nir: add bit_size info to nir_load_const_instr_create()
[mesa.git] / src / compiler / nir / nir.c
1 /*
2 * Copyright © 2014 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Connor Abbott (cwabbott0@gmail.com)
25 *
26 */
27
28 #include "nir.h"
29 #include "nir_control_flow_private.h"
30 #include <assert.h>
31
32 nir_shader *
33 nir_shader_create(void *mem_ctx,
34 gl_shader_stage stage,
35 const nir_shader_compiler_options *options)
36 {
37 nir_shader *shader = ralloc(mem_ctx, nir_shader);
38
39 exec_list_make_empty(&shader->uniforms);
40 exec_list_make_empty(&shader->inputs);
41 exec_list_make_empty(&shader->outputs);
42 exec_list_make_empty(&shader->shared);
43
44 shader->options = options;
45 memset(&shader->info, 0, sizeof(shader->info));
46
47 exec_list_make_empty(&shader->functions);
48 exec_list_make_empty(&shader->registers);
49 exec_list_make_empty(&shader->globals);
50 exec_list_make_empty(&shader->system_values);
51 shader->reg_alloc = 0;
52
53 shader->num_inputs = 0;
54 shader->num_outputs = 0;
55 shader->num_uniforms = 0;
56 shader->num_shared = 0;
57
58 shader->stage = stage;
59
60 return shader;
61 }
62
63 static nir_register *
64 reg_create(void *mem_ctx, struct exec_list *list)
65 {
66 nir_register *reg = ralloc(mem_ctx, nir_register);
67
68 list_inithead(&reg->uses);
69 list_inithead(&reg->defs);
70 list_inithead(&reg->if_uses);
71
72 reg->num_components = 0;
73 reg->bit_size = 32;
74 reg->num_array_elems = 0;
75 reg->is_packed = false;
76 reg->name = NULL;
77
78 exec_list_push_tail(list, &reg->node);
79
80 return reg;
81 }
82
83 nir_register *
84 nir_global_reg_create(nir_shader *shader)
85 {
86 nir_register *reg = reg_create(shader, &shader->registers);
87 reg->index = shader->reg_alloc++;
88 reg->is_global = true;
89
90 return reg;
91 }
92
93 nir_register *
94 nir_local_reg_create(nir_function_impl *impl)
95 {
96 nir_register *reg = reg_create(ralloc_parent(impl), &impl->registers);
97 reg->index = impl->reg_alloc++;
98 reg->is_global = false;
99
100 return reg;
101 }
102
103 void
104 nir_reg_remove(nir_register *reg)
105 {
106 exec_node_remove(&reg->node);
107 }
108
109 void
110 nir_shader_add_variable(nir_shader *shader, nir_variable *var)
111 {
112 switch (var->data.mode) {
113 case nir_var_all:
114 assert(!"invalid mode");
115 break;
116
117 case nir_var_local:
118 assert(!"nir_shader_add_variable cannot be used for local variables");
119 break;
120
121 case nir_var_param:
122 assert(!"nir_shader_add_variable cannot be used for function parameters");
123 break;
124
125 case nir_var_global:
126 exec_list_push_tail(&shader->globals, &var->node);
127 break;
128
129 case nir_var_shader_in:
130 exec_list_push_tail(&shader->inputs, &var->node);
131 break;
132
133 case nir_var_shader_out:
134 exec_list_push_tail(&shader->outputs, &var->node);
135 break;
136
137 case nir_var_uniform:
138 case nir_var_shader_storage:
139 exec_list_push_tail(&shader->uniforms, &var->node);
140 break;
141
142 case nir_var_shared:
143 assert(shader->stage == MESA_SHADER_COMPUTE);
144 exec_list_push_tail(&shader->shared, &var->node);
145 break;
146
147 case nir_var_system_value:
148 exec_list_push_tail(&shader->system_values, &var->node);
149 break;
150 }
151 }
152
153 nir_variable *
154 nir_variable_create(nir_shader *shader, nir_variable_mode mode,
155 const struct glsl_type *type, const char *name)
156 {
157 nir_variable *var = rzalloc(shader, nir_variable);
158 var->name = ralloc_strdup(var, name);
159 var->type = type;
160 var->data.mode = mode;
161
162 if ((mode == nir_var_shader_in && shader->stage != MESA_SHADER_VERTEX) ||
163 (mode == nir_var_shader_out && shader->stage != MESA_SHADER_FRAGMENT))
164 var->data.interpolation = INTERP_QUALIFIER_SMOOTH;
165
166 if (mode == nir_var_shader_in || mode == nir_var_uniform)
167 var->data.read_only = true;
168
169 nir_shader_add_variable(shader, var);
170
171 return var;
172 }
173
174 nir_variable *
175 nir_local_variable_create(nir_function_impl *impl,
176 const struct glsl_type *type, const char *name)
177 {
178 nir_variable *var = rzalloc(impl->function->shader, nir_variable);
179 var->name = ralloc_strdup(var, name);
180 var->type = type;
181 var->data.mode = nir_var_local;
182
183 nir_function_impl_add_variable(impl, var);
184
185 return var;
186 }
187
188 nir_function *
189 nir_function_create(nir_shader *shader, const char *name)
190 {
191 nir_function *func = ralloc(shader, nir_function);
192
193 exec_list_push_tail(&shader->functions, &func->node);
194
195 func->name = ralloc_strdup(func, name);
196 func->shader = shader;
197 func->num_params = 0;
198 func->params = NULL;
199 func->return_type = glsl_void_type();
200 func->impl = NULL;
201
202 return func;
203 }
204
205 void nir_src_copy(nir_src *dest, const nir_src *src, void *mem_ctx)
206 {
207 dest->is_ssa = src->is_ssa;
208 if (src->is_ssa) {
209 dest->ssa = src->ssa;
210 } else {
211 dest->reg.base_offset = src->reg.base_offset;
212 dest->reg.reg = src->reg.reg;
213 if (src->reg.indirect) {
214 dest->reg.indirect = ralloc(mem_ctx, nir_src);
215 nir_src_copy(dest->reg.indirect, src->reg.indirect, mem_ctx);
216 } else {
217 dest->reg.indirect = NULL;
218 }
219 }
220 }
221
222 void nir_dest_copy(nir_dest *dest, const nir_dest *src, nir_instr *instr)
223 {
224 /* Copying an SSA definition makes no sense whatsoever. */
225 assert(!src->is_ssa);
226
227 dest->is_ssa = false;
228
229 dest->reg.base_offset = src->reg.base_offset;
230 dest->reg.reg = src->reg.reg;
231 if (src->reg.indirect) {
232 dest->reg.indirect = ralloc(instr, nir_src);
233 nir_src_copy(dest->reg.indirect, src->reg.indirect, instr);
234 } else {
235 dest->reg.indirect = NULL;
236 }
237 }
238
239 void
240 nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src,
241 nir_alu_instr *instr)
242 {
243 nir_src_copy(&dest->src, &src->src, &instr->instr);
244 dest->abs = src->abs;
245 dest->negate = src->negate;
246 for (unsigned i = 0; i < 4; i++)
247 dest->swizzle[i] = src->swizzle[i];
248 }
249
250 void
251 nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src,
252 nir_alu_instr *instr)
253 {
254 nir_dest_copy(&dest->dest, &src->dest, &instr->instr);
255 dest->write_mask = src->write_mask;
256 dest->saturate = src->saturate;
257 }
258
259
260 static void
261 cf_init(nir_cf_node *node, nir_cf_node_type type)
262 {
263 exec_node_init(&node->node);
264 node->parent = NULL;
265 node->type = type;
266 }
267
268 nir_function_impl *
269 nir_function_impl_create_bare(nir_shader *shader)
270 {
