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
|
{-
(c) The GRASP/AQUA Project, Glasgow University, 1992-1998
\section[ConFold]{Constant Folder}
Conceptually, constant folding should be parameterized with the kind
of target machine to get identical behaviour during compilation time
and runtime. We cheat a little bit here...
ToDo:
check boundaries before folding, e.g. we can fold the Float addition
(i1 + i2) only if it results in a valid Float.
-}
{-# LANGUAGE CPP, RankNTypes, PatternSynonyms, ViewPatterns, RecordWildCards,
DeriveFunctor, LambdaCase, TypeApplications, MultiWayIf #-}
{-# OPTIONS_GHC -optc-DNON_POSIX_SOURCE -Wno-incomplete-uni-patterns #-}
module GHC.Core.Opt.ConstantFold
( primOpRules
, builtinRules
, caseRules
, EnableBignumRules (..)
)
where
#include "GhclibHsVersions.h"
import GHC.Prelude
import {-# SOURCE #-} GHC.Types.Id.Make ( mkPrimOpId, magicDictId, voidPrimId )
import GHC.Core
import GHC.Core.Make
import GHC.Types.Id
import GHC.Types.Literal
import GHC.Core.SimpleOpt ( exprIsLiteral_maybe )
import GHC.Builtin.PrimOps ( PrimOp(..), tagToEnumKey )
import GHC.Builtin.Types
import GHC.Builtin.Types.Prim
import GHC.Core.TyCon
( tyConDataCons_maybe, isAlgTyCon, isEnumerationTyCon
, isNewTyCon, unwrapNewTyCon_maybe, tyConDataCons
, tyConFamilySize )
import GHC.Core.DataCon ( dataConTagZ, dataConTyCon, dataConWrapId, dataConWorkId )
import GHC.Core.Utils ( eqExpr, cheapEqExpr, exprIsHNF, exprType
, stripTicksTop, stripTicksTopT, mkTicks, stripTicksE )
import GHC.Core.Unfold ( exprIsConApp_maybe )
import GHC.Core.Multiplicity
import GHC.Core.FVs
import GHC.Core.Type
import GHC.Types.Var.Set
import GHC.Types.Var.Env
import GHC.Types.Name.Occurrence ( occNameFS )
import GHC.Builtin.Names
import GHC.Data.Maybe ( orElse )
import GHC.Types.Name ( Name, nameOccName )
import GHC.Utils.Outputable
import GHC.Data.FastString
import GHC.Types.Basic
import GHC.Platform
import GHC.Utils.Misc
import GHC.Core.Coercion (mkUnbranchedAxInstCo,mkSymCo,Role(..))
import Control.Applicative ( Alternative(..) )
import Control.Monad
import Data.Bits as Bits
import qualified Data.ByteString as BS
import Data.Int
import Data.Ratio
import Data.Word
import Data.Maybe (fromMaybe)
{-
Note [Constant folding]
~~~~~~~~~~~~~~~~~~~~~~~
primOpRules generates a rewrite rule for each primop
These rules do what is often called "constant folding"
E.g. the rules for +# might say
4 +# 5 = 9
Well, of course you'd need a lot of rules if you did it
like that, so we use a BuiltinRule instead, so that we
can match in any two literal values. So the rule is really
more like
(Lit x) +# (Lit y) = Lit (x+#y)
where the (+#) on the rhs is done at compile time
That is why these rules are built in here.
-}
primOpRules :: Name -> PrimOp -> Maybe CoreRule
primOpRules nm = \case
TagToEnumOp -> mkPrimOpRule nm 2 [ tagToEnumRule ]
DataToTagOp -> mkPrimOpRule nm 2 [ dataToTagRule ]
-- Int operations
IntAddOp -> mkPrimOpRule nm 2 [ binaryLit (intOp2 (+))
, identityPlatform zeroi
, numFoldingRules IntAddOp intPrimOps
]
IntSubOp -> mkPrimOpRule nm 2 [ binaryLit (intOp2 (-))
, rightIdentityPlatform zeroi
, equalArgs >> retLit zeroi
, numFoldingRules IntSubOp intPrimOps
]
IntAddCOp -> mkPrimOpRule nm 2 [ binaryLit (intOpC2 (+))
, identityCPlatform zeroi ]
IntSubCOp -> mkPrimOpRule nm 2 [ binaryLit (intOpC2 (-))
, rightIdentityCPlatform zeroi
, equalArgs >> retLitNoC zeroi ]
IntMulOp -> mkPrimOpRule nm 2 [ binaryLit (intOp2 (*))
, zeroElem zeroi
, identityPlatform onei
, numFoldingRules IntMulOp intPrimOps
]
IntQuotOp -> mkPrimOpRule nm 2 [ nonZeroLit 1 >> binaryLit (intOp2 quot)
, leftZero zeroi
, rightIdentityPlatform onei
, equalArgs >> retLit onei ]
IntRemOp -> mkPrimOpRule nm 2 [ nonZeroLit 1 >> binaryLit (intOp2 rem)
, leftZero zeroi
, do l <- getLiteral 1
platform <- getPlatform
guard (l == onei platform)
retLit zeroi
, equalArgs >> retLit zeroi
, equalArgs >> retLit zeroi ]
AndIOp -> mkPrimOpRule nm 2 [ binaryLit (intOp2 (.&.))
, idempotent
, zeroElem zeroi ]
OrIOp -> mkPrimOpRule nm 2 [ binaryLit (intOp2 (.|.))
, idempotent
, identityPlatform zeroi ]
XorIOp -> mkPrimOpRule nm 2 [ binaryLit (intOp2 xor)
, identityPlatform zeroi
, equalArgs >> retLit zeroi ]
NotIOp -> mkPrimOpRule nm 1 [ unaryLit complementOp
, inversePrimOp NotIOp ]
IntNegOp -> mkPrimOpRule nm 1 [ unaryLit negOp
, inversePrimOp IntNegOp ]
ISllOp -> mkPrimOpRule nm 2 [ shiftRule LitNumInt (const Bits.shiftL)
, rightIdentityPlatform zeroi ]
ISraOp -> mkPrimOpRule nm 2 [ shiftRule LitNumInt (const Bits.shiftR)
, rightIdentityPlatform zeroi ]
ISrlOp -> mkPrimOpRule nm 2 [ shiftRule LitNumInt shiftRightLogical
, rightIdentityPlatform zeroi ]
-- Word operations
WordAddOp -> mkPrimOpRule nm 2 [ binaryLit (wordOp2 (+))
, identityPlatform zerow
, numFoldingRules WordAddOp wordPrimOps
]
WordSubOp -> mkPrimOpRule nm 2 [ binaryLit (wordOp2 (-))
, rightIdentityPlatform zerow
, equalArgs >> retLit zerow
, numFoldingRules WordSubOp wordPrimOps
]
WordAddCOp -> mkPrimOpRule nm 2 [ binaryLit (wordOpC2 (+))
, identityCPlatform zerow ]
WordSubCOp -> mkPrimOpRule nm 2 [ binaryLit (wordOpC2 (-))
, rightIdentityCPlatform zerow
, equalArgs >> retLitNoC zerow ]
WordMulOp -> mkPrimOpRule nm 2 [ binaryLit (wordOp2 (*))
, identityPlatform onew
, numFoldingRules WordMulOp wordPrimOps
]
WordQuotOp -> mkPrimOpRule nm 2 [ nonZeroLit 1 >> binaryLit (wordOp2 quot)
, rightIdentityPlatform onew ]
WordRemOp -> mkPrimOpRule nm 2 [ nonZeroLit 1 >> binaryLit (wordOp2 rem)
, leftZero zerow
, do l <- getLiteral 1
platform <- getPlatform
guard (l == onew platform)
retLit zerow
, equalArgs >> retLit zerow ]
AndOp -> mkPrimOpRule nm 2 [ binaryLit (wordOp2 (.&.))
, idempotent
, zeroElem zerow ]
OrOp -> mkPrimOpRule nm 2 [ binaryLit (wordOp2 (.|.))
, idempotent
, identityPlatform zerow ]
XorOp -> mkPrimOpRule nm 2 [ binaryLit (wordOp2 xor)
, identityPlatform zerow
, equalArgs >> retLit zerow ]
NotOp -> mkPrimOpRule nm 1 [ unaryLit complementOp
, inversePrimOp NotOp ]
SllOp -> mkPrimOpRule nm 2 [ shiftRule LitNumWord (const Bits.shiftL) ]
SrlOp -> mkPrimOpRule nm 2 [ shiftRule LitNumWord shiftRightLogical ]
-- coercions
Word2IntOp -> mkPrimOpRule nm 1 [ liftLitPlatform word2IntLit
, inversePrimOp Int2WordOp ]
Int2WordOp -> mkPrimOpRule nm 1 [ liftLitPlatform int2WordLit
, inversePrimOp Word2IntOp ]
Narrow8IntOp -> mkPrimOpRule nm 1 [ liftLit narrow8IntLit
, subsumedByPrimOp Narrow8IntOp
, Narrow8IntOp `subsumesPrimOp` Narrow16IntOp
, Narrow8IntOp `subsumesPrimOp` Narrow32IntOp
, narrowSubsumesAnd AndIOp Narrow8IntOp 8 ]
Narrow16IntOp -> mkPrimOpRule nm 1 [ liftLit narrow16IntLit
, subsumedByPrimOp Narrow8IntOp
, subsumedByPrimOp Narrow16IntOp
, Narrow16IntOp `subsumesPrimOp` Narrow32IntOp
, narrowSubsumesAnd AndIOp Narrow16IntOp 16 ]
Narrow32IntOp -> mkPrimOpRule nm 1 [ liftLit narrow32IntLit
, subsumedByPrimOp Narrow8IntOp
, subsumedByPrimOp Narrow16IntOp
, subsumedByPrimOp Narrow32IntOp
, removeOp32
, narrowSubsumesAnd AndIOp Narrow32IntOp 32 ]
Narrow8WordOp -> mkPrimOpRule nm 1 [ liftLit narrow8WordLit
, subsumedByPrimOp Narrow8WordOp
, Narrow8WordOp `subsumesPrimOp` Narrow16WordOp
, Narrow8WordOp `subsumesPrimOp` Narrow32WordOp
, narrowSubsumesAnd AndOp Narrow8WordOp 8 ]
Narrow16WordOp -> mkPrimOpRule nm 1 [ liftLit narrow16WordLit
, subsumedByPrimOp Narrow8WordOp
, subsumedByPrimOp Narrow16WordOp
, Narrow16WordOp `subsumesPrimOp` Narrow32WordOp
, narrowSubsumesAnd AndOp Narrow16WordOp 16 ]
Narrow32WordOp -> mkPrimOpRule nm 1 [ liftLit narrow32WordLit
, subsumedByPrimOp Narrow8WordOp
, subsumedByPrimOp Narrow16WordOp
, subsumedByPrimOp Narrow32WordOp
, removeOp32
, narrowSubsumesAnd AndOp Narrow32WordOp 32 ]
OrdOp -> mkPrimOpRule nm 1 [ liftLit char2IntLit
, inversePrimOp ChrOp ]
ChrOp -> mkPrimOpRule nm 1 [ do [Lit lit] <- getArgs
guard (litFitsInChar lit)
liftLit int2CharLit
, inversePrimOp OrdOp ]
Float2IntOp -> mkPrimOpRule nm 1 [ liftLit float2IntLit ]
Int2FloatOp -> mkPrimOpRule nm 1 [ liftLit int2FloatLit ]
Double2IntOp -> mkPrimOpRule nm 1 [ liftLit double2IntLit ]
Int2DoubleOp -> mkPrimOpRule nm 1 [ liftLit int2DoubleLit ]
-- SUP: Not sure what the standard says about precision in the following 2 cases
Float2DoubleOp -> mkPrimOpRule nm 1 [ liftLit float2DoubleLit ]
Double2FloatOp -> mkPrimOpRule nm 1 [ liftLit double2FloatLit ]
-- Float
FloatAddOp -> mkPrimOpRule nm 2 [ binaryLit (floatOp2 (+))
, identity zerof ]
FloatSubOp -> mkPrimOpRule nm 2 [ binaryLit (floatOp2 (-))
, rightIdentity zerof ]
FloatMulOp -> mkPrimOpRule nm 2 [ binaryLit (floatOp2 (*))
, identity onef
, strengthReduction twof FloatAddOp ]
-- zeroElem zerof doesn't hold because of NaN
FloatDivOp -> mkPrimOpRule nm 2 [ guardFloatDiv >> binaryLit (floatOp2 (/))
, rightIdentity onef ]
FloatNegOp -> mkPrimOpRule nm 1 [ unaryLit negOp
, inversePrimOp FloatNegOp ]
FloatDecode_IntOp -> mkPrimOpRule nm 1 [ unaryLit floatDecodeOp ]
-- Double
DoubleAddOp -> mkPrimOpRule nm 2 [ binaryLit (doubleOp2 (+))
, identity zerod ]
DoubleSubOp -> mkPrimOpRule nm 2 [ binaryLit (doubleOp2 (-))
, rightIdentity zerod ]
DoubleMulOp -> mkPrimOpRule nm 2 [ binaryLit (doubleOp2 (*))
, identity oned
, strengthReduction twod DoubleAddOp ]
-- zeroElem zerod doesn't hold because of NaN
DoubleDivOp -> mkPrimOpRule nm 2 [ guardDoubleDiv >> binaryLit (doubleOp2 (/))
, rightIdentity oned ]
DoubleNegOp -> mkPrimOpRule nm 1 [ unaryLit negOp
, inversePrimOp DoubleNegOp ]
DoubleDecode_Int64Op -> mkPrimOpRule nm 1 [ unaryLit doubleDecodeOp ]
-- Relational operators
IntEqOp -> mkRelOpRule nm (==) [ litEq True ]
IntNeOp -> mkRelOpRule nm (/=) [ litEq False ]
CharEqOp -> mkRelOpRule nm (==) [ litEq True ]
CharNeOp -> mkRelOpRule nm (/=) [ litEq False ]
IntGtOp -> mkRelOpRule nm (>) [ boundsCmp Gt ]
IntGeOp -> mkRelOpRule nm (>=) [ boundsCmp Ge ]
IntLeOp -> mkRelOpRule nm (<=) [ boundsCmp Le ]
IntLtOp -> mkRelOpRule nm (<) [ boundsCmp Lt ]
CharGtOp -> mkRelOpRule nm (>) [ boundsCmp Gt ]
CharGeOp -> mkRelOpRule nm (>=) [ boundsCmp Ge ]
CharLeOp -> mkRelOpRule nm (<=) [ boundsCmp Le ]
CharLtOp -> mkRelOpRule nm (<) [ boundsCmp Lt ]
