aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target/AArch64/AArch64ISelLowering.cpp
blob: 7c94d831cbae419f340b248e04bde0eec230ecd3 (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
//===-- AArch64ISelLowering.cpp - AArch64 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 implements the AArch64TargetLowering class.
//
//===----------------------------------------------------------------------===//

#include "AArch64ISelLowering.h"
#include "AArch64MachineFunctionInfo.h"
#include "AArch64PerfectShuffle.h"
#include "AArch64Subtarget.h"
#include "AArch64TargetMachine.h"
#include "AArch64TargetObjectFile.h"
#include "MCTargetDesc/AArch64AddressingModes.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Type.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetOptions.h"
using namespace llvm;

#define DEBUG_TYPE "aarch64-lower"

STATISTIC(NumTailCalls, "Number of tail calls");
STATISTIC(NumShiftInserts, "Number of vector shift inserts");

namespace {
enum AlignMode {
  StrictAlign,
  NoStrictAlign
};
}

static cl::opt<AlignMode>
Align(cl::desc("Load/store alignment support"),
      cl::Hidden, cl::init(NoStrictAlign),
      cl::values(
          clEnumValN(StrictAlign,   "aarch64-strict-align",
                     "Disallow all unaligned memory accesses"),
          clEnumValN(NoStrictAlign, "aarch64-no-strict-align",
                     "Allow unaligned memory accesses"),
          clEnumValEnd));

// Place holder until extr generation is tested fully.
static cl::opt<bool>
EnableAArch64ExtrGeneration("aarch64-extr-generation", cl::Hidden,
                          cl::desc("Allow AArch64 (or (shift)(shift))->extract"),
                          cl::init(true));

static cl::opt<bool>
EnableAArch64SlrGeneration("aarch64-shift-insert-generation", cl::Hidden,
                         cl::desc("Allow AArch64 SLI/SRI formation"),
                         cl::init(false));


AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM)
    : TargetLowering(TM) {
  Subtarget = &TM.getSubtarget<AArch64Subtarget>();

  // AArch64 doesn't have comparisons which set GPRs or setcc instructions, so
  // we have to make something up. Arbitrarily, choose ZeroOrOne.
  setBooleanContents(ZeroOrOneBooleanContent);
  // When comparing vectors the result sets the different elements in the
  // vector to all-one or all-zero.
  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);

  // Set up the register classes.
  addRegisterClass(MVT::i32, &AArch64::GPR32allRegClass);
  addRegisterClass(MVT::i64, &AArch64::GPR64allRegClass);

  if (Subtarget->hasFPARMv8()) {
    addRegisterClass(MVT::f16, &AArch64::FPR16RegClass);
    addRegisterClass(MVT::f32, &AArch64::FPR32RegClass);
    addRegisterClass(MVT::f64, &AArch64::FPR64RegClass);
    addRegisterClass(MVT::f128, &AArch64::FPR128RegClass);
  }

  if (Subtarget->hasNEON()) {
    addRegisterClass(MVT::v16i8, &AArch64::FPR8RegClass);
    addRegisterClass(MVT::v8i16, &AArch64::FPR16RegClass);
    // Someone set us up the NEON.
    addDRTypeForNEON(MVT::v2f32);
    addDRTypeForNEON(MVT::v8i8);
    addDRTypeForNEON(MVT::v4i16);
    addDRTypeForNEON(MVT::v2i32);
    addDRTypeForNEON(MVT::v1i64);
    addDRTypeForNEON(MVT::v1f64);
    addDRTypeForNEON(MVT::v4f16);

    addQRTypeForNEON(MVT::v4f32);
    addQRTypeForNEON(MVT::v2f64);
    addQRTypeForNEON(MVT::v16i8);
    addQRTypeForNEON(MVT::v8i16);
    addQRTypeForNEON(MVT::v4i32);
    addQRTypeForNEON(MVT::v2i64);
    addQRTypeForNEON(MVT::v8f16);
  }

  // Compute derived properties from the register classes
  computeRegisterProperties();

  // Provide all sorts of operation actions
  setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
  setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
  setOperationAction(ISD::SETCC, MVT::i32, Custom);
  setOperationAction(ISD::SETCC, MVT::i64, Custom);
  setOperationAction(ISD::SETCC, MVT::f32, Custom);
  setOperationAction(ISD::SETCC, MVT::f64, Custom);
  setOperationAction(ISD::BRCOND, MVT::Other, Expand);
  setOperationAction(ISD::BR_CC, MVT::i32, Custom);
  setOperationAction(ISD::BR_CC, MVT::i64, Custom);
  setOperationAction(ISD::BR_CC, MVT::f32, Custom);
  setOperationAction(ISD::BR_CC, MVT::f64, Custom);
  setOperationAction(ISD::SELECT, MVT::i32, Custom);
  setOperationAction(ISD::SELECT, MVT::i64, 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::i64, Custom);
  setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
  setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
  setOperationAction(ISD::BR_JT, MVT::Other, Expand);
  setOperationAction(ISD::JumpTable, MVT::i64, Custom);

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

  setOperationAction(ISD::FREM, MVT::f32, Expand);
  setOperationAction(ISD::FREM, MVT::f64, Expand);
  setOperationAction(ISD::FREM, MVT::f80, Expand);

  // Custom lowering hooks are needed for XOR
  // to fold it into CSINC/CSINV.
  setOperationAction(ISD::XOR, MVT::i32, Custom);
  setOperationAction(ISD::XOR, MVT::i64, Custom);

  // Virtually no operation on f128 is legal, but LLVM can't expand them when
  // there's a valid register class, so we need custom operations in most cases.
  setOperationAction(ISD::FABS, MVT::f128, Expand);
  setOperationAction(ISD::FADD, MVT::f128, Custom);
  setOperationAction(ISD::FCOPYSIGN, MVT::f128, Expand);
  setOperationAction(ISD::FCOS, MVT::f128, Expand);
  setOperationAction(ISD::FDIV, MVT::f128, Custom);
  setOperationAction(ISD::FMA, MVT::f128, Expand);
  setOperationAction(ISD::FMUL, MVT::f128, Custom);
  setOperationAction(ISD::FNEG, MVT::f128, Expand);
  setOperationAction(ISD::FPOW, MVT::f128, Expand);
  setOperationAction(ISD::FREM, MVT::f128, Expand);
  setOperationAction(ISD::FRINT, MVT::f128, Expand);
  setOperationAction(ISD::FSIN, MVT::f128, Expand);
  setOperationAction(ISD::FSINCOS, MVT::f128, Expand);
  setOperationAction(ISD::FSQRT, MVT::f128, Expand);
  setOperationAction(ISD::FSUB, MVT::f128, Custom);
  setOperationAction(ISD::FTRUNC, MVT::f128, Expand);
  setOperationAction(ISD::SETCC, MVT::f128, Custom);
  setOperationAction(ISD::BR_CC, MVT::f128, Custom);
  setOperationAction(ISD::SELECT, MVT::f128, Custom);
  setOperationAction(ISD::SELECT_CC, MVT::f128, Custom);
  setOperationAction(ISD::FP_EXTEND, MVT::f128, Custom);

  // Lowering for many of the conversions is actually specified by the non-f128
  // type. The LowerXXX function will be trivial when f128 isn't involved.
  setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
  setOperationAction(ISD::FP_TO_SINT, MVT::i128, Custom);
  setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
  setOperationAction(ISD::FP_TO_UINT, MVT::i64, Custom);
  setOperationAction(ISD::FP_TO_UINT, MVT::i128, Custom);
  setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
  setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
  setOperationAction(ISD::SINT_TO_FP, MVT::i128, Custom);
  setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
  setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
  setOperationAction(ISD::UINT_TO_FP, MVT::i128, Custom);
  setOperationAction(ISD::FP_ROUND, MVT::f32, Custom);
  setOperationAction(ISD::FP_ROUND, MVT::f64, Custom);

  // Variable arguments.
  setOperationAction(ISD::VASTART, MVT::Other, Custom);
  setOperationAction(ISD::VAARG, MVT::Other, Custom);
  setOperationAction(ISD::VACOPY, MVT::Other, Custom);
  setOperationAction(ISD::VAEND, MVT::Other, Expand);

  // Variable-sized objects.
  setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
  setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);

  // Exception handling.
  // FIXME: These are guesses. Has this been defined yet?
  setExceptionPointerRegister(AArch64::X0);
  setExceptionSelectorRegister(AArch64::X1);

  // Constant pool entries
  setOperationAction(ISD::ConstantPool, MVT::i64, Custom);

  // BlockAddress
  setOperationAction(ISD::BlockAddress, MVT::i64, Custom);

  // Add/Sub overflow ops with MVT::Glues are lowered to NZCV dependences.
  setOperationAction(ISD::ADDC, MVT::i32, Custom);
  setOperationAction(ISD::ADDE, MVT::i32, Custom);
  setOperationAction(ISD::SUBC, MVT::i32, Custom);
  setOperationAction(ISD::SUBE, MVT::i32, Custom);
  setOperationAction(ISD::ADDC, MVT::i64, Custom);
  setOperationAction(ISD::ADDE, MVT::i64, Custom);
  setOperationAction(ISD::SUBC, MVT::i64, Custom);
  setOperationAction(ISD::SUBE, MVT::i64, Custom);

  // AArch64 lacks both left-rotate and popcount instructions.
  setOperationAction(ISD::ROTL, MVT::i32, Expand);
  setOperationAction(ISD::ROTL, MVT::i64, Expand);

  // AArch64 doesn't have {U|S}MUL_LOHI.
  setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
  setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);


  // Expand the undefined-at-zero variants to cttz/ctlz to their defined-at-zero
  // counterparts, which AArch64 supports directly.
  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32, Expand);
  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Expand);
  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64, Expand);
  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i64, Expand);

  setOperationAction(ISD::CTPOP, MVT::i32, Custom);
  setOperationAction(ISD::CTPOP, MVT::i64, Custom);

  setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
  setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
  setOperationAction(ISD::SREM, MVT::i32, Expand);
  setOperationAction(ISD::SREM, MVT::i64, Expand);
  setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
  setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
  setOperationAction(ISD::UREM, MVT::i32, Expand);
  setOperationAction(ISD::UREM, MVT::i64, Expand);

  // Custom lower Add/Sub/Mul with overflow.
  setOperationAction(ISD::SADDO, MVT::i32, Custom);
  setOperationAction(ISD::SADDO, MVT::i64, Custom);
  setOperationAction(ISD::UADDO, MVT::i32, Custom);
  setOperationAction(ISD::UADDO, MVT::i64, Custom);
  setOperationAction(ISD::SSUBO, MVT::i32, Custom);
  setOperationAction(ISD::SSUBO, MVT::i64, Custom);
  setOperationAction(ISD::USUBO, MVT::i32, Custom);
  setOperationAction(ISD::USUBO, MVT::i64, Custom);
  setOperationAction(ISD::SMULO, MVT::i32, Custom);
  setOperationAction(ISD::SMULO, MVT::i64, Custom);
  setOperationAction(ISD::UMULO, MVT::i32, Custom);
  setOperationAction(ISD::UMULO, MVT::i64, Custom);

  setOperationAction(ISD::FSIN, MVT::f32, Expand);
  setOperationAction(ISD::FSIN, MVT::f64, Expand);
  setOperationAction(ISD::FCOS, MVT::f32, Expand);
  setOperationAction(ISD::FCOS, MVT::f64, Expand);
  setOperationAction(ISD::FPOW, MVT::f32, Expand);
  setOperationAction(ISD::FPOW, MVT::f64, Expand);
  setOperationAction(ISD::FCOPYSIGN, MVT::f64, Custom);
  setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom);

  // f16 is storage-only, so we promote operations to f32 if we know this is
  // valid, and ignore them otherwise. The operations not mentioned here will
  // fail to select, but this is not a major problem as no source language
  // should be emitting native f16 operations yet.
  setOperationAction(ISD::FADD, MVT::f16, Promote);
  setOperationAction(ISD::FDIV, MVT::f16, Promote);
  setOperationAction(ISD::FMUL, MVT::f16, Promote);
  setOperationAction(ISD::FSUB, MVT::f16, Promote);

  // v4f16 is also a storage-only type, so promote it to v4f32 when that is
  // known to be safe.
  setOperationAction(ISD::FADD, MVT::v4f16, Promote);
  setOperationAction(ISD::FSUB, MVT::v4f16, Promote);
  setOperationAction(ISD::FMUL, MVT::v4f16, Promote);
  setOperationAction(ISD::FDIV, MVT::v4f16, Promote);
  setOperationAction(ISD::FP_EXTEND, MVT::v4f16, Promote);
  setOperationAction(ISD::FP_ROUND, MVT::v4f16, Promote);
  AddPromotedToType(ISD::FADD, MVT::v4f16, MVT::v4f32);
  AddPromotedToType(ISD::FSUB, MVT::v4f16, MVT::v4f32);
  AddPromotedToType(ISD::FMUL, MVT::v4f16, MVT::v4f32);
  AddPromotedToType(ISD::FDIV, MVT::v4f16, MVT::v4f32);
  AddPromotedToType(ISD::FP_EXTEND, MVT::v4f16, MVT::v4f32);
  AddPromotedToType(ISD::FP_ROUND, MVT::v4f16, MVT::v4f32);

  // Expand all other v4f16 operations.
  // FIXME: We could generate better code by promoting some operations to
  // a pair of v4f32s
  setOperationAction(ISD::FABS, MVT::v4f16, Expand);
  setOperationAction(ISD::FCEIL, MVT::v4f16, Expand);
  setOperationAction(ISD::FCOPYSIGN, MVT::v4f16, Expand);
  setOperationAction(ISD::FCOS, MVT::v4f16, Expand);
  setOperationAction(ISD::FFLOOR, MVT::v4f16, Expand);
  setOperationAction(ISD::FMA, MVT::v4f16, Expand);
  setOperationAction(ISD::FNEARBYINT, MVT::v4f16, Expand);
  setOperationAction(ISD::FNEG, MVT::v4f16, Expand);
  setOperationAction(ISD::FPOW, MVT::v4f16, Expand);
  setOperationAction(ISD::FPOWI, MVT::v4f16, Expand);
  setOperationAction(ISD::FREM, MVT::v4f16, Expand);
  setOperationAction(ISD::FROUND, MVT::v4f16, Expand);
  setOperationAction(ISD::FRINT, MVT::v4f16, Expand);
  setOperationAction(ISD::FSIN, MVT::v4f16, Expand);
  setOperationAction(ISD::FSINCOS, MVT::v4f16, Expand);
  setOperationAction(ISD::FSQRT, MVT::v4f16, Expand);
  setOperationAction(ISD::FTRUNC, MVT::v4f16, Expand);
  setOperationAction(ISD::SETCC, MVT::v4f16, Expand);
  setOperationAction(ISD::BR_CC, MVT::v4f16, Expand);
  setOperationAction(ISD::SELECT, MVT::v4f16, Expand);
  setOperationAction(ISD::SELECT_CC, MVT::v4f16, Expand);
  setOperationAction(ISD::FEXP, MVT::v4f16, Expand);
  setOperationAction(ISD::FEXP2, MVT::v4f16, Expand);
  setOperationAction(ISD::FLOG, MVT::v4f16, Expand);
  setOperationAction(ISD::FLOG2, MVT::v4f16, Expand);
  setOperationAction(ISD::FLOG10, MVT::v4f16, Expand);


  // v8f16 is also a storage-only type, so expand it.
  setOperationAction(ISD::FABS, MVT::v8f16, Expand);
  setOperationAction(ISD::FADD, MVT::v8f16, Expand);
  setOperationAction(ISD::FCEIL, MVT::v8f16, Expand);
  setOperationAction(ISD::FCOPYSIGN, MVT::v8f16, Expand);
  setOperationAction(ISD::FCOS, MVT::v8f16, Expand);
  setOperationAction(ISD::FDIV, MVT::v8f16, Expand);
  setOperationAction(ISD::FFLOOR, MVT::v8f16, Expand);
  setOperationAction(ISD::FMA, MVT::v8f16, Expand);
  setOperationAction(ISD::FMUL, MVT::v8f16, Expand);
  setOperationAction(ISD::FNEARBYINT, MVT::v8f16, Expand);
  setOperationAction(ISD::FNEG, MVT::v8f16, Expand);
  setOperationAction(ISD::FPOW, MVT::v8f16, Expand);
  setOperationAction(ISD::FPOWI, MVT::v8f16, Expand);
  setOperationAction(ISD::FREM, MVT::v8f16, Expand);
  setOperationAction(ISD::FROUND, MVT::v8f16, Expand);
  setOperationAction(ISD::FRINT, MVT::v8f16, Expand);
  setOperationAction(ISD::FSIN, MVT::v8f16, Expand);
  setOperationAction(ISD::FSINCOS, MVT::v8f16, Expand);
  setOperationAction(ISD::FSQRT, MVT::v8f16, Expand);
  setOperationAction(ISD::FSUB, MVT::v8f16, Expand);
  setOperationAction(ISD::FTRUNC, MVT::v8f16, Expand);
  setOperationAction(ISD::SETCC, MVT::v8f16, Expand);
  setOperationAction(ISD::BR_CC, MVT::v8f16, Expand);
  setOperationAction(ISD::SELECT, MVT::v8f16, Expand);
  setOperationAction(ISD::SELECT_CC, MVT::v8f16, Expand);
  setOperationAction(ISD::FP_EXTEND, MVT::v8f16, Expand);
  setOperationAction(ISD::FEXP, MVT::v8f16, Expand);
  setOperationAction(ISD::FEXP2, MVT::v8f16, Expand);
  setOperationAction(ISD::FLOG, MVT::v8f16, Expand);
  setOperationAction(ISD::FLOG2, MVT::v8f16, Expand);
  setOperationAction(ISD::FLOG10, MVT::v8f16, Expand);

  // AArch64 has implementations of a lot of rounding-like FP operations.
  static MVT RoundingTypes[] = { MVT::f32, MVT::f64};
  for (unsigned I = 0; I < array_lengthof(RoundingTypes); ++I) {
    MVT Ty = RoundingTypes[I];
    setOperationAction(ISD::FFLOOR, Ty, Legal);
    setOperationAction(ISD::FNEARBYINT, Ty, Legal);
    setOperationAction(ISD::FCEIL, Ty, Legal);
    setOperationAction(ISD::FRINT, Ty, Legal);
    setOperationAction(ISD::FTRUNC, Ty, Legal);
    setOperationAction(ISD::FROUND, Ty, Legal);
  }

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

  if (Subtarget->isTargetMachO()) {
    // For iOS, we don't want to the normal expansion of a libcall to
    // sincos. We want to issue a libcall to __sincos_stret to avoid memory
    // traffic.
    setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
    setOperationAction(ISD::FSINCOS, MVT::f32, Custom);
  } else {
    setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
  }

  // AArch64 does not have floating-point extending loads, i1 sign-extending
  // load, floating-point truncating stores, or v2i32->v2i16 truncating store.
  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
  setLoadExtAction(ISD::EXTLOAD, MVT::f64, Expand);
  setLoadExtAction(ISD::EXTLOAD, MVT::f80, Expand);
  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Expand);
  setTruncStoreAction(MVT::f32, MVT::f16, Expand);
  setTruncStoreAction(MVT::f64, MVT::f32, Expand);
  setTruncStoreAction(MVT::f64, MVT::f16, Expand);
  setTruncStoreAction(MVT::f128, MVT::f80, Expand);
  setTruncStoreAction(MVT::f128, MVT::f64, Expand);
  setTruncStoreAction(MVT::f128, MVT::f32, Expand);
  setTruncStoreAction(MVT::f128, MVT::f16, Expand);

  setOperationAction(ISD::BITCAST, MVT::i16, Custom);
  setOperationAction(ISD::BITCAST, MVT::f16, Custom);

  // Indexed loads and stores are supported.
  for (unsigned im = (unsigned)ISD::PRE_INC;
       im != (unsigned)ISD::LAST_INDEXED_MODE; ++im) {
    setIndexedLoadAction(im, MVT::i8, Legal);
    setIndexedLoadAction(im, MVT::i16, Legal);
    setIndexedLoadAction(im, MVT::i32, Legal);
    setIndexedLoadAction(im, MVT::i64, Legal);
    setIndexedLoadAction(im, MVT::f64, Legal);
    setIndexedLoadAction(im, MVT::f32, Legal);
    setIndexedStoreAction(im, MVT::i8, Legal);
    setIndexedStoreAction(im, MVT::i16, Legal);
    setIndexedStoreAction(im, MVT::i32, Legal);
    setIndexedStoreAction(im, MVT::i64, Legal);
    setIndexedStoreAction(im, MVT::f64, Legal);
    setIndexedStoreAction(im, MVT::f32, Legal);
  }

  // Trap.
  setOperationAction(ISD::TRAP, MVT::Other, Legal);

  // We combine OR nodes for bitfield operations.
  setTargetDAGCombine(ISD::OR);

  // Vector add and sub nodes may conceal a high-half opportunity.
  // Also, try to fold ADD into CSINC/CSINV..
  setTargetDAGCombine(ISD::ADD);
  setTargetDAGCombine(ISD::SUB);

  setTargetDAGCombine(ISD::XOR);
  setTargetDAGCombine(ISD::SINT_TO_FP);
  setTargetDAGCombine(ISD::UINT_TO_FP);

  setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);

  setTargetDAGCombine(ISD::ANY_EXTEND);
  setTargetDAGCombine(ISD::ZERO_EXTEND);
  setTargetDAGCombine(ISD::SIGN_EXTEND);
  setTargetDAGCombine(ISD::BITCAST);
  setTargetDAGCombine(ISD::CONCAT_VECTORS);
  setTargetDAGCombine(ISD::STORE);

  setTargetDAGCombine(ISD::MUL);

  setTargetDAGCombine(ISD::SELECT);
  setTargetDAGCombine(ISD::VSELECT);

  setTargetDAGCombine(ISD::INTRINSIC_VOID);
  setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
  setTargetDAGCombine(ISD::INSERT_VECTOR_ELT);

  MaxStoresPerMemset = MaxStoresPerMemsetOptSize = 8;
  MaxStoresPerMemcpy = MaxStoresPerMemcpyOptSize = 4;
  MaxStoresPerMemmove = MaxStoresPerMemmoveOptSize = 4;

  setStackPointerRegisterToSaveRestore(AArch64::SP);

  setSchedulingPreference(Sched::Hybrid);

  // Enable TBZ/TBNZ
  MaskAndBranchFoldingIsLegal = true;

  setMinFunctionAlignment(2);

  RequireStrictAlign = (Align == StrictAlign);

  setHasExtractBitsInsn(true);

  if (Subtarget->hasNEON()) {
    // FIXME: v1f64 shouldn't be legal if we can avoid it, because it leads to
    // silliness like this:
    setOperationAction(ISD::FABS, MVT::v1f64, Expand);
    setOperationAction(ISD::FADD, MVT::v1f64, Expand);
    setOperationAction(ISD::FCEIL, MVT::v1f64, Expand);
    setOperationAction(ISD::FCOPYSIGN, MVT::v1f64, Expand);
    setOperationAction(ISD::FCOS, MVT::v1f64, Expand);
    setOperationAction(ISD::FDIV, MVT::v1f64, Expand);
    setOperationAction(ISD::FFLOOR, MVT::v1f64, Expand);
    setOperationAction(ISD::FMA, MVT::v1f64, Expand);
    setOperationAction(ISD::FMUL, MVT::v1f64, Expand);
    setOperationAction(ISD::FNEARBYINT, MVT::v1f64, Expand);
    setOperationAction(ISD::FNEG, MVT::v1f64, Expand);
    setOperationAction(ISD::FPOW, MVT::v1f64, Expand);
    setOperationAction(ISD::FREM, MVT::v1f64, Expand);
    setOperationAction(ISD::FROUND, MVT::v1f64, Expand);
    setOperationAction(ISD::FRINT, MVT::v1f64, Expand);
    setOperationAction(ISD::FSIN, MVT::v1f64, Expand);
    setOperationAction(ISD::FSINCOS, MVT::v1f64, Expand);
    setOperationAction(ISD::FSQRT, MVT::v1f64, Expand);
    setOperationAction(ISD::FSUB, MVT::v1f64, Expand);
    setOperationAction(ISD::FTRUNC, MVT::v1f64, Expand);
    setOperationAction(ISD::SETCC, MVT::v1f64, Expand);
    setOperationAction(ISD::BR_CC, MVT::v1f64, Expand);
    setOperationAction(ISD::SELECT, MVT::v1f64, Expand);
    setOperationAction(ISD::SELECT_CC, MVT::v1f64, Expand);
    setOperationAction(ISD::FP_EXTEND, MVT::v1f64, Expand);

    setOperationAction(ISD::FP_TO_SINT, MVT::v1i64, Expand);
    setOperationAction(ISD::FP_TO_UINT, MVT::v1i64, Expand);
    setOperationAction(ISD::SINT_TO_FP, MVT::v1i64, Expand);
    setOperationAction(ISD::UINT_TO_FP, MVT::v1i64, Expand);
    setOperationAction(ISD::FP_ROUND, MVT::v1f64, Expand);

    setOperationAction(ISD::MUL, MVT::v1i64, Expand);

    // AArch64 doesn't have a direct vector ->f32 conversion instructions for
    // elements smaller than i32, so promote the input to i32 first.
    setOperationAction(ISD::UINT_TO_FP, MVT::v4i8, Promote);
    setOperationAction(ISD::SINT_TO_FP, MVT::v4i8, Promote);
    setOperationAction(ISD::UINT_TO_FP, MVT::v4i16, Promote);
    setOperationAction(ISD::SINT_TO_FP, MVT::v4i16, Promote);
    // Similarly, there is no direct i32 -> f64 vector conversion instruction.
    setOperationAction(ISD::SINT_TO_FP, MVT::v2i32, Custom);
    setOperationAction(ISD::UINT_TO_FP, MVT::v2i32, Custom);
    setOperationAction(ISD::SINT_TO_FP, MVT::v2i64, Custom);
    setOperationAction(ISD::UINT_TO_FP, MVT::v2i64, Custom);

    // AArch64 doesn't have MUL.2d:
    setOperationAction(ISD::MUL, MVT::v2i64, Expand);
    // Custom handling for some quad-vector types to detect MULL.
    setOperationAction(ISD::MUL, MVT::v8i16, Custom);
    setOperationAction(ISD::MUL, MVT::v4i32, Custom);
    setOperationAction(ISD::MUL, MVT::v2i64, Custom);

    setOperationAction(ISD::ANY_EXTEND, MVT::v4i32, Legal);
    setTruncStoreAction(MVT::v2i32, MVT::v2i16, Expand);
    // Likewise, narrowing and extending vector loads/stores aren't handled
    // directly.
    for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
         VT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++VT) {

      setOperationAction(ISD::SIGN_EXTEND_INREG, (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);

      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);
    }

    // AArch64 has implementations of a lot of rounding-like FP operations.
    static MVT RoundingVecTypes[] = {MVT::v2f32, MVT::v4f32, MVT::v2f64 };
    for (unsigned I = 0; I < array_lengthof(RoundingVecTypes); ++I) {
      MVT Ty = RoundingVecTypes[I];
      setOperationAction(ISD::FFLOOR, Ty, Legal);
      setOperationAction(ISD::FNEARBYINT, Ty, Legal);
      setOperationAction(ISD::FCEIL, Ty, Legal);
      setOperationAction(ISD::FRINT, Ty, Legal);
      setOperationAction(ISD::FTRUNC, Ty, Legal);
      setOperationAction(ISD::FROUND, Ty, Legal);
    }
  }

  // Prefer likely predicted branches to selects on out-of-order cores.
  if (Subtarget->isCortexA57())
    PredictableSelectIsExpensive = true;
}

void AArch64TargetLowering::addTypeForNEON(EVT VT, EVT PromotedBitwiseVT) {
  if (VT == MVT::v2f32 || VT == MVT::v4f16) {
    setOperationAction(ISD::LOAD, VT.getSimpleVT(), Promote);
    AddPromotedToType(ISD::LOAD, VT.getSimpleVT(), MVT::v2i32);

    setOperationAction(ISD::STORE, VT.getSimpleVT(), Promote);
    AddPromotedToType(ISD::STORE, VT.getSimpleVT(), MVT::v2i32);
  } else if (VT == MVT::v2f64 || VT == MVT::v4f32 || VT == MVT::v8f16) {
    setOperationAction(ISD::LOAD, VT.getSimpleVT(), Promote);
    AddPromotedToType(ISD::LOAD, VT.getSimpleVT(), MVT::v2i64);

    setOperationAction(ISD::STORE, VT.getSimpleVT(), Promote);
    AddPromotedToType(ISD::STORE, VT.getSimpleVT(), MVT::v2i64);
  }

  // Mark vector float intrinsics as expand.
  if (VT == MVT::v2f32 || VT == MVT::v4f32 || VT == MVT::v2f64) {
    setOperationAction(ISD::FSIN, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FCOS, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FPOWI, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FPOW, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FLOG, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FLOG2, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FLOG10, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FEXP, VT.getSimpleVT(), Expand);
    setOperationAction(ISD::FEXP2, VT.getSimpleVT(), Expand);
  }

  setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::INSERT_VECTOR_ELT, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::BUILD_VECTOR, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::VECTOR_SHUFFLE, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::EXTRACT_SUBVECTOR, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::SRA, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::SRL, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::SHL, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::AND, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::OR, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::SETCC, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::CONCAT_VECTORS, VT.getSimpleVT(), Legal);

  setOperationAction(ISD::SELECT, VT.getSimpleVT(), Expand);
  setOperationAction(ISD::SELECT_CC, VT.getSimpleVT(), Expand);
  setOperationAction(ISD::VSELECT, VT.getSimpleVT(), Expand);
  setLoadExtAction(ISD::EXTLOAD, VT.getSimpleVT(), Expand);

  // CNT supports only B element sizes.
  if (VT != MVT::v8i8 && VT != MVT::v16i8)
    setOperationAction(ISD::CTPOP, VT.getSimpleVT(), Expand);

  setOperationAction(ISD::UDIV, VT.getSimpleVT(), Expand);
  setOperationAction(ISD::SDIV, VT.getSimpleVT(), Expand);
  setOperationAction(ISD::UREM, VT.getSimpleVT(), Expand);
  setOperationAction(ISD::SREM, VT.getSimpleVT(), Expand);
  setOperationAction(ISD::FREM, VT.getSimpleVT(), Expand);

  setOperationAction(ISD::FP_TO_SINT, VT.getSimpleVT(), Custom);
  setOperationAction(ISD::FP_TO_UINT, VT.getSimpleVT(), Custom);

  if (Subtarget->isLittleEndian()) {
    for (unsigned im = (unsigned)ISD::PRE_INC;
         im != (unsigned)ISD::LAST_INDEXED_MODE; ++im) {
      setIndexedLoadAction(im, VT.getSimpleVT(), Legal);
      setIndexedStoreAction(im, VT.getSimpleVT(), Legal);
    }
  }
}

void AArch64TargetLowering::addDRTypeForNEON(MVT VT) {
  addRegisterClass(VT, &AArch64::FPR64RegClass);
  addTypeForNEON(VT, MVT::v2i32);
}

void AArch64TargetLowering::addQRTypeForNEON(MVT VT) {
  addRegisterClass(VT, &AArch64::FPR128RegClass);
  addTypeForNEON(VT, MVT::v4i32);
}

EVT AArch64TargetLowering::getSetCCResultType(LLVMContext &, EVT VT) const {
  if (!VT.isVector())
    return MVT::i32;
  return VT.changeVectorElementTypeToInteger();
}

/// computeKnownBitsForTargetNode - Determine which of the bits specified in
/// Mask are known to be either zero or one and return them in the
/// KnownZero/KnownOne bitsets.
void AArch64TargetLowering::computeKnownBitsForTargetNode(
    const SDValue Op, APInt &KnownZero, APInt &KnownOne,
    const SelectionDAG &DAG, unsigned Depth) const {
  switch (Op.getOpcode()) {
  default:
    break;
  case AArch64ISD::CSEL: {
    APInt KnownZero2, KnownOne2;
    DAG.computeKnownBits(Op->getOperand(0), KnownZero, KnownOne, Depth + 1);
    DAG.computeKnownBits(Op->getOperand(1), KnownZero2, KnownOne2, Depth + 1);
    KnownZero &= KnownZero2;
    KnownOne &= KnownOne2;
    break;
  }
  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::aarch64_ldaxr:
    case Intrinsic::aarch64_ldxr: {
      unsigned BitWidth = KnownOne.getBitWidth();
      EVT VT = cast<MemIntrinsicSDNode>(Op)->getMemoryVT();
      unsigned MemBits = VT.getScalarType().getSizeInBits();
      KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits);
      return;
    }
    }
    break;
  }
  case ISD::INTRINSIC_WO_CHAIN:
  case ISD::INTRINSIC_VOID: {
    unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
    switch (IntNo) {
    default:
      break;
    case Intrinsic::aarch64_neon_umaxv:
    case Intrinsic::aarch64_neon_uminv: {
      // Figure out the datatype of the vector operand. The UMINV instruction
      // will zero extend the result, so we can mark as known zero all the
      // bits larger than the element datatype. 32-bit or larget doesn't need
      // this as those are legal types and will be handled by isel directly.
      MVT VT = Op.getOperand(1).getValueType().getSimpleVT();
      unsigned BitWidth = KnownZero.getBitWidth();
      if (VT == MVT::v8i8 || VT == MVT::v16i8) {
        assert(BitWidth >= 8 && "Unexpected width!");
        APInt Mask = APInt::getHighBitsSet(BitWidth, BitWidth - 8);
        KnownZero |= Mask;
      } else if (VT == MVT::v4i16 || VT == MVT::v8i16) {
        assert(BitWidth >= 16 && "Unexpected width!");
        APInt Mask = APInt::getHighBitsSet(BitWidth, BitWidth - 16);
        KnownZero |= Mask;
      }
      break;
    } break;
    }
  }
  }
}

MVT AArch64TargetLowering::getScalarShiftAmountTy(EVT LHSTy) const {
  return MVT::i64;
}

unsigned AArch64TargetLowering::getMaximalGlobalOffset() const {
  // FIXME: On AArch64, this depends on the type.
  // Basically, the addressable offsets are up to 4095 * Ty.getSizeInBytes().
  // and the offset has to be a multiple of the related size in bytes.
  return 4095;
}

FastISel *
AArch64TargetLowering::createFastISel(FunctionLoweringInfo &funcInfo,
                                      const TargetLibraryInfo *libInfo) const {
  return AArch64::createFastISel(funcInfo, libInfo);
}

const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
  switch (Opcode) {
  default:
    return nullptr;
  case AArch64ISD::CALL:              return "AArch64ISD::CALL";
  case AArch64ISD::ADRP:              return "AArch64ISD::ADRP";
  case AArch64ISD::ADDlow:            return "AArch64ISD::ADDlow";
  case AArch64ISD::LOADgot:           return "AArch64ISD::LOADgot";
  case AArch64ISD::RET_FLAG:          return "AArch64ISD::RET_FLAG";
  case AArch64ISD::BRCOND:            return "AArch64ISD::BRCOND";
  case AArch64ISD::CSEL:              return "AArch64ISD::CSEL";
  case AArch64ISD::FCSEL:             return "AArch64ISD::FCSEL";
  case AArch64ISD::CSINV:             return "AArch64ISD::CSINV";
  case AArch64ISD::CSNEG:             return "AArch64ISD::CSNEG";
  case AArch64ISD::CSINC:             return "AArch64ISD::CSINC";
  case AArch64ISD::THREAD_POINTER:    return "AArch64ISD::THREAD_POINTER";
  case AArch64ISD::TLSDESC_CALL:      return "AArch64ISD::TLSDESC_CALL";
  case AArch64ISD::ADC:               return "AArch64ISD::ADC";
  case AArch64ISD::SBC:               return "AArch64ISD::SBC";
  case AArch64ISD::ADDS:              return "AArch64ISD::ADDS";
  case AArch64ISD::SUBS:              return "AArch64ISD::SUBS";
  case AArch64ISD::ADCS:              return "AArch64ISD::ADCS";
  case AArch64ISD::SBCS:              return "AArch64ISD::SBCS";
  case AArch64ISD::ANDS:              return "AArch64ISD::ANDS";
  case AArch64ISD::FCMP:              return "AArch64ISD::FCMP";
  case AArch64ISD::FMIN:              return "AArch64ISD::FMIN";
  case AArch64ISD::FMAX:              return "AArch64ISD::FMAX";
  case AArch64ISD::DUP:               return "AArch64ISD::DUP";
  case AArch64ISD::DUPLANE8:          return "AArch64ISD::DUPLANE8";
  case AArch64ISD::DUPLANE16:         return "AArch64ISD::DUPLANE16";
  case AArch64ISD::DUPLANE32:         return "AArch64ISD::DUPLANE32";
  case AArch64ISD::DUPLANE64:         return "AArch64ISD::DUPLANE64";
  case AArch64ISD::MOVI:              return "AArch64ISD::MOVI";
  case AArch64ISD::MOVIshift:         return "AArch64ISD::MOVIshift";
  case AArch64ISD::MOVIedit:          return "AArch64ISD::MOVIedit";
  case AArch64ISD::MOVImsl:           return "AArch64ISD::MOVImsl";
  case AArch64ISD::FMOV:              return "AArch64ISD::FMOV";
  case AArch64ISD::MVNIshift:         return "AArch64ISD::MVNIshift";
  case AArch64ISD::MVNImsl:           return "AArch64ISD::MVNImsl";
  case AArch64ISD::BICi:              return "AArch64ISD::BICi";
  case AArch64ISD::ORRi:              return "AArch64ISD::ORRi";
  case AArch64ISD::BSL:               return "AArch64ISD::BSL";
  case AArch64ISD::NEG:               return "AArch64ISD::NEG";
  case AArch64ISD::EXTR:              return "AArch64ISD::EXTR";
  case AArch64ISD::ZIP1:              return "AArch64ISD::ZIP1";
  case AArch64ISD::ZIP2:              return "AArch64ISD::ZIP2";
  case AArch64ISD::UZP1:              return "AArch64ISD::UZP1";
  case AArch64ISD::UZP2:              return "AArch64ISD::UZP2";
  case AArch64ISD::TRN1:              return "AArch64ISD::TRN1";
  case AArch64ISD::TRN2:              return "AArch64ISD::TRN2";
  case AArch64ISD::REV16:             return "AArch64ISD::REV16";
  case AArch64ISD::REV32:             return "AArch64ISD::REV32";
  case AArch64ISD::REV64:             return "AArch64ISD::REV64";
  case AArch64ISD::EXT:               return "AArch64ISD::EXT";
  case AArch64ISD::VSHL:              return "AArch64ISD::VSHL";
  case AArch64ISD::VLSHR:             return "AArch64ISD::VLSHR";
  case AArch64ISD::VASHR:             return "AArch64ISD::VASHR";
  case AArch64ISD::CMEQ:              return "AArch64ISD::CMEQ";
  case AArch64ISD::CMGE:              return "AArch64ISD::CMGE";
  case AArch64ISD::CMGT:              return "AArch64ISD::CMGT";
  case AArch64ISD::CMHI:              return "AArch64ISD::CMHI";
  case AArch64ISD::CMHS:              return "AArch64ISD::CMHS";
  case AArch64ISD::FCMEQ:             return "AArch64ISD::FCMEQ";
  case AArch64ISD::FCMGE:             return "AArch64ISD::FCMGE";
  case AArch64ISD::FCMGT:             return "AArch64ISD::FCMGT";
  case AArch64ISD::CMEQz:             return "AArch64ISD::CMEQz";
  case AArch64ISD::CMGEz:             return "AArch64ISD::CMGEz";
  case AArch64ISD::CMGTz:             return "AArch64ISD::CMGTz";
  case AArch64ISD::CMLEz:             return "AArch64ISD::CMLEz";
  case AArch64ISD::CMLTz:             return "AArch64ISD::CMLTz";
  case AArch64ISD::FCMEQz:            return "AArch64ISD::FCMEQz";
  case AArch64ISD::FCMGEz:            return "AArch64ISD::FCMGEz";
  case AArch64ISD::FCMGTz:            return "AArch64ISD::FCMGTz";
  case AArch64ISD::FCMLEz:            return "AArch64ISD::FCMLEz";
  case AArch64ISD::FCMLTz:            return "AArch64ISD::FCMLTz";
  case AArch64ISD::NOT:               return "AArch64ISD::NOT";
  case AArch64ISD::BIT:               return "AArch64ISD::BIT";
  case AArch64ISD::CBZ:               return "AArch64ISD::CBZ";
  case AArch64ISD::CBNZ:              return "AArch64ISD::CBNZ";
  case AArch64ISD::TBZ:               return "AArch64ISD::TBZ";
  case AArch64ISD::TBNZ:              return "AArch64ISD::TBNZ";
  case AArch64ISD::TC_RETURN:         return "AArch64ISD::TC_RETURN";
  case AArch64ISD::SITOF:             return "AArch64ISD::SITOF";
  case AArch64ISD::UITOF:             return "AArch64ISD::UITOF";
  case AArch64ISD::NVCAST:            return "AArch64ISD::NVCAST";
  case AArch64ISD::SQSHL_I:           return "AArch64ISD::SQSHL_I";
  case AArch64ISD::UQSHL_I:           return "AArch64ISD::UQSHL_I";
  case AArch64ISD::SRSHR_I:           return "AArch64ISD::SRSHR_I";
  case AArch64ISD::URSHR_I:           return "AArch64ISD::URSHR_I";
  case AArch64ISD::SQSHLU_I:          return "AArch64ISD::SQSHLU_I";
  case AArch64ISD::WrapperLarge:      return "AArch64ISD::WrapperLarge";
  case AArch64ISD::LD2post:           return "AArch64ISD::LD2post";
  case AArch64ISD::LD3post:           return "AArch64ISD::LD3post";
  case AArch64ISD::LD4post:           return "AArch64ISD::LD4post";
  case AArch64ISD::ST2post:           return "AArch64ISD::ST2post";
  case AArch64ISD::ST3post:           return "AArch64ISD::ST3post";
  case AArch64ISD::ST4post:           return "AArch64ISD::ST4post";
  case AArch64ISD::LD1x2post:         return "AArch64ISD::LD1x2post";
  case AArch64ISD::LD1x3post:         return "AArch64ISD::LD1x3post";
  case AArch64ISD::LD1x4post:         return "AArch64ISD::LD1x4post";
  case AArch64ISD::ST1x2post:         return "AArch64ISD::ST1x2post";
  case AArch64ISD::ST1x3post:         return "AArch64ISD::ST1x3post";
  case AArch64ISD::ST1x4post:         return "AArch64ISD::ST1x4post";
  case AArch64ISD::LD1DUPpost:        return "AArch64ISD::LD1DUPpost";
  case AArch64ISD::LD2DUPpost:        return "AArch64ISD::LD2DUPpost";
  case AArch64ISD::LD3DUPpost:        return "AArch64ISD::LD3DUPpost";
  case AArch64ISD::LD4DUPpost:        return "AArch64ISD::LD4DUPpost";
  case AArch64ISD::LD1LANEpost:       return "AArch64ISD::LD1LANEpost";
  case AArch64ISD::LD2LANEpost:       return "AArch64ISD::LD2LANEpost";
  case AArch64ISD::LD3LANEpost:       return "AArch64ISD::LD3LANEpost";
  case AArch64ISD::LD4LANEpost:       return "AArch64ISD::LD4LANEpost";
  case AArch64ISD::ST2LANEpost:       return "AArch64ISD::ST2LANEpost";
  case AArch64ISD::ST3LANEpost:       return "AArch64ISD::ST3LANEpost";
  case AArch64ISD::ST4LANEpost:       return "AArch64ISD::ST4LANEpost";
  case AArch64ISD::SMULL:             return "AArch64ISD::SMULL";
  case AArch64ISD::UMULL:             return "AArch64ISD::UMULL";
  }
}

