nsz Git - libfirm/blob - ir/be/bespillbelady3.c

   1 /*
   2  * Copyright (C) 1995-2008 University of Karlsruhe.  All right reserved.
   3  *
   4  * This file is part of libFirm.
   5  *
   6  * This file may be distributed and/or modified under the terms of the
   7  * GNU General Public License version 2 as published by the Free Software
   8  * Foundation and appearing in the file LICENSE.GPL included in the
   9  * packaging of this file.
  10  *
  11  * Licensees holding valid libFirm Professional Edition licenses may use
  12  * this file in accordance with the libFirm Commercial License.
  13  * Agreement provided with the Software.
  14  *
  15  * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
  16  * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  17  * PURPOSE.
  18  */
  19
  20 /**
  21  * @file
  22  * @brief       MIN-algorithm with some twists (aka belady spiller v3)
  23  * @author      Matthias Braun
  24  * @date        15.01.2008
  25  * @version     $Id$
  26  *
  27  * TODO:
  28  *   - handle phis correctly, decide whether we should spill them ~ok
  29  *   - merge multiple start worksets of blocks ~ok
  30  *   - how to and when to do the tentative phase...
  31  */
  32 #include "config.h"
  33
  34 #include <stdbool.h>
  35
  36 #include "debug.h"
  37 #include "list.h"
  38 #include "pdeq.h"
  39
  40 #include "irnode_t.h"
  41 #include "irprintf.h"
  42 #include "iredges_t.h"
  43 #include "irloop_t.h"
  44 #include "irgwalk.h"
  45 #include "execfreq.h"
  46
  47 #include "bemodule.h"
  48 #include "bespill.h"
  49 #include "beutil.h"
  50 #include "bespilloptions.h"
  51 #include "besched_t.h"
  52 #include "be_t.h"
  53
  54 #ifndef NDEBUG
  55 #define EXPENSIVE_CHECKS
  56 #endif
  57
  58 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
  59
  60 typedef struct loop_edge_t loop_edge_t;
  61 struct loop_edge_t {
  62         ir_node     *block;
  63         int          pos;
  64         loop_edge_t *next;
  65 };
  66
  67 typedef struct loop_info_t loop_info_t;
  68 struct loop_info_t {
  69         ir_loop     *loop;
  70         loop_edge_t *entry_edges;
  71         loop_edge_t *exit_edges;
  72         size_t       max_register_pressure;
  73 };
  74
  75 typedef struct worklist_entry_t worklist_entry_t;
  76 struct worklist_entry_t {
  77         struct list_head  head;
  78         ir_node          *value;
  79         ir_node          *reload_point;
  80         ir_loop          *unused_livethrough_loop;
  81 };
  82
  83 typedef struct worklist_t worklist_t;
  84 struct worklist_t {
  85         struct list_head  live_values;
  86         size_t            n_live_values;
  87         ir_visited_t      visited;
  88 };
  89
  90 typedef struct block_info_t block_info_t;
  91 struct block_info_t {
  92         worklist_t *start_worklist;
  93         worklist_t *end_worklist;
  94 };
  95
  96 static const arch_register_class_t *cls;
  97 static struct obstack               obst;
  98 static spill_env_t                 *senv;
  99 static size_t                       n_regs;
 100 static size_t                       max_register_pressure;
 101 static bool                         tentative_mode;
 102 static bool                         should_have_reached_fixpoint;
 103 static bool                         do_push_unused_livethroughs;
 104 static ir_exec_freq                *exec_freq;
 105 static ir_visited_t                 worklist_visited;
 106
 107 static worklist_t *new_worklist(void)
 108 {
 109         worklist_t *worklist = obstack_alloc(&obst, sizeof(worklist[0]));
 110         memset(worklist, 0, sizeof(worklist[0]));
 111
 112         INIT_LIST_HEAD(&worklist->live_values);
 113         worklist->n_live_values    = 0;
 114
 115         return worklist;
 116 }
 117
 118 static void mark_irn_not_visited(ir_node *node)
 119 {
 120         set_irn_visited(node, get_irg_visited(current_ir_graph) - 1);
 121 }
 122
 123 static bool worklist_contains(const ir_node *node)
 124 {
 125         return irn_visited(node);
 126 }
 127
 128 static block_info_t *get_block_info(ir_node *block)
 129 {
 130         block_info_t *info = get_irn_link(block);
 131         if (info != NULL)
 132                 return info;
 133
 134         info = obstack_alloc(&obst, sizeof(info[0]));
 135         memset(info, 0, sizeof(info[0]));
 136         set_irn_link(block, info);
 137         return info;
 138 }
 139
 140 static void deactivate_worklist(const worklist_t *worklist)
 141 {
 142         struct list_head *entry;
 143
 144         list_for_each(entry, &worklist->live_values) {
 145                 worklist_entry_t *wl_entry
 146                         = list_entry(entry, worklist_entry_t, head);
 147                 assert(worklist_contains(wl_entry->value));
