1 /*
2 * Copyright © 2012 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 * Eric Anholt <eric@anholt.net>
25 *
26 */
27
28 #include "brw_cfg.h"
29
30 /** @file brw_cfg.cpp
31 *
32 * Walks the shader instructions generated and creates a set of basic
33 * blocks with successor/predecessor edges connecting them.
34 */
35
36 static bblock_t *
pop_stack(exec_list * list)37 pop_stack(exec_list *list)
38 {
39 bblock_link *link = (bblock_link *)list->get_tail();
40 bblock_t *block = link->block;
41 link->link.remove();
42
43 return block;
44 }
45
46 static exec_node *
link(void * mem_ctx,bblock_t * block)47 link(void *mem_ctx, bblock_t *block)
48 {
49 bblock_link *l = new(mem_ctx) bblock_link(block);
50 return &l->link;
51 }
52
bblock_t(cfg_t * cfg)53 bblock_t::bblock_t(cfg_t *cfg) :
54 cfg(cfg), idom(NULL), start_ip(0), end_ip(0), num(0), cycle_count(0)
55 {
56 instructions.make_empty();
57 parents.make_empty();
58 children.make_empty();
59 }
60
61 void
add_successor(void * mem_ctx,bblock_t * successor)62 bblock_t::add_successor(void *mem_ctx, bblock_t *successor)
63 {
64 successor->parents.push_tail(::link(mem_ctx, this));
65 children.push_tail(::link(mem_ctx, successor));
66 }
67
68 bool
is_predecessor_of(const bblock_t * block) const69 bblock_t::is_predecessor_of(const bblock_t *block) const
70 {
71 foreach_list_typed_safe (bblock_link, parent, link, &block->parents) {
72 if (parent->block == this) {
73 return true;
74 }
75 }
76
77 return false;
78 }
79
80 bool
is_successor_of(const bblock_t * block) const81 bblock_t::is_successor_of(const bblock_t *block) const
82 {
83 foreach_list_typed_safe (bblock_link, child, link, &block->children) {
84 if (child->block == this) {
85 return true;
86 }
87 }
88
89 return false;
90 }
91
92 static bool
ends_block(const backend_instruction * inst)93 ends_block(const backend_instruction *inst)
94 {
95 enum opcode op = inst->opcode;
96
97 return op == BRW_OPCODE_IF ||
98 op == BRW_OPCODE_ELSE ||
99 op == BRW_OPCODE_CONTINUE ||
100 op == BRW_OPCODE_BREAK ||
101 op == BRW_OPCODE_WHILE;
102 }
103
104 static bool
starts_block(const backend_instruction * inst)105 starts_block(const backend_instruction *inst)
106 {
107 enum opcode op = inst->opcode;
108
109 return op == BRW_OPCODE_DO ||
110 op == BRW_OPCODE_ENDIF;
111 }
112
113 bool
can_combine_with(const bblock_t * that) const114 bblock_t::can_combine_with(const bblock_t *that) const
115 {
116 if ((const bblock_t *)this->link.next != that)
117 return false;
118
119 if (ends_block(this->end()) ||
120 starts_block(that->start()))
121 return false;
122
123 return true;
124 }
125
126 void
combine_with(bblock_t * that)127 bblock_t::combine_with(bblock_t *that)
128 {
129 assert(this->can_combine_with(that));
130 foreach_list_typed (bblock_link, link, link, &this->children) {
131 assert(link->block == that);
132 }
133 foreach_list_typed (bblock_link, link, link, &that->parents) {
134 assert(link->block == this);
135 }
136
137 this->end_ip = that->end_ip;
138 this->instructions.append_list(&that->instructions);
139
140 this->cfg->remove_block(that);
141 }
