aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target/X86/X86FastISel.cpp
blob: 56bcfa30ff90722b81ba4d7c9894995872f78ace (plain)
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
//===-- X86FastISel.cpp - X86 FastISel implementation ---------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the X86-specific support for the FastISel class. Much
// of the target-specific code is generated by tablegen in the file
// X86GenFastISel.inc, which is #included here.
//
//===----------------------------------------------------------------------===//

#include "X86.h"
#include "X86CallingConv.h"
#include "X86InstrBuilder.h"
#include "X86MachineFunctionInfo.h"
#include "X86RegisterInfo.h"
#include "X86Subtarget.h"
#include "X86TargetMachine.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/FastISel.h"
#include "llvm/CodeGen/FunctionLoweringInfo.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/IR/CallSite.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/GetElementPtrTypeIterator.h"
#include "llvm/IR/GlobalAlias.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Operator.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Target/TargetOptions.h"
using namespace llvm;

namespace {

class X86FastISel final : public FastISel {
  /// Subtarget - Keep a pointer to the X86Subtarget around so that we can
  /// make the right decision when generating code for different targets.
  const X86Subtarget *Subtarget;

  /// X86ScalarSSEf32, X86ScalarSSEf64 - Select between SSE or x87
  /// floating point ops.
  /// When SSE is available, use it for f32 operations.
  /// When SSE2 is available, use it for f64 operations.
  bool X86ScalarSSEf64;
  bool X86ScalarSSEf32;

public:
  explicit X86FastISel(FunctionLoweringInfo &funcInfo,
                       const TargetLibraryInfo *libInfo)
    : FastISel(funcInfo, libInfo) {
    Subtarget = &TM.getSubtarget<X86Subtarget>();
    X86ScalarSSEf64 = Subtarget->hasSSE2();
    X86ScalarSSEf32 = Subtarget->hasSSE1();
  }

  bool TargetSelectInstruction(const Instruction *I) override;

  /// \brief The specified machine instr operand is a vreg, and that
  /// vreg is being provided by the specified load instruction.  If possible,
  /// try to fold the load as an operand to the instruction, returning true if
  /// possible.
  bool tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
                           const LoadInst *LI) override;

  bool FastLowerArguments() override;

#include "X86GenFastISel.inc"

private:
  bool X86FastEmitCompare(const Value *LHS, const Value *RHS, EVT VT);

  bool X86FastEmitLoad(EVT VT, const X86AddressMode &AM, unsigned &RR);

  bool X86FastEmitStore(EVT VT, const Value *Val, const X86AddressMode &AM,
                        bool Aligned = false);
  bool X86FastEmitStore(EVT VT, unsigned ValReg, const X86AddressMode &AM,
                        bool Aligned = false);

  bool X86FastEmitExtend(ISD::NodeType Opc, EVT DstVT, unsigned Src, EVT SrcVT,
                         unsigned &ResultReg);

  bool X86SelectAddress(const Value *V, X86AddressMode &AM);
  bool X86SelectCallAddress(const Value *V, X86AddressMode &AM);

  bool X86SelectLoad(const Instruction *I);

  bool X86SelectStore(const Instruction *I);

  bool X86SelectRet(const Instruction *I);

  bool X86SelectCmp(const Instruction *I);

  bool X86SelectZExt(const Instruction *I);

  bool X86SelectBranch(const Instruction *I);

  bool X86SelectShift(const Instruction *I);

  bool X86SelectDivRem(const Instruction *I);

  bool X86SelectSelect(const Instruction *I);

  bool X86SelectTrunc(const Instruction *I);

  bool X86SelectFPExt(const Instruction *I);
  bool X86SelectFPTrunc(const Instruction *I);

  bool X86VisitIntrinsicCall(const IntrinsicInst &I);
  bool X86SelectCall(const Instruction *I);

  bool DoSelectCall(const Instruction *I, const char *MemIntName);

  const X86InstrInfo *getInstrInfo() const {
    return getTargetMachine()->getInstrInfo();
  }
  const X86TargetMachine *getTargetMachine() const {
    return static_cast<const X86TargetMachine *>(&TM);
  }

  bool handleConstantAddresses(const Value *V, X86AddressMode &AM);

  unsigned TargetMaterializeConstant(const Constant *C) override;

  unsigned TargetMaterializeAlloca(const AllocaInst *C) override;

  unsigned TargetMaterializeFloatZero(const ConstantFP *CF) override;

  /// isScalarFPTypeInSSEReg - Return true if the specified scalar FP type is
  /// computed in an SSE register, not on the X87 floating point stack.
  bool isScalarFPTypeInSSEReg(EVT VT) const {
    return (VT == MVT::f64 && X86ScalarSSEf64) || // f64 is when SSE2
      (VT == MVT::f32 && X86ScalarSSEf32);   // f32 is when SSE1
  }

  bool isTypeLegal(Type *Ty, MVT &VT, bool AllowI1 = false);

  bool IsMemcpySmall(uint64_t Len);

  bool TryEmitSmallMemcpy(X86AddressMode DestAM,
                          X86AddressMode SrcAM, uint64_t Len);
};

} // end anonymous namespace.

bool X86FastISel::isTypeLegal(Type *Ty, MVT &VT, bool AllowI1) {
  EVT evt = TLI.getValueType(Ty, /*HandleUnknown=*/true);
  if (evt == MVT::Other || !evt.isSimple())
    // Unhandled type. Halt "fast" selection and bail.
    return false;

  VT = evt.getSimpleVT();
  // For now, require SSE/SSE2 for performing floating-point operations,
  // since x87 requires additional work.
  if (VT == MVT::f64 && !X86ScalarSSEf64)
    return false;
  if (VT == MVT::f32 && !X86ScalarSSEf32)
    return false;
  // Similarly, no f80 support yet.
  if (VT == MVT::f80)
    return false;
  // We only handle legal types. For example, on x86-32 the instruction
  // selector contains all of the 64-bit instructions from x86-64,
  // under the assumption that i64 won't be used if the target doesn't
  // support it.
  return (AllowI1 && VT == MVT::i1) || TLI.isTypeLegal(VT);
}

#include "X86GenCallingConv.inc"

/// X86FastEmitLoad - Emit a machine instruction to load a value of type VT.
/// The address is either pre-computed, i.e. Ptr, or a GlobalAddress, i.e. GV.
/// Return true and the result register by reference if it is possible.
bool X86FastISel::X86FastEmitLoad(EVT VT, const X86AddressMode &AM,
                                  unsigned &ResultReg) {
  // Get opcode and regclass of the output for the given load instruction.
  unsigned Opc = 0;
  const TargetRegisterClass *RC = nullptr;
  switch (VT.getSimpleVT().SimpleTy) {
  default: return false;
  case MVT::i1:
  case MVT::i8:
    Opc = X86::MOV8rm;
    RC  = &X86::GR8RegClass;
    break;
  case MVT::i16:
    Opc = X86::MOV16rm;
    RC  = &X86::GR16RegClass;
    break;
  case MVT::i32:
    Opc = X86::MOV32rm;
    RC  = &X86::GR32RegClass;
    break;
  case MVT::i64:
    // Must be in x86-64 mode.
    Opc = X86::MOV64rm;
    RC  = &X86::GR64RegClass;
    break;
  case MVT::f32:
    if (X86ScalarSSEf32) {
      Opc = Subtarget->hasAVX() ? X86::VMOVSSrm : X86::MOVSSrm;
      RC  = &X86::FR32RegClass;
    } else {
      Opc = X86::LD_Fp32m;
      RC  = &X86::RFP32RegClass;
    }
    break;
  case MVT::f64:
    if (X86ScalarSSEf64) {
      Opc = Subtarget->hasAVX() ? X86::VMOVSDrm : X86::MOVSDrm;
      RC  = &X86::FR64RegClass;
    } else {
      Opc = X86::LD_Fp64m;
      RC  = &X86::RFP64RegClass;
    }
    break;
  case MVT::f80:
    // No f80 support yet.
    return false;
  }

  ResultReg = createResultReg(RC);
  addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
                         DbgLoc, TII.get(Opc), ResultReg), AM);
  return true;
}

/// X86FastEmitStore - Emit a machine instruction to store a value Val of
/// type VT. The address is either pre-computed, consisted of a base ptr, Ptr
/// and a displacement offset, or a GlobalAddress,
/// i.e. V. Return true if it is possible.
bool
X86FastISel::X86FastEmitStore(EVT VT, unsigned ValReg,
                              const X86AddressMode &AM, bool Aligned) {
  // Get opcode and regclass of the output for the given store instruction.
  unsigned Opc = 0;
  switch (VT.getSimpleVT().SimpleTy) {
  case MVT::f80: // No f80 support yet.
  default: return false;
  case MVT::i1: {
    // Mask out all but lowest bit.
    unsigned AndResult = createResultReg(&X86::GR8RegClass);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
            TII.get(X86::AND8ri), AndResult).addReg(ValReg).addImm(1);
    ValReg = AndResult;
  }
  // FALLTHROUGH, handling i1 as i8.
  case MVT::i8:  Opc = X86::MOV8mr;  break;
  case MVT::i16: Opc = X86::MOV16mr; break;
  case MVT::i32: Opc = X86::MOV32mr; break;
  case MVT::i64: Opc = X86::MOV64mr; break; // Must be in x86-64 mode.
  case MVT::f32:
    Opc = X86ScalarSSEf32 ?
          (Subtarget->hasAVX() ? X86::VMOVSSmr : X86::MOVSSmr) : X86::ST_Fp32m;
    break;
  case MVT::f64:
    Opc = X86ScalarSSEf64 ?
          (Subtarget->hasAVX() ? X86::VMOVSDmr : X86::MOVSDmr) : X86::ST_Fp64m;
    break;
  case MVT::v4f32:
    if (Aligned)
      Opc = Subtarget->hasAVX() ? X86::VMOVAPSmr : X86::MOVAPSmr;
    else
      Opc = Subtarget->hasAVX() ? X86::VMOVUPSmr : X86::MOVUPSmr;
    break;
  case MVT::v2f64:
    if (Aligned)
      Opc = Subtarget->hasAVX() ? X86::VMOVAPDmr : X86::MOVAPDmr;
    else
      Opc = Subtarget->hasAVX() ? X86::VMOVUPDmr : X86::MOVUPDmr;
    break;
  case MVT::v4i32:
  case MVT::v2i64:
  case MVT::v8i16:
  case MVT::v16i8:
    if (Aligned)
      Opc = Subtarget->hasAVX() ? X86::VMOVDQAmr : X86::MOVDQAmr;
    else
      Opc = Subtarget->hasAVX() ? X86::VMOVDQUmr : X86::MOVDQUmr;
    break;
  }

  addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
                         DbgLoc, TII.get(Opc)), AM).addReg(ValReg);
  return true;
}

bool X86FastISel::X86FastEmitStore(EVT VT, const Value *Val,
                                   const X86AddressMode &AM, bool Aligned) {
  // Handle 'null' like i32/i64 0.
  if (isa<ConstantPointerNull>(Val))
    Val = Constant::getNullValue(DL.getIntPtrType(Val->getContext()));

  // If this is a store of a simple constant, fold the constant into the store.
  if (const ConstantInt *CI = dyn_cast<ConstantInt>(Val)) {
    unsigned Opc = 0;
    bool Signed = true;
    switch (VT.getSimpleVT().SimpleTy) {
    default: break;
    case MVT::i1:  Signed = false;     // FALLTHROUGH to handle as i8.
    case MVT::i8:  Opc = X86::MOV8mi;  break;
    case MVT::i16: Opc = X86::MOV16mi; break;
    case MVT::i32: Opc = X86::MOV32mi; break;
    case MVT::i64:
      // Must be a 32-bit sign extended value.
      if (isInt<32>(CI->getSExtValue()))
        Opc = X86::MOV64mi32;
      break;
    }

    if (Opc) {
      addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
                             DbgLoc, TII.get(Opc)), AM)
                             .addImm(Signed ? (uint64_t) CI->getSExtValue() :
                                              CI->getZExtValue());
      return true;
    }
  }

  unsigned ValReg = getRegForValue(Val);
  if (ValReg == 0)
    return false;

  return X86FastEmitStore(VT, ValReg, AM, Aligned);
}

/// X86FastEmitExtend - Emit a machine instruction to extend a value Src of
/// type SrcVT to type DstVT using the specified extension opcode Opc (e.g.
/// ISD::SIGN_EXTEND).
bool X86FastISel::X86FastEmitExtend(ISD::NodeType Opc, EVT DstVT,
                                    unsigned Src, EVT SrcVT,
                                    unsigned &ResultReg) {
  unsigned RR = FastEmit_r(SrcVT.getSimpleVT(), DstVT.getSimpleVT(), Opc,
                           Src, /*TODO: Kill=*/false);
  if (RR == 0)
    return false;