 148                 mark_irn_not_visited(wl_entry->value);
 149                 set_irn_link(wl_entry->value, NULL);
 150         }
 151 }
 152
 153 static void activate_worklist(const worklist_t *worklist)
 154 {
 155         struct list_head *entry;
 156
 157         list_for_each(entry, &worklist->live_values) {
 158                 worklist_entry_t *wl_entry
 159                         = list_entry(entry, worklist_entry_t, head);
 160                 assert(!worklist_contains(wl_entry->value));
 161                 mark_irn_visited(wl_entry->value);
 162                 set_irn_link(wl_entry->value, wl_entry);
 163         }
 164 }
 165
 166 /**
 167  * duplicate a worklist and directly activates it
 168  */
 169 static void fill_and_activate_worklist(worklist_t *new_worklist,
 170                 const worklist_t *worklist, ir_node *block, ir_node *succ_block,
 171                 int succ_pos)
 172 {
 173         ir_node          *reload_point  = NULL;
 174         size_t            n_live_values = 0;
 175         struct list_head *entry;
 176
 177         if (succ_block != NULL &&
 178                         (get_Block_n_cfgpreds(succ_block) > 1
 179                          || get_irn_n_edges_kind(block, EDGE_KIND_BLOCK) > 1)) {
 180                 reload_point = be_get_end_of_block_insertion_point(block);
 181         }
 182
 183         list_for_each(entry, &worklist->live_values) {
 184                 worklist_entry_t *wl_entry  = list_entry(entry, worklist_entry_t, head);
 185                 ir_node          *value     = wl_entry->value;
 186                 worklist_entry_t *new_entry;
 187
 188                 if (new_worklist->n_live_values >= n_regs)
 189                         break;
 190
 191                 if (is_Phi(value) && get_nodes_block(value) == succ_block) {
 192                         value = get_Phi_pred(value, succ_pos);
 193
 194                         /* can happen for unknown phi preds */
 195                         if (!arch_irn_consider_in_reg_alloc(cls, value))
 196                                 continue;
 197                 }
 198
 199                 if (irn_visited_else_mark(value))
 200                         continue;
 201
 202                 new_entry = obstack_alloc(&obst, sizeof(new_entry[0]));
 203                 memset(new_entry, 0, sizeof(new_entry[0]));
 204
 205                 new_entry->value = value;
 206                 if (reload_point != NULL) {
 207                         new_entry->reload_point = reload_point;
 208                 } else {
 209                         new_entry->reload_point = wl_entry->reload_point;
 210                 }
 211
 212                 list_add_tail(&new_entry->head, &new_worklist->live_values);
 213                 ++n_live_values;
 214
 215                 set_irn_link(value, new_entry);
 216                 new_worklist->n_live_values++;
 217         }
 218 }
 219
 220 static worklist_t *duplicate_worklist(const worklist_t *worklist)
 221 {
 222         worklist_t       *new_worklist;
 223         struct list_head *entry;
 224
 225         new_worklist = obstack_alloc(&obst, sizeof(new_worklist[0]));
 226         memset(new_worklist, 0, sizeof(new_worklist[0]));
 227         INIT_LIST_HEAD(&new_worklist->live_values);
 228         new_worklist->n_live_values = worklist->n_live_values;
 229
 230         list_for_each(entry, &worklist->live_values) {
 231                 worklist_entry_t *wl_entry  = list_entry(entry, worklist_entry_t, head);
 232                 worklist_entry_t *new_entry
 233                         = obstack_alloc(&obst, sizeof(new_entry[0]));
 234
 235                 memcpy(new_entry, wl_entry, sizeof(new_entry[0]));
 236                 list_add_tail(&new_entry->head, &new_worklist->live_values);
 237         }
 238
 239         return new_worklist;
 240 }
 241
 242 #ifdef DEBUG_libfirm
 243 static void print_worklist(const worklist_t *worklist, int level)
 244 {
 245         struct list_head *entry;
 246
 247         DB((dbg, level, "%d values: ", worklist->n_live_values));
 248         list_for_each(entry, &worklist->live_values) {
 249                 worklist_entry_t *wl_entry
 250                         = list_entry(entry, worklist_entry_t, head);
 251
 252                 DB((dbg, level, "%+F ", wl_entry->value));
 253         }
 254 }
 255 #endif
 256
 257
 258
 259 static void clear_loop_info(ir_loop *loop)
 260 {
 261         int n_elements = get_loop_n_elements(loop);
 262         int i;
 263
 264         loop->link = NULL;
 265         for (i = 0; i < n_elements; ++i) {
 266                 loop_element element = get_loop_element(loop, i);
 267                 if (*element.kind != k_ir_loop)
 268                         continue;
