Merge remote-tracking branch '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->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 {
474 nir_load_const_instr *instr = ralloc(shader, nir_load_const_instr);
475 instr_init(&instr->instr, nir_instr_type_load_const);
476
477 nir_ssa_def_init(&instr->instr, &instr->def, num_components, 32, NULL);
478
479 return instr;
480 }
481
482 nir_intrinsic_instr *
483 nir_intrinsic_instr_create(nir_shader *shader, nir_intrinsic_op op)
484 {
485 unsigned num_srcs = nir_intrinsic_infos[op].num_srcs;
486 nir_intrinsic_instr *instr =
487 ralloc_size(shader,
488 sizeof(nir_intrinsic_instr) + num_srcs * sizeof(nir_src));
489
490 instr_init(&instr->instr, nir_instr_type_intrinsic);
491 instr->intrinsic = op;
492
493 if (nir_intrinsic_infos[op].has_dest)
494 dest_init(&instr->dest);
495
496 for (unsigned i = 0; i < num_srcs; i++)
497 src_init(&instr->src[i]);
498
499 return instr;
500 }
501
502 nir_call_instr *
503 nir_call_instr_create(nir_shader *shader, nir_function *callee)
504 {
505 nir_call_instr *instr = ralloc(shader, nir_call_instr);
506 instr_init(&instr->instr, nir_instr_type_call);
507
508 instr->callee = callee;
509 instr->num_params = callee->num_params;
510 instr->params = ralloc_array(instr, nir_deref_var *, instr->num_params);
511 instr->return_deref = NULL;
512
513 return instr;
514 }
515
516 nir_tex_instr *
517 nir_tex_instr_create(nir_shader *shader, unsigned num_srcs)
518 {
519 nir_tex_instr *instr = rzalloc(shader, nir_tex_instr);
520 instr_init(&instr->instr, nir_instr_type_tex);
521
522 dest_init(&instr->dest);
523
524 instr->num_srcs = num_srcs;
525 instr->src = ralloc_array(instr, nir_tex_src, num_srcs);
526 for (unsigned i = 0; i < num_srcs; i++)
527 src_init(&instr->src[i].src);
528
529 instr->texture_index = 0;
530 instr->texture_array_size = 0;
531 instr->texture = NULL;
532 instr->sampler_index = 0;
533 instr->sampler = NULL;
534
535 return instr;
536 }
537
538 nir_phi_instr *
539 nir_phi_instr_create(nir_shader *shader)
540 {
541 nir_phi_instr *instr = ralloc(shader, nir_phi_instr);
542 instr_init(&instr->instr, nir_instr_type_phi);
543
544 dest_init(&instr->dest);
545 exec_list_make_empty(&instr->srcs);
546 return instr;
547 }
548
549 nir_parallel_copy_instr *
550 nir_parallel_copy_instr_create(nir_shader *shader)
551 {
552 nir_parallel_copy_instr *instr = ralloc(shader, nir_parallel_copy_instr);
553 instr_init(&instr->instr, nir_instr_type_parallel_copy);
554
555 exec_list_make_empty(&instr->entries);
556
557 return instr;
558 }
559
560 nir_ssa_undef_instr *
561 nir_ssa_undef_instr_create(nir_shader *shader, unsigned num_components)
562 {
563 nir_ssa_undef_instr *instr = ralloc(shader, nir_ssa_undef_instr);
564 instr_init(&instr->instr, nir_instr_type_ssa_undef);
565
566 nir_ssa_def_init(&instr->instr, &instr->def, num_components, 32, NULL);
567
568 return instr;
569 }
570
571 nir_deref_var *
572 nir_deref_var_create(void *mem_ctx, nir_variable *var)
573 {
574 nir_deref_var *deref = ralloc(mem_ctx, nir_deref_var);
575 deref->deref.deref_type = nir_deref_type_var;
576 deref->deref.child = NULL;
577 deref->deref.type = var->type;
578 deref->var = var;
579 return deref;
580 }
581
582 nir_deref_array *
583 nir_deref_array_create(void *mem_ctx)
584 {
585 nir_deref_array *deref = ralloc(mem_ctx, nir_deref_array);
586 deref->deref.deref_type = nir_deref_type_array;
587 deref->deref.child = NULL;
588 deref->deref_array_type = nir_deref_array_type_direct;
589 src_init(&deref->indirect);
590 deref->base_offset = 0;
591 return deref;
592 }
593
594 nir_deref_struct *
595 nir_deref_struct_create(void *mem_ctx, unsigned field_index)
596 {
597 nir_deref_struct *deref = ralloc(mem_ctx, nir_deref_struct);
598 deref->deref.deref_type = nir_deref_type_struct;
599 deref->deref.child = NULL;
600 deref->index = field_index;
601 return deref;
602 }
603
604 static nir_deref_var *
605 copy_deref_var(void *mem_ctx, nir_deref_var *deref)
606 {
607 nir_deref_var *ret = nir_deref_var_create(mem_ctx, deref->var);
608 ret->deref.type = deref->deref.type;
609 if (deref->deref.child)
610 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
611 return ret;
612 }
613
614 static nir_deref_array *
615 copy_deref_array(void *mem_ctx, nir_deref_array *deref)
616 {
617 nir_deref_array *ret = nir_deref_array_create(mem_ctx);
618 ret->base_offset = deref->base_offset;
619 ret->deref_array_type = deref->deref_array_type;
620 if (deref->deref_array_type == nir_deref_array_type_indirect) {
621 nir_src_copy(&ret->indirect, &deref->indirect, mem_ctx);
622 }
623 ret->deref.type = deref->deref.type;
624 if (deref->deref.child)
625 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
626 return ret;
627 }
628
629 static nir_deref_struct *
630 copy_deref_struct(void *mem_ctx, nir_deref_struct *deref)
631 {
632 nir_deref_struct *ret = nir_deref_struct_create(mem_ctx, deref->index);
633 ret->deref.type = deref->deref.type;
634 if (deref->deref.child)
635 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
636 return ret;
637 }
638
639 nir_deref *
640 nir_copy_deref(void *mem_ctx, nir_deref *deref)
641 {
642 switch (deref->deref_type) {
643 case nir_deref_type_var:
644 return &copy_deref_var(mem_ctx, nir_deref_as_var(deref))->deref;
645 case nir_deref_type_array:
646 return &copy_deref_array(mem_ctx, nir_deref_as_array(deref))->deref;
647 case nir_deref_type_struct:
648 return &copy_deref_struct(mem_ctx, nir_deref_as_struct(deref))->deref;
649 default:
650 unreachable("Invalid dereference type");
651 }
652
653 return NULL;
654 }
655
656 /* Returns a load_const instruction that represents the constant
657 * initializer for the given deref chain. The caller is responsible for
658 * ensuring that there actually is a constant initializer.
