pypy /rpython/jit/backend/x86/assembler.py

Language Python Lines 2665
MD5 Hash 8f12c41ad04bff39e6635c34d00d8a4f Estimated Cost $56,733 (why?)
Repository https://bitbucket.org/pypy/pypy/ View Raw File View Project SPDX
   1
   2
   3
   4
   5
   6
   7
   8
   9
  10
  11
  12
  13
  14
  15
  16
  17
  18
  19
  20
  21
  22
  23
  24
  25
  26
  27
  28
  29
  30
  31
  32
  33
  34
  35
  36
  37
  38
  39
  40
  41
  42
  43
  44
  45
  46
  47
  48
  49
  50
  51
  52
  53
  54
  55
  56
  57
  58
  59
  60
  61
  62
  63
  64
  65
  66
  67
  68
  69
  70
  71
  72
  73
  74
  75
  76
  77
  78
  79
  80
  81
  82
  83
  84
  85
  86
  87
  88
  89
  90
  91
  92
  93
  94
  95
  96
  97
  98
  99
 100
 101
 102
 103
 104
 105
 106
 107
 108
 109
 110
 111
 112
 113
 114
 115
 116
 117
 118
 119
 120
 121
 122
 123
 124
 125
 126
 127
 128
 129
 130
 131
 132
 133
 134
 135
 136
 137
 138
 139
 140
 141
 142
 143
 144
 145
 146
 147
 148
 149
 150
 151
 152
 153
 154
 155
 156
 157
 158
 159
 160
 161
 162
 163
 164
 165
 166
 167
 168
 169
 170
 171
 172
 173
 174
 175
 176
 177
 178
 179
 180
 181
 182
 183
 184
 185
 186
 187
 188
 189
 190
 191
 192
 193
 194
 195
 196
 197
 198
 199
 200
 201
 202
 203
 204
 205
 206
 207
 208
 209
 210
 211
 212
 213
 214
 215
 216
 217
 218
 219
 220
 221
 222
 223
 224
 225
 226
 227
 228
 229
 230
 231
 232
 233
 234
 235
 236
 237
 238
 239
 240
 241
 242
 243
 244
 245
 246
 247
 248
 249
 250
 251
 252
 253
 254
 255
 256
 257
 258
 259
 260
 261
 262
 263
 264
 265
 266
 267
 268
 269
 270
 271
 272
 273
 274
 275
 276
 277
 278
 279
 280
 281
 282
 283
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293
 294
 295
 296
 297
 298
 299
 300
 301
 302
 303
 304
 305
 306
 307
 308
 309
 310
 311
 312
 313
 314
 315
 316
 317
 318
 319
 320
 321
 322
 323
 324
 325
 326
 327
 328
 329
 330
 331
 332
 333
 334
 335
 336
 337
 338
 339
 340
 341
 342
 343
 344
 345
 346
 347
 348
 349
 350
 351
 352
 353
 354
 355
 356
 357
 358
 359
 360
 361
 362
 363
 364
 365
 366
 367
 368
 369
 370
 371
 372
 373
 374
 375
 376
 377
 378
 379
 380
 381
 382
 383
 384
 385
 386
 387
 388
 389
 390
 391
 392
 393
 394
 395
 396
 397
 398
 399
 400
 401
 402
 403
 404
 405
 406
 407
 408
 409
 410
 411
 412
 413
 414
 415
 416
 417
 418
 419
 420
 421
 422
 423
 424
 425
 426
 427
 428
 429
 430
 431
 432
 433
 434
 435
 436
 437
 438
 439
 440
 441
 442
 443
 444
 445
 446
 447
 448
 449
 450
 451
 452
 453
 454
 455
 456
 457
 458
 459
 460
 461
 462
 463
 464
 465
 466
 467
 468
 469
 470
 471
 472
 473
 474
 475
 476
 477
 478
 479
 480
 481
 482
 483
 484
 485
 486
 487
 488
 489
 490
 491
 492
 493
 494
 495
 496
 497
 498
 499
 500
 501
 502
 503
 504
 505
 506
 507
 508
 509
 510
 511
 512
 513
 514
 515
 516
 517
 518
 519
 520
 521
 522
 523
 524
 525
 526
 527
 528
 529
 530
 531
 532
 533
 534
 535
 536
 537
 538
 539
 540
 541
 542
 543
 544
 545
 546
 547
 548
 549
 550
 551
 552
 553
 554
 555
 556
 557
 558
 559
 560
 561
 562
 563
 564
 565
 566
 567
 568
 569
 570
 571
 572
 573
 574
 575
 576
 577
 578
 579
 580
 581
 582
 583
 584
 585
 586
 587
 588
 589
 590
 591
 592
 593
 594
 595
 596
 597
 598
 599
 600
 601
 602
 603
 604
 605
 606
 607
 608
 609
 610
 611
 612
 613
 614
 615
 616
 617
 618
 619
 620
 621
 622
 623
 624
 625
 626
 627
 628
 629
 630
 631
 632
 633
 634
 635
 636
 637
 638
 639
 640
 641
 642
 643
 644
 645
 646
 647
 648
 649
 650
 651
 652
 653
 654
 655
 656
 657
 658
 659
 660
 661
 662
 663
 664
 665
 666
 667
 668
 669
 670
 671
 672
 673
 674
 675
 676
 677
 678
 679
 680
 681
 682
 683
 684
 685
 686
 687
 688
 689
 690
 691
 692
 693
 694
 695
 696
 697
 698
 699
 700
 701
 702
 703
 704
 705
 706
 707
 708
 709
 710
 711
 712
 713
 714
 715
 716
 717
 718
 719
 720
 721
 722
 723
 724
 725
 726
 727
 728
 729
 730
 731
 732
 733
 734
 735
 736
 737
 738
 739
 740
 741
 742
 743
 744
 745
 746
 747
 748
 749
 750
 751
 752
 753
 754
 755
 756
 757
 758
 759
 760
 761
 762
 763
 764
 765
 766
 767
 768
 769
 770
 771
 772
 773
 774
 775
 776
 777
 778
 779
 780
 781
 782
 783
 784
 785
 786
 787
 788
 789
 790
 791
 792
 793
 794
 795
 796
 797
 798
 799
 800
 801
 802
 803
 804
 805
 806
 807
 808
 809
 810
 811
 812
 813
 814
 815
 816
 817
 818
 819
 820
 821
 822
 823
 824
 825
 826
 827
 828
 829
 830
 831
 832
 833
 834
 835
 836
 837
 838
 839
 840
 841
 842
 843
 844
 845
 846
 847
 848
 849
 850
 851
 852
 853
 854
 855
 856
 857
 858
 859
 860
 861
 862
 863
 864
 865
 866
 867
 868
 869
 870
 871
 872
 873
 874
 875
 876
 877
 878
 879
 880
 881
 882
 883
 884
 885
 886
 887
 888
 889
 890
 891
 892
 893
 894
 895
 896
 897
 898
 899
 900
 901
 902
 903
 904
 905
 906
 907
 908
 909
 910
 911
 912
 913
 914
 915
 916
 917
 918
 919
 920
 921
 922
 923
 924
 925
 926
 927
 928
 929
 930
 931
 932
 933
 934
 935
 936
 937
 938
 939
 940
 941
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978
 979
 980
 981
 982
 983
 984
 985
 986
 987
 988
 989
 990
 991
 992
 993
 994
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
import sys
import os
import py

from rpython.jit.backend.llsupport import symbolic, jitframe, rewrite
from rpython.jit.backend.llsupport.assembler import (GuardToken, BaseAssembler, debug_bridge)
from rpython.jit.backend.llsupport.asmmemmgr import MachineDataBlockWrapper
from rpython.jit.backend.llsupport.gcmap import allocate_gcmap
from rpython.jit.metainterp.history import (Const, VOID, ConstInt)
from rpython.jit.metainterp.history import AbstractFailDescr, INT, REF, FLOAT
from rpython.jit.metainterp.compile import ResumeGuardDescr
from rpython.rlib.rjitlog import rjitlog as jl
from rpython.rtyper.lltypesystem import lltype, rffi, rstr, llmemory
from rpython.rtyper.lltypesystem.lloperation import llop
from rpython.rtyper.annlowlevel import cast_instance_to_gcref
from rpython.rtyper import rclass
from rpython.rlib.jit import AsmInfo
from rpython.jit.backend.model import CompiledLoopToken
from rpython.jit.backend.x86.regalloc import (RegAlloc, get_ebp_ofs,
    gpr_reg_mgr_cls, xmm_reg_mgr_cls)
from rpython.jit.backend.llsupport.regalloc import (get_scale, valid_addressing_size)
from rpython.jit.backend.x86.arch import (FRAME_FIXED_SIZE, WORD, IS_X86_64,
                                       JITFRAME_FIXED_SIZE, IS_X86_32,
                                       PASS_ON_MY_FRAME, THREADLOCAL_OFS,
                                       DEFAULT_FRAME_BYTES)
from rpython.jit.backend.x86.regloc import (eax, ecx, edx, ebx, esp, ebp, esi,
    xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, r8, r9, r10, r11, edi,
    r12, r13, r14, r15, X86_64_SCRATCH_REG, X86_64_XMM_SCRATCH_REG,
    RegLoc, FrameLoc, ConstFloatLoc, ImmedLoc, AddressLoc, imm,
    imm0, imm1, FloatImmedLoc, RawEbpLoc, RawEspLoc)
from rpython.rlib.objectmodel import we_are_translated
from rpython.jit.backend.x86 import rx86, codebuf, callbuilder
from rpython.jit.backend.x86.vector_ext import VectorAssemblerMixin
from rpython.jit.backend.x86.callbuilder import follow_jump
from rpython.jit.metainterp.resoperation import rop
from rpython.jit.backend.x86 import support
from rpython.rlib.debug import debug_print, debug_start, debug_stop
from rpython.rlib import rgc
from rpython.jit.codewriter.effectinfo import EffectInfo
from rpython.jit.codewriter import longlong
from rpython.rlib.rarithmetic import intmask, r_uint
from rpython.rlib.objectmodel import compute_unique_id


class Assembler386(BaseAssembler, VectorAssemblerMixin):
    _regalloc = None
    _output_loop_log = None
    _second_tmp_reg = ecx

    DEBUG_FRAME_DEPTH = False

    def __init__(self, cpu, translate_support_code=False):
        BaseAssembler.__init__(self, cpu, translate_support_code)
        self.verbose = False
        self.loop_run_counters = []
        self.float_const_neg_addr = 0
        self.float_const_abs_addr = 0
        self.single_float_const_neg_addr = 0
        self.single_float_const_abs_addr = 0
        self.expand_byte_mask_addr = 0
        self.malloc_slowpath = 0
        self.malloc_slowpath_varsize = 0
        self.wb_slowpath = [0, 0, 0, 0, 0]
        self.setup_failure_recovery()
        self.datablockwrapper = None
        self.stack_check_slowpath = 0
        self.propagate_exception_path = 0
        self.teardown()

    def setup_once(self):
        BaseAssembler.setup_once(self)
        if self.cpu.supports_floats:
            support.ensure_sse2_floats()
            self._build_float_constants()

    def setup(self, looptoken):
        BaseAssembler.setup(self, looptoken)
        assert self.memcpy_addr != 0, "setup_once() not called?"
        self.current_clt = looptoken.compiled_loop_token
        self.pending_guard_tokens = []
        if WORD == 8:
            self.pending_memoryerror_trampoline_from = []
            self.error_trampoline_64 = 0
        self.mc = codebuf.MachineCodeBlockWrapper()
        #assert self.datablockwrapper is None --- but obscure case
        # possible, e.g. getting MemoryError and continuing
        allblocks = self.get_asmmemmgr_blocks(looptoken)
        self.datablockwrapper = MachineDataBlockWrapper(self.cpu.asmmemmgr,
                                                        allblocks)
        self.target_tokens_currently_compiling = {}
        self.frame_depth_to_patch = []

    def teardown(self):
        self.pending_guard_tokens = None
        if WORD == 8:
            self.pending_memoryerror_trampoline_from = None
        self.mc = None
        self.current_clt = None

    def _build_float_constants(self):
        # 0x80000000000000008000000000000000
        neg_const = '\x00\x00\x00\x00\x00\x00\x00\x80\x00\x00\x00\x00\x00\x00\x00\x80'
        # 0x7FFFFFFFFFFFFFFF7FFFFFFFFFFFFFFF
        abs_const = '\xFF\xFF\xFF\xFF\xFF\xFF\xFF\x7F\xFF\xFF\xFF\xFF\xFF\xFF\xFF\x7F'
        # 0x7FFFFFFF7FFFFFFF7FFFFFFF7FFFFFFF
        single_abs_const = '\xFF\xFF\xFF\x7F\xFF\xFF\xFF\x7F\xFF\xFF\xFF\x7F\xFF\xFF\xFF\x7F'
        # 0x80000000800000008000000080000000
        single_neg_const = '\x00\x00\x00\x80\x00\x00\x00\x80\x00\x00\x00\x80\x00\x00\x00\x80'
        zero_const = '\x00' * 16
        #
        data = neg_const + abs_const + \
               single_neg_const + single_abs_const + \
               zero_const
        datablockwrapper = MachineDataBlockWrapper(self.cpu.asmmemmgr, [])
        float_constants = datablockwrapper.malloc_aligned(len(data), alignment=16)
        datablockwrapper.done()
        addr = rffi.cast(rffi.CArrayPtr(lltype.Char), float_constants)
        for i in range(len(data)):
            addr[i] = data[i]
        self.float_const_neg_addr = float_constants
        self.float_const_abs_addr = float_constants + 16
        self.single_float_const_neg_addr = float_constants + 32
        self.single_float_const_abs_addr = float_constants + 48
        self.expand_byte_mask_addr = float_constants + 64

    def set_extra_stack_depth(self, mc, value):
        if self._is_asmgcc():
            extra_ofs = self.cpu.get_ofs_of_frame_field('jf_extra_stack_depth')
            mc.MOV_bi(extra_ofs, value)

    def build_frame_realloc_slowpath(self):
        mc = codebuf.MachineCodeBlockWrapper()
        self._push_all_regs_to_frame(mc, [], self.cpu.supports_floats)
        # the caller already did push_gcmap(store=True)

        if IS_X86_64:
            mc.MOV_rs(esi.value, WORD*2)
            # push first arg
            mc.MOV_rr(edi.value, ebp.value)
            align = callbuilder.align_stack_words(1)
            mc.SUB_ri(esp.value, (align - 1) * WORD)
        else:
            align = callbuilder.align_stack_words(3)
            mc.MOV_rs(eax.value, WORD * 2)
            mc.SUB_ri(esp.value, (align - 1) * WORD)
            mc.MOV_sr(WORD, eax.value)
            mc.MOV_sr(0, ebp.value)
        # align

        self.set_extra_stack_depth(mc, align * WORD)
        self._store_and_reset_exception(mc, None, ebx, ecx)

        mc.CALL(imm(self.cpu.realloc_frame))
        mc.MOV_rr(ebp.value, eax.value)
        self._restore_exception(mc, None, ebx, ecx)
        mc.ADD_ri(esp.value, (align - 1) * WORD)
        self.set_extra_stack_depth(mc, 0)

        gcrootmap = self.cpu.gc_ll_descr.gcrootmap
        if gcrootmap and gcrootmap.is_shadow_stack:
            self._load_shadowstack_top_in_ebx(mc, gcrootmap)
            mc.MOV_mr((ebx.value, -WORD), eax.value)

        self.pop_gcmap(mc)   # cancel the push_gcmap(store=True) in the caller
        self._pop_all_regs_from_frame(mc, [], self.cpu.supports_floats)
        mc.RET()
        self._frame_realloc_slowpath = mc.materialize(self.cpu, [])

    def _build_cond_call_slowpath(self, supports_floats, callee_only):
        """ This builds a general call slowpath, for whatever call happens to
        come.
        """
        mc = codebuf.MachineCodeBlockWrapper()
        # copy registers to the frame, with the exception of the
        # 'cond_call_register_arguments' and eax, because these have already
        # been saved by the caller.  Note that this is not symmetrical:
        # these 5 registers are saved by the caller but restored here at
        # the end of this function.
        self._push_all_regs_to_frame(mc, cond_call_register_arguments + [eax],
                                     supports_floats, callee_only)
        # the caller already did push_gcmap(store=True)
        if IS_X86_64:
            mc.SUB(esp, imm(WORD))     # alignment
            self.set_extra_stack_depth(mc, 2 * WORD)
            # the arguments are already in the correct registers
        else:
            # we want space for 4 arguments + call + alignment
            mc.SUB(esp, imm(WORD * 7))
            self.set_extra_stack_depth(mc, 8 * WORD)
            # store the arguments at the correct place in the stack
            for i in range(4):
                mc.MOV_sr(i * WORD, cond_call_register_arguments[i].value)
        mc.CALL(eax)
        self._reload_frame_if_necessary(mc)
        if IS_X86_64:
            mc.ADD(esp, imm(WORD))
        else:
            mc.ADD(esp, imm(WORD * 7))
        self.set_extra_stack_depth(mc, 0)
        self.pop_gcmap(mc)   # cancel the push_gcmap(store=True) in the caller
        self._pop_all_regs_from_frame(mc, [], supports_floats, callee_only)
        mc.RET()
        return mc.materialize(self.cpu, [])

    def _build_malloc_slowpath(self, kind):
        """ While arriving on slowpath, we have a gcpattern on stack 0.
        The arguments are passed in ecx and edx, as follows:

        kind == 'fixed': nursery_head in ecx and the size in (edx - ecx).

        kind == 'str/unicode': length of the string to allocate in edx.

        kind == 'var': length to allocate in edx, tid in ecx,
                       and itemsize in the stack 1 (position esp+WORD).

