gcc/tree-ssa-loop-ivcanon.c

   1 /* Induction variable canonicalization.
   2    Copyright (C) 2004, 2005 Free Software Foundation, Inc.
   3
   4 This file is part of GCC.
   5
   6 GCC is free software; you can redistribute it and/or modify it
   7 under the terms of the GNU General Public License as published by the
   8 Free Software Foundation; either version 2, or (at your option) any
   9 later version.
  10
  11 GCC is distributed in the hope that it will be useful, but WITHOUT
  12 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14 for more details.
  15
  16 You should have received a copy of the GNU General Public License
  17 along with GCC; see the file COPYING.  If not, write to the Free
  18 Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
  19 02110-1301, USA.  */
  20
  21 /* This pass detects the loops that iterate a constant number of times,
  22    adds a canonical induction variable (step -1, tested against 0)
  23    and replaces the exit test.  This enables the less powerful rtl
  24    level analysis to use this information.
  25
  26    This might spoil the code in some cases (by increasing register pressure).
  27    Note that in the case the new variable is not needed, ivopts will get rid
  28    of it, so it might only be a problem when there are no other linear induction
  29    variables.  In that case the created optimization possibilities are likely
  30    to pay up.
  31
  32    Additionally in case we detect that it is beneficial to unroll the
  33    loop completely, we do it right here to expose the optimization
  34    possibilities to the following passes.  */
  35
  36 #include "config.h"
  37 #include "system.h"
  38 #include "coretypes.h"
  39 #include "tm.h"
  40 #include "tree.h"
  41 #include "rtl.h"
  42 #include "tm_p.h"
  43 #include "hard-reg-set.h"
  44 #include "basic-block.h"
  45 #include "output.h"
  46 #include "diagnostic.h"
  47 #include "tree-flow.h"
  48 #include "tree-dump.h"
  49 #include "cfgloop.h"
  50 #include "tree-pass.h"
  51 #include "ggc.h"
  52 #include "tree-chrec.h"
  53 #include "tree-scalar-evolution.h"
  54 #include "params.h"
  55 #include "flags.h"
  56 #include "tree-inline.h"
  57
  58 /* Specifies types of loops that may be unrolled.  */
  59
  60 enum unroll_level
  61 {
  62   UL_SINGLE_ITER,       /* Only loops that exit immediately in the first
  63                            iteration.  */
  64   UL_NO_GROWTH,         /* Only loops whose unrolling will not cause increase
  65                            of code size.  */
  66   UL_ALL                /* All suitable loops.  */
  67 };
  68
  69 /* Adds a canonical induction variable to LOOP iterating NITER times.  EXIT
  70    is the exit edge whose condition is replaced.  */
  71
  72 static void
  73 create_canonical_iv (struct loop *loop, edge exit, tree niter)
  74 {
  75   edge in;
  76   tree cond, type, var;
  77   block_stmt_iterator incr_at;
  78   enum tree_code cmp;
  79
  80   if (dump_file && (dump_flags & TDF_DETAILS))
  81     {
  82       fprintf (dump_file, "Added canonical iv to loop %d, ", loop->num);
  83       print_generic_expr (dump_file, niter, TDF_SLIM);
  84       fprintf (dump_file, " iterations.\n");
  85     }
  86
  87   cond = last_stmt (exit->src);
  88   in = EDGE_SUCC (exit->src, 0);
  89   if (in == exit)
  90     in = EDGE_SUCC (exit->src, 1);
  91
  92   /* Note that we do not need to worry about overflows, since
  93      type of niter is always unsigned and all comparisons are
  94      just for equality/nonequality -- i.e. everything works
  95      with a modulo arithmetics.  */
  96
  97   type = TREE_TYPE (niter);
  98   niter = fold_build2 (PLUS_EXPR, type,
  99                        niter,
 100                        build_int_cst (type, 1));
 101   incr_at = bsi_last (in->src);
 102   create_iv (niter,
