1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955 6956 6957 6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049 7050 7051 7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064 7065 7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095 7096 7097 7098 7099 7100 7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119 7120 7121 7122 7123 7124 7125 7126 7127 7128 7129 7130 7131 7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154 7155 7156 7157 7158 7159 7160 7161 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188 7189 7190 7191 7192 7193 7194 7195 7196 7197 7198 7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219 7220 7221 7222 7223 7224 7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238 7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253 7254 7255 7256 7257 7258 7259 7260 7261 7262 7263 7264 7265 7266 7267 7268 7269 7270 7271 7272 7273 7274 7275 7276 7277 7278 7279 7280 7281 7282 7283 7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294 7295 7296 7297 7298 7299 7300 7301 7302 7303 7304 7305 7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320 7321 7322 7323 7324 7325 7326 7327 7328 7329 7330 7331 7332 7333 7334 7335 7336 7337 7338 7339 7340 7341 7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363 7364 7365 7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384 7385 7386 7387 7388 7389 7390 7391 7392 7393 7394 7395 7396 7397 7398 7399 7400 7401 7402 7403 7404 7405 7406 7407 7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423 7424 7425 7426 7427 7428 7429 7430 7431 7432 7433 7434 7435 7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452 7453 7454 7455 7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469 7470 7471 7472 7473 7474 7475 7476 7477 7478 7479 7480 7481 7482 7483 7484 7485 7486 7487 7488 7489 7490 7491 7492 7493 7494 7495 7496 7497 7498 7499 7500 7501 7502 7503 7504 7505 7506 7507 7508 7509 7510 7511 7512 7513 7514 7515 7516 7517 7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534 7535 7536 7537 7538 7539 7540 7541 7542 7543 7544 7545 7546 7547 7548 7549 7550 7551 7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566 7567 7568 7569 7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580 7581 7582 7583 7584 7585 7586 7587 7588 7589 7590 7591 7592 7593 7594 7595 7596 7597 7598 7599 7600 7601 7602 7603 7604 7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655 7656 7657 7658 7659 7660 7661 7662 7663 7664 7665 7666 7667 7668 7669 7670 7671 7672 7673 7674 7675 7676 7677 7678 7679 7680 7681 7682 7683 7684 7685 7686 7687 7688 7689 7690 7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718 7719 7720 7721 7722 7723 7724 7725 7726 7727 7728 7729 7730 7731 7732 7733 7734 7735 7736 7737 7738 7739 7740 7741 7742 7743 7744 7745 7746 7747 7748 7749 7750 7751 7752 7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767 7768 7769 7770 7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800 7801 7802 7803 7804 7805 7806 7807 7808 7809 7810 7811 7812 7813 7814 7815 7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839 7840 7841 7842 7843 7844 7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897 7898 7899 7900 7901 7902 7903 7904 7905 7906 7907 7908 7909 7910 7911 7912 7913 7914 7915 7916 7917 7918 7919 7920 7921 7922 7923 7924 7925 7926 7927 7928 7929 7930 7931 7932 7933 7934 7935 7936 7937 7938 7939 7940 7941 7942 7943 7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960 7961 7962 7963 7964 7965 7966 7967 7968 7969 7970 7971 7972 7973 7974 7975 7976 7977 7978 7979 7980 7981 7982 7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018 8019 8020 8021 8022 8023 8024 8025 8026 8027 8028 8029 8030 8031 8032 8033 8034 8035 8036 8037 8038 8039 8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061 8062 8063 8064 8065 8066 8067 8068 8069 8070 8071 8072 8073 8074 8075 8076 8077 8078 8079 8080 8081 8082 8083 8084 8085 8086 8087 8088 8089 8090 8091 8092 8093 8094 8095 8096 8097 8098 8099 8100 8101 8102 8103 8104 8105 8106 8107 8108 8109 8110 8111 8112 8113 8114 8115 8116 8117 8118 8119 8120 8121 8122 8123 8124 8125 8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136 8137 8138 8139 8140 8141 8142 8143 8144 8145 8146 8147 8148 8149 8150 8151 8152 8153 8154 8155 8156 8157 8158 8159 8160 8161 8162 8163 8164 8165 8166 8167 8168 8169 8170 8171 8172 8173 8174 8175 8176 8177 8178 8179 8180 8181 8182 8183 8184 8185 8186 8187 8188 8189 8190 8191 8192 8193 8194 8195 8196 8197 8198 8199 8200 8201 8202 8203 8204 8205 8206 8207 8208 8209 8210 8211 8212 8213 8214 8215 8216 8217 8218 8219 8220 8221 8222 8223 8224 8225 8226 8227 8228 8229 8230 8231 8232 8233 8234 8235 8236 8237 8238 8239 8240 8241 8242 8243 8244 8245 8246 8247 8248 8249 8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272 8273 8274 8275 8276 8277 8278 8279 8280 8281 8282 8283 8284 8285 8286 8287 8288 8289 8290 8291 8292 8293 8294 8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315 8316 8317 8318 8319 8320 8321 8322 8323 8324 8325 8326 8327 8328 8329 8330 8331 8332 8333 8334 8335 8336 8337 8338 8339 8340 8341 8342 8343 8344 8345 8346 8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402 8403 8404 8405 8406 8407 8408 8409 8410 8411 8412 8413 8414 8415 8416 8417 8418 8419 8420 8421 8422 8423 8424 8425 8426 8427 8428 8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441 8442 8443 8444 8445 8446 8447 8448 8449 8450 8451 8452 8453 8454 8455 8456 8457 8458 8459 8460 8461 8462 8463 8464 8465 8466 8467 8468 8469 8470 8471 8472 8473 8474 8475 8476 8477 8478 8479 8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492 8493 8494 8495 8496 8497 8498 8499 8500 8501 8502 8503 8504 8505 8506 8507 8508 8509 8510 8511 8512 8513 8514 8515 8516 8517 8518 8519 8520 8521 8522 8523 8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534 8535 8536 8537 8538 8539 8540 8541 8542 8543 8544 8545 8546 8547 8548 8549 8550 8551 8552 8553 8554 8555 8556 8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567 8568 8569 8570 8571 8572 8573 8574 8575 8576 8577 8578 8579 8580 8581 8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592 8593 8594 8595 8596 8597 8598 8599 8600 8601 8602 8603 8604 8605 8606 8607 8608 8609 8610 8611 8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623 8624 8625 8626 8627 8628 8629 8630 8631 8632 8633 8634 8635 8636 8637 8638 8639 8640 8641 8642 8643 8644 8645 8646 8647 8648 8649 8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662 8663 8664 8665 8666 8667 8668 8669 8670 8671 8672 8673 8674 8675 8676 8677 8678 8679 8680 8681 8682 8683 8684 8685 8686 8687 8688 8689 8690 8691 8692 8693 8694 8695 8696 8697 8698 8699 8700 8701 8702 8703 8704 8705 8706 8707 8708 8709 8710 8711 8712 8713 8714 8715 8716 8717 8718 8719 8720 8721 8722 8723 8724 8725 8726 8727 8728 8729 8730 8731 8732 8733 8734 8735 8736 8737 8738 8739 8740 8741 8742 8743 8744 8745 8746 8747 8748 8749 8750 8751 8752 8753 8754 8755 8756 8757 8758 8759 8760 8761 8762 8763 8764
|
/* Instruction scheduling pass.
Copyright (C) 1992, 1993, 1994, 1995, 1997 Free Software Foundation, Inc.
Contributed by Michael Tiemann (tiemann@cygnus.com) Enhanced by,
and currently maintained by, Jim Wilson (wilson@cygnus.com)
This file is part of GNU CC.
GNU CC 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, or (at your option)
any later version.
GNU CC 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 GNU CC; see the file COPYING. If not, write to the Free
the Free Software Foundation, 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA. */
/* Instruction scheduling pass.
This pass implements list scheduling within basic blocks. It is
run twice: (1) after flow analysis, but before register allocation,
and (2) after register allocation.
The first run performs interblock scheduling, moving insns between
different blocks in the same "region", and the second runs only
basic block scheduling.
Interblock motions performed are useful motions and speculative
motions, including speculative loads. Motions requiring code
duplication are not supported. The identification of motion type
and the check for validity of speculative motions requires
construction and analysis of the function's control flow graph.
The scheduler works as follows:
We compute insn priorities based on data dependencies. Flow
analysis only creates a fraction of the data-dependencies we must
observe: namely, only those dependencies which the combiner can be
expected to use. For this pass, we must therefore create the
remaining dependencies we need to observe: register dependencies,
memory dependencies, dependencies to keep function calls in order,
and the dependence between a conditional branch and the setting of
condition codes are all dealt with here.
The scheduler first traverses the data flow graph, starting with
the last instruction, and proceeding to the first, assigning values
to insn_priority as it goes. This sorts the instructions
topologically by data dependence.
Once priorities have been established, we order the insns using
list scheduling. This works as follows: starting with a list of
all the ready insns, and sorted according to priority number, we
schedule the insn from the end of the list by placing its
predecessors in the list according to their priority order. We
consider this insn scheduled by setting the pointer to the "end" of
the list to point to the previous insn. When an insn has no
predecessors, we either queue it until sufficient time has elapsed
or add it to the ready list. As the instructions are scheduled or
when stalls are introduced, the queue advances and dumps insns into
the ready list. When all insns down to the lowest priority have
been scheduled, the critical path of the basic block has been made
as short as possible. The remaining insns are then scheduled in
remaining slots.
Function unit conflicts are resolved during forward list scheduling
by tracking the time when each insn is committed to the schedule
and from that, the time the function units it uses must be free.
As insns on the ready list are considered for scheduling, those
that would result in a blockage of the already committed insns are
queued until no blockage will result.
The following list shows the order in which we want to break ties
among insns in the ready list:
1. choose insn with the longest path to end of bb, ties
broken by
2. choose insn with least contribution to register pressure,
ties broken by
3. prefer in-block upon interblock motion, ties broken by
4. prefer useful upon speculative motion, ties broken by
5. choose insn with largest control flow probability, ties
broken by
6. choose insn with the least dependences upon the previously
scheduled insn, or finally
7. choose insn with lowest UID.
Memory references complicate matters. Only if we can be certain
that memory references are not part of the data dependency graph
(via true, anti, or output dependence), can we move operations past
memory references. To first approximation, reads can be done
independently, while writes introduce dependencies. Better
approximations will yield fewer dependencies.
Before reload, an extended analysis of interblock data dependences
is required for interblock scheduling. This is performed in
compute_block_backward_dependences ().
Dependencies set up by memory references are treated in exactly the
same way as other dependencies, by using LOG_LINKS backward
dependences. LOG_LINKS are translated into INSN_DEPEND forward
dependences for the purpose of forward list scheduling.
Having optimized the critical path, we may have also unduly
extended the lifetimes of some registers. If an operation requires
that constants be loaded into registers, it is certainly desirable
to load those constants as early as necessary, but no earlier.
I.e., it will not do to load up a bunch of registers at the
beginning of a basic block only to use them at the end, if they
could be loaded later, since this may result in excessive register
utilization.
Note that since branches are never in basic blocks, but only end
basic blocks, this pass will not move branches. But that is ok,
since we can use GNU's delayed branch scheduling pass to take care
of this case.
Also note that no further optimizations based on algebraic
identities are performed, so this pass would be a good one to
perform instruction splitting, such as breaking up a multiply
instruction into shifts and adds where that is profitable.
Given the memory aliasing analysis that this pass should perform,
it should be possible to remove redundant stores to memory, and to
load values from registers instead of hitting memory.
Before reload, speculative insns are moved only if a 'proof' exists
that no exception will be caused by this, and if no live registers
exist that inhibit the motion (live registers constraints are not
represented by data dependence edges).
This pass must update information that subsequent passes expect to
be correct. Namely: reg_n_refs, reg_n_sets, reg_n_deaths,
reg_n_calls_crossed, and reg_live_length. Also, basic_block_head,
basic_block_end.
The information in the line number notes is carefully retained by
this pass. Notes that refer to the starting and ending of
exception regions are also carefully retained by this pass. All
other NOTE insns are grouped in their same relative order at the
beginning of basic blocks and regions that have been scheduled.
The main entry point for this pass is schedule_insns(), called for
each function. The work of the scheduler is organized in three
levels: (1) function level: insns are subject to splitting,
control-flow-graph is constructed, regions are computed (after
reload, each region is of one block), (2) region level: control
flow graph attributes required for interblock scheduling are
computed (dominators, reachability, etc.), data dependences and
priorities are computed, and (3) block level: insns in the block
are actually scheduled. */
#include <stdio.h>
#include "config.h"
#include "rtl.h"
#include "basic-block.h"
#include "regs.h"
#include "hard-reg-set.h"
#include "flags.h"
#include "insn-config.h"
#include "insn-attr.h"
#include "except.h"
extern char *reg_known_equiv_p;
extern rtx *reg_known_value;
#ifdef INSN_SCHEDULING
/* enable interblock scheduling code */
/* define INTERBLOCK_DEBUG for using the -fsched-max debugging facility */
/* #define INTERBLOCK_DEBUG */
/* target_units bitmask has 1 for each unit in the cpu. It should be
possible to compute this variable from the machine description.
But currently it is computed by examinning the insn list. Since
this is only needed for visualization, it seems an acceptable
solution. (For understanding the mapping of bits to units, see
definition of function_units[] in "insn-attrtab.c") */
static int target_units = 0;
/* issue_rate is the number of insns that can be scheduled in the same
machine cycle. It can be defined in the config/mach/mach.h file,
otherwise we set it to 1. */
static int issue_rate;
#ifndef ISSUE_RATE
#define ISSUE_RATE 1
#endif
/* sched_debug_count is used for debugging the scheduler by limiting
the number of scheduled insns. It is controlled by the option
-fsched-max-N (N is a number).
sched-verbose controls the amount of debugging output the
scheduler prints. It is controlled by -fsched-verbose-N:
N>0 and no -DSR : the output is directed to stderr.
N>=10 will direct the printouts to stderr (regardless of -dSR).
N=1: same as -dSR.
N=2: bb's probabilities, detailed ready list info, unit/insn info.
N=3: rtl at abort point, control-flow, regions info.
N=5: dependences info.
max_rgn_blocks and max_region_insns limit region size for
interblock scheduling. They are controlled by
-fsched-interblock-max-blocks-N, -fsched-interblock-max-insns-N */
#define MAX_RGN_BLOCKS 10
#define MAX_RGN_INSNS 100
static int sched_debug_count = -1;
static int sched_verbose_param = 0;
static int sched_verbose = 0;
static int max_rgn_blocks = MAX_RGN_BLOCKS;
static int max_rgn_insns = MAX_RGN_INSNS;
/* nr_inter/spec counts interblock/speculative motion for the function */
static int nr_inter, nr_spec;
/* debugging file. all printouts are sent to dump, which is always set,
either to stderr, or to the dump listing file (-dRS). */
static FILE *dump = 0;
/* fix_sched_param() is called from toplev.c upon detection
of the -fsched-***-N options. */
void
fix_sched_param (param, val)
char *param, *val;
{
if (!strcmp (param, "max"))
sched_debug_count = ((sched_debug_count == -1) ?
atoi (val) : sched_debug_count);
else if (!strcmp (param, "verbose"))
sched_verbose_param = atoi (val);
else if (!strcmp (param, "interblock-max-blocks"))
max_rgn_blocks = atoi (val);
else if (!strcmp (param, "interblock-max-insns"))
max_rgn_insns = atoi (val);
else
warning ("fix_sched_param: unknown param: %s", param);
}
/* Arrays set up by scheduling for the same respective purposes as
similar-named arrays set up by flow analysis. We work with these
arrays during the scheduling pass so we can compare values against
unscheduled code.
Values of these arrays are copied at the end of this pass into the
arrays set up by flow analysis. */
static int *sched_reg_n_calls_crossed;
static int *sched_reg_live_length;
static int *sched_reg_basic_block;
/* We need to know the current block number during the post scheduling
update of live register information so that we can also update
REG_BASIC_BLOCK if a register changes blocks. */
static int current_block_num;
/* Element N is the next insn that sets (hard or pseudo) register
N within the current basic block; or zero, if there is no
such insn. Needed for new registers which may be introduced
by splitting insns. */
static rtx *reg_last_uses;
static rtx *reg_last_sets;
static regset reg_pending_sets;
static int reg_pending_sets_all;
/* Vector indexed by INSN_UID giving the original ordering of the insns. */
static int *insn_luid;
#define INSN_LUID(INSN) (insn_luid[INSN_UID (INSN)])
/* Vector indexed by INSN_UID giving each instruction a priority. */
static int *insn_priority;
#define INSN_PRIORITY(INSN) (insn_priority[INSN_UID (INSN)])
static short *insn_costs;
#define INSN_COST(INSN) insn_costs[INSN_UID (INSN)]
/* Vector indexed by INSN_UID giving an encoding of the function units
used. */
static short *insn_units;
#define INSN_UNIT(INSN) insn_units[INSN_UID (INSN)]
/* Vector indexed by INSN_UID giving each instruction a register-weight.
This weight is an estimation of the insn contribution to registers pressure. */
static int *insn_reg_weight;
#define INSN_REG_WEIGHT(INSN) (insn_reg_weight[INSN_UID (INSN)])
/* Vector indexed by INSN_UID giving list of insns which
depend upon INSN. Unlike LOG_LINKS, it represents forward dependences. */
static rtx *insn_depend;
#define INSN_DEPEND(INSN) insn_depend[INSN_UID (INSN)]
/* Vector indexed by INSN_UID. Initialized to the number of incoming
edges in forward dependence graph (= number of LOG_LINKS). As
scheduling procedes, dependence counts are decreased. An
instruction moves to the ready list when its counter is zero. */
static int *insn_dep_count;
#define INSN_DEP_COUNT(INSN) (insn_dep_count[INSN_UID (INSN)])
/* Vector indexed by INSN_UID giving an encoding of the blockage range
function. The unit and the range are encoded. */
static unsigned int *insn_blockage;
#define INSN_BLOCKAGE(INSN) insn_blockage[INSN_UID (INSN)]
#define UNIT_BITS 5
#define BLOCKAGE_MASK ((1 << BLOCKAGE_BITS) - 1)
#define ENCODE_BLOCKAGE(U, R) \
((((U) << UNIT_BITS) << BLOCKAGE_BITS \
| MIN_BLOCKAGE_COST (R)) << BLOCKAGE_BITS \
| MAX_BLOCKAGE_COST (R))
#define UNIT_BLOCKED(B) ((B) >> (2 * BLOCKAGE_BITS))
#define BLOCKAGE_RANGE(B) \
(((((B) >> BLOCKAGE_BITS) & BLOCKAGE_MASK) << (HOST_BITS_PER_INT / 2)) \
| (B) & BLOCKAGE_MASK)
/* Encodings of the `<name>_unit_blockage_range' function. */
#define MIN_BLOCKAGE_COST(R) ((R) >> (HOST_BITS_PER_INT / 2))
#define MAX_BLOCKAGE_COST(R) ((R) & ((1 << (HOST_BITS_PER_INT / 2)) - 1))
#define DONE_PRIORITY -1
#define MAX_PRIORITY 0x7fffffff
#define TAIL_PRIORITY 0x7ffffffe
#define LAUNCH_PRIORITY 0x7f000001
#define DONE_PRIORITY_P(INSN) (INSN_PRIORITY (INSN) < 0)
#define LOW_PRIORITY_P(INSN) ((INSN_PRIORITY (INSN) & 0x7f000000) == 0)
/* Vector indexed by INSN_UID giving number of insns referring to this insn. */
static int *insn_ref_count;
#define INSN_REF_COUNT(INSN) (insn_ref_count[INSN_UID (INSN)])
/* Vector indexed by INSN_UID giving line-number note in effect for each
insn. For line-number notes, this indicates whether the note may be
reused. */
static rtx *line_note;
#define LINE_NOTE(INSN) (line_note[INSN_UID (INSN)])
/* Vector indexed by basic block number giving the starting line-number
for each basic block. */
static rtx *line_note_head;
/* List of important notes we must keep around. This is a pointer to the
last element in the list. */
static rtx note_list;
/* Regsets telling whether a given register is live or dead before the last
scheduled insn. Must scan the instructions once before scheduling to
determine what registers are live or dead at the end of the block. */
static regset bb_live_regs;
/* Regset telling whether a given register is live after the insn currently
being scheduled. Before processing an insn, this is equal to bb_live_regs
above. This is used so that we can find registers that are newly born/dead
after processing an insn. */
static regset old_live_regs;
/* The chain of REG_DEAD notes. REG_DEAD notes are removed from all insns
during the initial scan and reused later. If there are not exactly as
many REG_DEAD notes in the post scheduled code as there were in the
prescheduled code then we trigger an abort because this indicates a bug. */
static rtx dead_notes;
/* Queues, etc. */
/* An instruction is ready to be scheduled when all insns preceding it
have already been scheduled. It is important to ensure that all
insns which use its result will not be executed until its result
has been computed. An insn is maintained in one of four structures:
(P) the "Pending" set of insns which cannot be scheduled until
their dependencies have been satisfied.
(Q) the "Queued" set of insns that can be scheduled when sufficient
time has passed.
(R) the "Ready" list of unscheduled, uncommitted insns.
(S) the "Scheduled" list of insns.
Initially, all insns are either "Pending" or "Ready" depending on
whether their dependencies are satisfied.
Insns move from the "Ready" list to the "Scheduled" list as they
are committed to the schedule. As this occurs, the insns in the
"Pending" list have their dependencies satisfied and move to either
the "Ready" list or the "Queued" set depending on whether
sufficient time has passed to make them ready. As time passes,
insns move from the "Queued" set to the "Ready" list. Insns may
move from the "Ready" list to the "Queued" set if they are blocked
due to a function unit conflict.
The "Pending" list (P) are the insns in the INSN_DEPEND of the unscheduled
insns, i.e., those that are ready, queued, and pending.
The "Queued" set (Q) is implemented by the variable `insn_queue'.
The "Ready" list (R) is implemented by the variables `ready' and
`n_ready'.
The "Scheduled" list (S) is the new insn chain built by this pass.
The transition (R->S) is implemented in the scheduling loop in
`schedule_block' when the best insn to schedule is chosen.
The transition (R->Q) is implemented in `queue_insn' when an
insn is found to to have a function unit conflict with the already
committed insns.
The transitions (P->R and P->Q) are implemented in `schedule_insn' as
insns move from the ready list to the scheduled list.
The transition (Q->R) is implemented in 'queue_to_insn' as time
passes or stalls are introduced. */
/* Implement a circular buffer to delay instructions until sufficient
time has passed. INSN_QUEUE_SIZE is a power of two larger than
MAX_BLOCKAGE and MAX_READY_COST computed by genattr.c. This is the
longest time an isnsn may be queued. */
static rtx insn_queue[INSN_QUEUE_SIZE];
static int q_ptr = 0;
static int q_size = 0;
#define NEXT_Q(X) (((X)+1) & (INSN_QUEUE_SIZE-1))
#define NEXT_Q_AFTER(X, C) (((X)+C) & (INSN_QUEUE_SIZE-1))
/* Vector indexed by INSN_UID giving the minimum clock tick at which
the insn becomes ready. This is used to note timing constraints for
insns in the pending list. */
static int *insn_tick;
#define INSN_TICK(INSN) (insn_tick[INSN_UID (INSN)])
/* Data structure for keeping track of register information
during that register's life. */
struct sometimes
{
int regno;
int live_length;
int calls_crossed;
};
/* Forward declarations. */
static void add_dependence PROTO ((rtx, rtx, enum reg_note));
static void remove_dependence PROTO ((rtx, rtx));
static rtx find_insn_list PROTO ((rtx, rtx));
static int insn_unit PROTO ((rtx));
static unsigned int blockage_range PROTO ((int, rtx));
static void clear_units PROTO ((void));
static int actual_hazard_this_instance PROTO ((int, int, rtx, int, int));
static void schedule_unit PROTO ((int, rtx, int));
static int actual_hazard PROTO ((int, rtx, int, int));
static int potential_hazard PROTO ((int, rtx, int));
static int insn_cost PROTO ((rtx, rtx, rtx));
static int priority PROTO ((rtx));
static void free_pending_lists PROTO ((void));
static void add_insn_mem_dependence PROTO ((rtx *, rtx *, rtx, rtx));
static void flush_pending_lists PROTO ((rtx, int));
static void sched_analyze_1 PROTO ((rtx, rtx));
static void sched_analyze_2 PROTO ((rtx, rtx));
static void sched_analyze_insn PROTO ((rtx, rtx, rtx));
static void sched_analyze PROTO ((rtx, rtx));
static void sched_note_set PROTO ((int, rtx, int));
static int rank_for_schedule PROTO ((rtx *, rtx *));
static void swap_sort PROTO ((rtx *, int));
static void queue_insn PROTO ((rtx, int));
static int schedule_insn PROTO ((rtx, rtx *, int, int));
static void create_reg_dead_note PROTO ((rtx, rtx));
static void attach_deaths PROTO ((rtx, rtx, int));
static void attach_deaths_insn PROTO ((rtx));
static int new_sometimes_live PROTO ((struct sometimes *, int, int));
static void finish_sometimes_live PROTO ((struct sometimes *, int));
static int schedule_block PROTO ((int, int, int));
static rtx regno_use_in PROTO ((int, rtx));
static void split_hard_reg_notes PROTO ((rtx, rtx, rtx, rtx));
static void new_insn_dead_notes PROTO ((rtx, rtx, rtx, rtx));
static void update_n_sets PROTO ((rtx, int));
static void update_flow_info PROTO ((rtx, rtx, rtx, rtx));
/* Main entry point of this file. */
void schedule_insns PROTO ((FILE *));
/* Mapping of insns to their original block prior to scheduling. */
static int *insn_orig_block;
#define INSN_BLOCK(insn) (insn_orig_block[INSN_UID (insn)])
/* Some insns (e.g. call) are not allowed to move across blocks. */
static char *cant_move;
#define CANT_MOVE(insn) (cant_move[INSN_UID (insn)])
/* Control flow graph edges are kept in circular lists. */
typedef struct
{
int from_block;
int to_block;
int next_in;
int next_out;
}
edge;
static edge *edge_table;
#define NEXT_IN(edge) (edge_table[edge].next_in)
#define NEXT_OUT(edge) (edge_table[edge].next_out)
#define FROM_BLOCK(edge) (edge_table[edge].from_block)
#define TO_BLOCK(edge) (edge_table[edge].to_block)
/* Number of edges in the control flow graph. (in fact larger than
that by 1, since edge 0 is unused.) */
static int nr_edges;
/* Circular list of incoming/outgoing edges of a block */
static int *in_edges;
static int *out_edges;
#define IN_EDGES(block) (in_edges[block])
#define OUT_EDGES(block) (out_edges[block])
/* List of labels which cannot be deleted, needed for control
flow graph construction. */
extern rtx forced_labels;
static char is_cfg_nonregular PROTO ((void));
static int uses_reg_or_mem PROTO ((rtx));
void debug_control_flow PROTO ((void));
static void build_control_flow PROTO ((void));
static void build_jmp_edges PROTO ((rtx, int));
static void new_edge PROTO ((int, int));
/* A region is the main entity for interblock scheduling: insns
are allowed to move between blocks in the same region, along
control flow graph edges, in the 'up' direction. */
typedef struct
{
int rgn_nr_blocks; /* number of blocks in region */
int rgn_blocks; /* blocks in the region (actually index in rgn_bb_table) */
}
region;
/* Number of regions in the procedure */
static int nr_regions;
/* Table of region descriptions */
static region *rgn_table;
/* Array of lists of regions' blocks */
static int *rgn_bb_table;
/* Topological order of blocks in the region (if b2 is reachable from
b1, block_to_bb[b2] > block_to_bb[b1]).
Note: A basic block is always referred to by either block or b,
while its topological order name (in the region) is refered to by
bb.
*/
static int *block_to_bb;
/* The number of the region containing a block. */
static int *containing_rgn;
#define RGN_NR_BLOCKS(rgn) (rgn_table[rgn].rgn_nr_blocks)
#define RGN_BLOCKS(rgn) (rgn_table[rgn].rgn_blocks)
#define BLOCK_TO_BB(block) (block_to_bb[block])
#define CONTAINING_RGN(block) (containing_rgn[block])
void debug_regions PROTO ((void));
static void find_single_block_region PROTO ((void));
static void find_rgns PROTO ((void));
static int too_large PROTO ((int, int *, int *));
extern void debug_live PROTO ((int, int));
/* Blocks of the current region being scheduled. */
static int current_nr_blocks;
static int current_blocks;
/* The mapping from bb to block */
#define BB_TO_BLOCK(bb) (rgn_bb_table[current_blocks + (bb)])
/* Bit vectors and bitset operations are needed for computations on
the control flow graph. */
typedef unsigned HOST_WIDE_INT *bitset;
typedef struct
{
int *first_member; /* pointer to the list start in bitlst_table. */
int nr_members; /* the number of members of the bit list. */
}
bitlst;
static int bitlst_table_last;
static int bitlst_table_size;
static int *bitlst_table;
static char bitset_member PROTO ((bitset, int, int));
static void extract_bitlst PROTO ((bitset, int, bitlst *));
/* target info declarations.
The block currently being scheduled is referred to as the "target" block,
while other blocks in the region from which insns can be moved to the
target are called "source" blocks. The candidate structure holds info
about such sources: are they valid? Speculative? Etc. */
typedef bitlst bblst;
typedef struct
{
char is_valid;
char is_speculative;
int src_prob;
bblst split_bbs;
bblst update_bbs;
}
candidate;
static candidate *candidate_table;
/* A speculative motion requires checking live information on the path
from 'source' to 'target'. The split blocks are those to be checked.
After a speculative motion, live information should be modified in
the 'update' blocks.
Lists of split and update blocks for each candidate of the current
target are in array bblst_table */
static int *bblst_table, bblst_size, bblst_last;
#define IS_VALID(src) ( candidate_table[src].is_valid )
#define IS_SPECULATIVE(src) ( candidate_table[src].is_speculative )
#define SRC_PROB(src) ( candidate_table[src].src_prob )
/* The bb being currently scheduled. */
static int target_bb;
/* List of edges. */
typedef bitlst edgelst;
/* target info functions */
static void split_edges PROTO ((int, int, edgelst *));
static void compute_trg_info PROTO ((int));
void debug_candidate PROTO ((int));
void debug_candidates PROTO ((int));
/* Bit-set of bbs, where bit 'i' stands for bb 'i'. */
typedef bitset bbset;
/* Number of words of the bbset. */
static int bbset_size;
/* Dominators array: dom[i] contains the bbset of dominators of
bb i in the region. */
static bbset *dom;
/* bb 0 is the only region entry */
#define IS_RGN_ENTRY(bb) (!bb)
/* Is bb_src dominated by bb_trg. */
#define IS_DOMINATED(bb_src, bb_trg) \
( bitset_member (dom[bb_src], bb_trg, bbset_size) )
/* Probability: Prob[i] is a float in [0, 1] which is the probability
of bb i relative to the region entry. */
static float *prob;
/* The probability of bb_src, relative to bb_trg. Note, that while the
'prob[bb]' is a float in [0, 1], this macro returns an integer
in [0, 100]. */
#define GET_SRC_PROB(bb_src, bb_trg) ((int) (100.0 * (prob[bb_src] / \
prob[bb_trg])))
/* Bit-set of edges, where bit i stands for edge i. */
typedef bitset edgeset;
/* Number of edges in the region. */
static int rgn_nr_edges;
/* Array of size rgn_nr_edges. */
static int *rgn_edges;
/* Number of words in an edgeset. */
static int edgeset_size;
/* Mapping from each edge in the graph to its number in the rgn. */
static int *edge_to_bit;
#define EDGE_TO_BIT(edge) (edge_to_bit[edge])
/* The split edges of a source bb is different for each target
bb. In order to compute this efficiently, the 'potential-split edges'
are computed for each bb prior to scheduling a region. This is actually
the split edges of each bb relative to the region entry.
pot_split[bb] is the set of potential split edges of bb. */
static edgeset *pot_split;
/* For every bb, a set of its ancestor edges. */
static edgeset *ancestor_edges;
static void compute_dom_prob_ps PROTO ((int));
#define ABS_VALUE(x) (((x)<0)?(-(x)):(x))
#define INSN_PROBABILITY(INSN) (SRC_PROB (BLOCK_TO_BB (INSN_BLOCK (INSN))))
#define IS_SPECULATIVE_INSN(INSN) (IS_SPECULATIVE (BLOCK_TO_BB (INSN_BLOCK (INSN))))
#define INSN_BB(INSN) (BLOCK_TO_BB (INSN_BLOCK (INSN)))
/* parameters affecting the decision of rank_for_schedule() */
#define MIN_DIFF_PRIORITY 2
#define MIN_PROBABILITY 40
#define MIN_PROB_DIFF 10
/* speculative scheduling functions */
static int check_live_1 PROTO ((int, rtx));
static void update_live_1 PROTO ((int, rtx));
static int check_live PROTO ((rtx, int, int));
static void update_live PROTO ((rtx, int, int));
static void set_spec_fed PROTO ((rtx));
static int is_pfree PROTO ((rtx, int, int));
static int find_conditional_protection PROTO ((rtx, int));
static int is_conditionally_protected PROTO ((rtx, int, int));
static int may_trap_exp PROTO ((rtx, int));
static int classify_insn PROTO ((rtx));
static int is_exception_free PROTO ((rtx, int, int));
static char find_insn_mem_list PROTO ((rtx, rtx, rtx, rtx));
static void compute_block_forward_dependences PROTO ((int));
static void init_rgn_data_dependences PROTO ((int));
static void add_branch_dependences PROTO ((rtx, rtx));
static void compute_block_backward_dependences PROTO ((int));
void debug_dependencies PROTO ((void));
/* Notes handling mechanism:
=========================
Generally, NOTES are saved before scheduling and restored after scheduling.
