summaryrefslogtreecommitdiff
path: root/src/intel/vulkan/genX_cmd_buffer.c
blob: 0b4bbe328d20df11b2d6bad4091cd7011e3892ec (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
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
/*
 * 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 "util/fast_idiv_by_const.h"

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

/* We reserve :
 *    - GPR 14 for secondary command buffer returns
 *    - GPR 15 for conditional rendering
 */
#define GEN_MI_BUILDER_NUM_ALLOC_GPRS 14
#define __gen_get_batch_dwords anv_batch_emit_dwords
#define __gen_address_offset anv_address_add
#include "common/gen_mi_builder.h"

static void genX(flush_pipeline_select)(struct anv_cmd_buffer *cmd_buffer,
                                        uint32_t pipeline);

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;
   }
}

void
genX(cmd_buffer_emit_state_base_address)(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_device *device = cmd_buffer->device;
   UNUSED const struct gen_device_info *devinfo = &device->info;
   uint32_t mocs = isl_mocs(&device->isl_dev, 0);

   /* If we are emitting a new state base address we probably need to re-emit
    * binding tables.
    */
   cmd_buffer->state.descriptors_dirty |= ~0;

   /* 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;
#if GEN_GEN >= 12
      pc.TileCacheFlushEnable = true;
#endif
#if GEN_GEN == 12
      /* GEN:BUG:1606662791:
       *
       *   Software must program PIPE_CONTROL command with "HDC Pipeline
       *   Flush" prior to programming of the below two non-pipeline state :
       *      * STATE_BASE_ADDRESS
       *      * 3DSTATE_BINDING_TABLE_POOL_ALLOC
       */
      if (devinfo->revision == 0 /* A0 */)
         pc.HDCPipelineFlushEnable = true;
#endif
   }

#if GEN_GEN == 12
   /* GEN:BUG:1607854226:
    *
    *  Workaround the non pipelined state not applying in MEDIA/GPGPU pipeline
    *  mode by putting the pipeline temporarily in 3D mode.
    */
   uint32_t gen12_wa_pipeline = cmd_buffer->state.current_pipeline;
   genX(flush_pipeline_select_3d)(cmd_buffer);
#endif

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

      sba.StatelessDataPortAccessMOCS = mocs;

      sba.SurfaceStateBaseAddress =
         anv_cmd_buffer_surface_base_address(cmd_buffer);
      sba.SurfaceStateMOCS = mocs;
      sba.SurfaceStateBaseAddressModifyEnable = true;

      sba.DynamicStateBaseAddress =
         (struct anv_address) { device->dynamic_state_pool.block_pool.bo, 0 };
      sba.DynamicStateMOCS = mocs;
      sba.DynamicStateBaseAddressModifyEnable = true;

      sba.IndirectObjectBaseAddress = (struct anv_address) { NULL, 0 };
      sba.IndirectObjectMOCS = mocs;
      sba.IndirectObjectBaseAddressModifyEnable = true;

      sba.InstructionBaseAddress =
         (struct anv_address) { device->instruction_state_pool.block_pool.bo, 0 };
      sba.InstructionMOCS = 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.IndirectObjectBufferSize     = 0xfffff;
      if (device->physical->use_softpin) {
         /* With softpin, we use fixed addresses so we actually know how big
          * our base addresses are.
          */
         sba.DynamicStateBufferSize    = DYNAMIC_STATE_POOL_SIZE / 4096;
         sba.InstructionBufferSize     = INSTRUCTION_STATE_POOL_SIZE / 4096;
      } else {
         sba.DynamicStateBufferSize    = 0xfffff;
         sba.InstructionBufferSize     = 0xfffff;
      }
      sba.GeneralStateBufferSizeModifyEnable    = true;
      sba.IndirectObjectBufferSizeModifyEnable  = true;
      sba.DynamicStateBufferSizeModifyEnable    = true;
      sba.InstructionBuffersizeModifyEnable     = true;
#  else
      /* On gen7, we have upper bounds instead.  According to the docs,
       * setting an upper bound of zero means that no bounds checking is
       * performed so, in theory, we should be able to leave them zero.
       * However, border color is broken and the GPU bounds-checks anyway.
       * To avoid this and other potential problems, we may as well set it
       * for everything.
       */
      sba.GeneralStateAccessUpperBound =
         (struct anv_address) { .bo = NULL, .offset = 0xfffff000 };
      sba.GeneralStateAccessUpperBoundModifyEnable = true;
      sba.DynamicStateAccessUpperBound =
         (struct anv_address) { .bo = NULL, .offset = 0xfffff000 };
      sba.DynamicStateAccessUpperBoundModifyEnable = true;
      sba.InstructionAccessUpperBound =
         (struct anv_address) { .bo = NULL, .offset = 0xfffff000 };
      sba.InstructionAccessUpperBoundModifyEnable = true;
#  endif
#  if (GEN_GEN >= 9)
      if (cmd_buffer->device->physical->use_softpin) {
         sba.BindlessSurfaceStateBaseAddress = (struct anv_address) {
            .bo = device->surface_state_pool.block_pool.bo,
            .offset = 0,
         };
         sba.BindlessSurfaceStateSize = (1 << 20) - 1;
      } else {
         sba.BindlessSurfaceStateBaseAddress = ANV_NULL_ADDRESS;
         sba.BindlessSurfaceStateSize = 0;
      }
      sba.BindlessSurfaceStateMOCS = mocs;
      sba.BindlessSurfaceStateBaseAddressModifyEnable = true;
#  endif
#  if (GEN_GEN >= 10)
      sba.BindlessSamplerStateBaseAddress = (struct anv_address) { NULL, 0 };
      sba.BindlessSamplerStateMOCS = mocs;
      sba.BindlessSamplerStateBaseAddressModifyEnable = true;
      sba.BindlessSamplerStateBufferSize = 0;
#  endif
   }

#if GEN_GEN == 12
   /* GEN:BUG:1607854226:
    *
    *  Put the pipeline back into its current mode.
    */
   if (gen12_wa_pipeline != UINT32_MAX)
      genX(flush_pipeline_select)(cmd_buffer, gen12_wa_pipeline);
#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_reloc(struct anv_cmd_buffer *cmd_buffer,
                  struct anv_state state, struct anv_address addr)
{
   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,
                         addr.bo, addr.offset, NULL);
   if (result != VK_SUCCESS)
      anv_batch_set_error(&cmd_buffer->batch, result);
}

static void
add_surface_state_relocs(struct anv_cmd_buffer *cmd_buffer,
                         struct anv_surface_state state)
{
   const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;

   assert(!anv_address_is_null(state.address));
   add_surface_reloc(cmd_buffer, state.state, state.address);

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

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

static bool
isl_color_value_requires_conversion(union isl_color_value color,
                                    const struct isl_surf *surf,
                                    const struct isl_view *view)
{
   if (surf->format == view->format && isl_swizzle_is_identity(view->swizzle))
      return false;

   uint32_t surf_pack[4] = { 0, 0, 0, 0 };
   isl_color_value_pack(&color, surf->format, surf_pack);

   uint32_t view_pack[4] = { 0, 0, 0, 0 };
   union isl_color_value swiz_color =
      isl_color_value_swizzle_inv(color, view->swizzle);
   isl_color_value_pack(&swiz_color, view->format, view_pack);

   return memcmp(surf_pack, view_pack, sizeof(surf_pack)) != 0;
}

static bool
anv_can_fast_clear_color_view(struct anv_device * device,
                              struct anv_image_view *iview,
                              VkImageLayout layout,
                              union isl_color_value clear_color,
                              uint32_t num_layers,
                              VkRect2D render_area)
{
   if (iview->planes[0].isl.base_array_layer >=
       anv_image_aux_layers(iview->image, VK_IMAGE_ASPECT_COLOR_BIT,
                            iview->planes[0].isl.base_level))
      return false;

   /* Start by getting the fast clear type.  We use the first subpass
    * layout here because we don't want to fast-clear if the first subpass
    * to use the attachment can't handle fast-clears.
    */
   enum anv_fast_clear_type fast_clear_type =
      anv_layout_to_fast_clear_type(&device->info, iview->image,
                                    VK_IMAGE_ASPECT_COLOR_BIT,
                                    layout);
   switch (fast_clear_type) {
   case ANV_FAST_CLEAR_NONE:
      return false;
   case ANV_FAST_CLEAR_DEFAULT_VALUE:
      if (!isl_color_value_is_zero(clear_color, iview->planes[0].isl.format))
         return false;
      break;
   case ANV_FAST_CLEAR_ANY:
      break;
   }

   /* 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)
      return false;

   /* On Broadwell and earlier, we can only handle 0/1 clear colors */
   if (GEN_GEN <= 8 &&
       !isl_color_value_is_zero_one(clear_color, iview->planes[0].isl.format))
      return false;

   /* If the clear color is one that would require non-trivial format
    * conversion on resolve, we don't bother with the fast clear.  This
    * shouldn't be common as most clear colors are 0/1 and the most common
    * format re-interpretation is for sRGB.
    */
   if (isl_color_value_requires_conversion(clear_color,
                                           &iview->image->planes[0].surface.isl,
                                           &iview->planes[0].isl)) {
      anv_perf_warn(device, iview,
                    "Cannot fast-clear to colors which would require "
                    "format conversion on resolve");
      return false;
   }

   /* We only allow fast clears to the first slice of an image (level 0,
    * layer 0) and only for the entire slice.  This guarantees us that, at
    * any given time, there is only one clear color on any given image at
    * any given time.  At the time of our testing (Jan 17, 2018), there
    * were no known applications which would benefit from fast-clearing
    * more than just the first slice.
    */
   if (iview->planes[0].isl.base_level > 0 ||
       iview->planes[0].isl.base_array_layer > 0) {
      anv_perf_warn(device, iview->image,
                    "Rendering with multi-lod or multi-layer framebuffer "
                    "with LOAD_OP_LOAD and baseMipLevel > 0 or "
                    "baseArrayLayer > 0.  Not fast clearing.");
      return false;
   }

   if (num_layers > 1) {
      anv_perf_warn(device, iview->image,
                    "Rendering to a multi-layer framebuffer with "
                    "LOAD_OP_CLEAR.  Only fast-clearing the first slice");
   }

   return true;
}

static bool
anv_can_hiz_clear_ds_view(struct anv_device *device,
                          struct anv_image_view *iview,
                          VkImageLayout layout,
                          VkImageAspectFlags clear_aspects,
                          float depth_clear_value,
                          VkRect2D render_area)
{
   /* We don't do any HiZ or depth fast-clears on gen7 yet */
   if (GEN_GEN == 7)
      return false;

   /* If we're just clearing stencil, we can always HiZ clear */
   if (!(clear_aspects & VK_IMAGE_ASPECT_DEPTH_BIT))
      return true;

   /* We must have depth in order to have HiZ */
   if (!(iview->image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT))
      return false;

   const enum isl_aux_usage clear_aux_usage =
      anv_layout_to_aux_usage(&device->info, iview->image,
                              VK_IMAGE_ASPECT_DEPTH_BIT,
                              VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT,
                              layout);
   if (!blorp_can_hiz_clear_depth(&device->info,
                                  &iview->image->planes[0].surface.isl,
                                  clear_aux_usage,
                                  iview->planes[0].isl.base_level,
                                  iview->planes[0].isl.base_array_layer,
                                  render_area.offset.x,
                                  render_area.offset.y,
                                  render_area.offset.x +
                                  render_area.extent.width,
                                  render_area.offset.y +
                                  render_area.extent.height))
      return false;

   if (depth_clear_value != ANV_HZ_FC_VAL)
      return false;

   /* Only gen9+ supports returning ANV_HZ_FC_VAL when sampling a fast-cleared
    * portion of a HiZ buffer. Testing has revealed that Gen8 only supports
    * returning 0.0f. Gens prior to gen8 do not support this feature at all.
    */
   if (GEN_GEN == 8 && anv_can_sample_with_hiz(&device->info, iview->image))
      return false;

   /* If we got here, then we can fast clear */
   return true;
}

#define READ_ONCE(x) (*(volatile __typeof__(x) *)&(x))

#if GEN_GEN == 12
static void
anv_image_init_aux_tt(struct anv_cmd_buffer *cmd_buffer,
                      const struct anv_image *image,
                      VkImageAspectFlagBits aspect,
                      uint32_t base_level, uint32_t level_count,
                      uint32_t base_layer, uint32_t layer_count)
{
   uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);

   const struct anv_surface *surface = &image->planes[plane].surface;
   uint64_t base_address =
      anv_address_physical(anv_address_add(image->planes[plane].address,
                                           surface->offset));

   const struct isl_surf *isl_surf = &image->planes[plane].surface.isl;
   uint64_t format_bits = gen_aux_map_format_bits_for_isl_surf(isl_surf);

   /* We're about to live-update the AUX-TT.  We really don't want anyone else
    * trying to read it while we're doing this.  We could probably get away
    * with not having this stall in some cases if we were really careful but
    * it's better to play it safe.  Full stall the GPU.
    */
   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_END_OF_PIPE_SYNC_BIT;
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   for (uint32_t a = 0; a < layer_count; a++) {
      const uint32_t layer = base_layer + a;

      uint64_t start_offset_B = UINT64_MAX, end_offset_B = 0;
      for (uint32_t l = 0; l < level_count; l++) {
         const uint32_t level = base_level + l;

         uint32_t logical_array_layer, logical_z_offset_px;
         if (image->type == VK_IMAGE_TYPE_3D) {
            logical_array_layer = 0;

            /* If the given miplevel does not have this layer, then any higher
             * miplevels won't either because miplevels only get smaller the
             * higher the LOD.
             */
            assert(layer < image->extent.depth);
            if (layer >= anv_minify(image->extent.depth, level))
               break;
            logical_z_offset_px = layer;
         } else {
            assert(layer < image->array_size);
            logical_array_layer = layer;
            logical_z_offset_px = 0;
         }

         uint32_t slice_start_offset_B, slice_end_offset_B;
         isl_surf_get_image_range_B_tile(isl_surf, level,
                                         logical_array_layer,
                                         logical_z_offset_px,
                                         &slice_start_offset_B,
                                         &slice_end_offset_B);

         start_offset_B = MIN2(start_offset_B, slice_start_offset_B);
         end_offset_B = MAX2(end_offset_B, slice_end_offset_B);
      }

      /* Aux operates 64K at a time */
      start_offset_B = align_down_u64(start_offset_B, 64 * 1024);
      end_offset_B = align_u64(end_offset_B, 64 * 1024);

      for (uint64_t offset = start_offset_B;
           offset < end_offset_B; offset += 64 * 1024) {
         uint64_t address = base_address + offset;

         uint64_t aux_entry_addr64, *aux_entry_map;
         aux_entry_map = gen_aux_map_get_entry(cmd_buffer->device->aux_map_ctx,
                                               address, &aux_entry_addr64);

         assert(cmd_buffer->device->physical->use_softpin);
         struct anv_address aux_entry_address = {
            .bo = NULL,
            .offset = aux_entry_addr64,
         };

         const uint64_t old_aux_entry = READ_ONCE(*aux_entry_map);
         uint64_t new_aux_entry =
            (old_aux_entry & GEN_AUX_MAP_ADDRESS_MASK) | format_bits;

         if (isl_aux_usage_has_ccs(image->planes[plane].aux_usage))
            new_aux_entry |= GEN_AUX_MAP_ENTRY_VALID_BIT;

         gen_mi_store(&b, gen_mi_mem64(aux_entry_address),
                          gen_mi_imm(new_aux_entry));
      }
   }

   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_AUX_TABLE_INVALIDATE_BIT;
}
#endif /* GEN_GEN == 12 */

/* 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,
                        uint32_t base_layer, uint32_t layer_count,
                        VkImageLayout initial_layout,
                        VkImageLayout final_layout,
                        bool will_full_fast_clear)
{
   uint32_t depth_plane =
      anv_image_aspect_to_plane(image->aspects, VK_IMAGE_ASPECT_DEPTH_BIT);
   if (image->planes[depth_plane].aux_usage == ISL_AUX_USAGE_NONE)
      return;

#if GEN_GEN == 12
   if ((initial_layout == VK_IMAGE_LAYOUT_UNDEFINED ||
        initial_layout == VK_IMAGE_LAYOUT_PREINITIALIZED) &&
       cmd_buffer->device->physical->has_implicit_ccs &&
       cmd_buffer->device->info.has_aux_map) {
      anv_image_init_aux_tt(cmd_buffer, image, VK_IMAGE_ASPECT_DEPTH_BIT,
                            0, 1, base_layer, layer_count);
   }
#endif

   /* If will_full_fast_clear is set, the caller promises to fast-clear the
    * largest portion of the specified range as it can.  For depth images,
    * that means the entire image because we don't support multi-LOD HiZ.
    */
   assert(image->planes[0].surface.isl.levels == 1);
   if (will_full_fast_clear)
      return;

   const enum isl_aux_state initial_state =
      anv_layout_to_aux_state(&cmd_buffer->device->info, image,
                              VK_IMAGE_ASPECT_DEPTH_BIT,
                              initial_layout);
   const enum isl_aux_state final_state =
      anv_layout_to_aux_state(&cmd_buffer->device->info, image,
                              VK_IMAGE_ASPECT_DEPTH_BIT,
                              final_layout);

   const bool initial_depth_valid =
      isl_aux_state_has_valid_primary(initial_state);
   const bool initial_hiz_valid =
      isl_aux_state_has_valid_aux(initial_state);
   const bool final_needs_depth =
      isl_aux_state_has_valid_primary(final_state);
   const bool final_needs_hiz =
      isl_aux_state_has_valid_aux(final_state);

   /* Getting into the pass-through state for Depth is tricky and involves
    * both a resolve and an ambiguate.  We don't handle that state right now
    * as anv_layout_to_aux_state never returns it.
    */
   assert(final_state != ISL_AUX_STATE_PASS_THROUGH);

   if (final_needs_depth && !initial_depth_valid) {
      assert(initial_hiz_valid);
      anv_image_hiz_op(cmd_buffer, image, VK_IMAGE_ASPECT_DEPTH_BIT,
                       0, base_layer, layer_count, ISL_AUX_OP_FULL_RESOLVE);
   } else if (final_needs_hiz && !initial_hiz_valid) {
      assert(initial_depth_valid);
      anv_image_hiz_op(cmd_buffer, image, VK_IMAGE_ASPECT_DEPTH_BIT,
                       0, base_layer, layer_count, ISL_AUX_OP_AMBIGUATE);
   }
}

static inline bool
vk_image_layout_stencil_write_optimal(VkImageLayout layout)
{
   return layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
          layout == VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL ||
          layout == VK_IMAGE_LAYOUT_STENCIL_ATTACHMENT_OPTIMAL_KHR;
}

/* 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_stencil_buffer(struct anv_cmd_buffer *cmd_buffer,
                          const struct anv_image *image,
                          uint32_t base_level, uint32_t level_count,
                          uint32_t base_layer, uint32_t layer_count,
                          VkImageLayout initial_layout,
                          VkImageLayout final_layout,
                          bool will_full_fast_clear)
{
#if GEN_GEN == 7
   uint32_t plane = anv_image_aspect_to_plane(image->aspects,
                                              VK_IMAGE_ASPECT_STENCIL_BIT);

   /* On gen7, we have to store a texturable version of the stencil buffer in
    * a shadow whenever VK_IMAGE_USAGE_SAMPLED_BIT is set and copy back and
    * forth at strategic points. Stencil writes are only allowed in following
    * layouts:
    *
    *  - VK_IMAGE_LAYOUT_GENERAL
    *  - VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
    *  - VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
    *  - VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL
    *  - VK_IMAGE_LAYOUT_STENCIL_ATTACHMENT_OPTIMAL_KHR
    *
    * For general, we have no nice opportunity to transition so we do the copy
    * to the shadow unconditionally at the end of the subpass. For transfer
    * destinations, we can update it as part of the transfer op. For the other
    * layouts, we delay the copy until a transition into some other layout.
    */
   if (image->planes[plane].shadow_surface.isl.size_B > 0 &&
       vk_image_layout_stencil_write_optimal(initial_layout) &&
       !vk_image_layout_stencil_write_optimal(final_layout)) {
      anv_image_copy_to_shadow(cmd_buffer, image,
                               VK_IMAGE_ASPECT_STENCIL_BIT,
                               base_level, level_count,
                               base_layer, layer_count);
   }
#elif GEN_GEN == 12
   uint32_t plane = anv_image_aspect_to_plane(image->aspects,
                                              VK_IMAGE_ASPECT_STENCIL_BIT);
   if (image->planes[plane].aux_usage == ISL_AUX_USAGE_NONE)
      return;

   if ((initial_layout == VK_IMAGE_LAYOUT_UNDEFINED ||
        initial_layout == VK_IMAGE_LAYOUT_PREINITIALIZED) &&
       cmd_buffer->device->physical->has_implicit_ccs &&
       cmd_buffer->device->info.has_aux_map) {
      anv_image_init_aux_tt(cmd_buffer, image, VK_IMAGE_ASPECT_STENCIL_BIT,
                            base_level, level_count, base_layer, layer_count);

      /* If will_full_fast_clear is set, the caller promises to fast-clear the
       * largest portion of the specified range as it can.
       */
      if (will_full_fast_clear)
         return;

      for (uint32_t l = 0; l < level_count; l++) {
         const uint32_t level = base_level + l;
         const VkRect2D clear_rect = {
            .offset.x = 0,
            .offset.y = 0,
            .extent.width = anv_minify(image->extent.width, level),
            .extent.height = anv_minify(image->extent.height, level),
         };

         uint32_t aux_layers =
            anv_image_aux_layers(image, VK_IMAGE_ASPECT_STENCIL_BIT, level);
         uint32_t level_layer_count =
            MIN2(layer_count, aux_layers - base_layer);

         /* From Bspec's 3DSTATE_STENCIL_BUFFER_BODY > Stencil Compression
          * Enable:
          *
          *    "When enabled, Stencil Buffer needs to be initialized via
          *    stencil clear (HZ_OP) before any renderpass."
          */
         anv_image_hiz_clear(cmd_buffer, image, VK_IMAGE_ASPECT_STENCIL_BIT,
                             level, base_layer, level_layer_count,
                             clear_rect, 0 /* Stencil clear value */);
      }
   }
#endif
}

#define MI_PREDICATE_SRC0    0x2400
#define MI_PREDICATE_SRC1    0x2408
#define MI_PREDICATE_RESULT  0x2418

static void
set_image_compressed_bit(struct anv_cmd_buffer *cmd_buffer,
                         const struct anv_image *image,
                         VkImageAspectFlagBits aspect,
                         uint32_t level,
                         uint32_t base_layer, uint32_t layer_count,
                         bool compressed)
{
   uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);

   /* We only have compression tracking for CCS_E */
   if (image->planes[plane].aux_usage != ISL_AUX_USAGE_CCS_E)
      return;

   for (uint32_t a = 0; a < layer_count; a++) {
      uint32_t layer = base_layer + a;
      anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi) {
         sdi.Address = anv_image_get_compression_state_addr(cmd_buffer->device,
                                                            image, aspect,
                                                            level, layer);
         sdi.ImmediateData = compressed ? UINT32_MAX : 0;
      }
   }
}

static void
set_image_fast_clear_state(struct anv_cmd_buffer *cmd_buffer,
                           const struct anv_image *image,
                           VkImageAspectFlagBits aspect,
                           enum anv_fast_clear_type fast_clear)
{
   anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi) {
      sdi.Address = anv_image_get_fast_clear_type_addr(cmd_buffer->device,
                                                       image, aspect);
      sdi.ImmediateData = fast_clear;
   }

   /* Whenever we have fast-clear, we consider that slice to be compressed.
    * This makes building predicates much easier.
    */
   if (fast_clear != ANV_FAST_CLEAR_NONE)
      set_image_compressed_bit(cmd_buffer, image, aspect, 0, 0, 1, true);
}