  ResultReg = RR;
  return true;
}

bool X86FastISel::handleConstantAddresses(const Value *V, X86AddressMode &AM) {
  // Handle constant address.
  if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
    // Can't handle alternate code models yet.
    if (TM.getCodeModel() != CodeModel::Small)
      return false;

    // Can't handle TLS yet.
    if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV))
      if (GVar->isThreadLocal())
        return false;

    // Can't handle TLS yet, part 2 (this is slightly crazy, but this is how
    // it works...).
    if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
      if (const GlobalVariable *GVar =
              dyn_cast_or_null<GlobalVariable>(GA->getAliasee()))
        if (GVar->isThreadLocal())
          return false;

    // RIP-relative addresses can't have additional register operands, so if
    // we've already folded stuff into the addressing mode, just force the
    // global value into its own register, which we can use as the basereg.
    if (!Subtarget->isPICStyleRIPRel() ||
        (AM.Base.Reg == 0 && AM.IndexReg == 0)) {
      // Okay, we've committed to selecting this global. Set up the address.
      AM.GV = GV;

      // Allow the subtarget to classify the global.
      unsigned char GVFlags = Subtarget->ClassifyGlobalReference(GV, TM);

      // If this reference is relative to the pic base, set it now.
      if (isGlobalRelativeToPICBase(GVFlags)) {
        // FIXME: How do we know Base.Reg is free??
        AM.Base.Reg = getInstrInfo()->getGlobalBaseReg(FuncInfo.MF);
      }

      // Unless the ABI requires an extra load, return a direct reference to
      // the global.
      if (!isGlobalStubReference(GVFlags)) {
        if (Subtarget->isPICStyleRIPRel()) {
          // Use rip-relative addressing if we can.  Above we verified that the
          // base and index registers are unused.
          assert(AM.Base.Reg == 0 && AM.IndexReg == 0);
          AM.Base.Reg = X86::RIP;
        }
        AM.GVOpFlags = GVFlags;
        return true;
      }

      // Ok, we need to do a load from a stub.  If we've already loaded from
      // this stub, reuse the loaded pointer, otherwise emit the load now.
      DenseMap<const Value*, unsigned>::iterator I = LocalValueMap.find(V);
      unsigned LoadReg;
      if (I != LocalValueMap.end() && I->second != 0) {
        LoadReg = I->second;
      } else {
        // Issue load from stub.
        unsigned Opc = 0;
        const TargetRegisterClass *RC = nullptr;
        X86AddressMode StubAM;
        StubAM.Base.Reg = AM.Base.Reg;
        StubAM.GV = GV;
        StubAM.GVOpFlags = GVFlags;

        // Prepare for inserting code in the local-value area.
        SavePoint SaveInsertPt = enterLocalValueArea();

        if (TLI.getPointerTy() == MVT::i64) {
          Opc = X86::MOV64rm;
          RC  = &X86::GR64RegClass;

          if (Subtarget->isPICStyleRIPRel())
            StubAM.Base.Reg = X86::RIP;
        } else {
          Opc = X86::MOV32rm;
          RC  = &X86::GR32RegClass;
        }

        LoadReg = createResultReg(RC);
        MachineInstrBuilder LoadMI =
          BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), LoadReg);
        addFullAddress(LoadMI, StubAM);

        // Ok, back to normal mode.
        leaveLocalValueArea(SaveInsertPt);

        // Prevent loading GV stub multiple times in same MBB.
        LocalValueMap[V] = LoadReg;
      }

      // Now construct the final address. Note that the Disp, Scale,
      // and Index values may already be set here.
      AM.Base.Reg = LoadReg;
      AM.GV = nullptr;
      return true;
    }
  }

  // If all else fails, try to materialize the value in a register.
  if (!AM.GV || !Subtarget->isPICStyleRIPRel()) {
    if (AM.Base.Reg == 0) {
      AM.Base.Reg = getRegForValue(V);
      return AM.Base.Reg != 0;
    }
    if (AM.IndexReg == 0) {
      assert(AM.Scale == 1 && "Scale with no index!");
      AM.IndexReg = getRegForValue(V);
      return AM.IndexReg != 0;
    }
  }

  return false;
}

/// X86SelectAddress - Attempt to fill in an address from the given value.
///
bool X86FastISel::X86SelectAddress(const Value *V, X86AddressMode &AM) {
  SmallVector<const Value *, 32> GEPs;
redo_gep:
  const User *U = nullptr;
  unsigned Opcode = Instruction::UserOp1;
  if (const Instruction *I = dyn_cast<Instruction>(V)) {
    // Don't walk into other basic blocks; it's possible we haven't
    // visited them yet, so the instructions may not yet be assigned
    // virtual registers.
    if (FuncInfo.StaticAllocaMap.count(static_cast<const AllocaInst *>(V)) ||
        FuncInfo.MBBMap[I->getParent()] == FuncInfo.MBB) {
      Opcode = I->getOpcode();
      U = I;
    }
  } else if (const ConstantExpr *C = dyn_cast<ConstantExpr>(V)) {
    Opcode = C->getOpcode();
    U = C;
  }

  if (PointerType *Ty = dyn_cast<PointerType>(V->getType()))
    if (Ty->getAddressSpace() > 255)
      // Fast instruction selection doesn't support the special
      // address spaces.
      return false;

  switch (Opcode) {
  default: break;
  case Instruction::BitCast:
    // Look past bitcasts.
    return X86SelectAddress(U->getOperand(0), AM);

  case Instruction::IntToPtr:
    // Look past no-op inttoptrs.
    if (TLI.getValueType(U->getOperand(0)->getType()) == TLI.getPointerTy())
      return X86SelectAddress(U->getOperand(0), AM);
    break;

  case Instruction::PtrToInt:
    // Look past no-op ptrtoints.
    if (TLI.getValueType(U->getType()) == TLI.getPointerTy())
      return X86SelectAddress(U->getOperand(0), AM);
    break;

  case Instruction::Alloca: {
    // Do static allocas.
    const AllocaInst *A = cast<AllocaInst>(V);
    DenseMap<const AllocaInst*, int>::iterator SI =
      FuncInfo.StaticAllocaMap.find(A);
    if (SI != FuncInfo.StaticAllocaMap.end()) {
      AM.BaseType = X86AddressMode::FrameIndexBase;
      AM.Base.FrameIndex = SI->second;
      return true;
    }
    break;
  }

  case Instruction::Add: {
    // Adds of constants are common and easy enough.
    if (const ConstantInt *CI = dyn_cast<ConstantInt>(U->getOperand(1))) {
      uint64_t Disp = (int32_t)AM.Disp + (uint64_t)CI->getSExtValue();
      // They have to fit in the 32-bit signed displacement field though.
      if (isInt<32>(Disp)) {
        AM.Disp = (uint32_t)Disp;
        return X86SelectAddress(U->getOperand(0), AM);
      }
    }
    break;
  }

  case Instruction::GetElementPtr: {
    X86AddressMode SavedAM = AM;

    // Pattern-match simple GEPs.
    uint64_t Disp = (int32_t)AM.Disp;
    unsigned IndexReg = AM.IndexReg;
    unsigned Scale = AM.Scale;
    gep_type_iterator GTI = gep_type_begin(U);
    // Iterate through the indices, folding what we can. Constants can be
    // folded, and one dynamic index can be handled, if the scale is supported.
    for (User::const_op_iterator i = U->op_begin() + 1, e = U->op_end();
         i != e; ++i, ++GTI) {
      const Value *Op = *i;
      if (StructType *STy = dyn_cast<StructType>(*GTI)) {
        const StructLayout *SL = DL.getStructLayout(STy);
        Disp += SL->getElementOffset(cast<ConstantInt>(Op)->getZExtValue());
        continue;
      }

      // A array/variable index is always of the form i*S where S is the
      // constant scale size.  See if we can push the scale into immediates.
      uint64_t S = DL.getTypeAllocSize(GTI.getIndexedType());
      for (;;) {
        if (const ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
          // Constant-offset addressing.
          Disp += CI->getSExtValue() * S;
          break;
        }
        if (canFoldAddIntoGEP(U, Op)) {
          // A compatible add with a constant operand. Fold the constant.
          ConstantInt *CI =
            cast<ConstantInt>(cast<AddOperator>(Op)->getOperand(1));
          Disp += CI->getSExtValue() * S;
          // Iterate on the other operand.
          Op = cast<AddOperator>(Op)->getOperand(0);
          continue;
        }
        if (IndexReg == 0 &&
            (!AM.GV || !Subtarget->isPICStyleRIPRel()) &&
            (S == 1 || S == 2 || S == 4 || S == 8)) {
          // Scaled-index addressing.
          Scale = S;
          IndexReg = getRegForGEPIndex(Op).first;
          if (IndexReg == 0)
            return false;
          break;
        }
        // Unsupported.
        goto unsupported_gep;
      }
    }

    // Check for displacement overflow.
    if (!isInt<32>(Disp))
      break;

    AM.IndexReg = IndexReg;
    AM.Scale = Scale;
    AM.Disp = (uint32_t)Disp;
    GEPs.push_back(V);

    if (const GetElementPtrInst *GEP =
          dyn_cast<GetElementPtrInst>(U->getOperand(0))) {
      // Ok, the GEP indices were covered by constant-offset and scaled-index
      // addressing. Update the address state and move on to examining the base.
      V = GEP;
      goto redo_gep;
    } else if (X86SelectAddress(U->getOperand(0), AM)) {
      return true;
    }

    // If we couldn't merge the gep value into this addr mode, revert back to
    // our address and just match the value instead of completely failing.
    AM = SavedAM;

    for (SmallVectorImpl<const Value *>::reverse_iterator
           I = GEPs.rbegin(), E = GEPs.rend(); I != E; ++I)
      if (handleConstantAddresses(*I, AM))
        return true;

    return false;
  unsupported_gep:
    // Ok, the GEP indices weren't all covered.
    break;
  }
  }

  return handleConstantAddresses(V, AM);
}

/// X86SelectCallAddress - Attempt to fill in an address from the given value.
///
bool X86FastISel::X86SelectCallAddress(const Value *V, X86AddressMode &AM) {
  const User *U = nullptr;
  unsigned Opcode = Instruction::UserOp1;
  const Instruction *I = dyn_cast<Instruction>(V);
  // Record if the value is defined in the same basic block.
  //
  // This information is crucial to know whether or not folding an
  // operand is valid.
  // Indeed, FastISel generates or reuses a virtual register for all
  // operands of all instructions it selects. Obviously, the definition and
  // its uses must use the same virtual register otherwise the produced
  // code is incorrect.
  // Before instruction selection, FunctionLoweringInfo::set sets the virtual
  // registers for values that are alive across basic blocks. This ensures
  // that the values are consistently set between across basic block, even
  // if different instruction selection mechanisms are used (e.g., a mix of
  // SDISel and FastISel).
  // For values local to a basic block, the instruction selection process
  // generates these virtual registers with whatever method is appropriate
  // for its needs. In particular, FastISel and SDISel do not share the way
  // local virtual registers are set.
  // Therefore, this is impossible (or at least unsafe) to share values
  // between basic blocks unless they use the same instruction selection
  // method, which is not guarantee for X86.
  // Moreover, things like hasOneUse could not be used accurately, if we
  // allow to reference values across basic blocks whereas they are not
  // alive across basic blocks initially.
  bool InMBB = true;
  if (I) {
    Opcode = I->getOpcode();
    U = I;
    InMBB = I->getParent() == FuncInfo.MBB->getBasicBlock();
  } else if (const ConstantExpr *C = dyn_cast<ConstantExpr>(V)) {
    Opcode = C->getOpcode();
    U = C;
  }

  switch (Opcode) {
  default: break;
  case Instruction::BitCast:
    // Look past bitcasts if its operand is in the same BB.
    if (InMBB)
      return X86SelectCallAddress(U->getOperand(0), AM);
    break;

  case Instruction::IntToPtr:
    // Look past no-op inttoptrs if its operand is in the same BB.
    if (InMBB &&
        TLI.getValueType(U->getOperand(0)->getType()) == TLI.getPointerTy())
      return X86SelectCallAddress(U->getOperand(0), AM);
    break;

  case Instruction::PtrToInt:
    // Look past no-op ptrtoints if its operand is in the same BB.
    if (InMBB &&
        TLI.getValueType(U->getType()) == TLI.getPointerTy())
      return X86SelectCallAddress(U->getOperand(0), AM);
    break;
  }

  // Handle constant address.
  if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
    // Can't handle alternate code models yet.
    if (TM.getCodeModel() != CodeModel::Small)
      return false;

    // RIP-relative addresses can't have additional register operands.
    if (Subtarget->isPICStyleRIPRel() &&
        (AM.Base.Reg != 0 || AM.IndexReg != 0))
      return false;

