gcc/postreload-gcse.c

   1 /* Post reload partially redundant load elimination
   2    Copyright (C) 2004, 2005
   3    Free Software Foundation, Inc.
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 2, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING.  If not, write to the Free
  19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
  20 02111-1307, USA.  */
  21
  22 #include "config.h"
  23 #include "system.h"
  24 #include "coretypes.h"
  25 #include "tm.h"
  26 #include "toplev.h"
  27
  28 #include "rtl.h"
  29 #include "tree.h"
  30 #include "tm_p.h"
  31 #include "regs.h"
  32 #include "hard-reg-set.h"
  33 #include "flags.h"
  34 #include "real.h"
  35 #include "insn-config.h"
  36 #include "recog.h"
  37 #include "basic-block.h"
  38 #include "output.h"
  39 #include "function.h"
  40 #include "expr.h"
  41 #include "except.h"
  42 #include "intl.h"
  43 #include "obstack.h"
  44 #include "hashtab.h"
  45 #include "params.h"
  46
  47 /* The following code implements gcse after reload, the purpose of this
  48    pass is to cleanup redundant loads generated by reload and other
  49    optimizations that come after gcse. It searches for simple inter-block
  50    redundancies and tries to eliminate them by adding moves and loads
  51    in cold places.
  52
  53    Perform partially redundant load elimination, try to eliminate redundant
  54    loads created by the reload pass.  We try to look for full or partial
  55    redundant loads fed by one or more loads/stores in predecessor BBs,
  56    and try adding loads to make them fully redundant.  We also check if
  57    it's worth adding loads to be able to delete the redundant load.
  58
  59    Algorithm:
  60    1. Build available expressions hash table:
  61        For each load/store instruction, if the loaded/stored memory didn't
  62        change until the end of the basic block add this memory expression to
  63        the hash table.
  64    2. Perform Redundancy elimination:
  65       For each load instruction do the following:
  66          perform partial redundancy elimination, check if it's worth adding
  67          loads to make the load fully redundant.  If so add loads and
  68          register copies and delete the load.
  69    3. Delete instructions made redundant in step 2.
  70
  71    Future enhancement:
  72      If the loaded register is used/defined between load and some store,
  73      look for some other free register between load and all its stores,
  74      and replace the load with a copy from this register to the loaded
  75      register.
  76 */
  77 \f
  78
  79 /* Keep statistics of this pass.  */
  80 static struct
  81 {
  82   int moves_inserted;
  83   int copies_inserted;
  84   int insns_deleted;
  85 } stats;
  86
  87 /* We need to keep a hash table of expressions.  The table entries are of
  88    type 'struct expr', and for each expression there is a single linked
  89    list of occurrences.  */
  90
  91 /* The table itself.  */
  92 static htab_t expr_table;
  93
  94 /* Expression elements in the hash table.  */
  95 struct expr
  96 {
  97   /* The expression (SET_SRC for expressions, PATTERN for assignments).  */
  98   rtx expr;
  99
 100   /* The same hash for this entry.  */
 101   hashval_t hash;
 102
 103   /* List of available occurrence in basic blocks in the function.  */
 104   struct occr *avail_occr;
 105 };
 106
 107 static struct obstack expr_obstack;
 108
 109 /* Occurrence of an expression.
 110    There is at most one occurrence per basic block.  If a pattern appears
 111    more than once, the last appearance is used.  */
 112
 113 struct occr
 114 {
 115   /* Next occurrence of this expression.  */
 116   struct occr *next;
 117   /* The insn that computes the expression.  */
 118   rtx insn;
 119   /* Nonzero if this [anticipatable] occurrence has been deleted.  */
 120   char deleted_p;
 121 };
 122
 123 static struct obstack occr_obstack;
 124
 125 /* The following structure holds the information about the occurrences of
 126    the redundant instructions.  */
 127 struct unoccr
 128 {
 129   struct unoccr *next;
 130   edge pred;
 131   rtx insn;
 132 };
 133
 134 static struct obstack unoccr_obstack;
 135
 136 /* Array where each element is the CUID if the insn that last set the hard
 137    register with the number of the element, since the start of the current
 138    basic block.
 139
 140    This array is used during the building of the hash table (step 1) to
 141    determine if a reg is killed before the end of a basic block.
 142
 143    It is also used when eliminating partial redundancies (step 2) to see
 144    if a reg was modified since the start of a basic block.  */
 145 static int *reg_avail_info;
 146
 147 /* A list of insns that may modify memory within the current basic block.  */
 148 struct modifies_mem
 149 {
 150   rtx insn;
 151   struct modifies_mem *next;
 152 };
 153 static struct modifies_mem *modifies_mem_list;
 154
 155 /* The modifies_mem structs also go on an obstack, only this obstack is
 156    freed each time after completing the analysis or transformations on
 157    a basic block.  So we allocate a dummy modifies_mem_obstack_bottom
 158    object on the obstack to keep track of the bottom of the obstack.  */
 159 static struct obstack modifies_mem_obstack;
 160 static struct modifies_mem  *modifies_mem_obstack_bottom;
 161
 162 /* Mapping of insn UIDs to CUIDs.
