aboutsummaryrefslogtreecommitdiffstats
path: root/src/intel/blorp/blorp_blit.c
blob: 5bdf05cb0cc77032bc9b9efa7d6383b0a57f5125 (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
/*
 * Copyright © 2012 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 "compiler/nir/nir_builder.h"

#include "blorp_priv.h"
#include "brw_meta_util.h"

#define FILE_DEBUG_FLAG DEBUG_BLORP

/**
 * Enum to specify the order of arguments in a sampler message
 */
enum sampler_message_arg
{
   SAMPLER_MESSAGE_ARG_U_FLOAT,
   SAMPLER_MESSAGE_ARG_V_FLOAT,
   SAMPLER_MESSAGE_ARG_U_INT,
   SAMPLER_MESSAGE_ARG_V_INT,
   SAMPLER_MESSAGE_ARG_R_INT,
   SAMPLER_MESSAGE_ARG_SI_INT,
   SAMPLER_MESSAGE_ARG_MCS_INT,
   SAMPLER_MESSAGE_ARG_ZERO_INT,
};

struct brw_blorp_blit_vars {
   /* Input values from brw_blorp_wm_inputs */
   nir_variable *v_discard_rect;
   nir_variable *v_rect_grid;
   nir_variable *v_coord_transform;
   nir_variable *v_src_z;

   /* gl_FragCoord */
   nir_variable *frag_coord;

   /* gl_FragColor */
   nir_variable *color_out;
};

static void
brw_blorp_blit_vars_init(nir_builder *b, struct brw_blorp_blit_vars *v,
                         const struct brw_blorp_blit_prog_key *key)
{
    /* Blended and scaled blits never use pixel discard. */
    assert(!key->use_kill || !(key->blend && key->blit_scaled));

#define LOAD_INPUT(name, type)\
   v->v_##name = nir_variable_create(b->shader, nir_var_shader_in, \
                                     type, #name); \
   v->v_##name->data.interpolation = INTERP_MODE_FLAT; \
   v->v_##name->data.location = VARYING_SLOT_VAR0 + \
      offsetof(struct brw_blorp_wm_inputs, name) / (4 * sizeof(float));

   LOAD_INPUT(discard_rect, glsl_vec4_type())
   LOAD_INPUT(rect_grid, glsl_vec4_type())
   LOAD_INPUT(coord_transform, glsl_vec4_type())
   LOAD_INPUT(src_z, glsl_uint_type())

#undef LOAD_INPUT

   v->frag_coord = nir_variable_create(b->shader, nir_var_shader_in,
                                       glsl_vec4_type(), "gl_FragCoord");
   v->frag_coord->data.location = VARYING_SLOT_POS;
   v->frag_coord->data.origin_upper_left = true;

   v->color_out = nir_variable_create(b->shader, nir_var_shader_out,
                                      glsl_vec4_type(), "gl_FragColor");
   v->color_out->data.location = FRAG_RESULT_COLOR;
}

static nir_ssa_def *
blorp_blit_get_frag_coords(nir_builder *b,
                           const struct brw_blorp_blit_prog_key *key,
                           struct brw_blorp_blit_vars *v)
{
   nir_ssa_def *coord = nir_f2i(b, nir_load_var(b, v->frag_coord));

   if (key->persample_msaa_dispatch) {
      return nir_vec3(b, nir_channel(b, coord, 0), nir_channel(b, coord, 1),
                      nir_load_sample_id(b));
   } else {
      return nir_vec2(b, nir_channel(b, coord, 0), nir_channel(b, coord, 1));
   }
}

/**
 * Emit code to translate from destination (X, Y) coordinates to source (X, Y)
 * coordinates.
 */
static nir_ssa_def *
blorp_blit_apply_transform(nir_builder *b, nir_ssa_def *src_pos,
                           struct brw_blorp_blit_vars *v)
{
   nir_ssa_def *coord_transform = nir_load_var(b, v->v_coord_transform);

   nir_ssa_def *offset = nir_vec2(b, nir_channel(b, coord_transform, 1),
                                     nir_channel(b, coord_transform, 3));
   nir_ssa_def *mul = nir_vec2(b, nir_channel(b, coord_transform, 0),
                                  nir_channel(b, coord_transform, 2));

   return nir_ffma(b, src_pos, mul, offset);
}

static inline void
blorp_nir_discard_if_outside_rect(nir_builder *b, nir_ssa_def *pos,
                                  struct brw_blorp_blit_vars *v)
{
   nir_ssa_def *c0, *c1, *c2, *c3;
   nir_ssa_def *discard_rect = nir_load_var(b, v->v_discard_rect);
   nir_ssa_def *dst_x0 = nir_channel(b, discard_rect, 0);
   nir_ssa_def *dst_x1 = nir_channel(b, discard_rect, 1);
   nir_ssa_def *dst_y0 = nir_channel(b, discard_rect, 2);
   nir_ssa_def *dst_y1 = nir_channel(b, discard_rect, 3);

   c0 = nir_ult(b, nir_channel(b, pos, 0), dst_x0);
   c1 = nir_uge(b, nir_channel(b, pos, 0), dst_x1);
   c2 = nir_ult(b, nir_channel(b, pos, 1), dst_y0);
   c3 = nir_uge(b, nir_channel(b, pos, 1), dst_y1);

   nir_ssa_def *oob = nir_ior(b, nir_ior(b, c0, c1), nir_ior(b, c2, c3));

   nir_intrinsic_instr *discard =
      nir_intrinsic_instr_create(b->shader, nir_intrinsic_discard_if);
   discard->src[0] = nir_src_for_ssa(oob);
   nir_builder_instr_insert(b, &discard->instr);
}

static nir_tex_instr *
blorp_create_nir_tex_instr(nir_builder *b, struct brw_blorp_blit_vars *v,
                           nir_texop op, nir_ssa_def *pos, unsigned num_srcs,
                           nir_alu_type dst_type)
{
   nir_tex_instr *tex = nir_tex_instr_create(b->shader, num_srcs);

   tex->op = op;

   tex->dest_type = dst_type;
   tex->is_array = false;
   tex->is_shadow = false;

   /* Blorp only has one texture and it's bound at unit 0 */
   tex->texture = NULL;
   tex->sampler = NULL;
   tex->texture_index = 0;
   tex->sampler_index = 0;

   /* To properly handle 3-D and 2-D array textures, we pull the Z component
    * from an input.  TODO: This is a bit magic; we should probably make this
    * more explicit in the future.
    */
   assert(pos->num_components >= 2);
   pos = nir_vec3(b, nir_channel(b, pos, 0), nir_channel(b, pos, 1),
                     nir_load_var(b, v->v_src_z));

   tex->src[0].src_type = nir_tex_src_coord;
   tex->src[0].src = nir_src_for_ssa(pos);
   tex->coord_components = 3;

   nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, NULL);

   return tex;
}

static nir_ssa_def *
blorp_nir_tex(nir_builder *b, struct brw_blorp_blit_vars *v,
              nir_ssa_def *pos, nir_alu_type dst_type)
{
   nir_tex_instr *tex =
      blorp_create_nir_tex_instr(b, v, nir_texop_tex, pos, 2, dst_type);

   assert(pos->num_components == 2);
   tex->sampler_dim = GLSL_SAMPLER_DIM_2D;
   tex->src[1].src_type = nir_tex_src_lod;
   tex->src[1].src = nir_src_for_ssa(nir_imm_int(b, 0));

   nir_builder_instr_insert(b, &tex->instr);

   return &tex->dest.ssa;
}

static nir_ssa_def *
blorp_nir_txf(nir_builder *b, struct brw_blorp_blit_vars *v,
              nir_ssa_def *pos, nir_alu_type dst_type)
{
   nir_tex_instr *tex =
      blorp_create_nir_tex_instr(b, v, nir_texop_txf, pos, 2, dst_type);

   tex->sampler_dim = GLSL_SAMPLER_DIM_3D;
   tex->src[1].src_type = nir_tex_src_lod;
   tex->src[1].src = nir_src_for_ssa(nir_imm_int(b, 0));

   nir_builder_instr_insert(b, &tex->instr);

   return &tex->dest.ssa;
}

static nir_ssa_def *
blorp_nir_txf_ms(nir_builder *b, struct brw_blorp_blit_vars *v,
                 nir_ssa_def *pos, nir_ssa_def *mcs, nir_alu_type dst_type)
{
   nir_tex_instr *tex =
      blorp_create_nir_tex_instr(b, v, nir_texop_txf_ms, pos,
                                 mcs != NULL ? 3 : 2, dst_type);

   tex->sampler_dim = GLSL_SAMPLER_DIM_MS;