659 */
660 nir_load_const_instr *
661 nir_deref_get_const_initializer_load(nir_shader *shader, nir_deref_var *deref)
662 {
663 nir_constant *constant = deref->var->constant_initializer;
664 assert(constant);
665
666 const nir_deref *tail = &deref->deref;
667 unsigned matrix_offset = 0;
668 while (tail->child) {
669 switch (tail->child->deref_type) {
670 case nir_deref_type_array: {
671 nir_deref_array *arr = nir_deref_as_array(tail->child);
672 assert(arr->deref_array_type == nir_deref_array_type_direct);
673 if (glsl_type_is_matrix(tail->type)) {
674 assert(arr->deref.child == NULL);
675 matrix_offset = arr->base_offset;
676 } else {
677 constant = constant->elements[arr->base_offset];
678 }
679 break;
680 }
681
682 case nir_deref_type_struct: {
683 constant = constant->elements[nir_deref_as_struct(tail->child)->index];
684 break;
685 }
686
687 default:
688 unreachable("Invalid deref child type");
689 }
690
691 tail = tail->child;
692 }
693
694 nir_load_const_instr *load =
695 nir_load_const_instr_create(shader, glsl_get_vector_elements(tail->type));
696
697 matrix_offset *= load->def.num_components;
698 for (unsigned i = 0; i < load->def.num_components; i++) {
699 switch (glsl_get_base_type(tail->type)) {
700 case GLSL_TYPE_FLOAT:
701 case GLSL_TYPE_INT:
702 case GLSL_TYPE_UINT:
703 load->value.u32[i] = constant->value.u[matrix_offset + i];
704 break;
705 case GLSL_TYPE_BOOL:
706 load->value.u32[i] = constant->value.b[matrix_offset + i] ?
707 NIR_TRUE : NIR_FALSE;
708 break;
709 default:
710 unreachable("Invalid immediate type");
711 }
712 }
713
714 return load;
715 }
716
717 nir_function_impl *
718 nir_cf_node_get_function(nir_cf_node *node)
719 {
720 while (node->type != nir_cf_node_function) {
721 node = node->parent;
722 }
723
724 return nir_cf_node_as_function(node);
725 }
726
727 /* Reduces a cursor by trying to convert everything to after and trying to
728 * go up to block granularity when possible.
729 */
730 static nir_cursor
731 reduce_cursor(nir_cursor cursor)
732 {
733 switch (cursor.option) {
734 case nir_cursor_before_block:
735 assert(nir_cf_node_prev(&cursor.block->cf_node) == NULL ||
736 nir_cf_node_prev(&cursor.block->cf_node)->type != nir_cf_node_block);
737 if (exec_list_is_empty(&cursor.block->instr_list)) {
738 /* Empty block. After is as good as before. */
739 cursor.option = nir_cursor_after_block;
740 }
741 return cursor;
742
743 case nir_cursor_after_block:
744 return cursor;
745
746 case nir_cursor_before_instr: {
747 nir_instr *prev_instr = nir_instr_prev(cursor.instr);
748 if (prev_instr) {
749 /* Before this instruction is after the previous */
750 cursor.instr = prev_instr;
751 cursor.option = nir_cursor_after_instr;
752 } else {
753 /* No previous instruction. Switch to before block */
754 cursor.block = cursor.instr->block;
755 cursor.option = nir_cursor_before_block;
756 }
757 return reduce_cursor(cursor);
758 }
759
760 case nir_cursor_after_instr:
761 if (nir_instr_next(cursor.instr) == NULL) {
762 /* This is the last instruction, switch to after block */
763 cursor.option = nir_cursor_after_block;
764 cursor.block = cursor.instr->block;
765 }
766 return cursor;
767
768 default:
769 unreachable("Inavlid cursor option");
770 }
771 }
772
773 bool
774 nir_cursors_equal(nir_cursor a, nir_cursor b)
775 {
776 /* Reduced cursors should be unique */
777 a = reduce_cursor(a);
778 b = reduce_cursor(b);
779
780 return a.block == b.block && a.option == b.option;
781 }
782
783 static bool
784 add_use_cb(nir_src *src, void *state)
785 {
786 nir_instr *instr = state;
787
788 src->parent_instr = instr;
789 list_addtail(&src->use_link,
790 src->is_ssa ? &src->ssa->uses : &src->reg.reg->uses);
791
792 return true;
793 }
794
795 static bool
796 add_ssa_def_cb(nir_ssa_def *def, void *state)
797 {
798 nir_instr *instr = state;
799
800 if (instr->block && def->index == UINT_MAX) {
801 nir_function_impl *impl =
802 nir_cf_node_get_function(&instr->block->cf_node);
803
804 def->index = impl->ssa_alloc++;
805 }
806
807 return true;
808 }
809
810 static bool
811 add_reg_def_cb(nir_dest *dest, void *state)
812 {
813 nir_instr *instr = state;
814
815 if (!dest->is_ssa) {