FloatGtOp -> mkFloatingRelOpRule nm (>)
FloatGeOp -> mkFloatingRelOpRule nm (>=)
FloatLeOp -> mkFloatingRelOpRule nm (<=)
FloatLtOp -> mkFloatingRelOpRule nm (<)
FloatEqOp -> mkFloatingRelOpRule nm (==)
FloatNeOp -> mkFloatingRelOpRule nm (/=)
DoubleGtOp -> mkFloatingRelOpRule nm (>)
DoubleGeOp -> mkFloatingRelOpRule nm (>=)
DoubleLeOp -> mkFloatingRelOpRule nm (<=)
DoubleLtOp -> mkFloatingRelOpRule nm (<)
DoubleEqOp -> mkFloatingRelOpRule nm (==)
DoubleNeOp -> mkFloatingRelOpRule nm (/=)
WordGtOp -> mkRelOpRule nm (>) [ boundsCmp Gt ]
WordGeOp -> mkRelOpRule nm (>=) [ boundsCmp Ge ]
WordLeOp -> mkRelOpRule nm (<=) [ boundsCmp Le ]
WordLtOp -> mkRelOpRule nm (<) [ boundsCmp Lt ]
WordEqOp -> mkRelOpRule nm (==) [ litEq True ]
WordNeOp -> mkRelOpRule nm (/=) [ litEq False ]
AddrAddOp -> mkPrimOpRule nm 2 [ rightIdentityPlatform zeroi ]
SeqOp -> mkPrimOpRule nm 4 [ seqRule ]
SparkOp -> mkPrimOpRule nm 4 [ sparkRule ]
_ -> Nothing
{-
************************************************************************
* *
\subsection{Doing the business}
* *
************************************************************************
-}
-- useful shorthands
mkPrimOpRule :: Name -> Int -> [RuleM CoreExpr] -> Maybe CoreRule
mkPrimOpRule nm arity rules = Just $ mkBasicRule nm arity (msum rules)
mkRelOpRule :: Name -> (forall a . Ord a => a -> a -> Bool)
-> [RuleM CoreExpr] -> Maybe CoreRule
mkRelOpRule nm cmp extra
= mkPrimOpRule nm 2 $
binaryCmpLit cmp : equal_rule : extra
where
-- x `cmp` x does not depend on x, so
-- compute it for the arbitrary value 'True'
-- and use that result
equal_rule = do { equalArgs
; platform <- getPlatform
; return (if cmp True True
then trueValInt platform
else falseValInt platform) }
{- Note [Rules for floating-point comparisons]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We need different rules for floating-point values because for floats
it is not true that x = x (for NaNs); so we do not want the equal_rule
rule that mkRelOpRule uses.
Note also that, in the case of equality/inequality, we do /not/
want to switch to a case-expression. For example, we do not want
to convert
case (eqFloat# x 3.8#) of
True -> this
False -> that
to
case x of
3.8#::Float# -> this
_ -> that
See #9238. Reason: comparing floating-point values for equality
delicate, and we don't want to implement that delicacy in the code for
case expressions. So we make it an invariant of Core that a case
expression never scrutinises a Float# or Double#.
This transformation is what the litEq rule does;
see Note [The litEq rule: converting equality to case].
So we /refrain/ from using litEq for mkFloatingRelOpRule.
-}
mkFloatingRelOpRule :: Name -> (forall a . Ord a => a -> a -> Bool)
-> Maybe CoreRule
-- See Note [Rules for floating-point comparisons]
mkFloatingRelOpRule nm cmp
= mkPrimOpRule nm 2 [binaryCmpLit cmp]
-- common constants
zeroi, onei, zerow, onew :: Platform -> Literal
zeroi platform = mkLitInt platform 0
onei platform = mkLitInt platform 1
zerow platform = mkLitWord platform 0
onew platform = mkLitWord platform 1
zerof, onef, twof, zerod, oned, twod :: Literal
zerof = mkLitFloat 0.0
onef = mkLitFloat 1.0
twof = mkLitFloat 2.0
zerod = mkLitDouble 0.0
oned = mkLitDouble 1.0
twod = mkLitDouble 2.0
cmpOp :: Platform -> (forall a . Ord a => a -> a -> Bool)
-> Literal -> Literal -> Maybe CoreExpr
cmpOp platform cmp = go
where
done True = Just $ trueValInt platform
done False = Just $ falseValInt platform
-- These compares are at different types
go (LitChar i1) (LitChar i2) = done (i1 `cmp` i2)
go (LitFloat i1) (LitFloat i2) = done (i1 `cmp` i2)
go (LitDouble i1) (LitDouble i2) = done (i1 `cmp` i2)
go (LitNumber nt1 i1) (LitNumber nt2 i2)
| nt1 /= nt2 = Nothing
| otherwise = done (i1 `cmp` i2)
go _ _ = Nothing
--------------------------
negOp :: RuleOpts -> Literal -> Maybe CoreExpr -- Negate
negOp env = \case
(LitFloat 0.0) -> Nothing -- can't represent -0.0 as a Rational
(LitFloat f) -> Just (mkFloatVal env (-f))
(LitDouble 0.0) -> Nothing
(LitDouble d) -> Just (mkDoubleVal env (-d))
(LitNumber nt i)
| litNumIsSigned nt -> Just (Lit (mkLitNumberWrap (roPlatform env) nt (-i)))
_ -> Nothing
complementOp :: RuleOpts -> Literal -> Maybe CoreExpr -- Binary complement
complementOp env (LitNumber nt i) =
Just (Lit (mkLitNumberWrap (roPlatform env) nt (complement i)))
complementOp _ _ = Nothing
intOp2 :: (Integral a, Integral b)
=> (a -> b -> Integer)
-> RuleOpts -> Literal -> Literal -> Maybe CoreExpr
intOp2 = intOp2' . const
intOp2' :: (Integral a, Integral b)
=> (RuleOpts -> a -> b -> Integer)
-> RuleOpts -> Literal -> Literal -> Maybe CoreExpr
intOp2' op env (LitNumber LitNumInt i1) (LitNumber LitNumInt i2) =
let o = op env
in intResult (roPlatform env) (fromInteger i1 `o` fromInteger i2)
intOp2' _ _ _ _ = Nothing
intOpC2 :: (Integral a, Integral b)
=> (a -> b -> Integer)
-> RuleOpts -> Literal -> Literal -> Maybe CoreExpr
intOpC2 op env (LitNumber LitNumInt i1) (LitNumber LitNumInt i2) = do
intCResult (roPlatform env) (fromInteger i1 `op` fromInteger i2)
intOpC2 _ _ _ _ = Nothing
shiftRightLogical :: Platform -> Integer -> Int -> Integer
-- Shift right, putting zeros in rather than sign-propagating as Bits.shiftR would do
-- Do this by converting to Word and back. Obviously this won't work for big
-- values, but its ok as we use it here
shiftRightLogical platform x n =
case platformWordSize platform of
PW4 -> fromIntegral (fromInteger x `shiftR` n :: Word32)
PW8 -> fromIntegral (fromInteger x `shiftR` n :: Word64)
--------------------------
retLit :: (Platform -> Literal) -> RuleM CoreExpr
retLit l = do platform <- getPlatform
return $ Lit $ l platform
retLitNoC :: (Platform -> Literal) -> RuleM CoreExpr
retLitNoC l = do platform <- getPlatform
let lit = l platform
let ty = literalType lit
return $ mkCoreUbxTup [ty, ty] [Lit lit, Lit (zeroi platform)]
wordOp2 :: (Integral a, Integral b)
=> (a -> b -> Integer)
-> RuleOpts -> Literal -> Literal -> Maybe CoreExpr
wordOp2 op env (LitNumber LitNumWord w1) (LitNumber LitNumWord w2)
= wordResult (roPlatform env) (fromInteger w1 `op` fromInteger w2)
wordOp2 _ _ _ _ = Nothing
wordOpC2 :: (Integral a, Integral b)
=> (a -> b -> Integer)
-> RuleOpts -> Literal -> Literal -> Maybe CoreExpr
wordOpC2 op env (LitNumber LitNumWord w1) (LitNumber LitNumWord w2) =
wordCResult (roPlatform env) (fromInteger w1 `op` fromInteger w2)
wordOpC2 _ _ _ _ = Nothing
shiftRule :: LitNumType -- Type of the result, either LitNumInt or LitNumWord
-> (Platform -> Integer -> Int -> Integer)
-> RuleM CoreExpr
-- Shifts take an Int; hence third arg of op is Int
-- Used for shift primops
-- ISllOp, ISraOp, ISrlOp :: Int# -> Int# -> Int#
-- SllOp, SrlOp :: Word# -> Int# -> Word#
shiftRule lit_num_ty shift_op
= do { platform <- getPlatform
; [e1, Lit (LitNumber LitNumInt shift_len)] <- getArgs
; case e1 of
_ | shift_len == 0
-> return e1
-- See Note [Guarding against silly shifts]
| shift_len < 0 || shift_len > toInteger (platformWordSizeInBits platform)
-> return $ Lit $ mkLitNumberWrap platform lit_num_ty 0
-- Be sure to use lit_num_ty here, so we get a correctly typed zero
-- of type Int# or Word# resp. See #18589
-- Do the shift at type Integer, but shift length is Int
Lit (LitNumber nt x)
| 0 < shift_len
, shift_len <= toInteger (platformWordSizeInBits platform)
-> let op = shift_op platform
y = x `op` fromInteger shift_len
in liftMaybe $ Just (Lit (mkLitNumberWrap platform nt y))
_ -> mzero }
--------------------------
floatOp2 :: (Rational -> Rational -> Rational)
-> RuleOpts -> Literal -> Literal
-> Maybe (Expr CoreBndr)
floatOp2 op env (LitFloat f1) (LitFloat f2)
= Just (mkFloatVal env (f1 `op` f2))
floatOp2 _ _ _ _ = Nothing
--------------------------
floatDecodeOp :: RuleOpts -> Literal -> Maybe CoreExpr
floatDecodeOp env (LitFloat ((decodeFloat . fromRational @Float) -> (m, e)))
= Just $ mkCoreUbxTup [intPrimTy, intPrimTy]
[ mkIntVal (roPlatform env) (toInteger m)
, mkIntVal (roPlatform env) (toInteger e) ]
floatDecodeOp _ _
= Nothing
--------------------------
doubleOp2 :: (Rational -> Rational -> Rational)
-> RuleOpts -> Literal -> Literal
-> Maybe (Expr CoreBndr)
doubleOp2 op env (LitDouble f1) (LitDouble f2)
= Just (mkDoubleVal env (f1 `op` f2))
doubleOp2 _ _ _ _ = Nothing
--------------------------
doubleDecodeOp :: RuleOpts -> Literal -> Maybe CoreExpr
doubleDecodeOp env (LitDouble ((decodeFloat . fromRational @Double) -> (m, e)))
= Just $ mkCoreUbxTup [iNT64Ty, intPrimTy]
[ Lit (mkLitINT64 (roPlatform env) (toInteger m))
, mkIntVal platform (toInteger e) ]
where
platform = roPlatform env
(iNT64Ty, mkLitINT64)
| platformWordSizeInBits platform < 64
= (int64PrimTy, mkLitInt64Wrap)
| otherwise
= (intPrimTy , mkLitIntWrap)
doubleDecodeOp _ _
= Nothing
--------------------------
{- Note [The litEq rule: converting equality to case]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This stuff turns
n ==# 3#
into
case n of
3# -> True
m -> False
This is a Good Thing, because it allows case-of case things
to happen, and case-default absorption to happen. For
example:
if (n ==# 3#) || (n ==# 4#) then e1 else e2
will transform to
case n of
3# -> e1
4# -> e1
m -> e2
(modulo the usual precautions to avoid duplicating e1)
-}
litEq :: Bool -- True <=> equality, False <=> inequality
-> RuleM CoreExpr
litEq is_eq = msum
[ do [Lit lit, expr] <- getArgs
platform <- getPlatform
do_lit_eq platform lit expr
, do [expr, Lit lit] <- getArgs
platform <- getPlatform
do_lit_eq platform lit expr ]
where
do_lit_eq platform lit expr = do
guard (not (litIsLifted lit))
return (mkWildCase expr (unrestricted $ literalType lit) intPrimTy
[(DEFAULT, [], val_if_neq),
(LitAlt lit, [], val_if_eq)])
where
val_if_eq | is_eq = trueValInt platform
| otherwise = falseValInt platform
val_if_neq | is_eq = falseValInt platform
| otherwise = trueValInt platform
-- | Check if there is comparison with minBound or maxBound, that is
-- always true or false. For instance, an Int cannot be smaller than its
-- minBound, so we can replace such comparison with False.
