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
|
/*
* Copyright 2010 Christoph Bumiller
*
* 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 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 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.
*/
/* #define NV50_TGSI2NC_DEBUG */
/* XXX: need to clean this up so we get the typecasting right more naturally */
#include <unistd.h>
#include "nv50_context.h"
#include "nv50_pc.h"
#include "pipe/p_shader_tokens.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_util.h"
#include "util/u_simple_list.h"
#include "tgsi/tgsi_dump.h"
#define BLD_MAX_TEMPS 64
#define BLD_MAX_ADDRS 4
#define BLD_MAX_PREDS 4
#define BLD_MAX_IMMDS 128
#define BLD_MAX_COND_NESTING 4
#define BLD_MAX_LOOP_NESTING 4
#define BLD_MAX_CALL_NESTING 2
/* collects all values assigned to the same TGSI register */
struct bld_value_stack {
struct nv_value *top;
struct nv_value **body;
unsigned size;
uint16_t loop_use; /* 1 bit per loop level, indicates if used/defd */
uint16_t loop_def;
};
static INLINE void
bld_vals_push_val(struct bld_value_stack *stk, struct nv_value *val)
{
assert(!stk->size || (stk->body[stk->size - 1] != val));
if (!(stk->size % 8)) {
unsigned old_sz = (stk->size + 0) * sizeof(struct nv_value *);
unsigned new_sz = (stk->size + 8) * sizeof(struct nv_value *);
stk->body = (struct nv_value **)REALLOC(stk->body, old_sz, new_sz);
}
stk->body[stk->size++] = val;
}
static INLINE boolean
bld_vals_del_val(struct bld_value_stack *stk, struct nv_value *val)
{
unsigned i;
for (i = stk->size - 1; i >= 0; --i)
if (stk->body[i] == val)
break;
if (i < 0)
return FALSE;
if (i != stk->size - 1)
stk->body[i] = stk->body[stk->size - 1];
--stk->size; /* XXX: old size in REALLOC */
return TRUE;
}
static INLINE void
bld_vals_push(struct bld_value_stack *stk)
{
bld_vals_push_val(stk, stk->top);
stk->top = NULL;
}
static INLINE void
bld_push_values(struct bld_value_stack *stacks, int n)
{
int i, c;
for (i = 0; i < n; ++i)
for (c = 0; c < 4; ++c)
if (stacks[i * 4 + c].top)
bld_vals_push(&stacks[i * 4 + c]);
}
struct bld_context {
struct nv50_translation_info *ti;
struct nv_pc *pc;
struct nv_basic_block *b;
struct tgsi_parse_context parse[BLD_MAX_CALL_NESTING];
int call_lvl;
struct nv_basic_block *cond_bb[BLD_MAX_COND_NESTING];
struct nv_basic_block *join_bb[BLD_MAX_COND_NESTING];
struct nv_basic_block *else_bb[BLD_MAX_COND_NESTING];
int cond_lvl;
struct nv_basic_block *loop_bb[BLD_MAX_LOOP_NESTING];
struct nv_basic_block *brkt_bb[BLD_MAX_LOOP_NESTING];
int loop_lvl;
ubyte out_kind; /* CFG_EDGE_FORWARD, or FAKE in case of BREAK/CONT */
struct bld_value_stack tvs[BLD_MAX_TEMPS][4]; /* TGSI_FILE_TEMPORARY */
struct bld_value_stack avs[BLD_MAX_ADDRS][4]; /* TGSI_FILE_ADDRESS */
struct bld_value_stack pvs[BLD_MAX_PREDS][4]; /* TGSI_FILE_PREDICATE */
struct bld_value_stack ovs[PIPE_MAX_SHADER_OUTPUTS][4];
uint32_t outputs_written[(PIPE_MAX_SHADER_OUTPUTS + 31) / 32];
struct nv_value *frgcrd[4];
struct nv_value *sysval[4];
/* wipe on new BB */
struct nv_value *saved_addr[4][2];
struct nv_value *saved_inputs[128];
struct nv_value *saved_immd[BLD_MAX_IMMDS];
uint num_immds;
};
static INLINE ubyte
bld_stack_file(struct bld_context *bld, struct bld_value_stack *stk)
{
if (stk < &bld->avs[0][0])
return NV_FILE_GPR;
else
if (stk < &bld->pvs[0][0])
return NV_FILE_ADDR;
else
if (stk < &bld->ovs[0][0])
return NV_FILE_FLAGS;
else
return NV_FILE_OUT;
}
static INLINE struct nv_value *
bld_fetch(struct bld_context *bld, struct bld_value_stack *stk, int i, int c)
{
stk[i * 4 + c].loop_use |= 1 << bld->loop_lvl;
return stk[i * 4 + c].top;
}
static struct nv_value *
bld_loop_phi(struct bld_context *, struct bld_value_stack *, struct nv_value *);
/* If a variable is defined in a loop without prior use, we don't need
* a phi in the loop header to account for backwards flow.
*
* However, if this variable is then also used outside the loop, we do
* need a phi after all. But we must not use this phi's def inside the
* loop, so we can eliminate the phi if it is unused later.
*/
static INLINE void
bld_store(struct bld_context *bld, struct bld_value_stack *stk, int i, int c,
struct nv_value *val)
{
const uint16_t m = 1 << bld->loop_lvl;
stk = &stk[i * 4 + c];
if (bld->loop_lvl && !(m & (stk->loop_def | stk->loop_use)))
bld_loop_phi(bld, stk, val);
stk->top = val;
stk->loop_def |= 1 << bld->loop_lvl;
}
static INLINE void
bld_clear_def_use(struct bld_value_stack *stk, int n, int lvl)
{
int i;
const uint16_t mask = ~(1 << lvl);
for (i = 0; i < n * 4; ++i) {
stk[i].loop_def &= mask;
stk[i].loop_use &= mask;
}
}
#define FETCH_TEMP(i, c) bld_fetch(bld, &bld->tvs[0][0], i, c)
#define STORE_TEMP(i, c, v) bld_store(bld, &bld->tvs[0][0], i, c, (v))
#define FETCH_ADDR(i, c) bld_fetch(bld, &bld->avs[0][0], i, c)
#define STORE_ADDR(i, c, v) bld_store(bld, &bld->avs[0][0], i, c, (v))
#define FETCH_PRED(i, c) bld_fetch(bld, &bld->pvs[0][0], i, c)
#define STORE_PRED(i, c, v) bld_store(bld, &bld->pvs[0][0], i, c, (v))
#define STORE_OUTR(i, c, v) \
do { \
bld->ovs[i][c].top = (v); \
bld->outputs_written[(i) / 8] |= 1 << (((i) * 4 + (c)) % 32); \
} while (0)
static INLINE void
bld_warn_uninitialized(struct bld_context *bld, int kind,
struct bld_value_stack *stk, struct nv_basic_block *b)
{
long i = (stk - &bld->tvs[0][0]) / 4;
long c = (stk - &bld->tvs[0][0]) & 3;
if (c == 3)
c = -1;
debug_printf("WARNING: TEMP[%li].%c %s used uninitialized in BB:%i\n",
i, (int)('x' + c), kind ? "may be" : "is", b->id);
}
static INLINE struct nv_value *
bld_def(struct nv_instruction *i, int c, struct nv_value *value)
{
i->def[c] = value;
value->insn = i;
return value;
}
static INLINE struct nv_value *
find_by_bb(struct bld_value_stack *stack, struct nv_basic_block *b)
{
int i;
if (stack->top && stack->top->insn->bb == b)
return stack->top;
for (i = stack->size - 1; i >= 0; --i)
if (stack->body[i]->insn->bb == b)
return stack->body[i];
return NULL;
}
/* fetch value from stack that was defined in the specified basic block,
* or search for first definitions in all of its predecessors
*/
static void
fetch_by_bb(struct bld_value_stack *stack,
struct nv_value **vals, int *n,
struct nv_basic_block *b)
{
int i;
struct nv_value *val;
assert(*n < 16); /* MAX_COND_NESTING */
val = find_by_bb(stack, b);
if (val) {
for (i = 0; i < *n; ++i)
if (vals[i] == val)
return;
vals[(*n)++] = val;
return;
}
for (i = 0; i < b->num_in; ++i)
if (!IS_WALL_EDGE(b->in_kind[i]))
fetch_by_bb(stack, vals, n, b->in[i]);
}
static INLINE struct nv_value *
bld_load_imm_u32(struct bld_context *bld, uint32_t u);
static INLINE struct nv_value *
bld_undef(struct bld_context *bld, ubyte file)
{
struct nv_instruction *nvi = new_instruction(bld->pc, NV_OP_UNDEF);
return bld_def(nvi, 0, new_value(bld->pc, file, NV_TYPE_U32));
}
static struct nv_value *
bld_phi(struct bld_context *bld, struct nv_basic_block *b,
struct bld_value_stack *stack)
{
struct nv_basic_block *in;
struct nv_value *vals[16], *val;
struct nv_instruction *phi;
int i, j, n;
do {
i = n = 0;
fetch_by_bb(stack, vals, &n, b);
if (!n) {
bld_warn_uninitialized(bld, 0, stack, b);
return NULL;
}
if (n == 1) {
if (nvbb_dominated_by(b, vals[0]->insn->bb))
break;
bld_warn_uninitialized(bld, 1, stack, b);
/* back-tracking to insert missing value of other path */
in = b;
while (in->in[0]) {
if (in->num_in == 1) {
in = in->in[0];
} else {
if (!nvbb_reachable_by(in->in[0], vals[0]->insn->bb, b))
in = in->in[0];
else
if (!nvbb_reachable_by(in->in[1], vals[0]->insn->bb, b))
in = in->in[1];
else
in = in->in[0];
}
}
bld->pc->current_block = in;
/* should make this a no-op */
bld_vals_push_val(stack, bld_undef(bld, vals[0]->reg.file));
continue;
}
for (i = 0; i < n; ++i) {
/* if value dominates b, continue to the redefinitions */
if (nvbb_dominated_by(b, vals[i]->insn->bb))
continue;
/* if value dominates any in-block, b should be the dom frontier */
for (j = 0; j < b->num_in; ++j)
if (nvbb_dominated_by(b->in[j], vals[i]->insn->bb))
break;
/* otherwise, find the dominance frontier and put the phi there */
if (j == b->num_in) {
in = nvbb_dom_frontier(vals[i]->insn->bb);
val = bld_phi(bld, in, stack);
bld_vals_push_val(stack, val);
break;
}
}
} while(i < n);
bld->pc->current_block = b;
if (n == 1)
return vals[0];
phi = new_instruction(bld->pc, NV_OP_PHI);
bld_def(phi, 0, new_value(bld->pc, vals[0]->reg.file, vals[0]->reg.type));
for (i = 0; i < n; ++i)
phi->src[i] = new_ref(bld->pc, vals[i]);
return phi->def[0];
}
/* Insert a phi function in the loop header.
