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
|
/* Instruction scheduling pass.
Copyright (C) 1992, 93-97, 1998 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 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 after flow analysis, but before register allocation. 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 reverse 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. Among the remaining insns on
the ready list to be considered, the first one with the largest
potential for causing a subsequent blockage is chosen.
The following list shows the order in which we want to break ties
among insns in the ready list:
1. choose insn with lowest conflict cost, ties broken by
2. choose insn with the longest path to end of bb, ties broken by
3. choose insn that kills the most registers, ties broken by
4. choose insn that conflicts with the most ready insns, or finally
5. 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.
Dependencies set up by memory references are treated in exactly the
same way as other dependencies, by using LOG_LINKS.
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 do any branch scheduling. 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.
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 that have been scheduled. */
#include "config.h"
#include <stdio.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"
#ifdef INSN_SCHEDULING
/* 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;
/* 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 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_dead_regs;
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 following 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 LOG_LINKS 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 `schedule_select' 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 at the top of the scheduling
loop in `schedule_block' 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 rtx canon_rtx PROTO((rtx));
static int rtx_equal_for_memref_p PROTO((rtx, rtx));
static rtx find_symbolic_term PROTO((rtx));
static int memrefs_conflict_p PROTO((int, rtx, int, rtx,
HOST_WIDE_INT));
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 void prepare_unit PROTO((int));
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 int 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 birthing_insn_p PROTO((rtx));
static void adjust_priority PROTO((rtx));
static int schedule_insn PROTO((rtx, rtx *, int, int));
static int schedule_select PROTO((rtx *, int, int, FILE *));
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 rtx unlink_notes PROTO((rtx, rtx));
static int new_sometimes_live PROTO((struct sometimes *, int, int));
static void finish_sometimes_live PROTO((struct sometimes *, int));
static rtx reemit_notes PROTO((rtx, rtx));
static void schedule_block PROTO((int, FILE *));
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 *));
#endif /* INSN_SCHEDULING */
#define SIZE_FOR_MODE(X) (GET_MODE_SIZE (GET_MODE (X)))
/* Vector indexed by N giving the initial (unchanging) value known
for pseudo-register N. */
static rtx *reg_known_value;
/* Vector recording for each reg_known_value whether it is due to a
REG_EQUIV note. Future passes (viz., reload) may replace the
pseudo with the equivalent expression and so we account for the
dependences that would be introduced if that happens. */
/* ??? This is a problem only on the Convex. The REG_EQUIV notes created in
assign_parms mention the arg pointer, and there are explicit insns in the
RTL that modify the arg pointer. Thus we must ensure that such insns don't
get scheduled across each other because that would invalidate the REG_EQUIV
notes. One could argue that the REG_EQUIV notes are wrong, but solving
the problem in the scheduler will likely give better code, so we do it
here. */
static char *reg_known_equiv_p;
/* Indicates number of valid entries in reg_known_value. */
static int reg_known_value_size;
static rtx
canon_rtx (x)
rtx x;
{
/* Recursively look for equivalences. */
if (GET_CODE (x) == REG && REGNO (x) >= FIRST_PSEUDO_REGISTER
&& REGNO (x) <= reg_known_value_size)
return reg_known_value[REGNO (x)] == x
? x : canon_rtx (reg_known_value[REGNO (x)]);
else if (GET_CODE (x) == PLUS)
{
rtx x0 = canon_rtx (XEXP (x, 0));
rtx x1 = canon_rtx (XEXP (x, 1));
if (x0 != XEXP (x, 0) || x1 != XEXP (x, 1))
{
/* We can tolerate LO_SUMs being offset here; these
rtl are used for nothing other than comparisons. */
if (GET_CODE (x0) == CONST_INT)
return plus_constant_for_output (x1, INTVAL (x0));
else if (GET_CODE (x1) == CONST_INT)
return plus_constant_for_output (x0, INTVAL (x1));
return gen_rtx (PLUS, GET_MODE (x), x0, x1);
}
}
/* This gives us much better alias analysis when called from
the loop optimizer. Note we want to leave the original
MEM alone, but need to return the canonicalized MEM with
all the flags with their original values. */
else if (GET_CODE (x) == MEM)
{
rtx copy = copy_rtx (x);
XEXP (copy, 0) = canon_rtx (XEXP (copy, 0));
x = copy;
}
return x;
}
/* Set up all info needed to perform alias analysis on memory references. */
void
init_alias_analysis ()
{
int maxreg = max_reg_num ();
rtx insn;
rtx note;
rtx set;
reg_known_value_size = maxreg;
reg_known_value
= (rtx *) oballoc ((maxreg-FIRST_PSEUDO_REGISTER) * sizeof (rtx))
- FIRST_PSEUDO_REGISTER;
bzero ((char *) (reg_known_value + FIRST_PSEUDO_REGISTER),
(maxreg-FIRST_PSEUDO_REGISTER) * sizeof (rtx));
reg_known_equiv_p
= (char *) oballoc ((maxreg -FIRST_PSEUDO_REGISTER) * sizeof (char))
- FIRST_PSEUDO_REGISTER;
bzero (reg_known_equiv_p + FIRST_PSEUDO_REGISTER,
(maxreg - FIRST_PSEUDO_REGISTER) * sizeof (char));
/* Fill in the entries with known constant values. */
for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
if ((set = single_set (insn)) != 0
&& GET_CODE (SET_DEST (set)) == REG
&& REGNO (SET_DEST (set)) >= FIRST_PSEUDO_REGISTER
&& (((note = find_reg_note (insn, REG_EQUAL, 0)) != 0
&& REG_N_SETS (REGNO (SET_DEST (set))) == 1)
|| (note = find_reg_note (insn, REG_EQUIV, NULL_RTX)) != 0)
&& GET_CODE (XEXP (note, 0)) != EXPR_LIST)
{
int regno = REGNO (SET_DEST (set));
reg_known_value[regno] = XEXP (note, 0);
reg_known_equiv_p[regno] = REG_NOTE_KIND (note) == REG_EQUIV;
}
/* Fill in the remaining entries. */
while (--maxreg >= FIRST_PSEUDO_REGISTER)
if (reg_known_value[maxreg] == 0)
reg_known_value[maxreg] = regno_reg_rtx[maxreg];
}
/* Return 1 if X and Y are identical-looking rtx's.
We use the data in reg_known_value above to see if two registers with
different numbers are, in fact, equivalent. */
static int
rtx_equal_for_memref_p (x, y)
rtx x, y;
{
register int i;
register int j;
register enum rtx_code code;
register char *fmt;
if (x == 0 && y == 0)
return 1;
if (x == 0 || y == 0)
return 0;
x = canon_rtx (x);
y = canon_rtx (y);
if (x == y)
return 1;
code = GET_CODE (x);
/* Rtx's of different codes cannot be equal. */
if (code != GET_CODE (y))
return 0;
/* (MULT:SI x y) and (MULT:HI x y) are NOT equivalent.