        This function must preserve all registers apart from ecx and edx.
        """
        assert kind in ['fixed', 'str', 'unicode', 'var']
        mc = codebuf.MachineCodeBlockWrapper()
        self._push_all_regs_to_frame(mc, [ecx, edx], self.cpu.supports_floats)
        # the caller already did push_gcmap(store=True)
        #
        if kind == 'fixed':
            addr = self.cpu.gc_ll_descr.get_malloc_slowpath_addr()
        elif kind == 'str':
            addr = self.cpu.gc_ll_descr.get_malloc_fn_addr('malloc_str')
        elif kind == 'unicode':
            addr = self.cpu.gc_ll_descr.get_malloc_fn_addr('malloc_unicode')
        else:
            addr = self.cpu.gc_ll_descr.get_malloc_slowpath_array_addr()
        mc.SUB_ri(esp.value, 16 - WORD)  # restore 16-byte alignment
        # magically, the above is enough on X86_32 to reserve 3 stack places
        if kind == 'fixed':
            mc.SUB_rr(edx.value, ecx.value) # compute the size we want
            if IS_X86_32:
                mc.MOV_sr(0, edx.value)     # store the length
                if hasattr(self.cpu.gc_ll_descr, 'passes_frame'):
                    mc.MOV_sr(WORD, ebp.value)        # for tests only
            else:
                mc.MOV_rr(edi.value, edx.value)   # length argument
                if hasattr(self.cpu.gc_ll_descr, 'passes_frame'):
                    mc.MOV_rr(esi.value, ebp.value)   # for tests only
        elif kind == 'str' or kind == 'unicode':
            if IS_X86_32:
                # stack layout: [---][---][---][ret].. with 3 free stack places
                mc.MOV_sr(0, edx.value)     # store the length
            elif IS_X86_64:
                mc.MOV_rr(edi.value, edx.value)   # length argument
        else:
            if IS_X86_32:
                # stack layout: [---][---][---][ret][gcmap][itemsize]...
                mc.MOV_sr(WORD * 2, edx.value)  # store the length
                mc.MOV_sr(WORD * 1, ecx.value)  # store the tid
                mc.MOV_rs(edx.value, WORD * 5)  # load the itemsize
                mc.MOV_sr(WORD * 0, edx.value)  # store the itemsize
            else:
                # stack layout: [---][ret][gcmap][itemsize]...
                # (already in edx)              # length
                mc.MOV_rr(esi.value, ecx.value) # tid
                mc.MOV_rs(edi.value, WORD * 3)  # load the itemsize
        self.set_extra_stack_depth(mc, 16)
        mc.CALL(imm(follow_jump(addr)))
        self._reload_frame_if_necessary(mc)
        mc.ADD_ri(esp.value, 16 - WORD)
        self.set_extra_stack_depth(mc, 0)
        #
        mc.TEST_rr(eax.value, eax.value)
        mc.J_il(rx86.Conditions['Z'], 0xfffff) # patched later
        jz_location = mc.get_relative_pos()
        mc.MOV_rr(ecx.value, eax.value)
        #
        nursery_free_adr = self.cpu.gc_ll_descr.get_nursery_free_addr()
        self._pop_all_regs_from_frame(mc, [ecx, edx], self.cpu.supports_floats)
        mc.MOV(edx, heap(nursery_free_adr))   # load this in EDX
        self.pop_gcmap(mc)   # push_gcmap(store=True) done by the caller
        mc.RET()
        #
        # If the slowpath malloc failed, we raise a MemoryError that
        # always interrupts the current loop, as a "good enough"
        # approximation.  We have to adjust the esp a little, to point to
        # the correct "ret" arg
        offset = mc.get_relative_pos() - jz_location
        mc.overwrite32(jz_location-4, offset)
        # From now on this function is basically "merged" with
        # its caller and so contains DEFAULT_FRAME_BYTES bytes
        # plus my own return address, which we'll ignore next
        mc.force_frame_size(DEFAULT_FRAME_BYTES + WORD)
        mc.ADD_ri(esp.value, WORD)
        mc.JMP(imm(self.propagate_exception_path))
        #
        rawstart = mc.materialize(self.cpu, [])
        return rawstart

    def _build_propagate_exception_path(self):
        self.mc = codebuf.MachineCodeBlockWrapper()
        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        #
        # read and reset the current exception

        self._store_and_reset_exception(self.mc, eax)
        ofs = self.cpu.get_ofs_of_frame_field('jf_guard_exc')
        self.mc.MOV_br(ofs, eax.value)
        propagate_exception_descr = rffi.cast(lltype.Signed,
                  cast_instance_to_gcref(self.cpu.propagate_exception_descr))
        ofs = self.cpu.get_ofs_of_frame_field('jf_descr')
        self.mc.MOV(RawEbpLoc(ofs), imm(propagate_exception_descr))
        #
        self._call_footer()
        rawstart = self.mc.materialize(self.cpu, [])
        self.propagate_exception_path = rawstart
        self.mc = None

    def _build_stack_check_slowpath(self):
        _, _, slowpathaddr = self.cpu.insert_stack_check()
        if slowpathaddr == 0 or not self.cpu.propagate_exception_descr:
            return      # no stack check (for tests, or non-translated)
        #
        # make a regular function that is called from a point near the start
        # of an assembler function (after it adjusts the stack and saves
        # registers).
        mc = codebuf.MachineCodeBlockWrapper()
        #
        if IS_X86_64:
            mc.MOV_rr(edi.value, esp.value)
            mc.SUB_ri(esp.value, WORD)   # alignment
        #
        if IS_X86_32:
            mc.SUB_ri(esp.value, 2*WORD) # alignment
            mc.PUSH_r(esp.value)
        #
        # esp is now aligned to a multiple of 16 again
        mc.CALL(imm(follow_jump(slowpathaddr)))
        #
        if IS_X86_32:
            mc.ADD_ri(esp.value, 3*WORD)    # alignment
        else:
            mc.ADD_ri(esp.value, WORD)
        #
        mc.MOV(eax, heap(self.cpu.pos_exception()))
        mc.TEST_rr(eax.value, eax.value)
        mc.J_il8(rx86.Conditions['NZ'], 0)
        jnz_location = mc.get_relative_pos()
        #
        mc.RET()
        #
        # patch the JNZ above
        offset = mc.get_relative_pos() - jnz_location
        assert 0 < offset <= 127
        mc.overwrite(jnz_location-1, chr(offset))
        # From now on this function is basically "merged" with
        # its caller and so contains DEFAULT_FRAME_BYTES bytes
        # plus my own return address, which we'll ignore next
        mc.force_frame_size(DEFAULT_FRAME_BYTES + WORD)
        mc.ADD_ri(esp.value, WORD)
        mc.JMP(imm(self.propagate_exception_path))
        #
        rawstart = mc.materialize(self.cpu, [])
        self.stack_check_slowpath = rawstart

    def _build_wb_slowpath(self, withcards, withfloats=False, for_frame=False):
        descr = self.cpu.gc_ll_descr.write_barrier_descr
        exc0, exc1 = None, None
        if descr is None:
            return
        if not withcards:
            func = descr.get_write_barrier_fn(self.cpu)
        else:
            if descr.jit_wb_cards_set == 0:
                return
            func = descr.get_write_barrier_from_array_fn(self.cpu)
            if func == 0:
                return
        #
        # This builds a helper function called from the slow path of
        # write barriers.  It must save all registers, and optionally
        # all XMM registers.  It takes a single argument just pushed
        # on the stack even on X86_64.  It must restore stack alignment
        # accordingly.
        mc = codebuf.MachineCodeBlockWrapper()
        #
        if not for_frame:
            self._push_all_regs_to_frame(mc, [], withfloats, callee_only=True)
            if IS_X86_32:
                # we have 2 extra words on stack for retval and we pass 1 extra
                # arg, so we need to substract 2 words
                mc.SUB_ri(esp.value, 2 * WORD)
                mc.MOV_rs(eax.value, 3 * WORD) # 2 + 1
                mc.MOV_sr(0, eax.value)
            else:
                mc.MOV_rs(edi.value, WORD)
        else:
            # NOTE: don't save registers on the jitframe here!
            # It might override already-saved values that will be
            # restored later...
            #
            # This 'for_frame' version is called after a CALL.  It does not
            # need to save many registers: the registers that are anyway
            # destroyed by the call can be ignored (volatiles), and the
            # non-volatile registers won't be changed here.  It only needs
            # to save eax, maybe edx, and xmm0 (possible results of the call)
            # and two more non-volatile registers (used to store the RPython
            # exception that occurred in the CALL, if any).
            assert not withcards
            # we have one word to align
            mc.SUB_ri(esp.value, 7 * WORD) # align and reserve some space
            mc.MOV_sr(WORD, eax.value) # save for later
            if self.cpu.supports_floats:
                mc.MOVSD_sx(2 * WORD, xmm0.value)   # 32-bit: also 3 * WORD
            if IS_X86_32:
                mc.MOV_sr(4 * WORD, edx.value)
                mc.MOV_sr(0, ebp.value)
                exc0, exc1 = esi, edi
            else:
                mc.MOV_rr(edi.value, ebp.value)
                exc0, exc1 = ebx, r12
            mc.MOV(RawEspLoc(WORD * 5, REF), exc0)
            mc.MOV(RawEspLoc(WORD * 6, INT), exc1)
            # note that it's safe to store the exception in register,
            # since the call to write barrier can't collect
            # (and this is assumed a bit left and right here, like lack
            # of _reload_frame_if_necessary)
            self._store_and_reset_exception(mc, exc0, exc1)

        mc.CALL(imm(func))
        #
        if withcards:
            # A final TEST8 before the RET, for the caller.  Careful to
            # not follow this instruction with another one that changes
            # the status of the CPU flags!
            if IS_X86_32:
                mc.MOV_rs(eax.value, 3*WORD)
            else:
                mc.MOV_rs(eax.value, WORD)
            mc.TEST8(addr_add_const(eax, descr.jit_wb_if_flag_byteofs),
                     imm(-0x80))
        #

        if not for_frame:
            if IS_X86_32:
                # ADD touches CPU flags
                mc.LEA_rs(esp.value, 2 * WORD)
            self._pop_all_regs_from_frame(mc, [], withfloats, callee_only=True)
            mc.RET16_i(WORD)
            # Note that wb_slowpath[0..3] end with a RET16_i, which must be
            # taken care of in the caller by stack_frame_size_delta(-WORD)
        else:
            if IS_X86_32:
                mc.MOV_rs(edx.value, 4 * WORD)
            if self.cpu.supports_floats:
                mc.MOVSD_xs(xmm0.value, 2 * WORD)
            mc.MOV_rs(eax.value, WORD) # restore
            self._restore_exception(mc, exc0, exc1)
            mc.MOV(exc0, RawEspLoc(WORD * 5, REF))
            mc.MOV(exc1, RawEspLoc(WORD * 6, INT))
            mc.LEA_rs(esp.value, 7 * WORD)
            mc.RET()

        rawstart = mc.materialize(self.cpu, [])
        if for_frame:
            self.wb_slowpath[4] = rawstart
        else:
            self.wb_slowpath[withcards + 2 * withfloats] = rawstart

    @rgc.no_release_gil
    def assemble_loop(self, jd_id, unique_id, logger, loopname, inputargs,
                      operations, looptoken, log):
        '''adds the following attributes to looptoken:
               _ll_function_addr    (address of the generated func, as an int)
               _ll_loop_code       (debug: addr of the start of the ResOps)
               _x86_fullsize        (debug: full size including failure)
        '''
        # XXX this function is too longish and contains some code
        # duplication with assemble_bridge().  Also, we should think
        # about not storing on 'self' attributes that will live only
        # for the duration of compiling one loop or a one bridge.
        clt = CompiledLoopToken(self.cpu, looptoken.number)
        looptoken.compiled_loop_token = clt
        clt._debug_nbargs = len(inputargs)
        if not we_are_translated():
            # Arguments should be unique
            assert len(set(inputargs)) == len(inputargs)

        self.setup(looptoken)
        if self.cpu.HAS_CODEMAP:
            self.codemap_builder.enter_portal_frame(jd_id, unique_id,
                                                    self.mc.get_relative_pos())
        frame_info = self.datablockwrapper.malloc_aligned(
            jitframe.JITFRAMEINFO_SIZE, alignment=WORD)
        clt.frame_info = rffi.cast(jitframe.JITFRAMEINFOPTR, frame_info)
        clt.frame_info.clear() # for now

        if log:
            number = looptoken.number
            operations = self._inject_debugging_code(looptoken, operations,
                                                     'e', number)

        regalloc = RegAlloc(self, self.cpu.translate_support_code)
        #
        allgcrefs = []
        operations = regalloc.prepare_loop(inputargs, operations,
                                           looptoken, allgcrefs)
        self.reserve_gcref_table(allgcrefs)
        functionpos = self.mc.get_relative_pos()
        self._call_header_with_stack_check()
        self._check_frame_depth_debug(self.mc)
        looppos = self.mc.get_relative_pos()
        frame_depth_no_fixed_size = self._assemble(regalloc, inputargs,
                                                   operations)
        self.update_frame_depth(frame_depth_no_fixed_size + JITFRAME_FIXED_SIZE)
        #
        size_excluding_failure_stuff = self.mc.get_relative_pos()
        self.write_pending_failure_recoveries(regalloc)
        full_size = self.mc.get_relative_pos()
        #
        rawstart = self.materialize_loop(looptoken)
        self.patch_gcref_table(looptoken, rawstart)
        self.patch_stack_checks(frame_depth_no_fixed_size + JITFRAME_FIXED_SIZE,
                                rawstart)
        looptoken._ll_loop_code = looppos + rawstart
        debug_start("jit-backend-addr")
        debug_print("Loop %d (%s) has address 0x%x to 0x%x (bootstrap 0x%x)" % (
            looptoken.number, loopname,
            r_uint(rawstart + looppos),
            r_uint(rawstart + size_excluding_failure_stuff),
            r_uint(rawstart + functionpos)))
        debug_print("       gc table: 0x%x" % r_uint(self.gc_table_addr))
        debug_print("       function: 0x%x" % r_uint(rawstart + functionpos))
        debug_print("         resops: 0x%x" % r_uint(rawstart + looppos))
        debug_print("       failures: 0x%x" % r_uint(rawstart +
                                                 size_excluding_failure_stuff))
        debug_print("            end: 0x%x" % r_uint(rawstart + full_size))
        debug_stop("jit-backend-addr")
        self.patch_pending_failure_recoveries(rawstart)
        #
        ops_offset = self.mc.ops_offset
        if not we_are_translated():
            # used only by looptoken.dump() -- useful in tests
            looptoken._x86_rawstart = rawstart
            looptoken._x86_fullsize = full_size
            looptoken._x86_ops_offset = ops_offset
        looptoken._ll_function_addr = rawstart + functionpos

        if logger:
            log = logger.log_trace(jl.MARK_TRACE_ASM, None, self.mc)
            log.write(inputargs, operations, ops_offset=ops_offset)

