gcc/tree-ssa-sink.c

   1 /* Code sinking for trees
   2    Copyright (C) 2001-2014 Free Software Foundation, Inc.
   3    Contributed by Daniel Berlin <dan@dberlin.org>
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify
   8 it under the terms of the GNU General Public License as published by
   9 the Free Software Foundation; either version 3, or (at your option)
  10 any later version.
  11
  12 GCC is distributed in the hope that it will be useful,
  13 but WITHOUT ANY WARRANTY; without even the implied warranty of
  14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15 GNU General Public License for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING3.  If not see
  19 <http://www.gnu.org/licenses/>.  */
  20
  21 #include "config.h"
  22 #include "system.h"
  23 #include "coretypes.h"
  24 #include "tm.h"
  25 #include "tree.h"
  26 #include "stor-layout.h"
  27 #include "predict.h"
  28 #include "vec.h"
  29 #include "hashtab.h"
  30 #include "hash-set.h"
  31 #include "machmode.h"
  32 #include "hard-reg-set.h"
  33 #include "input.h"
  34 #include "function.h"
  35 #include "dominance.h"
  36 #include "cfg.h"
  37 #include "cfganal.h"
  38 #include "basic-block.h"
  39 #include "gimple-pretty-print.h"
  40 #include "tree-inline.h"
  41 #include "tree-ssa-alias.h"
  42 #include "internal-fn.h"
  43 #include "gimple-expr.h"
  44 #include "is-a.h"
  45 #include "gimple.h"
  46 #include "gimple-iterator.h"
  47 #include "gimple-ssa.h"
  48 #include "tree-cfg.h"
  49 #include "tree-phinodes.h"
  50 #include "ssa-iterators.h"
  51 #include "tree-iterator.h"
  52 #include "alloc-pool.h"
  53 #include "tree-pass.h"
  54 #include "flags.h"
  55 #include "cfgloop.h"
  56 #include "params.h"
  57
  58 /* TODO:
  59    1. Sinking store only using scalar promotion (IE without moving the RHS):
  60
  61    *q = p;
  62    p = p + 1;
  63    if (something)
  64      *q = <not p>;
  65    else
  66      y = *q;
  67
  68
  69    should become
  70    sinktemp = p;
  71    p = p + 1;
  72    if (something)
  73      *q = <not p>;
  74    else
  75    {
  76      *q = sinktemp;
  77      y = *q
  78    }
  79    Store copy propagation will take care of the store elimination above.
  80
  81
  82    2. Sinking using Partial Dead Code Elimination.  */
  83
  84
  85 static struct
  86 {
  87   /* The number of statements sunk down the flowgraph by code sinking.  */
  88   int sunk;
  89
  90 } sink_stats;
  91
  92
  93 /* Given a PHI, and one of its arguments (DEF), find the edge for
  94    that argument and return it.  If the argument occurs twice in the PHI node,
  95    we return NULL.  */
  96
  97 static basic_block
  98 find_bb_for_arg (gphi *phi, tree def)
  99 {
 100   size_t i;
 101   bool foundone = false;
 102   basic_block result = NULL;
 103   for (i = 0; i < gimple_phi_num_args (phi); i++)
 104     if (PHI_ARG_DEF (phi, i) == def)
 105       {
 106         if (foundone)
 107           return NULL;
 108         foundone = true;
 109         result = gimple_phi_arg_edge (phi, i)->src;
 110       }
 111   return result;
 112 }
 113
 114 /* When the first immediate use is in a statement, then return true if all
 115    immediate uses in IMM are in the same statement.