816 dest->reg.parent_instr = instr;
817 list_addtail(&dest->reg.def_link, &dest->reg.reg->defs);
818 }
819
820 return true;
821 }
822
823 static void
824 add_defs_uses(nir_instr *instr)
825 {
826 nir_foreach_src(instr, add_use_cb, instr);
827 nir_foreach_dest(instr, add_reg_def_cb, instr);
828 nir_foreach_ssa_def(instr, add_ssa_def_cb, instr);
829 }
830
831 void
832 nir_instr_insert(nir_cursor cursor, nir_instr *instr)
833 {
834 switch (cursor.option) {
835 case nir_cursor_before_block:
836 /* Only allow inserting jumps into empty blocks. */
837 if (instr->type == nir_instr_type_jump)
838 assert(exec_list_is_empty(&cursor.block->instr_list));
839
840 instr->block = cursor.block;
841 add_defs_uses(instr);
842 exec_list_push_head(&cursor.block->instr_list, &instr->node);
843 break;
844 case nir_cursor_after_block: {
845 /* Inserting instructions after a jump is illegal. */
846 nir_instr *last = nir_block_last_instr(cursor.block);
847 assert(last == NULL || last->type != nir_instr_type_jump);
848 (void) last;
849
850 instr->block = cursor.block;
851 add_defs_uses(instr);
852 exec_list_push_tail(&cursor.block->instr_list, &instr->node);
853 break;
854 }
855 case nir_cursor_before_instr:
856 assert(instr->type != nir_instr_type_jump);
857 instr->block = cursor.instr->block;
858 add_defs_uses(instr);
859 exec_node_insert_node_before(&cursor.instr->node, &instr->node);
860 break;
861 case nir_cursor_after_instr:
862 /* Inserting instructions after a jump is illegal. */
863 assert(cursor.instr->type != nir_instr_type_jump);
864
865 /* Only allow inserting jumps at the end of the block. */
866 if (instr->type == nir_instr_type_jump)
867 assert(cursor.instr == nir_block_last_instr(cursor.instr->block));
868
869 instr->block = cursor.instr->block;
870 add_defs_uses(instr);
871 exec_node_insert_after(&cursor.instr->node, &instr->node);
872 break;
873 }
874
875 if (instr->type == nir_instr_type_jump)
876 nir_handle_add_jump(instr->block);
877 }
878
879 static bool
880 src_is_valid(const nir_src *src)
881 {
882 return src->is_ssa ? (src->ssa != NULL) : (src->reg.reg != NULL);
883 }
884
885 static bool
886 remove_use_cb(nir_src *src, void *state)
887 {
888 if (src_is_valid(src))
889 list_del(&src->use_link);
890
891 return true;
892 }
893
894 static bool
895 remove_def_cb(nir_dest *dest, void *state)
896 {
897 if (!dest->is_ssa)
898 list_del(&dest->reg.def_link);
899
900 return true;
901 }
902
903 static void
904 remove_defs_uses(nir_instr *instr)
905 {
906 nir_foreach_dest(instr, remove_def_cb, instr);
907 nir_foreach_src(instr, remove_use_cb, instr);
908 }
909
910 void nir_instr_remove(nir_instr *instr)
911 {
912 remove_defs_uses(instr);
913 exec_node_remove(&instr->node);
914
915 if (instr->type == nir_instr_type_jump) {
916 nir_jump_instr *jump_instr = nir_instr_as_jump(instr);
917 nir_handle_remove_jump(instr->block, jump_instr->type);
918 }
919 }
920
921 /*@}*/
922
923 void
924 nir_index_local_regs(nir_function_impl *impl)
925 {
926 unsigned index = 0;
927 foreach_list_typed(nir_register, reg, node, &impl->registers) {
928 reg->index = index++;
929 }
930 impl->reg_alloc = index;
931 }
932
933 void
934 nir_index_global_regs(nir_shader *shader)
935 {
936 unsigned index = 0;
937 foreach_list_typed(nir_register, reg, node, &shader->registers) {
938 reg->index = index++;
939 }
940 shader->reg_alloc = index;
941 }
942
943 static bool
944 visit_alu_dest(nir_alu_instr *instr, nir_foreach_dest_cb cb, void *state)
945 {
946 return cb(&instr->dest.dest, state);
947 }
948
949 static bool
950 visit_intrinsic_dest(nir_intrinsic_instr *instr, nir_foreach_dest_cb cb,
951 void *state)
952 {
953 if (nir_intrinsic_infos[instr->intrinsic].has_dest)
954 return cb(&instr->dest, state);
955
956 return true;
957 }
958
959 static bool
960 visit_texture_dest(nir_tex_instr *instr, nir_foreach_dest_cb cb,
961 void *state)
962 {
963 return cb(&instr->dest, state);
964 }
965
966 static bool
967 visit_phi_dest(nir_phi_instr *instr, nir_foreach_dest_cb cb, void *state)
968 {
969 return cb(&instr->dest, state);
970 }
971
972 static bool
973 visit_parallel_copy_dest(nir_parallel_copy_instr *instr,
974 nir_foreach_dest_cb cb, void *state)
