1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
|
/*
* Copyright © 2015 Intel Corporation
*
* 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.
*
* Authors:
* Jason Ekstrand (jason@jlekstrand.net)
*
*/
#include "vtn_private.h"
#include "spirv_info.h"
#include "nir_deref.h"
#include <vulkan/vulkan_core.h>
static struct vtn_access_chain *
vtn_access_chain_create(struct vtn_builder *b, unsigned length)
{
struct vtn_access_chain *chain;
/* Subtract 1 from the length since there's already one built in */
size_t size = sizeof(*chain) +
(MAX2(length, 1) - 1) * sizeof(chain->link[0]);
chain = rzalloc_size(b, size);
chain->length = length;
return chain;
}
bool
vtn_pointer_uses_ssa_offset(struct vtn_builder *b,
struct vtn_pointer *ptr)
{
return ((ptr->mode == vtn_variable_mode_ubo ||
ptr->mode == vtn_variable_mode_ssbo) &&
b->options->lower_ubo_ssbo_access_to_offsets) ||
ptr->mode == vtn_variable_mode_push_constant ||
(ptr->mode == vtn_variable_mode_workgroup &&
b->options->lower_workgroup_access_to_offsets);
}
static bool
vtn_pointer_is_external_block(struct vtn_builder *b,
struct vtn_pointer *ptr)
{
return ptr->mode == vtn_variable_mode_ssbo ||
ptr->mode == vtn_variable_mode_ubo ||
ptr->mode == vtn_variable_mode_push_constant ||
(ptr->mode == vtn_variable_mode_workgroup &&
b->options->lower_workgroup_access_to_offsets);
}
static nir_ssa_def *
vtn_access_link_as_ssa(struct vtn_builder *b, struct vtn_access_link link,
unsigned stride, unsigned bit_size)
{
vtn_assert(stride > 0);
if (link.mode == vtn_access_mode_literal) {
return nir_imm_intN_t(&b->nb, link.id * stride, bit_size);
} else {
nir_ssa_def *ssa = vtn_ssa_value(b, link.id)->def;
if (ssa->bit_size != bit_size)
ssa = nir_i2i(&b->nb, ssa, bit_size);
if (stride != 1)
ssa = nir_imul_imm(&b->nb, ssa, stride);
return ssa;
}
}
static VkDescriptorType
vk_desc_type_for_mode(struct vtn_builder *b, enum vtn_variable_mode mode)
{
switch (mode) {
case vtn_variable_mode_ubo:
return VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
case vtn_variable_mode_ssbo:
return VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
default:
vtn_fail("Invalid mode for vulkan_resource_index");
}
}
static nir_ssa_def *
vtn_variable_resource_index(struct vtn_builder *b, struct vtn_variable *var,
nir_ssa_def *desc_array_index)
{
if (!desc_array_index) {
vtn_assert(glsl_type_is_struct(var->type->type));
desc_array_index = nir_imm_int(&b->nb, 0);
}
nir_intrinsic_instr *instr =
nir_intrinsic_instr_create(b->nb.shader,
nir_intrinsic_vulkan_resource_index);
instr->src[0] = nir_src_for_ssa(desc_array_index);
nir_intrinsic_set_desc_set(instr, var->descriptor_set);
nir_intrinsic_set_binding(instr, var->binding);
nir_intrinsic_set_desc_type(instr, vk_desc_type_for_mode(b, var->mode));
nir_ssa_dest_init(&instr->instr, &instr->dest, 1, 32, NULL);
nir_builder_instr_insert(&b->nb, &instr->instr);
return &instr->dest.ssa;
}
static nir_ssa_def *
vtn_resource_reindex(struct vtn_builder *b, enum vtn_variable_mode mode,
nir_ssa_def *base_index, nir_ssa_def *offset_index)
{
nir_intrinsic_instr *instr =
nir_intrinsic_instr_create(b->nb.shader,
nir_intrinsic_vulkan_resource_reindex);
instr->src[0] = nir_src_for_ssa(base_index);
instr->src[1] = nir_src_for_ssa(offset_index);
nir_intrinsic_set_desc_type(instr, vk_desc_type_for_mode(b, mode));
nir_ssa_dest_init(&instr->instr, &instr->dest, 1, 32, NULL);
nir_builder_instr_insert(&b->nb, &instr->instr);
return &instr->dest.ssa;
}
static nir_ssa_def *
vtn_descriptor_load(struct vtn_builder *b, enum vtn_variable_mode mode,
const struct glsl_type *desc_type, nir_ssa_def *desc_index)
{
nir_intrinsic_instr *desc_load =
nir_intrinsic_instr_create(b->nb.shader,
nir_intrinsic_load_vulkan_descriptor);
desc_load->src[0] = nir_src_for_ssa(desc_index);
desc_load->num_components = glsl_get_vector_elements(desc_type);
nir_intrinsic_set_desc_type(desc_load, vk_desc_type_for_mode(b, mode));
nir_ssa_dest_init(&desc_load->instr, &desc_load->dest,
desc_load->num_components,
glsl_get_bit_size(desc_type), NULL);
nir_builder_instr_insert(&b->nb, &desc_load->instr);
return &desc_load->dest.ssa;
}
/* Dereference the given base pointer by the access chain */
static struct vtn_pointer *
vtn_nir_deref_pointer_dereference(struct vtn_builder *b,
struct vtn_pointer *base,
struct vtn_access_chain *deref_chain)
{
struct vtn_type *type = base->type;
enum gl_access_qualifier access = base->access;
unsigned idx = 0;
nir_deref_instr *tail;
if (base->deref) {
tail = base->deref;
} else if (vtn_pointer_is_external_block(b, base)) {
nir_ssa_def *block_index = base->block_index;
/* We dereferencing an external block pointer. Correctness of this
* operation relies on one particular line in the SPIR-V spec, section
* entitled "Validation Rules for Shader Capabilities":
*
* "Block and BufferBlock decorations cannot decorate a structure
* type that is nested at any level inside another structure type
* decorated with Block or BufferBlock."
*
* This means that we can detect the point where we cross over from
* descriptor indexing to buffer indexing by looking for the block
* decorated struct type. Anything before the block decorated struct
* type is a descriptor indexing operation and anything after the block
* decorated struct is a buffer offset operation.
*/
/* Figure out the descriptor array index if any
*
* Some of the Vulkan CTS tests with hand-rolled SPIR-V have been known
* to forget the Block or BufferBlock decoration from time to time.
* It's more robust if we check for both !block_index and for the type
* to contain a block. This way there's a decent chance that arrays of
* UBOs/SSBOs will work correctly even if variable pointers are
* completley toast.
*/
nir_ssa_def *desc_arr_idx = NULL;
if (!block_index || vtn_type_contains_block(b, type)) {
/* If our type contains a block, then we're still outside the block
* and we need to process enough levels of dereferences to get inside
* of it.
*/
if (deref_chain->ptr_as_array) {
unsigned aoa_size = glsl_get_aoa_size(type->type);
desc_arr_idx = vtn_access_link_as_ssa(b, deref_chain->link[idx],
MAX2(aoa_size, 1), 32);
idx++;
}
for (; idx < deref_chain->length; idx++) {
if (type->base_type != vtn_base_type_array) {
vtn_assert(type->base_type == vtn_base_type_struct);
break;
}
unsigned aoa_size = glsl_get_aoa_size(type->array_element->type);
nir_ssa_def *arr_offset =
vtn_access_link_as_ssa(b, deref_chain->link[idx],
MAX2(aoa_size, 1), 32);
if (desc_arr_idx)
desc_arr_idx = nir_iadd(&b->nb, desc_arr_idx, arr_offset);
else
desc_arr_idx = arr_offset;
type = type->array_element;
access |= type->access;
}
}
if (!block_index) {
vtn_assert(base->var && base->type);
block_index = vtn_variable_resource_index(b, base->var, desc_arr_idx);
} else if (desc_arr_idx) {
block_index = vtn_resource_reindex(b, base->mode,
block_index, desc_arr_idx);
}
if (idx == deref_chain->length) {
/* The entire deref was consumed in finding the block index. Return
* a pointer which just has a block index and a later access chain
* will dereference deeper.
*/
struct vtn_pointer *ptr = rzalloc(b, struct vtn_pointer);
ptr->mode = base->mode;
ptr->type = type;
ptr->block_index = block_index;
ptr->access = access;
return ptr;
}
/* If we got here, there's more access chain to handle and we have the
* final block index. Insert a descriptor load and cast to a deref to
* start the deref chain.
*/
nir_ssa_def *desc =
vtn_descriptor_load(b, base->mode, base->ptr_type->type, block_index);
assert(base->mode == vtn_variable_mode_ssbo ||
base->mode == vtn_variable_mode_ubo);
nir_variable_mode nir_mode =
base->mode == vtn_variable_mode_ssbo ? nir_var_mem_ssbo : nir_var_mem_ubo;
tail = nir_build_deref_cast(&b->nb, desc, nir_mode, type->type,
base->ptr_type->stride);
} else {
assert(base->var && base->var->var);
tail = nir_build_deref_var(&b->nb, base->var->var);
if (base->ptr_type && base->ptr_type->type) {
tail->dest.ssa.num_components =
glsl_get_vector_elements(base->ptr_type->type);
tail->dest.ssa.bit_size = glsl_get_bit_size(base->ptr_type->type);
}
}
if (idx == 0 && deref_chain->ptr_as_array) {
/* We start with a deref cast to get the stride. Hopefully, we'll be
* able to delete that cast eventually.
