nir: glsl_get_bit_size() should take glsl_type
[mesa.git] / src / compiler / nir / nir.c
1 /*
2 * Copyright © 2014 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Connor Abbott (cwabbott0@gmail.com)
25 *
26 */
27
28 #include "nir.h"
29 #include "nir_control_flow_private.h"
30 #include <assert.h>
31
32 nir_shader *
33 nir_shader_create(void *mem_ctx,
34 gl_shader_stage stage,
35 const nir_shader_compiler_options *options)
36 {
37 nir_shader *shader = ralloc(mem_ctx, nir_shader);
38
39 exec_list_make_empty(&shader->uniforms);
40 exec_list_make_empty(&shader->inputs);
41 exec_list_make_empty(&shader->outputs);
42 exec_list_make_empty(&shader->shared);
43
44 shader->options = options;
45 memset(&shader->info, 0, sizeof(shader->info));
46
47 exec_list_make_empty(&shader->functions);
48 exec_list_make_empty(&shader->registers);
49 exec_list_make_empty(&shader->globals);
50 exec_list_make_empty(&shader->system_values);
51 shader->reg_alloc = 0;
52
53 shader->num_inputs = 0;
54 shader->num_outputs = 0;
55 shader->num_uniforms = 0;
56 shader->num_shared = 0;
57
58 shader->stage = stage;
59
60 return shader;
61 }
62
63 static nir_register *
64 reg_create(void *mem_ctx, struct exec_list *list)
65 {
66 nir_register *reg = ralloc(mem_ctx, nir_register);
67
68 list_inithead(&reg->uses);
69 list_inithead(&reg->defs);
70 list_inithead(&reg->if_uses);
71
72 reg->num_components = 0;
73 reg->bit_size = 32;
74 reg->num_array_elems = 0;
75 reg->is_packed = false;
76 reg->name = NULL;
77
78 exec_list_push_tail(list, &reg->node);
79
80 return reg;
81 }
82
83 nir_register *
84 nir_global_reg_create(nir_shader *shader)
85 {
86 nir_register *reg = reg_create(shader, &shader->registers);
87 reg->index = shader->reg_alloc++;
88 reg->is_global = true;
89
90 return reg;
91 }
92
93 nir_register *
94 nir_local_reg_create(nir_function_impl *impl)
95 {
96 nir_register *reg = reg_create(ralloc_parent(impl), &impl->registers);
97 reg->index = impl->reg_alloc++;
98 reg->is_global = false;
99
100 return reg;
101 }
102
103 void
104 nir_reg_remove(nir_register *reg)
105 {
106 exec_node_remove(&reg->node);
107 }
108
109 void
110 nir_shader_add_variable(nir_shader *shader, nir_variable *var)
111 {
112 switch (var->data.mode) {
113 case nir_var_all:
114 assert(!"invalid mode");
115 break;
116
117 case nir_var_local:
118 assert(!"nir_shader_add_variable cannot be used for local variables");
119 break;
120
121 case nir_var_param:
122 assert(!"nir_shader_add_variable cannot be used for function parameters");
123 break;
124
125 case nir_var_global:
126 exec_list_push_tail(&shader->globals, &var->node);
127 break;
128
129 case nir_var_shader_in:
130 exec_list_push_tail(&shader->inputs, &var->node);
131 break;
132
133 case nir_var_shader_out:
134 exec_list_push_tail(&shader->outputs, &var->node);
135 break;
136
137 case nir_var_uniform:
138 case nir_var_shader_storage:
139 exec_list_push_tail(&shader->uniforms, &var->node);
140 break;
141
142 case nir_var_shared:
143 assert(shader->stage == MESA_SHADER_COMPUTE);
144 exec_list_push_tail(&shader->shared, &var->node);
145 break;
146
147 case nir_var_system_value:
148 exec_list_push_tail(&shader->system_values, &var->node);
149 break;
150 }
151 }
152
153 nir_variable *
154 nir_variable_create(nir_shader *shader, nir_variable_mode mode,
155 const struct glsl_type *type, const char *name)
156 {
157 nir_variable *var = rzalloc(shader, nir_variable);
158 var->name = ralloc_strdup(var, name);
159 var->type = type;
160 var->data.mode = mode;
161
162 if ((mode == nir_var_shader_in && shader->stage != MESA_SHADER_VERTEX) ||
163 (mode == nir_var_shader_out && shader->stage != MESA_SHADER_FRAGMENT))
164 var->data.interpolation = INTERP_QUALIFIER_SMOOTH;
165
166 if (mode == nir_var_shader_in || mode == nir_var_uniform)
167 var->data.read_only = true;
168
169 nir_shader_add_variable(shader, var);
170
171 return var;
172 }
173
174 nir_variable *
175 nir_local_variable_create(nir_function_impl *impl,
176 const struct glsl_type *type, const char *name)
177 {
178 nir_variable *var = rzalloc(impl->function->shader, nir_variable);
179 var->name = ralloc_strdup(var, name);
180 var->type = type;
181 var->data.mode = nir_var_local;
182
183 nir_function_impl_add_variable(impl, var);
184
185 return var;
186 }
187
188 nir_function *
189 nir_function_create(nir_shader *shader, const char *name)
190 {
191 nir_function *func = ralloc(shader, nir_function);
192
193 exec_list_push_tail(&shader->functions, &func->node);
194
195 func->name = ralloc_strdup(func, name);
196 func->shader = shader;
197 func->num_params = 0;
198 func->params = NULL;
199 func->return_type = glsl_void_type();
200 func->impl = NULL;
201
202 return func;
203 }
204
205 void nir_src_copy(nir_src *dest, const nir_src *src, void *mem_ctx)
206 {
207 dest->is_ssa = src->is_ssa;
208 if (src->is_ssa) {
209 dest->ssa = src->ssa;
210 } else {
211 dest->reg.base_offset = src->reg.base_offset;
212 dest->reg.reg = src->reg.reg;
213 if (src->reg.indirect) {
214 dest->reg.indirect = ralloc(mem_ctx, nir_src);
215 nir_src_copy(dest->reg.indirect, src->reg.indirect, mem_ctx);
216 } else {
217 dest->reg.indirect = NULL;
218 }
219 }
220 }
221
222 void nir_dest_copy(nir_dest *dest, const nir_dest *src, nir_instr *instr)
223 {
224 /* Copying an SSA definition makes no sense whatsoever. */
225 assert(!src->is_ssa);
226
227 dest->is_ssa = false;
228
229 dest->reg.base_offset = src->reg.base_offset;
230 dest->reg.reg = src->reg.reg;
231 if (src->reg.indirect) {
232 dest->reg.indirect = ralloc(instr, nir_src);
233 nir_src_copy(dest->reg.indirect, src->reg.indirect, instr);
234 } else {
235 dest->reg.indirect = NULL;
236 }
237 }
238
239 void
240 nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src,
241 nir_alu_instr *instr)
242 {
243 nir_src_copy(&dest->src, &src->src, &instr->instr);
244 dest->abs = src->abs;
245 dest->negate = src->negate;
246 for (unsigned i = 0; i < 4; i++)
247 dest->swizzle[i] = src->swizzle[i];
248 }
249
250 void
251 nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src,
252 nir_alu_instr *instr)
253 {
254 nir_dest_copy(&dest->dest, &src->dest, &instr->instr);
255 dest->write_mask = src->write_mask;
256 dest->saturate = src->saturate;
257 }
258
259
260 static void
261 cf_init(nir_cf_node *node, nir_cf_node_type type)
262 {
263 exec_node_init(&node->node);
264 node->parent = NULL;
265 node->type = type;
266 }
267
268 nir_function_impl *
269 nir_function_impl_create_bare(nir_shader *shader)
270 {
271 nir_function_impl *impl = ralloc(shader, nir_function_impl);
272
