summaryrefslogtreecommitdiff
path: root/src/intel/blorp/blorp_clear.c
blob: 202d3839f73f1a0de2feba48db5d9ca06e8eda6f (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
/*
 * Copyright © 2013 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 "util/ralloc.h"

#include "main/macros.h" /* Needed for MAX3 and MAX2 for format_rgb9e5 */
#include "util/format_rgb9e5.h"
#include "util/format_srgb.h"

#include "blorp_priv.h"
#include "compiler/brw_eu_defines.h"

#include "blorp_nir_builder.h"

#define FILE_DEBUG_FLAG DEBUG_BLORP

#pragma pack(push, 1)
struct brw_blorp_const_color_prog_key
{
   enum blorp_shader_type shader_type; /* Must be BLORP_SHADER_TYPE_CLEAR */
   bool use_simd16_replicated_data;
   bool clear_rgb_as_red;
};
#pragma pack(pop)

static bool
blorp_params_get_clear_kernel(struct blorp_batch *batch,
                              struct blorp_params *params,
                              bool use_replicated_data,
                              bool clear_rgb_as_red)
{
   struct blorp_context *blorp = batch->blorp;

   const struct brw_blorp_const_color_prog_key blorp_key = {
      .shader_type = BLORP_SHADER_TYPE_CLEAR,
      .use_simd16_replicated_data = use_replicated_data,
      .clear_rgb_as_red = clear_rgb_as_red,
   };

   if (blorp->lookup_shader(batch, &blorp_key, sizeof(blorp_key),
                            &params->wm_prog_kernel, &params->wm_prog_data))
      return true;

   void *mem_ctx = ralloc_context(NULL);

   nir_builder b;
   blorp_nir_init_shader(&b, mem_ctx, MESA_SHADER_FRAGMENT, "BLORP-clear");

   nir_variable *v_color =
      BLORP_CREATE_NIR_INPUT(b.shader, clear_color, glsl_vec4_type());
   nir_ssa_def *color = nir_load_var(&b, v_color);

   if (clear_rgb_as_red) {
      nir_ssa_def *pos = nir_f2i32(&b, nir_load_frag_coord(&b));
      nir_ssa_def *comp = nir_umod(&b, nir_channel(&b, pos, 0),
                                       nir_imm_int(&b, 3));
      nir_ssa_def *color_component =
         nir_bcsel(&b, nir_ieq(&b, comp, nir_imm_int(&b, 0)),
                       nir_channel(&b, color, 0),
                       nir_bcsel(&b, nir_ieq(&b, comp, nir_imm_int(&b, 1)),
                                     nir_channel(&b, color, 1),
                                     nir_channel(&b, color, 2)));

      nir_ssa_def *u = nir_ssa_undef(&b, 1, 32);
      color = nir_vec4(&b, color_component, u, u, u);
   }

   nir_variable *frag_color = nir_variable_create(b.shader, nir_var_shader_out,
                                                  glsl_vec4_type(),
                                                  "gl_FragColor");
   frag_color->data.location = FRAG_RESULT_COLOR;
   nir_store_var(&b, frag_color, color, 0xf);

   struct brw_wm_prog_key wm_key;
   brw_blorp_init_wm_prog_key(&wm_key);

   struct brw_wm_prog_data prog_data;
   const unsigned *program =
      blorp_compile_fs(blorp, mem_ctx, b.shader, &wm_key, use_replicated_data,
                       &prog_data);

   bool result =
      blorp->upload_shader(batch, MESA_SHADER_FRAGMENT,
                           &blorp_key, sizeof(blorp_key),
                           program, prog_data.base.program_size,
                           &prog_data.base, sizeof(prog_data),
                           &params->wm_prog_kernel, &params->wm_prog_data);

   ralloc_free(mem_ctx);
   return result;
}

#pragma pack(push, 1)
struct layer_offset_vs_key {
   enum blorp_shader_type shader_type;
   unsigned num_inputs;
};
#pragma pack(pop)

/* In the case of doing attachment clears, we are using a surface state that
 * is handed to us so we can't set (and don't even know) the base array layer.
 * In order to do a layered clear in this scenario, we need some way of adding
 * the base array layer to the instance id.  Unfortunately, our hardware has
 * no real concept of "base instance", so we have to do it manually in a
 * vertex shader.
 */
static bool
blorp_params_get_layer_offset_vs(struct blorp_batch *batch,
                                 struct blorp_params *params)
{
   struct blorp_context *blorp = batch->blorp;
   struct layer_offset_vs_key blorp_key = {
      .shader_type = BLORP_SHADER_TYPE_LAYER_OFFSET_VS,
   };

   if (params->wm_prog_data)
      blorp_key.num_inputs = params->wm_prog_data->num_varying_inputs;

   if (blorp->lookup_shader(batch, &blorp_key, sizeof(blorp_key),
                            &params->vs_prog_kernel, &params->vs_prog_data))
      return true;

   void *mem_ctx = ralloc_context(NULL);

   nir_builder b;
   blorp_nir_init_shader(&b, mem_ctx, MESA_SHADER_VERTEX, "BLORP-layer-offset-vs");

   const struct glsl_type *uvec4_type = glsl_vector_type(GLSL_TYPE_UINT, 4);

   /* First we deal with the header which has instance and base instance */
   nir_variable *a_header = nir_variable_create(b.shader, nir_var_shader_in,
                                                uvec4_type, "header");
   a_header->data.location = VERT_ATTRIB_GENERIC0;

   nir_variable *v_layer = nir_variable_create(b.shader, nir_var_shader_out,
                                               glsl_int_type(), "layer_id");
   v_layer->data.location = VARYING_SLOT_LAYER;

   /* Compute the layer id */
   nir_ssa_def *header = nir_load_var(&b, a_header);
   nir_ssa_def *base_layer = nir_channel(&b, header, 0);
   nir_ssa_def *instance = nir_channel(&b, header, 1);
   nir_store_var(&b, v_layer, nir_iadd(&b, instance, base_layer), 0x1);

   /* Then we copy the vertex from the next slot to VARYING_SLOT_POS */
   nir_variable *a_vertex = nir_variable_create(b.shader, nir_var_shader_in,
                                                glsl_vec4_type(), "a_vertex");
   a_vertex->data.location = VERT_ATTRIB_GENERIC1;

   nir_variable *v_pos = nir_variable_create(b.shader, nir_var_shader_out,
                                             glsl_vec4_type(), "v_pos");
   v_pos->data.location = VARYING_SLOT_POS;

   nir_copy_var(&b, v_pos, a_vertex);