 103              build_int_cst (type, -1),
 104              NULL_TREE, loop,
 105              &incr_at, false, NULL, &var);
 106
 107   cmp = (exit->flags & EDGE_TRUE_VALUE) ? EQ_EXPR : NE_EXPR;
 108   COND_EXPR_COND (cond) = build2 (cmp, boolean_type_node,
 109                                   var,
 110                                   build_int_cst (type, 0));
 111   update_stmt (cond);
 112 }
 113
 114 /* Computes an estimated number of insns in LOOP.  */
 115
 116 unsigned
 117 tree_num_loop_insns (struct loop *loop)
 118 {
 119   basic_block *body = get_loop_body (loop);
 120   block_stmt_iterator bsi;
 121   unsigned size = 1, i;
 122
 123   for (i = 0; i < loop->num_nodes; i++)
 124     for (bsi = bsi_start (body[i]); !bsi_end_p (bsi); bsi_next (&bsi))
 125       size += estimate_num_insns (bsi_stmt (bsi));
 126   free (body);
 127
 128   return size;
 129 }
 130
 131 /* Estimate number of insns of completely unrolled loop.  We assume
 132    that the size of the unrolled loop is decreased in the
 133    following way (the numbers of insns are based on what
 134    estimate_num_insns returns for appropriate statements):
 135
 136    1) exit condition gets removed (2 insns)
 137    2) increment of the control variable gets removed (2 insns)
 138    3) All remaining statements are likely to get simplified
 139       due to constant propagation.  Hard to estimate; just
 140       as a heuristics we decrease the rest by 1/3.
 141
 142    NINSNS is the number of insns in the loop before unrolling.
 143    NUNROLL is the number of times the loop is unrolled.  */
 144
 145 static unsigned HOST_WIDE_INT
 146 estimated_unrolled_size (unsigned HOST_WIDE_INT ninsns,
 147                          unsigned HOST_WIDE_INT nunroll)
 148 {
 149   HOST_WIDE_INT unr_insns = 2 * ((HOST_WIDE_INT) ninsns - 4) / 3;
 150   if (unr_insns <= 0)
 151     unr_insns = 1;
 152   unr_insns *= (nunroll + 1);
 153
 154   return unr_insns;
 155 }
 156
 157 /* Tries to unroll LOOP completely, i.e. NITER times.
 158    UL determines which loops we are allowed to unroll.
 159    EXIT is the exit of the loop that should be eliminated.  */
 160
 161 static bool
 162 try_unroll_loop_completely (struct loop *loop,
 163                             edge exit, tree niter,
 164                             enum unroll_level ul)
 165 {
 166   unsigned HOST_WIDE_INT n_unroll, ninsns, max_unroll, unr_insns;
 167   tree old_cond, cond, dont_exit, do_exit;
 168
 169   if (loop->inner)
 170     return false;
 171
 172   if (!host_integerp (niter, 1))
 173     return false;
 174   n_unroll = tree_low_cst (niter, 1);
 175
 176   max_unroll = PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES);
 177   if (n_unroll > max_unroll)
 178     return false;
 179
 180   if (n_unroll)
 181     {
 182       if (ul == UL_SINGLE_ITER)
 183         return false;
 184
 185       ninsns = tree_num_loop_insns (loop);
 186
 187       if (n_unroll * ninsns
 188           > (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS))
 189         return false;
 190
 191       if (ul == UL_NO_GROWTH)
 192         {
 193           unr_insns = estimated_unrolled_size (ninsns, n_unroll);
 194
 195           if (dump_file && (dump_flags & TDF_DETAILS))
 196             {
 197               fprintf (dump_file, "  Loop size: %d\n", (int) ninsns);
 198               fprintf (dump_file, "  Estimated size after unrolling: %d\n",
 199                        (int) unr_insns);
 200             }
 201
 202           if (unr_insns > ninsns)
 203             {
 204               if (dump_file && (dump_flags & TDF_DETAILS))
 205                 fprintf (dump_file, "Not unrolling loop %d:\n", loop->num);
 206               return false;
 207             }
 208         }
 209     }
 210
 211   if (exit->flags & EDGE_TRUE_VALUE)
 212     {
 213       dont_exit = boolean_false_node;
