Merge remote-tracking branch 'mesa-public/master' into vulkan
[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->num_array_elems = 0;
74 reg->is_packed = false;
75 reg->name = NULL;
76
77 exec_list_push_tail(list, &reg->node);
78
79 return reg;
80 }
81
82 nir_register *
83 nir_global_reg_create(nir_shader *shader)
84 {
85 nir_register *reg = reg_create(shader, &shader->registers);
86 reg->index = shader->reg_alloc++;
87 reg->is_global = true;
88
89 return reg;
90 }
91
92 nir_register *
93 nir_local_reg_create(nir_function_impl *impl)
94 {
95 nir_register *reg = reg_create(ralloc_parent(impl), &impl->registers);
96 reg->index = impl->reg_alloc++;
97 reg->is_global = false;
98
99 return reg;
100 }
101
102 void
103 nir_reg_remove(nir_register *reg)
104 {
105 exec_node_remove(&reg->node);
106 }
107
108 void
109 nir_shader_add_variable(nir_shader *shader, nir_variable *var)
110 {
111 switch (var->data.mode) {
112 case nir_var_all:
113 assert(!"invalid mode");
114 break;
115
116 case nir_var_local:
117 assert(!"nir_shader_add_variable cannot be used for local variables");
118 break;
119
120 case nir_var_param:
121 assert(!"nir_shader_add_variable cannot be used for function parameters");
122 break;
123
124 case nir_var_global:
125 exec_list_push_tail(&shader->globals, &var->node);
126 break;
127
128 case nir_var_shader_in:
129 exec_list_push_tail(&shader->inputs, &var->node);
130 break;
131
132 case nir_var_shader_out:
133 exec_list_push_tail(&shader->outputs, &var->node);
134 break;
135
136 case nir_var_uniform:
137 case nir_var_shader_storage:
138 exec_list_push_tail(&shader->uniforms, &var->node);
139 break;
140
141 case nir_var_shared:
142 assert(shader->stage == MESA_SHADER_COMPUTE);
143 exec_list_push_tail(&shader->shared, &var->node);
144 break;
145
146 case nir_var_system_value:
147 exec_list_push_tail(&shader->system_values, &var->node);
148 break;
149 }
150 }
151
152 nir_variable *
153 nir_variable_create(nir_shader *shader, nir_variable_mode mode,
154 const struct glsl_type *type, const char *name)
155 {
156 nir_variable *var = rzalloc(shader, nir_variable);
157 var->name = ralloc_strdup(var, name);
158 var->type = type;
159 var->data.mode = mode;
160
161 if ((mode == nir_var_shader_in && shader->stage != MESA_SHADER_VERTEX) ||
162 (mode == nir_var_shader_out && shader->stage != MESA_SHADER_FRAGMENT))
163 var->data.interpolation = INTERP_QUALIFIER_SMOOTH;
164
165 if (mode == nir_var_shader_in || mode == nir_var_uniform)
166 var->data.read_only = true;
167
168 nir_shader_add_variable(shader, var);
169
170 return var;
171 }
172
173 nir_variable *
174 nir_local_variable_create(nir_function_impl *impl,
175 const struct glsl_type *type, const char *name)
176 {
177 nir_variable *var = rzalloc(impl->function->shader, nir_variable);
178 var->name = ralloc_strdup(var, name);
179 var->type = type;
180 var->data.mode = nir_var_local;
181
182 nir_function_impl_add_variable(impl, var);
183
184 return var;
185 }
186
187 nir_function *
188 nir_function_create(nir_shader *shader, const char *name)
189 {
190 nir_function *func = ralloc(shader, nir_function);
191
192 exec_list_push_tail(&shader->functions, &func->node);
193
194 func->name = ralloc_strdup(func, name);
195 func->shader = shader;
196 func->num_params = 0;
197 func->params = NULL;
198 func->return_type = glsl_void_type();
199 func->impl = NULL;
200
201 return func;
202 }
203
204 void nir_src_copy(nir_src *dest, const nir_src *src, void *mem_ctx)
205 {
206 dest->is_ssa = src->is_ssa;
207 if (src->is_ssa) {
208 dest->ssa = src->ssa;
209 } else {
210 dest->reg.base_offset = src->reg.base_offset;
211 dest->reg.reg = src->reg.reg;
212 if (src->reg.indirect) {
213 dest->reg.indirect = ralloc(mem_ctx, nir_src);
214 nir_src_copy(dest->reg.indirect, src->reg.indirect, mem_ctx);
215 } else {
216 dest->reg.indirect = NULL;
217 }
218 }
219 }
220
221 void nir_dest_copy(nir_dest *dest, const nir_dest *src, nir_instr *instr)
222 {
