summaryrefslogtreecommitdiff
path: root/src/intel/vulkan/genX_cmd_buffer.c
blob: de2e2140f27c33c553bb3fe24b4358dde7224a28 (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
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
/*
 * Copyright © 2015 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include <assert.h>
#include <stdbool.h>

#include "anv_private.h"
#include "vk_format_info.h"
#include "vk_util.h"

#include "common/gen_l3_config.h"
#include "genxml/gen_macros.h"
#include "genxml/genX_pack.h"

static void
emit_lrm(struct anv_batch *batch,
         uint32_t reg, struct anv_bo *bo, uint32_t offset)
{
   anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
      lrm.RegisterAddress  = reg;
      lrm.MemoryAddress    = (struct anv_address) { bo, offset };
   }
}

static void
emit_lri(struct anv_batch *batch, uint32_t reg, uint32_t imm)
{
   anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
      lri.RegisterOffset   = reg;
      lri.DataDWord        = imm;
   }
}

#if GEN_IS_HASWELL || GEN_GEN >= 8
static void
emit_lrr(struct anv_batch *batch, uint32_t dst, uint32_t src)
{
   anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_REG), lrr) {
      lrr.SourceRegisterAddress        = src;
      lrr.DestinationRegisterAddress   = dst;
   }
}
#endif

void
genX(cmd_buffer_emit_state_base_address)(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_device *device = cmd_buffer->device;

   /* Emit a render target cache flush.
    *
    * This isn't documented anywhere in the PRM.  However, it seems to be
    * necessary prior to changing the surface state base adress.  Without
    * this, we get GPU hangs when using multi-level command buffers which
    * clear depth, reset state base address, and then go render stuff.
    */
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      pc.DCFlushEnable = true;
      pc.RenderTargetCacheFlushEnable = true;
      pc.CommandStreamerStallEnable = true;
   }

   anv_batch_emit(&cmd_buffer->batch, GENX(STATE_BASE_ADDRESS), sba) {
      sba.GeneralStateBaseAddress = (struct anv_address) { NULL, 0 };
      sba.GeneralStateMemoryObjectControlState = GENX(MOCS);
      sba.GeneralStateBaseAddressModifyEnable = true;

      sba.SurfaceStateBaseAddress =
         anv_cmd_buffer_surface_base_address(cmd_buffer);
      sba.SurfaceStateMemoryObjectControlState = GENX(MOCS);
      sba.SurfaceStateBaseAddressModifyEnable = true;

      sba.DynamicStateBaseAddress =
         (struct anv_address) { &device->dynamic_state_pool.block_pool.bo, 0 };
      sba.DynamicStateMemoryObjectControlState = GENX(MOCS);
      sba.DynamicStateBaseAddressModifyEnable = true;

      sba.IndirectObjectBaseAddress = (struct anv_address) { NULL, 0 };
      sba.IndirectObjectMemoryObjectControlState = GENX(MOCS);
      sba.IndirectObjectBaseAddressModifyEnable = true;

      sba.InstructionBaseAddress =
         (struct anv_address) { &device->instruction_state_pool.block_pool.bo, 0 };
      sba.InstructionMemoryObjectControlState = GENX(MOCS);
      sba.InstructionBaseAddressModifyEnable = true;

#  if (GEN_GEN >= 8)
      /* Broadwell requires that we specify a buffer size for a bunch of
       * these fields.  However, since we will be growing the BO's live, we
       * just set them all to the maximum.
       */
      sba.GeneralStateBufferSize                = 0xfffff;
      sba.GeneralStateBufferSizeModifyEnable    = true;
      sba.DynamicStateBufferSize                = 0xfffff;
      sba.DynamicStateBufferSizeModifyEnable    = true;
      sba.IndirectObjectBufferSize              = 0xfffff;
      sba.IndirectObjectBufferSizeModifyEnable  = true;
      sba.InstructionBufferSize                 = 0xfffff;
      sba.InstructionBuffersizeModifyEnable     = true;
#  endif
   }

   /* After re-setting the surface state base address, we have to do some
    * cache flusing so that the sampler engine will pick up the new
    * SURFACE_STATE objects and binding tables. From the Broadwell PRM,
    * Shared Function > 3D Sampler > State > State Caching (page 96):
    *
    *    Coherency with system memory in the state cache, like the texture
    *    cache is handled partially by software. It is expected that the
    *    command stream or shader will issue Cache Flush operation or
    *    Cache_Flush sampler message to ensure that the L1 cache remains
    *    coherent with system memory.
    *
    *    [...]
    *
    *    Whenever the value of the Dynamic_State_Base_Addr,
    *    Surface_State_Base_Addr are altered, the L1 state cache must be
    *    invalidated to ensure the new surface or sampler state is fetched
    *    from system memory.
    *
    * The PIPE_CONTROL command has a "State Cache Invalidation Enable" bit
    * which, according the PIPE_CONTROL instruction documentation in the
    * Broadwell PRM:
    *
    *    Setting this bit is independent of any other bit in this packet.
    *    This bit controls the invalidation of the L1 and L2 state caches
    *    at the top of the pipe i.e. at the parsing time.
    *
    * Unfortunately, experimentation seems to indicate that state cache
    * invalidation through a PIPE_CONTROL does nothing whatsoever in
    * regards to surface state and binding tables.  In stead, it seems that
    * invalidating the texture cache is what is actually needed.
    *
    * XXX:  As far as we have been able to determine through
    * experimentation, shows that flush the texture cache appears to be
    * sufficient.  The theory here is that all of the sampling/rendering
    * units cache the binding table in the texture cache.  However, we have
    * yet to be able to actually confirm this.
    */
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      pc.TextureCacheInvalidationEnable = true;
      pc.ConstantCacheInvalidationEnable = true;
      pc.StateCacheInvalidationEnable = true;
   }
}

static void
add_surface_state_reloc(struct anv_cmd_buffer *cmd_buffer,
                        struct anv_state state,
                        struct anv_bo *bo, uint32_t offset)
{
   const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;

   VkResult result =
      anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
                         state.offset + isl_dev->ss.addr_offset, bo, offset);
   if (result != VK_SUCCESS)
      anv_batch_set_error(&cmd_buffer->batch, result);
}

static void
add_image_relocs(struct anv_cmd_buffer * const cmd_buffer,
                 const struct anv_image * const image,
                 const VkImageAspectFlags aspect_mask,
                 const enum isl_aux_usage aux_usage,
                 const struct anv_state state)
{
   const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;
   const uint32_t surf_offset = image->offset +
      anv_image_get_surface_for_aspect_mask(image, aspect_mask)->offset;

   add_surface_state_reloc(cmd_buffer, state, image->bo, surf_offset);

   if (aux_usage != ISL_AUX_USAGE_NONE) {
      uint32_t aux_offset = image->offset + image->aux_surface.offset;

      /* On gen7 and prior, the bottom 12 bits of the MCS base address are
       * used to store other information.  This should be ok, however, because
       * surface buffer addresses are always 4K page alinged.
       */
      assert((aux_offset & 0xfff) == 0);
      uint32_t *aux_addr_dw = state.map + isl_dev->ss.aux_addr_offset;
      aux_offset += *aux_addr_dw & 0xfff;

      VkResult result =
         anv_reloc_list_add(&cmd_buffer->surface_relocs,
                            &cmd_buffer->pool->alloc,
                            state.offset + isl_dev->ss.aux_addr_offset,
                            image->bo, aux_offset);
      if (result != VK_SUCCESS)
         anv_batch_set_error(&cmd_buffer->batch, result);
   }
}

static bool
color_is_zero_one(VkClearColorValue value, enum isl_format format)
{
   if (isl_format_has_int_channel(format)) {
      for (unsigned i = 0; i < 4; i++) {
         if (value.int32[i] != 0 && value.int32[i] != 1)
            return false;
      }
   } else {
      for (unsigned i = 0; i < 4; i++) {
         if (value.float32[i] != 0.0f && value.float32[i] != 1.0f)
            return false;
      }
   }

   return true;
}

static void
color_attachment_compute_aux_usage(struct anv_device *device,
                                   struct anv_attachment_state *att_state,
                                   struct anv_image_view *iview,
                                   VkRect2D render_area,
                                   union isl_color_value *fast_clear_color)
{
   if (iview->image->aux_surface.isl.size == 0) {
      att_state->aux_usage = ISL_AUX_USAGE_NONE;
      att_state->input_aux_usage = ISL_AUX_USAGE_NONE;
      att_state->fast_clear = false;
      return;
   } else if (iview->image->aux_usage == ISL_AUX_USAGE_MCS) {
      att_state->aux_usage = ISL_AUX_USAGE_MCS;
      att_state->input_aux_usage = ISL_AUX_USAGE_MCS;
      att_state->fast_clear = false;
      return;
   }

   assert(iview->image->aux_surface.isl.usage & ISL_SURF_USAGE_CCS_BIT);

   att_state->clear_color_is_zero_one =
      color_is_zero_one(att_state->clear_value.color, iview->isl.format);

   if (att_state->pending_clear_aspects == VK_IMAGE_ASPECT_COLOR_BIT) {
      /* Start off assuming fast clears are possible */
      att_state->fast_clear = true;

      /* Potentially, we could do partial fast-clears but doing so has crazy
       * alignment restrictions.  It's easier to just restrict to full size
       * fast clears for now.
       */
      if (render_area.offset.x != 0 ||
          render_area.offset.y != 0 ||
          render_area.extent.width != iview->extent.width ||
          render_area.extent.height != iview->extent.height)
         att_state->fast_clear = false;

      if (GEN_GEN <= 7) {
         /* On gen7, we can't do multi-LOD or multi-layer fast-clears.  We
          * technically can, but it comes with crazy restrictions that we
          * don't want to deal with now.
          */
         if (iview->isl.base_level > 0 ||
             iview->isl.base_array_layer > 0 ||
             iview->isl.array_len > 1)
            att_state->fast_clear = false;
      }

      /* On Broadwell and earlier, we can only handle 0/1 clear colors */
      if (GEN_GEN <= 8 && !att_state->clear_color_is_zero_one)
         att_state->fast_clear = false;

      if (att_state->fast_clear) {
         memcpy(fast_clear_color->u32, att_state->clear_value.color.uint32,
                sizeof(fast_clear_color->u32));
      }
   } else {
      att_state->fast_clear = false;
   }

   /**
    * TODO: Consider using a heuristic to determine if temporarily enabling
    * CCS_E for this image view would be beneficial.
    *
    * While fast-clear resolves and partial resolves are fairly cheap in the
    * case where you render to most of the pixels, full resolves are not
    * because they potentially involve reading and writing the entire
    * framebuffer.  If we can't texture with CCS_E, we should leave it off and
    * limit ourselves to fast clears.
    */
   if (iview->image->aux_usage == ISL_AUX_USAGE_CCS_E) {
      att_state->aux_usage = ISL_AUX_USAGE_CCS_E;
      att_state->input_aux_usage = ISL_AUX_USAGE_CCS_E;
   } else if (att_state->fast_clear) {
      att_state->aux_usage = ISL_AUX_USAGE_CCS_D;
      /* From the Sky Lake PRM, RENDER_SURFACE_STATE::AuxiliarySurfaceMode:
       *
       *    "If Number of Multisamples is MULTISAMPLECOUNT_1, AUX_CCS_D
       *    setting is only allowed if Surface Format supported for Fast
       *    Clear. In addition, if the surface is bound to the sampling
       *    engine, Surface Format must be supported for Render Target
       *    Compression for surfaces bound to the sampling engine."
       *
       * In other words, we can only sample from a fast-cleared image if it
       * also supports color compression.
       */
      if (isl_format_supports_ccs_e(&device->info, iview->isl.format))
         att_state->input_aux_usage = ISL_AUX_USAGE_CCS_D;
      else
         att_state->input_aux_usage = ISL_AUX_USAGE_NONE;
   } else {
      att_state->aux_usage = ISL_AUX_USAGE_NONE;
      att_state->input_aux_usage = ISL_AUX_USAGE_NONE;
   }
}

static bool
need_input_attachment_state(const struct anv_render_pass_attachment *att)
{
   if (!(att->usage & VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT))
      return false;

   /* We only allocate input attachment states for color surfaces. Compression
    * is not yet enabled for depth textures and stencil doesn't allow
    * compression so we can just use the texture surface state from the view.
    */
   return vk_format_is_color(att->format);
}

/* Transitions a HiZ-enabled depth buffer from one layout to another. Unless
 * the initial layout is undefined, the HiZ buffer and depth buffer will
 * represent the same data at the end of this operation.
 */
static void
transition_depth_buffer(struct anv_cmd_buffer *cmd_buffer,
                        const struct anv_image *image,
                        VkImageLayout initial_layout,
                        VkImageLayout final_layout)
{
   assert(image);

