Spaces:
Sleeping
Sleeping
File size: 99,371 Bytes
dc2106c |
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 |
"""Provides the `CCompilerOpt` class, used for handling the CPU/hardware
optimization, starting from parsing the command arguments, to managing the
relation between the CPU baseline and dispatch-able features,
also generating the required C headers and ending with compiling
the sources with proper compiler's flags.
`CCompilerOpt` doesn't provide runtime detection for the CPU features,
instead only focuses on the compiler side, but it creates abstract C headers
that can be used later for the final runtime dispatching process."""
import sys, io, os, re, textwrap, pprint, inspect, atexit, subprocess
class _Config:
"""An abstract class holds all configurable attributes of `CCompilerOpt`,
these class attributes can be used to change the default behavior
of `CCompilerOpt` in order to fit other requirements.
Attributes
----------
conf_nocache : bool
Set True to disable memory and file cache.
Default is False.
conf_noopt : bool
Set True to forces the optimization to be disabled,
in this case `CCompilerOpt` tends to generate all
expected headers in order to 'not' break the build.
Default is False.
conf_cache_factors : list
Add extra factors to the primary caching factors. The caching factors
are utilized to determine if there are changes had happened that
requires to discard the cache and re-updating it. The primary factors
are the arguments of `CCompilerOpt` and `CCompiler`'s properties(type, flags, etc).
Default is list of two items, containing the time of last modification
of `ccompiler_opt` and value of attribute "conf_noopt"
conf_tmp_path : str,
The path of temporary directory. Default is auto-created
temporary directory via ``tempfile.mkdtemp()``.
conf_check_path : str
The path of testing files. Each added CPU feature must have a
**C** source file contains at least one intrinsic or instruction that
related to this feature, so it can be tested against the compiler.
Default is ``./distutils/checks``.
conf_target_groups : dict
Extra tokens that can be reached from dispatch-able sources through
the special mark ``@targets``. Default is an empty dictionary.
**Notes**:
- case-insensitive for tokens and group names
- sign '#' must stick in the begin of group name and only within ``@targets``
**Example**:
.. code-block:: console
$ "@targets #avx_group other_tokens" > group_inside.c
>>> CCompilerOpt.conf_target_groups["avx_group"] = \\
"$werror $maxopt avx2 avx512f avx512_skx"
>>> cco = CCompilerOpt(cc_instance)
>>> cco.try_dispatch(["group_inside.c"])
conf_c_prefix : str
The prefix of public C definitions. Default is ``"NPY_"``.
conf_c_prefix_ : str
The prefix of internal C definitions. Default is ``"NPY__"``.
conf_cc_flags : dict
Nested dictionaries defining several compiler flags
that linked to some major functions, the main key
represent the compiler name and sub-keys represent
flags names. Default is already covers all supported
**C** compilers.
Sub-keys explained as follows:
"native": str or None
used by argument option `native`, to detect the current
machine support via the compiler.
"werror": str or None
utilized to treat warning as errors during testing CPU features
against the compiler and also for target's policy `$werror`
via dispatch-able sources.
"maxopt": str or None
utilized for target's policy '$maxopt' and the value should
contains the maximum acceptable optimization by the compiler.
e.g. in gcc `'-O3'`
**Notes**:
* case-sensitive for compiler names and flags
* use space to separate multiple flags
* any flag will tested against the compiler and it will skipped
if it's not applicable.
conf_min_features : dict
A dictionary defines the used CPU features for
argument option `'min'`, the key represent the CPU architecture
name e.g. `'x86'`. Default values provide the best effort
on wide range of users platforms.
**Note**: case-sensitive for architecture names.
conf_features : dict
Nested dictionaries used for identifying the CPU features.
the primary key is represented as a feature name or group name
that gathers several features. Default values covers all
supported features but without the major options like "flags",
these undefined options handle it by method `conf_features_partial()`.
Default value is covers almost all CPU features for *X86*, *IBM/Power64*
and *ARM 7/8*.
Sub-keys explained as follows:
"implies" : str or list, optional,
List of CPU feature names to be implied by it,
the feature name must be defined within `conf_features`.
Default is None.
"flags": str or list, optional
List of compiler flags. Default is None.
"detect": str or list, optional
List of CPU feature names that required to be detected
in runtime. By default, its the feature name or features
in "group" if its specified.
"implies_detect": bool, optional
If True, all "detect" of implied features will be combined.
Default is True. see `feature_detect()`.
"group": str or list, optional
Same as "implies" but doesn't require the feature name to be
defined within `conf_features`.
"interest": int, required
a key for sorting CPU features
"headers": str or list, optional
intrinsics C header file
"disable": str, optional
force disable feature, the string value should contains the
reason of disabling.
"autovec": bool or None, optional
True or False to declare that CPU feature can be auto-vectorized
by the compiler.
By default(None), treated as True if the feature contains at
least one applicable flag. see `feature_can_autovec()`
"extra_checks": str or list, optional
Extra test case names for the CPU feature that need to be tested
against the compiler.
Each test case must have a C file named ``extra_xxxx.c``, where
``xxxx`` is the case name in lower case, under 'conf_check_path'.
It should contain at least one intrinsic or function related to the test case.
If the compiler able to successfully compile the C file then `CCompilerOpt`
will add a C ``#define`` for it into the main dispatch header, e.g.
```#define {conf_c_prefix}_XXXX`` where ``XXXX`` is the case name in upper case.
**NOTES**:
* space can be used as separator with options that supports "str or list"
* case-sensitive for all values and feature name must be in upper-case.
* if flags aren't applicable, its will skipped rather than disable the
CPU feature
* the CPU feature will disabled if the compiler fail to compile
the test file
"""
conf_nocache = False
conf_noopt = False
conf_cache_factors = None
conf_tmp_path = None
conf_check_path = os.path.join(
os.path.dirname(os.path.realpath(__file__)), "checks"
)
conf_target_groups = {}
conf_c_prefix = 'NPY_'
conf_c_prefix_ = 'NPY__'
conf_cc_flags = dict(
gcc = dict(
# native should always fail on arm and ppc64,
# native usually works only with x86
native = '-march=native',
opt = '-O3',
werror = '-Werror'
),
clang = dict(
native = '-march=native',
opt = "-O3",
# One of the following flags needs to be applicable for Clang to
# guarantee the sanity of the testing process, however in certain
# cases `-Werror` gets skipped during the availability test due to
# "unused arguments" warnings.
# see https://github.com/numpy/numpy/issues/19624
werror = '-Werror-implicit-function-declaration -Werror'
),
icc = dict(
native = '-xHost',
opt = '-O3',
werror = '-Werror'
),
iccw = dict(
native = '/QxHost',
opt = '/O3',
werror = '/Werror'
),
msvc = dict(
native = None,
opt = '/O2',
werror = '/WX'
)
)
conf_min_features = dict(
x86 = "SSE SSE2",
x64 = "SSE SSE2 SSE3",
ppc64 = '', # play it safe
ppc64le = "VSX VSX2",
armhf = '', # play it safe
aarch64 = "NEON NEON_FP16 NEON_VFPV4 ASIMD"
)
conf_features = dict(
# X86
SSE = dict(
interest=1, headers="xmmintrin.h",
# enabling SSE without SSE2 is useless also
# it's non-optional for x86_64
implies="SSE2"
),
SSE2 = dict(interest=2, implies="SSE", headers="emmintrin.h"),
SSE3 = dict(interest=3, implies="SSE2", headers="pmmintrin.h"),
SSSE3 = dict(interest=4, implies="SSE3", headers="tmmintrin.h"),
SSE41 = dict(interest=5, implies="SSSE3", headers="smmintrin.h"),
POPCNT = dict(interest=6, implies="SSE41", headers="popcntintrin.h"),
SSE42 = dict(interest=7, implies="POPCNT"),
AVX = dict(
interest=8, implies="SSE42", headers="immintrin.h",
implies_detect=False
),
XOP = dict(interest=9, implies="AVX", headers="x86intrin.h"),
FMA4 = dict(interest=10, implies="AVX", headers="x86intrin.h"),
F16C = dict(interest=11, implies="AVX"),
FMA3 = dict(interest=12, implies="F16C"),
AVX2 = dict(interest=13, implies="F16C"),
AVX512F = dict(
interest=20, implies="FMA3 AVX2", implies_detect=False,
extra_checks="AVX512F_REDUCE"
),
AVX512CD = dict(interest=21, implies="AVX512F"),
AVX512_KNL = dict(
interest=40, implies="AVX512CD", group="AVX512ER AVX512PF",
detect="AVX512_KNL", implies_detect=False
),
AVX512_KNM = dict(
interest=41, implies="AVX512_KNL",
group="AVX5124FMAPS AVX5124VNNIW AVX512VPOPCNTDQ",
detect="AVX512_KNM", implies_detect=False
),
AVX512_SKX = dict(
interest=42, implies="AVX512CD", group="AVX512VL AVX512BW AVX512DQ",
detect="AVX512_SKX", implies_detect=False,
extra_checks="AVX512BW_MASK AVX512DQ_MASK"
),
AVX512_CLX = dict(
interest=43, implies="AVX512_SKX", group="AVX512VNNI",
detect="AVX512_CLX"
),
AVX512_CNL = dict(
interest=44, implies="AVX512_SKX", group="AVX512IFMA AVX512VBMI",
detect="AVX512_CNL", implies_detect=False
),
AVX512_ICL = dict(
interest=45, implies="AVX512_CLX AVX512_CNL",
group="AVX512VBMI2 AVX512BITALG AVX512VPOPCNTDQ",
detect="AVX512_ICL", implies_detect=False
),
# IBM/Power
## Power7/ISA 2.06
VSX = dict(interest=1, headers="altivec.h", extra_checks="VSX_ASM"),
## Power8/ISA 2.07
VSX2 = dict(interest=2, implies="VSX", implies_detect=False),
## Power9/ISA 3.00
VSX3 = dict(interest=3, implies="VSX2", implies_detect=False),
# ARM
NEON = dict(interest=1, headers="arm_neon.h"),
NEON_FP16 = dict(interest=2, implies="NEON"),
## FMA
NEON_VFPV4 = dict(interest=3, implies="NEON_FP16"),
## Advanced SIMD
ASIMD = dict(interest=4, implies="NEON_FP16 NEON_VFPV4", implies_detect=False),
## ARMv8.2 half-precision & vector arithm
ASIMDHP = dict(interest=5, implies="ASIMD"),
## ARMv8.2 dot product
ASIMDDP = dict(interest=6, implies="ASIMD"),
## ARMv8.2 Single & half-precision Multiply
ASIMDFHM = dict(interest=7, implies="ASIMDHP"),
)
def conf_features_partial(self):
"""Return a dictionary of supported CPU features by the platform,
and accumulate the rest of undefined options in `conf_features`,
the returned dict has same rules and notes in
class attribute `conf_features`, also its override
any options that been set in 'conf_features'.