223 /* Copying an SSA definition makes no sense whatsoever. */
224 assert(!src->is_ssa);
225
226 dest->is_ssa = false;
227
228 dest->reg.base_offset = src->reg.base_offset;
229 dest->reg.reg = src->reg.reg;
230 if (src->reg.indirect) {
231 dest->reg.indirect = ralloc(instr, nir_src);
232 nir_src_copy(dest->reg.indirect, src->reg.indirect, instr);
233 } else {
234 dest->reg.indirect = NULL;
235 }
236 }
237
238 void
239 nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src,
240 nir_alu_instr *instr)
241 {
242 nir_src_copy(&dest->src, &src->src, &instr->instr);
243 dest->abs = src->abs;
244 dest->negate = src->negate;
245 for (unsigned i = 0; i < 4; i++)
246 dest->swizzle[i] = src->swizzle[i];
247 }
248
249 void
250 nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src,
251 nir_alu_instr *instr)
252 {
253 nir_dest_copy(&dest->dest, &src->dest, &instr->instr);
254 dest->write_mask = src->write_mask;
255 dest->saturate = src->saturate;
256 }
257
258
259 static void
260 cf_init(nir_cf_node *node, nir_cf_node_type type)
261 {
262 exec_node_init(&node->node);
263 node->parent = NULL;
264 node->type = type;
265 }
266
267 nir_function_impl *
268 nir_function_impl_create_bare(nir_shader *shader)
269 {
270 nir_function_impl *impl = ralloc(shader, nir_function_impl);
271
272 impl->function = NULL;
273
274 cf_init(&impl->cf_node, nir_cf_node_function);
275
276 exec_list_make_empty(&impl->body);
277 exec_list_make_empty(&impl->registers);
278 exec_list_make_empty(&impl->locals);
279 impl->num_params = 0;
280 impl->params = NULL;
281 impl->return_var = NULL;
282 impl->reg_alloc = 0;
283 impl->ssa_alloc = 0;
284 impl->valid_metadata = nir_metadata_none;
285
286 /* create start & end blocks */
287 nir_block *start_block = nir_block_create(shader);
288 nir_block *end_block = nir_block_create(shader);
289 start_block->cf_node.parent = &impl->cf_node;
290 end_block->cf_node.parent = &impl->cf_node;
291 impl->end_block = end_block;
292
293 exec_list_push_tail(&impl->body, &start_block->cf_node.node);
294
295 start_block->successors[0] = end_block;
296 _mesa_set_add(end_block->predecessors, start_block);
297 return impl;
298 }
299
300 nir_function_impl *
301 nir_function_impl_create(nir_function *function)
302 {
303 assert(function->impl == NULL);
304
305 nir_function_impl *impl = nir_function_impl_create_bare(function->shader);
306
307 function->impl = impl;
308 impl->function = function;
309
310 impl->num_params = function->num_params;
311 impl->params = ralloc_array(function->shader,
312 nir_variable *, impl->num_params);
313
314 for (unsigned i = 0; i < impl->num_params; i++) {
315 impl->params[i] = rzalloc(function->shader, nir_variable);
316 impl->params[i]->type = function->params[i].type;
317 impl->params[i]->data.mode = nir_var_param;
318 impl->params[i]->data.location = i;
319 }
320
321 if (!glsl_type_is_void(function->return_type)) {
322 impl->return_var = rzalloc(function->shader, nir_variable);
323 impl->return_var->type = function->return_type;
324 impl->return_var->data.mode = nir_var_param;
325 impl->return_var->data.location = -1;
326 }
327
328 return impl;
329 }
330
331 nir_block *
332 nir_block_create(nir_shader *shader)
333 {
334 nir_block *block = ralloc(shader, nir_block);
335
336 cf_init(&block->cf_node, nir_cf_node_block);
337
338 block->successors[0] = block->successors[1] = NULL;
339 block->predecessors = _mesa_set_create(block, _mesa_hash_pointer,
340 _mesa_key_pointer_equal);
341 block->imm_dom = NULL;
342 /* XXX maybe it would be worth it to defer allocation? This
343 * way it doesn't get allocated for shader ref's that never run
344 * nir_calc_dominance? For example, state-tracker creates an
345 * initial IR, clones that, runs appropriate lowering pass, passes
346 * to driver which does common lowering/opt, and then stores ref
347 * which is later used to do state specific lowering and futher
348 * opt. Do any of the references not need dominance metadata?