142
143 void
dump(backend_shader * s) const144 bblock_t::dump(backend_shader *s) const
145 {
146 int ip = this->start_ip;
147 foreach_inst_in_block(backend_instruction, inst, this) {
148 fprintf(stderr, "%5d: ", ip);
149 s->dump_instruction(inst);
150 ip++;
151 }
152 }
153
cfg_t(exec_list * instructions)154 cfg_t::cfg_t(exec_list *instructions)
155 {
156 mem_ctx = ralloc_context(NULL);
157 block_list.make_empty();
158 blocks = NULL;
159 num_blocks = 0;
160 idom_dirty = true;
161 cycle_count = 0;
162
163 bblock_t *cur = NULL;
164 int ip = 0;
165
166 bblock_t *entry = new_block();
167 bblock_t *cur_if = NULL; /**< BB ending with IF. */
168 bblock_t *cur_else = NULL; /**< BB ending with ELSE. */
169 bblock_t *cur_endif = NULL; /**< BB starting with ENDIF. */
170 bblock_t *cur_do = NULL; /**< BB starting with DO. */
171 bblock_t *cur_while = NULL; /**< BB immediately following WHILE. */
172 exec_list if_stack, else_stack, do_stack, while_stack;
173 bblock_t *next;
174
175 set_next_block(&cur, entry, ip);
176
177 foreach_in_list_safe(backend_instruction, inst, instructions) {
178 /* set_next_block wants the post-incremented ip */
179 ip++;
180
181 inst->exec_node::remove();
182
183 switch (inst->opcode) {
184 case BRW_OPCODE_IF:
185 cur->instructions.push_tail(inst);
186
187 /* Push our information onto a stack so we can recover from
188 * nested ifs.
189 */
190 if_stack.push_tail(link(mem_ctx, cur_if));
191 else_stack.push_tail(link(mem_ctx, cur_else));
192
193 cur_if = cur;
194 cur_else = NULL;
195 cur_endif = NULL;
196
197 /* Set up our immediately following block, full of "then"
198 * instructions.
199 */
200 next = new_block();
201 cur_if->add_successor(mem_ctx, next);
202
203 set_next_block(&cur, next, ip);
204 break;
205
206 case BRW_OPCODE_ELSE:
207 cur->instructions.push_tail(inst);
208
209 cur_else = cur;
210
211 next = new_block();
212 assert(cur_if != NULL);
213 cur_if->add_successor(mem_ctx, next);
214
215 set_next_block(&cur, next, ip);
216 break;
217
218 case BRW_OPCODE_ENDIF: {
219 if (cur->instructions.is_empty()) {
220 /* New block was just created; use it. */
221 cur_endif = cur;
222 } else {
223 cur_endif = new_block();
224
225 cur->add_successor(mem_ctx, cur_endif);
226
227 set_next_block(&cur, cur_endif, ip - 1);
228 }
229
230 cur->instructions.push_tail(inst);
231
232 if (cur_else) {
233 cur_else->add_successor(mem_ctx, cur_endif);
234 } else {
235 assert(cur_if != NULL);
236 cur_if->add_successor(mem_ctx, cur_endif);
237 }
238
239 assert(cur_if->end()->opcode == BRW_OPCODE_IF);
240 assert(!cur_else || cur_else->end()->opcode == BRW_OPCODE_ELSE);
241
242 /* Pop the stack so we're in the previous if/else/endif */
243 cur_if = pop_stack(&if_stack);
244 cur_else = pop_stack(&else_stack);
245 break;
246 }
247 case BRW_OPCODE_DO:
248 /* Push our information onto a stack so we can recover from
249 * nested loops.
250 */
251 do_stack.push_tail(link(mem_ctx, cur_do));
252 while_stack.push_tail(link(mem_ctx, cur_while));
253
254 /* Set up the block just after the while. Don't know when exactly
255 * it will start, yet.