975 {
976 nir_foreach_parallel_copy_entry(instr, entry) {
977 if (!cb(&entry->dest, state))
978 return false;
979 }
980
981 return true;
982 }
983
984 bool
985 nir_foreach_dest(nir_instr *instr, nir_foreach_dest_cb cb, void *state)
986 {
987 switch (instr->type) {
988 case nir_instr_type_alu:
989 return visit_alu_dest(nir_instr_as_alu(instr), cb, state);
990 case nir_instr_type_intrinsic:
991 return visit_intrinsic_dest(nir_instr_as_intrinsic(instr), cb, state);
992 case nir_instr_type_tex:
993 return visit_texture_dest(nir_instr_as_tex(instr), cb, state);
994 case nir_instr_type_phi:
995 return visit_phi_dest(nir_instr_as_phi(instr), cb, state);
996 case nir_instr_type_parallel_copy:
997 return visit_parallel_copy_dest(nir_instr_as_parallel_copy(instr),
998 cb, state);
999
1000 case nir_instr_type_load_const:
1001 case nir_instr_type_ssa_undef:
1002 case nir_instr_type_call:
1003 case nir_instr_type_jump:
1004 break;
1005
1006 default:
1007 unreachable("Invalid instruction type");
1008 break;
1009 }
1010
1011 return true;
1012 }
1013
1014 struct foreach_ssa_def_state {
1015 nir_foreach_ssa_def_cb cb;
1016 void *client_state;
1017 };
1018
1019 static inline bool
1020 nir_ssa_def_visitor(nir_dest *dest, void *void_state)
1021 {
1022 struct foreach_ssa_def_state *state = void_state;
1023
1024 if (dest->is_ssa)
1025 return state->cb(&dest->ssa, state->client_state);
1026 else
1027 return true;
1028 }
1029
1030 bool
1031 nir_foreach_ssa_def(nir_instr *instr, nir_foreach_ssa_def_cb cb, void *state)
1032 {
1033 switch (instr->type) {
1034 case nir_instr_type_alu:
1035 case nir_instr_type_tex:
1036 case nir_instr_type_intrinsic:
1037 case nir_instr_type_phi:
1038 case nir_instr_type_parallel_copy: {
1039 struct foreach_ssa_def_state foreach_state = {cb, state};
1040 return nir_foreach_dest(instr, nir_ssa_def_visitor, &foreach_state);
1041 }
1042
1043 case nir_instr_type_load_const:
1044 return cb(&nir_instr_as_load_const(instr)->def, state);
1045 case nir_instr_type_ssa_undef:
1046 return cb(&nir_instr_as_ssa_undef(instr)->def, state);
1047 case nir_instr_type_call:
1048 case nir_instr_type_jump:
1049 return true;
1050 default:
1051 unreachable("Invalid instruction type");
1052 }
1053 }
1054
1055 static bool
1056 visit_src(nir_src *src, nir_foreach_src_cb cb, void *state)
1057 {
1058 if (!cb(src, state))
1059 return false;
1060 if (!src->is_ssa && src->reg.indirect)
1061 return cb(src->reg.indirect, state);
1062 return true;
1063 }
1064
1065 static bool
1066 visit_deref_array_src(nir_deref_array *deref, nir_foreach_src_cb cb,
1067 void *state)
1068 {
1069 if (deref->deref_array_type == nir_deref_array_type_indirect)
1070 return visit_src(&deref->indirect, cb, state);
1071 return true;
1072 }
1073
1074 static bool
1075 visit_deref_src(nir_deref_var *deref, nir_foreach_src_cb cb, void *state)
1076 {
1077 nir_deref *cur = &deref->deref;
1078 while (cur != NULL) {
1079 if (cur->deref_type == nir_deref_type_array) {
1080 if (!visit_deref_array_src(nir_deref_as_array(cur), cb, state))
1081 return false;
1082 }
1083
1084 cur = cur->child;
1085 }
1086
1087 return true;
1088 }
1089
1090 static bool
1091 visit_alu_src(nir_alu_instr *instr, nir_foreach_src_cb cb, void *state)
1092 {
1093 for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++)
1094 if (!visit_src(&instr->src[i].src, cb, state))
1095 return false;
1096
1097 return true;
1098 }
1099
1100 static bool
1101 visit_tex_src(nir_tex_instr *instr, nir_foreach_src_cb cb, void *state)
1102 {
1103 for (unsigned i = 0; i < instr->num_srcs; i++) {
1104 if (!visit_src(&instr->src[i].src, cb, state))
1105 return false;
1106 }
1107
1108 if (instr->texture != NULL) {
1109 if (!visit_deref_src(instr->texture, cb, state))
1110 return false;
1111 }
1112
1113 if (instr->sampler != NULL) {
1114 if (!visit_deref_src(instr->sampler, cb, state))
1115 return false;
1116 }
1117
1118 return true;
1119 }
1120
1121 static bool
1122 visit_intrinsic_src(nir_intrinsic_instr *instr, nir_foreach_src_cb cb,
1123 void *state)
1124 {
1125 unsigned num_srcs = nir_intrinsic_infos[instr->intrinsic].num_srcs;
1126 for (unsigned i = 0; i < num_srcs; i++) {