349 */
350 block->dom_frontier = _mesa_set_create(block, _mesa_hash_pointer,
351 _mesa_key_pointer_equal);
352
353 exec_list_make_empty(&block->instr_list);
354
355 return block;
356 }
357
358 static inline void
359 src_init(nir_src *src)
360 {
361 src->is_ssa = false;
362 src->reg.reg = NULL;
363 src->reg.indirect = NULL;
364 src->reg.base_offset = 0;
365 }
366
367 nir_if *
368 nir_if_create(nir_shader *shader)
369 {
370 nir_if *if_stmt = ralloc(shader, nir_if);
371
372 cf_init(&if_stmt->cf_node, nir_cf_node_if);
373 src_init(&if_stmt->condition);
374
375 nir_block *then = nir_block_create(shader);
376 exec_list_make_empty(&if_stmt->then_list);
377 exec_list_push_tail(&if_stmt->then_list, &then->cf_node.node);
378 then->cf_node.parent = &if_stmt->cf_node;
379
380 nir_block *else_stmt = nir_block_create(shader);
381 exec_list_make_empty(&if_stmt->else_list);
382 exec_list_push_tail(&if_stmt->else_list, &else_stmt->cf_node.node);
383 else_stmt->cf_node.parent = &if_stmt->cf_node;
384
385 return if_stmt;
386 }
387
388 nir_loop *
389 nir_loop_create(nir_shader *shader)
390 {
391 nir_loop *loop = ralloc(shader, nir_loop);
392
393 cf_init(&loop->cf_node, nir_cf_node_loop);
394
395 nir_block *body = nir_block_create(shader);
396 exec_list_make_empty(&loop->body);
397 exec_list_push_tail(&loop->body, &body->cf_node.node);
398 body->cf_node.parent = &loop->cf_node;
399
400 body->successors[0] = body;
401 _mesa_set_add(body->predecessors, body);
402
403 return loop;
404 }
405
406 static void
407 instr_init(nir_instr *instr, nir_instr_type type)
408 {
409 instr->type = type;
410 instr->block = NULL;
411 exec_node_init(&instr->node);
412 }
413
414 static void
415 dest_init(nir_dest *dest)
416 {
417 dest->is_ssa = false;
418 dest->reg.reg = NULL;
419 dest->reg.indirect = NULL;
420 dest->reg.base_offset = 0;
421 }
422
423 static void
424 alu_dest_init(nir_alu_dest *dest)
425 {
426 dest_init(&dest->dest);
427 dest->saturate = false;
428 dest->write_mask = 0xf;
429 }
430
431 static void
432 alu_src_init(nir_alu_src *src)
433 {
434 src_init(&src->src);
435 src->abs = src->negate = false;
436 src->swizzle[0] = 0;
437 src->swizzle[1] = 1;
438 src->swizzle[2] = 2;
439 src->swizzle[3] = 3;
440 }
441
442 nir_alu_instr *
443 nir_alu_instr_create(nir_shader *shader, nir_op op)
444 {
445 unsigned num_srcs = nir_op_infos[op].num_inputs;
446 nir_alu_instr *instr =
447 ralloc_size(shader,
448 sizeof(nir_alu_instr) + num_srcs * sizeof(nir_alu_src));
449
450 instr_init(&instr->instr, nir_instr_type_alu);
451 instr->op = op;
452 alu_dest_init(&instr->dest);
453 for (unsigned i = 0; i < num_srcs; i++)
454 alu_src_init(&instr->src[i]);
455
456 return instr;
457 }
458
459 nir_jump_instr *
460 nir_jump_instr_create(nir_shader *shader, nir_jump_type type)
461 {
462 nir_jump_instr *instr = ralloc(shader, nir_jump_instr);
463 instr_init(&instr->instr, nir_instr_type_jump);
464 instr->type = type;
465 return instr;
466 }
467
468 nir_load_const_instr *
469 nir_load_const_instr_create(nir_shader *shader, unsigned num_components)
470 {
471 nir_load_const_instr *instr = ralloc(shader, nir_load_const_instr);
472 instr_init(&instr->instr, nir_instr_type_load_const);
473
474 nir_ssa_def_init(&instr->instr, &instr->def, num_components, NULL);
475
476 return instr;
477 }
478
479 nir_intrinsic_instr *
480 nir_intrinsic_instr_create(nir_shader *shader, nir_intrinsic_op op)
481 {
482 unsigned num_srcs = nir_intrinsic_infos[op].num_srcs;
483 nir_intrinsic_instr *instr =
484 ralloc_size(shader,
485 sizeof(nir_intrinsic_instr) + num_srcs * sizeof(nir_src));
486
487 instr_init(&instr->instr, nir_instr_type_intrinsic);
488 instr->intrinsic = op;
489
490 if (nir_intrinsic_infos[op].has_dest)
491 dest_init(&instr->dest);
492
493 for (unsigned i = 0; i < num_srcs; i++)
494 src_init(&instr->src[i]);
495
496 return instr;
497 }
498
499 nir_call_instr *
500 nir_call_instr_create(nir_shader *shader, nir_function *callee)
501 {
502 nir_call_instr *instr = ralloc(shader, nir_call_instr);
503 instr_init(&instr->instr, nir_instr_type_call);
504
505 instr->callee = callee;
506 instr->num_params = callee->num_params;
507 instr->params = ralloc_array(instr, nir_deref_var *, instr->num_params);
508 instr->return_deref = NULL;
509
510 return instr;
511 }
512
513 nir_tex_instr *
514 nir_tex_instr_create(nir_shader *shader, unsigned num_srcs)