   /* Then we copy everything else */
   for (unsigned i = 0; i < blorp_key.num_inputs; i++) {
      nir_variable *a_in = nir_variable_create(b.shader, nir_var_shader_in,
                                               uvec4_type, "input");
      a_in->data.location = VERT_ATTRIB_GENERIC2 + i;

      nir_variable *v_out = nir_variable_create(b.shader, nir_var_shader_out,
                                                uvec4_type, "output");
      v_out->data.location = VARYING_SLOT_VAR0 + i;

      nir_copy_var(&b, v_out, a_in);
   }

   struct brw_vs_prog_data vs_prog_data;
   memset(&vs_prog_data, 0, sizeof(vs_prog_data));

   const unsigned *program =
      blorp_compile_vs(blorp, mem_ctx, b.shader, &vs_prog_data);

   bool result =
      blorp->upload_shader(batch, MESA_SHADER_VERTEX,
                           &blorp_key, sizeof(blorp_key),
                           program, vs_prog_data.base.base.program_size,
                           &vs_prog_data.base.base, sizeof(vs_prog_data),
                           &params->vs_prog_kernel, &params->vs_prog_data);

   ralloc_free(mem_ctx);
   return result;
}

/* The x0, y0, x1, and y1 parameters must already be populated with the render
 * area of the framebuffer to be cleared.
 */
static void
get_fast_clear_rect(const struct isl_device *dev,
                    const struct isl_surf *aux_surf,
                    unsigned *x0, unsigned *y0,
                    unsigned *x1, unsigned *y1)
{
   unsigned int x_align, y_align;
   unsigned int x_scaledown, y_scaledown;

   /* Only single sampled surfaces need to (and actually can) be resolved. */
   if (aux_surf->usage == ISL_SURF_USAGE_CCS_BIT) {
      /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
       * Target(s)", beneath the "Fast Color Clear" bullet (p327):
       *
       *     Clear pass must have a clear rectangle that must follow
       *     alignment rules in terms of pixels and lines as shown in the
       *     table below. Further, the clear-rectangle height and width
       *     must be multiple of the following dimensions. If the height
       *     and width of the render target being cleared do not meet these
       *     requirements, an MCS buffer can be created such that it
       *     follows the requirement and covers the RT.
       *
       * The alignment size in the table that follows is related to the
       * alignment size that is baked into the CCS surface format but with X
       * alignment multiplied by 16 and Y alignment multiplied by 32.
       */
      x_align = isl_format_get_layout(aux_surf->format)->bw;
      y_align = isl_format_get_layout(aux_surf->format)->bh;

      x_align *= 16;

      /* The line alignment requirement for Y-tiled is halved at SKL and again
       * at TGL.
       */
      if (dev->info->gen >= 12)
         y_align *= 8;
      else if (dev->info->gen >= 9)
         y_align *= 16;
      else
         y_align *= 32;

      /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
       * Target(s)", beneath the "Fast Color Clear" bullet (p327):
       *
       *     In order to optimize the performance MCS buffer (when bound to
       *     1X RT) clear similarly to MCS buffer clear for MSRT case,
       *     clear rect is required to be scaled by the following factors
       *     in the horizontal and vertical directions:
       *
       * The X and Y scale down factors in the table that follows are each
       * equal to half the alignment value computed above.
       */
      x_scaledown = x_align / 2;
      y_scaledown = y_align / 2;

      if (ISL_DEV_IS_HASWELL(dev)) {
         /* From BSpec: 3D-Media-GPGPU Engine > 3D Pipeline > Pixel > Pixel
          * Backend > MCS Buffer for Render Target(s) [DevIVB+] > Table "Color
          * Clear of Non-MultiSampled Render Target Restrictions":
          *
          *   Clear rectangle must be aligned to two times the number of
          *   pixels in the table shown below due to 16x16 hashing across the
          *   slice.
          *
          * This restriction is only documented to exist on HSW GT3 but
          * empirical evidence suggests that it's also needed GT2.
          */
         x_align *= 2;
         y_align *= 2;
      }
   } else {
      assert(aux_surf->usage == ISL_SURF_USAGE_MCS_BIT);

      /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
       * Target(s)", beneath the "MSAA Compression" bullet (p326):
       *
       *     Clear pass for this case requires that scaled down primitive
       *     is sent down with upper left co-ordinate to coincide with
       *     actual rectangle being cleared. For MSAA, clear rectangle’s
       *     height and width need to as show in the following table in
       *     terms of (width,height) of the RT.
       *
       *     MSAA  Width of Clear Rect  Height of Clear Rect
       *      2X     Ceil(1/8*width)      Ceil(1/2*height)
       *      4X     Ceil(1/8*width)      Ceil(1/2*height)
       *      8X     Ceil(1/2*width)      Ceil(1/2*height)
       *     16X         width            Ceil(1/2*height)
       *
       * The text "with upper left co-ordinate to coincide with actual
       * rectangle being cleared" is a little confusing--it seems to imply
       * that to clear a rectangle from (x,y) to (x+w,y+h), one needs to
       * feed the pipeline using the rectangle (x,y) to
       * (x+Ceil(w/N),y+Ceil(h/2)), where N is either 2 or 8 depending on
       * the number of samples.  Experiments indicate that this is not
       * quite correct; actually, what the hardware appears to do is to
       * align whatever rectangle is sent down the pipeline to the nearest
       * multiple of 2x2 blocks, and then scale it up by a factor of N
       * horizontally and 2 vertically.  So the resulting alignment is 4
       * vertically and either 4 or 16 horizontally, and the scaledown
       * factor is 2 vertically and either 2 or 8 horizontally.
       */
      switch (aux_surf->format) {
      case ISL_FORMAT_MCS_2X:
      case ISL_FORMAT_MCS_4X:
         x_scaledown = 8;
         break;
      case ISL_FORMAT_MCS_8X:
         x_scaledown = 2;
         break;
      case ISL_FORMAT_MCS_16X:
         x_scaledown = 1;
         break;
      default:
         unreachable("Unexpected MCS format for fast clear");
      }
      y_scaledown = 2;
      x_align = x_scaledown * 2;
      y_align = y_scaledown * 2;
   }

   *x0 = ROUND_DOWN_TO(*x0,  x_align) / x_scaledown;
   *y0 = ROUND_DOWN_TO(*y0, y_align) / y_scaledown;
   *x1 = ALIGN(*x1, x_align) / x_scaledown;
   *y1 = ALIGN(*y1, y_align) / y_scaledown;
}

void
blorp_fast_clear(struct blorp_batch *batch,
                 const struct blorp_surf *surf,
                 enum isl_format format, struct isl_swizzle swizzle,
                 uint32_t level, uint32_t start_layer, uint32_t num_layers,
                 uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1)
{
   struct blorp_params params;
   blorp_params_init(&params);
   params.num_layers = num_layers;

   params.x0 = x0;
   params.y0 = y0;
   params.x1 = x1;
   params.y1 = y1;

   memset(&params.wm_inputs.clear_color, 0xff, 4*sizeof(float));
   params.fast_clear_op = ISL_AUX_OP_FAST_CLEAR;

   get_fast_clear_rect(batch->blorp->isl_dev, surf->aux_surf,
                       &params.x0, &params.y0, &params.x1, &params.y1);

   if (!blorp_params_get_clear_kernel(batch, &params, true, false))
      return;

   brw_blorp_surface_info_init(batch->blorp, &params.dst, surf, level,
                               start_layer, format, true);
   params.num_samples = params.dst.surf.samples;