/* This is only really practical on haswell and above because it requires
 * MI math in order to get it correct.
 */
#if GEN_GEN >= 8 || GEN_IS_HASWELL
static void
anv_cmd_compute_resolve_predicate(struct anv_cmd_buffer *cmd_buffer,
                                  const struct anv_image *image,
                                  VkImageAspectFlagBits aspect,
                                  uint32_t level, uint32_t array_layer,
                                  enum isl_aux_op resolve_op,
                                  enum anv_fast_clear_type fast_clear_supported)
{
   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   const struct gen_mi_value fast_clear_type =
      gen_mi_mem32(anv_image_get_fast_clear_type_addr(cmd_buffer->device,
                                                      image, aspect));

   if (resolve_op == ISL_AUX_OP_FULL_RESOLVE) {
      /* In this case, we're doing a full resolve which means we want the
       * resolve to happen if any compression (including fast-clears) is
       * present.
       *
       * In order to simplify the logic a bit, we make the assumption that,
       * if the first slice has been fast-cleared, it is also marked as
       * compressed.  See also set_image_fast_clear_state.
       */
      const struct gen_mi_value compression_state =
         gen_mi_mem32(anv_image_get_compression_state_addr(cmd_buffer->device,
                                                           image, aspect,
                                                           level, array_layer));
      gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC0),
                       compression_state);
      gen_mi_store(&b, compression_state, gen_mi_imm(0));

      if (level == 0 && array_layer == 0) {
         /* If the predicate is true, we want to write 0 to the fast clear type
          * and, if it's false, leave it alone.  We can do this by writing
          *
          * clear_type = clear_type & ~predicate;
          */
         struct gen_mi_value new_fast_clear_type =
            gen_mi_iand(&b, fast_clear_type,
                            gen_mi_inot(&b, gen_mi_reg64(MI_PREDICATE_SRC0)));
         gen_mi_store(&b, fast_clear_type, new_fast_clear_type);
      }
   } else if (level == 0 && array_layer == 0) {
      /* In this case, we are doing a partial resolve to get rid of fast-clear
       * colors.  We don't care about the compression state but we do care
       * about how much fast clear is allowed by the final layout.
       */
      assert(resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE);
      assert(fast_clear_supported < ANV_FAST_CLEAR_ANY);

      /* We need to compute (fast_clear_supported < image->fast_clear) */
      struct gen_mi_value pred =
         gen_mi_ult(&b, gen_mi_imm(fast_clear_supported), fast_clear_type);
      gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC0),
                       gen_mi_value_ref(&b, pred));

      /* If the predicate is true, we want to write 0 to the fast clear type
       * and, if it's false, leave it alone.  We can do this by writing
       *
       * clear_type = clear_type & ~predicate;
       */
      struct gen_mi_value new_fast_clear_type =
         gen_mi_iand(&b, fast_clear_type, gen_mi_inot(&b, pred));
      gen_mi_store(&b, fast_clear_type, new_fast_clear_type);
   } else {
      /* In this case, we're trying to do a partial resolve on a slice that
       * doesn't have clear color.  There's nothing to do.
       */
      assert(resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE);
      return;
   }

   /* Set src1 to 0 and use a != condition */
   gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC1), gen_mi_imm(0));

   anv_batch_emit(&cmd_buffer->batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOADINV;
      mip.CombineOperation = COMBINE_SET;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }
}
#endif /* GEN_GEN >= 8 || GEN_IS_HASWELL */

#if GEN_GEN <= 8
static void
anv_cmd_simple_resolve_predicate(struct anv_cmd_buffer *cmd_buffer,
                                 const struct anv_image *image,
                                 VkImageAspectFlagBits aspect,
                                 uint32_t level, uint32_t array_layer,
                                 enum isl_aux_op resolve_op,
                                 enum anv_fast_clear_type fast_clear_supported)
{
   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   struct gen_mi_value fast_clear_type_mem =
      gen_mi_mem32(anv_image_get_fast_clear_type_addr(cmd_buffer->device,
                                                      image, aspect));

   /* This only works for partial resolves and only when the clear color is
    * all or nothing.  On the upside, this emits less command streamer code
    * and works on Ivybridge and Bay Trail.
    */
   assert(resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE);
   assert(fast_clear_supported != ANV_FAST_CLEAR_ANY);

   /* We don't support fast clears on anything other than the first slice. */
   if (level > 0 || array_layer > 0)
      return;

   /* On gen8, we don't have a concept of default clear colors because we
    * can't sample from CCS surfaces.  It's enough to just load the fast clear
    * state into the predicate register.
    */
   gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC0), fast_clear_type_mem);
   gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC1), gen_mi_imm(0));
   gen_mi_store(&b, fast_clear_type_mem, gen_mi_imm(0));

   anv_batch_emit(&cmd_buffer->batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOADINV;
      mip.CombineOperation = COMBINE_SET;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }
}
#endif /* GEN_GEN <= 8 */

static void
anv_cmd_predicated_ccs_resolve(struct anv_cmd_buffer *cmd_buffer,
                               const struct anv_image *image,
                               enum isl_format format,
                               struct isl_swizzle swizzle,
                               VkImageAspectFlagBits aspect,
                               uint32_t level, uint32_t array_layer,
                               enum isl_aux_op resolve_op,
                               enum anv_fast_clear_type fast_clear_supported)
{
   const uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);

#if GEN_GEN >= 9
   anv_cmd_compute_resolve_predicate(cmd_buffer, image,
                                     aspect, level, array_layer,
                                     resolve_op, fast_clear_supported);
#else /* GEN_GEN <= 8 */
   anv_cmd_simple_resolve_predicate(cmd_buffer, image,
                                    aspect, level, array_layer,
                                    resolve_op, fast_clear_supported);
#endif

   /* CCS_D only supports full resolves and BLORP will assert on us if we try
    * to do a partial resolve on a CCS_D surface.
    */
   if (resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE &&
       image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_D)
      resolve_op = ISL_AUX_OP_FULL_RESOLVE;

   anv_image_ccs_op(cmd_buffer, image, format, swizzle, aspect,
                    level, array_layer, 1, resolve_op, NULL, true);
}

static void
anv_cmd_predicated_mcs_resolve(struct anv_cmd_buffer *cmd_buffer,
                               const struct anv_image *image,
                               enum isl_format format,
                               struct isl_swizzle swizzle,
                               VkImageAspectFlagBits aspect,
                               uint32_t array_layer,
                               enum isl_aux_op resolve_op,
                               enum anv_fast_clear_type fast_clear_supported)
{
   assert(aspect == VK_IMAGE_ASPECT_COLOR_BIT);
   assert(resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE);

#if GEN_GEN >= 8 || GEN_IS_HASWELL
   anv_cmd_compute_resolve_predicate(cmd_buffer, image,
                                     aspect, 0, array_layer,
                                     resolve_op, fast_clear_supported);

   anv_image_mcs_op(cmd_buffer, image, format, swizzle, aspect,
                    array_layer, 1, resolve_op, NULL, true);
#else
   unreachable("MCS resolves are unsupported on Ivybridge and Bay Trail");
#endif
}

void
genX(cmd_buffer_mark_image_written)(struct anv_cmd_buffer *cmd_buffer,
                                    const struct anv_image *image,
                                    VkImageAspectFlagBits aspect,
                                    enum isl_aux_usage aux_usage,
                                    uint32_t level,
                                    uint32_t base_layer,
                                    uint32_t layer_count)
{
   /* The aspect must be exactly one of the image aspects. */
   assert(util_bitcount(aspect) == 1 && (aspect & image->aspects));

   /* The only compression types with more than just fast-clears are MCS,
    * CCS_E, and HiZ.  With HiZ we just trust the layout and don't actually
    * track the current fast-clear and compression state.  This leaves us
    * with just MCS and CCS_E.
    */
   if (aux_usage != ISL_AUX_USAGE_CCS_E &&
       aux_usage != ISL_AUX_USAGE_MCS)
      return;

   set_image_compressed_bit(cmd_buffer, image, aspect,
                            level, base_layer, layer_count, true);
}

static void
init_fast_clear_color(struct anv_cmd_buffer *cmd_buffer,
                      const struct anv_image *image,
                      VkImageAspectFlagBits aspect)
{
   assert(cmd_buffer && image);
   assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);

   set_image_fast_clear_state(cmd_buffer, image, aspect,
                              ANV_FAST_CLEAR_NONE);

   /* Initialize the struct fields that are accessed for fast-clears so that
    * the HW restrictions on the field values are satisfied.
    */
   struct anv_address addr =
      anv_image_get_clear_color_addr(cmd_buffer->device, image, aspect);

   if (GEN_GEN >= 9) {
      const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;
      const unsigned num_dwords = GEN_GEN >= 10 ?
                                  isl_dev->ss.clear_color_state_size / 4 :
                                  isl_dev->ss.clear_value_size / 4;
      for (unsigned i = 0; i < num_dwords; i++) {
         anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi) {
            sdi.Address = addr;
            sdi.Address.offset += i * 4;
            sdi.ImmediateData = 0;
         }
      }
   } else {
      anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi) {
         sdi.Address = addr;
         if (GEN_GEN >= 8 || GEN_IS_HASWELL) {
            /* 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.
             */
            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_GEN == 7) {
            /* On IVB, the dword containing the clear values also contains
             * other fields that must be zero or can be zero.
             */
            sdi.ImmediateData = 0;
         }
      }
   }
}

/* Copy the fast-clear value dword(s) between a surface state object and an
 * image's fast clear state buffer.
 */
static void
genX(copy_fast_clear_dwords)(struct anv_cmd_buffer *cmd_buffer,
                             struct anv_state surface_state,
                             const struct anv_image *image,
                             VkImageAspectFlagBits aspect,
                             bool copy_from_surface_state)
{
   assert(cmd_buffer && image);
   assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);

   struct anv_address ss_clear_addr = {
      .bo = cmd_buffer->device->surface_state_pool.block_pool.bo,
      .offset = surface_state.offset +
                cmd_buffer->device->isl_dev.ss.clear_value_offset,
   };
   const struct anv_address entry_addr =
      anv_image_get_clear_color_addr(cmd_buffer->device, image, aspect);
   unsigned copy_size = cmd_buffer->device->isl_dev.ss.clear_value_size;

#if GEN_GEN == 7
   /* On gen7, the combination of commands used here(MI_LOAD_REGISTER_MEM
    * and MI_STORE_REGISTER_MEM) can cause GPU hangs if any rendering is
    * in-flight when they are issued even if the memory touched is not
    * currently active for rendering.  The weird bit is that it is not the
    * MI_LOAD/STORE_REGISTER_MEM commands which hang but rather the in-flight
    * rendering hangs such that the next stalling command after the
    * MI_LOAD/STORE_REGISTER_MEM commands will catch the hang.
    *
    * It is unclear exactly why this hang occurs.  Both MI commands come with
    * warnings about the 3D pipeline but that doesn't seem to fully explain
    * it.  My (Jason's) best theory is that it has something to do with the
    * fact that we're using a GPU state register as our temporary and that
    * something with reading/writing it is causing problems.
    *
    * In order to work around this issue, we emit a PIPE_CONTROL with the
    * command streamer stall bit set.
    */
   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_CS_STALL_BIT;
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
#endif

   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   if (copy_from_surface_state) {
      gen_mi_memcpy(&b, entry_addr, ss_clear_addr, copy_size);
   } else {
      gen_mi_memcpy(&b, ss_clear_addr, entry_addr, copy_size);

      /* Updating a surface state object may require that the state cache be
       * invalidated. From the SKL PRM, Shared Functions -> State -> State
       * Caching:
       *
       *    Whenever the RENDER_SURFACE_STATE object in memory pointed to by
       *    the Binding Table Pointer (BTP) and Binding Table Index (BTI) is
       *    modified [...], the L1 state cache must be invalidated to ensure
       *    the new surface or sampler state is fetched from system memory.
       *
       * In testing, SKL doesn't actually seem to need this, but HSW does.
       */
      cmd_buffer->state.pending_pipe_bits |=
         ANV_PIPE_STATE_CACHE_INVALIDATE_BIT;
   }
}

/**
 * @brief Transitions a color buffer from one layout to another.
 *
 * See section 6.1.1. Image Layout Transitions of the Vulkan 1.0.50 spec for
 * more information.
 *
 * @param level_count VK_REMAINING_MIP_LEVELS isn't supported.
 * @param layer_count VK_REMAINING_ARRAY_LAYERS isn't supported. For 3D images,
 *                    this represents the maximum layers to transition at each
 *                    specified miplevel.
 */
static void
transition_color_buffer(struct anv_cmd_buffer *cmd_buffer,
                        const struct anv_image *image,
                        VkImageAspectFlagBits aspect,
                        const uint32_t base_level, uint32_t level_count,
                        uint32_t base_layer, uint32_t layer_count,
                        VkImageLayout initial_layout,
                        VkImageLayout final_layout,
                        bool will_full_fast_clear)
{
   struct anv_device *device = cmd_buffer->device;
   const struct gen_device_info *devinfo = &device->info;
   /* Validate the inputs. */
   assert(cmd_buffer);
   assert(image && image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
   /* These values aren't supported for simplicity's sake. */
   assert(level_count != VK_REMAINING_MIP_LEVELS &&
          layer_count != VK_REMAINING_ARRAY_LAYERS);
   /* Ensure the subresource range is valid. */
   UNUSED uint64_t last_level_num = base_level + level_count;
   const uint32_t max_depth = anv_minify(image->extent.depth, base_level);
   UNUSED const uint32_t image_layers = MAX2(image->array_size, max_depth);
   assert((uint64_t)base_layer + layer_count  <= image_layers);
   assert(last_level_num <= image->levels);
   /* The spec disallows these final layouts. */
   assert(final_layout != VK_IMAGE_LAYOUT_UNDEFINED &&
          final_layout != VK_IMAGE_LAYOUT_PREINITIALIZED);

   /* No work is necessary if the layout stays the same or if this subresource
    * range lacks auxiliary data.
    */
   if (initial_layout == final_layout)
      return;

   uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);

   if (image->planes[plane].shadow_surface.isl.size_B > 0 &&
       final_layout == VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL) {
      /* This surface is a linear compressed image with a tiled shadow surface
       * for texturing.  The client is about to use it in READ_ONLY_OPTIMAL so
       * we need to ensure the shadow copy is up-to-date.
       */
      assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
      assert(image->planes[plane].surface.isl.tiling == ISL_TILING_LINEAR);
      assert(image->planes[plane].shadow_surface.isl.tiling != ISL_TILING_LINEAR);
      assert(isl_format_is_compressed(image->planes[plane].surface.isl.format));
      assert(plane == 0);
      anv_image_copy_to_shadow(cmd_buffer, image,
                               VK_IMAGE_ASPECT_COLOR_BIT,
                               base_level, level_count,
                               base_layer, layer_count);
   }

   if (base_layer >= anv_image_aux_layers(image, aspect, base_level))
      return;

   assert(image->planes[plane].surface.isl.tiling != ISL_TILING_LINEAR);

   if (initial_layout == VK_IMAGE_LAYOUT_UNDEFINED ||
       initial_layout == VK_IMAGE_LAYOUT_PREINITIALIZED) {
#if GEN_GEN == 12
      if (device->physical->has_implicit_ccs && devinfo->has_aux_map) {
         anv_image_init_aux_tt(cmd_buffer, image, aspect,
                               base_level, level_count,
                               base_layer, layer_count);
      }
#else
      assert(!(device->physical->has_implicit_ccs && devinfo->has_aux_map));
#endif

      /* A subresource in the undefined layout may have been aliased and
       * populated with any arrangement of bits. Therefore, we must initialize
       * the related aux buffer and clear buffer entry with desirable values.
       * An initial layout of PREINITIALIZED is the same as UNDEFINED for
       * images with VK_IMAGE_TILING_OPTIMAL.
       *
       * Initialize the relevant clear buffer entries.
       */
      if (base_level == 0 && base_layer == 0)
         init_fast_clear_color(cmd_buffer, image, aspect);

      /* Initialize the aux buffers to enable correct rendering.  In order to
       * ensure that things such as storage images work correctly, aux buffers
       * need to be initialized to valid data.
       *
       * Having an aux buffer with invalid data is a problem for two reasons:
       *
       *  1) Having an invalid value in the buffer can confuse the hardware.
       *     For instance, with CCS_E on SKL, a two-bit CCS value of 2 is
       *     invalid and leads to the hardware doing strange things.  It
       *     doesn't hang as far as we can tell but rendering corruption can
       *     occur.
       *
       *  2) If this transition is into the GENERAL layout and we then use the
       *     image as a storage image, then we must have the aux buffer in the
       *     pass-through state so that, if we then go to texture from the
       *     image, we get the results of our storage image writes and not the
       *     fast clear color or other random data.
       *
       * For CCS both of the problems above are real demonstrable issues.  In
       * that case, the only thing we can do is to perform an ambiguate to
       * transition the aux surface into the pass-through state.
       *
       * For MCS, (2) is never an issue because we don't support multisampled
       * storage images.  In theory, issue (1) is a problem with MCS but we've
       * never seen it in the wild.  For 4x and 16x, all bit patters could, in
       * theory, be interpreted as something but we don't know that all bit
       * patterns are actually valid.  For 2x and 8x, you could easily end up
       * with the MCS referring to an invalid plane because not all bits of
       * the MCS value are actually used.  Even though we've never seen issues
       * in the wild, it's best to play it safe and initialize the MCS.  We
       * can use a fast-clear for MCS because we only ever touch from render
       * and texture (no image load store).
       */
      if (image->samples == 1) {
         for (uint32_t l = 0; l < level_count; l++) {
            const uint32_t level = base_level + l;

            uint32_t aux_layers = anv_image_aux_layers(image, aspect, level);
            if (base_layer >= aux_layers)
               break; /* We will only get fewer layers as level increases */
            uint32_t level_layer_count =
               MIN2(layer_count, aux_layers - base_layer);

            /* If will_full_fast_clear is set, the caller promises to
             * fast-clear the largest portion of the specified range as it can.
             * For color images, that means only the first LOD and array slice.
             */
            if (level == 0 && base_layer == 0 && will_full_fast_clear) {
               base_layer++;
               level_layer_count--;
               if (level_layer_count == 0)
                  continue;
            }

            anv_image_ccs_op(cmd_buffer, image,
                             image->planes[plane].surface.isl.format,
                             ISL_SWIZZLE_IDENTITY,
                             aspect, level, base_layer, level_layer_count,
                             ISL_AUX_OP_AMBIGUATE, NULL, false);

            if (image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_E) {
               set_image_compressed_bit(cmd_buffer, image, aspect,
                                        level, base_layer, level_layer_count,
                                        false);
            }
         }
      } else {
         if (image->samples == 4 || image->samples == 16) {
            anv_perf_warn(cmd_buffer->device, image,
                          "Doing a potentially unnecessary fast-clear to "
                          "define an MCS buffer.");
         }

         /* If will_full_fast_clear is set, the caller promises to fast-clear
          * the largest portion of the specified range as it can.
          */
         if (will_full_fast_clear)
            return;

         assert(base_level == 0 && level_count == 1);
         anv_image_mcs_op(cmd_buffer, image,
                          image->planes[plane].surface.isl.format,
                          ISL_SWIZZLE_IDENTITY,
                          aspect, base_layer, layer_count,
                          ISL_AUX_OP_FAST_CLEAR, NULL, false);
      }
      return;
   }

   const enum isl_aux_usage initial_aux_usage =
      anv_layout_to_aux_usage(devinfo, image, aspect, 0, initial_layout);
   const enum isl_aux_usage final_aux_usage =
      anv_layout_to_aux_usage(devinfo, image, aspect, 0, final_layout);

   /* The current code assumes that there is no mixing of CCS_E and CCS_D.
    * We can handle transitions between CCS_D/E to and from NONE.  What we
    * don't yet handle is switching between CCS_E and CCS_D within a given
    * image.  Doing so in a performant way requires more detailed aux state
    * tracking such as what is done in i965.  For now, just assume that we
    * only have one type of compression.
    */
   assert(initial_aux_usage == ISL_AUX_USAGE_NONE ||
          final_aux_usage == ISL_AUX_USAGE_NONE ||
          initial_aux_usage == final_aux_usage);

   /* If initial aux usage is NONE, there is nothing to resolve */
   if (initial_aux_usage == ISL_AUX_USAGE_NONE)
      return;

   enum isl_aux_op resolve_op = ISL_AUX_OP_NONE;

   /* If the initial layout supports more fast clear than the final layout
    * then we need at least a partial resolve.
    */
   const enum anv_fast_clear_type initial_fast_clear =
      anv_layout_to_fast_clear_type(devinfo, image, aspect, initial_layout);
   const enum anv_fast_clear_type final_fast_clear =
      anv_layout_to_fast_clear_type(devinfo, image, aspect, final_layout);
   if (final_fast_clear < initial_fast_clear)
      resolve_op = ISL_AUX_OP_PARTIAL_RESOLVE;

   if (initial_aux_usage == ISL_AUX_USAGE_CCS_E &&
       final_aux_usage != ISL_AUX_USAGE_CCS_E)
      resolve_op = ISL_AUX_OP_FULL_RESOLVE;

   if (resolve_op == ISL_AUX_OP_NONE)
      return;

   /* Perform a resolve to synchronize data between the main and aux buffer.
    * Before we begin, we must satisfy the cache flushing requirement specified
    * in the Sky Lake PRM Vol. 7, "MCS Buffer for Render Target(s)":
    *
    *    Any transition from any value in {Clear, Render, Resolve} to a
    *    different value in {Clear, Render, Resolve} requires end of pipe
    *    synchronization.
    *
    * We perform a flush of the write cache before and after the clear and
    * resolve operations to meet this requirement.
    *
    * Unlike other drawing, fast clear operations are not properly
    * synchronized. The first PIPE_CONTROL here likely ensures that the
    * contents of the previous render or clear hit the render target before we
    * resolve and the second likely ensures that the resolve is complete before
    * we do any more rendering or clearing.
    */
   cmd_buffer->state.pending_pipe_bits |=
      ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT | ANV_PIPE_END_OF_PIPE_SYNC_BIT;

   for (uint32_t l = 0; l < level_count; l++) {
      uint32_t level = base_level + l;

      uint32_t aux_layers = anv_image_aux_layers(image, aspect, level);
      if (base_layer >= aux_layers)
         break; /* We will only get fewer layers as level increases */
      uint32_t level_layer_count =
         MIN2(layer_count, aux_layers - base_layer);

      for (uint32_t a = 0; a < level_layer_count; a++) {
         uint32_t array_layer = base_layer + a;

         /* If will_full_fast_clear is set, the caller promises to fast-clear
          * the largest portion of the specified range as it can.  For color
          * images, that means only the first LOD and array slice.
          */
         if (level == 0 && array_layer == 0 && will_full_fast_clear)
            continue;

         if (image->samples == 1) {
            anv_cmd_predicated_ccs_resolve(cmd_buffer, image,
                                           image->planes[plane].surface.isl.format,
                                           ISL_SWIZZLE_IDENTITY,
                                           aspect, level, array_layer, resolve_op,
                                           final_fast_clear);
         } else {
            /* We only support fast-clear on the first layer so partial
             * resolves should not be used on other layers as they will use
             * the clear color stored in memory that is only valid for layer0.
             */
            if (resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE &&
                array_layer != 0)
               continue;

            anv_cmd_predicated_mcs_resolve(cmd_buffer, image,
                                           image->planes[plane].surface.isl.format,
                                           ISL_SWIZZLE_IDENTITY,
                                           aspect, array_layer, resolve_op,
                                           final_fast_clear);
         }
      }
   }

   cmd_buffer->state.pending_pipe_bits |=
      ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT | ANV_PIPE_END_OF_PIPE_SYNC_BIT;
}

static VkResult
genX(cmd_buffer_setup_attachments)(struct anv_cmd_buffer *cmd_buffer,
                                   const struct anv_render_pass *pass,
                                   const struct anv_framebuffer *framebuffer,
                                   const VkRenderPassBeginInfo *begin)
{
   struct anv_cmd_state *state = &cmd_buffer->state;

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

   if (pass->attachment_count > 0) {
      state->attachments = vk_zalloc(&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;
   }

   const VkRenderPassAttachmentBeginInfoKHR *attach_begin =
      vk_find_struct_const(begin, RENDER_PASS_ATTACHMENT_BEGIN_INFO_KHR);
   if (begin && !attach_begin)
      assert(pass->attachment_count == framebuffer->attachment_count);

   for (uint32_t i = 0; i < pass->attachment_count; ++i) {
      if (attach_begin && attach_begin->attachmentCount != 0) {
         assert(attach_begin->attachmentCount == pass->attachment_count);
         ANV_FROM_HANDLE(anv_image_view, iview, attach_begin->pAttachments[i]);
         state->attachments[i].image_view = iview;
      } else if (framebuffer && i < framebuffer->attachment_count) {
         state->attachments[i].image_view = framebuffer->attachments[i];
      } else {
         state->attachments[i].image_view = NULL;
      }
   }

   if (begin) {
      for (uint32_t i = 0; i < pass->attachment_count; ++i) {
         const struct anv_render_pass_attachment *pass_att = &pass->attachments[i];
         struct anv_attachment_state *att_state = &state->attachments[i];
         VkImageAspectFlags att_aspects = vk_format_aspects(pass_att->format);
         VkImageAspectFlags clear_aspects = 0;
         VkImageAspectFlags load_aspects = 0;

