aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target/ARM/ARMISelLowering.cpp
blob: 0d0d81fee56ccd62e8782eac6b6cba35aecb803b (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
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358
6359
6360
6361
6362
6363
6364
6365
6366
6367
6368
6369
6370
6371
6372
6373
6374
6375
6376
6377
6378
6379
6380
6381
6382
6383
6384
6385
6386
6387
6388
6389
6390
6391
6392
6393
6394
6395
6396
6397
6398
6399
6400
6401
6402
6403
6404
6405
6406
6407
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445
6446
6447
6448
6449
6450
6451
6452
6453
6454
6455
6456
6457
6458
6459
6460
6461
6462
6463
6464
6465
6466
6467
6468
6469
6470
6471
6472
6473
6474
6475
6476
6477
6478
6479
6480
6481
6482
6483
6484
6485
6486
6487
6488
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506
6507
6508
6509
6510
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524
6525
6526
6527
6528
6529
6530
6531
6532
6533
6534
6535
6536
6537
6538
6539
6540
6541
6542
6543
6544
6545
6546
6547
6548
6549
6550
6551
6552
6553
6554
6555
6556
6557
6558
6559
6560
6561
6562
6563
6564
6565
6566
6567
6568
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583
6584
6585
6586
6587
6588
6589
6590
6591
6592
6593
6594
6595
6596
6597
6598
6599
6600
6601
6602
6603
6604
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
6616
6617
6618
6619
6620
6621
6622
6623
6624
6625
6626
6627
6628
6629
6630
6631
6632
6633
6634
6635
6636
6637
6638
6639
6640
6641
6642
6643
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658
6659
6660
6661
6662
6663
6664
6665
6666
6667
6668
6669
6670
6671
6672
6673
6674
6675
6676
6677
6678
6679
6680
6681
6682
6683
6684
6685
6686
6687
6688
6689
6690
6691
6692
6693
6694
6695
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706
6707
6708
6709
6710
6711
6712
6713
6714
6715
6716
6717
6718
6719
6720
6721
6722
6723
6724
6725
6726
6727
6728
6729
6730
6731
6732
6733
6734
6735
6736
6737
6738
6739
6740
6741
6742
6743
6744
6745
6746
6747
6748
6749
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768
6769
6770
6771
6772
6773
6774
6775
6776
6777
6778
6779
6780
6781
6782
6783
6784
6785
6786
6787
6788
6789
6790
6791
6792
6793
6794
6795
6796
6797
6798
6799
6800
6801
6802
6803
6804
6805
6806
6807
6808
6809
6810
6811
6812
6813
6814
6815
6816
6817
6818
6819
6820
6821
6822
6823
6824
6825
6826
6827
6828
6829
6830
6831
6832
6833
6834
6835
6836
6837
6838
6839
6840
6841
6842
6843
6844
6845
6846
6847
6848
6849
6850
6851
6852
6853
6854
6855
6856
6857
6858
6859
6860
6861
6862
6863
6864
6865
6866
6867
6868
6869
6870
6871
6872
6873
6874
6875
6876
6877
6878
6879
6880
6881
6882
6883
6884
6885
6886
6887
6888
6889
6890
6891
6892
6893
6894
6895
6896
6897
6898
6899
6900
6901
6902
6903
6904
6905
6906
6907
6908
6909
6910
6911
6912
6913
6914
6915
6916
6917
6918
6919
6920
6921
6922
6923
6924
6925
6926
6927
6928
6929
6930
6931
6932
6933
6934
6935
6936
6937
6938
6939
6940
6941
6942
6943
6944
6945
6946
6947
6948
6949
6950
6951
6952
6953
6954
6955
6956
6957
6958
6959
6960
6961
6962
6963
6964
6965
6966
6967
6968
6969
6970
6971
6972
6973
6974
6975
6976
6977
6978
6979
6980
6981
6982
6983
6984
6985
6986
6987
6988
6989
6990
6991
6992
6993
6994
6995
6996
6997
6998
6999
7000
7001
7002
7003
7004
7005
7006
7007
7008
7009
7010
7011
7012
7013
7014
7015
7016
7017
7018
7019
7020
7021
7022
7023
7024
7025
7026
7027
7028
7029
7030
7031
7032
7033
7034
7035
7036
7037
7038
7039
7040
7041
7042
7043
7044
7045
7046
7047
7048
7049
7050
7051
7052
7053
7054
7055
7056
7057
7058
7059
7060
7061
7062
7063
7064
7065
7066
7067
7068
7069
7070
7071
7072
7073
7074
7075
7076
7077
7078
7079
7080
7081
7082
7083
7084
7085
7086
7087
7088
7089
7090
7091
7092
7093
7094
7095
7096
7097
7098
7099
7100
7101
7102
7103
7104
7105
7106
7107
7108
7109
7110
7111
7112
7113
7114
7115
7116
7117
7118
7119
7120
7121
7122
7123
7124
7125
7126
7127
7128
7129
7130
7131
7132
7133
7134
7135
7136
7137
7138
7139
7140
7141
7142
7143
7144
7145
7146
7147
7148
7149
7150
7151
7152
7153
7154
7155
7156
7157
7158
7159
7160
7161
7162
7163
7164
7165
7166
7167
7168
7169
7170
7171
7172
7173
7174
7175
7176
7177
7178
7179
7180
7181
7182
7183
7184
7185
7186
7187
7188
7189
7190
7191
7192
7193
7194
7195
7196
7197
7198
7199
7200
7201
7202
7203
7204
7205
7206
7207
7208
7209
7210
7211
7212
7213
7214
7215
7216
7217
7218
7219
7220
7221
7222
7223
7224
7225
7226
7227
7228
7229
7230
7231
7232
7233
7234
7235
7236
7237
7238
7239
7240
7241
7242
7243
7244
7245
7246
7247
7248
7249
7250
7251
7252
7253
7254
7255
7256
7257
7258
7259
7260
7261
7262
7263
7264
7265
7266
7267
7268
7269
7270
7271
7272
7273
7274
7275
7276
7277
7278
7279
7280
7281
7282
7283
7284
7285
7286
7287
7288
7289
7290
7291
7292
7293
7294
7295
7296
7297
7298
7299
7300
7301
7302
7303
7304
7305
7306
7307
7308
7309
7310
7311
7312
7313
7314
7315
7316
7317
7318
7319
7320
7321
7322
7323
7324
7325
7326
7327
7328
7329
7330
7331
7332
7333
7334
7335
7336
7337
7338
7339
7340
7341
7342
7343
7344
7345
7346
7347
7348
7349
7350
7351
7352
7353
7354
7355
7356
7357
7358
7359
7360
7361
7362
7363
7364
7365
7366
7367
7368
7369
7370
7371
7372
7373
7374
7375
7376
7377
7378
7379
7380
7381
7382
7383
7384
7385
7386
7387
7388
7389
7390
7391
7392
7393
7394
7395
7396
7397
7398
7399
7400
7401
7402
7403
7404
7405
7406
7407
7408
7409
7410
7411
7412
7413
7414
7415
7416
7417
7418
7419
7420
7421
7422
7423
7424
7425
7426
7427
7428
7429
7430
7431
7432
7433
7434
7435
7436
7437
7438
7439
7440
7441
7442
7443
7444
7445
7446
7447
7448
7449
7450
7451
7452
7453
7454
7455
7456
7457
7458
7459
7460
7461
7462
7463
7464
7465
7466
7467
7468
7469
7470
7471
7472
7473
7474
7475
7476
7477
7478
7479
7480
7481
7482
7483
7484
7485
7486
7487
7488
7489
7490
7491
7492
7493
7494
7495
7496
7497
7498
7499
7500
7501
7502
7503
7504
7505
7506
7507
7508
7509
7510
7511
7512
7513
7514
7515
7516
7517
7518
7519
7520
7521
7522
7523
7524
7525
7526
7527
7528
7529
7530
7531
7532
7533
7534
7535
7536
7537
7538
7539
7540
7541
7542
7543
7544
7545
7546
7547
7548
7549
7550
7551
7552
7553
7554
7555
7556
7557
7558
7559
7560
7561
7562
7563
7564
7565
7566
7567
7568
7569
7570
7571
7572
7573
7574
7575
7576
7577
7578
7579
7580
7581
7582
7583
7584
7585
7586
7587
7588
7589
7590
7591
7592
7593
7594
7595
7596
7597
7598
7599
7600
7601
7602
7603
7604
7605
7606
7607
7608
7609
7610
7611
7612
7613
7614
7615
7616
7617
7618
7619
7620
7621
7622
7623
7624
7625
7626
7627
7628
7629
7630
7631
7632
7633
7634
7635
7636
7637
7638
7639
7640
7641
7642
7643
7644
7645
7646
7647
7648
7649
7650
7651
7652
7653
7654
7655
7656
7657
7658
7659
7660
7661
7662
7663
7664
7665
7666
7667
7668
7669
7670
7671
7672
7673
7674
7675
7676
7677
7678
7679
7680
7681
7682
7683
7684
7685
7686
7687
7688
7689
7690
7691
7692
7693
7694
7695
7696
7697
7698
7699
7700
7701
7702
7703
7704
7705
7706
7707
7708
7709
7710
7711
7712
7713
7714
7715
7716
7717
7718
7719
7720
7721
7722
7723
7724
7725
7726
7727
7728
7729
7730
7731
7732
7733
7734
7735
7736
7737
7738
7739
7740
7741
7742
7743
7744
7745
7746
7747
7748
7749
7750
7751
7752
7753
7754
7755
7756
7757
7758
7759
7760
7761
7762
7763
7764
7765
7766
7767
7768
7769
7770
7771
7772
7773
7774
7775
7776
7777
7778
7779
7780
7781
7782
7783
7784
7785
7786
7787
7788
7789
7790
7791
7792
7793
7794
7795
7796
7797
7798
7799
7800
7801
7802
7803
7804
7805
7806
7807
7808
7809
7810
7811
7812
7813
7814
7815
7816
7817
7818
7819
7820
7821
7822
7823
7824
7825
7826
7827
7828
7829
7830
7831
7832
7833
7834
7835
7836
7837
7838
7839
7840
7841
7842
7843
7844
7845
7846
7847
7848
7849
7850
7851
7852
7853
7854
7855
7856
7857
7858
7859
7860
7861
7862
7863
7864
7865
7866
7867
7868
7869
7870
7871
7872
7873
7874
7875
7876
7877
7878
7879
7880
7881
7882
7883
7884
7885
7886
7887
7888
7889
7890
7891
7892
7893
7894
7895
7896
7897
7898
7899
7900
7901
7902
7903
7904
7905
7906
7907
7908
7909
7910
7911
7912
7913
7914
7915
7916
7917
7918
7919
7920
7921
7922
7923
7924
7925
7926
7927
7928
7929
7930
7931
7932
7933
7934
7935
7936
7937
7938
7939
7940
7941
7942
7943
7944
7945
7946
7947
7948
7949
7950
7951
7952
7953
7954
7955
7956
7957
7958
7959
7960
7961
7962
7963
7964
7965
7966
7967
7968
7969
7970
7971
7972
7973
7974
7975
7976
7977
7978
7979
7980
7981
7982
7983
7984
7985
7986
7987
7988
7989
7990
7991
7992
7993
7994
7995
7996
7997
7998
7999
8000
8001
8002
8003
8004
8005
8006
8007
8008
8009
8010
8011
8012
8013
8014
8015
8016
8017
8018
8019
8020
8021
8022
8023
8024
8025
8026
8027
8028
8029
8030
8031
8032
8033
8034
8035
8036
8037
8038
8039
8040
8041
8042
8043
8044
8045
8046
8047
8048
8049
8050
8051
8052
8053
8054
8055
8056
8057
8058
8059
8060
8061
8062
8063
8064
8065
8066
8067
8068
8069
8070
8071
8072
8073
8074
8075
8076
8077
8078
8079
8080
8081
8082
8083
8084
8085
8086
8087
8088
8089
8090
8091
8092
8093
8094
8095
8096
8097
8098
8099
8100
8101
8102
8103
8104
8105
8106
8107
8108
8109
8110
8111
8112
8113
8114
8115
8116
8117
8118
8119
8120
8121
8122
8123
8124
8125
8126
8127
8128
8129
8130
8131
8132
8133
8134
8135
8136
8137
8138
8139
8140
8141
8142
8143
8144
8145
8146
8147
8148
8149
8150
8151
8152
8153
8154
8155
8156
8157
8158
8159
8160
8161
8162
8163
8164
8165
8166
8167
8168
8169
8170
8171
8172
8173
8174
8175
8176
8177
8178
8179
8180
8181
8182
8183
8184
8185
8186
8187
8188
8189
8190
8191
8192
8193
8194
8195
8196
8197
8198
8199
8200
8201
8202
8203
8204
8205
8206
8207
8208
8209
8210
8211
8212
8213
8214
8215
8216
8217
8218
8219
8220
8221
8222
8223
8224
8225
8226
8227
8228
8229
8230
8231
8232
8233
8234
8235
8236
8237
8238
8239
8240
8241
8242
8243
8244
8245
8246
8247
8248
8249
8250
8251
8252
8253
8254
8255
8256
8257
8258
8259
8260
8261
8262
8263
8264
8265
8266
8267
8268
8269
8270
8271
8272
8273
8274
8275
8276
8277
8278
8279
8280
8281
8282
8283
8284
8285
8286
8287
8288
8289
8290
8291
8292
8293
8294
8295
8296
8297
8298
8299
8300
8301
8302
8303
8304
8305
8306
8307
8308
8309
8310
8311
8312
8313
8314
8315
8316
8317
8318
8319
8320
8321
8322
8323
8324
8325
8326
8327
8328
8329
8330
8331
8332
8333
8334
8335
8336
8337
8338
8339
8340
8341
8342
8343
8344
8345
8346
8347
8348
8349
8350
8351
8352
8353
8354
8355
8356
8357
8358
8359
8360
8361
8362
8363
8364
8365
8366
8367
8368
8369
8370
8371
8372
8373
8374
8375
8376
8377
8378
8379
8380
8381
8382
8383
8384
8385
8386
8387
8388
8389
8390
8391
8392
8393
8394
8395
8396
8397
8398
8399
8400
8401
8402
8403
8404
8405
8406
8407
8408
8409
8410
8411
8412
8413
8414
8415
8416
8417
8418
8419
8420
8421
8422
8423
8424
8425
8426
8427
8428
8429
8430
8431
8432
8433
8434
8435
8436
8437
8438
8439
8440
8441
8442
8443
8444
8445
8446
8447
8448
8449
8450
8451
8452
8453
8454
8455
8456
8457
8458
8459
8460
8461
8462
8463
8464
8465
8466
8467
8468
8469
8470
8471
8472
8473
8474
8475
8476
8477
8478
8479
8480
8481
8482
8483
8484
8485
8486
8487
8488
8489
8490
8491
8492
8493
8494
8495
8496
8497
8498
8499
8500
8501
8502
8503
8504
8505
8506
8507
8508
8509
8510
8511
8512
8513
8514
8515
8516
8517
8518
8519
8520
8521
8522
8523
8524
8525
8526
8527
8528
8529
8530
8531
8532
8533
8534
8535
8536
8537
8538
8539
8540
8541
8542
8543
8544
8545
8546
8547
8548
8549
8550
8551
8552
8553
8554
8555
8556
8557
8558
8559
8560
8561
8562
8563
8564
8565
8566
8567
8568
8569
8570
8571
8572
8573
8574
8575
8576
8577
8578
8579
8580
8581
8582
8583
8584
8585
8586
8587
8588
8589
8590
8591
8592
8593
8594
8595
8596
8597
8598
8599
8600
8601
8602
8603
8604
8605
8606
8607
8608
8609
8610
8611
8612
8613
8614
8615
8616
8617
8618
8619
8620
8621
8622
8623
8624
8625
8626
8627
8628
8629
8630
8631
8632
8633
8634
8635
8636
8637
8638
8639
8640
8641
8642
8643
8644
8645
8646
8647
8648
8649
8650
8651
8652
8653
8654
8655
8656
8657
8658
8659
8660
8661
8662
8663
8664
8665
8666
8667
8668
8669
8670
8671
8672
8673
8674
8675
8676
8677
8678
8679
8680
8681
8682
8683
8684
8685
8686
8687
8688
8689
8690
8691
8692
8693
8694
8695
8696
8697
8698
8699
8700
8701
8702
8703
8704
8705
8706
8707
8708
8709
8710
8711
8712
8713
8714
8715
8716
8717
8718
8719
8720
8721
8722
8723
8724
8725
8726
8727
8728
8729
8730
8731
8732
8733
8734
8735
8736
8737
8738
8739
8740
8741
8742
8743
8744
8745
8746
8747
8748
8749
8750
8751
8752
8753
8754
8755
8756
8757
8758
8759
8760
8761
8762
8763
8764
8765
8766
8767
8768
8769
8770
8771
8772
8773
8774
8775
8776
8777
8778
8779
8780
8781
8782
8783
8784
8785
8786
8787
8788
8789
8790
8791
8792
8793
8794
8795
8796
8797
8798
8799
8800
8801
8802
8803
8804
8805
8806
8807
8808
8809
8810
8811
8812
8813
8814
8815
8816
8817
8818
8819
8820
8821
8822
8823
8824
8825
8826
8827
8828
8829
8830
8831
8832
8833
8834
8835
8836
8837
8838
8839
8840
8841
8842
8843
8844
8845
8846
8847
8848
8849
8850
8851
8852
8853
8854
8855
8856
8857
8858
8859
8860
8861
8862
8863
8864
8865
8866
8867
8868
8869
8870
8871
8872
8873
8874
8875
8876
8877
8878
8879
8880
8881
8882
8883
8884
8885
8886
8887
8888
8889
8890
8891
8892
8893
8894
8895
8896
8897
8898
8899
8900
8901
8902
8903
8904
8905
8906
8907
8908
8909
8910
8911
8912
8913
8914
8915
8916
8917
8918
8919
8920
8921
8922
8923
8924
8925
8926
8927
8928
8929
8930
8931
8932
8933
8934
8935
8936
8937
8938
8939
8940
8941
8942
8943
8944
8945
8946
8947
8948
8949
8950
8951
8952
8953
8954
8955
8956
8957
8958
8959
8960
8961
8962
8963
8964
8965
8966
8967
8968
8969
8970
8971
8972
8973
8974
8975
8976
8977
8978
8979
8980
8981
8982
8983
8984
8985
8986
8987
8988
8989
8990
8991
8992
8993
8994
8995
8996
8997
8998
8999
9000
9001
9002
9003
9004
9005
9006
9007
9008
9009
9010
9011
9012
9013
9014
9015
9016
9017
9018
9019
9020
9021
9022
9023
9024
9025
9026
9027
9028
9029
9030
9031
9032
9033
9034
9035
9036
9037
9038
9039
9040
9041
9042
9043
9044
9045
9046
9047
9048
9049
9050
9051
9052
9053
9054
9055
9056
9057
9058
9059
9060
9061
9062
9063
9064
9065
9066
9067
9068
9069
9070
9071
9072
9073
9074
9075
9076
9077
9078
9079
9080
9081
9082
9083
9084
9085
9086
9087
9088
9089
9090
9091
9092
9093
9094
9095
9096
9097
9098
9099
9100
9101
9102
9103
9104
9105
9106
9107
9108
9109
9110
9111
9112
9113
9114
9115
9116
9117
9118
9119
9120
9121
9122
9123
9124
9125
9126
9127
9128
9129
9130
9131
9132
9133
9134
9135
9136
9137
9138
9139
9140
9141
9142
9143
9144
9145
9146
9147
9148
9149
9150
9151
9152
9153
9154
9155
9156
9157
9158
9159
9160
9161
9162
9163
9164
9165
9166
9167
9168
9169
9170
9171
9172
9173
9174
9175
9176
9177
9178
9179
9180
9181
9182
9183
9184
9185
9186
9187
9188
9189
9190
9191
9192
9193
9194
9195
9196
9197
9198
9199
9200
9201
9202
9203
9204
9205
9206
9207
9208
9209
9210
9211
9212
9213
9214
9215
9216
9217
9218
9219
9220
9221
9222
9223
9224
9225
9226
9227
9228
9229
9230
9231
9232
9233
9234
9235
9236
9237
9238
9239
9240
9241
9242
9243
9244
9245
9246
9247
9248
9249
9250
9251
9252
9253
9254
9255
9256
9257
9258
9259
9260
9261
9262
9263
9264
9265
9266
9267
9268
9269
9270
9271
9272
9273
9274
9275
9276
9277
9278
9279
9280
9281
9282
9283
9284
9285
9286
9287
9288
9289
9290
9291
9292
9293
9294
9295
9296
9297
9298
9299
9300
9301
9302
9303
9304
9305
9306
9307
9308
9309
9310
9311
9312
9313
9314
9315
9316
9317
9318
9319
9320
9321
9322
9323
9324
9325
9326
9327
9328
9329
9330
9331
9332
9333
9334
9335
9336
9337
9338
9339
9340
9341
9342
9343
9344
9345
9346
9347
9348
9349
9350
9351
9352
9353
9354
9355
9356
9357
9358
9359
9360
9361
9362
9363
9364
9365
9366
9367
9368
9369
9370
9371
9372
9373
9374
9375
9376
9377
9378
9379
9380
9381
9382
9383
9384
9385
9386
9387
9388
9389
9390
9391
9392
9393
9394
9395
9396
9397
9398
9399
9400
9401
9402
9403
9404
9405
9406
9407
9408
9409
9410
9411
9412
9413
9414
9415
9416
9417
9418
9419
9420
9421
9422
9423
9424
9425
9426
9427
9428
9429
9430
9431
9432
9433
9434
9435
9436
9437
9438
9439
9440
9441
9442
9443
9444
9445
9446
9447
9448
9449
9450
9451
9452
9453
9454
9455
9456
9457
9458
9459
9460
9461
9462
9463
9464
9465
9466
9467
9468
9469
9470
9471
9472
9473
9474
9475
9476
9477
9478
9479
9480
9481
9482
9483
9484
9485
9486
9487
9488
9489
9490
9491
9492
9493
9494
9495
9496
9497
9498
9499
9500
9501
9502
9503
9504
9505
9506
9507
9508
9509
9510
9511
9512
9513
9514
9515
9516
9517
9518
9519
9520
9521
9522
9523
9524
9525
9526
9527
9528
9529
9530
9531
9532
9533
9534
9535
9536
9537
9538
9539
9540
9541
9542
9543
9544
9545
9546
9547
9548
9549
9550
9551
9552
9553
9554
9555
9556
9557
9558
9559
9560
9561
9562
9563
9564
9565
9566
9567
9568
9569
9570
9571
9572
9573
9574
9575
9576
9577
9578
9579
9580
9581
9582
9583
9584
9585
9586
9587
9588
9589
9590
9591
9592
9593
9594
9595
9596
9597
9598
9599
9600
9601
9602
9603
9604
9605
9606
9607
9608
9609
9610
9611
9612
9613
9614
9615
9616
9617
9618
9619
9620
9621
9622
9623
9624
9625
9626
9627
9628
9629
9630
9631
9632
9633
9634
9635
9636
9637
9638
9639
9640
9641
9642
9643
9644
9645
9646
9647
9648
9649
9650
9651
9652
9653
9654
9655
9656
9657
9658
9659
9660
9661
9662
9663
9664
9665
9666
9667
9668
9669
9670
9671
9672
9673
9674
9675
9676
9677
9678
9679
9680
9681
9682
9683
9684
9685
9686
9687
9688
9689
9690
9691
9692
9693
9694
9695
9696
9697
9698
9699
9700
9701
9702
9703
9704
9705
9706
9707
9708
9709
9710
9711
9712
9713
9714
9715
9716
9717
9718
9719
9720
9721
9722
9723
9724
9725
9726
9727
9728
9729
9730
9731
9732
9733
9734
9735
9736
9737
9738
9739
9740
9741
9742
9743
9744
9745
9746
9747
9748
9749
9750
9751
9752
9753
9754
9755
9756
9757
9758
9759
9760
9761
9762
9763
9764
9765
9766
9767
9768
9769
9770
9771
9772
9773
9774
9775
9776
9777
9778
9779
9780
9781
9782
9783
9784
9785
9786
9787
9788
9789
9790
9791
9792
9793
9794
9795
9796
9797
9798
9799
9800
9801
9802
9803
9804
9805
9806
9807
9808
9809
9810
9811
9812
9813
9814
9815
9816
9817
9818
9819
9820
9821
9822
9823
9824
9825
9826
9827
9828
9829
9830
9831
9832
9833
9834
9835
9836
9837
9838
9839
9840
9841
9842
9843
9844
9845
9846
9847
9848
9849
9850
9851
9852
9853
9854
9855
9856
9857
9858
9859
9860
9861
9862
9863
9864
9865
9866
9867
9868
9869
9870
9871
9872
9873
9874
9875
9876
9877
9878
9879
9880
9881
9882
9883
9884
9885
9886
9887
9888
9889
9890
9891
9892
9893
9894
9895
9896
9897
9898
9899
9900
9901
9902
9903
9904
9905
9906
9907
9908
9909
9910
9911
9912
9913
9914
9915
9916
9917
9918
9919
9920
9921
9922
9923
9924
9925
9926
9927
9928
9929
9930
9931
9932
9933
9934
9935
9936
9937
9938
9939
9940
9941
9942
9943
9944
9945
9946
9947
9948
9949
9950
9951
9952
9953
9954
9955
9956
9957
9958
9959
9960
9961
9962
9963
9964
9965
9966
9967
9968
9969
9970
9971
9972
9973
9974
9975
9976
9977
9978
9979
9980
9981
9982
9983
9984
9985
9986
9987
9988
9989
9990
9991
9992
9993
9994
9995
9996
9997
9998
9999
10000
10001
10002
10003
10004
10005
10006
10007
10008
10009
10010
10011
10012
10013
10014
10015
10016
10017
10018
10019
10020
10021
10022
10023
10024
10025
10026
10027
10028
10029
10030
10031
10032
10033
10034
10035
10036
10037
10038
10039
10040
10041
10042
10043
10044
10045
10046
10047
10048
10049
10050
10051
10052
10053
10054
10055
10056
10057
10058
10059
10060
10061
10062
10063
10064
10065
10066
10067
10068
10069
10070
10071
10072
10073
10074
10075
10076
10077
10078
10079
10080
10081
10082
10083
10084
10085
10086
10087
10088
10089
10090
10091
10092
10093
10094
10095
10096
10097
10098
10099
10100
10101
10102
10103
10104
10105
10106
10107
10108
10109
10110
10111
10112
10113
10114
10115
10116
10117
10118
10119
10120
10121
10122
10123
10124
10125
10126
10127
10128
10129
10130
10131
10132
10133
10134
10135
10136
10137
10138
10139
10140
10141
10142
10143
10144
10145
10146
10147
10148
10149
10150
10151
10152
10153
10154
10155
10156
10157
10158
10159
10160
10161
10162
10163
10164
10165
10166
10167
10168
10169
10170
10171
10172
10173
10174
10175
10176
10177
10178
10179
10180
10181
10182
10183
10184
10185
10186
10187
10188
10189
10190
10191
10192
10193
10194
10195
10196
10197
10198
10199
10200
10201
10202
10203
10204
10205
10206
10207
10208
10209
10210
10211
10212
10213
10214
10215
10216
10217
10218
10219
10220
10221
10222
10223
10224
10225
10226
10227
10228
10229
10230
10231
10232
10233
10234
10235
10236
10237
10238
10239
10240
10241
10242
10243
10244
10245
10246
10247
10248
10249
10250
10251
10252
10253
10254
10255
10256
10257
10258
10259
10260
10261
10262
10263
10264
10265
10266
10267
10268
10269
10270
10271
10272
10273
10274
10275
10276
10277
10278
10279
10280
10281
10282
10283
10284
10285
10286
10287
10288
10289
10290
10291
10292
10293
10294
10295
10296
10297
10298
10299
10300
10301
10302
10303
10304
10305
10306
10307
10308
10309
10310
10311
10312
10313
10314
10315
10316
10317
10318
10319
10320
10321
10322
10323
10324
10325
10326
10327
10328
10329
10330
10331
10332
10333
10334
10335
10336
10337
10338
10339
10340
10341
10342
10343
10344
10345
10346
10347
10348
10349
10350
10351
10352
10353
10354
10355
10356
10357
10358
10359
10360
10361
10362
10363
10364
10365
10366
10367
10368
10369
10370
10371
10372
10373
10374
10375
10376
10377
10378
10379
10380
10381
10382
10383
10384
10385
10386
10387
10388
10389
10390
10391
10392
10393
10394
10395
10396
10397
10398
10399
10400
10401
10402
10403
10404
10405
10406
10407
10408
10409
10410
10411
10412
10413
10414
10415
10416
10417
10418
10419
10420
10421
10422
10423
10424
10425
10426
10427
10428
10429
10430
10431
10432
10433
10434
10435
10436
10437
10438
10439
10440
10441
10442
10443
10444
10445
10446
10447
10448
10449
10450
10451
10452
10453
10454
10455
10456
10457
10458
10459
10460
10461
10462
10463
10464
10465
10466
10467
10468
10469
10470
10471
10472
10473
10474
10475
10476
10477
10478
10479
10480
10481
10482
10483
10484
10485
10486
10487
10488
10489
10490
10491
10492
10493
10494
10495
10496
10497
10498
10499
10500
10501
10502
10503
10504
10505
10506
10507
10508
10509
10510
10511
10512
10513
10514
10515
10516
10517
10518
10519
10520
10521
10522
10523
10524
10525
10526
10527
10528
10529
10530
10531
10532
10533
10534
10535
10536
10537
10538
10539
10540
10541
10542
10543
10544
10545
10546
10547
10548
10549
10550
10551
10552
10553
10554
10555
10556
10557
10558
10559
10560
10561
10562
10563
10564
10565
10566
10567
10568
10569
10570
10571
10572
10573
10574
10575
10576
10577
10578
10579
10580
10581
10582
10583
10584
10585
10586
10587
10588
10589
10590
10591
10592
10593
10594
10595
10596
10597
10598
10599
10600
10601
10602
10603
10604
10605
10606
10607
10608
10609
10610
10611
10612
10613
10614
10615
10616
10617
10618
10619
10620
10621
10622
10623
10624
10625
10626
10627
10628
10629
10630
10631
10632
10633
10634
10635
10636
10637
10638
10639
10640
10641
10642
10643
10644
10645
10646
10647
10648
10649
10650
10651
10652
10653
10654
10655
10656
10657
10658
10659
10660
10661
10662
10663
10664
10665
10666
10667
10668
10669
10670
10671
10672
10673
10674
10675
10676
10677
10678
10679
10680
10681
10682
10683
10684
10685
10686
10687
10688
10689
10690
10691
10692
10693
10694
10695
10696
10697
10698
10699
10700
10701
10702
10703
10704
10705
10706
10707
10708
10709
10710
10711
10712
10713
10714
10715
10716
10717
10718
10719
10720
10721
10722
10723
10724
10725
10726
10727
10728
10729
10730
10731
10732
10733
10734
10735
10736
10737
10738
10739
10740
10741
10742
10743
10744
10745
10746
10747
10748
10749
10750
10751
10752
10753
10754
10755
10756
10757
10758
10759
10760
10761
10762
10763
10764
10765
10766
10767
10768
10769
10770
10771
10772
10773
10774
10775
10776
10777
10778
10779
10780
10781
10782
10783
10784
10785
10786
10787
10788
10789
10790
10791
10792
10793
10794
10795
10796
10797
10798
10799
10800
10801
10802
10803
10804
10805
10806
10807
10808
10809
10810
10811
10812
10813
10814
10815
10816
10817
10818
10819
10820
10821
10822
10823
10824
10825
10826
10827
10828
10829
10830
10831
10832
10833
10834
10835
10836
10837
10838
10839
10840
10841
10842
10843
10844
10845
10846
10847
10848
10849
10850
10851
10852
10853
10854
10855
10856
10857
10858
10859
10860
10861
10862
10863
10864
10865
10866
10867
10868
10869
10870
10871
10872
10873
10874
10875
10876
10877
10878
10879
10880
10881
10882
10883
10884
10885
10886
10887
10888
10889
10890
10891
10892
10893
10894
10895
10896
10897
10898
10899
10900
10901
10902
10903
10904
10905
10906
10907
10908
10909
10910
10911
10912
10913
10914
10915
10916
10917
10918
10919
10920
10921
10922
10923
10924
10925
10926
10927
10928
10929
10930
10931
10932
10933
10934
10935
10936
10937
10938
10939
10940
10941
10942
10943
10944
10945
10946
10947
10948
10949
10950
10951
10952
10953
10954
10955
10956
10957
10958
10959
10960
10961
10962
10963
10964
10965
10966
10967
10968
10969
10970
10971
10972
10973
10974
10975
10976
10977
10978
10979
10980
10981
10982
10983
10984
10985
10986
10987
10988
10989
10990
10991
10992
10993
10994
10995
10996
10997
10998
10999
11000
11001
11002
11003
11004
11005
11006
11007
11008
11009
11010
11011
11012
11013
11014
11015
11016
11017
11018
11019
11020
11021
11022
11023
11024
11025
11026
11027
11028
11029
11030
11031
11032
11033
11034
11035
11036
11037
11038
11039
11040
11041
11042
11043
11044
11045
11046
11047
11048
11049
11050
11051
11052
11053
11054
11055
11056
11057
11058
11059
11060
11061
11062
11063
11064
11065
11066
11067
11068
11069
11070
11071
11072
11073
11074
11075
11076
11077
11078
11079
11080
11081
11082
11083
11084
11085
11086
11087
11088
11089
11090
11091
11092
11093
11094
11095
11096
11097
11098
11099
11100
11101
11102
11103
11104
11105
11106
11107
11108
11109
11110
11111
11112
11113
11114
11115
11116
11117
11118
11119
11120
11121
11122
11123
11124
11125
11126
11127
11128
11129
11130
11131
11132
11133
11134
11135
11136
11137
11138
11139
11140
11141
11142
11143
11144
11145
11146
11147
11148
11149
11150
11151
11152
11153
11154
11155
11156
11157
11158
11159
11160
11161
11162
11163
11164
11165
11166
11167
11168
11169
11170
11171
11172
11173
11174
11175
11176
11177
11178
11179
11180
11181
11182
11183
11184
11185
11186
11187
11188
11189
11190
11191
11192
11193
11194
11195
11196
11197
11198
11199
11200
11201
11202
11203
11204
11205
11206
11207
11208
11209
11210
11211
11212
11213
11214
11215
11216
11217
11218
11219
11220
11221
11222
11223
11224
11225
11226
11227
11228
11229
11230
11231
11232
11233
11234
11235
11236
11237
11238
11239
11240
11241
11242
11243
11244
11245
11246
11247
11248
11249
11250
11251
11252
11253
11254
11255
11256
11257
11258
11259
11260
11261
11262
11263
11264
11265
11266
11267
11268
11269
11270
11271
11272
11273
11274
11275
11276
11277
11278
11279
11280
11281
11282
11283
11284
11285
11286
11287
11288
11289
//===-- ARMISelLowering.cpp - ARM DAG Lowering 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 interfaces that ARM uses to lower LLVM code into a
// selection DAG.
//
//===----------------------------------------------------------------------===//

#include "ARMISelLowering.h"
#include "ARMCallingConv.h"
#include "ARMConstantPoolValue.h"
#include "ARMMachineFunctionInfo.h"
#include "ARMPerfectShuffle.h"
#include "ARMSubtarget.h"
#include "ARMTargetMachine.h"
#include "ARMTargetObjectFile.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/IntrinsicLowering.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Type.h"
#include "llvm/MC/MCSectionMachO.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
#include <utility>
using namespace llvm;

#define DEBUG_TYPE "arm-isel"

STATISTIC(NumTailCalls, "Number of tail calls");
STATISTIC(NumMovwMovt, "Number of GAs materialized with movw + movt");
STATISTIC(NumLoopByVals, "Number of loops generated for byval arguments");

cl::opt<bool>
EnableARMLongCalls("arm-long-calls", cl::Hidden,
  cl::desc("Generate calls via indirect call instructions"),
  cl::init(false));

static cl::opt<bool>
ARMInterworking("arm-interworking", cl::Hidden,
  cl::desc("Enable / disable ARM interworking (for debugging only)"),
  cl::init(true));

namespace {
  class ARMCCState : public CCState {
  public:
    ARMCCState(CallingConv::ID CC, bool isVarArg, MachineFunction &MF,
               SmallVectorImpl<CCValAssign> &locs, LLVMContext &C,
               ParmContext PC)
        : CCState(CC, isVarArg, MF, locs, C) {
      assert(((PC == Call) || (PC == Prologue)) &&
             "ARMCCState users must specify whether their context is call"
             "or prologue generation.");
      CallOrPrologue = PC;
    }
  };
}

// The APCS parameter registers.
static const MCPhysReg GPRArgRegs[] = {
  ARM::R0, ARM::R1, ARM::R2, ARM::R3
};

void ARMTargetLowering::addTypeForNEON(MVT VT, MVT PromotedLdStVT,
                                       MVT PromotedBitwiseVT) {
  if (VT != PromotedLdStVT) {
    setOperationAction(ISD::LOAD, VT, Promote);
    AddPromotedToType (ISD::LOAD, VT, PromotedLdStVT);

    setOperationAction(ISD::STORE, VT, Promote);
    AddPromotedToType (ISD::STORE, VT, PromotedLdStVT);
  }

  MVT ElemTy = VT.getVectorElementType();
  if (ElemTy != MVT::i64 && ElemTy != MVT::f64)
    setOperationAction(ISD::SETCC, VT, Custom);
  setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
  setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
  if (ElemTy == MVT::i32) {
    setOperationAction(ISD::SINT_TO_FP, VT, Custom);
    setOperationAction(ISD::UINT_TO_FP, VT, Custom);
    setOperationAction(ISD::FP_TO_SINT, VT, Custom);
    setOperationAction(ISD::FP_TO_UINT, VT, Custom);
  } else {
    setOperationAction(ISD::SINT_TO_FP, VT, Expand);
    setOperationAction(ISD::UINT_TO_FP, VT, Expand);
    setOperationAction(ISD::FP_TO_SINT, VT, Expand);
    setOperationAction(ISD::FP_TO_UINT, VT, Expand);
  }
  setOperationAction(ISD::BUILD_VECTOR,      VT, Custom);
  setOperationAction(ISD::VECTOR_SHUFFLE,    VT, Custom);
  setOperationAction(ISD::CONCAT_VECTORS,    VT, Legal);
  setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Legal);
  setOperationAction(ISD::SELECT,            VT, Expand);
  setOperationAction(ISD::SELECT_CC,         VT, Expand);
  setOperationAction(ISD::VSELECT,           VT, Expand);
  setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
  if (VT.isInteger()) {
    setOperationAction(ISD::SHL, VT, Custom);
    setOperationAction(ISD::SRA, VT, Custom);
    setOperationAction(ISD::SRL, VT, Custom);
  }

  // Promote all bit-wise operations.
  if (VT.isInteger() && VT != PromotedBitwiseVT) {
    setOperationAction(ISD::AND, VT, Promote);
    AddPromotedToType (ISD::AND, VT, PromotedBitwiseVT);
    setOperationAction(ISD::OR,  VT, Promote);
    AddPromotedToType (ISD::OR,  VT, PromotedBitwiseVT);
    setOperationAction(ISD::XOR, VT, Promote);
    AddPromotedToType (ISD::XOR, VT, PromotedBitwiseVT);
  }

  // Neon does not support vector divide/remainder operations.
  setOperationAction(ISD::SDIV, VT, Expand);
  setOperationAction(ISD::UDIV, VT, Expand);
  setOperationAction(ISD::FDIV, VT, Expand);
  setOperationAction(ISD::SREM, VT, Expand);
  setOperationAction(ISD::UREM, VT, Expand);
  setOperationAction(ISD::FREM, VT, Expand);
}

void ARMTargetLowering::addDRTypeForNEON(MVT VT) {
  addRegisterClass(VT, &ARM::DPRRegClass);
  addTypeForNEON(VT, MVT::f64, MVT::v2i32);
}

void ARMTargetLowering::addQRTypeForNEON(MVT VT) {
  addRegisterClass(VT, &ARM::DPairRegClass);
  addTypeForNEON(VT, MVT::v2f64, MVT::v4i32);
}

ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM)
    : TargetLowering(TM) {
  Subtarget = &TM.getSubtarget<ARMSubtarget>();
  RegInfo = TM.getSubtargetImpl()->getRegisterInfo();
  Itins = TM.getSubtargetImpl()->getInstrItineraryData();

  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);

  if (Subtarget->isTargetMachO()) {
    // Uses VFP for Thumb libfuncs if available.
    if (Subtarget->isThumb() && Subtarget->hasVFP2() &&
        Subtarget->hasARMOps() && !TM.Options.UseSoftFloat) {
      // Single-precision floating-point arithmetic.
      setLibcallName(RTLIB::ADD_F32, "__addsf3vfp");
      setLibcallName(RTLIB::SUB_F32, "__subsf3vfp");
      setLibcallName(RTLIB::MUL_F32, "__mulsf3vfp");
      setLibcallName(RTLIB::DIV_F32, "__divsf3vfp");

      // Double-precision floating-point arithmetic.
      setLibcallName(RTLIB::ADD_F64, "__adddf3vfp");
      setLibcallName(RTLIB::SUB_F64, "__subdf3vfp");
      setLibcallName(RTLIB::MUL_F64, "__muldf3vfp");
      setLibcallName(RTLIB::DIV_F64, "__divdf3vfp");

      // Single-precision comparisons.
      setLibcallName(RTLIB::OEQ_F32, "__eqsf2vfp");
      setLibcallName(RTLIB::UNE_F32, "__nesf2vfp");
      setLibcallName(RTLIB::OLT_F32, "__ltsf2vfp");
      setLibcallName(RTLIB::OLE_F32, "__lesf2vfp");
      setLibcallName(RTLIB::OGE_F32, "__gesf2vfp");
      setLibcallName(RTLIB::OGT_F32, "__gtsf2vfp");
      setLibcallName(RTLIB::UO_F32,  "__unordsf2vfp");
      setLibcallName(RTLIB::O_F32,   "__unordsf2vfp");

      setCmpLibcallCC(RTLIB::OEQ_F32, ISD::SETNE);
      setCmpLibcallCC(RTLIB::UNE_F32, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OLT_F32, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OLE_F32, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OGE_F32, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OGT_F32, ISD::SETNE);
      setCmpLibcallCC(RTLIB::UO_F32,  ISD::SETNE);
      setCmpLibcallCC(RTLIB::O_F32,   ISD::SETEQ);

      // Double-precision comparisons.
      setLibcallName(RTLIB::OEQ_F64, "__eqdf2vfp");
      setLibcallName(RTLIB::UNE_F64, "__nedf2vfp");
      setLibcallName(RTLIB::OLT_F64, "__ltdf2vfp");
      setLibcallName(RTLIB::OLE_F64, "__ledf2vfp");
      setLibcallName(RTLIB::OGE_F64, "__gedf2vfp");
      setLibcallName(RTLIB::OGT_F64, "__gtdf2vfp");
      setLibcallName(RTLIB::UO_F64,  "__unorddf2vfp");
      setLibcallName(RTLIB::O_F64,   "__unorddf2vfp");

      setCmpLibcallCC(RTLIB::OEQ_F64, ISD::SETNE);
      setCmpLibcallCC(RTLIB::UNE_F64, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OLT_F64, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OLE_F64, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OGE_F64, ISD::SETNE);
      setCmpLibcallCC(RTLIB::OGT_F64, ISD::SETNE);
      setCmpLibcallCC(RTLIB::UO_F64,  ISD::SETNE);
      setCmpLibcallCC(RTLIB::O_F64,   ISD::SETEQ);

      // Floating-point to integer conversions.
      // i64 conversions are done via library routines even when generating VFP
      // instructions, so use the same ones.
      setLibcallName(RTLIB::FPTOSINT_F64_I32, "__fixdfsivfp");
      setLibcallName(RTLIB::FPTOUINT_F64_I32, "__fixunsdfsivfp");
      setLibcallName(RTLIB::FPTOSINT_F32_I32, "__fixsfsivfp");
      setLibcallName(RTLIB::FPTOUINT_F32_I32, "__fixunssfsivfp");

      // Conversions between floating types.
      setLibcallName(RTLIB::FPROUND_F64_F32, "__truncdfsf2vfp");
      setLibcallName(RTLIB::FPEXT_F32_F64,   "__extendsfdf2vfp");

      // Integer to floating-point conversions.
      // i64 conversions are done via library routines even when generating VFP
      // instructions, so use the same ones.
      // FIXME: There appears to be some naming inconsistency in ARM libgcc:
      // e.g., __floatunsidf vs. __floatunssidfvfp.
      setLibcallName(RTLIB::SINTTOFP_I32_F64, "__floatsidfvfp");
      setLibcallName(RTLIB::UINTTOFP_I32_F64, "__floatunssidfvfp");
      setLibcallName(RTLIB::SINTTOFP_I32_F32, "__floatsisfvfp");
      setLibcallName(RTLIB::UINTTOFP_I32_F32, "__floatunssisfvfp");
    }
  }

  // These libcalls are not available in 32-bit.
  setLibcallName(RTLIB::SHL_I128, nullptr);
  setLibcallName(RTLIB::SRL_I128, nullptr);
  setLibcallName(RTLIB::SRA_I128, nullptr);

  if (Subtarget->isAAPCS_ABI() && !Subtarget->isTargetMachO() &&
      !Subtarget->isTargetWindows()) {
    static const struct {
      const RTLIB::Libcall Op;
      const char * const Name;
      const CallingConv::ID CC;
      const ISD::CondCode Cond;
    } LibraryCalls[] = {
      // Double-precision floating-point arithmetic helper functions
      // RTABI chapter 4.1.2, Table 2
      { RTLIB::ADD_F64, "__aeabi_dadd", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::DIV_F64, "__aeabi_ddiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::MUL_F64, "__aeabi_dmul", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SUB_F64, "__aeabi_dsub", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },

      // Double-precision floating-point comparison helper functions
      // RTABI chapter 4.1.2, Table 3
      { RTLIB::OEQ_F64, "__aeabi_dcmpeq", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::UNE_F64, "__aeabi_dcmpeq", CallingConv::ARM_AAPCS, ISD::SETEQ },
      { RTLIB::OLT_F64, "__aeabi_dcmplt", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::OLE_F64, "__aeabi_dcmple", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::OGE_F64, "__aeabi_dcmpge", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::OGT_F64, "__aeabi_dcmpgt", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::UO_F64,  "__aeabi_dcmpun", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::O_F64,   "__aeabi_dcmpun", CallingConv::ARM_AAPCS, ISD::SETEQ },

      // Single-precision floating-point arithmetic helper functions
      // RTABI chapter 4.1.2, Table 4
      { RTLIB::ADD_F32, "__aeabi_fadd", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::DIV_F32, "__aeabi_fdiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::MUL_F32, "__aeabi_fmul", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SUB_F32, "__aeabi_fsub", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },

      // Single-precision floating-point comparison helper functions
      // RTABI chapter 4.1.2, Table 5
      { RTLIB::OEQ_F32, "__aeabi_fcmpeq", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::UNE_F32, "__aeabi_fcmpeq", CallingConv::ARM_AAPCS, ISD::SETEQ },
      { RTLIB::OLT_F32, "__aeabi_fcmplt", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::OLE_F32, "__aeabi_fcmple", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::OGE_F32, "__aeabi_fcmpge", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::OGT_F32, "__aeabi_fcmpgt", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::UO_F32,  "__aeabi_fcmpun", CallingConv::ARM_AAPCS, ISD::SETNE },
      { RTLIB::O_F32,   "__aeabi_fcmpun", CallingConv::ARM_AAPCS, ISD::SETEQ },

      // Floating-point to integer conversions.
      // RTABI chapter 4.1.2, Table 6
      { RTLIB::FPTOSINT_F64_I32, "__aeabi_d2iz",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPTOUINT_F64_I32, "__aeabi_d2uiz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPTOSINT_F64_I64, "__aeabi_d2lz",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPTOUINT_F64_I64, "__aeabi_d2ulz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPTOSINT_F32_I32, "__aeabi_f2iz",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPTOUINT_F32_I32, "__aeabi_f2uiz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPTOSINT_F32_I64, "__aeabi_f2lz",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPTOUINT_F32_I64, "__aeabi_f2ulz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },

      // Conversions between floating types.
      // RTABI chapter 4.1.2, Table 7
      { RTLIB::FPROUND_F64_F32, "__aeabi_d2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPROUND_F64_F16, "__aeabi_d2h", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::FPEXT_F32_F64,   "__aeabi_f2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },

      // Integer to floating-point conversions.
      // RTABI chapter 4.1.2, Table 8
      { RTLIB::SINTTOFP_I32_F64, "__aeabi_i2d",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UINTTOFP_I32_F64, "__aeabi_ui2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SINTTOFP_I64_F64, "__aeabi_l2d",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UINTTOFP_I64_F64, "__aeabi_ul2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SINTTOFP_I32_F32, "__aeabi_i2f",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UINTTOFP_I32_F32, "__aeabi_ui2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SINTTOFP_I64_F32, "__aeabi_l2f",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UINTTOFP_I64_F32, "__aeabi_ul2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },

      // Long long helper functions
      // RTABI chapter 4.2, Table 9
      { RTLIB::MUL_I64, "__aeabi_lmul", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SHL_I64, "__aeabi_llsl", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SRL_I64, "__aeabi_llsr", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SRA_I64, "__aeabi_lasr", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },

      // Integer division functions
      // RTABI chapter 4.3.1
      { RTLIB::SDIV_I8,  "__aeabi_idiv",     CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SDIV_I16, "__aeabi_idiv",     CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SDIV_I32, "__aeabi_idiv",     CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::SDIV_I64, "__aeabi_ldivmod",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UDIV_I8,  "__aeabi_uidiv",    CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UDIV_I16, "__aeabi_uidiv",    CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UDIV_I32, "__aeabi_uidiv",    CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::UDIV_I64, "__aeabi_uldivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },

      // Memory operations
      // RTABI chapter 4.3.4
      { RTLIB::MEMCPY,  "__aeabi_memcpy",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::MEMMOVE, "__aeabi_memmove", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
      { RTLIB::MEMSET,  "__aeabi_memset",  CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
    };

    for (const auto &LC : LibraryCalls) {
      setLibcallName(LC.Op, LC.Name);
      setLibcallCallingConv(LC.Op, LC.CC);
      if (LC.Cond != ISD::SETCC_INVALID)
        setCmpLibcallCC(LC.Op, LC.Cond);
    }
  }

  if (Subtarget->isTargetWindows()) {
    static const struct {
      const RTLIB::Libcall Op;
      const char * const Name;
      const CallingConv::ID CC;
    } LibraryCalls[] = {
      { RTLIB::FPTOSINT_F32_I64, "__stoi64", CallingConv::ARM_AAPCS_VFP },
      { RTLIB::FPTOSINT_F64_I64, "__dtoi64", CallingConv::ARM_AAPCS_VFP },
      { RTLIB::FPTOUINT_F32_I64, "__stou64", CallingConv::ARM_AAPCS_VFP },
      { RTLIB::FPTOUINT_F64_I64, "__dtou64", CallingConv::ARM_AAPCS_VFP },
      { RTLIB::SINTTOFP_I64_F32, "__i64tos", CallingConv::ARM_AAPCS_VFP },
      { RTLIB::SINTTOFP_I64_F64, "__i64tod", CallingConv::ARM_AAPCS_VFP },
      { RTLIB::UINTTOFP_I64_F32, "__u64tos", CallingConv::ARM_AAPCS_VFP },
      { RTLIB::UINTTOFP_I64_F64, "__u64tod", CallingConv::ARM_AAPCS_VFP },
    };

    for (const auto &LC : LibraryCalls) {
      setLibcallName(LC.Op, LC.Name);
      setLibcallCallingConv(LC.Op, LC.CC);
    }
  }

  // Use divmod compiler-rt calls for iOS 5.0 and later.
  if (Subtarget->getTargetTriple().isiOS() &&
      !Subtarget->getTargetTriple().isOSVersionLT(5, 0)) {
    setLibcallName(RTLIB::SDIVREM_I32, "__divmodsi4");
    setLibcallName(RTLIB::UDIVREM_I32, "__udivmodsi4");
  }

  // The half <-> float conversion functions are always soft-float, but are
  // needed for some targets which use a hard-float calling convention by
  // default.
  if (Subtarget->isAAPCS_ABI()) {
    setLibcallCallingConv(RTLIB::FPROUND_F32_F16, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::FPROUND_F64_F16, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::FPEXT_F16_F32, CallingConv::ARM_AAPCS);
  } else {
    setLibcallCallingConv(RTLIB::FPROUND_F32_F16, CallingConv::ARM_APCS);
    setLibcallCallingConv(RTLIB::FPROUND_F64_F16, CallingConv::ARM_APCS);
    setLibcallCallingConv(RTLIB::FPEXT_F16_F32, CallingConv::ARM_APCS);
  }

  if (Subtarget->isThumb1Only())
    addRegisterClass(MVT::i32, &ARM::tGPRRegClass);
  else
    addRegisterClass(MVT::i32, &ARM::GPRRegClass);
  if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
      !Subtarget->isThumb1Only()) {
    addRegisterClass(MVT::f32, &ARM::SPRRegClass);
    addRegisterClass(MVT::f64, &ARM::DPRRegClass);
  }

  for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
       VT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++VT) {
    for (unsigned InnerVT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
         InnerVT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++InnerVT)
      setTruncStoreAction((MVT::SimpleValueType)VT,
                          (MVT::SimpleValueType)InnerVT, Expand);
    setLoadExtAction(ISD::SEXTLOAD, (MVT::SimpleValueType)VT, Expand);
    setLoadExtAction(ISD::ZEXTLOAD, (MVT::SimpleValueType)VT, Expand);
    setLoadExtAction(ISD::EXTLOAD, (MVT::SimpleValueType)VT, Expand);

    setOperationAction(ISD::MULHS, (MVT::SimpleValueType)VT, Expand);
    setOperationAction(ISD::SMUL_LOHI, (MVT::SimpleValueType)VT, Expand);
    setOperationAction(ISD::MULHU, (MVT::SimpleValueType)VT, Expand);
    setOperationAction(ISD::UMUL_LOHI, (MVT::SimpleValueType)VT, Expand);

    setOperationAction(ISD::BSWAP, (MVT::SimpleValueType)VT, Expand);
  }

  setOperationAction(ISD::ConstantFP, MVT::f32, Custom);
  setOperationAction(ISD::ConstantFP, MVT::f64, Custom);

  if (Subtarget->hasNEON()) {
    addDRTypeForNEON(MVT::v2f32);
    addDRTypeForNEON(MVT::v8i8);
    addDRTypeForNEON(MVT::v4i16);
    addDRTypeForNEON(MVT::v2i32);
    addDRTypeForNEON(MVT::v1i64);

    addQRTypeForNEON(MVT::v4f32);
    addQRTypeForNEON(MVT::v2f64);
    addQRTypeForNEON(MVT::v16i8);
    addQRTypeForNEON(MVT::v8i16);
    addQRTypeForNEON(MVT::v4i32);
    addQRTypeForNEON(MVT::v2i64);

    // v2f64 is legal so that QR subregs can be extracted as f64 elements, but
    // neither Neon nor VFP support any arithmetic operations on it.
    // The same with v4f32. But keep in mind that vadd, vsub, vmul are natively
    // supported for v4f32.
    setOperationAction(ISD::FADD, MVT::v2f64, Expand);
    setOperationAction(ISD::FSUB, MVT::v2f64, Expand);
    setOperationAction(ISD::FMUL, MVT::v2f64, Expand);
    // FIXME: Code duplication: FDIV and FREM are expanded always, see
    // ARMTargetLowering::addTypeForNEON method for details.
    setOperationAction(ISD::FDIV, MVT::v2f64, Expand);
    setOperationAction(ISD::FREM, MVT::v2f64, Expand);
    // FIXME: Create unittest.
    // In another words, find a way when "copysign" appears in DAG with vector
    // operands.
    setOperationAction(ISD::FCOPYSIGN, MVT::v2f64, Expand);
    // FIXME: Code duplication: SETCC has custom operation action, see
    // ARMTargetLowering::addTypeForNEON method for details.
    setOperationAction(ISD::SETCC, MVT::v2f64, Expand);
    // FIXME: Create unittest for FNEG and for FABS.
    setOperationAction(ISD::FNEG, MVT::v2f64, Expand);
    setOperationAction(ISD::FABS, MVT::v2f64, Expand);
    setOperationAction(ISD::FSQRT, MVT::v2f64, Expand);
    setOperationAction(ISD::FSIN, MVT::v2f64, Expand);
    setOperationAction(ISD::FCOS, MVT::v2f64, Expand);
    setOperationAction(ISD::FPOWI, MVT::v2f64, Expand);
    setOperationAction(ISD::FPOW, MVT::v2f64, Expand);
    setOperationAction(ISD::FLOG, MVT::v2f64, Expand);
    setOperationAction(ISD::FLOG2, MVT::v2f64, Expand);
    setOperationAction(ISD::FLOG10, MVT::v2f64, Expand);
    setOperationAction(ISD::FEXP, MVT::v2f64, Expand);
    setOperationAction(ISD::FEXP2, MVT::v2f64, Expand);
    // FIXME: Create unittest for FCEIL, FTRUNC, FRINT, FNEARBYINT, FFLOOR.
    setOperationAction(ISD::FCEIL, MVT::v2f64, Expand);
    setOperationAction(ISD::FTRUNC, MVT::v2f64, Expand);
    setOperationAction(ISD::FRINT, MVT::v2f64, Expand);
    setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Expand);
    setOperationAction(ISD::FFLOOR, MVT::v2f64, Expand);
    setOperationAction(ISD::FMA, MVT::v2f64, Expand);

    setOperationAction(ISD::FSQRT, MVT::v4f32, Expand);
    setOperationAction(ISD::FSIN, MVT::v4f32, Expand);
    setOperationAction(ISD::FCOS, MVT::v4f32, Expand);
    setOperationAction(ISD::FPOWI, MVT::v4f32, Expand);
    setOperationAction(ISD::FPOW, MVT::v4f32, Expand);
    setOperationAction(ISD::FLOG, MVT::v4f32, Expand);
    setOperationAction(ISD::FLOG2, MVT::v4f32, Expand);
    setOperationAction(ISD::FLOG10, MVT::v4f32, Expand);
    setOperationAction(ISD::FEXP, MVT::v4f32, Expand);
    setOperationAction(ISD::FEXP2, MVT::v4f32, Expand);
    setOperationAction(ISD::FCEIL, MVT::v4f32, Expand);
    setOperationAction(ISD::FTRUNC, MVT::v4f32, Expand);
    setOperationAction(ISD::FRINT, MVT::v4f32, Expand);
    setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Expand);
    setOperationAction(ISD::FFLOOR, MVT::v4f32, Expand);

    // Mark v2f32 intrinsics.
    setOperationAction(ISD::FSQRT, MVT::v2f32, Expand);
    setOperationAction(ISD::FSIN, MVT::v2f32, Expand);
    setOperationAction(ISD::FCOS, MVT::v2f32, Expand);
    setOperationAction(ISD::FPOWI, MVT::v2f32, Expand);
    setOperationAction(ISD::FPOW, MVT::v2f32, Expand);
    setOperationAction(ISD::FLOG, MVT::v2f32, Expand);
    setOperationAction(ISD::FLOG2, MVT::v2f32, Expand);
    setOperationAction(ISD::FLOG10, MVT::v2f32, Expand);
    setOperationAction(ISD::FEXP, MVT::v2f32, Expand);
    setOperationAction(ISD::FEXP2, MVT::v2f32, Expand);
    setOperationAction(ISD::FCEIL, MVT::v2f32, Expand);
    setOperationAction(ISD::FTRUNC, MVT::v2f32, Expand);
    setOperationAction(ISD::FRINT, MVT::v2f32, Expand);
    setOperationAction(ISD::FNEARBYINT, MVT::v2f32, Expand);
    setOperationAction(ISD::FFLOOR, MVT::v2f32, Expand);

    // Neon does not support some operations on v1i64 and v2i64 types.
    setOperationAction(ISD::MUL, MVT::v1i64, Expand);
    // Custom handling for some quad-vector types to detect VMULL.
    setOperationAction(ISD::MUL, MVT::v8i16, Custom);
    setOperationAction(ISD::MUL, MVT::v4i32, Custom);
    setOperationAction(ISD::MUL, MVT::v2i64, Custom);
    // Custom handling for some vector types to avoid expensive expansions
    setOperationAction(ISD::SDIV, MVT::v4i16, Custom);
    setOperationAction(ISD::SDIV, MVT::v8i8, Custom);
    setOperationAction(ISD::UDIV, MVT::v4i16, Custom);
    setOperationAction(ISD::UDIV, MVT::v8i8, Custom);
    setOperationAction(ISD::SETCC, MVT::v1i64, Expand);
    setOperationAction(ISD::SETCC, MVT::v2i64, Expand);
    // Neon does not have single instruction SINT_TO_FP and UINT_TO_FP with
    // a destination type that is wider than the source, and nor does
    // it have a FP_TO_[SU]INT instruction with a narrower destination than
    // source.
    setOperationAction(ISD::SINT_TO_FP, MVT::v4i16, Custom);
    setOperationAction(ISD::UINT_TO_FP, MVT::v4i16, Custom);
    setOperationAction(ISD::FP_TO_UINT, MVT::v4i16, Custom);
    setOperationAction(ISD::FP_TO_SINT, MVT::v4i16, Custom);

    setOperationAction(ISD::FP_ROUND,   MVT::v2f32, Expand);
    setOperationAction(ISD::FP_EXTEND,  MVT::v2f64, Expand);

    // NEON does not have single instruction CTPOP for vectors with element
    // types wider than 8-bits.  However, custom lowering can leverage the
    // v8i8/v16i8 vcnt instruction.
    setOperationAction(ISD::CTPOP,      MVT::v2i32, Custom);
    setOperationAction(ISD::CTPOP,      MVT::v4i32, Custom);
    setOperationAction(ISD::CTPOP,      MVT::v4i16, Custom);
    setOperationAction(ISD::CTPOP,      MVT::v8i16, Custom);

    // NEON only has FMA instructions as of VFP4.
    if (!Subtarget->hasVFP4()) {
      setOperationAction(ISD::FMA, MVT::v2f32, Expand);
      setOperationAction(ISD::FMA, MVT::v4f32, Expand);
    }

    setTargetDAGCombine(ISD::INTRINSIC_VOID);
    setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
    setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
    setTargetDAGCombine(ISD::SHL);
    setTargetDAGCombine(ISD::SRL);
    setTargetDAGCombine(ISD::SRA);
    setTargetDAGCombine(ISD::SIGN_EXTEND);
    setTargetDAGCombine(ISD::ZERO_EXTEND);
    setTargetDAGCombine(ISD::ANY_EXTEND);
    setTargetDAGCombine(ISD::SELECT_CC);
    setTargetDAGCombine(ISD::BUILD_VECTOR);
    setTargetDAGCombine(ISD::VECTOR_SHUFFLE);
    setTargetDAGCombine(ISD::INSERT_VECTOR_ELT);
    setTargetDAGCombine(ISD::STORE);
    setTargetDAGCombine(ISD::FP_TO_SINT);
    setTargetDAGCombine(ISD::FP_TO_UINT);
    setTargetDAGCombine(ISD::FDIV);