273 impl->function = NULL;
274
275 cf_init(&impl->cf_node, nir_cf_node_function);
276
277 exec_list_make_empty(&impl->body);
278 exec_list_make_empty(&impl->registers);
279 exec_list_make_empty(&impl->locals);
280 impl->num_params = 0;
281 impl->params = NULL;
282 impl->return_var = NULL;
283 impl->reg_alloc = 0;
284 impl->ssa_alloc = 0;
285 impl->valid_metadata = nir_metadata_none;
286
287 /* create start & end blocks */
288 nir_block *start_block = nir_block_create(shader);
289 nir_block *end_block = nir_block_create(shader);
290 start_block->cf_node.parent = &impl->cf_node;
291 end_block->cf_node.parent = &impl->cf_node;
292 impl->end_block = end_block;
293
294 exec_list_push_tail(&impl->body, &start_block->cf_node.node);
295
296 start_block->successors[0] = end_block;
297 _mesa_set_add(end_block->predecessors, start_block);
298 return impl;
299 }
300
301 nir_function_impl *
302 nir_function_impl_create(nir_function *function)
303 {
304 assert(function->impl == NULL);
305
306 nir_function_impl *impl = nir_function_impl_create_bare(function->shader);
307
308 function->impl = impl;
309 impl->function = function;
310
311 impl->num_params = function->num_params;
312 impl->params = ralloc_array(function->shader,
313 nir_variable *, impl->num_params);
314
315 for (unsigned i = 0; i < impl->num_params; i++) {
316 impl->params[i] = rzalloc(function->shader, nir_variable);
317 impl->params[i]->type = function->params[i].type;
318 impl->params[i]->data.mode = nir_var_param;
319 impl->params[i]->data.location = i;
320 }
321
322 if (!glsl_type_is_void(function->return_type)) {
323 impl->return_var = rzalloc(function->shader, nir_variable);
324 impl->return_var->type = function->return_type;
325 impl->return_var->data.mode = nir_var_param;
326 impl->return_var->data.location = -1;
327 } else {
328 impl->return_var = NULL;
329 }
330
331 return impl;
332 }
333
334 nir_block *
335 nir_block_create(nir_shader *shader)
336 {
337 nir_block *block = ralloc(shader, nir_block);
338
339 cf_init(&block->cf_node, nir_cf_node_block);
340
341 block->successors[0] = block->successors[1] = NULL;
342 block->predecessors = _mesa_set_create(block, _mesa_hash_pointer,
343 _mesa_key_pointer_equal);
344 block->imm_dom = NULL;
345 /* XXX maybe it would be worth it to defer allocation? This
346 * way it doesn't get allocated for shader ref's that never run
347 * nir_calc_dominance? For example, state-tracker creates an
348 * initial IR, clones that, runs appropriate lowering pass, passes
349 * to driver which does common lowering/opt, and then stores ref
350 * which is later used to do state specific lowering and futher
351 * opt. Do any of the references not need dominance metadata?
352 */
353 block->dom_frontier = _mesa_set_create(block, _mesa_hash_pointer,
354 _mesa_key_pointer_equal);
355
356 exec_list_make_empty(&block->instr_list);
357
358 return block;
359 }
360
361 static inline void
362 src_init(nir_src *src)
363 {
364 src->is_ssa = false;
365 src->reg.reg = NULL;
366 src->reg.indirect = NULL;
367 src->reg.base_offset = 0;
368 }
369
370 nir_if *
371 nir_if_create(nir_shader *shader)
372 {
373 nir_if *if_stmt = ralloc(shader, nir_if);
374
375 cf_init(&if_stmt->cf_node, nir_cf_node_if);
376 src_init(&if_stmt->condition);
377
378 nir_block *then = nir_block_create(shader);
379 exec_list_make_empty(&if_stmt->then_list);
380 exec_list_push_tail(&if_stmt->then_list, &then->cf_node.node);
381 then->cf_node.parent = &if_stmt->cf_node;
382
383 nir_block *else_stmt = nir_block_create(shader);
384 exec_list_make_empty(&if_stmt->else_list);
385 exec_list_push_tail(&if_stmt->else_list, &else_stmt->cf_node.node);
386 else_stmt->cf_node.parent = &if_stmt->cf_node;
387
388 return if_stmt;
389 }
390
391 nir_loop *
392 nir_loop_create(nir_shader *shader)
393 {
394 nir_loop *loop = ralloc(shader, nir_loop);
395
396 cf_init(&loop->cf_node, nir_cf_node_loop);
397
398 nir_block *body = nir_block_create(shader);
399 exec_list_make_empty(&loop->body);
400 exec_list_push_tail(&loop->body, &body->cf_node.node);
401 body->cf_node.parent = &loop->cf_node;
402
403 body->successors[0] = body;
404 _mesa_set_add(body->predecessors, body);
405
406 return loop;
407 }
408
409 static void
410 instr_init(nir_instr *instr, nir_instr_type type)
411 {
412 instr->type = type;
413 instr->block = NULL;
414 exec_node_init(&instr->node);
415 }
416
417 static void
418 dest_init(nir_dest *dest)
419 {
420 dest->is_ssa = false;
421 dest->reg.reg = NULL;
422 dest->reg.indirect = NULL;
423 dest->reg.base_offset = 0;
424 }
425
426 static void
427 alu_dest_init(nir_alu_dest *dest)
428 {
429 dest_init(&dest->dest);
430 dest->saturate = false;
431 dest->write_mask = 0xf;
432 }
433
434 static void
435 alu_src_init(nir_alu_src *src)
436 {
437 src_init(&src->src);
438 src->abs = src->negate = false;
439 src->swizzle[0] = 0;
440 src->swizzle[1] = 1;
441 src->swizzle[2] = 2;
442 src->swizzle[3] = 3;
443 }
444
445 nir_alu_instr *
446 nir_alu_instr_create(nir_shader *shader, nir_op op)
447 {
448 unsigned num_srcs = nir_op_infos[op].num_inputs;
449 nir_alu_instr *instr =
450 ralloc_size(shader,
451 sizeof(nir_alu_instr) + num_srcs * sizeof(nir_alu_src));
452
453 instr_init(&instr->instr, nir_instr_type_alu);
454 instr->op = op;
455 alu_dest_init(&instr->dest);
456 for (unsigned i = 0; i < num_srcs; i++)
457 alu_src_init(&instr->src[i]);
458
459 return instr;
460 }
461
462 nir_jump_instr *
463 nir_jump_instr_create(nir_shader *shader, nir_jump_type type)
464 {
465 nir_jump_instr *instr = ralloc(shader, nir_jump_instr);
466 instr_init(&instr->instr, nir_instr_type_jump);
467 instr->type = type;
468 return instr;
469 }
470
471 nir_load_const_instr *
472 nir_load_const_instr_create(nir_shader *shader, unsigned num_components,
473 unsigned bit_size)
474 {
475 nir_load_const_instr *instr = ralloc(shader, nir_load_const_instr);
476 instr_init(&instr->instr, nir_instr_type_load_const);
477
478 nir_ssa_def_init(&instr->instr, &instr->def, num_components, bit_size, NULL);
479
480 return instr;
481 }
482
483 nir_intrinsic_instr *
484 nir_intrinsic_instr_create(nir_shader *shader, nir_intrinsic_op op)
485 {
486 unsigned num_srcs = nir_intrinsic_infos[op].num_srcs;
487 nir_intrinsic_instr *instr =
488 ralloc_size(shader,
489 sizeof(nir_intrinsic_instr) + num_srcs * sizeof(nir_src));
490
491 instr_init(&instr->instr, nir_instr_type_intrinsic);
492 instr->intrinsic = op;
493
494 if (nir_intrinsic_infos[op].has_dest)
495 dest_init(&instr->dest);
496
497 for (unsigned i = 0; i < num_srcs; i++)
498 src_init(&instr->src[i]);
