summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/radeonsi/si_texture.c
blob: 448e7497be540fa17e04c05b96ddb8780c3fec2d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
/*
 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
 * Copyright 2018 Advanced Micro Devices, Inc.
 * All Rights Reserved.
 *
 * 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
 * on the rights to use, copy, modify, merge, publish, distribute, sub
 * license, and/or sell copies of the Software, and to permit persons to whom
 * the Software is furnished to do so, subject to the following conditions:
 *
 * The 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 NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 */

#include "drm-uapi/drm_fourcc.h"
#include "si_pipe.h"
#include "si_query.h"
#include "sid.h"
#include "frontend/drm_driver.h"
#include "util/format/u_format.h"
#include "util/os_time.h"
#include "util/u_log.h"
#include "util/u_memory.h"
#include "util/u_pack_color.h"
#include "util/u_resource.h"
#include "util/u_surface.h"
#include "util/u_transfer.h"

#include <errno.h>
#include <inttypes.h>

#include "amd/addrlib/inc/addrinterface.h"

static enum radeon_surf_mode si_choose_tiling(struct si_screen *sscreen,
                                              const struct pipe_resource *templ,
                                              bool tc_compatible_htile);

bool si_prepare_for_dma_blit(struct si_context *sctx, struct si_texture *dst, unsigned dst_level,
                             unsigned dstx, unsigned dsty, unsigned dstz, struct si_texture *src,
                             unsigned src_level, const struct pipe_box *src_box)
{
   if (!sctx->sdma_cs)
      return false;

   if (dst->surface.bpe != src->surface.bpe)
      return false;

   /* MSAA: Blits don't exist in the real world. */
   if (src->buffer.b.b.nr_samples > 1 || dst->buffer.b.b.nr_samples > 1)
      return false;

   /* Depth-stencil surfaces:
    *   When dst is linear, the DB->CB copy preserves HTILE.
    *   When dst is tiled, the 3D path must be used to update HTILE.
    */
   if (src->is_depth || dst->is_depth)
      return false;

   /* DCC as:
    *   src: Use the 3D path. DCC decompression is expensive.
    *   dst: Use the 3D path to compress the pixels with DCC.
    */
   if (vi_dcc_enabled(src, src_level) || vi_dcc_enabled(dst, dst_level))
      return false;

   /* TMZ: mixing encrypted and non-encrypted buffer in a single command
    * doesn't seem supported.
    */
   if ((src->buffer.flags & RADEON_FLAG_ENCRYPTED) !=
       (dst->buffer.flags & RADEON_FLAG_ENCRYPTED))
      return false;

   /* CMASK as:
    *   src: Both texture and SDMA paths need decompression. Use SDMA.
    *   dst: If overwriting the whole texture, discard CMASK and use
    *        SDMA. Otherwise, use the 3D path.
    */
   if (dst->cmask_buffer && dst->dirty_level_mask & (1 << dst_level)) {
      /* The CMASK clear is only enabled for the first level. */
      assert(dst_level == 0);
      if (!util_texrange_covers_whole_level(&dst->buffer.b.b, dst_level, dstx, dsty, dstz,
                                            src_box->width, src_box->height, src_box->depth))
         return false;

      si_texture_discard_cmask(sctx->screen, dst);
   }

   /* All requirements are met. Prepare textures for SDMA. */
   if (src->cmask_buffer && src->dirty_level_mask & (1 << src_level))
      sctx->b.flush_resource(&sctx->b, &src->buffer.b.b);

   assert(!(src->dirty_level_mask & (1 << src_level)));
   assert(!(dst->dirty_level_mask & (1 << dst_level)));

   return true;
}

/* Same as resource_copy_region, except that both upsampling and downsampling are allowed. */
static void si_copy_region_with_blit(struct pipe_context *pipe, struct pipe_resource *dst,
                                     unsigned dst_level, unsigned dstx, unsigned dsty,
                                     unsigned dstz, struct pipe_resource *src, unsigned src_level,
                                     const struct pipe_box *src_box)
{
   struct pipe_blit_info blit;

   memset(&blit, 0, sizeof(blit));
   blit.src.resource = src;
   blit.src.format = src->format;
   blit.src.level = src_level;
   blit.src.box = *src_box;
   blit.dst.resource = dst;
   blit.dst.format = dst->format;
   blit.dst.level = dst_level;
   blit.dst.box.x = dstx;
   blit.dst.box.y = dsty;
   blit.dst.box.z = dstz;
   blit.dst.box.width = src_box->width;
   blit.dst.box.height = src_box->height;
   blit.dst.box.depth = src_box->depth;
   blit.mask = util_format_get_mask(dst->format);
   blit.filter = PIPE_TEX_FILTER_NEAREST;

   if (blit.mask) {
      pipe->blit(pipe, &blit);
   }
}

/* Copy from a full GPU texture to a transfer's staging one. */
static void si_copy_to_staging_texture(struct pipe_context *ctx, struct si_transfer *stransfer)
{
   struct si_context *sctx = (struct si_context *)ctx;
   struct pipe_transfer *transfer = (struct pipe_transfer *)stransfer;
   struct pipe_resource *dst = &stransfer->staging->b.b;
   struct pipe_resource *src = transfer->resource;

   if (src->nr_samples > 1 || ((struct si_texture *)src)->is_depth) {
      si_copy_region_with_blit(ctx, dst, 0, 0, 0, 0, src, transfer->level, &transfer->box);
      return;
   }

   sctx->dma_copy(ctx, dst, 0, 0, 0, 0, src, transfer->level, &transfer->box);
}

/* Copy from a transfer's staging texture to a full GPU one. */
static void si_copy_from_staging_texture(struct pipe_context *ctx, struct si_transfer *stransfer)
{
   struct si_context *sctx = (struct si_context *)ctx;
   struct pipe_transfer *transfer = (struct pipe_transfer *)stransfer;
   struct pipe_resource *dst = transfer->resource;
   struct pipe_resource *src = &stransfer->staging->b.b;
   struct pipe_box sbox;

   u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox);

   if (dst->nr_samples > 1 || ((struct si_texture *)dst)->is_depth) {
      si_copy_region_with_blit(ctx, dst, transfer->level, transfer->box.x, transfer->box.y,
                               transfer->box.z, src, 0, &sbox);
      return;
   }

   if (util_format_is_compressed(dst->format)) {
      sbox.width = util_format_get_nblocksx(dst->format, sbox.width);
      sbox.height = util_format_get_nblocksx(dst->format, sbox.height);
   }

   sctx->dma_copy(ctx, dst, transfer->level, transfer->box.x, transfer->box.y, transfer->box.z, src,
                  0, &sbox);
}

static unsigned si_texture_get_offset(struct si_screen *sscreen, struct si_texture *tex,
                                      unsigned level, const struct pipe_box *box, unsigned *stride,
                                      unsigned *layer_stride)
{
   if (sscreen->info.chip_class >= GFX9) {
      *stride = tex->surface.u.gfx9.surf_pitch * tex->surface.bpe;
      *layer_stride = tex->surface.u.gfx9.surf_slice_size;

      if (!box)
         return 0;

      /* Each texture is an array of slices. Each slice is an array
       * of mipmap levels. */
      return tex->surface.u.gfx9.surf_offset + box->z * tex->surface.u.gfx9.surf_slice_size +
             tex->surface.u.gfx9.offset[level] +
             (box->y / tex->surface.blk_h * tex->surface.u.gfx9.surf_pitch +
              box->x / tex->surface.blk_w) *
                tex->surface.bpe;
   } else {
      *stride = tex->surface.u.legacy.level[level].nblk_x * tex->surface.bpe;
      assert((uint64_t)tex->surface.u.legacy.level[level].slice_size_dw * 4 <= UINT_MAX);
      *layer_stride = (uint64_t)tex->surface.u.legacy.level[level].slice_size_dw * 4;

      if (!box)
         return tex->surface.u.legacy.level[level].offset;

      /* Each texture is an array of mipmap levels. Each level is
       * an array of slices. */
      return tex->surface.u.legacy.level[level].offset +
             box->z * (uint64_t)tex->surface.u.legacy.level[level].slice_size_dw * 4 +
             (box->y / tex->surface.blk_h * tex->surface.u.legacy.level[level].nblk_x +
              box->x / tex->surface.blk_w) *
                tex->surface.bpe;
   }
}

static int si_init_surface(struct si_screen *sscreen, struct radeon_surf *surface,
                           const struct pipe_resource *ptex, enum radeon_surf_mode array_mode,
                           bool is_imported, bool is_scanout, bool is_flushed_depth,
                           bool tc_compatible_htile)
{
   const struct util_format_description *desc = util_format_description(ptex->format);
   bool is_depth, is_stencil;
   int r;
   unsigned bpe, flags = 0;

   is_depth = util_format_has_depth(desc);
   is_stencil = util_format_has_stencil(desc);

   if (!is_flushed_depth && ptex->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
      bpe = 4; /* stencil is allocated separately */
   } else {
      bpe = util_format_get_blocksize(ptex->format);
      assert(util_is_power_of_two_or_zero(bpe));
   }

   if (!is_flushed_depth && is_depth) {
      flags |= RADEON_SURF_ZBUFFER;

      if (sscreen->debug_flags & DBG(NO_HYPERZ)) {
         flags |= RADEON_SURF_NO_HTILE;
      } else if (tc_compatible_htile &&
                 (sscreen->info.chip_class >= GFX9 || array_mode == RADEON_SURF_MODE_2D)) {
         /* TC-compatible HTILE only supports Z32_FLOAT.
          * GFX9 also supports Z16_UNORM.
          * On GFX8, promote Z16 to Z32. DB->CB copies will convert
          * the format for transfers.
          */
         if (sscreen->info.chip_class == GFX8)
            bpe = 4;

         flags |= RADEON_SURF_TC_COMPATIBLE_HTILE;
      }

      if (is_stencil)
         flags |= RADEON_SURF_SBUFFER;
   }

   if (sscreen->info.chip_class >= GFX8 &&
       (ptex->flags & SI_RESOURCE_FLAG_DISABLE_DCC ||
        (sscreen->info.chip_class < GFX10_3 &&
         ptex->format == PIPE_FORMAT_R9G9B9E5_FLOAT) ||
        (ptex->nr_samples >= 2 && !sscreen->dcc_msaa_allowed)))
      flags |= RADEON_SURF_DISABLE_DCC;

   /* Stoney: 128bpp MSAA textures randomly fail piglit tests with DCC. */
   if (sscreen->info.family == CHIP_STONEY && bpe == 16 && ptex->nr_samples >= 2)
      flags |= RADEON_SURF_DISABLE_DCC;

   /* GFX8: DCC clear for 4x and 8x MSAA array textures unimplemented. */
   if (sscreen->info.chip_class == GFX8 && ptex->nr_storage_samples >= 4 && ptex->array_size > 1)
      flags |= RADEON_SURF_DISABLE_DCC;

   /* GFX9: DCC clear for 4x and 8x MSAA textures unimplemented. */
   if (sscreen->info.chip_class == GFX9 &&
       (ptex->nr_storage_samples >= 4 ||
        (sscreen->info.family == CHIP_RAVEN && ptex->nr_storage_samples >= 2 && bpe < 4)))
      flags |= RADEON_SURF_DISABLE_DCC;

   /* TODO: GFX10: DCC causes corruption with MSAA. */
   if (sscreen->info.chip_class >= GFX10 && ptex->nr_storage_samples >= 2)
      flags |= RADEON_SURF_DISABLE_DCC;