    // It is legal to extload from v4i8 to v4i16 or v4i32.
    MVT Tys[6] = {MVT::v8i8, MVT::v4i8, MVT::v2i8,
                  MVT::v4i16, MVT::v2i16,
                  MVT::v2i32};
    for (unsigned i = 0; i < 6; ++i) {
      setLoadExtAction(ISD::EXTLOAD, Tys[i], Legal);
      setLoadExtAction(ISD::ZEXTLOAD, Tys[i], Legal);
      setLoadExtAction(ISD::SEXTLOAD, Tys[i], Legal);
    }
  }

  // ARM and Thumb2 support UMLAL/SMLAL.
  if (!Subtarget->isThumb1Only())
    setTargetDAGCombine(ISD::ADDC);

  if (Subtarget->isFPOnlySP()) {
    // When targetting a floating-point unit with only single-precision
    // operations, f64 is legal for the few double-precision instructions which
    // are present However, no double-precision operations other than moves,
    // loads and stores are provided by the hardware.
    setOperationAction(ISD::FADD,       MVT::f64, Expand);
    setOperationAction(ISD::FSUB,       MVT::f64, Expand);
    setOperationAction(ISD::FMUL,       MVT::f64, Expand);
    setOperationAction(ISD::FMA,        MVT::f64, Expand);
    setOperationAction(ISD::FDIV,       MVT::f64, Expand);
    setOperationAction(ISD::FREM,       MVT::f64, Expand);
    setOperationAction(ISD::FCOPYSIGN,  MVT::f64, Expand);
    setOperationAction(ISD::FGETSIGN,   MVT::f64, Expand);
    setOperationAction(ISD::FNEG,       MVT::f64, Expand);
    setOperationAction(ISD::FABS,       MVT::f64, Expand);
    setOperationAction(ISD::FSQRT,      MVT::f64, Expand);
    setOperationAction(ISD::FSIN,       MVT::f64, Expand);
    setOperationAction(ISD::FCOS,       MVT::f64, Expand);
    setOperationAction(ISD::FPOWI,      MVT::f64, Expand);
    setOperationAction(ISD::FPOW,       MVT::f64, Expand);
    setOperationAction(ISD::FLOG,       MVT::f64, Expand);
    setOperationAction(ISD::FLOG2,      MVT::f64, Expand);
    setOperationAction(ISD::FLOG10,     MVT::f64, Expand);
    setOperationAction(ISD::FEXP,       MVT::f64, Expand);
    setOperationAction(ISD::FEXP2,      MVT::f64, Expand);
    setOperationAction(ISD::FCEIL,      MVT::f64, Expand);
    setOperationAction(ISD::FTRUNC,     MVT::f64, Expand);
    setOperationAction(ISD::FRINT,      MVT::f64, Expand);
    setOperationAction(ISD::FNEARBYINT, MVT::f64, Expand);
    setOperationAction(ISD::FFLOOR,     MVT::f64, Expand);
    setOperationAction(ISD::FP_ROUND,   MVT::f32, Custom);
    setOperationAction(ISD::FP_EXTEND,  MVT::f64, Custom);
  }

  computeRegisterProperties();

  // ARM does not have floating-point extending loads.
  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);

  // ... or truncating stores
  setTruncStoreAction(MVT::f64, MVT::f32, Expand);
  setTruncStoreAction(MVT::f32, MVT::f16, Expand);
  setTruncStoreAction(MVT::f64, MVT::f16, Expand);

  // ARM does not have i1 sign extending load.
  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);

  // ARM supports all 4 flavors of integer indexed load / store.
  if (!Subtarget->isThumb1Only()) {
    for (unsigned im = (unsigned)ISD::PRE_INC;
         im != (unsigned)ISD::LAST_INDEXED_MODE; ++im) {
      setIndexedLoadAction(im,  MVT::i1,  Legal);
      setIndexedLoadAction(im,  MVT::i8,  Legal);
      setIndexedLoadAction(im,  MVT::i16, Legal);
      setIndexedLoadAction(im,  MVT::i32, Legal);
      setIndexedStoreAction(im, MVT::i1,  Legal);
      setIndexedStoreAction(im, MVT::i8,  Legal);
      setIndexedStoreAction(im, MVT::i16, Legal);
      setIndexedStoreAction(im, MVT::i32, Legal);
    }
  }

  setOperationAction(ISD::SADDO, MVT::i32, Custom);
  setOperationAction(ISD::UADDO, MVT::i32, Custom);
  setOperationAction(ISD::SSUBO, MVT::i32, Custom);
  setOperationAction(ISD::USUBO, MVT::i32, Custom);

  // i64 operation support.
  setOperationAction(ISD::MUL,     MVT::i64, Expand);
  setOperationAction(ISD::MULHU,   MVT::i32, Expand);
  if (Subtarget->isThumb1Only()) {
    setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
    setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
  }
  if (Subtarget->isThumb1Only() || !Subtarget->hasV6Ops()
      || (Subtarget->isThumb2() && !Subtarget->hasThumb2DSP()))
    setOperationAction(ISD::MULHS, MVT::i32, Expand);

  setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
  setOperationAction(ISD::SRA_PARTS, MVT::i32, Custom);
  setOperationAction(ISD::SRL_PARTS, MVT::i32, Custom);
  setOperationAction(ISD::SRL,       MVT::i64, Custom);
  setOperationAction(ISD::SRA,       MVT::i64, Custom);

  if (!Subtarget->isThumb1Only()) {
    // FIXME: We should do this for Thumb1 as well.
    setOperationAction(ISD::ADDC,    MVT::i32, Custom);
    setOperationAction(ISD::ADDE,    MVT::i32, Custom);
    setOperationAction(ISD::SUBC,    MVT::i32, Custom);
    setOperationAction(ISD::SUBE,    MVT::i32, Custom);
  }

  // ARM does not have ROTL.
  setOperationAction(ISD::ROTL,  MVT::i32, Expand);
  setOperationAction(ISD::CTTZ,  MVT::i32, Custom);
  setOperationAction(ISD::CTPOP, MVT::i32, Expand);
  if (!Subtarget->hasV5TOps() || Subtarget->isThumb1Only())
    setOperationAction(ISD::CTLZ, MVT::i32, Expand);

  // These just redirect to CTTZ and CTLZ on ARM.
  setOperationAction(ISD::CTTZ_ZERO_UNDEF  , MVT::i32  , Expand);
  setOperationAction(ISD::CTLZ_ZERO_UNDEF  , MVT::i32  , Expand);

  setOperationAction(ISD::READCYCLECOUNTER, MVT::i64, Custom);

  // Only ARMv6 has BSWAP.
  if (!Subtarget->hasV6Ops())
    setOperationAction(ISD::BSWAP, MVT::i32, Expand);

  if (!(Subtarget->hasDivide() && Subtarget->isThumb2()) &&
      !(Subtarget->hasDivideInARMMode() && !Subtarget->isThumb())) {
    // These are expanded into libcalls if the cpu doesn't have HW divider.
    setOperationAction(ISD::SDIV,  MVT::i32, Expand);
    setOperationAction(ISD::UDIV,  MVT::i32, Expand);
  }

  // FIXME: Also set divmod for SREM on EABI
  setOperationAction(ISD::SREM,  MVT::i32, Expand);
  setOperationAction(ISD::UREM,  MVT::i32, Expand);
  // Register based DivRem for AEABI (RTABI 4.2)
  if (Subtarget->isTargetAEABI()) {
    setLibcallName(RTLIB::SDIVREM_I8,  "__aeabi_idivmod");
    setLibcallName(RTLIB::SDIVREM_I16, "__aeabi_idivmod");
    setLibcallName(RTLIB::SDIVREM_I32, "__aeabi_idivmod");
    setLibcallName(RTLIB::SDIVREM_I64, "__aeabi_ldivmod");
    setLibcallName(RTLIB::UDIVREM_I8,  "__aeabi_uidivmod");
    setLibcallName(RTLIB::UDIVREM_I16, "__aeabi_uidivmod");
    setLibcallName(RTLIB::UDIVREM_I32, "__aeabi_uidivmod");
    setLibcallName(RTLIB::UDIVREM_I64, "__aeabi_uldivmod");

    setLibcallCallingConv(RTLIB::SDIVREM_I8, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::SDIVREM_I16, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::SDIVREM_I32, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::SDIVREM_I64, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::UDIVREM_I8, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::UDIVREM_I16, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::UDIVREM_I32, CallingConv::ARM_AAPCS);
    setLibcallCallingConv(RTLIB::UDIVREM_I64, CallingConv::ARM_AAPCS);

    setOperationAction(ISD::SDIVREM, MVT::i32, Custom);
    setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
  } else {
    setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
    setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
  }

  setOperationAction(ISD::GlobalAddress, MVT::i32,   Custom);
  setOperationAction(ISD::ConstantPool,  MVT::i32,   Custom);
  setOperationAction(ISD::GLOBAL_OFFSET_TABLE, MVT::i32, Custom);
  setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
  setOperationAction(ISD::BlockAddress, MVT::i32, Custom);

  setOperationAction(ISD::TRAP, MVT::Other, Legal);

  // Use the default implementation.
  setOperationAction(ISD::VASTART,            MVT::Other, Custom);
  setOperationAction(ISD::VAARG,              MVT::Other, Expand);
  setOperationAction(ISD::VACOPY,             MVT::Other, Expand);
  setOperationAction(ISD::VAEND,              MVT::Other, Expand);
  setOperationAction(ISD::STACKSAVE,          MVT::Other, Expand);
  setOperationAction(ISD::STACKRESTORE,       MVT::Other, Expand);

  if (!Subtarget->isTargetMachO()) {
    // Non-MachO platforms may return values in these registers via the
    // personality function.
    setExceptionPointerRegister(ARM::R0);
    setExceptionSelectorRegister(ARM::R1);
  }

  if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
    setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
  else
    setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);

  // ARMv6 Thumb1 (except for CPUs that support dmb / dsb) and earlier use
  // the default expansion. If we are targeting a single threaded system,
  // then set them all for expand so we can lower them later into their
  // non-atomic form.
  if (TM.Options.ThreadModel == ThreadModel::Single)
    setOperationAction(ISD::ATOMIC_FENCE,   MVT::Other, Expand);
  else if (Subtarget->hasAnyDataBarrier() && !Subtarget->isThumb1Only()) {
    // ATOMIC_FENCE needs custom lowering; the others should have been expanded
    // to ldrex/strex loops already.
    setOperationAction(ISD::ATOMIC_FENCE,     MVT::Other, Custom);

    // On v8, we have particularly efficient implementations of atomic fences
    // if they can be combined with nearby atomic loads and stores.
    if (!Subtarget->hasV8Ops()) {
      // Automatically insert fences (dmb ish) around ATOMIC_SWAP etc.
      setInsertFencesForAtomic(true);
    }
  } else {
    // If there's anything we can use as a barrier, go through custom lowering
    // for ATOMIC_FENCE.
    setOperationAction(ISD::ATOMIC_FENCE,   MVT::Other,
                       Subtarget->hasAnyDataBarrier() ? Custom : Expand);

    // Set them all for expansion, which will force libcalls.
    setOperationAction(ISD::ATOMIC_CMP_SWAP,  MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_SWAP,      MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_ADD,  MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_SUB,  MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_AND,  MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_OR,   MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_XOR,  MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i32, Expand);
    setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i32, Expand);
    // Mark ATOMIC_LOAD and ATOMIC_STORE custom so we can handle the
    // Unordered/Monotonic case.
    setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Custom);
    setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Custom);
  }

  setOperationAction(ISD::PREFETCH,         MVT::Other, Custom);

  // Requires SXTB/SXTH, available on v6 and up in both ARM and Thumb modes.
  if (!Subtarget->hasV6Ops()) {
    setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand);
    setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8,  Expand);
  }
  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);

  if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
      !Subtarget->isThumb1Only()) {
    // Turn f64->i64 into VMOVRRD, i64 -> f64 to VMOVDRR
    // iff target supports vfp2.
    setOperationAction(ISD::BITCAST, MVT::i64, Custom);
    setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
  }

  // We want to custom lower some of our intrinsics.
  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
  if (Subtarget->isTargetDarwin()) {
    setOperationAction(ISD::EH_SJLJ_SETJMP, MVT::i32, Custom);
    setOperationAction(ISD::EH_SJLJ_LONGJMP, MVT::Other, Custom);
    setLibcallName(RTLIB::UNWIND_RESUME, "_Unwind_SjLj_Resume");
  }

  setOperationAction(ISD::SETCC,     MVT::i32, Expand);
  setOperationAction(ISD::SETCC,     MVT::f32, Expand);
  setOperationAction(ISD::SETCC,     MVT::f64, Expand);
  setOperationAction(ISD::SELECT,    MVT::i32, Custom);
  setOperationAction(ISD::SELECT,    MVT::f32, Custom);
  setOperationAction(ISD::SELECT,    MVT::f64, Custom);
  setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
  setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
  setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);

  setOperationAction(ISD::BRCOND,    MVT::Other, Expand);
  setOperationAction(ISD::BR_CC,     MVT::i32,   Custom);
  setOperationAction(ISD::BR_CC,     MVT::f32,   Custom);
  setOperationAction(ISD::BR_CC,     MVT::f64,   Custom);
  setOperationAction(ISD::BR_JT,     MVT::Other, Custom);

  // We don't support sin/cos/fmod/copysign/pow
  setOperationAction(ISD::FSIN,      MVT::f64, Expand);
  setOperationAction(ISD::FSIN,      MVT::f32, Expand);
  setOperationAction(ISD::FCOS,      MVT::f32, Expand);
  setOperationAction(ISD::FCOS,      MVT::f64, Expand);
  setOperationAction(ISD::FSINCOS,   MVT::f64, Expand);
  setOperationAction(ISD::FSINCOS,   MVT::f32, Expand);
  setOperationAction(ISD::FREM,      MVT::f64, Expand);
  setOperationAction(ISD::FREM,      MVT::f32, Expand);
  if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
      !Subtarget->isThumb1Only()) {
    setOperationAction(ISD::FCOPYSIGN, MVT::f64, Custom);
    setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom);
  }
  setOperationAction(ISD::FPOW,      MVT::f64, Expand);
  setOperationAction(ISD::FPOW,      MVT::f32, Expand);

  if (!Subtarget->hasVFP4()) {
    setOperationAction(ISD::FMA, MVT::f64, Expand);
    setOperationAction(ISD::FMA, MVT::f32, Expand);
  }

  // Various VFP goodness
  if (!TM.Options.UseSoftFloat && !Subtarget->isThumb1Only()) {
    // int <-> fp are custom expanded into bit_convert + ARMISD ops.
    if (Subtarget->hasVFP2()) {
      setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
      setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
      setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
      setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
    }

    // FP-ARMv8 adds f64 <-> f16 conversion. Before that it should be expanded.
    if (!Subtarget->hasFPARMv8() || Subtarget->isFPOnlySP()) {
      setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
      setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
    }

    // fp16 is a special v7 extension that adds f16 <-> f32 conversions.
    if (!Subtarget->hasFP16()) {
      setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
      setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
    }
  }

  // Combine sin / cos into one node or libcall if possible.
  if (Subtarget->hasSinCos()) {
    setLibcallName(RTLIB::SINCOS_F32, "sincosf");
    setLibcallName(RTLIB::SINCOS_F64, "sincos");
    if (Subtarget->getTargetTriple().isiOS()) {
      // For iOS, we don't want to the normal expansion of a libcall to
      // sincos. We want to issue a libcall to __sincos_stret.
      setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
      setOperationAction(ISD::FSINCOS, MVT::f32, Custom);
    }
  }

  // FP-ARMv8 implements a lot of rounding-like FP operations.
  if (Subtarget->hasFPARMv8()) {
    setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
    setOperationAction(ISD::FCEIL, MVT::f32, Legal);
    setOperationAction(ISD::FROUND, MVT::f32, Legal);
    setOperationAction(ISD::FTRUNC, MVT::f32, Legal);
    setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
    setOperationAction(ISD::FRINT, MVT::f32, Legal);
    if (!Subtarget->isFPOnlySP()) {
      setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
      setOperationAction(ISD::FCEIL, MVT::f64, Legal);
      setOperationAction(ISD::FROUND, MVT::f64, Legal);
      setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
      setOperationAction(ISD::FNEARBYINT, MVT::f64, Legal);
      setOperationAction(ISD::FRINT, MVT::f64, Legal);
    }
  }
  // We have target-specific dag combine patterns for the following nodes:
  // ARMISD::VMOVRRD  - No need to call setTargetDAGCombine
  setTargetDAGCombine(ISD::ADD);
  setTargetDAGCombine(ISD::SUB);
  setTargetDAGCombine(ISD::MUL);
  setTargetDAGCombine(ISD::AND);
  setTargetDAGCombine(ISD::OR);
  setTargetDAGCombine(ISD::XOR);

  if (Subtarget->hasV6Ops())
    setTargetDAGCombine(ISD::SRL);

  setStackPointerRegisterToSaveRestore(ARM::SP);

  if (TM.Options.UseSoftFloat || Subtarget->isThumb1Only() ||
      !Subtarget->hasVFP2())
    setSchedulingPreference(Sched::RegPressure);
  else
    setSchedulingPreference(Sched::Hybrid);

  //// temporary - rewrite interface to use type
  MaxStoresPerMemset = 8;
  MaxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
  MaxStoresPerMemcpy = 4; // For @llvm.memcpy -> sequence of stores
  MaxStoresPerMemcpyOptSize = Subtarget->isTargetDarwin() ? 4 : 2;
  MaxStoresPerMemmove = 4; // For @llvm.memmove -> sequence of stores
  MaxStoresPerMemmoveOptSize = Subtarget->isTargetDarwin() ? 4 : 2;

  // On ARM arguments smaller than 4 bytes are extended, so all arguments
  // are at least 4 bytes aligned.
  setMinStackArgumentAlignment(4);

  // Prefer likely predicted branches to selects on out-of-order cores.
  PredictableSelectIsExpensive = Subtarget->isLikeA9();

  setMinFunctionAlignment(Subtarget->isThumb() ? 1 : 2);
}

// FIXME: It might make sense to define the representative register class as the
// nearest super-register that has a non-null superset. For example, DPR_VFP2 is
// a super-register of SPR, and DPR is a superset if DPR_VFP2. Consequently,
// SPR's representative would be DPR_VFP2. This should work well if register
// pressure tracking were modified such that a register use would increment the
// pressure of the register class's representative and all of it's super
// classes' representatives transitively. We have not implemented this because
// of the difficulty prior to coalescing of modeling operand register classes
// due to the common occurrence of cross class copies and subregister insertions
// and extractions.
std::pair<const TargetRegisterClass*, uint8_t>
ARMTargetLowering::findRepresentativeClass(MVT VT) const{
  const TargetRegisterClass *RRC = nullptr;
  uint8_t Cost = 1;
  switch (VT.SimpleTy) {
  default:
    return TargetLowering::findRepresentativeClass(VT);
  // Use DPR as representative register class for all floating point
  // and vector types. Since there are 32 SPR registers and 32 DPR registers so
  // the cost is 1 for both f32 and f64.
  case MVT::f32: case MVT::f64: case MVT::v8i8: case MVT::v4i16:
  case MVT::v2i32: case MVT::v1i64: case MVT::v2f32:
    RRC = &ARM::DPRRegClass;
    // When NEON is used for SP, only half of the register file is available
    // because operations that define both SP and DP results will be constrained
    // to the VFP2 class (D0-D15). We currently model this constraint prior to
    // coalescing by double-counting the SP regs. See the FIXME above.
    if (Subtarget->useNEONForSinglePrecisionFP())
      Cost = 2;
    break;
  case MVT::v16i8: case MVT::v8i16: case MVT::v4i32: case MVT::v2i64:
  case MVT::v4f32: case MVT::v2f64:
    RRC = &ARM::DPRRegClass;
    Cost = 2;
    break;
  case MVT::v4i64:
    RRC = &ARM::DPRRegClass;
    Cost = 4;
    break;
  case MVT::v8i64:
    RRC = &ARM::DPRRegClass;
    Cost = 8;
    break;
  }
  return std::make_pair(RRC, Cost);
}

const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const {
  switch (Opcode) {
  default: return nullptr;
  case ARMISD::Wrapper:       return "ARMISD::Wrapper";
  case ARMISD::WrapperPIC:    return "ARMISD::WrapperPIC";
  case ARMISD::WrapperJT:     return "ARMISD::WrapperJT";
  case ARMISD::CALL:          return "ARMISD::CALL";
  case ARMISD::CALL_PRED:     return "ARMISD::CALL_PRED";
  case ARMISD::CALL_NOLINK:   return "ARMISD::CALL_NOLINK";
  case ARMISD::tCALL:         return "ARMISD::tCALL";
  case ARMISD::BRCOND:        return "ARMISD::BRCOND";
  case ARMISD::BR_JT:         return "ARMISD::BR_JT";
  case ARMISD::BR2_JT:        return "ARMISD::BR2_JT";
  case ARMISD::RET_FLAG:      return "ARMISD::RET_FLAG";
  case ARMISD::INTRET_FLAG:   return "ARMISD::INTRET_FLAG";
  case ARMISD::PIC_ADD:       return "ARMISD::PIC_ADD";
  case ARMISD::CMP:           return "ARMISD::CMP";
  case ARMISD::CMN:           return "ARMISD::CMN";
  case ARMISD::CMPZ:          return "ARMISD::CMPZ";
  case ARMISD::CMPFP:         return "ARMISD::CMPFP";
  case ARMISD::CMPFPw0:       return "ARMISD::CMPFPw0";
  case ARMISD::BCC_i64:       return "ARMISD::BCC_i64";
  case ARMISD::FMSTAT:        return "ARMISD::FMSTAT";

  case ARMISD::CMOV:          return "ARMISD::CMOV";

  case ARMISD::RBIT:          return "ARMISD::RBIT";

  case ARMISD::FTOSI:         return "ARMISD::FTOSI";
  case ARMISD::FTOUI:         return "ARMISD::FTOUI";
  case ARMISD::SITOF:         return "ARMISD::SITOF";
  case ARMISD::UITOF:         return "ARMISD::UITOF";

  case ARMISD::SRL_FLAG:      return "ARMISD::SRL_FLAG";
  case ARMISD::SRA_FLAG:      return "ARMISD::SRA_FLAG";
  case ARMISD::RRX:           return "ARMISD::RRX";

  case ARMISD::ADDC:          return "ARMISD::ADDC";
  case ARMISD::ADDE:          return "ARMISD::ADDE";
  case ARMISD::SUBC:          return "ARMISD::SUBC";
  case ARMISD::SUBE:          return "ARMISD::SUBE";

  case ARMISD::VMOVRRD:       return "ARMISD::VMOVRRD";
  case ARMISD::VMOVDRR:       return "ARMISD::VMOVDRR";

  case ARMISD::EH_SJLJ_SETJMP: return "ARMISD::EH_SJLJ_SETJMP";
  case ARMISD::EH_SJLJ_LONGJMP:return "ARMISD::EH_SJLJ_LONGJMP";

  case ARMISD::TC_RETURN:     return "ARMISD::TC_RETURN";

  case ARMISD::THREAD_POINTER:return "ARMISD::THREAD_POINTER";

  case ARMISD::DYN_ALLOC:     return "ARMISD::DYN_ALLOC";

  case ARMISD::MEMBARRIER_MCR: return "ARMISD::MEMBARRIER_MCR";

  case ARMISD::PRELOAD:       return "ARMISD::PRELOAD";

  case ARMISD::WIN__CHKSTK:   return "ARMISD:::WIN__CHKSTK";

  case ARMISD::VCEQ:          return "ARMISD::VCEQ";
  case ARMISD::VCEQZ:         return "ARMISD::VCEQZ";
  case ARMISD::VCGE:          return "ARMISD::VCGE";
  case ARMISD::VCGEZ:         return "ARMISD::VCGEZ";
  case ARMISD::VCLEZ:         return "ARMISD::VCLEZ";
  case ARMISD::VCGEU:         return "ARMISD::VCGEU";
  case ARMISD::VCGT:          return "ARMISD::VCGT";
  case ARMISD::VCGTZ:         return "ARMISD::VCGTZ";
  case ARMISD::VCLTZ:         return "ARMISD::VCLTZ";
  case ARMISD::VCGTU:         return "ARMISD::VCGTU";
  case ARMISD::VTST:          return "ARMISD::VTST";

  case ARMISD::VSHL:          return "ARMISD::VSHL";
  case ARMISD::VSHRs:         return "ARMISD::VSHRs";
  case ARMISD::VSHRu:         return "ARMISD::VSHRu";
  case ARMISD::VRSHRs:        return "ARMISD::VRSHRs";
  case ARMISD::VRSHRu:        return "ARMISD::VRSHRu";
  case ARMISD::VRSHRN:        return "ARMISD::VRSHRN";
  case ARMISD::VQSHLs:        return "ARMISD::VQSHLs";
  case ARMISD::VQSHLu:        return "ARMISD::VQSHLu";
  case ARMISD::VQSHLsu:       return "ARMISD::VQSHLsu";
  case ARMISD::VQSHRNs:       return "ARMISD::VQSHRNs";
  case ARMISD::VQSHRNu:       return "ARMISD::VQSHRNu";
  case ARMISD::VQSHRNsu:      return "ARMISD::VQSHRNsu";
  case ARMISD::VQRSHRNs:      return "ARMISD::VQRSHRNs";
  case ARMISD::VQRSHRNu:      return "ARMISD::VQRSHRNu";
  case ARMISD::VQRSHRNsu:     return "ARMISD::VQRSHRNsu";
  case ARMISD::VGETLANEu:     return "ARMISD::VGETLANEu";
  case ARMISD::VGETLANEs:     return "ARMISD::VGETLANEs";
  case ARMISD::VMOVIMM:       return "ARMISD::VMOVIMM";
  case ARMISD::VMVNIMM:       return "ARMISD::VMVNIMM";
  case ARMISD::VMOVFPIMM:     return "ARMISD::VMOVFPIMM";
  case ARMISD::VDUP:          return "ARMISD::VDUP";
  case ARMISD::VDUPLANE:      return "ARMISD::VDUPLANE";
  case ARMISD::VEXT:          return "ARMISD::VEXT";
  case ARMISD::VREV64:        return "ARMISD::VREV64";
  case ARMISD::VREV32:        return "ARMISD::VREV32";
  case ARMISD::VREV16:        return "ARMISD::VREV16";
  case ARMISD::VZIP:          return "ARMISD::VZIP";
  case ARMISD::VUZP:          return "ARMISD::VUZP";
  case ARMISD::VTRN:          return "ARMISD::VTRN";
  case ARMISD::VTBL1:         return "ARMISD::VTBL1";
  case ARMISD::VTBL2:         return "ARMISD::VTBL2";
  case ARMISD::VMULLs:        return "ARMISD::VMULLs";
  case ARMISD::VMULLu:        return "ARMISD::VMULLu";
  case ARMISD::UMLAL:         return "ARMISD::UMLAL";
  case ARMISD::SMLAL:         return "ARMISD::SMLAL";
  case ARMISD::BUILD_VECTOR:  return "ARMISD::BUILD_VECTOR";
  case ARMISD::FMAX:          return "ARMISD::FMAX";
  case ARMISD::FMIN:          return "ARMISD::FMIN";
  case ARMISD::VMAXNM:        return "ARMISD::VMAX";
  case ARMISD::VMINNM:        return "ARMISD::VMIN";
  case ARMISD::BFI:           return "ARMISD::BFI";
  case ARMISD::VORRIMM:       return "ARMISD::VORRIMM";
  case ARMISD::VBICIMM:       return "ARMISD::VBICIMM";
  case ARMISD::VBSL:          return "ARMISD::VBSL";
  case ARMISD::VLD2DUP:       return "ARMISD::VLD2DUP";
  case ARMISD::VLD3DUP:       return "ARMISD::VLD3DUP";
  case ARMISD::VLD4DUP:       return "ARMISD::VLD4DUP";
  case ARMISD::VLD1_UPD:      return "ARMISD::VLD1_UPD";
  case ARMISD::VLD2_UPD:      return "ARMISD::VLD2_UPD";
  case ARMISD::VLD3_UPD:      return "ARMISD::VLD3_UPD";
  case ARMISD::VLD4_UPD:      return "ARMISD::VLD4_UPD";
  case ARMISD::VLD2LN_UPD:    return "ARMISD::VLD2LN_UPD";
  case ARMISD::VLD3LN_UPD:    return "ARMISD::VLD3LN_UPD";
  case ARMISD::VLD4LN_UPD:    return "ARMISD::VLD4LN_UPD";
  case ARMISD::VLD2DUP_UPD:   return "ARMISD::VLD2DUP_UPD";
  case ARMISD::VLD3DUP_UPD:   return "ARMISD::VLD3DUP_UPD";
  case ARMISD::VLD4DUP_UPD:   return "ARMISD::VLD4DUP_UPD";
  case ARMISD::VST1_UPD:      return "ARMISD::VST1_UPD";
  case ARMISD::VST2_UPD:      return "ARMISD::VST2_UPD";
  case ARMISD::VST3_UPD:      return "ARMISD::VST3_UPD";
  case ARMISD::VST4_UPD:      return "ARMISD::VST4_UPD";
  case ARMISD::VST2LN_UPD:    return "ARMISD::VST2LN_UPD";
  case ARMISD::VST3LN_UPD:    return "ARMISD::VST3LN_UPD";
  case ARMISD::VST4LN_UPD:    return "ARMISD::VST4LN_UPD";
  }
}

EVT ARMTargetLowering::getSetCCResultType(LLVMContext &, EVT VT) const {
  if (!VT.isVector()) return getPointerTy();
  return VT.changeVectorElementTypeToInteger();
}

/// getRegClassFor - Return the register class that should be used for the
/// specified value type.
const TargetRegisterClass *ARMTargetLowering::getRegClassFor(MVT VT) const {
  // Map v4i64 to QQ registers but do not make the type legal. Similarly map
  // v8i64 to QQQQ registers. v4i64 and v8i64 are only used for REG_SEQUENCE to
  // load / store 4 to 8 consecutive D registers.
  if (Subtarget->hasNEON()) {
    if (VT == MVT::v4i64)
      return &ARM::QQPRRegClass;
    if (VT == MVT::v8i64)
      return &ARM::QQQQPRRegClass;
  }
  return TargetLowering::getRegClassFor(VT);
}

// Create a fast isel object.
FastISel *
ARMTargetLowering::createFastISel(FunctionLoweringInfo &funcInfo,
                                  const TargetLibraryInfo *libInfo) const {
  return ARM::createFastISel(funcInfo, libInfo);
}

/// getMaximalGlobalOffset - Returns the maximal possible offset which can
/// be used for loads / stores from the global.
unsigned ARMTargetLowering::getMaximalGlobalOffset() const {
  return (Subtarget->isThumb1Only() ? 127 : 4095);
}

Sched::Preference ARMTargetLowering::getSchedulingPreference(SDNode *N) const {
  unsigned NumVals = N->getNumValues();
  if (!NumVals)
    return Sched::RegPressure;

  for (unsigned i = 0; i != NumVals; ++i) {
    EVT VT = N->getValueType(i);
    if (VT == MVT::Glue || VT == MVT::Other)
      continue;
    if (VT.isFloatingPoint() || VT.isVector())
      return Sched::ILP;
  }

  if (!N->isMachineOpcode())
    return Sched::RegPressure;

  // Load are scheduled for latency even if there instruction itinerary
  // is not available.
  const TargetInstrInfo *TII =
      getTargetMachine().getSubtargetImpl()->getInstrInfo();
  const MCInstrDesc &MCID = TII->get(N->getMachineOpcode());

  if (MCID.getNumDefs() == 0)
    return Sched::RegPressure;
  if (!Itins->isEmpty() &&
      Itins->getOperandCycle(MCID.getSchedClass(), 0) > 2)
    return Sched::ILP;

  return Sched::RegPressure;
}

//===----------------------------------------------------------------------===//
// Lowering Code
//===----------------------------------------------------------------------===//

/// IntCCToARMCC - Convert a DAG integer condition code to an ARM CC
static ARMCC::CondCodes IntCCToARMCC(ISD::CondCode CC) {
  switch (CC) {
  default: llvm_unreachable("Unknown condition code!");
  case ISD::SETNE:  return ARMCC::NE;
  case ISD::SETEQ:  return ARMCC::EQ;
  case ISD::SETGT:  return ARMCC::GT;
  case ISD::SETGE:  return ARMCC::GE;
  case ISD::SETLT:  return ARMCC::LT;
  case ISD::SETLE:  return ARMCC::LE;
  case ISD::SETUGT: return ARMCC::HI;
  case ISD::SETUGE: return ARMCC::HS;
  case ISD::SETULT: return ARMCC::LO;
  case ISD::SETULE: return ARMCC::LS;
  }
}

/// FPCCToARMCC - Convert a DAG fp condition code to an ARM CC.
static void FPCCToARMCC(ISD::CondCode CC, ARMCC::CondCodes &CondCode,
                        ARMCC::CondCodes &CondCode2) {
  CondCode2 = ARMCC::AL;
  switch (CC) {
  default: llvm_unreachable("Unknown FP condition!");
  case ISD::SETEQ:
  case ISD::SETOEQ: CondCode = ARMCC::EQ; break;
  case ISD::SETGT:
  case ISD::SETOGT: CondCode = ARMCC::GT; break;
  case ISD::SETGE:
  case ISD::SETOGE: CondCode = ARMCC::GE; break;
  case ISD::SETOLT: CondCode = ARMCC::MI; break;
  case ISD::SETOLE: CondCode = ARMCC::LS; break;
  case ISD::SETONE: CondCode = ARMCC::MI; CondCode2 = ARMCC::GT; break;
  case ISD::SETO:   CondCode = ARMCC::VC; break;
  case ISD::SETUO:  CondCode = ARMCC::VS; break;
  case ISD::SETUEQ: CondCode = ARMCC::EQ; CondCode2 = ARMCC::VS; break;
  case ISD::SETUGT: CondCode = ARMCC::HI; break;
  case ISD::SETUGE: CondCode = ARMCC::PL; break;
  case ISD::SETLT:
  case ISD::SETULT: CondCode = ARMCC::LT; break;
  case ISD::SETLE:
  case ISD::SETULE: CondCode = ARMCC::LE; break;
  case ISD::SETNE:
  case ISD::SETUNE: CondCode = ARMCC::NE; break;
  }
}

//===----------------------------------------------------------------------===//
//                      Calling Convention Implementation
//===----------------------------------------------------------------------===//

#include "ARMGenCallingConv.inc"

/// getEffectiveCallingConv - Get the effective calling convention, taking into
/// account presence of floating point hardware and calling convention
/// limitations, such as support for variadic functions.
CallingConv::ID
ARMTargetLowering::getEffectiveCallingConv(CallingConv::ID CC,
                                           bool isVarArg) const {
  switch (CC) {
  default:
    llvm_unreachable("Unsupported calling convention");
  case CallingConv::ARM_AAPCS:
  case CallingConv::ARM_APCS:
  case CallingConv::GHC:
    return CC;
  case CallingConv::ARM_AAPCS_VFP:
    return isVarArg ? CallingConv::ARM_AAPCS : CallingConv::ARM_AAPCS_VFP;
  case CallingConv::C:
    if (!Subtarget->isAAPCS_ABI())
      return CallingConv::ARM_APCS;
    else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() &&
             getTargetMachine().Options.FloatABIType == FloatABI::Hard &&
             !isVarArg)
      return CallingConv::ARM_AAPCS_VFP;
    else
      return CallingConv::ARM_AAPCS;
  case CallingConv::Fast:
    if (!Subtarget->isAAPCS_ABI()) {
      if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
        return CallingConv::Fast;
      return CallingConv::ARM_APCS;
    } else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
      return CallingConv::ARM_AAPCS_VFP;
    else
      return CallingConv::ARM_AAPCS;
  }
}

/// CCAssignFnForNode - Selects the correct CCAssignFn for the given
/// CallingConvention.
CCAssignFn *ARMTargetLowering::CCAssignFnForNode(CallingConv::ID CC,
                                                 bool Return,
                                                 bool isVarArg) const {
  switch (getEffectiveCallingConv(CC, isVarArg)) {
  default:
    llvm_unreachable("Unsupported calling convention");
  case CallingConv::ARM_APCS:
    return (Return ? RetCC_ARM_APCS : CC_ARM_APCS);
  case CallingConv::ARM_AAPCS:
    return (Return ? RetCC_ARM_AAPCS : CC_ARM_AAPCS);
  case CallingConv::ARM_AAPCS_VFP:
    return (Return ? RetCC_ARM_AAPCS_VFP : CC_ARM_AAPCS_VFP);
  case CallingConv::Fast:
    return (Return ? RetFastCC_ARM_APCS : FastCC_ARM_APCS);
  case CallingConv::GHC:
    return (Return ? RetCC_ARM_APCS : CC_ARM_APCS_GHC);
  }
}

/// LowerCallResult - Lower the result values of a call into the
/// appropriate copies out of appropriate physical registers.
SDValue
ARMTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
                                   CallingConv::ID CallConv, bool isVarArg,
                                   const SmallVectorImpl<ISD::InputArg> &Ins,
                                   SDLoc dl, SelectionDAG &DAG,
                                   SmallVectorImpl<SDValue> &InVals,
                                   bool isThisReturn, SDValue ThisVal) const {

  // Assign locations to each value returned by this call.
  SmallVector<CCValAssign, 16> RVLocs;
  ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
                    *DAG.getContext(), Call);
  CCInfo.AnalyzeCallResult(Ins,
                           CCAssignFnForNode(CallConv, /* Return*/ true,
                                             isVarArg));

  // Copy all of the result registers out of their specified physreg.
  for (unsigned i = 0; i != RVLocs.size(); ++i) {
    CCValAssign VA = RVLocs[i];

    // Pass 'this' value directly from the argument to return value, to avoid
    // reg unit interference
    if (i == 0 && isThisReturn) {
      assert(!VA.needsCustom() && VA.getLocVT() == MVT::i32 &&
             "unexpected return calling convention register assignment");
      InVals.push_back(ThisVal);
      continue;
    }

    SDValue Val;
    if (VA.needsCustom()) {
      // Handle f64 or half of a v2f64.
      SDValue Lo = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32,
                                      InFlag);
      Chain = Lo.getValue(1);
      InFlag = Lo.getValue(2);
      VA = RVLocs[++i]; // skip ahead to next loc
      SDValue Hi = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32,
                                      InFlag);
      Chain = Hi.getValue(1);
      InFlag = Hi.getValue(2);
      if (!Subtarget->isLittle())
        std::swap (Lo, Hi);
      Val = DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi);

      if (VA.getLocVT() == MVT::v2f64) {
        SDValue Vec = DAG.getNode(ISD::UNDEF, dl, MVT::v2f64);
        Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Vec, Val,
                          DAG.getConstant(0, MVT::i32));

        VA = RVLocs[++i]; // skip ahead to next loc
        Lo = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32, InFlag);
        Chain = Lo.getValue(1);
        InFlag = Lo.getValue(2);
        VA = RVLocs[++i]; // skip ahead to next loc
        Hi = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32, InFlag);
        Chain = Hi.getValue(1);
        InFlag = Hi.getValue(2);
        if (!Subtarget->isLittle())
          std::swap (Lo, Hi);
        Val = DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi);
        Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Vec, Val,
                          DAG.getConstant(1, MVT::i32));
      }
    } else {
      Val = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), VA.getLocVT(),
                               InFlag);
      Chain = Val.getValue(1);
      InFlag = Val.getValue(2);
    }

    switch (VA.getLocInfo()) {
    default: llvm_unreachable("Unknown loc info!");
    case CCValAssign::Full: break;
    case CCValAssign::BCvt:
      Val = DAG.getNode(ISD::BITCAST, dl, VA.getValVT(), Val);
      break;
    }

    InVals.push_back(Val);
  }

  return Chain;
}

/// LowerMemOpCallTo - Store the argument to the stack.
SDValue
ARMTargetLowering::LowerMemOpCallTo(SDValue Chain,
                                    SDValue StackPtr, SDValue Arg,
                                    SDLoc dl, SelectionDAG &DAG,
                                    const CCValAssign &VA,
                                    ISD::ArgFlagsTy Flags) const {
  unsigned LocMemOffset = VA.getLocMemOffset();
  SDValue PtrOff = DAG.getIntPtrConstant(LocMemOffset);
  PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, PtrOff);
  return DAG.getStore(Chain, dl, Arg, PtrOff,
                      MachinePointerInfo::getStack(LocMemOffset),
                      false, false, 0);
}

void ARMTargetLowering::PassF64ArgInRegs(SDLoc dl, SelectionDAG &DAG,
                                         SDValue Chain, SDValue &Arg,
                                         RegsToPassVector &RegsToPass,
                                         CCValAssign &VA, CCValAssign &NextVA,
                                         SDValue &StackPtr,
                                         SmallVectorImpl<SDValue> &MemOpChains,
                                         ISD::ArgFlagsTy Flags) const {

  SDValue fmrrd = DAG.getNode(ARMISD::VMOVRRD, dl,
                              DAG.getVTList(MVT::i32, MVT::i32), Arg);
  unsigned id = Subtarget->isLittle() ? 0 : 1;
  RegsToPass.push_back(std::make_pair(VA.getLocReg(), fmrrd.getValue(id)));

  if (NextVA.isRegLoc())
    RegsToPass.push_back(std::make_pair(NextVA.getLocReg(), fmrrd.getValue(1-id)));
  else {
    assert(NextVA.isMemLoc());
    if (!StackPtr.getNode())
      StackPtr = DAG.getCopyFromReg(Chain, dl, ARM::SP, getPointerTy());

    MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, fmrrd.getValue(1-id),
                                           dl, DAG, NextVA,
                                           Flags));
  }
}

/// LowerCall - Lowering a call into a callseq_start <-
/// ARMISD:CALL <- callseq_end chain. Also add input and output parameter
/// nodes.
SDValue
ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                             SmallVectorImpl<SDValue> &InVals) const {
  SelectionDAG &DAG                     = CLI.DAG;
  SDLoc &dl                          = CLI.DL;
  SmallVectorImpl<ISD::OutputArg> &Outs = CLI.Outs;
  SmallVectorImpl<SDValue> &OutVals     = CLI.OutVals;
  SmallVectorImpl<ISD::InputArg> &Ins   = CLI.Ins;
  SDValue Chain                         = CLI.Chain;
  SDValue Callee                        = CLI.Callee;
  bool &isTailCall                      = CLI.IsTailCall;
  CallingConv::ID CallConv              = CLI.CallConv;
  bool doesNotRet                       = CLI.DoesNotReturn;
  bool isVarArg                         = CLI.IsVarArg;

  MachineFunction &MF = DAG.getMachineFunction();
  bool isStructRet    = (Outs.empty()) ? false : Outs[0].Flags.isSRet();
  bool isThisReturn   = false;
  bool isSibCall      = false;

  // Disable tail calls if they're not supported.
  if (!Subtarget->supportsTailCall() || MF.getTarget().Options.DisableTailCalls)
    isTailCall = false;

  if (isTailCall) {
    // Check if it's really possible to do a tail call.
    isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv,
                    isVarArg, isStructRet, MF.getFunction()->hasStructRetAttr(),
                                                   Outs, OutVals, Ins, DAG);
    if (!isTailCall && CLI.CS && CLI.CS->isMustTailCall())
      report_fatal_error("failed to perform tail call elimination on a call "
                         "site marked musttail");
    // We don't support GuaranteedTailCallOpt for ARM, only automatically
    // detected sibcalls.
    if (isTailCall) {
      ++NumTailCalls;
      isSibCall = true;
    }
  }

  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                    *DAG.getContext(), Call);
  CCInfo.AnalyzeCallOperands(Outs,
                             CCAssignFnForNode(CallConv, /* Return*/ false,
                                               isVarArg));

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

  // For tail calls, memory operands are available in our caller's stack.
  if (isSibCall)
    NumBytes = 0;

  // Adjust the stack pointer for the new arguments...
  // These operations are automatically eliminated by the prolog/epilog pass
  if (!isSibCall)
    Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true),
                                 dl);

  SDValue StackPtr = DAG.getCopyFromReg(Chain, dl, ARM::SP, getPointerTy());

  RegsToPassVector RegsToPass;
  SmallVector<SDValue, 8> MemOpChains;

  // Walk the register/memloc assignments, inserting copies/loads.  In the case
  // of tail call optimization, arguments are handled later.
  for (unsigned i = 0, realArgIdx = 0, e = ArgLocs.size();
       i != e;
       ++i, ++realArgIdx) {
    CCValAssign &VA = ArgLocs[i];
    SDValue Arg = OutVals[realArgIdx];
    ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;
    bool isByVal = Flags.isByVal();

    // Promote the value if needed.
    switch (VA.getLocInfo()) {
    default: llvm_unreachable("Unknown loc info!");
    case CCValAssign::Full: break;
    case CCValAssign::SExt:
      Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
      break;
    case CCValAssign::ZExt:
      Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
      break;
    case CCValAssign::AExt:
      Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
      break;
    case CCValAssign::BCvt:
      Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
      break;
    }

    // f64 and v2f64 might be passed in i32 pairs and must be split into pieces
    if (VA.needsCustom()) {
      if (VA.getLocVT() == MVT::v2f64) {
        SDValue Op0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
                                  DAG.getConstant(0, MVT::i32));
        SDValue Op1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
                                  DAG.getConstant(1, MVT::i32));

        PassF64ArgInRegs(dl, DAG, Chain, Op0, RegsToPass,
                         VA, ArgLocs[++i], StackPtr, MemOpChains, Flags);

        VA = ArgLocs[++i]; // skip ahead to next loc
        if (VA.isRegLoc()) {
          PassF64ArgInRegs(dl, DAG, Chain, Op1, RegsToPass,
                           VA, ArgLocs[++i], StackPtr, MemOpChains, Flags);
        } else {
          assert(VA.isMemLoc());

          MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Op1,
                                                 dl, DAG, VA, Flags));
        }
      } else {
        PassF64ArgInRegs(dl, DAG, Chain, Arg, RegsToPass, VA, ArgLocs[++i],
                         StackPtr, MemOpChains, Flags);
      }
    } else if (VA.isRegLoc()) {
      if (realArgIdx == 0 && Flags.isReturned() && Outs[0].VT == MVT::i32) {
        assert(VA.getLocVT() == MVT::i32 &&
               "unexpected calling convention register assignment");
        assert(!Ins.empty() && Ins[0].VT == MVT::i32 &&
               "unexpected use of 'returned'");
        isThisReturn = true;
      }
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
    } else if (isByVal) {
      assert(VA.isMemLoc());
      unsigned offset = 0;

      // True if this byval aggregate will be split between registers
      // and memory.
      unsigned ByValArgsCount = CCInfo.getInRegsParamsCount();
      unsigned CurByValIdx = CCInfo.getInRegsParamsProcessed();

      if (CurByValIdx < ByValArgsCount) {

        unsigned RegBegin, RegEnd;
        CCInfo.getInRegsParamInfo(CurByValIdx, RegBegin, RegEnd);

        EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
        unsigned int i, j;
        for (i = 0, j = RegBegin; j < RegEnd; i++, j++) {
          SDValue Const = DAG.getConstant(4*i, MVT::i32);
          SDValue AddArg = DAG.getNode(ISD::ADD, dl, PtrVT, Arg, Const);
          SDValue Load = DAG.getLoad(PtrVT, dl, Chain, AddArg,
                                     MachinePointerInfo(),
                                     false, false, false,
                                     DAG.InferPtrAlignment(AddArg));
          MemOpChains.push_back(Load.getValue(1));
          RegsToPass.push_back(std::make_pair(j, Load));
        }

        // If parameter size outsides register area, "offset" value
        // helps us to calculate stack slot for remained part properly.
        offset = RegEnd - RegBegin;

        CCInfo.nextInRegsParam();
      }

      if (Flags.getByValSize() > 4*offset) {
        unsigned LocMemOffset = VA.getLocMemOffset();
        SDValue StkPtrOff = DAG.getIntPtrConstant(LocMemOffset);
        SDValue Dst = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr,
                                  StkPtrOff);
        SDValue SrcOffset = DAG.getIntPtrConstant(4*offset);
        SDValue Src = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg, SrcOffset);
        SDValue SizeNode = DAG.getConstant(Flags.getByValSize() - 4*offset,
                                           MVT::i32);
        SDValue AlignNode = DAG.getConstant(Flags.getByValAlign(), MVT::i32);

        SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
        SDValue Ops[] = { Chain, Dst, Src, SizeNode, AlignNode};
        MemOpChains.push_back(DAG.getNode(ARMISD::COPY_STRUCT_BYVAL, dl, VTs,
                                          Ops));
      }
    } else if (!isSibCall) {
      assert(VA.isMemLoc());

      MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Arg,
                                             dl, DAG, VA, Flags));
    }
  }

  if (!MemOpChains.empty())
    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);

  // Build a sequence of copy-to-reg nodes chained together with token chain
  // and flag operands which copy the outgoing args into the appropriate regs.
  SDValue InFlag;
  // Tail call byval lowering might overwrite argument registers so in case of
  // tail call optimization the copies to registers are lowered later.
  if (!isTailCall)
    for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
      Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
                               RegsToPass[i].second, InFlag);
      InFlag = Chain.getValue(1);
    }

  // For tail calls lower the arguments to the 'real' stack slot.
  if (isTailCall) {
    // Force all the incoming stack arguments to be loaded from the stack
    // before any new outgoing arguments are stored to the stack, because the
    // outgoing stack slots may alias the incoming argument stack slots, and
    // the alias isn't otherwise explicit. This is slightly more conservative
    // than necessary, because it means that each store effectively depends
    // on every argument instead of just those arguments it would clobber.

    // Do not flag preceding copytoreg stuff together with the following stuff.
    InFlag = SDValue();
    for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
      Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
                               RegsToPass[i].second, InFlag);
      InFlag = Chain.getValue(1);
    }
    InFlag = SDValue();
  }

  // If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
  // direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
  // node so that legalize doesn't hack it.
  bool isDirect = false;
  bool isARMFunc = false;
  bool isLocalARMFunc = false;
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();

  if (EnableARMLongCalls) {
    assert((Subtarget->isTargetWindows() ||
            getTargetMachine().getRelocationModel() == Reloc::Static) &&
           "long-calls with non-static relocation model!");
    // Handle a global address or an external symbol. If it's not one of
    // those, the target's already in a register, so we don't need to do
    // anything extra.
    if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
      const GlobalValue *GV = G->getGlobal();
      // Create a constant pool entry for the callee address
      unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
      ARMConstantPoolValue *CPV =
        ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue, 0);

      // Get the address of the callee into a register
      SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
      CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
      Callee = DAG.getLoad(getPointerTy(), dl,
                           DAG.getEntryNode(), CPAddr,
                           MachinePointerInfo::getConstantPool(),
                           false, false, false, 0);
    } else if (ExternalSymbolSDNode *S=dyn_cast<ExternalSymbolSDNode>(Callee)) {
      const char *Sym = S->getSymbol();

      // Create a constant pool entry for the callee address
      unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
      ARMConstantPoolValue *CPV =
        ARMConstantPoolSymbol::Create(*DAG.getContext(), Sym,
                                      ARMPCLabelIndex, 0);
      // Get the address of the callee into a register
      SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
      CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
      Callee = DAG.getLoad(getPointerTy(), dl,
                           DAG.getEntryNode(), CPAddr,
                           MachinePointerInfo::getConstantPool(),
                           false, false, false, 0);
    }
  } else if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
    const GlobalValue *GV = G->getGlobal();
    isDirect = true;
    bool isExt = GV->isDeclaration() || GV->isWeakForLinker();
    bool isStub = (isExt && Subtarget->isTargetMachO()) &&
                   getTargetMachine().getRelocationModel() != Reloc::Static;
    isARMFunc = !Subtarget->isThumb() || (isStub && !Subtarget->isMClass());
    // ARM call to a local ARM function is predicable.
    isLocalARMFunc = !Subtarget->isThumb() && (!isExt || !ARMInterworking);
    // tBX takes a register source operand.
    if (isStub && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
      assert(Subtarget->isTargetMachO() && "WrapperPIC use on non-MachO?");
      Callee = DAG.getNode(ARMISD::WrapperPIC, dl, getPointerTy(),
                           DAG.getTargetGlobalAddress(GV, dl, getPointerTy(),
                                                      0, ARMII::MO_NONLAZY));
      Callee = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(), Callee,
                           MachinePointerInfo::getGOT(), false, false, true, 0);
    } else if (Subtarget->isTargetCOFF()) {
      assert(Subtarget->isTargetWindows() &&
             "Windows is the only supported COFF target");
      unsigned TargetFlags = GV->hasDLLImportStorageClass()
                                 ? ARMII::MO_DLLIMPORT
                                 : ARMII::MO_NO_FLAG;
      Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), /*Offset=*/0,
                                          TargetFlags);
      if (GV->hasDLLImportStorageClass())
        Callee = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(),
                             DAG.getNode(ARMISD::Wrapper, dl, getPointerTy(),
                                         Callee), MachinePointerInfo::getGOT(),
                             false, false, false, 0);
    } else {
      // On ELF targets for PIC code, direct calls should go through the PLT
      unsigned OpFlags = 0;
      if (Subtarget->isTargetELF() &&
          getTargetMachine().getRelocationModel() == Reloc::PIC_)
        OpFlags = ARMII::MO_PLT;
      Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), 0, OpFlags);
    }
  } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
    isDirect = true;
    bool isStub = Subtarget->isTargetMachO() &&
                  getTargetMachine().getRelocationModel() != Reloc::Static;
    isARMFunc = !Subtarget->isThumb() || (isStub && !Subtarget->isMClass());
    // tBX takes a register source operand.
    const char *Sym = S->getSymbol();
    if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
      unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
      ARMConstantPoolValue *CPV =
        ARMConstantPoolSymbol::Create(*DAG.getContext(), Sym,
                                      ARMPCLabelIndex, 4);
      SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
      CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
      Callee = DAG.getLoad(getPointerTy(), dl,
                           DAG.getEntryNode(), CPAddr,
                           MachinePointerInfo::getConstantPool(),
                           false, false, false, 0);
      SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
      Callee = DAG.getNode(ARMISD::PIC_ADD, dl,
                           getPointerTy(), Callee, PICLabel);
    } else {
      unsigned OpFlags = 0;
      // On ELF targets for PIC code, direct calls should go through the PLT
      if (Subtarget->isTargetELF() &&
                  getTargetMachine().getRelocationModel() == Reloc::PIC_)
        OpFlags = ARMII::MO_PLT;
      Callee = DAG.getTargetExternalSymbol(Sym, getPointerTy(), OpFlags);
    }
  }

  // FIXME: handle tail calls differently.
  unsigned CallOpc;
  bool HasMinSizeAttr = MF.getFunction()->getAttributes().hasAttribute(
      AttributeSet::FunctionIndex, Attribute::MinSize);
  if (Subtarget->isThumb()) {
    if ((!isDirect || isARMFunc) && !Subtarget->hasV5TOps())
      CallOpc = ARMISD::CALL_NOLINK;
    else
      CallOpc = isARMFunc ? ARMISD::CALL : ARMISD::tCALL;
  } else {
    if (!isDirect && !Subtarget->hasV5TOps())
      CallOpc = ARMISD::CALL_NOLINK;
    else if (doesNotRet && isDirect && Subtarget->hasRAS() &&
               // Emit regular call when code size is the priority
               !HasMinSizeAttr)
      // "mov lr, pc; b _foo" to avoid confusing the RSP
      CallOpc = ARMISD::CALL_NOLINK;
    else
      CallOpc = isLocalARMFunc ? ARMISD::CALL_PRED : ARMISD::CALL;
  }

  std::vector<SDValue> Ops;
  Ops.push_back(Chain);
  Ops.push_back(Callee);

  // Add argument registers to the end of the list so that they are known live
  // into the call.
  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
    Ops.push_back(DAG.getRegister(RegsToPass[i].first,
                                  RegsToPass[i].second.getValueType()));

  // Add a register mask operand representing the call-preserved registers.
  if (!isTailCall) {
    const uint32_t *Mask;
    const TargetRegisterInfo *TRI =
        getTargetMachine().getSubtargetImpl()->getRegisterInfo();
    const ARMBaseRegisterInfo *ARI = static_cast<const ARMBaseRegisterInfo*>(TRI);
    if (isThisReturn) {
      // For 'this' returns, use the R0-preserving mask if applicable
      Mask = ARI->getThisReturnPreservedMask(CallConv);
      if (!Mask) {
        // Set isThisReturn to false if the calling convention is not one that
        // allows 'returned' to be modeled in this way, so LowerCallResult does
        // not try to pass 'this' straight through
        isThisReturn = false;
        Mask = ARI->getCallPreservedMask(CallConv);
      }
    } else
      Mask = ARI->getCallPreservedMask(CallConv);

    assert(Mask && "Missing call preserved mask for calling convention");
    Ops.push_back(DAG.getRegisterMask(Mask));
  }

  if (InFlag.getNode())
    Ops.push_back(InFlag);

  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
  if (isTailCall)
    return DAG.getNode(ARMISD::TC_RETURN, dl, NodeTys, Ops);

  // Returns a chain and a flag for retval copy to use.
  Chain = DAG.getNode(CallOpc, dl, NodeTys, Ops);
  InFlag = Chain.getValue(1);

  Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
                             DAG.getIntPtrConstant(0, true), InFlag, dl);
  if (!Ins.empty())
    InFlag = Chain.getValue(1);

  // Handle result values, copying them out of physregs into vregs that we
  // return.
  return LowerCallResult(Chain, InFlag, CallConv, isVarArg, Ins, dl, DAG,
                         InVals, isThisReturn,
                         isThisReturn ? OutVals[0] : SDValue());
}

/// HandleByVal - Every parameter *after* a byval parameter is passed
/// on the stack.  Remember the next parameter register to allocate,
/// and then confiscate the rest of the parameter registers to insure
/// this.
void
ARMTargetLowering::HandleByVal(
    CCState *State, unsigned &size, unsigned Align) const {
  unsigned reg = State->AllocateReg(GPRArgRegs, 4);
  assert((State->getCallOrPrologue() == Prologue ||
          State->getCallOrPrologue() == Call) &&
         "unhandled ParmContext");

  if ((ARM::R0 <= reg) && (reg <= ARM::R3)) {
    if (Subtarget->isAAPCS_ABI() && Align > 4) {
      unsigned AlignInRegs = Align / 4;
      unsigned Waste = (ARM::R4 - reg) % AlignInRegs;
      for (unsigned i = 0; i < Waste; ++i)
        reg = State->AllocateReg(GPRArgRegs, 4);
    }
    if (reg != 0) {
      unsigned excess = 4 * (ARM::R4 - reg);

      // Special case when NSAA != SP and parameter size greater than size of
      // all remained GPR regs. In that case we can't split parameter, we must
      // send it to stack. We also must set NCRN to R4, so waste all
      // remained registers.
      const unsigned NSAAOffset = State->getNextStackOffset();
      if (Subtarget->isAAPCS_ABI() && NSAAOffset != 0 && size > excess) {
        while (State->AllocateReg(GPRArgRegs, 4))
          ;
        return;
      }

      // First register for byval parameter is the first register that wasn't
      // allocated before this method call, so it would be "reg".
      // If parameter is small enough to be saved in range [reg, r4), then
      // the end (first after last) register would be reg + param-size-in-regs,
      // else parameter would be splitted between registers and stack,
      // end register would be r4 in this case.
      unsigned ByValRegBegin = reg;
      unsigned ByValRegEnd = (size < excess) ? reg + size/4 : (unsigned)ARM::R4;
      State->addInRegsParamInfo(ByValRegBegin, ByValRegEnd);
      // Note, first register is allocated in the beginning of function already,
      // allocate remained amount of registers we need.
      for (unsigned i = reg+1; i != ByValRegEnd; ++i)
        State->AllocateReg(GPRArgRegs, 4);
      // A byval parameter that is split between registers and memory needs its
      // size truncated here.
      // In the case where the entire structure fits in registers, we set the
      // size in memory to zero.
      if (size < excess)
        size = 0;
      else
        size -= excess;
    }
  }
}

/// MatchingStackOffset - Return true if the given stack call argument is
/// already available in the same position (relatively) of the caller's
/// incoming argument stack.
static
bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags,
                         MachineFrameInfo *MFI, const MachineRegisterInfo *MRI,
                         const TargetInstrInfo *TII) {
  unsigned Bytes = Arg.getValueType().getSizeInBits() / 8;
  int FI = INT_MAX;
  if (Arg.getOpcode() == ISD::CopyFromReg) {
    unsigned VR = cast<RegisterSDNode>(Arg.getOperand(1))->getReg();
    if (!TargetRegisterInfo::isVirtualRegister(VR))
      return false;
    MachineInstr *Def = MRI->getVRegDef(VR);
    if (!Def)
      return false;
    if (!Flags.isByVal()) {
      if (!TII->isLoadFromStackSlot(Def, FI))
        return false;
    } else {
      return false;
    }
  } else if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Arg)) {
    if (Flags.isByVal())
      // ByVal argument is passed in as a pointer but it's now being
      // dereferenced. e.g.
      // define @foo(%struct.X* %A) {
      //   tail call @bar(%struct.X* byval %A)
      // }
      return false;
    SDValue Ptr = Ld->getBasePtr();
    FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(Ptr);
    if (!FINode)
      return false;
    FI = FINode->getIndex();
  } else
    return false;

  assert(FI != INT_MAX);
  if (!MFI->isFixedObjectIndex(FI))
    return false;
  return Offset == MFI->getObjectOffset(FI) && Bytes == MFI->getObjectSize(FI);
}

/// IsEligibleForTailCallOptimization - Check whether the call is eligible
/// for tail call optimization. Targets which want to do tail call
/// optimization should implement this function.
bool
ARMTargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
                                                     CallingConv::ID CalleeCC,
                                                     bool isVarArg,
                                                     bool isCalleeStructRet,
                                                     bool isCallerStructRet,
                                    const SmallVectorImpl<ISD::OutputArg> &Outs,
                                    const SmallVectorImpl<SDValue> &OutVals,
                                    const SmallVectorImpl<ISD::InputArg> &Ins,
                                                     SelectionDAG& DAG) const {
  const Function *CallerF = DAG.getMachineFunction().getFunction();
  CallingConv::ID CallerCC = CallerF->getCallingConv();
  bool CCMatch = CallerCC == CalleeCC;

  // Look for obvious safe cases to perform tail call optimization that do not
  // require ABI changes. This is what gcc calls sibcall.

  // Do not sibcall optimize vararg calls unless the call site is not passing
  // any arguments.
  if (isVarArg && !Outs.empty())
    return false;

  // Exception-handling functions need a special set of instructions to indicate
  // a return to the hardware. Tail-calling another function would probably
  // break this.
  if (CallerF->hasFnAttribute("interrupt"))
    return false;

  // Also avoid sibcall optimization if either caller or callee uses struct
  // return semantics.
  if (isCalleeStructRet || isCallerStructRet)
    return false;

  // FIXME: Completely disable sibcall for Thumb1 since Thumb1RegisterInfo::
  // emitEpilogue is not ready for them. Thumb tail calls also use t2B, as
  // the Thumb1 16-bit unconditional branch doesn't have sufficient relocation
  // support in the assembler and linker to be used. This would need to be
  // fixed to fully support tail calls in Thumb1.
  //
  // Doing this is tricky, since the LDM/POP instruction on Thumb doesn't take
  // LR.  This means if we need to reload LR, it takes an extra instructions,
  // which outweighs the value of the tail call; but here we don't know yet
  // whether LR is going to be used.  Probably the right approach is to
  // generate the tail call here and turn it back into CALL/RET in
  // emitEpilogue if LR is used.

