summaryrefslogtreecommitdiff
path: root/src/compiler/spirv/vtn_cfg.c
blob: 4a07b35be53c4f93f57893e64fe70f8241450f8a (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
/*
 * Copyright © 2015 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include "vtn_private.h"
#include "nir/nir_vla.h"

static struct vtn_block *
vtn_block(struct vtn_builder *b, uint32_t value_id)
{
   return vtn_value(b, value_id, vtn_value_type_block)->block;
}

static struct vtn_pointer *
vtn_load_param_pointer(struct vtn_builder *b,
                       struct vtn_type *param_type,
                       uint32_t param_idx)
{
   struct vtn_type *ptr_type = param_type;
   if (param_type->base_type != vtn_base_type_pointer) {
      assert(param_type->base_type == vtn_base_type_image ||
             param_type->base_type == vtn_base_type_sampler);
      ptr_type = rzalloc(b, struct vtn_type);
      ptr_type->base_type = vtn_base_type_pointer;
      ptr_type->deref = param_type;
      ptr_type->storage_class = SpvStorageClassUniformConstant;
   }

   return vtn_pointer_from_ssa(b, nir_load_param(&b->nb, param_idx), ptr_type);
}

static unsigned
vtn_type_count_function_params(struct vtn_type *type)
{
   switch (type->base_type) {
   case vtn_base_type_array:
   case vtn_base_type_matrix:
      return type->length * vtn_type_count_function_params(type->array_element);

   case vtn_base_type_struct: {
      unsigned count = 0;
      for (unsigned i = 0; i < type->length; i++)
         count += vtn_type_count_function_params(type->members[i]);
      return count;
   }

   case vtn_base_type_sampled_image:
      return 2;

   default:
      return 1;
   }
}

static void
vtn_type_add_to_function_params(struct vtn_type *type,
                                nir_function *func,
                                unsigned *param_idx)
{
   static const nir_parameter nir_deref_param = {
      .num_components = 1,
      .bit_size = 32,
   };

   switch (type->base_type) {
   case vtn_base_type_array:
   case vtn_base_type_matrix:
      for (unsigned i = 0; i < type->length; i++)
         vtn_type_add_to_function_params(type->array_element, func, param_idx);
      break;

   case vtn_base_type_struct:
      for (unsigned i = 0; i < type->length; i++)
         vtn_type_add_to_function_params(type->members[i], func, param_idx);
      break;

   case vtn_base_type_sampled_image:
      func->params[(*param_idx)++] = nir_deref_param;
      func->params[(*param_idx)++] = nir_deref_param;
      break;

   case vtn_base_type_image:
   case vtn_base_type_sampler:
      func->params[(*param_idx)++] = nir_deref_param;
      break;

   case vtn_base_type_pointer:
      if (type->type) {
         func->params[(*param_idx)++] = (nir_parameter) {
            .num_components = glsl_get_vector_elements(type->type),
            .bit_size = glsl_get_bit_size(type->type),
         };
      } else {
         func->params[(*param_idx)++] = nir_deref_param;
      }
      break;

   default:
      func->params[(*param_idx)++] = (nir_parameter) {
         .num_components = glsl_get_vector_elements(type->type),
         .bit_size = glsl_get_bit_size(type->type),
      };
   }
}

static void
vtn_ssa_value_add_to_call_params(struct vtn_builder *b,
                                 struct vtn_ssa_value *value,
                                 struct vtn_type *type,
                                 nir_call_instr *call,
                                 unsigned *param_idx)
{
   switch (type->base_type) {
   case vtn_base_type_array:
   case vtn_base_type_matrix:
      for (unsigned i = 0; i < type->length; i++) {
         vtn_ssa_value_add_to_call_params(b, value->elems[i],
                                          type->array_element,
                                          call, param_idx);
      }
      break;

   case vtn_base_type_struct:
      for (unsigned i = 0; i < type->length; i++) {
         vtn_ssa_value_add_to_call_params(b, value->elems[i],
                                          type->members[i],
                                          call, param_idx);
      }
      break;

   default:
      call->params[(*param_idx)++] = nir_src_for_ssa(value->def);
      break;
   }
}

static void
vtn_ssa_value_load_function_param(struct vtn_builder *b,
                                  struct vtn_ssa_value *value,
                                  struct vtn_type *type,
                                  unsigned *param_idx)
{
   switch (type->base_type) {
   case vtn_base_type_array:
   case vtn_base_type_matrix:
      for (unsigned i = 0; i < type->length; i++) {
         vtn_ssa_value_load_function_param(b, value->elems[i],
                                           type->array_element, param_idx);
      }
      break;

   case vtn_base_type_struct:
      for (unsigned i = 0; i < type->length; i++) {
         vtn_ssa_value_load_function_param(b, value->elems[i],
                                           type->members[i], param_idx);
      }
      break;

   default:
      value->def = nir_load_param(&b->nb, (*param_idx)++);
      break;
   }
}

void
vtn_handle_function_call(struct vtn_builder *b, SpvOp opcode,
                         const uint32_t *w, unsigned count)
{
   struct vtn_type *res_type = vtn_value(b, w[1], vtn_value_type_type)->type;
   struct vtn_function *vtn_callee =
      vtn_value(b, w[3], vtn_value_type_function)->func;
   struct nir_function *callee = vtn_callee->impl->function;

   vtn_callee->referenced = true;

   nir_call_instr *call = nir_call_instr_create(b->nb.shader, callee);

   unsigned param_idx = 0;

   nir_deref_instr *ret_deref = NULL;
   struct vtn_type *ret_type = vtn_callee->type->return_type;
   if (ret_type->base_type != vtn_base_type_void) {
      nir_variable *ret_tmp =
         nir_local_variable_create(b->nb.impl,
                                   glsl_get_bare_type(ret_type->type),
                                   "return_tmp");
      ret_deref = nir_build_deref_var(&b->nb, ret_tmp);
      call->params[param_idx++] = nir_src_for_ssa(&ret_deref->dest.ssa);
   }

