.html, the GNU web page}.
Note especially @code{help} to list all flags and
@code{enableshared} to create shared, rather than static, libraries.
@code{configure} also accepts a few FFTWspecific flags, particularly:
@itemize @bullet
@item
@cindex floatingpoint precision
@code{enablefloat} Produces a singleprecision version of FFTW
(@code{float}) instead of the default doubleprecision (@code{double}).
@xref{Installing FFTW in both single and double precision}.
@item
@code{enabletypeprefix} Adds a @samp{d} or @samp{s} prefix to all
installed libraries and header files to indicate the floatingpoint
precision. @xref{Installing FFTW in both single and double
precision}. (@code{enabletypeprefix=<prefix>} lets you add an
arbitrary prefix.) By default, no prefix is used.
@item
@cindex threads
@code{enablethreads} Enables compilation and installation of the FFTW
threads library (@pxref{Multithreaded FFTW}), which provides a
simple interface to parallel transforms for SMP systems. (By default,
the threads routines are not compiled.)
@item
@code{withopenmp}, @code{withsgimp}
In conjunction with @code{enablethreads}, causes the multithreaded
FFTW library to use OpenMP or SGI MP compiler directives in order to
induce parallelism, rather than spawning its own threads directly.
(Useful especially for programs already employing such directives, in
order to minimize conflicts between different parallelization
mechanisms.)
@item
@cindex MPI
@code{enablempi} Enables compilation and installation of the FFTW MPI
library (@pxref{MPI FFTW}), which provides parallel transforms for
distributedmemory systems with MPI. (By default, the MPI routines are
not compiled.)
@item
@cindex Fortrancallable wrappers
@code{disablefortran} Disables inclusion of Fortrancallable wrapper
routines (@pxref{Calling FFTW from Fortran}) in the standard FFTW
libraries. These wrapper routines increase the library size by only a
negligible amount, so they are included by default as long as the
@code{configure} script finds a Fortran compiler on your system.
@item
@code{withgcc} Enables the use of @code{gcc}. By default, FFTW uses
the vendorsupplied @code{cc} compiler if present. Unfortunately,
@code{gcc} produces slower code than @code{cc} on many systems.
@item
@code{enablei386hacks} @xref{gcc and Pentium hacks}, below.
@item
@code{enablepentiumtimer} @xref{gcc and Pentium hacks}, below.
@end itemize
To force @code{configure} to use a particular C compiler (instead of the
@cindex compiler
default, usually @code{cc}), set the environment variable @code{CC} to
the name of the desired compiler before running @code{configure}; you
may also need to set the flags via the variable @code{CFLAGS}.
@cindex compiler flags
@node Installation on nonUnix Systems, Installing FFTW in both single and double precision, Installation on Unix, Installation and Customization
@section Installation on nonUnix Systems
It is quite straightforward to install FFTW even on nonUnix systems
lacking the niceties of the @code{configure} script. The FFTW Home Page
may include some FFTW packages preconfigured for particular
systems/compilers, and also contains installation notes sent in by
@cindex compiler
users. All you really need to do, though, is to compile all of the
@code{.c} files in the appropriate directories of the FFTW package.
(You needn't worry about the many extraneous files lying around.)
For the complex transforms, compile all of the @code{.c} files in the
@code{fftw} directory and link them into a library. Similarly, for the
real transforms, compile all of the @code{.c} files in the @code{rfftw}
directory into a library. Note that these sources @code{#include}
various files in the @code{fftw} and @code{rfftw} directories, so you
may need to set up the @code{#include} paths for your compiler
appropriately. Be sure to enable the highestpossible level of
optimization in your compiler.
@cindex floatingpoint precision
By default, FFTW is compiled for doubleprecision transforms. To work
in single precision rather than double precision, @code{#define} the
symbol @code{FFTW_ENABLE_FLOAT} in @code{fftw.h} (in the @code{fftw}
directory) and (re)compile FFTW.