The scheduler distinguishes between three types of notes:
(1) LINE_NUMBER notes, generated and used for debugging. Here,
before scheduling a region, a pointer to the LINE_NUMBER note is
added to the insn following it (in save_line_notes()), and the note
is removed (in rm_line_notes() and unlink_line_notes()). After
scheduling the region, this pointer is used for regeneration of
the LINE_NUMBER note (in restore_line_notes()).
(2) LOOP_BEGIN, LOOP_END, SETJMP, EHREGION_BEG, EHREGION_END notes:
Before scheduling a region, a pointer to the note is added to the insn
that follows or precedes it. (This happens as part of the data dependence
computation). After scheduling an insn, the pointer contained in it is
used for regenerating the corresponding note (in reemit_notes).
(3) All other notes (e.g. INSN_DELETED): Before scheduling a block,
these notes are put in a list (in rm_other_notes() and
unlink_other_notes ()). After scheduling the block, these notes are
inserted at the beginning of the block (in schedule_block()). */
static rtx unlink_other_notes PROTO ((rtx, rtx));
static rtx unlink_line_notes PROTO ((rtx, rtx));
static void rm_line_notes PROTO ((int));
static void save_line_notes PROTO ((int));
static void restore_line_notes PROTO ((int));
static void rm_redundant_line_notes PROTO ((void));
static void rm_other_notes PROTO ((rtx, rtx));
static rtx reemit_notes PROTO ((rtx, rtx));
static void get_block_head_tail PROTO ((int, rtx *, rtx *));
static void find_pre_sched_live PROTO ((int));
static void find_post_sched_live PROTO ((int));
static void update_reg_usage PROTO ((void));
void debug_ready_list PROTO ((rtx[], int));
static void init_target_units PROTO (());
static void insn_print_units PROTO ((rtx));
static int get_visual_tbl_length PROTO (());
static void init_block_visualization PROTO (());
static void print_block_visualization PROTO ((int, char *));
static void visualize_scheduled_insns PROTO ((int, int));
static void visualize_no_unit PROTO ((rtx));
static void visualize_stall_cycles PROTO ((int, int));
static void print_exp PROTO ((char *, rtx, int));
static void print_value PROTO ((char *, rtx, int));
static void print_pattern PROTO ((char *, rtx, int));
static void print_insn PROTO ((char *, rtx, int));
void debug_reg_vector PROTO ((regset));
static rtx move_insn1 PROTO ((rtx, rtx));
static rtx move_insn PROTO ((rtx, rtx));
static rtx group_leader PROTO ((rtx));
static int set_priorities PROTO ((int));
static void init_rtx_vector PROTO ((rtx **, rtx *, int, int));
static void schedule_region PROTO ((int));
static void split_block_insns PROTO ((int));
#endif /* INSN_SCHEDULING */
#define SIZE_FOR_MODE(X) (GET_MODE_SIZE (GET_MODE (X)))
/* Helper functions for instruction scheduling. */
/* An INSN_LIST containing all INSN_LISTs allocated but currently unused. */
static rtx unused_insn_list;
/* An EXPR_LIST containing all EXPR_LISTs allocated but currently unused. */
static rtx unused_expr_list;
static void free_list PROTO ((rtx *, rtx *));
static rtx alloc_INSN_LIST PROTO ((rtx, rtx));
static rtx alloc_EXPR_LIST PROTO ((int, rtx, rtx));
static void
free_list (listp, unused_listp)
rtx *listp, *unused_listp;
{
register rtx link, prev_link;
if (*listp == 0)
return;
prev_link = *listp;
link = XEXP (prev_link, 1);
while (link)
{
prev_link = link;
link = XEXP (link, 1);
}
XEXP (prev_link, 1) = *unused_listp;
*unused_listp = *listp;
*listp = 0;
}
rtx
alloc_INSN_LIST (val, next)
rtx val, next;
{
rtx r;
if (unused_insn_list)
{
r = unused_insn_list;
unused_insn_list = XEXP (r, 1);
XEXP (r, 0) = val;
XEXP (r, 1) = next;
PUT_REG_NOTE_KIND (r, VOIDmode);
}
else
r = gen_rtx (INSN_LIST, VOIDmode, val, next);
return r;
}
rtx
alloc_EXPR_LIST (kind, val, next)
int kind;
rtx val, next;
{
rtx r;
if (unused_insn_list)
{
r = unused_insn_list;
unused_insn_list = XEXP (r, 1);
XEXP (r, 0) = val;
XEXP (r, 1) = next;
PUT_REG_NOTE_KIND (r, kind);
}
else
r = gen_rtx (EXPR_LIST, kind, val, next);
return r;
}
/* Add ELEM wrapped in an INSN_LIST with reg note kind DEP_TYPE to the
LOG_LINKS of INSN, if not already there. DEP_TYPE indicates the type
of dependence that this link represents. */
static void
add_dependence (insn, elem, dep_type)
rtx insn;
rtx elem;
enum reg_note dep_type;
{
rtx link, next;
/* Don't depend an insn on itself. */
if (insn == elem)
return;
/* If elem is part of a sequence that must be scheduled together, then
make the dependence point to the last insn of the sequence.
When HAVE_cc0, it is possible for NOTEs to exist between users and
setters of the condition codes, so we must skip past notes here.
Otherwise, NOTEs are impossible here. */
next = NEXT_INSN (elem);
#ifdef HAVE_cc0
while (next && GET_CODE (next) == NOTE)
next = NEXT_INSN (next);
#endif
if (next && SCHED_GROUP_P (next)
&& GET_CODE (next) != CODE_LABEL)
{
/* Notes will never intervene here though, so don't bother checking
for them. */
/* We must reject CODE_LABELs, so that we don't get confused by one
that has LABEL_PRESERVE_P set, which is represented by the same
bit in the rtl as SCHED_GROUP_P. A CODE_LABEL can never be
SCHED_GROUP_P. */
while (NEXT_INSN (next) && SCHED_GROUP_P (NEXT_INSN (next))
&& GET_CODE (NEXT_INSN (next)) != CODE_LABEL)
next = NEXT_INSN (next);
/* Again, don't depend an insn on itself. */
if (insn == next)
return;
/* Make the dependence to NEXT, the last insn of the group, instead
of the original ELEM. */
elem = next;
}
#ifdef INSN_SCHEDULING
/* (This code is guarded by INSN_SCHEDULING, otherwise INSN_BB is undefined.)
No need for interblock dependences with calls, since
calls are not moved between blocks. Note: the edge where
elem is a CALL is still required. */
if (GET_CODE (insn) == CALL_INSN
&& (INSN_BB (elem) != INSN_BB (insn)))
return;
#endif
/* Check that we don't already have this dependence. */
for (link = LOG_LINKS (insn); link; link = XEXP (link, 1))
if (XEXP (link, 0) == elem)
{
/* If this is a more restrictive type of dependence than the existing
one, then change the existing dependence to this type. */
if ((int) dep_type < (int) REG_NOTE_KIND (link))
PUT_REG_NOTE_KIND (link, dep_type);
return;
}
/* Might want to check one level of transitivity to save conses. */
link = alloc_INSN_LIST (elem, LOG_LINKS (insn));
LOG_LINKS (insn) = link;
/* Insn dependency, not data dependency. */
PUT_REG_NOTE_KIND (link, dep_type);
}
/* Remove ELEM wrapped in an INSN_LIST from the LOG_LINKS
of INSN. Abort if not found. */
static void
remove_dependence (insn, elem)
rtx insn;
rtx elem;
{
rtx prev, link, next;
int found = 0;
for (prev = 0, link = LOG_LINKS (insn); link; link = next)
{
next = XEXP (link, 1);
if (XEXP (link, 0) == elem)
{
if (prev)
XEXP (prev, 1) = next;
else
LOG_LINKS (insn) = next;
XEXP (link, 1) = unused_insn_list;
unused_insn_list = link;
found = 1;
}
else
prev = link;
}
if (!found)
abort ();
return;
}
#ifndef INSN_SCHEDULING
void
schedule_insns (dump_file)
FILE *dump_file;
{
}
#else
#ifndef __GNUC__
#define __inline
#endif
/* Computation of memory dependencies. */
/* The *_insns and *_mems are paired lists. Each pending memory operation
will have a pointer to the MEM rtx on one list and a pointer to the
containing insn on the other list in the same place in the list. */
/* We can't use add_dependence like the old code did, because a single insn
may have multiple memory accesses, and hence needs to be on the list
once for each memory access. Add_dependence won't let you add an insn
to a list more than once. */
/* An INSN_LIST containing all insns with pending read operations. */
static rtx pending_read_insns;
/* An EXPR_LIST containing all MEM rtx's which are pending reads. */
static rtx pending_read_mems;
/* An INSN_LIST containing all insns with pending write operations. */
static rtx pending_write_insns;
/* An EXPR_LIST containing all MEM rtx's which are pending writes. */
static rtx pending_write_mems;
/* Indicates the combined length of the two pending lists. We must prevent
these lists from ever growing too large since the number of dependencies
produced is at least O(N*N), and execution time is at least O(4*N*N), as
a function of the length of these pending lists. */
static int pending_lists_length;
/* The last insn upon which all memory references must depend.
This is an insn which flushed the pending lists, creating a dependency
between it and all previously pending memory references. This creates
a barrier (or a checkpoint) which no memory reference is allowed to cross.
This includes all non constant CALL_INSNs. When we do interprocedural
alias analysis, this restriction can be relaxed.
This may also be an INSN that writes memory if the pending lists grow
too large. */
static rtx last_pending_memory_flush;
/* The last function call we have seen. All hard regs, and, of course,
the last function call, must depend on this. */
static rtx last_function_call;
/* The LOG_LINKS field of this is a list of insns which use a pseudo register
that does not already cross a call. We create dependencies between each
of those insn and the next call insn, to ensure that they won't cross a call
after scheduling is done. */
static rtx sched_before_next_call;
/* Pointer to the last instruction scheduled. Used by rank_for_schedule,
so that insns independent of the last scheduled insn will be preferred
over dependent instructions. */
static rtx last_scheduled_insn;
/* Data structures for the computation of data dependences in a regions. We
keep one copy of each of the declared above variables for each bb in the
region. Before analyzing the data dependences for a bb, its variables
are initialized as a function of the variables of its predecessors. When
the analysis for a bb completes, we save the contents of each variable X
to a corresponding bb_X[bb] variable. For example, pending_read_insns is
copied to bb_pending_read_insns[bb]. Another change is that few
variables are now a list of insns rather than a single insn:
last_pending_memory_flash, last_function_call, reg_last_sets. The
manipulation of these variables was changed appropriately. */
static rtx **bb_reg_last_uses;
static rtx **bb_reg_last_sets;
static rtx *bb_pending_read_insns;
static rtx *bb_pending_read_mems;
static rtx *bb_pending_write_insns;
static rtx *bb_pending_write_mems;
static int *bb_pending_lists_length;
static rtx *bb_last_pending_memory_flush;
static rtx *bb_last_function_call;
static rtx *bb_sched_before_next_call;
/* functions for construction of the control flow graph. */
/* Return 1 if control flow graph should not be constructed, 0 otherwise.
Estimate in nr_edges the number of edges on the graph.
We decide not to build the control flow graph if there is possibly more
than one entry to the function, or if computed branches exist. */
static char
is_cfg_nonregular ()
{
int b;
rtx insn;
RTX_CODE code;
rtx nonlocal_label_list = nonlocal_label_rtx_list ();
/* check for non local labels */
if (nonlocal_label_list)
{
return 1;
}
/* check for labels which cannot be deleted */
if (forced_labels)
{
return 1;
}
/* check for labels which probably cannot be deleted */
if (exception_handler_labels)
{
return 1;
}
/* check for labels referred to other thn by jumps */
for (b = 0; b < n_basic_blocks; b++)
for (insn = basic_block_head[b];; insn = NEXT_INSN (insn))
{
code = GET_CODE (insn);
if (GET_RTX_CLASS (code) == 'i')
{
rtx note;
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
if (REG_NOTE_KIND (note) == REG_LABEL)
{
return 1;
}
}
if (insn == basic_block_end[b])
break;
}
nr_edges = 0;
/* check for computed branches */
for (b = 0; b < n_basic_blocks; b++)
{
for (insn = basic_block_head[b];; insn = NEXT_INSN (insn))
{
if (GET_CODE (insn) == JUMP_INSN)
{
rtx pat = PATTERN (insn);
int i;
if (GET_CODE (pat) == PARALLEL)
{
int len = XVECLEN (pat, 0);
int has_use_labelref = 0;
for (i = len - 1; i >= 0; i--)
if (GET_CODE (XVECEXP (pat, 0, i)) == USE
&& (GET_CODE (XEXP (XVECEXP (pat, 0, i), 0))
== LABEL_REF))
{
nr_edges++;
has_use_labelref = 1;
}
if (!has_use_labelref)
for (i = len - 1; i >= 0; i--)
if (GET_CODE (XVECEXP (pat, 0, i)) == SET
&& SET_DEST (XVECEXP (pat, 0, i)) == pc_rtx
&& uses_reg_or_mem (SET_SRC (XVECEXP (pat, 0, i))))
{
return 1;
}
}
/* check for branch table */
else if (GET_CODE (pat) == ADDR_VEC
|| GET_CODE (pat) == ADDR_DIFF_VEC)
{
int diff_vec_p = GET_CODE (pat) == ADDR_DIFF_VEC;
int len = XVECLEN (pat, diff_vec_p);
nr_edges += len;
}
else
{
/* check for computed branch */
if (GET_CODE (pat) == SET
&& SET_DEST (pat) == pc_rtx
&& uses_reg_or_mem (SET_SRC (pat)))
{
return 1;
}
}
}
if (insn == basic_block_end[b])
break;
}
}
/* count for the fallthrough edges */
for (b = 0; b < n_basic_blocks; b++)
{
for (insn = PREV_INSN (basic_block_head[b]);
insn && GET_CODE (insn) == NOTE; insn = PREV_INSN (insn))
;
if (!insn && b != 0)
nr_edges++;
else if (insn && GET_CODE (insn) != BARRIER)
nr_edges++;
}
nr_edges++;
return 0;
}
/* Returns 1 if x uses a reg or a mem (function was taken from flow.c).
x is a target of a jump. Used for the detection of computed
branches. For each label seen, updates the edges estimation
counter nr_edges. */
static int
uses_reg_or_mem (x)
rtx x;
{
enum rtx_code code = GET_CODE (x);
int i, j;
char *fmt;
if (code == REG)
return 1;
if (code == MEM
&& !(GET_CODE (XEXP (x, 0)) == SYMBOL_REF
&& CONSTANT_POOL_ADDRESS_P (XEXP (x, 0))))
return 1;
if (code == IF_THEN_ELSE)
{
if (uses_reg_or_mem (XEXP (x, 1))
|| uses_reg_or_mem (XEXP (x, 2)))
return 1;
else
return 0;
}
if (code == LABEL_REF)
{
nr_edges++;
return 0;
}
fmt = GET_RTX_FORMAT (code);
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
{
if (fmt[i] == 'e'
&& uses_reg_or_mem (XEXP (x, i)))
return 1;
if (fmt[i] == 'E')
for (j = 0; j < XVECLEN (x, i); j++)
if (uses_reg_or_mem (XVECEXP (x, i, j)))
return 1;
}
return 0;
}
/* Print the control flow graph, for debugging purposes.
Callable from the debugger. */
void
debug_control_flow ()
{
int i, e, next;
fprintf (dump, ";; --------- CONTROL FLOW GRAPH --------- \n\n");
for (i = 0; i < n_basic_blocks; i++)
{
fprintf (dump, ";;\tBasic block %d: first insn %d, last %d.\n",
i,
INSN_UID (basic_block_head[i]),
INSN_UID (basic_block_end[i]));
fprintf (dump, ";;\tPredecessor blocks:");
for (e = IN_EDGES (i); e; e = next)
{
fprintf (dump, " %d", FROM_BLOCK (e));
next = NEXT_IN (e);
if (next == IN_EDGES (i))
break;
}
fprintf (dump, "\n;;\tSuccesor blocks:");
for (e = OUT_EDGES (i); e; e = next)
{
fprintf (dump, " %d", TO_BLOCK (e));
next = NEXT_OUT (e);
if (next == OUT_EDGES (i))
break;
}
fprintf (dump, " \n\n");
}
}
/* build the control flow graph. (also set nr_edges accurately) */
static void
build_control_flow ()
{
int i;
nr_edges = 0;
for (i = 0; i < n_basic_blocks; i++)
{
rtx insn;
insn = basic_block_end[i];
if (GET_CODE (insn) == JUMP_INSN)
{
build_jmp_edges (PATTERN (insn), i);
}
for (insn = PREV_INSN (basic_block_head[i]);
insn && GET_CODE (insn) == NOTE; insn = PREV_INSN (insn))
;
/* build fallthrough edges */
if (!insn && i != 0)
new_edge (i - 1, i);
else if (insn && GET_CODE (insn) != BARRIER)
new_edge (i - 1, i);
}
/* increment by 1, since edge 0 is unused. */
nr_edges++;
}
/* construct edges in the control flow graph, from 'source' block, to
blocks refered to by 'pattern'. */
static
void
build_jmp_edges (pattern, source)
rtx pattern;
int source;
{
register RTX_CODE code;
register int i;
register char *fmt;
code = GET_CODE (pattern);
if (code == LABEL_REF)
{
register rtx label = XEXP (pattern, 0);
register int target;
/* This can happen as a result of a syntax error
and a diagnostic has already been printed. */
if (INSN_UID (label) == 0)
return;
target = INSN_BLOCK (label);
new_edge (source, target);
return;
}
/* proper handling of ADDR_DIFF_VEC: do not add a non-existing edge
from the block containing the branch-on-table, to itself. */
if (code == ADDR_VEC
|| code == ADDR_DIFF_VEC)
{
int diff_vec_p = GET_CODE (pattern) == ADDR_DIFF_VEC;
int len = XVECLEN (pattern, diff_vec_p);
int k;
for (k = 0; k < len; k++)
{
rtx tem = XVECEXP (pattern, diff_vec_p, k);
build_jmp_edges (tem, source);
}
}
fmt = GET_RTX_FORMAT (code);
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
{
if (fmt[i] == 'e')
build_jmp_edges (XEXP (pattern, i), source);
if (fmt[i] == 'E')
{
register int j;
for (j = 0; j < XVECLEN (pattern, i); j++)
build_jmp_edges (XVECEXP (pattern, i, j), source);
}
}
}
/* construct an edge in the control flow graph, from 'source' to 'target'. */
static void
new_edge (source, target)
int source, target;
{
int e, next_edge;
int curr_edge, fst_edge;
/* check for duplicates */
fst_edge = curr_edge = OUT_EDGES (source);
while (curr_edge)
{
if (FROM_BLOCK (curr_edge) == source
&& TO_BLOCK (curr_edge) == target)
{
return;
}
curr_edge = NEXT_OUT (curr_edge);
if (fst_edge == curr_edge)
break;
}
e = ++nr_edges;
FROM_BLOCK (e) = source;
TO_BLOCK (e) = target;
if (OUT_EDGES (source))
{
next_edge = NEXT_OUT (OUT_EDGES (source));
NEXT_OUT (OUT_EDGES (source)) = e;
NEXT_OUT (e) = next_edge;
}
else
{
OUT_EDGES (source) = e;
NEXT_OUT (e) = e;
}
if (IN_EDGES (target))
{
next_edge = NEXT_IN (IN_EDGES (target));
NEXT_IN (IN_EDGES (target)) = e;
NEXT_IN (e) = next_edge;
}
else
{
IN_EDGES (target) = e;
NEXT_IN (e) = e;
}
}
/* BITSET macros for operations on the control flow graph. */
/* Compute bitwise union of two bitsets. */
#define BITSET_UNION(set1, set2, len) \
do { register bitset tp = set1, sp = set2; \
register int i; \
for (i = 0; i < len; i++) \
*(tp++) |= *(sp++); } while (0)
/* Compute bitwise intersection of two bitsets. */
#define BITSET_INTER(set1, set2, len) \
do { register bitset tp = set1, sp = set2; \
register int i; \
for (i = 0; i < len; i++) \
*(tp++) &= *(sp++); } while (0)
/* Compute bitwise difference of two bitsets. */
#define BITSET_DIFFER(set1, set2, len) \
do { register bitset tp = set1, sp = set2; \
register int i; \
for (i = 0; i < len; i++) \
*(tp++) &= ~*(sp++); } while (0)
/* Inverts every bit of bitset 'set' */
#define BITSET_INVERT(set, len) \
do { register bitset tmpset = set; \
register int i; \
for (i = 0; i < len; i++, tmpset++) \
*tmpset = ~*tmpset; } while (0)
/* Turn on the index'th bit in bitset set. */
#define BITSET_ADD(set, index, len) \
{ \
if (index >= HOST_BITS_PER_WIDE_INT * len) \
abort (); \
else \
set[index/HOST_BITS_PER_WIDE_INT] |= \
1 << (index % HOST_BITS_PER_WIDE_INT); \
}
/* Turn off the index'th bit in set. */
#define BITSET_REMOVE(set, index, len) \
{ \
if (index >= HOST_BITS_PER_WIDE_INT * len) \
abort (); \
else \
set[index/HOST_BITS_PER_WIDE_INT] &= \
~(1 << (index%HOST_BITS_PER_WIDE_INT)); \
}
/* Check if the index'th bit in bitset set is on. */
static char
bitset_member (set, index, len)
bitset set;
int index, len;
{
if (index >= HOST_BITS_PER_WIDE_INT * len)
abort ();
return (set[index / HOST_BITS_PER_WIDE_INT] &
1 << (index % HOST_BITS_PER_WIDE_INT)) ? 1 : 0;
}
/* Translate a bit-set SET to a list BL of the bit-set members. */
static void
extract_bitlst (set, len, bl)
bitset set;
int len;
bitlst *bl;
{
int i, j, offset;
unsigned HOST_WIDE_INT word;
/* bblst table space is reused in each call to extract_bitlst */
bitlst_table_last = 0;
bl->first_member = &bitlst_table[bitlst_table_last];
bl->nr_members = 0;
for (i = 0; i < len; i++)
{
word = set[i];
offset = i * HOST_BITS_PER_WIDE_INT;
for (j = 0; word; j++)
{
if (word & 1)
{
bitlst_table[bitlst_table_last++] = offset;
(bl->nr_members)++;
}
word >>= 1;
++offset;
}
}
}
/* functions for the construction of regions */
/* Print the regions, for debugging purposes. Callable from debugger. */
void
debug_regions ()
{
int rgn, bb;
fprintf (dump, "\n;; ------------ REGIONS ----------\n\n");
for (rgn = 0; rgn < nr_regions; rgn++)
{
fprintf (dump, ";;\trgn %d nr_blocks %d:\n", rgn,
rgn_table[rgn].rgn_nr_blocks);
fprintf (dump, ";;\tbb/block: ");
for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
{
current_blocks = RGN_BLOCKS (rgn);
if (bb != BLOCK_TO_BB (BB_TO_BLOCK (bb)))
abort ();
fprintf (dump, " %d/%d ", bb, BB_TO_BLOCK (bb));
}
fprintf (dump, "\n\n");
}
}
/* Build a single block region for each basic block in the function.
This allows for using the same code for interblock and basic block
scheduling. */
static void
find_single_block_region ()
{
int i;
for (i = 0; i < n_basic_blocks; i++)
{
rgn_bb_table[i] = i;
RGN_NR_BLOCKS (i) = 1;
RGN_BLOCKS (i) = i;
CONTAINING_RGN (i) = i;
BLOCK_TO_BB (i) = 0;
}
nr_regions = n_basic_blocks;
}
/* Update number of blocks and the estimate for number of insns
in the region. Return 1 if the region is "too large" for interblock
scheduling (compile time considerations), otherwise return 0. */
static int
too_large (block, num_bbs, num_insns)
int block, *num_bbs, *num_insns;
{
(*num_bbs)++;
(*num_insns) += (INSN_LUID (basic_block_end[block]) -
INSN_LUID (basic_block_head[block]));
if ((*num_bbs > max_rgn_blocks) || (*num_insns > max_rgn_insns))
return 1;
else
return 0;
}
/* Update_loop_relations(blk, hdr): Check if the loop headed by max_hdr[blk]
is still an inner loop. Put in max_hdr[blk] the header of the most inner
loop containing blk. */
#define UPDATE_LOOP_RELATIONS(blk, hdr) \
{ \
if (max_hdr[blk] == -1) \
max_hdr[blk] = hdr; \
else if (dfs_nr[max_hdr[blk]] > dfs_nr[hdr]) \
inner[hdr] = 0; \
else if (dfs_nr[max_hdr[blk]] < dfs_nr[hdr]) \
{ \
inner[max_hdr[blk]] = 0; \
max_hdr[blk] = hdr; \
} \
}
/* Find regions for interblock scheduling: a loop-free procedure, a reducible
inner loop, or a basic block not contained in any other region.
The procedures control flow graph is traversed twice.
First traversal, a DFS, finds the headers of inner loops in the graph,
and verifies that there are no unreacable blocks.
Second traversal processes headers of inner loops, checking that the
loop is reducible. The loop blocks that form a region are put into the
region's blocks list in topological order.
The following variables are changed by the function: rgn_nr, rgn_table,
rgn_bb_table, block_to_bb and containing_rgn. */
static void
find_rgns ()
{
int *max_hdr, *dfs_nr, *stack, *queue, *degree;
char *header, *inner, *passed, *in_stack, *in_queue, no_loops = 1;
int node, child, loop_head, i, j, fst_edge, head, tail;
int count = 0, sp, idx = 0, current_edge = out_edges[0];
int num_bbs, num_insns;
int too_large_failure;
char *reachable;
/*
The following data structures are computed by the first traversal and
are used by the second traversal:
header[i] - flag set if the block i is the header of a loop.
inner[i] - initially set. It is reset if the the block i is the header
of a non-inner loop.
max_hdr[i] - the header of the inner loop containing block i.
(for a block i not in an inner loop it may be -1 or the
header of the most inner loop containing the block).
These data structures are used by the first traversal only:
stack - non-recursive DFS implementation which uses a stack of edges.
sp - top of the stack of edges
dfs_nr[i] - the DFS ordering of block i.
in_stack[i] - flag set if the block i is in the DFS stack.
These data structures are used by the second traversal only:
queue - queue containing the blocks of the current region.
head and tail - queue boundaries.
in_queue[i] - flag set if the block i is in queue */
/* function's inner arrays allocation and initialization */
max_hdr = (int *) alloca (n_basic_blocks * sizeof (int));
dfs_nr = (int *) alloca (n_basic_blocks * sizeof (int));
bzero ((char *) dfs_nr, n_basic_blocks * sizeof (int));
stack = (int *) alloca (nr_edges * sizeof (int));
queue = (int *) alloca (n_basic_blocks * sizeof (int));
inner = (char *) alloca (n_basic_blocks * sizeof (char));
header = (char *) alloca (n_basic_blocks * sizeof (char));
bzero ((char *) header, n_basic_blocks * sizeof (char));
passed = (char *) alloca (nr_edges * sizeof (char));
bzero ((char *) passed, nr_edges * sizeof (char));
in_stack = (char *) alloca (nr_edges * sizeof (char));
bzero ((char *) in_stack, nr_edges * sizeof (char));
reachable = (char *) alloca (n_basic_blocks * sizeof (char));
bzero ((char *) reachable, n_basic_blocks * sizeof (char));
in_queue = (char *) alloca (n_basic_blocks * sizeof (char));
for (i = 0; i < n_basic_blocks; i++)
{
inner[i] = 1;
max_hdr[i] = -1;
}
/* First traversal: DFS, finds inner loops in control flow graph */
reachable[0] = 1;
sp = -1;
while (1)
{
if (current_edge == 0 || passed[current_edge])
{
/* Here, if current_edge < 0, this is a leaf block.
Otherwise current_edge was already passed. Note that in
the latter case, not only current_edge but also all its
NEXT_OUT edges are also passed. We have to "climb up on
edges in the stack", looking for the first (already
passed) edge whose NEXT_OUT was not passed yet. */
while (sp >= 0 && (current_edge == 0 || passed[current_edge]))
{
current_edge = stack[sp--];
node = FROM_BLOCK (current_edge);
child = TO_BLOCK (current_edge);
in_stack[child] = 0;
if (max_hdr[child] >= 0 && in_stack[max_hdr[child]])
UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
current_edge = NEXT_OUT (current_edge);
}
/* stack empty - the whole graph is traversed. */
if (sp < 0 && passed[current_edge])
break;
continue;
}
node = FROM_BLOCK (current_edge);
dfs_nr[node] = ++count;
in_stack[node] = 1;
child = TO_BLOCK (current_edge);
reachable[child] = 1;
/* found a loop header */
if (in_stack[child])
{
no_loops = 0;
header[child] = 1;
max_hdr[child] = child;
UPDATE_LOOP_RELATIONS (node, child);
passed[current_edge] = 1;
current_edge = NEXT_OUT (current_edge);
continue;
}
/* the child was already visited once, no need to go down from
it, everything is traversed there. */
if (dfs_nr[child])
{
if (max_hdr[child] >= 0 && in_stack[max_hdr[child]])
UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
passed[current_edge] = 1;
current_edge = NEXT_OUT (current_edge);
continue;
}
/* this is a step down in the dfs traversal */
stack[++sp] = current_edge;
passed[current_edge] = 1;
current_edge = OUT_EDGES (child);
} /* while (1); */
/* if there are unreachable blocks, or more than one entry to
the subroutine, give up on interblock scheduling */
for (i = 1; i < n_basic_blocks; i++)
{
if (reachable[i] == 0)
{
find_single_block_region ();
if (sched_verbose >= 3)
fprintf (stderr, "sched: warning: found an unreachable block %d \n", i);
return;
}
}
/* Second travsersal: find reducible inner loops, and sort
topologically the blocks of each region */
degree = dfs_nr; /* reuse dfs_nr array - it is not needed anymore */
bzero ((char *) in_queue, n_basic_blocks * sizeof (char));
if (no_loops)
header[0] = 1;
/* compute the in-degree of every block in the graph */
for (i = 0; i < n_basic_blocks; i++)
{
fst_edge = IN_EDGES (i);
if (fst_edge > 0)
{
degree[i] = 1;
current_edge = NEXT_IN (fst_edge);
while (fst_edge != current_edge)
{
++degree[i];
current_edge = NEXT_IN (current_edge);
}
}
else
degree[i] = 0;
}
/* pass through all graph blocks, looking for headers of inner loops */
for (i = 0; i < n_basic_blocks; i++)
{
if (header[i] && inner[i])
{
/* i is a header of a potentially reducible inner loop, or
block 0 in a subroutine with no loops at all */
head = tail = -1;
too_large_failure = 0;
loop_head = max_hdr[i];
/* decrease in_degree of all i's successors, (this is needed
for the topological ordering) */
fst_edge = current_edge = OUT_EDGES (i);
if (fst_edge > 0)
{
do
{
--degree[TO_BLOCK (current_edge)];
current_edge = NEXT_OUT (current_edge);
}
while (fst_edge != current_edge);
}
/* estimate # insns, and count # blocks in the region. */
num_bbs = 1;
num_insns = INSN_LUID (basic_block_end[i]) - INSN_LUID (basic_block_head[i]);
/* find all loop latches, if it is a true loop header, or
all leaves if the graph has no loops at all */
if (no_loops)
{
for (j = 0; j < n_basic_blocks; j++)
if (out_edges[j] == 0) /* a leaf */
{
queue[++tail] = j;
in_queue[j] = 1;
if (too_large (j, &num_bbs, &num_insns))
{
too_large_failure = 1;
break;
}
}
}
else
{
fst_edge = current_edge = IN_EDGES (i);
do
{
node = FROM_BLOCK (current_edge);
if (max_hdr[node] == loop_head && node != i) /* a latch */
{
queue[++tail] = node;
in_queue[node] = 1;
if (too_large (node, &num_bbs, &num_insns))
{
too_large_failure = 1;
break;
}
}
current_edge = NEXT_IN (current_edge);
}
while (fst_edge != current_edge);
}
/* Put in queue[] all blocks that belong to the loop. Check
that the loop is reducible, traversing back from the loop
latches up to the loop header. */
while (head < tail && !too_large_failure)
{
child = queue[++head];
fst_edge = current_edge = IN_EDGES (child);
do
{
node = FROM_BLOCK (current_edge);
if (max_hdr[node] != loop_head)
{ /* another entry to loop, it is irreducible */
tail = -1;
break;
}
else if (!in_queue[node] && node != i)
{
queue[++tail] = node;
in_queue[node] = 1;
if (too_large (node, &num_bbs, &num_insns))
{
too_large_failure = 1;
break;
}
}
current_edge = NEXT_IN (current_edge);
}
while (fst_edge != current_edge);
}
if (tail >= 0 && !too_large_failure)
{
/* Place the loop header into list of region blocks */
degree[i] = -1;
rgn_bb_table[idx] = i;
RGN_NR_BLOCKS (nr_regions) = num_bbs;
RGN_BLOCKS (nr_regions) = idx++;
CONTAINING_RGN (i) = nr_regions;
BLOCK_TO_BB (i) = count = 0;
/* remove blocks from queue[], (in topological order), when
their in_degree becomes 0. We scan the queue over and
over again until it is empty. Note: there may be a more
efficient way to do it. */
while (tail >= 0)
{
if (head < 0)
head = tail;
child = queue[head];
if (degree[child] == 0)
{
degree[child] = -1;
rgn_bb_table[idx++] = child;
BLOCK_TO_BB (child) = ++count;
CONTAINING_RGN (child) = nr_regions;
queue[head] = queue[tail--];
fst_edge = current_edge = OUT_EDGES (child);
if (fst_edge > 0)
{
do
{
--degree[TO_BLOCK (current_edge)];
current_edge = NEXT_OUT (current_edge);
}
while (fst_edge != current_edge);
}
}
else
--head;
}
++nr_regions;
}
}
}
/* define each of all other blocks as a region itself */
for (i = 0; i < n_basic_blocks; i++)
if (degree[i] >= 0)
{
rgn_bb_table[idx] = i;
RGN_NR_BLOCKS (nr_regions) = 1;
RGN_BLOCKS (nr_regions) = idx++;
CONTAINING_RGN (i) = nr_regions++;
BLOCK_TO_BB (i) = 0;
}
} /* find_rgns */
/* functions for regions scheduling information */
/* Compute dominators, probability, and potential-split-edges of bb.