   /* A transition is a no-op if HiZ is not enabled, or if the initial and
    * final layouts are equal.
    *
    * The undefined layout indicates that the user doesn't care about the data
    * that's currently in the buffer. Therefore, a data-preserving resolve
    * operation is not needed.
    */
   if (image->aux_usage != ISL_AUX_USAGE_HIZ || initial_layout == final_layout)
      return;

   const bool hiz_enabled = ISL_AUX_USAGE_HIZ ==
      anv_layout_to_aux_usage(&cmd_buffer->device->info, image, image->aspects,
                              initial_layout);
   const bool enable_hiz = ISL_AUX_USAGE_HIZ ==
      anv_layout_to_aux_usage(&cmd_buffer->device->info, image, image->aspects,
                              final_layout);

   enum blorp_hiz_op hiz_op;
   if (hiz_enabled && !enable_hiz) {
      hiz_op = BLORP_HIZ_OP_DEPTH_RESOLVE;
   } else if (!hiz_enabled && enable_hiz) {
      hiz_op = BLORP_HIZ_OP_HIZ_RESOLVE;
   } else {
      assert(hiz_enabled == enable_hiz);
      /* If the same buffer will be used, no resolves are necessary. */
      hiz_op = BLORP_HIZ_OP_NONE;
   }

   if (hiz_op != BLORP_HIZ_OP_NONE)
      anv_gen8_hiz_op_resolve(cmd_buffer, image, hiz_op);
}

static inline uint32_t
get_fast_clear_state_entry_offset(const struct anv_device *device,
                                  const struct anv_image *image,
                                  unsigned level)
{
   assert(device && image);
   assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
   assert(level < anv_image_aux_levels(image));
   const uint32_t offset = image->offset + image->aux_surface.offset +
                           image->aux_surface.isl.size +
                           anv_fast_clear_state_entry_size(device) * level;
   assert(offset < image->offset + image->size);
   return offset;
}

static void
init_fast_clear_state_entry(struct anv_cmd_buffer *cmd_buffer,
                            const struct anv_image *image,
                            unsigned level)
{
   assert(cmd_buffer && image);
   assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
   assert(level < anv_image_aux_levels(image));

   /* The fast clear value dword(s) will be copied into a surface state object.
    * Ensure that the restrictions of the fields in the dword(s) are followed.
    *
    * CCS buffers on SKL+ can have any value set for the clear colors.
    */
   if (image->samples == 1 && GEN_GEN >= 9)
      return;

   /* Other combinations of auxiliary buffers and platforms require specific
    * values in the clear value dword(s).
    */
   unsigned i = 0;
   for (; i < cmd_buffer->device->isl_dev.ss.clear_value_size; i += 4) {
      anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi) {
         const uint32_t entry_offset =
            get_fast_clear_state_entry_offset(cmd_buffer->device, image, level);
         sdi.Address = (struct anv_address) { image->bo, entry_offset + i };

         if (GEN_GEN >= 9) {
            /* MCS buffers on SKL+ can only have 1/0 clear colors. */
            assert(image->aux_usage == ISL_AUX_USAGE_MCS);
            sdi.ImmediateData = 0;
         } else if (GEN_VERSIONx10 >= 75) {
            /* Pre-SKL, the dword containing the clear values also contains
             * other fields, so we need to initialize those fields to match the
             * values that would be in a color attachment.
             */
            assert(i == 0);
            sdi.ImmediateData = ISL_CHANNEL_SELECT_RED   << 25 |
                                ISL_CHANNEL_SELECT_GREEN << 22 |
                                ISL_CHANNEL_SELECT_BLUE  << 19 |
                                ISL_CHANNEL_SELECT_ALPHA << 16;
         }  else if (GEN_VERSIONx10 == 70) {
            /* On IVB, the dword containing the clear values also contains
             * other fields that must be zero or can be zero.
             */
            assert(i == 0);
            sdi.ImmediateData = 0;
         }
      }
   }
}

static void
transition_color_buffer(struct anv_cmd_buffer *cmd_buffer,
                        const struct anv_image *image,
                        const uint32_t base_level, uint32_t level_count,
                        uint32_t base_layer, uint32_t layer_count,
                        VkImageLayout initial_layout,
                        VkImageLayout final_layout)
{
   assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);

   if (image->aux_surface.isl.size == 0 ||
       base_level >= anv_image_aux_levels(image))
      return;

   if (initial_layout != VK_IMAGE_LAYOUT_UNDEFINED &&
       initial_layout != VK_IMAGE_LAYOUT_PREINITIALIZED)
      return;

   /* A transition of a 3D subresource works on all slices at a time. */
   if (image->type == VK_IMAGE_TYPE_3D) {
      base_layer = 0;
      layer_count = anv_minify(image->extent.depth, base_level);
   }

   /* We're interested in the subresource range subset that has aux data. */
   level_count = MIN2(level_count, anv_image_aux_levels(image) - base_level);

   /* We're transitioning from an undefined layout. We must ensure that the
    * clear values buffer is filled with valid data.
    */
   for (unsigned l = 0; l < level_count; l++)
      init_fast_clear_state_entry(cmd_buffer, image, base_level + l);

   if (image->aux_usage == ISL_AUX_USAGE_CCS_E ||
       image->aux_usage == ISL_AUX_USAGE_MCS) {
      /* We're transitioning from an undefined layout so it doesn't really
       * matter what data ends up in the color buffer. We do, however, need to
       * ensure that the auxiliary surface is not in an undefined state. This
       * state is possible for CCS buffers SKL+ and MCS buffers with certain
       * sample counts that require certain bits to be reserved (2x and 8x).
       * One easy way to get to a valid state is to fast-clear the specified
       * range.
       *
       * Even for MCS buffers that have sample counts that don't require
       * certain bits to be reserved (4x and 8x), we're unsure if the hardware
       * will be okay with the sample mappings given by the undefined buffer.
       * We don't have any data to show that this is a problem, but we want to
       * avoid causing difficult-to-debug problems.
       */
      if (image->samples == 4 || image->samples == 16) {
         anv_perf_warn("Doing a potentially unnecessary fast-clear to define "
                       "an MCS buffer.");
      }

      anv_image_fast_clear(cmd_buffer, image, base_level, level_count,
                           base_layer, layer_count);
   }
}

/**
 * Setup anv_cmd_state::attachments for vkCmdBeginRenderPass.
 */
static VkResult
genX(cmd_buffer_setup_attachments)(struct anv_cmd_buffer *cmd_buffer,
                                   struct anv_render_pass *pass,
                                   const VkRenderPassBeginInfo *begin)
{
   const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;
   struct anv_cmd_state *state = &cmd_buffer->state;

   vk_free(&cmd_buffer->pool->alloc, state->attachments);

   if (pass->attachment_count > 0) {
      state->attachments = vk_alloc(&cmd_buffer->pool->alloc,
                                    pass->attachment_count *
                                         sizeof(state->attachments[0]),
                                    8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
      if (state->attachments == NULL) {
         /* Propagate VK_ERROR_OUT_OF_HOST_MEMORY to vkEndCommandBuffer */
         return anv_batch_set_error(&cmd_buffer->batch,
                                    VK_ERROR_OUT_OF_HOST_MEMORY);
      }
   } else {
      state->attachments = NULL;
   }

   /* Reserve one for the NULL state. */
   unsigned num_states = 1;
   for (uint32_t i = 0; i < pass->attachment_count; ++i) {
      if (vk_format_is_color(pass->attachments[i].format))
         num_states++;

      if (need_input_attachment_state(&pass->attachments[i]))
         num_states++;
   }

   const uint32_t ss_stride = align_u32(isl_dev->ss.size, isl_dev->ss.align);
   state->render_pass_states =
      anv_state_stream_alloc(&cmd_buffer->surface_state_stream,
                             num_states * ss_stride, isl_dev->ss.align);

   struct anv_state next_state = state->render_pass_states;
   next_state.alloc_size = isl_dev->ss.size;

   state->null_surface_state = next_state;
   next_state.offset += ss_stride;
   next_state.map += ss_stride;

   for (uint32_t i = 0; i < pass->attachment_count; ++i) {
      if (vk_format_is_color(pass->attachments[i].format)) {
         state->attachments[i].color_rt_state = next_state;
         next_state.offset += ss_stride;
         next_state.map += ss_stride;
      }

      if (need_input_attachment_state(&pass->attachments[i])) {
         state->attachments[i].input_att_state = next_state;
         next_state.offset += ss_stride;
         next_state.map += ss_stride;
      }
   }
   assert(next_state.offset == state->render_pass_states.offset +
                               state->render_pass_states.alloc_size);

   if (begin) {
      ANV_FROM_HANDLE(anv_framebuffer, framebuffer, begin->framebuffer);
      assert(pass->attachment_count == framebuffer->attachment_count);

      struct GENX(RENDER_SURFACE_STATE) null_ss = {
         .SurfaceType = SURFTYPE_NULL,
         .SurfaceArray = framebuffer->layers > 0,
         .SurfaceFormat = ISL_FORMAT_R8G8B8A8_UNORM,
#if GEN_GEN >= 8
         .TileMode = YMAJOR,
#else
         .TiledSurface = true,
#endif
         .Width = framebuffer->width - 1,
         .Height = framebuffer->height - 1,
         .Depth = framebuffer->layers - 1,
         .RenderTargetViewExtent = framebuffer->layers - 1,
      };
      GENX(RENDER_SURFACE_STATE_pack)(NULL, state->null_surface_state.map,
                                      &null_ss);

      for (uint32_t i = 0; i < pass->attachment_count; ++i) {
         struct anv_render_pass_attachment *att = &pass->attachments[i];
         VkImageAspectFlags att_aspects = vk_format_aspects(att->format);
         VkImageAspectFlags clear_aspects = 0;

         if (att_aspects == VK_IMAGE_ASPECT_COLOR_BIT) {
            /* color attachment */
            if (att->load_op == VK_ATTACHMENT_LOAD_OP_CLEAR) {
               clear_aspects |= VK_IMAGE_ASPECT_COLOR_BIT;
            }
         } else {
            /* depthstencil attachment */
            if ((att_aspects & VK_IMAGE_ASPECT_DEPTH_BIT) &&
                att->load_op == VK_ATTACHMENT_LOAD_OP_CLEAR) {
               clear_aspects |= VK_IMAGE_ASPECT_DEPTH_BIT;
            }
            if ((att_aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
                att->stencil_load_op == VK_ATTACHMENT_LOAD_OP_CLEAR) {
               clear_aspects |= VK_IMAGE_ASPECT_STENCIL_BIT;
            }
         }

         state->attachments[i].current_layout = att->initial_layout;
         state->attachments[i].pending_clear_aspects = clear_aspects;
         if (clear_aspects)
            state->attachments[i].clear_value = begin->pClearValues[i];

         struct anv_image_view *iview = framebuffer->attachments[i];
         anv_assert(iview->vk_format == att->format);

         union isl_color_value clear_color = { .u32 = { 0, } };
         if (att_aspects == VK_IMAGE_ASPECT_COLOR_BIT) {
            color_attachment_compute_aux_usage(cmd_buffer->device,
                                               &state->attachments[i],
                                               iview, begin->renderArea,
                                               &clear_color);

            struct isl_view view = iview->isl;
            view.usage |= ISL_SURF_USAGE_RENDER_TARGET_BIT;
            view.swizzle = anv_swizzle_for_render(view.swizzle);
            isl_surf_fill_state(isl_dev,
                                state->attachments[i].color_rt_state.map,
                                .surf = &iview->image->color_surface.isl,
                                .view = &view,
                                .aux_surf = &iview->image->aux_surface.isl,
                                .aux_usage = state->attachments[i].aux_usage,
                                .clear_color = clear_color,
                                .mocs = cmd_buffer->device->default_mocs);

            add_image_relocs(cmd_buffer, iview->image, iview->aspect_mask,
                             state->attachments[i].aux_usage,
                             state->attachments[i].color_rt_state);
         } else {
            /* This field will be initialized after the first subpass
             * transition.
             */
            state->attachments[i].aux_usage = ISL_AUX_USAGE_NONE;

            state->attachments[i].input_aux_usage = ISL_AUX_USAGE_NONE;
         }

         if (need_input_attachment_state(&pass->attachments[i])) {
            struct isl_view view = iview->isl;
            view.usage |= ISL_SURF_USAGE_TEXTURE_BIT;
            isl_surf_fill_state(isl_dev,
                                state->attachments[i].input_att_state.map,
                                .surf = &iview->image->color_surface.isl,
                                .view = &view,
                                .aux_surf = &iview->image->aux_surface.isl,
                                .aux_usage = state->attachments[i].input_aux_usage,
                                .clear_color = clear_color,
                                .mocs = cmd_buffer->device->default_mocs);

            add_image_relocs(cmd_buffer, iview->image, iview->aspect_mask,
                             state->attachments[i].input_aux_usage,
                             state->attachments[i].input_att_state);
         }
      }

      anv_state_flush(cmd_buffer->device, state->render_pass_states);
   }

   return VK_SUCCESS;
}

VkResult
genX(BeginCommandBuffer)(
    VkCommandBuffer                             commandBuffer,
    const VkCommandBufferBeginInfo*             pBeginInfo)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   /* If this is the first vkBeginCommandBuffer, we must *initialize* the
    * command buffer's state. Otherwise, we must *reset* its state. In both
    * cases we reset it.
    *
    * From the Vulkan 1.0 spec:
    *
    *    If a command buffer is in the executable state and the command buffer
    *    was allocated from a command pool with the
    *    VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT flag set, then
    *    vkBeginCommandBuffer implicitly resets the command buffer, behaving
    *    as if vkResetCommandBuffer had been called with
    *    VK_COMMAND_BUFFER_RESET_RELEASE_RESOURCES_BIT not set. It then puts
    *    the command buffer in the recording state.
    */
   anv_cmd_buffer_reset(cmd_buffer);

   cmd_buffer->usage_flags = pBeginInfo->flags;

   assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_SECONDARY ||
          !(cmd_buffer->usage_flags & VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT));

   genX(cmd_buffer_emit_state_base_address)(cmd_buffer);