256 */
257 cur_while = new_block();
258
259 if (cur->instructions.is_empty()) {
260 /* New block was just created; use it. */
261 cur_do = cur;
262 } else {
263 cur_do = new_block();
264
265 cur->add_successor(mem_ctx, cur_do);
266
267 set_next_block(&cur, cur_do, ip - 1);
268 }
269
270 cur->instructions.push_tail(inst);
271 break;
272
273 case BRW_OPCODE_CONTINUE:
274 cur->instructions.push_tail(inst);
275
276 assert(cur_do != NULL);
277 cur->add_successor(mem_ctx, cur_do);
278
279 next = new_block();
280 if (inst->predicate)
281 cur->add_successor(mem_ctx, next);
282
283 set_next_block(&cur, next, ip);
284 break;
285
286 case BRW_OPCODE_BREAK:
287 cur->instructions.push_tail(inst);
288
289 assert(cur_while != NULL);
290 cur->add_successor(mem_ctx, cur_while);
291
292 next = new_block();
293 if (inst->predicate)
294 cur->add_successor(mem_ctx, next);
295
296 set_next_block(&cur, next, ip);
297 break;
298
299 case BRW_OPCODE_WHILE:
300 cur->instructions.push_tail(inst);
301
302 assert(cur_do != NULL && cur_while != NULL);
303 cur->add_successor(mem_ctx, cur_do);
304
305 if (inst->predicate)
306 cur->add_successor(mem_ctx, cur_while);
307
308 set_next_block(&cur, cur_while, ip);
309
310 /* Pop the stack so we're in the previous loop */
311 cur_do = pop_stack(&do_stack);
312 cur_while = pop_stack(&while_stack);
313 break;
314
315 default:
316 cur->instructions.push_tail(inst);
317 break;
318 }
319 }
320
321 cur->end_ip = ip - 1;
322
323 make_block_array();
324 }
325
~cfg_t()326 cfg_t::~cfg_t()
327 {
328 ralloc_free(mem_ctx);
329 }
330
331 void
remove_block(bblock_t * block)332 cfg_t::remove_block(bblock_t *block)
333 {
334 foreach_list_typed_safe (bblock_link, predecessor, link, &block->parents) {
335 /* Remove block from all of its predecessors' successor lists. */
336 foreach_list_typed_safe (bblock_link, successor, link,
337 &predecessor->block->children) {
338 if (block == successor->block) {
339 successor->link.remove();
340 ralloc_free(successor);
341 }
342 }
343
344 /* Add removed-block's successors to its predecessors' successor lists. */
345 foreach_list_typed (bblock_link, successor, link, &block->children) {
346 if (!successor->block->is_successor_of(predecessor->block)) {
347 predecessor->block->children.push_tail(link(mem_ctx,
348 successor->block));
349 }
350 }
351 }
352
353 foreach_list_typed_safe (bblock_link, successor, link, &block->children) {
354 /* Remove block from all of its childrens' parents lists. */
355 foreach_list_typed_safe (bblock_link, predecessor, link,
356 &successor->block->parents) {
357 if (block == predecessor->block) {
358 predecessor->link.remove();
359 ralloc_free(predecessor);
360 }
361 }
362
363 /* Add removed-block's predecessors to its successors' predecessor lists. */
364 foreach_list_typed (bblock_link, predecessor, link, &block->parents) {
365 if (!predecessor->block->is_predecessor_of(successor->block)) {
366 successor->block->parents.push_tail(link(mem_ctx,
367 predecessor->block));
368 }
369 }
370 }
371
372 block->link.remove();
373
374 for (int b = block->num; b < this->num_blocks - 1; b++) {
375 this->blocks[b] = this->blocks[b + 1];
376 this->blocks[b]->num = b;
377 }
378
379 this->blocks[this->num_blocks - 1]->num = this->num_blocks - 2;
380 this->num_blocks--;
381 idom_dirty = true;
382 }
383
384 bblock_t *
new_block()385 cfg_t::new_block()
386 {
387 bblock_t *block = new(mem_ctx) bblock_t(this);
388
389 return block;
390 }
391
392 void