1127 if (!visit_src(&instr->src[i], cb, state))
1128 return false;
1129 }
1130
1131 unsigned num_vars =
1132 nir_intrinsic_infos[instr->intrinsic].num_variables;
1133 for (unsigned i = 0; i < num_vars; i++) {
1134 if (!visit_deref_src(instr->variables[i], cb, state))
1135 return false;
1136 }
1137
1138 return true;
1139 }
1140
1141 static bool
1142 visit_call_src(nir_call_instr *instr, nir_foreach_src_cb cb, void *state)
1143 {
1144 return true;
1145 }
1146
1147 static bool
1148 visit_load_const_src(nir_load_const_instr *instr, nir_foreach_src_cb cb,
1149 void *state)
1150 {
1151 return true;
1152 }
1153
1154 static bool
1155 visit_phi_src(nir_phi_instr *instr, nir_foreach_src_cb cb, void *state)
1156 {
1157 nir_foreach_phi_src(instr, src) {
1158 if (!visit_src(&src->src, cb, state))
1159 return false;
1160 }
1161
1162 return true;
1163 }
1164
1165 static bool
1166 visit_parallel_copy_src(nir_parallel_copy_instr *instr,
1167 nir_foreach_src_cb cb, void *state)
1168 {
1169 nir_foreach_parallel_copy_entry(instr, entry) {
1170 if (!visit_src(&entry->src, cb, state))
1171 return false;
1172 }
1173
1174 return true;
1175 }
1176
1177 typedef struct {
1178 void *state;
1179 nir_foreach_src_cb cb;
1180 } visit_dest_indirect_state;
1181
1182 static bool
1183 visit_dest_indirect(nir_dest *dest, void *_state)
1184 {
1185 visit_dest_indirect_state *state = (visit_dest_indirect_state *) _state;
1186
1187 if (!dest->is_ssa && dest->reg.indirect)
1188 return state->cb(dest->reg.indirect, state->state);
1189
1190 return true;
1191 }
1192
1193 bool
1194 nir_foreach_src(nir_instr *instr, nir_foreach_src_cb cb, void *state)
1195 {
1196 switch (instr->type) {
1197 case nir_instr_type_alu:
1198 if (!visit_alu_src(nir_instr_as_alu(instr), cb, state))
1199 return false;
1200 break;
1201 case nir_instr_type_intrinsic:
1202 if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr), cb, state))
1203 return false;
1204 break;
1205 case nir_instr_type_tex:
1206 if (!visit_tex_src(nir_instr_as_tex(instr), cb, state))
1207 return false;
1208 break;
1209 case nir_instr_type_call:
1210 if (!visit_call_src(nir_instr_as_call(instr), cb, state))
1211 return false;
1212 break;
1213 case nir_instr_type_load_const:
1214 if (!visit_load_const_src(nir_instr_as_load_const(instr), cb, state))
1215 return false;
1216 break;
1217 case nir_instr_type_phi:
1218 if (!visit_phi_src(nir_instr_as_phi(instr), cb, state))
1219 return false;
1220 break;
1221 case nir_instr_type_parallel_copy:
1222 if (!visit_parallel_copy_src(nir_instr_as_parallel_copy(instr),
1223 cb, state))
1224 return false;
1225 break;
1226 case nir_instr_type_jump:
1227 case nir_instr_type_ssa_undef:
1228 return true;
1229
1230 default:
1231 unreachable("Invalid instruction type");
1232 break;
1233 }
1234
1235 visit_dest_indirect_state dest_state;
1236 dest_state.state = state;
1237 dest_state.cb = cb;
1238 return nir_foreach_dest(instr, visit_dest_indirect, &dest_state);
1239 }
1240
1241 nir_const_value *
1242 nir_src_as_const_value(nir_src src)
1243 {
1244 if (!src.is_ssa)
1245 return NULL;
1246
1247 if (src.ssa->parent_instr->type != nir_instr_type_load_const)
1248 return NULL;
1249
1250 nir_load_const_instr *load = nir_instr_as_load_const(src.ssa->parent_instr);
1251
1252 return &load->value;
1253 }
1254
1255 /**
1256 * Returns true if the source is known to be dynamically uniform. Otherwise it
1257 * returns false which means it may or may not be dynamically uniform but it
1258 * can't be determined.
1259 */
1260 bool
1261 nir_src_is_dynamically_uniform(nir_src src)
1262 {
1263 if (!src.is_ssa)
1264 return false;
1265
1266 /* Constants are trivially dynamically uniform */
1267 if (src.ssa->parent_instr->type == nir_instr_type_load_const)
1268 return true;
1269
1270 /* As are uniform variables */
1271 if (src.ssa->parent_instr->type == nir_instr_type_intrinsic) {
1272 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(src.ssa->parent_instr);
1273
1274 if (intr->intrinsic == nir_intrinsic_load_uniform)
1275 return true;
1276 }
1277
1278 /* XXX: this could have many more tests, such as when a sampler function is
1279 * called with dynamically uniform arguments.