   /* If a swizzle was provided, we need to swizzle the clear color so that
    * the hardware color format conversion will work properly.
    */
   params.dst.clear_color =
      isl_color_value_swizzle_inv(params.dst.clear_color, swizzle);

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

void
blorp_clear(struct blorp_batch *batch,
            const struct blorp_surf *surf,
            enum isl_format format, struct isl_swizzle swizzle,
            uint32_t level, uint32_t start_layer, uint32_t num_layers,
            uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1,
            union isl_color_value clear_color,
            const bool color_write_disable[4])
{
   struct blorp_params params;
   blorp_params_init(&params);

   /* Manually apply the clear destination swizzle.  This way swizzled clears
    * will work for swizzles which we can't normally use for rendering and it
    * also ensures that they work on pre-Haswell hardware which can't swizlle
    * at all.
    */
   clear_color = isl_color_value_swizzle_inv(clear_color, swizzle);
   swizzle = ISL_SWIZZLE_IDENTITY;

   bool clear_rgb_as_red = false;
   if (format == ISL_FORMAT_R9G9B9E5_SHAREDEXP) {
      clear_color.u32[0] = float3_to_rgb9e5(clear_color.f32);
      format = ISL_FORMAT_R32_UINT;
   } else if (format == ISL_FORMAT_L8_UNORM_SRGB) {
      clear_color.f32[0] = util_format_linear_to_srgb_float(clear_color.f32[0]);
      format = ISL_FORMAT_R8_UNORM;
   } else if (format == ISL_FORMAT_A4B4G4R4_UNORM) {
      /* Broadwell and earlier cannot render to this format so we need to work
       * around it by swapping the colors around and using B4G4R4A4 instead.
       */
      const struct isl_swizzle ARGB = ISL_SWIZZLE(ALPHA, RED, GREEN, BLUE);
      clear_color = isl_color_value_swizzle_inv(clear_color, ARGB);
      format = ISL_FORMAT_B4G4R4A4_UNORM;
   } else if (isl_format_get_layout(format)->bpb % 3 == 0) {
      clear_rgb_as_red = true;
      if (format == ISL_FORMAT_R8G8B8_UNORM_SRGB) {
         clear_color.f32[0] = util_format_linear_to_srgb_float(clear_color.f32[0]);
         clear_color.f32[1] = util_format_linear_to_srgb_float(clear_color.f32[1]);
         clear_color.f32[2] = util_format_linear_to_srgb_float(clear_color.f32[2]);
      }
   }

   memcpy(&params.wm_inputs.clear_color, clear_color.f32, sizeof(float) * 4);

   bool use_simd16_replicated_data = true;

   /* From the SNB PRM (Vol4_Part1):
    *
    *     "Replicated data (Message Type = 111) is only supported when
    *      accessing tiled memory.  Using this Message Type to access linear
    *      (untiled) memory is UNDEFINED."
    */
   if (surf->surf->tiling == ISL_TILING_LINEAR)
      use_simd16_replicated_data = false;

   /* Replicated clears don't work yet before gen6 */
   if (batch->blorp->isl_dev->info->gen < 6)
      use_simd16_replicated_data = false;

   /* Constant color writes ignore everyting in blend and color calculator
    * state.  This is not documented.
    */
   if (color_write_disable) {
      for (unsigned i = 0; i < 4; i++) {
         params.color_write_disable[i] = color_write_disable[i];
         if (color_write_disable[i])
            use_simd16_replicated_data = false;
      }
   }

   if (!blorp_params_get_clear_kernel(batch, &params,
                                      use_simd16_replicated_data,
                                      clear_rgb_as_red))
      return;

   if (!blorp_ensure_sf_program(batch, &params))
      return;

   while (num_layers > 0) {
      brw_blorp_surface_info_init(batch->blorp, &params.dst, surf, level,
                                  start_layer, format, true);
      params.dst.view.swizzle = swizzle;

      params.x0 = x0;
      params.y0 = y0;
      params.x1 = x1;
      params.y1 = y1;

      if (params.dst.tile_x_sa || params.dst.tile_y_sa) {
         assert(params.dst.surf.samples == 1);
         assert(num_layers == 1);
         params.x0 += params.dst.tile_x_sa;
         params.y0 += params.dst.tile_y_sa;
         params.x1 += params.dst.tile_x_sa;
         params.y1 += params.dst.tile_y_sa;
      }

      /* The MinLOD and MinimumArrayElement don't work properly for cube maps.
       * Convert them to a single slice on gen4.
       */
      if (batch->blorp->isl_dev->info->gen == 4 &&
          (params.dst.surf.usage & ISL_SURF_USAGE_CUBE_BIT)) {
         blorp_surf_convert_to_single_slice(batch->blorp->isl_dev, &params.dst);
      }

      if (clear_rgb_as_red) {
         surf_fake_rgb_with_red(batch->blorp->isl_dev, &params.dst);
         params.x0 *= 3;
         params.x1 *= 3;
      }

      if (isl_format_is_compressed(params.dst.surf.format)) {
         blorp_surf_convert_to_uncompressed(batch->blorp->isl_dev, &params.dst,
                                            NULL, NULL, NULL, NULL);
                                            //&dst_x, &dst_y, &dst_w, &dst_h);
      }

      if (params.dst.tile_x_sa || params.dst.tile_y_sa) {
         /* Either we're on gen4 where there is no multisampling or the
          * surface is compressed which also implies no multisampling.
          * Therefore, sa == px and we don't need to do a conversion.
          */
         assert(params.dst.surf.samples == 1);
         params.x0 += params.dst.tile_x_sa;
         params.y0 += params.dst.tile_y_sa;
         params.x1 += params.dst.tile_x_sa;
         params.y1 += params.dst.tile_y_sa;
      }

      params.num_samples = params.dst.surf.samples;

      /* We may be restricted on the number of layers we can bind at any one
       * time.  In particular, Sandy Bridge has a maximum number of layers of
       * 512 but a maximum 3D texture size is much larger.
       */
      params.num_layers = MIN2(params.dst.view.array_len, num_layers);

      const unsigned max_image_width = 16 * 1024;
      if (params.dst.surf.logical_level0_px.width > max_image_width) {
         /* Clearing an RGB image as red multiplies the surface width by 3
          * so it may now be too wide for the hardware surface limits.  We
          * have to break the clear up into pieces in order to clear wide
          * images.
          */
         assert(clear_rgb_as_red);
         assert(params.dst.surf.dim == ISL_SURF_DIM_2D);
         assert(params.dst.surf.tiling == ISL_TILING_LINEAR);
         assert(params.dst.surf.logical_level0_px.depth == 1);
         assert(params.dst.surf.logical_level0_px.array_len == 1);
         assert(params.dst.surf.levels == 1);
         assert(params.dst.surf.samples == 1);
         assert(params.dst.tile_x_sa == 0 || params.dst.tile_y_sa == 0);
         assert(params.dst.aux_usage == ISL_AUX_USAGE_NONE);

