summaryrefslogtreecommitdiffstats
path: root/src/amd/common/ac_llvm_build.c
blob: c325dbbc8cb5d361b4f27a6ddd5c250985b7d2ae (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
/*
 * Copyright 2014 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 */
/* based on pieces from si_pipe.c and radeon_llvm_emit.c */
#include "ac_llvm_build.h"

#include <llvm-c/Core.h>

#include "c11/threads.h"

#include <assert.h>
#include <stdio.h>

#include "ac_llvm_util.h"
#include "ac_exp_param.h"
#include "util/bitscan.h"
#include "util/macros.h"
#include "util/u_atomic.h"
#include "sid.h"

#include "shader_enums.h"

#define AC_LLVM_INITIAL_CF_DEPTH 4

/* Data for if/else/endif and bgnloop/endloop control flow structures.
 */
struct ac_llvm_flow {
	/* Loop exit or next part of if/else/endif. */
	LLVMBasicBlockRef next_block;
	LLVMBasicBlockRef loop_entry_block;
};

/* Initialize module-independent parts of the context.
 *
 * The caller is responsible for initializing ctx::module and ctx::builder.
 */
void
ac_llvm_context_init(struct ac_llvm_context *ctx, LLVMContextRef context,
		     enum chip_class chip_class, enum radeon_family family)
{
	LLVMValueRef args[1];

	ctx->chip_class = chip_class;
	ctx->family = family;

	ctx->context = context;
	ctx->module = NULL;
	ctx->builder = NULL;

	ctx->voidt = LLVMVoidTypeInContext(ctx->context);
	ctx->i1 = LLVMInt1TypeInContext(ctx->context);
	ctx->i8 = LLVMInt8TypeInContext(ctx->context);
	ctx->i16 = LLVMIntTypeInContext(ctx->context, 16);
	ctx->i32 = LLVMIntTypeInContext(ctx->context, 32);
	ctx->i64 = LLVMIntTypeInContext(ctx->context, 64);
	ctx->f16 = LLVMHalfTypeInContext(ctx->context);
	ctx->f32 = LLVMFloatTypeInContext(ctx->context);
	ctx->f64 = LLVMDoubleTypeInContext(ctx->context);
	ctx->v2i32 = LLVMVectorType(ctx->i32, 2);
	ctx->v3i32 = LLVMVectorType(ctx->i32, 3);
	ctx->v4i32 = LLVMVectorType(ctx->i32, 4);
	ctx->v2f32 = LLVMVectorType(ctx->f32, 2);
	ctx->v4f32 = LLVMVectorType(ctx->f32, 4);
	ctx->v8i32 = LLVMVectorType(ctx->i32, 8);

	ctx->i32_0 = LLVMConstInt(ctx->i32, 0, false);
	ctx->i32_1 = LLVMConstInt(ctx->i32, 1, false);
	ctx->i64_0 = LLVMConstInt(ctx->i64, 0, false);
	ctx->i64_1 = LLVMConstInt(ctx->i64, 1, false);
	ctx->f32_0 = LLVMConstReal(ctx->f32, 0.0);
	ctx->f32_1 = LLVMConstReal(ctx->f32, 1.0);
	ctx->f64_0 = LLVMConstReal(ctx->f64, 0.0);
	ctx->f64_1 = LLVMConstReal(ctx->f64, 1.0);

	ctx->i1false = LLVMConstInt(ctx->i1, 0, false);
	ctx->i1true = LLVMConstInt(ctx->i1, 1, false);

	ctx->range_md_kind = LLVMGetMDKindIDInContext(ctx->context,
						     "range", 5);

	ctx->invariant_load_md_kind = LLVMGetMDKindIDInContext(ctx->context,
							       "invariant.load", 14);

	ctx->fpmath_md_kind = LLVMGetMDKindIDInContext(ctx->context, "fpmath", 6);

	args[0] = LLVMConstReal(ctx->f32, 2.5);
	ctx->fpmath_md_2p5_ulp = LLVMMDNodeInContext(ctx->context, args, 1);

	ctx->uniform_md_kind = LLVMGetMDKindIDInContext(ctx->context,
							"amdgpu.uniform", 14);

	ctx->empty_md = LLVMMDNodeInContext(ctx->context, NULL, 0);
}

void
ac_llvm_context_dispose(struct ac_llvm_context *ctx)
{
	free(ctx->flow);
	ctx->flow = NULL;
	ctx->flow_depth_max = 0;
}

int
ac_get_llvm_num_components(LLVMValueRef value)
{
	LLVMTypeRef type = LLVMTypeOf(value);
	unsigned num_components = LLVMGetTypeKind(type) == LLVMVectorTypeKind
	                              ? LLVMGetVectorSize(type)
	                              : 1;
	return num_components;
}

LLVMValueRef
ac_llvm_extract_elem(struct ac_llvm_context *ac,
		     LLVMValueRef value,
		     int index)
{
	if (LLVMGetTypeKind(LLVMTypeOf(value)) != LLVMVectorTypeKind) {
		assert(index == 0);
		return value;
	}

	return LLVMBuildExtractElement(ac->builder, value,
				       LLVMConstInt(ac->i32, index, false), "");
}

unsigned
ac_get_type_size(LLVMTypeRef type)
{
	LLVMTypeKind kind = LLVMGetTypeKind(type);

	switch (kind) {
	case LLVMIntegerTypeKind:
		return LLVMGetIntTypeWidth(type) / 8;
	case LLVMFloatTypeKind:
		return 4;
	case LLVMDoubleTypeKind:
	case LLVMPointerTypeKind:
		return 8;
	case LLVMVectorTypeKind:
		return LLVMGetVectorSize(type) *
		       ac_get_type_size(LLVMGetElementType(type));
	case LLVMArrayTypeKind:
		return LLVMGetArrayLength(type) *
		       ac_get_type_size(LLVMGetElementType(type));
	default:
		assert(0);
		return 0;
	}
}

static LLVMTypeRef to_integer_type_scalar(struct ac_llvm_context *ctx, LLVMTypeRef t)
{
	if (t == ctx->f16 || t == ctx->i16)
		return ctx->i16;
	else if (t == ctx->f32 || t == ctx->i32)
		return ctx->i32;
	else if (t == ctx->f64 || t == ctx->i64)
		return ctx->i64;
	else
		unreachable("Unhandled integer size");
}

LLVMTypeRef
ac_to_integer_type(struct ac_llvm_context *ctx, LLVMTypeRef t)
{
	if (LLVMGetTypeKind(t) == LLVMVectorTypeKind) {
		LLVMTypeRef elem_type = LLVMGetElementType(t);
		return LLVMVectorType(to_integer_type_scalar(ctx, elem_type),
		                      LLVMGetVectorSize(t));
	}
	return to_integer_type_scalar(ctx, t);
}

LLVMValueRef
ac_to_integer(struct ac_llvm_context *ctx, LLVMValueRef v)
{
	LLVMTypeRef type = LLVMTypeOf(v);
	return LLVMBuildBitCast(ctx->builder, v, ac_to_integer_type(ctx, type), "");
}

static LLVMTypeRef to_float_type_scalar(struct ac_llvm_context *ctx, LLVMTypeRef t)
{
	if (t == ctx->i16 || t == ctx->f16)
		return ctx->f16;
	else if (t == ctx->i32 || t == ctx->f32)
		return ctx->f32;
	else if (t == ctx->i64 || t == ctx->f64)
		return ctx->f64;
	else
		unreachable("Unhandled float size");
}

LLVMTypeRef
ac_to_float_type(struct ac_llvm_context *ctx, LLVMTypeRef t)
{
	if (LLVMGetTypeKind(t) == LLVMVectorTypeKind) {
		LLVMTypeRef elem_type = LLVMGetElementType(t);
		return LLVMVectorType(to_float_type_scalar(ctx, elem_type),
		                      LLVMGetVectorSize(t));
	}
	return to_float_type_scalar(ctx, t);
}

LLVMValueRef
ac_to_float(struct ac_llvm_context *ctx, LLVMValueRef v)
{
	LLVMTypeRef type = LLVMTypeOf(v);
	return LLVMBuildBitCast(ctx->builder, v, ac_to_float_type(ctx, type), "");
}


LLVMValueRef
ac_build_intrinsic(struct ac_llvm_context *ctx, const char *name,
		   LLVMTypeRef return_type, LLVMValueRef *params,
		   unsigned param_count, unsigned attrib_mask)
{
	LLVMValueRef function, call;
	bool set_callsite_attrs = HAVE_LLVM >= 0x0400 &&
				  !(attrib_mask & AC_FUNC_ATTR_LEGACY);

	function = LLVMGetNamedFunction(ctx->module, name);
	if (!function) {
		LLVMTypeRef param_types[32], function_type;
		unsigned i;

		assert(param_count <= 32);

		for (i = 0; i < param_count; ++i) {
			assert(params[i]);
			param_types[i] = LLVMTypeOf(params[i]);
		}
		function_type =
		    LLVMFunctionType(return_type, param_types, param_count, 0);
		function = LLVMAddFunction(ctx->module, name, function_type);

		LLVMSetFunctionCallConv(function, LLVMCCallConv);
		LLVMSetLinkage(function, LLVMExternalLinkage);

		if (!set_callsite_attrs)
			ac_add_func_attributes(ctx->context, function, attrib_mask);
	}

	call = LLVMBuildCall(ctx->builder, function, params, param_count, "");
	if (set_callsite_attrs)
		ac_add_func_attributes(ctx->context, call, attrib_mask);
	return call;
}