   /* We sometimes store vertex data in the dynamic state buffer for blorp
    * operations and our dynamic state stream may re-use data from previous
    * command buffers.  In order to prevent stale cache data, we flush the VF
    * cache.  We could do this on every blorp call but that's not really
    * needed as all of the data will get written by the CPU prior to the GPU
    * executing anything.  The chances are fairly high that they will use
    * blorp at least once per primary command buffer so it shouldn't be
    * wasted.
    */
   if (cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY)
      cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_VF_CACHE_INVALIDATE_BIT;

   VkResult result = VK_SUCCESS;
   if (cmd_buffer->usage_flags &
       VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT) {
      cmd_buffer->state.pass =
         anv_render_pass_from_handle(pBeginInfo->pInheritanceInfo->renderPass);
      cmd_buffer->state.subpass =
         &cmd_buffer->state.pass->subpasses[pBeginInfo->pInheritanceInfo->subpass];
      cmd_buffer->state.framebuffer = NULL;

      result = genX(cmd_buffer_setup_attachments)(cmd_buffer,
                                                  cmd_buffer->state.pass, NULL);

      cmd_buffer->state.dirty |= ANV_CMD_DIRTY_RENDER_TARGETS;
   }

   return result;
}

VkResult
genX(EndCommandBuffer)(
    VkCommandBuffer                             commandBuffer)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return cmd_buffer->batch.status;

   /* We want every command buffer to start with the PMA fix in a known state,
    * so we disable it at the end of the command buffer.
    */
   genX(cmd_buffer_enable_pma_fix)(cmd_buffer, false);

   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   anv_cmd_buffer_end_batch_buffer(cmd_buffer);

   return VK_SUCCESS;
}

void
genX(CmdExecuteCommands)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    commandBufferCount,
    const VkCommandBuffer*                      pCmdBuffers)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, primary, commandBuffer);

   assert(primary->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);

   if (anv_batch_has_error(&primary->batch))
      return;

   /* The secondary command buffers will assume that the PMA fix is disabled
    * when they begin executing.  Make sure this is true.
    */
   genX(cmd_buffer_enable_pma_fix)(primary, false);

   /* The secondary command buffer doesn't know which textures etc. have been
    * flushed prior to their execution.  Apply those flushes now.
    */
   genX(cmd_buffer_apply_pipe_flushes)(primary);

   for (uint32_t i = 0; i < commandBufferCount; i++) {
      ANV_FROM_HANDLE(anv_cmd_buffer, secondary, pCmdBuffers[i]);

      assert(secondary->level == VK_COMMAND_BUFFER_LEVEL_SECONDARY);
      assert(!anv_batch_has_error(&secondary->batch));

      if (secondary->usage_flags &
          VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT) {
         /* If we're continuing a render pass from the primary, we need to
          * copy the surface states for the current subpass into the storage
          * we allocated for them in BeginCommandBuffer.
          */
         struct anv_bo *ss_bo =
            &primary->device->surface_state_pool.block_pool.bo;
         struct anv_state src_state = primary->state.render_pass_states;
         struct anv_state dst_state = secondary->state.render_pass_states;
         assert(src_state.alloc_size == dst_state.alloc_size);

         genX(cmd_buffer_so_memcpy)(primary, ss_bo, dst_state.offset,
                                    ss_bo, src_state.offset,
                                    src_state.alloc_size);
      }

      anv_cmd_buffer_add_secondary(primary, secondary);
   }

   /* Each of the secondary command buffers will use its own state base
    * address.  We need to re-emit state base address for the primary after
    * all of the secondaries are done.
    *
    * TODO: Maybe we want to make this a dirty bit to avoid extra state base
    * address calls?
    */
   genX(cmd_buffer_emit_state_base_address)(primary);
}

#define IVB_L3SQCREG1_SQGHPCI_DEFAULT     0x00730000
#define VLV_L3SQCREG1_SQGHPCI_DEFAULT     0x00d30000
#define HSW_L3SQCREG1_SQGHPCI_DEFAULT     0x00610000

/**
 * Program the hardware to use the specified L3 configuration.
 */
void
genX(cmd_buffer_config_l3)(struct anv_cmd_buffer *cmd_buffer,
                           const struct gen_l3_config *cfg)
{
   assert(cfg);
   if (cfg == cmd_buffer->state.current_l3_config)
      return;

   if (unlikely(INTEL_DEBUG & DEBUG_L3)) {
      fprintf(stderr, "L3 config transition: ");
      gen_dump_l3_config(cfg, stderr);
   }

   const bool has_slm = cfg->n[GEN_L3P_SLM];

   /* According to the hardware docs, the L3 partitioning can only be changed
    * while the pipeline is completely drained and the caches are flushed,
    * which involves a first PIPE_CONTROL flush which stalls the pipeline...
    */
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      pc.DCFlushEnable = true;
      pc.PostSyncOperation = NoWrite;
      pc.CommandStreamerStallEnable = true;
   }

   /* ...followed by a second pipelined PIPE_CONTROL that initiates
    * invalidation of the relevant caches.  Note that because RO invalidation
    * happens at the top of the pipeline (i.e. right away as the PIPE_CONTROL
    * command is processed by the CS) we cannot combine it with the previous
    * stalling flush as the hardware documentation suggests, because that
    * would cause the CS to stall on previous rendering *after* RO
    * invalidation and wouldn't prevent the RO caches from being polluted by
    * concurrent rendering before the stall completes.  This intentionally
    * doesn't implement the SKL+ hardware workaround suggesting to enable CS
    * stall on PIPE_CONTROLs with the texture cache invalidation bit set for
    * GPGPU workloads because the previous and subsequent PIPE_CONTROLs
    * already guarantee that there is no concurrent GPGPU kernel execution
    * (see SKL HSD 2132585).
    */
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      pc.TextureCacheInvalidationEnable = true;
      pc.ConstantCacheInvalidationEnable = true;
      pc.InstructionCacheInvalidateEnable = true;
      pc.StateCacheInvalidationEnable = true;
      pc.PostSyncOperation = NoWrite;
   }

   /* Now send a third stalling flush to make sure that invalidation is
    * complete when the L3 configuration registers are modified.
    */
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      pc.DCFlushEnable = true;
      pc.PostSyncOperation = NoWrite;
      pc.CommandStreamerStallEnable = true;
   }

#if GEN_GEN >= 8

   assert(!cfg->n[GEN_L3P_IS] && !cfg->n[GEN_L3P_C] && !cfg->n[GEN_L3P_T]);

   uint32_t l3cr;
   anv_pack_struct(&l3cr, GENX(L3CNTLREG),
                   .SLMEnable = has_slm,
                   .URBAllocation = cfg->n[GEN_L3P_URB],
                   .ROAllocation = cfg->n[GEN_L3P_RO],
                   .DCAllocation = cfg->n[GEN_L3P_DC],
                   .AllAllocation = cfg->n[GEN_L3P_ALL]);

   /* Set up the L3 partitioning. */
   emit_lri(&cmd_buffer->batch, GENX(L3CNTLREG_num), l3cr);

#else

   const bool has_dc = cfg->n[GEN_L3P_DC] || cfg->n[GEN_L3P_ALL];
   const bool has_is = cfg->n[GEN_L3P_IS] || cfg->n[GEN_L3P_RO] ||
                       cfg->n[GEN_L3P_ALL];
   const bool has_c = cfg->n[GEN_L3P_C] || cfg->n[GEN_L3P_RO] ||
                      cfg->n[GEN_L3P_ALL];
   const bool has_t = cfg->n[GEN_L3P_T] || cfg->n[GEN_L3P_RO] ||
                      cfg->n[GEN_L3P_ALL];

   assert(!cfg->n[GEN_L3P_ALL]);

   /* When enabled SLM only uses a portion of the L3 on half of the banks,
    * the matching space on the remaining banks has to be allocated to a
    * client (URB for all validated configurations) set to the
    * lower-bandwidth 2-bank address hashing mode.
    */
   const struct gen_device_info *devinfo = &cmd_buffer->device->info;
   const bool urb_low_bw = has_slm && !devinfo->is_baytrail;
   assert(!urb_low_bw || cfg->n[GEN_L3P_URB] == cfg->n[GEN_L3P_SLM]);

   /* Minimum number of ways that can be allocated to the URB. */
   MAYBE_UNUSED const unsigned n0_urb = devinfo->is_baytrail ? 32 : 0;
   assert(cfg->n[GEN_L3P_URB] >= n0_urb);

   uint32_t l3sqcr1, l3cr2, l3cr3;
   anv_pack_struct(&l3sqcr1, GENX(L3SQCREG1),
                   .ConvertDC_UC = !has_dc,
                   .ConvertIS_UC = !has_is,
                   .ConvertC_UC = !has_c,
                   .ConvertT_UC = !has_t);
   l3sqcr1 |=
      GEN_IS_HASWELL ? HSW_L3SQCREG1_SQGHPCI_DEFAULT :
      devinfo->is_baytrail ? VLV_L3SQCREG1_SQGHPCI_DEFAULT :
      IVB_L3SQCREG1_SQGHPCI_DEFAULT;

   anv_pack_struct(&l3cr2, GENX(L3CNTLREG2),
                   .SLMEnable = has_slm,
                   .URBLowBandwidth = urb_low_bw,
                   .URBAllocation = cfg->n[GEN_L3P_URB] - n0_urb,
#if !GEN_IS_HASWELL
                   .ALLAllocation = cfg->n[GEN_L3P_ALL],
#endif
                   .ROAllocation = cfg->n[GEN_L3P_RO],
                   .DCAllocation = cfg->n[GEN_L3P_DC]);

   anv_pack_struct(&l3cr3, GENX(L3CNTLREG3),
                   .ISAllocation = cfg->n[GEN_L3P_IS],
                   .ISLowBandwidth = 0,
                   .CAllocation = cfg->n[GEN_L3P_C],
                   .CLowBandwidth = 0,
                   .TAllocation = cfg->n[GEN_L3P_T],
                   .TLowBandwidth = 0);

   /* Set up the L3 partitioning. */
   emit_lri(&cmd_buffer->batch, GENX(L3SQCREG1_num), l3sqcr1);
   emit_lri(&cmd_buffer->batch, GENX(L3CNTLREG2_num), l3cr2);
   emit_lri(&cmd_buffer->batch, GENX(L3CNTLREG3_num), l3cr3);

#if GEN_IS_HASWELL
   if (cmd_buffer->device->instance->physicalDevice.cmd_parser_version >= 4) {
      /* Enable L3 atomics on HSW if we have a DC partition, otherwise keep
       * them disabled to avoid crashing the system hard.
       */
      uint32_t scratch1, chicken3;
      anv_pack_struct(&scratch1, GENX(SCRATCH1),
                      .L3AtomicDisable = !has_dc);
      anv_pack_struct(&chicken3, GENX(CHICKEN3),
                      .L3AtomicDisableMask = true,
                      .L3AtomicDisable = !has_dc);
      emit_lri(&cmd_buffer->batch, GENX(SCRATCH1_num), scratch1);
      emit_lri(&cmd_buffer->batch, GENX(CHICKEN3_num), chicken3);
   }
#endif

#endif

   cmd_buffer->state.current_l3_config = cfg;
}

void
genX(cmd_buffer_apply_pipe_flushes)(struct anv_cmd_buffer *cmd_buffer)
{
   enum anv_pipe_bits bits = cmd_buffer->state.pending_pipe_bits;