 116    We could also do the case where  the first immediate use is in a phi node,
 117    and all the other uses are in phis in the same basic block, but this
 118    requires some expensive checking later (you have to make sure no def/vdef
 119    in the statement occurs for multiple edges in the various phi nodes it's
 120    used in, so that you only have one place you can sink it to.  */
 121
 122 static bool
 123 all_immediate_uses_same_place (def_operand_p def_p)
 124 {
 125   tree var = DEF_FROM_PTR (def_p);
 126   imm_use_iterator imm_iter;
 127   use_operand_p use_p;
 128
 129   gimple firstuse = NULL;
 130   FOR_EACH_IMM_USE_FAST (use_p, imm_iter, var)
 131     {
 132       if (is_gimple_debug (USE_STMT (use_p)))
 133         continue;
 134       if (firstuse == NULL)
 135         firstuse = USE_STMT (use_p);
 136       else
 137         if (firstuse != USE_STMT (use_p))
 138           return false;
 139     }
 140
 141   return true;
 142 }
 143
 144 /* Find the nearest common dominator of all of the immediate uses in IMM.  */
 145
 146 static basic_block
 147 nearest_common_dominator_of_uses (def_operand_p def_p, bool *debug_stmts)
 148 {
 149   tree var = DEF_FROM_PTR (def_p);
 150   bitmap blocks = BITMAP_ALLOC (NULL);
 151   basic_block commondom;
 152   unsigned int j;
 153   bitmap_iterator bi;
 154   imm_use_iterator imm_iter;
 155   use_operand_p use_p;
 156
 157   FOR_EACH_IMM_USE_FAST (use_p, imm_iter, var)
 158     {
 159       gimple usestmt = USE_STMT (use_p);
 160       basic_block useblock;
 161
 162       if (gphi *phi = dyn_cast <gphi *> (usestmt))
 163         {
 164           int idx = PHI_ARG_INDEX_FROM_USE (use_p);
 165
 166           useblock = gimple_phi_arg_edge (phi, idx)->src;
 167         }
 168       else if (is_gimple_debug (usestmt))
 169         {
 170           *debug_stmts = true;
 171           continue;
 172         }
 173       else
 174         {
 175           useblock = gimple_bb (usestmt);
 176         }
 177
 178       /* Short circuit. Nothing dominates the entry block.  */
 179       if (useblock == ENTRY_BLOCK_PTR_FOR_FN (cfun))
 180         {
 181           BITMAP_FREE (blocks);
 182           return NULL;
 183         }
 184       bitmap_set_bit (blocks, useblock->index);
 185     }
 186   commondom = BASIC_BLOCK_FOR_FN (cfun, bitmap_first_set_bit (blocks));
 187   EXECUTE_IF_SET_IN_BITMAP (blocks, 0, j, bi)
 188     commondom = nearest_common_dominator (CDI_DOMINATORS, commondom,
 189                                           BASIC_BLOCK_FOR_FN (cfun, j));
 190   BITMAP_FREE (blocks);
 191   return commondom;
 192 }
 193
 194 /* Given EARLY_BB and LATE_BB, two blocks in a path through the dominator
 195    tree, return the best basic block between them (inclusive) to place
 196    statements.
 197
 198    We want the most control dependent block in the shallowest loop nest.
 199
 200    If the resulting block is in a shallower loop nest, then use it.  Else
 201    only use the resulting block if it has significantly lower execution
 202    frequency than EARLY_BB to avoid gratutious statement movement.  We
 203    consider statements with VOPS more desirable to move.