 214       do_exit = boolean_true_node;
 215     }
 216   else
 217     {
 218       dont_exit = boolean_true_node;
 219       do_exit = boolean_false_node;
 220     }
 221   cond = last_stmt (exit->src);
 222
 223   if (n_unroll)
 224     {
 225       sbitmap wont_exit;
 226       edge *edges_to_remove = XNEWVEC (edge, n_unroll);
 227       unsigned int n_to_remove = 0;
 228
 229       old_cond = COND_EXPR_COND (cond);
 230       COND_EXPR_COND (cond) = dont_exit;
 231       update_stmt (cond);
 232       initialize_original_copy_tables ();
 233
 234       wont_exit = sbitmap_alloc (n_unroll + 1);
 235       sbitmap_ones (wont_exit);
 236       RESET_BIT (wont_exit, 0);
 237
 238       if (!tree_duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
 239                                                n_unroll, wont_exit,
 240                                                exit, edges_to_remove,
 241                                                &n_to_remove,
 242                                                DLTHE_FLAG_UPDATE_FREQ
 243                                                | DLTHE_FLAG_COMPLETTE_PEEL))
 244         {
 245           COND_EXPR_COND (cond) = old_cond;
 246           update_stmt (cond);
 247           free_original_copy_tables ();
 248           free (wont_exit);
 249           free (edges_to_remove);
 250           return false;
 251         }
 252       free (wont_exit);
 253       free (edges_to_remove);
 254       free_original_copy_tables ();
 255     }
 256
 257   COND_EXPR_COND (cond) = do_exit;
 258   update_stmt (cond);
 259
 260   update_ssa (TODO_update_ssa);
 261
 262   if (dump_file && (dump_flags & TDF_DETAILS))
 263     fprintf (dump_file, "Unrolled loop %d completely.\n", loop->num);
 264
 265   return true;
 266 }
 267
 268 /* Adds a canonical induction variable to LOOP if suitable.
 269    CREATE_IV is true if we may create a new iv.  UL determines
 270    which loops we are allowed to completely unroll.  If TRY_EVAL is true, we try
 271    to determine the number of iterations of a loop by direct evaluation.
 272    Returns true if cfg is changed.  */
 273
 274 static bool
 275 canonicalize_loop_induction_variables (struct loop *loop,
 276                                        bool create_iv, enum unroll_level ul,
 277                                        bool try_eval)
 278 {
 279   edge exit = NULL;
 280   tree niter;
 281
 282   niter = number_of_latch_executions (loop);
 283   if (TREE_CODE (niter) == INTEGER_CST)
 284     {
 285       exit = single_exit (loop);
 286       if (!just_once_each_iteration_p (loop, exit->src))
 287         return false;
 288     }
 289   else
 290     {
 291       /* If the loop has more than one exit, try checking all of them
 292          for # of iterations determinable through scev.  */
 293       if (!single_exit (loop))
 294         niter = find_loop_niter (loop, &exit);
 295
 296       /* Finally if everything else fails, try brute force evaluation.  */
 297       if (try_eval
 298           && (chrec_contains_undetermined (niter)
 299               || TREE_CODE (niter) != INTEGER_CST))
 300         niter = find_loop_niter_by_eval (loop, &exit);
 301
 302       if (chrec_contains_undetermined (niter)
 303           || TREE_CODE (niter) != INTEGER_CST)
 304         return false;
 305     }
 306
 307   if (dump_file && (dump_flags & TDF_DETAILS))
 308     {
 309       fprintf (dump_file, "Loop %d iterates ", loop->num);
 310       print_generic_expr (dump_file, niter, TDF_SLIM);
 311       fprintf (dump_file, " times.\n");
 312     }
 313
 314   if (try_unroll_loop_completely (loop, exit, niter, ul))
 315     return true;
 316
 317   if (create_iv)
 318     create_canonical_iv (loop, exit, niter);
 319
 320   return false;
 321 }
 322
 323 /* The main entry point of the pass.  Adds canonical induction variables