"""
if self.cc_noopt:
# optimization is disabled
return {}
on_x86 = self.cc_on_x86 or self.cc_on_x64
is_unix = self.cc_is_gcc or self.cc_is_clang
if on_x86 and is_unix: return dict(
SSE = dict(flags="-msse"),
SSE2 = dict(flags="-msse2"),
SSE3 = dict(flags="-msse3"),
SSSE3 = dict(flags="-mssse3"),
SSE41 = dict(flags="-msse4.1"),
POPCNT = dict(flags="-mpopcnt"),
SSE42 = dict(flags="-msse4.2"),
AVX = dict(flags="-mavx"),
F16C = dict(flags="-mf16c"),
XOP = dict(flags="-mxop"),
FMA4 = dict(flags="-mfma4"),
FMA3 = dict(flags="-mfma"),
AVX2 = dict(flags="-mavx2"),
AVX512F = dict(flags="-mavx512f"),
AVX512CD = dict(flags="-mavx512cd"),
AVX512_KNL = dict(flags="-mavx512er -mavx512pf"),
AVX512_KNM = dict(
flags="-mavx5124fmaps -mavx5124vnniw -mavx512vpopcntdq"
),
AVX512_SKX = dict(flags="-mavx512vl -mavx512bw -mavx512dq"),
AVX512_CLX = dict(flags="-mavx512vnni"),
AVX512_CNL = dict(flags="-mavx512ifma -mavx512vbmi"),
AVX512_ICL = dict(
flags="-mavx512vbmi2 -mavx512bitalg -mavx512vpopcntdq"
)
)
if on_x86 and self.cc_is_icc: return dict(
SSE = dict(flags="-msse"),
SSE2 = dict(flags="-msse2"),
SSE3 = dict(flags="-msse3"),
SSSE3 = dict(flags="-mssse3"),
SSE41 = dict(flags="-msse4.1"),
POPCNT = {},
SSE42 = dict(flags="-msse4.2"),
AVX = dict(flags="-mavx"),
F16C = {},
XOP = dict(disable="Intel Compiler doesn't support it"),
FMA4 = dict(disable="Intel Compiler doesn't support it"),
# Intel Compiler doesn't support AVX2 or FMA3 independently
FMA3 = dict(
implies="F16C AVX2", flags="-march=core-avx2"
),
AVX2 = dict(implies="FMA3", flags="-march=core-avx2"),
# Intel Compiler doesn't support AVX512F or AVX512CD independently
AVX512F = dict(
implies="AVX2 AVX512CD", flags="-march=common-avx512"
),
AVX512CD = dict(
implies="AVX2 AVX512F", flags="-march=common-avx512"
),
AVX512_KNL = dict(flags="-xKNL"),
AVX512_KNM = dict(flags="-xKNM"),
AVX512_SKX = dict(flags="-xSKYLAKE-AVX512"),
AVX512_CLX = dict(flags="-xCASCADELAKE"),
AVX512_CNL = dict(flags="-xCANNONLAKE"),
AVX512_ICL = dict(flags="-xICELAKE-CLIENT"),
)
if on_x86 and self.cc_is_iccw: return dict(
SSE = dict(flags="/arch:SSE"),
SSE2 = dict(flags="/arch:SSE2"),
SSE3 = dict(flags="/arch:SSE3"),
SSSE3 = dict(flags="/arch:SSSE3"),
SSE41 = dict(flags="/arch:SSE4.1"),
POPCNT = {},
SSE42 = dict(flags="/arch:SSE4.2"),
AVX = dict(flags="/arch:AVX"),
F16C = {},
XOP = dict(disable="Intel Compiler doesn't support it"),
FMA4 = dict(disable="Intel Compiler doesn't support it"),
# Intel Compiler doesn't support FMA3 or AVX2 independently
FMA3 = dict(
implies="F16C AVX2", flags="/arch:CORE-AVX2"
),
AVX2 = dict(
implies="FMA3", flags="/arch:CORE-AVX2"
),
# Intel Compiler doesn't support AVX512F or AVX512CD independently
AVX512F = dict(
implies="AVX2 AVX512CD", flags="/Qx:COMMON-AVX512"
),
AVX512CD = dict(
implies="AVX2 AVX512F", flags="/Qx:COMMON-AVX512"
),
AVX512_KNL = dict(flags="/Qx:KNL"),
AVX512_KNM = dict(flags="/Qx:KNM"),
AVX512_SKX = dict(flags="/Qx:SKYLAKE-AVX512"),
AVX512_CLX = dict(flags="/Qx:CASCADELAKE"),
AVX512_CNL = dict(flags="/Qx:CANNONLAKE"),
AVX512_ICL = dict(flags="/Qx:ICELAKE-CLIENT")
)
if on_x86 and self.cc_is_msvc: return dict(
SSE = dict(flags="/arch:SSE"),
SSE2 = dict(flags="/arch:SSE2"),
SSE3 = {},
SSSE3 = {},
SSE41 = {},
POPCNT = dict(headers="nmmintrin.h"),
SSE42 = {},
AVX = dict(flags="/arch:AVX"),
F16C = {},
XOP = dict(headers="ammintrin.h"),
FMA4 = dict(headers="ammintrin.h"),
# MSVC doesn't support FMA3 or AVX2 independently
FMA3 = dict(
implies="F16C AVX2", flags="/arch:AVX2"
),
AVX2 = dict(
implies="F16C FMA3", flags="/arch:AVX2"
),
# MSVC doesn't support AVX512F or AVX512CD independently,
# always generate instructions belong to (VL/VW/DQ)
AVX512F = dict(
implies="AVX2 AVX512CD AVX512_SKX", flags="/arch:AVX512"
),
AVX512CD = dict(
implies="AVX512F AVX512_SKX", flags="/arch:AVX512"
),
AVX512_KNL = dict(
disable="MSVC compiler doesn't support it"
),
AVX512_KNM = dict(
disable="MSVC compiler doesn't support it"
),
AVX512_SKX = dict(flags="/arch:AVX512"),
AVX512_CLX = {},
AVX512_CNL = {},
AVX512_ICL = {}
)
on_power = self.cc_on_ppc64le or self.cc_on_ppc64
if on_power:
partial = dict(
VSX = dict(
implies=("VSX2" if self.cc_on_ppc64le else ""),
flags="-mvsx"
),
VSX2 = dict(
flags="-mcpu=power8", implies_detect=False
),
VSX3 = dict(
flags="-mcpu=power9 -mtune=power9", implies_detect=False
)
)
if self.cc_is_clang:
partial["VSX"]["flags"] = "-maltivec -mvsx"
partial["VSX2"]["flags"] = "-mpower8-vector"
partial["VSX3"]["flags"] = "-mpower9-vector"
return partial
if self.cc_on_aarch64 and is_unix: return dict(
NEON = dict(
implies="NEON_FP16 NEON_VFPV4 ASIMD", autovec=True
),
NEON_FP16 = dict(
implies="NEON NEON_VFPV4 ASIMD", autovec=True
),
NEON_VFPV4 = dict(
implies="NEON NEON_FP16 ASIMD", autovec=True
),
ASIMD = dict(
implies="NEON NEON_FP16 NEON_VFPV4", autovec=True
),
ASIMDHP = dict(
flags="-march=armv8.2-a+fp16"
),
ASIMDDP = dict(
flags="-march=armv8.2-a+dotprod"
),
ASIMDFHM = dict(
flags="-march=armv8.2-a+fp16fml"
),
)
if self.cc_on_armhf and is_unix: return dict(
NEON = dict(
flags="-mfpu=neon"
),
NEON_FP16 = dict(
flags="-mfpu=neon-fp16 -mfp16-format=ieee"
),
NEON_VFPV4 = dict(
flags="-mfpu=neon-vfpv4",
),
ASIMD = dict(
flags="-mfpu=neon-fp-armv8 -march=armv8-a+simd",
),
ASIMDHP = dict(
flags="-march=armv8.2-a+fp16"
),
ASIMDDP = dict(
flags="-march=armv8.2-a+dotprod",
),
ASIMDFHM = dict(
flags="-march=armv8.2-a+fp16fml"
)
)
# TODO: ARM MSVC
return {}
def __init__(self):
if self.conf_tmp_path is None:
import tempfile, shutil
tmp = tempfile.mkdtemp()
def rm_temp():
try:
shutil.rmtree(tmp)
except IOError:
pass
atexit.register(rm_temp)
self.conf_tmp_path = tmp
if self.conf_cache_factors is None:
self.conf_cache_factors = [
os.path.getmtime(__file__),
self.conf_nocache
]
class _Distutils:
"""A helper class that provides a collection of fundamental methods
implemented in a top of Python and NumPy Distutils.
The idea behind this class is to gather all methods that it may
need to override in case of reuse 'CCompilerOpt' in environment
different than of what NumPy has.
Parameters
----------
ccompiler : `CCompiler`
The generate instance that returned from `distutils.ccompiler.new_compiler()`.