Assume that these values were already computed for bb's predecessors. */
static void
compute_dom_prob_ps (bb)
int bb;
{
int nxt_in_edge, fst_in_edge, pred;
int fst_out_edge, nxt_out_edge, nr_out_edges, nr_rgn_out_edges;
prob[bb] = 0.0;
if (IS_RGN_ENTRY (bb))
{
BITSET_ADD (dom[bb], 0, bbset_size);
prob[bb] = 1.0;
return;
}
fst_in_edge = nxt_in_edge = IN_EDGES (BB_TO_BLOCK (bb));
/* intialize dom[bb] to '111..1' */
BITSET_INVERT (dom[bb], bbset_size);
do
{
pred = FROM_BLOCK (nxt_in_edge);
BITSET_INTER (dom[bb], dom[BLOCK_TO_BB (pred)], bbset_size);
BITSET_UNION (ancestor_edges[bb], ancestor_edges[BLOCK_TO_BB (pred)],
edgeset_size);
BITSET_ADD (ancestor_edges[bb], EDGE_TO_BIT (nxt_in_edge), edgeset_size);
nr_out_edges = 1;
nr_rgn_out_edges = 0;
fst_out_edge = OUT_EDGES (pred);
nxt_out_edge = NEXT_OUT (fst_out_edge);
BITSET_UNION (pot_split[bb], pot_split[BLOCK_TO_BB (pred)],
edgeset_size);
BITSET_ADD (pot_split[bb], EDGE_TO_BIT (fst_out_edge), edgeset_size);
/* the successor doesn't belong the region? */
if (CONTAINING_RGN (TO_BLOCK (fst_out_edge)) !=
CONTAINING_RGN (BB_TO_BLOCK (bb)))
++nr_rgn_out_edges;
while (fst_out_edge != nxt_out_edge)
{
++nr_out_edges;
/* the successor doesn't belong the region? */
if (CONTAINING_RGN (TO_BLOCK (nxt_out_edge)) !=
CONTAINING_RGN (BB_TO_BLOCK (bb)))
++nr_rgn_out_edges;
BITSET_ADD (pot_split[bb], EDGE_TO_BIT (nxt_out_edge), edgeset_size);
nxt_out_edge = NEXT_OUT (nxt_out_edge);
}
/* now nr_rgn_out_edges is the number of region-exit edges from pred,
and nr_out_edges will be the number of pred out edges not leaving
the region. */
nr_out_edges -= nr_rgn_out_edges;
if (nr_rgn_out_edges > 0)
prob[bb] += 0.9 * prob[BLOCK_TO_BB (pred)] / nr_out_edges;
else
prob[bb] += prob[BLOCK_TO_BB (pred)] / nr_out_edges;
nxt_in_edge = NEXT_IN (nxt_in_edge);
}
while (fst_in_edge != nxt_in_edge);
BITSET_ADD (dom[bb], bb, bbset_size);
BITSET_DIFFER (pot_split[bb], ancestor_edges[bb], edgeset_size);
if (sched_verbose >= 2)
fprintf (dump, ";; bb_prob(%d, %d) = %3d\n", bb, BB_TO_BLOCK (bb), (int) (100.0 * prob[bb]));
} /* compute_dom_prob_ps */
/* functions for target info */
/* Compute in BL the list of split-edges of bb_src relatively to bb_trg.
Note that bb_trg dominates bb_src. */
static void
split_edges (bb_src, bb_trg, bl)
int bb_src;
int bb_trg;
edgelst *bl;
{
int es = edgeset_size;
edgeset src = (edgeset) alloca (es * sizeof (HOST_WIDE_INT));
while (es--)
src[es] = (pot_split[bb_src])[es];
BITSET_DIFFER (src, pot_split[bb_trg], edgeset_size);
extract_bitlst (src, edgeset_size, bl);
}
/* Find the valid candidate-source-blocks for the target block TRG, compute
their probability, and check if they are speculative or not.
For speculative sources, compute their update-blocks and split-blocks. */
static void
compute_trg_info (trg)
int trg;
{
register candidate *sp;
edgelst el;
int check_block, update_idx;
int i, j, k, fst_edge, nxt_edge;
/* define some of the fields for the target bb as well */
sp = candidate_table + trg;
sp->is_valid = 1;
sp->is_speculative = 0;
sp->src_prob = 100;
for (i = trg + 1; i < current_nr_blocks; i++)
{
sp = candidate_table + i;
sp->is_valid = IS_DOMINATED (i, trg);
if (sp->is_valid)
{
sp->src_prob = GET_SRC_PROB (i, trg);
sp->is_valid = (sp->src_prob >= MIN_PROBABILITY);
}
if (sp->is_valid)
{
split_edges (i, trg, &el);
sp->is_speculative = (el.nr_members) ? 1 : 0;
if (sp->is_speculative && !flag_schedule_speculative)
sp->is_valid = 0;
}
if (sp->is_valid)
{
sp->split_bbs.first_member = &bblst_table[bblst_last];
sp->split_bbs.nr_members = el.nr_members;
for (j = 0; j < el.nr_members; bblst_last++, j++)
bblst_table[bblst_last] =
TO_BLOCK (rgn_edges[el.first_member[j]]);
sp->update_bbs.first_member = &bblst_table[bblst_last];
update_idx = 0;
for (j = 0; j < el.nr_members; j++)
{
check_block = FROM_BLOCK (rgn_edges[el.first_member[j]]);
fst_edge = nxt_edge = OUT_EDGES (check_block);
do
{
for (k = 0; k < el.nr_members; k++)
if (EDGE_TO_BIT (nxt_edge) == el.first_member[k])
break;
if (k >= el.nr_members)
{
bblst_table[bblst_last++] = TO_BLOCK (nxt_edge);
update_idx++;
}
nxt_edge = NEXT_OUT (nxt_edge);
}
while (fst_edge != nxt_edge);
}
sp->update_bbs.nr_members = update_idx;
}
else
{
sp->split_bbs.nr_members = sp->update_bbs.nr_members = 0;
sp->is_speculative = 0;
sp->src_prob = 0;
}
}
} /* compute_trg_info */
/* Print candidates info, for debugging purposes. Callable from debugger. */
void
debug_candidate (i)
int i;
{
if (!candidate_table[i].is_valid)
return;
if (candidate_table[i].is_speculative)
{
int j;
fprintf (dump, "src b %d bb %d speculative \n", BB_TO_BLOCK (i), i);
fprintf (dump, "split path: ");
for (j = 0; j < candidate_table[i].split_bbs.nr_members; j++)
{
int b = candidate_table[i].split_bbs.first_member[j];
fprintf (dump, " %d ", b);
}
fprintf (dump, "\n");
fprintf (dump, "update path: ");
for (j = 0; j < candidate_table[i].update_bbs.nr_members; j++)
{
int b = candidate_table[i].update_bbs.first_member[j];
fprintf (dump, " %d ", b);
}
fprintf (dump, "\n");
}
else
{
fprintf (dump, " src %d equivalent\n", BB_TO_BLOCK (i));
}
}
/* Print candidates info, for debugging purposes. Callable from debugger. */
void
debug_candidates (trg)
int trg;
{
int i;
fprintf (dump, "----------- candidate table: target: b=%d bb=%d ---\n",
BB_TO_BLOCK (trg), trg);
for (i = trg + 1; i < current_nr_blocks; i++)
debug_candidate (i);
}
/* functions for speculative scheduing */
/* Return 0 if x is a set of a register alive in the beginning of one
of the split-blocks of src, otherwise return 1. */
static int
check_live_1 (src, x)
int src;
rtx x;
{
register i;
register int regno;
register rtx reg = SET_DEST (x);
if (reg == 0)
return 1;
while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
|| GET_CODE (reg) == SIGN_EXTRACT
|| GET_CODE (reg) == STRICT_LOW_PART)
reg = XEXP (reg, 0);
if (GET_CODE (reg) != REG)
return 1;
regno = REGNO (reg);
if (regno < FIRST_PSEUDO_REGISTER && global_regs[regno])
{
/* Global registers are assumed live */
return 0;
}
else
{
if (regno < FIRST_PSEUDO_REGISTER)
{
/* check for hard registers */
int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
while (--j >= 0)
{
for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
{
int b = candidate_table[src].split_bbs.first_member[i];
if (REGNO_REG_SET_P (basic_block_live_at_start[b], regno + j))
{
return 0;
}
}
}
}
else
{
/* check for psuedo registers */
for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
{
int b = candidate_table[src].split_bbs.first_member[i];
if (REGNO_REG_SET_P (basic_block_live_at_start[b], regno))
{
return 0;
}
}
}
}
return 1;
}
/* If x is a set of a register R, mark that R is alive in the beginning
of every update-block of src. */
static void
update_live_1 (src, x)
int src;
rtx x;
{
register i;
register int regno;
register rtx reg = SET_DEST (x);
if (reg == 0)
return;
while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
|| GET_CODE (reg) == SIGN_EXTRACT
|| GET_CODE (reg) == STRICT_LOW_PART)
reg = XEXP (reg, 0);
if (GET_CODE (reg) != REG)
return;
/* Global registers are always live, so the code below does not apply
to them. */
regno = REGNO (reg);
if (regno >= FIRST_PSEUDO_REGISTER || !global_regs[regno])
{
if (regno < FIRST_PSEUDO_REGISTER)
{
int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
while (--j >= 0)
{
for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
{
int b = candidate_table[src].update_bbs.first_member[i];
SET_REGNO_REG_SET (basic_block_live_at_start[b], regno + j);
}
}
}
else
{
for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
{
int b = candidate_table[src].update_bbs.first_member[i];
SET_REGNO_REG_SET (basic_block_live_at_start[b], regno);
}
}
}
}
/* Return 1 if insn can be speculatively moved from block src to trg,
otherwise return 0. Called before first insertion of insn to
ready-list or before the scheduling. */
static int
check_live (insn, src, trg)
rtx insn;
int src;
int trg;
{
/* find the registers set by instruction */
if (GET_CODE (PATTERN (insn)) == SET
|| GET_CODE (PATTERN (insn)) == CLOBBER)
return check_live_1 (src, PATTERN (insn));
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
{
int j;
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
if ((GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
&& !check_live_1 (src, XVECEXP (PATTERN (insn), 0, j)))
return 0;
return 1;
}
return 1;
}
/* Update the live registers info after insn was moved speculatively from
block src to trg. */
static void
update_live (insn, src, trg)
rtx insn;
int src, trg;
{
/* find the registers set by instruction */
if (GET_CODE (PATTERN (insn)) == SET
|| GET_CODE (PATTERN (insn)) == CLOBBER)
update_live_1 (src, PATTERN (insn));
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
{
int j;
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
update_live_1 (src, XVECEXP (PATTERN (insn), 0, j));
}
}
/* Exception Free Loads:
We define five classes of speculative loads: IFREE, IRISKY,
PFREE, PRISKY, and MFREE.
IFREE loads are loads that are proved to be exception-free, just
by examining the load insn. Examples for such loads are loads
from TOC and loads of global data.
IRISKY loads are loads that are proved to be exception-risky,
just by examining the load insn. Examples for such loads are
volatile loads and loads from shared memory.
PFREE loads are loads for which we can prove, by examining other
insns, that they are exception-free. Currently, this class consists
of loads for which we are able to find a "similar load", either in
the target block, or, if only one split-block exists, in that split
block. Load2 is similar to load1 if both have same single base
register. We identify only part of the similar loads, by finding
an insn upon which both load1 and load2 have a DEF-USE dependence.
PRISKY loads are loads for which we can prove, by examining other
insns, that they are exception-risky. Currently we have two proofs for
such loads. The first proof detects loads that are probably guarded by a
test on the memory address. This proof is based on the
backward and forward data dependence information for the region.
Let load-insn be the examined load.
Load-insn is PRISKY iff ALL the following hold:
- insn1 is not in the same block as load-insn
- there is a DEF-USE dependence chain (insn1, ..., load-insn)
- test-insn is either a compare or a branch, not in the same block as load-insn
- load-insn is reachable from test-insn
- there is a DEF-USE dependence chain (insn1, ..., test-insn)
This proof might fail when the compare and the load are fed
by an insn not in the region. To solve this, we will add to this
group all loads that have no input DEF-USE dependence.
The second proof detects loads that are directly or indirectly
fed by a speculative load. This proof is affected by the
scheduling process. We will use the flag fed_by_spec_load.
Initially, all insns have this flag reset. After a speculative
motion of an insn, if insn is either a load, or marked as
fed_by_spec_load, we will also mark as fed_by_spec_load every
insn1 for which a DEF-USE dependence (insn, insn1) exists. A
load which is fed_by_spec_load is also PRISKY.
MFREE (maybe-free) loads are all the remaining loads. They may be
exception-free, but we cannot prove it.
Now, all loads in IFREE and PFREE classes are considered
exception-free, while all loads in IRISKY and PRISKY classes are
considered exception-risky. As for loads in the MFREE class,
these are considered either exception-free or exception-risky,
depending on whether we are pessimistic or optimistic. We have
to take the pessimistic approach to assure the safety of
speculative scheduling, but we can take the optimistic approach
by invoking the -fsched_spec_load_dangerous option. */
enum INSN_TRAP_CLASS
{
TRAP_FREE = 0, IFREE = 1, PFREE_CANDIDATE = 2,
PRISKY_CANDIDATE = 3, IRISKY = 4, TRAP_RISKY = 5
};
#define WORST_CLASS(class1, class2) \
((class1 > class2) ? class1 : class2)
/* Indexed by INSN_UID, and set if there's DEF-USE dependence between */
/* some speculatively moved load insn and this one. */
char *fed_by_spec_load;
char *is_load_insn;
/* Non-zero if block bb_to is equal to, or reachable from block bb_from. */
#define IS_REACHABLE(bb_from, bb_to) \
(bb_from == bb_to \
|| IS_RGN_ENTRY (bb_from) \
|| (bitset_member (ancestor_edges[bb_to], \
EDGE_TO_BIT (IN_EDGES (BB_TO_BLOCK (bb_from))), \
edgeset_size)))
#define FED_BY_SPEC_LOAD(insn) (fed_by_spec_load[INSN_UID (insn)])
#define IS_LOAD_INSN(insn) (is_load_insn[INSN_UID (insn)])
/* Non-zero iff the address is comprised from at most 1 register */
#define CONST_BASED_ADDRESS_P(x) \
(GET_CODE (x) == REG \
|| ((GET_CODE (x) == PLUS || GET_CODE (x) == MINUS \
|| (GET_CODE (x) == LO_SUM)) \
&& (GET_CODE (XEXP (x, 0)) == CONST_INT \
|| GET_CODE (XEXP (x, 1)) == CONST_INT)))
/* Turns on the fed_by_spec_load flag for insns fed by load_insn. */
static void
set_spec_fed (load_insn)
rtx load_insn;
{
rtx link;
for (link = INSN_DEPEND (load_insn); link; link = XEXP (link, 1))
if (GET_MODE (link) == VOIDmode)
FED_BY_SPEC_LOAD (XEXP (link, 0)) = 1;
} /* set_spec_fed */
/* On the path from the insn to load_insn_bb, find a conditional branch */
/* depending on insn, that guards the speculative load. */
static int
find_conditional_protection (insn, load_insn_bb)
rtx insn;
int load_insn_bb;
{
rtx link;
/* iterate through DEF-USE forward dependences */
for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1))
{
rtx next = XEXP (link, 0);
if ((CONTAINING_RGN (INSN_BLOCK (next)) ==
CONTAINING_RGN (BB_TO_BLOCK (load_insn_bb)))
&& IS_REACHABLE (INSN_BB (next), load_insn_bb)
&& load_insn_bb != INSN_BB (next)
&& GET_MODE (link) == VOIDmode
&& (GET_CODE (next) == JUMP_INSN
|| find_conditional_protection (next, load_insn_bb)))
return 1;
}
return 0;
} /* find_conditional_protection */
/* Returns 1 if the same insn1 that participates in the computation
of load_insn's address is feeding a conditional branch that is
guarding on load_insn. This is true if we find a the two DEF-USE
chains:
insn1 -> ... -> conditional-branch
insn1 -> ... -> load_insn,
and if a flow path exist:
insn1 -> ... -> conditional-branch -> ... -> load_insn,
and if insn1 is on the path
region-entry -> ... -> bb_trg -> ... load_insn.
Locate insn1 by climbing on LOG_LINKS from load_insn.
Locate the branch by following INSN_DEPEND from insn1. */
static int
is_conditionally_protected (load_insn, bb_src, bb_trg)
rtx load_insn;
int bb_src, bb_trg;
{
rtx link;
for (link = LOG_LINKS (load_insn); link; link = XEXP (link, 1))
{
rtx insn1 = XEXP (link, 0);
/* must be a DEF-USE dependence upon non-branch */
if (GET_MODE (link) != VOIDmode
|| GET_CODE (insn1) == JUMP_INSN)
continue;
/* must exist a path: region-entry -> ... -> bb_trg -> ... load_insn */
if (INSN_BB (insn1) == bb_src
|| (CONTAINING_RGN (INSN_BLOCK (insn1))
!= CONTAINING_RGN (BB_TO_BLOCK (bb_src)))
|| (!IS_REACHABLE (bb_trg, INSN_BB (insn1))
&& !IS_REACHABLE (INSN_BB (insn1), bb_trg)))
continue;
/* now search for the conditional-branch */
if (find_conditional_protection (insn1, bb_src))
return 1;
/* recursive step: search another insn1, "above" current insn1. */
return is_conditionally_protected (insn1, bb_src, bb_trg);
}
/* the chain does not exsist */
return 0;
} /* is_conditionally_protected */
/* Returns 1 if a clue for "similar load" 'insn2' is found, and hence
load_insn can move speculatively from bb_src to bb_trg. All the
following must hold:
(1) both loads have 1 base register (PFREE_CANDIDATEs).
(2) load_insn and load1 have a def-use dependence upon
the same insn 'insn1'.
(3) either load2 is in bb_trg, or:
- there's only one split-block, and
- load1 is on the escape path, and
From all these we can conclude that the two loads access memory
addresses that differ at most by a constant, and hence if moving
load_insn would cause an exception, it would have been caused by
load2 anyhow. */
static int
is_pfree (load_insn, bb_src, bb_trg)
rtx load_insn;
int bb_src, bb_trg;
{
rtx back_link;
register candidate *candp = candidate_table + bb_src;
if (candp->split_bbs.nr_members != 1)
/* must have exactly one escape block */
return 0;
for (back_link = LOG_LINKS (load_insn);
back_link; back_link = XEXP (back_link, 1))
{
rtx insn1 = XEXP (back_link, 0);
if (GET_MODE (back_link) == VOIDmode)
{
/* found a DEF-USE dependence (insn1, load_insn) */
rtx fore_link;
for (fore_link = INSN_DEPEND (insn1);
fore_link; fore_link = XEXP (fore_link, 1))
{
rtx insn2 = XEXP (fore_link, 0);
if (GET_MODE (fore_link) == VOIDmode)
{
/* found a DEF-USE dependence (insn1, insn2) */
if (classify_insn (insn2) != PFREE_CANDIDATE)
/* insn2 not guaranteed to be a 1 base reg load */
continue;
if (INSN_BB (insn2) == bb_trg)
/* insn2 is the similar load, in the target block */
return 1;
if (*(candp->split_bbs.first_member) == INSN_BLOCK (insn2))
/* insn2 is a similar load, in a split-block */
return 1;
}
}
}
}
/* couldn't find a similar load */
return 0;
} /* is_pfree */
/* Returns a class that insn with GET_DEST(insn)=x may belong to,
as found by analyzing insn's expression. */
static int
may_trap_exp (x, is_store)
rtx x;
int is_store;
{
enum rtx_code code;
if (x == 0)
return TRAP_FREE;
code = GET_CODE (x);
if (is_store)
{
if (code == MEM)
return TRAP_RISKY;
else
return TRAP_FREE;
}
if (code == MEM)
{
/* The insn uses memory */
/* a volatile load */
if (MEM_VOLATILE_P (x))
return IRISKY;
/* an exception-free load */
if (!may_trap_p (x))
return IFREE;
/* a load with 1 base register, to be further checked */
if (CONST_BASED_ADDRESS_P (XEXP (x, 0)))
return PFREE_CANDIDATE;
/* no info on the load, to be further checked */
return PRISKY_CANDIDATE;
}
else
{
char *fmt;
int i, insn_class = TRAP_FREE;
/* neither store nor load, check if it may cause a trap */
if (may_trap_p (x))
return TRAP_RISKY;
/* recursive step: walk the insn... */
fmt = GET_RTX_FORMAT (code);
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
{
if (fmt[i] == 'e')
{
int tmp_class = may_trap_exp (XEXP (x, i), is_store);
insn_class = WORST_CLASS (insn_class, tmp_class);
}
else if (fmt[i] == 'E')
{
int j;
for (j = 0; j < XVECLEN (x, i); j++)
{
int tmp_class = may_trap_exp (XVECEXP (x, i, j), is_store);
insn_class = WORST_CLASS (insn_class, tmp_class);
if (insn_class == TRAP_RISKY || insn_class == IRISKY)
break;
}
}
if (insn_class == TRAP_RISKY || insn_class == IRISKY)
break;
}
return insn_class;
}
} /* may_trap_exp */
/* Classifies insn for the purpose of verifying that it can be
moved speculatively, by examining it's patterns, returning:
TRAP_RISKY: store, or risky non-load insn (e.g. division by variable).
TRAP_FREE: non-load insn.
IFREE: load from a globaly safe location.
IRISKY: volatile load.
PFREE_CANDIDATE, PRISKY_CANDIDATE: load that need to be checked for
being either PFREE or PRISKY. */
static int
classify_insn (insn)
rtx insn;
{
rtx pat = PATTERN (insn);
int tmp_class = TRAP_FREE;
int insn_class = TRAP_FREE;
enum rtx_code code;
if (GET_CODE (pat) == PARALLEL)
{
int i, len = XVECLEN (pat, 0);
for (i = len - 1; i >= 0; i--)
{
code = GET_CODE (XVECEXP (pat, 0, i));
switch (code)
{
case CLOBBER:
/* test if it is a 'store' */
tmp_class = may_trap_exp (XEXP (XVECEXP (pat, 0, i), 0), 1);
break;
case SET:
/* test if it is a store */
tmp_class = may_trap_exp (SET_DEST (XVECEXP (pat, 0, i)), 1);
if (tmp_class == TRAP_RISKY)
break;
/* test if it is a load */
tmp_class =
WORST_CLASS (tmp_class,
may_trap_exp (SET_SRC (XVECEXP (pat, 0, i)), 0));
default:;
}
insn_class = WORST_CLASS (insn_class, tmp_class);
if (insn_class == TRAP_RISKY || insn_class == IRISKY)
break;
}
}
else
{
code = GET_CODE (pat);
switch (code)
{
case CLOBBER:
/* test if it is a 'store' */
tmp_class = may_trap_exp (XEXP (pat, 0), 1);
break;
case SET:
/* test if it is a store */
tmp_class = may_trap_exp (SET_DEST (pat), 1);
if (tmp_class == TRAP_RISKY)
break;
/* test if it is a load */
tmp_class =
WORST_CLASS (tmp_class,
may_trap_exp (SET_SRC (pat), 0));
default:;
}
insn_class = tmp_class;
}
return insn_class;
} /* classify_insn */
/* Return 1 if load_insn is prisky (i.e. if load_insn is fed by
a load moved speculatively, or if load_insn is protected by
a compare on load_insn's address). */
static int
is_prisky (load_insn, bb_src, bb_trg)
rtx load_insn;
int bb_src, bb_trg;
{
if (FED_BY_SPEC_LOAD (load_insn))
return 1;
if (LOG_LINKS (load_insn) == NULL)
/* dependence may 'hide' out of the region. */
return 1;
if (is_conditionally_protected (load_insn, bb_src, bb_trg))
return 1;
return 0;
} /* is_prisky */
/* Insn is a candidate to be moved speculatively from bb_src to bb_trg.
Return 1 if insn is exception-free (and the motion is valid)
and 0 otherwise. */
static int
is_exception_free (insn, bb_src, bb_trg)
rtx insn;
int bb_src, bb_trg;
{
int insn_class = classify_insn (insn);
/* handle non-load insns */
switch (insn_class)
{
case TRAP_FREE:
return 1;
case TRAP_RISKY:
return 0;
default:;
}
/* handle loads */
if (!flag_schedule_speculative_load)
return 0;
IS_LOAD_INSN (insn) = 1;
switch (insn_class)
{
case IFREE:
return (1);
case IRISKY:
return 0;
case PFREE_CANDIDATE:
if (is_pfree (insn, bb_src, bb_trg))
return 1;
/* don't 'break' here: PFREE-candidate is also PRISKY-candidate */
case PRISKY_CANDIDATE:
if (!flag_schedule_speculative_load_dangerous
|| is_prisky (insn, bb_src, bb_trg))
return 0;
break;
default:;
}
return flag_schedule_speculative_load_dangerous;
} /* is_exception_free */
/* Process an insn's memory dependencies. There are four kinds of
dependencies:
(0) read dependence: read follows read
(1) true dependence: read follows write
(2) anti dependence: write follows read
(3) output dependence: write follows write
We are careful to build only dependencies which actually exist, and
use transitivity to avoid building too many links. */
/* Return the INSN_LIST containing INSN in LIST, or NULL
if LIST does not contain INSN. */
__inline static rtx
find_insn_list (insn, list)
rtx insn;
rtx list;
{
while (list)
{
if (XEXP (list, 0) == insn)
return list;
list = XEXP (list, 1);
}
return 0;
}
/* Return 1 if the pair (insn, x) is found in (LIST, LIST1), or 0 otherwise. */
__inline static char
find_insn_mem_list (insn, x, list, list1)
rtx insn, x;
rtx list, list1;
{
while (list)
{
if (XEXP (list, 0) == insn
&& XEXP (list1, 0) == x)
return 1;
list = XEXP (list, 1);
list1 = XEXP (list1, 1);
}
return 0;
}
/* Compute the function units used by INSN. This caches the value
returned by function_units_used. A function unit is encoded as the
unit number if the value is non-negative and the compliment of a
mask if the value is negative. A function unit index is the
non-negative encoding. */
__inline static int
insn_unit (insn)
rtx insn;
{
register int unit = INSN_UNIT (insn);
if (unit == 0)
{
recog_memoized (insn);
/* A USE insn, or something else we don't need to understand.
We can't pass these directly to function_units_used because it will
trigger a fatal error for unrecognizable insns. */
if (INSN_CODE (insn) < 0)
unit = -1;
else
{
unit = function_units_used (insn);
/* Increment non-negative values so we can cache zero. */
if (unit >= 0)
unit++;
}
/* We only cache 16 bits of the result, so if the value is out of
range, don't cache it. */
if (FUNCTION_UNITS_SIZE < HOST_BITS_PER_SHORT
|| unit >= 0
|| (~unit & ((1 << (HOST_BITS_PER_SHORT - 1)) - 1)) == 0)
INSN_UNIT (insn) = unit;
}
return (unit > 0 ? unit - 1 : unit);
}
/* Compute the blockage range for executing INSN on UNIT. This caches
the value returned by the blockage_range_function for the unit.
These values are encoded in an int where the upper half gives the
minimum value and the lower half gives the maximum value. */
__inline static unsigned int
blockage_range (unit, insn)
int unit;
rtx insn;
{
unsigned int blockage = INSN_BLOCKAGE (insn);
unsigned int range;
if (UNIT_BLOCKED (blockage) != unit + 1)
{
range = function_units[unit].blockage_range_function (insn);
/* We only cache the blockage range for one unit and then only if
the values fit. */
if (HOST_BITS_PER_INT >= UNIT_BITS + 2 * BLOCKAGE_BITS)
INSN_BLOCKAGE (insn) = ENCODE_BLOCKAGE (unit + 1, range);
}
else
range = BLOCKAGE_RANGE (blockage);
return range;
}
/* A vector indexed by function unit instance giving the last insn to use
the unit. The value of the function unit instance index for unit U
instance I is (U + I * FUNCTION_UNITS_SIZE). */
static rtx unit_last_insn[FUNCTION_UNITS_SIZE * MAX_MULTIPLICITY];
/* A vector indexed by function unit instance giving the minimum time when
the unit will unblock based on the maximum blockage cost. */
static int unit_tick[FUNCTION_UNITS_SIZE * MAX_MULTIPLICITY];
/* A vector indexed by function unit number giving the number of insns
that remain to use the unit. */
static int unit_n_insns[FUNCTION_UNITS_SIZE];
/* Reset the function unit state to the null state. */
static void
clear_units ()
{
bzero ((char *) unit_last_insn, sizeof (unit_last_insn));
bzero ((char *) unit_tick, sizeof (unit_tick));
bzero ((char *) unit_n_insns, sizeof (unit_n_insns));
}
/* Return the issue-delay of an insn */
__inline static int
insn_issue_delay (insn)
rtx insn;
{
rtx link;
int i, delay = 0;
int unit = insn_unit (insn);
/* efficiency note: in fact, we are working 'hard' to compute a
value that was available in md file, and is not available in
function_units[] structure. It would be nice to have this
value there, too. */
if (unit >= 0)
{
if (function_units[unit].blockage_range_function &&
function_units[unit].blockage_function)
delay = function_units[unit].blockage_function (insn, insn);
}
else
for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
if ((unit & 1) != 0 && function_units[i].blockage_range_function
&& function_units[i].blockage_function)
delay = MAX (delay, function_units[i].blockage_function (insn, insn));
return delay;
}
/* Return the actual hazard cost of executing INSN on the unit UNIT,
instance INSTANCE at time CLOCK if the previous actual hazard cost
was COST. */
__inline static int
actual_hazard_this_instance (unit, instance, insn, clock, cost)
int unit, instance, clock, cost;
rtx insn;
{
int tick = unit_tick[instance]; /* issue time of the last issued insn */
if (tick - clock > cost)
{
/* The scheduler is operating forward, so unit's last insn is the
executing insn and INSN is the candidate insn. We want a
more exact measure of the blockage if we execute INSN at CLOCK
given when we committed the execution of the unit's last insn.