/**
 * Given the i32 or vNi32 \p type, generate the textual name (e.g. for use with
 * intrinsic names).
 */
void ac_build_type_name_for_intr(LLVMTypeRef type, char *buf, unsigned bufsize)
{
	LLVMTypeRef elem_type = type;

	assert(bufsize >= 8);

	if (LLVMGetTypeKind(type) == LLVMVectorTypeKind) {
		int ret = snprintf(buf, bufsize, "v%u",
					LLVMGetVectorSize(type));
		if (ret < 0) {
			char *type_name = LLVMPrintTypeToString(type);
			fprintf(stderr, "Error building type name for: %s\n",
				type_name);
			return;
		}
		elem_type = LLVMGetElementType(type);
		buf += ret;
		bufsize -= ret;
	}
	switch (LLVMGetTypeKind(elem_type)) {
	default: break;
	case LLVMIntegerTypeKind:
		snprintf(buf, bufsize, "i%d", LLVMGetIntTypeWidth(elem_type));
		break;
	case LLVMFloatTypeKind:
		snprintf(buf, bufsize, "f32");
		break;
	case LLVMDoubleTypeKind:
		snprintf(buf, bufsize, "f64");
		break;
	}
}

/**
 * Helper function that builds an LLVM IR PHI node and immediately adds
 * incoming edges.
 */
LLVMValueRef
ac_build_phi(struct ac_llvm_context *ctx, LLVMTypeRef type,
	     unsigned count_incoming, LLVMValueRef *values,
	     LLVMBasicBlockRef *blocks)
{
	LLVMValueRef phi = LLVMBuildPhi(ctx->builder, type, "");
	LLVMAddIncoming(phi, values, blocks, count_incoming);
	return phi;
}

/* Prevent optimizations (at least of memory accesses) across the current
 * point in the program by emitting empty inline assembly that is marked as
 * having side effects.
 *
 * Optionally, a value can be passed through the inline assembly to prevent
 * LLVM from hoisting calls to ReadNone functions.
 */
void
ac_build_optimization_barrier(struct ac_llvm_context *ctx,
			      LLVMValueRef *pvgpr)
{
	static int counter = 0;

	LLVMBuilderRef builder = ctx->builder;
	char code[16];

	snprintf(code, sizeof(code), "; %d", p_atomic_inc_return(&counter));

	if (!pvgpr) {
		LLVMTypeRef ftype = LLVMFunctionType(ctx->voidt, NULL, 0, false);
		LLVMValueRef inlineasm = LLVMConstInlineAsm(ftype, code, "", true, false);
		LLVMBuildCall(builder, inlineasm, NULL, 0, "");
	} else {
		LLVMTypeRef ftype = LLVMFunctionType(ctx->i32, &ctx->i32, 1, false);
		LLVMValueRef inlineasm = LLVMConstInlineAsm(ftype, code, "=v,0", true, false);
		LLVMValueRef vgpr = *pvgpr;
		LLVMTypeRef vgpr_type = LLVMTypeOf(vgpr);
		unsigned vgpr_size = ac_get_type_size(vgpr_type);
		LLVMValueRef vgpr0;

		assert(vgpr_size % 4 == 0);

		vgpr = LLVMBuildBitCast(builder, vgpr, LLVMVectorType(ctx->i32, vgpr_size / 4), "");
		vgpr0 = LLVMBuildExtractElement(builder, vgpr, ctx->i32_0, "");
		vgpr0 = LLVMBuildCall(builder, inlineasm, &vgpr0, 1, "");
		vgpr = LLVMBuildInsertElement(builder, vgpr, vgpr0, ctx->i32_0, "");
		vgpr = LLVMBuildBitCast(builder, vgpr, vgpr_type, "");

		*pvgpr = vgpr;
	}
}

LLVMValueRef
ac_build_ballot(struct ac_llvm_context *ctx,
		LLVMValueRef value)
{
	LLVMValueRef args[3] = {
		value,
		ctx->i32_0,
		LLVMConstInt(ctx->i32, LLVMIntNE, 0)
	};

	/* We currently have no other way to prevent LLVM from lifting the icmp
	 * calls to a dominating basic block.
	 */
	ac_build_optimization_barrier(ctx, &args[0]);

	if (LLVMTypeOf(args[0]) != ctx->i32)
		args[0] = LLVMBuildBitCast(ctx->builder, args[0], ctx->i32, "");

	return ac_build_intrinsic(ctx,
				  "llvm.amdgcn.icmp.i32",
				  ctx->i64, args, 3,
				  AC_FUNC_ATTR_NOUNWIND |
				  AC_FUNC_ATTR_READNONE |
				  AC_FUNC_ATTR_CONVERGENT);
}

LLVMValueRef
ac_build_vote_all(struct ac_llvm_context *ctx, LLVMValueRef value)
{
	LLVMValueRef active_set = ac_build_ballot(ctx, ctx->i32_1);
	LLVMValueRef vote_set = ac_build_ballot(ctx, value);
	return LLVMBuildICmp(ctx->builder, LLVMIntEQ, vote_set, active_set, "");
}

LLVMValueRef
ac_build_vote_any(struct ac_llvm_context *ctx, LLVMValueRef value)
{
	LLVMValueRef vote_set = ac_build_ballot(ctx, value);
	return LLVMBuildICmp(ctx->builder, LLVMIntNE, vote_set,
			     LLVMConstInt(ctx->i64, 0, 0), "");
}

LLVMValueRef
ac_build_vote_eq(struct ac_llvm_context *ctx, LLVMValueRef value)
{
	LLVMValueRef active_set = ac_build_ballot(ctx, ctx->i32_1);
	LLVMValueRef vote_set = ac_build_ballot(ctx, value);

	LLVMValueRef all = LLVMBuildICmp(ctx->builder, LLVMIntEQ,
					 vote_set, active_set, "");
	LLVMValueRef none = LLVMBuildICmp(ctx->builder, LLVMIntEQ,
					  vote_set,
					  LLVMConstInt(ctx->i64, 0, 0), "");
	return LLVMBuildOr(ctx->builder, all, none, "");
}

LLVMValueRef
ac_build_varying_gather_values(struct ac_llvm_context *ctx, LLVMValueRef *values,
			       unsigned value_count, unsigned component)
{
	LLVMValueRef vec = NULL;

	if (value_count == 1) {
		return values[component];
	} else if (!value_count)
		unreachable("value_count is 0");

	for (unsigned i = component; i < value_count + component; i++) {
		LLVMValueRef value = values[i];

		if (i == component)
			vec = LLVMGetUndef( LLVMVectorType(LLVMTypeOf(value), value_count));
		LLVMValueRef index = LLVMConstInt(ctx->i32, i - component, false);
		vec = LLVMBuildInsertElement(ctx->builder, vec, value, index, "");
	}
	return vec;
}

LLVMValueRef
ac_build_gather_values_extended(struct ac_llvm_context *ctx,
				LLVMValueRef *values,
				unsigned value_count,
				unsigned value_stride,
				bool load,
				bool always_vector)
{
	LLVMBuilderRef builder = ctx->builder;
	LLVMValueRef vec = NULL;
	unsigned i;

	if (value_count == 1 && !always_vector) {
		if (load)
			return LLVMBuildLoad(builder, values[0], "");
		return values[0];
	} else if (!value_count)
		unreachable("value_count is 0");

	for (i = 0; i < value_count; i++) {
		LLVMValueRef value = values[i * value_stride];
		if (load)
			value = LLVMBuildLoad(builder, value, "");

		if (!i)
			vec = LLVMGetUndef( LLVMVectorType(LLVMTypeOf(value), value_count));
		LLVMValueRef index = LLVMConstInt(ctx->i32, i, false);
		vec = LLVMBuildInsertElement(builder, vec, value, index, "");
	}
	return vec;
}

LLVMValueRef
ac_build_gather_values(struct ac_llvm_context *ctx,
		       LLVMValueRef *values,
		       unsigned value_count)
{
	return ac_build_gather_values_extended(ctx, values, value_count, 1, false, false);
}

LLVMValueRef
ac_build_fdiv(struct ac_llvm_context *ctx,
	      LLVMValueRef num,
	      LLVMValueRef den)
{
	LLVMValueRef ret = LLVMBuildFDiv(ctx->builder, num, den, "");

	/* Use v_rcp_f32 instead of precise division. */
	if (!LLVMIsConstant(ret))
		LLVMSetMetadata(ret, ctx->fpmath_md_kind, ctx->fpmath_md_2p5_ulp);
	return ret;
}

/* Coordinates for cube map selection. sc, tc, and ma are as in Table 8.27
 * of the OpenGL 4.5 (Compatibility Profile) specification, except ma is
 * already multiplied by two. id is the cube face number.
 */
struct cube_selection_coords {
	LLVMValueRef stc[2];
	LLVMValueRef ma;
	LLVMValueRef id;
};

static void
build_cube_intrinsic(struct ac_llvm_context *ctx,
		     LLVMValueRef in[3],
		     struct cube_selection_coords *out)
{
	LLVMTypeRef f32 = ctx->f32;

	out->stc[1] = ac_build_intrinsic(ctx, "llvm.amdgcn.cubetc",
					 f32, in, 3, AC_FUNC_ATTR_READNONE);
	out->stc[0] = ac_build_intrinsic(ctx, "llvm.amdgcn.cubesc",
					 f32, in, 3, AC_FUNC_ATTR_READNONE);
	out->ma = ac_build_intrinsic(ctx, "llvm.amdgcn.cubema",
				     f32, in, 3, AC_FUNC_ATTR_READNONE);
	out->id = ac_build_intrinsic(ctx, "llvm.amdgcn.cubeid",
				     f32, in, 3, AC_FUNC_ATTR_READNONE);
}