         /* max_image_width rounded down to a multiple of 3 */
         const unsigned max_fake_rgb_width = (max_image_width / 3) * 3;
         const unsigned cpp =
            isl_format_get_layout(params.dst.surf.format)->bpb / 8;

         params.dst.surf.logical_level0_px.width = max_fake_rgb_width;
         params.dst.surf.phys_level0_sa.width = max_fake_rgb_width;

         uint32_t orig_x0 = params.x0, orig_x1 = params.x1;
         uint64_t orig_offset = params.dst.addr.offset;
         for (uint32_t x = orig_x0; x < orig_x1; x += max_fake_rgb_width) {
            /* Offset to the surface.  It's easy because we're linear */
            params.dst.addr.offset = orig_offset + x * cpp;

            params.x0 = 0;
            params.x1 = MIN2(orig_x1 - x, max_image_width);

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

      start_layer += params.num_layers;
      num_layers -= params.num_layers;
   }
}

static bool
blorp_clear_stencil_as_rgba(struct blorp_batch *batch,
                            const struct blorp_surf *surf,
                            uint32_t level, uint32_t start_layer,
                            uint32_t num_layers,
                            uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1,
                            uint8_t stencil_mask, uint8_t stencil_value)
{
   /* We only support separate W-tiled stencil for now */
   if (surf->surf->format != ISL_FORMAT_R8_UINT ||
       surf->surf->tiling != ISL_TILING_W)
      return false;

   /* Stencil mask support would require piles of shader magic */
   if (stencil_mask != 0xff)
      return false;

   if (surf->surf->samples > 1) {
      /* Adjust x0, y0, x1, and y1 to be in units of samples */
      assert(surf->surf->msaa_layout == ISL_MSAA_LAYOUT_INTERLEAVED);
      struct isl_extent2d msaa_px_size_sa =
         isl_get_interleaved_msaa_px_size_sa(surf->surf->samples);

      x0 *= msaa_px_size_sa.w;
      y0 *= msaa_px_size_sa.h;
      x1 *= msaa_px_size_sa.w;
      y1 *= msaa_px_size_sa.h;
   }

   /* W-tiles and Y-tiles have the same layout as far as cache lines are
    * concerned: both are 8x8 cache lines laid out Y-major.  The difference is
    * entirely in how the data is arranged withing the cache line.  W-tiling
    * is 8x8 pixels in a swizzled pattern while Y-tiling is 16B by 4 rows
    * regardless of image format size.  As long as everything is aligned to 8,
    * we can just treat the W-tiled image as Y-tiled, ignore the layout
    * difference within a cache line, and blast out data.
    */
   if (x0 % 8 != 0 || y0 % 8 != 0 || x1 % 8 != 0 || y1 % 8 != 0)
      return false;

   struct blorp_params params;
   blorp_params_init(&params);

   if (!blorp_params_get_clear_kernel(batch, &params, true, false))
      return false;

   memset(&params.wm_inputs.clear_color, stencil_value,
          sizeof(params.wm_inputs.clear_color));

   /* The Sandy Bridge PRM Vol. 4 Pt. 2, section 2.11.2.1.1 has the
    * following footnote to the format table:
    *
    *    128 BPE Formats cannot be Tiled Y when used as render targets
    *
    * We have to use RGBA16_UINT on SNB.
    */
   enum isl_format wide_format;
   if (ISL_DEV_GEN(batch->blorp->isl_dev) <= 6) {
      wide_format = ISL_FORMAT_R16G16B16A16_UINT;

      /* For RGBA16_UINT, we need to mask the stencil value otherwise, we risk
       * clamping giving us the wrong values
       */
      for (unsigned i = 0; i < 4; i++)
         params.wm_inputs.clear_color[i] &= 0xffff;
   } else {
      wide_format = ISL_FORMAT_R32G32B32A32_UINT;
   }

   for (uint32_t a = 0; a < num_layers; a++) {
      uint32_t layer = start_layer + a;

      brw_blorp_surface_info_init(batch->blorp, &params.dst, surf, level,
                                  layer, ISL_FORMAT_UNSUPPORTED, true);

      if (surf->surf->samples > 1)
         blorp_surf_fake_interleaved_msaa(batch->blorp->isl_dev, &params.dst);

      /* Make it Y-tiled */
      blorp_surf_retile_w_to_y(batch->blorp->isl_dev, &params.dst);

      unsigned wide_Bpp =
         isl_format_get_layout(wide_format)->bpb / 8;

      params.dst.view.format = params.dst.surf.format = wide_format;
      assert(params.dst.surf.logical_level0_px.width % wide_Bpp == 0);
      params.dst.surf.logical_level0_px.width /= wide_Bpp;
      assert(params.dst.tile_x_sa % wide_Bpp == 0);
      params.dst.tile_x_sa /= wide_Bpp;

      params.x0 = params.dst.tile_x_sa + x0 / (wide_Bpp / 2);
      params.y0 = params.dst.tile_y_sa + y0 / 2;
      params.x1 = params.dst.tile_x_sa + x1 / (wide_Bpp / 2);
      params.y1 = params.dst.tile_y_sa + y1 / 2;

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

   return true;
}

void
blorp_clear_depth_stencil(struct blorp_batch *batch,
                          const struct blorp_surf *depth,
                          const struct blorp_surf *stencil,
                          uint32_t level, uint32_t start_layer,
                          uint32_t num_layers,
                          uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1,
                          bool clear_depth, float depth_value,
                          uint8_t stencil_mask, uint8_t stencil_value)
{
   if (!clear_depth && blorp_clear_stencil_as_rgba(batch, stencil, level,
                                                   start_layer, num_layers,
                                                   x0, y0, x1, y1,
                                                   stencil_mask,
                                                   stencil_value))
      return;

   struct blorp_params params;
   blorp_params_init(&params);

   params.x0 = x0;
   params.y0 = y0;
   params.x1 = x1;
   params.y1 = y1;

   if (ISL_DEV_GEN(batch->blorp->isl_dev) == 6) {
      /* For some reason, Sandy Bridge gets occlusion queries wrong if we
       * don't have a shader.  In particular, it records samples even though
       * we disable statistics in 3DSTATE_WM.  Give it the usual clear shader
       * to work around the issue.
       */
      if (!blorp_params_get_clear_kernel(batch, &params, false, false))
         return;
   }

   while (num_layers > 0) {
      params.num_layers = num_layers;

      if (stencil_mask) {
         brw_blorp_surface_info_init(batch->blorp, &params.stencil, stencil,
                                     level, start_layer,
                                     ISL_FORMAT_UNSUPPORTED, true);
         params.stencil_mask = stencil_mask;
         params.stencil_ref = stencil_value;

         params.dst.surf.samples = params.stencil.surf.samples;
         params.dst.surf.logical_level0_px =
            params.stencil.surf.logical_level0_px;
         params.dst.view = params.stencil.view;

         params.num_samples = params.stencil.surf.samples;

