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
|
(*
Copyright (c) 2000
Cambridge University Technical Services Limited
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
This library 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
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*)
(* Adapted for the AMD64 from the I386 code generator. David C. J. Matthews 2006 *)
(*
Title: Code Generator Routines.
Author: Dave Matthews, Cambridge University Computer Laboratory
Copyright Cambridge University 1989
*)
(* This module contains the code vector and operations to insert code into
it. Each procedure is compiled into a separate segment. Initially it is
compiled into a fixed size segment, and then copied into a segment of the
correct size at the end.
This module contains all the definitions of the i386 opCodes and registers.
It uses "codeseg" to create and operate on the segment itself.
*)
(* 6-10/6/94 SPF added abstraction for opCodes, to replace old explicit hex(?) constants *)
(* 13/6/94 SPF started work on i386 version (was SPARC!) *)
(* August 2006 DCJM. Converted to AMD64. *)
functor AMD64CODECONS (
(*****************************************************************************)
(* DEBUG *)
(*****************************************************************************)
structure DEBUG :
sig
val assemblyCodeTag : bool Universal.tag
val compilerOutputTag: (string->unit) Universal.tag
val getParameter :
'a Universal.tag -> Universal.universal list -> 'a
end;
(*****************************************************************************)
(* MISC *)
(*****************************************************************************)
structure MISC :
sig
exception InternalError of string
end;
) :
(*****************************************************************************)
(* CODECONS export signature *)
(*****************************************************************************)
sig
type machineWord;
type short;
type code;
type reg; (* Machine registers *)
type address;
val regNone: reg;
val regResult: reg;
val regClosure: reg;
val regStackPtr: reg;
val regHandler: reg;
val regReturn: reg;
val regs: int; (* No of registers. *)
val argRegs: int; (* No of args in registers. *)
val regN: int -> reg;
val nReg: reg -> int;
val argReg: int -> reg;
val regEq: reg * reg -> bool;
val regNeq: reg * reg -> bool;
val regRepr: reg -> string;
type addrs
val codeCreate: bool * string * Universal.universal list -> code; (* makes the initial segment. *)
(* Operations. *)
type instrs;
val instrMove: instrs;
val instrAddA: instrs;
val instrSubA: instrs;
val instrRevSubA: instrs;
val instrMulA: instrs;
val instrAddW: instrs;
val instrSubW: instrs;
val instrRevSubW: instrs;
val instrMulW: instrs;
val instrDivW: instrs;
val instrModW: instrs;
val instrOrW: instrs;
val instrAndW: instrs;
val instrXorW: instrs;
val instrLoad: instrs;
val instrLoadB: instrs;
val instrVeclen: instrs;
val instrVecflags: instrs;
val instrPush: instrs;
val instrUpshiftW: instrs;
val instrDownshiftW: instrs;
val instrDownshiftArithW: instrs;
val instrGetFirstLong: instrs;
val instrStringLength: instrs;
val instrSetStringLength: instrs;
val instrBad: instrs;
(* Can the we use the same register as the source and destination
of an instructions? (it would be more flexible to make this
a function of type "instrs -> bool", but a simple flag will
suffice for now. SPF 17/1/97
*)
val canShareRegs : bool;
(* Enquire about operations. *)
val instrIsRR: instrs -> bool; (* Is the general form implemented? *)
val instrIsRI: instrs * machineWord -> bool; (* Is the immediate value ok? *)
(* Code generate operations. *)
val genRR: instrs * reg * reg * reg * code -> unit;
val genRI: instrs * reg * machineWord * reg * code -> unit;
type tests;
val testNeqW: tests;
val testEqW: tests;
val testGeqW: tests;
val testGtW: tests;
val testLeqW: tests;
val testLtW: tests;
val testNeqA: tests;
val testEqA: tests;
val testGeqA: tests;
val testGtA: tests;
val testLeqA: tests;
val testLtA: tests;
val Short: tests;
val Long: tests;
type labels; (* The source of a jump. *)
val noJump: labels;
(* Compare and branch for fixed and arbitrary precision. *)
val isCompRR: tests -> bool;
val isCompRI: tests * machineWord -> bool;
val compareAndBranchRR: reg * reg * tests * code -> labels;
val compareAndBranchRI: reg * machineWord * tests * code -> labels;
datatype storeWidth = STORE_WORD | STORE_BYTE
val genLoad: int * reg * reg * code -> unit;
val isIndexedStore: storeWidth -> bool
val genStore: reg * int * reg * storeWidth * reg * code -> unit;
val isStoreI: machineWord * storeWidth * bool -> bool;
val genStoreI: machineWord * int * reg * storeWidth * reg * code -> unit;
val genPush: reg * code -> unit;
val genLoadPush: int * reg * code -> unit;
val preferLoadPush: bool;
val genLoadCoderef: code * reg * code -> unit;
val genStackOffset: reg * int * code -> unit;
val allocStore: int * Word8.word * reg * code -> unit;
val setFlag: reg * code * Word8.word -> unit;
val completeSegment: code -> unit;
datatype callKinds =
Recursive
| ConstantFun of machineWord * bool
| CodeFun of code
| FullCall
val callFunction: callKinds * code -> unit;
val jumpToFunction: callKinds * reg * code -> unit;
val returnFromFunction: reg * int * code -> unit;
val raiseException: code -> unit;
type cseg;
val copyCode: code * int * reg list -> address;
val unconditionalBranch: code -> labels;
type handlerLab;
val loadHandlerAddress: reg * code -> handlerLab;
val fixupHandler: handlerLab * code -> unit;
val fixup: labels * code -> unit; (* Fix up a jump. *)
(* ic - Address for the next instruction in the segment. *)
val ic: code -> addrs;
val jumpback: addrs * bool * code -> unit; (* Backwards jump. *)
val resetStack: int * code -> unit; (* Set a pending reset *)
val procName: code -> string; (* Name of the procedure. *)
type cases
type jumpTableAddrs
val constrCases : int * addrs -> cases;
val useIndexedCase: int * int * int * bool -> bool;
val indexedCase : reg * reg * int * int * bool * code -> jumpTableAddrs;
val makeJumpTable : jumpTableAddrs * cases list * addrs * int * int * code -> unit;
val inlineAssignments: bool
val codeAddress: code -> address option
val traceContext: code -> string;
end (* CODECONS export signature *) =
let
(*****************************************************************************)
(* ADDRESS *)
(*****************************************************************************)
structure ADDRESS :
sig
type machineWord; (* NB *not* eqtype, 'cos it might be a closure *)
type short = Word.word;
type address;
type handler;
val wordEq : machineWord * machineWord -> bool
val isShort: 'a -> bool;
val toShort: 'a -> short;
val toMachineWord: 'a -> machineWord;
val offsetAddr : address * short -> handler
val alloc: (short * Word8.word * machineWord) -> address
val F_words : Word8.word
val lock : address -> unit;
val wordSize: int
end = Address;
(*****************************************************************************)
(* CODESEG *)
(*****************************************************************************)
structure CODESEG :
sig
type machineWord;
type short;
type address;
type cseg;
val csegMake: int -> cseg;
val csegConvertToCode: cseg -> unit;
val csegLock: cseg -> unit;
val csegGet: cseg * int -> Word8.word;
val csegSet: cseg * int * Word8.word -> unit;
val csegPutWord: cseg * int * machineWord -> unit;
val csegCopySeg: cseg * cseg * int * int -> unit;
val csegAddr: cseg -> address;
val csegPutConstant: cseg * int * machineWord * 'a -> unit;
end = CodeSeg;
in
(*****************************************************************************)
(* CODECONS functor body *)
(*****************************************************************************)
struct
open CODESEG;
open DEBUG;
open ADDRESS;
open MISC;
val toInt = Word.toIntX (* This previously just cast the value so continue to treat it as signed. *)
val isX64 = wordSize = 8 (* Generate X64 instructions if the word length is 8. *)
val short1 = toShort 1;
(* added SPF - take numbers OUT of code *)
(* These are defined here as explicit constants, so the *)
(* code-generator can in-line them as immediates (I think). *)
val exp2_3 = 0x8;
val exp2_6 = 0x40;
val exp2_7 = 0x80;
val exp2_8 = 0x100;
val exp2_16 = 0x10000;
val exp2_24 = 0x1000000;
val exp2_30 = 0x40000000;
val exp2_31 = 0x80000000;
val exp2_32 = 0x100000000;
val exp2_56 = 0x100000000000000;
val exp2_63 = 0x8000000000000000;
val exp2_64 = 0x10000000000000000;
(* Let's check that we got them right! *)
local
fun exp2 0 = 1
| exp2 n = 2 * exp2 (n - 1);
in
val UUU =
(
exp2_3 = exp2 3 andalso
exp2_6 = exp2 6 andalso
exp2_7 = exp2 7 andalso
exp2_8 = exp2 8 andalso
exp2_16 = exp2 16 andalso
exp2_24 = exp2 24 andalso
exp2_30 = exp2 30 andalso
exp2_31 = exp2 31 andalso
exp2_32 = exp2 32 andalso
exp2_56 = exp2 56 andalso
exp2_63 = exp2 63 andalso
exp2_64 = exp2 64
)
orelse raise InternalError "Powers of 2 incorrectly specified";
end;
(* tag a short constant *)
fun tag c = 2 * c + 1;
(* shift a short constant, but don't set tag bit *)
fun semitag c = 2 * c;
fun is8Bit n = ~ exp2_7 <= n andalso n < exp2_7;
infix 6 addrPlus addrMinus;
infix 4 (* ? *) addrLt; (* SPF 5/1/95 *)
infix 4 (* ? *) addrLe;
(* All indexes into the code vector have type "addrs" *)
(* This should be an abstype, but it's exported as an eqtype *)
datatype addrs = Addr of int
(* + is defined to add an integer to an address *)
fun (Addr a) addrPlus b = Addr (a + b);
(* The difference between two addresses is an integer *)
fun (Addr a) addrMinus (Addr b) = a - b;
fun (Addr a) addrLt (Addr b) = a < b;
fun (Addr a) addrLe (Addr b) = a <= b;
fun mkAddr n = Addr n; (* addr.up *)
fun getAddr (Addr a) = a; (* addr.down *)
val addrZero = mkAddr 0;
val addrLast = mkAddr (if isX64 then exp2_56 -1 else exp2_30 - 1); (* A big number. *)
(* The "value" points at the jump instruction, or rather at the
jump offset part of it. It is a ref because we may have to change
it if we have to put in a jump with a 32-bit offset. *)
datatype jumpFrom =
Jump8From of addrs
| Jump32From of addrs
type labels = (jumpFrom ref) list;
val noJump = []:labels;
(* This is the list of outstanding labels. Use a separate type from
"labels" for extra security. *)
type labList = (jumpFrom ref) list;
datatype setCodeseg =
Unset
| Set of cseg (* Used for completing forward references. *)
;
(* Constants which are too large to go inline in the code are put in
a list and put at the end of the code. They are arranged so that
the garbage collector can find them and change them as necessary.
A reference to a constant is treated like a forward reference to a
label. *)
(* A code list is used to hold a list of code-vectors which must have the
address of this code-vector put into it. *)
datatype const =
WVal of machineWord (* an existing constant *)
| CVal of code (* a forward-reference to another function *)
| HVal of addrs ref (* a handler *)
and ConstPosn =
InlineAbsolute (* The constant is within the code. *)
| InlineRelative (* The constant is within the code but is PC relative (call or jmp). *)
| ConstArea of int (* The constant is in the constant area. *)
and code = Code of
{ codeVec: cseg, (* This segment is used as a buffer. When the
procedure has been code generated it is
copied into a new segment of the correct size *)
ic: addrs ref, (* Pointer to first free location in "codevec" *)
constVec: (* Constants used in the code *)
{const: const, addrs: addrs, posn: ConstPosn} list ref,
numOfConsts: int ref, (* size of constVec *)
nonInlineConsts: int ref,
stackReset: int ref, (* Distance to reset the stack before the next instr. *)
pcOffset: int ref, (* Offset of code in final segment. *)
(* This is used also to test for identity of code segments. *)
labelList: labList ref, (* List of outstanding short branches. *)
longestBranch: addrs ref, (* Address of the earliest 1-byte branch. *)
procName: string, (* Name of the procedure. *)
otherCodes: code list ref, (* Other code vectors with forward references to this vector. *)
resultSeg: setCodeseg ref, (* The segment as the final result. *)
mustCheckStack: bool ref, (* Set to true if stack check must be done. *)
justComeFrom: labels ref, (* The label we have just jumped from. *)
exited: bool ref, (* False if we can fall-through to here *)
selfCalls: addrs list ref, (* List of recursive calls to patch up. *)
selfJumps: labels ref, (* List of recursive tail-calls to patch up. *)
noClosure: bool, (* should we make a closure from this? *)
branchCheck: addrs ref, (* the address we last fixed up to. I added
this to track down a bug and I've left it
in for security. DCJM 19/1/01. *)
printAssemblyCode:bool, (* Whether to print the code when we finish. *)
printStream: string->unit (* The stream to use *)
};
(* procName is exported. *)
fun procName (Code {procName,...}) = procName;
(* %ebp points to a structure that interfaces to the RTS. These are
offsets into that structure. *)
val MemRegLocalMPointer = 0
and MemRegHandlerRegister = wordSize
and MemRegLocalMbottom = wordSize*2
and MemRegStackLimit = wordSize*3
and MemRegHeapOverflowCall = wordSize*8
and MemRegStackOverflowCall = wordSize*9
and MemRegStackOverflowCallEx = wordSize*10
and MemRegRaiseException = wordSize*11
and MemRegRaiseDiv = wordSize*13
and MemRegArbEmulation = wordSize*14
(* Several operations are not generated immediately but recorded and
generated later. Labels (i.e. the destination of a branch) are recorded
in just_come_from. Adjustments to the real stack pointer are recorded
in stack_reset.
The order in which these "instructions" are assumed to happen is of
course significant. If just_come_from is not empty it is assumed to
have happened before anything else. After that the stack pointer is
adjusted and finally the next instruction is executed.