515 {
516 nir_tex_instr *instr = rzalloc(shader, nir_tex_instr);
517 instr_init(&instr->instr, nir_instr_type_tex);
518
519 dest_init(&instr->dest);
520
521 instr->num_srcs = num_srcs;
522 instr->src = ralloc_array(instr, nir_tex_src, num_srcs);
523 for (unsigned i = 0; i < num_srcs; i++)
524 src_init(&instr->src[i].src);
525
526 instr->texture_index = 0;
527 instr->texture_array_size = 0;
528 instr->texture = NULL;
529 instr->sampler_index = 0;
530 instr->sampler = NULL;
531
532 return instr;
533 }
534
535 nir_phi_instr *
536 nir_phi_instr_create(nir_shader *shader)
537 {
538 nir_phi_instr *instr = ralloc(shader, nir_phi_instr);
539 instr_init(&instr->instr, nir_instr_type_phi);
540
541 dest_init(&instr->dest);
542 exec_list_make_empty(&instr->srcs);
543 return instr;
544 }
545
546 nir_parallel_copy_instr *
547 nir_parallel_copy_instr_create(nir_shader *shader)
548 {
549 nir_parallel_copy_instr *instr = ralloc(shader, nir_parallel_copy_instr);
550 instr_init(&instr->instr, nir_instr_type_parallel_copy);
551
552 exec_list_make_empty(&instr->entries);
553
554 return instr;
555 }
556
557 nir_ssa_undef_instr *
558 nir_ssa_undef_instr_create(nir_shader *shader, unsigned num_components)
559 {
560 nir_ssa_undef_instr *instr = ralloc(shader, nir_ssa_undef_instr);
561 instr_init(&instr->instr, nir_instr_type_ssa_undef);
562
563 nir_ssa_def_init(&instr->instr, &instr->def, num_components, NULL);
564
565 return instr;
566 }
567
568 nir_deref_var *
569 nir_deref_var_create(void *mem_ctx, nir_variable *var)
570 {
571 nir_deref_var *deref = ralloc(mem_ctx, nir_deref_var);
572 deref->deref.deref_type = nir_deref_type_var;
573 deref->deref.child = NULL;
574 deref->deref.type = var->type;
575 deref->var = var;
576 return deref;
577 }
578
579 nir_deref_array *
580 nir_deref_array_create(void *mem_ctx)
581 {
582 nir_deref_array *deref = ralloc(mem_ctx, nir_deref_array);
583 deref->deref.deref_type = nir_deref_type_array;
584 deref->deref.child = NULL;
585 deref->deref_array_type = nir_deref_array_type_direct;
586 src_init(&deref->indirect);
587 deref->base_offset = 0;
588 return deref;
589 }
590
591 nir_deref_struct *
592 nir_deref_struct_create(void *mem_ctx, unsigned field_index)
593 {
594 nir_deref_struct *deref = ralloc(mem_ctx, nir_deref_struct);
595 deref->deref.deref_type = nir_deref_type_struct;
596 deref->deref.child = NULL;
597 deref->index = field_index;
598 return deref;
599 }
600
601 static nir_deref_var *
602 copy_deref_var(void *mem_ctx, nir_deref_var *deref)
603 {
604 nir_deref_var *ret = nir_deref_var_create(mem_ctx, deref->var);
605 ret->deref.type = deref->deref.type;
606 if (deref->deref.child)
607 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
608 return ret;
609 }
610
611 static nir_deref_array *
612 copy_deref_array(void *mem_ctx, nir_deref_array *deref)
613 {
614 nir_deref_array *ret = nir_deref_array_create(mem_ctx);
615 ret->base_offset = deref->base_offset;
616 ret->deref_array_type = deref->deref_array_type;
617 if (deref->deref_array_type == nir_deref_array_type_indirect) {
618 nir_src_copy(&ret->indirect, &deref->indirect, mem_ctx);
619 }
620 ret->deref.type = deref->deref.type;
621 if (deref->deref.child)
622 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
623 return ret;
624 }
625
626 static nir_deref_struct *
627 copy_deref_struct(void *mem_ctx, nir_deref_struct *deref)
628 {
629 nir_deref_struct *ret = nir_deref_struct_create(mem_ctx, deref->index);
630 ret->deref.type = deref->deref.type;
631 if (deref->deref.child)
632 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
633 return ret;
634 }
635
636 nir_deref *
637 nir_copy_deref(void *mem_ctx, nir_deref *deref)
638 {
639 switch (deref->deref_type) {
640 case nir_deref_type_var:
641 return &copy_deref_var(mem_ctx, nir_deref_as_var(deref))->deref;
642 case nir_deref_type_array:
643 return &copy_deref_array(mem_ctx, nir_deref_as_array(deref))->deref;
644 case nir_deref_type_struct:
645 return &copy_deref_struct(mem_ctx, nir_deref_as_struct(deref))->deref;
646 default:
647 unreachable("Invalid dereference type");
648 }
649
650 return NULL;
651 }
652
653 /* Returns a load_const instruction that represents the constant
654 * initializer for the given deref chain. The caller is responsible for
655 * ensuring that there actually is a constant initializer.
