summaryrefslogtreecommitdiff
path: root/src/freedreno/ir3/ir3.h
blob: fec6fe253c80aad6bb62d5a1b10b1a364bcd7f22 (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
/*
 * Copyright (c) 2013 Rob Clark <robdclark@gmail.com>
 *
 * 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, sublicense,
 * 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 above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS 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.
 */

#ifndef IR3_H_
#define IR3_H_

#include <stdint.h>
#include <stdbool.h>

#include "compiler/shader_enums.h"

#include "util/bitscan.h"
#include "util/list.h"
#include "util/set.h"
#include "util/u_debug.h"

#include "instr-a3xx.h"

/* low level intermediate representation of an adreno shader program */

struct ir3_compiler;
struct ir3;
struct ir3_instruction;
struct ir3_block;

struct ir3_info {
	void *data;              /* used internally in ir3 assembler */
	/* Size in bytes of the shader binary, including NIR constants and
	 * padding
	 */
	uint32_t size;
	/* byte offset from start of the shader to the NIR constant data. */
	uint32_t constant_data_offset;
	/* Size in dwords of the instructions. */
	uint16_t sizedwords;
	uint16_t instrs_count;   /* expanded to account for rpt's */
	uint16_t nops_count;     /* # of nop instructions, including nopN */
	uint16_t mov_count;
	uint16_t cov_count;
	/* NOTE: max_reg, etc, does not include registers not touched
	 * by the shader (ie. vertex fetched via VFD_DECODE but not
	 * touched by shader)
	 */
	int8_t   max_reg;   /* highest GPR # used by shader */
	int8_t   max_half_reg;
	int16_t  max_const;
	/* This is the maximum # of waves that can executed at once in one core,
	 * assuming that they are all executing this shader.
	 */
	int8_t   max_waves;
	bool     double_threadsize;
	bool     multi_dword_ldp_stp;

	/* number of sync bits: */
	uint16_t ss, sy;

	/* estimate of number of cycles stalled on (ss) */
	uint16_t sstall;

	uint16_t last_baryf;     /* instruction # of last varying fetch */

	/* Number of instructions of a given category: */
	uint16_t instrs_per_cat[8];
};

struct ir3_merge_set {
	uint16_t preferred_reg;
	uint16_t size;
	uint16_t alignment;

	unsigned interval_start;

	unsigned regs_count;
	struct ir3_register **regs;
};

struct ir3_register {
	enum {
		IR3_REG_CONST  = 0x001,
		IR3_REG_IMMED  = 0x002,
		IR3_REG_HALF   = 0x004,
		/* Shared registers have the same value for all threads when read.
		 * They can only be written when one thread is active (that is, inside
		 * a "getone" block).
		 */
		IR3_REG_SHARED = 0x008,
		IR3_REG_RELATIV= 0x010,
		IR3_REG_R      = 0x020,
		/* Most instructions, it seems, can do float abs/neg but not
		 * integer.  The CP pass needs to know what is intended (int or
		 * float) in order to do the right thing.  For this reason the
		 * abs/neg flags are split out into float and int variants.  In
		 * addition, .b (bitwise) operations, the negate is actually a
		 * bitwise not, so split that out into a new flag to make it
		 * more clear.
		 */
		IR3_REG_FNEG   = 0x040,
		IR3_REG_FABS   = 0x080,
		IR3_REG_SNEG   = 0x100,
		IR3_REG_SABS   = 0x200,
		IR3_REG_BNOT   = 0x400,
		/* (ei) flag, end-input?  Set on last bary, presumably to signal
		 * that the shader needs no more input:
		 */
		IR3_REG_EI     = 0x2000,
		/* meta-flags, for intermediate stages of IR, ie.
		 * before register assignment is done:
		 */
		IR3_REG_SSA    = 0x4000,   /* 'instr' is ptr to assigning instr */
		IR3_REG_ARRAY  = 0x8000,

		IR3_REG_DEST   = 0x10000,
		IR3_REG_KILL = 0x20000,
		IR3_REG_FIRST_KILL = 0x40000,
		IR3_REG_UNUSED = 0x80000,
	} flags;

	/* used for cat5 instructions, but also for internal/IR level
	 * tracking of what registers are read/written by an instruction.
	 * wrmask may be a bad name since it is used to represent both
	 * src and dst that touch multiple adjacent registers.
	 */
	unsigned wrmask : 16;  /* up to vec16 */

	/* for relative addressing, 32bits for array size is too small,
	 * but otoh we don't need to deal with disjoint sets, so instead
	 * use a simple size field (number of scalar components).
	 *
	 * Note the size field isn't important for relative const (since
	 * we don't have to do register allocation for constants).
	 */
	unsigned size : 16;

	/* normal registers:
	 * the component is in the low two bits of the reg #, so
	 * rN.x becomes: (N << 2) | x
	 */
	uint16_t num;
	uint16_t name;
	union {
		/* immediate: */
		int32_t  iim_val;
		uint32_t uim_val;
		float    fim_val;
		/* relative: */
		struct {
			uint16_t id;
			int16_t offset;
			uint16_t base;
		} array;
	};


	/* For IR3_REG_DEST, pointer back to the instruction containing this
	 * register.
	 */
	struct ir3_instruction *instr;

	/* For IR3_REG_SSA, src registers contain ptr back to assigning
	 * instruction.
	 *
	 * For IR3_REG_ARRAY, the pointer is back to the last dependent
	 * array access (although the net effect is the same, it points
	 * back to a previous instruction that we depend on).
	 */
	struct ir3_register *def;

	/* Pointer to another register in the instruction that must share the same
	 * physical register. Each destination can be tied with one source, and
	 * they must have "tied" pointing to each other.
	 */
	struct ir3_register *tied;

	unsigned merge_set_offset;
	struct ir3_merge_set *merge_set;
	unsigned interval_start, interval_end;
};

/*
 * Stupid/simple growable array implementation:
 */
#define DECLARE_ARRAY(type, name) \
	unsigned name ## _count, name ## _sz; \
	type * name;

#define array_insert(ctx, arr, ...) do { \
		if (arr ## _count == arr ## _sz) { \
			arr ## _sz = MAX2(2 * arr ## _sz, 16); \
			arr = reralloc_size(ctx, arr, arr ## _sz * sizeof(arr[0])); \
		} \
		arr[arr ##_count++] = __VA_ARGS__; \
	} while (0)