* For nested loops, we need to insert phi functions in all the outer
* loop headers if they don't have one yet.
*
* @def: redefinition from inside loop, or NULL if to be replaced later
*/
static struct nv_value *
bld_loop_phi(struct bld_context *bld, struct bld_value_stack *stack,
struct nv_value *def)
{
struct nv_instruction *phi;
struct nv_basic_block *bb = bld->pc->current_block;
struct nv_value *val = NULL;
if (bld->loop_lvl > 1) {
--bld->loop_lvl;
if (!((stack->loop_def | stack->loop_use) & (1 << bld->loop_lvl)))
val = bld_loop_phi(bld, stack, NULL);
++bld->loop_lvl;
}
if (!val)
val = bld_phi(bld, bld->pc->current_block, stack); /* old definition */
if (!val) {
bld->pc->current_block = bld->loop_bb[bld->loop_lvl - 1]->in[0];
val = bld_undef(bld, bld_stack_file(bld, stack));
}
bld->pc->current_block = bld->loop_bb[bld->loop_lvl - 1];
phi = new_instruction(bld->pc, NV_OP_PHI);
bld_def(phi, 0, new_value_like(bld->pc, val));
if (!def)
def = phi->def[0];
bld_vals_push_val(stack, phi->def[0]);
phi->target = (struct nv_basic_block *)stack; /* cheat */
nv_reference(bld->pc, &phi->src[0], val);
nv_reference(bld->pc, &phi->src[1], def);
bld->pc->current_block = bb;
return phi->def[0];
}
static INLINE struct nv_value *
bld_fetch_global(struct bld_context *bld, struct bld_value_stack *stack)
{
const uint16_t m = 1 << bld->loop_lvl;
const uint16_t use = stack->loop_use;
stack->loop_use |= m;
/* If neither used nor def'd inside the loop, build a phi in foresight,
* so we don't have to replace stuff later on, which requires tracking.
*/
if (bld->loop_lvl && !((use | stack->loop_def) & m))
return bld_loop_phi(bld, stack, NULL);
return bld_phi(bld, bld->pc->current_block, stack);
}
static INLINE struct nv_value *
bld_imm_u32(struct bld_context *bld, uint32_t u)
{
int i;
unsigned n = bld->num_immds;
for (i = 0; i < n; ++i)
if (bld->saved_immd[i]->reg.imm.u32 == u)
return bld->saved_immd[i];
assert(n < BLD_MAX_IMMDS);
bld->num_immds++;
bld->saved_immd[n] = new_value(bld->pc, NV_FILE_IMM, NV_TYPE_U32);
bld->saved_immd[n]->reg.imm.u32 = u;
return bld->saved_immd[n];
}
static void
bld_replace_value(struct nv_pc *, struct nv_basic_block *, struct nv_value *,
struct nv_value *);
/* Replace the source of the phi in the loop header by the last assignment,
* or eliminate the phi function if there is no assignment inside the loop.
*
* Redundancy situation 1 - (used) but (not redefined) value:
* %3 = phi %0, %3 = %3 is used
* %3 = phi %0, %4 = is new definition
*
* Redundancy situation 2 - (not used) but (redefined) value:
* %3 = phi %0, %2 = %2 is used, %3 could be used outside, deleted by DCE
*/
static void
bld_loop_end(struct bld_context *bld, struct nv_basic_block *bb)
{
struct nv_basic_block *save = bld->pc->current_block;
struct nv_instruction *phi, *next;
struct nv_value *val;
struct bld_value_stack *stk;
int i, s, n;
for (phi = bb->phi; phi && phi->opcode == NV_OP_PHI; phi = next) {
next = phi->next;
stk = (struct bld_value_stack *)phi->target;
phi->target = NULL;
for (s = 1, n = 0; n < bb->num_in; ++n) {
if (bb->in_kind[n] != CFG_EDGE_BACK)
continue;
assert(s < 4);
bld->pc->current_block = bb->in[n];
val = bld_fetch_global(bld, stk);
for (i = 0; i < 4; ++i)
if (phi->src[i] && phi->src[i]->value == val)
break;
if (i == 4)
nv_reference(bld->pc, &phi->src[s++], val);
}
bld->pc->current_block = save;
if (phi->src[0]->value == phi->def[0] ||
phi->src[0]->value == phi->src[1]->value)
s = 1;
else
if (phi->src[1]->value == phi->def[0])
s = 0;
else
continue;
if (s >= 0) {
/* eliminate the phi */
bld_vals_del_val(stk, phi->def[0]);
++bld->pc->pass_seq;
bld_replace_value(bld->pc, bb, phi->def[0], phi->src[s]->value);
nv_nvi_delete(phi);
}
}
}
static INLINE struct nv_value *
bld_imm_f32(struct bld_context *bld, float f)
{
return bld_imm_u32(bld, fui(f));
}
#define SET_TYPE(v, t) ((v)->reg.type = NV_TYPE_##t)
static struct nv_value *
bld_insn_1(struct bld_context *bld, uint opcode, struct nv_value *src0)
{
struct nv_instruction *insn = new_instruction(bld->pc, opcode);
assert(insn);
nv_reference(bld->pc, &insn->src[0], src0); /* NOTE: new_ref would suffice */
return bld_def(insn, 0, new_value(bld->pc, NV_FILE_GPR, src0->reg.type));
}
static struct nv_value *
bld_insn_2(struct bld_context *bld, uint opcode,
struct nv_value *src0, struct nv_value *src1)
{
struct nv_instruction *insn = new_instruction(bld->pc, opcode);
nv_reference(bld->pc, &insn->src[0], src0);
nv_reference(bld->pc, &insn->src[1], src1);
return bld_def(insn, 0, new_value(bld->pc, NV_FILE_GPR, src0->reg.type));
}
static struct nv_value *
bld_insn_3(struct bld_context *bld, uint opcode,
struct nv_value *src0, struct nv_value *src1,
struct nv_value *src2)
{
struct nv_instruction *insn = new_instruction(bld->pc, opcode);
nv_reference(bld->pc, &insn->src[0], src0);
nv_reference(bld->pc, &insn->src[1], src1);
nv_reference(bld->pc, &insn->src[2], src2);
return bld_def(insn, 0, new_value(bld->pc, NV_FILE_GPR, src0->reg.type));
}
static void
bld_lmem_store(struct bld_context *bld, struct nv_value *ptr, int ofst,
struct nv_value *val)
{
struct nv_instruction *insn = new_instruction(bld->pc, NV_OP_STA);
struct nv_value *loc;
loc = new_value(bld->pc, NV_FILE_MEM_L, NV_TYPE_U32);
loc->reg.id = ofst * 4;
nv_reference(bld->pc, &insn->src[0], loc);
nv_reference(bld->pc, &insn->src[1], val);
nv_reference(bld->pc, &insn->src[4], ptr);
}
static struct nv_value *
bld_lmem_load(struct bld_context *bld, struct nv_value *ptr, int ofst)
{
struct nv_value *loc, *val;
loc = new_value(bld->pc, NV_FILE_MEM_L, NV_TYPE_U32);
loc->reg.id = ofst * 4;