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

         att_state->current_layout = pass_att->initial_layout;
         att_state->current_stencil_layout = pass_att->stencil_initial_layout;
         att_state->pending_clear_aspects = clear_aspects;
         att_state->pending_load_aspects = load_aspects;
         if (clear_aspects)
            att_state->clear_value = begin->pClearValues[i];

         struct anv_image_view *iview = state->attachments[i].image_view;
         anv_assert(iview->vk_format == pass_att->format);

         const uint32_t num_layers = iview->planes[0].isl.array_len;
         att_state->pending_clear_views = (1 << num_layers) - 1;

         /* This will be initialized after the first subpass transition. */
         att_state->aux_usage = ISL_AUX_USAGE_NONE;

         att_state->fast_clear = false;
         if (clear_aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) {
            assert(clear_aspects == VK_IMAGE_ASPECT_COLOR_BIT);
            att_state->fast_clear =
               anv_can_fast_clear_color_view(cmd_buffer->device, iview,
                                             pass_att->first_subpass_layout,
                                             vk_to_isl_color(att_state->clear_value.color),
                                             framebuffer->layers,
                                             begin->renderArea);
         } else if (clear_aspects & (VK_IMAGE_ASPECT_DEPTH_BIT |
                                     VK_IMAGE_ASPECT_STENCIL_BIT)) {
            att_state->fast_clear =
               anv_can_hiz_clear_ds_view(cmd_buffer->device, iview,
                                         pass_att->first_subpass_layout,
                                         clear_aspects,
                                         att_state->clear_value.depthStencil.depth,
                                         begin->renderArea);
         }
      }
   }

   return VK_SUCCESS;
}

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

   /* Reserve one for the NULL state. */
   unsigned num_states = 1;
   for (uint32_t i = 0; i < subpass->attachment_count; i++) {
      uint32_t att = subpass->attachments[i].attachment;
      if (att == VK_ATTACHMENT_UNUSED)
         continue;

      assert(att < pass->attachment_count);
      if (!vk_format_is_color(pass->attachments[att].format))
         continue;

      const VkImageUsageFlagBits att_usage = subpass->attachments[i].usage;
      assert(util_bitcount(att_usage) == 1);

      if (att_usage == VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT ||
          att_usage == VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT)
         num_states++;
   }

   const uint32_t ss_stride = align_u32(isl_dev->ss.size, isl_dev->ss.align);
   state->attachment_states =
      anv_state_stream_alloc(&cmd_buffer->surface_state_stream,
                             num_states * ss_stride, isl_dev->ss.align);
   if (state->attachment_states.map == NULL) {
      return anv_batch_set_error(&cmd_buffer->batch,
                                 VK_ERROR_OUT_OF_DEVICE_MEMORY);
   }

   struct anv_state next_state = state->attachment_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 < subpass->attachment_count; i++) {
      uint32_t att = subpass->attachments[i].attachment;
      if (att == VK_ATTACHMENT_UNUSED)
         continue;

      assert(att < pass->attachment_count);
      if (!vk_format_is_color(pass->attachments[att].format))
         continue;

      const VkImageUsageFlagBits att_usage = subpass->attachments[i].usage;
      assert(util_bitcount(att_usage) == 1);

      if (att_usage == VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT)
         state->attachments[att].color.state = next_state;
      else if (att_usage == VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT)
         state->attachments[att].input.state = next_state;
      else
         continue;

      state->attachments[att].color.state = next_state;
      next_state.offset += ss_stride;
      next_state.map += ss_stride;
   }

   assert(next_state.offset == state->attachment_states.offset +
                               state->attachment_states.alloc_size);

   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;

   /* VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT must be ignored for
    * primary level command buffers.
    *
    * From the Vulkan 1.0 spec:
    *
    *    VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT specifies that a
    *    secondary command buffer is considered to be entirely inside a render
    *    pass. If this is a primary command buffer, then this bit is ignored.
    */
   if (cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY)
      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.
    *
    * There is also a workaround on gen8 which requires us to invalidate the
    * VF cache occasionally.  It's easier if we can assume we start with a
    * fresh cache (See also genX(cmd_buffer_set_binding_for_gen8_vb_flush).)
    */
   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_VF_CACHE_INVALIDATE_BIT;

   /* Re-emit the aux table register in every command buffer.  This way we're
    * ensured that we have the table even if this command buffer doesn't
    * initialize any images.
    */
   if (cmd_buffer->device->info.has_aux_map)
      cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_AUX_TABLE_INVALIDATE_BIT;

   /* We send an "Indirect State Pointers Disable" packet at
    * EndCommandBuffer, so all push contant packets are ignored during a
    * context restore. Documentation says after that command, we need to
    * emit push constants again before any rendering operation. So we
    * flag them dirty here to make sure they get emitted.
    */
   cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_ALL_GRAPHICS;

   VkResult result = VK_SUCCESS;
   if (cmd_buffer->usage_flags &
       VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT) {
      assert(pBeginInfo->pInheritanceInfo);
      ANV_FROM_HANDLE(anv_render_pass, pass,
                      pBeginInfo->pInheritanceInfo->renderPass);
      struct anv_subpass *subpass =
         &pass->subpasses[pBeginInfo->pInheritanceInfo->subpass];
      ANV_FROM_HANDLE(anv_framebuffer, framebuffer,
                      pBeginInfo->pInheritanceInfo->framebuffer);

      cmd_buffer->state.pass = pass;
      cmd_buffer->state.subpass = subpass;

      /* This is optional in the inheritance info. */
      cmd_buffer->state.framebuffer = framebuffer;

      result = genX(cmd_buffer_setup_attachments)(cmd_buffer, pass,
                                                  framebuffer, NULL);
      if (result != VK_SUCCESS)
         return result;

      result = genX(cmd_buffer_alloc_att_surf_states)(cmd_buffer, pass,
                                                      subpass);
      if (result != VK_SUCCESS)
         return result;

      /* Record that HiZ is enabled if we can. */
      if (cmd_buffer->state.framebuffer) {
         const struct anv_image_view * const iview =
            anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);

         if (iview) {
            VkImageLayout layout =
                cmd_buffer->state.subpass->depth_stencil_attachment->layout;

            enum isl_aux_usage aux_usage =
               anv_layout_to_aux_usage(&cmd_buffer->device->info, iview->image,
                                       VK_IMAGE_ASPECT_DEPTH_BIT,
                                       VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT,
                                       layout);

            cmd_buffer->state.hiz_enabled = isl_aux_usage_has_hiz(aux_usage);
         }
      }

      cmd_buffer->state.gfx.dirty |= ANV_CMD_DIRTY_RENDER_TARGETS;
   }

#if GEN_GEN >= 8 || GEN_IS_HASWELL
   if (cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_SECONDARY) {
      const VkCommandBufferInheritanceConditionalRenderingInfoEXT *conditional_rendering_info =
         vk_find_struct_const(pBeginInfo->pInheritanceInfo->pNext, COMMAND_BUFFER_INHERITANCE_CONDITIONAL_RENDERING_INFO_EXT);

      /* If secondary buffer supports conditional rendering
       * we should emit commands as if conditional rendering is enabled.
       */
      cmd_buffer->state.conditional_render_enabled =
         conditional_rendering_info && conditional_rendering_info->conditionalRenderingEnable;
   }
#endif

   return result;
}

/* From the PRM, Volume 2a:
 *
 *    "Indirect State Pointers Disable
 *
 *    At the completion of the post-sync operation associated with this pipe
 *    control packet, the indirect state pointers in the hardware are
 *    considered invalid; the indirect pointers are not saved in the context.
 *    If any new indirect state commands are executed in the command stream
 *    while the pipe control is pending, the new indirect state commands are
 *    preserved.
 *
 *    [DevIVB+]: Using Invalidate State Pointer (ISP) only inhibits context
 *    restoring of Push Constant (3DSTATE_CONSTANT_*) commands. Push Constant
 *    commands are only considered as Indirect State Pointers. Once ISP is
 *    issued in a context, SW must initialize by programming push constant
 *    commands for all the shaders (at least to zero length) before attempting
 *    any rendering operation for the same context."
 *
 * 3DSTATE_CONSTANT_* packets are restored during a context restore,
 * even though they point to a BO that has been already unreferenced at
 * the end of the previous batch buffer. This has been fine so far since
 * we are protected by these scratch page (every address not covered by
 * a BO should be pointing to the scratch page). But on CNL, it is
 * causing a GPU hang during context restore at the 3DSTATE_CONSTANT_*
 * instruction.
 *
 * The flag "Indirect State Pointers Disable" in PIPE_CONTROL tells the
 * hardware to ignore previous 3DSTATE_CONSTANT_* packets during a
 * context restore, so the mentioned hang doesn't happen. However,
 * software must program push constant commands for all stages prior to
 * rendering anything. So we flag them dirty in BeginCommandBuffer.
 *
 * Finally, we also make sure to stall at pixel scoreboard to make sure the
 * constants have been loaded into the EUs prior to disable the push constants
 * so that it doesn't hang a previous 3DPRIMITIVE.
 */
static void
emit_isp_disable(struct anv_cmd_buffer *cmd_buffer)
{
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
         pc.StallAtPixelScoreboard = true;
         pc.CommandStreamerStallEnable = true;
   }
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
         pc.IndirectStatePointersDisable = true;
         pc.CommandStreamerStallEnable = true;
   }
}

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);

   emit_isp_disable(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 GEN_GEN >= 8 || GEN_IS_HASWELL
      if (secondary->state.conditional_render_enabled) {
         if (!primary->state.conditional_render_enabled) {
            /* Secondary buffer is constructed as if it will be executed
             * with conditional rendering, we should satisfy this dependency
             * regardless of conditional rendering being enabled in primary.
             */
            struct gen_mi_builder b;
            gen_mi_builder_init(&b, &primary->batch);
            gen_mi_store(&b, gen_mi_reg64(ANV_PREDICATE_RESULT_REG),
                             gen_mi_imm(UINT64_MAX));
         }
      }
#endif

      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.attachment_states;
         struct anv_state dst_state = secondary->state.attachment_states;
         assert(src_state.alloc_size == dst_state.alloc_size);

         genX(cmd_buffer_so_memcpy)(primary,
                                    (struct anv_address) {
                                       .bo = ss_bo,
                                       .offset = dst_state.offset,
                                    },
                                    (struct anv_address) {
                                       .bo = ss_bo,
                                       .offset = src_state.offset,
                                    },
                                    src_state.alloc_size);
      }

      anv_cmd_buffer_add_secondary(primary, secondary);

      assert(secondary->perf_query_pool == NULL || primary->perf_query_pool == NULL ||
             secondary->perf_query_pool == primary->perf_query_pool);
      if (secondary->perf_query_pool)
         primary->perf_query_pool = secondary->perf_query_pool;
   }

   /* The secondary isn't counted in our VF cache tracking so we need to
    * invalidate the whole thing.
    */
   if (GEN_GEN >= 8 && GEN_GEN <= 9) {
      primary->state.pending_pipe_bits |=
         ANV_PIPE_CS_STALL_BIT | ANV_PIPE_VF_CACHE_INVALIDATE_BIT;
   }

   /* The secondary may have selected a different pipeline (3D or compute) and
    * may have changed the current L3$ configuration.  Reset our tracking
    * variables to invalid values to ensure that we re-emit these in the case
    * where we do any draws or compute dispatches from the primary after the
    * secondary has returned.
    */
   primary->state.current_pipeline = UINT32_MAX;
   primary->state.current_l3_config = NULL;
   primary->state.current_hash_scale = 0;

   /* 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 || GEN_GEN >= 12);
   if (cfg == cmd_buffer->state.current_l3_config)
      return;

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

   UNUSED 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]);

#if GEN_GEN >= 12
#define L3_ALLOCATION_REG GENX(L3ALLOC)
#define L3_ALLOCATION_REG_num GENX(L3ALLOC_num)
#else
#define L3_ALLOCATION_REG GENX(L3CNTLREG)
#define L3_ALLOCATION_REG_num GENX(L3CNTLREG_num)
#endif

   uint32_t l3cr;
   anv_pack_struct(&l3cr, L3_ALLOCATION_REG,
#if GEN_GEN < 11
                   .SLMEnable = has_slm,
#endif
#if GEN_GEN == 11
   /* WA_1406697149: Bit 9 "Error Detection Behavior Control" must be set
    * in L3CNTLREG register. The default setting of the bit is not the
    * desirable behavior.
   */
                   .ErrorDetectionBehaviorControl = true,
                   .UseFullWays = true,
#endif
                   .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, L3_ALLOCATION_REG_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. */
   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->physical->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)
{
   UNUSED const struct gen_device_info *devinfo = &cmd_buffer->device->info;
   enum anv_pipe_bits bits = cmd_buffer->state.pending_pipe_bits;

   if (cmd_buffer->device->physical->always_flush_cache)
      bits |= ANV_PIPE_FLUSH_BITS | ANV_PIPE_INVALIDATE_BITS;

   /*
    * From Sandybridge PRM, volume 2, "1.7.2 End-of-Pipe Synchronization":
    *
    *    Write synchronization is a special case of end-of-pipe
    *    synchronization that requires that the render cache and/or depth
    *    related caches are flushed to memory, where the data will become
    *    globally visible. This type of synchronization is required prior to
    *    SW (CPU) actually reading the result data from memory, or initiating
    *    an operation that will use as a read surface (such as a texture
    *    surface) a previous render target and/or depth/stencil buffer
    *
    *
    * From Haswell PRM, volume 2, part 1, "End-of-Pipe Synchronization":
    *
    *    Exercising the write cache flush bits (Render Target Cache Flush
    *    Enable, Depth Cache Flush Enable, DC Flush) in PIPE_CONTROL only
    *    ensures the write caches are flushed and doesn't guarantee the data
    *    is globally visible.
    *
    *    SW can track the completion of the end-of-pipe-synchronization by
    *    using "Notify Enable" and "PostSync Operation - Write Immediate
    *    Data" in the PIPE_CONTROL command.
    *
    * In other words, flushes are pipelined while invalidations are handled
    * immediately.  Therefore, if we're flushing anything then we need to
    * schedule an end-of-pipe sync before any invalidations can happen.
    */
   if (bits & ANV_PIPE_FLUSH_BITS)
      bits |= ANV_PIPE_NEEDS_END_OF_PIPE_SYNC_BIT;


   /* HSD 1209978178: docs say that before programming the aux table:
    *
    *    "Driver must ensure that the engine is IDLE but ensure it doesn't
    *    add extra flushes in the case it knows that the engine is already
    *    IDLE."
    */
   if (GEN_GEN == 12 && (bits & ANV_PIPE_AUX_TABLE_INVALIDATE_BIT))
      bits |= ANV_PIPE_NEEDS_END_OF_PIPE_SYNC_BIT;

   /* If we're going to do an invalidate and we have a pending end-of-pipe
    * sync that has yet to be resolved, we do the end-of-pipe sync now.
    */
   if ((bits & ANV_PIPE_INVALIDATE_BITS) &&
       (bits & ANV_PIPE_NEEDS_END_OF_PIPE_SYNC_BIT)) {
      bits |= ANV_PIPE_END_OF_PIPE_SYNC_BIT;
      bits &= ~ANV_PIPE_NEEDS_END_OF_PIPE_SYNC_BIT;
   }

   if (GEN_GEN >= 12 &&
       ((bits & ANV_PIPE_DEPTH_CACHE_FLUSH_BIT) ||
        (bits & ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT))) {
      /* From the PIPE_CONTROL instruction table, bit 28 (Tile Cache Flush
       * Enable):
       *
       *    Unified Cache (Tile Cache Disabled):
       *
       *    When the Color and Depth (Z) streams are enabled to be cached in
       *    the DC space of L2, Software must use "Render Target Cache Flush
       *    Enable" and "Depth Cache Flush Enable" along with "Tile Cache
       *    Flush" for getting the color and depth (Z) write data to be
       *    globally observable.  In this mode of operation it is not required
       *    to set "CS Stall" upon setting "Tile Cache Flush" bit.
       */
      bits |= ANV_PIPE_TILE_CACHE_FLUSH_BIT;
   }

   /* GEN:BUG:1409226450, Wait for EU to be idle before pipe control which
    * invalidates the instruction cache
    */
   if (GEN_GEN == 12 && (bits & ANV_PIPE_INSTRUCTION_CACHE_INVALIDATE_BIT))
      bits |= ANV_PIPE_CS_STALL_BIT | ANV_PIPE_STALL_AT_SCOREBOARD_BIT;

   if ((GEN_GEN >= 8 && GEN_GEN <= 9) &&
       (bits & ANV_PIPE_CS_STALL_BIT) &&
       (bits & ANV_PIPE_VF_CACHE_INVALIDATE_BIT)) {
      /* If we are doing a VF cache invalidate AND a CS stall (it must be
       * both) then we can reset our vertex cache tracking.
       */
      memset(cmd_buffer->state.gfx.vb_dirty_ranges, 0,
             sizeof(cmd_buffer->state.gfx.vb_dirty_ranges));
      memset(&cmd_buffer->state.gfx.ib_dirty_range, 0,
             sizeof(cmd_buffer->state.gfx.ib_dirty_range));
   }

   /* Project: SKL / Argument: LRI Post Sync Operation [23]
    *
    * "PIPECONTROL command with “Command Streamer Stall Enable” must be
    *  programmed prior to programming a PIPECONTROL command with "LRI
    *  Post Sync Operation" in GPGPU mode of operation (i.e when
    *  PIPELINE_SELECT command is set to GPGPU mode of operation)."
    *
    * The same text exists a few rows below for Post Sync Op.
    *
    * On Gen12 this is GEN:BUG:1607156449.
    */
   if (bits & ANV_PIPE_POST_SYNC_BIT) {
      if ((GEN_GEN == 9 || (GEN_GEN == 12 && devinfo->revision == 0 /* A0 */)) &&
          cmd_buffer->state.current_pipeline == GPGPU)
         bits |= ANV_PIPE_CS_STALL_BIT;
      bits &= ~ANV_PIPE_POST_SYNC_BIT;
   }

   if (bits & (ANV_PIPE_FLUSH_BITS | ANV_PIPE_CS_STALL_BIT |
               ANV_PIPE_END_OF_PIPE_SYNC_BIT)) {
      anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe) {
#if GEN_GEN >= 12
         pipe.TileCacheFlushEnable = bits & ANV_PIPE_TILE_CACHE_FLUSH_BIT;
#endif
         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;

         /* GEN:BUG:1409600907: "PIPE_CONTROL with Depth Stall Enable bit must
          * be set with any PIPE_CONTROL with Depth Flush Enable bit set.
          */
#if GEN_GEN >= 12
         pipe.DepthStallEnable =
            pipe.DepthCacheFlushEnable || (bits & ANV_PIPE_DEPTH_STALL_BIT);
#else
         pipe.DepthStallEnable = bits & ANV_PIPE_DEPTH_STALL_BIT;
#endif

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

         /* From Sandybridge PRM, volume 2, "1.7.3.1 Writing a Value to Memory":
          *
          *    "The most common action to perform upon reaching a
          *    synchronization point is to write a value out to memory. An
          *    immediate value (included with the synchronization command) may
          *    be written."
          *
          *
          * From Broadwell PRM, volume 7, "End-of-Pipe Synchronization":
          *
          *    "In case the data flushed out by the render engine is to be
          *    read back in to the render engine in coherent manner, then the
          *    render engine has to wait for the fence completion before
          *    accessing the flushed data. This can be achieved by following
          *    means on various products: PIPE_CONTROL command with CS Stall
          *    and the required write caches flushed with Post-Sync-Operation
          *    as Write Immediate Data.
          *
          *    Example:
          *       - Workload-1 (3D/GPGPU/MEDIA)
          *       - PIPE_CONTROL (CS Stall, Post-Sync-Operation Write
          *         Immediate Data, Required Write Cache Flush bits set)
          *       - Workload-2 (Can use the data produce or output by
          *         Workload-1)
          */
         if (bits & ANV_PIPE_END_OF_PIPE_SYNC_BIT) {
            pipe.CommandStreamerStallEnable = true;
            pipe.PostSyncOperation = WriteImmediateData;
            pipe.Address = cmd_buffer->device->workaround_address;
         }

         /*
          * 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 (pipe.CommandStreamerStallEnable &&
             !pipe.RenderTargetCacheFlushEnable &&
             !pipe.DepthCacheFlushEnable &&
             !pipe.StallAtPixelScoreboard &&
             !pipe.PostSyncOperation &&
             !pipe.DepthStallEnable &&
             !pipe.DCFlushEnable)
            pipe.StallAtPixelScoreboard = true;
      }

      /* If a render target flush was emitted, then we can toggle off the bit
       * saying that render target writes are ongoing.
       */
      if (bits & ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT)
         bits &= ~(ANV_PIPE_RENDER_TARGET_BUFFER_WRITES);

      if (GEN_IS_HASWELL) {
         /* Haswell needs addition work-arounds:
          *
          * From Haswell PRM, volume 2, part 1, "End-of-Pipe Synchronization":
          *
          *    Option 1:
          *    PIPE_CONTROL command with the CS Stall and the required write
          *    caches flushed with Post-SyncOperation as Write Immediate Data
          *    followed by eight dummy MI_STORE_DATA_IMM (write to scratch
          *    spce) commands.
          *
          *    Example:
          *       - Workload-1
          *       - PIPE_CONTROL (CS Stall, Post-Sync-Operation Write
          *         Immediate Data, Required Write Cache Flush bits set)
          *       - MI_STORE_DATA_IMM (8 times) (Dummy data, Scratch Address)
          *       - Workload-2 (Can use the data produce or output by
          *         Workload-1)
          *
          * Unfortunately, both the PRMs and the internal docs are a bit
          * out-of-date in this regard.  What the windows driver does (and
          * this appears to actually work) is to emit a register read from the
          * memory address written by the pipe control above.
          *
          * What register we load into doesn't matter.  We choose an indirect
          * rendering register because we know it always exists and it's one
          * of the first registers the command parser allows us to write.  If
          * you don't have command parser support in your kernel (pre-4.2),
          * this will get turned into MI_NOOP and you won't get the
          * workaround.  Unfortunately, there's just not much we can do in
          * that case.  This register is perfectly safe to write since we
          * always re-load all of the indirect draw registers right before
          * 3DPRIMITIVE when needed anyway.
          */
         anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
            lrm.RegisterAddress  = 0x243C; /* GEN7_3DPRIM_START_INSTANCE */
            lrm.MemoryAddress = cmd_buffer->device->workaround_address;
         }
      }

      bits &= ~(ANV_PIPE_FLUSH_BITS | ANV_PIPE_CS_STALL_BIT |
                ANV_PIPE_END_OF_PIPE_SYNC_BIT);
   }

   if (bits & ANV_PIPE_INVALIDATE_BITS) {
      /* From the SKL PRM, Vol. 2a, "PIPE_CONTROL",
       *
       *    "If the VF Cache Invalidation Enable is set to a 1 in a
       *    PIPE_CONTROL, a separate Null PIPE_CONTROL, all bitfields sets to
       *    0, with the VF Cache Invalidation Enable set to 0 needs to be sent
       *    prior to the PIPE_CONTROL with VF Cache Invalidation Enable set to
       *    a 1."
       *
       * This appears to hang Broadwell, so we restrict it to just gen9.
       */
      if (GEN_GEN == 9 && (bits & ANV_PIPE_VF_CACHE_INVALIDATE_BIT))
         anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe);

      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;

         /* From the SKL PRM, Vol. 2a, "PIPE_CONTROL",
          *
          *    "When VF Cache Invalidate is set “Post Sync Operation” must be
          *    enabled to “Write Immediate Data” or “Write PS Depth Count” or
          *    “Write Timestamp”.
          */
         if (GEN_GEN == 9 && pipe.VFCacheInvalidationEnable) {
            pipe.PostSyncOperation = WriteImmediateData;
            pipe.Address = cmd_buffer->device->workaround_address;
         }
      }