            # legacy
            if logger.logger_ops:
                logger.logger_ops.log_loop(inputargs, operations, 0,
                                           "rewritten", name=loopname,
                                           ops_offset=ops_offset)

        self.fixup_target_tokens(rawstart)
        self.teardown()
        # oprofile support
        if self.cpu.profile_agent is not None:
            name = "Loop # %s: %s" % (looptoken.number, loopname)
            self.cpu.profile_agent.native_code_written(name,
                                                       rawstart, full_size)
        return AsmInfo(ops_offset, rawstart + looppos,
                       size_excluding_failure_stuff - looppos, rawstart)

    @rgc.no_release_gil
    def assemble_bridge(self, faildescr, inputargs, operations,
                        original_loop_token, log, logger):
        if not we_are_translated():
            # Arguments should be unique
            assert len(set(inputargs)) == len(inputargs)

        self.setup(original_loop_token)
        if self.cpu.HAS_CODEMAP:
            self.codemap_builder.inherit_code_from_position(
                faildescr.adr_jump_offset)
        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        descr_number = compute_unique_id(faildescr)
        if log:
            operations = self._inject_debugging_code(faildescr, operations,
                                                     'b', descr_number)
        arglocs = self.rebuild_faillocs_from_descr(faildescr, inputargs)
        regalloc = RegAlloc(self, self.cpu.translate_support_code)
        allgcrefs = []
        operations = regalloc.prepare_bridge(inputargs, arglocs,
                                             operations,
                                             allgcrefs,
                                             self.current_clt.frame_info)
        self.reserve_gcref_table(allgcrefs)
        startpos = self.mc.get_relative_pos()
        self._check_frame_depth(self.mc, regalloc.get_gcmap())
        bridgestartpos = self.mc.get_relative_pos()
        self._update_at_exit(arglocs, inputargs, faildescr, regalloc)
        frame_depth_no_fixed_size = self._assemble(regalloc, inputargs, operations)
        codeendpos = self.mc.get_relative_pos()
        self.write_pending_failure_recoveries(regalloc)
        fullsize = self.mc.get_relative_pos()
        #
        rawstart = self.materialize_loop(original_loop_token)
        self.patch_gcref_table(original_loop_token, rawstart)
        self.patch_stack_checks(frame_depth_no_fixed_size + JITFRAME_FIXED_SIZE,
                                rawstart)
        debug_start("jit-backend-addr")
        debug_print("bridge out of Guard 0x%x has address 0x%x to 0x%x" %
                    (r_uint(descr_number), r_uint(rawstart + startpos),
                        r_uint(rawstart + codeendpos)))
        debug_print("       gc table: 0x%x" % r_uint(self.gc_table_addr))
        debug_print("    jump target: 0x%x" % r_uint(rawstart + startpos))
        debug_print("         resops: 0x%x" % r_uint(rawstart + bridgestartpos))
        debug_print("       failures: 0x%x" % r_uint(rawstart + codeendpos))
        debug_print("            end: 0x%x" % r_uint(rawstart + fullsize))
        debug_stop("jit-backend-addr")
        self.patch_pending_failure_recoveries(rawstart)
        # patch the jump from original guard
        self.patch_jump_for_descr(faildescr, rawstart + startpos)
        ops_offset = self.mc.ops_offset
        frame_depth = max(self.current_clt.frame_info.jfi_frame_depth,
                          frame_depth_no_fixed_size + JITFRAME_FIXED_SIZE)

        if logger:
            log = logger.log_trace(jl.MARK_TRACE_ASM, None, self.mc)
            log.write(inputargs, operations, ops_offset)
            # log that the already written bridge is stitched to a descr!
            logger.log_patch_guard(descr_number, rawstart)

            # legacy
            if logger.logger_ops:
                logger.logger_ops.log_bridge(inputargs, operations, "rewritten",
                                          faildescr, ops_offset=ops_offset)

        self.fixup_target_tokens(rawstart)
        self.update_frame_depth(frame_depth)
        self.teardown()
        # oprofile support
        if self.cpu.profile_agent is not None:
            name = "Bridge # %s" % (descr_number,)
            self.cpu.profile_agent.native_code_written(name,
                                                       rawstart, fullsize)
        return AsmInfo(ops_offset, startpos + rawstart, codeendpos - startpos, rawstart+bridgestartpos)

    def stitch_bridge(self, faildescr, target):
        """ Stitching means that one can enter a bridge with a complete different register
            allocation. This needs remapping which is done here for both normal registers
            and accumulation registers.
            Why? Because this only generates a very small junk of memory, instead of
            duplicating the loop assembler for each faildescr!
        """
        asminfo, bridge_faildescr, version, looptoken = target
        assert isinstance(bridge_faildescr, ResumeGuardDescr)
        assert isinstance(faildescr, ResumeGuardDescr)
        assert asminfo.rawstart != 0
        self.mc = codebuf.MachineCodeBlockWrapper()
        allblocks = self.get_asmmemmgr_blocks(looptoken)
        self.datablockwrapper = MachineDataBlockWrapper(self.cpu.asmmemmgr,
                                                   allblocks)
        frame_info = self.datablockwrapper.malloc_aligned(
            jitframe.JITFRAMEINFO_SIZE, alignment=WORD)

        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        # if accumulation is saved at the guard, we need to update it here!
        guard_locs = self.rebuild_faillocs_from_descr(faildescr, version.inputargs)
        bridge_locs = self.rebuild_faillocs_from_descr(bridge_faildescr, version.inputargs)
        #import pdb; pdb.set_trace()
        guard_accum_info = faildescr.rd_vector_info
        # O(n^2), but usually you only have at most 1 fail argument
        while guard_accum_info:
            bridge_accum_info = bridge_faildescr.rd_vector_info
            while bridge_accum_info:
                if bridge_accum_info.failargs_pos == guard_accum_info.failargs_pos:
                    # the mapping might be wrong!
                    if bridge_accum_info.location is not guard_accum_info.location:
                        self.mov(guard_accum_info.location, bridge_accum_info.location)
                bridge_accum_info = bridge_accum_info.next()
            guard_accum_info = guard_accum_info.next()

        # register mapping is most likely NOT valid, thus remap it in this
        # short piece of assembler
        assert len(guard_locs) == len(bridge_locs)
        for i,gloc in enumerate(guard_locs):
            bloc = bridge_locs[i]
            bstack = bloc.location_code() == 'b'
            gstack = gloc.location_code() == 'b'
            if bstack and gstack:
                pass
            elif gloc is not bloc:
                self.mov(gloc, bloc)
        offset = self.mc.get_relative_pos()
        self.mc.JMP_l(0)
        self.mc.writeimm32(0)
        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        rawstart = self.materialize_loop(looptoken)
        # update the jump (above) to the real trace
        self._patch_jump_to(rawstart + offset, asminfo.rawstart)
        # update the guard to jump right to this custom piece of assembler
        self.patch_jump_for_descr(faildescr, rawstart)

    def _patch_jump_to(self, adr_jump_offset, adr_new_target):
        assert adr_jump_offset != 0
        offset = adr_new_target - (adr_jump_offset + 5)
        mc = codebuf.MachineCodeBlockWrapper()
        mc.force_frame_size(DEFAULT_FRAME_BYTES)
        if rx86.fits_in_32bits(offset):
            mc.JMP_l(offset)
        else:
            mc.MOV_ri(X86_64_SCRATCH_REG.value, adr_new_target)
            mc.JMP_r(X86_64_SCRATCH_REG.value)
        mc.copy_to_raw_memory(adr_jump_offset)

    def reserve_gcref_table(self, allgcrefs):
        gcref_table_size = len(allgcrefs) * WORD
        if IS_X86_64:
            # align to a multiple of 16 and reserve space at the beginning
            # of the machine code for the gc table.  This lets us write
            # machine code with relative addressing (%rip - constant).
            gcref_table_size = (gcref_table_size + 15) & ~15
            mc = self.mc
            assert mc.get_relative_pos() == 0
            for i in range(gcref_table_size):
                mc.writechar('\x00')
        elif IS_X86_32:
            # allocate the gc table right now.  This lets us write
            # machine code with absolute 32-bit addressing.
            self.gc_table_addr = self.datablockwrapper.malloc_aligned(
                gcref_table_size, alignment=WORD)
        #
        self.setup_gcrefs_list(allgcrefs)

    def patch_gcref_table(self, looptoken, rawstart):
        if IS_X86_64:
            # the gc table is at the start of the machine code
            self.gc_table_addr = rawstart
        elif IS_X86_32:
            # the gc table was already allocated by reserve_gcref_table()
            rawstart = self.gc_table_addr
        #
        tracer = self.cpu.gc_ll_descr.make_gcref_tracer(rawstart,
                                                        self._allgcrefs)
        gcreftracers = self.get_asmmemmgr_gcreftracers(looptoken)
        gcreftracers.append(tracer)    # keepalive
        self.teardown_gcrefs_list()

    def write_pending_failure_recoveries(self, regalloc):
        # for each pending guard, generate the code of the recovery stub
        # at the end of self.mc.
        for tok in self.pending_guard_tokens:
            descr = tok.faildescr
            if descr.loop_version():
                startpos = self.mc.get_relative_pos()
                self.store_info_on_descr(startpos, tok)
            else:
                tok.pos_recovery_stub = self.generate_quick_failure(tok, regalloc)
        if WORD == 8 and len(self.pending_memoryerror_trampoline_from) > 0:
            self.error_trampoline_64 = self.generate_propagate_error_64()

    def patch_pending_failure_recoveries(self, rawstart):
        # after we wrote the assembler to raw memory, set up
        # tok.faildescr.adr_jump_offset to contain the raw address of
        # the 4-byte target field in the JMP/Jcond instruction, and patch
        # the field in question to point (initially) to the recovery stub
        clt = self.current_clt
        for tok in self.pending_guard_tokens:
            addr = rawstart + tok.pos_jump_offset
            tok.faildescr.adr_jump_offset = addr
            descr = tok.faildescr
            if descr.loop_version():
                continue # patch them later
            relative_target = tok.pos_recovery_stub - (tok.pos_jump_offset + 4)
            assert rx86.fits_in_32bits(relative_target)
            #
            if not tok.guard_not_invalidated():
                mc = codebuf.MachineCodeBlockWrapper()
                mc.writeimm32(relative_target)
                mc.copy_to_raw_memory(addr)
            else:
                # GUARD_NOT_INVALIDATED, record an entry in
                # clt.invalidate_positions of the form:
                #     (addr-in-the-code-of-the-not-yet-written-jump-target,
                #      relative-target-to-use)
                relpos = tok.pos_jump_offset
                clt.invalidate_positions.append((rawstart + relpos,
                                                 relative_target))
                # General idea: Although no code was generated by this
                # guard, the code might be patched with a "JMP rel32" to
                # the guard recovery code.  This recovery code is
                # already generated, and looks like the recovery code
                # for any guard, even if at first it has no jump to it.
                # So we may later write 5 bytes overriding the existing
                # instructions; this works because a CALL instruction
                # would also take at least 5 bytes.  If it could take
                # less, we would run into the issue that overwriting the
                # 5 bytes here might get a few nonsense bytes at the
                # return address of the following CALL.
        if WORD == 8:
            for pos_after_jz in self.pending_memoryerror_trampoline_from:
                assert self.error_trampoline_64 != 0     # only if non-empty
                mc = codebuf.MachineCodeBlockWrapper()
                mc.writeimm32(self.error_trampoline_64 - pos_after_jz)
                mc.copy_to_raw_memory(rawstart + pos_after_jz - 4)

    def update_frame_depth(self, frame_depth):
        baseofs = self.cpu.get_baseofs_of_frame_field()
        self.current_clt.frame_info.update_frame_depth(baseofs, frame_depth)

    def patch_stack_checks(self, framedepth, rawstart):
        for ofs in self.frame_depth_to_patch:
            self._patch_frame_depth(ofs + rawstart, framedepth)

    def _check_frame_depth(self, mc, gcmap):
        """ check if the frame is of enough depth to follow this bridge.
        Otherwise reallocate the frame in a helper.
        There are other potential solutions
        to that, but this one does not sound too bad.
        """
        descrs = self.cpu.gc_ll_descr.getframedescrs(self.cpu)
        ofs = self.cpu.unpack_fielddescr(descrs.arraydescr.lendescr)
        mc.CMP_bi(ofs, 0xffffff)     # force writing 32 bit
        stack_check_cmp_ofs = mc.get_relative_pos() - 4
        mc.J_il8(rx86.Conditions['GE'], 0)
        jg_location = mc.get_relative_pos()
        mc.MOV_si(WORD, 0xffffff)     # force writing 32 bit
        ofs2 = mc.get_relative_pos() - 4
        self.push_gcmap(mc, gcmap, store=True)
        mc.CALL(imm(self._frame_realloc_slowpath))
        # patch the JG above
        offset = mc.get_relative_pos() - jg_location
        assert 0 < offset <= 127
        mc.overwrite(jg_location-1, chr(offset))
        self.frame_depth_to_patch.append(stack_check_cmp_ofs)
        self.frame_depth_to_patch.append(ofs2)

    def _check_frame_depth_debug(self, mc):
        """ double check the depth size. It prints the error (and potentially
        segfaults later)
        """
        if not self.DEBUG_FRAME_DEPTH:
            return
        descrs = self.cpu.gc_ll_descr.getframedescrs(self.cpu)
        ofs = self.cpu.unpack_fielddescr(descrs.arraydescr.lendescr)
        mc.CMP_bi(ofs, 0xffffff)
        stack_check_cmp_ofs = mc.get_relative_pos() - 4
        mc.J_il8(rx86.Conditions['GE'], 0)
        jg_location = mc.get_relative_pos()
        mc.MOV_rr(edi.value, ebp.value)
        mc.MOV_ri(esi.value, 0xffffff)
        ofs2 = mc.get_relative_pos() - 4
        mc.CALL(imm(self.cpu.realloc_frame_crash))
        # patch the JG above
        offset = mc.get_relative_pos() - jg_location
        assert 0 < offset <= 127
        mc.overwrite(jg_location-1, chr(offset))
        self.frame_depth_to_patch.append(stack_check_cmp_ofs)
        self.frame_depth_to_patch.append(ofs2)

    def _patch_frame_depth(self, adr, allocated_depth):
        mc = codebuf.MachineCodeBlockWrapper()
        mc.writeimm32(allocated_depth)
        mc.copy_to_raw_memory(adr)

    def materialize_loop(self, looptoken):
        self.datablockwrapper.done()      # finish using cpu.asmmemmgr
        self.datablockwrapper = None
        allblocks = self.get_asmmemmgr_blocks(looptoken)
        size = self.mc.get_relative_pos()
        res = self.mc.materialize(self.cpu, allblocks,
                                  self.cpu.gc_ll_descr.gcrootmap)
        if self.cpu.HAS_CODEMAP:
            self.cpu.codemap.register_codemap(
                self.codemap_builder.get_final_bytecode(res, size))
        return res

    def patch_jump_for_descr(self, faildescr, adr_new_target):
        adr_jump_offset = faildescr.adr_jump_offset
        assert adr_jump_offset != 0
        offset = adr_new_target - (adr_jump_offset + 4)
        # If the new target fits within a rel32 of the jump, just patch
        # that. Otherwise, leave the original rel32 to the recovery stub in
        # place, but clobber the recovery stub with a jump to the real
        # target.
        mc = codebuf.MachineCodeBlockWrapper()
        mc.force_frame_size(DEFAULT_FRAME_BYTES)
        if rx86.fits_in_32bits(offset):
            mc.writeimm32(offset)
            mc.copy_to_raw_memory(adr_jump_offset)
        else:
            # "mov r11, addr; jmp r11" is up to 13 bytes, which fits in there
            # because we always write "mov r11, imm-as-8-bytes; call *r11" in
            # the first place.
            mc.MOV_ri(X86_64_SCRATCH_REG.value, adr_new_target)
            mc.JMP_r(X86_64_SCRATCH_REG.value)
            p = rffi.cast(rffi.INTP, adr_jump_offset)
            adr_target = adr_jump_offset + 4 + rffi.cast(lltype.Signed, p[0])
            mc.copy_to_raw_memory(adr_target)
        faildescr.adr_jump_offset = 0    # means "patched"

    def fixup_target_tokens(self, rawstart):
        for targettoken in self.target_tokens_currently_compiling:
            targettoken._ll_loop_code += rawstart
        self.target_tokens_currently_compiling = None

    def _assemble(self, regalloc, inputargs, operations):
        self._regalloc = regalloc
        self.guard_success_cc = rx86.cond_none
        regalloc.compute_hint_frame_locations(operations)
        regalloc.walk_operations(inputargs, operations)
        assert self.guard_success_cc == rx86.cond_none
        if we_are_translated() or self.cpu.dont_keepalive_stuff:
            self._regalloc = None   # else keep it around for debugging
        frame_depth = regalloc.get_final_frame_depth()
        jump_target_descr = regalloc.jump_target_descr
        if jump_target_descr is not None:
            tgt_depth = jump_target_descr._x86_clt.frame_info.jfi_frame_depth
            target_frame_depth = tgt_depth - JITFRAME_FIXED_SIZE
            frame_depth = max(frame_depth, target_frame_depth)
        return frame_depth

    def _call_header_vmprof(self):
        from rpython.rlib.rvmprof.rvmprof import cintf, VMPROF_JITTED_TAG