"""
def __init__(self, ccompiler):
self._ccompiler = ccompiler
def dist_compile(self, sources, flags, ccompiler=None, **kwargs):
"""Wrap CCompiler.compile()"""
assert(isinstance(sources, list))
assert(isinstance(flags, list))
flags = kwargs.pop("extra_postargs", []) + flags
if not ccompiler:
ccompiler = self._ccompiler
return ccompiler.compile(sources, extra_postargs=flags, **kwargs)
def dist_test(self, source, flags, macros=[]):
"""Return True if 'CCompiler.compile()' able to compile
a source file with certain flags.
"""
assert(isinstance(source, str))
from distutils.errors import CompileError
cc = self._ccompiler;
bk_spawn = getattr(cc, 'spawn', None)
if bk_spawn:
cc_type = getattr(self._ccompiler, "compiler_type", "")
if cc_type in ("msvc",):
setattr(cc, 'spawn', self._dist_test_spawn_paths)
else:
setattr(cc, 'spawn', self._dist_test_spawn)
test = False
try:
self.dist_compile(
[source], flags, macros=macros, output_dir=self.conf_tmp_path
)
test = True
except CompileError as e:
self.dist_log(str(e), stderr=True)
if bk_spawn:
setattr(cc, 'spawn', bk_spawn)
return test
def dist_info(self):
"""
Return a tuple containing info about (platform, compiler, extra_args),
required by the abstract class '_CCompiler' for discovering the
platform environment. This is also used as a cache factor in order
to detect any changes happening from outside.
"""
if hasattr(self, "_dist_info"):
return self._dist_info
cc_type = getattr(self._ccompiler, "compiler_type", '')
if cc_type in ("intelem", "intelemw"):
platform = "x86_64"
elif cc_type in ("intel", "intelw", "intele"):
platform = "x86"
else:
from distutils.util import get_platform
platform = get_platform()
cc_info = getattr(self._ccompiler, "compiler", getattr(self._ccompiler, "compiler_so", ''))
if not cc_type or cc_type == "unix":
if hasattr(cc_info, "__iter__"):
compiler = cc_info[0]
else:
compiler = str(cc_info)
else:
compiler = cc_type
if hasattr(cc_info, "__iter__") and len(cc_info) > 1:
extra_args = ' '.join(cc_info[1:])
else:
extra_args = os.environ.get("CFLAGS", "")
extra_args += os.environ.get("CPPFLAGS", "")
self._dist_info = (platform, compiler, extra_args)
return self._dist_info
@staticmethod
def dist_error(*args):
"""Raise a compiler error"""
from distutils.errors import CompileError
raise CompileError(_Distutils._dist_str(*args))
@staticmethod
def dist_fatal(*args):
"""Raise a distutils error"""
from distutils.errors import DistutilsError
raise DistutilsError(_Distutils._dist_str(*args))
@staticmethod
def dist_log(*args, stderr=False):
"""Print a console message"""
from numpy.distutils import log
out = _Distutils._dist_str(*args)
if stderr:
log.warn(out)
else:
log.info(out)
@staticmethod
def dist_load_module(name, path):
"""Load a module from file, required by the abstract class '_Cache'."""
from numpy.compat import npy_load_module
try:
return npy_load_module(name, path)
except Exception as e:
_Distutils.dist_log(e, stderr=True)
return None
@staticmethod
def _dist_str(*args):
"""Return a string to print by log and errors."""
def to_str(arg):
if not isinstance(arg, str) and hasattr(arg, '__iter__'):
ret = []
for a in arg:
ret.append(to_str(a))
return '('+ ' '.join(ret) + ')'
return str(arg)
stack = inspect.stack()[2]
start = "CCompilerOpt.%s[%d] : " % (stack.function, stack.lineno)
out = ' '.join([
to_str(a)
for a in (*args,)
])
return start + out
def _dist_test_spawn_paths(self, cmd, display=None):
"""
Fix msvc SDK ENV path same as distutils do
without it we get c1: fatal error C1356: unable to find mspdbcore.dll
"""
if not hasattr(self._ccompiler, "_paths"):
self._dist_test_spawn(cmd)
return
old_path = os.getenv("path")
try:
os.environ["path"] = self._ccompiler._paths
self._dist_test_spawn(cmd)
finally:
os.environ["path"] = old_path
_dist_warn_regex = re.compile(
# intel and msvc compilers don't raise
# fatal errors when flags are wrong or unsupported
".*("
"warning D9002|" # msvc, it should be work with any language.
"invalid argument for option" # intel
").*"
)
@staticmethod
def _dist_test_spawn(cmd, display=None):
from distutils.errors import CompileError
try:
o = subprocess.check_output(cmd, stderr=subprocess.STDOUT,
universal_newlines=True)
if o and re.match(_Distutils._dist_warn_regex, o):
_Distutils.dist_error(
"Flags in command", cmd ,"aren't supported by the compiler"
", output -> \n%s" % o
)
except subprocess.CalledProcessError as exc:
o = exc.output
s = exc.returncode
except OSError:
o = b''
s = 127
else:
return None
_Distutils.dist_error(
"Command", cmd, "failed with exit status %d output -> \n%s" % (
s, o
))
_share_cache = {}
class _Cache:
"""An abstract class handles caching functionality, provides two
levels of caching, in-memory by share instances attributes among
each other and by store attributes into files.
**Note**:
any attributes that start with ``_`` or ``conf_`` will be ignored.
Parameters
----------
cache_path: str or None
The path of cache file, if None then cache in file will disabled.
*factors:
The caching factors that need to utilize next to `conf_cache_factors`.
Attributes
----------
cache_private: set
Hold the attributes that need be skipped from "in-memory cache".
cache_infile: bool
Utilized during initializing this class, to determine if the cache was able
to loaded from the specified cache path in 'cache_path'.
"""
# skip attributes from cache
_cache_ignore = re.compile("^(_|conf_)")
def __init__(self, cache_path=None, *factors):
self.cache_me = {}
self.cache_private = set()
self.cache_infile = False
self._cache_path = None
if self.conf_nocache:
self.dist_log("cache is disabled by `Config`")
return
self._cache_hash = self.cache_hash(*factors, *self.conf_cache_factors)
self._cache_path = cache_path
if cache_path:
if os.path.exists(cache_path):
self.dist_log("load cache from file ->", cache_path)
cache_mod = self.dist_load_module("cache", cache_path)
if not cache_mod:
self.dist_log(
"unable to load the cache file as a module",
stderr=True
)
elif not hasattr(cache_mod, "hash") or \
not hasattr(cache_mod, "data"):
self.dist_log("invalid cache file", stderr=True)
elif self._cache_hash == cache_mod.hash:
self.dist_log("hit the file cache")
for attr, val in cache_mod.data.items():
setattr(self, attr, val)
self.cache_infile = True
else:
self.dist_log("miss the file cache")
if not self.cache_infile:
other_cache = _share_cache.get(self._cache_hash)
if other_cache:
self.dist_log("hit the memory cache")
for attr, val in other_cache.__dict__.items():
if attr in other_cache.cache_private or \
re.match(self._cache_ignore, attr):
continue
setattr(self, attr, val)
_share_cache[self._cache_hash] = self
atexit.register(self.cache_flush)
def __del__(self):
for h, o in _share_cache.items():
if o == self:
_share_cache.pop(h)
break
def cache_flush(self):
"""
Force update the cache.
"""
if not self._cache_path:
return
# TODO: don't write if the cache doesn't change
self.dist_log("write cache to path ->", self._cache_path)
cdict = self.__dict__.copy()
for attr in self.__dict__.keys():
if re.match(self._cache_ignore, attr):
cdict.pop(attr)
d = os.path.dirname(self._cache_path)
if not os.path.exists(d):
os.makedirs(d)
repr_dict = pprint.pformat(cdict, compact=True)
with open(self._cache_path, "w") as f:
f.write(textwrap.dedent("""\
# AUTOGENERATED DON'T EDIT
# Please make changes to the code generator \
(distutils/ccompiler_opt.py)
hash = {}
data = \\
""").format(self._cache_hash))
f.write(repr_dict)
def cache_hash(self, *factors):
# is there a built-in non-crypto hash?
# sdbm
chash = 0
for f in factors:
for char in str(f):
chash = ord(char) + (chash << 6) + (chash << 16) - chash
chash &= 0xFFFFFFFF
return chash
@staticmethod
def me(cb):
"""
A static method that can be treated as a decorator to
dynamically cache certain methods.
"""
def cache_wrap_me(self, *args, **kwargs):
# good for normal args
cache_key = str((
cb.__name__, *args, *kwargs.keys(), *kwargs.values()
))
if cache_key in self.cache_me:
return self.cache_me[cache_key]
ccb = cb(self, *args, **kwargs)
self.cache_me[cache_key] = ccb
return ccb
return cache_wrap_me
class _CCompiler:
"""A helper class for `CCompilerOpt` containing all utilities that
related to the fundamental compiler's functions.