   for (unsigned i = 0; i < vtn_callee->type->length; i++) {
      struct vtn_type *arg_type = vtn_callee->type->params[i];
      unsigned arg_id = w[4 + i];

      if (arg_type->base_type == vtn_base_type_sampled_image) {
         struct vtn_sampled_image *sampled_image =
            vtn_value(b, arg_id, vtn_value_type_sampled_image)->sampled_image;

         call->params[param_idx++] =
            nir_src_for_ssa(vtn_pointer_to_ssa(b, sampled_image->image));
         call->params[param_idx++] =
            nir_src_for_ssa(vtn_pointer_to_ssa(b, sampled_image->sampler));
      } else if (arg_type->base_type == vtn_base_type_pointer ||
                 arg_type->base_type == vtn_base_type_image ||
                 arg_type->base_type == vtn_base_type_sampler) {
         struct vtn_pointer *pointer =
            vtn_value(b, arg_id, vtn_value_type_pointer)->pointer;
         call->params[param_idx++] =
            nir_src_for_ssa(vtn_pointer_to_ssa(b, pointer));
      } else {
         vtn_ssa_value_add_to_call_params(b, vtn_ssa_value(b, arg_id),
                                          arg_type, call, &param_idx);
      }
   }
   assert(param_idx == call->num_params);

   nir_builder_instr_insert(&b->nb, &call->instr);

   if (ret_type->base_type == vtn_base_type_void) {
      vtn_push_value(b, w[2], vtn_value_type_undef);
   } else {
      vtn_push_ssa(b, w[2], res_type, vtn_local_load(b, ret_deref, 0));
   }
}

static bool
vtn_cfg_handle_prepass_instruction(struct vtn_builder *b, SpvOp opcode,
                                   const uint32_t *w, unsigned count)
{
   switch (opcode) {
   case SpvOpFunction: {
      vtn_assert(b->func == NULL);
      b->func = rzalloc(b, struct vtn_function);

      b->func->node.type = vtn_cf_node_type_function;
      b->func->node.parent = NULL;
      list_inithead(&b->func->body);
      b->func->control = w[3];

      UNUSED const struct glsl_type *result_type =
         vtn_value(b, w[1], vtn_value_type_type)->type->type;
      struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_function);
      val->func = b->func;

      b->func->type = vtn_value(b, w[4], vtn_value_type_type)->type;
      const struct vtn_type *func_type = b->func->type;

      vtn_assert(func_type->return_type->type == result_type);

      nir_function *func =
         nir_function_create(b->shader, ralloc_strdup(b->shader, val->name));

      unsigned num_params = 0;
      for (unsigned i = 0; i < func_type->length; i++)
         num_params += vtn_type_count_function_params(func_type->params[i]);

      /* Add one parameter for the function return value */
      if (func_type->return_type->base_type != vtn_base_type_void)
         num_params++;

      func->num_params = num_params;
      func->params = ralloc_array(b->shader, nir_parameter, num_params);

      unsigned idx = 0;
      if (func_type->return_type->base_type != vtn_base_type_void) {
         nir_address_format addr_format =
            vtn_mode_to_address_format(b, vtn_variable_mode_function);
         /* The return value is a regular pointer */
         func->params[idx++] = (nir_parameter) {
            .num_components = nir_address_format_num_components(addr_format),
            .bit_size = nir_address_format_bit_size(addr_format),
         };
      }

      for (unsigned i = 0; i < func_type->length; i++)
         vtn_type_add_to_function_params(func_type->params[i], func, &idx);
      assert(idx == num_params);

      b->func->impl = nir_function_impl_create(func);
      nir_builder_init(&b->nb, func->impl);
      b->nb.cursor = nir_before_cf_list(&b->func->impl->body);
      b->nb.exact = b->exact;

      b->func_param_idx = 0;

      /* The return value is the first parameter */
      if (func_type->return_type->base_type != vtn_base_type_void)
         b->func_param_idx++;
      break;
   }

   case SpvOpFunctionEnd:
      b->func->end = w;
      b->func = NULL;
      break;

   case SpvOpFunctionParameter: {
      struct vtn_type *type = vtn_value(b, w[1], vtn_value_type_type)->type;

      vtn_assert(b->func_param_idx < b->func->impl->function->num_params);

      if (type->base_type == vtn_base_type_sampled_image) {
         /* Sampled images are actually two parameters.  The first is the
          * image and the second is the sampler.
          */
         struct vtn_value *val =
            vtn_push_value(b, w[2], vtn_value_type_sampled_image);

         val->sampled_image = ralloc(b, struct vtn_sampled_image);

         struct vtn_type *image_type = rzalloc(b, struct vtn_type);
         image_type->base_type = vtn_base_type_image;
         image_type->type = type->type;

         struct vtn_type *sampler_type = rzalloc(b, struct vtn_type);
         sampler_type->base_type = vtn_base_type_sampler;
         sampler_type->type = glsl_bare_sampler_type();

         val->sampled_image->image =
            vtn_load_param_pointer(b, image_type, b->func_param_idx++);
         val->sampled_image->sampler =
            vtn_load_param_pointer(b, sampler_type, b->func_param_idx++);
      } else if (type->base_type == vtn_base_type_pointer &&
                 type->type != NULL) {
         /* This is a pointer with an actual storage type */
         nir_ssa_def *ssa_ptr = nir_load_param(&b->nb, b->func_param_idx++);
         vtn_push_value_pointer(b, w[2], vtn_pointer_from_ssa(b, ssa_ptr, type));
      } else if (type->base_type == vtn_base_type_pointer ||
                 type->base_type == vtn_base_type_image ||
                 type->base_type == vtn_base_type_sampler) {
         vtn_push_value_pointer(b, w[2], vtn_load_param_pointer(b, type, b->func_param_idx++));
      } else {
         /* We're a regular SSA value. */
         struct vtn_ssa_value *value = vtn_create_ssa_value(b, type->type);
         vtn_ssa_value_load_function_param(b, value, type, &b->func_param_idx);
         vtn_push_ssa(b, w[2], type, value);
      }
      break;
   }