MachineBasicBlock *
AArch64TargetLowering::EmitF128CSEL(MachineInstr *MI,
                                    MachineBasicBlock *MBB) const {
  // We materialise the F128CSEL pseudo-instruction as some control flow and a
  // phi node:

  // OrigBB:
  //     [... previous instrs leading to comparison ...]
  //     b.ne TrueBB
  //     b EndBB
  // TrueBB:
  //     ; Fallthrough
  // EndBB:
  //     Dest = PHI [IfTrue, TrueBB], [IfFalse, OrigBB]

  const TargetInstrInfo *TII =
      getTargetMachine().getSubtargetImpl()->getInstrInfo();
  MachineFunction *MF = MBB->getParent();
  const BasicBlock *LLVM_BB = MBB->getBasicBlock();
  DebugLoc DL = MI->getDebugLoc();
  MachineFunction::iterator It = MBB;
  ++It;

  unsigned DestReg = MI->getOperand(0).getReg();
  unsigned IfTrueReg = MI->getOperand(1).getReg();
  unsigned IfFalseReg = MI->getOperand(2).getReg();
  unsigned CondCode = MI->getOperand(3).getImm();
  bool NZCVKilled = MI->getOperand(4).isKill();

  MachineBasicBlock *TrueBB = MF->CreateMachineBasicBlock(LLVM_BB);
  MachineBasicBlock *EndBB = MF->CreateMachineBasicBlock(LLVM_BB);
  MF->insert(It, TrueBB);
  MF->insert(It, EndBB);

  // Transfer rest of current basic-block to EndBB
  EndBB->splice(EndBB->begin(), MBB, std::next(MachineBasicBlock::iterator(MI)),
                MBB->end());
  EndBB->transferSuccessorsAndUpdatePHIs(MBB);

  BuildMI(MBB, DL, TII->get(AArch64::Bcc)).addImm(CondCode).addMBB(TrueBB);
  BuildMI(MBB, DL, TII->get(AArch64::B)).addMBB(EndBB);
  MBB->addSuccessor(TrueBB);
  MBB->addSuccessor(EndBB);

  // TrueBB falls through to the end.
  TrueBB->addSuccessor(EndBB);

  if (!NZCVKilled) {
    TrueBB->addLiveIn(AArch64::NZCV);
    EndBB->addLiveIn(AArch64::NZCV);
  }

  BuildMI(*EndBB, EndBB->begin(), DL, TII->get(AArch64::PHI), DestReg)
      .addReg(IfTrueReg)
      .addMBB(TrueBB)
      .addReg(IfFalseReg)
      .addMBB(MBB);

  MI->eraseFromParent();
  return EndBB;
}

MachineBasicBlock *
AArch64TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
                                                 MachineBasicBlock *BB) const {
  switch (MI->getOpcode()) {
  default:
#ifndef NDEBUG
    MI->dump();
#endif
    llvm_unreachable("Unexpected instruction for custom inserter!");

  case AArch64::F128CSEL:
    return EmitF128CSEL(MI, BB);

  case TargetOpcode::STACKMAP:
  case TargetOpcode::PATCHPOINT:
    return emitPatchPoint(MI, BB);
  }
}

//===----------------------------------------------------------------------===//
// AArch64 Lowering private implementation.
//===----------------------------------------------------------------------===//

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

/// changeIntCCToAArch64CC - Convert a DAG integer condition code to an AArch64
/// CC
static AArch64CC::CondCode changeIntCCToAArch64CC(ISD::CondCode CC) {
  switch (CC) {
  default:
    llvm_unreachable("Unknown condition code!");
  case ISD::SETNE:
    return AArch64CC::NE;
  case ISD::SETEQ:
    return AArch64CC::EQ;
  case ISD::SETGT:
    return AArch64CC::GT;
  case ISD::SETGE:
    return AArch64CC::GE;
  case ISD::SETLT:
    return AArch64CC::LT;
  case ISD::SETLE:
    return AArch64CC::LE;
  case ISD::SETUGT:
    return AArch64CC::HI;
  case ISD::SETUGE:
    return AArch64CC::HS;
  case ISD::SETULT:
    return AArch64CC::LO;
  case ISD::SETULE:
    return AArch64CC::LS;
  }
}

/// changeFPCCToAArch64CC - Convert a DAG fp condition code to an AArch64 CC.
static void changeFPCCToAArch64CC(ISD::CondCode CC,
                                  AArch64CC::CondCode &CondCode,
                                  AArch64CC::CondCode &CondCode2) {
  CondCode2 = AArch64CC::AL;
  switch (CC) {
  default:
    llvm_unreachable("Unknown FP condition!");
  case ISD::SETEQ:
  case ISD::SETOEQ:
    CondCode = AArch64CC::EQ;
    break;
  case ISD::SETGT:
  case ISD::SETOGT:
    CondCode = AArch64CC::GT;
    break;
  case ISD::SETGE:
  case ISD::SETOGE:
    CondCode = AArch64CC::GE;
    break;
  case ISD::SETOLT:
    CondCode = AArch64CC::MI;
    break;
  case ISD::SETOLE:
    CondCode = AArch64CC::LS;
    break;
  case ISD::SETONE:
    CondCode = AArch64CC::MI;
    CondCode2 = AArch64CC::GT;
    break;
  case ISD::SETO:
    CondCode = AArch64CC::VC;
    break;
  case ISD::SETUO:
    CondCode = AArch64CC::VS;
    break;
  case ISD::SETUEQ:
    CondCode = AArch64CC::EQ;
    CondCode2 = AArch64CC::VS;
    break;
  case ISD::SETUGT:
    CondCode = AArch64CC::HI;
    break;
  case ISD::SETUGE:
    CondCode = AArch64CC::PL;
    break;
  case ISD::SETLT:
  case ISD::SETULT:
    CondCode = AArch64CC::LT;
    break;
  case ISD::SETLE:
  case ISD::SETULE:
    CondCode = AArch64CC::LE;
    break;
  case ISD::SETNE:
  case ISD::SETUNE:
    CondCode = AArch64CC::NE;
    break;
  }
}

/// changeVectorFPCCToAArch64CC - Convert a DAG fp condition code to an AArch64
/// CC usable with the vector instructions. Fewer operations are available
/// without a real NZCV register, so we have to use less efficient combinations
/// to get the same effect.
static void changeVectorFPCCToAArch64CC(ISD::CondCode CC,
                                        AArch64CC::CondCode &CondCode,
                                        AArch64CC::CondCode &CondCode2,
                                        bool &Invert) {
  Invert = false;
  switch (CC) {
  default:
    // Mostly the scalar mappings work fine.
    changeFPCCToAArch64CC(CC, CondCode, CondCode2);
    break;
  case ISD::SETUO:
    Invert = true; // Fallthrough
  case ISD::SETO:
    CondCode = AArch64CC::MI;
    CondCode2 = AArch64CC::GE;
    break;
  case ISD::SETUEQ:
  case ISD::SETULT:
  case ISD::SETULE:
  case ISD::SETUGT:
  case ISD::SETUGE:
    // All of the compare-mask comparisons are ordered, but we can switch
    // between the two by a double inversion. E.g. ULE == !OGT.
    Invert = true;
    changeFPCCToAArch64CC(getSetCCInverse(CC, false), CondCode, CondCode2);
    break;
  }
}

static bool isLegalArithImmed(uint64_t C) {
  // Matches AArch64DAGToDAGISel::SelectArithImmed().
  return (C >> 12 == 0) || ((C & 0xFFFULL) == 0 && C >> 24 == 0);
}

static SDValue emitComparison(SDValue LHS, SDValue RHS, ISD::CondCode CC,
                              SDLoc dl, SelectionDAG &DAG) {
  EVT VT = LHS.getValueType();

  if (VT.isFloatingPoint())
    return DAG.getNode(AArch64ISD::FCMP, dl, VT, LHS, RHS);

  // The CMP instruction is just an alias for SUBS, and representing it as
  // SUBS means that it's possible to get CSE with subtract operations.
  // A later phase can perform the optimization of setting the destination
  // register to WZR/XZR if it ends up being unused.
  unsigned Opcode = AArch64ISD::SUBS;

  if (RHS.getOpcode() == ISD::SUB && isa<ConstantSDNode>(RHS.getOperand(0)) &&
      cast<ConstantSDNode>(RHS.getOperand(0))->getZExtValue() == 0 &&
      (CC == ISD::SETEQ || CC == ISD::SETNE)) {
    // We'd like to combine a (CMP op1, (sub 0, op2) into a CMN instruction on
    // the grounds that "op1 - (-op2) == op1 + op2". However, the C and V flags
    // can be set differently by this operation. It comes down to whether
    // "SInt(~op2)+1 == SInt(~op2+1)" (and the same for UInt). If they are then
    // everything is fine. If not then the optimization is wrong. Thus general
    // comparisons are only valid if op2 != 0.

    // So, finally, the only LLVM-native comparisons that don't mention C and V
    // are SETEQ and SETNE. They're the only ones we can safely use CMN for in
    // the absence of information about op2.
    Opcode = AArch64ISD::ADDS;
    RHS = RHS.getOperand(1);
  } else if (LHS.getOpcode() == ISD::AND && isa<ConstantSDNode>(RHS) &&
             cast<ConstantSDNode>(RHS)->getZExtValue() == 0 &&
             !isUnsignedIntSetCC(CC)) {
    // Similarly, (CMP (and X, Y), 0) can be implemented with a TST
    // (a.k.a. ANDS) except that the flags are only guaranteed to work for one
    // of the signed comparisons.
    Opcode = AArch64ISD::ANDS;
    RHS = LHS.getOperand(1);
    LHS = LHS.getOperand(0);
  }

  return DAG.getNode(Opcode, dl, DAG.getVTList(VT, MVT::i32), LHS, RHS)
      .getValue(1);
}

static SDValue getAArch64Cmp(SDValue LHS, SDValue RHS, ISD::CondCode CC,
                             SDValue &AArch64cc, SelectionDAG &DAG, SDLoc dl) {
  SDValue Cmp;
  AArch64CC::CondCode AArch64CC;
  if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS.getNode())) {
    EVT VT = RHS.getValueType();
    uint64_t C = RHSC->getZExtValue();
    if (!isLegalArithImmed(C)) {
      // Constant does not fit, try adjusting it by one?
      switch (CC) {
      default:
        break;
      case ISD::SETLT:
      case ISD::SETGE:
        if ((VT == MVT::i32 && C != 0x80000000 &&
             isLegalArithImmed((uint32_t)(C - 1))) ||
            (VT == MVT::i64 && C != 0x80000000ULL &&
             isLegalArithImmed(C - 1ULL))) {
          CC = (CC == ISD::SETLT) ? ISD::SETLE : ISD::SETGT;
          C = (VT == MVT::i32) ? (uint32_t)(C - 1) : C - 1;
          RHS = DAG.getConstant(C, VT);
        }
        break;
      case ISD::SETULT:
      case ISD::SETUGE:
        if ((VT == MVT::i32 && C != 0 &&
             isLegalArithImmed((uint32_t)(C - 1))) ||
            (VT == MVT::i64 && C != 0ULL && isLegalArithImmed(C - 1ULL))) {
          CC = (CC == ISD::SETULT) ? ISD::SETULE : ISD::SETUGT;
          C = (VT == MVT::i32) ? (uint32_t)(C - 1) : C - 1;
          RHS = DAG.getConstant(C, VT);
        }
        break;
      case ISD::SETLE:
      case ISD::SETGT:
        if ((VT == MVT::i32 && C != INT32_MAX &&
             isLegalArithImmed((uint32_t)(C + 1))) ||
            (VT == MVT::i64 && C != INT64_MAX &&
             isLegalArithImmed(C + 1ULL))) {
          CC = (CC == ISD::SETLE) ? ISD::SETLT : ISD::SETGE;
          C = (VT == MVT::i32) ? (uint32_t)(C + 1) : C + 1;
          RHS = DAG.getConstant(C, VT);
        }
        break;
      case ISD::SETULE:
      case ISD::SETUGT:
        if ((VT == MVT::i32 && C != UINT32_MAX &&
             isLegalArithImmed((uint32_t)(C + 1))) ||
            (VT == MVT::i64 && C != UINT64_MAX &&
             isLegalArithImmed(C + 1ULL))) {
          CC = (CC == ISD::SETULE) ? ISD::SETULT : ISD::SETUGE;
          C = (VT == MVT::i32) ? (uint32_t)(C + 1) : C + 1;
          RHS = DAG.getConstant(C, VT);
        }
        break;
      }
    }
  }
  // The imm operand of ADDS is an unsigned immediate, in the range 0 to 4095.
  // For the i8 operand, the largest immediate is 255, so this can be easily
  // encoded in the compare instruction. For the i16 operand, however, the
  // largest immediate cannot be encoded in the compare.
  // Therefore, use a sign extending load and cmn to avoid materializing the -1
  // constant. For example,
  // movz w1, #65535
  // ldrh w0, [x0, #0]
  // cmp w0, w1
  // >
  // ldrsh w0, [x0, #0]
  // cmn w0, #1
  // Fundamental, we're relying on the property that (zext LHS) == (zext RHS)
  // if and only if (sext LHS) == (sext RHS). The checks are in place to ensure
  // both the LHS and RHS are truely zero extended and to make sure the
  // transformation is profitable.
  if ((CC == ISD::SETEQ || CC == ISD::SETNE) && isa<ConstantSDNode>(RHS)) {
    if ((cast<ConstantSDNode>(RHS)->getZExtValue() >> 16 == 0) &&
        isa<LoadSDNode>(LHS)) {
      if (cast<LoadSDNode>(LHS)->getExtensionType() == ISD::ZEXTLOAD &&
          cast<LoadSDNode>(LHS)->getMemoryVT() == MVT::i16 &&
          LHS.getNode()->hasNUsesOfValue(1, 0)) {
        int16_t ValueofRHS = cast<ConstantSDNode>(RHS)->getZExtValue();
        if (ValueofRHS < 0 && isLegalArithImmed(-ValueofRHS)) {
          SDValue SExt =
              DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, LHS.getValueType(), LHS,
                          DAG.getValueType(MVT::i16));
          Cmp = emitComparison(SExt,
                               DAG.getConstant(ValueofRHS, RHS.getValueType()),
                               CC, dl, DAG);
          AArch64CC = changeIntCCToAArch64CC(CC);
          AArch64cc = DAG.getConstant(AArch64CC, MVT::i32);
          return Cmp;
        }
      }
    }
  }
  Cmp = emitComparison(LHS, RHS, CC, dl, DAG);
  AArch64CC = changeIntCCToAArch64CC(CC);
  AArch64cc = DAG.getConstant(AArch64CC, MVT::i32);
  return Cmp;
}

static std::pair<SDValue, SDValue>
getAArch64XALUOOp(AArch64CC::CondCode &CC, SDValue Op, SelectionDAG &DAG) {
  assert((Op.getValueType() == MVT::i32 || Op.getValueType() == MVT::i64) &&
         "Unsupported value type");
  SDValue Value, Overflow;
  SDLoc DL(Op);
  SDValue LHS = Op.getOperand(0);
  SDValue RHS = Op.getOperand(1);
  unsigned Opc = 0;
  switch (Op.getOpcode()) {
  default:
    llvm_unreachable("Unknown overflow instruction!");
  case ISD::SADDO:
    Opc = AArch64ISD::ADDS;
    CC = AArch64CC::VS;
    break;
  case ISD::UADDO:
    Opc = AArch64ISD::ADDS;
    CC = AArch64CC::HS;
    break;
  case ISD::SSUBO:
    Opc = AArch64ISD::SUBS;
    CC = AArch64CC::VS;
    break;
  case ISD::USUBO:
    Opc = AArch64ISD::SUBS;
    CC = AArch64CC::LO;
    break;
  // Multiply needs a little bit extra work.
  case ISD::SMULO:
  case ISD::UMULO: {
    CC = AArch64CC::NE;
    bool IsSigned = (Op.getOpcode() == ISD::SMULO) ? true : false;
    if (Op.getValueType() == MVT::i32) {
      unsigned ExtendOpc = IsSigned ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
      // For a 32 bit multiply with overflow check we want the instruction
      // selector to generate a widening multiply (SMADDL/UMADDL). For that we
      // need to generate the following pattern:
      // (i64 add 0, (i64 mul (i64 sext|zext i32 %a), (i64 sext|zext i32 %b))
      LHS = DAG.getNode(ExtendOpc, DL, MVT::i64, LHS);
      RHS = DAG.getNode(ExtendOpc, DL, MVT::i64, RHS);
      SDValue Mul = DAG.getNode(ISD::MUL, DL, MVT::i64, LHS, RHS);
      SDValue Add = DAG.getNode(ISD::ADD, DL, MVT::i64, Mul,
                                DAG.getConstant(0, MVT::i64));
      // On AArch64 the upper 32 bits are always zero extended for a 32 bit
      // operation. We need to clear out the upper 32 bits, because we used a
      // widening multiply that wrote all 64 bits. In the end this should be a
      // noop.
      Value = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Add);
      if (IsSigned) {
        // The signed overflow check requires more than just a simple check for
        // any bit set in the upper 32 bits of the result. These bits could be
        // just the sign bits of a negative number. To perform the overflow
        // check we have to arithmetic shift right the 32nd bit of the result by
        // 31 bits. Then we compare the result to the upper 32 bits.
        SDValue UpperBits = DAG.getNode(ISD::SRL, DL, MVT::i64, Add,
                                        DAG.getConstant(32, MVT::i64));
        UpperBits = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, UpperBits);
        SDValue LowerBits = DAG.getNode(ISD::SRA, DL, MVT::i32, Value,
                                        DAG.getConstant(31, MVT::i64));
        // It is important that LowerBits is last, otherwise the arithmetic
        // shift will not be folded into the compare (SUBS).
        SDVTList VTs = DAG.getVTList(MVT::i32, MVT::i32);
        Overflow = DAG.getNode(AArch64ISD::SUBS, DL, VTs, UpperBits, LowerBits)
                       .getValue(1);
      } else {
        // The overflow check for unsigned multiply is easy. We only need to
        // check if any of the upper 32 bits are set. This can be done with a
        // CMP (shifted register). For that we need to generate the following
        // pattern:
        // (i64 AArch64ISD::SUBS i64 0, (i64 srl i64 %Mul, i64 32)
        SDValue UpperBits = DAG.getNode(ISD::SRL, DL, MVT::i64, Mul,
                                        DAG.getConstant(32, MVT::i64));
        SDVTList VTs = DAG.getVTList(MVT::i64, MVT::i32);
        Overflow =
            DAG.getNode(AArch64ISD::SUBS, DL, VTs, DAG.getConstant(0, MVT::i64),
                        UpperBits).getValue(1);
      }
      break;
    }
    assert(Op.getValueType() == MVT::i64 && "Expected an i64 value type");
    // For the 64 bit multiply
    Value = DAG.getNode(ISD::MUL, DL, MVT::i64, LHS, RHS);
    if (IsSigned) {
      SDValue UpperBits = DAG.getNode(ISD::MULHS, DL, MVT::i64, LHS, RHS);
      SDValue LowerBits = DAG.getNode(ISD::SRA, DL, MVT::i64, Value,
                                      DAG.getConstant(63, MVT::i64));
      // It is important that LowerBits is last, otherwise the arithmetic
      // shift will not be folded into the compare (SUBS).
      SDVTList VTs = DAG.getVTList(MVT::i64, MVT::i32);
      Overflow = DAG.getNode(AArch64ISD::SUBS, DL, VTs, UpperBits, LowerBits)
                     .getValue(1);
    } else {
      SDValue UpperBits = DAG.getNode(ISD::MULHU, DL, MVT::i64, LHS, RHS);
      SDVTList VTs = DAG.getVTList(MVT::i64, MVT::i32);
      Overflow =
          DAG.getNode(AArch64ISD::SUBS, DL, VTs, DAG.getConstant(0, MVT::i64),
                      UpperBits).getValue(1);
    }
    break;
  }
  } // switch (...)

  if (Opc) {
    SDVTList VTs = DAG.getVTList(Op->getValueType(0), MVT::i32);

    // Emit the AArch64 operation with overflow check.
    Value = DAG.getNode(Opc, DL, VTs, LHS, RHS);
    Overflow = Value.getValue(1);
  }
  return std::make_pair(Value, Overflow);
}

SDValue AArch64TargetLowering::LowerF128Call(SDValue Op, SelectionDAG &DAG,
                                             RTLIB::Libcall Call) const {
  SmallVector<SDValue, 2> Ops;
  for (unsigned i = 0, e = Op->getNumOperands(); i != e; ++i)
    Ops.push_back(Op.getOperand(i));

  return makeLibCall(DAG, Call, MVT::f128, &Ops[0], Ops.size(), false,
                     SDLoc(Op)).first;
}

static SDValue LowerXOR(SDValue Op, SelectionDAG &DAG) {
  SDValue Sel = Op.getOperand(0);
  SDValue Other = Op.getOperand(1);

  // If neither operand is a SELECT_CC, give up.
  if (Sel.getOpcode() != ISD::SELECT_CC)
    std::swap(Sel, Other);
  if (Sel.getOpcode() != ISD::SELECT_CC)
    return Op;

  // The folding we want to perform is:
  // (xor x, (select_cc a, b, cc, 0, -1) )
  //   -->
  // (csel x, (xor x, -1), cc ...)
  //
  // The latter will get matched to a CSINV instruction.

  ISD::CondCode CC = cast<CondCodeSDNode>(Sel.getOperand(4))->get();
  SDValue LHS = Sel.getOperand(0);
  SDValue RHS = Sel.getOperand(1);
  SDValue TVal = Sel.getOperand(2);
  SDValue FVal = Sel.getOperand(3);
  SDLoc dl(Sel);

  // FIXME: This could be generalized to non-integer comparisons.
  if (LHS.getValueType() != MVT::i32 && LHS.getValueType() != MVT::i64)
    return Op;

  ConstantSDNode *CFVal = dyn_cast<ConstantSDNode>(FVal);
  ConstantSDNode *CTVal = dyn_cast<ConstantSDNode>(TVal);

  // The the values aren't constants, this isn't the pattern we're looking for.
  if (!CFVal || !CTVal)
    return Op;

  // We can commute the SELECT_CC by inverting the condition.  This
  // might be needed to make this fit into a CSINV pattern.
  if (CTVal->isAllOnesValue() && CFVal->isNullValue()) {
    std::swap(TVal, FVal);
    std::swap(CTVal, CFVal);
    CC = ISD::getSetCCInverse(CC, true);
  }

  // If the constants line up, perform the transform!
  if (CTVal->isNullValue() && CFVal->isAllOnesValue()) {
    SDValue CCVal;
    SDValue Cmp = getAArch64Cmp(LHS, RHS, CC, CCVal, DAG, dl);

    FVal = Other;
    TVal = DAG.getNode(ISD::XOR, dl, Other.getValueType(), Other,
                       DAG.getConstant(-1ULL, Other.getValueType()));

    return DAG.getNode(AArch64ISD::CSEL, dl, Sel.getValueType(), FVal, TVal,
                       CCVal, Cmp);
  }

  return Op;
}

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

  // Let legalize expand this if it isn't a legal type yet.
  if (!DAG.getTargetLoweringInfo().isTypeLegal(VT))
    return SDValue();

  SDVTList VTs = DAG.getVTList(VT, MVT::i32);

  unsigned Opc;
  bool ExtraOp = false;
  switch (Op.getOpcode()) {
  default:
    llvm_unreachable("Invalid code");
  case ISD::ADDC:
    Opc = AArch64ISD::ADDS;
    break;
  case ISD::SUBC:
    Opc = AArch64ISD::SUBS;
    break;
  case ISD::ADDE:
    Opc = AArch64ISD::ADCS;
    ExtraOp = true;
    break;
  case ISD::SUBE:
    Opc = AArch64ISD::SBCS;
    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));
}

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

  AArch64CC::CondCode CC;
  // The actual operation that sets the overflow or carry flag.
  SDValue Value, Overflow;
  std::tie(Value, Overflow) = getAArch64XALUOOp(CC, Op, DAG);

  // We use 0 and 1 as false and true values.
  SDValue TVal = DAG.getConstant(1, MVT::i32);
  SDValue FVal = DAG.getConstant(0, MVT::i32);

  // We use an inverted condition, because the conditional select is inverted
  // too. This will allow it to be selected to a single instruction:
  // CSINC Wd, WZR, WZR, invert(cond).
  SDValue CCVal = DAG.getConstant(getInvertedCondCode(CC), MVT::i32);
  Overflow = DAG.getNode(AArch64ISD::CSEL, SDLoc(Op), MVT::i32, FVal, TVal,
                         CCVal, Overflow);

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

// Prefetch operands are:
// 1: Address to prefetch
// 2: bool isWrite
// 3: int locality (0 = no locality ... 3 = extreme locality)
// 4: bool isDataCache
static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG) {
  SDLoc DL(Op);
  unsigned IsWrite = cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue();
  unsigned Locality = cast<ConstantSDNode>(Op.getOperand(3))->getZExtValue();
  unsigned IsData = cast<ConstantSDNode>(Op.getOperand(4))->getZExtValue();

  bool IsStream = !Locality;
  // When the locality number is set
  if (Locality) {
    // The front-end should have filtered out the out-of-range values
    assert(Locality <= 3 && "Prefetch locality out-of-range");
    // The locality degree is the opposite of the cache speed.
    // Put the number the other way around.
    // The encoding starts at 0 for level 1
    Locality = 3 - Locality;
  }

  // built the mask value encoding the expected behavior.
  unsigned PrfOp = (IsWrite << 4) |     // Load/Store bit
                   (!IsData << 3) |     // IsDataCache bit
                   (Locality << 1) |    // Cache level bits
                   (unsigned)IsStream;  // Stream bit
  return DAG.getNode(AArch64ISD::PREFETCH, DL, MVT::Other, Op.getOperand(0),
                     DAG.getConstant(PrfOp, MVT::i32), Op.getOperand(1));
}

SDValue AArch64TargetLowering::LowerFP_EXTEND(SDValue Op,
                                              SelectionDAG &DAG) const {
  assert(Op.getValueType() == MVT::f128 && "Unexpected lowering");

  RTLIB::Libcall LC;
  LC = RTLIB::getFPEXT(Op.getOperand(0).getValueType(), Op.getValueType());

  return LowerF128Call(Op, DAG, LC);
}

SDValue AArch64TargetLowering::LowerFP_ROUND(SDValue Op,
                                             SelectionDAG &DAG) const {
  if (Op.getOperand(0).getValueType() != MVT::f128) {
    // It's legal except when f128 is involved
    return Op;
  }

  RTLIB::Libcall LC;
  LC = RTLIB::getFPROUND(Op.getOperand(0).getValueType(), Op.getValueType());

  // FP_ROUND node has a second operand indicating whether it is known to be
  // precise. That doesn't take part in the LibCall so we can't directly use
  // LowerF128Call.
  SDValue SrcVal = Op.getOperand(0);
  return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1,
                     /*isSigned*/ false, SDLoc(Op)).first;
}

static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
  // Warning: We maintain cost tables in AArch64TargetTransformInfo.cpp.
  // Any additional optimization in this function should be recorded
  // in the cost tables.
  EVT InVT = Op.getOperand(0).getValueType();
  EVT VT = Op.getValueType();

  if (VT.getSizeInBits() < InVT.getSizeInBits()) {
    SDLoc dl(Op);
    SDValue Cv =
        DAG.getNode(Op.getOpcode(), dl, InVT.changeVectorElementTypeToInteger(),
                    Op.getOperand(0));
    return DAG.getNode(ISD::TRUNCATE, dl, VT, Cv);
  }

  if (VT.getSizeInBits() > InVT.getSizeInBits()) {
    SDLoc dl(Op);
    MVT ExtVT =
        MVT::getVectorVT(MVT::getFloatingPointVT(VT.getScalarSizeInBits()),
                         VT.getVectorNumElements());
    SDValue Ext = DAG.getNode(ISD::FP_EXTEND, dl, ExtVT, Op.getOperand(0));
    return DAG.getNode(Op.getOpcode(), dl, VT, Ext);
  }

  // Type changing conversions are illegal.
  return Op;
}

SDValue AArch64TargetLowering::LowerFP_TO_INT(SDValue Op,
                                              SelectionDAG &DAG) const {
  if (Op.getOperand(0).getValueType().isVector())
    return LowerVectorFP_TO_INT(Op, DAG);

  if (Op.getOperand(0).getValueType() != MVT::f128) {
    // It's legal except when f128 is involved
    return Op;
  }

  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());

  SmallVector<SDValue, 2> Ops;
  for (unsigned i = 0, e = Op->getNumOperands(); i != e; ++i)
    Ops.push_back(Op.getOperand(i));

  return makeLibCall(DAG, LC, Op.getValueType(), &Ops[0], Ops.size(), false,
                     SDLoc(Op)).first;
}

static SDValue LowerVectorINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
  // Warning: We maintain cost tables in AArch64TargetTransformInfo.cpp.
  // Any additional optimization in this function should be recorded
  // in the cost tables.
  EVT VT = Op.getValueType();
  SDLoc dl(Op);
  SDValue In = Op.getOperand(0);
  EVT InVT = In.getValueType();

  if (VT.getSizeInBits() < InVT.getSizeInBits()) {
    MVT CastVT =
        MVT::getVectorVT(MVT::getFloatingPointVT(InVT.getScalarSizeInBits()),
                         InVT.getVectorNumElements());
    In = DAG.getNode(Op.getOpcode(), dl, CastVT, In);
    return DAG.getNode(ISD::FP_ROUND, dl, VT, In, DAG.getIntPtrConstant(0));
  }

  if (VT.getSizeInBits() > InVT.getSizeInBits()) {
    unsigned CastOpc =
        Op.getOpcode() == ISD::SINT_TO_FP ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
    EVT CastVT = VT.changeVectorElementTypeToInteger();
    In = DAG.getNode(CastOpc, dl, CastVT, In);
    return DAG.getNode(Op.getOpcode(), dl, VT, In);
  }

  return Op;
}

SDValue AArch64TargetLowering::LowerINT_TO_FP(SDValue Op,
                                            SelectionDAG &DAG) const {
  if (Op.getValueType().isVector())
    return LowerVectorINT_TO_FP(Op, DAG);

  // i128 conversions are libcalls.
  if (Op.getOperand(0).getValueType() == MVT::i128)
    return SDValue();

  // Other conversions are legal, unless it's to the completely software-based
  // fp128.
  if (Op.getValueType() != MVT::f128)
    return Op;

  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 LowerF128Call(Op, DAG, LC);
}

SDValue AArch64TargetLowering::LowerFSINCOS(SDValue Op,
                                            SelectionDAG &DAG) const {
  // For iOS, we want to call an alternative entry point: __sincos_stret,
  // which returns the values in two S / D registers.
  SDLoc dl(Op);
  SDValue Arg = Op.getOperand(0);
  EVT ArgVT = Arg.getValueType();
  Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());

  ArgListTy Args;
  ArgListEntry 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());

  StructType *RetTy = StructType::get(ArgTy, ArgTy, nullptr);
  TargetLowering::CallLoweringInfo CLI(DAG);
  CLI.setDebugLoc(dl).setChain(DAG.getEntryNode())
    .setCallee(CallingConv::Fast, RetTy, Callee, std::move(Args), 0);

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

static SDValue LowerBITCAST(SDValue Op, SelectionDAG &DAG) {
  if (Op.getValueType() != MVT::f16)
    return SDValue();

  assert(Op.getOperand(0).getValueType() == MVT::i16);
  SDLoc DL(Op);

  Op = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, Op.getOperand(0));
  Op = DAG.getNode(ISD::BITCAST, DL, MVT::f32, Op);
  return SDValue(
      DAG.getMachineNode(TargetOpcode::EXTRACT_SUBREG, DL, MVT::f16, Op,
                         DAG.getTargetConstant(AArch64::hsub, MVT::i32)),
      0);
}

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;
  }
}

static SDValue addRequiredExtensionForVectorMULL(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);
}

static bool isExtendedBUILD_VECTOR(SDNode *N, SelectionDAG &DAG,
                                   bool isSigned) {
  EVT VT = N->getValueType(0);

  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;
}

static SDValue skipExtensionForVectorMULL(SDNode *N, SelectionDAG &DAG) {
  if (N->getOpcode() == ISD::SIGN_EXTEND || N->getOpcode() == ISD::ZERO_EXTEND)
    return addRequiredExtensionForVectorMULL(N->getOperand(0), DAG,
                                             N->getOperand(0)->getValueType(0),
                                             N->getValueType(0),
                                             N->getOpcode());

  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 isSignExtended(SDNode *N, SelectionDAG &DAG) {
  if (N->getOpcode() == ISD::SIGN_EXTEND)
    return true;
  if (isExtendedBUILD_VECTOR(N, DAG, true))
    return true;
  return false;
}

static bool isZeroExtended(SDNode *N, SelectionDAG &DAG) {
  if (N->getOpcode() == ISD::ZERO_EXTEND)
    return true;
  if (isExtendedBUILD_VECTOR(N, DAG, false))
    return true;
  return false;
}

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 = AArch64ISD::SMULL;
  else {
    bool isN0ZExt = isZeroExtended(N0, DAG);
    bool isN1ZExt = isZeroExtended(N1, DAG);
    if (isN0ZExt && isN1ZExt)
      NewOpc = AArch64ISD::UMULL;
    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 = AArch64ISD::SMULL;
        isMLA = true;
      } else if (isN1ZExt && isAddSubZExt(N0, DAG)) {
        NewOpc =  AArch64ISD::UMULL;
        isMLA = true;
      } else if (isN0ZExt && isAddSubZExt(N1, DAG)) {
        std::swap(N0, N1);
        NewOpc =  AArch64ISD::UMULL;
        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 S/UMULL instruction
  SDLoc DL(Op);
  SDValue Op0;
  SDValue Op1 = skipExtensionForVectorMULL(N1, DAG);
  if (!isMLA) {
    Op0 = skipExtensionForVectorMULL(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 (S/UMULL A, C) + (S/UMULL B, C) during
  // isel lowering to take advantage of no-stall back to back s/umul + s/umla.
  // This is true for CPUs with accumulate forwarding such as Cortex-A53/A57
  SDValue N00 = skipExtensionForVectorMULL(N0->getOperand(0).getNode(), DAG);
  SDValue N01 = skipExtensionForVectorMULL(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));
}

SDValue AArch64TargetLowering::LowerOperation(SDValue Op,
                                              SelectionDAG &DAG) const {
  switch (Op.getOpcode()) {
  default:
    llvm_unreachable("unimplemented operand");
    return SDValue();
  case ISD::BITCAST:
    return LowerBITCAST(Op, DAG);
  case ISD::GlobalAddress:
    return LowerGlobalAddress(Op, DAG);
  case ISD::GlobalTLSAddress:
    return LowerGlobalTLSAddress(Op, DAG);
  case ISD::SETCC:
    return LowerSETCC(Op, DAG);
  case ISD::BR_CC:
    return LowerBR_CC(Op, DAG);
  case ISD::SELECT:
    return LowerSELECT(Op, DAG);
  case ISD::SELECT_CC:
    return LowerSELECT_CC(Op, DAG);
  case ISD::JumpTable:
    return LowerJumpTable(Op, DAG);
  case ISD::ConstantPool:
    return LowerConstantPool(Op, DAG);
  case ISD::BlockAddress:
    return LowerBlockAddress(Op, DAG);
  case ISD::VASTART:
    return LowerVASTART(Op, DAG);
  case ISD::VACOPY:
    return LowerVACOPY(Op, DAG);
  case ISD::VAARG:
    return LowerVAARG(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:
  case ISD::SMULO:
  case ISD::UMULO:
    return LowerXALUO(Op, DAG);
  case ISD::FADD:
    return LowerF128Call(Op, DAG, RTLIB::ADD_F128);
  case ISD::FSUB:
    return LowerF128Call(Op, DAG, RTLIB::SUB_F128);
  case ISD::FMUL:
    return LowerF128Call(Op, DAG, RTLIB::MUL_F128);
  case ISD::FDIV:
    return LowerF128Call(Op, DAG, RTLIB::DIV_F128);
  case ISD::FP_ROUND:
    return LowerFP_ROUND(Op, DAG);
  case ISD::FP_EXTEND:
    return LowerFP_EXTEND(Op, DAG);
  case ISD::FRAMEADDR:
    return LowerFRAMEADDR(Op, DAG);
  case ISD::RETURNADDR:
    return LowerRETURNADDR(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::BUILD_VECTOR:
    return LowerBUILD_VECTOR(Op, DAG);
  case ISD::VECTOR_SHUFFLE:
    return LowerVECTOR_SHUFFLE(Op, DAG);
  case ISD::EXTRACT_SUBVECTOR:
    return LowerEXTRACT_SUBVECTOR(Op, DAG);
  case ISD::SRA:
  case ISD::SRL:
  case ISD::SHL:
    return LowerVectorSRA_SRL_SHL(Op, DAG);
  case ISD::SHL_PARTS:
    return LowerShiftLeftParts(Op, DAG);
  case ISD::SRL_PARTS:
  case ISD::SRA_PARTS:
    return LowerShiftRightParts(Op, DAG);
  case ISD::CTPOP:
    return LowerCTPOP(Op, DAG);
  case ISD::FCOPYSIGN:
    return LowerFCOPYSIGN(Op, DAG);
  case ISD::AND:
    return LowerVectorAND(Op, DAG);
  case ISD::OR:
    return LowerVectorOR(Op, DAG);
  case ISD::XOR:
    return LowerXOR(Op, DAG);
  case ISD::PREFETCH:
    return LowerPREFETCH(Op, DAG);
  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::FSINCOS:
    return LowerFSINCOS(Op, DAG);
  case ISD::MUL:
    return LowerMUL(Op, DAG);
  }
}

/// getFunctionAlignment - Return the Log2 alignment of this function.
unsigned AArch64TargetLowering::getFunctionAlignment(const Function *F) const {
  return 2;
}

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

#include "AArch64GenCallingConv.inc"

/// Selects the correct CCAssignFn for a given CallingConvention value.
CCAssignFn *AArch64TargetLowering::CCAssignFnForCall(CallingConv::ID CC,
                                                     bool IsVarArg) const {
  switch (CC) {
  default:
    llvm_unreachable("Unsupported calling convention.");
  case CallingConv::WebKit_JS:
    return CC_AArch64_WebKit_JS;
  case CallingConv::C:
  case CallingConv::Fast:
    if (!Subtarget->isTargetDarwin())
      return CC_AArch64_AAPCS;
    return IsVarArg ? CC_AArch64_DarwinPCS_VarArg : CC_AArch64_DarwinPCS;
  }
}

SDValue AArch64TargetLowering::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();

  // Assign locations to all of the incoming arguments.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                 *DAG.getContext());

  // At this point, Ins[].VT may already be promoted to i32. To correctly
  // handle passing i8 as i8 instead of i32 on stack, we pass in both i32 and
  // i8 to CC_AArch64_AAPCS with i32 being ValVT and i8 being LocVT.
  // Since AnalyzeFormalArguments uses Ins[].VT for both ValVT and LocVT, here
  // we use a special version of AnalyzeFormalArguments to pass in ValVT and
  // LocVT.
  unsigned NumArgs = Ins.size();
  Function::const_arg_iterator CurOrigArg = MF.getFunction()->arg_begin();
  unsigned CurArgIdx = 0;
  for (unsigned i = 0; i != NumArgs; ++i) {
    MVT ValVT = Ins[i].VT;
    std::advance(CurOrigArg, Ins[i].OrigArgIndex - CurArgIdx);
    CurArgIdx = Ins[i].OrigArgIndex;

    // Get type of the original argument.
    EVT ActualVT = getValueType(CurOrigArg->getType(), /*AllowUnknown*/ true);
    MVT ActualMVT = ActualVT.isSimple() ? ActualVT.getSimpleVT() : MVT::Other;
    // If ActualMVT is i1/i8/i16, we should set LocVT to i8/i8/i16.
    if (ActualMVT == MVT::i1 || ActualMVT == MVT::i8)
      ValVT = MVT::i8;
    else if (ActualMVT == MVT::i16)
      ValVT = MVT::i16;