    // Can't handle DbgLocLImport.
    if (GV->hasDLLImportStorageClass())
      return false;

    // Can't handle TLS.
    if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV))
      if (GVar->isThreadLocal())
        return false;

    // Okay, we've committed to selecting this global. Set up the basic address.
    AM.GV = GV;

    // No ABI requires an extra load for anything other than DLLImport, which
    // we rejected above. Return a direct reference to the global.
    if (Subtarget->isPICStyleRIPRel()) {
      // Use rip-relative addressing if we can.  Above we verified that the
      // base and index registers are unused.
      assert(AM.Base.Reg == 0 && AM.IndexReg == 0);
      AM.Base.Reg = X86::RIP;
    } else if (Subtarget->isPICStyleStubPIC()) {
      AM.GVOpFlags = X86II::MO_PIC_BASE_OFFSET;
    } else if (Subtarget->isPICStyleGOT()) {
      AM.GVOpFlags = X86II::MO_GOTOFF;
    }

    return true;
  }

  // If all else fails, try to materialize the value in a register.
  if (!AM.GV || !Subtarget->isPICStyleRIPRel()) {
    if (AM.Base.Reg == 0) {
      AM.Base.Reg = getRegForValue(V);
      return AM.Base.Reg != 0;
    }
    if (AM.IndexReg == 0) {
      assert(AM.Scale == 1 && "Scale with no index!");
      AM.IndexReg = getRegForValue(V);
      return AM.IndexReg != 0;
    }
  }

  return false;
}


/// X86SelectStore - Select and emit code to implement store instructions.
bool X86FastISel::X86SelectStore(const Instruction *I) {
  // Atomic stores need special handling.
  const StoreInst *S = cast<StoreInst>(I);

  if (S->isAtomic())
    return false;

  unsigned SABIAlignment =
    DL.getABITypeAlignment(S->getValueOperand()->getType());
  bool Aligned = S->getAlignment() == 0 || S->getAlignment() >= SABIAlignment;

  MVT VT;
  if (!isTypeLegal(I->getOperand(0)->getType(), VT, /*AllowI1=*/true))
    return false;

  X86AddressMode AM;
  if (!X86SelectAddress(I->getOperand(1), AM))
    return false;

  return X86FastEmitStore(VT, I->getOperand(0), AM, Aligned);
}

/// X86SelectRet - Select and emit code to implement ret instructions.
bool X86FastISel::X86SelectRet(const Instruction *I) {
  const ReturnInst *Ret = cast<ReturnInst>(I);
  const Function &F = *I->getParent()->getParent();
  const X86MachineFunctionInfo *X86MFInfo =
      FuncInfo.MF->getInfo<X86MachineFunctionInfo>();

  if (!FuncInfo.CanLowerReturn)
    return false;

  CallingConv::ID CC = F.getCallingConv();
  if (CC != CallingConv::C &&
      CC != CallingConv::Fast &&
      CC != CallingConv::X86_FastCall &&
      CC != CallingConv::X86_64_SysV)
    return false;

  if (Subtarget->isCallingConvWin64(CC))
    return false;

  // Don't handle popping bytes on return for now.
  if (X86MFInfo->getBytesToPopOnReturn() != 0)
    return false;

  // fastcc with -tailcallopt is intended to provide a guaranteed
  // tail call optimization. Fastisel doesn't know how to do that.
  if (CC == CallingConv::Fast && TM.Options.GuaranteedTailCallOpt)
    return false;

  // Let SDISel handle vararg functions.
  if (F.isVarArg())
    return false;

  // Build a list of return value registers.
  SmallVector<unsigned, 4> RetRegs;

  if (Ret->getNumOperands() > 0) {
    SmallVector<ISD::OutputArg, 4> Outs;
    GetReturnInfo(F.getReturnType(), F.getAttributes(), Outs, TLI);

    // Analyze operands of the call, assigning locations to each operand.
    SmallVector<CCValAssign, 16> ValLocs;
    CCState CCInfo(CC, F.isVarArg(), *FuncInfo.MF, TM, ValLocs,
                   I->getContext());
    CCInfo.AnalyzeReturn(Outs, RetCC_X86);

    const Value *RV = Ret->getOperand(0);
    unsigned Reg = getRegForValue(RV);
    if (Reg == 0)
      return false;

    // Only handle a single return value for now.
    if (ValLocs.size() != 1)
      return false;

    CCValAssign &VA = ValLocs[0];

    // Don't bother handling odd stuff for now.
    if (VA.getLocInfo() != CCValAssign::Full)
      return false;
    // Only handle register returns for now.
    if (!VA.isRegLoc())
      return false;

    // The calling-convention tables for x87 returns don't tell
    // the whole story.
    if (VA.getLocReg() == X86::ST0 || VA.getLocReg() == X86::ST1)
      return false;

    unsigned SrcReg = Reg + VA.getValNo();
    EVT SrcVT = TLI.getValueType(RV->getType());
    EVT DstVT = VA.getValVT();
    // Special handling for extended integers.
    if (SrcVT != DstVT) {
      if (SrcVT != MVT::i1 && SrcVT != MVT::i8 && SrcVT != MVT::i16)
        return false;

      if (!Outs[0].Flags.isZExt() && !Outs[0].Flags.isSExt())
        return false;

      assert(DstVT == MVT::i32 && "X86 should always ext to i32");

      if (SrcVT == MVT::i1) {
        if (Outs[0].Flags.isSExt())
          return false;
        SrcReg = FastEmitZExtFromI1(MVT::i8, SrcReg, /*TODO: Kill=*/false);
        SrcVT = MVT::i8;
      }
      unsigned Op = Outs[0].Flags.isZExt() ? ISD::ZERO_EXTEND :
                                             ISD::SIGN_EXTEND;
      SrcReg = FastEmit_r(SrcVT.getSimpleVT(), DstVT.getSimpleVT(), Op,
                          SrcReg, /*TODO: Kill=*/false);
    }

    // Make the copy.
    unsigned DstReg = VA.getLocReg();
    const TargetRegisterClass* SrcRC = MRI.getRegClass(SrcReg);
    // Avoid a cross-class copy. This is very unlikely.
    if (!SrcRC->contains(DstReg))
      return false;
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
            DstReg).addReg(SrcReg);

    // Add register to return instruction.
    RetRegs.push_back(VA.getLocReg());
  }

  // The x86-64 ABI for returning structs by value requires that we copy
  // the sret argument into %rax for the return. We saved the argument into
  // a virtual register in the entry block, so now we copy the value out
  // and into %rax. We also do the same with %eax for Win32.
  if (F.hasStructRetAttr() &&
      (Subtarget->is64Bit() || Subtarget->isTargetKnownWindowsMSVC())) {
    unsigned Reg = X86MFInfo->getSRetReturnReg();
    assert(Reg &&
           "SRetReturnReg should have been set in LowerFormalArguments()!");
    unsigned RetReg = Subtarget->is64Bit() ? X86::RAX : X86::EAX;
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
            RetReg).addReg(Reg);
    RetRegs.push_back(RetReg);
  }

  // Now emit the RET.
  MachineInstrBuilder MIB =
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Subtarget->is64Bit() ? X86::RETQ : X86::RETL));
  for (unsigned i = 0, e = RetRegs.size(); i != e; ++i)
    MIB.addReg(RetRegs[i], RegState::Implicit);
  return true;
}

/// X86SelectLoad - Select and emit code to implement load instructions.
///
bool X86FastISel::X86SelectLoad(const Instruction *I)  {
  // Atomic loads need special handling.
  if (cast<LoadInst>(I)->isAtomic())
    return false;

  MVT VT;
  if (!isTypeLegal(I->getType(), VT, /*AllowI1=*/true))
    return false;

  X86AddressMode AM;
  if (!X86SelectAddress(I->getOperand(0), AM))
    return false;

  unsigned ResultReg = 0;
  if (X86FastEmitLoad(VT, AM, ResultReg)) {
    UpdateValueMap(I, ResultReg);
    return true;
  }
  return false;
}

static unsigned X86ChooseCmpOpcode(EVT VT, const X86Subtarget *Subtarget) {
  bool HasAVX = Subtarget->hasAVX();
  bool X86ScalarSSEf32 = Subtarget->hasSSE1();
  bool X86ScalarSSEf64 = Subtarget->hasSSE2();

  switch (VT.getSimpleVT().SimpleTy) {
  default:       return 0;
  case MVT::i8:  return X86::CMP8rr;
  case MVT::i16: return X86::CMP16rr;
  case MVT::i32: return X86::CMP32rr;
  case MVT::i64: return X86::CMP64rr;
  case MVT::f32:
    return X86ScalarSSEf32 ? (HasAVX ? X86::VUCOMISSrr : X86::UCOMISSrr) : 0;
  case MVT::f64:
    return X86ScalarSSEf64 ? (HasAVX ? X86::VUCOMISDrr : X86::UCOMISDrr) : 0;
  }
}

/// X86ChooseCmpImmediateOpcode - If we have a comparison with RHS as the RHS
/// of the comparison, return an opcode that works for the compare (e.g.
/// CMP32ri) otherwise return 0.
static unsigned X86ChooseCmpImmediateOpcode(EVT VT, const ConstantInt *RHSC) {
  switch (VT.getSimpleVT().SimpleTy) {
  // Otherwise, we can't fold the immediate into this comparison.
  default: return 0;
  case MVT::i8: return X86::CMP8ri;
  case MVT::i16: return X86::CMP16ri;
  case MVT::i32: return X86::CMP32ri;
  case MVT::i64:
    // 64-bit comparisons are only valid if the immediate fits in a 32-bit sext
    // field.
    if ((int)RHSC->getSExtValue() == RHSC->getSExtValue())
      return X86::CMP64ri32;
    return 0;
  }
}

bool X86FastISel::X86FastEmitCompare(const Value *Op0, const Value *Op1,
                                     EVT VT) {
  unsigned Op0Reg = getRegForValue(Op0);
  if (Op0Reg == 0) return false;

  // Handle 'null' like i32/i64 0.
  if (isa<ConstantPointerNull>(Op1))
    Op1 = Constant::getNullValue(DL.getIntPtrType(Op0->getContext()));

  // We have two options: compare with register or immediate.  If the RHS of
  // the compare is an immediate that we can fold into this compare, use
  // CMPri, otherwise use CMPrr.
  if (const ConstantInt *Op1C = dyn_cast<ConstantInt>(Op1)) {
    if (unsigned CompareImmOpc = X86ChooseCmpImmediateOpcode(VT, Op1C)) {
      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CompareImmOpc))
        .addReg(Op0Reg)
        .addImm(Op1C->getSExtValue());
      return true;
    }
  }

  unsigned CompareOpc = X86ChooseCmpOpcode(VT, Subtarget);
  if (CompareOpc == 0) return false;

  unsigned Op1Reg = getRegForValue(Op1);
  if (Op1Reg == 0) return false;
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CompareOpc))
    .addReg(Op0Reg)
    .addReg(Op1Reg);

  return true;
}

bool X86FastISel::X86SelectCmp(const Instruction *I) {
  const CmpInst *CI = cast<CmpInst>(I);