   case SpvOpLabel: {
      vtn_assert(b->block == NULL);
      b->block = rzalloc(b, struct vtn_block);
      b->block->node.type = vtn_cf_node_type_block;
      b->block->label = w;
      vtn_push_value(b, w[1], vtn_value_type_block)->block = b->block;

      if (b->func->start_block == NULL) {
         /* This is the first block encountered for this function.  In this
          * case, we set the start block and add it to the list of
          * implemented functions that we'll walk later.
          */
         b->func->start_block = b->block;
         list_addtail(&b->func->node.link, &b->functions);
      }
      break;
   }

   case SpvOpSelectionMerge:
   case SpvOpLoopMerge:
      vtn_assert(b->block && b->block->merge == NULL);
      b->block->merge = w;
      break;

   case SpvOpBranch:
   case SpvOpBranchConditional:
   case SpvOpSwitch:
   case SpvOpKill:
   case SpvOpReturn:
   case SpvOpReturnValue:
   case SpvOpUnreachable:
      vtn_assert(b->block && b->block->branch == NULL);
      b->block->branch = w;
      b->block = NULL;
      break;

   default:
      /* Continue on as per normal */
      return true;
   }

   return true;
}

/* This function performs a depth-first search of the cases and puts them
 * in fall-through order.
 */
static void
vtn_order_case(struct vtn_switch *swtch, struct vtn_case *cse)
{
   if (cse->visited)
      return;

   cse->visited = true;

   list_del(&cse->node.link);

   if (cse->fallthrough) {
      vtn_order_case(swtch, cse->fallthrough);

      /* If we have a fall-through, place this case right before the case it
       * falls through to.  This ensures that fallthroughs come one after
       * the other.  These two can never get separated because that would
       * imply something else falling through to the same case.  Also, this
       * can't break ordering because the DFS ensures that this case is
       * visited before anything that falls through to it.
       */
      list_addtail(&cse->node.link, &cse->fallthrough->node.link);
   } else {
      list_add(&cse->node.link, &swtch->cases);
   }
}

static void
vtn_switch_order_cases(struct vtn_switch *swtch)
{
   struct list_head cases;
   list_replace(&swtch->cases, &cases);
   list_inithead(&swtch->cases);
   while (!list_is_empty(&cases)) {
      struct vtn_case *cse =
         list_first_entry(&cases, struct vtn_case, node.link);
      vtn_order_case(swtch, cse);
   }
}

static void
vtn_block_set_merge_cf_node(struct vtn_builder *b, struct vtn_block *block,
                            struct vtn_cf_node *cf_node)
{
   vtn_fail_if(block->merge_cf_node != NULL,
               "The merge block declared by a header block cannot be a "
               "merge block declared by any other header block.");

   block->merge_cf_node = cf_node;
}

#define VTN_DECL_CF_NODE_FIND(_type)                        \
static inline struct vtn_##_type *                          \
vtn_cf_node_find_##_type(struct vtn_cf_node *node)          \
{                                                           \
   while (node && node->type != vtn_cf_node_type_##_type)   \
      node = node->parent;                                  \
   return (struct vtn_##_type *)node;                       \
}

VTN_DECL_CF_NODE_FIND(if)
VTN_DECL_CF_NODE_FIND(loop)
VTN_DECL_CF_NODE_FIND(case)
VTN_DECL_CF_NODE_FIND(switch)
VTN_DECL_CF_NODE_FIND(function)

static enum vtn_branch_type
vtn_handle_branch(struct vtn_builder *b,
                  struct vtn_cf_node *cf_parent,
                  struct vtn_block *target_block)
{
   struct vtn_loop *loop = vtn_cf_node_find_loop(cf_parent);

   /* Detect a loop back-edge first.  That way none of the code below
    * accidentally operates on a loop back-edge.
    */
   if (loop && target_block == loop->header_block)
      return vtn_branch_type_loop_back_edge;

   /* Try to detect fall-through */
   if (target_block->switch_case) {
      /* When it comes to handling switch cases, we can break calls to
       * vtn_handle_branch into two cases: calls from within a case construct
       * and calls for the jump to each case construct.  In the second case,
       * cf_parent is the vtn_switch itself and vtn_cf_node_find_case() will
       * return the outer switch case in which this switch is contained.  It's
       * fine if the target block is a switch case from an outer switch as
       * long as it is also the switch break for this switch.
       */
      struct vtn_case *switch_case = vtn_cf_node_find_case(cf_parent);

      /* This doesn't get called for the OpSwitch */
      vtn_fail_if(switch_case == NULL,
                  "A switch case can only be entered through an OpSwitch or "
                  "falling through from another switch case.");

      /* Because block->switch_case is only set on the entry block for a given
       * switch case, we only ever get here if we're jumping to the start of a
       * switch case.  It's possible, however, that a switch case could jump
       * to itself via a back-edge.  That *should* get caught by the loop
       * handling case above but if we have a back edge without a loop merge,
       * we could en up here.
       */
      vtn_fail_if(target_block->switch_case == switch_case,
                  "A switch cannot fall-through to itself.  Likely, there is "
                  "a back-edge which is not to a loop header.");

      vtn_fail_if(target_block->switch_case->node.parent !=
                     switch_case->node.parent,
                  "A switch case fall-through must come from the same "
                  "OpSwitch construct");

      vtn_fail_if(switch_case->fallthrough != NULL &&
                  switch_case->fallthrough != target_block->switch_case,
                  "Each case construct can have at most one branch to "
                  "another case construct");

      switch_case->fallthrough = target_block->switch_case;