val = bld_insn_1(bld, NV_OP_LDA, loc);
nv_reference(bld->pc, &val->insn->src[4], ptr);
return val;
}
#define BLD_INSN_1_EX(d, op, dt, s0, s0t) \
do { \
(d) = bld_insn_1(bld, (NV_OP_##op), (s0)); \
(d)->reg.type = NV_TYPE_##dt; \
(d)->insn->src[0]->typecast = NV_TYPE_##s0t; \
} while(0)
#define BLD_INSN_2_EX(d, op, dt, s0, s0t, s1, s1t) \
do { \
(d) = bld_insn_2(bld, (NV_OP_##op), (s0), (s1)); \
(d)->reg.type = NV_TYPE_##dt; \
(d)->insn->src[0]->typecast = NV_TYPE_##s0t; \
(d)->insn->src[1]->typecast = NV_TYPE_##s1t; \
} while(0)
static struct nv_value *
bld_pow(struct bld_context *bld, struct nv_value *x, struct nv_value *e)
{
struct nv_value *val;
BLD_INSN_1_EX(val, LG2, F32, x, F32);
BLD_INSN_2_EX(val, MUL, F32, e, F32, val, F32);
val = bld_insn_1(bld, NV_OP_PREEX2, val);
val = bld_insn_1(bld, NV_OP_EX2, val);
return val;
}
static INLINE struct nv_value *
bld_load_imm_f32(struct bld_context *bld, float f)
{
return bld_insn_1(bld, NV_OP_MOV, bld_imm_f32(bld, f));
}
static INLINE struct nv_value *
bld_load_imm_u32(struct bld_context *bld, uint32_t u)
{
return bld_insn_1(bld, NV_OP_MOV, bld_imm_u32(bld, u));
}
static struct nv_value *
bld_get_address(struct bld_context *bld, int id, struct nv_value *indirect)
{
int i;
struct nv_instruction *nvi;
for (i = 0; i < 4; ++i) {
if (!bld->saved_addr[i][0])
break;
if (bld->saved_addr[i][1] == indirect) {
nvi = bld->saved_addr[i][0]->insn;
if (nvi->src[0]->value->reg.imm.u32 == id)
return bld->saved_addr[i][0];
}
}
i &= 3;
bld->saved_addr[i][0] = bld_load_imm_u32(bld, id);
bld->saved_addr[i][0]->reg.file = NV_FILE_ADDR;
bld->saved_addr[i][1] = indirect;
return bld->saved_addr[i][0];
}
static struct nv_value *
bld_predicate(struct bld_context *bld, struct nv_value *src, boolean bool_only)
{
struct nv_instruction *s0i, *nvi = src->insn;
if (!nvi) {
nvi = bld_insn_1(bld,
(src->reg.file == NV_FILE_IMM) ? NV_OP_MOV : NV_OP_LDA,
src)->insn;
src = nvi->def[0];
} else
if (bool_only) {
while (nvi->opcode == NV_OP_ABS || nvi->opcode == NV_OP_NEG ||
nvi->opcode == NV_OP_CVT) {
s0i = nvi->src[0]->value->insn;
if (!s0i ||
s0i->opcode == NV_OP_LDA ||
s0i->opcode == NV_OP_MOV ||
s0i->opcode == NV_OP_PHI)
break;
nvi = s0i;
assert(!nvi->flags_src);
}
}
if (nvi->opcode == NV_OP_LDA ||
nvi->opcode == NV_OP_MOV ||
nvi->opcode == NV_OP_PHI || nvi->bb != bld->pc->current_block) {
nvi = new_instruction(bld->pc, NV_OP_CVT);
nv_reference(bld->pc, &nvi->src[0], src);
}
if (!nvi->flags_def) {
nvi->flags_def = new_value(bld->pc, NV_FILE_FLAGS, NV_TYPE_U16);
nvi->flags_def->insn = nvi;
}
return nvi->flags_def;
}
static void
bld_kil(struct bld_context *bld, struct nv_value *src)
{
struct nv_instruction *nvi;
src = bld_predicate(bld, src, FALSE);
nvi = new_instruction(bld->pc, NV_OP_KIL);
nvi->fixed = 1;
nvi->flags_src = new_ref(bld->pc, src);
nvi->cc = NV_CC_LT;
}
static void
bld_flow(struct bld_context *bld, uint opcode, ubyte cc,
struct nv_value *src, struct nv_basic_block *target,
boolean plan_reconverge)
{
struct nv_instruction *nvi;
if (plan_reconverge)
new_instruction(bld->pc, NV_OP_JOINAT)->fixed = 1;
nvi = new_instruction(bld->pc, opcode);
nvi->is_terminator = 1;
nvi->cc = cc;
nvi->target = target;
if (src)
nvi->flags_src = new_ref(bld->pc, src);
}
static ubyte
translate_setcc(unsigned opcode)
{
switch (opcode) {
case TGSI_OPCODE_SLT: return NV_CC_LT;
case TGSI_OPCODE_SGE: return NV_CC_GE;
case TGSI_OPCODE_SEQ: return NV_CC_EQ;
case TGSI_OPCODE_SGT: return NV_CC_GT;
case TGSI_OPCODE_SLE: return NV_CC_LE;
case TGSI_OPCODE_SNE: return NV_CC_NE | NV_CC_U;
case TGSI_OPCODE_STR: return NV_CC_TR;
case TGSI_OPCODE_SFL: return NV_CC_FL;
case TGSI_OPCODE_ISLT: return NV_CC_LT;
case TGSI_OPCODE_ISGE: return NV_CC_GE;
case TGSI_OPCODE_USEQ: return NV_CC_EQ;
case TGSI_OPCODE_USGE: return NV_CC_GE;
case TGSI_OPCODE_USLT: return NV_CC_LT;
case TGSI_OPCODE_USNE: return NV_CC_NE;
default:
assert(0);
return NV_CC_FL;
}
}
static uint
translate_opcode(uint opcode)
{
switch (opcode) {
case TGSI_OPCODE_ABS: return NV_OP_ABS;
case TGSI_OPCODE_ADD:
case TGSI_OPCODE_SUB:
case TGSI_OPCODE_UADD: return NV_OP_ADD;
case TGSI_OPCODE_AND: return NV_OP_AND;
case TGSI_OPCODE_EX2: return NV_OP_EX2;
case TGSI_OPCODE_CEIL: return NV_OP_CEIL;
case TGSI_OPCODE_FLR: return NV_OP_FLOOR;
case TGSI_OPCODE_TRUNC: return NV_OP_TRUNC;
case TGSI_OPCODE_COS: return NV_OP_COS;
case TGSI_OPCODE_SIN: return NV_OP_SIN;
case TGSI_OPCODE_DDX: return NV_OP_DFDX;
case TGSI_OPCODE_DDY: return NV_OP_DFDY;
case TGSI_OPCODE_F2I:
case TGSI_OPCODE_F2U:
case TGSI_OPCODE_I2F:
case TGSI_OPCODE_U2F: return NV_OP_CVT;
case TGSI_OPCODE_INEG: return NV_OP_NEG;
case TGSI_OPCODE_LG2: return NV_OP_LG2;
case TGSI_OPCODE_ISHR:
case TGSI_OPCODE_USHR: return NV_OP_SHR;
case TGSI_OPCODE_MAD:
case TGSI_OPCODE_UMAD: return NV_OP_MAD;
case TGSI_OPCODE_MAX:
case TGSI_OPCODE_IMAX:
case TGSI_OPCODE_UMAX: return NV_OP_MAX;
case TGSI_OPCODE_MIN:
case TGSI_OPCODE_IMIN:
case TGSI_OPCODE_UMIN: return NV_OP_MIN;
case TGSI_OPCODE_MUL:
case TGSI_OPCODE_UMUL: return NV_OP_MUL;
case TGSI_OPCODE_OR: return NV_OP_OR;
case TGSI_OPCODE_RCP: return NV_OP_RCP;
case TGSI_OPCODE_RSQ: return NV_OP_RSQ;
case TGSI_OPCODE_SAD: return NV_OP_SAD;
case TGSI_OPCODE_SHL: return NV_OP_SHL;
case TGSI_OPCODE_SLT:
case TGSI_OPCODE_SGE:
case TGSI_OPCODE_SEQ:
case TGSI_OPCODE_SGT:
case TGSI_OPCODE_SLE:
case TGSI_OPCODE_SNE:
case TGSI_OPCODE_ISLT:
case TGSI_OPCODE_ISGE:
case TGSI_OPCODE_USEQ:
case TGSI_OPCODE_USGE:
case TGSI_OPCODE_USLT:
case TGSI_OPCODE_USNE: return NV_OP_SET;
case TGSI_OPCODE_TEX: return NV_OP_TEX;
case TGSI_OPCODE_TXP: return NV_OP_TEX;
case TGSI_OPCODE_TXB: return NV_OP_TXB;