struct ir3_instruction {
	struct ir3_block *block;
	opc_t opc;
	enum {
		/* (sy) flag is set on first instruction, and after sample
		 * instructions (probably just on RAW hazard).
		 */
		IR3_INSTR_SY    = 0x001,
		/* (ss) flag is set on first instruction, and first instruction
		 * to depend on the result of "long" instructions (RAW hazard):
		 *
		 *   rcp, rsq, log2, exp2, sin, cos, sqrt
		 *
		 * It seems to synchronize until all in-flight instructions are
		 * completed, for example:
		 *
		 *   rsq hr1.w, hr1.w
		 *   add.f hr2.z, (neg)hr2.z, hc0.y
		 *   mul.f hr2.w, (neg)hr2.y, (neg)hr2.y
		 *   rsq hr2.x, hr2.x
		 *   (rpt1)nop
		 *   mad.f16 hr2.w, hr2.z, hr2.z, hr2.w
		 *   nop
		 *   mad.f16 hr2.w, (neg)hr0.w, (neg)hr0.w, hr2.w
		 *   (ss)(rpt2)mul.f hr1.x, (r)hr1.x, hr1.w
		 *   (rpt2)mul.f hr0.x, (neg)(r)hr0.x, hr2.x
		 *
		 * The last mul.f does not have (ss) set, presumably because the
		 * (ss) on the previous instruction does the job.
		 *
		 * The blob driver also seems to set it on WAR hazards, although
		 * not really clear if this is needed or just blob compiler being
		 * sloppy.  So far I haven't found a case where removing the (ss)
		 * causes problems for WAR hazard, but I could just be getting
		 * lucky:
		 *
		 *   rcp r1.y, r3.y
		 *   (ss)(rpt2)mad.f32 r3.y, (r)c9.x, r1.x, (r)r3.z
		 *
		 */
		IR3_INSTR_SS    = 0x002,
		/* (jp) flag is set on jump targets:
		 */
		IR3_INSTR_JP    = 0x004,
		IR3_INSTR_UL    = 0x008,
		IR3_INSTR_3D    = 0x010,
		IR3_INSTR_A     = 0x020,
		IR3_INSTR_O     = 0x040,
		IR3_INSTR_P     = 0x080,
		IR3_INSTR_S     = 0x100,
		IR3_INSTR_S2EN  = 0x200,
		IR3_INSTR_G     = 0x400,
		IR3_INSTR_SAT   = 0x800,
		/* (cat5/cat6) Bindless */
		IR3_INSTR_B     = 0x1000,
		/* (cat5/cat6) nonuniform */
		IR3_INSTR_NONUNIF    = 0x02000,
		/* (cat5-only) Get some parts of the encoding from a1.x */
		IR3_INSTR_A1EN       = 0x04000,
		/* meta-flags, for intermediate stages of IR, ie.
		 * before register assignment is done:
		 */
		IR3_INSTR_MARK       = 0x08000,
		IR3_INSTR_UNUSED     = 0x10000,
	} flags;
	uint8_t repeat;
	uint8_t nop;
#ifdef DEBUG
	unsigned regs_max, srcs_max, dsts_max;
#endif
	unsigned regs_count, srcs_count, dsts_count;
	struct ir3_register **regs;
	struct ir3_register **dsts;
	struct ir3_register **srcs;
	union {
		struct {
			char inv1, inv2;
			char comp1, comp2;
			int  immed;
			struct ir3_block *target;
			const char *target_label;
			brtype_t brtype;
			unsigned idx;  /* for brac.N */
		} cat0;
		struct {
			type_t src_type, dst_type;
			round_t round;
		} cat1;
		struct {
			enum {
				IR3_COND_LT = 0,
				IR3_COND_LE = 1,
				IR3_COND_GT = 2,
				IR3_COND_GE = 3,
				IR3_COND_EQ = 4,
				IR3_COND_NE = 5,
			} condition;
		} cat2;
		struct {
			unsigned samp, tex;
			unsigned tex_base : 3;
			type_t type;
		} cat5;
		struct {
			type_t type;
			/* TODO remove dst_offset and handle as a ir3_register
			 * which might be IMMED, similar to how src_offset is
			 * handled.
			 */
			int dst_offset;
			int iim_val : 3;      /* for ldgb/stgb, # of components */
			unsigned d : 3;       /* for ldc, component offset */
			bool typed : 1;
			unsigned base : 3;
		} cat6;
		struct {
			unsigned w : 1;       /* write */
			unsigned r : 1;       /* read */
			unsigned l : 1;       /* local */
			unsigned g : 1;       /* global */
		} cat7;
		/* for meta-instructions, just used to hold extra data
		 * before instruction scheduling, etc
		 */
		struct {
			int off;              /* component/offset */
		} split;
		struct {
			/* Per-source index back to the entry in the
			 * ir3_shader_variant::outputs table.
			 */
			unsigned *outidxs;
		} end;
		struct {
			/* used to temporarily hold reference to nir_phi_instr
			 * until we resolve the phi srcs
			 */
			void *nphi;
		} phi;
		struct {
			unsigned samp, tex;
			unsigned input_offset;
			unsigned samp_base : 3;
			unsigned tex_base : 3;
		} prefetch;
		struct {
			/* maps back to entry in ir3_shader_variant::inputs table: */
			int inidx;
			/* for sysvals, identifies the sysval type.  Mostly so we can
			 * identify the special cases where a sysval should not be DCE'd
			 * (currently, just pre-fs texture fetch)
			 */
			gl_system_value sysval;
		} input;
	};

	/* When we get to the RA stage, we need instruction's position/name: */
	uint16_t ip;
	uint16_t name;

	/* used for per-pass extra instruction data.
	 *
	 * TODO we should remove the per-pass data like this and 'use_count'
	 * and do something similar to what RA does w/ ir3_ra_instr_data..
	 * ie. use the ir3_count_instructions pass, and then use instr->ip
	 * to index into a table of pass-private data.
	 */
	void *data;

	/**
	 * Valid if pass calls ir3_find_ssa_uses().. see foreach_ssa_use()
	 */
	struct set *uses;

	int use_count;      /* currently just updated/used by cp */

	/* an instruction can reference at most one address register amongst
	 * it's src/dst registers.  Beyond that, you need to insert mov's.
	 *
	 * NOTE: do not write this directly, use ir3_instr_set_address()
	 */
	struct ir3_register *address;

	/* Tracking for additional dependent instructions.  Used to handle
	 * barriers, WAR hazards for arrays/SSBOs/etc.
	 */
	DECLARE_ARRAY(struct ir3_instruction *, deps);

	/*
	 * From PoV of instruction scheduling, not execution (ie. ignores global/
	 * local distinction):
	 *                            shared  image  atomic  SSBO  everything
	 *   barrier()/            -   R/W     R/W    R/W     R/W       X
	 *     groupMemoryBarrier()
	 *     memoryBarrier()
	 *     (but only images declared coherent?)
	 *   memoryBarrierAtomic() -                  R/W
	 *   memoryBarrierBuffer() -                          R/W
	 *   memoryBarrierImage()  -           R/W
	 *   memoryBarrierShared() -   R/W
	 *
	 * TODO I think for SSBO/image/shared, in cases where we can determine
	 * which variable is accessed, we don't need to care about accesses to
	 * different variables (unless declared coherent??)
	 */
	enum {
		IR3_BARRIER_EVERYTHING = 1 << 0,
		IR3_BARRIER_SHARED_R   = 1 << 1,
		IR3_BARRIER_SHARED_W   = 1 << 2,
		IR3_BARRIER_IMAGE_R    = 1 << 3,
		IR3_BARRIER_IMAGE_W    = 1 << 4,
		IR3_BARRIER_BUFFER_R   = 1 << 5,
		IR3_BARRIER_BUFFER_W   = 1 << 6,
		IR3_BARRIER_ARRAY_R    = 1 << 7,
		IR3_BARRIER_ARRAY_W    = 1 << 8,
		IR3_BARRIER_PRIVATE_R  = 1 << 9,
		IR3_BARRIER_PRIVATE_W  = 1 << 10,
	} barrier_class, barrier_conflict;

	/* Entry in ir3_block's instruction list: */
	struct list_head node;

	uint32_t serialno;

	// TODO only computerator/assembler:
	int line;
};

struct ir3 {
	struct ir3_compiler *compiler;
	gl_shader_stage type;

	DECLARE_ARRAY(struct ir3_instruction *, inputs);

	/* Track bary.f (and ldlv) instructions.. this is needed in
	 * scheduling to ensure that all varying fetches happen before
	 * any potential kill instructions.  The hw gets grumpy if all
	 * threads in a group are killed before the last bary.f gets
	 * a chance to signal end of input (ei).
	 */
	DECLARE_ARRAY(struct ir3_instruction *, baryfs);

	/* Track all indirect instructions (read and write).  To avoid
	 * deadlock scenario where an address register gets scheduled,
	 * but other dependent src instructions cannot be scheduled due
	 * to dependency on a *different* address register value, the
	 * scheduler needs to ensure that all dependencies other than
	 * the instruction other than the address register are scheduled
	 * before the one that writes the address register.  Having a
	 * convenient list of instructions that reference some address
	 * register simplifies this.
	 */
	DECLARE_ARRAY(struct ir3_instruction *, a0_users);

	/* same for a1.x: */
	DECLARE_ARRAY(struct ir3_instruction *, a1_users);

	/* and same for instructions that consume predicate register: */
	DECLARE_ARRAY(struct ir3_instruction *, predicates);

	/* Track texture sample instructions which need texture state
	 * patched in (for astc-srgb workaround):
	 */
	DECLARE_ARRAY(struct ir3_instruction *, astc_srgb);

	/* List of blocks: */
	struct list_head block_list;

	/* List of ir3_array's: */
	struct list_head array_list;