These libraries should be linked with any program that uses the
corresponding transforms. The required header files, @code{fftw.h} and
@code{rfftw.h}, are located in the @code{fftw} and @code{rfftw}
directories respectively; you may want to put them with the libraries,
or wherever header files normally go on your system.
FFTW includes test programs, @code{fftw_test} and @code{rfftw_test}, in
@pindex fftw_test
@pindex rfftw_test
the @code{tests} directory. These are compiled and linked like any
program using FFTW, except that they use additional header files located
in the @code{fftw} and @code{rfftw} directories, so you will need to set
your compiler @code{#include} paths appropriately. @code{fftw_test} is
compiled from @code{fftw_test.c} and @code{test_main.c}, while
@code{rfftw_test} is compiled from @code{rfftw_test.c} and
@code{test_main.c}. When you run these programs, you will be prompted
interactively for various possible tests to perform; see also
@code{tests/README} for more information.
@node Installing FFTW in both single and double precision, gcc and Pentium hacks, Installation on nonUnix Systems, Installation and Customization
@section Installing FFTW in both single and double precision
@cindex floatingpoint precision
It is often useful to install both single and doubleprecision versions
of the FFTW libraries on the same machine, and we provide a convenient
mechanism for achieving this on Unix systems.
@fpindex configure
When the @code{enabletypeprefix} option of configure is used, the
FFTW libraries and header files are installed with a prefix of @samp{d}
or @samp{s}, depending upon whether you compiled in double or single
precision. Then, instead of linking your program with @code{lrfftw
lfftw}, for example, you would link with @code{ldrfftw ldfftw} to use
the doubleprecision version or with @code{lsrfftw lsfftw} to use the
singleprecision version. Also, you would @code{#include}
@code{<drfftw.h>} or @code{<srfftw.h>} instead of @code{<rfftw.h>}, and
so on.
@emph{The names of FFTW functions, data types, and constants remain
unchanged!} You still call, for instance, @code{fftw_one} and not
@code{dfftw_one}. Only the names of header files and libraries are
modified. One consequence of this is that @emph{you @b{cannot} use both
the single and doubleprecision FFTW libraries in the same program,
simultaneously,} as the function names would conflict.
So, to install both the single and doubleprecision libraries on the
same machine, you would do:
@example
./configure enabletypeprefix @i{[ other options ]}
make
make install
make clean
./configure enablefloat enabletypeprefix @i{[ other options ]}
make
make install
@end example
@node gcc and Pentium hacks, Customizing the timer, Installing FFTW in both single and double precision, Installation and Customization
@section @code{gcc} and Pentium hacks
@cindex Pentium hack
The @code{configure} option @code{enablei386hacks} enables specific
optimizations for the Pentium and later x86 CPUs under gcc, which can
significantly improve performance of doubleprecision transforms.
Specifically, we have tested these hacks on Linux with @code{gcc} 2.[789]
and versions of @code{egcs} since 1.0.3. These optimizations affect
only the performance and not the correctness of FFTW (i.e. it is always
safe to try them out).
These hacks provide a workaround to the incorrect alignment of local
@code{double} variables in @code{gcc}. The
compiler aligns these
@cindex compiler
variables to multiples of 4 bytes, but execution is much faster (on
Pentium and PentiumPro) if @code{double}s are aligned to a multiple of 8
bytes. By carefully counting the number of variables allocated by the
compiler in performancecritical regions of the code, we have been able
to introduce dummy allocations (using @code{alloca}) that align the
stack properly. The hack depends crucially on the compiler flags that
are used. For example, it won't work without
@code{fomitframepointer}.
In principle, these hacks are no longer required under @code{gcc}
versions 2.95 and later, which automatically align the stack correctly
(see @code{mpreferredstackboundary} in the @code{gcc} manual).