*)
val initialCodeSize = 15; (* words. Initial size of segment. *)
(* Test for identity of the code segments by testing whether
the pcOffset ref is the same. N.B. NOT its contents. *)
infix is;
fun (Code{pcOffset=a, ...}) is (Code{pcOffset=b, ...}) = a=b;
(* create and initialise a code segment *)
fun codeCreate (noClosure : bool, name : string, parameters) : code =
Code
{
codeVec = csegMake initialCodeSize, (* a byte array *)
ic = ref addrZero,
constVec = ref [],
numOfConsts = ref 0,
nonInlineConsts = ref 0,
stackReset = ref 0,
pcOffset = ref 0, (* only non-zero after code is copied *)
labelList = ref [],
longestBranch = ref addrLast, (* None so far *)
procName = name,
otherCodes = ref [],
resultSeg = ref Unset, (* Not yet done *)
mustCheckStack = ref false,
justComeFrom = ref [],
exited = ref false,
selfCalls = ref [],
selfJumps = ref [],
noClosure = noClosure,
branchCheck = ref addrZero,
printAssemblyCode = DEBUG.getParameter DEBUG.assemblyCodeTag parameters,
printStream = DEBUG.getParameter DEBUG.compilerOutputTag parameters
};
(* Put 1 unsigned byte at a given offset in the segment. *)
fun set8u (b : int, addr, seg) =
let
val a = getAddr addr;
in
csegSet (seg, a, Word8.fromInt b)
end;
(* Put 1 signed byte at a given offset in the segment. *)
fun set8s (b : int, addr, seg) =
let
val a = getAddr addr;
val b' = if b < 0 then b + exp2_8 else b;
in
csegSet (seg, a, Word8.fromInt b')
end;
(* Get 1 unsigned byte from the given offset in the segment. *)
fun get8u (a: int, seg: cseg) : int = Word8.toInt (csegGet (seg, a));
(* Get 1 signed byte from the given offset in the segment. *)
fun get8s (a: int, seg: cseg) : int = Word8.toIntX (csegGet (seg, a));
(* Put 4 bytes at a given offset in the segment. *)
(* b0 is the least significant byte. *)
fun set4Bytes (b3, b2, b1, b0, addr, seg) =
let
val a = getAddr addr;
in
(* Little-endian *)
csegSet (seg, a, Word8.fromInt b0);
csegSet (seg, a + 1, Word8.fromInt b1);
csegSet (seg, a + 2, Word8.fromInt b2);
csegSet (seg, a + 3, Word8.fromInt b3)
end;
(* Put 1 unsigned word at a given offset in the segment. *)
fun set32u (ival: int, addr: addrs, seg) : unit =
let
val topHw = ival div exp2_16;
val bottomHw = ival mod exp2_16;
val b3 = topHw div exp2_8;
val b2 = topHw mod exp2_8;
val b1 = bottomHw div exp2_8;
val b0 = bottomHw mod exp2_8;
in
set4Bytes (b3, b2, b1, b0, addr, seg)
end;
fun setBytes(seg, ival, offset, 0) = ()
| setBytes(seg, ival, offset, count) =
(
csegSet(seg, offset, Word8.fromInt(ival mod exp2_8));
setBytes(seg, ival div exp2_8, offset+1, count-1)
);
fun setWordU (ival: int, addr: addrs, seg) : unit =
setBytes(seg, ival, getAddr addr, wordSize)
fun set64u (ival: int, addr: addrs, seg) : unit =
setBytes(seg, ival, getAddr addr, 8)
fun set64s (ival: int, addr: addrs, seg) : unit =
let
val topByte = (ival div exp2_56) mod exp2_8
in
setBytes(seg, ival, getAddr addr, 7);
setBytes(seg, if topByte < 0 then topByte + exp2_8 else topByte, getAddr addr + 7, 1)
end
(* Put 1 signed word at a given offset in the segment. *)
fun set32s (ival: int, addr: addrs, seg) : unit =
let
val topHw = ival div exp2_16;
val bottomHw = ival mod exp2_16;
val b3 = topHw div exp2_8;
val b2 = topHw mod exp2_8;
val b1 = bottomHw div exp2_8;
val b0 = bottomHw mod exp2_8;
val b3' = if b3 < 0 then b3 + exp2_8 else b3;
in
set4Bytes (b3', b2, b1, b0, addr, seg)
end;
(* Get 1 signed 32 bit word from the given offset in the segment. *)
fun get32s (a: int, seg: cseg) : int =
let
val b0 = Word8.toInt (csegGet (seg, a));
val b1 = Word8.toInt (csegGet (seg, a + 1));
val b2 = Word8.toInt (csegGet (seg, a + 2));
val b3 = Word8.toInt (csegGet (seg, a + 3));
val b3' = if b3 >= exp2_7 then b3 - exp2_8 else b3;
val topHw = (b3' * exp2_8) + b2;
val bottomHw = (b1 * exp2_8) + b0;
in
(topHw * exp2_16) + bottomHw
end;
fun get64s (a: int, seg: cseg) : int =
let
val b0 = Word8.toInt (csegGet (seg, a));
val b1 = Word8.toInt (csegGet (seg, a + 1));
val b2 = Word8.toInt (csegGet (seg, a + 2));
val b3 = Word8.toInt (csegGet (seg, a + 3));
val b4 = Word8.toInt (csegGet (seg, a + 4));
val b5 = Word8.toInt (csegGet (seg, a + 5));
val b6 = Word8.toInt (csegGet (seg, a + 6));
val b7 = Word8.toInt (csegGet (seg, a + 7));
val b7' = if b7 >= exp2_7 then b7 - exp2_8 else b7;
in
((((((((b7' * exp2_8 + b6) * exp2_8 + b5) * exp2_8 + b4) * exp2_8 + b3)
* exp2_8 + b2) * exp2_8) + b1) * exp2_8) + b0
end;
(* Test whether a tagged value will fit into a 32-bit signed constant. *)
fun isTagged32bitS(a: machineWord) =
if isShort a
then let val aI = toInt(toShort a) in ~exp2_30 <= aI andalso aI < exp2_30 end
else false
(* Code-generate a byte. *)
fun gen8u (ival: int, Code {ic, codeVec, ...}) : unit =
if 0 <= ival andalso ival < exp2_8
then let
val icVal = !ic;
in
ic := icVal addrPlus 1;
set8u (ival, icVal, codeVec)
end
else raise InternalError "gen8u: invalid byte";
(* Used for signed byte values. *)
fun gen8s (ival: int, Code {ic, codeVec, ...}) =
if ~exp2_7 <= ival andalso ival < exp2_7
then let
val icVal = !ic;
in
ic := icVal addrPlus 1;
set8s (ival, icVal, codeVec)
end
else raise InternalError "gen8s: invalid byte";
(* Code-generate a 32-bit word. *)
fun gen32u (ival: int, Code {ic, codeVec, ...}) : unit =
if 0 <= ival andalso (isShort ival orelse ival < exp2_32)
(* Note: This previously was 0 <= ival andalso ival < exp2_32
The only reason for adding isShort ival is that doing so
avoids almost all the arbitrary precision emulation traps
that used to be generated here. *)
then let
val icVal = !ic;
in
ic := icVal addrPlus 4;
set32u (ival, icVal, codeVec)
end
else raise InternalError "gen32u: invalid word";
fun gen32s (ival: int, Code {ic, codeVec, ...}) : unit =
(* We really only need to check this on the 64-bit machine and it would otherwise
be a hot-spot for arbitrary precision arithmetic on 32-bit m/c. *)
if not isX64 orelse ~exp2_31 <= ival andalso ival < exp2_31
then let
val icVal = !ic;
in
ic := icVal addrPlus 4;
set32s (ival, icVal, codeVec)
end
else raise InternalError "gen32s: invalid word";
fun gen64u (ival: int, Code {ic, codeVec, ...}) : unit =
if 0 <= ival andalso (isShort ival orelse ival < exp2_64)
then let
val icVal = !ic;
in
ic := icVal addrPlus 8;
set64u (ival, icVal, codeVec)
end
else raise InternalError "gen32u: invalid word";
fun gen64s (ival: int, Code {ic, codeVec, ...}) : unit =
let
val icVal = !ic;
in
ic := icVal addrPlus 8;
set64s (ival, icVal, codeVec)
end
datatype mode =
Based (* mod = 0 *)
| Based8 (* mod = 1 *)
| Based32 (* mod = 2 *)
| Register (* mod = 3 *)
;
(* Put together the three fields which make up the mod r/m byte. *)
fun modrm (md : mode, rg: int, rm : int) : int =
let
val modField =
case md of
Based => 0 * exp2_6 (* compile-time evaluated! *)
| Based8 => 1 * exp2_6
| Based32 => 2 * exp2_6
| Register => 3 * exp2_6
;
in
modField + (rg * exp2_3) + rm
end;
fun genmodrm (md : mode, rg: int, rm : int, cvec) : unit =
gen8u (modrm (md, rg, rm), cvec);
(* REX prefix *)
fun rex {w,r,x,b} =
0x40 + (if w then 8 else 0) + (if r then 4 else 0) + (if x then 2 else 0) + (if b then 1 else 0);
datatype scaleFactor =
Scale1 (* s = 0 *)
| Scale2 (* s = 1 *)
| Scale4 (* s = 2 *)
| Scale8 (* s = 3 *)
;
(* Put together the three fields which make up the s-i-b byte. *)
fun sib (s : scaleFactor, i: int, b : int) : int =
let
val sizeField =
case s of
Scale1 => 0 * exp2_6
| Scale2 => 1 * exp2_6
| Scale4 => 2 * exp2_6
| Scale8 => 3 * exp2_6
;
in
sizeField + (i * exp2_3) + b
end;
fun gensib (s : scaleFactor, i: int, b : int, cvec : code) : unit =
gen8u (sib (s, i, b), cvec);
fun scSet (Set x) = x | scSet _ = raise InternalError "scSet";
(* Add a constant to the list along with its address. We mustn't put
the constant directly in the code since at this stage the code is
simply a byte segment and if we have a garbage collection the value
won't be updated. *)
fun addConstToVec (valu: const, posn: ConstPosn,
cvec as Code{numOfConsts, constVec, ic, nonInlineConsts, ...}): unit =
let
(* Inline constants are in the body of the code. Non-inline constants are
stored in the constant vector at the end of the code. The value that goes
in here is the PC-relative offset of the constant. *)
val realPosn =
case posn of
ConstArea _ => (nonInlineConsts := ! nonInlineConsts + 1; ConstArea(!nonInlineConsts))
| p => p
val isInline = case posn of ConstArea _ => false | _ => true
in
numOfConsts := ! numOfConsts + 1;
constVec := {const = valu, addrs = !ic, posn = realPosn} :: ! constVec;
(* We must put a valid tagged integer in here because we might
get a garbage collection after we have copied this code into
the new code segment but before we've put in the real constant.
If this is a relative branch we need to point this at itself.
Until it is set to the relative offset of the destination it
needs to contain an address within the code and this could
be the last instruction. *)
if isInline andalso wordSize = 8
then gen64s (tag 0, cvec)
else gen32s (case posn of InlineRelative => ~5 | _ => tag 0, cvec)
end
abstype reg = Reg of int*bool (* registers. *)
with
(* These are the real registers we have. The AMD extension encodes the
additional registers through the REX prefix. *)
val eax = Reg (0, false);
val ecx = Reg (1, false);
val edx = Reg (2, false);
val ebx = Reg (3, false);
val esp = Reg (4, false);
val ebp = Reg (5, false);
val esi = Reg (6, false);
val edi = Reg (7, false);
val r8 = Reg (0, true);
val r9 = Reg (1, true);
val r10 = Reg (2, true);
val r11 = Reg (3, true);
val r12 = Reg (4, true);
val r13 = Reg (5, true);
val r14 = Reg (6, true);
val r15 = Reg (7, true);
(* Not real registers. *)
val regNone = Reg (8, true);
val regHandler = Reg (9, true);
val regResult = eax; (* Result of function call goes in here. *)
val regClosure = edx; (* Addr. of closure for fn. call goes here. *)
val regStackPtr = esp;
val regReturn = regNone;
fun getReg (Reg rf) = rf; (* reg.down *)
fun mkReg nf = Reg nf; (* reg.up *)
(* The number of general registers. Includes result, closure, code,
return and arg regs but not stackptr, handler, stack limit
or heap ptrs. *)
val regs = 13 (* eax, ebx, ecx, edx, esi, edi, r8-r14 *);
(* N.B. The encoding of registers as integers here is used to
encode the register modification sets. It MUST match the
encoding used for IO functions in the registerMaskVector in
the RTS assembly code section. *)
(* The nth register (counting from 0). *)
fun regN i =
if i < 0 orelse i >= regs then raise InternalError "Bad register number"
else if i <= 3 then mkReg (i, false)
else if i = 4 orelse i = 5 then mkReg (i+2, false)
else mkReg (i-6, true)
infix 7 regEq regNeq;
fun (Reg a) regEq (Reg b) = a = b;
fun (Reg a) regNeq (Reg b) = a <> b;
(* The number of the register. *)
fun nReg(Reg(r, false)) = if r >= 6 then r-2 else r
| nReg(Reg(r, true)) = r+6
fun regRepr (r as Reg(n, false)) =
if r regEq eax then "%rax" else
if r regEq ebx then "%rbx" else
if r regEq ecx then "%rcx" else
if r regEq edx then "%rdx" else
if r regEq esp then "%rsp" else
if r regEq ebp then "%rbp" else
if r regEq esi then "%rsi" else
if r regEq edi then "%rdi" else "Unknown"
| regRepr (r as Reg(n, true)) =
"%r" ^ Int.toString(n+8);
fun regPrint r = TextIO.output (TextIO.stdOut, regRepr r);
(* Arguments are passed in eax, ebx, r8, r9 and r10. *)
val argRegs = 5;
fun argReg 0 = eax
| argReg 1 = ebx
| argReg 2 = r8
| argReg 3 = r9
| argReg 4 = r10
| argReg i = raise InternalError ("Not arg reg " ^ Int.toString i);
end (* reg *);
datatype arithOp =
ADD
| OR
| ADC
| SBB
| AND
| SUB
| XOR
| CMP
;
fun arithOpToInt ADD = 0
| arithOpToInt OR = 1
| arithOpToInt ADC = 2
| arithOpToInt SBB = 3
| arithOpToInt AND = 4
| arithOpToInt SUB = 5
| arithOpToInt XOR = 6
| arithOpToInt CMP = 7
;
(* Primary opCodes. N.B. only opCodes actually used are listed here.
If new instruction are added check they will be handled by the
run-time system in the event of trap. *)
datatype opCode =
Group1_8_A
| Group1_32_A
| JMP_8
| JMP_32
| CALL_32
| MOVL_A_R
| MOVL_R_A
| MOVB_R_A
| PUSH_R of int
| POP_R of int
| Group5
| NOP
| LEAL
| MOVL_32_64_R of int
| MOVL_32_A
| MOVB_8_A
| ESCAPE
| POP_A
| RET
| RET_16
| JO
| JE
| JNE
| JL
| JGE
| JLE
| JG
| JB
| JNB
| JNA
| JA
| Arith of arithOp * int
| STC
| Group3_A
| Group3_a
| Group2_8_A
| Group2_1_A
| PUSH_8
| PUSH_32
| TEST_ACC8
| MOVZX (* Needs ESCAPE code *) (* B6 *)
;
fun opToInt opn =
case opn of
Group1_8_A => 0x83
| Group1_32_A => 0x81
| JMP_8 => 0xeb
| JMP_32 => 0xe9
| CALL_32 => 0xe8
| MOVL_A_R => 0x8b
| MOVL_R_A => 0x89
| MOVB_R_A => 0x88
| PUSH_R reg => 0x50 + reg
| POP_R reg => 0x58 + reg
| Group5 => 0xff
| NOP => 0x90
| LEAL => 0x8d
| MOVL_32_64_R reg => 0xb8 + reg
| MOVL_32_A => 0xc7
| MOVB_8_A => 0xc6
| ESCAPE => 0x0f
| POP_A => 0x8f
| RET => 0xc3
| RET_16 => 0xc2
| JO => 0x70
| JB => 0x72
| JNB => 0x73
| JE => 0x74
| JNE => 0x75
| JNA => 0x76
| JA => 0x77
| JL => 0x7c
| JGE => 0x7d
| JLE => 0x7e
| JG => 0x7f
| Arith (ao,dw) => arithOpToInt ao * 8 + dw
| STC => 0xf9
| Group3_A => 0xf7
| Group3_a => 0xf6
| Group2_8_A => 0xc1
| Group2_1_A => 0xd1
| PUSH_8 => 0x6a
| PUSH_32 => 0x68
| TEST_ACC8 => 0xa8
| MOVZX => 0xb6 (* Needs ESCAPE code *)
;
(* ...
val eax = Reg 0;
val ecx = Reg 1;
val edx = Reg 2;
val ebx = Reg 3;
val esp = Reg 4; (* also used for "SIB used" and "no index" *)
val ebp = Reg 5; (* also used for "absolute" *)
val esi = Reg 6;
val edi = Reg 7;
type basereg = reg; {0,1,2,3,6,7 only}
type indexreg = reg; {0,1,2,3,5,6,7 only}
The i386 family has a horrendous collection of not-quite-orthogonal addressing modes.