The blockage value is given by either the unit's max blockage
constant, blockage range function, or blockage function. Use
the most exact form for the given unit. */
if (function_units[unit].blockage_range_function)
{
if (function_units[unit].blockage_function)
tick += (function_units[unit].blockage_function
(unit_last_insn[instance], insn)
- function_units[unit].max_blockage);
else
tick += ((int) MAX_BLOCKAGE_COST (blockage_range (unit, insn))
- function_units[unit].max_blockage);
}
if (tick - clock > cost)
cost = tick - clock;
}
return cost;
}
/* Record INSN as having begun execution on the units encoded by UNIT at
time CLOCK. */
__inline static void
schedule_unit (unit, insn, clock)
int unit, clock;
rtx insn;
{
int i;
if (unit >= 0)
{
int instance = unit;
#if MAX_MULTIPLICITY > 1
/* Find the first free instance of the function unit and use that
one. We assume that one is free. */
for (i = function_units[unit].multiplicity - 1; i > 0; i--)
{
if (!actual_hazard_this_instance (unit, instance, insn, clock, 0))
break;
instance += FUNCTION_UNITS_SIZE;
}
#endif
unit_last_insn[instance] = insn;
unit_tick[instance] = (clock + function_units[unit].max_blockage);
}
else
for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
if ((unit & 1) != 0)
schedule_unit (i, insn, clock);
}
/* Return the actual hazard cost of executing INSN on the units encoded by
UNIT at time CLOCK if the previous actual hazard cost was COST. */
__inline static int
actual_hazard (unit, insn, clock, cost)
int unit, clock, cost;
rtx insn;
{
int i;
if (unit >= 0)
{
/* Find the instance of the function unit with the minimum hazard. */
int instance = unit;
int best_cost = actual_hazard_this_instance (unit, instance, insn,
clock, cost);
int this_cost;
#if MAX_MULTIPLICITY > 1
if (best_cost > cost)
{
for (i = function_units[unit].multiplicity - 1; i > 0; i--)
{
instance += FUNCTION_UNITS_SIZE;
this_cost = actual_hazard_this_instance (unit, instance, insn,
clock, cost);
if (this_cost < best_cost)
{
best_cost = this_cost;
if (this_cost <= cost)
break;
}
}
}
#endif
cost = MAX (cost, best_cost);
}
else
for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
if ((unit & 1) != 0)
cost = actual_hazard (i, insn, clock, cost);
return cost;
}
/* Return the potential hazard cost of executing an instruction on the
units encoded by UNIT if the previous potential hazard cost was COST.
An insn with a large blockage time is chosen in preference to one
with a smaller time; an insn that uses a unit that is more likely
to be used is chosen in preference to one with a unit that is less
used. We are trying to minimize a subsequent actual hazard. */
__inline static int
potential_hazard (unit, insn, cost)
int unit, cost;
rtx insn;
{
int i, ncost;
unsigned int minb, maxb;
if (unit >= 0)
{
minb = maxb = function_units[unit].max_blockage;
if (maxb > 1)
{
if (function_units[unit].blockage_range_function)
{
maxb = minb = blockage_range (unit, insn);
maxb = MAX_BLOCKAGE_COST (maxb);
minb = MIN_BLOCKAGE_COST (minb);
}
if (maxb > 1)
{
/* Make the number of instructions left dominate. Make the
minimum delay dominate the maximum delay. If all these
are the same, use the unit number to add an arbitrary
ordering. Other terms can be added. */
ncost = minb * 0x40 + maxb;
ncost *= (unit_n_insns[unit] - 1) * 0x1000 + unit;
if (ncost > cost)
cost = ncost;
}
}
}
else
for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
if ((unit & 1) != 0)
cost = potential_hazard (i, insn, cost);
return cost;
}
/* Compute cost of executing INSN given the dependence LINK on the insn USED.
This is the number of cycles between instruction issue and
instruction results. */
__inline static int
insn_cost (insn, link, used)
rtx insn, link, used;
{
register int cost = INSN_COST (insn);
if (cost == 0)
{
recog_memoized (insn);
/* A USE insn, or something else we don't need to understand.
We can't pass these directly to result_ready_cost because it will
trigger a fatal error for unrecognizable insns. */
if (INSN_CODE (insn) < 0)
{
INSN_COST (insn) = 1;
return 1;
}
else
{
cost = result_ready_cost (insn);
if (cost < 1)
cost = 1;
INSN_COST (insn) = cost;
}
}
/* in this case estimate cost without caring how insn is used. */
if (link == 0 && used == 0)
return cost;
/* A USE insn should never require the value used to be computed. This
allows the computation of a function's result and parameter values to
overlap the return and call. */
recog_memoized (used);
if (INSN_CODE (used) < 0)
LINK_COST_FREE (link) = 1;
/* If some dependencies vary the cost, compute the adjustment. Most
commonly, the adjustment is complete: either the cost is ignored
(in the case of an output- or anti-dependence), or the cost is
unchanged. These values are cached in the link as LINK_COST_FREE
and LINK_COST_ZERO. */
if (LINK_COST_FREE (link))
cost = 1;
#ifdef ADJUST_COST
else if (!LINK_COST_ZERO (link))
{
int ncost = cost;
ADJUST_COST (used, link, insn, ncost);
if (ncost <= 1)
LINK_COST_FREE (link) = ncost = 1;
if (cost == ncost)
LINK_COST_ZERO (link) = 1;
cost = ncost;
}
#endif
return cost;
}
/* Compute the priority number for INSN. */
static int
priority (insn)
rtx insn;
{
int this_priority;
rtx link;
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
return 0;
if ((this_priority = INSN_PRIORITY (insn)) == 0)
{
if (INSN_DEPEND (insn) == 0)
this_priority = insn_cost (insn, 0, 0);
else
for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1))
{
rtx next;
int next_priority;
if (RTX_INTEGRATED_P (link))
continue;
next = XEXP (link, 0);
/* critical path is meaningful in block boundaries only */
if (INSN_BLOCK (next) != INSN_BLOCK (insn))
continue;
next_priority = insn_cost (insn, link, next) + priority (next);
if (next_priority > this_priority)
this_priority = next_priority;
}
INSN_PRIORITY (insn) = this_priority;
}
return this_priority;
}
/* Remove all INSN_LISTs and EXPR_LISTs from the pending lists and add
them to the unused_*_list variables, so that they can be reused. */
static void
free_pending_lists ()
{
if (current_nr_blocks <= 1)
{
free_list (&pending_read_insns, &unused_insn_list);
free_list (&pending_write_insns, &unused_insn_list);
free_list (&pending_read_mems, &unused_expr_list);
free_list (&pending_write_mems, &unused_expr_list);
}
else
{
/* interblock scheduling */
int bb;
for (bb = 0; bb < current_nr_blocks; bb++)
{
free_list (&bb_pending_read_insns[bb], &unused_insn_list);
free_list (&bb_pending_write_insns[bb], &unused_insn_list);
free_list (&bb_pending_read_mems[bb], &unused_expr_list);
free_list (&bb_pending_write_mems[bb], &unused_expr_list);
}
}
}
/* Add an INSN and MEM reference pair to a pending INSN_LIST and MEM_LIST.
The MEM is a memory reference contained within INSN, which we are saving
so that we can do memory aliasing on it. */
static void
add_insn_mem_dependence (insn_list, mem_list, insn, mem)
rtx *insn_list, *mem_list, insn, mem;
{
register rtx link;
link = alloc_INSN_LIST (insn, *insn_list);
*insn_list = link;
link = alloc_EXPR_LIST (VOIDmode, mem, *mem_list);
*mem_list = link;
pending_lists_length++;
}
/* Make a dependency between every memory reference on the pending lists
and INSN, thus flushing the pending lists. If ONLY_WRITE, don't flush
the read list. */
static void
flush_pending_lists (insn, only_write)
rtx insn;
int only_write;
{
rtx u;
rtx link;
while (pending_read_insns && ! only_write)
{
add_dependence (insn, XEXP (pending_read_insns, 0), REG_DEP_ANTI);
link = pending_read_insns;
pending_read_insns = XEXP (pending_read_insns, 1);
XEXP (link, 1) = unused_insn_list;
unused_insn_list = link;
link = pending_read_mems;
pending_read_mems = XEXP (pending_read_mems, 1);
XEXP (link, 1) = unused_expr_list;
unused_expr_list = link;
}
while (pending_write_insns)
{
add_dependence (insn, XEXP (pending_write_insns, 0), REG_DEP_ANTI);
link = pending_write_insns;
pending_write_insns = XEXP (pending_write_insns, 1);
XEXP (link, 1) = unused_insn_list;
unused_insn_list = link;
link = pending_write_mems;
pending_write_mems = XEXP (pending_write_mems, 1);
XEXP (link, 1) = unused_expr_list;
unused_expr_list = link;
}
pending_lists_length = 0;
/* last_pending_memory_flush is now a list of insns */
for (u = last_pending_memory_flush; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
free_list (&last_pending_memory_flush, &unused_insn_list);
last_pending_memory_flush = alloc_INSN_LIST (insn, NULL_RTX);
}
/* Analyze a single SET or CLOBBER rtx, X, creating all dependencies generated
by the write to the destination of X, and reads of everything mentioned. */
static void
sched_analyze_1 (x, insn)
rtx x;
rtx insn;
{
register int regno;
register rtx dest = SET_DEST (x);
if (dest == 0)
return;
while (GET_CODE (dest) == STRICT_LOW_PART || GET_CODE (dest) == SUBREG
|| GET_CODE (dest) == ZERO_EXTRACT || GET_CODE (dest) == SIGN_EXTRACT)
{
if (GET_CODE (dest) == ZERO_EXTRACT || GET_CODE (dest) == SIGN_EXTRACT)
{
/* The second and third arguments are values read by this insn. */
sched_analyze_2 (XEXP (dest, 1), insn);
sched_analyze_2 (XEXP (dest, 2), insn);
}
dest = SUBREG_REG (dest);
}
if (GET_CODE (dest) == REG)
{
register int i;
regno = REGNO (dest);
/* A hard reg in a wide mode may really be multiple registers.
If so, mark all of them just like the first. */
if (regno < FIRST_PSEUDO_REGISTER)
{
i = HARD_REGNO_NREGS (regno, GET_MODE (dest));
while (--i >= 0)
{
rtx u;
for (u = reg_last_uses[regno + i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
reg_last_uses[regno + i] = 0;
for (u = reg_last_sets[regno + i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_OUTPUT);
SET_REGNO_REG_SET (reg_pending_sets, regno + i);
if ((call_used_regs[regno + i] || global_regs[regno + i]))
/* Function calls clobber all call_used regs. */
for (u = last_function_call; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
}
}
else
{
rtx u;
for (u = reg_last_uses[regno]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
reg_last_uses[regno] = 0;
for (u = reg_last_sets[regno]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_OUTPUT);
SET_REGNO_REG_SET (reg_pending_sets, regno);
/* Pseudos that are REG_EQUIV to something may be replaced
by that during reloading. We need only add dependencies for
the address in the REG_EQUIV note. */
if (!reload_completed
&& reg_known_equiv_p[regno]
&& GET_CODE (reg_known_value[regno]) == MEM)
sched_analyze_2 (XEXP (reg_known_value[regno], 0), insn);
/* Don't let it cross a call after scheduling if it doesn't
already cross one. */
if (REG_N_CALLS_CROSSED (regno) == 0)
for (u = last_function_call; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
}
}
else if (GET_CODE (dest) == MEM)
{
/* Writing memory. */
if (pending_lists_length > 32)
{
/* Flush all pending reads and writes to prevent the pending lists
from getting any larger. Insn scheduling runs too slowly when
these lists get long. The number 32 was chosen because it
seems like a reasonable number. When compiling GCC with itself,
this flush occurs 8 times for sparc, and 10 times for m88k using
the number 32. */
flush_pending_lists (insn, 0);
}
else
{
rtx u;
rtx pending, pending_mem;
pending = pending_read_insns;
pending_mem = pending_read_mems;
while (pending)
{
/* If a dependency already exists, don't create a new one. */
if (!find_insn_list (XEXP (pending, 0), LOG_LINKS (insn)))
if (anti_dependence (XEXP (pending_mem, 0), dest))
add_dependence (insn, XEXP (pending, 0), REG_DEP_ANTI);
pending = XEXP (pending, 1);
pending_mem = XEXP (pending_mem, 1);
}
pending = pending_write_insns;
pending_mem = pending_write_mems;
while (pending)
{
/* If a dependency already exists, don't create a new one. */
if (!find_insn_list (XEXP (pending, 0), LOG_LINKS (insn)))
if (output_dependence (XEXP (pending_mem, 0), dest))
add_dependence (insn, XEXP (pending, 0), REG_DEP_OUTPUT);
pending = XEXP (pending, 1);
pending_mem = XEXP (pending_mem, 1);
}
for (u = last_pending_memory_flush; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
add_insn_mem_dependence (&pending_write_insns, &pending_write_mems,
insn, dest);
}
sched_analyze_2 (XEXP (dest, 0), insn);
}
/* Analyze reads. */
if (GET_CODE (x) == SET)
sched_analyze_2 (SET_SRC (x), insn);
}
/* Analyze the uses of memory and registers in rtx X in INSN. */
static void
sched_analyze_2 (x, insn)
rtx x;
rtx insn;
{
register int i;
register int j;
register enum rtx_code code;
register char *fmt;
if (x == 0)
return;
code = GET_CODE (x);
switch (code)
{
case CONST_INT:
case CONST_DOUBLE:
case SYMBOL_REF:
case CONST:
case LABEL_REF:
/* Ignore constants. Note that we must handle CONST_DOUBLE here
because it may have a cc0_rtx in its CONST_DOUBLE_CHAIN field, but
this does not mean that this insn is using cc0. */
return;
#ifdef HAVE_cc0
case CC0:
{
rtx link, prev;
/* User of CC0 depends on immediately preceding insn. */
SCHED_GROUP_P (insn) = 1;
/* There may be a note before this insn now, but all notes will
be removed before we actually try to schedule the insns, so
it won't cause a problem later. We must avoid it here though. */
prev = prev_nonnote_insn (insn);
/* Make a copy of all dependencies on the immediately previous insn,
and add to this insn. This is so that all the dependencies will
apply to the group. Remove an explicit dependence on this insn
as SCHED_GROUP_P now represents it. */
if (find_insn_list (prev, LOG_LINKS (insn)))
remove_dependence (insn, prev);
for (link = LOG_LINKS (prev); link; link = XEXP (link, 1))
add_dependence (insn, XEXP (link, 0), REG_NOTE_KIND (link));
return;
}
#endif
case REG:
{
rtx u;
int regno = REGNO (x);
if (regno < FIRST_PSEUDO_REGISTER)
{
int i;
i = HARD_REGNO_NREGS (regno, GET_MODE (x));
while (--i >= 0)
{
reg_last_uses[regno + i]
= alloc_INSN_LIST (insn, reg_last_uses[regno + i]);
for (u = reg_last_sets[regno + i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), 0);
if ((call_used_regs[regno + i] || global_regs[regno + i]))
/* Function calls clobber all call_used regs. */
for (u = last_function_call; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
}
}
else
{
reg_last_uses[regno]
= alloc_INSN_LIST (insn, reg_last_uses[regno]);
for (u = reg_last_sets[regno]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), 0);
/* Pseudos that are REG_EQUIV to something may be replaced
by that during reloading. We need only add dependencies for
the address in the REG_EQUIV note. */
if (!reload_completed
&& reg_known_equiv_p[regno]
&& GET_CODE (reg_known_value[regno]) == MEM)
sched_analyze_2 (XEXP (reg_known_value[regno], 0), insn);
/* If the register does not already cross any calls, then add this
insn to the sched_before_next_call list so that it will still
not cross calls after scheduling. */
if (REG_N_CALLS_CROSSED (regno) == 0)
add_dependence (sched_before_next_call, insn, REG_DEP_ANTI);
}
return;
}
case MEM:
{
/* Reading memory. */
rtx u;
rtx pending, pending_mem;
pending = pending_read_insns;
pending_mem = pending_read_mems;
while (pending)
{
/* If a dependency already exists, don't create a new one. */
if (!find_insn_list (XEXP (pending, 0), LOG_LINKS (insn)))
if (read_dependence (XEXP (pending_mem, 0), x))
add_dependence (insn, XEXP (pending, 0), REG_DEP_ANTI);
pending = XEXP (pending, 1);
pending_mem = XEXP (pending_mem, 1);
}
pending = pending_write_insns;
pending_mem = pending_write_mems;
while (pending)
{
/* If a dependency already exists, don't create a new one. */
if (!find_insn_list (XEXP (pending, 0), LOG_LINKS (insn)))
if (true_dependence (XEXP (pending_mem, 0), VOIDmode,
x, rtx_varies_p))
add_dependence (insn, XEXP (pending, 0), 0);
pending = XEXP (pending, 1);
pending_mem = XEXP (pending_mem, 1);
}
for (u = last_pending_memory_flush; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
/* Always add these dependencies to pending_reads, since
this insn may be followed by a write. */
add_insn_mem_dependence (&pending_read_insns, &pending_read_mems,
insn, x);
/* Take advantage of tail recursion here. */
sched_analyze_2 (XEXP (x, 0), insn);
return;
}
case ASM_OPERANDS:
case ASM_INPUT:
case UNSPEC_VOLATILE:
case TRAP_IF:
{
rtx u;
/* Traditional and volatile asm instructions must be considered to use
and clobber all hard registers, all pseudo-registers and all of
memory. So must TRAP_IF and UNSPEC_VOLATILE operations.
Consider for instance a volatile asm that changes the fpu rounding
mode. An insn should not be moved across this even if it only uses
pseudo-regs because it might give an incorrectly rounded result. */
if (code != ASM_OPERANDS || MEM_VOLATILE_P (x))
{
int max_reg = max_reg_num ();
for (i = 0; i < max_reg; i++)
{
for (u = reg_last_uses[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
reg_last_uses[i] = 0;
/* reg_last_sets[r] is now a list of insns */
for (u = reg_last_sets[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), 0);
}
reg_pending_sets_all = 1;
flush_pending_lists (insn, 0);
}
/* For all ASM_OPERANDS, we must traverse the vector of input operands.
We can not just fall through here since then we would be confused
by the ASM_INPUT rtx inside ASM_OPERANDS, which do not indicate
traditional asms unlike their normal usage. */
if (code == ASM_OPERANDS)
{
for (j = 0; j < ASM_OPERANDS_INPUT_LENGTH (x); j++)
sched_analyze_2 (ASM_OPERANDS_INPUT (x, j), insn);
return;
}
break;
}
case PRE_DEC:
case POST_DEC:
case PRE_INC:
case POST_INC:
/* These both read and modify the result. We must handle them as writes
to get proper dependencies for following instructions. We must handle
them as reads to get proper dependencies from this to previous
instructions. Thus we need to pass them to both sched_analyze_1
and sched_analyze_2. We must call sched_analyze_2 first in order
to get the proper antecedent for the read. */
sched_analyze_2 (XEXP (x, 0), insn);
sched_analyze_1 (x, insn);
return;
}
/* Other cases: walk the insn. */
fmt = GET_RTX_FORMAT (code);
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
{
if (fmt[i] == 'e')
sched_analyze_2 (XEXP (x, i), insn);
else if (fmt[i] == 'E')
for (j = 0; j < XVECLEN (x, i); j++)
sched_analyze_2 (XVECEXP (x, i, j), insn);
}
}
/* Analyze an INSN with pattern X to find all dependencies. */
static void
sched_analyze_insn (x, insn, loop_notes)
rtx x, insn;
rtx loop_notes;
{
register RTX_CODE code = GET_CODE (x);
rtx link;
int maxreg = max_reg_num ();
int i;
if (code == SET || code == CLOBBER)
sched_analyze_1 (x, insn);
else if (code == PARALLEL)
{
register int i;
for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
{
code = GET_CODE (XVECEXP (x, 0, i));
if (code == SET || code == CLOBBER)
sched_analyze_1 (XVECEXP (x, 0, i), insn);
else
sched_analyze_2 (XVECEXP (x, 0, i), insn);
}
}
else
sched_analyze_2 (x, insn);
/* Mark registers CLOBBERED or used by called function. */
if (GET_CODE (insn) == CALL_INSN)
for (link = CALL_INSN_FUNCTION_USAGE (insn); link; link = XEXP (link, 1))
{
if (GET_CODE (XEXP (link, 0)) == CLOBBER)
sched_analyze_1 (XEXP (link, 0), insn);
else
sched_analyze_2 (XEXP (link, 0), insn);
}
/* If there is a {LOOP,EHREGION}_{BEG,END} note in the middle of a basic block, then
we must be sure that no instructions are scheduled across it.
Otherwise, the reg_n_refs info (which depends on loop_depth) would
become incorrect. */
if (loop_notes)
{
int max_reg = max_reg_num ();
rtx link;
for (i = 0; i < max_reg; i++)
{
rtx u;
for (u = reg_last_uses[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
reg_last_uses[i] = 0;
/* reg_last_sets[r] is now a list of insns */
for (u = reg_last_sets[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), 0);
}
reg_pending_sets_all = 1;
flush_pending_lists (insn, 0);
link = loop_notes;
while (XEXP (link, 1))
link = XEXP (link, 1);
XEXP (link, 1) = REG_NOTES (insn);
REG_NOTES (insn) = loop_notes;
}
/* After reload, it is possible for an instruction to have a REG_DEAD note
for a register that actually dies a few instructions earlier. For
example, this can happen with SECONDARY_MEMORY_NEEDED reloads.
In this case, we must consider the insn to use the register mentioned
in the REG_DEAD note. Otherwise, we may accidentally move this insn
after another insn that sets the register, thus getting obviously invalid
rtl. This confuses reorg which believes that REG_DEAD notes are still
meaningful.
??? We would get better code if we fixed reload to put the REG_DEAD
notes in the right places, but that may not be worth the effort. */
if (reload_completed)
{
rtx note;
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
if (REG_NOTE_KIND (note) == REG_DEAD)
sched_analyze_2 (XEXP (note, 0), insn);
}
EXECUTE_IF_SET_IN_REG_SET (reg_pending_sets, 0, i,
{
/* reg_last_sets[r] is now a list of insns */
free_list (®_last_sets[i], &unused_insn_list);
reg_last_sets[i]
= alloc_INSN_LIST (insn, NULL_RTX);
});
CLEAR_REG_SET (reg_pending_sets);
if (reg_pending_sets_all)
{
for (i = 0; i < maxreg; i++)
{
/* reg_last_sets[r] is now a list of insns */
free_list (®_last_sets[i], &unused_insn_list);
reg_last_sets[i] = alloc_INSN_LIST (insn, NULL_RTX);
}
reg_pending_sets_all = 0;
}
/* Handle function calls and function returns created by the epilogue
threading code. */
if (GET_CODE (insn) == CALL_INSN || GET_CODE (insn) == JUMP_INSN)
{
rtx dep_insn;
rtx prev_dep_insn;
/* When scheduling instructions, we make sure calls don't lose their
accompanying USE insns by depending them one on another in order.
Also, we must do the same thing for returns created by the epilogue
threading code. Note this code works only in this special case,
because other passes make no guarantee that they will never emit
an instruction between a USE and a RETURN. There is such a guarantee
for USE instructions immediately before a call. */
prev_dep_insn = insn;
dep_insn = PREV_INSN (insn);
while (GET_CODE (dep_insn) == INSN
&& GET_CODE (PATTERN (dep_insn)) == USE
&& GET_CODE (XEXP (PATTERN (dep_insn), 0)) == REG)
{
SCHED_GROUP_P (prev_dep_insn) = 1;
/* Make a copy of all dependencies on dep_insn, and add to insn.
This is so that all of the dependencies will apply to the
group. */
for (link = LOG_LINKS (dep_insn); link; link = XEXP (link, 1))
add_dependence (insn, XEXP (link, 0), REG_NOTE_KIND (link));
prev_dep_insn = dep_insn;
dep_insn = PREV_INSN (dep_insn);
}
}
}
/* Analyze every insn between HEAD and TAIL inclusive, creating LOG_LINKS
for every dependency. */
static void
sched_analyze (head, tail)
rtx head, tail;
{
register rtx insn;
register rtx u;
rtx loop_notes = 0;
for (insn = head;; insn = NEXT_INSN (insn))
{
if (GET_CODE (insn) == INSN || GET_CODE (insn) == JUMP_INSN)
{
sched_analyze_insn (PATTERN (insn), insn, loop_notes);
loop_notes = 0;
}
else if (GET_CODE (insn) == CALL_INSN)
{
rtx x;
register int i;
CANT_MOVE (insn) = 1;
/* Any instruction using a hard register which may get clobbered
by a call needs to be marked as dependent on this call.
This prevents a use of a hard return reg from being moved
past a void call (i.e. it does not explicitly set the hard
return reg). */
/* If this call is followed by a NOTE_INSN_SETJMP, then assume that
all registers, not just hard registers, may be clobbered by this
call. */
/* Insn, being a CALL_INSN, magically depends on
`last_function_call' already. */
if (NEXT_INSN (insn) && GET_CODE (NEXT_INSN (insn)) == NOTE
&& NOTE_LINE_NUMBER (NEXT_INSN (insn)) == NOTE_INSN_SETJMP)
{
int max_reg = max_reg_num ();
for (i = 0; i < max_reg; i++)
{
for (u = reg_last_uses[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
reg_last_uses[i] = 0;
/* reg_last_sets[r] is now a list of insns */
for (u = reg_last_sets[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), 0);
}
reg_pending_sets_all = 1;
/* Add a pair of fake REG_NOTE which we will later
convert back into a NOTE_INSN_SETJMP note. See
reemit_notes for why we use a pair of NOTEs. */
REG_NOTES (insn) = alloc_EXPR_LIST (REG_DEAD,
GEN_INT (0),
REG_NOTES (insn));
REG_NOTES (insn) = alloc_EXPR_LIST (REG_DEAD,
GEN_INT (NOTE_INSN_SETJMP),
REG_NOTES (insn));
}
else
{
for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
if (call_used_regs[i] || global_regs[i])
{
for (u = reg_last_uses[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
reg_last_uses[i] = 0;
/* reg_last_sets[r] is now a list of insns */
for (u = reg_last_sets[i]; u; u = XEXP (u, 1))
add_dependence (insn, XEXP (u, 0), REG_DEP_ANTI);
SET_REGNO_REG_SET (reg_pending_sets, i);
}
}
/* For each insn which shouldn't cross a call, add a dependence
between that insn and this call insn. */
x = LOG_LINKS (sched_before_next_call);
while (x)
{
add_dependence (insn, XEXP (x, 0), REG_DEP_ANTI);
x = XEXP (x, 1);
}
LOG_LINKS (sched_before_next_call) = 0;
sched_analyze_insn (PATTERN (insn), insn, loop_notes);
loop_notes = 0;
/* In the absence of interprocedural alias analysis, we must flush
all pending reads and writes, and start new dependencies starting
from here. But only flush writes for constant calls (which may
be passed a pointer to something we haven't written yet). */
flush_pending_lists (insn, CONST_CALL_P (insn));
/* Depend this function call (actually, the user of this
function call) on all hard register clobberage. */
/* last_function_call is now a list of insns */
free_list(&last_function_call, &unused_insn_list);
last_function_call = alloc_INSN_LIST (insn, NULL_RTX);
}
/* See comments on reemit_notes as to why we do this. */
else if (GET_CODE (insn) == NOTE
&& (NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_BEG
|| NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_END
|| NOTE_LINE_NUMBER (insn) == NOTE_INSN_EH_REGION_BEG
|| NOTE_LINE_NUMBER (insn) == NOTE_INSN_EH_REGION_END
|| (NOTE_LINE_NUMBER (insn) == NOTE_INSN_SETJMP
&& GET_CODE (PREV_INSN (insn)) != CALL_INSN)))
{
loop_notes = alloc_EXPR_LIST (REG_DEAD,
GEN_INT (NOTE_BLOCK_NUMBER (insn)),
loop_notes);
loop_notes = alloc_EXPR_LIST (REG_DEAD,
GEN_INT (NOTE_LINE_NUMBER (insn)),
loop_notes);
CONST_CALL_P (loop_notes) = CONST_CALL_P (insn);
}
if (insn == tail)
return;
}
abort ();
}
/* Called when we see a set of a register. If death is true, then we are
scanning backwards. Mark that register as unborn. If nobody says
otherwise, that is how things will remain. If death is false, then we
are scanning forwards. Mark that register as being born. */
static void
sched_note_set (b, x, death)
int b;
rtx x;
int death;
{
register int regno;
register rtx reg = SET_DEST (x);
int subreg_p = 0;
if (reg == 0)
return;
while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == STRICT_LOW_PART
|| GET_CODE (reg) == SIGN_EXTRACT || GET_CODE (reg) == ZERO_EXTRACT)
{
/* Must treat modification of just one hardware register of a multi-reg
value or just a byte field of a register exactly the same way that
mark_set_1 in flow.c does, i.e. anything except a paradoxical subreg
does not kill the entire register. */
if (GET_CODE (reg) != SUBREG
|| REG_SIZE (SUBREG_REG (reg)) > REG_SIZE (reg))
subreg_p = 1;
reg = SUBREG_REG (reg);
}
if (GET_CODE (reg) != REG)
return;
/* Global registers are always live, so the code below does not apply
to them. */
regno = REGNO (reg);
if (regno >= FIRST_PSEUDO_REGISTER || !global_regs[regno])
{
if (death)
{
/* If we only set part of the register, then this set does not
kill it. */
if (subreg_p)
return;
/* Try killing this register. */
if (regno < FIRST_PSEUDO_REGISTER)
{
int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
while (--j >= 0)
{
CLEAR_REGNO_REG_SET (bb_live_regs, regno + j);
}
}
else
{
/* Recompute REG_BASIC_BLOCK as we update all the other
dataflow information. */
if (sched_reg_basic_block[regno] == REG_BLOCK_UNKNOWN)
sched_reg_basic_block[regno] = current_block_num;
else if (sched_reg_basic_block[regno] != current_block_num)
sched_reg_basic_block[regno] = REG_BLOCK_GLOBAL;
CLEAR_REGNO_REG_SET (bb_live_regs, regno);
}
}
else
{
/* Make the register live again. */
if (regno < FIRST_PSEUDO_REGISTER)
{
int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
while (--j >= 0)
{
SET_REGNO_REG_SET (bb_live_regs, regno + j);
}
}
else
{
SET_REGNO_REG_SET (bb_live_regs, regno);
}
}
}
}
/* Macros and functions for keeping the priority queue sorted, and
dealing with queueing and dequeueing of instructions. */
#define SCHED_SORT(READY, N_READY) \
do { if ((N_READY) == 2) \
swap_sort (READY, N_READY); \
else if ((N_READY) > 2) \
qsort (READY, N_READY, sizeof (rtx), rank_for_schedule); } \
while (0)
/* Returns a positive value if x is preferred; returns a negative value if
y is preferred. Should never return 0, since that will make the sort
unstable. */
static int
rank_for_schedule (x, y)
rtx *x, *y;
{
rtx tmp = *y;
rtx tmp2 = *x;
rtx link;
int tmp_class, tmp2_class;
int val, priority_val, spec_val, prob_val, weight_val;
/* schedule reverse is a stress test of the scheduler correctness,
controlled by -fsched-reverse option. */
if ((reload_completed && flag_schedule_reverse_after_reload) ||
(!reload_completed && flag_schedule_reverse_before_reload))
return INSN_LUID (tmp2) - INSN_LUID (tmp);
/* prefer insn with higher priority */
priority_val = INSN_PRIORITY (tmp2) - INSN_PRIORITY (tmp);
if (priority_val)
return priority_val;
/* prefer an insn with smaller contribution to registers-pressure */
if (!reload_completed &&
(weight_val = INSN_REG_WEIGHT (tmp) - INSN_REG_WEIGHT (tmp2)))
return (weight_val);
/* some comparison make sense in interblock scheduling only */
if (INSN_BB (tmp) != INSN_BB (tmp2))
{
/* prefer an inblock motion on an interblock motion */
if ((INSN_BB (tmp2) == target_bb) && (INSN_BB (tmp) != target_bb))
return 1;
if ((INSN_BB (tmp) == target_bb) && (INSN_BB (tmp2) != target_bb))
return -1;
/* prefer a useful motion on a speculative one */
if ((spec_val = IS_SPECULATIVE_INSN (tmp) - IS_SPECULATIVE_INSN (tmp2)))
return (spec_val);
/* prefer a more probable (speculative) insn */
prob_val = INSN_PROBABILITY (tmp2) - INSN_PROBABILITY (tmp);
if (prob_val)
return (prob_val);
}
/* compare insns based on their relation to the last-scheduled-insn */
if (last_scheduled_insn)
{
/* Classify the instructions into three classes:
1) Data dependent on last schedule insn.