#if GEN_GEN == 12
      if ((bits & ANV_PIPE_AUX_TABLE_INVALIDATE_BIT) &&
          cmd_buffer->device->info.has_aux_map) {
         anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
            lri.RegisterOffset = GENX(GFX_CCS_AUX_INV_num);
            lri.DataDWord = 1;
         }
      }
#endif

      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;

      uint32_t base_layer, layer_count;
      if (image->type == VK_IMAGE_TYPE_3D) {
         base_layer = 0;
         layer_count = anv_minify(image->extent.depth, range->baseMipLevel);
      } else {
         base_layer = range->baseArrayLayer;
         layer_count = anv_get_layerCount(image, range);
      }

      if (range->aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT) {
         transition_depth_buffer(cmd_buffer, image,
                                 base_layer, layer_count,
                                 pImageMemoryBarriers[i].oldLayout,
                                 pImageMemoryBarriers[i].newLayout,
                                 false /* will_full_fast_clear */);
      }

      if (range->aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT) {
         transition_stencil_buffer(cmd_buffer, image,
                                   range->baseMipLevel,
                                   anv_get_levelCount(image, range),
                                   base_layer, layer_count,
                                   pImageMemoryBarriers[i].oldLayout,
                                   pImageMemoryBarriers[i].newLayout,
                                   false /* will_full_fast_clear */);
      }

      if (range->aspectMask & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) {
         VkImageAspectFlags color_aspects =
            anv_image_expand_aspects(image, range->aspectMask);
         uint32_t aspect_bit;
         anv_foreach_image_aspect_bit(aspect_bit, image, color_aspects) {
            transition_color_buffer(cmd_buffer, image, 1UL << aspect_bit,
                                    range->baseMipLevel,
                                    anv_get_levelCount(image, range),
                                    base_layer, layer_count,
                                    pImageMemoryBarriers[i].oldLayout,
                                    pImageMemoryBarriers[i].newLayout,
                                    false /* will_full_fast_clear */);
         }
      }
   }

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

static void
cmd_buffer_alloc_push_constants(struct anv_cmd_buffer *cmd_buffer)
{
   VkShaderStageFlags stages =
      cmd_buffer->state.gfx.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.gfx.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 =
      util_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.gfx.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 struct anv_address
anv_descriptor_set_address(struct anv_cmd_buffer *cmd_buffer,
                           struct anv_descriptor_set *set)
{
   if (set->pool) {
      /* This is a normal descriptor set */
      return (struct anv_address) {
         .bo = set->pool->bo,
         .offset = set->desc_mem.offset,
      };
   } else {
      /* This is a push descriptor set.  We have to flag it as used on the GPU
       * so that the next time we push descriptors, we grab a new memory.
       */
      struct anv_push_descriptor_set *push_set =
         (struct anv_push_descriptor_set *)set;
      push_set->set_used_on_gpu = true;

      return (struct anv_address) {
         .bo = cmd_buffer->dynamic_state_stream.state_pool->block_pool.bo,
         .offset = set->desc_mem.offset,
      };
   }
}

static VkResult
emit_binding_table(struct anv_cmd_buffer *cmd_buffer,
                   struct anv_cmd_pipeline_state *pipe_state,
                   struct anv_shader_bin *shader,
                   struct anv_state *bt_state)
{
   struct anv_subpass *subpass = cmd_buffer->state.subpass;
   uint32_t state_offset;

   struct anv_pipeline_bind_map *map = &shader->bind_map;
   if (map->surface_count == 0) {
      *bt_state = (struct anv_state) { 0, };
      return VK_SUCCESS;
   }

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

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

   /* We only need to emit relocs if we're not using softpin.  If we are using
    * softpin then we always keep all user-allocated memory objects resident.
    */
   const bool need_client_mem_relocs =
      !cmd_buffer->device->physical->use_softpin;
   struct anv_push_constants *push = &pipe_state->push_constants;

   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;

      switch (binding->set) {
      case ANV_DESCRIPTOR_SET_NULL:
         bt_map[s] = 0;
         break;

      case ANV_DESCRIPTOR_SET_COLOR_ATTACHMENTS:
         /* Color attachment binding */
         assert(shader->stage == MESA_SHADER_FRAGMENT);
         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.state;
            }
         } else {
            surface_state = cmd_buffer->state.null_surface_state;
         }

         assert(surface_state.map);
         bt_map[s] = surface_state.offset + state_offset;
         break;

      case ANV_DESCRIPTOR_SET_SHADER_CONSTANTS: {
         struct anv_state surface_state =
            anv_cmd_buffer_alloc_surface_state(cmd_buffer);

         struct anv_address constant_data = {
            .bo = cmd_buffer->device->instruction_state_pool.block_pool.bo,
            .offset = shader->kernel.offset +
                      shader->prog_data->const_data_offset,
         };
         unsigned constant_data_size = shader->prog_data->const_data_size;

         const enum isl_format format =
            anv_isl_format_for_descriptor_type(cmd_buffer->device,
                                               VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
         anv_fill_buffer_surface_state(cmd_buffer->device,
                                       surface_state, format,
                                       ISL_SURF_USAGE_CONSTANT_BUFFER_BIT,
                                       constant_data, constant_data_size, 1);

         assert(surface_state.map);
         bt_map[s] = surface_state.offset + state_offset;
         add_surface_reloc(cmd_buffer, surface_state, constant_data);
         break;
      }

      case ANV_DESCRIPTOR_SET_NUM_WORK_GROUPS: {
         /* This is always the first binding for compute shaders */
         assert(shader->stage == MESA_SHADER_COMPUTE && s == 0);

         struct anv_state surface_state =
            anv_cmd_buffer_alloc_surface_state(cmd_buffer);

         const enum isl_format format =
            anv_isl_format_for_descriptor_type(cmd_buffer->device,
                                               VK_DESCRIPTOR_TYPE_STORAGE_BUFFER);
         anv_fill_buffer_surface_state(cmd_buffer->device, surface_state,
                                       format,
                                       ISL_SURF_USAGE_CONSTANT_BUFFER_BIT,
                                       cmd_buffer->state.compute.num_workgroups,
                                       12, 1);

         assert(surface_state.map);
         bt_map[s] = surface_state.offset + state_offset;
         if (need_client_mem_relocs) {
            add_surface_reloc(cmd_buffer, surface_state,
                              cmd_buffer->state.compute.num_workgroups);
         }
         break;
      }

      case ANV_DESCRIPTOR_SET_DESCRIPTORS: {
         /* This is a descriptor set buffer so the set index is actually
          * given by binding->binding.  (Yes, that's confusing.)
          */
         struct anv_descriptor_set *set =
            pipe_state->descriptors[binding->index];
         assert(set->desc_mem.alloc_size);
         assert(set->desc_surface_state.alloc_size);
         bt_map[s] = set->desc_surface_state.offset + state_offset;
         add_surface_reloc(cmd_buffer, set->desc_surface_state,
                           anv_descriptor_set_address(cmd_buffer, set));
         break;
      }

      default: {
         assert(binding->set < MAX_SETS);
         const struct anv_descriptor_set *set =
            pipe_state->descriptors[binding->set];
         if (binding->index >= set->descriptor_count) {
            /* From the Vulkan spec section entitled "DescriptorSet and
             * Binding Assignment":
             *
             *    "If the array is runtime-sized, then array elements greater
             *    than or equal to the size of that binding in the bound
             *    descriptor set must not be used."
             *
             * Unfortunately, the compiler isn't smart enough to figure out
             * when a dynamic binding isn't used so it may grab the whole
             * array and stick it in the binding table.  In this case, it's
             * safe to just skip those bindings that are OOB.
             */
            assert(binding->index < set->layout->descriptor_count);
            continue;
         }
         const struct anv_descriptor *desc = &set->descriptors[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: {
            if (desc->image_view) {
               struct anv_surface_state sstate =
                  (desc->layout == VK_IMAGE_LAYOUT_GENERAL) ?
                  desc->image_view->planes[binding->plane].general_sampler_surface_state :
                  desc->image_view->planes[binding->plane].optimal_sampler_surface_state;
               surface_state = sstate.state;
               assert(surface_state.alloc_size);
               if (need_client_mem_relocs)
                  add_surface_state_relocs(cmd_buffer, sstate);
            } else {
               surface_state = cmd_buffer->device->null_surface_state;
            }
            break;
         }
         case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
            assert(shader->stage == MESA_SHADER_FRAGMENT);
            assert(desc->image_view != NULL);
            if ((desc->image_view->aspect_mask & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) == 0) {
               /* For depth and stencil input attachments, we treat it like any
                * old texture that a user may have bound.
                */
               assert(desc->image_view->n_planes == 1);
               struct anv_surface_state sstate =
                  (desc->layout == VK_IMAGE_LAYOUT_GENERAL) ?
                  desc->image_view->planes[0].general_sampler_surface_state :
                  desc->image_view->planes[0].optimal_sampler_surface_state;
               surface_state = sstate.state;
               assert(surface_state.alloc_size);
               if (need_client_mem_relocs)
                  add_surface_state_relocs(cmd_buffer, sstate);
            } 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.state;
            }
            break;

         case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: {
            if (desc->image_view) {
               struct anv_surface_state sstate = (binding->write_only)
                  ? desc->image_view->planes[binding->plane].writeonly_storage_surface_state
                  : desc->image_view->planes[binding->plane].storage_surface_state;
               surface_state = sstate.state;
               assert(surface_state.alloc_size);
               if (need_client_mem_relocs)
                  add_surface_state_relocs(cmd_buffer, sstate);
            } else {
               surface_state = cmd_buffer->device->null_surface_state;
            }
            break;
         }

         case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
         case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
         case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
            if (desc->buffer_view) {
               surface_state = desc->buffer_view->surface_state;
               assert(surface_state.alloc_size);
               if (need_client_mem_relocs) {
                  add_surface_reloc(cmd_buffer, surface_state,
                                    desc->buffer_view->address);
               }
            } else {
               surface_state = cmd_buffer->device->null_surface_state;
            }
            break;

         case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
         case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
            if (desc->buffer) {
               /* Compute the offset within the buffer */
               uint32_t dynamic_offset =
                  push->dynamic_offsets[binding->dynamic_offset_index];
               uint64_t offset = desc->offset + dynamic_offset;
               /* 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);

               /* Align the range for consistency */
               if (desc->type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC)
                  range = align_u32(range, ANV_UBO_ALIGNMENT);

               struct anv_address address =
                  anv_address_add(desc->buffer->address, offset);

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

               isl_surf_usage_flags_t usage =
                  desc->type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC ?
                  ISL_SURF_USAGE_CONSTANT_BUFFER_BIT :
                  ISL_SURF_USAGE_STORAGE_BIT;

               anv_fill_buffer_surface_state(cmd_buffer->device, surface_state,
                                             format, usage, address, range, 1);
               if (need_client_mem_relocs)
                  add_surface_reloc(cmd_buffer, surface_state, address);
            } else {
               surface_state = cmd_buffer->device->null_surface_state;
            }
            break;
         }

         case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
            if (desc->buffer_view) {
               surface_state = (binding->write_only)
                  ? desc->buffer_view->writeonly_storage_surface_state
                  : desc->buffer_view->storage_surface_state;
               assert(surface_state.alloc_size);
               if (need_client_mem_relocs) {
                  add_surface_reloc(cmd_buffer, surface_state,
                                    desc->buffer_view->address);
               }
            } else {
               surface_state = cmd_buffer->device->null_surface_state;
            }
            break;

         default:
            assert(!"Invalid descriptor type");
            continue;
         }
         assert(surface_state.map);
         bt_map[s] = surface_state.offset + state_offset;
         break;
      }
      }
   }

   return VK_SUCCESS;
}

static VkResult
emit_samplers(struct anv_cmd_buffer *cmd_buffer,
              struct anv_cmd_pipeline_state *pipe_state,
              struct anv_shader_bin *shader,
              struct anv_state *state)
{
   struct anv_pipeline_bind_map *map = &shader->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];
      const struct anv_descriptor *desc =
         &pipe_state->descriptors[binding->set]->descriptors[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[binding->plane], sizeof(sampler->state[0]));
   }

   return VK_SUCCESS;
}

static uint32_t
flush_descriptor_sets(struct anv_cmd_buffer *cmd_buffer,
                      struct anv_cmd_pipeline_state *pipe_state,
                      struct anv_shader_bin **shaders,
                      uint32_t num_shaders)
{
   const VkShaderStageFlags dirty = cmd_buffer->state.descriptors_dirty;
   VkShaderStageFlags flushed = 0;

   VkResult result = VK_SUCCESS;
   for (uint32_t i = 0; i < num_shaders; i++) {
      if (!shaders[i])
         continue;

      gl_shader_stage stage = shaders[i]->stage;
      VkShaderStageFlags vk_stage = mesa_to_vk_shader_stage(stage);
      if ((vk_stage & dirty) == 0)
         continue;

      result = emit_samplers(cmd_buffer, pipe_state, shaders[i],
                             &cmd_buffer->state.samplers[stage]);
      if (result != VK_SUCCESS)
         break;
      result = emit_binding_table(cmd_buffer, pipe_state, shaders[i],
                                  &cmd_buffer->state.binding_tables[stage]);
      if (result != VK_SUCCESS)
         break;

      flushed |= vk_stage;
   }

   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 */
      flushed = 0;

      for (uint32_t i = 0; i < num_shaders; i++) {
         if (!shaders[i])
            continue;

         gl_shader_stage stage = shaders[i]->stage;

         result = emit_samplers(cmd_buffer, pipe_state, shaders[i],
                                &cmd_buffer->state.samplers[stage]);
         if (result != VK_SUCCESS) {
            anv_batch_set_error(&cmd_buffer->batch, result);
            return 0;
         }
         result = emit_binding_table(cmd_buffer, pipe_state, shaders[i],
                                     &cmd_buffer->state.binding_tables[stage]);
         if (result != VK_SUCCESS) {
            anv_batch_set_error(&cmd_buffer->batch, result);
            return 0;
         }

         flushed |= mesa_to_vk_shader_stage(stage);
      }
   }

   cmd_buffer->state.descriptors_dirty &= ~flushed;

   return flushed;
}

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) {
      assert(s < ARRAY_SIZE(binding_table_opcodes));
      assert(binding_table_opcodes[s] > 0);

      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 struct anv_address
get_push_range_address(struct anv_cmd_buffer *cmd_buffer,
                       const struct anv_shader_bin *shader,
                       const struct anv_push_range *range)
{
   struct anv_cmd_graphics_state *gfx_state = &cmd_buffer->state.gfx;
   switch (range->set) {
   case ANV_DESCRIPTOR_SET_DESCRIPTORS: {
      /* This is a descriptor set buffer so the set index is
       * actually given by binding->binding.  (Yes, that's
       * confusing.)
       */
      struct anv_descriptor_set *set =
         gfx_state->base.descriptors[range->index];
      return anv_descriptor_set_address(cmd_buffer, set);
   }

   case ANV_DESCRIPTOR_SET_PUSH_CONSTANTS: {
      if (gfx_state->base.push_constants_state.alloc_size == 0) {
         gfx_state->base.push_constants_state =
            anv_cmd_buffer_gfx_push_constants(cmd_buffer);
      }
      return (struct anv_address) {
         .bo = cmd_buffer->device->dynamic_state_pool.block_pool.bo,
         .offset = gfx_state->base.push_constants_state.offset,
      };
   }

   case ANV_DESCRIPTOR_SET_SHADER_CONSTANTS:
      return (struct anv_address) {
         .bo = cmd_buffer->device->instruction_state_pool.block_pool.bo,
         .offset = shader->kernel.offset +
                   shader->prog_data->const_data_offset,
      };

   default: {
      assert(range->set < MAX_SETS);
      struct anv_descriptor_set *set =
         gfx_state->base.descriptors[range->set];
      const struct anv_descriptor *desc =
         &set->descriptors[range->index];

      if (desc->type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER) {
         if (desc->buffer_view)
            return desc->buffer_view->address;
      } else {
         assert(desc->type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC);
         if (desc->buffer) {
            const struct anv_push_constants *push =
               &gfx_state->base.push_constants;
            uint32_t dynamic_offset =
               push->dynamic_offsets[range->dynamic_offset_index];
            return anv_address_add(desc->buffer->address,
                                   desc->offset + dynamic_offset);
         }
      }

      /* For NULL UBOs, we just return an address in the workaround BO.  We do
       * writes to it for workarounds but always at the bottom.  The higher
       * bytes should be all zeros.
       */
      assert(range->length * 32 <= 2048);
      return (struct anv_address) {
         .bo = cmd_buffer->device->workaround_bo,
         .offset = 1024,
      };
   }
   }
}


/** Returns the size in bytes of the bound buffer
 *
 * The range is relative to the start of the buffer, not the start of the
 * range.  The returned range may be smaller than
 *
 *    (range->start + range->length) * 32;
 */
static uint32_t
get_push_range_bound_size(struct anv_cmd_buffer *cmd_buffer,
                          const struct anv_shader_bin *shader,
                          const struct anv_push_range *range)
{
   assert(shader->stage != MESA_SHADER_COMPUTE);
   const struct anv_cmd_graphics_state *gfx_state = &cmd_buffer->state.gfx;
   switch (range->set) {
   case ANV_DESCRIPTOR_SET_DESCRIPTORS: {
      struct anv_descriptor_set *set =
         gfx_state->base.descriptors[range->index];
      assert(range->start * 32 < set->desc_mem.alloc_size);
      assert((range->start + range->length) * 32 <= set->desc_mem.alloc_size);
      return set->desc_mem.alloc_size;
   }

   case ANV_DESCRIPTOR_SET_PUSH_CONSTANTS:
      return (range->start + range->length) * 32;

   case ANV_DESCRIPTOR_SET_SHADER_CONSTANTS:
      return ALIGN(shader->prog_data->const_data_size, ANV_UBO_ALIGNMENT);

   default: {
      assert(range->set < MAX_SETS);
      struct anv_descriptor_set *set =
         gfx_state->base.descriptors[range->set];
      const struct anv_descriptor *desc =
         &set->descriptors[range->index];

      if (desc->type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER) {
         if (!desc->buffer_view)
            return 0;

         if (range->start * 32 > desc->buffer_view->range)
            return 0;

         return desc->buffer_view->range;
      } else {
         if (!desc->buffer)
            return 0;

         assert(desc->type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC);
         /* Compute the offset within the buffer */
         const struct anv_push_constants *push =
            &gfx_state->base.push_constants;
         uint32_t dynamic_offset =
            push->dynamic_offsets[range->dynamic_offset_index];
         uint64_t offset = desc->offset + dynamic_offset;
         /* Clamp to the buffer size */
         offset = MIN2(offset, desc->buffer->size);
         /* Clamp the range to the buffer size */
         uint32_t bound_range = MIN2(desc->range, desc->buffer->size - offset);

         /* Align the range for consistency */
         bound_range = align_u32(bound_range, ANV_UBO_ALIGNMENT);

         return bound_range;
      }
   }
   }
}

static void
cmd_buffer_emit_push_constant(struct anv_cmd_buffer *cmd_buffer,
                              gl_shader_stage stage,
                              struct anv_address *buffers,
                              unsigned buffer_count)
{
   const struct anv_cmd_graphics_state *gfx_state = &cmd_buffer->state.gfx;
   const struct anv_graphics_pipeline *pipeline = gfx_state->pipeline;

   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,
   };

   assert(stage < ARRAY_SIZE(push_constant_opcodes));
   assert(push_constant_opcodes[stage] > 0);

   anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CONSTANT_VS), c) {
      c._3DCommandSubOpcode = push_constant_opcodes[stage];

      if (anv_pipeline_has_stage(pipeline, stage)) {
         const struct anv_pipeline_bind_map *bind_map =
            &pipeline->shaders[stage]->bind_map;

#if GEN_GEN >= 9
         /* This field exists since Gen8.  However, the Broadwell PRM says:
          *
          *    "Constant Buffer Object Control State must be always programmed
          *    to zero."
          *
          * This restriction does not exist on any newer platforms.
          *
          * We only have one MOCS field for the whole packet, not one per
          * buffer.  We could go out of our way here to walk over all of the
          * buffers and see if any of them are used externally and use the
          * external MOCS.  However, the notion that someone would use the
          * same bit of memory for both scanout and a UBO is nuts.  Let's not
          * bother and assume it's all internal.
          */
         c.MOCS = isl_mocs(&cmd_buffer->device->isl_dev, 0);
#endif

#if GEN_GEN >= 8 || GEN_IS_HASWELL
         /* The Skylake PRM contains the following restriction:
          *
          *    "The driver must ensure The following case does not occur
          *     without a flush to the 3D engine: 3DSTATE_CONSTANT_* with
          *     buffer 3 read length equal to zero committed followed by a
          *     3DSTATE_CONSTANT_* with buffer 0 read length not equal to
          *     zero committed."
          *
          * To avoid this, we program the buffers in the highest slots.
          * This way, slot 0 is only used if slot 3 is also used.
          */
         assert(buffer_count <= 4);
         const unsigned shift = 4 - buffer_count;
         for (unsigned i = 0; i < buffer_count; i++) {
            const struct anv_push_range *range = &bind_map->push_ranges[i];

            /* At this point we only have non-empty ranges */
            assert(range->length > 0);

            /* For Ivy Bridge, make sure we only set the first range (actual
             * push constants)
             */
            assert((GEN_GEN >= 8 || GEN_IS_HASWELL) || i == 0);

            c.ConstantBody.ReadLength[i + shift] = range->length;
            c.ConstantBody.Buffer[i + shift] =
               anv_address_add(buffers[i], range->start * 32);
         }
#else
         /* For Ivy Bridge, push constants are relative to dynamic state
          * base address and we only ever push actual push constants.
          */
         if (bind_map->push_ranges[0].length > 0) {
            assert(buffer_count == 1);
            assert(bind_map->push_ranges[0].set ==
                   ANV_DESCRIPTOR_SET_PUSH_CONSTANTS);
            assert(buffers[0].bo ==
                   cmd_buffer->device->dynamic_state_pool.block_pool.bo);
            c.ConstantBody.ReadLength[0] = bind_map->push_ranges[0].length;
            c.ConstantBody.Buffer[0].bo = NULL;
            c.ConstantBody.Buffer[0].offset = buffers[0].offset;
         }
         assert(bind_map->push_ranges[1].length == 0);
         assert(bind_map->push_ranges[2].length == 0);
         assert(bind_map->push_ranges[3].length == 0);
#endif
      }
   }
}

#if GEN_GEN >= 12
static void
cmd_buffer_emit_push_constant_all(struct anv_cmd_buffer *cmd_buffer,
                                  uint32_t shader_mask,
                                  struct anv_address *buffers,
                                  uint32_t buffer_count)
{
   if (buffer_count == 0) {
      anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CONSTANT_ALL), c) {
         c.ShaderUpdateEnable = shader_mask;
         c.MOCS = isl_mocs(&cmd_buffer->device->isl_dev, 0);
      }
      return;
   }

   const struct anv_cmd_graphics_state *gfx_state = &cmd_buffer->state.gfx;
   const struct anv_graphics_pipeline *pipeline = gfx_state->pipeline;

   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,
   };

   gl_shader_stage stage = vk_to_mesa_shader_stage(shader_mask);
   assert(stage < ARRAY_SIZE(push_constant_opcodes));
   assert(push_constant_opcodes[stage] > 0);

   const struct anv_pipeline_bind_map *bind_map =
      &pipeline->shaders[stage]->bind_map;

   uint32_t *dw;
   const uint32_t buffer_mask = (1 << buffer_count) - 1;
   const uint32_t num_dwords = 2 + 2 * buffer_count;

   dw = anv_batch_emitn(&cmd_buffer->batch, num_dwords,
                        GENX(3DSTATE_CONSTANT_ALL),
                        .ShaderUpdateEnable = shader_mask,
                        .PointerBufferMask = buffer_mask,
                        .MOCS = isl_mocs(&cmd_buffer->device->isl_dev, 0));

   for (int i = 0; i < buffer_count; i++) {
      const struct anv_push_range *range = &bind_map->push_ranges[i];
      GENX(3DSTATE_CONSTANT_ALL_DATA_pack)(
         &cmd_buffer->batch, dw + 2 + i * 2,
         &(struct GENX(3DSTATE_CONSTANT_ALL_DATA)) {
            .PointerToConstantBuffer =
               anv_address_add(buffers[i], range->start * 32),
            .ConstantBufferReadLength = range->length,
         });
   }
}
#endif

static void
cmd_buffer_flush_push_constants(struct anv_cmd_buffer *cmd_buffer,
                                VkShaderStageFlags dirty_stages)
{
   VkShaderStageFlags flushed = 0;
   struct anv_cmd_graphics_state *gfx_state = &cmd_buffer->state.gfx;
   const struct anv_graphics_pipeline *pipeline = gfx_state->pipeline;