1280 */
1281 return false;
1282 }
1283
1284 static void
1285 src_remove_all_uses(nir_src *src)
1286 {
1287 for (; src; src = src->is_ssa ? NULL : src->reg.indirect) {
1288 if (!src_is_valid(src))
1289 continue;
1290
1291 list_del(&src->use_link);
1292 }
1293 }
1294
1295 static void
1296 src_add_all_uses(nir_src *src, nir_instr *parent_instr, nir_if *parent_if)
1297 {
1298 for (; src; src = src->is_ssa ? NULL : src->reg.indirect) {
1299 if (!src_is_valid(src))
1300 continue;
1301
1302 if (parent_instr) {
1303 src->parent_instr = parent_instr;
1304 if (src->is_ssa)
1305 list_addtail(&src->use_link, &src->ssa->uses);
1306 else
1307 list_addtail(&src->use_link, &src->reg.reg->uses);
1308 } else {
1309 assert(parent_if);
1310 src->parent_if = parent_if;
1311 if (src->is_ssa)
1312 list_addtail(&src->use_link, &src->ssa->if_uses);
1313 else
1314 list_addtail(&src->use_link, &src->reg.reg->if_uses);
1315 }
1316 }
1317 }
1318
1319 void
1320 nir_instr_rewrite_src(nir_instr *instr, nir_src *src, nir_src new_src)
1321 {
1322 assert(!src_is_valid(src) || src->parent_instr == instr);
1323
1324 src_remove_all_uses(src);
1325 *src = new_src;
1326 src_add_all_uses(src, instr, NULL);
1327 }
1328
1329 void
1330 nir_instr_move_src(nir_instr *dest_instr, nir_src *dest, nir_src *src)
1331 {
1332 assert(!src_is_valid(dest) || dest->parent_instr == dest_instr);
1333
1334 src_remove_all_uses(dest);
1335 src_remove_all_uses(src);
1336 *dest = *src;
1337 *src = NIR_SRC_INIT;
1338 src_add_all_uses(dest, dest_instr, NULL);
1339 }
1340
1341 void
1342 nir_if_rewrite_condition(nir_if *if_stmt, nir_src new_src)
1343 {
1344 nir_src *src = &if_stmt->condition;
1345 assert(!src_is_valid(src) || src->parent_if == if_stmt);
1346
1347 src_remove_all_uses(src);
1348 *src = new_src;
1349 src_add_all_uses(src, NULL, if_stmt);
1350 }
1351
1352 void
1353 nir_instr_rewrite_dest(nir_instr *instr, nir_dest *dest, nir_dest new_dest)
1354 {
1355 if (dest->is_ssa) {
1356 /* We can only overwrite an SSA destination if it has no uses. */
1357 assert(list_empty(&dest->ssa.uses) && list_empty(&dest->ssa.if_uses));
1358 } else {
1359 list_del(&dest->reg.def_link);
1360 if (dest->reg.indirect)
1361 src_remove_all_uses(dest->reg.indirect);
1362 }
1363
1364 /* We can't re-write with an SSA def */
1365 assert(!new_dest.is_ssa);
1366
1367 nir_dest_copy(dest, &new_dest, instr);
1368
1369 dest->reg.parent_instr = instr;
1370 list_addtail(&dest->reg.def_link, &new_dest.reg.reg->defs);
1371
1372 if (dest->reg.indirect)
1373 src_add_all_uses(dest->reg.indirect, instr, NULL);
1374 }
1375
1376 /* note: does *not* take ownership of 'name' */
1377 void
1378 nir_ssa_def_init(nir_instr *instr, nir_ssa_def *def,
1379 unsigned num_components,
1380 unsigned bit_size, const char *name)
1381 {
1382 def->name = ralloc_strdup(instr, name);
1383 def->parent_instr = instr;
1384 list_inithead(&def->uses);
1385 list_inithead(&def->if_uses);
1386 def->num_components = num_components;
1387 def->bit_size = bit_size;
1388
1389 if (instr->block) {
1390 nir_function_impl *impl =
1391 nir_cf_node_get_function(&instr->block->cf_node);
1392
1393 def->index = impl->ssa_alloc++;
1394 } else {
1395 def->index = UINT_MAX;
1396 }
1397 }
1398
1399 /* note: does *not* take ownership of 'name' */
1400 void
1401 nir_ssa_dest_init(nir_instr *instr, nir_dest *dest,
1402 unsigned num_components, unsigned bit_size,
1403 const char *name)
1404 {
1405 dest->is_ssa = true;
1406 nir_ssa_def_init(instr, &dest->ssa, num_components, bit_size, name);
1407 }
1408
1409 void
1410 nir_ssa_def_rewrite_uses(nir_ssa_def *def, nir_src new_src)
1411 {
1412 assert(!new_src.is_ssa || def != new_src.ssa);
1413
1414 nir_foreach_use_safe(def, use_src)
1415 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1416
1417 nir_foreach_if_use_safe(def, use_src)
1418 nir_if_rewrite_condition(use_src->parent_if, new_src);
1419 }
1420
1421 static bool
1422 is_instr_between(nir_instr *start, nir_instr *end, nir_instr *between)
1423 {
1424 assert(start->block == end->block);
1425
1426 if (between->block != start->block)
1427 return false;
1428
1429 /* Search backwards looking for "between" */
1430 while (start != end) {
1431 if (between == end)
1432 return true;
1433
1434 end = nir_instr_prev(end);
1435 assert(end);
1436 }
1437
1438 return false;
1439 }
1440
1441 /* Replaces all uses of the given SSA def with the given source but only if
1442 * the use comes after the after_me instruction. This can be useful if you
1443 * are emitting code to fix up the result of some instruction: you can freely
1444 * use the result in that code and then call rewrite_uses_after and pass the
1445 * last fixup instruction as after_me and it will replace all of the uses you
1446 * want without touching the fixup code.
1447 *
1448 * This function assumes that after_me is in the same block as
1449 * def->parent_instr and that after_me comes after def->parent_instr.
1450 */
1451 void
1452 nir_ssa_def_rewrite_uses_after(nir_ssa_def *def, nir_src new_src,
1453 nir_instr *after_me)
1454 {
1455 assert(!new_src.is_ssa || def != new_src.ssa);
1456
1457 nir_foreach_use_safe(def, use_src) {
1458 assert(use_src->parent_instr != def->parent_instr);
1459 /* Since def already dominates all of its uses, the only way a use can
1460 * not be dominated by after_me is if it is between def and after_me in
1461 * the instruction list.