      /* We don't immediately return vtn_branch_type_switch_fallthrough
       * because it may also be a loop or switch break for an inner loop or
       * switch and that takes precedence.
       */
   }

   if (loop && target_block == loop->cont_block)
      return vtn_branch_type_loop_continue;

   /* We walk blocks as a breadth-first search on the control-flow construct
    * tree where, when we find a construct, we add the vtn_cf_node for that
    * construct and continue iterating at the merge target block (if any).
    * Therefore, we want merges whose with parent == cf_parent to be treated
    * as regular branches.  We only want to consider merges if they break out
    * of the current CF construct.
    */
   if (target_block->merge_cf_node != NULL &&
       target_block->merge_cf_node->parent != cf_parent) {
      switch (target_block->merge_cf_node->type) {
      case vtn_cf_node_type_if:
         for (struct vtn_cf_node *node = cf_parent;
              node != target_block->merge_cf_node; node = node->parent) {
            vtn_fail_if(node == NULL || node->type != vtn_cf_node_type_if,
                        "Branching to the merge block of a selection "
                        "construct can only be used to break out of a "
                        "selection construct");

            struct vtn_if *if_stmt = vtn_cf_node_as_if(node);

            /* This should be guaranteed by our iteration */
            assert(if_stmt->merge_block != target_block);

            vtn_fail_if(if_stmt->merge_block != NULL,
                        "Branching to the merge block of a selection "
                        "construct can only be used to break out of the "
                        "inner most nested selection level");
         }
         return vtn_branch_type_if_merge;

      case vtn_cf_node_type_loop:
         vtn_fail_if(target_block->merge_cf_node != &loop->node,
                     "Loop breaks can only break out of the inner most "
                     "nested loop level");
         return vtn_branch_type_loop_break;

      case vtn_cf_node_type_switch: {
         struct vtn_switch *swtch = vtn_cf_node_find_switch(cf_parent);
         vtn_fail_if(target_block->merge_cf_node != &swtch->node,
                     "Switch breaks can only break out of the inner most "
                     "nested switch level");
         return vtn_branch_type_switch_break;
      }

      default:
         unreachable("Invalid CF node type for a merge");
      }
   }

   if (target_block->switch_case)
      return vtn_branch_type_switch_fallthrough;

   return vtn_branch_type_none;
}

struct vtn_cfg_work_item {
   struct list_head link;

   struct vtn_cf_node *cf_parent;
   struct list_head *cf_list;
   struct vtn_block *start_block;
};

static void
vtn_add_cfg_work_item(struct vtn_builder *b,
                      struct list_head *work_list,
                      struct vtn_cf_node *cf_parent,
                      struct list_head *cf_list,
                      struct vtn_block *start_block)
{
   struct vtn_cfg_work_item *work = ralloc(b, struct vtn_cfg_work_item);
   work->cf_parent = cf_parent;
   work->cf_list = cf_list;
   work->start_block = start_block;
   list_addtail(&work->link, work_list);
}

/* returns the default block */
static void
vtn_parse_switch(struct vtn_builder *b,
                 struct vtn_switch *swtch,
                 const uint32_t *branch,
                 struct list_head *case_list)
{
   const uint32_t *branch_end = branch + (branch[0] >> SpvWordCountShift);

   struct vtn_value *sel_val = vtn_untyped_value(b, branch[1]);
   vtn_fail_if(!sel_val->type ||
               sel_val->type->base_type != vtn_base_type_scalar,
               "Selector of OpSwitch must have a type of OpTypeInt");

   nir_alu_type sel_type =
      nir_get_nir_type_for_glsl_type(sel_val->type->type);
   vtn_fail_if(nir_alu_type_get_base_type(sel_type) != nir_type_int &&
               nir_alu_type_get_base_type(sel_type) != nir_type_uint,
               "Selector of OpSwitch must have a type of OpTypeInt");

   struct hash_table *block_to_case = _mesa_pointer_hash_table_create(b);

   bool is_default = true;
   const unsigned bitsize = nir_alu_type_get_type_size(sel_type);
   for (const uint32_t *w = branch + 2; w < branch_end;) {
      uint64_t literal = 0;
      if (!is_default) {
         if (bitsize <= 32) {
            literal = *(w++);
         } else {
            assert(bitsize == 64);
            literal = vtn_u64_literal(w);
            w += 2;
         }
      }
      struct vtn_block *case_block = vtn_block(b, *(w++));

      struct hash_entry *case_entry =
         _mesa_hash_table_search(block_to_case, case_block);

      struct vtn_case *cse;
      if (case_entry) {
         cse = case_entry->data;
      } else {
         cse = rzalloc(b, struct vtn_case);

         cse->node.type = vtn_cf_node_type_case;
         cse->node.parent = swtch ? &swtch->node : NULL;
         cse->block = case_block;
         list_inithead(&cse->body);
         util_dynarray_init(&cse->values, b);

         list_addtail(&cse->node.link, case_list);
         _mesa_hash_table_insert(block_to_case, case_block, cse);
      }

      if (is_default) {
         cse->is_default = true;
      } else {
         util_dynarray_append(&cse->values, uint64_t, literal);
      }

      is_default = false;
   }

   _mesa_hash_table_destroy(block_to_case, NULL);
}

/* Processes a block and returns the next block to process or NULL if we've
 * reached the end of the construct.
 */
static struct vtn_block *
vtn_process_block(struct vtn_builder *b,
                  struct list_head *work_list,
                  struct vtn_cf_node *cf_parent,
                  struct list_head *cf_list,
                  struct vtn_block *block)
{
   if (!list_is_empty(cf_list)) {
      /* vtn_process_block() acts like an iterator: it processes the given
       * block and then returns the next block to process.  For a given
       * control-flow construct, vtn_build_cfg() calls vtn_process_block()
       * repeatedly until it finally returns NULL.  Therefore, we know that
       * the only blocks on which vtn_process_block() can be called are either
       * the first block in a construct or a block that vtn_process_block()
       * returned for the current construct.  If cf_list is empty then we know
       * that we're processing the first block in the construct and we have to
       * add it to the list.
       *
       * If cf_list is not empty, then it must be the block returned by the
       * previous call to vtn_process_block().  We know a priori that
       * vtn_process_block only returns either normal branches
       * (vtn_branch_type_none) or merge target blocks.
       */
      switch (vtn_handle_branch(b, cf_parent, block)) {
      case vtn_branch_type_none:
         /* For normal branches, we want to process them and add them to the
          * current construct.  Merge target blocks also look like normal
          * branches from the perspective of this construct.  See also
          * vtn_handle_branch().
          */
         break;