boundsCmp :: Comparison -> RuleM CoreExpr
boundsCmp op = do
platform <- getPlatform
[a, b] <- getArgs
liftMaybe $ mkRuleFn platform op a b
data Comparison = Gt | Ge | Lt | Le
mkRuleFn :: Platform -> Comparison -> CoreExpr -> CoreExpr -> Maybe CoreExpr
mkRuleFn platform Gt (Lit lit) _ | isMinBound platform lit = Just $ falseValInt platform
mkRuleFn platform Le (Lit lit) _ | isMinBound platform lit = Just $ trueValInt platform
mkRuleFn platform Ge _ (Lit lit) | isMinBound platform lit = Just $ trueValInt platform
mkRuleFn platform Lt _ (Lit lit) | isMinBound platform lit = Just $ falseValInt platform
mkRuleFn platform Ge (Lit lit) _ | isMaxBound platform lit = Just $ trueValInt platform
mkRuleFn platform Lt (Lit lit) _ | isMaxBound platform lit = Just $ falseValInt platform
mkRuleFn platform Gt _ (Lit lit) | isMaxBound platform lit = Just $ falseValInt platform
mkRuleFn platform Le _ (Lit lit) | isMaxBound platform lit = Just $ trueValInt platform
mkRuleFn _ _ _ _ = Nothing
isMinBound :: Platform -> Literal -> Bool
isMinBound _ (LitChar c) = c == minBound
isMinBound platform (LitNumber nt i) = case nt of
LitNumInt -> i == platformMinInt platform
LitNumInt64 -> i == toInteger (minBound :: Int64)
LitNumWord -> i == 0
LitNumWord64 -> i == 0
LitNumNatural -> i == 0
LitNumInteger -> False
isMinBound _ _ = False
isMaxBound :: Platform -> Literal -> Bool
isMaxBound _ (LitChar c) = c == maxBound
isMaxBound platform (LitNumber nt i) = case nt of
LitNumInt -> i == platformMaxInt platform
LitNumInt64 -> i == toInteger (maxBound :: Int64)
LitNumWord -> i == platformMaxWord platform
LitNumWord64 -> i == toInteger (maxBound :: Word64)
LitNumNatural -> False
LitNumInteger -> False
isMaxBound _ _ = False
-- | Create an Int literal expression while ensuring the given Integer is in the
-- target Int range
intResult :: Platform -> Integer -> Maybe CoreExpr
intResult platform result = Just (intResult' platform result)
intResult' :: Platform -> Integer -> CoreExpr
intResult' platform result = Lit (mkLitIntWrap platform result)
-- | Create an unboxed pair of an Int literal expression, ensuring the given
-- Integer is in the target Int range and the corresponding overflow flag
-- (@0#@/@1#@) if it wasn't.
intCResult :: Platform -> Integer -> Maybe CoreExpr
intCResult platform result = Just (mkPair [Lit lit, Lit c])
where
mkPair = mkCoreUbxTup [intPrimTy, intPrimTy]
(lit, b) = mkLitIntWrapC platform result
c = if b then onei platform else zeroi platform
-- | Create a Word literal expression while ensuring the given Integer is in the
-- target Word range
wordResult :: Platform -> Integer -> Maybe CoreExpr
wordResult platform result = Just (wordResult' platform result)
wordResult' :: Platform -> Integer -> CoreExpr
wordResult' platform result = Lit (mkLitWordWrap platform result)
-- | Create an unboxed pair of a Word literal expression, ensuring the given
-- Integer is in the target Word range and the corresponding carry flag
-- (@0#@/@1#@) if it wasn't.
wordCResult :: Platform -> Integer -> Maybe CoreExpr
wordCResult platform result = Just (mkPair [Lit lit, Lit c])
where
mkPair = mkCoreUbxTup [wordPrimTy, intPrimTy]
(lit, b) = mkLitWordWrapC platform result
c = if b then onei platform else zeroi platform
inversePrimOp :: PrimOp -> RuleM CoreExpr
inversePrimOp primop = do
[Var primop_id `App` e] <- getArgs
matchPrimOpId primop primop_id
return e
subsumesPrimOp :: PrimOp -> PrimOp -> RuleM CoreExpr
this `subsumesPrimOp` that = do
[Var primop_id `App` e] <- getArgs
matchPrimOpId that primop_id
return (Var (mkPrimOpId this) `App` e)
subsumedByPrimOp :: PrimOp -> RuleM CoreExpr
subsumedByPrimOp primop = do
[e@(Var primop_id `App` _)] <- getArgs
matchPrimOpId primop primop_id
return e
-- | narrow subsumes bitwise `and` with full mask (cf #16402):
--
-- narrowN (x .&. m)
-- m .&. (2^N-1) = 2^N-1
-- ==> narrowN x
--
-- e.g. narrow16 (x .&. 0xFFFF)
-- ==> narrow16 x
--
narrowSubsumesAnd :: PrimOp -> PrimOp -> Int -> RuleM CoreExpr
narrowSubsumesAnd and_primop narrw n = do
[Var primop_id `App` x `App` y] <- getArgs
matchPrimOpId and_primop primop_id
let mask = bit n -1
g v (Lit (LitNumber _ m)) = do
guard (m .&. mask == mask)
return (Var (mkPrimOpId narrw) `App` v)
g _ _ = mzero
g x y <|> g y x
idempotent :: RuleM CoreExpr
idempotent = do [e1, e2] <- getArgs
guard $ cheapEqExpr e1 e2
return e1
{-
Note [Guarding against silly shifts]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider this code:
import Data.Bits( (.|.), shiftL )
chunkToBitmap :: [Bool] -> Word32
chunkToBitmap chunk = foldr (.|.) 0 [ 1 `shiftL` n | (True,n) <- zip chunk [0..] ]
This optimises to:
Shift.$wgo = \ (w_sCS :: GHC.Prim.Int#) (w1_sCT :: [GHC.Types.Bool]) ->
case w1_sCT of _ {
[] -> 0##;
: x_aAW xs_aAX ->
case x_aAW of _ {
GHC.Types.False ->
case w_sCS of wild2_Xh {
__DEFAULT -> Shift.$wgo (GHC.Prim.+# wild2_Xh 1) xs_aAX;
9223372036854775807 -> 0## };
GHC.Types.True ->
case GHC.Prim.>=# w_sCS 64 of _ {
GHC.Types.False ->
case w_sCS of wild3_Xh {
__DEFAULT ->
case Shift.$wgo (GHC.Prim.+# wild3_Xh 1) xs_aAX of ww_sCW { __DEFAULT ->
GHC.Prim.or# (GHC.Prim.narrow32Word#
(GHC.Prim.uncheckedShiftL# 1## wild3_Xh))
ww_sCW
};
9223372036854775807 ->
GHC.Prim.narrow32Word#
!!!!--> (GHC.Prim.uncheckedShiftL# 1## 9223372036854775807)
};
GHC.Types.True ->
case w_sCS of wild3_Xh {
__DEFAULT -> Shift.$wgo (GHC.Prim.+# wild3_Xh 1) xs_aAX;
9223372036854775807 -> 0##
} } } }
Note the massive shift on line "!!!!". It can't happen, because we've checked
that w < 64, but the optimiser didn't spot that. We DO NOT want to constant-fold this!
Moreover, if the programmer writes (n `uncheckedShiftL` 9223372036854775807), we
can't constant fold it, but if it gets to the assembler we get
Error: operand type mismatch for `shl'
So the best thing to do is to rewrite the shift with a call to error,
when the second arg is large. However, in general we cannot do this; consider
this case
let x = I# (uncheckedIShiftL# n 80)
in ...
Here x contains an invalid shift and consequently we would like to rewrite it
as follows:
let x = I# (error "invalid shift)
in ...
This was originally done in the fix to #16449 but this breaks the let/app
invariant (see Note [Core let/app invariant] in GHC.Core) as noted in #16742.
For the reasons discussed in Note [Checking versus non-checking primops] (in
the PrimOp module) there is no safe way rewrite the argument of I# such that
it bottoms.
Consequently we instead take advantage of the fact that large shifts are
undefined behavior (see associated documentation in primops.txt.pp) and
transform the invalid shift into an "obviously incorrect" value.
There are two cases:
- Shifting fixed-width things: the primops ISll, Sll, etc
These are handled by shiftRule.
We are happy to shift by any amount up to wordSize but no more.
- Shifting Bignums (Integer, Natural): these are handled by bignum_shift.
Here we could in principle shift by any amount, but we arbitrary
limit the shift to 4 bits; in particular we do not want shift by a
huge amount, which can happen in code like that above.
The two cases are more different in their code paths that is comfortable,
but that is only a historical accident.