  // Thumb1 PIC calls to external symbols use BX, so they can be tail calls,
  // but we need to make sure there are enough registers; the only valid
  // registers are the 4 used for parameters.  We don't currently do this
  // case.
  if (Subtarget->isThumb1Only())
    return false;

  // Externally-defined functions with weak linkage should not be
  // tail-called on ARM when the OS does not support dynamic
  // pre-emption of symbols, as the AAELF spec requires normal calls
  // to undefined weak functions to be replaced with a NOP or jump to the
  // next instruction. The behaviour of branch instructions in this
  // situation (as used for tail calls) is implementation-defined, so we
  // cannot rely on the linker replacing the tail call with a return.
  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
    const GlobalValue *GV = G->getGlobal();
    if (GV->hasExternalWeakLinkage())
      return false;
  }

  // If the calling conventions do not match, then we'd better make sure the
  // results are returned in the same way as what the caller expects.
  if (!CCMatch) {
    SmallVector<CCValAssign, 16> RVLocs1;
    ARMCCState CCInfo1(CalleeCC, false, DAG.getMachineFunction(), RVLocs1,
                       *DAG.getContext(), Call);
    CCInfo1.AnalyzeCallResult(Ins, CCAssignFnForNode(CalleeCC, true, isVarArg));

    SmallVector<CCValAssign, 16> RVLocs2;
    ARMCCState CCInfo2(CallerCC, false, DAG.getMachineFunction(), RVLocs2,
                       *DAG.getContext(), Call);
    CCInfo2.AnalyzeCallResult(Ins, CCAssignFnForNode(CallerCC, true, isVarArg));

    if (RVLocs1.size() != RVLocs2.size())
      return false;
    for (unsigned i = 0, e = RVLocs1.size(); i != e; ++i) {
      if (RVLocs1[i].isRegLoc() != RVLocs2[i].isRegLoc())
        return false;
      if (RVLocs1[i].getLocInfo() != RVLocs2[i].getLocInfo())
        return false;
      if (RVLocs1[i].isRegLoc()) {
        if (RVLocs1[i].getLocReg() != RVLocs2[i].getLocReg())
          return false;
      } else {
        if (RVLocs1[i].getLocMemOffset() != RVLocs2[i].getLocMemOffset())
          return false;
      }
    }
  }

  // If Caller's vararg or byval argument has been split between registers and
  // stack, do not perform tail call, since part of the argument is in caller's
  // local frame.
  const ARMFunctionInfo *AFI_Caller = DAG.getMachineFunction().
                                      getInfo<ARMFunctionInfo>();
  if (AFI_Caller->getArgRegsSaveSize())
    return false;

  // If the callee takes no arguments then go on to check the results of the
  // call.
  if (!Outs.empty()) {
    // Check if stack adjustment is needed. For now, do not do this if any
    // argument is passed on the stack.
    SmallVector<CCValAssign, 16> ArgLocs;
    ARMCCState CCInfo(CalleeCC, isVarArg, DAG.getMachineFunction(), ArgLocs,
                      *DAG.getContext(), Call);
    CCInfo.AnalyzeCallOperands(Outs,
                               CCAssignFnForNode(CalleeCC, false, isVarArg));
    if (CCInfo.getNextStackOffset()) {
      MachineFunction &MF = DAG.getMachineFunction();

      // Check if the arguments are already laid out in the right way as
      // the caller's fixed stack objects.
      MachineFrameInfo *MFI = MF.getFrameInfo();
      const MachineRegisterInfo *MRI = &MF.getRegInfo();
      const TargetInstrInfo *TII =
          getTargetMachine().getSubtargetImpl()->getInstrInfo();
      for (unsigned i = 0, realArgIdx = 0, e = ArgLocs.size();
           i != e;
           ++i, ++realArgIdx) {
        CCValAssign &VA = ArgLocs[i];
        EVT RegVT = VA.getLocVT();
        SDValue Arg = OutVals[realArgIdx];
        ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;
        if (VA.getLocInfo() == CCValAssign::Indirect)
          return false;
        if (VA.needsCustom()) {
          // f64 and vector types are split into multiple registers or
          // register/stack-slot combinations.  The types will not match
          // the registers; give up on memory f64 refs until we figure
          // out what to do about this.
          if (!VA.isRegLoc())
            return false;
          if (!ArgLocs[++i].isRegLoc())
            return false;
          if (RegVT == MVT::v2f64) {
            if (!ArgLocs[++i].isRegLoc())
              return false;
            if (!ArgLocs[++i].isRegLoc())
              return false;
          }
        } else if (!VA.isRegLoc()) {
          if (!MatchingStackOffset(Arg, VA.getLocMemOffset(), Flags,
                                   MFI, MRI, TII))
            return false;
        }
      }
    }
  }

  return true;
}

bool
ARMTargetLowering::CanLowerReturn(CallingConv::ID CallConv,
                                  MachineFunction &MF, bool isVarArg,
                                  const SmallVectorImpl<ISD::OutputArg> &Outs,
                                  LLVMContext &Context) const {
  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCInfo(CallConv, isVarArg, MF, RVLocs, Context);
  return CCInfo.CheckReturn(Outs, CCAssignFnForNode(CallConv, /*Return=*/true,
                                                    isVarArg));
}

static SDValue LowerInterruptReturn(SmallVectorImpl<SDValue> &RetOps,
                                    SDLoc DL, SelectionDAG &DAG) {
  const MachineFunction &MF = DAG.getMachineFunction();
  const Function *F = MF.getFunction();

  StringRef IntKind = F->getFnAttribute("interrupt").getValueAsString();

  // See ARM ARM v7 B1.8.3. On exception entry LR is set to a possibly offset
  // version of the "preferred return address". These offsets affect the return
  // instruction if this is a return from PL1 without hypervisor extensions.
  //    IRQ/FIQ: +4     "subs pc, lr, #4"
  //    SWI:     0      "subs pc, lr, #0"
  //    ABORT:   +4     "subs pc, lr, #4"
  //    UNDEF:   +4/+2  "subs pc, lr, #0"
  // UNDEF varies depending on where the exception came from ARM or Thumb
  // mode. Alongside GCC, we throw our hands up in disgust and pretend it's 0.

  int64_t LROffset;
  if (IntKind == "" || IntKind == "IRQ" || IntKind == "FIQ" ||
      IntKind == "ABORT")
    LROffset = 4;
  else if (IntKind == "SWI" || IntKind == "UNDEF")
    LROffset = 0;
  else
    report_fatal_error("Unsupported interrupt attribute. If present, value "
                       "must be one of: IRQ, FIQ, SWI, ABORT or UNDEF");

  RetOps.insert(RetOps.begin() + 1, DAG.getConstant(LROffset, MVT::i32, false));

  return DAG.getNode(ARMISD::INTRET_FLAG, DL, MVT::Other, RetOps);
}

SDValue
ARMTargetLowering::LowerReturn(SDValue Chain,
                               CallingConv::ID CallConv, bool isVarArg,
                               const SmallVectorImpl<ISD::OutputArg> &Outs,
                               const SmallVectorImpl<SDValue> &OutVals,
                               SDLoc dl, SelectionDAG &DAG) const {

  // CCValAssign - represent the assignment of the return value to a location.
  SmallVector<CCValAssign, 16> RVLocs;

  // CCState - Info about the registers and stack slots.
  ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
                    *DAG.getContext(), Call);

  // Analyze outgoing return values.
  CCInfo.AnalyzeReturn(Outs, CCAssignFnForNode(CallConv, /* Return */ true,
                                               isVarArg));

  SDValue Flag;
  SmallVector<SDValue, 4> RetOps;
  RetOps.push_back(Chain); // Operand #0 = Chain (updated below)
  bool isLittleEndian = Subtarget->isLittle();

  MachineFunction &MF = DAG.getMachineFunction();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
  AFI->setReturnRegsCount(RVLocs.size());

  // Copy the result values into the output registers.
  for (unsigned i = 0, realRVLocIdx = 0;
       i != RVLocs.size();
       ++i, ++realRVLocIdx) {
    CCValAssign &VA = RVLocs[i];
    assert(VA.isRegLoc() && "Can only return in registers!");

    SDValue Arg = OutVals[realRVLocIdx];

    switch (VA.getLocInfo()) {
    default: llvm_unreachable("Unknown loc info!");
    case CCValAssign::Full: break;
    case CCValAssign::BCvt:
      Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
      break;
    }

    if (VA.needsCustom()) {
      if (VA.getLocVT() == MVT::v2f64) {
        // Extract the first half and return it in two registers.
        SDValue Half = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
                                   DAG.getConstant(0, MVT::i32));
        SDValue HalfGPRs = DAG.getNode(ARMISD::VMOVRRD, dl,
                                       DAG.getVTList(MVT::i32, MVT::i32), Half);

        Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
                                 HalfGPRs.getValue(isLittleEndian ? 0 : 1),
                                 Flag);
        Flag = Chain.getValue(1);
        RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
        VA = RVLocs[++i]; // skip ahead to next loc
        Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
                                 HalfGPRs.getValue(isLittleEndian ? 1 : 0),
                                 Flag);
        Flag = Chain.getValue(1);
        RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
        VA = RVLocs[++i]; // skip ahead to next loc

        // Extract the 2nd half and fall through to handle it as an f64 value.
        Arg = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
                          DAG.getConstant(1, MVT::i32));
      }
      // Legalize ret f64 -> ret 2 x i32.  We always have fmrrd if f64 is
      // available.
      SDValue fmrrd = DAG.getNode(ARMISD::VMOVRRD, dl,
                                  DAG.getVTList(MVT::i32, MVT::i32), Arg);
      Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
                               fmrrd.getValue(isLittleEndian ? 0 : 1),
                               Flag);
      Flag = Chain.getValue(1);
      RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
      VA = RVLocs[++i]; // skip ahead to next loc
      Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
                               fmrrd.getValue(isLittleEndian ? 1 : 0),
                               Flag);
    } else
      Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), Arg, Flag);

    // Guarantee that all emitted copies are
    // stuck together, avoiding something bad.
    Flag = Chain.getValue(1);
    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
  }

  // Update chain and glue.
  RetOps[0] = Chain;
  if (Flag.getNode())
    RetOps.push_back(Flag);

  // CPUs which aren't M-class use a special sequence to return from
  // exceptions (roughly, any instruction setting pc and cpsr simultaneously,
  // though we use "subs pc, lr, #N").
  //
  // M-class CPUs actually use a normal return sequence with a special
  // (hardware-provided) value in LR, so the normal code path works.
  if (DAG.getMachineFunction().getFunction()->hasFnAttribute("interrupt") &&
      !Subtarget->isMClass()) {
    if (Subtarget->isThumb1Only())
      report_fatal_error("interrupt attribute is not supported in Thumb1");
    return LowerInterruptReturn(RetOps, dl, DAG);
  }

  return DAG.getNode(ARMISD::RET_FLAG, dl, MVT::Other, RetOps);
}

bool ARMTargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
  if (N->getNumValues() != 1)
    return false;
  if (!N->hasNUsesOfValue(1, 0))
    return false;

  SDValue TCChain = Chain;
  SDNode *Copy = *N->use_begin();
  if (Copy->getOpcode() == ISD::CopyToReg) {
    // If the copy has a glue operand, we conservatively assume it isn't safe to
    // perform a tail call.
    if (Copy->getOperand(Copy->getNumOperands()-1).getValueType() == MVT::Glue)
      return false;
    TCChain = Copy->getOperand(0);
  } else if (Copy->getOpcode() == ARMISD::VMOVRRD) {
    SDNode *VMov = Copy;
    // f64 returned in a pair of GPRs.
    SmallPtrSet<SDNode*, 2> Copies;
    for (SDNode::use_iterator UI = VMov->use_begin(), UE = VMov->use_end();
         UI != UE; ++UI) {
      if (UI->getOpcode() != ISD::CopyToReg)
        return false;
      Copies.insert(*UI);
    }
    if (Copies.size() > 2)
      return false;

    for (SDNode::use_iterator UI = VMov->use_begin(), UE = VMov->use_end();
         UI != UE; ++UI) {
      SDValue UseChain = UI->getOperand(0);
      if (Copies.count(UseChain.getNode()))
        // Second CopyToReg
        Copy = *UI;
      else {
        // We are at the top of this chain.
        // If the copy has a glue operand, we conservatively assume it
        // isn't safe to perform a tail call.
        if (UI->getOperand(UI->getNumOperands()-1).getValueType() == MVT::Glue)
          return false;
        // First CopyToReg
        TCChain = UseChain;
      }
    }
  } else if (Copy->getOpcode() == ISD::BITCAST) {
    // f32 returned in a single GPR.
    if (!Copy->hasOneUse())
      return false;
    Copy = *Copy->use_begin();
    if (Copy->getOpcode() != ISD::CopyToReg || !Copy->hasNUsesOfValue(1, 0))
      return false;
    // If the copy has a glue operand, we conservatively assume it isn't safe to
    // perform a tail call.
    if (Copy->getOperand(Copy->getNumOperands()-1).getValueType() == MVT::Glue)
      return false;
    TCChain = Copy->getOperand(0);
  } else {
    return false;
  }

  bool HasRet = false;
  for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end();
       UI != UE; ++UI) {
    if (UI->getOpcode() != ARMISD::RET_FLAG &&
        UI->getOpcode() != ARMISD::INTRET_FLAG)
      return false;
    HasRet = true;
  }

  if (!HasRet)
    return false;

  Chain = TCChain;
  return true;
}

bool ARMTargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
  if (!Subtarget->supportsTailCall())
    return false;

  if (!CI->isTailCall() || getTargetMachine().Options.DisableTailCalls)
    return false;

  return !Subtarget->isThumb1Only();
}

// ConstantPool, JumpTable, GlobalAddress, and ExternalSymbol are lowered as
// their target counterpart wrapped in the ARMISD::Wrapper node. Suppose N is
// one of the above mentioned nodes. It has to be wrapped because otherwise
// Select(N) returns N. So the raw TargetGlobalAddress nodes, etc. can only
// be used to form addressing mode. These wrapped nodes will be selected
// into MOVi.
static SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) {
  EVT PtrVT = Op.getValueType();
  // FIXME there is no actual debug info here
  SDLoc dl(Op);
  ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
  SDValue Res;
  if (CP->isMachineConstantPoolEntry())
    Res = DAG.getTargetConstantPool(CP->getMachineCPVal(), PtrVT,
                                    CP->getAlignment());
  else
    Res = DAG.getTargetConstantPool(CP->getConstVal(), PtrVT,
                                    CP->getAlignment());
  return DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Res);
}

unsigned ARMTargetLowering::getJumpTableEncoding() const {
  return MachineJumpTableInfo::EK_Inline;
}

SDValue ARMTargetLowering::LowerBlockAddress(SDValue Op,
                                             SelectionDAG &DAG) const {
  MachineFunction &MF = DAG.getMachineFunction();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
  unsigned ARMPCLabelIndex = 0;
  SDLoc DL(Op);
  EVT PtrVT = getPointerTy();
  const BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
  Reloc::Model RelocM = getTargetMachine().getRelocationModel();
  SDValue CPAddr;
  if (RelocM == Reloc::Static) {
    CPAddr = DAG.getTargetConstantPool(BA, PtrVT, 4);
  } else {
    unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
    ARMPCLabelIndex = AFI->createPICLabelUId();
    ARMConstantPoolValue *CPV =
      ARMConstantPoolConstant::Create(BA, ARMPCLabelIndex,
                                      ARMCP::CPBlockAddress, PCAdj);
    CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
  }
  CPAddr = DAG.getNode(ARMISD::Wrapper, DL, PtrVT, CPAddr);
  SDValue Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), CPAddr,
                               MachinePointerInfo::getConstantPool(),
                               false, false, false, 0);
  if (RelocM == Reloc::Static)
    return Result;
  SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
  return DAG.getNode(ARMISD::PIC_ADD, DL, PtrVT, Result, PICLabel);
}

// Lower ISD::GlobalTLSAddress using the "general dynamic" model
SDValue
ARMTargetLowering::LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA,
                                                 SelectionDAG &DAG) const {
  SDLoc dl(GA);
  EVT PtrVT = getPointerTy();
  unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8;
  MachineFunction &MF = DAG.getMachineFunction();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
  unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
  ARMConstantPoolValue *CPV =
    ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex,
                                    ARMCP::CPValue, PCAdj, ARMCP::TLSGD, true);
  SDValue Argument = DAG.getTargetConstantPool(CPV, PtrVT, 4);
  Argument = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Argument);
  Argument = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Argument,
                         MachinePointerInfo::getConstantPool(),
                         false, false, false, 0);
  SDValue Chain = Argument.getValue(1);

  SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
  Argument = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Argument, PICLabel);

  // call __tls_get_addr.
  ArgListTy Args;
  ArgListEntry Entry;
  Entry.Node = Argument;
  Entry.Ty = (Type *) Type::getInt32Ty(*DAG.getContext());
  Args.push_back(Entry);

  // FIXME: is there useful debug info available here?
  TargetLowering::CallLoweringInfo CLI(DAG);
  CLI.setDebugLoc(dl).setChain(Chain)
    .setCallee(CallingConv::C, Type::getInt32Ty(*DAG.getContext()),
               DAG.getExternalSymbol("__tls_get_addr", PtrVT), std::move(Args),
               0);

  std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
  return CallResult.first;
}

// Lower ISD::GlobalTLSAddress using the "initial exec" or
// "local exec" model.
SDValue
ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA,
                                        SelectionDAG &DAG,
                                        TLSModel::Model model) const {
  const GlobalValue *GV = GA->getGlobal();
  SDLoc dl(GA);
  SDValue Offset;
  SDValue Chain = DAG.getEntryNode();
  EVT PtrVT = getPointerTy();
  // Get the Thread Pointer
  SDValue ThreadPointer = DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);

  if (model == TLSModel::InitialExec) {
    MachineFunction &MF = DAG.getMachineFunction();
    ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
    unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
    // Initial exec model.
    unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8;
    ARMConstantPoolValue *CPV =
      ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex,
                                      ARMCP::CPValue, PCAdj, ARMCP::GOTTPOFF,
                                      true);
    Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
    Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
    Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
                         MachinePointerInfo::getConstantPool(),
                         false, false, false, 0);
    Chain = Offset.getValue(1);

    SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
    Offset = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Offset, PICLabel);

    Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
                         MachinePointerInfo::getConstantPool(),
                         false, false, false, 0);
  } else {
    // local exec model
    assert(model == TLSModel::LocalExec);
    ARMConstantPoolValue *CPV =
      ARMConstantPoolConstant::Create(GV, ARMCP::TPOFF);
    Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
    Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
    Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
                         MachinePointerInfo::getConstantPool(),
                         false, false, false, 0);
  }

  // The address of the thread local variable is the add of the thread
  // pointer with the offset of the variable.
  return DAG.getNode(ISD::ADD, dl, PtrVT, ThreadPointer, Offset);
}

SDValue
ARMTargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const {
  // TODO: implement the "local dynamic" model
  assert(Subtarget->isTargetELF() &&
         "TLS not implemented for non-ELF targets");
  GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);

  TLSModel::Model model = getTargetMachine().getTLSModel(GA->getGlobal());

  switch (model) {
    case TLSModel::GeneralDynamic:
    case TLSModel::LocalDynamic:
      return LowerToTLSGeneralDynamicModel(GA, DAG);
    case TLSModel::InitialExec:
    case TLSModel::LocalExec:
      return LowerToTLSExecModels(GA, DAG, model);
  }
  llvm_unreachable("bogus TLS model");
}

SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op,
                                                 SelectionDAG &DAG) const {
  EVT PtrVT = getPointerTy();
  SDLoc dl(Op);
  const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
  if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
    bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
    ARMConstantPoolValue *CPV =
      ARMConstantPoolConstant::Create(GV,
                                      UseGOTOFF ? ARMCP::GOTOFF : ARMCP::GOT);
    SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
    CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
    SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(),
                                 CPAddr,
                                 MachinePointerInfo::getConstantPool(),
                                 false, false, false, 0);
    SDValue Chain = Result.getValue(1);
    SDValue GOT = DAG.getGLOBAL_OFFSET_TABLE(PtrVT);
    Result = DAG.getNode(ISD::ADD, dl, PtrVT, Result, GOT);
    if (!UseGOTOFF)
      Result = DAG.getLoad(PtrVT, dl, Chain, Result,
                           MachinePointerInfo::getGOT(),
                           false, false, false, 0);
    return Result;
  }

  // If we have T2 ops, we can materialize the address directly via movt/movw
  // pair. This is always cheaper.
  if (Subtarget->useMovt(DAG.getMachineFunction())) {
    ++NumMovwMovt;
    // FIXME: Once remat is capable of dealing with instructions with register
    // operands, expand this into two nodes.
    return DAG.getNode(ARMISD::Wrapper, dl, PtrVT,
                       DAG.getTargetGlobalAddress(GV, dl, PtrVT));
  } else {
    SDValue CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
    CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
    return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
                       MachinePointerInfo::getConstantPool(),
                       false, false, false, 0);
  }
}

SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op,
                                                    SelectionDAG &DAG) const {
  EVT PtrVT = getPointerTy();
  SDLoc dl(Op);
  const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
  Reloc::Model RelocM = getTargetMachine().getRelocationModel();

  if (Subtarget->useMovt(DAG.getMachineFunction()))
    ++NumMovwMovt;

  // FIXME: Once remat is capable of dealing with instructions with register
  // operands, expand this into multiple nodes
  unsigned Wrapper =
      RelocM == Reloc::PIC_ ? ARMISD::WrapperPIC : ARMISD::Wrapper;

  SDValue G = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, ARMII::MO_NONLAZY);
  SDValue Result = DAG.getNode(Wrapper, dl, PtrVT, G);

  if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
    Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Result,
                         MachinePointerInfo::getGOT(), false, false, false, 0);
  return Result;
}

SDValue ARMTargetLowering::LowerGlobalAddressWindows(SDValue Op,
                                                     SelectionDAG &DAG) const {
  assert(Subtarget->isTargetWindows() && "non-Windows COFF is not supported");
  assert(Subtarget->useMovt(DAG.getMachineFunction()) &&
         "Windows on ARM expects to use movw/movt");

  const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
  const ARMII::TOF TargetFlags =
    (GV->hasDLLImportStorageClass() ? ARMII::MO_DLLIMPORT : ARMII::MO_NO_FLAG);
  EVT PtrVT = getPointerTy();
  SDValue Result;
  SDLoc DL(Op);

  ++NumMovwMovt;

  // FIXME: Once remat is capable of dealing with instructions with register
  // operands, expand this into two nodes.
  Result = DAG.getNode(ARMISD::Wrapper, DL, PtrVT,
                       DAG.getTargetGlobalAddress(GV, DL, PtrVT, /*Offset=*/0,
                                                  TargetFlags));
  if (GV->hasDLLImportStorageClass())
    Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), Result,
                         MachinePointerInfo::getGOT(), false, false, false, 0);
  return Result;
}

SDValue ARMTargetLowering::LowerGLOBAL_OFFSET_TABLE(SDValue Op,
                                                    SelectionDAG &DAG) const {
  assert(Subtarget->isTargetELF() &&
         "GLOBAL OFFSET TABLE not implemented for non-ELF targets");
  MachineFunction &MF = DAG.getMachineFunction();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
  unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
  EVT PtrVT = getPointerTy();
  SDLoc dl(Op);
  unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
  ARMConstantPoolValue *CPV =
    ARMConstantPoolSymbol::Create(*DAG.getContext(), "_GLOBAL_OFFSET_TABLE_",
                                  ARMPCLabelIndex, PCAdj);
  SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
  CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
  SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
                               MachinePointerInfo::getConstantPool(),
                               false, false, false, 0);
  SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
  return DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
}

SDValue
ARMTargetLowering::LowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const {
  SDLoc dl(Op);
  SDValue Val = DAG.getConstant(0, MVT::i32);
  return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl,
                     DAG.getVTList(MVT::i32, MVT::Other), Op.getOperand(0),
                     Op.getOperand(1), Val);
}

SDValue
ARMTargetLowering::LowerEH_SJLJ_LONGJMP(SDValue Op, SelectionDAG &DAG) const {
  SDLoc dl(Op);
  return DAG.getNode(ARMISD::EH_SJLJ_LONGJMP, dl, MVT::Other, Op.getOperand(0),
                     Op.getOperand(1), DAG.getConstant(0, MVT::i32));
}

SDValue
ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
                                          const ARMSubtarget *Subtarget) const {
  unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
  SDLoc dl(Op);
  switch (IntNo) {
  default: return SDValue();    // Don't custom lower most intrinsics.
  case Intrinsic::arm_rbit: {
    assert(Op.getOperand(1).getValueType() == MVT::i32 &&
           "RBIT intrinsic must have i32 type!");
    return DAG.getNode(ARMISD::RBIT, dl, MVT::i32, Op.getOperand(1));
  }
  case Intrinsic::arm_thread_pointer: {
    EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
    return DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);
  }
  case Intrinsic::eh_sjlj_lsda: {
    MachineFunction &MF = DAG.getMachineFunction();
    ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
    unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
    EVT PtrVT = getPointerTy();
    Reloc::Model RelocM = getTargetMachine().getRelocationModel();
    SDValue CPAddr;
    unsigned PCAdj = (RelocM != Reloc::PIC_)
      ? 0 : (Subtarget->isThumb() ? 4 : 8);
    ARMConstantPoolValue *CPV =
      ARMConstantPoolConstant::Create(MF.getFunction(), ARMPCLabelIndex,
                                      ARMCP::CPLSDA, PCAdj);
    CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
    CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
    SDValue Result =
      DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
                  MachinePointerInfo::getConstantPool(),
                  false, false, false, 0);

    if (RelocM == Reloc::PIC_) {
      SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
      Result = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
    }
    return Result;
  }
  case Intrinsic::arm_neon_vmulls:
  case Intrinsic::arm_neon_vmullu: {
    unsigned NewOpc = (IntNo == Intrinsic::arm_neon_vmulls)
      ? ARMISD::VMULLs : ARMISD::VMULLu;
    return DAG.getNode(NewOpc, SDLoc(Op), Op.getValueType(),
                       Op.getOperand(1), Op.getOperand(2));
  }
  }
}

static SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG,
                                 const ARMSubtarget *Subtarget) {
  // FIXME: handle "fence singlethread" more efficiently.
  SDLoc dl(Op);
  if (!Subtarget->hasDataBarrier()) {
    // Some ARMv6 cpus can support data barriers with an mcr instruction.
    // Thumb1 and pre-v6 ARM mode use a libcall instead and should never get
    // here.
    assert(Subtarget->hasV6Ops() && !Subtarget->isThumb() &&
           "Unexpected ISD::ATOMIC_FENCE encountered. Should be libcall!");
    return DAG.getNode(ARMISD::MEMBARRIER_MCR, dl, MVT::Other, Op.getOperand(0),
                       DAG.getConstant(0, MVT::i32));
  }

  ConstantSDNode *OrdN = cast<ConstantSDNode>(Op.getOperand(1));
  AtomicOrdering Ord = static_cast<AtomicOrdering>(OrdN->getZExtValue());
  ARM_MB::MemBOpt Domain = ARM_MB::ISH;
  if (Subtarget->isMClass()) {
    // Only a full system barrier exists in the M-class architectures.
    Domain = ARM_MB::SY;
  } else if (Subtarget->isSwift() && Ord == Release) {
    // Swift happens to implement ISHST barriers in a way that's compatible with
    // Release semantics but weaker than ISH so we'd be fools not to use
    // it. Beware: other processors probably don't!
    Domain = ARM_MB::ISHST;
  }

  return DAG.getNode(ISD::INTRINSIC_VOID, dl, MVT::Other, Op.getOperand(0),
                     DAG.getConstant(Intrinsic::arm_dmb, MVT::i32),
                     DAG.getConstant(Domain, MVT::i32));
}

static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG,
                             const ARMSubtarget *Subtarget) {
  // ARM pre v5TE and Thumb1 does not have preload instructions.
  if (!(Subtarget->isThumb2() ||
        (!Subtarget->isThumb1Only() && Subtarget->hasV5TEOps())))
    // Just preserve the chain.
    return Op.getOperand(0);

  SDLoc dl(Op);
  unsigned isRead = ~cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue() & 1;
  if (!isRead &&
      (!Subtarget->hasV7Ops() || !Subtarget->hasMPExtension()))
    // ARMv7 with MP extension has PLDW.
    return Op.getOperand(0);

  unsigned isData = cast<ConstantSDNode>(Op.getOperand(4))->getZExtValue();
  if (Subtarget->isThumb()) {
    // Invert the bits.
    isRead = ~isRead & 1;
    isData = ~isData & 1;
  }

  return DAG.getNode(ARMISD::PRELOAD, dl, MVT::Other, Op.getOperand(0),
                     Op.getOperand(1), DAG.getConstant(isRead, MVT::i32),
                     DAG.getConstant(isData, MVT::i32));
}

static SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) {
  MachineFunction &MF = DAG.getMachineFunction();
  ARMFunctionInfo *FuncInfo = MF.getInfo<ARMFunctionInfo>();

  // vastart just stores the address of the VarArgsFrameIndex slot into the
  // memory location argument.
  SDLoc dl(Op);
  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
  SDValue FR = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
  const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
  return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1),
                      MachinePointerInfo(SV), false, false, 0);
}

SDValue
ARMTargetLowering::GetF64FormalArgument(CCValAssign &VA, CCValAssign &NextVA,
                                        SDValue &Root, SelectionDAG &DAG,
                                        SDLoc dl) const {
  MachineFunction &MF = DAG.getMachineFunction();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();

  const TargetRegisterClass *RC;
  if (AFI->isThumb1OnlyFunction())
    RC = &ARM::tGPRRegClass;
  else
    RC = &ARM::GPRRegClass;

  // Transform the arguments stored in physical registers into virtual ones.
  unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
  SDValue ArgValue = DAG.getCopyFromReg(Root, dl, Reg, MVT::i32);

  SDValue ArgValue2;
  if (NextVA.isMemLoc()) {
    MachineFrameInfo *MFI = MF.getFrameInfo();
    int FI = MFI->CreateFixedObject(4, NextVA.getLocMemOffset(), true);

    // Create load node to retrieve arguments from the stack.
    SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
    ArgValue2 = DAG.getLoad(MVT::i32, dl, Root, FIN,
                            MachinePointerInfo::getFixedStack(FI),
                            false, false, false, 0);
  } else {
    Reg = MF.addLiveIn(NextVA.getLocReg(), RC);
    ArgValue2 = DAG.getCopyFromReg(Root, dl, Reg, MVT::i32);
  }
  if (!Subtarget->isLittle())
    std::swap (ArgValue, ArgValue2);
  return DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, ArgValue, ArgValue2);
}

void
ARMTargetLowering::computeRegArea(CCState &CCInfo, MachineFunction &MF,
                                  unsigned InRegsParamRecordIdx,
                                  unsigned ArgSize,
                                  unsigned &ArgRegsSize,
                                  unsigned &ArgRegsSaveSize)
  const {
  unsigned NumGPRs;
  if (InRegsParamRecordIdx < CCInfo.getInRegsParamsCount()) {
    unsigned RBegin, REnd;
    CCInfo.getInRegsParamInfo(InRegsParamRecordIdx, RBegin, REnd);
    NumGPRs = REnd - RBegin;
  } else {
    unsigned int firstUnalloced;
    firstUnalloced = CCInfo.getFirstUnallocated(GPRArgRegs,
                                                sizeof(GPRArgRegs) /
                                                sizeof(GPRArgRegs[0]));
    NumGPRs = (firstUnalloced <= 3) ? (4 - firstUnalloced) : 0;
  }

  unsigned Align = MF.getTarget()
                       .getSubtargetImpl()
                       ->getFrameLowering()
                       ->getStackAlignment();
  ArgRegsSize = NumGPRs * 4;

  // If parameter is split between stack and GPRs...
  if (NumGPRs && Align > 4 &&
      (ArgRegsSize < ArgSize ||
        InRegsParamRecordIdx >= CCInfo.getInRegsParamsCount())) {
    // Add padding for part of param recovered from GPRs.  For example,
    // if Align == 8, its last byte must be at address K*8 - 1.
    // We need to do it, since remained (stack) part of parameter has
    // stack alignment, and we need to "attach" "GPRs head" without gaps
    // to it:
    // Stack:
    // |---- 8 bytes block ----| |---- 8 bytes block ----| |---- 8 bytes...
    // [ [padding] [GPRs head] ] [        Tail passed via stack       ....
    //
    ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
    unsigned Padding =
        OffsetToAlignment(ArgRegsSize + AFI->getArgRegsSaveSize(), Align);
    ArgRegsSaveSize = ArgRegsSize + Padding;
  } else
    // We don't need to extend regs save size for byval parameters if they
    // are passed via GPRs only.
    ArgRegsSaveSize = ArgRegsSize;
}

// The remaining GPRs hold either the beginning of variable-argument
// data, or the beginning of an aggregate passed by value (usually
// byval).  Either way, we allocate stack slots adjacent to the data
// provided by our caller, and store the unallocated registers there.
// If this is a variadic function, the va_list pointer will begin with
// these values; otherwise, this reassembles a (byval) structure that
// was split between registers and memory.
// Return: The frame index registers were stored into.
int
ARMTargetLowering::StoreByValRegs(CCState &CCInfo, SelectionDAG &DAG,
                                  SDLoc dl, SDValue &Chain,
                                  const Value *OrigArg,
                                  unsigned InRegsParamRecordIdx,
                                  unsigned OffsetFromOrigArg,
                                  unsigned ArgOffset,
                                  unsigned ArgSize,
                                  bool ForceMutable,
                                  unsigned ByValStoreOffset,
                                  unsigned TotalArgRegsSaveSize) const {

  // Currently, two use-cases possible:
  // Case #1. Non-var-args function, and we meet first byval parameter.
  //          Setup first unallocated register as first byval register;
  //          eat all remained registers
  //          (these two actions are performed by HandleByVal method).
  //          Then, here, we initialize stack frame with
  //          "store-reg" instructions.
  // Case #2. Var-args function, that doesn't contain byval parameters.
  //          The same: eat all remained unallocated registers,
  //          initialize stack frame.

  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
  unsigned firstRegToSaveIndex, lastRegToSaveIndex;
  unsigned RBegin, REnd;
  if (InRegsParamRecordIdx < CCInfo.getInRegsParamsCount()) {
    CCInfo.getInRegsParamInfo(InRegsParamRecordIdx, RBegin, REnd);
    firstRegToSaveIndex = RBegin - ARM::R0;
    lastRegToSaveIndex = REnd - ARM::R0;
  } else {
    firstRegToSaveIndex = CCInfo.getFirstUnallocated
      (GPRArgRegs, array_lengthof(GPRArgRegs));
    lastRegToSaveIndex = 4;
  }

  unsigned ArgRegsSize, ArgRegsSaveSize;
  computeRegArea(CCInfo, MF, InRegsParamRecordIdx, ArgSize,
                 ArgRegsSize, ArgRegsSaveSize);

  // Store any by-val regs to their spots on the stack so that they may be
  // loaded by deferencing the result of formal parameter pointer or va_next.
  // Note: once stack area for byval/varargs registers
  // was initialized, it can't be initialized again.
  if (ArgRegsSaveSize) {
    unsigned Padding = ArgRegsSaveSize - ArgRegsSize;

    if (Padding) {
      assert(AFI->getStoredByValParamsPadding() == 0 &&
             "The only parameter may be padded.");
      AFI->setStoredByValParamsPadding(Padding);
    }

    int FrameIndex = MFI->CreateFixedObject(ArgRegsSaveSize,
                                            Padding +
                                              ByValStoreOffset -
                                              (int64_t)TotalArgRegsSaveSize,
                                            false);
    SDValue FIN = DAG.getFrameIndex(FrameIndex, getPointerTy());
    if (Padding) {
       MFI->CreateFixedObject(Padding,
                              ArgOffset + ByValStoreOffset -
                                (int64_t)ArgRegsSaveSize,
                              false);
    }

    SmallVector<SDValue, 4> MemOps;
    for (unsigned i = 0; firstRegToSaveIndex < lastRegToSaveIndex;
         ++firstRegToSaveIndex, ++i) {
      const TargetRegisterClass *RC;
      if (AFI->isThumb1OnlyFunction())
        RC = &ARM::tGPRRegClass;
      else
        RC = &ARM::GPRRegClass;

      unsigned VReg = MF.addLiveIn(GPRArgRegs[firstRegToSaveIndex], RC);
      SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
      SDValue Store =
        DAG.getStore(Val.getValue(1), dl, Val, FIN,
                     MachinePointerInfo(OrigArg, OffsetFromOrigArg + 4*i),
                     false, false, 0);
      MemOps.push_back(Store);
      FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(), FIN,
                        DAG.getConstant(4, getPointerTy()));
    }

    AFI->setArgRegsSaveSize(ArgRegsSaveSize + AFI->getArgRegsSaveSize());

    if (!MemOps.empty())
      Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
    return FrameIndex;
  } else {
    if (ArgSize == 0) {
      // We cannot allocate a zero-byte object for the first variadic argument,
      // so just make up a size.
      ArgSize = 4;
    }
    // This will point to the next argument passed via stack.
    return MFI->CreateFixedObject(
      ArgSize, ArgOffset, !ForceMutable);
  }
}

// Setup stack frame, the va_list pointer will start from.
void
ARMTargetLowering::VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG,
                                        SDLoc dl, SDValue &Chain,
                                        unsigned ArgOffset,
                                        unsigned TotalArgRegsSaveSize,
                                        bool ForceMutable) const {
  MachineFunction &MF = DAG.getMachineFunction();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();

  // Try to store any remaining integer argument regs
  // to their spots on the stack so that they may be loaded by deferencing
  // the result of va_next.
  // If there is no regs to be stored, just point address after last
  // argument passed via stack.
  int FrameIndex =
    StoreByValRegs(CCInfo, DAG, dl, Chain, nullptr,
                   CCInfo.getInRegsParamsCount(), 0, ArgOffset, 0, ForceMutable,
                   0, TotalArgRegsSaveSize);

  AFI->setVarArgsFrameIndex(FrameIndex);
}

SDValue
ARMTargetLowering::LowerFormalArguments(SDValue Chain,
                                        CallingConv::ID CallConv, bool isVarArg,
                                        const SmallVectorImpl<ISD::InputArg>
                                          &Ins,
                                        SDLoc dl, SelectionDAG &DAG,
                                        SmallVectorImpl<SDValue> &InVals)
                                          const {
  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();

  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();

  // Assign locations to all of the incoming arguments.
  SmallVector<CCValAssign, 16> ArgLocs;
  ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                    *DAG.getContext(), Prologue);
  CCInfo.AnalyzeFormalArguments(Ins,
                                CCAssignFnForNode(CallConv, /* Return*/ false,
                                                  isVarArg));

  SmallVector<SDValue, 16> ArgValues;
  int lastInsIndex = -1;
  SDValue ArgValue;
  Function::const_arg_iterator CurOrigArg = MF.getFunction()->arg_begin();
  unsigned CurArgIdx = 0;

  // Initially ArgRegsSaveSize is zero.
  // Then we increase this value each time we meet byval parameter.
  // We also increase this value in case of varargs function.
  AFI->setArgRegsSaveSize(0);

  unsigned ByValStoreOffset = 0;
  unsigned TotalArgRegsSaveSize = 0;
  unsigned ArgRegsSaveSizeMaxAlign = 4;

  // Calculate the amount of stack space that we need to allocate to store
  // byval and variadic arguments that are passed in registers.
  // We need to know this before we allocate the first byval or variadic
  // argument, as they will be allocated a stack slot below the CFA (Canonical
  // Frame Address, the stack pointer at entry to the function).
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];
    if (VA.isMemLoc()) {
      int index = VA.getValNo();
      if (index != lastInsIndex) {
        ISD::ArgFlagsTy Flags = Ins[index].Flags;
        if (Flags.isByVal()) {
          unsigned ExtraArgRegsSize;
          unsigned ExtraArgRegsSaveSize;
          computeRegArea(CCInfo, MF, CCInfo.getInRegsParamsProcessed(),
                         Flags.getByValSize(),
                         ExtraArgRegsSize, ExtraArgRegsSaveSize);

          TotalArgRegsSaveSize += ExtraArgRegsSaveSize;
          if (Flags.getByValAlign() > ArgRegsSaveSizeMaxAlign)
              ArgRegsSaveSizeMaxAlign = Flags.getByValAlign();
          CCInfo.nextInRegsParam();
        }
        lastInsIndex = index;
      }
    }
  }
  CCInfo.rewindByValRegsInfo();
  lastInsIndex = -1;
  if (isVarArg && MFI->hasVAStart()) {
    unsigned ExtraArgRegsSize;
    unsigned ExtraArgRegsSaveSize;
    computeRegArea(CCInfo, MF, CCInfo.getInRegsParamsCount(), 0,
                   ExtraArgRegsSize, ExtraArgRegsSaveSize);
    TotalArgRegsSaveSize += ExtraArgRegsSaveSize;
  }
  // If the arg regs save area contains N-byte aligned values, the
  // bottom of it must be at least N-byte aligned.
  TotalArgRegsSaveSize = RoundUpToAlignment(TotalArgRegsSaveSize, ArgRegsSaveSizeMaxAlign);
  TotalArgRegsSaveSize = std::min(TotalArgRegsSaveSize, 16U);

  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];
    std::advance(CurOrigArg, Ins[VA.getValNo()].OrigArgIndex - CurArgIdx);
    CurArgIdx = Ins[VA.getValNo()].OrigArgIndex;
    // Arguments stored in registers.
    if (VA.isRegLoc()) {
      EVT RegVT = VA.getLocVT();

      if (VA.needsCustom()) {
        // f64 and vector types are split up into multiple registers or
        // combinations of registers and stack slots.
        if (VA.getLocVT() == MVT::v2f64) {
          SDValue ArgValue1 = GetF64FormalArgument(VA, ArgLocs[++i],
                                                   Chain, DAG, dl);
          VA = ArgLocs[++i]; // skip ahead to next loc
          SDValue ArgValue2;
          if (VA.isMemLoc()) {
            int FI = MFI->CreateFixedObject(8, VA.getLocMemOffset(), true);
            SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
            ArgValue2 = DAG.getLoad(MVT::f64, dl, Chain, FIN,
                                    MachinePointerInfo::getFixedStack(FI),
                                    false, false, false, 0);
          } else {
            ArgValue2 = GetF64FormalArgument(VA, ArgLocs[++i],
                                             Chain, DAG, dl);
          }
          ArgValue = DAG.getNode(ISD::UNDEF, dl, MVT::v2f64);
          ArgValue = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64,
                                 ArgValue, ArgValue1, DAG.getIntPtrConstant(0));
          ArgValue = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64,
                                 ArgValue, ArgValue2, DAG.getIntPtrConstant(1));
        } else
          ArgValue = GetF64FormalArgument(VA, ArgLocs[++i], Chain, DAG, dl);

      } else {
        const TargetRegisterClass *RC;

        if (RegVT == MVT::f32)
          RC = &ARM::SPRRegClass;
        else if (RegVT == MVT::f64)
          RC = &ARM::DPRRegClass;
        else if (RegVT == MVT::v2f64)
          RC = &ARM::QPRRegClass;
        else if (RegVT == MVT::i32)
          RC = AFI->isThumb1OnlyFunction() ?
            (const TargetRegisterClass*)&ARM::tGPRRegClass :
            (const TargetRegisterClass*)&ARM::GPRRegClass;
        else
          llvm_unreachable("RegVT not supported by FORMAL_ARGUMENTS Lowering");

        // Transform the arguments in physical registers into virtual ones.
        unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
        ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);
      }

      // If this is an 8 or 16-bit value, it is really passed promoted
      // to 32 bits.  Insert an assert[sz]ext to capture this, then
      // truncate to the right size.
      switch (VA.getLocInfo()) {
      default: llvm_unreachable("Unknown loc info!");
      case CCValAssign::Full: break;
      case CCValAssign::BCvt:
        ArgValue = DAG.getNode(ISD::BITCAST, dl, VA.getValVT(), ArgValue);
        break;
      case CCValAssign::SExt:
        ArgValue = DAG.getNode(ISD::AssertSext, dl, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
        ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
        break;
      case CCValAssign::ZExt:
        ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
        ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
        break;
      }

      InVals.push_back(ArgValue);

    } else { // VA.isRegLoc()

      // sanity check
      assert(VA.isMemLoc());
      assert(VA.getValVT() != MVT::i64 && "i64 should already be lowered");

      int index = ArgLocs[i].getValNo();

      // Some Ins[] entries become multiple ArgLoc[] entries.
      // Process them only once.
      if (index != lastInsIndex)
        {
          ISD::ArgFlagsTy Flags = Ins[index].Flags;
          // FIXME: For now, all byval parameter objects are marked mutable.
          // This can be changed with more analysis.
          // In case of tail call optimization mark all arguments mutable.
          // Since they could be overwritten by lowering of arguments in case of
          // a tail call.
          if (Flags.isByVal()) {
            unsigned CurByValIndex = CCInfo.getInRegsParamsProcessed();

            ByValStoreOffset = RoundUpToAlignment(ByValStoreOffset, Flags.getByValAlign());
            int FrameIndex = StoreByValRegs(
                CCInfo, DAG, dl, Chain, CurOrigArg,
                CurByValIndex,
                Ins[VA.getValNo()].PartOffset,
                VA.getLocMemOffset(),
                Flags.getByValSize(),
                true /*force mutable frames*/,
                ByValStoreOffset,
                TotalArgRegsSaveSize);
            ByValStoreOffset += Flags.getByValSize();
            ByValStoreOffset = std::min(ByValStoreOffset, 16U);
            InVals.push_back(DAG.getFrameIndex(FrameIndex, getPointerTy()));
            CCInfo.nextInRegsParam();
          } else {
            unsigned FIOffset = VA.getLocMemOffset();
            int FI = MFI->CreateFixedObject(VA.getLocVT().getSizeInBits()/8,
                                            FIOffset, true);

            // Create load nodes to retrieve arguments from the stack.
            SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
            InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN,
                                         MachinePointerInfo::getFixedStack(FI),
                                         false, false, false, 0));
          }
          lastInsIndex = index;
        }
    }
  }

  // varargs
  if (isVarArg && MFI->hasVAStart())
    VarArgStyleRegisters(CCInfo, DAG, dl, Chain,
                         CCInfo.getNextStackOffset(),
                         TotalArgRegsSaveSize);

  AFI->setArgumentStackSize(CCInfo.getNextStackOffset());

  return Chain;
}

/// isFloatingPointZero - Return true if this is +0.0.
static bool isFloatingPointZero(SDValue Op) {
  if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Op))
    return CFP->getValueAPF().isPosZero();
  else if (ISD::isEXTLoad(Op.getNode()) || ISD::isNON_EXTLoad(Op.getNode())) {
    // Maybe this has already been legalized into the constant pool?
    if (Op.getOperand(1).getOpcode() == ARMISD::Wrapper) {
      SDValue WrapperOp = Op.getOperand(1).getOperand(0);
      if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(WrapperOp))
        if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CP->getConstVal()))
          return CFP->getValueAPF().isPosZero();
    }
  } else if (Op->getOpcode() == ISD::BITCAST &&
             Op->getValueType(0) == MVT::f64) {
    // Handle (ISD::BITCAST (ARMISD::VMOVIMM (ISD::TargetConstant 0)) MVT::f64)
    // created by LowerConstantFP().
    SDValue BitcastOp = Op->getOperand(0);
    if (BitcastOp->getOpcode() == ARMISD::VMOVIMM) {
      SDValue MoveOp = BitcastOp->getOperand(0);
      if (MoveOp->getOpcode() == ISD::TargetConstant &&
          cast<ConstantSDNode>(MoveOp)->getZExtValue() == 0) {
        return true;
      }
    }
  }
  return false;
}

/// Returns appropriate ARM CMP (cmp) and corresponding condition code for
/// the given operands.
SDValue
ARMTargetLowering::getARMCmp(SDValue LHS, SDValue RHS, ISD::CondCode CC,
                             SDValue &ARMcc, SelectionDAG &DAG,
                             SDLoc dl) const {
  if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS.getNode())) {
    unsigned C = RHSC->getZExtValue();
    if (!isLegalICmpImmediate(C)) {
      // Constant does not fit, try adjusting it by one?
      switch (CC) {
      default: break;
      case ISD::SETLT:
      case ISD::SETGE:
        if (C != 0x80000000 && isLegalICmpImmediate(C-1)) {
          CC = (CC == ISD::SETLT) ? ISD::SETLE : ISD::SETGT;
          RHS = DAG.getConstant(C-1, MVT::i32);
        }
        break;
      case ISD::SETULT:
      case ISD::SETUGE:
        if (C != 0 && isLegalICmpImmediate(C-1)) {
          CC = (CC == ISD::SETULT) ? ISD::SETULE : ISD::SETUGT;
          RHS = DAG.getConstant(C-1, MVT::i32);
        }
        break;
      case ISD::SETLE:
      case ISD::SETGT:
        if (C != 0x7fffffff && isLegalICmpImmediate(C+1)) {
          CC = (CC == ISD::SETLE) ? ISD::SETLT : ISD::SETGE;
          RHS = DAG.getConstant(C+1, MVT::i32);
        }
        break;
      case ISD::SETULE:
      case ISD::SETUGT:
        if (C != 0xffffffff && isLegalICmpImmediate(C+1)) {
          CC = (CC == ISD::SETULE) ? ISD::SETULT : ISD::SETUGE;
          RHS = DAG.getConstant(C+1, MVT::i32);
        }
        break;
      }
    }
  }

  ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
  ARMISD::NodeType CompareType;
  switch (CondCode) {
  default:
    CompareType = ARMISD::CMP;
    break;
  case ARMCC::EQ:
  case ARMCC::NE:
    // Uses only Z Flag
    CompareType = ARMISD::CMPZ;
    break;
  }
  ARMcc = DAG.getConstant(CondCode, MVT::i32);
  return DAG.getNode(CompareType, dl, MVT::Glue, LHS, RHS);
}

/// Returns a appropriate VFP CMP (fcmp{s|d}+fmstat) for the given operands.
SDValue
ARMTargetLowering::getVFPCmp(SDValue LHS, SDValue RHS, SelectionDAG &DAG,
                             SDLoc dl) const {
  assert(!Subtarget->isFPOnlySP() || RHS.getValueType() != MVT::f64);
  SDValue Cmp;
  if (!isFloatingPointZero(RHS))
    Cmp = DAG.getNode(ARMISD::CMPFP, dl, MVT::Glue, LHS, RHS);
  else
    Cmp = DAG.getNode(ARMISD::CMPFPw0, dl, MVT::Glue, LHS);
  return DAG.getNode(ARMISD::FMSTAT, dl, MVT::Glue, Cmp);
}

/// duplicateCmp - Glue values can have only one use, so this function
/// duplicates a comparison node.
SDValue
ARMTargetLowering::duplicateCmp(SDValue Cmp, SelectionDAG &DAG) const {
  unsigned Opc = Cmp.getOpcode();
  SDLoc DL(Cmp);
  if (Opc == ARMISD::CMP || Opc == ARMISD::CMPZ)
    return DAG.getNode(Opc, DL, MVT::Glue, Cmp.getOperand(0),Cmp.getOperand(1));

  assert(Opc == ARMISD::FMSTAT && "unexpected comparison operation");
  Cmp = Cmp.getOperand(0);
  Opc = Cmp.getOpcode();
  if (Opc == ARMISD::CMPFP)
    Cmp = DAG.getNode(Opc, DL, MVT::Glue, Cmp.getOperand(0),Cmp.getOperand(1));
  else {
    assert(Opc == ARMISD::CMPFPw0 && "unexpected operand of FMSTAT");
    Cmp = DAG.getNode(Opc, DL, MVT::Glue, Cmp.getOperand(0));
  }
  return DAG.getNode(ARMISD::FMSTAT, DL, MVT::Glue, Cmp);
}

std::pair<SDValue, SDValue>
ARMTargetLowering::getARMXALUOOp(SDValue Op, SelectionDAG &DAG,
                                 SDValue &ARMcc) const {
  assert(Op.getValueType() == MVT::i32 &&  "Unsupported value type");

  SDValue Value, OverflowCmp;
  SDValue LHS = Op.getOperand(0);
  SDValue RHS = Op.getOperand(1);


  // FIXME: We are currently always generating CMPs because we don't support
  // generating CMN through the backend. This is not as good as the natural
  // CMP case because it causes a register dependency and cannot be folded
  // later.

  switch (Op.getOpcode()) {
  default:
    llvm_unreachable("Unknown overflow instruction!");
  case ISD::SADDO:
    ARMcc = DAG.getConstant(ARMCC::VC, MVT::i32);
    Value = DAG.getNode(ISD::ADD, SDLoc(Op), Op.getValueType(), LHS, RHS);
    OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, Value, LHS);
    break;
  case ISD::UADDO:
    ARMcc = DAG.getConstant(ARMCC::HS, MVT::i32);
    Value = DAG.getNode(ISD::ADD, SDLoc(Op), Op.getValueType(), LHS, RHS);
    OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, Value, LHS);
    break;
  case ISD::SSUBO:
    ARMcc = DAG.getConstant(ARMCC::VC, MVT::i32);
    Value = DAG.getNode(ISD::SUB, SDLoc(Op), Op.getValueType(), LHS, RHS);
    OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, LHS, RHS);
    break;
  case ISD::USUBO:
    ARMcc = DAG.getConstant(ARMCC::HS, MVT::i32);
    Value = DAG.getNode(ISD::SUB, SDLoc(Op), Op.getValueType(), LHS, RHS);
    OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, LHS, RHS);
    break;
  } // switch (...)

  return std::make_pair(Value, OverflowCmp);
}


SDValue
ARMTargetLowering::LowerXALUO(SDValue Op, SelectionDAG &DAG) const {
  // Let legalize expand this if it isn't a legal type yet.
  if (!DAG.getTargetLoweringInfo().isTypeLegal(Op.getValueType()))
    return SDValue();

  SDValue Value, OverflowCmp;
  SDValue ARMcc;
  std::tie(Value, OverflowCmp) = getARMXALUOOp(Op, DAG, ARMcc);
  SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
  // We use 0 and 1 as false and true values.
  SDValue TVal = DAG.getConstant(1, MVT::i32);
  SDValue FVal = DAG.getConstant(0, MVT::i32);
  EVT VT = Op.getValueType();

  SDValue Overflow = DAG.getNode(ARMISD::CMOV, SDLoc(Op), VT, TVal, FVal,
                                 ARMcc, CCR, OverflowCmp);

  SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::i32);
  return DAG.getNode(ISD::MERGE_VALUES, SDLoc(Op), VTs, Value, Overflow);
}


SDValue ARMTargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
  SDValue Cond = Op.getOperand(0);
  SDValue SelectTrue = Op.getOperand(1);
  SDValue SelectFalse = Op.getOperand(2);
  SDLoc dl(Op);
  unsigned Opc = Cond.getOpcode();

  if (Cond.getResNo() == 1 &&
      (Opc == ISD::SADDO || Opc == ISD::UADDO || Opc == ISD::SSUBO ||
       Opc == ISD::USUBO)) {
    if (!DAG.getTargetLoweringInfo().isTypeLegal(Cond->getValueType(0)))
      return SDValue();

    SDValue Value, OverflowCmp;
    SDValue ARMcc;
    std::tie(Value, OverflowCmp) = getARMXALUOOp(Cond, DAG, ARMcc);
    SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
    EVT VT = Op.getValueType();

    return getCMOV(SDLoc(Op), VT, SelectTrue, SelectFalse, ARMcc, CCR,
                   OverflowCmp, DAG);
  }

  // Convert:
  //
  //   (select (cmov 1, 0, cond), t, f) -> (cmov t, f, cond)
  //   (select (cmov 0, 1, cond), t, f) -> (cmov f, t, cond)
  //
  if (Cond.getOpcode() == ARMISD::CMOV && Cond.hasOneUse()) {
    const ConstantSDNode *CMOVTrue =
      dyn_cast<ConstantSDNode>(Cond.getOperand(0));
    const ConstantSDNode *CMOVFalse =
      dyn_cast<ConstantSDNode>(Cond.getOperand(1));

    if (CMOVTrue && CMOVFalse) {
      unsigned CMOVTrueVal = CMOVTrue->getZExtValue();
      unsigned CMOVFalseVal = CMOVFalse->getZExtValue();

      SDValue True;
      SDValue False;
      if (CMOVTrueVal == 1 && CMOVFalseVal == 0) {
        True = SelectTrue;
        False = SelectFalse;
      } else if (CMOVTrueVal == 0 && CMOVFalseVal == 1) {
        True = SelectFalse;
        False = SelectTrue;
      }

      if (True.getNode() && False.getNode()) {
        EVT VT = Op.getValueType();
        SDValue ARMcc = Cond.getOperand(2);
        SDValue CCR = Cond.getOperand(3);
        SDValue Cmp = duplicateCmp(Cond.getOperand(4), DAG);
        assert(True.getValueType() == VT);
        return getCMOV(dl, VT, True, False, ARMcc, CCR, Cmp, DAG);
      }
    }
  }

  // ARM's BooleanContents value is UndefinedBooleanContent. Mask out the
  // undefined bits before doing a full-word comparison with zero.
  Cond = DAG.getNode(ISD::AND, dl, Cond.getValueType(), Cond,
                     DAG.getConstant(1, Cond.getValueType()));

  return DAG.getSelectCC(dl, Cond,
                         DAG.getConstant(0, Cond.getValueType()),
                         SelectTrue, SelectFalse, ISD::SETNE);
}

static ISD::CondCode getInverseCCForVSEL(ISD::CondCode CC) {
  if (CC == ISD::SETNE)
    return ISD::SETEQ;
  return ISD::getSetCCInverse(CC, true);
}

static void checkVSELConstraints(ISD::CondCode CC, ARMCC::CondCodes &CondCode,
                                 bool &swpCmpOps, bool &swpVselOps) {
  // Start by selecting the GE condition code for opcodes that return true for
  // 'equality'
  if (CC == ISD::SETUGE || CC == ISD::SETOGE || CC == ISD::SETOLE ||
      CC == ISD::SETULE)
    CondCode = ARMCC::GE;

  // and GT for opcodes that return false for 'equality'.
  else if (CC == ISD::SETUGT || CC == ISD::SETOGT || CC == ISD::SETOLT ||
           CC == ISD::SETULT)
    CondCode = ARMCC::GT;

  // Since we are constrained to GE/GT, if the opcode contains 'less', we need
  // to swap the compare operands.
  if (CC == ISD::SETOLE || CC == ISD::SETULE || CC == ISD::SETOLT ||
      CC == ISD::SETULT)
    swpCmpOps = true;

  // Both GT and GE are ordered comparisons, and return false for 'unordered'.
  // If we have an unordered opcode, we need to swap the operands to the VSEL
  // instruction (effectively negating the condition).
  //
  // This also has the effect of swapping which one of 'less' or 'greater'
  // returns true, so we also swap the compare operands. It also switches
  // whether we return true for 'equality', so we compensate by picking the
  // opposite condition code to our original choice.
  if (CC == ISD::SETULE || CC == ISD::SETULT || CC == ISD::SETUGE ||
      CC == ISD::SETUGT) {
    swpCmpOps = !swpCmpOps;
    swpVselOps = !swpVselOps;
    CondCode = CondCode == ARMCC::GT ? ARMCC::GE : ARMCC::GT;
  }

  // 'ordered' is 'anything but unordered', so use the VS condition code and
  // swap the VSEL operands.
  if (CC == ISD::SETO) {
    CondCode = ARMCC::VS;
    swpVselOps = true;
  }

  // 'unordered or not equal' is 'anything but equal', so use the EQ condition
  // code and swap the VSEL operands.
  if (CC == ISD::SETUNE) {
    CondCode = ARMCC::EQ;
    swpVselOps = true;
  }
}

SDValue ARMTargetLowering::getCMOV(SDLoc dl, EVT VT, SDValue FalseVal,
                                   SDValue TrueVal, SDValue ARMcc, SDValue CCR,
                                   SDValue Cmp, SelectionDAG &DAG) const {
  if (Subtarget->isFPOnlySP() && VT == MVT::f64) {
    FalseVal = DAG.getNode(ARMISD::VMOVRRD, dl,
                           DAG.getVTList(MVT::i32, MVT::i32), FalseVal);
    TrueVal = DAG.getNode(ARMISD::VMOVRRD, dl,
                          DAG.getVTList(MVT::i32, MVT::i32), TrueVal);

    SDValue TrueLow = TrueVal.getValue(0);
    SDValue TrueHigh = TrueVal.getValue(1);
    SDValue FalseLow = FalseVal.getValue(0);
    SDValue FalseHigh = FalseVal.getValue(1);

    SDValue Low = DAG.getNode(ARMISD::CMOV, dl, MVT::i32, FalseLow, TrueLow,
                              ARMcc, CCR, Cmp);
    SDValue High = DAG.getNode(ARMISD::CMOV, dl, MVT::i32, FalseHigh, TrueHigh,
                               ARMcc, CCR, duplicateCmp(Cmp, DAG));

    return DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Low, High);
  } else {
    return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR,
                       Cmp);
  }
}

SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
  EVT VT = Op.getValueType();
  SDValue LHS = Op.getOperand(0);
  SDValue RHS = Op.getOperand(1);
  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
  SDValue TrueVal = Op.getOperand(2);
  SDValue FalseVal = Op.getOperand(3);
  SDLoc dl(Op);

  if (Subtarget->isFPOnlySP() && LHS.getValueType() == MVT::f64) {
    DAG.getTargetLoweringInfo().softenSetCCOperands(DAG, MVT::f64, LHS, RHS, CC,
                                                    dl);

    // If softenSetCCOperands only returned one value, we should compare it to
    // zero.
    if (!RHS.getNode()) {
      RHS = DAG.getConstant(0, LHS.getValueType());
      CC = ISD::SETNE;
    }
  }

  if (LHS.getValueType() == MVT::i32) {
    // Try to generate VSEL on ARMv8.
    // The VSEL instruction can't use all the usual ARM condition
    // codes: it only has two bits to select the condition code, so it's
    // constrained to use only GE, GT, VS and EQ.
    //
    // To implement all the various ISD::SETXXX opcodes, we sometimes need to
    // swap the operands of the previous compare instruction (effectively
    // inverting the compare condition, swapping 'less' and 'greater') and
    // sometimes need to swap the operands to the VSEL (which inverts the
    // condition in the sense of firing whenever the previous condition didn't)
    if (getSubtarget()->hasFPARMv8() && (TrueVal.getValueType() == MVT::f32 ||
                                      TrueVal.getValueType() == MVT::f64)) {
      ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
      if (CondCode == ARMCC::LT || CondCode == ARMCC::LE ||
          CondCode == ARMCC::VC || CondCode == ARMCC::NE) {
        CC = getInverseCCForVSEL(CC);
        std::swap(TrueVal, FalseVal);
      }
    }

    SDValue ARMcc;
    SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
    SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
    return getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG);
  }

  ARMCC::CondCodes CondCode, CondCode2;
  FPCCToARMCC(CC, CondCode, CondCode2);

