src/mesa/drivers/dri/i965/brw_cfg.cpp

   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 *
  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 *
  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
  53 bblock_t::bblock_t(cfg_t *cfg) :
  54    cfg(cfg), idom(NULL), start_ip(0), end_ip(0), num(0)
  55 {
  56    instructions.make_empty();
  57    parents.make_empty();
  58    children.make_empty();
  59 }
  60
  61 void
  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
  69 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
  81 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
  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
 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
 114 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
 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
 144 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
 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
 162    bblock_t *cur = NULL;
 163    int ip = 0;
 164
 165    bblock_t *entry = new_block();
 166    bblock_t *cur_if = NULL;    /**< BB ending with IF. */
 167    bblock_t *cur_else = NULL;  /**< BB ending with ELSE. */
 168    bblock_t *cur_endif = NULL; /**< BB starting with ENDIF. */
 169    bblock_t *cur_do = NULL;    /**< BB starting with DO. */
 170    bblock_t *cur_while = NULL; /**< BB immediately following WHILE. */
 171    exec_list if_stack, else_stack, do_stack, while_stack;
 172    bblock_t *next;
 173
 174    set_next_block(&cur, entry, ip);
 175
 176    foreach_in_list_safe(backend_instruction, inst, instructions) {
 177       /* set_next_block wants the post-incremented ip */
 178       ip++;
 179
 180       inst->exec_node::remove();
 181
 182       switch (inst->opcode) {
 183       case BRW_OPCODE_IF:
 184          cur->instructions.push_tail(inst);
 185
 186          /* Push our information onto a stack so we can recover from
 187           * nested ifs.
 188           */
 189          if_stack.push_tail(link(mem_ctx, cur_if));
 190          else_stack.push_tail(link(mem_ctx, cur_else));
 191
 192          cur_if = cur;
 193          cur_else = NULL;
 194          cur_endif = NULL;
 195
 196          /* Set up our immediately following block, full of "then"
 197           * instructions.
 198           */
 199          next = new_block();
 200          cur_if->add_successor(mem_ctx, next);
 201
 202          set_next_block(&cur, next, ip);
 203          break;
 204
 205       case BRW_OPCODE_ELSE:
 206          cur->instructions.push_tail(inst);
 207
 208          cur_else = cur;
 209
 210          next = new_block();
 211          assert(cur_if != NULL);
 212          cur_if->add_successor(mem_ctx, next);
 213
 214          set_next_block(&cur, next, ip);
 215          break;
 216
 217       case BRW_OPCODE_ENDIF: {
 218          if (cur->instructions.is_empty()) {
 219             /* New block was just created; use it. */
 220             cur_endif = cur;
 221          } else {
 222             cur_endif = new_block();
 223
 224             cur->add_successor(mem_ctx, cur_endif);
 225
 226             set_next_block(&cur, cur_endif, ip - 1);
 227          }
 228
 229          cur->instructions.push_tail(inst);
 230
 231          if (cur_else) {
 232             cur_else->add_successor(mem_ctx, cur_endif);
 233          } else {
 234             assert(cur_if != NULL);
 235             cur_if->add_successor(mem_ctx, cur_endif);
 236          }
 237
 238          assert(cur_if->end()->opcode == BRW_OPCODE_IF);
 239          assert(!cur_else || cur_else->end()->opcode == BRW_OPCODE_ELSE);
 240
 241          /* Pop the stack so we're in the previous if/else/endif */
 242          cur_if = pop_stack(&if_stack);
 243          cur_else = pop_stack(&else_stack);
 244          break;
 245       }
 246       case BRW_OPCODE_DO:
 247          /* Push our information onto a stack so we can recover from
 248           * nested loops.
 249           */
 250          do_stack.push_tail(link(mem_ctx, cur_do));
 251          while_stack.push_tail(link(mem_ctx, cur_while));
 252
 253          /* Set up the block just after the while.  Don't know when exactly
 254           * it will start, yet.