case TGSI_OPCODE_TXL: return NV_OP_TXL;
case TGSI_OPCODE_XOR: return NV_OP_XOR;
default:
return NV_OP_NOP;
}
}
static ubyte
infer_src_type(unsigned opcode)
{
switch (opcode) {
case TGSI_OPCODE_MOV:
case TGSI_OPCODE_AND:
case TGSI_OPCODE_OR:
case TGSI_OPCODE_XOR:
case TGSI_OPCODE_SAD:
case TGSI_OPCODE_U2F:
case TGSI_OPCODE_UADD:
case TGSI_OPCODE_UDIV:
case TGSI_OPCODE_UMOD:
case TGSI_OPCODE_UMAD:
case TGSI_OPCODE_UMUL:
case TGSI_OPCODE_UMAX:
case TGSI_OPCODE_UMIN:
case TGSI_OPCODE_USEQ:
case TGSI_OPCODE_USGE:
case TGSI_OPCODE_USLT:
case TGSI_OPCODE_USNE:
case TGSI_OPCODE_USHR:
return NV_TYPE_U32;
case TGSI_OPCODE_I2F:
case TGSI_OPCODE_IDIV:
case TGSI_OPCODE_IMAX:
case TGSI_OPCODE_IMIN:
case TGSI_OPCODE_INEG:
case TGSI_OPCODE_ISGE:
case TGSI_OPCODE_ISHR:
case TGSI_OPCODE_ISLT:
return NV_TYPE_S32;
default:
return NV_TYPE_F32;
}
}
static ubyte
infer_dst_type(unsigned opcode)
{
switch (opcode) {
case TGSI_OPCODE_MOV:
case TGSI_OPCODE_F2U:
case TGSI_OPCODE_AND:
case TGSI_OPCODE_OR:
case TGSI_OPCODE_XOR:
case TGSI_OPCODE_SAD:
case TGSI_OPCODE_UADD:
case TGSI_OPCODE_UDIV:
case TGSI_OPCODE_UMOD:
case TGSI_OPCODE_UMAD:
case TGSI_OPCODE_UMUL:
case TGSI_OPCODE_UMAX:
case TGSI_OPCODE_UMIN:
case TGSI_OPCODE_USEQ:
case TGSI_OPCODE_USGE:
case TGSI_OPCODE_USLT:
case TGSI_OPCODE_USNE:
case TGSI_OPCODE_USHR:
return NV_TYPE_U32;
case TGSI_OPCODE_F2I:
case TGSI_OPCODE_IDIV:
case TGSI_OPCODE_IMAX:
case TGSI_OPCODE_IMIN:
case TGSI_OPCODE_INEG:
case TGSI_OPCODE_ISGE:
case TGSI_OPCODE_ISHR:
case TGSI_OPCODE_ISLT:
return NV_TYPE_S32;
default:
return NV_TYPE_F32;
}
}
static void
emit_store(struct bld_context *bld, const struct tgsi_full_instruction *inst,
unsigned chan, struct nv_value *value)
{
struct nv_value *ptr;
const struct tgsi_full_dst_register *reg = &inst->Dst[0];
if (reg->Register.Indirect) {
ptr = FETCH_ADDR(reg->Indirect.Index,
tgsi_util_get_src_register_swizzle(®->Indirect, 0));
} else {
ptr = NULL;
}
assert(chan < 4);
if (inst->Instruction.Opcode != TGSI_OPCODE_MOV)
value->reg.type = infer_dst_type(inst->Instruction.Opcode);
switch (inst->Instruction.Saturate) {
case TGSI_SAT_NONE:
break;
case TGSI_SAT_ZERO_ONE:
BLD_INSN_1_EX(value, SAT, F32, value, F32);
break;
case TGSI_SAT_MINUS_PLUS_ONE:
value = bld_insn_2(bld, NV_OP_MAX, value, bld_load_imm_f32(bld, -1.0f));
value = bld_insn_2(bld, NV_OP_MIN, value, bld_load_imm_f32(bld, 1.0f));
value->reg.type = NV_TYPE_F32;
break;
}
switch (reg->Register.File) {
case TGSI_FILE_OUTPUT:
value = bld_insn_1(bld, NV_OP_MOV, value);
value->reg.file = bld->ti->output_file;
if (bld->ti->p->type == PIPE_SHADER_FRAGMENT) {
STORE_OUTR(reg->Register.Index, chan, value);
} else {
value->insn->fixed = 1;
value->reg.id = bld->ti->output_map[reg->Register.Index][chan];
}
break;
case TGSI_FILE_TEMPORARY:
assert(reg->Register.Index < BLD_MAX_TEMPS);
value->reg.file = NV_FILE_GPR;
if (value->insn->bb != bld->pc->current_block)
value = bld_insn_1(bld, NV_OP_MOV, value);
if (bld->ti->store_to_memory)
bld_lmem_store(bld, ptr, reg->Register.Index * 4 + chan, value);
else
STORE_TEMP(reg->Register.Index, chan, value);
break;
case TGSI_FILE_ADDRESS:
assert(reg->Register.Index < BLD_MAX_ADDRS);
value->reg.file = NV_FILE_ADDR;
STORE_ADDR(reg->Register.Index, chan, value);
break;
}
}
static INLINE uint32_t
bld_is_output_written(struct bld_context *bld, int i, int c)
{
if (c < 0)
return bld->outputs_written[i / 8] & (0xf << ((i * 4) % 32));
return bld->outputs_written[i / 8] & (1 << ((i * 4 + c) % 32));
}
static void
bld_export_outputs(struct bld_context *bld)
{
struct nv_value *vals[4];
struct nv_instruction *nvi;
int i, c, n;
bld_push_values(&bld->ovs[0][0], PIPE_MAX_SHADER_OUTPUTS);
for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; ++i) {
if (!bld_is_output_written(bld, i, -1))
continue;
for (n = 0, c = 0; c < 4; ++c) {
if (!bld_is_output_written(bld, i, c))
continue;
vals[n] = bld_fetch_global(bld, &bld->ovs[i][c]);
assert(vals[n]);
vals[n] = bld_insn_1(bld, NV_OP_MOV, vals[n]);
vals[n++]->reg.id = bld->ti->output_map[i][c];
}
assert(n);
(nvi = new_instruction(bld->pc, NV_OP_EXPORT))->fixed = 1;
for (c = 0; c < n; ++c)
nvi->src[c] = new_ref(bld->pc, vals[c]);
}
}
static void
bld_new_block(struct bld_context *bld, struct nv_basic_block *b)
{
int i;
bld_push_values(&bld->tvs[0][0], BLD_MAX_TEMPS);
bld_push_values(&bld->avs[0][0], BLD_MAX_ADDRS);
bld_push_values(&bld->pvs[0][0], BLD_MAX_PREDS);
bld_push_values(&bld->ovs[0][0], PIPE_MAX_SHADER_OUTPUTS);
bld->pc->current_block = b;
for (i = 0; i < 4; ++i)
bld->saved_addr[i][0] = NULL;
for (i = 0; i < 128; ++i)
bld->saved_inputs[i] = NULL;
bld->out_kind = CFG_EDGE_FORWARD;
}
static struct nv_value *
bld_saved_input(struct bld_context *bld, unsigned i, unsigned c)
{
unsigned idx = bld->ti->input_map[i][c];
if (bld->ti->p->type != PIPE_SHADER_FRAGMENT)
return NULL;
if (bld->saved_inputs[idx])
return bld->saved_inputs[idx];
return NULL;
}
static struct nv_value *
bld_interpolate(struct bld_context *bld, unsigned mode, struct nv_value *val)
{
if (val->reg.id == 255) {
/* gl_FrontFacing: 0/~0 to -1.0/+1.0 */
val = bld_insn_1(bld, NV_OP_LINTERP, val);
val = bld_insn_2(bld, NV_OP_SHL, val, bld_imm_u32(bld, 31));
val->insn->src[0]->typecast = NV_TYPE_U32;
val = bld_insn_2(bld, NV_OP_XOR, val, bld_imm_f32(bld, -1.0f));
val->insn->src[0]->typecast = NV_TYPE_U32;
} else
if (mode & (NV50_INTERP_LINEAR | NV50_INTERP_FLAT))
val = bld_insn_1(bld, NV_OP_LINTERP, val);
else
val = bld_insn_2(bld, NV_OP_PINTERP, val, bld->frgcrd[3]);
val->insn->flat = (mode & NV50_INTERP_FLAT) ? 1 : 0;
val->insn->centroid = (mode & NV50_INTERP_CENTROID) ? 1 : 0;