  // Try to generate VSEL on ARMv8.
  if (getSubtarget()->hasFPARMv8() && (TrueVal.getValueType() == MVT::f32 ||
                                    TrueVal.getValueType() == MVT::f64)) {
    // We can select VMAXNM/VMINNM from a compare followed by a select with the
    // same operands, as follows:
    //   c = fcmp [ogt, olt, ugt, ult] a, b
    //   select c, a, b
    // We only do this in unsafe-fp-math, because signed zeros and NaNs are
    // handled differently than the original code sequence.
    if (getTargetMachine().Options.UnsafeFPMath) {
      if (LHS == TrueVal && RHS == FalseVal) {
        if (CC == ISD::SETOGT || CC == ISD::SETUGT)
          return DAG.getNode(ARMISD::VMAXNM, dl, VT, TrueVal, FalseVal);
        if (CC == ISD::SETOLT || CC == ISD::SETULT)
          return DAG.getNode(ARMISD::VMINNM, dl, VT, TrueVal, FalseVal);
      } else if (LHS == FalseVal && RHS == TrueVal) {
        if (CC == ISD::SETOLT || CC == ISD::SETULT)
          return DAG.getNode(ARMISD::VMAXNM, dl, VT, TrueVal, FalseVal);
        if (CC == ISD::SETOGT || CC == ISD::SETUGT)
          return DAG.getNode(ARMISD::VMINNM, dl, VT, TrueVal, FalseVal);
      }
    }

    bool swpCmpOps = false;
    bool swpVselOps = false;
    checkVSELConstraints(CC, CondCode, swpCmpOps, swpVselOps);

    if (CondCode == ARMCC::GT || CondCode == ARMCC::GE ||
        CondCode == ARMCC::VS || CondCode == ARMCC::EQ) {
      if (swpCmpOps)
        std::swap(LHS, RHS);
      if (swpVselOps)
        std::swap(TrueVal, FalseVal);
    }
  }

  SDValue ARMcc = DAG.getConstant(CondCode, MVT::i32);
  SDValue Cmp = getVFPCmp(LHS, RHS, DAG, dl);
  SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
  SDValue Result = getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG);
  if (CondCode2 != ARMCC::AL) {
    SDValue ARMcc2 = DAG.getConstant(CondCode2, MVT::i32);
    // FIXME: Needs another CMP because flag can have but one use.
    SDValue Cmp2 = getVFPCmp(LHS, RHS, DAG, dl);
    Result = getCMOV(dl, VT, Result, TrueVal, ARMcc2, CCR, Cmp2, DAG);
  }
  return Result;
}

/// canChangeToInt - Given the fp compare operand, return true if it is suitable
/// to morph to an integer compare sequence.
static bool canChangeToInt(SDValue Op, bool &SeenZero,
                           const ARMSubtarget *Subtarget) {
  SDNode *N = Op.getNode();
  if (!N->hasOneUse())
    // Otherwise it requires moving the value from fp to integer registers.
    return false;
  if (!N->getNumValues())
    return false;
  EVT VT = Op.getValueType();
  if (VT != MVT::f32 && !Subtarget->isFPBrccSlow())
    // f32 case is generally profitable. f64 case only makes sense when vcmpe +
    // vmrs are very slow, e.g. cortex-a8.
    return false;

  if (isFloatingPointZero(Op)) {
    SeenZero = true;
    return true;
  }
  return ISD::isNormalLoad(N);
}

static SDValue bitcastf32Toi32(SDValue Op, SelectionDAG &DAG) {
  if (isFloatingPointZero(Op))
    return DAG.getConstant(0, MVT::i32);

  if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Op))
    return DAG.getLoad(MVT::i32, SDLoc(Op),
                       Ld->getChain(), Ld->getBasePtr(), Ld->getPointerInfo(),
                       Ld->isVolatile(), Ld->isNonTemporal(),
                       Ld->isInvariant(), Ld->getAlignment());

  llvm_unreachable("Unknown VFP cmp argument!");
}

static void expandf64Toi32(SDValue Op, SelectionDAG &DAG,
                           SDValue &RetVal1, SDValue &RetVal2) {
  if (isFloatingPointZero(Op)) {
    RetVal1 = DAG.getConstant(0, MVT::i32);
    RetVal2 = DAG.getConstant(0, MVT::i32);
    return;
  }

  if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Op)) {
    SDValue Ptr = Ld->getBasePtr();
    RetVal1 = DAG.getLoad(MVT::i32, SDLoc(Op),
                          Ld->getChain(), Ptr,
                          Ld->getPointerInfo(),
                          Ld->isVolatile(), Ld->isNonTemporal(),
                          Ld->isInvariant(), Ld->getAlignment());

    EVT PtrType = Ptr.getValueType();
    unsigned NewAlign = MinAlign(Ld->getAlignment(), 4);
    SDValue NewPtr = DAG.getNode(ISD::ADD, SDLoc(Op),
                                 PtrType, Ptr, DAG.getConstant(4, PtrType));
    RetVal2 = DAG.getLoad(MVT::i32, SDLoc(Op),
                          Ld->getChain(), NewPtr,
                          Ld->getPointerInfo().getWithOffset(4),
                          Ld->isVolatile(), Ld->isNonTemporal(),
                          Ld->isInvariant(), NewAlign);
    return;
  }

  llvm_unreachable("Unknown VFP cmp argument!");
}

/// OptimizeVFPBrcond - With -enable-unsafe-fp-math, it's legal to optimize some
/// f32 and even f64 comparisons to integer ones.
SDValue
ARMTargetLowering::OptimizeVFPBrcond(SDValue Op, SelectionDAG &DAG) const {
  SDValue Chain = Op.getOperand(0);
  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(1))->get();
  SDValue LHS = Op.getOperand(2);
  SDValue RHS = Op.getOperand(3);
  SDValue Dest = Op.getOperand(4);
  SDLoc dl(Op);

  bool LHSSeenZero = false;
  bool LHSOk = canChangeToInt(LHS, LHSSeenZero, Subtarget);
  bool RHSSeenZero = false;
  bool RHSOk = canChangeToInt(RHS, RHSSeenZero, Subtarget);
  if (LHSOk && RHSOk && (LHSSeenZero || RHSSeenZero)) {
    // If unsafe fp math optimization is enabled and there are no other uses of
    // the CMP operands, and the condition code is EQ or NE, we can optimize it
    // to an integer comparison.
    if (CC == ISD::SETOEQ)
      CC = ISD::SETEQ;
    else if (CC == ISD::SETUNE)
      CC = ISD::SETNE;

    SDValue Mask = DAG.getConstant(0x7fffffff, MVT::i32);
    SDValue ARMcc;
    if (LHS.getValueType() == MVT::f32) {
      LHS = DAG.getNode(ISD::AND, dl, MVT::i32,
                        bitcastf32Toi32(LHS, DAG), Mask);
      RHS = DAG.getNode(ISD::AND, dl, MVT::i32,
                        bitcastf32Toi32(RHS, DAG), Mask);
      SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
      SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
      return DAG.getNode(ARMISD::BRCOND, dl, MVT::Other,
                         Chain, Dest, ARMcc, CCR, Cmp);
    }

    SDValue LHS1, LHS2;
    SDValue RHS1, RHS2;
    expandf64Toi32(LHS, DAG, LHS1, LHS2);
    expandf64Toi32(RHS, DAG, RHS1, RHS2);
    LHS2 = DAG.getNode(ISD::AND, dl, MVT::i32, LHS2, Mask);
    RHS2 = DAG.getNode(ISD::AND, dl, MVT::i32, RHS2, Mask);
    ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
    ARMcc = DAG.getConstant(CondCode, MVT::i32);
    SDVTList VTList = DAG.getVTList(MVT::Other, MVT::Glue);
    SDValue Ops[] = { Chain, ARMcc, LHS1, LHS2, RHS1, RHS2, Dest };
    return DAG.getNode(ARMISD::BCC_i64, dl, VTList, Ops);
  }

  return SDValue();
}

SDValue ARMTargetLowering::LowerBR_CC(SDValue Op, SelectionDAG &DAG) const {
  SDValue Chain = Op.getOperand(0);
  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(1))->get();
  SDValue LHS = Op.getOperand(2);
  SDValue RHS = Op.getOperand(3);
  SDValue Dest = Op.getOperand(4);
  SDLoc dl(Op);

  if (Subtarget->isFPOnlySP() && LHS.getValueType() == MVT::f64) {
    DAG.getTargetLoweringInfo().softenSetCCOperands(DAG, MVT::f64, LHS, RHS, CC,
                                                    dl);

    // If softenSetCCOperands only returned one value, we should compare it to
    // zero.
    if (!RHS.getNode()) {
      RHS = DAG.getConstant(0, LHS.getValueType());
      CC = ISD::SETNE;
    }
  }

  if (LHS.getValueType() == MVT::i32) {
    SDValue ARMcc;
    SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
    SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
    return DAG.getNode(ARMISD::BRCOND, dl, MVT::Other,
                       Chain, Dest, ARMcc, CCR, Cmp);
  }

  assert(LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64);

  if (getTargetMachine().Options.UnsafeFPMath &&
      (CC == ISD::SETEQ || CC == ISD::SETOEQ ||
       CC == ISD::SETNE || CC == ISD::SETUNE)) {
    SDValue Result = OptimizeVFPBrcond(Op, DAG);
    if (Result.getNode())
      return Result;
  }

  ARMCC::CondCodes CondCode, CondCode2;
  FPCCToARMCC(CC, CondCode, CondCode2);

  SDValue ARMcc = DAG.getConstant(CondCode, MVT::i32);
  SDValue Cmp = getVFPCmp(LHS, RHS, DAG, dl);
  SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
  SDVTList VTList = DAG.getVTList(MVT::Other, MVT::Glue);
  SDValue Ops[] = { Chain, Dest, ARMcc, CCR, Cmp };
  SDValue Res = DAG.getNode(ARMISD::BRCOND, dl, VTList, Ops);
  if (CondCode2 != ARMCC::AL) {
    ARMcc = DAG.getConstant(CondCode2, MVT::i32);
    SDValue Ops[] = { Res, Dest, ARMcc, CCR, Res.getValue(1) };
    Res = DAG.getNode(ARMISD::BRCOND, dl, VTList, Ops);
  }
  return Res;
}

SDValue ARMTargetLowering::LowerBR_JT(SDValue Op, SelectionDAG &DAG) const {
  SDValue Chain = Op.getOperand(0);
  SDValue Table = Op.getOperand(1);
  SDValue Index = Op.getOperand(2);
  SDLoc dl(Op);

  EVT PTy = getPointerTy();
  JumpTableSDNode *JT = cast<JumpTableSDNode>(Table);
  ARMFunctionInfo *AFI = DAG.getMachineFunction().getInfo<ARMFunctionInfo>();
  SDValue UId = DAG.getConstant(AFI->createJumpTableUId(), PTy);
  SDValue JTI = DAG.getTargetJumpTable(JT->getIndex(), PTy);
  Table = DAG.getNode(ARMISD::WrapperJT, dl, MVT::i32, JTI, UId);
  Index = DAG.getNode(ISD::MUL, dl, PTy, Index, DAG.getConstant(4, PTy));
  SDValue Addr = DAG.getNode(ISD::ADD, dl, PTy, Index, Table);
  if (Subtarget->isThumb2()) {
    // Thumb2 uses a two-level jump. That is, it jumps into the jump table
    // which does another jump to the destination. This also makes it easier
    // to translate it to TBB / TBH later.
    // FIXME: This might not work if the function is extremely large.
    return DAG.getNode(ARMISD::BR2_JT, dl, MVT::Other, Chain,
                       Addr, Op.getOperand(2), JTI, UId);
  }
  if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
    Addr = DAG.getLoad((EVT)MVT::i32, dl, Chain, Addr,
                       MachinePointerInfo::getJumpTable(),
                       false, false, false, 0);
    Chain = Addr.getValue(1);
    Addr = DAG.getNode(ISD::ADD, dl, PTy, Addr, Table);
    return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI, UId);
  } else {
    Addr = DAG.getLoad(PTy, dl, Chain, Addr,
                       MachinePointerInfo::getJumpTable(),
                       false, false, false, 0);
    Chain = Addr.getValue(1);
    return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI, UId);
  }
}

static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  SDLoc dl(Op);

  if (Op.getValueType().getVectorElementType() == MVT::i32) {
    if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::f32)
      return Op;
    return DAG.UnrollVectorOp(Op.getNode());
  }

  assert(Op.getOperand(0).getValueType() == MVT::v4f32 &&
         "Invalid type for custom lowering!");
  if (VT != MVT::v4i16)
    return DAG.UnrollVectorOp(Op.getNode());

  Op = DAG.getNode(Op.getOpcode(), dl, MVT::v4i32, Op.getOperand(0));
  return DAG.getNode(ISD::TRUNCATE, dl, VT, Op);
}

SDValue ARMTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) const {
  EVT VT = Op.getValueType();
  if (VT.isVector())
    return LowerVectorFP_TO_INT(Op, DAG);

  if (Subtarget->isFPOnlySP() && Op.getOperand(0).getValueType() == MVT::f64) {
    RTLIB::Libcall LC;
    if (Op.getOpcode() == ISD::FP_TO_SINT)
      LC = RTLIB::getFPTOSINT(Op.getOperand(0).getValueType(),
                              Op.getValueType());
    else
      LC = RTLIB::getFPTOUINT(Op.getOperand(0).getValueType(),
                              Op.getValueType());
    return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1,
                       /*isSigned*/ false, SDLoc(Op)).first;
  }

  SDLoc dl(Op);
  unsigned Opc;

  switch (Op.getOpcode()) {
  default: llvm_unreachable("Invalid opcode!");
  case ISD::FP_TO_SINT:
    Opc = ARMISD::FTOSI;
    break;
  case ISD::FP_TO_UINT:
    Opc = ARMISD::FTOUI;
    break;
  }
  Op = DAG.getNode(Opc, dl, MVT::f32, Op.getOperand(0));
  return DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
}

static SDValue LowerVectorINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  SDLoc dl(Op);

  if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::i32) {
    if (VT.getVectorElementType() == MVT::f32)
      return Op;
    return DAG.UnrollVectorOp(Op.getNode());
  }

  assert(Op.getOperand(0).getValueType() == MVT::v4i16 &&
         "Invalid type for custom lowering!");
  if (VT != MVT::v4f32)
    return DAG.UnrollVectorOp(Op.getNode());

  unsigned CastOpc;
  unsigned Opc;
  switch (Op.getOpcode()) {
  default: llvm_unreachable("Invalid opcode!");
  case ISD::SINT_TO_FP:
    CastOpc = ISD::SIGN_EXTEND;
    Opc = ISD::SINT_TO_FP;
    break;
  case ISD::UINT_TO_FP:
    CastOpc = ISD::ZERO_EXTEND;
    Opc = ISD::UINT_TO_FP;
    break;
  }

  Op = DAG.getNode(CastOpc, dl, MVT::v4i32, Op.getOperand(0));
  return DAG.getNode(Opc, dl, VT, Op);
}

SDValue ARMTargetLowering::LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const {
  EVT VT = Op.getValueType();
  if (VT.isVector())
    return LowerVectorINT_TO_FP(Op, DAG);

  if (Subtarget->isFPOnlySP() && Op.getValueType() == MVT::f64) {
    RTLIB::Libcall LC;
    if (Op.getOpcode() == ISD::SINT_TO_FP)
      LC = RTLIB::getSINTTOFP(Op.getOperand(0).getValueType(),
                              Op.getValueType());
    else
      LC = RTLIB::getUINTTOFP(Op.getOperand(0).getValueType(),
                              Op.getValueType());
    return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1,
                       /*isSigned*/ false, SDLoc(Op)).first;
  }

  SDLoc dl(Op);
  unsigned Opc;

  switch (Op.getOpcode()) {
  default: llvm_unreachable("Invalid opcode!");
  case ISD::SINT_TO_FP:
    Opc = ARMISD::SITOF;
    break;
  case ISD::UINT_TO_FP:
    Opc = ARMISD::UITOF;
    break;
  }

  Op = DAG.getNode(ISD::BITCAST, dl, MVT::f32, Op.getOperand(0));
  return DAG.getNode(Opc, dl, VT, Op);
}

SDValue ARMTargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
  // Implement fcopysign with a fabs and a conditional fneg.
  SDValue Tmp0 = Op.getOperand(0);
  SDValue Tmp1 = Op.getOperand(1);
  SDLoc dl(Op);
  EVT VT = Op.getValueType();
  EVT SrcVT = Tmp1.getValueType();
  bool InGPR = Tmp0.getOpcode() == ISD::BITCAST ||
    Tmp0.getOpcode() == ARMISD::VMOVDRR;
  bool UseNEON = !InGPR && Subtarget->hasNEON();

  if (UseNEON) {
    // Use VBSL to copy the sign bit.
    unsigned EncodedVal = ARM_AM::createNEONModImm(0x6, 0x80);
    SDValue Mask = DAG.getNode(ARMISD::VMOVIMM, dl, MVT::v2i32,
                               DAG.getTargetConstant(EncodedVal, MVT::i32));
    EVT OpVT = (VT == MVT::f32) ? MVT::v2i32 : MVT::v1i64;
    if (VT == MVT::f64)
      Mask = DAG.getNode(ARMISD::VSHL, dl, OpVT,
                         DAG.getNode(ISD::BITCAST, dl, OpVT, Mask),
                         DAG.getConstant(32, MVT::i32));
    else /*if (VT == MVT::f32)*/
      Tmp0 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2f32, Tmp0);
    if (SrcVT == MVT::f32) {
      Tmp1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2f32, Tmp1);
      if (VT == MVT::f64)
        Tmp1 = DAG.getNode(ARMISD::VSHL, dl, OpVT,
                           DAG.getNode(ISD::BITCAST, dl, OpVT, Tmp1),
                           DAG.getConstant(32, MVT::i32));
    } else if (VT == MVT::f32)
      Tmp1 = DAG.getNode(ARMISD::VSHRu, dl, MVT::v1i64,
                         DAG.getNode(ISD::BITCAST, dl, MVT::v1i64, Tmp1),
                         DAG.getConstant(32, MVT::i32));
    Tmp0 = DAG.getNode(ISD::BITCAST, dl, OpVT, Tmp0);
    Tmp1 = DAG.getNode(ISD::BITCAST, dl, OpVT, Tmp1);

    SDValue AllOnes = DAG.getTargetConstant(ARM_AM::createNEONModImm(0xe, 0xff),
                                            MVT::i32);
    AllOnes = DAG.getNode(ARMISD::VMOVIMM, dl, MVT::v8i8, AllOnes);
    SDValue MaskNot = DAG.getNode(ISD::XOR, dl, OpVT, Mask,
                                  DAG.getNode(ISD::BITCAST, dl, OpVT, AllOnes));

    SDValue Res = DAG.getNode(ISD::OR, dl, OpVT,
                              DAG.getNode(ISD::AND, dl, OpVT, Tmp1, Mask),
                              DAG.getNode(ISD::AND, dl, OpVT, Tmp0, MaskNot));
    if (VT == MVT::f32) {
      Res = DAG.getNode(ISD::BITCAST, dl, MVT::v2f32, Res);
      Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f32, Res,
                        DAG.getConstant(0, MVT::i32));
    } else {
      Res = DAG.getNode(ISD::BITCAST, dl, MVT::f64, Res);
    }

    return Res;
  }

  // Bitcast operand 1 to i32.
  if (SrcVT == MVT::f64)
    Tmp1 = DAG.getNode(ARMISD::VMOVRRD, dl, DAG.getVTList(MVT::i32, MVT::i32),
                       Tmp1).getValue(1);
  Tmp1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Tmp1);

  // Or in the signbit with integer operations.
  SDValue Mask1 = DAG.getConstant(0x80000000, MVT::i32);
  SDValue Mask2 = DAG.getConstant(0x7fffffff, MVT::i32);
  Tmp1 = DAG.getNode(ISD::AND, dl, MVT::i32, Tmp1, Mask1);
  if (VT == MVT::f32) {
    Tmp0 = DAG.getNode(ISD::AND, dl, MVT::i32,
                       DAG.getNode(ISD::BITCAST, dl, MVT::i32, Tmp0), Mask2);
    return DAG.getNode(ISD::BITCAST, dl, MVT::f32,
                       DAG.getNode(ISD::OR, dl, MVT::i32, Tmp0, Tmp1));
  }

  // f64: Or the high part with signbit and then combine two parts.
  Tmp0 = DAG.getNode(ARMISD::VMOVRRD, dl, DAG.getVTList(MVT::i32, MVT::i32),
                     Tmp0);
  SDValue Lo = Tmp0.getValue(0);
  SDValue Hi = DAG.getNode(ISD::AND, dl, MVT::i32, Tmp0.getValue(1), Mask2);
  Hi = DAG.getNode(ISD::OR, dl, MVT::i32, Hi, Tmp1);
  return DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi);
}

SDValue ARMTargetLowering::LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const{
  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  MFI->setReturnAddressIsTaken(true);

  if (verifyReturnAddressArgumentIsConstant(Op, DAG))
    return SDValue();

  EVT VT = Op.getValueType();
  SDLoc dl(Op);
  unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
  if (Depth) {
    SDValue FrameAddr = LowerFRAMEADDR(Op, DAG);
    SDValue Offset = DAG.getConstant(4, MVT::i32);
    return DAG.getLoad(VT, dl, DAG.getEntryNode(),
                       DAG.getNode(ISD::ADD, dl, VT, FrameAddr, Offset),
                       MachinePointerInfo(), false, false, false, 0);
  }

  // Return LR, which contains the return address. Mark it an implicit live-in.
  unsigned Reg = MF.addLiveIn(ARM::LR, getRegClassFor(MVT::i32));
  return DAG.getCopyFromReg(DAG.getEntryNode(), dl, Reg, VT);
}

SDValue ARMTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
  const ARMBaseRegisterInfo &ARI =
    *static_cast<const ARMBaseRegisterInfo*>(RegInfo);
  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  MFI->setFrameAddressIsTaken(true);

  EVT VT = Op.getValueType();
  SDLoc dl(Op);  // FIXME probably not meaningful
  unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
  unsigned FrameReg = ARI.getFrameRegister(MF);
  SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg, VT);
  while (Depth--)
    FrameAddr = DAG.getLoad(VT, dl, DAG.getEntryNode(), FrameAddr,
                            MachinePointerInfo(),
                            false, false, false, 0);
  return FrameAddr;
}

// FIXME? Maybe this could be a TableGen attribute on some registers and
// this table could be generated automatically from RegInfo.
unsigned ARMTargetLowering::getRegisterByName(const char* RegName,
                                              EVT VT) const {
  unsigned Reg = StringSwitch<unsigned>(RegName)
                       .Case("sp", ARM::SP)
                       .Default(0);
  if (Reg)
    return Reg;
  report_fatal_error("Invalid register name global variable");
}

/// ExpandBITCAST - If the target supports VFP, this function is called to
/// expand a bit convert where either the source or destination type is i64 to
/// use a VMOVDRR or VMOVRRD node.  This should not be done when the non-i64
/// operand type is illegal (e.g., v2f32 for a target that doesn't support
/// vectors), since the legalizer won't know what to do with that.
static SDValue ExpandBITCAST(SDNode *N, SelectionDAG &DAG) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDLoc dl(N);
  SDValue Op = N->getOperand(0);

  // This function is only supposed to be called for i64 types, either as the
  // source or destination of the bit convert.
  EVT SrcVT = Op.getValueType();
  EVT DstVT = N->getValueType(0);
  assert((SrcVT == MVT::i64 || DstVT == MVT::i64) &&
         "ExpandBITCAST called for non-i64 type");

  // Turn i64->f64 into VMOVDRR.
  if (SrcVT == MVT::i64 && TLI.isTypeLegal(DstVT)) {
    SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, Op,
                             DAG.getConstant(0, MVT::i32));
    SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, Op,
                             DAG.getConstant(1, MVT::i32));
    return DAG.getNode(ISD::BITCAST, dl, DstVT,
                       DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi));
  }

  // Turn f64->i64 into VMOVRRD.
  if (DstVT == MVT::i64 && TLI.isTypeLegal(SrcVT)) {
    SDValue Cvt;
    if (TLI.isBigEndian() && SrcVT.isVector() &&
        SrcVT.getVectorNumElements() > 1)
      Cvt = DAG.getNode(ARMISD::VMOVRRD, dl,
                        DAG.getVTList(MVT::i32, MVT::i32),
                        DAG.getNode(ARMISD::VREV64, dl, SrcVT, Op));
    else
      Cvt = DAG.getNode(ARMISD::VMOVRRD, dl,
                        DAG.getVTList(MVT::i32, MVT::i32), Op);
    // Merge the pieces into a single i64 value.
    return DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Cvt, Cvt.getValue(1));
  }

  return SDValue();
}

/// getZeroVector - Returns a vector of specified type with all zero elements.
/// Zero vectors are used to represent vector negation and in those cases
/// will be implemented with the NEON VNEG instruction.  However, VNEG does
/// not support i64 elements, so sometimes the zero vectors will need to be
/// explicitly constructed.  Regardless, use a canonical VMOV to create the
/// zero vector.
static SDValue getZeroVector(EVT VT, SelectionDAG &DAG, SDLoc dl) {
  assert(VT.isVector() && "Expected a vector type");
  // The canonical modified immediate encoding of a zero vector is....0!
  SDValue EncodedVal = DAG.getTargetConstant(0, MVT::i32);
  EVT VmovVT = VT.is128BitVector() ? MVT::v4i32 : MVT::v2i32;
  SDValue Vmov = DAG.getNode(ARMISD::VMOVIMM, dl, VmovVT, EncodedVal);
  return DAG.getNode(ISD::BITCAST, dl, VT, Vmov);
}

/// LowerShiftRightParts - Lower SRA_PARTS, which returns two
/// i32 values and take a 2 x i32 value to shift plus a shift amount.
SDValue ARMTargetLowering::LowerShiftRightParts(SDValue Op,
                                                SelectionDAG &DAG) const {
  assert(Op.getNumOperands() == 3 && "Not a double-shift!");
  EVT VT = Op.getValueType();
  unsigned VTBits = VT.getSizeInBits();
  SDLoc dl(Op);
  SDValue ShOpLo = Op.getOperand(0);
  SDValue ShOpHi = Op.getOperand(1);
  SDValue ShAmt  = Op.getOperand(2);
  SDValue ARMcc;
  unsigned Opc = (Op.getOpcode() == ISD::SRA_PARTS) ? ISD::SRA : ISD::SRL;

  assert(Op.getOpcode() == ISD::SRA_PARTS || Op.getOpcode() == ISD::SRL_PARTS);

  SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
                                 DAG.getConstant(VTBits, MVT::i32), ShAmt);
  SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, ShAmt);
  SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
                                   DAG.getConstant(VTBits, MVT::i32));
  SDValue Tmp2 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, RevShAmt);
  SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
  SDValue TrueVal = DAG.getNode(Opc, dl, VT, ShOpHi, ExtraShAmt);

  SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
  SDValue Cmp = getARMCmp(ExtraShAmt, DAG.getConstant(0, MVT::i32), ISD::SETGE,
                          ARMcc, DAG, dl);
  SDValue Hi = DAG.getNode(Opc, dl, VT, ShOpHi, ShAmt);
  SDValue Lo = DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc,
                           CCR, Cmp);

  SDValue Ops[2] = { Lo, Hi };
  return DAG.getMergeValues(Ops, dl);
}

/// LowerShiftLeftParts - Lower SHL_PARTS, which returns two
/// i32 values and take a 2 x i32 value to shift plus a shift amount.
SDValue ARMTargetLowering::LowerShiftLeftParts(SDValue Op,
                                               SelectionDAG &DAG) const {
  assert(Op.getNumOperands() == 3 && "Not a double-shift!");
  EVT VT = Op.getValueType();
  unsigned VTBits = VT.getSizeInBits();
  SDLoc dl(Op);
  SDValue ShOpLo = Op.getOperand(0);
  SDValue ShOpHi = Op.getOperand(1);
  SDValue ShAmt  = Op.getOperand(2);
  SDValue ARMcc;

  assert(Op.getOpcode() == ISD::SHL_PARTS);
  SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
                                 DAG.getConstant(VTBits, MVT::i32), ShAmt);
  SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, RevShAmt);
  SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
                                   DAG.getConstant(VTBits, MVT::i32));
  SDValue Tmp2 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, ShAmt);
  SDValue Tmp3 = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ExtraShAmt);

  SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
  SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
  SDValue Cmp = getARMCmp(ExtraShAmt, DAG.getConstant(0, MVT::i32), ISD::SETGE,
                          ARMcc, DAG, dl);
  SDValue Lo = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ShAmt);
  SDValue Hi = DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, Tmp3, ARMcc,
                           CCR, Cmp);

  SDValue Ops[2] = { Lo, Hi };
  return DAG.getMergeValues(Ops, dl);
}

SDValue ARMTargetLowering::LowerFLT_ROUNDS_(SDValue Op,
                                            SelectionDAG &DAG) const {
  // The rounding mode is in bits 23:22 of the FPSCR.
  // The ARM rounding mode value to FLT_ROUNDS mapping is 0->1, 1->2, 2->3, 3->0
  // The formula we use to implement this is (((FPSCR + 1 << 22) >> 22) & 3)
  // so that the shift + and get folded into a bitfield extract.
  SDLoc dl(Op);
  SDValue FPSCR = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::i32,
                              DAG.getConstant(Intrinsic::arm_get_fpscr,
                                              MVT::i32));
  SDValue FltRounds = DAG.getNode(ISD::ADD, dl, MVT::i32, FPSCR,
                                  DAG.getConstant(1U << 22, MVT::i32));
  SDValue RMODE = DAG.getNode(ISD::SRL, dl, MVT::i32, FltRounds,
                              DAG.getConstant(22, MVT::i32));
  return DAG.getNode(ISD::AND, dl, MVT::i32, RMODE,
                     DAG.getConstant(3, MVT::i32));
}

static SDValue LowerCTTZ(SDNode *N, SelectionDAG &DAG,
                         const ARMSubtarget *ST) {
  EVT VT = N->getValueType(0);
  SDLoc dl(N);

  if (!ST->hasV6T2Ops())
    return SDValue();

  SDValue rbit = DAG.getNode(ARMISD::RBIT, dl, VT, N->getOperand(0));
  return DAG.getNode(ISD::CTLZ, dl, VT, rbit);
}

/// getCTPOP16BitCounts - Returns a v8i8/v16i8 vector containing the bit-count
/// for each 16-bit element from operand, repeated.  The basic idea is to
/// leverage vcnt to get the 8-bit counts, gather and add the results.
///
/// Trace for v4i16:
/// input    = [v0    v1    v2    v3   ] (vi 16-bit element)
/// cast: N0 = [w0 w1 w2 w3 w4 w5 w6 w7] (v0 = [w0 w1], wi 8-bit element)
/// vcnt: N1 = [b0 b1 b2 b3 b4 b5 b6 b7] (bi = bit-count of 8-bit element wi)
/// vrev: N2 = [b1 b0 b3 b2 b5 b4 b7 b6]
///            [b0 b1 b2 b3 b4 b5 b6 b7]
///           +[b1 b0 b3 b2 b5 b4 b7 b6]
/// N3=N1+N2 = [k0 k0 k1 k1 k2 k2 k3 k3] (k0 = b0+b1 = bit-count of 16-bit v0,
/// vuzp:    = [k0 k1 k2 k3 k0 k1 k2 k3]  each ki is 8-bits)
static SDValue getCTPOP16BitCounts(SDNode *N, SelectionDAG &DAG) {
  EVT VT = N->getValueType(0);
  SDLoc DL(N);

  EVT VT8Bit = VT.is64BitVector() ? MVT::v8i8 : MVT::v16i8;
  SDValue N0 = DAG.getNode(ISD::BITCAST, DL, VT8Bit, N->getOperand(0));
  SDValue N1 = DAG.getNode(ISD::CTPOP, DL, VT8Bit, N0);
  SDValue N2 = DAG.getNode(ARMISD::VREV16, DL, VT8Bit, N1);
  SDValue N3 = DAG.getNode(ISD::ADD, DL, VT8Bit, N1, N2);
  return DAG.getNode(ARMISD::VUZP, DL, VT8Bit, N3, N3);
}

/// lowerCTPOP16BitElements - Returns a v4i16/v8i16 vector containing the
/// bit-count for each 16-bit element from the operand.  We need slightly
/// different sequencing for v4i16 and v8i16 to stay within NEON's available
/// 64/128-bit registers.
///
/// Trace for v4i16:
/// input           = [v0    v1    v2    v3    ] (vi 16-bit element)
/// v8i8: BitCounts = [k0 k1 k2 k3 k0 k1 k2 k3 ] (ki is the bit-count of vi)
/// v8i16:Extended  = [k0    k1    k2    k3    k0    k1    k2    k3    ]
/// v4i16:Extracted = [k0    k1    k2    k3    ]
static SDValue lowerCTPOP16BitElements(SDNode *N, SelectionDAG &DAG) {
  EVT VT = N->getValueType(0);
  SDLoc DL(N);

  SDValue BitCounts = getCTPOP16BitCounts(N, DAG);
  if (VT.is64BitVector()) {
    SDValue Extended = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v8i16, BitCounts);
    return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i16, Extended,
                       DAG.getIntPtrConstant(0));
  } else {
    SDValue Extracted = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v8i8,
                                    BitCounts, DAG.getIntPtrConstant(0));
    return DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v8i16, Extracted);
  }
}

/// lowerCTPOP32BitElements - Returns a v2i32/v4i32 vector containing the
/// bit-count for each 32-bit element from the operand.  The idea here is
/// to split the vector into 16-bit elements, leverage the 16-bit count
/// routine, and then combine the results.
///
/// Trace for v2i32 (v4i32 similar with Extracted/Extended exchanged):
/// input    = [v0    v1    ] (vi: 32-bit elements)
/// Bitcast  = [w0 w1 w2 w3 ] (wi: 16-bit elements, v0 = [w0 w1])
/// Counts16 = [k0 k1 k2 k3 ] (ki: 16-bit elements, bit-count of wi)
/// vrev: N0 = [k1 k0 k3 k2 ]
///            [k0 k1 k2 k3 ]
///       N1 =+[k1 k0 k3 k2 ]
///            [k0 k2 k1 k3 ]
///       N2 =+[k1 k3 k0 k2 ]
///            [k0    k2    k1    k3    ]
/// Extended =+[k1    k3    k0    k2    ]
///            [k0    k2    ]
/// Extracted=+[k1    k3    ]
///
static SDValue lowerCTPOP32BitElements(SDNode *N, SelectionDAG &DAG) {
  EVT VT = N->getValueType(0);
  SDLoc DL(N);

  EVT VT16Bit = VT.is64BitVector() ? MVT::v4i16 : MVT::v8i16;

  SDValue Bitcast = DAG.getNode(ISD::BITCAST, DL, VT16Bit, N->getOperand(0));
  SDValue Counts16 = lowerCTPOP16BitElements(Bitcast.getNode(), DAG);
  SDValue N0 = DAG.getNode(ARMISD::VREV32, DL, VT16Bit, Counts16);
  SDValue N1 = DAG.getNode(ISD::ADD, DL, VT16Bit, Counts16, N0);
  SDValue N2 = DAG.getNode(ARMISD::VUZP, DL, VT16Bit, N1, N1);

  if (VT.is64BitVector()) {
    SDValue Extended = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v4i32, N2);
    return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v2i32, Extended,
                       DAG.getIntPtrConstant(0));
  } else {
    SDValue Extracted = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i16, N2,
                                    DAG.getIntPtrConstant(0));
    return DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v4i32, Extracted);
  }
}

static SDValue LowerCTPOP(SDNode *N, SelectionDAG &DAG,
                          const ARMSubtarget *ST) {
  EVT VT = N->getValueType(0);

  assert(ST->hasNEON() && "Custom ctpop lowering requires NEON.");
  assert((VT == MVT::v2i32 || VT == MVT::v4i32 ||
          VT == MVT::v4i16 || VT == MVT::v8i16) &&
         "Unexpected type for custom ctpop lowering");

  if (VT.getVectorElementType() == MVT::i32)
    return lowerCTPOP32BitElements(N, DAG);
  else
    return lowerCTPOP16BitElements(N, DAG);
}

static SDValue LowerShift(SDNode *N, SelectionDAG &DAG,
                          const ARMSubtarget *ST) {
  EVT VT = N->getValueType(0);
  SDLoc dl(N);

  if (!VT.isVector())
    return SDValue();

  // Lower vector shifts on NEON to use VSHL.
  assert(ST->hasNEON() && "unexpected vector shift");

  // Left shifts translate directly to the vshiftu intrinsic.
  if (N->getOpcode() == ISD::SHL)
    return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT,
                       DAG.getConstant(Intrinsic::arm_neon_vshiftu, MVT::i32),
                       N->getOperand(0), N->getOperand(1));

  assert((N->getOpcode() == ISD::SRA ||
          N->getOpcode() == ISD::SRL) && "unexpected vector shift opcode");

  // NEON uses the same intrinsics for both left and right shifts.  For
  // right shifts, the shift amounts are negative, so negate the vector of
  // shift amounts.
  EVT ShiftVT = N->getOperand(1).getValueType();
  SDValue NegatedCount = DAG.getNode(ISD::SUB, dl, ShiftVT,
                                     getZeroVector(ShiftVT, DAG, dl),
                                     N->getOperand(1));
  Intrinsic::ID vshiftInt = (N->getOpcode() == ISD::SRA ?
                             Intrinsic::arm_neon_vshifts :
                             Intrinsic::arm_neon_vshiftu);
  return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT,
                     DAG.getConstant(vshiftInt, MVT::i32),
                     N->getOperand(0), NegatedCount);
}

static SDValue Expand64BitShift(SDNode *N, SelectionDAG &DAG,
                                const ARMSubtarget *ST) {
  EVT VT = N->getValueType(0);
  SDLoc dl(N);

  // We can get here for a node like i32 = ISD::SHL i32, i64
  if (VT != MVT::i64)
    return SDValue();

  assert((N->getOpcode() == ISD::SRL || N->getOpcode() == ISD::SRA) &&
         "Unknown shift to lower!");

  // We only lower SRA, SRL of 1 here, all others use generic lowering.
  if (!isa<ConstantSDNode>(N->getOperand(1)) ||
      cast<ConstantSDNode>(N->getOperand(1))->getZExtValue() != 1)
    return SDValue();

  // If we are in thumb mode, we don't have RRX.
  if (ST->isThumb1Only()) return SDValue();

  // Okay, we have a 64-bit SRA or SRL of 1.  Lower this to an RRX expr.
  SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, N->getOperand(0),
                           DAG.getConstant(0, MVT::i32));
  SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, N->getOperand(0),
                           DAG.getConstant(1, MVT::i32));

  // First, build a SRA_FLAG/SRL_FLAG op, which shifts the top part by one and
  // captures the result into a carry flag.
  unsigned Opc = N->getOpcode() == ISD::SRL ? ARMISD::SRL_FLAG:ARMISD::SRA_FLAG;
  Hi = DAG.getNode(Opc, dl, DAG.getVTList(MVT::i32, MVT::Glue), Hi);

  // The low part is an ARMISD::RRX operand, which shifts the carry in.
  Lo = DAG.getNode(ARMISD::RRX, dl, MVT::i32, Lo, Hi.getValue(1));

  // Merge the pieces into a single i64 value.
 return DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Lo, Hi);
}

static SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG) {
  SDValue TmpOp0, TmpOp1;
  bool Invert = false;
  bool Swap = false;
  unsigned Opc = 0;

  SDValue Op0 = Op.getOperand(0);
  SDValue Op1 = Op.getOperand(1);
  SDValue CC = Op.getOperand(2);
  EVT VT = Op.getValueType();
  ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
  SDLoc dl(Op);

  if (Op1.getValueType().isFloatingPoint()) {
    switch (SetCCOpcode) {
    default: llvm_unreachable("Illegal FP comparison");
    case ISD::SETUNE:
    case ISD::SETNE:  Invert = true; // Fallthrough
    case ISD::SETOEQ:
    case ISD::SETEQ:  Opc = ARMISD::VCEQ; break;
    case ISD::SETOLT:
    case ISD::SETLT: Swap = true; // Fallthrough
    case ISD::SETOGT:
    case ISD::SETGT:  Opc = ARMISD::VCGT; break;
    case ISD::SETOLE:
    case ISD::SETLE:  Swap = true; // Fallthrough
    case ISD::SETOGE:
    case ISD::SETGE: Opc = ARMISD::VCGE; break;
    case ISD::SETUGE: Swap = true; // Fallthrough
    case ISD::SETULE: Invert = true; Opc = ARMISD::VCGT; break;
    case ISD::SETUGT: Swap = true; // Fallthrough
    case ISD::SETULT: Invert = true; Opc = ARMISD::VCGE; break;
    case ISD::SETUEQ: Invert = true; // Fallthrough
    case ISD::SETONE:
      // Expand this to (OLT | OGT).
      TmpOp0 = Op0;
      TmpOp1 = Op1;
      Opc = ISD::OR;
      Op0 = DAG.getNode(ARMISD::VCGT, dl, VT, TmpOp1, TmpOp0);
      Op1 = DAG.getNode(ARMISD::VCGT, dl, VT, TmpOp0, TmpOp1);
      break;
    case ISD::SETUO: Invert = true; // Fallthrough
    case ISD::SETO:
      // Expand this to (OLT | OGE).
      TmpOp0 = Op0;
      TmpOp1 = Op1;
      Opc = ISD::OR;
      Op0 = DAG.getNode(ARMISD::VCGT, dl, VT, TmpOp1, TmpOp0);
      Op1 = DAG.getNode(ARMISD::VCGE, dl, VT, TmpOp0, TmpOp1);
      break;
    }
  } else {
    // Integer comparisons.
    switch (SetCCOpcode) {
    default: llvm_unreachable("Illegal integer comparison");
    case ISD::SETNE:  Invert = true;
    case ISD::SETEQ:  Opc = ARMISD::VCEQ; break;
    case ISD::SETLT:  Swap = true;
    case ISD::SETGT:  Opc = ARMISD::VCGT; break;
    case ISD::SETLE:  Swap = true;
    case ISD::SETGE:  Opc = ARMISD::VCGE; break;
    case ISD::SETULT: Swap = true;
    case ISD::SETUGT: Opc = ARMISD::VCGTU; break;
    case ISD::SETULE: Swap = true;
    case ISD::SETUGE: Opc = ARMISD::VCGEU; break;
    }

    // Detect VTST (Vector Test Bits) = icmp ne (and (op0, op1), zero).
    if (Opc == ARMISD::VCEQ) {

      SDValue AndOp;
      if (ISD::isBuildVectorAllZeros(Op1.getNode()))
        AndOp = Op0;
      else if (ISD::isBuildVectorAllZeros(Op0.getNode()))
        AndOp = Op1;

      // Ignore bitconvert.
      if (AndOp.getNode() && AndOp.getOpcode() == ISD::BITCAST)
        AndOp = AndOp.getOperand(0);

      if (AndOp.getNode() && AndOp.getOpcode() == ISD::AND) {
        Opc = ARMISD::VTST;
        Op0 = DAG.getNode(ISD::BITCAST, dl, VT, AndOp.getOperand(0));
        Op1 = DAG.getNode(ISD::BITCAST, dl, VT, AndOp.getOperand(1));
        Invert = !Invert;
      }
    }
  }

  if (Swap)
    std::swap(Op0, Op1);

  // If one of the operands is a constant vector zero, attempt to fold the
  // comparison to a specialized compare-against-zero form.
  SDValue SingleOp;
  if (ISD::isBuildVectorAllZeros(Op1.getNode()))
    SingleOp = Op0;
  else if (ISD::isBuildVectorAllZeros(Op0.getNode())) {
    if (Opc == ARMISD::VCGE)
      Opc = ARMISD::VCLEZ;
    else if (Opc == ARMISD::VCGT)
      Opc = ARMISD::VCLTZ;
    SingleOp = Op1;
  }

  SDValue Result;
  if (SingleOp.getNode()) {
    switch (Opc) {
    case ARMISD::VCEQ:
      Result = DAG.getNode(ARMISD::VCEQZ, dl, VT, SingleOp); break;
    case ARMISD::VCGE:
      Result = DAG.getNode(ARMISD::VCGEZ, dl, VT, SingleOp); break;
    case ARMISD::VCLEZ:
      Result = DAG.getNode(ARMISD::VCLEZ, dl, VT, SingleOp); break;
    case ARMISD::VCGT:
      Result = DAG.getNode(ARMISD::VCGTZ, dl, VT, SingleOp); break;
    case ARMISD::VCLTZ:
      Result = DAG.getNode(ARMISD::VCLTZ, dl, VT, SingleOp); break;
    default:
      Result = DAG.getNode(Opc, dl, VT, Op0, Op1);
    }
  } else {
     Result = DAG.getNode(Opc, dl, VT, Op0, Op1);
  }

  if (Invert)
    Result = DAG.getNOT(dl, Result, VT);

  return Result;
}

/// isNEONModifiedImm - Check if the specified splat value corresponds to a
/// valid vector constant for a NEON instruction with a "modified immediate"
/// operand (e.g., VMOV).  If so, return the encoded value.
static SDValue isNEONModifiedImm(uint64_t SplatBits, uint64_t SplatUndef,
                                 unsigned SplatBitSize, SelectionDAG &DAG,
                                 EVT &VT, bool is128Bits, NEONModImmType type) {
  unsigned OpCmode, Imm;

  // SplatBitSize is set to the smallest size that splats the vector, so a
  // zero vector will always have SplatBitSize == 8.  However, NEON modified
  // immediate instructions others than VMOV do not support the 8-bit encoding
  // of a zero vector, and the default encoding of zero is supposed to be the
  // 32-bit version.
  if (SplatBits == 0)
    SplatBitSize = 32;

  switch (SplatBitSize) {
  case 8:
    if (type != VMOVModImm)
      return SDValue();
    // Any 1-byte value is OK.  Op=0, Cmode=1110.
    assert((SplatBits & ~0xff) == 0 && "one byte splat value is too big");
    OpCmode = 0xe;
    Imm = SplatBits;
    VT = is128Bits ? MVT::v16i8 : MVT::v8i8;
    break;

  case 16:
    // NEON's 16-bit VMOV supports splat values where only one byte is nonzero.
    VT = is128Bits ? MVT::v8i16 : MVT::v4i16;
    if ((SplatBits & ~0xff) == 0) {
      // Value = 0x00nn: Op=x, Cmode=100x.
      OpCmode = 0x8;
      Imm = SplatBits;
      break;
    }
    if ((SplatBits & ~0xff00) == 0) {
      // Value = 0xnn00: Op=x, Cmode=101x.
      OpCmode = 0xa;
      Imm = SplatBits >> 8;
      break;
    }
    return SDValue();

  case 32:
    // NEON's 32-bit VMOV supports splat values where:
    // * only one byte is nonzero, or
    // * the least significant byte is 0xff and the second byte is nonzero, or
    // * the least significant 2 bytes are 0xff and the third is nonzero.
    VT = is128Bits ? MVT::v4i32 : MVT::v2i32;
    if ((SplatBits & ~0xff) == 0) {
      // Value = 0x000000nn: Op=x, Cmode=000x.
      OpCmode = 0;
      Imm = SplatBits;
      break;
    }
    if ((SplatBits & ~0xff00) == 0) {
      // Value = 0x0000nn00: Op=x, Cmode=001x.
      OpCmode = 0x2;
      Imm = SplatBits >> 8;
      break;
    }
    if ((SplatBits & ~0xff0000) == 0) {
      // Value = 0x00nn0000: Op=x, Cmode=010x.
      OpCmode = 0x4;
      Imm = SplatBits >> 16;
      break;
    }
    if ((SplatBits & ~0xff000000) == 0) {
      // Value = 0xnn000000: Op=x, Cmode=011x.
      OpCmode = 0x6;
      Imm = SplatBits >> 24;
      break;
    }

    // cmode == 0b1100 and cmode == 0b1101 are not supported for VORR or VBIC
    if (type == OtherModImm) return SDValue();

    if ((SplatBits & ~0xffff) == 0 &&
        ((SplatBits | SplatUndef) & 0xff) == 0xff) {
      // Value = 0x0000nnff: Op=x, Cmode=1100.
      OpCmode = 0xc;
      Imm = SplatBits >> 8;
      break;
    }

    if ((SplatBits & ~0xffffff) == 0 &&
        ((SplatBits | SplatUndef) & 0xffff) == 0xffff) {
      // Value = 0x00nnffff: Op=x, Cmode=1101.
      OpCmode = 0xd;
      Imm = SplatBits >> 16;
      break;
    }

    // Note: there are a few 32-bit splat values (specifically: 00ffff00,
    // ff000000, ff0000ff, and ffff00ff) that are valid for VMOV.I64 but not
    // VMOV.I32.  A (very) minor optimization would be to replicate the value
    // and fall through here to test for a valid 64-bit splat.  But, then the
    // caller would also need to check and handle the change in size.
    return SDValue();

  case 64: {
    if (type != VMOVModImm)
      return SDValue();
    // NEON has a 64-bit VMOV splat where each byte is either 0 or 0xff.
    uint64_t BitMask = 0xff;
    uint64_t Val = 0;
    unsigned ImmMask = 1;
    Imm = 0;
    for (int ByteNum = 0; ByteNum < 8; ++ByteNum) {
      if (((SplatBits | SplatUndef) & BitMask) == BitMask) {
        Val |= BitMask;
        Imm |= ImmMask;
      } else if ((SplatBits & BitMask) != 0) {
        return SDValue();
      }
      BitMask <<= 8;
      ImmMask <<= 1;
    }

    if (DAG.getTargetLoweringInfo().isBigEndian())
      // swap higher and lower 32 bit word
      Imm = ((Imm & 0xf) << 4) | ((Imm & 0xf0) >> 4);

    // Op=1, Cmode=1110.
    OpCmode = 0x1e;
    VT = is128Bits ? MVT::v2i64 : MVT::v1i64;
    break;
  }

  default:
    llvm_unreachable("unexpected size for isNEONModifiedImm");
  }

  unsigned EncodedVal = ARM_AM::createNEONModImm(OpCmode, Imm);
  return DAG.getTargetConstant(EncodedVal, MVT::i32);
}

SDValue ARMTargetLowering::LowerConstantFP(SDValue Op, SelectionDAG &DAG,
                                           const ARMSubtarget *ST) const {
  if (!ST->hasVFP3())
    return SDValue();

  bool IsDouble = Op.getValueType() == MVT::f64;
  ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Op);

  // Use the default (constant pool) lowering for double constants when we have
  // an SP-only FPU
  if (IsDouble && Subtarget->isFPOnlySP())
    return SDValue();

  // Try splatting with a VMOV.f32...
  APFloat FPVal = CFP->getValueAPF();
  int ImmVal = IsDouble ? ARM_AM::getFP64Imm(FPVal) : ARM_AM::getFP32Imm(FPVal);

  if (ImmVal != -1) {
    if (IsDouble || !ST->useNEONForSinglePrecisionFP()) {
      // We have code in place to select a valid ConstantFP already, no need to
      // do any mangling.
      return Op;
    }

    // It's a float and we are trying to use NEON operations where
    // possible. Lower it to a splat followed by an extract.
    SDLoc DL(Op);
    SDValue NewVal = DAG.getTargetConstant(ImmVal, MVT::i32);
    SDValue VecConstant = DAG.getNode(ARMISD::VMOVFPIMM, DL, MVT::v2f32,
                                      NewVal);
    return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecConstant,
                       DAG.getConstant(0, MVT::i32));
  }

  // The rest of our options are NEON only, make sure that's allowed before
  // proceeding..
  if (!ST->hasNEON() || (!IsDouble && !ST->useNEONForSinglePrecisionFP()))
    return SDValue();

  EVT VMovVT;
  uint64_t iVal = FPVal.bitcastToAPInt().getZExtValue();

  // It wouldn't really be worth bothering for doubles except for one very
  // important value, which does happen to match: 0.0. So make sure we don't do
  // anything stupid.
  if (IsDouble && (iVal & 0xffffffff) != (iVal >> 32))
    return SDValue();

  // Try a VMOV.i32 (FIXME: i8, i16, or i64 could work too).
  SDValue NewVal = isNEONModifiedImm(iVal & 0xffffffffU, 0, 32, DAG, VMovVT,
                                     false, VMOVModImm);
  if (NewVal != SDValue()) {
    SDLoc DL(Op);
    SDValue VecConstant = DAG.getNode(ARMISD::VMOVIMM, DL, VMovVT,
                                      NewVal);
    if (IsDouble)
      return DAG.getNode(ISD::BITCAST, DL, MVT::f64, VecConstant);

    // It's a float: cast and extract a vector element.
    SDValue VecFConstant = DAG.getNode(ISD::BITCAST, DL, MVT::v2f32,
                                       VecConstant);
    return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecFConstant,
                       DAG.getConstant(0, MVT::i32));
  }

  // Finally, try a VMVN.i32
  NewVal = isNEONModifiedImm(~iVal & 0xffffffffU, 0, 32, DAG, VMovVT,
                             false, VMVNModImm);
  if (NewVal != SDValue()) {
    SDLoc DL(Op);
    SDValue VecConstant = DAG.getNode(ARMISD::VMVNIMM, DL, VMovVT, NewVal);

    if (IsDouble)
      return DAG.getNode(ISD::BITCAST, DL, MVT::f64, VecConstant);

    // It's a float: cast and extract a vector element.
    SDValue VecFConstant = DAG.getNode(ISD::BITCAST, DL, MVT::v2f32,
                                       VecConstant);
    return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecFConstant,
                       DAG.getConstant(0, MVT::i32));
  }

  return SDValue();
}

// check if an VEXT instruction can handle the shuffle mask when the
// vector sources of the shuffle are the same.
static bool isSingletonVEXTMask(ArrayRef<int> M, EVT VT, unsigned &Imm) {
  unsigned NumElts = VT.getVectorNumElements();

  // Assume that the first shuffle index is not UNDEF.  Fail if it is.
  if (M[0] < 0)
    return false;

  Imm = M[0];

  // If this is a VEXT shuffle, the immediate value is the index of the first
  // element.  The other shuffle indices must be the successive elements after
  // the first one.
  unsigned ExpectedElt = Imm;
  for (unsigned i = 1; i < NumElts; ++i) {
    // Increment the expected index.  If it wraps around, just follow it
    // back to index zero and keep going.
    ++ExpectedElt;
    if (ExpectedElt == NumElts)
      ExpectedElt = 0;

    if (M[i] < 0) continue; // ignore UNDEF indices
    if (ExpectedElt != static_cast<unsigned>(M[i]))
      return false;
  }

  return true;
}


static bool isVEXTMask(ArrayRef<int> M, EVT VT,
                       bool &ReverseVEXT, unsigned &Imm) {
  unsigned NumElts = VT.getVectorNumElements();
  ReverseVEXT = false;

  // Assume that the first shuffle index is not UNDEF.  Fail if it is.
  if (M[0] < 0)
    return false;

  Imm = M[0];

  // If this is a VEXT shuffle, the immediate value is the index of the first
  // element.  The other shuffle indices must be the successive elements after
  // the first one.
  unsigned ExpectedElt = Imm;
  for (unsigned i = 1; i < NumElts; ++i) {
    // Increment the expected index.  If it wraps around, it may still be
    // a VEXT but the source vectors must be swapped.
    ExpectedElt += 1;
    if (ExpectedElt == NumElts * 2) {
      ExpectedElt = 0;
      ReverseVEXT = true;
    }

    if (M[i] < 0) continue; // ignore UNDEF indices
    if (ExpectedElt != static_cast<unsigned>(M[i]))
      return false;
  }

  // Adjust the index value if the source operands will be swapped.
  if (ReverseVEXT)
    Imm -= NumElts;

  return true;
}

/// isVREVMask - Check if a vector shuffle corresponds to a VREV
/// instruction with the specified blocksize.  (The order of the elements
/// within each block of the vector is reversed.)
static bool isVREVMask(ArrayRef<int> M, EVT VT, unsigned BlockSize) {
  assert((BlockSize==16 || BlockSize==32 || BlockSize==64) &&
         "Only possible block sizes for VREV are: 16, 32, 64");

  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
  if (EltSz == 64)
    return false;

  unsigned NumElts = VT.getVectorNumElements();
  unsigned BlockElts = M[0] + 1;
  // If the first shuffle index is UNDEF, be optimistic.
  if (M[0] < 0)
    BlockElts = BlockSize / EltSz;

  if (BlockSize <= EltSz || BlockSize != BlockElts * EltSz)
    return false;

  for (unsigned i = 0; i < NumElts; ++i) {
    if (M[i] < 0) continue; // ignore UNDEF indices
    if ((unsigned) M[i] != (i - i%BlockElts) + (BlockElts - 1 - i%BlockElts))
      return false;
  }

  return true;
}

static bool isVTBLMask(ArrayRef<int> M, EVT VT) {
  // We can handle <8 x i8> vector shuffles. If the index in the mask is out of
  // range, then 0 is placed into the resulting vector. So pretty much any mask
  // of 8 elements can work here.
  return VT == MVT::v8i8 && M.size() == 8;
}

static bool isVTRNMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
  if (EltSz == 64)
    return false;

  unsigned NumElts = VT.getVectorNumElements();
  WhichResult = (M[0] == 0 ? 0 : 1);
  for (unsigned i = 0; i < NumElts; i += 2) {
    if ((M[i] >= 0 && (unsigned) M[i] != i + WhichResult) ||
        (M[i+1] >= 0 && (unsigned) M[i+1] != i + NumElts + WhichResult))
      return false;
  }
  return true;
}

/// isVTRN_v_undef_Mask - Special case of isVTRNMask for canonical form of
/// "vector_shuffle v, v", i.e., "vector_shuffle v, undef".
/// Mask is e.g., <0, 0, 2, 2> instead of <0, 4, 2, 6>.
static bool isVTRN_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult){
  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
  if (EltSz == 64)
    return false;

  unsigned NumElts = VT.getVectorNumElements();
  WhichResult = (M[0] == 0 ? 0 : 1);
  for (unsigned i = 0; i < NumElts; i += 2) {
    if ((M[i] >= 0 && (unsigned) M[i] != i + WhichResult) ||
        (M[i+1] >= 0 && (unsigned) M[i+1] != i + WhichResult))
      return false;
  }
  return true;
}

static bool isVUZPMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
  if (EltSz == 64)
    return false;

  unsigned NumElts = VT.getVectorNumElements();
  WhichResult = (M[0] == 0 ? 0 : 1);
  for (unsigned i = 0; i != NumElts; ++i) {
    if (M[i] < 0) continue; // ignore UNDEF indices
    if ((unsigned) M[i] != 2 * i + WhichResult)
      return false;
  }

  // VUZP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
  if (VT.is64BitVector() && EltSz == 32)
    return false;

  return true;
}

/// isVUZP_v_undef_Mask - Special case of isVUZPMask for canonical form of
/// "vector_shuffle v, v", i.e., "vector_shuffle v, undef".
/// Mask is e.g., <0, 2, 0, 2> instead of <0, 2, 4, 6>,
static bool isVUZP_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult){
  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
  if (EltSz == 64)
    return false;

  unsigned Half = VT.getVectorNumElements() / 2;
  WhichResult = (M[0] == 0 ? 0 : 1);
  for (unsigned j = 0; j != 2; ++j) {
    unsigned Idx = WhichResult;
    for (unsigned i = 0; i != Half; ++i) {
      int MIdx = M[i + j * Half];
      if (MIdx >= 0 && (unsigned) MIdx != Idx)
        return false;
      Idx += 2;
    }
  }

  // VUZP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
  if (VT.is64BitVector() && EltSz == 32)
    return false;

  return true;
}

static bool isVZIPMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
  if (EltSz == 64)
    return false;

  unsigned NumElts = VT.getVectorNumElements();
  WhichResult = (M[0] == 0 ? 0 : 1);
  unsigned Idx = WhichResult * NumElts / 2;
  for (unsigned i = 0; i != NumElts; i += 2) {
    if ((M[i] >= 0 && (unsigned) M[i] != Idx) ||
        (M[i+1] >= 0 && (unsigned) M[i+1] != Idx + NumElts))
      return false;
    Idx += 1;
  }

  // VZIP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
  if (VT.is64BitVector() && EltSz == 32)
    return false;

  return true;
}

/// isVZIP_v_undef_Mask - Special case of isVZIPMask for canonical form of
/// "vector_shuffle v, v", i.e., "vector_shuffle v, undef".
/// Mask is e.g., <0, 0, 1, 1> instead of <0, 4, 1, 5>.
static bool isVZIP_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult){
  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
  if (EltSz == 64)
    return false;

  unsigned NumElts = VT.getVectorNumElements();
  WhichResult = (M[0] == 0 ? 0 : 1);
  unsigned Idx = WhichResult * NumElts / 2;
  for (unsigned i = 0; i != NumElts; i += 2) {
    if ((M[i] >= 0 && (unsigned) M[i] != Idx) ||
        (M[i+1] >= 0 && (unsigned) M[i+1] != Idx))
      return false;
    Idx += 1;
  }

  // VZIP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
  if (VT.is64BitVector() && EltSz == 32)
    return false;

  return true;
}

/// \return true if this is a reverse operation on an vector.
static bool isReverseMask(ArrayRef<int> M, EVT VT) {
  unsigned NumElts = VT.getVectorNumElements();
  // Make sure the mask has the right size.
  if (NumElts != M.size())
      return false;

  // Look for <15, ..., 3, -1, 1, 0>.
  for (unsigned i = 0; i != NumElts; ++i)
    if (M[i] >= 0 && M[i] != (int) (NumElts - 1 - i))
      return false;

  return true;
}

// If N is an integer constant that can be moved into a register in one
// instruction, return an SDValue of such a constant (will become a MOV
// instruction).  Otherwise return null.
static SDValue IsSingleInstrConstant(SDValue N, SelectionDAG &DAG,
                                     const ARMSubtarget *ST, SDLoc dl) {
  uint64_t Val;
  if (!isa<ConstantSDNode>(N))
    return SDValue();
  Val = cast<ConstantSDNode>(N)->getZExtValue();

  if (ST->isThumb1Only()) {
    if (Val <= 255 || ~Val <= 255)
      return DAG.getConstant(Val, MVT::i32);
  } else {
    if (ARM_AM::getSOImmVal(Val) != -1 || ARM_AM::getSOImmVal(~Val) != -1)
      return DAG.getConstant(Val, MVT::i32);
  }
  return SDValue();
}

// If this is a case we can't handle, return null and let the default
// expansion code take care of it.
SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
                                             const ARMSubtarget *ST) const {
  BuildVectorSDNode *BVN = cast<BuildVectorSDNode>(Op.getNode());
  SDLoc dl(Op);
  EVT VT = Op.getValueType();

  APInt SplatBits, SplatUndef;
  unsigned SplatBitSize;
  bool HasAnyUndefs;
  if (BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
    if (SplatBitSize <= 64) {
      // Check if an immediate VMOV works.
      EVT VmovVT;
      SDValue Val = isNEONModifiedImm(SplatBits.getZExtValue(),
                                      SplatUndef.getZExtValue(), SplatBitSize,
                                      DAG, VmovVT, VT.is128BitVector(),
                                      VMOVModImm);
      if (Val.getNode()) {
        SDValue Vmov = DAG.getNode(ARMISD::VMOVIMM, dl, VmovVT, Val);
        return DAG.getNode(ISD::BITCAST, dl, VT, Vmov);
      }

      // Try an immediate VMVN.
      uint64_t NegatedImm = (~SplatBits).getZExtValue();
      Val = isNEONModifiedImm(NegatedImm,
                                      SplatUndef.getZExtValue(), SplatBitSize,
                                      DAG, VmovVT, VT.is128BitVector(),
                                      VMVNModImm);
      if (Val.getNode()) {
        SDValue Vmov = DAG.getNode(ARMISD::VMVNIMM, dl, VmovVT, Val);
        return DAG.getNode(ISD::BITCAST, dl, VT, Vmov);
      }