*/
tail = nir_build_deref_cast(&b->nb, &tail->dest.ssa, tail->mode,
tail->type, base->ptr_type->stride);
nir_ssa_def *index = vtn_access_link_as_ssa(b, deref_chain->link[0], 1,
tail->dest.ssa.bit_size);
tail = nir_build_deref_ptr_as_array(&b->nb, tail, index);
idx++;
}
for (; idx < deref_chain->length; idx++) {
if (glsl_type_is_struct(type->type)) {
vtn_assert(deref_chain->link[idx].mode == vtn_access_mode_literal);
unsigned field = deref_chain->link[idx].id;
tail = nir_build_deref_struct(&b->nb, tail, field);
type = type->members[field];
} else {
nir_ssa_def *arr_index =
vtn_access_link_as_ssa(b, deref_chain->link[idx], 1,
tail->dest.ssa.bit_size);
tail = nir_build_deref_array(&b->nb, tail, arr_index);
type = type->array_element;
}
access |= type->access;
}
struct vtn_pointer *ptr = rzalloc(b, struct vtn_pointer);
ptr->mode = base->mode;
ptr->type = type;
ptr->var = base->var;
ptr->deref = tail;
ptr->access = access;
return ptr;
}
static struct vtn_pointer *
vtn_ssa_offset_pointer_dereference(struct vtn_builder *b,
struct vtn_pointer *base,
struct vtn_access_chain *deref_chain)
{
nir_ssa_def *block_index = base->block_index;
nir_ssa_def *offset = base->offset;
struct vtn_type *type = base->type;
enum gl_access_qualifier access = base->access;
unsigned idx = 0;
if (base->mode == vtn_variable_mode_ubo ||
base->mode == vtn_variable_mode_ssbo) {
if (!block_index) {
vtn_assert(base->var && base->type);
nir_ssa_def *desc_arr_idx;
if (glsl_type_is_array(type->type)) {
if (deref_chain->length >= 1) {
desc_arr_idx =
vtn_access_link_as_ssa(b, deref_chain->link[0], 1, 32);
idx++;
/* This consumes a level of type */
type = type->array_element;
access |= type->access;
} else {
/* This is annoying. We've been asked for a pointer to the
* array of UBOs/SSBOs and not a specifc buffer. Return a
* pointer with a descriptor index of 0 and we'll have to do
* a reindex later to adjust it to the right thing.
*/
desc_arr_idx = nir_imm_int(&b->nb, 0);
}
} else if (deref_chain->ptr_as_array) {
/* You can't have a zero-length OpPtrAccessChain */
vtn_assert(deref_chain->length >= 1);
desc_arr_idx = vtn_access_link_as_ssa(b, deref_chain->link[0], 1, 32);
} else {
/* We have a regular non-array SSBO. */
desc_arr_idx = NULL;
}
block_index = vtn_variable_resource_index(b, base->var, desc_arr_idx);
} else if (deref_chain->ptr_as_array &&
type->base_type == vtn_base_type_struct && type->block) {
/* We are doing an OpPtrAccessChain on a pointer to a struct that is
* decorated block. This is an interesting corner in the SPIR-V
* spec. One interpretation would be that they client is clearly
* trying to treat that block as if it's an implicit array of blocks
* repeated in the buffer. However, the SPIR-V spec for the
* OpPtrAccessChain says:
*
* "Base is treated as the address of the first element of an
* array, and the Element element’s address is computed to be the
* base for the Indexes, as per OpAccessChain."
*
* Taken literally, that would mean that your struct type is supposed
* to be treated as an array of such a struct and, since it's
* decorated block, that means an array of blocks which corresponds
* to an array descriptor. Therefore, we need to do a reindex
* operation to add the index from the first link in the access chain
* to the index we recieved.
*
* The downside to this interpretation (there always is one) is that
* this might be somewhat surprising behavior to apps if they expect
* the implicit array behavior described above.
*/
vtn_assert(deref_chain->length >= 1);
nir_ssa_def *offset_index =
vtn_access_link_as_ssa(b, deref_chain->link[0], 1, 32);
idx++;
block_index = vtn_resource_reindex(b, base->mode,
block_index, offset_index);
}
}
if (!offset) {
if (base->mode == vtn_variable_mode_workgroup) {
/* SLM doesn't need nor have a block index */
vtn_assert(!block_index);
/* We need the variable for the base offset */
vtn_assert(base->var);
/* We need ptr_type for size and alignment */
vtn_assert(base->ptr_type);
/* Assign location on first use so that we don't end up bloating SLM
* address space for variables which are never statically used.
*/
if (base->var->shared_location < 0) {
vtn_assert(base->ptr_type->length > 0 && base->ptr_type->align > 0);
b->shader->num_shared = vtn_align_u32(b->shader->num_shared,
base->ptr_type->align);
base->var->shared_location = b->shader->num_shared;
b->shader->num_shared += base->ptr_type->length;
}
offset = nir_imm_int(&b->nb, base->var->shared_location);
} else if (base->mode == vtn_variable_mode_push_constant) {
/* Push constants neither need nor have a block index */
vtn_assert(!block_index);
/* Start off with at the start of the push constant block. */
offset = nir_imm_int(&b->nb, 0);
} else {
/* The code above should have ensured a block_index when needed. */
vtn_assert(block_index);
/* Start off with at the start of the buffer. */
offset = nir_imm_int(&b->nb, 0);
}
}
if (deref_chain->ptr_as_array && idx == 0) {
/* We need ptr_type for the stride */
vtn_assert(base->ptr_type);
/* We need at least one element in the chain */
vtn_assert(deref_chain->length >= 1);
nir_ssa_def *elem_offset =
vtn_access_link_as_ssa(b, deref_chain->link[idx],
base->ptr_type->stride, offset->bit_size);
offset = nir_iadd(&b->nb, offset, elem_offset);
idx++;
}
for (; idx < deref_chain->length; idx++) {
switch (glsl_get_base_type(type->type)) {
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
case GLSL_TYPE_UINT16:
case GLSL_TYPE_INT16:
case GLSL_TYPE_UINT8:
case GLSL_TYPE_INT8:
case GLSL_TYPE_UINT64:
case GLSL_TYPE_INT64:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_FLOAT16:
case GLSL_TYPE_DOUBLE:
case GLSL_TYPE_BOOL:
case GLSL_TYPE_ARRAY: {
nir_ssa_def *elem_offset =
vtn_access_link_as_ssa(b, deref_chain->link[idx],
type->stride, offset->bit_size);
offset = nir_iadd(&b->nb, offset, elem_offset);
type = type->array_element;
access |= type->access;
break;
}
case GLSL_TYPE_STRUCT: {
vtn_assert(deref_chain->link[idx].mode == vtn_access_mode_literal);
unsigned member = deref_chain->link[idx].id;
offset = nir_iadd_imm(&b->nb, offset, type->offsets[member]);
type = type->members[member];
access |= type->access;
break;
}
default:
vtn_fail("Invalid type for deref");
}
}
struct vtn_pointer *ptr = rzalloc(b, struct vtn_pointer);
ptr->mode = base->mode;
ptr->type = type;
ptr->block_index = block_index;
ptr->offset = offset;
ptr->access = access;
return ptr;
}
/* Dereference the given base pointer by the access chain */
static struct vtn_pointer *
vtn_pointer_dereference(struct vtn_builder *b,
struct vtn_pointer *base,
struct vtn_access_chain *deref_chain)
{
if (vtn_pointer_uses_ssa_offset(b, base)) {
return vtn_ssa_offset_pointer_dereference(b, base, deref_chain);
} else {
return vtn_nir_deref_pointer_dereference(b, base, deref_chain);
}
}
struct vtn_pointer *
vtn_pointer_for_variable(struct vtn_builder *b,
struct vtn_variable *var, struct vtn_type *ptr_type)
{
struct vtn_pointer *pointer = rzalloc(b, struct vtn_pointer);
pointer->mode = var->mode;
pointer->type = var->type;
vtn_assert(ptr_type->base_type == vtn_base_type_pointer);
vtn_assert(ptr_type->deref->type == var->type->type);
pointer->ptr_type = ptr_type;
pointer->var = var;
pointer->access = var->access | var->type->access;
return pointer;
}
/* Returns an atomic_uint type based on the original uint type. The returned
* type will be equivalent to the original one but will have an atomic_uint
* type as leaf instead of an uint.
*
* Manages uint scalars, arrays, and arrays of arrays of any nested depth.
*/
static const struct glsl_type *
repair_atomic_type(const struct glsl_type *type)
{
assert(glsl_get_base_type(glsl_without_array(type)) == GLSL_TYPE_UINT);
assert(glsl_type_is_scalar(glsl_without_array(type)));
if (glsl_type_is_array(type)) {
const struct glsl_type *atomic =
repair_atomic_type(glsl_get_array_element(type));
return glsl_array_type(atomic, glsl_get_length(type),
glsl_get_explicit_stride(type));
} else {
return glsl_atomic_uint_type();
}
}
nir_deref_instr *
vtn_pointer_to_deref(struct vtn_builder *b, struct vtn_pointer *ptr)
{
if (b->wa_glslang_179) {
/* Do on-the-fly copy propagation for samplers. */
if (ptr->var && ptr->var->copy_prop_sampler)
return vtn_pointer_to_deref(b, ptr->var->copy_prop_sampler);
}
vtn_assert(!vtn_pointer_uses_ssa_offset(b, ptr));
if (!ptr->deref) {
struct vtn_access_chain chain = {
.length = 0,
};
ptr = vtn_nir_deref_pointer_dereference(b, ptr, &chain);
}
return ptr->deref;
}
static void
_vtn_local_load_store(struct vtn_builder *b, bool load, nir_deref_instr *deref,
struct vtn_ssa_value *inout)
{
if (glsl_type_is_vector_or_scalar(deref->type)) {
if (load) {
inout->def = nir_load_deref(&b->nb, deref);
} else {
nir_store_deref(&b->nb, deref, inout->def, ~0);
}
} else if (glsl_type_is_array(deref->type) ||
glsl_type_is_matrix(deref->type)) {
unsigned elems = glsl_get_length(deref->type);
for (unsigned i = 0; i < elems; i++) {
nir_deref_instr *child =
nir_build_deref_array(&b->nb, deref, nir_imm_int(&b->nb, i));
_vtn_local_load_store(b, load, child, inout->elems[i]);
}
} else {
vtn_assert(glsl_type_is_struct(deref->type));
unsigned elems = glsl_get_length(deref->type);
for (unsigned i = 0; i < elems; i++) {
nir_deref_instr *child = nir_build_deref_struct(&b->nb, deref, i);
_vtn_local_load_store(b, load, child, inout->elems[i]);
}
}
}
nir_deref_instr *
vtn_nir_deref(struct vtn_builder *b, uint32_t id)
{
struct vtn_pointer *ptr = vtn_value(b, id, vtn_value_type_pointer)->pointer;
return vtn_pointer_to_deref(b, ptr);
}
/*
* Gets the NIR-level deref tail, which may have as a child an array deref
* selecting which component due to OpAccessChain supporting per-component
* indexing in SPIR-V.