   /* Flushes are pipelined while invalidations are handled immediately.
    * Therefore, if we're flushing anything then we need to schedule a stall
    * before any invalidations can happen.
    */
   if (bits & ANV_PIPE_FLUSH_BITS)
      bits |= ANV_PIPE_NEEDS_CS_STALL_BIT;

   /* If we're going to do an invalidate and we have a pending CS stall that
    * has yet to be resolved, we do the CS stall now.
    */
   if ((bits & ANV_PIPE_INVALIDATE_BITS) &&
       (bits & ANV_PIPE_NEEDS_CS_STALL_BIT)) {
      bits |= ANV_PIPE_CS_STALL_BIT;
      bits &= ~ANV_PIPE_NEEDS_CS_STALL_BIT;
   }

   if (bits & (ANV_PIPE_FLUSH_BITS | ANV_PIPE_CS_STALL_BIT)) {
      anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe) {
         pipe.DepthCacheFlushEnable = bits & ANV_PIPE_DEPTH_CACHE_FLUSH_BIT;
         pipe.DCFlushEnable = bits & ANV_PIPE_DATA_CACHE_FLUSH_BIT;
         pipe.RenderTargetCacheFlushEnable =
            bits & ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT;

         pipe.DepthStallEnable = bits & ANV_PIPE_DEPTH_STALL_BIT;
         pipe.CommandStreamerStallEnable = bits & ANV_PIPE_CS_STALL_BIT;
         pipe.StallAtPixelScoreboard = bits & ANV_PIPE_STALL_AT_SCOREBOARD_BIT;

         /*
          * According to the Broadwell documentation, any PIPE_CONTROL with the
          * "Command Streamer Stall" bit set must also have another bit set,
          * with five different options:
          *
          *  - Render Target Cache Flush
          *  - Depth Cache Flush
          *  - Stall at Pixel Scoreboard
          *  - Post-Sync Operation
          *  - Depth Stall
          *  - DC Flush Enable
          *
          * I chose "Stall at Pixel Scoreboard" since that's what we use in
          * mesa and it seems to work fine. The choice is fairly arbitrary.
          */
         if ((bits & ANV_PIPE_CS_STALL_BIT) &&
             !(bits & (ANV_PIPE_FLUSH_BITS | ANV_PIPE_DEPTH_STALL_BIT |
                       ANV_PIPE_STALL_AT_SCOREBOARD_BIT)))
            pipe.StallAtPixelScoreboard = true;
      }

      bits &= ~(ANV_PIPE_FLUSH_BITS | ANV_PIPE_CS_STALL_BIT);
   }

   if (bits & ANV_PIPE_INVALIDATE_BITS) {
      anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe) {
         pipe.StateCacheInvalidationEnable =
            bits & ANV_PIPE_STATE_CACHE_INVALIDATE_BIT;
         pipe.ConstantCacheInvalidationEnable =
            bits & ANV_PIPE_CONSTANT_CACHE_INVALIDATE_BIT;
         pipe.VFCacheInvalidationEnable =
            bits & ANV_PIPE_VF_CACHE_INVALIDATE_BIT;
         pipe.TextureCacheInvalidationEnable =
            bits & ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT;
         pipe.InstructionCacheInvalidateEnable =
            bits & ANV_PIPE_INSTRUCTION_CACHE_INVALIDATE_BIT;
      }

      bits &= ~ANV_PIPE_INVALIDATE_BITS;
   }

   cmd_buffer->state.pending_pipe_bits = bits;
}

void genX(CmdPipelineBarrier)(
    VkCommandBuffer                             commandBuffer,
    VkPipelineStageFlags                        srcStageMask,
    VkPipelineStageFlags                        destStageMask,
    VkBool32                                    byRegion,
    uint32_t                                    memoryBarrierCount,
    const VkMemoryBarrier*                      pMemoryBarriers,
    uint32_t                                    bufferMemoryBarrierCount,
    const VkBufferMemoryBarrier*                pBufferMemoryBarriers,
    uint32_t                                    imageMemoryBarrierCount,
    const VkImageMemoryBarrier*                 pImageMemoryBarriers)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   /* XXX: Right now, we're really dumb and just flush whatever categories
    * the app asks for.  One of these days we may make this a bit better
    * but right now that's all the hardware allows for in most areas.
    */
   VkAccessFlags src_flags = 0;
   VkAccessFlags dst_flags = 0;

   for (uint32_t i = 0; i < memoryBarrierCount; i++) {
      src_flags |= pMemoryBarriers[i].srcAccessMask;
      dst_flags |= pMemoryBarriers[i].dstAccessMask;
   }

   for (uint32_t i = 0; i < bufferMemoryBarrierCount; i++) {
      src_flags |= pBufferMemoryBarriers[i].srcAccessMask;
      dst_flags |= pBufferMemoryBarriers[i].dstAccessMask;
   }

   for (uint32_t i = 0; i < imageMemoryBarrierCount; i++) {
      src_flags |= pImageMemoryBarriers[i].srcAccessMask;
      dst_flags |= pImageMemoryBarriers[i].dstAccessMask;
      ANV_FROM_HANDLE(anv_image, image, pImageMemoryBarriers[i].image);
      const VkImageSubresourceRange *range =
         &pImageMemoryBarriers[i].subresourceRange;

      if (range->aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT) {
         transition_depth_buffer(cmd_buffer, image,
                                 pImageMemoryBarriers[i].oldLayout,
                                 pImageMemoryBarriers[i].newLayout);
      } else if (range->aspectMask == VK_IMAGE_ASPECT_COLOR_BIT) {
         transition_color_buffer(cmd_buffer, image,
                                 range->baseMipLevel,
                                 anv_get_levelCount(image, range),
                                 range->baseArrayLayer,
                                 anv_get_layerCount(image, range),
                                 pImageMemoryBarriers[i].oldLayout,
                                 pImageMemoryBarriers[i].newLayout);
      }
   }

   cmd_buffer->state.pending_pipe_bits |=
      anv_pipe_flush_bits_for_access_flags(src_flags) |
      anv_pipe_invalidate_bits_for_access_flags(dst_flags);
}

static void
cmd_buffer_alloc_push_constants(struct anv_cmd_buffer *cmd_buffer)
{
   VkShaderStageFlags stages = cmd_buffer->state.pipeline->active_stages;

   /* In order to avoid thrash, we assume that vertex and fragment stages
    * always exist.  In the rare case where one is missing *and* the other
    * uses push concstants, this may be suboptimal.  However, avoiding stalls
    * seems more important.
    */
   stages |= VK_SHADER_STAGE_FRAGMENT_BIT | VK_SHADER_STAGE_VERTEX_BIT;

   if (stages == cmd_buffer->state.push_constant_stages)
      return;

#if GEN_GEN >= 8
   const unsigned push_constant_kb = 32;
#elif GEN_IS_HASWELL
   const unsigned push_constant_kb = cmd_buffer->device->info.gt == 3 ? 32 : 16;
#else
   const unsigned push_constant_kb = 16;
#endif

   const unsigned num_stages =
      _mesa_bitcount(stages & VK_SHADER_STAGE_ALL_GRAPHICS);
   unsigned size_per_stage = push_constant_kb / num_stages;

   /* Broadwell+ and Haswell gt3 require that the push constant sizes be in
    * units of 2KB.  Incidentally, these are the same platforms that have
    * 32KB worth of push constant space.
    */
   if (push_constant_kb == 32)
      size_per_stage &= ~1u;

   uint32_t kb_used = 0;
   for (int i = MESA_SHADER_VERTEX; i < MESA_SHADER_FRAGMENT; i++) {
      unsigned push_size = (stages & (1 << i)) ? size_per_stage : 0;
      anv_batch_emit(&cmd_buffer->batch,
                     GENX(3DSTATE_PUSH_CONSTANT_ALLOC_VS), alloc) {
         alloc._3DCommandSubOpcode  = 18 + i;
         alloc.ConstantBufferOffset = (push_size > 0) ? kb_used : 0;
         alloc.ConstantBufferSize   = push_size;
      }
      kb_used += push_size;
   }

   anv_batch_emit(&cmd_buffer->batch,
                  GENX(3DSTATE_PUSH_CONSTANT_ALLOC_PS), alloc) {
      alloc.ConstantBufferOffset = kb_used;
      alloc.ConstantBufferSize = push_constant_kb - kb_used;
   }

   cmd_buffer->state.push_constant_stages = stages;

   /* From the BDW PRM for 3DSTATE_PUSH_CONSTANT_ALLOC_VS:
    *
    *    "The 3DSTATE_CONSTANT_VS must be reprogrammed prior to
    *    the next 3DPRIMITIVE command after programming the
    *    3DSTATE_PUSH_CONSTANT_ALLOC_VS"
    *
    * Since 3DSTATE_PUSH_CONSTANT_ALLOC_VS is programmed as part of
    * pipeline setup, we need to dirty push constants.
    */
   cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_ALL_GRAPHICS;
}

static VkResult
emit_binding_table(struct anv_cmd_buffer *cmd_buffer,
                   gl_shader_stage stage,
                   struct anv_state *bt_state)
{
   struct anv_subpass *subpass = cmd_buffer->state.subpass;
   struct anv_pipeline *pipeline;
   uint32_t bias, state_offset;

   switch (stage) {
   case  MESA_SHADER_COMPUTE:
      pipeline = cmd_buffer->state.compute_pipeline;
      bias = 1;
      break;
   default:
      pipeline = cmd_buffer->state.pipeline;
      bias = 0;
      break;
   }

   if (!anv_pipeline_has_stage(pipeline, stage)) {
      *bt_state = (struct anv_state) { 0, };
      return VK_SUCCESS;
   }

   struct anv_pipeline_bind_map *map = &pipeline->shaders[stage]->bind_map;
   if (bias + map->surface_count == 0) {
      *bt_state = (struct anv_state) { 0, };
      return VK_SUCCESS;
   }

   *bt_state = anv_cmd_buffer_alloc_binding_table(cmd_buffer,
                                                  bias + map->surface_count,
                                                  &state_offset);
   uint32_t *bt_map = bt_state->map;

   if (bt_state->map == NULL)
      return VK_ERROR_OUT_OF_DEVICE_MEMORY;

   if (stage == MESA_SHADER_COMPUTE &&
       get_cs_prog_data(cmd_buffer->state.compute_pipeline)->uses_num_work_groups) {
      struct anv_bo *bo = cmd_buffer->state.num_workgroups_bo;
      uint32_t bo_offset = cmd_buffer->state.num_workgroups_offset;

      struct anv_state surface_state;
      surface_state =
         anv_cmd_buffer_alloc_surface_state(cmd_buffer);

      const enum isl_format format =
         anv_isl_format_for_descriptor_type(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER);
      anv_fill_buffer_surface_state(cmd_buffer->device, surface_state,
                                    format, bo_offset, 12, 1);

      bt_map[0] = surface_state.offset + state_offset;
      add_surface_state_reloc(cmd_buffer, surface_state, bo, bo_offset);
   }

   if (map->surface_count == 0)
      goto out;

   if (map->image_count > 0) {
      VkResult result =
         anv_cmd_buffer_ensure_push_constant_field(cmd_buffer, stage, images);
      if (result != VK_SUCCESS)
         return result;

      cmd_buffer->state.push_constants_dirty |= 1 << stage;
   }

   uint32_t image = 0;
   for (uint32_t s = 0; s < map->surface_count; s++) {
      struct anv_pipeline_binding *binding = &map->surface_to_descriptor[s];

      struct anv_state surface_state;

      if (binding->set == ANV_DESCRIPTOR_SET_COLOR_ATTACHMENTS) {
         /* Color attachment binding */
         assert(stage == MESA_SHADER_FRAGMENT);
         assert(binding->binding == 0);
         if (binding->index < subpass->color_count) {
            const unsigned att =
               subpass->color_attachments[binding->index].attachment;