(REG:SI x) and (REG:HI x) are NOT equivalent. */
if (GET_MODE (x) != GET_MODE (y))
return 0;
/* REG, LABEL_REF, and SYMBOL_REF can be compared nonrecursively. */
if (code == REG)
return REGNO (x) == REGNO (y);
if (code == LABEL_REF)
return XEXP (x, 0) == XEXP (y, 0);
if (code == SYMBOL_REF)
return XSTR (x, 0) == XSTR (y, 0);
/* For commutative operations, the RTX match if the operand match in any
order. Also handle the simple binary and unary cases without a loop. */
if (code == EQ || code == NE || GET_RTX_CLASS (code) == 'c')
return ((rtx_equal_for_memref_p (XEXP (x, 0), XEXP (y, 0))
&& rtx_equal_for_memref_p (XEXP (x, 1), XEXP (y, 1)))
|| (rtx_equal_for_memref_p (XEXP (x, 0), XEXP (y, 1))
&& rtx_equal_for_memref_p (XEXP (x, 1), XEXP (y, 0))));
else if (GET_RTX_CLASS (code) == '<' || GET_RTX_CLASS (code) == '2')
return (rtx_equal_for_memref_p (XEXP (x, 0), XEXP (y, 0))
&& rtx_equal_for_memref_p (XEXP (x, 1), XEXP (y, 1)));
else if (GET_RTX_CLASS (code) == '1')
return rtx_equal_for_memref_p (XEXP (x, 0), XEXP (y, 0));
/* Compare the elements. If any pair of corresponding elements
fail to match, return 0 for the whole things. */
fmt = GET_RTX_FORMAT (code);
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
{
switch (fmt[i])
{
case 'w':
if (XWINT (x, i) != XWINT (y, i))
return 0;
break;
case 'n':
case 'i':
if (XINT (x, i) != XINT (y, i))
return 0;
break;
case 'V':
case 'E':
/* Two vectors must have the same length. */
if (XVECLEN (x, i) != XVECLEN (y, i))
return 0;
/* And the corresponding elements must match. */
for (j = 0; j < XVECLEN (x, i); j++)
if (rtx_equal_for_memref_p (XVECEXP (x, i, j), XVECEXP (y, i, j)) == 0)
return 0;
break;
case 'e':
if (rtx_equal_for_memref_p (XEXP (x, i), XEXP (y, i)) == 0)
return 0;
break;
case 'S':
case 's':
if (strcmp (XSTR (x, i), XSTR (y, i)))
return 0;
break;
case 'u':
/* These are just backpointers, so they don't matter. */
break;
case '0':
break;
/* It is believed that rtx's at this level will never
contain anything but integers and other rtx's,
except for within LABEL_REFs and SYMBOL_REFs. */
default:
abort ();
}
}
return 1;
}
/* Given an rtx X, find a SYMBOL_REF or LABEL_REF within
X and return it, or return 0 if none found. */
static rtx
find_symbolic_term (x)
rtx x;
{
register int i;
register enum rtx_code code;
register char *fmt;
code = GET_CODE (x);
if (code == SYMBOL_REF || code == LABEL_REF)
return x;
if (GET_RTX_CLASS (code) == 'o')
return 0;
fmt = GET_RTX_FORMAT (code);
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
{
rtx t;
if (fmt[i] == 'e')
{
t = find_symbolic_term (XEXP (x, i));
if (t != 0)
return t;
}
else if (fmt[i] == 'E')
break;
}
return 0;
}
/* Return nonzero if X and Y (memory addresses) could reference the
same location in memory. C is an offset accumulator. When
C is nonzero, we are testing aliases between X and Y + C.
XSIZE is the size in bytes of the X reference,
similarly YSIZE is the size in bytes for Y.
If XSIZE or YSIZE is zero, we do not know the amount of memory being
referenced (the reference was BLKmode), so make the most pessimistic
assumptions.
We recognize the following cases of non-conflicting memory:
(1) addresses involving the frame pointer cannot conflict
with addresses involving static variables.
(2) static variables with different addresses cannot conflict.
Nice to notice that varying addresses cannot conflict with fp if no
local variables had their addresses taken, but that's too hard now. */
/* ??? In Fortran, references to a array parameter can never conflict with
another array parameter. */
static int
memrefs_conflict_p (xsize, x, ysize, y, c)
rtx x, y;
int xsize, ysize;
HOST_WIDE_INT c;
{
if (GET_CODE (x) == HIGH)
x = XEXP (x, 0);
else if (GET_CODE (x) == LO_SUM)
x = XEXP (x, 1);
else
x = canon_rtx (x);
if (GET_CODE (y) == HIGH)
y = XEXP (y, 0);
else if (GET_CODE (y) == LO_SUM)
y = XEXP (y, 1);
else
y = canon_rtx (y);
if (rtx_equal_for_memref_p (x, y))
return (xsize == 0 || ysize == 0
|| (c >= 0 && xsize > c) || (c < 0 && ysize+c > 0));
if (y == frame_pointer_rtx || y == hard_frame_pointer_rtx
|| y == stack_pointer_rtx)
{
rtx t = y;
int tsize = ysize;
y = x; ysize = xsize;
x = t; xsize = tsize;
}
if (x == frame_pointer_rtx || x == hard_frame_pointer_rtx
|| x == stack_pointer_rtx)
{
rtx y1;
if (CONSTANT_P (y))
return 0;
if (GET_CODE (y) == PLUS
&& canon_rtx (XEXP (y, 0)) == x
&& (y1 = canon_rtx (XEXP (y, 1)))
&& GET_CODE (y1) == CONST_INT)
{
c += INTVAL (y1);
return (xsize == 0 || ysize == 0
|| (c >= 0 && xsize > c) || (c < 0 && ysize+c > 0));
}
if (GET_CODE (y) == PLUS
&& (y1 = canon_rtx (XEXP (y, 0)))
&& CONSTANT_P (y1))
return 0;
return 1;
}
if (GET_CODE (x) == PLUS)
{
/* The fact that X is canonicalized means that this
PLUS rtx is canonicalized. */
rtx x0 = XEXP (x, 0);
rtx x1 = XEXP (x, 1);
if (GET_CODE (y) == PLUS)
{
/* The fact that Y is canonicalized means that this
PLUS rtx is canonicalized. */
rtx y0 = XEXP (y, 0);
rtx y1 = XEXP (y, 1);
if (rtx_equal_for_memref_p (x1, y1))
return memrefs_conflict_p (xsize, x0, ysize, y0, c);
if (rtx_equal_for_memref_p (x0, y0))
return memrefs_conflict_p (xsize, x1, ysize, y1, c);
if (GET_CODE (x1) == CONST_INT)
if (GET_CODE (y1) == CONST_INT)
return memrefs_conflict_p (xsize, x0, ysize, y0,
c - INTVAL (x1) + INTVAL (y1));
else
return memrefs_conflict_p (xsize, x0, ysize, y, c - INTVAL (x1));
else if (GET_CODE (y1) == CONST_INT)
return memrefs_conflict_p (xsize, x, ysize, y0, c + INTVAL (y1));
/* Handle case where we cannot understand iteration operators,
but we notice that the base addresses are distinct objects. */
x = find_symbolic_term (x);
if (x == 0)
return 1;
y = find_symbolic_term (y);
if (y == 0)
return 1;
return rtx_equal_for_memref_p (x, y);
}
else if (GET_CODE (x1) == CONST_INT)
return memrefs_conflict_p (xsize, x0, ysize, y, c - INTVAL (x1));
}
else if (GET_CODE (y) == PLUS)
{
/* The fact that Y is canonicalized means that this
PLUS rtx is canonicalized. */
rtx y0 = XEXP (y, 0);
rtx y1 = XEXP (y, 1);
if (GET_CODE (y1) == CONST_INT)
return memrefs_conflict_p (xsize, x, ysize, y0, c + INTVAL (y1));
else
return 1;
}
if (GET_CODE (x) == GET_CODE (y) && GET_CODE (x) == MULT)
{
/* Handle cases where we expect the second operands to be the
same, and check only whether the first operand would conflict
or not. */
rtx x0, y0;
rtx x1 = canon_rtx (XEXP (x, 1));
rtx y1 = canon_rtx (XEXP (y, 1));
if (! rtx_equal_for_memref_p (x1, y1))
return 1;
x0 = canon_rtx (XEXP (x, 0));
y0 = canon_rtx (XEXP (y, 0));
if (rtx_equal_for_memref_p (x0, y0))
return (xsize == 0 || ysize == 0
|| (c >= 0 && xsize > c) || (c < 0 && ysize+c > 0));
/* Can't properly adjust our sizes. */
if (GET_CODE (x1) != CONST_INT)
return 1;
xsize /= INTVAL (x1);
ysize /= INTVAL (x1);
c /= INTVAL (x1);
return memrefs_conflict_p (xsize, x0, ysize, y0, c);
}
if (CONSTANT_P (x))
{
if (GET_CODE (x) == CONST_INT && GET_CODE (y) == CONST_INT)
{
c += (INTVAL (y) - INTVAL (x));
return (xsize == 0 || ysize == 0
|| (c >= 0 && xsize > c) || (c < 0 && ysize+c > 0));
}
if (GET_CODE (x) == CONST)
{
if (GET_CODE (y) == CONST)
return memrefs_conflict_p (xsize, canon_rtx (XEXP (x, 0)),
ysize, canon_rtx (XEXP (y, 0)), c);
else
return memrefs_conflict_p (xsize, canon_rtx (XEXP (x, 0)),
ysize, y, c);
}
if (GET_CODE (y) == CONST)
return memrefs_conflict_p (xsize, x, ysize,
canon_rtx (XEXP (y, 0)), c);
if (CONSTANT_P (y))
return (rtx_equal_for_memref_p (x, y)
&& (xsize == 0 || ysize == 0
|| (c >= 0 && xsize > c) || (c < 0 && ysize+c > 0)));
return 1;
}
return 1;
}
/* Functions to compute memory dependencies.