656 */
657 nir_load_const_instr *
658 nir_deref_get_const_initializer_load(nir_shader *shader, nir_deref_var *deref)
659 {
660 nir_constant *constant = deref->var->constant_initializer;
661 assert(constant);
662
663 const nir_deref *tail = &deref->deref;
664 unsigned matrix_offset = 0;
665 while (tail->child) {
666 switch (tail->child->deref_type) {
667 case nir_deref_type_array: {
668 nir_deref_array *arr = nir_deref_as_array(tail->child);
669 assert(arr->deref_array_type == nir_deref_array_type_direct);
670 if (glsl_type_is_matrix(tail->type)) {
671 assert(arr->deref.child == NULL);
672 matrix_offset = arr->base_offset;
673 } else {
674 constant = constant->elements[arr->base_offset];
675 }
676 break;
677 }
678
679 case nir_deref_type_struct: {
680 constant = constant->elements[nir_deref_as_struct(tail->child)->index];
681 break;
682 }
683
684 default:
685 unreachable("Invalid deref child type");
686 }
687
688 tail = tail->child;
689 }
690
691 nir_load_const_instr *load =
692 nir_load_const_instr_create(shader, glsl_get_vector_elements(tail->type));
693
694 matrix_offset *= load->def.num_components;
695 for (unsigned i = 0; i < load->def.num_components; i++) {
696 switch (glsl_get_base_type(tail->type)) {
697 case GLSL_TYPE_FLOAT:
698 case GLSL_TYPE_INT:
699 case GLSL_TYPE_UINT:
700 load->value.u[i] = constant->value.u[matrix_offset + i];
701 break;
702 case GLSL_TYPE_BOOL:
703 load->value.u[i] = constant->value.b[matrix_offset + i] ?
704 NIR_TRUE : NIR_FALSE;
705 break;
706 default:
707 unreachable("Invalid immediate type");
708 }
709 }
710
711 return load;
712 }
713
714 nir_function_impl *
715 nir_cf_node_get_function(nir_cf_node *node)
716 {
717 while (node->type != nir_cf_node_function) {
718 node = node->parent;
719 }
720
721 return nir_cf_node_as_function(node);
722 }
723
724 /* Reduces a cursor by trying to convert everything to after and trying to
725 * go up to block granularity when possible.
726 */
727 static nir_cursor
728 reduce_cursor(nir_cursor cursor)
729 {
730 switch (cursor.option) {
731 case nir_cursor_before_block:
732 if (exec_list_is_empty(&cursor.block->instr_list)) {
733 /* Empty block. After is as good as before. */
734 cursor.option = nir_cursor_after_block;
735 } else {
736 /* Try to switch to after the previous block if there is one.
737 * (This isn't likely, but it can happen.)