 255           */
 256          cur_while = new_block();
 257
 258          if (cur->instructions.is_empty()) {
 259             /* New block was just created; use it. */
 260             cur_do = cur;
 261          } else {
 262             cur_do = new_block();
 263
 264             cur->add_successor(mem_ctx, cur_do);
 265
 266             set_next_block(&cur, cur_do, ip - 1);
 267          }
 268
 269          cur->instructions.push_tail(inst);
 270          break;
 271
 272       case BRW_OPCODE_CONTINUE:
 273          cur->instructions.push_tail(inst);
 274
 275          assert(cur_do != NULL);
 276          cur->add_successor(mem_ctx, cur_do);
 277
 278          next = new_block();
 279          if (inst->predicate)
 280             cur->add_successor(mem_ctx, next);
 281
 282          set_next_block(&cur, next, ip);
 283          break;
 284
 285       case BRW_OPCODE_BREAK:
 286          cur->instructions.push_tail(inst);
 287
 288          assert(cur_while != NULL);
 289          cur->add_successor(mem_ctx, cur_while);
 290
 291          next = new_block();
 292          if (inst->predicate)
 293             cur->add_successor(mem_ctx, next);
 294
 295          set_next_block(&cur, next, ip);
 296          break;
 297
 298       case BRW_OPCODE_WHILE:
 299          cur->instructions.push_tail(inst);
 300
 301          assert(cur_do != NULL && cur_while != NULL);
 302          cur->add_successor(mem_ctx, cur_do);
 303
 304          if (inst->predicate)
 305             cur->add_successor(mem_ctx, cur_while);
 306
 307          set_next_block(&cur, cur_while, ip);
 308
 309          /* Pop the stack so we're in the previous loop */
 310          cur_do = pop_stack(&do_stack);
 311          cur_while = pop_stack(&while_stack);
 312          break;
 313
 314       default:
 315          cur->instructions.push_tail(inst);
 316          break;
 317       }
 318    }
 319
 320    cur->end_ip = ip - 1;
 321
 322    make_block_array();
 323 }
 324
 325 cfg_t::~cfg_t()
 326 {
 327    ralloc_free(mem_ctx);
 328 }
 329
 330 void
 331 cfg_t::remove_block(bblock_t *block)
 332 {
 333    foreach_list_typed_safe (bblock_link, predecessor, link, &block->parents) {
 334       /* Remove block from all of its predecessors' successor lists. */
 335       foreach_list_typed_safe (bblock_link, successor, link,
 336                                &predecessor->block->children) {
 337          if (block == successor->block) {
 338             successor->link.remove();
 339             ralloc_free(successor);
 340          }
 341       }
 342
 343       /* Add removed-block's successors to its predecessors' successor lists. */
 344       foreach_list_typed (bblock_link, successor, link, &block->children) {
 345          if (!successor->block->is_successor_of(predecessor->block)) {
 346             predecessor->block->children.push_tail(link(mem_ctx,
 347                                                         successor->block));
 348          }
 349       }
 350    }
 351
 352    foreach_list_typed_safe (bblock_link, successor, link, &block->children) {
 353       /* Remove block from all of its childrens' parents lists. */
 354       foreach_list_typed_safe (bblock_link, predecessor, link,
 355                                &successor->block->parents) {
 356          if (block == predecessor->block) {
 357             predecessor->link.remove();
 358             ralloc_free(predecessor);
 359          }
 360       }
 361
 362       /* Add removed-block's predecessors to its successors' predecessor lists. */
 363       foreach_list_typed (bblock_link, predecessor, link, &block->parents) {
 364          if (!predecessor->block->is_predecessor_of(successor->block)) {
 365             successor->block->parents.push_tail(link(mem_ctx,
 366                                                      predecessor->block));
 367          }
 368       }
 369    }
 370
 371    block->link.remove();
 372
 373    for (int b = block->num; b < this->num_blocks - 1; b++) {
 374       this->blocks[b] = this->blocks[b + 1];
 375       this->blocks[b]->num = b;
 376    }
 377
 378    this->blocks[this->num_blocks - 1]->num = this->num_blocks - 2;
 379    this->num_blocks--;