   /* Shared textures must always set up DCC.
    * If it's not present, it will be disabled by
    * si_get_opaque_metadata later.
    */
   if (!is_imported && (sscreen->debug_flags & DBG(NO_DCC)))
      flags |= RADEON_SURF_DISABLE_DCC;

   if (is_scanout) {
      /* This should catch bugs in gallium users setting incorrect flags. */
      assert(ptex->nr_samples <= 1 && ptex->array_size == 1 && ptex->depth0 == 1 &&
             ptex->last_level == 0 && !(flags & RADEON_SURF_Z_OR_SBUFFER));

      flags |= RADEON_SURF_SCANOUT;
   }

   if (ptex->bind & PIPE_BIND_SHARED)
      flags |= RADEON_SURF_SHAREABLE;
   if (is_imported)
      flags |= RADEON_SURF_IMPORTED | RADEON_SURF_SHAREABLE;
   if (sscreen->debug_flags & DBG(NO_FMASK))
      flags |= RADEON_SURF_NO_FMASK;

   if (sscreen->info.chip_class == GFX9 && (ptex->flags & SI_RESOURCE_FLAG_FORCE_MICRO_TILE_MODE)) {
      flags |= RADEON_SURF_FORCE_MICRO_TILE_MODE;
      surface->micro_tile_mode = SI_RESOURCE_FLAG_MICRO_TILE_MODE_GET(ptex->flags);
   }

   if (ptex->flags & SI_RESOURCE_FLAG_FORCE_MSAA_TILING) {
      flags |= RADEON_SURF_FORCE_SWIZZLE_MODE;

      if (sscreen->info.chip_class >= GFX10)
         surface->u.gfx9.surf.swizzle_mode = ADDR_SW_64KB_R_X;
   }

   r = sscreen->ws->surface_init(sscreen->ws, ptex, flags, bpe, array_mode, surface);
   if (r) {
      return r;
   }

   return 0;
}

void si_eliminate_fast_color_clear(struct si_context *sctx, struct si_texture *tex,
                                   bool *ctx_flushed)
{
   struct si_screen *sscreen = sctx->screen;
   struct pipe_context *ctx = &sctx->b;

   if (ctx == sscreen->aux_context)
      simple_mtx_lock(&sscreen->aux_context_lock);

   unsigned n = sctx->num_decompress_calls;
   ctx->flush_resource(ctx, &tex->buffer.b.b);

   /* Flush only if any fast clear elimination took place. */
   bool flushed = false;
   if (n != sctx->num_decompress_calls)
   {
      ctx->flush(ctx, NULL, 0);
      flushed = true;
   }
   if (ctx_flushed)
      *ctx_flushed = flushed;

   if (ctx == sscreen->aux_context)
      simple_mtx_unlock(&sscreen->aux_context_lock);
}

void si_texture_discard_cmask(struct si_screen *sscreen, struct si_texture *tex)
{
   if (!tex->cmask_buffer)
      return;

   assert(tex->buffer.b.b.nr_samples <= 1);

   /* Disable CMASK. */
   tex->cmask_base_address_reg = tex->buffer.gpu_address >> 8;
   tex->dirty_level_mask = 0;

   tex->cb_color_info &= ~S_028C70_FAST_CLEAR(1);

   if (tex->cmask_buffer != &tex->buffer)
      si_resource_reference(&tex->cmask_buffer, NULL);

   tex->cmask_buffer = NULL;

   /* Notify all contexts about the change. */
   p_atomic_inc(&sscreen->dirty_tex_counter);
   p_atomic_inc(&sscreen->compressed_colortex_counter);
}

static bool si_can_disable_dcc(struct si_texture *tex)
{
   /* We can't disable DCC if it can be written by another process. */
   return tex->surface.dcc_offset &&
          (!tex->buffer.b.is_shared ||
           !(tex->buffer.external_usage & PIPE_HANDLE_USAGE_FRAMEBUFFER_WRITE));
}

static bool si_texture_discard_dcc(struct si_screen *sscreen, struct si_texture *tex)
{
   if (!si_can_disable_dcc(tex))
      return false;

   assert(tex->dcc_separate_buffer == NULL);

   /* Disable DCC. */
   ac_surface_zero_dcc_fields(&tex->surface);

   /* Notify all contexts about the change. */
   p_atomic_inc(&sscreen->dirty_tex_counter);
   return true;
}

/**
 * Disable DCC for the texture. (first decompress, then discard metadata).
 *
 * There is unresolved multi-context synchronization issue between
 * screen::aux_context and the current context. If applications do this with
 * multiple contexts, it's already undefined behavior for them and we don't
 * have to worry about that. The scenario is:
 *
 * If context 1 disables DCC and context 2 has queued commands that write
 * to the texture via CB with DCC enabled, and the order of operations is
 * as follows:
 *   context 2 queues draw calls rendering to the texture, but doesn't flush
 *   context 1 disables DCC and flushes
 *   context 1 & 2 reset descriptors and FB state
 *   context 2 flushes (new compressed tiles written by the draw calls)
 *   context 1 & 2 read garbage, because DCC is disabled, yet there are
 *   compressed tiled
 *
 * \param sctx  the current context if you have one, or sscreen->aux_context
 *              if you don't.
 */
bool si_texture_disable_dcc(struct si_context *sctx, struct si_texture *tex)
{
   struct si_screen *sscreen = sctx->screen;

   if (!sctx->has_graphics)
      return si_texture_discard_dcc(sscreen, tex);

   if (!si_can_disable_dcc(tex))
      return false;

   if (&sctx->b == sscreen->aux_context)
      simple_mtx_lock(&sscreen->aux_context_lock);

   /* Decompress DCC. */
   si_decompress_dcc(sctx, tex);
   sctx->b.flush(&sctx->b, NULL, 0);

   if (&sctx->b == sscreen->aux_context)
      simple_mtx_unlock(&sscreen->aux_context_lock);

   return si_texture_discard_dcc(sscreen, tex);
}

static void si_reallocate_texture_inplace(struct si_context *sctx, struct si_texture *tex,
                                          unsigned new_bind_flag, bool invalidate_storage)
{
   struct pipe_screen *screen = sctx->b.screen;
   struct si_texture *new_tex;
   struct pipe_resource templ = tex->buffer.b.b;
   unsigned i;

   templ.bind |= new_bind_flag;

   if (tex->buffer.b.is_shared || tex->num_planes > 1)
      return;

   if (new_bind_flag == PIPE_BIND_LINEAR) {
      if (tex->surface.is_linear)
         return;

      /* This fails with MSAA, depth, and compressed textures. */
      if (si_choose_tiling(sctx->screen, &templ, false) != RADEON_SURF_MODE_LINEAR_ALIGNED)
         return;
   }

   new_tex = (struct si_texture *)screen->resource_create(screen, &templ);
   if (!new_tex)
      return;

   /* Copy the pixels to the new texture. */
   if (!invalidate_storage) {
      for (i = 0; i <= templ.last_level; i++) {
         struct pipe_box box;

         u_box_3d(0, 0, 0, u_minify(templ.width0, i), u_minify(templ.height0, i),
                  util_num_layers(&templ, i), &box);

         sctx->dma_copy(&sctx->b, &new_tex->buffer.b.b, i, 0, 0, 0, &tex->buffer.b.b, i, &box);
      }
   }

   if (new_bind_flag == PIPE_BIND_LINEAR) {
      si_texture_discard_cmask(sctx->screen, tex);
      si_texture_discard_dcc(sctx->screen, tex);
   }

   /* Replace the structure fields of tex. */
   tex->buffer.b.b.bind = templ.bind;
   pb_reference(&tex->buffer.buf, new_tex->buffer.buf);
   tex->buffer.gpu_address = new_tex->buffer.gpu_address;
   tex->buffer.vram_usage = new_tex->buffer.vram_usage;
   tex->buffer.gart_usage = new_tex->buffer.gart_usage;
   tex->buffer.bo_size = new_tex->buffer.bo_size;
   tex->buffer.bo_alignment = new_tex->buffer.bo_alignment;
   tex->buffer.domains = new_tex->buffer.domains;
   tex->buffer.flags = new_tex->buffer.flags;

   tex->surface = new_tex->surface;
   si_texture_reference(&tex->flushed_depth_texture, new_tex->flushed_depth_texture);

   tex->surface.fmask_offset = new_tex->surface.fmask_offset;
   tex->surface.cmask_offset = new_tex->surface.cmask_offset;
   tex->cmask_base_address_reg = new_tex->cmask_base_address_reg;

   if (tex->cmask_buffer == &tex->buffer)
      tex->cmask_buffer = NULL;
   else
      si_resource_reference(&tex->cmask_buffer, NULL);

   if (new_tex->cmask_buffer == &new_tex->buffer)
      tex->cmask_buffer = &tex->buffer;
   else
      si_resource_reference(&tex->cmask_buffer, new_tex->cmask_buffer);

   tex->surface.dcc_offset = new_tex->surface.dcc_offset;
   tex->cb_color_info = new_tex->cb_color_info;
   memcpy(tex->color_clear_value, new_tex->color_clear_value, sizeof(tex->color_clear_value));
   tex->last_msaa_resolve_target_micro_mode = new_tex->last_msaa_resolve_target_micro_mode;

   tex->surface.htile_offset = new_tex->surface.htile_offset;
   tex->depth_clear_value = new_tex->depth_clear_value;
   tex->dirty_level_mask = new_tex->dirty_level_mask;
   tex->stencil_dirty_level_mask = new_tex->stencil_dirty_level_mask;
   tex->db_render_format = new_tex->db_render_format;
   tex->stencil_clear_value = new_tex->stencil_clear_value;
   tex->tc_compatible_htile = new_tex->tc_compatible_htile;
   tex->depth_cleared = new_tex->depth_cleared;
   tex->stencil_cleared = new_tex->stencil_cleared;
   tex->upgraded_depth = new_tex->upgraded_depth;
   tex->db_compatible = new_tex->db_compatible;
   tex->can_sample_z = new_tex->can_sample_z;
   tex->can_sample_s = new_tex->can_sample_s;

   tex->separate_dcc_dirty = new_tex->separate_dcc_dirty;
   tex->displayable_dcc_dirty = new_tex->displayable_dcc_dirty;
   tex->dcc_gather_statistics = new_tex->dcc_gather_statistics;
   si_resource_reference(&tex->dcc_separate_buffer, new_tex->dcc_separate_buffer);
   si_resource_reference(&tex->last_dcc_separate_buffer, new_tex->last_dcc_separate_buffer);
   si_resource_reference(&tex->dcc_retile_buffer, new_tex->dcc_retile_buffer);

   if (new_bind_flag == PIPE_BIND_LINEAR) {
      assert(!tex->surface.htile_offset);
      assert(!tex->cmask_buffer);
      assert(!tex->surface.fmask_size);
      assert(!tex->surface.dcc_offset);
      assert(!tex->is_depth);
   }

   si_texture_reference(&new_tex, NULL);

   p_atomic_inc(&sctx->screen->dirty_tex_counter);
}

static void si_set_tex_bo_metadata(struct si_screen *sscreen, struct si_texture *tex)
{
   struct pipe_resource *res = &tex->buffer.b.b;
   struct radeon_bo_metadata md;

   memset(&md, 0, sizeof(md));

   assert(tex->dcc_separate_buffer == NULL);
   assert(tex->surface.fmask_size == 0);

   static const unsigned char swizzle[] = {PIPE_SWIZZLE_X, PIPE_SWIZZLE_Y, PIPE_SWIZZLE_Z,
                                           PIPE_SWIZZLE_W};
   bool is_array = util_texture_is_array(res->target);
   uint32_t desc[8];

   sscreen->make_texture_descriptor(sscreen, tex, true, res->target, res->format, swizzle, 0,
                                    res->last_level, 0, is_array ? res->array_size - 1 : 0,
                                    res->width0, res->height0, res->depth0, desc, NULL);
   si_set_mutable_tex_desc_fields(sscreen, tex, &tex->surface.u.legacy.level[0], 0, 0,
                                  tex->surface.blk_w, false, false, desc);

   ac_surface_get_umd_metadata(&sscreen->info, &tex->surface,
                               tex->buffer.b.b.last_level + 1,
                               desc, &md.size_metadata, md.metadata);
   sscreen->ws->buffer_set_metadata(tex->buffer.buf, &md, &tex->surface);
}

static bool si_has_displayable_dcc(struct si_texture *tex)
{
   struct si_screen *sscreen = (struct si_screen *)tex->buffer.b.b.screen;

   if (sscreen->info.chip_class <= GFX8)
      return false;

   return tex->surface.is_displayable && tex->surface.dcc_offset;
}

static bool si_resource_get_param(struct pipe_screen *screen, struct pipe_context *context,
                                  struct pipe_resource *resource, unsigned plane, unsigned layer,
                                  enum pipe_resource_param param, unsigned handle_usage,
                                  uint64_t *value)
{
   for (unsigned i = 0; i < plane; i++)
      resource = resource->next;