 269
 270                 clear_loop_info(element.son);
 271         }
 272 }
 273
 274 static loop_info_t *get_loop_info(ir_loop *loop)
 275 {
 276         loop_info_t *info = loop->link;
 277         if (info != NULL)
 278                 return info;
 279
 280         info = obstack_alloc(&obst, sizeof(info[0]));
 281         memset(info, 0, sizeof(info[0]));
 282         info->loop = loop;
 283         loop->link = info;
 284         return info;
 285 }
 286
 287 /**
 288  * constructs loop in+out edges when called in a block walker
 289  */
 290 static void construct_loop_edges(ir_node *block, void *data)
 291 {
 292         int      n_cfgpreds = get_Block_n_cfgpreds(block);
 293         ir_loop *loop       = get_irn_loop(block);
 294         int      i;
 295
 296         (void) data;
 297
 298         for (i = 0; i < n_cfgpreds; ++i) {
 299                 ir_node     *cfgpred_block = get_Block_cfgpred_block(block, i);
 300                 ir_loop     *cfgpred_loop  = get_irn_loop(cfgpred_block);
 301                 loop_edge_t *edge;
 302                 bool        is_exit_edge;
 303                 ir_loop     *l, *goal;
 304
 305                 if (cfgpred_loop == loop)
 306                         continue;
 307
 308                 /* critical edges are splitted, so we can't jump from 1 loop
 309                  * directly into another without going out of it */
 310                 assert(get_loop_depth(cfgpred_loop) != get_loop_depth(loop));
 311
 312                 /* edge out of loop */
 313                 if (get_loop_depth(cfgpred_loop) > get_loop_depth(loop)) {
 314                         is_exit_edge = true;
 315                         l            = cfgpred_loop;
 316                         goal         = loop;
 317                 } else {
 318                         is_exit_edge = false;
 319                         l            = loop;
 320                         goal         = cfgpred_loop;
 321                 }
 322
 323                 /* this might be a jump out/in multiple loops */
 324                 do {
 325                         loop_info_t *l_info = get_loop_info(l);
 326
 327                         edge = obstack_alloc(&obst, sizeof(edge[0]));
 328                         memset(edge, 0, sizeof(edge[0]));
 329                         edge->block = block;
 330                         edge->pos   = i;
 331
 332                         if (is_exit_edge) {
 333                                 edge->next         = l_info->exit_edges;
 334                                 l_info->exit_edges = edge;
 335                                 assert(get_loop_depth(loop) < get_loop_depth(l));
 336                         } else {
 337                                 edge->next          = l_info->entry_edges;
 338                                 l_info->entry_edges = edge;
 339                                 assert(get_loop_depth(cfgpred_loop) < get_loop_depth(l));
 340                         }
 341                         l = get_loop_outer_loop(l);
 342                 } while(l != goal);
 343         }
 344 }
 345
 346
 347
 348
 349 static void place_reload(worklist_entry_t *entry)
 350 {
 351         if (tentative_mode)
 352                 return;
 353
 354         DB((dbg, LEVEL_1, "reload of %+F before %+F\n", entry->value,
 355                 entry->reload_point));
 356         assert(entry->reload_point != NULL);
 357         be_add_reload(senv, entry->value, entry->reload_point, cls, 1);
 358         entry->reload_point = NULL;
 359 }
 360
 361 /**
 362  * makes sure the worklist contains not more than n_regs - room_needed entries
 363  */
 364 static void make_room(worklist_t *worklist, size_t room_needed)
 365 {
 366         int               i;
 367         int               spills_needed;
 368         struct list_head *entry;
 369
 370         spills_needed = worklist->n_live_values + room_needed - n_regs;
 371         if (spills_needed <= 0)
 372                 return;
 373
 374         entry = worklist->live_values.next;
 375         for(i = spills_needed; i > 0; --i) {
 376                 struct list_head *next = entry->next;
 377                 worklist_entry_t *wl_entry
 378                         = list_entry(entry, worklist_entry_t, head);
 379
 380                 assert(worklist_contains(wl_entry->value));
 381                 mark_irn_not_visited(wl_entry->value);
 382                 place_reload(wl_entry);
 383                 list_del(entry);
 384
 385                 entry = next;
 386         }
 387         worklist->n_live_values -= spills_needed;
 388 }
 389
 390 /**
 391  * a value was used, so bring it to the back of the worklist (which might
 392  * result in a spill of another value).