        # tloc = address of pypy_threadlocal_s
        if IS_X86_32:
            # Can't use esi here, its old value is not saved yet.
            # But we can use eax and ecx.
            self.mc.MOV_rs(edx.value, THREADLOCAL_OFS)
            tloc = edx
            old = ecx
        else:
            # The thread-local value is already in esi.
            # We should avoid if possible to use ecx or edx because they
            # would be used to pass arguments #3 and #4 (even though, so
            # far, the assembler only receives two arguments).
            tloc = esi
            old = r11
        # eax = address in the stack of a 3-words struct vmprof_stack_s
        self.mc.LEA_rs(eax.value, (FRAME_FIXED_SIZE - 4) * WORD)
        # old = current value of vmprof_tl_stack
        offset = cintf.vmprof_tl_stack.getoffset()
        self.mc.MOV_rm(old.value, (tloc.value, offset))
        # eax->next = old
        self.mc.MOV_mr((eax.value, 0), old.value)
        # eax->value = my esp
        self.mc.MOV_mr((eax.value, WORD), esp.value)
        # eax->kind = VMPROF_JITTED_TAG
        self.mc.MOV_mi((eax.value, WORD * 2), VMPROF_JITTED_TAG)
        # save in vmprof_tl_stack the new eax
        self.mc.MOV_mr((tloc.value, offset), eax.value)

    def _call_footer_vmprof(self):
        from rpython.rlib.rvmprof.rvmprof import cintf
        # edx = address of pypy_threadlocal_s
        self.mc.MOV_rs(edx.value, THREADLOCAL_OFS)
        self.mc.AND_ri(edx.value, ~1)
        # eax = (our local vmprof_tl_stack).next
        self.mc.MOV_rs(eax.value, (FRAME_FIXED_SIZE - 4 + 0) * WORD)
        # save in vmprof_tl_stack the value eax
        offset = cintf.vmprof_tl_stack.getoffset()
        self.mc.MOV_mr((edx.value, offset), eax.value)

    def _call_header(self):
        self.mc.SUB_ri(esp.value, FRAME_FIXED_SIZE * WORD)
        self.mc.MOV_sr(PASS_ON_MY_FRAME * WORD, ebp.value)
        if IS_X86_64:
            self.mc.MOV_sr(THREADLOCAL_OFS, esi.value)
        if self.cpu.translate_support_code:
            self._call_header_vmprof()     # on X86_64, this uses esi
        if IS_X86_64:
            self.mc.MOV_rr(ebp.value, edi.value)
        else:
            self.mc.MOV_rs(ebp.value, (FRAME_FIXED_SIZE + 1) * WORD)

        for i, loc in enumerate(self.cpu.CALLEE_SAVE_REGISTERS):
            self.mc.MOV_sr((PASS_ON_MY_FRAME + i + 1) * WORD, loc.value)

        gcrootmap = self.cpu.gc_ll_descr.gcrootmap
        if gcrootmap and gcrootmap.is_shadow_stack:
            self._call_header_shadowstack(gcrootmap)

    def _call_header_with_stack_check(self):
        self._call_header()
        if self.stack_check_slowpath == 0:
            pass                # no stack check (e.g. not translated)
        else:
            endaddr, lengthaddr, _ = self.cpu.insert_stack_check()
            self.mc.MOV(eax, heap(endaddr))             # MOV eax, [start]
            self.mc.SUB(eax, esp)                       # SUB eax, current
            self.mc.CMP(eax, heap(lengthaddr))          # CMP eax, [length]
            self.mc.J_il8(rx86.Conditions['BE'], 0)     # JBE .skip
            jb_location = self.mc.get_relative_pos()
            self.mc.CALL(imm(self.stack_check_slowpath))# CALL slowpath
            # patch the JB above                        # .skip:
            offset = self.mc.get_relative_pos() - jb_location
            assert 0 < offset <= 127
            self.mc.overwrite(jb_location-1, chr(offset))
            #

    def _call_footer(self):
        # the return value is the jitframe
        if self.cpu.translate_support_code:
            self._call_footer_vmprof()
        self.mc.MOV_rr(eax.value, ebp.value)

        gcrootmap = self.cpu.gc_ll_descr.gcrootmap
        if gcrootmap and gcrootmap.is_shadow_stack:
            self._call_footer_shadowstack(gcrootmap)

        for i in range(len(self.cpu.CALLEE_SAVE_REGISTERS)-1, -1, -1):
            self.mc.MOV_rs(self.cpu.CALLEE_SAVE_REGISTERS[i].value,
                           (i + 1 + PASS_ON_MY_FRAME) * WORD)

        self.mc.MOV_rs(ebp.value, PASS_ON_MY_FRAME * WORD)
        self.mc.ADD_ri(esp.value, FRAME_FIXED_SIZE * WORD)
        self.mc.RET()

    def _load_shadowstack_top_in_ebx(self, mc, gcrootmap):
        """Loads the shadowstack top in ebx, and returns an integer
        that gives the address of the stack top.  If this integer doesn't
        fit in 32 bits, it will be loaded in r11.
        """
        rst = gcrootmap.get_root_stack_top_addr()
        if rx86.fits_in_32bits(rst):
            mc.MOV_rj(ebx.value, rst)            # MOV ebx, [rootstacktop]
        else:
            mc.MOV_ri(X86_64_SCRATCH_REG.value, rst) # MOV r11, rootstacktop
            mc.MOV_rm(ebx.value, (X86_64_SCRATCH_REG.value, 0))
            # MOV ebx, [r11]
        #
        return rst

    def _call_header_shadowstack(self, gcrootmap):
        rst = self._load_shadowstack_top_in_ebx(self.mc, gcrootmap)
        self.mc.MOV_mr((ebx.value, 0), ebp.value)      # MOV [ebx], ebp
        self.mc.ADD_ri(ebx.value, WORD)
        if rx86.fits_in_32bits(rst):
            self.mc.MOV_jr(rst, ebx.value)            # MOV [rootstacktop], ebx
        else:
            # The integer 'rst' doesn't fit in 32 bits, so we know that
            # _load_shadowstack_top_in_ebx() above loaded it in r11.
            # Reuse it.  Be careful not to overwrite r11 in the middle!
            self.mc.MOV_mr((X86_64_SCRATCH_REG.value, 0),
                           ebx.value) # MOV [r11], ebx

    def _call_footer_shadowstack(self, gcrootmap):
        rst = gcrootmap.get_root_stack_top_addr()
        if rx86.fits_in_32bits(rst):
            self.mc.SUB_ji8(rst, WORD)       # SUB [rootstacktop], WORD
        else:
            self.mc.MOV_ri(ebx.value, rst)           # MOV ebx, rootstacktop
            self.mc.SUB_mi8((ebx.value, 0), WORD)  # SUB [ebx], WORD

    def redirect_call_assembler(self, oldlooptoken, newlooptoken):
        # some minimal sanity checking
        old_nbargs = oldlooptoken.compiled_loop_token._debug_nbargs
        new_nbargs = newlooptoken.compiled_loop_token._debug_nbargs
        assert old_nbargs == new_nbargs
        # we overwrite the instructions at the old _ll_function_addr
        # to start with a JMP to the new _ll_function_addr.
        # Ideally we should rather patch all existing CALLs, but well.
        oldadr = oldlooptoken._ll_function_addr
        target = newlooptoken._ll_function_addr
        # copy frame-info data
        baseofs = self.cpu.get_baseofs_of_frame_field()
        newlooptoken.compiled_loop_token.update_frame_info(
            oldlooptoken.compiled_loop_token, baseofs)
        mc = codebuf.MachineCodeBlockWrapper()
        mc.JMP(imm(follow_jump(target)))
        if WORD == 4:         # keep in sync with prepare_loop()
            assert mc.get_relative_pos() == 5
        else:
            assert mc.get_relative_pos() <= 13
        mc.copy_to_raw_memory(oldadr)

    def dump(self, text):
        if not self.verbose:
            return
        pos = self.mc.get_relative_pos()
        print >> sys.stderr, ' 0x%x  %s' % (pos, text)

    # ------------------------------------------------------------

    def mov(self, from_loc, to_loc):
        from_xmm = isinstance(from_loc, RegLoc) and from_loc.is_xmm
        to_xmm = isinstance(to_loc, RegLoc) and to_loc.is_xmm
        if from_xmm or to_xmm:
            if from_xmm and to_xmm:
                # copy 128-bit from -> to
                self.mc.MOVAPD(to_loc, from_loc)
            else:
                self.mc.MOVSD(to_loc, from_loc)
        else:
            assert to_loc is not ebp
            self.mc.MOV(to_loc, from_loc)

    regalloc_mov = mov # legacy interface

    def regalloc_push(self, loc):
        if isinstance(loc, RegLoc) and loc.is_xmm:
            self.mc.SUB_ri(esp.value, 8)   # = size of doubles
            self.mc.MOVSD_sx(0, loc.value)
        elif WORD == 4 and isinstance(loc, FrameLoc) and loc.get_width() == 8:
            # XXX evil trick
            self.mc.PUSH_b(loc.value + 4)
            self.mc.PUSH_b(loc.value)
        else:
            self.mc.PUSH(loc)

    def regalloc_pop(self, loc):
        if isinstance(loc, RegLoc) and loc.is_xmm:
            self.mc.MOVSD_xs(loc.value, 0)
            self.mc.ADD_ri(esp.value, 8)   # = size of doubles
        elif WORD == 4 and isinstance(loc, FrameLoc) and loc.get_width() == 8:
            # XXX evil trick
            self.mc.POP_b(loc.value)
            self.mc.POP_b(loc.value + 4)
        else:
            self.mc.POP(loc)

    def regalloc_immedmem2mem(self, from_loc, to_loc):
        # move a ConstFloatLoc directly to a FrameLoc, as two MOVs
        # (even on x86-64, because the immediates are encoded as 32 bits)
        assert isinstance(from_loc, ConstFloatLoc)
        low_part  = rffi.cast(rffi.CArrayPtr(rffi.INT), from_loc.value)[0]
        high_part = rffi.cast(rffi.CArrayPtr(rffi.INT), from_loc.value)[1]
        low_part  = intmask(low_part)
        high_part = intmask(high_part)
        if isinstance(to_loc, RawEbpLoc):
            self.mc.MOV32_bi(to_loc.value,     low_part)
            self.mc.MOV32_bi(to_loc.value + 4, high_part)
        else:
            assert isinstance(to_loc, RawEspLoc)
            self.mc.MOV32_si(to_loc.value,     low_part)
            self.mc.MOV32_si(to_loc.value + 4, high_part)

    def regalloc_perform(self, op, arglocs, resloc):
        genop_list[op.getopnum()](self, op, arglocs, resloc)

    def regalloc_perform_discard(self, op, arglocs):
        genop_discard_list[op.getopnum()](self, op, arglocs)

    def regalloc_perform_llong(self, op, arglocs, resloc):
        effectinfo = op.getdescr().get_extra_info()
        oopspecindex = effectinfo.oopspecindex
        genop_llong_list[oopspecindex](self, op, arglocs, resloc)

    def regalloc_perform_math(self, op, arglocs, resloc):
        effectinfo = op.getdescr().get_extra_info()
        oopspecindex = effectinfo.oopspecindex
        genop_math_list[oopspecindex](self, op, arglocs, resloc)

    def regalloc_perform_guard(self, guard_op, faillocs, arglocs, resloc,
                               frame_depth):
        faildescr = guard_op.getdescr()
        assert isinstance(faildescr, AbstractFailDescr)
        failargs = guard_op.getfailargs()
        guard_opnum = guard_op.getopnum()
        guard_token = self.implement_guard_recovery(guard_opnum,
                                                    faildescr, failargs,
                                                    faillocs, frame_depth)
        genop_guard_list[guard_opnum](self, guard_op, guard_token,
                                      arglocs, resloc)
        if not we_are_translated():
            # must be added by the genop_guard_list[]()
            assert guard_token is self.pending_guard_tokens[-1]

    def load_effective_addr(self, sizereg, baseofs, scale, result, frm=imm0):
        self.mc.LEA(result, addr_add(frm, sizereg, baseofs, scale))

    def _unaryop(asmop):
        def genop_unary(self, op, arglocs, resloc):
            getattr(self.mc, asmop)(arglocs[0])
        return genop_unary

    def _binaryop(asmop):
        def genop_binary(self, op, arglocs, result_loc):
            getattr(self.mc, asmop)(arglocs[0], arglocs[1])
        return genop_binary

    def _binaryop_or_lea(asmop, is_add):
        def genop_binary_or_lea(self, op, arglocs, result_loc):
            # use a regular ADD or SUB if result_loc is arglocs[0],
            # and a LEA only if different.
            if result_loc is arglocs[0]:
                getattr(self.mc, asmop)(arglocs[0], arglocs[1])
            else:
                loc = arglocs[0]
                argloc = arglocs[1]
                assert isinstance(loc, RegLoc)
                assert isinstance(argloc, ImmedLoc)
                assert isinstance(result_loc, RegLoc)
                delta = argloc.value
                if not is_add:    # subtraction
                    delta = -delta
                self.mc.LEA_rm(result_loc.value, (loc.value, delta))
        return genop_binary_or_lea

    def flush_cc(self, cond, result_loc):
        # After emitting a instruction that leaves a boolean result in
        # a condition code (cc), call this.  In the common case, result_loc
        # will be set to ebp by the regalloc, which in this case means
        # "propagate it between this operation and the next guard by keeping
        # it in the cc".  In the uncommon case, result_loc is another
        # register, and we emit a load from the cc into this register.
        assert self.guard_success_cc == rx86.cond_none
        if result_loc is ebp:
            self.guard_success_cc = cond
        else:
            self.mc.MOV_ri(result_loc.value, 0)
            rl = result_loc.lowest8bits()
            self.mc.SET_ir(cond, rl.value)