   struct si_screen *sscreen = (struct si_screen *)screen;
   struct si_texture *tex = (struct si_texture *)resource;
   struct winsys_handle whandle;

   switch (param) {
   case PIPE_RESOURCE_PARAM_NPLANES:
      *value = resource->target == PIPE_BUFFER ? 1 : tex->num_planes;
      return true;

   case PIPE_RESOURCE_PARAM_STRIDE:
      if (resource->target == PIPE_BUFFER)
         *value = 0;
      else if (sscreen->info.chip_class >= GFX9)
         *value = tex->surface.u.gfx9.surf_pitch * tex->surface.bpe;
      else
         *value = tex->surface.u.legacy.level[0].nblk_x * tex->surface.bpe;
      return true;

   case PIPE_RESOURCE_PARAM_OFFSET:
      if (resource->target == PIPE_BUFFER)
         *value = 0;
      else if (sscreen->info.chip_class >= GFX9)
         *value = tex->surface.u.gfx9.surf_offset + layer * tex->surface.u.gfx9.surf_slice_size;
      else
         *value = tex->surface.u.legacy.level[0].offset +
                  layer * (uint64_t)tex->surface.u.legacy.level[0].slice_size_dw * 4;
      return true;

   case PIPE_RESOURCE_PARAM_MODIFIER:
      *value = DRM_FORMAT_MOD_INVALID;
      return true;

   case PIPE_RESOURCE_PARAM_HANDLE_TYPE_SHARED:
   case PIPE_RESOURCE_PARAM_HANDLE_TYPE_KMS:
   case PIPE_RESOURCE_PARAM_HANDLE_TYPE_FD:
      memset(&whandle, 0, sizeof(whandle));

      if (param == PIPE_RESOURCE_PARAM_HANDLE_TYPE_SHARED)
         whandle.type = WINSYS_HANDLE_TYPE_SHARED;
      else if (param == PIPE_RESOURCE_PARAM_HANDLE_TYPE_KMS)
         whandle.type = WINSYS_HANDLE_TYPE_KMS;
      else if (param == PIPE_RESOURCE_PARAM_HANDLE_TYPE_FD)
         whandle.type = WINSYS_HANDLE_TYPE_FD;

      if (!screen->resource_get_handle(screen, context, resource, &whandle, handle_usage))
         return false;

      *value = whandle.handle;
      return true;
   }
   return false;
}

static void si_texture_get_info(struct pipe_screen *screen, struct pipe_resource *resource,
                                unsigned *pstride, unsigned *poffset)
{
   uint64_t value;

   if (pstride) {
      si_resource_get_param(screen, NULL, resource, 0, 0, PIPE_RESOURCE_PARAM_STRIDE, 0, &value);
      *pstride = value;
   }

   if (poffset) {
      si_resource_get_param(screen, NULL, resource, 0, 0, PIPE_RESOURCE_PARAM_OFFSET, 0, &value);
      *poffset = value;
   }
}

static bool si_texture_get_handle(struct pipe_screen *screen, struct pipe_context *ctx,
                                  struct pipe_resource *resource, struct winsys_handle *whandle,
                                  unsigned usage)
{
   struct si_screen *sscreen = (struct si_screen *)screen;
   struct si_context *sctx;
   struct si_resource *res = si_resource(resource);
   struct si_texture *tex = (struct si_texture *)resource;
   bool update_metadata = false;
   unsigned stride, offset, slice_size;
   bool flush = false;

   ctx = threaded_context_unwrap_sync(ctx);
   sctx = (struct si_context *)(ctx ? ctx : sscreen->aux_context);

   if (resource->target != PIPE_BUFFER) {
      /* Individual planes are chained pipe_resource instances. */
      for (unsigned i = 0; i < whandle->plane; i++) {
         resource = resource->next;
         res = si_resource(resource);
         tex = (struct si_texture *)resource;
      }

      /* This is not supported now, but it might be required for OpenCL
       * interop in the future.
       */
      if (resource->nr_samples > 1 || tex->is_depth)
         return false;

      /* Move a suballocated texture into a non-suballocated allocation. */
      if (sscreen->ws->buffer_is_suballocated(res->buf) || tex->surface.tile_swizzle ||
          (tex->buffer.flags & RADEON_FLAG_NO_INTERPROCESS_SHARING &&
           sscreen->info.has_local_buffers)) {
         assert(!res->b.is_shared);
         si_reallocate_texture_inplace(sctx, tex, PIPE_BIND_SHARED, false);
         flush = true;
         assert(res->b.b.bind & PIPE_BIND_SHARED);
         assert(res->flags & RADEON_FLAG_NO_SUBALLOC);
         assert(!(res->flags & RADEON_FLAG_NO_INTERPROCESS_SHARING));
         assert(tex->surface.tile_swizzle == 0);
      }

      /* Since shader image stores don't support DCC on GFX8,
       * disable it for external clients that want write
       * access.
       */
      if ((usage & PIPE_HANDLE_USAGE_SHADER_WRITE && tex->surface.dcc_offset) ||
          /* Displayable DCC requires an explicit flush. */
          (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) && si_has_displayable_dcc(tex))) {
         if (si_texture_disable_dcc(sctx, tex)) {
            update_metadata = true;
            /* si_texture_disable_dcc flushes the context */
            flush = false;
         }
      }

      if (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) &&
          (tex->cmask_buffer || tex->surface.dcc_offset)) {
         /* Eliminate fast clear (both CMASK and DCC) */
         bool flushed;
         si_eliminate_fast_color_clear(sctx, tex, &flushed);
         /* eliminate_fast_color_clear sometimes flushes the context */
         if (flushed)
            flush = false;

         /* Disable CMASK if flush_resource isn't going
          * to be called.
          */
         if (tex->cmask_buffer)
            si_texture_discard_cmask(sscreen, tex);
      }

      /* Set metadata. */
      if ((!res->b.is_shared || update_metadata) && whandle->offset == 0)
         si_set_tex_bo_metadata(sscreen, tex);

      if (sscreen->info.chip_class >= GFX9) {
         slice_size = tex->surface.u.gfx9.surf_slice_size;
      } else {
         slice_size = (uint64_t)tex->surface.u.legacy.level[0].slice_size_dw * 4;
      }
   } else {
      /* Buffer exports are for the OpenCL interop. */
      /* Move a suballocated buffer into a non-suballocated allocation. */
      if (sscreen->ws->buffer_is_suballocated(res->buf) ||
          /* A DMABUF export always fails if the BO is local. */
          (tex->buffer.flags & RADEON_FLAG_NO_INTERPROCESS_SHARING &&
           sscreen->info.has_local_buffers)) {
         assert(!res->b.is_shared);

         /* Allocate a new buffer with PIPE_BIND_SHARED. */
         struct pipe_resource templ = res->b.b;
         templ.bind |= PIPE_BIND_SHARED;

         struct pipe_resource *newb = screen->resource_create(screen, &templ);
         if (!newb)
            return false;

         /* Copy the old buffer contents to the new one. */
         struct pipe_box box;
         u_box_1d(0, newb->width0, &box);
         sctx->b.resource_copy_region(&sctx->b, newb, 0, 0, 0, 0, &res->b.b, 0, &box);
         flush = true;
         /* Move the new buffer storage to the old pipe_resource. */
         si_replace_buffer_storage(&sctx->b, &res->b.b, newb);
         pipe_resource_reference(&newb, NULL);

         assert(res->b.b.bind & PIPE_BIND_SHARED);
         assert(res->flags & RADEON_FLAG_NO_SUBALLOC);
      }

      /* Buffers */
      slice_size = 0;
   }

   si_texture_get_info(screen, resource, &stride, &offset);

   if (flush)
      sctx->b.flush(&sctx->b, NULL, 0);

   if (res->b.is_shared) {
      /* USAGE_EXPLICIT_FLUSH must be cleared if at least one user
       * doesn't set it.
       */
      res->external_usage |= usage & ~PIPE_HANDLE_USAGE_EXPLICIT_FLUSH;
      if (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH))
         res->external_usage &= ~PIPE_HANDLE_USAGE_EXPLICIT_FLUSH;
   } else {
      res->b.is_shared = true;
      res->external_usage = usage;
   }

   whandle->stride = stride;
   whandle->offset = offset + slice_size * whandle->layer;

   return sscreen->ws->buffer_get_handle(sscreen->ws, res->buf, whandle);
}

static void si_texture_destroy(struct pipe_screen *screen, struct pipe_resource *ptex)
{
   struct si_texture *tex = (struct si_texture *)ptex;
   struct si_resource *resource = &tex->buffer;

   si_texture_reference(&tex->flushed_depth_texture, NULL);

   if (tex->cmask_buffer != &tex->buffer) {
      si_resource_reference(&tex->cmask_buffer, NULL);
   }
   pb_reference(&resource->buf, NULL);
   si_resource_reference(&tex->dcc_separate_buffer, NULL);
   si_resource_reference(&tex->last_dcc_separate_buffer, NULL);
   si_resource_reference(&tex->dcc_retile_buffer, NULL);
   FREE(tex);
}

static const struct u_resource_vtbl si_texture_vtbl;

void si_print_texture_info(struct si_screen *sscreen, struct si_texture *tex,
                           struct u_log_context *log)
{
   int i;

   /* Common parameters. */
   u_log_printf(log,
                "  Info: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, "
                "blk_h=%u, array_size=%u, last_level=%u, "
                "bpe=%u, nsamples=%u, flags=0x%x, %s\n",
                tex->buffer.b.b.width0, tex->buffer.b.b.height0, tex->buffer.b.b.depth0,
                tex->surface.blk_w, tex->surface.blk_h, tex->buffer.b.b.array_size,
                tex->buffer.b.b.last_level, tex->surface.bpe, tex->buffer.b.b.nr_samples,
                tex->surface.flags, util_format_short_name(tex->buffer.b.b.format));

   if (sscreen->info.chip_class >= GFX9) {
      u_log_printf(log,
                   "  Surf: size=%" PRIu64 ", slice_size=%" PRIu64 ", "
                   "alignment=%u, swmode=%u, epitch=%u, pitch=%u\n",
                   tex->surface.surf_size, tex->surface.u.gfx9.surf_slice_size,
                   tex->surface.surf_alignment, tex->surface.u.gfx9.surf.swizzle_mode,
                   tex->surface.u.gfx9.surf.epitch, tex->surface.u.gfx9.surf_pitch);

      if (tex->surface.fmask_offset) {
         u_log_printf(log,
                      "  FMASK: offset=%" PRIu64 ", size=%" PRIu64 ", "
                      "alignment=%u, swmode=%u, epitch=%u\n",
                      tex->surface.fmask_offset, tex->surface.fmask_size,
                      tex->surface.fmask_alignment, tex->surface.u.gfx9.fmask.swizzle_mode,
                      tex->surface.u.gfx9.fmask.epitch);
      }

      if (tex->cmask_buffer) {
         u_log_printf(log,
                      "  CMask: offset=%" PRIu64 ", size=%u, "
                      "alignment=%u\n",
                      tex->surface.cmask_offset, tex->surface.cmask_size,
                      tex->surface.cmask_alignment);
      }

      if (tex->surface.htile_offset) {
         u_log_printf(log,
                      "  HTile: offset=%" PRIu64 ", size=%u, alignment=%u\n",
                      tex->surface.htile_offset, tex->surface.htile_size,
                      tex->surface.htile_alignment);
      }

      if (tex->surface.dcc_offset) {
         u_log_printf(log,
                      "  DCC: offset=%" PRIu64 ", size=%u, "
                      "alignment=%u, pitch_max=%u, num_dcc_levels=%u\n",
                      tex->surface.dcc_offset, tex->surface.dcc_size, tex->surface.dcc_alignment,
                      tex->surface.u.gfx9.display_dcc_pitch_max, tex->surface.num_dcc_levels);
      }

      if (tex->surface.u.gfx9.stencil_offset) {
         u_log_printf(log, "  Stencil: offset=%" PRIu64 ", swmode=%u, epitch=%u\n",
                      tex->surface.u.gfx9.stencil_offset, tex->surface.u.gfx9.stencil.swizzle_mode,
                      tex->surface.u.gfx9.stencil.epitch);
      }
      return;
   }