************************************************************************
* *
\subsection{Vaguely generic functions}
* *
************************************************************************
-}
mkBasicRule :: Name -> Int -> RuleM CoreExpr -> CoreRule
-- Gives the Rule the same name as the primop itself
mkBasicRule op_name n_args rm
= BuiltinRule { ru_name = occNameFS (nameOccName op_name),
ru_fn = op_name,
ru_nargs = n_args,
ru_try = runRuleM rm }
newtype RuleM r = RuleM
{ runRuleM :: RuleOpts -> InScopeEnv -> Id -> [CoreExpr] -> Maybe r }
deriving (Functor)
instance Applicative RuleM where
pure x = RuleM $ \_ _ _ _ -> Just x
(<*>) = ap
instance Monad RuleM where
RuleM f >>= g
= RuleM $ \env iu fn args ->
case f env iu fn args of
Nothing -> Nothing
Just r -> runRuleM (g r) env iu fn args
instance MonadFail RuleM where
fail _ = mzero
instance Alternative RuleM where
empty = RuleM $ \_ _ _ _ -> Nothing
RuleM f1 <|> RuleM f2 = RuleM $ \env iu fn args ->
f1 env iu fn args <|> f2 env iu fn args
instance MonadPlus RuleM
getPlatform :: RuleM Platform
getPlatform = roPlatform <$> getEnv
getEnv :: RuleM RuleOpts
getEnv = RuleM $ \env _ _ _ -> Just env
liftMaybe :: Maybe a -> RuleM a
liftMaybe Nothing = mzero
liftMaybe (Just x) = return x
liftLit :: (Literal -> Literal) -> RuleM CoreExpr
liftLit f = liftLitPlatform (const f)
liftLitPlatform :: (Platform -> Literal -> Literal) -> RuleM CoreExpr
liftLitPlatform f = do
platform <- getPlatform
[Lit lit] <- getArgs
return $ Lit (f platform lit)
removeOp32 :: RuleM CoreExpr
removeOp32 = do
platform <- getPlatform
case platformWordSize platform of
PW4 -> do
[e] <- getArgs
return e
PW8 ->
mzero
getArgs :: RuleM [CoreExpr]
getArgs = RuleM $ \_ _ _ args -> Just args
getInScopeEnv :: RuleM InScopeEnv
getInScopeEnv = RuleM $ \_ iu _ _ -> Just iu
getFunction :: RuleM Id
getFunction = RuleM $ \_ _ fn _ -> Just fn
isLiteral :: CoreExpr -> RuleM Literal
isLiteral e = do
env <- getInScopeEnv
case exprIsLiteral_maybe env e of
Nothing -> mzero
Just l -> pure l
isNumberLiteral :: CoreExpr -> RuleM Integer
isNumberLiteral e = isLiteral e >>= \case
LitNumber _ x -> pure x
_ -> mzero
isIntegerLiteral :: CoreExpr -> RuleM Integer
isIntegerLiteral e = isLiteral e >>= \case
LitNumber LitNumInteger x -> pure x
_ -> mzero
isNaturalLiteral :: CoreExpr -> RuleM Integer
isNaturalLiteral e = isLiteral e >>= \case
LitNumber LitNumNatural x -> pure x
_ -> mzero
isWordLiteral :: CoreExpr -> RuleM Integer
isWordLiteral e = isLiteral e >>= \case
LitNumber LitNumWord x -> pure x
_ -> mzero
isIntLiteral :: CoreExpr -> RuleM Integer
isIntLiteral e = isLiteral e >>= \case
LitNumber LitNumInt x -> pure x
_ -> mzero
-- return the n-th argument of this rule, if it is a literal
-- argument indices start from 0
getLiteral :: Int -> RuleM Literal
getLiteral n = RuleM $ \_ _ _ exprs -> case drop n exprs of
(Lit l:_) -> Just l
_ -> Nothing
unaryLit :: (RuleOpts -> Literal -> Maybe CoreExpr) -> RuleM CoreExpr
unaryLit op = do
env <- getEnv
[Lit l] <- getArgs
liftMaybe $ op env (convFloating env l)
binaryLit :: (RuleOpts -> Literal -> Literal -> Maybe CoreExpr) -> RuleM CoreExpr
binaryLit op = do
env <- getEnv
[Lit l1, Lit l2] <- getArgs
liftMaybe $ op env (convFloating env l1) (convFloating env l2)
binaryCmpLit :: (forall a . Ord a => a -> a -> Bool) -> RuleM CoreExpr
binaryCmpLit op = do
platform <- getPlatform
binaryLit (\_ -> cmpOp platform op)
leftIdentity :: Literal -> RuleM CoreExpr
leftIdentity id_lit = leftIdentityPlatform (const id_lit)
rightIdentity :: Literal -> RuleM CoreExpr
rightIdentity id_lit = rightIdentityPlatform (const id_lit)
identity :: Literal -> RuleM CoreExpr
identity lit = leftIdentity lit `mplus` rightIdentity lit
leftIdentityPlatform :: (Platform -> Literal) -> RuleM CoreExpr
leftIdentityPlatform id_lit = do
platform <- getPlatform
[Lit l1, e2] <- getArgs
guard $ l1 == id_lit platform
return e2
-- | Left identity rule for PrimOps like 'IntAddC' and 'WordAddC', where, in
-- addition to the result, we have to indicate that no carry/overflow occurred.
leftIdentityCPlatform :: (Platform -> Literal) -> RuleM CoreExpr
leftIdentityCPlatform id_lit = do
platform <- getPlatform
[Lit l1, e2] <- getArgs
guard $ l1 == id_lit platform
let no_c = Lit (zeroi platform)
return (mkCoreUbxTup [exprType e2, intPrimTy] [e2, no_c])
rightIdentityPlatform :: (Platform -> Literal) -> RuleM CoreExpr
rightIdentityPlatform id_lit = do
platform <- getPlatform
[e1, Lit l2] <- getArgs
guard $ l2 == id_lit platform
return e1
-- | Right identity rule for PrimOps like 'IntSubC' and 'WordSubC', where, in
-- addition to the result, we have to indicate that no carry/overflow occurred.
rightIdentityCPlatform :: (Platform -> Literal) -> RuleM CoreExpr
rightIdentityCPlatform id_lit = do
platform <- getPlatform
[e1, Lit l2] <- getArgs
guard $ l2 == id_lit platform
let no_c = Lit (zeroi platform)
return (mkCoreUbxTup [exprType e1, intPrimTy] [e1, no_c])
identityPlatform :: (Platform -> Literal) -> RuleM CoreExpr
identityPlatform lit =
leftIdentityPlatform lit `mplus` rightIdentityPlatform lit
-- | Identity rule for PrimOps like 'IntAddC' and 'WordAddC', where, in addition
-- to the result, we have to indicate that no carry/overflow occurred.
identityCPlatform :: (Platform -> Literal) -> RuleM CoreExpr
identityCPlatform lit =
leftIdentityCPlatform lit `mplus` rightIdentityCPlatform lit
leftZero :: (Platform -> Literal) -> RuleM CoreExpr
leftZero zero = do
platform <- getPlatform
[Lit l1, _] <- getArgs
guard $ l1 == zero platform
return $ Lit l1
rightZero :: (Platform -> Literal) -> RuleM CoreExpr
rightZero zero = do
platform <- getPlatform
[_, Lit l2] <- getArgs
guard $ l2 == zero platform
return $ Lit l2
zeroElem :: (Platform -> Literal) -> RuleM CoreExpr
zeroElem lit = leftZero lit `mplus` rightZero lit
equalArgs :: RuleM ()
equalArgs = do
[e1, e2] <- getArgs
guard $ e1 `cheapEqExpr` e2
nonZeroLit :: Int -> RuleM ()
nonZeroLit n = getLiteral n >>= guard . not . isZeroLit
-- When excess precision is not requested, cut down the precision of the
-- Rational value to that of Float/Double. We confuse host architecture
-- and target architecture here, but it's convenient (and wrong :-).
convFloating :: RuleOpts -> Literal -> Literal
convFloating env (LitFloat f) | not (roExcessRationalPrecision env) =
LitFloat (toRational (fromRational f :: Float ))
convFloating env (LitDouble d) | not (roExcessRationalPrecision env) =
LitDouble (toRational (fromRational d :: Double))
convFloating _ l = l
guardFloatDiv :: RuleM ()
guardFloatDiv = do
[Lit (LitFloat f1), Lit (LitFloat f2)] <- getArgs
guard $ (f1 /=0 || f2 > 0) -- see Note [negative zero]
&& f2 /= 0 -- avoid NaN and Infinity/-Infinity
guardDoubleDiv :: RuleM ()
guardDoubleDiv = do
[Lit (LitDouble d1), Lit (LitDouble d2)] <- getArgs
guard $ (d1 /=0 || d2 > 0) -- see Note [negative zero]
&& d2 /= 0 -- avoid NaN and Infinity/-Infinity
-- Note [negative zero] Avoid (0 / -d), otherwise 0/(-1) reduces to
-- zero, but we might want to preserve the negative zero here which
-- is representable in Float/Double but not in (normalised)
-- Rational. (#3676) Perhaps we should generate (0 :% (-1)) instead?
strengthReduction :: Literal -> PrimOp -> RuleM CoreExpr
strengthReduction two_lit add_op = do -- Note [Strength reduction]
arg <- msum [ do [arg, Lit mult_lit] <- getArgs
guard (mult_lit == two_lit)
return arg
, do [Lit mult_lit, arg] <- getArgs
guard (mult_lit == two_lit)
return arg ]
return $ Var (mkPrimOpId add_op) `App` arg `App` arg
-- Note [Strength reduction]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- This rule turns floating point multiplications of the form 2.0 * x and
-- x * 2.0 into x + x addition, because addition costs less than multiplication.
-- See #7116
-- Note [What's true and false]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- trueValInt and falseValInt represent true and false values returned by
-- comparison primops for Char, Int, Word, Integer, Double, Float and Addr.
-- True is represented as an unboxed 1# literal, while false is represented
-- as 0# literal.
-- We still need Bool data constructors (True and False) to use in a rule
-- for constant folding of equal Strings
trueValInt, falseValInt :: Platform -> Expr CoreBndr
trueValInt platform = Lit $ onei platform -- see Note [What's true and false]
falseValInt platform = Lit $ zeroi platform
trueValBool, falseValBool :: Expr CoreBndr
trueValBool = Var trueDataConId -- see Note [What's true and false]
falseValBool = Var falseDataConId
ltVal, eqVal, gtVal :: Expr CoreBndr
ltVal = Var ordLTDataConId
eqVal = Var ordEQDataConId
gtVal = Var ordGTDataConId
mkIntVal :: Platform -> Integer -> Expr CoreBndr
mkIntVal platform i = Lit (mkLitInt platform i)
mkFloatVal :: RuleOpts -> Rational -> Expr CoreBndr
mkFloatVal env f = Lit (convFloating env (LitFloat f))
mkDoubleVal :: RuleOpts -> Rational -> Expr CoreBndr
mkDoubleVal env d = Lit (convFloating env (LitDouble d))
matchPrimOpId :: PrimOp -> Id -> RuleM ()
matchPrimOpId op id = do
op' <- liftMaybe $ isPrimOpId_maybe id
guard $ op == op'
{-
************************************************************************
* *
\subsection{Special rules for seq, tagToEnum, dataToTag}
* *
************************************************************************
Note [tagToEnum#]
~~~~~~~~~~~~~~~~~
Nasty check to ensure that tagToEnum# is applied to a type that is an
enumeration TyCon. Unification may refine the type later, but this
check won't see that, alas. It's crude but it works.
Here's are two cases that should fail
f :: forall a. a
f = tagToEnum# 0 -- Can't do tagToEnum# at a type variable
g :: Int
g = tagToEnum# 0 -- Int is not an enumeration
We used to make this check in the type inference engine, but it's quite
ugly to do so, because the delayed constraint solving means that we don't
really know what's going on until the end. It's very much a corner case
because we don't expect the user to call tagToEnum# at all; we merely
generate calls in derived instances of Enum. So we compromise: a
rewrite rule rewrites a bad instance of tagToEnum# to an error call,
and emits a warning.
-}
tagToEnumRule :: RuleM CoreExpr
-- If data T a = A | B | C
-- then tagToEnum# (T ty) 2# --> B ty
tagToEnumRule = do
[Type ty, Lit (LitNumber LitNumInt i)] <- getArgs
case splitTyConApp_maybe ty of
Just (tycon, tc_args) | isEnumerationTyCon tycon -> do
let tag = fromInteger i
correct_tag dc = (dataConTagZ dc) == tag
(dc:rest) <- return $ filter correct_tag (tyConDataCons_maybe tycon `orElse` [])
ASSERT(null rest) return ()
return $ mkTyApps (Var (dataConWorkId dc)) tc_args
-- See Note [tagToEnum#]
_ -> WARN( True, text "tagToEnum# on non-enumeration type" <+> ppr ty )
return $ mkRuntimeErrorApp rUNTIME_ERROR_ID ty "tagToEnum# on non-enumeration type"
------------------------------
dataToTagRule :: RuleM CoreExpr
-- See Note [dataToTag#] in primops.txt.pp
dataToTagRule = a `mplus` b
where
-- dataToTag (tagToEnum x) ==> x
a = do
[Type ty1, Var tag_to_enum `App` Type ty2 `App` tag] <- getArgs
guard $ tag_to_enum `hasKey` tagToEnumKey
guard $ ty1 `eqType` ty2
return tag
-- dataToTag (K e1 e2) ==> tag-of K
-- This also works (via exprIsConApp_maybe) for
-- dataToTag x
-- where x's unfolding is a constructor application
b = do
dflags <- getPlatform
[_, val_arg] <- getArgs
in_scope <- getInScopeEnv
(_,floats, dc,_,_) <- liftMaybe $ exprIsConApp_maybe in_scope val_arg
ASSERT( not (isNewTyCon (dataConTyCon dc)) ) return ()
return $ wrapFloats floats (mkIntVal dflags (toInteger (dataConTagZ dc)))
{- Note [dataToTag# magic]
~~~~~~~~~~~~~~~~~~~~~~~~~~
The primop dataToTag# is unusual because it evaluates its argument.