Register mode:
(1) reg mod = 3; r/m = getReg reg
DS-relative addressing modes:
(2) DS:[basereg] mod = 0; r/m = getReg basereg
(3) DS:[basereg + disp8] mod = 1; r/m = getReg basereg
(4) DS:[basereg + disp32] mod = 2; r/m = getReg basereg
(2a) DS:[basereg] mod = 0; r/m = 4; s = ?; i = 4; b = getReg basereg
(3a) DS:[basereg + disp8] mod = 1; r/m = 4; s = ?; i = 4; b = getReg basereg
(4a) DS:[basereg + disp32] mod = 2; r/m = 4; s = ?; i = 4; b = getReg basereg
(5) DS:[basereg + (scale * indexreg)] mod = 0; r/m = 4; s = scale; i = getReg indexreg; b = getReg basereg
(6) DS:[basereg + (scale * indexreg) + disp8] mod = 1; r/m = 4; s = scale; i = getReg indexreg; b = getReg basereg
(7) DS:[basereg + (scale * indexreg) + disp32] mod = 2; r/m = 4; s = scale; i = getReg indexreg; b = getReg basereg
(8) DS:disp32 mod = 0; r/m = 5
(8a) DS:[disp32] mod = 0; r/m = 4; s = ?; i = 4; b = 5
(9) DS:[disp32 + (scale * indexreg)] mod = 0; r/m = 4; s = scale; i = getReg indexreg; b = 5
SS-relative addressing modes:
(10) SS:[ebp + disp8] mod = 1; r/m = 5
(11) SS:[ebp + disp32] mod = 2; r/m = 5
(12) SS:[ebp + (scale * indexreg) + disp8] mod = 1; r/m = 4; s = scale; i = getReg indexreg; b = 5
(13) SS:[ebp + (scale * indexreg) + disp32] mod = 2; r/m = 4; s = scale; i = getReg indexreg; b = 5
(14) SS:[esp + (scale * indexreg)] mod = 0; r/m = 4; s = scale; i = getReg indexreg; b = 4
(15) SS:[esp + (scale * indexreg) + disp8] mod = 1; r/m = 4; s = scale; i = getReg indexreg; b = 4
(16) SS:[esp + (scale * indexreg) + disp32] mod = 2; r/m = 4; s = scale; i = getReg indexreg; b = 4
... *)
(* Make a reference to another procedure. Usually this will be a forward reference but
it may have been compiled already, in which case we can put the code address in now. *)
fun codeConst (Code {resultSeg = ref(Set seg), ... }, posn, into) =
(* Already done. *) addConstToVec (WVal (toMachineWord(csegAddr seg)), posn, into)
| codeConst (r as Code {otherCodes, ... }, posn, into) = (* forward *)
(* Add the referring procedure onto the list of the procedure
referred to if it is not already there. This makes sure that when
the referring procedure is finished and its address is known the
address will be plugged in to every procedure which needs it. *)
let
fun onList x [] = false
| onList x (c::cs) = (x is c) orelse onList x cs ;
val codeList = ! otherCodes;
in
if onList into codeList then () else otherCodes := into :: codeList;
addConstToVec (CVal r, posn, into)
end;
(* Removes a label from the list when it has been fixed up
or converted to the long form. *)
fun removeLabel (lab:addrs, Code{longestBranch, labelList, ... }) : unit =
let
fun removeEntry ([]: labList) : labList = []
| removeEntry ((entry as ref (Jump32From addr)) :: t) =
removeEntry t (* we discard long jumps *)
| removeEntry ((entry as ref (Jump8From addr)) :: t) =
if lab = addr
then removeEntry t
else
(
if addr addrLt !longestBranch
then longestBranch := addr
else ();
entry :: removeEntry t
) (* removeEntry *);
in
(* Must also find the new longest branch. *)
longestBranch := addrLast;
labelList := removeEntry (! labelList)
end;
(* Fix up the list of labels. *)
fun reallyFixBranches ([] : labels) cvec = ()
| reallyFixBranches (h::t) (cvec as Code{codeVec=cseg, ic, branchCheck, ...}) =
((case !h of
Jump8From addr =>
let
val offset : int = get8s (getAddr addr, cseg);
val diff : int = (!ic addrMinus addr) - 1;
in
branchCheck := !ic;
if is8Bit diff then () else raise InternalError "jump too large";
if offset <> 0
then raise InternalError "reallyFixBranches: jump already patched"
else set8s (diff, addr, cseg);
removeLabel (addr, cvec)
end
| Jump32From addr =>
let
val offset : int = get32s (getAddr addr, cseg);
val diff : int = (!ic addrMinus addr) - 4;
in
branchCheck := !ic;
if offset <> 0
then raise InternalError "reallyFixBranches: jump already patched"
else
(* A zero offset is more than simply redundant, it can
introduce zero words into the code which could be
taken as markers. It will not normally be produced
but can occur in very unusual cases. The only example
I've seen is a branch extension in a complicated series
of andalsos and orelses where the branch extension was
followed by an unconditional branch which was then backed
up by check_labs. We simply fill it with no-ops. *)
if diff = 0
then let
val a = getAddr addr;
val nop = Word8.fromInt (opToInt NOP);
in
csegSet (cseg, a - 1, nop);
csegSet (cseg, a, nop);
csegSet (cseg, a + 1, nop);
csegSet (cseg, a + 2, nop);
csegSet (cseg, a + 3, nop)
end
else
set32s (diff, addr, cseg)
end
);
reallyFixBranches t cvec
)
fun fixRecursiveBranches (cvec, target, []) = ()
| fixRecursiveBranches (cvec as Code{codeVec=cseg, ...}, target, addrH :: addrT) =
((case !addrH of
Jump8From addr =>
let
val offset : int = get8s (getAddr addr, cseg);
val diff : int = (target addrMinus addr) - 1;
in
if offset <> 0
then raise InternalError "fixRecursiveBranches: already patched"
else ();
if is8Bit diff
then set8s (diff, addr, cseg)
else raise InternalError "fixRecursiveBranches: branch too large"
end
| Jump32From addr =>
let
val offset : int = get32s (getAddr addr, cseg);
val diff : int = (target addrMinus addr) - 4;
in
if offset <> 0
then raise InternalError "fixRecursiveBranches: already patched"
else ();
if diff <> 0
then set32s (diff, addr, cseg)
else raise InternalError "fixRecursiveBranches: zero offset"
end
);
fixRecursiveBranches (cvec, target, addrT)
);
(* The address is the offset of the offset, not the instruction itself. *)
fun fixRecursiveCalls (cvec, target, []) = ()
| fixRecursiveCalls (cvec as Code{codeVec=cseg, ...}, target, addrH :: addrT) =
let
val instr : int = get8u (getAddr addrH - 1, cseg);
val offset : int = get32s (getAddr addrH, cseg);
val diff : int = (target addrMinus addrH) - 4;
in
if instr <> opToInt CALL_32
then raise InternalError "fixRecursiveCalls: not a call instruction"
else if offset <> 0
then raise InternalError "fixRecursiveCalls: already patched"
else if diff = 0
then raise InternalError "fixRecursiveCalls: zero offset"
else set32s (diff, addrH, cseg);
fixRecursiveCalls (cvec, target, addrT)
end;
(* Deal with a pending fix-up. *)
fun reallyFixup (cvec as Code{justComeFrom=ref [], ... }) = ()
| reallyFixup (cvec as Code{justComeFrom=jcf as ref labs, exited, ... }) =
(exited := false; reallyFixBranches labs cvec; jcf := []);
(* Adds the displacement to the stack pointer before an
instruction is generated. *)
fun resetSp (cvec as Code{stackReset, ...}) =
( (* Any pending jumps must be taken first. *)
reallyFixup cvec;
let
val sr = !stackReset * wordSize; (* Offset in bytes. *)
in
stackReset := 0;
if sr < 0 then raise InternalError "Negative stack reset" else
if is8Bit sr
then (* Can use one byte immediate *)
(
gen8u(rex{w=true, r=false, x=false, b=false}, cvec);
gen8u(opToInt Group1_8_A (* group1, 8-bit immediate *), cvec);
genmodrm(Register, arithOpToInt ADD, #1 (getReg esp), cvec);
gen8s(sr, cvec)
)
else (* Need 32 bit immediate. *)
(
gen8u(rex{w=true, r=false, x=false, b=false}, cvec);
gen8u(opToInt Group1_32_A (* group1, 32-bit immediate *), cvec);
genmodrm(Register, arithOpToInt ADD, #1 (getReg esp), cvec);
gen32s(sr, cvec)
)
end
) (* resetSp *);
(* Do any pending instructions, but only fix up branches if there
are instructions in the pipe-line. *)
fun flushQueue (Code{stackReset = ref 0, ...}) = ()
| flushQueue cvec = resetSp cvec;
(* Makes a new label. (no longer returns a list SPF) *)
fun makeShortLabel (addr: addrs, Code{longestBranch, labelList ,...}) : jumpFrom ref =
let
val lab = ref (Jump8From addr);
in
if addr addrLt ! longestBranch
then longestBranch := addr
else ();
labelList := lab :: ! labelList;
lab
end;
(* Apparently fix up jumps - actually just record where we have come from *)
fun fixup (labs:labels, cvec as Code{justComeFrom, exited, ic, branchCheck, procName, ...}) =
let
(* If the jump we are fixing up is immediately preceding,
we can remove it. It is particularly important to remove
32 bit jumps to the next instruction because they would
put a word of all zeros in the code, and that could be mistaken
for a marker word. *)
fun checkLabs [] = []
| checkLabs ((lab as ref (Jump8From addr))::labs) =
if !ic addrMinus addr = 1
then
(
if !ic addrLe !branchCheck
then raise InternalError "Backing up too far (8bit)"
else ();
ic := addr addrPlus ~1; (* Back up over the opCode *)
removeLabel (addr, cvec);
exited := false;
checkLabs labs
)
else lab :: checkLabs labs
| checkLabs ((lab as ref (Jump32From addr))::labs) =
if !ic addrMinus addr = 4
then
(
if !ic addrLe !branchCheck
then raise InternalError "Backing up too far (32bit)"
else ();
ic := addr addrPlus ~1; (* Back up over the opCode *)
exited := false;
checkLabs labs
)
else lab :: checkLabs labs
fun doCheck labs =
(* Repeatedly check the labels until we are no longer backing up.
We may have several to back up if we have just extended some
branches and then immediately fix them up. DCJM 19/1/01. *)
let
val lastIc = !ic
val newLabs = checkLabs labs
in
if lastIc = !ic then newLabs
else doCheck newLabs
end
in
case labs of
[] => () (* we're not actually jumping from anywhere *)
| _ =>
(
(* Any pending stack reset must be done now.
That may involve fixing up pending jumps because
so take effect before the stack adjustment. *)
flushQueue cvec;
(* Add together the jumps to here and remove redundant jumps. *)
justComeFrom := doCheck (labs @ !justComeFrom)
)
end;
fun checkBranchList
(cvec as Code{longestBranch, justComeFrom,
exited, ic, stackReset, labelList, ...})
(branched:bool) (size:int) =
(* If the longest branch is close to going out of range it must
be converted into a long form. *)
(* If we have just made an unconditional branch then we make the
distance shorter. *)
let
val maxDiff = (if branched then 100 else 127 - 5) - size;
(* See if we must extend some branches. If we are going to fix up a label
immediately we don't normally extend it. The exception is if we have
to extend some other labels in which case we may have to extend this
because the jumps we add may push this label out of range.
DCJM 9/4/01. *)
local
val icOffset =
if branched then !ic else !ic addrPlus 2 (* Size of the initial branch. *)
fun checkLab (lab as ref (Jump8From addr), n) =
if List.exists (fn a => a = lab) (! justComeFrom)
then n (* Don't include it here. *)
else if (icOffset addrMinus addr) + n > (100 - size) then n+5 else n
| checkLab (_, n) = n
(* Extending one branch may extend others. We need to process the list in
reverse order. *)
in
val jumpSpace = List.foldr checkLab 0 (!labelList)
end
(* Go down the list converting any long labels, and finding the
longest remaining. *)
fun convertLabels ([]:labList) : labList = []
| convertLabels (lab::labs) =
let
(* Process the list starting at the end. The reason for this
is that more recent labels appear before earlier ones.
We must put the earliest labels in first because they may
be about to go out of range. *)
val convertRest = convertLabels labs;
in
(* Now do this entry. *)
case !lab of
Jump32From addr => (* shouldn't happen? *)
convertRest
| Jump8From addr =>
(* If we are about to fix this label up we don't need to extend it except that we
must extend it if we are going to put in more branch extensions which will take
it out of range. DCJM 9/4/01. *)
if List.exists (fn a => a = lab) (! justComeFrom)
andalso (jumpSpace = 0 orelse !ic addrMinus addr < 127 - jumpSpace)
then lab :: convertRest
else if !ic addrMinus addr > (100 - size)
then (* Getting close - convert it. *)
(
reallyFixBranches [lab] cvec; (* fix up short jump to here *)
gen8u (opToInt JMP_32, cvec);
gen32u (0, cvec); (* long jump to final destination *)
lab := Jump32From (!ic addrPlus ~4);
(* Return the rest of the list. *)
convertRest
)
else
(
(* Not ready to remove this. Just find out if this is an
earlier branch and continue. *)
if addr addrLt ! longestBranch
then longestBranch := addr
else ();
lab :: convertRest
)
end (* convertLabels *);
in
if !ic addrMinus ! longestBranch > maxDiff
then let
(* Must save the stack-reset, otherwise "fixup" will try
to reset it. *)
val sr = !stackReset;
val _ = stackReset := 0;
(* Must skip round the branches unless we have just taken an
unconditional branch. *)
val lab =
if branched then []
else
(
exited := true;
gen8u (opToInt JMP_8, cvec);
gen8u (0, cvec);
[makeShortLabel (!ic addrPlus ~1, cvec)]
);
in
(* Find the new longest branch. *)
longestBranch := addrLast; (* Initial value. *)
labelList := convertLabels (!labelList);
fixup (lab, cvec); (* Continue with normal processing. *)
stackReset := sr (* Restore old value. *)
end
else ()
end;
(* Do all the outstanding operations including fixing up the branches. *)
fun doPending (cvec as Code{exited, stackReset=ref stackReset, ...}, size) : unit =
let
val mustReset = stackReset <> 0;
in
(* If we have not exited and there are branches coming in here
then we fix them up before jumping round any branch extensions. *)
if ! exited then () else reallyFixup cvec;
checkBranchList cvec (! exited) (if mustReset then size + 6 else size);
(* Fix up any incoming branches, including a jump round any
branch extensions. *)
reallyFixup cvec;
flushQueue cvec
end;
(* Generate an opCode byte after doing any pending operations. *)
fun genop (opb:opCode, rx, cvec) =
(
doPending (cvec, 15);
case rx of
NONE => ()
| SOME rxx => gen8u(rex rxx, cvec);
(* 15 is maximum size of an instruction;. It's also big
enough for a comparison and the following conditional
branch. *)
gen8u (opToInt opb, cvec)
);
(* This has to be done quite carefully if we are to be able to back-up
over jumps that point to the next instruction in fixup. We have to
guarantee that if we back up we haven't already set a jump to point
beyond where we're backing up. See below for more explanation.
DCJM 19/1/01.*)
fun putConditional (br: opCode, cvec as Code{ic, ...}) : jumpFrom ref =
(
flushQueue cvec; (* Do any stack adjustments. *)
gen8u (opToInt br, cvec); (* Don't use genop. *)
gen8u (0, cvec);
makeShortLabel (!ic addrPlus ~1, cvec)
);
(* Generates an unconditional branch. *)
fun unconditionalBranch (cvec as Code {justComeFrom, exited, ...}): labels =
let
(* If we have just jumped here we may be able to avoid generating a
jump instruction. *)
val U : unit = flushQueue cvec; (* Do any pending instructions. *)
val labs = ! justComeFrom;
in
justComeFrom := [];
(* We may get the sequence: jmp L1; L2: jmp L3.
If this is the "jmp L3" we can simply remember everything
that was supposed to jump to L2 and replace it with
jumps to L3. *)
(* This code has one disadvantage. If we have several short branches
coming here we don't record against the branches themselves that
they're all going to the same place. If we have to extend them
we put in separate long branches for each rather than pointing
them all at the same branch. This doesn't increase run-time
but makes the code larger than it need be. DCJM 1/1/01. *)
if ! exited
then labs
else
let
(* The code here has gone through various versions. The original
version always fixed up pending branches so that if we had a
short branch coming here we might avoid having to extend it.
A subsequent version separated out long and short branches
coming here and fixed up short branches but added long ones
onto the label list. I discovered a bug with this which
occurred when we put in branch extension code before an
unconditional branch and then backed up over the unconditional
branch and over one of the extended branches. Since we'd
already fixed up (really fixed up) the branch round the
branch extensions we ended up with that branch now pointing into
the middle of the code we subsequently generated.