   u_log_printf(log,
                "  Layout: size=%" PRIu64 ", alignment=%u, bankw=%u, "
                "bankh=%u, nbanks=%u, mtilea=%u, tilesplit=%u, pipeconfig=%u, scanout=%u\n",
                tex->surface.surf_size, tex->surface.surf_alignment, tex->surface.u.legacy.bankw,
                tex->surface.u.legacy.bankh, tex->surface.u.legacy.num_banks,
                tex->surface.u.legacy.mtilea, tex->surface.u.legacy.tile_split,
                tex->surface.u.legacy.pipe_config, (tex->surface.flags & RADEON_SURF_SCANOUT) != 0);

   if (tex->surface.fmask_offset)
      u_log_printf(
         log,
         "  FMask: offset=%" PRIu64 ", size=%" PRIu64 ", alignment=%u, pitch_in_pixels=%u, "
         "bankh=%u, slice_tile_max=%u, tile_mode_index=%u\n",
         tex->surface.fmask_offset, tex->surface.fmask_size, tex->surface.fmask_alignment,
         tex->surface.u.legacy.fmask.pitch_in_pixels, tex->surface.u.legacy.fmask.bankh,
         tex->surface.u.legacy.fmask.slice_tile_max, tex->surface.u.legacy.fmask.tiling_index);

   if (tex->cmask_buffer)
      u_log_printf(log,
                   "  CMask: offset=%" PRIu64 ", size=%u, alignment=%u, "
                   "slice_tile_max=%u\n",
                   tex->surface.cmask_offset, tex->surface.cmask_size, tex->surface.cmask_alignment,
                   tex->surface.u.legacy.cmask_slice_tile_max);

   if (tex->surface.htile_offset)
      u_log_printf(log,
                   "  HTile: offset=%" PRIu64 ", size=%u, "
                   "alignment=%u, TC_compatible = %u\n",
                   tex->surface.htile_offset, tex->surface.htile_size, tex->surface.htile_alignment,
                   tex->tc_compatible_htile);

   if (tex->surface.dcc_offset) {
      u_log_printf(log, "  DCC: offset=%" PRIu64 ", size=%u, alignment=%u\n",
                   tex->surface.dcc_offset, tex->surface.dcc_size, tex->surface.dcc_alignment);
      for (i = 0; i <= tex->buffer.b.b.last_level; i++)
         u_log_printf(log,
                      "  DCCLevel[%i]: enabled=%u, offset=%u, "
                      "fast_clear_size=%u\n",
                      i, i < tex->surface.num_dcc_levels, tex->surface.u.legacy.level[i].dcc_offset,
                      tex->surface.u.legacy.level[i].dcc_fast_clear_size);
   }

   for (i = 0; i <= tex->buffer.b.b.last_level; i++)
      u_log_printf(log,
                   "  Level[%i]: offset=%" PRIu64 ", slice_size=%" PRIu64 ", "
                   "npix_x=%u, npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
                   "mode=%u, tiling_index = %u\n",
                   i, tex->surface.u.legacy.level[i].offset,
                   (uint64_t)tex->surface.u.legacy.level[i].slice_size_dw * 4,
                   u_minify(tex->buffer.b.b.width0, i), u_minify(tex->buffer.b.b.height0, i),
                   u_minify(tex->buffer.b.b.depth0, i), tex->surface.u.legacy.level[i].nblk_x,
                   tex->surface.u.legacy.level[i].nblk_y, tex->surface.u.legacy.level[i].mode,
                   tex->surface.u.legacy.tiling_index[i]);

   if (tex->surface.has_stencil) {
      u_log_printf(log, "  StencilLayout: tilesplit=%u\n",
                   tex->surface.u.legacy.stencil_tile_split);
      for (i = 0; i <= tex->buffer.b.b.last_level; i++) {
         u_log_printf(log,
                      "  StencilLevel[%i]: offset=%" PRIu64 ", "
                      "slice_size=%" PRIu64 ", npix_x=%u, "
                      "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
                      "mode=%u, tiling_index = %u\n",
                      i, tex->surface.u.legacy.stencil_level[i].offset,
                      (uint64_t)tex->surface.u.legacy.stencil_level[i].slice_size_dw * 4,
                      u_minify(tex->buffer.b.b.width0, i), u_minify(tex->buffer.b.b.height0, i),
                      u_minify(tex->buffer.b.b.depth0, i),
                      tex->surface.u.legacy.stencil_level[i].nblk_x,
                      tex->surface.u.legacy.stencil_level[i].nblk_y,
                      tex->surface.u.legacy.stencil_level[i].mode,
                      tex->surface.u.legacy.stencil_tiling_index[i]);
      }
   }
}

/**
 * Common function for si_texture_create and si_texture_from_handle.
 *
 * \param screen	screen
 * \param base		resource template
 * \param surface	radeon_surf
 * \param plane0	if a non-zero plane is being created, this is the first plane
 * \param imported_buf	from si_texture_from_handle
 * \param offset	offset for non-zero planes or imported buffers
 * \param alloc_size	the size to allocate if plane0 != NULL
 * \param alignment	alignment for the allocation
 */
static struct si_texture *si_texture_create_object(struct pipe_screen *screen,
                                                   const struct pipe_resource *base,
                                                   const struct radeon_surf *surface,
                                                   const struct si_texture *plane0,
                                                   struct pb_buffer *imported_buf,
                                                   uint64_t offset, unsigned pitch_in_bytes,
                                                   uint64_t alloc_size, unsigned alignment)
{
   struct si_texture *tex;
   struct si_resource *resource;
   struct si_screen *sscreen = (struct si_screen *)screen;

   tex = CALLOC_STRUCT(si_texture);
   if (!tex)
      goto error;

   resource = &tex->buffer;
   resource->b.b = *base;
   resource->b.vtbl = &si_texture_vtbl;
   pipe_reference_init(&resource->b.b.reference, 1);
   resource->b.b.screen = screen;

   /* don't include stencil-only formats which we don't support for rendering */
   tex->is_depth = util_format_has_depth(util_format_description(tex->buffer.b.b.format));
   tex->surface = *surface;

   /* On GFX8, HTILE uses different tiling depending on the TC_COMPATIBLE_HTILE
    * setting, so we have to enable it if we enabled it at allocation.
    *
    * GFX9 and later use the same tiling for both, so TC-compatible HTILE can be
    * enabled on demand.
    */
   tex->tc_compatible_htile = sscreen->info.chip_class == GFX8 &&
                              tex->surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE;

   /* TC-compatible HTILE:
    * - GFX8 only supports Z32_FLOAT.
    * - GFX9 only supports Z32_FLOAT and Z16_UNORM. */
   if (tex->surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE) {
      if (sscreen->info.chip_class >= GFX9 && base->format == PIPE_FORMAT_Z16_UNORM)
         tex->db_render_format = base->format;
      else {
         tex->db_render_format = PIPE_FORMAT_Z32_FLOAT;
         tex->upgraded_depth = base->format != PIPE_FORMAT_Z32_FLOAT &&
                               base->format != PIPE_FORMAT_Z32_FLOAT_S8X24_UINT;
      }
   } else {
      tex->db_render_format = base->format;
   }

   /* Applies to GCN. */
   tex->last_msaa_resolve_target_micro_mode = tex->surface.micro_tile_mode;

   /* Disable separate DCC at the beginning. DRI2 doesn't reuse buffers
    * between frames, so the only thing that can enable separate DCC
    * with DRI2 is multiple slow clears within a frame.
    */
   tex->ps_draw_ratio = 0;

   ac_surface_override_offset_stride(&sscreen->info, &tex->surface,
                                     tex->buffer.b.b.last_level + 1,
                                     offset, pitch_in_bytes / tex->surface.bpe);

   if (tex->is_depth) {
      if (sscreen->info.chip_class >= GFX9) {
         tex->can_sample_z = true;
         tex->can_sample_s = true;

         /* Stencil texturing with HTILE doesn't work
          * with mipmapping on Navi10-14. */
         if (sscreen->info.chip_class == GFX10 && base->last_level > 0)
            tex->htile_stencil_disabled = true;
      } else {
         tex->can_sample_z = !tex->surface.u.legacy.depth_adjusted;
         tex->can_sample_s = !tex->surface.u.legacy.stencil_adjusted;
      }

      tex->db_compatible = surface->flags & RADEON_SURF_ZBUFFER;
   } else {
      if (tex->surface.cmask_offset) {
         tex->cb_color_info |= S_028C70_FAST_CLEAR(1);
         tex->cmask_buffer = &tex->buffer;
      }
   }

   if (plane0) {
      /* The buffer is shared with the first plane. */
      resource->bo_size = plane0->buffer.bo_size;
      resource->bo_alignment = plane0->buffer.bo_alignment;
      resource->flags = plane0->buffer.flags;
      resource->domains = plane0->buffer.domains;
      resource->vram_usage = plane0->buffer.vram_usage;
      resource->gart_usage = plane0->buffer.gart_usage;

      pb_reference(&resource->buf, plane0->buffer.buf);
      resource->gpu_address = plane0->buffer.gpu_address;
   } else if (!(surface->flags & RADEON_SURF_IMPORTED)) {
      /* Create the backing buffer. */
      si_init_resource_fields(sscreen, resource, alloc_size, alignment);

      if (!si_alloc_resource(sscreen, resource))
         goto error;
   } else {
      resource->buf = imported_buf;
      resource->gpu_address = sscreen->ws->buffer_get_virtual_address(resource->buf);
      resource->bo_size = imported_buf->size;
      resource->bo_alignment = imported_buf->alignment;
      resource->domains = sscreen->ws->buffer_get_initial_domain(resource->buf);
      if (resource->domains & RADEON_DOMAIN_VRAM)
         resource->vram_usage = resource->bo_size;
      else if (resource->domains & RADEON_DOMAIN_GTT)
         resource->gart_usage = resource->bo_size;
      if (sscreen->ws->buffer_get_flags)
         resource->flags = sscreen->ws->buffer_get_flags(resource->buf);
   }

   if (tex->cmask_buffer) {
      /* Initialize the cmask to 0xCC (= compressed state). */
      si_screen_clear_buffer(sscreen, &tex->cmask_buffer->b.b, tex->surface.cmask_offset,
                             tex->surface.cmask_size, 0xCCCCCCCC);
   }
   if (tex->surface.htile_offset) {
      uint32_t clear_value = 0;

      if (sscreen->info.chip_class >= GFX9 || tex->tc_compatible_htile)
         clear_value = 0x0000030F;

      si_screen_clear_buffer(sscreen, &tex->buffer.b.b, tex->surface.htile_offset,
                             tex->surface.htile_size, clear_value);
   }

   /* Initialize DCC only if the texture is not being imported. */
   if (!(surface->flags & RADEON_SURF_IMPORTED) && tex->surface.dcc_offset) {
      /* Clear DCC to black for all tiles with DCC enabled.
       *
       * This fixes corruption in 3DMark Slingshot Extreme, which
       * uses uninitialized textures, causing corruption.
       */
      if (tex->surface.num_dcc_levels == tex->buffer.b.b.last_level + 1 &&
          tex->buffer.b.b.nr_samples <= 2) {
         /* Simple case - all tiles have DCC enabled. */
         si_screen_clear_buffer(sscreen, &tex->buffer.b.b, tex->surface.dcc_offset,
                                tex->surface.dcc_size, DCC_CLEAR_COLOR_0000);
      } else if (sscreen->info.chip_class >= GFX9) {
         /* Clear to uncompressed. Clearing this to black is complicated. */
         si_screen_clear_buffer(sscreen, &tex->buffer.b.b, tex->surface.dcc_offset,
                                tex->surface.dcc_size, DCC_UNCOMPRESSED);
      } else {
         /* GFX8: Initialize mipmap levels and multisamples separately. */
         if (tex->buffer.b.b.nr_samples >= 2) {
            /* Clearing this to black is complicated. */
            si_screen_clear_buffer(sscreen, &tex->buffer.b.b, tex->surface.dcc_offset,
                                   tex->surface.dcc_size, DCC_UNCOMPRESSED);
         } else {
            /* Clear the enabled mipmap levels to black. */
            unsigned size = 0;

            for (unsigned i = 0; i < tex->surface.num_dcc_levels; i++) {
               if (!tex->surface.u.legacy.level[i].dcc_fast_clear_size)
                  break;

               size = tex->surface.u.legacy.level[i].dcc_offset +
                      tex->surface.u.legacy.level[i].dcc_fast_clear_size;
            }