      case vtn_branch_type_loop_continue:
      case vtn_branch_type_switch_fallthrough:
         /* The two cases where we can get early exits from a construct that
          * are not to that construct's merge target are loop continues and
          * switch fall-throughs.  In these cases, we need to break out of the
          * current construct by returning NULL.
          */
         return NULL;

      default:
         /* The only way we can get here is if something was used as two kinds
          * of merges at the same time and that's illegal.
          */
         vtn_fail("A block was used as a merge target from two or more "
                  "structured control-flow constructs");
      }
   }

   /* Once a block has been processed, it is placed into and the list link
    * will point to something non-null.  If we see a node we've already
    * processed here, it either exists in multiple functions or it's an
    * invalid back-edge.
    */
   if (block->node.parent != NULL) {
      vtn_fail_if(vtn_cf_node_find_function(&block->node) !=
                  vtn_cf_node_find_function(cf_parent),
                  "A block cannot exist in two functions at the "
                  "same time");

      vtn_fail("Invalid back or cross-edge in the CFG");
   }

   if (block->merge && (*block->merge & SpvOpCodeMask) == SpvOpLoopMerge &&
       block->loop == NULL) {
      vtn_fail_if((*block->branch & SpvOpCodeMask) != SpvOpBranch &&
                  (*block->branch & SpvOpCodeMask) != SpvOpBranchConditional,
                  "An OpLoopMerge instruction must immediately precede "
                  "either an OpBranch or OpBranchConditional instruction.");

      struct vtn_loop *loop = rzalloc(b, struct vtn_loop);

      loop->node.type = vtn_cf_node_type_loop;
      loop->node.parent = cf_parent;
      list_inithead(&loop->body);
      list_inithead(&loop->cont_body);
      loop->header_block = block;
      loop->break_block = vtn_block(b, block->merge[1]);
      loop->cont_block = vtn_block(b, block->merge[2]);
      loop->control = block->merge[3];

      list_addtail(&loop->node.link, cf_list);
      block->loop = loop;

      /* Note: The work item for the main loop body will start with the
       * current block as its start block.  If we weren't careful, we would
       * get here again and end up in an infinite loop.  This is why we set
       * block->loop above and check for it before creating one.  This way,
       * we only create the loop once and the second iteration that tries to
       * handle this loop goes to the cases below and gets handled as a
       * regular block.
       */
      vtn_add_cfg_work_item(b, work_list, &loop->node,
                            &loop->body, loop->header_block);

      /* For continue targets, SPIR-V guarantees the following:
       *
       *  - the Continue Target must dominate the back-edge block
       *  - the back-edge block must post dominate the Continue Target
       *
       * If the header block is the same as the continue target, this
       * condition is trivially satisfied and there is no real continue
       * section.
       */
      if (loop->cont_block != loop->header_block) {
         vtn_add_cfg_work_item(b, work_list, &loop->node,
                               &loop->cont_body, loop->cont_block);
      }

      vtn_block_set_merge_cf_node(b, loop->break_block, &loop->node);

      return loop->break_block;
   }

   /* Add the block to the CF list */
   block->node.parent = cf_parent;
   list_addtail(&block->node.link, cf_list);

   switch (*block->branch & SpvOpCodeMask) {
   case SpvOpBranch: {
      struct vtn_block *branch_block = vtn_block(b, block->branch[1]);

      block->branch_type = vtn_handle_branch(b, cf_parent, branch_block);

      if (block->branch_type == vtn_branch_type_none)
         return branch_block;
      else
         return NULL;
   }

   case SpvOpReturn:
   case SpvOpReturnValue:
      block->branch_type = vtn_branch_type_return;
      return NULL;

   case SpvOpKill:
      b->has_kill = true;
      block->branch_type = vtn_branch_type_discard;
      return NULL;

   case SpvOpBranchConditional: {
      struct vtn_value *cond_val = vtn_untyped_value(b, block->branch[1]);
      vtn_fail_if(!cond_val->type ||
                  cond_val->type->base_type != vtn_base_type_scalar ||
                  cond_val->type->type != glsl_bool_type(),
                  "Condition must be a Boolean type scalar");

      struct vtn_block *then_block = vtn_block(b, block->branch[2]);
      struct vtn_block *else_block = vtn_block(b, block->branch[3]);

      if (then_block == else_block) {
         /* This is uncommon but it can happen.  We treat this the same way as
          * an unconditional branch.
          */
         block->branch_type = vtn_handle_branch(b, cf_parent, then_block);

         if (block->branch_type == vtn_branch_type_none)
            return then_block;
         else
            return NULL;
      }

      struct vtn_if *if_stmt = rzalloc(b, struct vtn_if);

      if_stmt->node.type = vtn_cf_node_type_if;
      if_stmt->node.parent = cf_parent;
      if_stmt->condition = block->branch[1];
      list_inithead(&if_stmt->then_body);
      list_inithead(&if_stmt->else_body);

      list_addtail(&if_stmt->node.link, cf_list);

      if (block->merge &&
          (*block->merge & SpvOpCodeMask) == SpvOpSelectionMerge) {
         /* We may not always have a merge block and that merge doesn't
          * technically have to be an OpSelectionMerge.  We could have a block
          * with an OpLoopMerge which ends in an OpBranchConditional.
          */
         if_stmt->merge_block = vtn_block(b, block->merge[1]);
         vtn_block_set_merge_cf_node(b, if_stmt->merge_block, &if_stmt->node);