Attributes
----------
cc_on_x86 : bool
True when the target architecture is 32-bit x86
cc_on_x64 : bool
True when the target architecture is 64-bit x86
cc_on_ppc64 : bool
True when the target architecture is 64-bit big-endian PowerPC
cc_on_armhf : bool
True when the target architecture is 32-bit ARMv7+
cc_on_aarch64 : bool
True when the target architecture is 64-bit Armv8-a+
cc_on_noarch : bool
True when the target architecture is unknown or not supported
cc_is_gcc : bool
True if the compiler is GNU or
if the compiler is unknown
cc_is_clang : bool
True if the compiler is Clang
cc_is_icc : bool
True if the compiler is Intel compiler (unix like)
cc_is_iccw : bool
True if the compiler is Intel compiler (msvc like)
cc_is_nocc : bool
True if the compiler isn't supported directly,
Note: that cause a fail-back to gcc
cc_has_debug : bool
True if the compiler has debug flags
cc_has_native : bool
True if the compiler has native flags
cc_noopt : bool
True if the compiler has definition 'DISABLE_OPT*',
or 'cc_on_noarch' is True
cc_march : str
The target architecture name, or "unknown" if
the architecture isn't supported
cc_name : str
The compiler name, or "unknown" if the compiler isn't supported
cc_flags : dict
Dictionary containing the initialized flags of `_Config.conf_cc_flags`
"""
def __init__(self):
if hasattr(self, "cc_is_cached"):
return
# attr regex
detect_arch = (
("cc_on_x64", ".*(x|x86_|amd)64.*"),
("cc_on_x86", ".*(win32|x86|i386|i686).*"),
("cc_on_ppc64le", ".*(powerpc|ppc)64(el|le).*"),
("cc_on_ppc64", ".*(powerpc|ppc)64.*"),
("cc_on_aarch64", ".*(aarch64|arm64).*"),
("cc_on_armhf", ".*arm.*"),
# undefined platform
("cc_on_noarch", ""),
)
detect_compiler = (
("cc_is_gcc", r".*(gcc|gnu\-g).*"),
("cc_is_clang", ".*clang.*"),
("cc_is_iccw", ".*(intelw|intelemw|iccw).*"), # intel msvc like
("cc_is_icc", ".*(intel|icc).*"), # intel unix like
("cc_is_msvc", ".*msvc.*"),
# undefined compiler will be treat it as gcc
("cc_is_nocc", ""),
)
detect_args = (
("cc_has_debug", ".*(O0|Od|ggdb|coverage|debug:full).*"),
("cc_has_native", ".*(-march=native|-xHost|/QxHost).*"),
# in case if the class run with -DNPY_DISABLE_OPTIMIZATION
("cc_noopt", ".*DISABLE_OPT.*"),
)
dist_info = self.dist_info()
platform, compiler_info, extra_args = dist_info
# set False to all attrs
for section in (detect_arch, detect_compiler, detect_args):
for attr, rgex in section:
setattr(self, attr, False)
for detect, searchin in ((detect_arch, platform), (detect_compiler, compiler_info)):
for attr, rgex in detect:
if rgex and not re.match(rgex, searchin, re.IGNORECASE):
continue
setattr(self, attr, True)
break
for attr, rgex in detect_args:
if rgex and not re.match(rgex, extra_args, re.IGNORECASE):
continue
setattr(self, attr, True)
if self.cc_on_noarch:
self.dist_log(
"unable to detect CPU architecture which lead to disable the optimization. "
f"check dist_info:<<\n{dist_info}\n>>",
stderr=True
)
self.cc_noopt = True
if self.conf_noopt:
self.dist_log("Optimization is disabled by the Config", stderr=True)
self.cc_noopt = True
if self.cc_is_nocc:
"""
mingw can be treated as a gcc, and also xlc even if it based on clang,
but still has the same gcc optimization flags.
"""
self.dist_log(
"unable to detect compiler type which leads to treating it as GCC. "
"this is a normal behavior if you're using gcc-like compiler such as MinGW or IBM/XLC."
f"check dist_info:<<\n{dist_info}\n>>",
stderr=True
)
self.cc_is_gcc = True
self.cc_march = "unknown"
for arch in ("x86", "x64", "ppc64", "ppc64le", "armhf", "aarch64"):
if getattr(self, "cc_on_" + arch):
self.cc_march = arch
break
self.cc_name = "unknown"
for name in ("gcc", "clang", "iccw", "icc", "msvc"):
if getattr(self, "cc_is_" + name):
self.cc_name = name
break
self.cc_flags = {}
compiler_flags = self.conf_cc_flags.get(self.cc_name)
if compiler_flags is None:
self.dist_fatal(
"undefined flag for compiler '%s', "
"leave an empty dict instead" % self.cc_name
)
for name, flags in compiler_flags.items():
self.cc_flags[name] = nflags = []
if flags:
assert(isinstance(flags, str))
flags = flags.split()
for f in flags:
if self.cc_test_flags([f]):
nflags.append(f)
self.cc_is_cached = True
@_Cache.me
def cc_test_flags(self, flags):
"""
Returns True if the compiler supports 'flags'.
"""
assert(isinstance(flags, list))
self.dist_log("testing flags", flags)
test_path = os.path.join(self.conf_check_path, "test_flags.c")
test = self.dist_test(test_path, flags)
if not test:
self.dist_log("testing failed", stderr=True)
return test
def cc_normalize_flags(self, flags):
"""
Remove the conflicts that caused due gathering implied features flags.
Parameters
----------
'flags' list, compiler flags
flags should be sorted from the lowest to the highest interest.
Returns
-------
list, filtered from any conflicts.
Examples
--------
>>> self.cc_normalize_flags(['-march=armv8.2-a+fp16', '-march=armv8.2-a+dotprod'])
['armv8.2-a+fp16+dotprod']
>>> self.cc_normalize_flags(
['-msse', '-msse2', '-msse3', '-mssse3', '-msse4.1', '-msse4.2', '-mavx', '-march=core-avx2']
)
['-march=core-avx2']
"""
assert(isinstance(flags, list))
if self.cc_is_gcc or self.cc_is_clang or self.cc_is_icc:
return self._cc_normalize_unix(flags)
if self.cc_is_msvc or self.cc_is_iccw:
return self._cc_normalize_win(flags)
return flags
_cc_normalize_unix_mrgx = re.compile(
# 1- to check the highest of
r"^(-mcpu=|-march=|-x[A-Z0-9\-])"
)
_cc_normalize_unix_frgx = re.compile(
# 2- to remove any flags starts with
# -march, -mcpu, -x(INTEL) and '-m' without '='
r"^(?!(-mcpu=|-march=|-x[A-Z0-9\-]))(?!-m[a-z0-9\-\.]*.$)"
)
_cc_normalize_unix_krgx = re.compile(
# 3- keep only the highest of
r"^(-mfpu|-mtune)"
)
_cc_normalize_arch_ver = re.compile(
r"[0-9.]"
)
def _cc_normalize_unix(self, flags):
def ver_flags(f):
# arch ver subflag
# -march=armv8.2-a+fp16fml
tokens = f.split('+')
ver = float('0' + ''.join(
re.findall(self._cc_normalize_arch_ver, tokens[0])
))
return ver, tokens[0], tokens[1:]
if len(flags) <= 1:
return flags
# get the highest matched flag
for i, cur_flag in enumerate(reversed(flags)):
if not re.match(self._cc_normalize_unix_mrgx, cur_flag):
continue
lower_flags = flags[:-(i+1)]
upper_flags = flags[-i:]
filterd = list(filter(
self._cc_normalize_unix_frgx.search, lower_flags
))
# gather subflags
ver, arch, subflags = ver_flags(cur_flag)
if ver > 0 and len(subflags) > 0:
for xflag in lower_flags:
xver, _, xsubflags = ver_flags(xflag)
if ver == xver:
subflags = xsubflags + subflags
cur_flag = arch + '+' + '+'.join(subflags)
flags = filterd + [cur_flag]
if i > 0:
flags += upper_flags
break
# to remove overridable flags
final_flags = []
matched = set()
for f in reversed(flags):
match = re.match(self._cc_normalize_unix_krgx, f)
if not match:
pass
elif match[0] in matched:
continue
else:
matched.add(match[0])
final_flags.insert(0, f)
return final_flags
_cc_normalize_win_frgx = re.compile(
r"^(?!(/arch\:|/Qx\:))"
)
_cc_normalize_win_mrgx = re.compile(
r"^(/arch|/Qx:)"
)
def _cc_normalize_win(self, flags):
for i, f in enumerate(reversed(flags)):
if not re.match(self._cc_normalize_win_mrgx, f):
continue
i += 1
return list(filter(
self._cc_normalize_win_frgx.search, flags[:-i]
)) + flags[-i:]
return flags
class _Feature:
"""A helper class for `CCompilerOpt` that managing CPU features.
Attributes
----------
feature_supported : dict
Dictionary containing all CPU features that supported
by the platform, according to the specified values in attribute
`_Config.conf_features` and `_Config.conf_features_partial()`
feature_min : set
The minimum support of CPU features, according to
the specified values in attribute `_Config.conf_min_features`.
"""
def __init__(self):
if hasattr(self, "feature_is_cached"):
return
self.feature_supported = pfeatures = self.conf_features_partial()
for feature_name in list(pfeatures.keys()):
feature = pfeatures[feature_name]
cfeature = self.conf_features[feature_name]
feature.update({
k:v for k,v in cfeature.items() if k not in feature
})
disabled = feature.get("disable")
if disabled is not None:
pfeatures.pop(feature_name)
self.dist_log(
"feature '%s' is disabled," % feature_name,
disabled, stderr=True
)
continue
# list is used internally for these options
for option in (
"implies", "group", "detect", "headers", "flags", "extra_checks"
) :
oval = feature.get(option)
if isinstance(oval, str):
feature[option] = oval.split()
self.feature_min = set()
min_f = self.conf_min_features.get(self.cc_march, "")
for F in min_f.upper().split():
if F in self.feature_supported:
self.feature_min.add(F)
self.feature_is_cached = True
def feature_names(self, names=None, force_flags=None, macros=[]):
"""
Returns a set of CPU feature names that supported by platform and the **C** compiler.
Parameters
----------
names: sequence or None, optional
Specify certain CPU features to test it against the **C** compiler.
if None(default), it will test all current supported features.
**Note**: feature names must be in upper-case.
force_flags: list or None, optional
If None(default), default compiler flags for every CPU feature will
be used during the test.
macros : list of tuples, optional
A list of C macro definitions.
"""
assert(
names is None or (
not isinstance(names, str) and
hasattr(names, "__iter__")
)
)
assert(force_flags is None or isinstance(force_flags, list))
if names is None:
names = self.feature_supported.keys()
supported_names = set()
for f in names:
if self.feature_is_supported(
f, force_flags=force_flags, macros=macros
):
supported_names.add(f)
return supported_names
def feature_is_exist(self, name):
"""
Returns True if a certain feature is exist and covered within
`_Config.conf_features`.