#if GEN_GEN >= 12
   uint32_t nobuffer_stages = 0;
#endif

   /* Compute robust pushed register access mask for each stage. */
   if (cmd_buffer->device->robust_buffer_access) {
      anv_foreach_stage(stage, dirty_stages) {
         if (!anv_pipeline_has_stage(pipeline, stage))
            continue;

         const struct anv_shader_bin *shader = pipeline->shaders[stage];
         const struct anv_pipeline_bind_map *bind_map = &shader->bind_map;
         struct anv_push_constants *push = &gfx_state->base.push_constants;

         push->push_reg_mask[stage] = 0;
         /* Start of the current range in the shader, relative to the start of
          * push constants in the shader.
          */
         unsigned range_start_reg = 0;
         for (unsigned i = 0; i < 4; i++) {
            const struct anv_push_range *range = &bind_map->push_ranges[i];
            if (range->length == 0)
               continue;

            unsigned bound_size =
               get_push_range_bound_size(cmd_buffer, shader, range);
            if (bound_size >= range->start * 32) {
               unsigned bound_regs =
                  MIN2(DIV_ROUND_UP(bound_size, 32) - range->start,
                       range->length);
               assert(range_start_reg + bound_regs <= 64);
               push->push_reg_mask[stage] |= BITFIELD64_RANGE(range_start_reg,
                                                              bound_regs);
            }

            cmd_buffer->state.push_constants_dirty |=
               mesa_to_vk_shader_stage(stage);

            range_start_reg += range->length;
         }
      }
   }

   /* Resets the push constant state so that we allocate a new one if
    * needed.
    */
   gfx_state->base.push_constants_state = ANV_STATE_NULL;

   anv_foreach_stage(stage, dirty_stages) {
      unsigned buffer_count = 0;
      flushed |= mesa_to_vk_shader_stage(stage);
      UNUSED uint32_t max_push_range = 0;

      struct anv_address buffers[4] = {};
      if (anv_pipeline_has_stage(pipeline, stage)) {
         const struct anv_shader_bin *shader = pipeline->shaders[stage];
         const struct anv_pipeline_bind_map *bind_map = &shader->bind_map;

         /* We have to gather buffer addresses as a second step because the
          * loop above puts data into the push constant area and the call to
          * get_push_range_address is what locks our push constants and copies
          * them into the actual GPU buffer.  If we did the two loops at the
          * same time, we'd risk only having some of the sizes in the push
          * constant buffer when we did the copy.
          */
         for (unsigned i = 0; i < 4; i++) {
            const struct anv_push_range *range = &bind_map->push_ranges[i];
            if (range->length == 0)
               break;

            buffers[i] = get_push_range_address(cmd_buffer, shader, range);
            max_push_range = MAX2(max_push_range, range->length);
            buffer_count++;
         }

         /* We have at most 4 buffers but they should be tightly packed */
         for (unsigned i = buffer_count; i < 4; i++)
            assert(bind_map->push_ranges[i].length == 0);
      }

#if GEN_GEN >= 12
      /* If this stage doesn't have any push constants, emit it later in a
       * single CONSTANT_ALL packet.
       */
      if (buffer_count == 0) {
         nobuffer_stages |= 1 << stage;
         continue;
      }

      /* The Constant Buffer Read Length field from 3DSTATE_CONSTANT_ALL
       * contains only 5 bits, so we can only use it for buffers smaller than
       * 32.
       */
      if (max_push_range < 32) {
         cmd_buffer_emit_push_constant_all(cmd_buffer, 1 << stage,
                                           buffers, buffer_count);
         continue;
      }
#endif

      cmd_buffer_emit_push_constant(cmd_buffer, stage, buffers, buffer_count);
   }

#if GEN_GEN >= 12
   if (nobuffer_stages)
      cmd_buffer_emit_push_constant_all(cmd_buffer, nobuffer_stages, NULL, 0);
#endif

   cmd_buffer->state.push_constants_dirty &= ~flushed;
}

static void
cmd_buffer_emit_clip(struct anv_cmd_buffer *cmd_buffer)
{
   const uint32_t clip_states =
#if GEN_GEN <= 7
      ANV_CMD_DIRTY_DYNAMIC_FRONT_FACE |
      ANV_CMD_DIRTY_DYNAMIC_CULL_MODE |
#endif
      ANV_CMD_DIRTY_DYNAMIC_VIEWPORT |
      ANV_CMD_DIRTY_PIPELINE;

   if ((cmd_buffer->state.gfx.dirty & clip_states) == 0)
      return;

#if GEN_GEN <= 7
   const struct anv_dynamic_state *d = &cmd_buffer->state.gfx.dynamic;
#endif
   struct GENX(3DSTATE_CLIP) clip = {
      GENX(3DSTATE_CLIP_header),
#if GEN_GEN <= 7
      .FrontWinding = genX(vk_to_gen_front_face)[d->front_face],
      .CullMode     = genX(vk_to_gen_cullmode)[d->cull_mode],
#endif
   };
   uint32_t dwords[GENX(3DSTATE_CLIP_length)];

   struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline;
   const struct brw_vue_prog_data *last =
      anv_pipeline_get_last_vue_prog_data(pipeline);
   if (last->vue_map.slots_valid & VARYING_BIT_VIEWPORT) {
      clip.MaximumVPIndex =
         cmd_buffer->state.gfx.dynamic.viewport.count > 0 ?
         cmd_buffer->state.gfx.dynamic.viewport.count - 1 : 0;
   }

   GENX(3DSTATE_CLIP_pack)(NULL, dwords, &clip);
   anv_batch_emit_merge(&cmd_buffer->batch, dwords,
                        pipeline->gen7.clip);
}

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

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

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

   genX(cmd_buffer_emit_hashing_mode)(cmd_buffer, UINT_MAX, UINT_MAX, 1);

   genX(flush_pipeline_select_3d)(cmd_buffer);

   /* Apply any pending pipeline flushes we may have.  We want to apply them
    * now because, if any of those flushes are for things like push constants,
    * the GPU will read the state at weird times.
    */
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   uint32_t vb_emit = cmd_buffer->state.gfx.vb_dirty & pipeline->vb_used;
   if (cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_PIPELINE)
      vb_emit |= pipeline->vb_used;

   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;

         /* If dynamic, use stride/size from vertex binding, otherwise use
          * stride/size that was setup in the pipeline object.
          */
         bool dynamic_stride = cmd_buffer->state.gfx.dynamic.dyn_vbo_stride;
         bool dynamic_size = cmd_buffer->state.gfx.dynamic.dyn_vbo_size;

         struct GENX(VERTEX_BUFFER_STATE) state;
         if (buffer) {
            uint32_t stride = dynamic_stride ?
               cmd_buffer->state.vertex_bindings[vb].stride : pipeline->vb[vb].stride;
            /* From the Vulkan spec (vkCmdBindVertexBuffers2EXT):
             *
             * "If pname:pSizes is not NULL then pname:pSizes[i] specifies
             * the bound size of the vertex buffer starting from the corresponding
             * elements of pname:pBuffers[i] plus pname:pOffsets[i]."
             */
            UNUSED uint32_t size = dynamic_size ?
               cmd_buffer->state.vertex_bindings[vb].size : buffer->size - offset;

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

               .MOCS = anv_mocs(cmd_buffer->device, buffer->address.bo,
                                ISL_SURF_USAGE_VERTEX_BUFFER_BIT),
#if GEN_GEN <= 7
               .BufferAccessType = pipeline->vb[vb].instanced ? INSTANCEDATA : VERTEXDATA,
               .InstanceDataStepRate = pipeline->vb[vb].instance_divisor,
#endif
               .AddressModifyEnable = true,
               .BufferPitch = stride,
               .BufferStartingAddress = anv_address_add(buffer->address, offset),
               .NullVertexBuffer = offset >= buffer->size,

#if GEN_GEN >= 8
               .BufferSize = size,
#else
               /* XXX: to handle dynamic offset for older gens we might want
                * to modify Endaddress, but there are issues when doing so:
                *
                * https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7439
                */
               .EndAddress = anv_address_add(buffer->address, buffer->size - 1),
#endif
            };
         } else {
            state = (struct GENX(VERTEX_BUFFER_STATE)) {
               .VertexBufferIndex = vb,
               .NullVertexBuffer = true,
            };
         }

#if GEN_GEN >= 8 && GEN_GEN <= 9
         genX(cmd_buffer_set_binding_for_gen8_vb_flush)(cmd_buffer, vb,
                                                        state.BufferStartingAddress,
                                                        state.BufferSize);
#endif

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

   cmd_buffer->state.gfx.vb_dirty &= ~vb_emit;

   if ((cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_XFB_ENABLE) ||
       (GEN_GEN == 7 && (cmd_buffer->state.gfx.dirty &
                         ANV_CMD_DIRTY_PIPELINE))) {
      /* We don't need any per-buffer dirty tracking because you're not
       * allowed to bind different XFB buffers while XFB is enabled.
       */
      for (unsigned idx = 0; idx < MAX_XFB_BUFFERS; idx++) {
         struct anv_xfb_binding *xfb = &cmd_buffer->state.xfb_bindings[idx];
         anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_SO_BUFFER), sob) {
#if GEN_GEN < 12
            sob.SOBufferIndex = idx;
#else
            sob._3DCommandOpcode = 0;
            sob._3DCommandSubOpcode = SO_BUFFER_INDEX_0_CMD + idx;
#endif

            if (cmd_buffer->state.xfb_enabled && xfb->buffer && xfb->size != 0) {
               sob.MOCS = isl_mocs(&cmd_buffer->device->isl_dev, 0);
               sob.SurfaceBaseAddress = anv_address_add(xfb->buffer->address,
                                                        xfb->offset);
#if GEN_GEN >= 8
               sob.SOBufferEnable = true;
               sob.StreamOffsetWriteEnable = false;
               /* Size is in DWords - 1 */
               sob.SurfaceSize = DIV_ROUND_UP(xfb->size, 4) - 1;
#else
               /* We don't have SOBufferEnable in 3DSTATE_SO_BUFFER on Gen7 so
                * we trust in SurfaceEndAddress = SurfaceBaseAddress = 0 (the
                * default for an empty SO_BUFFER packet) to disable them.
                */
               sob.SurfacePitch = pipeline->gen7.xfb_bo_pitch[idx];
               sob.SurfaceEndAddress = anv_address_add(xfb->buffer->address,
                                                       xfb->offset + xfb->size);
#endif
            }
         }
      }

      /* CNL and later require a CS stall after 3DSTATE_SO_BUFFER */
      if (GEN_GEN >= 10)
         cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_CS_STALL_BIT;
   }

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

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

   if (cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_PIPELINE)
      cmd_buffer->state.gfx.primitive_topology = pipeline->topology;

#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           = cmd_buffer->device->workaround_address;
      }
   }
#endif

   /* Render targets live in the same binding table as fragment descriptors */
   if (cmd_buffer->state.gfx.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,
                                    &cmd_buffer->state.gfx.base,
                                    pipeline->shaders,
                                    ARRAY_SIZE(pipeline->shaders));
   }

   if (dirty || cmd_buffer->state.push_constants_dirty) {
      /* Because we're pushing UBOs, we have to push whenever either
       * descriptors or push constants is dirty.
       */
      dirty |= cmd_buffer->state.push_constants_dirty;
      dirty &= ANV_STAGE_MASK & VK_SHADER_STAGE_ALL_GRAPHICS;
      cmd_buffer_flush_push_constants(cmd_buffer, dirty);
   }

   if (dirty)
      cmd_buffer_emit_descriptor_pointers(cmd_buffer, dirty);

   cmd_buffer_emit_clip(cmd_buffer);

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

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

   genX(cmd_buffer_flush_dynamic_state)(cmd_buffer);
}

static void
emit_vertex_bo(struct anv_cmd_buffer *cmd_buffer,
               struct anv_address addr,
               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,
         .MOCS = addr.bo ? anv_mocs(cmd_buffer->device, addr.bo,
                                    ISL_SURF_USAGE_VERTEX_BUFFER_BIT) : 0,
         .NullVertexBuffer = size == 0,
#if (GEN_GEN >= 8)
         .BufferStartingAddress = addr,
         .BufferSize = size
#else
         .BufferStartingAddress = addr,
         .EndAddress = anv_address_add(addr, size),
#endif
      });

   genX(cmd_buffer_set_binding_for_gen8_vb_flush)(cmd_buffer,
                                                  index, addr, size);
}

static void
emit_base_vertex_instance_bo(struct anv_cmd_buffer *cmd_buffer,
                             struct anv_address addr)
{
   emit_vertex_bo(cmd_buffer, addr, addr.bo ? 8 : 0, ANV_SVGS_VB_INDEX);
}

static void
emit_base_vertex_instance(struct anv_cmd_buffer *cmd_buffer,
                          uint32_t base_vertex, uint32_t base_instance)
{
   if (base_vertex == 0 && base_instance == 0) {
      emit_base_vertex_instance_bo(cmd_buffer, ANV_NULL_ADDRESS);
   } else {
      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;

      struct anv_address addr = {
         .bo = cmd_buffer->device->dynamic_state_pool.block_pool.bo,
         .offset = id_state.offset,
      };

      emit_base_vertex_instance_bo(cmd_buffer, addr);
   }
}

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;

   struct anv_address addr = {
      .bo = cmd_buffer->device->dynamic_state_pool.block_pool.bo,
      .offset = state.offset,
   };

   emit_vertex_bo(cmd_buffer, addr, 4, ANV_DRAWID_VB_INDEX);
}

static void
update_dirty_vbs_for_gen8_vb_flush(struct anv_cmd_buffer *cmd_buffer,
                                   uint32_t access_type)
{
   struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline;
   const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);

   uint64_t vb_used = pipeline->vb_used;
   if (vs_prog_data->uses_firstvertex ||
       vs_prog_data->uses_baseinstance)
      vb_used |= 1ull << ANV_SVGS_VB_INDEX;
   if (vs_prog_data->uses_drawid)
      vb_used |= 1ull << ANV_DRAWID_VB_INDEX;

   genX(cmd_buffer_update_dirty_vbs_for_gen8_vb_flush)(cmd_buffer,
                                                       access_type == RANDOM,
                                                       vb_used);
}

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_graphics_pipeline *pipeline = cmd_buffer->state.gfx.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 (cmd_buffer->state.conditional_render_enabled)
      genX(cmd_emit_conditional_render_predicate)(cmd_buffer);

   if (vs_prog_data->uses_firstvertex ||
       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);

   /* Emitting draw index or vertex index BOs may result in needing
    * additional VF cache flushes.
    */
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

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

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

   update_dirty_vbs_for_gen8_vb_flush(cmd_buffer, SEQUENTIAL);
}

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_graphics_pipeline *pipeline = cmd_buffer->state.gfx.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 (cmd_buffer->state.conditional_render_enabled)
      genX(cmd_emit_conditional_render_predicate)(cmd_buffer);

   if (vs_prog_data->uses_firstvertex ||
       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);

   /* Emitting draw index or vertex index BOs may result in needing
    * additional VF cache flushes.
    */
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

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

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

   update_dirty_vbs_for_gen8_vb_flush(cmd_buffer, RANDOM);
}

/* 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

void genX(CmdDrawIndirectByteCountEXT)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    instanceCount,
    uint32_t                                    firstInstance,
    VkBuffer                                    counterBuffer,
    VkDeviceSize                                counterBufferOffset,
    uint32_t                                    counterOffset,
    uint32_t                                    vertexStride)
{
#if GEN_IS_HASWELL || GEN_GEN >= 8
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_buffer, counter_buffer, counterBuffer);
   struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline;
   const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);

   /* firstVertex is always zero for this draw function */
   const uint32_t firstVertex = 0;

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

   genX(cmd_buffer_flush_state)(cmd_buffer);

   if (vs_prog_data->uses_firstvertex ||
       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);

   /* Emitting draw index or vertex index BOs may result in needing
    * additional VF cache flushes.
    */
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

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

   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);
   struct gen_mi_value count =
      gen_mi_mem32(anv_address_add(counter_buffer->address,
                                   counterBufferOffset));
   if (counterOffset)
      count = gen_mi_isub(&b, count, gen_mi_imm(counterOffset));
   count = gen_mi_udiv32_imm(&b, count, vertexStride);
   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_VERTEX_COUNT), count);

   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_VERTEX),
                    gen_mi_imm(firstVertex));
   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_INSTANCE_COUNT),
                    gen_mi_imm(instanceCount));
   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_INSTANCE),
                    gen_mi_imm(firstInstance));
   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_BASE_VERTEX), gen_mi_imm(0));

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

   update_dirty_vbs_for_gen8_vb_flush(cmd_buffer, SEQUENTIAL);
#endif /* GEN_IS_HASWELL || GEN_GEN >= 8 */
}

static void
load_indirect_parameters(struct anv_cmd_buffer *cmd_buffer,
                         struct anv_address addr,
                         bool indexed)
{
   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_VERTEX_COUNT),
                    gen_mi_mem32(anv_address_add(addr, 0)));

   struct gen_mi_value instance_count = gen_mi_mem32(anv_address_add(addr, 4));
   unsigned view_count = anv_subpass_view_count(cmd_buffer->state.subpass);
   if (view_count > 1) {
#if GEN_IS_HASWELL || GEN_GEN >= 8
      instance_count = gen_mi_imul_imm(&b, instance_count, view_count);
#else
      anv_finishme("Multiview + indirect draw requires MI_MATH; "
                   "MI_MATH is not supported on Ivy Bridge");
#endif
   }
   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_INSTANCE_COUNT), instance_count);

   gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_VERTEX),
                    gen_mi_mem32(anv_address_add(addr, 8)));

   if (indexed) {
      gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_BASE_VERTEX),
                       gen_mi_mem32(anv_address_add(addr, 12)));
      gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_INSTANCE),
                       gen_mi_mem32(anv_address_add(addr, 16)));
   } else {
      gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_INSTANCE),
                       gen_mi_mem32(anv_address_add(addr, 12)));
      gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_BASE_VERTEX), gen_mi_imm(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_graphics_pipeline *pipeline = cmd_buffer->state.gfx.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 (cmd_buffer->state.conditional_render_enabled)
      genX(cmd_emit_conditional_render_predicate)(cmd_buffer);

   for (uint32_t i = 0; i < drawCount; i++) {
      struct anv_address draw = anv_address_add(buffer->address, offset);

      if (vs_prog_data->uses_firstvertex ||
          vs_prog_data->uses_baseinstance)
         emit_base_vertex_instance_bo(cmd_buffer, anv_address_add(draw, 8));
      if (vs_prog_data->uses_drawid)
         emit_draw_index(cmd_buffer, i);

      /* Emitting draw index or vertex index BOs may result in needing
       * additional VF cache flushes.
       */
      genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

      load_indirect_parameters(cmd_buffer, draw, false);

      anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
         prim.IndirectParameterEnable  = true;
         prim.PredicateEnable          = cmd_buffer->state.conditional_render_enabled;
         prim.VertexAccessType         = SEQUENTIAL;
         prim.PrimitiveTopologyType    = cmd_buffer->state.gfx.primitive_topology;
      }

      update_dirty_vbs_for_gen8_vb_flush(cmd_buffer, SEQUENTIAL);

      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_graphics_pipeline *pipeline = cmd_buffer->state.gfx.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 (cmd_buffer->state.conditional_render_enabled)
      genX(cmd_emit_conditional_render_predicate)(cmd_buffer);

   for (uint32_t i = 0; i < drawCount; i++) {
      struct anv_address draw = anv_address_add(buffer->address, offset);

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

      /* Emitting draw index or vertex index BOs may result in needing
       * additional VF cache flushes.
       */
      genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

      load_indirect_parameters(cmd_buffer, draw, true);

      anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
         prim.IndirectParameterEnable  = true;
         prim.PredicateEnable          = cmd_buffer->state.conditional_render_enabled;
         prim.VertexAccessType         = RANDOM;
         prim.PrimitiveTopologyType    = cmd_buffer->state.gfx.primitive_topology;
      }

      update_dirty_vbs_for_gen8_vb_flush(cmd_buffer, RANDOM);

      offset += stride;
   }
}

static struct gen_mi_value
prepare_for_draw_count_predicate(struct anv_cmd_buffer *cmd_buffer,
                                 struct gen_mi_builder *b,
                                 struct anv_address count_address,
                                 const bool conditional_render_enabled)
{
   struct gen_mi_value ret = gen_mi_imm(0);

   if (conditional_render_enabled) {
#if GEN_GEN >= 8 || GEN_IS_HASWELL
      ret = gen_mi_new_gpr(b);
      gen_mi_store(b, gen_mi_value_ref(b, ret), gen_mi_mem32(count_address));
#endif
   } else {
      /* Upload the current draw count from the draw parameters buffer to
       * MI_PREDICATE_SRC0.
       */
      gen_mi_store(b, gen_mi_reg64(MI_PREDICATE_SRC0),
                      gen_mi_mem32(count_address));

      gen_mi_store(b, gen_mi_reg32(MI_PREDICATE_SRC1 + 4), gen_mi_imm(0));
   }

   return ret;
}

static void
emit_draw_count_predicate(struct anv_cmd_buffer *cmd_buffer,
                          struct gen_mi_builder *b,
                          uint32_t draw_index)
{
   /* Upload the index of the current primitive to MI_PREDICATE_SRC1. */
   gen_mi_store(b, gen_mi_reg32(MI_PREDICATE_SRC1), gen_mi_imm(draw_index));

   if (draw_index == 0) {
      anv_batch_emit(&cmd_buffer->batch, GENX(MI_PREDICATE), mip) {
         mip.LoadOperation    = LOAD_LOADINV;
         mip.CombineOperation = COMBINE_SET;
         mip.CompareOperation = COMPARE_SRCS_EQUAL;
      }
   } else {
      /* While draw_index < draw_count the predicate's result will be
       *  (draw_index == draw_count) ^ TRUE = TRUE
       * When draw_index == draw_count the result is
       *  (TRUE) ^ TRUE = FALSE
       * After this all results will be:
       *  (FALSE) ^ FALSE = FALSE
       */
      anv_batch_emit(&cmd_buffer->batch, GENX(MI_PREDICATE), mip) {
         mip.LoadOperation    = LOAD_LOAD;
         mip.CombineOperation = COMBINE_XOR;
         mip.CompareOperation = COMPARE_SRCS_EQUAL;
      }
   }
}

#if GEN_GEN >= 8 || GEN_IS_HASWELL
static void
emit_draw_count_predicate_with_conditional_render(
                          struct anv_cmd_buffer *cmd_buffer,
                          struct gen_mi_builder *b,
                          uint32_t draw_index,
                          struct gen_mi_value max)
{
   struct gen_mi_value pred = gen_mi_ult(b, gen_mi_imm(draw_index), max);
   pred = gen_mi_iand(b, pred, gen_mi_reg64(ANV_PREDICATE_RESULT_REG));

#if GEN_GEN >= 8
   gen_mi_store(b, gen_mi_reg32(MI_PREDICATE_RESULT), pred);
#else
   /* MI_PREDICATE_RESULT is not whitelisted in i915 command parser
    * so we emit MI_PREDICATE to set it.
    */

   gen_mi_store(b, gen_mi_reg64(MI_PREDICATE_SRC0), pred);
   gen_mi_store(b, gen_mi_reg64(MI_PREDICATE_SRC1), gen_mi_imm(0));

   anv_batch_emit(&cmd_buffer->batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOADINV;
      mip.CombineOperation = COMBINE_SET;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }
#endif
}
#endif

void genX(CmdDrawIndirectCount)(
    VkCommandBuffer                             commandBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset,
    VkBuffer                                    _countBuffer,
    VkDeviceSize                                countBufferOffset,
    uint32_t                                    maxDrawCount,
    uint32_t                                    stride)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   ANV_FROM_HANDLE(anv_buffer, count_buffer, _countBuffer);
   struct anv_cmd_state *cmd_state = &cmd_buffer->state;
   struct anv_graphics_pipeline *pipeline = cmd_state->gfx.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);