*/
static nir_deref_instr *
get_deref_tail(nir_deref_instr *deref)
{
if (deref->deref_type != nir_deref_type_array)
return deref;
nir_deref_instr *parent =
nir_instr_as_deref(deref->parent.ssa->parent_instr);
if (glsl_type_is_vector(parent->type))
return parent;
else
return deref;
}
struct vtn_ssa_value *
vtn_local_load(struct vtn_builder *b, nir_deref_instr *src)
{
nir_deref_instr *src_tail = get_deref_tail(src);
struct vtn_ssa_value *val = vtn_create_ssa_value(b, src_tail->type);
_vtn_local_load_store(b, true, src_tail, val);
if (src_tail != src) {
val->type = src->type;
if (nir_src_is_const(src->arr.index))
val->def = vtn_vector_extract(b, val->def,
nir_src_as_uint(src->arr.index));
else
val->def = vtn_vector_extract_dynamic(b, val->def, src->arr.index.ssa);
}
return val;
}
void
vtn_local_store(struct vtn_builder *b, struct vtn_ssa_value *src,
nir_deref_instr *dest)
{
nir_deref_instr *dest_tail = get_deref_tail(dest);
if (dest_tail != dest) {
struct vtn_ssa_value *val = vtn_create_ssa_value(b, dest_tail->type);
_vtn_local_load_store(b, true, dest_tail, val);
if (nir_src_is_const(dest->arr.index))
val->def = vtn_vector_insert(b, val->def, src->def,
nir_src_as_uint(dest->arr.index));
else
val->def = vtn_vector_insert_dynamic(b, val->def, src->def,
dest->arr.index.ssa);
_vtn_local_load_store(b, false, dest_tail, val);
} else {
_vtn_local_load_store(b, false, dest_tail, src);
}
}
nir_ssa_def *
vtn_pointer_to_offset(struct vtn_builder *b, struct vtn_pointer *ptr,
nir_ssa_def **index_out)
{
assert(vtn_pointer_uses_ssa_offset(b, ptr));
if (!ptr->offset) {
struct vtn_access_chain chain = {
.length = 0,
};
ptr = vtn_ssa_offset_pointer_dereference(b, ptr, &chain);
}
*index_out = ptr->block_index;
return ptr->offset;
}
/* Tries to compute the size of an interface block based on the strides and
* offsets that are provided to us in the SPIR-V source.
*/
static unsigned
vtn_type_block_size(struct vtn_builder *b, struct vtn_type *type)
{
enum glsl_base_type base_type = glsl_get_base_type(type->type);
switch (base_type) {
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
case GLSL_TYPE_UINT16:
case GLSL_TYPE_INT16:
case GLSL_TYPE_UINT8:
case GLSL_TYPE_INT8:
case GLSL_TYPE_UINT64:
case GLSL_TYPE_INT64:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_FLOAT16:
case GLSL_TYPE_BOOL:
case GLSL_TYPE_DOUBLE: {
unsigned cols = type->row_major ? glsl_get_vector_elements(type->type) :
glsl_get_matrix_columns(type->type);
if (cols > 1) {
vtn_assert(type->stride > 0);
return type->stride * cols;
} else {
unsigned type_size = glsl_get_bit_size(type->type) / 8;
return glsl_get_vector_elements(type->type) * type_size;
}
}
case GLSL_TYPE_STRUCT:
case GLSL_TYPE_INTERFACE: {
unsigned size = 0;
unsigned num_fields = glsl_get_length(type->type);
for (unsigned f = 0; f < num_fields; f++) {
unsigned field_end = type->offsets[f] +
vtn_type_block_size(b, type->members[f]);
size = MAX2(size, field_end);
}
return size;
}
case GLSL_TYPE_ARRAY:
vtn_assert(type->stride > 0);
vtn_assert(glsl_get_length(type->type) > 0);
return type->stride * glsl_get_length(type->type);
default:
vtn_fail("Invalid block type");
return 0;
}
}
static void
_vtn_load_store_tail(struct vtn_builder *b, nir_intrinsic_op op, bool load,
nir_ssa_def *index, nir_ssa_def *offset,
unsigned access_offset, unsigned access_size,
struct vtn_ssa_value **inout, const struct glsl_type *type,
enum gl_access_qualifier access)
{
nir_intrinsic_instr *instr = nir_intrinsic_instr_create(b->nb.shader, op);
instr->num_components = glsl_get_vector_elements(type);
/* Booleans usually shouldn't show up in external memory in SPIR-V.
* However, they do for certain older GLSLang versions and can for shared
* memory when we lower access chains internally.
*/
const unsigned data_bit_size = glsl_type_is_boolean(type) ? 32 :
glsl_get_bit_size(type);
int src = 0;
if (!load) {
nir_intrinsic_set_write_mask(instr, (1 << instr->num_components) - 1);
instr->src[src++] = nir_src_for_ssa((*inout)->def);
}
if (op == nir_intrinsic_load_push_constant) {
nir_intrinsic_set_base(instr, access_offset);
nir_intrinsic_set_range(instr, access_size);
}
if (op == nir_intrinsic_load_ssbo ||
op == nir_intrinsic_store_ssbo) {
nir_intrinsic_set_access(instr, access);
}
/* With extensions like relaxed_block_layout, we really can't guarantee
* much more than scalar alignment.
*/
if (op != nir_intrinsic_load_push_constant)
nir_intrinsic_set_align(instr, data_bit_size / 8, 0);
if (index)
instr->src[src++] = nir_src_for_ssa(index);
if (op == nir_intrinsic_load_push_constant) {
/* We need to subtract the offset from where the intrinsic will load the
* data. */
instr->src[src++] =
nir_src_for_ssa(nir_isub(&b->nb, offset,
nir_imm_int(&b->nb, access_offset)));
} else {
instr->src[src++] = nir_src_for_ssa(offset);
}
if (load) {
nir_ssa_dest_init(&instr->instr, &instr->dest,
instr->num_components, data_bit_size, NULL);
(*inout)->def = &instr->dest.ssa;
}
nir_builder_instr_insert(&b->nb, &instr->instr);
if (load && glsl_get_base_type(type) == GLSL_TYPE_BOOL)
(*inout)->def = nir_ine(&b->nb, (*inout)->def, nir_imm_int(&b->nb, 0));
}
static void
_vtn_block_load_store(struct vtn_builder *b, nir_intrinsic_op op, bool load,
nir_ssa_def *index, nir_ssa_def *offset,
unsigned access_offset, unsigned access_size,
struct vtn_type *type, enum gl_access_qualifier access,
struct vtn_ssa_value **inout)
{
if (load && *inout == NULL)
*inout = vtn_create_ssa_value(b, type->type);
enum glsl_base_type base_type = glsl_get_base_type(type->type);
switch (base_type) {
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
case GLSL_TYPE_UINT16:
case GLSL_TYPE_INT16:
case GLSL_TYPE_UINT8:
case GLSL_TYPE_INT8:
case GLSL_TYPE_UINT64:
case GLSL_TYPE_INT64:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_FLOAT16:
case GLSL_TYPE_DOUBLE:
case GLSL_TYPE_BOOL:
/* This is where things get interesting. At this point, we've hit
* a vector, a scalar, or a matrix.
*/
if (glsl_type_is_matrix(type->type)) {
/* Loading the whole matrix */
struct vtn_ssa_value *transpose;
unsigned num_ops, vec_width, col_stride;
if (type->row_major) {
num_ops = glsl_get_vector_elements(type->type);
vec_width = glsl_get_matrix_columns(type->type);
col_stride = type->array_element->stride;
if (load) {
const struct glsl_type *transpose_type =
glsl_matrix_type(base_type, vec_width, num_ops);
*inout = vtn_create_ssa_value(b, transpose_type);
} else {
transpose = vtn_ssa_transpose(b, *inout);
inout = &transpose;
}
} else {
num_ops = glsl_get_matrix_columns(type->type);
vec_width = glsl_get_vector_elements(type->type);
col_stride = type->stride;
}
for (unsigned i = 0; i < num_ops; i++) {
nir_ssa_def *elem_offset =
nir_iadd_imm(&b->nb, offset, i * col_stride);
_vtn_load_store_tail(b, op, load, index, elem_offset,
access_offset, access_size,
&(*inout)->elems[i],
glsl_vector_type(base_type, vec_width),
type->access | access);
}
if (load && type->row_major)
*inout = vtn_ssa_transpose(b, *inout);
} else {
unsigned elems = glsl_get_vector_elements(type->type);
unsigned type_size = glsl_get_bit_size(type->type) / 8;
if (elems == 1 || type->stride == type_size) {
/* This is a tightly-packed normal scalar or vector load */
vtn_assert(glsl_type_is_vector_or_scalar(type->type));
_vtn_load_store_tail(b, op, load, index, offset,
access_offset, access_size,
inout, type->type,
type->access | access);
} else {
/* This is a strided load. We have to load N things separately.
* This is the single column of a row-major matrix case.