Parameters
----------
'name': str
feature name in uppercase.
"""
assert(name.isupper())
return name in self.conf_features
def feature_sorted(self, names, reverse=False):
"""
Sort a list of CPU features ordered by the lowest interest.
Parameters
----------
'names': sequence
sequence of supported feature names in uppercase.
'reverse': bool, optional
If true, the sorted features is reversed. (highest interest)
Returns
-------
list, sorted CPU features
"""
def sort_cb(k):
if isinstance(k, str):
return self.feature_supported[k]["interest"]
# multiple features
rank = max([self.feature_supported[f]["interest"] for f in k])
# FIXME: that's not a safe way to increase the rank for
# multi targets
rank += len(k) -1
return rank
return sorted(names, reverse=reverse, key=sort_cb)
def feature_implies(self, names, keep_origins=False):
"""
Return a set of CPU features that implied by 'names'
Parameters
----------
names: str or sequence of str
CPU feature name(s) in uppercase.
keep_origins: bool
if False(default) then the returned set will not contain any
features from 'names'. This case happens only when two features
imply each other.
Examples
--------
>>> self.feature_implies("SSE3")
{'SSE', 'SSE2'}
>>> self.feature_implies("SSE2")
{'SSE'}
>>> self.feature_implies("SSE2", keep_origins=True)
# 'SSE2' found here since 'SSE' and 'SSE2' imply each other
{'SSE', 'SSE2'}
"""
def get_implies(name, _caller=set()):
implies = set()
d = self.feature_supported[name]
for i in d.get("implies", []):
implies.add(i)
if i in _caller:
# infinity recursive guard since
# features can imply each other
continue
_caller.add(name)
implies = implies.union(get_implies(i, _caller))
return implies
if isinstance(names, str):
implies = get_implies(names)
names = [names]
else:
assert(hasattr(names, "__iter__"))
implies = set()
for n in names:
implies = implies.union(get_implies(n))
if not keep_origins:
implies.difference_update(names)
return implies
def feature_implies_c(self, names):
"""same as feature_implies() but combining 'names'"""
if isinstance(names, str):
names = set((names,))
else:
names = set(names)
return names.union(self.feature_implies(names))
def feature_ahead(self, names):
"""
Return list of features in 'names' after remove any
implied features and keep the origins.
Parameters
----------
'names': sequence
sequence of CPU feature names in uppercase.
Returns
-------
list of CPU features sorted as-is 'names'
Examples
--------
>>> self.feature_ahead(["SSE2", "SSE3", "SSE41"])
["SSE41"]
# assume AVX2 and FMA3 implies each other and AVX2
# is the highest interest
>>> self.feature_ahead(["SSE2", "SSE3", "SSE41", "AVX2", "FMA3"])
["AVX2"]
# assume AVX2 and FMA3 don't implies each other
>>> self.feature_ahead(["SSE2", "SSE3", "SSE41", "AVX2", "FMA3"])
["AVX2", "FMA3"]
"""
assert(
not isinstance(names, str)
and hasattr(names, '__iter__')
)
implies = self.feature_implies(names, keep_origins=True)
ahead = [n for n in names if n not in implies]
if len(ahead) == 0:
# return the highest interested feature
# if all features imply each other
ahead = self.feature_sorted(names, reverse=True)[:1]
return ahead
def feature_untied(self, names):
"""
same as 'feature_ahead()' but if both features implied each other
and keep the highest interest.
Parameters
----------
'names': sequence
sequence of CPU feature names in uppercase.
Returns
-------
list of CPU features sorted as-is 'names'
Examples
--------
>>> self.feature_untied(["SSE2", "SSE3", "SSE41"])
["SSE2", "SSE3", "SSE41"]
# assume AVX2 and FMA3 implies each other
>>> self.feature_untied(["SSE2", "SSE3", "SSE41", "FMA3", "AVX2"])
["SSE2", "SSE3", "SSE41", "AVX2"]
"""
assert(
not isinstance(names, str)
and hasattr(names, '__iter__')
)
final = []
for n in names:
implies = self.feature_implies(n)
tied = [
nn for nn in final
if nn in implies and n in self.feature_implies(nn)
]
if tied:
tied = self.feature_sorted(tied + [n])
if n not in tied[1:]:
continue
final.remove(tied[:1][0])
final.append(n)
return final
def feature_get_til(self, names, keyisfalse):
"""
same as `feature_implies_c()` but stop collecting implied
features when feature's option that provided through
parameter 'keyisfalse' is False, also sorting the returned
features.
"""
def til(tnames):
# sort from highest to lowest interest then cut if "key" is False
tnames = self.feature_implies_c(tnames)
tnames = self.feature_sorted(tnames, reverse=True)
for i, n in enumerate(tnames):
if not self.feature_supported[n].get(keyisfalse, True):
tnames = tnames[:i+1]
break
return tnames
if isinstance(names, str) or len(names) <= 1:
names = til(names)
# normalize the sort
names.reverse()
return names
names = self.feature_ahead(names)
names = {t for n in names for t in til(n)}
return self.feature_sorted(names)
def feature_detect(self, names):
"""
Return a list of CPU features that required to be detected
sorted from the lowest to highest interest.
"""
names = self.feature_get_til(names, "implies_detect")
detect = []
for n in names:
d = self.feature_supported[n]
detect += d.get("detect", d.get("group", [n]))
return detect
@_Cache.me
def feature_flags(self, names):
"""
Return a list of CPU features flags sorted from the lowest
to highest interest.
"""
names = self.feature_sorted(self.feature_implies_c(names))
flags = []
for n in names:
d = self.feature_supported[n]
f = d.get("flags", [])
if not f or not self.cc_test_flags(f):
continue
flags += f
return self.cc_normalize_flags(flags)
@_Cache.me
def feature_test(self, name, force_flags=None, macros=[]):
"""
Test a certain CPU feature against the compiler through its own
check file.
Parameters
----------
name: str
Supported CPU feature name.
force_flags: list or None, optional
If None(default), the returned flags from `feature_flags()`
will be used.
macros : list of tuples, optional
A list of C macro definitions.
"""
if force_flags is None:
force_flags = self.feature_flags(name)
self.dist_log(
"testing feature '%s' with flags (%s)" % (
name, ' '.join(force_flags)
))
# Each CPU feature must have C source code contains at
# least one intrinsic or instruction related to this feature.
test_path = os.path.join(
self.conf_check_path, "cpu_%s.c" % name.lower()
)
if not os.path.exists(test_path):
self.dist_fatal("feature test file is not exist", test_path)
test = self.dist_test(
test_path, force_flags + self.cc_flags["werror"], macros=macros
)
if not test:
self.dist_log("testing failed", stderr=True)
return test
@_Cache.me
def feature_is_supported(self, name, force_flags=None, macros=[]):
"""
Check if a certain CPU feature is supported by the platform and compiler.
Parameters
----------
name: str
CPU feature name in uppercase.
force_flags: list or None, optional
If None(default), default compiler flags for every CPU feature will
be used during test.
macros : list of tuples, optional
A list of C macro definitions.
"""
assert(name.isupper())
assert(force_flags is None or isinstance(force_flags, list))
supported = name in self.feature_supported
if supported:
for impl in self.feature_implies(name):
if not self.feature_test(impl, force_flags, macros=macros):
return False
if not self.feature_test(name, force_flags, macros=macros):
return False
return supported
@_Cache.me
def feature_can_autovec(self, name):
"""
check if the feature can be auto-vectorized by the compiler
"""
assert(isinstance(name, str))
d = self.feature_supported[name]
can = d.get("autovec", None)
if can is None:
valid_flags = [
self.cc_test_flags([f]) for f in d.get("flags", [])
]
can = valid_flags and any(valid_flags)
return can
@_Cache.me
def feature_extra_checks(self, name):
"""
Return a list of supported extra checks after testing them against
the compiler.
Parameters
----------
names: str
CPU feature name in uppercase.
"""
assert isinstance(name, str)
d = self.feature_supported[name]
extra_checks = d.get("extra_checks", [])
if not extra_checks:
return []
self.dist_log("Testing extra checks for feature '%s'" % name, extra_checks)
flags = self.feature_flags(name)
available = []
not_available = []
for chk in extra_checks:
test_path = os.path.join(
self.conf_check_path, "extra_%s.c" % chk.lower()
)
if not os.path.exists(test_path):
self.dist_fatal("extra check file does not exist", test_path)
is_supported = self.dist_test(test_path, flags + self.cc_flags["werror"])
if is_supported:
available.append(chk)
else:
not_available.append(chk)
if not_available:
self.dist_log("testing failed for checks", not_available, stderr=True)
return available
def feature_c_preprocessor(self, feature_name, tabs=0):
"""
Generate C preprocessor definitions and include headers of a CPU feature.
Parameters
----------
'feature_name': str
CPU feature name in uppercase.
'tabs': int
if > 0, align the generated strings to the right depend on number of tabs.
Returns
-------
str, generated C preprocessor
Examples
--------
>>> self.feature_c_preprocessor("SSE3")
/** SSE3 **/
#define NPY_HAVE_SSE3 1
#include <pmmintrin.h>
"""
assert(feature_name.isupper())
feature = self.feature_supported.get(feature_name)
assert(feature is not None)
prepr = [
"/** %s **/" % feature_name,
"#define %sHAVE_%s 1" % (self.conf_c_prefix, feature_name)
]
prepr += [
"#include <%s>" % h for h in feature.get("headers", [])
]
extra_defs = feature.get("group", [])
extra_defs += self.feature_extra_checks(feature_name)
for edef in extra_defs:
# Guard extra definitions in case of duplicate with
# another feature
prepr += [
"#ifndef %sHAVE_%s" % (self.conf_c_prefix, edef),
"\t#define %sHAVE_%s 1" % (self.conf_c_prefix, edef),
"#endif",
]
if tabs > 0:
prepr = [('\t'*tabs) + l for l in prepr]
return '\n'.join(prepr)
class _Parse:
"""A helper class that parsing main arguments of `CCompilerOpt`,
also parsing configuration statements in dispatch-able sources.