We could get a similar situation if we have a conditional
branch immediately before this instruction and back up over
both, for example (if exp then () else (); ...). In that case
we have to make sure we haven't already fixed up another branch
to come here. Instead we must always add it onto the label list
so that we only (really) fix it up when we generate something other
than a branch. DCJM 19/1/01. *)
val br = putConditional (JMP_8, cvec);
in
exited := true;
br :: labs
end
end; (* unconditionalBranch *)
fun genSelfBranch (cvec as Code{justComeFrom, exited, ic, ... }) : labels =
let
(* Do any pending instructions. *)
val U : unit = flushQueue cvec;
(* Can we get into the prelude with an 8-bit jump? *)
(* Conservative estimation needs to allow for:
(1) stack check (10 bytes)
(2) 1 byte instruction + 1 byte offset (2 bytes)
(3) possible programmer arithmetic error (6 bytes)
*)
fun isNearPrelude addr = getAddr addr <= 110;
(* Is the jump long enough to reach back into the prelude? *)
fun isLongJump (Jump32From _ ) = true
| isLongJump (Jump8From addr) = isNearPrelude (addr addrPlus ~1)
val labs = ! justComeFrom;
val longJumps = List.filter (fn r => isLongJump (!r)) labs;
val shortJumps = List.filter (fn r => not (isLongJump (!r))) labs;
(* remove the "long enough" 8-bit jumps from the
list of pending jumps to extend. *)
fun tidy [] = ()
| tidy (ref (Jump32From _) :: t) = tidy t
| tidy (ref (Jump8From a) :: t) = (removeLabel (a, cvec); tidy t);
val U : unit = tidy longJumps;
(* do we actually need to insert a jump into the codestream? *)
val needsJump =
case shortJumps of
[] => not (! exited)
| _ => true;
in
if needsJump
then let
(* fix up pending short jumps to here *)
val U : unit = justComeFrom := shortJumps;
val U : unit = doPending (cvec, 5);
(* Now decide whether we can use an 8-bit jump here. *)
(* N.B. we use gen8u here, not genop, because the latter
calls "doPending (cvec, 12)" which could change the
results of our "isNearPrelude (!ic)" test. *)
val br =
if isNearPrelude (!ic)
then
(
gen8u (opToInt JMP_8, cvec);
gen8u (0, cvec);
ref (Jump8From (!ic addrPlus ~1))
)
else
(
gen8u (opToInt JMP_32, cvec);
gen32u (0, cvec);
ref (Jump32From (!ic addrPlus ~4))
);
in
justComeFrom := [];
exited := true;
br :: longJumps
end
else
(
justComeFrom := [];
exited := true;
longJumps
)
end; (* genSelfBranch *)
(* Exported. Adds in the reset. Does not actually generate code. *)
fun resetStack (offset, Code{stackReset, ...}) : unit =
stackReset := ! stackReset + offset;
(* Generate the modrm and, if necessary, sib bytes followed by the displacement. *)
fun genEACode (offset: int, rb: int, r: int, cvec) : unit =
let
val offsetCode =
(* don't generate [ebp] (use [ebp+0]) 'cos it doesn't exist! *)
if offset = 0 andalso rb <> 0x5
then Based (* no disp field *)
else if is8Bit offset
then Based8 (* use 8-bit disp field *)
else Based32; (* use 32-bit disp field *)
in
if rb = 0x4 (* Code for esp and r12 *)
then (* Need to use s-i-b byte. *)
(
genmodrm (offsetCode, r, 4 (* use s-i-b *), cvec);
gensib (Scale1, 4 (* no index *), rb, cvec)
)
else genmodrm(offsetCode, r, rb, cvec);
(* generate the disp field (if any) *)
case offsetCode of
Based8 => gen8s (offset, cvec)
| Based32 => gen32s (offset, cvec)
| _ => ()
end;
(* Generate a opcode plus a modrm byte. *)
fun genOpEA(opb:opCode, offset: int, rb: reg, r: reg, cvec): unit =
let
val (rbC, rbX) = getReg rb
val (rrC, rrX) = getReg r
in
doPending (cvec, 15);
(* For the moment always put in a REX prefix. *)
gen8u(rex{w=true, r=rrX, b=rbX, x = false}, cvec);
(* Generate the escape codes for the opcodes that need them. *)
case opb of
MOVZX => gen8u(opToInt ESCAPE, cvec)
| _ => ();
if offset < 0 andalso rb regEq esp then raise InternalError "Negative stack offset" else ();
gen8u(opToInt opb, cvec);
genEACode(offset, rbC, rrC, cvec)
end;
(* Generate a opcode plus a second modrm byte but where the "register" field in
the modrm byte is actually a code. *)
fun genOpPlus2(opb:opCode, offset: int, rb: reg, op2: int, cvec): unit =
let
val (rbC, rbX) = getReg rb
in
doPending (cvec, 15);
(* For the moment always put in a REX prefix. *)
(* If (opb = Group5 andalso op2 = 6 (* push *) orelse opb = POP_A)
andalso not rbX then we don't need it. *)
gen8u(rex{w=true, r=false, b=rbX, x = false}, cvec);
gen8u(opToInt opb, cvec);
genEACode(offset, rbC, op2, cvec)
end;
fun genOpReg(opb:opCode, rd: reg, rs: reg, cvec) =
let
val (rbC, rbX) = getReg rs
val (rrC, rrX) = getReg rd
in
doPending (cvec, 15);
(* For the moment always put in a REX prefix. *)
gen8u(rex{w=true, r=rrX, b=rbX, x = false}, cvec);
gen8u(opToInt opb, cvec);
genmodrm(Register, rrC, rbC, cvec)
end;
fun genOpRegPlus2(opb:opCode, rd: reg, op2: int, cvec) =
let
val (rrC, rrX) = getReg rd
in
doPending (cvec, 15);
(* For the moment always put in a REX prefix. *)
gen8u(rex{w=true, r=false, b=rrX, x = false}, cvec);
gen8u(opToInt opb, cvec);
genmodrm(Register, op2, rrC, cvec)
end;
(* Similar to genOpEA, but used when there is an index register.
rb may be regNone if no base register is required (used
with leal to tag values). *)
fun genOpIndexed (opb:opCode, offset: int, rb: reg, ri: reg, size : scaleFactor, rd: reg, cvec) =
let
val (rbC, rbX) = getReg rb
val (riC, riX) = getReg ri
val (rrC, rrX) = getReg rd
val offsetCode = (* If rb is ebp or r13 we must put in a one byte displacement. *)
if rb regEq regNone orelse offset = 0 andalso rbC <> 0x5
then Based (* no disp field *)
else if is8Bit offset
then Based8 (* use 8-bit disp field *)
else Based32; (* use 32-bit disp field *)
val basefield =
if rb regEq regNone
then 5 (* no base register *)
else rbC;
in
doPending (cvec, 15);
(* For the moment always put in a REX prefix. *)
gen8u(rex{w=true, r=rrX, b=rbX andalso rb regNeq regNone, x=riX}, cvec);
(* Generate the ESCAPE code if needed. *)
case opb of
MOVZX => gen8u(opToInt ESCAPE, cvec)
| _ => ();
gen8u(opToInt opb, cvec);
genmodrm (offsetCode, rrC, 4 (* s-i-b *), cvec);
gensib (size, riC, basefield, cvec);
(* generate the disp field (if any) *)
case offsetCode of
Based8 => gen8s (offset, cvec)
| Based32 => gen32s (offset, cvec)
| _ => if rb regEq regNone (* 32 bit absolute used as base *)
then gen32s (offset, cvec)
else ()
end;
fun genPushPop(opc: int->opCode, r: reg, cvec) =
let
val (rc, rx) = getReg r
in
(* These are always 64-bit but a REX prefix may be needed for the register. *)
genop(opc rc, if rx then SOME{w=true, b = true, x=false, r = false } else NONE, cvec)
end;
(* Tag the value in register r *)
fun genTag (r : reg, cvec) : unit =
genOpIndexed(LEAL, 1, r, r, Scale1, r, cvec);
fun genImmed (opn: arithOp, rd: reg, imm: int, cvec) : unit =
if is8Bit imm
then (* Can use one byte immediate *)
(
genOpRegPlus2 (Group1_8_A, rd, arithOpToInt opn, cvec);
gen8s (imm, cvec)
)
else (* Need 32 bit immediate. *)
(
genOpRegPlus2 (Group1_32_A, rd, arithOpToInt opn, cvec);
gen32s(imm, cvec)
);
fun genReg (opn: arithOp, rd: reg, rs: reg, cvec) : unit =
genOpReg (Arith (opn, 3 (* r/m to reg *)), rd, rs, cvec)
(* generate padding no-ops to align to n modulo 4 *)
fun align (n:int, cvec as Code{ic, ...}) =
while n - getAddr (!ic) mod 4 <> 0
do genop (NOP, NONE, cvec);
(* Exported. - movl offset(rb),rd. *)
fun genLoad (offset: int, rb: reg, rd: reg, cvec) : unit =
if rd regEq regHandler (* Not a real register. *)
then
(
(* pushl offset(rb); popl 4(ebp) *)
(* This only happens when we are popping the handler so we
could simply pop it straight. *)
genOpPlus2(Group5, offset, rb, 6 (* push *), cvec);
genOpPlus2(POP_A, MemRegHandlerRegister, ebp, 0, cvec)
)
else genOpEA(MOVL_A_R, offset, rb, rd, cvec);
datatype storeWidth = STORE_WORD | STORE_BYTE
fun isIndexedStore _ = true (* Yes, for both word and byte. *)
(* Exported - Can we store the value without going through a register?
Only if it is short and will fit in 32 bits. *)
fun isStoreI (cnstnt: machineWord, _, _) = isTagged32bitS cnstnt;
(* Store an immediate value at a given address and offset. *)
fun genStoreI (cnstnt: machineWord, offset: int, rb: reg, STORE_WORD, ri: reg,
cvec as Code{ic, ...}) =
(
(* There was a problem here on the 32-bit machine where we could have
an immediate that was nearly all zero combining with zeros in other
bytes to give a words of zeros. That shouldn't be a problem on 64-bits
because we don't have 64-bit immediates. *)
if ri regEq regNone
then genOpPlus2(MOVL_32_A, offset, rb, 0, cvec)
else genOpIndexed(MOVL_32_A, offset-wordSize div 2, rb, ri,
if wordSize = 4 then Scale2 else Scale4, mkReg(0, false), cvec);
if isShort cnstnt
then gen32s (tag(toInt (toShort cnstnt)), cvec)
else addConstToVec (WVal cnstnt, InlineAbsolute, cvec)
)
| genStoreI (cnstnt: machineWord, offset: int, rb: reg, STORE_BYTE, ri: reg, cvec) : unit =
if not (isShort cnstnt)
then (* This should never happen. *)
raise InternalError "genStoreI: storing long constant as a byte"
else
let
val v = toInt (toShort cnstnt)
in
if ri regEq regNone
then
(
genOpPlus2 (MOVB_8_A, offset, rb, 0, cvec);
gen8u (v, cvec)
)
else
(
(* Untag the index first. *)
genOpRegPlus2 (Group2_1_A, ri, 5 (* shr *), cvec);
genOpIndexed(MOVB_8_A, offset, rb, ri, Scale1, mkReg(0, false), cvec);
gen8u (v, cvec);
(* Retag the index. *)
genTag (ri, cvec)
)
end
(* Exported. *)
(* Store a value on the stack. This is used when the registers need to be
saved, for more than 4 arguments or to push an exception handler. *)
fun genPush (r:reg, cvec) : unit =
if r regEq regHandler (* Not a real register. *)
then genOpPlus2(Group5, MemRegHandlerRegister, ebp, 6 (* push *), cvec)
else genPushPop (PUSH_R, r, cvec);
(* Exported. Load a value and push it on the stack. Used when all
the allocatable registers have run out.
Also used if preferLoadPush is true. *)
fun genLoadPush (offset: int, rb: reg, cvec) : unit =
genOpPlus2(Group5, offset, rb, 6 (* push *), cvec)
val preferLoadPush : bool = true; (* It's cheap. *)
(* Call the function. Must ensure that the return address
is on a word + 2 byte boundary. *)
(* Call the function. Must ensure that the return address
is on a word + 2 byte boundary. *)
fun genSelfCall (cvec as Code{ic, ...}) : addrs =
(
(* Make sure anything pending is done first. *)
(* 15 comes from the maximum instruction size (12)
used in genop, together with up to 3 nops. *)
doPending (cvec, 15+3);
(* Ensure the return address is aligned on
a word + 2 byte boundary. *)
align (1, cvec);
genop (CALL_32, NONE, cvec); (* 1 byte *)
gen32u (0, cvec); (* 4 bytes *)
if getAddr (!ic) mod 4 <> 2
then raise InternalError "genSelfCall: call not aligned"
else ();
!ic addrPlus ~4
);
(* Register/register move. *)
fun genMove (rd:reg, rs:reg, cvec) : unit =
(* Because we're using the register to EA instruction here the destination
register is encoded in the "base" part and the source register in
the "reg" part. *)
genOpReg (MOVL_R_A, rs,rd, cvec)
(* Add a register to a constant. *)
fun genLeal (rd:reg, rs:reg, offset:int, cvec) : unit =
genOpEA (LEAL, offset, rs, rd, cvec);
(* Exported. - movl rs,offset(rb) / movb rs,offset(rb) *)
fun genStore (rs: reg, offset: int, rb: reg, STORE_WORD, ri, cvec) : unit =
if ri regEq regNone
then genOpEA (MOVL_R_A, offset, rb, rs, cvec)
else genOpIndexed(MOVL_R_A, offset- wordSize div 2, rb, ri,
if wordSize = 4 then Scale2 else Scale4, rs, cvec)
| genStore (rs: reg, offset: int, rb: reg, STORE_BYTE, ri, cvec) : unit =
let
(* Only some of the 32 bit registers can be used to store from.
Eax, Ebx, Ecx and Edx are fine and the code used for those
registers corresponds to their low-order byte. The other
registers can't be used. In the absence of some way of
telling the higher-level code-generator about this or finding
out if a suitable register is free if we are given one of
the other registers we just have to make the best of it. *)
val regToStore =
if #1 (getReg rs) <= 3 then rs (* No problem *)
else if rb regNeq eax andalso ri regNeq eax then eax
else if rb regNeq ebx andalso ri regNeq ebx then ebx
else ecx (* Must be free *)
in
if rs regEq regToStore then ()
else (* Have to move the value into the the right register. *)
(
genPush(regToStore, cvec);
genMove(regToStore, rs, cvec)
);
(* The value we store has to be untagged. *)
genOpRegPlus2 (Group2_1_A, regToStore, 5 (* shr *), cvec);
if ri regEq regNone
then ()
else (* Untag the index as well. *)
genOpRegPlus2 (Group2_1_A, ri, 5 (* shr *), cvec);
if ri regEq regNone
then genOpEA (MOVB_R_A, offset, rb, regToStore, cvec)
else genOpIndexed(MOVB_R_A, offset, rb, ri, Scale1, regToStore, cvec);
(* Restore the original value, either by popping or by retagging.
This ensures we don't have a bad value around and also
restores the original value since it may still be wanted. *)
if rs regEq regToStore
then genTag(rs, cvec)
else genPushPop(POP_R, regToStore, cvec);
if ri regEq regNone
then ()
else (* retag the index as well. *) genTag (ri, cvec)
end;
(* Move an immediate value into a register. The immediate value may be
any 32-bit number. *)
fun genMoveI (rd:reg, immed:int, cvec) =
(* There may be better ways e.g. xorl rd,rd; addl immed,rd. *)
if not isX64 andalso isTagged32bitS(toMachineWord immed)
then (* This is better on X64 but longer than a 32 bit immediate on i386 *)
(
genOpRegPlus2 (MOVL_32_A, rd, 0, cvec);
gen32s (immed, cvec)
)
else
let
val (rc, rx) = getReg rd
in
genop (MOVL_32_64_R rc, SOME {w=true, r=false, b=rx, x=false}, cvec);
if isX64 then gen64s (immed, cvec) else gen32s (immed, cvec)
end;
(* Exported. *)
fun genLoadCoderef (c:code, rd:reg, cvec) : unit =
let
val (rc, rx) = getReg rd
in
genop (MOVL_32_64_R rc, SOME {w=true, r=false, b=rx, x=false}, cvec);
codeConst (c, InlineAbsolute, cvec)
end
type handlerLab = addrs ref;
(* Exported. Loads the address of the destination of a branch. Used to
put in the address of the exception handler.
We used to have pushAddress in place of this which pushed the
address at the same time. On this architecture it can save an
instruction but it's a problem on machines where we have to load
the address into a register - we don't have a spare checked
register available. *)
fun loadHandlerAddress (rd, cvec) : handlerLab =
let
val lab = ref addrZero;
val (rc, rx) = getReg rd
in
genop (MOVL_32_64_R rc, SOME {w=true, r=false, b=rx, x=false}, cvec);
addConstToVec (HVal lab, InlineAbsolute, cvec);
lab
end;
(* Exported *)
fun fixupHandler (lab:handlerLab, cvec as Code{exited, ic, branchCheck, ...}) : unit =
(
(* Make sure anything pending is done first. *)
(* 15 comes from maximum instruction size + up to 3 nops. *)
doPending (cvec, 15+3);
(* Ensure the return address is aligned onto a word + 2 byte
boundary. *)
align (2, cvec);
exited := false;
branchCheck := !ic;
lab := !ic
);
datatype callKinds =
Recursive (* The function calls itself. *)
| ConstantFun of machineWord * bool (* A pre-compiled or io function. *)
| CodeFun of code (* A static link call. *)
| FullCall (* Full closure call *)
(*****************************************************************************
Calling conventions:
FullCall:
the caller loads the function's closure into regClosure and then
(the code here) does an indirect jump through it.
Recursive:
the caller loads its own function's closure/static-link into regClosure
and the code here does a jump to the start of the code.
ConstantFun:
a direct or indirect call through the given address. If possible the
caller will have done the indirection for us and passed false as the
indirection value. The exception is calls to IO functions where the
address of the code itself is invalid. If the closure/static-link
value is needed that will already have been loaded.
CodeFun:
the same as ConstantFun except that this is used only for static-link
calls so is never indirect.