            /* Mipmap levels with DCC. */
            if (size) {
               si_screen_clear_buffer(sscreen, &tex->buffer.b.b, tex->surface.dcc_offset, size,
                                      DCC_CLEAR_COLOR_0000);
            }
            /* Mipmap levels without DCC. */
            if (size != tex->surface.dcc_size) {
               si_screen_clear_buffer(sscreen, &tex->buffer.b.b, tex->surface.dcc_offset + size,
                                      tex->surface.dcc_size - size, DCC_UNCOMPRESSED);
            }
         }
      }
   }

   /* Initialize displayable DCC that requires the retile blit. */
   if (tex->surface.display_dcc_offset) {
      /* Uninitialized DCC can hang the display hw.
       * Clear to white to indicate that. */
      si_screen_clear_buffer(sscreen, &tex->buffer.b.b, tex->surface.display_dcc_offset,
                             tex->surface.u.gfx9.display_dcc_size, DCC_CLEAR_COLOR_1111);

      /* Upload the DCC retile map.
       * Use a staging buffer for the upload, because
       * the buffer backing the texture is unmappable.
       */
      bool use_uint16 = tex->surface.u.gfx9.dcc_retile_use_uint16;
      unsigned num_elements = tex->surface.u.gfx9.dcc_retile_num_elements;
      unsigned dcc_retile_map_size = num_elements * (use_uint16 ? 2 : 4);

      tex->dcc_retile_buffer = si_aligned_buffer_create(screen, 0, PIPE_USAGE_DEFAULT,
                                                        dcc_retile_map_size,
                                                        sscreen->info.tcc_cache_line_size);
      struct si_resource *buf = si_aligned_buffer_create(screen, 0, PIPE_USAGE_STREAM,
                                                         dcc_retile_map_size,
                                                         sscreen->info.tcc_cache_line_size);
      void *map = sscreen->ws->buffer_map(buf->buf, NULL, PIPE_TRANSFER_WRITE);

      /* Upload the retile map into the staging buffer. */
      memcpy(map, tex->surface.u.gfx9.dcc_retile_map, dcc_retile_map_size);

      /* Copy the staging buffer to the buffer backing the texture. */
      struct si_context *sctx = (struct si_context *)sscreen->aux_context;

      simple_mtx_lock(&sscreen->aux_context_lock);
      si_sdma_copy_buffer(sctx, &tex->dcc_retile_buffer->b.b, &buf->b.b, 0,
                          0, buf->b.b.width0);
      sscreen->aux_context->flush(sscreen->aux_context, NULL, 0);
      simple_mtx_unlock(&sscreen->aux_context_lock);

      si_resource_reference(&buf, NULL);
   }

   /* Initialize the CMASK base register value. */
   tex->cmask_base_address_reg = (tex->buffer.gpu_address + tex->surface.cmask_offset) >> 8;

   if (sscreen->debug_flags & DBG(VM)) {
      fprintf(stderr,
              "VM start=0x%" PRIX64 "  end=0x%" PRIX64
              " | Texture %ix%ix%i, %i levels, %i samples, %s\n",
              tex->buffer.gpu_address, tex->buffer.gpu_address + tex->buffer.buf->size,
              base->width0, base->height0, util_num_layers(base, 0), base->last_level + 1,
              base->nr_samples ? base->nr_samples : 1, util_format_short_name(base->format));
   }

   if (sscreen->debug_flags & DBG(TEX)) {
      puts("Texture:");
      struct u_log_context log;
      u_log_context_init(&log);
      si_print_texture_info(sscreen, tex, &log);
      u_log_new_page_print(&log, stdout);
      fflush(stdout);
      u_log_context_destroy(&log);
   }

   return tex;

error:
   FREE(tex);
   return NULL;
}

static enum radeon_surf_mode si_choose_tiling(struct si_screen *sscreen,
                                              const struct pipe_resource *templ,
                                              bool tc_compatible_htile)
{
   const struct util_format_description *desc = util_format_description(templ->format);
   bool force_tiling = templ->flags & SI_RESOURCE_FLAG_FORCE_MSAA_TILING;
   bool is_depth_stencil = util_format_is_depth_or_stencil(templ->format) &&
                           !(templ->flags & SI_RESOURCE_FLAG_FLUSHED_DEPTH);

   /* MSAA resources must be 2D tiled. */
   if (templ->nr_samples > 1)
      return RADEON_SURF_MODE_2D;

   /* Transfer resources should be linear. */
   if (templ->flags & SI_RESOURCE_FLAG_FORCE_LINEAR)
      return RADEON_SURF_MODE_LINEAR_ALIGNED;

   /* Avoid Z/S decompress blits by forcing TC-compatible HTILE on GFX8,
    * which requires 2D tiling.
    */
   if (sscreen->info.chip_class == GFX8 && tc_compatible_htile)
      return RADEON_SURF_MODE_2D;

   /* Handle common candidates for the linear mode.
    * Compressed textures and DB surfaces must always be tiled.
    */
   if (!force_tiling && !is_depth_stencil && !util_format_is_compressed(templ->format)) {
      if (sscreen->debug_flags & DBG(NO_TILING))
         return RADEON_SURF_MODE_LINEAR_ALIGNED;

      /* Tiling doesn't work with the 422 (SUBSAMPLED) formats. */
      if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED)
         return RADEON_SURF_MODE_LINEAR_ALIGNED;

      /* Cursors are linear on AMD GCN.
       * (XXX double-check, maybe also use RADEON_SURF_SCANOUT) */
      if (templ->bind & PIPE_BIND_CURSOR)
         return RADEON_SURF_MODE_LINEAR_ALIGNED;

      if (templ->bind & PIPE_BIND_LINEAR)
         return RADEON_SURF_MODE_LINEAR_ALIGNED;

      /* Textures with a very small height are recommended to be linear. */
      if (templ->target == PIPE_TEXTURE_1D || templ->target == PIPE_TEXTURE_1D_ARRAY ||
          /* Only very thin and long 2D textures should benefit from
           * linear_aligned. */
          (templ->width0 > 8 && templ->height0 <= 2))
         return RADEON_SURF_MODE_LINEAR_ALIGNED;

      /* Textures likely to be mapped often. */
      if (templ->usage == PIPE_USAGE_STAGING || templ->usage == PIPE_USAGE_STREAM)
         return RADEON_SURF_MODE_LINEAR_ALIGNED;
   }

   /* Make small textures 1D tiled. */
   if (templ->width0 <= 16 || templ->height0 <= 16 || (sscreen->debug_flags & DBG(NO_2D_TILING)))
      return RADEON_SURF_MODE_1D;

   /* The allocator will switch to 1D if needed. */
   return RADEON_SURF_MODE_2D;
}

struct pipe_resource *si_texture_create(struct pipe_screen *screen,
                                        const struct pipe_resource *templ)
{
   struct si_screen *sscreen = (struct si_screen *)screen;
   bool is_zs = util_format_is_depth_or_stencil(templ->format);

   if (templ->nr_samples >= 2) {
      /* This is hackish (overwriting the const pipe_resource template),
       * but should be harmless and gallium frontends can also see
       * the overriden number of samples in the created pipe_resource.
       */
      if (is_zs && sscreen->eqaa_force_z_samples) {
         ((struct pipe_resource *)templ)->nr_samples =
            ((struct pipe_resource *)templ)->nr_storage_samples = sscreen->eqaa_force_z_samples;
      } else if (!is_zs && sscreen->eqaa_force_color_samples) {
         ((struct pipe_resource *)templ)->nr_samples = sscreen->eqaa_force_coverage_samples;
         ((struct pipe_resource *)templ)->nr_storage_samples = sscreen->eqaa_force_color_samples;
      }
   }

   bool is_flushed_depth = templ->flags & SI_RESOURCE_FLAG_FLUSHED_DEPTH ||
                           templ->flags & SI_RESOURCE_FLAG_FORCE_LINEAR;
   bool tc_compatible_htile =
      sscreen->info.chip_class >= GFX8 &&
      /* There are issues with TC-compatible HTILE on Tonga (and
       * Iceland is the same design), and documented bug workarounds
       * don't help. For example, this fails:
       *   piglit/bin/tex-miplevel-selection 'texture()' 2DShadow -auto
       */
      sscreen->info.family != CHIP_TONGA && sscreen->info.family != CHIP_ICELAND &&
      (templ->flags & PIPE_RESOURCE_FLAG_TEXTURING_MORE_LIKELY) &&
      !(sscreen->debug_flags & DBG(NO_HYPERZ)) && !is_flushed_depth &&
      templ->nr_samples <= 1 && /* TC-compat HTILE is less efficient with MSAA */
      is_zs;
   enum radeon_surf_mode tile_mode = si_choose_tiling(sscreen, templ, tc_compatible_htile);

   /* This allocates textures with multiple planes like NV12 in 1 buffer. */
   enum
   {
      SI_TEXTURE_MAX_PLANES = 3
   };
   struct radeon_surf surface[SI_TEXTURE_MAX_PLANES] = {};
   struct pipe_resource plane_templ[SI_TEXTURE_MAX_PLANES];
   uint64_t plane_offset[SI_TEXTURE_MAX_PLANES] = {};
   uint64_t total_size = 0;
   unsigned max_alignment = 0;
   unsigned num_planes = util_format_get_num_planes(templ->format);
   assert(num_planes <= SI_TEXTURE_MAX_PLANES);

   /* Compute texture or plane layouts and offsets. */
   for (unsigned i = 0; i < num_planes; i++) {
      plane_templ[i] = *templ;
      plane_templ[i].format = util_format_get_plane_format(templ->format, i);
      plane_templ[i].width0 = util_format_get_plane_width(templ->format, i, templ->width0);
      plane_templ[i].height0 = util_format_get_plane_height(templ->format, i, templ->height0);

      /* Multi-plane allocations need PIPE_BIND_SHARED, because we can't
       * reallocate the storage to add PIPE_BIND_SHARED, because it's
       * shared by 3 pipe_resources.
       */
      if (num_planes > 1)
         plane_templ[i].bind |= PIPE_BIND_SHARED;

      if (si_init_surface(sscreen, &surface[i], &plane_templ[i], tile_mode, false,
                          plane_templ[i].bind & PIPE_BIND_SCANOUT, is_flushed_depth,
                          tc_compatible_htile))
         return NULL;

      plane_offset[i] = align64(total_size, surface[i].surf_alignment);
      total_size = plane_offset[i] + surface[i].total_size;
      max_alignment = MAX2(max_alignment, surface[i].surf_alignment);
   }

   struct si_texture *plane0 = NULL, *last_plane = NULL;

   for (unsigned i = 0; i < num_planes; i++) {
      struct si_texture *tex =
         si_texture_create_object(screen, &plane_templ[i], &surface[i], plane0, NULL,
                                  plane_offset[i], 0, total_size, max_alignment);
      if (!tex) {
         si_texture_reference(&plane0, NULL);
         return NULL;
      }

      tex->plane_index = i;
      tex->num_planes = num_planes;

      if (!plane0) {
         plane0 = last_plane = tex;
      } else {
         last_plane->buffer.b.b.next = &tex->buffer.b.b;
         last_plane = tex;
      }
   }

   return (struct pipe_resource *)plane0;
}

static struct pipe_resource *si_texture_from_winsys_buffer(struct si_screen *sscreen,
                                                           const struct pipe_resource *templ,
                                                           struct pb_buffer *buf, unsigned stride,
                                                           uint64_t offset, unsigned usage,
                                                           bool dedicated)
{
   struct radeon_surf surface = {};
   struct radeon_bo_metadata metadata = {};
   struct si_texture *tex;
   int r;