However, we have encountered a
@uref{http://egcs.cygnus.com/ml/gccbugs/199911/msg00259.html,bug} in
the stack alignment of versions 2.95.[012] that causes FFTW's stack to
be misaligned under some circumstances. The @code{configure} script
automatically detects this bug and disables @code{gcc}'s stack alignment
in favor of our own hacks when @code{enablei386hacks} is used.
The @code{fftw_test} program outputs speed measurements that you can use
to see if these hacks are beneficial.
@pindex fftw_test
@cindex benchmark
The @code{configure} option @code{enablepentiumtimer} enables the
use of the Pentium and PentiumPro cycle counter for timing purposes. In
order to get correct results, you must define @code{FFTW_CYCLES_PER_SEC}
in @code{fftw/config.h} to be the clock speed of your processor; the
resulting FFTW library will be nonportable. The use of this option is
deprecated. On serious operating systems (such as Linux), FFTW uses
@code{gettimeofday()}, which has enough resolution and is portable.
(Note that Win32 has its own highresolution timing routines as well.
FFTW contains unsupported code to use these routines.)
@node Customizing the timer, Generating your own code, gcc and Pentium hacks, Installation and Customization
@section Customizing the timer
@cindex timer, customization of
FFTW needs a reasonablyprecise clock in order to find the optimal way
to compute a transform. On Unix systems, @code{configure} looks for
@code{gettimeofday} and other systemspecific timers. If it does not
find any high resolution clock, it defaults to using the @code{clock()}
function, which is very portable, but forces FFTW to run for a long time
in order to get reliable measurements.
@ffindex gettimeofday
@ffindex clock
If your machine supports a highresolution clock not recognized by FFTW,
it is therefore advisable to use it. You must edit
@code{fftw/fftwint.h}. There are a few macros you must redefine. The
code is documented and should be selfexplanatory. (By the way,
@code{fftwint} stands for @code{fftwinternal}, but for some
inexplicable reason people are still using primitive systems with 8.3
filenames.)
Even if you don't install highresolution timing code, we still
recommend that you look at the @code{FFTW_TIME_MIN} constant in
@ctindex FFTW_TIME_MIN
@code{fftw/fftwint.h}. This constant holds the minimum time interval (in
seconds) required to get accurate timing measurements, and should be (at
least) several hundred times the resolution of your clock. The default
constants are on the conservative side, and may cause FFTW to take
longer than necessary when you create a plan. Set @code{FFTW_TIME_MIN}
to whatever is appropriate on your system (be sure to set the
@emph{right} @code{FFTW_TIME_MIN}@dots{}there are several definitions in
@code{fftwint.h}, corresponding to different platforms and timers).
As an aid in checking the resolution of your clock, you can use the
@code{tests/fftw_test} program with the @code{t} option
(c.f. @code{tests/README}). Remember, the mere fact that your clock
reports times in, say, picoseconds, does not mean that it is actually
@emph{accurate} to that resolution.
@node Generating your own code, , Customizing the timer, Installation and Customization
@section Generating your own code
@cindex Caml
@cindex ML
@cindex code generator
If you know that you will only use transforms of a certain size (say,
powers of @math{2}) and want to reduce the size of the library, you can
reconfigure FFTW to support only those sizes you are interested in. You
may even generate code to enable efficient transforms of a size not
supported by the default distribution. The default distribution
supports transforms of any size, but not all sizes are equally fast.
The default installation of FFTW is best at handling sizes of the form
@ifinfo
@math{2^a 3^b 5^c 7^d 11^e 13^f},
@end ifinfo
@tex
$2^a 3^b 5^c 7^d 11^e 13^f$,
@end tex
@ifhtml
2<SUP>a</SUP> 3<SUP>b</SUP> 5<SUP>c</SUP> 7<SUP>d</SUP>
11<SUP>e</SUP> 13<SUP>f</SUP>,
@end ifhtml
where @math{e+f} is either @math{0} or
@math{1}, and the other exponents are arbitrary. Other sizes are
computed by means of a slow, generalpurpose routine. However, if you
have an application that requires fast transforms of size, say,
@code{17}, there is a way to generate specialized code to handle that.