   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);
   struct anv_address count_address =
      anv_address_add(count_buffer->address, countBufferOffset);
   struct gen_mi_value max =
      prepare_for_draw_count_predicate(cmd_buffer, &b, count_address,
                                       cmd_state->conditional_render_enabled);

   for (uint32_t i = 0; i < maxDrawCount; i++) {
      struct anv_address draw = anv_address_add(buffer->address, offset);

#if GEN_GEN >= 8 || GEN_IS_HASWELL
      if (cmd_state->conditional_render_enabled) {
         emit_draw_count_predicate_with_conditional_render(
            cmd_buffer, &b, i, gen_mi_value_ref(&b, max));
      } else {
         emit_draw_count_predicate(cmd_buffer, &b, i);
      }
#else
      emit_draw_count_predicate(cmd_buffer, &b, i);
#endif

      if (vs_prog_data->uses_firstvertex ||
          vs_prog_data->uses_baseinstance)
         emit_base_vertex_instance_bo(cmd_buffer, anv_address_add(draw, 8));
      if (vs_prog_data->uses_drawid)
         emit_draw_index(cmd_buffer, i);

      /* Emitting draw index or vertex index BOs may result in needing
       * additional VF cache flushes.
       */
      genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

      load_indirect_parameters(cmd_buffer, draw, false);

      anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
         prim.IndirectParameterEnable  = true;
         prim.PredicateEnable          = true;
         prim.VertexAccessType         = SEQUENTIAL;
         prim.PrimitiveTopologyType    = cmd_buffer->state.gfx.primitive_topology;
      }

      update_dirty_vbs_for_gen8_vb_flush(cmd_buffer, SEQUENTIAL);

      offset += stride;
   }

   gen_mi_value_unref(&b, max);
}

void genX(CmdDrawIndexedIndirectCount)(
    VkCommandBuffer                             commandBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset,
    VkBuffer                                    _countBuffer,
    VkDeviceSize                                countBufferOffset,
    uint32_t                                    maxDrawCount,
    uint32_t                                    stride)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   ANV_FROM_HANDLE(anv_buffer, count_buffer, _countBuffer);
   struct anv_cmd_state *cmd_state = &cmd_buffer->state;
   struct anv_graphics_pipeline *pipeline = cmd_state->gfx.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);

   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);
   struct anv_address count_address =
      anv_address_add(count_buffer->address, countBufferOffset);
   struct gen_mi_value max =
      prepare_for_draw_count_predicate(cmd_buffer, &b, count_address,
                                       cmd_state->conditional_render_enabled);

   for (uint32_t i = 0; i < maxDrawCount; i++) {
      struct anv_address draw = anv_address_add(buffer->address, offset);

#if GEN_GEN >= 8 || GEN_IS_HASWELL
      if (cmd_state->conditional_render_enabled) {
         emit_draw_count_predicate_with_conditional_render(
            cmd_buffer, &b, i, gen_mi_value_ref(&b, max));
      } else {
         emit_draw_count_predicate(cmd_buffer, &b, i);
      }
#else
      emit_draw_count_predicate(cmd_buffer, &b, i);
#endif

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

      /* Emitting draw index or vertex index BOs may result in needing
       * additional VF cache flushes.
       */
      genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

      load_indirect_parameters(cmd_buffer, draw, true);

      anv_batch_emit(&cmd_buffer->batch, GENX(3DPRIMITIVE), prim) {
         prim.IndirectParameterEnable  = true;
         prim.PredicateEnable          = true;
         prim.VertexAccessType         = RANDOM;
         prim.PrimitiveTopologyType    = cmd_buffer->state.gfx.primitive_topology;
      }

      update_dirty_vbs_for_gen8_vb_flush(cmd_buffer, RANDOM);

      offset += stride;
   }

   gen_mi_value_unref(&b, max);
}

void genX(CmdBeginTransformFeedbackEXT)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    firstCounterBuffer,
    uint32_t                                    counterBufferCount,
    const VkBuffer*                             pCounterBuffers,
    const VkDeviceSize*                         pCounterBufferOffsets)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   assert(firstCounterBuffer < MAX_XFB_BUFFERS);
   assert(counterBufferCount <= MAX_XFB_BUFFERS);
   assert(firstCounterBuffer + counterBufferCount <= MAX_XFB_BUFFERS);

   /* From the SKL PRM Vol. 2c, SO_WRITE_OFFSET:
    *
    *    "Ssoftware must ensure that no HW stream output operations can be in
    *    process or otherwise pending at the point that the MI_LOAD/STORE
    *    commands are processed. This will likely require a pipeline flush."
    */
   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_CS_STALL_BIT;
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   for (uint32_t idx = 0; idx < MAX_XFB_BUFFERS; idx++) {
      /* If we have a counter buffer, this is a resume so we need to load the
       * value into the streamout offset register.  Otherwise, this is a begin
       * and we need to reset it to zero.
       */
      if (pCounterBuffers &&
          idx >= firstCounterBuffer &&
          idx - firstCounterBuffer < counterBufferCount &&
          pCounterBuffers[idx - firstCounterBuffer] != VK_NULL_HANDLE) {
         uint32_t cb_idx = idx - firstCounterBuffer;
         ANV_FROM_HANDLE(anv_buffer, counter_buffer, pCounterBuffers[cb_idx]);
         uint64_t offset = pCounterBufferOffsets ?
                           pCounterBufferOffsets[cb_idx] : 0;

         anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
            lrm.RegisterAddress  = GENX(SO_WRITE_OFFSET0_num) + idx * 4;
            lrm.MemoryAddress    = anv_address_add(counter_buffer->address,
                                                   offset);
         }
      } else {
         anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
            lri.RegisterOffset   = GENX(SO_WRITE_OFFSET0_num) + idx * 4;
            lri.DataDWord        = 0;
         }
      }
   }

   cmd_buffer->state.xfb_enabled = true;
   cmd_buffer->state.gfx.dirty |= ANV_CMD_DIRTY_XFB_ENABLE;
}

void genX(CmdEndTransformFeedbackEXT)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    firstCounterBuffer,
    uint32_t                                    counterBufferCount,
    const VkBuffer*                             pCounterBuffers,
    const VkDeviceSize*                         pCounterBufferOffsets)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   assert(firstCounterBuffer < MAX_XFB_BUFFERS);
   assert(counterBufferCount <= MAX_XFB_BUFFERS);
   assert(firstCounterBuffer + counterBufferCount <= MAX_XFB_BUFFERS);

   /* From the SKL PRM Vol. 2c, SO_WRITE_OFFSET:
    *
    *    "Ssoftware must ensure that no HW stream output operations can be in
    *    process or otherwise pending at the point that the MI_LOAD/STORE
    *    commands are processed. This will likely require a pipeline flush."
    */
   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_CS_STALL_BIT;
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   for (uint32_t cb_idx = 0; cb_idx < counterBufferCount; cb_idx++) {
      unsigned idx = firstCounterBuffer + cb_idx;

      /* If we have a counter buffer, this is a resume so we need to load the
       * value into the streamout offset register.  Otherwise, this is a begin
       * and we need to reset it to zero.
       */
      if (pCounterBuffers &&
          cb_idx < counterBufferCount &&
          pCounterBuffers[cb_idx] != VK_NULL_HANDLE) {
         ANV_FROM_HANDLE(anv_buffer, counter_buffer, pCounterBuffers[cb_idx]);
         uint64_t offset = pCounterBufferOffsets ?
                           pCounterBufferOffsets[cb_idx] : 0;

         anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_REGISTER_MEM), srm) {
            srm.MemoryAddress    = anv_address_add(counter_buffer->address,
                                                   offset);
            srm.RegisterAddress  = GENX(SO_WRITE_OFFSET0_num) + idx * 4;
         }
      }
   }

   cmd_buffer->state.xfb_enabled = false;
   cmd_buffer->state.gfx.dirty |= ANV_CMD_DIRTY_XFB_ENABLE;
}

void
genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_compute_pipeline *pipeline = cmd_buffer->state.compute.pipeline;

   assert(pipeline->cs);

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

   genX(flush_pipeline_select_gpgpu)(cmd_buffer);

   /* Apply any pending pipeline flushes we may have.  We want to apply them
    * now because, if any of those flushes are for things like push constants,
    * the GPU will read the state at weird times.
    */
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   if (cmd_buffer->state.compute.pipeline_dirty) {
      /* 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->base.batch);

      /* The workgroup size of the pipeline affects our push constant layout
       * so flag push constants as dirty if we change the pipeline.
       */
      cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_COMPUTE_BIT;
   }

   if ((cmd_buffer->state.descriptors_dirty & VK_SHADER_STAGE_COMPUTE_BIT) ||
       cmd_buffer->state.compute.pipeline_dirty) {
      flush_descriptor_sets(cmd_buffer,
                            &cmd_buffer->state.compute.base,
                            &pipeline->cs, 1);

      uint32_t iface_desc_data_dw[GENX(INTERFACE_DESCRIPTOR_DATA_length)];
      struct GENX(INTERFACE_DESCRIPTOR_DATA) desc = {
         .BindingTablePointer =
            cmd_buffer->state.binding_tables[MESA_SHADER_COMPUTE].offset,
         .SamplerStatePointer =
            cmd_buffer->state.samplers[MESA_SHADER_COMPUTE].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;
      }
   }

   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.push_constants_dirty &= ~VK_SHADER_STAGE_COMPUTE_BIT;
   }

   cmd_buffer->state.compute.pipeline_dirty = false;

   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->physical->cmd_parser_version < required_version) {
      return vk_errorf(device, device->physical,
                       VK_ERROR_FEATURE_NOT_PRESENT,
                       "cmd parser version %d is required for %s",
                       required_version, function);
   } else {
      return VK_SUCCESS;
   }
}

#endif

static void
anv_cmd_buffer_push_base_group_id(struct anv_cmd_buffer *cmd_buffer,
                                  uint32_t baseGroupX,
                                  uint32_t baseGroupY,
                                  uint32_t baseGroupZ)
{
   if (anv_batch_has_error(&cmd_buffer->batch))
      return;

   struct anv_push_constants *push =
      &cmd_buffer->state.compute.base.push_constants;
   if (push->cs.base_work_group_id[0] != baseGroupX ||
       push->cs.base_work_group_id[1] != baseGroupY ||
       push->cs.base_work_group_id[2] != baseGroupZ) {
      push->cs.base_work_group_id[0] = baseGroupX;
      push->cs.base_work_group_id[1] = baseGroupY;
      push->cs.base_work_group_id[2] = baseGroupZ;

      cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_COMPUTE_BIT;
   }
}

void genX(CmdDispatch)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    x,
    uint32_t                                    y,
    uint32_t                                    z)
{
   genX(CmdDispatchBase)(commandBuffer, 0, 0, 0, x, y, z);
}

static inline void
emit_gpgpu_walker(struct anv_cmd_buffer *cmd_buffer,
                  const struct anv_compute_pipeline *pipeline, bool indirect,
                  const struct brw_cs_prog_data *prog_data,
                  uint32_t groupCountX, uint32_t groupCountY,
                  uint32_t groupCountZ)
{
   bool predicate = (GEN_GEN <= 7 && indirect) ||
      cmd_buffer->state.conditional_render_enabled;
   const struct anv_cs_parameters cs_params = anv_cs_parameters(pipeline);

   anv_batch_emit(&cmd_buffer->batch, GENX(GPGPU_WALKER), ggw) {
      ggw.IndirectParameterEnable      = indirect;
      ggw.PredicateEnable              = predicate;
      ggw.SIMDSize                     = cs_params.simd_size / 16;
      ggw.ThreadDepthCounterMaximum    = 0;
      ggw.ThreadHeightCounterMaximum   = 0;
      ggw.ThreadWidthCounterMaximum    = cs_params.threads - 1;
      ggw.ThreadGroupIDXDimension      = groupCountX;
      ggw.ThreadGroupIDYDimension      = groupCountY;
      ggw.ThreadGroupIDZDimension      = groupCountZ;
      ggw.RightExecutionMask           = pipeline->cs_right_mask;
      ggw.BottomExecutionMask          = 0xffffffff;
   }

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

void genX(CmdDispatchBase)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    baseGroupX,
    uint32_t                                    baseGroupY,
    uint32_t                                    baseGroupZ,
    uint32_t                                    groupCountX,
    uint32_t                                    groupCountY,
    uint32_t                                    groupCountZ)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   struct anv_compute_pipeline *pipeline = cmd_buffer->state.compute.pipeline;
   const struct brw_cs_prog_data *prog_data = get_cs_prog_data(pipeline);

   anv_cmd_buffer_push_base_group_id(cmd_buffer, baseGroupX,
                                     baseGroupY, baseGroupZ);

   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] = groupCountX;
      sizes[1] = groupCountY;
      sizes[2] = groupCountZ;
      cmd_buffer->state.compute.num_workgroups = (struct anv_address) {
         .bo = cmd_buffer->device->dynamic_state_pool.block_pool.bo,
         .offset = state.offset,
      };

      /* The num_workgroups buffer goes in the binding table */
      cmd_buffer->state.descriptors_dirty |= VK_SHADER_STAGE_COMPUTE_BIT;
   }

   genX(cmd_buffer_flush_compute_state)(cmd_buffer);

   if (cmd_buffer->state.conditional_render_enabled)
      genX(cmd_emit_conditional_render_predicate)(cmd_buffer);

   emit_gpgpu_walker(cmd_buffer, pipeline, false, prog_data, groupCountX,
                     groupCountY, groupCountZ);
}

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

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_compute_pipeline *pipeline = cmd_buffer->state.compute.pipeline;
   const struct brw_cs_prog_data *prog_data = get_cs_prog_data(pipeline);
   struct anv_address addr = anv_address_add(buffer->address, offset);
   UNUSED struct anv_batch *batch = &cmd_buffer->batch;

   anv_cmd_buffer_push_base_group_id(cmd_buffer, 0, 0, 0);

#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.compute.num_workgroups = addr;

      /* The num_workgroups buffer goes in the binding table */
      cmd_buffer->state.descriptors_dirty |= VK_SHADER_STAGE_COMPUTE_BIT;
   }

   genX(cmd_buffer_flush_compute_state)(cmd_buffer);

   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   struct gen_mi_value size_x = gen_mi_mem32(anv_address_add(addr, 0));
   struct gen_mi_value size_y = gen_mi_mem32(anv_address_add(addr, 4));
   struct gen_mi_value size_z = gen_mi_mem32(anv_address_add(addr, 8));

   gen_mi_store(&b, gen_mi_reg32(GPGPU_DISPATCHDIMX), size_x);
   gen_mi_store(&b, gen_mi_reg32(GPGPU_DISPATCHDIMY), size_y);
   gen_mi_store(&b, gen_mi_reg32(GPGPU_DISPATCHDIMZ), size_z);

#if GEN_GEN <= 7
   /* predicate = (compute_dispatch_indirect_x_size == 0); */
   gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC0), size_x);
   gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC1), gen_mi_imm(0));
   anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOAD;
      mip.CombineOperation = COMBINE_SET;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }

   /* predicate |= (compute_dispatch_indirect_y_size == 0); */
   gen_mi_store(&b, gen_mi_reg32(MI_PREDICATE_SRC0), size_y);
   anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOAD;
      mip.CombineOperation = COMBINE_OR;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }

   /* predicate |= (compute_dispatch_indirect_z_size == 0); */
   gen_mi_store(&b, gen_mi_reg32(MI_PREDICATE_SRC0), size_z);
   anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOAD;
      mip.CombineOperation = COMBINE_OR;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }

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

#if GEN_IS_HASWELL
   if (cmd_buffer->state.conditional_render_enabled) {
      /* predicate &= !(conditional_rendering_predicate == 0); */
      gen_mi_store(&b, gen_mi_reg32(MI_PREDICATE_SRC0),
                       gen_mi_reg32(ANV_PREDICATE_RESULT_REG));
      anv_batch_emit(batch, GENX(MI_PREDICATE), mip) {
         mip.LoadOperation    = LOAD_LOADINV;
         mip.CombineOperation = COMBINE_AND;
         mip.CompareOperation = COMPARE_SRCS_EQUAL;
      }
   }
#endif

#else /* GEN_GEN > 7 */
   if (cmd_buffer->state.conditional_render_enabled)
      genX(cmd_emit_conditional_render_predicate)(cmd_buffer);
#endif

   emit_gpgpu_walker(cmd_buffer, pipeline, true, prog_data, 0, 0, 0);
}

static void
genX(flush_pipeline_select)(struct anv_cmd_buffer *cmd_buffer,
                            uint32_t pipeline)
{
   UNUSED const struct gen_device_info *devinfo = &cmd_buffer->device->info;

   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

#if GEN_GEN == 9
   if (pipeline == _3D) {
      /* There is a mid-object preemption workaround which requires you to
       * re-emit MEDIA_VFE_STATE after switching from GPGPU to 3D.  However,
       * even without preemption, we have issues with geometry flickering when
       * GPGPU and 3D are back-to-back and this seems to fix it.  We don't
       * really know why.
       */
      const uint32_t subslices =
         MAX2(cmd_buffer->device->physical->subslice_total, 1);
      anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_VFE_STATE), vfe) {
         vfe.MaximumNumberofThreads =
            devinfo->max_cs_threads * subslices - 1;
         vfe.NumberofURBEntries     = 2;
         vfe.URBEntryAllocationSize = 2;
      }

      /* We just emitted a dummy MEDIA_VFE_STATE so now that packet is
       * invalid. Set the compute pipeline to dirty to force a re-emit of the
       * pipeline in case we get back-to-back dispatch calls with the same
       * pipeline and a PIPELINE_SELECT in between.
       */
      cmd_buffer->state.compute.pipeline_dirty = true;
   }
#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;
#if GEN_GEN >= 12
      pc.TileCacheFlushEnable = true;

      /* GEN:BUG:1409600907: "PIPE_CONTROL with Depth Stall Enable bit must be
       * set with any PIPE_CONTROL with Depth Flush Enable bit set.
       */
      pc.DepthStallEnable = true;
#endif
   }

   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;
#if GEN_GEN >= 12
      pc.TileCacheFlushEnable = true;
#endif
   }

   anv_batch_emit(&cmd_buffer->batch, GENX(PIPELINE_SELECT), ps) {
#if GEN_GEN >= 9
      ps.MaskBits = GEN_GEN >= 12 ? 0x13 : 3;
      ps.MediaSamplerDOPClockGateEnable = GEN_GEN >= 12;
#endif
      ps.PipelineSelection = pipeline;
   }

#if GEN_GEN == 9
   if (devinfo->is_geminilake) {
      /* Project: DevGLK
       *
       * "This chicken bit works around a hardware issue with barrier logic
       *  encountered when switching between GPGPU and 3D pipelines.  To
       *  workaround the issue, this mode bit should be set after a pipeline
       *  is selected."
       */
      uint32_t scec;
      anv_pack_struct(&scec, GENX(SLICE_COMMON_ECO_CHICKEN1),
                      .GLKBarrierMode =
                          pipeline == GPGPU ? GLK_BARRIER_MODE_GPGPU
                                            : GLK_BARRIER_MODE_3D_HULL,
                      .GLKBarrierModeMask = 1);
      emit_lri(&cmd_buffer->batch, GENX(SLICE_COMMON_ECO_CHICKEN1_num), scec);
   }
#endif

   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;
#if GEN_GEN >= 12
      pipe.TileCacheFlushEnable = true;
#endif
   }
   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pipe) {
      pipe.DepthStallEnable = true;
   }
}

/* From the Skylake PRM, 3DSTATE_VERTEX_BUFFERS:
 *
 *    "The VF cache needs to be invalidated before binding and then using
 *    Vertex Buffers that overlap with any previously bound Vertex Buffer
 *    (at a 64B granularity) since the last invalidation.  A VF cache
 *    invalidate is performed by setting the "VF Cache Invalidation Enable"
 *    bit in PIPE_CONTROL."
 *
 * This is implemented by carefully tracking all vertex and index buffer
 * bindings and flushing if the cache ever ends up with a range in the cache
 * that would exceed 4 GiB.  This is implemented in three parts:
 *
 *    1. genX(cmd_buffer_set_binding_for_gen8_vb_flush)() which must be called
 *       every time a 3DSTATE_VERTEX_BUFFER packet is emitted and informs the
 *       tracking code of the new binding.  If this new binding would cause
 *       the cache to have a too-large range on the next draw call, a pipeline
 *       stall and VF cache invalidate are added to pending_pipeline_bits.
 *
 *    2. genX(cmd_buffer_apply_pipe_flushes)() resets the cache tracking to
 *       empty whenever we emit a VF invalidate.
 *
 *    3. genX(cmd_buffer_update_dirty_vbs_for_gen8_vb_flush)() must be called
 *       after every 3DPRIMITIVE and copies the bound range into the dirty
 *       range for each used buffer.  This has to be a separate step because
 *       we don't always re-bind all buffers and so 1. can't know which
 *       buffers are actually bound.
 */
void
genX(cmd_buffer_set_binding_for_gen8_vb_flush)(struct anv_cmd_buffer *cmd_buffer,
                                               int vb_index,
                                               struct anv_address vb_address,
                                               uint32_t vb_size)
{
   if (GEN_GEN < 8 || GEN_GEN > 9 ||
       !cmd_buffer->device->physical->use_softpin)
      return;

   struct anv_vb_cache_range *bound, *dirty;
   if (vb_index == -1) {
      bound = &cmd_buffer->state.gfx.ib_bound_range;
      dirty = &cmd_buffer->state.gfx.ib_dirty_range;
   } else {
      assert(vb_index >= 0);
      assert(vb_index < ARRAY_SIZE(cmd_buffer->state.gfx.vb_bound_ranges));
      assert(vb_index < ARRAY_SIZE(cmd_buffer->state.gfx.vb_dirty_ranges));
      bound = &cmd_buffer->state.gfx.vb_bound_ranges[vb_index];
      dirty = &cmd_buffer->state.gfx.vb_dirty_ranges[vb_index];
   }

   if (vb_size == 0) {
      bound->start = 0;
      bound->end = 0;
      return;
   }

   assert(vb_address.bo && (vb_address.bo->flags & EXEC_OBJECT_PINNED));
   bound->start = gen_48b_address(anv_address_physical(vb_address));
   bound->end = bound->start + vb_size;
   assert(bound->end > bound->start); /* No overflow */

   /* Align everything to a cache line */
   bound->start &= ~(64ull - 1ull);
   bound->end = align_u64(bound->end, 64);

   /* Compute the dirty range */
   dirty->start = MIN2(dirty->start, bound->start);
   dirty->end = MAX2(dirty->end, bound->end);

   /* If our range is larger than 32 bits, we have to flush */
   assert(bound->end - bound->start <= (1ull << 32));
   if (dirty->end - dirty->start > (1ull << 32)) {
      cmd_buffer->state.pending_pipe_bits |=
         ANV_PIPE_CS_STALL_BIT | ANV_PIPE_VF_CACHE_INVALIDATE_BIT;
   }
}

void
genX(cmd_buffer_update_dirty_vbs_for_gen8_vb_flush)(struct anv_cmd_buffer *cmd_buffer,
                                                    uint32_t access_type,
                                                    uint64_t vb_used)
{
   if (GEN_GEN < 8 || GEN_GEN > 9 ||
       !cmd_buffer->device->physical->use_softpin)
      return;

   if (access_type == RANDOM) {
      /* We have an index buffer */
      struct anv_vb_cache_range *bound = &cmd_buffer->state.gfx.ib_bound_range;
      struct anv_vb_cache_range *dirty = &cmd_buffer->state.gfx.ib_dirty_range;

      if (bound->end > bound->start) {
         dirty->start = MIN2(dirty->start, bound->start);
         dirty->end = MAX2(dirty->end, bound->end);
      }
   }

   uint64_t mask = vb_used;
   while (mask) {
      int i = u_bit_scan64(&mask);
      assert(i >= 0);
      assert(i < ARRAY_SIZE(cmd_buffer->state.gfx.vb_bound_ranges));
      assert(i < ARRAY_SIZE(cmd_buffer->state.gfx.vb_dirty_ranges));

      struct anv_vb_cache_range *bound, *dirty;
      bound = &cmd_buffer->state.gfx.vb_bound_ranges[i];
      dirty = &cmd_buffer->state.gfx.vb_dirty_ranges[i];

      if (bound->end > bound->start) {
         dirty->start = MIN2(dirty->start, bound->start);
         dirty->end = MAX2(dirty->end, bound->end);
      }
   }
}