499
500 return instr;
501 }
502
503 nir_call_instr *
504 nir_call_instr_create(nir_shader *shader, nir_function *callee)
505 {
506 nir_call_instr *instr = ralloc(shader, nir_call_instr);
507 instr_init(&instr->instr, nir_instr_type_call);
508
509 instr->callee = callee;
510 instr->num_params = callee->num_params;
511 instr->params = ralloc_array(instr, nir_deref_var *, instr->num_params);
512 instr->return_deref = NULL;
513
514 return instr;
515 }
516
517 nir_tex_instr *
518 nir_tex_instr_create(nir_shader *shader, unsigned num_srcs)
519 {
520 nir_tex_instr *instr = rzalloc(shader, nir_tex_instr);
521 instr_init(&instr->instr, nir_instr_type_tex);
522
523 dest_init(&instr->dest);
524
525 instr->num_srcs = num_srcs;
526 instr->src = ralloc_array(instr, nir_tex_src, num_srcs);
527 for (unsigned i = 0; i < num_srcs; i++)
528 src_init(&instr->src[i].src);
529
530 instr->texture_index = 0;
531 instr->texture_array_size = 0;
532 instr->texture = NULL;
533 instr->sampler_index = 0;
534 instr->sampler = NULL;
535
536 return instr;
537 }
538
539 nir_phi_instr *
540 nir_phi_instr_create(nir_shader *shader)
541 {
542 nir_phi_instr *instr = ralloc(shader, nir_phi_instr);
543 instr_init(&instr->instr, nir_instr_type_phi);
544
545 dest_init(&instr->dest);
546 exec_list_make_empty(&instr->srcs);
547 return instr;
548 }
549
550 nir_parallel_copy_instr *
551 nir_parallel_copy_instr_create(nir_shader *shader)
552 {
553 nir_parallel_copy_instr *instr = ralloc(shader, nir_parallel_copy_instr);
554 instr_init(&instr->instr, nir_instr_type_parallel_copy);
555
556 exec_list_make_empty(&instr->entries);
557
558 return instr;
559 }
560
561 nir_ssa_undef_instr *
562 nir_ssa_undef_instr_create(nir_shader *shader,
563 unsigned num_components,
564 unsigned bit_size)
565 {
566 nir_ssa_undef_instr *instr = ralloc(shader, nir_ssa_undef_instr);
567 instr_init(&instr->instr, nir_instr_type_ssa_undef);
568
569 nir_ssa_def_init(&instr->instr, &instr->def, num_components, bit_size, NULL);
570
571 return instr;
572 }
573
574 nir_deref_var *
575 nir_deref_var_create(void *mem_ctx, nir_variable *var)
576 {
577 nir_deref_var *deref = ralloc(mem_ctx, nir_deref_var);
578 deref->deref.deref_type = nir_deref_type_var;
579 deref->deref.child = NULL;
580 deref->deref.type = var->type;
581 deref->var = var;
582 return deref;
583 }
584
585 nir_deref_array *
586 nir_deref_array_create(void *mem_ctx)
587 {
588 nir_deref_array *deref = ralloc(mem_ctx, nir_deref_array);
589 deref->deref.deref_type = nir_deref_type_array;
590 deref->deref.child = NULL;
591 deref->deref_array_type = nir_deref_array_type_direct;
592 src_init(&deref->indirect);
593 deref->base_offset = 0;
594 return deref;
595 }
596
597 nir_deref_struct *
598 nir_deref_struct_create(void *mem_ctx, unsigned field_index)
599 {
600 nir_deref_struct *deref = ralloc(mem_ctx, nir_deref_struct);
601 deref->deref.deref_type = nir_deref_type_struct;
602 deref->deref.child = NULL;
603 deref->index = field_index;
604 return deref;
605 }
606
607 static nir_deref_var *
608 copy_deref_var(void *mem_ctx, nir_deref_var *deref)
609 {
610 nir_deref_var *ret = nir_deref_var_create(mem_ctx, deref->var);
611 ret->deref.type = deref->deref.type;
612 if (deref->deref.child)
613 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
614 return ret;
615 }
616
617 static nir_deref_array *
618 copy_deref_array(void *mem_ctx, nir_deref_array *deref)
619 {
620 nir_deref_array *ret = nir_deref_array_create(mem_ctx);
621 ret->base_offset = deref->base_offset;
622 ret->deref_array_type = deref->deref_array_type;
623 if (deref->deref_array_type == nir_deref_array_type_indirect) {
624 nir_src_copy(&ret->indirect, &deref->indirect, mem_ctx);
625 }
626 ret->deref.type = deref->deref.type;
627 if (deref->deref.child)
628 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
629 return ret;
630 }
631
632 static nir_deref_struct *
633 copy_deref_struct(void *mem_ctx, nir_deref_struct *deref)
634 {
635 nir_deref_struct *ret = nir_deref_struct_create(mem_ctx, deref->index);
636 ret->deref.type = deref->deref.type;
637 if (deref->deref.child)
638 ret->deref.child = nir_copy_deref(ret, deref->deref.child);
639 return ret;
640 }
641
642 nir_deref *
643 nir_copy_deref(void *mem_ctx, nir_deref *deref)
644 {
645 switch (deref->deref_type) {
646 case nir_deref_type_var:
647 return &copy_deref_var(mem_ctx, nir_deref_as_var(deref))->deref;
648 case nir_deref_type_array:
649 return &copy_deref_array(mem_ctx, nir_deref_as_array(deref))->deref;
650 case nir_deref_type_struct:
651 return &copy_deref_struct(mem_ctx, nir_deref_as_struct(deref))->deref;
652 default:
653 unreachable("Invalid dereference type");
654 }
655
656 return NULL;
657 }
658
659 /* Returns a load_const instruction that represents the constant
660 * initializer for the given deref chain. The caller is responsible for
661 * ensuring that there actually is a constant initializer.
662 */
663 nir_load_const_instr *
664 nir_deref_get_const_initializer_load(nir_shader *shader, nir_deref_var *deref)
665 {
666 nir_constant *constant = deref->var->constant_initializer;
667 assert(constant);
668
669 const nir_deref *tail = &deref->deref;
670 unsigned matrix_offset = 0;
671 while (tail->child) {
672 switch (tail->child->deref_type) {
673 case nir_deref_type_array: {
674 nir_deref_array *arr = nir_deref_as_array(tail->child);
675 assert(arr->deref_array_type == nir_deref_array_type_direct);
676 if (glsl_type_is_matrix(tail->type)) {
677 assert(arr->deref.child == NULL);
678 matrix_offset = arr->base_offset;
679 } else {
680 constant = constant->elements[arr->base_offset];
681 }
682 break;
683 }
684
685 case nir_deref_type_struct: {
686 constant = constant->elements[nir_deref_as_struct(tail->child)->index];
687 break;
688 }
689
690 default:
691 unreachable("Invalid deref child type");
692 }
693
694 tail = tail->child;
695 }
696
697 unsigned bit_size = glsl_get_bit_size(tail->type);
698 nir_load_const_instr *load =
699 nir_load_const_instr_create(shader, glsl_get_vector_elements(tail->type),
700 bit_size);
701
702 matrix_offset *= load->def.num_components;
703 for (unsigned i = 0; i < load->def.num_components; i++) {
704 switch (glsl_get_base_type(tail->type)) {
705 case GLSL_TYPE_FLOAT:
706 case GLSL_TYPE_INT:
707 case GLSL_TYPE_UINT:
708 load->value.u32[i] = constant->value.u[matrix_offset + i];
709 break;
710 case GLSL_TYPE_DOUBLE:
711 load->value.f64[i] = constant->value.d[matrix_offset + i];
712 break;
713 case GLSL_TYPE_BOOL:
714 load->value.u32[i] = constant->value.b[matrix_offset + i] ?