Only `SeqOp` shares that property. (Other primops do not do anything
as fancy as argument evaluation.) The special handling for dataToTag#
is:
* GHC.Core.Utils.exprOkForSpeculation has a special case for DataToTagOp,
(actually in app_ok). Most primops with lifted arguments do not
evaluate those arguments, but DataToTagOp and SeqOp are two
exceptions. We say that they are /never/ ok-for-speculation,
regardless of the evaluated-ness of their argument.
See GHC.Core.Utils Note [exprOkForSpeculation and SeqOp/DataToTagOp]
* There is a special case for DataToTagOp in GHC.StgToCmm.Expr.cgExpr,
that evaluates its argument and then extracts the tag from
the returned value.
* An application like (dataToTag# (Just x)) is optimised by
dataToTagRule in GHC.Core.Opt.ConstantFold.
* A case expression like
case (dataToTag# e) of <alts>
gets transformed t
case e of <transformed alts>
by GHC.Core.Opt.ConstantFold.caseRules; see Note [caseRules for dataToTag]
See #15696 for a long saga.
-}
{- *********************************************************************
* *
unsafeEqualityProof
* *
********************************************************************* -}
-- unsafeEqualityProof k t t ==> UnsafeRefl (Refl t)
-- That is, if the two types are equal, it's not unsafe!
unsafeEqualityProofRule :: RuleM CoreExpr
unsafeEqualityProofRule
= do { [Type rep, Type t1, Type t2] <- getArgs
; guard (t1 `eqType` t2)
; fn <- getFunction
; let (_, ue) = splitForAllTys (idType fn)
tc = tyConAppTyCon ue -- tycon: UnsafeEquality
(dc:_) = tyConDataCons tc -- data con: UnsafeRefl
-- UnsafeRefl :: forall (r :: RuntimeRep) (a :: TYPE r).
-- UnsafeEquality r a a
; return (mkTyApps (Var (dataConWrapId dc)) [rep, t1]) }
{- *********************************************************************
* *
Rules for seq# and spark#
* *
********************************************************************* -}
{- Note [seq# magic]
~~~~~~~~~~~~~~~~~~~~
The primop
seq# :: forall a s . a -> State# s -> (# State# s, a #)
is /not/ the same as the Prelude function seq :: a -> b -> b
as you can see from its type. In fact, seq# is the implementation
mechanism for 'evaluate'
evaluate :: a -> IO a
evaluate a = IO $ \s -> seq# a s
The semantics of seq# is
* evaluate its first argument
* and return it
Things to note
* Why do we need a primop at all? That is, instead of
case seq# x s of (# x, s #) -> blah
why not instead say this?
case x of { DEFAULT -> blah)
Reason (see #5129): if we saw
catch# (\s -> case x of { DEFAULT -> raiseIO# exn s }) handler
then we'd drop the 'case x' because the body of the case is bottom
anyway. But we don't want to do that; the whole /point/ of
seq#/evaluate is to evaluate 'x' first in the IO monad.
In short, we /always/ evaluate the first argument and never
just discard it.
* Why return the value? So that we can control sharing of seq'd
values: in
let x = e in x `seq` ... x ...
We don't want to inline x, so better to represent it as
let x = e in case seq# x RW of (# _, x' #) -> ... x' ...
also it matches the type of rseq in the Eval monad.
Implementing seq#. The compiler has magic for SeqOp in
- GHC.Core.Opt.ConstantFold.seqRule: eliminate (seq# <whnf> s)
- GHC.StgToCmm.Expr.cgExpr, and cgCase: special case for seq#
- GHC.Core.Utils.exprOkForSpeculation;
see Note [exprOkForSpeculation and SeqOp/DataToTagOp] in GHC.Core.Utils
- Simplify.addEvals records evaluated-ness for the result; see
Note [Adding evaluatedness info to pattern-bound variables]
in GHC.Core.Opt.Simplify
-}
seqRule :: RuleM CoreExpr
seqRule = do
[Type ty_a, Type _ty_s, a, s] <- getArgs
guard $ exprIsHNF a
return $ mkCoreUbxTup [exprType s, ty_a] [s, a]
-- spark# :: forall a s . a -> State# s -> (# State# s, a #)
sparkRule :: RuleM CoreExpr
sparkRule = seqRule -- reduce on HNF, just the same
-- XXX perhaps we shouldn't do this, because a spark eliminated by
-- this rule won't be counted as a dud at runtime?
{-
************************************************************************
* *
\subsection{Built in rules}
* *
************************************************************************
Note [Scoping for Builtin rules]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
When compiling a (base-package) module that defines one of the
functions mentioned in the RHS of a built-in rule, there's a danger
that we'll see
f = ...(eq String x)....
....and lower down...
eqString = ...
Then a rewrite would give
f = ...(eqString x)...
....and lower down...
eqString = ...
and lo, eqString is not in scope. This only really matters when we
get to code generation. But the occurrence analyser does a GlomBinds
step when necessary, that does a new SCC analysis on the whole set of
bindings (see occurAnalysePgm), which sorts out the dependency, so all
is fine.
-}
newtype EnableBignumRules = EnableBignumRules Bool
builtinRules :: EnableBignumRules -> [CoreRule]
-- Rules for non-primops that can't be expressed using a RULE pragma
builtinRules enableBignumRules
= [BuiltinRule { ru_name = fsLit "AppendLitString",
ru_fn = unpackCStringFoldrName,
ru_nargs = 4, ru_try = match_append_lit_C },
BuiltinRule { ru_name = fsLit "AppendLitStringUtf8",
ru_fn = unpackCStringFoldrUtf8Name,
ru_nargs = 4, ru_try = match_append_lit_utf8 },
BuiltinRule { ru_name = fsLit "EqString", ru_fn = eqStringName,
ru_nargs = 2, ru_try = match_eq_string },
BuiltinRule { ru_name = fsLit "CStringLength", ru_fn = cstringLengthName,
ru_nargs = 1, ru_try = match_cstring_length },
BuiltinRule { ru_name = fsLit "Inline", ru_fn = inlineIdName,
ru_nargs = 2, ru_try = \_ _ _ -> match_inline },
BuiltinRule { ru_name = fsLit "MagicDict", ru_fn = idName magicDictId,
ru_nargs = 4, ru_try = \_ _ _ -> match_magicDict },
mkBasicRule unsafeEqualityProofName 3 unsafeEqualityProofRule,
mkBasicRule divIntName 2 $ msum
[ nonZeroLit 1 >> binaryLit (intOp2 div)
, leftZero zeroi
, do
[arg, Lit (LitNumber LitNumInt d)] <- getArgs
Just n <- return $ exactLog2 d
platform <- getPlatform
return $ Var (mkPrimOpId ISraOp) `App` arg `App` mkIntVal platform n
],
mkBasicRule modIntName 2 $ msum
[ nonZeroLit 1 >> binaryLit (intOp2 mod)
, leftZero zeroi
, do
[arg, Lit (LitNumber LitNumInt d)] <- getArgs
Just _ <- return $ exactLog2 d
platform <- getPlatform
return $ Var (mkPrimOpId AndIOp)
`App` arg `App` mkIntVal platform (d - 1)
]
]
++ builtinBignumRules enableBignumRules
{-# NOINLINE builtinRules #-}
-- there is no benefit to inlining these yet, despite this, GHC produces
-- unfoldings for this regardless since the floated list entries look small.
builtinBignumRules :: EnableBignumRules -> [CoreRule]
builtinBignumRules (EnableBignumRules False) = []
builtinBignumRules _ =
[ -- conversions
lit_to_integer "Word# -> Integer" integerFromWordName
, lit_to_integer "Int64# -> Integer" integerFromInt64Name
, lit_to_integer "Word64# -> Integer" integerFromWord64Name
, lit_to_integer "Natural -> Integer" integerFromNaturalName
, integer_to_lit "Integer -> Word# (wrap)" integerToWordName mkWordLitWrap
, integer_to_lit "Integer -> Int# (wrap)" integerToIntName mkIntLitWrap
, integer_to_lit "Integer -> Word64# (wrap)" integerToWord64Name (\_ -> mkWord64LitWord64 . fromInteger)
, integer_to_lit "Integer -> Int64# (wrap)" integerToInt64Name (\_ -> mkInt64LitInt64 . fromInteger)
, integer_to_lit "Integer -> Float#" integerToFloatName (\_ -> mkFloatLitFloat . fromInteger)
, integer_to_lit "Integer -> Double#" integerToDoubleName (\_ -> mkDoubleLitDouble . fromInteger)
, integer_to_natural "Integer -> Natural (clamp)" integerToNaturalClampName False True
, integer_to_natural "Integer -> Natural (wrap)" integerToNaturalName False False
, integer_to_natural "Integer -> Natural (throw)" integerToNaturalThrowName True False
, lit_to_natural "Word# -> Natural" naturalNSName
, natural_to_word "Natural -> Word# (wrap)" naturalToWordName False
, natural_to_word "Natural -> Word# (clamp)" naturalToWordClampName True
-- comparisons (return an unlifted Int#)
, integer_cmp "integerEq#" integerEqName (==)
, integer_cmp "integerNe#" integerNeName (/=)
, integer_cmp "integerLe#" integerLeName (<=)
, integer_cmp "integerGt#" integerGtName (>)
, integer_cmp "integerLt#" integerLtName (<)
, integer_cmp "integerGe#" integerGeName (>=)
, natural_cmp "naturalEq#" naturalEqName (==)
, natural_cmp "naturalNe#" naturalNeName (/=)
, natural_cmp "naturalLe#" naturalLeName (<=)
, natural_cmp "naturalGt#" naturalGtName (>)
, natural_cmp "naturalLt#" naturalLtName (<)
, natural_cmp "naturalGe#" naturalGeName (>=)
-- comparisons (return an Ordering)
, bignum_compare "integerCompare" integerCompareName
, bignum_compare "naturalCompare" naturalCompareName
-- binary operations
, integer_binop "integerAdd" integerAddName (+)
, integer_binop "integerSub" integerSubName (-)
, integer_binop "integerMul" integerMulName (*)
, integer_binop "integerGcd" integerGcdName gcd
, integer_binop "integerLcm" integerLcmName lcm
, integer_binop "integerAnd" integerAndName (.&.)
, integer_binop "integerOr" integerOrName (.|.)
, integer_binop "integerXor" integerXorName xor
, natural_binop "naturalAdd" naturalAddName (+)
, natural_binop "naturalMul" naturalMulName (*)
, natural_binop "naturalGcd" naturalGcdName gcd
, natural_binop "naturalLcm" naturalLcmName lcm
, natural_binop "naturalAnd" naturalAndName (.&.)
, natural_binop "naturalOr" naturalOrName (.|.)
, natural_binop "naturalXor" naturalXorName xor
-- Natural subtraction: it's a binop but it can fail because of underflow so
-- we have several primitives to handle here.