271 nir_function_impl *impl = ralloc(shader, nir_function_impl);
272
273 impl->function = NULL;
274
275 cf_init(&impl->cf_node, nir_cf_node_function);
276
277 exec_list_make_empty(&impl->body);
278 exec_list_make_empty(&impl->registers);
279 exec_list_make_empty(&impl->locals);
280 impl->num_params = 0;
281 impl->params = NULL;
282 impl->return_var = NULL;
283 impl->reg_alloc = 0;
284 impl->ssa_alloc = 0;
285 impl->valid_metadata = nir_metadata_none;
286
287 /* create start & end blocks */
288 nir_block *start_block = nir_block_create(shader);
289 nir_block *end_block = nir_block_create(shader);
290 start_block->cf_node.parent = &impl->cf_node;
291 end_block->cf_node.parent = &impl->cf_node;
292 impl->end_block = end_block;
293
294 exec_list_push_tail(&impl->body, &start_block->cf_node.node);
295
296 start_block->successors[0] = end_block;
297 _mesa_set_add(end_block->predecessors, start_block);
298 return impl;
299 }
300
301 nir_function_impl *
302 nir_function_impl_create(nir_function *function)
303 {
304 assert(function->impl == NULL);
305
306 nir_function_impl *impl = nir_function_impl_create_bare(function->shader);
307
308 function->impl = impl;
309 impl->function = function;
310
311 impl->num_params = function->num_params;
312 impl->params = ralloc_array(function->shader,
313 nir_variable *, impl->num_params);
314
315 for (unsigned i = 0; i < impl->num_params; i++) {
316 impl->params[i] = rzalloc(function->shader, nir_variable);
317 impl->params[i]->type = function->params[i].type;
318 impl->params[i]->data.mode = nir_var_param;
319 impl->params[i]->data.location = i;
320 }
321
322 if (!glsl_type_is_void(function->return_type)) {
323 impl->return_var = rzalloc(function->shader, nir_variable);
324 impl->return_var->type = function->return_type;
325 impl->return_var->data.mode = nir_var_param;
326 impl->return_var->data.location = -1;
327 } else {
328 impl->return_var = NULL;
329 }
330
331 return impl;
332 }
333
334 nir_block *
335 nir_block_create(nir_shader *shader)
336 {
337 nir_block *block = ralloc(shader, nir_block);
338
339 cf_init(&block->cf_node, nir_cf_node_block);
340
341 block->successors[0] = block->successors[1] = NULL;
342 block->predecessors = _mesa_set_create(block, _mesa_hash_pointer,
343 _mesa_key_pointer_equal);
344 block->imm_dom = NULL;
345 /* XXX maybe it would be worth it to defer allocation? This
346 * way it doesn't get allocated for shader ref's that never run
347 * nir_calc_dominance? For example, state-tracker creates an
348 * initial IR, clones that, runs appropriate lowering pass, passes
349 * to driver which does common lowering/opt, and then stores ref
350 * which is later used to do state specific lowering and futher
351 * opt. Do any of the references not need dominance metadata?
352 */
353 block->dom_frontier = _mesa_set_create(block, _mesa_hash_pointer,
354 _mesa_key_pointer_equal);
355
356 exec_list_make_empty(&block->instr_list);
357
358 return block;
359 }
360
361 static inline void
362 src_init(nir_src *src)
363 {
364 src->is_ssa = false;
365 src->reg.reg = NULL;
366 src->reg.indirect = NULL;
367 src->reg.base_offset = 0;
368 }
369
370 nir_if *
371 nir_if_create(nir_shader *shader)
372 {
373 nir_if *if_stmt = ralloc(shader, nir_if);
374
375 cf_init(&if_stmt->cf_node, nir_cf_node_if);
376 src_init(&if_stmt->condition);
377
378 nir_block *then = nir_block_create(shader);
379 exec_list_make_empty(&if_stmt->then_list);
380 exec_list_push_tail(&if_stmt->then_list, &then->cf_node.node);
381 then->cf_node.parent = &if_stmt->cf_node;
382
383 nir_block *else_stmt = nir_block_create(shader);
384 exec_list_make_empty(&if_stmt->else_list);
385 exec_list_push_tail(&if_stmt->else_list, &else_stmt->cf_node.node);
386 else_stmt->cf_node.parent = &if_stmt->cf_node;
387
388 return if_stmt;
389 }
390
391 nir_loop *
392 nir_loop_create(nir_shader *shader)
393 {
394 nir_loop *loop = ralloc(shader, nir_loop);
395
396 cf_init(&loop->cf_node, nir_cf_node_loop);
397
398 nir_block *body = nir_block_create(shader);
399 exec_list_make_empty(&loop->body);
400 exec_list_push_tail(&loop->body, &body->cf_node.node);
401 body->cf_node.parent = &loop->cf_node;
402
403 body->successors[0] = body;
404 _mesa_set_add(body->predecessors, body);
405
406 return loop;
407 }
408
409 static void
410 instr_init(nir_instr *instr, nir_instr_type type)
411 {
412 instr->type = type;
413 instr->block = NULL;
414 exec_node_init(&instr->node);
415 }
416
417 static void
418 dest_init(nir_dest *dest)
419 {
420 dest->is_ssa = false;
421 dest->reg.reg = NULL;
422 dest->reg.indirect = NULL;
423 dest->reg.base_offset = 0;
424 }
425
426 static void
427 alu_dest_init(nir_alu_dest *dest)
428 {
429 dest_init(&dest->dest);
430 dest->saturate = false;
431 dest->write_mask = 0xf;
432 }
433
434 static void
435 alu_src_init(nir_alu_src *src)
436 {
437 src_init(&src->src);
438 src->abs = src->negate = false;
439 src->swizzle[0] = 0;
440 src->swizzle[1] = 1;
441 src->swizzle[2] = 2;
442 src->swizzle[3] = 3;
443 }
444
445 nir_alu_instr *
446 nir_alu_instr_create(nir_shader *shader, nir_op op)
447 {
448 unsigned num_srcs = nir_op_infos[op].num_inputs;
449 nir_alu_instr *instr =
450 ralloc_size(shader,
451 sizeof(nir_alu_instr) + num_srcs * sizeof(nir_alu_src));
452
453 instr_init(&instr->instr, nir_instr_type_alu);
454 instr->op = op;
455 alu_dest_init(&instr->dest);
456 for (unsigned i = 0; i < num_srcs; i++)
457 alu_src_init(&instr->src[i]);
458
459 return instr;
460 }
461
462 nir_jump_instr *
463 nir_jump_instr_create(nir_shader *shader, nir_jump_type type)
464 {
465 nir_jump_instr *instr = ralloc(shader, nir_jump_instr);
466 instr_init(&instr->instr, nir_instr_type_jump);
467 instr->type = type;
468 return instr;
469 }
470
471 nir_load_const_instr *
472 nir_load_const_instr_create(nir_shader *shader, unsigned num_components,
473 unsigned bit_size)
474 {
475 nir_load_const_instr *instr = ralloc(shader, nir_load_const_instr);
476 instr_init(&instr->instr, nir_instr_type_load_const);
477
478 nir_ssa_def_init(&instr->instr, &instr->def, num_components, bit_size, NULL);
479
480 return instr;
481 }
482
483 nir_intrinsic_instr *
484 nir_intrinsic_instr_create(nir_shader *shader, nir_intrinsic_op op)
485 {
486 unsigned num_srcs = nir_intrinsic_infos[op].num_srcs;
487 nir_intrinsic_instr *instr =
488 ralloc_size(shader,
489 sizeof(nir_intrinsic_instr) + num_srcs * sizeof(nir_src));
490
491 instr_init(&instr->instr, nir_instr_type_intrinsic);
492 instr->intrinsic = op;
493
494 if (nir_intrinsic_infos[op].has_dest)