    def _cmpop(cond, rev_cond):
        cond = rx86.Conditions[cond]
        rev_cond = rx86.Conditions[rev_cond]
        #
        def genop_cmp(self, op, arglocs, result_loc):
            if isinstance(op.getarg(0), Const):
                self.mc.CMP(arglocs[1], arglocs[0])
                self.flush_cc(rev_cond, result_loc)
            else:
                self.mc.CMP(arglocs[0], arglocs[1])
                self.flush_cc(cond, result_loc)
        return genop_cmp

    def _if_parity_clear_zero_and_carry(self):
        self.mc.J_il8(rx86.Conditions['NP'], 0)
        jnp_location = self.mc.get_relative_pos()
        # CMP EBP, 0: as EBP cannot be null here, that operation should
        # always clear zero and carry
        self.mc.CMP_ri(ebp.value, 0)
        # patch the JNP above
        offset = self.mc.get_relative_pos() - jnp_location
        assert 0 < offset <= 127
        self.mc.overwrite(jnp_location-1, chr(offset))

    def _cmpop_float(cond, rev_cond):
        is_ne           = cond == 'NE'
        need_direct_p   = 'A' not in cond
        need_rev_p      = 'A' not in rev_cond
        cond_contains_e = ('E' in cond) ^ ('N' in cond)
        cond            = rx86.Conditions[cond]
        rev_cond        = rx86.Conditions[rev_cond]
        #
        def genop_cmp_float(self, op, arglocs, result_loc):
            if need_direct_p:
                direct_case = not isinstance(arglocs[1], RegLoc)
            else:
                direct_case = isinstance(arglocs[0], RegLoc)
            if direct_case:
                self.mc.UCOMISD(arglocs[0], arglocs[1])
                checkcond = cond
                need_p = need_direct_p
            else:
                self.mc.UCOMISD(arglocs[1], arglocs[0])
                checkcond = rev_cond
                need_p = need_rev_p
            if need_p:
                self._if_parity_clear_zero_and_carry()
            self.flush_cc(checkcond, result_loc)
        return genop_cmp_float

    def simple_call(self, fnloc, arglocs, result_loc=eax):
        if result_loc is xmm0:
            result_type = FLOAT
            result_size = 8
        elif result_loc is None:
            result_type = VOID
            result_size = 0
        else:
            result_type = INT
            result_size = WORD
        cb = callbuilder.CallBuilder(self, fnloc, arglocs,
                                     result_loc, result_type,
                                     result_size)
        cb.emit()

    def simple_call_no_collect(self, fnloc, arglocs):
        cb = callbuilder.CallBuilder(self, fnloc, arglocs)
        cb.emit_no_collect()

    def _reload_frame_if_necessary(self, mc, shadowstack_reg=None):
        gcrootmap = self.cpu.gc_ll_descr.gcrootmap
        if gcrootmap:
            if gcrootmap.is_shadow_stack:
                if shadowstack_reg is None:
                    rst = gcrootmap.get_root_stack_top_addr()
                    mc.MOV(ecx, heap(rst))
                    shadowstack_reg = ecx
                mc.MOV(ebp, mem(shadowstack_reg, -WORD))
        wbdescr = self.cpu.gc_ll_descr.write_barrier_descr
        if gcrootmap and wbdescr:
            # frame never uses card marking, so we enforce this is not
            # an array
            self._write_barrier_fastpath(mc, wbdescr, [ebp], array=False,
                                         is_frame=True)

    genop_int_neg = _unaryop("NEG")
    genop_int_invert = _unaryop("NOT")
    genop_int_add = _binaryop_or_lea("ADD", is_add=True)
    genop_nursery_ptr_increment = _binaryop_or_lea('ADD', is_add=True)
    genop_int_sub = _binaryop_or_lea("SUB", is_add=False)
    genop_int_mul = _binaryop("IMUL")
    genop_int_or  = _binaryop("OR")
    genop_int_xor = _binaryop("XOR")
    genop_int_lshift = _binaryop("SHL")
    genop_int_rshift = _binaryop("SAR")
    genop_uint_rshift = _binaryop("SHR")
    genop_float_add = _binaryop("ADDSD")
    genop_float_sub = _binaryop('SUBSD')
    genop_float_mul = _binaryop('MULSD')
    genop_float_truediv = _binaryop('DIVSD')

    def genop_uint_mul_high(self, op, arglocs, result_loc):
        self.mc.MUL(arglocs[0])

    def genop_int_and(self, op, arglocs, result_loc):
        arg1 = arglocs[1]
        if IS_X86_64 and (isinstance(arg1, ImmedLoc) and
                          arg1.value == (1 << 32) - 1):
            # special case
            self.mc.MOV32(arglocs[0], arglocs[0])
        else:
            self.mc.AND(arglocs[0], arg1)

    genop_int_lt = _cmpop("L", "G")
    genop_int_le = _cmpop("LE", "GE")
    genop_int_eq = _cmpop("E", "E")
    genop_int_ne = _cmpop("NE", "NE")
    genop_int_gt = _cmpop("G", "L")
    genop_int_ge = _cmpop("GE", "LE")
    genop_ptr_eq = genop_instance_ptr_eq = genop_int_eq
    genop_ptr_ne = genop_instance_ptr_ne = genop_int_ne

    genop_uint_gt = _cmpop("A", "B")
    genop_uint_lt = _cmpop("B", "A")
    genop_uint_le = _cmpop("BE", "AE")
    genop_uint_ge = _cmpop("AE", "BE")

    genop_float_lt = _cmpop_float("B", "A")
    genop_float_le = _cmpop_float("BE","AE")
    genop_float_eq = _cmpop_float("E", "E")
    genop_float_ne = _cmpop_float("NE", "NE")
    genop_float_gt = _cmpop_float("A", "B")
    genop_float_ge = _cmpop_float("AE","BE")

    def genop_math_sqrt(self, op, arglocs, resloc):
        self.mc.SQRTSD(arglocs[0], resloc)

    def genop_int_signext(self, op, arglocs, resloc):
        argloc, numbytesloc = arglocs
        assert isinstance(numbytesloc, ImmedLoc)
        assert isinstance(resloc, RegLoc)
        if numbytesloc.value == 1:
            if isinstance(argloc, RegLoc):
                if WORD == 4 and argloc.value >= 4:
                    # meh, can't read the lowest byte of esi or edi on 32-bit
                    if resloc is not argloc:
                        self.mc.MOV(resloc, argloc)
                        argloc = resloc
                    if resloc.value >= 4:
                        # still annoyed, hack needed
                        self.mc.SHL_ri(resloc.value, 24)
                        self.mc.SAR_ri(resloc.value, 24)
                        return
                argloc = argloc.lowest8bits()
            self.mc.MOVSX8(resloc, argloc)
        elif numbytesloc.value == 2:
            self.mc.MOVSX16(resloc, argloc)
        elif IS_X86_64 and numbytesloc.value == 4:
            self.mc.MOVSX32(resloc, argloc)
        else:
            raise AssertionError("bad number of bytes")

    def genop_float_neg(self, op, arglocs, resloc):
        # Following what gcc does: res = x ^ 0x8000000000000000
        self.mc.XORPD(arglocs[0], heap(self.float_const_neg_addr))

    def genop_float_abs(self, op, arglocs, resloc):
        # Following what gcc does: res = x & 0x7FFFFFFFFFFFFFFF
        self.mc.ANDPD(arglocs[0], heap(self.float_const_abs_addr))

    def genop_cast_float_to_int(self, op, arglocs, resloc):
        self.mc.CVTTSD2SI(resloc, arglocs[0])

    def genop_cast_int_to_float(self, op, arglocs, resloc):
        self.mc.CVTSI2SD(resloc, arglocs[0])

    def genop_cast_float_to_singlefloat(self, op, arglocs, resloc):
        loc0, loctmp = arglocs
        self.mc.CVTSD2SS(loctmp, loc0)
        assert isinstance(resloc, RegLoc)
        assert isinstance(loctmp, RegLoc)
        self.mc.MOVD32_rx(resloc.value, loctmp.value)

    def genop_cast_singlefloat_to_float(self, op, arglocs, resloc):
        loc0, = arglocs
        assert isinstance(resloc, RegLoc)
        assert isinstance(loc0, RegLoc)
        self.mc.MOVD32_xr(resloc.value, loc0.value)
        self.mc.CVTSS2SD_xx(resloc.value, resloc.value)

    def genop_convert_float_bytes_to_longlong(self, op, arglocs, resloc):
        loc0, = arglocs
        if longlong.is_64_bit:
            assert isinstance(resloc, RegLoc)
            assert isinstance(loc0, RegLoc)
            self.mc.MOVDQ(resloc, loc0)
        else:
            self.mov(loc0, resloc)

    def genop_convert_longlong_bytes_to_float(self, op, arglocs, resloc):
        loc0, = arglocs
        if longlong.is_64_bit:
            assert isinstance(resloc, RegLoc)
            assert isinstance(loc0, RegLoc)
            self.mc.MOVDQ(resloc, loc0)
        else:
            self.mov(loc0, resloc)

    def test_location(self, loc):
        assert not isinstance(loc, ImmedLoc)
        if isinstance(loc, RegLoc):
            self.mc.TEST_rr(loc.value, loc.value)   # more compact
        else:
            self.mc.CMP(loc, imm0)         # works from memory too

    def genop_int_is_true(self, op, arglocs, resloc):
        self.test_location(arglocs[0])
        self.flush_cc(rx86.Conditions['NZ'], resloc)

    def genop_int_is_zero(self, op, arglocs, resloc):
        self.test_location(arglocs[0])
        self.flush_cc(rx86.Conditions['Z'], resloc)

    def _genop_same_as(self, op, arglocs, resloc):
        self.mov(arglocs[0], resloc)
    genop_same_as_i = _genop_same_as
    genop_same_as_r = _genop_same_as
    genop_same_as_f = _genop_same_as
    genop_cast_ptr_to_int = _genop_same_as
    genop_cast_int_to_ptr = _genop_same_as

    def _patch_load_from_gc_table(self, index):
        # must be called immediately after a "p"-mode instruction
        # has been emitted.  64-bit mode only.
        assert IS_X86_64
        address_in_buffer = index * WORD   # at the start of the buffer
        p_location = self.mc.get_relative_pos()
        offset = address_in_buffer - p_location
        self.mc.overwrite32(p_location-4, offset)

    def _addr_from_gc_table(self, index):
        # get the address of the gc table entry 'index'.  32-bit mode only.
        assert IS_X86_32
        return self.gc_table_addr + index * WORD

    def genop_load_from_gc_table(self, op, arglocs, resloc):
        index = op.getarg(0).getint()
        assert isinstance(resloc, RegLoc)
        if IS_X86_64:
            self.mc.MOV_rp(resloc.value, 0)    # %rip-relative
            self._patch_load_from_gc_table(index)
        elif IS_X86_32:
            self.mc.MOV_rj(resloc.value, self._addr_from_gc_table(index))

    def genop_int_force_ge_zero(self, op, arglocs, resloc):
        self.mc.TEST(arglocs[0], arglocs[0])
        self.mov(imm0, resloc)
        self.mc.CMOVNS(resloc, arglocs[0])

    genop_llong_add = _binaryop("PADDQ")
    genop_llong_sub = _binaryop("PSUBQ")
    genop_llong_and = _binaryop("PAND")
    genop_llong_or  = _binaryop("POR")
    genop_llong_xor = _binaryop("PXOR")

    def genop_llong_to_int(self, op, arglocs, resloc):
        loc = arglocs[0]
        assert isinstance(resloc, RegLoc)
        if isinstance(loc, RegLoc):
            self.mc.MOVD32_rx(resloc.value, loc.value)
        elif isinstance(loc, FrameLoc):
            self.mc.MOV_rb(resloc.value, loc.value)
        else:
            not_implemented("llong_to_int: %s" % (loc,))

    def genop_llong_from_int(self, op, arglocs, resloc):
        loc1, loc2 = arglocs
        if isinstance(loc1, ConstFloatLoc):
            assert loc2 is None
            self.mc.MOVSD(resloc, loc1)
        else:
            assert isinstance(loc1, RegLoc)
            assert isinstance(loc2, RegLoc)
            assert isinstance(resloc, RegLoc)
            self.mc.MOVD32_xr(loc2.value, loc1.value)
            self.mc.PSRAD_xi(loc2.value, 31)    # -> 0 or -1
            self.mc.MOVD32_xr(resloc.value, loc1.value)
            self.mc.PUNPCKLDQ_xx(resloc.value, loc2.value)

    def genop_llong_from_uint(self, op, arglocs, resloc):
        loc1, = arglocs
        assert isinstance(resloc, RegLoc)
        assert isinstance(loc1, RegLoc)
        self.mc.MOVD32_xr(resloc.value, loc1.value)   # zero-extending

    def genop_llong_eq(self, op, arglocs, resloc):
        loc1, loc2, locxtmp = arglocs
        self.mc.MOVSD(locxtmp, loc1)
        self.mc.PCMPEQD(locxtmp, loc2)
        self.mc.PMOVMSKB_rx(resloc.value, locxtmp.value)
        # Now the lower 8 bits of resloc contain 0x00, 0x0F, 0xF0 or 0xFF
        # depending on the result of the comparison of each of the two
        # double-words of loc1 and loc2.  The higher 8 bits contain random
        # results.  We want to map 0xFF to 1, and 0x00, 0x0F and 0xF0 to 0.
        self.mc.CMP8_ri(resloc.value | rx86.BYTE_REG_FLAG, -1)
        self.mc.SBB_rr(resloc.value, resloc.value)
        self.mc.ADD_ri(resloc.value, 1)

    def genop_llong_ne(self, op, arglocs, resloc):
        loc1, loc2, locxtmp = arglocs
        self.mc.MOVSD(locxtmp, loc1)
        self.mc.PCMPEQD(locxtmp, loc2)
        self.mc.PMOVMSKB_rx(resloc.value, locxtmp.value)
        # Now the lower 8 bits of resloc contain 0x00, 0x0F, 0xF0 or 0xFF
        # depending on the result of the comparison of each of the two
        # double-words of loc1 and loc2.  The higher 8 bits contain random
        # results.  We want to map 0xFF to 0, and 0x00, 0x0F and 0xF0 to 1.
        self.mc.CMP8_ri(resloc.value | rx86.BYTE_REG_FLAG, -1)
        self.mc.SBB_rr(resloc.value, resloc.value)
        self.mc.NEG_r(resloc.value)

    def genop_llong_lt(self, op, arglocs, resloc):
        # XXX just a special case for now: "x < 0"
        loc1, = arglocs
        self.mc.PMOVMSKB_rx(resloc.value, loc1.value)
        self.mc.SHR_ri(resloc.value, 7)
        self.mc.AND_ri(resloc.value, 1)

    # ----------

    def genop_discard_check_memory_error(self, op, arglocs):
        reg = arglocs[0]
        self.mc.TEST(reg, reg)
        if WORD == 4:
            self.mc.J_il(rx86.Conditions['Z'], self.propagate_exception_path)
            self.mc.add_pending_relocation()
        elif WORD == 8:
            self.mc.J_il(rx86.Conditions['Z'], 0)
            pos = self.mc.get_relative_pos()
            self.pending_memoryerror_trampoline_from.append(pos)