 204
 205    This pass would obviously benefit from PDO as it utilizes block
 206    frequencies.  It would also benefit from recomputing frequencies
 207    if profile data is not available since frequencies often get out
 208    of sync with reality.  */
 209
 210 static basic_block
 211 select_best_block (basic_block early_bb,
 212                    basic_block late_bb,
 213                    gimple stmt)
 214 {
 215   basic_block best_bb = late_bb;
 216   basic_block temp_bb = late_bb;
 217   int threshold;
 218
 219   while (temp_bb != early_bb)
 220     {
 221       /* If we've moved into a lower loop nest, then that becomes
 222          our best block.  */
 223       if (bb_loop_depth (temp_bb) < bb_loop_depth (best_bb))
 224         best_bb = temp_bb;
 225
 226       /* Walk up the dominator tree, hopefully we'll find a shallower
 227          loop nest.  */
 228       temp_bb = get_immediate_dominator (CDI_DOMINATORS, temp_bb);
 229     }
 230
 231   /* If we found a shallower loop nest, then we always consider that
 232      a win.  This will always give us the most control dependent block
 233      within that loop nest.  */
 234   if (bb_loop_depth (best_bb) < bb_loop_depth (early_bb))
 235     return best_bb;
 236
 237   /* Get the sinking threshold.  If the statement to be moved has memory
 238      operands, then increase the threshold by 7% as those are even more
 239      profitable to avoid, clamping at 100%.  */
 240   threshold = PARAM_VALUE (PARAM_SINK_FREQUENCY_THRESHOLD);
 241   if (gimple_vuse (stmt) || gimple_vdef (stmt))
 242     {
 243       threshold += 7;
 244       if (threshold > 100)
 245         threshold = 100;
 246     }
 247
 248   /* If BEST_BB is at the same nesting level, then require it to have
 249      significantly lower execution frequency to avoid gratutious movement.  */
 250   if (bb_loop_depth (best_bb) == bb_loop_depth (early_bb)
 251       && best_bb->frequency < (early_bb->frequency * threshold / 100.0))
 252     return best_bb;
 253
 254   /* No better block found, so return EARLY_BB, which happens to be the
 255      statement's original block.  */
 256   return early_bb;
 257 }
 258
 259 /* Given a statement (STMT) and the basic block it is currently in (FROMBB),
 260    determine the location to sink the statement to, if any.
 261    Returns true if there is such location; in that case, TOGSI points to the
 262    statement before that STMT should be moved.  */
 263
 264 static bool
 265 statement_sink_location (gimple stmt, basic_block frombb,
 266                          gimple_stmt_iterator *togsi)
 267 {
 268   gimple use;
 269   use_operand_p one_use = NULL_USE_OPERAND_P;
 270   basic_block sinkbb;
 271   use_operand_p use_p;
 272   def_operand_p def_p;
 273   ssa_op_iter iter;
 274   imm_use_iterator imm_iter;
 275
 276   /* We only can sink assignments.  */
 277   if (!is_gimple_assign (stmt))
 278     return false;
 279
 280   /* We only can sink stmts with a single definition.  */
 281   def_p = single_ssa_def_operand (stmt, SSA_OP_ALL_DEFS);
 282   if (def_p == NULL_DEF_OPERAND_P)
 283     return false;
 284
 285   /* Return if there are no immediate uses of this stmt.  */
 286   if (has_zero_uses (DEF_FROM_PTR (def_p)))
 287     return false;
 288
 289   /* There are a few classes of things we can't or don't move, some because we
 290      don't have code to handle it, some because it's not profitable and some
 291      because it's not legal.
 292
 293      We can't sink things that may be global stores, at least not without
 294      calculating a lot more information, because we may cause it to no longer
 295      be seen by an external routine that needs it depending on where it gets
 296      moved to.
 297
 298      We can't sink statements that end basic blocks without splitting the
 299      incoming edge for the sink location to place it there.
 300
 301      We can't sink statements that have volatile operands.
 302
 303      We don't want to sink dead code, so anything with 0 immediate uses is not
 304      sunk.
 305
 306      Don't sink BLKmode assignments if current function has any local explicit
 307      register variables, as BLKmode assignments may involve memcpy or memset
 308      calls or, on some targets, inline expansion thereof that sometimes need
 309      to use specific hard registers.