   tex->src[1].src_type = nir_tex_src_ms_index;
   if (pos->num_components == 2) {
      tex->src[1].src = nir_src_for_ssa(nir_imm_int(b, 0));
   } else {
      assert(pos->num_components == 3);
      tex->src[1].src = nir_src_for_ssa(nir_channel(b, pos, 2));
   }

   if (mcs) {
      tex->src[2].src_type = nir_tex_src_ms_mcs;
      tex->src[2].src = nir_src_for_ssa(mcs);
   }

   nir_builder_instr_insert(b, &tex->instr);

   return &tex->dest.ssa;
}

static nir_ssa_def *
blorp_nir_txf_ms_mcs(nir_builder *b, struct brw_blorp_blit_vars *v, nir_ssa_def *pos)
{
   nir_tex_instr *tex =
      blorp_create_nir_tex_instr(b, v, nir_texop_txf_ms_mcs,
                                 pos, 1, nir_type_int);

   tex->sampler_dim = GLSL_SAMPLER_DIM_MS;

   nir_builder_instr_insert(b, &tex->instr);

   return &tex->dest.ssa;
}

static nir_ssa_def *
nir_mask_shift_or(struct nir_builder *b, nir_ssa_def *dst, nir_ssa_def *src,
                  uint32_t src_mask, int src_left_shift)
{
   nir_ssa_def *masked = nir_iand(b, src, nir_imm_int(b, src_mask));

   nir_ssa_def *shifted;
   if (src_left_shift > 0) {
      shifted = nir_ishl(b, masked, nir_imm_int(b, src_left_shift));
   } else if (src_left_shift < 0) {
      shifted = nir_ushr(b, masked, nir_imm_int(b, -src_left_shift));
   } else {
      assert(src_left_shift == 0);
      shifted = masked;
   }

   return nir_ior(b, dst, shifted);
}

/**
 * Emit code to compensate for the difference between Y and W tiling.
 *
 * This code modifies the X and Y coordinates according to the formula:
 *
 *   (X', Y', S') = detile(W-MAJOR, tile(Y-MAJOR, X, Y, S))
 *
 * (See brw_blorp_build_nir_shader).
 */
static inline nir_ssa_def *
blorp_nir_retile_y_to_w(nir_builder *b, nir_ssa_def *pos)
{
   assert(pos->num_components == 2);
   nir_ssa_def *x_Y = nir_channel(b, pos, 0);
   nir_ssa_def *y_Y = nir_channel(b, pos, 1);

   /* Given X and Y coordinates that describe an address using Y tiling,
    * translate to the X and Y coordinates that describe the same address
    * using W tiling.
    *
    * If we break down the low order bits of X and Y, using a
    * single letter to represent each low-order bit:
    *
    *   X = A << 7 | 0bBCDEFGH
    *   Y = J << 5 | 0bKLMNP                                       (1)
    *
    * Then we can apply the Y tiling formula to see the memory offset being
    * addressed:
    *
    *   offset = (J * tile_pitch + A) << 12 | 0bBCDKLMNPEFGH       (2)
    *
    * If we apply the W detiling formula to this memory location, that the
    * corresponding X' and Y' coordinates are:
    *
    *   X' = A << 6 | 0bBCDPFH                                     (3)
    *   Y' = J << 6 | 0bKLMNEG
    *
    * Combining (1) and (3), we see that to transform (X, Y) to (X', Y'),
    * we need to make the following computation:
    *
    *   X' = (X & ~0b1011) >> 1 | (Y & 0b1) << 2 | X & 0b1         (4)
    *   Y' = (Y & ~0b1) << 1 | (X & 0b1000) >> 2 | (X & 0b10) >> 1
    */
   nir_ssa_def *x_W = nir_imm_int(b, 0);
   x_W = nir_mask_shift_or(b, x_W, x_Y, 0xfffffff4, -1);
   x_W = nir_mask_shift_or(b, x_W, y_Y, 0x1, 2);
   x_W = nir_mask_shift_or(b, x_W, x_Y, 0x1, 0);

   nir_ssa_def *y_W = nir_imm_int(b, 0);
   y_W = nir_mask_shift_or(b, y_W, y_Y, 0xfffffffe, 1);
   y_W = nir_mask_shift_or(b, y_W, x_Y, 0x8, -2);
   y_W = nir_mask_shift_or(b, y_W, x_Y, 0x2, -1);

   return nir_vec2(b, x_W, y_W);
}

/**
 * Emit code to compensate for the difference between Y and W tiling.
 *
 * This code modifies the X and Y coordinates according to the formula:
 *
 *   (X', Y', S') = detile(Y-MAJOR, tile(W-MAJOR, X, Y, S))
 *
 * (See brw_blorp_build_nir_shader).
 */
static inline nir_ssa_def *
blorp_nir_retile_w_to_y(nir_builder *b, nir_ssa_def *pos)
{
   assert(pos->num_components == 2);
   nir_ssa_def *x_W = nir_channel(b, pos, 0);
   nir_ssa_def *y_W = nir_channel(b, pos, 1);

   /* Applying the same logic as above, but in reverse, we obtain the
    * formulas:
    *
    * X' = (X & ~0b101) << 1 | (Y & 0b10) << 2 | (Y & 0b1) << 1 | X & 0b1
    * Y' = (Y & ~0b11) >> 1 | (X & 0b100) >> 2
    */
   nir_ssa_def *x_Y = nir_imm_int(b, 0);
   x_Y = nir_mask_shift_or(b, x_Y, x_W, 0xfffffffa, 1);
   x_Y = nir_mask_shift_or(b, x_Y, y_W, 0x2, 2);
   x_Y = nir_mask_shift_or(b, x_Y, y_W, 0x1, 1);
   x_Y = nir_mask_shift_or(b, x_Y, x_W, 0x1, 0);

   nir_ssa_def *y_Y = nir_imm_int(b, 0);
   y_Y = nir_mask_shift_or(b, y_Y, y_W, 0xfffffffc, -1);
   y_Y = nir_mask_shift_or(b, y_Y, x_W, 0x4, -2);

   return nir_vec2(b, x_Y, y_Y);
}

/**
 * Emit code to compensate for the difference between MSAA and non-MSAA
 * surfaces.
 *
 * This code modifies the X and Y coordinates according to the formula:
 *
 *   (X', Y', S') = encode_msaa(num_samples, IMS, X, Y, S)
 *
 * (See brw_blorp_blit_program).
 */
static inline nir_ssa_def *
blorp_nir_encode_msaa(nir_builder *b, nir_ssa_def *pos,
                      unsigned num_samples, enum isl_msaa_layout layout)
{
   assert(pos->num_components == 2 || pos->num_components == 3);

   switch (layout) {
   case ISL_MSAA_LAYOUT_NONE:
      assert(pos->num_components == 2);
      return pos;
   case ISL_MSAA_LAYOUT_ARRAY:
      /* No translation needed */
      return pos;
   case ISL_MSAA_LAYOUT_INTERLEAVED: {
      nir_ssa_def *x_in = nir_channel(b, pos, 0);
      nir_ssa_def *y_in = nir_channel(b, pos, 1);
      nir_ssa_def *s_in = pos->num_components == 2 ? nir_imm_int(b, 0) :
                                                     nir_channel(b, pos, 2);

      nir_ssa_def *x_out = nir_imm_int(b, 0);
      nir_ssa_def *y_out = nir_imm_int(b, 0);
      switch (num_samples) {
      case 2:
      case 4:
         /* encode_msaa(2, IMS, X, Y, S) = (X', Y', 0)
          *   where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1)
          *         Y' = Y
          *
          * encode_msaa(4, IMS, X, Y, S) = (X', Y', 0)
          *   where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1)
          *         Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1)
          */
         x_out = nir_mask_shift_or(b, x_out, x_in, 0xfffffffe, 1);
         x_out = nir_mask_shift_or(b, x_out, s_in, 0x1, 1);
         x_out = nir_mask_shift_or(b, x_out, x_in, 0x1, 0);
         if (num_samples == 2) {
            y_out = y_in;
         } else {
            y_out = nir_mask_shift_or(b, y_out, y_in, 0xfffffffe, 1);
            y_out = nir_mask_shift_or(b, y_out, s_in, 0x2, 0);
            y_out = nir_mask_shift_or(b, y_out, y_in, 0x1, 0);
         }
         break;

      case 8:
         /* encode_msaa(8, IMS, X, Y, S) = (X', Y', 0)
          *   where X' = (X & ~0b1) << 2 | (S & 0b100) | (S & 0b1) << 1
          *              | (X & 0b1)
          *         Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1)
          */
         x_out = nir_mask_shift_or(b, x_out, x_in, 0xfffffffe, 2);
         x_out = nir_mask_shift_or(b, x_out, s_in, 0x4, 0);
         x_out = nir_mask_shift_or(b, x_out, s_in, 0x1, 1);
         x_out = nir_mask_shift_or(b, x_out, x_in, 0x1, 0);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0xfffffffe, 1);
         y_out = nir_mask_shift_or(b, y_out, s_in, 0x2, 0);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0x1, 0);
         break;