   /* Ignore metadata for non-zero planes. */
   if (offset != 0)
      dedicated = false;

   if (dedicated) {
      sscreen->ws->buffer_get_metadata(buf, &metadata, &surface);
   } else {
      /**
       * The bo metadata is unset for un-dedicated images. So we fall
       * back to linear. See answer to question 5 of the
       * VK_KHX_external_memory spec for some details.
       *
       * It is possible that this case isn't going to work if the
       * surface pitch isn't correctly aligned by default.
       *
       * In order to support it correctly we require multi-image
       * metadata to be syncrhonized between radv and radeonsi. The
       * semantics of associating multiple image metadata to a memory
       * object on the vulkan export side are not concretely defined
       * either.
       *
       * All the use cases we are aware of at the moment for memory
       * objects use dedicated allocations. So lets keep the initial
       * implementation simple.
       *
       * A possible alternative is to attempt to reconstruct the
       * tiling information when the TexParameter TEXTURE_TILING_EXT
       * is set.
       */
      metadata.mode = RADEON_SURF_MODE_LINEAR_ALIGNED;
   }

   r = si_init_surface(sscreen, &surface, templ, metadata.mode, true,
                       surface.flags & RADEON_SURF_SCANOUT, false, false);
   if (r)
      return NULL;

   tex = si_texture_create_object(&sscreen->b, templ, &surface, NULL, buf,
                                  offset, stride, 0, 0);
   if (!tex)
      return NULL;

   tex->buffer.b.is_shared = true;
   tex->buffer.external_usage = usage;
   tex->num_planes = 1;

   /* Account for multiple planes with lowered yuv import. */
   struct pipe_resource *next_plane = tex->buffer.b.b.next;
   while(next_plane) {
      struct si_texture *next_tex = (struct si_texture *)next_plane;
      ++next_tex->num_planes;
      ++tex->num_planes;
      next_plane = next_plane->next;
   }

   if (!ac_surface_set_umd_metadata(&sscreen->info, &tex->surface,
                                    tex->buffer.b.b.nr_storage_samples,
                                    tex->buffer.b.b.last_level + 1,
                                    metadata.size_metadata,
                                    metadata.metadata)) {
      si_texture_reference(&tex, NULL);
      return NULL;
   }

   /* Displayable DCC requires an explicit flush. */
   if (dedicated && offset == 0 && !(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) &&
       si_has_displayable_dcc(tex)) {
      /* TODO: do we need to decompress DCC? */
      if (si_texture_discard_dcc(sscreen, tex)) {
         /* Update BO metadata after disabling DCC. */
         si_set_tex_bo_metadata(sscreen, tex);
      }
   }

   assert(tex->surface.tile_swizzle == 0);
   return &tex->buffer.b.b;
}

static struct pipe_resource *si_texture_from_handle(struct pipe_screen *screen,
                                                    const struct pipe_resource *templ,
                                                    struct winsys_handle *whandle, unsigned usage)
{
   struct si_screen *sscreen = (struct si_screen *)screen;
   struct pb_buffer *buf = NULL;

   /* Support only 2D textures without mipmaps */
   if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT &&
        templ->target != PIPE_TEXTURE_2D_ARRAY) ||
       templ->last_level != 0)
      return NULL;

   buf = sscreen->ws->buffer_from_handle(sscreen->ws, whandle, sscreen->info.max_alignment);
   if (!buf)
      return NULL;

   return si_texture_from_winsys_buffer(sscreen, templ, buf, whandle->stride, whandle->offset,
                                        usage, true);
}

bool si_init_flushed_depth_texture(struct pipe_context *ctx, struct pipe_resource *texture)
{
   struct si_texture *tex = (struct si_texture *)texture;
   struct pipe_resource resource;
   enum pipe_format pipe_format = texture->format;

   assert(!tex->flushed_depth_texture);

   if (!tex->can_sample_z && tex->can_sample_s) {
      switch (pipe_format) {
      case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
         /* Save memory by not allocating the S plane. */
         pipe_format = PIPE_FORMAT_Z32_FLOAT;
         break;
      case PIPE_FORMAT_Z24_UNORM_S8_UINT:
      case PIPE_FORMAT_S8_UINT_Z24_UNORM:
         /* Save memory bandwidth by not copying the
          * stencil part during flush.
          *
          * This potentially increases memory bandwidth
          * if an application uses both Z and S texturing
          * simultaneously (a flushed Z24S8 texture
          * would be stored compactly), but how often
          * does that really happen?
          */
         pipe_format = PIPE_FORMAT_Z24X8_UNORM;
         break;
      default:;
      }
   } else if (!tex->can_sample_s && tex->can_sample_z) {
      assert(util_format_has_stencil(util_format_description(pipe_format)));

      /* DB->CB copies to an 8bpp surface don't work. */
      pipe_format = PIPE_FORMAT_X24S8_UINT;
   }

   memset(&resource, 0, sizeof(resource));
   resource.target = texture->target;
   resource.format = pipe_format;
   resource.width0 = texture->width0;
   resource.height0 = texture->height0;
   resource.depth0 = texture->depth0;
   resource.array_size = texture->array_size;
   resource.last_level = texture->last_level;
   resource.nr_samples = texture->nr_samples;
   resource.usage = PIPE_USAGE_DEFAULT;
   resource.bind = texture->bind & ~PIPE_BIND_DEPTH_STENCIL;
   resource.flags = texture->flags | SI_RESOURCE_FLAG_FLUSHED_DEPTH;

   tex->flushed_depth_texture =
      (struct si_texture *)ctx->screen->resource_create(ctx->screen, &resource);
   if (!tex->flushed_depth_texture) {
      PRINT_ERR("failed to create temporary texture to hold flushed depth\n");
      return false;
   }
   return true;
}

/**
 * Initialize the pipe_resource descriptor to be of the same size as the box,
 * which is supposed to hold a subregion of the texture "orig" at the given
 * mipmap level.
 */
static void si_init_temp_resource_from_box(struct pipe_resource *res, struct pipe_resource *orig,
                                           const struct pipe_box *box, unsigned level,
                                           unsigned usage, unsigned flags)
{
   memset(res, 0, sizeof(*res));
   res->format = orig->format;
   res->width0 = box->width;
   res->height0 = box->height;
   res->depth0 = 1;
   res->array_size = 1;
   res->usage = usage;
   res->flags = flags;

   if (flags & SI_RESOURCE_FLAG_FORCE_LINEAR && util_format_is_compressed(orig->format)) {
      /* Transfer resources are allocated with linear tiling, which is
       * not supported for compressed formats.
       */
      unsigned blocksize = util_format_get_blocksize(orig->format);

      if (blocksize == 8) {
         res->format = PIPE_FORMAT_R16G16B16A16_UINT;
      } else {
         assert(blocksize == 16);
         res->format = PIPE_FORMAT_R32G32B32A32_UINT;
      }

      res->width0 = util_format_get_nblocksx(orig->format, box->width);
      res->height0 = util_format_get_nblocksy(orig->format, box->height);
   }

   /* We must set the correct texture target and dimensions for a 3D box. */
   if (box->depth > 1 && util_max_layer(orig, level) > 0) {
      res->target = PIPE_TEXTURE_2D_ARRAY;
      res->array_size = box->depth;
   } else {
      res->target = PIPE_TEXTURE_2D;
   }
}

static bool si_can_invalidate_texture(struct si_screen *sscreen, struct si_texture *tex,
                                      unsigned transfer_usage, const struct pipe_box *box)
{
   return !tex->buffer.b.is_shared && !(tex->surface.flags & RADEON_SURF_IMPORTED) &&
          !(transfer_usage & PIPE_TRANSFER_READ) && tex->buffer.b.b.last_level == 0 &&
          util_texrange_covers_whole_level(&tex->buffer.b.b, 0, box->x, box->y, box->z, box->width,
                                           box->height, box->depth);
}

static void si_texture_invalidate_storage(struct si_context *sctx, struct si_texture *tex)
{
   struct si_screen *sscreen = sctx->screen;

   /* There is no point in discarding depth and tiled buffers. */
   assert(!tex->is_depth);
   assert(tex->surface.is_linear);

   /* Reallocate the buffer in the same pipe_resource. */
   si_alloc_resource(sscreen, &tex->buffer);

   /* Initialize the CMASK base address (needed even without CMASK). */
   tex->cmask_base_address_reg = (tex->buffer.gpu_address + tex->surface.cmask_offset) >> 8;

   p_atomic_inc(&sscreen->dirty_tex_counter);

   sctx->num_alloc_tex_transfer_bytes += tex->surface.total_size;
}

static void *si_texture_transfer_map(struct pipe_context *ctx, struct pipe_resource *texture,
                                     unsigned level, unsigned usage, const struct pipe_box *box,
                                     struct pipe_transfer **ptransfer)
{
   struct si_context *sctx = (struct si_context *)ctx;
   struct si_texture *tex = (struct si_texture *)texture;
   struct si_transfer *trans;
   struct si_resource *buf;
   unsigned offset = 0;
   char *map;
   bool use_staging_texture = false;

   assert(!(texture->flags & SI_RESOURCE_FLAG_FORCE_LINEAR));
   assert(box->width && box->height && box->depth);

   if (tex->is_depth) {
      /* Depth textures use staging unconditionally. */
      use_staging_texture = true;
   } else {
      /* Degrade the tile mode if we get too many transfers on APUs.
       * On dGPUs, the staging texture is always faster.
       * Only count uploads that are at least 4x4 pixels large.
       */
      if (!sctx->screen->info.has_dedicated_vram && level == 0 && box->width >= 4 &&
          box->height >= 4 && p_atomic_inc_return(&tex->num_level0_transfers) == 10) {
         bool can_invalidate = si_can_invalidate_texture(sctx->screen, tex, usage, box);

         si_reallocate_texture_inplace(sctx, tex, PIPE_BIND_LINEAR, can_invalidate);
      }

      /* Tiled textures need to be converted into a linear texture for CPU
       * access. The staging texture is always linear and is placed in GART.
       *
       * Reading from VRAM or GTT WC is slow, always use the staging
       * texture in this case.
       *
       * Use the staging texture for uploads if the underlying BO
       * is busy.
       */
      if (!tex->surface.is_linear || (tex->buffer.flags & RADEON_FLAG_ENCRYPTED))
         use_staging_texture = true;
      else if (usage & PIPE_TRANSFER_READ)
         use_staging_texture =
            tex->buffer.domains & RADEON_DOMAIN_VRAM || tex->buffer.flags & RADEON_FLAG_GTT_WC;
      /* Write & linear only: */
      else if (si_rings_is_buffer_referenced(sctx, tex->buffer.buf, RADEON_USAGE_READWRITE) ||
               !sctx->ws->buffer_wait(tex->buffer.buf, 0, RADEON_USAGE_READWRITE)) {
         /* It's busy. */
         if (si_can_invalidate_texture(sctx->screen, tex, usage, box))
            si_texture_invalidate_storage(sctx, tex);
         else
            use_staging_texture = true;
      }
   }

   trans = CALLOC_STRUCT(si_transfer);
   if (!trans)
      return NULL;
   pipe_resource_reference(&trans->b.b.resource, texture);
   trans->b.b.level = level;
   trans->b.b.usage = usage;
   trans->b.b.box = *box;

   if (use_staging_texture) {
      struct pipe_resource resource;
      struct si_texture *staging;
      unsigned bo_usage = usage & PIPE_TRANSFER_READ ? PIPE_USAGE_STAGING : PIPE_USAGE_STREAM;
      unsigned bo_flags = SI_RESOURCE_FLAG_FORCE_LINEAR;

      /* The pixel shader has a bad access pattern for linear textures.
       * If a pixel shader is used to blit to/from staging, don't disable caches.
       *
       * MSAA, depth/stencil textures, and compressed textures use the pixel shader
       * to blit.
       */
      if (texture->nr_samples <= 1 &&
          !tex->is_depth &&
          !util_format_is_compressed(texture->format) &&
          /* Texture uploads with DCC use the pixel shader to blit */
          (!(usage & PIPE_TRANSFER_WRITE) || !vi_dcc_enabled(tex, level)))
         bo_flags |= SI_RESOURCE_FLAG_UNCACHED;

      si_init_temp_resource_from_box(&resource, texture, box, level, bo_usage,
                                     bo_flags);

      /* Since depth-stencil textures don't support linear tiling,
       * blit from ZS to color and vice versa. u_blitter will do
       * the packing for these formats.
       */
      if (tex->is_depth)
         resource.format = util_blitter_get_color_format_for_zs(resource.format);