Since we process the insns in execution order, we can build tables
to keep track of what registers are fixed (and not aliased), what registers
are varying in known ways, and what registers are varying in unknown
ways.
If both memory references are volatile, then there must always be a
dependence between the two references, since their order can not be
changed. A volatile and non-volatile reference can be interchanged
though.
A MEM_IN_STRUCT reference at a non-QImode non-AND varying address can never
conflict with a non-MEM_IN_STRUCT reference at a fixed address. We must
allow QImode aliasing because the ANSI C standard allows character
pointers to alias anything. We are assuming that characters are
always QImode here. We also must allow AND addresses, because they may
generate accesses outside the object being referenced. This is used to
generate aligned addresses from unaligned addresses, for instance, the
alpha storeqi_unaligned pattern. */
/* Read dependence: X is read after read in MEM takes place. There can
only be a dependence here if both reads are volatile. */
int
read_dependence (mem, x)
rtx mem;
rtx x;
{
return MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem);
}
/* True dependence: X is read after store in MEM takes place. */
int
true_dependence (mem, x)
rtx mem;
rtx x;
{
/* If X is an unchanging read, then it can't possibly conflict with any
non-unchanging store. It may conflict with an unchanging write though,
because there may be a single store to this address to initialize it.
Just fall through to the code below to resolve the case where we have
both an unchanging read and an unchanging write. This won't handle all
cases optimally, but the possible performance loss should be
negligible. */
x = canon_rtx (x);
mem = canon_rtx (mem);
if (RTX_UNCHANGING_P (x) && ! RTX_UNCHANGING_P (mem))
return 0;
return ((MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem))
|| (memrefs_conflict_p (SIZE_FOR_MODE (mem), XEXP (mem, 0),
SIZE_FOR_MODE (x), XEXP (x, 0), 0)
&& ! (MEM_IN_STRUCT_P (mem) && rtx_addr_varies_p (mem)
&& GET_MODE (mem) != QImode
&& GET_CODE (XEXP (mem, 0)) != AND
&& ! MEM_IN_STRUCT_P (x) && ! rtx_addr_varies_p (x))
&& ! (MEM_IN_STRUCT_P (x) && rtx_addr_varies_p (x)
&& GET_MODE (x) != QImode
&& GET_CODE (XEXP (x, 0)) != AND
&& ! MEM_IN_STRUCT_P (mem) && ! rtx_addr_varies_p (mem))));
}
/* Anti dependence: X is written after read in MEM takes place. */
int
anti_dependence (mem, x)
rtx mem;
rtx x;
{
/* If MEM is an unchanging read, then it can't possibly conflict with
the store to X, because there is at most one store to MEM, and it must
have occurred somewhere before MEM. */
x = canon_rtx (x);
mem = canon_rtx (mem);
if (RTX_UNCHANGING_P (mem))
return 0;
return ((MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem))
|| (memrefs_conflict_p (SIZE_FOR_MODE (mem), XEXP (mem, 0),
SIZE_FOR_MODE (x), XEXP (x, 0), 0)
&& ! (MEM_IN_STRUCT_P (mem) && rtx_addr_varies_p (mem)
&& GET_MODE (mem) != QImode
&& GET_CODE (XEXP (mem, 0)) != AND
&& ! MEM_IN_STRUCT_P (x) && ! rtx_addr_varies_p (x))
&& ! (MEM_IN_STRUCT_P (x) && rtx_addr_varies_p (x)
&& GET_MODE (x) != QImode
&& GET_CODE (XEXP (x, 0)) != AND
&& ! MEM_IN_STRUCT_P (mem) && ! rtx_addr_varies_p (mem))));
}
/* Output dependence: X is written after store in MEM takes place. */
int
output_dependence (mem, x)
rtx mem;
rtx x;
{
x = canon_rtx (x);
mem = canon_rtx (mem);
return ((MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem))
|| (memrefs_conflict_p (SIZE_FOR_MODE (mem), XEXP (mem, 0),
SIZE_FOR_MODE (x), XEXP (x, 0), 0)
&& ! (MEM_IN_STRUCT_P (mem) && rtx_addr_varies_p (mem)
&& GET_MODE (mem) != QImode
&& GET_CODE (XEXP (mem, 0)) != AND
&& ! MEM_IN_STRUCT_P (x) && ! rtx_addr_varies_p (x))
&& ! (MEM_IN_STRUCT_P (x) && rtx_addr_varies_p (x)
&& GET_MODE (x) != QImode
&& GET_CODE (XEXP (x, 0)) != AND
&& ! MEM_IN_STRUCT_P (mem) && ! rtx_addr_varies_p (mem))));
}
/* Helper functions for instruction scheduling. */
/* 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;
}
/* 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 = rtx_alloc (INSN_LIST);
/* Insn dependency, not data dependency. */
PUT_REG_NOTE_KIND (link, dep_type);
XEXP (link, 0) = elem;
XEXP (link, 1) = LOG_LINKS (insn);
LOG_LINKS (insn) = link;
}
/* 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;
int found = 0;
for (prev = 0, link = LOG_LINKS (insn); link; link = XEXP (link, 1))
{
if (XEXP (link, 0) == elem)
{
RTX_INTEGRATED_P (link) = 1;
if (prev)
XEXP (prev, 1) = XEXP (link, 1);
else
LOG_LINKS (insn) = XEXP (link, 1);
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;
/* 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;
/* 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;
/* 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;
}
/* 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));
}
/* Record an insn as one that will use the units encoded by UNIT. */
__inline static void
prepare_unit (unit)
int unit;
{
int i;
if (unit >= 0)
unit_n_insns[unit]++;
else
for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
if ((unit & 1) != 0)
prepare_unit (i);
}
/* 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];
if (tick - clock > cost)
{
/* The scheduler is operating in reverse, so INSN is the executing
insn and the unit's last 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
(insn, unit_last_insn[instance])
- 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 virtual cycles taken 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;
}
}
/* 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;
{
if (insn && GET_RTX_CLASS (GET_CODE (insn)) == 'i')
{
int prev_priority;
int max_priority;
int this_priority = INSN_PRIORITY (insn);
rtx prev;
if (this_priority > 0)
return this_priority;
max_priority = 1;
/* Nonzero if these insns must be scheduled together. */
if (SCHED_GROUP_P (insn))
{
prev = insn;
while (SCHED_GROUP_P (prev))
{
prev = PREV_INSN (prev);
INSN_REF_COUNT (prev) += 1;
}
}
for (prev = LOG_LINKS (insn); prev; prev = XEXP (prev, 1))
{
rtx x = XEXP (prev, 0);
/* If this was a duplicate of a dependence we already deleted,
ignore it. */
if (RTX_INTEGRATED_P (prev))
continue;
/* A dependence pointing to a note or deleted insn is always
obsolete, because sched_analyze_insn will have created any
necessary new dependences which replace it. Notes and deleted
insns can be created when instructions are deleted by insn
splitting, or by register allocation. */
if (GET_CODE (x) == NOTE || INSN_DELETED_P (x))
{
remove_dependence (insn, x);
continue;
}
/* Clear the link cost adjustment bits. */
LINK_COST_FREE (prev) = 0;
#ifdef ADJUST_COST
LINK_COST_ZERO (prev) = 0;
#endif
/* This priority calculation was chosen because it results in the
least instruction movement, and does not hurt the performance
of the resulting code compared to the old algorithm.