495 dest_init(&instr->dest);
496
497 for (unsigned i = 0; i < num_srcs; i++)
498 src_init(&instr->src[i]);
499
500 return instr;
501 }
502
503 nir_call_instr *
504 nir_call_instr_create(nir_shader *shader, nir_function *callee)
505 {
506 nir_call_instr *instr = ralloc(shader, nir_call_instr);
507 instr_init(&instr->instr, nir_instr_type_call);
508
509 instr->callee = callee;
510 instr->num_params = callee->num_params;
511 instr->params = ralloc_array(instr, nir_deref_var *, instr->num_params);
512 instr->return_deref = NULL;
513
514 return instr;
515 }
516
517 nir_tex_instr *
518 nir_tex_instr_create(nir_shader *shader, unsigned num_srcs)
519 {
520 nir_tex_instr *instr = rzalloc(shader, nir_tex_instr);
521 instr_init(&instr->instr, nir_instr_type_tex);
522
523 dest_init(&instr->dest);
524
525 instr->num_srcs = num_srcs;
526 instr->src = ralloc_array(instr, nir_tex_src, num_srcs);
527 for (unsigned i = 0; i < num_srcs; i++)
528 src_init(&instr->src[i].src);
529
530 instr->texture_index = 0;
531 instr->texture_array_size = 0;
532 instr->texture = NULL;
533 instr->sampler_index = 0;
534 instr->sampler = NULL;
535
536 return instr;
537 }
538
539 nir_phi_instr *
540 nir_phi_instr_create(nir_shader *shader)
541 {
542 nir_phi_instr *instr = ralloc(shader, nir_phi_instr);
543 instr_init(&instr->instr, nir_instr_type_phi);
544
545 dest_init(&instr->dest);
546 exec_list_make_empty(&instr->srcs);
547 return instr;
548 }
549
550 nir_parallel_copy_instr *
551 nir_parallel_copy_instr_create(nir_shader *shader)
552 {
553 nir_parallel_copy_instr *instr = ralloc(shader, nir_parallel_copy_instr);
554 instr_init(&instr->instr, nir_instr_type_parallel_copy);
555
556 exec_list_make_empty(&instr->entries);
557
558 return instr;
559 }
560
561 nir_ssa_undef_instr *
562 nir_ssa_undef_instr_create(nir_shader *shader,
563 unsigned num_components,
564 unsigned bit_size)
565 {
566 nir_ssa_undef_instr *instr = ralloc(shader, nir_ssa_undef_instr);
567 instr_init(&instr->instr, nir_instr_type_ssa_undef);
568
569 nir_ssa_def_init(&instr->instr, &instr->def, num_components, bit_size, NULL);
570
571 return instr;
572 }
573
574 nir_deref_var *
575 nir_deref_var_create(void *mem_ctx, nir_variable *var)
576 {
577 nir_deref_var *deref = ralloc(mem_ctx, nir_deref_var);
578 deref->deref.deref_type = nir_deref_type_var;
579 deref->deref.child = NULL;
580 deref->deref.type = var->type;
581 deref->var = var;
582 return deref;
583 }
584
585 nir_deref_array *
586 nir_deref_array_create(void *mem_ctx)
587 {
588 nir_deref_array *deref = ralloc(mem_ctx, nir_deref_array);
589 deref->deref.deref_type = nir_deref_type_array;
590 deref->deref.child = NULL;
591 deref->deref_array_type = nir_deref_array_type_direct;
592 src_init(&deref->indirect);
593 deref->base_offset = 0;
594 return deref;
595 }
596
597 nir_deref_struct *
598 nir_deref_struct_create(void *mem_ctx, unsigned field_index)
599 {
600 nir_deref_struct *deref = ralloc(mem_ctx, nir_deref_struct);
601 deref->deref.deref_type = nir_deref_type_struct;
602 deref->deref.child = NULL;
603 deref->index = field_index;
604 return deref;
605 }
606
607 static nir_deref_var *
608 copy_deref_var(void *mem_ctx, nir_deref_var *deref)
609 {
610 nir_deref_var *ret = nir_deref_var_create(mem_ctx, deref->var);
611 ret->deref.type = deref->deref.type;
612 if (deref->deref.child)
613 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
614 return ret;
615 }
616
617 static nir_deref_array *
618 copy_deref_array(void *mem_ctx, nir_deref_array *deref)
619 {
620 nir_deref_array *ret = nir_deref_array_create(mem_ctx);
621 ret->base_offset = deref->base_offset;
622 ret->deref_array_type = deref->deref_array_type;
623 if (deref->deref_array_type == nir_deref_array_type_indirect) {
624 nir_src_copy(&ret->indirect, &deref->indirect, mem_ctx);
625 }
626 ret->deref.type = deref->deref.type;
627 if (deref->deref.child)
628 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
629 return ret;
630 }
631
632 static nir_deref_struct *
633 copy_deref_struct(void *mem_ctx, nir_deref_struct *deref)
634 {
635 nir_deref_struct *ret = nir_deref_struct_create(mem_ctx, deref->index);
636 ret->deref.type = deref->deref.type;
637 if (deref->deref.child)
638 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
639 return ret;
640 }
641
642 nir_deref *
643 nir_copy_deref(void *mem_ctx, nir_deref *deref)
644 {
645 switch (deref->deref_type) {
646 case nir_deref_type_var:
647 return &copy_deref_var(mem_ctx, nir_deref_as_var(deref))->deref;
648 case nir_deref_type_array:
649 return &copy_deref_array(mem_ctx, nir_deref_as_array(deref))->deref;
650 case nir_deref_type_struct:
651 return &copy_deref_struct(mem_ctx, nir_deref_as_struct(deref))->deref;
652 default:
653 unreachable("Invalid dereference type");
654 }
655
656 return NULL;
657 }
658
659 /* Returns a load_const instruction that represents the constant
660 * initializer for the given deref chain. The caller is responsible for
661 * ensuring that there actually is a constant initializer.
662 */
663 nir_load_const_instr *
664 nir_deref_get_const_initializer_load(nir_shader *shader, nir_deref_var *deref)
665 {
666 nir_constant *constant = deref->var->constant_initializer;
667 assert(constant);
668
669 const nir_deref *tail = &deref->deref;
670 unsigned matrix_offset = 0;
671 while (tail->child) {
672 switch (tail->child->deref_type) {
673 case nir_deref_type_array: {
674 nir_deref_array *arr = nir_deref_as_array(tail->child);
675 assert(arr->deref_array_type == nir_deref_array_type_direct);
676 if (glsl_type_is_matrix(tail->type)) {
677 assert(arr->deref.child == NULL);
678 matrix_offset = arr->base_offset;
679 } else {
680 constant = constant->elements[arr->base_offset];
681 }
682 break;
683 }
684
685 case nir_deref_type_struct: {
686 constant = constant->elements[nir_deref_as_struct(tail->child)->index];
687 break;
688 }
689
690 default:
691 unreachable("Invalid deref child type");
692 }
693
694 tail = tail->child;
695 }
696
697 nir_load_const_instr *load =
698 nir_load_const_instr_create(shader, glsl_get_vector_elements(tail->type),
699 32);
700
701 matrix_offset *= load->def.num_components;
702 for (unsigned i = 0; i < load->def.num_components; i++) {
703 switch (glsl_get_base_type(tail->type)) {
704 case GLSL_TYPE_FLOAT:
705 case GLSL_TYPE_INT:
706 case GLSL_TYPE_UINT:
707 load->value.u32[i] = constant->value.u[matrix_offset + i];
708 break;
709 case GLSL_TYPE_BOOL:
710 load->value.u32[i] = constant->value.b[matrix_offset + i] ?