    CCAssignFn *AssignFn = CCAssignFnForCall(CallConv, /*IsVarArg=*/false);
    bool Res =
        AssignFn(i, ValVT, ValVT, CCValAssign::Full, Ins[i].Flags, CCInfo);
    assert(!Res && "Call operand has unhandled type");
    (void)Res;
  }
  assert(ArgLocs.size() == Ins.size());
  SmallVector<SDValue, 16> ArgValues;
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];

    if (Ins[i].Flags.isByVal()) {
      // Byval is used for HFAs in the PCS, but the system should work in a
      // non-compliant manner for larger structs.
      EVT PtrTy = getPointerTy();
      int Size = Ins[i].Flags.getByValSize();
      unsigned NumRegs = (Size + 7) / 8;

      // FIXME: This works on big-endian for composite byvals, which are the common
      // case. It should also work for fundamental types too.
      unsigned FrameIdx =
        MFI->CreateFixedObject(8 * NumRegs, VA.getLocMemOffset(), false);
      SDValue FrameIdxN = DAG.getFrameIndex(FrameIdx, PtrTy);
      InVals.push_back(FrameIdxN);

      continue;
    }
    
    if (VA.isRegLoc()) {
      // Arguments stored in registers.
      EVT RegVT = VA.getLocVT();

      SDValue ArgValue;
      const TargetRegisterClass *RC;

      if (RegVT == MVT::i32)
        RC = &AArch64::GPR32RegClass;
      else if (RegVT == MVT::i64)
        RC = &AArch64::GPR64RegClass;
      else if (RegVT == MVT::f16)
        RC = &AArch64::FPR16RegClass;
      else if (RegVT == MVT::f32)
        RC = &AArch64::FPR32RegClass;
      else if (RegVT == MVT::f64 || RegVT.is64BitVector())
        RC = &AArch64::FPR64RegClass;
      else if (RegVT == MVT::f128 || RegVT.is128BitVector())
        RC = &AArch64::FPR128RegClass;
      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, 16 or 32-bit value, it is really passed promoted
      // to 64 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::AExt:
      case CCValAssign::SExt:
      case CCValAssign::ZExt:
        // SelectionDAGBuilder will insert appropriate AssertZExt & AssertSExt
        // nodes after our lowering.
        assert(RegVT == Ins[i].VT && "incorrect register location selected");
        break;
      }

      InVals.push_back(ArgValue);

    } else { // VA.isRegLoc()
      assert(VA.isMemLoc() && "CCValAssign is neither reg nor mem");
      unsigned ArgOffset = VA.getLocMemOffset();
      unsigned ArgSize = VA.getValVT().getSizeInBits() / 8;

      uint32_t BEAlign = 0;
      if (ArgSize < 8 && !Subtarget->isLittleEndian())
        BEAlign = 8 - ArgSize;

      int FI = MFI->CreateFixedObject(ArgSize, ArgOffset + BEAlign, true);

      // Create load nodes to retrieve arguments from the stack.
      SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
      SDValue ArgValue;

      // For NON_EXTLOAD, generic code in getLoad assert(ValVT == MemVT)
      ISD::LoadExtType ExtType = ISD::NON_EXTLOAD;
      MVT MemVT = VA.getValVT();

      switch (VA.getLocInfo()) {
      default:
        break;
      case CCValAssign::BCvt:
        MemVT = VA.getLocVT();
        break;
      case CCValAssign::SExt:
        ExtType = ISD::SEXTLOAD;
        break;
      case CCValAssign::ZExt:
        ExtType = ISD::ZEXTLOAD;
        break;
      case CCValAssign::AExt:
        ExtType = ISD::EXTLOAD;
        break;
      }

      ArgValue = DAG.getExtLoad(ExtType, DL, VA.getLocVT(), Chain, FIN,
                                MachinePointerInfo::getFixedStack(FI),
                                MemVT, false, false, false, 0);

      InVals.push_back(ArgValue);
    }
  }

  // varargs
  if (isVarArg) {
    if (!Subtarget->isTargetDarwin()) {
      // The AAPCS variadic function ABI is identical to the non-variadic
      // one. As a result there may be more arguments in registers and we should
      // save them for future reference.
      saveVarArgRegisters(CCInfo, DAG, DL, Chain);
    }

    AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
    // This will point to the next argument passed via stack.
    unsigned StackOffset = CCInfo.getNextStackOffset();
    // We currently pass all varargs at 8-byte alignment.
    StackOffset = ((StackOffset + 7) & ~7);
    AFI->setVarArgsStackIndex(MFI->CreateFixedObject(4, StackOffset, true));
  }

  AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();
  unsigned StackArgSize = CCInfo.getNextStackOffset();
  bool TailCallOpt = MF.getTarget().Options.GuaranteedTailCallOpt;
  if (DoesCalleeRestoreStack(CallConv, TailCallOpt)) {
    // This is a non-standard ABI so by fiat I say we're allowed to make full
    // use of the stack area to be popped, which must be aligned to 16 bytes in
    // any case:
    StackArgSize = RoundUpToAlignment(StackArgSize, 16);

    // If we're expected to restore the stack (e.g. fastcc) then we'll be adding
    // a multiple of 16.
    FuncInfo->setArgumentStackToRestore(StackArgSize);

    // This realignment carries over to the available bytes below. Our own
    // callers will guarantee the space is free by giving an aligned value to
    // CALLSEQ_START.
  }
  // Even if we're not expected to free up the space, it's useful to know how
  // much is there while considering tail calls (because we can reuse it).
  FuncInfo->setBytesInStackArgArea(StackArgSize);

  return Chain;
}

void AArch64TargetLowering::saveVarArgRegisters(CCState &CCInfo,
                                                SelectionDAG &DAG, SDLoc DL,
                                                SDValue &Chain) const {
  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();

  SmallVector<SDValue, 8> MemOps;

  static const MCPhysReg GPRArgRegs[] = { AArch64::X0, AArch64::X1, AArch64::X2,
                                          AArch64::X3, AArch64::X4, AArch64::X5,
                                          AArch64::X6, AArch64::X7 };
  static const unsigned NumGPRArgRegs = array_lengthof(GPRArgRegs);
  unsigned FirstVariadicGPR =
      CCInfo.getFirstUnallocated(GPRArgRegs, NumGPRArgRegs);

  unsigned GPRSaveSize = 8 * (NumGPRArgRegs - FirstVariadicGPR);
  int GPRIdx = 0;
  if (GPRSaveSize != 0) {
    GPRIdx = MFI->CreateStackObject(GPRSaveSize, 8, false);

    SDValue FIN = DAG.getFrameIndex(GPRIdx, getPointerTy());

    for (unsigned i = FirstVariadicGPR; i < NumGPRArgRegs; ++i) {
      unsigned VReg = MF.addLiveIn(GPRArgRegs[i], &AArch64::GPR64RegClass);
      SDValue Val = DAG.getCopyFromReg(Chain, DL, VReg, MVT::i64);
      SDValue Store =
          DAG.getStore(Val.getValue(1), DL, Val, FIN,
                       MachinePointerInfo::getStack(i * 8), false, false, 0);
      MemOps.push_back(Store);
      FIN = DAG.getNode(ISD::ADD, DL, getPointerTy(), FIN,
                        DAG.getConstant(8, getPointerTy()));
    }
  }
  FuncInfo->setVarArgsGPRIndex(GPRIdx);
  FuncInfo->setVarArgsGPRSize(GPRSaveSize);

  if (Subtarget->hasFPARMv8()) {
    static const MCPhysReg FPRArgRegs[] = {
        AArch64::Q0, AArch64::Q1, AArch64::Q2, AArch64::Q3,
        AArch64::Q4, AArch64::Q5, AArch64::Q6, AArch64::Q7};
    static const unsigned NumFPRArgRegs = array_lengthof(FPRArgRegs);
    unsigned FirstVariadicFPR =
        CCInfo.getFirstUnallocated(FPRArgRegs, NumFPRArgRegs);

    unsigned FPRSaveSize = 16 * (NumFPRArgRegs - FirstVariadicFPR);
    int FPRIdx = 0;
    if (FPRSaveSize != 0) {
      FPRIdx = MFI->CreateStackObject(FPRSaveSize, 16, false);

      SDValue FIN = DAG.getFrameIndex(FPRIdx, getPointerTy());

      for (unsigned i = FirstVariadicFPR; i < NumFPRArgRegs; ++i) {
        unsigned VReg = MF.addLiveIn(FPRArgRegs[i], &AArch64::FPR128RegClass);
        SDValue Val = DAG.getCopyFromReg(Chain, DL, VReg, MVT::f128);

        SDValue Store =
            DAG.getStore(Val.getValue(1), DL, Val, FIN,
                         MachinePointerInfo::getStack(i * 16), false, false, 0);
        MemOps.push_back(Store);
        FIN = DAG.getNode(ISD::ADD, DL, getPointerTy(), FIN,
                          DAG.getConstant(16, getPointerTy()));
      }
    }
    FuncInfo->setVarArgsFPRIndex(FPRIdx);
    FuncInfo->setVarArgsFPRSize(FPRSaveSize);
  }

  if (!MemOps.empty()) {
    Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, MemOps);
  }
}

/// LowerCallResult - Lower the result values of a call into the
/// appropriate copies out of appropriate physical registers.
SDValue AArch64TargetLowering::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 {
  CCAssignFn *RetCC = CallConv == CallingConv::WebKit_JS
                          ? RetCC_AArch64_WebKit_JS
                          : RetCC_AArch64_AAPCS;
  // Assign locations to each value returned by this call.
  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
                 *DAG.getContext());
  CCInfo.AnalyzeCallResult(Ins, RetCC);

  // 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::i64 &&
             "unexpected return calling convention register assignment");
      InVals.push_back(ThisVal);
      continue;
    }

    SDValue 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;
}

bool AArch64TargetLowering::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 {
  // For CallingConv::C this function knows whether the ABI needs
  // changing. That's not true for other conventions so they will have to opt in
  // manually.
  if (!IsTailCallConvention(CalleeCC) && CalleeCC != CallingConv::C)
    return false;

  const MachineFunction &MF = DAG.getMachineFunction();
  const Function *CallerF = MF.getFunction();
  CallingConv::ID CallerCC = CallerF->getCallingConv();
  bool CCMatch = CallerCC == CalleeCC;

  // Byval parameters hand the function a pointer directly into the stack area
  // we want to reuse during a tail call. Working around this *is* possible (see
  // X86) but less efficient and uglier in LowerCall.
  for (Function::const_arg_iterator i = CallerF->arg_begin(),
                                    e = CallerF->arg_end();
       i != e; ++i)
    if (i->hasByValAttr())
      return false;

  if (getTargetMachine().Options.GuaranteedTailCallOpt) {
    if (IsTailCallConvention(CalleeCC) && CCMatch)
      return true;
    return false;
  }

  // Externally-defined functions with weak linkage should not be
  // tail-called on AArch64 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;
  }

  // Now we search for cases where we can use a tail call without changing the
  // ABI. Sibcall is used in some places (particularly gcc) to refer to this
  // concept.

  // I want anyone implementing a new calling convention to think long and hard
  // about this assert.
  assert((!isVarArg || CalleeCC == CallingConv::C) &&
         "Unexpected variadic calling convention");

  if (isVarArg && !Outs.empty()) {
    // At least two cases here: if caller is fastcc then we can't have any
    // memory arguments (we'd be expected to clean up the stack afterwards). If
    // caller is C then we could potentially use its argument area.

    // FIXME: for now we take the most conservative of these in both cases:
    // disallow all variadic memory operands.
    SmallVector<CCValAssign, 16> ArgLocs;
    CCState CCInfo(CalleeCC, isVarArg, DAG.getMachineFunction(), ArgLocs,
                   *DAG.getContext());

    CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForCall(CalleeCC, true));
    for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i)
      if (!ArgLocs[i].isRegLoc())
        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;
    CCState CCInfo1(CalleeCC, false, DAG.getMachineFunction(), RVLocs1,
                    *DAG.getContext());
    CCInfo1.AnalyzeCallResult(Ins, CCAssignFnForCall(CalleeCC, isVarArg));

    SmallVector<CCValAssign, 16> RVLocs2;
    CCState CCInfo2(CallerCC, false, DAG.getMachineFunction(), RVLocs2,
                    *DAG.getContext());
    CCInfo2.AnalyzeCallResult(Ins, CCAssignFnForCall(CallerCC, 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;
      }
    }
  }

  // Nothing more to check if the callee is taking no arguments
  if (Outs.empty())
    return true;

  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CalleeCC, isVarArg, DAG.getMachineFunction(), ArgLocs,
                 *DAG.getContext());

  CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForCall(CalleeCC, isVarArg));

  const AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();

  // If the stack arguments for this call would fit into our own save area then
  // the call can be made tail.
  return CCInfo.getNextStackOffset() <= FuncInfo->getBytesInStackArgArea();
}

SDValue AArch64TargetLowering::addTokenForArgument(SDValue Chain,
                                                   SelectionDAG &DAG,
                                                   MachineFrameInfo *MFI,
                                                   int ClobberedFI) const {
  SmallVector<SDValue, 8> ArgChains;
  int64_t FirstByte = MFI->getObjectOffset(ClobberedFI);
  int64_t LastByte = FirstByte + MFI->getObjectSize(ClobberedFI) - 1;

  // Include the original chain at the beginning of the list. When this is
  // used by target LowerCall hooks, this helps legalize find the
  // CALLSEQ_BEGIN node.
  ArgChains.push_back(Chain);

  // Add a chain value for each stack argument corresponding
  for (SDNode::use_iterator U = DAG.getEntryNode().getNode()->use_begin(),
                            UE = DAG.getEntryNode().getNode()->use_end();
       U != UE; ++U)
    if (LoadSDNode *L = dyn_cast<LoadSDNode>(*U))
      if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(L->getBasePtr()))
        if (FI->getIndex() < 0) {
          int64_t InFirstByte = MFI->getObjectOffset(FI->getIndex());
          int64_t InLastByte = InFirstByte;
          InLastByte += MFI->getObjectSize(FI->getIndex()) - 1;

          if ((InFirstByte <= FirstByte && FirstByte <= InLastByte) ||
              (FirstByte <= InFirstByte && InFirstByte <= LastByte))
            ArgChains.push_back(SDValue(L, 1));
        }

  // Build a tokenfactor for all the chains.
  return DAG.getNode(ISD::TokenFactor, SDLoc(Chain), MVT::Other, ArgChains);
}

bool AArch64TargetLowering::DoesCalleeRestoreStack(CallingConv::ID CallCC,
                                                   bool TailCallOpt) const {
  return CallCC == CallingConv::Fast && TailCallOpt;
}

bool AArch64TargetLowering::IsTailCallConvention(CallingConv::ID CallCC) const {
  return CallCC == CallingConv::Fast;
}

/// LowerCall - Lower a call to a callseq_start + CALL + callseq_end chain,
/// and add input and output parameter nodes.
SDValue
AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
                                 SmallVectorImpl<SDValue> &InVals) const {
  SelectionDAG &DAG = CLI.DAG;
  SDLoc &DL = CLI.DL;
  SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
  SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
  SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
  SDValue Chain = CLI.Chain;
  SDValue Callee = CLI.Callee;
  bool &IsTailCall = CLI.IsTailCall;
  CallingConv::ID CallConv = CLI.CallConv;
  bool IsVarArg = CLI.IsVarArg;

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

  AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();
  bool TailCallOpt = MF.getTarget().Options.GuaranteedTailCallOpt;
  bool IsSibCall = 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");

    // A sibling call is one where we're under the usual C ABI and not planning
    // to change that but can still do a tail call:
    if (!TailCallOpt && IsTailCall)
      IsSibCall = true;

    if (IsTailCall)
      ++NumTailCalls;
  }

  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), ArgLocs,
                 *DAG.getContext());

  if (IsVarArg) {
    // Handle fixed and variable vector arguments differently.
    // Variable vector arguments always go into memory.
    unsigned NumArgs = Outs.size();

    for (unsigned i = 0; i != NumArgs; ++i) {
      MVT ArgVT = Outs[i].VT;
      ISD::ArgFlagsTy ArgFlags = Outs[i].Flags;
      CCAssignFn *AssignFn = CCAssignFnForCall(CallConv,
                                               /*IsVarArg=*/ !Outs[i].IsFixed);
      bool Res = AssignFn(i, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo);
      assert(!Res && "Call operand has unhandled type");
      (void)Res;
    }
  } else {
    // At this point, Outs[].VT may already be promoted to i32. To correctly
    // handle passing i8 as i8 instead of i32 on stack, we pass in both i32 and
    // i8 to CC_AArch64_AAPCS with i32 being ValVT and i8 being LocVT.
    // Since AnalyzeCallOperands uses Ins[].VT for both ValVT and LocVT, here
    // we use a special version of AnalyzeCallOperands to pass in ValVT and
    // LocVT.
    unsigned NumArgs = Outs.size();
    for (unsigned i = 0; i != NumArgs; ++i) {
      MVT ValVT = Outs[i].VT;
      // Get type of the original argument.
      EVT ActualVT = getValueType(CLI.getArgs()[Outs[i].OrigArgIndex].Ty,
                                  /*AllowUnknown*/ true);
      MVT ActualMVT = ActualVT.isSimple() ? ActualVT.getSimpleVT() : ValVT;
      ISD::ArgFlagsTy ArgFlags = Outs[i].Flags;
      // If ActualMVT is i1/i8/i16, we should set LocVT to i8/i8/i16.
      if (ActualMVT == MVT::i1 || ActualMVT == MVT::i8)
        ValVT = MVT::i8;
      else if (ActualMVT == MVT::i16)
        ValVT = MVT::i16;

      CCAssignFn *AssignFn = CCAssignFnForCall(CallConv, /*IsVarArg=*/false);
      bool Res = AssignFn(i, ValVT, ValVT, CCValAssign::Full, ArgFlags, CCInfo);
      assert(!Res && "Call operand has unhandled type");
      (void)Res;
    }
  }

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

  if (IsSibCall) {
    // Since we're not changing the ABI to make this a tail call, the memory
    // operands are already available in the caller's incoming argument space.
    NumBytes = 0;
  }

  // FPDiff is the byte offset of the call's argument area from the callee's.
  // Stores to callee stack arguments will be placed in FixedStackSlots offset
  // by this amount for a tail call. In a sibling call it must be 0 because the
  // caller will deallocate the entire stack and the callee still expects its
  // arguments to begin at SP+0. Completely unused for non-tail calls.
  int FPDiff = 0;

  if (IsTailCall && !IsSibCall) {
    unsigned NumReusableBytes = FuncInfo->getBytesInStackArgArea();

    // Since callee will pop argument stack as a tail call, we must keep the
    // popped size 16-byte aligned.
    NumBytes = RoundUpToAlignment(NumBytes, 16);

    // FPDiff will be negative if this tail call requires more space than we
    // would automatically have in our incoming argument space. Positive if we
    // can actually shrink the stack.
    FPDiff = NumReusableBytes - NumBytes;

    // The stack pointer must be 16-byte aligned at all times it's used for a
    // memory operation, which in practice means at *all* times and in
    // particular across call boundaries. Therefore our own arguments started at
    // a 16-byte aligned SP and the delta applied for the tail call should
    // satisfy the same constraint.
    assert(FPDiff % 16 == 0 && "unaligned stack on tail call");
  }

  // 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, AArch64::SP, getPointerTy());

  SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
  SmallVector<SDValue, 8> MemOpChains;

  // Walk the register/memloc assignments, inserting copies/loads.
  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;

    // 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:
      if (Outs[realArgIdx].ArgVT == MVT::i1) {
        // AAPCS requires i1 to be zero-extended to 8-bits by the caller.
        Arg = DAG.getNode(ISD::TRUNCATE, DL, MVT::i1, Arg);
        Arg = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i8, Arg);
      }
      Arg = DAG.getNode(ISD::ANY_EXTEND, DL, VA.getLocVT(), Arg);
      break;
    case CCValAssign::BCvt:
      Arg = DAG.getNode(ISD::BITCAST, DL, VA.getLocVT(), Arg);
      break;
    case CCValAssign::FPExt:
      Arg = DAG.getNode(ISD::FP_EXTEND, DL, VA.getLocVT(), Arg);
      break;
    }

    if (VA.isRegLoc()) {
      if (realArgIdx == 0 && Flags.isReturned() && Outs[0].VT == MVT::i64) {
        assert(VA.getLocVT() == MVT::i64 &&
               "unexpected calling convention register assignment");
        assert(!Ins.empty() && Ins[0].VT == MVT::i64 &&
               "unexpected use of 'returned'");
        IsThisReturn = true;
      }
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
    } else {
      assert(VA.isMemLoc());

      SDValue DstAddr;
      MachinePointerInfo DstInfo;

      // FIXME: This works on big-endian for composite byvals, which are the
      // common case. It should also work for fundamental types too.
      uint32_t BEAlign = 0;
      unsigned OpSize = Flags.isByVal() ? Flags.getByValSize() * 8
                                        : VA.getValVT().getSizeInBits();
      OpSize = (OpSize + 7) / 8;
      if (!Subtarget->isLittleEndian() && !Flags.isByVal()) {
        if (OpSize < 8)
          BEAlign = 8 - OpSize;
      }
      unsigned LocMemOffset = VA.getLocMemOffset();
      int32_t Offset = LocMemOffset + BEAlign;
      SDValue PtrOff = DAG.getIntPtrConstant(Offset);
      PtrOff = DAG.getNode(ISD::ADD, DL, getPointerTy(), StackPtr, PtrOff);

      if (IsTailCall) {
        Offset = Offset + FPDiff;
        int FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset, true);

        DstAddr = DAG.getFrameIndex(FI, getPointerTy());
        DstInfo = MachinePointerInfo::getFixedStack(FI);

        // Make sure any stack arguments overlapping with where we're storing
        // are loaded before this eventual operation. Otherwise they'll be
        // clobbered.
        Chain = addTokenForArgument(Chain, DAG, MF.getFrameInfo(), FI);
      } else {
        SDValue PtrOff = DAG.getIntPtrConstant(Offset);

        DstAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), StackPtr, PtrOff);
        DstInfo = MachinePointerInfo::getStack(LocMemOffset);
      }

      if (Outs[i].Flags.isByVal()) {
        SDValue SizeNode =
            DAG.getConstant(Outs[i].Flags.getByValSize(), MVT::i64);
        SDValue Cpy = DAG.getMemcpy(
            Chain, DL, DstAddr, Arg, SizeNode, Outs[i].Flags.getByValAlign(),
            /*isVol = */ false,
            /*AlwaysInline = */ false, DstInfo, MachinePointerInfo());

        MemOpChains.push_back(Cpy);
      } else {
        // Since we pass i1/i8/i16 as i1/i8/i16 on stack and Arg is already
        // promoted to a legal register type i32, we should truncate Arg back to
        // i1/i8/i16.
        if (VA.getValVT() == MVT::i1 || VA.getValVT() == MVT::i8 ||
            VA.getValVT() == MVT::i16)
          Arg = DAG.getNode(ISD::TRUNCATE, DL, VA.getValVT(), Arg);

        SDValue Store =
            DAG.getStore(Chain, DL, Arg, DstAddr, DstInfo, false, false, 0);
        MemOpChains.push_back(Store);
      }
    }
  }

  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;
  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);
  }

  // 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.
  if (getTargetMachine().getCodeModel() == CodeModel::Large &&
      Subtarget->isTargetMachO()) {
    if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
      const GlobalValue *GV = G->getGlobal();
      bool InternalLinkage = GV->hasInternalLinkage();
      if (InternalLinkage)
        Callee = DAG.getTargetGlobalAddress(GV, DL, getPointerTy(), 0, 0);
      else {
        Callee = DAG.getTargetGlobalAddress(GV, DL, getPointerTy(), 0,
                                            AArch64II::MO_GOT);
        Callee = DAG.getNode(AArch64ISD::LOADgot, DL, getPointerTy(), Callee);
      }
    } else if (ExternalSymbolSDNode *S =
                   dyn_cast<ExternalSymbolSDNode>(Callee)) {
      const char *Sym = S->getSymbol();
      Callee =
          DAG.getTargetExternalSymbol(Sym, getPointerTy(), AArch64II::MO_GOT);
      Callee = DAG.getNode(AArch64ISD::LOADgot, DL, getPointerTy(), Callee);
    }
  } else if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
    const GlobalValue *GV = G->getGlobal();
    Callee = DAG.getTargetGlobalAddress(GV, DL, getPointerTy(), 0, 0);
  } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
    const char *Sym = S->getSymbol();
    Callee = DAG.getTargetExternalSymbol(Sym, getPointerTy(), 0);
  }

  // We don't usually want to end the call-sequence here because we would tidy
  // the frame up *after* the call, however in the ABI-changing tail-call case
  // we've carefully laid out the parameters so that when sp is reset they'll be
  // in the correct location.
  if (IsTailCall && !IsSibCall) {
    Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
                               DAG.getIntPtrConstant(0, true), InFlag, DL);
    InFlag = Chain.getValue(1);
  }

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

  if (IsTailCall) {
    // Each tail call may have to adjust the stack by a different amount, so
    // this information must travel along with the operation for eventual
    // consumption by emitEpilogue.
    Ops.push_back(DAG.getTargetConstant(FPDiff, MVT::i32));
  }

  // 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.
  const uint32_t *Mask;
  const TargetRegisterInfo *TRI =
      getTargetMachine().getSubtargetImpl()->getRegisterInfo();
  const AArch64RegisterInfo *ARI =
      static_cast<const AArch64RegisterInfo *>(TRI);
  if (IsThisReturn) {
    // For 'this' returns, use the X0-preserving mask if applicable
    Mask = ARI->getThisReturnPreservedMask(CallConv);
    if (!Mask) {
      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 we're doing a tall call, use a TC_RETURN here rather than an
  // actual call instruction.
  if (IsTailCall)
    return DAG.getNode(AArch64ISD::TC_RETURN, DL, NodeTys, Ops);

  // Returns a chain and a flag for retval copy to use.
  Chain = DAG.getNode(AArch64ISD::CALL, DL, NodeTys, Ops);
  InFlag = Chain.getValue(1);

  uint64_t CalleePopBytes = DoesCalleeRestoreStack(CallConv, TailCallOpt)
                                ? RoundUpToAlignment(NumBytes, 16)
                                : 0;

  Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
                             DAG.getIntPtrConstant(CalleePopBytes, 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());
}

bool AArch64TargetLowering::CanLowerReturn(
    CallingConv::ID CallConv, MachineFunction &MF, bool isVarArg,
    const SmallVectorImpl<ISD::OutputArg> &Outs, LLVMContext &Context) const {
  CCAssignFn *RetCC = CallConv == CallingConv::WebKit_JS
                          ? RetCC_AArch64_WebKit_JS
                          : RetCC_AArch64_AAPCS;
  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCInfo(CallConv, isVarArg, MF, RVLocs, Context);
  return CCInfo.CheckReturn(Outs, RetCC);
}

SDValue
AArch64TargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
                                   bool isVarArg,
                                   const SmallVectorImpl<ISD::OutputArg> &Outs,
                                   const SmallVectorImpl<SDValue> &OutVals,
                                   SDLoc DL, SelectionDAG &DAG) const {
  CCAssignFn *RetCC = CallConv == CallingConv::WebKit_JS
                          ? RetCC_AArch64_WebKit_JS
                          : RetCC_AArch64_AAPCS;
  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
                 *DAG.getContext());
  CCInfo.AnalyzeReturn(Outs, RetCC);

  // Copy the result values into the output registers.
  SDValue Flag;
  SmallVector<SDValue, 4> RetOps(1, Chain);
  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:
      if (Outs[i].ArgVT == MVT::i1) {
        // AAPCS requires i1 to be zero-extended to i8 by the producer of the
        // value. This is strictly redundant on Darwin (which uses "zeroext
        // i1"), but will be optimised out before ISel.
        Arg = DAG.getNode(ISD::TRUNCATE, DL, MVT::i1, Arg);
        Arg = DAG.getNode(ISD::ZERO_EXTEND, DL, VA.getLocVT(), Arg);
      }
      break;
    case CCValAssign::BCvt:
      Arg = DAG.getNode(ISD::BITCAST, DL, VA.getLocVT(), Arg);
      break;
    }

    Chain = DAG.getCopyToReg(Chain, DL, VA.getLocReg(), Arg, Flag);
    Flag = Chain.getValue(1);
    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
  }

  RetOps[0] = Chain; // Update chain.

  // Add the flag if we have it.
  if (Flag.getNode())
    RetOps.push_back(Flag);

  return DAG.getNode(AArch64ISD::RET_FLAG, DL, MVT::Other, RetOps);
}

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

SDValue AArch64TargetLowering::LowerGlobalAddress(SDValue Op,
                                                  SelectionDAG &DAG) const {
  EVT PtrVT = getPointerTy();
  SDLoc DL(Op);
  const GlobalAddressSDNode *GN = cast<GlobalAddressSDNode>(Op);
  const GlobalValue *GV = GN->getGlobal();
  unsigned char OpFlags =
      Subtarget->ClassifyGlobalReference(GV, getTargetMachine());

  assert(cast<GlobalAddressSDNode>(Op)->getOffset() == 0 &&
         "unexpected offset in global node");

  // This also catched the large code model case for Darwin.
  if ((OpFlags & AArch64II::MO_GOT) != 0) {
    SDValue GotAddr = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, OpFlags);
    // FIXME: Once remat is capable of dealing with instructions with register
    // operands, expand this into two nodes instead of using a wrapper node.
    return DAG.getNode(AArch64ISD::LOADgot, DL, PtrVT, GotAddr);
  }

  if ((OpFlags & AArch64II::MO_CONSTPOOL) != 0) {
    assert(getTargetMachine().getCodeModel() == CodeModel::Small &&
           "use of MO_CONSTPOOL only supported on small model");
    SDValue Hi = DAG.getTargetConstantPool(GV, PtrVT, 0, 0, AArch64II::MO_PAGE);
    SDValue ADRP = DAG.getNode(AArch64ISD::ADRP, DL, PtrVT, Hi);
    unsigned char LoFlags = AArch64II::MO_PAGEOFF | AArch64II::MO_NC;
    SDValue Lo = DAG.getTargetConstantPool(GV, PtrVT, 0, 0, LoFlags);
    SDValue PoolAddr = DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, ADRP, Lo);
    SDValue GlobalAddr = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), PoolAddr,
                                     MachinePointerInfo::getConstantPool(),
                                     /*isVolatile=*/ false,
                                     /*isNonTemporal=*/ true,
                                     /*isInvariant=*/ true, 8);
    if (GN->getOffset() != 0)
      return DAG.getNode(ISD::ADD, DL, PtrVT, GlobalAddr,
                         DAG.getConstant(GN->getOffset(), PtrVT));
    return GlobalAddr;
  }

  if (getTargetMachine().getCodeModel() == CodeModel::Large) {
    const unsigned char MO_NC = AArch64II::MO_NC;
    return DAG.getNode(
        AArch64ISD::WrapperLarge, DL, PtrVT,
        DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_G3),
        DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_G2 | MO_NC),
        DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_G1 | MO_NC),
        DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_G0 | MO_NC));
  } else {
    // Use ADRP/ADD or ADRP/LDR for everything else: the small model on ELF and
    // the only correct model on Darwin.
    SDValue Hi = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
                                            OpFlags | AArch64II::MO_PAGE);
    unsigned char LoFlags = OpFlags | AArch64II::MO_PAGEOFF | AArch64II::MO_NC;
    SDValue Lo = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, LoFlags);

    SDValue ADRP = DAG.getNode(AArch64ISD::ADRP, DL, PtrVT, Hi);
    return DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, ADRP, Lo);
  }
}

/// \brief Convert a TLS address reference into the correct sequence of loads
/// and calls to compute the variable's address (for Darwin, currently) and
/// return an SDValue containing the final node.

/// Darwin only has one TLS scheme which must be capable of dealing with the
/// fully general situation, in the worst case. This means:
///     + "extern __thread" declaration.
///     + Defined in a possibly unknown dynamic library.
///
/// The general system is that each __thread variable has a [3 x i64] descriptor
/// which contains information used by the runtime to calculate the address. The
/// only part of this the compiler needs to know about is the first xword, which
/// contains a function pointer that must be called with the address of the
/// entire descriptor in "x0".
///
/// Since this descriptor may be in a different unit, in general even the
/// descriptor must be accessed via an indirect load. The "ideal" code sequence
/// is:
///     adrp x0, _var@TLVPPAGE
///     ldr x0, [x0, _var@TLVPPAGEOFF]   ; x0 now contains address of descriptor
///     ldr x1, [x0]                     ; x1 contains 1st entry of descriptor,
///                                      ; the function pointer
///     blr x1                           ; Uses descriptor address in x0
///     ; Address of _var is now in x0.
///
/// If the address of _var's descriptor *is* known to the linker, then it can
/// change the first "ldr" instruction to an appropriate "add x0, x0, #imm" for
/// a slight efficiency gain.
SDValue
AArch64TargetLowering::LowerDarwinGlobalTLSAddress(SDValue Op,
                                                   SelectionDAG &DAG) const {
  assert(Subtarget->isTargetDarwin() && "TLS only supported on Darwin");

  SDLoc DL(Op);
  MVT PtrVT = getPointerTy();
  const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();

  SDValue TLVPAddr =
      DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_TLS);
  SDValue DescAddr = DAG.getNode(AArch64ISD::LOADgot, DL, PtrVT, TLVPAddr);

  // The first entry in the descriptor is a function pointer that we must call
  // to obtain the address of the variable.
  SDValue Chain = DAG.getEntryNode();
  SDValue FuncTLVGet =
      DAG.getLoad(MVT::i64, DL, Chain, DescAddr, MachinePointerInfo::getGOT(),
                  false, true, true, 8);
  Chain = FuncTLVGet.getValue(1);

  MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
  MFI->setAdjustsStack(true);

  // TLS calls preserve all registers except those that absolutely must be
  // trashed: X0 (it takes an argument), LR (it's a call) and NZCV (let's not be
  // silly).
  const TargetRegisterInfo *TRI =
      getTargetMachine().getSubtargetImpl()->getRegisterInfo();
  const AArch64RegisterInfo *ARI =
      static_cast<const AArch64RegisterInfo *>(TRI);
  const uint32_t *Mask = ARI->getTLSCallPreservedMask();

  // Finally, we can make the call. This is just a degenerate version of a
  // normal AArch64 call node: x0 takes the address of the descriptor, and
  // returns the address of the variable in this thread.
  Chain = DAG.getCopyToReg(Chain, DL, AArch64::X0, DescAddr, SDValue());
  Chain =
      DAG.getNode(AArch64ISD::CALL, DL, DAG.getVTList(MVT::Other, MVT::Glue),
                  Chain, FuncTLVGet, DAG.getRegister(AArch64::X0, MVT::i64),
                  DAG.getRegisterMask(Mask), Chain.getValue(1));
  return DAG.getCopyFromReg(Chain, DL, AArch64::X0, PtrVT, Chain.getValue(1));
}

/// When accessing thread-local variables under either the general-dynamic or
/// local-dynamic system, we make a "TLS-descriptor" call. The variable will
/// have a descriptor, accessible via a PC-relative ADRP, and whose first entry
/// is a function pointer to carry out the resolution. This function takes the
/// address of the descriptor in X0 and returns the TPIDR_EL0 offset in X0. All
/// other registers (except LR, NZCV) are preserved.
///
/// Thus, the ideal call sequence on AArch64 is:
///
///     adrp x0, :tlsdesc:thread_var
///     ldr x8, [x0, :tlsdesc_lo12:thread_var]
///     add x0, x0, :tlsdesc_lo12:thread_var
///     .tlsdesccall thread_var
///     blr x8
///     (TPIDR_EL0 offset now in x0).
///
/// The ".tlsdesccall" directive instructs the assembler to insert a particular
/// relocation to help the linker relax this sequence if it turns out to be too
/// conservative.
///
/// FIXME: we currently produce an extra, duplicated, ADRP instruction, but this
/// is harmless.
SDValue AArch64TargetLowering::LowerELFTLSDescCall(SDValue SymAddr,
                                                   SDValue DescAddr, SDLoc DL,
                                                   SelectionDAG &DAG) const {
  EVT PtrVT = getPointerTy();

  // The function we need to call is simply the first entry in the GOT for this
  // descriptor, load it in preparation.
  SDValue Func = DAG.getNode(AArch64ISD::LOADgot, DL, PtrVT, SymAddr);

  // TLS calls preserve all registers except those that absolutely must be
  // trashed: X0 (it takes an argument), LR (it's a call) and NZCV (let's not be
  // silly).
  const TargetRegisterInfo *TRI =
      getTargetMachine().getSubtargetImpl()->getRegisterInfo();
  const AArch64RegisterInfo *ARI =
      static_cast<const AArch64RegisterInfo *>(TRI);
  const uint32_t *Mask = ARI->getTLSCallPreservedMask();

  // The function takes only one argument: the address of the descriptor itself
  // in X0.
  SDValue Glue, Chain;
  Chain = DAG.getCopyToReg(DAG.getEntryNode(), DL, AArch64::X0, DescAddr, Glue);
  Glue = Chain.getValue(1);

  // We're now ready to populate the argument list, as with a normal call:
  SmallVector<SDValue, 6> Ops;
  Ops.push_back(Chain);
  Ops.push_back(Func);
  Ops.push_back(SymAddr);
  Ops.push_back(DAG.getRegister(AArch64::X0, PtrVT));
  Ops.push_back(DAG.getRegisterMask(Mask));
  Ops.push_back(Glue);

  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
  Chain = DAG.getNode(AArch64ISD::TLSDESC_CALL, DL, NodeTys, Ops);
  Glue = Chain.getValue(1);

  return DAG.getCopyFromReg(Chain, DL, AArch64::X0, PtrVT, Glue);
}

SDValue
AArch64TargetLowering::LowerELFGlobalTLSAddress(SDValue Op,
                                                SelectionDAG &DAG) const {
  assert(Subtarget->isTargetELF() && "This function expects an ELF target");
  assert(getTargetMachine().getCodeModel() == CodeModel::Small &&
         "ELF TLS only supported in small memory model");
  const GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);

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

  SDValue TPOff;
  EVT PtrVT = getPointerTy();
  SDLoc DL(Op);
  const GlobalValue *GV = GA->getGlobal();

  SDValue ThreadBase = DAG.getNode(AArch64ISD::THREAD_POINTER, DL, PtrVT);

  if (Model == TLSModel::LocalExec) {
    SDValue HiVar = DAG.getTargetGlobalAddress(
        GV, DL, PtrVT, 0, AArch64II::MO_TLS | AArch64II::MO_G1);
    SDValue LoVar = DAG.getTargetGlobalAddress(
        GV, DL, PtrVT, 0,
        AArch64II::MO_TLS | AArch64II::MO_G0 | AArch64II::MO_NC);

    TPOff = SDValue(DAG.getMachineNode(AArch64::MOVZXi, DL, PtrVT, HiVar,
                                       DAG.getTargetConstant(16, MVT::i32)),
                    0);
    TPOff = SDValue(DAG.getMachineNode(AArch64::MOVKXi, DL, PtrVT, TPOff, LoVar,
                                       DAG.getTargetConstant(0, MVT::i32)),
                    0);
  } else if (Model == TLSModel::InitialExec) {
    TPOff = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_TLS);
    TPOff = DAG.getNode(AArch64ISD::LOADgot, DL, PtrVT, TPOff);
  } else if (Model == TLSModel::LocalDynamic) {
    // Local-dynamic accesses proceed in two phases. A general-dynamic TLS
    // descriptor call against the special symbol _TLS_MODULE_BASE_ to calculate
    // the beginning of the module's TLS region, followed by a DTPREL offset
    // calculation.