  MVT VT;
  if (!isTypeLegal(I->getOperand(0)->getType(), VT))
    return false;

  unsigned ResultReg = createResultReg(&X86::GR8RegClass);
  unsigned SetCCOpc;
  bool SwapArgs;  // false -> compare Op0, Op1.  true -> compare Op1, Op0.
  switch (CI->getPredicate()) {
  case CmpInst::FCMP_OEQ: {
    if (!X86FastEmitCompare(CI->getOperand(0), CI->getOperand(1), VT))
      return false;

    unsigned EReg = createResultReg(&X86::GR8RegClass);
    unsigned NPReg = createResultReg(&X86::GR8RegClass);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETEr), EReg);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
            TII.get(X86::SETNPr), NPReg);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
            TII.get(X86::AND8rr), ResultReg).addReg(NPReg).addReg(EReg);
    UpdateValueMap(I, ResultReg);
    return true;
  }
  case CmpInst::FCMP_UNE: {
    if (!X86FastEmitCompare(CI->getOperand(0), CI->getOperand(1), VT))
      return false;

    unsigned NEReg = createResultReg(&X86::GR8RegClass);
    unsigned PReg = createResultReg(&X86::GR8RegClass);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETNEr), NEReg);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETPr), PReg);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::OR8rr),ResultReg)
      .addReg(PReg).addReg(NEReg);
    UpdateValueMap(I, ResultReg);
    return true;
  }
  case CmpInst::FCMP_OGT: SwapArgs = false; SetCCOpc = X86::SETAr;  break;
  case CmpInst::FCMP_OGE: SwapArgs = false; SetCCOpc = X86::SETAEr; break;
  case CmpInst::FCMP_OLT: SwapArgs = true;  SetCCOpc = X86::SETAr;  break;
  case CmpInst::FCMP_OLE: SwapArgs = true;  SetCCOpc = X86::SETAEr; break;
  case CmpInst::FCMP_ONE: SwapArgs = false; SetCCOpc = X86::SETNEr; break;
  case CmpInst::FCMP_ORD: SwapArgs = false; SetCCOpc = X86::SETNPr; break;
  case CmpInst::FCMP_UNO: SwapArgs = false; SetCCOpc = X86::SETPr;  break;
  case CmpInst::FCMP_UEQ: SwapArgs = false; SetCCOpc = X86::SETEr;  break;
  case CmpInst::FCMP_UGT: SwapArgs = true;  SetCCOpc = X86::SETBr;  break;
  case CmpInst::FCMP_UGE: SwapArgs = true;  SetCCOpc = X86::SETBEr; break;
  case CmpInst::FCMP_ULT: SwapArgs = false; SetCCOpc = X86::SETBr;  break;
  case CmpInst::FCMP_ULE: SwapArgs = false; SetCCOpc = X86::SETBEr; break;

  case CmpInst::ICMP_EQ:  SwapArgs = false; SetCCOpc = X86::SETEr;  break;
  case CmpInst::ICMP_NE:  SwapArgs = false; SetCCOpc = X86::SETNEr; break;
  case CmpInst::ICMP_UGT: SwapArgs = false; SetCCOpc = X86::SETAr;  break;
  case CmpInst::ICMP_UGE: SwapArgs = false; SetCCOpc = X86::SETAEr; break;
  case CmpInst::ICMP_ULT: SwapArgs = false; SetCCOpc = X86::SETBr;  break;
  case CmpInst::ICMP_ULE: SwapArgs = false; SetCCOpc = X86::SETBEr; break;
  case CmpInst::ICMP_SGT: SwapArgs = false; SetCCOpc = X86::SETGr;  break;
  case CmpInst::ICMP_SGE: SwapArgs = false; SetCCOpc = X86::SETGEr; break;
  case CmpInst::ICMP_SLT: SwapArgs = false; SetCCOpc = X86::SETLr;  break;
  case CmpInst::ICMP_SLE: SwapArgs = false; SetCCOpc = X86::SETLEr; break;
  default:
    return false;
  }

  const Value *Op0 = CI->getOperand(0), *Op1 = CI->getOperand(1);
  if (SwapArgs)
    std::swap(Op0, Op1);

  // Emit a compare of Op0/Op1.
  if (!X86FastEmitCompare(Op0, Op1, VT))
    return false;

  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(SetCCOpc), ResultReg);
  UpdateValueMap(I, ResultReg);
  return true;
}

bool X86FastISel::X86SelectZExt(const Instruction *I) {
  EVT DstVT = TLI.getValueType(I->getType());
  if (!TLI.isTypeLegal(DstVT))
    return false;

  unsigned ResultReg = getRegForValue(I->getOperand(0));
  if (ResultReg == 0)
    return false;

  // Handle zero-extension from i1 to i8, which is common.
  MVT SrcVT = TLI.getSimpleValueType(I->getOperand(0)->getType());
  if (SrcVT.SimpleTy == MVT::i1) {
    // Set the high bits to zero.
    ResultReg = FastEmitZExtFromI1(MVT::i8, ResultReg, /*TODO: Kill=*/false);
    SrcVT = MVT::i8;

    if (ResultReg == 0)
      return false;
  }

  if (DstVT == MVT::i64) {
    // Handle extension to 64-bits via sub-register shenanigans.
    unsigned MovInst;

    switch (SrcVT.SimpleTy) {
    case MVT::i8:  MovInst = X86::MOVZX32rr8;  break;
    case MVT::i16: MovInst = X86::MOVZX32rr16; break;
    case MVT::i32: MovInst = X86::MOV32rr;     break;
    default: llvm_unreachable("Unexpected zext to i64 source type");
    }

    unsigned Result32 = createResultReg(&X86::GR32RegClass);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(MovInst), Result32)
      .addReg(ResultReg);

    ResultReg = createResultReg(&X86::GR64RegClass);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::SUBREG_TO_REG),
            ResultReg)
      .addImm(0).addReg(Result32).addImm(X86::sub_32bit);
  } else if (DstVT != MVT::i8) {
    ResultReg = FastEmit_r(MVT::i8, DstVT.getSimpleVT(), ISD::ZERO_EXTEND,
                           ResultReg, /*Kill=*/true);
    if (ResultReg == 0)
      return false;
  }

  UpdateValueMap(I, ResultReg);
  return true;
}


bool X86FastISel::X86SelectBranch(const Instruction *I) {
  // Unconditional branches are selected by tablegen-generated code.
  // Handle a conditional branch.
  const BranchInst *BI = cast<BranchInst>(I);
  MachineBasicBlock *TrueMBB = FuncInfo.MBBMap[BI->getSuccessor(0)];
  MachineBasicBlock *FalseMBB = FuncInfo.MBBMap[BI->getSuccessor(1)];

  // Fold the common case of a conditional branch with a comparison
  // in the same block (values defined on other blocks may not have
  // initialized registers).
  if (const CmpInst *CI = dyn_cast<CmpInst>(BI->getCondition())) {
    if (CI->hasOneUse() && CI->getParent() == I->getParent()) {
      EVT VT = TLI.getValueType(CI->getOperand(0)->getType());

      // Try to take advantage of fallthrough opportunities.
      CmpInst::Predicate Predicate = CI->getPredicate();
      if (FuncInfo.MBB->isLayoutSuccessor(TrueMBB)) {
        std::swap(TrueMBB, FalseMBB);
        Predicate = CmpInst::getInversePredicate(Predicate);
      }

      bool SwapArgs;  // false -> compare Op0, Op1.  true -> compare Op1, Op0.
      unsigned BranchOpc; // Opcode to jump on, e.g. "X86::JA"

      switch (Predicate) {
      case CmpInst::FCMP_OEQ:
        std::swap(TrueMBB, FalseMBB);
        Predicate = CmpInst::FCMP_UNE;
        // FALL THROUGH
      case CmpInst::FCMP_UNE: SwapArgs = false; BranchOpc = X86::JNE_4; break;
      case CmpInst::FCMP_OGT: SwapArgs = false; BranchOpc = X86::JA_4;  break;
      case CmpInst::FCMP_OGE: SwapArgs = false; BranchOpc = X86::JAE_4; break;
      case CmpInst::FCMP_OLT: SwapArgs = true;  BranchOpc = X86::JA_4;  break;
      case CmpInst::FCMP_OLE: SwapArgs = true;  BranchOpc = X86::JAE_4; break;
      case CmpInst::FCMP_ONE: SwapArgs = false; BranchOpc = X86::JNE_4; break;
      case CmpInst::FCMP_ORD: SwapArgs = false; BranchOpc = X86::JNP_4; break;
      case CmpInst::FCMP_UNO: SwapArgs = false; BranchOpc = X86::JP_4;  break;
      case CmpInst::FCMP_UEQ: SwapArgs = false; BranchOpc = X86::JE_4;  break;
      case CmpInst::FCMP_UGT: SwapArgs = true;  BranchOpc = X86::JB_4;  break;
      case CmpInst::FCMP_UGE: SwapArgs = true;  BranchOpc = X86::JBE_4; break;
      case CmpInst::FCMP_ULT: SwapArgs = false; BranchOpc = X86::JB_4;  break;
      case CmpInst::FCMP_ULE: SwapArgs = false; BranchOpc = X86::JBE_4; break;

      case CmpInst::ICMP_EQ:  SwapArgs = false; BranchOpc = X86::JE_4;  break;
      case CmpInst::ICMP_NE:  SwapArgs = false; BranchOpc = X86::JNE_4; break;
      case CmpInst::ICMP_UGT: SwapArgs = false; BranchOpc = X86::JA_4;  break;
      case CmpInst::ICMP_UGE: SwapArgs = false; BranchOpc = X86::JAE_4; break;
      case CmpInst::ICMP_ULT: SwapArgs = false; BranchOpc = X86::JB_4;  break;
      case CmpInst::ICMP_ULE: SwapArgs = false; BranchOpc = X86::JBE_4; break;
      case CmpInst::ICMP_SGT: SwapArgs = false; BranchOpc = X86::JG_4;  break;
      case CmpInst::ICMP_SGE: SwapArgs = false; BranchOpc = X86::JGE_4; break;
      case CmpInst::ICMP_SLT: SwapArgs = false; BranchOpc = X86::JL_4;  break;
      case CmpInst::ICMP_SLE: SwapArgs = false; BranchOpc = X86::JLE_4; break;
      default:
        return false;
      }

      const Value *Op0 = CI->getOperand(0), *Op1 = CI->getOperand(1);
      if (SwapArgs)
        std::swap(Op0, Op1);

      // Emit a compare of the LHS and RHS, setting the flags.
      if (!X86FastEmitCompare(Op0, Op1, VT))
        return false;

      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(BranchOpc))
        .addMBB(TrueMBB);

      if (Predicate == CmpInst::FCMP_UNE) {
        // X86 requires a second branch to handle UNE (and OEQ,
        // which is mapped to UNE above).
        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::JP_4))
          .addMBB(TrueMBB);
      }

      FastEmitBranch(FalseMBB, DbgLoc);
      FuncInfo.MBB->addSuccessor(TrueMBB);
      return true;
    }
  } else if (TruncInst *TI = dyn_cast<TruncInst>(BI->getCondition())) {
    // Handle things like "%cond = trunc i32 %X to i1 / br i1 %cond", which
    // typically happen for _Bool and C++ bools.
    MVT SourceVT;
    if (TI->hasOneUse() && TI->getParent() == I->getParent() &&
        isTypeLegal(TI->getOperand(0)->getType(), SourceVT)) {
      unsigned TestOpc = 0;
      switch (SourceVT.SimpleTy) {
      default: break;
      case MVT::i8:  TestOpc = X86::TEST8ri; break;
      case MVT::i16: TestOpc = X86::TEST16ri; break;
      case MVT::i32: TestOpc = X86::TEST32ri; break;
      case MVT::i64: TestOpc = X86::TEST64ri32; break;
      }
      if (TestOpc) {
        unsigned OpReg = getRegForValue(TI->getOperand(0));
        if (OpReg == 0) return false;
        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TestOpc))
          .addReg(OpReg).addImm(1);

        unsigned JmpOpc = X86::JNE_4;
        if (FuncInfo.MBB->isLayoutSuccessor(TrueMBB)) {
          std::swap(TrueMBB, FalseMBB);
          JmpOpc = X86::JE_4;
        }

        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(JmpOpc))
          .addMBB(TrueMBB);
        FastEmitBranch(FalseMBB, DbgLoc);
        FuncInfo.MBB->addSuccessor(TrueMBB);
        return true;
      }
    }
  }

  // Otherwise do a clumsy setcc and re-test it.
  // Note that i1 essentially gets ANY_EXTEND'ed to i8 where it isn't used
  // in an explicit cast, so make sure to handle that correctly.
  unsigned OpReg = getRegForValue(BI->getCondition());
  if (OpReg == 0) return false;

  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::TEST8ri))
    .addReg(OpReg).addImm(1);
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::JNE_4))
    .addMBB(TrueMBB);
  FastEmitBranch(FalseMBB, DbgLoc);
  FuncInfo.MBB->addSuccessor(TrueMBB);
  return true;
}

bool X86FastISel::X86SelectShift(const Instruction *I) {
  unsigned CReg = 0, OpReg = 0;
  const TargetRegisterClass *RC = nullptr;
  if (I->getType()->isIntegerTy(8)) {
    CReg = X86::CL;
    RC = &X86::GR8RegClass;
    switch (I->getOpcode()) {
    case Instruction::LShr: OpReg = X86::SHR8rCL; break;
    case Instruction::AShr: OpReg = X86::SAR8rCL; break;
    case Instruction::Shl:  OpReg = X86::SHL8rCL; break;
    default: return false;
    }
  } else if (I->getType()->isIntegerTy(16)) {
    CReg = X86::CX;
    RC = &X86::GR16RegClass;
    switch (I->getOpcode()) {
    case Instruction::LShr: OpReg = X86::SHR16rCL; break;
    case Instruction::AShr: OpReg = X86::SAR16rCL; break;
    case Instruction::Shl:  OpReg = X86::SHL16rCL; break;
    default: return false;
    }
  } else if (I->getType()->isIntegerTy(32)) {
    CReg = X86::ECX;
    RC = &X86::GR32RegClass;
    switch (I->getOpcode()) {
    case Instruction::LShr: OpReg = X86::SHR32rCL; break;
    case Instruction::AShr: OpReg = X86::SAR32rCL; break;
    case Instruction::Shl:  OpReg = X86::SHL32rCL; break;
    default: return false;
    }
  } else if (I->getType()->isIntegerTy(64)) {
    CReg = X86::RCX;
    RC = &X86::GR64RegClass;
    switch (I->getOpcode()) {
    case Instruction::LShr: OpReg = X86::SHR64rCL; break;
    case Instruction::AShr: OpReg = X86::SAR64rCL; break;
    case Instruction::Shl:  OpReg = X86::SHL64rCL; break;
    default: return false;
    }
  } else {
    return false;
  }