 310
 311   */
 312   if (stmt_ends_bb_p (stmt)
 313       || gimple_has_side_effects (stmt)
 314       || gimple_has_volatile_ops (stmt)
 315       || (cfun->has_local_explicit_reg_vars
 316           && TYPE_MODE (TREE_TYPE (gimple_assign_lhs (stmt))) == BLKmode))
 317     return false;
 318
 319   if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (DEF_FROM_PTR (def_p)))
 320     return false;
 321
 322   FOR_EACH_SSA_USE_OPERAND (use_p, stmt, iter, SSA_OP_ALL_USES)
 323     {
 324       tree use = USE_FROM_PTR (use_p);
 325       if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (use))
 326         return false;
 327     }
 328
 329   use = NULL;
 330
 331   /* If stmt is a store the one and only use needs to be the VOP
 332      merging PHI node.  */
 333   if (virtual_operand_p (DEF_FROM_PTR (def_p)))
 334     {
 335       FOR_EACH_IMM_USE_FAST (use_p, imm_iter, DEF_FROM_PTR (def_p))
 336         {
 337           gimple use_stmt = USE_STMT (use_p);
 338
 339           /* A killing definition is not a use.  */
 340           if ((gimple_has_lhs (use_stmt)
 341                && operand_equal_p (gimple_assign_lhs (stmt),
 342                                    gimple_get_lhs (use_stmt), 0))
 343               || stmt_kills_ref_p (use_stmt, gimple_assign_lhs (stmt)))
 344             {
 345               /* If use_stmt is or might be a nop assignment then USE_STMT
 346                  acts as a use as well as definition.  */
 347               if (stmt != use_stmt
 348                   && ref_maybe_used_by_stmt_p (use_stmt,
 349                                                gimple_assign_lhs (stmt)))
 350                 return false;
 351               continue;
 352             }
 353
 354           if (gimple_code (use_stmt) != GIMPLE_PHI)
 355             return false;
 356
 357           if (use
 358               && use != use_stmt)
 359             return false;
 360
 361           use = use_stmt;
 362         }
 363       if (!use)
 364         return false;
 365     }
 366   /* If all the immediate uses are not in the same place, find the nearest
 367      common dominator of all the immediate uses.  For PHI nodes, we have to
 368      find the nearest common dominator of all of the predecessor blocks, since
 369      that is where insertion would have to take place.  */
 370   else if (gimple_vuse (stmt)
 371            || !all_immediate_uses_same_place (def_p))
 372     {
 373       bool debug_stmts = false;
 374       basic_block commondom = nearest_common_dominator_of_uses (def_p,
 375                                                                 &debug_stmts);
 376
 377       if (commondom == frombb)
 378         return false;
 379
 380       /* If this is a load then do not sink past any stores.
 381          ???  This is overly simple but cheap.  We basically look
 382          for an existing load with the same VUSE in the path to one
 383          of the sink candidate blocks and we adjust commondom to the
 384          nearest to commondom.  */
 385       if (gimple_vuse (stmt))
 386         {
 387           /* Do not sink loads from hard registers.  */
 388           if (gimple_assign_single_p (stmt)
 389               && TREE_CODE (gimple_assign_rhs1 (stmt)) == VAR_DECL
 390               && DECL_HARD_REGISTER (gimple_assign_rhs1 (stmt)))
 391             return false;
 392
 393           imm_use_iterator imm_iter;
 394           use_operand_p use_p;
 395           basic_block found = NULL;
 396           FOR_EACH_IMM_USE_FAST (use_p, imm_iter, gimple_vuse (stmt))
 397             {
 398               gimple use_stmt = USE_STMT (use_p);
 399               basic_block bb = gimple_bb (use_stmt);
 400               /* For PHI nodes the block we know sth about
 401                  is the incoming block with the use.  */
 402               if (gimple_code (use_stmt) == GIMPLE_PHI)
 403                 bb = EDGE_PRED (bb, PHI_ARG_INDEX_FROM_USE (use_p))->src;
 404               /* Any dominator of commondom would be ok with
 405                  adjusting commondom to that block.  */
 406               bb = nearest_common_dominator (CDI_DOMINATORS, bb, commondom);
 407               if (!found)
 408                 found = bb;
 409               else if (dominated_by_p (CDI_DOMINATORS, bb, found))
 410                 found = bb;
 411               /* If we can't improve, stop.  */
 412               if (found == commondom)
 413                 break;
 414             }
 415           commondom = found;
 416           if (commondom == frombb)
 417             return false;
 418         }
 419
 420       /* Our common dominator has to be dominated by frombb in order to be a