    // These accesses will need deduplicating if there's more than one.
    AArch64FunctionInfo *MFI =
        DAG.getMachineFunction().getInfo<AArch64FunctionInfo>();
    MFI->incNumLocalDynamicTLSAccesses();

    // Accesses used in this sequence go via the TLS descriptor which lives in
    // the GOT. Prepare an address we can use to handle this.
    SDValue HiDesc = DAG.getTargetExternalSymbol(
        "_TLS_MODULE_BASE_", PtrVT, AArch64II::MO_TLS | AArch64II::MO_PAGE);
    SDValue LoDesc = DAG.getTargetExternalSymbol(
        "_TLS_MODULE_BASE_", PtrVT,
        AArch64II::MO_TLS | AArch64II::MO_PAGEOFF | AArch64II::MO_NC);

    // First argument to the descriptor call is the address of the descriptor
    // itself.
    SDValue DescAddr = DAG.getNode(AArch64ISD::ADRP, DL, PtrVT, HiDesc);
    DescAddr = DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, DescAddr, LoDesc);

    // The call needs a relocation too for linker relaxation. It doesn't make
    // sense to call it MO_PAGE or MO_PAGEOFF though so we need another copy of
    // the address.
    SDValue SymAddr = DAG.getTargetExternalSymbol("_TLS_MODULE_BASE_", PtrVT,
                                                  AArch64II::MO_TLS);

    // Now we can calculate the offset from TPIDR_EL0 to this module's
    // thread-local area.
    TPOff = LowerELFTLSDescCall(SymAddr, DescAddr, DL, DAG);

    // Now use :dtprel_whatever: operations to calculate this variable's offset
    // in its thread-storage area.
    SDValue HiVar = DAG.getTargetGlobalAddress(
        GV, DL, MVT::i64, 0, AArch64II::MO_TLS | AArch64II::MO_G1);
    SDValue LoVar = DAG.getTargetGlobalAddress(
        GV, DL, MVT::i64, 0,
        AArch64II::MO_TLS | AArch64II::MO_G0 | AArch64II::MO_NC);

    SDValue DTPOff =
        SDValue(DAG.getMachineNode(AArch64::MOVZXi, DL, PtrVT, HiVar,
                                   DAG.getTargetConstant(16, MVT::i32)),
                0);
    DTPOff =
        SDValue(DAG.getMachineNode(AArch64::MOVKXi, DL, PtrVT, DTPOff, LoVar,
                                   DAG.getTargetConstant(0, MVT::i32)),
                0);

    TPOff = DAG.getNode(ISD::ADD, DL, PtrVT, TPOff, DTPOff);
  } else if (Model == TLSModel::GeneralDynamic) {
    // Accesses used in this sequence go via the TLS descriptor which lives in
    // the GOT. Prepare an address we can use to handle this.
    SDValue HiDesc = DAG.getTargetGlobalAddress(
        GV, DL, PtrVT, 0, AArch64II::MO_TLS | AArch64II::MO_PAGE);
    SDValue LoDesc = DAG.getTargetGlobalAddress(
        GV, DL, PtrVT, 0,
        AArch64II::MO_TLS | AArch64II::MO_PAGEOFF | AArch64II::MO_NC);

    // First argument to the descriptor call is the address of the descriptor
    // itself.
    SDValue DescAddr = DAG.getNode(AArch64ISD::ADRP, DL, PtrVT, HiDesc);
    DescAddr = DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, DescAddr, LoDesc);

    // The call needs a relocation too for linker relaxation. It doesn't make
    // sense to call it MO_PAGE or MO_PAGEOFF though so we need another copy of
    // the address.
    SDValue SymAddr =
        DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_TLS);

    // Finally we can make a call to calculate the offset from tpidr_el0.
    TPOff = LowerELFTLSDescCall(SymAddr, DescAddr, DL, DAG);
  } else
    llvm_unreachable("Unsupported ELF TLS access model");

  return DAG.getNode(ISD::ADD, DL, PtrVT, ThreadBase, TPOff);
}

SDValue AArch64TargetLowering::LowerGlobalTLSAddress(SDValue Op,
                                                     SelectionDAG &DAG) const {
  if (Subtarget->isTargetDarwin())
    return LowerDarwinGlobalTLSAddress(Op, DAG);
  else if (Subtarget->isTargetELF())
    return LowerELFGlobalTLSAddress(Op, DAG);

  llvm_unreachable("Unexpected platform trying to use TLS");
}
SDValue AArch64TargetLowering::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);

  // Handle f128 first, since lowering it will result in comparing the return
  // value of a libcall against zero, which is just what the rest of LowerBR_CC
  // is expecting to deal with.
  if (LHS.getValueType() == MVT::f128) {
    softenSetCCOperands(DAG, MVT::f128, LHS, RHS, CC, dl);

    // If softenSetCCOperands returned a scalar, we need to compare the result
    // against zero to select between true and false values.
    if (!RHS.getNode()) {
      RHS = DAG.getConstant(0, LHS.getValueType());
      CC = ISD::SETNE;
    }
  }

  // Optimize {s|u}{add|sub|mul}.with.overflow feeding into a branch
  // instruction.
  unsigned Opc = LHS.getOpcode();
  if (LHS.getResNo() == 1 && isa<ConstantSDNode>(RHS) &&
      cast<ConstantSDNode>(RHS)->isOne() &&
      (Opc == ISD::SADDO || Opc == ISD::UADDO || Opc == ISD::SSUBO ||
       Opc == ISD::USUBO || Opc == ISD::SMULO || Opc == ISD::UMULO)) {
    assert((CC == ISD::SETEQ || CC == ISD::SETNE) &&
           "Unexpected condition code.");
    // Only lower legal XALUO ops.
    if (!DAG.getTargetLoweringInfo().isTypeLegal(LHS->getValueType(0)))
      return SDValue();

    // The actual operation with overflow check.
    AArch64CC::CondCode OFCC;
    SDValue Value, Overflow;
    std::tie(Value, Overflow) = getAArch64XALUOOp(OFCC, LHS.getValue(0), DAG);

    if (CC == ISD::SETNE)
      OFCC = getInvertedCondCode(OFCC);
    SDValue CCVal = DAG.getConstant(OFCC, MVT::i32);

    return DAG.getNode(AArch64ISD::BRCOND, SDLoc(LHS), MVT::Other, Chain, Dest,
                       CCVal, Overflow);
  }

  if (LHS.getValueType().isInteger()) {
    assert((LHS.getValueType() == RHS.getValueType()) &&
           (LHS.getValueType() == MVT::i32 || LHS.getValueType() == MVT::i64));

    // If the RHS of the comparison is zero, we can potentially fold this
    // to a specialized branch.
    const ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS);
    if (RHSC && RHSC->getZExtValue() == 0) {
      if (CC == ISD::SETEQ) {
        // See if we can use a TBZ to fold in an AND as well.
        // TBZ has a smaller branch displacement than CBZ.  If the offset is
        // out of bounds, a late MI-layer pass rewrites branches.
        // 403.gcc is an example that hits this case.
        if (LHS.getOpcode() == ISD::AND &&
            isa<ConstantSDNode>(LHS.getOperand(1)) &&
            isPowerOf2_64(LHS.getConstantOperandVal(1))) {
          SDValue Test = LHS.getOperand(0);
          uint64_t Mask = LHS.getConstantOperandVal(1);
          return DAG.getNode(AArch64ISD::TBZ, dl, MVT::Other, Chain, Test,
                             DAG.getConstant(Log2_64(Mask), MVT::i64), Dest);
        }

        return DAG.getNode(AArch64ISD::CBZ, dl, MVT::Other, Chain, LHS, Dest);
      } else if (CC == ISD::SETNE) {
        // See if we can use a TBZ to fold in an AND as well.
        // TBZ has a smaller branch displacement than CBZ.  If the offset is
        // out of bounds, a late MI-layer pass rewrites branches.
        // 403.gcc is an example that hits this case.
        if (LHS.getOpcode() == ISD::AND &&
            isa<ConstantSDNode>(LHS.getOperand(1)) &&
            isPowerOf2_64(LHS.getConstantOperandVal(1))) {
          SDValue Test = LHS.getOperand(0);
          uint64_t Mask = LHS.getConstantOperandVal(1);
          return DAG.getNode(AArch64ISD::TBNZ, dl, MVT::Other, Chain, Test,
                             DAG.getConstant(Log2_64(Mask), MVT::i64), Dest);
        }

        return DAG.getNode(AArch64ISD::CBNZ, dl, MVT::Other, Chain, LHS, Dest);
      } else if (CC == ISD::SETLT && LHS.getOpcode() != ISD::AND) {
        // Don't combine AND since emitComparison converts the AND to an ANDS
        // (a.k.a. TST) and the test in the test bit and branch instruction
        // becomes redundant.  This would also increase register pressure.
        uint64_t Mask = LHS.getValueType().getSizeInBits() - 1;
        return DAG.getNode(AArch64ISD::TBNZ, dl, MVT::Other, Chain, LHS,
                           DAG.getConstant(Mask, MVT::i64), Dest);
      }
    }
    if (RHSC && RHSC->getSExtValue() == -1 && CC == ISD::SETGT &&
        LHS.getOpcode() != ISD::AND) {
      // Don't combine AND since emitComparison converts the AND to an ANDS
      // (a.k.a. TST) and the test in the test bit and branch instruction
      // becomes redundant.  This would also increase register pressure.
      uint64_t Mask = LHS.getValueType().getSizeInBits() - 1;
      return DAG.getNode(AArch64ISD::TBZ, dl, MVT::Other, Chain, LHS,
                         DAG.getConstant(Mask, MVT::i64), Dest);
    }

    SDValue CCVal;
    SDValue Cmp = getAArch64Cmp(LHS, RHS, CC, CCVal, DAG, dl);
    return DAG.getNode(AArch64ISD::BRCOND, dl, MVT::Other, Chain, Dest, CCVal,
                       Cmp);
  }

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

  // Unfortunately, the mapping of LLVM FP CC's onto AArch64 CC's isn't totally
  // clean.  Some of them require two branches to implement.
  SDValue Cmp = emitComparison(LHS, RHS, CC, dl, DAG);
  AArch64CC::CondCode CC1, CC2;
  changeFPCCToAArch64CC(CC, CC1, CC2);
  SDValue CC1Val = DAG.getConstant(CC1, MVT::i32);
  SDValue BR1 =
      DAG.getNode(AArch64ISD::BRCOND, dl, MVT::Other, Chain, Dest, CC1Val, Cmp);
  if (CC2 != AArch64CC::AL) {
    SDValue CC2Val = DAG.getConstant(CC2, MVT::i32);
    return DAG.getNode(AArch64ISD::BRCOND, dl, MVT::Other, BR1, Dest, CC2Val,
                       Cmp);
  }

  return BR1;
}

SDValue AArch64TargetLowering::LowerFCOPYSIGN(SDValue Op,
                                              SelectionDAG &DAG) const {
  EVT VT = Op.getValueType();
  SDLoc DL(Op);

  SDValue In1 = Op.getOperand(0);
  SDValue In2 = Op.getOperand(1);
  EVT SrcVT = In2.getValueType();
  if (SrcVT != VT) {
    if (SrcVT == MVT::f32 && VT == MVT::f64)
      In2 = DAG.getNode(ISD::FP_EXTEND, DL, VT, In2);
    else if (SrcVT == MVT::f64 && VT == MVT::f32)
      In2 = DAG.getNode(ISD::FP_ROUND, DL, VT, In2, DAG.getIntPtrConstant(0));
    else
      // FIXME: Src type is different, bail out for now. Can VT really be a
      // vector type?
      return SDValue();
  }

  EVT VecVT;
  EVT EltVT;
  SDValue EltMask, VecVal1, VecVal2;
  if (VT == MVT::f32 || VT == MVT::v2f32 || VT == MVT::v4f32) {
    EltVT = MVT::i32;
    VecVT = MVT::v4i32;
    EltMask = DAG.getConstant(0x80000000ULL, EltVT);

    if (!VT.isVector()) {
      VecVal1 = DAG.getTargetInsertSubreg(AArch64::ssub, DL, VecVT,
                                          DAG.getUNDEF(VecVT), In1);
      VecVal2 = DAG.getTargetInsertSubreg(AArch64::ssub, DL, VecVT,
                                          DAG.getUNDEF(VecVT), In2);
    } else {
      VecVal1 = DAG.getNode(ISD::BITCAST, DL, VecVT, In1);
      VecVal2 = DAG.getNode(ISD::BITCAST, DL, VecVT, In2);
    }
  } else if (VT == MVT::f64 || VT == MVT::v2f64) {
    EltVT = MVT::i64;
    VecVT = MVT::v2i64;

    // We want to materialize a mask with the the high bit set, but the AdvSIMD
    // immediate moves cannot materialize that in a single instruction for
    // 64-bit elements. Instead, materialize zero and then negate it.
    EltMask = DAG.getConstant(0, EltVT);

    if (!VT.isVector()) {
      VecVal1 = DAG.getTargetInsertSubreg(AArch64::dsub, DL, VecVT,
                                          DAG.getUNDEF(VecVT), In1);
      VecVal2 = DAG.getTargetInsertSubreg(AArch64::dsub, DL, VecVT,
                                          DAG.getUNDEF(VecVT), In2);
    } else {
      VecVal1 = DAG.getNode(ISD::BITCAST, DL, VecVT, In1);
      VecVal2 = DAG.getNode(ISD::BITCAST, DL, VecVT, In2);
    }
  } else {
    llvm_unreachable("Invalid type for copysign!");
  }

  std::vector<SDValue> BuildVectorOps;
  for (unsigned i = 0; i < VecVT.getVectorNumElements(); ++i)
    BuildVectorOps.push_back(EltMask);

  SDValue BuildVec = DAG.getNode(ISD::BUILD_VECTOR, DL, VecVT, BuildVectorOps);

  // If we couldn't materialize the mask above, then the mask vector will be
  // the zero vector, and we need to negate it here.
  if (VT == MVT::f64 || VT == MVT::v2f64) {
    BuildVec = DAG.getNode(ISD::BITCAST, DL, MVT::v2f64, BuildVec);
    BuildVec = DAG.getNode(ISD::FNEG, DL, MVT::v2f64, BuildVec);
    BuildVec = DAG.getNode(ISD::BITCAST, DL, MVT::v2i64, BuildVec);
  }

  SDValue Sel =
      DAG.getNode(AArch64ISD::BIT, DL, VecVT, VecVal1, VecVal2, BuildVec);

  if (VT == MVT::f32)
    return DAG.getTargetExtractSubreg(AArch64::ssub, DL, VT, Sel);
  else if (VT == MVT::f64)
    return DAG.getTargetExtractSubreg(AArch64::dsub, DL, VT, Sel);
  else
    return DAG.getNode(ISD::BITCAST, DL, VT, Sel);
}

SDValue AArch64TargetLowering::LowerCTPOP(SDValue Op, SelectionDAG &DAG) const {
  if (DAG.getMachineFunction().getFunction()->getAttributes().hasAttribute(
          AttributeSet::FunctionIndex, Attribute::NoImplicitFloat))
    return SDValue();

  if (!Subtarget->hasNEON())
    return SDValue();

  // While there is no integer popcount instruction, it can
  // be more efficiently lowered to the following sequence that uses
  // AdvSIMD registers/instructions as long as the copies to/from
  // the AdvSIMD registers are cheap.
  //  FMOV    D0, X0        // copy 64-bit int to vector, high bits zero'd
  //  CNT     V0.8B, V0.8B  // 8xbyte pop-counts
  //  ADDV    B0, V0.8B     // sum 8xbyte pop-counts
  //  UMOV    X0, V0.B[0]   // copy byte result back to integer reg
  SDValue Val = Op.getOperand(0);
  SDLoc DL(Op);
  EVT VT = Op.getValueType();
  SDValue ZeroVec = DAG.getUNDEF(MVT::v8i8);

  SDValue VecVal;
  if (VT == MVT::i32) {
    VecVal = DAG.getNode(ISD::BITCAST, DL, MVT::f32, Val);
    VecVal = DAG.getTargetInsertSubreg(AArch64::ssub, DL, MVT::v8i8, ZeroVec,
                                       VecVal);
  } else {
    VecVal = DAG.getNode(ISD::BITCAST, DL, MVT::v8i8, Val);
  }

  SDValue CtPop = DAG.getNode(ISD::CTPOP, DL, MVT::v8i8, VecVal);
  SDValue UaddLV = DAG.getNode(
      ISD::INTRINSIC_WO_CHAIN, DL, MVT::i32,
      DAG.getConstant(Intrinsic::aarch64_neon_uaddlv, MVT::i32), CtPop);

  if (VT == MVT::i64)
    UaddLV = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, UaddLV);
  return UaddLV;
}

SDValue AArch64TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {

  if (Op.getValueType().isVector())
    return LowerVSETCC(Op, DAG);

  SDValue LHS = Op.getOperand(0);
  SDValue RHS = Op.getOperand(1);
  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
  SDLoc dl(Op);

  // We chose ZeroOrOneBooleanContents, so use zero and one.
  EVT VT = Op.getValueType();
  SDValue TVal = DAG.getConstant(1, VT);
  SDValue FVal = DAG.getConstant(0, VT);

  // Handle f128 first, since one possible outcome is a normal integer
  // comparison which gets picked up by the next if statement.
  if (LHS.getValueType() == MVT::f128) {
    softenSetCCOperands(DAG, MVT::f128, LHS, RHS, CC, dl);

    // If softenSetCCOperands returned a scalar, use it.
    if (!RHS.getNode()) {
      assert(LHS.getValueType() == Op.getValueType() &&
             "Unexpected setcc expansion!");
      return LHS;
    }
  }

  if (LHS.getValueType().isInteger()) {
    SDValue CCVal;
    SDValue Cmp =
        getAArch64Cmp(LHS, RHS, ISD::getSetCCInverse(CC, true), CCVal, DAG, dl);

    // Note that we inverted the condition above, so we reverse the order of
    // the true and false operands here.  This will allow the setcc to be
    // matched to a single CSINC instruction.
    return DAG.getNode(AArch64ISD::CSEL, dl, VT, FVal, TVal, CCVal, Cmp);
  }

  // Now we know we're dealing with FP values.
  assert(LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64);

  // If that fails, we'll need to perform an FCMP + CSEL sequence.  Go ahead
  // and do the comparison.
  SDValue Cmp = emitComparison(LHS, RHS, CC, dl, DAG);

  AArch64CC::CondCode CC1, CC2;
  changeFPCCToAArch64CC(CC, CC1, CC2);
  if (CC2 == AArch64CC::AL) {
    changeFPCCToAArch64CC(ISD::getSetCCInverse(CC, false), CC1, CC2);
    SDValue CC1Val = DAG.getConstant(CC1, MVT::i32);

    // Note that we inverted the condition above, so we reverse the order of
    // the true and false operands here.  This will allow the setcc to be
    // matched to a single CSINC instruction.
    return DAG.getNode(AArch64ISD::CSEL, dl, VT, FVal, TVal, CC1Val, Cmp);
  } else {
    // Unfortunately, the mapping of LLVM FP CC's onto AArch64 CC's isn't
    // totally clean.  Some of them require two CSELs to implement.  As is in
    // this case, we emit the first CSEL and then emit a second using the output
    // of the first as the RHS.  We're effectively OR'ing the two CC's together.

    // FIXME: It would be nice if we could match the two CSELs to two CSINCs.
    SDValue CC1Val = DAG.getConstant(CC1, MVT::i32);
    SDValue CS1 =
        DAG.getNode(AArch64ISD::CSEL, dl, VT, TVal, FVal, CC1Val, Cmp);

    SDValue CC2Val = DAG.getConstant(CC2, MVT::i32);
    return DAG.getNode(AArch64ISD::CSEL, dl, VT, TVal, CS1, CC2Val, Cmp);
  }
}

/// A SELECT_CC operation is really some kind of max or min if both values being
/// compared are, in some sense, equal to the results in either case. However,
/// it is permissible to compare f32 values and produce directly extended f64
/// values.
///
/// Extending the comparison operands would also be allowed, but is less likely
/// to happen in practice since their use is right here. Note that truncate
/// operations would *not* be semantically equivalent.
static bool selectCCOpsAreFMaxCompatible(SDValue Cmp, SDValue Result) {
  if (Cmp == Result)
    return true;

  ConstantFPSDNode *CCmp = dyn_cast<ConstantFPSDNode>(Cmp);
  ConstantFPSDNode *CResult = dyn_cast<ConstantFPSDNode>(Result);
  if (CCmp && CResult && Cmp.getValueType() == MVT::f32 &&
      Result.getValueType() == MVT::f64) {
    bool Lossy;
    APFloat CmpVal = CCmp->getValueAPF();
    CmpVal.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &Lossy);
    return CResult->getValueAPF().bitwiseIsEqual(CmpVal);
  }

  return Result->getOpcode() == ISD::FP_EXTEND && Result->getOperand(0) == Cmp;
}

SDValue AArch64TargetLowering::LowerSELECT(SDValue Op,
                                           SelectionDAG &DAG) const {
  SDValue CC = Op->getOperand(0);
  SDValue TVal = Op->getOperand(1);
  SDValue FVal = Op->getOperand(2);
  SDLoc DL(Op);

  unsigned Opc = CC.getOpcode();
  // Optimize {s|u}{add|sub|mul}.with.overflow feeding into a select
  // instruction.
  if (CC.getResNo() == 1 &&
      (Opc == ISD::SADDO || Opc == ISD::UADDO || Opc == ISD::SSUBO ||
       Opc == ISD::USUBO || Opc == ISD::SMULO || Opc == ISD::UMULO)) {
    // Only lower legal XALUO ops.
    if (!DAG.getTargetLoweringInfo().isTypeLegal(CC->getValueType(0)))
      return SDValue();

    AArch64CC::CondCode OFCC;
    SDValue Value, Overflow;
    std::tie(Value, Overflow) = getAArch64XALUOOp(OFCC, CC.getValue(0), DAG);
    SDValue CCVal = DAG.getConstant(OFCC, MVT::i32);

    return DAG.getNode(AArch64ISD::CSEL, DL, Op.getValueType(), TVal, FVal,
                       CCVal, Overflow);
  }

  if (CC.getOpcode() == ISD::SETCC)
    return DAG.getSelectCC(DL, CC.getOperand(0), CC.getOperand(1), TVal, FVal,
                           cast<CondCodeSDNode>(CC.getOperand(2))->get());
  else
    return DAG.getSelectCC(DL, CC, DAG.getConstant(0, CC.getValueType()), TVal,
                           FVal, ISD::SETNE);
}

SDValue AArch64TargetLowering::LowerSELECT_CC(SDValue Op,
                                              SelectionDAG &DAG) const {
  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
  SDValue LHS = Op.getOperand(0);
  SDValue RHS = Op.getOperand(1);
  SDValue TVal = Op.getOperand(2);
  SDValue FVal = Op.getOperand(3);
  SDLoc dl(Op);

  // Handle f128 first, because it will result in a comparison of some RTLIB
  // call result against zero.
  if (LHS.getValueType() == MVT::f128) {
    softenSetCCOperands(DAG, MVT::f128, LHS, RHS, CC, dl);

    // If softenSetCCOperands returned a scalar, we need to compare the result
    // against zero to select between true and false values.
    if (!RHS.getNode()) {
      RHS = DAG.getConstant(0, LHS.getValueType());
      CC = ISD::SETNE;
    }
  }

  // Handle integers first.
  if (LHS.getValueType().isInteger()) {
    assert((LHS.getValueType() == RHS.getValueType()) &&
           (LHS.getValueType() == MVT::i32 || LHS.getValueType() == MVT::i64));

    unsigned Opcode = AArch64ISD::CSEL;

    // If both the TVal and the FVal are constants, see if we can swap them in
    // order to for a CSINV or CSINC out of them.
    ConstantSDNode *CFVal = dyn_cast<ConstantSDNode>(FVal);
    ConstantSDNode *CTVal = dyn_cast<ConstantSDNode>(TVal);

    if (CTVal && CFVal && CTVal->isAllOnesValue() && CFVal->isNullValue()) {
      std::swap(TVal, FVal);
      std::swap(CTVal, CFVal);
      CC = ISD::getSetCCInverse(CC, true);
    } else if (CTVal && CFVal && CTVal->isOne() && CFVal->isNullValue()) {
      std::swap(TVal, FVal);
      std::swap(CTVal, CFVal);
      CC = ISD::getSetCCInverse(CC, true);
    } else if (TVal.getOpcode() == ISD::XOR) {
      // If TVal is a NOT we want to swap TVal and FVal so that we can match
      // with a CSINV rather than a CSEL.
      ConstantSDNode *CVal = dyn_cast<ConstantSDNode>(TVal.getOperand(1));

      if (CVal && CVal->isAllOnesValue()) {
        std::swap(TVal, FVal);
        std::swap(CTVal, CFVal);
        CC = ISD::getSetCCInverse(CC, true);
      }
    } else if (TVal.getOpcode() == ISD::SUB) {
      // If TVal is a negation (SUB from 0) we want to swap TVal and FVal so
      // that we can match with a CSNEG rather than a CSEL.
      ConstantSDNode *CVal = dyn_cast<ConstantSDNode>(TVal.getOperand(0));

      if (CVal && CVal->isNullValue()) {
        std::swap(TVal, FVal);
        std::swap(CTVal, CFVal);
        CC = ISD::getSetCCInverse(CC, true);
      }
    } else if (CTVal && CFVal) {
      const int64_t TrueVal = CTVal->getSExtValue();
      const int64_t FalseVal = CFVal->getSExtValue();
      bool Swap = false;

      // If both TVal and FVal are constants, see if FVal is the
      // inverse/negation/increment of TVal and generate a CSINV/CSNEG/CSINC
      // instead of a CSEL in that case.
      if (TrueVal == ~FalseVal) {
        Opcode = AArch64ISD::CSINV;
      } else if (TrueVal == -FalseVal) {
        Opcode = AArch64ISD::CSNEG;
      } else if (TVal.getValueType() == MVT::i32) {
        // If our operands are only 32-bit wide, make sure we use 32-bit
        // arithmetic for the check whether we can use CSINC. This ensures that
        // the addition in the check will wrap around properly in case there is
        // an overflow (which would not be the case if we do the check with
        // 64-bit arithmetic).
        const uint32_t TrueVal32 = CTVal->getZExtValue();
        const uint32_t FalseVal32 = CFVal->getZExtValue();

        if ((TrueVal32 == FalseVal32 + 1) || (TrueVal32 + 1 == FalseVal32)) {
          Opcode = AArch64ISD::CSINC;

          if (TrueVal32 > FalseVal32) {
            Swap = true;
          }
        }
        // 64-bit check whether we can use CSINC.
      } else if ((TrueVal == FalseVal + 1) || (TrueVal + 1 == FalseVal)) {
        Opcode = AArch64ISD::CSINC;

        if (TrueVal > FalseVal) {
          Swap = true;
        }
      }

      // Swap TVal and FVal if necessary.
      if (Swap) {
        std::swap(TVal, FVal);
        std::swap(CTVal, CFVal);
        CC = ISD::getSetCCInverse(CC, true);
      }

      if (Opcode != AArch64ISD::CSEL) {
        // Drop FVal since we can get its value by simply inverting/negating
        // TVal.
        FVal = TVal;
      }
    }

    SDValue CCVal;
    SDValue Cmp = getAArch64Cmp(LHS, RHS, CC, CCVal, DAG, dl);

    EVT VT = Op.getValueType();
    return DAG.getNode(Opcode, dl, VT, TVal, FVal, CCVal, Cmp);
  }

  // Now we know we're dealing with FP values.
  assert(LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64);
  assert(LHS.getValueType() == RHS.getValueType());
  EVT VT = Op.getValueType();

  // Try to match this select into a max/min operation, which have dedicated
  // opcode in the instruction set.
  // FIXME: This is not correct in the presence of NaNs, so we only enable this
  // in no-NaNs mode.
  if (getTargetMachine().Options.NoNaNsFPMath) {
    SDValue MinMaxLHS = TVal, MinMaxRHS = FVal;
    if (selectCCOpsAreFMaxCompatible(LHS, MinMaxRHS) &&
        selectCCOpsAreFMaxCompatible(RHS, MinMaxLHS)) {
      CC = ISD::getSetCCSwappedOperands(CC);
      std::swap(MinMaxLHS, MinMaxRHS);
    }

    if (selectCCOpsAreFMaxCompatible(LHS, MinMaxLHS) &&
        selectCCOpsAreFMaxCompatible(RHS, MinMaxRHS)) {
      switch (CC) {
      default:
        break;
      case ISD::SETGT:
      case ISD::SETGE:
      case ISD::SETUGT:
      case ISD::SETUGE:
      case ISD::SETOGT:
      case ISD::SETOGE:
        return DAG.getNode(AArch64ISD::FMAX, dl, VT, MinMaxLHS, MinMaxRHS);
        break;
      case ISD::SETLT:
      case ISD::SETLE:
      case ISD::SETULT:
      case ISD::SETULE:
      case ISD::SETOLT:
      case ISD::SETOLE:
        return DAG.getNode(AArch64ISD::FMIN, dl, VT, MinMaxLHS, MinMaxRHS);
        break;
      }
    }
  }

  // If that fails, we'll need to perform an FCMP + CSEL sequence.  Go ahead
  // and do the comparison.
  SDValue Cmp = emitComparison(LHS, RHS, CC, dl, DAG);

  // Unfortunately, the mapping of LLVM FP CC's onto AArch64 CC's isn't totally
  // clean.  Some of them require two CSELs to implement.
  AArch64CC::CondCode CC1, CC2;
  changeFPCCToAArch64CC(CC, CC1, CC2);
  SDValue CC1Val = DAG.getConstant(CC1, MVT::i32);
  SDValue CS1 = DAG.getNode(AArch64ISD::CSEL, dl, VT, TVal, FVal, CC1Val, Cmp);

  // If we need a second CSEL, emit it, using the output of the first as the
  // RHS.  We're effectively OR'ing the two CC's together.
  if (CC2 != AArch64CC::AL) {
    SDValue CC2Val = DAG.getConstant(CC2, MVT::i32);
    return DAG.getNode(AArch64ISD::CSEL, dl, VT, TVal, CS1, CC2Val, Cmp);
  }

  // Otherwise, return the output of the first CSEL.
  return CS1;
}

SDValue AArch64TargetLowering::LowerJumpTable(SDValue Op,
                                              SelectionDAG &DAG) const {
  // Jump table entries as PC relative offsets. No additional tweaking
  // is necessary here. Just get the address of the jump table.
  JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
  EVT PtrVT = getPointerTy();
  SDLoc DL(Op);

  if (getTargetMachine().getCodeModel() == CodeModel::Large &&
      !Subtarget->isTargetMachO()) {
    const unsigned char MO_NC = AArch64II::MO_NC;
    return DAG.getNode(
        AArch64ISD::WrapperLarge, DL, PtrVT,
        DAG.getTargetJumpTable(JT->getIndex(), PtrVT, AArch64II::MO_G3),
        DAG.getTargetJumpTable(JT->getIndex(), PtrVT, AArch64II::MO_G2 | MO_NC),
        DAG.getTargetJumpTable(JT->getIndex(), PtrVT, AArch64II::MO_G1 | MO_NC),
        DAG.getTargetJumpTable(JT->getIndex(), PtrVT,
                               AArch64II::MO_G0 | MO_NC));
  }

  SDValue Hi =
      DAG.getTargetJumpTable(JT->getIndex(), PtrVT, AArch64II::MO_PAGE);
  SDValue Lo = DAG.getTargetJumpTable(JT->getIndex(), PtrVT,
                                      AArch64II::MO_PAGEOFF | AArch64II::MO_NC);
  SDValue ADRP = DAG.getNode(AArch64ISD::ADRP, DL, PtrVT, Hi);
  return DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, ADRP, Lo);
}

SDValue AArch64TargetLowering::LowerConstantPool(SDValue Op,
                                                 SelectionDAG &DAG) const {
  ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
  EVT PtrVT = getPointerTy();
  SDLoc DL(Op);

  if (getTargetMachine().getCodeModel() == CodeModel::Large) {
    // Use the GOT for the large code model on iOS.
    if (Subtarget->isTargetMachO()) {
      SDValue GotAddr = DAG.getTargetConstantPool(
          CP->getConstVal(), PtrVT, CP->getAlignment(), CP->getOffset(),
          AArch64II::MO_GOT);
      return DAG.getNode(AArch64ISD::LOADgot, DL, PtrVT, GotAddr);
    }

    const unsigned char MO_NC = AArch64II::MO_NC;
    return DAG.getNode(
        AArch64ISD::WrapperLarge, DL, PtrVT,
        DAG.getTargetConstantPool(CP->getConstVal(), PtrVT, CP->getAlignment(),
                                  CP->getOffset(), AArch64II::MO_G3),
        DAG.getTargetConstantPool(CP->getConstVal(), PtrVT, CP->getAlignment(),
                                  CP->getOffset(), AArch64II::MO_G2 | MO_NC),
        DAG.getTargetConstantPool(CP->getConstVal(), PtrVT, CP->getAlignment(),
                                  CP->getOffset(), AArch64II::MO_G1 | MO_NC),
        DAG.getTargetConstantPool(CP->getConstVal(), PtrVT, CP->getAlignment(),
                                  CP->getOffset(), AArch64II::MO_G0 | MO_NC));
  } else {
    // Use ADRP/ADD or ADRP/LDR for everything else: the small memory model on
    // ELF, the only valid one on Darwin.
    SDValue Hi =
        DAG.getTargetConstantPool(CP->getConstVal(), PtrVT, CP->getAlignment(),
                                  CP->getOffset(), AArch64II::MO_PAGE);
    SDValue Lo = DAG.getTargetConstantPool(
        CP->getConstVal(), PtrVT, CP->getAlignment(), CP->getOffset(),
        AArch64II::MO_PAGEOFF | AArch64II::MO_NC);

    SDValue ADRP = DAG.getNode(AArch64ISD::ADRP, DL, PtrVT, Hi);
    return DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, ADRP, Lo);
  }
}

SDValue AArch64TargetLowering::LowerBlockAddress(SDValue Op,
                                               SelectionDAG &DAG) const {
  const BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
  EVT PtrVT = getPointerTy();
  SDLoc DL(Op);
  if (getTargetMachine().getCodeModel() == CodeModel::Large &&
      !Subtarget->isTargetMachO()) {
    const unsigned char MO_NC = AArch64II::MO_NC;
    return DAG.getNode(
        AArch64ISD::WrapperLarge, DL, PtrVT,
        DAG.getTargetBlockAddress(BA, PtrVT, 0, AArch64II::MO_G3),
        DAG.getTargetBlockAddress(BA, PtrVT, 0, AArch64II::MO_G2 | MO_NC),
        DAG.getTargetBlockAddress(BA, PtrVT, 0, AArch64II::MO_G1 | MO_NC),
        DAG.getTargetBlockAddress(BA, PtrVT, 0, AArch64II::MO_G0 | MO_NC));
  } else {
    SDValue Hi = DAG.getTargetBlockAddress(BA, PtrVT, 0, AArch64II::MO_PAGE);
    SDValue Lo = DAG.getTargetBlockAddress(BA, PtrVT, 0, AArch64II::MO_PAGEOFF |
                                                             AArch64II::MO_NC);
    SDValue ADRP = DAG.getNode(AArch64ISD::ADRP, DL, PtrVT, Hi);
    return DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, ADRP, Lo);
  }
}

SDValue AArch64TargetLowering::LowerDarwin_VASTART(SDValue Op,
                                                 SelectionDAG &DAG) const {
  AArch64FunctionInfo *FuncInfo =
      DAG.getMachineFunction().getInfo<AArch64FunctionInfo>();

  SDLoc DL(Op);
  SDValue FR =
      DAG.getFrameIndex(FuncInfo->getVarArgsStackIndex(), getPointerTy());
  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 AArch64TargetLowering::LowerAAPCS_VASTART(SDValue Op,
                                                SelectionDAG &DAG) const {
  // The layout of the va_list struct is specified in the AArch64 Procedure Call
  // Standard, section B.3.
  MachineFunction &MF = DAG.getMachineFunction();
  AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();
  SDLoc DL(Op);

  SDValue Chain = Op.getOperand(0);
  SDValue VAList = Op.getOperand(1);
  const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
  SmallVector<SDValue, 4> MemOps;

  // void *__stack at offset 0
  SDValue Stack =
      DAG.getFrameIndex(FuncInfo->getVarArgsStackIndex(), getPointerTy());
  MemOps.push_back(DAG.getStore(Chain, DL, Stack, VAList,
                                MachinePointerInfo(SV), false, false, 8));

  // void *__gr_top at offset 8
  int GPRSize = FuncInfo->getVarArgsGPRSize();
  if (GPRSize > 0) {
    SDValue GRTop, GRTopAddr;

    GRTopAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
                            DAG.getConstant(8, getPointerTy()));

    GRTop = DAG.getFrameIndex(FuncInfo->getVarArgsGPRIndex(), getPointerTy());
    GRTop = DAG.getNode(ISD::ADD, DL, getPointerTy(), GRTop,
                        DAG.getConstant(GPRSize, getPointerTy()));

    MemOps.push_back(DAG.getStore(Chain, DL, GRTop, GRTopAddr,
                                  MachinePointerInfo(SV, 8), false, false, 8));
  }

  // void *__vr_top at offset 16
  int FPRSize = FuncInfo->getVarArgsFPRSize();
  if (FPRSize > 0) {
    SDValue VRTop, VRTopAddr;
    VRTopAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
                            DAG.getConstant(16, getPointerTy()));

    VRTop = DAG.getFrameIndex(FuncInfo->getVarArgsFPRIndex(), getPointerTy());
    VRTop = DAG.getNode(ISD::ADD, DL, getPointerTy(), VRTop,
                        DAG.getConstant(FPRSize, getPointerTy()));

    MemOps.push_back(DAG.getStore(Chain, DL, VRTop, VRTopAddr,
                                  MachinePointerInfo(SV, 16), false, false, 8));
  }

  // int __gr_offs at offset 24
  SDValue GROffsAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
                                   DAG.getConstant(24, getPointerTy()));
  MemOps.push_back(DAG.getStore(Chain, DL, DAG.getConstant(-GPRSize, MVT::i32),
                                GROffsAddr, MachinePointerInfo(SV, 24), false,
                                false, 4));

  // int __vr_offs at offset 28
  SDValue VROffsAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
                                   DAG.getConstant(28, getPointerTy()));
  MemOps.push_back(DAG.getStore(Chain, DL, DAG.getConstant(-FPRSize, MVT::i32),
                                VROffsAddr, MachinePointerInfo(SV, 28), false,
                                false, 4));

  return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, MemOps);
}

SDValue AArch64TargetLowering::LowerVASTART(SDValue Op,
                                            SelectionDAG &DAG) const {
  return Subtarget->isTargetDarwin() ? LowerDarwin_VASTART(Op, DAG)
                                     : LowerAAPCS_VASTART(Op, DAG);
}

SDValue AArch64TargetLowering::LowerVACOPY(SDValue Op,
                                           SelectionDAG &DAG) const {
  // AAPCS has three pointers and two ints (= 32 bytes), Darwin has single
  // pointer.
  unsigned VaListSize = Subtarget->isTargetDarwin() ? 8 : 32;
  const Value *DestSV = cast<SrcValueSDNode>(Op.getOperand(3))->getValue();
  const Value *SrcSV = cast<SrcValueSDNode>(Op.getOperand(4))->getValue();

  return DAG.getMemcpy(Op.getOperand(0), SDLoc(Op), Op.getOperand(1),
                       Op.getOperand(2), DAG.getConstant(VaListSize, MVT::i32),
                       8, false, false, MachinePointerInfo(DestSV),
                       MachinePointerInfo(SrcSV));
}

SDValue AArch64TargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
  assert(Subtarget->isTargetDarwin() &&
         "automatic va_arg instruction only works on Darwin");

  const Value *V = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
  EVT VT = Op.getValueType();
  SDLoc DL(Op);
  SDValue Chain = Op.getOperand(0);
  SDValue Addr = Op.getOperand(1);
  unsigned Align = Op.getConstantOperandVal(3);

  SDValue VAList = DAG.getLoad(getPointerTy(), DL, Chain, Addr,
                               MachinePointerInfo(V), false, false, false, 0);
  Chain = VAList.getValue(1);

  if (Align > 8) {
    assert(((Align & (Align - 1)) == 0) && "Expected Align to be a power of 2");
    VAList = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
                         DAG.getConstant(Align - 1, getPointerTy()));
    VAList = DAG.getNode(ISD::AND, DL, getPointerTy(), VAList,
                         DAG.getConstant(-(int64_t)Align, getPointerTy()));
  }

  Type *ArgTy = VT.getTypeForEVT(*DAG.getContext());
  uint64_t ArgSize = getDataLayout()->getTypeAllocSize(ArgTy);

  // Scalar integer and FP values smaller than 64 bits are implicitly extended
  // up to 64 bits.  At the very least, we have to increase the striding of the
  // vaargs list to match this, and for FP values we need to introduce
  // FP_ROUND nodes as well.
  if (VT.isInteger() && !VT.isVector())
    ArgSize = 8;
  bool NeedFPTrunc = false;
  if (VT.isFloatingPoint() && !VT.isVector() && VT != MVT::f64) {
    ArgSize = 8;
    NeedFPTrunc = true;
  }

  // Increment the pointer, VAList, to the next vaarg
  SDValue VANext = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
                               DAG.getConstant(ArgSize, getPointerTy()));
  // Store the incremented VAList to the legalized pointer
  SDValue APStore = DAG.getStore(Chain, DL, VANext, Addr, MachinePointerInfo(V),
                                 false, false, 0);

  // Load the actual argument out of the pointer VAList
  if (NeedFPTrunc) {
    // Load the value as an f64.
    SDValue WideFP = DAG.getLoad(MVT::f64, DL, APStore, VAList,
                                 MachinePointerInfo(), false, false, false, 0);
    // Round the value down to an f32.
    SDValue NarrowFP = DAG.getNode(ISD::FP_ROUND, DL, VT, WideFP.getValue(0),
                                   DAG.getIntPtrConstant(1));
    SDValue Ops[] = { NarrowFP, WideFP.getValue(1) };
    // Merge the rounded value with the chain output of the load.
    return DAG.getMergeValues(Ops, DL);
  }

  return DAG.getLoad(VT, DL, APStore, VAList, MachinePointerInfo(), false,
                     false, false, 0);
}

SDValue AArch64TargetLowering::LowerFRAMEADDR(SDValue Op,
                                              SelectionDAG &DAG) const {
  MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
  MFI->setFrameAddressIsTaken(true);

  EVT VT = Op.getValueType();
  SDLoc DL(Op);
  unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
  SDValue FrameAddr =
      DAG.getCopyFromReg(DAG.getEntryNode(), DL, AArch64::FP, 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 AArch64TargetLowering::getRegisterByName(const char* RegName,
                                                  EVT VT) const {
  unsigned Reg = StringSwitch<unsigned>(RegName)
                       .Case("sp", AArch64::SP)
                       .Default(0);
  if (Reg)
    return Reg;
  report_fatal_error("Invalid register name global variable");
}

SDValue AArch64TargetLowering::LowerRETURNADDR(SDValue Op,
                                               SelectionDAG &DAG) const {
  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  MFI->setReturnAddressIsTaken(true);

  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(8, getPointerTy());
    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(AArch64::LR, &AArch64::GPR64RegClass);
  return DAG.getCopyFromReg(DAG.getEntryNode(), DL, Reg, VT);
}

/// LowerShiftRightParts - Lower SRA_PARTS, which returns two
/// i64 values and take a 2 x i64 value to shift plus a shift amount.
SDValue AArch64TargetLowering::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::i64,
                                 DAG.getConstant(VTBits, MVT::i64), ShAmt);
  SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, ShAmt);
  SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i64, ShAmt,
                                   DAG.getConstant(VTBits, MVT::i64));
  SDValue Tmp2 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, RevShAmt);

  SDValue Cmp = emitComparison(ExtraShAmt, DAG.getConstant(0, MVT::i64),
                               ISD::SETGE, dl, DAG);
  SDValue CCVal = DAG.getConstant(AArch64CC::GE, MVT::i32);

  SDValue FalseValLo = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
  SDValue TrueValLo = DAG.getNode(Opc, dl, VT, ShOpHi, ExtraShAmt);
  SDValue Lo =
      DAG.getNode(AArch64ISD::CSEL, dl, VT, TrueValLo, FalseValLo, CCVal, Cmp);

  // AArch64 shifts larger than the register width are wrapped rather than
  // clamped, so we can't just emit "hi >> x".
  SDValue FalseValHi = DAG.getNode(Opc, dl, VT, ShOpHi, ShAmt);
  SDValue TrueValHi = Opc == ISD::SRA
                          ? DAG.getNode(Opc, dl, VT, ShOpHi,
                                        DAG.getConstant(VTBits - 1, MVT::i64))
                          : DAG.getConstant(0, VT);
  SDValue Hi =
      DAG.getNode(AArch64ISD::CSEL, dl, VT, TrueValHi, FalseValHi, CCVal, Cmp);

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

/// LowerShiftLeftParts - Lower SHL_PARTS, which returns two
/// i64 values and take a 2 x i64 value to shift plus a shift amount.
SDValue AArch64TargetLowering::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::i64,
                                 DAG.getConstant(VTBits, MVT::i64), ShAmt);
  SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, RevShAmt);
  SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i64, ShAmt,
                                   DAG.getConstant(VTBits, MVT::i64));
  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 Cmp = emitComparison(ExtraShAmt, DAG.getConstant(0, MVT::i64),
                               ISD::SETGE, dl, DAG);
  SDValue CCVal = DAG.getConstant(AArch64CC::GE, MVT::i32);
  SDValue Hi =
      DAG.getNode(AArch64ISD::CSEL, dl, VT, Tmp3, FalseVal, CCVal, Cmp);

  // AArch64 shifts of larger than register sizes are wrapped rather than
  // clamped, so we can't just emit "lo << a" if a is too big.
  SDValue TrueValLo = DAG.getConstant(0, VT);
  SDValue FalseValLo = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ShAmt);
  SDValue Lo =
      DAG.getNode(AArch64ISD::CSEL, dl, VT, TrueValLo, FalseValLo, CCVal, Cmp);

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

bool AArch64TargetLowering::isOffsetFoldingLegal(
    const GlobalAddressSDNode *GA) const {
  // The AArch64 target doesn't support folding offsets into global addresses.
  return false;
}

bool AArch64TargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
  // We can materialize #0.0 as fmov $Rd, XZR for 64-bit and 32-bit cases.
  // FIXME: We should be able to handle f128 as well with a clever lowering.
  if (Imm.isPosZero() && (VT == MVT::f64 || VT == MVT::f32))
    return true;

  if (VT == MVT::f64)
    return AArch64_AM::getFP64Imm(Imm) != -1;
  else if (VT == MVT::f32)
    return AArch64_AM::getFP32Imm(Imm) != -1;
  return false;
}

//===----------------------------------------------------------------------===//
//                          AArch64 Optimization Hooks
//===----------------------------------------------------------------------===//

//===----------------------------------------------------------------------===//
//                          AArch64 Inline Assembly Support
//===----------------------------------------------------------------------===//

// Table of Constraints
// TODO: This is the current set of constraints supported by ARM for the
// compiler, not all of them may make sense, e.g. S may be difficult to support.
//
// r - A general register
// w - An FP/SIMD register of some size in the range v0-v31
// x - An FP/SIMD register of some size in the range v0-v15
// I - Constant that can be used with an ADD instruction
// J - Constant that can be used with a SUB instruction
// K - Constant that can be used with a 32-bit logical instruction
// L - Constant that can be used with a 64-bit logical instruction
// M - Constant that can be used as a 32-bit MOV immediate
// N - Constant that can be used as a 64-bit MOV immediate
// Q - A memory reference with base register and no offset
// S - A symbolic address
// Y - Floating point constant zero
// Z - Integer constant zero
//
//   Note that general register operands will be output using their 64-bit x
// register name, whatever the size of the variable, unless the asm operand
// is prefixed by the %w modifier. Floating-point and SIMD register operands
// will be output with the v prefix unless prefixed by the %b, %h, %s, %d or
// %q modifier.