      case 16:
         /* encode_msaa(16, IMS, X, Y, S) = (X', Y', 0)
          *   where X' = (X & ~0b1) << 2 | (S & 0b100) | (S & 0b1) << 1
          *              | (X & 0b1)
          *         Y' = (Y & ~0b1) << 2 | (S & 0b1000) >> 1 (S & 0b10)
          *              | (Y & 0b1)
          */
         x_out = nir_mask_shift_or(b, x_out, x_in, 0xfffffffe, 2);
         x_out = nir_mask_shift_or(b, x_out, s_in, 0x4, 0);
         x_out = nir_mask_shift_or(b, x_out, s_in, 0x1, 1);
         x_out = nir_mask_shift_or(b, x_out, x_in, 0x1, 0);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0xfffffffe, 2);
         y_out = nir_mask_shift_or(b, y_out, s_in, 0x8, -1);
         y_out = nir_mask_shift_or(b, y_out, s_in, 0x2, 0);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0x1, 0);
         break;

      default:
         unreachable("Invalid number of samples for IMS layout");
      }

      return nir_vec2(b, x_out, y_out);
   }

   default:
      unreachable("Invalid MSAA layout");
   }
}

/**
 * Emit code to compensate for the difference between MSAA and non-MSAA
 * surfaces.
 *
 * This code modifies the X and Y coordinates according to the formula:
 *
 *   (X', Y', S) = decode_msaa(num_samples, IMS, X, Y, S)
 *
 * (See brw_blorp_blit_program).
 */
static inline nir_ssa_def *
blorp_nir_decode_msaa(nir_builder *b, nir_ssa_def *pos,
                      unsigned num_samples, enum isl_msaa_layout layout)
{
   assert(pos->num_components == 2 || pos->num_components == 3);

   switch (layout) {
   case ISL_MSAA_LAYOUT_NONE:
      /* No translation necessary, and S should already be zero. */
      assert(pos->num_components == 2);
      return pos;
   case ISL_MSAA_LAYOUT_ARRAY:
      /* No translation necessary. */
      return pos;
   case ISL_MSAA_LAYOUT_INTERLEAVED: {
      assert(pos->num_components == 2);

      nir_ssa_def *x_in = nir_channel(b, pos, 0);
      nir_ssa_def *y_in = nir_channel(b, pos, 1);

      nir_ssa_def *x_out = nir_imm_int(b, 0);
      nir_ssa_def *y_out = nir_imm_int(b, 0);
      nir_ssa_def *s_out = nir_imm_int(b, 0);
      switch (num_samples) {
      case 2:
      case 4:
         /* decode_msaa(2, IMS, X, Y, 0) = (X', Y', S)
          *   where X' = (X & ~0b11) >> 1 | (X & 0b1)
          *         S = (X & 0b10) >> 1
          *
          * decode_msaa(4, IMS, X, Y, 0) = (X', Y', S)
          *   where X' = (X & ~0b11) >> 1 | (X & 0b1)
          *         Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
          *         S = (Y & 0b10) | (X & 0b10) >> 1
          */
         x_out = nir_mask_shift_or(b, x_out, x_in, 0xfffffffc, -1);
         x_out = nir_mask_shift_or(b, x_out, x_in, 0x1, 0);
         if (num_samples == 2) {
            y_out = y_in;
            s_out = nir_mask_shift_or(b, s_out, x_in, 0x2, -1);
         } else {
            y_out = nir_mask_shift_or(b, y_out, y_in, 0xfffffffc, -1);
            y_out = nir_mask_shift_or(b, y_out, y_in, 0x1, 0);
            s_out = nir_mask_shift_or(b, s_out, x_in, 0x2, -1);
            s_out = nir_mask_shift_or(b, s_out, y_in, 0x2, 0);
         }
         break;

      case 8:
         /* decode_msaa(8, IMS, X, Y, 0) = (X', Y', S)
          *   where X' = (X & ~0b111) >> 2 | (X & 0b1)
          *         Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
          *         S = (X & 0b100) | (Y & 0b10) | (X & 0b10) >> 1
          */
         x_out = nir_mask_shift_or(b, x_out, x_in, 0xfffffff8, -2);
         x_out = nir_mask_shift_or(b, x_out, x_in, 0x1, 0);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0xfffffffc, -1);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0x1, 0);
         s_out = nir_mask_shift_or(b, s_out, x_in, 0x4, 0);
         s_out = nir_mask_shift_or(b, s_out, y_in, 0x2, 0);
         s_out = nir_mask_shift_or(b, s_out, x_in, 0x2, -1);
         break;

      case 16:
         /* decode_msaa(16, IMS, X, Y, 0) = (X', Y', S)
          *   where X' = (X & ~0b111) >> 2 | (X & 0b1)
          *         Y' = (Y & ~0b111) >> 2 | (Y & 0b1)
          *         S = (Y & 0b100) << 1 | (X & 0b100) |
          *             (Y & 0b10) | (X & 0b10) >> 1
          */
         x_out = nir_mask_shift_or(b, x_out, x_in, 0xfffffff8, -2);
         x_out = nir_mask_shift_or(b, x_out, x_in, 0x1, 0);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0xfffffff8, -2);
         y_out = nir_mask_shift_or(b, y_out, y_in, 0x1, 0);
         s_out = nir_mask_shift_or(b, s_out, y_in, 0x4, 1);
         s_out = nir_mask_shift_or(b, s_out, x_in, 0x4, 0);
         s_out = nir_mask_shift_or(b, s_out, y_in, 0x2, 0);
         s_out = nir_mask_shift_or(b, s_out, x_in, 0x2, -1);
         break;

      default:
         unreachable("Invalid number of samples for IMS layout");
      }

      return nir_vec3(b, x_out, y_out, s_out);
   }

   default:
      unreachable("Invalid MSAA layout");
   }
}

/**
 * Count the number of trailing 1 bits in the given value.  For example:
 *
 * count_trailing_one_bits(0) == 0
 * count_trailing_one_bits(7) == 3
 * count_trailing_one_bits(11) == 2
 */
static inline int count_trailing_one_bits(unsigned value)
{
#ifdef HAVE___BUILTIN_CTZ
   return __builtin_ctz(~value);
#else
   return _mesa_bitcount(value & ~(value + 1));
#endif
}

static nir_ssa_def *
blorp_nir_manual_blend_average(nir_builder *b, struct brw_blorp_blit_vars *v,
                               nir_ssa_def *pos, unsigned tex_samples,
                               enum isl_aux_usage tex_aux_usage,
                               nir_alu_type dst_type)
{
   /* If non-null, this is the outer-most if statement */
   nir_if *outer_if = NULL;

   nir_variable *color =
      nir_local_variable_create(b->impl, glsl_vec4_type(), "color");

   nir_ssa_def *mcs = NULL;
   if (tex_aux_usage == ISL_AUX_USAGE_MCS)
      mcs = blorp_nir_txf_ms_mcs(b, v, pos);

   /* We add together samples using a binary tree structure, e.g. for 4x MSAA:
    *
    *   result = ((sample[0] + sample[1]) + (sample[2] + sample[3])) / 4
    *
    * This ensures that when all samples have the same value, no numerical
    * precision is lost, since each addition operation always adds two equal
    * values, and summing two equal floating point values does not lose
    * precision.
    *
    * We perform this computation by treating the texture_data array as a
    * stack and performing the following operations:
    *
    * - push sample 0 onto stack
    * - push sample 1 onto stack
    * - add top two stack entries
    * - push sample 2 onto stack
    * - push sample 3 onto stack
    * - add top two stack entries
    * - add top two stack entries
    * - divide top stack entry by 4
    *
    * Note that after pushing sample i onto the stack, the number of add
    * operations we do is equal to the number of trailing 1 bits in i.  This
    * works provided the total number of samples is a power of two, which it
    * always is for i965.
    *
    * For integer formats, we replace the add operations with average
    * operations and skip the final division.
    */
   nir_ssa_def *texture_data[5];
   unsigned stack_depth = 0;
   for (unsigned i = 0; i < tex_samples; ++i) {
      assert(stack_depth == _mesa_bitcount(i)); /* Loop invariant */