         /* We may be restricted on the number of layers we can bind at any
          * one time.  In particular, Sandy Bridge has a maximum number of
          * layers of 512 but a maximum 3D texture size is much larger.
          */
         if (params.stencil.view.array_len < params.num_layers)
            params.num_layers = params.stencil.view.array_len;
      }

      if (clear_depth) {
         brw_blorp_surface_info_init(batch->blorp, &params.depth, depth,
                                     level, start_layer,
                                     ISL_FORMAT_UNSUPPORTED, true);
         params.z = depth_value;
         params.depth_format =
            isl_format_get_depth_format(depth->surf->format, false);

         params.dst.surf.samples = params.depth.surf.samples;
         params.dst.surf.logical_level0_px =
            params.depth.surf.logical_level0_px;
         params.dst.view = params.depth.view;

         params.num_samples = params.depth.surf.samples;

         /* We may be restricted on the number of layers we can bind at any
          * one time.  In particular, Sandy Bridge has a maximum number of
          * layers of 512 but a maximum 3D texture size is much larger.
          */
         if (params.depth.view.array_len < params.num_layers)
            params.num_layers = params.depth.view.array_len;
      }

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

      start_layer += params.num_layers;
      num_layers -= params.num_layers;
   }
}

bool
blorp_can_hiz_clear_depth(const struct gen_device_info *devinfo,
                          const struct isl_surf *surf,
                          enum isl_aux_usage aux_usage,
                          uint32_t level, uint32_t layer,
                          uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1)
{
   /* This function currently doesn't support any gen prior to gen8 */
   assert(devinfo->gen >= 8);

   if (devinfo->gen == 8 && surf->format == ISL_FORMAT_R16_UNORM) {
      /* Apply the D16 alignment restrictions. On BDW, HiZ has an 8x4 sample
       * block with the following property: as the number of samples increases,
       * the number of pixels representable by this block decreases by a factor
       * of the sample dimensions. Sample dimensions scale following the MSAA
       * interleaved pattern.
       *
       * Sample|Sample|Pixel
       * Count |Dim   |Dim
       * ===================
       *    1  | 1x1  | 8x4
       *    2  | 2x1  | 4x4
       *    4  | 2x2  | 4x2
       *    8  | 4x2  | 2x2
       *   16  | 4x4  | 2x1
       *
       * Table: Pixel Dimensions in a HiZ Sample Block Pre-SKL
       */
      const struct isl_extent2d sa_block_dim =
         isl_get_interleaved_msaa_px_size_sa(surf->samples);
      const uint8_t align_px_w = 8 / sa_block_dim.w;
      const uint8_t align_px_h = 4 / sa_block_dim.h;

      /* Fast depth clears clear an entire sample block at a time. As a result,
       * the rectangle must be aligned to the dimensions of the encompassing
       * pixel block for a successful operation.
       *
       * Fast clears can still work if the upper-left corner is aligned and the
       * bottom-rigtht corner touches the edge of a depth buffer whose extent
       * is unaligned. This is because each miplevel in the depth buffer is
       * padded by the Pixel Dim (similar to a standard compressed texture).
       * In this case, the clear rectangle could be padded by to match the full
       * depth buffer extent but to support multiple clearing techniques, we
       * chose to be unaware of the depth buffer's extent and thus don't handle
       * this case.
       */
      if (x0 % align_px_w || y0 % align_px_h ||
          x1 % align_px_w || y1 % align_px_h)
         return false;
   } else if (aux_usage == ISL_AUX_USAGE_HIZ_CCS_WT) {
      /* We have to set the WM_HZ_OP::FullSurfaceDepthandStencilClear bit
       * whenever we clear an uninitialized HIZ buffer (as some drivers
       * currently do). However, this bit seems liable to clear 16x8 pixels in
       * the ZCS on Gen12 - greater than the slice alignments for depth
       * buffers.
       */
      assert(surf->image_alignment_el.w % 16 != 0 ||
             surf->image_alignment_el.h % 8 != 0);

      /* This is the hypothesis behind some corruption that was seen with the
       * amd_vertex_shader_layer-layered-depth-texture-render piglit test.
       *
       * From the Compressed Depth Buffers section of the Bspec, under the
       * Gen12 texture performant and ZCS columns:
       *
       *    Update with clear at either 16x8 or 8x4 granularity, based on
       *    fs_clr or otherwise.
       *
       * There are a number of ways to avoid full surface CCS clears that
       * overlap other slices, but for now we choose to disable fast-clears
       * when an initializing clear could hit another miplevel.
       *
       * NOTE: Because the CCS compresses the depth buffer and not a version
       * of it that has been rearranged with different alignments (like Gen8+
       * HIZ), we have to make sure that the x0 and y0 are at least 16x8
       * aligned in the context of the entire surface.
       */
      uint32_t slice_x0, slice_y0;
      isl_surf_get_image_offset_el(surf, level,
                                   surf->dim == ISL_SURF_DIM_3D ? 0 : layer,
                                   surf->dim == ISL_SURF_DIM_3D ? layer: 0,
                                   &slice_x0, &slice_y0);
      const bool max_x1_y1 =
         x1 == minify(surf->logical_level0_px.width, level) && 
         y1 == minify(surf->logical_level0_px.height, level);
      const uint32_t haligned_x1 = ALIGN(x1, surf->image_alignment_el.w);
      const uint32_t valigned_y1 = ALIGN(y1, surf->image_alignment_el.h);
      const bool unaligned = (slice_x0 + x0) % 16 || (slice_y0 + y0) % 8 ||
                             (max_x1_y1 ? haligned_x1 % 16 || valigned_y1 % 8 :
                              x1 % 16 || y1 % 8);
      const bool partial_clear = x0 > 0 || y0 > 0 || !max_x1_y1;
      const bool multislice_surf = surf->levels > 1 ||
                                   surf->logical_level0_px.depth > 1 ||
                                   surf->logical_level0_px.array_len > 1;

      if (unaligned && (partial_clear || multislice_surf))
         return false;
   }

   return isl_aux_usage_has_hiz(aux_usage);
}

void
blorp_hiz_clear_depth_stencil(struct blorp_batch *batch,
                              const struct blorp_surf *depth,
                              const struct blorp_surf *stencil,
                              uint32_t level,
                              uint32_t start_layer, uint32_t num_layers,
                              uint32_t x0, uint32_t y0,
                              uint32_t x1, uint32_t y1,
                              bool clear_depth, float depth_value,
                              bool clear_stencil, uint8_t stencil_value)
{
   struct blorp_params params;
   blorp_params_init(&params);