 163    CUIDs are like UIDs except they increase monotonically in each basic
 164    block, have no gaps, and only apply to real insns.  */
 165 static int *uid_cuid;
 166 #define INSN_CUID(INSN) (uid_cuid[INSN_UID (INSN)])
 167 \f
 168
 169 /* Helpers for memory allocation/freeing.  */
 170 static void alloc_mem (void);
 171 static void free_mem (void);
 172
 173 /* Support for hash table construction and transformations.  */
 174 static bool oprs_unchanged_p (rtx, rtx, bool);
 175 static void record_last_reg_set_info (rtx, int);
 176 static void record_last_mem_set_info (rtx);
 177 static void record_last_set_info (rtx, rtx, void *);
 178 static void record_opr_changes (rtx);
 179
 180 static void find_mem_conflicts (rtx, rtx, void *);
 181 static int load_killed_in_block_p (int, rtx, bool);
 182 static void reset_opr_set_tables (void);
 183
 184 /* Hash table support.  */
 185 static hashval_t hash_expr (rtx, int *);
 186 static hashval_t hash_expr_for_htab (const void *);
 187 static int expr_equiv_p (const void *, const void *);
 188 static void insert_expr_in_table (rtx, rtx);
 189 static struct expr *lookup_expr_in_table (rtx);
 190 static int dump_hash_table_entry (void **, void *);
 191 static void dump_hash_table (FILE *);
 192
 193 /* Helpers for eliminate_partially_redundant_load.  */
 194 static bool reg_killed_on_edge (rtx, edge);
 195 static bool reg_used_on_edge (rtx, edge);
 196
 197 static rtx reg_set_between_after_reload_p (rtx, rtx, rtx);
 198 static rtx reg_used_between_after_reload_p (rtx, rtx, rtx);
 199 static rtx get_avail_load_store_reg (rtx);
 200
 201 static bool bb_has_well_behaved_predecessors (basic_block);
 202 static struct occr* get_bb_avail_insn (basic_block, struct occr *);
 203 static void hash_scan_set (rtx);
 204 static void compute_hash_table (void);
 205
 206 /* The work horses of this pass.  */
 207 static void eliminate_partially_redundant_load (basic_block,
 208                                                 rtx,
 209                                                 struct expr *);
 210 static void eliminate_partially_redundant_loads (void);
 211 \f
 212
 213 /* Allocate memory for the CUID mapping array and register/memory
 214    tracking tables.  */
 215
 216 static void
 217 alloc_mem (void)
 218 {
 219   int i;
 220   basic_block bb;
 221   rtx insn;
 222
 223   /* Find the largest UID and create a mapping from UIDs to CUIDs.  */
 224   uid_cuid = xcalloc (get_max_uid () + 1, sizeof (int));
 225   i = 0;
 226   FOR_EACH_BB (bb)
 227     FOR_BB_INSNS (bb, insn)
 228       {
 229         if (INSN_P (insn))
 230           uid_cuid[INSN_UID (insn)] = i++;
 231         else
 232           uid_cuid[INSN_UID (insn)] = i;
 233       }
 234
 235   /* Allocate the available expressions hash table.  We don't want to
 236      make the hash table too small, but unnecessarily making it too large
 237      also doesn't help.  The i/4 is a gcse.c relic, and seems like a
 238      reasonable choice.  */
 239   expr_table = htab_create (MAX (i / 4, 13),
 240                             hash_expr_for_htab, expr_equiv_p, NULL);
 241
 242   /* We allocate everything on obstacks because we often can roll back
 243      the whole obstack to some point.  Freeing obstacks is very fast.  */
 244   gcc_obstack_init (&expr_obstack);
 245   gcc_obstack_init (&occr_obstack);
 246   gcc_obstack_init (&unoccr_obstack);
 247   gcc_obstack_init (&modifies_mem_obstack);
 248
 249   /* Working array used to track the last set for each register
 250      in the current block.  */
 251   reg_avail_info = (int *) xmalloc (FIRST_PSEUDO_REGISTER * sizeof (int));
 252
 253   /* Put a dummy modifies_mem object on the modifies_mem_obstack, so we
 254      can roll it back in reset_opr_set_tables.  */
 255   modifies_mem_obstack_bottom =
 256     (struct modifies_mem *) obstack_alloc (&modifies_mem_obstack,
 257                                            sizeof (struct modifies_mem));
 258 }
 259
 260 /* Free memory allocated by alloc_mem.  */
 261
 262 static void
 263 free_mem (void)
 264 {
 265   free (uid_cuid);
 266
 267   htab_delete (expr_table);
 268
 269   obstack_free (&expr_obstack, NULL);
 270   obstack_free (&occr_obstack, NULL);
 271   obstack_free (&unoccr_obstack, NULL);
 272   obstack_free (&modifies_mem_obstack, NULL);
 273
 274   free (reg_avail_info);
 275 }
 276 \f
 277
 278 /* Hash expression X.
 279    DO_NOT_RECORD_P is a boolean indicating if a volatile operand is found
 280    or if the expression contains something we don't want to insert in the
 281    table.  */
 282
 283 static hashval_t
 284 hash_expr (rtx x, int *do_not_record_p)
 285 {
 286   *do_not_record_p = 0;
 287   return hash_rtx (x, GET_MODE (x), do_not_record_p,
 288                    NULL,  /*have_reg_qty=*/false);
 289 }
 290
 291 /* Callback for hashtab.
 292    Return the hash value for expression EXP.  We don't actually hash
 293    here, we just return the cached hash value.  */
 294
 295 static hashval_t
 296 hash_expr_for_htab (const void *expp)
 297 {
 298   struct expr *exp = (struct expr *) expp;
 299   return exp->hash;
 300 }
 301
 302 /* Callbach for hashtab.
 303    Return nonzero if exp1 is equivalent to exp2.  */
 304
 305 static int
 306 expr_equiv_p (const void *exp1p, const void *exp2p)
 307 {
 308   struct expr *exp1 = (struct expr *) exp1p;
 309   struct expr *exp2 = (struct expr *) exp2p;
 310   int equiv_p = exp_equiv_p (exp1->expr, exp2->expr, 0, true);
 311   if (equiv_p
 312       && exp1->hash != exp2->hash)
 313     abort ();
 314   return equiv_p;
 315 }
 316 \f
 317
 318 /* Insert expression X in INSN in the hash TABLE.
 319    If it is already present, record it as the last occurrence in INSN's
 320    basic block.  */
 321
 322 static void
 323 insert_expr_in_table (rtx x, rtx insn)
 324 {
 325   int do_not_record_p;
 326   hashval_t hash;
 327   struct expr *cur_expr, **slot;
 328   struct occr *avail_occr, *last_occr = NULL;
 329
 330   hash = hash_expr (x, &do_not_record_p);
 331
 332   /* Do not insert expression in the table if it contains volatile operands,
 333      or if hash_expr determines the expression is something we don't want
 334      to or can't handle.  */
 335   if (do_not_record_p)
 336     return;
 337
 338   /* We anticipate that redundant expressions are rare, so for convenience
 339      allocate a new hash table element here already and set its fields.
