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
|
/*
* 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(&sampled_image->image->deref->dest.ssa);
call->params[param_idx++] =
nir_src_for_ssa(&sampled_image->sampler->deref->dest.ssa);
} 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, ¶m_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 *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, 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;
}
static void
vtn_add_case(struct vtn_builder *b, struct vtn_switch *swtch,
struct vtn_block *break_block,
uint32_t block_id, uint64_t val, bool is_default)
{
struct vtn_block *case_block = vtn_block(b, block_id);
/* Don't create dummy cases that just break */
if (case_block == break_block)
return;
if (case_block->switch_case == NULL) {
struct vtn_case *c = ralloc(b, struct vtn_case);
c->node.type = vtn_cf_node_type_case;
c->node.parent = &swtch->node;
list_inithead(&c->body);
c->start_block = case_block;
c->fallthrough = NULL;
util_dynarray_init(&c->values, b);
c->is_default = false;
c->visited = false;
list_addtail(&c->node.link, &swtch->cases);
case_block->switch_case = c;
}
if (is_default) {
case_block->switch_case->is_default = true;
} else {
util_dynarray_append(&case_block->switch_case->values, uint64_t, val);
}
}
/* 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 enum vtn_branch_type
vtn_get_branch_type(struct vtn_builder *b,
struct vtn_block *block,
struct vtn_case *swcase, struct vtn_block *switch_break,
struct vtn_block *loop_break, struct vtn_block *loop_cont)
{
if (block->switch_case) {
/* This branch is actually a fallthrough */
vtn_assert(swcase->fallthrough == NULL ||
swcase->fallthrough == block->switch_case);
swcase->fallthrough = block->switch_case;
return vtn_branch_type_switch_fallthrough;
} else if (block == loop_break) {
return vtn_branch_type_loop_break;
} else if (block == loop_cont) {
return vtn_branch_type_loop_continue;
} else if (block == switch_break) {
return vtn_branch_type_switch_break;
} else {
return vtn_branch_type_none;
}
}
static void
vtn_cfg_walk_blocks(struct vtn_builder *b,
struct vtn_cf_node *cf_parent,
struct list_head *cf_list,
struct vtn_block *start, struct vtn_case *switch_case,
struct vtn_block *switch_break,
struct vtn_block *loop_break, struct vtn_block *loop_cont,
struct vtn_block *end)
{
struct vtn_block *block = start;
while (block != end) {
if (block->merge && (*block->merge & SpvOpCodeMask) == SpvOpLoopMerge &&
!block->loop) {
struct vtn_loop *loop = ralloc(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->control = block->merge[3];
list_addtail(&loop->node.link, cf_list);
block->loop = loop;
struct vtn_block *new_loop_break = vtn_block(b, block->merge[1]);
struct vtn_block *new_loop_cont = vtn_block(b, block->merge[2]);
/* Note: This recursive call will start with the current block as
* its start block. If we weren't careful, we would get here
* again and end up in infinite recursion. 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 call that
* tries to handle this loop goes to the cases below and gets
* handled as a regular block.
*
* Note: When we make the recursive walk calls, we pass NULL for
* the switch break since you have to break out of the loop first.
* We do, however, still pass the current switch case because it's
* possible that the merge block for the loop is the start of
* another case.
*/
vtn_cfg_walk_blocks(b, &loop->node, &loop->body,
block, switch_case, NULL,
new_loop_break, new_loop_cont, NULL );
vtn_cfg_walk_blocks(b, &loop->node, &loop->cont_body,
new_loop_cont, NULL, NULL,
new_loop_break, NULL, block);
enum vtn_branch_type branch_type =
vtn_get_branch_type(b, new_loop_break, switch_case, switch_break,
loop_break, loop_cont);
if (branch_type != vtn_branch_type_none) {
/* Stop walking through the CFG when this inner loop's break block
* ends up as the same block as the outer loop's continue block
* because we are already going to visit it.