      /* Push sample i onto the stack */
      assert(stack_depth < ARRAY_SIZE(texture_data));

      nir_ssa_def *ms_pos = nir_vec3(b, nir_channel(b, pos, 0),
                                        nir_channel(b, pos, 1),
                                        nir_imm_int(b, i));
      texture_data[stack_depth++] = blorp_nir_txf_ms(b, v, ms_pos, mcs, dst_type);

      if (i == 0 && tex_aux_usage == ISL_AUX_USAGE_MCS) {
         /* The Ivy Bridge PRM, Vol4 Part1 p27 (Multisample Control Surface)
          * suggests an optimization:
          *
          *     "A simple optimization with probable large return in
          *     performance is to compare the MCS value to zero (indicating
          *     all samples are on sample slice 0), and sample only from
          *     sample slice 0 using ld2dss if MCS is zero."
          *
          * Note that in the case where the MCS value is zero, sampling from
          * sample slice 0 using ld2dss and sampling from sample 0 using
          * ld2dms are equivalent (since all samples are on sample slice 0).
          * Since we have already sampled from sample 0, all we need to do is
          * skip the remaining fetches and averaging if MCS is zero.
          */
         nir_ssa_def *mcs_zero =
            nir_ieq(b, nir_channel(b, mcs, 0), nir_imm_int(b, 0));
         if (tex_samples == 16) {
            mcs_zero = nir_iand(b, mcs_zero,
               nir_ieq(b, nir_channel(b, mcs, 1), nir_imm_int(b, 0)));
         }

         nir_if *if_stmt = nir_if_create(b->shader);
         if_stmt->condition = nir_src_for_ssa(mcs_zero);
         nir_cf_node_insert(b->cursor, &if_stmt->cf_node);

         b->cursor = nir_after_cf_list(&if_stmt->then_list);
         nir_store_var(b, color, texture_data[0], 0xf);

         b->cursor = nir_after_cf_list(&if_stmt->else_list);
         outer_if = if_stmt;
      }

      for (int j = 0; j < count_trailing_one_bits(i); j++) {
         assert(stack_depth >= 2);
         --stack_depth;

         assert(dst_type == nir_type_float);
         texture_data[stack_depth - 1] =
            nir_fadd(b, texture_data[stack_depth - 1],
                        texture_data[stack_depth]);
      }
   }

   /* We should have just 1 sample on the stack now. */
   assert(stack_depth == 1);

   texture_data[0] = nir_fmul(b, texture_data[0],
                              nir_imm_float(b, 1.0 / tex_samples));

   nir_store_var(b, color, texture_data[0], 0xf);

   if (outer_if)
      b->cursor = nir_after_cf_node(&outer_if->cf_node);

   return nir_load_var(b, color);
}

static inline nir_ssa_def *
nir_imm_vec2(nir_builder *build, float x, float y)
{
   nir_const_value v;

   memset(&v, 0, sizeof(v));
   v.f32[0] = x;
   v.f32[1] = y;

   return nir_build_imm(build, 4, 32, v);
}

static nir_ssa_def *
blorp_nir_manual_blend_bilinear(nir_builder *b, nir_ssa_def *pos,
                                unsigned tex_samples,
                                const struct brw_blorp_blit_prog_key *key,
                                struct brw_blorp_blit_vars *v)
{
   nir_ssa_def *pos_xy = nir_channels(b, pos, 0x3);
   nir_ssa_def *rect_grid = nir_load_var(b, v->v_rect_grid);
   nir_ssa_def *scale = nir_imm_vec2(b, key->x_scale, key->y_scale);

   /* Translate coordinates to lay out the samples in a rectangular  grid
    * roughly corresponding to sample locations.
    */
   pos_xy = nir_fmul(b, pos_xy, scale);
   /* Adjust coordinates so that integers represent pixel centers rather
    * than pixel edges.
    */
   pos_xy = nir_fadd(b, pos_xy, nir_imm_float(b, -0.5));
   /* Clamp the X, Y texture coordinates to properly handle the sampling of
    * texels on texture edges.
    */
   pos_xy = nir_fmin(b, nir_fmax(b, pos_xy, nir_imm_float(b, 0.0)),
                        nir_vec2(b, nir_channel(b, rect_grid, 0),
                                    nir_channel(b, rect_grid, 1)));

   /* Store the fractional parts to be used as bilinear interpolation
    * coefficients.
    */
   nir_ssa_def *frac_xy = nir_ffract(b, pos_xy);
   /* Round the float coordinates down to nearest integer */
   pos_xy = nir_fdiv(b, nir_ftrunc(b, pos_xy), scale);

   nir_ssa_def *tex_data[4];
   for (unsigned i = 0; i < 4; ++i) {
      float sample_off_x = (float)(i & 0x1) / key->x_scale;
      float sample_off_y = (float)((i >> 1) & 0x1) / key->y_scale;
      nir_ssa_def *sample_off = nir_imm_vec2(b, sample_off_x, sample_off_y);

      nir_ssa_def *sample_coords = nir_fadd(b, pos_xy, sample_off);
      nir_ssa_def *sample_coords_int = nir_f2i(b, sample_coords);

      /* The MCS value we fetch has to match up with the pixel that we're
       * sampling from. Since we sample from different pixels in each
       * iteration of this "for" loop, the call to mcs_fetch() should be
       * here inside the loop after computing the pixel coordinates.
       */
      nir_ssa_def *mcs = NULL;
      if (key->tex_aux_usage == ISL_AUX_USAGE_MCS)
         mcs = blorp_nir_txf_ms_mcs(b, v, sample_coords_int);

      /* Compute sample index and map the sample index to a sample number.
       * Sample index layout shows the numbering of slots in a rectangular
       * grid of samples with in a pixel. Sample number layout shows the
       * rectangular grid of samples roughly corresponding to the real sample
       * locations with in a pixel.
       * In case of 4x MSAA, layout of sample indices matches the layout of
       * sample numbers:
       *           ---------
       *           | 0 | 1 |
       *           ---------
       *           | 2 | 3 |
       *           ---------
       *
       * In case of 8x MSAA the two layouts don't match.
       * sample index layout :  ---------    sample number layout :  ---------
       *                        | 0 | 1 |                            | 3 | 7 |
       *                        ---------                            ---------
       *                        | 2 | 3 |                            | 5 | 0 |
       *                        ---------                            ---------
       *                        | 4 | 5 |                            | 1 | 2 |
       *                        ---------                            ---------
       *                        | 6 | 7 |                            | 4 | 6 |
       *                        ---------                            ---------
       *
       * Fortunately, this can be done fairly easily as:
       * S' = (0x17306425 >> (S * 4)) & 0xf
       *
       * In the case of 16x MSAA the two layouts don't match.
       * Sample index layout:                Sample number layout:
       * ---------------------               ---------------------
       * |  0 |  1 |  2 |  3 |               | 15 | 10 |  9 |  7 |
       * ---------------------               ---------------------
       * |  4 |  5 |  6 |  7 |               |  4 |  1 |  3 | 13 |
       * ---------------------               ---------------------
       * |  8 |  9 | 10 | 11 |               | 12 |  2 |  0 |  6 |
       * ---------------------               ---------------------
       * | 12 | 13 | 14 | 15 |               | 11 |  8 |  5 | 14 |
       * ---------------------               ---------------------
       *
       * This is equivalent to
       * S' = (0xe58b602cd31479af >> (S * 4)) & 0xf
       */
      nir_ssa_def *frac = nir_ffract(b, sample_coords);
      nir_ssa_def *sample =
         nir_fdot2(b, frac, nir_imm_vec2(b, key->x_scale,
                                            key->x_scale * key->y_scale));
      sample = nir_f2i(b, sample);

      if (tex_samples == 8) {
         sample = nir_iand(b, nir_ishr(b, nir_imm_int(b, 0x64210573),
                                       nir_ishl(b, sample, nir_imm_int(b, 2))),
                           nir_imm_int(b, 0xf));
      } else if (tex_samples == 16) {
         nir_ssa_def *sample_low =
            nir_iand(b, nir_ishr(b, nir_imm_int(b, 0xd31479af),
                                 nir_ishl(b, sample, nir_imm_int(b, 2))),
                     nir_imm_int(b, 0xf));
         nir_ssa_def *sample_high =
            nir_iand(b, nir_ishr(b, nir_imm_int(b, 0xe58b602c),
                                 nir_ishl(b, nir_iadd(b, sample,
                                                      nir_imm_int(b, -8)),
                                          nir_imm_int(b, 2))),
                     nir_imm_int(b, 0xf));

         sample = nir_bcsel(b, nir_ilt(b, sample, nir_imm_int(b, 8)),
                            sample_low, sample_high);
      }
      nir_ssa_def *pos_ms = nir_vec3(b, nir_channel(b, sample_coords_int, 0),
                                        nir_channel(b, sample_coords_int, 1),
                                        sample);
      tex_data[i] = blorp_nir_txf_ms(b, v, pos_ms, mcs, key->texture_data_type);
   }

   nir_ssa_def *frac_x = nir_channel(b, frac_xy, 0);
   nir_ssa_def *frac_y = nir_channel(b, frac_xy, 1);
   return nir_flrp(b, nir_flrp(b, tex_data[0], tex_data[1], frac_x),
                      nir_flrp(b, tex_data[2], tex_data[3], frac_x),
                      frac_y);
}