  MVT VT;
  if (!isTypeLegal(I->getType(), VT))
    return false;

  unsigned Op0Reg = getRegForValue(I->getOperand(0));
  if (Op0Reg == 0) return false;

  unsigned Op1Reg = getRegForValue(I->getOperand(1));
  if (Op1Reg == 0) return false;
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
          CReg).addReg(Op1Reg);

  // The shift instruction uses X86::CL. If we defined a super-register
  // of X86::CL, emit a subreg KILL to precisely describe what we're doing here.
  if (CReg != X86::CL)
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
            TII.get(TargetOpcode::KILL), X86::CL)
      .addReg(CReg, RegState::Kill);

  unsigned ResultReg = createResultReg(RC);
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(OpReg), ResultReg)
    .addReg(Op0Reg);
  UpdateValueMap(I, ResultReg);
  return true;
}

bool X86FastISel::X86SelectDivRem(const Instruction *I) {
  const static unsigned NumTypes = 4; // i8, i16, i32, i64
  const static unsigned NumOps   = 4; // SDiv, SRem, UDiv, URem
  const static bool S = true;  // IsSigned
  const static bool U = false; // !IsSigned
  const static unsigned Copy = TargetOpcode::COPY;
  // For the X86 DIV/IDIV instruction, in most cases the dividend
  // (numerator) must be in a specific register pair highreg:lowreg,
  // producing the quotient in lowreg and the remainder in highreg.
  // For most data types, to set up the instruction, the dividend is
  // copied into lowreg, and lowreg is sign-extended or zero-extended
  // into highreg.  The exception is i8, where the dividend is defined
  // as a single register rather than a register pair, and we
  // therefore directly sign-extend or zero-extend the dividend into
  // lowreg, instead of copying, and ignore the highreg.
  const static struct DivRemEntry {
    // The following portion depends only on the data type.
    const TargetRegisterClass *RC;
    unsigned LowInReg;  // low part of the register pair
    unsigned HighInReg; // high part of the register pair
    // The following portion depends on both the data type and the operation.
    struct DivRemResult {
    unsigned OpDivRem;        // The specific DIV/IDIV opcode to use.
    unsigned OpSignExtend;    // Opcode for sign-extending lowreg into
                              // highreg, or copying a zero into highreg.
    unsigned OpCopy;          // Opcode for copying dividend into lowreg, or
                              // zero/sign-extending into lowreg for i8.
    unsigned DivRemResultReg; // Register containing the desired result.
    bool IsOpSigned;          // Whether to use signed or unsigned form.
    } ResultTable[NumOps];
  } OpTable[NumTypes] = {
    { &X86::GR8RegClass,  X86::AX,  0, {
        { X86::IDIV8r,  0,            X86::MOVSX16rr8, X86::AL,  S }, // SDiv
        { X86::IDIV8r,  0,            X86::MOVSX16rr8, X86::AH,  S }, // SRem
        { X86::DIV8r,   0,            X86::MOVZX16rr8, X86::AL,  U }, // UDiv
        { X86::DIV8r,   0,            X86::MOVZX16rr8, X86::AH,  U }, // URem
      }
    }, // i8
    { &X86::GR16RegClass, X86::AX,  X86::DX, {
        { X86::IDIV16r, X86::CWD,     Copy,            X86::AX,  S }, // SDiv
        { X86::IDIV16r, X86::CWD,     Copy,            X86::DX,  S }, // SRem
        { X86::DIV16r,  X86::MOV32r0, Copy,            X86::AX,  U }, // UDiv
        { X86::DIV16r,  X86::MOV32r0, Copy,            X86::DX,  U }, // URem
      }
    }, // i16
    { &X86::GR32RegClass, X86::EAX, X86::EDX, {
        { X86::IDIV32r, X86::CDQ,     Copy,            X86::EAX, S }, // SDiv
        { X86::IDIV32r, X86::CDQ,     Copy,            X86::EDX, S }, // SRem
        { X86::DIV32r,  X86::MOV32r0, Copy,            X86::EAX, U }, // UDiv
        { X86::DIV32r,  X86::MOV32r0, Copy,            X86::EDX, U }, // URem
      }
    }, // i32
    { &X86::GR64RegClass, X86::RAX, X86::RDX, {
        { X86::IDIV64r, X86::CQO,     Copy,            X86::RAX, S }, // SDiv
        { X86::IDIV64r, X86::CQO,     Copy,            X86::RDX, S }, // SRem
        { X86::DIV64r,  X86::MOV32r0, Copy,            X86::RAX, U }, // UDiv
        { X86::DIV64r,  X86::MOV32r0, Copy,            X86::RDX, U }, // URem
      }
    }, // i64
  };

  MVT VT;
  if (!isTypeLegal(I->getType(), VT))
    return false;

  unsigned TypeIndex, OpIndex;
  switch (VT.SimpleTy) {
  default: return false;
  case MVT::i8:  TypeIndex = 0; break;
  case MVT::i16: TypeIndex = 1; break;
  case MVT::i32: TypeIndex = 2; break;
  case MVT::i64: TypeIndex = 3;
    if (!Subtarget->is64Bit())
      return false;
    break;
  }

  switch (I->getOpcode()) {
  default: llvm_unreachable("Unexpected div/rem opcode");
  case Instruction::SDiv: OpIndex = 0; break;
  case Instruction::SRem: OpIndex = 1; break;
  case Instruction::UDiv: OpIndex = 2; break;
  case Instruction::URem: OpIndex = 3; break;
  }

  const DivRemEntry &TypeEntry = OpTable[TypeIndex];
  const DivRemEntry::DivRemResult &OpEntry = TypeEntry.ResultTable[OpIndex];
  unsigned Op0Reg = getRegForValue(I->getOperand(0));
  if (Op0Reg == 0)
    return false;
  unsigned Op1Reg = getRegForValue(I->getOperand(1));
  if (Op1Reg == 0)
    return false;

  // Move op0 into low-order input register.
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
          TII.get(OpEntry.OpCopy), TypeEntry.LowInReg).addReg(Op0Reg);
  // Zero-extend or sign-extend into high-order input register.
  if (OpEntry.OpSignExtend) {
    if (OpEntry.IsOpSigned)
      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
              TII.get(OpEntry.OpSignExtend));
    else {
      unsigned Zero32 = createResultReg(&X86::GR32RegClass);
      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
              TII.get(X86::MOV32r0), Zero32);

      // Copy the zero into the appropriate sub/super/identical physical
      // register. Unfortunately the operations needed are not uniform enough to
      // fit neatly into the table above.
      if (VT.SimpleTy == MVT::i16) {
        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                TII.get(Copy), TypeEntry.HighInReg)
          .addReg(Zero32, 0, X86::sub_16bit);
      } else if (VT.SimpleTy == MVT::i32) {
        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                TII.get(Copy), TypeEntry.HighInReg)
            .addReg(Zero32);
      } else if (VT.SimpleTy == MVT::i64) {
        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                TII.get(TargetOpcode::SUBREG_TO_REG), TypeEntry.HighInReg)
            .addImm(0).addReg(Zero32).addImm(X86::sub_32bit);
      }
    }
  }
  // Generate the DIV/IDIV instruction.
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
          TII.get(OpEntry.OpDivRem)).addReg(Op1Reg);
  // For i8 remainder, we can't reference AH directly, as we'll end
  // up with bogus copies like %R9B = COPY %AH. Reference AX
  // instead to prevent AH references in a REX instruction.
  //
  // The current assumption of the fast register allocator is that isel
  // won't generate explicit references to the GPR8_NOREX registers. If
  // the allocator and/or the backend get enhanced to be more robust in
  // that regard, this can be, and should be, removed.
  unsigned ResultReg = 0;
  if ((I->getOpcode() == Instruction::SRem ||
       I->getOpcode() == Instruction::URem) &&
      OpEntry.DivRemResultReg == X86::AH && Subtarget->is64Bit()) {
    unsigned SourceSuperReg = createResultReg(&X86::GR16RegClass);
    unsigned ResultSuperReg = createResultReg(&X86::GR16RegClass);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
            TII.get(Copy), SourceSuperReg).addReg(X86::AX);

    // Shift AX right by 8 bits instead of using AH.
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SHR16ri),
            ResultSuperReg).addReg(SourceSuperReg).addImm(8);

    // Now reference the 8-bit subreg of the result.
    ResultReg = FastEmitInst_extractsubreg(MVT::i8, ResultSuperReg,
                                           /*Kill=*/true, X86::sub_8bit);
  }
  // Copy the result out of the physreg if we haven't already.
  if (!ResultReg) {
    ResultReg = createResultReg(TypeEntry.RC);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Copy), ResultReg)
        .addReg(OpEntry.DivRemResultReg);
  }
  UpdateValueMap(I, ResultReg);

  return true;
}

bool X86FastISel::X86SelectSelect(const Instruction *I) {
  MVT VT;
  if (!isTypeLegal(I->getType(), VT))
    return false;

  // We only use cmov here, if we don't have a cmov instruction bail.
  if (!Subtarget->hasCMov()) return false;

  unsigned Opc = 0;
  const TargetRegisterClass *RC = nullptr;
  if (VT == MVT::i16) {
    Opc = X86::CMOVE16rr;
    RC = &X86::GR16RegClass;
  } else if (VT == MVT::i32) {
    Opc = X86::CMOVE32rr;
    RC = &X86::GR32RegClass;
  } else if (VT == MVT::i64) {
    Opc = X86::CMOVE64rr;
    RC = &X86::GR64RegClass;
  } else {
    return false;
  }

  unsigned Op0Reg = getRegForValue(I->getOperand(0));
  if (Op0Reg == 0) return false;
  unsigned Op1Reg = getRegForValue(I->getOperand(1));
  if (Op1Reg == 0) return false;
  unsigned Op2Reg = getRegForValue(I->getOperand(2));
  if (Op2Reg == 0) return false;

  // Selects operate on i1, however, Op0Reg is 8 bits width and may contain
  // garbage. Indeed, only the less significant bit is supposed to be accurate.
  // If we read more than the lsb, we may see non-zero values whereas lsb
  // is zero. Therefore, we have to truncate Op0Reg to i1 for the select.
  // This is achieved by performing TEST against 1.
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::TEST8ri))
    .addReg(Op0Reg).addImm(1);
  unsigned ResultReg = createResultReg(RC);
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
    .addReg(Op1Reg).addReg(Op2Reg);
  UpdateValueMap(I, ResultReg);
  return true;
}

bool X86FastISel::X86SelectFPExt(const Instruction *I) {
  // fpext from float to double.
  if (X86ScalarSSEf64 &&
      I->getType()->isDoubleTy()) {
    const Value *V = I->getOperand(0);
    if (V->getType()->isFloatTy()) {
      unsigned OpReg = getRegForValue(V);
      if (OpReg == 0) return false;
      unsigned ResultReg = createResultReg(&X86::FR64RegClass);
      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
              TII.get(X86::CVTSS2SDrr), ResultReg)
        .addReg(OpReg);
      UpdateValueMap(I, ResultReg);
      return true;
    }
  }

  return false;
}

bool X86FastISel::X86SelectFPTrunc(const Instruction *I) {
  if (X86ScalarSSEf64) {
    if (I->getType()->isFloatTy()) {
      const Value *V = I->getOperand(0);
      if (V->getType()->isDoubleTy()) {
        unsigned OpReg = getRegForValue(V);
        if (OpReg == 0) return false;
        unsigned ResultReg = createResultReg(&X86::FR32RegClass);
        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                TII.get(X86::CVTSD2SSrr), ResultReg)
          .addReg(OpReg);
        UpdateValueMap(I, ResultReg);
        return true;
      }
    }
  }

  return false;
}

bool X86FastISel::X86SelectTrunc(const Instruction *I) {
  EVT SrcVT = TLI.getValueType(I->getOperand(0)->getType());
  EVT DstVT = TLI.getValueType(I->getType());