   /* This requires WM_HZ_OP which only exists on gen8+ */
   assert(ISL_DEV_GEN(batch->blorp->isl_dev) >= 8);

   params.hiz_op = ISL_AUX_OP_FAST_CLEAR;
   params.num_layers = 1;

   params.x0 = x0;
   params.y0 = y0;
   params.x1 = x1;
   params.y1 = y1;

   for (uint32_t l = 0; l < num_layers; l++) {
      const uint32_t layer = start_layer + l;
      if (clear_stencil) {
         brw_blorp_surface_info_init(batch->blorp, &params.stencil, stencil,
                                     level, layer,
                                     ISL_FORMAT_UNSUPPORTED, true);
         params.stencil_mask = 0xff;
         params.stencil_ref = stencil_value;
         params.num_samples = params.stencil.surf.samples;
      }

      if (clear_depth) {
         /* If we're clearing depth, we must have HiZ */
         assert(depth && isl_aux_usage_has_hiz(depth->aux_usage));

         brw_blorp_surface_info_init(batch->blorp, &params.depth, depth,
                                     level, layer,
                                     ISL_FORMAT_UNSUPPORTED, true);
         params.depth.clear_color.f32[0] = depth_value;
         params.depth_format =
            isl_format_get_depth_format(depth->surf->format, false);
         params.num_samples = params.depth.surf.samples;
      }

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

/* Given a depth stencil attachment, this function performs a fast depth clear
 * on a depth portion and a regular clear on the stencil portion. When
 * performing a fast depth clear on the depth portion, the HiZ buffer is simply
 * tagged as cleared so the depth clear value is not actually needed.
 */
void
blorp_gen8_hiz_clear_attachments(struct blorp_batch *batch,
                                 uint32_t num_samples,
                                 uint32_t x0, uint32_t y0,
                                 uint32_t x1, uint32_t y1,
                                 bool clear_depth, bool clear_stencil,
                                 uint8_t stencil_value)
{
   assert(batch->flags & BLORP_BATCH_NO_EMIT_DEPTH_STENCIL);

   struct blorp_params params;
   blorp_params_init(&params);
   params.num_layers = 1;
   params.hiz_op = ISL_AUX_OP_FAST_CLEAR;
   params.x0 = x0;
   params.y0 = y0;
   params.x1 = x1;
   params.y1 = y1;
   params.num_samples = num_samples;
   params.depth.enabled = clear_depth;
   params.stencil.enabled = clear_stencil;
   params.stencil_ref = stencil_value;
   batch->blorp->exec(batch, &params);
}

/** Clear active color/depth/stencili attachments
 *
 * This function performs a clear operation on the currently bound
 * color/depth/stencil attachments.  It is assumed that any information passed
 * in here is valid, consistent, and in-bounds relative to the currently
 * attached depth/stencil.  The binding_table_offset parameter is the 32-bit
 * offset relative to surface state base address where pre-baked binding table
 * that we are to use lives.  If clear_color is false, binding_table_offset
 * must point to a binding table with one entry which is a valid null surface
 * that matches the currently bound depth and stencil.
 */
void
blorp_clear_attachments(struct blorp_batch *batch,
                        uint32_t binding_table_offset,
                        enum isl_format depth_format,
                        uint32_t num_samples,
                        uint32_t start_layer, uint32_t num_layers,
                        uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1,
                        bool clear_color, union isl_color_value color_value,
                        bool clear_depth, float depth_value,
                        uint8_t stencil_mask, uint8_t stencil_value)
{
   struct blorp_params params;
   blorp_params_init(&params);

   assert(batch->flags & BLORP_BATCH_NO_EMIT_DEPTH_STENCIL);

   params.x0 = x0;
   params.y0 = y0;
   params.x1 = x1;
   params.y1 = y1;

   params.use_pre_baked_binding_table = true;
   params.pre_baked_binding_table_offset = binding_table_offset;

   params.num_layers = num_layers;
   params.num_samples = num_samples;

   if (clear_color) {
      params.dst.enabled = true;

      memcpy(&params.wm_inputs.clear_color, color_value.f32, sizeof(float) * 4);

      /* Unfortunately, without knowing whether or not our destination surface
       * is tiled or not, we have to assume it may be linear.  This means no
       * SIMD16_REPDATA for us. :-(
       */
      if (!blorp_params_get_clear_kernel(batch, &params, false, false))
         return;
   }

   if (clear_depth) {
      params.depth.enabled = true;

      params.z = depth_value;
      params.depth_format = isl_format_get_depth_format(depth_format, false);
   }

   if (stencil_mask) {
      params.stencil.enabled = true;

      params.stencil_mask = stencil_mask;
      params.stencil_ref = stencil_value;
   }

   if (!blorp_params_get_layer_offset_vs(batch, &params))
      return;

   params.vs_inputs.base_layer = start_layer;

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

void
blorp_ccs_resolve(struct blorp_batch *batch,
                  struct blorp_surf *surf, uint32_t level,
                  uint32_t start_layer, uint32_t num_layers,
                  enum isl_format format,
                  enum isl_aux_op resolve_op)
{
   struct blorp_params params;

   blorp_params_init(&params);
   brw_blorp_surface_info_init(batch->blorp, &params.dst, surf,
                               level, start_layer, format, true);

   /* From the Ivy Bridge PRM, Vol2 Part1 11.9 "Render Target Resolve":
    *
    *     A rectangle primitive must be scaled down by the following factors
    *     with respect to render target being resolved.
    *
    * The scaledown factors in the table that follows are related to the block
    * size of the CCS format.  For IVB and HSW, we divide by two, for BDW we
    * multiply by 8 and 16. On Sky Lake, we multiply by 8.
    */
   const struct isl_format_layout *aux_fmtl =
      isl_format_get_layout(params.dst.aux_surf.format);
   assert(aux_fmtl->txc == ISL_TXC_CCS);

   unsigned x_scaledown, y_scaledown;
   if (ISL_DEV_GEN(batch->blorp->isl_dev) >= 12) {
      x_scaledown = aux_fmtl->bw * 8;
      y_scaledown = aux_fmtl->bh * 4;
   } else if (ISL_DEV_GEN(batch->blorp->isl_dev) >= 9) {
      x_scaledown = aux_fmtl->bw * 8;
      y_scaledown = aux_fmtl->bh * 8;
   } else if (ISL_DEV_GEN(batch->blorp->isl_dev) >= 8) {
      x_scaledown = aux_fmtl->bw * 8;
      y_scaledown = aux_fmtl->bh * 16;
   } else {
      x_scaledown = aux_fmtl->bw / 2;
      y_scaledown = aux_fmtl->bh / 2;
   }
   params.x0 = params.y0 = 0;
   params.x1 = minify(params.dst.surf.logical_level0_px.width, level);
   params.y1 = minify(params.dst.surf.logical_level0_px.height, level);
   params.x1 = ALIGN(params.x1, x_scaledown) / x_scaledown;
   params.y1 = ALIGN(params.y1, y_scaledown) / y_scaledown;

   if (batch->blorp->isl_dev->info->gen >= 10) {
      assert(resolve_op == ISL_AUX_OP_FULL_RESOLVE ||
             resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE ||
             resolve_op == ISL_AUX_OP_AMBIGUATE);
   } else if (batch->blorp->isl_dev->info->gen >= 9) {
      assert(resolve_op == ISL_AUX_OP_FULL_RESOLVE ||
             resolve_op == ISL_AUX_OP_PARTIAL_RESOLVE);
   } else {
      /* Broadwell and earlier do not have a partial resolve */
      assert(resolve_op == ISL_AUX_OP_FULL_RESOLVE);
   }
   params.fast_clear_op = resolve_op;
   params.num_layers = num_layers;