738 */
739 nir_cf_node *prev_node = nir_cf_node_prev(&cursor.block->cf_node);
740 if (prev_node && prev_node->type == nir_cf_node_block) {
741 cursor.block = nir_cf_node_as_block(prev_node);
742 cursor.option = nir_cursor_after_block;
743 }
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 void
1381 nir_ssa_def_init(nir_instr *instr, nir_ssa_def *def,
1382 unsigned num_components, const char *name)
1383 {
1384 def->name = name;
1385 def->parent_instr = instr;
1386 list_inithead(&def->uses);
1387 list_inithead(&def->if_uses);
1388 def->num_components = num_components;
1389
1390 if (instr->block) {
1391 nir_function_impl *impl =
1392 nir_cf_node_get_function(&instr->block->cf_node);
1393
1394 def->index = impl->ssa_alloc++;
1395 } else {
1396 def->index = UINT_MAX;
1397 }
1398 }
1399
1400 void
1401 nir_ssa_dest_init(nir_instr *instr, nir_dest *dest,
1402 unsigned num_components, const char *name)
1403 {
1404 dest->is_ssa = true;
1405 nir_ssa_def_init(instr, &dest->ssa, num_components, name);
1406 }
1407
1408 void
1409 nir_ssa_def_rewrite_uses(nir_ssa_def *def, nir_src new_src)
1410 {
1411 assert(!new_src.is_ssa || def != new_src.ssa);
1412
1413 nir_foreach_use_safe(def, use_src)
1414 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1415
1416 nir_foreach_if_use_safe(def, use_src)
1417 nir_if_rewrite_condition(use_src->parent_if, new_src);
1418 }
1419
1420 static bool
1421 is_instr_between(nir_instr *start, nir_instr *end, nir_instr *between)
1422 {
1423 assert(start->block == end->block);
1424
1425 if (between->block != start->block)
1426 return false;
1427
1428 /* Search backwards looking for "between" */
1429 while (start != end) {
1430 if (between == end)
1431 return true;
1432
1433 end = nir_instr_prev(end);
1434 assert(end);
1435 }
1436
1437 return false;
1438 }
1439
1440 /* Replaces all uses of the given SSA def with the given source but only if
1441 * the use comes after the after_me instruction. This can be useful if you
1442 * are emitting code to fix up the result of some instruction: you can freely
1443 * use the result in that code and then call rewrite_uses_after and pass the
1444 * last fixup instruction as after_me and it will replace all of the uses you
1445 * want without touching the fixup code.
1446 *
1447 * This function assumes that after_me is in the same block as
1448 * def->parent_instr and that after_me comes after def->parent_instr.
1449 */
1450 void
1451 nir_ssa_def_rewrite_uses_after(nir_ssa_def *def, nir_src new_src,
1452 nir_instr *after_me)
1453 {
1454 assert(!new_src.is_ssa || def != new_src.ssa);
1455
1456 nir_foreach_use_safe(def, use_src) {
1457 assert(use_src->parent_instr != def->parent_instr);
1458 /* Since def already dominates all of its uses, the only way a use can
1459 * not be dominated by after_me is if it is between def and after_me in
1460 * the instruction list.
1461 */
1462 if (!is_instr_between(def->parent_instr, after_me, use_src->parent_instr))
1463 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1464 }
1465
1466 nir_foreach_if_use_safe(def, use_src)
1467 nir_if_rewrite_condition(use_src->parent_if, new_src);
1468 }
1469
1470 static bool foreach_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1471 bool reverse, void *state);
1472
1473 static inline bool
1474 foreach_if(nir_if *if_stmt, nir_foreach_block_cb cb, bool reverse, void *state)
1475 {
1476 if (reverse) {
1477 foreach_list_typed_reverse_safe(nir_cf_node, node, node,
1478 &if_stmt->else_list) {
1479 if (!foreach_cf_node(node, cb, reverse, state))
1480 return false;
1481 }
1482
1483 foreach_list_typed_reverse_safe(nir_cf_node, node, node,
1484 &if_stmt->then_list) {
1485 if (!foreach_cf_node(node, cb, reverse, state))
1486 return false;
1487 }
1488 } else {
1489 foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt->then_list) {
1490 if (!foreach_cf_node(node, cb, reverse, state))
1491 return false;
1492 }
1493
1494 foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt->else_list) {
1495 if (!foreach_cf_node(node, cb, reverse, state))
1496 return false;
1497 }
1498 }
1499
1500 return true;
1501 }
1502
1503 static inline bool
1504 foreach_loop(nir_loop *loop, nir_foreach_block_cb cb, bool reverse, void *state)