2) Anti/Output dependent on last scheduled insn.
3) Independent of last scheduled insn, or has latency of one.
Choose the insn from the highest numbered class if different. */
link = find_insn_list (tmp, INSN_DEPEND (last_scheduled_insn));
if (link == 0 || insn_cost (last_scheduled_insn, link, tmp) == 1)
tmp_class = 3;
else if (REG_NOTE_KIND (link) == 0) /* Data dependence. */
tmp_class = 1;
else
tmp_class = 2;
link = find_insn_list (tmp2, INSN_DEPEND (last_scheduled_insn));
if (link == 0 || insn_cost (last_scheduled_insn, link, tmp2) == 1)
tmp2_class = 3;
else if (REG_NOTE_KIND (link) == 0) /* Data dependence. */
tmp2_class = 1;
else
tmp2_class = 2;
if ((val = tmp2_class - tmp_class))
return val;
}
/* If insns are equally good, sort by INSN_LUID (original insn order),
so that we make the sort stable. This minimizes instruction movement,
thus minimizing sched's effect on debugging and cross-jumping. */
return INSN_LUID (tmp) - INSN_LUID (tmp2);
}
/* Resort the array A in which only element at index N may be out of order. */
__inline static void
swap_sort (a, n)
rtx *a;
int n;
{
rtx insn = a[n - 1];
int i = n - 2;
while (i >= 0 && rank_for_schedule (a + i, &insn) >= 0)
{
a[i + 1] = a[i];
i -= 1;
}
a[i + 1] = insn;
}
static int max_priority;
/* Add INSN to the insn queue so that it can be executed at least
N_CYCLES after the currently executing insn. Preserve insns
chain for debugging purposes. */
__inline static void
queue_insn (insn, n_cycles)
rtx insn;
int n_cycles;
{
int next_q = NEXT_Q_AFTER (q_ptr, n_cycles);
rtx link = alloc_INSN_LIST (insn, insn_queue[next_q]);
insn_queue[next_q] = link;
q_size += 1;
if (sched_verbose >= 2)
{
fprintf (dump, ";;\t\tReady-->Q: insn %d: ", INSN_UID (insn));
if (INSN_BB (insn) != target_bb)
fprintf (dump, "(b%d) ", INSN_BLOCK (insn));
fprintf (dump, "queued for %d cycles.\n", n_cycles);
}
}
/* Return nonzero if PAT is the pattern of an insn which makes a
register live. */
__inline static int
birthing_insn_p (pat)
rtx pat;
{
int j;
if (reload_completed == 1)
return 0;
if (GET_CODE (pat) == SET
&& GET_CODE (SET_DEST (pat)) == REG)
{
rtx dest = SET_DEST (pat);
int i = REGNO (dest);
/* It would be more accurate to use refers_to_regno_p or
reg_mentioned_p to determine when the dest is not live before this
insn. */
if (REGNO_REG_SET_P (bb_live_regs, i))
return (REG_N_SETS (i) == 1);
return 0;
}
if (GET_CODE (pat) == PARALLEL)
{
for (j = 0; j < XVECLEN (pat, 0); j++)
if (birthing_insn_p (XVECEXP (pat, 0, j)))
return 1;
}
return 0;
}
/* PREV is an insn that is ready to execute. Adjust its priority if that
will help shorten register lifetimes. */
__inline static void
adjust_priority (prev)
rtx prev;
{
/* Trying to shorten register lives after reload has completed
is useless and wrong. It gives inaccurate schedules. */
if (reload_completed == 0)
{
rtx note;
int n_deaths = 0;
/* ??? This code has no effect, because REG_DEAD notes are removed
before we ever get here. */
for (note = REG_NOTES (prev); note; note = XEXP (note, 1))
if (REG_NOTE_KIND (note) == REG_DEAD)
n_deaths += 1;
/* Defer scheduling insns which kill registers, since that
shortens register lives. Prefer scheduling insns which
make registers live for the same reason. */
switch (n_deaths)
{
default:
INSN_PRIORITY (prev) >>= 3;
break;
case 3:
INSN_PRIORITY (prev) >>= 2;
break;
case 2:
case 1:
INSN_PRIORITY (prev) >>= 1;
break;
case 0:
if (birthing_insn_p (PATTERN (prev)))
{
int max = max_priority;
if (max > INSN_PRIORITY (prev))
INSN_PRIORITY (prev) = max;
}
break;
}
#ifdef ADJUST_PRIORITY
ADJUST_PRIORITY (prev);
#endif
}
}
/* INSN is the "currently executing insn". Launch each insn which was
waiting on INSN. READY is a vector of insns which are ready to fire.
N_READY is the number of elements in READY. CLOCK is the current
cycle. */
static int
schedule_insn (insn, ready, n_ready, clock)
rtx insn;
rtx *ready;
int n_ready;
int clock;
{
rtx link;
int unit;
unit = insn_unit (insn);
if (sched_verbose >= 2)
{
fprintf (dump, ";;\t\t--> scheduling insn <<<%d>>> on unit ", INSN_UID (insn));
insn_print_units (insn);
fprintf (dump, "\n");
}
if (sched_verbose && unit == -1)
visualize_no_unit (insn);
if (MAX_BLOCKAGE > 1 || issue_rate > 1 || sched_verbose)
schedule_unit (unit, insn, clock);
if (INSN_DEPEND (insn) == 0)
return n_ready;
/* This is used by the function adjust_priority above. */
if (n_ready > 0)
max_priority = MAX (INSN_PRIORITY (ready[0]), INSN_PRIORITY (insn));
else
max_priority = INSN_PRIORITY (insn);
for (link = INSN_DEPEND (insn); link != 0; link = XEXP (link, 1))
{
rtx next = XEXP (link, 0);
int cost = insn_cost (insn, link, next);
INSN_TICK (next) = MAX (INSN_TICK (next), clock + cost);
if ((INSN_DEP_COUNT (next) -= 1) == 0)
{
int effective_cost = INSN_TICK (next) - clock;
/* For speculative insns, before inserting to ready/queue,
check live, exception-free, and issue-delay */
if (INSN_BB (next) != target_bb
&& (!IS_VALID (INSN_BB (next))
|| CANT_MOVE (next)
|| (IS_SPECULATIVE_INSN (next)
&& (insn_issue_delay (next) > 3
|| !check_live (next, INSN_BB (next), target_bb)
|| !is_exception_free (next, INSN_BB (next), target_bb)))))
continue;
if (sched_verbose >= 2)
{
fprintf (dump, ";;\t\tdependences resolved: insn %d ", INSN_UID (next));
if (current_nr_blocks > 1 && INSN_BB (next) != target_bb)
fprintf (dump, "/b%d ", INSN_BLOCK (next));
if (effective_cost <= 1)
fprintf (dump, "into ready\n");
else
fprintf (dump, "into queue with cost=%d\n", effective_cost);
}
/* Adjust the priority of NEXT and either put it on the ready
list or queue it. */
adjust_priority (next);
if (effective_cost <= 1)
ready[n_ready++] = next;
else
queue_insn (next, effective_cost);
}
}
return n_ready;
}
/* Add a REG_DEAD note for REG to INSN, reusing a REG_DEAD note from the
dead_notes list. */
static void
create_reg_dead_note (reg, insn)
rtx reg, insn;
{
rtx link;
/* The number of registers killed after scheduling must be the same as the
number of registers killed before scheduling. The number of REG_DEAD
notes may not be conserved, i.e. two SImode hard register REG_DEAD notes
might become one DImode hard register REG_DEAD note, but the number of
registers killed will be conserved.
We carefully remove REG_DEAD notes from the dead_notes list, so that
there will be none left at the end. If we run out early, then there
is a bug somewhere in flow, combine and/or sched. */
if (dead_notes == 0)
{
if (current_nr_blocks <= 1)
abort ();
else
link = alloc_EXPR_LIST (REG_DEAD, NULL_RTX, NULL_RTX);
}
else
{
/* Number of regs killed by REG. */
int regs_killed = (REGNO (reg) >= FIRST_PSEUDO_REGISTER ? 1
: HARD_REGNO_NREGS (REGNO (reg), GET_MODE (reg)));
/* Number of regs killed by REG_DEAD notes taken off the list. */
int reg_note_regs;
link = dead_notes;
reg_note_regs = (REGNO (XEXP (link, 0)) >= FIRST_PSEUDO_REGISTER ? 1
: HARD_REGNO_NREGS (REGNO (XEXP (link, 0)),
GET_MODE (XEXP (link, 0))));
while (reg_note_regs < regs_killed)
{
link = XEXP (link, 1);
/* LINK might be zero if we killed more registers after scheduling
than before, and the last hard register we kill is actually
multiple hard regs.
This is normal for interblock scheduling, so deal with it in
that case, else abort. */
if (link == NULL_RTX && current_nr_blocks <= 1)
abort ();
else if (link == NULL_RTX)
link = alloc_EXPR_LIST (REG_DEAD, gen_rtx (REG, word_mode, 0),
NULL_RTX);
reg_note_regs += (REGNO (XEXP (link, 0)) >= FIRST_PSEUDO_REGISTER ? 1
: HARD_REGNO_NREGS (REGNO (XEXP (link, 0)),
GET_MODE (XEXP (link, 0))));
}
dead_notes = XEXP (link, 1);
/* If we took too many regs kills off, put the extra ones back. */
while (reg_note_regs > regs_killed)
{
rtx temp_reg, temp_link;
temp_reg = gen_rtx (REG, word_mode, 0);
temp_link = alloc_EXPR_LIST (REG_DEAD, temp_reg, dead_notes);
dead_notes = temp_link;
reg_note_regs--;
}
}
XEXP (link, 0) = reg;
XEXP (link, 1) = REG_NOTES (insn);
REG_NOTES (insn) = link;
}
/* Subroutine on attach_deaths_insn--handles the recursive search
through INSN. If SET_P is true, then x is being modified by the insn. */
static void
attach_deaths (x, insn, set_p)
rtx x;
rtx insn;
int set_p;
{
register int i;
register int j;
register enum rtx_code code;
register char *fmt;
if (x == 0)
return;
code = GET_CODE (x);
switch (code)
{
case CONST_INT:
case CONST_DOUBLE:
case LABEL_REF:
case SYMBOL_REF:
case CONST:
case CODE_LABEL:
case PC:
case CC0:
/* Get rid of the easy cases first. */
return;
case REG:
{
/* If the register dies in this insn, queue that note, and mark
this register as needing to die. */
/* This code is very similar to mark_used_1 (if set_p is false)
and mark_set_1 (if set_p is true) in flow.c. */
register int regno;
int some_needed;
int all_needed;
if (set_p)
return;
regno = REGNO (x);
all_needed = some_needed = REGNO_REG_SET_P (old_live_regs, regno);
if (regno < FIRST_PSEUDO_REGISTER)
{
int n;
n = HARD_REGNO_NREGS (regno, GET_MODE (x));
while (--n > 0)
{
int needed = (REGNO_REG_SET_P (old_live_regs, regno + n));
some_needed |= needed;
all_needed &= needed;
}
}
/* If it wasn't live before we started, then add a REG_DEAD note.
We must check the previous lifetime info not the current info,
because we may have to execute this code several times, e.g.
once for a clobber (which doesn't add a note) and later
for a use (which does add a note).
Always make the register live. We must do this even if it was
live before, because this may be an insn which sets and uses
the same register, in which case the register has already been
killed, so we must make it live again.
Global registers are always live, and should never have a REG_DEAD
note added for them, so none of the code below applies to them. */
if (regno >= FIRST_PSEUDO_REGISTER || ! global_regs[regno])
{
/* Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
STACK_POINTER_REGNUM, since these are always considered to be
live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
if (regno != FRAME_POINTER_REGNUM
#if HARD_FRAME_POINTER_REGNUM != FRAME_POINTER_REGNUM
&& ! (regno == HARD_FRAME_POINTER_REGNUM)
#endif
#if ARG_POINTER_REGNUM != FRAME_POINTER_REGNUM
&& ! (regno == ARG_POINTER_REGNUM && fixed_regs[regno])
#endif
&& regno != STACK_POINTER_REGNUM)
{
/* ??? It is perhaps a dead_or_set_p bug that it does
not check for REG_UNUSED notes itself. This is necessary
for the case where the SET_DEST is a subreg of regno, as
dead_or_set_p handles subregs specially. */
if (! all_needed && ! dead_or_set_p (insn, x)
&& ! find_reg_note (insn, REG_UNUSED, x))
{
/* Check for the case where the register dying partially
overlaps the register set by this insn. */
if (regno < FIRST_PSEUDO_REGISTER
&& HARD_REGNO_NREGS (regno, GET_MODE (x)) > 1)
{
int n = HARD_REGNO_NREGS (regno, GET_MODE (x));
while (--n >= 0)
some_needed |= dead_or_set_regno_p (insn, regno + n);
}
/* If none of the words in X is needed, make a REG_DEAD
note. Otherwise, we must make partial REG_DEAD
notes. */
if (! some_needed)
create_reg_dead_note (x, insn);
else
{
int i;
/* Don't make a REG_DEAD note for a part of a
register that is set in the insn. */
for (i = HARD_REGNO_NREGS (regno, GET_MODE (x)) - 1;
i >= 0; i--)
if (! REGNO_REG_SET_P (old_live_regs, regno+i)
&& ! dead_or_set_regno_p (insn, regno + i))
create_reg_dead_note (gen_rtx (REG,
reg_raw_mode[regno + i],
regno + i),
insn);
}
}
}
if (regno < FIRST_PSEUDO_REGISTER)
{
int j = HARD_REGNO_NREGS (regno, GET_MODE (x));
while (--j >= 0)
{
SET_REGNO_REG_SET (bb_live_regs, regno + j);
}
}
else
{
/* Recompute REG_BASIC_BLOCK as we update all the other
dataflow information. */
if (sched_reg_basic_block[regno] == REG_BLOCK_UNKNOWN)
sched_reg_basic_block[regno] = current_block_num;
else if (sched_reg_basic_block[regno] != current_block_num)
sched_reg_basic_block[regno] = REG_BLOCK_GLOBAL;
SET_REGNO_REG_SET (bb_live_regs, regno);
}
}
return;
}
case MEM:
/* Handle tail-recursive case. */
attach_deaths (XEXP (x, 0), insn, 0);
return;
case SUBREG:
case STRICT_LOW_PART:
/* These two cases preserve the value of SET_P, so handle them
separately. */
attach_deaths (XEXP (x, 0), insn, set_p);
return;
case ZERO_EXTRACT:
case SIGN_EXTRACT:
/* This case preserves the value of SET_P for the first operand, but
clears it for the other two. */
attach_deaths (XEXP (x, 0), insn, set_p);
attach_deaths (XEXP (x, 1), insn, 0);
attach_deaths (XEXP (x, 2), insn, 0);
return;
default:
/* Other cases: walk the insn. */
fmt = GET_RTX_FORMAT (code);
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
{
if (fmt[i] == 'e')
attach_deaths (XEXP (x, i), insn, 0);
else if (fmt[i] == 'E')
for (j = 0; j < XVECLEN (x, i); j++)
attach_deaths (XVECEXP (x, i, j), insn, 0);
}
}
}
/* After INSN has executed, add register death notes for each register
that is dead after INSN. */
static void
attach_deaths_insn (insn)
rtx insn;
{
rtx x = PATTERN (insn);
register RTX_CODE code = GET_CODE (x);
rtx link;
if (code == SET)
{
attach_deaths (SET_SRC (x), insn, 0);
/* A register might die here even if it is the destination, e.g.
it is the target of a volatile read and is otherwise unused.
Hence we must always call attach_deaths for the SET_DEST. */
attach_deaths (SET_DEST (x), insn, 1);
}
else if (code == PARALLEL)
{
register int i;
for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
{
code = GET_CODE (XVECEXP (x, 0, i));
if (code == SET)
{
attach_deaths (SET_SRC (XVECEXP (x, 0, i)), insn, 0);
attach_deaths (SET_DEST (XVECEXP (x, 0, i)), insn, 1);
}
/* Flow does not add REG_DEAD notes to registers that die in
clobbers, so we can't either. */
else if (code != CLOBBER)
attach_deaths (XVECEXP (x, 0, i), insn, 0);
}
}
/* If this is a CLOBBER, only add REG_DEAD notes to registers inside a
MEM being clobbered, just like flow. */
else if (code == CLOBBER && GET_CODE (XEXP (x, 0)) == MEM)
attach_deaths (XEXP (XEXP (x, 0), 0), insn, 0);
/* Otherwise don't add a death note to things being clobbered. */
else if (code != CLOBBER)
attach_deaths (x, insn, 0);
/* Make death notes for things used in the called function. */
if (GET_CODE (insn) == CALL_INSN)
for (link = CALL_INSN_FUNCTION_USAGE (insn); link; link = XEXP (link, 1))
attach_deaths (XEXP (XEXP (link, 0), 0), insn,
GET_CODE (XEXP (link, 0)) == CLOBBER);
}
/* functions for handlnig of notes */
/* Delete notes beginning with INSN and put them in the chain
of notes ended by NOTE_LIST.
Returns the insn following the notes. */
static rtx
unlink_other_notes (insn, tail)
rtx insn, tail;
{
rtx prev = PREV_INSN (insn);
while (insn != tail && GET_CODE (insn) == NOTE)
{
rtx next = NEXT_INSN (insn);
/* Delete the note from its current position. */
if (prev)
NEXT_INSN (prev) = next;
if (next)
PREV_INSN (next) = prev;
/* Don't save away NOTE_INSN_SETJMPs, because they must remain
immediately after the call they follow. We use a fake
(REG_DEAD (const_int -1)) note to remember them.
Likewise with NOTE_INSN_{LOOP,EHREGION}_{BEG, END}. */
if (NOTE_LINE_NUMBER (insn) != NOTE_INSN_SETJMP
&& NOTE_LINE_NUMBER (insn) != NOTE_INSN_LOOP_BEG
&& NOTE_LINE_NUMBER (insn) != NOTE_INSN_LOOP_END
&& NOTE_LINE_NUMBER (insn) != NOTE_INSN_EH_REGION_BEG
&& NOTE_LINE_NUMBER (insn) != NOTE_INSN_EH_REGION_END)
{
/* Insert the note at the end of the notes list. */
PREV_INSN (insn) = note_list;
if (note_list)
NEXT_INSN (note_list) = insn;
note_list = insn;
}
insn = next;
}
return insn;
}
/* Delete line notes beginning with INSN. Record line-number notes so
they can be reused. Returns the insn following the notes. */
static rtx
unlink_line_notes (insn, tail)
rtx insn, tail;
{
rtx prev = PREV_INSN (insn);
while (insn != tail && GET_CODE (insn) == NOTE)
{
rtx next = NEXT_INSN (insn);
if (write_symbols != NO_DEBUG && NOTE_LINE_NUMBER (insn) > 0)
{
/* Delete the note from its current position. */
if (prev)
NEXT_INSN (prev) = next;
if (next)
PREV_INSN (next) = prev;
/* Record line-number notes so they can be reused. */
LINE_NOTE (insn) = insn;
}
else
prev = insn;
insn = next;
}
return insn;
}
/* Return the head and tail pointers of BB. */
__inline static void
get_block_head_tail (bb, headp, tailp)
int bb;
rtx *headp;
rtx *tailp;
{
rtx head;
rtx tail;
int b;
b = BB_TO_BLOCK (bb);
/* HEAD and TAIL delimit the basic block being scheduled. */
head = basic_block_head[b];
tail = basic_block_end[b];
/* Don't include any notes or labels at the beginning of the
basic block, or notes at the ends of basic blocks. */
while (head != tail)
{
if (GET_CODE (head) == NOTE)
head = NEXT_INSN (head);
else if (GET_CODE (tail) == NOTE)
tail = PREV_INSN (tail);
else if (GET_CODE (head) == CODE_LABEL)
head = NEXT_INSN (head);
else
break;
}
*headp = head;
*tailp = tail;
}
/* Delete line notes from bb. Save them so they can be later restored
(in restore_line_notes ()). */
static void
rm_line_notes (bb)
int bb;
{
rtx next_tail;
rtx tail;
rtx head;
rtx insn;
get_block_head_tail (bb, &head, &tail);
if (head == tail
&& (GET_RTX_CLASS (GET_CODE (head)) != 'i'))
return;
next_tail = NEXT_INSN (tail);
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
{
rtx prev;
/* Farm out notes, and maybe save them in NOTE_LIST.
This is needed to keep the debugger from
getting completely deranged. */
if (GET_CODE (insn) == NOTE)
{
prev = insn;
insn = unlink_line_notes (insn, next_tail);
if (prev == tail)
abort ();
if (prev == head)
abort ();
if (insn == next_tail)
abort ();
}
}
}
/* Save line number notes for each insn in bb. */
static void
save_line_notes (bb)
int bb;
{
rtx head, tail;
rtx next_tail;
/* We must use the true line number for the first insn in the block
that was computed and saved at the start of this pass. We can't
use the current line number, because scheduling of the previous
block may have changed the current line number. */
rtx line = line_note_head[BB_TO_BLOCK (bb)];
rtx insn;
get_block_head_tail (bb, &head, &tail);
next_tail = NEXT_INSN (tail);
for (insn = basic_block_head[BB_TO_BLOCK (bb)];
insn != next_tail;
insn = NEXT_INSN (insn))
if (GET_CODE (insn) == NOTE && NOTE_LINE_NUMBER (insn) > 0)
line = insn;
else
LINE_NOTE (insn) = line;
}
/* After bb was scheduled, insert line notes into the insns list. */
static void
restore_line_notes (bb)
int bb;
{
rtx line, note, prev, new;
int added_notes = 0;
int b;
rtx head, next_tail, insn;
b = BB_TO_BLOCK (bb);
head = basic_block_head[b];
next_tail = NEXT_INSN (basic_block_end[b]);
/* Determine the current line-number. We want to know the current
line number of the first insn of the block here, in case it is
different from the true line number that was saved earlier. If
different, then we need a line number note before the first insn
of this block. If it happens to be the same, then we don't want to
emit another line number note here. */
for (line = head; line; line = PREV_INSN (line))
if (GET_CODE (line) == NOTE && NOTE_LINE_NUMBER (line) > 0)
break;
/* Walk the insns keeping track of the current line-number and inserting
the line-number notes as needed. */
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
if (GET_CODE (insn) == NOTE && NOTE_LINE_NUMBER (insn) > 0)
line = insn;
/* This used to emit line number notes before every non-deleted note.
However, this confuses a debugger, because line notes not separated
by real instructions all end up at the same address. I can find no
use for line number notes before other notes, so none are emitted. */
else if (GET_CODE (insn) != NOTE
&& (note = LINE_NOTE (insn)) != 0
&& note != line
&& (line == 0
|| NOTE_LINE_NUMBER (note) != NOTE_LINE_NUMBER (line)
|| NOTE_SOURCE_FILE (note) != NOTE_SOURCE_FILE (line)))
{
line = note;
prev = PREV_INSN (insn);
if (LINE_NOTE (note))
{
/* Re-use the original line-number note. */
LINE_NOTE (note) = 0;
PREV_INSN (note) = prev;
NEXT_INSN (prev) = note;
PREV_INSN (insn) = note;
NEXT_INSN (note) = insn;
}
else
{
added_notes++;
new = emit_note_after (NOTE_LINE_NUMBER (note), prev);
NOTE_SOURCE_FILE (new) = NOTE_SOURCE_FILE (note);
RTX_INTEGRATED_P (new) = RTX_INTEGRATED_P (note);
}
}
if (sched_verbose && added_notes)
fprintf (dump, ";; added %d line-number notes\n", added_notes);
}
/* After scheduling the function, delete redundant line notes from the
insns list. */
static void
rm_redundant_line_notes ()
{
rtx line = 0;
rtx insn = get_insns ();
int active_insn = 0;
int notes = 0;
/* Walk the insns deleting redundant line-number notes. Many of these
are already present. The remainder tend to occur at basic
block boundaries. */
for (insn = get_last_insn (); insn; insn = PREV_INSN (insn))
if (GET_CODE (insn) == NOTE && NOTE_LINE_NUMBER (insn) > 0)
{
/* If there are no active insns following, INSN is redundant. */
if (active_insn == 0)
{
notes++;
NOTE_SOURCE_FILE (insn) = 0;
NOTE_LINE_NUMBER (insn) = NOTE_INSN_DELETED;
}
/* If the line number is unchanged, LINE is redundant. */
else if (line
&& NOTE_LINE_NUMBER (line) == NOTE_LINE_NUMBER (insn)
&& NOTE_SOURCE_FILE (line) == NOTE_SOURCE_FILE (insn))
{
notes++;
NOTE_SOURCE_FILE (line) = 0;
NOTE_LINE_NUMBER (line) = NOTE_INSN_DELETED;
line = insn;
}
else
line = insn;
active_insn = 0;
}
else if (!((GET_CODE (insn) == NOTE
&& NOTE_LINE_NUMBER (insn) == NOTE_INSN_DELETED)
|| (GET_CODE (insn) == INSN
&& (GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER))))
active_insn++;
if (sched_verbose && notes)
fprintf (dump, ";; deleted %d line-number notes\n", notes);
}
/* Delete notes between head and tail and put them in the chain
of notes ended by NOTE_LIST. */
static void
rm_other_notes (head, tail)
rtx head;
rtx tail;
{
rtx next_tail;
rtx insn;
if (head == tail
&& (GET_RTX_CLASS (GET_CODE (head)) != 'i'))
return;
next_tail = NEXT_INSN (tail);
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
{
rtx prev;
/* Farm out notes, and maybe save them in NOTE_LIST.
This is needed to keep the debugger from
getting completely deranged. */
if (GET_CODE (insn) == NOTE)
{
prev = insn;
insn = unlink_other_notes (insn, next_tail);
if (prev == tail)
abort ();
if (prev == head)
abort ();
if (insn == next_tail)
abort ();
}
}
}
/* Constructor for `sometimes' data structure. */
static int
new_sometimes_live (regs_sometimes_live, regno, sometimes_max)
struct sometimes *regs_sometimes_live;
int regno;
int sometimes_max;
{
register struct sometimes *p;
/* There should never be a register greater than max_regno here. If there
is, it means that a define_split has created a new pseudo reg. This
is not allowed, since there will not be flow info available for any
new register, so catch the error here. */
if (regno >= max_regno)
abort ();
p = ®s_sometimes_live[sometimes_max];
p->regno = regno;
p->live_length = 0;
p->calls_crossed = 0;
sometimes_max++;
return sometimes_max;
}
/* Count lengths of all regs we are currently tracking,
and find new registers no longer live. */
static void
finish_sometimes_live (regs_sometimes_live, sometimes_max)
struct sometimes *regs_sometimes_live;
int sometimes_max;
{
int i;
for (i = 0; i < sometimes_max; i++)
{
register struct sometimes *p = ®s_sometimes_live[i];
int regno = p->regno;
sched_reg_live_length[regno] += p->live_length;
sched_reg_n_calls_crossed[regno] += p->calls_crossed;
}
}
/* functions for computation of registers live/usage info */
/* It is assumed that prior to scheduling basic_block_live_at_start (b)
contains the registers that are alive at the entry to b.
Two passes follow: The first pass is performed before the scheduling
of a region. It scans each block of the region forward, computing
the set of registers alive at the end of the basic block and
discard REG_DEAD notes (done by find_pre_sched_live ()).
The second path is invoked after scheduling all region blocks.
It scans each block of the region backward, a block being traversed
only after its succesors in the region. When the set of registers
live at the end of a basic block may be changed by the scheduling
(this may happen for multiple blocks region), it is computed as
the union of the registers live at the start of its succesors.
The last-use information is updated by inserting REG_DEAD notes.
(done by find_post_sched_live ()) */
/* Scan all the insns to be scheduled, removing register death notes.
Register death notes end up in DEAD_NOTES.