      /* Create the temporary texture. */
      staging = (struct si_texture *)ctx->screen->resource_create(ctx->screen, &resource);
      if (!staging) {
         PRINT_ERR("failed to create temporary texture to hold untiled copy\n");
         goto fail_trans;
      }
      trans->staging = &staging->buffer;

      /* Just get the strides. */
      si_texture_get_offset(sctx->screen, staging, 0, NULL, &trans->b.b.stride,
                            &trans->b.b.layer_stride);

      if (usage & PIPE_TRANSFER_READ)
         si_copy_to_staging_texture(ctx, trans);
      else
         usage |= PIPE_TRANSFER_UNSYNCHRONIZED;

      buf = trans->staging;
   } else {
      /* the resource is mapped directly */
      offset = si_texture_get_offset(sctx->screen, tex, level, box, &trans->b.b.stride,
                                     &trans->b.b.layer_stride);
      buf = &tex->buffer;
   }

   /* Always unmap texture CPU mappings on 32-bit architectures, so that
    * we don't run out of the CPU address space.
    */
   if (sizeof(void *) == 4)
      usage |= RADEON_TRANSFER_TEMPORARY;

   if (!(map = si_buffer_map_sync_with_rings(sctx, buf, usage)))
      goto fail_trans;

   *ptransfer = &trans->b.b;
   return map + offset;

fail_trans:
   si_resource_reference(&trans->staging, NULL);
   pipe_resource_reference(&trans->b.b.resource, NULL);
   FREE(trans);
   return NULL;
}

static void si_texture_transfer_unmap(struct pipe_context *ctx, struct pipe_transfer *transfer)
{
   struct si_context *sctx = (struct si_context *)ctx;
   struct si_transfer *stransfer = (struct si_transfer *)transfer;
   struct pipe_resource *texture = transfer->resource;
   struct si_texture *tex = (struct si_texture *)texture;

   /* Always unmap texture CPU mappings on 32-bit architectures, so that
    * we don't run out of the CPU address space.
    */
   if (sizeof(void *) == 4) {
      struct si_resource *buf = stransfer->staging ? stransfer->staging : &tex->buffer;

      sctx->ws->buffer_unmap(buf->buf);
   }

   if ((transfer->usage & PIPE_TRANSFER_WRITE) && stransfer->staging)
      si_copy_from_staging_texture(ctx, stransfer);

   if (stransfer->staging) {
      sctx->num_alloc_tex_transfer_bytes += stransfer->staging->buf->size;
      si_resource_reference(&stransfer->staging, NULL);
   }

   /* Heuristic for {upload, draw, upload, draw, ..}:
    *
    * Flush the gfx IB if we've allocated too much texture storage.
    *
    * The idea is that we don't want to build IBs that use too much
    * memory and put pressure on the kernel memory manager and we also
    * want to make temporary and invalidated buffers go idle ASAP to
    * decrease the total memory usage or make them reusable. The memory
    * usage will be slightly higher than given here because of the buffer
    * cache in the winsys.
    *
    * The result is that the kernel memory manager is never a bottleneck.
    */
   if (sctx->num_alloc_tex_transfer_bytes > sctx->screen->info.gart_size / 4) {
      si_flush_gfx_cs(sctx, RADEON_FLUSH_ASYNC_START_NEXT_GFX_IB_NOW, NULL);
      sctx->num_alloc_tex_transfer_bytes = 0;
   }

   pipe_resource_reference(&transfer->resource, NULL);
   FREE(transfer);
}

static const struct u_resource_vtbl si_texture_vtbl = {
   NULL,                            /* get_handle */
   si_texture_destroy,              /* resource_destroy */
   si_texture_transfer_map,         /* transfer_map */
   u_default_transfer_flush_region, /* transfer_flush_region */
   si_texture_transfer_unmap,       /* transfer_unmap */
};

/* Return if it's allowed to reinterpret one format as another with DCC enabled.
 */
bool vi_dcc_formats_compatible(struct si_screen *sscreen, enum pipe_format format1,
                               enum pipe_format format2)
{
   const struct util_format_description *desc1, *desc2;

   /* No format change - exit early. */
   if (format1 == format2)
      return true;

   format1 = si_simplify_cb_format(format1);
   format2 = si_simplify_cb_format(format2);

   /* Check again after format adjustments. */
   if (format1 == format2)
      return true;

   desc1 = util_format_description(format1);
   desc2 = util_format_description(format2);

   if (desc1->layout != UTIL_FORMAT_LAYOUT_PLAIN || desc2->layout != UTIL_FORMAT_LAYOUT_PLAIN)
      return false;

   /* Float and non-float are totally incompatible. */
   if ((desc1->channel[0].type == UTIL_FORMAT_TYPE_FLOAT) !=
       (desc2->channel[0].type == UTIL_FORMAT_TYPE_FLOAT))
      return false;

   /* Channel sizes must match across DCC formats.
    * Comparing just the first 2 channels should be enough.
    */
   if (desc1->channel[0].size != desc2->channel[0].size ||
       (desc1->nr_channels >= 2 && desc1->channel[1].size != desc2->channel[1].size))
      return false;

   /* Everything below is not needed if the driver never uses the DCC
    * clear code with the value of 1.
    */

   /* If the clear values are all 1 or all 0, this constraint can be
    * ignored. */
   if (vi_alpha_is_on_msb(sscreen, format1) != vi_alpha_is_on_msb(sscreen, format2))
      return false;

   /* Channel types must match if the clear value of 1 is used.
    * The type categories are only float, signed, unsigned.
    * NORM and INT are always compatible.
    */
   if (desc1->channel[0].type != desc2->channel[0].type ||
       (desc1->nr_channels >= 2 && desc1->channel[1].type != desc2->channel[1].type))
      return false;

   return true;
}

bool vi_dcc_formats_are_incompatible(struct pipe_resource *tex, unsigned level,
                                     enum pipe_format view_format)
{
   struct si_texture *stex = (struct si_texture *)tex;

   return vi_dcc_enabled(stex, level) &&
          !vi_dcc_formats_compatible((struct si_screen *)tex->screen, tex->format, view_format);
}

/* This can't be merged with the above function, because
 * vi_dcc_formats_compatible should be called only when DCC is enabled. */
void vi_disable_dcc_if_incompatible_format(struct si_context *sctx, struct pipe_resource *tex,
                                           unsigned level, enum pipe_format view_format)
{
   struct si_texture *stex = (struct si_texture *)tex;

   if (vi_dcc_formats_are_incompatible(tex, level, view_format))
      if (!si_texture_disable_dcc(sctx, stex))
         si_decompress_dcc(sctx, stex);
}

struct pipe_surface *si_create_surface_custom(struct pipe_context *pipe,
                                              struct pipe_resource *texture,
                                              const struct pipe_surface *templ, unsigned width0,
                                              unsigned height0, unsigned width, unsigned height)
{
   struct si_surface *surface = CALLOC_STRUCT(si_surface);

   if (!surface)
      return NULL;

   assert(templ->u.tex.first_layer <= util_max_layer(texture, templ->u.tex.level));
   assert(templ->u.tex.last_layer <= util_max_layer(texture, templ->u.tex.level));

   pipe_reference_init(&surface->base.reference, 1);
   pipe_resource_reference(&surface->base.texture, texture);
   surface->base.context = pipe;
   surface->base.format = templ->format;
   surface->base.width = width;
   surface->base.height = height;
   surface->base.u = templ->u;

   surface->width0 = width0;
   surface->height0 = height0;

   surface->dcc_incompatible =
      texture->target != PIPE_BUFFER &&
      vi_dcc_formats_are_incompatible(texture, templ->u.tex.level, templ->format);
   return &surface->base;
}

static struct pipe_surface *si_create_surface(struct pipe_context *pipe, struct pipe_resource *tex,
                                              const struct pipe_surface *templ)
{
   unsigned level = templ->u.tex.level;
   unsigned width = u_minify(tex->width0, level);
   unsigned height = u_minify(tex->height0, level);
   unsigned width0 = tex->width0;
   unsigned height0 = tex->height0;

   if (tex->target != PIPE_BUFFER && templ->format != tex->format) {
      const struct util_format_description *tex_desc = util_format_description(tex->format);
      const struct util_format_description *templ_desc = util_format_description(templ->format);

      assert(tex_desc->block.bits == templ_desc->block.bits);

      /* Adjust size of surface if and only if the block width or
       * height is changed. */
      if (tex_desc->block.width != templ_desc->block.width ||
          tex_desc->block.height != templ_desc->block.height) {
         unsigned nblks_x = util_format_get_nblocksx(tex->format, width);
         unsigned nblks_y = util_format_get_nblocksy(tex->format, height);

         width = nblks_x * templ_desc->block.width;
         height = nblks_y * templ_desc->block.height;

         width0 = util_format_get_nblocksx(tex->format, width0);
         height0 = util_format_get_nblocksy(tex->format, height0);
      }
   }

   return si_create_surface_custom(pipe, tex, templ, width0, height0, width, height);
}

static void si_surface_destroy(struct pipe_context *pipe, struct pipe_surface *surface)
{
   pipe_resource_reference(&surface->texture, NULL);
   FREE(surface);
}

unsigned si_translate_colorswap(enum pipe_format format, bool do_endian_swap)
{
   const struct util_format_description *desc = util_format_description(format);

#define HAS_SWIZZLE(chan, swz) (desc->swizzle[chan] == PIPE_SWIZZLE_##swz)

   if (format == PIPE_FORMAT_R11G11B10_FLOAT) /* isn't plain */
      return V_028C70_SWAP_STD;

   if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
      return ~0U;

   switch (desc->nr_channels) {
   case 1:
      if (HAS_SWIZZLE(0, X))
         return V_028C70_SWAP_STD; /* X___ */
      else if (HAS_SWIZZLE(3, X))
         return V_028C70_SWAP_ALT_REV; /* ___X */
      break;
   case 2:
      if ((HAS_SWIZZLE(0, X) && HAS_SWIZZLE(1, Y)) || (HAS_SWIZZLE(0, X) && HAS_SWIZZLE(1, NONE)) ||
          (HAS_SWIZZLE(0, NONE) && HAS_SWIZZLE(1, Y)))
         return V_028C70_SWAP_STD; /* XY__ */
      else if ((HAS_SWIZZLE(0, Y) && HAS_SWIZZLE(1, X)) ||
               (HAS_SWIZZLE(0, Y) && HAS_SWIZZLE(1, NONE)) ||
               (HAS_SWIZZLE(0, NONE) && HAS_SWIZZLE(1, X)))
         /* YX__ */
         return (do_endian_swap ? V_028C70_SWAP_STD : V_028C70_SWAP_STD_REV);
      else if (HAS_SWIZZLE(0, X) && HAS_SWIZZLE(3, Y))
         return V_028C70_SWAP_ALT; /* X__Y */
      else if (HAS_SWIZZLE(0, Y) && HAS_SWIZZLE(3, X))
         return V_028C70_SWAP_ALT_REV; /* Y__X */
      break;
   case 3:
      if (HAS_SWIZZLE(0, X))
         return (do_endian_swap ? V_028C70_SWAP_STD_REV : V_028C70_SWAP_STD);
      else if (HAS_SWIZZLE(0, Z))
         return V_028C70_SWAP_STD_REV; /* ZYX */
      break;
   case 4:
      /* check the middle channels, the 1st and 4th channel can be NONE */
      if (HAS_SWIZZLE(1, Y) && HAS_SWIZZLE(2, Z)) {
         return V_028C70_SWAP_STD; /* XYZW */
      } else if (HAS_SWIZZLE(1, Z) && HAS_SWIZZLE(2, Y)) {
         return V_028C70_SWAP_STD_REV; /* WZYX */
      } else if (HAS_SWIZZLE(1, Y) && HAS_SWIZZLE(2, X)) {
         return V_028C70_SWAP_ALT; /* ZYXW */
      } else if (HAS_SWIZZLE(1, Z) && HAS_SWIZZLE(2, W)) {
         /* YZWX */
         if (desc->is_array)
            return V_028C70_SWAP_ALT_REV;
         else
            return (do_endian_swap ? V_028C70_SWAP_ALT : V_028C70_SWAP_ALT_REV);
      }
      break;
   }
   return ~0U;
}