1462 */
1463 if (!is_instr_between(def->parent_instr, after_me, use_src->parent_instr))
1464 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1465 }
1466
1467 nir_foreach_if_use_safe(def, use_src)
1468 nir_if_rewrite_condition(use_src->parent_if, new_src);
1469 }
1470
1471 static bool foreach_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1472 bool reverse, void *state);
1473
1474 static inline bool
1475 foreach_if(nir_if *if_stmt, nir_foreach_block_cb cb, bool reverse, void *state)
1476 {
1477 if (reverse) {
1478 foreach_list_typed_reverse_safe(nir_cf_node, node, node,
1479 &if_stmt->else_list) {
1480 if (!foreach_cf_node(node, cb, reverse, state))
1481 return false;
1482 }
1483
1484 foreach_list_typed_reverse_safe(nir_cf_node, node, node,
1485 &if_stmt->then_list) {
1486 if (!foreach_cf_node(node, cb, reverse, state))
1487 return false;
1488 }
1489 } else {
1490 foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt->then_list) {
1491 if (!foreach_cf_node(node, cb, reverse, state))
1492 return false;
1493 }
1494
1495 foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt->else_list) {
1496 if (!foreach_cf_node(node, cb, reverse, state))
1497 return false;
1498 }
1499 }
1500
1501 return true;
1502 }
1503
1504 static inline bool
1505 foreach_loop(nir_loop *loop, nir_foreach_block_cb cb, bool reverse, void *state)
1506 {
1507 if (reverse) {
1508 foreach_list_typed_reverse_safe(nir_cf_node, node, node, &loop->body) {
1509 if (!foreach_cf_node(node, cb, reverse, state))
1510 return false;
1511 }
1512 } else {
1513 foreach_list_typed_safe(nir_cf_node, node, node, &loop->body) {
1514 if (!foreach_cf_node(node, cb, reverse, state))
1515 return false;
1516 }
1517 }
1518
1519 return true;
1520 }
1521
1522 static bool
1523 foreach_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1524 bool reverse, void *state)
1525 {
1526 switch (node->type) {
1527 case nir_cf_node_block:
1528 return cb(nir_cf_node_as_block(node), state);
1529 case nir_cf_node_if:
1530 return foreach_if(nir_cf_node_as_if(node), cb, reverse, state);
1531 case nir_cf_node_loop:
1532 return foreach_loop(nir_cf_node_as_loop(node), cb, reverse, state);
1533 break;
1534
1535 default:
1536 unreachable("Invalid CFG node type");
1537 break;
1538 }
1539
1540 return false;
1541 }
1542
1543 bool
1544 nir_foreach_block_in_cf_node(nir_cf_node *node, nir_foreach_block_cb cb,
1545 void *state)
1546 {
1547 return foreach_cf_node(node, cb, false, state);
1548 }
1549
1550 bool
1551 nir_foreach_block(nir_function_impl *impl, nir_foreach_block_cb cb, void *state)
1552 {
1553 foreach_list_typed_safe(nir_cf_node, node, node, &impl->body) {
1554 if (!foreach_cf_node(node, cb, false, state))
1555 return false;
1556 }
1557
1558 return cb(impl->end_block, state);
1559 }
1560
1561 bool
1562 nir_foreach_block_reverse(nir_function_impl *impl, nir_foreach_block_cb cb,
1563 void *state)
1564 {
1565 if (!cb(impl->end_block, state))
1566 return false;
1567
1568 foreach_list_typed_reverse_safe(nir_cf_node, node, node, &impl->body) {
1569 if (!foreach_cf_node(node, cb, true, state))
1570 return false;
1571 }
1572
1573 return true;
1574 }
1575
1576 nir_if *
1577 nir_block_get_following_if(nir_block *block)
1578 {
1579 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1580 return NULL;
1581
1582 if (nir_cf_node_is_last(&block->cf_node))
1583 return NULL;
1584
1585 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1586
1587 if (next_node->type != nir_cf_node_if)
1588 return NULL;
1589
1590 return nir_cf_node_as_if(next_node);
1591 }
1592
1593 nir_loop *
1594 nir_block_get_following_loop(nir_block *block)
1595 {
1596 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1597 return NULL;
1598
1599 if (nir_cf_node_is_last(&block->cf_node))
1600 return NULL;
1601
1602 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1603
1604 if (next_node->type != nir_cf_node_loop)
1605 return NULL;
1606
1607 return nir_cf_node_as_loop(next_node);
1608 }
1609 static bool
1610 index_block(nir_block *block, void *state)
1611 {
1612 unsigned *index = state;
1613 block->index = (*index)++;
1614 return true;
1615 }
1616
1617 void
1618 nir_index_blocks(nir_function_impl *impl)
1619 {
1620 unsigned index = 0;
1621
1622 if (impl->valid_metadata & nir_metadata_block_index)
1623 return;
1624
1625 nir_foreach_block(impl, index_block, &index);
1626
1627 impl->num_blocks = index;
1628 }
1629
1630 static bool
1631 index_ssa_def_cb(nir_ssa_def *def, void *state)
1632 {
1633 unsigned *index = (unsigned *) state;
1634 def->index = (*index)++;
1635
1636 return true;
1637 }
1638
1639 static bool
1640 index_ssa_block(nir_block *block, void *state)
1641 {
1642 nir_foreach_instr(block, instr)
1643 nir_foreach_ssa_def(instr, index_ssa_def_cb, state);
1644
1645 return true;
1646 }
1647
1648 /**
1649 * The indices are applied top-to-bottom which has the very nice property
1650 * that, if A dominates B, then A->index <= B->index.
1651 */
1652 void
1653 nir_index_ssa_defs(nir_function_impl *impl)
1654 {
1655 unsigned index = 0;
1656 nir_foreach_block(impl, index_ssa_block, &index);
1657 impl->ssa_alloc = index;
1658 }
1659
1660 static bool
1661 index_instrs_block(nir_block *block, void *state)
1662 {
1663 unsigned *index = state;
1664 nir_foreach_instr(block, instr)
1665 instr->index = (*index)++;
1666
1667 return true;
1668 }
1669
1670 /**
1671 * The indices are applied top-to-bottom which has the very nice property
1672 * that, if A dominates B, then A->index <= B->index.