    # ----------

    def load_from_mem(self, resloc, source_addr, size_loc, sign_loc):
        assert isinstance(resloc, RegLoc)
        size = size_loc.value
        sign = sign_loc.value
        if resloc.is_xmm:
            self.mc.MOVSD(resloc, source_addr)
        elif size == WORD:
            self.mc.MOV(resloc, source_addr)
        elif size == 1:
            if sign:
                self.mc.MOVSX8(resloc, source_addr)
            else:
                self.mc.MOVZX8(resloc, source_addr)
        elif size == 2:
            if sign:
                self.mc.MOVSX16(resloc, source_addr)
            else:
                self.mc.MOVZX16(resloc, source_addr)
        elif IS_X86_64 and size == 4:
            if sign:
                self.mc.MOVSX32(resloc, source_addr)
            else:
                self.mc.MOV32(resloc, source_addr)    # zero-extending
        else:
            not_implemented("load_from_mem size = %d" % size)

    def save_into_mem(self, dest_addr, value_loc, size_loc):
        size = size_loc.value
        if isinstance(value_loc, RegLoc) and value_loc.is_xmm:
            self.mc.MOVSD(dest_addr, value_loc)
        elif size == 1:
            self.mc.MOV8(dest_addr, value_loc.lowest8bits())
        elif size == 2:
            self.mc.MOV16(dest_addr, value_loc)
        elif size == 4:
            self.mc.MOV32(dest_addr, value_loc)
        elif size == 8:
            if IS_X86_64:
                self.mc.MOV(dest_addr, value_loc)
            else:
                assert isinstance(value_loc, FloatImmedLoc)
                self.mc.MOV(dest_addr, value_loc.low_part_loc())
                self.mc.MOV(dest_addr.add_offset(4), value_loc.high_part_loc())
        else:
            not_implemented("save_into_mem size = %d" % size)

    def _genop_getfield(self, op, arglocs, resloc):
        base_loc, ofs_loc, size_loc, sign_loc = arglocs
        assert isinstance(size_loc, ImmedLoc)
        source_addr = AddressLoc(base_loc, ofs_loc)
        self.load_from_mem(resloc, source_addr, size_loc, sign_loc)

    genop_getfield_gc_i = _genop_getfield
    genop_getfield_gc_r = _genop_getfield
    genop_getfield_gc_f = _genop_getfield
    genop_getfield_raw_i = _genop_getfield
    genop_getfield_raw_f = _genop_getfield

    def _genop_gc_load(self, op, arglocs, resloc):
        base_loc, ofs_loc, size_loc, sign_loc = arglocs
        assert isinstance(size_loc, ImmedLoc)
        src_addr = addr_add(base_loc, ofs_loc, 0, 0)
        self.load_from_mem(resloc, src_addr, size_loc, sign_loc)

    genop_gc_load_i = _genop_gc_load
    genop_gc_load_r = _genop_gc_load
    genop_gc_load_f = _genop_gc_load

    def _genop_gc_load_indexed(self, op, arglocs, resloc):
        base_loc, ofs_loc, scale_loc, offset_loc, size_loc, sign_loc = arglocs
        assert isinstance(scale_loc, ImmedLoc)
        scale = get_scale(scale_loc.value)
        src_addr = addr_add(base_loc, ofs_loc, offset_loc.value, scale)
        self.load_from_mem(resloc, src_addr, size_loc, sign_loc)

    genop_gc_load_indexed_i = _genop_gc_load_indexed
    genop_gc_load_indexed_r = _genop_gc_load_indexed
    genop_gc_load_indexed_f = _genop_gc_load_indexed

    def _imul_const_scaled(self, mc, targetreg, sourcereg, itemsize):
        """Produce one operation to do roughly
               targetreg = sourcereg * itemsize
           except that the targetreg may still need shifting by 0,1,2,3.
        """
        if (itemsize & 7) == 0:
            shift = 3
        elif (itemsize & 3) == 0:
            shift = 2
        elif (itemsize & 1) == 0:
            shift = 1
        else:
            shift = 0
        itemsize >>= shift
        #
        if valid_addressing_size(itemsize - 1):
            mc.LEA_ra(targetreg, (sourcereg, sourcereg,
                                  get_scale(itemsize - 1), 0))
        elif valid_addressing_size(itemsize):
            mc.LEA_ra(targetreg, (rx86.NO_BASE_REGISTER, sourcereg,
                                  get_scale(itemsize), 0))
        else:
            mc.IMUL_rri(targetreg, sourcereg, itemsize)
        #
        return shift

    def genop_discard_increment_debug_counter(self, op, arglocs):
        # The argument should be an immediate address.  This should
        # generate code equivalent to a GETFIELD_RAW, an ADD(1), and a
        # SETFIELD_RAW.  Here we use the direct from-memory-to-memory
        # increment operation of x86.
        base_loc, = arglocs
        self.mc.INC(mem(base_loc, 0))

    def genop_discard_gc_store(self, op, arglocs):
        base_loc, ofs_loc, value_loc, size_loc = arglocs
        assert isinstance(size_loc, ImmedLoc)
        scale = get_scale(size_loc.value)
        dest_addr = AddressLoc(base_loc, ofs_loc, 0, 0)
        self.save_into_mem(dest_addr, value_loc, size_loc)

    def genop_discard_gc_store_indexed(self, op, arglocs):
        base_loc, ofs_loc, value_loc, factor_loc, offset_loc, size_loc = arglocs
        assert isinstance(size_loc, ImmedLoc)
        scale = get_scale(factor_loc.value)
        dest_addr = AddressLoc(base_loc, ofs_loc, scale, offset_loc.value)
        self.save_into_mem(dest_addr, value_loc, size_loc)

    # genop_discard_setfield_raw = genop_discard_setfield_gc

    def genop_math_read_timestamp(self, op, arglocs, resloc):
        self.mc.RDTSC()
        if longlong.is_64_bit:
            self.mc.SHL_ri(edx.value, 32)
            self.mc.OR_rr(edx.value, eax.value)
        else:
            loc1, = arglocs
            self.mc.MOVD32_xr(loc1.value, edx.value)
            self.mc.MOVD32_xr(resloc.value, eax.value)
            self.mc.PUNPCKLDQ_xx(resloc.value, loc1.value)

    def genop_guard_guard_true(self, guard_op, guard_token, locs, resloc):
        self.implement_guard(guard_token)
    genop_guard_guard_nonnull = genop_guard_guard_true

    def genop_guard_guard_false(self, guard_op, guard_token, locs, resloc):
        self.guard_success_cc = rx86.invert_condition(self.guard_success_cc)
        self.implement_guard(guard_token)
    genop_guard_guard_isnull = genop_guard_guard_false

    def genop_guard_guard_no_exception(self, guard_op, guard_token, locs, ign):
        self.mc.CMP(heap(self.cpu.pos_exception()), imm0)
        self.guard_success_cc = rx86.Conditions['Z']
        self.implement_guard(guard_token)
        # If the previous operation was a COND_CALL, overwrite its conditional
        # jump to jump over this GUARD_NO_EXCEPTION as well, if we can
        if self._find_nearby_operation(-1).getopnum() == rop.COND_CALL:
            jmp_adr = self.previous_cond_call_jcond
            offset = self.mc.get_relative_pos() - jmp_adr
            if offset <= 127:
                self.mc.overwrite(jmp_adr-1, chr(offset))

    def genop_guard_guard_not_invalidated(self, guard_op, guard_token,
                                          locs, ign):
        pos = self.mc.get_relative_pos() + 1 # after potential jmp
        guard_token.pos_jump_offset = pos
        self.pending_guard_tokens.append(guard_token)

    def genop_guard_guard_exception(self, guard_op, guard_token, locs, resloc):
        loc = locs[0]
        loc1 = locs[1]
        self.mc.MOV(loc1, heap(self.cpu.pos_exception()))
        self.mc.CMP(loc1, loc)
        self.guard_success_cc = rx86.Conditions['E']
        self.implement_guard(guard_token)
        self._store_and_reset_exception(self.mc, resloc)

    def genop_save_exc_class(self, op, arglocs, resloc):
        self.mc.MOV(resloc, heap(self.cpu.pos_exception()))

    def genop_save_exception(self, op, arglocs, resloc):
        self._store_and_reset_exception(self.mc, resloc)

    def genop_discard_restore_exception(self, op, arglocs):
        self._restore_exception(self.mc, arglocs[1], arglocs[0])

    def _store_and_reset_exception(self, mc, excvalloc=None, exctploc=None,
                                   tmploc=None):
        """ Resest the exception. If excvalloc is None, then store it on the
        frame in jf_guard_exc
        """
        if excvalloc is not None:
            assert excvalloc.is_core_reg()
            mc.MOV(excvalloc, heap(self.cpu.pos_exc_value()))
        elif tmploc is not None: # if both are None, just ignore
            ofs = self.cpu.get_ofs_of_frame_field('jf_guard_exc')
            mc.MOV(tmploc, heap(self.cpu.pos_exc_value()))
            mc.MOV(RawEbpLoc(ofs), tmploc)
        if exctploc is not None:
            assert exctploc.is_core_reg()
            mc.MOV(exctploc, heap(self.cpu.pos_exception()))

        mc.MOV(heap(self.cpu.pos_exception()), imm0)
        mc.MOV(heap(self.cpu.pos_exc_value()), imm0)

    def _restore_exception(self, mc, excvalloc, exctploc, tmploc=None):
        if excvalloc is not None:
            mc.MOV(heap(self.cpu.pos_exc_value()), excvalloc)
        else:
            assert tmploc is not None
            ofs = self.cpu.get_ofs_of_frame_field('jf_guard_exc')
            mc.MOV(tmploc, RawEbpLoc(ofs))
            mc.MOV_bi(ofs, 0)
            mc.MOV(heap(self.cpu.pos_exc_value()), tmploc)
        mc.MOV(heap(self.cpu.pos_exception()), exctploc)

    def genop_int_add_ovf(self, op, arglocs, resloc):
        self.genop_int_add(op, arglocs, resloc)
        self.guard_success_cc = rx86.Conditions['NO']

    def genop_int_sub_ovf(self, op, arglocs, resloc):
        self.genop_int_sub(op, arglocs, resloc)
        self.guard_success_cc = rx86.Conditions['NO']

    def genop_int_mul_ovf(self, op, arglocs, resloc):
        self.genop_int_mul(op, arglocs, resloc)
        self.guard_success_cc = rx86.Conditions['NO']

    genop_guard_guard_no_overflow = genop_guard_guard_true
    genop_guard_guard_overflow    = genop_guard_guard_false

    def genop_guard_guard_value(self, guard_op, guard_token, locs, ign):
        if guard_op.getarg(0).type == FLOAT:
            assert guard_op.getarg(1).type == FLOAT
            self.mc.UCOMISD(locs[0], locs[1])
        else:
            self.mc.CMP(locs[0], locs[1])
        self.guard_success_cc = rx86.Conditions['E']
        self.implement_guard(guard_token)

    def _cmp_guard_class(self, locs):
        loc_ptr = locs[0]
        loc_classptr = locs[1]
        offset = self.cpu.vtable_offset
        if offset is not None:
            self.mc.CMP(mem(loc_ptr, offset), loc_classptr)
        else:
            assert isinstance(loc_classptr, ImmedLoc)
            classptr = loc_classptr.value
            expected_typeid = (self.cpu.gc_ll_descr
                    .get_typeid_from_classptr_if_gcremovetypeptr(classptr))
            self._cmp_guard_gc_type(loc_ptr, ImmedLoc(expected_typeid))

    def _cmp_guard_gc_type(self, loc_ptr, loc_expected_typeid):
        # Note that the typeid half-word is at offset 0 on a little-endian
        # machine; it would be at offset 2 or 4 on a big-endian machine.
        assert self.cpu.supports_guard_gc_type
        if IS_X86_32:
            self.mc.CMP16(mem(loc_ptr, 0), loc_expected_typeid)
        else:
            assert isinstance(loc_expected_typeid, ImmedLoc)
            self.mc.CMP32_mi((loc_ptr.value, 0), loc_expected_typeid.value)

    def genop_guard_guard_class(self, guard_op, guard_token, locs, ign):
        self._cmp_guard_class(locs)
        self.guard_success_cc = rx86.Conditions['E']
        self.implement_guard(guard_token)

    def genop_guard_guard_nonnull_class(self, guard_op, guard_token, locs, ign):
        self.mc.CMP(locs[0], imm1)
        # Patched below
        self.mc.J_il8(rx86.Conditions['B'], 0)
        jb_location = self.mc.get_relative_pos()
        self._cmp_guard_class(locs)
        # patch the JB above
        offset = self.mc.get_relative_pos() - jb_location
        assert 0 < offset <= 127
        self.mc.overwrite(jb_location-1, chr(offset))
        #
        self.guard_success_cc = rx86.Conditions['E']
        self.implement_guard(guard_token)

    def genop_guard_guard_gc_type(self, guard_op, guard_token, locs, ign):
        self._cmp_guard_gc_type(locs[0], locs[1])
        self.guard_success_cc = rx86.Conditions['E']
        self.implement_guard(guard_token)

    def genop_guard_guard_is_object(self, guard_op, guard_token, locs, ign):
        assert self.cpu.supports_guard_gc_type
        [loc_object, loc_typeid] = locs
        # idea: read the typeid, fetch the field 'infobits' from the big
        # typeinfo table, and check the flag 'T_IS_RPYTHON_INSTANCE'.
        if IS_X86_32:
            self.mc.MOVZX16(loc_typeid, mem(loc_object, 0))
        else:
            self.mc.MOV32(loc_typeid, mem(loc_object, 0))
        #
        base_type_info, shift_by, sizeof_ti = (
            self.cpu.gc_ll_descr.get_translated_info_for_typeinfo())
        infobits_offset, IS_OBJECT_FLAG = (
            self.cpu.gc_ll_descr.get_translated_info_for_guard_is_object())
        loc_infobits = addr_add(imm(base_type_info), loc_typeid,
                                scale=shift_by, offset=infobits_offset)
        self.mc.TEST8(loc_infobits, imm(IS_OBJECT_FLAG))
        #
        self.guard_success_cc = rx86.Conditions['NZ']
        self.implement_guard(guard_token)

    def genop_guard_guard_subclass(self, guard_op, guard_token, locs, ign):
        assert self.cpu.supports_guard_gc_type
        [loc_object, loc_check_against_class, loc_tmp] = locs
        assert isinstance(loc_object, RegLoc)
        assert isinstance(loc_tmp, RegLoc)
        offset = self.cpu.vtable_offset
        offset2 = self.cpu.subclassrange_min_offset
        if offset is not None:
            # read this field to get the vtable pointer
            self.mc.MOV_rm(loc_tmp.value, (loc_object.value, offset))
            # read the vtable's subclassrange_min field
            self.mc.MOV_rm(loc_tmp.value, (loc_tmp.value, offset2))
        else:
            # read the typeid
            if IS_X86_32:
                self.mc.MOVZX16(loc_tmp, mem(loc_object, 0))
            else:
                self.mc.MOV32(loc_tmp, mem(loc_object, 0))
            # read the vtable's subclassrange_min field, as a single
            # step with the correct offset
            base_type_info, shift_by, sizeof_ti = (
                self.cpu.gc_ll_descr.get_translated_info_for_typeinfo())
            self.mc.MOV(loc_tmp, addr_add(imm(base_type_info), loc_tmp,
                                          scale = shift_by,
                                          offset = sizeof_ti + offset2))
        # get the two bounds to check against
        vtable_ptr = loc_check_against_class.getint()
        vtable_ptr = rffi.cast(rclass.CLASSTYPE, vtable_ptr)
        check_min = vtable_ptr.subclassrange_min
        check_max = vtable_ptr.subclassrange_max
        # check by doing the unsigned comparison (tmp - min) < (max - min)
        self.mc.SUB_ri(loc_tmp.value, check_min)
        self.mc.CMP_ri(loc_tmp.value, check_max - check_min)
        # the guard passes if we get a result of "below"
        self.guard_success_cc = rx86.Conditions['B']
        self.implement_guard(guard_token)

    def implement_guard_recovery(self, guard_opnum, faildescr, failargs,
                                 fail_locs, frame_depth):
        gcmap = allocate_gcmap(self, frame_depth, JITFRAME_FIXED_SIZE)
        faildescrindex = self.get_gcref_from_faildescr(faildescr)
        return GuardToken(self.cpu, gcmap, faildescr, failargs, fail_locs,
                          guard_opnum, frame_depth, faildescrindex)

    def generate_propagate_error_64(self):
        assert WORD == 8
        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        startpos = self.mc.get_relative_pos()
        self.mc.JMP(imm(self.propagate_exception_path))
        return startpos