Parameters
----------
cpu_baseline: str or None
minimal set of required CPU features or special options.
cpu_dispatch: str or None
dispatched set of additional CPU features or special options.
Special options can be:
- **MIN**: Enables the minimum CPU features that utilized via `_Config.conf_min_features`
- **MAX**: Enables all supported CPU features by the Compiler and platform.
- **NATIVE**: Enables all CPU features that supported by the current machine.
- **NONE**: Enables nothing
- **Operand +/-**: remove or add features, useful with options **MAX**, **MIN** and **NATIVE**.
NOTE: operand + is only added for nominal reason.
NOTES:
- Case-insensitive among all CPU features and special options.
- Comma or space can be used as a separator.
- If the CPU feature is not supported by the user platform or compiler,
it will be skipped rather than raising a fatal error.
- Any specified CPU features to 'cpu_dispatch' will be skipped if its part of CPU baseline features
- 'cpu_baseline' force enables implied features.
Attributes
----------
parse_baseline_names : list
Final CPU baseline's feature names(sorted from low to high)
parse_baseline_flags : list
Compiler flags of baseline features
parse_dispatch_names : list
Final CPU dispatch-able feature names(sorted from low to high)
parse_target_groups : dict
Dictionary containing initialized target groups that configured
through class attribute `conf_target_groups`.
The key is represent the group name and value is a tuple
contains three items :
- bool, True if group has the 'baseline' option.
- list, list of CPU features.
- list, list of extra compiler flags.
"""
def __init__(self, cpu_baseline, cpu_dispatch):
self._parse_policies = dict(
# POLICY NAME, (HAVE, NOT HAVE, [DEB])
KEEP_BASELINE = (
None, self._parse_policy_not_keepbase,
[]
),
KEEP_SORT = (
self._parse_policy_keepsort,
self._parse_policy_not_keepsort,
[]
),
MAXOPT = (
self._parse_policy_maxopt, None,
[]
),
WERROR = (
self._parse_policy_werror, None,
[]
),
AUTOVEC = (
self._parse_policy_autovec, None,
["MAXOPT"]
)
)
if hasattr(self, "parse_is_cached"):
return
self.parse_baseline_names = []
self.parse_baseline_flags = []
self.parse_dispatch_names = []
self.parse_target_groups = {}
if self.cc_noopt:
# skip parsing baseline and dispatch args and keep parsing target groups
cpu_baseline = cpu_dispatch = None
self.dist_log("check requested baseline")
if cpu_baseline is not None:
cpu_baseline = self._parse_arg_features("cpu_baseline", cpu_baseline)
baseline_names = self.feature_names(cpu_baseline)
self.parse_baseline_flags = self.feature_flags(baseline_names)
self.parse_baseline_names = self.feature_sorted(
self.feature_implies_c(baseline_names)
)
self.dist_log("check requested dispatch-able features")
if cpu_dispatch is not None:
cpu_dispatch_ = self._parse_arg_features("cpu_dispatch", cpu_dispatch)
cpu_dispatch = {
f for f in cpu_dispatch_
if f not in self.parse_baseline_names
}
conflict_baseline = cpu_dispatch_.difference(cpu_dispatch)
self.parse_dispatch_names = self.feature_sorted(
self.feature_names(cpu_dispatch)
)
if len(conflict_baseline) > 0:
self.dist_log(
"skip features", conflict_baseline, "since its part of baseline"
)
self.dist_log("initialize targets groups")
for group_name, tokens in self.conf_target_groups.items():
self.dist_log("parse target group", group_name)
GROUP_NAME = group_name.upper()
if not tokens or not tokens.strip():
# allow empty groups, useful in case if there's a need
# to disable certain group since '_parse_target_tokens()'
# requires at least one valid target
self.parse_target_groups[GROUP_NAME] = (
False, [], []
)
continue
has_baseline, features, extra_flags = \
self._parse_target_tokens(tokens)
self.parse_target_groups[GROUP_NAME] = (
has_baseline, features, extra_flags
)
self.parse_is_cached = True
def parse_targets(self, source):
"""
Fetch and parse configuration statements that required for
defining the targeted CPU features, statements should be declared
in the top of source in between **C** comment and start
with a special mark **@targets**.
Configuration statements are sort of keywords representing
CPU features names, group of statements and policies, combined
together to determine the required optimization.
Parameters
----------
source: str
the path of **C** source file.
Returns
-------
- bool, True if group has the 'baseline' option
- list, list of CPU features
- list, list of extra compiler flags
"""
self.dist_log("looking for '@targets' inside -> ", source)
# get lines between /*@targets and */
with open(source) as fd:
tokens = ""
max_to_reach = 1000 # good enough, isn't?
start_with = "@targets"
start_pos = -1
end_with = "*/"
end_pos = -1
for current_line, line in enumerate(fd):
if current_line == max_to_reach:
self.dist_fatal("reached the max of lines")
break
if start_pos == -1:
start_pos = line.find(start_with)
if start_pos == -1:
continue
start_pos += len(start_with)
tokens += line
end_pos = line.find(end_with)
if end_pos != -1:
end_pos += len(tokens) - len(line)
break
if start_pos == -1:
self.dist_fatal("expected to find '%s' within a C comment" % start_with)
if end_pos == -1:
self.dist_fatal("expected to end with '%s'" % end_with)
tokens = tokens[start_pos:end_pos]
return self._parse_target_tokens(tokens)
_parse_regex_arg = re.compile(r'\s|,|([+-])')
def _parse_arg_features(self, arg_name, req_features):
if not isinstance(req_features, str):
self.dist_fatal("expected a string in '%s'" % arg_name)
final_features = set()
# space and comma can be used as a separator
tokens = list(filter(None, re.split(self._parse_regex_arg, req_features)))
append = True # append is the default
for tok in tokens:
if tok[0] in ("#", "$"):
self.dist_fatal(
arg_name, "target groups and policies "
"aren't allowed from arguments, "
"only from dispatch-able sources"
)
if tok == '+':
append = True
continue
if tok == '-':
append = False
continue
TOK = tok.upper() # we use upper-case internally
features_to = set()
if TOK == "NONE":
pass
elif TOK == "NATIVE":
native = self.cc_flags["native"]
if not native:
self.dist_fatal(arg_name,
"native option isn't supported by the compiler"
)
features_to = self.feature_names(
force_flags=native, macros=[("DETECT_FEATURES", 1)]
)
elif TOK == "MAX":
features_to = self.feature_supported.keys()
elif TOK == "MIN":
features_to = self.feature_min
else:
if TOK in self.feature_supported:
features_to.add(TOK)
else:
if not self.feature_is_exist(TOK):
self.dist_fatal(arg_name,
", '%s' isn't a known feature or option" % tok
)
if append:
final_features = final_features.union(features_to)
else:
final_features = final_features.difference(features_to)
append = True # back to default
return final_features
_parse_regex_target = re.compile(r'\s|[*,/]|([()])')
def _parse_target_tokens(self, tokens):
assert(isinstance(tokens, str))
final_targets = [] # to keep it sorted as specified
extra_flags = []
has_baseline = False
skipped = set()
policies = set()
multi_target = None
tokens = list(filter(None, re.split(self._parse_regex_target, tokens)))
if not tokens:
self.dist_fatal("expected one token at least")
for tok in tokens:
TOK = tok.upper()
ch = tok[0]
if ch in ('+', '-'):
self.dist_fatal(
"+/- are 'not' allowed from target's groups or @targets, "
"only from cpu_baseline and cpu_dispatch parms"
)
elif ch == '$':
if multi_target is not None:
self.dist_fatal(
"policies aren't allowed inside multi-target '()'"
", only CPU features"
)
policies.add(self._parse_token_policy(TOK))
elif ch == '#':
if multi_target is not None:
self.dist_fatal(
"target groups aren't allowed inside multi-target '()'"
", only CPU features"
)
has_baseline, final_targets, extra_flags = \
self._parse_token_group(TOK, has_baseline, final_targets, extra_flags)
elif ch == '(':
if multi_target is not None:
self.dist_fatal("unclosed multi-target, missing ')'")
multi_target = set()
elif ch == ')':
if multi_target is None:
self.dist_fatal("multi-target opener '(' wasn't found")
targets = self._parse_multi_target(multi_target)
if targets is None:
skipped.add(tuple(multi_target))
else:
if len(targets) == 1:
targets = targets[0]
if targets and targets not in final_targets:
final_targets.append(targets)
multi_target = None # back to default
else:
if TOK == "BASELINE":
if multi_target is not None:
self.dist_fatal("baseline isn't allowed inside multi-target '()'")
has_baseline = True
continue
if multi_target is not None:
multi_target.add(TOK)
continue
if not self.feature_is_exist(TOK):
self.dist_fatal("invalid target name '%s'" % TOK)
is_enabled = (
TOK in self.parse_baseline_names or
TOK in self.parse_dispatch_names
)
if is_enabled:
if TOK not in final_targets:
final_targets.append(TOK)
continue
skipped.add(TOK)
if multi_target is not None:
self.dist_fatal("unclosed multi-target, missing ')'")
if skipped:
self.dist_log(
"skip targets", skipped,
"not part of baseline or dispatch-able features"
)
final_targets = self.feature_untied(final_targets)
# add polices dependencies
for p in list(policies):
_, _, deps = self._parse_policies[p]
for d in deps:
if d in policies:
continue
self.dist_log(
"policy '%s' force enables '%s'" % (
p, d
))
policies.add(d)
# release policies filtrations
for p, (have, nhave, _) in self._parse_policies.items():
func = None
if p in policies:
func = have
self.dist_log("policy '%s' is ON" % p)
else:
func = nhave
if not func:
continue
has_baseline, final_targets, extra_flags = func(