/**
 * Update the pixel hashing modes that determine the balancing of PS threads
 * across subslices and slices.
 *
 * \param width Width bound of the rendering area (already scaled down if \p
 *              scale is greater than 1).
 * \param height Height bound of the rendering area (already scaled down if \p
 *               scale is greater than 1).
 * \param scale The number of framebuffer samples that could potentially be
 *              affected by an individual channel of the PS thread.  This is
 *              typically one for single-sampled rendering, but for operations
 *              like CCS resolves and fast clears a single PS invocation may
 *              update a huge number of pixels, in which case a finer
 *              balancing is desirable in order to maximally utilize the
 *              bandwidth available.  UINT_MAX can be used as shorthand for
 *              "finest hashing mode available".
 */
void
genX(cmd_buffer_emit_hashing_mode)(struct anv_cmd_buffer *cmd_buffer,
                                   unsigned width, unsigned height,
                                   unsigned scale)
{
#if GEN_GEN == 9
   const struct gen_device_info *devinfo = &cmd_buffer->device->info;
   const unsigned slice_hashing[] = {
      /* Because all Gen9 platforms with more than one slice require
       * three-way subslice hashing, a single "normal" 16x16 slice hashing
       * block is guaranteed to suffer from substantial imbalance, with one
       * subslice receiving twice as much work as the other two in the
       * slice.
       *
       * The performance impact of that would be particularly severe when
       * three-way hashing is also in use for slice balancing (which is the
       * case for all Gen9 GT4 platforms), because one of the slices
       * receives one every three 16x16 blocks in either direction, which
       * is roughly the periodicity of the underlying subslice imbalance
       * pattern ("roughly" because in reality the hardware's
       * implementation of three-way hashing doesn't do exact modulo 3
       * arithmetic, which somewhat decreases the magnitude of this effect
       * in practice).  This leads to a systematic subslice imbalance
       * within that slice regardless of the size of the primitive.  The
       * 32x32 hashing mode guarantees that the subslice imbalance within a
       * single slice hashing block is minimal, largely eliminating this
       * effect.
       */
      _32x32,
      /* Finest slice hashing mode available. */
      NORMAL
   };
   const unsigned subslice_hashing[] = {
      /* 16x16 would provide a slight cache locality benefit especially
       * visible in the sampler L1 cache efficiency of low-bandwidth
       * non-LLC platforms, but it comes at the cost of greater subslice
       * imbalance for primitives of dimensions approximately intermediate
       * between 16x4 and 16x16.
       */
      _16x4,
      /* Finest subslice hashing mode available. */
      _8x4
   };
   /* Dimensions of the smallest hashing block of a given hashing mode.  If
    * the rendering area is smaller than this there can't possibly be any
    * benefit from switching to this mode, so we optimize out the
    * transition.
    */
   const unsigned min_size[][2] = {
         { 16, 4 },
         { 8, 4 }
   };
   const unsigned idx = scale > 1;

   if (cmd_buffer->state.current_hash_scale != scale &&
       (width > min_size[idx][0] || height > min_size[idx][1])) {
      uint32_t gt_mode;

      anv_pack_struct(&gt_mode, GENX(GT_MODE),
                      .SliceHashing = (devinfo->num_slices > 1 ? slice_hashing[idx] : 0),
                      .SliceHashingMask = (devinfo->num_slices > 1 ? -1 : 0),
                      .SubsliceHashing = subslice_hashing[idx],
                      .SubsliceHashingMask = -1);

      cmd_buffer->state.pending_pipe_bits |=
         ANV_PIPE_CS_STALL_BIT | ANV_PIPE_STALL_AT_SCOREBOARD_BIT;
      genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

      emit_lri(&cmd_buffer->batch, GENX(GT_MODE_num), gt_mode);

      cmd_buffer->state.current_hash_scale = scale;
   }
#endif
}

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 = { };

   if (iview)
      info.view = &iview->planes[0].isl;

   if (image && (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT)) {
      uint32_t depth_plane =
         anv_image_aspect_to_plane(image->aspects, VK_IMAGE_ASPECT_DEPTH_BIT);
      const struct anv_surface *surface = &image->planes[depth_plane].surface;

      info.depth_surf = &surface->isl;

      info.depth_address =
         anv_batch_emit_reloc(&cmd_buffer->batch,
                              dw + device->isl_dev.ds.depth_offset / 4,
                              image->planes[depth_plane].address.bo,
                              image->planes[depth_plane].address.offset +
                              surface->offset);
      info.mocs =
         anv_mocs(device, image->planes[depth_plane].address.bo,
                  ISL_SURF_USAGE_DEPTH_BIT);

      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_NONE) {
         assert(isl_aux_usage_has_hiz(info.hiz_usage));
         info.hiz_surf = &image->planes[depth_plane].aux_surface.isl;

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

         info.depth_clear_value = ANV_HZ_FC_VAL;
      }
   }

   if (image && (image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT)) {
      uint32_t stencil_plane =
         anv_image_aspect_to_plane(image->aspects, VK_IMAGE_ASPECT_STENCIL_BIT);
      const struct anv_surface *surface = &image->planes[stencil_plane].surface;

      info.stencil_surf = &surface->isl;

      info.stencil_aux_usage = image->planes[stencil_plane].aux_usage;
      info.stencil_address =
         anv_batch_emit_reloc(&cmd_buffer->batch,
                              dw + device->isl_dev.ds.stencil_offset / 4,
                              image->planes[stencil_plane].address.bo,
                              image->planes[stencil_plane].address.offset +
                              surface->offset);
      info.mocs =
         anv_mocs(device, image->planes[stencil_plane].address.bo,
                  ISL_SURF_USAGE_STENCIL_BIT);
   }

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

   if (GEN_GEN >= 12) {
      cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_POST_SYNC_BIT;
      genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

      /* GEN:BUG:1408224581
       *
       * Workaround: Gen12LP Astep only An additional pipe control with
       * post-sync = store dword operation would be required.( w/a is to
       * have an additional pipe control after the stencil state whenever
       * the surface state bits of this state is changing).
       */
      anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
         pc.PostSyncOperation = WriteImmediateData;
         pc.Address = cmd_buffer->device->workaround_address;
      }
   }
   cmd_buffer->state.hiz_enabled = isl_aux_usage_has_hiz(info.hiz_usage);
}

/**
 * This ANDs the view mask of the current subpass with the pending clear
 * views in the attachment to get the mask of views active in the subpass
 * that still need to be cleared.
 */
static inline uint32_t
get_multiview_subpass_clear_mask(const struct anv_cmd_state *cmd_state,
                                 const struct anv_attachment_state *att_state)
{
   return cmd_state->subpass->view_mask & att_state->pending_clear_views;
}

static inline bool
do_first_layer_clear(const struct anv_cmd_state *cmd_state,
                     const struct anv_attachment_state *att_state)
{
   if (!cmd_state->subpass->view_mask)
      return true;

   uint32_t pending_clear_mask =
      get_multiview_subpass_clear_mask(cmd_state, att_state);

   return pending_clear_mask & 1;
}

static inline bool
current_subpass_is_last_for_attachment(const struct anv_cmd_state *cmd_state,
                                       uint32_t att_idx)
{
   const uint32_t last_subpass_idx =
      cmd_state->pass->attachments[att_idx].last_subpass_idx;
   const struct anv_subpass *last_subpass =
      &cmd_state->pass->subpasses[last_subpass_idx];
   return last_subpass == cmd_state->subpass;
}

static void
cmd_buffer_begin_subpass(struct anv_cmd_buffer *cmd_buffer,
                         uint32_t subpass_id)
{
   struct anv_cmd_state *cmd_state = &cmd_buffer->state;
   struct anv_render_pass *pass = cmd_state->pass;
   struct anv_subpass *subpass = &pass->subpasses[subpass_id];
   cmd_state->subpass = subpass;

   cmd_buffer->state.gfx.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.gfx.vb_dirty |= ~0;

   /* It is possible to start a render pass with an old pipeline.  Because the
    * render pass and subpass index are both baked into the pipeline, this is
    * highly unlikely.  In order to do so, it requires that you have a render
    * pass with a single subpass and that you use that render pass twice
    * back-to-back and use the same pipeline at the start of the second render
    * pass as at the end of the first.  In order to avoid unpredictable issues
    * with this edge case, we just dirty the pipeline at the start of every
    * subpass.
    */
   cmd_buffer->state.gfx.dirty |= ANV_CMD_DIRTY_PIPELINE;

   /* Accumulate any subpass flushes that need to happen before the subpass */
   cmd_buffer->state.pending_pipe_bits |=
      cmd_buffer->state.pass->subpass_flushes[subpass_id];

   VkRect2D render_area = cmd_buffer->state.render_area;
   struct anv_framebuffer *fb = cmd_buffer->state.framebuffer;

   bool is_multiview = subpass->view_mask != 0;

   for (uint32_t i = 0; i < subpass->attachment_count; ++i) {
      const uint32_t a = subpass->attachments[i].attachment;
      if (a == VK_ATTACHMENT_UNUSED)
         continue;

      assert(a < cmd_state->pass->attachment_count);
      struct anv_attachment_state *att_state = &cmd_state->attachments[a];

      struct anv_image_view *iview = cmd_state->attachments[a].image_view;
      const struct anv_image *image = iview->image;

      VkImageLayout target_layout = subpass->attachments[i].layout;
      VkImageLayout target_stencil_layout =
         subpass->attachments[i].stencil_layout;

      uint32_t level = iview->planes[0].isl.base_level;
      uint32_t width = anv_minify(iview->image->extent.width, level);
      uint32_t height = anv_minify(iview->image->extent.height, level);
      bool full_surface_draw =
         render_area.offset.x == 0 && render_area.offset.y == 0 &&
         render_area.extent.width == width &&
         render_area.extent.height == height;

      uint32_t base_layer, layer_count;
      if (image->type == VK_IMAGE_TYPE_3D) {
         base_layer = 0;
         layer_count = anv_minify(iview->image->extent.depth, level);
      } else {
         base_layer = iview->planes[0].isl.base_array_layer;
         layer_count = fb->layers;
      }

      if (image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) {
         bool will_full_fast_clear =
            (att_state->pending_clear_aspects & VK_IMAGE_ASPECT_COLOR_BIT) &&
            att_state->fast_clear && full_surface_draw;

         assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
         transition_color_buffer(cmd_buffer, image, VK_IMAGE_ASPECT_COLOR_BIT,
                                 level, 1, base_layer, layer_count,
                                 att_state->current_layout, target_layout,
                                 will_full_fast_clear);
         att_state->aux_usage =
            anv_layout_to_aux_usage(&cmd_buffer->device->info, image,
                                    VK_IMAGE_ASPECT_COLOR_BIT,
                                    VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
                                    target_layout);
      }

      if (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT) {
         bool will_full_fast_clear =
            (att_state->pending_clear_aspects & VK_IMAGE_ASPECT_DEPTH_BIT) &&
            att_state->fast_clear && full_surface_draw;

         transition_depth_buffer(cmd_buffer, image,
                                 base_layer, layer_count,
                                 att_state->current_layout, target_layout,
                                 will_full_fast_clear);
         att_state->aux_usage =
            anv_layout_to_aux_usage(&cmd_buffer->device->info, image,
                                    VK_IMAGE_ASPECT_DEPTH_BIT,
                                    VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT,
                                    target_layout);
      }

      if (image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
         bool will_full_fast_clear =
            (att_state->pending_clear_aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
            att_state->fast_clear && full_surface_draw;

         transition_stencil_buffer(cmd_buffer, image,
                                   level, 1, base_layer, layer_count,
                                   att_state->current_stencil_layout,
                                   target_stencil_layout,
                                   will_full_fast_clear);
      }
      att_state->current_layout = target_layout;
      att_state->current_stencil_layout = target_stencil_layout;

      if (att_state->pending_clear_aspects & VK_IMAGE_ASPECT_COLOR_BIT) {
         assert(att_state->pending_clear_aspects == VK_IMAGE_ASPECT_COLOR_BIT);

         /* Multi-planar images are not supported as attachments */
         assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
         assert(image->n_planes == 1);

         uint32_t base_clear_layer = iview->planes[0].isl.base_array_layer;
         uint32_t clear_layer_count = fb->layers;

         if (att_state->fast_clear &&
             do_first_layer_clear(cmd_state, att_state)) {
            /* We only support fast-clears on the first layer */
            assert(level == 0 && base_layer == 0);

            union isl_color_value clear_color = {};
            anv_clear_color_from_att_state(&clear_color, att_state, iview);
            if (iview->image->samples == 1) {
               anv_image_ccs_op(cmd_buffer, image,
                                iview->planes[0].isl.format,
                                iview->planes[0].isl.swizzle,
                                VK_IMAGE_ASPECT_COLOR_BIT,
                                0, 0, 1, ISL_AUX_OP_FAST_CLEAR,
                                &clear_color,
                                false);
            } else {
               anv_image_mcs_op(cmd_buffer, image,
                                iview->planes[0].isl.format,
                                iview->planes[0].isl.swizzle,
                                VK_IMAGE_ASPECT_COLOR_BIT,
                                0, 1, ISL_AUX_OP_FAST_CLEAR,
                                &clear_color,
                                false);
            }
            base_clear_layer++;
            clear_layer_count--;
            if (is_multiview)
               att_state->pending_clear_views &= ~1;

            if (isl_color_value_is_zero(clear_color,
                                        iview->planes[0].isl.format)) {
               /* This image has the auxiliary buffer enabled. We can mark the
                * subresource as not needing a resolve because the clear color
                * will match what's in every RENDER_SURFACE_STATE object when
                * it's being used for sampling.
                */
               set_image_fast_clear_state(cmd_buffer, iview->image,
                                          VK_IMAGE_ASPECT_COLOR_BIT,
                                          ANV_FAST_CLEAR_DEFAULT_VALUE);
            } else {
               set_image_fast_clear_state(cmd_buffer, iview->image,
                                          VK_IMAGE_ASPECT_COLOR_BIT,
                                          ANV_FAST_CLEAR_ANY);
            }
         }

         /* From the VkFramebufferCreateInfo spec:
          *
          * "If the render pass uses multiview, then layers must be one and each
          *  attachment requires a number of layers that is greater than the
          *  maximum bit index set in the view mask in the subpasses in which it
          *  is used."
          *
          * So if multiview is active we ignore the number of layers in the
          * framebuffer and instead we honor the view mask from the subpass.
          */
         if (is_multiview) {
            assert(image->n_planes == 1);
            uint32_t pending_clear_mask =
               get_multiview_subpass_clear_mask(cmd_state, att_state);

            uint32_t layer_idx;
            for_each_bit(layer_idx, pending_clear_mask) {
               uint32_t layer =
                  iview->planes[0].isl.base_array_layer + layer_idx;

               anv_image_clear_color(cmd_buffer, image,
                                     VK_IMAGE_ASPECT_COLOR_BIT,
                                     att_state->aux_usage,
                                     iview->planes[0].isl.format,
                                     iview->planes[0].isl.swizzle,
                                     level, layer, 1,
                                     render_area,
                                     vk_to_isl_color(att_state->clear_value.color));
            }

            att_state->pending_clear_views &= ~pending_clear_mask;
         } else if (clear_layer_count > 0) {
            assert(image->n_planes == 1);
            anv_image_clear_color(cmd_buffer, image, VK_IMAGE_ASPECT_COLOR_BIT,
                                  att_state->aux_usage,
                                  iview->planes[0].isl.format,
                                  iview->planes[0].isl.swizzle,
                                  level, base_clear_layer, clear_layer_count,
                                  render_area,
                                  vk_to_isl_color(att_state->clear_value.color));
         }
      } else if (att_state->pending_clear_aspects & (VK_IMAGE_ASPECT_DEPTH_BIT |
                                                     VK_IMAGE_ASPECT_STENCIL_BIT)) {
         if (att_state->fast_clear &&
             (att_state->pending_clear_aspects & VK_IMAGE_ASPECT_DEPTH_BIT)) {
            /* We currently only support HiZ for single-LOD images */
            assert(isl_aux_usage_has_hiz(iview->image->planes[0].aux_usage));
            assert(iview->planes[0].isl.base_level == 0);
            assert(iview->planes[0].isl.levels == 1);
         }

         if (is_multiview) {
            uint32_t pending_clear_mask =
              get_multiview_subpass_clear_mask(cmd_state, att_state);

            uint32_t layer_idx;
            for_each_bit(layer_idx, pending_clear_mask) {
               uint32_t layer =
                  iview->planes[0].isl.base_array_layer + layer_idx;

               if (att_state->fast_clear) {
                  anv_image_hiz_clear(cmd_buffer, image,
                                      att_state->pending_clear_aspects,
                                      level, layer, 1, render_area,
                                      att_state->clear_value.depthStencil.stencil);
               } else {
                  anv_image_clear_depth_stencil(cmd_buffer, image,
                                                att_state->pending_clear_aspects,
                                                att_state->aux_usage,
                                                level, layer, 1, render_area,
                                                att_state->clear_value.depthStencil.depth,
                                                att_state->clear_value.depthStencil.stencil);
               }
            }

            att_state->pending_clear_views &= ~pending_clear_mask;
         } else {
            if (att_state->fast_clear) {
               anv_image_hiz_clear(cmd_buffer, image,
                                   att_state->pending_clear_aspects,
                                   level, base_layer, layer_count,
                                   render_area,
                                   att_state->clear_value.depthStencil.stencil);
            } else {
               anv_image_clear_depth_stencil(cmd_buffer, image,
                                             att_state->pending_clear_aspects,
                                             att_state->aux_usage,
                                             level, base_layer, layer_count,
                                             render_area,
                                             att_state->clear_value.depthStencil.depth,
                                             att_state->clear_value.depthStencil.stencil);
            }
         }
      } else  {
         assert(att_state->pending_clear_aspects == 0);
      }

      /* If multiview is enabled, then we are only done clearing when we no
       * longer have pending layers to clear, or when we have processed the
       * last subpass that uses this attachment.
       */
      if (!is_multiview ||
          att_state->pending_clear_views == 0 ||
          current_subpass_is_last_for_attachment(cmd_state, a)) {
         att_state->pending_clear_aspects = 0;
      }

      att_state->pending_load_aspects = 0;
   }

   /* We've transitioned all our images possibly fast clearing them.  Now we
    * can fill out the surface states that we will use as render targets
    * during actual subpass rendering.
    */
   VkResult result = genX(cmd_buffer_alloc_att_surf_states)(cmd_buffer,
                                                            pass, subpass);
   if (result != VK_SUCCESS)
      return;

   isl_null_fill_state(&cmd_buffer->device->isl_dev,
                       cmd_state->null_surface_state.map,
                       isl_extent3d(fb->width, fb->height, fb->layers));

   for (uint32_t i = 0; i < subpass->attachment_count; ++i) {
      const uint32_t att = subpass->attachments[i].attachment;
      if (att == VK_ATTACHMENT_UNUSED)
         continue;

      assert(att < cmd_state->pass->attachment_count);
      struct anv_render_pass_attachment *pass_att = &pass->attachments[att];
      struct anv_attachment_state *att_state = &cmd_state->attachments[att];
      struct anv_image_view *iview = att_state->image_view;

      if (!vk_format_is_color(pass_att->format))
         continue;

      const VkImageUsageFlagBits att_usage = subpass->attachments[i].usage;
      assert(util_bitcount(att_usage) == 1);

      struct anv_surface_state *surface_state;
      isl_surf_usage_flags_t isl_surf_usage;
      enum isl_aux_usage isl_aux_usage;
      if (att_usage == VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT) {
         surface_state = &att_state->color;
         isl_surf_usage = ISL_SURF_USAGE_RENDER_TARGET_BIT;
         isl_aux_usage = att_state->aux_usage;
      } else if (att_usage == VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT) {
         surface_state = &att_state->input;
         isl_surf_usage = ISL_SURF_USAGE_TEXTURE_BIT;
         isl_aux_usage =
            anv_layout_to_aux_usage(&cmd_buffer->device->info, iview->image,
                                    VK_IMAGE_ASPECT_COLOR_BIT,
                                    VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT,
                                    att_state->current_layout);
      } else {
         continue;
      }

      /* We had better have a surface state when we get here */
      assert(surface_state->state.map);

      union isl_color_value clear_color = { .u32 = { 0, } };
      if (pass_att->load_op == VK_ATTACHMENT_LOAD_OP_CLEAR &&
          att_state->fast_clear)
         anv_clear_color_from_att_state(&clear_color, att_state, iview);

      anv_image_fill_surface_state(cmd_buffer->device,
                                   iview->image,
                                   VK_IMAGE_ASPECT_COLOR_BIT,
                                   &iview->planes[0].isl,
                                   isl_surf_usage,
                                   isl_aux_usage,
                                   &clear_color,
                                   0,
                                   surface_state,
                                   NULL);

      add_surface_state_relocs(cmd_buffer, *surface_state);

      if (GEN_GEN < 10 &&
          pass_att->load_op == VK_ATTACHMENT_LOAD_OP_LOAD &&
          iview->image->planes[0].aux_usage != ISL_AUX_USAGE_NONE &&
          iview->planes[0].isl.base_level == 0 &&
          iview->planes[0].isl.base_array_layer == 0) {
         genX(copy_fast_clear_dwords)(cmd_buffer, surface_state->state,
                                      iview->image,
                                      VK_IMAGE_ASPECT_COLOR_BIT,
                                      false /* copy to ss */);
      }
   }

#if GEN_GEN >= 11
   /* The PIPE_CONTROL command description says:
    *
    *    "Whenever a Binding Table Index (BTI) used by a Render Taget Message
    *     points to a different RENDER_SURFACE_STATE, SW must issue a Render
    *     Target Cache Flush by enabling this bit. When render target flush
    *     is set due to new association of BTI, PS Scoreboard Stall bit must
    *     be set in this packet."
    */
   cmd_buffer->state.pending_pipe_bits |=
      ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT |
      ANV_PIPE_STALL_AT_SCOREBOARD_BIT;
#endif

#if GEN_GEN == 12
   /* GEN:BUG:14010455700
    *
    * ISL will change some CHICKEN registers depending on the depth surface
    * format, along with emitting the depth and stencil packets. In that case,
    * we want to do a depth flush and stall, so the pipeline is not using these
    * settings while we change the registers.
    */
   cmd_buffer->state.pending_pipe_bits |=
      ANV_PIPE_DEPTH_CACHE_FLUSH_BIT |
      ANV_PIPE_DEPTH_STALL_BIT |
      ANV_PIPE_END_OF_PIPE_SYNC_BIT;
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
#endif

   cmd_buffer_emit_depth_stencil(cmd_buffer);
}

static enum blorp_filter
vk_to_blorp_resolve_mode(VkResolveModeFlagBitsKHR vk_mode)
{
   switch (vk_mode) {
   case VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR:
      return BLORP_FILTER_SAMPLE_0;
   case VK_RESOLVE_MODE_AVERAGE_BIT_KHR:
      return BLORP_FILTER_AVERAGE;
   case VK_RESOLVE_MODE_MIN_BIT_KHR:
      return BLORP_FILTER_MIN_SAMPLE;
   case VK_RESOLVE_MODE_MAX_BIT_KHR:
      return BLORP_FILTER_MAX_SAMPLE;
   default:
      return BLORP_FILTER_NONE;
   }
}

static void
cmd_buffer_end_subpass(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_cmd_state *cmd_state = &cmd_buffer->state;
   struct anv_subpass *subpass = cmd_state->subpass;
   uint32_t subpass_id = anv_get_subpass_id(&cmd_buffer->state);
   struct anv_framebuffer *fb = cmd_buffer->state.framebuffer;