    def generate_quick_failure(self, guardtok, regalloc):
        """ Gather information about failure
        """
        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        startpos = self.mc.get_relative_pos()
        #
        self._update_at_exit(guardtok.fail_locs, guardtok.failargs,
                             guardtok.faildescr, regalloc)
        #
        faildescrindex, target = self.store_info_on_descr(startpos, guardtok)
        if IS_X86_64:
            self.mc.PUSH_p(0)     # %rip-relative
            self._patch_load_from_gc_table(faildescrindex)
        elif IS_X86_32:
            self.mc.PUSH_j(self._addr_from_gc_table(faildescrindex))
        self.push_gcmap(self.mc, guardtok.gcmap, push=True)
        self.mc.JMP(imm(target))
        return startpos

    def push_gcmap(self, mc, gcmap, push=False, store=False):
        if push:
            mc.PUSH(imm(rffi.cast(lltype.Signed, gcmap)))
        else:
            assert store
            ofs = self.cpu.get_ofs_of_frame_field('jf_gcmap')
            mc.MOV(raw_stack(ofs), imm(rffi.cast(lltype.Signed, gcmap)))

    def pop_gcmap(self, mc):
        ofs = self.cpu.get_ofs_of_frame_field('jf_gcmap')
        mc.MOV_bi(ofs, 0)

    def new_stack_loc(self, i, tp):
        base_ofs = self.cpu.get_baseofs_of_frame_field()
        return FrameLoc(i, get_ebp_ofs(base_ofs, i), tp)

    def setup_failure_recovery(self):
        self.failure_recovery_code = [0, 0, 0, 0]

    def _push_all_regs_to_frame(self, mc, ignored_regs, withfloats,
                                callee_only=False):
        # Push all general purpose registers
        base_ofs = self.cpu.get_baseofs_of_frame_field()
        if callee_only:
            regs = gpr_reg_mgr_cls.save_around_call_regs
        else:
            regs = gpr_reg_mgr_cls.all_regs
        for gpr in regs:
            if gpr not in ignored_regs:
                v = gpr_reg_mgr_cls.all_reg_indexes[gpr.value]
                mc.MOV_br(v * WORD + base_ofs, gpr.value)
        if withfloats:
            if IS_X86_64:
                coeff = 1
            else:
                coeff = 2
            # Push all XMM regs
            ofs = len(gpr_reg_mgr_cls.all_regs)
            for i in range(len(xmm_reg_mgr_cls.all_regs)):
                mc.MOVSD_bx((ofs + i * coeff) * WORD + base_ofs, i)

    def _pop_all_regs_from_frame(self, mc, ignored_regs, withfloats,
                                 callee_only=False):
        # Pop all general purpose registers
        base_ofs = self.cpu.get_baseofs_of_frame_field()
        if callee_only:
            regs = gpr_reg_mgr_cls.save_around_call_regs
        else:
            regs = gpr_reg_mgr_cls.all_regs
        for gpr in regs:
            if gpr not in ignored_regs:
                v = gpr_reg_mgr_cls.all_reg_indexes[gpr.value]
                mc.MOV_rb(gpr.value, v * WORD + base_ofs)
        if withfloats:
            # Pop all XMM regs
            if IS_X86_64:
                coeff = 1
            else:
                coeff = 2
            ofs = len(gpr_reg_mgr_cls.all_regs)
            for i in range(len(xmm_reg_mgr_cls.all_regs)):
                mc.MOVSD_xb(i, (ofs + i * coeff) * WORD + base_ofs)

    def _build_failure_recovery(self, exc, withfloats=False):
        mc = codebuf.MachineCodeBlockWrapper()
        # this is jumped to, from a stack that has DEFAULT_FRAME_BYTES
        # followed by 2 extra words just pushed
        mc.force_frame_size(DEFAULT_FRAME_BYTES + 2 * WORD)
        self.mc = mc

        self._push_all_regs_to_frame(mc, [], withfloats)

        if exc:
            # We might have an exception pending.  Load it into ebx...
            mc.MOV(ebx, heap(self.cpu.pos_exc_value()))
            mc.MOV(heap(self.cpu.pos_exception()), imm0)
            mc.MOV(heap(self.cpu.pos_exc_value()), imm0)
            # ...and save ebx into 'jf_guard_exc'
            offset = self.cpu.get_ofs_of_frame_field('jf_guard_exc')
            mc.MOV_br(offset, ebx.value)

        # fill in the jf_descr and jf_gcmap fields of the frame according
        # to which failure we are resuming from.  These are constants
        # pushed on the stack just before we jump to the current helper,
        # in generate_quick_failure().
        ofs = self.cpu.get_ofs_of_frame_field('jf_descr')
        ofs2 = self.cpu.get_ofs_of_frame_field('jf_gcmap')
        mc.POP_b(ofs2)
        mc.POP_b(ofs)

        # now we return from the complete frame, which starts from
        # _call_header_with_stack_check().  The _call_footer below does it.
        self._call_footer()
        rawstart = mc.materialize(self.cpu, [])
        self.failure_recovery_code[exc + 2 * withfloats] = rawstart
        self.mc = None

    def genop_finish(self, op, arglocs, result_loc):
        base_ofs = self.cpu.get_baseofs_of_frame_field()
        if len(arglocs) > 0:
            [return_val] = arglocs
            if op.getarg(0).type == FLOAT and not IS_X86_64:
                size = WORD * 2
            else:
                size = WORD
            self.save_into_mem(raw_stack(base_ofs), return_val, imm(size))
        ofs = self.cpu.get_ofs_of_frame_field('jf_descr')
        
        descr = op.getdescr()
        faildescrindex = self.get_gcref_from_faildescr(descr)
        if IS_X86_64:
            self.mc.MOV_rp(eax.value, 0)
            self._patch_load_from_gc_table(faildescrindex)
        elif IS_X86_32:
            self.mc.MOV_rj(eax.value, self._addr_from_gc_table(faildescrindex))
        self.mov(eax, RawEbpLoc(ofs))

        arglist = op.getarglist()
        if arglist and arglist[0].type == REF:
            if self._finish_gcmap:
                # we're returning with a guard_not_forced_2, and
                # additionally we need to say that eax/rax contains
                # a reference too:
                self._finish_gcmap[0] |= r_uint(1)
                gcmap = self._finish_gcmap
            else:
                gcmap = self.gcmap_for_finish
            self.push_gcmap(self.mc, gcmap, store=True)
        elif self._finish_gcmap:
            # we're returning with a guard_not_forced_2
            gcmap = self._finish_gcmap
            self.push_gcmap(self.mc, gcmap, store=True)
        else:
            # note that the 0 here is redundant, but I would rather
            # keep that one and kill all the others
            ofs = self.cpu.get_ofs_of_frame_field('jf_gcmap')
            self.mc.MOV_bi(ofs, 0)
        # exit function
        self._call_footer()

    def implement_guard(self, guard_token):
        # These jumps are patched later.
        assert self.guard_success_cc >= 0
        self.mc.J_il(rx86.invert_condition(self.guard_success_cc), 0)
        self.guard_success_cc = rx86.cond_none
        guard_token.pos_jump_offset = self.mc.get_relative_pos() - 4
        self.pending_guard_tokens.append(guard_token)

    def _genop_real_call(self, op, arglocs, resloc):
        self._genop_call(op, arglocs, resloc)
    genop_call_i = _genop_real_call
    genop_call_r = _genop_real_call
    genop_call_f = _genop_real_call
    genop_call_n = _genop_real_call

    def _genop_call(self, op, arglocs, resloc, is_call_release_gil=False):
        from rpython.jit.backend.llsupport.descr import CallDescr

        func_index = 2 + is_call_release_gil
        cb = callbuilder.CallBuilder(self, arglocs[func_index],
                                     arglocs[func_index+1:], resloc)

        descr = op.getdescr()
        assert isinstance(descr, CallDescr)
        cb.callconv = descr.get_call_conv()
        cb.argtypes = descr.get_arg_types()
        cb.restype  = descr.get_result_type()
        sizeloc = arglocs[0]
        assert isinstance(sizeloc, ImmedLoc)
        cb.ressize = sizeloc.value
        signloc = arglocs[1]
        assert isinstance(signloc, ImmedLoc)
        cb.ressign = signloc.value

        if is_call_release_gil:
            saveerrloc = arglocs[2]
            assert isinstance(saveerrloc, ImmedLoc)
            cb.emit_call_release_gil(saveerrloc.value)
        else:
            effectinfo = descr.get_extra_info()
            if effectinfo is None or effectinfo.check_can_collect():
                cb.emit()
            else:
                cb.emit_no_collect()

    def _store_force_index(self, guard_op):
        assert (guard_op.getopnum() == rop.GUARD_NOT_FORCED or
                guard_op.getopnum() == rop.GUARD_NOT_FORCED_2)
        faildescr = guard_op.getdescr()
        ofs = self.cpu.get_ofs_of_frame_field('jf_force_descr')

        faildescrindex = self.get_gcref_from_faildescr(faildescr)
        if IS_X86_64:
            self.mc.MOV_rp(X86_64_SCRATCH_REG.value, 0)
            self._patch_load_from_gc_table(faildescrindex)
            self.mc.MOV(raw_stack(ofs), X86_64_SCRATCH_REG)
        elif IS_X86_32:
            # XXX need a scratch reg here for efficiency; be more clever
            self.mc.PUSH_j(self._addr_from_gc_table(faildescrindex))
            self.mc.POP(raw_stack(ofs))

    def _find_nearby_operation(self, delta):
        regalloc = self._regalloc
        return regalloc.operations[regalloc.rm.position + delta]

    def genop_guard_guard_not_forced(self, guard_op, guard_token, locs, resloc):
        ofs = self.cpu.get_ofs_of_frame_field('jf_descr')
        self.mc.CMP_bi(ofs, 0)
        self.guard_success_cc = rx86.Conditions['E']
        self.implement_guard(guard_token)

    def _genop_call_may_force(self, op, arglocs, result_loc):
        self._store_force_index(self._find_nearby_operation(+1))
        self._genop_call(op, arglocs, result_loc)
    genop_call_may_force_i = _genop_call_may_force
    genop_call_may_force_r = _genop_call_may_force
    genop_call_may_force_f = _genop_call_may_force
    genop_call_may_force_n = _genop_call_may_force

    def _genop_call_release_gil(self, op, arglocs, result_loc):
        self._store_force_index(self._find_nearby_operation(+1))
        self._genop_call(op, arglocs, result_loc, is_call_release_gil=True)
    genop_call_release_gil_i = _genop_call_release_gil
    genop_call_release_gil_f = _genop_call_release_gil
    genop_call_release_gil_n = _genop_call_release_gil

    def imm(self, v):
        return imm(v)

    # ------------------- CALL ASSEMBLER --------------------------

    def _genop_call_assembler(self, op, arglocs, result_loc):
        if len(arglocs) == 2:
            [argloc, vloc] = arglocs
        else:
            [argloc] = arglocs
            vloc = self.imm(0)
        self._store_force_index(self._find_nearby_operation(+1))
        self.call_assembler(op, argloc, vloc, result_loc, eax)
    genop_call_assembler_i = _genop_call_assembler
    genop_call_assembler_r = _genop_call_assembler
    genop_call_assembler_f = _genop_call_assembler
    genop_call_assembler_n = _genop_call_assembler

    def _call_assembler_emit_call(self, addr, argloc, _):
        threadlocal_loc = RawEspLoc(THREADLOCAL_OFS, INT)
        if self._is_asmgcc():
            # We need to remove the bit "already seen during the
            # previous minor collection" instead of passing this
            # value directly.
            if IS_X86_64:
                tmploc = esi    # already the correct place
                if argloc is tmploc:
                    # this case is theoretical only so far: in practice,
                    # argloc is always eax, never esi
                    self.mc.MOV_rr(edi.value, esi.value)
                    argloc = edi
            else:
                tmploc = eax
                if tmploc is argloc:
                    tmploc = edx
            self.mc.MOV(tmploc, threadlocal_loc)
            self.mc.AND_ri(tmploc.value, ~1)
            threadlocal_loc = tmploc
        #
        self.simple_call(addr, [argloc, threadlocal_loc])

    def _call_assembler_emit_helper_call(self, addr, arglocs, result_loc):
        self.simple_call(addr, arglocs, result_loc)

    def _call_assembler_check_descr(self, value, tmploc):
        ofs = self.cpu.get_ofs_of_frame_field('jf_descr')
        self.mc.CMP(mem(eax, ofs), imm(value))
        # patched later
        self.mc.J_il8(rx86.Conditions['E'], 0) # goto B if we get 'done_with_this_frame'
        return self.mc.get_relative_pos()

    def _call_assembler_patch_je(self, result_loc, je_location):
        if (IS_X86_32 and isinstance(result_loc, FrameLoc) and
            result_loc.type == FLOAT):
            self.mc.FSTPL_b(result_loc.value)
        self.mc.JMP_l8(0) # jump to done, patched later
        jmp_location = self.mc.get_relative_pos()
        #
        offset = jmp_location - je_location
        assert 0 < offset <= 127
        self.mc.overwrite(je_location - 1, chr(offset))
        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        #
        return jmp_location

    def _call_assembler_load_result(self, op, result_loc):
        if op.type != 'v':
            # load the return value from the dead frame's value index 0
            kind = op.type
            descr = self.cpu.getarraydescr_for_frame(kind)
            ofs = self.cpu.unpack_arraydescr(descr)
            if kind == FLOAT:
                self.mc.MOVSD_xm(xmm0.value, (eax.value, ofs))
                if result_loc is not xmm0:
                    self.mc.MOVSD(result_loc, xmm0)
            else:
                assert result_loc is eax
                self.mc.MOV_rm(eax.value, (eax.value, ofs))

    def _call_assembler_patch_jmp(self, jmp_location):
        offset = self.mc.get_relative_pos() - jmp_location
        assert 0 <= offset <= 127
        self.mc.overwrite(jmp_location - 1, chr(offset))

    # ------------------- END CALL ASSEMBLER -----------------------

    def _write_barrier_fastpath(self, mc, descr, arglocs, array=False,
                                is_frame=False):
        # Write code equivalent to write_barrier() in the GC: it checks
        # a flag in the object at arglocs[0], and if set, it calls a
        # helper piece of assembler.  The latter saves registers as needed
        # and call the function remember_young_pointer() from the GC.
        if we_are_translated():
            cls = self.cpu.gc_ll_descr.has_write_barrier_class()
            assert cls is not None and isinstance(descr, cls)
        #
        card_marking = False
        mask = descr.jit_wb_if_flag_singlebyte
        if array and descr.jit_wb_cards_set != 0:
            # assumptions the rest of the function depends on:
            assert (descr.jit_wb_cards_set_byteofs ==
                    descr.jit_wb_if_flag_byteofs)
            assert descr.jit_wb_cards_set_singlebyte == -0x80
            card_marking = True
            mask = descr.jit_wb_if_flag_singlebyte | -0x80
        #
        loc_base = arglocs[0]
        if is_frame:
            assert loc_base is ebp
            loc = raw_stack(descr.jit_wb_if_flag_byteofs)
        else:
            loc = addr_add_const(loc_base, descr.jit_wb_if_flag_byteofs)
        mc.TEST8(loc, imm(mask))
        mc.J_il8(rx86.Conditions['Z'], 0) # patched later
        jz_location = mc.get_relative_pos()

        # for cond_call_gc_wb_array, also add another fast path:
        # if GCFLAG_CARDS_SET, then we can just set one bit and be done
        if card_marking:
            # GCFLAG_CARDS_SET is in this byte at 0x80, so this fact can
            # been checked by the status flags of the previous TEST8
            mc.J_il8(rx86.Conditions['S'], 0) # patched later
            js_location = mc.get_relative_pos()
        else:
            js_location = 0