has_baseline, final_targets, extra_flags
)
return has_baseline, final_targets, extra_flags
def _parse_token_policy(self, token):
"""validate policy token"""
if len(token) <= 1 or token[-1:] == token[0]:
self.dist_fatal("'$' must stuck in the begin of policy name")
token = token[1:]
if token not in self._parse_policies:
self.dist_fatal(
"'%s' is an invalid policy name, available policies are" % token,
self._parse_policies.keys()
)
return token
def _parse_token_group(self, token, has_baseline, final_targets, extra_flags):
"""validate group token"""
if len(token) <= 1 or token[-1:] == token[0]:
self.dist_fatal("'#' must stuck in the begin of group name")
token = token[1:]
ghas_baseline, gtargets, gextra_flags = self.parse_target_groups.get(
token, (False, None, [])
)
if gtargets is None:
self.dist_fatal(
"'%s' is an invalid target group name, " % token + \
"available target groups are",
self.parse_target_groups.keys()
)
if ghas_baseline:
has_baseline = True
# always keep sorting as specified
final_targets += [f for f in gtargets if f not in final_targets]
extra_flags += [f for f in gextra_flags if f not in extra_flags]
return has_baseline, final_targets, extra_flags
def _parse_multi_target(self, targets):
"""validate multi targets that defined between parentheses()"""
# remove any implied features and keep the origins
if not targets:
self.dist_fatal("empty multi-target '()'")
if not all([
self.feature_is_exist(tar) for tar in targets
]) :
self.dist_fatal("invalid target name in multi-target", targets)
if not all([
(
tar in self.parse_baseline_names or
tar in self.parse_dispatch_names
)
for tar in targets
]) :
return None
targets = self.feature_ahead(targets)
if not targets:
return None
# force sort multi targets, so it can be comparable
targets = self.feature_sorted(targets)
targets = tuple(targets) # hashable
return targets
def _parse_policy_not_keepbase(self, has_baseline, final_targets, extra_flags):
"""skip all baseline features"""
skipped = []
for tar in final_targets[:]:
is_base = False
if isinstance(tar, str):
is_base = tar in self.parse_baseline_names
else:
# multi targets
is_base = all([
f in self.parse_baseline_names
for f in tar
])
if is_base:
skipped.append(tar)
final_targets.remove(tar)
if skipped:
self.dist_log("skip baseline features", skipped)
return has_baseline, final_targets, extra_flags
def _parse_policy_keepsort(self, has_baseline, final_targets, extra_flags):
"""leave a notice that $keep_sort is on"""
self.dist_log(
"policy 'keep_sort' is on, dispatch-able targets", final_targets, "\n"
"are 'not' sorted depend on the highest interest but"
"as specified in the dispatch-able source or the extra group"
)
return has_baseline, final_targets, extra_flags
def _parse_policy_not_keepsort(self, has_baseline, final_targets, extra_flags):
"""sorted depend on the highest interest"""
final_targets = self.feature_sorted(final_targets, reverse=True)
return has_baseline, final_targets, extra_flags
def _parse_policy_maxopt(self, has_baseline, final_targets, extra_flags):
"""append the compiler optimization flags"""
if self.cc_has_debug:
self.dist_log("debug mode is detected, policy 'maxopt' is skipped.")
elif self.cc_noopt:
self.dist_log("optimization is disabled, policy 'maxopt' is skipped.")
else:
flags = self.cc_flags["opt"]
if not flags:
self.dist_log(
"current compiler doesn't support optimization flags, "
"policy 'maxopt' is skipped", stderr=True
)
else:
extra_flags += flags
return has_baseline, final_targets, extra_flags
def _parse_policy_werror(self, has_baseline, final_targets, extra_flags):
"""force warnings to treated as errors"""
flags = self.cc_flags["werror"]
if not flags:
self.dist_log(
"current compiler doesn't support werror flags, "
"warnings will 'not' treated as errors", stderr=True
)
else:
self.dist_log("compiler warnings are treated as errors")
extra_flags += flags
return has_baseline, final_targets, extra_flags
def _parse_policy_autovec(self, has_baseline, final_targets, extra_flags):
"""skip features that has no auto-vectorized support by compiler"""
skipped = []
for tar in final_targets[:]:
if isinstance(tar, str):
can = self.feature_can_autovec(tar)
else: # multiple target
can = all([
self.feature_can_autovec(t)
for t in tar
])
if not can:
final_targets.remove(tar)
skipped.append(tar)
if skipped:
self.dist_log("skip non auto-vectorized features", skipped)
return has_baseline, final_targets, extra_flags
class CCompilerOpt(_Config, _Distutils, _Cache, _CCompiler, _Feature, _Parse):
"""
A helper class for `CCompiler` aims to provide extra build options
to effectively control of compiler optimizations that are directly
related to CPU features.
"""
def __init__(self, ccompiler, cpu_baseline="min", cpu_dispatch="max", cache_path=None):
_Config.__init__(self)
_Distutils.__init__(self, ccompiler)
_Cache.__init__(self, cache_path, self.dist_info(), cpu_baseline, cpu_dispatch)
_CCompiler.__init__(self)
_Feature.__init__(self)
if not self.cc_noopt and self.cc_has_native:
self.dist_log(
"native flag is specified through environment variables. "
"force cpu-baseline='native'"
)
cpu_baseline = "native"
_Parse.__init__(self, cpu_baseline, cpu_dispatch)
# keep the requested features untouched, need it later for report
# and trace purposes
self._requested_baseline = cpu_baseline
self._requested_dispatch = cpu_dispatch
# key is the dispatch-able source and value is a tuple
# contains two items (has_baseline[boolean], dispatched-features[list])
self.sources_status = getattr(self, "sources_status", {})
# every instance should has a separate one
self.cache_private.add("sources_status")
# set it at the end to make sure the cache writing was done after init
# this class
self.hit_cache = hasattr(self, "hit_cache")
def is_cached(self):
"""
Returns True if the class loaded from the cache file
"""
return self.cache_infile and self.hit_cache
def cpu_baseline_flags(self):
"""
Returns a list of final CPU baseline compiler flags
"""
return self.parse_baseline_flags
def cpu_baseline_names(self):
"""
return a list of final CPU baseline feature names
"""
return self.parse_baseline_names
def cpu_dispatch_names(self):
"""
return a list of final CPU dispatch feature names
"""
return self.parse_dispatch_names
def try_dispatch(self, sources, src_dir=None, ccompiler=None, **kwargs):
"""
Compile one or more dispatch-able sources and generates object files,
also generates abstract C config headers and macros that
used later for the final runtime dispatching process.
The mechanism behind it is to takes each source file that specified
in 'sources' and branching it into several files depend on
special configuration statements that must be declared in the
top of each source which contains targeted CPU features,
then it compiles every branched source with the proper compiler flags.
Parameters
----------
sources : list
Must be a list of dispatch-able sources file paths,
and configuration statements must be declared inside
each file.
src_dir : str
Path of parent directory for the generated headers and wrapped sources.
If None(default) the files will generated in-place.
ccompiler: CCompiler
Distutils `CCompiler` instance to be used for compilation.
If None (default), the provided instance during the initialization
will be used instead.
**kwargs : any
Arguments to pass on to the `CCompiler.compile()`
Returns
-------
list : generated object files
Raises
------
CompileError
Raises by `CCompiler.compile()` on compiling failure.
DistutilsError
Some errors during checking the sanity of configuration statements.
See Also
--------
parse_targets :
Parsing the configuration statements of dispatch-able sources.
"""
to_compile = {}
baseline_flags = self.cpu_baseline_flags()
include_dirs = kwargs.setdefault("include_dirs", [])
for src in sources:
output_dir = os.path.dirname(src)
if src_dir:
if not output_dir.startswith(src_dir):
output_dir = os.path.join(src_dir, output_dir)
if output_dir not in include_dirs:
# To allow including the generated config header(*.dispatch.h)
# by the dispatch-able sources
include_dirs.append(output_dir)
has_baseline, targets, extra_flags = self.parse_targets(src)
nochange = self._generate_config(output_dir, src, targets, has_baseline)
for tar in targets:
tar_src = self._wrap_target(output_dir, src, tar, nochange=nochange)
flags = tuple(extra_flags + self.feature_flags(tar))
to_compile.setdefault(flags, []).append(tar_src)
if has_baseline:
flags = tuple(extra_flags + baseline_flags)
to_compile.setdefault(flags, []).append(src)
self.sources_status[src] = (has_baseline, targets)
# For these reasons, the sources are compiled in a separate loop:
# - Gathering all sources with the same flags to benefit from
# the parallel compiling as much as possible.
# - To generate all config headers of the dispatchable sources,
# before the compilation in case if there are dependency relationships
# among them.
objects = []
for flags, srcs in to_compile.items():
objects += self.dist_compile(
srcs, list(flags), ccompiler=ccompiler, **kwargs
)
return objects
def generate_dispatch_header(self, header_path):
"""
Generate the dispatch header which contains the #definitions and headers
for platform-specific instruction-sets for the enabled CPU baseline and
dispatch-able features.
Its highly recommended to take a look at the generated header
also the generated source files via `try_dispatch()`
in order to get the full picture.