      // Use vmov.f32 to materialize other v2f32 and v4f32 splats.
      if ((VT == MVT::v2f32 || VT == MVT::v4f32) && SplatBitSize == 32) {
        int ImmVal = ARM_AM::getFP32Imm(SplatBits);
        if (ImmVal != -1) {
          SDValue Val = DAG.getTargetConstant(ImmVal, MVT::i32);
          return DAG.getNode(ARMISD::VMOVFPIMM, dl, VT, Val);
        }
      }
    }
  }

  // Scan through the operands to see if only one value is used.
  //
  // As an optimisation, even if more than one value is used it may be more
  // profitable to splat with one value then change some lanes.
  //
  // Heuristically we decide to do this if the vector has a "dominant" value,
  // defined as splatted to more than half of the lanes.
  unsigned NumElts = VT.getVectorNumElements();
  bool isOnlyLowElement = true;
  bool usesOnlyOneValue = true;
  bool hasDominantValue = false;
  bool isConstant = true;

  // Map of the number of times a particular SDValue appears in the
  // element list.
  DenseMap<SDValue, unsigned> ValueCounts;
  SDValue Value;
  for (unsigned i = 0; i < NumElts; ++i) {
    SDValue V = Op.getOperand(i);
    if (V.getOpcode() == ISD::UNDEF)
      continue;
    if (i > 0)
      isOnlyLowElement = false;
    if (!isa<ConstantFPSDNode>(V) && !isa<ConstantSDNode>(V))
      isConstant = false;

    ValueCounts.insert(std::make_pair(V, 0));
    unsigned &Count = ValueCounts[V];

    // Is this value dominant? (takes up more than half of the lanes)
    if (++Count > (NumElts / 2)) {
      hasDominantValue = true;
      Value = V;
    }
  }
  if (ValueCounts.size() != 1)
    usesOnlyOneValue = false;
  if (!Value.getNode() && ValueCounts.size() > 0)
    Value = ValueCounts.begin()->first;

  if (ValueCounts.size() == 0)
    return DAG.getUNDEF(VT);

  // Loads are better lowered with insert_vector_elt/ARMISD::BUILD_VECTOR.
  // Keep going if we are hitting this case.
  if (isOnlyLowElement && !ISD::isNormalLoad(Value.getNode()))
    return DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value);

  unsigned EltSize = VT.getVectorElementType().getSizeInBits();

  // Use VDUP for non-constant splats.  For f32 constant splats, reduce to
  // i32 and try again.
  if (hasDominantValue && EltSize <= 32) {
    if (!isConstant) {
      SDValue N;

      // If we are VDUPing a value that comes directly from a vector, that will
      // cause an unnecessary move to and from a GPR, where instead we could
      // just use VDUPLANE. We can only do this if the lane being extracted
      // is at a constant index, as the VDUP from lane instructions only have
      // constant-index forms.
      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
          isa<ConstantSDNode>(Value->getOperand(1))) {
        // We need to create a new undef vector to use for the VDUPLANE if the
        // size of the vector from which we get the value is different than the
        // size of the vector that we need to create. We will insert the element
        // such that the register coalescer will remove unnecessary copies.
        if (VT != Value->getOperand(0).getValueType()) {
          ConstantSDNode *constIndex;
          constIndex = dyn_cast<ConstantSDNode>(Value->getOperand(1));
          assert(constIndex && "The index is not a constant!");
          unsigned index = constIndex->getAPIntValue().getLimitedValue() %
                             VT.getVectorNumElements();
          N =  DAG.getNode(ARMISD::VDUPLANE, dl, VT,
                 DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, DAG.getUNDEF(VT),
                        Value, DAG.getConstant(index, MVT::i32)),
                           DAG.getConstant(index, MVT::i32));
        } else
          N = DAG.getNode(ARMISD::VDUPLANE, dl, VT,
                        Value->getOperand(0), Value->getOperand(1));
      } else
        N = DAG.getNode(ARMISD::VDUP, dl, VT, Value);

      if (!usesOnlyOneValue) {
        // The dominant value was splatted as 'N', but we now have to insert
        // all differing elements.
        for (unsigned I = 0; I < NumElts; ++I) {
          if (Op.getOperand(I) == Value)
            continue;
          SmallVector<SDValue, 3> Ops;
          Ops.push_back(N);
          Ops.push_back(Op.getOperand(I));
          Ops.push_back(DAG.getConstant(I, MVT::i32));
          N = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, Ops);
        }
      }
      return N;
    }
    if (VT.getVectorElementType().isFloatingPoint()) {
      SmallVector<SDValue, 8> Ops;
      for (unsigned i = 0; i < NumElts; ++i)
        Ops.push_back(DAG.getNode(ISD::BITCAST, dl, MVT::i32,
                                  Op.getOperand(i)));
      EVT VecVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, NumElts);
      SDValue Val = DAG.getNode(ISD::BUILD_VECTOR, dl, VecVT, Ops);
      Val = LowerBUILD_VECTOR(Val, DAG, ST);
      if (Val.getNode())
        return DAG.getNode(ISD::BITCAST, dl, VT, Val);
    }
    if (usesOnlyOneValue) {
      SDValue Val = IsSingleInstrConstant(Value, DAG, ST, dl);
      if (isConstant && Val.getNode())
        return DAG.getNode(ARMISD::VDUP, dl, VT, Val);
    }
  }

  // If all elements are constants and the case above didn't get hit, fall back
  // to the default expansion, which will generate a load from the constant
  // pool.
  if (isConstant)
    return SDValue();

  // Empirical tests suggest this is rarely worth it for vectors of length <= 2.
  if (NumElts >= 4) {
    SDValue shuffle = ReconstructShuffle(Op, DAG);
    if (shuffle != SDValue())
      return shuffle;
  }

  // Vectors with 32- or 64-bit elements can be built by directly assigning
  // the subregisters.  Lower it to an ARMISD::BUILD_VECTOR so the operands
  // will be legalized.
  if (EltSize >= 32) {
    // Do the expansion with floating-point types, since that is what the VFP
    // registers are defined to use, and since i64 is not legal.
    EVT EltVT = EVT::getFloatingPointVT(EltSize);
    EVT VecVT = EVT::getVectorVT(*DAG.getContext(), EltVT, NumElts);
    SmallVector<SDValue, 8> Ops;
    for (unsigned i = 0; i < NumElts; ++i)
      Ops.push_back(DAG.getNode(ISD::BITCAST, dl, EltVT, Op.getOperand(i)));
    SDValue Val = DAG.getNode(ARMISD::BUILD_VECTOR, dl, VecVT, Ops);
    return DAG.getNode(ISD::BITCAST, dl, VT, Val);
  }

  // If all else fails, just use a sequence of INSERT_VECTOR_ELT when we
  // know the default expansion would otherwise fall back on something even
  // worse. For a vector with one or two non-undef values, that's
  // scalar_to_vector for the elements followed by a shuffle (provided the
  // shuffle is valid for the target) and materialization element by element
  // on the stack followed by a load for everything else.
  if (!isConstant && !usesOnlyOneValue) {
    SDValue Vec = DAG.getUNDEF(VT);
    for (unsigned i = 0 ; i < NumElts; ++i) {
      SDValue V = Op.getOperand(i);
      if (V.getOpcode() == ISD::UNDEF)
        continue;
      SDValue LaneIdx = DAG.getConstant(i, MVT::i32);
      Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, Vec, V, LaneIdx);
    }
    return Vec;
  }

  return SDValue();
}

// Gather data to see if the operation can be modelled as a
// shuffle in combination with VEXTs.
SDValue ARMTargetLowering::ReconstructShuffle(SDValue Op,
                                              SelectionDAG &DAG) const {
  SDLoc dl(Op);
  EVT VT = Op.getValueType();
  unsigned NumElts = VT.getVectorNumElements();

  SmallVector<SDValue, 2> SourceVecs;
  SmallVector<unsigned, 2> MinElts;
  SmallVector<unsigned, 2> MaxElts;

  for (unsigned i = 0; i < NumElts; ++i) {
    SDValue V = Op.getOperand(i);
    if (V.getOpcode() == ISD::UNDEF)
      continue;
    else if (V.getOpcode() != ISD::EXTRACT_VECTOR_ELT) {
      // A shuffle can only come from building a vector from various
      // elements of other vectors.
      return SDValue();
    } else if (V.getOperand(0).getValueType().getVectorElementType() !=
               VT.getVectorElementType()) {
      // This code doesn't know how to handle shuffles where the vector
      // element types do not match (this happens because type legalization
      // promotes the return type of EXTRACT_VECTOR_ELT).
      // FIXME: It might be appropriate to extend this code to handle
      // mismatched types.
      return SDValue();
    }

    // Record this extraction against the appropriate vector if possible...
    SDValue SourceVec = V.getOperand(0);
    // If the element number isn't a constant, we can't effectively
    // analyze what's going on.
    if (!isa<ConstantSDNode>(V.getOperand(1)))
      return SDValue();
    unsigned EltNo = cast<ConstantSDNode>(V.getOperand(1))->getZExtValue();
    bool FoundSource = false;
    for (unsigned j = 0; j < SourceVecs.size(); ++j) {
      if (SourceVecs[j] == SourceVec) {
        if (MinElts[j] > EltNo)
          MinElts[j] = EltNo;
        if (MaxElts[j] < EltNo)
          MaxElts[j] = EltNo;
        FoundSource = true;
        break;
      }
    }

    // Or record a new source if not...
    if (!FoundSource) {
      SourceVecs.push_back(SourceVec);
      MinElts.push_back(EltNo);
      MaxElts.push_back(EltNo);
    }
  }

  // Currently only do something sane when at most two source vectors
  // involved.
  if (SourceVecs.size() > 2)
    return SDValue();

  SDValue ShuffleSrcs[2] = {DAG.getUNDEF(VT), DAG.getUNDEF(VT) };
  int VEXTOffsets[2] = {0, 0};

  // This loop extracts the usage patterns of the source vectors
  // and prepares appropriate SDValues for a shuffle if possible.
  for (unsigned i = 0; i < SourceVecs.size(); ++i) {
    if (SourceVecs[i].getValueType() == VT) {
      // No VEXT necessary
      ShuffleSrcs[i] = SourceVecs[i];
      VEXTOffsets[i] = 0;
      continue;
    } else if (SourceVecs[i].getValueType().getVectorNumElements() < NumElts) {
      // It probably isn't worth padding out a smaller vector just to
      // break it down again in a shuffle.
      return SDValue();
    }

    // Since only 64-bit and 128-bit vectors are legal on ARM and
    // we've eliminated the other cases...
    assert(SourceVecs[i].getValueType().getVectorNumElements() == 2*NumElts &&
           "unexpected vector sizes in ReconstructShuffle");

    if (MaxElts[i] - MinElts[i] >= NumElts) {
      // Span too large for a VEXT to cope
      return SDValue();
    }

    if (MinElts[i] >= NumElts) {
      // The extraction can just take the second half
      VEXTOffsets[i] = NumElts;
      ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
                                   SourceVecs[i],
                                   DAG.getIntPtrConstant(NumElts));
    } else if (MaxElts[i] < NumElts) {
      // The extraction can just take the first half
      VEXTOffsets[i] = 0;
      ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
                                   SourceVecs[i],
                                   DAG.getIntPtrConstant(0));
    } else {
      // An actual VEXT is needed
      VEXTOffsets[i] = MinElts[i];
      SDValue VEXTSrc1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
                                     SourceVecs[i],
                                     DAG.getIntPtrConstant(0));
      SDValue VEXTSrc2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
                                     SourceVecs[i],
                                     DAG.getIntPtrConstant(NumElts));
      ShuffleSrcs[i] = DAG.getNode(ARMISD::VEXT, dl, VT, VEXTSrc1, VEXTSrc2,
                                   DAG.getConstant(VEXTOffsets[i], MVT::i32));
    }
  }

  SmallVector<int, 8> Mask;

  for (unsigned i = 0; i < NumElts; ++i) {
    SDValue Entry = Op.getOperand(i);
    if (Entry.getOpcode() == ISD::UNDEF) {
      Mask.push_back(-1);
      continue;
    }

    SDValue ExtractVec = Entry.getOperand(0);
    int ExtractElt = cast<ConstantSDNode>(Op.getOperand(i)
                                          .getOperand(1))->getSExtValue();
    if (ExtractVec == SourceVecs[0]) {
      Mask.push_back(ExtractElt - VEXTOffsets[0]);
    } else {
      Mask.push_back(ExtractElt + NumElts - VEXTOffsets[1]);
    }
  }

  // Final check before we try to produce nonsense...
  if (isShuffleMaskLegal(Mask, VT))
    return DAG.getVectorShuffle(VT, dl, ShuffleSrcs[0], ShuffleSrcs[1],
                                &Mask[0]);

  return SDValue();
}

/// isShuffleMaskLegal - Targets can use this to indicate that they only
/// support *some* VECTOR_SHUFFLE operations, those with specific masks.
/// By default, if a target supports the VECTOR_SHUFFLE node, all mask values
/// are assumed to be legal.
bool
ARMTargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &M,
                                      EVT VT) const {
  if (VT.getVectorNumElements() == 4 &&
      (VT.is128BitVector() || VT.is64BitVector())) {
    unsigned PFIndexes[4];
    for (unsigned i = 0; i != 4; ++i) {
      if (M[i] < 0)
        PFIndexes[i] = 8;
      else
        PFIndexes[i] = M[i];
    }

    // Compute the index in the perfect shuffle table.
    unsigned PFTableIndex =
      PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
    unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
    unsigned Cost = (PFEntry >> 30);

    if (Cost <= 4)
      return true;
  }

  bool ReverseVEXT;
  unsigned Imm, WhichResult;

  unsigned EltSize = VT.getVectorElementType().getSizeInBits();
  return (EltSize >= 32 ||
          ShuffleVectorSDNode::isSplatMask(&M[0], VT) ||
          isVREVMask(M, VT, 64) ||
          isVREVMask(M, VT, 32) ||
          isVREVMask(M, VT, 16) ||
          isVEXTMask(M, VT, ReverseVEXT, Imm) ||
          isVTBLMask(M, VT) ||
          isVTRNMask(M, VT, WhichResult) ||
          isVUZPMask(M, VT, WhichResult) ||
          isVZIPMask(M, VT, WhichResult) ||
          isVTRN_v_undef_Mask(M, VT, WhichResult) ||
          isVUZP_v_undef_Mask(M, VT, WhichResult) ||
          isVZIP_v_undef_Mask(M, VT, WhichResult) ||
          ((VT == MVT::v8i16 || VT == MVT::v16i8) && isReverseMask(M, VT)));
}

/// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit
/// the specified operations to build the shuffle.
static SDValue GeneratePerfectShuffle(unsigned PFEntry, SDValue LHS,
                                      SDValue RHS, SelectionDAG &DAG,
                                      SDLoc dl) {
  unsigned OpNum = (PFEntry >> 26) & 0x0F;
  unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
  unsigned RHSID = (PFEntry >>  0) & ((1 << 13)-1);

  enum {
    OP_COPY = 0, // Copy, used for things like <u,u,u,3> to say it is <0,1,2,3>
    OP_VREV,
    OP_VDUP0,
    OP_VDUP1,
    OP_VDUP2,
    OP_VDUP3,
    OP_VEXT1,
    OP_VEXT2,
    OP_VEXT3,
    OP_VUZPL, // VUZP, left result
    OP_VUZPR, // VUZP, right result
    OP_VZIPL, // VZIP, left result
    OP_VZIPR, // VZIP, right result
    OP_VTRNL, // VTRN, left result
    OP_VTRNR  // VTRN, right result
  };

  if (OpNum == OP_COPY) {
    if (LHSID == (1*9+2)*9+3) return LHS;
    assert(LHSID == ((4*9+5)*9+6)*9+7 && "Illegal OP_COPY!");
    return RHS;
  }

  SDValue OpLHS, OpRHS;
  OpLHS = GeneratePerfectShuffle(PerfectShuffleTable[LHSID], LHS, RHS, DAG, dl);
  OpRHS = GeneratePerfectShuffle(PerfectShuffleTable[RHSID], LHS, RHS, DAG, dl);
  EVT VT = OpLHS.getValueType();

  switch (OpNum) {
  default: llvm_unreachable("Unknown shuffle opcode!");
  case OP_VREV:
    // VREV divides the vector in half and swaps within the half.
    if (VT.getVectorElementType() == MVT::i32 ||
        VT.getVectorElementType() == MVT::f32)
      return DAG.getNode(ARMISD::VREV64, dl, VT, OpLHS);
    // vrev <4 x i16> -> VREV32
    if (VT.getVectorElementType() == MVT::i16)
      return DAG.getNode(ARMISD::VREV32, dl, VT, OpLHS);
    // vrev <4 x i8> -> VREV16
    assert(VT.getVectorElementType() == MVT::i8);
    return DAG.getNode(ARMISD::VREV16, dl, VT, OpLHS);
  case OP_VDUP0:
  case OP_VDUP1:
  case OP_VDUP2:
  case OP_VDUP3:
    return DAG.getNode(ARMISD::VDUPLANE, dl, VT,
                       OpLHS, DAG.getConstant(OpNum-OP_VDUP0, MVT::i32));
  case OP_VEXT1:
  case OP_VEXT2:
  case OP_VEXT3:
    return DAG.getNode(ARMISD::VEXT, dl, VT,
                       OpLHS, OpRHS,
                       DAG.getConstant(OpNum-OP_VEXT1+1, MVT::i32));
  case OP_VUZPL:
  case OP_VUZPR:
    return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
                       OpLHS, OpRHS).getValue(OpNum-OP_VUZPL);
  case OP_VZIPL:
  case OP_VZIPR:
    return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
                       OpLHS, OpRHS).getValue(OpNum-OP_VZIPL);
  case OP_VTRNL:
  case OP_VTRNR:
    return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
                       OpLHS, OpRHS).getValue(OpNum-OP_VTRNL);
  }
}

static SDValue LowerVECTOR_SHUFFLEv8i8(SDValue Op,
                                       ArrayRef<int> ShuffleMask,
                                       SelectionDAG &DAG) {
  // Check to see if we can use the VTBL instruction.
  SDValue V1 = Op.getOperand(0);
  SDValue V2 = Op.getOperand(1);
  SDLoc DL(Op);

  SmallVector<SDValue, 8> VTBLMask;
  for (ArrayRef<int>::iterator
         I = ShuffleMask.begin(), E = ShuffleMask.end(); I != E; ++I)
    VTBLMask.push_back(DAG.getConstant(*I, MVT::i32));

  if (V2.getNode()->getOpcode() == ISD::UNDEF)
    return DAG.getNode(ARMISD::VTBL1, DL, MVT::v8i8, V1,
                       DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v8i8, VTBLMask));

  return DAG.getNode(ARMISD::VTBL2, DL, MVT::v8i8, V1, V2,
                     DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v8i8, VTBLMask));
}

static SDValue LowerReverse_VECTOR_SHUFFLEv16i8_v8i16(SDValue Op,
                                                      SelectionDAG &DAG) {
  SDLoc DL(Op);
  SDValue OpLHS = Op.getOperand(0);
  EVT VT = OpLHS.getValueType();

  assert((VT == MVT::v8i16 || VT == MVT::v16i8) &&
         "Expect an v8i16/v16i8 type");
  OpLHS = DAG.getNode(ARMISD::VREV64, DL, VT, OpLHS);
  // For a v16i8 type: After the VREV, we have got <8, ...15, 8, ..., 0>. Now,
  // extract the first 8 bytes into the top double word and the last 8 bytes
  // into the bottom double word. The v8i16 case is similar.
  unsigned ExtractNum = (VT == MVT::v16i8) ? 8 : 4;
  return DAG.getNode(ARMISD::VEXT, DL, VT, OpLHS, OpLHS,
                     DAG.getConstant(ExtractNum, MVT::i32));
}

static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
  SDValue V1 = Op.getOperand(0);
  SDValue V2 = Op.getOperand(1);
  SDLoc dl(Op);
  EVT VT = Op.getValueType();
  ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op.getNode());

  // Convert shuffles that are directly supported on NEON to target-specific
  // DAG nodes, instead of keeping them as shuffles and matching them again
  // during code selection.  This is more efficient and avoids the possibility
  // of inconsistencies between legalization and selection.
  // FIXME: floating-point vectors should be canonicalized to integer vectors
  // of the same time so that they get CSEd properly.
  ArrayRef<int> ShuffleMask = SVN->getMask();

  unsigned EltSize = VT.getVectorElementType().getSizeInBits();
  if (EltSize <= 32) {
    if (ShuffleVectorSDNode::isSplatMask(&ShuffleMask[0], VT)) {
      int Lane = SVN->getSplatIndex();
      // If this is undef splat, generate it via "just" vdup, if possible.
      if (Lane == -1) Lane = 0;

      // Test if V1 is a SCALAR_TO_VECTOR.
      if (Lane == 0 && V1.getOpcode() == ISD::SCALAR_TO_VECTOR) {
        return DAG.getNode(ARMISD::VDUP, dl, VT, V1.getOperand(0));
      }
      // Test if V1 is a BUILD_VECTOR which is equivalent to a SCALAR_TO_VECTOR
      // (and probably will turn into a SCALAR_TO_VECTOR once legalization
      // reaches it).
      if (Lane == 0 && V1.getOpcode() == ISD::BUILD_VECTOR &&
          !isa<ConstantSDNode>(V1.getOperand(0))) {
        bool IsScalarToVector = true;
        for (unsigned i = 1, e = V1.getNumOperands(); i != e; ++i)
          if (V1.getOperand(i).getOpcode() != ISD::UNDEF) {
            IsScalarToVector = false;
            break;
          }
        if (IsScalarToVector)
          return DAG.getNode(ARMISD::VDUP, dl, VT, V1.getOperand(0));
      }
      return DAG.getNode(ARMISD::VDUPLANE, dl, VT, V1,
                         DAG.getConstant(Lane, MVT::i32));
    }

    bool ReverseVEXT;
    unsigned Imm;
    if (isVEXTMask(ShuffleMask, VT, ReverseVEXT, Imm)) {
      if (ReverseVEXT)
        std::swap(V1, V2);
      return DAG.getNode(ARMISD::VEXT, dl, VT, V1, V2,
                         DAG.getConstant(Imm, MVT::i32));
    }

    if (isVREVMask(ShuffleMask, VT, 64))
      return DAG.getNode(ARMISD::VREV64, dl, VT, V1);
    if (isVREVMask(ShuffleMask, VT, 32))
      return DAG.getNode(ARMISD::VREV32, dl, VT, V1);
    if (isVREVMask(ShuffleMask, VT, 16))
      return DAG.getNode(ARMISD::VREV16, dl, VT, V1);

    if (V2->getOpcode() == ISD::UNDEF &&
        isSingletonVEXTMask(ShuffleMask, VT, Imm)) {
      return DAG.getNode(ARMISD::VEXT, dl, VT, V1, V1,
                         DAG.getConstant(Imm, MVT::i32));
    }

    // Check for Neon shuffles that modify both input vectors in place.
    // If both results are used, i.e., if there are two shuffles with the same
    // source operands and with masks corresponding to both results of one of
    // these operations, DAG memoization will ensure that a single node is
    // used for both shuffles.
    unsigned WhichResult;
    if (isVTRNMask(ShuffleMask, VT, WhichResult))
      return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
                         V1, V2).getValue(WhichResult);
    if (isVUZPMask(ShuffleMask, VT, WhichResult))
      return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
                         V1, V2).getValue(WhichResult);
    if (isVZIPMask(ShuffleMask, VT, WhichResult))
      return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
                         V1, V2).getValue(WhichResult);

    if (isVTRN_v_undef_Mask(ShuffleMask, VT, WhichResult))
      return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
                         V1, V1).getValue(WhichResult);
    if (isVUZP_v_undef_Mask(ShuffleMask, VT, WhichResult))
      return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
                         V1, V1).getValue(WhichResult);
    if (isVZIP_v_undef_Mask(ShuffleMask, VT, WhichResult))
      return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
                         V1, V1).getValue(WhichResult);
  }

  // If the shuffle is not directly supported and it has 4 elements, use
  // the PerfectShuffle-generated table to synthesize it from other shuffles.
  unsigned NumElts = VT.getVectorNumElements();
  if (NumElts == 4) {
    unsigned PFIndexes[4];
    for (unsigned i = 0; i != 4; ++i) {
      if (ShuffleMask[i] < 0)
        PFIndexes[i] = 8;
      else
        PFIndexes[i] = ShuffleMask[i];
    }

    // Compute the index in the perfect shuffle table.
    unsigned PFTableIndex =
      PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
    unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
    unsigned Cost = (PFEntry >> 30);

    if (Cost <= 4)
      return GeneratePerfectShuffle(PFEntry, V1, V2, DAG, dl);
  }

  // Implement shuffles with 32- or 64-bit elements as ARMISD::BUILD_VECTORs.
  if (EltSize >= 32) {
    // Do the expansion with floating-point types, since that is what the VFP
    // registers are defined to use, and since i64 is not legal.
    EVT EltVT = EVT::getFloatingPointVT(EltSize);
    EVT VecVT = EVT::getVectorVT(*DAG.getContext(), EltVT, NumElts);
    V1 = DAG.getNode(ISD::BITCAST, dl, VecVT, V1);
    V2 = DAG.getNode(ISD::BITCAST, dl, VecVT, V2);
    SmallVector<SDValue, 8> Ops;
    for (unsigned i = 0; i < NumElts; ++i) {
      if (ShuffleMask[i] < 0)
        Ops.push_back(DAG.getUNDEF(EltVT));
      else
        Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT,
                                  ShuffleMask[i] < (int)NumElts ? V1 : V2,
                                  DAG.getConstant(ShuffleMask[i] & (NumElts-1),
                                                  MVT::i32)));
    }
    SDValue Val = DAG.getNode(ARMISD::BUILD_VECTOR, dl, VecVT, Ops);
    return DAG.getNode(ISD::BITCAST, dl, VT, Val);
  }

  if ((VT == MVT::v8i16 || VT == MVT::v16i8) && isReverseMask(ShuffleMask, VT))
    return LowerReverse_VECTOR_SHUFFLEv16i8_v8i16(Op, DAG);

  if (VT == MVT::v8i8) {
    SDValue NewOp = LowerVECTOR_SHUFFLEv8i8(Op, ShuffleMask, DAG);
    if (NewOp.getNode())
      return NewOp;
  }

  return SDValue();
}

static SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
  // INSERT_VECTOR_ELT is legal only for immediate indexes.
  SDValue Lane = Op.getOperand(2);
  if (!isa<ConstantSDNode>(Lane))
    return SDValue();

  return Op;
}

static SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
  // EXTRACT_VECTOR_ELT is legal only for immediate indexes.
  SDValue Lane = Op.getOperand(1);
  if (!isa<ConstantSDNode>(Lane))
    return SDValue();

  SDValue Vec = Op.getOperand(0);
  if (Op.getValueType() == MVT::i32 &&
      Vec.getValueType().getVectorElementType().getSizeInBits() < 32) {
    SDLoc dl(Op);
    return DAG.getNode(ARMISD::VGETLANEu, dl, MVT::i32, Vec, Lane);
  }

  return Op;
}

static SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) {
  // The only time a CONCAT_VECTORS operation can have legal types is when
  // two 64-bit vectors are concatenated to a 128-bit vector.
  assert(Op.getValueType().is128BitVector() && Op.getNumOperands() == 2 &&
         "unexpected CONCAT_VECTORS");
  SDLoc dl(Op);
  SDValue Val = DAG.getUNDEF(MVT::v2f64);
  SDValue Op0 = Op.getOperand(0);
  SDValue Op1 = Op.getOperand(1);
  if (Op0.getOpcode() != ISD::UNDEF)
    Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Val,
                      DAG.getNode(ISD::BITCAST, dl, MVT::f64, Op0),
                      DAG.getIntPtrConstant(0));
  if (Op1.getOpcode() != ISD::UNDEF)
    Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Val,
                      DAG.getNode(ISD::BITCAST, dl, MVT::f64, Op1),
                      DAG.getIntPtrConstant(1));
  return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Val);
}

/// isExtendedBUILD_VECTOR - Check if N is a constant BUILD_VECTOR where each
/// element has been zero/sign-extended, depending on the isSigned parameter,
/// from an integer type half its size.
static bool isExtendedBUILD_VECTOR(SDNode *N, SelectionDAG &DAG,
                                   bool isSigned) {
  // A v2i64 BUILD_VECTOR will have been legalized to a BITCAST from v4i32.
  EVT VT = N->getValueType(0);
  if (VT == MVT::v2i64 && N->getOpcode() == ISD::BITCAST) {
    SDNode *BVN = N->getOperand(0).getNode();
    if (BVN->getValueType(0) != MVT::v4i32 ||
        BVN->getOpcode() != ISD::BUILD_VECTOR)
      return false;
    unsigned LoElt = DAG.getTargetLoweringInfo().isBigEndian() ? 1 : 0;
    unsigned HiElt = 1 - LoElt;
    ConstantSDNode *Lo0 = dyn_cast<ConstantSDNode>(BVN->getOperand(LoElt));
    ConstantSDNode *Hi0 = dyn_cast<ConstantSDNode>(BVN->getOperand(HiElt));
    ConstantSDNode *Lo1 = dyn_cast<ConstantSDNode>(BVN->getOperand(LoElt+2));
    ConstantSDNode *Hi1 = dyn_cast<ConstantSDNode>(BVN->getOperand(HiElt+2));
    if (!Lo0 || !Hi0 || !Lo1 || !Hi1)
      return false;
    if (isSigned) {
      if (Hi0->getSExtValue() == Lo0->getSExtValue() >> 32 &&
          Hi1->getSExtValue() == Lo1->getSExtValue() >> 32)
        return true;
    } else {
      if (Hi0->isNullValue() && Hi1->isNullValue())
        return true;
    }
    return false;
  }

  if (N->getOpcode() != ISD::BUILD_VECTOR)
    return false;

  for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
    SDNode *Elt = N->getOperand(i).getNode();
    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Elt)) {
      unsigned EltSize = VT.getVectorElementType().getSizeInBits();
      unsigned HalfSize = EltSize / 2;
      if (isSigned) {
        if (!isIntN(HalfSize, C->getSExtValue()))
          return false;
      } else {
        if (!isUIntN(HalfSize, C->getZExtValue()))
          return false;
      }
      continue;
    }
    return false;
  }

  return true;
}

/// isSignExtended - Check if a node is a vector value that is sign-extended
/// or a constant BUILD_VECTOR with sign-extended elements.
static bool isSignExtended(SDNode *N, SelectionDAG &DAG) {
  if (N->getOpcode() == ISD::SIGN_EXTEND || ISD::isSEXTLoad(N))
    return true;
  if (isExtendedBUILD_VECTOR(N, DAG, true))
    return true;
  return false;
}

/// isZeroExtended - Check if a node is a vector value that is zero-extended
/// or a constant BUILD_VECTOR with zero-extended elements.
static bool isZeroExtended(SDNode *N, SelectionDAG &DAG) {
  if (N->getOpcode() == ISD::ZERO_EXTEND || ISD::isZEXTLoad(N))
    return true;
  if (isExtendedBUILD_VECTOR(N, DAG, false))
    return true;
  return false;
}

static EVT getExtensionTo64Bits(const EVT &OrigVT) {
  if (OrigVT.getSizeInBits() >= 64)
    return OrigVT;

  assert(OrigVT.isSimple() && "Expecting a simple value type");

  MVT::SimpleValueType OrigSimpleTy = OrigVT.getSimpleVT().SimpleTy;
  switch (OrigSimpleTy) {
  default: llvm_unreachable("Unexpected Vector Type");
  case MVT::v2i8:
  case MVT::v2i16:
     return MVT::v2i32;
  case MVT::v4i8:
    return  MVT::v4i16;
  }
}

/// AddRequiredExtensionForVMULL - Add a sign/zero extension to extend the total
/// value size to 64 bits. We need a 64-bit D register as an operand to VMULL.
/// We insert the required extension here to get the vector to fill a D register.
static SDValue AddRequiredExtensionForVMULL(SDValue N, SelectionDAG &DAG,
                                            const EVT &OrigTy,
                                            const EVT &ExtTy,
                                            unsigned ExtOpcode) {
  // The vector originally had a size of OrigTy. It was then extended to ExtTy.
  // We expect the ExtTy to be 128-bits total. If the OrigTy is less than
  // 64-bits we need to insert a new extension so that it will be 64-bits.
  assert(ExtTy.is128BitVector() && "Unexpected extension size");
  if (OrigTy.getSizeInBits() >= 64)
    return N;

  // Must extend size to at least 64 bits to be used as an operand for VMULL.
  EVT NewVT = getExtensionTo64Bits(OrigTy);

  return DAG.getNode(ExtOpcode, SDLoc(N), NewVT, N);
}

/// SkipLoadExtensionForVMULL - return a load of the original vector size that
/// does not do any sign/zero extension. If the original vector is less
/// than 64 bits, an appropriate extension will be added after the load to
/// reach a total size of 64 bits. We have to add the extension separately
/// because ARM does not have a sign/zero extending load for vectors.
static SDValue SkipLoadExtensionForVMULL(LoadSDNode *LD, SelectionDAG& DAG) {
  EVT ExtendedTy = getExtensionTo64Bits(LD->getMemoryVT());

  // The load already has the right type.
  if (ExtendedTy == LD->getMemoryVT())
    return DAG.getLoad(LD->getMemoryVT(), SDLoc(LD), LD->getChain(),
                LD->getBasePtr(), LD->getPointerInfo(), LD->isVolatile(),
                LD->isNonTemporal(), LD->isInvariant(),
                LD->getAlignment());

  // We need to create a zextload/sextload. We cannot just create a load
  // followed by a zext/zext node because LowerMUL is also run during normal
  // operation legalization where we can't create illegal types.
  return DAG.getExtLoad(LD->getExtensionType(), SDLoc(LD), ExtendedTy,
                        LD->getChain(), LD->getBasePtr(), LD->getPointerInfo(),
                        LD->getMemoryVT(), LD->isVolatile(), LD->isInvariant(),
                        LD->isNonTemporal(), LD->getAlignment());
}

/// SkipExtensionForVMULL - For a node that is a SIGN_EXTEND, ZERO_EXTEND,
/// extending load, or BUILD_VECTOR with extended elements, return the
/// unextended value. The unextended vector should be 64 bits so that it can
/// be used as an operand to a VMULL instruction. If the original vector size
/// before extension is less than 64 bits we add a an extension to resize
/// the vector to 64 bits.
static SDValue SkipExtensionForVMULL(SDNode *N, SelectionDAG &DAG) {
  if (N->getOpcode() == ISD::SIGN_EXTEND || N->getOpcode() == ISD::ZERO_EXTEND)
    return AddRequiredExtensionForVMULL(N->getOperand(0), DAG,
                                        N->getOperand(0)->getValueType(0),
                                        N->getValueType(0),
                                        N->getOpcode());

  if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N))
    return SkipLoadExtensionForVMULL(LD, DAG);

  // Otherwise, the value must be a BUILD_VECTOR.  For v2i64, it will
  // have been legalized as a BITCAST from v4i32.
  if (N->getOpcode() == ISD::BITCAST) {
    SDNode *BVN = N->getOperand(0).getNode();
    assert(BVN->getOpcode() == ISD::BUILD_VECTOR &&
           BVN->getValueType(0) == MVT::v4i32 && "expected v4i32 BUILD_VECTOR");
    unsigned LowElt = DAG.getTargetLoweringInfo().isBigEndian() ? 1 : 0;
    return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), MVT::v2i32,
                       BVN->getOperand(LowElt), BVN->getOperand(LowElt+2));
  }
  // Construct a new BUILD_VECTOR with elements truncated to half the size.
  assert(N->getOpcode() == ISD::BUILD_VECTOR && "expected BUILD_VECTOR");
  EVT VT = N->getValueType(0);
  unsigned EltSize = VT.getVectorElementType().getSizeInBits() / 2;
  unsigned NumElts = VT.getVectorNumElements();
  MVT TruncVT = MVT::getIntegerVT(EltSize);
  SmallVector<SDValue, 8> Ops;
  for (unsigned i = 0; i != NumElts; ++i) {
    ConstantSDNode *C = cast<ConstantSDNode>(N->getOperand(i));
    const APInt &CInt = C->getAPIntValue();
    // Element types smaller than 32 bits are not legal, so use i32 elements.
    // The values are implicitly truncated so sext vs. zext doesn't matter.
    Ops.push_back(DAG.getConstant(CInt.zextOrTrunc(32), MVT::i32));
  }
  return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N),
                     MVT::getVectorVT(TruncVT, NumElts), Ops);
}

static bool isAddSubSExt(SDNode *N, SelectionDAG &DAG) {
  unsigned Opcode = N->getOpcode();
  if (Opcode == ISD::ADD || Opcode == ISD::SUB) {
    SDNode *N0 = N->getOperand(0).getNode();
    SDNode *N1 = N->getOperand(1).getNode();
    return N0->hasOneUse() && N1->hasOneUse() &&
      isSignExtended(N0, DAG) && isSignExtended(N1, DAG);
  }
  return false;
}

static bool isAddSubZExt(SDNode *N, SelectionDAG &DAG) {
  unsigned Opcode = N->getOpcode();
  if (Opcode == ISD::ADD || Opcode == ISD::SUB) {
    SDNode *N0 = N->getOperand(0).getNode();
    SDNode *N1 = N->getOperand(1).getNode();
    return N0->hasOneUse() && N1->hasOneUse() &&
      isZeroExtended(N0, DAG) && isZeroExtended(N1, DAG);
  }
  return false;
}

static SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) {
  // Multiplications are only custom-lowered for 128-bit vectors so that
  // VMULL can be detected.  Otherwise v2i64 multiplications are not legal.
  EVT VT = Op.getValueType();
  assert(VT.is128BitVector() && VT.isInteger() &&
         "unexpected type for custom-lowering ISD::MUL");
  SDNode *N0 = Op.getOperand(0).getNode();
  SDNode *N1 = Op.getOperand(1).getNode();
  unsigned NewOpc = 0;
  bool isMLA = false;
  bool isN0SExt = isSignExtended(N0, DAG);
  bool isN1SExt = isSignExtended(N1, DAG);
  if (isN0SExt && isN1SExt)
    NewOpc = ARMISD::VMULLs;
  else {
    bool isN0ZExt = isZeroExtended(N0, DAG);
    bool isN1ZExt = isZeroExtended(N1, DAG);
    if (isN0ZExt && isN1ZExt)
      NewOpc = ARMISD::VMULLu;
    else if (isN1SExt || isN1ZExt) {
      // Look for (s/zext A + s/zext B) * (s/zext C). We want to turn these
      // into (s/zext A * s/zext C) + (s/zext B * s/zext C)
      if (isN1SExt && isAddSubSExt(N0, DAG)) {
        NewOpc = ARMISD::VMULLs;
        isMLA = true;
      } else if (isN1ZExt && isAddSubZExt(N0, DAG)) {
        NewOpc = ARMISD::VMULLu;
        isMLA = true;
      } else if (isN0ZExt && isAddSubZExt(N1, DAG)) {
        std::swap(N0, N1);
        NewOpc = ARMISD::VMULLu;
        isMLA = true;
      }
    }

    if (!NewOpc) {
      if (VT == MVT::v2i64)
        // Fall through to expand this.  It is not legal.
        return SDValue();
      else
        // Other vector multiplications are legal.
        return Op;
    }
  }

  // Legalize to a VMULL instruction.
  SDLoc DL(Op);
  SDValue Op0;
  SDValue Op1 = SkipExtensionForVMULL(N1, DAG);
  if (!isMLA) {
    Op0 = SkipExtensionForVMULL(N0, DAG);
    assert(Op0.getValueType().is64BitVector() &&
           Op1.getValueType().is64BitVector() &&
           "unexpected types for extended operands to VMULL");
    return DAG.getNode(NewOpc, DL, VT, Op0, Op1);
  }

  // Optimizing (zext A + zext B) * C, to (VMULL A, C) + (VMULL B, C) during
  // isel lowering to take advantage of no-stall back to back vmul + vmla.
  //   vmull q0, d4, d6
  //   vmlal q0, d5, d6
  // is faster than
  //   vaddl q0, d4, d5
  //   vmovl q1, d6
  //   vmul  q0, q0, q1
  SDValue N00 = SkipExtensionForVMULL(N0->getOperand(0).getNode(), DAG);
  SDValue N01 = SkipExtensionForVMULL(N0->getOperand(1).getNode(), DAG);
  EVT Op1VT = Op1.getValueType();
  return DAG.getNode(N0->getOpcode(), DL, VT,
                     DAG.getNode(NewOpc, DL, VT,
                               DAG.getNode(ISD::BITCAST, DL, Op1VT, N00), Op1),
                     DAG.getNode(NewOpc, DL, VT,
                               DAG.getNode(ISD::BITCAST, DL, Op1VT, N01), Op1));
}

static SDValue
LowerSDIV_v4i8(SDValue X, SDValue Y, SDLoc dl, SelectionDAG &DAG) {
  // Convert to float
  // float4 xf = vcvt_f32_s32(vmovl_s16(a.lo));
  // float4 yf = vcvt_f32_s32(vmovl_s16(b.lo));
  X = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, X);
  Y = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, Y);
  X = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, X);
  Y = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, Y);
  // Get reciprocal estimate.
  // float4 recip = vrecpeq_f32(yf);
  Y = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
                   DAG.getConstant(Intrinsic::arm_neon_vrecpe, MVT::i32), Y);
  // Because char has a smaller range than uchar, we can actually get away
  // without any newton steps.  This requires that we use a weird bias
  // of 0xb000, however (again, this has been exhaustively tested).
  // float4 result = as_float4(as_int4(xf*recip) + 0xb000);
  X = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, X, Y);
  X = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, X);
  Y = DAG.getConstant(0xb000, MVT::i32);
  Y = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Y, Y, Y, Y);
  X = DAG.getNode(ISD::ADD, dl, MVT::v4i32, X, Y);
  X = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, X);
  // Convert back to short.
  X = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::v4i32, X);
  X = DAG.getNode(ISD::TRUNCATE, dl, MVT::v4i16, X);
  return X;
}

static SDValue
LowerSDIV_v4i16(SDValue N0, SDValue N1, SDLoc dl, SelectionDAG &DAG) {
  SDValue N2;
  // Convert to float.
  // float4 yf = vcvt_f32_s32(vmovl_s16(y));
  // float4 xf = vcvt_f32_s32(vmovl_s16(x));
  N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, N0);
  N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, N1);
  N0 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N0);
  N1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N1);

  // Use reciprocal estimate and one refinement step.
  // float4 recip = vrecpeq_f32(yf);
  // recip *= vrecpsq_f32(yf, recip);
  N2 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
                   DAG.getConstant(Intrinsic::arm_neon_vrecpe, MVT::i32), N1);
  N1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
                   DAG.getConstant(Intrinsic::arm_neon_vrecps, MVT::i32),
                   N1, N2);
  N2 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N1, N2);
  // Because short has a smaller range than ushort, we can actually get away
  // with only a single newton step.  This requires that we use a weird bias
  // of 89, however (again, this has been exhaustively tested).
  // float4 result = as_float4(as_int4(xf*recip) + 0x89);
  N0 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N0, N2);
  N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, N0);
  N1 = DAG.getConstant(0x89, MVT::i32);
  N1 = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, N1, N1, N1, N1);
  N0 = DAG.getNode(ISD::ADD, dl, MVT::v4i32, N0, N1);
  N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, N0);
  // Convert back to integer and return.
  // return vmovn_s32(vcvt_s32_f32(result));
  N0 = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::v4i32, N0);
  N0 = DAG.getNode(ISD::TRUNCATE, dl, MVT::v4i16, N0);
  return N0;
}

static SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  assert((VT == MVT::v4i16 || VT == MVT::v8i8) &&
         "unexpected type for custom-lowering ISD::SDIV");

  SDLoc dl(Op);
  SDValue N0 = Op.getOperand(0);
  SDValue N1 = Op.getOperand(1);
  SDValue N2, N3;

  if (VT == MVT::v8i8) {
    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v8i16, N0);
    N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v8i16, N1);

    N2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
                     DAG.getIntPtrConstant(4));
    N3 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
                     DAG.getIntPtrConstant(4));
    N0 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
                     DAG.getIntPtrConstant(0));
    N1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
                     DAG.getIntPtrConstant(0));

    N0 = LowerSDIV_v4i8(N0, N1, dl, DAG); // v4i16
    N2 = LowerSDIV_v4i8(N2, N3, dl, DAG); // v4i16

    N0 = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v8i16, N0, N2);
    N0 = LowerCONCAT_VECTORS(N0, DAG);

    N0 = DAG.getNode(ISD::TRUNCATE, dl, MVT::v8i8, N0);
    return N0;
  }
  return LowerSDIV_v4i16(N0, N1, dl, DAG);
}

static SDValue LowerUDIV(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  assert((VT == MVT::v4i16 || VT == MVT::v8i8) &&
         "unexpected type for custom-lowering ISD::UDIV");

  SDLoc dl(Op);
  SDValue N0 = Op.getOperand(0);
  SDValue N1 = Op.getOperand(1);
  SDValue N2, N3;

  if (VT == MVT::v8i8) {
    N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v8i16, N0);
    N1 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v8i16, N1);

    N2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
                     DAG.getIntPtrConstant(4));
    N3 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
                     DAG.getIntPtrConstant(4));
    N0 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
                     DAG.getIntPtrConstant(0));
    N1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
                     DAG.getIntPtrConstant(0));

    N0 = LowerSDIV_v4i16(N0, N1, dl, DAG); // v4i16
    N2 = LowerSDIV_v4i16(N2, N3, dl, DAG); // v4i16

    N0 = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v8i16, N0, N2);
    N0 = LowerCONCAT_VECTORS(N0, DAG);

    N0 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v8i8,
                     DAG.getConstant(Intrinsic::arm_neon_vqmovnsu, MVT::i32),
                     N0);
    return N0;
  }

  // v4i16 sdiv ... Convert to float.
  // float4 yf = vcvt_f32_s32(vmovl_u16(y));
  // float4 xf = vcvt_f32_s32(vmovl_u16(x));
  N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v4i32, N0);
  N1 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v4i32, N1);
  N0 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N0);
  SDValue BN1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N1);

  // Use reciprocal estimate and two refinement steps.
  // float4 recip = vrecpeq_f32(yf);
  // recip *= vrecpsq_f32(yf, recip);
  // recip *= vrecpsq_f32(yf, recip);
  N2 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
                   DAG.getConstant(Intrinsic::arm_neon_vrecpe, MVT::i32), BN1);
  N1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
                   DAG.getConstant(Intrinsic::arm_neon_vrecps, MVT::i32),
                   BN1, N2);
  N2 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N1, N2);
  N1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
                   DAG.getConstant(Intrinsic::arm_neon_vrecps, MVT::i32),
                   BN1, N2);
  N2 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N1, N2);
  // Simply multiplying by the reciprocal estimate can leave us a few ulps
  // too low, so we add 2 ulps (exhaustive testing shows that this is enough,
  // and that it will never cause us to return an answer too large).
  // float4 result = as_float4(as_int4(xf*recip) + 2);
  N0 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N0, N2);
  N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, N0);
  N1 = DAG.getConstant(2, MVT::i32);
  N1 = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, N1, N1, N1, N1);
  N0 = DAG.getNode(ISD::ADD, dl, MVT::v4i32, N0, N1);
  N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, N0);
  // Convert back to integer and return.
  // return vmovn_u32(vcvt_s32_f32(result));
  N0 = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::v4i32, N0);
  N0 = DAG.getNode(ISD::TRUNCATE, dl, MVT::v4i16, N0);
  return N0;
}

static SDValue LowerADDC_ADDE_SUBC_SUBE(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getNode()->getValueType(0);
  SDVTList VTs = DAG.getVTList(VT, MVT::i32);

  unsigned Opc;
  bool ExtraOp = false;
  switch (Op.getOpcode()) {
  default: llvm_unreachable("Invalid code");
  case ISD::ADDC: Opc = ARMISD::ADDC; break;
  case ISD::ADDE: Opc = ARMISD::ADDE; ExtraOp = true; break;
  case ISD::SUBC: Opc = ARMISD::SUBC; break;
  case ISD::SUBE: Opc = ARMISD::SUBE; ExtraOp = true; break;
  }

  if (!ExtraOp)
    return DAG.getNode(Opc, SDLoc(Op), VTs, Op.getOperand(0),
                       Op.getOperand(1));
  return DAG.getNode(Opc, SDLoc(Op), VTs, Op.getOperand(0),
                     Op.getOperand(1), Op.getOperand(2));
}

SDValue ARMTargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const {
  assert(Subtarget->isTargetDarwin());

  // For iOS, we want to call an alternative entry point: __sincos_stret,
  // return values are passed via sret.
  SDLoc dl(Op);
  SDValue Arg = Op.getOperand(0);
  EVT ArgVT = Arg.getValueType();
  Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());

  MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();

  // Pair of floats / doubles used to pass the result.
  StructType *RetTy = StructType::get(ArgTy, ArgTy, nullptr);

  // Create stack object for sret.
  const uint64_t ByteSize = TLI.getDataLayout()->getTypeAllocSize(RetTy);
  const unsigned StackAlign = TLI.getDataLayout()->getPrefTypeAlignment(RetTy);
  int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false);
  SDValue SRet = DAG.getFrameIndex(FrameIdx, TLI.getPointerTy());

  ArgListTy Args;
  ArgListEntry Entry;

  Entry.Node = SRet;
  Entry.Ty = RetTy->getPointerTo();
  Entry.isSExt = false;
  Entry.isZExt = false;
  Entry.isSRet = true;
  Args.push_back(Entry);

  Entry.Node = Arg;
  Entry.Ty = ArgTy;
  Entry.isSExt = false;
  Entry.isZExt = false;
  Args.push_back(Entry);

  const char *LibcallName  = (ArgVT == MVT::f64)
  ? "__sincos_stret" : "__sincosf_stret";
  SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy());

  TargetLowering::CallLoweringInfo CLI(DAG);
  CLI.setDebugLoc(dl).setChain(DAG.getEntryNode())
    .setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()), Callee,
               std::move(Args), 0)
    .setDiscardResult();

  std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);

  SDValue LoadSin = DAG.getLoad(ArgVT, dl, CallResult.second, SRet,
                                MachinePointerInfo(), false, false, false, 0);

  // Address of cos field.
  SDValue Add = DAG.getNode(ISD::ADD, dl, getPointerTy(), SRet,
                            DAG.getIntPtrConstant(ArgVT.getStoreSize()));
  SDValue LoadCos = DAG.getLoad(ArgVT, dl, LoadSin.getValue(1), Add,
                                MachinePointerInfo(), false, false, false, 0);

  SDVTList Tys = DAG.getVTList(ArgVT, ArgVT);
  return DAG.getNode(ISD::MERGE_VALUES, dl, Tys,
                     LoadSin.getValue(0), LoadCos.getValue(0));
}

static SDValue LowerAtomicLoadStore(SDValue Op, SelectionDAG &DAG) {
  // Monotonic load/store is legal for all targets
  if (cast<AtomicSDNode>(Op)->getOrdering() <= Monotonic)
    return Op;

  // Acquire/Release load/store is not legal for targets without a
  // dmb or equivalent available.
  return SDValue();
}

static void ReplaceREADCYCLECOUNTER(SDNode *N,
                                    SmallVectorImpl<SDValue> &Results,
                                    SelectionDAG &DAG,
                                    const ARMSubtarget *Subtarget) {
  SDLoc DL(N);
  SDValue Cycles32, OutChain;

  if (Subtarget->hasPerfMon()) {
    // Under Power Management extensions, the cycle-count is:
    //    mrc p15, #0, <Rt>, c9, c13, #0
    SDValue Ops[] = { N->getOperand(0), // Chain
                      DAG.getConstant(Intrinsic::arm_mrc, MVT::i32),
                      DAG.getConstant(15, MVT::i32),
                      DAG.getConstant(0, MVT::i32),
                      DAG.getConstant(9, MVT::i32),
                      DAG.getConstant(13, MVT::i32),
                      DAG.getConstant(0, MVT::i32)
    };

    Cycles32 = DAG.getNode(ISD::INTRINSIC_W_CHAIN, DL,
                           DAG.getVTList(MVT::i32, MVT::Other), Ops);
    OutChain = Cycles32.getValue(1);
  } else {
    // Intrinsic is defined to return 0 on unsupported platforms. Technically
    // there are older ARM CPUs that have implementation-specific ways of
    // obtaining this information (FIXME!).
    Cycles32 = DAG.getConstant(0, MVT::i32);
    OutChain = DAG.getEntryNode();
  }


  SDValue Cycles64 = DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64,
                                 Cycles32, DAG.getConstant(0, MVT::i32));
  Results.push_back(Cycles64);
  Results.push_back(OutChain);
}

SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
  switch (Op.getOpcode()) {
  default: llvm_unreachable("Don't know how to custom lower this!");
  case ISD::ConstantPool:  return LowerConstantPool(Op, DAG);
  case ISD::BlockAddress:  return LowerBlockAddress(Op, DAG);
  case ISD::GlobalAddress:
    switch (Subtarget->getTargetTriple().getObjectFormat()) {
    default: llvm_unreachable("unknown object format");
    case Triple::COFF:
      return LowerGlobalAddressWindows(Op, DAG);
    case Triple::ELF:
      return LowerGlobalAddressELF(Op, DAG);
    case Triple::MachO:
      return LowerGlobalAddressDarwin(Op, DAG);
    }
  case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
  case ISD::SELECT:        return LowerSELECT(Op, DAG);
  case ISD::SELECT_CC:     return LowerSELECT_CC(Op, DAG);
  case ISD::BR_CC:         return LowerBR_CC(Op, DAG);
  case ISD::BR_JT:         return LowerBR_JT(Op, DAG);
  case ISD::VASTART:       return LowerVASTART(Op, DAG);
  case ISD::ATOMIC_FENCE:  return LowerATOMIC_FENCE(Op, DAG, Subtarget);
  case ISD::PREFETCH:      return LowerPREFETCH(Op, DAG, Subtarget);
  case ISD::SINT_TO_FP:
  case ISD::UINT_TO_FP:    return LowerINT_TO_FP(Op, DAG);
  case ISD::FP_TO_SINT:
  case ISD::FP_TO_UINT:    return LowerFP_TO_INT(Op, DAG);
  case ISD::FCOPYSIGN:     return LowerFCOPYSIGN(Op, DAG);
  case ISD::RETURNADDR:    return LowerRETURNADDR(Op, DAG);
  case ISD::FRAMEADDR:     return LowerFRAMEADDR(Op, DAG);
  case ISD::GLOBAL_OFFSET_TABLE: return LowerGLOBAL_OFFSET_TABLE(Op, DAG);
  case ISD::EH_SJLJ_SETJMP: return LowerEH_SJLJ_SETJMP(Op, DAG);
  case ISD::EH_SJLJ_LONGJMP: return LowerEH_SJLJ_LONGJMP(Op, DAG);
  case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG,
                                                               Subtarget);
  case ISD::BITCAST:       return ExpandBITCAST(Op.getNode(), DAG);
  case ISD::SHL:
  case ISD::SRL:
  case ISD::SRA:           return LowerShift(Op.getNode(), DAG, Subtarget);
  case ISD::SHL_PARTS:     return LowerShiftLeftParts(Op, DAG);
  case ISD::SRL_PARTS:
  case ISD::SRA_PARTS:     return LowerShiftRightParts(Op, DAG);
  case ISD::CTTZ:          return LowerCTTZ(Op.getNode(), DAG, Subtarget);
  case ISD::CTPOP:         return LowerCTPOP(Op.getNode(), DAG, Subtarget);
  case ISD::SETCC:         return LowerVSETCC(Op, DAG);
  case ISD::ConstantFP:    return LowerConstantFP(Op, DAG, Subtarget);
  case ISD::BUILD_VECTOR:  return LowerBUILD_VECTOR(Op, DAG, Subtarget);
  case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
  case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR_ELT(Op, DAG);
  case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
  case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
  case ISD::FLT_ROUNDS_:   return LowerFLT_ROUNDS_(Op, DAG);
  case ISD::MUL:           return LowerMUL(Op, DAG);
  case ISD::SDIV:          return LowerSDIV(Op, DAG);
  case ISD::UDIV:          return LowerUDIV(Op, DAG);
  case ISD::ADDC:
  case ISD::ADDE:
  case ISD::SUBC:
  case ISD::SUBE:          return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
  case ISD::SADDO:
  case ISD::UADDO:
  case ISD::SSUBO:
  case ISD::USUBO:
    return LowerXALUO(Op, DAG);
  case ISD::ATOMIC_LOAD:
  case ISD::ATOMIC_STORE:  return LowerAtomicLoadStore(Op, DAG);
  case ISD::FSINCOS:       return LowerFSINCOS(Op, DAG);
  case ISD::SDIVREM:
  case ISD::UDIVREM:       return LowerDivRem(Op, DAG);
  case ISD::DYNAMIC_STACKALLOC:
    if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
      return LowerDYNAMIC_STACKALLOC(Op, DAG);
    llvm_unreachable("Don't know how to custom lower this!");
  case ISD::FP_ROUND: return LowerFP_ROUND(Op, DAG);
  case ISD::FP_EXTEND: return LowerFP_EXTEND(Op, DAG);
  }
}

/// ReplaceNodeResults - Replace the results of node with an illegal result
/// type with new values built out of custom code.
void ARMTargetLowering::ReplaceNodeResults(SDNode *N,
                                           SmallVectorImpl<SDValue>&Results,
                                           SelectionDAG &DAG) const {
  SDValue Res;
  switch (N->getOpcode()) {
  default:
    llvm_unreachable("Don't know how to custom expand this!");
  case ISD::BITCAST:
    Res = ExpandBITCAST(N, DAG);
    break;
  case ISD::SRL:
  case ISD::SRA:
    Res = Expand64BitShift(N, DAG, Subtarget);
    break;
  case ISD::READCYCLECOUNTER:
    ReplaceREADCYCLECOUNTER(N, Results, DAG, Subtarget);
    return;
  }
  if (Res.getNode())
    Results.push_back(Res);
}

//===----------------------------------------------------------------------===//
//                           ARM Scheduler Hooks
//===----------------------------------------------------------------------===//

/// SetupEntryBlockForSjLj - Insert code into the entry block that creates and
/// registers the function context.
void ARMTargetLowering::
SetupEntryBlockForSjLj(MachineInstr *MI, MachineBasicBlock *MBB,
                       MachineBasicBlock *DispatchBB, int FI) const {
  const TargetInstrInfo *TII =
      getTargetMachine().getSubtargetImpl()->getInstrInfo();
  DebugLoc dl = MI->getDebugLoc();
  MachineFunction *MF = MBB->getParent();
  MachineRegisterInfo *MRI = &MF->getRegInfo();
  MachineConstantPool *MCP = MF->getConstantPool();
  ARMFunctionInfo *AFI = MF->getInfo<ARMFunctionInfo>();
  const Function *F = MF->getFunction();

  bool isThumb = Subtarget->isThumb();
  bool isThumb2 = Subtarget->isThumb2();

  unsigned PCLabelId = AFI->createPICLabelUId();
  unsigned PCAdj = (isThumb || isThumb2) ? 4 : 8;
  ARMConstantPoolValue *CPV =
    ARMConstantPoolMBB::Create(F->getContext(), DispatchBB, PCLabelId, PCAdj);
  unsigned CPI = MCP->getConstantPoolIndex(CPV, 4);

  const TargetRegisterClass *TRC = isThumb ?
    (const TargetRegisterClass*)&ARM::tGPRRegClass :
    (const TargetRegisterClass*)&ARM::GPRRegClass;

  // Grab constant pool and fixed stack memory operands.
  MachineMemOperand *CPMMO =
    MF->getMachineMemOperand(MachinePointerInfo::getConstantPool(),
                             MachineMemOperand::MOLoad, 4, 4);

  MachineMemOperand *FIMMOSt =
    MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI),
                             MachineMemOperand::MOStore, 4, 4);

  // Load the address of the dispatch MBB into the jump buffer.
  if (isThumb2) {
    // Incoming value: jbuf
    //   ldr.n  r5, LCPI1_1
    //   orr    r5, r5, #1
    //   add    r5, pc
    //   str    r5, [$jbuf, #+4] ; &jbuf[1]
    unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2LDRpci), NewVReg1)
                   .addConstantPoolIndex(CPI)
                   .addMemOperand(CPMMO));
    // Set the low bit because of thumb mode.
    unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
    AddDefaultCC(
      AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2ORRri), NewVReg2)
                     .addReg(NewVReg1, RegState::Kill)
                     .addImm(0x01)));
    unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
    BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg3)
      .addReg(NewVReg2, RegState::Kill)
      .addImm(PCLabelId);
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2STRi12))
                   .addReg(NewVReg3, RegState::Kill)
                   .addFrameIndex(FI)
                   .addImm(36)  // &jbuf[1] :: pc
                   .addMemOperand(FIMMOSt));
  } else if (isThumb) {
    // Incoming value: jbuf
    //   ldr.n  r1, LCPI1_4
    //   add    r1, pc
    //   mov    r2, #1
    //   orrs   r1, r2
    //   add    r2, $jbuf, #+4 ; &jbuf[1]
    //   str    r1, [r2]
    unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tLDRpci), NewVReg1)
                   .addConstantPoolIndex(CPI)
                   .addMemOperand(CPMMO));
    unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
    BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg2)
      .addReg(NewVReg1, RegState::Kill)
      .addImm(PCLabelId);
    // Set the low bit because of thumb mode.
    unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tMOVi8), NewVReg3)
                   .addReg(ARM::CPSR, RegState::Define)
                   .addImm(1));
    unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tORR), NewVReg4)
                   .addReg(ARM::CPSR, RegState::Define)
                   .addReg(NewVReg2, RegState::Kill)
                   .addReg(NewVReg3, RegState::Kill));
    unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
    BuildMI(*MBB, MI, dl, TII->get(ARM::tADDframe), NewVReg5)
            .addFrameIndex(FI)
            .addImm(36); // &jbuf[1] :: pc
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tSTRi))
                   .addReg(NewVReg4, RegState::Kill)
                   .addReg(NewVReg5, RegState::Kill)
                   .addImm(0)
                   .addMemOperand(FIMMOSt));
  } else {
    // Incoming value: jbuf
    //   ldr  r1, LCPI1_1
    //   add  r1, pc, r1
    //   str  r1, [$jbuf, #+4] ; &jbuf[1]
    unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::LDRi12),  NewVReg1)
                   .addConstantPoolIndex(CPI)
                   .addImm(0)
                   .addMemOperand(CPMMO));
    unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::PICADD), NewVReg2)
                   .addReg(NewVReg1, RegState::Kill)
                   .addImm(PCLabelId));
    AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::STRi12))
                   .addReg(NewVReg2, RegState::Kill)
                   .addFrameIndex(FI)
                   .addImm(36)  // &jbuf[1] :: pc
                   .addMemOperand(FIMMOSt));
  }
}

MachineBasicBlock *ARMTargetLowering::
EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
  const TargetInstrInfo *TII =
      getTargetMachine().getSubtargetImpl()->getInstrInfo();
  DebugLoc dl = MI->getDebugLoc();
  MachineFunction *MF = MBB->getParent();
  MachineRegisterInfo *MRI = &MF->getRegInfo();
  ARMFunctionInfo *AFI = MF->getInfo<ARMFunctionInfo>();
  MachineFrameInfo *MFI = MF->getFrameInfo();
  int FI = MFI->getFunctionContextIndex();

  const TargetRegisterClass *TRC = Subtarget->isThumb() ?
    (const TargetRegisterClass*)&ARM::tGPRRegClass :
    (const TargetRegisterClass*)&ARM::GPRnopcRegClass;

  // Get a mapping of the call site numbers to all of the landing pads they're
  // associated with.
  DenseMap<unsigned, SmallVector<MachineBasicBlock*, 2> > CallSiteNumToLPad;
  unsigned MaxCSNum = 0;
  MachineModuleInfo &MMI = MF->getMMI();
  for (MachineFunction::iterator BB = MF->begin(), E = MF->end(); BB != E;
       ++BB) {
    if (!BB->isLandingPad()) continue;

    // FIXME: We should assert that the EH_LABEL is the first MI in the landing
    // pad.
    for (MachineBasicBlock::iterator
           II = BB->begin(), IE = BB->end(); II != IE; ++II) {
      if (!II->isEHLabel()) continue;

      MCSymbol *Sym = II->getOperand(0).getMCSymbol();
      if (!MMI.hasCallSiteLandingPad(Sym)) continue;

      SmallVectorImpl<unsigned> &CallSiteIdxs = MMI.getCallSiteLandingPad(Sym);
      for (SmallVectorImpl<unsigned>::iterator
             CSI = CallSiteIdxs.begin(), CSE = CallSiteIdxs.end();
           CSI != CSE; ++CSI) {
        CallSiteNumToLPad[*CSI].push_back(BB);
        MaxCSNum = std::max(MaxCSNum, *CSI);
      }
      break;
    }
  }

  // Get an ordered list of the machine basic blocks for the jump table.
  std::vector<MachineBasicBlock*> LPadList;
  SmallPtrSet<MachineBasicBlock*, 64> InvokeBBs;
  LPadList.reserve(CallSiteNumToLPad.size());
  for (unsigned I = 1; I <= MaxCSNum; ++I) {
    SmallVectorImpl<MachineBasicBlock*> &MBBList = CallSiteNumToLPad[I];
    for (SmallVectorImpl<MachineBasicBlock*>::iterator
           II = MBBList.begin(), IE = MBBList.end(); II != IE; ++II) {
      LPadList.push_back(*II);
      InvokeBBs.insert((*II)->pred_begin(), (*II)->pred_end());
    }
  }

  assert(!LPadList.empty() &&
         "No landing pad destinations for the dispatch jump table!");

  // Create the jump table and associated information.
  MachineJumpTableInfo *JTI =
    MF->getOrCreateJumpTableInfo(MachineJumpTableInfo::EK_Inline);
  unsigned MJTI = JTI->createJumpTableIndex(LPadList);
  unsigned UId = AFI->createJumpTableUId();
  Reloc::Model RelocM = getTargetMachine().getRelocationModel();

  // Create the MBBs for the dispatch code.