   /* We are done with the previous subpass and all rendering directly to that
    * subpass is now complete.  Zero out all the surface states so we don't
    * accidentally use them between now and the next subpass.
    */
   for (uint32_t i = 0; i < cmd_state->pass->attachment_count; ++i) {
      memset(&cmd_state->attachments[i].color, 0,
             sizeof(cmd_state->attachments[i].color));
      memset(&cmd_state->attachments[i].input, 0,
             sizeof(cmd_state->attachments[i].input));
   }
   cmd_state->null_surface_state = ANV_STATE_NULL;
   cmd_state->attachment_states = ANV_STATE_NULL;

   for (uint32_t i = 0; i < subpass->attachment_count; ++i) {
      const uint32_t a = subpass->attachments[i].attachment;
      if (a == VK_ATTACHMENT_UNUSED)
         continue;

      assert(a < cmd_state->pass->attachment_count);
      struct anv_attachment_state *att_state = &cmd_state->attachments[a];
      struct anv_image_view *iview = att_state->image_view;

      assert(util_bitcount(subpass->attachments[i].usage) == 1);
      if (subpass->attachments[i].usage ==
          VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT) {
         /* We assume that if we're ending a subpass, we did do some rendering
          * so we may end up with compressed data.
          */
         genX(cmd_buffer_mark_image_written)(cmd_buffer, iview->image,
                                             VK_IMAGE_ASPECT_COLOR_BIT,
                                             att_state->aux_usage,
                                             iview->planes[0].isl.base_level,
                                             iview->planes[0].isl.base_array_layer,
                                             fb->layers);
      } else if (subpass->attachments[i].usage ==
                 VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) {
         /* We may be writing depth or stencil so we need to mark the surface.
          * Unfortunately, there's no way to know at this point whether the
          * depth or stencil tests used will actually write to the surface.
          *
          * Even though stencil may be plane 1, it always shares a base_level
          * with depth.
          */
         const struct isl_view *ds_view = &iview->planes[0].isl;
         if (iview->aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) {
            genX(cmd_buffer_mark_image_written)(cmd_buffer, iview->image,
                                                VK_IMAGE_ASPECT_DEPTH_BIT,
                                                att_state->aux_usage,
                                                ds_view->base_level,
                                                ds_view->base_array_layer,
                                                fb->layers);
         }
         if (iview->aspect_mask & VK_IMAGE_ASPECT_STENCIL_BIT) {
            /* Even though stencil may be plane 1, it always shares a
             * base_level with depth.
             */
            genX(cmd_buffer_mark_image_written)(cmd_buffer, iview->image,
                                                VK_IMAGE_ASPECT_STENCIL_BIT,
                                                ISL_AUX_USAGE_NONE,
                                                ds_view->base_level,
                                                ds_view->base_array_layer,
                                                fb->layers);
         }
      }
   }

   if (subpass->has_color_resolve) {
      /* We are about to do some MSAA resolves.  We need to flush so that the
       * result of writes to the MSAA color attachments show up in the sampler
       * when we blit to the single-sampled resolve target.
       */
      cmd_buffer->state.pending_pipe_bits |=
         ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT |
         ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT;

      for (uint32_t i = 0; i < subpass->color_count; ++i) {
         uint32_t src_att = subpass->color_attachments[i].attachment;
         uint32_t dst_att = subpass->resolve_attachments[i].attachment;

         if (dst_att == VK_ATTACHMENT_UNUSED)
            continue;

         assert(src_att < cmd_buffer->state.pass->attachment_count);
         assert(dst_att < cmd_buffer->state.pass->attachment_count);

         if (cmd_buffer->state.attachments[dst_att].pending_clear_aspects) {
            /* From the Vulkan 1.0 spec:
             *
             *    If the first use of an attachment in a render pass is as a
             *    resolve attachment, then the loadOp is effectively ignored
             *    as the resolve is guaranteed to overwrite all pixels in the
             *    render area.
             */
            cmd_buffer->state.attachments[dst_att].pending_clear_aspects = 0;
         }

         struct anv_image_view *src_iview = cmd_state->attachments[src_att].image_view;
         struct anv_image_view *dst_iview = cmd_state->attachments[dst_att].image_view;

         const VkRect2D render_area = cmd_buffer->state.render_area;

         enum isl_aux_usage src_aux_usage =
            cmd_buffer->state.attachments[src_att].aux_usage;
         enum isl_aux_usage dst_aux_usage =
            cmd_buffer->state.attachments[dst_att].aux_usage;

         assert(src_iview->aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT &&
                dst_iview->aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT);

         anv_image_msaa_resolve(cmd_buffer,
                                src_iview->image, src_aux_usage,
                                src_iview->planes[0].isl.base_level,
                                src_iview->planes[0].isl.base_array_layer,
                                dst_iview->image, dst_aux_usage,
                                dst_iview->planes[0].isl.base_level,
                                dst_iview->planes[0].isl.base_array_layer,
                                VK_IMAGE_ASPECT_COLOR_BIT,
                                render_area.offset.x, render_area.offset.y,
                                render_area.offset.x, render_area.offset.y,
                                render_area.extent.width,
                                render_area.extent.height,
                                fb->layers, BLORP_FILTER_NONE);
      }
   }

   if (subpass->ds_resolve_attachment) {
      /* We are about to do some MSAA resolves.  We need to flush so that the
       * result of writes to the MSAA depth attachments show up in the sampler
       * when we blit to the single-sampled resolve target.
       */
      cmd_buffer->state.pending_pipe_bits |=
         ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT |
         ANV_PIPE_DEPTH_CACHE_FLUSH_BIT;

      uint32_t src_att = subpass->depth_stencil_attachment->attachment;
      uint32_t dst_att = subpass->ds_resolve_attachment->attachment;

      assert(src_att < cmd_buffer->state.pass->attachment_count);
      assert(dst_att < cmd_buffer->state.pass->attachment_count);

      if (cmd_buffer->state.attachments[dst_att].pending_clear_aspects) {
         /* From the Vulkan 1.0 spec:
          *
          *    If the first use of an attachment in a render pass is as a
          *    resolve attachment, then the loadOp is effectively ignored
          *    as the resolve is guaranteed to overwrite all pixels in the
          *    render area.
          */
         cmd_buffer->state.attachments[dst_att].pending_clear_aspects = 0;
      }

      struct anv_image_view *src_iview = cmd_state->attachments[src_att].image_view;
      struct anv_image_view *dst_iview = cmd_state->attachments[dst_att].image_view;

      const VkRect2D render_area = cmd_buffer->state.render_area;

      struct anv_attachment_state *src_state =
         &cmd_state->attachments[src_att];
      struct anv_attachment_state *dst_state =
         &cmd_state->attachments[dst_att];

      if ((src_iview->image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT) &&
          subpass->depth_resolve_mode != VK_RESOLVE_MODE_NONE_KHR) {

         /* MSAA resolves sample from the source attachment.  Transition the
          * depth attachment first to get rid of any HiZ that we may not be
          * able to handle.
          */
         transition_depth_buffer(cmd_buffer, src_iview->image,
                                 src_iview->planes[0].isl.base_array_layer,
                                 fb->layers,
                                 src_state->current_layout,
                                 VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
                                 false /* will_full_fast_clear */);
         src_state->aux_usage =
            anv_layout_to_aux_usage(&cmd_buffer->device->info, src_iview->image,
                                    VK_IMAGE_ASPECT_DEPTH_BIT,
                                    VK_IMAGE_USAGE_TRANSFER_SRC_BIT,
                                    VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
         src_state->current_layout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;

         /* MSAA resolves write to the resolve attachment as if it were any
          * other transfer op.  Transition the resolve attachment accordingly.
          */
         VkImageLayout dst_initial_layout = dst_state->current_layout;

         /* If our render area is the entire size of the image, we're going to
          * blow it all away so we can claim the initial layout is UNDEFINED
          * and we'll get a HiZ ambiguate instead of a resolve.
          */
         if (dst_iview->image->type != VK_IMAGE_TYPE_3D &&
             render_area.offset.x == 0 && render_area.offset.y == 0 &&
             render_area.extent.width == dst_iview->extent.width &&
             render_area.extent.height == dst_iview->extent.height)
            dst_initial_layout = VK_IMAGE_LAYOUT_UNDEFINED;

         transition_depth_buffer(cmd_buffer, dst_iview->image,
                                 dst_iview->planes[0].isl.base_array_layer,
                                 fb->layers,
                                 dst_initial_layout,
                                 VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
                                 false /* will_full_fast_clear */);
         dst_state->aux_usage =
            anv_layout_to_aux_usage(&cmd_buffer->device->info, dst_iview->image,
                                    VK_IMAGE_ASPECT_DEPTH_BIT,
                                    VK_IMAGE_USAGE_TRANSFER_DST_BIT,
                                    VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
         dst_state->current_layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;

         enum blorp_filter filter =
            vk_to_blorp_resolve_mode(subpass->depth_resolve_mode);

         anv_image_msaa_resolve(cmd_buffer,
                                src_iview->image, src_state->aux_usage,
                                src_iview->planes[0].isl.base_level,
                                src_iview->planes[0].isl.base_array_layer,
                                dst_iview->image, dst_state->aux_usage,
                                dst_iview->planes[0].isl.base_level,
                                dst_iview->planes[0].isl.base_array_layer,
                                VK_IMAGE_ASPECT_DEPTH_BIT,
                                render_area.offset.x, render_area.offset.y,
                                render_area.offset.x, render_area.offset.y,
                                render_area.extent.width,
                                render_area.extent.height,
                                fb->layers, filter);
      }

      if ((src_iview->image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
          subpass->stencil_resolve_mode != VK_RESOLVE_MODE_NONE_KHR) {

         src_state->current_stencil_layout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
         dst_state->current_stencil_layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;

         enum isl_aux_usage src_aux_usage = ISL_AUX_USAGE_NONE;
         uint32_t plane = anv_image_aspect_to_plane(dst_iview->image->aspects,
                                                    VK_IMAGE_ASPECT_STENCIL_BIT);
         enum isl_aux_usage dst_aux_usage =
            dst_iview->image->planes[plane].aux_usage;

         enum blorp_filter filter =
            vk_to_blorp_resolve_mode(subpass->stencil_resolve_mode);

         anv_image_msaa_resolve(cmd_buffer,
                                src_iview->image, src_aux_usage,
                                src_iview->planes[0].isl.base_level,
                                src_iview->planes[0].isl.base_array_layer,
                                dst_iview->image, dst_aux_usage,
                                dst_iview->planes[0].isl.base_level,
                                dst_iview->planes[0].isl.base_array_layer,
                                VK_IMAGE_ASPECT_STENCIL_BIT,
                                render_area.offset.x, render_area.offset.y,
                                render_area.offset.x, render_area.offset.y,
                                render_area.extent.width,
                                render_area.extent.height,
                                fb->layers, filter);
      }
   }

#if GEN_GEN == 7
   /* On gen7, we have to store a texturable version of the stencil buffer in
    * a shadow whenever VK_IMAGE_USAGE_SAMPLED_BIT is set and copy back and
    * forth at strategic points. Stencil writes are only allowed in following
    * layouts:
    *
    *  - VK_IMAGE_LAYOUT_GENERAL
    *  - VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
    *  - VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
    *  - VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL
    *  - VK_IMAGE_LAYOUT_STENCIL_ATTACHMENT_OPTIMAL_KHR
    *
    * For general, we have no nice opportunity to transition so we do the copy
    * to the shadow unconditionally at the end of the subpass. For transfer
    * destinations, we can update it as part of the transfer op. For the other
    * layouts, we delay the copy until a transition into some other layout.
    */
   if (subpass->depth_stencil_attachment) {
      uint32_t a = subpass->depth_stencil_attachment->attachment;
      assert(a != VK_ATTACHMENT_UNUSED);

      struct anv_attachment_state *att_state = &cmd_state->attachments[a];
      struct anv_image_view *iview = cmd_state->attachments[a].image_view;;
      const struct anv_image *image = iview->image;

      if (image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
         uint32_t plane = anv_image_aspect_to_plane(image->aspects,
                                                    VK_IMAGE_ASPECT_STENCIL_BIT);

         if (image->planes[plane].shadow_surface.isl.size_B > 0 &&
             att_state->current_stencil_layout == VK_IMAGE_LAYOUT_GENERAL) {
            assert(image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT);
            anv_image_copy_to_shadow(cmd_buffer, image,
                                     VK_IMAGE_ASPECT_STENCIL_BIT,
                                     iview->planes[plane].isl.base_level, 1,
                                     iview->planes[plane].isl.base_array_layer,
                                     fb->layers);
         }
      }
   }
#endif /* GEN_GEN == 7 */

   for (uint32_t i = 0; i < subpass->attachment_count; ++i) {
      const uint32_t a = subpass->attachments[i].attachment;
      if (a == VK_ATTACHMENT_UNUSED)
         continue;

      if (cmd_state->pass->attachments[a].last_subpass_idx != subpass_id)
         continue;

      assert(a < cmd_state->pass->attachment_count);
      struct anv_attachment_state *att_state = &cmd_state->attachments[a];
      struct anv_image_view *iview = cmd_state->attachments[a].image_view;
      const struct anv_image *image = iview->image;

      /* Transition the image into the final layout for this render pass */
      VkImageLayout target_layout =
         cmd_state->pass->attachments[a].final_layout;
      VkImageLayout target_stencil_layout =
         cmd_state->pass->attachments[a].stencil_final_layout;

      uint32_t base_layer, layer_count;
      if (image->type == VK_IMAGE_TYPE_3D) {
         base_layer = 0;
         layer_count = anv_minify(iview->image->extent.depth,
                                  iview->planes[0].isl.base_level);
      } else {
         base_layer = iview->planes[0].isl.base_array_layer;
         layer_count = fb->layers;
      }

      if (image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) {
         assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
         transition_color_buffer(cmd_buffer, image, VK_IMAGE_ASPECT_COLOR_BIT,
                                 iview->planes[0].isl.base_level, 1,
                                 base_layer, layer_count,
                                 att_state->current_layout, target_layout,
                                 false /* will_full_fast_clear */);
      }

      if (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT) {
         transition_depth_buffer(cmd_buffer, image,
                                 base_layer, layer_count,
                                 att_state->current_layout, target_layout,
                                 false /* will_full_fast_clear */);
      }

      if (image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
         transition_stencil_buffer(cmd_buffer, image,
                                   iview->planes[0].isl.base_level, 1,
                                   base_layer, layer_count,
                                   att_state->current_stencil_layout,
                                   target_stencil_layout,
                                   false /* will_full_fast_clear */);
      }
   }

   /* Accumulate any subpass flushes that need to happen after the subpass.
    * Yes, they do get accumulated twice in the NextSubpass case but since
    * genX_CmdNextSubpass just calls end/begin back-to-back, we just end up
    * ORing the bits in twice so it's harmless.
    */
   cmd_buffer->state.pending_pipe_bits |=
      cmd_buffer->state.pass->subpass_flushes[subpass_id + 1];
}

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);
   VkResult result;

   cmd_buffer->state.framebuffer = framebuffer;
   cmd_buffer->state.pass = pass;
   cmd_buffer->state.render_area = pRenderPassBegin->renderArea;

   result = genX(cmd_buffer_setup_attachments)(cmd_buffer, pass,
                                               framebuffer,
                                               pRenderPassBegin);
   if (result != VK_SUCCESS) {
      assert(anv_batch_has_error(&cmd_buffer->batch));
      return;
   }

   genX(flush_pipeline_select_3d)(cmd_buffer);

   cmd_buffer_begin_subpass(cmd_buffer, 0);
}

void genX(CmdBeginRenderPass2)(
    VkCommandBuffer                             commandBuffer,
    const VkRenderPassBeginInfo*                pRenderPassBeginInfo,
    const VkSubpassBeginInfoKHR*                pSubpassBeginInfo)
{
   genX(CmdBeginRenderPass)(commandBuffer, pRenderPassBeginInfo,
                            pSubpassBeginInfo->contents);
}

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);

   uint32_t prev_subpass = anv_get_subpass_id(&cmd_buffer->state);
   cmd_buffer_end_subpass(cmd_buffer);
   cmd_buffer_begin_subpass(cmd_buffer, prev_subpass + 1);
}

void genX(CmdNextSubpass2)(
    VkCommandBuffer                             commandBuffer,
    const VkSubpassBeginInfoKHR*                pSubpassBeginInfo,
    const VkSubpassEndInfoKHR*                  pSubpassEndInfo)
{
   genX(CmdNextSubpass)(commandBuffer, pSubpassBeginInfo->contents);
}

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

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

   cmd_buffer_end_subpass(cmd_buffer);

   cmd_buffer->state.hiz_enabled = false;

#ifndef NDEBUG
   anv_dump_add_attachments(cmd_buffer);
#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;
}

void genX(CmdEndRenderPass2)(
    VkCommandBuffer                             commandBuffer,
    const VkSubpassEndInfoKHR*                  pSubpassEndInfo)
{
   genX(CmdEndRenderPass)(commandBuffer);
}

void
genX(cmd_emit_conditional_render_predicate)(struct anv_cmd_buffer *cmd_buffer)
{
#if GEN_GEN >= 8 || GEN_IS_HASWELL
   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC0),
                    gen_mi_reg32(ANV_PREDICATE_RESULT_REG));
   gen_mi_store(&b, gen_mi_reg64(MI_PREDICATE_SRC1), gen_mi_imm(0));

   anv_batch_emit(&cmd_buffer->batch, GENX(MI_PREDICATE), mip) {
      mip.LoadOperation    = LOAD_LOADINV;
      mip.CombineOperation = COMBINE_SET;
      mip.CompareOperation = COMPARE_SRCS_EQUAL;
   }
#endif
}

#if GEN_GEN >= 8 || GEN_IS_HASWELL
void genX(CmdBeginConditionalRenderingEXT)(
   VkCommandBuffer                             commandBuffer,
   const VkConditionalRenderingBeginInfoEXT*   pConditionalRenderingBegin)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, pConditionalRenderingBegin->buffer);
   struct anv_cmd_state *cmd_state = &cmd_buffer->state;
   struct anv_address value_address =
      anv_address_add(buffer->address, pConditionalRenderingBegin->offset);

   const bool isInverted = pConditionalRenderingBegin->flags &
                           VK_CONDITIONAL_RENDERING_INVERTED_BIT_EXT;

   cmd_state->conditional_render_enabled = true;

   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   struct gen_mi_builder b;
   gen_mi_builder_init(&b, &cmd_buffer->batch);

   /* Section 19.4 of the Vulkan 1.1.85 spec says:
    *
    *    If the value of the predicate in buffer memory changes
    *    while conditional rendering is active, the rendering commands
    *    may be discarded in an implementation-dependent way.
    *    Some implementations may latch the value of the predicate
    *    upon beginning conditional rendering while others
    *    may read it before every rendering command.
    *
    * So it's perfectly fine to read a value from the buffer once.
    */
   struct gen_mi_value value =  gen_mi_mem32(value_address);

   /* Precompute predicate result, it is necessary to support secondary
    * command buffers since it is unknown if conditional rendering is
    * inverted when populating them.
    */
   gen_mi_store(&b, gen_mi_reg64(ANV_PREDICATE_RESULT_REG),
                    isInverted ? gen_mi_uge(&b, gen_mi_imm(0), value) :
                                 gen_mi_ult(&b, gen_mi_imm(0), value));
}

void genX(CmdEndConditionalRenderingEXT)(
	VkCommandBuffer                             commandBuffer)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   struct anv_cmd_state *cmd_state = &cmd_buffer->state;

   cmd_state->conditional_render_enabled = false;
}
#endif

/* Set of stage bits for which are pipelined, i.e. they get queued by the
 * command streamer for later execution.
 */
#define ANV_PIPELINE_STAGE_PIPELINED_BITS \
   (VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | \
    VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | \
    VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT | \
    VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT | \
    VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT | \
    VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | \
    VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | \
    VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | \
    VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | \
    VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | \
    VK_PIPELINE_STAGE_TRANSFER_BIT | \
    VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT | \
    VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT | \
    VK_PIPELINE_STAGE_ALL_COMMANDS_BIT)

void genX(CmdSetEvent)(
    VkCommandBuffer                             commandBuffer,
    VkEvent                                     _event,
    VkPipelineStageFlags                        stageMask)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_event, event, _event);

   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_POST_SYNC_BIT;
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
         pc.StallAtPixelScoreboard = true;
         pc.CommandStreamerStallEnable = true;
      }

      pc.DestinationAddressType  = DAT_PPGTT,
      pc.PostSyncOperation       = WriteImmediateData,
      pc.Address = (struct anv_address) {
         cmd_buffer->device->dynamic_state_pool.block_pool.bo,
         event->state.offset
      };
      pc.ImmediateData           = VK_EVENT_SET;
   }
}

void genX(CmdResetEvent)(
    VkCommandBuffer                             commandBuffer,
    VkEvent                                     _event,
    VkPipelineStageFlags                        stageMask)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
   ANV_FROM_HANDLE(anv_event, event, _event);

   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_POST_SYNC_BIT;
   genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);

   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
      if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
         pc.StallAtPixelScoreboard = true;
         pc.CommandStreamerStallEnable = true;
      }

      pc.DestinationAddressType  = DAT_PPGTT;
      pc.PostSyncOperation       = WriteImmediateData;
      pc.Address = (struct anv_address) {
         cmd_buffer->device->dynamic_state_pool.block_pool.bo,
         event->state.offset
      };
      pc.ImmediateData           = VK_EVENT_RESET;
   }
}

void genX(CmdWaitEvents)(
    VkCommandBuffer                             commandBuffer,
    uint32_t                                    eventCount,
    const VkEvent*                              pEvents,
    VkPipelineStageFlags                        srcStageMask,
    VkPipelineStageFlags                        destStageMask,
    uint32_t                                    memoryBarrierCount,
    const VkMemoryBarrier*                      pMemoryBarriers,
    uint32_t                                    bufferMemoryBarrierCount,
    const VkBufferMemoryBarrier*                pBufferMemoryBarriers,
    uint32_t                                    imageMemoryBarrierCount,
    const VkImageMemoryBarrier*                 pImageMemoryBarriers)
{
#if GEN_GEN >= 8
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   for (uint32_t i = 0; i < eventCount; i++) {
      ANV_FROM_HANDLE(anv_event, event, pEvents[i]);

      anv_batch_emit(&cmd_buffer->batch, GENX(MI_SEMAPHORE_WAIT), sem) {
         sem.WaitMode            = PollingMode,
         sem.CompareOperation    = COMPARE_SAD_EQUAL_SDD,
         sem.SemaphoreDataDword  = VK_EVENT_SET,
         sem.SemaphoreAddress = (struct anv_address) {
            cmd_buffer->device->dynamic_state_pool.block_pool.bo,
            event->state.offset
         };
      }
   }
#else
   anv_finishme("Implement events on gen7");
#endif

   genX(CmdPipelineBarrier)(commandBuffer, srcStageMask, destStageMask,
                            false, /* byRegion */
                            memoryBarrierCount, pMemoryBarriers,
                            bufferMemoryBarrierCount, pBufferMemoryBarriers,
                            imageMemoryBarrierCount, pImageMemoryBarriers);
}

VkResult genX(CmdSetPerformanceOverrideINTEL)(
    VkCommandBuffer                             commandBuffer,
    const VkPerformanceOverrideInfoINTEL*       pOverrideInfo)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);

   switch (pOverrideInfo->type) {
   case VK_PERFORMANCE_OVERRIDE_TYPE_NULL_HARDWARE_INTEL: {
      uint32_t dw;

#if GEN_GEN >= 9
      anv_pack_struct(&dw, GENX(CS_DEBUG_MODE2),
                      ._3DRenderingInstructionDisable = pOverrideInfo->enable,
                      .MediaInstructionDisable = pOverrideInfo->enable,
                      ._3DRenderingInstructionDisableMask = true,
                      .MediaInstructionDisableMask = true);
      emit_lri(&cmd_buffer->batch, GENX(CS_DEBUG_MODE2_num), dw);
#else
      anv_pack_struct(&dw, GENX(INSTPM),
                      ._3DRenderingInstructionDisable = pOverrideInfo->enable,
                      .MediaInstructionDisable = pOverrideInfo->enable,
                      ._3DRenderingInstructionDisableMask = true,
                      .MediaInstructionDisableMask = true);
      emit_lri(&cmd_buffer->batch, GENX(INSTPM_num), dw);
#endif
      break;
   }

   case VK_PERFORMANCE_OVERRIDE_TYPE_FLUSH_GPU_CACHES_INTEL:
      if (pOverrideInfo->enable) {
         /* FLUSH ALL THE THINGS! As requested by the MDAPI team. */
         cmd_buffer->state.pending_pipe_bits |=
            ANV_PIPE_FLUSH_BITS |
            ANV_PIPE_INVALIDATE_BITS;
         genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
      }
      break;

   default:
      unreachable("Invalid override");
   }

   return VK_SUCCESS;
}

VkResult genX(CmdSetPerformanceStreamMarkerINTEL)(
    VkCommandBuffer                             commandBuffer,
    const VkPerformanceStreamMarkerInfoINTEL*   pMarkerInfo)
{
   /* TODO: Waiting on the register to write, might depend on generation. */

   return VK_SUCCESS;
}