  // This code only handles truncation to byte.
  if (DstVT != MVT::i8 && DstVT != MVT::i1)
    return false;
  if (!TLI.isTypeLegal(SrcVT))
    return false;

  unsigned InputReg = getRegForValue(I->getOperand(0));
  if (!InputReg)
    // Unhandled operand.  Halt "fast" selection and bail.
    return false;

  if (SrcVT == MVT::i8) {
    // Truncate from i8 to i1; no code needed.
    UpdateValueMap(I, InputReg);
    return true;
  }

  if (!Subtarget->is64Bit()) {
    // If we're on x86-32; we can't extract an i8 from a general register.
    // First issue a copy to GR16_ABCD or GR32_ABCD.
    const TargetRegisterClass *CopyRC = (SrcVT == MVT::i16) ?
      (const TargetRegisterClass*)&X86::GR16_ABCDRegClass :
      (const TargetRegisterClass*)&X86::GR32_ABCDRegClass;
    unsigned CopyReg = createResultReg(CopyRC);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(TargetOpcode::COPY),
            CopyReg).addReg(InputReg);
    InputReg = CopyReg;
  }

  // Issue an extract_subreg.
  unsigned ResultReg = FastEmitInst_extractsubreg(MVT::i8,
                                                  InputReg, /*Kill=*/true,
                                                  X86::sub_8bit);
  if (!ResultReg)
    return false;

  UpdateValueMap(I, ResultReg);
  return true;
}

bool X86FastISel::IsMemcpySmall(uint64_t Len) {
  return Len <= (Subtarget->is64Bit() ? 32 : 16);
}

bool X86FastISel::TryEmitSmallMemcpy(X86AddressMode DestAM,
                                     X86AddressMode SrcAM, uint64_t Len) {

  // Make sure we don't bloat code by inlining very large memcpy's.
  if (!IsMemcpySmall(Len))
    return false;

  bool i64Legal = Subtarget->is64Bit();

  // We don't care about alignment here since we just emit integer accesses.
  while (Len) {
    MVT VT;
    if (Len >= 8 && i64Legal)
      VT = MVT::i64;
    else if (Len >= 4)
      VT = MVT::i32;
    else if (Len >= 2)
      VT = MVT::i16;
    else {
      VT = MVT::i8;
    }

    unsigned Reg;
    bool RV = X86FastEmitLoad(VT, SrcAM, Reg);
    RV &= X86FastEmitStore(VT, Reg, DestAM);
    assert(RV && "Failed to emit load or store??");

    unsigned Size = VT.getSizeInBits()/8;
    Len -= Size;
    DestAM.Disp += Size;
    SrcAM.Disp += Size;
  }

  return true;
}

bool X86FastISel::X86VisitIntrinsicCall(const IntrinsicInst &I) {
  // FIXME: Handle more intrinsics.
  switch (I.getIntrinsicID()) {
  default: return false;
  case Intrinsic::memcpy: {
    const MemCpyInst &MCI = cast<MemCpyInst>(I);
    // Don't handle volatile or variable length memcpys.
    if (MCI.isVolatile())
      return false;

    if (isa<ConstantInt>(MCI.getLength())) {
      // Small memcpy's are common enough that we want to do them
      // without a call if possible.
      uint64_t Len = cast<ConstantInt>(MCI.getLength())->getZExtValue();
      if (IsMemcpySmall(Len)) {
        X86AddressMode DestAM, SrcAM;
        if (!X86SelectAddress(MCI.getRawDest(), DestAM) ||
            !X86SelectAddress(MCI.getRawSource(), SrcAM))
          return false;
        TryEmitSmallMemcpy(DestAM, SrcAM, Len);
        return true;
      }
    }

    unsigned SizeWidth = Subtarget->is64Bit() ? 64 : 32;
    if (!MCI.getLength()->getType()->isIntegerTy(SizeWidth))
      return false;

    if (MCI.getSourceAddressSpace() > 255 || MCI.getDestAddressSpace() > 255)
      return false;

    return DoSelectCall(&I, "memcpy");
  }
  case Intrinsic::memset: {
    const MemSetInst &MSI = cast<MemSetInst>(I);

    if (MSI.isVolatile())
      return false;

    unsigned SizeWidth = Subtarget->is64Bit() ? 64 : 32;
    if (!MSI.getLength()->getType()->isIntegerTy(SizeWidth))
      return false;

    if (MSI.getDestAddressSpace() > 255)
      return false;

    return DoSelectCall(&I, "memset");
  }
  case Intrinsic::stackprotector: {
    // Emit code to store the stack guard onto the stack.
    EVT PtrTy = TLI.getPointerTy();

    const Value *Op1 = I.getArgOperand(0); // The guard's value.
    const AllocaInst *Slot = cast<AllocaInst>(I.getArgOperand(1));

    MFI.setStackProtectorIndex(FuncInfo.StaticAllocaMap[Slot]);

    // Grab the frame index.
    X86AddressMode AM;
    if (!X86SelectAddress(Slot, AM)) return false;
    if (!X86FastEmitStore(PtrTy, Op1, AM)) return false;
    return true;
  }
  case Intrinsic::dbg_declare: {
    const DbgDeclareInst *DI = cast<DbgDeclareInst>(&I);
    X86AddressMode AM;
    assert(DI->getAddress() && "Null address should be checked earlier!");
    if (!X86SelectAddress(DI->getAddress(), AM))
      return false;
    const MCInstrDesc &II = TII.get(TargetOpcode::DBG_VALUE);
    // FIXME may need to add RegState::Debug to any registers produced,
    // although ESP/EBP should be the only ones at the moment.
    addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, II), AM).
      addImm(0).addMetadata(DI->getVariable());
    return true;
  }
  case Intrinsic::trap: {
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::TRAP));
    return true;
  }
  case Intrinsic::sadd_with_overflow:
  case Intrinsic::uadd_with_overflow: {
    // FIXME: Should fold immediates.

    // Replace "add with overflow" intrinsics with an "add" instruction followed
    // by a seto/setc instruction.
    const Function *Callee = I.getCalledFunction();
    Type *RetTy =
      cast<StructType>(Callee->getReturnType())->getTypeAtIndex(unsigned(0));

    MVT VT;
    if (!isTypeLegal(RetTy, VT))
      return false;

    const Value *Op1 = I.getArgOperand(0);
    const Value *Op2 = I.getArgOperand(1);
    unsigned Reg1 = getRegForValue(Op1);
    unsigned Reg2 = getRegForValue(Op2);

    if (Reg1 == 0 || Reg2 == 0)
      // FIXME: Handle values *not* in registers.
      return false;

    unsigned OpC = 0;
    if (VT == MVT::i32)
      OpC = X86::ADD32rr;
    else if (VT == MVT::i64)
      OpC = X86::ADD64rr;
    else
      return false;

    // The call to CreateRegs builds two sequential registers, to store the
    // both the returned values.
    unsigned ResultReg = FuncInfo.CreateRegs(I.getType());
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(OpC), ResultReg)
      .addReg(Reg1).addReg(Reg2);

    unsigned Opc = X86::SETBr;
    if (I.getIntrinsicID() == Intrinsic::sadd_with_overflow)
      Opc = X86::SETOr;
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc),
            ResultReg + 1);

    UpdateValueMap(&I, ResultReg, 2);
    return true;
  }
  }
}

bool X86FastISel::FastLowerArguments() {
  if (!FuncInfo.CanLowerReturn)
    return false;

  const Function *F = FuncInfo.Fn;
  if (F->isVarArg())
    return false;

  CallingConv::ID CC = F->getCallingConv();
  if (CC != CallingConv::C)
    return false;

  if (Subtarget->isCallingConvWin64(CC))
    return false;

  if (!Subtarget->is64Bit())
    return false;
  
  // Only handle simple cases. i.e. Up to 6 i32/i64 scalar arguments.
  unsigned Idx = 1;
  for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
       I != E; ++I, ++Idx) {
    if (Idx > 6)
      return false;

    if (F->getAttributes().hasAttribute(Idx, Attribute::ByVal) ||
        F->getAttributes().hasAttribute(Idx, Attribute::InReg) ||
        F->getAttributes().hasAttribute(Idx, Attribute::StructRet) ||
        F->getAttributes().hasAttribute(Idx, Attribute::Nest))
      return false;

    Type *ArgTy = I->getType();
    if (ArgTy->isStructTy() || ArgTy->isArrayTy() || ArgTy->isVectorTy())
      return false;

    EVT ArgVT = TLI.getValueType(ArgTy);
    if (!ArgVT.isSimple()) return false;
    switch (ArgVT.getSimpleVT().SimpleTy) {
    case MVT::i32:
    case MVT::i64:
      break;
    default:
      return false;
    }
  }

  static const MCPhysReg GPR32ArgRegs[] = {
    X86::EDI, X86::ESI, X86::EDX, X86::ECX, X86::R8D, X86::R9D
  };
  static const MCPhysReg GPR64ArgRegs[] = {
    X86::RDI, X86::RSI, X86::RDX, X86::RCX, X86::R8 , X86::R9
  };

  Idx = 0;
  const TargetRegisterClass *RC32 = TLI.getRegClassFor(MVT::i32);
  const TargetRegisterClass *RC64 = TLI.getRegClassFor(MVT::i64);
  for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
       I != E; ++I, ++Idx) {
    bool is32Bit = TLI.getValueType(I->getType()) == MVT::i32;
    const TargetRegisterClass *RC = is32Bit ? RC32 : RC64;
    unsigned SrcReg = is32Bit ? GPR32ArgRegs[Idx] : GPR64ArgRegs[Idx];
    unsigned DstReg = FuncInfo.MF->addLiveIn(SrcReg, RC);
    // FIXME: Unfortunately it's necessary to emit a copy from the livein copy.
    // Without this, EmitLiveInCopies may eliminate the livein if its only
    // use is a bitcast (which isn't turned into an instruction).
    unsigned ResultReg = createResultReg(RC);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
            TII.get(TargetOpcode::COPY),
            ResultReg).addReg(DstReg, getKillRegState(true));
    UpdateValueMap(I, ResultReg);
  }
  return true;
}

bool X86FastISel::X86SelectCall(const Instruction *I) {
  const CallInst *CI = cast<CallInst>(I);
  const Value *Callee = CI->getCalledValue();

  // Can't handle inline asm yet.
  if (isa<InlineAsm>(Callee))
    return false;

  // Handle intrinsic calls.
  if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI))
    return X86VisitIntrinsicCall(*II);

  // Allow SelectionDAG isel to handle tail calls.
  if (cast<CallInst>(I)->isTailCall())
    return false;

  return DoSelectCall(I, nullptr);
}

static unsigned computeBytesPoppedByCallee(const X86Subtarget &Subtarget,
                                           const ImmutableCallSite &CS) {
  if (Subtarget.is64Bit())
    return 0;
  if (Subtarget.getTargetTriple().isOSMSVCRT())
    return 0;
  CallingConv::ID CC = CS.getCallingConv();
  if (CC == CallingConv::Fast || CC == CallingConv::GHC)
    return 0;
  if (!CS.paramHasAttr(1, Attribute::StructRet))
    return 0;
  if (CS.paramHasAttr(1, Attribute::InReg))
    return 0;
  return 4;
}

// Select either a call, or an llvm.memcpy/memmove/memset intrinsic
bool X86FastISel::DoSelectCall(const Instruction *I, const char *MemIntName) {
  const CallInst *CI = cast<CallInst>(I);
  const Value *Callee = CI->getCalledValue();

  // Handle only C and fastcc calling conventions for now.
  ImmutableCallSite CS(CI);
  CallingConv::ID CC = CS.getCallingConv();
  bool isWin64 = Subtarget->isCallingConvWin64(CC);
  if (CC != CallingConv::C && CC != CallingConv::Fast &&
      CC != CallingConv::X86_FastCall && CC != CallingConv::X86_64_Win64 &&
      CC != CallingConv::X86_64_SysV)
    return false;

  // fastcc with -tailcallopt is intended to provide a guaranteed
  // tail call optimization. Fastisel doesn't know how to do that.
  if (CC == CallingConv::Fast && TM.Options.GuaranteedTailCallOpt)
    return false;

  PointerType *PT = cast<PointerType>(CS.getCalledValue()->getType());
  FunctionType *FTy = cast<FunctionType>(PT->getElementType());
  bool isVarArg = FTy->isVarArg();

  // Don't know how to handle Win64 varargs yet.  Nothing special needed for
  // x86-32.  Special handling for x86-64 is implemented.
  if (isVarArg && isWin64)
    return false;

  // Don't know about inalloca yet.
  if (CS.hasInAllocaArgument())
    return false;

  // Fast-isel doesn't know about callee-pop yet.
  if (X86::isCalleePop(CC, Subtarget->is64Bit(), isVarArg,
                       TM.Options.GuaranteedTailCallOpt))
    return false;

  // Check whether the function can return without sret-demotion.
  SmallVector<ISD::OutputArg, 4> Outs;
  GetReturnInfo(I->getType(), CS.getAttributes(), Outs, TLI);
  bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
                                           *FuncInfo.MF, FTy->isVarArg(),
                                           Outs, FTy->getContext());
  if (!CanLowerReturn)
    return false;