 324    to the suitable loops.  */
 325
 326 unsigned int
 327 canonicalize_induction_variables (void)
 328 {
 329   loop_iterator li;
 330   struct loop *loop;
 331   bool changed = false;
 332
 333   FOR_EACH_LOOP (li, loop, 0)
 334     {
 335       changed |= canonicalize_loop_induction_variables (loop,
 336                                                         true, UL_SINGLE_ITER,
 337                                                         true);
 338     }
 339
 340   /* Clean up the information about numbers of iterations, since brute force
 341      evaluation could reveal new information.  */
 342   scev_reset ();
 343
 344   if (changed)
 345     return TODO_cleanup_cfg;
 346   return 0;
 347 }
 348
 349 /* Unroll LOOPS completely if they iterate just few times.  Unless
 350    MAY_INCREASE_SIZE is true, perform the unrolling only if the
 351    size of the code does not increase.  */
 352
 353 unsigned int
 354 tree_unroll_loops_completely (bool may_increase_size)
 355 {
 356   loop_iterator li;
 357   struct loop *loop;
 358   bool changed = false;
 359   enum unroll_level ul;
 360
 361   FOR_EACH_LOOP (li, loop, 0)
 362     {
 363       if (may_increase_size && maybe_hot_bb_p (loop->header))
 364         ul = UL_ALL;
 365       else
 366         ul = UL_NO_GROWTH;
 367       changed |= canonicalize_loop_induction_variables (loop,
 368                                                         false, ul,
 369                                                         !flag_tree_loop_ivcanon);
 370     }
 371
 372   /* Clean up the information about numbers of iterations, since complete
 373      unrolling might have invalidated it.  */
 374   scev_reset ();
 375
 376   if (changed)
 377     return TODO_cleanup_cfg;
 378   return 0;
 379 }
 380
 381 /* Checks whether LOOP is empty.  */
 382
 383 static bool
 384 empty_loop_p (struct loop *loop)
 385 {
 386   edge exit;
 387   struct tree_niter_desc niter;
 388   tree phi, def;
 389   basic_block *body;
 390   block_stmt_iterator bsi;
 391   unsigned i;
 392   tree stmt;
 393
 394   /* If the loop has multiple exits, it is too hard for us to handle.
 395      Similarly, if the exit is not dominating, we cannot determine
 396      whether the loop is not infinite.  */
 397   exit = single_dom_exit (loop);
 398   if (!exit)
 399     return false;
 400
 401   /* The loop must be finite.  */
 402   if (!number_of_iterations_exit (loop, exit, &niter, false))
 403     return false;
 404
 405   /* Values of all loop exit phi nodes must be invariants.  */
 406   for (phi = phi_nodes (exit->dest); phi; phi = PHI_CHAIN (phi))
 407     {
 408       if (!is_gimple_reg (PHI_RESULT (phi)))
 409         continue;
 410
 411       def = PHI_ARG_DEF_FROM_EDGE (phi, exit);
 412
 413       if (!expr_invariant_in_loop_p (loop, def))
 414         return false;
 415     }
 416
 417   /* And there should be no memory modifying or from other reasons
 418      unremovable statements.  */
 419   body = get_loop_body (loop);
 420   for (i = 0; i < loop->num_nodes; i++)
 421     {
 422       /* Irreducible region might be infinite.  */
 423       if (body[i]->flags & BB_IRREDUCIBLE_LOOP)
 424         {
 425           free (body);
 426           return false;
 427         }
 428
 429       for (bsi = bsi_start (body[i]); !bsi_end_p (bsi); bsi_next (&bsi))
 430         {
 431           stmt = bsi_stmt (bsi);
 432           if (!ZERO_SSA_OPERANDS (stmt, SSA_OP_VIRTUAL_DEFS)
 433               || stmt_ann (stmt)->has_volatile_ops)
 434             {
 435               free (body);
 436               return false;
 437             }
 438
 439           /* Also, asm statements and calls may have side effects and we
 440              cannot change the number of times they are executed.  */
 441           switch (TREE_CODE (stmt))
 442             {
 443             case RETURN_EXPR:
 444             case GIMPLE_MODIFY_STMT:
 445               stmt = get_call_expr_in (stmt);
 446               if (!stmt)
 447                 break;
 448
 449             case CALL_EXPR:
 450               if (TREE_SIDE_EFFECTS (stmt))
 451                 {
 452                   free (body);
 453                   return false;
 454                 }
 455               break;
 456
 457             case ASM_EXPR:
 458               /* We cannot remove volatile assembler.  */
 459               if (ASM_VOLATILE_P (stmt))
 460                 {
 461                   free (body);
 462                   return false;
 463                 }
 464               break;
 465
 466             default:
 467               break;
 468             }
 469         }
 470       }
 471   free (body);
 472
 473   return true;
 474 }
 475
 476 /* Remove LOOP by making it exit in the first iteration.  */
 477
 478 static void
 479 remove_empty_loop (struct loop *loop)
 480 {
 481   edge exit = single_dom_exit (loop), non_exit;
 482   tree cond_stmt = last_stmt (exit->src);
 483   tree do_exit;
 484   basic_block *body;
 485   unsigned n_before, freq_in, freq_h;
 486   gcov_type exit_count = exit->count;
 487
 488   non_exit = EDGE_SUCC (exit->src, 0);
 489   if (non_exit == exit)
 490     non_exit = EDGE_SUCC (exit->src, 1);
 491
 492   if (exit->flags & EDGE_TRUE_VALUE)
 493     do_exit = boolean_true_node;
 494   else
 495     do_exit = boolean_false_node;
 496
 497   COND_EXPR_COND (cond_stmt) = do_exit;
 498   update_stmt (cond_stmt);
 499
 500   /* Let us set the probabilities of the edges coming from the exit block.  */
 501   exit->probability = REG_BR_PROB_BASE;
 502   non_exit->probability = 0;
 503   non_exit->count = 0;
 504
 505   /* Update frequencies and counts.  Everything before
 506      the exit needs to be scaled FREQ_IN/FREQ_H times,
 507      where FREQ_IN is the frequency of the entry edge
 508      and FREQ_H is the frequency of the loop header.
 509      Everything after the exit has zero frequency.  */
 510   freq_h = loop->header->frequency;
 511   freq_in = EDGE_FREQUENCY (loop_preheader_edge (loop));
 512   if (freq_h != 0)
 513     {
 514       body = get_loop_body_in_dom_order (loop);
 515       for (n_before = 1; n_before <= loop->num_nodes; n_before++)
 516         if (body[n_before - 1] == exit->src)
 517           break;
 518       scale_bbs_frequencies_int (body, n_before, freq_in, freq_h);
 519       scale_bbs_frequencies_int (body + n_before, loop->num_nodes - n_before,
 520                                  0, 1);
 521       free (body);
 522     }
 523
 524   /* Number of executions of exit is not changed, thus we need to restore
 525      the original value.  */
 526   exit->count = exit_count;
 527 }
 528
 529 /* Removes LOOP if it is empty.  Returns true if LOOP is removed.  CHANGED
 530    is set to true if LOOP or any of its subloops is removed.  */
 531
 532 static bool
 533 try_remove_empty_loop (struct loop *loop, bool *changed)
 534 {
 535   bool nonempty_subloop = false;
 536   struct loop *sub;
 537
 538   /* First, all subloops must be removed.  */
 539   for (sub = loop->inner; sub; sub = sub->next)
 540     nonempty_subloop |= !try_remove_empty_loop (sub, changed);
 541
 542   if (nonempty_subloop || !empty_loop_p (loop))
 543     return false;
 544
 545   remove_empty_loop (loop);
 546   *changed = true;
 547   return true;
 548 }
 549
 550 /* Remove the empty loops.  */
 551
 552 unsigned int
 553 remove_empty_loops (void)
 554 {
 555   bool changed = false;
 556   struct loop *loop;
 557
 558   for (loop = current_loops->tree_root->inner; loop; loop = loop->next)
 559     try_remove_empty_loop (loop, &changed);
 560
 561   if (changed)
 562     {
 563       scev_reset ();
 564       return TODO_cleanup_cfg;
 565     }
 566   return 0;
 567 }