set_next_block(bblock_t ** cur,bblock_t * block,int ip)393 cfg_t::set_next_block(bblock_t **cur, bblock_t *block, int ip)
394 {
395 if (*cur) {
396 (*cur)->end_ip = ip - 1;
397 }
398
399 block->start_ip = ip;
400 block->num = num_blocks++;
401 block_list.push_tail(&block->link);
402 *cur = block;
403 }
404
405 void
make_block_array()406 cfg_t::make_block_array()
407 {
408 blocks = ralloc_array(mem_ctx, bblock_t *, num_blocks);
409
410 int i = 0;
411 foreach_block (block, this) {
412 blocks[i++] = block;
413 }
414 assert(i == num_blocks);
415 }
416
417 void
dump(backend_shader * s)418 cfg_t::dump(backend_shader *s)
419 {
420 if (idom_dirty)
421 calculate_idom();
422
423 foreach_block (block, this) {
424 if (block->idom)
425 fprintf(stderr, "START B%d IDOM(B%d)", block->num, block->idom->num);
426 else
427 fprintf(stderr, "START B%d IDOM(none)", block->num);
428
429 foreach_list_typed(bblock_link, link, link, &block->parents) {
430 fprintf(stderr, " <-B%d",
431 link->block->num);
432 }
433 fprintf(stderr, "\n");
434 if (s != NULL)
435 block->dump(s);
436 fprintf(stderr, "END B%d", block->num);
437 foreach_list_typed(bblock_link, link, link, &block->children) {
438 fprintf(stderr, " ->B%d",
439 link->block->num);
440 }
441 fprintf(stderr, "\n");
442 }
443 }
444
445 /* Calculates the immediate dominator of each block, according to "A Simple,
446 * Fast Dominance Algorithm" by Keith D. Cooper, Timothy J. Harvey, and Ken
447 * Kennedy.
448 *
449 * The authors claim that for control flow graphs of sizes normally encountered
450 * (less than 1000 nodes) that this algorithm is significantly faster than
451 * others like Lengauer-Tarjan.
452 */
453 void
calculate_idom()454 cfg_t::calculate_idom()
455 {
456 foreach_block(block, this) {
457 block->idom = NULL;
458 }
459 blocks[0]->idom = blocks[0];
460
461 bool changed;
462 do {
463 changed = false;
464
465 foreach_block(block, this) {
466 if (block->num == 0)
467 continue;
468
469 bblock_t *new_idom = NULL;
470 foreach_list_typed(bblock_link, parent, link, &block->parents) {
471 if (parent->block->idom) {
472 if (new_idom == NULL) {
473 new_idom = parent->block;
474 } else if (parent->block->idom != NULL) {
475 new_idom = intersect(parent->block, new_idom);
476 }
477 }
478 }
479
480 if (block->idom != new_idom) {
481 block->idom = new_idom;
482 changed = true;
483 }
484 }
485 } while (changed);
486
487 idom_dirty = false;
488 }
489
490 bblock_t *
intersect(bblock_t * b1,bblock_t * b2)491 cfg_t::intersect(bblock_t *b1, bblock_t *b2)
492 {
493 /* Note, the comparisons here are the opposite of what the paper says
494 * because we index blocks from beginning -> end (i.e. reverse post-order)
495 * instead of post-order like they assume.
496 */
497 while (b1->num != b2->num) {
498 while (b1->num > b2->num)
499 b1 = b1->idom;
500 while (b2->num > b1->num)
501 b2 = b2->idom;
502 }
503 assert(b1);
504 return b1;
505 }
506
507 void
dump_cfg()508 cfg_t::dump_cfg()
509 {
510 printf("digraph CFG {\n");
511 for (int b = 0; b < num_blocks; b++) {
512 bblock_t *block = this->blocks[b];
513
514 foreach_list_typed_safe (bblock_link, child, link, &block->children) {
515 printf("\t%d -> %d\n", b, child->block->num);
516 }
517 }
518 printf("}\n");
519 }
520
521 void
dump_domtree()522 cfg_t::dump_domtree()
523 {
524 printf("digraph DominanceTree {\n");
525 foreach_block(block, this) {
526 if (block->idom) {
527 printf("\t%d -> %d\n", block->idom->num, block->num);
528 }
529 }
530 printf("}\n");
531 }
532