This makes the sched algorithm more stable, which results
in better code, because there is less register pressure,
cross jumping is more likely to work, and debugging is easier.
When all instructions have a latency of 1, there is no need to
move any instructions. Subtracting one here ensures that in such
cases all instructions will end up with a priority of one, and
hence no scheduling will be done.
The original code did not subtract the one, and added the
insn_cost of the current instruction to its priority (e.g.
move the insn_cost call down to the end). */
prev_priority = priority (x) + insn_cost (x, prev, insn) - 1;
if (prev_priority > max_priority)
max_priority = prev_priority;
INSN_REF_COUNT (x) += 1;
}
prepare_unit (insn_unit (insn));
INSN_PRIORITY (insn) = max_priority;
return INSN_PRIORITY (insn);
}
return 0;
}
/* 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 ()
{
register rtx link, prev_link;
if (pending_read_insns)
{
prev_link = pending_read_insns;
link = XEXP (prev_link, 1);
while (link)
{
prev_link = link;
link = XEXP (link, 1);
}
XEXP (prev_link, 1) = unused_insn_list;
unused_insn_list = pending_read_insns;
pending_read_insns = 0;
}
if (pending_write_insns)
{
prev_link = pending_write_insns;
link = XEXP (prev_link, 1);
while (link)
{
prev_link = link;
link = XEXP (link, 1);
}
XEXP (prev_link, 1) = unused_insn_list;
unused_insn_list = pending_write_insns;
pending_write_insns = 0;
}
if (pending_read_mems)
{
prev_link = pending_read_mems;
link = XEXP (prev_link, 1);
while (link)
{
prev_link = link;
link = XEXP (link, 1);
}
XEXP (prev_link, 1) = unused_expr_list;
unused_expr_list = pending_read_mems;
pending_read_mems = 0;
}
if (pending_write_mems)
{
prev_link = pending_write_mems;
link = XEXP (prev_link, 1);
while (link)
{
prev_link = link;
link = XEXP (link, 1);
}
XEXP (prev_link, 1) = unused_expr_list;
unused_expr_list = pending_write_mems;
pending_write_mems = 0;
}
}
/* 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;
if (unused_insn_list)
{
link = unused_insn_list;
unused_insn_list = XEXP (link, 1);
}
else
link = rtx_alloc (INSN_LIST);
XEXP (link, 0) = insn;
XEXP (link, 1) = *insn_list;
*insn_list = link;
if (unused_expr_list)
{
link = unused_expr_list;
unused_expr_list = XEXP (link, 1);
}
else
link = rtx_alloc (EXPR_LIST);
XEXP (link, 0) = mem;
XEXP (link, 1) = *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 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;
if (last_pending_memory_flush)
add_dependence (insn, last_pending_memory_flush, REG_DEP_ANTI);
last_pending_memory_flush = insn;
}
/* 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;
if (reg_last_sets[regno + i])
add_dependence (insn, reg_last_sets[regno + i],
REG_DEP_OUTPUT);
SET_REGNO_REG_SET (reg_pending_sets, regno + i);
if ((call_used_regs[i] || global_regs[i])
&& last_function_call)
/* Function calls clobber all call_used regs. */
add_dependence (insn, last_function_call, 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;
if (reg_last_sets[regno])
add_dependence (insn, reg_last_sets[regno], 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 && last_function_call)
add_dependence (insn, last_function_call, 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 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);
}
if (last_pending_memory_flush)
add_dependence (insn, last_pending_memory_flush, 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:
{
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]
= gen_rtx (INSN_LIST, VOIDmode,
insn, reg_last_uses[regno + i]);
if (reg_last_sets[regno + i])
add_dependence (insn, reg_last_sets[regno + i], 0);
if ((call_used_regs[regno + i] || global_regs[regno + i])
&& last_function_call)
/* Function calls clobber all call_used regs. */
add_dependence (insn, last_function_call, REG_DEP_ANTI);
}
}
else
{
reg_last_uses[regno]
= gen_rtx (INSN_LIST, VOIDmode, insn, reg_last_uses[regno]);
if (reg_last_sets[regno])
add_dependence (insn, reg_last_sets[regno], 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 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), x))
add_dependence (insn, XEXP (pending, 0), 0);
pending = XEXP (pending, 1);
pending_mem = XEXP (pending_mem, 1);
}
if (last_pending_memory_flush)
add_dependence (insn, last_pending_memory_flush, 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;
if (reg_last_sets[i])
add_dependence (insn, reg_last_sets[i], 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;
default:
break;
}
/* 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;
if (reg_last_sets[i])
add_dependence (insn, reg_last_sets[i], 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[i] = insn;
});
CLEAR_REG_SET (reg_pending_sets);
if (reg_pending_sets_all)
{
for (i = 0; i < maxreg; i++)
reg_last_sets[i] = insn;
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 int
sched_analyze (head, tail)
rtx head, tail;
{
register rtx insn;
register int n_insns = 0;
register rtx u;
register int luid = 0;
rtx loop_notes = 0;
for (insn = head; ; insn = NEXT_INSN (insn))
{
INSN_LUID (insn) = luid++;
if (GET_CODE (insn) == INSN || GET_CODE (insn) == JUMP_INSN)
{
sched_analyze_insn (PATTERN (insn), insn, loop_notes);
loop_notes = 0;
n_insns += 1;
}
else if (GET_CODE (insn) == CALL_INSN)
{
rtx x;
register int i;
/* 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;
if (reg_last_sets[i])
add_dependence (insn, reg_last_sets[i], 0);
}
reg_pending_sets_all = 1;
/* Add a pair of fake REG_NOTEs which we will later
convert back into a NOTE_INSN_SETJMP note. See
reemit_notes for why we use a pair of of NOTEs. */
REG_NOTES (insn) = gen_rtx (EXPR_LIST, REG_DEAD,
GEN_INT (0),
REG_NOTES (insn));
REG_NOTES (insn) = gen_rtx (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;
if (reg_last_sets[i])
add_dependence (insn, reg_last_sets[i], 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 = insn;
n_insns += 1;
}
/* 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 = gen_rtx (EXPR_LIST, REG_DEAD,
GEN_INT (NOTE_BLOCK_NUMBER (insn)), loop_notes);
loop_notes = gen_rtx (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 n_insns;
}
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);
SET_REGNO_REG_SET (bb_dead_regs, regno + j);
}
}
else
{
CLEAR_REGNO_REG_SET (bb_live_regs, regno);