1505 {
1506 if (reverse) {
1507 foreach_list_typed_reverse_safe(nir_cf_node, node, node, &loop->body) {
1508 if (!foreach_cf_node(node, cb, reverse, state))
1509 return false;
1510 }
1511 } else {
1512 foreach_list_typed_safe(nir_cf_node, node, node, &loop->body) {
1513 if (!foreach_cf_node(node, cb, reverse, state))
1514 return false;
1515 }
1516 }
1517
1518 return true;
1519 }
1520
1521 static bool
1522 foreach_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1523 bool reverse, void *state)
1524 {
1525 switch (node->type) {
1526 case nir_cf_node_block:
1527 return cb(nir_cf_node_as_block(node), state);
1528 case nir_cf_node_if:
1529 return foreach_if(nir_cf_node_as_if(node), cb, reverse, state);
1530 case nir_cf_node_loop:
1531 return foreach_loop(nir_cf_node_as_loop(node), cb, reverse, state);
1532 break;
1533
1534 default:
1535 unreachable("Invalid CFG node type");
1536 break;
1537 }
1538
1539 return false;
1540 }
1541
1542 bool
1543 nir_foreach_block_in_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1544 void *state)
1545 {
1546 return foreach_cf_node(node, cb, false, state);
1547 }
1548
1549 bool
1550 nir_foreach_block(nir_function_impl *impl, nir_foreach_block_cb cb, void *state)
1551 {
1552 foreach_list_typed_safe(nir_cf_node, node, node, &impl->body) {
1553 if (!foreach_cf_node(node, cb, false, state))
1554 return false;
1555 }
1556
1557 return cb(impl->end_block, state);
1558 }
1559
1560 bool
1561 nir_foreach_block_reverse(nir_function_impl *impl, nir_foreach_block_cb cb,
1562 void *state)
1563 {
1564 if (!cb(impl->end_block, state))
1565 return false;
1566
1567 foreach_list_typed_reverse_safe(nir_cf_node, node, node, &impl->body) {
1568 if (!foreach_cf_node(node, cb, true, state))
1569 return false;
1570 }
1571
1572 return true;
1573 }
1574
1575 nir_if *
1576 nir_block_get_following_if(nir_block *block)
1577 {
1578 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1579 return NULL;
1580
1581 if (nir_cf_node_is_last(&block->cf_node))
1582 return NULL;
1583
1584 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1585
1586 if (next_node->type != nir_cf_node_if)
1587 return NULL;
1588
1589 return nir_cf_node_as_if(next_node);
1590 }
1591
1592 nir_loop *
1593 nir_block_get_following_loop(nir_block *block)
1594 {
1595 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1596 return NULL;
1597
1598 if (nir_cf_node_is_last(&block->cf_node))
1599 return NULL;
1600
1601 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1602
1603 if (next_node->type != nir_cf_node_loop)
1604 return NULL;
1605
1606 return nir_cf_node_as_loop(next_node);
1607 }
1608 static bool
1609 index_block(nir_block *block, void *state)
1610 {
1611 unsigned *index = state;
1612 block->index = (*index)++;
1613 return true;
1614 }
1615
1616 void
1617 nir_index_blocks(nir_function_impl *impl)
1618 {
1619 unsigned index = 0;
1620
1621 if (impl->valid_metadata & nir_metadata_block_index)
1622 return;
1623
1624 nir_foreach_block(impl, index_block, &index);
1625
1626 impl->num_blocks = index;
1627 }
1628
1629 static bool
1630 index_ssa_def_cb(nir_ssa_def *def, void *state)
1631 {
1632 unsigned *index = (unsigned *) state;
1633 def->index = (*index)++;
1634
1635 return true;
1636 }
1637
1638 static bool
1639 index_ssa_block(nir_block *block, void *state)
1640 {
1641 nir_foreach_instr(block, instr)
1642 nir_foreach_ssa_def(instr, index_ssa_def_cb, state);
1643
1644 return true;
1645 }
1646
1647 /**
1648 * The indices are applied top-to-bottom which has the very nice property
1649 * that, if A dominates B, then A->index <= B->index.
1650 */
1651 void
1652 nir_index_ssa_defs(nir_function_impl *impl)
1653 {
1654 unsigned index = 0;
1655 nir_foreach_block(impl, index_ssa_block, &index);
1656 impl->ssa_alloc = index;
1657 }
1658
1659 static bool
1660 index_instrs_block(nir_block *block, void *state)
1661 {
1662 unsigned *index = state;
1663 nir_foreach_instr(block, instr)
1664 instr->index = (*index)++;
1665
1666 return true;
1667 }
1668
1669 /**
1670 * The indices are applied top-to-bottom which has the very nice property
1671 * that, if A dominates B, then A->index <= B->index.