/// getConstraintType - Given a constraint letter, return the type of
/// constraint it is for this target.
AArch64TargetLowering::ConstraintType
AArch64TargetLowering::getConstraintType(const std::string &Constraint) const {
  if (Constraint.size() == 1) {
    switch (Constraint[0]) {
    default:
      break;
    case 'z':
      return C_Other;
    case 'x':
    case 'w':
      return C_RegisterClass;
    // An address with a single base register. Due to the way we
    // currently handle addresses it is the same as 'r'.
    case 'Q':
      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
AArch64TargetLowering::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 'x':
  case 'w':
    if (type->isFloatingPointTy() || type->isVectorTy())
      weight = CW_Register;
    break;
  case 'z':
    weight = CW_Constant;
    break;
  }
  return weight;
}

std::pair<unsigned, const TargetRegisterClass *>
AArch64TargetLowering::getRegForInlineAsmConstraint(
    const std::string &Constraint, MVT VT) const {
  if (Constraint.size() == 1) {
    switch (Constraint[0]) {
    case 'r':
      if (VT.getSizeInBits() == 64)
        return std::make_pair(0U, &AArch64::GPR64commonRegClass);
      return std::make_pair(0U, &AArch64::GPR32commonRegClass);
    case 'w':
      if (VT == MVT::f32)
        return std::make_pair(0U, &AArch64::FPR32RegClass);
      if (VT.getSizeInBits() == 64)
        return std::make_pair(0U, &AArch64::FPR64RegClass);
      if (VT.getSizeInBits() == 128)
        return std::make_pair(0U, &AArch64::FPR128RegClass);
      break;
    // The instructions that this constraint is designed for can
    // only take 128-bit registers so just use that regclass.
    case 'x':
      if (VT.getSizeInBits() == 128)
        return std::make_pair(0U, &AArch64::FPR128_loRegClass);
      break;
    }
  }
  if (StringRef("{cc}").equals_lower(Constraint))
    return std::make_pair(unsigned(AArch64::NZCV), &AArch64::CCRRegClass);

  // Use the default implementation in TargetLowering to convert the register
  // constraint into a member of a register class.
  std::pair<unsigned, const TargetRegisterClass *> Res;
  Res = TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);

  // Not found as a standard register?
  if (!Res.second) {
    unsigned Size = Constraint.size();
    if ((Size == 4 || Size == 5) && Constraint[0] == '{' &&
        tolower(Constraint[1]) == 'v' && Constraint[Size - 1] == '}') {
      const std::string Reg =
          std::string(&Constraint[2], &Constraint[Size - 1]);
      int RegNo = atoi(Reg.c_str());
      if (RegNo >= 0 && RegNo <= 31) {
        // v0 - v31 are aliases of q0 - q31.
        // By default we'll emit v0-v31 for this unless there's a modifier where
        // we'll emit the correct register as well.
        Res.first = AArch64::FPR128RegClass.getRegister(RegNo);
        Res.second = &AArch64::FPR128RegClass;
      }
    }
  }

  return Res;
}

/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector.  If it is invalid, don't add anything to Ops.
void AArch64TargetLowering::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;

  // This set of constraints deal with valid constants for various instructions.
  // Validate and return a target constant for them if we can.
  case 'z': {
    // 'z' maps to xzr or wzr so it needs an input of 0.
    ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
    if (!C || C->getZExtValue() != 0)
      return;

    if (Op.getValueType() == MVT::i64)
      Result = DAG.getRegister(AArch64::XZR, MVT::i64);
    else
      Result = DAG.getRegister(AArch64::WZR, MVT::i32);
    break;
  }

  case 'I':
  case 'J':
  case 'K':
  case 'L':
  case 'M':
  case 'N':
    ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
    if (!C)
      return;

    // Grab the value and do some validation.
    uint64_t CVal = C->getZExtValue();
    switch (ConstraintLetter) {
    // The I constraint applies only to simple ADD or SUB immediate operands:
    // i.e. 0 to 4095 with optional shift by 12
    // The J constraint applies only to ADD or SUB immediates that would be
    // valid when negated, i.e. if [an add pattern] were to be output as a SUB
    // instruction [or vice versa], in other words -1 to -4095 with optional
    // left shift by 12.
    case 'I':
      if (isUInt<12>(CVal) || isShiftedUInt<12, 12>(CVal))
        break;
      return;
    case 'J': {
      uint64_t NVal = -C->getSExtValue();
      if (isUInt<12>(NVal) || isShiftedUInt<12, 12>(NVal)) {
        CVal = C->getSExtValue();
        break;
      }
      return;
    }
    // The K and L constraints apply *only* to logical immediates, including
    // what used to be the MOVI alias for ORR (though the MOVI alias has now
    // been removed and MOV should be used). So these constraints have to
    // distinguish between bit patterns that are valid 32-bit or 64-bit
    // "bitmask immediates": for example 0xaaaaaaaa is a valid bimm32 (K), but
    // not a valid bimm64 (L) where 0xaaaaaaaaaaaaaaaa would be valid, and vice
    // versa.
    case 'K':
      if (AArch64_AM::isLogicalImmediate(CVal, 32))
        break;
      return;
    case 'L':
      if (AArch64_AM::isLogicalImmediate(CVal, 64))
        break;
      return;
    // The M and N constraints are a superset of K and L respectively, for use
    // with the MOV (immediate) alias. As well as the logical immediates they
    // also match 32 or 64-bit immediates that can be loaded either using a
    // *single* MOVZ or MOVN , such as 32-bit 0x12340000, 0x00001234, 0xffffedca
    // (M) or 64-bit 0x1234000000000000 (N) etc.
    // As a note some of this code is liberally stolen from the asm parser.
    case 'M': {
      if (!isUInt<32>(CVal))
        return;
      if (AArch64_AM::isLogicalImmediate(CVal, 32))
        break;
      if ((CVal & 0xFFFF) == CVal)
        break;
      if ((CVal & 0xFFFF0000ULL) == CVal)
        break;
      uint64_t NCVal = ~(uint32_t)CVal;
      if ((NCVal & 0xFFFFULL) == NCVal)
        break;
      if ((NCVal & 0xFFFF0000ULL) == NCVal)
        break;
      return;
    }
    case 'N': {
      if (AArch64_AM::isLogicalImmediate(CVal, 64))
        break;
      if ((CVal & 0xFFFFULL) == CVal)
        break;
      if ((CVal & 0xFFFF0000ULL) == CVal)
        break;
      if ((CVal & 0xFFFF00000000ULL) == CVal)
        break;
      if ((CVal & 0xFFFF000000000000ULL) == CVal)
        break;
      uint64_t NCVal = ~CVal;
      if ((NCVal & 0xFFFFULL) == NCVal)
        break;
      if ((NCVal & 0xFFFF0000ULL) == NCVal)
        break;
      if ((NCVal & 0xFFFF00000000ULL) == NCVal)
        break;
      if ((NCVal & 0xFFFF000000000000ULL) == NCVal)
        break;
      return;
    }
    default:
      return;
    }

    // All assembler immediates are 64-bit integers.
    Result = DAG.getTargetConstant(CVal, MVT::i64);
    break;
  }

  if (Result.getNode()) {
    Ops.push_back(Result);
    return;
  }

  return TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
}

//===----------------------------------------------------------------------===//
//                     AArch64 Advanced SIMD Support
//===----------------------------------------------------------------------===//

/// WidenVector - Given a value in the V64 register class, produce the
/// equivalent value in the V128 register class.
static SDValue WidenVector(SDValue V64Reg, SelectionDAG &DAG) {
  EVT VT = V64Reg.getValueType();
  unsigned NarrowSize = VT.getVectorNumElements();
  MVT EltTy = VT.getVectorElementType().getSimpleVT();
  MVT WideTy = MVT::getVectorVT(EltTy, 2 * NarrowSize);
  SDLoc DL(V64Reg);

  return DAG.getNode(ISD::INSERT_SUBVECTOR, DL, WideTy, DAG.getUNDEF(WideTy),
                     V64Reg, DAG.getConstant(0, MVT::i32));
}

/// getExtFactor - Determine the adjustment factor for the position when
/// generating an "extract from vector registers" instruction.
static unsigned getExtFactor(SDValue &V) {
  EVT EltType = V.getValueType().getVectorElementType();
  return EltType.getSizeInBits() / 8;
}

/// NarrowVector - Given a value in the V128 register class, produce the
/// equivalent value in the V64 register class.
static SDValue NarrowVector(SDValue V128Reg, SelectionDAG &DAG) {
  EVT VT = V128Reg.getValueType();
  unsigned WideSize = VT.getVectorNumElements();
  MVT EltTy = VT.getVectorElementType().getSimpleVT();
  MVT NarrowTy = MVT::getVectorVT(EltTy, WideSize / 2);
  SDLoc DL(V128Reg);

  return DAG.getTargetExtractSubreg(AArch64::dsub, DL, NarrowTy, V128Reg);
}

// Gather data to see if the operation can be modelled as a
// shuffle in combination with VEXTs.
SDValue AArch64TargetLowering::ReconstructShuffle(SDValue Op,
                                                  SelectionDAG &DAG) const {
  assert(Op.getOpcode() == ISD::BUILD_VECTOR && "Unknown opcode!");
  SDLoc dl(Op);
  EVT VT = Op.getValueType();
  unsigned NumElts = VT.getVectorNumElements();

  struct ShuffleSourceInfo {
    SDValue Vec;
    unsigned MinElt;
    unsigned MaxElt;

    // We may insert some combination of BITCASTs and VEXT nodes to force Vec to
    // be compatible with the shuffle we intend to construct. As a result
    // ShuffleVec will be some sliding window into the original Vec.
    SDValue ShuffleVec;

    // Code should guarantee that element i in Vec starts at element "WindowBase
    // + i * WindowScale in ShuffleVec".
    int WindowBase;
    int WindowScale;

    bool operator ==(SDValue OtherVec) { return Vec == OtherVec; }
    ShuffleSourceInfo(SDValue Vec)
        : Vec(Vec), MinElt(UINT_MAX), MaxElt(0), ShuffleVec(Vec), WindowBase(0),
          WindowScale(1) {}
  };

  // First gather all vectors used as an immediate source for this BUILD_VECTOR
  // node.
  SmallVector<ShuffleSourceInfo, 2> Sources;
  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();
    }

    // Add this element source to the list if it's not already there.
    SDValue SourceVec = V.getOperand(0);
    auto Source = std::find(Sources.begin(), Sources.end(), SourceVec);
    if (Source == Sources.end())
      Source = Sources.insert(Sources.end(), ShuffleSourceInfo(SourceVec));

    // Update the minimum and maximum lane number seen.
    unsigned EltNo = cast<ConstantSDNode>(V.getOperand(1))->getZExtValue();
    Source->MinElt = std::min(Source->MinElt, EltNo);
    Source->MaxElt = std::max(Source->MaxElt, EltNo);
  }

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

  // Find out the smallest element size among result and two sources, and use
  // it as element size to build the shuffle_vector.
  EVT SmallestEltTy = VT.getVectorElementType();
  for (auto &Source : Sources) {
    EVT SrcEltTy = Source.Vec.getValueType().getVectorElementType();
    if (SrcEltTy.bitsLT(SmallestEltTy)) {
      SmallestEltTy = SrcEltTy;
    }
  }
  unsigned ResMultiplier =
      VT.getVectorElementType().getSizeInBits() / SmallestEltTy.getSizeInBits();
  NumElts = VT.getSizeInBits() / SmallestEltTy.getSizeInBits();
  EVT ShuffleVT = EVT::getVectorVT(*DAG.getContext(), SmallestEltTy, NumElts);

  // If the source vector is too wide or too narrow, we may nevertheless be able
  // to construct a compatible shuffle either by concatenating it with UNDEF or
  // extracting a suitable range of elements.
  for (auto &Src : Sources) {
    EVT SrcVT = Src.ShuffleVec.getValueType();

    if (SrcVT.getSizeInBits() == VT.getSizeInBits())
      continue;

    // This stage of the search produces a source with the same element type as
    // the original, but with a total width matching the BUILD_VECTOR output.
    EVT EltVT = SrcVT.getVectorElementType();
    unsigned NumSrcElts = VT.getSizeInBits() / EltVT.getSizeInBits();
    EVT DestVT = EVT::getVectorVT(*DAG.getContext(), EltVT, NumSrcElts);

    if (SrcVT.getSizeInBits() < VT.getSizeInBits()) {
      assert(2 * SrcVT.getSizeInBits() == VT.getSizeInBits());
      // We can pad out the smaller vector for free, so if it's part of a
      // shuffle...
      Src.ShuffleVec =
          DAG.getNode(ISD::CONCAT_VECTORS, dl, DestVT, Src.ShuffleVec,
                      DAG.getUNDEF(Src.ShuffleVec.getValueType()));
      continue;
    }

    assert(SrcVT.getSizeInBits() == 2 * VT.getSizeInBits());

    if (Src.MaxElt - Src.MinElt >= NumSrcElts) {
      // Span too large for a VEXT to cope
      return SDValue();
    }

    if (Src.MinElt >= NumSrcElts) {
      // The extraction can just take the second half
      Src.ShuffleVec =
          DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT, Src.ShuffleVec,
                      DAG.getIntPtrConstant(NumSrcElts));
      Src.WindowBase = -NumSrcElts;
    } else if (Src.MaxElt < NumSrcElts) {
      // The extraction can just take the first half
      Src.ShuffleVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT,
                                   Src.ShuffleVec, DAG.getIntPtrConstant(0));
    } else {
      // An actual VEXT is needed
      SDValue VEXTSrc1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT,
                                     Src.ShuffleVec, DAG.getIntPtrConstant(0));
      SDValue VEXTSrc2 =
          DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT, Src.ShuffleVec,
                      DAG.getIntPtrConstant(NumSrcElts));
      unsigned Imm = Src.MinElt * getExtFactor(VEXTSrc1);

      Src.ShuffleVec = DAG.getNode(AArch64ISD::EXT, dl, DestVT, VEXTSrc1,
                                   VEXTSrc2, DAG.getConstant(Imm, MVT::i32));
      Src.WindowBase = -Src.MinElt;
    }
  }

  // Another possible incompatibility occurs from the vector element types. We
  // can fix this by bitcasting the source vectors to the same type we intend
  // for the shuffle.
  for (auto &Src : Sources) {
    EVT SrcEltTy = Src.ShuffleVec.getValueType().getVectorElementType();
    if (SrcEltTy == SmallestEltTy)
      continue;
    assert(ShuffleVT.getVectorElementType() == SmallestEltTy);
    Src.ShuffleVec = DAG.getNode(ISD::BITCAST, dl, ShuffleVT, Src.ShuffleVec);
    Src.WindowScale = SrcEltTy.getSizeInBits() / SmallestEltTy.getSizeInBits();
    Src.WindowBase *= Src.WindowScale;
  }

  // Final sanity check before we try to actually produce a shuffle.
  DEBUG(
    for (auto Src : Sources)
      assert(Src.ShuffleVec.getValueType() == ShuffleVT);
  );

  // The stars all align, our next step is to produce the mask for the shuffle.
  SmallVector<int, 8> Mask(ShuffleVT.getVectorNumElements(), -1);
  int BitsPerShuffleLane = ShuffleVT.getVectorElementType().getSizeInBits();
  for (unsigned i = 0; i < VT.getVectorNumElements(); ++i) {
    SDValue Entry = Op.getOperand(i);
    if (Entry.getOpcode() == ISD::UNDEF)
      continue;

    auto Src = std::find(Sources.begin(), Sources.end(), Entry.getOperand(0));
    int EltNo = cast<ConstantSDNode>(Entry.getOperand(1))->getSExtValue();

    // EXTRACT_VECTOR_ELT performs an implicit any_ext; BUILD_VECTOR an implicit
    // trunc. So only std::min(SrcBits, DestBits) actually get defined in this
    // segment.
    EVT OrigEltTy = Entry.getOperand(0).getValueType().getVectorElementType();
    int BitsDefined = std::min(OrigEltTy.getSizeInBits(),
                               VT.getVectorElementType().getSizeInBits());
    int LanesDefined = BitsDefined / BitsPerShuffleLane;

    // This source is expected to fill ResMultiplier lanes of the final shuffle,
    // starting at the appropriate offset.
    int *LaneMask = &Mask[i * ResMultiplier];

    int ExtractBase = EltNo * Src->WindowScale + Src->WindowBase;
    ExtractBase += NumElts * (Src - Sources.begin());
    for (int j = 0; j < LanesDefined; ++j)
      LaneMask[j] = ExtractBase + j;
  }

  // Final check before we try to produce nonsense...
  if (!isShuffleMaskLegal(Mask, ShuffleVT))
    return SDValue();

  SDValue ShuffleOps[] = { DAG.getUNDEF(ShuffleVT), DAG.getUNDEF(ShuffleVT) };
  for (unsigned i = 0; i < Sources.size(); ++i)
    ShuffleOps[i] = Sources[i].ShuffleVec;

  SDValue Shuffle = DAG.getVectorShuffle(ShuffleVT, dl, ShuffleOps[0],
                                         ShuffleOps[1], &Mask[0]);
  return DAG.getNode(ISD::BITCAST, dl, VT, Shuffle);
}

// check if an EXT instruction can handle the shuffle mask when the
// vector sources of the shuffle are the same.
static bool isSingletonEXTMask(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;
}

// check if an EXT instruction can handle the shuffle mask when the
// vector sources of the shuffle are different.
static bool isEXTMask(ArrayRef<int> M, EVT VT, bool &ReverseEXT,
                      unsigned &Imm) {
  // Look for the first non-undef element.
  const int *FirstRealElt = std::find_if(M.begin(), M.end(),
      [](int Elt) {return Elt >= 0;});

  // Benefit form APInt to handle overflow when calculating expected element.
  unsigned NumElts = VT.getVectorNumElements();
  unsigned MaskBits = APInt(32, NumElts * 2).logBase2();
  APInt ExpectedElt = APInt(MaskBits, *FirstRealElt + 1);
  // The following shuffle indices must be the successive elements after the
  // first real element.
  const int *FirstWrongElt = std::find_if(FirstRealElt + 1, M.end(),
      [&](int Elt) {return Elt != ExpectedElt++ && Elt != -1;});
  if (FirstWrongElt != M.end())
    return false;

  // The index of an EXT is the first element if it is not UNDEF.
  // Watch out for the beginning UNDEFs. The EXT index should be the expected
  // value of the first element.  E.g. 
  // <-1, -1, 3, ...> is treated as <1, 2, 3, ...>.
  // <-1, -1, 0, 1, ...> is treated as <2*NumElts-2, 2*NumElts-1, 0, 1, ...>.
  // ExpectedElt is the last mask index plus 1.
  Imm = ExpectedElt.getZExtValue();

  // There are two difference cases requiring to reverse input vectors.
  // For example, for vector <4 x i32> we have the following cases,
  // Case 1: shufflevector(<4 x i32>,<4 x i32>,<-1, -1, -1, 0>)
  // Case 2: shufflevector(<4 x i32>,<4 x i32>,<-1, -1, 7, 0>)
  // For both cases, we finally use mask <5, 6, 7, 0>, which requires
  // to reverse two input vectors.
  if (Imm < NumElts)
    ReverseEXT = true;
  else
    Imm -= NumElts;

  return true;
}

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

  return true;
}

static bool isUZPMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  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;
  }

  return true;
}

static bool isTRNMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  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;
}

/// isZIP_v_undef_Mask - Special case of isZIPMask 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 isZIP_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  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;
  }

  return true;
}

/// isUZP_v_undef_Mask - Special case of isUZPMask 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 isUZP_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  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;
    }
  }

  return true;
}

/// isTRN_v_undef_Mask - Special case of isTRNMask 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 isTRN_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
  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 isINSMask(ArrayRef<int> M, int NumInputElements,
                      bool &DstIsLeft, int &Anomaly) {
  if (M.size() != static_cast<size_t>(NumInputElements))
    return false;

  int NumLHSMatch = 0, NumRHSMatch = 0;
  int LastLHSMismatch = -1, LastRHSMismatch = -1;

  for (int i = 0; i < NumInputElements; ++i) {
    if (M[i] == -1) {
      ++NumLHSMatch;
      ++NumRHSMatch;
      continue;
    }

    if (M[i] == i)
      ++NumLHSMatch;
    else
      LastLHSMismatch = i;

    if (M[i] == i + NumInputElements)
      ++NumRHSMatch;
    else
      LastRHSMismatch = i;
  }

  if (NumLHSMatch == NumInputElements - 1) {
    DstIsLeft = true;
    Anomaly = LastLHSMismatch;
    return true;
  } else if (NumRHSMatch == NumInputElements - 1) {
    DstIsLeft = false;
    Anomaly = LastRHSMismatch;
    return true;
  }

  return false;
}

static bool isConcatMask(ArrayRef<int> Mask, EVT VT, bool SplitLHS) {
  if (VT.getSizeInBits() != 128)
    return false;

  unsigned NumElts = VT.getVectorNumElements();

  for (int I = 0, E = NumElts / 2; I != E; I++) {
    if (Mask[I] != I)
      return false;
  }

  int Offset = NumElts / 2;
  for (int I = NumElts / 2, E = NumElts; I != E; I++) {
    if (Mask[I] != I + SplitLHS * Offset)
      return false;
  }

  return true;
}

static SDValue tryFormConcatFromShuffle(SDValue Op, SelectionDAG &DAG) {
  SDLoc DL(Op);
  EVT VT = Op.getValueType();
  SDValue V0 = Op.getOperand(0);
  SDValue V1 = Op.getOperand(1);
  ArrayRef<int> Mask = cast<ShuffleVectorSDNode>(Op)->getMask();

  if (VT.getVectorElementType() != V0.getValueType().getVectorElementType() ||
      VT.getVectorElementType() != V1.getValueType().getVectorElementType())
    return SDValue();

  bool SplitV0 = V0.getValueType().getSizeInBits() == 128;

  if (!isConcatMask(Mask, VT, SplitV0))
    return SDValue();

  EVT CastVT = EVT::getVectorVT(*DAG.getContext(), VT.getVectorElementType(),
                                VT.getVectorNumElements() / 2);
  if (SplitV0) {
    V0 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, CastVT, V0,
                     DAG.getConstant(0, MVT::i64));
  }
  if (V1.getValueType().getSizeInBits() == 128) {
    V1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, CastVT, V1,
                     DAG.getConstant(0, MVT::i64));
  }
  return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, V0, V1);
}

/// 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(AArch64ISD::REV64, dl, VT, OpLHS);
    // vrev <4 x i16> -> REV32
    if (VT.getVectorElementType() == MVT::i16 ||
        VT.getVectorElementType() == MVT::f16)
      return DAG.getNode(AArch64ISD::REV32, dl, VT, OpLHS);
    // vrev <4 x i8> -> REV16
    assert(VT.getVectorElementType() == MVT::i8);
    return DAG.getNode(AArch64ISD::REV16, dl, VT, OpLHS);
  case OP_VDUP0:
  case OP_VDUP1:
  case OP_VDUP2:
  case OP_VDUP3: {
    EVT EltTy = VT.getVectorElementType();
    unsigned Opcode;
    if (EltTy == MVT::i8)
      Opcode = AArch64ISD::DUPLANE8;
    else if (EltTy == MVT::i16)
      Opcode = AArch64ISD::DUPLANE16;
    else if (EltTy == MVT::i32 || EltTy == MVT::f32)
      Opcode = AArch64ISD::DUPLANE32;
    else if (EltTy == MVT::i64 || EltTy == MVT::f64)
      Opcode = AArch64ISD::DUPLANE64;
    else
      llvm_unreachable("Invalid vector element type?");

    if (VT.getSizeInBits() == 64)
      OpLHS = WidenVector(OpLHS, DAG);
    SDValue Lane = DAG.getConstant(OpNum - OP_VDUP0, MVT::i64);
    return DAG.getNode(Opcode, dl, VT, OpLHS, Lane);
  }
  case OP_VEXT1:
  case OP_VEXT2:
  case OP_VEXT3: {
    unsigned Imm = (OpNum - OP_VEXT1 + 1) * getExtFactor(OpLHS);
    return DAG.getNode(AArch64ISD::EXT, dl, VT, OpLHS, OpRHS,
                       DAG.getConstant(Imm, MVT::i32));
  }
  case OP_VUZPL:
    return DAG.getNode(AArch64ISD::UZP1, dl, DAG.getVTList(VT, VT), OpLHS,
                       OpRHS);
  case OP_VUZPR:
    return DAG.getNode(AArch64ISD::UZP2, dl, DAG.getVTList(VT, VT), OpLHS,
                       OpRHS);
  case OP_VZIPL:
    return DAG.getNode(AArch64ISD::ZIP1, dl, DAG.getVTList(VT, VT), OpLHS,
                       OpRHS);
  case OP_VZIPR:
    return DAG.getNode(AArch64ISD::ZIP2, dl, DAG.getVTList(VT, VT), OpLHS,
                       OpRHS);
  case OP_VTRNL:
    return DAG.getNode(AArch64ISD::TRN1, dl, DAG.getVTList(VT, VT), OpLHS,
                       OpRHS);
  case OP_VTRNR:
    return DAG.getNode(AArch64ISD::TRN2, dl, DAG.getVTList(VT, VT), OpLHS,
                       OpRHS);
  }
}

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

  EVT EltVT = Op.getValueType().getVectorElementType();
  unsigned BytesPerElt = EltVT.getSizeInBits() / 8;

  SmallVector<SDValue, 8> TBLMask;
  for (int Val : ShuffleMask) {
    for (unsigned Byte = 0; Byte < BytesPerElt; ++Byte) {
      unsigned Offset = Byte + Val * BytesPerElt;
      TBLMask.push_back(DAG.getConstant(Offset, MVT::i32));
    }
  }

  MVT IndexVT = MVT::v8i8;
  unsigned IndexLen = 8;
  if (Op.getValueType().getSizeInBits() == 128) {
    IndexVT = MVT::v16i8;
    IndexLen = 16;
  }

  SDValue V1Cst = DAG.getNode(ISD::BITCAST, DL, IndexVT, V1);
  SDValue V2Cst = DAG.getNode(ISD::BITCAST, DL, IndexVT, V2);

  SDValue Shuffle;
  if (V2.getNode()->getOpcode() == ISD::UNDEF) {
    if (IndexLen == 8)
      V1Cst = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v16i8, V1Cst, V1Cst);
    Shuffle = DAG.getNode(
        ISD::INTRINSIC_WO_CHAIN, DL, IndexVT,
        DAG.getConstant(Intrinsic::aarch64_neon_tbl1, MVT::i32), V1Cst,
        DAG.getNode(ISD::BUILD_VECTOR, DL, IndexVT,
                    makeArrayRef(TBLMask.data(), IndexLen)));
  } else {
    if (IndexLen == 8) {
      V1Cst = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v16i8, V1Cst, V2Cst);
      Shuffle = DAG.getNode(
          ISD::INTRINSIC_WO_CHAIN, DL, IndexVT,
          DAG.getConstant(Intrinsic::aarch64_neon_tbl1, MVT::i32), V1Cst,
          DAG.getNode(ISD::BUILD_VECTOR, DL, IndexVT,
                      makeArrayRef(TBLMask.data(), IndexLen)));
    } else {
      // FIXME: We cannot, for the moment, emit a TBL2 instruction because we
      // cannot currently represent the register constraints on the input
      // table registers.
      //  Shuffle = DAG.getNode(AArch64ISD::TBL2, DL, IndexVT, V1Cst, V2Cst,
      //                   DAG.getNode(ISD::BUILD_VECTOR, DL, IndexVT,
      //                               &TBLMask[0], IndexLen));
      Shuffle = DAG.getNode(
          ISD::INTRINSIC_WO_CHAIN, DL, IndexVT,
          DAG.getConstant(Intrinsic::aarch64_neon_tbl2, MVT::i32), V1Cst, V2Cst,
          DAG.getNode(ISD::BUILD_VECTOR, DL, IndexVT,
                      makeArrayRef(TBLMask.data(), IndexLen)));
    }
  }
  return DAG.getNode(ISD::BITCAST, DL, Op.getValueType(), Shuffle);
}

static unsigned getDUPLANEOp(EVT EltType) {
  if (EltType == MVT::i8)
    return AArch64ISD::DUPLANE8;
  if (EltType == MVT::i16 || EltType == MVT::f16)
    return AArch64ISD::DUPLANE16;
  if (EltType == MVT::i32 || EltType == MVT::f32)
    return AArch64ISD::DUPLANE32;
  if (EltType == MVT::i64 || EltType == MVT::f64)
    return AArch64ISD::DUPLANE64;

  llvm_unreachable("Invalid vector element type?");
}

SDValue AArch64TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
                                                   SelectionDAG &DAG) const {
  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.
  ArrayRef<int> ShuffleMask = SVN->getMask();

  SDValue V1 = Op.getOperand(0);
  SDValue V2 = Op.getOperand(1);

  if (ShuffleVectorSDNode::isSplatMask(&ShuffleMask[0],
                                       V1.getValueType().getSimpleVT())) {
    int Lane = SVN->getSplatIndex();
    // If this is undef splat, generate it via "just" vdup, if possible.
    if (Lane == -1)
      Lane = 0;

    if (Lane == 0 && V1.getOpcode() == ISD::SCALAR_TO_VECTOR)
      return DAG.getNode(AArch64ISD::DUP, dl, V1.getValueType(),
                         V1.getOperand(0));
    // Test if V1 is a BUILD_VECTOR and the lane being referenced is a non-
    // constant. If so, we can just reference the lane's definition directly.
    if (V1.getOpcode() == ISD::BUILD_VECTOR &&
        !isa<ConstantSDNode>(V1.getOperand(Lane)))
      return DAG.getNode(AArch64ISD::DUP, dl, VT, V1.getOperand(Lane));

    // Otherwise, duplicate from the lane of the input vector.
    unsigned Opcode = getDUPLANEOp(V1.getValueType().getVectorElementType());

    // SelectionDAGBuilder may have "helpfully" already extracted or conatenated
    // to make a vector of the same size as this SHUFFLE. We can ignore the
    // extract entirely, and canonicalise the concat using WidenVector.
    if (V1.getOpcode() == ISD::EXTRACT_SUBVECTOR) {
      Lane += cast<ConstantSDNode>(V1.getOperand(1))->getZExtValue();
      V1 = V1.getOperand(0);
    } else if (V1.getOpcode() == ISD::CONCAT_VECTORS) {
      unsigned Idx = Lane >= (int)VT.getVectorNumElements() / 2;
      Lane -= Idx * VT.getVectorNumElements() / 2;
      V1 = WidenVector(V1.getOperand(Idx), DAG);
    } else if (VT.getSizeInBits() == 64)
      V1 = WidenVector(V1, DAG);

    return DAG.getNode(Opcode, dl, VT, V1, DAG.getConstant(Lane, MVT::i64));
  }

  if (isREVMask(ShuffleMask, VT, 64))
    return DAG.getNode(AArch64ISD::REV64, dl, V1.getValueType(), V1, V2);
  if (isREVMask(ShuffleMask, VT, 32))
    return DAG.getNode(AArch64ISD::REV32, dl, V1.getValueType(), V1, V2);
  if (isREVMask(ShuffleMask, VT, 16))
    return DAG.getNode(AArch64ISD::REV16, dl, V1.getValueType(), V1, V2);

  bool ReverseEXT = false;
  unsigned Imm;
  if (isEXTMask(ShuffleMask, VT, ReverseEXT, Imm)) {
    if (ReverseEXT)
      std::swap(V1, V2);
    Imm *= getExtFactor(V1);
    return DAG.getNode(AArch64ISD::EXT, dl, V1.getValueType(), V1, V2,
                       DAG.getConstant(Imm, MVT::i32));
  } else if (V2->getOpcode() == ISD::UNDEF &&
             isSingletonEXTMask(ShuffleMask, VT, Imm)) {
    Imm *= getExtFactor(V1);
    return DAG.getNode(AArch64ISD::EXT, dl, V1.getValueType(), V1, V1,
                       DAG.getConstant(Imm, MVT::i32));
  }

  unsigned WhichResult;
  if (isZIPMask(ShuffleMask, VT, WhichResult)) {
    unsigned Opc = (WhichResult == 0) ? AArch64ISD::ZIP1 : AArch64ISD::ZIP2;
    return DAG.getNode(Opc, dl, V1.getValueType(), V1, V2);
  }
  if (isUZPMask(ShuffleMask, VT, WhichResult)) {
    unsigned Opc = (WhichResult == 0) ? AArch64ISD::UZP1 : AArch64ISD::UZP2;
    return DAG.getNode(Opc, dl, V1.getValueType(), V1, V2);
  }
  if (isTRNMask(ShuffleMask, VT, WhichResult)) {
    unsigned Opc = (WhichResult == 0) ? AArch64ISD::TRN1 : AArch64ISD::TRN2;
    return DAG.getNode(Opc, dl, V1.getValueType(), V1, V2);
  }

  if (isZIP_v_undef_Mask(ShuffleMask, VT, WhichResult)) {
    unsigned Opc = (WhichResult == 0) ? AArch64ISD::ZIP1 : AArch64ISD::ZIP2;
    return DAG.getNode(Opc, dl, V1.getValueType(), V1, V1);
  }
  if (isUZP_v_undef_Mask(ShuffleMask, VT, WhichResult)) {
    unsigned Opc = (WhichResult == 0) ? AArch64ISD::UZP1 : AArch64ISD::UZP2;
    return DAG.getNode(Opc, dl, V1.getValueType(), V1, V1);
  }
  if (isTRN_v_undef_Mask(ShuffleMask, VT, WhichResult)) {
    unsigned Opc = (WhichResult == 0) ? AArch64ISD::TRN1 : AArch64ISD::TRN2;
    return DAG.getNode(Opc, dl, V1.getValueType(), V1, V1);
  }

  SDValue Concat = tryFormConcatFromShuffle(Op, DAG);
  if (Concat.getNode())
    return Concat;

  bool DstIsLeft;
  int Anomaly;
  int NumInputElements = V1.getValueType().getVectorNumElements();
  if (isINSMask(ShuffleMask, NumInputElements, DstIsLeft, Anomaly)) {
    SDValue DstVec = DstIsLeft ? V1 : V2;
    SDValue DstLaneV = DAG.getConstant(Anomaly, MVT::i64);

    SDValue SrcVec = V1;
    int SrcLane = ShuffleMask[Anomaly];
    if (SrcLane >= NumInputElements) {
      SrcVec = V2;
      SrcLane -= VT.getVectorNumElements();
    }
    SDValue SrcLaneV = DAG.getConstant(SrcLane, MVT::i64);

    EVT ScalarVT = VT.getVectorElementType();

    if (ScalarVT.getSizeInBits() < 32 && ScalarVT.isInteger())
      ScalarVT = MVT::i32;

    return DAG.getNode(
        ISD::INSERT_VECTOR_ELT, dl, VT, DstVec,
        DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, ScalarVT, SrcVec, SrcLaneV),
        DstLaneV);
  }

  // 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);
  }

  return GenerateTBL(Op, ShuffleMask, DAG);
}

static bool resolveBuildVector(BuildVectorSDNode *BVN, APInt &CnstBits,
                               APInt &UndefBits) {
  EVT VT = BVN->getValueType(0);
  APInt SplatBits, SplatUndef;
  unsigned SplatBitSize;
  bool HasAnyUndefs;
  if (BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
    unsigned NumSplats = VT.getSizeInBits() / SplatBitSize;

    for (unsigned i = 0; i < NumSplats; ++i) {
      CnstBits <<= SplatBitSize;
      UndefBits <<= SplatBitSize;
      CnstBits |= SplatBits.zextOrTrunc(VT.getSizeInBits());
      UndefBits |= (SplatBits ^ SplatUndef).zextOrTrunc(VT.getSizeInBits());
    }

    return true;
  }

  return false;
}

SDValue AArch64TargetLowering::LowerVectorAND(SDValue Op,
                                              SelectionDAG &DAG) const {
  BuildVectorSDNode *BVN =
      dyn_cast<BuildVectorSDNode>(Op.getOperand(1).getNode());
  SDValue LHS = Op.getOperand(0);
  SDLoc dl(Op);
  EVT VT = Op.getValueType();

  if (!BVN)
    return Op;

  APInt CnstBits(VT.getSizeInBits(), 0);
  APInt UndefBits(VT.getSizeInBits(), 0);
  if (resolveBuildVector(BVN, CnstBits, UndefBits)) {
    // We only have BIC vector immediate instruction, which is and-not.
    CnstBits = ~CnstBits;

    // We make use of a little bit of goto ickiness in order to avoid having to
    // duplicate the immediate matching logic for the undef toggled case.
    bool SecondTry = false;
  AttemptModImm:

    if (CnstBits.getHiBits(64) == CnstBits.getLoBits(64)) {
      CnstBits = CnstBits.zextOrTrunc(64);
      uint64_t CnstVal = CnstBits.getZExtValue();

      if (AArch64_AM::isAdvSIMDModImmType1(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType1(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::BICi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType2(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType2(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::BICi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType3(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType3(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::BICi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(16, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType4(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType4(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::BICi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(24, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType5(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType5(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::BICi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType6(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType6(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::BICi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }
    }

    if (SecondTry)
      goto FailedModImm;
    SecondTry = true;
    CnstBits = ~UndefBits;
    goto AttemptModImm;
  }

// We can always fall back to a non-immediate AND.
FailedModImm:
  return Op;
}

// Specialized code to quickly find if PotentialBVec is a BuildVector that
// consists of only the same constant int value, returned in reference arg
// ConstVal
static bool isAllConstantBuildVector(const SDValue &PotentialBVec,
                                     uint64_t &ConstVal) {
  BuildVectorSDNode *Bvec = dyn_cast<BuildVectorSDNode>(PotentialBVec);
  if (!Bvec)
    return false;
  ConstantSDNode *FirstElt = dyn_cast<ConstantSDNode>(Bvec->getOperand(0));
  if (!FirstElt)
    return false;
  EVT VT = Bvec->getValueType(0);
  unsigned NumElts = VT.getVectorNumElements();
  for (unsigned i = 1; i < NumElts; ++i)
    if (dyn_cast<ConstantSDNode>(Bvec->getOperand(i)) != FirstElt)
      return false;
  ConstVal = FirstElt->getZExtValue();
  return true;
}

static unsigned getIntrinsicID(const SDNode *N) {
  unsigned Opcode = N->getOpcode();
  switch (Opcode) {
  default:
    return Intrinsic::not_intrinsic;
  case ISD::INTRINSIC_WO_CHAIN: {
    unsigned IID = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
    if (IID < Intrinsic::num_intrinsics)
      return IID;
    return Intrinsic::not_intrinsic;
  }
  }
}