The directory @code{gensrc} contains all the programs and scripts that
were used to generate FFTW. In particular, the program
@code{gensrc/genfft.ml} was used to generate the code that FFTW uses to
compute the transforms. We do not expect casual users to use it.
@code{genfft} is a rather sophisticated program that generates directed
acyclic graphs of FFT algorithms and performs algebraic simplifications
on them. @code{genfft} is written in Objective Caml, a dialect of ML.
Objective Caml is described at @uref{http://pauillac.inria.fr/ocaml/}
and can be downloaded from from @uref{ftp://ftp.inria.fr/lang/camllight}.
@pindex genfft
@cindex Caml
If you have Objective Caml installed, you can type @code{sh
bootstrap.sh} in the toplevel directory to regenerate the files. If
you change the @code{gensrc/config} file, you can optimize FFTW for
sizes that are not currently supported efficiently (say, 17 or 19).
We do not provide more details about the codegeneration process, since
we do not expect that users will need to generate their own code.
However, feel free to contact us at @email{fftw@@fftw.org} if
you are interested in the subject.
@cindex monadic programming
You might find it interesting to learn Caml and/or some modern
programming techniques that we used in the generator (including monadic
programming), especially if you heard the rumor that Java and
objectoriented programming are the latest advancement in the field.
The internal operation of the codelet generator is described in the
paper, ``A Fast Fourier Transform Compiler,'' by M. Frigo, which is
available from the @uref{http://www.fftw.org,FFTW home page}
and will appear in the @cite{Proceedings of the 1999 ACM SIGPLAN
Conference on Programming Language Design and Implementation (PLDI)}.
@c ************************************************************
@node Acknowledgments, License and Copyright, Installation and Customization, Top
@chapter Acknowledgments
Matteo Frigo was supported in part by the Defense Advanced Research
Projects Agency (DARPA) under Grants N000149410985 and
F306029710270, and by a Digital Equipment Corporation Fellowship.
Steven G. Johnson was supported in part by a DoD NDSEG Fellowship, an
MIT Karl Taylor Compton Fellowship, and by the Materials Research
Science and Engineering Center program of the National Science
Foundation under award DMR9400334.
Both authors were also supported in part by their respective
girlfriends, by the letters ``Q'' and ``R'', and by the number 12.
@cindex girlfriends
We are grateful to SUN Microsystems Inc.@ for its donation of a cluster
of 9 8processor Ultra HPC 5000 SMPs (24 Gflops peak). These machines
served as the primary platform for the development of earlier versions
of FFTW.
We thank Intel Corporation for donating a fourprocessor Pentium Pro
machine. We thank the Linux community for giving us a decent OS to run
on that machine.
The @code{genfft} program was written using Objective Caml, a dialect of
ML. Objective Caml is a small and elegant language developed by Xavier
Leroy. The implementation is available from @code{ftp.inria.fr} in the
directory @code{lang/camllight}. We used versions 1.07 and 2.00 of the
software. In previous releases of FFTW, @code{genfft} was written in
Caml Light, by the same authors. An even earlier implementation of
@code{genfft} was written in Scheme, but Caml is definitely better for
this kind of application.
@pindex genfft
@cindex Caml
@cindex LISP
FFTW uses many tools from the GNU project, including @code{automake},
@code{texinfo}, and @code{libtool}.
Prof.@ Charles E.@ Leiserson of MIT provided continuous support and
encouragement. This program would not exist without him. Charles also
proposed the name ``codelets'' for the basic FFT blocks.
Prof.@ John D.@ Joannopoulos of MIT demonstrated continuing tolerance of
Steven's ``extracurricular'' computerscience activities. Steven's
chances at a physics degree would not exist without him.