*****************************************************************************)
(* Call a function. *)
fun callFunction (callKind,
cvec as Code {selfCalls, mustCheckStack, ic, ... }) : unit =
(* If we ever call a function we must do a stack check. *)
(
mustCheckStack := true;
case callKind of
Recursive =>
let
val lab : addrs = genSelfCall cvec;
in
selfCalls := lab :: ! selfCalls
end
| FullCall => (* Indirect call through closure reg. *)
(
(* Make sure anything pending is done first. *)
(* 18 comes from the maximum instruction size (12) used in genop,
together with up to 3 nops. *)
doPending (cvec, 15+3);
(* Ensure the return address is aligned on
a word + 2 byte boundary. *)
align (0, cvec);
genop (Group5, NONE, cvec);
genmodrm(Based, 2 (* call *), #1 (getReg regClosure), cvec)
)
| CodeFun c =>
(
(* Make sure anything pending is done first. *)
doPending (cvec, 15+3);
(* Ensure the return address is aligned on
a word + 2 byte boundary. *)
align (0, cvec);
genop (Group5, NONE, cvec);
genmodrm(Based, 2 (* call *), 5 (* PC rel *), cvec);
codeConst (c, ConstArea 0, cvec)
)
| ConstantFun(w, false) =>
(
(* Make sure anything pending is done first. *)
doPending (cvec, 15+3);
(* Ensure the return address is aligned on
a word + 2 byte boundary. *)
align (0, cvec);
genop (Group5, NONE, cvec);
genmodrm(Based, 2 (* call *), 5 (* PC rel *), cvec);
addConstToVec (WVal w, ConstArea 0, cvec)
)
| ConstantFun(w, true) =>
let
val (rc, rx) = getReg regClosure
in
genop (MOVL_32_64_R rc, SOME {w=true, r=false, b=rx, x=false}, cvec);
addConstToVec (WVal w, InlineAbsolute, cvec);
(* Make sure anything pending is done first. *)
doPending (cvec, 15+3);
(* Ensure the return address is aligned on
a word + 2 byte boundary. *)
align (0, cvec);
genop (Group5, NONE, cvec);
genmodrm(Based, 2 (* call *), rc, cvec)
end;
if getAddr (!ic) mod 4 <> 2 (* This can be 8byte + 2 or 8byte + 6. *)
then raise InternalError "callFunction: call not aligned"
else ()
);
(* Exported. Tail recursive jump to a function.
N.B. stack checking is used both to ensure that the stack does
not overflow and also as a way for the RTS to interrupt the code
at a safe place. The RTS can set the stack limit "register" at any
time but the code will only take a trap when it next checks the
stack. The only way to break out of infinite loops is for the
user to type control-C and some time later for the code to do a
stack check. We need to make sure that we check the stack in any
function that my be recursive, directly or indirectly.
*)
fun jumpToFunction (callKind, returnReg,
cvec as Code{selfJumps, exited, mustCheckStack, ...}) =
( (* Must push the return register? *)
if returnReg regNeq regNone
then genPushPop(PUSH_R, returnReg, cvec)
else ();
case callKind of
Recursive =>
let
val U : unit = mustCheckStack := true;
val lab = genSelfBranch cvec;
in
selfJumps := lab @ ! selfJumps
end
| FullCall =>
( (* Full closure call *)
mustCheckStack := true;
genop (Group5, NONE, cvec);
genmodrm(Based, 4 (* jmp *), #1 (getReg regClosure), cvec)
)
| CodeFun c =>
(
mustCheckStack := true; (* May be recursive. *)
genop (Group5, NONE, cvec);
genmodrm(Based, 4 (* jmp *), 5 (* PC rel *), cvec);
codeConst (c, ConstArea 0, cvec)
)
| ConstantFun(w, false) =>
(
mustCheckStack := true; (* May be recursive. *)
genop (Group5, NONE, cvec);
genmodrm(Based, 4 (* jmp *), 5 (* PC rel *), cvec);
addConstToVec (WVal w, ConstArea 0, cvec)
)
| ConstantFun(w, true) =>
(* Indirect jumps are used to call into the RTS. No need
to check the stack. *)
let
val (rc, rx) = getReg regClosure
in
genop (MOVL_32_64_R rc, SOME {w=true, r=false, b=rx, x=false}, cvec);
addConstToVec (WVal w, InlineAbsolute, cvec);
genop (Group5, NONE, cvec);
genmodrm(Based, 4 (* jmp *), rc, cvec)
end;
exited := true (* We're not coming back. *)
);
(* Exported. Return and remove args. *)
fun returnFromFunction (resReg, args, cvec as Code{exited, ...}) : unit =
(if resReg regNeq regNone
then raise InternalError "Wrong argument"
else if args = 0
then genop (RET, NONE, cvec)
else let
val offset = args * wordSize
in
genop (RET_16, NONE, cvec);
gen8u (offset mod exp2_8, cvec);
gen8s (offset div exp2_8, cvec);
(* Assume it will fit in 16 bits. *)
if offset > 32767
then raise InternalError "Return offset too large"
else ()
end;
exited := true (* We're not coming back. *)
);
(* Exported. The exception argument has already been loaded into eax *)
(* Call, rather than jump to, the exception code so that we have
the address of the caller if we need to produce an exception
trace. *)
fun raiseException cvec =
(
doPending (cvec, 15+3);
(* Since we're calling we put the "return address" on a word+2 byte
boundary. This is never actually used as a return address but
it's probably best to make sure it's properly aligned. It probably
simplifies exception tracing which is the reason it's there. *)
align (3, cvec);
genop(Group5, NONE, cvec);
genmodrm (Based8, 2 (* call *), #1 (getReg ebp), cvec);
gen8u (MemRegRaiseException, cvec)
)
(* Exported. Set a register to a particular offset in the stack. *)
fun genStackOffset (reg, byteOffset, cvec) : unit =
if byteOffset = 0
then genMove (reg, regStackPtr, cvec)
else genLeal (reg, regStackPtr, byteOffset, cvec)
(* Only used for while-loops. *)
fun jumpback (lab, stackCheck, cvec as Code{exited, ic, ...}) : unit =
(
(* Put in a stack check. This is used to allow the code to be interrupted. *)
if stackCheck
then
(
(* cmp reg,16(%ebp)*)
genOpEA(Arith (CMP, 3), MemRegStackLimit, ebp, esp, cvec);
(* jnb 3 *)
let
val lab = [putConditional (JNB, cvec)]
in
(* call *)
genop(Group5, NONE, cvec);
genmodrm (Based8, 2 (* call *), #1 (getReg ebp), cvec);
gen8u (MemRegStackOverflowCall, cvec);
fixup (lab, cvec)
end
)
else ();
(* Do any pending instructions before calculating the offset, just
in case we put in some instructions first. *)
doPending (cvec, 15);
let
val offset = lab addrMinus (!ic); (* Negative *)
val offset2 = offset - 2;
in
if is8Bit offset2
then
(
genop (JMP_8, NONE, cvec);
gen8s (offset2, cvec)
)
else
(
genop (JMP_32, NONE, cvec);
gen32s (offset - 5, cvec)
)
end;
exited := true
);
(* Allocate store and put the resulting pointer in the result register. *)
fun allocStore (size, flag, resultReg, cvec) : unit =
let
val bytes = (size + 1) * wordSize;
val lengthWord = size + (Word8.toInt flag * exp2_56); (* Size + mutable. *)
in
(* subl (size+1)*4,r15; cmpl r15,8(%ebp); jnb 1f;
call 40[%ebp]; 1f: movl size,-4(r15); movl r15,r *)
genLeal (r15, r15, ~ bytes, cvec); (* TODO: What if it's too big to fit? *)
genOpEA(Arith (CMP, 3 (* r/m to reg *)), MemRegLocalMbottom, ebp, r15, cvec);
let
val lab = [putConditional (JNB, cvec)]
in
(* If we don't have enough store for this allocation we call this
function. *)
genop (Group5, NONE, cvec);
genmodrm(Based8, 2 (* call *), #1 (getReg ebp), cvec);
gen8s (MemRegHeapOverflowCall, cvec);
fixup (lab, cvec)
end;
(* Set the size field of a newly allocated piece of store. *)
genOpPlus2 (MOVL_32_A, ~wordSize, r15, 0, cvec);
gen32s (size, cvec);
(* Set the flag byte separately. *)
if flag <> 0w0
then
(
genOpPlus2 (MOVB_8_A, ~1, r15, 0, cvec);
gen8s (Word8.toInt flag, cvec)
)
else ();
(* TODO: What if the length won't fit in 32 bits? *)
genMove(resultReg, r15, cvec)
end;
(* Remove the mutable bit by clearing the flag byte. *)
fun setFlag (baseReg, cvec, flag) : unit =
genStoreI (toMachineWord flag, ~1, baseReg, STORE_BYTE, regNone, cvec);
(* Small tuples and closures are created by allocating the space and
storing into it without setting the mutable bit. This is safe
provided there are no traps until all the values have been stored,
and gcode checks for this by loading all the values, apart from
constants, into registers. We have to make sure that we don't mess
this up by reordering a load instruction (which might cause a
persistent store trap) before the final store. Gcode calls
"completeSegment" after the last store and we flush the queue just
to be on the safe side. *)
(* This comment applied to the (very) old persistent store system and
is no longer relevant. I've left the comment in because there may
be code that assumes that this is still necessary. DCJM June 2006. *)
val completeSegment = flushQueue;
datatype instrs =
instrMove
| instrAddA
| instrSubA
| instrRevSubA
| instrMulA
| instrAddW
| instrSubW
| instrRevSubW
| instrMulW
| instrDivW
| instrModW
| instrOrW
| instrAndW
| instrXorW
| instrLoad
| instrLoadB
| instrVeclen
| instrVecflags
| instrPush
| instrUpshiftW (* logical shift left *)
| instrDownshiftW (* logical shift right *)
| instrDownshiftArithW (* arithmetic shift right *)
| instrGetFirstLong
| instrStringLength
| instrSetStringLength
| instrBad;
(* Can the we use the same register as the source and destination
of an instructions? On this machine - no. *)
val canShareRegs : bool = false;
(* Is there a general register/register operation? Some operations may not
be implemented because this machine does not have a suitable instruction
or simply because they have not yet been added to the code generator. It
is possible for an instruction to be implemented as a register/immediate
operation but not as a register/register operation (e.g. multiply) *)
fun instrIsRR instrUpshiftW = false (* General shifts require CL register *)
| instrIsRR instrDownshiftW = false
| instrIsRR instrDownshiftArithW = false
(* | instrIsRR instrMulA = false (* Too complicated to implement *)
| instrIsRR instrMulW = false (* Too complicated to implement *)
| instrIsRR instrDivW = false
| instrIsRR instrModW = false
*)
| instrIsRR _ = true
;
datatype tests =
Length of opCode (* always a conditional jump *)
| Arb of opCode
| Wrd of opCode;
val Short = Length JNE;
val Long = Length JE;
val testNeqW = Wrd JNE;
val testEqW = Wrd JE;
val testGeqW = Wrd JNB; (* These are UNsigned comparisons *)
val testGtW = Wrd JA;
val testLeqW = Wrd JNA;
val testLtW = Wrd JB;
val testNeqA = Arb JNE;
val testEqA = Arb JE;
val testGeqA = Arb JGE;
val testGtA = Arb JG;
val testLeqA = Arb JLE;
val testLtA = Arb JL;
(* Is this argument acceptable as an immediate or should it be loaded into a register? *)
local
fun isPower2 n =
let
fun p2 i = if i < n then p2 (i*2) else i = n
in
n > 0 andalso p2 1
end
in
fun instrIsRI (i, cnstnt) =
case i of
instrBad => false
| instrRevSubA => false
| instrRevSubW => false
| instrMove => true
| instrPush => true
| instrMulA => isShort cnstnt andalso toInt (toShort cnstnt) = 2
| instrMulW => isShort cnstnt andalso isPower2(toInt (toShort cnstnt))
| instrDivW => isShort cnstnt andalso isPower2(toInt (toShort cnstnt))
| instrModW => isShort cnstnt andalso isPower2(toInt (toShort cnstnt))
| instrSetStringLength => false (* The string length is untagged and so
it's not safe to put it inline. *)
| _ => isTagged32bitS cnstnt (* All others must be short *)
end;
(* Test a single argument and trap if it is long. The result is
the instruction address of the trap, and is used to jump back to
if the instruction overflows. *)
fun tagTest1 (r: reg, cvec as Code{ic, ...}) =
let
val (regNum, rx) = getReg r;
in
if r regEq eax
then (* Special instruction for testing accumulator. Can use an
8-bit test. *)
(
genop (TEST_ACC8, NONE, cvec);
gen8u (1, cvec)
)
else
( (* We can use a REX code to force it to always use the low order byte. *)
genop (Group3_a,
if rx orelse regNum >= 4 then SOME{w=false, r=false, b=rx, x=false} else NONE, cvec);
genmodrm (Register, 0 (* test *), regNum, cvec);
gen8u(1, cvec)
);
let
val lab = putConditional (JNE, cvec); (* generates code *)
val jumpback = !ic;
in
genop(Group5, NONE, cvec);
genmodrm (Based8, 2 (* call *), #1 (getReg ebp), cvec);
gen8u (MemRegArbEmulation, cvec);
fixup ([lab], cvec);
jumpback
end
end; (* tagTest1 *)
(* Test a pair of arguments and trap if either is long. The result is
the instruction address of the trap, and is used to jump back to
if the instruction overflows. *)
fun tagTest2 (rd:reg, r1:reg, r2:reg, useOr:bool, cvec as Code{ic, ...}) : addrs =
let
(* In most cases we have to trap if EITHER is long, and so we AND
together the arguments. However if we are comparing two arbitrary
precision values for (in)equality, we need only trap if BOTH are
long, since if one is long and the other not then they are both
definitely different. *)
val testOP = if useOr then OR else AND;
in
(* If the destination register is not the same as either source
register we can use that directly, otherwise we have to push it. *)
(* bug-fixed SPF 3/1/95 - now checks both r1 and r2. *)
if rd regEq r1
then
(
genPush (rd, cvec);
genReg (testOP, rd, r2, cvec)
)
else if rd regEq r2
then
(
genPush (rd, cvec);
genReg (testOP, rd, r1, cvec)
)
else
(
genMove (rd, r1, cvec);
genReg (testOP, rd, r2, cvec)
);
if rd regEq eax
then (* Special instruction to test the accumulator. *)
(
genop (TEST_ACC8, NONE, cvec);
gen8u (1, cvec)
)
else
let
val (regNum, rx) = getReg rd
in (* We can use a REX code to force it to always use the low order byte. *)
genop (Group3_a,
if rx orelse regNum >= 4 then SOME{w=false, r=false, b=rx, x=false} else NONE, cvec);
genmodrm (Register, 0 (* test *), regNum, cvec);
gen8u(1, cvec)
end;
(* restore rd if it's used as a source register *)
if (rd regEq r1) orelse (rd regEq r2)
then genPushPop(POP_R, rd, cvec)
else ();
let
val lab = putConditional (JNE, cvec); (* generates code *)
val jumpback = !ic;
in
genop(Group5, NONE, cvec);
genmodrm (Based8, 2 (* call *), #1 (getReg ebp), cvec);
gen8u (MemRegArbEmulation, cvec);
fixup ([lab], cvec);
jumpback
end
end;
(* generate the "jump on overflow" used to
implement arbitrary-precision operations. *)
fun genJO8 (addr, cvec as Code{ic, ...}) =
let
val U : unit = doPending (cvec, 15);
val here = !ic;
(* jump address calculations are relative to the value
of the program counter *after* the instruction *)
val offset = addr addrMinus (here addrPlus 2);
in
gen8u (opToInt JO, cvec);
gen8s (offset, cvec)
end
(* All these can be handled. *)
fun isCompRR tc = true;
(* Is this argument acceptable as an immediate or should it be loaded into a register? *)
fun isCompRI (tc, cnstnt) =
case tc of
Length _ => true
| _ => isTagged32bitS cnstnt; (* The compare instruction sign-extends immediates. *)
(* Fixed and arbitrary precision comparisons. *)
fun compareAndBranchRR (r1, r2, tc, cvec) : labels =
case tc of
Wrd opc =>
(
genReg (CMP, r1, r2, cvec);
[putConditional (opc, cvec)]
)
| Arb opc =>
let
(* Test the tags. If we are testing for equality we can OR the
tags and only trap if both arguments are long. Try to use
eax, ebx, ecx or edx for rd because then we can use a single
byte test. *)
val rd = if #1 (getReg r1) > 4 then r2 else r1;
val useOr =
case opc of
JE => true
| JNE => true
| _ => false
;
in
tagTest2 (rd, r1, r2, useOr, cvec);
genReg (CMP, r1, r2, cvec);
[putConditional (opc, cvec)]
end
| _ =>
raise InternalError "Should not have short/long here";
fun compareAndBranchRI (r, cnstnt, tc, cvec) : labels =
let
val c = toInt (toShort cnstnt);
in
case tc of
Length opc =>
(
(* Since we're only interested in the bottom bit we could use an
8-bit test, provided the value was in eax, ebx, ecx, edx. *)
genOpRegPlus2 (Group3_A, r, 0, cvec);
gen32u (1, cvec);
[putConditional (opc, cvec)]
)
| Wrd opc =>
(
genImmed (CMP, r, tag c, cvec);
[putConditional (opc, cvec)]
)
| Arb opc =>
(
(* Can do tests for (in)equality on arbitrary precision values
without checking the tags since the values we're comparing
against are short constants. *)
case opc of
JE => ()
| JNE => ()
| _ => (tagTest1 (r, cvec) : addrs; ())
;
genImmed (CMP, r, tag c, cvec);
[putConditional (opc, cvec)]
)
end;
val inlineAssignments = true;
(* Common code for div and mod word. They are identical apart from
getting the result. *)
fun divmodWord(isDiv: bool, r1:reg, r2:reg, rd:reg, cvec) : unit =
let
(* The divisor needs to be a different register from
either eax or edx. It also needs to be different from
r1 since we're going to modify divReg before we load r1
into eax. *)
val divReg =
if rd regNeq eax andalso rd regNeq edx
then rd
else if r1 regNeq ecx
then ecx
else esi
in
(* This is a bit complicated because the result is always placed
in the EDX:EAX register pair so we have to save one or both. *)
if rd regNeq eax
then genPush(eax, cvec) else ();
if rd regNeq edx
then genPush(edx, cvec) else ();
if divReg regNeq rd
then genPush(divReg, cvec) else ();
(* Untag, but don't shift, the divisor and dividend. *)
if r2 regEq eax
then
(
if divReg regEq r1
then raise InternalError "Assertion failed: Invalid registers"
else ();
(* We use move followed by substraction since that tests the
result for zero. *)
genMove(divReg, r2, cvec);
genImmed (SUB, divReg, 1, cvec);
genLeal (eax, r1, ~1, cvec)
)
else
(
genLeal (eax, r1, ~1, cvec);
genMove(divReg, r2, cvec);
genImmed (SUB, divReg, 1, cvec)
);
let
val lab = [putConditional (JNE, cvec)]
in
(* call *) (* Use a call so we can get an exception trace. *)
genop(Group5, NONE, cvec);
genmodrm (Based8, 2 (* call *), #1 (getReg ebp), cvec);
gen8u (MemRegRaiseDiv, cvec);
fixup (lab, cvec)
end;
(* Do the division. *)
genReg (XOR, edx, edx, cvec);
genOpRegPlus2 (Group3_A, divReg, 6 (* div *), cvec);
if isDiv
then (* Tag the result into the result register. *)
genOpIndexed(LEAL, 1, eax, eax, Scale1, rd, cvec)
else (* Add the tag back into the remainder. *)
genLeal (rd, edx, 1, cvec);
(* Restore the saved registers. N.B. This also has
the effect of making sure that both eax and edx contain
valid values. *)
if divReg regNeq rd
then genPushPop(POP_R, divReg, cvec) else ();
if rd regNeq edx
then genPushPop(POP_R, edx, cvec) else ();
if rd regNeq eax
then genPushPop(POP_R, eax, cvec) else ()
end
(* General register/register operation. *)
fun genRR (inst, r1:reg, r2:reg, rd:reg, cvec) : unit =
(
if (rd regEq r1) orelse (rd regEq r2)
then raise InternalError "Registers must be different"
else ();
case inst of
instrMove =>
(* Move from one register to another. r2 is ignored *)
if rd regEq regHandler (* Not a real register. *)
then genStore (r1, MemRegHandlerRegister, ebp, STORE_WORD, regNone, cvec)
else genMove (rd, r1, cvec)
| instrPush =>
(* Both rd and r2 are ignored. *)
genPush (r1, cvec)
| instrAddA =>
(* Arbitrary precision addition. *)
(* If either argument is long, or if both arguments are short
but the result overflows, the code branchs to "addr". This
executes a trap which gets us into the run-time system which
then emulates the instructions, using long arithmetic.