 421          trivially safe place to put this statement, since it has multiple
 422          uses.  */
 423       if (!dominated_by_p (CDI_DOMINATORS, commondom, frombb))
 424         return false;
 425
 426       commondom = select_best_block (frombb, commondom, stmt);
 427
 428       if (commondom == frombb)
 429         return false;
 430
 431       *togsi = gsi_after_labels (commondom);
 432
 433       return true;
 434     }
 435   else
 436     {
 437       FOR_EACH_IMM_USE_FAST (one_use, imm_iter, DEF_FROM_PTR (def_p))
 438         {
 439           if (is_gimple_debug (USE_STMT (one_use)))
 440             continue;
 441           break;
 442         }
 443       use = USE_STMT (one_use);
 444
 445       if (gimple_code (use) != GIMPLE_PHI)
 446         {
 447           sinkbb = gimple_bb (use);
 448           sinkbb = select_best_block (frombb, gimple_bb (use), stmt);
 449
 450           if (sinkbb == frombb)
 451             return false;
 452
 453           *togsi = gsi_for_stmt (use);
 454
 455           return true;
 456         }
 457     }
 458
 459   sinkbb = find_bb_for_arg (as_a <gphi *> (use), DEF_FROM_PTR (def_p));
 460
 461   /* This can happen if there are multiple uses in a PHI.  */
 462   if (!sinkbb)
 463     return false;
 464
 465   sinkbb = select_best_block (frombb, sinkbb, stmt);
 466   if (!sinkbb || sinkbb == frombb)
 467     return false;
 468
 469   /* If the latch block is empty, don't make it non-empty by sinking
 470      something into it.  */
 471   if (sinkbb == frombb->loop_father->latch
 472       && empty_block_p (sinkbb))
 473     return false;
 474
 475   *togsi = gsi_after_labels (sinkbb);
 476
 477   return true;
 478 }
 479
 480 /* Perform code sinking on BB */
 481
 482 static void
 483 sink_code_in_bb (basic_block bb)
 484 {
 485   basic_block son;
 486   gimple_stmt_iterator gsi;
 487   edge_iterator ei;
 488   edge e;
 489   bool last = true;
 490
 491   /* If this block doesn't dominate anything, there can't be any place to sink
 492      the statements to.  */
 493   if (first_dom_son (CDI_DOMINATORS, bb) == NULL)
 494     goto earlyout;
 495
 496   /* We can't move things across abnormal edges, so don't try.  */
 497   FOR_EACH_EDGE (e, ei, bb->succs)
 498     if (e->flags & EDGE_ABNORMAL)
 499       goto earlyout;
 500
 501   for (gsi = gsi_last_bb (bb); !gsi_end_p (gsi);)
 502     {
 503       gimple stmt = gsi_stmt (gsi);
 504       gimple_stmt_iterator togsi;
 505
 506       if (!statement_sink_location (stmt, bb, &togsi))
 507         {
 508           if (!gsi_end_p (gsi))
 509             gsi_prev (&gsi);
 510           last = false;
 511           continue;
 512         }
 513       if (dump_file)
 514         {
 515           fprintf (dump_file, "Sinking ");
 516           print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS);
 517           fprintf (dump_file, " from bb %d to bb %d\n",
 518                    bb->index, (gsi_bb (togsi))->index);
 519         }
 520
 521       /* Update virtual operands of statements in the path we
 522          do not sink to.  */
 523       if (gimple_vdef (stmt))
 524         {
 525           imm_use_iterator iter;
 526           use_operand_p use_p;
 527           gimple vuse_stmt;
 528
 529           FOR_EACH_IMM_USE_STMT (vuse_stmt, iter, gimple_vdef (stmt))
 530             if (gimple_code (vuse_stmt) != GIMPLE_PHI)
 531               FOR_EACH_IMM_USE_ON_STMT (use_p, iter)
 532                 SET_USE (use_p, gimple_vuse (stmt));
 533         }
 534
 535       /* If this is the end of the basic block, we need to insert at the end
 536          of the basic block.  */
 537       if (gsi_end_p (togsi))
 538         gsi_move_to_bb_end (&gsi, gsi_bb (togsi));
 539       else
 540         gsi_move_before (&gsi, &togsi);
 541
 542       sink_stats.sunk++;
 543
 544       /* If we've just removed the last statement of the BB, the
 545          gsi_end_p() test below would fail, but gsi_prev() would have
 546          succeeded, and we want it to succeed.  So we keep track of
 547          whether we're at the last statement and pick up the new last
 548          statement.  */
 549       if (last)
 550         {
 551           gsi = gsi_last_bb (bb);
 552           continue;
 553         }
 554
 555       last = false;
 556       if (!gsi_end_p (gsi))
 557         gsi_prev (&gsi);
 558
 559     }
 560  earlyout:
 561   for (son = first_dom_son (CDI_POST_DOMINATORS, bb);
 562        son;
 563        son = next_dom_son (CDI_POST_DOMINATORS, son))
 564     {
 565       sink_code_in_bb (son);
 566     }
 567 }
 568
 569 /* Perform code sinking.