/**
 * Generator for WM programs used in BLORP blits.
 *
 * The bulk of the work done by the WM program is to wrap and unwrap the
 * coordinate transformations used by the hardware to store surfaces in
 * memory.  The hardware transforms a pixel location (X, Y, S) (where S is the
 * sample index for a multisampled surface) to a memory offset by the
 * following formulas:
 *
 *   offset = tile(tiling_format, encode_msaa(num_samples, layout, X, Y, S))
 *   (X, Y, S) = decode_msaa(num_samples, layout, detile(tiling_format, offset))
 *
 * For a single-sampled surface, or for a multisampled surface using
 * INTEL_MSAA_LAYOUT_UMS, encode_msaa() and decode_msaa are the identity
 * function:
 *
 *   encode_msaa(1, NONE, X, Y, 0) = (X, Y, 0)
 *   decode_msaa(1, NONE, X, Y, 0) = (X, Y, 0)
 *   encode_msaa(n, UMS, X, Y, S) = (X, Y, S)
 *   decode_msaa(n, UMS, X, Y, S) = (X, Y, S)
 *
 * For a 4x multisampled surface using INTEL_MSAA_LAYOUT_IMS, encode_msaa()
 * embeds the sample number into bit 1 of the X and Y coordinates:
 *
 *   encode_msaa(4, IMS, X, Y, S) = (X', Y', 0)
 *     where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1)
 *           Y' = (Y & ~0b1 ) << 1 | (S & 0b10) | (Y & 0b1)
 *   decode_msaa(4, IMS, X, Y, 0) = (X', Y', S)
 *     where X' = (X & ~0b11) >> 1 | (X & 0b1)
 *           Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
 *           S = (Y & 0b10) | (X & 0b10) >> 1
 *
 * For an 8x multisampled surface using INTEL_MSAA_LAYOUT_IMS, encode_msaa()
 * embeds the sample number into bits 1 and 2 of the X coordinate and bit 1 of
 * the Y coordinate:
 *
 *   encode_msaa(8, IMS, X, Y, S) = (X', Y', 0)
 *     where X' = (X & ~0b1) << 2 | (S & 0b100) | (S & 0b1) << 1 | (X & 0b1)
 *           Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1)
 *   decode_msaa(8, IMS, X, Y, 0) = (X', Y', S)
 *     where X' = (X & ~0b111) >> 2 | (X & 0b1)
 *           Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
 *           S = (X & 0b100) | (Y & 0b10) | (X & 0b10) >> 1
 *
 * For X tiling, tile() combines together the low-order bits of the X and Y
 * coordinates in the pattern 0byyyxxxxxxxxx, creating 4k tiles that are 512
 * bytes wide and 8 rows high:
 *
 *   tile(x_tiled, X, Y, S) = A
 *     where A = tile_num << 12 | offset
 *           tile_num = (Y' >> 3) * tile_pitch + (X' >> 9)
 *           offset = (Y' & 0b111) << 9
 *                    | (X & 0b111111111)
 *           X' = X * cpp
 *           Y' = Y + S * qpitch
 *   detile(x_tiled, A) = (X, Y, S)
 *     where X = X' / cpp
 *           Y = Y' % qpitch
 *           S = Y' / qpitch
 *           Y' = (tile_num / tile_pitch) << 3
 *                | (A & 0b111000000000) >> 9
 *           X' = (tile_num % tile_pitch) << 9
 *                | (A & 0b111111111)
 *
 * (In all tiling formulas, cpp is the number of bytes occupied by a single
 * sample ("chars per pixel"), tile_pitch is the number of 4k tiles required
 * to fill the width of the surface, and qpitch is the spacing (in rows)
 * between array slices).
 *
 * For Y tiling, tile() combines together the low-order bits of the X and Y
 * coordinates in the pattern 0bxxxyyyyyxxxx, creating 4k tiles that are 128
 * bytes wide and 32 rows high:
 *
 *   tile(y_tiled, X, Y, S) = A
 *     where A = tile_num << 12 | offset
 *           tile_num = (Y' >> 5) * tile_pitch + (X' >> 7)
 *           offset = (X' & 0b1110000) << 5
 *                    | (Y' & 0b11111) << 4
 *                    | (X' & 0b1111)
 *           X' = X * cpp
 *           Y' = Y + S * qpitch
 *   detile(y_tiled, A) = (X, Y, S)
 *     where X = X' / cpp
 *           Y = Y' % qpitch
 *           S = Y' / qpitch
 *           Y' = (tile_num / tile_pitch) << 5
 *                | (A & 0b111110000) >> 4
 *           X' = (tile_num % tile_pitch) << 7
 *                | (A & 0b111000000000) >> 5
 *                | (A & 0b1111)
 *
 * For W tiling, tile() combines together the low-order bits of the X and Y
 * coordinates in the pattern 0bxxxyyyyxyxyx, creating 4k tiles that are 64
 * bytes wide and 64 rows high (note that W tiling is only used for stencil
 * buffers, which always have cpp = 1 and S=0):
 *
 *   tile(w_tiled, X, Y, S) = A
 *     where A = tile_num << 12 | offset
 *           tile_num = (Y' >> 6) * tile_pitch + (X' >> 6)
 *           offset = (X' & 0b111000) << 6
 *                    | (Y' & 0b111100) << 3
 *                    | (X' & 0b100) << 2
 *                    | (Y' & 0b10) << 2
 *                    | (X' & 0b10) << 1
 *                    | (Y' & 0b1) << 1
 *                    | (X' & 0b1)
 *           X' = X * cpp = X
 *           Y' = Y + S * qpitch
 *   detile(w_tiled, A) = (X, Y, S)
 *     where X = X' / cpp = X'
 *           Y = Y' % qpitch = Y'
 *           S = Y / qpitch = 0
 *           Y' = (tile_num / tile_pitch) << 6
 *                | (A & 0b111100000) >> 3
 *                | (A & 0b1000) >> 2
 *                | (A & 0b10) >> 1
 *           X' = (tile_num % tile_pitch) << 6
 *                | (A & 0b111000000000) >> 6
 *                | (A & 0b10000) >> 2
 *                | (A & 0b100) >> 1
 *                | (A & 0b1)
 *
 * Finally, for a non-tiled surface, tile() simply combines together the X and
 * Y coordinates in the natural way:
 *
 *   tile(untiled, X, Y, S) = A
 *     where A = Y * pitch + X'
 *           X' = X * cpp
 *           Y' = Y + S * qpitch
 *   detile(untiled, A) = (X, Y, S)
 *     where X = X' / cpp
 *           Y = Y' % qpitch
 *           S = Y' / qpitch
 *           X' = A % pitch
 *           Y' = A / pitch
 *
 * (In these formulas, pitch is the number of bytes occupied by a single row
 * of samples).
 */
static nir_shader *
brw_blorp_build_nir_shader(struct blorp_context *blorp,
                           const struct brw_blorp_blit_prog_key *key)
{
   const struct gen_device_info *devinfo = blorp->isl_dev->info;
   nir_ssa_def *src_pos, *dst_pos, *color;

   /* Sanity checks */
   if (key->dst_tiled_w && key->rt_samples > 1) {
      /* If the destination image is W tiled and multisampled, then the thread
       * must be dispatched once per sample, not once per pixel.  This is
       * necessary because after conversion between W and Y tiling, there's no
       * guarantee that all samples corresponding to a single pixel will still
       * be together.
       */
      assert(key->persample_msaa_dispatch);
   }

   if (key->blend) {
      /* We are blending, which means we won't have an opportunity to
       * translate the tiling and sample count for the texture surface.  So
       * the surface state for the texture must be configured with the correct
       * tiling and sample count.
       */
      assert(!key->src_tiled_w);
      assert(key->tex_samples == key->src_samples);
      assert(key->tex_layout == key->src_layout);
      assert(key->tex_samples > 0);
   }

   if (key->persample_msaa_dispatch) {
      /* It only makes sense to do persample dispatch if the render target is
       * configured as multisampled.
       */
      assert(key->rt_samples > 0);
   }

   /* Make sure layout is consistent with sample count */
   assert((key->tex_layout == ISL_MSAA_LAYOUT_NONE) ==
          (key->tex_samples <= 1));
   assert((key->rt_layout == ISL_MSAA_LAYOUT_NONE) ==
          (key->rt_samples <= 1));
   assert((key->src_layout == ISL_MSAA_LAYOUT_NONE) ==
          (key->src_samples <= 1));
   assert((key->dst_layout == ISL_MSAA_LAYOUT_NONE) ==
          (key->dst_samples <= 1));

   nir_builder b;
   nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);

   struct brw_blorp_blit_vars v;
   brw_blorp_blit_vars_init(&b, &v, key);

   dst_pos = blorp_blit_get_frag_coords(&b, key, &v);