            /* From the Vulkan 1.0.46 spec:
             *
             *    "If any color or depth/stencil attachments are
             *    VK_ATTACHMENT_UNUSED, then no writes occur for those
             *    attachments."
             */
            if (att == VK_ATTACHMENT_UNUSED) {
               surface_state = cmd_buffer->state.null_surface_state;
            } else {
               surface_state = cmd_buffer->state.attachments[att].color_rt_state;
            }
         } else {
            surface_state = cmd_buffer->state.null_surface_state;
         }

         bt_map[bias + s] = surface_state.offset + state_offset;
         continue;
      }

      struct anv_descriptor_set *set =
         cmd_buffer->state.descriptors[binding->set];
      uint32_t offset = set->layout->binding[binding->binding].descriptor_index;
      struct anv_descriptor *desc = &set->descriptors[offset + binding->index];

      switch (desc->type) {
      case VK_DESCRIPTOR_TYPE_SAMPLER:
         /* Nothing for us to do here */
         continue;

      case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
      case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
         surface_state = desc->aux_usage == ISL_AUX_USAGE_NONE ?
            desc->image_view->no_aux_sampler_surface_state :
            desc->image_view->sampler_surface_state;
         assert(surface_state.alloc_size);
         add_image_relocs(cmd_buffer, desc->image_view->image,
                          desc->image_view->aspect_mask,
                          desc->aux_usage, surface_state);
         break;
      case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
         assert(stage == MESA_SHADER_FRAGMENT);
         if (desc->image_view->aspect_mask != VK_IMAGE_ASPECT_COLOR_BIT) {
            /* For depth and stencil input attachments, we treat it like any
             * old texture that a user may have bound.
             */
            surface_state = desc->aux_usage == ISL_AUX_USAGE_NONE ?
               desc->image_view->no_aux_sampler_surface_state :
               desc->image_view->sampler_surface_state;
            assert(surface_state.alloc_size);
            add_image_relocs(cmd_buffer, desc->image_view->image,
                             desc->image_view->aspect_mask,
                             desc->aux_usage, surface_state);
         } else {
            /* For color input attachments, we create the surface state at
             * vkBeginRenderPass time so that we can include aux and clear
             * color information.
             */
            assert(binding->input_attachment_index < subpass->input_count);
            const unsigned subpass_att = binding->input_attachment_index;
            const unsigned att = subpass->input_attachments[subpass_att].attachment;
            surface_state = cmd_buffer->state.attachments[att].input_att_state;
         }
         break;

      case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: {
         surface_state = (binding->write_only)
            ? desc->image_view->writeonly_storage_surface_state
            : desc->image_view->storage_surface_state;
         assert(surface_state.alloc_size);
         add_image_relocs(cmd_buffer, desc->image_view->image,
                          desc->image_view->aspect_mask,
                          desc->image_view->image->aux_usage, surface_state);

         struct brw_image_param *image_param =
            &cmd_buffer->state.push_constants[stage]->images[image++];

         *image_param = desc->image_view->storage_image_param;
         image_param->surface_idx = bias + s;
         break;
      }

      case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
      case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
      case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
         surface_state = desc->buffer_view->surface_state;
         assert(surface_state.alloc_size);
         add_surface_state_reloc(cmd_buffer, surface_state,
                                 desc->buffer_view->bo,
                                 desc->buffer_view->offset);
         break;

      case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
      case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
         uint32_t dynamic_offset_idx =
            pipeline->layout->set[binding->set].dynamic_offset_start +
            set->layout->binding[binding->binding].dynamic_offset_index +
            binding->index;

         /* Compute the offset within the buffer */
         uint64_t offset = desc->offset +
            cmd_buffer->state.dynamic_offsets[dynamic_offset_idx];
         /* Clamp to the buffer size */
         offset = MIN2(offset, desc->buffer->size);
         /* Clamp the range to the buffer size */
         uint32_t range = MIN2(desc->range, desc->buffer->size - offset);

         surface_state =
            anv_state_stream_alloc(&cmd_buffer->surface_state_stream, 64, 64);
         enum isl_format format =
            anv_isl_format_for_descriptor_type(desc->type);

         anv_fill_buffer_surface_state(cmd_buffer->device, surface_state,
                                       format, offset, range, 1);
         add_surface_state_reloc(cmd_buffer, surface_state,
                                 desc->buffer->bo,
                                 desc->buffer->offset + offset);
         break;
      }

      case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
         surface_state = (binding->write_only)
            ? desc->buffer_view->writeonly_storage_surface_state
            : desc->buffer_view->storage_surface_state;
         assert(surface_state.alloc_size);
         add_surface_state_reloc(cmd_buffer, surface_state,
                                 desc->buffer_view->bo,
                                 desc->buffer_view->offset);

         struct brw_image_param *image_param =
            &cmd_buffer->state.push_constants[stage]->images[image++];

         *image_param = desc->buffer_view->storage_image_param;
         image_param->surface_idx = bias + s;
         break;

      default:
         assert(!"Invalid descriptor type");
         continue;
      }

      bt_map[bias + s] = surface_state.offset + state_offset;
   }
   assert(image == map->image_count);

 out:
   anv_state_flush(cmd_buffer->device, *bt_state);

   return VK_SUCCESS;
}

static VkResult
emit_samplers(struct anv_cmd_buffer *cmd_buffer,
              gl_shader_stage stage,
              struct anv_state *state)
{
   struct anv_pipeline *pipeline;

   if (stage == MESA_SHADER_COMPUTE)
      pipeline = cmd_buffer->state.compute_pipeline;
   else
      pipeline = cmd_buffer->state.pipeline;

   if (!anv_pipeline_has_stage(pipeline, stage)) {
      *state = (struct anv_state) { 0, };
      return VK_SUCCESS;
   }

   struct anv_pipeline_bind_map *map = &pipeline->shaders[stage]->bind_map;
   if (map->sampler_count == 0) {
      *state = (struct anv_state) { 0, };
      return VK_SUCCESS;
   }

   uint32_t size = map->sampler_count * 16;
   *state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, 32);

   if (state->map == NULL)
      return VK_ERROR_OUT_OF_DEVICE_MEMORY;

   for (uint32_t s = 0; s < map->sampler_count; s++) {
      struct anv_pipeline_binding *binding = &map->sampler_to_descriptor[s];
      struct anv_descriptor_set *set =
         cmd_buffer->state.descriptors[binding->set];
      uint32_t offset = set->layout->binding[binding->binding].descriptor_index;
      struct anv_descriptor *desc = &set->descriptors[offset + binding->index];

      if (desc->type != VK_DESCRIPTOR_TYPE_SAMPLER &&
          desc->type != VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
         continue;

      struct anv_sampler *sampler = desc->sampler;

      /* This can happen if we have an unfilled slot since TYPE_SAMPLER
       * happens to be zero.
       */
      if (sampler == NULL)
         continue;

      memcpy(state->map + (s * 16),
             sampler->state, sizeof(sampler->state));
   }

   anv_state_flush(cmd_buffer->device, *state);

   return VK_SUCCESS;
}

static uint32_t
flush_descriptor_sets(struct anv_cmd_buffer *cmd_buffer)
{
   VkShaderStageFlags dirty = cmd_buffer->state.descriptors_dirty &
                              cmd_buffer->state.pipeline->active_stages;

   VkResult result = VK_SUCCESS;
   anv_foreach_stage(s, dirty) {
      result = emit_samplers(cmd_buffer, s, &cmd_buffer->state.samplers[s]);
      if (result != VK_SUCCESS)
         break;
      result = emit_binding_table(cmd_buffer, s,
                                  &cmd_buffer->state.binding_tables[s]);
      if (result != VK_SUCCESS)
         break;
   }

   if (result != VK_SUCCESS) {
      assert(result == VK_ERROR_OUT_OF_DEVICE_MEMORY);

      result = anv_cmd_buffer_new_binding_table_block(cmd_buffer);
      if (result != VK_SUCCESS)
         return 0;

      /* Re-emit state base addresses so we get the new surface state base
       * address before we start emitting binding tables etc.
       */
      genX(cmd_buffer_emit_state_base_address)(cmd_buffer);

      /* Re-emit all active binding tables */
      dirty |= cmd_buffer->state.pipeline->active_stages;
      anv_foreach_stage(s, dirty) {
         result = emit_samplers(cmd_buffer, s, &cmd_buffer->state.samplers[s]);
         if (result != VK_SUCCESS) {
            anv_batch_set_error(&cmd_buffer->batch, result);
            return 0;
         }
         result = emit_binding_table(cmd_buffer, s,
                                     &cmd_buffer->state.binding_tables[s]);
         if (result != VK_SUCCESS) {
            anv_batch_set_error(&cmd_buffer->batch, result);
            return 0;
         }
      }
   }

   cmd_buffer->state.descriptors_dirty &= ~dirty;

   return dirty;
}

static void
cmd_buffer_emit_descriptor_pointers(struct anv_cmd_buffer *cmd_buffer,
                                    uint32_t stages)
{
   static const uint32_t sampler_state_opcodes[] = {
      [MESA_SHADER_VERTEX]                      = 43,
      [MESA_SHADER_TESS_CTRL]                   = 44, /* HS */
      [MESA_SHADER_TESS_EVAL]                   = 45, /* DS */
      [MESA_SHADER_GEOMETRY]                    = 46,
      [MESA_SHADER_FRAGMENT]                    = 47,
      [MESA_SHADER_COMPUTE]                     = 0,
   };

   static const uint32_t binding_table_opcodes[] = {
      [MESA_SHADER_VERTEX]                      = 38,
      [MESA_SHADER_TESS_CTRL]                   = 39,
      [MESA_SHADER_TESS_EVAL]                   = 40,
      [MESA_SHADER_GEOMETRY]                    = 41,
      [MESA_SHADER_FRAGMENT]                    = 42,
      [MESA_SHADER_COMPUTE]                     = 0,
   };

   anv_foreach_stage(s, stages) {
      if (cmd_buffer->state.samplers[s].alloc_size > 0) {
         anv_batch_emit(&cmd_buffer->batch,
                        GENX(3DSTATE_SAMPLER_STATE_POINTERS_VS), ssp) {
            ssp._3DCommandSubOpcode = sampler_state_opcodes[s];
            ssp.PointertoVSSamplerState = cmd_buffer->state.samplers[s].offset;
         }
      }

      /* Always emit binding table pointers if we're asked to, since on SKL
       * this is what flushes push constants. */
      anv_batch_emit(&cmd_buffer->batch,
                     GENX(3DSTATE_BINDING_TABLE_POINTERS_VS), btp) {
         btp._3DCommandSubOpcode = binding_table_opcodes[s];
         btp.PointertoVSBindingTable = cmd_buffer->state.binding_tables[s].offset;
      }
   }
}

static uint32_t
cmd_buffer_flush_push_constants(struct anv_cmd_buffer *cmd_buffer)
{
   static const uint32_t push_constant_opcodes[] = {
      [MESA_SHADER_VERTEX]                      = 21,
      [MESA_SHADER_TESS_CTRL]                   = 25, /* HS */
      [MESA_SHADER_TESS_EVAL]                   = 26, /* DS */
      [MESA_SHADER_GEOMETRY]                    = 22,
      [MESA_SHADER_FRAGMENT]                    = 23,
      [MESA_SHADER_COMPUTE]                     = 0,
   };

   VkShaderStageFlags flushed = 0;

   anv_foreach_stage(stage, cmd_buffer->state.push_constants_dirty) {
      if (stage == MESA_SHADER_COMPUTE)
         continue;

      struct anv_state state = anv_cmd_buffer_push_constants(cmd_buffer, stage);

      if (state.offset == 0) {
         anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CONSTANT_VS), c)
            c._3DCommandSubOpcode = push_constant_opcodes[stage];
      } else {
         anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CONSTANT_VS), c) {
            c._3DCommandSubOpcode = push_constant_opcodes[stage],
            c.ConstantBody = (struct GENX(3DSTATE_CONSTANT_BODY)) {
#if GEN_GEN >= 9
               .Buffer[2] = { &cmd_buffer->device->dynamic_state_pool.block_pool.bo, state.offset },
               .ReadLength[2] = DIV_ROUND_UP(state.alloc_size, 32),
#else
               .Buffer[0] = { .offset = state.offset },
               .ReadLength[0] = DIV_ROUND_UP(state.alloc_size, 32),
#endif
            };
         }
      }

      flushed |= mesa_to_vk_shader_stage(stage);
   }

   cmd_buffer->state.push_constants_dirty &= ~VK_SHADER_STAGE_ALL_GRAPHICS;

   return flushed;
}

void
genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
   uint32_t *p;

   uint32_t vb_emit = cmd_buffer->state.vb_dirty & pipeline->vb_used;

   assert((pipeline->active_stages & VK_SHADER_STAGE_COMPUTE_BIT) == 0);

   genX(cmd_buffer_config_l3)(cmd_buffer, pipeline->urb.l3_config);

   genX(flush_pipeline_select_3d)(cmd_buffer);

   if (vb_emit) {
      const uint32_t num_buffers = __builtin_popcount(vb_emit);
      const uint32_t num_dwords = 1 + num_buffers * 4;

      p = anv_batch_emitn(&cmd_buffer->batch, num_dwords,
                          GENX(3DSTATE_VERTEX_BUFFERS));
      uint32_t vb, i = 0;
      for_each_bit(vb, vb_emit) {
         struct anv_buffer *buffer = cmd_buffer->state.vertex_bindings[vb].buffer;
         uint32_t offset = cmd_buffer->state.vertex_bindings[vb].offset;

         struct GENX(VERTEX_BUFFER_STATE) state = {
            .VertexBufferIndex = vb,

#if GEN_GEN >= 8
            .MemoryObjectControlState = GENX(MOCS),
#else
            .BufferAccessType = pipeline->instancing_enable[vb] ? INSTANCEDATA : VERTEXDATA,
            /* Our implementation of VK_KHR_multiview uses instancing to draw
             * the different views.  If the client asks for instancing, we
             * need to use the Instance Data Step Rate to ensure that we
             * repeat the client's per-instance data once for each view.
             */
            .InstanceDataStepRate = anv_subpass_view_count(pipeline->subpass),
            .VertexBufferMemoryObjectControlState = GENX(MOCS),
#endif