711 NIR_TRUE : NIR_FALSE;
712 break;
713 default:
714 unreachable("Invalid immediate type");
715 }
716 }
717
718 return load;
719 }
720
721 nir_function_impl *
722 nir_cf_node_get_function(nir_cf_node *node)
723 {
724 while (node->type != nir_cf_node_function) {
725 node = node->parent;
726 }
727
728 return nir_cf_node_as_function(node);
729 }
730
731 /* Reduces a cursor by trying to convert everything to after and trying to
732 * go up to block granularity when possible.
733 */
734 static nir_cursor
735 reduce_cursor(nir_cursor cursor)
736 {
737 switch (cursor.option) {
738 case nir_cursor_before_block:
739 assert(nir_cf_node_prev(&cursor.block->cf_node) == NULL ||
740 nir_cf_node_prev(&cursor.block->cf_node)->type != nir_cf_node_block);
741 if (exec_list_is_empty(&cursor.block->instr_list)) {
742 /* Empty block. After is as good as before. */
743 cursor.option = nir_cursor_after_block;
744 }
745 return cursor;
746
747 case nir_cursor_after_block:
748 return cursor;
749
750 case nir_cursor_before_instr: {
751 nir_instr *prev_instr = nir_instr_prev(cursor.instr);
752 if (prev_instr) {
753 /* Before this instruction is after the previous */
754 cursor.instr = prev_instr;
755 cursor.option = nir_cursor_after_instr;
756 } else {
757 /* No previous instruction. Switch to before block */
758 cursor.block = cursor.instr->block;
759 cursor.option = nir_cursor_before_block;
760 }
761 return reduce_cursor(cursor);
762 }
763
764 case nir_cursor_after_instr:
765 if (nir_instr_next(cursor.instr) == NULL) {
766 /* This is the last instruction, switch to after block */
767 cursor.option = nir_cursor_after_block;
768 cursor.block = cursor.instr->block;
769 }
770 return cursor;
771
772 default:
773 unreachable("Inavlid cursor option");
774 }
775 }
776
777 bool
778 nir_cursors_equal(nir_cursor a, nir_cursor b)
779 {
780 /* Reduced cursors should be unique */
781 a = reduce_cursor(a);
782 b = reduce_cursor(b);
783
784 return a.block == b.block && a.option == b.option;
785 }
786
787 static bool
788 add_use_cb(nir_src *src, void *state)
789 {
790 nir_instr *instr = state;
791
792 src->parent_instr = instr;
793 list_addtail(&src->use_link,
794 src->is_ssa ? &src->ssa->uses : &src->reg.reg->uses);
795
796 return true;
797 }
798
799 static bool
800 add_ssa_def_cb(nir_ssa_def *def, void *state)
801 {
802 nir_instr *instr = state;
803
804 if (instr->block && def->index == UINT_MAX) {
805 nir_function_impl *impl =
806 nir_cf_node_get_function(&instr->block->cf_node);
807
808 def->index = impl->ssa_alloc++;
809 }
810
811 return true;
812 }
813
814 static bool
815 add_reg_def_cb(nir_dest *dest, void *state)
816 {
817 nir_instr *instr = state;
818
819 if (!dest->is_ssa) {
820 dest->reg.parent_instr = instr;
821 list_addtail(&dest->reg.def_link, &dest->reg.reg->defs);
822 }
823
824 return true;
825 }
826
827 static void
828 add_defs_uses(nir_instr *instr)
829 {
830 nir_foreach_src(instr, add_use_cb, instr);
831 nir_foreach_dest(instr, add_reg_def_cb, instr);
832 nir_foreach_ssa_def(instr, add_ssa_def_cb, instr);
833 }
834
835 void
836 nir_instr_insert(nir_cursor cursor, nir_instr *instr)
837 {
838 switch (cursor.option) {
839 case nir_cursor_before_block:
840 /* Only allow inserting jumps into empty blocks. */
841 if (instr->type == nir_instr_type_jump)
842 assert(exec_list_is_empty(&cursor.block->instr_list));
843
844 instr->block = cursor.block;
845 add_defs_uses(instr);
846 exec_list_push_head(&cursor.block->instr_list, &instr->node);
847 break;
848 case nir_cursor_after_block: {
849 /* Inserting instructions after a jump is illegal. */
850 nir_instr *last = nir_block_last_instr(cursor.block);
851 assert(last == NULL || last->type != nir_instr_type_jump);
852 (void) last;
853
854 instr->block = cursor.block;
855 add_defs_uses(instr);
856 exec_list_push_tail(&cursor.block->instr_list, &instr->node);
857 break;
858 }
859 case nir_cursor_before_instr:
860 assert(instr->type != nir_instr_type_jump);
861 instr->block = cursor.instr->block;
862 add_defs_uses(instr);
863 exec_node_insert_node_before(&cursor.instr->node, &instr->node);
864 break;
865 case nir_cursor_after_instr:
866 /* Inserting instructions after a jump is illegal. */
867 assert(cursor.instr->type != nir_instr_type_jump);
868
869 /* Only allow inserting jumps at the end of the block. */
870 if (instr->type == nir_instr_type_jump)
871 assert(cursor.instr == nir_block_last_instr(cursor.instr->block));
872
873 instr->block = cursor.instr->block;
874 add_defs_uses(instr);
875 exec_node_insert_after(&cursor.instr->node, &instr->node);
876 break;
877 }
878
879 if (instr->type == nir_instr_type_jump)
880 nir_handle_add_jump(instr->block);
881 }
882
883 static bool
884 src_is_valid(const nir_src *src)
885 {
886 return src->is_ssa ? (src->ssa != NULL) : (src->reg.reg != NULL);
887 }
888
889 static bool
890 remove_use_cb(nir_src *src, void *state)
891 {
892 if (src_is_valid(src))
893 list_del(&src->use_link);
894
895 return true;
896 }
897
898 static bool
899 remove_def_cb(nir_dest *dest, void *state)
900 {
901 if (!dest->is_ssa)
902 list_del(&dest->reg.def_link);
903
904 return true;
905 }
906
907 static void
908 remove_defs_uses(nir_instr *instr)
909 {
910 nir_foreach_dest(instr, remove_def_cb, instr);
911 nir_foreach_src(instr, remove_use_cb, instr);
912 }
913
914 void nir_instr_remove(nir_instr *instr)
915 {
916 remove_defs_uses(instr);
917 exec_node_remove(&instr->node);
918
919 if (instr->type == nir_instr_type_jump) {
920 nir_jump_instr *jump_instr = nir_instr_as_jump(instr);
921 nir_handle_remove_jump(instr->block, jump_instr->type);
922 }
923 }
924
925 /*@}*/
926
927 void
928 nir_index_local_regs(nir_function_impl *impl)
929 {
930 unsigned index = 0;
931 foreach_list_typed(nir_register, reg, node, &impl->registers) {
932 reg->index = index++;
933 }
934 impl->reg_alloc = index;
935 }
936
937 void
938 nir_index_global_regs(nir_shader *shader)
939 {
940 unsigned index = 0;
941 foreach_list_typed(nir_register, reg, node, &shader->registers) {
942 reg->index = index++;
943 }
944 shader->reg_alloc = index;
945 }
946
947 static bool
948 visit_alu_dest(nir_alu_instr *instr, nir_foreach_dest_cb cb, void *state)
949 {
950 return cb(&instr->dest.dest, state);
951 }
952
953 static bool
954 visit_intrinsic_dest(nir_intrinsic_instr *instr, nir_foreach_dest_cb cb,
955 void *state)
956 {
957 if (nir_intrinsic_infos[instr->intrinsic].has_dest)