   /* Render target and texture hardware don't support W tiling until Gen8. */
   const bool rt_tiled_w = false;
   const bool tex_tiled_w = devinfo->gen >= 8 && key->src_tiled_w;

   /* The address that data will be written to is determined by the
    * coordinates supplied to the WM thread and the tiling and sample count of
    * the render target, according to the formula:
    *
    * (X, Y, S) = decode_msaa(rt_samples, detile(rt_tiling, offset))
    *
    * If the actual tiling and sample count of the destination surface are not
    * the same as the configuration of the render target, then these
    * coordinates are wrong and we have to adjust them to compensate for the
    * difference.
    */
   if (rt_tiled_w != key->dst_tiled_w ||
       key->rt_samples != key->dst_samples ||
       key->rt_layout != key->dst_layout) {
      dst_pos = blorp_nir_encode_msaa(&b, dst_pos, key->rt_samples,
                                      key->rt_layout);
      /* Now (X, Y, S) = detile(rt_tiling, offset) */
      if (rt_tiled_w != key->dst_tiled_w)
         dst_pos = blorp_nir_retile_y_to_w(&b, dst_pos);
      /* Now (X, Y, S) = detile(rt_tiling, offset) */
      dst_pos = blorp_nir_decode_msaa(&b, dst_pos, key->dst_samples,
                                      key->dst_layout);
   }

   /* Now (X, Y, S) = decode_msaa(dst_samples, detile(dst_tiling, offset)).
    *
    * That is: X, Y and S now contain the true coordinates and sample index of
    * the data that the WM thread should output.
    *
    * If we need to kill pixels that are outside the destination rectangle,
    * now is the time to do it.
    */
   if (key->use_kill) {
      assert(!(key->blend && key->blit_scaled));
      blorp_nir_discard_if_outside_rect(&b, dst_pos, &v);
   }

   src_pos = blorp_blit_apply_transform(&b, nir_i2f(&b, dst_pos), &v);
   if (dst_pos->num_components == 3) {
      /* The sample coordinate is an integer that we want left alone but
       * blorp_blit_apply_transform() blindly applies the transform to all
       * three coordinates.  Grab the original sample index.
       */
      src_pos = nir_vec3(&b, nir_channel(&b, src_pos, 0),
                             nir_channel(&b, src_pos, 1),
                             nir_channel(&b, dst_pos, 2));
   }

   /* If the source image is not multisampled, then we want to fetch sample
    * number 0, because that's the only sample there is.
    */
   if (key->src_samples == 1)
      src_pos = nir_channels(&b, src_pos, 0x3);

   /* X, Y, and S are now the coordinates of the pixel in the source image
    * that we want to texture from.  Exception: if we are blending, then S is
    * irrelevant, because we are going to fetch all samples.
    */
   if (key->blend && !key->blit_scaled) {
      /* Resolves (effecively) use texelFetch, so we need integers and we
       * don't care about the sample index if we got one.
       */
      src_pos = nir_f2i(&b, nir_channels(&b, src_pos, 0x3));

      if (devinfo->gen == 6) {
         /* Because gen6 only supports 4x interleved MSAA, we can do all the
          * blending we need with a single linear-interpolated texture lookup
          * at the center of the sample. The texture coordinates to be odd
          * integers so that they correspond to the center of a 2x2 block
          * representing the four samples that maxe up a pixel.  So we need
          * to multiply our X and Y coordinates each by 2 and then add 1.
          */
         src_pos = nir_ishl(&b, src_pos, nir_imm_int(&b, 1));
         src_pos = nir_iadd(&b, src_pos, nir_imm_int(&b, 1));
         src_pos = nir_i2f(&b, src_pos);
         color = blorp_nir_tex(&b, &v, src_pos, key->texture_data_type);
      } else {
         /* Gen7+ hardware doesn't automaticaly blend. */
         color = blorp_nir_manual_blend_average(&b, &v, src_pos, key->src_samples,
                                                key->tex_aux_usage,
                                                key->texture_data_type);
      }
   } else if (key->blend && key->blit_scaled) {
      assert(!key->use_kill);
      color = blorp_nir_manual_blend_bilinear(&b, src_pos, key->src_samples, key, &v);
   } else {
      if (key->bilinear_filter) {
         color = blorp_nir_tex(&b, &v, src_pos, key->texture_data_type);
      } else {
         /* We're going to use texelFetch, so we need integers */
         if (src_pos->num_components == 2) {
            src_pos = nir_f2i(&b, src_pos);
         } else {
            assert(src_pos->num_components == 3);
            src_pos = nir_vec3(&b, nir_channel(&b, nir_f2i(&b, src_pos), 0),
                                   nir_channel(&b, nir_f2i(&b, src_pos), 1),
                                   nir_channel(&b, src_pos, 2));
         }

         /* We aren't blending, which means we just want to fetch a single
          * sample from the source surface.  The address that we want to fetch
          * from is related to the X, Y and S values according to the formula:
          *
          * (X, Y, S) = decode_msaa(src_samples, detile(src_tiling, offset)).
          *
          * If the actual tiling and sample count of the source surface are
          * not the same as the configuration of the texture, then we need to
          * adjust the coordinates to compensate for the difference.
          */
         if (tex_tiled_w != key->src_tiled_w ||
             key->tex_samples != key->src_samples ||
             key->tex_layout != key->src_layout) {
            src_pos = blorp_nir_encode_msaa(&b, src_pos, key->src_samples,
                                            key->src_layout);
            /* Now (X, Y, S) = detile(src_tiling, offset) */
            if (tex_tiled_w != key->src_tiled_w)
               src_pos = blorp_nir_retile_w_to_y(&b, src_pos);
            /* Now (X, Y, S) = detile(tex_tiling, offset) */
            src_pos = blorp_nir_decode_msaa(&b, src_pos, key->tex_samples,
                                            key->tex_layout);
         }

         /* Now (X, Y, S) = decode_msaa(tex_samples, detile(tex_tiling, offset)).
          *
          * In other words: X, Y, and S now contain values which, when passed to
          * the texturing unit, will cause data to be read from the correct
          * memory location.  So we can fetch the texel now.
          */
         if (key->src_samples == 1) {
            color = blorp_nir_txf(&b, &v, src_pos, key->texture_data_type);
         } else {
            nir_ssa_def *mcs = NULL;
            if (key->tex_aux_usage == ISL_AUX_USAGE_MCS)
               mcs = blorp_nir_txf_ms_mcs(&b, &v, src_pos);

            color = blorp_nir_txf_ms(&b, &v, src_pos, mcs, key->texture_data_type);
         }
      }
   }

   nir_store_var(&b, v.color_out, color, 0xf);

   return b.shader;
}

static void
brw_blorp_get_blit_kernel(struct blorp_context *blorp,
                          struct blorp_params *params,
                          const struct brw_blorp_blit_prog_key *prog_key)
{
   if (blorp->lookup_shader(blorp, prog_key, sizeof(*prog_key),
                            &params->wm_prog_kernel, &params->wm_prog_data))
      return;

   const unsigned *program;
   unsigned program_size;
   struct brw_blorp_prog_data prog_data;