#ifdef DEBUG
	unsigned block_count;
#endif
	unsigned instr_count;
};

struct ir3_array {
	struct list_head node;
	unsigned length;
	unsigned id;

	struct nir_register *r;

	/* To avoid array write's from getting DCE'd, keep track of the
	 * most recent write.  Any array access depends on the most
	 * recent write.  This way, nothing depends on writes after the
	 * last read.  But all the writes that happen before that have
	 * something depending on them
	 */
	struct ir3_register *last_write;

	/* extra stuff used in RA pass: */
	unsigned base;      /* base vreg name */
	unsigned reg;       /* base physical reg */
	uint16_t start_ip, end_ip;

	/* Indicates if half-precision */
	bool half;

	bool unused;
};

struct ir3_array * ir3_lookup_array(struct ir3 *ir, unsigned id);

struct ir3_block {
	struct list_head node;
	struct ir3 *shader;

	const struct nir_block *nblock;

	struct list_head instr_list;  /* list of ir3_instruction */

	/* each block has either one or two successors.. in case of
	 * two successors, 'condition' decides which one to follow.
	 * A block preceding an if/else has two successors.
	 */
	struct ir3_instruction *condition;
	struct ir3_block *successors[2];

	DECLARE_ARRAY(struct ir3_block *, predecessors);

	uint16_t start_ip, end_ip;

	/* Track instructions which do not write a register but other-
	 * wise must not be discarded (such as kill, stg, etc)
	 */
	DECLARE_ARRAY(struct ir3_instruction *, keeps);

	/* used for per-pass extra block data.  Mainly used right
	 * now in RA step to track livein/liveout.
	 */
	void *data;

	uint32_t index;

	struct ir3_block *imm_dom;
	DECLARE_ARRAY(struct ir3_block *, dom_children);

	uint32_t dom_pre_index;
	uint32_t dom_post_index;

#ifdef DEBUG
	uint32_t serialno;
#endif
};

static inline uint32_t
block_id(struct ir3_block *block)
{
#ifdef DEBUG
	return block->serialno;
#else
	return (uint32_t)(unsigned long)block;
#endif
}

static inline struct ir3_block *
ir3_start_block(struct ir3 *ir)
{
	return list_first_entry(&ir->block_list, struct ir3_block, node);
}

void ir3_block_add_predecessor(struct ir3_block *block, struct ir3_block *pred);
void ir3_block_remove_predecessor(struct ir3_block *block, struct ir3_block *pred);
unsigned ir3_block_get_pred_index(struct ir3_block *block, struct ir3_block *pred);

void ir3_calc_dominance(struct ir3 *ir);
bool ir3_block_dominates(struct ir3_block *a, struct ir3_block *b);

struct ir3_shader_variant;

struct ir3 * ir3_create(struct ir3_compiler *compiler, struct ir3_shader_variant *v);
void ir3_destroy(struct ir3 *shader);

void ir3_collect_info(struct ir3_shader_variant *v);
void * ir3_alloc(struct ir3 *shader, int sz);

unsigned ir3_get_reg_dependent_max_waves(const struct ir3_compiler *compiler,
										 unsigned reg_count, bool double_threadsize);

unsigned ir3_get_reg_independent_max_waves(struct ir3_shader_variant *v,
										   bool double_threadsize);

bool ir3_should_double_threadsize(struct ir3_shader_variant *v,
								  unsigned regs_count);

struct ir3_block * ir3_block_create(struct ir3 *shader);

struct ir3_instruction * ir3_instr_create(struct ir3_block *block,
		opc_t opc, int ndst, int nsrc);
struct ir3_instruction * ir3_instr_clone(struct ir3_instruction *instr);
void ir3_instr_add_dep(struct ir3_instruction *instr, struct ir3_instruction *dep);
const char *ir3_instr_name(struct ir3_instruction *instr);

struct ir3_register * ir3_src_create(struct ir3_instruction *instr,
		int num, int flags);
struct ir3_register * ir3_dst_create(struct ir3_instruction *instr,
		int num, int flags);
struct ir3_register * ir3_reg_clone(struct ir3 *shader,
		struct ir3_register *reg);

static inline void ir3_reg_tie(struct ir3_register *dst, struct ir3_register *src)
{
	assert(!dst->tied && !src->tied);
	dst->tied = src;
	src->tied = dst;
}

void ir3_reg_set_last_array(struct ir3_instruction *instr,
							struct ir3_register *reg,
							struct ir3_register *last_write);

void ir3_instr_set_address(struct ir3_instruction *instr,
		struct ir3_instruction *addr);

static inline bool ir3_instr_check_mark(struct ir3_instruction *instr)
{
	if (instr->flags & IR3_INSTR_MARK)
		return true;  /* already visited */
	instr->flags |= IR3_INSTR_MARK;
	return false;
}

void ir3_block_clear_mark(struct ir3_block *block);
void ir3_clear_mark(struct ir3 *shader);

unsigned ir3_count_instructions(struct ir3 *ir);
unsigned ir3_count_instructions_ra(struct ir3 *ir);

/**
 * Move 'instr' to just before 'after'
 */
static inline void
ir3_instr_move_before(struct ir3_instruction *instr,
		struct ir3_instruction *after)
{
	list_delinit(&instr->node);
	list_addtail(&instr->node, &after->node);
}

/**
 * Move 'instr' to just after 'before':
 */
static inline void
ir3_instr_move_after(struct ir3_instruction *instr,
		struct ir3_instruction *before)
{
	list_delinit(&instr->node);
	list_add(&instr->node, &before->node);
}

void ir3_find_ssa_uses(struct ir3 *ir, void *mem_ctx, bool falsedeps);

void ir3_set_dst_type(struct ir3_instruction *instr, bool half);
void ir3_fixup_src_type(struct ir3_instruction *instr);

bool ir3_valid_flags(struct ir3_instruction *instr, unsigned n, unsigned flags);

#include "util/set.h"
#define foreach_ssa_use(__use, __instr) \
	for (struct ir3_instruction *__use = (void *)~0; \
	     __use && (__instr)->uses; __use = NULL) \
		set_foreach ((__instr)->uses, __entry) \
			if ((__use = (void *)__entry->key))

static inline uint32_t reg_num(const struct ir3_register *reg)
{
	return reg->num >> 2;
}

static inline uint32_t reg_comp(const struct ir3_register *reg)
{
	return reg->num & 0x3;
}

static inline bool is_flow(struct ir3_instruction *instr)
{
	return (opc_cat(instr->opc) == 0);
}

static inline bool is_kill_or_demote(struct ir3_instruction *instr)
{
	return instr->opc == OPC_KILL || instr->opc == OPC_DEMOTE;
}

static inline bool is_nop(struct ir3_instruction *instr)
{
	return instr->opc == OPC_NOP;
}

static inline bool is_same_type_reg(struct ir3_register *reg1,
		struct ir3_register *reg2)
{
	unsigned type_reg1 = (reg1->flags & (IR3_REG_SHARED | IR3_REG_HALF));
	unsigned type_reg2 = (reg2->flags & (IR3_REG_SHARED | IR3_REG_HALF));

	if (type_reg1 ^ type_reg2)
		return false;
	else
		return true;
}

/* Is it a non-transformative (ie. not type changing) mov?  This can
 * also include absneg.s/absneg.f, which for the most part can be
 * treated as a mov (single src argument).
 */
static inline bool is_same_type_mov(struct ir3_instruction *instr)
{
	struct ir3_register *dst;

	switch (instr->opc) {
	case OPC_MOV:
		if (instr->cat1.src_type != instr->cat1.dst_type)
			return false;
		/* If the type of dest reg and src reg are different,
		 * it shouldn't be considered as same type mov
		 */
		if (!is_same_type_reg(instr->dsts[0], instr->srcs[0]))
			return false;
		break;
	case OPC_ABSNEG_F:
	case OPC_ABSNEG_S:
		if (instr->flags & IR3_INSTR_SAT)
			return false;
		/* If the type of dest reg and src reg are different,
		 * it shouldn't be considered as same type mov
		 */
		if (!is_same_type_reg(instr->dsts[0], instr->srcs[0]))
			return false;
		break;
	default:
		return false;
	}

	dst = instr->dsts[0];