 393  */
 394 static void val_used(worklist_t *worklist, ir_node *value, ir_node *sched_point)
 395 {
 396         /* already in the worklist? move around, otherwise add at back */
 397         worklist_entry_t *entry = get_irn_link(value);
 398
 399         assert(arch_irn_consider_in_reg_alloc(cls, value));
 400
 401         if (worklist_contains(value)) {
 402                 assert(entry != NULL);
 403
 404                 list_del(&entry->head);
 405         } else {
 406                 if (entry == NULL) {
 407                         entry        = obstack_alloc(&obst, sizeof(entry[0]));
 408                         memset(entry, 0, sizeof(entry[0]));
 409
 410                         entry->value = value;
 411                         set_irn_link(value, entry);
 412                 }
 413
 414                 ++worklist->n_live_values;
 415                 mark_irn_visited(value);
 416         }
 417
 418         entry->reload_point = sched_point;
 419         list_add_tail(&entry->head, &worklist->live_values);
 420 }
 421
 422 static void worklist_remove(worklist_t *worklist, ir_node *value)
 423 {
 424         worklist_entry_t *entry = get_irn_link(value);
 425         assert(entry != NULL);
 426         list_del(&entry->head);
 427         --worklist->n_live_values;
 428
 429         assert(worklist_contains(value));
 430         mark_irn_not_visited(value);
 431 }
 432
 433 static void update_max_pressure(worklist_t *worklist)
 434 {
 435         if (worklist->n_live_values > max_register_pressure)
 436                 max_register_pressure = worklist->n_live_values;
 437 }
 438
 439 static void do_spilling(ir_node *block, worklist_t *worklist)
 440 {
 441         ir_node *node;
 442
 443         assert(worklist != NULL);
 444
 445         sched_foreach_reverse(block, node) {
 446                 int    i, arity;
 447                 size_t n_defs = 0;
 448
 449                 DB((dbg, LEVEL_2, "\t%+F... ", node));
 450                 update_max_pressure(worklist);
 451
 452                 if (is_Phi(node)) {
 453                         ir_node *node2;
 454                         /* TODO: if we have some free registers, then we could decide to
 455                          * not spill some phis (but not for phis where at least 1 input is
 456                          * themselfes) */
 457
 458                         /* we have to spill all phis that are not live */
 459                         sched_foreach_reverse_from(node, node2) {
 460                                 assert(is_Phi(node2));
 461
 462                                 if (worklist_contains(node2))
 463                                         continue;
 464                                 if (!arch_irn_consider_in_reg_alloc(cls, node2))
 465                                         continue;
 466
 467                                 if (!tentative_mode)
 468                                         be_spill_phi(senv, node2);
 469                         }
 470                         DB((dbg, LEVEL_2, "\n"));
 471                         break;
 472                 }
 473
 474                 /* remove values defined by this instruction from the workset. Values
 475                  * defined but not in the workset need free registers */
 476                 if (get_irn_mode(node) == mode_T) {
 477                         const ir_edge_t *edge;
 478
 479                         foreach_out_edge(node, edge) {
 480                                 ir_node *proj = get_edge_src_irn(edge);
 481                                 if (!arch_irn_consider_in_reg_alloc(cls, proj))
 482                                         continue;
 483                                 if (worklist_contains(proj)) {
 484                                         worklist_remove(worklist, proj);
 485                                 } else {
 486                                         ++n_defs;
 487                                 }
 488                         }
 489                 } else if (arch_irn_consider_in_reg_alloc(cls, node)) {
 490                         if (worklist_contains(node)) {
 491                                 worklist_remove(worklist, node);
 492                         } else {
 493                                 n_defs = 1;
 494                         }
 495                 }
 496
 497                 /* make sure we have enough free registers for the spills */
 498                 make_room(worklist, n_defs);
 499
 500                 /* put all values used by the instruction into the workset */
 501                 arity = get_irn_arity(node);
 502                 for(i = 0; i < arity; ++i) {
 503                         ir_node *use = get_irn_n(node, i);
 504
 505                         if (!arch_irn_consider_in_reg_alloc(cls, use))
 506                                 continue;
 507
 508                         val_used(worklist, use, node);
 509                 }
 510
 511                 /* we might have too many values in the worklist now and need to spill
 512                  * some */
 513                 make_room(worklist, 0);
 514
 515 #ifdef DEBUG_libfirm
 516                 print_worklist(worklist, LEVEL_2);
 517                 DB((dbg, LEVEL_2, "\n"));
 518 #endif
 519         }
 520
 521         update_max_pressure(worklist);
 522
 523 #ifdef DEBUG_libfirm
 524         DB((dbg, LEVEL_1, "worklist at begin of %+F:", block));
 525         print_worklist(worklist, LEVEL_1);
 526         DB((dbg, LEVEL_1, "\n"));
 527 #endif
 528 }
 529
 530 static bool worklists_equal(const worklist_t *wl1, const worklist_t *wl2)
 531 {
 532         const struct list_head *i1 = &wl1->live_values;
 533         const struct list_head *i2 = &wl2->live_values;
 534
 535         for ( ; i1 != &wl1->live_values && i2 != &wl2->live_values;
 536                         i1 = i1->next, i2 = i2->next) {
 537                 worklist_entry_t *entry1 = list_entry(i1, worklist_entry_t, head);
 538                 worklist_entry_t *entry2 = list_entry(i2, worklist_entry_t, head);
 539
 540                 if (entry1->value != entry2->value)
 541                         return false;
 542         }
 543         /* both lists at the end */
 544         if (i1 != &wl1->live_values || i2 != &wl2->live_values)
 545                 return false;
 546
 547         return true;
 548 }
 549
 550 static bool fill_start_worklist(worklist_t *new_worklist, ir_node *block)
 551 {
 552         double           best_execfreq   = -1;