         if_stmt->control = block->merge[2];
      }

      if_stmt->then_type = vtn_handle_branch(b, &if_stmt->node, then_block);
      if (if_stmt->then_type == vtn_branch_type_none) {
         vtn_add_cfg_work_item(b, work_list, &if_stmt->node,
                               &if_stmt->then_body, then_block);
      }

      if_stmt->else_type = vtn_handle_branch(b, &if_stmt->node, else_block);
      if (if_stmt->else_type == vtn_branch_type_none) {
         vtn_add_cfg_work_item(b, work_list, &if_stmt->node,
                               &if_stmt->else_body, else_block);
      }

      return if_stmt->merge_block;
   }

   case SpvOpSwitch: {
      struct vtn_switch *swtch = rzalloc(b, struct vtn_switch);

      swtch->node.type = vtn_cf_node_type_switch;
      swtch->node.parent = cf_parent;
      swtch->selector = block->branch[1];
      list_inithead(&swtch->cases);

      list_addtail(&swtch->node.link, cf_list);

      /* We may not always have a merge block */
      if (block->merge) {
         vtn_fail_if((*block->merge & SpvOpCodeMask) != SpvOpSelectionMerge,
                     "An OpLoopMerge instruction must immediately precede "
                     "either an OpBranch or OpBranchConditional "
                     "instruction.");
         swtch->break_block = vtn_block(b, block->merge[1]);
         vtn_block_set_merge_cf_node(b, swtch->break_block, &swtch->node);
      }

      /* First, we go through and record all of the cases. */
      vtn_parse_switch(b, swtch, block->branch, &swtch->cases);

      /* Gather the branch types for the switch */
      vtn_foreach_cf_node(case_node, &swtch->cases) {
         struct vtn_case *cse = vtn_cf_node_as_case(case_node);

         cse->type = vtn_handle_branch(b, &swtch->node, cse->block);
         switch (cse->type) {
         case vtn_branch_type_none:
            /* This is a "real" cases which has stuff in it */
            vtn_fail_if(cse->block->switch_case != NULL,
                        "OpSwitch has a case which is also in another "
                        "OpSwitch construct");
            cse->block->switch_case = cse;
            vtn_add_cfg_work_item(b, work_list, &cse->node,
                                  &cse->body, cse->block);
            break;

         case vtn_branch_type_switch_break:
         case vtn_branch_type_loop_break:
         case vtn_branch_type_loop_continue:
            /* Switch breaks as well as loop breaks and continues can be
             * used to break out of a switch construct or as direct targets
             * of the OpSwitch.
             */
            break;

         default:
            vtn_fail("Target of OpSwitch is not a valid structured exit "
                     "from the switch construct.");
         }
      }

      return swtch->break_block;
   }

   case SpvOpUnreachable:
      return NULL;

   default:
      vtn_fail("Block did not end with a valid branch instruction");
   }
}

void
vtn_build_cfg(struct vtn_builder *b, const uint32_t *words, const uint32_t *end)
{
   vtn_foreach_instruction(b, words, end,
                           vtn_cfg_handle_prepass_instruction);

   vtn_foreach_cf_node(func_node, &b->functions) {
      struct vtn_function *func = vtn_cf_node_as_function(func_node);

      /* We build the CFG for each function by doing a breadth-first search on
       * the control-flow graph.  We keep track of our state using a worklist.
       * Doing a BFS ensures that we visit each structured control-flow
       * construct and its merge node before we visit the stuff inside the
       * construct.
       */
      struct list_head work_list;
      list_inithead(&work_list);
      vtn_add_cfg_work_item(b, &work_list, &func->node, &func->body,
                            func->start_block);

      while (!list_is_empty(&work_list)) {
         struct vtn_cfg_work_item *work =
            list_first_entry(&work_list, struct vtn_cfg_work_item, link);
         list_del(&work->link);

         for (struct vtn_block *block = work->start_block; block; ) {
            block = vtn_process_block(b, &work_list, work->cf_parent,
                                      work->cf_list, block);
         }
      }
   }
}

static bool
vtn_handle_phis_first_pass(struct vtn_builder *b, SpvOp opcode,
                           const uint32_t *w, unsigned count)
{
   if (opcode == SpvOpLabel)
      return true; /* Nothing to do */

   /* If this isn't a phi node, stop. */
   if (opcode != SpvOpPhi)
      return false;

   /* For handling phi nodes, we do a poor-man's out-of-ssa on the spot.
    * For each phi, we create a variable with the appropreate type and
    * do a load from that variable.  Then, in a second pass, we add
    * stores to that variable to each of the predecessor blocks.
    *
    * We could do something more intelligent here.  However, in order to
    * handle loops and things properly, we really need dominance
    * information.  It would end up basically being the into-SSA
    * algorithm all over again.  It's easier if we just let
    * lower_vars_to_ssa do that for us instead of repeating it here.
    */
   struct vtn_type *type = vtn_value(b, w[1], vtn_value_type_type)->type;
   nir_variable *phi_var =
      nir_local_variable_create(b->nb.impl, type->type, "phi");
   _mesa_hash_table_insert(b->phi_table, w, phi_var);

   vtn_push_ssa(b, w[2], type,
                vtn_local_load(b, nir_build_deref_var(&b->nb, phi_var), 0));

   return true;
}

static bool
vtn_handle_phi_second_pass(struct vtn_builder *b, SpvOp opcode,
                           const uint32_t *w, unsigned count)
{
   if (opcode != SpvOpPhi)
      return true;

   struct hash_entry *phi_entry = _mesa_hash_table_search(b->phi_table, w);