*/
vtn_assert(branch_type == vtn_branch_type_loop_continue);
return;
}
block = new_loop_break;
continue;
}
vtn_assert(block->node.link.next == NULL);
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_get_branch_type(b, branch_block,
switch_case, switch_break,
loop_break, loop_cont);
if (block->branch_type != vtn_branch_type_none)
return;
block = branch_block;
continue;
}
case SpvOpReturn:
case SpvOpReturnValue:
block->branch_type = vtn_branch_type_return;
return;
case SpvOpKill:
block->branch_type = vtn_branch_type_discard;
return;
case SpvOpBranchConditional: {
struct vtn_block *then_block = vtn_block(b, block->branch[2]);
struct vtn_block *else_block = vtn_block(b, block->branch[3]);
struct vtn_if *if_stmt = ralloc(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) {
if_stmt->control = block->merge[2];
} else {
if_stmt->control = SpvSelectionControlMaskNone;
}
if_stmt->then_type = vtn_get_branch_type(b, then_block,
switch_case, switch_break,
loop_break, loop_cont);
if_stmt->else_type = vtn_get_branch_type(b, else_block,
switch_case, switch_break,
loop_break, loop_cont);
if (then_block == else_block) {
block->branch_type = if_stmt->then_type;
if (block->branch_type == vtn_branch_type_none) {
block = then_block;
continue;
} else {
return;
}
} else if (if_stmt->then_type == vtn_branch_type_none &&
if_stmt->else_type == vtn_branch_type_none) {
/* Neither side of the if is something we can short-circuit. */
vtn_assert((*block->merge & SpvOpCodeMask) == SpvOpSelectionMerge);
struct vtn_block *merge_block = vtn_block(b, block->merge[1]);
vtn_cfg_walk_blocks(b, &if_stmt->node, &if_stmt->then_body,
then_block, switch_case, switch_break,
loop_break, loop_cont, merge_block);
vtn_cfg_walk_blocks(b, &if_stmt->node, &if_stmt->else_body,
else_block, switch_case, switch_break,
loop_break, loop_cont, merge_block);
enum vtn_branch_type merge_type =
vtn_get_branch_type(b, merge_block, switch_case, switch_break,
loop_break, loop_cont);
if (merge_type == vtn_branch_type_none) {
block = merge_block;
continue;
} else {
return;
}
} else if (if_stmt->then_type != vtn_branch_type_none &&
if_stmt->else_type != vtn_branch_type_none) {
/* Both sides were short-circuited. We're done here. */
return;
} else {
/* Exeactly one side of the branch could be short-circuited.
* We set the branch up as a predicated break/continue and we
* continue on with the other side as if it were what comes
* after the if.
*/
if (if_stmt->then_type == vtn_branch_type_none) {
block = then_block;
} else {
block = else_block;
}
continue;
}
vtn_fail("Should have returned or continued");
}
case SpvOpSwitch: {
vtn_assert((*block->merge & SpvOpCodeMask) == SpvOpSelectionMerge);
struct vtn_block *break_block = vtn_block(b, block->merge[1]);
struct vtn_switch *swtch = ralloc(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);
/* First, we go through and record all of the cases. */
const uint32_t *branch_end =
block->branch + (block->branch[0] >> SpvWordCountShift);
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,
"Selector of OpSelect must have a type of OpTypeInt");
nir_alu_type cond_type =
nir_get_nir_type_for_glsl_type(cond_val->type->type);
vtn_fail_if(nir_alu_type_get_base_type(cond_type) != nir_type_int &&
nir_alu_type_get_base_type(cond_type) != nir_type_uint,
"Selector of OpSelect must have a type of OpTypeInt");
bool is_default = true;
const unsigned bitsize = nir_alu_type_get_type_size(cond_type);
for (const uint32_t *w = block->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;
}
}
uint32_t block_id = *(w++);
vtn_add_case(b, swtch, break_block, block_id, literal, is_default);
is_default = false;
}
/* Now, we go through and walk the blocks. While we walk through
* the blocks, we also gather the much-needed fall-through
* information.
*/
vtn_foreach_cf_node(case_node, &swtch->cases) {
struct vtn_case *cse = vtn_cf_node_as_case(case_node);
vtn_assert(cse->start_block != break_block);
vtn_cfg_walk_blocks(b, &cse->node, &cse->body, cse->start_block,
cse, break_block, loop_break, loop_cont, NULL);
}
/* Finally, we walk over all of the cases one more time and put
* them in fall-through order.
*/
for (const uint32_t *w = block->branch + 2; w < branch_end;) {
struct vtn_block *case_block = vtn_block(b, *w);
if (bitsize <= 32) {
w += 2;
} else {
assert(bitsize == 64);
w += 3;
}
if (case_block == break_block)
continue;
vtn_assert(case_block->switch_case);
vtn_order_case(swtch, case_block->switch_case);
}
enum vtn_branch_type branch_type =
vtn_get_branch_type(b, break_block, switch_case, NULL,
loop_break, loop_cont);
if (branch_type != vtn_branch_type_none) {
/* It is possible that the break is actually the continue block
* for the containing loop. In this case, we need to bail and let
* the loop parsing code handle the continue properly.
*/
vtn_assert(branch_type == vtn_branch_type_loop_continue);
return;
}
block = break_block;
continue;
}
case SpvOpUnreachable:
return;
default:
vtn_fail("Unhandled opcode");
}
}
}
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(node, &b->functions) {
struct vtn_function *func = vtn_cf_node_as_function(node);
vtn_cfg_walk_blocks(b, &func->node, &func->body, func->start_block,
NULL, NULL, NULL, NULL, NULL);
}
}
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);
vtn_assert(phi_entry);
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_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_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);
/* 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)
nir_repair_ssa_impl(func->impl);
func->emitted = true;
}
|