// Attempt to form a vector S[LR]I from (or (and X, BvecC1), (lsl Y, C2)),
// to (SLI X, Y, C2), where X and Y have matching vector types, BvecC1 is a
// BUILD_VECTORs with constant element C1, C2 is a constant, and C1 == ~C2.
// Also, logical shift right -> sri, with the same structure.
static SDValue tryLowerToSLI(SDNode *N, SelectionDAG &DAG) {
  EVT VT = N->getValueType(0);

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

  SDLoc DL(N);

  // Is the first op an AND?
  const SDValue And = N->getOperand(0);
  if (And.getOpcode() != ISD::AND)
    return SDValue();

  // Is the second op an shl or lshr?
  SDValue Shift = N->getOperand(1);
  // This will have been turned into: AArch64ISD::VSHL vector, #shift
  // or AArch64ISD::VLSHR vector, #shift
  unsigned ShiftOpc = Shift.getOpcode();
  if ((ShiftOpc != AArch64ISD::VSHL && ShiftOpc != AArch64ISD::VLSHR))
    return SDValue();
  bool IsShiftRight = ShiftOpc == AArch64ISD::VLSHR;

  // Is the shift amount constant?
  ConstantSDNode *C2node = dyn_cast<ConstantSDNode>(Shift.getOperand(1));
  if (!C2node)
    return SDValue();

  // Is the and mask vector all constant?
  uint64_t C1;
  if (!isAllConstantBuildVector(And.getOperand(1), C1))
    return SDValue();

  // Is C1 == ~C2, taking into account how much one can shift elements of a
  // particular size?
  uint64_t C2 = C2node->getZExtValue();
  unsigned ElemSizeInBits = VT.getVectorElementType().getSizeInBits();
  if (C2 > ElemSizeInBits)
    return SDValue();
  unsigned ElemMask = (1 << ElemSizeInBits) - 1;
  if ((C1 & ElemMask) != (~C2 & ElemMask))
    return SDValue();

  SDValue X = And.getOperand(0);
  SDValue Y = Shift.getOperand(0);

  unsigned Intrin =
      IsShiftRight ? Intrinsic::aarch64_neon_vsri : Intrinsic::aarch64_neon_vsli;
  SDValue ResultSLI =
      DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, VT,
                  DAG.getConstant(Intrin, MVT::i32), X, Y, Shift.getOperand(1));

  DEBUG(dbgs() << "aarch64-lower: transformed: \n");
  DEBUG(N->dump(&DAG));
  DEBUG(dbgs() << "into: \n");
  DEBUG(ResultSLI->dump(&DAG));

  ++NumShiftInserts;
  return ResultSLI;
}

SDValue AArch64TargetLowering::LowerVectorOR(SDValue Op,
                                             SelectionDAG &DAG) const {
  // Attempt to form a vector S[LR]I from (or (and X, C1), (lsl Y, C2))
  if (EnableAArch64SlrGeneration) {
    SDValue Res = tryLowerToSLI(Op.getNode(), DAG);
    if (Res.getNode())
      return Res;
  }

  BuildVectorSDNode *BVN =
      dyn_cast<BuildVectorSDNode>(Op.getOperand(0).getNode());
  SDValue LHS = Op.getOperand(1);
  SDLoc dl(Op);
  EVT VT = Op.getValueType();

  // OR commutes, so try swapping the operands.
  if (!BVN) {
    LHS = Op.getOperand(0);
    BVN = dyn_cast<BuildVectorSDNode>(Op.getOperand(1).getNode());
  }
  if (!BVN)
    return Op;

  APInt CnstBits(VT.getSizeInBits(), 0);
  APInt UndefBits(VT.getSizeInBits(), 0);
  if (resolveBuildVector(BVN, CnstBits, UndefBits)) {
    // We make use of a little bit of goto ickiness in order to avoid having to
    // duplicate the immediate matching logic for the undef toggled case.
    bool SecondTry = false;
  AttemptModImm:

    if (CnstBits.getHiBits(64) == CnstBits.getLoBits(64)) {
      CnstBits = CnstBits.zextOrTrunc(64);
      uint64_t CnstVal = CnstBits.getZExtValue();

      if (AArch64_AM::isAdvSIMDModImmType1(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType1(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::ORRi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType2(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType2(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::ORRi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType3(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType3(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::ORRi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(16, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType4(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType4(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::ORRi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(24, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType5(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType5(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::ORRi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType6(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType6(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::ORRi, dl, MovTy, LHS,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }
    }

    if (SecondTry)
      goto FailedModImm;
    SecondTry = true;
    CnstBits = UndefBits;
    goto AttemptModImm;
  }

// We can always fall back to a non-immediate OR.
FailedModImm:
  return Op;
}

// Normalize the operands of BUILD_VECTOR. The value of constant operands will
// be truncated to fit element width.
static SDValue NormalizeBuildVector(SDValue Op,
                                    SelectionDAG &DAG) {
  assert(Op.getOpcode() == ISD::BUILD_VECTOR && "Unknown opcode!");
  SDLoc dl(Op);
  EVT VT = Op.getValueType();
  EVT EltTy= VT.getVectorElementType();

  if (EltTy.isFloatingPoint() || EltTy.getSizeInBits() > 16)
    return Op;

  SmallVector<SDValue, 16> Ops;
  for (unsigned I = 0, E = VT.getVectorNumElements(); I != E; ++I) {
    SDValue Lane = Op.getOperand(I);
    if (Lane.getOpcode() == ISD::Constant) {
      APInt LowBits(EltTy.getSizeInBits(),
                    cast<ConstantSDNode>(Lane)->getZExtValue());
      Lane = DAG.getConstant(LowBits.getZExtValue(), MVT::i32);
    }
    Ops.push_back(Lane);
  }
  return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
}

SDValue AArch64TargetLowering::LowerBUILD_VECTOR(SDValue Op,
                                                 SelectionDAG &DAG) const {
  SDLoc dl(Op);
  EVT VT = Op.getValueType();
  Op = NormalizeBuildVector(Op, DAG);
  BuildVectorSDNode *BVN = cast<BuildVectorSDNode>(Op.getNode());

  APInt CnstBits(VT.getSizeInBits(), 0);
  APInt UndefBits(VT.getSizeInBits(), 0);
  if (resolveBuildVector(BVN, CnstBits, UndefBits)) {
    // We make use of a little bit of goto ickiness in order to avoid having to
    // duplicate the immediate matching logic for the undef toggled case.
    bool SecondTry = false;
  AttemptModImm:

    if (CnstBits.getHiBits(64) == CnstBits.getLoBits(64)) {
      CnstBits = CnstBits.zextOrTrunc(64);
      uint64_t CnstVal = CnstBits.getZExtValue();

      // Certain magic vector constants (used to express things like NOT
      // and NEG) are passed through unmodified.  This allows codegen patterns
      // for these operations to match.  Special-purpose patterns will lower
      // these immediates to MOVIs if it proves necessary.
      if (VT.isInteger() && (CnstVal == 0 || CnstVal == ~0ULL))
        return Op;

      // The many faces of MOVI...
      if (AArch64_AM::isAdvSIMDModImmType10(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType10(CnstVal);
        if (VT.getSizeInBits() == 128) {
          SDValue Mov = DAG.getNode(AArch64ISD::MOVIedit, dl, MVT::v2i64,
                                    DAG.getConstant(CnstVal, MVT::i32));
          return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
        }

        // Support the V64 version via subregister insertion.
        SDValue Mov = DAG.getNode(AArch64ISD::MOVIedit, dl, MVT::f64,
                                  DAG.getConstant(CnstVal, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType1(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType1(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType2(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType2(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType3(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType3(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(16, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType4(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType4(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(24, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType5(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType5(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType6(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType6(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType7(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType7(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVImsl, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(264, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType8(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType8(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVImsl, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(272, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType9(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType9(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v16i8 : MVT::v8i8;
        SDValue Mov = DAG.getNode(AArch64ISD::MOVI, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      // The few faces of FMOV...
      if (AArch64_AM::isAdvSIMDModImmType11(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType11(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4f32 : MVT::v2f32;
        SDValue Mov = DAG.getNode(AArch64ISD::FMOV, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType12(CnstVal) &&
          VT.getSizeInBits() == 128) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType12(CnstVal);
        SDValue Mov = DAG.getNode(AArch64ISD::FMOV, dl, MVT::v2f64,
                                  DAG.getConstant(CnstVal, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      // The many faces of MVNI...
      CnstVal = ~CnstVal;
      if (AArch64_AM::isAdvSIMDModImmType1(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType1(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType2(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType2(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType3(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType3(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(16, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType4(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType4(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(24, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType5(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType5(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(0, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType6(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType6(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v8i16 : MVT::v4i16;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNIshift, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(8, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType7(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType7(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNImsl, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(264, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }

      if (AArch64_AM::isAdvSIMDModImmType8(CnstVal)) {
        CnstVal = AArch64_AM::encodeAdvSIMDModImmType8(CnstVal);
        MVT MovTy = (VT.getSizeInBits() == 128) ? MVT::v4i32 : MVT::v2i32;
        SDValue Mov = DAG.getNode(AArch64ISD::MVNImsl, dl, MovTy,
                                  DAG.getConstant(CnstVal, MVT::i32),
                                  DAG.getConstant(272, MVT::i32));
        return DAG.getNode(AArch64ISD::NVCAST, dl, VT, Mov);
      }
    }

    if (SecondTry)
      goto FailedModImm;
    SecondTry = true;
    CnstBits = UndefBits;
    goto AttemptModImm;
  }
FailedModImm:

  // Scan through the operands to find some interesting properties we can
  // exploit:
  //   1) If only one value is used, we can use a DUP, or
  //   2) if only the low element is not undef, we can just insert that, or
  //   3) if only one constant value is used (w/ some non-constant lanes),
  //      we can splat the constant value into the whole vector then fill
  //      in the non-constant lanes.
  //   4) FIXME: If different constant values are used, but we can intelligently
  //             select the values we'll be overwriting for the non-constant
  //             lanes such that we can directly materialize the vector
  //             some other way (MOVI, e.g.), we can be sneaky.
  unsigned NumElts = VT.getVectorNumElements();
  bool isOnlyLowElement = true;
  bool usesOnlyOneValue = true;
  bool usesOnlyOneConstantValue = true;
  bool isConstant = true;
  unsigned NumConstantLanes = 0;
  SDValue Value;
  SDValue ConstantValue;
  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;

    if (isa<ConstantSDNode>(V) || isa<ConstantFPSDNode>(V)) {
      ++NumConstantLanes;
      if (!ConstantValue.getNode())
        ConstantValue = V;
      else if (ConstantValue != V)
        usesOnlyOneConstantValue = false;
    }

    if (!Value.getNode())
      Value = V;
    else if (V != Value)
      usesOnlyOneValue = false;
  }

  if (!Value.getNode())
    return DAG.getUNDEF(VT);

  if (isOnlyLowElement)
    return DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value);

  // Use DUP for non-constant splats.  For f32 constant splats, reduce to
  // i32 and try again.
  if (usesOnlyOneValue) {
    if (!isConstant) {
      if (Value.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
          Value.getValueType() != VT)
        return DAG.getNode(AArch64ISD::DUP, dl, VT, Value);

      // This is actually a DUPLANExx operation, which keeps everything vectory.

      // DUPLANE works on 128-bit vectors, widen it if necessary.
      SDValue Lane = Value.getOperand(1);
      Value = Value.getOperand(0);
      if (Value.getValueType().getSizeInBits() == 64)
        Value = WidenVector(Value, DAG);

      unsigned Opcode = getDUPLANEOp(VT.getVectorElementType());
      return DAG.getNode(Opcode, dl, VT, Value, Lane);
    }

    if (VT.getVectorElementType().isFloatingPoint()) {
      SmallVector<SDValue, 8> Ops;
      MVT NewType =
          (VT.getVectorElementType() == MVT::f32) ? MVT::i32 : MVT::i64;
      for (unsigned i = 0; i < NumElts; ++i)
        Ops.push_back(DAG.getNode(ISD::BITCAST, dl, NewType, Op.getOperand(i)));
      EVT VecVT = EVT::getVectorVT(*DAG.getContext(), NewType, NumElts);
      SDValue Val = DAG.getNode(ISD::BUILD_VECTOR, dl, VecVT, Ops);
      Val = LowerBUILD_VECTOR(Val, DAG);
      if (Val.getNode())
        return DAG.getNode(ISD::BITCAST, dl, VT, Val);
    }
  }

  // If there was only one constant value used and for more than one lane,
  // start by splatting that value, then replace the non-constant lanes. This
  // is better than the default, which will perform a separate initialization
  // for each lane.
  if (NumConstantLanes > 0 && usesOnlyOneConstantValue) {
    SDValue Val = DAG.getNode(AArch64ISD::DUP, dl, VT, ConstantValue);
    // Now insert the non-constant lanes.
    for (unsigned i = 0; i < NumElts; ++i) {
      SDValue V = Op.getOperand(i);
      SDValue LaneIdx = DAG.getConstant(i, MVT::i64);
      if (!isa<ConstantSDNode>(V) && !isa<ConstantFPSDNode>(V)) {
        // Note that type legalization likely mucked about with the VT of the
        // source operand, so we may have to convert it here before inserting.
        Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, Val, V, LaneIdx);
      }
    }
    return 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;
  }

  // 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);
    SDValue Op0 = Op.getOperand(0);
    unsigned ElemSize = VT.getVectorElementType().getSizeInBits();
    unsigned i = 0;
    // For 32 and 64 bit types, use INSERT_SUBREG for lane zero to
    // a) Avoid a RMW dependency on the full vector register, and
    // b) Allow the register coalescer to fold away the copy if the
    //    value is already in an S or D register.
    if (Op0.getOpcode() != ISD::UNDEF && (ElemSize == 32 || ElemSize == 64)) {
      unsigned SubIdx = ElemSize == 32 ? AArch64::ssub : AArch64::dsub;
      MachineSDNode *N =
          DAG.getMachineNode(TargetOpcode::INSERT_SUBREG, dl, VT, Vec, Op0,
                             DAG.getTargetConstant(SubIdx, MVT::i32));
      Vec = SDValue(N, 0);
      ++i;
    }
    for (; i < NumElts; ++i) {
      SDValue V = Op.getOperand(i);
      if (V.getOpcode() == ISD::UNDEF)
        continue;
      SDValue LaneIdx = DAG.getConstant(i, MVT::i64);
      Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, Vec, V, LaneIdx);
    }
    return Vec;
  }

  // Just use the default expansion. We failed to find a better alternative.
  return SDValue();
}

SDValue AArch64TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op,
                                                      SelectionDAG &DAG) const {
  assert(Op.getOpcode() == ISD::INSERT_VECTOR_ELT && "Unknown opcode!");

  // Check for non-constant or out of range lane.
  EVT VT = Op.getOperand(0).getValueType();
  ConstantSDNode *CI = dyn_cast<ConstantSDNode>(Op.getOperand(2));
  if (!CI || CI->getZExtValue() >= VT.getVectorNumElements())
    return SDValue();


  // Insertion/extraction are legal for V128 types.
  if (VT == MVT::v16i8 || VT == MVT::v8i16 || VT == MVT::v4i32 ||
      VT == MVT::v2i64 || VT == MVT::v4f32 || VT == MVT::v2f64 ||
      VT == MVT::v8f16)
    return Op;

  if (VT != MVT::v8i8 && VT != MVT::v4i16 && VT != MVT::v2i32 &&
      VT != MVT::v1i64 && VT != MVT::v2f32 && VT != MVT::v4f16)
    return SDValue();

  // For V64 types, we perform insertion by expanding the value
  // to a V128 type and perform the insertion on that.
  SDLoc DL(Op);
  SDValue WideVec = WidenVector(Op.getOperand(0), DAG);
  EVT WideTy = WideVec.getValueType();

  SDValue Node = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, WideTy, WideVec,
                             Op.getOperand(1), Op.getOperand(2));
  // Re-narrow the resultant vector.
  return NarrowVector(Node, DAG);
}

SDValue
AArch64TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
                                               SelectionDAG &DAG) const {
  assert(Op.getOpcode() == ISD::EXTRACT_VECTOR_ELT && "Unknown opcode!");

  // Check for non-constant or out of range lane.
  EVT VT = Op.getOperand(0).getValueType();
  ConstantSDNode *CI = dyn_cast<ConstantSDNode>(Op.getOperand(1));
  if (!CI || CI->getZExtValue() >= VT.getVectorNumElements())
    return SDValue();


  // Insertion/extraction are legal for V128 types.
  if (VT == MVT::v16i8 || VT == MVT::v8i16 || VT == MVT::v4i32 ||
      VT == MVT::v2i64 || VT == MVT::v4f32 || VT == MVT::v2f64 ||
      VT == MVT::v8f16)
    return Op;

  if (VT != MVT::v8i8 && VT != MVT::v4i16 && VT != MVT::v2i32 &&
      VT != MVT::v1i64 && VT != MVT::v2f32 && VT != MVT::v4f16)
    return SDValue();

  // For V64 types, we perform extraction by expanding the value
  // to a V128 type and perform the extraction on that.
  SDLoc DL(Op);
  SDValue WideVec = WidenVector(Op.getOperand(0), DAG);
  EVT WideTy = WideVec.getValueType();

  EVT ExtrTy = WideTy.getVectorElementType();
  if (ExtrTy == MVT::i16 || ExtrTy == MVT::i8)
    ExtrTy = MVT::i32;

  // For extractions, we just return the result directly.
  return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ExtrTy, WideVec,
                     Op.getOperand(1));
}

SDValue AArch64TargetLowering::LowerEXTRACT_SUBVECTOR(SDValue Op,
                                                      SelectionDAG &DAG) const {
  EVT VT = Op.getOperand(0).getValueType();
  SDLoc dl(Op);
  // Just in case...
  if (!VT.isVector())
    return SDValue();

  ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(Op.getOperand(1));
  if (!Cst)
    return SDValue();
  unsigned Val = Cst->getZExtValue();

  unsigned Size = Op.getValueType().getSizeInBits();
  if (Val == 0) {
    switch (Size) {
    case 8:
      return DAG.getTargetExtractSubreg(AArch64::bsub, dl, Op.getValueType(),
                                        Op.getOperand(0));
    case 16:
      return DAG.getTargetExtractSubreg(AArch64::hsub, dl, Op.getValueType(),
                                        Op.getOperand(0));
    case 32:
      return DAG.getTargetExtractSubreg(AArch64::ssub, dl, Op.getValueType(),
                                        Op.getOperand(0));
    case 64:
      return DAG.getTargetExtractSubreg(AArch64::dsub, dl, Op.getValueType(),
                                        Op.getOperand(0));
    default:
      llvm_unreachable("Unexpected vector type in extract_subvector!");
    }
  }
  // If this is extracting the upper 64-bits of a 128-bit vector, we match
  // that directly.
  if (Size == 64 && Val * VT.getVectorElementType().getSizeInBits() == 64)
    return Op;

  return SDValue();
}

bool AArch64TargetLowering::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 DummyBool;
  int DummyInt;
  unsigned DummyUnsigned;

  return (ShuffleVectorSDNode::isSplatMask(&M[0], VT) || isREVMask(M, VT, 64) ||
          isREVMask(M, VT, 32) || isREVMask(M, VT, 16) ||
          isEXTMask(M, VT, DummyBool, DummyUnsigned) ||
          // isTBLMask(M, VT) || // FIXME: Port TBL support from ARM.
          isTRNMask(M, VT, DummyUnsigned) || isUZPMask(M, VT, DummyUnsigned) ||
          isZIPMask(M, VT, DummyUnsigned) ||
          isTRN_v_undef_Mask(M, VT, DummyUnsigned) ||
          isUZP_v_undef_Mask(M, VT, DummyUnsigned) ||
          isZIP_v_undef_Mask(M, VT, DummyUnsigned) ||
          isINSMask(M, VT.getVectorNumElements(), DummyBool, DummyInt) ||
          isConcatMask(M, VT, VT.getSizeInBits() == 128));
}

/// 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));
}

SDValue AArch64TargetLowering::LowerVectorSRA_SRL_SHL(SDValue Op,
                                                      SelectionDAG &DAG) const {
  EVT VT = Op.getValueType();
  SDLoc DL(Op);
  int64_t Cnt;

  if (!Op.getOperand(1).getValueType().isVector())
    return Op;
  unsigned EltSize = VT.getVectorElementType().getSizeInBits();

  switch (Op.getOpcode()) {
  default:
    llvm_unreachable("unexpected shift opcode");

  case ISD::SHL:
    if (isVShiftLImm(Op.getOperand(1), VT, false, Cnt) && Cnt < EltSize)
      return DAG.getNode(AArch64ISD::VSHL, SDLoc(Op), VT, Op.getOperand(0),
                         DAG.getConstant(Cnt, MVT::i32));
    return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, VT,
                       DAG.getConstant(Intrinsic::aarch64_neon_ushl, MVT::i32),
                       Op.getOperand(0), Op.getOperand(1));
  case ISD::SRA:
  case ISD::SRL:
    // Right shift immediate
    if (isVShiftRImm(Op.getOperand(1), VT, false, false, Cnt) &&
        Cnt < EltSize) {
      unsigned Opc =
          (Op.getOpcode() == ISD::SRA) ? AArch64ISD::VASHR : AArch64ISD::VLSHR;
      return DAG.getNode(Opc, SDLoc(Op), VT, Op.getOperand(0),
                         DAG.getConstant(Cnt, MVT::i32));
    }

    // Right shift register.  Note, there is not a shift right register
    // instruction, but the shift left register instruction takes a signed
    // value, where negative numbers specify a right shift.
    unsigned Opc = (Op.getOpcode() == ISD::SRA) ? Intrinsic::aarch64_neon_sshl
                                                : Intrinsic::aarch64_neon_ushl;
    // negate the shift amount
    SDValue NegShift = DAG.getNode(AArch64ISD::NEG, DL, VT, Op.getOperand(1));
    SDValue NegShiftLeft =
        DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, VT,
                    DAG.getConstant(Opc, MVT::i32), Op.getOperand(0), NegShift);
    return NegShiftLeft;
  }

  return SDValue();
}

static SDValue EmitVectorComparison(SDValue LHS, SDValue RHS,
                                    AArch64CC::CondCode CC, bool NoNans, EVT VT,
                                    SDLoc dl, SelectionDAG &DAG) {
  EVT SrcVT = LHS.getValueType();

  BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(RHS.getNode());
  APInt CnstBits(VT.getSizeInBits(), 0);
  APInt UndefBits(VT.getSizeInBits(), 0);
  bool IsCnst = BVN && resolveBuildVector(BVN, CnstBits, UndefBits);
  bool IsZero = IsCnst && (CnstBits == 0);

  if (SrcVT.getVectorElementType().isFloatingPoint()) {
    switch (CC) {
    default:
      return SDValue();
    case AArch64CC::NE: {
      SDValue Fcmeq;
      if (IsZero)
        Fcmeq = DAG.getNode(AArch64ISD::FCMEQz, dl, VT, LHS);
      else
        Fcmeq = DAG.getNode(AArch64ISD::FCMEQ, dl, VT, LHS, RHS);
      return DAG.getNode(AArch64ISD::NOT, dl, VT, Fcmeq);
    }
    case AArch64CC::EQ:
      if (IsZero)
        return DAG.getNode(AArch64ISD::FCMEQz, dl, VT, LHS);
      return DAG.getNode(AArch64ISD::FCMEQ, dl, VT, LHS, RHS);
    case AArch64CC::GE:
      if (IsZero)
        return DAG.getNode(AArch64ISD::FCMGEz, dl, VT, LHS);
      return DAG.getNode(AArch64ISD::FCMGE, dl, VT, LHS, RHS);
    case AArch64CC::GT:
      if (IsZero)
        return DAG.getNode(AArch64ISD::FCMGTz, dl, VT, LHS);
      return DAG.getNode(AArch64ISD::FCMGT, dl, VT, LHS, RHS);
    case AArch64CC::LS:
      if (IsZero)
        return DAG.getNode(AArch64ISD::FCMLEz, dl, VT, LHS);
      return DAG.getNode(AArch64ISD::FCMGE, dl, VT, RHS, LHS);
    case AArch64CC::LT:
      if (!NoNans)
        return SDValue();
    // If we ignore NaNs then we can use to the MI implementation.
    // Fallthrough.
    case AArch64CC::MI:
      if (IsZero)
        return DAG.getNode(AArch64ISD::FCMLTz, dl, VT, LHS);
      return DAG.getNode(AArch64ISD::FCMGT, dl, VT, RHS, LHS);
    }
  }

  switch (CC) {
  default:
    return SDValue();
  case AArch64CC::NE: {
    SDValue Cmeq;
    if (IsZero)
      Cmeq = DAG.getNode(AArch64ISD::CMEQz, dl, VT, LHS);
    else
      Cmeq = DAG.getNode(AArch64ISD::CMEQ, dl, VT, LHS, RHS);
    return DAG.getNode(AArch64ISD::NOT, dl, VT, Cmeq);
  }
  case AArch64CC::EQ:
    if (IsZero)
      return DAG.getNode(AArch64ISD::CMEQz, dl, VT, LHS);
    return DAG.getNode(AArch64ISD::CMEQ, dl, VT, LHS, RHS);
  case AArch64CC::GE:
    if (IsZero)
      return DAG.getNode(AArch64ISD::CMGEz, dl, VT, LHS);
    return DAG.getNode(AArch64ISD::CMGE, dl, VT, LHS, RHS);
  case AArch64CC::GT:
    if (IsZero)
      return DAG.getNode(AArch64ISD::CMGTz, dl, VT, LHS);
    return DAG.getNode(AArch64ISD::CMGT, dl, VT, LHS, RHS);
  case AArch64CC::LE:
    if (IsZero)
      return DAG.getNode(AArch64ISD::CMLEz, dl, VT, LHS);
    return DAG.getNode(AArch64ISD::CMGE, dl, VT, RHS, LHS);
  case AArch64CC::LS:
    return DAG.getNode(AArch64ISD::CMHS, dl, VT, RHS, LHS);
  case AArch64CC::LO:
    return DAG.getNode(AArch64ISD::CMHI, dl, VT, RHS, LHS);
  case AArch64CC::LT:
    if (IsZero)
      return DAG.getNode(AArch64ISD::CMLTz, dl, VT, LHS);
    return DAG.getNode(AArch64ISD::CMGT, dl, VT, RHS, LHS);
  case AArch64CC::HI:
    return DAG.getNode(AArch64ISD::CMHI, dl, VT, LHS, RHS);
  case AArch64CC::HS:
    return DAG.getNode(AArch64ISD::CMHS, dl, VT, LHS, RHS);
  }
}

SDValue AArch64TargetLowering::LowerVSETCC(SDValue Op,
                                           SelectionDAG &DAG) const {
  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
  SDValue LHS = Op.getOperand(0);
  SDValue RHS = Op.getOperand(1);
  SDLoc dl(Op);

  if (LHS.getValueType().getVectorElementType().isInteger()) {
    assert(LHS.getValueType() == RHS.getValueType());
    AArch64CC::CondCode AArch64CC = changeIntCCToAArch64CC(CC);
    return EmitVectorComparison(LHS, RHS, AArch64CC, false, Op.getValueType(),
                                dl, DAG);
  }

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

  // Unfortunately, the mapping of LLVM FP CC's onto AArch64 CC's isn't totally
  // clean.  Some of them require two branches to implement.
  AArch64CC::CondCode CC1, CC2;
  bool ShouldInvert;
  changeVectorFPCCToAArch64CC(CC, CC1, CC2, ShouldInvert);

  bool NoNaNs = getTargetMachine().Options.NoNaNsFPMath;
  SDValue Cmp =
      EmitVectorComparison(LHS, RHS, CC1, NoNaNs, Op.getValueType(), dl, DAG);
  if (!Cmp.getNode())
    return SDValue();

  if (CC2 != AArch64CC::AL) {
    SDValue Cmp2 =
        EmitVectorComparison(LHS, RHS, CC2, NoNaNs, Op.getValueType(), dl, DAG);
    if (!Cmp2.getNode())
      return SDValue();

    Cmp = DAG.getNode(ISD::OR, dl, Cmp.getValueType(), Cmp, Cmp2);
  }

  if (ShouldInvert)
    return Cmp = DAG.getNOT(dl, Cmp, Cmp.getValueType());

  return Cmp;
}

/// getTgtMemIntrinsic - Represent NEON load and store intrinsics as
/// MemIntrinsicNodes.  The associated MachineMemOperands record the alignment
/// specified in the intrinsic calls.
bool AArch64TargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
                                               const CallInst &I,
                                               unsigned Intrinsic) const {
  switch (Intrinsic) {
  case Intrinsic::aarch64_neon_ld2:
  case Intrinsic::aarch64_neon_ld3:
  case Intrinsic::aarch64_neon_ld4:
  case Intrinsic::aarch64_neon_ld1x2:
  case Intrinsic::aarch64_neon_ld1x3:
  case Intrinsic::aarch64_neon_ld1x4:
  case Intrinsic::aarch64_neon_ld2lane:
  case Intrinsic::aarch64_neon_ld3lane:
  case Intrinsic::aarch64_neon_ld4lane:
  case Intrinsic::aarch64_neon_ld2r:
  case Intrinsic::aarch64_neon_ld3r:
  case Intrinsic::aarch64_neon_ld4r: {
    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(I.getNumArgOperands() - 1);
    Info.offset = 0;
    Info.align = 0;
    Info.vol = false; // volatile loads with NEON intrinsics not supported
    Info.readMem = true;
    Info.writeMem = false;
    return true;
  }
  case Intrinsic::aarch64_neon_st2:
  case Intrinsic::aarch64_neon_st3:
  case Intrinsic::aarch64_neon_st4:
  case Intrinsic::aarch64_neon_st1x2:
  case Intrinsic::aarch64_neon_st1x3:
  case Intrinsic::aarch64_neon_st1x4:
  case Intrinsic::aarch64_neon_st2lane:
  case Intrinsic::aarch64_neon_st3lane:
  case Intrinsic::aarch64_neon_st4lane: {
    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(I.getNumArgOperands() - 1);
    Info.offset = 0;
    Info.align = 0;
    Info.vol = false; // volatile stores with NEON intrinsics not supported
    Info.readMem = false;
    Info.writeMem = true;
    return true;
  }
  case Intrinsic::aarch64_ldaxr:
  case Intrinsic::aarch64_ldxr: {
    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::aarch64_stlxr:
  case Intrinsic::aarch64_stxr: {
    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::aarch64_ldaxp:
  case Intrinsic::aarch64_ldxp: {
    Info.opc = ISD::INTRINSIC_W_CHAIN;
    Info.memVT = MVT::i128;
    Info.ptrVal = I.getArgOperand(0);
    Info.offset = 0;
    Info.align = 16;
    Info.vol = true;
    Info.readMem = true;
    Info.writeMem = false;
    return true;
  }
  case Intrinsic::aarch64_stlxp:
  case Intrinsic::aarch64_stxp: {
    Info.opc = ISD::INTRINSIC_W_CHAIN;
    Info.memVT = MVT::i128;
    Info.ptrVal = I.getArgOperand(2);
    Info.offset = 0;
    Info.align = 16;
    Info.vol = true;
    Info.readMem = false;
    Info.writeMem = true;
    return true;
  }
  default:
    break;
  }

  return false;
}

// Truncations from 64-bit GPR to 32-bit GPR is free.
bool AArch64TargetLowering::isTruncateFree(Type *Ty1, Type *Ty2) const {
  if (!Ty1->isIntegerTy() || !Ty2->isIntegerTy())
    return false;
  unsigned NumBits1 = Ty1->getPrimitiveSizeInBits();
  unsigned NumBits2 = Ty2->getPrimitiveSizeInBits();
  return NumBits1 > NumBits2;
}
bool AArch64TargetLowering::isTruncateFree(EVT VT1, EVT VT2) const {
  if (VT1.isVector() || VT2.isVector() || !VT1.isInteger() || !VT2.isInteger())
    return false;
  unsigned NumBits1 = VT1.getSizeInBits();
  unsigned NumBits2 = VT2.getSizeInBits();
  return NumBits1 > NumBits2;
}

// All 32-bit GPR operations implicitly zero the high-half of the corresponding
// 64-bit GPR.
bool AArch64TargetLowering::isZExtFree(Type *Ty1, Type *Ty2) const {
  if (!Ty1->isIntegerTy() || !Ty2->isIntegerTy())
    return false;
  unsigned NumBits1 = Ty1->getPrimitiveSizeInBits();
  unsigned NumBits2 = Ty2->getPrimitiveSizeInBits();
  return NumBits1 == 32 && NumBits2 == 64;
}
bool AArch64TargetLowering::isZExtFree(EVT VT1, EVT VT2) const {
  if (VT1.isVector() || VT2.isVector() || !VT1.isInteger() || !VT2.isInteger())
    return false;
  unsigned NumBits1 = VT1.getSizeInBits();
  unsigned NumBits2 = VT2.getSizeInBits();
  return NumBits1 == 32 && NumBits2 == 64;
}

bool AArch64TargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
  EVT VT1 = Val.getValueType();
  if (isZExtFree(VT1, VT2)) {
    return true;
  }

  if (Val.getOpcode() != ISD::LOAD)
    return false;

  // 8-, 16-, and 32-bit integer loads all implicitly zero-extend.
  return (VT1.isSimple() && !VT1.isVector() && VT1.isInteger() &&
          VT2.isSimple() && !VT2.isVector() && VT2.isInteger() &&
          VT1.getSizeInBits() <= 32);
}

bool AArch64TargetLowering::hasPairedLoad(Type *LoadedType,
                                          unsigned &RequiredAligment) const {
  if (!LoadedType->isIntegerTy() && !LoadedType->isFloatTy())
    return false;
  // Cyclone supports unaligned accesses.
  RequiredAligment = 0;
  unsigned NumBits = LoadedType->getPrimitiveSizeInBits();
  return NumBits == 32 || NumBits == 64;
}

bool AArch64TargetLowering::hasPairedLoad(EVT LoadedType,
                                          unsigned &RequiredAligment) const {
  if (!LoadedType.isSimple() ||
      (!LoadedType.isInteger() && !LoadedType.isFloatingPoint()))
    return false;
  // Cyclone supports unaligned accesses.
  RequiredAligment = 0;
  unsigned NumBits = LoadedType.getSizeInBits();
  return NumBits == 32 || NumBits == 64;
}

static bool memOpAlign(unsigned DstAlign, unsigned SrcAlign,
                       unsigned AlignCheck) {
  return ((SrcAlign == 0 || SrcAlign % AlignCheck == 0) &&
          (DstAlign == 0 || DstAlign % AlignCheck == 0));
}

EVT AArch64TargetLowering::getOptimalMemOpType(uint64_t Size, unsigned DstAlign,
                                               unsigned SrcAlign, bool IsMemset,
                                               bool ZeroMemset,
                                               bool MemcpyStrSrc,
                                               MachineFunction &MF) const {
  // Don't use AdvSIMD to implement 16-byte memset. It would have taken one
  // instruction to materialize the v2i64 zero and one store (with restrictive
  // addressing mode). Just do two i64 store of zero-registers.
  bool Fast;
  const Function *F = MF.getFunction();
  if (Subtarget->hasFPARMv8() && !IsMemset && Size >= 16 &&
      !F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
                                       Attribute::NoImplicitFloat) &&
      (memOpAlign(SrcAlign, DstAlign, 16) ||
       (allowsMisalignedMemoryAccesses(MVT::f128, 0, 1, &Fast) && Fast)))
    return MVT::f128;

  return Size >= 8 ? MVT::i64 : MVT::i32;
}

// 12-bit optionally shifted immediates are legal for adds.
bool AArch64TargetLowering::isLegalAddImmediate(int64_t Immed) const {
  if ((Immed >> 12) == 0 || ((Immed & 0xfff) == 0 && Immed >> 24 == 0))
    return true;
  return false;
}

// Integer comparisons are implemented with ADDS/SUBS, so the range of valid
// immediates is the same as for an add or a sub.
bool AArch64TargetLowering::isLegalICmpImmediate(int64_t Immed) const {
  if (Immed < 0)
    Immed *= -1;
  return isLegalAddImmediate(Immed);
}

/// isLegalAddressingMode - Return true if the addressing mode represented
/// by AM is legal for this target, for a load/store of the specified type.
bool AArch64TargetLowering::isLegalAddressingMode(const AddrMode &AM,
                                                  Type *Ty) const {
  // AArch64 has five basic addressing modes:
  //  reg
  //  reg + 9-bit signed offset
  //  reg + SIZE_IN_BYTES * 12-bit unsigned offset
  //  reg1 + reg2
  //  reg + SIZE_IN_BYTES * reg

  // No global is ever allowed as a base.
  if (AM.BaseGV)
    return false;

  // No reg+reg+imm addressing.
  if (AM.HasBaseReg && AM.BaseOffs && AM.Scale)
    return false;

  // check reg + imm case:
  // i.e., reg + 0, reg + imm9, reg + SIZE_IN_BYTES * uimm12
  uint64_t NumBytes = 0;
  if (Ty->isSized()) {
    uint64_t NumBits = getDataLayout()->getTypeSizeInBits(Ty);
    NumBytes = NumBits / 8;
    if (!isPowerOf2_64(NumBits))
      NumBytes = 0;
  }

  if (!AM.Scale) {
    int64_t Offset = AM.BaseOffs;

    // 9-bit signed offset
    if (Offset >= -(1LL << 9) && Offset <= (1LL << 9) - 1)
      return true;

    // 12-bit unsigned offset
    unsigned shift = Log2_64(NumBytes);
    if (NumBytes && Offset > 0 && (Offset / NumBytes) <= (1LL << 12) - 1 &&
        // Must be a multiple of NumBytes (NumBytes is a power of 2)
        (Offset >> shift) << shift == Offset)
      return true;
    return false;
  }

  // Check reg1 + SIZE_IN_BYTES * reg2 and reg1 + reg2

  if (!AM.Scale || AM.Scale == 1 ||
      (AM.Scale > 0 && (uint64_t)AM.Scale == NumBytes))
    return true;
  return false;
}

int AArch64TargetLowering::getScalingFactorCost(const AddrMode &AM,
                                                Type *Ty) const {
  // Scaling factors are not free at all.
  // Operands                     | Rt Latency
  // -------------------------------------------
  // Rt, [Xn, Xm]                 | 4
  // -------------------------------------------
  // Rt, [Xn, Xm, lsl #imm]       | Rn: 4 Rm: 5
  // Rt, [Xn, Wm, <extend> #imm]  |
  if (isLegalAddressingMode(AM, Ty))
    // Scale represents reg2 * scale, thus account for 1 if
    // it is not equal to 0 or 1.
    return AM.Scale != 0 && AM.Scale != 1;
  return -1;
}

bool AArch64TargetLowering::isFMAFasterThanFMulAndFAdd(EVT VT) const {
  VT = VT.getScalarType();

  if (!VT.isSimple())
    return false;

  switch (VT.getSimpleVT().SimpleTy) {
  case MVT::f32:
  case MVT::f64:
    return true;
  default:
    break;
  }

  return false;
}

const MCPhysReg *
AArch64TargetLowering::getScratchRegisters(CallingConv::ID) const {
  // LR is a callee-save register, but we must treat it as clobbered by any call
  // site. Hence we include LR in the scratch registers, which are in turn added
  // as implicit-defs for stackmaps and patchpoints.
  static const MCPhysReg ScratchRegs[] = {
    AArch64::X16, AArch64::X17, AArch64::LR, 0
  };
  return ScratchRegs;
}

bool
AArch64TargetLowering::isDesirableToCommuteWithShift(const SDNode *N) const {
  EVT VT = N->getValueType(0);
    // If N is unsigned bit extraction: ((x >> C) & mask), then do not combine
    // it with shift to let it be lowered to UBFX.
  if (N->getOpcode() == ISD::AND && (VT == MVT::i32 || VT == MVT::i64) &&
      isa<ConstantSDNode>(N->getOperand(1))) {
    uint64_t TruncMask = N->getConstantOperandVal(1);
    if (isMask_64(TruncMask) &&
      N->getOperand(0).getOpcode() == ISD::SRL &&
      isa<ConstantSDNode>(N->getOperand(0)->getOperand(1)))
      return false;
  }
  return true;
}

bool AArch64TargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
                                                              Type *Ty) const {
  assert(Ty->isIntegerTy());

  unsigned BitSize = Ty->getPrimitiveSizeInBits();
  if (BitSize == 0)
    return false;

  int64_t Val = Imm.getSExtValue();
  if (Val == 0 || AArch64_AM::isLogicalImmediate(Val, BitSize))
    return true;

  if ((int64_t)Val < 0)
    Val = ~Val;
  if (BitSize == 32)
    Val &= (1LL << 32) - 1;

  unsigned LZ = countLeadingZeros((uint64_t)Val);
  unsigned Shift = (63 - LZ) / 16;
  // MOVZ is free so return true for one or fewer MOVK.
  return (Shift < 3) ? true : false;
}

// Generate SUBS and CSEL for integer abs.
static SDValue performIntegerAbsCombine(SDNode *N, SelectionDAG &DAG) {
  EVT VT = N->getValueType(0);

  SDValue N0 = N->getOperand(0);
  SDValue N1 = N->getOperand(1);
  SDLoc DL(N);