   /* It's possible that this phi is in an unreachable block in which case it
    * may never have been emitted and therefore may not be in the hash table.
    * In this case, there's no var for it and it's safe to just bail.
    */
   if (phi_entry == NULL)
      return true;

   nir_variable *phi_var = phi_entry->data;

   for (unsigned i = 3; i < count; i += 2) {
      struct vtn_block *pred = vtn_block(b, w[i + 1]);

      /* If block does not have end_nop, that is because it is an unreacheable
       * block, and hence it is not worth to handle it */
      if (!pred->end_nop)
         continue;

      b->nb.cursor = nir_after_instr(&pred->end_nop->instr);

      struct vtn_ssa_value *src = vtn_ssa_value(b, w[i]);

      vtn_local_store(b, src, nir_build_deref_var(&b->nb, phi_var), 0);
   }

   return true;
}

static void
vtn_emit_branch(struct vtn_builder *b, enum vtn_branch_type branch_type,
                nir_variable *switch_fall_var, bool *has_switch_break)
{
   switch (branch_type) {
   case vtn_branch_type_if_merge:
      break; /* Nothing to do */
   case vtn_branch_type_switch_break:
      nir_store_var(&b->nb, switch_fall_var, nir_imm_false(&b->nb), 1);
      *has_switch_break = true;
      break;
   case vtn_branch_type_switch_fallthrough:
      break; /* Nothing to do */
   case vtn_branch_type_loop_break:
      nir_jump(&b->nb, nir_jump_break);
      break;
   case vtn_branch_type_loop_continue:
      nir_jump(&b->nb, nir_jump_continue);
      break;
   case vtn_branch_type_loop_back_edge:
      break;
   case vtn_branch_type_return:
      nir_jump(&b->nb, nir_jump_return);
      break;
   case vtn_branch_type_discard: {
      nir_intrinsic_instr *discard =
         nir_intrinsic_instr_create(b->nb.shader, nir_intrinsic_discard);
      nir_builder_instr_insert(&b->nb, &discard->instr);
      break;
   }
   default:
      vtn_fail("Invalid branch type");
   }
}

static nir_ssa_def *
vtn_switch_case_condition(struct vtn_builder *b, struct vtn_switch *swtch,
                          nir_ssa_def *sel, struct vtn_case *cse)
{
   if (cse->is_default) {
      nir_ssa_def *any = nir_imm_false(&b->nb);
      vtn_foreach_cf_node(other_node, &swtch->cases) {
         struct vtn_case *other = vtn_cf_node_as_case(other_node);
         if (other->is_default)
            continue;

         any = nir_ior(&b->nb, any,
                       vtn_switch_case_condition(b, swtch, sel, other));
      }
      return nir_inot(&b->nb, any);
   } else {
      nir_ssa_def *cond = nir_imm_false(&b->nb);
      util_dynarray_foreach(&cse->values, uint64_t, val) {
         nir_ssa_def *imm = nir_imm_intN_t(&b->nb, *val, sel->bit_size);
         cond = nir_ior(&b->nb, cond, nir_ieq(&b->nb, sel, imm));
      }
      return cond;
   }
}

static nir_loop_control
vtn_loop_control(struct vtn_builder *b, struct vtn_loop *vtn_loop)
{
   if (vtn_loop->control == SpvLoopControlMaskNone)
      return nir_loop_control_none;
   else if (vtn_loop->control & SpvLoopControlDontUnrollMask)
      return nir_loop_control_dont_unroll;
   else if (vtn_loop->control & SpvLoopControlUnrollMask)
      return nir_loop_control_unroll;
   else if (vtn_loop->control & SpvLoopControlDependencyInfiniteMask ||
            vtn_loop->control & SpvLoopControlDependencyLengthMask ||
            vtn_loop->control & SpvLoopControlMinIterationsMask ||
            vtn_loop->control & SpvLoopControlMaxIterationsMask ||
            vtn_loop->control & SpvLoopControlIterationMultipleMask ||
            vtn_loop->control & SpvLoopControlPeelCountMask ||
            vtn_loop->control & SpvLoopControlPartialCountMask) {
      /* We do not do anything special with these yet. */
      return nir_loop_control_none;
   } else {
      vtn_fail("Invalid loop control");
   }
}

static nir_selection_control
vtn_selection_control(struct vtn_builder *b, struct vtn_if *vtn_if)
{
   if (vtn_if->control == SpvSelectionControlMaskNone)
      return nir_selection_control_none;
   else if (vtn_if->control & SpvSelectionControlDontFlattenMask)
      return nir_selection_control_dont_flatten;
   else if (vtn_if->control & SpvSelectionControlFlattenMask)
      return nir_selection_control_flatten;
   else
      vtn_fail("Invalid selection control");
}

static void
vtn_emit_cf_list(struct vtn_builder *b, struct list_head *cf_list,
                 nir_variable *switch_fall_var, bool *has_switch_break,
                 vtn_instruction_handler handler)
{
   vtn_foreach_cf_node(node, cf_list) {
      switch (node->type) {
      case vtn_cf_node_type_block: {
         struct vtn_block *block = vtn_cf_node_as_block(node);

         const uint32_t *block_start = block->label;
         const uint32_t *block_end = block->merge ? block->merge :
                                                    block->branch;

         block_start = vtn_foreach_instruction(b, block_start, block_end,
                                               vtn_handle_phis_first_pass);

         vtn_foreach_instruction(b, block_start, block_end, handler);

         block->end_nop = nir_intrinsic_instr_create(b->nb.shader,
                                                     nir_intrinsic_nop);
         nir_builder_instr_insert(&b->nb, &block->end_nop->instr);

         if ((*block->branch & SpvOpCodeMask) == SpvOpReturnValue) {
            vtn_fail_if(b->func->type->return_type->base_type ==
                        vtn_base_type_void,
                        "Return with a value from a function returning void");
            struct vtn_ssa_value *src = vtn_ssa_value(b, block->branch[1]);
            const struct glsl_type *ret_type =
               glsl_get_bare_type(b->func->type->return_type->type);
            nir_deref_instr *ret_deref =
               nir_build_deref_cast(&b->nb, nir_load_param(&b->nb, 0),
                                    nir_var_function_temp, ret_type, 0);
            vtn_local_store(b, src, ret_deref, 0);
         }