*/
vtn_assert(type->stride > type_size);
vtn_assert(type->stride % type_size == 0);
nir_ssa_def *per_comp[4];
for (unsigned i = 0; i < elems; i++) {
nir_ssa_def *elem_offset =
nir_iadd_imm(&b->nb, offset, i * type->stride);
struct vtn_ssa_value *comp, temp_val;
if (!load) {
temp_val.def = nir_channel(&b->nb, (*inout)->def, i);
temp_val.type = glsl_scalar_type(base_type);
}
comp = &temp_val;
_vtn_load_store_tail(b, op, load, index, elem_offset,
access_offset, access_size,
&comp, glsl_scalar_type(base_type),
type->access | access);
per_comp[i] = comp->def;
}
if (load) {
if (*inout == NULL)
*inout = vtn_create_ssa_value(b, type->type);
(*inout)->def = nir_vec(&b->nb, per_comp, elems);
}
}
}
return;
case GLSL_TYPE_ARRAY: {
unsigned elems = glsl_get_length(type->type);
for (unsigned i = 0; i < elems; i++) {
nir_ssa_def *elem_off =
nir_iadd_imm(&b->nb, offset, i * type->stride);
_vtn_block_load_store(b, op, load, index, elem_off,
access_offset, access_size,
type->array_element,
type->array_element->access | access,
&(*inout)->elems[i]);
}
return;
}
case GLSL_TYPE_STRUCT: {
unsigned elems = glsl_get_length(type->type);
for (unsigned i = 0; i < elems; i++) {
nir_ssa_def *elem_off =
nir_iadd_imm(&b->nb, offset, type->offsets[i]);
_vtn_block_load_store(b, op, load, index, elem_off,
access_offset, access_size,
type->members[i],
type->members[i]->access | access,
&(*inout)->elems[i]);
}
return;
}
default:
vtn_fail("Invalid block member type");
}
}
static struct vtn_ssa_value *
vtn_block_load(struct vtn_builder *b, struct vtn_pointer *src)
{
nir_intrinsic_op op;
unsigned access_offset = 0, access_size = 0;
switch (src->mode) {
case vtn_variable_mode_ubo:
op = nir_intrinsic_load_ubo;
break;
case vtn_variable_mode_ssbo:
op = nir_intrinsic_load_ssbo;
break;
case vtn_variable_mode_push_constant:
op = nir_intrinsic_load_push_constant;
access_size = b->shader->num_uniforms;
break;
case vtn_variable_mode_workgroup:
op = nir_intrinsic_load_shared;
break;
default:
vtn_fail("Invalid block variable mode");
}
nir_ssa_def *offset, *index = NULL;
offset = vtn_pointer_to_offset(b, src, &index);
struct vtn_ssa_value *value = NULL;
_vtn_block_load_store(b, op, true, index, offset,
access_offset, access_size,
src->type, src->access, &value);
return value;
}
static void
vtn_block_store(struct vtn_builder *b, struct vtn_ssa_value *src,
struct vtn_pointer *dst)
{
nir_intrinsic_op op;
switch (dst->mode) {
case vtn_variable_mode_ssbo:
op = nir_intrinsic_store_ssbo;
break;
case vtn_variable_mode_workgroup:
op = nir_intrinsic_store_shared;
break;
default:
vtn_fail("Invalid block variable mode");
}
nir_ssa_def *offset, *index = NULL;
offset = vtn_pointer_to_offset(b, dst, &index);
_vtn_block_load_store(b, op, false, index, offset,
0, 0, dst->type, dst->access, &src);
}
static void
_vtn_variable_load_store(struct vtn_builder *b, bool load,
struct vtn_pointer *ptr,
struct vtn_ssa_value **inout)
{
enum glsl_base_type base_type = glsl_get_base_type(ptr->type->type);
switch (base_type) {
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
case GLSL_TYPE_UINT16:
case GLSL_TYPE_INT16:
case GLSL_TYPE_UINT8:
case GLSL_TYPE_INT8:
case GLSL_TYPE_UINT64:
case GLSL_TYPE_INT64:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_FLOAT16:
case GLSL_TYPE_BOOL:
case GLSL_TYPE_DOUBLE:
if (glsl_type_is_vector_or_scalar(ptr->type->type)) {
/* We hit a vector or scalar; go ahead and emit the load[s] */
nir_deref_instr *deref = vtn_pointer_to_deref(b, ptr);
if (vtn_pointer_is_external_block(b, ptr)) {
/* If it's external, we call nir_load/store_deref directly. The
* vtn_local_load/store helpers are too clever and do magic to
* avoid array derefs of vectors. That magic is both less
* efficient than the direct load/store and, in the case of
* stores, is broken because it creates a race condition if two
* threads are writing to different components of the same vector
* due to the load+insert+store it uses to emulate the array
* deref.
*/
if (load) {
*inout = vtn_create_ssa_value(b, ptr->type->type);
(*inout)->def = nir_load_deref(&b->nb, deref);
} else {
nir_store_deref(&b->nb, deref, (*inout)->def, ~0);
}
} else {
if (load) {
*inout = vtn_local_load(b, deref);
} else {
vtn_local_store(b, *inout, deref);
}
}
return;
}
/* Fall through */
case GLSL_TYPE_ARRAY:
case GLSL_TYPE_STRUCT: {
unsigned elems = glsl_get_length(ptr->type->type);
if (load) {
vtn_assert(*inout == NULL);
*inout = rzalloc(b, struct vtn_ssa_value);
(*inout)->type = ptr->type->type;
(*inout)->elems = rzalloc_array(b, struct vtn_ssa_value *, elems);
}
struct vtn_access_chain chain = {
.length = 1,
.link = {
{ .mode = vtn_access_mode_literal, },
}
};
for (unsigned i = 0; i < elems; i++) {
chain.link[0].id = i;
struct vtn_pointer *elem = vtn_pointer_dereference(b, ptr, &chain);
_vtn_variable_load_store(b, load, elem, &(*inout)->elems[i]);
}
return;
}
default:
vtn_fail("Invalid access chain type");
}
}
struct vtn_ssa_value *
vtn_variable_load(struct vtn_builder *b, struct vtn_pointer *src)
{
if (vtn_pointer_uses_ssa_offset(b, src)) {
return vtn_block_load(b, src);
} else {
struct vtn_ssa_value *val = NULL;
_vtn_variable_load_store(b, true, src, &val);
return val;
}
}
void
vtn_variable_store(struct vtn_builder *b, struct vtn_ssa_value *src,
struct vtn_pointer *dest)
{
if (vtn_pointer_uses_ssa_offset(b, dest)) {
vtn_assert(dest->mode == vtn_variable_mode_ssbo ||
dest->mode == vtn_variable_mode_workgroup);
vtn_block_store(b, src, dest);
} else {
_vtn_variable_load_store(b, false, dest, &src);
}
}
static void
_vtn_variable_copy(struct vtn_builder *b, struct vtn_pointer *dest,
struct vtn_pointer *src)
{
vtn_assert(src->type->type == dest->type->type);
enum glsl_base_type base_type = glsl_get_base_type(src->type->type);
switch (base_type) {
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
case GLSL_TYPE_UINT16:
case GLSL_TYPE_INT16:
case GLSL_TYPE_UINT8:
case GLSL_TYPE_INT8:
case GLSL_TYPE_UINT64:
case GLSL_TYPE_INT64:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_FLOAT16:
case GLSL_TYPE_DOUBLE:
case GLSL_TYPE_BOOL:
/* At this point, we have a scalar, vector, or matrix so we know that
* there cannot be any structure splitting still in the way. By
* stopping at the matrix level rather than the vector level, we
* ensure that matrices get loaded in the optimal way even if they
* are storred row-major in a UBO.
*/
vtn_variable_store(b, vtn_variable_load(b, src), dest);
return;
case GLSL_TYPE_ARRAY:
case GLSL_TYPE_STRUCT: {
struct vtn_access_chain chain = {
.length = 1,
.link = {
{ .mode = vtn_access_mode_literal, },
}
};
unsigned elems = glsl_get_length(src->type->type);
for (unsigned i = 0; i < elems; i++) {
chain.link[0].id = i;
struct vtn_pointer *src_elem =
vtn_pointer_dereference(b, src, &chain);
struct vtn_pointer *dest_elem =
vtn_pointer_dereference(b, dest, &chain);
_vtn_variable_copy(b, dest_elem, src_elem);
}
return;
}
default:
vtn_fail("Invalid access chain type");
}
}
static void
vtn_variable_copy(struct vtn_builder *b, struct vtn_pointer *dest,
struct vtn_pointer *src)
{
/* TODO: At some point, we should add a special-case for when we can
* just emit a copy_var intrinsic.
*/
_vtn_variable_copy(b, dest, src);
}
static void
set_mode_system_value(struct vtn_builder *b, nir_variable_mode *mode)
{
vtn_assert(*mode == nir_var_system_value || *mode == nir_var_shader_in);
*mode = nir_var_system_value;
}
static void
vtn_get_builtin_location(struct vtn_builder *b,
SpvBuiltIn builtin, int *location,
nir_variable_mode *mode)
{
switch (builtin) {
case SpvBuiltInPosition:
*location = VARYING_SLOT_POS;
break;
case SpvBuiltInPointSize:
*location = VARYING_SLOT_PSIZ;
break;
case SpvBuiltInClipDistance:
*location = VARYING_SLOT_CLIP_DIST0; /* XXX CLIP_DIST1? */
break;
case SpvBuiltInCullDistance:
*location = VARYING_SLOT_CULL_DIST0;
break;
case SpvBuiltInVertexId:
case SpvBuiltInVertexIndex:
/* The Vulkan spec defines VertexIndex to be non-zero-based and doesn't
* allow VertexId. The ARB_gl_spirv spec defines VertexId to be the
* same as gl_VertexID, which is non-zero-based, and removes
* VertexIndex. Since they're both defined to be non-zero-based, we use
* SYSTEM_VALUE_VERTEX_ID for both.