Andrew Sterian contributed the Windows timing code.
Didier Miras reported a bug in the test procedure used in FFTW 1.2. We
now use a completely different test algorithm by Funda Ergun that does
not require a separate FFT program to compare against.
Wolfgang Reimer contributed the Pentium cycle counter and a few fixes
that help portability.
MingChang Liu uncovered a wellhidden bug in the complex transforms of
FFTW 2.0 and supplied a patch to correct it.
The FFTW FAQ was written in @code{bfnn} (Bizarre Format With No Name)
and formatted using the tools developed by Ian Jackson for the Linux
FAQ.
@emph{We are especially thankful to all of our users for their
continuing support, feedback, and interest during our development of
FFTW.}
@c ************************************************************
@node License and Copyright, Concept Index, Acknowledgments, Top
@chapter License and Copyright
FFTW is copyright @copyright{} 19971999 Massachusetts Institute of
Technology.
FFTW is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
59 Temple Place, Suite 330, Boston, MA 021111307 USA. You can also
find the @uref{http://www.gnu.org/copyleft/gpl.html, GPL on the GNU web
site}.
In addition, we kindly ask you to acknowledge FFTW and its authors in
any program or publication in which you use FFTW. (You are not
@emph{required} to do so; it is up to your common sense to decide
whether you want to comply with this request or not.)
Nonfree versions of FFTW are available under terms different than the
General Public License. (e.g. they do not require you to accompany any
object code using FFTW with the corresponding source code.) For these
alternate terms you must purchase a license from MIT's Technology
Licensing Office. Users interested in such a license should contact us
(@email{fftw@@fftw.org}) for more information.
@node Concept Index, Library Index, License and Copyright, Top
@chapter Concept Index
@printindex cp
@node Library Index, , Concept Index, Top
@chapter Library Index
@printindex fn
@c ************************************************************
@contents
@bye
@c LocalWords: texinfo fftw texi Exp setfilename settitle setchapternewpage
@c LocalWords: syncodeindex fn cp vr pg tp ifinfo titlepage sp Matteo Frigo
@c LocalWords: vskip pt filll dir detailmenu halfcomplex DFT fftwnd rfftw gcc
@c LocalWords: rfftwnd Pentium PentiumPro NJ FFT cindex ESSL FFTW's emph uref
@c LocalWords: http lcs mit edu benchfft benchFFT pindex dfn iftex tex cdot
@c LocalWords: ifhtml nbsp TR Sep Proc ICASSP Tukey Rader's ref xref int FFTs
@c LocalWords: findex tindex vindex im strided pxref DC lfftw lm samp const
@c LocalWords: nx ny nz ldots rk ik hc freq lrfftw datatypes cdots pointwise
@c LocalWords: pinv ij rote increaseth Ecclesiastes nerd stdout fopen printf
@c LocalWords: fclose dimension's malloc sizeof comp lang www eskimo com scs
@c LocalWords: faq html README Cilk SMP cilk POSIX Solaris BeOS MacOS mpi mcs
@c LocalWords: anl gov mutex THREADSAFE struct leq conj lt hbox istride lg rt
@c LocalWords: ostride subsubheading howmany idist odist exp IMG SRC gif IFFT
@c LocalWords: frftw dist datatype datatype's nc noindent callback putc getc
@c LocalWords: EOF Pentia PentiumPro's codelets CFLAGS cc Omax config org toc
@c LocalWords: pentium preconfigured egcs alloca fomit SEC gettimeofday MIN
@c LocalWords: picoseconds Caml gensrc genfft pauillac inria fr ocaml ftp sh
@c LocalWords: caml DoD NDSEG DMR HPC SMPs Gflops automake libtool Leiserson
@c LocalWords: Sterian Didier Miras Funda Ergun Reimer bfnn copyleft gpl
@c LocalWords: printindex LocalWords