 553         worklist_t      *best_worklist   = NULL;
 554         ir_node         *best_succ_block = NULL;
 555         int              best_pos;
 556         const ir_edge_t *edge;
 557
 558         /* construct worklist */
 559         foreach_block_succ(block, edge) {
 560                 ir_node      *succ_block = get_edge_src_irn(edge);
 561                 double       execfreq    = get_block_execfreq(exec_freq, succ_block);
 562                 block_info_t *block_info;
 563                 worklist_t   *succ_worklist;
 564
 565                 if (execfreq < best_execfreq)
 566                         continue;
 567
 568                 block_info    = get_block_info(succ_block);
 569                 succ_worklist = block_info->start_worklist;
 570
 571                 if (succ_worklist == NULL || succ_worklist->visited >= worklist_visited)
 572                         continue;
 573
 574                 best_execfreq   = execfreq;
 575                 best_worklist   = succ_worklist;
 576                 best_succ_block = succ_block;
 577                 best_pos        = get_edge_src_pos(edge);
 578         }
 579
 580         if (best_worklist == NULL)
 581                 return false;
 582         best_worklist->visited = worklist_visited;
 583
 584         fill_and_activate_worklist(new_worklist, best_worklist, block,
 585                         best_succ_block, best_pos);
 586         return true;
 587 }
 588
 589 static worklist_t *construct_start_worklist(ir_node *block)
 590 {
 591         worklist_t *worklist = new_worklist();
 592
 593         ++worklist_visited;
 594
 595         while(fill_start_worklist(worklist, block)) {
 596                 if (worklist->n_live_values >= n_regs)
 597                         break;
 598         }
 599
 600         return worklist;
 601 }
 602
 603 static void process_block(ir_node *block, void *env)
 604 {
 605         block_info_t    *block_info;
 606         worklist_t      *worklist;
 607         int              n_preds;
 608
 609         (void) env;
 610         DB((dbg, LEVEL_1, "Processing %+F\n", block));
 611
 612         worklist   = construct_start_worklist(block);
 613         block_info = get_block_info(block);
 614
 615 #ifdef DEBUG_libfirm
 616         DB((dbg, LEVEL_1, "worklist at end of %+F:", block));
 617         print_worklist(worklist, LEVEL_1);
 618         DB((dbg, LEVEL_1, "\n"));
 619
 620         if (should_have_reached_fixpoint) {
 621                 assert(worklists_equal(block_info->end_worklist, worklist));
 622         }
 623 #endif
 624         block_info->end_worklist = duplicate_worklist(worklist);
 625
 626         do_spilling(block, worklist);
 627         deactivate_worklist(worklist);
 628
 629 #ifdef DEBUG_libfirm
 630         if (should_have_reached_fixpoint) {
 631                 assert(worklists_equal(block_info->start_worklist, worklist));
 632         }
 633 #endif
 634         block_info->start_worklist = worklist;
 635
 636         /* we shouldn't have any live values left at the start block */
 637         n_preds = get_Block_n_cfgpreds(block);
 638         //assert(n_preds != 0 || worklist->n_live_values == 0);
 639 }
 640
 641 typedef struct block_or_loop_t block_or_loop_t;
 642 struct block_or_loop_t {
 643         union {
 644                 ir_node *block;
 645                 ir_loop *loop;
 646         } v;
 647         bool is_loop;
 648 };
 649
 650 static block_or_loop_t *loop_blocks;
 651 static ir_loop         *current_loop;
 652
 653 static void find_blocks(ir_node *block);
 654
 655 static void find_in_loop(ir_loop *loop, ir_node *entry)
 656 {
 657         loop_info_t     *loop_info = get_loop_info(loop);
 658         block_or_loop_t block_or_loop;
 659         loop_edge_t     *edge;
 660
 661         /* simply mark 1 block in the loop to indicate that the loop was already
 662          * processed */
 663         ir_node *some_block = loop_info->entry_edges->block;
 664         if (Block_block_visited(some_block))
 665                 return;
 666
 667         block_or_loop.v.loop  = loop;
 668         block_or_loop.is_loop = true;
 669         ARR_APP1(block_or_loop_t, loop_blocks, block_or_loop);
 670
 671 #ifndef NDEBUG
 672         {
 673                 /* we should be 1 of the entry blocks */
 674                 loop_edge_t *edge  = loop_info->entry_edges;
 675                 bool         found = false;
 676                 for ( ; edge != NULL; edge = edge->next) {
 677                         if (edge->block == entry)
 678                                 found = true;
 679                 }
 680                 assert(found);
 681         }
 682 #else
 683         (void) entry;
 684 #endif
 685
 686         /* check all loop successors */
 687         for (edge = loop_info->exit_edges; edge != NULL; edge = edge->next) {
 688                 ir_node *succ      = edge->block;
 689                 ir_loop *succ_loop = get_irn_loop(succ);
 690
 691                 if (succ_loop == current_loop) {
 692                         find_blocks(succ);
 693                 } else {
 694                         assert(get_loop_depth(succ_loop) < get_loop_depth(current_loop));
 695                 }
 696         }
 697 }
 698
 699 static void find_blocks(ir_node *block)
 700 {
 701         const ir_edge_t *edge;
 702         block_or_loop_t block_or_loop;
 703
 704         if (Block_block_visited(block))
 705                 return;
 706
 707         block_or_loop.v.block = block;
 708         block_or_loop.is_loop = false;
 709
 710         ARR_APP1(block_or_loop_t, loop_blocks, block_or_loop);
 711         mark_Block_block_visited(block);
 712
 713         foreach_block_succ(block, edge) {
 714                 ir_node *succ = get_edge_src_irn(edge);
 715
 716                 /* is it part of the current loop? */
 717                 ir_loop *loop = get_irn_loop(succ);
 718                 if (loop != current_loop) {
 719                         /* a sub-loop? */
 720                         if (get_loop_depth(loop) > get_loop_depth(current_loop)) {
 721                                 find_in_loop(loop, succ);
 722                         } else {
 723                                 /* parent loop: we're not interested in the block */
 724                         }
 725                 } else {
 726                         find_blocks(succ);
 727                 }
 728         }
 729 }
 730
 731 /**
 732  * append an entry to a worklist. WARNING: The entry must not already be in the
 733  * worklist.