958 return cb(&instr->dest, state);
959
960 return true;
961 }
962
963 static bool
964 visit_texture_dest(nir_tex_instr *instr, nir_foreach_dest_cb cb,
965 void *state)
966 {
967 return cb(&instr->dest, state);
968 }
969
970 static bool
971 visit_phi_dest(nir_phi_instr *instr, nir_foreach_dest_cb cb, void *state)
972 {
973 return cb(&instr->dest, state);
974 }
975
976 static bool
977 visit_parallel_copy_dest(nir_parallel_copy_instr *instr,
978 nir_foreach_dest_cb cb, void *state)
979 {
980 nir_foreach_parallel_copy_entry(instr, entry) {
981 if (!cb(&entry->dest, state))
982 return false;
983 }
984
985 return true;
986 }
987
988 bool
989 nir_foreach_dest(nir_instr *instr, nir_foreach_dest_cb cb, void *state)
990 {
991 switch (instr->type) {
992 case nir_instr_type_alu:
993 return visit_alu_dest(nir_instr_as_alu(instr), cb, state);
994 case nir_instr_type_intrinsic:
995 return visit_intrinsic_dest(nir_instr_as_intrinsic(instr), cb, state);
996 case nir_instr_type_tex:
997 return visit_texture_dest(nir_instr_as_tex(instr), cb, state);
998 case nir_instr_type_phi:
999 return visit_phi_dest(nir_instr_as_phi(instr), cb, state);
1000 case nir_instr_type_parallel_copy:
1001 return visit_parallel_copy_dest(nir_instr_as_parallel_copy(instr),
1002 cb, state);
1003
1004 case nir_instr_type_load_const:
1005 case nir_instr_type_ssa_undef:
1006 case nir_instr_type_call:
1007 case nir_instr_type_jump:
1008 break;
1009
1010 default:
1011 unreachable("Invalid instruction type");
1012 break;
1013 }
1014
1015 return true;
1016 }
1017
1018 struct foreach_ssa_def_state {
1019 nir_foreach_ssa_def_cb cb;
1020 void *client_state;
1021 };
1022
1023 static inline bool
1024 nir_ssa_def_visitor(nir_dest *dest, void *void_state)
1025 {
1026 struct foreach_ssa_def_state *state = void_state;
1027
1028 if (dest->is_ssa)
1029 return state->cb(&dest->ssa, state->client_state);
1030 else
1031 return true;
1032 }
1033
1034 bool
1035 nir_foreach_ssa_def(nir_instr *instr, nir_foreach_ssa_def_cb cb, void *state)
1036 {
1037 switch (instr->type) {
1038 case nir_instr_type_alu:
1039 case nir_instr_type_tex:
1040 case nir_instr_type_intrinsic:
1041 case nir_instr_type_phi:
1042 case nir_instr_type_parallel_copy: {
1043 struct foreach_ssa_def_state foreach_state = {cb, state};
1044 return nir_foreach_dest(instr, nir_ssa_def_visitor, &foreach_state);
1045 }
1046
1047 case nir_instr_type_load_const:
1048 return cb(&nir_instr_as_load_const(instr)->def, state);
1049 case nir_instr_type_ssa_undef:
1050 return cb(&nir_instr_as_ssa_undef(instr)->def, state);
1051 case nir_instr_type_call:
1052 case nir_instr_type_jump:
1053 return true;
1054 default:
1055 unreachable("Invalid instruction type");
1056 }
1057 }
1058
1059 static bool
1060 visit_src(nir_src *src, nir_foreach_src_cb cb, void *state)
1061 {
1062 if (!cb(src, state))
1063 return false;
1064 if (!src->is_ssa && src->reg.indirect)
1065 return cb(src->reg.indirect, state);
1066 return true;
1067 }
1068
1069 static bool
1070 visit_deref_array_src(nir_deref_array *deref, nir_foreach_src_cb cb,
1071 void *state)
1072 {
1073 if (deref->deref_array_type == nir_deref_array_type_indirect)
1074 return visit_src(&deref->indirect, cb, state);
1075 return true;
1076 }
1077
1078 static bool
1079 visit_deref_src(nir_deref_var *deref, nir_foreach_src_cb cb, void *state)
1080 {
1081 nir_deref *cur = &deref->deref;
1082 while (cur != NULL) {
1083 if (cur->deref_type == nir_deref_type_array) {
1084 if (!visit_deref_array_src(nir_deref_as_array(cur), cb, state))
1085 return false;
1086 }
1087
1088 cur = cur->child;
1089 }
1090
1091 return true;
1092 }
1093
1094 static bool
1095 visit_alu_src(nir_alu_instr *instr, nir_foreach_src_cb cb, void *state)
1096 {
1097 for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++)
1098 if (!visit_src(&instr->src[i].src, cb, state))
1099 return false;
1100
1101 return true;
1102 }
1103
1104 static bool
1105 visit_tex_src(nir_tex_instr *instr, nir_foreach_src_cb cb, void *state)
1106 {
1107 for (unsigned i = 0; i < instr->num_srcs; i++) {
1108 if (!visit_src(&instr->src[i].src, cb, state))
1109 return false;
1110 }
1111
1112 if (instr->texture != NULL) {
1113 if (!visit_deref_src(instr->texture, cb, state))
1114 return false;
1115 }
1116
1117 if (instr->sampler != NULL) {
1118 if (!visit_deref_src(instr->sampler, cb, state))
1119 return false;
1120 }
1121
1122 return true;
1123 }
1124
1125 static bool
1126 visit_intrinsic_src(nir_intrinsic_instr *instr, nir_foreach_src_cb cb,
1127 void *state)
1128 {
1129 unsigned num_srcs = nir_intrinsic_infos[instr->intrinsic].num_srcs;
1130 for (unsigned i = 0; i < num_srcs; i++) {
1131 if (!visit_src(&instr->src[i], cb, state))
1132 return false;
1133 }
1134
1135 unsigned num_vars =
1136 nir_intrinsic_infos[instr->intrinsic].num_variables;
1137 for (unsigned i = 0; i < num_vars; i++) {
1138 if (!visit_deref_src(instr->variables[i], cb, state))
1139 return false;
1140 }
1141
1142 return true;
1143 }
1144
1145 static bool
1146 visit_call_src(nir_call_instr *instr, nir_foreach_src_cb cb, void *state)
1147 {
1148 return true;
1149 }
1150
1151 static bool
1152 visit_load_const_src(nir_load_const_instr *instr, nir_foreach_src_cb cb,
1153 void *state)
1154 {
1155 return true;
1156 }
1157
1158 static bool
1159 visit_phi_src(nir_phi_instr *instr, nir_foreach_src_cb cb, void *state)
1160 {
1161 nir_foreach_phi_src(instr, src) {
1162 if (!visit_src(&src->src, cb, state))
1163 return false;
1164 }
1165
1166 return true;
1167 }
1168
1169 static bool
1170 visit_parallel_copy_src(nir_parallel_copy_instr *instr,
1171 nir_foreach_src_cb cb, void *state)
1172 {
1173 nir_foreach_parallel_copy_entry(instr, entry) {
1174 if (!visit_src(&entry->src, cb, state))
1175 return false;
1176 }
1177
1178 return true;
1179 }
1180
1181 typedef struct {
1182 void *state;
1183 nir_foreach_src_cb cb;
1184 } visit_dest_indirect_state;
1185
1186 static bool
1187 visit_dest_indirect(nir_dest *dest, void *_state)
1188 {
1189 visit_dest_indirect_state *state = (visit_dest_indirect_state *) _state;
1190
1191 if (!dest->is_ssa && dest->reg.indirect)
1192 return state->cb(dest->reg.indirect, state->state);
1193
1194 return true;
1195 }
1196
1197 bool
1198 nir_foreach_src(nir_instr *instr, nir_foreach_src_cb cb, void *state)
1199 {
1200 switch (instr->type) {
1201 case nir_instr_type_alu:
1202 if (!visit_alu_src(nir_instr_as_alu(instr), cb, state))
1203 return false;
1204 break;
1205 case nir_instr_type_intrinsic:
1206 if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr), cb, state))
1207 return false;
1208 break;
1209 case nir_instr_type_tex:
1210 if (!visit_tex_src(nir_instr_as_tex(instr), cb, state))
1211 return false;
1212 break;
1213 case nir_instr_type_call:
1214 if (!visit_call_src(nir_instr_as_call(instr), cb, state))
1215 return false;
1216 break;
1217 case nir_instr_type_load_const:
1218 if (!visit_load_const_src(nir_instr_as_load_const(instr), cb, state))
1219 return false;
1220 break;
1221 case nir_instr_type_phi:
1222 if (!visit_phi_src(nir_instr_as_phi(instr), cb, state))
1223 return false;
1224 break;
1225 case nir_instr_type_parallel_copy:
1226 if (!visit_parallel_copy_src(nir_instr_as_parallel_copy(instr),
1227 cb, state))
1228 return false;
1229 break;
1230 case nir_instr_type_jump:
1231 case nir_instr_type_ssa_undef:
1232 return true;
1233
1234 default:
1235 unreachable("Invalid instruction type");
1236 break;
1237 }
1238
1239 visit_dest_indirect_state dest_state;
1240 dest_state.state = state;
1241 dest_state.cb = cb;
1242 return nir_foreach_dest(instr, visit_dest_indirect, &dest_state);
1243 }
1244
1245 nir_const_value *
1246 nir_src_as_const_value(nir_src src)
1247 {
1248 if (!src.is_ssa)
1249 return NULL;
1250
1251 if (src.ssa->parent_instr->type != nir_instr_type_load_const)
1252 return NULL;
1253
1254 nir_load_const_instr *load = nir_instr_as_load_const(src.ssa->parent_instr);
1255
1256 return &load->value;
1257 }
1258
1259 /**
1260 * Returns true if the source is known to be dynamically uniform. Otherwise it
1261 * returns false which means it may or may not be dynamically uniform but it
1262 * can't be determined.
1263 */
1264 bool
1265 nir_src_is_dynamically_uniform(nir_src src)
1266 {
1267 if (!src.is_ssa)
1268 return false;
1269
1270 /* Constants are trivially dynamically uniform */
1271 if (src.ssa->parent_instr->type == nir_instr_type_load_const)
1272 return true;
1273
1274 /* As are uniform variables */
1275 if (src.ssa->parent_instr->type == nir_instr_type_intrinsic) {
1276 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(src.ssa->parent_instr);
1277
1278 if (intr->intrinsic == nir_intrinsic_load_uniform)
1279 return true;
1280 }
1281
1282 /* XXX: this could have many more tests, such as when a sampler function is
1283 * called with dynamically uniform arguments.
1284 */
1285 return false;
1286 }
1287
1288 static void
1289 src_remove_all_uses(nir_src *src)
1290 {
1291 for (; src; src = src->is_ssa ? NULL : src->reg.indirect) {
1292 if (!src_is_valid(src))
1293 continue;
1294
1295 list_del(&src->use_link);
1296 }
1297 }
1298
1299 static void
1300 src_add_all_uses(nir_src *src, nir_instr *parent_instr, nir_if *parent_if)
1301 {
1302 for (; src; src = src->is_ssa ? NULL : src->reg.indirect) {
1303 if (!src_is_valid(src))
1304 continue;
1305
1306 if (parent_instr) {
1307 src->parent_instr = parent_instr;
1308 if (src->is_ssa)
1309 list_addtail(&src->use_link, &src->ssa->uses);
1310 else
1311 list_addtail(&src->use_link, &src->reg.reg->uses);
1312 } else {
1313 assert(parent_if);
1314 src->parent_if = parent_if;
1315 if (src->is_ssa)
1316 list_addtail(&src->use_link, &src->ssa->if_uses);
1317 else
1318 list_addtail(&src->use_link, &src->reg.reg->if_uses);
1319 }
1320 }
1321 }
1322
1323 void
1324 nir_instr_rewrite_src(nir_instr *instr, nir_src *src, nir_src new_src)
1325 {
1326 assert(!src_is_valid(src) || src->parent_instr == instr);
1327
1328 src_remove_all_uses(src);
1329 *src = new_src;
1330 src_add_all_uses(src, instr, NULL);
1331 }
1332
1333 void
1334 nir_instr_move_src(nir_instr *dest_instr, nir_src *dest, nir_src *src)
1335 {
1336 assert(!src_is_valid(dest) || dest->parent_instr == dest_instr);
1337
1338 src_remove_all_uses(dest);
1339 src_remove_all_uses(src);
1340 *dest = *src;
1341 *src = NIR_SRC_INIT;
1342 src_add_all_uses(dest, dest_instr, NULL);
1343 }
1344
1345 void
1346 nir_if_rewrite_condition(nir_if *if_stmt, nir_src new_src)
1347 {
1348 nir_src *src = &if_stmt->condition;
1349 assert(!src_is_valid(src) || src->parent_if == if_stmt);
1350
1351 src_remove_all_uses(src);
1352 *src = new_src;
1353 src_add_all_uses(src, NULL, if_stmt);
1354 }
1355
1356 void
1357 nir_instr_rewrite_dest(nir_instr *instr, nir_dest *dest, nir_dest new_dest)
1358 {
1359 if (dest->is_ssa) {
1360 /* We can only overwrite an SSA destination if it has no uses. */
1361 assert(list_empty(&dest->ssa.uses) && list_empty(&dest->ssa.if_uses));
1362 } else {
1363 list_del(&dest->reg.def_link);
1364 if (dest->reg.indirect)
1365 src_remove_all_uses(dest->reg.indirect);
1366 }
1367
1368 /* We can't re-write with an SSA def */
1369 assert(!new_dest.is_ssa);
1370
1371 nir_dest_copy(dest, &new_dest, instr);
1372
1373 dest->reg.parent_instr = instr;
1374 list_addtail(&dest->reg.def_link, &new_dest.reg.reg->defs);
1375
1376 if (dest->reg.indirect)
1377 src_add_all_uses(dest->reg.indirect, instr, NULL);
1378 }
1379
1380 /* note: does *not* take ownership of 'name' */
1381 void
1382 nir_ssa_def_init(nir_instr *instr, nir_ssa_def *def,
1383 unsigned num_components,
1384 unsigned bit_size, const char *name)
1385 {
1386 def->name = ralloc_strdup(instr, name);
1387 def->parent_instr = instr;
1388 list_inithead(&def->uses);
1389 list_inithead(&def->if_uses);
1390 def->num_components = num_components;
1391 def->bit_size = bit_size;
1392
1393 if (instr->block) {
1394 nir_function_impl *impl =
1395 nir_cf_node_get_function(&instr->block->cf_node);
1396
1397 def->index = impl->ssa_alloc++;
1398 } else {
1399 def->index = UINT_MAX;
1400 }
1401 }
1402
1403 /* note: does *not* take ownership of 'name' */
1404 void
1405 nir_ssa_dest_init(nir_instr *instr, nir_dest *dest,
1406 unsigned num_components, unsigned bit_size,
1407 const char *name)
1408 {
1409 dest->is_ssa = true;
1410 nir_ssa_def_init(instr, &dest->ssa, num_components, bit_size, name);
1411 }
1412
1413 void
1414 nir_ssa_def_rewrite_uses(nir_ssa_def *def, nir_src new_src)
1415 {
1416 assert(!new_src.is_ssa || def != new_src.ssa);
1417
1418 nir_foreach_use_safe(def, use_src)
1419 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1420
1421 nir_foreach_if_use_safe(def, use_src)
1422 nir_if_rewrite_condition(use_src->parent_if, new_src);
1423 }
1424
1425 static bool