/**
 * Build a manual selection sequence for cube face sc/tc coordinates and
 * major axis vector (multiplied by 2 for consistency) for the given
 * vec3 \p coords, for the face implied by \p selcoords.
 *
 * For the major axis, we always adjust the sign to be in the direction of
 * selcoords.ma; i.e., a positive out_ma means that coords is pointed towards
 * the selcoords major axis.
 */
static void build_cube_select(struct ac_llvm_context *ctx,
			      const struct cube_selection_coords *selcoords,
			      const LLVMValueRef *coords,
			      LLVMValueRef *out_st,
			      LLVMValueRef *out_ma)
{
	LLVMBuilderRef builder = ctx->builder;
	LLVMTypeRef f32 = LLVMTypeOf(coords[0]);
	LLVMValueRef is_ma_positive;
	LLVMValueRef sgn_ma;
	LLVMValueRef is_ma_z, is_not_ma_z;
	LLVMValueRef is_ma_y;
	LLVMValueRef is_ma_x;
	LLVMValueRef sgn;
	LLVMValueRef tmp;

	is_ma_positive = LLVMBuildFCmp(builder, LLVMRealUGE,
		selcoords->ma, LLVMConstReal(f32, 0.0), "");
	sgn_ma = LLVMBuildSelect(builder, is_ma_positive,
		LLVMConstReal(f32, 1.0), LLVMConstReal(f32, -1.0), "");

	is_ma_z = LLVMBuildFCmp(builder, LLVMRealUGE, selcoords->id, LLVMConstReal(f32, 4.0), "");
	is_not_ma_z = LLVMBuildNot(builder, is_ma_z, "");
	is_ma_y = LLVMBuildAnd(builder, is_not_ma_z,
		LLVMBuildFCmp(builder, LLVMRealUGE, selcoords->id, LLVMConstReal(f32, 2.0), ""), "");
	is_ma_x = LLVMBuildAnd(builder, is_not_ma_z, LLVMBuildNot(builder, is_ma_y, ""), "");

	/* Select sc */
	tmp = LLVMBuildSelect(builder, is_ma_x, coords[2], coords[0], "");
	sgn = LLVMBuildSelect(builder, is_ma_y, LLVMConstReal(f32, 1.0),
		LLVMBuildSelect(builder, is_ma_z, sgn_ma,
			LLVMBuildFNeg(builder, sgn_ma, ""), ""), "");
	out_st[0] = LLVMBuildFMul(builder, tmp, sgn, "");

	/* Select tc */
	tmp = LLVMBuildSelect(builder, is_ma_y, coords[2], coords[1], "");
	sgn = LLVMBuildSelect(builder, is_ma_y, sgn_ma,
		LLVMConstReal(f32, -1.0), "");
	out_st[1] = LLVMBuildFMul(builder, tmp, sgn, "");

	/* Select ma */
	tmp = LLVMBuildSelect(builder, is_ma_z, coords[2],
		LLVMBuildSelect(builder, is_ma_y, coords[1], coords[0], ""), "");
	tmp = ac_build_intrinsic(ctx, "llvm.fabs.f32",
				 ctx->f32, &tmp, 1, AC_FUNC_ATTR_READNONE);
	*out_ma = LLVMBuildFMul(builder, tmp, LLVMConstReal(f32, 2.0), "");
}

void
ac_prepare_cube_coords(struct ac_llvm_context *ctx,
		       bool is_deriv, bool is_array, bool is_lod,
		       LLVMValueRef *coords_arg,
		       LLVMValueRef *derivs_arg)
{

	LLVMBuilderRef builder = ctx->builder;
	struct cube_selection_coords selcoords;
	LLVMValueRef coords[3];
	LLVMValueRef invma;

	if (is_array && !is_lod) {
		LLVMValueRef tmp = coords_arg[3];
		tmp = ac_build_intrinsic(ctx, "llvm.rint.f32", ctx->f32, &tmp, 1, 0);

		/* Section 8.9 (Texture Functions) of the GLSL 4.50 spec says:
		 *
		 *    "For Array forms, the array layer used will be
		 *
		 *       max(0, min(d−1, floor(layer+0.5)))
		 *
		 *     where d is the depth of the texture array and layer
		 *     comes from the component indicated in the tables below.
		 *     Workaroudn for an issue where the layer is taken from a
		 *     helper invocation which happens to fall on a different
		 *     layer due to extrapolation."
		 *
		 * VI and earlier attempt to implement this in hardware by
		 * clamping the value of coords[2] = (8 * layer) + face.
		 * Unfortunately, this means that the we end up with the wrong
		 * face when clamping occurs.
		 *
		 * Clamp the layer earlier to work around the issue.
		 */
		if (ctx->chip_class <= VI) {
			LLVMValueRef ge0;
			ge0 = LLVMBuildFCmp(builder, LLVMRealOGE, tmp, ctx->f32_0, "");
			tmp = LLVMBuildSelect(builder, ge0, tmp, ctx->f32_0, "");
		}

		coords_arg[3] = tmp;
	}

	build_cube_intrinsic(ctx, coords_arg, &selcoords);

	invma = ac_build_intrinsic(ctx, "llvm.fabs.f32",
			ctx->f32, &selcoords.ma, 1, AC_FUNC_ATTR_READNONE);
	invma = ac_build_fdiv(ctx, LLVMConstReal(ctx->f32, 1.0), invma);

	for (int i = 0; i < 2; ++i)
		coords[i] = LLVMBuildFMul(builder, selcoords.stc[i], invma, "");

	coords[2] = selcoords.id;

	if (is_deriv && derivs_arg) {
		LLVMValueRef derivs[4];
		int axis;

		/* Convert cube derivatives to 2D derivatives. */
		for (axis = 0; axis < 2; axis++) {
			LLVMValueRef deriv_st[2];
			LLVMValueRef deriv_ma;

			/* Transform the derivative alongside the texture
			 * coordinate. Mathematically, the correct formula is
			 * as follows. Assume we're projecting onto the +Z face
			 * and denote by dx/dh the derivative of the (original)
			 * X texture coordinate with respect to horizontal
			 * window coordinates. The projection onto the +Z face
			 * plane is:
			 *
			 *   f(x,z) = x/z
			 *
			 * Then df/dh = df/dx * dx/dh + df/dz * dz/dh
			 *            = 1/z * dx/dh - x/z * 1/z * dz/dh.
			 *
			 * This motivatives the implementation below.
			 *
			 * Whether this actually gives the expected results for
			 * apps that might feed in derivatives obtained via
			 * finite differences is anyone's guess. The OpenGL spec
			 * seems awfully quiet about how textureGrad for cube
			 * maps should be handled.
			 */
			build_cube_select(ctx, &selcoords, &derivs_arg[axis * 3],
					  deriv_st, &deriv_ma);

			deriv_ma = LLVMBuildFMul(builder, deriv_ma, invma, "");

			for (int i = 0; i < 2; ++i)
				derivs[axis * 2 + i] =
					LLVMBuildFSub(builder,
						LLVMBuildFMul(builder, deriv_st[i], invma, ""),
						LLVMBuildFMul(builder, deriv_ma, coords[i], ""), "");
		}

		memcpy(derivs_arg, derivs, sizeof(derivs));
	}

	/* Shift the texture coordinate. This must be applied after the
	 * derivative calculation.
	 */
	for (int i = 0; i < 2; ++i)
		coords[i] = LLVMBuildFAdd(builder, coords[i], LLVMConstReal(ctx->f32, 1.5), "");

	if (is_array) {
		/* for cube arrays coord.z = coord.w(array_index) * 8 + face */
		/* coords_arg.w component - array_index for cube arrays */
		LLVMValueRef tmp = LLVMBuildFMul(ctx->builder, coords_arg[3], LLVMConstReal(ctx->f32, 8.0), "");
		coords[2] = LLVMBuildFAdd(ctx->builder, tmp, coords[2], "");
	}

	memcpy(coords_arg, coords, sizeof(coords));
}


LLVMValueRef
ac_build_fs_interp(struct ac_llvm_context *ctx,
		   LLVMValueRef llvm_chan,
		   LLVMValueRef attr_number,
		   LLVMValueRef params,
		   LLVMValueRef i,
		   LLVMValueRef j)
{
	LLVMValueRef args[5];
	LLVMValueRef p1;
	
	if (HAVE_LLVM < 0x0400) {
		LLVMValueRef ij[2];
		ij[0] = LLVMBuildBitCast(ctx->builder, i, ctx->i32, "");
		ij[1] = LLVMBuildBitCast(ctx->builder, j, ctx->i32, "");

		args[0] = llvm_chan;
		args[1] = attr_number;
		args[2] = params;
		args[3] = ac_build_gather_values(ctx, ij, 2);
		return ac_build_intrinsic(ctx, "llvm.SI.fs.interp",
					  ctx->f32, args, 4,
					  AC_FUNC_ATTR_READNONE);
	}

	args[0] = i;
	args[1] = llvm_chan;
	args[2] = attr_number;
	args[3] = params;

	p1 = ac_build_intrinsic(ctx, "llvm.amdgcn.interp.p1",
				ctx->f32, args, 4, AC_FUNC_ATTR_READNONE);

	args[0] = p1;
	args[1] = j;
	args[2] = llvm_chan;
	args[3] = attr_number;
	args[4] = params;

	return ac_build_intrinsic(ctx, "llvm.amdgcn.interp.p2",
				  ctx->f32, args, 5, AC_FUNC_ATTR_READNONE);
}