   /* Note: there is no need to initialize push constants because it doesn't
    * matter what data gets dispatched to the render target.  However, we must
    * ensure that the fragment shader delivers the data using the "replicated
    * color" message.
    */

   if (!blorp_params_get_clear_kernel(batch, &params, true, false))
      return;

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

static nir_ssa_def *
blorp_nir_bit(nir_builder *b, nir_ssa_def *src, unsigned bit)
{
   return nir_iand(b, nir_ushr(b, src, nir_imm_int(b, bit)),
                      nir_imm_int(b, 1));
}

#pragma pack(push, 1)
struct blorp_mcs_partial_resolve_key
{
   enum blorp_shader_type shader_type;
   bool indirect_clear_color;
   bool int_format;
   uint32_t num_samples;
};
#pragma pack(pop)

static bool
blorp_params_get_mcs_partial_resolve_kernel(struct blorp_batch *batch,
                                            struct blorp_params *params)
{
   struct blorp_context *blorp = batch->blorp;
   const struct blorp_mcs_partial_resolve_key blorp_key = {
      .shader_type = BLORP_SHADER_TYPE_MCS_PARTIAL_RESOLVE,
      .indirect_clear_color = params->dst.clear_color_addr.buffer != NULL,
      .int_format = isl_format_has_int_channel(params->dst.view.format),
      .num_samples = params->num_samples,
   };

   if (blorp->lookup_shader(batch, &blorp_key, sizeof(blorp_key),
                            &params->wm_prog_kernel, &params->wm_prog_data))
      return true;

   void *mem_ctx = ralloc_context(NULL);

   nir_builder b;
   blorp_nir_init_shader(&b, mem_ctx, MESA_SHADER_FRAGMENT,
                         "BLORP-mcs-partial-resolve");

   nir_variable *v_color =
      BLORP_CREATE_NIR_INPUT(b.shader, clear_color, glsl_vec4_type());

   nir_variable *frag_color =
      nir_variable_create(b.shader, nir_var_shader_out,
                          glsl_vec4_type(), "gl_FragColor");
   frag_color->data.location = FRAG_RESULT_COLOR;

   /* Do an MCS fetch and check if it is equal to the magic clear value */
   nir_ssa_def *mcs =
      blorp_nir_txf_ms_mcs(&b, nir_f2i32(&b, nir_load_frag_coord(&b)),
                               nir_load_layer_id(&b));
   nir_ssa_def *is_clear =
      blorp_nir_mcs_is_clear_color(&b, mcs, blorp_key.num_samples);

   /* If we aren't the clear value, discard. */
   nir_intrinsic_instr *discard =
      nir_intrinsic_instr_create(b.shader, nir_intrinsic_discard_if);
   discard->src[0] = nir_src_for_ssa(nir_inot(&b, is_clear));
   nir_builder_instr_insert(&b, &discard->instr);

   nir_ssa_def *clear_color = nir_load_var(&b, v_color);
   if (blorp_key.indirect_clear_color && blorp->isl_dev->info->gen <= 8) {
      /* Gen7-8 clear colors are stored as single 0/1 bits */
      clear_color = nir_vec4(&b, blorp_nir_bit(&b, clear_color, 31),
                                 blorp_nir_bit(&b, clear_color, 30),
                                 blorp_nir_bit(&b, clear_color, 29),
                                 blorp_nir_bit(&b, clear_color, 28));

      if (!blorp_key.int_format)
         clear_color = nir_i2f32(&b, clear_color);
   }
   nir_store_var(&b, frag_color, clear_color, 0xf);

   struct brw_wm_prog_key wm_key;
   brw_blorp_init_wm_prog_key(&wm_key);
   wm_key.base.tex.compressed_multisample_layout_mask = 1;
   wm_key.base.tex.msaa_16 = blorp_key.num_samples == 16;
   wm_key.multisample_fbo = true;

   struct brw_wm_prog_data prog_data;
   const unsigned *program =
      blorp_compile_fs(blorp, mem_ctx, b.shader, &wm_key, false,
                       &prog_data);

   bool result =
      blorp->upload_shader(batch, MESA_SHADER_FRAGMENT,
                           &blorp_key, sizeof(blorp_key),
                           program, prog_data.base.program_size,
                           &prog_data.base, sizeof(prog_data),
                           &params->wm_prog_kernel, &params->wm_prog_data);

   ralloc_free(mem_ctx);
   return result;
}

void
blorp_mcs_partial_resolve(struct blorp_batch *batch,
                          struct blorp_surf *surf,
                          enum isl_format format,
                          uint32_t start_layer, uint32_t num_layers)
{
   struct blorp_params params;
   blorp_params_init(&params);

   assert(batch->blorp->isl_dev->info->gen >= 7);

   params.x0 = 0;
   params.y0 = 0;
   params.x1 = surf->surf->logical_level0_px.width;
   params.y1 = surf->surf->logical_level0_px.height;

   brw_blorp_surface_info_init(batch->blorp, &params.src, surf, 0,
                               start_layer, format, false);
   brw_blorp_surface_info_init(batch->blorp, &params.dst, surf, 0,
                               start_layer, format, true);

   params.num_samples = params.dst.surf.samples;
   params.num_layers = num_layers;
   params.dst_clear_color_as_input = surf->clear_color_addr.buffer != NULL;

   memcpy(&params.wm_inputs.clear_color,
          surf->clear_color.f32, sizeof(float) * 4);

   if (!blorp_params_get_mcs_partial_resolve_kernel(batch, &params))
      return;