  // Shove the dispatch's address into the return slot in the function context.
  MachineBasicBlock *DispatchBB = MF->CreateMachineBasicBlock();
  DispatchBB->setIsLandingPad();

  MachineBasicBlock *TrapBB = MF->CreateMachineBasicBlock();
  unsigned trap_opcode;
  if (Subtarget->isThumb())
    trap_opcode = ARM::tTRAP;
  else
    trap_opcode = Subtarget->useNaClTrap() ? ARM::TRAPNaCl : ARM::TRAP;

  BuildMI(TrapBB, dl, TII->get(trap_opcode));
  DispatchBB->addSuccessor(TrapBB);

  MachineBasicBlock *DispContBB = MF->CreateMachineBasicBlock();
  DispatchBB->addSuccessor(DispContBB);

  // Insert and MBBs.
  MF->insert(MF->end(), DispatchBB);
  MF->insert(MF->end(), DispContBB);
  MF->insert(MF->end(), TrapBB);

  // Insert code into the entry block that creates and registers the function
  // context.
  SetupEntryBlockForSjLj(MI, MBB, DispatchBB, FI);

  MachineMemOperand *FIMMOLd =
    MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI),
                             MachineMemOperand::MOLoad |
                             MachineMemOperand::MOVolatile, 4, 4);

  MachineInstrBuilder MIB;
  MIB = BuildMI(DispatchBB, dl, TII->get(ARM::Int_eh_sjlj_dispatchsetup));

  const ARMBaseInstrInfo *AII = static_cast<const ARMBaseInstrInfo*>(TII);
  const ARMBaseRegisterInfo &RI = AII->getRegisterInfo();

  // Add a register mask with no preserved registers.  This results in all
  // registers being marked as clobbered.
  MIB.addRegMask(RI.getNoPreservedMask());

  unsigned NumLPads = LPadList.size();
  if (Subtarget->isThumb2()) {
    unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2LDRi12), NewVReg1)
                   .addFrameIndex(FI)
                   .addImm(4)
                   .addMemOperand(FIMMOLd));

    if (NumLPads < 256) {
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPri))
                     .addReg(NewVReg1)
                     .addImm(LPadList.size()));
    } else {
      unsigned VReg1 = MRI->createVirtualRegister(TRC);
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2MOVi16), VReg1)
                     .addImm(NumLPads & 0xFFFF));

      unsigned VReg2 = VReg1;
      if ((NumLPads & 0xFFFF0000) != 0) {
        VReg2 = MRI->createVirtualRegister(TRC);
        AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2MOVTi16), VReg2)
                       .addReg(VReg1)
                       .addImm(NumLPads >> 16));
      }

      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPrr))
                     .addReg(NewVReg1)
                     .addReg(VReg2));
    }

    BuildMI(DispatchBB, dl, TII->get(ARM::t2Bcc))
      .addMBB(TrapBB)
      .addImm(ARMCC::HI)
      .addReg(ARM::CPSR);

    unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::t2LEApcrelJT),NewVReg3)
                   .addJumpTableIndex(MJTI)
                   .addImm(UId));

    unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
    AddDefaultCC(
      AddDefaultPred(
        BuildMI(DispContBB, dl, TII->get(ARM::t2ADDrs), NewVReg4)
        .addReg(NewVReg3, RegState::Kill)
        .addReg(NewVReg1)
        .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));

    BuildMI(DispContBB, dl, TII->get(ARM::t2BR_JT))
      .addReg(NewVReg4, RegState::Kill)
      .addReg(NewVReg1)
      .addJumpTableIndex(MJTI)
      .addImm(UId);
  } else if (Subtarget->isThumb()) {
    unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tLDRspi), NewVReg1)
                   .addFrameIndex(FI)
                   .addImm(1)
                   .addMemOperand(FIMMOLd));

    if (NumLPads < 256) {
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPi8))
                     .addReg(NewVReg1)
                     .addImm(NumLPads));
    } else {
      MachineConstantPool *ConstantPool = MF->getConstantPool();
      Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
      const Constant *C = ConstantInt::get(Int32Ty, NumLPads);

      // MachineConstantPool wants an explicit alignment.
      unsigned Align = getDataLayout()->getPrefTypeAlignment(Int32Ty);
      if (Align == 0)
        Align = getDataLayout()->getTypeAllocSize(C->getType());
      unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);

      unsigned VReg1 = MRI->createVirtualRegister(TRC);
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tLDRpci))
                     .addReg(VReg1, RegState::Define)
                     .addConstantPoolIndex(Idx));
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPr))
                     .addReg(NewVReg1)
                     .addReg(VReg1));
    }

    BuildMI(DispatchBB, dl, TII->get(ARM::tBcc))
      .addMBB(TrapBB)
      .addImm(ARMCC::HI)
      .addReg(ARM::CPSR);

    unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLSLri), NewVReg2)
                   .addReg(ARM::CPSR, RegState::Define)
                   .addReg(NewVReg1)
                   .addImm(2));

    unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLEApcrelJT), NewVReg3)
                   .addJumpTableIndex(MJTI)
                   .addImm(UId));

    unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg4)
                   .addReg(ARM::CPSR, RegState::Define)
                   .addReg(NewVReg2, RegState::Kill)
                   .addReg(NewVReg3));

    MachineMemOperand *JTMMOLd =
      MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(),
                               MachineMemOperand::MOLoad, 4, 4);

    unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLDRi), NewVReg5)
                   .addReg(NewVReg4, RegState::Kill)
                   .addImm(0)
                   .addMemOperand(JTMMOLd));

    unsigned NewVReg6 = NewVReg5;
    if (RelocM == Reloc::PIC_) {
      NewVReg6 = MRI->createVirtualRegister(TRC);
      AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6)
                     .addReg(ARM::CPSR, RegState::Define)
                     .addReg(NewVReg5, RegState::Kill)
                     .addReg(NewVReg3));
    }

    BuildMI(DispContBB, dl, TII->get(ARM::tBR_JTr))
      .addReg(NewVReg6, RegState::Kill)
      .addJumpTableIndex(MJTI)
      .addImm(UId);
  } else {
    unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::LDRi12), NewVReg1)
                   .addFrameIndex(FI)
                   .addImm(4)
                   .addMemOperand(FIMMOLd));

    if (NumLPads < 256) {
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPri))
                     .addReg(NewVReg1)
                     .addImm(NumLPads));
    } else if (Subtarget->hasV6T2Ops() && isUInt<16>(NumLPads)) {
      unsigned VReg1 = MRI->createVirtualRegister(TRC);
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::MOVi16), VReg1)
                     .addImm(NumLPads & 0xFFFF));

      unsigned VReg2 = VReg1;
      if ((NumLPads & 0xFFFF0000) != 0) {
        VReg2 = MRI->createVirtualRegister(TRC);
        AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::MOVTi16), VReg2)
                       .addReg(VReg1)
                       .addImm(NumLPads >> 16));
      }

      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPrr))
                     .addReg(NewVReg1)
                     .addReg(VReg2));
    } else {
      MachineConstantPool *ConstantPool = MF->getConstantPool();
      Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
      const Constant *C = ConstantInt::get(Int32Ty, NumLPads);

      // MachineConstantPool wants an explicit alignment.
      unsigned Align = getDataLayout()->getPrefTypeAlignment(Int32Ty);
      if (Align == 0)
        Align = getDataLayout()->getTypeAllocSize(C->getType());
      unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);

      unsigned VReg1 = MRI->createVirtualRegister(TRC);
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::LDRcp))
                     .addReg(VReg1, RegState::Define)
                     .addConstantPoolIndex(Idx)
                     .addImm(0));
      AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPrr))
                     .addReg(NewVReg1)
                     .addReg(VReg1, RegState::Kill));
    }

    BuildMI(DispatchBB, dl, TII->get(ARM::Bcc))
      .addMBB(TrapBB)
      .addImm(ARMCC::HI)
      .addReg(ARM::CPSR);

    unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
    AddDefaultCC(
      AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::MOVsi), NewVReg3)
                     .addReg(NewVReg1)
                     .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));
    unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::LEApcrelJT), NewVReg4)
                   .addJumpTableIndex(MJTI)
                   .addImm(UId));

    MachineMemOperand *JTMMOLd =
      MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(),
                               MachineMemOperand::MOLoad, 4, 4);
    unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
    AddDefaultPred(
      BuildMI(DispContBB, dl, TII->get(ARM::LDRrs), NewVReg5)
      .addReg(NewVReg3, RegState::Kill)
      .addReg(NewVReg4)
      .addImm(0)
      .addMemOperand(JTMMOLd));

    if (RelocM == Reloc::PIC_) {
      BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
        .addReg(NewVReg5, RegState::Kill)
        .addReg(NewVReg4)
        .addJumpTableIndex(MJTI)
        .addImm(UId);
    } else {
      BuildMI(DispContBB, dl, TII->get(ARM::BR_JTr))
        .addReg(NewVReg5, RegState::Kill)
        .addJumpTableIndex(MJTI)
        .addImm(UId);
    }
  }

  // Add the jump table entries as successors to the MBB.
  SmallPtrSet<MachineBasicBlock*, 8> SeenMBBs;
  for (std::vector<MachineBasicBlock*>::iterator
         I = LPadList.begin(), E = LPadList.end(); I != E; ++I) {
    MachineBasicBlock *CurMBB = *I;
    if (SeenMBBs.insert(CurMBB).second)
      DispContBB->addSuccessor(CurMBB);
  }

  // N.B. the order the invoke BBs are processed in doesn't matter here.
  const MCPhysReg *SavedRegs = RI.getCalleeSavedRegs(MF);
  SmallVector<MachineBasicBlock*, 64> MBBLPads;
  for (MachineBasicBlock *BB : InvokeBBs) {

    // Remove the landing pad successor from the invoke block and replace it
    // with the new dispatch block.
    SmallVector<MachineBasicBlock*, 4> Successors(BB->succ_begin(),
                                                  BB->succ_end());
    while (!Successors.empty()) {
      MachineBasicBlock *SMBB = Successors.pop_back_val();
      if (SMBB->isLandingPad()) {
        BB->removeSuccessor(SMBB);
        MBBLPads.push_back(SMBB);
      }
    }

    BB->addSuccessor(DispatchBB);

    // Find the invoke call and mark all of the callee-saved registers as
    // 'implicit defined' so that they're spilled. This prevents code from
    // moving instructions to before the EH block, where they will never be
    // executed.
    for (MachineBasicBlock::reverse_iterator
           II = BB->rbegin(), IE = BB->rend(); II != IE; ++II) {
      if (!II->isCall()) continue;

      DenseMap<unsigned, bool> DefRegs;
      for (MachineInstr::mop_iterator
             OI = II->operands_begin(), OE = II->operands_end();
           OI != OE; ++OI) {
        if (!OI->isReg()) continue;
        DefRegs[OI->getReg()] = true;
      }

      MachineInstrBuilder MIB(*MF, &*II);

      for (unsigned i = 0; SavedRegs[i] != 0; ++i) {
        unsigned Reg = SavedRegs[i];
        if (Subtarget->isThumb2() &&
            !ARM::tGPRRegClass.contains(Reg) &&
            !ARM::hGPRRegClass.contains(Reg))
          continue;
        if (Subtarget->isThumb1Only() && !ARM::tGPRRegClass.contains(Reg))
          continue;
        if (!Subtarget->isThumb() && !ARM::GPRRegClass.contains(Reg))
          continue;
        if (!DefRegs[Reg])
          MIB.addReg(Reg, RegState::ImplicitDefine | RegState::Dead);
      }

      break;
    }
  }

  // Mark all former landing pads as non-landing pads. The dispatch is the only
  // landing pad now.
  for (SmallVectorImpl<MachineBasicBlock*>::iterator
         I = MBBLPads.begin(), E = MBBLPads.end(); I != E; ++I)
    (*I)->setIsLandingPad(false);

  // The instruction is gone now.
  MI->eraseFromParent();

  return MBB;
}

static
MachineBasicBlock *OtherSucc(MachineBasicBlock *MBB, MachineBasicBlock *Succ) {
  for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(),
       E = MBB->succ_end(); I != E; ++I)
    if (*I != Succ)
      return *I;
  llvm_unreachable("Expecting a BB with two successors!");
}

/// Return the load opcode for a given load size. If load size >= 8,
/// neon opcode will be returned.
static unsigned getLdOpcode(unsigned LdSize, bool IsThumb1, bool IsThumb2) {
  if (LdSize >= 8)
    return LdSize == 16 ? ARM::VLD1q32wb_fixed
                        : LdSize == 8 ? ARM::VLD1d32wb_fixed : 0;
  if (IsThumb1)
    return LdSize == 4 ? ARM::tLDRi
                       : LdSize == 2 ? ARM::tLDRHi
                                     : LdSize == 1 ? ARM::tLDRBi : 0;
  if (IsThumb2)
    return LdSize == 4 ? ARM::t2LDR_POST
                       : LdSize == 2 ? ARM::t2LDRH_POST
                                     : LdSize == 1 ? ARM::t2LDRB_POST : 0;
  return LdSize == 4 ? ARM::LDR_POST_IMM
                     : LdSize == 2 ? ARM::LDRH_POST
                                   : LdSize == 1 ? ARM::LDRB_POST_IMM : 0;
}

/// Return the store opcode for a given store size. If store size >= 8,
/// neon opcode will be returned.
static unsigned getStOpcode(unsigned StSize, bool IsThumb1, bool IsThumb2) {
  if (StSize >= 8)
    return StSize == 16 ? ARM::VST1q32wb_fixed
                        : StSize == 8 ? ARM::VST1d32wb_fixed : 0;
  if (IsThumb1)
    return StSize == 4 ? ARM::tSTRi
                       : StSize == 2 ? ARM::tSTRHi
                                     : StSize == 1 ? ARM::tSTRBi : 0;
  if (IsThumb2)
    return StSize == 4 ? ARM::t2STR_POST
                       : StSize == 2 ? ARM::t2STRH_POST
                                     : StSize == 1 ? ARM::t2STRB_POST : 0;
  return StSize == 4 ? ARM::STR_POST_IMM
                     : StSize == 2 ? ARM::STRH_POST
                                   : StSize == 1 ? ARM::STRB_POST_IMM : 0;
}

/// Emit a post-increment load operation with given size. The instructions
/// will be added to BB at Pos.
static void emitPostLd(MachineBasicBlock *BB, MachineInstr *Pos,
                       const TargetInstrInfo *TII, DebugLoc dl,
                       unsigned LdSize, unsigned Data, unsigned AddrIn,
                       unsigned AddrOut, bool IsThumb1, bool IsThumb2) {
  unsigned LdOpc = getLdOpcode(LdSize, IsThumb1, IsThumb2);
  assert(LdOpc != 0 && "Should have a load opcode");
  if (LdSize >= 8) {
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
                       .addReg(AddrOut, RegState::Define).addReg(AddrIn)
                       .addImm(0));
  } else if (IsThumb1) {
    // load + update AddrIn
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
                       .addReg(AddrIn).addImm(0));
    MachineInstrBuilder MIB =
        BuildMI(*BB, Pos, dl, TII->get(ARM::tADDi8), AddrOut);
    MIB = AddDefaultT1CC(MIB);
    MIB.addReg(AddrIn).addImm(LdSize);
    AddDefaultPred(MIB);
  } else if (IsThumb2) {
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
                       .addReg(AddrOut, RegState::Define).addReg(AddrIn)
                       .addImm(LdSize));
  } else { // arm
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
                       .addReg(AddrOut, RegState::Define).addReg(AddrIn)
                       .addReg(0).addImm(LdSize));
  }
}

/// Emit a post-increment store operation with given size. The instructions
/// will be added to BB at Pos.
static void emitPostSt(MachineBasicBlock *BB, MachineInstr *Pos,
                       const TargetInstrInfo *TII, DebugLoc dl,
                       unsigned StSize, unsigned Data, unsigned AddrIn,
                       unsigned AddrOut, bool IsThumb1, bool IsThumb2) {
  unsigned StOpc = getStOpcode(StSize, IsThumb1, IsThumb2);
  assert(StOpc != 0 && "Should have a store opcode");
  if (StSize >= 8) {
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
                       .addReg(AddrIn).addImm(0).addReg(Data));
  } else if (IsThumb1) {
    // store + update AddrIn
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc)).addReg(Data)
                       .addReg(AddrIn).addImm(0));
    MachineInstrBuilder MIB =
        BuildMI(*BB, Pos, dl, TII->get(ARM::tADDi8), AddrOut);
    MIB = AddDefaultT1CC(MIB);
    MIB.addReg(AddrIn).addImm(StSize);
    AddDefaultPred(MIB);
  } else if (IsThumb2) {
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
                       .addReg(Data).addReg(AddrIn).addImm(StSize));
  } else { // arm
    AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
                       .addReg(Data).addReg(AddrIn).addReg(0)
                       .addImm(StSize));
  }
}

MachineBasicBlock *
ARMTargetLowering::EmitStructByval(MachineInstr *MI,
                                   MachineBasicBlock *BB) const {
  // This pseudo instruction has 3 operands: dst, src, size
  // We expand it to a loop if size > Subtarget->getMaxInlineSizeThreshold().
  // Otherwise, we will generate unrolled scalar copies.
  const TargetInstrInfo *TII =
      getTargetMachine().getSubtargetImpl()->getInstrInfo();
  const BasicBlock *LLVM_BB = BB->getBasicBlock();
  MachineFunction::iterator It = BB;
  ++It;

  unsigned dest = MI->getOperand(0).getReg();
  unsigned src = MI->getOperand(1).getReg();
  unsigned SizeVal = MI->getOperand(2).getImm();
  unsigned Align = MI->getOperand(3).getImm();
  DebugLoc dl = MI->getDebugLoc();

  MachineFunction *MF = BB->getParent();
  MachineRegisterInfo &MRI = MF->getRegInfo();
  unsigned UnitSize = 0;
  const TargetRegisterClass *TRC = nullptr;
  const TargetRegisterClass *VecTRC = nullptr;

  bool IsThumb1 = Subtarget->isThumb1Only();
  bool IsThumb2 = Subtarget->isThumb2();

  if (Align & 1) {
    UnitSize = 1;
  } else if (Align & 2) {
    UnitSize = 2;
  } else {
    // Check whether we can use NEON instructions.
    if (!MF->getFunction()->getAttributes().
          hasAttribute(AttributeSet::FunctionIndex,
                       Attribute::NoImplicitFloat) &&
        Subtarget->hasNEON()) {
      if ((Align % 16 == 0) && SizeVal >= 16)
        UnitSize = 16;
      else if ((Align % 8 == 0) && SizeVal >= 8)
        UnitSize = 8;
    }
    // Can't use NEON instructions.
    if (UnitSize == 0)
      UnitSize = 4;
  }

  // Select the correct opcode and register class for unit size load/store
  bool IsNeon = UnitSize >= 8;
  TRC = (IsThumb1 || IsThumb2) ? (const TargetRegisterClass *)&ARM::tGPRRegClass
                               : (const TargetRegisterClass *)&ARM::GPRRegClass;
  if (IsNeon)
    VecTRC = UnitSize == 16
                 ? (const TargetRegisterClass *)&ARM::DPairRegClass
                 : UnitSize == 8
                       ? (const TargetRegisterClass *)&ARM::DPRRegClass
                       : nullptr;

  unsigned BytesLeft = SizeVal % UnitSize;
  unsigned LoopSize = SizeVal - BytesLeft;

  if (SizeVal <= Subtarget->getMaxInlineSizeThreshold()) {
    // Use LDR and STR to copy.
    // [scratch, srcOut] = LDR_POST(srcIn, UnitSize)
    // [destOut] = STR_POST(scratch, destIn, UnitSize)
    unsigned srcIn = src;
    unsigned destIn = dest;
    for (unsigned i = 0; i < LoopSize; i+=UnitSize) {
      unsigned srcOut = MRI.createVirtualRegister(TRC);
      unsigned destOut = MRI.createVirtualRegister(TRC);
      unsigned scratch = MRI.createVirtualRegister(IsNeon ? VecTRC : TRC);
      emitPostLd(BB, MI, TII, dl, UnitSize, scratch, srcIn, srcOut,
                 IsThumb1, IsThumb2);
      emitPostSt(BB, MI, TII, dl, UnitSize, scratch, destIn, destOut,
                 IsThumb1, IsThumb2);
      srcIn = srcOut;
      destIn = destOut;
    }

    // Handle the leftover bytes with LDRB and STRB.
    // [scratch, srcOut] = LDRB_POST(srcIn, 1)
    // [destOut] = STRB_POST(scratch, destIn, 1)
    for (unsigned i = 0; i < BytesLeft; i++) {
      unsigned srcOut = MRI.createVirtualRegister(TRC);
      unsigned destOut = MRI.createVirtualRegister(TRC);
      unsigned scratch = MRI.createVirtualRegister(TRC);
      emitPostLd(BB, MI, TII, dl, 1, scratch, srcIn, srcOut,
                 IsThumb1, IsThumb2);
      emitPostSt(BB, MI, TII, dl, 1, scratch, destIn, destOut,
                 IsThumb1, IsThumb2);
      srcIn = srcOut;
      destIn = destOut;
    }
    MI->eraseFromParent();   // The instruction is gone now.
    return BB;
  }

  // Expand the pseudo op to a loop.
  // thisMBB:
  //   ...
  //   movw varEnd, # --> with thumb2
  //   movt varEnd, #
  //   ldrcp varEnd, idx --> without thumb2
  //   fallthrough --> loopMBB
  // loopMBB:
  //   PHI varPhi, varEnd, varLoop
  //   PHI srcPhi, src, srcLoop
  //   PHI destPhi, dst, destLoop
  //   [scratch, srcLoop] = LDR_POST(srcPhi, UnitSize)
  //   [destLoop] = STR_POST(scratch, destPhi, UnitSize)
  //   subs varLoop, varPhi, #UnitSize
  //   bne loopMBB
  //   fallthrough --> exitMBB
  // exitMBB:
  //   epilogue to handle left-over bytes
  //   [scratch, srcOut] = LDRB_POST(srcLoop, 1)
  //   [destOut] = STRB_POST(scratch, destLoop, 1)
  MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
  MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
  MF->insert(It, loopMBB);
  MF->insert(It, exitMBB);

  // Transfer the remainder of BB and its successor edges to exitMBB.
  exitMBB->splice(exitMBB->begin(), BB,
                  std::next(MachineBasicBlock::iterator(MI)), BB->end());
  exitMBB->transferSuccessorsAndUpdatePHIs(BB);

  // Load an immediate to varEnd.
  unsigned varEnd = MRI.createVirtualRegister(TRC);
  if (IsThumb2) {
    unsigned Vtmp = varEnd;
    if ((LoopSize & 0xFFFF0000) != 0)
      Vtmp = MRI.createVirtualRegister(TRC);
    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2MOVi16), Vtmp)
                       .addImm(LoopSize & 0xFFFF));

    if ((LoopSize & 0xFFFF0000) != 0)
      AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2MOVTi16), varEnd)
                         .addReg(Vtmp).addImm(LoopSize >> 16));
  } else {
    MachineConstantPool *ConstantPool = MF->getConstantPool();
    Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
    const Constant *C = ConstantInt::get(Int32Ty, LoopSize);

    // MachineConstantPool wants an explicit alignment.
    unsigned Align = getDataLayout()->getPrefTypeAlignment(Int32Ty);
    if (Align == 0)
      Align = getDataLayout()->getTypeAllocSize(C->getType());
    unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);

    if (IsThumb1)
      AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::tLDRpci)).addReg(
          varEnd, RegState::Define).addConstantPoolIndex(Idx));
    else
      AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::LDRcp)).addReg(
          varEnd, RegState::Define).addConstantPoolIndex(Idx).addImm(0));
  }
  BB->addSuccessor(loopMBB);

  // Generate the loop body:
  //   varPhi = PHI(varLoop, varEnd)
  //   srcPhi = PHI(srcLoop, src)
  //   destPhi = PHI(destLoop, dst)
  MachineBasicBlock *entryBB = BB;
  BB = loopMBB;
  unsigned varLoop = MRI.createVirtualRegister(TRC);
  unsigned varPhi = MRI.createVirtualRegister(TRC);
  unsigned srcLoop = MRI.createVirtualRegister(TRC);
  unsigned srcPhi = MRI.createVirtualRegister(TRC);
  unsigned destLoop = MRI.createVirtualRegister(TRC);
  unsigned destPhi = MRI.createVirtualRegister(TRC);

  BuildMI(*BB, BB->begin(), dl, TII->get(ARM::PHI), varPhi)
    .addReg(varLoop).addMBB(loopMBB)
    .addReg(varEnd).addMBB(entryBB);
  BuildMI(BB, dl, TII->get(ARM::PHI), srcPhi)
    .addReg(srcLoop).addMBB(loopMBB)
    .addReg(src).addMBB(entryBB);
  BuildMI(BB, dl, TII->get(ARM::PHI), destPhi)
    .addReg(destLoop).addMBB(loopMBB)
    .addReg(dest).addMBB(entryBB);

  //   [scratch, srcLoop] = LDR_POST(srcPhi, UnitSize)
  //   [destLoop] = STR_POST(scratch, destPhi, UnitSiz)
  unsigned scratch = MRI.createVirtualRegister(IsNeon ? VecTRC : TRC);
  emitPostLd(BB, BB->end(), TII, dl, UnitSize, scratch, srcPhi, srcLoop,
             IsThumb1, IsThumb2);
  emitPostSt(BB, BB->end(), TII, dl, UnitSize, scratch, destPhi, destLoop,
             IsThumb1, IsThumb2);

  // Decrement loop variable by UnitSize.
  if (IsThumb1) {
    MachineInstrBuilder MIB =
        BuildMI(*BB, BB->end(), dl, TII->get(ARM::tSUBi8), varLoop);
    MIB = AddDefaultT1CC(MIB);
    MIB.addReg(varPhi).addImm(UnitSize);
    AddDefaultPred(MIB);
  } else {
    MachineInstrBuilder MIB =
        BuildMI(*BB, BB->end(), dl,
                TII->get(IsThumb2 ? ARM::t2SUBri : ARM::SUBri), varLoop);
    AddDefaultCC(AddDefaultPred(MIB.addReg(varPhi).addImm(UnitSize)));
    MIB->getOperand(5).setReg(ARM::CPSR);
    MIB->getOperand(5).setIsDef(true);
  }
  BuildMI(*BB, BB->end(), dl,
          TII->get(IsThumb1 ? ARM::tBcc : IsThumb2 ? ARM::t2Bcc : ARM::Bcc))
      .addMBB(loopMBB).addImm(ARMCC::NE).addReg(ARM::CPSR);

  // loopMBB can loop back to loopMBB or fall through to exitMBB.
  BB->addSuccessor(loopMBB);
  BB->addSuccessor(exitMBB);

  // Add epilogue to handle BytesLeft.
  BB = exitMBB;
  MachineInstr *StartOfExit = exitMBB->begin();

  //   [scratch, srcOut] = LDRB_POST(srcLoop, 1)
  //   [destOut] = STRB_POST(scratch, destLoop, 1)
  unsigned srcIn = srcLoop;
  unsigned destIn = destLoop;
  for (unsigned i = 0; i < BytesLeft; i++) {
    unsigned srcOut = MRI.createVirtualRegister(TRC);
    unsigned destOut = MRI.createVirtualRegister(TRC);
    unsigned scratch = MRI.createVirtualRegister(TRC);
    emitPostLd(BB, StartOfExit, TII, dl, 1, scratch, srcIn, srcOut,
               IsThumb1, IsThumb2);
    emitPostSt(BB, StartOfExit, TII, dl, 1, scratch, destIn, destOut,
               IsThumb1, IsThumb2);
    srcIn = srcOut;
    destIn = destOut;
  }

  MI->eraseFromParent();   // The instruction is gone now.
  return BB;
}

MachineBasicBlock *
ARMTargetLowering::EmitLowered__chkstk(MachineInstr *MI,
                                       MachineBasicBlock *MBB) const {
  const TargetMachine &TM = getTargetMachine();
  const TargetInstrInfo &TII = *TM.getSubtargetImpl()->getInstrInfo();
  DebugLoc DL = MI->getDebugLoc();

  assert(Subtarget->isTargetWindows() &&
         "__chkstk is only supported on Windows");
  assert(Subtarget->isThumb2() && "Windows on ARM requires Thumb-2 mode");

  // __chkstk takes the number of words to allocate on the stack in R4, and
  // returns the stack adjustment in number of bytes in R4.  This will not
  // clober any other registers (other than the obvious lr).
  //
  // Although, technically, IP should be considered a register which may be
  // clobbered, the call itself will not touch it.  Windows on ARM is a pure
  // thumb-2 environment, so there is no interworking required.  As a result, we
  // do not expect a veneer to be emitted by the linker, clobbering IP.
  //
  // Each module receives its own copy of __chkstk, so no import thunk is
  // required, again, ensuring that IP is not clobbered.
  //
  // Finally, although some linkers may theoretically provide a trampoline for
  // out of range calls (which is quite common due to a 32M range limitation of
  // branches for Thumb), we can generate the long-call version via
  // -mcmodel=large, alleviating the need for the trampoline which may clobber
  // IP.

  switch (TM.getCodeModel()) {
  case CodeModel::Small:
  case CodeModel::Medium:
  case CodeModel::Default:
  case CodeModel::Kernel:
    BuildMI(*MBB, MI, DL, TII.get(ARM::tBL))
      .addImm((unsigned)ARMCC::AL).addReg(0)
      .addExternalSymbol("__chkstk")
      .addReg(ARM::R4, RegState::Implicit | RegState::Kill)
      .addReg(ARM::R4, RegState::Implicit | RegState::Define)
      .addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
    break;
  case CodeModel::Large:
  case CodeModel::JITDefault: {
    MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
    unsigned Reg = MRI.createVirtualRegister(&ARM::rGPRRegClass);

    BuildMI(*MBB, MI, DL, TII.get(ARM::t2MOVi32imm), Reg)
      .addExternalSymbol("__chkstk");
    BuildMI(*MBB, MI, DL, TII.get(ARM::tBLXr))
      .addImm((unsigned)ARMCC::AL).addReg(0)
      .addReg(Reg, RegState::Kill)
      .addReg(ARM::R4, RegState::Implicit | RegState::Kill)
      .addReg(ARM::R4, RegState::Implicit | RegState::Define)
      .addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
    break;
  }
  }

  AddDefaultCC(AddDefaultPred(BuildMI(*MBB, MI, DL, TII.get(ARM::t2SUBrr),
                                      ARM::SP)
                              .addReg(ARM::SP).addReg(ARM::R4)));

  MI->eraseFromParent();
  return MBB;
}

MachineBasicBlock *
ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
                                               MachineBasicBlock *BB) const {
  const TargetInstrInfo *TII =
      getTargetMachine().getSubtargetImpl()->getInstrInfo();
  DebugLoc dl = MI->getDebugLoc();
  bool isThumb2 = Subtarget->isThumb2();
  switch (MI->getOpcode()) {
  default: {
    MI->dump();
    llvm_unreachable("Unexpected instr type to insert");
  }
  // The Thumb2 pre-indexed stores have the same MI operands, they just
  // define them differently in the .td files from the isel patterns, so
  // they need pseudos.
  case ARM::t2STR_preidx:
    MI->setDesc(TII->get(ARM::t2STR_PRE));
    return BB;
  case ARM::t2STRB_preidx:
    MI->setDesc(TII->get(ARM::t2STRB_PRE));
    return BB;
  case ARM::t2STRH_preidx:
    MI->setDesc(TII->get(ARM::t2STRH_PRE));
    return BB;

  case ARM::STRi_preidx:
  case ARM::STRBi_preidx: {
    unsigned NewOpc = MI->getOpcode() == ARM::STRi_preidx ?
      ARM::STR_PRE_IMM : ARM::STRB_PRE_IMM;
    // Decode the offset.
    unsigned Offset = MI->getOperand(4).getImm();
    bool isSub = ARM_AM::getAM2Op(Offset) == ARM_AM::sub;
    Offset = ARM_AM::getAM2Offset(Offset);
    if (isSub)
      Offset = -Offset;

    MachineMemOperand *MMO = *MI->memoperands_begin();
    BuildMI(*BB, MI, dl, TII->get(NewOpc))
      .addOperand(MI->getOperand(0))  // Rn_wb
      .addOperand(MI->getOperand(1))  // Rt
      .addOperand(MI->getOperand(2))  // Rn
      .addImm(Offset)                 // offset (skip GPR==zero_reg)
      .addOperand(MI->getOperand(5))  // pred
      .addOperand(MI->getOperand(6))
      .addMemOperand(MMO);
    MI->eraseFromParent();
    return BB;
  }
  case ARM::STRr_preidx:
  case ARM::STRBr_preidx:
  case ARM::STRH_preidx: {
    unsigned NewOpc;
    switch (MI->getOpcode()) {
    default: llvm_unreachable("unexpected opcode!");
    case ARM::STRr_preidx: NewOpc = ARM::STR_PRE_REG; break;
    case ARM::STRBr_preidx: NewOpc = ARM::STRB_PRE_REG; break;
    case ARM::STRH_preidx: NewOpc = ARM::STRH_PRE; break;
    }
    MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(NewOpc));
    for (unsigned i = 0; i < MI->getNumOperands(); ++i)
      MIB.addOperand(MI->getOperand(i));
    MI->eraseFromParent();
    return BB;
  }

  case ARM::tMOVCCr_pseudo: {
    // To "insert" a SELECT_CC instruction, we actually have to insert the
    // diamond control-flow pattern.  The incoming instruction knows the
    // destination vreg to set, the condition code register to branch on, the
    // true/false values to select between, and a branch opcode to use.
    const BasicBlock *LLVM_BB = BB->getBasicBlock();
    MachineFunction::iterator It = BB;
    ++It;

    //  thisMBB:
    //  ...
    //   TrueVal = ...
    //   cmpTY ccX, r1, r2
    //   bCC copy1MBB
    //   fallthrough --> copy0MBB
    MachineBasicBlock *thisMBB  = BB;
    MachineFunction *F = BB->getParent();
    MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
    MachineBasicBlock *sinkMBB  = F->CreateMachineBasicBlock(LLVM_BB);
    F->insert(It, copy0MBB);
    F->insert(It, sinkMBB);

    // Transfer the remainder of BB and its successor edges to sinkMBB.
    sinkMBB->splice(sinkMBB->begin(), BB,
                    std::next(MachineBasicBlock::iterator(MI)), BB->end());
    sinkMBB->transferSuccessorsAndUpdatePHIs(BB);

    BB->addSuccessor(copy0MBB);
    BB->addSuccessor(sinkMBB);

    BuildMI(BB, dl, TII->get(ARM::tBcc)).addMBB(sinkMBB)
      .addImm(MI->getOperand(3).getImm()).addReg(MI->getOperand(4).getReg());

    //  copy0MBB:
    //   %FalseValue = ...
    //   # fallthrough to sinkMBB
    BB = copy0MBB;

    // Update machine-CFG edges
    BB->addSuccessor(sinkMBB);

    //  sinkMBB:
    //   %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
    //  ...
    BB = sinkMBB;
    BuildMI(*BB, BB->begin(), dl,
            TII->get(ARM::PHI), MI->getOperand(0).getReg())
      .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB)
      .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);

    MI->eraseFromParent();   // The pseudo instruction is gone now.
    return BB;
  }

  case ARM::BCCi64:
  case ARM::BCCZi64: {
    // If there is an unconditional branch to the other successor, remove it.
    BB->erase(std::next(MachineBasicBlock::iterator(MI)), BB->end());

    // Compare both parts that make up the double comparison separately for
    // equality.
    bool RHSisZero = MI->getOpcode() == ARM::BCCZi64;

    unsigned LHS1 = MI->getOperand(1).getReg();
    unsigned LHS2 = MI->getOperand(2).getReg();
    if (RHSisZero) {
      AddDefaultPred(BuildMI(BB, dl,
                             TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
                     .addReg(LHS1).addImm(0));
      BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
        .addReg(LHS2).addImm(0)
        .addImm(ARMCC::EQ).addReg(ARM::CPSR);
    } else {
      unsigned RHS1 = MI->getOperand(3).getReg();
      unsigned RHS2 = MI->getOperand(4).getReg();
      AddDefaultPred(BuildMI(BB, dl,
                             TII->get(isThumb2 ? ARM::t2CMPrr : ARM::CMPrr))
                     .addReg(LHS1).addReg(RHS1));
      BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPrr : ARM::CMPrr))
        .addReg(LHS2).addReg(RHS2)
        .addImm(ARMCC::EQ).addReg(ARM::CPSR);
    }

    MachineBasicBlock *destMBB = MI->getOperand(RHSisZero ? 3 : 5).getMBB();
    MachineBasicBlock *exitMBB = OtherSucc(BB, destMBB);
    if (MI->getOperand(0).getImm() == ARMCC::NE)
      std::swap(destMBB, exitMBB);

    BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
      .addMBB(destMBB).addImm(ARMCC::EQ).addReg(ARM::CPSR);
    if (isThumb2)
      AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2B)).addMBB(exitMBB));
    else
      BuildMI(BB, dl, TII->get(ARM::B)) .addMBB(exitMBB);

    MI->eraseFromParent();   // The pseudo instruction is gone now.
    return BB;
  }

  case ARM::Int_eh_sjlj_setjmp:
  case ARM::Int_eh_sjlj_setjmp_nofp:
  case ARM::tInt_eh_sjlj_setjmp:
  case ARM::t2Int_eh_sjlj_setjmp:
  case ARM::t2Int_eh_sjlj_setjmp_nofp:
    EmitSjLjDispatchBlock(MI, BB);
    return BB;

  case ARM::ABS:
  case ARM::t2ABS: {
    // To insert an ABS instruction, we have to insert the
    // diamond control-flow pattern.  The incoming instruction knows the
    // source vreg to test against 0, the destination vreg to set,
    // the condition code register to branch on, the
    // true/false values to select between, and a branch opcode to use.
    // It transforms
    //     V1 = ABS V0
    // into
    //     V2 = MOVS V0
    //     BCC                      (branch to SinkBB if V0 >= 0)
    //     RSBBB: V3 = RSBri V2, 0  (compute ABS if V2 < 0)
    //     SinkBB: V1 = PHI(V2, V3)
    const BasicBlock *LLVM_BB = BB->getBasicBlock();
    MachineFunction::iterator BBI = BB;
    ++BBI;
    MachineFunction *Fn = BB->getParent();
    MachineBasicBlock *RSBBB = Fn->CreateMachineBasicBlock(LLVM_BB);
    MachineBasicBlock *SinkBB  = Fn->CreateMachineBasicBlock(LLVM_BB);
    Fn->insert(BBI, RSBBB);
    Fn->insert(BBI, SinkBB);

    unsigned int ABSSrcReg = MI->getOperand(1).getReg();
    unsigned int ABSDstReg = MI->getOperand(0).getReg();
    bool isThumb2 = Subtarget->isThumb2();
    MachineRegisterInfo &MRI = Fn->getRegInfo();
    // In Thumb mode S must not be specified if source register is the SP or
    // PC and if destination register is the SP, so restrict register class
    unsigned NewRsbDstReg = MRI.createVirtualRegister(isThumb2 ?
      (const TargetRegisterClass*)&ARM::rGPRRegClass :
      (const TargetRegisterClass*)&ARM::GPRRegClass);

    // Transfer the remainder of BB and its successor edges to sinkMBB.
    SinkBB->splice(SinkBB->begin(), BB,
                   std::next(MachineBasicBlock::iterator(MI)), BB->end());
    SinkBB->transferSuccessorsAndUpdatePHIs(BB);

    BB->addSuccessor(RSBBB);
    BB->addSuccessor(SinkBB);

    // fall through to SinkMBB
    RSBBB->addSuccessor(SinkBB);

    // insert a cmp at the end of BB
    AddDefaultPred(BuildMI(BB, dl,
                           TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
                   .addReg(ABSSrcReg).addImm(0));

    // insert a bcc with opposite CC to ARMCC::MI at the end of BB
    BuildMI(BB, dl,
      TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)).addMBB(SinkBB)
      .addImm(ARMCC::getOppositeCondition(ARMCC::MI)).addReg(ARM::CPSR);

    // insert rsbri in RSBBB
    // Note: BCC and rsbri will be converted into predicated rsbmi
    // by if-conversion pass
    BuildMI(*RSBBB, RSBBB->begin(), dl,
      TII->get(isThumb2 ? ARM::t2RSBri : ARM::RSBri), NewRsbDstReg)
      .addReg(ABSSrcReg, RegState::Kill)
      .addImm(0).addImm((unsigned)ARMCC::AL).addReg(0).addReg(0);

    // insert PHI in SinkBB,
    // reuse ABSDstReg to not change uses of ABS instruction
    BuildMI(*SinkBB, SinkBB->begin(), dl,
      TII->get(ARM::PHI), ABSDstReg)
      .addReg(NewRsbDstReg).addMBB(RSBBB)
      .addReg(ABSSrcReg).addMBB(BB);

    // remove ABS instruction
    MI->eraseFromParent();

    // return last added BB
    return SinkBB;
  }
  case ARM::COPY_STRUCT_BYVAL_I32:
    ++NumLoopByVals;
    return EmitStructByval(MI, BB);
  case ARM::WIN__CHKSTK:
    return EmitLowered__chkstk(MI, BB);
  }
}

void ARMTargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI,
                                                      SDNode *Node) const {
  const MCInstrDesc *MCID = &MI->getDesc();
  // Adjust potentially 's' setting instructions after isel, i.e. ADC, SBC, RSB,
  // RSC. Coming out of isel, they have an implicit CPSR def, but the optional
  // operand is still set to noreg. If needed, set the optional operand's
  // register to CPSR, and remove the redundant implicit def.
  //
  // e.g. ADCS (..., CPSR<imp-def>) -> ADC (... opt:CPSR<def>).

  // Rename pseudo opcodes.
  unsigned NewOpc = convertAddSubFlagsOpcode(MI->getOpcode());
  if (NewOpc) {
    const ARMBaseInstrInfo *TII = static_cast<const ARMBaseInstrInfo *>(
        getTargetMachine().getSubtargetImpl()->getInstrInfo());
    MCID = &TII->get(NewOpc);

    assert(MCID->getNumOperands() == MI->getDesc().getNumOperands() + 1 &&
           "converted opcode should be the same except for cc_out");

    MI->setDesc(*MCID);

    // Add the optional cc_out operand
    MI->addOperand(MachineOperand::CreateReg(0, /*isDef=*/true));
  }
  unsigned ccOutIdx = MCID->getNumOperands() - 1;

  // Any ARM instruction that sets the 's' bit should specify an optional
  // "cc_out" operand in the last operand position.
  if (!MI->hasOptionalDef() || !MCID->OpInfo[ccOutIdx].isOptionalDef()) {
    assert(!NewOpc && "Optional cc_out operand required");
    return;
  }
  // Look for an implicit def of CPSR added by MachineInstr ctor. Remove it
  // since we already have an optional CPSR def.
  bool definesCPSR = false;
  bool deadCPSR = false;
  for (unsigned i = MCID->getNumOperands(), e = MI->getNumOperands();
       i != e; ++i) {
    const MachineOperand &MO = MI->getOperand(i);
    if (MO.isReg() && MO.isDef() && MO.getReg() == ARM::CPSR) {
      definesCPSR = true;
      if (MO.isDead())
        deadCPSR = true;
      MI->RemoveOperand(i);
      break;
    }
  }
  if (!definesCPSR) {
    assert(!NewOpc && "Optional cc_out operand required");
    return;
  }
  assert(deadCPSR == !Node->hasAnyUseOfValue(1) && "inconsistent dead flag");
  if (deadCPSR) {
    assert(!MI->getOperand(ccOutIdx).getReg() &&
           "expect uninitialized optional cc_out operand");
    return;
  }

  // If this instruction was defined with an optional CPSR def and its dag node
  // had a live implicit CPSR def, then activate the optional CPSR def.
  MachineOperand &MO = MI->getOperand(ccOutIdx);
  MO.setReg(ARM::CPSR);
  MO.setIsDef(true);
}

//===----------------------------------------------------------------------===//
//                           ARM Optimization Hooks
//===----------------------------------------------------------------------===//

// Helper function that checks if N is a null or all ones constant.
static inline bool isZeroOrAllOnes(SDValue N, bool AllOnes) {
  ConstantSDNode *C = dyn_cast<ConstantSDNode>(N);
  if (!C)
    return false;
  return AllOnes ? C->isAllOnesValue() : C->isNullValue();
}

// Return true if N is conditionally 0 or all ones.
// Detects these expressions where cc is an i1 value:
//
//   (select cc 0, y)   [AllOnes=0]
//   (select cc y, 0)   [AllOnes=0]
//   (zext cc)          [AllOnes=0]
//   (sext cc)          [AllOnes=0/1]
//   (select cc -1, y)  [AllOnes=1]
//   (select cc y, -1)  [AllOnes=1]
//
// Invert is set when N is the null/all ones constant when CC is false.
// OtherOp is set to the alternative value of N.
static bool isConditionalZeroOrAllOnes(SDNode *N, bool AllOnes,
                                       SDValue &CC, bool &Invert,
                                       SDValue &OtherOp,
                                       SelectionDAG &DAG) {
  switch (N->getOpcode()) {
  default: return false;
  case ISD::SELECT: {
    CC = N->getOperand(0);
    SDValue N1 = N->getOperand(1);
    SDValue N2 = N->getOperand(2);
    if (isZeroOrAllOnes(N1, AllOnes)) {
      Invert = false;
      OtherOp = N2;
      return true;
    }
    if (isZeroOrAllOnes(N2, AllOnes)) {
      Invert = true;
      OtherOp = N1;
      return true;
    }
    return false;
  }
  case ISD::ZERO_EXTEND:
    // (zext cc) can never be the all ones value.
    if (AllOnes)
      return false;
    // Fall through.
  case ISD::SIGN_EXTEND: {
    EVT VT = N->getValueType(0);
    CC = N->getOperand(0);
    if (CC.getValueType() != MVT::i1)
      return false;
    Invert = !AllOnes;
    if (AllOnes)
      // When looking for an AllOnes constant, N is an sext, and the 'other'
      // value is 0.
      OtherOp = DAG.getConstant(0, VT);
    else if (N->getOpcode() == ISD::ZERO_EXTEND)
      // When looking for a 0 constant, N can be zext or sext.
      OtherOp = DAG.getConstant(1, VT);
    else
      OtherOp = DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), VT);
    return true;
  }
  }
}

// Combine a constant select operand into its use:
//
//   (add (select cc, 0, c), x)  -> (select cc, x, (add, x, c))
//   (sub x, (select cc, 0, c))  -> (select cc, x, (sub, x, c))
//   (and (select cc, -1, c), x) -> (select cc, x, (and, x, c))  [AllOnes=1]
//   (or  (select cc, 0, c), x)  -> (select cc, x, (or, x, c))
//   (xor (select cc, 0, c), x)  -> (select cc, x, (xor, x, c))
//
// The transform is rejected if the select doesn't have a constant operand that
// is null, or all ones when AllOnes is set.
//
// Also recognize sext/zext from i1:
//
//   (add (zext cc), x) -> (select cc (add x, 1), x)
//   (add (sext cc), x) -> (select cc (add x, -1), x)
//
// These transformations eventually create predicated instructions.
//
// @param N       The node to transform.
// @param Slct    The N operand that is a select.
// @param OtherOp The other N operand (x above).
// @param DCI     Context.
// @param AllOnes Require the select constant to be all ones instead of null.
// @returns The new node, or SDValue() on failure.
static
SDValue combineSelectAndUse(SDNode *N, SDValue Slct, SDValue OtherOp,
                            TargetLowering::DAGCombinerInfo &DCI,
                            bool AllOnes = false) {
  SelectionDAG &DAG = DCI.DAG;
  EVT VT = N->getValueType(0);
  SDValue NonConstantVal;
  SDValue CCOp;
  bool SwapSelectOps;
  if (!isConditionalZeroOrAllOnes(Slct.getNode(), AllOnes, CCOp, SwapSelectOps,
                                  NonConstantVal, DAG))
    return SDValue();

  // Slct is now know to be the desired identity constant when CC is true.
  SDValue TrueVal = OtherOp;
  SDValue FalseVal = DAG.getNode(N->getOpcode(), SDLoc(N), VT,
                                 OtherOp, NonConstantVal);
  // Unless SwapSelectOps says CC should be false.
  if (SwapSelectOps)
    std::swap(TrueVal, FalseVal);

  return DAG.getNode(ISD::SELECT, SDLoc(N), VT,
                     CCOp, TrueVal, FalseVal);
}

// Attempt combineSelectAndUse on each operand of a commutative operator N.
static
SDValue combineSelectAndUseCommutative(SDNode *N, bool AllOnes,
                                       TargetLowering::DAGCombinerInfo &DCI) {
  SDValue N0 = N->getOperand(0);
  SDValue N1 = N->getOperand(1);
  if (N0.getNode()->hasOneUse()) {
    SDValue Result = combineSelectAndUse(N, N0, N1, DCI, AllOnes);
    if (Result.getNode())
      return Result;
  }
  if (N1.getNode()->hasOneUse()) {
    SDValue Result = combineSelectAndUse(N, N1, N0, DCI, AllOnes);
    if (Result.getNode())
      return Result;
  }
  return SDValue();
}

// AddCombineToVPADDL- For pair-wise add on neon, use the vpaddl instruction
// (only after legalization).
static SDValue AddCombineToVPADDL(SDNode *N, SDValue N0, SDValue N1,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const ARMSubtarget *Subtarget) {

  // Only perform optimization if after legalize, and if NEON is available. We
  // also expected both operands to be BUILD_VECTORs.
  if (DCI.isBeforeLegalize() || !Subtarget->hasNEON()
      || N0.getOpcode() != ISD::BUILD_VECTOR
      || N1.getOpcode() != ISD::BUILD_VECTOR)
    return SDValue();

  // Check output type since VPADDL operand elements can only be 8, 16, or 32.
  EVT VT = N->getValueType(0);
  if (!VT.isInteger() || VT.getVectorElementType() == MVT::i64)
    return SDValue();

  // Check that the vector operands are of the right form.
  // N0 and N1 are BUILD_VECTOR nodes with N number of EXTRACT_VECTOR
  // operands, where N is the size of the formed vector.
  // Each EXTRACT_VECTOR should have the same input vector and odd or even
  // index such that we have a pair wise add pattern.

  // Grab the vector that all EXTRACT_VECTOR nodes should be referencing.
  if (N0->getOperand(0)->getOpcode() != ISD::EXTRACT_VECTOR_ELT)
    return SDValue();
  SDValue Vec = N0->getOperand(0)->getOperand(0);
  SDNode *V = Vec.getNode();
  unsigned nextIndex = 0;

  // For each operands to the ADD which are BUILD_VECTORs,
  // check to see if each of their operands are an EXTRACT_VECTOR with
  // the same vector and appropriate index.
  for (unsigned i = 0, e = N0->getNumOperands(); i != e; ++i) {
    if (N0->getOperand(i)->getOpcode() == ISD::EXTRACT_VECTOR_ELT
        && N1->getOperand(i)->getOpcode() == ISD::EXTRACT_VECTOR_ELT) {

      SDValue ExtVec0 = N0->getOperand(i);
      SDValue ExtVec1 = N1->getOperand(i);

      // First operand is the vector, verify its the same.
      if (V != ExtVec0->getOperand(0).getNode() ||
          V != ExtVec1->getOperand(0).getNode())
        return SDValue();

      // Second is the constant, verify its correct.
      ConstantSDNode *C0 = dyn_cast<ConstantSDNode>(ExtVec0->getOperand(1));
      ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(ExtVec1->getOperand(1));

      // For the constant, we want to see all the even or all the odd.
      if (!C0 || !C1 || C0->getZExtValue() != nextIndex
          || C1->getZExtValue() != nextIndex+1)
        return SDValue();

      // Increment index.
      nextIndex+=2;
    } else
      return SDValue();
  }

  // Create VPADDL node.
  SelectionDAG &DAG = DCI.DAG;
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();

  // Build operand list.
  SmallVector<SDValue, 8> Ops;
  Ops.push_back(DAG.getConstant(Intrinsic::arm_neon_vpaddls,
                                TLI.getPointerTy()));

  // Input is the vector.
  Ops.push_back(Vec);

  // Get widened type and narrowed type.
  MVT widenType;
  unsigned numElem = VT.getVectorNumElements();
  
  EVT inputLaneType = Vec.getValueType().getVectorElementType();
  switch (inputLaneType.getSimpleVT().SimpleTy) {
    case MVT::i8: widenType = MVT::getVectorVT(MVT::i16, numElem); break;
    case MVT::i16: widenType = MVT::getVectorVT(MVT::i32, numElem); break;
    case MVT::i32: widenType = MVT::getVectorVT(MVT::i64, numElem); break;
    default:
      llvm_unreachable("Invalid vector element type for padd optimization.");
  }

  SDValue tmp = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N), widenType, Ops);
  unsigned ExtOp = VT.bitsGT(tmp.getValueType()) ? ISD::ANY_EXTEND : ISD::TRUNCATE;
  return DAG.getNode(ExtOp, SDLoc(N), VT, tmp);
}

static SDValue findMUL_LOHI(SDValue V) {
  if (V->getOpcode() == ISD::UMUL_LOHI ||
      V->getOpcode() == ISD::SMUL_LOHI)
    return V;
  return SDValue();
}

static SDValue AddCombineTo64bitMLAL(SDNode *AddcNode,
                                     TargetLowering::DAGCombinerInfo &DCI,
                                     const ARMSubtarget *Subtarget) {

  if (Subtarget->isThumb1Only()) return SDValue();

  // Only perform the checks after legalize when the pattern is available.
  if (DCI.isBeforeLegalize()) return SDValue();

  // Look for multiply add opportunities.
  // The pattern is a ISD::UMUL_LOHI followed by two add nodes, where
  // each add nodes consumes a value from ISD::UMUL_LOHI and there is
  // a glue link from the first add to the second add.
  // If we find this pattern, we can replace the U/SMUL_LOHI, ADDC, and ADDE by
  // a S/UMLAL instruction.
  //          loAdd   UMUL_LOHI
  //            \    / :lo    \ :hi
  //             \  /          \          [no multiline comment]
  //              ADDC         |  hiAdd
  //                 \ :glue  /  /
  //                  \      /  /
  //                    ADDE
  //
  assert(AddcNode->getOpcode() == ISD::ADDC && "Expect an ADDC");
  SDValue AddcOp0 = AddcNode->getOperand(0);
  SDValue AddcOp1 = AddcNode->getOperand(1);

  // Check if the two operands are from the same mul_lohi node.
  if (AddcOp0.getNode() == AddcOp1.getNode())
    return SDValue();

  assert(AddcNode->getNumValues() == 2 &&
         AddcNode->getValueType(0) == MVT::i32 &&
         "Expect ADDC with two result values. First: i32");

  // Check that we have a glued ADDC node.
  if (AddcNode->getValueType(1) != MVT::Glue)
    return SDValue();

  // Check that the ADDC adds the low result of the S/UMUL_LOHI.
  if (AddcOp0->getOpcode() != ISD::UMUL_LOHI &&
      AddcOp0->getOpcode() != ISD::SMUL_LOHI &&
      AddcOp1->getOpcode() != ISD::UMUL_LOHI &&
      AddcOp1->getOpcode() != ISD::SMUL_LOHI)
    return SDValue();

  // Look for the glued ADDE.
  SDNode* AddeNode = AddcNode->getGluedUser();
  if (!AddeNode)
    return SDValue();

  // Make sure it is really an ADDE.
  if (AddeNode->getOpcode() != ISD::ADDE)
    return SDValue();

  assert(AddeNode->getNumOperands() == 3 &&
         AddeNode->getOperand(2).getValueType() == MVT::Glue &&
         "ADDE node has the wrong inputs");

  // Check for the triangle shape.
  SDValue AddeOp0 = AddeNode->getOperand(0);
  SDValue AddeOp1 = AddeNode->getOperand(1);

  // Make sure that the ADDE operands are not coming from the same node.
  if (AddeOp0.getNode() == AddeOp1.getNode())
    return SDValue();

  // Find the MUL_LOHI node walking up ADDE's operands.
  bool IsLeftOperandMUL = false;
  SDValue MULOp = findMUL_LOHI(AddeOp0);
  if (MULOp == SDValue())
   MULOp = findMUL_LOHI(AddeOp1);
  else
    IsLeftOperandMUL = true;
  if (MULOp == SDValue())
     return SDValue();

  // Figure out the right opcode.
  unsigned Opc = MULOp->getOpcode();
  unsigned FinalOpc = (Opc == ISD::SMUL_LOHI) ? ARMISD::SMLAL : ARMISD::UMLAL;

  // Figure out the high and low input values to the MLAL node.
  SDValue* HiMul = &MULOp;
  SDValue* HiAdd = nullptr;
  SDValue* LoMul = nullptr;
  SDValue* LowAdd = nullptr;

  if (IsLeftOperandMUL)
    HiAdd = &AddeOp1;
  else
    HiAdd = &AddeOp0;


  if (AddcOp0->getOpcode() == Opc) {
    LoMul = &AddcOp0;
    LowAdd = &AddcOp1;
  }
  if (AddcOp1->getOpcode() == Opc) {
    LoMul = &AddcOp1;
    LowAdd = &AddcOp0;
  }

  if (!LoMul)
    return SDValue();

  if (LoMul->getNode() != HiMul->getNode())
    return SDValue();

  // Create the merged node.
  SelectionDAG &DAG = DCI.DAG;

  // Build operand list.
  SmallVector<SDValue, 8> Ops;
  Ops.push_back(LoMul->getOperand(0));
  Ops.push_back(LoMul->getOperand(1));
  Ops.push_back(*LowAdd);
  Ops.push_back(*HiAdd);

  SDValue MLALNode =  DAG.getNode(FinalOpc, SDLoc(AddcNode),
                                 DAG.getVTList(MVT::i32, MVT::i32), Ops);

  // Replace the ADDs' nodes uses by the MLA node's values.
  SDValue HiMLALResult(MLALNode.getNode(), 1);
  DAG.ReplaceAllUsesOfValueWith(SDValue(AddeNode, 0), HiMLALResult);

  SDValue LoMLALResult(MLALNode.getNode(), 0);
  DAG.ReplaceAllUsesOfValueWith(SDValue(AddcNode, 0), LoMLALResult);

  // Return original node to notify the driver to stop replacing.
  SDValue resNode(AddcNode, 0);
  return resNode;
}

/// PerformADDCCombine - Target-specific dag combine transform from
/// ISD::ADDC, ISD::ADDE, and ISD::MUL_LOHI to MLAL.
static SDValue PerformADDCCombine(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const ARMSubtarget *Subtarget) {

  return AddCombineTo64bitMLAL(N, DCI, Subtarget);

}

/// PerformADDCombineWithOperands - Try DAG combinations for an ADD with
/// operands N0 and N1.  This is a helper for PerformADDCombine that is
/// called with the default operands, and if that fails, with commuted
/// operands.
static SDValue PerformADDCombineWithOperands(SDNode *N, SDValue N0, SDValue N1,
                                          TargetLowering::DAGCombinerInfo &DCI,
                                          const ARMSubtarget *Subtarget){

  // Attempt to create vpaddl for this add.
  SDValue Result = AddCombineToVPADDL(N, N0, N1, DCI, Subtarget);
  if (Result.getNode())
    return Result;

  // fold (add (select cc, 0, c), x) -> (select cc, x, (add, x, c))
  if (N0.getNode()->hasOneUse()) {
    SDValue Result = combineSelectAndUse(N, N0, N1, DCI);
    if (Result.getNode()) return Result;
  }
  return SDValue();
}

/// PerformADDCombine - Target-specific dag combine xforms for ISD::ADD.
///
static SDValue PerformADDCombine(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const ARMSubtarget *Subtarget) {
  SDValue N0 = N->getOperand(0);
  SDValue N1 = N->getOperand(1);

  // First try with the default operand order.
  SDValue Result = PerformADDCombineWithOperands(N, N0, N1, DCI, Subtarget);
  if (Result.getNode())
    return Result;

  // If that didn't work, try again with the operands commuted.
  return PerformADDCombineWithOperands(N, N1, N0, DCI, Subtarget);
}

/// PerformSUBCombine - Target-specific dag combine xforms for ISD::SUB.
///
static SDValue PerformSUBCombine(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI) {
  SDValue N0 = N->getOperand(0);
  SDValue N1 = N->getOperand(1);

  // fold (sub x, (select cc, 0, c)) -> (select cc, x, (sub, x, c))
  if (N1.getNode()->hasOneUse()) {
    SDValue Result = combineSelectAndUse(N, N1, N0, DCI);
    if (Result.getNode()) return Result;
  }

  return SDValue();
}

/// PerformVMULCombine
/// Distribute (A + B) * C to (A * C) + (B * C) to take advantage of the
/// special multiplier accumulator forwarding.
///   vmul d3, d0, d2
///   vmla d3, d1, d2
/// is faster than
///   vadd d3, d0, d1
///   vmul d3, d3, d2
//  However, for (A + B) * (A + B),
//    vadd d2, d0, d1
//    vmul d3, d0, d2
//    vmla d3, d1, d2
//  is slower than
//    vadd d2, d0, d1
//    vmul d3, d2, d2
static SDValue PerformVMULCombine(SDNode *N,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const ARMSubtarget *Subtarget) {
  if (!Subtarget->hasVMLxForwarding())
    return SDValue();

  SelectionDAG &DAG = DCI.DAG;
  SDValue N0 = N->getOperand(0);
  SDValue N1 = N->getOperand(1);
  unsigned Opcode = N0.getOpcode();
  if (Opcode != ISD::ADD && Opcode != ISD::SUB &&
      Opcode != ISD::FADD && Opcode != ISD::FSUB) {
    Opcode = N1.getOpcode();
    if (Opcode != ISD::ADD && Opcode != ISD::SUB &&
        Opcode != ISD::FADD && Opcode != ISD::FSUB)
      return SDValue();
    std::swap(N0, N1);
  }

  if (N0 == N1)
    return SDValue();

  EVT VT = N->getValueType(0);
  SDLoc DL(N);
  SDValue N00 = N0->getOperand(0);
  SDValue N01 = N0->getOperand(1);
  return DAG.getNode(Opcode, DL, VT,
                     DAG.getNode(ISD::MUL, DL, VT, N00, N1),
                     DAG.getNode(ISD::MUL, DL, VT, N01, N1));
}

static SDValue PerformMULCombine(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const ARMSubtarget *Subtarget) {
  SelectionDAG &DAG = DCI.DAG;

  if (Subtarget->isThumb1Only())
    return SDValue();

  if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
    return SDValue();

  EVT VT = N->getValueType(0);
  if (VT.is64BitVector() || VT.is128BitVector())
    return PerformVMULCombine(N, DCI, Subtarget);
  if (VT != MVT::i32)
    return SDValue();

  ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1));
  if (!C)
    return SDValue();

  int64_t MulAmt = C->getSExtValue();
  unsigned ShiftAmt = countTrailingZeros<uint64_t>(MulAmt);

  ShiftAmt = ShiftAmt & (32 - 1);
  SDValue V = N->getOperand(0);
  SDLoc DL(N);

  SDValue Res;
  MulAmt >>= ShiftAmt;

  if (MulAmt >= 0) {
    if (isPowerOf2_32(MulAmt - 1)) {
      // (mul x, 2^N + 1) => (add (shl x, N), x)
      Res = DAG.getNode(ISD::ADD, DL, VT,
                        V,
                        DAG.getNode(ISD::SHL, DL, VT,
                                    V,
                                    DAG.getConstant(Log2_32(MulAmt - 1),
                                                    MVT::i32)));
    } else if (isPowerOf2_32(MulAmt + 1)) {
      // (mul x, 2^N - 1) => (sub (shl x, N), x)
      Res = DAG.getNode(ISD::SUB, DL, VT,
                        DAG.getNode(ISD::SHL, DL, VT,
                                    V,
                                    DAG.getConstant(Log2_32(MulAmt + 1),
                                                    MVT::i32)),
                        V);
    } else
      return SDValue();
  } else {
    uint64_t MulAmtAbs = -MulAmt;
    if (isPowerOf2_32(MulAmtAbs + 1)) {
      // (mul x, -(2^N - 1)) => (sub x, (shl x, N))
      Res = DAG.getNode(ISD::SUB, DL, VT,
                        V,
                        DAG.getNode(ISD::SHL, DL, VT,
                                    V,
                                    DAG.getConstant(Log2_32(MulAmtAbs + 1),
                                                    MVT::i32)));
    } else if (isPowerOf2_32(MulAmtAbs - 1)) {
      // (mul x, -(2^N + 1)) => - (add (shl x, N), x)
      Res = DAG.getNode(ISD::ADD, DL, VT,
                        V,
                        DAG.getNode(ISD::SHL, DL, VT,
                                    V,
                                    DAG.getConstant(Log2_32(MulAmtAbs-1),
                                                    MVT::i32)));
      Res = DAG.getNode(ISD::SUB, DL, VT,
                        DAG.getConstant(0, MVT::i32),Res);

    } else
      return SDValue();
  }

  if (ShiftAmt != 0)
    Res = DAG.getNode(ISD::SHL, DL, VT,
                      Res, DAG.getConstant(ShiftAmt, MVT::i32));

  // Do not add new nodes to DAG combiner worklist.
  DCI.CombineTo(N, Res, false);
  return SDValue();
}

static SDValue PerformANDCombine(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const ARMSubtarget *Subtarget) {

  // Attempt to use immediate-form VBIC
  BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N->getOperand(1));
  SDLoc dl(N);
  EVT VT = N->getValueType(0);
  SelectionDAG &DAG = DCI.DAG;

  if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
    return SDValue();

  APInt SplatBits, SplatUndef;
  unsigned SplatBitSize;
  bool HasAnyUndefs;
  if (BVN &&
      BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
    if (SplatBitSize <= 64) {
      EVT VbicVT;
      SDValue Val = isNEONModifiedImm((~SplatBits).getZExtValue(),
                                      SplatUndef.getZExtValue(), SplatBitSize,
                                      DAG, VbicVT, VT.is128BitVector(),
                                      OtherModImm);
      if (Val.getNode()) {
        SDValue Input =
          DAG.getNode(ISD::BITCAST, dl, VbicVT, N->getOperand(0));
        SDValue Vbic = DAG.getNode(ARMISD::VBICIMM, dl, VbicVT, Input, Val);
        return DAG.getNode(ISD::BITCAST, dl, VT, Vbic);
      }
    }
  }

  if (!Subtarget->isThumb1Only()) {
    // fold (and (select cc, -1, c), x) -> (select cc, x, (and, x, c))
    SDValue Result = combineSelectAndUseCommutative(N, true, DCI);
    if (Result.getNode())
      return Result;
  }

  return SDValue();
}

/// PerformORCombine - Target-specific dag combine xforms for ISD::OR
static SDValue PerformORCombine(SDNode *N,
                                TargetLowering::DAGCombinerInfo &DCI,
                                const ARMSubtarget *Subtarget) {
  // Attempt to use immediate-form VORR
  BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N->getOperand(1));
  SDLoc dl(N);
  EVT VT = N->getValueType(0);
  SelectionDAG &DAG = DCI.DAG;

  if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
    return SDValue();

  APInt SplatBits, SplatUndef;
  unsigned SplatBitSize;
  bool HasAnyUndefs;
  if (BVN && Subtarget->hasNEON() &&
      BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
    if (SplatBitSize <= 64) {
      EVT VorrVT;
      SDValue Val = isNEONModifiedImm(SplatBits.getZExtValue(),
                                      SplatUndef.getZExtValue(), SplatBitSize,
                                      DAG, VorrVT, VT.is128BitVector(),
                                      OtherModImm);
      if (Val.getNode()) {
        SDValue Input =
          DAG.getNode(ISD::BITCAST, dl, VorrVT, N->getOperand(0));
        SDValue Vorr = DAG.getNode(ARMISD::VORRIMM, dl, VorrVT, Input, Val);
        return DAG.getNode(ISD::BITCAST, dl, VT, Vorr);
      }
    }
  }

  if (!Subtarget->isThumb1Only()) {
    // fold (or (select cc, 0, c), x) -> (select cc, x, (or, x, c))
    SDValue Result = combineSelectAndUseCommutative(N, false, DCI);
    if (Result.getNode())
      return Result;
  }

  // The code below optimizes (or (and X, Y), Z).
  // The AND operand needs to have a single user to make these optimizations
  // profitable.
  SDValue N0 = N->getOperand(0);
  if (N0.getOpcode() != ISD::AND || !N0.hasOneUse())
    return SDValue();
  SDValue N1 = N->getOperand(1);

  // (or (and B, A), (and C, ~A)) => (VBSL A, B, C) when A is a constant.
  if (Subtarget->hasNEON() && N1.getOpcode() == ISD::AND && VT.isVector() &&
      DAG.getTargetLoweringInfo().isTypeLegal(VT)) {
    APInt SplatUndef;
    unsigned SplatBitSize;
    bool HasAnyUndefs;

    APInt SplatBits0, SplatBits1;
    BuildVectorSDNode *BVN0 = dyn_cast<BuildVectorSDNode>(N0->getOperand(1));
    BuildVectorSDNode *BVN1 = dyn_cast<BuildVectorSDNode>(N1->getOperand(1));
    // Ensure that the second operand of both ands are constants
    if (BVN0 && BVN0->isConstantSplat(SplatBits0, SplatUndef, SplatBitSize,
                                      HasAnyUndefs) && !HasAnyUndefs) {
        if (BVN1 && BVN1->isConstantSplat(SplatBits1, SplatUndef, SplatBitSize,
                                          HasAnyUndefs) && !HasAnyUndefs) {
            // Ensure that the bit width of the constants are the same and that
            // the splat arguments are logical inverses as per the pattern we
            // are trying to simplify.
            if (SplatBits0.getBitWidth() == SplatBits1.getBitWidth() &&
                SplatBits0 == ~SplatBits1) {
                // Canonicalize the vector type to make instruction selection
                // simpler.
                EVT CanonicalVT = VT.is128BitVector() ? MVT::v4i32 : MVT::v2i32;
                SDValue Result = DAG.getNode(ARMISD::VBSL, dl, CanonicalVT,
                                             N0->getOperand(1),
                                             N0->getOperand(0),
                                             N1->getOperand(0));
                return DAG.getNode(ISD::BITCAST, dl, VT, Result);
            }
        }
    }
  }

  // Try to use the ARM/Thumb2 BFI (bitfield insert) instruction when
  // reasonable.