SET_REGNO_REG_SET (bb_dead_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);
CLEAR_REGNO_REG_SET (bb_dead_regs, regno + j);
}
}
else
{
SET_REGNO_REG_SET (bb_live_regs, regno);
CLEAR_REGNO_REG_SET (bb_dead_regs, regno);
}
}
}
}
/* Macros and functions for keeping the priority queue sorted, and
dealing with queueing and dequeueing of instructions. */
#define SCHED_SORT(READY, NEW_READY, OLD_READY) \
do { if ((NEW_READY) - (OLD_READY) == 1) \
swap_sort (READY, NEW_READY); \
else if ((NEW_READY) - (OLD_READY) > 1) \
qsort (READY, NEW_READY, sizeof (rtx), rank_for_schedule); } \
while (0)
/* Returns a positive value if y is preferred; returns a negative value if
x 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 value;
/* Choose the instruction with the highest priority, if different. */
if (value = INSN_PRIORITY (tmp) - INSN_PRIORITY (tmp2))
return value;
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, LOG_LINKS (last_scheduled_insn));
if (link == 0 || insn_cost (tmp, link, last_scheduled_insn) == 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, LOG_LINKS (last_scheduled_insn));
if (link == 0 || insn_cost (tmp2, link, last_scheduled_insn) == 1)
tmp2_class = 3;
else if (REG_NOTE_KIND (link) == 0) /* Data dependence. */
tmp2_class = 1;
else
tmp2_class = 2;
if (value = tmp_class - tmp2_class)
return value;
}
/* 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 fires at least N_CYCLES
before the currently executing insn. */
__inline static void
queue_insn (insn, n_cycles)
rtx insn;
int n_cycles;
{
int next_q = NEXT_Q_AFTER (q_ptr, n_cycles);
NEXT_INSN (insn) = insn_queue[next_q];
insn_queue[next_q] = insn;
q_size += 1;
}
/* 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 (in the backwards dataflow sense). READY is a
vector of insns which are ready to fire. N_READY is the number of
elements in READY. CLOCK is the current virtual cycle. */
static int
schedule_insn (insn, ready, n_ready, clock)
rtx insn;
rtx *ready;
int n_ready;
int clock;
{
rtx link;
int new_ready = n_ready;
if (MAX_BLOCKAGE > 1)
schedule_unit (insn_unit (insn), insn, clock);
if (LOG_LINKS (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 = LOG_LINKS (insn); link != 0; link = XEXP (link, 1))
{
rtx prev = XEXP (link, 0);
int cost = insn_cost (prev, link, insn);
if ((INSN_REF_COUNT (prev) -= 1) != 0)
{
/* We satisfied one requirement to fire PREV. Record the earliest
time when PREV can fire. No need to do this if the cost is 1,
because PREV can fire no sooner than the next cycle. */
if (cost > 1)
INSN_TICK (prev) = MAX (INSN_TICK (prev), clock + cost);
}
else
{
/* We satisfied the last requirement to fire PREV. Ensure that all
timing requirements are satisfied. */
if (INSN_TICK (prev) - clock > cost)
cost = INSN_TICK (prev) - clock;
/* Adjust the priority of PREV and either put it on the ready
list or queue it. */
adjust_priority (prev);
if (cost <= 1)
ready[new_ready++] = prev;
else
queue_insn (prev, cost);
}
}
return new_ready;
}
/* Given N_READY insns in the ready list READY at time CLOCK, queue
those that are blocked due to function unit hazards and rearrange
the remaining ones to minimize subsequent function unit hazards. */
static int
schedule_select (ready, n_ready, clock, file)
rtx *ready;
int n_ready, clock;
FILE *file;
{
int pri = INSN_PRIORITY (ready[0]);
int i, j, k, q, cost, best_cost, best_insn = 0, new_ready = n_ready;
rtx insn;
/* Work down the ready list in groups of instructions with the same
priority value. Queue insns in the group that are blocked and
select among those that remain for the one with the largest
potential hazard. */
for (i = 0; i < n_ready; i = j)
{
int opri = pri;
for (j = i + 1; j < n_ready; j++)
if ((pri = INSN_PRIORITY (ready[j])) != opri)
break;
/* Queue insns in the group that are blocked. */
for (k = i, q = 0; k < j; k++)
{
insn = ready[k];
if ((cost = actual_hazard (insn_unit (insn), insn, clock, 0)) != 0)
{
q++;
ready[k] = 0;
queue_insn (insn, cost);
if (file)
fprintf (file, "\n;; blocking insn %d for %d cycles",
INSN_UID (insn), cost);
}
}
new_ready -= q;
/* Check the next group if all insns were queued. */
if (j - i - q == 0)
continue;
/* If more than one remains, select the first one with the largest
potential hazard. */
else if (j - i - q > 1)
{
best_cost = -1;
for (k = i; k < j; k++)
{
if ((insn = ready[k]) == 0)
continue;
if ((cost = potential_hazard (insn_unit (insn), insn, 0))
> best_cost)
{
best_cost = cost;
best_insn = k;
}
}
}
/* We have found a suitable insn to schedule. */
break;
}
/* Move the best insn to be front of the ready list. */
if (best_insn != 0)
{
if (file)
{
fprintf (file, ", now");
for (i = 0; i < n_ready; i++)
if (ready[i])
fprintf (file, " %d", INSN_UID (ready[i]));
fprintf (file, "\n;; insn %d has a greater potential hazard",
INSN_UID (ready[best_insn]));
}
for (i = best_insn; i > 0; i--)
{
insn = ready[i-1];
ready[i-1] = ready[i];
ready[i] = insn;
}
}
/* Compact the ready list. */
if (new_ready < n_ready)
for (i = j = 0; i < n_ready; i++)
if (ready[i])
ready[j++] = ready[i];
return new_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 1
abort ();
#else
link = rtx_alloc (EXPR_LIST);
PUT_REG_NOTE_KIND (link, REG_DEAD);
#endif
}
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);
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 = rtx_alloc (EXPR_LIST);
PUT_REG_NOTE_KIND (temp_link, REG_DEAD);
XEXP (temp_link, 0) = temp_reg;
XEXP (temp_link, 1) = 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)
{
if (! all_needed && ! dead_or_set_p (insn, 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)
{
CLEAR_REGNO_REG_SET (bb_dead_regs, regno + j);
SET_REGNO_REG_SET (bb_live_regs, regno + j);
}
}
else
{
CLEAR_REGNO_REG_SET (bb_dead_regs, regno);
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:
attach_deaths (SUBREG_REG (x), insn,
set_p && ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (x)))
<= UNITS_PER_WORD)
|| (GET_MODE_SIZE (GET_MODE (SUBREG_REG (x)))
== GET_MODE_SIZE (GET_MODE ((x))))));
return;
case STRICT_LOW_PART:
attach_deaths (XEXP (x, 0), insn, 0);
return;
case ZERO_EXTRACT:
case SIGN_EXTRACT:
attach_deaths (XEXP (x, 0), insn, 0);
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);
}
/* Delete notes beginning with INSN and maybe put them in the chain
of notes ended by NOTE_LIST.