  // Check pattern of XOR(ADD(X,Y), Y) where Y is SRA(X, size(X)-1)
  // and change it to SUB and CSEL.
  if (VT.isInteger() && N->getOpcode() == ISD::XOR &&
      N0.getOpcode() == ISD::ADD && N0.getOperand(1) == N1 &&
      N1.getOpcode() == ISD::SRA && N1.getOperand(0) == N0.getOperand(0))
    if (ConstantSDNode *Y1C = dyn_cast<ConstantSDNode>(N1.getOperand(1)))
      if (Y1C->getAPIntValue() == VT.getSizeInBits() - 1) {
        SDValue Neg = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT),
                                  N0.getOperand(0));
        // Generate SUBS & CSEL.
        SDValue Cmp =
            DAG.getNode(AArch64ISD::SUBS, DL, DAG.getVTList(VT, MVT::i32),
                        N0.getOperand(0), DAG.getConstant(0, VT));
        return DAG.getNode(AArch64ISD::CSEL, DL, VT, N0.getOperand(0), Neg,
                           DAG.getConstant(AArch64CC::PL, MVT::i32),
                           SDValue(Cmp.getNode(), 1));
      }
  return SDValue();
}

// performXorCombine - Attempts to handle integer ABS.
static SDValue performXorCombine(SDNode *N, SelectionDAG &DAG,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const AArch64Subtarget *Subtarget) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();

  return performIntegerAbsCombine(N, DAG);
}

SDValue
AArch64TargetLowering::BuildSDIVPow2(SDNode *N, const APInt &Divisor,
                                     SelectionDAG &DAG,
                                     std::vector<SDNode *> *Created) const {
  // fold (sdiv X, pow2)
  EVT VT = N->getValueType(0);
  if ((VT != MVT::i32 && VT != MVT::i64) ||
      !(Divisor.isPowerOf2() || (-Divisor).isPowerOf2()))
    return SDValue();

  SDLoc DL(N);
  SDValue N0 = N->getOperand(0);
  unsigned Lg2 = Divisor.countTrailingZeros();
  SDValue Zero = DAG.getConstant(0, VT);
  SDValue Pow2MinusOne = DAG.getConstant((1ULL << Lg2) - 1, VT);

  // Add (N0 < 0) ? Pow2 - 1 : 0;
  SDValue CCVal;
  SDValue Cmp = getAArch64Cmp(N0, Zero, ISD::SETLT, CCVal, DAG, DL);
  SDValue Add = DAG.getNode(ISD::ADD, DL, VT, N0, Pow2MinusOne);
  SDValue CSel = DAG.getNode(AArch64ISD::CSEL, DL, VT, Add, N0, CCVal, Cmp);

  if (Created) {
    Created->push_back(Cmp.getNode());
    Created->push_back(Add.getNode());
    Created->push_back(CSel.getNode());
  }

  // Divide by pow2.
  SDValue SRA =
      DAG.getNode(ISD::SRA, DL, VT, CSel, DAG.getConstant(Lg2, MVT::i64));

  // If we're dividing by a positive value, we're done.  Otherwise, we must
  // negate the result.
  if (Divisor.isNonNegative())
    return SRA;

  if (Created)
    Created->push_back(SRA.getNode());
  return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT), SRA);
}

static SDValue performMulCombine(SDNode *N, SelectionDAG &DAG,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const AArch64Subtarget *Subtarget) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();

  // Multiplication of a power of two plus/minus one can be done more
  // cheaply as as shift+add/sub. For now, this is true unilaterally. If
  // future CPUs have a cheaper MADD instruction, this may need to be
  // gated on a subtarget feature. For Cyclone, 32-bit MADD is 4 cycles and
  // 64-bit is 5 cycles, so this is always a win.
  if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1))) {
    APInt Value = C->getAPIntValue();
    EVT VT = N->getValueType(0);
    if (Value.isNonNegative()) {
      // (mul x, 2^N + 1) => (add (shl x, N), x)
      APInt VM1 = Value - 1;
      if (VM1.isPowerOf2()) {
        SDValue ShiftedVal =
            DAG.getNode(ISD::SHL, SDLoc(N), VT, N->getOperand(0),
                        DAG.getConstant(VM1.logBase2(), MVT::i64));
        return DAG.getNode(ISD::ADD, SDLoc(N), VT, ShiftedVal,
                           N->getOperand(0));
      }
      // (mul x, 2^N - 1) => (sub (shl x, N), x)
      APInt VP1 = Value + 1;
      if (VP1.isPowerOf2()) {
        SDValue ShiftedVal =
            DAG.getNode(ISD::SHL, SDLoc(N), VT, N->getOperand(0),
                        DAG.getConstant(VP1.logBase2(), MVT::i64));
        return DAG.getNode(ISD::SUB, SDLoc(N), VT, ShiftedVal,
                           N->getOperand(0));
      }
    } else {
      // (mul x, -(2^N + 1)) => - (add (shl x, N), x)
      APInt VNM1 = -Value - 1;
      if (VNM1.isPowerOf2()) {
        SDValue ShiftedVal =
            DAG.getNode(ISD::SHL, SDLoc(N), VT, N->getOperand(0),
                        DAG.getConstant(VNM1.logBase2(), MVT::i64));
        SDValue Add =
            DAG.getNode(ISD::ADD, SDLoc(N), VT, ShiftedVal, N->getOperand(0));
        return DAG.getNode(ISD::SUB, SDLoc(N), VT, DAG.getConstant(0, VT), Add);
      }
      // (mul x, -(2^N - 1)) => (sub x, (shl x, N))
      APInt VNP1 = -Value + 1;
      if (VNP1.isPowerOf2()) {
        SDValue ShiftedVal =
            DAG.getNode(ISD::SHL, SDLoc(N), VT, N->getOperand(0),
                        DAG.getConstant(VNP1.logBase2(), MVT::i64));
        return DAG.getNode(ISD::SUB, SDLoc(N), VT, N->getOperand(0),
                           ShiftedVal);
      }
    }
  }
  return SDValue();
}

static SDValue performVectorCompareAndMaskUnaryOpCombine(SDNode *N,
                                                         SelectionDAG &DAG) {
  // Take advantage of vector comparisons producing 0 or -1 in each lane to
  // optimize away operation when it's from a constant.
  //
  // The general transformation is:
  //    UNARYOP(AND(VECTOR_CMP(x,y), constant)) -->
  //       AND(VECTOR_CMP(x,y), constant2)
  //    constant2 = UNARYOP(constant)

  // Early exit if this isn't a vector operation, the operand of the
  // unary operation isn't a bitwise AND, or if the sizes of the operations
  // aren't the same.
  EVT VT = N->getValueType(0);
  if (!VT.isVector() || N->getOperand(0)->getOpcode() != ISD::AND ||
      N->getOperand(0)->getOperand(0)->getOpcode() != ISD::SETCC ||
      VT.getSizeInBits() != N->getOperand(0)->getValueType(0).getSizeInBits())
    return SDValue();

  // Now check that the other operand of the AND is a constant. We could
  // make the transformation for non-constant splats as well, but it's unclear
  // that would be a benefit as it would not eliminate any operations, just
  // perform one more step in scalar code before moving to the vector unit.
  if (BuildVectorSDNode *BV =
          dyn_cast<BuildVectorSDNode>(N->getOperand(0)->getOperand(1))) {
    // Bail out if the vector isn't a constant.
    if (!BV->isConstant())
      return SDValue();

    // Everything checks out. Build up the new and improved node.
    SDLoc DL(N);
    EVT IntVT = BV->getValueType(0);
    // Create a new constant of the appropriate type for the transformed
    // DAG.
    SDValue SourceConst = DAG.getNode(N->getOpcode(), DL, VT, SDValue(BV, 0));
    // The AND node needs bitcasts to/from an integer vector type around it.
    SDValue MaskConst = DAG.getNode(ISD::BITCAST, DL, IntVT, SourceConst);
    SDValue NewAnd = DAG.getNode(ISD::AND, DL, IntVT,
                                 N->getOperand(0)->getOperand(0), MaskConst);
    SDValue Res = DAG.getNode(ISD::BITCAST, DL, VT, NewAnd);
    return Res;
  }

  return SDValue();
}

static SDValue performIntToFpCombine(SDNode *N, SelectionDAG &DAG) {
  // First try to optimize away the conversion when it's conditionally from
  // a constant. Vectors only.
  SDValue Res = performVectorCompareAndMaskUnaryOpCombine(N, DAG);
  if (Res != SDValue())
    return Res;

  EVT VT = N->getValueType(0);
  if (VT != MVT::f32 && VT != MVT::f64)
    return SDValue();

  // Only optimize when the source and destination types have the same width.
  if (VT.getSizeInBits() != N->getOperand(0).getValueType().getSizeInBits())
    return SDValue();

  // If the result of an integer load is only used by an integer-to-float
  // conversion, use a fp load instead and a AdvSIMD scalar {S|U}CVTF instead.
  // This eliminates an "integer-to-vector-move UOP and improve throughput.
  SDValue N0 = N->getOperand(0);
  if (ISD::isNormalLoad(N0.getNode()) && N0.hasOneUse() &&
      // Do not change the width of a volatile load.
      !cast<LoadSDNode>(N0)->isVolatile()) {
    LoadSDNode *LN0 = cast<LoadSDNode>(N0);
    SDValue Load = DAG.getLoad(VT, SDLoc(N), LN0->getChain(), LN0->getBasePtr(),
                               LN0->getPointerInfo(), LN0->isVolatile(),
                               LN0->isNonTemporal(), LN0->isInvariant(),
                               LN0->getAlignment());

    // Make sure successors of the original load stay after it by updating them
    // to use the new Chain.
    DAG.ReplaceAllUsesOfValueWith(SDValue(LN0, 1), Load.getValue(1));

    unsigned Opcode =
        (N->getOpcode() == ISD::SINT_TO_FP) ? AArch64ISD::SITOF : AArch64ISD::UITOF;
    return DAG.getNode(Opcode, SDLoc(N), VT, Load);
  }

  return SDValue();
}

/// An EXTR instruction is made up of two shifts, ORed together. This helper
/// searches for and classifies those shifts.
static bool findEXTRHalf(SDValue N, SDValue &Src, uint32_t &ShiftAmount,
                         bool &FromHi) {
  if (N.getOpcode() == ISD::SHL)
    FromHi = false;
  else if (N.getOpcode() == ISD::SRL)
    FromHi = true;
  else
    return false;

  if (!isa<ConstantSDNode>(N.getOperand(1)))
    return false;

  ShiftAmount = N->getConstantOperandVal(1);
  Src = N->getOperand(0);
  return true;
}

/// EXTR instruction extracts a contiguous chunk of bits from two existing
/// registers viewed as a high/low pair. This function looks for the pattern:
/// (or (shl VAL1, #N), (srl VAL2, #RegWidth-N)) and replaces it with an
/// EXTR. Can't quite be done in TableGen because the two immediates aren't
/// independent.
static SDValue tryCombineToEXTR(SDNode *N,
                                TargetLowering::DAGCombinerInfo &DCI) {
  SelectionDAG &DAG = DCI.DAG;
  SDLoc DL(N);
  EVT VT = N->getValueType(0);

  assert(N->getOpcode() == ISD::OR && "Unexpected root");

  if (VT != MVT::i32 && VT != MVT::i64)
    return SDValue();

  SDValue LHS;
  uint32_t ShiftLHS = 0;
  bool LHSFromHi = 0;
  if (!findEXTRHalf(N->getOperand(0), LHS, ShiftLHS, LHSFromHi))
    return SDValue();

  SDValue RHS;
  uint32_t ShiftRHS = 0;
  bool RHSFromHi = 0;
  if (!findEXTRHalf(N->getOperand(1), RHS, ShiftRHS, RHSFromHi))
    return SDValue();

  // If they're both trying to come from the high part of the register, they're
  // not really an EXTR.
  if (LHSFromHi == RHSFromHi)
    return SDValue();

  if (ShiftLHS + ShiftRHS != VT.getSizeInBits())
    return SDValue();

  if (LHSFromHi) {
    std::swap(LHS, RHS);
    std::swap(ShiftLHS, ShiftRHS);
  }

  return DAG.getNode(AArch64ISD::EXTR, DL, VT, LHS, RHS,
                     DAG.getConstant(ShiftRHS, MVT::i64));
}

static SDValue tryCombineToBSL(SDNode *N,
                                TargetLowering::DAGCombinerInfo &DCI) {
  EVT VT = N->getValueType(0);
  SelectionDAG &DAG = DCI.DAG;
  SDLoc DL(N);

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

  SDValue N0 = N->getOperand(0);
  if (N0.getOpcode() != ISD::AND)
    return SDValue();

  SDValue N1 = N->getOperand(1);
  if (N1.getOpcode() != ISD::AND)
    return SDValue();

  // We only have to look for constant vectors here since the general, variable
  // case can be handled in TableGen.
  unsigned Bits = VT.getVectorElementType().getSizeInBits();
  uint64_t BitMask = Bits == 64 ? -1ULL : ((1ULL << Bits) - 1);
  for (int i = 1; i >= 0; --i)
    for (int j = 1; j >= 0; --j) {
      BuildVectorSDNode *BVN0 = dyn_cast<BuildVectorSDNode>(N0->getOperand(i));
      BuildVectorSDNode *BVN1 = dyn_cast<BuildVectorSDNode>(N1->getOperand(j));
      if (!BVN0 || !BVN1)
        continue;

      bool FoundMatch = true;
      for (unsigned k = 0; k < VT.getVectorNumElements(); ++k) {
        ConstantSDNode *CN0 = dyn_cast<ConstantSDNode>(BVN0->getOperand(k));
        ConstantSDNode *CN1 = dyn_cast<ConstantSDNode>(BVN1->getOperand(k));
        if (!CN0 || !CN1 ||
            CN0->getZExtValue() != (BitMask & ~CN1->getZExtValue())) {
          FoundMatch = false;
          break;
        }
      }

      if (FoundMatch)
        return DAG.getNode(AArch64ISD::BSL, DL, VT, SDValue(BVN0, 0),
                           N0->getOperand(1 - i), N1->getOperand(1 - j));
    }

  return SDValue();
}

static SDValue performORCombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI,
                                const AArch64Subtarget *Subtarget) {
  // Attempt to form an EXTR from (or (shl VAL1, #N), (srl VAL2, #RegWidth-N))
  if (!EnableAArch64ExtrGeneration)
    return SDValue();
  SelectionDAG &DAG = DCI.DAG;
  EVT VT = N->getValueType(0);

  if (!DAG.getTargetLoweringInfo().isTypeLegal(VT))
    return SDValue();

  SDValue Res = tryCombineToEXTR(N, DCI);
  if (Res.getNode())
    return Res;

  Res = tryCombineToBSL(N, DCI);
  if (Res.getNode())
    return Res;

  return SDValue();
}

static SDValue performBitcastCombine(SDNode *N,
                                     TargetLowering::DAGCombinerInfo &DCI,
                                     SelectionDAG &DAG) {
  // Wait 'til after everything is legalized to try this. That way we have
  // legal vector types and such.
  if (DCI.isBeforeLegalizeOps())
    return SDValue();

  // Remove extraneous bitcasts around an extract_subvector.
  // For example,
  //    (v4i16 (bitconvert
  //             (extract_subvector (v2i64 (bitconvert (v8i16 ...)), (i64 1)))))
  //  becomes
  //    (extract_subvector ((v8i16 ...), (i64 4)))

  // Only interested in 64-bit vectors as the ultimate result.
  EVT VT = N->getValueType(0);
  if (!VT.isVector())
    return SDValue();
  if (VT.getSimpleVT().getSizeInBits() != 64)
    return SDValue();
  // Is the operand an extract_subvector starting at the beginning or halfway
  // point of the vector? A low half may also come through as an
  // EXTRACT_SUBREG, so look for that, too.
  SDValue Op0 = N->getOperand(0);
  if (Op0->getOpcode() != ISD::EXTRACT_SUBVECTOR &&
      !(Op0->isMachineOpcode() &&
        Op0->getMachineOpcode() == AArch64::EXTRACT_SUBREG))
    return SDValue();
  uint64_t idx = cast<ConstantSDNode>(Op0->getOperand(1))->getZExtValue();
  if (Op0->getOpcode() == ISD::EXTRACT_SUBVECTOR) {
    if (Op0->getValueType(0).getVectorNumElements() != idx && idx != 0)
      return SDValue();
  } else if (Op0->getMachineOpcode() == AArch64::EXTRACT_SUBREG) {
    if (idx != AArch64::dsub)
      return SDValue();
    // The dsub reference is equivalent to a lane zero subvector reference.
    idx = 0;
  }
  // Look through the bitcast of the input to the extract.
  if (Op0->getOperand(0)->getOpcode() != ISD::BITCAST)
    return SDValue();
  SDValue Source = Op0->getOperand(0)->getOperand(0);
  // If the source type has twice the number of elements as our destination
  // type, we know this is an extract of the high or low half of the vector.
  EVT SVT = Source->getValueType(0);
  if (SVT.getVectorNumElements() != VT.getVectorNumElements() * 2)
    return SDValue();

  DEBUG(dbgs() << "aarch64-lower: bitcast extract_subvector simplification\n");

  // Create the simplified form to just extract the low or high half of the
  // vector directly rather than bothering with the bitcasts.
  SDLoc dl(N);
  unsigned NumElements = VT.getVectorNumElements();
  if (idx) {
    SDValue HalfIdx = DAG.getConstant(NumElements, MVT::i64);
    return DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, Source, HalfIdx);
  } else {
    SDValue SubReg = DAG.getTargetConstant(AArch64::dsub, MVT::i32);
    return SDValue(DAG.getMachineNode(TargetOpcode::EXTRACT_SUBREG, dl, VT,
                                      Source, SubReg),
                   0);
  }
}

static SDValue performConcatVectorsCombine(SDNode *N,
                                           TargetLowering::DAGCombinerInfo &DCI,
                                           SelectionDAG &DAG) {
  // Wait 'til after everything is legalized to try this. That way we have
  // legal vector types and such.
  if (DCI.isBeforeLegalizeOps())
    return SDValue();

  SDLoc dl(N);
  EVT VT = N->getValueType(0);

  // If we see a (concat_vectors (v1x64 A), (v1x64 A)) it's really a vector
  // splat. The indexed instructions are going to be expecting a DUPLANE64, so
  // canonicalise to that.
  if (N->getOperand(0) == N->getOperand(1) && VT.getVectorNumElements() == 2) {
    assert(VT.getVectorElementType().getSizeInBits() == 64);
    return DAG.getNode(AArch64ISD::DUPLANE64, dl, VT,
                       WidenVector(N->getOperand(0), DAG),
                       DAG.getConstant(0, MVT::i64));
  }

  // Canonicalise concat_vectors so that the right-hand vector has as few
  // bit-casts as possible before its real operation. The primary matching
  // destination for these operations will be the narrowing "2" instructions,
  // which depend on the operation being performed on this right-hand vector.
  // For example,
  //    (concat_vectors LHS,  (v1i64 (bitconvert (v4i16 RHS))))
  // becomes
  //    (bitconvert (concat_vectors (v4i16 (bitconvert LHS)), RHS))

  SDValue Op1 = N->getOperand(1);
  if (Op1->getOpcode() != ISD::BITCAST)
    return SDValue();
  SDValue RHS = Op1->getOperand(0);
  MVT RHSTy = RHS.getValueType().getSimpleVT();
  // If the RHS is not a vector, this is not the pattern we're looking for.
  if (!RHSTy.isVector())
    return SDValue();

  DEBUG(dbgs() << "aarch64-lower: concat_vectors bitcast simplification\n");

  MVT ConcatTy = MVT::getVectorVT(RHSTy.getVectorElementType(),
                                  RHSTy.getVectorNumElements() * 2);
  return DAG.getNode(
      ISD::BITCAST, dl, VT,
      DAG.getNode(ISD::CONCAT_VECTORS, dl, ConcatTy,
                  DAG.getNode(ISD::BITCAST, dl, RHSTy, N->getOperand(0)), RHS));
}

static SDValue tryCombineFixedPointConvert(SDNode *N,
                                           TargetLowering::DAGCombinerInfo &DCI,
                                           SelectionDAG &DAG) {
  // Wait 'til after everything is legalized to try this. That way we have
  // legal vector types and such.
  if (DCI.isBeforeLegalizeOps())
    return SDValue();
  // Transform a scalar conversion of a value from a lane extract into a
  // lane extract of a vector conversion. E.g., from foo1 to foo2:
  // double foo1(int64x2_t a) { return vcvtd_n_f64_s64(a[1], 9); }
  // double foo2(int64x2_t a) { return vcvtq_n_f64_s64(a, 9)[1]; }
  //
  // The second form interacts better with instruction selection and the
  // register allocator to avoid cross-class register copies that aren't
  // coalescable due to a lane reference.

  // Check the operand and see if it originates from a lane extract.
  SDValue Op1 = N->getOperand(1);
  if (Op1.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
    // Yep, no additional predication needed. Perform the transform.
    SDValue IID = N->getOperand(0);
    SDValue Shift = N->getOperand(2);
    SDValue Vec = Op1.getOperand(0);
    SDValue Lane = Op1.getOperand(1);
    EVT ResTy = N->getValueType(0);
    EVT VecResTy;
    SDLoc DL(N);

    // The vector width should be 128 bits by the time we get here, even
    // if it started as 64 bits (the extract_vector handling will have
    // done so).
    assert(Vec.getValueType().getSizeInBits() == 128 &&
           "unexpected vector size on extract_vector_elt!");
    if (Vec.getValueType() == MVT::v4i32)
      VecResTy = MVT::v4f32;
    else if (Vec.getValueType() == MVT::v2i64)
      VecResTy = MVT::v2f64;
    else
      llvm_unreachable("unexpected vector type!");

    SDValue Convert =
        DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, VecResTy, IID, Vec, Shift);
    return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ResTy, Convert, Lane);
  }
  return SDValue();
}

// AArch64 high-vector "long" operations are formed by performing the non-high
// version on an extract_subvector of each operand which gets the high half:
//
//  (longop2 LHS, RHS) == (longop (extract_high LHS), (extract_high RHS))
//
// However, there are cases which don't have an extract_high explicitly, but
// have another operation that can be made compatible with one for free. For
// example:
//
//  (dupv64 scalar) --> (extract_high (dup128 scalar))
//
// This routine does the actual conversion of such DUPs, once outer routines
// have determined that everything else is in order.
static SDValue tryExtendDUPToExtractHigh(SDValue N, SelectionDAG &DAG) {
  // We can handle most types of duplicate, but the lane ones have an extra
  // operand saying *which* lane, so we need to know.
  bool IsDUPLANE;
  switch (N.getOpcode()) {
  case AArch64ISD::DUP:
    IsDUPLANE = false;
    break;
  case AArch64ISD::DUPLANE8:
  case AArch64ISD::DUPLANE16:
  case AArch64ISD::DUPLANE32:
  case AArch64ISD::DUPLANE64:
    IsDUPLANE = true;
    break;
  default:
    return SDValue();
  }

  MVT NarrowTy = N.getSimpleValueType();
  if (!NarrowTy.is64BitVector())
    return SDValue();

  MVT ElementTy = NarrowTy.getVectorElementType();
  unsigned NumElems = NarrowTy.getVectorNumElements();
  MVT NewDUPVT = MVT::getVectorVT(ElementTy, NumElems * 2);

  SDValue NewDUP;
  if (IsDUPLANE)
    NewDUP = DAG.getNode(N.getOpcode(), SDLoc(N), NewDUPVT, N.getOperand(0),
                         N.getOperand(1));
  else
    NewDUP = DAG.getNode(AArch64ISD::DUP, SDLoc(N), NewDUPVT, N.getOperand(0));

  return DAG.getNode(ISD::EXTRACT_SUBVECTOR, SDLoc(N.getNode()), NarrowTy,
                     NewDUP, DAG.getConstant(NumElems, MVT::i64));
}

static bool isEssentiallyExtractSubvector(SDValue N) {
  if (N.getOpcode() == ISD::EXTRACT_SUBVECTOR)
    return true;

  return N.getOpcode() == ISD::BITCAST &&
         N.getOperand(0).getOpcode() == ISD::EXTRACT_SUBVECTOR;
}

/// \brief Helper structure to keep track of ISD::SET_CC operands.
struct GenericSetCCInfo {
  const SDValue *Opnd0;
  const SDValue *Opnd1;
  ISD::CondCode CC;
};

/// \brief Helper structure to keep track of a SET_CC lowered into AArch64 code.
struct AArch64SetCCInfo {
  const SDValue *Cmp;
  AArch64CC::CondCode CC;
};

/// \brief Helper structure to keep track of SetCC information.
union SetCCInfo {
  GenericSetCCInfo Generic;
  AArch64SetCCInfo AArch64;
};

/// \brief Helper structure to be able to read SetCC information.  If set to
/// true, IsAArch64 field, Info is a AArch64SetCCInfo, otherwise Info is a
/// GenericSetCCInfo.
struct SetCCInfoAndKind {
  SetCCInfo Info;
  bool IsAArch64;
};

/// \brief Check whether or not \p Op is a SET_CC operation, either a generic or
/// an
/// AArch64 lowered one.
/// \p SetCCInfo is filled accordingly.
/// \post SetCCInfo is meanginfull only when this function returns true.
/// \return True when Op is a kind of SET_CC operation.
static bool isSetCC(SDValue Op, SetCCInfoAndKind &SetCCInfo) {
  // If this is a setcc, this is straight forward.
  if (Op.getOpcode() == ISD::SETCC) {
    SetCCInfo.Info.Generic.Opnd0 = &Op.getOperand(0);
    SetCCInfo.Info.Generic.Opnd1 = &Op.getOperand(1);
    SetCCInfo.Info.Generic.CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
    SetCCInfo.IsAArch64 = false;
    return true;
  }
  // Otherwise, check if this is a matching csel instruction.
  // In other words:
  // - csel 1, 0, cc
  // - csel 0, 1, !cc
  if (Op.getOpcode() != AArch64ISD::CSEL)
    return false;
  // Set the information about the operands.
  // TODO: we want the operands of the Cmp not the csel
  SetCCInfo.Info.AArch64.Cmp = &Op.getOperand(3);
  SetCCInfo.IsAArch64 = true;
  SetCCInfo.Info.AArch64.CC = static_cast<AArch64CC::CondCode>(
      cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue());

  // Check that the operands matches the constraints:
  // (1) Both operands must be constants.
  // (2) One must be 1 and the other must be 0.
  ConstantSDNode *TValue = dyn_cast<ConstantSDNode>(Op.getOperand(0));
  ConstantSDNode *FValue = dyn_cast<ConstantSDNode>(Op.getOperand(1));

  // Check (1).
  if (!TValue || !FValue)
    return false;

  // Check (2).
  if (!TValue->isOne()) {
    // Update the comparison when we are interested in !cc.
    std::swap(TValue, FValue);
    SetCCInfo.Info.AArch64.CC =
        AArch64CC::getInvertedCondCode(SetCCInfo.Info.AArch64.CC);
  }
  return TValue->isOne() && FValue->isNullValue();
}

// Returns true if Op is setcc or zext of setcc.
static bool isSetCCOrZExtSetCC(const SDValue& Op, SetCCInfoAndKind &Info) {
  if (isSetCC(Op, Info))
    return true;
  return ((Op.getOpcode() == ISD::ZERO_EXTEND) &&
    isSetCC(Op->getOperand(0), Info));
}

// The folding we want to perform is:
// (add x, [zext] (setcc cc ...) )
//   -->
// (csel x, (add x, 1), !cc ...)
//
// The latter will get matched to a CSINC instruction.
static SDValue performSetccAddFolding(SDNode *Op, SelectionDAG &DAG) {
  assert(Op && Op->getOpcode() == ISD::ADD && "Unexpected operation!");
  SDValue LHS = Op->getOperand(0);
  SDValue RHS = Op->getOperand(1);
  SetCCInfoAndKind InfoAndKind;

  // If neither operand is a SET_CC, give up.
  if (!isSetCCOrZExtSetCC(LHS, InfoAndKind)) {
    std::swap(LHS, RHS);
    if (!isSetCCOrZExtSetCC(LHS, InfoAndKind))
      return SDValue();
  }

  // FIXME: This could be generatized to work for FP comparisons.
  EVT CmpVT = InfoAndKind.IsAArch64
                  ? InfoAndKind.Info.AArch64.Cmp->getOperand(0).getValueType()
                  : InfoAndKind.Info.Generic.Opnd0->getValueType();
  if (CmpVT != MVT::i32 && CmpVT != MVT::i64)
    return SDValue();

  SDValue CCVal;
  SDValue Cmp;
  SDLoc dl(Op);
  if (InfoAndKind.IsAArch64) {
    CCVal = DAG.getConstant(
        AArch64CC::getInvertedCondCode(InfoAndKind.Info.AArch64.CC), MVT::i32);
    Cmp = *InfoAndKind.Info.AArch64.Cmp;
  } else
    Cmp = getAArch64Cmp(*InfoAndKind.Info.Generic.Opnd0,
                      *InfoAndKind.Info.Generic.Opnd1,
                      ISD::getSetCCInverse(InfoAndKind.Info.Generic.CC, true),
                      CCVal, DAG, dl);

  EVT VT = Op->getValueType(0);
  LHS = DAG.getNode(ISD::ADD, dl, VT, RHS, DAG.getConstant(1, VT));
  return DAG.getNode(AArch64ISD::CSEL, dl, VT, RHS, LHS, CCVal, Cmp);
}

// The basic add/sub long vector instructions have variants with "2" on the end
// which act on the high-half of their inputs. They are normally matched by
// patterns like:
//
// (add (zeroext (extract_high LHS)),
//      (zeroext (extract_high RHS)))
// -> uaddl2 vD, vN, vM
//
// However, if one of the extracts is something like a duplicate, this
// instruction can still be used profitably. This function puts the DAG into a
// more appropriate form for those patterns to trigger.
static SDValue performAddSubLongCombine(SDNode *N,
                                        TargetLowering::DAGCombinerInfo &DCI,
                                        SelectionDAG &DAG) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();

  MVT VT = N->getSimpleValueType(0);
  if (!VT.is128BitVector()) {
    if (N->getOpcode() == ISD::ADD)
      return performSetccAddFolding(N, DAG);
    return SDValue();
  }

  // Make sure both branches are extended in the same way.
  SDValue LHS = N->getOperand(0);
  SDValue RHS = N->getOperand(1);
  if ((LHS.getOpcode() != ISD::ZERO_EXTEND &&
       LHS.getOpcode() != ISD::SIGN_EXTEND) ||
      LHS.getOpcode() != RHS.getOpcode())
    return SDValue();

  unsigned ExtType = LHS.getOpcode();

  // It's not worth doing if at least one of the inputs isn't already an
  // extract, but we don't know which it'll be so we have to try both.
  if (isEssentiallyExtractSubvector(LHS.getOperand(0))) {
    RHS = tryExtendDUPToExtractHigh(RHS.getOperand(0), DAG);
    if (!RHS.getNode())
      return SDValue();

    RHS = DAG.getNode(ExtType, SDLoc(N), VT, RHS);
  } else if (isEssentiallyExtractSubvector(RHS.getOperand(0))) {
    LHS = tryExtendDUPToExtractHigh(LHS.getOperand(0), DAG);
    if (!LHS.getNode())
      return SDValue();

    LHS = DAG.getNode(ExtType, SDLoc(N), VT, LHS);
  }

  return DAG.getNode(N->getOpcode(), SDLoc(N), VT, LHS, RHS);
}

// Massage DAGs which we can use the high-half "long" operations on into
// something isel will recognize better. E.g.
//
// (aarch64_neon_umull (extract_high vec) (dupv64 scalar)) -->
//   (aarch64_neon_umull (extract_high (v2i64 vec)))
//                     (extract_high (v2i64 (dup128 scalar)))))
//
static SDValue tryCombineLongOpWithDup(unsigned IID, SDNode *N,
                                       TargetLowering::DAGCombinerInfo &DCI,
                                       SelectionDAG &DAG) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();

  SDValue LHS = N->getOperand(1);
  SDValue RHS = N->getOperand(2);
  assert(LHS.getValueType().is64BitVector() &&
         RHS.getValueType().is64BitVector() &&
         "unexpected shape for long operation");

  // Either node could be a DUP, but it's not worth doing both of them (you'd
  // just as well use the non-high version) so look for a corresponding extract
  // operation on the other "wing".
  if (isEssentiallyExtractSubvector(LHS)) {
    RHS = tryExtendDUPToExtractHigh(RHS, DAG);
    if (!RHS.getNode())
      return SDValue();
  } else if (isEssentiallyExtractSubvector(RHS)) {
    LHS = tryExtendDUPToExtractHigh(LHS, DAG);
    if (!LHS.getNode())
      return SDValue();
  }

  return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N), N->getValueType(0),
                     N->getOperand(0), LHS, RHS);
}

static SDValue tryCombineShiftImm(unsigned IID, SDNode *N, SelectionDAG &DAG) {
  MVT ElemTy = N->getSimpleValueType(0).getScalarType();
  unsigned ElemBits = ElemTy.getSizeInBits();

  int64_t ShiftAmount;
  if (BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N->getOperand(2))) {
    APInt SplatValue, SplatUndef;
    unsigned SplatBitSize;
    bool HasAnyUndefs;
    if (!BVN->isConstantSplat(SplatValue, SplatUndef, SplatBitSize,
                              HasAnyUndefs, ElemBits) ||
        SplatBitSize != ElemBits)
      return SDValue();

    ShiftAmount = SplatValue.getSExtValue();
  } else if (ConstantSDNode *CVN = dyn_cast<ConstantSDNode>(N->getOperand(2))) {
    ShiftAmount = CVN->getSExtValue();
  } else
    return SDValue();

  unsigned Opcode;
  bool IsRightShift;
  switch (IID) {
  default:
    llvm_unreachable("Unknown shift intrinsic");
  case Intrinsic::aarch64_neon_sqshl:
    Opcode = AArch64ISD::SQSHL_I;
    IsRightShift = false;
    break;
  case Intrinsic::aarch64_neon_uqshl:
    Opcode = AArch64ISD::UQSHL_I;
    IsRightShift = false;
    break;
  case Intrinsic::aarch64_neon_srshl:
    Opcode = AArch64ISD::SRSHR_I;
    IsRightShift = true;
    break;
  case Intrinsic::aarch64_neon_urshl:
    Opcode = AArch64ISD::URSHR_I;
    IsRightShift = true;
    break;
  case Intrinsic::aarch64_neon_sqshlu:
    Opcode = AArch64ISD::SQSHLU_I;
    IsRightShift = false;
    break;
  }

  if (IsRightShift && ShiftAmount <= -1 && ShiftAmount >= -(int)ElemBits)
    return DAG.getNode(Opcode, SDLoc(N), N->getValueType(0), N->getOperand(1),
                       DAG.getConstant(-ShiftAmount, MVT::i32));
  else if (!IsRightShift && ShiftAmount >= 0 && ShiftAmount < ElemBits)
    return DAG.getNode(Opcode, SDLoc(N), N->getValueType(0), N->getOperand(1),
                       DAG.getConstant(ShiftAmount, MVT::i32));

  return SDValue();
}

// The CRC32[BH] instructions ignore the high bits of their data operand. Since
// the intrinsics must be legal and take an i32, this means there's almost
// certainly going to be a zext in the DAG which we can eliminate.
static SDValue tryCombineCRC32(unsigned Mask, SDNode *N, SelectionDAG &DAG) {
  SDValue AndN = N->getOperand(2);
  if (AndN.getOpcode() != ISD::AND)
    return SDValue();

  ConstantSDNode *CMask = dyn_cast<ConstantSDNode>(AndN.getOperand(1));
  if (!CMask || CMask->getZExtValue() != Mask)
    return SDValue();

  return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N), MVT::i32,
                     N->getOperand(0), N->getOperand(1), AndN.getOperand(0));
}

static SDValue performIntrinsicCombine(SDNode *N,
                                       TargetLowering::DAGCombinerInfo &DCI,
                                       const AArch64Subtarget *Subtarget) {
  SelectionDAG &DAG = DCI.DAG;
  unsigned IID = getIntrinsicID(N);
  switch (IID) {
  default:
    break;
  case Intrinsic::aarch64_neon_vcvtfxs2fp:
  case Intrinsic::aarch64_neon_vcvtfxu2fp:
    return tryCombineFixedPointConvert(N, DCI, DAG);
    break;
  case Intrinsic::aarch64_neon_fmax:
    return DAG.getNode(AArch64ISD::FMAX, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), N->getOperand(2));
  case Intrinsic::aarch64_neon_fmin:
    return DAG.getNode(AArch64ISD::FMIN, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), N->getOperand(2));
  case Intrinsic::aarch64_neon_smull:
  case Intrinsic::aarch64_neon_umull:
  case Intrinsic::aarch64_neon_pmull:
  case Intrinsic::aarch64_neon_sqdmull:
    return tryCombineLongOpWithDup(IID, N, DCI, DAG);
  case Intrinsic::aarch64_neon_sqshl:
  case Intrinsic::aarch64_neon_uqshl:
  case Intrinsic::aarch64_neon_sqshlu:
  case Intrinsic::aarch64_neon_srshl:
  case Intrinsic::aarch64_neon_urshl:
    return tryCombineShiftImm(IID, N, DAG);
  case Intrinsic::aarch64_crc32b:
  case Intrinsic::aarch64_crc32cb:
    return tryCombineCRC32(0xff, N, DAG);
  case Intrinsic::aarch64_crc32h:
  case Intrinsic::aarch64_crc32ch:
    return tryCombineCRC32(0xffff, N, DAG);
  }
  return SDValue();
}

static SDValue performExtendCombine(SDNode *N,
                                    TargetLowering::DAGCombinerInfo &DCI,
                                    SelectionDAG &DAG) {
  // If we see something like (zext (sabd (extract_high ...), (DUP ...))) then
  // we can convert that DUP into another extract_high (of a bigger DUP), which
  // helps the backend to decide that an sabdl2 would be useful, saving a real
  // extract_high operation.
  if (!DCI.isBeforeLegalizeOps() && N->getOpcode() == ISD::ZERO_EXTEND &&
      N->getOperand(0).getOpcode() == ISD::INTRINSIC_WO_CHAIN) {
    SDNode *ABDNode = N->getOperand(0).getNode();
    unsigned IID = getIntrinsicID(ABDNode);
    if (IID == Intrinsic::aarch64_neon_sabd ||
        IID == Intrinsic::aarch64_neon_uabd) {
      SDValue NewABD = tryCombineLongOpWithDup(IID, ABDNode, DCI, DAG);
      if (!NewABD.getNode())
        return SDValue();

      return DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N), N->getValueType(0),
                         NewABD);
    }
  }

  // This is effectively a custom type legalization for AArch64.
  //
  // Type legalization will split an extend of a small, legal, type to a larger
  // illegal type by first splitting the destination type, often creating
  // illegal source types, which then get legalized in isel-confusing ways,
  // leading to really terrible codegen. E.g.,
  //   %result = v8i32 sext v8i8 %value
  // becomes
  //   %losrc = extract_subreg %value, ...
  //   %hisrc = extract_subreg %value, ...
  //   %lo = v4i32 sext v4i8 %losrc
  //   %hi = v4i32 sext v4i8 %hisrc
  // Things go rapidly downhill from there.
  //
  // For AArch64, the [sz]ext vector instructions can only go up one element
  // size, so we can, e.g., extend from i8 to i16, but to go from i8 to i32
  // take two instructions.
  //
  // This implies that the most efficient way to do the extend from v8i8
  // to two v4i32 values is to first extend the v8i8 to v8i16, then do
  // the normal splitting to happen for the v8i16->v8i32.