  // BFI is only available on V6T2+
  if (Subtarget->isThumb1Only() || !Subtarget->hasV6T2Ops())
    return SDValue();

  SDLoc DL(N);
  // 1) or (and A, mask), val => ARMbfi A, val, mask
  //      iff (val & mask) == val
  //
  // 2) or (and A, mask), (and B, mask2) => ARMbfi A, (lsr B, amt), mask
  //  2a) iff isBitFieldInvertedMask(mask) && isBitFieldInvertedMask(~mask2)
  //          && mask == ~mask2
  //  2b) iff isBitFieldInvertedMask(~mask) && isBitFieldInvertedMask(mask2)
  //          && ~mask == mask2
  //  (i.e., copy a bitfield value into another bitfield of the same width)

  if (VT != MVT::i32)
    return SDValue();

  SDValue N00 = N0.getOperand(0);

  // The value and the mask need to be constants so we can verify this is
  // actually a bitfield set. If the mask is 0xffff, we can do better
  // via a movt instruction, so don't use BFI in that case.
  SDValue MaskOp = N0.getOperand(1);
  ConstantSDNode *MaskC = dyn_cast<ConstantSDNode>(MaskOp);
  if (!MaskC)
    return SDValue();
  unsigned Mask = MaskC->getZExtValue();
  if (Mask == 0xffff)
    return SDValue();
  SDValue Res;
  // Case (1): or (and A, mask), val => ARMbfi A, val, mask
  ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
  if (N1C) {
    unsigned Val = N1C->getZExtValue();
    if ((Val & ~Mask) != Val)
      return SDValue();

    if (ARM::isBitFieldInvertedMask(Mask)) {
      Val >>= countTrailingZeros(~Mask);

      Res = DAG.getNode(ARMISD::BFI, DL, VT, N00,
                        DAG.getConstant(Val, MVT::i32),
                        DAG.getConstant(Mask, MVT::i32));

      // Do not add new nodes to DAG combiner worklist.
      DCI.CombineTo(N, Res, false);
      return SDValue();
    }
  } else if (N1.getOpcode() == ISD::AND) {
    // case (2) or (and A, mask), (and B, mask2) => ARMbfi A, (lsr B, amt), mask
    ConstantSDNode *N11C = dyn_cast<ConstantSDNode>(N1.getOperand(1));
    if (!N11C)
      return SDValue();
    unsigned Mask2 = N11C->getZExtValue();

    // Mask and ~Mask2 (or reverse) must be equivalent for the BFI pattern
    // as is to match.
    if (ARM::isBitFieldInvertedMask(Mask) &&
        (Mask == ~Mask2)) {
      // The pack halfword instruction works better for masks that fit it,
      // so use that when it's available.
      if (Subtarget->hasT2ExtractPack() &&
          (Mask == 0xffff || Mask == 0xffff0000))
        return SDValue();
      // 2a
      unsigned amt = countTrailingZeros(Mask2);
      Res = DAG.getNode(ISD::SRL, DL, VT, N1.getOperand(0),
                        DAG.getConstant(amt, MVT::i32));
      Res = DAG.getNode(ARMISD::BFI, DL, VT, N00, Res,
                        DAG.getConstant(Mask, MVT::i32));
      // Do not add new nodes to DAG combiner worklist.
      DCI.CombineTo(N, Res, false);
      return SDValue();
    } else if (ARM::isBitFieldInvertedMask(~Mask) &&
               (~Mask == Mask2)) {
      // The pack halfword instruction works better for masks that fit it,
      // so use that when it's available.
      if (Subtarget->hasT2ExtractPack() &&
          (Mask2 == 0xffff || Mask2 == 0xffff0000))
        return SDValue();
      // 2b
      unsigned lsb = countTrailingZeros(Mask);
      Res = DAG.getNode(ISD::SRL, DL, VT, N00,
                        DAG.getConstant(lsb, MVT::i32));
      Res = DAG.getNode(ARMISD::BFI, DL, VT, N1.getOperand(0), Res,
                        DAG.getConstant(Mask2, MVT::i32));
      // Do not add new nodes to DAG combiner worklist.
      DCI.CombineTo(N, Res, false);
      return SDValue();
    }
  }

  if (DAG.MaskedValueIsZero(N1, MaskC->getAPIntValue()) &&
      N00.getOpcode() == ISD::SHL && isa<ConstantSDNode>(N00.getOperand(1)) &&
      ARM::isBitFieldInvertedMask(~Mask)) {
    // Case (3): or (and (shl A, #shamt), mask), B => ARMbfi B, A, ~mask
    // where lsb(mask) == #shamt and masked bits of B are known zero.
    SDValue ShAmt = N00.getOperand(1);
    unsigned ShAmtC = cast<ConstantSDNode>(ShAmt)->getZExtValue();
    unsigned LSB = countTrailingZeros(Mask);
    if (ShAmtC != LSB)
      return SDValue();

    Res = DAG.getNode(ARMISD::BFI, DL, VT, N1, N00.getOperand(0),
                      DAG.getConstant(~Mask, MVT::i32));

    // Do not add new nodes to DAG combiner worklist.
    DCI.CombineTo(N, Res, false);
  }

  return SDValue();
}

static SDValue PerformXORCombine(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const ARMSubtarget *Subtarget) {
  EVT VT = N->getValueType(0);
  SelectionDAG &DAG = DCI.DAG;

  if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
    return SDValue();

  if (!Subtarget->isThumb1Only()) {
    // fold (xor (select cc, 0, c), x) -> (select cc, x, (xor, x, c))
    SDValue Result = combineSelectAndUseCommutative(N, false, DCI);
    if (Result.getNode())
      return Result;
  }

  return SDValue();
}

/// PerformBFICombine - (bfi A, (and B, Mask1), Mask2) -> (bfi A, B, Mask2) iff
/// the bits being cleared by the AND are not demanded by the BFI.
static SDValue PerformBFICombine(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI) {
  SDValue N1 = N->getOperand(1);
  if (N1.getOpcode() == ISD::AND) {
    ConstantSDNode *N11C = dyn_cast<ConstantSDNode>(N1.getOperand(1));
    if (!N11C)
      return SDValue();
    unsigned InvMask = cast<ConstantSDNode>(N->getOperand(2))->getZExtValue();
    unsigned LSB = countTrailingZeros(~InvMask);
    unsigned Width = (32 - countLeadingZeros(~InvMask)) - LSB;
    unsigned Mask = (1 << Width)-1;
    unsigned Mask2 = N11C->getZExtValue();
    if ((Mask & (~Mask2)) == 0)
      return DCI.DAG.getNode(ARMISD::BFI, SDLoc(N), N->getValueType(0),
                             N->getOperand(0), N1.getOperand(0),
                             N->getOperand(2));
  }
  return SDValue();
}

/// PerformVMOVRRDCombine - Target-specific dag combine xforms for
/// ARMISD::VMOVRRD.
static SDValue PerformVMOVRRDCombine(SDNode *N,
                                     TargetLowering::DAGCombinerInfo &DCI,
                                     const ARMSubtarget *Subtarget) {
  // vmovrrd(vmovdrr x, y) -> x,y
  SDValue InDouble = N->getOperand(0);
  if (InDouble.getOpcode() == ARMISD::VMOVDRR && !Subtarget->isFPOnlySP())
    return DCI.CombineTo(N, InDouble.getOperand(0), InDouble.getOperand(1));

  // vmovrrd(load f64) -> (load i32), (load i32)
  SDNode *InNode = InDouble.getNode();
  if (ISD::isNormalLoad(InNode) && InNode->hasOneUse() &&
      InNode->getValueType(0) == MVT::f64 &&
      InNode->getOperand(1).getOpcode() == ISD::FrameIndex &&
      !cast<LoadSDNode>(InNode)->isVolatile()) {
    // TODO: Should this be done for non-FrameIndex operands?
    LoadSDNode *LD = cast<LoadSDNode>(InNode);

    SelectionDAG &DAG = DCI.DAG;
    SDLoc DL(LD);
    SDValue BasePtr = LD->getBasePtr();
    SDValue NewLD1 = DAG.getLoad(MVT::i32, DL, LD->getChain(), BasePtr,
                                 LD->getPointerInfo(), LD->isVolatile(),
                                 LD->isNonTemporal(), LD->isInvariant(),
                                 LD->getAlignment());

    SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i32, BasePtr,
                                    DAG.getConstant(4, MVT::i32));
    SDValue NewLD2 = DAG.getLoad(MVT::i32, DL, NewLD1.getValue(1), OffsetPtr,
                                 LD->getPointerInfo(), LD->isVolatile(),
                                 LD->isNonTemporal(), LD->isInvariant(),
                                 std::min(4U, LD->getAlignment() / 2));

    DAG.ReplaceAllUsesOfValueWith(SDValue(LD, 1), NewLD2.getValue(1));
    if (DCI.DAG.getTargetLoweringInfo().isBigEndian())
      std::swap (NewLD1, NewLD2);
    SDValue Result = DCI.CombineTo(N, NewLD1, NewLD2);
    return Result;
  }

  return SDValue();
}

/// PerformVMOVDRRCombine - Target-specific dag combine xforms for
/// ARMISD::VMOVDRR.  This is also used for BUILD_VECTORs with 2 operands.
static SDValue PerformVMOVDRRCombine(SDNode *N, SelectionDAG &DAG) {
  // N=vmovrrd(X); vmovdrr(N:0, N:1) -> bit_convert(X)
  SDValue Op0 = N->getOperand(0);
  SDValue Op1 = N->getOperand(1);
  if (Op0.getOpcode() == ISD::BITCAST)
    Op0 = Op0.getOperand(0);
  if (Op1.getOpcode() == ISD::BITCAST)
    Op1 = Op1.getOperand(0);
  if (Op0.getOpcode() == ARMISD::VMOVRRD &&
      Op0.getNode() == Op1.getNode() &&
      Op0.getResNo() == 0 && Op1.getResNo() == 1)
    return DAG.getNode(ISD::BITCAST, SDLoc(N),
                       N->getValueType(0), Op0.getOperand(0));
  return SDValue();
}

/// PerformSTORECombine - Target-specific dag combine xforms for
/// ISD::STORE.
static SDValue PerformSTORECombine(SDNode *N,
                                   TargetLowering::DAGCombinerInfo &DCI) {
  StoreSDNode *St = cast<StoreSDNode>(N);
  if (St->isVolatile())
    return SDValue();

  // Optimize trunc store (of multiple scalars) to shuffle and store.  First,
  // pack all of the elements in one place.  Next, store to memory in fewer
  // chunks.
  SDValue StVal = St->getValue();
  EVT VT = StVal.getValueType();
  if (St->isTruncatingStore() && VT.isVector()) {
    SelectionDAG &DAG = DCI.DAG;
    const TargetLowering &TLI = DAG.getTargetLoweringInfo();
    EVT StVT = St->getMemoryVT();
    unsigned NumElems = VT.getVectorNumElements();
    assert(StVT != VT && "Cannot truncate to the same type");
    unsigned FromEltSz = VT.getVectorElementType().getSizeInBits();
    unsigned ToEltSz = StVT.getVectorElementType().getSizeInBits();

    // From, To sizes and ElemCount must be pow of two
    if (!isPowerOf2_32(NumElems * FromEltSz * ToEltSz)) return SDValue();

    // We are going to use the original vector elt for storing.
    // Accumulated smaller vector elements must be a multiple of the store size.
    if (0 != (NumElems * FromEltSz) % ToEltSz) return SDValue();

    unsigned SizeRatio  = FromEltSz / ToEltSz;
    assert(SizeRatio * NumElems * ToEltSz == VT.getSizeInBits());

    // Create a type on which we perform the shuffle.
    EVT WideVecVT = EVT::getVectorVT(*DAG.getContext(), StVT.getScalarType(),
                                     NumElems*SizeRatio);
    assert(WideVecVT.getSizeInBits() == VT.getSizeInBits());

    SDLoc DL(St);
    SDValue WideVec = DAG.getNode(ISD::BITCAST, DL, WideVecVT, StVal);
    SmallVector<int, 8> ShuffleVec(NumElems * SizeRatio, -1);
    for (unsigned i = 0; i < NumElems; ++i)
      ShuffleVec[i] = TLI.isBigEndian() ? (i+1) * SizeRatio - 1 : i * SizeRatio;

    // Can't shuffle using an illegal type.
    if (!TLI.isTypeLegal(WideVecVT)) return SDValue();

    SDValue Shuff = DAG.getVectorShuffle(WideVecVT, DL, WideVec,
                                DAG.getUNDEF(WideVec.getValueType()),
                                ShuffleVec.data());
    // At this point all of the data is stored at the bottom of the
    // register. We now need to save it to mem.

    // Find the largest store unit
    MVT StoreType = MVT::i8;
    for (unsigned tp = MVT::FIRST_INTEGER_VALUETYPE;
         tp < MVT::LAST_INTEGER_VALUETYPE; ++tp) {
      MVT Tp = (MVT::SimpleValueType)tp;
      if (TLI.isTypeLegal(Tp) && Tp.getSizeInBits() <= NumElems * ToEltSz)
        StoreType = Tp;
    }
    // Didn't find a legal store type.
    if (!TLI.isTypeLegal(StoreType))
      return SDValue();

    // Bitcast the original vector into a vector of store-size units
    EVT StoreVecVT = EVT::getVectorVT(*DAG.getContext(),
            StoreType, VT.getSizeInBits()/EVT(StoreType).getSizeInBits());
    assert(StoreVecVT.getSizeInBits() == VT.getSizeInBits());
    SDValue ShuffWide = DAG.getNode(ISD::BITCAST, DL, StoreVecVT, Shuff);
    SmallVector<SDValue, 8> Chains;
    SDValue Increment = DAG.getConstant(StoreType.getSizeInBits()/8,
                                        TLI.getPointerTy());
    SDValue BasePtr = St->getBasePtr();

    // Perform one or more big stores into memory.
    unsigned E = (ToEltSz*NumElems)/StoreType.getSizeInBits();
    for (unsigned I = 0; I < E; I++) {
      SDValue SubVec = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
                                   StoreType, ShuffWide,
                                   DAG.getIntPtrConstant(I));
      SDValue Ch = DAG.getStore(St->getChain(), DL, SubVec, BasePtr,
                                St->getPointerInfo(), St->isVolatile(),
                                St->isNonTemporal(), St->getAlignment());
      BasePtr = DAG.getNode(ISD::ADD, DL, BasePtr.getValueType(), BasePtr,
                            Increment);
      Chains.push_back(Ch);
    }
    return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
  }

  if (!ISD::isNormalStore(St))
    return SDValue();

  // Split a store of a VMOVDRR into two integer stores to avoid mixing NEON and
  // ARM stores of arguments in the same cache line.
  if (StVal.getNode()->getOpcode() == ARMISD::VMOVDRR &&
      StVal.getNode()->hasOneUse()) {
    SelectionDAG  &DAG = DCI.DAG;
    bool isBigEndian = DAG.getTargetLoweringInfo().isBigEndian();
    SDLoc DL(St);
    SDValue BasePtr = St->getBasePtr();
    SDValue NewST1 = DAG.getStore(St->getChain(), DL,
                                  StVal.getNode()->getOperand(isBigEndian ? 1 : 0 ),
                                  BasePtr, St->getPointerInfo(), St->isVolatile(),
                                  St->isNonTemporal(), St->getAlignment());

    SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i32, BasePtr,
                                    DAG.getConstant(4, MVT::i32));
    return DAG.getStore(NewST1.getValue(0), DL,
                        StVal.getNode()->getOperand(isBigEndian ? 0 : 1),
                        OffsetPtr, St->getPointerInfo(), St->isVolatile(),
                        St->isNonTemporal(),
                        std::min(4U, St->getAlignment() / 2));
  }

  if (StVal.getValueType() != MVT::i64 ||
      StVal.getNode()->getOpcode() != ISD::EXTRACT_VECTOR_ELT)
    return SDValue();

  // Bitcast an i64 store extracted from a vector to f64.
  // Otherwise, the i64 value will be legalized to a pair of i32 values.
  SelectionDAG &DAG = DCI.DAG;
  SDLoc dl(StVal);
  SDValue IntVec = StVal.getOperand(0);
  EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f64,
                                 IntVec.getValueType().getVectorNumElements());
  SDValue Vec = DAG.getNode(ISD::BITCAST, dl, FloatVT, IntVec);
  SDValue ExtElt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64,
                               Vec, StVal.getOperand(1));
  dl = SDLoc(N);
  SDValue V = DAG.getNode(ISD::BITCAST, dl, MVT::i64, ExtElt);
  // Make the DAGCombiner fold the bitcasts.
  DCI.AddToWorklist(Vec.getNode());
  DCI.AddToWorklist(ExtElt.getNode());
  DCI.AddToWorklist(V.getNode());
  return DAG.getStore(St->getChain(), dl, V, St->getBasePtr(),
                      St->getPointerInfo(), St->isVolatile(),
                      St->isNonTemporal(), St->getAlignment(),
                      St->getAAInfo());
}

/// hasNormalLoadOperand - Check if any of the operands of a BUILD_VECTOR node
/// are normal, non-volatile loads.  If so, it is profitable to bitcast an
/// i64 vector to have f64 elements, since the value can then be loaded
/// directly into a VFP register.
static bool hasNormalLoadOperand(SDNode *N) {
  unsigned NumElts = N->getValueType(0).getVectorNumElements();
  for (unsigned i = 0; i < NumElts; ++i) {
    SDNode *Elt = N->getOperand(i).getNode();
    if (ISD::isNormalLoad(Elt) && !cast<LoadSDNode>(Elt)->isVolatile())
      return true;
  }
  return false;
}

/// PerformBUILD_VECTORCombine - Target-specific dag combine xforms for
/// ISD::BUILD_VECTOR.
static SDValue PerformBUILD_VECTORCombine(SDNode *N,
                                          TargetLowering::DAGCombinerInfo &DCI,
                                          const ARMSubtarget *Subtarget) {
  // build_vector(N=ARMISD::VMOVRRD(X), N:1) -> bit_convert(X):
  // VMOVRRD is introduced when legalizing i64 types.  It forces the i64 value
  // into a pair of GPRs, which is fine when the value is used as a scalar,
  // but if the i64 value is converted to a vector, we need to undo the VMOVRRD.
  SelectionDAG &DAG = DCI.DAG;
  if (N->getNumOperands() == 2) {
    SDValue RV = PerformVMOVDRRCombine(N, DAG);
    if (RV.getNode())
      return RV;
  }

  // Load i64 elements as f64 values so that type legalization does not split
  // them up into i32 values.
  EVT VT = N->getValueType(0);
  if (VT.getVectorElementType() != MVT::i64 || !hasNormalLoadOperand(N))
    return SDValue();
  SDLoc dl(N);
  SmallVector<SDValue, 8> Ops;
  unsigned NumElts = VT.getVectorNumElements();
  for (unsigned i = 0; i < NumElts; ++i) {
    SDValue V = DAG.getNode(ISD::BITCAST, dl, MVT::f64, N->getOperand(i));
    Ops.push_back(V);
    // Make the DAGCombiner fold the bitcast.
    DCI.AddToWorklist(V.getNode());
  }
  EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f64, NumElts);
  SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, dl, FloatVT, Ops);
  return DAG.getNode(ISD::BITCAST, dl, VT, BV);
}

/// \brief Target-specific dag combine xforms for ARMISD::BUILD_VECTOR.
static SDValue
PerformARMBUILD_VECTORCombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI) {
  // ARMISD::BUILD_VECTOR is introduced when legalizing ISD::BUILD_VECTOR.
  // At that time, we may have inserted bitcasts from integer to float.
  // If these bitcasts have survived DAGCombine, change the lowering of this
  // BUILD_VECTOR in something more vector friendly, i.e., that does not
  // force to use floating point types.

  // Make sure we can change the type of the vector.
  // This is possible iff:
  // 1. The vector is only used in a bitcast to a integer type. I.e.,
  //    1.1. Vector is used only once.
  //    1.2. Use is a bit convert to an integer type.
  // 2. The size of its operands are 32-bits (64-bits are not legal).
  EVT VT = N->getValueType(0);
  EVT EltVT = VT.getVectorElementType();

  // Check 1.1. and 2.
  if (EltVT.getSizeInBits() != 32 || !N->hasOneUse())
    return SDValue();

  // By construction, the input type must be float.
  assert(EltVT == MVT::f32 && "Unexpected type!");

  // Check 1.2.
  SDNode *Use = *N->use_begin();
  if (Use->getOpcode() != ISD::BITCAST ||
      Use->getValueType(0).isFloatingPoint())
    return SDValue();

  // Check profitability.
  // Model is, if more than half of the relevant operands are bitcast from
  // i32, turn the build_vector into a sequence of insert_vector_elt.
  // Relevant operands are everything that is not statically
  // (i.e., at compile time) bitcasted.
  unsigned NumOfBitCastedElts = 0;
  unsigned NumElts = VT.getVectorNumElements();
  unsigned NumOfRelevantElts = NumElts;
  for (unsigned Idx = 0; Idx < NumElts; ++Idx) {
    SDValue Elt = N->getOperand(Idx);
    if (Elt->getOpcode() == ISD::BITCAST) {
      // Assume only bit cast to i32 will go away.
      if (Elt->getOperand(0).getValueType() == MVT::i32)
        ++NumOfBitCastedElts;
    } else if (Elt.getOpcode() == ISD::UNDEF || isa<ConstantSDNode>(Elt))
      // Constants are statically casted, thus do not count them as
      // relevant operands.
      --NumOfRelevantElts;
  }

  // Check if more than half of the elements require a non-free bitcast.
  if (NumOfBitCastedElts <= NumOfRelevantElts / 2)
    return SDValue();

  SelectionDAG &DAG = DCI.DAG;
  // Create the new vector type.
  EVT VecVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, NumElts);
  // Check if the type is legal.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  if (!TLI.isTypeLegal(VecVT))
    return SDValue();

  // Combine:
  // ARMISD::BUILD_VECTOR E1, E2, ..., EN.
  // => BITCAST INSERT_VECTOR_ELT
  //                      (INSERT_VECTOR_ELT (...), (BITCAST EN-1), N-1),
  //                      (BITCAST EN), N.
  SDValue Vec = DAG.getUNDEF(VecVT);
  SDLoc dl(N);
  for (unsigned Idx = 0 ; Idx < NumElts; ++Idx) {
    SDValue V = N->getOperand(Idx);
    if (V.getOpcode() == ISD::UNDEF)
      continue;
    if (V.getOpcode() == ISD::BITCAST &&
        V->getOperand(0).getValueType() == MVT::i32)
      // Fold obvious case.
      V = V.getOperand(0);
    else {
      V = DAG.getNode(ISD::BITCAST, SDLoc(V), MVT::i32, V);
      // Make the DAGCombiner fold the bitcasts.
      DCI.AddToWorklist(V.getNode());
    }
    SDValue LaneIdx = DAG.getConstant(Idx, MVT::i32);
    Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VecVT, Vec, V, LaneIdx);
  }
  Vec = DAG.getNode(ISD::BITCAST, dl, VT, Vec);
  // Make the DAGCombiner fold the bitcasts.
  DCI.AddToWorklist(Vec.getNode());
  return Vec;
}

/// PerformInsertEltCombine - Target-specific dag combine xforms for
/// ISD::INSERT_VECTOR_ELT.
static SDValue PerformInsertEltCombine(SDNode *N,
                                       TargetLowering::DAGCombinerInfo &DCI) {
  // Bitcast an i64 load inserted into a vector to f64.
  // Otherwise, the i64 value will be legalized to a pair of i32 values.
  EVT VT = N->getValueType(0);
  SDNode *Elt = N->getOperand(1).getNode();
  if (VT.getVectorElementType() != MVT::i64 ||
      !ISD::isNormalLoad(Elt) || cast<LoadSDNode>(Elt)->isVolatile())
    return SDValue();

  SelectionDAG &DAG = DCI.DAG;
  SDLoc dl(N);
  EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f64,
                                 VT.getVectorNumElements());
  SDValue Vec = DAG.getNode(ISD::BITCAST, dl, FloatVT, N->getOperand(0));
  SDValue V = DAG.getNode(ISD::BITCAST, dl, MVT::f64, N->getOperand(1));
  // Make the DAGCombiner fold the bitcasts.
  DCI.AddToWorklist(Vec.getNode());
  DCI.AddToWorklist(V.getNode());
  SDValue InsElt = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, FloatVT,
                               Vec, V, N->getOperand(2));
  return DAG.getNode(ISD::BITCAST, dl, VT, InsElt);
}

/// PerformVECTOR_SHUFFLECombine - Target-specific dag combine xforms for
/// ISD::VECTOR_SHUFFLE.
static SDValue PerformVECTOR_SHUFFLECombine(SDNode *N, SelectionDAG &DAG) {
  // The LLVM shufflevector instruction does not require the shuffle mask
  // length to match the operand vector length, but ISD::VECTOR_SHUFFLE does
  // have that requirement.  When translating to ISD::VECTOR_SHUFFLE, if the
  // operands do not match the mask length, they are extended by concatenating
  // them with undef vectors.  That is probably the right thing for other
  // targets, but for NEON it is better to concatenate two double-register
  // size vector operands into a single quad-register size vector.  Do that
  // transformation here:
  //   shuffle(concat(v1, undef), concat(v2, undef)) ->
  //   shuffle(concat(v1, v2), undef)
  SDValue Op0 = N->getOperand(0);
  SDValue Op1 = N->getOperand(1);
  if (Op0.getOpcode() != ISD::CONCAT_VECTORS ||
      Op1.getOpcode() != ISD::CONCAT_VECTORS ||
      Op0.getNumOperands() != 2 ||
      Op1.getNumOperands() != 2)
    return SDValue();
  SDValue Concat0Op1 = Op0.getOperand(1);
  SDValue Concat1Op1 = Op1.getOperand(1);
  if (Concat0Op1.getOpcode() != ISD::UNDEF ||
      Concat1Op1.getOpcode() != ISD::UNDEF)
    return SDValue();
  // Skip the transformation if any of the types are illegal.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT VT = N->getValueType(0);
  if (!TLI.isTypeLegal(VT) ||
      !TLI.isTypeLegal(Concat0Op1.getValueType()) ||
      !TLI.isTypeLegal(Concat1Op1.getValueType()))
    return SDValue();

  SDValue NewConcat = DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(N), VT,
                                  Op0.getOperand(0), Op1.getOperand(0));
  // Translate the shuffle mask.
  SmallVector<int, 16> NewMask;
  unsigned NumElts = VT.getVectorNumElements();
  unsigned HalfElts = NumElts/2;
  ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N);
  for (unsigned n = 0; n < NumElts; ++n) {
    int MaskElt = SVN->getMaskElt(n);
    int NewElt = -1;
    if (MaskElt < (int)HalfElts)
      NewElt = MaskElt;
    else if (MaskElt >= (int)NumElts && MaskElt < (int)(NumElts + HalfElts))
      NewElt = HalfElts + MaskElt - NumElts;
    NewMask.push_back(NewElt);
  }
  return DAG.getVectorShuffle(VT, SDLoc(N), NewConcat,
                              DAG.getUNDEF(VT), NewMask.data());
}

/// CombineBaseUpdate - Target-specific DAG combine function for VLDDUP and
/// NEON load/store intrinsics to merge base address updates.
static SDValue CombineBaseUpdate(SDNode *N,
                                 TargetLowering::DAGCombinerInfo &DCI) {
  if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
    return SDValue();

  SelectionDAG &DAG = DCI.DAG;
  bool isIntrinsic = (N->getOpcode() == ISD::INTRINSIC_VOID ||
                      N->getOpcode() == ISD::INTRINSIC_W_CHAIN);
  unsigned AddrOpIdx = (isIntrinsic ? 2 : 1);
  SDValue Addr = N->getOperand(AddrOpIdx);

  // Search for a use of the address operand that is an increment.
  for (SDNode::use_iterator UI = Addr.getNode()->use_begin(),
         UE = Addr.getNode()->use_end(); UI != UE; ++UI) {
    SDNode *User = *UI;
    if (User->getOpcode() != ISD::ADD ||
        UI.getUse().getResNo() != Addr.getResNo())
      continue;

    // Check that the add is independent of the load/store.  Otherwise, folding
    // it would create a cycle.
    if (User->isPredecessorOf(N) || N->isPredecessorOf(User))
      continue;

    // Find the new opcode for the updating load/store.
    bool isLoad = true;
    bool isLaneOp = false;
    unsigned NewOpc = 0;
    unsigned NumVecs = 0;
    if (isIntrinsic) {
      unsigned IntNo = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
      switch (IntNo) {
      default: llvm_unreachable("unexpected intrinsic for Neon base update");
      case Intrinsic::arm_neon_vld1:     NewOpc = ARMISD::VLD1_UPD;
        NumVecs = 1; break;
      case Intrinsic::arm_neon_vld2:     NewOpc = ARMISD::VLD2_UPD;
        NumVecs = 2; break;
      case Intrinsic::arm_neon_vld3:     NewOpc = ARMISD::VLD3_UPD;
        NumVecs = 3; break;
      case Intrinsic::arm_neon_vld4:     NewOpc = ARMISD::VLD4_UPD;
        NumVecs = 4; break;
      case Intrinsic::arm_neon_vld2lane: NewOpc = ARMISD::VLD2LN_UPD;
        NumVecs = 2; isLaneOp = true; break;
      case Intrinsic::arm_neon_vld3lane: NewOpc = ARMISD::VLD3LN_UPD;
        NumVecs = 3; isLaneOp = true; break;
      case Intrinsic::arm_neon_vld4lane: NewOpc = ARMISD::VLD4LN_UPD;
        NumVecs = 4; isLaneOp = true; break;
      case Intrinsic::arm_neon_vst1:     NewOpc = ARMISD::VST1_UPD;
        NumVecs = 1; isLoad = false; break;
      case Intrinsic::arm_neon_vst2:     NewOpc = ARMISD::VST2_UPD;
        NumVecs = 2; isLoad = false; break;
      case Intrinsic::arm_neon_vst3:     NewOpc = ARMISD::VST3_UPD;
        NumVecs = 3; isLoad = false; break;
      case Intrinsic::arm_neon_vst4:     NewOpc = ARMISD::VST4_UPD;
        NumVecs = 4; isLoad = false; break;
      case Intrinsic::arm_neon_vst2lane: NewOpc = ARMISD::VST2LN_UPD;
        NumVecs = 2; isLoad = false; isLaneOp = true; break;
      case Intrinsic::arm_neon_vst3lane: NewOpc = ARMISD::VST3LN_UPD;
        NumVecs = 3; isLoad = false; isLaneOp = true; break;
      case Intrinsic::arm_neon_vst4lane: NewOpc = ARMISD::VST4LN_UPD;
        NumVecs = 4; isLoad = false; isLaneOp = true; break;
      }
    } else {
      isLaneOp = true;
      switch (N->getOpcode()) {
      default: llvm_unreachable("unexpected opcode for Neon base update");
      case ARMISD::VLD2DUP: NewOpc = ARMISD::VLD2DUP_UPD; NumVecs = 2; break;
      case ARMISD::VLD3DUP: NewOpc = ARMISD::VLD3DUP_UPD; NumVecs = 3; break;
      case ARMISD::VLD4DUP: NewOpc = ARMISD::VLD4DUP_UPD; NumVecs = 4; break;
      }
    }

    // Find the size of memory referenced by the load/store.
    EVT VecTy;
    if (isLoad)
      VecTy = N->getValueType(0);
    else
      VecTy = N->getOperand(AddrOpIdx+1).getValueType();
    unsigned NumBytes = NumVecs * VecTy.getSizeInBits() / 8;
    if (isLaneOp)
      NumBytes /= VecTy.getVectorNumElements();

    // If the increment is a constant, it must match the memory ref size.
    SDValue Inc = User->getOperand(User->getOperand(0) == Addr ? 1 : 0);
    if (ConstantSDNode *CInc = dyn_cast<ConstantSDNode>(Inc.getNode())) {
      uint64_t IncVal = CInc->getZExtValue();
      if (IncVal != NumBytes)
        continue;
    } else if (NumBytes >= 3 * 16) {
      // VLD3/4 and VST3/4 for 128-bit vectors are implemented with two
      // separate instructions that make it harder to use a non-constant update.
      continue;
    }

    // Create the new updating load/store node.
    EVT Tys[6];
    unsigned NumResultVecs = (isLoad ? NumVecs : 0);
    unsigned n;
    for (n = 0; n < NumResultVecs; ++n)
      Tys[n] = VecTy;
    Tys[n++] = MVT::i32;
    Tys[n] = MVT::Other;
    SDVTList SDTys = DAG.getVTList(makeArrayRef(Tys, NumResultVecs+2));
    SmallVector<SDValue, 8> Ops;
    Ops.push_back(N->getOperand(0)); // incoming chain
    Ops.push_back(N->getOperand(AddrOpIdx));
    Ops.push_back(Inc);
    for (unsigned i = AddrOpIdx + 1; i < N->getNumOperands(); ++i) {
      Ops.push_back(N->getOperand(i));
    }
    MemIntrinsicSDNode *MemInt = cast<MemIntrinsicSDNode>(N);
    SDValue UpdN = DAG.getMemIntrinsicNode(NewOpc, SDLoc(N), SDTys,
                                           Ops, MemInt->getMemoryVT(),
                                           MemInt->getMemOperand());

    // Update the uses.
    std::vector<SDValue> NewResults;
    for (unsigned i = 0; i < NumResultVecs; ++i) {
      NewResults.push_back(SDValue(UpdN.getNode(), i));
    }
    NewResults.push_back(SDValue(UpdN.getNode(), NumResultVecs+1)); // chain
    DCI.CombineTo(N, NewResults);
    DCI.CombineTo(User, SDValue(UpdN.getNode(), NumResultVecs));

    break;
  }
  return SDValue();
}

/// CombineVLDDUP - For a VDUPLANE node N, check if its source operand is a
/// vldN-lane (N > 1) intrinsic, and if all the other uses of that intrinsic
/// are also VDUPLANEs.  If so, combine them to a vldN-dup operation and
/// return true.
static bool CombineVLDDUP(SDNode *N, TargetLowering::DAGCombinerInfo &DCI) {
  SelectionDAG &DAG = DCI.DAG;
  EVT VT = N->getValueType(0);
  // vldN-dup instructions only support 64-bit vectors for N > 1.
  if (!VT.is64BitVector())
    return false;

  // Check if the VDUPLANE operand is a vldN-dup intrinsic.
  SDNode *VLD = N->getOperand(0).getNode();
  if (VLD->getOpcode() != ISD::INTRINSIC_W_CHAIN)
    return false;
  unsigned NumVecs = 0;
  unsigned NewOpc = 0;
  unsigned IntNo = cast<ConstantSDNode>(VLD->getOperand(1))->getZExtValue();
  if (IntNo == Intrinsic::arm_neon_vld2lane) {
    NumVecs = 2;
    NewOpc = ARMISD::VLD2DUP;
  } else if (IntNo == Intrinsic::arm_neon_vld3lane) {
    NumVecs = 3;
    NewOpc = ARMISD::VLD3DUP;
  } else if (IntNo == Intrinsic::arm_neon_vld4lane) {
    NumVecs = 4;
    NewOpc = ARMISD::VLD4DUP;
  } else {
    return false;
  }

  // First check that all the vldN-lane uses are VDUPLANEs and that the lane
  // numbers match the load.
  unsigned VLDLaneNo =
    cast<ConstantSDNode>(VLD->getOperand(NumVecs+3))->getZExtValue();
  for (SDNode::use_iterator UI = VLD->use_begin(), UE = VLD->use_end();
       UI != UE; ++UI) {
    // Ignore uses of the chain result.
    if (UI.getUse().getResNo() == NumVecs)
      continue;
    SDNode *User = *UI;
    if (User->getOpcode() != ARMISD::VDUPLANE ||
        VLDLaneNo != cast<ConstantSDNode>(User->getOperand(1))->getZExtValue())
      return false;
  }

  // Create the vldN-dup node.
  EVT Tys[5];
  unsigned n;
  for (n = 0; n < NumVecs; ++n)
    Tys[n] = VT;
  Tys[n] = MVT::Other;
  SDVTList SDTys = DAG.getVTList(makeArrayRef(Tys, NumVecs+1));
  SDValue Ops[] = { VLD->getOperand(0), VLD->getOperand(2) };
  MemIntrinsicSDNode *VLDMemInt = cast<MemIntrinsicSDNode>(VLD);
  SDValue VLDDup = DAG.getMemIntrinsicNode(NewOpc, SDLoc(VLD), SDTys,
                                           Ops, VLDMemInt->getMemoryVT(),
                                           VLDMemInt->getMemOperand());

  // Update the uses.
  for (SDNode::use_iterator UI = VLD->use_begin(), UE = VLD->use_end();
       UI != UE; ++UI) {
    unsigned ResNo = UI.getUse().getResNo();
    // Ignore uses of the chain result.
    if (ResNo == NumVecs)
      continue;
    SDNode *User = *UI;
    DCI.CombineTo(User, SDValue(VLDDup.getNode(), ResNo));
  }

  // Now the vldN-lane intrinsic is dead except for its chain result.
  // Update uses of the chain.
  std::vector<SDValue> VLDDupResults;
  for (unsigned n = 0; n < NumVecs; ++n)
    VLDDupResults.push_back(SDValue(VLDDup.getNode(), n));
  VLDDupResults.push_back(SDValue(VLDDup.getNode(), NumVecs));
  DCI.CombineTo(VLD, VLDDupResults);

  return true;
}

/// PerformVDUPLANECombine - Target-specific dag combine xforms for
/// ARMISD::VDUPLANE.
static SDValue PerformVDUPLANECombine(SDNode *N,
                                      TargetLowering::DAGCombinerInfo &DCI) {
  SDValue Op = N->getOperand(0);

  // If the source is a vldN-lane (N > 1) intrinsic, and all the other uses
  // of that intrinsic are also VDUPLANEs, combine them to a vldN-dup operation.
  if (CombineVLDDUP(N, DCI))
    return SDValue(N, 0);

  // If the source is already a VMOVIMM or VMVNIMM splat, the VDUPLANE is
  // redundant.  Ignore bit_converts for now; element sizes are checked below.
  while (Op.getOpcode() == ISD::BITCAST)
    Op = Op.getOperand(0);
  if (Op.getOpcode() != ARMISD::VMOVIMM && Op.getOpcode() != ARMISD::VMVNIMM)
    return SDValue();

  // Make sure the VMOV element size is not bigger than the VDUPLANE elements.
  unsigned EltSize = Op.getValueType().getVectorElementType().getSizeInBits();
  // The canonical VMOV for a zero vector uses a 32-bit element size.
  unsigned Imm = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
  unsigned EltBits;
  if (ARM_AM::decodeNEONModImm(Imm, EltBits) == 0)
    EltSize = 8;
  EVT VT = N->getValueType(0);
  if (EltSize > VT.getVectorElementType().getSizeInBits())
    return SDValue();

  return DCI.DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Op);
}

// isConstVecPow2 - Return true if each vector element is a power of 2, all
// elements are the same constant, C, and Log2(C) ranges from 1 to 32.
static bool isConstVecPow2(SDValue ConstVec, bool isSigned, uint64_t &C)
{
  integerPart cN;
  integerPart c0 = 0;
  for (unsigned I = 0, E = ConstVec.getValueType().getVectorNumElements();
       I != E; I++) {
    ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(ConstVec.getOperand(I));
    if (!C)
      return false;

    bool isExact;
    APFloat APF = C->getValueAPF();
    if (APF.convertToInteger(&cN, 64, isSigned, APFloat::rmTowardZero, &isExact)
        != APFloat::opOK || !isExact)
      return false;

    c0 = (I == 0) ? cN : c0;
    if (!isPowerOf2_64(cN) || c0 != cN || Log2_64(c0) < 1 || Log2_64(c0) > 32)
      return false;
  }
  C = c0;
  return true;
}

/// PerformVCVTCombine - VCVT (floating-point to fixed-point, Advanced SIMD)
/// can replace combinations of VMUL and VCVT (floating-point to integer)
/// when the VMUL has a constant operand that is a power of 2.
///
/// Example (assume d17 = <float 8.000000e+00, float 8.000000e+00>):
///  vmul.f32        d16, d17, d16
///  vcvt.s32.f32    d16, d16
/// becomes:
///  vcvt.s32.f32    d16, d16, #3
static SDValue PerformVCVTCombine(SDNode *N,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const ARMSubtarget *Subtarget) {
  SelectionDAG &DAG = DCI.DAG;
  SDValue Op = N->getOperand(0);

  if (!Subtarget->hasNEON() || !Op.getValueType().isVector() ||
      Op.getOpcode() != ISD::FMUL)
    return SDValue();

  uint64_t C;
  SDValue N0 = Op->getOperand(0);
  SDValue ConstVec = Op->getOperand(1);
  bool isSigned = N->getOpcode() == ISD::FP_TO_SINT;

  if (ConstVec.getOpcode() != ISD::BUILD_VECTOR ||
      !isConstVecPow2(ConstVec, isSigned, C))
    return SDValue();

  MVT FloatTy = Op.getSimpleValueType().getVectorElementType();
  MVT IntTy = N->getSimpleValueType(0).getVectorElementType();
  if (FloatTy.getSizeInBits() != 32 || IntTy.getSizeInBits() > 32) {
    // These instructions only exist converting from f32 to i32. We can handle
    // smaller integers by generating an extra truncate, but larger ones would
    // be lossy.
    return SDValue();
  }

  unsigned IntrinsicOpcode = isSigned ? Intrinsic::arm_neon_vcvtfp2fxs :
    Intrinsic::arm_neon_vcvtfp2fxu;
  unsigned NumLanes = Op.getValueType().getVectorNumElements();
  SDValue FixConv =  DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N),
                                 NumLanes == 2 ? MVT::v2i32 : MVT::v4i32,
                                 DAG.getConstant(IntrinsicOpcode, MVT::i32), N0,
                                 DAG.getConstant(Log2_64(C), MVT::i32));

  if (IntTy.getSizeInBits() < FloatTy.getSizeInBits())
    FixConv = DAG.getNode(ISD::TRUNCATE, SDLoc(N), N->getValueType(0), FixConv);

  return FixConv;
}

/// PerformVDIVCombine - VCVT (fixed-point to floating-point, Advanced SIMD)
/// can replace combinations of VCVT (integer to floating-point) and VDIV
/// when the VDIV has a constant operand that is a power of 2.
///
/// Example (assume d17 = <float 8.000000e+00, float 8.000000e+00>):
///  vcvt.f32.s32    d16, d16
///  vdiv.f32        d16, d17, d16
/// becomes:
///  vcvt.f32.s32    d16, d16, #3
static SDValue PerformVDIVCombine(SDNode *N,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const ARMSubtarget *Subtarget) {
  SelectionDAG &DAG = DCI.DAG;
  SDValue Op = N->getOperand(0);
  unsigned OpOpcode = Op.getNode()->getOpcode();

  if (!Subtarget->hasNEON() || !N->getValueType(0).isVector() ||
      (OpOpcode != ISD::SINT_TO_FP && OpOpcode != ISD::UINT_TO_FP))
    return SDValue();

  uint64_t C;
  SDValue ConstVec = N->getOperand(1);
  bool isSigned = OpOpcode == ISD::SINT_TO_FP;

  if (ConstVec.getOpcode() != ISD::BUILD_VECTOR ||
      !isConstVecPow2(ConstVec, isSigned, C))
    return SDValue();

  MVT FloatTy = N->getSimpleValueType(0).getVectorElementType();
  MVT IntTy = Op.getOperand(0).getSimpleValueType().getVectorElementType();
  if (FloatTy.getSizeInBits() != 32 || IntTy.getSizeInBits() > 32) {
    // These instructions only exist converting from i32 to f32. We can handle
    // smaller integers by generating an extra extend, but larger ones would
    // be lossy.
    return SDValue();
  }

  SDValue ConvInput = Op.getOperand(0);
  unsigned NumLanes = Op.getValueType().getVectorNumElements();
  if (IntTy.getSizeInBits() < FloatTy.getSizeInBits())
    ConvInput = DAG.getNode(isSigned ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND,
                            SDLoc(N), NumLanes == 2 ? MVT::v2i32 : MVT::v4i32,
                            ConvInput);

  unsigned IntrinsicOpcode = isSigned ? Intrinsic::arm_neon_vcvtfxs2fp :
    Intrinsic::arm_neon_vcvtfxu2fp;
  return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N),
                     Op.getValueType(),
                     DAG.getConstant(IntrinsicOpcode, MVT::i32),
                     ConvInput, DAG.getConstant(Log2_64(C), MVT::i32));
}

/// Getvshiftimm - Check if this is a valid build_vector for the immediate
/// operand of a vector shift operation, where all the elements of the
/// build_vector must have the same constant integer value.
static bool getVShiftImm(SDValue Op, unsigned ElementBits, int64_t &Cnt) {
  // Ignore bit_converts.
  while (Op.getOpcode() == ISD::BITCAST)
    Op = Op.getOperand(0);
  BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(Op.getNode());
  APInt SplatBits, SplatUndef;
  unsigned SplatBitSize;
  bool HasAnyUndefs;
  if (! BVN || ! BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize,
                                      HasAnyUndefs, ElementBits) ||
      SplatBitSize > ElementBits)
    return false;
  Cnt = SplatBits.getSExtValue();
  return true;
}

/// isVShiftLImm - Check if this is a valid build_vector for the immediate
/// operand of a vector shift left operation.  That value must be in the range:
///   0 <= Value < ElementBits for a left shift; or
///   0 <= Value <= ElementBits for a long left shift.
static bool isVShiftLImm(SDValue Op, EVT VT, bool isLong, int64_t &Cnt) {
  assert(VT.isVector() && "vector shift count is not a vector type");
  unsigned ElementBits = VT.getVectorElementType().getSizeInBits();
  if (! getVShiftImm(Op, ElementBits, Cnt))
    return false;
  return (Cnt >= 0 && (isLong ? Cnt-1 : Cnt) < ElementBits);
}

/// isVShiftRImm - Check if this is a valid build_vector for the immediate
/// operand of a vector shift right operation.  For a shift opcode, the value
/// is positive, but for an intrinsic the value count must be negative. The
/// absolute value must be in the range:
///   1 <= |Value| <= ElementBits for a right shift; or
///   1 <= |Value| <= ElementBits/2 for a narrow right shift.
static bool isVShiftRImm(SDValue Op, EVT VT, bool isNarrow, bool isIntrinsic,
                         int64_t &Cnt) {
  assert(VT.isVector() && "vector shift count is not a vector type");
  unsigned ElementBits = VT.getVectorElementType().getSizeInBits();
  if (! getVShiftImm(Op, ElementBits, Cnt))
    return false;
  if (isIntrinsic)
    Cnt = -Cnt;
  return (Cnt >= 1 && Cnt <= (isNarrow ? ElementBits/2 : ElementBits));
}

/// PerformIntrinsicCombine - ARM-specific DAG combining for intrinsics.
static SDValue PerformIntrinsicCombine(SDNode *N, SelectionDAG &DAG) {
  unsigned IntNo = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
  switch (IntNo) {
  default:
    // Don't do anything for most intrinsics.
    break;

  // Vector shifts: check for immediate versions and lower them.
  // Note: This is done during DAG combining instead of DAG legalizing because
  // the build_vectors for 64-bit vector element shift counts are generally
  // not legal, and it is hard to see their values after they get legalized to
  // loads from a constant pool.
  case Intrinsic::arm_neon_vshifts:
  case Intrinsic::arm_neon_vshiftu:
  case Intrinsic::arm_neon_vrshifts:
  case Intrinsic::arm_neon_vrshiftu:
  case Intrinsic::arm_neon_vrshiftn:
  case Intrinsic::arm_neon_vqshifts:
  case Intrinsic::arm_neon_vqshiftu:
  case Intrinsic::arm_neon_vqshiftsu:
  case Intrinsic::arm_neon_vqshiftns:
  case Intrinsic::arm_neon_vqshiftnu:
  case Intrinsic::arm_neon_vqshiftnsu:
  case Intrinsic::arm_neon_vqrshiftns:
  case Intrinsic::arm_neon_vqrshiftnu:
  case Intrinsic::arm_neon_vqrshiftnsu: {
    EVT VT = N->getOperand(1).getValueType();
    int64_t Cnt;
    unsigned VShiftOpc = 0;

    switch (IntNo) {
    case Intrinsic::arm_neon_vshifts:
    case Intrinsic::arm_neon_vshiftu:
      if (isVShiftLImm(N->getOperand(2), VT, false, Cnt)) {
        VShiftOpc = ARMISD::VSHL;
        break;
      }
      if (isVShiftRImm(N->getOperand(2), VT, false, true, Cnt)) {
        VShiftOpc = (IntNo == Intrinsic::arm_neon_vshifts ?
                     ARMISD::VSHRs : ARMISD::VSHRu);
        break;
      }
      return SDValue();

    case Intrinsic::arm_neon_vrshifts:
    case Intrinsic::arm_neon_vrshiftu:
      if (isVShiftRImm(N->getOperand(2), VT, false, true, Cnt))
        break;
      return SDValue();

    case Intrinsic::arm_neon_vqshifts:
    case Intrinsic::arm_neon_vqshiftu:
      if (isVShiftLImm(N->getOperand(2), VT, false, Cnt))
        break;
      return SDValue();

    case Intrinsic::arm_neon_vqshiftsu:
      if (isVShiftLImm(N->getOperand(2), VT, false, Cnt))
        break;
      llvm_unreachable("invalid shift count for vqshlu intrinsic");

    case Intrinsic::arm_neon_vrshiftn:
    case Intrinsic::arm_neon_vqshiftns:
    case Intrinsic::arm_neon_vqshiftnu:
    case Intrinsic::arm_neon_vqshiftnsu:
    case Intrinsic::arm_neon_vqrshiftns:
    case Intrinsic::arm_neon_vqrshiftnu:
    case Intrinsic::arm_neon_vqrshiftnsu:
      // Narrowing shifts require an immediate right shift.
      if (isVShiftRImm(N->getOperand(2), VT, true, true, Cnt))
        break;
      llvm_unreachable("invalid shift count for narrowing vector shift "
                       "intrinsic");

    default:
      llvm_unreachable("unhandled vector shift");
    }

    switch (IntNo) {
    case Intrinsic::arm_neon_vshifts:
    case Intrinsic::arm_neon_vshiftu:
      // Opcode already set above.
      break;
    case Intrinsic::arm_neon_vrshifts:
      VShiftOpc = ARMISD::VRSHRs; break;
    case Intrinsic::arm_neon_vrshiftu:
      VShiftOpc = ARMISD::VRSHRu; break;
    case Intrinsic::arm_neon_vrshiftn:
      VShiftOpc = ARMISD::VRSHRN; break;
    case Intrinsic::arm_neon_vqshifts:
      VShiftOpc = ARMISD::VQSHLs; break;
    case Intrinsic::arm_neon_vqshiftu:
      VShiftOpc = ARMISD::VQSHLu; break;
    case Intrinsic::arm_neon_vqshiftsu:
      VShiftOpc = ARMISD::VQSHLsu; break;
    case Intrinsic::arm_neon_vqshiftns:
      VShiftOpc = ARMISD::VQSHRNs; break;
    case Intrinsic::arm_neon_vqshiftnu:
      VShiftOpc = ARMISD::VQSHRNu; break;
    case Intrinsic::arm_neon_vqshiftnsu:
      VShiftOpc = ARMISD::VQSHRNsu; break;
    case Intrinsic::arm_neon_vqrshiftns:
      VShiftOpc = ARMISD::VQRSHRNs; break;
    case Intrinsic::arm_neon_vqrshiftnu:
      VShiftOpc = ARMISD::VQRSHRNu; break;
    case Intrinsic::arm_neon_vqrshiftnsu:
      VShiftOpc = ARMISD::VQRSHRNsu; break;
    }

    return DAG.getNode(VShiftOpc, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), DAG.getConstant(Cnt, MVT::i32));
  }

  case Intrinsic::arm_neon_vshiftins: {
    EVT VT = N->getOperand(1).getValueType();
    int64_t Cnt;
    unsigned VShiftOpc = 0;

    if (isVShiftLImm(N->getOperand(3), VT, false, Cnt))
      VShiftOpc = ARMISD::VSLI;
    else if (isVShiftRImm(N->getOperand(3), VT, false, true, Cnt))
      VShiftOpc = ARMISD::VSRI;
    else {
      llvm_unreachable("invalid shift count for vsli/vsri intrinsic");
    }

    return DAG.getNode(VShiftOpc, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), N->getOperand(2),
                       DAG.getConstant(Cnt, MVT::i32));
  }

  case Intrinsic::arm_neon_vqrshifts:
  case Intrinsic::arm_neon_vqrshiftu:
    // No immediate versions of these to check for.
    break;
  }

  return SDValue();
}

/// PerformShiftCombine - Checks for immediate versions of vector shifts and
/// lowers them.  As with the vector shift intrinsics, this is done during DAG
/// combining instead of DAG legalizing because the build_vectors for 64-bit
/// vector element shift counts are generally not legal, and it is hard to see
/// their values after they get legalized to loads from a constant pool.
static SDValue PerformShiftCombine(SDNode *N, SelectionDAG &DAG,
                                   const ARMSubtarget *ST) {
  EVT VT = N->getValueType(0);
  if (N->getOpcode() == ISD::SRL && VT == MVT::i32 && ST->hasV6Ops()) {
    // Canonicalize (srl (bswap x), 16) to (rotr (bswap x), 16) if the high
    // 16-bits of x is zero. This optimizes rev + lsr 16 to rev16.
    SDValue N1 = N->getOperand(1);
    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N1)) {
      SDValue N0 = N->getOperand(0);
      if (C->getZExtValue() == 16 && N0.getOpcode() == ISD::BSWAP &&
          DAG.MaskedValueIsZero(N0.getOperand(0),
                                APInt::getHighBitsSet(32, 16)))
        return DAG.getNode(ISD::ROTR, SDLoc(N), VT, N0, N1);
    }
  }

  // Nothing to be done for scalar shifts.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  if (!VT.isVector() || !TLI.isTypeLegal(VT))
    return SDValue();

  assert(ST->hasNEON() && "unexpected vector shift");
  int64_t Cnt;

  switch (N->getOpcode()) {
  default: llvm_unreachable("unexpected shift opcode");

  case ISD::SHL:
    if (isVShiftLImm(N->getOperand(1), VT, false, Cnt))
      return DAG.getNode(ARMISD::VSHL, SDLoc(N), VT, N->getOperand(0),
                         DAG.getConstant(Cnt, MVT::i32));
    break;

  case ISD::SRA:
  case ISD::SRL:
    if (isVShiftRImm(N->getOperand(1), VT, false, false, Cnt)) {
      unsigned VShiftOpc = (N->getOpcode() == ISD::SRA ?
                            ARMISD::VSHRs : ARMISD::VSHRu);
      return DAG.getNode(VShiftOpc, SDLoc(N), VT, N->getOperand(0),
                         DAG.getConstant(Cnt, MVT::i32));
    }
  }
  return SDValue();
}

/// PerformExtendCombine - Target-specific DAG combining for ISD::SIGN_EXTEND,
/// ISD::ZERO_EXTEND, and ISD::ANY_EXTEND.
static SDValue PerformExtendCombine(SDNode *N, SelectionDAG &DAG,
                                    const ARMSubtarget *ST) {
  SDValue N0 = N->getOperand(0);

  // Check for sign- and zero-extensions of vector extract operations of 8-
  // and 16-bit vector elements.  NEON supports these directly.  They are
  // handled during DAG combining because type legalization will promote them
  // to 32-bit types and it is messy to recognize the operations after that.
  if (ST->hasNEON() && N0.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
    SDValue Vec = N0.getOperand(0);
    SDValue Lane = N0.getOperand(1);
    EVT VT = N->getValueType(0);
    EVT EltVT = N0.getValueType();
    const TargetLowering &TLI = DAG.getTargetLoweringInfo();

    if (VT == MVT::i32 &&
        (EltVT == MVT::i8 || EltVT == MVT::i16) &&
        TLI.isTypeLegal(Vec.getValueType()) &&
        isa<ConstantSDNode>(Lane)) {

      unsigned Opc = 0;
      switch (N->getOpcode()) {
      default: llvm_unreachable("unexpected opcode");
      case ISD::SIGN_EXTEND:
        Opc = ARMISD::VGETLANEs;
        break;
      case ISD::ZERO_EXTEND:
      case ISD::ANY_EXTEND:
        Opc = ARMISD::VGETLANEu;
        break;
      }
      return DAG.getNode(Opc, SDLoc(N), VT, Vec, Lane);
    }
  }

  return SDValue();
}

/// PerformSELECT_CCCombine - Target-specific DAG combining for ISD::SELECT_CC
/// to match f32 max/min patterns to use NEON vmax/vmin instructions.
static SDValue PerformSELECT_CCCombine(SDNode *N, SelectionDAG &DAG,
                                       const ARMSubtarget *ST) {
  // If the target supports NEON, try to use vmax/vmin instructions for f32
  // selects like "x < y ? x : y".  Unless the NoNaNsFPMath option is set,
  // be careful about NaNs:  NEON's vmax/vmin return NaN if either operand is
  // a NaN; only do the transformation when it matches that behavior.