LLVMValueRef
ac_build_fs_interp_mov(struct ac_llvm_context *ctx,
		       LLVMValueRef parameter,
		       LLVMValueRef llvm_chan,
		       LLVMValueRef attr_number,
		       LLVMValueRef params)
{
	LLVMValueRef args[4];
	if (HAVE_LLVM < 0x0400) {
		args[0] = llvm_chan;
		args[1] = attr_number;
		args[2] = params;

		return ac_build_intrinsic(ctx,
					  "llvm.SI.fs.constant",
					  ctx->f32, args, 3,
					  AC_FUNC_ATTR_READNONE);
	}

	args[0] = parameter;
	args[1] = llvm_chan;
	args[2] = attr_number;
	args[3] = params;

	return ac_build_intrinsic(ctx, "llvm.amdgcn.interp.mov",
				  ctx->f32, args, 4, AC_FUNC_ATTR_READNONE);
}

LLVMValueRef
ac_build_gep0(struct ac_llvm_context *ctx,
	      LLVMValueRef base_ptr,
	      LLVMValueRef index)
{
	LLVMValueRef indices[2] = {
		LLVMConstInt(ctx->i32, 0, 0),
		index,
	};
	return LLVMBuildGEP(ctx->builder, base_ptr,
			    indices, 2, "");
}

void
ac_build_indexed_store(struct ac_llvm_context *ctx,
		       LLVMValueRef base_ptr, LLVMValueRef index,
		       LLVMValueRef value)
{
	LLVMBuildStore(ctx->builder, value,
		       ac_build_gep0(ctx, base_ptr, index));
}

/**
 * Build an LLVM bytecode indexed load using LLVMBuildGEP + LLVMBuildLoad.
 * It's equivalent to doing a load from &base_ptr[index].
 *
 * \param base_ptr  Where the array starts.
 * \param index     The element index into the array.
 * \param uniform   Whether the base_ptr and index can be assumed to be
 *                  dynamically uniform (i.e. load to an SGPR)
 * \param invariant Whether the load is invariant (no other opcodes affect it)
 */
static LLVMValueRef
ac_build_load_custom(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
		     LLVMValueRef index, bool uniform, bool invariant)
{
	LLVMValueRef pointer, result;

	pointer = ac_build_gep0(ctx, base_ptr, index);
	if (uniform)
		LLVMSetMetadata(pointer, ctx->uniform_md_kind, ctx->empty_md);
	result = LLVMBuildLoad(ctx->builder, pointer, "");
	if (invariant)
		LLVMSetMetadata(result, ctx->invariant_load_md_kind, ctx->empty_md);
	return result;
}

LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
			   LLVMValueRef index)
{
	return ac_build_load_custom(ctx, base_ptr, index, false, false);
}

LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx,
				     LLVMValueRef base_ptr, LLVMValueRef index)
{
	return ac_build_load_custom(ctx, base_ptr, index, false, true);
}

LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx,
				   LLVMValueRef base_ptr, LLVMValueRef index)
{
	return ac_build_load_custom(ctx, base_ptr, index, true, true);
}

/* TBUFFER_STORE_FORMAT_{X,XY,XYZ,XYZW} <- the suffix is selected by num_channels=1..4.
 * The type of vdata must be one of i32 (num_channels=1), v2i32 (num_channels=2),
 * or v4i32 (num_channels=3,4).
 */
void
ac_build_buffer_store_dword(struct ac_llvm_context *ctx,
			    LLVMValueRef rsrc,
			    LLVMValueRef vdata,
			    unsigned num_channels,
			    LLVMValueRef voffset,
			    LLVMValueRef soffset,
			    unsigned inst_offset,
			    bool glc,
			    bool slc,
			    bool writeonly_memory,
			    bool swizzle_enable_hint)
{
	/* SWIZZLE_ENABLE requires that soffset isn't folded into voffset
	 * (voffset is swizzled, but soffset isn't swizzled).
	 * llvm.amdgcn.buffer.store doesn't have a separate soffset parameter.
	 */
	if (!swizzle_enable_hint) {
		/* Split 3 channel stores, becase LLVM doesn't support 3-channel
		 * intrinsics. */
		if (num_channels == 3) {
			LLVMValueRef v[3], v01;

			for (int i = 0; i < 3; i++) {
				v[i] = LLVMBuildExtractElement(ctx->builder, vdata,
						LLVMConstInt(ctx->i32, i, 0), "");
			}
			v01 = ac_build_gather_values(ctx, v, 2);

			ac_build_buffer_store_dword(ctx, rsrc, v01, 2, voffset,
						    soffset, inst_offset, glc, slc,
						    writeonly_memory, swizzle_enable_hint);
			ac_build_buffer_store_dword(ctx, rsrc, v[2], 1, voffset,
						    soffset, inst_offset + 8,
						    glc, slc,
						    writeonly_memory, swizzle_enable_hint);
			return;
		}

		unsigned func = CLAMP(num_channels, 1, 3) - 1;
		static const char *types[] = {"f32", "v2f32", "v4f32"};
		char name[256];
		LLVMValueRef offset = soffset;

		if (inst_offset)
			offset = LLVMBuildAdd(ctx->builder, offset,
					      LLVMConstInt(ctx->i32, inst_offset, 0), "");
		if (voffset)
			offset = LLVMBuildAdd(ctx->builder, offset, voffset, "");

		LLVMValueRef args[] = {
			ac_to_float(ctx, vdata),
			LLVMBuildBitCast(ctx->builder, rsrc, ctx->v4i32, ""),
			LLVMConstInt(ctx->i32, 0, 0),
			offset,
			LLVMConstInt(ctx->i1, glc, 0),
			LLVMConstInt(ctx->i1, slc, 0),
		};

		snprintf(name, sizeof(name), "llvm.amdgcn.buffer.store.%s",
			 types[func]);

		ac_build_intrinsic(ctx, name, ctx->voidt,
				   args, ARRAY_SIZE(args),
				   writeonly_memory ?
					   AC_FUNC_ATTR_INACCESSIBLE_MEM_ONLY :
					   AC_FUNC_ATTR_WRITEONLY);
		return;
	}

	static unsigned dfmt[] = {
		V_008F0C_BUF_DATA_FORMAT_32,
		V_008F0C_BUF_DATA_FORMAT_32_32,
		V_008F0C_BUF_DATA_FORMAT_32_32_32,
		V_008F0C_BUF_DATA_FORMAT_32_32_32_32
	};
	assert(num_channels >= 1 && num_channels <= 4);

	LLVMValueRef args[] = {
		rsrc,
		vdata,
		LLVMConstInt(ctx->i32, num_channels, 0),
		voffset ? voffset : LLVMGetUndef(ctx->i32),
		soffset,
		LLVMConstInt(ctx->i32, inst_offset, 0),
		LLVMConstInt(ctx->i32, dfmt[num_channels - 1], 0),
		LLVMConstInt(ctx->i32, V_008F0C_BUF_NUM_FORMAT_UINT, 0),
		LLVMConstInt(ctx->i32, voffset != NULL, 0),
		LLVMConstInt(ctx->i32, 0, 0), /* idxen */
		LLVMConstInt(ctx->i32, glc, 0),
		LLVMConstInt(ctx->i32, slc, 0),
		LLVMConstInt(ctx->i32, 0, 0), /* tfe*/
	};

	/* The instruction offset field has 12 bits */
	assert(voffset || inst_offset < (1 << 12));

	/* The intrinsic is overloaded, we need to add a type suffix for overloading to work. */
	unsigned func = CLAMP(num_channels, 1, 3) - 1;
	const char *types[] = {"i32", "v2i32", "v4i32"};
	char name[256];
	snprintf(name, sizeof(name), "llvm.SI.tbuffer.store.%s", types[func]);

	ac_build_intrinsic(ctx, name, ctx->voidt,
			   args, ARRAY_SIZE(args),
			   AC_FUNC_ATTR_LEGACY);
}

LLVMValueRef
ac_build_buffer_load(struct ac_llvm_context *ctx,
		     LLVMValueRef rsrc,
		     int num_channels,
		     LLVMValueRef vindex,
		     LLVMValueRef voffset,
		     LLVMValueRef soffset,
		     unsigned inst_offset,
		     unsigned glc,
		     unsigned slc,
		     bool can_speculate,
		     bool allow_smem)
{
	LLVMValueRef offset = LLVMConstInt(ctx->i32, inst_offset, 0);
	if (voffset)
		offset = LLVMBuildAdd(ctx->builder, offset, voffset, "");
	if (soffset)
		offset = LLVMBuildAdd(ctx->builder, offset, soffset, "");

	/* TODO: VI and later generations can use SMEM with GLC=1.*/
	if (allow_smem && !glc && !slc) {
		assert(vindex == NULL);

		LLVMValueRef result[4];

		for (int i = 0; i < num_channels; i++) {
			if (i) {
				offset = LLVMBuildAdd(ctx->builder, offset,
						      LLVMConstInt(ctx->i32, 4, 0), "");
			}
			LLVMValueRef args[2] = {rsrc, offset};
			result[i] = ac_build_intrinsic(ctx, "llvm.SI.load.const.v4i32",
						       ctx->f32, args, 2,
						       AC_FUNC_ATTR_READNONE |
						       AC_FUNC_ATTR_LEGACY);
		}
		if (num_channels == 1)
			return result[0];

		if (num_channels == 3)
			result[num_channels++] = LLVMGetUndef(ctx->f32);
		return ac_build_gather_values(ctx, result, num_channels);
	}

	unsigned func = CLAMP(num_channels, 1, 3) - 1;

	LLVMValueRef args[] = {
		LLVMBuildBitCast(ctx->builder, rsrc, ctx->v4i32, ""),
		vindex ? vindex : LLVMConstInt(ctx->i32, 0, 0),
		offset,
		LLVMConstInt(ctx->i1, glc, 0),
		LLVMConstInt(ctx->i1, slc, 0)
	};