  // Materialize callee address in a register. FIXME: GV address can be
  // handled with a CALLpcrel32 instead.
  X86AddressMode CalleeAM;
  if (!X86SelectCallAddress(Callee, CalleeAM))
    return false;
  unsigned CalleeOp = 0;
  const GlobalValue *GV = nullptr;
  if (CalleeAM.GV != nullptr) {
    GV = CalleeAM.GV;
  } else if (CalleeAM.Base.Reg != 0) {
    CalleeOp = CalleeAM.Base.Reg;
  } else
    return false;

  // Deal with call operands first.
  SmallVector<const Value *, 8> ArgVals;
  SmallVector<unsigned, 8> Args;
  SmallVector<MVT, 8> ArgVTs;
  SmallVector<ISD::ArgFlagsTy, 8> ArgFlags;
  unsigned arg_size = CS.arg_size();
  Args.reserve(arg_size);
  ArgVals.reserve(arg_size);
  ArgVTs.reserve(arg_size);
  ArgFlags.reserve(arg_size);
  for (ImmutableCallSite::arg_iterator i = CS.arg_begin(), e = CS.arg_end();
       i != e; ++i) {
    // If we're lowering a mem intrinsic instead of a regular call, skip the
    // last two arguments, which should not passed to the underlying functions.
    if (MemIntName && e-i <= 2)
      break;
    Value *ArgVal = *i;
    ISD::ArgFlagsTy Flags;
    unsigned AttrInd = i - CS.arg_begin() + 1;
    if (CS.paramHasAttr(AttrInd, Attribute::SExt))
      Flags.setSExt();
    if (CS.paramHasAttr(AttrInd, Attribute::ZExt))
      Flags.setZExt();

    if (CS.paramHasAttr(AttrInd, Attribute::ByVal)) {
      PointerType *Ty = cast<PointerType>(ArgVal->getType());
      Type *ElementTy = Ty->getElementType();
      unsigned FrameSize = DL.getTypeAllocSize(ElementTy);
      unsigned FrameAlign = CS.getParamAlignment(AttrInd);
      if (!FrameAlign)
        FrameAlign = TLI.getByValTypeAlignment(ElementTy);
      Flags.setByVal();
      Flags.setByValSize(FrameSize);
      Flags.setByValAlign(FrameAlign);
      if (!IsMemcpySmall(FrameSize))
        return false;
    }

    if (CS.paramHasAttr(AttrInd, Attribute::InReg))
      Flags.setInReg();
    if (CS.paramHasAttr(AttrInd, Attribute::Nest))
      Flags.setNest();

    // If this is an i1/i8/i16 argument, promote to i32 to avoid an extra
    // instruction.  This is safe because it is common to all fastisel supported
    // calling conventions on x86.
    if (ConstantInt *CI = dyn_cast<ConstantInt>(ArgVal)) {
      if (CI->getBitWidth() == 1 || CI->getBitWidth() == 8 ||
          CI->getBitWidth() == 16) {
        if (Flags.isSExt())
          ArgVal = ConstantExpr::getSExt(CI,Type::getInt32Ty(CI->getContext()));
        else
          ArgVal = ConstantExpr::getZExt(CI,Type::getInt32Ty(CI->getContext()));
      }
    }

    unsigned ArgReg;

    // Passing bools around ends up doing a trunc to i1 and passing it.
    // Codegen this as an argument + "and 1".
    if (ArgVal->getType()->isIntegerTy(1) && isa<TruncInst>(ArgVal) &&
        cast<TruncInst>(ArgVal)->getParent() == I->getParent() &&
        ArgVal->hasOneUse()) {
      ArgVal = cast<TruncInst>(ArgVal)->getOperand(0);
      ArgReg = getRegForValue(ArgVal);
      if (ArgReg == 0) return false;

      MVT ArgVT;
      if (!isTypeLegal(ArgVal->getType(), ArgVT)) return false;

      ArgReg = FastEmit_ri(ArgVT, ArgVT, ISD::AND, ArgReg,
                           ArgVal->hasOneUse(), 1);
    } else {
      ArgReg = getRegForValue(ArgVal);
    }

    if (ArgReg == 0) return false;

    Type *ArgTy = ArgVal->getType();
    MVT ArgVT;
    if (!isTypeLegal(ArgTy, ArgVT))
      return false;
    if (ArgVT == MVT::x86mmx)
      return false;
    unsigned OriginalAlignment = DL.getABITypeAlignment(ArgTy);
    Flags.setOrigAlign(OriginalAlignment);

    Args.push_back(ArgReg);
    ArgVals.push_back(ArgVal);
    ArgVTs.push_back(ArgVT);
    ArgFlags.push_back(Flags);
  }

  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CC, isVarArg, *FuncInfo.MF, TM, ArgLocs,
                 I->getParent()->getContext());

  // Allocate shadow area for Win64
  if (isWin64)
    CCInfo.AllocateStack(32, 8);

  CCInfo.AnalyzeCallOperands(ArgVTs, ArgFlags, CC_X86);

  // Get a count of how many bytes are to be pushed on the stack.
  unsigned NumBytes = CCInfo.getNextStackOffset();

  // Issue CALLSEQ_START
  unsigned AdjStackDown = TII.getCallFrameSetupOpcode();
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(AdjStackDown))
    .addImm(NumBytes);

  // Process argument: walk the register/memloc assignments, inserting
  // copies / loads.
  SmallVector<unsigned, 4> RegArgs;
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];
    unsigned Arg = Args[VA.getValNo()];
    EVT ArgVT = ArgVTs[VA.getValNo()];

    // Promote the value if needed.
    switch (VA.getLocInfo()) {
    case CCValAssign::Full: break;
    case CCValAssign::SExt: {
      assert(VA.getLocVT().isInteger() && !VA.getLocVT().isVector() &&
             "Unexpected extend");
      bool Emitted = X86FastEmitExtend(ISD::SIGN_EXTEND, VA.getLocVT(),
                                       Arg, ArgVT, Arg);
      assert(Emitted && "Failed to emit a sext!"); (void)Emitted;
      ArgVT = VA.getLocVT();
      break;
    }
    case CCValAssign::ZExt: {
      assert(VA.getLocVT().isInteger() && !VA.getLocVT().isVector() &&
             "Unexpected extend");
      bool Emitted = X86FastEmitExtend(ISD::ZERO_EXTEND, VA.getLocVT(),
                                       Arg, ArgVT, Arg);
      assert(Emitted && "Failed to emit a zext!"); (void)Emitted;
      ArgVT = VA.getLocVT();
      break;
    }
    case CCValAssign::AExt: {
      assert(VA.getLocVT().isInteger() && !VA.getLocVT().isVector() &&
             "Unexpected extend");
      bool Emitted = X86FastEmitExtend(ISD::ANY_EXTEND, VA.getLocVT(),
                                       Arg, ArgVT, Arg);
      if (!Emitted)
        Emitted = X86FastEmitExtend(ISD::ZERO_EXTEND, VA.getLocVT(),
                                    Arg, ArgVT, Arg);
      if (!Emitted)
        Emitted = X86FastEmitExtend(ISD::SIGN_EXTEND, VA.getLocVT(),
                                    Arg, ArgVT, Arg);

      assert(Emitted && "Failed to emit a aext!"); (void)Emitted;
      ArgVT = VA.getLocVT();
      break;
    }
    case CCValAssign::BCvt: {
      unsigned BC = FastEmit_r(ArgVT.getSimpleVT(), VA.getLocVT(),
                               ISD::BITCAST, Arg, /*TODO: Kill=*/false);
      assert(BC != 0 && "Failed to emit a bitcast!");
      Arg = BC;
      ArgVT = VA.getLocVT();
      break;
    }
    case CCValAssign::VExt: 
      // VExt has not been implemented, so this should be impossible to reach
      // for now.  However, fallback to Selection DAG isel once implemented.
      return false;
    case CCValAssign::Indirect:
      // FIXME: Indirect doesn't need extending, but fast-isel doesn't fully
      // support this.
      return false;
    case CCValAssign::FPExt:
      llvm_unreachable("Unexpected loc info!");
    }

    if (VA.isRegLoc()) {
      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
              TII.get(TargetOpcode::COPY), VA.getLocReg()).addReg(Arg);
      RegArgs.push_back(VA.getLocReg());
    } else {
      unsigned LocMemOffset = VA.getLocMemOffset();
      X86AddressMode AM;
      const X86RegisterInfo *RegInfo = static_cast<const X86RegisterInfo*>(
          getTargetMachine()->getRegisterInfo());
      AM.Base.Reg = RegInfo->getStackRegister();
      AM.Disp = LocMemOffset;
      const Value *ArgVal = ArgVals[VA.getValNo()];
      ISD::ArgFlagsTy Flags = ArgFlags[VA.getValNo()];

      if (Flags.isByVal()) {
        X86AddressMode SrcAM;
        SrcAM.Base.Reg = Arg;
        bool Res = TryEmitSmallMemcpy(AM, SrcAM, Flags.getByValSize());
        assert(Res && "memcpy length already checked!"); (void)Res;
      } else if (isa<ConstantInt>(ArgVal) || isa<ConstantPointerNull>(ArgVal)) {
        // If this is a really simple value, emit this with the Value* version
        // of X86FastEmitStore.  If it isn't simple, we don't want to do this,
        // as it can cause us to reevaluate the argument.
        if (!X86FastEmitStore(ArgVT, ArgVal, AM))
          return false;
      } else {
        if (!X86FastEmitStore(ArgVT, Arg, AM))
          return false;
      }
    }
  }

  // ELF / PIC requires GOT in the EBX register before function calls via PLT
  // GOT pointer.
  if (Subtarget->isPICStyleGOT()) {
    unsigned Base = getInstrInfo()->getGlobalBaseReg(FuncInfo.MF);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
            TII.get(TargetOpcode::COPY), X86::EBX).addReg(Base);
  }

  if (Subtarget->is64Bit() && isVarArg && !isWin64) {
    // Count the number of XMM registers allocated.
    static const MCPhysReg XMMArgRegs[] = {
      X86::XMM0, X86::XMM1, X86::XMM2, X86::XMM3,
      X86::XMM4, X86::XMM5, X86::XMM6, X86::XMM7
    };
    unsigned NumXMMRegs = CCInfo.getFirstUnallocated(XMMArgRegs, 8);
    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::MOV8ri),
            X86::AL).addImm(NumXMMRegs);
  }

  // Issue the call.
  MachineInstrBuilder MIB;
  if (CalleeOp) {
    // Register-indirect call.
    unsigned CallOpc;
    if (Subtarget->is64Bit())
      CallOpc = X86::CALL64r;
    else
      CallOpc = X86::CALL32r;
    MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CallOpc))
      .addReg(CalleeOp);

  } else {
    // Direct call.
    assert(GV && "Not a direct call");
    unsigned CallOpc;
    if (Subtarget->is64Bit())
      CallOpc = X86::CALL64pcrel32;
    else
      CallOpc = X86::CALLpcrel32;

    // See if we need any target-specific flags on the GV operand.
    unsigned char OpFlags = 0;

    // On ELF targets, in both X86-64 and X86-32 mode, direct calls to
    // external symbols most go through the PLT in PIC mode.  If the symbol
    // has hidden or protected visibility, or if it is static or local, then
    // we don't need to use the PLT - we can directly call it.
    if (Subtarget->isTargetELF() &&
        TM.getRelocationModel() == Reloc::PIC_ &&
        GV->hasDefaultVisibility() && !GV->hasLocalLinkage()) {
      OpFlags = X86II::MO_PLT;
    } else if (Subtarget->isPICStyleStubAny() &&
               (GV->isDeclaration() || GV->isWeakForLinker()) &&
               (!Subtarget->getTargetTriple().isMacOSX() ||
                Subtarget->getTargetTriple().isMacOSXVersionLT(10, 5))) {
      // PC-relative references to external symbols should go through $stub,
      // unless we're building with the leopard linker or later, which
      // automatically synthesizes these stubs.
      OpFlags = X86II::MO_DARWIN_STUB;
    }


    MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CallOpc));
    if (MemIntName)
      MIB.addExternalSymbol(MemIntName, OpFlags);
    else
      MIB.addGlobalAddress(GV, 0, OpFlags);
  }

  // Add a register mask with the call-preserved registers.
  // Proper defs for return values will be added by setPhysRegsDeadExcept().
  MIB.addRegMask(TRI.getCallPreservedMask(CS.getCallingConv()));

  // Add an implicit use GOT pointer in EBX.
  if (Subtarget->isPICStyleGOT())
    MIB.addReg(X86::EBX, RegState::Implicit);

  if (Subtarget->is64Bit() && isVarArg && !isWin64)
    MIB.addReg(X86::AL, RegState::Implicit);