  // For now only do this when using NEON for FP operations; if using VFP, it
  // is not obvious that the benefit outweighs the cost of switching to the
  // NEON pipeline.
  if (!ST->hasNEON() || !ST->useNEONForSinglePrecisionFP() ||
      N->getValueType(0) != MVT::f32)
    return SDValue();

  SDValue CondLHS = N->getOperand(0);
  SDValue CondRHS = N->getOperand(1);
  SDValue LHS = N->getOperand(2);
  SDValue RHS = N->getOperand(3);
  ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(4))->get();

  unsigned Opcode = 0;
  bool IsReversed;
  if (DAG.isEqualTo(LHS, CondLHS) && DAG.isEqualTo(RHS, CondRHS)) {
    IsReversed = false; // x CC y ? x : y
  } else if (DAG.isEqualTo(LHS, CondRHS) && DAG.isEqualTo(RHS, CondLHS)) {
    IsReversed = true ; // x CC y ? y : x
  } else {
    return SDValue();
  }

  bool IsUnordered;
  switch (CC) {
  default: break;
  case ISD::SETOLT:
  case ISD::SETOLE:
  case ISD::SETLT:
  case ISD::SETLE:
  case ISD::SETULT:
  case ISD::SETULE:
    // If LHS is NaN, an ordered comparison will be false and the result will
    // be the RHS, but vmin(NaN, RHS) = NaN.  Avoid this by checking that LHS
    // != NaN.  Likewise, for unordered comparisons, check for RHS != NaN.
    IsUnordered = (CC == ISD::SETULT || CC == ISD::SETULE);
    if (!DAG.isKnownNeverNaN(IsUnordered ? RHS : LHS))
      break;
    // For less-than-or-equal comparisons, "+0 <= -0" will be true but vmin
    // will return -0, so vmin can only be used for unsafe math or if one of
    // the operands is known to be nonzero.
    if ((CC == ISD::SETLE || CC == ISD::SETOLE || CC == ISD::SETULE) &&
        !DAG.getTarget().Options.UnsafeFPMath &&
        !(DAG.isKnownNeverZero(LHS) || DAG.isKnownNeverZero(RHS)))
      break;
    Opcode = IsReversed ? ARMISD::FMAX : ARMISD::FMIN;
    break;

  case ISD::SETOGT:
  case ISD::SETOGE:
  case ISD::SETGT:
  case ISD::SETGE:
  case ISD::SETUGT:
  case ISD::SETUGE:
    // If LHS is NaN, an ordered comparison will be false and the result will
    // be the RHS, but vmax(NaN, RHS) = NaN.  Avoid this by checking that LHS
    // != NaN.  Likewise, for unordered comparisons, check for RHS != NaN.
    IsUnordered = (CC == ISD::SETUGT || CC == ISD::SETUGE);
    if (!DAG.isKnownNeverNaN(IsUnordered ? RHS : LHS))
      break;
    // For greater-than-or-equal comparisons, "-0 >= +0" will be true but vmax
    // will return +0, so vmax can only be used for unsafe math or if one of
    // the operands is known to be nonzero.
    if ((CC == ISD::SETGE || CC == ISD::SETOGE || CC == ISD::SETUGE) &&
        !DAG.getTarget().Options.UnsafeFPMath &&
        !(DAG.isKnownNeverZero(LHS) || DAG.isKnownNeverZero(RHS)))
      break;
    Opcode = IsReversed ? ARMISD::FMIN : ARMISD::FMAX;
    break;
  }

  if (!Opcode)
    return SDValue();
  return DAG.getNode(Opcode, SDLoc(N), N->getValueType(0), LHS, RHS);
}

/// PerformCMOVCombine - Target-specific DAG combining for ARMISD::CMOV.
SDValue
ARMTargetLowering::PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const {
  SDValue Cmp = N->getOperand(4);
  if (Cmp.getOpcode() != ARMISD::CMPZ)
    // Only looking at EQ and NE cases.
    return SDValue();

  EVT VT = N->getValueType(0);
  SDLoc dl(N);
  SDValue LHS = Cmp.getOperand(0);
  SDValue RHS = Cmp.getOperand(1);
  SDValue FalseVal = N->getOperand(0);
  SDValue TrueVal = N->getOperand(1);
  SDValue ARMcc = N->getOperand(2);
  ARMCC::CondCodes CC =
    (ARMCC::CondCodes)cast<ConstantSDNode>(ARMcc)->getZExtValue();

  // Simplify
  //   mov     r1, r0
  //   cmp     r1, x
  //   mov     r0, y
  //   moveq   r0, x
  // to
  //   cmp     r0, x
  //   movne   r0, y
  //
  //   mov     r1, r0
  //   cmp     r1, x
  //   mov     r0, x
  //   movne   r0, y
  // to
  //   cmp     r0, x
  //   movne   r0, y
  /// FIXME: Turn this into a target neutral optimization?
  SDValue Res;
  if (CC == ARMCC::NE && FalseVal == RHS && FalseVal != LHS) {
    Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, TrueVal, ARMcc,
                      N->getOperand(3), Cmp);
  } else if (CC == ARMCC::EQ && TrueVal == RHS) {
    SDValue ARMcc;
    SDValue NewCmp = getARMCmp(LHS, RHS, ISD::SETNE, ARMcc, DAG, dl);
    Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, FalseVal, ARMcc,
                      N->getOperand(3), NewCmp);
  }

  if (Res.getNode()) {
    APInt KnownZero, KnownOne;
    DAG.computeKnownBits(SDValue(N,0), KnownZero, KnownOne);
    // Capture demanded bits information that would be otherwise lost.
    if (KnownZero == 0xfffffffe)
      Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res,
                        DAG.getValueType(MVT::i1));
    else if (KnownZero == 0xffffff00)
      Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res,
                        DAG.getValueType(MVT::i8));
    else if (KnownZero == 0xffff0000)
      Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res,
                        DAG.getValueType(MVT::i16));
  }

  return Res;
}

SDValue ARMTargetLowering::PerformDAGCombine(SDNode *N,
                                             DAGCombinerInfo &DCI) const {
  switch (N->getOpcode()) {
  default: break;
  case ISD::ADDC:       return PerformADDCCombine(N, DCI, Subtarget);
  case ISD::ADD:        return PerformADDCombine(N, DCI, Subtarget);
  case ISD::SUB:        return PerformSUBCombine(N, DCI);
  case ISD::MUL:        return PerformMULCombine(N, DCI, Subtarget);
  case ISD::OR:         return PerformORCombine(N, DCI, Subtarget);
  case ISD::XOR:        return PerformXORCombine(N, DCI, Subtarget);
  case ISD::AND:        return PerformANDCombine(N, DCI, Subtarget);
  case ARMISD::BFI:     return PerformBFICombine(N, DCI);
  case ARMISD::VMOVRRD: return PerformVMOVRRDCombine(N, DCI, Subtarget);
  case ARMISD::VMOVDRR: return PerformVMOVDRRCombine(N, DCI.DAG);
  case ISD::STORE:      return PerformSTORECombine(N, DCI);
  case ISD::BUILD_VECTOR: return PerformBUILD_VECTORCombine(N, DCI, Subtarget);
  case ISD::INSERT_VECTOR_ELT: return PerformInsertEltCombine(N, DCI);
  case ISD::VECTOR_SHUFFLE: return PerformVECTOR_SHUFFLECombine(N, DCI.DAG);
  case ARMISD::VDUPLANE: return PerformVDUPLANECombine(N, DCI);
  case ISD::FP_TO_SINT:
  case ISD::FP_TO_UINT: return PerformVCVTCombine(N, DCI, Subtarget);
  case ISD::FDIV:       return PerformVDIVCombine(N, DCI, Subtarget);
  case ISD::INTRINSIC_WO_CHAIN: return PerformIntrinsicCombine(N, DCI.DAG);
  case ISD::SHL:
  case ISD::SRA:
  case ISD::SRL:        return PerformShiftCombine(N, DCI.DAG, Subtarget);
  case ISD::SIGN_EXTEND:
  case ISD::ZERO_EXTEND:
  case ISD::ANY_EXTEND: return PerformExtendCombine(N, DCI.DAG, Subtarget);
  case ISD::SELECT_CC:  return PerformSELECT_CCCombine(N, DCI.DAG, Subtarget);
  case ARMISD::CMOV: return PerformCMOVCombine(N, DCI.DAG);
  case ARMISD::VLD2DUP:
  case ARMISD::VLD3DUP:
  case ARMISD::VLD4DUP:
    return CombineBaseUpdate(N, DCI);
  case ARMISD::BUILD_VECTOR:
    return PerformARMBUILD_VECTORCombine(N, DCI);
  case ISD::INTRINSIC_VOID:
  case ISD::INTRINSIC_W_CHAIN:
    switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
    case Intrinsic::arm_neon_vld1:
    case Intrinsic::arm_neon_vld2:
    case Intrinsic::arm_neon_vld3:
    case Intrinsic::arm_neon_vld4:
    case Intrinsic::arm_neon_vld2lane:
    case Intrinsic::arm_neon_vld3lane:
    case Intrinsic::arm_neon_vld4lane:
    case Intrinsic::arm_neon_vst1:
    case Intrinsic::arm_neon_vst2:
    case Intrinsic::arm_neon_vst3:
    case Intrinsic::arm_neon_vst4:
    case Intrinsic::arm_neon_vst2lane:
    case Intrinsic::arm_neon_vst3lane:
    case Intrinsic::arm_neon_vst4lane:
      return CombineBaseUpdate(N, DCI);
    default: break;
    }
    break;
  }
  return SDValue();
}

bool ARMTargetLowering::isDesirableToTransformToIntegerOp(unsigned Opc,
                                                          EVT VT) const {
  return (VT == MVT::f32) && (Opc == ISD::LOAD || Opc == ISD::STORE);
}

bool ARMTargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
                                                       unsigned,
                                                       unsigned,
                                                       bool *Fast) const {
  // The AllowsUnaliged flag models the SCTLR.A setting in ARM cpus
  bool AllowsUnaligned = Subtarget->allowsUnalignedMem();

  switch (VT.getSimpleVT().SimpleTy) {
  default:
    return false;
  case MVT::i8:
  case MVT::i16:
  case MVT::i32: {
    // Unaligned access can use (for example) LRDB, LRDH, LDR
    if (AllowsUnaligned) {
      if (Fast)
        *Fast = Subtarget->hasV7Ops();
      return true;
    }
    return false;
  }
  case MVT::f64:
  case MVT::v2f64: {
    // For any little-endian targets with neon, we can support unaligned ld/st
    // of D and Q (e.g. {D0,D1}) registers by using vld1.i8/vst1.i8.
    // A big-endian target may also explicitly support unaligned accesses
    if (Subtarget->hasNEON() && (AllowsUnaligned || isLittleEndian())) {
      if (Fast)
        *Fast = true;
      return true;
    }
    return false;
  }
  }
}

static bool memOpAlign(unsigned DstAlign, unsigned SrcAlign,
                       unsigned AlignCheck) {
  return ((SrcAlign == 0 || SrcAlign % AlignCheck == 0) &&
          (DstAlign == 0 || DstAlign % AlignCheck == 0));
}

EVT ARMTargetLowering::getOptimalMemOpType(uint64_t Size,
                                           unsigned DstAlign, unsigned SrcAlign,
                                           bool IsMemset, bool ZeroMemset,
                                           bool MemcpyStrSrc,
                                           MachineFunction &MF) const {
  const Function *F = MF.getFunction();

  // See if we can use NEON instructions for this...
  if ((!IsMemset || ZeroMemset) &&
      Subtarget->hasNEON() &&
      !F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
                                       Attribute::NoImplicitFloat)) {
    bool Fast;
    if (Size >= 16 &&
        (memOpAlign(SrcAlign, DstAlign, 16) ||
         (allowsMisalignedMemoryAccesses(MVT::v2f64, 0, 1, &Fast) && Fast))) {
      return MVT::v2f64;
    } else if (Size >= 8 &&
               (memOpAlign(SrcAlign, DstAlign, 8) ||
                (allowsMisalignedMemoryAccesses(MVT::f64, 0, 1, &Fast) &&
                 Fast))) {
      return MVT::f64;
    }
  }

  // Lowering to i32/i16 if the size permits.
  if (Size >= 4)
    return MVT::i32;
  else if (Size >= 2)
    return MVT::i16;

  // Let the target-independent logic figure it out.
  return MVT::Other;
}

bool ARMTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
  if (Val.getOpcode() != ISD::LOAD)
    return false;

  EVT VT1 = Val.getValueType();
  if (!VT1.isSimple() || !VT1.isInteger() ||
      !VT2.isSimple() || !VT2.isInteger())
    return false;

  switch (VT1.getSimpleVT().SimpleTy) {
  default: break;
  case MVT::i1:
  case MVT::i8:
  case MVT::i16:
    // 8-bit and 16-bit loads implicitly zero-extend to 32-bits.
    return true;
  }

  return false;
}

bool ARMTargetLowering::allowTruncateForTailCall(Type *Ty1, Type *Ty2) const {
  if (!Ty1->isIntegerTy() || !Ty2->isIntegerTy())
    return false;

  if (!isTypeLegal(EVT::getEVT(Ty1)))
    return false;

  assert(Ty1->getPrimitiveSizeInBits() <= 64 && "i128 is probably not a noop");

  // Assuming the caller doesn't have a zeroext or signext return parameter,
  // truncation all the way down to i1 is valid.
  return true;
}


static bool isLegalT1AddressImmediate(int64_t V, EVT VT) {
  if (V < 0)
    return false;

  unsigned Scale = 1;
  switch (VT.getSimpleVT().SimpleTy) {
  default: return false;
  case MVT::i1:
  case MVT::i8:
    // Scale == 1;
    break;
  case MVT::i16:
    // Scale == 2;
    Scale = 2;
    break;
  case MVT::i32:
    // Scale == 4;
    Scale = 4;
    break;
  }

  if ((V & (Scale - 1)) != 0)
    return false;
  V /= Scale;
  return V == (V & ((1LL << 5) - 1));
}

static bool isLegalT2AddressImmediate(int64_t V, EVT VT,
                                      const ARMSubtarget *Subtarget) {
  bool isNeg = false;
  if (V < 0) {
    isNeg = true;
    V = - V;
  }

  switch (VT.getSimpleVT().SimpleTy) {
  default: return false;
  case MVT::i1:
  case MVT::i8:
  case MVT::i16:
  case MVT::i32:
    // + imm12 or - imm8
    if (isNeg)
      return V == (V & ((1LL << 8) - 1));
    return V == (V & ((1LL << 12) - 1));
  case MVT::f32:
  case MVT::f64:
    // Same as ARM mode. FIXME: NEON?
    if (!Subtarget->hasVFP2())
      return false;
    if ((V & 3) != 0)
      return false;
    V >>= 2;
    return V == (V & ((1LL << 8) - 1));
  }
}

/// isLegalAddressImmediate - Return true if the integer value can be used
/// as the offset of the target addressing mode for load / store of the
/// given type.
static bool isLegalAddressImmediate(int64_t V, EVT VT,
                                    const ARMSubtarget *Subtarget) {
  if (V == 0)
    return true;

  if (!VT.isSimple())
    return false;

  if (Subtarget->isThumb1Only())
    return isLegalT1AddressImmediate(V, VT);
  else if (Subtarget->isThumb2())
    return isLegalT2AddressImmediate(V, VT, Subtarget);

  // ARM mode.
  if (V < 0)
    V = - V;
  switch (VT.getSimpleVT().SimpleTy) {
  default: return false;
  case MVT::i1:
  case MVT::i8:
  case MVT::i32:
    // +- imm12
    return V == (V & ((1LL << 12) - 1));
  case MVT::i16:
    // +- imm8
    return V == (V & ((1LL << 8) - 1));
  case MVT::f32:
  case MVT::f64:
    if (!Subtarget->hasVFP2()) // FIXME: NEON?
      return false;
    if ((V & 3) != 0)
      return false;
    V >>= 2;
    return V == (V & ((1LL << 8) - 1));
  }
}

bool ARMTargetLowering::isLegalT2ScaledAddressingMode(const AddrMode &AM,
                                                      EVT VT) const {
  int Scale = AM.Scale;
  if (Scale < 0)
    return false;

  switch (VT.getSimpleVT().SimpleTy) {
  default: return false;
  case MVT::i1:
  case MVT::i8:
  case MVT::i16:
  case MVT::i32:
    if (Scale == 1)
      return true;
    // r + r << imm
    Scale = Scale & ~1;
    return Scale == 2 || Scale == 4 || Scale == 8;
  case MVT::i64:
    // r + r
    if (((unsigned)AM.HasBaseReg + Scale) <= 2)
      return true;
    return false;
  case MVT::isVoid:
    // Note, we allow "void" uses (basically, uses that aren't loads or
    // stores), because arm allows folding a scale into many arithmetic
    // operations.  This should be made more precise and revisited later.

    // Allow r << imm, but the imm has to be a multiple of two.
    if (Scale & 1) return false;
    return isPowerOf2_32(Scale);
  }
}

/// isLegalAddressingMode - Return true if the addressing mode represented
/// by AM is legal for this target, for a load/store of the specified type.
bool ARMTargetLowering::isLegalAddressingMode(const AddrMode &AM,
                                              Type *Ty) const {
  EVT VT = getValueType(Ty, true);
  if (!isLegalAddressImmediate(AM.BaseOffs, VT, Subtarget))
    return false;

  // Can never fold addr of global into load/store.
  if (AM.BaseGV)
    return false;

  switch (AM.Scale) {
  case 0:  // no scale reg, must be "r+i" or "r", or "i".
    break;
  case 1:
    if (Subtarget->isThumb1Only())
      return false;
    // FALL THROUGH.
  default:
    // ARM doesn't support any R+R*scale+imm addr modes.
    if (AM.BaseOffs)
      return false;

    if (!VT.isSimple())
      return false;

    if (Subtarget->isThumb2())
      return isLegalT2ScaledAddressingMode(AM, VT);

    int Scale = AM.Scale;
    switch (VT.getSimpleVT().SimpleTy) {
    default: return false;
    case MVT::i1:
    case MVT::i8:
    case MVT::i32:
      if (Scale < 0) Scale = -Scale;
      if (Scale == 1)
        return true;
      // r + r << imm
      return isPowerOf2_32(Scale & ~1);
    case MVT::i16:
    case MVT::i64:
      // r + r
      if (((unsigned)AM.HasBaseReg + Scale) <= 2)
        return true;
      return false;

    case MVT::isVoid:
      // Note, we allow "void" uses (basically, uses that aren't loads or
      // stores), because arm allows folding a scale into many arithmetic
      // operations.  This should be made more precise and revisited later.

      // Allow r << imm, but the imm has to be a multiple of two.
      if (Scale & 1) return false;
      return isPowerOf2_32(Scale);
    }
  }
  return true;
}

/// isLegalICmpImmediate - Return true if the specified immediate is legal
/// icmp immediate, that is the target has icmp instructions which can compare
/// a register against the immediate without having to materialize the
/// immediate into a register.
bool ARMTargetLowering::isLegalICmpImmediate(int64_t Imm) const {
  // Thumb2 and ARM modes can use cmn for negative immediates.
  if (!Subtarget->isThumb())
    return ARM_AM::getSOImmVal(llvm::abs64(Imm)) != -1;
  if (Subtarget->isThumb2())
    return ARM_AM::getT2SOImmVal(llvm::abs64(Imm)) != -1;
  // Thumb1 doesn't have cmn, and only 8-bit immediates.
  return Imm >= 0 && Imm <= 255;
}

/// isLegalAddImmediate - Return true if the specified immediate is a legal add
/// *or sub* immediate, that is the target has add or sub instructions which can
/// add a register with the immediate without having to materialize the
/// immediate into a register.
bool ARMTargetLowering::isLegalAddImmediate(int64_t Imm) const {
  // Same encoding for add/sub, just flip the sign.
  int64_t AbsImm = llvm::abs64(Imm);
  if (!Subtarget->isThumb())
    return ARM_AM::getSOImmVal(AbsImm) != -1;
  if (Subtarget->isThumb2())
    return ARM_AM::getT2SOImmVal(AbsImm) != -1;
  // Thumb1 only has 8-bit unsigned immediate.
  return AbsImm >= 0 && AbsImm <= 255;
}

static bool getARMIndexedAddressParts(SDNode *Ptr, EVT VT,
                                      bool isSEXTLoad, SDValue &Base,
                                      SDValue &Offset, bool &isInc,
                                      SelectionDAG &DAG) {
  if (Ptr->getOpcode() != ISD::ADD && Ptr->getOpcode() != ISD::SUB)
    return false;

  if (VT == MVT::i16 || ((VT == MVT::i8 || VT == MVT::i1) && isSEXTLoad)) {
    // AddressingMode 3
    Base = Ptr->getOperand(0);
    if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Ptr->getOperand(1))) {
      int RHSC = (int)RHS->getZExtValue();
      if (RHSC < 0 && RHSC > -256) {
        assert(Ptr->getOpcode() == ISD::ADD);
        isInc = false;
        Offset = DAG.getConstant(-RHSC, RHS->getValueType(0));
        return true;
      }
    }
    isInc = (Ptr->getOpcode() == ISD::ADD);
    Offset = Ptr->getOperand(1);
    return true;
  } else if (VT == MVT::i32 || VT == MVT::i8 || VT == MVT::i1) {
    // AddressingMode 2
    if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Ptr->getOperand(1))) {
      int RHSC = (int)RHS->getZExtValue();
      if (RHSC < 0 && RHSC > -0x1000) {
        assert(Ptr->getOpcode() == ISD::ADD);
        isInc = false;
        Offset = DAG.getConstant(-RHSC, RHS->getValueType(0));
        Base = Ptr->getOperand(0);
        return true;
      }
    }

    if (Ptr->getOpcode() == ISD::ADD) {
      isInc = true;
      ARM_AM::ShiftOpc ShOpcVal=
        ARM_AM::getShiftOpcForNode(Ptr->getOperand(0).getOpcode());
      if (ShOpcVal != ARM_AM::no_shift) {
        Base = Ptr->getOperand(1);
        Offset = Ptr->getOperand(0);
      } else {
        Base = Ptr->getOperand(0);
        Offset = Ptr->getOperand(1);
      }
      return true;
    }

    isInc = (Ptr->getOpcode() == ISD::ADD);
    Base = Ptr->getOperand(0);
    Offset = Ptr->getOperand(1);
    return true;
  }

  // FIXME: Use VLDM / VSTM to emulate indexed FP load / store.
  return false;
}

static bool getT2IndexedAddressParts(SDNode *Ptr, EVT VT,
                                     bool isSEXTLoad, SDValue &Base,
                                     SDValue &Offset, bool &isInc,
                                     SelectionDAG &DAG) {
  if (Ptr->getOpcode() != ISD::ADD && Ptr->getOpcode() != ISD::SUB)
    return false;

  Base = Ptr->getOperand(0);
  if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Ptr->getOperand(1))) {
    int RHSC = (int)RHS->getZExtValue();
    if (RHSC < 0 && RHSC > -0x100) { // 8 bits.
      assert(Ptr->getOpcode() == ISD::ADD);
      isInc = false;
      Offset = DAG.getConstant(-RHSC, RHS->getValueType(0));
      return true;
    } else if (RHSC > 0 && RHSC < 0x100) { // 8 bit, no zero.
      isInc = Ptr->getOpcode() == ISD::ADD;
      Offset = DAG.getConstant(RHSC, RHS->getValueType(0));
      return true;
    }
  }

  return false;
}

/// getPreIndexedAddressParts - returns true by value, base pointer and
/// offset pointer and addressing mode by reference if the node's address
/// can be legally represented as pre-indexed load / store address.
bool
ARMTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
                                             SDValue &Offset,
                                             ISD::MemIndexedMode &AM,
                                             SelectionDAG &DAG) const {
  if (Subtarget->isThumb1Only())
    return false;

  EVT VT;
  SDValue Ptr;
  bool isSEXTLoad = false;
  if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
    Ptr = LD->getBasePtr();
    VT  = LD->getMemoryVT();
    isSEXTLoad = LD->getExtensionType() == ISD::SEXTLOAD;
  } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
    Ptr = ST->getBasePtr();
    VT  = ST->getMemoryVT();
  } else
    return false;

  bool isInc;
  bool isLegal = false;
  if (Subtarget->isThumb2())
    isLegal = getT2IndexedAddressParts(Ptr.getNode(), VT, isSEXTLoad, Base,
                                       Offset, isInc, DAG);
  else
    isLegal = getARMIndexedAddressParts(Ptr.getNode(), VT, isSEXTLoad, Base,
                                        Offset, isInc, DAG);
  if (!isLegal)
    return false;

  AM = isInc ? ISD::PRE_INC : ISD::PRE_DEC;
  return true;
}

/// getPostIndexedAddressParts - returns true by value, base pointer and
/// offset pointer and addressing mode by reference if this node can be
/// combined with a load / store to form a post-indexed load / store.
bool ARMTargetLowering::getPostIndexedAddressParts(SDNode *N, SDNode *Op,
                                                   SDValue &Base,
                                                   SDValue &Offset,
                                                   ISD::MemIndexedMode &AM,
                                                   SelectionDAG &DAG) const {
  if (Subtarget->isThumb1Only())
    return false;

  EVT VT;
  SDValue Ptr;
  bool isSEXTLoad = false;
  if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
    VT  = LD->getMemoryVT();
    Ptr = LD->getBasePtr();
    isSEXTLoad = LD->getExtensionType() == ISD::SEXTLOAD;
  } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
    VT  = ST->getMemoryVT();
    Ptr = ST->getBasePtr();
  } else
    return false;

  bool isInc;
  bool isLegal = false;
  if (Subtarget->isThumb2())
    isLegal = getT2IndexedAddressParts(Op, VT, isSEXTLoad, Base, Offset,
                                       isInc, DAG);
  else
    isLegal = getARMIndexedAddressParts(Op, VT, isSEXTLoad, Base, Offset,
                                        isInc, DAG);
  if (!isLegal)
    return false;

  if (Ptr != Base) {
    // Swap base ptr and offset to catch more post-index load / store when
    // it's legal. In Thumb2 mode, offset must be an immediate.
    if (Ptr == Offset && Op->getOpcode() == ISD::ADD &&
        !Subtarget->isThumb2())
      std::swap(Base, Offset);

    // Post-indexed load / store update the base pointer.
    if (Ptr != Base)
      return false;
  }

  AM = isInc ? ISD::POST_INC : ISD::POST_DEC;
  return true;
}

void ARMTargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
                                                      APInt &KnownZero,
                                                      APInt &KnownOne,
                                                      const SelectionDAG &DAG,
                                                      unsigned Depth) const {
  unsigned BitWidth = KnownOne.getBitWidth();
  KnownZero = KnownOne = APInt(BitWidth, 0);
  switch (Op.getOpcode()) {
  default: break;
  case ARMISD::ADDC:
  case ARMISD::ADDE:
  case ARMISD::SUBC:
  case ARMISD::SUBE:
    // These nodes' second result is a boolean
    if (Op.getResNo() == 0)
      break;
    KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - 1);
    break;
  case ARMISD::CMOV: {
    // Bits are known zero/one if known on the LHS and RHS.
    DAG.computeKnownBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
    if (KnownZero == 0 && KnownOne == 0) return;

    APInt KnownZeroRHS, KnownOneRHS;
    DAG.computeKnownBits(Op.getOperand(1), KnownZeroRHS, KnownOneRHS, Depth+1);
    KnownZero &= KnownZeroRHS;
    KnownOne  &= KnownOneRHS;
    return;
  }
  case ISD::INTRINSIC_W_CHAIN: {
    ConstantSDNode *CN = cast<ConstantSDNode>(Op->getOperand(1));
    Intrinsic::ID IntID = static_cast<Intrinsic::ID>(CN->getZExtValue());
    switch (IntID) {
    default: return;
    case Intrinsic::arm_ldaex:
    case Intrinsic::arm_ldrex: {
      EVT VT = cast<MemIntrinsicSDNode>(Op)->getMemoryVT();
      unsigned MemBits = VT.getScalarType().getSizeInBits();
      KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits);
      return;
    }
    }
  }
  }
}

//===----------------------------------------------------------------------===//
//                           ARM Inline Assembly Support
//===----------------------------------------------------------------------===//

bool ARMTargetLowering::ExpandInlineAsm(CallInst *CI) const {
  // Looking for "rev" which is V6+.
  if (!Subtarget->hasV6Ops())
    return false;

  InlineAsm *IA = cast<InlineAsm>(CI->getCalledValue());
  std::string AsmStr = IA->getAsmString();
  SmallVector<StringRef, 4> AsmPieces;
  SplitString(AsmStr, AsmPieces, ";\n");

  switch (AsmPieces.size()) {
  default: return false;
  case 1:
    AsmStr = AsmPieces[0];
    AsmPieces.clear();
    SplitString(AsmStr, AsmPieces, " \t,");

    // rev $0, $1
    if (AsmPieces.size() == 3 &&
        AsmPieces[0] == "rev" && AsmPieces[1] == "$0" && AsmPieces[2] == "$1" &&
        IA->getConstraintString().compare(0, 4, "=l,l") == 0) {
      IntegerType *Ty = dyn_cast<IntegerType>(CI->getType());
      if (Ty && Ty->getBitWidth() == 32)
        return IntrinsicLowering::LowerToByteSwap(CI);
    }
    break;
  }

  return false;
}

/// getConstraintType - Given a constraint letter, return the type of
/// constraint it is for this target.
ARMTargetLowering::ConstraintType
ARMTargetLowering::getConstraintType(const std::string &Constraint) const {
  if (Constraint.size() == 1) {
    switch (Constraint[0]) {
    default:  break;
    case 'l': return C_RegisterClass;
    case 'w': return C_RegisterClass;
    case 'h': return C_RegisterClass;
    case 'x': return C_RegisterClass;
    case 't': return C_RegisterClass;
    case 'j': return C_Other; // Constant for movw.
      // An address with a single base register. Due to the way we
      // currently handle addresses it is the same as an 'r' memory constraint.
    case 'Q': return C_Memory;
    }
  } else if (Constraint.size() == 2) {
    switch (Constraint[0]) {
    default: break;
    // All 'U+' constraints are addresses.
    case 'U': return C_Memory;
    }
  }
  return TargetLowering::getConstraintType(Constraint);
}

/// Examine constraint type and operand type and determine a weight value.
/// This object must already have been set up with the operand type
/// and the current alternative constraint selected.
TargetLowering::ConstraintWeight
ARMTargetLowering::getSingleConstraintMatchWeight(
    AsmOperandInfo &info, const char *constraint) const {
  ConstraintWeight weight = CW_Invalid;
  Value *CallOperandVal = info.CallOperandVal;
    // If we don't have a value, we can't do a match,
    // but allow it at the lowest weight.
  if (!CallOperandVal)
    return CW_Default;
  Type *type = CallOperandVal->getType();
  // Look at the constraint type.
  switch (*constraint) {
  default:
    weight = TargetLowering::getSingleConstraintMatchWeight(info, constraint);
    break;
  case 'l':
    if (type->isIntegerTy()) {
      if (Subtarget->isThumb())
        weight = CW_SpecificReg;
      else
        weight = CW_Register;
    }
    break;
  case 'w':
    if (type->isFloatingPointTy())
      weight = CW_Register;
    break;
  }
  return weight;
}

typedef std::pair<unsigned, const TargetRegisterClass*> RCPair;
RCPair
ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
                                                MVT VT) const {
  if (Constraint.size() == 1) {
    // GCC ARM Constraint Letters
    switch (Constraint[0]) {
    case 'l': // Low regs or general regs.
      if (Subtarget->isThumb())
        return RCPair(0U, &ARM::tGPRRegClass);
      return RCPair(0U, &ARM::GPRRegClass);
    case 'h': // High regs or no regs.
      if (Subtarget->isThumb())
        return RCPair(0U, &ARM::hGPRRegClass);
      break;
    case 'r':
      if (Subtarget->isThumb1Only())
        return RCPair(0U, &ARM::tGPRRegClass);
      return RCPair(0U, &ARM::GPRRegClass);
    case 'w':
      if (VT == MVT::Other)
        break;
      if (VT == MVT::f32)
        return RCPair(0U, &ARM::SPRRegClass);
      if (VT.getSizeInBits() == 64)
        return RCPair(0U, &ARM::DPRRegClass);
      if (VT.getSizeInBits() == 128)
        return RCPair(0U, &ARM::QPRRegClass);
      break;
    case 'x':
      if (VT == MVT::Other)
        break;
      if (VT == MVT::f32)
        return RCPair(0U, &ARM::SPR_8RegClass);
      if (VT.getSizeInBits() == 64)
        return RCPair(0U, &ARM::DPR_8RegClass);
      if (VT.getSizeInBits() == 128)
        return RCPair(0U, &ARM::QPR_8RegClass);
      break;
    case 't':
      if (VT == MVT::f32)
        return RCPair(0U, &ARM::SPRRegClass);
      break;
    }
  }
  if (StringRef("{cc}").equals_lower(Constraint))
    return std::make_pair(unsigned(ARM::CPSR), &ARM::CCRRegClass);

  return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
}

/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector.  If it is invalid, don't add anything to Ops.
void ARMTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
                                                     std::string &Constraint,
                                                     std::vector<SDValue>&Ops,
                                                     SelectionDAG &DAG) const {
  SDValue Result;

  // Currently only support length 1 constraints.
  if (Constraint.length() != 1) return;

  char ConstraintLetter = Constraint[0];
  switch (ConstraintLetter) {
  default: break;
  case 'j':
  case 'I': case 'J': case 'K': case 'L':
  case 'M': case 'N': case 'O':
    ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
    if (!C)
      return;

    int64_t CVal64 = C->getSExtValue();
    int CVal = (int) CVal64;
    // None of these constraints allow values larger than 32 bits.  Check
    // that the value fits in an int.
    if (CVal != CVal64)
      return;

    switch (ConstraintLetter) {
      case 'j':
        // Constant suitable for movw, must be between 0 and
        // 65535.
        if (Subtarget->hasV6T2Ops())
          if (CVal >= 0 && CVal <= 65535)
            break;
        return;
      case 'I':
        if (Subtarget->isThumb1Only()) {
          // This must be a constant between 0 and 255, for ADD
          // immediates.
          if (CVal >= 0 && CVal <= 255)
            break;
        } else if (Subtarget->isThumb2()) {
          // A constant that can be used as an immediate value in a
          // data-processing instruction.
          if (ARM_AM::getT2SOImmVal(CVal) != -1)
            break;
        } else {
          // A constant that can be used as an immediate value in a
          // data-processing instruction.
          if (ARM_AM::getSOImmVal(CVal) != -1)
            break;
        }
        return;

      case 'J':
        if (Subtarget->isThumb()) {  // FIXME thumb2
          // This must be a constant between -255 and -1, for negated ADD
          // immediates. This can be used in GCC with an "n" modifier that
          // prints the negated value, for use with SUB instructions. It is
          // not useful otherwise but is implemented for compatibility.
          if (CVal >= -255 && CVal <= -1)
            break;
        } else {
          // This must be a constant between -4095 and 4095. It is not clear
          // what this constraint is intended for. Implemented for
          // compatibility with GCC.
          if (CVal >= -4095 && CVal <= 4095)
            break;
        }
        return;

      case 'K':
        if (Subtarget->isThumb1Only()) {
          // A 32-bit value where only one byte has a nonzero value. Exclude
          // zero to match GCC. This constraint is used by GCC internally for
          // constants that can be loaded with a move/shift combination.
          // It is not useful otherwise but is implemented for compatibility.
          if (CVal != 0 && ARM_AM::isThumbImmShiftedVal(CVal))
            break;
        } else if (Subtarget->isThumb2()) {
          // A constant whose bitwise inverse can be used as an immediate
          // value in a data-processing instruction. This can be used in GCC
          // with a "B" modifier that prints the inverted value, for use with
          // BIC and MVN instructions. It is not useful otherwise but is
          // implemented for compatibility.
          if (ARM_AM::getT2SOImmVal(~CVal) != -1)
            break;
        } else {
          // A constant whose bitwise inverse can be used as an immediate
          // value in a data-processing instruction. This can be used in GCC
          // with a "B" modifier that prints the inverted value, for use with
          // BIC and MVN instructions. It is not useful otherwise but is
          // implemented for compatibility.
          if (ARM_AM::getSOImmVal(~CVal) != -1)
            break;
        }
        return;

      case 'L':
        if (Subtarget->isThumb1Only()) {
          // This must be a constant between -7 and 7,
          // for 3-operand ADD/SUB immediate instructions.
          if (CVal >= -7 && CVal < 7)
            break;
        } else if (Subtarget->isThumb2()) {
          // A constant whose negation can be used as an immediate value in a
          // data-processing instruction. This can be used in GCC with an "n"
          // modifier that prints the negated value, for use with SUB
          // instructions. It is not useful otherwise but is implemented for
          // compatibility.
          if (ARM_AM::getT2SOImmVal(-CVal) != -1)
            break;
        } else {
          // A constant whose negation can be used as an immediate value in a
          // data-processing instruction. This can be used in GCC with an "n"
          // modifier that prints the negated value, for use with SUB
          // instructions. It is not useful otherwise but is implemented for
          // compatibility.
          if (ARM_AM::getSOImmVal(-CVal) != -1)
            break;
        }
        return;

      case 'M':
        if (Subtarget->isThumb()) { // FIXME thumb2
          // This must be a multiple of 4 between 0 and 1020, for
          // ADD sp + immediate.
          if ((CVal >= 0 && CVal <= 1020) && ((CVal & 3) == 0))
            break;
        } else {
          // A power of two or a constant between 0 and 32.  This is used in
          // GCC for the shift amount on shifted register operands, but it is
          // useful in general for any shift amounts.
          if ((CVal >= 0 && CVal <= 32) || ((CVal & (CVal - 1)) == 0))
            break;
        }
        return;

      case 'N':
        if (Subtarget->isThumb()) {  // FIXME thumb2
          // This must be a constant between 0 and 31, for shift amounts.
          if (CVal >= 0 && CVal <= 31)
            break;
        }
        return;

      case 'O':
        if (Subtarget->isThumb()) {  // FIXME thumb2
          // This must be a multiple of 4 between -508 and 508, for
          // ADD/SUB sp = sp + immediate.
          if ((CVal >= -508 && CVal <= 508) && ((CVal & 3) == 0))
            break;
        }
        return;
    }
    Result = DAG.getTargetConstant(CVal, Op.getValueType());
    break;
  }

  if (Result.getNode()) {
    Ops.push_back(Result);
    return;
  }
  return TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
}

SDValue ARMTargetLowering::LowerDivRem(SDValue Op, SelectionDAG &DAG) const {
  assert(Subtarget->isTargetAEABI() && "Register-based DivRem lowering only");
  unsigned Opcode = Op->getOpcode();
  assert((Opcode == ISD::SDIVREM || Opcode == ISD::UDIVREM) &&
         "Invalid opcode for Div/Rem lowering");
  bool isSigned = (Opcode == ISD::SDIVREM);
  EVT VT = Op->getValueType(0);
  Type *Ty = VT.getTypeForEVT(*DAG.getContext());

  RTLIB::Libcall LC;
  switch (VT.getSimpleVT().SimpleTy) {
  default: llvm_unreachable("Unexpected request for libcall!");
  case MVT::i8:  LC = isSigned ? RTLIB::SDIVREM_I8  : RTLIB::UDIVREM_I8;  break;
  case MVT::i16: LC = isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
  case MVT::i32: LC = isSigned ? RTLIB::SDIVREM_I32 : RTLIB::UDIVREM_I32; break;
  case MVT::i64: LC = isSigned ? RTLIB::SDIVREM_I64 : RTLIB::UDIVREM_I64; break;
  }

  SDValue InChain = DAG.getEntryNode();

  TargetLowering::ArgListTy Args;
  TargetLowering::ArgListEntry Entry;
  for (unsigned i = 0, e = Op->getNumOperands(); i != e; ++i) {
    EVT ArgVT = Op->getOperand(i).getValueType();
    Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
    Entry.Node = Op->getOperand(i);
    Entry.Ty = ArgTy;
    Entry.isSExt = isSigned;
    Entry.isZExt = !isSigned;
    Args.push_back(Entry);
  }

  SDValue Callee = DAG.getExternalSymbol(getLibcallName(LC),
                                         getPointerTy());

  Type *RetTy = (Type*)StructType::get(Ty, Ty, nullptr);

  SDLoc dl(Op);
  TargetLowering::CallLoweringInfo CLI(DAG);
  CLI.setDebugLoc(dl).setChain(InChain)
    .setCallee(getLibcallCallingConv(LC), RetTy, Callee, std::move(Args), 0)
    .setInRegister().setSExtResult(isSigned).setZExtResult(!isSigned);

  std::pair<SDValue, SDValue> CallInfo = LowerCallTo(CLI);
  return CallInfo.first;
}

SDValue
ARMTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const {
  assert(Subtarget->isTargetWindows() && "unsupported target platform");
  SDLoc DL(Op);

  // Get the inputs.
  SDValue Chain = Op.getOperand(0);
  SDValue Size  = Op.getOperand(1);

  SDValue Words = DAG.getNode(ISD::SRL, DL, MVT::i32, Size,
                              DAG.getConstant(2, MVT::i32));

  SDValue Flag;
  Chain = DAG.getCopyToReg(Chain, DL, ARM::R4, Words, Flag);
  Flag = Chain.getValue(1);

  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
  Chain = DAG.getNode(ARMISD::WIN__CHKSTK, DL, NodeTys, Chain, Flag);

  SDValue NewSP = DAG.getCopyFromReg(Chain, DL, ARM::SP, MVT::i32);
  Chain = NewSP.getValue(1);

  SDValue Ops[2] = { NewSP, Chain };
  return DAG.getMergeValues(Ops, DL);
}

SDValue ARMTargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const {
  assert(Op.getValueType() == MVT::f64 && Subtarget->isFPOnlySP() &&
         "Unexpected type for custom-lowering FP_EXTEND");

  RTLIB::Libcall LC;
  LC = RTLIB::getFPEXT(Op.getOperand(0).getValueType(), Op.getValueType());

  SDValue SrcVal = Op.getOperand(0);
  return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1,
                     /*isSigned*/ false, SDLoc(Op)).first;
}

SDValue ARMTargetLowering::LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const {
  assert(Op.getOperand(0).getValueType() == MVT::f64 &&
         Subtarget->isFPOnlySP() &&
         "Unexpected type for custom-lowering FP_ROUND");

  RTLIB::Libcall LC;
  LC = RTLIB::getFPROUND(Op.getOperand(0).getValueType(), Op.getValueType());

  SDValue SrcVal = Op.getOperand(0);
  return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1,
                     /*isSigned*/ false, SDLoc(Op)).first;
}

bool
ARMTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
  // The ARM target isn't yet aware of offsets.
  return false;
}

bool ARM::isBitFieldInvertedMask(unsigned v) {
  if (v == 0xffffffff)
    return false;

  // there can be 1's on either or both "outsides", all the "inside"
  // bits must be 0's
  unsigned TO = CountTrailingOnes_32(v);
  unsigned LO = CountLeadingOnes_32(v);
  v = (v >> TO) << TO;
  v = (v << LO) >> LO;
  return v == 0;
}

/// isFPImmLegal - Returns true if the target can instruction select the
/// specified FP immediate natively. If false, the legalizer will
/// materialize the FP immediate as a load from a constant pool.
bool ARMTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
  if (!Subtarget->hasVFP3())
    return false;
  if (VT == MVT::f32)
    return ARM_AM::getFP32Imm(Imm) != -1;
  if (VT == MVT::f64 && !Subtarget->isFPOnlySP())
    return ARM_AM::getFP64Imm(Imm) != -1;
  return false;
}

/// getTgtMemIntrinsic - Represent NEON load and store intrinsics as
/// MemIntrinsicNodes.  The associated MachineMemOperands record the alignment
/// specified in the intrinsic calls.
bool ARMTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
                                           const CallInst &I,
                                           unsigned Intrinsic) const {
  switch (Intrinsic) {
  case Intrinsic::arm_neon_vld1:
  case Intrinsic::arm_neon_vld2:
  case Intrinsic::arm_neon_vld3:
  case Intrinsic::arm_neon_vld4:
  case Intrinsic::arm_neon_vld2lane:
  case Intrinsic::arm_neon_vld3lane:
  case Intrinsic::arm_neon_vld4lane: {
    Info.opc = ISD::INTRINSIC_W_CHAIN;
    // Conservatively set memVT to the entire set of vectors loaded.
    uint64_t NumElts = getDataLayout()->getTypeAllocSize(I.getType()) / 8;
    Info.memVT = EVT::getVectorVT(I.getType()->getContext(), MVT::i64, NumElts);
    Info.ptrVal = I.getArgOperand(0);
    Info.offset = 0;
    Value *AlignArg = I.getArgOperand(I.getNumArgOperands() - 1);
    Info.align = cast<ConstantInt>(AlignArg)->getZExtValue();
    Info.vol = false; // volatile loads with NEON intrinsics not supported
    Info.readMem = true;
    Info.writeMem = false;
    return true;
  }
  case Intrinsic::arm_neon_vst1:
  case Intrinsic::arm_neon_vst2:
  case Intrinsic::arm_neon_vst3:
  case Intrinsic::arm_neon_vst4:
  case Intrinsic::arm_neon_vst2lane:
  case Intrinsic::arm_neon_vst3lane:
  case Intrinsic::arm_neon_vst4lane: {
    Info.opc = ISD::INTRINSIC_VOID;
    // Conservatively set memVT to the entire set of vectors stored.
    unsigned NumElts = 0;
    for (unsigned ArgI = 1, ArgE = I.getNumArgOperands(); ArgI < ArgE; ++ArgI) {
      Type *ArgTy = I.getArgOperand(ArgI)->getType();
      if (!ArgTy->isVectorTy())
        break;
      NumElts += getDataLayout()->getTypeAllocSize(ArgTy) / 8;
    }
    Info.memVT = EVT::getVectorVT(I.getType()->getContext(), MVT::i64, NumElts);
    Info.ptrVal = I.getArgOperand(0);
    Info.offset = 0;
    Value *AlignArg = I.getArgOperand(I.getNumArgOperands() - 1);
    Info.align = cast<ConstantInt>(AlignArg)->getZExtValue();
    Info.vol = false; // volatile stores with NEON intrinsics not supported
    Info.readMem = false;
    Info.writeMem = true;
    return true;
  }
  case Intrinsic::arm_ldaex:
  case Intrinsic::arm_ldrex: {
    PointerType *PtrTy = cast<PointerType>(I.getArgOperand(0)->getType());
    Info.opc = ISD::INTRINSIC_W_CHAIN;
    Info.memVT = MVT::getVT(PtrTy->getElementType());
    Info.ptrVal = I.getArgOperand(0);
    Info.offset = 0;
    Info.align = getDataLayout()->getABITypeAlignment(PtrTy->getElementType());
    Info.vol = true;
    Info.readMem = true;
    Info.writeMem = false;
    return true;
  }
  case Intrinsic::arm_stlex:
  case Intrinsic::arm_strex: {
    PointerType *PtrTy = cast<PointerType>(I.getArgOperand(1)->getType());
    Info.opc = ISD::INTRINSIC_W_CHAIN;
    Info.memVT = MVT::getVT(PtrTy->getElementType());
    Info.ptrVal = I.getArgOperand(1);
    Info.offset = 0;
    Info.align = getDataLayout()->getABITypeAlignment(PtrTy->getElementType());
    Info.vol = true;
    Info.readMem = false;
    Info.writeMem = true;
    return true;
  }
  case Intrinsic::arm_stlexd:
  case Intrinsic::arm_strexd: {
    Info.opc = ISD::INTRINSIC_W_CHAIN;
    Info.memVT = MVT::i64;
    Info.ptrVal = I.getArgOperand(2);
    Info.offset = 0;
    Info.align = 8;
    Info.vol = true;
    Info.readMem = false;
    Info.writeMem = true;
    return true;
  }
  case Intrinsic::arm_ldaexd:
  case Intrinsic::arm_ldrexd: {
    Info.opc = ISD::INTRINSIC_W_CHAIN;
    Info.memVT = MVT::i64;
    Info.ptrVal = I.getArgOperand(0);
    Info.offset = 0;
    Info.align = 8;
    Info.vol = true;
    Info.readMem = true;
    Info.writeMem = false;
    return true;
  }
  default:
    break;
  }

  return false;
}

/// \brief Returns true if it is beneficial to convert a load of a constant
/// to just the constant itself.
bool ARMTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
                                                          Type *Ty) const {
  assert(Ty->isIntegerTy());

  unsigned Bits = Ty->getPrimitiveSizeInBits();
  if (Bits == 0 || Bits > 32)
    return false;
  return true;
}

bool ARMTargetLowering::hasLoadLinkedStoreConditional() const { return true; }

Instruction* ARMTargetLowering::makeDMB(IRBuilder<> &Builder,
                                        ARM_MB::MemBOpt Domain) const {
  Module *M = Builder.GetInsertBlock()->getParent()->getParent();

  // First, if the target has no DMB, see what fallback we can use.
  if (!Subtarget->hasDataBarrier()) {
    // Some ARMv6 cpus can support data barriers with an mcr instruction.
    // Thumb1 and pre-v6 ARM mode use a libcall instead and should never get
    // here.
    if (Subtarget->hasV6Ops() && !Subtarget->isThumb()) {
      Function *MCR = llvm::Intrinsic::getDeclaration(M, Intrinsic::arm_mcr);
      Value* args[6] = {Builder.getInt32(15), Builder.getInt32(0),
                        Builder.getInt32(0), Builder.getInt32(7),
                        Builder.getInt32(10), Builder.getInt32(5)};
      return Builder.CreateCall(MCR, args);
    } else {
      // Instead of using barriers, atomic accesses on these subtargets use
      // libcalls.
      llvm_unreachable("makeDMB on a target so old that it has no barriers");
    }
  } else {
    Function *DMB = llvm::Intrinsic::getDeclaration(M, Intrinsic::arm_dmb);
    // Only a full system barrier exists in the M-class architectures.
    Domain = Subtarget->isMClass() ? ARM_MB::SY : Domain;
    Constant *CDomain = Builder.getInt32(Domain);
    return Builder.CreateCall(DMB, CDomain);
  }
}

// Based on http://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
Instruction* ARMTargetLowering::emitLeadingFence(IRBuilder<> &Builder,
                                         AtomicOrdering Ord, bool IsStore,
                                         bool IsLoad) const {
  if (!getInsertFencesForAtomic())
    return nullptr;

  switch (Ord) {
  case NotAtomic:
  case Unordered:
    llvm_unreachable("Invalid fence: unordered/non-atomic");
  case Monotonic:
  case Acquire:
    return nullptr; // Nothing to do
  case SequentiallyConsistent:
    if (!IsStore)
      return nullptr; // Nothing to do
    /*FALLTHROUGH*/
  case Release:
  case AcquireRelease:
    if (Subtarget->isSwift())
      return makeDMB(Builder, ARM_MB::ISHST);
    // FIXME: add a comment with a link to documentation justifying this.
    else
      return makeDMB(Builder, ARM_MB::ISH);
  }
  llvm_unreachable("Unknown fence ordering in emitLeadingFence");
}

Instruction* ARMTargetLowering::emitTrailingFence(IRBuilder<> &Builder,
                                          AtomicOrdering Ord, bool IsStore,
                                          bool IsLoad) const {
  if (!getInsertFencesForAtomic())
    return nullptr;

  switch (Ord) {
  case NotAtomic:
  case Unordered:
    llvm_unreachable("Invalid fence: unordered/not-atomic");
  case Monotonic:
  case Release:
    return nullptr; // Nothing to do
  case Acquire:
  case AcquireRelease:
  case SequentiallyConsistent:
    return makeDMB(Builder, ARM_MB::ISH);
  }
  llvm_unreachable("Unknown fence ordering in emitTrailingFence");
}

// Loads and stores less than 64-bits are already atomic; ones above that
// are doomed anyway, so defer to the default libcall and blame the OS when
// things go wrong. Cortex M doesn't have ldrexd/strexd though, so don't emit
// anything for those.
bool ARMTargetLowering::shouldExpandAtomicStoreInIR(StoreInst *SI) const {
  unsigned Size = SI->getValueOperand()->getType()->getPrimitiveSizeInBits();
  return (Size == 64) && !Subtarget->isMClass();
}

// Loads and stores less than 64-bits are already atomic; ones above that
// are doomed anyway, so defer to the default libcall and blame the OS when
// things go wrong. Cortex M doesn't have ldrexd/strexd though, so don't emit
// anything for those.
// FIXME: ldrd and strd are atomic if the CPU has LPAE (e.g. A15 has that
// guarantee, see DDI0406C ARM architecture reference manual,
// sections A8.8.72-74 LDRD)
bool ARMTargetLowering::shouldExpandAtomicLoadInIR(LoadInst *LI) const {
  unsigned Size = LI->getType()->getPrimitiveSizeInBits();
  return (Size == 64) && !Subtarget->isMClass();
}

// For the real atomic operations, we have ldrex/strex up to 32 bits,
// and up to 64 bits on the non-M profiles
bool ARMTargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
  unsigned Size = AI->getType()->getPrimitiveSizeInBits();
  return Size <= (Subtarget->isMClass() ? 32U : 64U);
}

// This has so far only been implemented for MachO.
bool ARMTargetLowering::useLoadStackGuardNode() const {
  return Subtarget->getTargetTriple().getObjectFormat() == Triple::MachO;
}

bool ARMTargetLowering::canCombineStoreAndExtract(Type *VectorTy, Value *Idx,
                                                  unsigned &Cost) const {
  // If we do not have NEON, vector types are not natively supported.
  if (!Subtarget->hasNEON())
    return false;

  // Floating point values and vector values map to the same register file.
  // Therefore, althought we could do a store extract of a vector type, this is
  // better to leave at float as we have more freedom in the addressing mode for
  // those.
  if (VectorTy->isFPOrFPVectorTy())
    return false;

  // If the index is unknown at compile time, this is very expensive to lower
  // and it is not possible to combine the store with the extract.
  if (!isa<ConstantInt>(Idx))
    return false;

  assert(VectorTy->isVectorTy() && "VectorTy is not a vector type");
  unsigned BitWidth = cast<VectorType>(VectorTy)->getBitWidth();
  // We can do a store + vector extract on any vector that fits perfectly in a D
  // or Q register.
  if (BitWidth == 64 || BitWidth == 128) {
    Cost = 0;
    return true;
  }
  return false;
}

Value *ARMTargetLowering::emitLoadLinked(IRBuilder<> &Builder, Value *Addr,
                                         AtomicOrdering Ord) const {
  Module *M = Builder.GetInsertBlock()->getParent()->getParent();
  Type *ValTy = cast<PointerType>(Addr->getType())->getElementType();
  bool IsAcquire = isAtLeastAcquire(Ord);

  // Since i64 isn't legal and intrinsics don't get type-lowered, the ldrexd
  // intrinsic must return {i32, i32} and we have to recombine them into a
  // single i64 here.
  if (ValTy->getPrimitiveSizeInBits() == 64) {
    Intrinsic::ID Int =
        IsAcquire ? Intrinsic::arm_ldaexd : Intrinsic::arm_ldrexd;
    Function *Ldrex = llvm::Intrinsic::getDeclaration(M, Int);

    Addr = Builder.CreateBitCast(Addr, Type::getInt8PtrTy(M->getContext()));
    Value *LoHi = Builder.CreateCall(Ldrex, Addr, "lohi");

    Value *Lo = Builder.CreateExtractValue(LoHi, 0, "lo");
    Value *Hi = Builder.CreateExtractValue(LoHi, 1, "hi");
    if (!Subtarget->isLittle())
      std::swap (Lo, Hi);
    Lo = Builder.CreateZExt(Lo, ValTy, "lo64");
    Hi = Builder.CreateZExt(Hi, ValTy, "hi64");
    return Builder.CreateOr(
        Lo, Builder.CreateShl(Hi, ConstantInt::get(ValTy, 32)), "val64");
  }

  Type *Tys[] = { Addr->getType() };
  Intrinsic::ID Int = IsAcquire ? Intrinsic::arm_ldaex : Intrinsic::arm_ldrex;
  Function *Ldrex = llvm::Intrinsic::getDeclaration(M, Int, Tys);

  return Builder.CreateTruncOrBitCast(
      Builder.CreateCall(Ldrex, Addr),
      cast<PointerType>(Addr->getType())->getElementType());
}

Value *ARMTargetLowering::emitStoreConditional(IRBuilder<> &Builder, Value *Val,
                                               Value *Addr,
                                               AtomicOrdering Ord) const {
  Module *M = Builder.GetInsertBlock()->getParent()->getParent();
  bool IsRelease = isAtLeastRelease(Ord);

  // Since the intrinsics must have legal type, the i64 intrinsics take two
  // parameters: "i32, i32". We must marshal Val into the appropriate form
  // before the call.
  if (Val->getType()->getPrimitiveSizeInBits() == 64) {
    Intrinsic::ID Int =
        IsRelease ? Intrinsic::arm_stlexd : Intrinsic::arm_strexd;
    Function *Strex = Intrinsic::getDeclaration(M, Int);
    Type *Int32Ty = Type::getInt32Ty(M->getContext());

    Value *Lo = Builder.CreateTrunc(Val, Int32Ty, "lo");
    Value *Hi = Builder.CreateTrunc(Builder.CreateLShr(Val, 32), Int32Ty, "hi");
    if (!Subtarget->isLittle())
      std::swap (Lo, Hi);
    Addr = Builder.CreateBitCast(Addr, Type::getInt8PtrTy(M->getContext()));
    return Builder.CreateCall3(Strex, Lo, Hi, Addr);
  }

  Intrinsic::ID Int = IsRelease ? Intrinsic::arm_stlex : Intrinsic::arm_strex;
  Type *Tys[] = { Addr->getType() };
  Function *Strex = Intrinsic::getDeclaration(M, Int, Tys);

  return Builder.CreateCall2(
      Strex, Builder.CreateZExtOrBitCast(
                 Val, Strex->getFunctionType()->getParamType(0)),
      Addr);
}

enum HABaseType {
  HA_UNKNOWN = 0,
  HA_FLOAT,
  HA_DOUBLE,
  HA_VECT64,
  HA_VECT128
};

static bool isHomogeneousAggregate(Type *Ty, HABaseType &Base,
                                   uint64_t &Members) {
  if (const StructType *ST = dyn_cast<StructType>(Ty)) {
    for (unsigned i = 0; i < ST->getNumElements(); ++i) {
      uint64_t SubMembers = 0;
      if (!isHomogeneousAggregate(ST->getElementType(i), Base, SubMembers))
        return false;
      Members += SubMembers;
    }
  } else if (const ArrayType *AT = dyn_cast<ArrayType>(Ty)) {
    uint64_t SubMembers = 0;
    if (!isHomogeneousAggregate(AT->getElementType(), Base, SubMembers))
      return false;
    Members += SubMembers * AT->getNumElements();
  } else if (Ty->isFloatTy()) {
    if (Base != HA_UNKNOWN && Base != HA_FLOAT)
      return false;
    Members = 1;
    Base = HA_FLOAT;
  } else if (Ty->isDoubleTy()) {
    if (Base != HA_UNKNOWN && Base != HA_DOUBLE)
      return false;
    Members = 1;
    Base = HA_DOUBLE;
  } else if (const VectorType *VT = dyn_cast<VectorType>(Ty)) {
    Members = 1;
    switch (Base) {
    case HA_FLOAT:
    case HA_DOUBLE:
      return false;
    case HA_VECT64:
      return VT->getBitWidth() == 64;
    case HA_VECT128:
      return VT->getBitWidth() == 128;
    case HA_UNKNOWN:
      switch (VT->getBitWidth()) {
      case 64:
        Base = HA_VECT64;
        return true;
      case 128:
        Base = HA_VECT128;
        return true;
      default:
        return false;
      }
    }
  }

  return (Members > 0 && Members <= 4);
}

/// \brief Return true if a type is an AAPCS-VFP homogeneous aggregate.
bool ARMTargetLowering::functionArgumentNeedsConsecutiveRegisters(
    Type *Ty, CallingConv::ID CallConv, bool isVarArg) const {
  if (getEffectiveCallingConv(CallConv, isVarArg) !=
      CallingConv::ARM_AAPCS_VFP)
    return false;

  HABaseType Base = HA_UNKNOWN;
  uint64_t Members = 0;
  bool result = isHomogeneousAggregate(Ty, Base, Members);
  DEBUG(dbgs() << "isHA: " << result << " "; Ty->dump());
  return result;
}