	LLVMTypeRef types[] = {ctx->f32, LLVMVectorType(ctx->f32, 2),
			       ctx->v4f32};
	const char *type_names[] = {"f32", "v2f32", "v4f32"};
	char name[256];

	snprintf(name, sizeof(name), "llvm.amdgcn.buffer.load.%s",
		 type_names[func]);

	return ac_build_intrinsic(ctx, name, types[func], args,
				  ARRAY_SIZE(args),
				  ac_get_load_intr_attribs(can_speculate));
}

LLVMValueRef ac_build_buffer_load_format(struct ac_llvm_context *ctx,
					 LLVMValueRef rsrc,
					 LLVMValueRef vindex,
					 LLVMValueRef voffset,
					 bool can_speculate)
{
	LLVMValueRef args [] = {
		LLVMBuildBitCast(ctx->builder, rsrc, ctx->v4i32, ""),
		vindex,
		voffset,
		ctx->i1false, /* glc */
		ctx->i1false, /* slc */
	};

	return ac_build_intrinsic(ctx,
				  "llvm.amdgcn.buffer.load.format.v4f32",
				  ctx->v4f32, args, ARRAY_SIZE(args),
				  ac_get_load_intr_attribs(can_speculate));
}

/**
 * Set range metadata on an instruction.  This can only be used on load and
 * call instructions.  If you know an instruction can only produce the values
 * 0, 1, 2, you would do set_range_metadata(value, 0, 3);
 * \p lo is the minimum value inclusive.
 * \p hi is the maximum value exclusive.
 */
static void set_range_metadata(struct ac_llvm_context *ctx,
			       LLVMValueRef value, unsigned lo, unsigned hi)
{
	LLVMValueRef range_md, md_args[2];
	LLVMTypeRef type = LLVMTypeOf(value);
	LLVMContextRef context = LLVMGetTypeContext(type);

	md_args[0] = LLVMConstInt(type, lo, false);
	md_args[1] = LLVMConstInt(type, hi, false);
	range_md = LLVMMDNodeInContext(context, md_args, 2);
	LLVMSetMetadata(value, ctx->range_md_kind, range_md);
}

LLVMValueRef
ac_get_thread_id(struct ac_llvm_context *ctx)
{
	LLVMValueRef tid;

	LLVMValueRef tid_args[2];
	tid_args[0] = LLVMConstInt(ctx->i32, 0xffffffff, false);
	tid_args[1] = LLVMConstInt(ctx->i32, 0, false);
	tid_args[1] = ac_build_intrinsic(ctx,
					 "llvm.amdgcn.mbcnt.lo", ctx->i32,
					 tid_args, 2, AC_FUNC_ATTR_READNONE);

	tid = ac_build_intrinsic(ctx, "llvm.amdgcn.mbcnt.hi",
				 ctx->i32, tid_args,
				 2, AC_FUNC_ATTR_READNONE);
	set_range_metadata(ctx, tid, 0, 64);
	return tid;
}

/*
 * SI implements derivatives using the local data store (LDS)
 * All writes to the LDS happen in all executing threads at
 * the same time. TID is the Thread ID for the current
 * thread and is a value between 0 and 63, representing
 * the thread's position in the wavefront.
 *
 * For the pixel shader threads are grouped into quads of four pixels.
 * The TIDs of the pixels of a quad are:
 *
 *  +------+------+
 *  |4n + 0|4n + 1|
 *  +------+------+
 *  |4n + 2|4n + 3|
 *  +------+------+
 *
 * So, masking the TID with 0xfffffffc yields the TID of the top left pixel
 * of the quad, masking with 0xfffffffd yields the TID of the top pixel of
 * the current pixel's column, and masking with 0xfffffffe yields the TID
 * of the left pixel of the current pixel's row.
 *
 * Adding 1 yields the TID of the pixel to the right of the left pixel, and
 * adding 2 yields the TID of the pixel below the top pixel.
 */
LLVMValueRef
ac_build_ddxy(struct ac_llvm_context *ctx,
	      uint32_t mask,
	      int idx,
	      LLVMValueRef val)
{
	LLVMValueRef tl, trbl, args[2];
	LLVMValueRef result;

	if (ctx->chip_class >= VI) {
		LLVMValueRef thread_id, tl_tid, trbl_tid;
		thread_id = ac_get_thread_id(ctx);

		tl_tid = LLVMBuildAnd(ctx->builder, thread_id,
				      LLVMConstInt(ctx->i32, mask, false), "");

		trbl_tid = LLVMBuildAdd(ctx->builder, tl_tid,
					LLVMConstInt(ctx->i32, idx, false), "");

		args[0] = LLVMBuildMul(ctx->builder, tl_tid,
				       LLVMConstInt(ctx->i32, 4, false), "");
		args[1] = val;
		tl = ac_build_intrinsic(ctx,
					"llvm.amdgcn.ds.bpermute", ctx->i32,
					args, 2,
					AC_FUNC_ATTR_READNONE |
					AC_FUNC_ATTR_CONVERGENT);

		args[0] = LLVMBuildMul(ctx->builder, trbl_tid,
				       LLVMConstInt(ctx->i32, 4, false), "");
		trbl = ac_build_intrinsic(ctx,
					  "llvm.amdgcn.ds.bpermute", ctx->i32,
					  args, 2,
					  AC_FUNC_ATTR_READNONE |
					  AC_FUNC_ATTR_CONVERGENT);
	} else {
		uint32_t masks[2] = {};

		switch (mask) {
		case AC_TID_MASK_TOP_LEFT:
			masks[0] = 0x8000;
			if (idx == 1)
				masks[1] = 0x8055;
			else
				masks[1] = 0x80aa;

			break;
		case AC_TID_MASK_TOP:
			masks[0] = 0x8044;
			masks[1] = 0x80ee;
			break;
		case AC_TID_MASK_LEFT:
			masks[0] = 0x80a0;
			masks[1] = 0x80f5;
			break;
		default:
			assert(0);
		}

		args[0] = val;
		args[1] = LLVMConstInt(ctx->i32, masks[0], false);

		tl = ac_build_intrinsic(ctx,
					"llvm.amdgcn.ds.swizzle", ctx->i32,
					args, 2,
					AC_FUNC_ATTR_READNONE |
					AC_FUNC_ATTR_CONVERGENT);

		args[1] = LLVMConstInt(ctx->i32, masks[1], false);
		trbl = ac_build_intrinsic(ctx,
					"llvm.amdgcn.ds.swizzle", ctx->i32,
					args, 2,
					AC_FUNC_ATTR_READNONE |
					AC_FUNC_ATTR_CONVERGENT);
	}

	tl = LLVMBuildBitCast(ctx->builder, tl, ctx->f32, "");
	trbl = LLVMBuildBitCast(ctx->builder, trbl, ctx->f32, "");
	result = LLVMBuildFSub(ctx->builder, trbl, tl, "");
	return result;
}

void
ac_build_sendmsg(struct ac_llvm_context *ctx,
		 uint32_t msg,
		 LLVMValueRef wave_id)
{
	LLVMValueRef args[2];
	const char *intr_name = (HAVE_LLVM < 0x0400) ? "llvm.SI.sendmsg" : "llvm.amdgcn.s.sendmsg";
	args[0] = LLVMConstInt(ctx->i32, msg, false);
	args[1] = wave_id;
	ac_build_intrinsic(ctx, intr_name, ctx->voidt, args, 2, 0);
}

LLVMValueRef
ac_build_imsb(struct ac_llvm_context *ctx,
	      LLVMValueRef arg,
	      LLVMTypeRef dst_type)
{
	const char *intr_name = (HAVE_LLVM < 0x0400) ? "llvm.AMDGPU.flbit.i32" :
						       "llvm.amdgcn.sffbh.i32";
	LLVMValueRef msb = ac_build_intrinsic(ctx, intr_name,
					      dst_type, &arg, 1,
					      AC_FUNC_ATTR_READNONE);

	/* The HW returns the last bit index from MSB, but NIR/TGSI wants
	 * the index from LSB. Invert it by doing "31 - msb". */
	msb = LLVMBuildSub(ctx->builder, LLVMConstInt(ctx->i32, 31, false),
			   msb, "");

	LLVMValueRef all_ones = LLVMConstInt(ctx->i32, -1, true);
	LLVMValueRef cond = LLVMBuildOr(ctx->builder,
					LLVMBuildICmp(ctx->builder, LLVMIntEQ,
						      arg, LLVMConstInt(ctx->i32, 0, 0), ""),
					LLVMBuildICmp(ctx->builder, LLVMIntEQ,
						      arg, all_ones, ""), "");

	return LLVMBuildSelect(ctx->builder, cond, all_ones, msb, "");
}

LLVMValueRef
ac_build_umsb(struct ac_llvm_context *ctx,
	      LLVMValueRef arg,
	      LLVMTypeRef dst_type)
{
	LLVMValueRef args[2] = {
		arg,
		ctx->i1true,
	};
	LLVMValueRef msb = ac_build_intrinsic(ctx, "llvm.ctlz.i32",
					      dst_type, args, ARRAY_SIZE(args),
					      AC_FUNC_ATTR_READNONE);

	/* The HW returns the last bit index from MSB, but TGSI/NIR wants
	 * the index from LSB. Invert it by doing "31 - msb". */
	msb = LLVMBuildSub(ctx->builder, LLVMConstInt(ctx->i32, 31, false),
			   msb, "");

	/* check for zero */
	return LLVMBuildSelect(ctx->builder,
			       LLVMBuildICmp(ctx->builder, LLVMIntEQ, arg,
					     LLVMConstInt(ctx->i32, 0, 0), ""),
			       LLVMConstInt(ctx->i32, -1, true), msb, "");
}

LLVMValueRef ac_build_fmin(struct ac_llvm_context *ctx, LLVMValueRef a,
			   LLVMValueRef b)
{
	LLVMValueRef args[2] = {a, b};
	return ac_build_intrinsic(ctx, "llvm.minnum.f32", ctx->f32, args, 2,
				  AC_FUNC_ATTR_READNONE);
}