	/* mov's that write to a0 or p0.x are special: */
	if (dst->num == regid(REG_P0, 0))
		return false;
	if (reg_num(dst) == REG_A0)
		return false;

	if (dst->flags & (IR3_REG_RELATIV | IR3_REG_ARRAY))
		return false;

	return true;
}

/* A move from const, which changes size but not type, can also be
 * folded into dest instruction in some cases.
 */
static inline bool is_const_mov(struct ir3_instruction *instr)
{
	if (instr->opc != OPC_MOV)
		return false;

	if (!(instr->srcs[0]->flags & IR3_REG_CONST))
		return false;

	type_t src_type = instr->cat1.src_type;
	type_t dst_type = instr->cat1.dst_type;

	return (type_float(src_type) && type_float(dst_type)) ||
		(type_uint(src_type) && type_uint(dst_type)) ||
		(type_sint(src_type) && type_sint(dst_type));
}

static inline bool is_alu(struct ir3_instruction *instr)
{
	return (1 <= opc_cat(instr->opc)) && (opc_cat(instr->opc) <= 3);
}

static inline bool is_sfu(struct ir3_instruction *instr)
{
	return (opc_cat(instr->opc) == 4);
}

static inline bool is_tex(struct ir3_instruction *instr)
{
	return (opc_cat(instr->opc) == 5);
}

static inline bool is_tex_or_prefetch(struct ir3_instruction *instr)
{
	return is_tex(instr) || (instr->opc == OPC_META_TEX_PREFETCH);
}

static inline bool is_mem(struct ir3_instruction *instr)
{
	return (opc_cat(instr->opc) == 6);
}

static inline bool is_barrier(struct ir3_instruction *instr)
{
	return (opc_cat(instr->opc) == 7);
}

static inline bool
is_half(struct ir3_instruction *instr)
{
	return !!(instr->dsts[0]->flags & IR3_REG_HALF);
}

static inline bool
is_shared(struct ir3_instruction *instr)
{
	return !!(instr->dsts[0]->flags & IR3_REG_SHARED);
}

static inline bool
is_store(struct ir3_instruction *instr)
{
	/* these instructions, the "destination" register is
	 * actually a source, the address to store to.
	 */
	switch (instr->opc) {
	case OPC_STG:
	case OPC_STGB:
	case OPC_STIB:
	case OPC_STP:
	case OPC_STL:
	case OPC_STLW:
	case OPC_L2G:
	case OPC_G2L:
		return true;
	default:
		return false;
	}
}

static inline bool is_load(struct ir3_instruction *instr)
{
	switch (instr->opc) {
	case OPC_LDG:
	case OPC_LDGB:
	case OPC_LDIB:
	case OPC_LDL:
	case OPC_LDP:
	case OPC_L2G:
	case OPC_LDLW:
	case OPC_LDC:
	case OPC_LDLV:
		/* probably some others too.. */
		return true;
	default:
		return false;
	}
}

static inline bool is_input(struct ir3_instruction *instr)
{
	/* in some cases, ldlv is used to fetch varying without
	 * interpolation.. fortunately inloc is the first src
	 * register in either case
	 */
	switch (instr->opc) {
	case OPC_LDLV:
	case OPC_BARY_F:
		return true;
	default:
		return false;
	}
}

static inline bool is_bool(struct ir3_instruction *instr)
{
	switch (instr->opc) {
	case OPC_CMPS_F:
	case OPC_CMPS_S:
	case OPC_CMPS_U:
		return true;
	default:
		return false;
	}
}

static inline opc_t
cat3_half_opc(opc_t opc)
{
	switch (opc) {
	case OPC_MAD_F32: return OPC_MAD_F16;
	case OPC_SEL_B32: return OPC_SEL_B16;
	case OPC_SEL_S32: return OPC_SEL_S16;
	case OPC_SEL_F32: return OPC_SEL_F16;
	case OPC_SAD_S32: return OPC_SAD_S16;
	default:          return opc;
	}
}

static inline opc_t
cat3_full_opc(opc_t opc)
{
	switch (opc) {
	case OPC_MAD_F16: return OPC_MAD_F32;
	case OPC_SEL_B16: return OPC_SEL_B32;
	case OPC_SEL_S16: return OPC_SEL_S32;
	case OPC_SEL_F16: return OPC_SEL_F32;
	case OPC_SAD_S16: return OPC_SAD_S32;
	default:          return opc;
	}
}

static inline opc_t
cat4_half_opc(opc_t opc)
{
	switch (opc) {
	case OPC_RSQ:  return OPC_HRSQ;
	case OPC_LOG2: return OPC_HLOG2;
	case OPC_EXP2: return OPC_HEXP2;
	default:       return opc;
	}
}

static inline opc_t
cat4_full_opc(opc_t opc)
{
	switch (opc) {
	case OPC_HRSQ:  return OPC_RSQ;
	case OPC_HLOG2: return OPC_LOG2;
	case OPC_HEXP2: return OPC_EXP2;
	default:        return opc;
	}
}

static inline bool is_meta(struct ir3_instruction *instr)
{
	return (opc_cat(instr->opc) == -1);
}

static inline unsigned reg_elems(const struct ir3_register *reg)
{
	if (reg->flags & IR3_REG_ARRAY)
		return reg->size;
	else
		return util_last_bit(reg->wrmask);
}

static inline unsigned
reg_elem_size(const struct ir3_register *reg)
{
	return (reg->flags & IR3_REG_HALF) ? 1 : 2;
}

static inline unsigned
reg_size(const struct ir3_register *reg)
{
	return reg_elems(reg) * reg_elem_size(reg);
}

static inline unsigned dest_regs(struct ir3_instruction *instr)
{
	if ((instr->dsts_count == 0) || is_store(instr) || is_flow(instr))
		return 0;

	return util_last_bit(instr->dsts[0]->wrmask);
}

static inline bool
writes_gpr(struct ir3_instruction *instr)
{
	if (dest_regs(instr) == 0)
		return false;
	/* is dest a normal temp register: */
	struct ir3_register *reg = instr->dsts[0];
	debug_assert(!(reg->flags & (IR3_REG_CONST | IR3_REG_IMMED)));
	if ((reg_num(reg) == REG_A0) ||
			(reg->num == regid(REG_P0, 0)))
		return false;
	return true;
}

static inline bool writes_addr0(struct ir3_instruction *instr)
{
	if (instr->dsts_count > 0) {
		struct ir3_register *dst = instr->dsts[0];
		return dst->num == regid(REG_A0, 0);
	}
	return false;
}

static inline bool writes_addr1(struct ir3_instruction *instr)
{
	if (instr->dsts_count > 0) {
		struct ir3_register *dst = instr->dsts[0];
		return dst->num == regid(REG_A0, 1);
	}
	return false;
}

static inline bool writes_pred(struct ir3_instruction *instr)
{
	if (instr->dsts_count > 0) {
		struct ir3_register *dst = instr->dsts[0];
		return reg_num(dst) == REG_P0;
	}
	return false;
}

/* Is it something other than a normal register. Shared regs, p0, and a0/a1
 * are considered special here. Special registers are always accessed with one
 * size and never alias normal registers, even though a naive calculation
 * would sometimes make it seem like e.g. r30.z aliases a0.x.
 */
static inline bool is_reg_special(const struct ir3_register *reg)
{
	return (reg->flags & IR3_REG_SHARED) ||
		(reg_num(reg) == REG_A0) || (reg_num(reg) == REG_P0);
}