Recreate the register life information for the end of this basic
block. */
static void
find_pre_sched_live (bb)
int bb;
{
rtx insn, next_tail, head, tail;
int b = BB_TO_BLOCK (bb);
get_block_head_tail (bb, &head, &tail);
COPY_REG_SET (bb_live_regs, basic_block_live_at_start[b]);
next_tail = NEXT_INSN (tail);
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
{
rtx prev, next, link;
int reg_weight = 0;
/* Handle register life information. */
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i')
{
/* See if the register gets born here. */
/* We must check for registers being born before we check for
registers dying. It is possible for a register to be born and
die in the same insn, e.g. reading from a volatile memory
location into an otherwise unused register. Such a register
must be marked as dead after this insn. */
if (GET_CODE (PATTERN (insn)) == SET
|| GET_CODE (PATTERN (insn)) == CLOBBER)
{
sched_note_set (b, PATTERN (insn), 0);
reg_weight++;
}
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
{
int j;
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
{
sched_note_set (b, XVECEXP (PATTERN (insn), 0, j), 0);
reg_weight++;
}
/* ??? This code is obsolete and should be deleted. It
is harmless though, so we will leave it in for now. */
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == USE)
sched_note_set (b, XVECEXP (PATTERN (insn), 0, j), 0);
}
/* Each call cobbers (makes live) all call-clobbered regs
that are not global or fixed. Note that the function-value
reg is a call_clobbered reg. */
if (GET_CODE (insn) == CALL_INSN)
{
int j;
for (j = 0; j < FIRST_PSEUDO_REGISTER; j++)
if (call_used_regs[j] && !global_regs[j]
&& ! fixed_regs[j])
{
SET_REGNO_REG_SET (bb_live_regs, j);
}
}
/* Need to know what registers this insn kills. */
for (prev = 0, link = REG_NOTES (insn); link; link = next)
{
next = XEXP (link, 1);
if ((REG_NOTE_KIND (link) == REG_DEAD
|| REG_NOTE_KIND (link) == REG_UNUSED)
/* Verify that the REG_NOTE has a valid value. */
&& GET_CODE (XEXP (link, 0)) == REG)
{
register int regno = REGNO (XEXP (link, 0));
reg_weight--;
/* Only unlink REG_DEAD notes; leave REG_UNUSED notes
alone. */
if (REG_NOTE_KIND (link) == REG_DEAD)
{
if (prev)
XEXP (prev, 1) = next;
else
REG_NOTES (insn) = next;
XEXP (link, 1) = dead_notes;
dead_notes = link;
}
else
prev = link;
if (regno < FIRST_PSEUDO_REGISTER)
{
int j = HARD_REGNO_NREGS (regno,
GET_MODE (XEXP (link, 0)));
while (--j >= 0)
{
CLEAR_REGNO_REG_SET (bb_live_regs, regno+j);
}
}
else
{
CLEAR_REGNO_REG_SET (bb_live_regs, regno);
}
}
else
prev = link;
}
}
INSN_REG_WEIGHT (insn) = reg_weight;
}
}
/* Update register life and usage information for block bb
after scheduling. Put register dead notes back in the code. */
static void
find_post_sched_live (bb)
int bb;
{
int sometimes_max;
int j, i;
int b;
rtx insn;
rtx head, tail, prev_head, next_tail;
register struct sometimes *regs_sometimes_live;
b = BB_TO_BLOCK (bb);
/* compute live regs at the end of bb as a function of its successors. */
if (current_nr_blocks > 1)
{
int e;
int first_edge;
first_edge = e = OUT_EDGES (b);
CLEAR_REG_SET (bb_live_regs);
if (e)
do
{
int b_succ;
b_succ = TO_BLOCK (e);
IOR_REG_SET (bb_live_regs, basic_block_live_at_start[b_succ]);
e = NEXT_OUT (e);
}
while (e != first_edge);
}
get_block_head_tail (bb, &head, &tail);
next_tail = NEXT_INSN (tail);
prev_head = PREV_INSN (head);
for (i = FIRST_PSEUDO_REGISTER; i < max_regno; i++)
if (REGNO_REG_SET_P (bb_live_regs, i))
sched_reg_basic_block[i] = REG_BLOCK_GLOBAL;
/* if the block is empty, same regs are alive at its end and its start.
since this is not guaranteed after interblock scheduling, make sure they
are truly identical. */
if (NEXT_INSN (prev_head) == tail
&& (GET_RTX_CLASS (GET_CODE (tail)) != 'i'))
{
if (current_nr_blocks > 1)
COPY_REG_SET (basic_block_live_at_start[b], bb_live_regs);
return;
}
b = BB_TO_BLOCK (bb);
current_block_num = b;
/* Keep track of register lives. */
old_live_regs = ALLOCA_REG_SET ();
regs_sometimes_live
= (struct sometimes *) alloca (max_regno * sizeof (struct sometimes));
sometimes_max = 0;
/* initiate "sometimes" data, starting with registers live at end */
sometimes_max = 0;
COPY_REG_SET (old_live_regs, bb_live_regs);
EXECUTE_IF_SET_IN_REG_SET (bb_live_regs, 0, j,
{
sometimes_max
= new_sometimes_live (regs_sometimes_live,
j, sometimes_max);
});
/* scan insns back, computing regs live info */
for (insn = tail; insn != prev_head; insn = PREV_INSN (insn))
{
/* First we kill registers set by this insn, and then we
make registers used by this insn live. This is the opposite
order used above because we are traversing the instructions
backwards. */
/* Strictly speaking, we should scan REG_UNUSED notes and make
every register mentioned there live, however, we will just
kill them again immediately below, so there doesn't seem to
be any reason why we bother to do this. */
/* See if this is the last notice we must take of a register. */
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
continue;
if (GET_CODE (PATTERN (insn)) == SET
|| GET_CODE (PATTERN (insn)) == CLOBBER)
sched_note_set (b, PATTERN (insn), 1);
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
{
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
sched_note_set (b, XVECEXP (PATTERN (insn), 0, j), 1);
}
/* This code keeps life analysis information up to date. */
if (GET_CODE (insn) == CALL_INSN)
{
register struct sometimes *p;
/* A call kills all call used registers that are not
global or fixed, except for those mentioned in the call
pattern which will be made live again later. */
for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
if (call_used_regs[i] && ! global_regs[i]
&& ! fixed_regs[i])
{
CLEAR_REGNO_REG_SET (bb_live_regs, i);
}
/* Regs live at the time of a call instruction must not
go in a register clobbered by calls. Record this for
all regs now live. Note that insns which are born or
die in a call do not cross a call, so this must be done
after the killings (above) and before the births
(below). */
p = regs_sometimes_live;
for (i = 0; i < sometimes_max; i++, p++)
if (REGNO_REG_SET_P (bb_live_regs, p->regno))
p->calls_crossed += 1;
}
/* Make every register used live, and add REG_DEAD notes for
registers which were not live before we started. */
attach_deaths_insn (insn);
/* Find registers now made live by that instruction. */
EXECUTE_IF_AND_COMPL_IN_REG_SET (bb_live_regs, old_live_regs, 0, j,
{
sometimes_max
= new_sometimes_live (regs_sometimes_live,
j, sometimes_max);
});
IOR_REG_SET (old_live_regs, bb_live_regs);
/* Count lengths of all regs we are worrying about now,
and handle registers no longer live. */
for (i = 0; i < sometimes_max; i++)
{
register struct sometimes *p = ®s_sometimes_live[i];
int regno = p->regno;
p->live_length += 1;
if (!REGNO_REG_SET_P (bb_live_regs, regno))
{
/* This is the end of one of this register's lifetime
segments. Save the lifetime info collected so far,
and clear its bit in the old_live_regs entry. */
sched_reg_live_length[regno] += p->live_length;
sched_reg_n_calls_crossed[regno] += p->calls_crossed;
CLEAR_REGNO_REG_SET (old_live_regs, p->regno);
/* Delete the reg_sometimes_live entry for this reg by
copying the last entry over top of it. */
*p = regs_sometimes_live[--sometimes_max];
/* ...and decrement i so that this newly copied entry
will be processed. */
i--;
}
}
}
finish_sometimes_live (regs_sometimes_live, sometimes_max);
/* In interblock scheduling, basic_block_live_at_start may have changed. */
if (current_nr_blocks > 1)
COPY_REG_SET (basic_block_live_at_start[b], bb_live_regs);
FREE_REG_SET (old_live_regs);
} /* find_post_sched_live */
/* After scheduling the subroutine, restore information about uses of
registers. */
static void
update_reg_usage ()
{
int regno;
if (n_basic_blocks > 0)
for (regno = FIRST_PSEUDO_REGISTER; regno < max_regno; regno++)
if (REGNO_REG_SET_P (basic_block_live_at_start[0], regno))
sched_reg_basic_block[regno] = REG_BLOCK_GLOBAL;
for (regno = 0; regno < max_regno; regno++)
if (sched_reg_live_length[regno])
{
if (sched_verbose)
{
if (REG_LIVE_LENGTH (regno) > sched_reg_live_length[regno])
fprintf (dump,
";; register %d life shortened from %d to %d\n",
regno, REG_LIVE_LENGTH (regno),
sched_reg_live_length[regno]);
/* Negative values are special; don't overwrite the current
reg_live_length value if it is negative. */
else if (REG_LIVE_LENGTH (regno) < sched_reg_live_length[regno]
&& REG_LIVE_LENGTH (regno) >= 0)
fprintf (dump,
";; register %d life extended from %d to %d\n",
regno, REG_LIVE_LENGTH (regno),
sched_reg_live_length[regno]);
if (!REG_N_CALLS_CROSSED (regno)
&& sched_reg_n_calls_crossed[regno])
fprintf (dump,
";; register %d now crosses calls\n", regno);
else if (REG_N_CALLS_CROSSED (regno)
&& !sched_reg_n_calls_crossed[regno]
&& REG_BASIC_BLOCK (regno) != REG_BLOCK_GLOBAL)
fprintf (dump,
";; register %d no longer crosses calls\n", regno);
if (REG_BASIC_BLOCK (regno) != sched_reg_basic_block[regno]
&& sched_reg_basic_block[regno] != REG_BLOCK_UNKNOWN
&& REG_BASIC_BLOCK(regno) != REG_BLOCK_UNKNOWN)
fprintf (dump,
";; register %d changed basic block from %d to %d\n",
regno, REG_BASIC_BLOCK(regno),
sched_reg_basic_block[regno]);
}
/* Negative values are special; don't overwrite the current
reg_live_length value if it is negative. */
if (REG_LIVE_LENGTH (regno) >= 0)
REG_LIVE_LENGTH (regno) = sched_reg_live_length[regno];
if (sched_reg_basic_block[regno] != REG_BLOCK_UNKNOWN
&& REG_BASIC_BLOCK(regno) != REG_BLOCK_UNKNOWN)
REG_BASIC_BLOCK(regno) = sched_reg_basic_block[regno];
/* We can't change the value of reg_n_calls_crossed to zero for
pseudos which are live in more than one block.
This is because combine might have made an optimization which
invalidated basic_block_live_at_start and reg_n_calls_crossed,
but it does not update them. If we update reg_n_calls_crossed
here, the two variables are now inconsistent, and this might
confuse the caller-save code into saving a register that doesn't
need to be saved. This is only a problem when we zero calls
crossed for a pseudo live in multiple basic blocks.
Alternatively, we could try to correctly update basic block live
at start here in sched, but that seems complicated.
Note: it is possible that a global register became local, as result
of interblock motion, but will remain marked as a global register. */
if (sched_reg_n_calls_crossed[regno]
|| REG_BASIC_BLOCK (regno) != REG_BLOCK_GLOBAL)
REG_N_CALLS_CROSSED (regno) = sched_reg_n_calls_crossed[regno];
}
}
/* Scheduling clock, modified in schedule_block() and queue_to_ready () */
static int clock_var;
/* Move insns that became ready to fire from queue to ready list. */
static int
queue_to_ready (ready, n_ready)
rtx ready[];
int n_ready;
{
rtx insn;
rtx link;
q_ptr = NEXT_Q (q_ptr);
/* Add all pending insns that can be scheduled without stalls to the
ready list. */
for (link = insn_queue[q_ptr]; link; link = XEXP (link, 1))
{
insn = XEXP (link, 0);
q_size -= 1;
if (sched_verbose >= 2)
fprintf (dump, ";;\t\tQ-->Ready: insn %d: ", INSN_UID (insn));
if (sched_verbose >= 2 && INSN_BB (insn) != target_bb)
fprintf (dump, "(b%d) ", INSN_BLOCK (insn));
ready[n_ready++] = insn;
if (sched_verbose >= 2)
fprintf (dump, "moving to ready without stalls\n");
}
insn_queue[q_ptr] = 0;
/* If there are no ready insns, stall until one is ready and add all
of the pending insns at that point to the ready list. */
if (n_ready == 0)
{
register int stalls;
for (stalls = 1; stalls < INSN_QUEUE_SIZE; stalls++)
{
if ((link = insn_queue[NEXT_Q_AFTER (q_ptr, stalls)]))
{
for (; link; link = XEXP (link, 1))
{
insn = XEXP (link, 0);
q_size -= 1;
if (sched_verbose >= 2)
fprintf (dump, ";;\t\tQ-->Ready: insn %d: ", INSN_UID (insn));
if (sched_verbose >= 2 && INSN_BB (insn) != target_bb)
fprintf (dump, "(b%d) ", INSN_BLOCK (insn));
ready[n_ready++] = insn;
if (sched_verbose >= 2)
fprintf (dump, "moving to ready with %d stalls\n", stalls);
}
insn_queue[NEXT_Q_AFTER (q_ptr, stalls)] = 0;
if (n_ready)
break;
}
}
if (sched_verbose && stalls)
visualize_stall_cycles (BB_TO_BLOCK (target_bb), stalls);
q_ptr = NEXT_Q_AFTER (q_ptr, stalls);
clock_var += stalls;
}
return n_ready;
}
/* Print the ready list for debugging purposes. Callable from debugger. */
extern void
debug_ready_list (ready, n_ready)
rtx ready[];
int n_ready;
{
int i;
for (i = 0; i < n_ready; i++)
{
fprintf (dump, " %d", INSN_UID (ready[i]));
if (current_nr_blocks > 1 && INSN_BB (ready[i]) != target_bb)
fprintf (dump, "/b%d", INSN_BLOCK (ready[i]));
}
fprintf (dump, "\n");
}
/* Print names of units on which insn can/should execute, for debugging. */
static void
insn_print_units (insn)
rtx insn;
{
int i;
int unit = insn_unit (insn);
if (unit == -1)
fprintf (dump, "none");
else if (unit >= 0)
fprintf (dump, "%s", function_units[unit].name);
else
{
fprintf (dump, "[");
for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
if (unit & 1)
{
fprintf (dump, "%s", function_units[i].name);
if (unit != 1)
fprintf (dump, " ");
}
fprintf (dump, "]");
}
}
/* MAX_VISUAL_LINES is the maximum number of lines in visualization table
of a basic block. If more lines are needed, table is splitted to two.
n_visual_lines is the number of lines printed so far for a block.
visual_tbl contains the block visualization info.
vis_no_unit holds insns in a cycle that are not mapped to any unit. */
#define MAX_VISUAL_LINES 100
#define INSN_LEN 30
int n_visual_lines;
char *visual_tbl;
int n_vis_no_unit;
rtx vis_no_unit[10];
/* Finds units that are in use in this fuction. Required only
for visualization. */
static void
init_target_units ()
{
rtx insn;
int unit;
for (insn = get_last_insn (); insn; insn = PREV_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
continue;
unit = insn_unit (insn);
if (unit < 0)
target_units |= ~unit;
else
target_units |= (1 << unit);
}
}
/* Return the length of the visualization table */
static int
get_visual_tbl_length ()
{
int unit, i;
int n, n1;
char *s;
/* compute length of one field in line */
s = (char *) alloca (INSN_LEN + 5);
sprintf (s, " %33s", "uname");
n1 = strlen (s);
/* compute length of one line */
n = strlen (";; ");
n += n1;
for (unit = 0; unit < FUNCTION_UNITS_SIZE; unit++)
if (function_units[unit].bitmask & target_units)
for (i = 0; i < function_units[unit].multiplicity; i++)
n += n1;
n += n1;
n += strlen ("\n") + 2;
/* compute length of visualization string */
return (MAX_VISUAL_LINES * n);
}
/* Init block visualization debugging info */
static void
init_block_visualization ()
{
strcpy (visual_tbl, "");
n_visual_lines = 0;
n_vis_no_unit = 0;
}
#define BUF_LEN 256
/* This recognizes rtx, I classified as expressions. These are always */
/* represent some action on values or results of other expression, */
/* that may be stored in objects representing values. */
static void
print_exp (buf, x, verbose)
char *buf;
rtx x;
int verbose;
{
char t1[BUF_LEN], t2[BUF_LEN], t3[BUF_LEN];
switch (GET_CODE (x))
{
case PLUS:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s+%s", t1, t2);
break;
case LO_SUM:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%sl+%s", t1, t2);
break;
case MINUS:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s-%s", t1, t2);
break;
case COMPARE:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s??%s", t1, t2);
break;
case NEG:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "-%s", t1);
break;
case MULT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s*%s", t1, t2);
break;
case DIV:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s/%s", t1, t2);
break;
case UDIV:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%su/%s", t1, t2);
break;
case MOD:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s%%%s", t1, t2);
break;
case UMOD:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%su%%%s", t1, t2);
break;
case SMIN:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "smin (%s, %s)", t1, t2);
break;
case SMAX:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "smax(%s,%s)", t1, t2);
break;
case UMIN:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "umin (%s, %s)", t1, t2);
break;
case UMAX:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "umax(%s,%s)", t1, t2);
break;
case NOT:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "!%s", t1);
break;
case AND:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s&%s", t1, t2);
break;
case IOR:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s|%s", t1, t2);
break;
case XOR:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s^%s", t1, t2);
break;
case ASHIFT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s<<%s", t1, t2);
break;
case LSHIFTRT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s0>%s", t1, t2);
break;
case ASHIFTRT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s>>%s", t1, t2);
break;
case ROTATE:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s<-<%s", t1, t2);
break;
case ROTATERT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s>->%s", t1, t2);
break;
case ABS:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "abs(%s)", t1);
break;
case SQRT:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "sqrt(%s)", t1);
break;
case FFS:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "ffs(%s)", t1);
break;
case EQ:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s == %s", t1, t2);
break;
case NE:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s!=%s", t1, t2);
break;
case GT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s>%s", t1, t2);
break;
case GTU:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s>u%s", t1, t2);
break;
case LT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s<%s", t1, t2);
break;
case LTU:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s<u%s", t1, t2);
break;
case GE:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s>=%s", t1, t2);
break;
case GEU:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s>=u%s", t1, t2);
break;
case LE:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s<=%s", t1, t2);
break;
case LEU:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "%s<=u%s", t1, t2);
break;
case SIGN_EXTRACT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
print_value (t3, XEXP (x, 2), verbose);
if (verbose)
sprintf (buf, "sign_extract(%s,%s,%s)", t1, t2, t3);
else
sprintf (buf, "sxt(%s,%s,%s)", t1, t2, t3);
break;
case ZERO_EXTRACT:
print_value (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
print_value (t3, XEXP (x, 2), verbose);
if (verbose)
sprintf (buf, "zero_extract(%s,%s,%s)", t1, t2, t3);
else
sprintf (buf, "zxt(%s,%s,%s)", t1, t2, t3);
break;
case SIGN_EXTEND:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "sign_extend(%s)", t1);
else
sprintf (buf, "sxn(%s)", t1);
break;
case ZERO_EXTEND:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "zero_extend(%s)", t1);
else
sprintf (buf, "zxn(%s)", t1);
break;
case FLOAT_EXTEND:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "float_extend(%s)", t1);
else
sprintf (buf, "fxn(%s)", t1);
break;
case TRUNCATE:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "trunc(%s)", t1);
else
sprintf (buf, "trn(%s)", t1);
break;
case FLOAT_TRUNCATE:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "float_trunc(%s)", t1);
else
sprintf (buf, "ftr(%s)", t1);
break;
case FLOAT:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "float(%s)", t1);
else
sprintf (buf, "flt(%s)", t1);
break;
case UNSIGNED_FLOAT:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "uns_float(%s)", t1);
else
sprintf (buf, "ufl(%s)", t1);
break;
case FIX:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "fix(%s)", t1);
break;
case UNSIGNED_FIX:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
sprintf (buf, "uns_fix(%s)", t1);
else
sprintf (buf, "ufx(%s)", t1);
break;
case PRE_DEC:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "--%s", t1);
break;
case PRE_INC:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "++%s", t1);
break;
case POST_DEC:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "%s--", t1);
break;
case POST_INC:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "%s++", t1);
break;
case CALL:
print_value (t1, XEXP (x, 0), verbose);
if (verbose)
{
print_value (t2, XEXP (x, 1), verbose);
sprintf (buf, "call %s argc:%s", t1, t2);
}
else
sprintf (buf, "call %s", t1);
break;
case IF_THEN_ELSE:
print_exp (t1, XEXP (x, 0), verbose);
print_value (t2, XEXP (x, 1), verbose);
print_value (t3, XEXP (x, 2), verbose);
sprintf (buf, "{(%s)?%s:%s}", t1, t2, t3);
break;
case TRAP_IF:
print_value (t1, TRAP_CONDITION (x), verbose);
sprintf (buf, "trap_if %s", t1);
break;
case UNSPEC:
{
int i;
sprintf (t1, "unspec{");
for (i = 0; i < XVECLEN (x, 0); i++)
{
print_pattern (t2, XVECEXP (x, 0, i), verbose);
sprintf (t3, "%s%s;", t1, t2);
strcpy (t1, t3);
}
sprintf (buf, "%s}", t1);
}
break;
case UNSPEC_VOLATILE:
{
int i;
sprintf (t1, "unspec/v{");
for (i = 0; i < XVECLEN (x, 0); i++)
{
print_pattern (t2, XVECEXP (x, 0, i), verbose);
sprintf (t3, "%s%s;", t1, t2);
strcpy (t1, t3);
}
sprintf (buf, "%s}", t1);
}
break;
default:
/* if (verbose) debug_rtx (x); else sprintf (buf, "$$$"); */
sprintf (buf, "$$$");
}
} /* print_exp */
/* Prints rtxes, i customly classified as values. They're constants, */
/* registers, labels, symbols and memory accesses. */
static void
print_value (buf, x, verbose)
char *buf;
rtx x;
int verbose;
{
char t[BUF_LEN];
switch (GET_CODE (x))
{
case CONST_INT:
sprintf (buf, "%Xh", INTVAL (x));
break;
case CONST_DOUBLE:
print_value (t, XEXP (x, 0), verbose);
sprintf (buf, "<%s>", t);
break;
case CONST_STRING:
sprintf (buf, "\"%s\"", (char *) XEXP (x, 0));
break;
case SYMBOL_REF:
sprintf (buf, "`%s'", (char *) XEXP (x, 0));
break;
case LABEL_REF:
sprintf (buf, "L%d", INSN_UID (XEXP (x, 0)));
break;
case CONST:
print_value (buf, XEXP (x, 0), verbose);
break;
case HIGH:
print_value (buf, XEXP (x, 0), verbose);
break;
case REG:
if (GET_MODE (x) == SFmode
|| GET_MODE (x) == DFmode
|| GET_MODE (x) == XFmode
|| GET_MODE (x) == TFmode)
strcpy (t, "fr");
else
strcpy (t, "r");
sprintf (buf, "%s%d", t, REGNO (x));
break;
case SUBREG:
print_value (t, XEXP (x, 0), verbose);
sprintf (buf, "%s#%d", t, SUBREG_WORD (x));
break;
case SCRATCH:
sprintf (buf, "scratch");
break;
case CC0:
sprintf (buf, "cc0");
break;
case PC:
sprintf (buf, "pc");
break;
case MEM:
print_value (t, XEXP (x, 0), verbose);
sprintf (buf, "[%s]", t);
break;
default:
print_exp (buf, x, verbose);
}
} /* print_value */
/* The next step in insn detalization, its pattern recognition */
static void
print_pattern (buf, x, verbose)
char *buf;
rtx x;
int verbose;
{
char t1[BUF_LEN], t2[BUF_LEN], t3[BUF_LEN];
switch (GET_CODE (x))
{
case SET:
print_value (t1, SET_DEST (x), verbose);
print_value (t2, SET_SRC (x), verbose);
sprintf (buf, "%s=%s", t1, t2);
break;
case RETURN:
sprintf (buf, "return");
break;
case CALL:
print_exp (buf, x, verbose);
break;
case CLOBBER:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "clobber %s", t1);
break;
case USE:
print_value (t1, XEXP (x, 0), verbose);
sprintf (buf, "use %s", t1);
break;
case PARALLEL:
{
int i;
sprintf (t1, "{");
for (i = 0; i < XVECLEN (x, 0); i++)
{
print_pattern (t2, XVECEXP (x, 0, i), verbose);
sprintf (t3, "%s%s;", t1, t2);
strcpy (t1, t3);
}
sprintf (buf, "%s}", t1);
}
break;
case SEQUENCE:
{
int i;
sprintf (t1, "%%{");
for (i = 0; i < XVECLEN (x, 0); i++)
{
print_insn (t2, XVECEXP (x, 0, i), verbose);
sprintf (t3, "%s%s;", t1, t2);
strcpy (t1, t3);
}
sprintf (buf, "%s%%}", t1);
}
break;
case ASM_INPUT:
sprintf (buf, "asm {%s}", XEXP (x, 0));
break;
case ADDR_VEC:
break;
case ADDR_DIFF_VEC:
print_value (buf, XEXP (x, 0), verbose);
break;
case TRAP_IF:
print_value (t1, TRAP_CONDITION (x), verbose);
sprintf (buf, "trap_if %s", t1);
break;
case UNSPEC:
{
int i;
sprintf (t1, "unspec{");
for (i = 0; i < XVECLEN (x, 0); i++)
{
print_pattern (t2, XVECEXP (x, 0, i), verbose);
sprintf (t3, "%s%s;", t1, t2);
strcpy (t1, t3);
}
sprintf (buf, "%s}", t1);
}
break;
case UNSPEC_VOLATILE:
{
int i;
sprintf (t1, "unspec/v{");
for (i = 0; i < XVECLEN (x, 0); i++)
{
print_pattern (t2, XVECEXP (x, 0, i), verbose);
sprintf (t3, "%s%s;", t1, t2);
strcpy (t1, t3);
}
sprintf (buf, "%s}", t1);
}
break;
default:
print_value (buf, x, verbose);
}
} /* print_pattern */
/* This is the main function in rtl visualization mechanism. It
accepts an rtx and tries to recognize it as an insn, then prints it
properly in human readable form, resembling assembler mnemonics. */
/* For every insn it prints its UID and BB the insn belongs */
/* too. (probably the last "option" should be extended somehow, since */
/* it depends now on sched.c inner variables ...) */
static void
print_insn (buf, x, verbose)
char *buf;
rtx x;
int verbose;
{
char t[BUF_LEN];
rtx insn = x;
switch (GET_CODE (x))
{
case INSN:
print_pattern (t, PATTERN (x), verbose);
if (verbose)
sprintf (buf, "b%d: i% 4d: %s", INSN_BB (x),
INSN_UID (x), t);
else
sprintf (buf, "%-4d %s", INSN_UID (x), t);
break;
case JUMP_INSN:
print_pattern (t, PATTERN (x), verbose);
if (verbose)
sprintf (buf, "b%d: i% 4d: jump %s", INSN_BB (x),
INSN_UID (x), t);
else
sprintf (buf, "%-4d %s", INSN_UID (x), t);
break;
case CALL_INSN:
x = PATTERN (insn);
if (GET_CODE (x) == PARALLEL)
{
x = XVECEXP (x, 0, 0);
print_pattern (t, x, verbose);
}
else
strcpy (t, "call <...>");
if (verbose)
sprintf (buf, "b%d: i% 4d: %s", INSN_BB (insn),
INSN_UID (insn), t);
else
sprintf (buf, "%-4d %s", INSN_UID (insn), t);
break;
case CODE_LABEL:
sprintf (buf, "L%d:", INSN_UID (x));
break;
case BARRIER:
sprintf (buf, "i% 4d: barrier", INSN_UID (x));
break;
case NOTE:
if (NOTE_LINE_NUMBER (x) > 0)
sprintf (buf, "%4d note \"%s\" %d", INSN_UID (x),
NOTE_SOURCE_FILE (x), NOTE_LINE_NUMBER (x));
else
sprintf (buf, "%4d %s", INSN_UID (x),
GET_NOTE_INSN_NAME (NOTE_LINE_NUMBER (x)));
break;
default:
if (verbose)
{
sprintf (buf, "Not an INSN at all\n");
debug_rtx (x);
}
else
sprintf (buf, "i%-4d <What?>", INSN_UID (x));
}
} /* print_insn */
void
print_insn_chain (rtx_first)
rtx rtx_first;
{
register rtx tmp_rtx;
char str[BUF_LEN];
strcpy (str, "(nil)\n");
if (rtx_first != 0)
switch (GET_CODE (rtx_first))
{
case INSN:
case JUMP_INSN:
case CALL_INSN:
case NOTE:
case CODE_LABEL:
case BARRIER:
for (tmp_rtx = rtx_first; tmp_rtx != NULL;
tmp_rtx = NEXT_INSN (tmp_rtx))
{
print_insn (str, tmp_rtx, 0);
printf ("%s\n", str);
}
break;
default:
print_insn (str, rtx_first, 0);
printf ("%s\n", str);
}
} /* print_insn_chain */
/* Print visualization debugging info */
static void
print_block_visualization (b, s)
int b;
char *s;
{
int unit, i;
/* print header */
fprintf (dump, "\n;; ==================== scheduling visualization for block %d %s \n", b, s);
/* Print names of units */
fprintf (dump, ";; %-8s", "clock");
for (unit = 0; unit < FUNCTION_UNITS_SIZE; unit++)
if (function_units[unit].bitmask & target_units)
for (i = 0; i < function_units[unit].multiplicity; i++)
fprintf (dump, " %-33s", function_units[unit].name);
fprintf (dump, " %-8s\n", "no-unit");
fprintf (dump, ";; %-8s", "=====");
for (unit = 0; unit < FUNCTION_UNITS_SIZE; unit++)
if (function_units[unit].bitmask & target_units)
for (i = 0; i < function_units[unit].multiplicity; i++)
fprintf (dump, " %-33s", "==============================");
fprintf (dump, " %-8s\n", "=======");
/* Print insns in each cycle */
fprintf (dump, "%s\n", visual_tbl);
}
/* Print insns in the 'no_unit' column of visualization */
static void
visualize_no_unit (insn)
rtx insn;
{
vis_no_unit[n_vis_no_unit] = insn;
n_vis_no_unit++;
}
/* Print insns scheduled in clock, for visualization. */
static void
visualize_scheduled_insns (b, clock)
int b, clock;
{
int i, unit;
/* if no more room, split table into two */
if (n_visual_lines >= MAX_VISUAL_LINES)
{
print_block_visualization (b, "(incomplete)");
init_block_visualization ();
}
n_visual_lines++;
sprintf (visual_tbl + strlen (visual_tbl), ";; %-8d", clock);
for (unit = 0; unit < FUNCTION_UNITS_SIZE; unit++)
if (function_units[unit].bitmask & target_units)
for (i = 0; i < function_units[unit].multiplicity; i++)
{
int instance = unit + i * FUNCTION_UNITS_SIZE;
rtx insn = unit_last_insn[instance];
/* print insns that still keep the unit busy */
if (insn &&
actual_hazard_this_instance (unit, instance, insn, clock, 0))
{
char str[BUF_LEN];
print_insn (str, insn, 0);
str[INSN_LEN] = '\0';
sprintf (visual_tbl + strlen (visual_tbl), " %-33s", str);
}
else
sprintf (visual_tbl + strlen (visual_tbl), " %-33s", "------------------------------");
}
/* print insns that are not assigned to any unit */
for (i = 0; i < n_vis_no_unit; i++)
sprintf (visual_tbl + strlen (visual_tbl), " %-8d",
INSN_UID (vis_no_unit[i]));
n_vis_no_unit = 0;
sprintf (visual_tbl + strlen (visual_tbl), "\n");
}
/* Print stalled cycles */
static void
visualize_stall_cycles (b, stalls)
int b, stalls;
{
int i;
/* if no more room, split table into two */
if (n_visual_lines >= MAX_VISUAL_LINES)
{
print_block_visualization (b, "(incomplete)");
init_block_visualization ();
}
n_visual_lines++;
sprintf (visual_tbl + strlen (visual_tbl), ";; ");
for (i = 0; i < stalls; i++)
sprintf (visual_tbl + strlen (visual_tbl), ".");
sprintf (visual_tbl + strlen (visual_tbl), "\n");
}
/* move_insn1: Remove INSN from insn chain, and link it after LAST insn */
static rtx
move_insn1 (insn, last)
rtx insn, last;
{
NEXT_INSN (PREV_INSN (insn)) = NEXT_INSN (insn);
PREV_INSN (NEXT_INSN (insn)) = PREV_INSN (insn);
NEXT_INSN (insn) = NEXT_INSN (last);
PREV_INSN (NEXT_INSN (last)) = insn;
NEXT_INSN (last) = insn;
PREV_INSN (insn) = last;
return insn;
}
/* Search INSN for fake REG_DEAD note pairs for NOTE_INSN_SETJMP,
NOTE_INSN_{LOOP,EHREGION}_{BEG,END}; and convert them back into
NOTEs. The REG_DEAD note following first one is contains the saved
value for NOTE_BLOCK_NUMBER which is useful for
NOTE_INSN_EH_REGION_{BEG,END} NOTEs. LAST is the last instruction
output by the instruction scheduler. Return the new value of LAST. */
static rtx
reemit_notes (insn, last)
rtx insn;
rtx last;
{
rtx note, retval;
retval = last;
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
{
if (REG_NOTE_KIND (note) == REG_DEAD
&& GET_CODE (XEXP (note, 0)) == CONST_INT)
{
if (INTVAL (XEXP (note, 0)) == NOTE_INSN_SETJMP)
{
retval = emit_note_after (INTVAL (XEXP (note, 0)), insn);
CONST_CALL_P (retval) = CONST_CALL_P (note);
remove_note (insn, note);
note = XEXP (note, 1);
}
else
{
last = emit_note_before (INTVAL (XEXP (note, 0)), last);
remove_note (insn, note);
note = XEXP (note, 1);
NOTE_BLOCK_NUMBER (last) = INTVAL (XEXP (note, 0));
}
remove_note (insn, note);
}
}
return retval;
}
/* Move INSN, and all insns which should be issued before it,
due to SCHED_GROUP_P flag. Reemit notes if needed.
Return the last insn emitted by the scheduler, which is the
return value from the first call to reemit_notes. */
static rtx
move_insn (insn, last)
rtx insn, last;
{
rtx retval = NULL;
/* If INSN has SCHED_GROUP_P set, then issue it and any other
insns with SCHED_GROUP_P set first. */
while (SCHED_GROUP_P (insn))
{
rtx prev = PREV_INSN (insn);
/* Move a SCHED_GROUP_P insn. */
move_insn1 (insn, last);
/* If this is the first call to reemit_notes, then record
its return value. */
if (retval == NULL_RTX)
retval = reemit_notes (insn, insn);
else
reemit_notes (insn, insn);
insn = prev;
}
/* Now move the first non SCHED_GROUP_P insn. */
move_insn1 (insn, last);
/* If this is the first call to reemit_notes, then record
its return value. */
if (retval == NULL_RTX)
retval = reemit_notes (insn, insn);
else
reemit_notes (insn, insn);
return retval;
}
/* Return an insn which represents a SCHED_GROUP, which is
the last insn in the group. */
static rtx
group_leader (insn)
rtx insn;
{
rtx prev;
do
{
prev = insn;
insn = next_nonnote_insn (insn);
}
while (insn && SCHED_GROUP_P (insn) && (GET_CODE (insn) != CODE_LABEL));
return prev;
}
/* Use forward list scheduling to rearrange insns of block BB in region RGN,
possibly bringing insns from subsequent blocks in the same region.