1672 */
1673 unsigned
1674 nir_index_instrs(nir_function_impl *impl)
1675 {
1676 unsigned index = 0;
1677 nir_foreach_block(impl, index_instrs_block, &index);
1678 return index;
1679 }
1680
1681 nir_intrinsic_op
1682 nir_intrinsic_from_system_value(gl_system_value val)
1683 {
1684 switch (val) {
1685 case SYSTEM_VALUE_VERTEX_ID:
1686 return nir_intrinsic_load_vertex_id;
1687 case SYSTEM_VALUE_INSTANCE_ID:
1688 return nir_intrinsic_load_instance_id;
1689 case SYSTEM_VALUE_DRAW_ID:
1690 return nir_intrinsic_load_draw_id;
1691 case SYSTEM_VALUE_BASE_INSTANCE:
1692 return nir_intrinsic_load_base_instance;
1693 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
1694 return nir_intrinsic_load_vertex_id_zero_base;
1695 case SYSTEM_VALUE_BASE_VERTEX:
1696 return nir_intrinsic_load_base_vertex;
1697 case SYSTEM_VALUE_INVOCATION_ID:
1698 return nir_intrinsic_load_invocation_id;
1699 case SYSTEM_VALUE_FRONT_FACE:
1700 return nir_intrinsic_load_front_face;
1701 case SYSTEM_VALUE_SAMPLE_ID:
1702 return nir_intrinsic_load_sample_id;
1703 case SYSTEM_VALUE_SAMPLE_POS:
1704 return nir_intrinsic_load_sample_pos;
1705 case SYSTEM_VALUE_SAMPLE_MASK_IN:
1706 return nir_intrinsic_load_sample_mask_in;
1707 case SYSTEM_VALUE_LOCAL_INVOCATION_ID:
1708 return nir_intrinsic_load_local_invocation_id;
1709 case SYSTEM_VALUE_WORK_GROUP_ID:
1710 return nir_intrinsic_load_work_group_id;
1711 case SYSTEM_VALUE_NUM_WORK_GROUPS:
1712 return nir_intrinsic_load_num_work_groups;
1713 case SYSTEM_VALUE_PRIMITIVE_ID:
1714 return nir_intrinsic_load_primitive_id;
1715 case SYSTEM_VALUE_TESS_COORD:
1716 return nir_intrinsic_load_tess_coord;
1717 case SYSTEM_VALUE_TESS_LEVEL_OUTER:
1718 return nir_intrinsic_load_tess_level_outer;
1719 case SYSTEM_VALUE_TESS_LEVEL_INNER:
1720 return nir_intrinsic_load_tess_level_inner;
1721 case SYSTEM_VALUE_VERTICES_IN:
1722 return nir_intrinsic_load_patch_vertices_in;
1723 case SYSTEM_VALUE_HELPER_INVOCATION:
1724 return nir_intrinsic_load_helper_invocation;
1725 default:
1726 unreachable("system value does not directly correspond to intrinsic");
1727 }
1728 }
1729
1730 gl_system_value
1731 nir_system_value_from_intrinsic(nir_intrinsic_op intrin)
1732 {
1733 switch (intrin) {
1734 case nir_intrinsic_load_vertex_id:
1735 return SYSTEM_VALUE_VERTEX_ID;
1736 case nir_intrinsic_load_instance_id:
1737 return SYSTEM_VALUE_INSTANCE_ID;
1738 case nir_intrinsic_load_draw_id:
1739 return SYSTEM_VALUE_DRAW_ID;
1740 case nir_intrinsic_load_base_instance:
1741 return SYSTEM_VALUE_BASE_INSTANCE;
1742 case nir_intrinsic_load_vertex_id_zero_base:
1743 return SYSTEM_VALUE_VERTEX_ID_ZERO_BASE;
1744 case nir_intrinsic_load_base_vertex:
1745 return SYSTEM_VALUE_BASE_VERTEX;
1746 case nir_intrinsic_load_invocation_id:
1747 return SYSTEM_VALUE_INVOCATION_ID;
1748 case nir_intrinsic_load_front_face:
1749 return SYSTEM_VALUE_FRONT_FACE;
1750 case nir_intrinsic_load_sample_id:
1751 return SYSTEM_VALUE_SAMPLE_ID;
1752 case nir_intrinsic_load_sample_pos:
1753 return SYSTEM_VALUE_SAMPLE_POS;
1754 case nir_intrinsic_load_sample_mask_in:
1755 return SYSTEM_VALUE_SAMPLE_MASK_IN;
1756 case nir_intrinsic_load_local_invocation_id:
1757 return SYSTEM_VALUE_LOCAL_INVOCATION_ID;
1758 case nir_intrinsic_load_num_work_groups:
1759 return SYSTEM_VALUE_NUM_WORK_GROUPS;
1760 case nir_intrinsic_load_work_group_id:
1761 return SYSTEM_VALUE_WORK_GROUP_ID;
1762 case nir_intrinsic_load_primitive_id:
1763 return SYSTEM_VALUE_PRIMITIVE_ID;
1764 case nir_intrinsic_load_tess_coord:
1765 return SYSTEM_VALUE_TESS_COORD;
1766 case nir_intrinsic_load_tess_level_outer:
1767 return SYSTEM_VALUE_TESS_LEVEL_OUTER;
1768 case nir_intrinsic_load_tess_level_inner:
1769 return SYSTEM_VALUE_TESS_LEVEL_INNER;
1770 case nir_intrinsic_load_patch_vertices_in:
1771 return SYSTEM_VALUE_VERTICES_IN;
1772 case nir_intrinsic_load_helper_invocation:
1773 return SYSTEM_VALUE_HELPER_INVOCATION;
1774 default:
1775 unreachable("intrinsic doesn't produce a system value");
1776 }
1777 }