*/
*location = SYSTEM_VALUE_VERTEX_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInInstanceIndex:
*location = SYSTEM_VALUE_INSTANCE_INDEX;
set_mode_system_value(b, mode);
break;
case SpvBuiltInInstanceId:
*location = SYSTEM_VALUE_INSTANCE_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInPrimitiveId:
if (b->shader->info.stage == MESA_SHADER_FRAGMENT) {
vtn_assert(*mode == nir_var_shader_in);
*location = VARYING_SLOT_PRIMITIVE_ID;
} else if (*mode == nir_var_shader_out) {
*location = VARYING_SLOT_PRIMITIVE_ID;
} else {
*location = SYSTEM_VALUE_PRIMITIVE_ID;
set_mode_system_value(b, mode);
}
break;
case SpvBuiltInInvocationId:
*location = SYSTEM_VALUE_INVOCATION_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInLayer:
*location = VARYING_SLOT_LAYER;
if (b->shader->info.stage == MESA_SHADER_FRAGMENT)
*mode = nir_var_shader_in;
else if (b->shader->info.stage == MESA_SHADER_GEOMETRY)
*mode = nir_var_shader_out;
else if (b->options && b->options->caps.shader_viewport_index_layer &&
(b->shader->info.stage == MESA_SHADER_VERTEX ||
b->shader->info.stage == MESA_SHADER_TESS_EVAL))
*mode = nir_var_shader_out;
else
vtn_fail("invalid stage for SpvBuiltInLayer");
break;
case SpvBuiltInViewportIndex:
*location = VARYING_SLOT_VIEWPORT;
if (b->shader->info.stage == MESA_SHADER_GEOMETRY)
*mode = nir_var_shader_out;
else if (b->options && b->options->caps.shader_viewport_index_layer &&
(b->shader->info.stage == MESA_SHADER_VERTEX ||
b->shader->info.stage == MESA_SHADER_TESS_EVAL))
*mode = nir_var_shader_out;
else if (b->shader->info.stage == MESA_SHADER_FRAGMENT)
*mode = nir_var_shader_in;
else
vtn_fail("invalid stage for SpvBuiltInViewportIndex");
break;
case SpvBuiltInTessLevelOuter:
*location = VARYING_SLOT_TESS_LEVEL_OUTER;
break;
case SpvBuiltInTessLevelInner:
*location = VARYING_SLOT_TESS_LEVEL_INNER;
break;
case SpvBuiltInTessCoord:
*location = SYSTEM_VALUE_TESS_COORD;
set_mode_system_value(b, mode);
break;
case SpvBuiltInPatchVertices:
*location = SYSTEM_VALUE_VERTICES_IN;
set_mode_system_value(b, mode);
break;
case SpvBuiltInFragCoord:
*location = VARYING_SLOT_POS;
vtn_assert(*mode == nir_var_shader_in);
break;
case SpvBuiltInPointCoord:
*location = VARYING_SLOT_PNTC;
vtn_assert(*mode == nir_var_shader_in);
break;
case SpvBuiltInFrontFacing:
*location = SYSTEM_VALUE_FRONT_FACE;
set_mode_system_value(b, mode);
break;
case SpvBuiltInSampleId:
*location = SYSTEM_VALUE_SAMPLE_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInSamplePosition:
*location = SYSTEM_VALUE_SAMPLE_POS;
set_mode_system_value(b, mode);
break;
case SpvBuiltInSampleMask:
if (*mode == nir_var_shader_out) {
*location = FRAG_RESULT_SAMPLE_MASK;
} else {
*location = SYSTEM_VALUE_SAMPLE_MASK_IN;
set_mode_system_value(b, mode);
}
break;
case SpvBuiltInFragDepth:
*location = FRAG_RESULT_DEPTH;
vtn_assert(*mode == nir_var_shader_out);
break;
case SpvBuiltInHelperInvocation:
*location = SYSTEM_VALUE_HELPER_INVOCATION;
set_mode_system_value(b, mode);
break;
case SpvBuiltInNumWorkgroups:
*location = SYSTEM_VALUE_NUM_WORK_GROUPS;
set_mode_system_value(b, mode);
break;
case SpvBuiltInWorkgroupSize:
*location = SYSTEM_VALUE_LOCAL_GROUP_SIZE;
set_mode_system_value(b, mode);
break;
case SpvBuiltInWorkgroupId:
*location = SYSTEM_VALUE_WORK_GROUP_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInLocalInvocationId:
*location = SYSTEM_VALUE_LOCAL_INVOCATION_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInLocalInvocationIndex:
*location = SYSTEM_VALUE_LOCAL_INVOCATION_INDEX;
set_mode_system_value(b, mode);
break;
case SpvBuiltInGlobalInvocationId:
*location = SYSTEM_VALUE_GLOBAL_INVOCATION_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInBaseVertex:
/* OpenGL gl_BaseVertex (SYSTEM_VALUE_BASE_VERTEX) is not the same
* semantic as SPIR-V BaseVertex (SYSTEM_VALUE_FIRST_VERTEX).
*/
*location = SYSTEM_VALUE_FIRST_VERTEX;
set_mode_system_value(b, mode);
break;
case SpvBuiltInBaseInstance:
*location = SYSTEM_VALUE_BASE_INSTANCE;
set_mode_system_value(b, mode);
break;
case SpvBuiltInDrawIndex:
*location = SYSTEM_VALUE_DRAW_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupSize:
*location = SYSTEM_VALUE_SUBGROUP_SIZE;
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupId:
*location = SYSTEM_VALUE_SUBGROUP_ID;
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupLocalInvocationId:
*location = SYSTEM_VALUE_SUBGROUP_INVOCATION;
set_mode_system_value(b, mode);
break;
case SpvBuiltInNumSubgroups:
*location = SYSTEM_VALUE_NUM_SUBGROUPS;
set_mode_system_value(b, mode);
break;
case SpvBuiltInDeviceIndex:
*location = SYSTEM_VALUE_DEVICE_INDEX;
set_mode_system_value(b, mode);
break;
case SpvBuiltInViewIndex:
*location = SYSTEM_VALUE_VIEW_INDEX;
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupEqMask:
*location = SYSTEM_VALUE_SUBGROUP_EQ_MASK,
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupGeMask:
*location = SYSTEM_VALUE_SUBGROUP_GE_MASK,
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupGtMask:
*location = SYSTEM_VALUE_SUBGROUP_GT_MASK,
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupLeMask:
*location = SYSTEM_VALUE_SUBGROUP_LE_MASK,
set_mode_system_value(b, mode);
break;
case SpvBuiltInSubgroupLtMask:
*location = SYSTEM_VALUE_SUBGROUP_LT_MASK,
set_mode_system_value(b, mode);
break;
case SpvBuiltInFragStencilRefEXT:
*location = FRAG_RESULT_STENCIL;
vtn_assert(*mode == nir_var_shader_out);
break;
case SpvBuiltInWorkDim:
*location = SYSTEM_VALUE_WORK_DIM;
set_mode_system_value(b, mode);
break;
case SpvBuiltInGlobalSize:
*location = SYSTEM_VALUE_GLOBAL_GROUP_SIZE;
set_mode_system_value(b, mode);
break;
default:
vtn_fail("unsupported builtin: %u", builtin);
}
}
static void
apply_var_decoration(struct vtn_builder *b,
struct nir_variable_data *var_data,
const struct vtn_decoration *dec)
{
switch (dec->decoration) {
case SpvDecorationRelaxedPrecision:
break; /* FIXME: Do nothing with this for now. */
case SpvDecorationNoPerspective:
var_data->interpolation = INTERP_MODE_NOPERSPECTIVE;
break;
case SpvDecorationFlat:
var_data->interpolation = INTERP_MODE_FLAT;
break;
case SpvDecorationCentroid:
var_data->centroid = true;
break;
case SpvDecorationSample:
var_data->sample = true;
break;
case SpvDecorationInvariant:
var_data->invariant = true;
break;
case SpvDecorationConstant:
var_data->read_only = true;
break;
case SpvDecorationNonReadable:
var_data->image.access |= ACCESS_NON_READABLE;
break;
case SpvDecorationNonWritable:
var_data->read_only = true;
var_data->image.access |= ACCESS_NON_WRITEABLE;
break;
case SpvDecorationRestrict:
var_data->image.access |= ACCESS_RESTRICT;
break;
case SpvDecorationVolatile:
var_data->image.access |= ACCESS_VOLATILE;
break;
case SpvDecorationCoherent:
var_data->image.access |= ACCESS_COHERENT;
break;
case SpvDecorationComponent:
var_data->location_frac = dec->literals[0];
break;
case SpvDecorationIndex:
var_data->index = dec->literals[0];
break;
case SpvDecorationBuiltIn: {
SpvBuiltIn builtin = dec->literals[0];
nir_variable_mode mode = var_data->mode;
vtn_get_builtin_location(b, builtin, &var_data->location, &mode);
var_data->mode = mode;
switch (builtin) {
case SpvBuiltInTessLevelOuter:
case SpvBuiltInTessLevelInner:
var_data->compact = true;
break;
case SpvBuiltInFragCoord:
var_data->pixel_center_integer = b->pixel_center_integer;
/* fallthrough */
case SpvBuiltInSamplePosition:
var_data->origin_upper_left = b->origin_upper_left;
break;
default:
break;
}
}
case SpvDecorationSpecId:
case SpvDecorationRowMajor:
case SpvDecorationColMajor:
case SpvDecorationMatrixStride:
case SpvDecorationAliased:
case SpvDecorationUniform:
case SpvDecorationLinkageAttributes:
break; /* Do nothing with these here */
case SpvDecorationPatch:
var_data->patch = true;
break;
case SpvDecorationLocation:
vtn_fail("Handled above");
case SpvDecorationBlock:
case SpvDecorationBufferBlock:
case SpvDecorationArrayStride:
case SpvDecorationGLSLShared:
case SpvDecorationGLSLPacked:
break; /* These can apply to a type but we don't care about them */
case SpvDecorationBinding:
case SpvDecorationDescriptorSet:
case SpvDecorationNoContraction:
case SpvDecorationInputAttachmentIndex:
vtn_warn("Decoration not allowed for variable or structure member: %s",
spirv_decoration_to_string(dec->decoration));
break;
case SpvDecorationXfbBuffer:
var_data->explicit_xfb_buffer = true;
var_data->xfb_buffer = dec->literals[0];
var_data->always_active_io = true;
break;
case SpvDecorationXfbStride:
var_data->explicit_xfb_stride = true;
var_data->xfb_stride = dec->literals[0];
break;
case SpvDecorationOffset:
var_data->explicit_offset = true;
var_data->offset = dec->literals[0];
break;
case SpvDecorationStream:
var_data->stream = dec->literals[0];
break;
case SpvDecorationCPacked:
case SpvDecorationSaturatedConversion:
case SpvDecorationFuncParamAttr:
case SpvDecorationFPRoundingMode:
case SpvDecorationFPFastMathMode:
case SpvDecorationAlignment:
if (b->shader->info.stage != MESA_SHADER_KERNEL) {
vtn_warn("Decoration only allowed for CL-style kernels: %s",
spirv_decoration_to_string(dec->decoration));
}
break;
case SpvDecorationHlslSemanticGOOGLE:
/* HLSL semantic decorations can safely be ignored by the driver. */
break;
default:
vtn_fail("Unhandled decoration");
}
}
static void
var_is_patch_cb(struct vtn_builder *b, struct vtn_value *val, int member,
const struct vtn_decoration *dec, void *out_is_patch)
{
if (dec->decoration == SpvDecorationPatch) {
*((bool *) out_is_patch) = true;
}
}
static void
var_decoration_cb(struct vtn_builder *b, struct vtn_value *val, int member,
const struct vtn_decoration *dec, void *void_var)
{
struct vtn_variable *vtn_var = void_var;
/* Handle decorations that apply to a vtn_variable as a whole */
switch (dec->decoration) {
case SpvDecorationBinding:
vtn_var->binding = dec->literals[0];
vtn_var->explicit_binding = true;
return;
case SpvDecorationDescriptorSet:
vtn_var->descriptor_set = dec->literals[0];
return;
case SpvDecorationInputAttachmentIndex:
vtn_var->input_attachment_index = dec->literals[0];
return;
case SpvDecorationPatch:
vtn_var->patch = true;
break;
case SpvDecorationOffset:
vtn_var->offset = dec->literals[0];
break;
case SpvDecorationNonWritable:
vtn_var->access |= ACCESS_NON_WRITEABLE;
break;
case SpvDecorationNonReadable:
vtn_var->access |= ACCESS_NON_READABLE;
break;
case SpvDecorationVolatile:
vtn_var->access |= ACCESS_VOLATILE;
break;
case SpvDecorationCoherent:
vtn_var->access |= ACCESS_COHERENT;
break;
case SpvDecorationHlslCounterBufferGOOGLE:
/* HLSL semantic decorations can safely be ignored by the driver. */
break;
default:
break;
}
if (val->value_type == vtn_value_type_pointer) {
assert(val->pointer->var == void_var);
assert(member == -1);
} else {
assert(val->value_type == vtn_value_type_type);
}
/* Location is odd. If applied to a split structure, we have to walk the
* whole thing and accumulate the location. It's easier to handle as a
* special case.
*/
if (dec->decoration == SpvDecorationLocation) {
unsigned location = dec->literals[0];
bool is_vertex_input = false;
if (b->shader->info.stage == MESA_SHADER_FRAGMENT &&
vtn_var->mode == vtn_variable_mode_output) {
location += FRAG_RESULT_DATA0;
} else if (b->shader->info.stage == MESA_SHADER_VERTEX &&
vtn_var->mode == vtn_variable_mode_input) {
is_vertex_input = true;
location += VERT_ATTRIB_GENERIC0;
} else if (vtn_var->mode == vtn_variable_mode_input ||
vtn_var->mode == vtn_variable_mode_output) {
location += vtn_var->patch ? VARYING_SLOT_PATCH0 : VARYING_SLOT_VAR0;
} else if (vtn_var->mode != vtn_variable_mode_uniform) {
vtn_warn("Location must be on input, output, uniform, sampler or "
"image variable");
return;
}
if (vtn_var->var->num_members == 0) {
/* This handles the member and lone variable cases */
vtn_var->var->data.location = location;
} else {
/* This handles the structure member case */
assert(vtn_var->var->members);
for (unsigned i = 0; i < vtn_var->var->num_members; i++) {
vtn_var->var->members[i].location = location;
const struct glsl_type *member_type =
glsl_get_struct_field(vtn_var->var->interface_type, i);
location += glsl_count_attribute_slots(member_type,
is_vertex_input);
}
}
return;
} else {
if (vtn_var->var) {
if (vtn_var->var->num_members == 0) {
/* We call this function on types as well as variables and not all
* struct types get split so we can end up having stray member
* decorations; just ignore them.
*/
if (member == -1)
apply_var_decoration(b, &vtn_var->var->data, dec);
} else if (member >= 0) {
/* Member decorations must come from a type */
assert(val->value_type == vtn_value_type_type);
apply_var_decoration(b, &vtn_var->var->members[member], dec);
} else {
unsigned length =
glsl_get_length(glsl_without_array(vtn_var->type->type));
for (unsigned i = 0; i < length; i++)
apply_var_decoration(b, &vtn_var->var->members[i], dec);
}
} else {
/* A few variables, those with external storage, have no actual
* nir_variables associated with them. Fortunately, all decorations
* we care about for those variables are on the type only.
*/
vtn_assert(vtn_var->mode == vtn_variable_mode_ubo ||
vtn_var->mode == vtn_variable_mode_ssbo ||
vtn_var->mode == vtn_variable_mode_push_constant ||
(vtn_var->mode == vtn_variable_mode_workgroup &&
b->options->lower_workgroup_access_to_offsets));
}
}
}
static enum vtn_variable_mode
vtn_storage_class_to_mode(struct vtn_builder *b,
SpvStorageClass class,
struct vtn_type *interface_type,
nir_variable_mode *nir_mode_out)
{
enum vtn_variable_mode mode;
nir_variable_mode nir_mode;
switch (class) {
case SpvStorageClassUniform:
if (interface_type->block) {
mode = vtn_variable_mode_ubo;
nir_mode = nir_var_mem_ubo;
} else if (interface_type->buffer_block) {
mode = vtn_variable_mode_ssbo;
nir_mode = nir_var_mem_ssbo;
} else {
/* Default-block uniforms, coming from gl_spirv */
mode = vtn_variable_mode_uniform;
nir_mode = nir_var_uniform;
}
break;
case SpvStorageClassStorageBuffer:
mode = vtn_variable_mode_ssbo;
nir_mode = nir_var_mem_ssbo;
break;
case SpvStorageClassUniformConstant:
mode = vtn_variable_mode_uniform;
nir_mode = nir_var_uniform;
break;
case SpvStorageClassPushConstant:
mode = vtn_variable_mode_push_constant;
nir_mode = nir_var_uniform;
break;
case SpvStorageClassInput:
mode = vtn_variable_mode_input;
nir_mode = nir_var_shader_in;
break;
case SpvStorageClassOutput:
mode = vtn_variable_mode_output;
nir_mode = nir_var_shader_out;
break;
case SpvStorageClassPrivate:
mode = vtn_variable_mode_private;
nir_mode = nir_var_shader_temp;
break;
case SpvStorageClassFunction:
mode = vtn_variable_mode_function;
nir_mode = nir_var_function_temp;
break;
case SpvStorageClassWorkgroup:
mode = vtn_variable_mode_workgroup;
nir_mode = nir_var_mem_shared;
break;
case SpvStorageClassAtomicCounter:
mode = vtn_variable_mode_uniform;
nir_mode = nir_var_uniform;
break;
case SpvStorageClassCrossWorkgroup:
mode = vtn_variable_mode_cross_workgroup;
nir_mode = nir_var_mem_global;
break;
case SpvStorageClassGeneric:
default:
vtn_fail("Unhandled variable storage class");
}
if (nir_mode_out)
*nir_mode_out = nir_mode;
return mode;
}
nir_ssa_def *
vtn_pointer_to_ssa(struct vtn_builder *b, struct vtn_pointer *ptr)
{
if (vtn_pointer_uses_ssa_offset(b, ptr)) {
/* This pointer needs to have a pointer type with actual storage */
vtn_assert(ptr->ptr_type);
vtn_assert(ptr->ptr_type->type);
if (!ptr->offset) {
/* If we don't have an offset then we must be a pointer to the variable
* itself.
*/
vtn_assert(!ptr->offset && !ptr->block_index);
struct vtn_access_chain chain = {
.length = 0,
};
ptr = vtn_ssa_offset_pointer_dereference(b, ptr, &chain);
}
vtn_assert(ptr->offset);
if (ptr->block_index) {
vtn_assert(ptr->mode == vtn_variable_mode_ubo ||
ptr->mode == vtn_variable_mode_ssbo);
return nir_vec2(&b->nb, ptr->block_index, ptr->offset);
} else {
vtn_assert(ptr->mode == vtn_variable_mode_workgroup);
return ptr->offset;
}
} else {
if (vtn_pointer_is_external_block(b, ptr) &&
vtn_type_contains_block(b, ptr->type)) {
const unsigned bit_size = glsl_get_bit_size(ptr->ptr_type->type);
const unsigned num_components =
glsl_get_vector_elements(ptr->ptr_type->type);
/* In this case, we're looking for a block index and not an actual
* deref.
*/
if (!ptr->block_index) {
/* If we don't have a block_index then we must be a pointer to the
* variable itself.
*/
vtn_assert(!ptr->deref);
struct vtn_access_chain chain = {
.length = 0,
};
ptr = vtn_nir_deref_pointer_dereference(b, ptr, &chain);
}
/* A block index is just a 32-bit value but the pointer has some
* other dimensionality. Cram it in there and we'll unpack it later
* in vtn_pointer_from_ssa.