 734  */
 735 static void worklist_append(worklist_t *worklist, ir_node *value,
 736                             ir_node *reload_point,
 737                                                         ir_loop *unused_livethrough_loop)
 738 {
 739         worklist_entry_t *entry     = obstack_alloc(&obst, sizeof(entry[0]));
 740         memset(entry, 0, sizeof(entry[0]));
 741
 742 #ifdef EXPENSIVE_CHECKS
 743         {
 744                 struct list_head *entry;
 745                 list_for_each(entry, &worklist->live_values) {
 746                         worklist_entry_t *wl_entry
 747                                 = list_entry(entry, worklist_entry_t, head);
 748                         assert(wl_entry->value != value);
 749                 }
 750         }
 751 #endif
 752
 753         entry->value                   = value;
 754         entry->reload_point            = reload_point;
 755         entry->unused_livethrough_loop = unused_livethrough_loop;
 756         list_add_tail(&entry->head, &worklist->live_values);
 757         ++worklist->n_live_values;
 758         assert(worklist->n_live_values <= n_regs);
 759 }
 760
 761 static void push_unused_livethrough(loop_info_t *loop_info, ir_node *value)
 762 {
 763         loop_edge_t *edge;
 764         ++worklist_visited;
 765
 766         /* add the value to all loop exit and entry blocks */
 767         for (edge = loop_info->exit_edges; edge != NULL; edge = edge->next) {
 768                 ir_node            *block
 769                         = get_Block_cfgpred_block(edge->block, edge->pos);
 770                 const block_info_t *info     = get_block_info(block);
 771                 worklist_t         *worklist = info->end_worklist;
 772                 ir_node            *reload_point = NULL;
 773
 774                 if (worklist->visited >= worklist_visited)
 775                         continue;
 776                 worklist->visited = worklist_visited;
 777
 778                 /* TODO: we need a smarter mechanism here, that makes the reloader place
 779                  * reload nodes on all loop exits... */
 780
 781                 worklist_append(worklist, value, reload_point, loop_info->loop);
 782         }
 783         edge = loop_info->entry_edges;
 784         for ( ; edge != NULL; edge = edge->next) {
 785                 ir_node            *entry_block = edge->block;
 786                 const block_info_t *info        = get_block_info(entry_block);
 787                 worklist_t         *worklist    = info->start_worklist;
 788                 ir_node            *pred_block;
 789                 ir_node            *reload_point;
 790
 791                 if (worklist->visited >= worklist_visited)
 792                         continue;
 793                 worklist->visited = worklist_visited;
 794
 795                 pred_block   = get_Block_cfgpred_block(entry_block, edge->pos);
 796                 reload_point = be_get_end_of_block_insertion_point(pred_block);
 797
 798                 worklist_append(worklist, value, reload_point, loop_info->loop);
 799         }
 800
 801         set_irn_link(value, NULL);
 802         ++loop_info->max_register_pressure;
 803 }
 804
 805 static void push_unused_livethroughs(loop_info_t *loop_info)
 806 {
 807         loop_edge_t *edge;
 808
 809         /* we can only push unused livethroughs if register pressure inside the loop
 810          * was low enough */
 811         if (loop_info->max_register_pressure >= n_regs)
 812                 return;
 813
 814         /* find unused livethroughs: register pressure in the loop was low enough
 815          * which means that we had no spills which implies that at every point in
 816          * the loop all*/
 817         for (edge = loop_info->exit_edges; edge != NULL; edge = edge->next) {
 818                 ir_node            *block          = edge->block;
 819                 const block_info_t *info           = get_block_info(block);
 820                 worklist_t         *start_worklist = info->start_worklist;
 821                 ir_node            *exit_block;
 822                 const block_info_t *exit_info;
 823                 worklist_t         *end_worklist;
 824                 struct list_head   *entry;
 825
 826                 if (start_worklist == NULL)
 827                         continue;
 828
 829                 exit_block   = get_Block_cfgpred_block(edge->block, edge->pos);
 830                 exit_info    = get_block_info(exit_block);
 831                 end_worklist = exit_info->end_worklist;
 832
 833                 activate_worklist(end_worklist);
 834                 /* all values contained in the start_worklist, which are not available