Returns the insn following the notes. */
static rtx
unlink_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;
if (write_symbols != NO_DEBUG && NOTE_LINE_NUMBER (insn) > 0)
/* Record line-number notes so they can be reused. */
LINE_NOTE (insn) = insn;
/* 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}. */
else 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;
}
/* 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;
}
}
/* 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;
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)
{
CONST_CALL_P (emit_note_after (INTVAL (XEXP (note, 0)), insn))
= 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 last;
}
/* Use modified list scheduling to rearrange insns in basic block
B. FILE, if nonzero, is where we dump interesting output about
this pass. */
static void
schedule_block (b, file)
int b;
FILE *file;
{
rtx insn, last;
rtx *ready, link;
int i, j, n_ready = 0, new_ready, n_insns;
int sched_n_insns = 0;
int clock;
#define NEED_NOTHING 0
#define NEED_HEAD 1
#define NEED_TAIL 2
int new_needs;
/* HEAD and TAIL delimit the region being scheduled. */
rtx head = basic_block_head[b];
rtx tail = basic_block_end[b];
/* PREV_HEAD and NEXT_TAIL are the boundaries of the insns
being scheduled. When the insns have been ordered,
these insns delimit where the new insns are to be
spliced back into the insn chain. */
rtx next_tail;
rtx prev_head;
/* Keep life information accurate. */
register struct sometimes *regs_sometimes_live;
int sometimes_max;
if (file)
fprintf (file, ";;\t -- basic block number %d from %d to %d --\n",
b, INSN_UID (basic_block_head[b]), INSN_UID (basic_block_end[b]));
i = max_reg_num ();
reg_last_uses = (rtx *) alloca (i * sizeof (rtx));
bzero ((char *) reg_last_uses, i * sizeof (rtx));
reg_last_sets = (rtx *) alloca (i * sizeof (rtx));
bzero ((char *) reg_last_sets, i * sizeof (rtx));
reg_pending_sets = ALLOCA_REG_SET ();
CLEAR_REG_SET (reg_pending_sets);
reg_pending_sets_all = 0;
clear_units ();
/* Remove certain insns at the beginning from scheduling,
by advancing HEAD. */
/* At the start of a function, before reload has run, don't delay getting
parameters from hard registers into pseudo registers. */
if (reload_completed == 0 && b == 0)
{
while (head != tail
&& GET_CODE (head) == NOTE
&& NOTE_LINE_NUMBER (head) != NOTE_INSN_FUNCTION_BEG)
head = NEXT_INSN (head);
while (head != tail
&& GET_CODE (head) == INSN
&& GET_CODE (PATTERN (head)) == SET)
{
rtx src = SET_SRC (PATTERN (head));
while (GET_CODE (src) == SUBREG
|| GET_CODE (src) == SIGN_EXTEND
|| GET_CODE (src) == ZERO_EXTEND
|| GET_CODE (src) == SIGN_EXTRACT
|| GET_CODE (src) == ZERO_EXTRACT)
src = XEXP (src, 0);
if (GET_CODE (src) != REG
|| REGNO (src) >= FIRST_PSEUDO_REGISTER)
break;
/* Keep this insn from ever being scheduled. */
INSN_REF_COUNT (head) = 1;
head = NEXT_INSN (head);
}
}
/* 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;
}
/* 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_CODE (head) == NOTE || GET_CODE (head) == CODE_LABEL))
goto ret;
#if 0
/* This short-cut doesn't work. It does not count call insns crossed by
registers in reg_sometimes_live. It does not mark these registers as
dead if they die in this block. It does not mark these registers live
(or create new reg_sometimes_live entries if necessary) if they are born
in this block.
The easy solution is to just always schedule a block. This block only
has one insn, so this won't slow down this pass by much. */
if (head == tail)
goto ret;
#endif
/* Now HEAD through TAIL are the insns actually to be rearranged;
Let PREV_HEAD and NEXT_TAIL enclose them. */
prev_head = PREV_INSN (head);
next_tail = NEXT_INSN (tail);
/* Initialize basic block data structures. */
dead_notes = 0;
pending_read_insns = 0;
pending_read_mems = 0;
pending_write_insns = 0;
pending_write_mems = 0;
pending_lists_length = 0;
last_pending_memory_flush = 0;
last_function_call = 0;
last_scheduled_insn = 0;
LOG_LINKS (sched_before_next_call) = 0;
n_insns = sched_analyze (head, tail);
if (n_insns == 0)
{
free_pending_lists ();
goto ret;
}
/* Allocate vector to hold insns to be rearranged (except those
insns which are controlled by an insn with SCHED_GROUP_P set).
All these insns are included between ORIG_HEAD and ORIG_TAIL,
as those variables ultimately are set up. */
ready = (rtx *) alloca ((n_insns+1) * sizeof (rtx));
/* TAIL is now the last of the insns to be rearranged.
Put those insns into the READY vector. */
insn = tail;
/* 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. */
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)
{
priority (insn);
if (last == 0)
{
ready[n_ready++] = insn;
INSN_PRIORITY (insn) = TAIL_PRIORITY - i;
INSN_REF_COUNT (insn) = 0;
}
else if (! find_insn_list (insn, LOG_LINKS (last)))
{
add_dependence (last, insn, REG_DEP_ANTI);
INSN_REF_COUNT (insn)++;
}
last = insn;
/* Skip over insns that are part of a group. */
while (SCHED_GROUP_P (insn))
{
insn = prev_nonnote_insn (insn);
priority (insn);
}
}
insn = PREV_INSN (insn);
/* Don't overrun the bounds of the basic block. */
if (insn == prev_head)
break;
}
/* Assign priorities to instructions. Also check whether they
are in priority order already. If so then I will be nonnegative.
We use this shortcut only before reloading. */
#if 0
i = reload_completed ? DONE_PRIORITY : MAX_PRIORITY;
#endif
for (; insn != prev_head; insn = PREV_INSN (insn))
{
if (GET_RTX_CLASS (GET_CODE (insn)) == 'i')
{
priority (insn);
if (INSN_REF_COUNT (insn) == 0)
{
if (last == 0)
ready[n_ready++] = insn;
else
{
/* Make this dependent on the last of the instructions
that must remain in order at the end of the block. */
add_dependence (last, insn, REG_DEP_ANTI);
INSN_REF_COUNT (insn) = 1;
}
}
if (SCHED_GROUP_P (insn))
{
while (SCHED_GROUP_P (insn))
{
insn = prev_nonnote_insn (insn);
priority (insn);
}
continue;
}
#if 0
if (i < 0)
continue;
if (INSN_PRIORITY (insn) < i)
i = INSN_PRIORITY (insn);
else if (INSN_PRIORITY (insn) > i)
i = DONE_PRIORITY;
#endif
}
}
#if 0
/* This short-cut doesn't work. It does not count call insns crossed by
registers in reg_sometimes_live. It does not mark these registers as
dead if they die in this block. It does not mark these registers live
(or create new reg_sometimes_live entries if necessary) if they are born
in this block.