/* returns defining instruction for reg */
/* TODO better name */
static inline struct ir3_instruction *ssa(struct ir3_register *reg)
{
	if ((reg->flags & (IR3_REG_SSA | IR3_REG_ARRAY)) && reg->def)
		return reg->def->instr;
	return NULL;
}

static inline bool conflicts(struct ir3_register *a,
		struct ir3_register *b)
{
	return (a && b) && (a->def != b->def);
}

static inline bool reg_gpr(struct ir3_register *r)
{
	if (r->flags & (IR3_REG_CONST | IR3_REG_IMMED))
		return false;
	if ((reg_num(r) == REG_A0) || (reg_num(r) == REG_P0))
		return false;
	return true;
}

static inline type_t half_type(type_t type)
{
	switch (type) {
	case TYPE_F32: return TYPE_F16;
	case TYPE_U32: return TYPE_U16;
	case TYPE_S32: return TYPE_S16;
	case TYPE_F16:
	case TYPE_U16:
	case TYPE_S16:
		return type;
	default:
		assert(0);
		return ~0;
	}
}

static inline type_t full_type(type_t type)
{
	switch (type) {
	case TYPE_F16: return TYPE_F32;
	case TYPE_U16: return TYPE_U32;
	case TYPE_S16: return TYPE_S32;
	case TYPE_F32:
	case TYPE_U32:
	case TYPE_S32:
		return type;
	default:
		assert(0);
		return ~0;
	}
}

/* some cat2 instructions (ie. those which are not float) can embed an
 * immediate:
 */
static inline bool ir3_cat2_int(opc_t opc)
{
	switch (opc) {
	case OPC_ADD_U:
	case OPC_ADD_S:
	case OPC_SUB_U:
	case OPC_SUB_S:
	case OPC_CMPS_U:
	case OPC_CMPS_S:
	case OPC_MIN_U:
	case OPC_MIN_S:
	case OPC_MAX_U:
	case OPC_MAX_S:
	case OPC_CMPV_U:
	case OPC_CMPV_S:
	case OPC_MUL_U24:
	case OPC_MUL_S24:
	case OPC_MULL_U:
	case OPC_CLZ_S:
	case OPC_ABSNEG_S:
	case OPC_AND_B:
	case OPC_OR_B:
	case OPC_NOT_B:
	case OPC_XOR_B:
	case OPC_BFREV_B:
	case OPC_CLZ_B:
	case OPC_SHL_B:
	case OPC_SHR_B:
	case OPC_ASHR_B:
	case OPC_MGEN_B:
	case OPC_GETBIT_B:
	case OPC_CBITS_B:
	case OPC_BARY_F:
		return true;

	default:
		return false;
	}
}

/* map cat2 instruction to valid abs/neg flags: */
static inline unsigned ir3_cat2_absneg(opc_t opc)
{
	switch (opc) {
	case OPC_ADD_F:
	case OPC_MIN_F:
	case OPC_MAX_F:
	case OPC_MUL_F:
	case OPC_SIGN_F:
	case OPC_CMPS_F:
	case OPC_ABSNEG_F:
	case OPC_CMPV_F:
	case OPC_FLOOR_F:
	case OPC_CEIL_F:
	case OPC_RNDNE_F:
	case OPC_RNDAZ_F:
	case OPC_TRUNC_F:
	case OPC_BARY_F:
		return IR3_REG_FABS | IR3_REG_FNEG;

	case OPC_ADD_U:
	case OPC_ADD_S:
	case OPC_SUB_U:
	case OPC_SUB_S:
	case OPC_CMPS_U:
	case OPC_CMPS_S:
	case OPC_MIN_U:
	case OPC_MIN_S:
	case OPC_MAX_U:
	case OPC_MAX_S:
	case OPC_CMPV_U:
	case OPC_CMPV_S:
	case OPC_MUL_U24:
	case OPC_MUL_S24:
	case OPC_MULL_U:
	case OPC_CLZ_S:
		return 0;

	case OPC_ABSNEG_S:
		return IR3_REG_SABS | IR3_REG_SNEG;

	case OPC_AND_B:
	case OPC_OR_B:
	case OPC_NOT_B:
	case OPC_XOR_B:
	case OPC_BFREV_B:
	case OPC_CLZ_B:
	case OPC_SHL_B:
	case OPC_SHR_B:
	case OPC_ASHR_B:
	case OPC_MGEN_B:
	case OPC_GETBIT_B:
	case OPC_CBITS_B:
		return IR3_REG_BNOT;

	default:
		return 0;
	}
}

/* map cat3 instructions to valid abs/neg flags: */
static inline unsigned ir3_cat3_absneg(opc_t opc)
{
	switch (opc) {
	case OPC_MAD_F16:
	case OPC_MAD_F32:
	case OPC_SEL_F16:
	case OPC_SEL_F32:
		return IR3_REG_FNEG;

	case OPC_MAD_U16:
	case OPC_MADSH_U16:
	case OPC_MAD_S16:
	case OPC_MADSH_M16:
	case OPC_MAD_U24:
	case OPC_MAD_S24:
	case OPC_SEL_S16:
	case OPC_SEL_S32:
	case OPC_SAD_S16:
	case OPC_SAD_S32:
		/* neg *may* work on 3rd src.. */

	case OPC_SEL_B16:
	case OPC_SEL_B32:

	default:
		return 0;
	}
}

/* Return the type (float, int, or uint) the op uses when converting from the
 * internal result of the op (which is assumed to be the same size as the
 * sources) to the destination when they are not the same size. If F32 it does
 * a floating-point conversion, if U32 it does a truncation/zero-extension, if
 * S32 it does a truncation/sign-extension. "can_fold" will be false if it
 * doesn't do anything sensible or is unknown.
 */
static inline type_t
ir3_output_conv_type(struct ir3_instruction *instr, bool *can_fold)
{
	*can_fold = true;
	switch (instr->opc) {
	case OPC_ADD_F:
	case OPC_MUL_F:
	case OPC_BARY_F:
	case OPC_MAD_F32:
	case OPC_MAD_F16:
		return TYPE_F32;

	case OPC_ADD_U:
	case OPC_SUB_U:
	case OPC_MIN_U:
	case OPC_MAX_U:
	case OPC_AND_B:
	case OPC_OR_B:
	case OPC_NOT_B:
	case OPC_XOR_B:
	case OPC_MUL_U24:
	case OPC_MULL_U:
	case OPC_SHL_B:
	case OPC_SHR_B:
	case OPC_ASHR_B:
	case OPC_MAD_U24:
	/* Comparison ops zero-extend/truncate their results, so consider them as
	 * unsigned here.
	 */
	case OPC_CMPS_F:
	case OPC_CMPV_F:
	case OPC_CMPS_U:
	case OPC_CMPS_S:
		return TYPE_U32;

	case OPC_ADD_S:
	case OPC_SUB_S:
	case OPC_MIN_S:
	case OPC_MAX_S:
	case OPC_ABSNEG_S:
	case OPC_MUL_S24:
	case OPC_MAD_S24:
		return TYPE_S32;

	/* We assume that any move->move folding that could be done was done by
	 * NIR.
	 */
	case OPC_MOV:
	default:
		*can_fold = false;
		return TYPE_U32;
	}
}

/* Return the src and dst types for the conversion which is already folded
 * into the op. We can assume that instr has folded in a conversion from
 * ir3_output_conv_src_type() to ir3_output_conv_dst_type(). Only makes sense
 * to call if ir3_output_conv_type() returns can_fold = true.
 */
static inline type_t
ir3_output_conv_src_type(struct ir3_instruction *instr, type_t base_type)
{
	switch (instr->opc) {
	case OPC_CMPS_F:
	case OPC_CMPV_F:
	case OPC_CMPS_U:
	case OPC_CMPS_S:
		/* Comparisons only return 0/1 and the size of the comparison sources
		 * is irrelevant, never consider them as having an output conversion
		 * by returning a type with the dest size here:
		 */
		return (instr->dsts[0]->flags & IR3_REG_HALF) ? half_type(base_type) :
			full_type(base_type);

	case OPC_BARY_F:
		/* bary.f doesn't have an explicit source, but we can assume here that
		 * the varying data it reads is in fp32.
		 *
		 * This may be fp16 on older gen's depending on some register
		 * settings, but it's probably not worth plumbing that through for a
		 * small improvement that NIR would hopefully handle for us anyway.
		 */
		return TYPE_F32;

	default:
		return (instr->dsts[1]->flags & IR3_REG_HALF) ? half_type(base_type) :
			full_type(base_type);
	}
}

static inline type_t
ir3_output_conv_dst_type(struct ir3_instruction *instr, type_t base_type)
{
	return (instr->dsts[0]->flags & IR3_REG_HALF) ? half_type(base_type) :
		full_type(base_type);
}