715 NIR_TRUE : NIR_FALSE;
716 break;
717 default:
718 unreachable("Invalid immediate type");
719 }
720 }
721
722 return load;
723 }
724
725 nir_function_impl *
726 nir_cf_node_get_function(nir_cf_node *node)
727 {
728 while (node->type != nir_cf_node_function) {
729 node = node->parent;
730 }
731
732 return nir_cf_node_as_function(node);
733 }
734
735 /* Reduces a cursor by trying to convert everything to after and trying to
736 * go up to block granularity when possible.
737 */
738 static nir_cursor
739 reduce_cursor(nir_cursor cursor)
740 {
741 switch (cursor.option) {
742 case nir_cursor_before_block:
743 assert(nir_cf_node_prev(&cursor.block->cf_node) == NULL ||
744 nir_cf_node_prev(&cursor.block->cf_node)->type != nir_cf_node_block);
745 if (exec_list_is_empty(&cursor.block->instr_list)) {
746 /* Empty block. After is as good as before. */
747 cursor.option = nir_cursor_after_block;
748 }
749 return cursor;
750
751 case nir_cursor_after_block:
752 return cursor;
753
754 case nir_cursor_before_instr: {
755 nir_instr *prev_instr = nir_instr_prev(cursor.instr);
756 if (prev_instr) {
757 /* Before this instruction is after the previous */
758 cursor.instr = prev_instr;
759 cursor.option = nir_cursor_after_instr;
760 } else {
761 /* No previous instruction. Switch to before block */
762 cursor.block = cursor.instr->block;
763 cursor.option = nir_cursor_before_block;
764 }
765 return reduce_cursor(cursor);
766 }
767
768 case nir_cursor_after_instr:
769 if (nir_instr_next(cursor.instr) == NULL) {
770 /* This is the last instruction, switch to after block */
771 cursor.option = nir_cursor_after_block;
772 cursor.block = cursor.instr->block;
773 }
774 return cursor;
775
776 default:
777 unreachable("Inavlid cursor option");
778 }
779 }
780
781 bool
782 nir_cursors_equal(nir_cursor a, nir_cursor b)
783 {
784 /* Reduced cursors should be unique */
785 a = reduce_cursor(a);
786 b = reduce_cursor(b);
787
788 return a.block == b.block && a.option == b.option;
789 }
790
791 static bool
792 add_use_cb(nir_src *src, void *state)
793 {
794 nir_instr *instr = state;
795
796 src->parent_instr = instr;
797 list_addtail(&src->use_link,
798 src->is_ssa ? &src->ssa->uses : &src->reg.reg->uses);
799
800 return true;
801 }
802
803 static bool
804 add_ssa_def_cb(nir_ssa_def *def, void *state)
805 {
806 nir_instr *instr = state;
807
808 if (instr->block && def->index == UINT_MAX) {
809 nir_function_impl *impl =
810 nir_cf_node_get_function(&instr->block->cf_node);
811
812 def->index = impl->ssa_alloc++;
813 }
814
815 return true;
816 }
817
818 static bool
819 add_reg_def_cb(nir_dest *dest, void *state)
820 {
821 nir_instr *instr = state;
822
823 if (!dest->is_ssa) {
824 dest->reg.parent_instr = instr;
825 list_addtail(&dest->reg.def_link, &dest->reg.reg->defs);
826 }
827
828 return true;
829 }
830
831 static void
832 add_defs_uses(nir_instr *instr)
833 {
834 nir_foreach_src(instr, add_use_cb, instr);
835 nir_foreach_dest(instr, add_reg_def_cb, instr);
836 nir_foreach_ssa_def(instr, add_ssa_def_cb, instr);
837 }
838
839 void
840 nir_instr_insert(nir_cursor cursor, nir_instr *instr)
841 {
842 switch (cursor.option) {
843 case nir_cursor_before_block:
844 /* Only allow inserting jumps into empty blocks. */
845 if (instr->type == nir_instr_type_jump)
846 assert(exec_list_is_empty(&cursor.block->instr_list));
847
848 instr->block = cursor.block;
849 add_defs_uses(instr);
850 exec_list_push_head(&cursor.block->instr_list, &instr->node);
851 break;
852 case nir_cursor_after_block: {
853 /* Inserting instructions after a jump is illegal. */
854 nir_instr *last = nir_block_last_instr(cursor.block);
855 assert(last == NULL || last->type != nir_instr_type_jump);
856 (void) last;
857
858 instr->block = cursor.block;
859 add_defs_uses(instr);
860 exec_list_push_tail(&cursor.block->instr_list, &instr->node);
861 break;
862 }
863 case nir_cursor_before_instr:
864 assert(instr->type != nir_instr_type_jump);
865 instr->block = cursor.instr->block;
866 add_defs_uses(instr);
867 exec_node_insert_node_before(&cursor.instr->node, &instr->node);
868 break;
869 case nir_cursor_after_instr:
870 /* Inserting instructions after a jump is illegal. */
871 assert(cursor.instr->type != nir_instr_type_jump);
872
873 /* Only allow inserting jumps at the end of the block. */
874 if (instr->type == nir_instr_type_jump)
875 assert(cursor.instr == nir_block_last_instr(cursor.instr->block));
876
877 instr->block = cursor.instr->block;
878 add_defs_uses(instr);
879 exec_node_insert_after(&cursor.instr->node, &instr->node);
880 break;
881 }
882
883 if (instr->type == nir_instr_type_jump)
884 nir_handle_add_jump(instr->block);
885 }
886
887 static bool
888 src_is_valid(const nir_src *src)
889 {
890 return src->is_ssa ? (src->ssa != NULL) : (src->reg.reg != NULL);
891 }
892
893 static bool
894 remove_use_cb(nir_src *src, void *state)
895 {
896 if (src_is_valid(src))
897 list_del(&src->use_link);
898
899 return true;
900 }
901
902 static bool
903 remove_def_cb(nir_dest *dest, void *state)
904 {
905 if (!dest->is_ssa)
906 list_del(&dest->reg.def_link);
907
908 return true;
909 }
910
911 static void
912 remove_defs_uses(nir_instr *instr)
913 {
914 nir_foreach_dest(instr, remove_def_cb, instr);
915 nir_foreach_src(instr, remove_use_cb, instr);
916 }
917
918 void nir_instr_remove(nir_instr *instr)
919 {
920 remove_defs_uses(instr);
921 exec_node_remove(&instr->node);
922
923 if (instr->type == nir_instr_type_jump) {
924 nir_jump_instr *jump_instr = nir_instr_as_jump(instr);
925 nir_handle_remove_jump(instr->block, jump_instr->type);
926 }
927 }
928
929 /*@}*/
930
931 void
932 nir_index_local_regs(nir_function_impl *impl)
933 {
934 unsigned index = 0;
935 foreach_list_typed(nir_register, reg, node, &impl->registers) {
936 reg->index = index++;
937 }
938 impl->reg_alloc = index;
939 }
940
941 void
942 nir_index_global_regs(nir_shader *shader)
943 {
944 unsigned index = 0;
945 foreach_list_typed(nir_register, reg, node, &shader->registers) {
946 reg->index = index++;
947 }
948 shader->reg_alloc = index;
949 }
950
951 static bool
952 visit_alu_dest(nir_alu_instr *instr, nir_foreach_dest_cb cb, void *state)
953 {
954 return cb(&instr->dest.dest, state);
955 }
956
957 static bool
958 visit_intrinsic_dest(nir_intrinsic_instr *instr, nir_foreach_dest_cb cb,
959 void *state)
960 {
961 if (nir_intrinsic_infos[instr->intrinsic].has_dest)
962 return cb(&instr->dest, state);
963
964 return true;
965 }
966
967 static bool
968 visit_texture_dest(nir_tex_instr *instr, nir_foreach_dest_cb cb,
969 void *state)
970 {
971 return cb(&instr->dest, state);
972 }
973
974 static bool
975 visit_phi_dest(nir_phi_instr *instr, nir_foreach_dest_cb cb, void *state)
976 {
977 return cb(&instr->dest, state);
978 }
979
980 static bool
981 visit_parallel_copy_dest(nir_parallel_copy_instr *instr,
982 nir_foreach_dest_cb cb, void *state)
983 {
984 nir_foreach_parallel_copy_entry(entry, instr) {
985 if (!cb(&entry->dest, state))
986 return false;
987 }
988
989 return true;
990 }
991
992 bool
993 nir_foreach_dest(nir_instr *instr, nir_foreach_dest_cb cb, void *state)
994 {
995 switch (instr->type) {
996 case nir_instr_type_alu:
997 return visit_alu_dest(nir_instr_as_alu(instr), cb, state);
998 case nir_instr_type_intrinsic:
999 return visit_intrinsic_dest(nir_instr_as_intrinsic(instr), cb, state);
1000 case nir_instr_type_tex:
1001 return visit_texture_dest(nir_instr_as_tex(instr), cb, state);
1002 case nir_instr_type_phi:
1003 return visit_phi_dest(nir_instr_as_phi(instr), cb, state);
1004 case nir_instr_type_parallel_copy:
1005 return visit_parallel_copy_dest(nir_instr_as_parallel_copy(instr),
1006 cb, state);
1007
1008 case nir_instr_type_load_const:
1009 case nir_instr_type_ssa_undef:
1010 case nir_instr_type_call:
1011 case nir_instr_type_jump:
1012 break;
1013
1014 default:
1015 unreachable("Invalid instruction type");
1016 break;
1017 }
1018
1019 return true;
1020 }
1021
1022 struct foreach_ssa_def_state {
1023 nir_foreach_ssa_def_cb cb;
1024 void *client_state;
1025 };
1026
1027 static inline bool
1028 nir_ssa_def_visitor(nir_dest *dest, void *void_state)
1029 {
1030 struct foreach_ssa_def_state *state = void_state;
1031
1032 if (dest->is_ssa)
1033 return state->cb(&dest->ssa, state->client_state);
1034 else
1035 return true;
1036 }
1037
1038 bool
1039 nir_foreach_ssa_def(nir_instr *instr, nir_foreach_ssa_def_cb cb, void *state)
1040 {
1041 switch (instr->type) {
1042 case nir_instr_type_alu:
1043 case nir_instr_type_tex:
1044 case nir_instr_type_intrinsic:
1045 case nir_instr_type_phi:
1046 case nir_instr_type_parallel_copy: {
1047 struct foreach_ssa_def_state foreach_state = {cb, state};
1048 return nir_foreach_dest(instr, nir_ssa_def_visitor, &foreach_state);
1049 }
1050
1051 case nir_instr_type_load_const:
1052 return cb(&nir_instr_as_load_const(instr)->def, state);
1053 case nir_instr_type_ssa_undef:
1054 return cb(&nir_instr_as_ssa_undef(instr)->def, state);
1055 case nir_instr_type_call:
1056 case nir_instr_type_jump:
1057 return true;
1058 default:
1059 unreachable("Invalid instruction type");
1060 }
1061 }
1062
1063 static bool
1064 visit_src(nir_src *src, nir_foreach_src_cb cb, void *state)
1065 {
1066 if (!cb(src, state))
1067 return false;
1068 if (!src->is_ssa && src->reg.indirect)
1069 return cb(src->reg.indirect, state);
1070 return true;
1071 }
1072
1073 static bool
1074 visit_deref_array_src(nir_deref_array *deref, nir_foreach_src_cb cb,
1075 void *state)
1076 {
1077 if (deref->deref_array_type == nir_deref_array_type_indirect)
1078 return visit_src(&deref->indirect, cb, state);
1079 return true;
1080 }
1081
1082 static bool
1083 visit_deref_src(nir_deref_var *deref, nir_foreach_src_cb cb, void *state)
1084 {
1085 nir_deref *cur = &deref->deref;
1086 while (cur != NULL) {
1087 if (cur->deref_type == nir_deref_type_array) {
1088 if (!visit_deref_array_src(nir_deref_as_array(cur), cb, state))
1089 return false;
1090 }
1091
1092 cur = cur->child;
1093 }
1094
1095 return true;
1096 }
1097
1098 static bool
1099 visit_alu_src(nir_alu_instr *instr, nir_foreach_src_cb cb, void *state)
1100 {
1101 for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++)
1102 if (!visit_src(&instr->src[i].src, cb, state))
1103 return false;
1104
1105 return true;
1106 }
1107
1108 static bool
1109 visit_tex_src(nir_tex_instr *instr, nir_foreach_src_cb cb, void *state)
1110 {
1111 for (unsigned i = 0; i < instr->num_srcs; i++) {
1112 if (!visit_src(&instr->src[i].src, cb, state))
1113 return false;
1114 }
1115
1116 if (instr->texture != NULL) {
1117 if (!visit_deref_src(instr->texture, cb, state))
1118 return false;
1119 }
1120
1121 if (instr->sampler != NULL) {
1122 if (!visit_deref_src(instr->sampler, cb, state))
1123 return false;
1124 }
1125
1126 return true;
1127 }
1128
1129 static bool
1130 visit_intrinsic_src(nir_intrinsic_instr *instr, nir_foreach_src_cb cb,
1131 void *state)
1132 {
1133 unsigned num_srcs = nir_intrinsic_infos[instr->intrinsic].num_srcs;
1134 for (unsigned i = 0; i < num_srcs; i++) {
1135 if (!visit_src(&instr->src[i], cb, state))
1136 return false;
1137 }
1138
1139 unsigned num_vars =
1140 nir_intrinsic_infos[instr->intrinsic].num_variables;
1141 for (unsigned i = 0; i < num_vars; i++) {
1142 if (!visit_deref_src(instr->variables[i], cb, state))
1143 return false;
1144 }
1145
1146 return true;
1147 }
1148
1149 static bool
1150 visit_call_src(nir_call_instr *instr, nir_foreach_src_cb cb, void *state)
1151 {
1152 return true;
1153 }
1154
1155 static bool
1156 visit_load_const_src(nir_load_const_instr *instr, nir_foreach_src_cb cb,
1157 void *state)
1158 {
1159 return true;
1160 }
1161
1162 static bool
1163 visit_phi_src(nir_phi_instr *instr, nir_foreach_src_cb cb, void *state)
1164 {
1165 nir_foreach_phi_src(src, instr) {
1166 if (!visit_src(&src->src, cb, state))
1167 return false;
1168 }
1169
1170 return true;
1171 }
1172
1173 static bool
1174 visit_parallel_copy_src(nir_parallel_copy_instr *instr,
1175 nir_foreach_src_cb cb, void *state)
1176 {
1177 nir_foreach_parallel_copy_entry(entry, instr) {
1178 if (!visit_src(&entry->src, cb, state))
1179 return false;
1180 }
1181
1182 return true;
1183 }
1184
1185 typedef struct {
1186 void *state;
1187 nir_foreach_src_cb cb;
1188 } visit_dest_indirect_state;
1189
1190 static bool
1191 visit_dest_indirect(nir_dest *dest, void *_state)
1192 {
1193 visit_dest_indirect_state *state = (visit_dest_indirect_state *) _state;
1194
1195 if (!dest->is_ssa && dest->reg.indirect)
1196 return state->cb(dest->reg.indirect, state->state);
1197
1198 return true;
1199 }
1200
1201 bool
1202 nir_foreach_src(nir_instr *instr, nir_foreach_src_cb cb, void *state)
1203 {
1204 switch (instr->type) {
1205 case nir_instr_type_alu:
1206 if (!visit_alu_src(nir_instr_as_alu(instr), cb, state))
1207 return false;
1208 break;
1209 case nir_instr_type_intrinsic:
1210 if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr), cb, state))
1211 return false;
1212 break;
1213 case nir_instr_type_tex:
1214 if (!visit_tex_src(nir_instr_as_tex(instr), cb, state))
1215 return false;
1216 break;
1217 case nir_instr_type_call:
1218 if (!visit_call_src(nir_instr_as_call(instr), cb, state))
1219 return false;
1220 break;
1221 case nir_instr_type_load_const:
1222 if (!visit_load_const_src(nir_instr_as_load_const(instr), cb, state))
1223 return false;
1224 break;
1225 case nir_instr_type_phi:
1226 if (!visit_phi_src(nir_instr_as_phi(instr), cb, state))
1227 return false;
1228 break;
1229 case nir_instr_type_parallel_copy:
1230 if (!visit_parallel_copy_src(nir_instr_as_parallel_copy(instr),
1231 cb, state))
1232 return false;
1233 break;
1234 case nir_instr_type_jump:
1235 case nir_instr_type_ssa_undef:
1236 return true;
1237
1238 default:
1239 unreachable("Invalid instruction type");
1240 break;
1241 }
1242
1243 visit_dest_indirect_state dest_state;
1244 dest_state.state = state;
1245 dest_state.cb = cb;
1246 return nir_foreach_dest(instr, visit_dest_indirect, &dest_state);
1247 }
1248
1249 nir_const_value *
1250 nir_src_as_const_value(nir_src src)
1251 {
1252 if (!src.is_ssa)
1253 return NULL;
1254
1255 if (src.ssa->parent_instr->type != nir_instr_type_load_const)
1256 return NULL;
1257
1258 nir_load_const_instr *load = nir_instr_as_load_const(src.ssa->parent_instr);
1259
1260 return &load->value;
1261 }
1262
1263 /**
1264 * Returns true if the source is known to be dynamically uniform. Otherwise it
1265 * returns false which means it may or may not be dynamically uniform but it
1266 * can't be determined.