1426 is_instr_between(nir_instr *start, nir_instr *end, nir_instr *between)
1427 {
1428 assert(start->block == end->block);
1429
1430 if (between->block != start->block)
1431 return false;
1432
1433 /* Search backwards looking for "between" */
1434 while (start != end) {
1435 if (between == end)
1436 return true;
1437
1438 end = nir_instr_prev(end);
1439 assert(end);
1440 }
1441
1442 return false;
1443 }
1444
1445 /* Replaces all uses of the given SSA def with the given source but only if
1446 * the use comes after the after_me instruction. This can be useful if you
1447 * are emitting code to fix up the result of some instruction: you can freely
1448 * use the result in that code and then call rewrite_uses_after and pass the
1449 * last fixup instruction as after_me and it will replace all of the uses you
1450 * want without touching the fixup code.
1451 *
1452 * This function assumes that after_me is in the same block as
1453 * def->parent_instr and that after_me comes after def->parent_instr.
1454 */
1455 void
1456 nir_ssa_def_rewrite_uses_after(nir_ssa_def *def, nir_src new_src,
1457 nir_instr *after_me)
1458 {
1459 assert(!new_src.is_ssa || def != new_src.ssa);
1460
1461 nir_foreach_use_safe(def, use_src) {
1462 assert(use_src->parent_instr != def->parent_instr);
1463 /* Since def already dominates all of its uses, the only way a use can
1464 * not be dominated by after_me is if it is between def and after_me in
1465 * the instruction list.
1466 */
1467 if (!is_instr_between(def->parent_instr, after_me, use_src->parent_instr))
1468 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1469 }
1470
1471 nir_foreach_if_use_safe(def, use_src)
1472 nir_if_rewrite_condition(use_src->parent_if, new_src);
1473 }
1474
1475 static bool foreach_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1476 bool reverse, void *state);
1477
1478 static inline bool
1479 foreach_if(nir_if *if_stmt, nir_foreach_block_cb cb, bool reverse, void *state)
1480 {
1481 if (reverse) {
1482 foreach_list_typed_reverse_safe(nir_cf_node, node, node,
1483 &if_stmt->else_list) {
1484 if (!foreach_cf_node(node, cb, reverse, state))
1485 return false;
1486 }
1487
1488 foreach_list_typed_reverse_safe(nir_cf_node, node, node,
1489 &if_stmt->then_list) {
1490 if (!foreach_cf_node(node, cb, reverse, state))
1491 return false;
1492 }
1493 } else {
1494 foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt->then_list) {
1495 if (!foreach_cf_node(node, cb, reverse, state))
1496 return false;
1497 }
1498
1499 foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt->else_list) {
1500 if (!foreach_cf_node(node, cb, reverse, state))
1501 return false;
1502 }
1503 }
1504
1505 return true;
1506 }
1507
1508 static inline bool
1509 foreach_loop(nir_loop *loop, nir_foreach_block_cb cb, bool reverse, void *state)
1510 {
1511 if (reverse) {
1512 foreach_list_typed_reverse_safe(nir_cf_node, node, node, &loop->body) {
1513 if (!foreach_cf_node(node, cb, reverse, state))
1514 return false;
1515 }
1516 } else {
1517 foreach_list_typed_safe(nir_cf_node, node, node, &loop->body) {
1518 if (!foreach_cf_node(node, cb, reverse, state))
1519 return false;
1520 }
1521 }
1522
1523 return true;
1524 }
1525
1526 static bool
1527 foreach_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1528 bool reverse, void *state)
1529 {
1530 switch (node->type) {
1531 case nir_cf_node_block:
1532 return cb(nir_cf_node_as_block(node), state);
1533 case nir_cf_node_if:
1534 return foreach_if(nir_cf_node_as_if(node), cb, reverse, state);
1535 case nir_cf_node_loop:
1536 return foreach_loop(nir_cf_node_as_loop(node), cb, reverse, state);
1537 break;
1538
1539 default:
1540 unreachable("Invalid CFG node type");
1541 break;
1542 }
1543
1544 return false;
1545 }
1546
1547 bool
1548 nir_foreach_block_in_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1549 void *state)
1550 {
1551 return foreach_cf_node(node, cb, false, state);
1552 }
1553
1554 bool
1555 nir_foreach_block(nir_function_impl *impl, nir_foreach_block_cb cb, void *state)
1556 {
1557 foreach_list_typed_safe(nir_cf_node, node, node, &impl->body) {
1558 if (!foreach_cf_node(node, cb, false, state))
1559 return false;
1560 }
1561
1562 return cb(impl->end_block, state);
1563 }
1564
1565 bool
1566 nir_foreach_block_reverse(nir_function_impl *impl, nir_foreach_block_cb cb,
1567 void *state)
1568 {
1569 if (!cb(impl->end_block, state))
1570 return false;
1571
1572 foreach_list_typed_reverse_safe(nir_cf_node, node, node, &impl->body) {
1573 if (!foreach_cf_node(node, cb, true, state))
1574 return false;
1575 }
1576
1577 return true;
1578 }
1579
1580 nir_if *
1581 nir_block_get_following_if(nir_block *block)
1582 {
1583 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1584 return NULL;
1585
1586 if (nir_cf_node_is_last(&block->cf_node))
1587 return NULL;
1588
1589 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1590
1591 if (next_node->type != nir_cf_node_if)
1592 return NULL;
1593
1594 return nir_cf_node_as_if(next_node);
1595 }
1596
1597 nir_loop *
1598 nir_block_get_following_loop(nir_block *block)
1599 {
1600 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1601 return NULL;
1602
1603 if (nir_cf_node_is_last(&block->cf_node))
1604 return NULL;
1605
1606 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1607
1608 if (next_node->type != nir_cf_node_loop)
1609 return NULL;
1610
1611 return nir_cf_node_as_loop(next_node);
1612 }
1613 static bool
1614 index_block(nir_block *block, void *state)
1615 {
1616 unsigned *index = state;
1617 block->index = (*index)++;
1618 return true;
1619 }
1620
1621 void
1622 nir_index_blocks(nir_function_impl *impl)
1623 {
1624 unsigned index = 0;
1625
1626 if (impl->valid_metadata & nir_metadata_block_index)
1627 return;
1628
1629 nir_foreach_block(impl, index_block, &index);
1630
1631 impl->num_blocks = index;
1632 }
1633
1634 static bool
1635 index_ssa_def_cb(nir_ssa_def *def, void *state)
1636 {
1637 unsigned *index = (unsigned *) state;
1638 def->index = (*index)++;
1639
1640 return true;
1641 }
1642
1643 static bool
1644 index_ssa_block(nir_block *block, void *state)
1645 {
1646 nir_foreach_instr(block, instr)
1647 nir_foreach_ssa_def(instr, index_ssa_def_cb, state);
1648
1649 return true;
1650 }
1651
1652 /**
1653 * The indices are applied top-to-bottom which has the very nice property
1654 * that, if A dominates B, then A->index <= B->index.