return val;
}
static struct nv_value *
emit_fetch(struct bld_context *bld, const struct tgsi_full_instruction *insn,
const unsigned s, const unsigned chan)
{
const struct tgsi_full_src_register *src = &insn->Src[s];
struct nv_value *res;
struct nv_value *ptr = NULL;
unsigned idx, swz, dim_idx, ind_idx, ind_swz;
ubyte type = infer_src_type(insn->Instruction.Opcode);
idx = src->Register.Index;
swz = tgsi_util_get_full_src_register_swizzle(src, chan);
dim_idx = -1;
ind_idx = -1;
ind_swz = 0;
if (src->Register.Indirect) {
ind_idx = src->Indirect.Index;
ind_swz = tgsi_util_get_src_register_swizzle(&src->Indirect, 0);
ptr = FETCH_ADDR(ind_idx, ind_swz);
}
if (idx >= (128 / 4) && src->Register.File == TGSI_FILE_CONSTANT)
ptr = bld_get_address(bld, (idx * 16) & ~0x1ff, ptr);
switch (src->Register.File) {
case TGSI_FILE_CONSTANT:
dim_idx = src->Dimension.Index ? src->Dimension.Index + 2 : 1;
assert(dim_idx < 14);
assert(dim_idx == 1); /* for now */
res = new_value(bld->pc, NV_FILE_MEM_C(dim_idx), type);
res->reg.type = type;
res->reg.id = (idx * 4 + swz) & 127;
res = bld_insn_1(bld, NV_OP_LDA, res);
if (ptr)
res->insn->src[4] = new_ref(bld->pc, ptr);
break;
case TGSI_FILE_IMMEDIATE:
assert(idx < bld->ti->immd32_nr);
res = bld_load_imm_u32(bld, bld->ti->immd32[idx * 4 + swz]);
switch (bld->ti->immd32_ty[idx]) {
case TGSI_IMM_FLOAT32: res->reg.type = NV_TYPE_F32; break;
case TGSI_IMM_UINT32: res->reg.type = NV_TYPE_U32; break;
case TGSI_IMM_INT32: res->reg.type = NV_TYPE_S32; break;
default:
res->reg.type = type;
break;
}
break;
case TGSI_FILE_INPUT:
res = bld_saved_input(bld, idx, swz);
if (res && (insn->Instruction.Opcode != TGSI_OPCODE_TXP))
return res;
res = new_value(bld->pc, bld->ti->input_file, type);
res->reg.id = bld->ti->input_map[idx][swz];
if (res->reg.file == NV_FILE_MEM_V) {
res = bld_interpolate(bld, bld->ti->interp_mode[idx], res);
} else {
assert(src->Dimension.Dimension == 0);
res = bld_insn_1(bld, NV_OP_LDA, res);
assert(res->reg.type == type);
}
bld->saved_inputs[bld->ti->input_map[idx][swz]] = res;
break;
case TGSI_FILE_TEMPORARY:
if (bld->ti->store_to_memory)
res = bld_lmem_load(bld, ptr, idx * 4 + swz);
else
res = bld_fetch_global(bld, &bld->tvs[idx][swz]);
break;
case TGSI_FILE_ADDRESS:
res = bld_fetch_global(bld, &bld->avs[idx][swz]);
break;
case TGSI_FILE_PREDICATE:
res = bld_fetch_global(bld, &bld->pvs[idx][swz]);
break;
default:
NOUVEAU_ERR("illegal/unhandled src reg file: %d\n", src->Register.File);
abort();
break;
}
if (!res)
return bld_undef(bld, NV_FILE_GPR);
switch (tgsi_util_get_full_src_register_sign_mode(src, chan)) {
case TGSI_UTIL_SIGN_KEEP:
break;
case TGSI_UTIL_SIGN_CLEAR:
res = bld_insn_1(bld, NV_OP_ABS, res);
break;
case TGSI_UTIL_SIGN_TOGGLE:
res = bld_insn_1(bld, NV_OP_NEG, res);
break;
case TGSI_UTIL_SIGN_SET:
res = bld_insn_1(bld, NV_OP_ABS, res);
res = bld_insn_1(bld, NV_OP_NEG, res);
break;
default:
NOUVEAU_ERR("illegal/unhandled src reg sign mode\n");
abort();
break;
}
return res;
}
static void
bld_lit(struct bld_context *bld, struct nv_value *dst0[4],
const struct tgsi_full_instruction *insn)
{
struct nv_value *val0, *zero;
unsigned mask = insn->Dst[0].Register.WriteMask;
if (mask & ((1 << 0) | (1 << 3)))
dst0[3] = dst0[0] = bld_load_imm_f32(bld, 1.0f);
if (mask & (3 << 1)) {
zero = bld_load_imm_f32(bld, 0.0f);
val0 = bld_insn_2(bld, NV_OP_MAX, emit_fetch(bld, insn, 0, 0), zero);
if (mask & (1 << 1))
dst0[1] = val0;
}
if (mask & (1 << 2)) {
struct nv_value *val1, *val3, *src1, *src3;
struct nv_value *pos128 = bld_load_imm_f32(bld, 127.999999f);
struct nv_value *neg128 = bld_load_imm_f32(bld, -127.999999f);
src1 = emit_fetch(bld, insn, 0, 1);
src3 = emit_fetch(bld, insn, 0, 3);
val0->insn->flags_def = new_value(bld->pc, NV_FILE_FLAGS, NV_TYPE_U16);
val0->insn->flags_def->insn = val0->insn;
val1 = bld_insn_2(bld, NV_OP_MAX, src1, zero);
val3 = bld_insn_2(bld, NV_OP_MAX, src3, neg128);
val3 = bld_insn_2(bld, NV_OP_MIN, val3, pos128);
val3 = bld_pow(bld, val1, val3);
dst0[2] = bld_insn_1(bld, NV_OP_MOV, zero);
dst0[2]->insn->cc = NV_CC_LE;
dst0[2]->insn->flags_src = new_ref(bld->pc, val0->insn->flags_def);
dst0[2] = bld_insn_2(bld, NV_OP_SELECT, val3, dst0[2]);
}
}
static INLINE void
get_tex_dim(const struct tgsi_full_instruction *insn, int *dim, int *arg)
{
switch (insn->Texture.Texture) {
case TGSI_TEXTURE_1D:
*arg = *dim = 1;
break;
case TGSI_TEXTURE_SHADOW1D:
*dim = 1;
*arg = 2;
break;
case TGSI_TEXTURE_UNKNOWN:
case TGSI_TEXTURE_2D:
case TGSI_TEXTURE_RECT:
*arg = *dim = 2;
break;
case TGSI_TEXTURE_SHADOW2D:
case TGSI_TEXTURE_SHADOWRECT:
*dim = 2;
*arg = 3;
break;
case TGSI_TEXTURE_3D:
case TGSI_TEXTURE_CUBE:
*dim = *arg = 3;
break;
default:
assert(0);
break;
}
}
static void
load_proj_tex_coords(struct bld_context *bld,
struct nv_value *t[4], int dim,
const struct tgsi_full_instruction *insn)
{
int c, mask = 0;
t[3] = emit_fetch(bld, insn, 0, 3);
if (t[3]->insn->opcode == NV_OP_PINTERP) {
t[3]->insn->opcode = NV_OP_LINTERP;
nv_reference(bld->pc, &t[3]->insn->src[1], NULL);
}
t[3] = bld_insn_1(bld, NV_OP_RCP, t[3]);
for (c = 0; c < dim; ++c) {
t[c] = emit_fetch(bld, insn, 0, c);
if (t[c]->insn->opcode == NV_OP_LINTERP)
t[c]->insn->opcode = NV_OP_PINTERP;
if (t[c]->insn->opcode == NV_OP_PINTERP)
nv_reference(bld->pc, &t[c]->insn->src[1], t[3]);
else
mask |= 1 << c;
}
for (c = 0; mask; ++c, mask >>= 1) {
if (!(mask & 1))
continue;
t[c] = bld_insn_2(bld, NV_OP_MUL, t[c], t[3]);
}
}
/* For a quad of threads / top left, top right, bottom left, bottom right
* pixels, do a different operation, and take src0 from a specific thread.