1673 */
1674 unsigned
1675 nir_index_instrs(nir_function_impl *impl)
1676 {
1677 unsigned index = 0;
1678 nir_foreach_block(impl, index_instrs_block, &index);
1679 return index;
1680 }
1681
1682 nir_intrinsic_op
1683 nir_intrinsic_from_system_value(gl_system_value val)
1684 {
1685 switch (val) {
1686 case SYSTEM_VALUE_VERTEX_ID:
1687 return nir_intrinsic_load_vertex_id;
1688 case SYSTEM_VALUE_INSTANCE_ID:
1689 return nir_intrinsic_load_instance_id;
1690 case SYSTEM_VALUE_DRAW_ID:
1691 return nir_intrinsic_load_draw_id;
1692 case SYSTEM_VALUE_BASE_INSTANCE:
1693 return nir_intrinsic_load_base_instance;
1694 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
1695 return nir_intrinsic_load_vertex_id_zero_base;
1696 case SYSTEM_VALUE_BASE_VERTEX:
1697 return nir_intrinsic_load_base_vertex;
1698 case SYSTEM_VALUE_INVOCATION_ID:
1699 return nir_intrinsic_load_invocation_id;
1700 case SYSTEM_VALUE_FRONT_FACE:
1701 return nir_intrinsic_load_front_face;
1702 case SYSTEM_VALUE_SAMPLE_ID:
1703 return nir_intrinsic_load_sample_id;
1704 case SYSTEM_VALUE_SAMPLE_POS:
1705 return nir_intrinsic_load_sample_pos;
1706 case SYSTEM_VALUE_SAMPLE_MASK_IN:
1707 return nir_intrinsic_load_sample_mask_in;
1708 case SYSTEM_VALUE_LOCAL_INVOCATION_ID:
1709 return nir_intrinsic_load_local_invocation_id;
1710 case SYSTEM_VALUE_WORK_GROUP_ID:
1711 return nir_intrinsic_load_work_group_id;
1712 case SYSTEM_VALUE_NUM_WORK_GROUPS:
1713 return nir_intrinsic_load_num_work_groups;
1714 case SYSTEM_VALUE_PRIMITIVE_ID:
1715 return nir_intrinsic_load_primitive_id;
1716 case SYSTEM_VALUE_TESS_COORD:
1717 return nir_intrinsic_load_tess_coord;
1718 case SYSTEM_VALUE_TESS_LEVEL_OUTER:
1719 return nir_intrinsic_load_tess_level_outer;
1720 case SYSTEM_VALUE_TESS_LEVEL_INNER:
1721 return nir_intrinsic_load_tess_level_inner;
1722 case SYSTEM_VALUE_VERTICES_IN:
1723 return nir_intrinsic_load_patch_vertices_in;
1724 case SYSTEM_VALUE_HELPER_INVOCATION:
1725 return nir_intrinsic_load_helper_invocation;
1726 default:
1727 unreachable("system value does not directly correspond to intrinsic");
1728 }
1729 }
1730
1731 gl_system_value
1732 nir_system_value_from_intrinsic(nir_intrinsic_op intrin)
1733 {
1734 switch (intrin) {
1735 case nir_intrinsic_load_vertex_id:
1736 return SYSTEM_VALUE_VERTEX_ID;
1737 case nir_intrinsic_load_instance_id:
1738 return SYSTEM_VALUE_INSTANCE_ID;
1739 case nir_intrinsic_load_draw_id:
1740 return SYSTEM_VALUE_DRAW_ID;
1741 case nir_intrinsic_load_base_instance:
1742 return SYSTEM_VALUE_BASE_INSTANCE;
1743 case nir_intrinsic_load_vertex_id_zero_base:
1744 return SYSTEM_VALUE_VERTEX_ID_ZERO_BASE;
1745 case nir_intrinsic_load_base_vertex:
1746 return SYSTEM_VALUE_BASE_VERTEX;
1747 case nir_intrinsic_load_invocation_id:
1748 return SYSTEM_VALUE_INVOCATION_ID;
1749 case nir_intrinsic_load_front_face:
1750 return SYSTEM_VALUE_FRONT_FACE;
1751 case nir_intrinsic_load_sample_id:
1752 return SYSTEM_VALUE_SAMPLE_ID;
1753 case nir_intrinsic_load_sample_pos:
1754 return SYSTEM_VALUE_SAMPLE_POS;
1755 case nir_intrinsic_load_sample_mask_in:
1756 return SYSTEM_VALUE_SAMPLE_MASK_IN;
1757 case nir_intrinsic_load_local_invocation_id:
1758 return SYSTEM_VALUE_LOCAL_INVOCATION_ID;
1759 case nir_intrinsic_load_num_work_groups:
1760 return SYSTEM_VALUE_NUM_WORK_GROUPS;
1761 case nir_intrinsic_load_work_group_id:
1762 return SYSTEM_VALUE_WORK_GROUP_ID;
1763 case nir_intrinsic_load_primitive_id:
1764 return SYSTEM_VALUE_PRIMITIVE_ID;
1765 case nir_intrinsic_load_tess_coord:
1766 return SYSTEM_VALUE_TESS_COORD;
1767 case nir_intrinsic_load_tess_level_outer:
1768 return SYSTEM_VALUE_TESS_LEVEL_OUTER;
1769 case nir_intrinsic_load_tess_level_inner:
1770 return SYSTEM_VALUE_TESS_LEVEL_INNER;
1771 case nir_intrinsic_load_patch_vertices_in:
1772 return SYSTEM_VALUE_VERTICES_IN;
1773 case nir_intrinsic_load_helper_invocation:
1774 return SYSTEM_VALUE_HELPER_INVOCATION;
1775 default:
1776 unreachable("intrinsic doesn't produce a system value");
1777 }
1778 }