Isn't that cute? To make it work, we have to be sure that
the source and destination registers are different, because
otherwise we wouldn't be able to perform the emulation
following an arithmetic overflow.
Warning: since the RTS can only emulate a few instructions,
we have to be very careful about what code we generate here.
For example, we use 2-byte "leal" instructions for tagging and
untagging rather than 1-byte "add" instructions because the
emulation has to treat these two operations differently.
SPF 4/1/95 *)
let
val addr = tagTest2 (rd, r1, r2, false, cvec); (* generates code *)
in
(* Do the actual operation after removing a tag from one arg. *)
genLeal (rd, r1, ~1, cvec);
genReg (ADD, rd, r2, cvec);
genJO8 (addr, cvec)
end
| instrSubA =>
(* Arbitrary precision subtraction. *)
let
val addr = tagTest2 (rd, r1, r2, false, cvec); (* generates code *)
in
(* Do the actual operation after removing a tag from one arg. *)
genMove (rd, r1, cvec);
genReg (SUB, rd, r2, cvec);
genLeal (rd, rd, 1, cvec); (* Put back the tag. *)
genJO8 (addr, cvec)
end
| instrRevSubA =>
(* Arbitrary precision subtraction. *)
let
val addr = tagTest2 (rd, r1, r2, false, cvec); (* generates code *)
in
(* Do the actual operation after removing a tag from one arg. *)
genMove (rd, r2, cvec);
genReg (SUB, rd, r1, cvec);
genLeal (rd, rd, 1, cvec); (* Put back the tag. *)
genJO8 (addr, cvec)
end
| instrMulA =>
(* Arbitrary precision multiplication. *)
let
val addr = tagTest2 (rd, r1, r2, false, cvec); (* generates code *)
in
(* This is a bit complicated because the result is always placed
in the EDX:EAX register pair so we have to save one or both. *)
(* If the multiply overflows we need to be able to recover the
original arguments in order to emulate the instruction. *)
if rd regNeq eax
then genPush(eax, cvec) else ();
if rd regNeq edx
then genPush(edx, cvec) else ();
if r2 regEq edx
then
(
(* Untag, but don't shift the multiplicand. *)
genLeal (eax, r1, ~1, cvec);
(* Shift down the multiplier. *)
genOpRegPlus2 (Group2_1_A, edx, 7 (* sar *), cvec)
)
else (* r2 <> edx *)
(
(* Shift down the multiplier. *)
if r1 regNeq edx
then genMove(edx, r1, cvec) else ();
genOpRegPlus2 (Group2_1_A, edx, 7 (* sar *), cvec);
(* Untag, but don't shift the multiplicand. *)
genLeal (eax, r2, ~1, cvec)
);
(* Do the multiplication. *)
genOpRegPlus2 (Group3_A, edx, 5 (* imull *), cvec);
(* Add back the tag, but don't shift. *)
genLeal (rd, eax, 1, cvec);
(* Restore the saved registers. N.B. This also has
the effect of making sure that both eax and edx contain
valid values. *)
if rd regNeq edx
then genPushPop(POP_R, edx, cvec)
else ();
if rd regNeq eax
then genPushPop(POP_R, eax, cvec)
else ();
genJO8 (addr, cvec) (* Check for overflow. *)
end
| instrAddW =>
(* Fixed precision addition. (Doesn't test for overflow.) *)
(
(* Remove the tag from one argument, then add in the other. *)
(* This could be done using a single leal instruction:
leal rd,[r1+r2-1] *)
genLeal (rd, r2, ~1, cvec);
genReg (ADD, rd, r1, cvec)
)
| instrSubW =>
(* Fixed precision subtraction. (Doesn't test for overflow.) *)
(
genMove (rd, r1, cvec);
genReg (SUB, rd, r2, cvec);
genImmed (ADD, rd, 1, cvec)
)
| instrRevSubW =>
(* Fixed precision subtraction. (Doesn't test for overflow.) *)
(
genMove (rd, r2, cvec);
genReg (SUB, rd, r1, cvec);
genImmed (ADD, rd, 1, cvec)
)
| instrMulW =>
(* Fixed precision multiplication. (Doesn't test for overflow.) *)
(
(* This is a bit complicated because the result is always placed
in the EDX:EAX register pair so we have to save one or both. *)
if rd regNeq eax
then genPush(eax, cvec) else ();
if rd regNeq edx
then genPush(edx, cvec) else ();
if r2 regEq edx
then
(
(* Untag, but don't shift the multiplicand. *)
genLeal (eax, r1, ~1, cvec);
(* Shift down the multiplier. *)
genOpRegPlus2 (Group2_1_A, edx, 5 (* shr *), cvec)
)
else
(
(* Shift down the multiplier. *)
if r1 regNeq edx
then genMove(edx, r1, cvec) else ();
genOpRegPlus2 (Group2_1_A, edx, 5 (* shr *), cvec);
(* Untag, but don't shift the multiplicand. *)
genLeal (eax, r2, ~1, cvec)
);
(* Do the multiplication. *)
genOpRegPlus2 (Group3_A, edx, 4 (* mull *), cvec);
(* Add back the tag, but don't shift. *)
genLeal (rd, eax, 1, cvec);
(* Restore the saved registers. N.B. This also has
the effect of making sure that both eax and edx contain
valid values. *)
if rd regNeq edx
then genPushPop(POP_R, edx, cvec)
else ();
if rd regNeq eax
then genPushPop(POP_R, eax, cvec)
else ()
)
| instrDivW =>
(* Fixed precision division. (Doesn't test for overflow.) *)
divmodWord(true, r1, r2, rd, cvec)
| instrModW =>
(* Fixed precision remainder. (Doesn't test for overflow.) *)
divmodWord(false, r1, r2, rd, cvec)
| instrOrW =>
(* Logical or. *)
(
genMove (rd, r1, cvec);
genReg (OR, rd, r2, cvec)
)
| instrAndW =>
(* Logical and. *)
(
genMove (rd, r1, cvec);
genReg (AND, rd, r2, cvec)
)
| instrXorW =>
(
(* Must remove the tag from one argument. *)
genLeal (rd, r2, ~1, cvec);
genReg (XOR, rd, r1, cvec)
)
| instrLoad =>
(* Load a word. *)
(* The index is already multiplied by 2, so we need only multiply
by two again to give a word offset. Then we have to subtract
2 to account for the tag. *)
genOpIndexed(MOVL_A_R, ~(wordSize div 2), r1, r2, if wordSize = 4 then Scale2 else Scale4, rd, cvec)
| instrLoadB =>
(* Load a byte. *)
(
(* mov r2,rd; shrl $1,rd; movzl 0(r1,rd,1),rd; leal 1(,rd,2),rd *)
genMove (rd, r2, cvec);
genOpRegPlus2 (Group2_1_A, rd, 5 (* shr *), cvec);
genOpIndexed (MOVZX, 0, r1, rd, Scale1, rd, cvec);
(* Tag the result. *)
genTag (rd, cvec)
)
| instrSetStringLength => (* Set the length word of a string. *)
(
(* The length is untagged. *)
genOpRegPlus2 (Group2_1_A, r2, 5 (* shr *), cvec);
genOpEA (MOVL_R_A, 0, r1, r2, cvec);
(* Restore the original value. This ensures we don't have a
bad value around and also restores the original value
since it may still be wanted. *)
genTag(r2, cvec)
)
| _ =>
(* bad and unimplemented instrs *)
raise InternalError "Unimplemented instruction"
); (* end genRR *)
(* Register/immediate operations. In many of these operations we have to tag
the immediate value. *)
fun genRI (inst, rs, constnt, rd, cvec) : unit =
let
(* log2 function for special cases of powers of 2. *)
fun log2 n =
let
fun l2 i j =
if i < n then l2 (i*2) (j+1) else if i = n then j
else raise InternalError "Not a power of two"
in
if n <= 0 then raise InternalError "Not a power of two"
else l2 1 0
end
in
if rd regEq rs andalso (case inst of instrPush => false | _ => true)
then raise InternalError "Registers must be different"
else ();
case inst of
instrMove =>
if isShort constnt
then
(* Load a constant into a register. rs is ignored. *)
let
val c = toInt (toShort constnt);
val tagged = tag c;
in
genMoveI (rd, tagged, cvec)
end
else
let
val (rc, rx) = getReg rd
in
genop (MOVL_32_64_R rc, SOME {w=true, r=false, b=rx, x=false}, cvec);
addConstToVec (WVal constnt, InlineAbsolute, cvec) (* Remember this constant and address. *)
end
| instrPush =>
if isTagged32bitS constnt
then
(* Both rd and rs are ignored. *)
let
val c = toInt (toShort constnt);
val tagged = tag c;
in
if not (is8Bit tagged)
then
(
genop (PUSH_32, NONE, cvec);
gen32s(tagged, cvec)
)
else
(
genop (PUSH_8, NONE, cvec);
gen8s (tagged, cvec)
)
end
else if isX64
then (* Put it in the constant area. *)
(
genop (Group5, NONE, cvec);
genmodrm(Based, 6 (* push *), 5 (* PC rel *), cvec);
addConstToVec (WVal constnt, ConstArea 0, cvec)
)
else (* 32-bit *)
(
genop (PUSH_32, NONE, cvec);
addConstToVec (WVal constnt, InlineAbsolute, cvec)
)
| instrAddA =>
(* Arbitrary precision addition. *)
let
val c = toInt (toShort constnt);
val addr = tagTest1 (rs, cvec);
in
genMove (rd, rs, cvec);
genImmed (ADD, rd, semitag c, cvec);
genJO8 (addr, cvec)
end
| instrSubA =>
(* Arbitrary precision subtraction. *)
let
val c = toInt (toShort constnt);
val addr = tagTest1 (rs, cvec);
in
genMove (rd, rs, cvec);
genImmed (SUB, rd, semitag c, cvec);
genJO8 (addr, cvec)
end
| instrAddW =>
(* Fixed precision addition - doesn't check for overflow. *)
(* The argument is shifted but not tagged *)
let
val c = toInt (toShort constnt)
in
genLeal (rd, rs, semitag c, cvec)
end
| instrSubW =>
(* Fixed precision subtraction - doesn't check for overflow. *)
(* The argument is shifted but not tagged. *)
let
val c = toInt (toShort constnt)
in
genLeal (rd, rs, ~ (semitag c), cvec)
end
(* Now removed. This is no longer safe now that we look for constants
in the code.
| instrRevSubW =>
(* Fixed precision reverse subtraction - doesn't check for overflow. *)
(
genMoveI (rd, semitag c + 2, cvec);
genReg (SUB, rd, rs, cvec)
)
*)
| instrOrW =>
(* Logical or. *)
let
val c = toInt (toShort constnt);
val tagged = tag c;
in
genMove (rd, rs, cvec);
genImmed (OR, rd, tagged, cvec)
end
| instrAndW =>
(* Logical and. *)
let
val c = toInt (toShort constnt);
val tagged = tag c;
in
genMove (rd, rs, cvec);
genImmed (AND, rd, tagged, cvec)
end
| instrXorW =>
(* Constant must be shifted but not tagged. *)
let
val c = toInt (toShort constnt)
in
genMove (rd, rs, cvec);
genImmed (XOR, rd, semitag c, cvec)
end
| instrUpshiftW =>
(* Word shift of more than 63 (unsigned) is defined to return zero
for the logical shifts and either 0 or all ones for the arithmetic
shift. The i386 shift instructions mask the shift value instead. *)
let
val c = toInt (toShort constnt)
in
if c < 0 orelse c > 63
then genMoveI (rd, tag 0, cvec)
else
(
genLeal (rd, rs, ~1, cvec); (* Remove the tag. *)
genOpRegPlus2 (Group2_8_A, rd, 4 (* shl *), cvec);
gen8s (c, cvec);
genImmed (OR, rd, tag 0, cvec) (* Put in the tag *)
)
end
| instrDownshiftW =>
let
val c = toInt (toShort constnt)
in
if c < 0 orelse c > 63
then genMoveI (rd, tag 0, cvec)
else
(
genMove (rd, rs, cvec);
genOpRegPlus2 (Group2_8_A, rd, 5 (* shr *), cvec);
gen8s (c, cvec);
genImmed (OR, rd, tag 0, cvec) (* Put in the tag *)
)
end
| instrDownshiftArithW =>
(* In this case it's easiest to set the shift to 63. *)
let
val c = toInt (toShort constnt)
in
genMove (rd, rs, cvec);
genOpRegPlus2 (Group2_8_A, rd, 7 (* sar *), cvec);
gen8s (if c < 0 orelse c > 63 then 63 else c, cvec);
genImmed (OR, rd, tag 0, cvec) (* Put in the tag *)
end
| instrMulA =>
(* We only handle multiplication by two at the moment. We
could handle a wider range but it's not that easy particularly
because the overflow flag is not defined on shifts of more than
one. *)
let
val c = toInt (toShort constnt)
val addr = tagTest1 (rs, cvec);
in
if c = 2 then () else raise InternalError "Multiply not implemented";
(* Do the actual operation after removing a tag from one arg. *)
genLeal (rd, rs, ~1, cvec);
genReg (ADD, rd, rs, cvec);
genJO8 (addr, cvec)
end
| instrMulW =>
let
val c = toInt (toShort constnt)
(* We only handle multiplication by powers of two at the moment.