1267 */
1268 bool
1269 nir_src_is_dynamically_uniform(nir_src src)
1270 {
1271 if (!src.is_ssa)
1272 return false;
1273
1274 /* Constants are trivially dynamically uniform */
1275 if (src.ssa->parent_instr->type == nir_instr_type_load_const)
1276 return true;
1277
1278 /* As are uniform variables */
1279 if (src.ssa->parent_instr->type == nir_instr_type_intrinsic) {
1280 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(src.ssa->parent_instr);
1281
1282 if (intr->intrinsic == nir_intrinsic_load_uniform)
1283 return true;
1284 }
1285
1286 /* XXX: this could have many more tests, such as when a sampler function is
1287 * called with dynamically uniform arguments.
1288 */
1289 return false;
1290 }
1291
1292 static void
1293 src_remove_all_uses(nir_src *src)
1294 {
1295 for (; src; src = src->is_ssa ? NULL : src->reg.indirect) {
1296 if (!src_is_valid(src))
1297 continue;
1298
1299 list_del(&src->use_link);
1300 }
1301 }
1302
1303 static void
1304 src_add_all_uses(nir_src *src, nir_instr *parent_instr, nir_if *parent_if)
1305 {
1306 for (; src; src = src->is_ssa ? NULL : src->reg.indirect) {
1307 if (!src_is_valid(src))
1308 continue;
1309
1310 if (parent_instr) {
1311 src->parent_instr = parent_instr;
1312 if (src->is_ssa)
1313 list_addtail(&src->use_link, &src->ssa->uses);
1314 else
1315 list_addtail(&src->use_link, &src->reg.reg->uses);
1316 } else {
1317 assert(parent_if);
1318 src->parent_if = parent_if;
1319 if (src->is_ssa)
1320 list_addtail(&src->use_link, &src->ssa->if_uses);
1321 else
1322 list_addtail(&src->use_link, &src->reg.reg->if_uses);
1323 }
1324 }
1325 }
1326
1327 void
1328 nir_instr_rewrite_src(nir_instr *instr, nir_src *src, nir_src new_src)
1329 {
1330 assert(!src_is_valid(src) || src->parent_instr == instr);
1331
1332 src_remove_all_uses(src);
1333 *src = new_src;
1334 src_add_all_uses(src, instr, NULL);
1335 }
1336
1337 void
1338 nir_instr_move_src(nir_instr *dest_instr, nir_src *dest, nir_src *src)
1339 {
1340 assert(!src_is_valid(dest) || dest->parent_instr == dest_instr);
1341
1342 src_remove_all_uses(dest);
1343 src_remove_all_uses(src);
1344 *dest = *src;
1345 *src = NIR_SRC_INIT;
1346 src_add_all_uses(dest, dest_instr, NULL);
1347 }
1348
1349 void
1350 nir_if_rewrite_condition(nir_if *if_stmt, nir_src new_src)
1351 {
1352 nir_src *src = &if_stmt->condition;
1353 assert(!src_is_valid(src) || src->parent_if == if_stmt);
1354
1355 src_remove_all_uses(src);
1356 *src = new_src;
1357 src_add_all_uses(src, NULL, if_stmt);
1358 }
1359
1360 void
1361 nir_instr_rewrite_dest(nir_instr *instr, nir_dest *dest, nir_dest new_dest)
1362 {
1363 if (dest->is_ssa) {
1364 /* We can only overwrite an SSA destination if it has no uses. */
1365 assert(list_empty(&dest->ssa.uses) && list_empty(&dest->ssa.if_uses));
1366 } else {
1367 list_del(&dest->reg.def_link);
1368 if (dest->reg.indirect)
1369 src_remove_all_uses(dest->reg.indirect);
1370 }
1371
1372 /* We can't re-write with an SSA def */
1373 assert(!new_dest.is_ssa);
1374
1375 nir_dest_copy(dest, &new_dest, instr);
1376
1377 dest->reg.parent_instr = instr;
1378 list_addtail(&dest->reg.def_link, &new_dest.reg.reg->defs);
1379
1380 if (dest->reg.indirect)
1381 src_add_all_uses(dest->reg.indirect, instr, NULL);
1382 }
1383
1384 /* note: does *not* take ownership of 'name' */
1385 void
1386 nir_ssa_def_init(nir_instr *instr, nir_ssa_def *def,
1387 unsigned num_components,
1388 unsigned bit_size, const char *name)
1389 {
1390 def->name = ralloc_strdup(instr, name);
1391 def->parent_instr = instr;
1392 list_inithead(&def->uses);
1393 list_inithead(&def->if_uses);
1394 def->num_components = num_components;
1395 def->bit_size = bit_size;
1396
1397 if (instr->block) {
1398 nir_function_impl *impl =
1399 nir_cf_node_get_function(&instr->block->cf_node);
1400
1401 def->index = impl->ssa_alloc++;
1402 } else {
1403 def->index = UINT_MAX;
1404 }
1405 }
1406
1407 /* note: does *not* take ownership of 'name' */
1408 void
1409 nir_ssa_dest_init(nir_instr *instr, nir_dest *dest,
1410 unsigned num_components, unsigned bit_size,
1411 const char *name)
1412 {
1413 dest->is_ssa = true;
1414 nir_ssa_def_init(instr, &dest->ssa, num_components, bit_size, name);
1415 }
1416
1417 void
1418 nir_ssa_def_rewrite_uses(nir_ssa_def *def, nir_src new_src)
1419 {
1420 assert(!new_src.is_ssa || def != new_src.ssa);
1421
1422 nir_foreach_use_safe(use_src, def)
1423 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1424
1425 nir_foreach_if_use_safe(use_src, def)
1426 nir_if_rewrite_condition(use_src->parent_if, new_src);
1427 }
1428
1429 static bool
1430 is_instr_between(nir_instr *start, nir_instr *end, nir_instr *between)
1431 {
1432 assert(start->block == end->block);
1433
1434 if (between->block != start->block)
1435 return false;
1436
1437 /* Search backwards looking for "between" */
1438 while (start != end) {
1439 if (between == end)
1440 return true;
1441
1442 end = nir_instr_prev(end);
1443 assert(end);
1444 }
1445
1446 return false;
1447 }
1448
1449 /* Replaces all uses of the given SSA def with the given source but only if
1450 * the use comes after the after_me instruction. This can be useful if you
1451 * are emitting code to fix up the result of some instruction: you can freely
1452 * use the result in that code and then call rewrite_uses_after and pass the
1453 * last fixup instruction as after_me and it will replace all of the uses you
1454 * want without touching the fixup code.