*/
#define QOP_ADD 0
#define QOP_SUBR 1
#define QOP_SUB 2
#define QOP_MOV1 3
#define QOP(a, b, c, d) \
((QOP_##a << 0) | (QOP_##b << 2) | (QOP_##c << 4) | (QOP_##d << 6))
static INLINE struct nv_value *
bld_quadop(struct bld_context *bld, ubyte qop, struct nv_value *src0, int lane,
struct nv_value *src1, boolean wp)
{
struct nv_value *val = bld_insn_2(bld, NV_OP_QUADOP, src0, src1);
val->insn->lanes = lane;
val->insn->quadop = qop;
if (wp) {
val->insn->flags_def = new_value(bld->pc, NV_FILE_FLAGS, NV_TYPE_U16);
val->insn->flags_def->insn = val->insn;
}
return val;
}
static INLINE struct nv_value *
bld_cmov(struct bld_context *bld,
struct nv_value *src, ubyte cc, struct nv_value *cr)
{
src = bld_insn_1(bld, NV_OP_MOV, src);
src->insn->cc = cc;
src->insn->flags_src = new_ref(bld->pc, cr);
return src;
}
static struct nv_instruction *
emit_tex(struct bld_context *bld, uint opcode,
struct nv_value *dst[4], struct nv_value *t_in[4],
int argc, int tic, int tsc, int cube)
{
struct nv_value *t[4];
struct nv_instruction *nvi;
int c;
/* the inputs to a tex instruction must be separate values */
for (c = 0; c < argc; ++c) {
t[c] = bld_insn_1(bld, NV_OP_MOV, t_in[c]);
t[c]->reg.type = NV_TYPE_F32;
t[c]->insn->fixed = 1;
}
nvi = new_instruction(bld->pc, opcode);
for (c = 0; c < 4; ++c)
dst[c] = bld_def(nvi, c, new_value(bld->pc, NV_FILE_GPR, NV_TYPE_F32));
for (c = 0; c < argc; ++c)
nvi->src[c] = new_ref(bld->pc, t[c]);
nvi->tex_t = tic;
nvi->tex_s = tsc;
nvi->tex_mask = 0xf;
nvi->tex_cube = cube;
nvi->tex_live = 0;
nvi->tex_argc = argc;
return nvi;
}
static void
bld_texlod_sequence(struct bld_context *bld,
struct nv_value *dst[4], struct nv_value *t[4], int arg,
int tic, int tsc, int cube)
{
emit_tex(bld, NV_OP_TXL, dst, t, arg, tic, tsc, cube); /* TODO */
}
/* The lanes of a quad are grouped by the bit in the condition register
* they have set, which is selected by differing bias values.
* Move the input values for TEX into a new register set for each group
* and execute TEX only for a specific group.
* We always need to use 4 new registers for the inputs/outputs because
* the implicitly calculated derivatives must be correct.
*/
static void
bld_texbias_sequence(struct bld_context *bld,
struct nv_value *dst[4], struct nv_value *t[4], int arg,
int tic, int tsc, int cube)
{
struct nv_instruction *sel, *tex;
struct nv_value *bit[4], *cr[4], *res[4][4], *val;
int l, c;
const ubyte cc[4] = { NV_CC_EQ, NV_CC_S, NV_CC_C, NV_CC_O };
for (l = 0; l < 4; ++l) {
bit[l] = bld_load_imm_u32(bld, 1 << l);
val = bld_quadop(bld, QOP(SUBR, SUBR, SUBR, SUBR),
t[arg - 1], l, t[arg - 1], TRUE);
cr[l] = bld_cmov(bld, bit[l], NV_CC_EQ, val->insn->flags_def);
cr[l]->reg.file = NV_FILE_FLAGS;
cr[l]->reg.type = NV_TYPE_U16;
}
sel = new_instruction(bld->pc, NV_OP_SELECT);
for (l = 0; l < 4; ++l)
sel->src[l] = new_ref(bld->pc, cr[l]);
bld_def(sel, 0, new_value(bld->pc, NV_FILE_FLAGS, NV_TYPE_U16));
for (l = 0; l < 4; ++l) {
tex = emit_tex(bld, NV_OP_TXB, dst, t, arg, tic, tsc, cube);
tex->cc = cc[l];
tex->flags_src = new_ref(bld->pc, sel->def[0]);
for (c = 0; c < 4; ++c)
res[l][c] = tex->def[c];
}
for (l = 0; l < 4; ++l)
for (c = 0; c < 4; ++c)
res[l][c] = bld_cmov(bld, res[l][c], cc[l], sel->def[0]);
for (c = 0; c < 4; ++c) {
sel = new_instruction(bld->pc, NV_OP_SELECT);
for (l = 0; l < 4; ++l)
sel->src[l] = new_ref(bld->pc, res[l][c]);
bld_def(sel, 0, (dst[c] = new_value(bld->pc, NV_FILE_GPR, NV_TYPE_F32)));
}
}
static boolean
bld_is_constant(struct nv_value *val)
{
if (val->reg.file == NV_FILE_IMM)
return TRUE;
return val->insn && nvcg_find_constant(val->insn->src[0]);
}
static void
bld_tex(struct bld_context *bld, struct nv_value *dst0[4],
const struct tgsi_full_instruction *insn)
{
struct nv_value *t[4], *s[3];
uint opcode = translate_opcode(insn->Instruction.Opcode);
int arg, dim, c;
const int tic = insn->Src[1].Register.Index;
const int tsc = 0;
const int cube = (insn->Texture.Texture == TGSI_TEXTURE_CUBE) ? 1 : 0;
get_tex_dim(insn, &dim, &arg);
if (!cube && insn->Instruction.Opcode == TGSI_OPCODE_TXP)
load_proj_tex_coords(bld, t, dim, insn);
else
for (c = 0; c < dim; ++c)
t[c] = emit_fetch(bld, insn, 0, c);
if (cube) {
assert(dim >= 3);
for (c = 0; c < 3; ++c)
s[c] = bld_insn_1(bld, NV_OP_ABS, t[c]);
s[0] = bld_insn_2(bld, NV_OP_MAX, s[0], s[1]);
s[0] = bld_insn_2(bld, NV_OP_MAX, s[0], s[2]);
s[0] = bld_insn_1(bld, NV_OP_RCP, s[0]);
for (c = 0; c < 3; ++c)
t[c] = bld_insn_2(bld, NV_OP_MUL, t[c], s[0]);
}
if (arg != dim)
t[dim] = emit_fetch(bld, insn, 0, 2);
if (opcode == NV_OP_TXB || opcode == NV_OP_TXL) {
t[arg++] = emit_fetch(bld, insn, 0, 3);
if ((bld->ti->p->type == PIPE_SHADER_FRAGMENT) &&
!bld_is_constant(t[arg - 1])) {
if (opcode == NV_OP_TXB)
bld_texbias_sequence(bld, dst0, t, arg, tic, tsc, cube);
else
bld_texlod_sequence(bld, dst0, t, arg, tic, tsc, cube);
return;
}
}
emit_tex(bld, opcode, dst0, t, arg, tic, tsc, cube);
}
static INLINE struct nv_value *
bld_dot(struct bld_context *bld, const struct tgsi_full_instruction *insn,
int n)
{
struct nv_value *dotp, *src0, *src1;
int c;
src0 = emit_fetch(bld, insn, 0, 0);
src1 = emit_fetch(bld, insn, 1, 0);
dotp = bld_insn_2(bld, NV_OP_MUL, src0, src1);
for (c = 1; c < n; ++c) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = emit_fetch(bld, insn, 1, c);
dotp = bld_insn_3(bld, NV_OP_MAD, src0, src1, dotp);
}
return dotp;
}
#define FOR_EACH_DST0_ENABLED_CHANNEL(chan, inst) \
for (chan = 0; chan < 4; ++chan) \
if ((inst)->Dst[0].Register.WriteMask & (1 << chan))
static void
bld_instruction(struct bld_context *bld,
const struct tgsi_full_instruction *insn)
{
struct nv_value *src0;
struct nv_value *src1;
struct nv_value *src2;
struct nv_value *dst0[4];
struct nv_value *temp;
int c;
uint opcode = translate_opcode(insn->Instruction.Opcode);
#ifdef NV50_TGSI2NC_DEBUG
debug_printf("bld_instruction:"); tgsi_dump_instruction(insn, 1);
#endif
switch (insn->Instruction.Opcode) {
case TGSI_OPCODE_ADD:
case TGSI_OPCODE_MAX:
case TGSI_OPCODE_MIN:
case TGSI_OPCODE_MUL:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = emit_fetch(bld, insn, 1, c);
dst0[c] = bld_insn_2(bld, opcode, src0, src1);
}
break;
case TGSI_OPCODE_ARL:
src1 = bld_imm_u32(bld, 4);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
(temp = bld_insn_1(bld, NV_OP_FLOOR, src0))->reg.type = NV_TYPE_S32;
dst0[c] = bld_insn_2(bld, NV_OP_SHL, temp, src1);
}
break;
case TGSI_OPCODE_CMP:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = emit_fetch(bld, insn, 1, c);
src2 = emit_fetch(bld, insn, 2, c);
src0 = bld_predicate(bld, src0, FALSE);
src1 = bld_insn_1(bld, NV_OP_MOV, src1);
src1->insn->flags_src = new_ref(bld->pc, src0);
src1->insn->cc = NV_CC_LT;
src2 = bld_insn_1(bld, NV_OP_MOV, src2);
src2->insn->flags_src = new_ref(bld->pc, src0);
src2->insn->cc = NV_CC_GE;
dst0[c] = bld_insn_2(bld, NV_OP_SELECT, src1, src2);
}
break;
case TGSI_OPCODE_COS:
case TGSI_OPCODE_SIN:
src0 = emit_fetch(bld, insn, 0, 0);
temp = bld_insn_1(bld, NV_OP_PRESIN, src0);
if (insn->Dst[0].Register.WriteMask & 7)
temp = bld_insn_1(bld, opcode, temp);
for (c = 0; c < 3; ++c)
if (insn->Dst[0].Register.WriteMask & (1 << c))
dst0[c] = temp;
if (!(insn->Dst[0].Register.WriteMask & (1 << 3)))
break;
src0 = emit_fetch(bld, insn, 0, 3);
temp = bld_insn_1(bld, NV_OP_PRESIN, src0);
dst0[3] = bld_insn_1(bld, opcode, temp);
break;
case TGSI_OPCODE_DP2:
temp = bld_dot(bld, insn, 2);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_DP3:
temp = bld_dot(bld, insn, 3);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_DP4:
temp = bld_dot(bld, insn, 4);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_DPH:
src0 = bld_dot(bld, insn, 3);
src1 = emit_fetch(bld, insn, 1, 3);
temp = bld_insn_2(bld, NV_OP_ADD, src0, src1);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_DST:
if (insn->Dst[0].Register.WriteMask & 1)
dst0[0] = bld_imm_f32(bld, 1.0f);
if (insn->Dst[0].Register.WriteMask & 2) {
src0 = emit_fetch(bld, insn, 0, 1);
src1 = emit_fetch(bld, insn, 1, 1);
dst0[1] = bld_insn_2(bld, NV_OP_MUL, src0, src1);
}
if (insn->Dst[0].Register.WriteMask & 4)
dst0[2] = emit_fetch(bld, insn, 0, 2);
if (insn->Dst[0].Register.WriteMask & 8)
dst0[3] = emit_fetch(bld, insn, 1, 3);
break;
case TGSI_OPCODE_EX2:
src0 = emit_fetch(bld, insn, 0, 0);
temp = bld_insn_1(bld, NV_OP_PREEX2, src0);
temp = bld_insn_1(bld, NV_OP_EX2, temp);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_FRC:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
dst0[c] = bld_insn_1(bld, NV_OP_FLOOR, src0);
dst0[c] = bld_insn_2(bld, NV_OP_SUB, src0, dst0[c]);
}
break;
case TGSI_OPCODE_KIL:
for (c = 0; c < 4; ++c) {
src0 = emit_fetch(bld, insn, 0, c);
bld_kil(bld, src0);
}
break;
case TGSI_OPCODE_KILP:
(new_instruction(bld->pc, NV_OP_KIL))->fixed = 1;
break;
case TGSI_OPCODE_IF:
{
struct nv_basic_block *b = new_basic_block(bld->pc);
nvbb_attach_block(bld->pc->current_block, b, CFG_EDGE_FORWARD);
bld->join_bb[bld->cond_lvl] = bld->pc->current_block;
bld->cond_bb[bld->cond_lvl] = bld->pc->current_block;
src1 = bld_predicate(bld, emit_fetch(bld, insn, 0, 0), TRUE);
bld_flow(bld, NV_OP_BRA, NV_CC_EQ, src1, NULL, (bld->cond_lvl == 0));
++bld->cond_lvl;
bld_new_block(bld, b);
}
break;
case TGSI_OPCODE_ELSE:
{
struct nv_basic_block *b = new_basic_block(bld->pc);
--bld->cond_lvl;
nvbb_attach_block(bld->join_bb[bld->cond_lvl], b, CFG_EDGE_FORWARD);
bld->cond_bb[bld->cond_lvl]->exit->target = b;
bld->cond_bb[bld->cond_lvl] = bld->pc->current_block;
new_instruction(bld->pc, NV_OP_BRA)->is_terminator = 1;
++bld->cond_lvl;
bld_new_block(bld, b);
}
break;
case TGSI_OPCODE_ENDIF:
{
struct nv_basic_block *b = new_basic_block(bld->pc);
--bld->cond_lvl;
nvbb_attach_block(bld->pc->current_block, b, bld->out_kind);
nvbb_attach_block(bld->cond_bb[bld->cond_lvl], b, CFG_EDGE_FORWARD);
bld->cond_bb[bld->cond_lvl]->exit->target = b;
bld_new_block(bld, b);
if (!bld->cond_lvl && bld->join_bb[bld->cond_lvl]) {
bld->join_bb[bld->cond_lvl]->exit->prev->target = b;
new_instruction(bld->pc, NV_OP_JOIN)->is_join = TRUE;
}
}
break;
case TGSI_OPCODE_BGNLOOP:
{
struct nv_basic_block *bl = new_basic_block(bld->pc);
struct nv_basic_block *bb = new_basic_block(bld->pc);
bld->loop_bb[bld->loop_lvl] = bl;
bld->brkt_bb[bld->loop_lvl] = bb;
bld_flow(bld, NV_OP_BREAKADDR, NV_CC_TR, NULL, bb, FALSE);
nvbb_attach_block(bld->pc->current_block, bl, CFG_EDGE_LOOP_ENTER);
bld_new_block(bld, bld->loop_bb[bld->loop_lvl++]);
if (bld->loop_lvl == bld->pc->loop_nesting_bound)
bld->pc->loop_nesting_bound++;
bld_clear_def_use(&bld->tvs[0][0], BLD_MAX_TEMPS, bld->loop_lvl);
bld_clear_def_use(&bld->avs[0][0], BLD_MAX_ADDRS, bld->loop_lvl);
bld_clear_def_use(&bld->pvs[0][0], BLD_MAX_PREDS, bld->loop_lvl);
}
break;
case TGSI_OPCODE_BRK:
{
struct nv_basic_block *bb = bld->brkt_bb[bld->loop_lvl - 1];
bld_flow(bld, NV_OP_BREAK, NV_CC_TR, NULL, bb, FALSE);
if (bld->out_kind == CFG_EDGE_FORWARD) /* else we already had BRK/CONT */
nvbb_attach_block(bld->pc->current_block, bb, CFG_EDGE_LOOP_LEAVE);
bld->out_kind = CFG_EDGE_FAKE;
}
break;
case TGSI_OPCODE_CONT:
{
struct nv_basic_block *bb = bld->loop_bb[bld->loop_lvl - 1];
bld_flow(bld, NV_OP_BRA, NV_CC_TR, NULL, bb, FALSE);
nvbb_attach_block(bld->pc->current_block, bb, CFG_EDGE_BACK);
if ((bb = bld->join_bb[bld->cond_lvl - 1])) {
bld->join_bb[bld->cond_lvl - 1] = NULL;
nv_nvi_delete(bb->exit->prev);
}
bld->out_kind = CFG_EDGE_FAKE;
}
break;
case TGSI_OPCODE_ENDLOOP:
{
struct nv_basic_block *bb = bld->loop_bb[bld->loop_lvl - 1];
bld_flow(bld, NV_OP_BRA, NV_CC_TR, NULL, bb, FALSE);
nvbb_attach_block(bld->pc->current_block, bb, CFG_EDGE_BACK);
bld_loop_end(bld, bb); /* replace loop-side operand of the phis */
bld_new_block(bld, bld->brkt_bb[--bld->loop_lvl]);
}
break;
case TGSI_OPCODE_ABS:
case TGSI_OPCODE_CEIL:
case TGSI_OPCODE_FLR:
case TGSI_OPCODE_TRUNC:
case TGSI_OPCODE_DDX:
case TGSI_OPCODE_DDY:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
dst0[c] = bld_insn_1(bld, opcode, src0);
}
break;
case TGSI_OPCODE_LIT:
bld_lit(bld, dst0, insn);
break;
case TGSI_OPCODE_LRP:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = emit_fetch(bld, insn, 1, c);
src2 = emit_fetch(bld, insn, 2, c);
dst0[c] = bld_insn_2(bld, NV_OP_SUB, src1, src2);
dst0[c] = bld_insn_3(bld, NV_OP_MAD, dst0[c], src0, src2);
}
break;
case TGSI_OPCODE_MOV:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = emit_fetch(bld, insn, 0, c);
break;
case TGSI_OPCODE_MAD:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = emit_fetch(bld, insn, 1, c);
src2 = emit_fetch(bld, insn, 2, c);
dst0[c] = bld_insn_3(bld, opcode, src0, src1, src2);
}
break;
case TGSI_OPCODE_POW:
src0 = emit_fetch(bld, insn, 0, 0);
src1 = emit_fetch(bld, insn, 1, 0);
temp = bld_pow(bld, src0, src1);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_RCP:
case TGSI_OPCODE_LG2:
src0 = emit_fetch(bld, insn, 0, 0);
temp = bld_insn_1(bld, opcode, src0);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_RSQ:
src0 = emit_fetch(bld, insn, 0, 0);
temp = bld_insn_1(bld, NV_OP_ABS, src0);
temp = bld_insn_1(bld, NV_OP_RSQ, temp);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_SLT:
case TGSI_OPCODE_SGE:
case TGSI_OPCODE_SEQ:
case TGSI_OPCODE_SGT:
case TGSI_OPCODE_SLE:
case TGSI_OPCODE_SNE:
case TGSI_OPCODE_ISLT:
case TGSI_OPCODE_ISGE:
case TGSI_OPCODE_USEQ:
case TGSI_OPCODE_USGE:
case TGSI_OPCODE_USLT:
case TGSI_OPCODE_USNE:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = emit_fetch(bld, insn, 1, c);
dst0[c] = bld_insn_2(bld, NV_OP_SET, src0, src1);
dst0[c]->insn->set_cond = translate_setcc(insn->Instruction.Opcode);
dst0[c]->reg.type = infer_dst_type(insn->Instruction.Opcode);
dst0[c]->insn->src[0]->typecast =
dst0[c]->insn->src[1]->typecast =
infer_src_type(insn->Instruction.Opcode);
if (dst0[c]->reg.type != NV_TYPE_F32)
break;
dst0[c] = bld_insn_1(bld, NV_OP_ABS, dst0[c]);
dst0[c]->insn->src[0]->typecast = NV_TYPE_S32;
dst0[c]->reg.type = NV_TYPE_S32;
dst0[c] = bld_insn_1(bld, NV_OP_CVT, dst0[c]);
dst0[c]->reg.type = NV_TYPE_F32;
}
break;
case TGSI_OPCODE_SCS:
if (insn->Dst[0].Register.WriteMask & 0x3) {
src0 = emit_fetch(bld, insn, 0, 0);
temp = bld_insn_1(bld, NV_OP_PRESIN, src0);
if (insn->Dst[0].Register.WriteMask & 0x1)
dst0[0] = bld_insn_1(bld, NV_OP_COS, temp);
if (insn->Dst[0].Register.WriteMask & 0x2)
dst0[1] = bld_insn_1(bld, NV_OP_SIN, temp);
}
if (insn->Dst[0].Register.WriteMask & 0x4)
dst0[2] = bld_imm_f32(bld, 0.0f);
if (insn->Dst[0].Register.WriteMask & 0x8)
dst0[3] = bld_imm_f32(bld, 1.0f);
break;
case TGSI_OPCODE_SSG:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = bld_predicate(bld, src0, FALSE);
temp = bld_insn_2(bld, NV_OP_AND, src0, bld_imm_u32(bld, 0x80000000));
temp = bld_insn_2(bld, NV_OP_OR, temp, bld_imm_f32(bld, 1.0f));
dst0[c] = bld_insn_2(bld, NV_OP_XOR, temp, temp);
dst0[c]->insn->cc = NV_CC_EQ;
nv_reference(bld->pc, &dst0[c]->insn->flags_src, src1);
}
break;
case TGSI_OPCODE_SUB:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
src1 = emit_fetch(bld, insn, 1, c);
dst0[c] = bld_insn_2(bld, NV_OP_ADD, src0, src1);
dst0[c]->insn->src[1]->mod ^= NV_MOD_NEG;
}
break;
case TGSI_OPCODE_TEX:
case TGSI_OPCODE_TXB:
case TGSI_OPCODE_TXL:
case TGSI_OPCODE_TXP:
bld_tex(bld, dst0, insn);
break;
case TGSI_OPCODE_XPD:
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
if (c == 3) {
dst0[3] = bld_imm_f32(bld, 1.0f);
break;
}
src0 = emit_fetch(bld, insn, 1, (c + 1) % 3);
src1 = emit_fetch(bld, insn, 0, (c + 2) % 3);
dst0[c] = bld_insn_2(bld, NV_OP_MUL, src0, src1);
src0 = emit_fetch(bld, insn, 0, (c + 1) % 3);
src1 = emit_fetch(bld, insn, 1, (c + 2) % 3);
dst0[c] = bld_insn_3(bld, NV_OP_MAD, src0, src1, dst0[c]);
dst0[c]->insn->src[2]->mod ^= NV_MOD_NEG;
}
break;
case TGSI_OPCODE_RET:
(new_instruction(bld->pc, NV_OP_RET))->fixed = 1;
break;
case TGSI_OPCODE_END:
if (bld->ti->p->type == PIPE_SHADER_FRAGMENT)
bld_export_outputs(bld);
break;
default:
NOUVEAU_ERR("unhandled opcode %u\n", insn->Instruction.Opcode);
abort();
break;
}
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
emit_store(bld, insn, c, dst0[c]);
}
static INLINE void
bld_free_value_trackers(struct bld_value_stack *base, int n)
{
int i, c;
for (i = 0; i < n; ++i)
for (c = 0; c < 4; ++c)
if (base[i * 4 + c].body)
FREE(base[i * 4 + c].body);
}
int
nv50_tgsi_to_nc(struct nv_pc *pc, struct nv50_translation_info *ti)
{
struct bld_context *bld = CALLOC_STRUCT(bld_context);
int c;
unsigned ip;
pc->root[0] = pc->current_block = new_basic_block(pc);
bld->pc = pc;
bld->ti = ti;
pc->loop_nesting_bound = 1;
c = util_bitcount(bld->ti->p->fp.interp >> 24);
if (c && ti->p->type == PIPE_SHADER_FRAGMENT) {
bld->frgcrd[3] = new_value(pc, NV_FILE_MEM_V, NV_TYPE_F32);
bld->frgcrd[3]->reg.id = c - 1;
bld->frgcrd[3] = bld_insn_1(bld, NV_OP_LINTERP, bld->frgcrd[3]);
bld->frgcrd[3] = bld_insn_1(bld, NV_OP_RCP, bld->frgcrd[3]);
}
for (ip = 0; ip < ti->inst_nr; ++ip)
bld_instruction(bld, &ti->insns[ip]);
bld_free_value_trackers(&bld->tvs[0][0], BLD_MAX_TEMPS);
bld_free_value_trackers(&bld->avs[0][0], BLD_MAX_ADDRS);
bld_free_value_trackers(&bld->pvs[0][0], BLD_MAX_PREDS);
bld_free_value_trackers(&bld->ovs[0][0], PIPE_MAX_SHADER_OUTPUTS);
FREE(bld);
return 0;
}
/* If a variable is assigned in a loop, replace all references to the value
* from outside the loop with a phi value.
*/
static void
bld_replace_value(struct nv_pc *pc, struct nv_basic_block *b,
struct nv_value *old_val,
struct nv_value *new_val)
{
struct nv_instruction *nvi;
for (nvi = b->phi ? b->phi : b->entry; nvi; nvi = nvi->next) {
int s;
for (s = 0; s < 5; ++s) {
if (!nvi->src[s])
continue;
if (nvi->src[s]->value == old_val)
nv_reference(pc, &nvi->src[s], new_val);
}
if (nvi->flags_src && nvi->flags_src->value == old_val)
nv_reference(pc, &nvi->flags_src, new_val);
}
b->pass_seq = pc->pass_seq;
if (b->out[0] && b->out[0]->pass_seq < pc->pass_seq)
bld_replace_value(pc, b->out[0], old_val, new_val);
if (b->out[1] && b->out[1]->pass_seq < pc->pass_seq)
bld_replace_value(pc, b->out[1], old_val, new_val);
}
|