            .AddressModifyEnable = true,
            .BufferPitch = pipeline->binding_stride[vb],
            .BufferStartingAddress = { buffer->bo, buffer->offset + offset },

#if GEN_GEN >= 8
            .BufferSize = buffer->size - offset
#else
            .EndAddress = { buffer->bo, buffer->offset + buffer->size - 1},
#endif
         };

         GENX(VERTEX_BUFFER_STATE_pack)(&cmd_buffer->batch, &p[1 + i * 4], &state);
         i++;
      }
   }

   cmd_buffer->state.vb_dirty &= ~vb_emit;

   if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_PIPELINE) {
      anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);

      /* The exact descriptor layout is pulled from the pipeline, so we need
       * to re-emit binding tables on every pipeline change.
       */
      cmd_buffer->state.descriptors_dirty |=
         cmd_buffer->state.pipeline->active_stages;

      /* If the pipeline changed, we may need to re-allocate push constant
       * space in the URB.
       */
      cmd_buffer_alloc_push_constants(cmd_buffer);
   }

#if GEN_GEN <= 7
   if (cmd_buffer->state.descriptors_dirty & VK_SHADER_STAGE_VERTEX_BIT ||
       cmd_buffer->state.push_constants_dirty & VK_SHADER_STAGE_VERTEX_BIT) {
      /* From the IVB PRM Vol. 2, Part 1, Section 3.2.1:
       *
       *    "A PIPE_CONTROL with Post-Sync Operation set to 1h and a depth
       *    stall needs to be sent just prior to any 3DSTATE_VS,
       *    3DSTATE_URB_VS, 3DSTATE_CONSTANT_VS,
       *    3DSTATE_BINDING_TABLE_POINTER_VS,
       *    3DSTATE_SAMPLER_STATE_POINTER_VS command.  Only one
       *    PIPE_CONTROL needs to be sent before any combination of VS
       *    associated 3DSTATE."
       */
      anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
         pc.DepthStallEnable  = true;
         pc.PostSyncOperation = WriteImmediateData;
         pc.Address           =
            (struct anv_address) { &cmd_buffer->device->workaround_bo, 0 };
      }
   }
#endif

   /* Render targets live in the same binding table as fragment descriptors */
   if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_RENDER_TARGETS)
      cmd_buffer->state.descriptors_dirty |= VK_SHADER_STAGE_FRAGMENT_BIT;

   /* We emit the binding tables and sampler tables first, then emit push
    * constants and then finally emit binding table and sampler table
    * pointers.  It has to happen in this order, since emitting the binding
    * tables may change the push constants (in case of storage images). After
    * emitting push constants, on SKL+ we have to emit the corresponding
    * 3DSTATE_BINDING_TABLE_POINTER_* for the push constants to take effect.
    */
   uint32_t dirty = 0;
   if (cmd_buffer->state.descriptors_dirty)
      dirty = flush_descriptor_sets(cmd_buffer);

   if (cmd_buffer->state.push_constants_dirty) {
#if GEN_GEN >= 9
      /* On Sky Lake and later, the binding table pointers commands are
       * what actually flush the changes to push constant state so we need
       * to dirty them so they get re-emitted below.
       */
      dirty |= cmd_buffer_flush_push_constants(cmd_buffer);
#else
      cmd_buffer_flush_push_constants(cmd_buffer);
#endif
   }

   if (dirty)
      cmd_buffer_emit_descriptor_pointers(cmd_buffer, dirty);

   if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_VIEWPORT)
      gen8_cmd_buffer_emit_viewport(cmd_buffer);

   if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_DYNAMIC_VIEWPORT |
                                  ANV_CMD_DIRTY_PIPELINE)) {
      gen8_cmd_buffer_emit_depth_viewport(cmd_buffer,
                                          pipeline->depth_clamp_enable);
   }

   if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_SCISSOR)
      gen7_cmd_buffer_emit_scissor(cmd_buffer);

   genX(cmd_buffer_flush_dynamic_state)(cmd_buffer);

   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
}

static void
emit_vertex_bo(struct anv_cmd_buffer *cmd_buffer,
               struct anv_bo *bo, uint32_t offset,
               uint32_t size, uint32_t index)
{
   uint32_t *p = anv_batch_emitn(&cmd_buffer->batch, 5,
                                 GENX(3DSTATE_VERTEX_BUFFERS));

   GENX(VERTEX_BUFFER_STATE_pack)(&cmd_buffer->batch, p + 1,
      &(struct GENX(VERTEX_BUFFER_STATE)) {
         .VertexBufferIndex = index,
         .AddressModifyEnable = true,
         .BufferPitch = 0,
#if (GEN_GEN >= 8)
         .MemoryObjectControlState = GENX(MOCS),
         .BufferStartingAddress = { bo, offset },
         .BufferSize = size
#else
         .VertexBufferMemoryObjectControlState = GENX(MOCS),
         .BufferStartingAddress = { bo, offset },
         .EndAddress = { bo, offset + size },
#endif
      });
}

static void
emit_base_vertex_instance_bo(struct anv_cmd_buffer *cmd_buffer,
                             struct anv_bo *bo, uint32_t offset)
{
   emit_vertex_bo(cmd_buffer, bo, offset, 8, ANV_SVGS_VB_INDEX);
}

static void
emit_base_vertex_instance(struct anv_cmd_buffer *cmd_buffer,
                          uint32_t base_vertex, uint32_t base_instance)
{
   struct anv_state id_state =
      anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, 8, 4);

   ((uint32_t *)id_state.map)[0] = base_vertex;
   ((uint32_t *)id_state.map)[1] = base_instance;

   anv_state_flush(cmd_buffer->device, id_state);

   emit_base_vertex_instance_bo(cmd_buffer,
      &cmd_buffer->device->dynamic_state_pool.block_pool.bo, id_state.offset);
}

static void
emit_draw_index(struct anv_cmd_buffer *cmd_buffer, uint32_t draw_index)
{
   struct anv_state state =
      anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, 4, 4);

   ((uint32_t *)state.map)[0] = draw_index;

   anv_state_flush(cmd_buffer->device, state);

   emit_vertex_bo(cmd_buffer,
                  &cmd_buffer->device->dynamic_state_pool.block_pool.bo,
                  state.offset, 4, ANV_DRAWID_VB_INDEX);
}

void genX(CmdDraw)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    vertexCount,
    uint32_t                                    instanceCount,
    uint32_t                                    firstVertex,
    uint32_t                                    firstInstance)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
   const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   genX(cmd_buffer_flush_state)(cmd_buffer);

   if (vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance)
      emit_base_vertex_instance(cmd_buffer, firstVertex, firstInstance);
   if (vs_prog_data->uses_drawid)
      emit_draw_index(cmd_buffer, 0);

   /* Our implementation of VK_KHR_multiview uses instancing to draw the
    * different views.  We need to multiply instanceCount by the view count.
    */
   instanceCount *= anv_subpass_view_count(cmd_buffer->state.subpass);

   anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
      prim.VertexAccessType         = SEQUENTIAL;
      prim.PrimitiveTopologyType    = pipeline->topology;
      prim.VertexCountPerInstance   = vertexCount;
      prim.StartVertexLocation      = firstVertex;
      prim.InstanceCount            = instanceCount;
      prim.StartInstanceLocation    = firstInstance;
      prim.BaseVertexLocation       = 0;
   }
}

void genX(CmdDrawIndexed)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    indexCount,
    uint32_t                                    instanceCount,
    uint32_t                                    firstIndex,
    int32_t                                     vertexOffset,
    uint32_t                                    firstInstance)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
   const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   genX(cmd_buffer_flush_state)(cmd_buffer);

   if (vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance)
      emit_base_vertex_instance(cmd_buffer, vertexOffset, firstInstance);
   if (vs_prog_data->uses_drawid)
      emit_draw_index(cmd_buffer, 0);

   /* Our implementation of VK_KHR_multiview uses instancing to draw the
    * different views.  We need to multiply instanceCount by the view count.
    */
   instanceCount *= anv_subpass_view_count(cmd_buffer->state.subpass);

   anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
      prim.VertexAccessType         = RANDOM;
      prim.PrimitiveTopologyType    = pipeline->topology;
      prim.VertexCountPerInstance   = indexCount;
      prim.StartVertexLocation      = firstIndex;
      prim.InstanceCount            = instanceCount;
      prim.StartInstanceLocation    = firstInstance;
      prim.BaseVertexLocation       = vertexOffset;
   }
}

/* Auto-Draw / Indirect Registers */
#define GEN7_3DPRIM_END_OFFSET          0x2420
#define GEN7_3DPRIM_START_VERTEX        0x2430
#define GEN7_3DPRIM_VERTEX_COUNT        0x2434
#define GEN7_3DPRIM_INSTANCE_COUNT      0x2438
#define GEN7_3DPRIM_START_INSTANCE      0x243C
#define GEN7_3DPRIM_BASE_VERTEX         0x2440

/* MI_MATH only exists on Haswell+ */
#if GEN_IS_HASWELL || GEN_GEN >= 8

static uint32_t
mi_alu(uint32_t opcode, uint32_t op1, uint32_t op2)
{
   struct GENX(MI_MATH_ALU_INSTRUCTION) instr = {
      .ALUOpcode = opcode,
      .Operand1 = op1,
      .Operand2 = op2,
   };

   uint32_t dw;
   GENX(MI_MATH_ALU_INSTRUCTION_pack)(NULL, &dw, &instr);

   return dw;
}

#define CS_GPR(n) (0x2600 + (n) * 8)

/* Emit dwords to multiply GPR0 by N */
static void
build_alu_multiply_gpr0(uint32_t *dw, unsigned *dw_count, uint32_t N)
{
   VK_OUTARRAY_MAKE(out, dw, dw_count);

#define append_alu(opcode, operand1, operand2) \
   vk_outarray_append(&out, alu_dw) *alu_dw = mi_alu(opcode, operand1, operand2)

   assert(N > 0);
   unsigned top_bit = 31 - __builtin_clz(N);
   for (int i = top_bit - 1; i >= 0; i--) {
      /* We get our initial data in GPR0 and we write the final data out to
       * GPR0 but we use GPR1 as our scratch register.
       */
      unsigned src_reg = i == top_bit - 1 ? MI_ALU_REG0 : MI_ALU_REG1;
      unsigned dst_reg = i == 0 ? MI_ALU_REG0 : MI_ALU_REG1;

      /* Shift the current value left by 1 */
      append_alu(MI_ALU_LOAD, MI_ALU_SRCA, src_reg);
      append_alu(MI_ALU_LOAD, MI_ALU_SRCB, src_reg);
      append_alu(MI_ALU_ADD, 0, 0);

      if (N & (1 << i)) {
         /* Store ACCU to R1 and add R0 to R1 */
         append_alu(MI_ALU_STORE, MI_ALU_REG1, MI_ALU_ACCU);
         append_alu(MI_ALU_LOAD, MI_ALU_SRCA, MI_ALU_REG0);
         append_alu(MI_ALU_LOAD, MI_ALU_SRCB, MI_ALU_REG1);
         append_alu(MI_ALU_ADD, 0, 0);
      }

      append_alu(MI_ALU_STORE, dst_reg, MI_ALU_ACCU);
   }

#undef append_alu
}

static void
emit_mul_gpr0(struct anv_batch *batch, uint32_t N)
{
   uint32_t num_dwords;
   build_alu_multiply_gpr0(NULL, &num_dwords, N);

   uint32_t *dw = anv_batch_emitn(batch, 1 + num_dwords, GENX(MI_MATH));
   build_alu_multiply_gpr0(dw + 1, &num_dwords, N);
}