LLVMValueRef ac_build_fmax(struct ac_llvm_context *ctx, LLVMValueRef a,
			   LLVMValueRef b)
{
	LLVMValueRef args[2] = {a, b};
	return ac_build_intrinsic(ctx, "llvm.maxnum.f32", ctx->f32, args, 2,
				  AC_FUNC_ATTR_READNONE);
}

LLVMValueRef ac_build_umin(struct ac_llvm_context *ctx, LLVMValueRef a,
			   LLVMValueRef b)
{
	LLVMValueRef cmp = LLVMBuildICmp(ctx->builder, LLVMIntULE, a, b, "");
	return LLVMBuildSelect(ctx->builder, cmp, a, b, "");
}

LLVMValueRef ac_build_clamp(struct ac_llvm_context *ctx, LLVMValueRef value)
{
	if (HAVE_LLVM >= 0x0500) {
		return ac_build_fmin(ctx, ac_build_fmax(ctx, value, ctx->f32_0),
				     ctx->f32_1);
	}

	LLVMValueRef args[3] = {
		value,
		LLVMConstReal(ctx->f32, 0),
		LLVMConstReal(ctx->f32, 1),
	};

	return ac_build_intrinsic(ctx, "llvm.AMDGPU.clamp.", ctx->f32, args, 3,
				  AC_FUNC_ATTR_READNONE |
				  AC_FUNC_ATTR_LEGACY);
}

void ac_build_export(struct ac_llvm_context *ctx, struct ac_export_args *a)
{
	LLVMValueRef args[9];

	if (HAVE_LLVM >= 0x0500) {
		args[0] = LLVMConstInt(ctx->i32, a->target, 0);
		args[1] = LLVMConstInt(ctx->i32, a->enabled_channels, 0);

		if (a->compr) {
			LLVMTypeRef i16 = LLVMInt16TypeInContext(ctx->context);
			LLVMTypeRef v2i16 = LLVMVectorType(i16, 2);

			args[2] = LLVMBuildBitCast(ctx->builder, a->out[0],
						   v2i16, "");
			args[3] = LLVMBuildBitCast(ctx->builder, a->out[1],
						   v2i16, "");
			args[4] = LLVMConstInt(ctx->i1, a->done, 0);
			args[5] = LLVMConstInt(ctx->i1, a->valid_mask, 0);

			ac_build_intrinsic(ctx, "llvm.amdgcn.exp.compr.v2i16",
					   ctx->voidt, args, 6, 0);
		} else {
			args[2] = a->out[0];
			args[3] = a->out[1];
			args[4] = a->out[2];
			args[5] = a->out[3];
			args[6] = LLVMConstInt(ctx->i1, a->done, 0);
			args[7] = LLVMConstInt(ctx->i1, a->valid_mask, 0);

			ac_build_intrinsic(ctx, "llvm.amdgcn.exp.f32",
					   ctx->voidt, args, 8, 0);
		}
		return;
	}

	args[0] = LLVMConstInt(ctx->i32, a->enabled_channels, 0);
	args[1] = LLVMConstInt(ctx->i32, a->valid_mask, 0);
	args[2] = LLVMConstInt(ctx->i32, a->done, 0);
	args[3] = LLVMConstInt(ctx->i32, a->target, 0);
	args[4] = LLVMConstInt(ctx->i32, a->compr, 0);
	memcpy(args + 5, a->out, sizeof(a->out[0]) * 4);

	ac_build_intrinsic(ctx, "llvm.SI.export", ctx->voidt, args, 9,
			   AC_FUNC_ATTR_LEGACY);
}

LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx,
				   struct ac_image_args *a)
{
	LLVMTypeRef dst_type;
	LLVMValueRef args[11];
	unsigned num_args = 0;
	const char *name = NULL;
	char intr_name[128], type[64];

	if (HAVE_LLVM >= 0x0400) {
		bool sample = a->opcode == ac_image_sample ||
			      a->opcode == ac_image_gather4 ||
			      a->opcode == ac_image_get_lod;

		if (sample)
			args[num_args++] = ac_to_float(ctx, a->addr);
		else
			args[num_args++] = a->addr;

		args[num_args++] = a->resource;
		if (sample)
			args[num_args++] = a->sampler;
		args[num_args++] = LLVMConstInt(ctx->i32, a->dmask, 0);
		if (sample)
			args[num_args++] = LLVMConstInt(ctx->i1, a->unorm, 0);
		args[num_args++] = ctx->i1false; /* glc */
		args[num_args++] = ctx->i1false; /* slc */
		args[num_args++] = ctx->i1false; /* lwe */
		args[num_args++] = LLVMConstInt(ctx->i1, a->da, 0);

		switch (a->opcode) {
		case ac_image_sample:
			name = "llvm.amdgcn.image.sample";
			break;
		case ac_image_gather4:
			name = "llvm.amdgcn.image.gather4";
			break;
		case ac_image_load:
			name = "llvm.amdgcn.image.load";
			break;
		case ac_image_load_mip:
			name = "llvm.amdgcn.image.load.mip";
			break;
		case ac_image_get_lod:
			name = "llvm.amdgcn.image.getlod";
			break;
		case ac_image_get_resinfo:
			name = "llvm.amdgcn.image.getresinfo";
			break;
		default:
			unreachable("invalid image opcode");
		}

		ac_build_type_name_for_intr(LLVMTypeOf(args[0]), type,
					    sizeof(type));

		snprintf(intr_name, sizeof(intr_name), "%s%s%s%s.v4f32.%s.v8i32",
			name,
			a->compare ? ".c" : "",
			a->bias ? ".b" :
			a->lod ? ".l" :
			a->deriv ? ".d" :
			a->level_zero ? ".lz" : "",
			a->offset ? ".o" : "",
			type);

		LLVMValueRef result =
			ac_build_intrinsic(ctx, intr_name,
					   ctx->v4f32, args, num_args,
					   AC_FUNC_ATTR_READNONE);
		if (!sample) {
			result = LLVMBuildBitCast(ctx->builder, result,
						  ctx->v4i32, "");
		}
		return result;
	}

	args[num_args++] = a->addr;
	args[num_args++] = a->resource;

	if (a->opcode == ac_image_load ||
	    a->opcode == ac_image_load_mip ||
	    a->opcode == ac_image_get_resinfo) {
		dst_type = ctx->v4i32;
	} else {
		dst_type = ctx->v4f32;
		args[num_args++] = a->sampler;
	}

	args[num_args++] = LLVMConstInt(ctx->i32, a->dmask, 0);
	args[num_args++] = LLVMConstInt(ctx->i32, a->unorm, 0);
	args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* r128 */
	args[num_args++] = LLVMConstInt(ctx->i32, a->da, 0);
	args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* glc */
	args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* slc */
	args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* tfe */
	args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* lwe */

	switch (a->opcode) {
	case ac_image_sample:
		name = "llvm.SI.image.sample";
		break;
	case ac_image_gather4:
		name = "llvm.SI.gather4";
		break;
	case ac_image_load:
		name = "llvm.SI.image.load";
		break;
	case ac_image_load_mip:
		name = "llvm.SI.image.load.mip";
		break;
	case ac_image_get_lod:
		name = "llvm.SI.getlod";
		break;
	case ac_image_get_resinfo:
		name = "llvm.SI.getresinfo";
		break;
	}

	ac_build_type_name_for_intr(LLVMTypeOf(a->addr), type, sizeof(type));
	snprintf(intr_name, sizeof(intr_name), "%s%s%s%s.%s",
		name,
		a->compare ? ".c" : "",
		a->bias ? ".b" :
		a->lod ? ".l" :
		a->deriv ? ".d" :
		a->level_zero ? ".lz" : "",
		a->offset ? ".o" : "",
		type);

	return ac_build_intrinsic(ctx, intr_name,
				  dst_type, args, num_args,
				  AC_FUNC_ATTR_READNONE |
				  AC_FUNC_ATTR_LEGACY);
}

LLVMValueRef ac_build_cvt_pkrtz_f16(struct ac_llvm_context *ctx,
				    LLVMValueRef args[2])
{
	if (HAVE_LLVM >= 0x0500) {
		LLVMTypeRef v2f16 =
			LLVMVectorType(LLVMHalfTypeInContext(ctx->context), 2);
		LLVMValueRef res =
			ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pkrtz",
					   v2f16, args, 2,
					   AC_FUNC_ATTR_READNONE);
		return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
	}

	return ac_build_intrinsic(ctx, "llvm.SI.packf16", ctx->i32, args, 2,
				  AC_FUNC_ATTR_READNONE |
				  AC_FUNC_ATTR_LEGACY);
}

LLVMValueRef ac_build_wqm_vote(struct ac_llvm_context *ctx, LLVMValueRef i1)
{
	assert(HAVE_LLVM >= 0x0600);
	return ac_build_intrinsic(ctx, "llvm.amdgcn.wqm.vote", ctx->i1,
				  &i1, 1, AC_FUNC_ATTR_READNONE);
}

void ac_build_kill_if_false(struct ac_llvm_context *ctx, LLVMValueRef i1)
{
	if (HAVE_LLVM >= 0x0600) {
		ac_build_intrinsic(ctx, "llvm.amdgcn.kill", ctx->voidt,
				   &i1, 1, 0);
		return;
	}

	LLVMValueRef value = LLVMBuildSelect(ctx->builder, i1,
					     LLVMConstReal(ctx->f32, 1),
					     LLVMConstReal(ctx->f32, -1), "");
	ac_build_intrinsic(ctx, "llvm.AMDGPU.kill", ctx->voidt,
			   &value, 1, AC_FUNC_ATTR_LEGACY);
}