  // Add implicit physical register uses to the call.
  for (unsigned i = 0, e = RegArgs.size(); i != e; ++i)
    MIB.addReg(RegArgs[i], RegState::Implicit);

  // Issue CALLSEQ_END
  unsigned AdjStackUp = TII.getCallFrameDestroyOpcode();
  const unsigned NumBytesCallee = computeBytesPoppedByCallee(*Subtarget, CS);
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(AdjStackUp))
    .addImm(NumBytes).addImm(NumBytesCallee);

  // Build info for return calling conv lowering code.
  // FIXME: This is practically a copy-paste from TargetLowering::LowerCallTo.
  SmallVector<ISD::InputArg, 32> Ins;
  SmallVector<EVT, 4> RetTys;
  ComputeValueVTs(TLI, I->getType(), RetTys);
  for (unsigned i = 0, e = RetTys.size(); i != e; ++i) {
    EVT VT = RetTys[i];
    MVT RegisterVT = TLI.getRegisterType(I->getParent()->getContext(), VT);
    unsigned NumRegs = TLI.getNumRegisters(I->getParent()->getContext(), VT);
    for (unsigned j = 0; j != NumRegs; ++j) {
      ISD::InputArg MyFlags;
      MyFlags.VT = RegisterVT;
      MyFlags.Used = !CS.getInstruction()->use_empty();
      if (CS.paramHasAttr(0, Attribute::SExt))
        MyFlags.Flags.setSExt();
      if (CS.paramHasAttr(0, Attribute::ZExt))
        MyFlags.Flags.setZExt();
      if (CS.paramHasAttr(0, Attribute::InReg))
        MyFlags.Flags.setInReg();
      Ins.push_back(MyFlags);
    }
  }

  // Now handle call return values.
  SmallVector<unsigned, 4> UsedRegs;
  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCRetInfo(CC, false, *FuncInfo.MF, TM, RVLocs,
                    I->getParent()->getContext());
  unsigned ResultReg = FuncInfo.CreateRegs(I->getType());
  CCRetInfo.AnalyzeCallResult(Ins, RetCC_X86);
  for (unsigned i = 0; i != RVLocs.size(); ++i) {
    EVT CopyVT = RVLocs[i].getValVT();
    unsigned CopyReg = ResultReg + i;

    // If this is a call to a function that returns an fp value on the x87 fp
    // stack, but where we prefer to use the value in xmm registers, copy it
    // out as F80 and use a truncate to move it from fp stack reg to xmm reg.
    if ((RVLocs[i].getLocReg() == X86::ST0 ||
         RVLocs[i].getLocReg() == X86::ST1)) {
      if (isScalarFPTypeInSSEReg(RVLocs[i].getValVT())) {
        CopyVT = MVT::f80;
        CopyReg = createResultReg(&X86::RFP80RegClass);
      }
      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
              TII.get(X86::FpPOP_RETVAL), CopyReg);
    } else {
      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
              TII.get(TargetOpcode::COPY),
              CopyReg).addReg(RVLocs[i].getLocReg());
      UsedRegs.push_back(RVLocs[i].getLocReg());
    }

    if (CopyVT != RVLocs[i].getValVT()) {
      // Round the F80 the right size, which also moves to the appropriate xmm
      // register. This is accomplished by storing the F80 value in memory and
      // then loading it back. Ewww...
      EVT ResVT = RVLocs[i].getValVT();
      unsigned Opc = ResVT == MVT::f32 ? X86::ST_Fp80m32 : X86::ST_Fp80m64;
      unsigned MemSize = ResVT.getSizeInBits()/8;
      int FI = MFI.CreateStackObject(MemSize, MemSize, false);
      addFrameReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                                TII.get(Opc)), FI)
        .addReg(CopyReg);
      Opc = ResVT == MVT::f32 ? X86::MOVSSrm : X86::MOVSDrm;
      addFrameReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                                TII.get(Opc), ResultReg + i), FI);
    }
  }

  if (RVLocs.size())
    UpdateValueMap(I, ResultReg, RVLocs.size());

  // Set all unused physreg defs as dead.
  static_cast<MachineInstr *>(MIB)->setPhysRegsDeadExcept(UsedRegs, TRI);

  return true;
}


bool
X86FastISel::TargetSelectInstruction(const Instruction *I)  {
  switch (I->getOpcode()) {
  default: break;
  case Instruction::Load:
    return X86SelectLoad(I);
  case Instruction::Store:
    return X86SelectStore(I);
  case Instruction::Ret:
    return X86SelectRet(I);
  case Instruction::ICmp:
  case Instruction::FCmp:
    return X86SelectCmp(I);
  case Instruction::ZExt:
    return X86SelectZExt(I);
  case Instruction::Br:
    return X86SelectBranch(I);
  case Instruction::Call:
    return X86SelectCall(I);
  case Instruction::LShr:
  case Instruction::AShr:
  case Instruction::Shl:
    return X86SelectShift(I);
  case Instruction::SDiv:
  case Instruction::UDiv:
  case Instruction::SRem:
  case Instruction::URem:
    return X86SelectDivRem(I);
  case Instruction::Select:
    return X86SelectSelect(I);
  case Instruction::Trunc:
    return X86SelectTrunc(I);
  case Instruction::FPExt:
    return X86SelectFPExt(I);
  case Instruction::FPTrunc:
    return X86SelectFPTrunc(I);
  case Instruction::IntToPtr: // Deliberate fall-through.
  case Instruction::PtrToInt: {
    EVT SrcVT = TLI.getValueType(I->getOperand(0)->getType());
    EVT DstVT = TLI.getValueType(I->getType());
    if (DstVT.bitsGT(SrcVT))
      return X86SelectZExt(I);
    if (DstVT.bitsLT(SrcVT))
      return X86SelectTrunc(I);
    unsigned Reg = getRegForValue(I->getOperand(0));
    if (Reg == 0) return false;
    UpdateValueMap(I, Reg);
    return true;
  }
  }

  return false;
}

unsigned X86FastISel::TargetMaterializeConstant(const Constant *C) {
  MVT VT;
  if (!isTypeLegal(C->getType(), VT))
    return 0;

  // Can't handle alternate code models yet.
  if (TM.getCodeModel() != CodeModel::Small)
    return 0;

  // Get opcode and regclass of the output for the given load instruction.
  unsigned Opc = 0;
  const TargetRegisterClass *RC = nullptr;
  switch (VT.SimpleTy) {
  default: return 0;
  case MVT::i8:
    Opc = X86::MOV8rm;
    RC  = &X86::GR8RegClass;
    break;
  case MVT::i16:
    Opc = X86::MOV16rm;
    RC  = &X86::GR16RegClass;
    break;
  case MVT::i32:
    Opc = X86::MOV32rm;
    RC  = &X86::GR32RegClass;
    break;
  case MVT::i64:
    // Must be in x86-64 mode.
    Opc = X86::MOV64rm;
    RC  = &X86::GR64RegClass;
    break;
  case MVT::f32:
    if (X86ScalarSSEf32) {
      Opc = Subtarget->hasAVX() ? X86::VMOVSSrm : X86::MOVSSrm;
      RC  = &X86::FR32RegClass;
    } else {
      Opc = X86::LD_Fp32m;
      RC  = &X86::RFP32RegClass;
    }
    break;
  case MVT::f64:
    if (X86ScalarSSEf64) {
      Opc = Subtarget->hasAVX() ? X86::VMOVSDrm : X86::MOVSDrm;
      RC  = &X86::FR64RegClass;
    } else {
      Opc = X86::LD_Fp64m;
      RC  = &X86::RFP64RegClass;
    }
    break;
  case MVT::f80:
    // No f80 support yet.
    return 0;
  }

  // Materialize addresses with LEA instructions.
  if (isa<GlobalValue>(C)) {
    X86AddressMode AM;
    if (X86SelectAddress(C, AM)) {
      // If the expression is just a basereg, then we're done, otherwise we need
      // to emit an LEA.
      if (AM.BaseType == X86AddressMode::RegBase &&
          AM.IndexReg == 0 && AM.Disp == 0 && AM.GV == nullptr)
        return AM.Base.Reg;

      Opc = TLI.getPointerTy() == MVT::i32 ? X86::LEA32r : X86::LEA64r;
      unsigned ResultReg = createResultReg(RC);
      addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                             TII.get(Opc), ResultReg), AM);
      return ResultReg;
    }
    return 0;
  }

  // MachineConstantPool wants an explicit alignment.
  unsigned Align = DL.getPrefTypeAlignment(C->getType());
  if (Align == 0) {
    // Alignment of vector types.  FIXME!
    Align = DL.getTypeAllocSize(C->getType());
  }

  // x86-32 PIC requires a PIC base register for constant pools.
  unsigned PICBase = 0;
  unsigned char OpFlag = 0;
  if (Subtarget->isPICStyleStubPIC()) { // Not dynamic-no-pic
    OpFlag = X86II::MO_PIC_BASE_OFFSET;
    PICBase = getInstrInfo()->getGlobalBaseReg(FuncInfo.MF);
  } else if (Subtarget->isPICStyleGOT()) {
    OpFlag = X86II::MO_GOTOFF;
    PICBase = getInstrInfo()->getGlobalBaseReg(FuncInfo.MF);
  } else if (Subtarget->isPICStyleRIPRel() &&
             TM.getCodeModel() == CodeModel::Small) {
    PICBase = X86::RIP;
  }

  // Create the load from the constant pool.
  unsigned MCPOffset = MCP.getConstantPoolIndex(C, Align);
  unsigned ResultReg = createResultReg(RC);
  addConstantPoolReference(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                                   TII.get(Opc), ResultReg),
                           MCPOffset, PICBase, OpFlag);

  return ResultReg;
}

unsigned X86FastISel::TargetMaterializeAlloca(const AllocaInst *C) {
  // Fail on dynamic allocas. At this point, getRegForValue has already
  // checked its CSE maps, so if we're here trying to handle a dynamic
  // alloca, we're not going to succeed. X86SelectAddress has a
  // check for dynamic allocas, because it's called directly from
  // various places, but TargetMaterializeAlloca also needs a check
  // in order to avoid recursion between getRegForValue,
  // X86SelectAddrss, and TargetMaterializeAlloca.
  if (!FuncInfo.StaticAllocaMap.count(C))
    return 0;
  assert(C->isStaticAlloca() && "dynamic alloca in the static alloca map?");

  X86AddressMode AM;
  if (!X86SelectAddress(C, AM))
    return 0;
  unsigned Opc = Subtarget->is64Bit() ? X86::LEA64r : X86::LEA32r;
  const TargetRegisterClass* RC = TLI.getRegClassFor(TLI.getPointerTy());
  unsigned ResultReg = createResultReg(RC);
  addFullAddress(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                         TII.get(Opc), ResultReg), AM);
  return ResultReg;
}

unsigned X86FastISel::TargetMaterializeFloatZero(const ConstantFP *CF) {
  MVT VT;
  if (!isTypeLegal(CF->getType(), VT))
    return 0;

  // Get opcode and regclass for the given zero.
  unsigned Opc = 0;
  const TargetRegisterClass *RC = nullptr;
  switch (VT.SimpleTy) {
  default: return 0;
  case MVT::f32:
    if (X86ScalarSSEf32) {
      Opc = X86::FsFLD0SS;
      RC  = &X86::FR32RegClass;
    } else {
      Opc = X86::LD_Fp032;
      RC  = &X86::RFP32RegClass;
    }
    break;
  case MVT::f64:
    if (X86ScalarSSEf64) {
      Opc = X86::FsFLD0SD;
      RC  = &X86::FR64RegClass;
    } else {
      Opc = X86::LD_Fp064;
      RC  = &X86::RFP64RegClass;
    }
    break;
  case MVT::f80:
    // No f80 support yet.
    return 0;
  }

  unsigned ResultReg = createResultReg(RC);
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg);
  return ResultReg;
}


bool X86FastISel::tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
                                      const LoadInst *LI) {
  X86AddressMode AM;
  if (!X86SelectAddress(LI->getOperand(0), AM))
    return false;

  const X86InstrInfo &XII = (const X86InstrInfo&)TII;

  unsigned Size = DL.getTypeAllocSize(LI->getType());
  unsigned Alignment = LI->getAlignment();

  SmallVector<MachineOperand, 8> AddrOps;
  AM.getFullAddress(AddrOps);

  MachineInstr *Result =
    XII.foldMemoryOperandImpl(*FuncInfo.MF, MI, OpNo, AddrOps, Size, Alignment);
  if (!Result) return false;

  FuncInfo.MBB->insert(FuncInfo.InsertPt, Result);
  MI->eraseFromParent();
  return true;
}


namespace llvm {
  FastISel *X86::createFastISel(FunctionLoweringInfo &funcInfo,
                                const TargetLibraryInfo *libInfo) {
    return new X86FastISel(funcInfo, libInfo);
  }
}