#endif /* GEN_IS_HASWELL || GEN_GEN >= 8 */

static void
load_indirect_parameters(struct anv_cmd_buffer *cmd_buffer,
                         struct anv_buffer *buffer, uint64_t offset,
                         bool indexed)
{
   struct anv_batch *batch = &cmd_buffer->batch;
   struct anv_bo *bo = buffer->bo;
   uint32_t bo_offset = buffer->offset + offset;

   emit_lrm(batch, GEN7_3DPRIM_VERTEX_COUNT, bo, bo_offset);

   unsigned view_count = anv_subpass_view_count(cmd_buffer->state.subpass);
   if (view_count > 1) {
#if GEN_IS_HASWELL || GEN_GEN >= 8
      emit_lrm(batch, CS_GPR(0), bo, bo_offset + 4);
      emit_mul_gpr0(batch, view_count);
      emit_lrr(batch, GEN7_3DPRIM_INSTANCE_COUNT, CS_GPR(0));
#else
      anv_finishme("Multiview + indirect draw requires MI_MATH\n"
                   "MI_MATH is not supported on Ivy Bridge");
      emit_lrm(batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
#endif
   } else {
      emit_lrm(batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
   }

   emit_lrm(batch, GEN7_3DPRIM_START_VERTEX, bo, bo_offset + 8);

   if (indexed) {
      emit_lrm(batch, GEN7_3DPRIM_BASE_VERTEX, bo, bo_offset + 12);
      emit_lrm(batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 16);
   } else {
      emit_lrm(batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 12);
      emit_lri(batch, GEN7_3DPRIM_BASE_VERTEX, 0);
   }
}

void genX(CmdDrawIndirect)(
    VkCommandBuffer                             commandBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset,
    uint32_t                                    drawCount,
    uint32_t                                    stride)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
   const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   genX(cmd_buffer_flush_state)(cmd_buffer);

   for (uint32_t i = 0; i < drawCount; i++) {
      struct anv_bo *bo = buffer->bo;
      uint32_t bo_offset = buffer->offset + offset;

      if (vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance)
         emit_base_vertex_instance_bo(cmd_buffer, bo, bo_offset + 8);
      if (vs_prog_data->uses_drawid)
         emit_draw_index(cmd_buffer, i);

      load_indirect_parameters(cmd_buffer, buffer, offset, false);

      anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
         prim.IndirectParameterEnable  = true;
         prim.VertexAccessType         = SEQUENTIAL;
         prim.PrimitiveTopologyType    = pipeline->topology;
      }

      offset += stride;
   }
}

void genX(CmdDrawIndexedIndirect)(
    VkCommandBuffer                             commandBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset,
    uint32_t                                    drawCount,
    uint32_t                                    stride)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
   const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   genX(cmd_buffer_flush_state)(cmd_buffer);

   for (uint32_t i = 0; i < drawCount; i++) {
      struct anv_bo *bo = buffer->bo;
      uint32_t bo_offset = buffer->offset + offset;

      /* TODO: We need to stomp base vertex to 0 somehow */
      if (vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance)
         emit_base_vertex_instance_bo(cmd_buffer, bo, bo_offset + 12);
      if (vs_prog_data->uses_drawid)
         emit_draw_index(cmd_buffer, i);

      load_indirect_parameters(cmd_buffer, buffer, offset, true);

      anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
         prim.IndirectParameterEnable  = true;
         prim.VertexAccessType         = RANDOM;
         prim.PrimitiveTopologyType    = pipeline->topology;
      }

      offset += stride;
   }
}

static VkResult
flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   struct anv_state surfaces = { 0, }, samplers = { 0, };
   VkResult result;

   result = emit_binding_table(cmd_buffer, MESA_SHADER_COMPUTE, &surfaces);
   if (result != VK_SUCCESS) {
      assert(result == VK_ERROR_OUT_OF_DEVICE_MEMORY);

      result = anv_cmd_buffer_new_binding_table_block(cmd_buffer);
      if (result != VK_SUCCESS)
         return result;

      /* Re-emit state base addresses so we get the new surface state base
       * address before we start emitting binding tables etc.
       */
      genX(cmd_buffer_emit_state_base_address)(cmd_buffer);

      result = emit_binding_table(cmd_buffer, MESA_SHADER_COMPUTE, &surfaces);
      if (result != VK_SUCCESS) {
         anv_batch_set_error(&cmd_buffer->batch, result);
         return result;
      }
   }

   result = emit_samplers(cmd_buffer, MESA_SHADER_COMPUTE, &samplers);
   if (result != VK_SUCCESS) {
      anv_batch_set_error(&cmd_buffer->batch, result);
      return result;
   }

   uint32_t iface_desc_data_dw[GENX(INTERFACE_DESCRIPTOR_DATA_length)];
   struct GENX(INTERFACE_DESCRIPTOR_DATA) desc = {
      .BindingTablePointer = surfaces.offset,
      .SamplerStatePointer = samplers.offset,
   };
   GENX(INTERFACE_DESCRIPTOR_DATA_pack)(NULL, iface_desc_data_dw, &desc);

   struct anv_state state =
      anv_cmd_buffer_merge_dynamic(cmd_buffer, iface_desc_data_dw,
                                   pipeline->interface_descriptor_data,
                                   GENX(INTERFACE_DESCRIPTOR_DATA_length),
                                   64);

   uint32_t size = GENX(INTERFACE_DESCRIPTOR_DATA_length) * sizeof(uint32_t);
   anv_batch_emit(&cmd_buffer->batch,
                  GENX(MEDIA_INTERFACE_DESCRIPTOR_LOAD), mid) {
      mid.InterfaceDescriptorTotalLength        = size;
      mid.InterfaceDescriptorDataStartAddress   = state.offset;
   }

   return VK_SUCCESS;
}

void
genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   MAYBE_UNUSED VkResult result;

   assert(pipeline->active_stages == VK_SHADER_STAGE_COMPUTE_BIT);

   genX(cmd_buffer_config_l3)(cmd_buffer, pipeline->urb.l3_config);

   genX(flush_pipeline_select_gpgpu)(cmd_buffer);

   if (cmd_buffer->state.compute_dirty & ANV_CMD_DIRTY_PIPELINE) {
      /* From the Sky Lake PRM Vol 2a, MEDIA_VFE_STATE:
       *
       *    "A stalling PIPE_CONTROL is required before MEDIA_VFE_STATE unless
       *    the only bits that are changed are scoreboard related: Scoreboard
       *    Enable, Scoreboard Type, Scoreboard Mask, Scoreboard * Delta. For
       *    these scoreboard related states, a MEDIA_STATE_FLUSH is
       *    sufficient."
       */
      cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_CS_STALL_BIT;
      genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

      anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);
   }

   if ((cmd_buffer->state.descriptors_dirty & VK_SHADER_STAGE_COMPUTE_BIT) ||
       (cmd_buffer->state.compute_dirty & ANV_CMD_DIRTY_PIPELINE)) {
      /* FIXME: figure out descriptors for gen7 */
      result = flush_compute_descriptor_set(cmd_buffer);
      if (result != VK_SUCCESS)
         return;

      cmd_buffer->state.descriptors_dirty &= ~VK_SHADER_STAGE_COMPUTE_BIT;
   }

   if (cmd_buffer->state.push_constants_dirty & VK_SHADER_STAGE_COMPUTE_BIT) {
      struct anv_state push_state =
         anv_cmd_buffer_cs_push_constants(cmd_buffer);

      if (push_state.alloc_size) {
         anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_CURBE_LOAD), curbe) {
            curbe.CURBETotalDataLength    = push_state.alloc_size;
            curbe.CURBEDataStartAddress   = push_state.offset;
         }
      }
   }

   cmd_buffer->state.compute_dirty = 0;

   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
}

#if GEN_GEN == 7

static VkResult
verify_cmd_parser(const struct anv_device *device,
                  int required_version,
                  const char *function)
{
   if (device->instance->physicalDevice.cmd_parser_version < required_version) {
      return vk_errorf(VK_ERROR_FEATURE_NOT_PRESENT,
                       "cmd parser version %d is required for %s",
                       required_version, function);
   } else {
      return VK_SUCCESS;
   }
}

#endif

void genX(CmdDispatch)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    x,
    uint32_t                                    y,
    uint32_t                                    z)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   const struct brw_cs_prog_data *prog_data = get_cs_prog_data(pipeline);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   if (prog_data->uses_num_work_groups) {
      struct anv_state state =
         anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, 12, 4);
      uint32_t *sizes = state.map;
      sizes[0] = x;
      sizes[1] = y;
      sizes[2] = z;
      anv_state_flush(cmd_buffer->device, state);
      cmd_buffer->state.num_workgroups_offset = state.offset;
      cmd_buffer->state.num_workgroups_bo =
         &cmd_buffer->device->dynamic_state_pool.block_pool.bo;
   }

   genX(cmd_buffer_flush_compute_state)(cmd_buffer);

   anv_batch_emit(&cmd_buffer->batch, GENX(GPGPU_WALKER), ggw) {
      ggw.SIMDSize                     = prog_data->simd_size / 16;
      ggw.ThreadDepthCounterMaximum    = 0;
      ggw.ThreadHeightCounterMaximum   = 0;
      ggw.ThreadWidthCounterMaximum    = prog_data->threads - 1;
      ggw.ThreadGroupIDXDimension      = x;
      ggw.ThreadGroupIDYDimension      = y;
      ggw.ThreadGroupIDZDimension      = z;
      ggw.RightExecutionMask           = pipeline->cs_right_mask;
      ggw.BottomExecutionMask          = 0xffffffff;
   }

   anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_STATE_FLUSH), msf);
}

#define GPGPU_DISPATCHDIMX 0x2500
#define GPGPU_DISPATCHDIMY 0x2504
#define GPGPU_DISPATCHDIMZ 0x2508

#define MI_PREDICATE_SRC0  0x2400
#define MI_PREDICATE_SRC1  0x2408

void genX(CmdDispatchIndirect)(
    VkCommandBuffer                             commandBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   const struct brw_cs_prog_data *prog_data = get_cs_prog_data(pipeline);
   struct anv_bo *bo = buffer->bo;
   uint32_t bo_offset = buffer->offset + offset;
   struct anv_batch *batch = &cmd_buffer->batch;

#if GEN_GEN == 7
   /* Linux 4.4 added command parser version 5 which allows the GPGPU
    * indirect dispatch registers to be written.
    */
   if (verify_cmd_parser(cmd_buffer->device, 5,
                         "vkCmdDispatchIndirect") != VK_SUCCESS)
      return;
#endif

   if (prog_data->uses_num_work_groups) {
      cmd_buffer->state.num_workgroups_offset = bo_offset;
      cmd_buffer->state.num_workgroups_bo = bo;
   }

   genX(cmd_buffer_flush_compute_state)(cmd_buffer);

   emit_lrm(batch, GPGPU_DISPATCHDIMX, bo, bo_offset);
   emit_lrm(batch, GPGPU_DISPATCHDIMY, bo, bo_offset + 4);
   emit_lrm(batch, GPGPU_DISPATCHDIMZ, bo, bo_offset + 8);

#if GEN_GEN <= 7
   /* Clear upper 32-bits of SRC0 and all 64-bits of SRC1 */
   emit_lri(batch, MI_PREDICATE_SRC0 + 4, 0);
   emit_lri(batch, MI_PREDICATE_SRC1 + 0, 0);
   emit_lri(batch, MI_PREDICATE_SRC1 + 4, 0);

   /* Load compute_dispatch_indirect_x_size into SRC0 */
   emit_lrm(batch, MI_PREDICATE_SRC0, bo, bo_offset + 0);

   /* predicate = (compute_dispatch_indirect_x_size == 0); */
   anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOAD;
      mip.CombineOperation = COMBINE_SET;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }

   /* Load compute_dispatch_indirect_y_size into SRC0 */
   emit_lrm(batch, MI_PREDICATE_SRC0, bo, bo_offset + 4);

   /* predicate |= (compute_dispatch_indirect_y_size == 0); */
   anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOAD;
      mip.CombineOperation = COMBINE_OR;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }

   /* Load compute_dispatch_indirect_z_size into SRC0 */
   emit_lrm(batch, MI_PREDICATE_SRC0, bo, bo_offset + 8);

   /* predicate |= (compute_dispatch_indirect_z_size == 0); */
   anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOAD;
      mip.CombineOperation = COMBINE_OR;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }

   /* predicate = !predicate; */
#define COMPARE_FALSE                           1
   anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOADINV;
      mip.CombineOperation = COMBINE_OR;
      mip.CompareOperation = COMPARE_FALSE;
   }
#endif

   anv_batch_emit(batch, GENX(GPGPU_WALKER), ggw) {
      ggw.IndirectParameterEnable      = true;
      ggw.PredicateEnable              = GEN_GEN <= 7;
      ggw.SIMDSize                     = prog_data->simd_size / 16;
      ggw.ThreadDepthCounterMaximum    = 0;
      ggw.ThreadHeightCounterMaximum   = 0;
      ggw.ThreadWidthCounterMaximum    = prog_data->threads - 1;
      ggw.RightExecutionMask           = pipeline->cs_right_mask;
      ggw.BottomExecutionMask          = 0xffffffff;
   }

   anv_batch_emit(batch, GENX(MEDIA_STATE_FLUSH), msf);
}

static void
genX(flush_pipeline_select)(struct anv_cmd_buffer *cmd_buffer,
                            uint32_t pipeline)
{
   if (cmd_buffer->state.current_pipeline == pipeline)
      return;