/* Some instructions have signed/unsigned variants which are identical except
 * for whether the folded conversion sign-extends or zero-extends, and we can
 * fold in a mismatching move by rewriting the opcode. Return the opcode to
 * switch signedness, and whether one exists.
 */
static inline opc_t
ir3_try_swap_signedness(opc_t opc, bool *can_swap)
{
	switch (opc) {
#define PAIR(u, s)		\
	case OPC_##u:		\
		return OPC_##s;	\
	case OPC_##s:		\
		return OPC_##u;
	PAIR(ADD_U, ADD_S)
	PAIR(SUB_U, SUB_S)
	/* Note: these are only identical when the sources are half, but that's
	 * the only case we call this function for anyway.
	 */
	PAIR(MUL_U24, MUL_S24)

	default:
		*can_swap = false;
		return opc;
	}
}

#define MASK(n) ((1 << (n)) - 1)

/* iterator for an instructions's sources (reg), also returns src #: */
#define foreach_src_n(__srcreg, __n, __instr) \
	if ((__instr)->srcs_count) \
		for (struct ir3_register *__srcreg = (void *)~0; __srcreg; __srcreg = NULL) \
			for (unsigned __cnt = (__instr)->srcs_count, __n = 0; __n < __cnt; __n++) \
				if ((__srcreg = (__instr)->srcs[__n]))

/* iterator for an instructions's sources (reg): */
#define foreach_src(__srcreg, __instr) \
	foreach_src_n(__srcreg, __i, __instr)

static inline unsigned __ssa_src_cnt(struct ir3_instruction *instr)
{
	return instr->srcs_count + instr->deps_count;
}

static inline bool __is_false_dep(struct ir3_instruction *instr, unsigned n)
{
	if (n >= instr->srcs_count)
		return true;
	return false;
}

static inline struct ir3_instruction **
__ssa_srcp_n(struct ir3_instruction *instr, unsigned n)
{
	if (__is_false_dep(instr, n))
		return &instr->deps[n - instr->srcs_count];
	if (ssa(instr->srcs[n]))
		return &instr->srcs[n]->def->instr;
	return NULL;
}

#define foreach_ssa_srcp_n(__srcp, __n, __instr) \
	for (struct ir3_instruction **__srcp = (void *)~0; __srcp; __srcp = NULL) \
		for (unsigned __cnt = __ssa_src_cnt(__instr), __n = 0; __n < __cnt; __n++) \
			if ((__srcp = __ssa_srcp_n(__instr, __n)))

#define foreach_ssa_srcp(__srcp, __instr) \
	foreach_ssa_srcp_n(__srcp, __i, __instr)

/* iterator for an instruction's SSA sources (instr), also returns src #: */
#define foreach_ssa_src_n(__srcinst, __n, __instr) \
	for (struct ir3_instruction *__srcinst = (void *)~0; __srcinst; __srcinst = NULL) \
		foreach_ssa_srcp_n(__srcp, __n, __instr) \
			if ((__srcinst = *__srcp))

/* iterator for an instruction's SSA sources (instr): */
#define foreach_ssa_src(__srcinst, __instr) \
	foreach_ssa_src_n(__srcinst, __i, __instr)

/* iterators for shader inputs: */
#define foreach_input_n(__ininstr, __cnt, __ir) \
	for (struct ir3_instruction *__ininstr = (void *)~0; __ininstr; __ininstr = NULL) \
		for (unsigned __cnt = 0; __cnt < (__ir)->inputs_count; __cnt++) \
			if ((__ininstr = (__ir)->inputs[__cnt]))
#define foreach_input(__ininstr, __ir) \
	foreach_input_n(__ininstr, __i, __ir)

/* iterators for instructions: */
#define foreach_instr(__instr, __list) \
	list_for_each_entry(struct ir3_instruction, __instr, __list, node)
#define foreach_instr_rev(__instr, __list) \
	list_for_each_entry_rev(struct ir3_instruction, __instr, __list, node)
#define foreach_instr_safe(__instr, __list) \
	list_for_each_entry_safe(struct ir3_instruction, __instr, __list, node)

/* iterators for blocks: */
#define foreach_block(__block, __list) \
	list_for_each_entry(struct ir3_block, __block, __list, node)
#define foreach_block_safe(__block, __list) \
	list_for_each_entry_safe(struct ir3_block, __block, __list, node)
#define foreach_block_rev(__block, __list) \
	list_for_each_entry_rev(struct ir3_block, __block, __list, node)

/* iterators for arrays: */
#define foreach_array(__array, __list) \
	list_for_each_entry(struct ir3_array, __array, __list, node)
#define foreach_array_safe(__array, __list) \
	list_for_each_entry_safe(struct ir3_array, __array, __list, node)

#define IR3_PASS(ir, pass, ...) ({ \
		bool progress = pass(ir, ##__VA_ARGS__); \
		if (progress) { \
			ir3_debug_print(ir, "AFTER: " #pass); \
			ir3_validate(ir); \
		} \
		progress; \
	})

/* validate: */
void ir3_validate(struct ir3 *ir);

/* dump: */
void ir3_print(struct ir3 *ir);
void ir3_print_instr(struct ir3_instruction *instr);

/* delay calculation: */
int ir3_delayslots(struct ir3_instruction *assigner,
		struct ir3_instruction *consumer, unsigned n, bool soft);
unsigned ir3_delay_calc_prera(struct ir3_block *block, struct ir3_instruction *instr);
unsigned ir3_delay_calc_postra(struct ir3_block *block, struct ir3_instruction *instr,
		bool soft, bool mergedregs);
unsigned ir3_delay_calc_exact(struct ir3_block *block, struct ir3_instruction *instr,
		bool mergedregs);
void ir3_remove_nops(struct ir3 *ir);

/* dead code elimination: */
struct ir3_shader_variant;
bool ir3_dce(struct ir3 *ir, struct ir3_shader_variant *so);

/* fp16 conversion folding */
bool ir3_cf(struct ir3 *ir);

/* copy-propagate: */
bool ir3_cp(struct ir3 *ir, struct ir3_shader_variant *so);
bool ir3_cp_postsched(struct ir3 *ir);

/* common subexpression elimination: */
bool ir3_cse(struct ir3 *ir);

/* Make arrays SSA */
bool ir3_array_to_ssa(struct ir3 *ir);

/* scheduling: */
bool ir3_sched_add_deps(struct ir3 *ir);
int ir3_sched(struct ir3 *ir);

struct ir3_context;
bool ir3_postsched(struct ir3 *ir, struct ir3_shader_variant *v);

/* register assignment: */
int ir3_ra(struct ir3_shader_variant *v);

/* legalize: */
bool ir3_legalize(struct ir3 *ir, struct ir3_shader_variant *so, int *max_bary);

static inline bool
ir3_has_latency_to_hide(struct ir3 *ir)
{
	/* VS/GS/TCS/TESS  co-exist with frag shader invocations, but we don't
	 * know the nature of the fragment shader.  Just assume it will have
	 * latency to hide:
	 */
	if (ir->type != MESA_SHADER_FRAGMENT)
		return true;

	foreach_block (block, &ir->block_list) {
		foreach_instr (instr, &block->instr_list) {
			if (is_tex_or_prefetch(instr))
				return true;

			if (is_load(instr)) {
				switch (instr->opc) {
				case OPC_LDLV:
				case OPC_LDL:
				case OPC_LDLW:
					break;
				default:
					return true;
				}
			}
		}
	}

	return false;
}

/* ************************************************************************* */
/* instruction helpers */