 340      If we don't do this, we need a hack with a static struct expr.  Anyway,
 341      obstack_free is really fast and one more obstack_alloc doesn't hurt if
 342      we're going to see more expressions later on.  */
 343   cur_expr = (struct expr *) obstack_alloc (&expr_obstack,
 344                                             sizeof (struct expr));
 345   cur_expr->expr = x;
 346   cur_expr->hash = hash;
 347   cur_expr->avail_occr = NULL;
 348
 349   slot = (struct expr **) htab_find_slot_with_hash (expr_table, cur_expr,
 350                                                     hash, INSERT);
 351
 352   if (! (*slot))
 353     /* The expression isn't found, so insert it.  */
 354     *slot = cur_expr;
 355   else
 356     {
 357       /* The expression is already in the table, so roll back the
 358          obstack and use the existing table entry.  */
 359       obstack_free (&expr_obstack, cur_expr);
 360       cur_expr = *slot;
 361     }
 362
 363   /* Search for another occurrence in the same basic block.  */
 364   avail_occr = cur_expr->avail_occr;
 365   while (avail_occr && BLOCK_NUM (avail_occr->insn) != BLOCK_NUM (insn))
 366     {
 367       /* If an occurrence isn't found, save a pointer to the end of
 368          the list.  */
 369       last_occr = avail_occr;
 370       avail_occr = avail_occr->next;
 371     }
 372
 373   if (avail_occr)
 374     /* Found another instance of the expression in the same basic block.
 375        Prefer this occurrence to the currently recorded one.  We want
 376        the last one in the block and the block is scanned from start
 377        to end.  */
 378     avail_occr->insn = insn;
 379   else
 380     {
 381       /* First occurrence of this expression in this basic block.  */
 382       avail_occr = (struct occr *) obstack_alloc (&occr_obstack,
 383                                                   sizeof (struct occr));
 384
 385       /* First occurrence of this expression in any block?  */
 386       if (cur_expr->avail_occr == NULL)
 387         cur_expr->avail_occr = avail_occr;
 388       else
 389         last_occr->next = avail_occr;
 390
 391       avail_occr->insn = insn;
 392       avail_occr->next = NULL;
 393       avail_occr->deleted_p = 0;
 394     }
 395 }
 396 \f
 397
 398 /* Lookup pattern PAT in the expression hash table.
 399    The result is a pointer to the table entry, or NULL if not found.  */
 400
 401 static struct expr *
 402 lookup_expr_in_table (rtx pat)
 403 {
 404   int do_not_record_p;
 405   struct expr **slot, *tmp_expr;
 406   hashval_t hash = hash_expr (pat, &do_not_record_p);
 407
 408   if (do_not_record_p)
 409     return NULL;
 410
 411   tmp_expr = (struct expr *) obstack_alloc (&expr_obstack,
 412                                             sizeof (struct expr));
 413   tmp_expr->expr = pat;
 414   tmp_expr->hash = hash;
 415   tmp_expr->avail_occr = NULL;
 416
 417   slot = (struct expr **) htab_find_slot_with_hash (expr_table, tmp_expr,
 418                                                     hash, INSERT);
 419   obstack_free (&expr_obstack, tmp_expr);
 420
 421   if (!slot)
 422     return NULL;
 423   else
 424     return (*slot);
 425 }
 426 \f
 427
 428 /* Dump all expressions and occurrences that are currently in the
 429    expression hash table to FILE.  */
 430
 431 /* This helper is called via htab_traverse.  */
 432 static int
 433 dump_hash_table_entry (void **slot, void *filep)
 434 {
 435   struct expr *expr = (struct expr *) *slot;
 436   FILE *file = (FILE *) filep;
 437   struct occr *occr;
 438
 439   fprintf (file, "expr: ");
 440   print_rtl (file, expr->expr);
 441   fprintf (file,"\nhashcode: %u\n", expr->hash);
 442   fprintf (file,"list of occurences:\n");
 443   occr = expr->avail_occr;
 444   while (occr)
 445     {
 446       rtx insn = occr->insn;
 447       print_rtl_single (file, insn);
 448       fprintf (file, "\n");
 449       occr = occr->next;
 450     }
 451   fprintf (file, "\n");
 452   return 1;
 453 }
 454
 455 static void
 456 dump_hash_table (FILE *file)
 457 {
 458   fprintf (file, "\n\nexpression hash table\n");
 459   fprintf (file, "size %ld, %ld elements, %f collision/search ratio\n",
 460            (long) htab_size (expr_table),
 461            (long) htab_elements (expr_table),
 462            htab_collisions (expr_table));
 463   if (htab_elements (expr_table) > 0)
 464     {
 465       fprintf (file, "\n\ntable entries:\n");
 466       htab_traverse (expr_table, dump_hash_table_entry, file);
 467     }
 468   fprintf (file, "\n");
 469 }
 470 \f
 471
 472 /* Return nonzero if the operands of expression X are unchanged
 473    1) from the start of INSN's basic block up to but not including INSN
 474       if AFTER_INSN is false, or
 475    2) from INSN to the end of INSN's basic block if AFTER_INSN is true.  */
 476
 477 static bool
 478 oprs_unchanged_p (rtx x, rtx insn, bool after_insn)
 479 {
 480   int i, j;
 481   enum rtx_code code;
 482   const char *fmt;
 483
 484   if (x == 0)
 485     return 1;
 486
 487   code = GET_CODE (x);
 488   switch (code)
 489     {
 490     case REG:
 491 #ifdef ENABLE_CHECKING
 492       /* We are called after register allocation.  */
 493       if (REGNO (x) >= FIRST_PSEUDO_REGISTER)
 494         abort ();
 495 #endif
 496       if (after_insn)
 497         /* If the last CUID setting the insn is less than the CUID of
 498            INSN, then reg X is not changed in or after INSN.  */
 499         return reg_avail_info[REGNO (x)] < INSN_CUID (insn);
 500       else
 501         /* Reg X is not set before INSN in the current basic block if
 502            we have not yet recorded the CUID of an insn that touches
 503            the reg.  */
 504         return reg_avail_info[REGNO (x)] == 0;
 505
 506     case MEM:
 507       if (load_killed_in_block_p (INSN_CUID (insn), x, after_insn))
 508         return 0;
 509       else
 510         return oprs_unchanged_p (XEXP (x, 0), insn, after_insn);
 511
 512     case PC:
 513     case CC0: /*FIXME*/
 514     case CONST:
 515     case CONST_INT:
 516     case CONST_DOUBLE:
 517     case CONST_VECTOR:
 518     case SYMBOL_REF:
 519     case LABEL_REF:
 520     case ADDR_VEC:
 521     case ADDR_DIFF_VEC:
 522       return 1;
 523
 524     case PRE_DEC:
 525     case PRE_INC:
 526     case POST_DEC:
 527     case POST_INC:
 528     case PRE_MODIFY:
 529     case POST_MODIFY:
 530       if (after_insn)
 531         return 0;
 532       break;
 533
 534     default:
 535       break;
 536     }
 537
 538   for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
 539     {
 540       if (fmt[i] == 'e')
 541         {
 542           if (! oprs_unchanged_p (XEXP (x, i), insn, after_insn))
 543             return 0;
 544         }
 545       else if (fmt[i] == 'E')
 546         for (j = 0; j < XVECLEN (x, i); j++)
 547           if (! oprs_unchanged_p (XVECEXP (x, i, j), insn, after_insn))
 548             return 0;
 549     }
 550
 551   return 1;
 552 }
 553 \f
 554
 555 /* Used for communication between find_mem_conflicts and
 556    load_killed_in_block_p.  Nonzero if find_mem_conflicts finds a
 557    conflict between two memory references.
 558    This is a bit of a hack to work around the limitations of note_stores.  */
 559 static int mems_conflict_p;
 560
 561 /* DEST is the output of an instruction.  If it is a memory reference, and
 562    possibly conflicts with the load found in DATA, then set mems_conflict_p
 563    to a nonzero value.  */
 564
 565 static void
 566 find_mem_conflicts (rtx dest, rtx setter ATTRIBUTE_UNUSED,
 567                     void *data)
 568 {
 569   rtx mem_op = (rtx) data;
 570
 571   while (GET_CODE (dest) == SUBREG
 572          || GET_CODE (dest) == ZERO_EXTRACT
 573          || GET_CODE (dest) == STRICT_LOW_PART)
 574     dest = XEXP (dest, 0);
 575
 576   /* If DEST is not a MEM, then it will not conflict with the load.  Note
 577      that function calls are assumed to clobber memory, but are handled
 578      elsewhere.  */
 579   if (! MEM_P (dest))
 580     return;
 581
 582   if (true_dependence (dest, GET_MODE (dest), mem_op,
 583                        rtx_addr_varies_p))
 584     mems_conflict_p = 1;
 585 }
 586 \f
 587
 588 /* Return nonzero if the expression in X (a memory reference) is killed
 589    in the current basic block before (if AFTER_INSN is false) or after
 590    (if AFTER_INSN is true) the insn with the CUID in UID_LIMIT.
 591
 592    This function assumes that the modifies_mem table is flushed when
 593    the hash table construction or redundancy elimination phases start
 594    processing a new basic block.  */
 595
 596 static int
 597 load_killed_in_block_p (int uid_limit, rtx x, bool after_insn)
 598 {
 599   struct modifies_mem *list_entry = modifies_mem_list;
 600
 601   while (list_entry)
 602     {
 603       rtx setter = list_entry->insn;
 604
 605       /* Ignore entries in the list that do not apply.  */
 606       if ((after_insn
 607            && INSN_CUID (setter) < uid_limit)
 608           || (! after_insn
 609               && INSN_CUID (setter) > uid_limit))
 610         {
 611           list_entry = list_entry->next;
 612           continue;
 613         }
 614
 615       /* If SETTER is a call everything is clobbered.  Note that calls
 616          to pure functions are never put on the list, so we need not
 617          worry about them.  */
 618       if (CALL_P (setter))
 619         return 1;
 620
 621       /* SETTER must be an insn of some kind that sets memory.  Call
 622          note_stores to examine each hunk of memory that is modified.
 623          It will set mems_conflict_p to nonzero if there may be a
 624          conflict between X and SETTER.  */
 625       mems_conflict_p = 0;
 626       note_stores (PATTERN (setter), find_mem_conflicts, x);
 627       if (mems_conflict_p)
 628         return 1;
 629
 630       list_entry = list_entry->next;
 631     }
 632   return 0;
 633 }
 634 \f
 635
 636 /* Record register first/last/block set information for REGNO in INSN.  */
 637
 638 static inline void
 639 record_last_reg_set_info (rtx insn, int regno)
 640 {
 641   reg_avail_info[regno] = INSN_CUID (insn);
 642 }
 643
 644
 645 /* Record memory modification information for INSN.  We do not actually care
 646    about the memory location(s) that are set, or even how they are set (consider
 647    a CALL_INSN).  We merely need to record which insns modify memory.  */
 648
 649 static void
 650 record_last_mem_set_info (rtx insn)
 651 {
 652   struct modifies_mem *list_entry;
 653
 654   list_entry = (struct modifies_mem *) obstack_alloc (&modifies_mem_obstack,
 655                                                       sizeof (struct modifies_mem));
 656   list_entry->insn = insn;
 657   list_entry->next = modifies_mem_list;
 658   modifies_mem_list = list_entry;
 659 }
 660
 661 /* Called from compute_hash_table via note_stores to handle one
 662    SET or CLOBBER in an insn.  DATA is really the instruction in which
 663    the SET is taking place.  */
 664
 665 static void
 666 record_last_set_info (rtx dest, rtx setter ATTRIBUTE_UNUSED, void *data)
 667 {
 668   rtx last_set_insn = (rtx) data;
 669
 670   if (GET_CODE (dest) == SUBREG)
 671     dest = SUBREG_REG (dest);
 672
 673   if (REG_P (dest))
 674     record_last_reg_set_info (last_set_insn, REGNO (dest));
 675   else if (MEM_P (dest)
 676            /* Ignore pushes, they clobber nothing.  */
 677            && ! push_operand (dest, GET_MODE (dest)))
 678     record_last_mem_set_info (last_set_insn);
 679 }
 680
 681
 682 /* Reset tables used to keep track of what's still available since the
 683    start of the block.  */
 684
 685 static void
 686 reset_opr_set_tables (void)
 687 {
 688   memset (reg_avail_info, 0, FIRST_PSEUDO_REGISTER * sizeof (int));
 689   obstack_free (&modifies_mem_obstack, modifies_mem_obstack_bottom);
 690   modifies_mem_list = NULL;
 691 }
 692 \f
 693
 694 /* Record things set by INSN.