The easy solution is to just always schedule a block. These blocks tend
to be very short, so this doesn't slow down this pass by much. */
/* If existing order is good, don't bother to reorder. */
if (i != DONE_PRIORITY)
{
if (file)
fprintf (file, ";; already scheduled\n");
if (reload_completed == 0)
{
for (i = 0; i < sometimes_max; i++)
regs_sometimes_live[i].live_length += n_insns;
finish_sometimes_live (regs_sometimes_live, sometimes_max);
}
free_pending_lists ();
goto ret;
}
#endif
/* Scan all the insns to be scheduled, removing NOTE insns
and register death notes.
Line number NOTE insns end up in NOTE_LIST.
Register death notes end up in DEAD_NOTES.
Recreate the register life information for the end of this basic
block. */
if (reload_completed == 0)
{
COPY_REG_SET (bb_live_regs, basic_block_live_at_start[b]);
CLEAR_REG_SET (bb_dead_regs);
if (b == 0)
{
/* This is the first block in the function. There may be insns
before head that we can't schedule. We still need to examine
them though for accurate register lifetime analysis. */
/* We don't want to remove any REG_DEAD notes as the code below
does. */
for (insn = basic_block_head[b]; insn != head;
insn = NEXT_INSN (insn))
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);
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);
/* ??? 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 clobbers (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);
CLEAR_REGNO_REG_SET (bb_dead_regs, j);
}
}
for (link = REG_NOTES (insn); link; link = 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));
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);
SET_REGNO_REG_SET (bb_dead_regs, regno + j);
}
}
else
{
CLEAR_REGNO_REG_SET (bb_live_regs, regno);
SET_REGNO_REG_SET (bb_dead_regs, regno);
}
}
}
}
}
}
/* If debugging information is being produced, keep track of the line
number notes for each insn. */
if (write_symbols != NO_DEBUG)
{
/* 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[b];
for (insn = basic_block_head[b];
insn != next_tail;
insn = NEXT_INSN (insn))
if (GET_CODE (insn) == NOTE && NOTE_LINE_NUMBER (insn) > 0)
line = insn;
else
LINE_NOTE (insn) = line;
}
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
{
rtx prev, next, link;
/* Farm out notes. This is needed to keep the debugger from
getting completely deranged. */
if (GET_CODE (insn) == NOTE)
{
prev = insn;
insn = unlink_notes (insn, next_tail);
if (prev == tail)
abort ();
if (prev == head)
abort ();
if (insn == next_tail)
abort ();
}
if (reload_completed == 0
&& 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);
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);
/* ??? 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 clobbers (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);
CLEAR_REGNO_REG_SET (bb_dead_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));
/* 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);
SET_REGNO_REG_SET (bb_dead_regs, regno + j);
}
}
else
{
CLEAR_REGNO_REG_SET (bb_live_regs, regno);
SET_REGNO_REG_SET (bb_dead_regs, regno);
}
}
else
prev = link;
}
}
}
if (reload_completed == 0)
{
/* 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;
/* Start with registers live at end. */
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);
});
}
SCHED_SORT (ready, n_ready, 1);
if (file)
{
fprintf (file, ";; ready list initially:\n;; ");
for (i = 0; i < n_ready; i++)
fprintf (file, "%d ", INSN_UID (ready[i]));
fprintf (file, "\n\n");
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
if (INSN_PRIORITY (insn) > 0)
fprintf (file, ";; insn[%4d]: priority = %4d, ref_count = %4d\n",
INSN_UID (insn), INSN_PRIORITY (insn),
INSN_REF_COUNT (insn));
}
/* Now HEAD and TAIL are going to become disconnected
entirely from the insn chain. */
tail = 0;
/* Q_SIZE will always be zero here. */
q_ptr = 0; clock = 0;
bzero ((char *) insn_queue, sizeof (insn_queue));
/* Now, perform list scheduling. */
/* Where we start inserting insns is after TAIL. */
last = next_tail;
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;
new_ready = n_ready;
while (sched_n_insns < n_insns)
{
q_ptr = NEXT_Q (q_ptr); clock++;
/* Add all pending insns that can be scheduled without stalls to the
ready list. */
for (insn = insn_queue[q_ptr]; insn; insn = NEXT_INSN (insn))
{
if (file)
fprintf (file, ";; launching %d before %d with no stalls at T-%d\n",
INSN_UID (insn), INSN_UID (last), clock);
ready[new_ready++] = insn;
q_size -= 1;
}
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 (new_ready == 0)
{
register int stalls;
for (stalls = 1; stalls < INSN_QUEUE_SIZE; stalls++)
if (insn = insn_queue[NEXT_Q_AFTER (q_ptr, stalls)])
{
for (; insn; insn = NEXT_INSN (insn))
{
if (file)
fprintf (file, ";; launching %d before %d with %d stalls at T-%d\n",
INSN_UID (insn), INSN_UID (last), stalls, clock);
ready[new_ready++] = insn;
q_size -= 1;
}
insn_queue[NEXT_Q_AFTER (q_ptr, stalls)] = 0;
break;
}
q_ptr = NEXT_Q_AFTER (q_ptr, stalls); clock += stalls;
}
/* There should be some instructions waiting to fire. */
if (new_ready == 0)
abort ();
if (file)
{
fprintf (file, ";; ready list at T-%d:", clock);
for (i = 0; i < new_ready; i++)
fprintf (file, " %d (%x)",
INSN_UID (ready[i]), INSN_PRIORITY (ready[i]));
}
/* Sort the ready list and choose the best insn to schedule. Select
which insn should issue in this cycle and queue those that are
blocked by function unit hazards.