   /* Try and compile with NIR first.  If that fails, fall back to the old
    * method of building shaders manually.
    */
   nir_shader *nir = brw_blorp_build_nir_shader(blorp, prog_key);
   struct brw_wm_prog_key wm_key;
   brw_blorp_init_wm_prog_key(&wm_key);
   wm_key.tex.compressed_multisample_layout_mask =
      prog_key->tex_aux_usage == ISL_AUX_USAGE_MCS;
   wm_key.tex.msaa_16 = prog_key->tex_samples == 16;
   wm_key.multisample_fbo = prog_key->rt_samples > 1;

   program = brw_blorp_compile_nir_shader(blorp, nir, &wm_key, false,
                                          &prog_data, &program_size);

   blorp->upload_shader(blorp, prog_key, sizeof(*prog_key),
                        program, program_size,
                        &prog_data, sizeof(prog_data),
                        &params->wm_prog_kernel, &params->wm_prog_data);
}

static void
brw_blorp_setup_coord_transform(struct brw_blorp_coord_transform *xform,
                                GLfloat src0, GLfloat src1,
                                GLfloat dst0, GLfloat dst1,
                                bool mirror)
{
   float scale = (src1 - src0) / (dst1 - dst0);
   if (!mirror) {
      /* When not mirroring a coordinate (say, X), we need:
       *   src_x - src_x0 = (dst_x - dst_x0 + 0.5) * scale
       * Therefore:
       *   src_x = src_x0 + (dst_x - dst_x0 + 0.5) * scale
       *
       * blorp program uses "round toward zero" to convert the
       * transformed floating point coordinates to integer coordinates,
       * whereas the behaviour we actually want is "round to nearest",
       * so 0.5 provides the necessary correction.
       */
      xform->multiplier = scale;
      xform->offset = src0 + (-dst0 + 0.5f) * scale;
   } else {
      /* When mirroring X we need:
       *   src_x - src_x0 = dst_x1 - dst_x - 0.5
       * Therefore:
       *   src_x = src_x0 + (dst_x1 -dst_x - 0.5) * scale
       */
      xform->multiplier = -scale;
      xform->offset = src0 + (dst1 - 0.5f) * scale;
   }
}

static void
surf_convert_to_single_slice(const struct isl_device *isl_dev,
                             struct brw_blorp_surface_info *info)
{
   /* Just bail if we have nothing to do. */
   if (info->surf.dim == ISL_SURF_DIM_2D &&
       info->view.base_level == 0 && info->view.base_array_layer == 0 &&
       info->surf.levels == 0 && info->surf.logical_level0_px.array_len == 0)
      return;

   uint32_t x_offset_sa, y_offset_sa;
   isl_surf_get_image_offset_sa(&info->surf, info->view.base_level,
                                info->view.base_array_layer, 0,
                                &x_offset_sa, &y_offset_sa);

   uint32_t byte_offset;
   isl_tiling_get_intratile_offset_sa(isl_dev, info->surf.tiling,
                                      info->view.format, info->surf.row_pitch,
                                      x_offset_sa, y_offset_sa,
                                      &byte_offset,
                                      &info->tile_x_sa, &info->tile_y_sa);
   info->addr.offset += byte_offset;

   /* TODO: Once this file gets converted to C, we shouls just use designated
    * initializers.
    */
   struct isl_surf_init_info init_info = { 0, };

   init_info.dim = ISL_SURF_DIM_2D;
   init_info.format = ISL_FORMAT_R8_UINT;
   init_info.width =
      minify(info->surf.logical_level0_px.width, info->view.base_level);
   init_info.height =
      minify(info->surf.logical_level0_px.height, info->view.base_level);
   init_info.depth = 1;
   init_info.levels = 1;
   init_info.array_len = 1;
   init_info.samples = info->surf.samples;
   init_info.min_pitch = info->surf.row_pitch;
   init_info.usage = info->surf.usage;
   init_info.tiling_flags = 1 << info->surf.tiling;

   isl_surf_init_s(isl_dev, &info->surf, &init_info);
   assert(info->surf.row_pitch == init_info.min_pitch);

   /* The view is also different now. */
   info->view.base_level = 0;
   info->view.levels = 1;
   info->view.base_array_layer = 0;
   info->view.array_len = 1;
}

static void
surf_fake_interleaved_msaa(const struct isl_device *isl_dev,
                           struct brw_blorp_surface_info *info)
{
   assert(info->surf.msaa_layout == ISL_MSAA_LAYOUT_INTERLEAVED);

   /* First, we need to convert it to a simple 1-level 1-layer 2-D surface */
   surf_convert_to_single_slice(isl_dev, info);

   info->surf.logical_level0_px = info->surf.phys_level0_sa;
   info->surf.samples = 1;
   info->surf.msaa_layout = ISL_MSAA_LAYOUT_NONE;
}

static void
surf_retile_w_to_y(const struct isl_device *isl_dev,
                   struct brw_blorp_surface_info *info)
{
   assert(info->surf.tiling == ISL_TILING_W);

   /* First, we need to convert it to a simple 1-level 1-layer 2-D surface */
   surf_convert_to_single_slice(isl_dev, info);

   /* On gen7+, we don't have interleaved multisampling for color render
    * targets so we have to fake it.
    *
    * TODO: Are we sure we don't also need to fake it on gen6?
    */
   if (isl_dev->info->gen > 6 &&
       info->surf.msaa_layout == ISL_MSAA_LAYOUT_INTERLEAVED) {
      info->surf.logical_level0_px = info->surf.phys_level0_sa;
      info->surf.samples = 1;
      info->surf.msaa_layout = ISL_MSAA_LAYOUT_NONE;
   }

   if (isl_dev->info->gen == 6) {
      /* Gen6 stencil buffers have a very large alignment coming in from the
       * miptree.  It's out-of-bounds for what the surface state can handle.
       * Since we have a single layer and level, it doesn't really matter as
       * long as we don't pass a bogus value into isl_surf_fill_state().
       */
      info->surf.image_alignment_el = isl_extent3d(4, 2, 1);
   }

   /* Now that we've converted everything to a simple 2-D surface with only
    * one miplevel, we can go about retiling it.
    */
   const unsigned x_align = 8, y_align = info->surf.samples != 0 ? 8 : 4;
   info->surf.tiling = ISL_TILING_Y0;
   info->surf.logical_level0_px.width =
      ALIGN(info->surf.logical_level0_px.width, x_align) * 2;
   info->surf.logical_level0_px.height =
      ALIGN(info->surf.logical_level0_px.height, y_align) / 2;
   info->tile_x_sa *= 2;
   info->tile_y_sa /= 2;
}

void
blorp_blit(struct blorp_batch *batch,
           const struct blorp_surf *src_surf,
           unsigned src_level, unsigned src_layer,
           enum isl_format src_format, struct isl_swizzle src_swizzle,
           const struct blorp_surf *dst_surf,
           unsigned dst_level, unsigned dst_layer,
           enum isl_format dst_format,
           float src_x0, float src_y0,
           float src_x1, float src_y1,
           float dst_x0, float dst_y0,
           float dst_x1, float dst_y1,
           GLenum filter, bool mirror_x, bool mirror_y)
{
   const struct gen_device_info *devinfo = batch->blorp->isl_dev->info;

   struct blorp_params params;
   blorp_params_init(&params);

   brw_blorp_surface_info_init(batch->blorp, &params.src, src_surf, src_level,
                               src_layer, src_format, false);
   brw_blorp_surface_info_init(batch->blorp, &params.dst, dst_surf, dst_level,
                               dst_layer, dst_format, true);

   struct brw_blorp_blit_prog_key wm_prog_key;
   memset(&wm_prog_key, 0, sizeof(wm_prog_key));

   if (isl_format_has_sint_channel(params.src.view.format)) {
      wm_prog_key.texture_data_type = nir_type_int;
   } else if (isl_format_has_uint_channel(params.src.view.format)) {
      wm_prog_key.texture_data_type = nir_type_uint;
   } else {
      wm_prog_key.texture_data_type = nir_type_float;
   }

   /* Scaled blitting or not. */
   wm_prog_key.blit_scaled =
      ((dst_x1 - dst_x0) == (src_x1 - src_x0) &&
       (dst_y1 - dst_y0) == (src_y1 - src_y0)) ? false : true;

   /* Scaling factors used for bilinear filtering in multisample scaled
    * blits.
    */
   if (params.src.surf.samples == 16)
      wm_prog_key.x_scale = 4.0f;
   else
      wm_prog_key.x_scale = 2.0f;
   wm_prog_key.y_scale = params.src.surf.samples / wm_prog_key.x_scale;

   if (filter == GL_LINEAR &&
       params.src.surf.samples <= 1 && params.dst.surf.samples <= 1)
      wm_prog_key.bilinear_filter = true;

   if ((params.src.surf.usage & ISL_SURF_USAGE_DEPTH_BIT) == 0 &&
       (params.src.surf.usage & ISL_SURF_USAGE_STENCIL_BIT) == 0 &&
       !isl_format_has_int_channel(params.src.surf.format) &&
       params.src.surf.samples > 1 && params.dst.surf.samples <= 1) {
      /* We are downsampling a non-integer color buffer, so blend.
       *
       * Regarding integer color buffers, the OpenGL ES 3.2 spec says:
       *
       *    "If the source formats are integer types or stencil values, a
       *    single sample's value is selected for each pixel."
       *
       * This implies we should not blend in that case.
       */
      wm_prog_key.blend = true;
   }

   /* src_samples and dst_samples are the true sample counts */
   wm_prog_key.src_samples = params.src.surf.samples;
   wm_prog_key.dst_samples = params.dst.surf.samples;

   wm_prog_key.tex_aux_usage = params.src.aux_usage;

   /* src_layout and dst_layout indicate the true MSAA layout used by src and
    * dst.
    */
   wm_prog_key.src_layout = params.src.surf.msaa_layout;
   wm_prog_key.dst_layout = params.dst.surf.msaa_layout;