 695    This data is used by oprs_unchanged_p.  */
 696
 697 static void
 698 record_opr_changes (rtx insn)
 699 {
 700   rtx note;
 701
 702   /* Find all stores and record them.  */
 703   note_stores (PATTERN (insn), record_last_set_info, insn);
 704
 705   /* Also record autoincremented REGs for this insn as changed.  */
 706   for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
 707     if (REG_NOTE_KIND (note) == REG_INC)
 708       record_last_reg_set_info (insn, REGNO (XEXP (note, 0)));
 709
 710   /* Finally, if this is a call, record all call clobbers.  */
 711   if (CALL_P (insn))
 712     {
 713       unsigned int regno;
 714
 715       for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
 716         if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
 717           record_last_reg_set_info (insn, regno);
 718
 719       if (! CONST_OR_PURE_CALL_P (insn))
 720         record_last_mem_set_info (insn);
 721     }
 722 }
 723 \f
 724
 725 /* Scan the pattern of INSN and add an entry to the hash TABLE.
 726    After reload we are interested in loads/stores only.  */
 727
 728 static void
 729 hash_scan_set (rtx insn)
 730 {
 731   rtx pat = PATTERN (insn);
 732   rtx src = SET_SRC (pat);
 733   rtx dest = SET_DEST (pat);
 734
 735   /* We are only interested in loads and stores.  */
 736   if (! MEM_P (src) && ! MEM_P (dest))
 737     return;
 738
 739   /* Don't mess with jumps and nops.  */
 740   if (JUMP_P (insn) || set_noop_p (pat))
 741     return;
 742
 743 #ifdef ENABLE_CHEKCING
 744   /* We shouldn't have any EH_REGION notes post reload.  */
 745   if (find_reg_note (insn, REG_EH_REGION, NULL_RTX))
 746     abort ();
 747 #endif
 748
 749   if (REG_P (dest))
 750     {
 751       if (/* Don't CSE something if we can't do a reg/reg copy.  */
 752           can_copy_p (GET_MODE (dest))
 753           /* Is SET_SRC something we want to gcse?  */
 754           && general_operand (src, GET_MODE (src))
 755           /* An expression is not available if its operands are
 756              subsequently modified, including this insn.  */
 757           && oprs_unchanged_p (src, insn, true))
 758         {
 759           insert_expr_in_table (src, insn);
 760         }
 761     }
 762   else if (REG_P (src))
 763     {
 764       /* Only record sets of pseudo-regs in the hash table.  */
 765       if (/* Don't CSE something if we can't do a reg/reg copy.  */
 766           can_copy_p (GET_MODE (src))
 767           /* Is SET_DEST something we want to gcse?  */
 768           && general_operand (dest, GET_MODE (dest))
 769           && ! (flag_float_store && FLOAT_MODE_P (GET_MODE (dest)))
 770           /* Check if the memory expression is killed after insn.  */
 771           && ! load_killed_in_block_p (INSN_CUID (insn) + 1, dest, true)
 772           && oprs_unchanged_p (XEXP (dest, 0), insn, true))
 773         {
 774           insert_expr_in_table (dest, insn);
 775         }
 776     }
 777 }
 778 \f
 779
 780 /* Create hash table of memory expressions available at end of basic
 781    blocks.  Basically you should think of this hash table as the
 782    representation of AVAIL_OUT.  This is the set of expressions that
 783    is generated in a basic block and not killed before the end of the
 784    same basic block.  Notice that this is really a local computation.  */
 785
 786 static void
 787 compute_hash_table (void)
 788 {
 789   basic_block bb;
 790
 791   FOR_EACH_BB (bb)
 792     {
 793       rtx insn;
 794
 795       /* First pass over the instructions records information used to
 796          determine when registers and memory are last set.
 797          Since we compute a "local" AVAIL_OUT, reset the tables that
 798          help us keep track of what has been modified since the start
 799          of the block.  */
 800       reset_opr_set_tables ();
 801       FOR_BB_INSNS (bb, insn)
 802         {
 803           if (INSN_P (insn))
 804             record_opr_changes (insn);
 805         }
 806
 807       /* The next pass actually builds the hash table.  */
 808       FOR_BB_INSNS (bb, insn)
 809         if (INSN_P (insn) && GET_CODE (PATTERN (insn)) == SET)
 810           hash_scan_set (insn);
 811     }
 812 }
 813 \f
 814
 815 /* Check if register REG is killed in any insn waiting to be inserted on
 816    edge E.  This function is required to check that our data flow analysis
 817    is still valid prior to commit_edge_insertions.  */
 818
 819 static bool
 820 reg_killed_on_edge (rtx reg, edge e)
 821 {
 822   rtx insn;
 823
 824   for (insn = e->insns.r; insn; insn = NEXT_INSN (insn))
 825     if (INSN_P (insn) && reg_set_p (reg, insn))
 826       return true;
 827
 828   return false;
 829 }
 830
 831 /* Similar to above - check if register REG is used in any insn waiting
 832    to be inserted on edge E.