, natural_sub "naturalSubUnsafe" naturalSubUnsafeName
, natural_sub "naturalSubThrow" naturalSubThrowName
, mkRule "naturalSub" naturalSubName 2 $ do
[a0,a1] <- getArgs
x <- isNaturalLiteral a0
y <- isNaturalLiteral a1
-- return an unboxed sum: (# (# #) | Natural #)
let ret n v = pure $ mkCoreUbxSum 2 n [voidPrimTy,naturalTy] v
if x < y
then ret 1 $ Var voidPrimId
else ret 2 $ Lit (mkLitNatural (x - y))
-- unary operations
, bignum_unop "integerNegate" integerNegateName mkLitInteger negate
, bignum_unop "integerAbs" integerAbsName mkLitInteger abs
, bignum_unop "integerSignum" integerSignumName mkLitInteger signum
, bignum_unop "integerComplement" integerComplementName mkLitInteger complement
, bignum_unop "naturalSignum" naturalSignumName mkLitNatural signum
, mkRule "naturalNegate" naturalNegateName 1 $ do
[a0] <- getArgs
x <- isNaturalLiteral a0
guard (x == 0) -- negate is only valid for (0 :: Natural)
pure a0
, bignum_popcount "integerPopCount" integerPopCountName mkLitIntWrap
, bignum_popcount "naturalPopCount" naturalPopCountName mkLitWordWrap
-- identity passthrough
, id_passthrough "Int# -> Integer -> Int#" integerToIntName integerISName
, id_passthrough "Word# -> Integer -> Word#" integerToWordName integerFromWordName
, id_passthrough "Int64# -> Integer -> Int64#" integerToInt64Name integerFromInt64Name
, id_passthrough "Word64# -> Integer -> Word64#" integerToWord64Name integerFromWord64Name
, id_passthrough "Word# -> Natural -> Word#" naturalToWordName naturalNSName
-- identity passthrough with a conversion that can be done directly instead
, small_passthrough "Int# -> Integer -> Word#"
integerISName integerToWordName (mkPrimOpId Int2WordOp)
, small_passthrough "Int# -> Integer -> Float#"
integerISName integerToFloatName (mkPrimOpId Int2FloatOp)
, small_passthrough "Int# -> Integer -> Double#"
integerISName integerToDoubleName (mkPrimOpId Int2DoubleOp)
, small_passthrough "Word# -> Natural -> Int#"
naturalNSName naturalToWordName (mkPrimOpId Word2IntOp)
-- Bits.bit
, bignum_bit "integerBit" integerBitName mkLitInteger
, bignum_bit "naturalBit" naturalBitName mkLitNatural
-- Bits.testBit
, bignum_testbit "integerTestBit" integerTestBitName
, bignum_testbit "naturalTestBit" naturalTestBitName
-- Bits.shift
, bignum_shift "integerShiftL" integerShiftLName shiftL mkLitInteger
, bignum_shift "integerShiftR" integerShiftRName shiftR mkLitInteger
, bignum_shift "naturalShiftL" naturalShiftLName shiftL mkLitNatural
, bignum_shift "naturalShiftR" naturalShiftRName shiftR mkLitNatural
-- division
, divop_one "integerQuot" integerQuotName quot mkLitInteger
, divop_one "integerRem" integerRemName rem mkLitInteger
, divop_one "integerDiv" integerDivName div mkLitInteger
, divop_one "integerMod" integerModName mod mkLitInteger
, divop_both "integerDivMod" integerDivModName divMod mkLitInteger integerTy
, divop_both "integerQuotRem" integerQuotRemName quotRem mkLitInteger integerTy
, divop_one "naturalQuot" naturalQuotName quot mkLitNatural
, divop_one "naturalRem" naturalRemName rem mkLitNatural
, divop_both "naturalQuotRem" naturalQuotRemName quotRem mkLitNatural naturalTy
-- conversions from Rational for Float/Double literals
, rational_to "rationalToFloat" rationalToFloatName mkFloatExpr
, rational_to "rationalToDouble" rationalToDoubleName mkDoubleExpr
-- conversions from Integer for Float/Double literals
, integer_encode_float "integerEncodeFloat" integerEncodeFloatName mkFloatLitFloat
, integer_encode_float "integerEncodeDouble" integerEncodeDoubleName mkDoubleLitDouble
]
where
-- The rule is matching against an occurrence of a data constructor in a
-- Core expression. It must match either its worker name or its wrapper
-- name, /not/ the DataCon name itself, which is different.
-- See Note [Data Constructor Naming] in GHC.Core.DataCon and #19892
--
-- But data constructor wrappers deliberately inline late; See Note
-- [Activation for data constructor wrappers] in GHC.Types.Id.Make.
-- Suppose there is a wrapper and the rule matches on the worker: the
-- wrapper won't be inlined until rules have finished firing and the rule
-- will never fire.
--
-- Hence the rule must match on the wrapper, if there is one, otherwise on
-- the worker. That is exactly the dataConWrapId for the data constructor.
-- The data constructor may or may not have a wrapper, but if not
-- dataConWrapId will return the worker
--
integerISName = idName (dataConWrapId integerISDataCon)
naturalNSName = idName (dataConWrapId naturalNSDataCon)
mkRule str name nargs f = BuiltinRule
{ ru_name = fsLit str
, ru_fn = name
, ru_nargs = nargs
, ru_try = runRuleM f
}
integer_to_lit str name convert = mkRule str name 1 $ do
[a0] <- getArgs
platform <- getPlatform
x <- isIntegerLiteral a0
pure (convert platform x)
natural_to_word str name clamp = mkRule str name 1 $ do
[a0] <- getArgs
n <- isNaturalLiteral a0
platform <- getPlatform
if clamp && not (platformInWordRange platform n)
then pure (Lit (mkLitWord platform (platformMaxWord platform)))
else pure (Lit (mkLitWordWrap platform n))
integer_to_natural str name thrw clamp = mkRule str name 1 $ do
[a0] <- getArgs
x <- isIntegerLiteral a0
if | x >= 0 -> pure $ Lit $ mkLitNatural x
| thrw -> mzero
| clamp -> pure $ Lit $ mkLitNatural 0 -- clamp to 0
| otherwise -> pure $ Lit $ mkLitNatural (abs x) -- negate/wrap
lit_to_integer str name = mkRule str name 1 $ do
[a0] <- getArgs
isLiteral a0 >>= \case
-- convert any numeric literal into an Integer literal
LitNumber _ i -> pure (Lit (mkLitInteger i))
_ -> mzero
lit_to_natural str name = mkRule str name 1 $ do
[a0] <- getArgs
isLiteral a0 >>= \case
-- convert any *positive* numeric literal into a Natural literal
LitNumber _ i | i >= 0 -> pure (Lit (mkLitNatural i))
_ -> mzero
integer_binop str name op = mkRule str name 2 $ do
[a0,a1] <- getArgs
x <- isIntegerLiteral a0
y <- isIntegerLiteral a1
pure (Lit (mkLitInteger (x `op` y)))
natural_binop str name op = mkRule str name 2 $ do
[a0,a1] <- getArgs
x <- isNaturalLiteral a0
y <- isNaturalLiteral a1
pure (Lit (mkLitNatural (x `op` y)))
natural_sub str name = mkRule str name 2 $ do
[a0,a1] <- getArgs
x <- isNaturalLiteral a0
y <- isNaturalLiteral a1
guard (x >= y)
pure (Lit (mkLitNatural (x - y)))
integer_cmp str name op = mkRule str name 2 $ do
platform <- getPlatform
[a0,a1] <- getArgs
x <- isIntegerLiteral a0
y <- isIntegerLiteral a1
pure $ if x `op` y
then trueValInt platform
else falseValInt platform
natural_cmp str name op = mkRule str name 2 $ do
platform <- getPlatform
[a0,a1] <- getArgs
x <- isNaturalLiteral a0
y <- isNaturalLiteral a1
pure $ if x `op` y
then trueValInt platform
else falseValInt platform
bignum_compare str name = mkRule str name 2 $ do
[a0,a1] <- getArgs
x <- isNumberLiteral a0
y <- isNumberLiteral a1
pure $ case x `compare` y of
LT -> ltVal
EQ -> eqVal
GT -> gtVal
bignum_unop str name mk_lit op = mkRule str name 1 $ do
[a0] <- getArgs
x <- isNumberLiteral a0
pure $ Lit (mk_lit (op x))
bignum_popcount str name mk_lit = mkRule str name 1 $ do
platform <- getPlatform
-- We use a host Int to compute the popCount. If we compile on a 32-bit
-- host for a 64-bit target, the result may be different than if computed
-- by the target. So we disable this rule if sizes don't match.
guard (platformWordSizeInBits platform == finiteBitSize (0 :: Word))
[a0] <- getArgs
x <- isNumberLiteral a0
pure $ Lit (mk_lit platform (fromIntegral (popCount x)))
id_passthrough str to_x from_x = mkRule str to_x 1 $ do
[App (Var f) x] <- getArgs
guard (idName f == from_x)
pure x
small_passthrough str from_x to_y x_to_y = mkRule str to_y 1 $ do
[App (Var f) x] <- getArgs
guard (idName f == from_x)
pure $ App (Var x_to_y) x
bignum_bit str name mk_lit = mkRule str name 1 $ do
[a0] <- getArgs
platform <- getPlatform
n <- isNumberLiteral a0
-- Make sure n is positive and small enough to yield a decently
-- small number. Attempting to construct the Integer for
-- (integerBit 9223372036854775807#)
-- would be a bad idea (#14959)
guard (n >= 0 && n <= fromIntegral (platformWordSizeInBits platform))
-- it's safe to convert a target Int value into a host Int value
-- to perform the "bit" operation because n is very small (<= 64).
pure $ Lit (mk_lit (bit (fromIntegral n)))
bignum_testbit str name = mkRule str name 2 $ do
[a0,a1] <- getArgs
platform <- getPlatform
x <- isNumberLiteral a0
n <- isNumberLiteral a1
-- ensure that we can store 'n' in a host Int
guard (n >= 0 && n <= fromIntegral (maxBound :: Int))
pure $ if testBit x (fromIntegral n)
then trueValInt platform
else falseValInt platform
bignum_shift str name shift_op mk_lit = mkRule str name 2 $ do
[a0,a1] <- getArgs
x <- isNumberLiteral a0
n <- isWordLiteral a1
-- See Note [Guarding against silly shifts]
-- Restrict constant-folding of shifts on Integers, somewhat arbitrary.
-- We can get huge shifts in inaccessible code (#15673)
guard (n <= 4)
pure $ Lit (mk_lit (x `shift_op` fromIntegral n))
divop_one str name divop mk_lit = mkRule str name 2 $ do
[a0,a1] <- getArgs
n <- isNumberLiteral a0
d <- isNumberLiteral a1
guard (d /= 0)
pure $ Lit (mk_lit (n `divop` d))
divop_both str name divop mk_lit ty = mkRule str name 2 $ do
[a0,a1] <- getArgs
n <- isNumberLiteral a0
d <- isNumberLiteral a1
guard (d /= 0)
let (r,s) = n `divop` d
pure $ mkCoreUbxTup [ty,ty] [Lit (mk_lit r), Lit (mk_lit s)]
integer_encode_float :: RealFloat a => String -> Name -> (a -> CoreExpr) -> CoreRule
integer_encode_float str name mk_lit = mkRule str name 2 $ do
[a0,a1] <- getArgs
x <- isIntegerLiteral a0
y <- isIntLiteral a1
-- check that y (a target Int) is in the host Int range
guard (y <= fromIntegral (maxBound :: Int))
pure (mk_lit $ encodeFloat x (fromInteger y))
rational_to :: RealFloat a => String -> Name -> (a -> CoreExpr) -> CoreRule
rational_to str name mk_lit = mkRule str name 2 $ do
-- This turns `rationalToFloat n d` where `n` and `d` are literals into
-- a literal Float (and similarly for Double).