1455 *
1456 * This function assumes that after_me is in the same block as
1457 * def->parent_instr and that after_me comes after def->parent_instr.
1458 */
1459 void
1460 nir_ssa_def_rewrite_uses_after(nir_ssa_def *def, nir_src new_src,
1461 nir_instr *after_me)
1462 {
1463 assert(!new_src.is_ssa || def != new_src.ssa);
1464
1465 nir_foreach_use_safe(use_src, def) {
1466 assert(use_src->parent_instr != def->parent_instr);
1467 /* Since def already dominates all of its uses, the only way a use can
1468 * not be dominated by after_me is if it is between def and after_me in
1469 * the instruction list.
1470 */
1471 if (!is_instr_between(def->parent_instr, after_me, use_src->parent_instr))
1472 nir_instr_rewrite_src(use_src->parent_instr, use_src, new_src);
1473 }
1474
1475 nir_foreach_if_use_safe(use_src, def)
1476 nir_if_rewrite_condition(use_src->parent_if, new_src);
1477 }
1478
1479 uint8_t
1480 nir_ssa_def_components_read(nir_ssa_def *def)
1481 {
1482 uint8_t read_mask = 0;
1483 nir_foreach_use(use, def) {
1484 if (use->parent_instr->type == nir_instr_type_alu) {
1485 nir_alu_instr *alu = nir_instr_as_alu(use->parent_instr);
1486 nir_alu_src *alu_src = exec_node_data(nir_alu_src, use, src);
1487 int src_idx = alu_src - &alu->src[0];
1488 assert(src_idx >= 0 && src_idx < nir_op_infos[alu->op].num_inputs);
1489
1490 for (unsigned c = 0; c < 4; c++) {
1491 if (!nir_alu_instr_channel_used(alu, src_idx, c))
1492 continue;
1493
1494 read_mask |= (1 << alu_src->swizzle[c]);
1495 }
1496 } else {
1497 return (1 << def->num_components) - 1;
1498 }
1499 }
1500
1501 return read_mask;
1502 }
1503
1504 nir_block *
1505 nir_block_cf_tree_next(nir_block *block)
1506 {
1507 if (block == NULL) {
1508 /* nir_foreach_block_safe() will call this function on a NULL block
1509 * after the last iteration, but it won't use the result so just return
1510 * NULL here.
1511 */
1512 return NULL;
1513 }
1514
1515 nir_cf_node *cf_next = nir_cf_node_next(&block->cf_node);
1516 if (cf_next)
1517 return nir_cf_node_cf_tree_first(cf_next);
1518
1519 nir_cf_node *parent = block->cf_node.parent;
1520
1521 switch (parent->type) {
1522 case nir_cf_node_if: {
1523 /* Are we at the end of the if? Go to the beginning of the else */
1524 nir_if *if_stmt = nir_cf_node_as_if(parent);
1525 if (&block->cf_node == nir_if_last_then_node(if_stmt))
1526 return nir_cf_node_as_block(nir_if_first_else_node(if_stmt));
1527
1528 assert(&block->cf_node == nir_if_last_else_node(if_stmt));
1529 /* fall through */
1530 }
1531
1532 case nir_cf_node_loop:
1533 return nir_cf_node_as_block(nir_cf_node_next(parent));
1534
1535 case nir_cf_node_function:
1536 return NULL;
1537
1538 default:
1539 unreachable("unknown cf node type");
1540 }
1541 }
1542
1543 nir_block *
1544 nir_block_cf_tree_prev(nir_block *block)
1545 {
1546 if (block == NULL) {
1547 /* do this for consistency with nir_block_cf_tree_next() */
1548 return NULL;
1549 }
1550
1551 nir_cf_node *cf_prev = nir_cf_node_prev(&block->cf_node);
1552 if (cf_prev)
1553 return nir_cf_node_cf_tree_last(cf_prev);
1554
1555 nir_cf_node *parent = block->cf_node.parent;
1556
1557 switch (parent->type) {
1558 case nir_cf_node_if: {
1559 /* Are we at the beginning of the else? Go to the end of the if */
1560 nir_if *if_stmt = nir_cf_node_as_if(parent);
1561 if (&block->cf_node == nir_if_first_else_node(if_stmt))
1562 return nir_cf_node_as_block(nir_if_last_then_node(if_stmt));
1563
1564 assert(&block->cf_node == nir_if_first_then_node(if_stmt));
1565 /* fall through */
1566 }
1567
1568 case nir_cf_node_loop:
1569 return nir_cf_node_as_block(nir_cf_node_prev(parent));
1570
1571 case nir_cf_node_function:
1572 return NULL;
1573
1574 default:
1575 unreachable("unknown cf node type");
1576 }
1577 }
1578
1579 nir_block *nir_cf_node_cf_tree_first(nir_cf_node *node)
1580 {
1581 switch (node->type) {
1582 case nir_cf_node_function: {
1583 nir_function_impl *impl = nir_cf_node_as_function(node);
1584 return nir_start_block(impl);
1585 }
1586
1587 case nir_cf_node_if: {
1588 nir_if *if_stmt = nir_cf_node_as_if(node);
1589 return nir_cf_node_as_block(nir_if_first_then_node(if_stmt));
1590 }
1591
1592 case nir_cf_node_loop: {
1593 nir_loop *loop = nir_cf_node_as_loop(node);
1594 return nir_cf_node_as_block(nir_loop_first_cf_node(loop));
1595 }
1596
1597 case nir_cf_node_block: {
1598 return nir_cf_node_as_block(node);
1599 }
1600
1601 default:
1602 unreachable("unknown node type");
1603 }
1604 }
1605
1606 nir_block *nir_cf_node_cf_tree_last(nir_cf_node *node)
1607 {
1608 switch (node->type) {
1609 case nir_cf_node_function: {
1610 nir_function_impl *impl = nir_cf_node_as_function(node);
1611 return nir_impl_last_block(impl);
1612 }
1613
1614 case nir_cf_node_if: {
1615 nir_if *if_stmt = nir_cf_node_as_if(node);
1616 return nir_cf_node_as_block(nir_if_last_else_node(if_stmt));
1617 }
1618
1619 case nir_cf_node_loop: {
1620 nir_loop *loop = nir_cf_node_as_loop(node);
1621 return nir_cf_node_as_block(nir_loop_last_cf_node(loop));
1622 }
1623
1624 case nir_cf_node_block: {
1625 return nir_cf_node_as_block(node);
1626 }
1627
1628 default:
1629 unreachable("unknown node type");
1630 }
1631 }
1632
1633 nir_block *nir_cf_node_cf_tree_next(nir_cf_node *node)
1634 {
1635 if (node->type == nir_cf_node_block)
1636 return nir_cf_node_cf_tree_first(nir_cf_node_next(node));
1637 else if (node->type == nir_cf_node_function)
1638 return NULL;
1639 else
1640 return nir_cf_node_as_block(nir_cf_node_next(node));
1641 }
1642
1643 nir_if *
1644 nir_block_get_following_if(nir_block *block)
1645 {
1646 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1647 return NULL;
1648
1649 if (nir_cf_node_is_last(&block->cf_node))
1650 return NULL;
1651
1652 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1653
1654 if (next_node->type != nir_cf_node_if)
1655 return NULL;
1656
1657 return nir_cf_node_as_if(next_node);
1658 }
1659
1660 nir_loop *
1661 nir_block_get_following_loop(nir_block *block)
1662 {
1663 if (exec_node_is_tail_sentinel(&block->cf_node.node))
1664 return NULL;
1665
1666 if (nir_cf_node_is_last(&block->cf_node))
1667 return NULL;
1668
1669 nir_cf_node *next_node = nir_cf_node_next(&block->cf_node);
1670
1671 if (next_node->type != nir_cf_node_loop)
1672 return NULL;
1673
1674 return nir_cf_node_as_loop(next_node);
1675 }
1676
1677 void
1678 nir_index_blocks(nir_function_impl *impl)
1679 {
1680 unsigned index = 0;
1681
1682 if (impl->valid_metadata & nir_metadata_block_index)
1683 return;
1684
1685 nir_foreach_block(block, impl) {
1686 block->index = index++;
1687 }
1688
1689 impl->num_blocks = index;
1690 }
1691
1692 static bool
1693 index_ssa_def_cb(nir_ssa_def *def, void *state)
1694 {
1695 unsigned *index = (unsigned *) state;
1696 def->index = (*index)++;
1697
1698 return true;
1699 }
1700
1701 /**
1702 * The indices are applied top-to-bottom which has the very nice property
1703 * that, if A dominates B, then A->index <= B->index.