*/
const unsigned swiz[4] = { 0, };
return nir_swizzle(&b->nb, nir_u2u(&b->nb, ptr->block_index, bit_size),
swiz, num_components, false);
} else {
return &vtn_pointer_to_deref(b, ptr)->dest.ssa;
}
}
}
struct vtn_pointer *
vtn_pointer_from_ssa(struct vtn_builder *b, nir_ssa_def *ssa,
struct vtn_type *ptr_type)
{
vtn_assert(ssa->num_components <= 2 && ssa->bit_size == 32);
vtn_assert(ptr_type->base_type == vtn_base_type_pointer);
struct vtn_type *interface_type = ptr_type->deref;
while (interface_type->base_type == vtn_base_type_array)
interface_type = interface_type->array_element;
struct vtn_pointer *ptr = rzalloc(b, struct vtn_pointer);
nir_variable_mode nir_mode;
ptr->mode = vtn_storage_class_to_mode(b, ptr_type->storage_class,
interface_type, &nir_mode);
ptr->type = ptr_type->deref;
ptr->ptr_type = ptr_type;
if (b->wa_glslang_179) {
/* To work around https://github.com/KhronosGroup/glslang/issues/179 we
* need to whack the mode because it creates a function parameter with
* the Function storage class even though it's a pointer to a sampler.
* If we don't do this, then NIR won't get rid of the deref_cast for us.
*/
if (ptr->mode == vtn_variable_mode_function &&
(ptr->type->base_type == vtn_base_type_sampler ||
ptr->type->base_type == vtn_base_type_sampled_image)) {
ptr->mode = vtn_variable_mode_uniform;
nir_mode = nir_var_uniform;
}
}
if (vtn_pointer_uses_ssa_offset(b, ptr)) {
/* This pointer type needs to have actual storage */
vtn_assert(ptr_type->type);
if (ptr->mode == vtn_variable_mode_ubo ||
ptr->mode == vtn_variable_mode_ssbo) {
vtn_assert(ssa->num_components == 2);
ptr->block_index = nir_channel(&b->nb, ssa, 0);
ptr->offset = nir_channel(&b->nb, ssa, 1);
} else {
vtn_assert(ssa->num_components == 1);
ptr->block_index = NULL;
ptr->offset = ssa;
}
} else {
const struct glsl_type *deref_type = ptr_type->deref->type;
if (!vtn_pointer_is_external_block(b, ptr)) {
assert(ssa->bit_size == 32 && ssa->num_components == 1);
ptr->deref = nir_build_deref_cast(&b->nb, ssa, nir_mode,
glsl_get_bare_type(deref_type), 0);
} else if (vtn_type_contains_block(b, ptr->type)) {
/* This is a pointer to somewhere in an array of blocks, not a
* pointer to somewhere inside the block. We squashed it into a
* random vector type before so just pick off the first channel and
* cast it back to 32 bits.
*/
ptr->block_index = nir_u2u32(&b->nb, nir_channel(&b->nb, ssa, 0));
} else {
/* This is a pointer to something internal or a pointer inside a
* block. It's just a regular cast.
*/
ptr->deref = nir_build_deref_cast(&b->nb, ssa, nir_mode,
ptr_type->deref->type,
ptr_type->stride);
ptr->deref->dest.ssa.num_components =
glsl_get_vector_elements(ptr_type->type);
ptr->deref->dest.ssa.bit_size = glsl_get_bit_size(ptr_type->type);
}
}
return ptr;
}
static bool
is_per_vertex_inout(const struct vtn_variable *var, gl_shader_stage stage)
{
if (var->patch || !glsl_type_is_array(var->type->type))
return false;
if (var->mode == vtn_variable_mode_input) {
return stage == MESA_SHADER_TESS_CTRL ||
stage == MESA_SHADER_TESS_EVAL ||
stage == MESA_SHADER_GEOMETRY;
}
if (var->mode == vtn_variable_mode_output)
return stage == MESA_SHADER_TESS_CTRL;
return false;
}
static void
vtn_create_variable(struct vtn_builder *b, struct vtn_value *val,
struct vtn_type *ptr_type, SpvStorageClass storage_class,
nir_constant *initializer)
{
vtn_assert(ptr_type->base_type == vtn_base_type_pointer);
struct vtn_type *type = ptr_type->deref;
struct vtn_type *without_array = type;
while(glsl_type_is_array(without_array->type))
without_array = without_array->array_element;
enum vtn_variable_mode mode;
nir_variable_mode nir_mode;
mode = vtn_storage_class_to_mode(b, storage_class, without_array, &nir_mode);
switch (mode) {
case vtn_variable_mode_ubo:
/* There's no other way to get vtn_variable_mode_ubo */
vtn_assert(without_array->block);
b->shader->info.num_ubos++;
break;
case vtn_variable_mode_ssbo:
if (storage_class == SpvStorageClassStorageBuffer &&
!without_array->block) {
if (b->variable_pointers) {
vtn_fail("Variables in the StorageBuffer storage class must "
"have a struct type with the Block decoration");
} else {
/* If variable pointers are not present, it's still malformed
* SPIR-V but we can parse it and do the right thing anyway.
* Since some of the 8-bit storage tests have bugs in this are,
* just make it a warning for now.
*/
vtn_warn("Variables in the StorageBuffer storage class must "
"have a struct type with the Block decoration");
}
}
b->shader->info.num_ssbos++;
break;
case vtn_variable_mode_uniform:
if (glsl_type_is_image(without_array->type))
b->shader->info.num_images++;
else if (glsl_type_is_sampler(without_array->type))
b->shader->info.num_textures++;
break;
case vtn_variable_mode_push_constant:
b->shader->num_uniforms = vtn_type_block_size(b, type);
break;
default:
/* No tallying is needed */
break;
}
struct vtn_variable *var = rzalloc(b, struct vtn_variable);
var->type = type;
var->mode = mode;
vtn_assert(val->value_type == vtn_value_type_pointer);
val->pointer = vtn_pointer_for_variable(b, var, ptr_type);
switch (var->mode) {
case vtn_variable_mode_function:
case vtn_variable_mode_private:
case vtn_variable_mode_uniform:
/* For these, we create the variable normally */
var->var = rzalloc(b->shader, nir_variable);
var->var->name = ralloc_strdup(var->var, val->name);
if (storage_class == SpvStorageClassAtomicCounter) {
/* Need to tweak the nir type here as at vtn_handle_type we don't
* have the access to storage_class, that is the one that points us
* that is an atomic uint.
*/
var->var->type = repair_atomic_type(var->type->type);
} else {
/* Private variables don't have any explicit layout but some layouts
* may have leaked through due to type deduplication in the SPIR-V.
*/
var->var->type = glsl_get_bare_type(var->type->type);
}
var->var->data.mode = nir_mode;
var->var->data.location = -1;
var->var->interface_type = NULL;
break;
case vtn_variable_mode_workgroup:
if (b->options->lower_workgroup_access_to_offsets) {
var->shared_location = -1;
} else {
/* Create the variable normally */
var->var = rzalloc(b->shader, nir_variable);
var->var->name = ralloc_strdup(var->var, val->name);
/* Workgroup variables don't have any explicit layout but some
* layouts may have leaked through due to type deduplication in the
* SPIR-V.
*/
var->var->type = glsl_get_bare_type(var->type->type);
var->var->data.mode = nir_var_mem_shared;
}
break;
case vtn_variable_mode_input:
case vtn_variable_mode_output: {
/* In order to know whether or not we're a per-vertex inout, we need
* the patch qualifier. This means walking the variable decorations
* early before we actually create any variables. Not a big deal.
*
* GLSLang really likes to place decorations in the most interior
* thing it possibly can. In particular, if you have a struct, it
* will place the patch decorations on the struct members. This
* should be handled by the variable splitting below just fine.
*
* If you have an array-of-struct, things get even more weird as it
* will place the patch decorations on the struct even though it's
* inside an array and some of the members being patch and others not
* makes no sense whatsoever. Since the only sensible thing is for
* it to be all or nothing, we'll call it patch if any of the members
* are declared patch.
*/
var->patch = false;
vtn_foreach_decoration(b, val, var_is_patch_cb, &var->patch);
if (glsl_type_is_array(var->type->type) &&
glsl_type_is_struct(without_array->type)) {
vtn_foreach_decoration(b, vtn_value(b, without_array->id,
vtn_value_type_type),
var_is_patch_cb, &var->patch);
}
/* For inputs and outputs, we immediately split structures. This
* is for a couple of reasons. For one, builtins may all come in
* a struct and we really want those split out into separate
* variables. For another, interpolation qualifiers can be
* applied to members of the top-level struct ane we need to be
* able to preserve that information.
*/
struct vtn_type *interface_type = var->type;
if (is_per_vertex_inout(var, b->shader->info.stage)) {
/* In Geometry shaders (and some tessellation), inputs come
* in per-vertex arrays. However, some builtins come in
* non-per-vertex, hence the need for the is_array check. In
* any case, there are no non-builtin arrays allowed so this
* check should be sufficient.
*/
interface_type = var->type->array_element;
}
var->var = rzalloc(b->shader, nir_variable);
var->var->name = ralloc_strdup(var->var, val->name);
/* In Vulkan, shader I/O variables don't have any explicit layout but
* some layouts may have leaked through due to type deduplication in
* the SPIR-V.
*/
var->var->type = glsl_get_bare_type(var->type->type);
var->var->interface_type = interface_type->type;
var->var->data.mode = nir_mode;
var->var->data.patch = var->patch;
if (interface_type->base_type == vtn_base_type_struct &&
interface_type->block) {
/* It's a struct. Set it up as per-member. */
var->var->num_members = glsl_get_length(interface_type->type);
var->var->members = rzalloc_array(var->var, struct nir_variable_data,
var->var->num_members);
for (unsigned i = 0; i < var->var->num_members; i++) {
var->var->members[i].mode = nir_mode;
var->var->members[i].patch = var->patch;
}
}
/* For inputs and outputs, we need to grab locations and builtin
* information from the interface type.