/* creates SSA src of correct type (ie. half vs full precision) */
static inline struct ir3_register * __ssa_src(struct ir3_instruction *instr,
		struct ir3_instruction *src, unsigned flags)
{
	struct ir3_register *reg;
	if (src->dsts[0]->flags & IR3_REG_HALF)
		flags |= IR3_REG_HALF;
	reg = ir3_src_create(instr, INVALID_REG, IR3_REG_SSA | flags);
	reg->def = src->dsts[0];
	reg->wrmask = src->dsts[0]->wrmask;
	return reg;
}

static inline struct ir3_register * __ssa_dst(struct ir3_instruction *instr)
{
	struct ir3_register *reg = ir3_dst_create(instr, INVALID_REG, IR3_REG_SSA);
	reg->instr = instr;
	return reg;
}

static inline struct ir3_instruction *
create_immed_typed(struct ir3_block *block, uint32_t val, type_t type)
{
	struct ir3_instruction *mov;
	unsigned flags = (type_size(type) < 32) ? IR3_REG_HALF : 0;

	mov = ir3_instr_create(block, OPC_MOV, 1, 1);
	mov->cat1.src_type = type;
	mov->cat1.dst_type = type;
	__ssa_dst(mov)->flags |= flags;
	ir3_src_create(mov, 0, IR3_REG_IMMED | flags)->uim_val = val;

	return mov;
}

static inline struct ir3_instruction *
create_immed(struct ir3_block *block, uint32_t val)
{
	return create_immed_typed(block, val, TYPE_U32);
}

static inline struct ir3_instruction *
create_uniform_typed(struct ir3_block *block, unsigned n, type_t type)
{
	struct ir3_instruction *mov;
	unsigned flags = (type_size(type) < 32) ? IR3_REG_HALF : 0;

	mov = ir3_instr_create(block, OPC_MOV, 1, 1);
	mov->cat1.src_type = type;
	mov->cat1.dst_type = type;
	__ssa_dst(mov)->flags |= flags;
	ir3_src_create(mov, n, IR3_REG_CONST | flags);

	return mov;
}

static inline struct ir3_instruction *
create_uniform(struct ir3_block *block, unsigned n)
{
	return create_uniform_typed(block, n, TYPE_F32);
}

static inline struct ir3_instruction *
create_uniform_indirect(struct ir3_block *block, int n, type_t type,
		struct ir3_instruction *address)
{
	struct ir3_instruction *mov;

	mov = ir3_instr_create(block, OPC_MOV, 1, 1);
	mov->cat1.src_type = type;
	mov->cat1.dst_type = type;
	__ssa_dst(mov);
	ir3_src_create(mov, 0, IR3_REG_CONST | IR3_REG_RELATIV)->array.offset = n;

	ir3_instr_set_address(mov, address);

	return mov;
}

static inline struct ir3_instruction *
ir3_MOV(struct ir3_block *block, struct ir3_instruction *src, type_t type)
{
	struct ir3_instruction *instr = ir3_instr_create(block, OPC_MOV, 1, 1);
	unsigned flags = (type_size(type) < 32) ? IR3_REG_HALF : 0;

	__ssa_dst(instr)->flags |= flags;
	if (src->dsts[0]->flags & IR3_REG_ARRAY) {
		struct ir3_register *src_reg = __ssa_src(instr, src, IR3_REG_ARRAY);
		src_reg->array = src->dsts[0]->array;
	} else {
		__ssa_src(instr, src, src->dsts[0]->flags & IR3_REG_SHARED);
	}
	debug_assert(!(src->dsts[0]->flags & IR3_REG_RELATIV));
	instr->cat1.src_type = type;
	instr->cat1.dst_type = type;
	return instr;
}

static inline struct ir3_instruction *
ir3_COV(struct ir3_block *block, struct ir3_instruction *src,
		type_t src_type, type_t dst_type)
{
	struct ir3_instruction *instr = ir3_instr_create(block, OPC_MOV, 1, 1);
	unsigned dst_flags = (type_size(dst_type) < 32) ? IR3_REG_HALF : 0;
	unsigned src_flags = (type_size(src_type) < 32) ? IR3_REG_HALF : 0;

	debug_assert((src->dsts[0]->flags & IR3_REG_HALF) == src_flags);

	__ssa_dst(instr)->flags |= dst_flags;
	__ssa_src(instr, src, 0);
	instr->cat1.src_type = src_type;
	instr->cat1.dst_type = dst_type;
	debug_assert(!(src->dsts[0]->flags & IR3_REG_ARRAY));
	return instr;
}

static inline struct ir3_instruction *
ir3_MOVMSK(struct ir3_block *block, unsigned components)
{
	struct ir3_instruction *instr = ir3_instr_create(block, OPC_MOVMSK, 1, 0);

	struct ir3_register *dst = __ssa_dst(instr);
	dst->flags |= IR3_REG_SHARED;
	dst->wrmask = (1 << components) - 1;
	return instr;
}

static inline struct ir3_instruction *
ir3_NOP(struct ir3_block *block)
{
	return ir3_instr_create(block, OPC_NOP, 0, 0);
}

#define IR3_INSTR_0 0

#define __INSTR0(flag, name, opc)                                        \
static inline struct ir3_instruction *                                   \
ir3_##name(struct ir3_block *block)                                      \
{                                                                        \
	struct ir3_instruction *instr =                                      \
		ir3_instr_create(block, opc, 1, 0);                              \
	instr->flags |= flag;                                                \
	return instr;                                                        \
}
#define INSTR0F(f, name)    __INSTR0(IR3_INSTR_##f, name##_##f, OPC_##name)
#define INSTR0(name)        __INSTR0(0, name, OPC_##name)

#define __INSTR1(flag, name, opc)                                        \
static inline struct ir3_instruction *                                   \
ir3_##name(struct ir3_block *block,                                      \
		struct ir3_instruction *a, unsigned aflags)                      \
{                                                                        \
	struct ir3_instruction *instr =                                      \
		ir3_instr_create(block, opc, 1, 1);                              \
	__ssa_dst(instr);                                                    \
	__ssa_src(instr, a, aflags);                                         \
	instr->flags |= flag;                                                \
	return instr;                                                        \
}
#define INSTR1F(f, name)    __INSTR1(IR3_INSTR_##f, name##_##f, OPC_##name)
#define INSTR1(name)        __INSTR1(0, name, OPC_##name)

#define __INSTR2(flag, name, opc)                                        \
static inline struct ir3_instruction *                                   \
ir3_##name(struct ir3_block *block,                                      \
		struct ir3_instruction *a, unsigned aflags,                      \
		struct ir3_instruction *b, unsigned bflags)                      \
{                                                                        \
	struct ir3_instruction *instr =                                      \
		ir3_instr_create(block, opc, 1, 2);                              \
	__ssa_dst(instr);                                                    \
	__ssa_src(instr, a, aflags);                                         \
	__ssa_src(instr, b, bflags);                                         \
	instr->flags |= flag;                                                \
	return instr;                                                        \
}
#define INSTR2F(f, name)    __INSTR2(IR3_INSTR_##f, name##_##f, OPC_##name)
#define INSTR2(name)        __INSTR2(0, name, OPC_##name)

#define __INSTR3(flag, name, opc)                                        \
static inline struct ir3_instruction *                                   \
ir3_##name(struct ir3_block *block,                                      \
		struct ir3_instruction *a, unsigned aflags,                      \
		struct ir3_instruction *b, unsigned bflags,                      \
		struct ir3_instruction *c, unsigned cflags)                      \
{                                                                        \
	struct ir3_instruction *instr =                                      \
		ir3_instr_create(block, opc, 1, 3);                              \
	__ssa_dst(instr);                                                    \
	__ssa_src(instr, a, aflags);                                         \
	__ssa_src(instr, b, bflags);                                         \
	__ssa_src(instr, c, cflags);                                         \
	instr->flags |= flag;                                                \
	return instr;                                                        \
}
#define INSTR3F(f, name)    __INSTR3(IR3_INSTR_##f, name##_##f, OPC_##name)
#define INSTR3(name)        __INSTR3(0, name, OPC_##name)

#define __INSTR4(flag, name, opc)                                        \
static inline struct ir3_instruction *                                   \
ir3_##name(struct ir3_block *block,                                      \
		struct ir3_instruction *a, unsigned aflags,                      \
		struct ir3_instruction *b, unsigned bflags,                      \
		struct ir3_instruction *c, unsigned cflags,                      \
		struct ir3_instruction *d, unsigned dflags)                      \
{                                                                        \
	struct ir3_instruction *instr =                                      \
		ir3_instr_create(block, opc, 1, 4);                              \
	__ssa_dst(instr);                                                    \
	__ssa_src(instr, a, aflags);                                         \
	__ssa_src(instr, b, bflags);                                         \
	__ssa_src(instr, c, cflags);                                         \
	__ssa_src(instr, d, dflags);                                         \
	instr->flags |= flag;                                                \
	return instr;                                                        \
}
#define INSTR4F(f, name)    __INSTR4(IR3_INSTR_##f, name##_##f, OPC_##name)
#define INSTR4(name)        __INSTR4(0, name, OPC_##name)