 835                  * in the end_worklist, must be unused livethroughs */
 836
 837                 list_for_each(entry, &start_worklist->live_values) {
 838                         worklist_entry_t *wl_entry
 839                                 = list_entry(entry, worklist_entry_t, head);
 840                         ir_node *value = wl_entry->value;
 841
 842                         if (loop_info->max_register_pressure >= n_regs)
 843                                 break;
 844
 845
 846                         if (!worklist_contains(value)) {
 847                                 /* add it to all loop exits */
 848                                 DB((dbg, LEVEL_2, "Found unused livethrough: %+F (regpressure in loop %d)\n", value, loop_info->max_register_pressure));
 849                                 push_unused_livethrough(loop_info, value);
 850                                 /* the value should now be part of end_worklist, so mark it */
 851                                 mark_irn_visited(value);
 852                         }
 853                 }
 854
 855                 deactivate_worklist(end_worklist);
 856         }
 857 }
 858
 859 static void process_block_or_loop(const block_or_loop_t *block_or_loop)
 860 {
 861         if (block_or_loop->is_loop) {
 862                 loop_info_t *loop_info = get_loop_info(block_or_loop->v.loop);
 863
 864                 if (do_push_unused_livethroughs)
 865                         push_unused_livethroughs(loop_info);
 866
 867                 if (loop_info->max_register_pressure > max_register_pressure)
 868                         max_register_pressure = loop_info->max_register_pressure;
 869
 870                 return;
 871         }
 872         process_block(block_or_loop->v.block, NULL);
 873 }
 874
 875 static void process_loop(ir_loop *loop)
 876 {
 877         int         n_elements = get_loop_n_elements(loop);
 878         int         i, len;
 879         loop_info_t *loop_info;
 880         loop_edge_t *edge;
 881         ir_node     *some_block;
 882
 883         /* first handle all sub-loops */
 884         for (i = 0; i < n_elements; ++i) {
 885                 loop_element element = get_loop_element(loop, i);
 886                 if (*element.kind != k_ir_loop)
 887                         continue;
 888
 889                 process_loop(element.son);
 890         }
 891
 892         /* create a postorder of the blocks */
 893         loop_info = get_loop_info(loop);
 894         edge      = loop_info->entry_edges;
 895         if (edge != NULL) {
 896                 some_block = edge->block;
 897         } else {
 898                 assert(loop == get_irg_loop(current_ir_graph));
 899                 some_block = get_irg_start_block(current_ir_graph);
 900         }
 901
 902         loop_blocks  = NEW_ARR_F(block_or_loop_t,0);
 903         current_loop = loop;
 904
 905         ir_reserve_resources(current_ir_graph, IR_RESOURCE_BLOCK_VISITED);
 906         inc_irg_block_visited(current_ir_graph);
 907         find_blocks(some_block);
 908         /* for endless loops the end-block might be unreachable */
 909         if (loop == get_irg_loop(current_ir_graph)) {
 910                 find_blocks(get_irg_end_block(current_ir_graph));
 911         }
 912         ir_free_resources(current_ir_graph, IR_RESOURCE_BLOCK_VISITED);
 913
 914         DB((dbg, LEVEL_3, "Block List for loop %p\n", loop));
 915         len = ARR_LEN(loop_blocks);
 916         for (i = 0; i < len; ++i) {
 917                 block_or_loop_t *block_or_loop = &loop_blocks[i];
 918                 if (block_or_loop->is_loop) {
 919                         DB((dbg, LEVEL_3, " L-%p", block_or_loop->v.loop));
 920                 } else {
 921                         DB((dbg, LEVEL_3, " B-%+F", block_or_loop->v.block));
 922                 }
 923         }
 924         DB((dbg, LEVEL_3, "\n"));
 925
 926         max_register_pressure = 0;
 927
 928         /* run1: tentative phase */
 929         tentative_mode = true;
 930         for (i = len-1; i >= 0; --i) {
 931                 process_block_or_loop(&loop_blocks[i]);
 932         }
 933
 934         /* run2: tentative phase - should reach fixpoint */
 935         tentative_mode = true;
 936         for (i = len-1; i >= 0; --i) {
 937                 process_block_or_loop(&loop_blocks[i]);
 938         }
 939
 940 #ifndef NDEBUG
 941         /* run3: tentative phase - check fixpoint */
 942         tentative_mode               = true;
 943         should_have_reached_fixpoint = true;
 944         for (i = len-1; i >= 0; --i) {
 945                 process_block_or_loop(&loop_blocks[i]);
 946         }
 947         should_have_reached_fixpoint = false;
 948 #endif
 949
 950         /* run4: add spills/reloads */