Return number of insns scheduled. */
static int
schedule_block (bb, rgn, rgn_n_insns)
int bb;
int rgn;
int rgn_n_insns;
{
/* Local variables. */
rtx insn, last;
rtx *ready;
int i;
int n_ready = 0;
int can_issue_more;
/* flow block of this bb */
int b = BB_TO_BLOCK (bb);
/* target_n_insns == number of insns in b before scheduling starts.
sched_target_n_insns == how many of b's insns were scheduled.
sched_n_insns == how many insns were scheduled in b */
int target_n_insns = 0;
int sched_target_n_insns = 0;
int sched_n_insns = 0;
#define NEED_NOTHING 0
#define NEED_HEAD 1
#define NEED_TAIL 2
int new_needs;
/* head/tail info for this block */
rtx prev_head;
rtx next_tail;
rtx head;
rtx tail;
int bb_src;
/* We used to have code to avoid getting parameters moved from hard
argument registers into pseudos.
However, it was removed when it proved to be of marginal benefit
and caused problems because schedule_block and compute_forward_dependences
had different notions of what the "head" insn was. */
get_block_head_tail (bb, &head, &tail);
/* Interblock scheduling could have moved the original head insn from this
block into a proceeding block. This may also cause schedule_block and
compute_forward_dependences to have different notions of what the
"head" insn was.
If the interblock movement happened to make this block start with
some notes (LOOP, EH or SETJMP) before the first real insn, then
HEAD will have various special notes attached to it which must be
removed so that we don't end up with extra copies of the notes. */
if (GET_RTX_CLASS (GET_CODE (head)) == 'i')
{
rtx note;
for (note = REG_NOTES (head); note; note = XEXP (note, 1))
if (REG_NOTE_KIND (note) == REG_DEAD
&& GET_CODE (XEXP (note, 0)) == CONST_INT)
remove_note (head, note);
}
next_tail = NEXT_INSN (tail);
prev_head = PREV_INSN (head);
/* If the only insn left is a NOTE or a CODE_LABEL, then there is no need
to schedule this block. */
if (head == tail
&& (GET_RTX_CLASS (GET_CODE (head)) != 'i'))
return (sched_n_insns);
/* debug info */
if (sched_verbose)
{
fprintf (dump, ";; ======================================================\n");
fprintf (dump,
";; -- basic block %d from %d to %d -- %s reload\n",
b, INSN_UID (basic_block_head[b]),
INSN_UID (basic_block_end[b]),
(reload_completed ? "after" : "before"));
fprintf (dump, ";; ======================================================\n");
if (sched_debug_count >= 0)
fprintf (dump, ";;\t -- sched_debug_count=%d\n", sched_debug_count);
fprintf (dump, "\n");
visual_tbl = (char *) alloca (get_visual_tbl_length ());
init_block_visualization ();
}
/* remove remaining note insns from the block, save them in
note_list. These notes are restored at the end of
schedule_block (). */
note_list = 0;
rm_other_notes (head, tail);
target_bb = bb;
/* prepare current target block info */
if (current_nr_blocks > 1)
{
candidate_table = (candidate *) alloca (current_nr_blocks * sizeof (candidate));
bblst_last = 0;
/* ??? It is not clear why bblst_size is computed this way. The original
number was clearly too small as it resulted in compiler failures.
Multiplying by the original number by 2 (to account for update_bbs
members) seems to be a reasonable solution. */
/* ??? Or perhaps there is a bug somewhere else in this file? */
bblst_size = (current_nr_blocks - bb) * rgn_nr_edges * 2;
bblst_table = (int *) alloca (bblst_size * sizeof (int));
bitlst_table_last = 0;
bitlst_table_size = rgn_nr_edges;
bitlst_table = (int *) alloca (rgn_nr_edges * sizeof (int));
compute_trg_info (bb);
}
clear_units ();
/* Allocate the ready list */
ready = (rtx *) alloca ((rgn_n_insns + 1) * sizeof (rtx));
/* Print debugging information. */
if (sched_verbose >= 5)
debug_dependencies ();
/* Initialize ready list with all 'ready' insns in target block.
Count number of insns in the target block being scheduled. */
n_ready = 0;
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
{
rtx next;
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
continue;
next = NEXT_INSN (insn);
if (INSN_DEP_COUNT (insn) == 0
&& (SCHED_GROUP_P (next) == 0 || GET_RTX_CLASS (GET_CODE (next)) != 'i'))
ready[n_ready++] = insn;
if (!(SCHED_GROUP_P (insn)))
target_n_insns++;
}
/* Add to ready list all 'ready' insns in valid source blocks.
For speculative insns, check-live, exception-free, and
issue-delay. */
for (bb_src = bb + 1; bb_src < current_nr_blocks; bb_src++)
if (IS_VALID (bb_src))
{
rtx src_head;
rtx src_next_tail;
rtx tail, head;
get_block_head_tail (bb_src, &head, &tail);
src_next_tail = NEXT_INSN (tail);
src_head = head;
if (head == tail
&& (GET_RTX_CLASS (GET_CODE (head)) != 'i'))
continue;
for (insn = src_head; insn != src_next_tail; insn = NEXT_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
continue;
if (!CANT_MOVE (insn)
&& (!IS_SPECULATIVE_INSN (insn)
|| (insn_issue_delay (insn) <= 3
&& check_live (insn, bb_src, target_bb)
&& is_exception_free (insn, bb_src, target_bb))))
{
rtx next;
next = NEXT_INSN (insn);
if (INSN_DEP_COUNT (insn) == 0
&& (SCHED_GROUP_P (next) == 0
|| GET_RTX_CLASS (GET_CODE (next)) != 'i'))
ready[n_ready++] = insn;
}
}
}
/* no insns scheduled in this block yet */
last_scheduled_insn = 0;
/* Sort the ready list */
SCHED_SORT (ready, n_ready);
if (sched_verbose >= 2)
{
fprintf (dump, ";;\t\tReady list initially: ");
debug_ready_list (ready, n_ready);
}
/* Q_SIZE is the total number of insns in the queue. */
q_ptr = 0;
q_size = 0;
clock_var = 0;
bzero ((char *) insn_queue, sizeof (insn_queue));
/* We start inserting insns after PREV_HEAD. */
last = prev_head;
/* Initialize INSN_QUEUE, LIST and NEW_NEEDS. */
new_needs = (NEXT_INSN (prev_head) == basic_block_head[b]
? NEED_HEAD : NEED_NOTHING);
if (PREV_INSN (next_tail) == basic_block_end[b])
new_needs |= NEED_TAIL;
/* loop until all the insns in BB are scheduled. */
while (sched_target_n_insns < target_n_insns)
{
int b1;
#ifdef INTERBLOCK_DEBUG
if (sched_debug_count == 0)
break;
#endif
clock_var++;
/* Add to the ready list all pending insns that can be issued now.
If there are no ready insns, increment clock until one
is ready and add all pending insns at that point to the ready
list. */
n_ready = queue_to_ready (ready, n_ready);
if (n_ready == 0)
abort ();
if (sched_verbose >= 2)
{
fprintf (dump, ";;\t\tReady list after queue_to_ready: ");
debug_ready_list (ready, n_ready);
}
/* Sort the ready list. */
SCHED_SORT (ready, n_ready);
if (sched_verbose)
{
fprintf (dump, ";;\tReady list (t =%3d): ", clock_var);
debug_ready_list (ready, n_ready);
}
/* Issue insns from ready list.
It is important to count down from n_ready, because n_ready may change
as insns are issued. */
can_issue_more = issue_rate;
for (i = n_ready - 1; i >= 0 && can_issue_more; i--)
{
rtx insn = ready[i];
int cost = actual_hazard (insn_unit (insn), insn, clock_var, 0);
if (cost > 1)
{
queue_insn (insn, cost);
ready[i] = ready[--n_ready]; /* remove insn from ready list */
}
else if (cost == 0)
{
#ifdef INTERBLOCK_DEBUG
if (sched_debug_count == 0)
break;
#endif
/* an interblock motion? */
if (INSN_BB (insn) != target_bb)
{
rtx temp;
if (IS_SPECULATIVE_INSN (insn))
{
if (!check_live (insn, INSN_BB (insn), target_bb))
{
/* speculative motion, live check failed, remove
insn from ready list */
ready[i] = ready[--n_ready];
continue;
}
update_live (insn, INSN_BB (insn), target_bb);
/* for speculative load, mark insns fed by it. */
if (IS_LOAD_INSN (insn) || FED_BY_SPEC_LOAD (insn))
set_spec_fed (insn);
nr_spec++;
}
nr_inter++;
temp = insn;
while (SCHED_GROUP_P (temp))
temp = PREV_INSN (temp);
/* Update source block boundaries. */
b1 = INSN_BLOCK (temp);
if (temp == basic_block_head[b1]
&& insn == basic_block_end[b1])
{
/* We moved all the insns in the basic block.
Emit a note after the last insn and update the
begin/end boundaries to point to the note. */
emit_note_after (NOTE_INSN_DELETED, insn);
basic_block_end[b1] = NEXT_INSN (insn);
basic_block_head[b1] = NEXT_INSN (insn);
}
else if (insn == basic_block_end[b1])
{
/* We took insns from the end of the basic block,
so update the end of block boundary so that it
points to the first insn we did not move. */
basic_block_end[b1] = PREV_INSN (temp);
}
else if (temp == basic_block_head[b1])
{
/* We took insns from the start of the basic block,
so update the start of block boundary so that
it points to the first insn we did not move. */
basic_block_head[b1] = NEXT_INSN (insn);
}
}
else
{
/* in block motion */
sched_target_n_insns++;
}
last_scheduled_insn = insn;
last = move_insn (insn, last);
sched_n_insns++;
can_issue_more--;
#ifdef INTERBLOCK_DEBUG
if (sched_debug_count > 0)
sched_debug_count--;
#endif
n_ready = schedule_insn (insn, ready, n_ready, clock_var);
/* remove insn from ready list */
ready[i] = ready[--n_ready];
/* close this block after scheduling its jump */
if (GET_CODE (last_scheduled_insn) == JUMP_INSN)
break;
}
}
/* debug info */
if (sched_verbose)
{
visualize_scheduled_insns (b, clock_var);
#ifdef INTERBLOCK_DEBUG
if (sched_debug_count == 0)
fprintf (dump, "........ sched_debug_count == 0 .................\n");
#endif
}
}
/* debug info */
if (sched_verbose)
{
fprintf (dump, ";;\tReady list (final): ");
debug_ready_list (ready, n_ready);
print_block_visualization (b, "");
}
/* Sanity check -- queue must be empty now. Meaningless if region has
multiple bbs, or if scheduling stopped by sched_debug_count. */
if (current_nr_blocks > 1)
#ifdef INTERBLOCK_DEBUG
if (sched_debug_count != 0)
#endif
if (!flag_schedule_interblock && q_size != 0)
abort ();
/* update head/tail boundaries. */
head = NEXT_INSN (prev_head);
tail = last;
#ifdef INTERBLOCK_DEBUG
if (sched_debug_count == 0)
/* compensate for stopping scheduling prematurely */
for (i = sched_target_n_insns; i < target_n_insns; i++)
tail = move_insn (group_leader (NEXT_INSN (tail)), tail);
#endif
/* Restore-other-notes: NOTE_LIST is the end of a chain of notes
previously found among the insns. Insert them at the beginning
of the insns. */
if (note_list != 0)
{
rtx note_head = note_list;
while (PREV_INSN (note_head))
{
note_head = PREV_INSN (note_head);
}
PREV_INSN (note_head) = PREV_INSN (head);
NEXT_INSN (PREV_INSN (head)) = note_head;
PREV_INSN (head) = note_list;
NEXT_INSN (note_list) = head;
head = note_head;
}
/* update target block boundaries. */
if (new_needs & NEED_HEAD)
basic_block_head[b] = head;
if (new_needs & NEED_TAIL)
basic_block_end[b] = tail;
/* debugging */
if (sched_verbose)
{
fprintf (dump, ";; total time = %d\n;; new basic block head = %d\n",
clock_var, INSN_UID (basic_block_head[b]));
fprintf (dump, ";; new basic block end = %d\n\n",
INSN_UID (basic_block_end[b]));
}
return (sched_n_insns);
} /* schedule_block () */
/* print the bit-set of registers, S. callable from debugger */
extern void
debug_reg_vector (s)
regset s;
{
int regno;
EXECUTE_IF_SET_IN_REG_SET (s, 0, regno,
{
fprintf (dump, " %d", regno);
});
fprintf (dump, "\n");
}
/* Use the backward dependences from LOG_LINKS to build
forward dependences in INSN_DEPEND. */
static void
compute_block_forward_dependences (bb)
int bb;
{
rtx insn, link;
rtx tail, head;
rtx next_tail;
enum reg_note dep_type;
get_block_head_tail (bb, &head, &tail);
next_tail = NEXT_INSN (tail);
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
continue;
insn = group_leader (insn);
for (link = LOG_LINKS (insn); link; link = XEXP (link, 1))
{
rtx x = group_leader (XEXP (link, 0));
rtx new_link;
if (x != XEXP (link, 0))
continue;
/* Ignore dependences upon deleted insn */
if (GET_CODE (x) == NOTE || INSN_DELETED_P (x))
continue;
if (find_insn_list (insn, INSN_DEPEND (x)))
continue;
new_link = alloc_INSN_LIST (insn, INSN_DEPEND (x));
dep_type = REG_NOTE_KIND (link);
PUT_REG_NOTE_KIND (new_link, dep_type);
INSN_DEPEND (x) = new_link;
INSN_DEP_COUNT (insn) += 1;
}
}
}
/* Initialize variables for region data dependence analysis.
n_bbs is the number of region blocks */
__inline static void
init_rgn_data_dependences (n_bbs)
int n_bbs;
{
int bb;
/* variables for which one copy exists for each block */
bzero ((char *) bb_pending_read_insns, n_bbs * sizeof (rtx));
bzero ((char *) bb_pending_read_mems, n_bbs * sizeof (rtx));
bzero ((char *) bb_pending_write_insns, n_bbs * sizeof (rtx));
bzero ((char *) bb_pending_write_mems, n_bbs * sizeof (rtx));
bzero ((char *) bb_pending_lists_length, n_bbs * sizeof (rtx));
bzero ((char *) bb_last_pending_memory_flush, n_bbs * sizeof (rtx));
bzero ((char *) bb_last_function_call, n_bbs * sizeof (rtx));
bzero ((char *) bb_sched_before_next_call, n_bbs * sizeof (rtx));
/* Create an insn here so that we can hang dependencies off of it later. */
for (bb = 0; bb < n_bbs; bb++)
{
bb_sched_before_next_call[bb] =
gen_rtx (INSN, VOIDmode, 0, NULL_RTX, NULL_RTX,
NULL_RTX, 0, NULL_RTX, NULL_RTX);
LOG_LINKS (bb_sched_before_next_call[bb]) = 0;
}
}
/* Add dependences so that branches are scheduled to run last in their block */
static void
add_branch_dependences (head, tail)
rtx head, tail;
{
rtx insn, last;
/* For all branches, calls, uses, and cc0 setters, force them to remain
in order at the end of the block by adding dependencies and giving
the last a high priority. There may be notes present, and prev_head
may also be a note.
Branches must obviously remain at the end. Calls should remain at the
end since moving them results in worse register allocation. Uses remain
at the end to ensure proper register allocation. cc0 setters remaim
at the end because they can't be moved away from their cc0 user. */
insn = tail;
last = 0;
while (GET_CODE (insn) == CALL_INSN || GET_CODE (insn) == JUMP_INSN
|| (GET_CODE (insn) == INSN
&& (GET_CODE (PATTERN (insn)) == USE
#ifdef HAVE_cc0
|| sets_cc0_p (PATTERN (insn))
#endif
))
|| GET_CODE (insn) == NOTE)
{
if (GET_CODE (insn) != NOTE)
{
if (last != 0
&& !find_insn_list (insn, LOG_LINKS (last)))
{
add_dependence (last, insn, REG_DEP_ANTI);
INSN_REF_COUNT (insn)++;
}
CANT_MOVE (insn) = 1;
last = insn;
/* Skip over insns that are part of a group.
Make each insn explicitly depend on the previous insn.
This ensures that only the group header will ever enter
the ready queue (and, when scheduled, will automatically
schedule the SCHED_GROUP_P block). */
while (SCHED_GROUP_P (insn))
{
rtx temp = prev_nonnote_insn (insn);
add_dependence (insn, temp, REG_DEP_ANTI);
insn = temp;
}
}
/* Don't overrun the bounds of the basic block. */
if (insn == head)
break;
insn = PREV_INSN (insn);
}
/* make sure these insns are scheduled last in their block */
insn = last;
if (insn != 0)
while (insn != head)
{
insn = prev_nonnote_insn (insn);
if (INSN_REF_COUNT (insn) != 0)
continue;
if (!find_insn_list (last, LOG_LINKS (insn)))
add_dependence (last, insn, REG_DEP_ANTI);
INSN_REF_COUNT (insn) = 1;
/* Skip over insns that are part of a group. */
while (SCHED_GROUP_P (insn))
insn = prev_nonnote_insn (insn);
}
}
/* Compute bacward dependences inside BB. In a multiple blocks region:
(1) a bb is analyzed after its predecessors, and (2) the lists in
effect at the end of bb (after analyzing for bb) are inherited by
bb's successrs.
Specifically for reg-reg data dependences, the block insns are
scanned by sched_analyze () top-to-bottom. Two lists are
naintained by sched_analyze (): reg_last_defs[] for register DEFs,
and reg_last_uses[] for register USEs.
When analysis is completed for bb, we update for its successors:
; - DEFS[succ] = Union (DEFS [succ], DEFS [bb])
; - USES[succ] = Union (USES [succ], DEFS [bb])
The mechanism for computing mem-mem data dependence is very
similar, and the result is interblock dependences in the region. */
static void
compute_block_backward_dependences (bb)
int bb;
{
int b;
rtx x;
rtx head, tail;
int max_reg = max_reg_num ();
b = BB_TO_BLOCK (bb);
if (current_nr_blocks == 1)
{
reg_last_uses = (rtx *) alloca (max_reg * sizeof (rtx));
reg_last_sets = (rtx *) alloca (max_reg * sizeof (rtx));
bzero ((char *) reg_last_uses, max_reg * sizeof (rtx));
bzero ((char *) reg_last_sets, max_reg * sizeof (rtx));
pending_read_insns = 0;
pending_read_mems = 0;
pending_write_insns = 0;
pending_write_mems = 0;
pending_lists_length = 0;
last_function_call = 0;
last_pending_memory_flush = 0;
sched_before_next_call
= gen_rtx (INSN, VOIDmode, 0, NULL_RTX, NULL_RTX,
NULL_RTX, 0, NULL_RTX, NULL_RTX);
LOG_LINKS (sched_before_next_call) = 0;
}
else
{
reg_last_uses = bb_reg_last_uses[bb];
reg_last_sets = bb_reg_last_sets[bb];
pending_read_insns = bb_pending_read_insns[bb];
pending_read_mems = bb_pending_read_mems[bb];
pending_write_insns = bb_pending_write_insns[bb];
pending_write_mems = bb_pending_write_mems[bb];
pending_lists_length = bb_pending_lists_length[bb];
last_function_call = bb_last_function_call[bb];
last_pending_memory_flush = bb_last_pending_memory_flush[bb];
sched_before_next_call = bb_sched_before_next_call[bb];
}
/* do the analysis for this block */
get_block_head_tail (bb, &head, &tail);
sched_analyze (head, tail);
add_branch_dependences (head, tail);
if (current_nr_blocks > 1)
{
int e, first_edge;
int b_succ, bb_succ;
int reg;
rtx link_insn, link_mem;
rtx u;
/* these lists should point to the right place, for correct freeing later. */
bb_pending_read_insns[bb] = pending_read_insns;
bb_pending_read_mems[bb] = pending_read_mems;
bb_pending_write_insns[bb] = pending_write_insns;
bb_pending_write_mems[bb] = pending_write_mems;
/* bb's structures are inherited by it's successors */
first_edge = e = OUT_EDGES (b);
if (e > 0)
do
{
b_succ = TO_BLOCK (e);
bb_succ = BLOCK_TO_BB (b_succ);
/* only bbs "below" bb, in the same region, are interesting */
if (CONTAINING_RGN (b) != CONTAINING_RGN (b_succ)
|| bb_succ <= bb)
{
e = NEXT_OUT (e);
continue;
}
for (reg = 0; reg < max_reg; reg++)
{
/* reg-last-uses lists are inherited by bb_succ */
for (u = reg_last_uses[reg]; u; u = XEXP (u, 1))
{
if (find_insn_list (XEXP (u, 0), (bb_reg_last_uses[bb_succ])[reg]))
continue;
(bb_reg_last_uses[bb_succ])[reg]
= alloc_INSN_LIST (XEXP (u, 0),
(bb_reg_last_uses[bb_succ])[reg]);
}
/* reg-last-defs lists are inherited by bb_succ */
for (u = reg_last_sets[reg]; u; u = XEXP (u, 1))
{
if (find_insn_list (XEXP (u, 0), (bb_reg_last_sets[bb_succ])[reg]))
continue;
(bb_reg_last_sets[bb_succ])[reg]
= alloc_INSN_LIST (XEXP (u, 0),
(bb_reg_last_sets[bb_succ])[reg]);
}
}
/* mem read/write lists are inherited by bb_succ */
link_insn = pending_read_insns;
link_mem = pending_read_mems;
while (link_insn)
{
if (!(find_insn_mem_list (XEXP (link_insn, 0), XEXP (link_mem, 0),
bb_pending_read_insns[bb_succ],
bb_pending_read_mems[bb_succ])))
add_insn_mem_dependence (&bb_pending_read_insns[bb_succ],
&bb_pending_read_mems[bb_succ],
XEXP (link_insn, 0), XEXP (link_mem, 0));
link_insn = XEXP (link_insn, 1);
link_mem = XEXP (link_mem, 1);
}
link_insn = pending_write_insns;
link_mem = pending_write_mems;
while (link_insn)
{
if (!(find_insn_mem_list (XEXP (link_insn, 0), XEXP (link_mem, 0),
bb_pending_write_insns[bb_succ],
bb_pending_write_mems[bb_succ])))
add_insn_mem_dependence (&bb_pending_write_insns[bb_succ],
&bb_pending_write_mems[bb_succ],
XEXP (link_insn, 0), XEXP (link_mem, 0));
link_insn = XEXP (link_insn, 1);
link_mem = XEXP (link_mem, 1);
}
/* last_function_call is inherited by bb_succ */
for (u = last_function_call; u; u = XEXP (u, 1))
{
if (find_insn_list (XEXP (u, 0), bb_last_function_call[bb_succ]))
continue;
bb_last_function_call[bb_succ]
= alloc_INSN_LIST (XEXP (u, 0),
bb_last_function_call[bb_succ]);
}
/* last_pending_memory_flush is inherited by bb_succ */
for (u = last_pending_memory_flush; u; u = XEXP (u, 1))
{
if (find_insn_list (XEXP (u, 0), bb_last_pending_memory_flush[bb_succ]))
continue;
bb_last_pending_memory_flush[bb_succ]
= alloc_INSN_LIST (XEXP (u, 0),
bb_last_pending_memory_flush[bb_succ]);
}
/* sched_before_next_call is inherited by bb_succ */
x = LOG_LINKS (sched_before_next_call);
for (; x; x = XEXP (x, 1))
add_dependence (bb_sched_before_next_call[bb_succ],
XEXP (x, 0), REG_DEP_ANTI);
e = NEXT_OUT (e);
}
while (e != first_edge);
}
/* Free up the INSN_LISTs */
for (b = 0; b < max_reg; ++b)
{
free_list (®_last_sets[b], &unused_insn_list);
free_list (®_last_uses[b], &unused_insn_list);
}
/* Assert that we won't need bb_reg_last_* for this block anymore. */
if (current_nr_blocks > 1)
{
bb_reg_last_uses[bb] = (rtx *) NULL_RTX;
bb_reg_last_sets[bb] = (rtx *) NULL_RTX;
}
}
/* Print dependences for debugging, callable from debugger */
void
debug_dependencies ()
{
int bb;
fprintf (dump, ";; --------------- forward dependences: ------------ \n");
for (bb = 0; bb < current_nr_blocks; bb++)
{
if (1)
{
rtx head, tail;
rtx next_tail;
rtx insn;
get_block_head_tail (bb, &head, &tail);
next_tail = NEXT_INSN (tail);
fprintf (dump, "\n;; --- Region Dependences --- b %d bb %d \n",
BB_TO_BLOCK (bb), bb);
fprintf (dump, ";; %7s%6s%6s%6s%6s%6s%11s%6s\n",
"insn", "code", "bb", "dep", "prio", "cost", "blockage", "units");
fprintf (dump, ";; %7s%6s%6s%6s%6s%6s%11s%6s\n",
"----", "----", "--", "---", "----", "----", "--------", "-----");
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
{
rtx link;
int unit, range;
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
{
int n;
fprintf (dump, ";; %6d ", INSN_UID (insn));
if (GET_CODE (insn) == NOTE)
{
n = NOTE_LINE_NUMBER (insn);
if (n < 0)
fprintf (dump, "%s\n", GET_NOTE_INSN_NAME (n));
else
fprintf (dump, "line %d, file %s\n", n,
NOTE_SOURCE_FILE (insn));
}
else
fprintf (dump, " {%s}\n", GET_RTX_NAME (GET_CODE (insn)));
continue;
}
unit = insn_unit (insn);
range = (unit < 0
|| function_units[unit].blockage_range_function == 0) ? 0 :
function_units[unit].blockage_range_function (insn);
fprintf (dump,
";; %s%5d%6d%6d%6d%6d%6d %3d -%3d ",
(SCHED_GROUP_P (insn) ? "+" : " "),
INSN_UID (insn),
INSN_CODE (insn),
INSN_BB (insn),
INSN_DEP_COUNT (insn),
INSN_PRIORITY (insn),
insn_cost (insn, 0, 0),
(int) MIN_BLOCKAGE_COST (range),
(int) MAX_BLOCKAGE_COST (range));
insn_print_units (insn);
fprintf (dump, "\t: ");
for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1))
fprintf (dump, "%d ", INSN_UID (XEXP (link, 0)));
fprintf (dump, "\n");
}
}
}
fprintf (dump, "\n");
}
/* Set_priorities: compute priority of each insn in the block */
static int
set_priorities (bb)
int bb;
{
rtx insn;
int n_insn;
rtx tail;
rtx prev_head;
rtx head;
get_block_head_tail (bb, &head, &tail);
prev_head = PREV_INSN (head);
if (head == tail
&& (GET_RTX_CLASS (GET_CODE (head)) != 'i'))
return 0;
n_insn = 0;
for (insn = tail; insn != prev_head; insn = PREV_INSN (insn))
{
if (GET_CODE (insn) == NOTE)
continue;
if (!(SCHED_GROUP_P (insn)))
n_insn++;
(void) priority (insn);
}
return n_insn;
}
/* Make each element of VECTOR point at an rtx-vector,
taking the space for all those rtx-vectors from SPACE.
SPACE is of type (rtx *), but it is really as long as NELTS rtx-vectors.
BYTES_PER_ELT is the number of bytes in one rtx-vector.
(this is the same as init_regset_vector () in flow.c) */
static void
init_rtx_vector (vector, space, nelts, bytes_per_elt)
rtx **vector;
rtx *space;
int nelts;
int bytes_per_elt;
{
register int i;
register rtx *p = space;
for (i = 0; i < nelts; i++)
{
vector[i] = p;
p += bytes_per_elt / sizeof (*p);
}
}
/* Schedule a region. A region is either an inner loop, a loop-free
subroutine, or a single basic block. Each bb in the region is
scheduled after its flow predecessors. */
static void
schedule_region (rgn)
int rgn;
{
int bb;
int rgn_n_insns = 0;
int sched_rgn_n_insns = 0;
/* set variables for the current region */
current_nr_blocks = RGN_NR_BLOCKS (rgn);
current_blocks = RGN_BLOCKS (rgn);
reg_pending_sets = ALLOCA_REG_SET ();
reg_pending_sets_all = 0;
/* initializations for region data dependence analyisis */
if (current_nr_blocks > 1)
{
rtx *space;
int maxreg = max_reg_num ();
bb_reg_last_uses = (rtx **) alloca (current_nr_blocks * sizeof (rtx *));
space = (rtx *) alloca (current_nr_blocks * maxreg * sizeof (rtx));
bzero ((char *) space, current_nr_blocks * maxreg * sizeof (rtx));
init_rtx_vector (bb_reg_last_uses, space, current_nr_blocks, maxreg * sizeof (rtx *));
bb_reg_last_sets = (rtx **) alloca (current_nr_blocks * sizeof (rtx *));
space = (rtx *) alloca (current_nr_blocks * maxreg * sizeof (rtx));
bzero ((char *) space, current_nr_blocks * maxreg * sizeof (rtx));
init_rtx_vector (bb_reg_last_sets, space, current_nr_blocks, maxreg * sizeof (rtx *));
bb_pending_read_insns = (rtx *) alloca (current_nr_blocks * sizeof (rtx));
bb_pending_read_mems = (rtx *) alloca (current_nr_blocks * sizeof (rtx));
bb_pending_write_insns = (rtx *) alloca (current_nr_blocks * sizeof (rtx));
bb_pending_write_mems = (rtx *) alloca (current_nr_blocks * sizeof (rtx));
bb_pending_lists_length = (int *) alloca (current_nr_blocks * sizeof (int));
bb_last_pending_memory_flush = (rtx *) alloca (current_nr_blocks * sizeof (rtx));
bb_last_function_call = (rtx *) alloca (current_nr_blocks * sizeof (rtx));
bb_sched_before_next_call = (rtx *) alloca (current_nr_blocks * sizeof (rtx));
init_rgn_data_dependences (current_nr_blocks);
}
/* compute LOG_LINKS */
for (bb = 0; bb < current_nr_blocks; bb++)
compute_block_backward_dependences (bb);
/* compute INSN_DEPEND */
for (bb = current_nr_blocks - 1; bb >= 0; bb--)
compute_block_forward_dependences (bb);
/* Delete line notes, compute live-regs at block end, and set priorities. */
dead_notes = 0;
for (bb = 0; bb < current_nr_blocks; bb++)
{
if (reload_completed == 0)
find_pre_sched_live (bb);
if (write_symbols != NO_DEBUG)
{
save_line_notes (bb);
rm_line_notes (bb);
}
rgn_n_insns += set_priorities (bb);
}
/* compute interblock info: probabilities, split-edges, dominators, etc. */
if (current_nr_blocks > 1)
{
int i;
prob = (float *) alloca ((current_nr_blocks) * sizeof (float));
bbset_size = current_nr_blocks / HOST_BITS_PER_WIDE_INT + 1;
dom = (bbset *) alloca (current_nr_blocks * sizeof (bbset));
for (i = 0; i < current_nr_blocks; i++)
{
dom[i] = (bbset) alloca (bbset_size * sizeof (HOST_WIDE_INT));
bzero ((char *) dom[i], bbset_size * sizeof (HOST_WIDE_INT));
}
/* edge to bit */
rgn_nr_edges = 0;
edge_to_bit = (int *) alloca (nr_edges * sizeof (int));
for (i = 1; i < nr_edges; i++)
if (CONTAINING_RGN (FROM_BLOCK (i)) == rgn)
EDGE_TO_BIT (i) = rgn_nr_edges++;
rgn_edges = (int *) alloca (rgn_nr_edges * sizeof (int));
rgn_nr_edges = 0;
for (i = 1; i < nr_edges; i++)
if (CONTAINING_RGN (FROM_BLOCK (i)) == (rgn))
rgn_edges[rgn_nr_edges++] = i;
/* split edges */
edgeset_size = rgn_nr_edges / HOST_BITS_PER_WIDE_INT + 1;
pot_split = (edgeset *) alloca (current_nr_blocks * sizeof (edgeset));
ancestor_edges = (edgeset *) alloca (current_nr_blocks * sizeof (edgeset));
for (i = 0; i < current_nr_blocks; i++)
{
pot_split[i] =
(edgeset) alloca (edgeset_size * sizeof (HOST_WIDE_INT));
bzero ((char *) pot_split[i],
edgeset_size * sizeof (HOST_WIDE_INT));
ancestor_edges[i] =
(edgeset) alloca (edgeset_size * sizeof (HOST_WIDE_INT));
bzero ((char *) ancestor_edges[i],
edgeset_size * sizeof (HOST_WIDE_INT));
}
/* compute probabilities, dominators, split_edges */
for (bb = 0; bb < current_nr_blocks; bb++)
compute_dom_prob_ps (bb);
}
/* now we can schedule all blocks */
for (bb = 0; bb < current_nr_blocks; bb++)
{
sched_rgn_n_insns += schedule_block (bb, rgn, rgn_n_insns);
#ifdef USE_C_ALLOCA
alloca (0);
#endif
}
#ifdef INTERBLOCK_DEBUG
if (sched_debug_count != 0)
#endif
/* sanity check: verify that all region insns were scheduled */
if (sched_rgn_n_insns != rgn_n_insns)
abort ();
/* update register life and usage information */
if (reload_completed == 0)
{
for (bb = current_nr_blocks - 1; bb >= 0; bb--)
find_post_sched_live (bb);
if (current_nr_blocks <= 1)
/* Sanity check. There should be no REG_DEAD notes leftover at the end.