 833    Assumes no such insn can be a CALL_INSN; if so call reg_used_between_p
 834    with PREV(insn),NEXT(insn) instead of calling reg_overlap_mentioned_p.  */
 835
 836 static bool
 837 reg_used_on_edge (rtx reg, edge e)
 838 {
 839   rtx insn;
 840
 841   for (insn = e->insns.r; insn; insn = NEXT_INSN (insn))
 842     if (INSN_P (insn) && reg_overlap_mentioned_p (reg, PATTERN (insn)))
 843       return true;
 844
 845   return false;
 846 }
 847 \f
 848
 849 /* Return the insn that sets register REG or clobbers it in between
 850    FROM_INSN and TO_INSN (exclusive of those two).
 851    Just like reg_set_between but for hard registers and not pseudos.  */
 852
 853 static rtx
 854 reg_set_between_after_reload_p (rtx reg, rtx from_insn, rtx to_insn)
 855 {
 856   rtx insn;
 857
 858 #ifdef ENABLE_CHECKING
 859   /* We are called after register allocation.  */
 860   if (!REG_P (reg) || REGNO (reg) >= FIRST_PSEUDO_REGISTER)
 861     abort ();
 862 #endif
 863
 864   if (from_insn == to_insn)
 865     return NULL_RTX;
 866
 867   for (insn = NEXT_INSN (from_insn);
 868        insn != to_insn;
 869        insn = NEXT_INSN (insn))
 870     if (INSN_P (insn))
 871       {
 872         if (set_of (reg, insn) != NULL_RTX)
 873           return insn;
 874         if ((CALL_P (insn)
 875               && call_used_regs[REGNO (reg)])
 876             || find_reg_fusage (insn, CLOBBER, reg))
 877           return insn;
 878
 879         if (FIND_REG_INC_NOTE (insn, reg))
 880           return insn;
 881       }
 882
 883   return NULL_RTX;
 884 }
 885
 886 /* Return the insn that uses register REG in between FROM_INSN and TO_INSN
 887    (exclusive of those two). Similar to reg_used_between but for hard
 888    registers and not pseudos.  */
 889
 890 static rtx
 891 reg_used_between_after_reload_p (rtx reg, rtx from_insn, rtx to_insn)
 892 {
 893   rtx insn;
 894
 895 #ifdef ENABLE_CHECKING
 896   /* We are called after register allocation.  */
 897   if (!REG_P (reg) || REGNO (reg) >= FIRST_PSEUDO_REGISTER)
 898     abort ();
 899 #endif
 900
 901   if (from_insn == to_insn)
 902     return NULL_RTX;
 903
 904   for (insn = NEXT_INSN (from_insn);
 905        insn != to_insn;
 906        insn = NEXT_INSN (insn))
 907     if (INSN_P (insn))
 908       {
 909         if (reg_overlap_mentioned_p (reg, PATTERN (insn))
 910             || (CALL_P (insn)
 911                 && call_used_regs[REGNO (reg)])
 912             || find_reg_fusage (insn, USE, reg)
 913             || find_reg_fusage (insn, CLOBBER, reg))
 914           return insn;
 915
 916         if (FIND_REG_INC_NOTE (insn, reg))
 917           return insn;
 918       }
 919
 920   return NULL_RTX;
 921 }
 922
 923 /* Return true if REG is used, set, or killed between the beginning of
 924    basic block BB and UP_TO_INSN.  Caches the result in reg_avail_info.  */
 925
 926 static bool
 927 reg_set_or_used_since_bb_start (rtx reg, basic_block bb, rtx up_to_insn)
 928 {
 929   rtx insn, start = PREV_INSN (BB_HEAD (bb));
 930
 931   if (reg_avail_info[REGNO (reg)] != 0)
 932     return true;
 933
 934   insn = reg_used_between_after_reload_p (reg, start, up_to_insn);
 935   if (! insn)
 936     insn = reg_set_between_after_reload_p (reg, start, up_to_insn);
 937
 938   if (insn)
 939     reg_avail_info[REGNO (reg)] = INSN_CUID (insn);
 940
 941   return insn != NULL_RTX;
 942 }
 943
 944 /* Return the loaded/stored register of a load/store instruction.  */
 945
 946 static rtx
 947 get_avail_load_store_reg (rtx insn)
 948 {
 949   if (REG_P (SET_DEST (PATTERN (insn))))  /* A load.  */
 950     return SET_DEST(PATTERN(insn));
 951   if (REG_P (SET_SRC (PATTERN (insn))))  /* A store.  */
 952     return SET_SRC (PATTERN (insn));
 953   abort ();
 954 }
 955
 956 /* Return nonzero if the predecessors of BB are "well behaved".  */
 957
 958 static bool
 959 bb_has_well_behaved_predecessors (basic_block bb)
 960 {
 961   edge pred;
 962   edge_iterator ei;
 963
 964   if (EDGE_COUNT (bb->preds) == 0)
 965     return false;
 966
 967   FOR_EACH_EDGE (pred, ei, bb->preds)
 968     {
 969       if ((pred->flags & EDGE_ABNORMAL) && EDGE_CRITICAL_P (pred))
 970         return false;
 971
 972       if (JUMP_TABLE_DATA_P (BB_END (pred->src)))
 973         return false;
 974     }
 975   return true;
 976 }
 977
 978
 979 /* Search for the occurrences of expression in BB.  */
 980
 981 static struct occr*
 982 get_bb_avail_insn (basic_block bb, struct occr *occr)
 983 {
 984   for (; occr != NULL; occr = occr->next)
 985     if (BLOCK_FOR_INSN (occr->insn) == bb)
 986       return occr;
 987   return NULL;
 988 }
 989
 990
 991 /* This handles the case where several stores feed a partially redundant
 992    load. It checks if the redundancy elimination is possible and if it's
 993    worth it.
 994
 995    Redundancy elimination is possible if,
 996    1) None of the operands of an insn have been modified since the start
 997       of the current basic block.