/* cat0 instructions: */
INSTR1(B)
INSTR0(JUMP)
INSTR1(KILL)
INSTR1(DEMOTE)
INSTR0(END)
INSTR0(CHSH)
INSTR0(CHMASK)
INSTR1(PREDT)
INSTR0(PREDF)
INSTR0(PREDE)

/* cat2 instructions, most 2 src but some 1 src: */
INSTR2(ADD_F)
INSTR2(MIN_F)
INSTR2(MAX_F)
INSTR2(MUL_F)
INSTR1(SIGN_F)
INSTR2(CMPS_F)
INSTR1(ABSNEG_F)
INSTR2(CMPV_F)
INSTR1(FLOOR_F)
INSTR1(CEIL_F)
INSTR1(RNDNE_F)
INSTR1(RNDAZ_F)
INSTR1(TRUNC_F)
INSTR2(ADD_U)
INSTR2(ADD_S)
INSTR2(SUB_U)
INSTR2(SUB_S)
INSTR2(CMPS_U)
INSTR2(CMPS_S)
INSTR2(MIN_U)
INSTR2(MIN_S)
INSTR2(MAX_U)
INSTR2(MAX_S)
INSTR1(ABSNEG_S)
INSTR2(AND_B)
INSTR2(OR_B)
INSTR1(NOT_B)
INSTR2(XOR_B)
INSTR2(CMPV_U)
INSTR2(CMPV_S)
INSTR2(MUL_U24)
INSTR2(MUL_S24)
INSTR2(MULL_U)
INSTR1(BFREV_B)
INSTR1(CLZ_S)
INSTR1(CLZ_B)
INSTR2(SHL_B)
INSTR2(SHR_B)
INSTR2(ASHR_B)
INSTR2(BARY_F)
INSTR2(MGEN_B)
INSTR2(GETBIT_B)
INSTR1(SETRM)
INSTR1(CBITS_B)
INSTR2(SHB)
INSTR2(MSAD)

/* cat3 instructions: */
INSTR3(MAD_U16)
INSTR3(MADSH_U16)
INSTR3(MAD_S16)
INSTR3(MADSH_M16)
INSTR3(MAD_U24)
INSTR3(MAD_S24)
INSTR3(MAD_F16)
INSTR3(MAD_F32)
/* NOTE: SEL_B32 checks for zero vs nonzero */
INSTR3(SEL_B16)
INSTR3(SEL_B32)
INSTR3(SEL_S16)
INSTR3(SEL_S32)
INSTR3(SEL_F16)
INSTR3(SEL_F32)
INSTR3(SAD_S16)
INSTR3(SAD_S32)

/* cat4 instructions: */
INSTR1(RCP)
INSTR1(RSQ)
INSTR1(HRSQ)
INSTR1(LOG2)
INSTR1(HLOG2)
INSTR1(EXP2)
INSTR1(HEXP2)
INSTR1(SIN)
INSTR1(COS)
INSTR1(SQRT)

/* cat5 instructions: */
INSTR1(DSX)
INSTR1(DSXPP_MACRO)
INSTR1(DSY)
INSTR1(DSYPP_MACRO)
INSTR1F(3D, DSX)
INSTR1F(3D, DSY)
INSTR1(RGETPOS)

static inline struct ir3_instruction *
ir3_SAM(struct ir3_block *block, opc_t opc, type_t type,
		unsigned wrmask, unsigned flags, struct ir3_instruction *samp_tex,
		struct ir3_instruction *src0, struct ir3_instruction *src1)
{
	struct ir3_instruction *sam;
	unsigned nreg = 0;

	if (flags & IR3_INSTR_S2EN) {
		nreg++;
	}
	if (src0) {
		nreg++;
	}
	if (src1) {
		nreg++;
	}

	sam = ir3_instr_create(block, opc, 1, nreg);
	sam->flags |= flags;
	__ssa_dst(sam)->wrmask = wrmask;
	if (flags & IR3_INSTR_S2EN) {
		__ssa_src(sam, samp_tex, (flags & IR3_INSTR_B) ? 0 : IR3_REG_HALF);
	}
	if (src0) {
		__ssa_src(sam, src0, 0);
	}
	if (src1) {
		__ssa_src(sam, src1, 0);
	}
	sam->cat5.type  = type;

	return sam;
}

/* cat6 instructions: */
INSTR2(LDLV)
INSTR3(LDG)
INSTR3(LDL)
INSTR3(LDLW)
INSTR3(LDP)
INSTR3(STG)
INSTR3(STL)
INSTR3(STLW)
INSTR3(STP)
INSTR1(RESINFO)
INSTR1(RESFMT)
INSTR2(ATOMIC_ADD)
INSTR2(ATOMIC_SUB)
INSTR2(ATOMIC_XCHG)
INSTR2(ATOMIC_INC)
INSTR2(ATOMIC_DEC)
INSTR2(ATOMIC_CMPXCHG)
INSTR2(ATOMIC_MIN)
INSTR2(ATOMIC_MAX)
INSTR2(ATOMIC_AND)
INSTR2(ATOMIC_OR)
INSTR2(ATOMIC_XOR)
INSTR2(LDC)
#if GPU >= 600
INSTR3(STIB);
INSTR2(LDIB);
INSTR3F(G, ATOMIC_ADD)
INSTR3F(G, ATOMIC_SUB)
INSTR3F(G, ATOMIC_XCHG)
INSTR3F(G, ATOMIC_INC)
INSTR3F(G, ATOMIC_DEC)
INSTR3F(G, ATOMIC_CMPXCHG)
INSTR3F(G, ATOMIC_MIN)
INSTR3F(G, ATOMIC_MAX)
INSTR3F(G, ATOMIC_AND)
INSTR3F(G, ATOMIC_OR)
INSTR3F(G, ATOMIC_XOR)
#elif GPU >= 400
INSTR3(LDGB)
INSTR4(STGB)
INSTR4(STIB)
INSTR4F(G, ATOMIC_ADD)
INSTR4F(G, ATOMIC_SUB)
INSTR4F(G, ATOMIC_XCHG)
INSTR4F(G, ATOMIC_INC)
INSTR4F(G, ATOMIC_DEC)
INSTR4F(G, ATOMIC_CMPXCHG)
INSTR4F(G, ATOMIC_MIN)
INSTR4F(G, ATOMIC_MAX)
INSTR4F(G, ATOMIC_AND)
INSTR4F(G, ATOMIC_OR)
INSTR4F(G, ATOMIC_XOR)
#endif

INSTR4F(G, STG)

/* cat7 instructions: */
INSTR0(BAR)
INSTR0(FENCE)

/* ************************************************************************* */
#include "regmask.h"

static inline void regmask_set(regmask_t *regmask, struct ir3_register *reg)
{
	bool half = reg->flags & IR3_REG_HALF;
	if (reg->flags & IR3_REG_RELATIV) {
		for (unsigned i = 0; i < reg->size; i++)
			__regmask_set(regmask, half, reg->array.base + i);
	} else {
		for (unsigned mask = reg->wrmask, n = reg->num; mask; mask >>= 1, n++)
			if (mask & 1)
				__regmask_set(regmask, half, n);
	}
}

static inline bool regmask_get(regmask_t *regmask,
		struct ir3_register *reg)
{
	bool half = reg->flags & IR3_REG_HALF;
	if (reg->flags & IR3_REG_RELATIV) {
		for (unsigned i = 0; i < reg->size; i++)
			if (__regmask_get(regmask, half, reg->array.base + i))
				return true;
	} else {
		for (unsigned mask = reg->wrmask, n = reg->num; mask; mask >>= 1, n++)
			if (mask & 1)
				if (__regmask_get(regmask, half, n))
					return true;
	}
	return false;
}
/* ************************************************************************* */

#endif /* IR3_H_ */