In practice, this can occur as the result of bugs in flow, combine.c,
and/or sched.c. The values of the REG_DEAD notes remaining are
meaningless, because dead_notes is just used as a free list. */
if (dead_notes != 0)
abort ();
}
/* restore line notes. */
if (write_symbols != NO_DEBUG)
{
for (bb = 0; bb < current_nr_blocks; bb++)
restore_line_notes (bb);
}
/* Done with this region */
free_pending_lists ();
FREE_REG_SET (reg_pending_sets);
}
/* Subroutine of split_hard_reg_notes. Searches X for any reference to
REGNO, returning the rtx of the reference found if any. Otherwise,
returns 0. */
static rtx
regno_use_in (regno, x)
int regno;
rtx x;
{
register char *fmt;
int i, j;
rtx tem;
if (GET_CODE (x) == REG && REGNO (x) == regno)
return x;
fmt = GET_RTX_FORMAT (GET_CODE (x));
for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0; i--)
{
if (fmt[i] == 'e')
{
if ((tem = regno_use_in (regno, XEXP (x, i))))
return tem;
}
else if (fmt[i] == 'E')
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
if ((tem = regno_use_in (regno, XVECEXP (x, i, j))))
return tem;
}
return 0;
}
/* Subroutine of update_flow_info. Determines whether any new REG_NOTEs are
needed for the hard register mentioned in the note. This can happen
if the reference to the hard register in the original insn was split into
several smaller hard register references in the split insns. */
static void
split_hard_reg_notes (note, first, last, orig_insn)
rtx note, first, last, orig_insn;
{
rtx reg, temp, link;
int n_regs, i, new_reg;
rtx insn;
/* Assume that this is a REG_DEAD note. */
if (REG_NOTE_KIND (note) != REG_DEAD)
abort ();
reg = XEXP (note, 0);
n_regs = HARD_REGNO_NREGS (REGNO (reg), GET_MODE (reg));
for (i = 0; i < n_regs; i++)
{
new_reg = REGNO (reg) + i;
/* Check for references to new_reg in the split insns. */
for (insn = last;; insn = PREV_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i'
&& (temp = regno_use_in (new_reg, PATTERN (insn))))
{
/* Create a new reg dead note ere. */
link = alloc_EXPR_LIST (REG_DEAD, temp, REG_NOTES (insn));
REG_NOTES (insn) = link;
/* If killed multiple registers here, then add in the excess. */
i += HARD_REGNO_NREGS (REGNO (temp), GET_MODE (temp)) - 1;
break;
}
/* It isn't mentioned anywhere, so no new reg note is needed for
this register. */
if (insn == first)
break;
}
}
}
/* Subroutine of update_flow_info. Determines whether a SET or CLOBBER in an
insn created by splitting needs a REG_DEAD or REG_UNUSED note added. */
static void
new_insn_dead_notes (pat, insn, last, orig_insn)
rtx pat, insn, last, orig_insn;
{
rtx dest, tem, set;
/* PAT is either a CLOBBER or a SET here. */
dest = XEXP (pat, 0);
while (GET_CODE (dest) == ZERO_EXTRACT || GET_CODE (dest) == SUBREG
|| GET_CODE (dest) == STRICT_LOW_PART
|| GET_CODE (dest) == SIGN_EXTRACT)
dest = XEXP (dest, 0);
if (GET_CODE (dest) == REG)
{
for (tem = last; tem != insn; tem = PREV_INSN (tem))
{
if (GET_RTX_CLASS (GET_CODE (tem)) == 'i'
&& reg_overlap_mentioned_p (dest, PATTERN (tem))
&& (set = single_set (tem)))
{
rtx tem_dest = SET_DEST (set);
while (GET_CODE (tem_dest) == ZERO_EXTRACT
|| GET_CODE (tem_dest) == SUBREG
|| GET_CODE (tem_dest) == STRICT_LOW_PART
|| GET_CODE (tem_dest) == SIGN_EXTRACT)
tem_dest = XEXP (tem_dest, 0);
if (!rtx_equal_p (tem_dest, dest))
{
/* Use the same scheme as combine.c, don't put both REG_DEAD
and REG_UNUSED notes on the same insn. */
if (!find_regno_note (tem, REG_UNUSED, REGNO (dest))
&& !find_regno_note (tem, REG_DEAD, REGNO (dest)))
{
rtx note = alloc_EXPR_LIST (REG_DEAD, dest,
REG_NOTES (tem));
REG_NOTES (tem) = note;
}
/* The reg only dies in one insn, the last one that uses
it. */
break;
}
else if (reg_overlap_mentioned_p (dest, SET_SRC (set)))
/* We found an instruction that both uses the register,
and sets it, so no new REG_NOTE is needed for this set. */
break;
}
}
/* If this is a set, it must die somewhere, unless it is the dest of
the original insn, and hence is live after the original insn. Abort
if it isn't supposed to be live after the original insn.
If this is a clobber, then just add a REG_UNUSED note. */
if (tem == insn)
{
int live_after_orig_insn = 0;
rtx pattern = PATTERN (orig_insn);
int i;
if (GET_CODE (pat) == CLOBBER)
{
rtx note = alloc_EXPR_LIST (REG_UNUSED, dest, REG_NOTES (insn));
REG_NOTES (insn) = note;
return;
}
/* The original insn could have multiple sets, so search the
insn for all sets. */
if (GET_CODE (pattern) == SET)
{
if (reg_overlap_mentioned_p (dest, SET_DEST (pattern)))
live_after_orig_insn = 1;
}
else if (GET_CODE (pattern) == PARALLEL)
{
for (i = 0; i < XVECLEN (pattern, 0); i++)
if (GET_CODE (XVECEXP (pattern, 0, i)) == SET
&& reg_overlap_mentioned_p (dest,
SET_DEST (XVECEXP (pattern,
0, i))))
live_after_orig_insn = 1;
}
if (!live_after_orig_insn)
abort ();
}
}
}
/* Subroutine of update_flow_info. Update the value of reg_n_sets for all
registers modified by X. INC is -1 if the containing insn is being deleted,
and is 1 if the containing insn is a newly generated insn. */
static void
update_n_sets (x, inc)
rtx x;
int inc;
{
rtx dest = SET_DEST (x);
while (GET_CODE (dest) == STRICT_LOW_PART || GET_CODE (dest) == SUBREG
|| GET_CODE (dest) == ZERO_EXTRACT || GET_CODE (dest) == SIGN_EXTRACT)
dest = SUBREG_REG (dest);
if (GET_CODE (dest) == REG)
{
int regno = REGNO (dest);
if (regno < FIRST_PSEUDO_REGISTER)
{
register int i;
int endregno = regno + HARD_REGNO_NREGS (regno, GET_MODE (dest));
for (i = regno; i < endregno; i++)
REG_N_SETS (i) += inc;
}
else
REG_N_SETS (regno) += inc;
}
}
/* Updates all flow-analysis related quantities (including REG_NOTES) for
the insns from FIRST to LAST inclusive that were created by splitting
ORIG_INSN. NOTES are the original REG_NOTES. */
static void
update_flow_info (notes, first, last, orig_insn)
rtx notes;
rtx first, last;
rtx orig_insn;
{
rtx insn, note;
rtx next;
rtx orig_dest, temp;
rtx set;
/* Get and save the destination set by the original insn. */
orig_dest = single_set (orig_insn);
if (orig_dest)
orig_dest = SET_DEST (orig_dest);
/* Move REG_NOTES from the original insn to where they now belong. */
for (note = notes; note; note = next)
{
next = XEXP (note, 1);
switch (REG_NOTE_KIND (note))
{
case REG_DEAD:
case REG_UNUSED:
/* Move these notes from the original insn to the last new insn where
the register is now set. */
for (insn = last;; insn = PREV_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i'
&& reg_mentioned_p (XEXP (note, 0), PATTERN (insn)))
{
/* If this note refers to a multiple word hard register, it
may have been split into several smaller hard register
references, so handle it specially. */
temp = XEXP (note, 0);
if (REG_NOTE_KIND (note) == REG_DEAD
&& GET_CODE (temp) == REG
&& REGNO (temp) < FIRST_PSEUDO_REGISTER
&& HARD_REGNO_NREGS (REGNO (temp), GET_MODE (temp)) > 1)
split_hard_reg_notes (note, first, last, orig_insn);
else
{
XEXP (note, 1) = REG_NOTES (insn);
REG_NOTES (insn) = note;
}
/* Sometimes need to convert REG_UNUSED notes to REG_DEAD
notes. */
/* ??? This won't handle multiple word registers correctly,
but should be good enough for now. */
if (REG_NOTE_KIND (note) == REG_UNUSED
&& GET_CODE (XEXP (note, 0)) != SCRATCH
&& !dead_or_set_p (insn, XEXP (note, 0)))
PUT_REG_NOTE_KIND (note, REG_DEAD);
/* The reg only dies in one insn, the last one that uses
it. */
break;
}
/* It must die somewhere, fail it we couldn't find where it died.
If this is a REG_UNUSED note, then it must be a temporary
register that was not needed by this instantiation of the
pattern, so we can safely ignore it. */
if (insn == first)
{
/* After reload, REG_DEAD notes come sometimes an
instruction after the register actually dies. */
if (reload_completed && REG_NOTE_KIND (note) == REG_DEAD)
{
XEXP (note, 1) = REG_NOTES (insn);
REG_NOTES (insn) = note;
break;
}
if (REG_NOTE_KIND (note) != REG_UNUSED)
abort ();
break;
}
}
break;
case REG_WAS_0:
/* This note applies to the dest of the original insn. Find the
first new insn that now has the same dest, and move the note
there. */
if (!orig_dest)
abort ();
for (insn = first;; insn = NEXT_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i'
&& (temp = single_set (insn))
&& rtx_equal_p (SET_DEST (temp), orig_dest))
{
XEXP (note, 1) = REG_NOTES (insn);
REG_NOTES (insn) = note;
/* The reg is only zero before one insn, the first that
uses it. */
break;
}
/* If this note refers to a multiple word hard
register, it may have been split into several smaller
hard register references. We could split the notes,
but simply dropping them is good enough. */
if (GET_CODE (orig_dest) == REG
&& REGNO (orig_dest) < FIRST_PSEUDO_REGISTER
&& HARD_REGNO_NREGS (REGNO (orig_dest),
GET_MODE (orig_dest)) > 1)
break;
/* It must be set somewhere, fail if we couldn't find where it
was set. */
if (insn == last)
abort ();
}
break;
case REG_EQUAL:
case REG_EQUIV:
/* A REG_EQUIV or REG_EQUAL note on an insn with more than one
set is meaningless. Just drop the note. */
if (!orig_dest)
break;
case REG_NO_CONFLICT:
/* These notes apply to the dest of the original insn. Find the last
new insn that now has the same dest, and move the note there. */
if (!orig_dest)
abort ();
for (insn = last;; insn = PREV_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i'
&& (temp = single_set (insn))
&& rtx_equal_p (SET_DEST (temp), orig_dest))
{
XEXP (note, 1) = REG_NOTES (insn);
REG_NOTES (insn) = note;
/* Only put this note on one of the new insns. */
break;
}
/* The original dest must still be set someplace. Abort if we
couldn't find it. */
if (insn == first)
{
/* However, if this note refers to a multiple word hard
register, it may have been split into several smaller
hard register references. We could split the notes,
but simply dropping them is good enough. */
if (GET_CODE (orig_dest) == REG
&& REGNO (orig_dest) < FIRST_PSEUDO_REGISTER
&& HARD_REGNO_NREGS (REGNO (orig_dest),
GET_MODE (orig_dest)) > 1)
break;
/* Likewise for multi-word memory references. */
if (GET_CODE (orig_dest) == MEM
&& SIZE_FOR_MODE (orig_dest) > MOVE_MAX)
break;
abort ();
}
}
break;
case REG_LIBCALL:
/* Move a REG_LIBCALL note to the first insn created, and update
the corresponding REG_RETVAL note. */
XEXP (note, 1) = REG_NOTES (first);
REG_NOTES (first) = note;
insn = XEXP (note, 0);
note = find_reg_note (insn, REG_RETVAL, NULL_RTX);
if (note)
XEXP (note, 0) = first;
break;
case REG_EXEC_COUNT:
/* Move a REG_EXEC_COUNT note to the first insn created. */
XEXP (note, 1) = REG_NOTES (first);
REG_NOTES (first) = note;
break;
case REG_RETVAL:
/* Move a REG_RETVAL note to the last insn created, and update
the corresponding REG_LIBCALL note. */
XEXP (note, 1) = REG_NOTES (last);
REG_NOTES (last) = note;
insn = XEXP (note, 0);
note = find_reg_note (insn, REG_LIBCALL, NULL_RTX);
if (note)
XEXP (note, 0) = last;
break;
case REG_NONNEG:
case REG_BR_PROB:
/* This should be moved to whichever instruction is a JUMP_INSN. */
for (insn = last;; insn = PREV_INSN (insn))
{
if (GET_CODE (insn) == JUMP_INSN)
{
XEXP (note, 1) = REG_NOTES (insn);
REG_NOTES (insn) = note;
/* Only put this note on one of the new insns. */
break;
}
/* Fail if we couldn't find a JUMP_INSN. */
if (insn == first)
abort ();
}
break;
case REG_INC:
/* reload sometimes leaves obsolete REG_INC notes around. */
if (reload_completed)
break;
/* This should be moved to whichever instruction now has the
increment operation. */
abort ();
case REG_LABEL:
/* Should be moved to the new insn(s) which use the label. */
for (insn = first; insn != NEXT_INSN (last); insn = NEXT_INSN (insn))
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i'
&& reg_mentioned_p (XEXP (note, 0), PATTERN (insn)))
{
REG_NOTES (insn) = alloc_EXPR_LIST (REG_LABEL,
XEXP (note, 0),
REG_NOTES (insn));
}
break;
case REG_CC_SETTER:
case REG_CC_USER:
/* These two notes will never appear until after reorg, so we don't
have to handle them here. */
default:
abort ();
}
}
/* Each new insn created, except the last, has a new set. If the destination
is a register, then this reg is now live across several insns, whereas
previously the dest reg was born and died within the same insn. To
reflect this, we now need a REG_DEAD note on the insn where this
dest reg dies.
Similarly, the new insns may have clobbers that need REG_UNUSED notes. */
for (insn = first; insn != last; insn = NEXT_INSN (insn))
{
rtx pat;
int i;
pat = PATTERN (insn);
if (GET_CODE (pat) == SET || GET_CODE (pat) == CLOBBER)
new_insn_dead_notes (pat, insn, last, orig_insn);
else if (GET_CODE (pat) == PARALLEL)
{
for (i = 0; i < XVECLEN (pat, 0); i++)
if (GET_CODE (XVECEXP (pat, 0, i)) == SET
|| GET_CODE (XVECEXP (pat, 0, i)) == CLOBBER)
new_insn_dead_notes (XVECEXP (pat, 0, i), insn, last, orig_insn);
}
}
/* If any insn, except the last, uses the register set by the last insn,
then we need a new REG_DEAD note on that insn. In this case, there
would not have been a REG_DEAD note for this register in the original
insn because it was used and set within one insn. */
set = single_set (last);
if (set)
{
rtx dest = SET_DEST (set);
while (GET_CODE (dest) == ZERO_EXTRACT || GET_CODE (dest) == SUBREG
|| GET_CODE (dest) == STRICT_LOW_PART
|| GET_CODE (dest) == SIGN_EXTRACT)
dest = XEXP (dest, 0);
if (GET_CODE (dest) == REG
/* Global registers are always live, so the code below does not
apply to them. */
&& (REGNO (dest) >= FIRST_PSEUDO_REGISTER
|| ! global_regs[REGNO (dest)]))
{
rtx stop_insn = PREV_INSN (first);
/* If the last insn uses the register that it is setting, then
we don't want to put a REG_DEAD note there. Search backwards
to find the first insn that sets but does not use DEST. */
insn = last;
if (reg_overlap_mentioned_p (dest, SET_SRC (set)))
{
for (insn = PREV_INSN (insn); insn != first;
insn = PREV_INSN (insn))
{
if ((set = single_set (insn))
&& reg_mentioned_p (dest, SET_DEST (set))
&& ! reg_overlap_mentioned_p (dest, SET_SRC (set)))
break;
}
}
/* Now find the first insn that uses but does not set DEST. */
for (insn = PREV_INSN (insn); insn != stop_insn;
insn = PREV_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i'
&& reg_mentioned_p (dest, PATTERN (insn))
&& (set = single_set (insn)))
{
rtx insn_dest = SET_DEST (set);
while (GET_CODE (insn_dest) == ZERO_EXTRACT
|| GET_CODE (insn_dest) == SUBREG
|| GET_CODE (insn_dest) == STRICT_LOW_PART
|| GET_CODE (insn_dest) == SIGN_EXTRACT)
insn_dest = XEXP (insn_dest, 0);
if (insn_dest != dest)
{
note = alloc_EXPR_LIST (REG_DEAD, dest, REG_NOTES (insn));
REG_NOTES (insn) = note;
/* The reg only dies in one insn, the last one
that uses it. */
break;
}
}
}
}
}
/* If the original dest is modifying a multiple register target, and the
original instruction was split such that the original dest is now set
by two or more SUBREG sets, then the split insns no longer kill the
destination of the original insn.
In this case, if there exists an instruction in the same basic block,
before the split insn, which uses the original dest, and this use is
killed by the original insn, then we must remove the REG_DEAD note on
this insn, because it is now superfluous.
This does not apply when a hard register gets split, because the code
knows how to handle overlapping hard registers properly. */
if (orig_dest && GET_CODE (orig_dest) == REG)
{
int found_orig_dest = 0;
int found_split_dest = 0;
for (insn = first;; insn = NEXT_INSN (insn))
{
rtx pat;
int i;
/* I'm not sure if this can happen, but let's be safe. */
if (GET_RTX_CLASS (GET_CODE (insn)) != 'i')
continue;
pat = PATTERN (insn);
i = GET_CODE (pat) == PARALLEL ? XVECLEN (pat, 0) : 0;
set = pat;
for (;;)
{
if (GET_CODE (set) == SET)
{
if (GET_CODE (SET_DEST (set)) == REG
&& REGNO (SET_DEST (set)) == REGNO (orig_dest))
{
found_orig_dest = 1;
break;
}
else if (GET_CODE (SET_DEST (set)) == SUBREG
&& SUBREG_REG (SET_DEST (set)) == orig_dest)
{
found_split_dest = 1;
break;
}
}
if (--i < 0)
break;
set = XVECEXP (pat, 0, i);
}
if (insn == last)
break;
}
if (found_split_dest)
{
/* Search backwards from FIRST, looking for the first insn that uses
the original dest. Stop if we pass a CODE_LABEL or a JUMP_INSN.
If we find an insn, and it has a REG_DEAD note, then delete the
note. */
for (insn = first; insn; insn = PREV_INSN (insn))
{
if (GET_CODE (insn) == CODE_LABEL
|| GET_CODE (insn) == JUMP_INSN)
break;
else if (GET_RTX_CLASS (GET_CODE (insn)) == 'i'
&& reg_mentioned_p (orig_dest, insn))
{
note = find_regno_note (insn, REG_DEAD, REGNO (orig_dest));
if (note)
remove_note (insn, note);
}
}
}
else if (!found_orig_dest)
{
/* This should never happen. */
abort ();
}
}
/* Update reg_n_sets. This is necessary to prevent local alloc from
converting REG_EQUAL notes to REG_EQUIV when splitting has modified
a reg from set once to set multiple times. */
{
rtx x = PATTERN (orig_insn);
RTX_CODE code = GET_CODE (x);
if (code == SET || code == CLOBBER)
update_n_sets (x, -1);
else if (code == PARALLEL)
{
int i;
for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
{
code = GET_CODE (XVECEXP (x, 0, i));
if (code == SET || code == CLOBBER)
update_n_sets (XVECEXP (x, 0, i), -1);
}
}
for (insn = first;; insn = NEXT_INSN (insn))
{
x = PATTERN (insn);
code = GET_CODE (x);
if (code == SET || code == CLOBBER)
update_n_sets (x, 1);
else if (code == PARALLEL)
{
int i;
for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
{
code = GET_CODE (XVECEXP (x, 0, i));
if (code == SET || code == CLOBBER)
update_n_sets (XVECEXP (x, 0, i), 1);
}
}
if (insn == last)
break;
}
}
}
/* Do the splitting of insns in the block b. */
static void
split_block_insns (b)
int b;
{
rtx insn, next;
for (insn = basic_block_head[b];; insn = next)
{
rtx prev;
rtx set;
/* Can't use `next_real_insn' because that
might go across CODE_LABELS and short-out basic blocks. */
next = NEXT_INSN (insn);
if (GET_CODE (insn) != INSN)
{
if (insn == basic_block_end[b])
break;
continue;
}
/* Don't split no-op move insns. These should silently disappear
later in final. Splitting such insns would break the code
that handles REG_NO_CONFLICT blocks. */
set = single_set (insn);
if (set && rtx_equal_p (SET_SRC (set), SET_DEST (set)))
{
if (insn == basic_block_end[b])
break;
/* Nops get in the way while scheduling, so delete them now if
register allocation has already been done. It is too risky
to try to do this before register allocation, and there are
unlikely to be very many nops then anyways. */
if (reload_completed)
{
PUT_CODE (insn, NOTE);
NOTE_LINE_NUMBER (insn) = NOTE_INSN_DELETED;
NOTE_SOURCE_FILE (insn) = 0;
}
continue;
}
/* Split insns here to get max fine-grain parallelism. */
prev = PREV_INSN (insn);
/* It is probably not worthwhile to try to split again in
the second pass. However, if flag_schedule_insns is not set,
the first and only (if any) scheduling pass is after reload. */
if (reload_completed == 0 || ! flag_schedule_insns)
{
rtx last, first = PREV_INSN (insn);
rtx notes = REG_NOTES (insn);
last = try_split (PATTERN (insn), insn, 1);
if (last != insn)
{
/* try_split returns the NOTE that INSN became. */
first = NEXT_INSN (first);
update_flow_info (notes, first, last, insn);
PUT_CODE (insn, NOTE);
NOTE_SOURCE_FILE (insn) = 0;
NOTE_LINE_NUMBER (insn) = NOTE_INSN_DELETED;
if (insn == basic_block_head[b])
basic_block_head[b] = first;
if (insn == basic_block_end[b])
{
basic_block_end[b] = last;
break;
}
}
}
if (insn == basic_block_end[b])
break;
}
}
/* The one entry point in this file. DUMP_FILE is the dump file for
this pass. */
void
schedule_insns (dump_file)
FILE *dump_file;
{
int max_uid;
int b;
int i;
rtx insn;
int rgn;
int luid;
/* disable speculative loads in their presence if cc0 defined */
#ifdef HAVE_cc0
flag_schedule_speculative_load = 0;
#endif
/* Taking care of this degenerate case makes the rest of
this code simpler. */
if (n_basic_blocks == 0)
return;
/* set dump and sched_verbose for the desired debugging output. If no
dump-file was specified, but -fsched-verbose-N (any N), print to stderr.
For -fsched-verbose-N, N>=10, print everything to stderr. */
sched_verbose = sched_verbose_param;
if (sched_verbose_param == 0 && dump_file)
sched_verbose = 1;
dump = ((sched_verbose_param >= 10 || !dump_file) ? stderr : dump_file);
nr_inter = 0;
nr_spec = 0;
/* Initialize the unused_*_lists. We can't use the ones left over from
the previous function, because gcc has freed that memory. We can use
the ones left over from the first sched pass in the second pass however,
so only clear them on the first sched pass. The first pass is before
reload if flag_schedule_insns is set, otherwise it is afterwards. */
if (reload_completed == 0 || !flag_schedule_insns)
{
unused_insn_list = 0;
unused_expr_list = 0;
}
/* initialize issue_rate */
issue_rate = ISSUE_RATE;
/* do the splitting first for all blocks */
for (b = 0; b < n_basic_blocks; b++)
split_block_insns (b);
max_uid = (get_max_uid () + 1);
cant_move = (char *) alloca (max_uid * sizeof (char));
bzero ((char *) cant_move, max_uid * sizeof (char));
fed_by_spec_load = (char *) alloca (max_uid * sizeof (char));
bzero ((char *) fed_by_spec_load, max_uid * sizeof (char));
is_load_insn = (char *) alloca (max_uid * sizeof (char));
bzero ((char *) is_load_insn, max_uid * sizeof (char));
insn_orig_block = (int *) alloca (max_uid * sizeof (int));
insn_luid = (int *) alloca (max_uid * sizeof (int));
luid = 0;
for (b = 0; b < n_basic_blocks; b++)
for (insn = basic_block_head[b];; insn = NEXT_INSN (insn))
{
INSN_BLOCK (insn) = b;
INSN_LUID (insn) = luid++;
if (insn == basic_block_end[b])
break;
}
/* after reload, remove inter-blocks dependences computed before reload. */
if (reload_completed)
{
int b;
rtx insn;
for (b = 0; b < n_basic_blocks; b++)
for (insn = basic_block_head[b];; insn = NEXT_INSN (insn))
{
rtx link, prev;
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i')
{
prev = NULL_RTX;
link = LOG_LINKS (insn);
while (link)
{
rtx x = XEXP (link, 0);
if (INSN_BLOCK (x) != b)
{
remove_dependence (insn, x);
link = prev ? XEXP (prev, 1) : LOG_LINKS (insn);
}
else
prev = link, link = XEXP (prev, 1);
}
}
if (insn == basic_block_end[b])
break;
}
}
nr_regions = 0;
rgn_table = (region *) alloca ((n_basic_blocks) * sizeof (region));
rgn_bb_table = (int *) alloca ((n_basic_blocks) * sizeof (int));
block_to_bb = (int *) alloca ((n_basic_blocks) * sizeof (int));
containing_rgn = (int *) alloca ((n_basic_blocks) * sizeof (int));
/* compute regions for scheduling */
if (reload_completed
|| n_basic_blocks == 1
|| !flag_schedule_interblock)
{
find_single_block_region ();
}
else
{
/* an estimation for nr_edges is computed in is_cfg_nonregular () */
nr_edges = 0;
/* verify that a 'good' control flow graph can be built */
if (is_cfg_nonregular ()
|| nr_edges <= 1)
{
find_single_block_region ();
}
else
{
/* build control flow graph */
in_edges = (int *) alloca (n_basic_blocks * sizeof (int));
out_edges = (int *) alloca (n_basic_blocks * sizeof (int));
bzero ((char *) in_edges, n_basic_blocks * sizeof (int));
bzero ((char *) out_edges, n_basic_blocks * sizeof (int));
edge_table =
(edge *) alloca ((nr_edges) * sizeof (edge));
bzero ((char *) edge_table,
((nr_edges) * sizeof (edge)));
build_control_flow ();
/* identify reducible inner loops and compute regions */
find_rgns ();
if (sched_verbose >= 3)
{
debug_control_flow ();
debug_regions ();
}
}
}
/* Allocate data for this pass. See comments, above,
for what these vectors do. */
insn_priority = (int *) alloca (max_uid * sizeof (int));
insn_reg_weight = (int *) alloca (max_uid * sizeof (int));
insn_tick = (int *) alloca (max_uid * sizeof (int));
insn_costs = (short *) alloca (max_uid * sizeof (short));
insn_units = (short *) alloca (max_uid * sizeof (short));
insn_blockage = (unsigned int *) alloca (max_uid * sizeof (unsigned int));
insn_ref_count = (int *) alloca (max_uid * sizeof (int));
/* Allocate for forward dependencies */
insn_dep_count = (int *) alloca (max_uid * sizeof (int));
insn_depend = (rtx *) alloca (max_uid * sizeof (rtx));
if (reload_completed == 0)
{
int i;
sched_reg_n_calls_crossed = (int *) alloca (max_regno * sizeof (int));
sched_reg_live_length = (int *) alloca (max_regno * sizeof (int));
sched_reg_basic_block = (int *) alloca (max_regno * sizeof (int));
bb_live_regs = ALLOCA_REG_SET ();
bzero ((char *) sched_reg_n_calls_crossed, max_regno * sizeof (int));
bzero ((char *) sched_reg_live_length, max_regno * sizeof (int));
for (i = 0; i < max_regno; i++)
sched_reg_basic_block[i] = REG_BLOCK_UNKNOWN;
}
else
{
sched_reg_n_calls_crossed = 0;
sched_reg_live_length = 0;
bb_live_regs = 0;
}
init_alias_analysis ();
if (write_symbols != NO_DEBUG)
{
rtx line;
line_note = (rtx *) alloca (max_uid * sizeof (rtx));
bzero ((char *) line_note, max_uid * sizeof (rtx));
line_note_head = (rtx *) alloca (n_basic_blocks * sizeof (rtx));
bzero ((char *) line_note_head, n_basic_blocks * sizeof (rtx));
/* Save-line-note-head:
Determine the line-number at the start of each basic block.
This must be computed and saved now, because after a basic block's
predecessor has been scheduled, it is impossible to accurately
determine the correct line number for the first insn of the block. */
for (b = 0; b < n_basic_blocks; b++)
for (line = basic_block_head[b]; line; line = PREV_INSN (line))
if (GET_CODE (line) == NOTE && NOTE_LINE_NUMBER (line) > 0)
{
line_note_head[b] = line;
break;
}
}
bzero ((char *) insn_priority, max_uid * sizeof (int));
bzero ((char *) insn_reg_weight, max_uid * sizeof (int));
bzero ((char *) insn_tick, max_uid * sizeof (int));
bzero ((char *) insn_costs, max_uid * sizeof (short));
bzero ((char *) insn_units, max_uid * sizeof (short));
bzero ((char *) insn_blockage, max_uid * sizeof (unsigned int));
bzero ((char *) insn_ref_count, max_uid * sizeof (int));
/* Initialize for forward dependencies */
bzero ((char *) insn_depend, max_uid * sizeof (rtx));
bzero ((char *) insn_dep_count, max_uid * sizeof (int));
/* Find units used in this fuction, for visualization */
if (sched_verbose)
init_target_units ();
/* ??? Add a NOTE after the last insn of the last basic block. It is not
known why this is done. */
insn = basic_block_end[n_basic_blocks - 1];
if (NEXT_INSN (insn) == 0
|| (GET_CODE (insn) != NOTE
&& GET_CODE (insn) != CODE_LABEL
/* Don't emit a NOTE if it would end up between an unconditional
jump and a BARRIER. */
&& !(GET_CODE (insn) == JUMP_INSN
&& GET_CODE (NEXT_INSN (insn)) == BARRIER)))
emit_note_after (NOTE_INSN_DELETED, basic_block_end[n_basic_blocks - 1]);
/* Schedule every region in the subroutine */
for (rgn = 0; rgn < nr_regions; rgn++)
{
schedule_region (rgn);
#ifdef USE_C_ALLOCA
alloca (0);
#endif
}
/* Reposition the prologue and epilogue notes in case we moved the
prologue/epilogue insns. */
if (reload_completed)
reposition_prologue_and_epilogue_notes (get_insns ());
/* delete redundant line notes. */
if (write_symbols != NO_DEBUG)
rm_redundant_line_notes ();
/* Update information about uses of registers in the subroutine. */
if (reload_completed == 0)
update_reg_usage ();
if (sched_verbose)
{
if (reload_completed == 0 && flag_schedule_interblock)
{
fprintf (dump, "\n;; Procedure interblock/speculative motions == %d/%d \n",
nr_inter, nr_spec);
}
else
{
if (nr_inter > 0)
abort ();
}
fprintf (dump, "\n\n");
}
if (bb_live_regs)
FREE_REG_SET (bb_live_regs);
}
#endif /* INSN_SCHEDULING */
|