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

/** Clear a CCS to the "uncompressed" state
 *
 * This pass is the CCS equivalent of a "HiZ resolve".  It sets the CCS values
 * for a given layer/level of a surface to 0x0 which is the "uncompressed"
 * state which tells the sampler to go look at the main surface.
 */
void
blorp_ccs_ambiguate(struct blorp_batch *batch,
                    struct blorp_surf *surf,
                    uint32_t level, uint32_t layer)
{
   if (ISL_DEV_GEN(batch->blorp->isl_dev) >= 10) {
      /* On gen10 and above, we have a hardware resolve op for this */
      return blorp_ccs_resolve(batch, surf, level, layer, 1,
                               surf->surf->format, ISL_AUX_OP_AMBIGUATE);
   }

   struct blorp_params params;
   blorp_params_init(&params);

   assert(ISL_DEV_GEN(batch->blorp->isl_dev) >= 7);

   const struct isl_format_layout *aux_fmtl =
      isl_format_get_layout(surf->aux_surf->format);
   assert(aux_fmtl->txc == ISL_TXC_CCS);

   params.dst = (struct brw_blorp_surface_info) {
      .enabled = true,
      .addr = surf->aux_addr,
      .view = {
         .usage = ISL_SURF_USAGE_RENDER_TARGET_BIT,
         .format = ISL_FORMAT_R32G32B32A32_UINT,
         .base_level = 0,
         .base_array_layer = 0,
         .levels = 1,
         .array_len = 1,
         .swizzle = ISL_SWIZZLE_IDENTITY,
      },
   };

   uint32_t z = 0;
   if (surf->surf->dim == ISL_SURF_DIM_3D) {
      z = layer;
      layer = 0;
   }

   uint32_t offset_B, x_offset_el, y_offset_el;
   isl_surf_get_image_offset_el(surf->aux_surf, level, layer, z,
                                &x_offset_el, &y_offset_el);
   isl_tiling_get_intratile_offset_el(surf->aux_surf->tiling, aux_fmtl->bpb,
                                      surf->aux_surf->row_pitch_B,
                                      x_offset_el, y_offset_el,
                                      &offset_B, &x_offset_el, &y_offset_el);
   params.dst.addr.offset += offset_B;

   const uint32_t width_px =
      minify(surf->aux_surf->logical_level0_px.width, level);
   const uint32_t height_px =
      minify(surf->aux_surf->logical_level0_px.height, level);
   const uint32_t width_el = DIV_ROUND_UP(width_px, aux_fmtl->bw);
   const uint32_t height_el = DIV_ROUND_UP(height_px, aux_fmtl->bh);

   struct isl_tile_info ccs_tile_info;
   isl_surf_get_tile_info(surf->aux_surf, &ccs_tile_info);

   /* We're going to map it as a regular RGBA32_UINT surface.  We need to
    * downscale a good deal.  We start by computing the area on the CCS to
    * clear in units of Y-tiled cache lines.
    */
   uint32_t x_offset_cl, y_offset_cl, width_cl, height_cl;
   if (ISL_DEV_GEN(batch->blorp->isl_dev) >= 8) {
      /* From the Sky Lake PRM Vol. 12 in the section on planes:
       *
       *    "The Color Control Surface (CCS) contains the compression status
       *    of the cache-line pairs. The compression state of the cache-line
       *    pair is specified by 2 bits in the CCS.  Each CCS cache-line
       *    represents an area on the main surface of 16x16 sets of 128 byte
       *    Y-tiled cache-line-pairs. CCS is always Y tiled."
       *
       * Each 2-bit surface element in the CCS corresponds to a single
       * cache-line pair in the main surface.  This means that 16x16 el block
       * in the CCS maps to a Y-tiled cache line.  Fortunately, CCS layouts
       * are calculated with a very large alignment so we can round up to a
       * whole cache line without worrying about overdraw.
       */

      /* On Broadwell and above, a CCS tile is the same as a Y tile when
       * viewed at the cache-line granularity.  Fortunately, the horizontal
       * and vertical alignment requirements of the CCS are such that we can
       * align to an entire cache line without worrying about crossing over
       * from one LOD to another.
       */
      const uint32_t x_el_per_cl = ccs_tile_info.logical_extent_el.w / 8;
      const uint32_t y_el_per_cl = ccs_tile_info.logical_extent_el.h / 8;
      assert(surf->aux_surf->image_alignment_el.w % x_el_per_cl == 0);
      assert(surf->aux_surf->image_alignment_el.h % y_el_per_cl == 0);

      assert(x_offset_el % x_el_per_cl == 0);
      assert(y_offset_el % y_el_per_cl == 0);
      x_offset_cl = x_offset_el / x_el_per_cl;
      y_offset_cl = y_offset_el / y_el_per_cl;
      width_cl = DIV_ROUND_UP(width_el, x_el_per_cl);
      height_cl = DIV_ROUND_UP(height_el, y_el_per_cl);
   } else {
      /* On gen7, the CCS tiling is not so nice.  However, there we are
       * guaranteed that we only have a single level and slice so we don't
       * have to worry about it and can just align to a whole tile.
       */
      assert(surf->aux_surf->logical_level0_px.depth == 1);
      assert(surf->aux_surf->logical_level0_px.array_len == 1);
      assert(x_offset_el == 0 && y_offset_el == 0);
      const uint32_t width_tl =
         DIV_ROUND_UP(width_el, ccs_tile_info.logical_extent_el.w);
      const uint32_t height_tl =
         DIV_ROUND_UP(height_el, ccs_tile_info.logical_extent_el.h);
      x_offset_cl = 0;
      y_offset_cl = 0;
      width_cl = width_tl * 8;
      height_cl = height_tl * 8;
   }

   /* We're going to use a RGBA32 format so as to write data as quickly as
    * possible.  A y-tiled cache line will then be 1x4 px.
    */
   const uint32_t x_offset_rgba_px = x_offset_cl;
   const uint32_t y_offset_rgba_px = y_offset_cl * 4;
   const uint32_t width_rgba_px = width_cl;
   const uint32_t height_rgba_px = height_cl * 4;

   ASSERTED bool ok =
      isl_surf_init(batch->blorp->isl_dev, &params.dst.surf,
                    .dim = ISL_SURF_DIM_2D,
                    .format = ISL_FORMAT_R32G32B32A32_UINT,
                    .width = width_rgba_px + x_offset_rgba_px,
                    .height = height_rgba_px + y_offset_rgba_px,
                    .depth = 1,
                    .levels = 1,
                    .array_len = 1,
                    .samples = 1,
                    .row_pitch_B = surf->aux_surf->row_pitch_B,
                    .usage = ISL_SURF_USAGE_RENDER_TARGET_BIT,
                    .tiling_flags = ISL_TILING_Y0_BIT);
   assert(ok);

   params.x0 = x_offset_rgba_px;
   params.y0 = y_offset_rgba_px;
   params.x1 = x_offset_rgba_px + width_rgba_px;
   params.y1 = y_offset_rgba_px + height_rgba_px;

   /* A CCS value of 0 means "uncompressed." */
   memset(&params.wm_inputs.clear_color, 0,
          sizeof(params.wm_inputs.clear_color));

   if (!blorp_params_get_clear_kernel(batch, &params, true, false))
      return;

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