LLVMValueRef ac_build_bfe(struct ac_llvm_context *ctx, LLVMValueRef input,
			  LLVMValueRef offset, LLVMValueRef width,
			  bool is_signed)
{
	LLVMValueRef args[] = {
		input,
		offset,
		width,
	};

	if (HAVE_LLVM >= 0x0500) {
		return ac_build_intrinsic(ctx,
					  is_signed ? "llvm.amdgcn.sbfe.i32" :
						      "llvm.amdgcn.ubfe.i32",
					  ctx->i32, args, 3,
					  AC_FUNC_ATTR_READNONE);
	}

	return ac_build_intrinsic(ctx,
				  is_signed ? "llvm.AMDGPU.bfe.i32" :
					      "llvm.AMDGPU.bfe.u32",
				  ctx->i32, args, 3,
				  AC_FUNC_ATTR_READNONE |
				  AC_FUNC_ATTR_LEGACY);
}

void ac_build_waitcnt(struct ac_llvm_context *ctx, unsigned simm16)
{
	LLVMValueRef args[1] = {
		LLVMConstInt(ctx->i32, simm16, false),
	};
	ac_build_intrinsic(ctx, "llvm.amdgcn.s.waitcnt",
			   ctx->voidt, args, 1, 0);
}

void ac_get_image_intr_name(const char *base_name,
			    LLVMTypeRef data_type,
			    LLVMTypeRef coords_type,
			    LLVMTypeRef rsrc_type,
			    char *out_name, unsigned out_len)
{
        char coords_type_name[8];

        ac_build_type_name_for_intr(coords_type, coords_type_name,
                            sizeof(coords_type_name));

        if (HAVE_LLVM <= 0x0309) {
                snprintf(out_name, out_len, "%s.%s", base_name, coords_type_name);
        } else {
                char data_type_name[8];
                char rsrc_type_name[8];

                ac_build_type_name_for_intr(data_type, data_type_name,
                                        sizeof(data_type_name));
                ac_build_type_name_for_intr(rsrc_type, rsrc_type_name,
                                        sizeof(rsrc_type_name));
                snprintf(out_name, out_len, "%s.%s.%s.%s", base_name,
                         data_type_name, coords_type_name, rsrc_type_name);
        }
}

#define AC_EXP_TARGET (HAVE_LLVM >= 0x0500 ? 0 : 3)
#define AC_EXP_OUT0 (HAVE_LLVM >= 0x0500 ? 2 : 5)

enum ac_ir_type {
	AC_IR_UNDEF,
	AC_IR_CONST,
	AC_IR_VALUE,
};

struct ac_vs_exp_chan
{
	LLVMValueRef value;
	float const_float;
	enum ac_ir_type type;
};

struct ac_vs_exp_inst {
	unsigned offset;
	LLVMValueRef inst;
	struct ac_vs_exp_chan chan[4];
};

struct ac_vs_exports {
	unsigned num;
	struct ac_vs_exp_inst exp[VARYING_SLOT_MAX];
};

/* Return true if the PARAM export has been eliminated. */
static bool ac_eliminate_const_output(uint8_t *vs_output_param_offset,
				      uint32_t num_outputs,
				      struct ac_vs_exp_inst *exp)
{
	unsigned i, default_val; /* SPI_PS_INPUT_CNTL_i.DEFAULT_VAL */
	bool is_zero[4] = {}, is_one[4] = {};

	for (i = 0; i < 4; i++) {
		/* It's a constant expression. Undef outputs are eliminated too. */
		if (exp->chan[i].type == AC_IR_UNDEF) {
			is_zero[i] = true;
			is_one[i] = true;
		} else if (exp->chan[i].type == AC_IR_CONST) {
			if (exp->chan[i].const_float == 0)
				is_zero[i] = true;
			else if (exp->chan[i].const_float == 1)
				is_one[i] = true;
			else
				return false; /* other constant */
		} else
			return false;
	}

	/* Only certain combinations of 0 and 1 can be eliminated. */
	if (is_zero[0] && is_zero[1] && is_zero[2])
		default_val = is_zero[3] ? 0 : 1;
	else if (is_one[0] && is_one[1] && is_one[2])
		default_val = is_zero[3] ? 2 : 3;
	else
		return false;

	/* The PARAM export can be represented as DEFAULT_VAL. Kill it. */
	LLVMInstructionEraseFromParent(exp->inst);

	/* Change OFFSET to DEFAULT_VAL. */
	for (i = 0; i < num_outputs; i++) {
		if (vs_output_param_offset[i] == exp->offset) {
			vs_output_param_offset[i] =
				AC_EXP_PARAM_DEFAULT_VAL_0000 + default_val;
			break;
		}
	}
	return true;
}

static bool ac_eliminate_duplicated_output(uint8_t *vs_output_param_offset,
					   uint32_t num_outputs,
					   struct ac_vs_exports *processed,
				           struct ac_vs_exp_inst *exp)
{
	unsigned p, copy_back_channels = 0;

	/* See if the output is already in the list of processed outputs.
	 * The LLVMValueRef comparison relies on SSA.
	 */
	for (p = 0; p < processed->num; p++) {
		bool different = false;

		for (unsigned j = 0; j < 4; j++) {
			struct ac_vs_exp_chan *c1 = &processed->exp[p].chan[j];
			struct ac_vs_exp_chan *c2 = &exp->chan[j];

			/* Treat undef as a match. */
			if (c2->type == AC_IR_UNDEF)
				continue;

			/* If c1 is undef but c2 isn't, we can copy c2 to c1
			 * and consider the instruction duplicated.
			 */
			if (c1->type == AC_IR_UNDEF) {
				copy_back_channels |= 1 << j;
				continue;
			}

			/* Test whether the channels are not equal. */
			if (c1->type != c2->type ||
			    (c1->type == AC_IR_CONST &&
			     c1->const_float != c2->const_float) ||
			    (c1->type == AC_IR_VALUE &&
			     c1->value != c2->value)) {
				different = true;
				break;
			}
		}
		if (!different)
			break;

		copy_back_channels = 0;
	}
	if (p == processed->num)
		return false;

	/* If a match was found, but the matching export has undef where the new
	 * one has a normal value, copy the normal value to the undef channel.
	 */
	struct ac_vs_exp_inst *match = &processed->exp[p];

	while (copy_back_channels) {
		unsigned chan = u_bit_scan(&copy_back_channels);

		assert(match->chan[chan].type == AC_IR_UNDEF);
		LLVMSetOperand(match->inst, AC_EXP_OUT0 + chan,
			       exp->chan[chan].value);
		match->chan[chan] = exp->chan[chan];
	}

	/* The PARAM export is duplicated. Kill it. */
	LLVMInstructionEraseFromParent(exp->inst);

	/* Change OFFSET to the matching export. */
	for (unsigned i = 0; i < num_outputs; i++) {
		if (vs_output_param_offset[i] == exp->offset) {
			vs_output_param_offset[i] = match->offset;
			break;
		}
	}
	return true;
}

void ac_optimize_vs_outputs(struct ac_llvm_context *ctx,
			    LLVMValueRef main_fn,
			    uint8_t *vs_output_param_offset,
			    uint32_t num_outputs,
			    uint8_t *num_param_exports)
{
	LLVMBasicBlockRef bb;
	bool removed_any = false;
	struct ac_vs_exports exports;

	exports.num = 0;

	/* Process all LLVM instructions. */
	bb = LLVMGetFirstBasicBlock(main_fn);
	while (bb) {
		LLVMValueRef inst = LLVMGetFirstInstruction(bb);

		while (inst) {
			LLVMValueRef cur = inst;
			inst = LLVMGetNextInstruction(inst);
			struct ac_vs_exp_inst exp;

			if (LLVMGetInstructionOpcode(cur) != LLVMCall)
				continue;

			LLVMValueRef callee = ac_llvm_get_called_value(cur);

			if (!ac_llvm_is_function(callee))
				continue;

			const char *name = LLVMGetValueName(callee);
			unsigned num_args = LLVMCountParams(callee);

			/* Check if this is an export instruction. */
			if ((num_args != 9 && num_args != 8) ||
			    (strcmp(name, "llvm.SI.export") &&
			     strcmp(name, "llvm.amdgcn.exp.f32")))
				continue;

			LLVMValueRef arg = LLVMGetOperand(cur, AC_EXP_TARGET);
			unsigned target = LLVMConstIntGetZExtValue(arg);

			if (target < V_008DFC_SQ_EXP_PARAM)
				continue;

			target -= V_008DFC_SQ_EXP_PARAM;

			/* Parse the instruction. */
			memset(&exp, 0, sizeof(exp));
			exp.offset = target;
			exp.inst = cur;

			for (unsigned i = 0; i < 4; i++) {
				LLVMValueRef v = LLVMGetOperand(cur, AC_EXP_OUT0 + i);

				exp.chan[i].value = v;

				if (LLVMIsUndef(v)) {
					exp.chan[i].type = AC_IR_UNDEF;
				} else if (LLVMIsAConstantFP(v)) {
					LLVMBool loses_info;
					exp.chan[i].type = AC_IR_CONST;
					exp.chan[i].const_float =
						LLVMConstRealGetDouble(v, &loses_info);
				} else {
					exp.chan[i].type = AC_IR_VALUE;
				}
			}

			/* Eliminate constant and duplicated PARAM exports. */
			if (ac_eliminate_const_output(vs_output_param_offset,
						      num_outputs, &exp) ||
			    ac_eliminate_duplicated_output(vs_output_param_offset,
							   num_outputs, &exports,
							   &exp)) {
				removed_any = true;
			} else {
				exports.exp[exports.num++] = exp;
			}
		}
		bb = LLVMGetNextBasicBlock(bb);
	}