  // This is pre-legalization to catch some cases where the default
  // type legalization will create ill-tempered code.
  if (!DCI.isBeforeLegalizeOps())
    return SDValue();

  // We're only interested in cleaning things up for non-legal vector types
  // here. If both the source and destination are legal, things will just
  // work naturally without any fiddling.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT ResVT = N->getValueType(0);
  if (!ResVT.isVector() || TLI.isTypeLegal(ResVT))
    return SDValue();
  // If the vector type isn't a simple VT, it's beyond the scope of what
  // we're  worried about here. Let legalization do its thing and hope for
  // the best.
  SDValue Src = N->getOperand(0);
  EVT SrcVT = Src->getValueType(0);
  if (!ResVT.isSimple() || !SrcVT.isSimple())
    return SDValue();

  // If the source VT is a 64-bit vector, we can play games and get the
  // better results we want.
  if (SrcVT.getSizeInBits() != 64)
    return SDValue();

  unsigned SrcEltSize = SrcVT.getVectorElementType().getSizeInBits();
  unsigned ElementCount = SrcVT.getVectorNumElements();
  SrcVT = MVT::getVectorVT(MVT::getIntegerVT(SrcEltSize * 2), ElementCount);
  SDLoc DL(N);
  Src = DAG.getNode(N->getOpcode(), DL, SrcVT, Src);

  // Now split the rest of the operation into two halves, each with a 64
  // bit source.
  EVT LoVT, HiVT;
  SDValue Lo, Hi;
  unsigned NumElements = ResVT.getVectorNumElements();
  assert(!(NumElements & 1) && "Splitting vector, but not in half!");
  LoVT = HiVT = EVT::getVectorVT(*DAG.getContext(),
                                 ResVT.getVectorElementType(), NumElements / 2);

  EVT InNVT = EVT::getVectorVT(*DAG.getContext(), SrcVT.getVectorElementType(),
                               LoVT.getVectorNumElements());
  Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, Src,
                   DAG.getIntPtrConstant(0));
  Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, Src,
                   DAG.getIntPtrConstant(InNVT.getVectorNumElements()));
  Lo = DAG.getNode(N->getOpcode(), DL, LoVT, Lo);
  Hi = DAG.getNode(N->getOpcode(), DL, HiVT, Hi);

  // Now combine the parts back together so we still have a single result
  // like the combiner expects.
  return DAG.getNode(ISD::CONCAT_VECTORS, DL, ResVT, Lo, Hi);
}

/// Replace a splat of a scalar to a vector store by scalar stores of the scalar
/// value. The load store optimizer pass will merge them to store pair stores.
/// This has better performance than a splat of the scalar followed by a split
/// vector store. Even if the stores are not merged it is four stores vs a dup,
/// followed by an ext.b and two stores.
static SDValue replaceSplatVectorStore(SelectionDAG &DAG, StoreSDNode *St) {
  SDValue StVal = St->getValue();
  EVT VT = StVal.getValueType();

  // Don't replace floating point stores, they possibly won't be transformed to
  // stp because of the store pair suppress pass.
  if (VT.isFloatingPoint())
    return SDValue();

  // Check for insert vector elements.
  if (StVal.getOpcode() != ISD::INSERT_VECTOR_ELT)
    return SDValue();

  // We can express a splat as store pair(s) for 2 or 4 elements.
  unsigned NumVecElts = VT.getVectorNumElements();
  if (NumVecElts != 4 && NumVecElts != 2)
    return SDValue();
  SDValue SplatVal = StVal.getOperand(1);
  unsigned RemainInsertElts = NumVecElts - 1;

  // Check that this is a splat.
  while (--RemainInsertElts) {
    SDValue NextInsertElt = StVal.getOperand(0);
    if (NextInsertElt.getOpcode() != ISD::INSERT_VECTOR_ELT)
      return SDValue();
    if (NextInsertElt.getOperand(1) != SplatVal)
      return SDValue();
    StVal = NextInsertElt;
  }
  unsigned OrigAlignment = St->getAlignment();
  unsigned EltOffset = NumVecElts == 4 ? 4 : 8;
  unsigned Alignment = std::min(OrigAlignment, EltOffset);

  // Create scalar stores. This is at least as good as the code sequence for a
  // split unaligned store wich is a dup.s, ext.b, and two stores.
  // Most of the time the three stores should be replaced by store pair
  // instructions (stp).
  SDLoc DL(St);
  SDValue BasePtr = St->getBasePtr();
  SDValue NewST1 =
      DAG.getStore(St->getChain(), DL, SplatVal, BasePtr, St->getPointerInfo(),
                   St->isVolatile(), St->isNonTemporal(), St->getAlignment());

  unsigned Offset = EltOffset;
  while (--NumVecElts) {
    SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i64, BasePtr,
                                    DAG.getConstant(Offset, MVT::i64));
    NewST1 = DAG.getStore(NewST1.getValue(0), DL, SplatVal, OffsetPtr,
                          St->getPointerInfo(), St->isVolatile(),
                          St->isNonTemporal(), Alignment);
    Offset += EltOffset;
  }
  return NewST1;
}

static SDValue performSTORECombine(SDNode *N,
                                   TargetLowering::DAGCombinerInfo &DCI,
                                   SelectionDAG &DAG,
                                   const AArch64Subtarget *Subtarget) {
  if (!DCI.isBeforeLegalize())
    return SDValue();

  StoreSDNode *S = cast<StoreSDNode>(N);
  if (S->isVolatile())
    return SDValue();

  // Cyclone has bad performance on unaligned 16B stores when crossing line and
  // page boundries. We want to split such stores.
  if (!Subtarget->isCyclone())
    return SDValue();

  // Don't split at Oz.
  MachineFunction &MF = DAG.getMachineFunction();
  bool IsMinSize = MF.getFunction()->getAttributes().hasAttribute(
      AttributeSet::FunctionIndex, Attribute::MinSize);
  if (IsMinSize)
    return SDValue();

  SDValue StVal = S->getValue();
  EVT VT = StVal.getValueType();

  // Don't split v2i64 vectors. Memcpy lowering produces those and splitting
  // those up regresses performance on micro-benchmarks and olden/bh.
  if (!VT.isVector() || VT.getVectorNumElements() < 2 || VT == MVT::v2i64)
    return SDValue();

  // Split unaligned 16B stores. They are terrible for performance.
  // Don't split stores with alignment of 1 or 2. Code that uses clang vector
  // extensions can use this to mark that it does not want splitting to happen
  // (by underspecifying alignment to be 1 or 2). Furthermore, the chance of
  // eliminating alignment hazards is only 1 in 8 for alignment of 2.
  if (VT.getSizeInBits() != 128 || S->getAlignment() >= 16 ||
      S->getAlignment() <= 2)
    return SDValue();

  // If we get a splat of a scalar convert this vector store to a store of
  // scalars. They will be merged into store pairs thereby removing two
  // instructions.
  SDValue ReplacedSplat = replaceSplatVectorStore(DAG, S);
  if (ReplacedSplat != SDValue())
    return ReplacedSplat;

  SDLoc DL(S);
  unsigned NumElts = VT.getVectorNumElements() / 2;
  // Split VT into two.
  EVT HalfVT =
      EVT::getVectorVT(*DAG.getContext(), VT.getVectorElementType(), NumElts);
  SDValue SubVector0 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, HalfVT, StVal,
                                   DAG.getIntPtrConstant(0));
  SDValue SubVector1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, HalfVT, StVal,
                                   DAG.getIntPtrConstant(NumElts));
  SDValue BasePtr = S->getBasePtr();
  SDValue NewST1 =
      DAG.getStore(S->getChain(), DL, SubVector0, BasePtr, S->getPointerInfo(),
                   S->isVolatile(), S->isNonTemporal(), S->getAlignment());
  SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i64, BasePtr,
                                  DAG.getConstant(8, MVT::i64));
  return DAG.getStore(NewST1.getValue(0), DL, SubVector1, OffsetPtr,
                      S->getPointerInfo(), S->isVolatile(), S->isNonTemporal(),
                      S->getAlignment());
}

/// Target-specific DAG combine function for post-increment LD1 (lane) and
/// post-increment LD1R.
static SDValue performPostLD1Combine(SDNode *N,
                                     TargetLowering::DAGCombinerInfo &DCI,
                                     bool IsLaneOp) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();

  SelectionDAG &DAG = DCI.DAG;
  EVT VT = N->getValueType(0);

  unsigned LoadIdx = IsLaneOp ? 1 : 0;
  SDNode *LD = N->getOperand(LoadIdx).getNode();
  // If it is not LOAD, can not do such combine.
  if (LD->getOpcode() != ISD::LOAD)
    return SDValue();

  LoadSDNode *LoadSDN = cast<LoadSDNode>(LD);
  EVT MemVT = LoadSDN->getMemoryVT();
  // Check if memory operand is the same type as the vector element.
  if (MemVT != VT.getVectorElementType())
    return SDValue();

  // Check if there are other uses. If so, do not combine as it will introduce
  // an extra load.
  for (SDNode::use_iterator UI = LD->use_begin(), UE = LD->use_end(); UI != UE;
       ++UI) {
    if (UI.getUse().getResNo() == 1) // Ignore uses of the chain result.
      continue;
    if (*UI != N)
      return SDValue();
  }

  SDValue Addr = LD->getOperand(1);
  SDValue Vector = N->getOperand(0);
  // 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.  Otherwise, folding it
    // would create a cycle.
    if (User->isPredecessorOf(LD) || LD->isPredecessorOf(User))
      continue;
    // Also check that add is not used in the vector operand.  This would also
    // create a cycle.
    if (User->isPredecessorOf(Vector.getNode()))
      continue;

    // 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())) {
      uint32_t IncVal = CInc->getZExtValue();
      unsigned NumBytes = VT.getScalarSizeInBits() / 8;
      if (IncVal != NumBytes)
        continue;
      Inc = DAG.getRegister(AArch64::XZR, MVT::i64);
    }

    SmallVector<SDValue, 8> Ops;
    Ops.push_back(LD->getOperand(0));  // Chain
    if (IsLaneOp) {
      Ops.push_back(Vector);           // The vector to be inserted
      Ops.push_back(N->getOperand(2)); // The lane to be inserted in the vector
    }
    Ops.push_back(Addr);
    Ops.push_back(Inc);

    EVT Tys[3] = { VT, MVT::i64, MVT::Other };
    SDVTList SDTys = DAG.getVTList(Tys);
    unsigned NewOp = IsLaneOp ? AArch64ISD::LD1LANEpost : AArch64ISD::LD1DUPpost;
    SDValue UpdN = DAG.getMemIntrinsicNode(NewOp, SDLoc(N), SDTys, Ops,
                                           MemVT,
                                           LoadSDN->getMemOperand());

    // Update the uses.
    std::vector<SDValue> NewResults;
    NewResults.push_back(SDValue(LD, 0));             // The result of load
    NewResults.push_back(SDValue(UpdN.getNode(), 2)); // Chain
    DCI.CombineTo(LD, NewResults);
    DCI.CombineTo(N, SDValue(UpdN.getNode(), 0));     // Dup/Inserted Result
    DCI.CombineTo(User, SDValue(UpdN.getNode(), 1));  // Write back register

    break;
  }
  return SDValue();
}

/// Target-specific DAG combine function for NEON load/store intrinsics
/// to merge base address updates.
static SDValue performNEONPostLDSTCombine(SDNode *N,
                                          TargetLowering::DAGCombinerInfo &DCI,
                                          SelectionDAG &DAG) {
  if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
    return SDValue();

  unsigned AddrOpIdx = N->getNumOperands() - 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 IsStore = false;
    bool IsLaneOp = false;
    bool IsDupOp = false;
    unsigned NewOpc = 0;
    unsigned NumVecs = 0;
    unsigned IntNo = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
    switch (IntNo) {
    default: llvm_unreachable("unexpected intrinsic for Neon base update");
    case Intrinsic::aarch64_neon_ld2:       NewOpc = AArch64ISD::LD2post;
      NumVecs = 2; break;
    case Intrinsic::aarch64_neon_ld3:       NewOpc = AArch64ISD::LD3post;
      NumVecs = 3; break;
    case Intrinsic::aarch64_neon_ld4:       NewOpc = AArch64ISD::LD4post;
      NumVecs = 4; break;
    case Intrinsic::aarch64_neon_st2:       NewOpc = AArch64ISD::ST2post;
      NumVecs = 2; IsStore = true; break;
    case Intrinsic::aarch64_neon_st3:       NewOpc = AArch64ISD::ST3post;
      NumVecs = 3; IsStore = true; break;
    case Intrinsic::aarch64_neon_st4:       NewOpc = AArch64ISD::ST4post;
      NumVecs = 4; IsStore = true; break;
    case Intrinsic::aarch64_neon_ld1x2:     NewOpc = AArch64ISD::LD1x2post;
      NumVecs = 2; break;
    case Intrinsic::aarch64_neon_ld1x3:     NewOpc = AArch64ISD::LD1x3post;
      NumVecs = 3; break;
    case Intrinsic::aarch64_neon_ld1x4:     NewOpc = AArch64ISD::LD1x4post;
      NumVecs = 4; break;
    case Intrinsic::aarch64_neon_st1x2:     NewOpc = AArch64ISD::ST1x2post;
      NumVecs = 2; IsStore = true; break;
    case Intrinsic::aarch64_neon_st1x3:     NewOpc = AArch64ISD::ST1x3post;
      NumVecs = 3; IsStore = true; break;
    case Intrinsic::aarch64_neon_st1x4:     NewOpc = AArch64ISD::ST1x4post;
      NumVecs = 4; IsStore = true; break;
    case Intrinsic::aarch64_neon_ld2r:      NewOpc = AArch64ISD::LD2DUPpost;
      NumVecs = 2; IsDupOp = true; break;
    case Intrinsic::aarch64_neon_ld3r:      NewOpc = AArch64ISD::LD3DUPpost;
      NumVecs = 3; IsDupOp = true; break;
    case Intrinsic::aarch64_neon_ld4r:      NewOpc = AArch64ISD::LD4DUPpost;
      NumVecs = 4; IsDupOp = true; break;
    case Intrinsic::aarch64_neon_ld2lane:   NewOpc = AArch64ISD::LD2LANEpost;
      NumVecs = 2; IsLaneOp = true; break;
    case Intrinsic::aarch64_neon_ld3lane:   NewOpc = AArch64ISD::LD3LANEpost;
      NumVecs = 3; IsLaneOp = true; break;
    case Intrinsic::aarch64_neon_ld4lane:   NewOpc = AArch64ISD::LD4LANEpost;
      NumVecs = 4; IsLaneOp = true; break;
    case Intrinsic::aarch64_neon_st2lane:   NewOpc = AArch64ISD::ST2LANEpost;
      NumVecs = 2; IsStore = true; IsLaneOp = true; break;
    case Intrinsic::aarch64_neon_st3lane:   NewOpc = AArch64ISD::ST3LANEpost;
      NumVecs = 3; IsStore = true; IsLaneOp = true; break;
    case Intrinsic::aarch64_neon_st4lane:   NewOpc = AArch64ISD::ST4LANEpost;
      NumVecs = 4; IsStore = true; IsLaneOp = true; break;
    }

    EVT VecTy;
    if (IsStore)
      VecTy = N->getOperand(2).getValueType();
    else
      VecTy = N->getValueType(0);

    // 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())) {
      uint32_t IncVal = CInc->getZExtValue();
      unsigned NumBytes = NumVecs * VecTy.getSizeInBits() / 8;
      if (IsLaneOp || IsDupOp)
        NumBytes /= VecTy.getVectorNumElements();
      if (IncVal != NumBytes)
        continue;
      Inc = DAG.getRegister(AArch64::XZR, MVT::i64);
    }
    SmallVector<SDValue, 8> Ops;
    Ops.push_back(N->getOperand(0)); // Incoming chain
    // Load lane and store have vector list as input.
    if (IsLaneOp || IsStore)
      for (unsigned i = 2; i < AddrOpIdx; ++i)
        Ops.push_back(N->getOperand(i));
    Ops.push_back(Addr); // Base register
    Ops.push_back(Inc);

    // Return Types.
    EVT Tys[6];
    unsigned NumResultVecs = (IsStore ? 0 : NumVecs);
    unsigned n;
    for (n = 0; n < NumResultVecs; ++n)
      Tys[n] = VecTy;
    Tys[n++] = MVT::i64;  // Type of write back register
    Tys[n] = MVT::Other;  // Type of the chain
    SDVTList SDTys = DAG.getVTList(makeArrayRef(Tys, NumResultVecs + 2));

    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));
    DCI.CombineTo(N, NewResults);
    DCI.CombineTo(User, SDValue(UpdN.getNode(), NumResultVecs));

    break;
  }
  return SDValue();
}

// Checks to see if the value is the prescribed width and returns information
// about its extension mode.
static
bool checkValueWidth(SDValue V, unsigned width, ISD::LoadExtType &ExtType) {
  ExtType = ISD::NON_EXTLOAD;
  switch(V.getNode()->getOpcode()) {
  default:
    return false;
  case ISD::LOAD: {
    LoadSDNode *LoadNode = cast<LoadSDNode>(V.getNode());
    if ((LoadNode->getMemoryVT() == MVT::i8 && width == 8)
       || (LoadNode->getMemoryVT() == MVT::i16 && width == 16)) {
      ExtType = LoadNode->getExtensionType();
      return true;
    }
    return false;
  }
  case ISD::AssertSext: {
    VTSDNode *TypeNode = cast<VTSDNode>(V.getNode()->getOperand(1));
    if ((TypeNode->getVT() == MVT::i8 && width == 8)
       || (TypeNode->getVT() == MVT::i16 && width == 16)) {
      ExtType = ISD::SEXTLOAD;
      return true;
    }
    return false;
  }
  case ISD::AssertZext: {
    VTSDNode *TypeNode = cast<VTSDNode>(V.getNode()->getOperand(1));
    if ((TypeNode->getVT() == MVT::i8 && width == 8)
       || (TypeNode->getVT() == MVT::i16 && width == 16)) {
      ExtType = ISD::ZEXTLOAD;
      return true;
    }
    return false;
  }
  case ISD::Constant:
  case ISD::TargetConstant: {
    if (std::abs(cast<ConstantSDNode>(V.getNode())->getSExtValue()) <
        1LL << (width - 1))
      return true;
    return false;
  }
  }

  return true;
}

// This function does a whole lot of voodoo to determine if the tests are
// equivalent without and with a mask. Essentially what happens is that given a
// DAG resembling:
//
//  +-------------+ +-------------+ +-------------+ +-------------+
//  |    Input    | | AddConstant | | CompConstant| |     CC      |
//  +-------------+ +-------------+ +-------------+ +-------------+
//           |           |           |               |
//           V           V           |    +----------+
//          +-------------+  +----+  |    |
//          |     ADD     |  |0xff|  |    |
//          +-------------+  +----+  |    |
//                  |           |    |    |
//                  V           V    |    |
//                 +-------------+   |    |
//                 |     AND     |   |    |
//                 +-------------+   |    |
//                      |            |    |
//                      +-----+      |    |
//                            |      |    |
//                            V      V    V
//                           +-------------+
//                           |     CMP     |
//                           +-------------+
//
// The AND node may be safely removed for some combinations of inputs. In
// particular we need to take into account the extension type of the Input,
// the exact values of AddConstant, CompConstant, and CC, along with the nominal
// width of the input (this can work for any width inputs, the above graph is
// specific to 8 bits.
//
// The specific equations were worked out by generating output tables for each
// AArch64CC value in terms of and AddConstant (w1), CompConstant(w2). The
// problem was simplified by working with 4 bit inputs, which means we only
// needed to reason about 24 distinct bit patterns: 8 patterns unique to zero
// extension (8,15), 8 patterns unique to sign extensions (-8,-1), and 8
// patterns present in both extensions (0,7). For every distinct set of
// AddConstant and CompConstants bit patterns we can consider the masked and
// unmasked versions to be equivalent if the result of this function is true for
// all 16 distinct bit patterns of for the current extension type of Input (w0).
//
//   sub      w8, w0, w1
//   and      w10, w8, #0x0f
//   cmp      w8, w2
//   cset     w9, AArch64CC
//   cmp      w10, w2
//   cset     w11, AArch64CC
//   cmp      w9, w11
//   cset     w0, eq
//   ret
//
// Since the above function shows when the outputs are equivalent it defines
// when it is safe to remove the AND. Unfortunately it only runs on AArch64 and
// would be expensive to run during compiles. The equations below were written
// in a test harness that confirmed they gave equivalent outputs to the above
// for all inputs function, so they can be used determine if the removal is
// legal instead.
//
// isEquivalentMaskless() is the code for testing if the AND can be removed
// factored out of the DAG recognition as the DAG can take several forms.

static
bool isEquivalentMaskless(unsigned CC, unsigned width,
                          ISD::LoadExtType ExtType, signed AddConstant,
                          signed CompConstant) {
  // By being careful about our equations and only writing the in term
  // symbolic values and well known constants (0, 1, -1, MaxUInt) we can
  // make them generally applicable to all bit widths.
  signed MaxUInt = (1 << width);

  // For the purposes of these comparisons sign extending the type is
  // equivalent to zero extending the add and displacing it by half the integer
  // width. Provided we are careful and make sure our equations are valid over
  // the whole range we can just adjust the input and avoid writing equations
  // for sign extended inputs.
  if (ExtType == ISD::SEXTLOAD)
    AddConstant -= (1 << (width-1));

  switch(CC) {
  case AArch64CC::LE:
  case AArch64CC::GT: {
    if ((AddConstant == 0) ||
        (CompConstant == MaxUInt - 1 && AddConstant < 0) ||
        (AddConstant >= 0 && CompConstant < 0) ||
        (AddConstant <= 0 && CompConstant <= 0 && CompConstant < AddConstant))
      return true;
  } break;
  case AArch64CC::LT:
  case AArch64CC::GE: {
    if ((AddConstant == 0) ||
        (AddConstant >= 0 && CompConstant <= 0) ||
        (AddConstant <= 0 && CompConstant <= 0 && CompConstant <= AddConstant))
      return true;
  } break;
  case AArch64CC::HI:
  case AArch64CC::LS: {
    if ((AddConstant >= 0 && CompConstant < 0) ||
       (AddConstant <= 0 && CompConstant >= -1 &&
        CompConstant < AddConstant + MaxUInt))
      return true;
  } break;
  case AArch64CC::PL:
  case AArch64CC::MI: {
    if ((AddConstant == 0) ||
        (AddConstant > 0 && CompConstant <= 0) ||
        (AddConstant < 0 && CompConstant <= AddConstant))
      return true;
  } break;
  case AArch64CC::LO:
  case AArch64CC::HS: {
    if ((AddConstant >= 0 && CompConstant <= 0) ||
        (AddConstant <= 0 && CompConstant >= 0 &&
         CompConstant <= AddConstant + MaxUInt))
      return true;
  } break;
  case AArch64CC::EQ:
  case AArch64CC::NE: {
    if ((AddConstant > 0 && CompConstant < 0) ||
        (AddConstant < 0 && CompConstant >= 0 &&
         CompConstant < AddConstant + MaxUInt) ||
        (AddConstant >= 0 && CompConstant >= 0 &&
         CompConstant >= AddConstant) ||
        (AddConstant <= 0 && CompConstant < 0 && CompConstant < AddConstant))

      return true;
  } break;
  case AArch64CC::VS:
  case AArch64CC::VC:
  case AArch64CC::AL:
  case AArch64CC::NV:
    return true;
  case AArch64CC::Invalid:
    break;
  }

  return false;
}

static
SDValue performCONDCombine(SDNode *N,
                           TargetLowering::DAGCombinerInfo &DCI,
                           SelectionDAG &DAG, unsigned CCIndex,
                           unsigned CmpIndex) {
  unsigned CC = cast<ConstantSDNode>(N->getOperand(CCIndex))->getSExtValue();
  SDNode *SubsNode = N->getOperand(CmpIndex).getNode();
  unsigned CondOpcode = SubsNode->getOpcode();

  if (CondOpcode != AArch64ISD::SUBS)
    return SDValue();

  // There is a SUBS feeding this condition. Is it fed by a mask we can
  // use?

  SDNode *AndNode = SubsNode->getOperand(0).getNode();
  unsigned MaskBits = 0;

  if (AndNode->getOpcode() != ISD::AND)
    return SDValue();

  if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(AndNode->getOperand(1))) {
    uint32_t CNV = CN->getZExtValue();
    if (CNV == 255)
      MaskBits = 8;
    else if (CNV == 65535)
      MaskBits = 16;
  }

  if (!MaskBits)
    return SDValue();

  SDValue AddValue = AndNode->getOperand(0);

  if (AddValue.getOpcode() != ISD::ADD)
    return SDValue();

  // The basic dag structure is correct, grab the inputs and validate them.

  SDValue AddInputValue1 = AddValue.getNode()->getOperand(0);
  SDValue AddInputValue2 = AddValue.getNode()->getOperand(1);
  SDValue SubsInputValue = SubsNode->getOperand(1);

  // The mask is present and the provenance of all the values is a smaller type,
  // lets see if the mask is superfluous.

  if (!isa<ConstantSDNode>(AddInputValue2.getNode()) ||
      !isa<ConstantSDNode>(SubsInputValue.getNode()))
    return SDValue();

  ISD::LoadExtType ExtType;

  if (!checkValueWidth(SubsInputValue, MaskBits, ExtType) ||
      !checkValueWidth(AddInputValue2, MaskBits, ExtType) ||
      !checkValueWidth(AddInputValue1, MaskBits, ExtType) )
    return SDValue();

  if(!isEquivalentMaskless(CC, MaskBits, ExtType,
                cast<ConstantSDNode>(AddInputValue2.getNode())->getSExtValue(),
                cast<ConstantSDNode>(SubsInputValue.getNode())->getSExtValue()))
    return SDValue();

  // The AND is not necessary, remove it.

  SDVTList VTs = DAG.getVTList(SubsNode->getValueType(0),
                               SubsNode->getValueType(1));
  SDValue Ops[] = { AddValue, SubsNode->getOperand(1) };

  SDValue NewValue = DAG.getNode(CondOpcode, SDLoc(SubsNode), VTs, Ops);
  DAG.ReplaceAllUsesWith(SubsNode, NewValue.getNode());

  return SDValue(N, 0);
}

// Optimize compare with zero and branch.
static SDValue performBRCONDCombine(SDNode *N,
                                    TargetLowering::DAGCombinerInfo &DCI,
                                    SelectionDAG &DAG) {
  SDValue NV = performCONDCombine(N, DCI, DAG, 2, 3);
  if (NV.getNode())
    N = NV.getNode();
  SDValue Chain = N->getOperand(0);
  SDValue Dest = N->getOperand(1);
  SDValue CCVal = N->getOperand(2);
  SDValue Cmp = N->getOperand(3);

  assert(isa<ConstantSDNode>(CCVal) && "Expected a ConstantSDNode here!");
  unsigned CC = cast<ConstantSDNode>(CCVal)->getZExtValue();
  if (CC != AArch64CC::EQ && CC != AArch64CC::NE)
    return SDValue();

  unsigned CmpOpc = Cmp.getOpcode();
  if (CmpOpc != AArch64ISD::ADDS && CmpOpc != AArch64ISD::SUBS)
    return SDValue();

  // Only attempt folding if there is only one use of the flag and no use of the
  // value.
  if (!Cmp->hasNUsesOfValue(0, 0) || !Cmp->hasNUsesOfValue(1, 1))
    return SDValue();

  SDValue LHS = Cmp.getOperand(0);
  SDValue RHS = Cmp.getOperand(1);

  assert(LHS.getValueType() == RHS.getValueType() &&
         "Expected the value type to be the same for both operands!");
  if (LHS.getValueType() != MVT::i32 && LHS.getValueType() != MVT::i64)
    return SDValue();

  if (isa<ConstantSDNode>(LHS) && cast<ConstantSDNode>(LHS)->isNullValue())
    std::swap(LHS, RHS);

  if (!isa<ConstantSDNode>(RHS) || !cast<ConstantSDNode>(RHS)->isNullValue())
    return SDValue();

  if (LHS.getOpcode() == ISD::SHL || LHS.getOpcode() == ISD::SRA ||
      LHS.getOpcode() == ISD::SRL)
    return SDValue();

  // Fold the compare into the branch instruction.
  SDValue BR;
  if (CC == AArch64CC::EQ)
    BR = DAG.getNode(AArch64ISD::CBZ, SDLoc(N), MVT::Other, Chain, LHS, Dest);
  else
    BR = DAG.getNode(AArch64ISD::CBNZ, SDLoc(N), MVT::Other, Chain, LHS, Dest);

  // Do not add new nodes to DAG combiner worklist.
  DCI.CombineTo(N, BR, false);

  return SDValue();
}

// vselect (v1i1 setcc) ->
//     vselect (v1iXX setcc)  (XX is the size of the compared operand type)
// FIXME: Currently the type legalizer can't handle VSELECT having v1i1 as
// condition. If it can legalize "VSELECT v1i1" correctly, no need to combine
// such VSELECT.
static SDValue performVSelectCombine(SDNode *N, SelectionDAG &DAG) {
  SDValue N0 = N->getOperand(0);
  EVT CCVT = N0.getValueType();

  if (N0.getOpcode() != ISD::SETCC || CCVT.getVectorNumElements() != 1 ||
      CCVT.getVectorElementType() != MVT::i1)
    return SDValue();

  EVT ResVT = N->getValueType(0);
  EVT CmpVT = N0.getOperand(0).getValueType();
  // Only combine when the result type is of the same size as the compared
  // operands.
  if (ResVT.getSizeInBits() != CmpVT.getSizeInBits())
    return SDValue();

  SDValue IfTrue = N->getOperand(1);
  SDValue IfFalse = N->getOperand(2);
  SDValue SetCC =
      DAG.getSetCC(SDLoc(N), CmpVT.changeVectorElementTypeToInteger(),
                   N0.getOperand(0), N0.getOperand(1),
                   cast<CondCodeSDNode>(N0.getOperand(2))->get());
  return DAG.getNode(ISD::VSELECT, SDLoc(N), ResVT, SetCC,
                     IfTrue, IfFalse);
}

/// A vector select: "(select vL, vR, (setcc LHS, RHS))" is best performed with
/// the compare-mask instructions rather than going via NZCV, even if LHS and
/// RHS are really scalar. This replaces any scalar setcc in the above pattern
/// with a vector one followed by a DUP shuffle on the result.
static SDValue performSelectCombine(SDNode *N, SelectionDAG &DAG) {
  SDValue N0 = N->getOperand(0);
  EVT ResVT = N->getValueType(0);

  if (N0.getOpcode() != ISD::SETCC || N0.getValueType() != MVT::i1)
    return SDValue();

  // If NumMaskElts == 0, the comparison is larger than select result. The
  // largest real NEON comparison is 64-bits per lane, which means the result is
  // at most 32-bits and an illegal vector. Just bail out for now.
  EVT SrcVT = N0.getOperand(0).getValueType();
  int NumMaskElts = ResVT.getSizeInBits() / SrcVT.getSizeInBits();
  if (!ResVT.isVector() || NumMaskElts == 0)
    return SDValue();

  SrcVT = EVT::getVectorVT(*DAG.getContext(), SrcVT, NumMaskElts);
  EVT CCVT = SrcVT.changeVectorElementTypeToInteger();

  // First perform a vector comparison, where lane 0 is the one we're interested
  // in.
  SDLoc DL(N0);
  SDValue LHS =
      DAG.getNode(ISD::SCALAR_TO_VECTOR, DL, SrcVT, N0.getOperand(0));
  SDValue RHS =
      DAG.getNode(ISD::SCALAR_TO_VECTOR, DL, SrcVT, N0.getOperand(1));
  SDValue SetCC = DAG.getNode(ISD::SETCC, DL, CCVT, LHS, RHS, N0.getOperand(2));

  // Now duplicate the comparison mask we want across all other lanes.
  SmallVector<int, 8> DUPMask(CCVT.getVectorNumElements(), 0);
  SDValue Mask = DAG.getVectorShuffle(CCVT, DL, SetCC, SetCC, DUPMask.data());
  Mask = DAG.getNode(ISD::BITCAST, DL,
                     ResVT.changeVectorElementTypeToInteger(), Mask);

  return DAG.getSelect(DL, ResVT, Mask, N->getOperand(1), N->getOperand(2));
}

SDValue AArch64TargetLowering::PerformDAGCombine(SDNode *N,
                                                 DAGCombinerInfo &DCI) const {
  SelectionDAG &DAG = DCI.DAG;
  switch (N->getOpcode()) {
  default:
    break;
  case ISD::ADD:
  case ISD::SUB:
    return performAddSubLongCombine(N, DCI, DAG);
  case ISD::XOR:
    return performXorCombine(N, DAG, DCI, Subtarget);
  case ISD::MUL:
    return performMulCombine(N, DAG, DCI, Subtarget);
  case ISD::SINT_TO_FP:
  case ISD::UINT_TO_FP:
    return performIntToFpCombine(N, DAG);
  case ISD::OR:
    return performORCombine(N, DCI, Subtarget);
  case ISD::INTRINSIC_WO_CHAIN:
    return performIntrinsicCombine(N, DCI, Subtarget);
  case ISD::ANY_EXTEND:
  case ISD::ZERO_EXTEND:
  case ISD::SIGN_EXTEND:
    return performExtendCombine(N, DCI, DAG);
  case ISD::BITCAST:
    return performBitcastCombine(N, DCI, DAG);
  case ISD::CONCAT_VECTORS:
    return performConcatVectorsCombine(N, DCI, DAG);
  case ISD::SELECT:
    return performSelectCombine(N, DAG);
  case ISD::VSELECT:
    return performVSelectCombine(N, DCI.DAG);
  case ISD::STORE:
    return performSTORECombine(N, DCI, DAG, Subtarget);
  case AArch64ISD::BRCOND:
    return performBRCONDCombine(N, DCI, DAG);
  case AArch64ISD::CSEL:
    return performCONDCombine(N, DCI, DAG, 2, 3);
  case AArch64ISD::DUP:
    return performPostLD1Combine(N, DCI, false);
  case ISD::INSERT_VECTOR_ELT:
    return performPostLD1Combine(N, DCI, true);
  case ISD::INTRINSIC_VOID:
  case ISD::INTRINSIC_W_CHAIN:
    switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
    case Intrinsic::aarch64_neon_ld2:
    case Intrinsic::aarch64_neon_ld3:
    case Intrinsic::aarch64_neon_ld4:
    case Intrinsic::aarch64_neon_ld1x2:
    case Intrinsic::aarch64_neon_ld1x3:
    case Intrinsic::aarch64_neon_ld1x4:
    case Intrinsic::aarch64_neon_ld2lane:
    case Intrinsic::aarch64_neon_ld3lane:
    case Intrinsic::aarch64_neon_ld4lane:
    case Intrinsic::aarch64_neon_ld2r:
    case Intrinsic::aarch64_neon_ld3r:
    case Intrinsic::aarch64_neon_ld4r:
    case Intrinsic::aarch64_neon_st2:
    case Intrinsic::aarch64_neon_st3:
    case Intrinsic::aarch64_neon_st4:
    case Intrinsic::aarch64_neon_st1x2:
    case Intrinsic::aarch64_neon_st1x3:
    case Intrinsic::aarch64_neon_st1x4:
    case Intrinsic::aarch64_neon_st2lane:
    case Intrinsic::aarch64_neon_st3lane:
    case Intrinsic::aarch64_neon_st4lane:
      return performNEONPostLDSTCombine(N, DCI, DAG);
    default:
      break;
    }
  }
  return SDValue();
}

// Check if the return value is used as only a return value, as otherwise
// we can't perform a tail-call. In particular, we need to check for
// target ISD nodes that are returns and any other "odd" constructs
// that the generic analysis code won't necessarily catch.
bool AArch64TargetLowering::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() != ISD::FP_EXTEND)
    return false;

  bool HasRet = false;
  for (SDNode *Node : Copy->uses()) {
    if (Node->getOpcode() != AArch64ISD::RET_FLAG)
      return false;
    HasRet = true;
  }

  if (!HasRet)
    return false;

  Chain = TCChain;
  return true;
}

// Return whether the an instruction can potentially be optimized to a tail
// call. This will cause the optimizers to attempt to move, or duplicate,
// return instructions to help enable tail call optimizations for this
// instruction.
bool AArch64TargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
  if (!CI->isTailCall())
    return false;

  return true;
}

bool AArch64TargetLowering::getIndexedAddressParts(SDNode *Op, SDValue &Base,
                                                   SDValue &Offset,
                                                   ISD::MemIndexedMode &AM,
                                                   bool &IsInc,
                                                   SelectionDAG &DAG) const {
  if (Op->getOpcode() != ISD::ADD && Op->getOpcode() != ISD::SUB)
    return false;

  Base = Op->getOperand(0);
  // All of the indexed addressing mode instructions take a signed
  // 9 bit immediate offset.
  if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Op->getOperand(1))) {
    int64_t RHSC = (int64_t)RHS->getZExtValue();
    if (RHSC >= 256 || RHSC <= -256)
      return false;
    IsInc = (Op->getOpcode() == ISD::ADD);
    Offset = Op->getOperand(1);
    return true;
  }
  return false;
}

bool AArch64TargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
                                                      SDValue &Offset,
                                                      ISD::MemIndexedMode &AM,
                                                      SelectionDAG &DAG) const {
  EVT VT;
  SDValue Ptr;
  if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
    VT = LD->getMemoryVT();
    Ptr = LD->getBasePtr();
  } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
    VT = ST->getMemoryVT();
    Ptr = ST->getBasePtr();
  } else
    return false;

  bool IsInc;
  if (!getIndexedAddressParts(Ptr.getNode(), Base, Offset, AM, IsInc, DAG))
    return false;
  AM = IsInc ? ISD::PRE_INC : ISD::PRE_DEC;
  return true;
}

bool AArch64TargetLowering::getPostIndexedAddressParts(
    SDNode *N, SDNode *Op, SDValue &Base, SDValue &Offset,
    ISD::MemIndexedMode &AM, SelectionDAG &DAG) const {
  EVT VT;
  SDValue Ptr;
  if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
    VT = LD->getMemoryVT();
    Ptr = LD->getBasePtr();
  } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
    VT = ST->getMemoryVT();
    Ptr = ST->getBasePtr();
  } else
    return false;

  bool IsInc;
  if (!getIndexedAddressParts(Op, Base, Offset, AM, IsInc, DAG))
    return false;
  // Post-indexing updates the base, so it's not a valid transform
  // if that's not the same as the load's pointer.
  if (Ptr != Base)
    return false;
  AM = IsInc ? ISD::POST_INC : ISD::POST_DEC;
  return true;
}

static void ReplaceBITCASTResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
                                  SelectionDAG &DAG) {
  if (N->getValueType(0) != MVT::i16)
    return;

  SDLoc DL(N);
  SDValue Op = N->getOperand(0);
  assert(Op.getValueType() == MVT::f16 &&
         "Inconsistent bitcast? Only 16-bit types should be i16 or f16");
  Op = SDValue(
      DAG.getMachineNode(TargetOpcode::INSERT_SUBREG, DL, MVT::f32,
                         DAG.getUNDEF(MVT::i32), Op,
                         DAG.getTargetConstant(AArch64::hsub, MVT::i32)),
      0);
  Op = DAG.getNode(ISD::BITCAST, DL, MVT::i32, Op);
  Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i16, Op));
}

void AArch64TargetLowering::ReplaceNodeResults(
    SDNode *N, SmallVectorImpl<SDValue> &Results, SelectionDAG &DAG) const {
  switch (N->getOpcode()) {
  default:
    llvm_unreachable("Don't know how to custom expand this");
  case ISD::BITCAST:
    ReplaceBITCASTResults(N, Results, DAG);
    return;
  case ISD::FP_TO_UINT:
  case ISD::FP_TO_SINT:
    assert(N->getValueType(0) == MVT::i128 && "unexpected illegal conversion");
    // Let normal code take care of it by not adding anything to Results.
    return;
  }
}

bool AArch64TargetLowering::useLoadStackGuardNode() const {
  return true;
}

TargetLoweringBase::LegalizeTypeAction
AArch64TargetLowering::getPreferredVectorAction(EVT VT) const {
  MVT SVT = VT.getSimpleVT();
  // During type legalization, we prefer to widen v1i8, v1i16, v1i32  to v8i8,
  // v4i16, v2i32 instead of to promote.
  if (SVT == MVT::v1i8 || SVT == MVT::v1i16 || SVT == MVT::v1i32
      || SVT == MVT::v1f32)
    return TypeWidenVector;

  return TargetLoweringBase::getPreferredVectorAction(VT);
}

// Loads and stores less than 128-bits are already atomic; ones above that
// are doomed anyway, so defer to the default libcall and blame the OS when
// things go wrong.
bool AArch64TargetLowering::shouldExpandAtomicStoreInIR(StoreInst *SI) const {
  unsigned Size = SI->getValueOperand()->getType()->getPrimitiveSizeInBits();
  return Size == 128;
}

// Loads and stores less than 128-bits are already atomic; ones above that
// are doomed anyway, so defer to the default libcall and blame the OS when
// things go wrong.
bool AArch64TargetLowering::shouldExpandAtomicLoadInIR(LoadInst *LI) const {
  unsigned Size = LI->getType()->getPrimitiveSizeInBits();
  return Size == 128;
}

// For the real atomic operations, we have ldxr/stxr up to 128 bits,
bool AArch64TargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
  unsigned Size = AI->getType()->getPrimitiveSizeInBits();
  return Size <= 128;
}

bool AArch64TargetLowering::hasLoadLinkedStoreConditional() const {
  return true;
}

Value *AArch64TargetLowering::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 i128 isn't legal and intrinsics don't get type-lowered, the ldrexd
  // intrinsic must return {i64, i64} and we have to recombine them into a
  // single i128 here.
  if (ValTy->getPrimitiveSizeInBits() == 128) {
    Intrinsic::ID Int =
        IsAcquire ? Intrinsic::aarch64_ldaxp : Intrinsic::aarch64_ldxp;
    Function *Ldxr = llvm::Intrinsic::getDeclaration(M, Int);

    Addr = Builder.CreateBitCast(Addr, Type::getInt8PtrTy(M->getContext()));
    Value *LoHi = Builder.CreateCall(Ldxr, Addr, "lohi");

    Value *Lo = Builder.CreateExtractValue(LoHi, 0, "lo");
    Value *Hi = Builder.CreateExtractValue(LoHi, 1, "hi");
    Lo = Builder.CreateZExt(Lo, ValTy, "lo64");
    Hi = Builder.CreateZExt(Hi, ValTy, "hi64");
    return Builder.CreateOr(
        Lo, Builder.CreateShl(Hi, ConstantInt::get(ValTy, 64)), "val64");
  }

  Type *Tys[] = { Addr->getType() };
  Intrinsic::ID Int =
      IsAcquire ? Intrinsic::aarch64_ldaxr : Intrinsic::aarch64_ldxr;
  Function *Ldxr = llvm::Intrinsic::getDeclaration(M, Int, Tys);

  return Builder.CreateTruncOrBitCast(
      Builder.CreateCall(Ldxr, Addr),
      cast<PointerType>(Addr->getType())->getElementType());
}

Value *AArch64TargetLowering::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 i128 intrinsics take two
  // parameters: "i64, i64". We must marshal Val into the appropriate form
  // before the call.
  if (Val->getType()->getPrimitiveSizeInBits() == 128) {
    Intrinsic::ID Int =
        IsRelease ? Intrinsic::aarch64_stlxp : Intrinsic::aarch64_stxp;
    Function *Stxr = Intrinsic::getDeclaration(M, Int);
    Type *Int64Ty = Type::getInt64Ty(M->getContext());

    Value *Lo = Builder.CreateTrunc(Val, Int64Ty, "lo");
    Value *Hi = Builder.CreateTrunc(Builder.CreateLShr(Val, 64), Int64Ty, "hi");
    Addr = Builder.CreateBitCast(Addr, Type::getInt8PtrTy(M->getContext()));
    return Builder.CreateCall3(Stxr, Lo, Hi, Addr);
  }

  Intrinsic::ID Int =
      IsRelease ? Intrinsic::aarch64_stlxr : Intrinsic::aarch64_stxr;
  Type *Tys[] = { Addr->getType() };
  Function *Stxr = Intrinsic::getDeclaration(M, Int, Tys);

  return Builder.CreateCall2(
      Stxr, Builder.CreateZExtOrBitCast(
                Val, Stxr->getFunctionType()->getParamType(0)),
      Addr);
}