 998    2) In any predecessor of the current basic block, the same expression
 999       is generated.
1000
1001    See the function body for the heuristics that determine if eliminating
1002    a redundancy is also worth doing, assuming it is possible.  */
1003
1004 static void
1005 eliminate_partially_redundant_load (basic_block bb, rtx insn,
1006                                     struct expr *expr)
1007 {
1008   edge pred;
1009   rtx avail_insn = NULL_RTX;
1010   rtx avail_reg;
1011   rtx dest, pat;
1012   struct occr *a_occr;
1013   struct unoccr *occr, *avail_occrs = NULL;
1014   struct unoccr *unoccr, *unavail_occrs = NULL, *rollback_unoccr = NULL;
1015   int npred_ok = 0;
1016   gcov_type ok_count = 0; /* Redundant load execution count.  */
1017   gcov_type critical_count = 0; /* Execution count of critical edges.  */
1018   edge_iterator ei;
1019
1020   /* The execution count of the loads to be added to make the
1021      load fully redundant.  */
1022   gcov_type not_ok_count = 0;
1023   basic_block pred_bb;
1024
1025   pat = PATTERN (insn);
1026   dest = SET_DEST (pat);
1027
1028   /* Check that the loaded register is not used, set, or killed from the
1029      beginning of the block.  */
1030   if (reg_set_or_used_since_bb_start (dest, bb, insn))
1031     return;
1032
1033   /* Check potential for replacing load with copy for predecessors.  */
1034   FOR_EACH_EDGE (pred, ei, bb->preds)
1035     {
1036       rtx next_pred_bb_end;
1037
1038       avail_insn = NULL_RTX;
1039       pred_bb = pred->src;
1040       next_pred_bb_end = NEXT_INSN (BB_END (pred_bb));
1041       for (a_occr = get_bb_avail_insn (pred_bb, expr->avail_occr); a_occr;
1042            a_occr = get_bb_avail_insn (pred_bb, a_occr->next))
1043         {
1044           /* Check if the loaded register is not used.  */
1045           avail_insn = a_occr->insn;
1046           if (! (avail_reg = get_avail_load_store_reg (avail_insn)))
1047             abort ();
1048           /* Make sure we can generate a move from register avail_reg to
1049              dest.  */
1050           extract_insn (gen_move_insn (copy_rtx (dest),
1051                                        copy_rtx (avail_reg)));
1052           if (! constrain_operands (1)
1053               || reg_killed_on_edge (avail_reg, pred)
1054               || reg_used_on_edge (dest, pred))
1055             {
1056               avail_insn = NULL;
1057               continue;
1058             }
1059           if (! reg_set_between_after_reload_p (avail_reg, avail_insn,
1060                                                 next_pred_bb_end))
1061             /* AVAIL_INSN remains non-null.  */
1062             break;
1063           else
1064             avail_insn = NULL;
1065         }
1066
1067       if (EDGE_CRITICAL_P (pred))
1068         critical_count += pred->count;
1069
1070       if (avail_insn != NULL_RTX)
1071         {
1072           npred_ok++;
1073           ok_count += pred->count;
1074           occr = (struct unoccr *) obstack_alloc (&unoccr_obstack,
1075                                                   sizeof (struct occr));
1076           occr->insn = avail_insn;
1077           occr->pred = pred;
1078           occr->next = avail_occrs;
1079           avail_occrs = occr;
1080           if (! rollback_unoccr)
1081             rollback_unoccr = occr;
1082         }
1083       else
1084         {
1085           not_ok_count += pred->count;
1086           unoccr = (struct unoccr *) obstack_alloc (&unoccr_obstack,
1087                                                     sizeof (struct unoccr));
1088           unoccr->insn = NULL_RTX;
1089           unoccr->pred = pred;
1090           unoccr->next = unavail_occrs;
1091           unavail_occrs = unoccr;
1092           if (! rollback_unoccr)
1093             rollback_unoccr = unoccr;
1094         }
1095     }
1096
1097   if (/* No load can be replaced by copy.  */
1098       npred_ok == 0
1099       /* Prevent exploding the code.  */
1100       || (optimize_size && npred_ok > 1))
1101     goto cleanup;
1102
1103   /* Check if it's worth applying the partial redundancy elimination.  */
1104   if (ok_count < GCSE_AFTER_RELOAD_PARTIAL_FRACTION * not_ok_count)
1105     goto cleanup;
1106   if (ok_count < GCSE_AFTER_RELOAD_CRITICAL_FRACTION * critical_count)
1107     goto cleanup;
1108
1109   /* Generate moves to the loaded register from where
1110      the memory is available.  */
1111   for (occr = avail_occrs; occr; occr = occr->next)
1112     {
1113       avail_insn = occr->insn;
1114       pred = occr->pred;
1115       /* Set avail_reg to be the register having the value of the
1116          memory.  */
1117       avail_reg = get_avail_load_store_reg (avail_insn);
1118       if (! avail_reg)
1119         abort ();
1120
1121       insert_insn_on_edge (gen_move_insn (copy_rtx (dest),
1122                                           copy_rtx (avail_reg)),
1123                            pred);
1124       stats.moves_inserted++;
1125
1126       if (dump_file)
1127         fprintf (dump_file,
1128                  "generating move from %d to %d on edge from %d to %d\n",
1129                  REGNO (avail_reg),
1130                  REGNO (dest),
1131                  pred->src->index,
1132                  pred->dest->index);
1133     }
1134
1135   /* Regenerate loads where the memory is unavailable.  */
1136   for (unoccr = unavail_occrs; unoccr; unoccr = unoccr->next)
1137     {
1138       pred = unoccr->pred;
1139       insert_insn_on_edge (copy_insn (PATTERN (insn)), pred);
1140       stats.copies_inserted++;
1141
1142       if (dump_file)
1143         {
1144           fprintf (dump_file,
1145                    "generating on edge from %d to %d a copy of load: ",
1146                    pred->src->index,
1147                    pred->dest->index);