 570    This moves code down the flowgraph when we know it would be
 571    profitable to do so, or it wouldn't increase the number of
 572    executions of the statement.
 573
 574    IE given
 575
 576    a_1 = b + c;
 577    if (<something>)
 578    {
 579    }
 580    else
 581    {
 582      foo (&b, &c);
 583      a_5 = b + c;
 584    }
 585    a_6 = PHI (a_5, a_1);
 586    USE a_6.
 587
 588    we'll transform this into:
 589
 590    if (<something>)
 591    {
 592       a_1 = b + c;
 593    }
 594    else
 595    {
 596       foo (&b, &c);
 597       a_5 = b + c;
 598    }
 599    a_6 = PHI (a_5, a_1);
 600    USE a_6.
 601
 602    Note that this reduces the number of computations of a = b + c to 1
 603    when we take the else edge, instead of 2.
 604 */
 605 namespace {
 606
 607 const pass_data pass_data_sink_code =
 608 {
 609   GIMPLE_PASS, /* type */
 610   "sink", /* name */
 611   OPTGROUP_NONE, /* optinfo_flags */
 612   TV_TREE_SINK, /* tv_id */
 613   /* PROP_no_crit_edges is ensured by running split_critical_edges in
 614      pass_data_sink_code::execute ().  */
 615   ( PROP_cfg | PROP_ssa ), /* properties_required */
 616   0, /* properties_provided */
 617   0, /* properties_destroyed */
 618   0, /* todo_flags_start */
 619   TODO_update_ssa, /* todo_flags_finish */
 620 };
 621
 622 class pass_sink_code : public gimple_opt_pass
 623 {
 624 public:
 625   pass_sink_code (gcc::context *ctxt)
 626     : gimple_opt_pass (pass_data_sink_code, ctxt)
 627   {}
 628
 629   /* opt_pass methods: */
 630   virtual bool gate (function *) { return flag_tree_sink != 0; }
 631   virtual unsigned int execute (function *);
 632
 633 }; // class pass_sink_code
 634
 635 unsigned int
 636 pass_sink_code::execute (function *fun)
 637 {
 638   loop_optimizer_init (LOOPS_NORMAL);
 639   split_critical_edges ();
 640   connect_infinite_loops_to_exit ();
 641   memset (&sink_stats, 0, sizeof (sink_stats));
 642   calculate_dominance_info (CDI_DOMINATORS);
 643   calculate_dominance_info (CDI_POST_DOMINATORS);
 644   sink_code_in_bb (EXIT_BLOCK_PTR_FOR_FN (fun));
 645   statistics_counter_event (fun, "Sunk statements", sink_stats.sunk);
 646   free_dominance_info (CDI_POST_DOMINATORS);
 647   remove_fake_exit_edges ();
 648   loop_optimizer_finalize ();
 649
 650   return 0;
 651 }
 652
 653 } // anon namespace
 654
 655 gimple_opt_pass *
 656 make_pass_sink_code (gcc::context *ctxt)
 657 {
 658   return new pass_sink_code (ctxt);
 659 }