[a0,a1] <- getArgs
n <- isIntegerLiteral a0
d <- isIntegerLiteral a1
-- it's important to not match d == 0, because that may represent a
-- literal "0/0" or similar, and we can't produce a literal value for
-- NaN or +-Inf
guard (d /= 0)
pure $ mk_lit (fromRational (n % d))
---------------------------------------------------
-- The rule is this:
-- unpackFoldrCString*# "foo"# c (unpackFoldrCString*# "baz"# c n)
-- = unpackFoldrCString*# "foobaz"# c n
--
-- See also Note [String literals in GHC] in CString.hs
-- CString version
match_append_lit_C :: RuleFun
match_append_lit_C = match_append_lit unpackCStringFoldrIdKey
-- CStringUTF8 version
match_append_lit_utf8 :: RuleFun
match_append_lit_utf8 = match_append_lit unpackCStringFoldrUtf8IdKey
{-# INLINE match_append_lit #-}
match_append_lit :: Unique -> RuleFun
match_append_lit foldVariant _ id_unf _
[ Type ty1
, lit1
, c1
, e2
]
-- N.B. Ensure that we strip off any ticks (e.g. source notes) from the
-- `lit` and `c` arguments, lest this may fail to fire when building with
-- -g3. See #16740.
| (strTicks, Var unpk `App` Type ty2
`App` lit2
`App` c2
`App` n) <- stripTicksTop tickishFloatable e2
, unpk `hasKey` foldVariant
, Just (LitString s1) <- exprIsLiteral_maybe id_unf lit1
, Just (LitString s2) <- exprIsLiteral_maybe id_unf lit2
, let freeVars = (mkInScopeSet (exprFreeVars c1 `unionVarSet` exprFreeVars c2))
in eqExpr freeVars c1 c2
, (c1Ticks, c1') <- stripTicksTop tickishFloatable c1
, c2Ticks <- stripTicksTopT tickishFloatable c2
= ASSERT( ty1 `eqType` ty2 )
Just $ mkTicks strTicks
$ Var unpk `App` Type ty1
`App` Lit (LitString (s1 `BS.append` s2))
`App` mkTicks (c1Ticks ++ c2Ticks) c1'
`App` n
match_append_lit _ _ _ _ _ = Nothing
---------------------------------------------------
-- The rule is this:
-- eqString (unpackCString# (Lit s1)) (unpackCString# (Lit s2)) = s1==s2
-- Also matches unpackCStringUtf8#
match_eq_string :: RuleFun
match_eq_string _ id_unf _
[Var unpk1 `App` lit1, Var unpk2 `App` lit2]
| unpk_key1 <- getUnique unpk1
, unpk_key2 <- getUnique unpk2
, unpk_key1 == unpk_key2
-- For now we insist the literals have to agree in their encoding
-- to keep the rule simple. But we could check if the decoded strings
-- compare equal in here as well.
, unpk_key1 `elem` [unpackCStringUtf8IdKey, unpackCStringIdKey]
, Just (LitString s1) <- exprIsLiteral_maybe id_unf lit1
, Just (LitString s2) <- exprIsLiteral_maybe id_unf lit2
= Just (if s1 == s2 then trueValBool else falseValBool)
match_eq_string _ _ _ _ = Nothing
-----------------------------------------------------------------------
-- Illustration of this rule:
--
-- cstringLength# "foobar"# --> 6
-- cstringLength# "fizz\NULzz"# --> 4
--
-- Nota bene: Addr# literals are suffixed by a NUL byte when they are
-- compiled to read-only data sections. That's why cstringLength# is
-- well defined on Addr# literals that do not explicitly have an embedded
-- NUL byte.
--
-- See GHC issue #5218, MR 2165, and bytestring PR 191. This is particularly
-- helpful when using OverloadedStrings to create a ByteString since the
-- function computing the length of such ByteStrings can often be constant
-- folded.
match_cstring_length :: RuleFun
match_cstring_length env id_unf _ [lit1]
| Just (LitString str) <- exprIsLiteral_maybe id_unf lit1
-- If elemIndex returns Just, it has the index of the first embedded NUL
-- in the string. If no NUL bytes are present (the common case) then use
-- full length of the byte string.
= let len = fromMaybe (BS.length str) (BS.elemIndex 0 str)
in Just (Lit (mkLitInt (roPlatform env) (fromIntegral len)))
match_cstring_length _ _ _ _ = Nothing
---------------------------------------------------
{- Note [inlineId magic]
~~~~~~~~~~~~~~~~~~~~~~~~
The call 'inline f' arranges that 'f' is inlined, regardless of
its size. More precisely, the call 'inline f' rewrites to the
right-hand side of 'f's definition. This allows the programmer to
control inlining from a particular call site rather than the
definition site of the function.
The moving parts are simple:
* A very simple definition in the library base:GHC.Magic
{-# NOINLINE[0] inline #-}
inline :: a -> a
inline x = x
So in phase 0, 'inline' will be inlined, so its use imposes
no overhead.
* A rewrite rule, in GHC.Core.Opt.ConstantFold, which makes
(inline f) inline, implemented by match_inline.
The rule for the 'inline' function is this:
inline f_ty (f a b c) = <f's unfolding> a b c
(if f has an unfolding, EVEN if it's a loop breaker)
It's important to allow the argument to 'inline' to have args itself
(a) because its more forgiving to allow the programmer to write
either inline f a b c
or inline (f a b c)
(b) because a polymorphic f wll get a type argument that the
programmer can't avoid, so the call may look like
inline (map @Int @Bool) g xs
Also, don't forget about 'inline's type argument!
-}
match_inline :: [Expr CoreBndr] -> Maybe (Expr CoreBndr)
match_inline (Type _ : e : _)
| (Var f, args1) <- collectArgs e,
Just unf <- maybeUnfoldingTemplate (realIdUnfolding f)
-- Ignore the IdUnfoldingFun here!
= Just (mkApps unf args1)
match_inline _ = Nothing
---------------------------------------------------
-- See Note [magicDictId magic] in "GHC.Types.Id.Make"
-- for a description of what is going on here.
match_magicDict :: [Expr CoreBndr] -> Maybe (Expr CoreBndr)
match_magicDict [Type _, (stripTicksE (const True) -> (Var wrap `App` Type a `App` Type _ `App` f)), x, y ]
| Just (_, fieldTy, _) <- splitFunTy_maybe $ dropForAlls $ idType wrap
, Just (_, dictTy, _) <- splitFunTy_maybe fieldTy
, Just dictTc <- tyConAppTyCon_maybe dictTy
, Just (_,_,co) <- unwrapNewTyCon_maybe dictTc
= Just
$ f `App` Cast x (mkSymCo (mkUnbranchedAxInstCo Representational co [a] []))
`App` y
match_magicDict _ = Nothing
--------------------------------------------------------
-- Note [Constant folding through nested expressions]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- We use rewrites rules to perform constant folding. It means that we don't
-- have a global view of the expression we are trying to optimise. As a
-- consequence we only perform local (small-step) transformations that either:
-- 1) reduce the number of operations
-- 2) rearrange the expression to increase the odds that other rules will
-- match
--
-- We don't try to handle more complex expression optimisation cases that would
-- require a global view. For example, rewriting expressions to increase
-- sharing (e.g., Horner's method); optimisations that require local
-- transformations increasing the number of operations; rearrangements to
-- cancel/factorize terms (e.g., (a+b-a-b) isn't rearranged to reduce to 0).
--
-- We already have rules to perform constant folding on expressions with the
-- following shape (where a and/or b are literals):
--
-- D) op
-- /\
-- / \
-- / \
-- a b
--
-- To support nested expressions, we match three other shapes of expression
-- trees:
--
-- A) op1 B) op1 C) op1
-- /\ /\ /\
-- / \ / \ / \
-- / \ / \ / \
-- a op2 op2 c op2 op3
-- /\ /\ /\ /\
-- / \ / \ / \ / \
-- b c a b a b c d
--
--
-- R1) +/- simplification:
-- ops = + or -, two literals (not siblings)
--
-- Examples:
-- A: 5 + (10-x) ==> 15-x
-- B: (10+x) + 5 ==> 15+x
-- C: (5+a)-(5-b) ==> 0+(a+b)
--
-- R2) * simplification
-- ops = *, two literals (not siblings)
--
-- Examples:
-- A: 5 * (10*x) ==> 50*x
-- B: (10*x) * 5 ==> 50*x
-- C: (5*a)*(5*b) ==> 25*(a*b)
--
-- R3) * distribution over +/-
-- op1 = *, op2 = + or -, two literals (not siblings)
--
-- This transformation doesn't reduce the number of operations but switches
-- the outer and the inner operations so that the outer is (+) or (-) instead
-- of (*). It increases the odds that other rules will match after this one.
--
-- Examples:
-- A: 5 * (10-x) ==> 50 - (5*x)
-- B: (10+x) * 5 ==> 50 + (5*x)
-- C: Not supported as it would increase the number of operations:
-- (5+a)*(5-b) ==> 25 - 5*b + 5*a - a*b
--
-- R4) Simple factorization
--
-- op1 = + or -, op2/op3 = *,
-- one literal for each innermost * operation (except in the D case),
-- the two other terms are equals
--
-- Examples:
-- A: x - (10*x) ==> (-9)*x
-- B: (10*x) + x ==> 11*x
-- C: (5*x)-(x*3) ==> 2*x
-- D: x+x ==> 2*x
--
-- R5) +/- propagation
--
-- ops = + or -, one literal
--
-- This transformation doesn't reduce the number of operations but propagates
-- the constant to the outer level. It increases the odds that other rules
-- will match after this one.
--
-- Examples:
-- A: x - (10-y) ==> (x+y) - 10
-- B: (10+x) - y ==> 10 + (x-y)
-- C: N/A (caught by the A and B cases)
--
--------------------------------------------------------
-- | Rules to perform constant folding into nested expressions
--
--See Note [Constant folding through nested expressions]
numFoldingRules :: PrimOp -> (Platform -> PrimOps) -> RuleM CoreExpr
numFoldingRules op dict = do
env <- getEnv
if not (roNumConstantFolding env)
then mzero
else do
[e1,e2] <- getArgs
platform <- getPlatform
let PrimOps{..} = dict platform
case BinOpApp e1 op e2 of
-- R1) +/- simplification
x :++: (y :++: v) -> return $ mkL (x+y) `add` v
x :++: (L y :-: v) -> return $ mkL (x+y) `sub` v
x :++: (v :-: L y) -> return $ mkL (x-y) `add` v
L x :-: (y :++: v) -> return $ mkL (x-y) `sub` v
L x :-: (L y :-: v) -> return $ mkL (x-y) `add` v
L x :-: (v :-: L y) -> return $ mkL (x+y) `sub` v
(y :++: v) :-: L x -> return $ mkL (y-x) `add` v
(L y :-: v) :-: L x -> return $ mkL (y-x) `sub` v
(v :-: L y) :-: L x -> return $ mkL (0-y-x) `add` v
(x :++: w) :+: (y :++: v) -> return $ mkL (x+y) `add` (w `add` v)
(w :-: L x) :+: (L y :-: v) -> return $ mkL (y-x) `add` (w `sub` v)
(w :-: L x) :+: (v :-: L y) -> return $ mkL (0-x-y) `add` (w `add` v)
(L x :-: w) :+: (L y :-: v) -> return $ mkL (x+y) `sub` (w `add` v)
(L x :-: w) :+: (v :-: L y) -> return $ mkL (x-y) `add` (v `sub` w)
(w :-: L x) :+: (y :++: v) -> return $ mkL (y-x) `add` (w `add` v)
(L x :-: w) :+: (y :++: v) -> return $ mkL (x+y) `add` (v `sub` w)
(y :++: v) :+: (w :-: L x) -> return $ mkL (y-x) `add` (w `add` v)
(y :++: v) :+: (L x :-: w) -> return $ mkL (x+y) `add` (v `sub` w)
(v :-: L y) :-: (w :-: L x) -> return $ mkL (x-y) `add` (v `sub` w)
(v :-: L y) :-: (L x :-: w) -> return $ mkL (0-x-y) `add` (v `add` w)
(L y :-: v) :-: (w :-: L x) -> return $ mkL (x+y) `sub` (v `add` w)
(L y :-: v) :-: (L x :-: w) -> return $ mkL (y-x) `add` (w `sub` v)
(x :++: w) :-: (y :++: v) -> return $ mkL (x-y) `add` (w `sub` v)