1148           print_rtl (dump_file, PATTERN (insn));
1149           fprintf (dump_file, "\n");
1150         }
1151     }
1152
1153   /* Delete the insn if it is not available in this block and mark it
1154      for deletion if it is available. If insn is available it may help
1155      discover additional redundancies, so mark it for later deletion.  */
1156   for (a_occr = get_bb_avail_insn (bb, expr->avail_occr);
1157        a_occr && (a_occr->insn != insn);
1158        a_occr = get_bb_avail_insn (bb, a_occr->next));
1159
1160   if (!a_occr)
1161     delete_insn (insn);
1162   else
1163     a_occr->deleted_p = 1;
1164
1165 cleanup:
1166   if (rollback_unoccr)
1167     obstack_free (&unoccr_obstack, rollback_unoccr);
1168 }
1169
1170 /* Performing the redundancy elimination as described before.  */
1171
1172 static void
1173 eliminate_partially_redundant_loads (void)
1174 {
1175   rtx insn;
1176   basic_block bb;
1177
1178   /* Note we start at block 1.  */
1179
1180   if (ENTRY_BLOCK_PTR->next_bb == EXIT_BLOCK_PTR)
1181     return;
1182
1183   FOR_BB_BETWEEN (bb,
1184                   ENTRY_BLOCK_PTR->next_bb->next_bb,
1185                   EXIT_BLOCK_PTR,
1186                   next_bb)
1187     {
1188       /* Don't try anything on basic blocks with strange predecessors.  */
1189       if (! bb_has_well_behaved_predecessors (bb))
1190         continue;
1191
1192       /* Do not try anything on cold basic blocks.  */
1193       if (probably_cold_bb_p (bb))
1194         continue;
1195
1196       /* Reset the table of things changed since the start of the current
1197          basic block.  */
1198       reset_opr_set_tables ();
1199
1200       /* Look at all insns in the current basic block and see if there are
1201          any loads in it that we can record.  */
1202       FOR_BB_INSNS (bb, insn)
1203         {
1204           /* Is it a load - of the form (set (reg) (mem))?  */
1205           if (NONJUMP_INSN_P (insn)
1206               && GET_CODE (PATTERN (insn)) == SET
1207               && REG_P (SET_DEST (PATTERN (insn)))
1208               && MEM_P (SET_SRC (PATTERN (insn))))
1209             {
1210               rtx pat = PATTERN (insn);
1211               rtx src = SET_SRC (pat);
1212               struct expr *expr;
1213
1214               if (!MEM_VOLATILE_P (src)
1215                   && GET_MODE (src) != BLKmode
1216                   && general_operand (src, GET_MODE (src))
1217                   /* Are the operands unchanged since the start of the
1218                      block?  */
1219                   && oprs_unchanged_p (src, insn, false)
1220                   && !(flag_non_call_exceptions && may_trap_p (src))
1221                   && !side_effects_p (src)
1222                   /* Is the expression recorded?  */
1223                   && (expr = lookup_expr_in_table (src)) != NULL)
1224                 {
1225                   /* We now have a load (insn) and an available memory at
1226                      its BB start (expr). Try to remove the loads if it is
1227                      redundant.  */
1228                   eliminate_partially_redundant_load (bb, insn, expr);
1229                 }
1230             }
1231
1232           /* Keep track of everything modified by this insn, so that we
1233              know what has been modified since the start of the current
1234              basic block.  */
1235           if (INSN_P (insn))
1236             record_opr_changes (insn);
1237         }
1238     }
1239
1240   commit_edge_insertions ();
1241 }
1242
1243 /* Go over the expression hash table and delete insns that were
1244    marked for later deletion.  */
1245
1246 /* This helper is called via htab_traverse.  */
1247 static int
1248 delete_redundant_insns_1 (void **slot, void *data ATTRIBUTE_UNUSED)
1249 {
1250   struct expr *expr = (struct expr *) *slot;
1251   struct occr *occr;
1252
1253   for (occr = expr->avail_occr; occr != NULL; occr = occr->next)
1254     {
1255       if (occr->deleted_p)
1256         {
1257           delete_insn (occr->insn);
1258           stats.insns_deleted++;
1259
1260           if (dump_file)
1261             {
1262               fprintf (dump_file, "deleting insn:\n");
1263               print_rtl_single (dump_file, occr->insn);
1264               fprintf (dump_file, "\n");
1265             }
1266         }
1267     }
1268
1269   return 1;
1270 }
1271
1272 static void
1273 delete_redundant_insns (void)
1274 {
1275   htab_traverse (expr_table, delete_redundant_insns_1, NULL);
1276   if (dump_file)
1277     fprintf (dump_file, "\n");
1278 }
1279
1280 /* Main entry point of the GCSE after reload - clean some redundant loads
1281    due to spilling.  */
1282
1283 void
1284 gcse_after_reload_main (rtx f ATTRIBUTE_UNUSED)
1285 {
1286   memset (&stats, 0, sizeof (stats));
1287
1288   /* Allocate ememory for this pass.
1289      Also computes and initializes the insns' CUIDs.  */
1290   alloc_mem ();
1291
1292   /* We need alias analysis.  */
1293   init_alias_analysis ();
1294
1295   compute_hash_table ();
1296
1297   if (dump_file)
1298     dump_hash_table (dump_file);
1299
1300   if (htab_elements (expr_table) > 0)
1301     {
1302       eliminate_partially_redundant_loads ();
1303       delete_redundant_insns ();
1304
1305       if (dump_file)
1306         {
1307           fprintf (dump_file, "GCSE AFTER RELOAD stats:\n");
1308           fprintf (dump_file, "copies inserted: %d\n", stats.copies_inserted);
1309           fprintf (dump_file, "moves inserted:  %d\n", stats.moves_inserted);
1310           fprintf (dump_file, "insns deleted:   %d\n", stats.insns_deleted);
1311           fprintf (dump_file, "\n\n");
1312         }
1313     }
1314
1315   /* We are finished with alias.  */
1316   end_alias_analysis ();
1317
1318   free_mem ();
1319 }
1320