N_READY holds the number of items that were scheduled the last time,
minus the one instruction scheduled on the last loop iteration; it
is not modified for any other reason in this loop. */
SCHED_SORT (ready, new_ready, n_ready);
if (MAX_BLOCKAGE > 1)
{
new_ready = schedule_select (ready, new_ready, clock, file);
if (new_ready == 0)
{
if (file)
fprintf (file, "\n");
/* We must set n_ready here, to ensure that sorting always
occurs when we come back to the SCHED_SORT line above. */
n_ready = 0;
continue;
}
}
n_ready = new_ready;
last_scheduled_insn = insn = ready[0];
/* The first insn scheduled becomes the new tail. */
if (tail == 0)
tail = insn;
if (file)
{
fprintf (file, ", now");
for (i = 0; i < n_ready; i++)
fprintf (file, " %d", INSN_UID (ready[i]));
fprintf (file, "\n");
}
if (DONE_PRIORITY_P (insn))
abort ();
if (reload_completed == 0)
{
/* Process this insn, and each insn linked to this one which must
be immediately output after this insn. */
do
{
/* 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_CODE (PATTERN (insn)) == SET
|| GET_CODE (PATTERN (insn)) == CLOBBER)
sched_note_set (b, PATTERN (insn), 1);
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), 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);
SET_REGNO_REG_SET (bb_dead_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, i,
{
sometimes_max
= new_sometimes_live (regs_sometimes_live,
i, 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, p->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--;
}
}
link = insn;
insn = PREV_INSN (insn);
}
while (SCHED_GROUP_P (link));
/* Set INSN back to the insn we are scheduling now. */
insn = ready[0];
}
/* Schedule INSN. Remove it from the ready list. */
ready += 1;
n_ready -= 1;
sched_n_insns += 1;
NEXT_INSN (insn) = last;
PREV_INSN (last) = insn;
/* Everything that precedes INSN now either becomes "ready", if
it can execute immediately before INSN, or "pending", if
there must be a delay. Give INSN high enough priority that
at least one (maybe more) reg-killing insns can be launched
ahead of all others. Mark INSN as scheduled by changing its
priority to -1. */
INSN_PRIORITY (insn) = LAUNCH_PRIORITY;
new_ready = schedule_insn (insn, ready, n_ready, clock);
INSN_PRIORITY (insn) = DONE_PRIORITY;
/* Schedule all prior insns that must not be moved. */
if (SCHED_GROUP_P (insn))
{
/* Disable these insns from being launched, in case one of the
insns in the group has a dependency on an earlier one. */
link = insn;
while (SCHED_GROUP_P (link))
{
/* Disable these insns from being launched by anybody. */
link = PREV_INSN (link);
INSN_REF_COUNT (link) = 0;
}
/* Now handle each group insn like the main insn was handled
above. */
link = insn;
while (SCHED_GROUP_P (link))
{
link = PREV_INSN (link);
sched_n_insns += 1;
/* ??? Why don't we set LAUNCH_PRIORITY here? */
new_ready = schedule_insn (link, ready, new_ready, clock);
INSN_PRIORITY (link) = DONE_PRIORITY;
}
}
/* Put back NOTE_INSN_SETJMP,
NOTE_INSN_{LOOP,EHREGION}_{BEGIN,END} notes. */
/* To prime the loop. We need to handle INSN and all the insns in the
sched group. */
last = NEXT_INSN (insn);
do
{
insn = PREV_INSN (last);
/* Maintain a valid chain so emit_note_before works.
This is necessary because PREV_INSN (insn) isn't valid
(if ! SCHED_GROUP_P) and if it points to an insn already
scheduled, a circularity will result. */
if (! SCHED_GROUP_P (insn))
{
NEXT_INSN (prev_head) = insn;
PREV_INSN (insn) = prev_head;
}
last = reemit_notes (insn, insn);
}
while (SCHED_GROUP_P (insn));
}
if (q_size != 0)
abort ();
if (reload_completed == 0)
finish_sometimes_live (regs_sometimes_live, sometimes_max);
/* HEAD is now the first insn in the chain of insns that
been scheduled by the loop above.
TAIL is the last of those insns. */
head = last;
/* 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 (head) = note_list;
NEXT_INSN (note_list) = head;
head = note_head;
}
/* 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 1
if (dead_notes != 0)
abort ();
#endif
if (new_needs & NEED_HEAD)
basic_block_head[b] = head;
PREV_INSN (head) = prev_head;
NEXT_INSN (prev_head) = head;
if (new_needs & NEED_TAIL)
basic_block_end[b] = tail;
NEXT_INSN (tail) = next_tail;
PREV_INSN (next_tail) = tail;
/* Restore the line-number notes of each insn. */
if (write_symbols != NO_DEBUG)
{
rtx line, note, prev, new;
int notes = 0;
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
{
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 (file && notes)
fprintf (file, ";; added %d line-number notes\n", notes);
}
if (file)
{
fprintf (file, ";; total time = %d\n;; new basic block head = %d\n;; new basic block end = %d\n\n",
clock, INSN_UID (basic_block_head[b]), INSN_UID (basic_block_end[b]));
}
/* Yow! We're done! */
free_pending_lists ();
ret:
FREE_REG_SET (reg_pending_sets);
FREE_REG_SET (old_live_regs);
return;
}
/* 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 here. */
link = rtx_alloc (EXPR_LIST);
PUT_REG_NOTE_KIND (link, REG_DEAD);
XEXP (link, 0) = temp;
XEXP (link, 1) = 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 = rtx_alloc (EXPR_LIST);
PUT_REG_NOTE_KIND (note, REG_DEAD);
XEXP (note, 0) = dest;
XEXP (note, 1) = 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 = rtx_alloc (EXPR_LIST);
PUT_REG_NOTE_KIND (note, REG_UNUSED);
XEXP (note, 0) = dest;
XEXP (note, 1) = 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) = gen_rtx (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 = rtx_alloc (EXPR_LIST);
PUT_REG_NOTE_KIND (note, REG_DEAD);
XEXP (note, 0) = dest;
XEXP (note, 1) = 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 = PATTERN (insn);
int 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;
}
}
}
/* 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 = MAX_INSNS_PER_SPLIT * (get_max_uid () + 1);
int b;
int i;
rtx insn;
/* Taking care of this degenerate case makes the rest of
this code simpler. */
if (n_basic_blocks == 0)
return;
/* Create an insn here so that we can hang dependencies off of it later. */
sched_before_next_call
= gen_rtx (INSN, VOIDmode, 0, NULL_RTX, NULL_RTX,
NULL_RTX, 0, NULL_RTX, NULL_RTX);
/* 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;
}
/* We create no insns here, only reorder them, so we
remember how far we can cut back the stack on exit. */
/* Allocate data for this pass. See comments, above,
for what these vectors do. */
insn_luid = (int *) alloca (max_uid * sizeof (int));
insn_priority = (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));
if (reload_completed == 0)
{
sched_reg_n_calls_crossed = (int *) alloca (max_regno * sizeof (int));
sched_reg_live_length = (int *) alloca (max_regno * sizeof (int));
bb_dead_regs = ALLOCA_REG_SET ();
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));
init_alias_analysis ();
}
else
{
sched_reg_n_calls_crossed = 0;
sched_reg_live_length = 0;
bb_dead_regs = 0;
bb_live_regs = 0;
if (! flag_schedule_insns)
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));
/* 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_luid, max_uid * sizeof (int));
bzero ((char *) insn_priority, 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));
/* Schedule each basic block, block by block. */
/* ??? Add a NOTE after the last insn of the last basic block. It is not
known why this is done. */
/* ??? Perhaps it's done to ensure NEXT_TAIL in schedule_block is a
valid insn. */
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]);
for (b = 0; b < n_basic_blocks; b++)
{
rtx insn, next;
note_list = 0;
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;
}
schedule_block (b, dump_file);
#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 ());
if (write_symbols != NO_DEBUG)
{
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 (dump_file && notes)
fprintf (dump_file, ";; deleted %d line-number notes\n", notes);
}
if (reload_completed == 0)
{
int regno;
for (regno = 0; regno < max_regno; regno++)
if (sched_reg_live_length[regno])
{
if (dump_file)
{
if (REG_LIVE_LENGTH (regno) > sched_reg_live_length[regno])
fprintf (dump_file,
";; 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_file,
";; 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_file,
";; 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_file,
";; register %d no longer crosses calls\n", 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];
/* 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. */
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];
}
}
if (reload_completed == 0)
{
FREE_REG_SET (bb_dead_regs);
FREE_REG_SET (bb_live_regs);
}
}
#endif /* INSN_SCHEDULING */
|