         if (block->branch_type != vtn_branch_type_none) {
            vtn_emit_branch(b, block->branch_type,
                            switch_fall_var, has_switch_break);
            return;
         }

         break;
      }

      case vtn_cf_node_type_if: {
         struct vtn_if *vtn_if = vtn_cf_node_as_if(node);
         bool sw_break = false;

         nir_if *nif =
            nir_push_if(&b->nb, vtn_ssa_value(b, vtn_if->condition)->def);

         nif->control = vtn_selection_control(b, vtn_if);

         if (vtn_if->then_type == vtn_branch_type_none) {
            vtn_emit_cf_list(b, &vtn_if->then_body,
                             switch_fall_var, &sw_break, handler);
         } else {
            vtn_emit_branch(b, vtn_if->then_type, switch_fall_var, &sw_break);
         }

         nir_push_else(&b->nb, nif);
         if (vtn_if->else_type == vtn_branch_type_none) {
            vtn_emit_cf_list(b, &vtn_if->else_body,
                             switch_fall_var, &sw_break, handler);
         } else {
            vtn_emit_branch(b, vtn_if->else_type, switch_fall_var, &sw_break);
         }

         nir_pop_if(&b->nb, nif);

         /* If we encountered a switch break somewhere inside of the if,
          * then it would have been handled correctly by calling
          * emit_cf_list or emit_branch for the interrior.  However, we
          * need to predicate everything following on wether or not we're
          * still going.
          */
         if (sw_break) {
            *has_switch_break = true;
            nir_push_if(&b->nb, nir_load_var(&b->nb, switch_fall_var));
         }
         break;
      }

      case vtn_cf_node_type_loop: {
         struct vtn_loop *vtn_loop = vtn_cf_node_as_loop(node);

         nir_loop *loop = nir_push_loop(&b->nb);
         loop->control = vtn_loop_control(b, vtn_loop);

         vtn_emit_cf_list(b, &vtn_loop->body, NULL, NULL, handler);

         if (!list_is_empty(&vtn_loop->cont_body)) {
            /* If we have a non-trivial continue body then we need to put
             * it at the beginning of the loop with a flag to ensure that
             * it doesn't get executed in the first iteration.
             */
            nir_variable *do_cont =
               nir_local_variable_create(b->nb.impl, glsl_bool_type(), "cont");

            b->nb.cursor = nir_before_cf_node(&loop->cf_node);
            nir_store_var(&b->nb, do_cont, nir_imm_false(&b->nb), 1);

            b->nb.cursor = nir_before_cf_list(&loop->body);

            nir_if *cont_if =
               nir_push_if(&b->nb, nir_load_var(&b->nb, do_cont));

            vtn_emit_cf_list(b, &vtn_loop->cont_body, NULL, NULL, handler);

            nir_pop_if(&b->nb, cont_if);

            nir_store_var(&b->nb, do_cont, nir_imm_true(&b->nb), 1);

            b->has_loop_continue = true;
         }

         nir_pop_loop(&b->nb, loop);
         break;
      }

      case vtn_cf_node_type_switch: {
         struct vtn_switch *vtn_switch = vtn_cf_node_as_switch(node);

         /* Before we can emit anything, we need to sort the list of cases in
          * fall-through order.
          */
         vtn_switch_order_cases(vtn_switch);

         /* First, we create a variable to keep track of whether or not the
          * switch is still going at any given point.  Any switch breaks
          * will set this variable to false.
          */
         nir_variable *fall_var =
            nir_local_variable_create(b->nb.impl, glsl_bool_type(), "fall");
         nir_store_var(&b->nb, fall_var, nir_imm_false(&b->nb), 1);

         nir_ssa_def *sel = vtn_ssa_value(b, vtn_switch->selector)->def;

         /* Now we can walk the list of cases and actually emit code */
         vtn_foreach_cf_node(case_node, &vtn_switch->cases) {
            struct vtn_case *cse = vtn_cf_node_as_case(case_node);

            /* Figure out the condition */
            nir_ssa_def *cond =
               vtn_switch_case_condition(b, vtn_switch, sel, cse);
            /* Take fallthrough into account */
            cond = nir_ior(&b->nb, cond, nir_load_var(&b->nb, fall_var));

            nir_if *case_if = nir_push_if(&b->nb, cond);

            bool has_break = false;
            nir_store_var(&b->nb, fall_var, nir_imm_true(&b->nb), 1);
            vtn_emit_cf_list(b, &cse->body, fall_var, &has_break, handler);
            (void)has_break; /* We don't care */

            nir_pop_if(&b->nb, case_if);
         }

         break;
      }

      default:
         vtn_fail("Invalid CF node type");
      }
   }
}

void
vtn_function_emit(struct vtn_builder *b, struct vtn_function *func,
                  vtn_instruction_handler instruction_handler)
{
   nir_builder_init(&b->nb, func->impl);
   b->func = func;
   b->nb.cursor = nir_after_cf_list(&func->impl->body);
   b->nb.exact = b->exact;
   b->has_loop_continue = false;
   b->phi_table = _mesa_pointer_hash_table_create(b);

   vtn_emit_cf_list(b, &func->body, NULL, NULL, instruction_handler);

   vtn_foreach_instruction(b, func->start_block->label, func->end,
                           vtn_handle_phi_second_pass);

   nir_rematerialize_derefs_in_use_blocks_impl(func->impl);

   /* Continue blocks for loops get inserted before the body of the loop
    * but instructions in the continue may use SSA defs in the loop body.
    * Therefore, we need to repair SSA to insert the needed phi nodes.
    */
   if (b->has_loop_continue || b->has_kill)
      nir_repair_ssa_impl(func->impl);

   func->emitted = true;
}