        # Write only a CALL to the helper prepared in advance, passing it as
        # argument the address of the structure we are writing into
        # (the first argument to COND_CALL_GC_WB).
        helper_num = card_marking
        if is_frame:
            helper_num = 4
        elif self._regalloc is not None and self._regalloc.xrm.reg_bindings:
            helper_num += 2
        if self.wb_slowpath[helper_num] == 0:    # tests only
            assert not we_are_translated()
            self.cpu.gc_ll_descr.write_barrier_descr = descr
            self._build_wb_slowpath(card_marking,
                                    bool(self._regalloc.xrm.reg_bindings))
            assert self.wb_slowpath[helper_num] != 0
        #
        if not is_frame:
            mc.PUSH(loc_base)
        mc.CALL(imm(self.wb_slowpath[helper_num]))
        if not is_frame:
            mc.stack_frame_size_delta(-WORD)

        if card_marking:
            # The helper ends again with a check of the flag in the object.
            # So here, we can simply write again a 'JNS', which will be
            # taken if GCFLAG_CARDS_SET is still not set.
            mc.J_il8(rx86.Conditions['NS'], 0) # patched later
            jns_location = mc.get_relative_pos()
            #
            # patch the JS above
            offset = mc.get_relative_pos() - js_location
            assert 0 < offset <= 127
            mc.overwrite(js_location-1, chr(offset))
            #
            # case GCFLAG_CARDS_SET: emit a few instructions to do
            # directly the card flag setting
            loc_index = arglocs[1]
            if isinstance(loc_index, RegLoc):
                if IS_X86_64 and isinstance(loc_base, RegLoc):
                    # copy loc_index into r11
                    tmp1 = X86_64_SCRATCH_REG
                    mc.MOV_rr(tmp1.value, loc_index.value)
                    final_pop = False
                else:
                    # must save the register loc_index before it is mutated
                    mc.PUSH_r(loc_index.value)
                    tmp1 = loc_index
                    final_pop = True
                # SHR tmp, card_page_shift
                mc.SHR_ri(tmp1.value, descr.jit_wb_card_page_shift)
                # XOR tmp, -8
                mc.XOR_ri(tmp1.value, -8)
                # BTS [loc_base], tmp
                mc.BTS(addr_add_const(loc_base, 0), tmp1)
                # done
                if final_pop:
                    mc.POP_r(loc_index.value)
                #
            elif isinstance(loc_index, ImmedLoc):
                byte_index = loc_index.value >> descr.jit_wb_card_page_shift
                byte_ofs = ~(byte_index >> 3)
                byte_val = 1 << (byte_index & 7)
                mc.OR8(addr_add_const(loc_base, byte_ofs), imm(byte_val))
            else:
                raise AssertionError("index is neither RegLoc nor ImmedLoc")
            #
            # patch the JNS above
            offset = mc.get_relative_pos() - jns_location
            assert 0 < offset <= 127
            mc.overwrite(jns_location-1, chr(offset))

        # patch the JZ above
        offset = mc.get_relative_pos() - jz_location
        assert 0 < offset <= 127
        mc.overwrite(jz_location-1, chr(offset))

    def genop_discard_cond_call_gc_wb(self, op, arglocs):
        self._write_barrier_fastpath(self.mc, op.getdescr(), arglocs)

    def genop_discard_cond_call_gc_wb_array(self, op, arglocs):
        self._write_barrier_fastpath(self.mc, op.getdescr(), arglocs,
                                     array=True)

    def not_implemented_op_discard(self, op, arglocs):
        not_implemented("not implemented operation: %s" % op.getopname())

    def not_implemented_op(self, op, arglocs, resloc):
        not_implemented("not implemented operation with res: %s" %
                        op.getopname())

    def not_implemented_op_guard(self, guard_op, guard_token, locs, resloc):
        not_implemented("not implemented operation (guard): %s" %
                        guard_op.getopname())

    def closing_jump(self, target_token):
        target = target_token._ll_loop_code
        if target_token in self.target_tokens_currently_compiling:
            curpos = self.mc.get_relative_pos() + 5
            self.mc.JMP_l(target - curpos)
        else:
            self.mc.JMP(imm(target))

    def label(self):
        self._check_frame_depth_debug(self.mc)

    def cond_call(self, op, gcmap, imm_func, arglocs):
        assert self.guard_success_cc >= 0
        self.mc.J_il8(rx86.invert_condition(self.guard_success_cc), 0)
                                                            # patched later
        jmp_adr = self.mc.get_relative_pos()
        self.guard_success_cc = rx86.cond_none
        #
        self.push_gcmap(self.mc, gcmap, store=True)
        #
        # first save away the 4 registers from 'cond_call_register_arguments'
        # plus the register 'eax'
        base_ofs = self.cpu.get_baseofs_of_frame_field()
        should_be_saved = self._regalloc.rm.reg_bindings.values()
        for gpr in cond_call_register_arguments + [eax]:
            if gpr not in should_be_saved:
                continue
            v = gpr_reg_mgr_cls.all_reg_indexes[gpr.value]
            self.mc.MOV_br(v * WORD + base_ofs, gpr.value)
        #
        # load the 0-to-4 arguments into these registers
        from rpython.jit.backend.x86.jump import remap_frame_layout
        remap_frame_layout(self, arglocs,
                           cond_call_register_arguments[:len(arglocs)],
                           X86_64_SCRATCH_REG if IS_X86_64 else None)
        #
        # load the constant address of the function to call into eax
        self.mc.MOV(eax, imm_func)
        #
        # figure out which variant of cond_call_slowpath to call, and call it
        callee_only = False
        floats = False
        if self._regalloc is not None:
            for reg in self._regalloc.rm.reg_bindings.values():
                if reg not in self._regalloc.rm.save_around_call_regs:
                    break
            else:
                callee_only = True
            if self._regalloc.xrm.reg_bindings:
                floats = True
        cond_call_adr = self.cond_call_slowpath[floats * 2 + callee_only]
        self.mc.CALL(imm(follow_jump(cond_call_adr)))
        # restoring the registers saved above, and doing pop_gcmap(), is left
        # to the cond_call_slowpath helper.  We never have any result value.
        offset = self.mc.get_relative_pos() - jmp_adr
        assert 0 < offset <= 127
        self.mc.overwrite(jmp_adr-1, chr(offset))
        # might be overridden again to skip over the following
        # guard_no_exception too
        self.previous_cond_call_jcond = jmp_adr

    def malloc_cond(self, nursery_free_adr, nursery_top_adr, size, gcmap):
        assert size & (WORD-1) == 0     # must be correctly aligned
        self.mc.MOV(ecx, heap(nursery_free_adr))
        self.mc.LEA_rm(edx.value, (ecx.value, size))
        self.mc.CMP(edx, heap(nursery_top_adr))
        self.mc.J_il8(rx86.Conditions['NA'], 0) # patched later
        jmp_adr = self.mc.get_relative_pos()
        # save the gcmap
        self.push_gcmap(self.mc, gcmap, store=True)
        self.mc.CALL(imm(follow_jump(self.malloc_slowpath)))
        offset = self.mc.get_relative_pos() - jmp_adr
        assert 0 < offset <= 127
        self.mc.overwrite(jmp_adr-1, chr(offset))
        self.mc.MOV(heap(nursery_free_adr), edx)

    def malloc_cond_varsize_frame(self, nursery_free_adr, nursery_top_adr,
                                  sizeloc, gcmap):
        if sizeloc is ecx:
            self.mc.MOV(edx, sizeloc)
            sizeloc = edx
        self.mc.MOV(ecx, heap(nursery_free_adr))
        if sizeloc is edx:
            self.mc.ADD_rr(edx.value, ecx.value)
        else:
            self.mc.LEA_ra(edx.value, (ecx.value, sizeloc.value, 0, 0))
        self.mc.CMP(edx, heap(nursery_top_adr))
        self.mc.J_il8(rx86.Conditions['NA'], 0) # patched later
        jmp_adr = self.mc.get_relative_pos()
        # save the gcmap
        self.push_gcmap(self.mc, gcmap, store=True)
        self.mc.CALL(imm(follow_jump(self.malloc_slowpath)))
        offset = self.mc.get_relative_pos() - jmp_adr
        assert 0 < offset <= 127
        self.mc.overwrite(jmp_adr-1, chr(offset))
        self.mc.MOV(heap(nursery_free_adr), edx)

    def malloc_cond_varsize(self, kind, nursery_free_adr, nursery_top_adr,
                            lengthloc, itemsize, maxlength, gcmap,
                            arraydescr):
        from rpython.jit.backend.llsupport.descr import ArrayDescr
        assert isinstance(arraydescr, ArrayDescr)

        # lengthloc is the length of the array, which we must not modify!
        assert lengthloc is not ecx and lengthloc is not edx
        if isinstance(lengthloc, RegLoc):
            varsizeloc = lengthloc
        else:
            self.mc.MOV(edx, lengthloc)
            varsizeloc = edx

        self.mc.CMP(varsizeloc, imm(maxlength))
        self.mc.J_il8(rx86.Conditions['A'], 0) # patched later
        jmp_adr0 = self.mc.get_relative_pos()

        self.mc.MOV(ecx, heap(nursery_free_adr))
        if valid_addressing_size(itemsize):
            shift = get_scale(itemsize)
        else:
            shift = self._imul_const_scaled(self.mc, edx.value,
                                            varsizeloc.value, itemsize)
            varsizeloc = edx

        # now varsizeloc is a register != ecx.  The size of
        # the variable part of the array is (varsizeloc << shift)
        assert arraydescr.basesize >= self.gc_minimal_size_in_nursery
        constsize = arraydescr.basesize + self.gc_size_of_header
        force_realignment = (itemsize % WORD) != 0
        if force_realignment:
            constsize += WORD - 1
        self.mc.LEA_ra(edx.value, (ecx.value, varsizeloc.value, shift,
                                   constsize))
        if force_realignment:
            self.mc.AND_ri(edx.value, ~(WORD - 1))
        # now edx contains the total size in bytes, rounded up to a multiple
        # of WORD, plus nursery_free_adr
        self.mc.CMP(edx, heap(nursery_top_adr))
        self.mc.J_il8(rx86.Conditions['NA'], 0) # patched later
        jmp_adr1 = self.mc.get_relative_pos()
        #
        offset = self.mc.get_relative_pos() - jmp_adr0
        assert 0 < offset <= 127
        self.mc.overwrite(jmp_adr0-1, chr(offset))
        # save the gcmap
        self.push_gcmap(self.mc, gcmap, store=True)
        if kind == rewrite.FLAG_ARRAY:
            self.mc.MOV_si(WORD, itemsize)
            self.mc.MOV(edx, lengthloc)
            self.mc.MOV_ri(ecx.value, arraydescr.tid)
            addr = self.malloc_slowpath_varsize
        else:
            if kind == rewrite.FLAG_STR:
                addr = self.malloc_slowpath_str
            else:
                assert kind == rewrite.FLAG_UNICODE
                addr = self.malloc_slowpath_unicode
            self.mc.MOV(edx, lengthloc)
        self.mc.CALL(imm(follow_jump(addr)))
        self.mc.JMP_l8(0)      # jump to done, patched later
        jmp_location = self.mc.get_relative_pos()
        #
        offset = self.mc.get_relative_pos() - jmp_adr1
        assert 0 < offset <= 127
        self.mc.overwrite(jmp_adr1-1, chr(offset))
        self.mc.force_frame_size(DEFAULT_FRAME_BYTES)
        # write down the tid, but not if it's the result of the CALL
        self.mc.MOV(mem(ecx, 0), imm(arraydescr.tid))
        # while we're at it, this line is not needed if we've done the CALL
        self.mc.MOV(heap(nursery_free_adr), edx)
        #
        offset = self.mc.get_relative_pos() - jmp_location
        assert 0 < offset <= 127
        self.mc.overwrite(jmp_location - 1, chr(offset))

    def store_force_descr(self, op, fail_locs, frame_depth):
        guard_token = self.implement_guard_recovery(op.opnum,
                                                    op.getdescr(),
                                                    op.getfailargs(),
                                                    fail_locs, frame_depth)
        self._finish_gcmap = guard_token.gcmap
        self._store_force_index(op)
        self.store_info_on_descr(0, guard_token)

    def force_token(self, reg):
        # XXX kill me
        assert isinstance(reg, RegLoc)
        self.mc.MOV_rr(reg.value, ebp.value)

    def threadlocalref_get(self, offset, resloc, size, sign):
        # This loads the stack location THREADLOCAL_OFS into a
        # register, and then read the word at the given offset.
        # It is only supported if 'translate_support_code' is
        # true; otherwise, the execute_token() was done with a
        # dummy value for the stack location THREADLOCAL_OFS
        # 
        assert self.cpu.translate_support_code
        assert isinstance(resloc, RegLoc)
        self.mc.MOV_rs(resloc.value, THREADLOCAL_OFS)
        if self._is_asmgcc():
            self.mc.AND_ri(resloc.value, ~1)
        self.load_from_mem(resloc, addr_add_const(resloc, offset),
                           imm(size), imm(sign))

    def genop_discard_zero_array(self, op, arglocs):
        (base_loc, startindex_loc, bytes_loc,
         itemsize_loc, baseofs_loc, null_loc) = arglocs
        assert isinstance(bytes_loc, ImmedLoc)
        assert isinstance(itemsize_loc, ImmedLoc)
        assert isinstance(baseofs_loc, ImmedLoc)
        assert isinstance(null_loc, RegLoc) and null_loc.is_xmm
        baseofs = baseofs_loc.value
        nbytes = bytes_loc.value
        assert valid_addressing_size(itemsize_loc.value)
        scale = get_scale(itemsize_loc.value)
        null_reg_cleared = False
        i = 0
        while i < nbytes:
            addr = addr_add(base_loc, startindex_loc, baseofs + i, scale)
            current = nbytes - i
            if current >= 16 and self.cpu.supports_floats:
                current = 16
                if not null_reg_cleared:
                    self.mc.XORPS_xx(null_loc.value, null_loc.value)
                    null_reg_cleared = True
                self.mc.MOVUPS(addr, null_loc)
            else:
                if current >= WORD:
                    current = WORD
                elif current >= 4:
                    current = 4
                elif current >= 2:
                    current = 2
                self.save_into_mem(addr, imm0, imm(current))
            i += current


genop_discard_list = [Assembler386.not_implemented_op_discard] * rop._LAST
genop_list = [Assembler386.not_implemented_op] * rop._LAST
genop_llong_list = {}
genop_math_list = {}
genop_tlref_list = {}
genop_guard_list = [Assembler386.not_implemented_op_guard] * rop._LAST

import itertools
iterate = itertools.chain(Assembler386.__dict__.iteritems(),
                          VectorAssemblerMixin.__dict__.iteritems())
for name, value in iterate:
    if name.startswith('genop_discard_'):
        opname = name[len('genop_discard_'):]
        num = getattr(rop, opname.upper())
        genop_discard_list[num] = value
    elif name.startswith('genop_guard_'):
        opname = name[len('genop_guard_'):]
        num = getattr(rop, opname.upper())
        genop_guard_list[num] = value
    elif name.startswith('genop_llong_'):
        opname = name[len('genop_llong_'):]
        num = getattr(EffectInfo, 'OS_LLONG_' + opname.upper())
        genop_llong_list[num] = value
    elif name.startswith('genop_math_'):
        opname = name[len('genop_math_'):]
        num = getattr(EffectInfo, 'OS_MATH_' + opname.upper())
        genop_math_list[num] = value
    elif name.startswith('genop_'):
        opname = name[len('genop_'):]
        num = getattr(rop, opname.upper())
        genop_list[num] = value

# XXX: ri386 migration shims:
def addr_add(reg_or_imm1, reg_or_imm2, offset=0, scale=0):
    return AddressLoc(reg_or_imm1, reg_or_imm2, scale, offset)

def addr_add_const(reg_or_imm1, offset):
    return AddressLoc(reg_or_imm1, imm0, 0, offset)

def mem(loc, offset):
    return AddressLoc(loc, imm0, 0, offset)

def raw_stack(offset, type=INT):
    return RawEbpLoc(offset, type)

def heap(addr):
    return AddressLoc(ImmedLoc(addr), imm0, 0, 0)

def not_implemented(msg):
    msg = '[x86/asm] %s\n' % msg
    if we_are_translated():
        llop.debug_print(lltype.Void, msg)
    raise NotImplementedError(msg)

cond_call_register_arguments = [edi, esi, edx, ecx]

class BridgeAlreadyCompiled(Exception):
    pass
Back to Top