This is easier than for arbitrary precision multiplication
because we don't have to detect overflow. *)
val log2c = log2 c
in
if log2c = 0 (* Multiplying by one??? *)
then genMove (rd, rs, cvec)
else if log2c = 1
then (* Multiplying by 2. *)
genOpIndexed (LEAL, ~1, rs, rs, Scale1, rd, cvec)
else if log2c = 2
then (* Multiplying by 4 *)
genOpIndexed(LEAL, ~3, regNone, rs, Scale4, rd, cvec)
else if log2c = 3
then (* Multiplying by 8 *)
genOpIndexed(LEAL, ~7, regNone, rs, Scale8, rd, cvec)
else (* Other powers of 2. *)
(
genMove (rd, rs, cvec);
genOpRegPlus2 (Group2_8_A, rd, 4 (* shl *), cvec);
gen8s (log2c, cvec);
genLeal (rd, rd, 1-c, cvec) (* Remove the shifted tag and add the tag. *)
)
end
| instrDivW =>
let
val c = toInt (toShort constnt)
(* We only handle division by powers of two at the moment. *)
val log2c = log2 c
in
if log2c = 0 (* Dividing by one??? *)
then genMove (rd, rs, cvec)
else (* Other powers of 2. *)
(
genMove (rd, rs, cvec);
genOpRegPlus2 (Group2_8_A, rd, 5 (* shr *), cvec);
gen8s (log2c, cvec);
(* Set the tag bit, which may already be set as
a result of shifting a bit into it. *)
genImmed (OR, rd, tag 0, cvec)
)
end
| instrModW =>
let
val c = toInt (toShort constnt)
val ASSERT = log2 c (* Check it's a power of 2. *)
val tagged = tag (c-1)
in
genMove (rd, rs, cvec);
genImmed (AND, rd, tagged, cvec)
end
| instrLoad =>
let
val c = toInt (toShort constnt)
in
(* Offset is words so multiply by word size to get byte offset. *)
genLoad (c * wordSize, rs, rd, cvec)
end
| instrLoadB =>
(* Load a byte. *)
let
val c = toInt (toShort constnt)
in
genOpEA(MOVZX (* 2 byte opcode *), c, rs, rd, cvec);
(* Tag the result. *)
genTag (rd, cvec)
end
| instrVeclen =>
(
genLoad (~wordSize, rs, rd, cvec);
(* length only occupies the least significant 56 bits
- the other bits are flags. We can't AND it with exp2_56-1
because that's too big so we shift it up and down again. *)
genOpRegPlus2 (Group2_8_A, rd, 4 (* shl *),cvec);
gen8s(8, cvec);
genOpRegPlus2 (Group2_8_A, rd, 5 (* shr *), cvec);
gen8s(8, cvec);
(* Tag the result. *)
genTag (rd, cvec)
)
| instrVecflags =>
(* Load the flags byte. *)
(
genOpEA (MOVZX (* 2 byte opcode *), ~1, rs, rd, cvec);
(* Tag the result. *)
genTag (rd, cvec)
)
| instrGetFirstLong =>
let
(* Get the first word of a long integer. We've already
checked that it is long. *)
(* Test the "sign bit" of the object. *)
val _ =
(
genOpPlus2(Group3_a, ~1, rs, 0 (* test *), cvec);
gen8u (16, cvec)
)
(* Load the unsigned, untagged, little-endian value. *)
val _ = genLoad (0, rs, rd, cvec);
(* Skip if the sign bit wasn't set. *)
val l1 = [putConditional (JE, cvec)]
in
genOpRegPlus2(Group3_A, rd, 3 (* neg *), cvec);
fixup(l1, cvec);
genTag (rd, cvec)
end
| instrStringLength =>
let
(* If it's tagged the result is 1 otherwise we need to load
the length word and tag it. *)
val l1 = compareAndBranchRI (rs, toMachineWord 0 (* Unused *), Long, cvec)
val _ = genMoveI (rd, tag 1, cvec);
val l2 = unconditionalBranch cvec
in
fixup(l1, cvec);
genLoad (0, rs, rd, cvec); (* Load the length word. *)
genTag (rd, cvec); (* And tag the result. *)
fixup(l2, cvec)
end
| _ =>
(* bad *)
raise InternalError "Unimplemented instruction"
end; (* genRI *)
type cases = int * addrs;
(* On this architecture, the jumptable is physically inserted into
the code as a vector of address offsets. The function "indexedCase"
generates the space for the table and "makeJumpTable" inserts
the actual entries, once the addresses are known.
SPF 23/11/1997
Now changed to use a vector of jump instructions. These are padded
out to 8 bytes with no-ops. The reason for the change is to ensure
that the code segment only contains instructions so that we can scan
for addresses within the code. It also simplifies and speeds up
the indexed jump at the expense of doubling the size of the table
itself. DCJM 1/1/2001
*)
type jumpTableAddrs = addrs;
fun constrCases (p as (i,a)) = p;
type caseList = cases list;
fun useIndexedCase (min:int, max:int, numberOfCases:int, exhaustive:bool) =
isShort min andalso
isShort max andalso
numberOfCases > 7 andalso
numberOfCases >= (max - min) div 3;
fun indexedCase (r1:reg, r2:reg, min:int, max:int,
exhaustive:bool, cvec as Code{exited, ic, ...}) : jumpTableAddrs =
let
val rangeCheck =
if exhaustive then []
else let
val taggedMin = tag min;
val taggedMax = tag max;
in
(* Is it long? *)
genOpRegPlus2 (Group3_A, r1, 0(* test *), cvec);
gen32u(1, cvec);
(* Need to check whether the branch is in range. *)
let
val l1 = putConditional (JE, cvec);
(* Compare with the minimum. *)
val UUU = genImmed(CMP, r1, taggedMin, cvec);
val l2 = putConditional (JL, cvec);
(* Compare with the maximum. *)
val UUU = genImmed(CMP, r1, taggedMax, cvec);
val l3 = putConditional (JG, cvec);
in
[l1, l2, l3]
end
end;
val lab = ref addrZero;
val (rc2, rx2) = getReg r2
in
(* Load the address of the jump table. *)
genop (MOVL_32_64_R rc2, SOME {w=true, r=false, b=rx2, x=false}, cvec);
addConstToVec (HVal lab, InlineAbsolute, cvec);
(* Compute the jump address. The index is a tagged
integer so it is already multiplied by 2. We need to
multiply by four to get the correct size. We subtract off
the minimum value and also the shifted tag. *)
genOpIndexed (LEAL, min * ~8 - 4, r2, r1, Scale4, r2, cvec);
(* Jump into the jump table. Since each entry in the table
is 8 bytes long r2 will still be on a word + 2 byte
boundary. *)
genop (Group5, if rx2 then SOME{w=false, r=false, b=rx2, x=false} else NONE, cvec);
genmodrm(Register, 4 (* jmp *), rc2, cvec);
exited := true;
(* There's a very good chance that we will now extend the branches for
the "out of range" checks. The code to do that doesn't know
that all these branches will come to the same point so will generate three
separate long branches. We could combine them but it's hardly worth it. *)
doPending (cvec,
(max - min + 1) * 8 (* size of table. *) + 3 (* Maximum alignment *));
(* The start address must be on a two byte boundary so that the
address we've loaded is a valid code address. *)
while getAddr (!ic) mod 4 <> 2 do genop (NOP, NONE, cvec);
let
fun initialiseTable i =
if i > max then () (* Done *)
else
(
gen8u (opToInt JMP_32, cvec);
gen32u (0, cvec);
(* Add no-ops to make it 8 bytes. *)
gen8u (opToInt NOP, cvec);
gen8u (opToInt NOP, cvec);
gen8u (opToInt NOP, cvec);
initialiseTable (i+1)
)
val here = !ic;
in
lab := here;
initialiseTable min;
fixup (rangeCheck, cvec); (* The default case comes in here. *)
here
end
end;
fun makeJumpTable (startTab:jumpTableAddrs, cl:caseList, default:addrs,
min : int, max : int, Code{codeVec, ...}) : unit =
let
fun putCase i addr =
let
val addrOfJmp = startTab addrPlus ((i - min) * 8)
val jumpOffset = (addr addrMinus addrOfJmp) - 5 (* From end of instr. *)
in
set32s(jumpOffset, addrOfJmp addrPlus 1, codeVec)
end
(* Initialise to the default. *)
fun putInDefaults i =
if i <= max
then (putCase i default; putInDefaults(i+1))
else ()
(* Overwrite the defaults by the cases. N.B. We've generated
the list in reverse order so if we have any duplicates we
will correctly overwrite the later cases with earlier ones. *)
fun putInCases [] = ()
| putInCases ((i, a) :: t) = (putCase i a; putInCases t)
in
putInDefaults min;
putInCases cl
end;
fun printCode (Code{procName, numOfConsts, pcOffset, constVec, printStream, ...}) seg endcode =
let
val print = printStream
val ptr = ref 0;
(* prints a string representation of a number *)
fun printHex v = print(Int.fmt StringCvt.HEX v)
infix 3 +:= ;
fun (x +:= y) = (x := !x + (y:int));
fun print32 () =
let
val valu = get32s (!ptr, seg);
val U : unit = (ptr +:= 4);
in
if valu = tag 0 andalso !numOfConsts <> 0
then
(* May be a reference to a code-segment we haven't generated yet.
In that case we try to print the name of the function rather
than simply printing "1". It might be nice to print the
function name in other cases but that might be complicated. *)
let
val caddr = !ptr - 4
fun findRef [] = (* Not there - probably really tagged 0 *) printHex valu
| findRef ({const = CVal(Code{procName, ...}), addrs, ...} :: rest) =
if caddr = getAddr addrs + ! pcOffset*4
then print("=" ^ procName)
else findRef rest
| findRef (_ :: rest) = findRef rest
in
findRef(! constVec)
end
else printHex valu
end;
fun print64 () =
let
val valu = get64s(!ptr, seg);
in
printHex valu;
ptr +:= 8
end
fun get16s (a: int, seg: cseg) : int =
let
val b0 = Word8.toInt (csegGet (seg, a));
val b1 = Word8.toInt (csegGet (seg, a + 1));
val b1' = if b1 >= exp2_7 then b1 - exp2_8 else b1;
in
(b1' * exp2_8) + b0
end;
fun print16 () =
let
val valu = get16s (!ptr, seg);
val U : unit = (ptr +:= 2);
in
printHex valu
end;
fun print8 () =
let
val valu = get8s (!ptr, seg);
val U : unit = ptr +:= 1;
in
printHex valu
end;
fun printJmp () =
let
val valu = get8s (!ptr, seg);
val U : unit = ptr +:= 1;
in
printHex (valu + !ptr)
end;
(* Print an effective address. *)
fun printEA rex =
let
val modrm = Word8.toInt (csegGet (seg, !ptr));
val U : unit = (ptr +:= 1);
val md = modrm div 64;
val rm = modrm mod 8;
(* Decode the Rex prefix if present. *)
val rexW = IntInf.andb(rex, 0x8) <> 0
val rexR = IntInf.andb(rex, 0x4) <> 0
val rexX = IntInf.andb(rex, 0x2) <> 0
val rexB = IntInf.andb(rex, 0x1) <> 0
in
if md = 3
then print (regRepr (mkReg(rm, rexB)))
else if rm = 4
then let (* s-i-b present. *)
val sib = Word8.toInt (csegGet (seg, !ptr));
val U : unit = (ptr +:= 1);
val ss = sib div 64;
val index = (sib div 8) mod 8;
val base = sib mod 8;
in
if md = 1
then print8 ()
else if md = 2 orelse base = 5 (* andalso md=0 *)
then print32 ()
else ();
print "(";
if md <> 0 orelse base <> 5
then print (regRepr (mkReg(base, rexB)))
else ();
if index <> 4 (* No index. *)
then
print ("," ^ regRepr (mkReg(index, rexX)) ^
(if ss = 0 then ",1"
else if ss = 1 then ",2"
else if ss = 2 then ",4"
else ",8"))
else ();
print ")"
end
else (* no s-i-b. *) if md = 0 andalso rm = 5
then (* PC relative. *)
(print "(%rip+"; print32 (); print ")")
else (* register plus offset. *)
(
if md = 1
then print8 ()
else if md = 2
then print32 ()
else ();
print ("(" ^ regRepr (mkReg(rm, rexB)) ^ ")")
)
end;
fun printArith opc =
print
(case opc of
0 => "add"
| 1 => "or"
| 2 => "adc"
| 3 => "sbb"
| 4 => "and"
| 5 => "sub"
| 6 => "xor"
| _ => "cmp"
);
in
if procName = "" (* No name *) then print "?" else print procName;
print ":\n";
while !ptr < endcode do
let
val U : unit = printHex (!ptr); (* The address. *)
val U : unit = print "\t";
(* See if we have a REX byte. *)
val rex =
let
val b = get8u (!ptr, seg);
in
if b >= 0x40 andalso b <= 0x4f
then (ptr := !ptr + 1; b)
else 0
end
val rexW = IntInf.andb(rex, 0x8) <> 0
val rexR = IntInf.andb(rex, 0x4) <> 0
val rexX = IntInf.andb(rex, 0x2) <> 0
val rexB = IntInf.andb(rex, 0x1) <> 0
val opByte : int = get8u (!ptr, seg);
val U : unit = ptr +:= 1;
in
if opByte = opToInt Group1_8_A orelse
opByte = opToInt Group1_32_A
then let
(* Opcode is determined by next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
in
printArith ((nb div 8) mod 8);
print "_rev\t";
printEA rex; (* These are the wrong way round. *)
print ",";
if opByte = opToInt Group1_8_A
then print8 ()
else print32 ()
end : unit
else if opByte = opToInt JE
then (print "je\t"; printJmp()) : unit
else if opByte = opToInt JNE
then (print "jne\t"; printJmp()) : unit
else if opByte = opToInt JO
then (print "jo\t"; printJmp()) : unit
else if opByte = opToInt JL
then (print "jl\t"; printJmp()) : unit
else if opByte = opToInt JG
then (print "jg\t"; printJmp()) : unit
else if opByte = opToInt JLE
then (print "jle\t"; printJmp()) : unit
else if opByte = opToInt JGE
then (print "jge\t"; printJmp()) : unit
else if opByte = opToInt JB
then (print "jb\t"; printJmp()) : unit
else if opByte = opToInt JA
then (print "ja\t"; printJmp()) : unit
else if opByte = opToInt JNA
then (print "jna\t"; printJmp()) : unit
else if opByte = opToInt JNB
then (print "jnb\t"; printJmp()) : unit
else if opByte = opToInt JMP_8
then (print "jmp\t"; printJmp()) : unit
else if opByte = opToInt JMP_32
then let
val valu = get32s (!ptr, seg);
val U : unit = (ptr +:= 4);
in
print "jmp\t";
printHex (!ptr + valu)
end : unit
else if opByte = opToInt CALL_32
then let
val valu = get32s (!ptr, seg);
val U : unit = (ptr +:= 4);
in
print "call\t";
printHex (!ptr + valu)
end : unit
else if opByte = opToInt MOVL_A_R
then let
(* Register is in next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val reg = (nb div 8) mod 8;
in
print "movl\t";
printEA rex;
print ",";
print (regRepr (mkReg(reg, rexR)))
end : unit
else if opByte mod 8 = 3 andalso
opByte < 3 * 16 + 15 (* 0x3f *)
then let
(* Register is in next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val reg = (nb div 8) mod 8;
in
printArith ((opByte div 8) mod 8);
print "\t";
printEA rex;
print ",";
print (regRepr (mkReg(reg, rexR)))
end : unit
else if opByte = opToInt MOVL_R_A
then let
(* Register is in next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val reg = (nb div 8) mod 8;
in
print "movl\t";
print (regRepr (mkReg(reg, rexR)));
print ",";
printEA rex
end : unit
else if opByte = opToInt MOVB_R_A
then let
(* Register is in next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val reg = (nb div 8) mod 8;
in
print "movb\t";
if rexX
then print ("r" ^ Int.toString(reg+8) ^ "B")
else case reg of
0 => print "%al"
| 1 => print "%cl"
| 2 => print "%dl"
| 3 => print "%bl"
(* If there is a REX byte these select the low byte of the registers. *)
| 4 => print (if rex = 0 then "%ah" else "%sil")
| 5 => print (if rex = 0 then "%ch" else "%dil")
| 6 => print (if rex = 0 then "%dh" else "%bpl")
| 7 => print (if rex = 0 then "%bh" else "%spl")
| _ => print "Unknown register";
print ",";
printEA rex
end : unit
else if opByte >= opToInt (PUSH_R 0) andalso
opByte <= opToInt (PUSH_R 7)
then print ("pushl\t" ^ regRepr (mkReg (opByte mod 8, rexB))) : unit
else if opByte >= opToInt (POP_R 0) andalso
opByte <= opToInt (POP_R 7)
then print ("pop\t" ^ regRepr (mkReg (opByte mod 8, rexB))) : unit
else if opByte = opToInt NOP
then print "nop" : unit
else if opByte = opToInt LEAL
then let
(* Register is in next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val reg = (nb div 8) mod 8;
in
print "leal\t";
printEA rex;
print ",";
print (regRepr (mkReg(reg, rexR)))
end : unit
else if opByte >= opToInt (MOVL_32_64_R 0) andalso
opByte <= opToInt (MOVL_32_64_R 7)
then
(
print "movl\t";
if rexW then print64 () else print32 ();
print("," ^ regRepr (mkReg (opByte mod 8, rexB)))
) : unit
else if opByte = opToInt MOVL_32_A
then
(
print "movl_rev\t";
printEA rex; (* These are the wrong way round. *)
print ",";
print32 ()
) : unit
else if opByte = opToInt MOVB_8_A
then
(
print "movb_rev\t";
printEA rex; (* These are the wrong way round. *)
print ",";
print8 ()
) : unit
else if opByte = opToInt PUSH_32
then (print "push\t"; print32 ()) : unit
else if opByte = opToInt PUSH_8
then (print "push\t"; print8 ()) : unit
else if opByte = opToInt Group5
then let
(* Opcode is determined by next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val opc = (nb div 8) mod 8;
in
print
(case opc of
2 => "call"
| 4 => "jmp"
| 6 => "push"
| _ => "???"