(w :-: L x) :-: (y :++: v) -> return $ mkL (0-y-x) `add` (w `sub` v)
(L x :-: w) :-: (y :++: v) -> return $ mkL (x-y) `sub` (v `add` w)
(y :++: v) :-: (w :-: L x) -> return $ mkL (y+x) `add` (v `sub` w)
(y :++: v) :-: (L x :-: w) -> return $ mkL (y-x) `add` (v `add` w)
-- R2) * simplification
x :**: (y :**: v) -> return $ mkL (x*y) `mul` v
(x :**: w) :*: (y :**: v) -> return $ mkL (x*y) `mul` (w `mul` v)
-- R3) * distribution over +/-
x :**: (y :++: v) -> return $ mkL (x*y) `add` (mkL x `mul` v)
x :**: (L y :-: v) -> return $ mkL (x*y) `sub` (mkL x `mul` v)
x :**: (v :-: L y) -> return $ (mkL x `mul` v) `sub` mkL (x*y)
-- R4) Simple factorization
v :+: w
| w `cheapEqExpr` v -> return $ mkL 2 `mul` v
w :+: (y :**: v)
| w `cheapEqExpr` v -> return $ mkL (1+y) `mul` v
w :-: (y :**: v)
| w `cheapEqExpr` v -> return $ mkL (1-y) `mul` v
(y :**: v) :+: w
| w `cheapEqExpr` v -> return $ mkL (y+1) `mul` v
(y :**: v) :-: w
| w `cheapEqExpr` v -> return $ mkL (y-1) `mul` v
(x :**: w) :+: (y :**: v)
| w `cheapEqExpr` v -> return $ mkL (x+y) `mul` v
(x :**: w) :-: (y :**: v)
| w `cheapEqExpr` v -> return $ mkL (x-y) `mul` v
-- R5) +/- propagation
w :+: (y :++: v) -> return $ mkL y `add` (w `add` v)
(y :++: v) :+: w -> return $ mkL y `add` (w `add` v)
w :-: (y :++: v) -> return $ (w `sub` v) `sub` mkL y
(y :++: v) :-: w -> return $ mkL y `add` (v `sub` w)
w :-: (L y :-: v) -> return $ (w `add` v) `sub` mkL y
(L y :-: v) :-: w -> return $ mkL y `sub` (w `add` v)
w :+: (L y :-: v) -> return $ mkL y `add` (w `sub` v)
w :+: (v :-: L y) -> return $ (w `add` v) `sub` mkL y
(L y :-: v) :+: w -> return $ mkL y `add` (w `sub` v)
(v :-: L y) :+: w -> return $ (w `add` v) `sub` mkL y
_ -> mzero
-- | Match the application of a binary primop
pattern BinOpApp :: Arg CoreBndr -> PrimOp -> Arg CoreBndr -> CoreExpr
pattern BinOpApp x op y = OpVal op `App` x `App` y
-- | Match a primop
pattern OpVal :: PrimOp -> Arg CoreBndr
pattern OpVal op <- Var (isPrimOpId_maybe -> Just op) where
OpVal op = Var (mkPrimOpId op)
-- | Match a literal
pattern L :: Integer -> Arg CoreBndr
pattern L l <- Lit (isLitValue_maybe -> Just l)
-- | Match an addition
pattern (:+:) :: Arg CoreBndr -> Arg CoreBndr -> CoreExpr
pattern x :+: y <- BinOpApp x (isAddOp -> True) y
-- | Match an addition with a literal (handle commutativity)
pattern (:++:) :: Integer -> Arg CoreBndr -> CoreExpr
pattern l :++: x <- (isAdd -> Just (l,x))
isAdd :: CoreExpr -> Maybe (Integer,CoreExpr)
isAdd e = case e of
L l :+: x -> Just (l,x)
x :+: L l -> Just (l,x)
_ -> Nothing
-- | Match a multiplication
pattern (:*:) :: Arg CoreBndr -> Arg CoreBndr -> CoreExpr
pattern x :*: y <- BinOpApp x (isMulOp -> True) y
-- | Match a multiplication with a literal (handle commutativity)
pattern (:**:) :: Integer -> Arg CoreBndr -> CoreExpr
pattern l :**: x <- (isMul -> Just (l,x))
isMul :: CoreExpr -> Maybe (Integer,CoreExpr)
isMul e = case e of
L l :*: x -> Just (l,x)
x :*: L l -> Just (l,x)
_ -> Nothing
-- | Match a subtraction
pattern (:-:) :: Arg CoreBndr -> Arg CoreBndr -> CoreExpr
pattern x :-: y <- BinOpApp x (isSubOp -> True) y
isSubOp :: PrimOp -> Bool
isSubOp IntSubOp = True
isSubOp WordSubOp = True
isSubOp _ = False
isAddOp :: PrimOp -> Bool
isAddOp IntAddOp = True
isAddOp WordAddOp = True
isAddOp _ = False
isMulOp :: PrimOp -> Bool
isMulOp IntMulOp = True
isMulOp WordMulOp = True
isMulOp _ = False
-- | Explicit "type-class"-like dictionary for numeric primops
--
-- Depends on Platform because creating a literal value depends on Platform
data PrimOps = PrimOps
{ add :: CoreExpr -> CoreExpr -> CoreExpr -- ^ Add two numbers
, sub :: CoreExpr -> CoreExpr -> CoreExpr -- ^ Sub two numbers
, mul :: CoreExpr -> CoreExpr -> CoreExpr -- ^ Multiply two numbers
, mkL :: Integer -> CoreExpr -- ^ Create a literal value
}
intPrimOps :: Platform -> PrimOps
intPrimOps platform = PrimOps
{ add = \x y -> BinOpApp x IntAddOp y
, sub = \x y -> BinOpApp x IntSubOp y
, mul = \x y -> BinOpApp x IntMulOp y
, mkL = intResult' platform
}
wordPrimOps :: Platform -> PrimOps
wordPrimOps platform = PrimOps
{ add = \x y -> BinOpApp x WordAddOp y
, sub = \x y -> BinOpApp x WordSubOp y
, mul = \x y -> BinOpApp x WordMulOp y
, mkL = wordResult' platform
}
--------------------------------------------------------
-- Constant folding through case-expressions
--
-- cf Scrutinee Constant Folding in simplCore/GHC.Core.Opt.Simplify.Utils
--------------------------------------------------------
-- | Match the scrutinee of a case and potentially return a new scrutinee and a
-- function to apply to each literal alternative.
caseRules :: Platform
-> CoreExpr -- Scrutinee
-> Maybe ( CoreExpr -- New scrutinee
, AltCon -> Maybe AltCon -- How to fix up the alt pattern
-- Nothing <=> Unreachable
-- See Note [Unreachable caseRules alternatives]
, Id -> CoreExpr) -- How to reconstruct the original scrutinee
-- from the new case-binder
-- e.g case e of b {
-- ...;
-- con bs -> rhs;
-- ... }
-- ==>
-- case e' of b' {
-- ...;
-- fixup_altcon[con] bs -> let b = mk_orig[b] in rhs;
-- ... }
caseRules platform (App (App (Var f) v) (Lit l)) -- v `op` x#
| Just op <- isPrimOpId_maybe f
, Just x <- isLitValue_maybe l
, Just adjust_lit <- adjustDyadicRight op x
= Just (v, tx_lit_con platform adjust_lit
, \v -> (App (App (Var f) (Var v)) (Lit l)))
caseRules platform (App (App (Var f) (Lit l)) v) -- x# `op` v
| Just op <- isPrimOpId_maybe f
, Just x <- isLitValue_maybe l
, Just adjust_lit <- adjustDyadicLeft x op
= Just (v, tx_lit_con platform adjust_lit
, \v -> (App (App (Var f) (Lit l)) (Var v)))
caseRules platform (App (Var f) v ) -- op v
| Just op <- isPrimOpId_maybe f
, Just adjust_lit <- adjustUnary op
= Just (v, tx_lit_con platform adjust_lit
, \v -> App (Var f) (Var v))
-- See Note [caseRules for tagToEnum]
caseRules platform (App (App (Var f) type_arg) v)
| Just TagToEnumOp <- isPrimOpId_maybe f
= Just (v, tx_con_tte platform
, \v -> (App (App (Var f) type_arg) (Var v)))
-- See Note [caseRules for dataToTag]
caseRules _ (App (App (Var f) (Type ty)) v) -- dataToTag x
| Just DataToTagOp <- isPrimOpId_maybe f
, Just (tc, _) <- tcSplitTyConApp_maybe ty
, isAlgTyCon tc
= Just (v, tx_con_dtt ty
, \v -> App (App (Var f) (Type ty)) (Var v))
caseRules _ _ = Nothing
tx_lit_con :: Platform -> (Integer -> Integer) -> AltCon -> Maybe AltCon
tx_lit_con _ _ DEFAULT = Just DEFAULT
tx_lit_con platform adjust (LitAlt l) = Just $ LitAlt (mapLitValue platform adjust l)
tx_lit_con _ _ alt = pprPanic "caseRules" (ppr alt)
-- NB: mapLitValue uses mkLitIntWrap etc, to ensure that the
-- literal alternatives remain in Word/Int target ranges
-- (See Note [Word/Int underflow/overflow] in GHC.Types.Literal and #13172).
adjustDyadicRight :: PrimOp -> Integer -> Maybe (Integer -> Integer)
-- Given (x `op` lit) return a function 'f' s.t. f (x `op` lit) = x
adjustDyadicRight op lit
= case op of
WordAddOp -> Just (\y -> y-lit )
IntAddOp -> Just (\y -> y-lit )
WordSubOp -> Just (\y -> y+lit )
IntSubOp -> Just (\y -> y+lit )
XorOp -> Just (\y -> y `xor` lit)
XorIOp -> Just (\y -> y `xor` lit)
_ -> Nothing
adjustDyadicLeft :: Integer -> PrimOp -> Maybe (Integer -> Integer)
-- Given (lit `op` x) return a function 'f' s.t. f (lit `op` x) = x
adjustDyadicLeft lit op
= case op of
WordAddOp -> Just (\y -> y-lit )
IntAddOp -> Just (\y -> y-lit )
WordSubOp -> Just (\y -> lit-y )
IntSubOp -> Just (\y -> lit-y )
XorOp -> Just (\y -> y `xor` lit)
XorIOp -> Just (\y -> y `xor` lit)
_ -> Nothing
adjustUnary :: PrimOp -> Maybe (Integer -> Integer)
-- Given (op x) return a function 'f' s.t. f (op x) = x
adjustUnary op
= case op of
NotOp -> Just (\y -> complement y)
NotIOp -> Just (\y -> complement y)
IntNegOp -> Just (\y -> negate y )
_ -> Nothing
tx_con_tte :: Platform -> AltCon -> Maybe AltCon
tx_con_tte _ DEFAULT = Just DEFAULT
tx_con_tte _ alt@(LitAlt {}) = pprPanic "caseRules" (ppr alt)
tx_con_tte platform (DataAlt dc) -- See Note [caseRules for tagToEnum]
= Just $ LitAlt $ mkLitInt platform $ toInteger $ dataConTagZ dc
tx_con_dtt :: Type -> AltCon -> Maybe AltCon
tx_con_dtt _ DEFAULT = Just DEFAULT
tx_con_dtt ty (LitAlt (LitNumber LitNumInt i))
| tag >= 0
, tag < n_data_cons
= Just (DataAlt (data_cons !! tag)) -- tag is zero-indexed, as is (!!)
| otherwise
= Nothing
where
tag = fromInteger i :: ConTagZ
tc = tyConAppTyCon ty
n_data_cons = tyConFamilySize tc
data_cons = tyConDataCons tc
tx_con_dtt _ alt = pprPanic "caseRules" (ppr alt)
{- Note [caseRules for tagToEnum]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We want to transform
case tagToEnum x of
False -> e1
True -> e2
into
case x of
0# -> e1
1# -> e2
This rule eliminates a lot of boilerplate. For
if (x>y) then e2 else e1
we generate
case tagToEnum (x ># y) of
False -> e1
True -> e2
and it is nice to then get rid of the tagToEnum.
Beware (#14768): avoid the temptation to map constructor 0 to
DEFAULT, in the hope of getting this
case (x ># y) of
DEFAULT -> e1
1# -> e2
That fails utterly in the case of
data Colour = Red | Green | Blue
case tagToEnum x of
DEFAULT -> e1
Red -> e2
We don't want to get this!
case x of
DEFAULT -> e1
DEFAULT -> e2
Instead, we deal with turning one branch into DEFAULT in GHC.Core.Opt.Simplify.Utils
(add_default in mkCase3).
Note [caseRules for dataToTag]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
See also Note [dataToTag#] in primpops.txt.pp
We want to transform
case dataToTag x of
DEFAULT -> e1
1# -> e2
into
case x of
DEFAULT -> e1
(:) _ _ -> e2
Note the need for some wildcard binders in
the 'cons' case.
For the time, we only apply this transformation when the type of `x` is a type
headed by a normal tycon. In particular, we do not apply this in the case of a
data family tycon, since that would require carefully applying coercion(s)
between the data family and the data family instance's representation type,
which caseRules isn't currently engineered to handle (#14680).
Note [Unreachable caseRules alternatives]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Take care if we see something like
case dataToTag x of
DEFAULT -> e1
-1# -> e2
100 -> e3
because there isn't a data constructor with tag -1 or 100. In this case the
out-of-range alternative is dead code -- we know the range of tags for x.
Hence caseRules returns (AltCon -> Maybe AltCon), with Nothing indicating
an alternative that is unreachable.
You may wonder how this can happen: check out #15436.
-}
|