/* PIPELINE_STAT-BASED DCC ENABLEMENT FOR DISPLAYABLE SURFACES */

static void vi_dcc_clean_up_context_slot(struct si_context *sctx, int slot)
{
   int i;

   if (sctx->dcc_stats[slot].query_active)
      vi_separate_dcc_stop_query(sctx, sctx->dcc_stats[slot].tex);

   for (i = 0; i < ARRAY_SIZE(sctx->dcc_stats[slot].ps_stats); i++)
      if (sctx->dcc_stats[slot].ps_stats[i]) {
         sctx->b.destroy_query(&sctx->b, sctx->dcc_stats[slot].ps_stats[i]);
         sctx->dcc_stats[slot].ps_stats[i] = NULL;
      }

   si_texture_reference(&sctx->dcc_stats[slot].tex, NULL);
}

/**
 * Return the per-context slot where DCC statistics queries for the texture live.
 */
static unsigned vi_get_context_dcc_stats_index(struct si_context *sctx, struct si_texture *tex)
{
   int i, empty_slot = -1;

   /* Remove zombie textures (textures kept alive by this array only). */
   for (i = 0; i < ARRAY_SIZE(sctx->dcc_stats); i++)
      if (sctx->dcc_stats[i].tex && sctx->dcc_stats[i].tex->buffer.b.b.reference.count == 1)
         vi_dcc_clean_up_context_slot(sctx, i);

   /* Find the texture. */
   for (i = 0; i < ARRAY_SIZE(sctx->dcc_stats); i++) {
      /* Return if found. */
      if (sctx->dcc_stats[i].tex == tex) {
         sctx->dcc_stats[i].last_use_timestamp = os_time_get();
         return i;
      }

      /* Record the first seen empty slot. */
      if (empty_slot == -1 && !sctx->dcc_stats[i].tex)
         empty_slot = i;
   }

   /* Not found. Remove the oldest member to make space in the array. */
   if (empty_slot == -1) {
      int oldest_slot = 0;

      /* Find the oldest slot. */
      for (i = 1; i < ARRAY_SIZE(sctx->dcc_stats); i++)
         if (sctx->dcc_stats[oldest_slot].last_use_timestamp >
             sctx->dcc_stats[i].last_use_timestamp)
            oldest_slot = i;

      /* Clean up the oldest slot. */
      vi_dcc_clean_up_context_slot(sctx, oldest_slot);
      empty_slot = oldest_slot;
   }

   /* Add the texture to the new slot. */
   si_texture_reference(&sctx->dcc_stats[empty_slot].tex, tex);
   sctx->dcc_stats[empty_slot].last_use_timestamp = os_time_get();
   return empty_slot;
}

static struct pipe_query *vi_create_resuming_pipestats_query(struct si_context *sctx)
{
   struct si_query_hw *query =
      (struct si_query_hw *)sctx->b.create_query(&sctx->b, PIPE_QUERY_PIPELINE_STATISTICS, 0);

   query->flags |= SI_QUERY_HW_FLAG_BEGIN_RESUMES;
   return (struct pipe_query *)query;
}

/**
 * Called when binding a color buffer.
 */
void vi_separate_dcc_start_query(struct si_context *sctx, struct si_texture *tex)
{
   unsigned i = vi_get_context_dcc_stats_index(sctx, tex);

   assert(!sctx->dcc_stats[i].query_active);

   if (!sctx->dcc_stats[i].ps_stats[0])
      sctx->dcc_stats[i].ps_stats[0] = vi_create_resuming_pipestats_query(sctx);

   /* begin or resume the query */
   sctx->b.begin_query(&sctx->b, sctx->dcc_stats[i].ps_stats[0]);
   sctx->dcc_stats[i].query_active = true;
}

/**
 * Called when unbinding a color buffer.
 */
void vi_separate_dcc_stop_query(struct si_context *sctx, struct si_texture *tex)
{
   unsigned i = vi_get_context_dcc_stats_index(sctx, tex);

   assert(sctx->dcc_stats[i].query_active);
   assert(sctx->dcc_stats[i].ps_stats[0]);

   /* pause or end the query */
   sctx->b.end_query(&sctx->b, sctx->dcc_stats[i].ps_stats[0]);
   sctx->dcc_stats[i].query_active = false;
}

static bool vi_should_enable_separate_dcc(struct si_texture *tex)
{
   /* The minimum number of fullscreen draws per frame that is required
    * to enable DCC. */
   return tex->ps_draw_ratio + tex->num_slow_clears >= 5;
}

/* Called by fast clear. */
void vi_separate_dcc_try_enable(struct si_context *sctx, struct si_texture *tex)
{
   /* The intent is to use this with shared displayable back buffers,
    * but it's not strictly limited only to them.
    */
   if (!tex->buffer.b.is_shared ||
       !(tex->buffer.external_usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) ||
       tex->buffer.b.b.target != PIPE_TEXTURE_2D || tex->buffer.b.b.last_level > 0 ||
       !tex->surface.dcc_size || sctx->screen->debug_flags & DBG(NO_DCC) ||
       sctx->screen->debug_flags & DBG(NO_DCC_FB))
      return;

   assert(sctx->chip_class >= GFX8);

   if (tex->surface.dcc_offset)
      return; /* already enabled */

   /* Enable the DCC stat gathering. */
   if (!tex->dcc_gather_statistics) {
      tex->dcc_gather_statistics = true;
      vi_separate_dcc_start_query(sctx, tex);
   }

   if (!vi_should_enable_separate_dcc(tex))
      return; /* stats show that DCC decompression is too expensive */

   assert(tex->surface.num_dcc_levels);
   assert(!tex->dcc_separate_buffer);

   si_texture_discard_cmask(sctx->screen, tex);

   /* Get a DCC buffer. */
   if (tex->last_dcc_separate_buffer) {
      assert(tex->dcc_gather_statistics);
      assert(!tex->dcc_separate_buffer);
      tex->dcc_separate_buffer = tex->last_dcc_separate_buffer;
      tex->last_dcc_separate_buffer = NULL;
   } else {
      tex->dcc_separate_buffer =
         si_aligned_buffer_create(sctx->b.screen, SI_RESOURCE_FLAG_UNMAPPABLE, PIPE_USAGE_DEFAULT,
                                  tex->surface.dcc_size, tex->surface.dcc_alignment);
      if (!tex->dcc_separate_buffer)
         return;
   }

   /* dcc_offset is the absolute GPUVM address. */
   tex->surface.dcc_offset = tex->dcc_separate_buffer->gpu_address;

   /* no need to flag anything since this is called by fast clear that
    * flags framebuffer state
    */
}

/**
 * Called by pipe_context::flush_resource, the place where DCC decompression
 * takes place.
 */
void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx, struct si_texture *tex)
{
   struct si_context *sctx = (struct si_context *)ctx;
   struct pipe_query *tmp;
   unsigned i = vi_get_context_dcc_stats_index(sctx, tex);
   bool query_active = sctx->dcc_stats[i].query_active;
   bool disable = false;

   if (sctx->dcc_stats[i].ps_stats[2]) {
      union pipe_query_result result;

      /* Read the results. */
      struct pipe_query *query = sctx->dcc_stats[i].ps_stats[2];
      ctx->get_query_result(ctx, query, true, &result);
      si_query_buffer_reset(sctx, &((struct si_query_hw *)query)->buffer);

      /* Compute the approximate number of fullscreen draws. */
      tex->ps_draw_ratio = result.pipeline_statistics.ps_invocations /
                           (tex->buffer.b.b.width0 * tex->buffer.b.b.height0);
      sctx->last_tex_ps_draw_ratio = tex->ps_draw_ratio;

      disable = tex->dcc_separate_buffer && !vi_should_enable_separate_dcc(tex);
   }

   tex->num_slow_clears = 0;

   /* stop the statistics query for ps_stats[0] */
   if (query_active)
      vi_separate_dcc_stop_query(sctx, tex);

   /* Move the queries in the queue by one. */
   tmp = sctx->dcc_stats[i].ps_stats[2];
   sctx->dcc_stats[i].ps_stats[2] = sctx->dcc_stats[i].ps_stats[1];
   sctx->dcc_stats[i].ps_stats[1] = sctx->dcc_stats[i].ps_stats[0];
   sctx->dcc_stats[i].ps_stats[0] = tmp;

   /* create and start a new query as ps_stats[0] */
   if (query_active)
      vi_separate_dcc_start_query(sctx, tex);

   if (disable) {
      assert(!tex->last_dcc_separate_buffer);
      tex->last_dcc_separate_buffer = tex->dcc_separate_buffer;
      tex->dcc_separate_buffer = NULL;
      tex->surface.dcc_offset = 0;
      /* no need to flag anything since this is called after
       * decompression that re-sets framebuffer state
       */
   }
}

static struct pipe_memory_object *
si_memobj_from_handle(struct pipe_screen *screen, struct winsys_handle *whandle, bool dedicated)
{
   struct si_screen *sscreen = (struct si_screen *)screen;
   struct si_memory_object *memobj = CALLOC_STRUCT(si_memory_object);
   struct pb_buffer *buf = NULL;

   if (!memobj)
      return NULL;

   buf = sscreen->ws->buffer_from_handle(sscreen->ws, whandle, sscreen->info.max_alignment);
   if (!buf) {
      free(memobj);
      return NULL;
   }

   memobj->b.dedicated = dedicated;
   memobj->buf = buf;
   memobj->stride = whandle->stride;

   return (struct pipe_memory_object *)memobj;
}

static void si_memobj_destroy(struct pipe_screen *screen, struct pipe_memory_object *_memobj)
{
   struct si_memory_object *memobj = (struct si_memory_object *)_memobj;

   pb_reference(&memobj->buf, NULL);
   free(memobj);
}

static struct pipe_resource *si_texture_from_memobj(struct pipe_screen *screen,
                                                    const struct pipe_resource *templ,
                                                    struct pipe_memory_object *_memobj,
                                                    uint64_t offset)
{
   struct si_screen *sscreen = (struct si_screen *)screen;
   struct si_memory_object *memobj = (struct si_memory_object *)_memobj;
   struct pipe_resource *tex = si_texture_from_winsys_buffer(
      sscreen, templ, memobj->buf, memobj->stride, offset,
      PIPE_HANDLE_USAGE_FRAMEBUFFER_WRITE | PIPE_HANDLE_USAGE_SHADER_WRITE, memobj->b.dedicated);
   if (!tex)
      return NULL;

   /* si_texture_from_winsys_buffer doesn't increment refcount of
    * memobj->buf, so increment it here.
    */
   struct pb_buffer *buf = NULL;
   pb_reference(&buf, memobj->buf);
   return tex;
}

static bool si_check_resource_capability(struct pipe_screen *screen, struct pipe_resource *resource,
                                         unsigned bind)
{
   struct si_texture *tex = (struct si_texture *)resource;

   /* Buffers only support the linear flag. */
   if (resource->target == PIPE_BUFFER)
      return (bind & ~PIPE_BIND_LINEAR) == 0;

   if (bind & PIPE_BIND_LINEAR && !tex->surface.is_linear)
      return false;

   if (bind & PIPE_BIND_SCANOUT && !tex->surface.is_displayable)
      return false;

   /* TODO: PIPE_BIND_CURSOR - do we care? */
   return true;
}

void si_init_screen_texture_functions(struct si_screen *sscreen)
{
   sscreen->b.resource_from_handle = si_texture_from_handle;
   sscreen->b.resource_get_handle = si_texture_get_handle;
   sscreen->b.resource_get_param = si_resource_get_param;
   sscreen->b.resource_get_info = si_texture_get_info;
   sscreen->b.resource_from_memobj = si_texture_from_memobj;
   sscreen->b.memobj_create_from_handle = si_memobj_from_handle;
   sscreen->b.memobj_destroy = si_memobj_destroy;
   sscreen->b.check_resource_capability = si_check_resource_capability;
}

void si_init_context_texture_functions(struct si_context *sctx)
{
   sctx->b.create_surface = si_create_surface;
   sctx->b.surface_destroy = si_surface_destroy;
}