);
print "\t";
printEA rex
end : unit
else if opByte = opToInt Group3_A
then let
(* Opcode is determined by next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val opc = (nb div 8) mod 8;
in
print
(case opc of
0 => "testl"
| 3 => "negl"
| 4 => "mull"
| 5 => "imull"
| 6 => "divl"
| 7 => "idivl"
| _ => "???"
);
print "\t";
printEA rex;
if opc = 0 then (print ","; print32 ()) else ()
end : unit
else if opByte = opToInt Group3_a
then let
(* Opcode is determined by next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val opc = (nb div 8) mod 8;
in
print
(case opc of
0 => "testb"
| 3 => "negb"
| _ => "???"
);
print "\t";
printEA rex;
if opc = 0 then (print ","; print8 ()) else ()
end : unit
else if opByte = opToInt Group2_8_A
then let
(* Opcode is determined by next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val opc = (nb div 8) mod 8;
in
print
(case opc of
4 => "shl"
| 5 => "shr"
| 7 => "sar"
| _ => "???"
);
print "\t";
printEA rex; (* These are the wrong way round. *)
print ",";
print8 ()
end : unit
else if opByte = opToInt Group2_1_A
then let
(* Opcode is determined by next byte. *)
val nb = Word8.toInt (csegGet (seg, !ptr));
val opc = (nb div 8) mod 8;
in
print
(case opc of
5 => "shr"
| 7 => "sar"
| _ => "???"
);
print "\t1,";
printEA rex
end : unit
else if opByte = opToInt ESCAPE
then let
(* Opcode is in next byte. *)
val opByte2 = Word8.toInt (csegGet (seg, !ptr));
val U : unit = (ptr +:= 1);
in
if opByte2 = 11 * 16 + 6 (* 0xb6 *)
then let
val nb = Word8.toInt (csegGet (seg, !ptr));
val reg = (nb div 8) mod 8;
in
print "movzl\t";
printEA rex;
print ",";
print (regRepr (mkReg(reg, rexR)))
end : unit
else if opByte2 >= 8 * 16 (* 0x80 *) andalso
opByte2 <= 8 * 16 + 15 (* 0x8f *)
then let
val valu = get32s (!ptr, seg);
val U : unit = (ptr +:= 4);
in
print
(if opByte2 = 8 * 16 (* 0x80 *)
then "jo\t"
else if opByte2 = 8 * 16 + 4 (* 0x84 *)
then "je\t"
else if opByte2 = 8 * 16 + 5 (* 0x85 *)
then "jne\t"
else if opByte2 = 8 * 16 + 12 (* 0x8c *)
then "jl\t"
else if opByte2 = 8 * 16 + 13 (* 0x8d *)
then "jge\t"
else if opByte2 = 8 * 16 + 14 (* 0x8e *)
then "jle\t"
else if opByte2 = 8 * 16 + 15 (* 0x8f *)
then "jg\t"
else if opByte2 = 8 * 16 + 2 (* 0x82 *)
then "jb\t"
else if opByte2 = 8 * 16 + 3 (* 0x83 *)
then "jnb\t"
else if opByte2 = 8 * 16 + 6 (* 0x86 *)
then "jna\t"
else if opByte2 = 8 * 16 + 7 (* 0x87 *)
then "ja\t"
else "???\t"
);
printHex (!ptr + valu)
end : unit
else (print "esc\t"; printHex opByte2) : unit
end (* ESCAPE *)
else if opByte = opToInt POP_A
then (print "pop\t"; printEA rex) : unit
else if opByte = opToInt RET
then print "ret" : unit
else if opByte = opToInt STC
then print "stc" : unit
else if opByte = opToInt RET_16
then (print "ret\t"; print16 ()) : unit
else if opByte = opToInt TEST_ACC8
then (print "testb\t%al,"; print8 ())
else printHex opByte : unit;
print "\n" : unit
end; (* end of while loop *)
print "\n"
end (* printCode *);
(* constLabels - fill in a constant in the code. *)
fun constLabels (Code{resultSeg=ref rseg, pcOffset=ref offset, ic = ref endByte, ...},
addr : addrs, value : machineWord, posn: ConstPosn) : unit =
let
val seg = scSet rseg; (* The address of the segment. *)
val constAddr = addr addrPlus offset*wordSize;
in
case posn of
InlineAbsolute =>
csegPutConstant (seg, getAddr constAddr, value, false)
| InlineRelative =>
csegPutConstant (seg, getAddr constAddr, value, true)
| ConstArea nonInlineCount =>
(* Not inline. Put the constant in the constant area and set the original address
to be the relative offset to the constant itself. *)
let
val addrOfConst = getAddr (endByte addrPlus (offset + nonInlineCount-1 + 2+1+1) * wordSize);
in
csegPutConstant (seg, addrOfConst, value, false);
set32s(addrOfConst - getAddr constAddr - 4, constAddr, seg)
end
end;
(* Fix up references from other vectors to this one. *)
fun fixOtherRefs (refTo as Code{otherCodes=ref otherCodes, ...}, value) =
let
fun fixRef (refFrom as
Code{numOfConsts = noc, constVec = ref constVec,
resultSeg = ref resultSeg, ...}) =
let
fun putConst {const = CVal cCode, addrs, posn, ...} =
if cCode is refTo
then (* A reference to this one. *)
(
(* Fix up the forward reference. *)
constLabels (refFrom, addrs, value, posn);
(* decrement the "pending references" count *)
noc := !noc - 1
)
else ()
| putConst _ = ();
in
(* look down its list of forward references until we find ourselves. *)
List.app putConst constVec;
(* If this function has no more references we can lock it. *)
if !noc = 0
then csegLock (scSet resultSeg)
else ()
end (* fixRef *);
in
(* For each `code' which needs a forward reference to `refTo' fixing up. *)
List.app fixRef otherCodes
end; (* fixOtherRefs *)
(***************************************************************************)
(* copyCode *)
(***************************************************************************)
(* The stack limit register is set at least twice this far from the
end of the stack so we can simply compare the stack pointer with
the stack limit register if we need less than this much. Setting
it at twice this value means that procedures which use up to this
much stack and do not call any other procedures do not need to
check the stack at all. *)
val minStackCheck = 20;
(* Adds the constants onto the code, and copies the code into a new segment *)
fun copyCode (cvec as
Code{pcOffset,
codeVec,
noClosure,
selfCalls = ref selfCalls,
selfJumps = ref selfJumps,
mustCheckStack = ref callsAProc,
numOfConsts,
nonInlineConsts = ref constsInConstArea,
ic,
constVec = ref constVec,
resultSeg,
procName,
printAssemblyCode,
printStream,
...},
stackRequired, registerSet) : address =
let
(* This aligns ic onto a fullword boundary. *)
val U : unit = while getAddr (!ic) mod wordSize <> 0 do genop(NOP, NONE, cvec);
val endic = !ic; (* Remember end *)
val U : unit = if wordSize = 8 then gen64u(0, cvec) else gen32u (0, cvec); (* Marker - 0 (changes !ic) *)
(* Prelude consists of
1) nops to make it a whole number of words
2) stack checking code
*)
local
(* little-endian *)
fun getBytes (0, x) = []
| getBytes (n, x) = (x mod exp2_8) :: getBytes (n - 1, x div exp2_8);
fun testRegAndTrap (reg, entryPt) =
[
rex{w=true,r=false,x=false,b=false},
(* cmp reg,16(%ebp)*)
opToInt(Arith (CMP, 3)),
modrm (Based8, #1 (getReg reg), #1 (getReg ebp)),
MemRegStackLimit,
(* jnb 3 *)
opToInt JNB,
3,
(* call *)
opToInt Group5,
modrm (Based8, 2 (* call *), #1 (getReg ebp)),
entryPt
];
val stackCheckCode : int list =
if stackRequired >= minStackCheck
then let
val stackByteAdjust = ~wordSize * stackRequired;
val loadEdiCode : int list =
if is8Bit stackByteAdjust
then
[
rex{w=true,r=false,x=false,b=false},
opToInt LEAL,
modrm (Based8, #1(getReg edi), 4), (* Need s-i-b byte for %esp *)
sib (Scale1, 4 (* no index *), #1(getReg esp))
]
@ getBytes (1, stackByteAdjust)
else
[
rex{w=true,r=false,x=false,b=false},
opToInt LEAL,
modrm (Based32, #1(getReg edi), 4), (* Need s-i-b byte for %esp *)
sib (Scale1, 4 (* no index *), #1(getReg esp))
]
@ getBytes (4, stackByteAdjust);
val testEdiCode : int list =
testRegAndTrap (edi, MemRegStackOverflowCallEx)
in
loadEdiCode @ testEdiCode
(* The effect of this sequence is to generate an
overflow trap if sp < sl *)
end
else if callsAProc (* check for user interrupt *)
then testRegAndTrap (esp, MemRegStackOverflowCall)
else (* no stack check required *)
[];
(*****************************************************************************
Functions now have up to 2 entry points:
(1) Standard entry point
(2) Self-call entry point - doesn't change %ecx
Entry point 1 is always the first word of the actual code.
Entry point 2 can be at various offsets (if it is needed at all),
but that's OK because it is only used for calls within the procedure
itself.
*****************************************************************************)
val nopCode : int list =
let
(* Add sufficient No-ops to round this to a full word. *)
val len = length stackCheckCode mod wordSize
in
if len = 0
then []
else List.tabulate(wordSize - len, fn _ => opToInt NOP)
end
in
val preludeCode = nopCode @ stackCheckCode;
val wordsForPrelude = length preludeCode div wordSize
(* +5 for code size, profile count, function name, register mask and constants count *)
val segSize = (getAddr (!ic)) div wordSize + constsInConstArea + wordsForPrelude + 5;
(* byte offset of L2 label relative to start of post-prelude code. *)
val L2Addr = mkAddr (~ (length stackCheckCode));
end; (* local *)
(* fix-up all the self-calls *)
val U : unit =
fixRecursiveCalls (cvec, L2Addr, selfCalls);
val U : unit =
fixRecursiveBranches (cvec, L2Addr, selfJumps);
(* Now make the byte segment that we'll turn into the code segment *)
val seg : cseg = csegMake segSize;
val offset = wordsForPrelude;
val _ = resultSeg := Set seg;
(* Copy the code into the new segment. *)
val _ = pcOffset := offset;
val _ = csegCopySeg (codeVec, seg, getAddr (! ic), offset);
(* insert prelude code into segment *)
local
val ptr = ref 0;
(* Generate the prelude. *)
fun putPrelude (b: int) : unit =
let
val a = !ptr
in
csegSet (seg, a, Word8.fromInt b);
ptr := a + 1
end;
fun putPreludeList [] = ()
| putPreludeList (w::ws) = (putPrelude w; putPreludeList ws);
in
val U : unit = putPreludeList preludeCode
end;
local
val endOfCode (* words *) = (getAddr (! ic)) div wordSize + offset;
in
(* Byte offset of start of code. *)
local
val byteEndofcode = endOfCode * wordSize;
val addr = mkAddr byteEndofcode;
in
val U : unit = setWordU (byteEndofcode, addr, seg)
end;
(* Put in the number of constants. This must go in before we actually put
in any constants. *)
local
val addr = mkAddr ((endOfCode + 3 + constsInConstArea + 1) * wordSize);
in
val U : unit = setWordU(2 + constsInConstArea, addr, seg)
end;
(* Next the profile count. *)
local
val addr = mkAddr ((endOfCode + 1) * wordSize);
in
val U : unit = setWordU (0, addr, seg)
end;
(* Now we've filled in all the C integers; now we need to convert the segment
into a proper code segment before it's safe to put in any ML values.
SPF 13/2/97
*)
val U : unit = csegConvertToCode seg;
local
(* why do we treat the empty string as a special case? SPF 15/7/94 *)
(* This is so that profiling can print "<anon>". Note that a
tagged zero *is* a legal string (it's "\000"). SPF 14/10/94 *)
val nameWord : machineWord = if procName = "" then toMachineWord 0 else toMachineWord procName;
in
val _ = csegPutWord (seg, endOfCode + 2, nameWord);
end;
local
(* Encode the register mask. This encoding must be the same
as the one used for assembly code segments. *)
fun encodeReg(r, n: short): short =
let
open Word
infix << orb
val reg = 0w1 << Word.fromInt (nReg r)
in
reg orb n
end
val regSet = List.foldl encodeReg 0w0 registerSet
in
val U : unit = csegPutWord (seg, endOfCode + 3, toMachineWord regSet);
end;
end; (* scope of endOfCode *)
in
let
(* and then copy the objects from the constant list. *)
fun putConst {const = WVal c, addrs, posn, ...} =
( (* Can put these in now. *)
constLabels (cvec, addrs, c, posn);
numOfConsts := ! numOfConsts - 1
)
| putConst {const = HVal(ref hv), addrs, posn, ...} =
let
(* on the PC, we don't add the extra 2 (we do on the Sparc) *)
(* SPF 24/4/95 *)
val handlerByteOffset = getAddr hv + offset * wordSize;
(* The following comment applies to offsetAddr *)
(* Special function to add to an address.
This only works if the resulting value is
in a code segment and is on a word + 2 byte boundary. *)
val handlerAddr : handler =
offsetAddr (csegAddr seg, toShort handlerByteOffset);
in
constLabels (cvec, addrs, toMachineWord handlerAddr, posn);
numOfConsts := ! numOfConsts - 1
end
(* forward-reference - fix up later when we compile
the referenced code *)
| putConst {const = CVal _, ...} = ()
val _ = List.app putConst constVec;
(* Switch off "mutable" bit now if we have no
forward or recursive references to fix-up *)
val _ = if ! numOfConsts = 0 then csegLock seg else ();
(* Do we need to make a closure, or just return the code? *)
val addr : address =
if noClosure
then csegAddr seg
else let
val addr : address = alloc (short1, F_words, toMachineWord (csegAddr seg));
(* Logically unnecessary; however the RTS currently allocates everything
as mutable because Dave's code assumed that things were done this
way and I'm not completely sure that everything that needs a mutable
allocation actually asks for it yet. SPF 19/2/97
*)
val U : unit = lock addr;
in
addr
end
(* Now we know the address of this object we can fix up
any forward references outstanding. This is put in here
because there may be directly recursive references. *)
val U : unit = fixOtherRefs (cvec, toMachineWord addr);
val U : unit =
if printAssemblyCode
then (* print out the code *)
(
printCode cvec seg ((getAddr endic) + offset * wordSize);
printStream "Register set = [";
List.app (fn r => (printStream " "; regPrint r)) registerSet;
printStream "]\n\n"
)
else ();
in
addr
end (* the result *)
end (* copyCode *);
(* ic function exported to gencode. Currently only used for backward jumps. *)
fun ic (cvec as Code {exited, ic=ic', branchCheck, ...}) =
( (* Make sure any pending operations are done. *)
doPending (cvec, 0);
exited := false; (* We may be jumping here. *)
branchCheck := !ic';
! ic' (* After any pending operations. *)
);
fun codeAddress (cvec: code) : address option =
(* This is used to find the register set for a function which was
originally a forward reference. If it has now been compiled we
can get the code. *)
case cvec of
Code {resultSeg = ref (Set cseg), ...} => SOME(csegAddr cseg)
| Code {resultSeg = ref Unset, ...} =>
(* We haven't compiled this yet: assume worst case. *) NONE
fun traceContext (Code {procName, ic = ref ic, ...}) =
(* Function name and code offset to help tracing. *)
procName ^ ":" ^ Int.fmt StringCvt.HEX (getAddr ic)
end (* struct *)
end (* CODECONS *);
|