"""
self.dist_log("generate CPU dispatch header: (%s)" % header_path)
baseline_names = self.cpu_baseline_names()
dispatch_names = self.cpu_dispatch_names()
baseline_len = len(baseline_names)
dispatch_len = len(dispatch_names)
header_dir = os.path.dirname(header_path)
if not os.path.exists(header_dir):
self.dist_log(
f"dispatch header dir {header_dir} does not exist, creating it",
stderr=True
)
os.makedirs(header_dir)
with open(header_path, 'w') as f:
baseline_calls = ' \\\n'.join([
(
"\t%sWITH_CPU_EXPAND_(MACRO_TO_CALL(%s, __VA_ARGS__))"
) % (self.conf_c_prefix, f)
for f in baseline_names
])
dispatch_calls = ' \\\n'.join([
(
"\t%sWITH_CPU_EXPAND_(MACRO_TO_CALL(%s, __VA_ARGS__))"
) % (self.conf_c_prefix, f)
for f in dispatch_names
])
f.write(textwrap.dedent("""\
/*
* AUTOGENERATED DON'T EDIT
* Please make changes to the code generator (distutils/ccompiler_opt.py)
*/
#define {pfx}WITH_CPU_BASELINE "{baseline_str}"
#define {pfx}WITH_CPU_DISPATCH "{dispatch_str}"
#define {pfx}WITH_CPU_BASELINE_N {baseline_len}
#define {pfx}WITH_CPU_DISPATCH_N {dispatch_len}
#define {pfx}WITH_CPU_EXPAND_(X) X
#define {pfx}WITH_CPU_BASELINE_CALL(MACRO_TO_CALL, ...) \\
{baseline_calls}
#define {pfx}WITH_CPU_DISPATCH_CALL(MACRO_TO_CALL, ...) \\
{dispatch_calls}
""").format(
pfx=self.conf_c_prefix, baseline_str=" ".join(baseline_names),
dispatch_str=" ".join(dispatch_names), baseline_len=baseline_len,
dispatch_len=dispatch_len, baseline_calls=baseline_calls,
dispatch_calls=dispatch_calls
))
baseline_pre = ''
for name in baseline_names:
baseline_pre += self.feature_c_preprocessor(name, tabs=1) + '\n'
dispatch_pre = ''
for name in dispatch_names:
dispatch_pre += textwrap.dedent("""\
#ifdef {pfx}CPU_TARGET_{name}
{pre}
#endif /*{pfx}CPU_TARGET_{name}*/
""").format(
pfx=self.conf_c_prefix_, name=name, pre=self.feature_c_preprocessor(
name, tabs=1
))
f.write(textwrap.dedent("""\
/******* baseline features *******/
{baseline_pre}
/******* dispatch features *******/
{dispatch_pre}
""").format(
pfx=self.conf_c_prefix_, baseline_pre=baseline_pre,
dispatch_pre=dispatch_pre
))
def report(self, full=False):
report = []
platform_rows = []
baseline_rows = []
dispatch_rows = []
report.append(("Platform", platform_rows))
report.append(("", ""))
report.append(("CPU baseline", baseline_rows))
report.append(("", ""))
report.append(("CPU dispatch", dispatch_rows))
########## platform ##########
platform_rows.append(("Architecture", (
"unsupported" if self.cc_on_noarch else self.cc_march)
))
platform_rows.append(("Compiler", (
"unix-like" if self.cc_is_nocc else self.cc_name)
))
########## baseline ##########
if self.cc_noopt:
baseline_rows.append(("Requested", "optimization disabled"))
else:
baseline_rows.append(("Requested", repr(self._requested_baseline)))
baseline_names = self.cpu_baseline_names()
baseline_rows.append((
"Enabled", (' '.join(baseline_names) if baseline_names else "none")
))
baseline_flags = self.cpu_baseline_flags()
baseline_rows.append((
"Flags", (' '.join(baseline_flags) if baseline_flags else "none")
))
extra_checks = []
for name in baseline_names:
extra_checks += self.feature_extra_checks(name)
baseline_rows.append((
"Extra checks", (' '.join(extra_checks) if extra_checks else "none")
))
########## dispatch ##########
if self.cc_noopt:
baseline_rows.append(("Requested", "optimization disabled"))
else:
dispatch_rows.append(("Requested", repr(self._requested_dispatch)))
dispatch_names = self.cpu_dispatch_names()
dispatch_rows.append((
"Enabled", (' '.join(dispatch_names) if dispatch_names else "none")
))
########## Generated ##########
# TODO:
# - collect object names from 'try_dispatch()'
# then get size of each object and printed
# - give more details about the features that not
# generated due compiler support
# - find a better output's design.
#
target_sources = {}
for source, (_, targets) in self.sources_status.items():
for tar in targets:
target_sources.setdefault(tar, []).append(source)
if not full or not target_sources:
generated = ""
for tar in self.feature_sorted(target_sources):
sources = target_sources[tar]
name = tar if isinstance(tar, str) else '(%s)' % ' '.join(tar)
generated += name + "[%d] " % len(sources)
dispatch_rows.append(("Generated", generated[:-1] if generated else "none"))
else:
dispatch_rows.append(("Generated", ''))
for tar in self.feature_sorted(target_sources):
sources = target_sources[tar]
pretty_name = tar if isinstance(tar, str) else '(%s)' % ' '.join(tar)
flags = ' '.join(self.feature_flags(tar))
implies = ' '.join(self.feature_sorted(self.feature_implies(tar)))
detect = ' '.join(self.feature_detect(tar))
extra_checks = []
for name in ((tar,) if isinstance(tar, str) else tar):
extra_checks += self.feature_extra_checks(name)
extra_checks = (' '.join(extra_checks) if extra_checks else "none")
dispatch_rows.append(('', ''))
dispatch_rows.append((pretty_name, implies))
dispatch_rows.append(("Flags", flags))
dispatch_rows.append(("Extra checks", extra_checks))
dispatch_rows.append(("Detect", detect))
for src in sources:
dispatch_rows.append(("", src))
###############################
# TODO: add support for 'markdown' format
text = []
secs_len = [len(secs) for secs, _ in report]
cols_len = [len(col) for _, rows in report for col, _ in rows]
tab = ' ' * 2
pad = max(max(secs_len), max(cols_len))
for sec, rows in report:
if not sec:
text.append("") # empty line
continue
sec += ' ' * (pad - len(sec))
text.append(sec + tab + ': ')
for col, val in rows:
col += ' ' * (pad - len(col))
text.append(tab + col + ': ' + val)
return '\n'.join(text)
def _wrap_target(self, output_dir, dispatch_src, target, nochange=False):
assert(isinstance(target, (str, tuple)))
if isinstance(target, str):
ext_name = target_name = target
else:
# multi-target
ext_name = '.'.join(target)
target_name = '__'.join(target)
wrap_path = os.path.join(output_dir, os.path.basename(dispatch_src))
wrap_path = "{0}.{2}{1}".format(*os.path.splitext(wrap_path), ext_name.lower())
if nochange and os.path.exists(wrap_path):
return wrap_path
self.dist_log("wrap dispatch-able target -> ", wrap_path)
# sorting for readability
features = self.feature_sorted(self.feature_implies_c(target))
target_join = "#define %sCPU_TARGET_" % self.conf_c_prefix_
target_defs = [target_join + f for f in features]
target_defs = '\n'.join(target_defs)
with open(wrap_path, "w") as fd:
fd.write(textwrap.dedent("""\
/**
* AUTOGENERATED DON'T EDIT
* Please make changes to the code generator \
(distutils/ccompiler_opt.py)
*/
#define {pfx}CPU_TARGET_MODE
#define {pfx}CPU_TARGET_CURRENT {target_name}
{target_defs}
#include "{path}"
""").format(
pfx=self.conf_c_prefix_, target_name=target_name,
path=os.path.abspath(dispatch_src), target_defs=target_defs
))
return wrap_path
def _generate_config(self, output_dir, dispatch_src, targets, has_baseline=False):
config_path = os.path.basename(dispatch_src)
config_path = os.path.splitext(config_path)[0] + '.h'
config_path = os.path.join(output_dir, config_path)
# check if targets didn't change to avoid recompiling
cache_hash = self.cache_hash(targets, has_baseline)
try:
with open(config_path) as f:
last_hash = f.readline().split("cache_hash:")
if len(last_hash) == 2 and int(last_hash[1]) == cache_hash:
return True
except IOError:
pass
self.dist_log("generate dispatched config -> ", config_path)
dispatch_calls = []
for tar in targets:
if isinstance(tar, str):
target_name = tar
else: # multi target
target_name = '__'.join([t for t in tar])
req_detect = self.feature_detect(tar)
req_detect = '&&'.join([
"CHK(%s)" % f for f in req_detect
])
dispatch_calls.append(
"\t%sCPU_DISPATCH_EXPAND_(CB((%s), %s, __VA_ARGS__))" % (
self.conf_c_prefix_, req_detect, target_name
))
dispatch_calls = ' \\\n'.join(dispatch_calls)
if has_baseline:
baseline_calls = (
"\t%sCPU_DISPATCH_EXPAND_(CB(__VA_ARGS__))"
) % self.conf_c_prefix_
else:
baseline_calls = ''
with open(config_path, "w") as fd:
fd.write(textwrap.dedent("""\
// cache_hash:{cache_hash}
/**
* AUTOGENERATED DON'T EDIT
* Please make changes to the code generator (distutils/ccompiler_opt.py)
*/
#ifndef {pfx}CPU_DISPATCH_EXPAND_
#define {pfx}CPU_DISPATCH_EXPAND_(X) X
#endif
#undef {pfx}CPU_DISPATCH_BASELINE_CALL
#undef {pfx}CPU_DISPATCH_CALL
#define {pfx}CPU_DISPATCH_BASELINE_CALL(CB, ...) \\
{baseline_calls}
#define {pfx}CPU_DISPATCH_CALL(CHK, CB, ...) \\
{dispatch_calls}
""").format(
pfx=self.conf_c_prefix_, baseline_calls=baseline_calls,
dispatch_calls=dispatch_calls, cache_hash=cache_hash
))
return False
def new_ccompiler_opt(compiler, dispatch_hpath, **kwargs):
"""
Create a new instance of 'CCompilerOpt' and generate the dispatch header
which contains the #definitions and headers of platform-specific instruction-sets for
the enabled CPU baseline and dispatch-able features.
Parameters
----------
compiler : CCompiler instance
dispatch_hpath : str
path of the dispatch header
**kwargs: passed as-is to `CCompilerOpt(...)`
Returns
-------
new instance of CCompilerOpt
"""
opt = CCompilerOpt(compiler, **kwargs)
if not os.path.exists(dispatch_hpath) or not opt.is_cached():
opt.generate_dispatch_header(dispatch_hpath)
return opt
|