 380    idom_dirty = true;
 381 }
 382
 383 bblock_t *
 384 cfg_t::new_block()
 385 {
 386    bblock_t *block = new(mem_ctx) bblock_t(this);
 387
 388    return block;
 389 }
 390
 391 void
 392 cfg_t::set_next_block(bblock_t **cur, bblock_t *block, int ip)
 393 {
 394    if (*cur) {
 395       (*cur)->end_ip = ip - 1;
 396    }
 397
 398    block->start_ip = ip;
 399    block->num = num_blocks++;
 400    block_list.push_tail(&block->link);
 401    *cur = block;
 402 }
 403
 404 void
 405 cfg_t::make_block_array()
 406 {
 407    blocks = ralloc_array(mem_ctx, bblock_t *, num_blocks);
 408
 409    int i = 0;
 410    foreach_block (block, this) {
 411       blocks[i++] = block;
 412    }
 413    assert(i == num_blocks);
 414 }
 415
 416 void
 417 cfg_t::dump(backend_shader *s)
 418 {
 419    if (idom_dirty)
 420       calculate_idom();
 421
 422    foreach_block (block, this) {
 423       if (block->idom)
 424          fprintf(stderr, "START B%d IDOM(B%d)", block->num, block->idom->num);
 425       else
 426          fprintf(stderr, "START B%d IDOM(none)", block->num);
 427
 428       foreach_list_typed(bblock_link, link, link, &block->parents) {
 429          fprintf(stderr, " <-B%d",
 430                  link->block->num);
 431       }
 432       fprintf(stderr, "\n");
 433       if (s != NULL)
 434          block->dump(s);
 435       fprintf(stderr, "END B%d", block->num);
 436       foreach_list_typed(bblock_link, link, link, &block->children) {
 437          fprintf(stderr, " ->B%d",
 438                  link->block->num);
 439       }
 440       fprintf(stderr, "\n");
 441    }
 442 }
 443
 444 /* Calculates the immediate dominator of each block, according to "A Simple,
 445  * Fast Dominance Algorithm" by Keith D. Cooper, Timothy J. Harvey, and Ken
 446  * Kennedy.
 447  *
 448  * The authors claim that for control flow graphs of sizes normally encountered
 449  * (less than 1000 nodes) that this algorithm is significantly faster than
 450  * others like Lengauer-Tarjan.
 451  */
 452 void
 453 cfg_t::calculate_idom()
 454 {
 455    foreach_block(block, this) {
 456       block->idom = NULL;
 457    }
 458    blocks[0]->idom = blocks[0];
 459
 460    bool changed;
 461    do {
 462       changed = false;
 463
 464       foreach_block(block, this) {
 465          if (block->num == 0)
 466             continue;
 467
 468          bblock_t *new_idom = NULL;
 469          foreach_list_typed(bblock_link, parent, link, &block->parents) {
 470             if (parent->block->idom) {
 471                if (new_idom == NULL) {
 472                   new_idom = parent->block;
 473                } else if (parent->block->idom != NULL) {
 474                   new_idom = intersect(parent->block, new_idom);
 475                }
 476             }
 477          }
 478
 479          if (block->idom != new_idom) {
 480             block->idom = new_idom;
 481             changed = true;
 482          }
 483       }
 484    } while (changed);
 485
 486    idom_dirty = false;
 487 }
 488
 489 bblock_t *
 490 cfg_t::intersect(bblock_t *b1, bblock_t *b2)
 491 {
 492    /* Note, the comparisons here are the opposite of what the paper says
 493     * because we index blocks from beginning -> end (i.e. reverse post-order)
 494     * instead of post-order like they assume.
 495     */
 496    while (b1->num != b2->num) {
 497       while (b1->num > b2->num)
 498          b1 = b1->idom;
 499       while (b2->num > b1->num)
 500          b2 = b2->idom;
 501    }
 502    assert(b1);
 503    return b1;
 504 }
 505
 506 void
 507 cfg_t::dump_cfg()
 508 {
 509    printf("digraph CFG {\n");
 510    for (int b = 0; b < num_blocks; b++) {
 511       bblock_t *block = this->blocks[b];
 512
 513       foreach_list_typed_safe (bblock_link, child, link, &block->children) {
 514          printf("\t%d -> %d\n", b, child->block->num);
 515       }
 516    }
 517    printf("}\n");
 518 }
 519
 520 void
 521 cfg_t::dump_domtree()
 522 {
 523    printf("digraph DominanceTree {\n");
 524    foreach_block(block, this) {
 525       if (block->idom) {
 526          printf("\t%d -> %d\n", block->idom->num, block->num);
 527       }
 528    }
 529    printf("}\n");
 530 }