1655 */
1656 void
1657 nir_index_ssa_defs(nir_function_impl *impl)
1658 {
1659 unsigned index = 0;
1660 nir_foreach_block(impl, index_ssa_block, &index);
1661 impl->ssa_alloc = index;
1662 }
1663
1664 static bool
1665 index_instrs_block(nir_block *block, void *state)
1666 {
1667 unsigned *index = state;
1668 nir_foreach_instr(block, instr)
1669 instr->index = (*index)++;
1670
1671 return true;
1672 }
1673
1674 /**
1675 * The indices are applied top-to-bottom which has the very nice property
1676 * that, if A dominates B, then A->index <= B->index.
1677 */
1678 unsigned
1679 nir_index_instrs(nir_function_impl *impl)
1680 {
1681 unsigned index = 0;
1682 nir_foreach_block(impl, index_instrs_block, &index);
1683 return index;
1684 }
1685
1686 nir_intrinsic_op
1687 nir_intrinsic_from_system_value(gl_system_value val)
1688 {
1689 switch (val) {
1690 case SYSTEM_VALUE_VERTEX_ID:
1691 return nir_intrinsic_load_vertex_id;
1692 case SYSTEM_VALUE_INSTANCE_ID:
1693 return nir_intrinsic_load_instance_id;
1694 case SYSTEM_VALUE_DRAW_ID:
1695 return nir_intrinsic_load_draw_id;
1696 case SYSTEM_VALUE_BASE_INSTANCE:
1697 return nir_intrinsic_load_base_instance;
1698 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
1699 return nir_intrinsic_load_vertex_id_zero_base;
1700 case SYSTEM_VALUE_BASE_VERTEX:
1701 return nir_intrinsic_load_base_vertex;
1702 case SYSTEM_VALUE_INVOCATION_ID:
1703 return nir_intrinsic_load_invocation_id;
1704 case SYSTEM_VALUE_FRONT_FACE:
1705 return nir_intrinsic_load_front_face;
1706 case SYSTEM_VALUE_SAMPLE_ID:
1707 return nir_intrinsic_load_sample_id;
1708 case SYSTEM_VALUE_SAMPLE_POS:
1709 return nir_intrinsic_load_sample_pos;
1710 case SYSTEM_VALUE_SAMPLE_MASK_IN:
1711 return nir_intrinsic_load_sample_mask_in;
1712 case SYSTEM_VALUE_LOCAL_INVOCATION_ID:
1713 return nir_intrinsic_load_local_invocation_id;
1714 case SYSTEM_VALUE_WORK_GROUP_ID:
1715 return nir_intrinsic_load_work_group_id;
1716 case SYSTEM_VALUE_NUM_WORK_GROUPS:
1717 return nir_intrinsic_load_num_work_groups;
1718 case SYSTEM_VALUE_PRIMITIVE_ID:
1719 return nir_intrinsic_load_primitive_id;
1720 case SYSTEM_VALUE_TESS_COORD:
1721 return nir_intrinsic_load_tess_coord;
1722 case SYSTEM_VALUE_TESS_LEVEL_OUTER:
1723 return nir_intrinsic_load_tess_level_outer;
1724 case SYSTEM_VALUE_TESS_LEVEL_INNER:
1725 return nir_intrinsic_load_tess_level_inner;
1726 case SYSTEM_VALUE_VERTICES_IN:
1727 return nir_intrinsic_load_patch_vertices_in;
1728 case SYSTEM_VALUE_HELPER_INVOCATION:
1729 return nir_intrinsic_load_helper_invocation;
1730 default:
1731 unreachable("system value does not directly correspond to intrinsic");
1732 }
1733 }
1734
1735 gl_system_value
1736 nir_system_value_from_intrinsic(nir_intrinsic_op intrin)
1737 {
1738 switch (intrin) {
1739 case nir_intrinsic_load_vertex_id:
1740 return SYSTEM_VALUE_VERTEX_ID;
1741 case nir_intrinsic_load_instance_id:
1742 return SYSTEM_VALUE_INSTANCE_ID;
1743 case nir_intrinsic_load_draw_id:
1744 return SYSTEM_VALUE_DRAW_ID;
1745 case nir_intrinsic_load_base_instance:
1746 return SYSTEM_VALUE_BASE_INSTANCE;
1747 case nir_intrinsic_load_vertex_id_zero_base:
1748 return SYSTEM_VALUE_VERTEX_ID_ZERO_BASE;
1749 case nir_intrinsic_load_base_vertex:
1750 return SYSTEM_VALUE_BASE_VERTEX;
1751 case nir_intrinsic_load_invocation_id:
1752 return SYSTEM_VALUE_INVOCATION_ID;
1753 case nir_intrinsic_load_front_face:
1754 return SYSTEM_VALUE_FRONT_FACE;
1755 case nir_intrinsic_load_sample_id:
1756 return SYSTEM_VALUE_SAMPLE_ID;
1757 case nir_intrinsic_load_sample_pos:
1758 return SYSTEM_VALUE_SAMPLE_POS;
1759 case nir_intrinsic_load_sample_mask_in:
1760 return SYSTEM_VALUE_SAMPLE_MASK_IN;
1761 case nir_intrinsic_load_local_invocation_id:
1762 return SYSTEM_VALUE_LOCAL_INVOCATION_ID;
1763 case nir_intrinsic_load_num_work_groups:
1764 return SYSTEM_VALUE_NUM_WORK_GROUPS;
1765 case nir_intrinsic_load_work_group_id:
1766 return SYSTEM_VALUE_WORK_GROUP_ID;
1767 case nir_intrinsic_load_primitive_id:
1768 return SYSTEM_VALUE_PRIMITIVE_ID;
1769 case nir_intrinsic_load_tess_coord:
1770 return SYSTEM_VALUE_TESS_COORD;
1771 case nir_intrinsic_load_tess_level_outer:
1772 return SYSTEM_VALUE_TESS_LEVEL_OUTER;
1773 case nir_intrinsic_load_tess_level_inner:
1774 return SYSTEM_VALUE_TESS_LEVEL_INNER;
1775 case nir_intrinsic_load_patch_vertices_in:
1776 return SYSTEM_VALUE_VERTICES_IN;
1777 case nir_intrinsic_load_helper_invocation:
1778 return SYSTEM_VALUE_HELPER_INVOCATION;
1779 default:
1780 unreachable("intrinsic doesn't produce a system value");
1781 }
1782 }