1704 */
1705 void
1706 nir_index_ssa_defs(nir_function_impl *impl)
1707 {
1708 unsigned index = 0;
1709
1710 nir_foreach_block(block, impl) {
1711 nir_foreach_instr(instr, block)
1712 nir_foreach_ssa_def(instr, index_ssa_def_cb, &index);
1713 }
1714
1715 impl->ssa_alloc = index;
1716 }
1717
1718 /**
1719 * The indices are applied top-to-bottom which has the very nice property
1720 * that, if A dominates B, then A->index <= B->index.
1721 */
1722 unsigned
1723 nir_index_instrs(nir_function_impl *impl)
1724 {
1725 unsigned index = 0;
1726
1727 nir_foreach_block(block, impl) {
1728 nir_foreach_instr(instr, block)
1729 instr->index = index++;
1730 }
1731
1732 return index;
1733 }
1734
1735 nir_intrinsic_op
1736 nir_intrinsic_from_system_value(gl_system_value val)
1737 {
1738 switch (val) {
1739 case SYSTEM_VALUE_VERTEX_ID:
1740 return nir_intrinsic_load_vertex_id;
1741 case SYSTEM_VALUE_INSTANCE_ID:
1742 return nir_intrinsic_load_instance_id;
1743 case SYSTEM_VALUE_DRAW_ID:
1744 return nir_intrinsic_load_draw_id;
1745 case SYSTEM_VALUE_BASE_INSTANCE:
1746 return nir_intrinsic_load_base_instance;
1747 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
1748 return nir_intrinsic_load_vertex_id_zero_base;
1749 case SYSTEM_VALUE_BASE_VERTEX:
1750 return nir_intrinsic_load_base_vertex;
1751 case SYSTEM_VALUE_INVOCATION_ID:
1752 return nir_intrinsic_load_invocation_id;
1753 case SYSTEM_VALUE_FRONT_FACE:
1754 return nir_intrinsic_load_front_face;
1755 case SYSTEM_VALUE_SAMPLE_ID:
1756 return nir_intrinsic_load_sample_id;
1757 case SYSTEM_VALUE_SAMPLE_POS:
1758 return nir_intrinsic_load_sample_pos;
1759 case SYSTEM_VALUE_SAMPLE_MASK_IN:
1760 return nir_intrinsic_load_sample_mask_in;
1761 case SYSTEM_VALUE_LOCAL_INVOCATION_ID:
1762 return nir_intrinsic_load_local_invocation_id;
1763 case SYSTEM_VALUE_WORK_GROUP_ID:
1764 return nir_intrinsic_load_work_group_id;
1765 case SYSTEM_VALUE_NUM_WORK_GROUPS:
1766 return nir_intrinsic_load_num_work_groups;
1767 case SYSTEM_VALUE_PRIMITIVE_ID:
1768 return nir_intrinsic_load_primitive_id;
1769 case SYSTEM_VALUE_TESS_COORD:
1770 return nir_intrinsic_load_tess_coord;
1771 case SYSTEM_VALUE_TESS_LEVEL_OUTER:
1772 return nir_intrinsic_load_tess_level_outer;
1773 case SYSTEM_VALUE_TESS_LEVEL_INNER:
1774 return nir_intrinsic_load_tess_level_inner;
1775 case SYSTEM_VALUE_VERTICES_IN:
1776 return nir_intrinsic_load_patch_vertices_in;
1777 case SYSTEM_VALUE_HELPER_INVOCATION:
1778 return nir_intrinsic_load_helper_invocation;
1779 default:
1780 unreachable("system value does not directly correspond to intrinsic");
1781 }
1782 }
1783
1784 gl_system_value
1785 nir_system_value_from_intrinsic(nir_intrinsic_op intrin)
1786 {
1787 switch (intrin) {
1788 case nir_intrinsic_load_vertex_id:
1789 return SYSTEM_VALUE_VERTEX_ID;
1790 case nir_intrinsic_load_instance_id:
1791 return SYSTEM_VALUE_INSTANCE_ID;
1792 case nir_intrinsic_load_draw_id:
1793 return SYSTEM_VALUE_DRAW_ID;
1794 case nir_intrinsic_load_base_instance:
1795 return SYSTEM_VALUE_BASE_INSTANCE;
1796 case nir_intrinsic_load_vertex_id_zero_base:
1797 return SYSTEM_VALUE_VERTEX_ID_ZERO_BASE;
1798 case nir_intrinsic_load_base_vertex:
1799 return SYSTEM_VALUE_BASE_VERTEX;
1800 case nir_intrinsic_load_invocation_id:
1801 return SYSTEM_VALUE_INVOCATION_ID;
1802 case nir_intrinsic_load_front_face:
1803 return SYSTEM_VALUE_FRONT_FACE;
1804 case nir_intrinsic_load_sample_id:
1805 return SYSTEM_VALUE_SAMPLE_ID;
1806 case nir_intrinsic_load_sample_pos:
1807 return SYSTEM_VALUE_SAMPLE_POS;
1808 case nir_intrinsic_load_sample_mask_in:
1809 return SYSTEM_VALUE_SAMPLE_MASK_IN;
1810 case nir_intrinsic_load_local_invocation_id:
1811 return SYSTEM_VALUE_LOCAL_INVOCATION_ID;
1812 case nir_intrinsic_load_num_work_groups:
1813 return SYSTEM_VALUE_NUM_WORK_GROUPS;
1814 case nir_intrinsic_load_work_group_id:
1815 return SYSTEM_VALUE_WORK_GROUP_ID;
1816 case nir_intrinsic_load_primitive_id:
1817 return SYSTEM_VALUE_PRIMITIVE_ID;
1818 case nir_intrinsic_load_tess_coord:
1819 return SYSTEM_VALUE_TESS_COORD;
1820 case nir_intrinsic_load_tess_level_outer:
1821 return SYSTEM_VALUE_TESS_LEVEL_OUTER;
1822 case nir_intrinsic_load_tess_level_inner:
1823 return SYSTEM_VALUE_TESS_LEVEL_INNER;
1824 case nir_intrinsic_load_patch_vertices_in:
1825 return SYSTEM_VALUE_VERTICES_IN;
1826 case nir_intrinsic_load_helper_invocation:
1827 return SYSTEM_VALUE_HELPER_INVOCATION;
1828 default:
1829 unreachable("intrinsic doesn't produce a system value");
1830 }
1831 }