 951         tentative_mode              = false;
 952         do_push_unused_livethroughs = true;
 953         for (i = len-1; i >= 0; --i) {
 954                 process_block_or_loop(&loop_blocks[i]);
 955         }
 956         do_push_unused_livethroughs = false;
 957
 958         loop_info->max_register_pressure = max_register_pressure;
 959         DB((dbg, LEVEL_2, "Regpressure in loop %p: %u\n", loop,
 960                                 (unsigned) max_register_pressure));
 961
 962         DEL_ARR_F(loop_blocks);
 963 }
 964
 965 static void fix_block_borders(ir_node *block, void *data)
 966 {
 967         block_info_t *block_info     = get_block_info(block);
 968         worklist_t   *start_worklist = block_info->start_worklist;
 969         int           n_cfgpreds     = get_Block_n_cfgpreds(block);
 970         ir_loop      *loop           = get_irn_loop(block);
 971         int           i;
 972
 973         (void) data;
 974         assert(start_worklist != NULL);
 975
 976         for (i = 0; i < n_cfgpreds; ++i) {
 977                 ir_node      *pred_block      = get_Block_cfgpred_block(block, i);
 978                 block_info_t *pred_block_info = get_block_info(pred_block);
 979                 worklist_t   *end_worklist    = pred_block_info->end_worklist;
 980                 ir_loop      *pred_loop       = get_irn_loop(pred_block);
 981                 bool          is_loop_entry   = false;
 982                 struct list_head *entry;
 983
 984                 assert(end_worklist != NULL);
 985
 986                 if (get_loop_depth(pred_loop) < get_loop_depth(loop)) {
 987                         is_loop_entry = true;
 988                 }
 989
 990                 /* reload missing values */
 991                 activate_worklist(end_worklist);
 992
 993                 list_for_each(entry, &start_worklist->live_values) {
 994                         worklist_entry_t *wl_entry
 995                                 = list_entry(entry, worklist_entry_t, head);
 996                         ir_node          *value = wl_entry->value;
 997
 998                         if (is_Phi(value) && get_nodes_block(value) == block) {
 999                                 value = get_irn_n(value, i);
1000
1001                                 /* we might have unknowns as argument for the phi */
1002                                 if (!arch_irn_consider_in_reg_alloc(cls, value))
1003                                         continue;
1004                         }
1005
1006                         if (worklist_contains(value))
1007                                 continue;
1008                         if (wl_entry->unused_livethrough_loop != NULL && !is_loop_entry)
1009                                 continue;
1010
1011                         be_add_reload_on_edge(senv, value, block, i, cls, 1);
1012                 }
1013
1014                 deactivate_worklist(end_worklist);
1015         }
1016 }
1017
1018 static void be_spill_belady3(be_irg_t *birg, const arch_register_class_t *ncls)
1019 {
1020         ir_graph *irg = be_get_birg_irg(birg);
1021
1022         cls    = ncls;
1023         n_regs = cls->n_regs - be_put_ignore_regs(birg, cls, NULL);
1024
1025         /* shortcut for register classes with ignore regs only */
1026         if (n_regs == 0)
1027                 return;
1028
1029         worklist_visited = 0;
1030         exec_freq        = be_get_birg_exec_freq(birg);
1031
1032         be_clear_links(irg);
1033         ir_reserve_resources(irg, IR_RESOURCE_IRN_VISITED | IR_RESOURCE_IRN_LINK
1034                         | IR_RESOURCE_LOOP_LINK);
1035         inc_irg_visited(irg);
1036
1037         obstack_init(&obst);
1038         senv = be_new_spill_env(birg);
1039
1040         assure_cf_loop(irg);
1041         clear_loop_info(get_irg_loop(irg));
1042         irg_block_walk_graph(irg, construct_loop_edges, NULL, NULL);
1043
1044         process_loop(get_irg_loop(current_ir_graph));
1045
1046         irg_block_walk_graph(irg, fix_block_borders, NULL, NULL);
1047
1048         ir_free_resources(irg, IR_RESOURCE_IRN_VISITED | IR_RESOURCE_IRN_LINK
1049                         | IR_RESOURCE_LOOP_LINK);
1050
1051         be_insert_spills_reloads(senv);
1052
1053         obstack_free(&obst, NULL);
1054
1055         /* clean up */
1056         be_delete_spill_env(senv);
1057 }
1058
1059 void be_init_spillbelady3(void)
1060 {
1061         static be_spiller_t belady3_spiller = {
1062                 be_spill_belady3
1063         };
1064
1065         be_register_spiller("belady3", &belady3_spiller);
1066         FIRM_DBG_REGISTER(dbg, "firm.be.spill.belady3");
1067 }
1068
1069 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_spillbelady3);