#if GEN_GEN >= 8 && GEN_GEN < 10
   /* From the Broadwell PRM, Volume 2a: Instructions, PIPELINE_SELECT:
    *
    *   Software must clear the COLOR_CALC_STATE Valid field in
    *   3DSTATE_CC_STATE_POINTERS command prior to send a PIPELINE_SELECT
    *   with Pipeline Select set to GPGPU.
    *
    * The internal hardware docs recommend the same workaround for Gen9
    * hardware too.
    */
   if (pipeline == GPGPU)
      anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), t);
#endif

   /* From "BXML » GT » MI » vol1a GPU Overview » [Instruction]
    * PIPELINE_SELECT [DevBWR+]":
    *
    *   Project: DEVSNB+
    *
    *   Software must ensure all the write caches are flushed through a
    *   stalling PIPE_CONTROL command followed by another PIPE_CONTROL
    *   command to invalidate read only caches prior to programming
    *   MI_PIPELINE_SELECT command to change the Pipeline Select Mode.
    */
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      pc.RenderTargetCacheFlushEnable  = true;
      pc.DepthCacheFlushEnable         = true;
      pc.DCFlushEnable                 = true;
      pc.PostSyncOperation             = NoWrite;
      pc.CommandStreamerStallEnable    = true;
   }

   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      pc.TextureCacheInvalidationEnable   = true;
      pc.ConstantCacheInvalidationEnable  = true;
      pc.StateCacheInvalidationEnable     = true;
      pc.InstructionCacheInvalidateEnable = true;
      pc.PostSyncOperation                = NoWrite;
   }

   anv_batch_emit(&cmd_buffer->batch, GENX(PIPELINE_SELECT), ps) {
#if GEN_GEN >= 9
      ps.MaskBits = 3;
#endif
      ps.PipelineSelection = pipeline;
   }

   cmd_buffer->state.current_pipeline = pipeline;
}

void
genX(flush_pipeline_select_3d)(struct anv_cmd_buffer *cmd_buffer)
{
   genX(flush_pipeline_select)(cmd_buffer, _3D);
}

void
genX(flush_pipeline_select_gpgpu)(struct anv_cmd_buffer *cmd_buffer)
{
   genX(flush_pipeline_select)(cmd_buffer, GPGPU);
}

void
genX(cmd_buffer_emit_gen7_depth_flush)(struct anv_cmd_buffer *cmd_buffer)
{
   if (GEN_GEN >= 8)
      return;

   /* From the Haswell PRM, documentation for 3DSTATE_DEPTH_BUFFER:
    *
    *    "Restriction: Prior to changing Depth/Stencil Buffer state (i.e., any
    *    combination of 3DSTATE_DEPTH_BUFFER, 3DSTATE_CLEAR_PARAMS,
    *    3DSTATE_STENCIL_BUFFER, 3DSTATE_HIER_DEPTH_BUFFER) SW must first
    *    issue a pipelined depth stall (PIPE_CONTROL with Depth Stall bit
    *    set), followed by a pipelined depth cache flush (PIPE_CONTROL with
    *    Depth Flush Bit set, followed by another pipelined depth stall
    *    (PIPE_CONTROL with Depth Stall Bit set), unless SW can otherwise
    *    guarantee that the pipeline from WM onwards is already flushed (e.g.,
    *    via a preceding MI_FLUSH)."
    */
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe) {
      pipe.DepthStallEnable = true;
   }
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe) {
      pipe.DepthCacheFlushEnable = true;
   }
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe) {
      pipe.DepthStallEnable = true;
   }
}

static void
cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_device *device = cmd_buffer->device;
   const struct anv_image_view *iview =
      anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
   const struct anv_image *image = iview ? iview->image : NULL;

   /* FIXME: Width and Height are wrong */

   genX(cmd_buffer_emit_gen7_depth_flush)(cmd_buffer);

   uint32_t *dw = anv_batch_emit_dwords(&cmd_buffer->batch,
                                        device->isl_dev.ds.size / 4);
   if (dw == NULL)
      return;

   struct isl_depth_stencil_hiz_emit_info info = {
      .mocs = device->default_mocs,
   };

   if (iview)
      info.view = &iview->isl;

   if (image && (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT)) {
      info.depth_surf = &image->depth_surface.isl;

      info.depth_address =
         anv_batch_emit_reloc(&cmd_buffer->batch,
                              dw + device->isl_dev.ds.depth_offset / 4,
                              image->bo,
                              image->offset + image->depth_surface.offset);

      const uint32_t ds =
         cmd_buffer->state.subpass->depth_stencil_attachment.attachment;
      info.hiz_usage = cmd_buffer->state.attachments[ds].aux_usage;
      if (info.hiz_usage == ISL_AUX_USAGE_HIZ) {
         info.hiz_surf = &image->aux_surface.isl;

         info.hiz_address =
            anv_batch_emit_reloc(&cmd_buffer->batch,
                                 dw + device->isl_dev.ds.hiz_offset / 4,
                                 image->bo,
                                 image->offset + image->aux_surface.offset);

         info.depth_clear_value = ANV_HZ_FC_VAL;
      }
   }

   if (image && (image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT)) {
      info.stencil_surf = &image->stencil_surface.isl;

      info.stencil_address =
         anv_batch_emit_reloc(&cmd_buffer->batch,
                              dw + device->isl_dev.ds.stencil_offset / 4,
                              image->bo,
                              image->offset + image->stencil_surface.offset);
   }

   isl_emit_depth_stencil_hiz_s(&device->isl_dev, dw, &info);

   cmd_buffer->state.hiz_enabled = info.hiz_usage == ISL_AUX_USAGE_HIZ;
}


/**
 * @brief Perform any layout transitions required at the beginning and/or end
 *        of the current subpass for depth buffers.
 *
 * TODO: Consider preprocessing the attachment reference array at render pass
 *       create time to determine if no layout transition is needed at the
 *       beginning and/or end of each subpass.
 *
 * @param cmd_buffer The command buffer the transition is happening within.
 * @param subpass_end If true, marks that the transition is happening at the
 *                    end of the subpass.
 */
static void
cmd_buffer_subpass_transition_layouts(struct anv_cmd_buffer * const cmd_buffer,
                                      const bool subpass_end)
{
   /* We need a non-NULL command buffer. */
   assert(cmd_buffer);

   const struct anv_cmd_state * const cmd_state = &cmd_buffer->state;
   const struct anv_subpass * const subpass = cmd_state->subpass;

   /* This function must be called within a subpass. */
   assert(subpass);

   /* If there are attachment references, the array shouldn't be NULL.
    */
   if (subpass->attachment_count > 0)
      assert(subpass->attachments);

   /* Iterate over the array of attachment references. */
   for (const VkAttachmentReference *att_ref = subpass->attachments;
        att_ref < subpass->attachments + subpass->attachment_count; att_ref++) {

      /* If the attachment is unused, we can't perform a layout transition. */
      if (att_ref->attachment == VK_ATTACHMENT_UNUSED)
         continue;

      /* This attachment index shouldn't go out of bounds. */
      assert(att_ref->attachment < cmd_state->pass->attachment_count);

      const struct anv_render_pass_attachment * const att_desc =
         &cmd_state->pass->attachments[att_ref->attachment];
      struct anv_attachment_state * const att_state =
         &cmd_buffer->state.attachments[att_ref->attachment];

      /* The attachment should not be used in a subpass after its last. */
      assert(att_desc->last_subpass_idx >= anv_get_subpass_id(cmd_state));

      if (subpass_end && anv_get_subpass_id(cmd_state) <
          att_desc->last_subpass_idx) {
         /* We're calling this function on a buffer twice in one subpass and
          * this is not the last use of the buffer. The layout should not have
          * changed from the first call and no transition is necessary.
          */
         assert(att_ref->layout == att_state->current_layout);
         continue;
      }

      /* Get the appropriate target layout for this attachment. */
      const VkImageLayout target_layout = subpass_end ?
         att_desc->final_layout : att_ref->layout;

      /* The attachment index must be less than the number of attachments
       * within the framebuffer.
       */
      assert(att_ref->attachment < cmd_state->framebuffer->attachment_count);

      const struct anv_image_view * const iview =
         cmd_state->framebuffer->attachments[att_ref->attachment];
      const struct anv_image * const image = iview->image;

      /* Perform the layout transition. */
      if (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT) {
         transition_depth_buffer(cmd_buffer, image,
                                 att_state->current_layout, target_layout);
         att_state->aux_usage =
            anv_layout_to_aux_usage(&cmd_buffer->device->info, image,
                                    image->aspects, target_layout);
      } else if (image->aspects == VK_IMAGE_ASPECT_COLOR_BIT) {
         transition_color_buffer(cmd_buffer, image,
                                 iview->isl.base_level, 1,
                                 iview->isl.base_array_layer,
                                 iview->isl.array_len,
                                 att_state->current_layout, target_layout);
      }

      att_state->current_layout = target_layout;
   }
}

static void
genX(cmd_buffer_set_subpass)(struct anv_cmd_buffer *cmd_buffer,
                             struct anv_subpass *subpass)
{
   cmd_buffer->state.subpass = subpass;

   cmd_buffer->state.dirty |= ANV_CMD_DIRTY_RENDER_TARGETS;

   /* Our implementation of VK_KHR_multiview uses instancing to draw the
    * different views.  If the client asks for instancing, we need to use the
    * Instance Data Step Rate to ensure that we repeat the client's
    * per-instance data once for each view.  Since this bit is in
    * VERTEX_BUFFER_STATE on gen7, we need to dirty vertex buffers at the top
    * of each subpass.
    */
   if (GEN_GEN == 7)
      cmd_buffer->state.vb_dirty |= ~0;

   /* Perform transitions to the subpass layout before any writes have
    * occurred.
    */
   cmd_buffer_subpass_transition_layouts(cmd_buffer, false);

   cmd_buffer_emit_depth_stencil(cmd_buffer);

   anv_cmd_buffer_clear_subpass(cmd_buffer);
}

void genX(CmdBeginRenderPass)(
    VkCommandBuffer                             commandBuffer,
    const VkRenderPassBeginInfo*                pRenderPassBegin,
    VkSubpassContents                           contents)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_render_pass, pass, pRenderPassBegin->renderPass);
   ANV_FROM_HANDLE(anv_framebuffer, framebuffer, pRenderPassBegin->framebuffer);

   cmd_buffer->state.framebuffer = framebuffer;
   cmd_buffer->state.pass = pass;
   cmd_buffer->state.render_area = pRenderPassBegin->renderArea;
   VkResult result =
      genX(cmd_buffer_setup_attachments)(cmd_buffer, pass, pRenderPassBegin);

   /* If we failed to setup the attachments we should not try to go further */
   if (result != VK_SUCCESS) {
      assert(anv_batch_has_error(&cmd_buffer->batch));
      return;
   }

   genX(flush_pipeline_select_3d)(cmd_buffer);

   genX(cmd_buffer_set_subpass)(cmd_buffer, pass->subpasses);

   cmd_buffer->state.pending_pipe_bits |=
      cmd_buffer->state.pass->subpass_flushes[0];
}

void genX(CmdNextSubpass)(
    VkCommandBuffer                             commandBuffer,
    VkSubpassContents                           contents)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);

   anv_cmd_buffer_resolve_subpass(cmd_buffer);

   /* Perform transitions to the final layout after all writes have occurred.
    */
   cmd_buffer_subpass_transition_layouts(cmd_buffer, true);

   genX(cmd_buffer_set_subpass)(cmd_buffer, cmd_buffer->state.subpass + 1);

   uint32_t subpass_id = anv_get_subpass_id(&cmd_buffer->state);
   cmd_buffer->state.pending_pipe_bits |=
      cmd_buffer->state.pass->subpass_flushes[subpass_id];
}

void genX(CmdEndRenderPass)(
    VkCommandBuffer                             commandBuffer)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   anv_cmd_buffer_resolve_subpass(cmd_buffer);

   /* Perform transitions to the final layout after all writes have occurred.
    */
   cmd_buffer_subpass_transition_layouts(cmd_buffer, true);

   cmd_buffer->state.pending_pipe_bits |=
      cmd_buffer->state.pass->subpass_flushes[cmd_buffer->state.pass->subpass_count];

   cmd_buffer->state.hiz_enabled = false;

#ifndef NDEBUG
   anv_dump_add_framebuffer(cmd_buffer, cmd_buffer->state.framebuffer);
#endif

   /* Remove references to render pass specific state. This enables us to
    * detect whether or not we're in a renderpass.
    */
   cmd_buffer->state.framebuffer = NULL;
   cmd_buffer->state.pass = NULL;
   cmd_buffer->state.subpass = NULL;
}