	/* Remove holes in export memory due to removed PARAM exports.
	 * This is done by renumbering all PARAM exports.
	 */
	if (removed_any) {
		uint8_t old_offset[VARYING_SLOT_MAX];
		unsigned out, i;

		/* Make a copy of the offsets. We need the old version while
		 * we are modifying some of them. */
		memcpy(old_offset, vs_output_param_offset,
		       sizeof(old_offset));

		for (i = 0; i < exports.num; i++) {
			unsigned offset = exports.exp[i].offset;

			/* Update vs_output_param_offset. Multiple outputs can
			 * have the same offset.
			 */
			for (out = 0; out < num_outputs; out++) {
				if (old_offset[out] == offset)
					vs_output_param_offset[out] = i;
			}

			/* Change the PARAM offset in the instruction. */
			LLVMSetOperand(exports.exp[i].inst, AC_EXP_TARGET,
				       LLVMConstInt(ctx->i32,
						    V_008DFC_SQ_EXP_PARAM + i, 0));
		}
		*num_param_exports = exports.num;
	}
}

void ac_init_exec_full_mask(struct ac_llvm_context *ctx)
{
	LLVMValueRef full_mask = LLVMConstInt(ctx->i64, ~0ull, 0);
	ac_build_intrinsic(ctx,
			   "llvm.amdgcn.init.exec", ctx->voidt,
			   &full_mask, 1, AC_FUNC_ATTR_CONVERGENT);
}

void ac_declare_lds_as_pointer(struct ac_llvm_context *ctx)
{
	unsigned lds_size = ctx->chip_class >= CIK ? 65536 : 32768;
	ctx->lds = LLVMBuildIntToPtr(ctx->builder, ctx->i32_0,
				     LLVMPointerType(LLVMArrayType(ctx->i32, lds_size / 4), AC_LOCAL_ADDR_SPACE),
				     "lds");
}

LLVMValueRef ac_lds_load(struct ac_llvm_context *ctx,
			 LLVMValueRef dw_addr)
{
	return ac_build_load(ctx, ctx->lds, dw_addr);
}

void ac_lds_store(struct ac_llvm_context *ctx,
		  LLVMValueRef dw_addr,
		  LLVMValueRef value)
{
	value = ac_to_integer(ctx, value);
	ac_build_indexed_store(ctx, ctx->lds,
			       dw_addr, value);
}

LLVMValueRef ac_find_lsb(struct ac_llvm_context *ctx,
			 LLVMTypeRef dst_type,
			 LLVMValueRef src0)
{
	LLVMValueRef params[2] = {
		src0,

		/* The value of 1 means that ffs(x=0) = undef, so LLVM won't
		 * add special code to check for x=0. The reason is that
		 * the LLVM behavior for x=0 is different from what we
		 * need here. However, LLVM also assumes that ffs(x) is
		 * in [0, 31], but GLSL expects that ffs(0) = -1, so
		 * a conditional assignment to handle 0 is still required.
		 *
		 * The hardware already implements the correct behavior.
		 */
		LLVMConstInt(ctx->i1, 1, false),
	};

	LLVMValueRef lsb = ac_build_intrinsic(ctx, "llvm.cttz.i32", ctx->i32,
					      params, 2,
					      AC_FUNC_ATTR_READNONE);

	/* TODO: We need an intrinsic to skip this conditional. */
	/* Check for zero: */
	return LLVMBuildSelect(ctx->builder, LLVMBuildICmp(ctx->builder,
							   LLVMIntEQ, src0,
							   ctx->i32_0, ""),
			       LLVMConstInt(ctx->i32, -1, 0), lsb, "");
}

static struct ac_llvm_flow *
get_current_flow(struct ac_llvm_context *ctx)
{
	if (ctx->flow_depth > 0)
		return &ctx->flow[ctx->flow_depth - 1];
	return NULL;
}

static struct ac_llvm_flow *
get_innermost_loop(struct ac_llvm_context *ctx)
{
	for (unsigned i = ctx->flow_depth; i > 0; --i) {
		if (ctx->flow[i - 1].loop_entry_block)
			return &ctx->flow[i - 1];
	}
	return NULL;
}

static struct ac_llvm_flow *
push_flow(struct ac_llvm_context *ctx)
{
	struct ac_llvm_flow *flow;

	if (ctx->flow_depth >= ctx->flow_depth_max) {
		unsigned new_max = MAX2(ctx->flow_depth << 1,
					AC_LLVM_INITIAL_CF_DEPTH);

		ctx->flow = realloc(ctx->flow, new_max * sizeof(*ctx->flow));
		ctx->flow_depth_max = new_max;
	}

	flow = &ctx->flow[ctx->flow_depth];
	ctx->flow_depth++;

	flow->next_block = NULL;
	flow->loop_entry_block = NULL;
	return flow;
}

static void set_basicblock_name(LLVMBasicBlockRef bb, const char *base,
				int label_id)
{
	char buf[32];
	snprintf(buf, sizeof(buf), "%s%d", base, label_id);
	LLVMSetValueName(LLVMBasicBlockAsValue(bb), buf);
}

/* Append a basic block at the level of the parent flow.
 */
static LLVMBasicBlockRef append_basic_block(struct ac_llvm_context *ctx,
					    const char *name)
{
	assert(ctx->flow_depth >= 1);

	if (ctx->flow_depth >= 2) {
		struct ac_llvm_flow *flow = &ctx->flow[ctx->flow_depth - 2];

		return LLVMInsertBasicBlockInContext(ctx->context,
						     flow->next_block, name);
	}

	LLVMValueRef main_fn =
		LLVMGetBasicBlockParent(LLVMGetInsertBlock(ctx->builder));
	return LLVMAppendBasicBlockInContext(ctx->context, main_fn, name);
}

/* Emit a branch to the given default target for the current block if
 * applicable -- that is, if the current block does not already contain a
 * branch from a break or continue.
 */
static void emit_default_branch(LLVMBuilderRef builder,
				LLVMBasicBlockRef target)
{
	if (!LLVMGetBasicBlockTerminator(LLVMGetInsertBlock(builder)))
		 LLVMBuildBr(builder, target);
}

void ac_build_bgnloop(struct ac_llvm_context *ctx, int label_id)
{
	struct ac_llvm_flow *flow = push_flow(ctx);
	flow->loop_entry_block = append_basic_block(ctx, "LOOP");
	flow->next_block = append_basic_block(ctx, "ENDLOOP");
	set_basicblock_name(flow->loop_entry_block, "loop", label_id);
	LLVMBuildBr(ctx->builder, flow->loop_entry_block);
	LLVMPositionBuilderAtEnd(ctx->builder, flow->loop_entry_block);
}

void ac_build_break(struct ac_llvm_context *ctx)
{
	struct ac_llvm_flow *flow = get_innermost_loop(ctx);
	LLVMBuildBr(ctx->builder, flow->next_block);
}

void ac_build_continue(struct ac_llvm_context *ctx)
{
	struct ac_llvm_flow *flow = get_innermost_loop(ctx);
	LLVMBuildBr(ctx->builder, flow->loop_entry_block);
}

void ac_build_else(struct ac_llvm_context *ctx, int label_id)
{
	struct ac_llvm_flow *current_branch = get_current_flow(ctx);
	LLVMBasicBlockRef endif_block;

	assert(!current_branch->loop_entry_block);

	endif_block = append_basic_block(ctx, "ENDIF");
	emit_default_branch(ctx->builder, endif_block);

	LLVMPositionBuilderAtEnd(ctx->builder, current_branch->next_block);
	set_basicblock_name(current_branch->next_block, "else", label_id);

	current_branch->next_block = endif_block;
}

void ac_build_endif(struct ac_llvm_context *ctx, int label_id)
{
	struct ac_llvm_flow *current_branch = get_current_flow(ctx);

	assert(!current_branch->loop_entry_block);

	emit_default_branch(ctx->builder, current_branch->next_block);
	LLVMPositionBuilderAtEnd(ctx->builder, current_branch->next_block);
	set_basicblock_name(current_branch->next_block, "endif", label_id);

	ctx->flow_depth--;
}

void ac_build_endloop(struct ac_llvm_context *ctx, int label_id)
{
	struct ac_llvm_flow *current_loop = get_current_flow(ctx);

	assert(current_loop->loop_entry_block);

	emit_default_branch(ctx->builder, current_loop->loop_entry_block);

	LLVMPositionBuilderAtEnd(ctx->builder, current_loop->next_block);
	set_basicblock_name(current_loop->next_block, "endloop", label_id);
	ctx->flow_depth--;
}

static void if_cond_emit(struct ac_llvm_context *ctx, LLVMValueRef cond,
			 int label_id)
{
	struct ac_llvm_flow *flow = push_flow(ctx);
	LLVMBasicBlockRef if_block;

	if_block = append_basic_block(ctx, "IF");
	flow->next_block = append_basic_block(ctx, "ELSE");
	set_basicblock_name(if_block, "if", label_id);
	LLVMBuildCondBr(ctx->builder, cond, if_block, flow->next_block);
	LLVMPositionBuilderAtEnd(ctx->builder, if_block);
}

void ac_build_if(struct ac_llvm_context *ctx, LLVMValueRef value,
		 int label_id)
{
	LLVMValueRef cond = LLVMBuildFCmp(ctx->builder, LLVMRealUNE,
					  value, ctx->f32_0, "");
	if_cond_emit(ctx, cond, label_id);
}

void ac_build_uif(struct ac_llvm_context *ctx, LLVMValueRef value,
		  int label_id)
{
	LLVMValueRef cond = LLVMBuildICmp(ctx->builder, LLVMIntNE,
					  ac_to_integer(ctx, value),
					  ctx->i32_0, "");
	if_cond_emit(ctx, cond, label_id);
}