*/
vtn_foreach_decoration(b, vtn_value(b, interface_type->id,
vtn_value_type_type),
var_decoration_cb, var);
break;
}
case vtn_variable_mode_ubo:
case vtn_variable_mode_ssbo:
case vtn_variable_mode_push_constant:
case vtn_variable_mode_cross_workgroup:
/* These don't need actual variables. */
break;
}
if (initializer) {
var->var->constant_initializer =
nir_constant_clone(initializer, var->var);
}
vtn_foreach_decoration(b, val, var_decoration_cb, var);
if (var->mode == vtn_variable_mode_uniform) {
/* XXX: We still need the binding information in the nir_variable
* for these. We should fix that.
*/
var->var->data.binding = var->binding;
var->var->data.explicit_binding = var->explicit_binding;
var->var->data.descriptor_set = var->descriptor_set;
var->var->data.index = var->input_attachment_index;
var->var->data.offset = var->offset;
if (glsl_type_is_image(without_array->type))
var->var->data.image.format = without_array->image_format;
}
if (var->mode == vtn_variable_mode_function) {
vtn_assert(var->var != NULL && var->var->members == NULL);
nir_function_impl_add_variable(b->nb.impl, var->var);
} else if (var->var) {
nir_shader_add_variable(b->shader, var->var);
} else {
vtn_assert(vtn_pointer_is_external_block(b, val->pointer));
}
}
static void
vtn_assert_types_equal(struct vtn_builder *b, SpvOp opcode,
struct vtn_type *dst_type,
struct vtn_type *src_type)
{
if (dst_type->id == src_type->id)
return;
if (vtn_types_compatible(b, dst_type, src_type)) {
/* Early versions of GLSLang would re-emit types unnecessarily and you
* would end up with OpLoad, OpStore, or OpCopyMemory opcodes which have
* mismatched source and destination types.
*
* https://github.com/KhronosGroup/glslang/issues/304
* https://github.com/KhronosGroup/glslang/issues/307
* https://bugs.freedesktop.org/show_bug.cgi?id=104338
* https://bugs.freedesktop.org/show_bug.cgi?id=104424
*/
vtn_warn("Source and destination types of %s do not have the same "
"ID (but are compatible): %u vs %u",
spirv_op_to_string(opcode), dst_type->id, src_type->id);
return;
}
vtn_fail("Source and destination types of %s do not match: %s vs. %s",
spirv_op_to_string(opcode),
glsl_get_type_name(dst_type->type),
glsl_get_type_name(src_type->type));
}
void
vtn_handle_variables(struct vtn_builder *b, SpvOp opcode,
const uint32_t *w, unsigned count)
{
switch (opcode) {
case SpvOpUndef: {
struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_undef);
val->type = vtn_value(b, w[1], vtn_value_type_type)->type;
break;
}
case SpvOpVariable: {
struct vtn_type *ptr_type = vtn_value(b, w[1], vtn_value_type_type)->type;
struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_pointer);
SpvStorageClass storage_class = w[3];
nir_constant *initializer = NULL;
if (count > 4)
initializer = vtn_value(b, w[4], vtn_value_type_constant)->constant;
vtn_create_variable(b, val, ptr_type, storage_class, initializer);
break;
}
case SpvOpAccessChain:
case SpvOpPtrAccessChain:
case SpvOpInBoundsAccessChain: {
struct vtn_access_chain *chain = vtn_access_chain_create(b, count - 4);
chain->ptr_as_array = (opcode == SpvOpPtrAccessChain);
unsigned idx = 0;
for (int i = 4; i < count; i++) {
struct vtn_value *link_val = vtn_untyped_value(b, w[i]);
if (link_val->value_type == vtn_value_type_constant) {
chain->link[idx].mode = vtn_access_mode_literal;
switch (glsl_get_bit_size(link_val->type->type)) {
case 8:
chain->link[idx].id = link_val->constant->values[0].i8[0];
break;
case 16:
chain->link[idx].id = link_val->constant->values[0].i16[0];
break;
case 32:
chain->link[idx].id = link_val->constant->values[0].i32[0];
break;
case 64:
chain->link[idx].id = link_val->constant->values[0].i64[0];
break;
default:
vtn_fail("Invalid bit size");
}
} else {
chain->link[idx].mode = vtn_access_mode_id;
chain->link[idx].id = w[i];
}
idx++;
}
struct vtn_type *ptr_type = vtn_value(b, w[1], vtn_value_type_type)->type;
struct vtn_value *base_val = vtn_untyped_value(b, w[3]);
if (base_val->value_type == vtn_value_type_sampled_image) {
/* This is rather insane. SPIR-V allows you to use OpSampledImage
* to combine an array of images with a single sampler to get an
* array of sampled images that all share the same sampler.
* Fortunately, this means that we can more-or-less ignore the
* sampler when crawling the access chain, but it does leave us
* with this rather awkward little special-case.
*/
struct vtn_value *val =
vtn_push_value(b, w[2], vtn_value_type_sampled_image);
val->sampled_image = ralloc(b, struct vtn_sampled_image);
val->sampled_image->type = base_val->sampled_image->type;
val->sampled_image->image =
vtn_pointer_dereference(b, base_val->sampled_image->image, chain);
val->sampled_image->sampler = base_val->sampled_image->sampler;
} else {
vtn_assert(base_val->value_type == vtn_value_type_pointer);
struct vtn_value *val =
vtn_push_value(b, w[2], vtn_value_type_pointer);
val->pointer = vtn_pointer_dereference(b, base_val->pointer, chain);
val->pointer->ptr_type = ptr_type;
}
break;
}
case SpvOpCopyMemory: {
struct vtn_value *dest = vtn_value(b, w[1], vtn_value_type_pointer);
struct vtn_value *src = vtn_value(b, w[2], vtn_value_type_pointer);
vtn_assert_types_equal(b, opcode, dest->type->deref, src->type->deref);
vtn_variable_copy(b, dest->pointer, src->pointer);
break;
}
case SpvOpLoad: {
struct vtn_type *res_type =
vtn_value(b, w[1], vtn_value_type_type)->type;
struct vtn_value *src_val = vtn_value(b, w[3], vtn_value_type_pointer);
struct vtn_pointer *src = src_val->pointer;
vtn_assert_types_equal(b, opcode, res_type, src_val->type->deref);
if (glsl_type_is_image(res_type->type) ||
glsl_type_is_sampler(res_type->type)) {
vtn_push_value(b, w[2], vtn_value_type_pointer)->pointer = src;
return;
}
vtn_push_ssa(b, w[2], res_type, vtn_variable_load(b, src));
break;
}
case SpvOpStore: {
struct vtn_value *dest_val = vtn_value(b, w[1], vtn_value_type_pointer);
struct vtn_pointer *dest = dest_val->pointer;
struct vtn_value *src_val = vtn_untyped_value(b, w[2]);
/* OpStore requires us to actually have a storage type */
vtn_fail_if(dest->type->type == NULL,
"Invalid destination type for OpStore");
if (glsl_get_base_type(dest->type->type) == GLSL_TYPE_BOOL &&
glsl_get_base_type(src_val->type->type) == GLSL_TYPE_UINT) {
/* Early versions of GLSLang would use uint types for UBOs/SSBOs but
* would then store them to a local variable as bool. Work around
* the issue by doing an implicit conversion.
*
* https://github.com/KhronosGroup/glslang/issues/170
* https://bugs.freedesktop.org/show_bug.cgi?id=104424
*/
vtn_warn("OpStore of value of type OpTypeInt to a pointer to type "
"OpTypeBool. Doing an implicit conversion to work around "
"the problem.");
struct vtn_ssa_value *bool_ssa =
vtn_create_ssa_value(b, dest->type->type);
bool_ssa->def = nir_i2b(&b->nb, vtn_ssa_value(b, w[2])->def);
vtn_variable_store(b, bool_ssa, dest);
break;
}
vtn_assert_types_equal(b, opcode, dest_val->type->deref, src_val->type);
if (glsl_type_is_sampler(dest->type->type)) {
if (b->wa_glslang_179) {
vtn_warn("OpStore of a sampler detected. Doing on-the-fly copy "
"propagation to workaround the problem.");
vtn_assert(dest->var->copy_prop_sampler == NULL);
dest->var->copy_prop_sampler =
vtn_value(b, w[2], vtn_value_type_pointer)->pointer;
} else {
vtn_fail("Vulkan does not allow OpStore of a sampler or image.");
}
break;
}
struct vtn_ssa_value *src = vtn_ssa_value(b, w[2]);
vtn_variable_store(b, src, dest);
break;
}
case SpvOpArrayLength: {
struct vtn_pointer *ptr =
vtn_value(b, w[3], vtn_value_type_pointer)->pointer;
const uint32_t offset = ptr->var->type->offsets[w[4]];
const uint32_t stride = ptr->var->type->members[w[4]]->stride;
if (!ptr->block_index) {
struct vtn_access_chain chain = {
.length = 0,
};
ptr = vtn_ssa_offset_pointer_dereference(b, ptr, &chain);
vtn_assert(ptr->block_index);
}
nir_intrinsic_instr *instr =
nir_intrinsic_instr_create(b->nb.shader,
nir_intrinsic_get_buffer_size);
instr->src[0] = nir_src_for_ssa(ptr->block_index);
nir_ssa_dest_init(&instr->instr, &instr->dest, 1, 32, NULL);
nir_builder_instr_insert(&b->nb, &instr->instr);
nir_ssa_def *buf_size = &instr->dest.ssa;
/* array_length = max(buffer_size - offset, 0) / stride */
nir_ssa_def *array_length =
nir_idiv(&b->nb,
nir_imax(&b->nb,
nir_isub(&b->nb,
buf_size,
nir_imm_int(&b->nb, offset)),
nir_imm_int(&b->nb, 0u)),
nir_imm_int(&b->nb, stride));
struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_ssa);
val->ssa = vtn_create_ssa_value(b, glsl_uint_type());
val->ssa->def = array_length;
break;
}
case SpvOpCopyMemorySized:
default:
vtn_fail("Unhandled opcode");
}
}
|