   /* Round floating point values to nearest integer to avoid "off by one texel"
    * kind of errors when blitting.
    */
   params.x0 = params.wm_inputs.discard_rect.x0 = roundf(dst_x0);
   params.y0 = params.wm_inputs.discard_rect.y0 = roundf(dst_y0);
   params.x1 = params.wm_inputs.discard_rect.x1 = roundf(dst_x1);
   params.y1 = params.wm_inputs.discard_rect.y1 = roundf(dst_y1);

   params.wm_inputs.rect_grid.x1 =
      minify(params.src.surf.logical_level0_px.width, src_level) *
      wm_prog_key.x_scale - 1.0f;
   params.wm_inputs.rect_grid.y1 =
      minify(params.src.surf.logical_level0_px.height, src_level) *
      wm_prog_key.y_scale - 1.0f;

   brw_blorp_setup_coord_transform(&params.wm_inputs.coord_transform[0],
                                   src_x0, src_x1, dst_x0, dst_x1, mirror_x);
   brw_blorp_setup_coord_transform(&params.wm_inputs.coord_transform[1],
                                   src_y0, src_y1, dst_y0, dst_y1, mirror_y);

   /* For some texture types, we need to pass the layer through the sampler. */
   params.wm_inputs.src_z = params.src.z_offset;

   if (devinfo->gen > 6 &&
       params.dst.surf.msaa_layout == ISL_MSAA_LAYOUT_INTERLEAVED) {
      assert(params.dst.surf.samples > 1);

      /* We must expand the rectangle we send through the rendering pipeline,
       * to account for the fact that we are mapping the destination region as
       * single-sampled when it is in fact multisampled.  We must also align
       * it to a multiple of the multisampling pattern, because the
       * differences between multisampled and single-sampled surface formats
       * will mean that pixels are scrambled within the multisampling pattern.
       * TODO: what if this makes the coordinates too large?
       *
       * Note: this only works if the destination surface uses the IMS layout.
       * If it's UMS, then we have no choice but to set up the rendering
       * pipeline as multisampled.
       */
      switch (params.dst.surf.samples) {
      case 2:
         params.x0 = ROUND_DOWN_TO(params.x0 * 2, 4);
         params.y0 = ROUND_DOWN_TO(params.y0, 4);
         params.x1 = ALIGN(params.x1 * 2, 4);
         params.y1 = ALIGN(params.y1, 4);
         break;
      case 4:
         params.x0 = ROUND_DOWN_TO(params.x0 * 2, 4);
         params.y0 = ROUND_DOWN_TO(params.y0 * 2, 4);
         params.x1 = ALIGN(params.x1 * 2, 4);
         params.y1 = ALIGN(params.y1 * 2, 4);
         break;
      case 8:
         params.x0 = ROUND_DOWN_TO(params.x0 * 4, 8);
         params.y0 = ROUND_DOWN_TO(params.y0 * 2, 4);
         params.x1 = ALIGN(params.x1 * 4, 8);
         params.y1 = ALIGN(params.y1 * 2, 4);
         break;
      case 16:
         params.x0 = ROUND_DOWN_TO(params.x0 * 4, 8);
         params.y0 = ROUND_DOWN_TO(params.y0 * 4, 8);
         params.x1 = ALIGN(params.x1 * 4, 8);
         params.y1 = ALIGN(params.y1 * 4, 8);
         break;
      default:
         unreachable("Unrecognized sample count in brw_blorp_blit_params ctor");
      }

      surf_fake_interleaved_msaa(batch->blorp->isl_dev, &params.dst);

      wm_prog_key.use_kill = true;
   }

   if (params.dst.surf.tiling == ISL_TILING_W) {
      /* We must modify the rectangle we send through the rendering pipeline
       * (and the size and x/y offset of the destination surface), to account
       * for the fact that we are mapping it as Y-tiled when it is in fact
       * W-tiled.
       *
       * Both Y tiling and W tiling can be understood as organizations of
       * 32-byte sub-tiles; within each 32-byte sub-tile, the layout of pixels
       * is different, but the layout of the 32-byte sub-tiles within the 4k
       * tile is the same (8 sub-tiles across by 16 sub-tiles down, in
       * column-major order).  In Y tiling, the sub-tiles are 16 bytes wide
       * and 2 rows high; in W tiling, they are 8 bytes wide and 4 rows high.
       *
       * Therefore, to account for the layout differences within the 32-byte
       * sub-tiles, we must expand the rectangle so the X coordinates of its
       * edges are multiples of 8 (the W sub-tile width), and its Y
       * coordinates of its edges are multiples of 4 (the W sub-tile height).
       * Then we need to scale the X and Y coordinates of the rectangle to
       * account for the differences in aspect ratio between the Y and W
       * sub-tiles.  We need to modify the layer width and height similarly.
       *
       * A correction needs to be applied when MSAA is in use: since
       * INTEL_MSAA_LAYOUT_IMS uses an interleaving pattern whose height is 4,
       * we need to align the Y coordinates to multiples of 8, so that when
       * they are divided by two they are still multiples of 4.
       *
       * Note: Since the x/y offset of the surface will be applied using the
       * SURFACE_STATE command packet, it will be invisible to the swizzling
       * code in the shader; therefore it needs to be in a multiple of the
       * 32-byte sub-tile size.  Fortunately it is, since the sub-tile is 8
       * pixels wide and 4 pixels high (when viewed as a W-tiled stencil
       * buffer), and the miplevel alignment used for stencil buffers is 8
       * pixels horizontally and either 4 or 8 pixels vertically (see
       * intel_horizontal_texture_alignment_unit() and
       * intel_vertical_texture_alignment_unit()).
       *
       * Note: Also, since the SURFACE_STATE command packet can only apply
       * offsets that are multiples of 4 pixels horizontally and 2 pixels
       * vertically, it is important that the offsets will be multiples of
       * these sizes after they are converted into Y-tiled coordinates.
       * Fortunately they will be, since we know from above that the offsets
       * are a multiple of the 32-byte sub-tile size, and in Y-tiled
       * coordinates the sub-tile is 16 pixels wide and 2 pixels high.
       *
       * TODO: what if this makes the coordinates (or the texture size) too
       * large?
       */
      const unsigned x_align = 8, y_align = params.dst.surf.samples != 0 ? 8 : 4;
      params.x0 = ROUND_DOWN_TO(params.x0, x_align) * 2;
      params.y0 = ROUND_DOWN_TO(params.y0, y_align) / 2;
      params.x1 = ALIGN(params.x1, x_align) * 2;
      params.y1 = ALIGN(params.y1, y_align) / 2;

      /* Retile the surface to Y-tiled */
      surf_retile_w_to_y(batch->blorp->isl_dev, &params.dst);

      wm_prog_key.dst_tiled_w = true;
      wm_prog_key.use_kill = true;

      if (params.dst.surf.samples > 1) {
         /* If the destination surface is a W-tiled multisampled stencil
          * buffer that we're mapping as Y tiled, then we need to arrange for
          * the WM program to run once per sample rather than once per pixel,
          * because the memory layout of related samples doesn't match between
          * W and Y tiling.
          */
         wm_prog_key.persample_msaa_dispatch = true;
      }
   }

   if (devinfo->gen < 8 && params.src.surf.tiling == ISL_TILING_W) {
      /* On Haswell and earlier, we have to fake W-tiled sources as Y-tiled.
       * Broadwell adds support for sampling from stencil.
       *
       * See the comments above concerning x/y offset alignment for the
       * destination surface.
       *
       * TODO: what if this makes the texture size too large?
       */
      surf_retile_w_to_y(batch->blorp->isl_dev, &params.src);

      wm_prog_key.src_tiled_w = true;
   }

   /* tex_samples and rt_samples are the sample counts that are set up in
    * SURFACE_STATE.
    */
   wm_prog_key.tex_samples = params.src.surf.samples;
   wm_prog_key.rt_samples  = params.dst.surf.samples;

   /* tex_layout and rt_layout indicate the MSAA layout the GPU pipeline will
    * use to access the source and destination surfaces.
    */
   wm_prog_key.tex_layout = params.src.surf.msaa_layout;
   wm_prog_key.rt_layout = params.dst.surf.msaa_layout;

   if (params.src.surf.samples > 0 && params.dst.surf.samples > 1) {
      /* We are blitting from a multisample buffer to a multisample buffer, so
       * we must preserve samples within a pixel.  This means we have to
       * arrange for the WM program to run once per sample rather than once
       * per pixel.
       */
      wm_prog_key.persample_msaa_dispatch = true;
   }

   brw_blorp_get_blit_kernel(batch->blorp, &params, &wm_prog_key);

   params.src.view.swizzle = src_swizzle;

   batch->blorp->exec(batch, &params);
}