gcc/tree-ssa-sink.c

   1 /* Code sinking for trees
   2    Copyright (C) 2001-2017 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 "backend.h"
  25 #include "tree.h"
  26 #include "gimple.h"
  27 #include "cfghooks.h"
  28 #include "tree-pass.h"
  29 #include "ssa.h"
  30 #include "gimple-pretty-print.h"
  31 #include "fold-const.h"
  32 #include "stor-layout.h"
  33 #include "cfganal.h"
  34 #include "gimple-iterator.h"
  35 #include "tree-cfg.h"
  36 #include "cfgloop.h"
  37 #include "params.h"
  38
  39 /* TODO:
  40    1. Sinking store only using scalar promotion (IE without moving the RHS):
  41
  42    *q = p;
  43    p = p + 1;
  44    if (something)
  45      *q = <not p>;
  46    else
  47      y = *q;
  48
  49
  50    should become
  51    sinktemp = p;
  52    p = p + 1;
  53    if (something)
  54      *q = <not p>;
  55    else
  56    {
  57      *q = sinktemp;
  58      y = *q
  59    }
  60    Store copy propagation will take care of the store elimination above.
  61
  62
  63    2. Sinking using Partial Dead Code Elimination.  */
  64
  65
  66 static struct
  67 {
  68   /* The number of statements sunk down the flowgraph by code sinking.  */
  69   int sunk;
  70
  71 } sink_stats;
  72
  73
  74 /* Given a PHI, and one of its arguments (DEF), find the edge for
  75    that argument and return it.  If the argument occurs twice in the PHI node,
  76    we return NULL.  */
  77
  78 static basic_block
  79 find_bb_for_arg (gphi *phi, tree def)
  80 {
  81   size_t i;
  82   bool foundone = false;
  83   basic_block result = NULL;
  84   for (i = 0; i < gimple_phi_num_args (phi); i++)
  85     if (PHI_ARG_DEF (phi, i) == def)
  86       {
  87         if (foundone)
  88           return NULL;
  89         foundone = true;
  90         result = gimple_phi_arg_edge (phi, i)->src;
  91       }
  92   return result;
  93 }
  94
  95 /* When the first immediate use is in a statement, then return true if all
  96    immediate uses in IMM are in the same statement.
  97    We could also do the case where  the first immediate use is in a phi node,
  98    and all the other uses are in phis in the same basic block, but this
  99    requires some expensive checking later (you have to make sure no def/vdef
 100    in the statement occurs for multiple edges in the various phi nodes it's
 101    used in, so that you only have one place you can sink it to.  */
 102
 103 static bool
 104 all_immediate_uses_same_place (def_operand_p def_p)
 105 {
 106   tree var = DEF_FROM_PTR (def_p);
 107   imm_use_iterator imm_iter;
 108   use_operand_p use_p;
 109
 110   gimple *firstuse = NULL;
 111   FOR_EACH_IMM_USE_FAST (use_p, imm_iter, var)
 112     {
 113       if (is_gimple_debug (USE_STMT (use_p)))
 114         continue;
 115       if (firstuse == NULL)
 116         firstuse = USE_STMT (use_p);
 117       else
 118         if (firstuse != USE_STMT (use_p))
 119           return false;
 120     }
 121
 122   return true;
 123 }
 124
 125 /* Find the nearest common dominator of all of the immediate uses in IMM.  */
 126
 127 static basic_block
 128 nearest_common_dominator_of_uses (def_operand_p def_p, bool *debug_stmts)
 129 {
 130   tree var = DEF_FROM_PTR (def_p);
 131   auto_bitmap blocks;
 132   basic_block commondom;
 133   unsigned int j;
 134   bitmap_iterator bi;
 135   imm_use_iterator imm_iter;
 136   use_operand_p use_p;
 137
 138   FOR_EACH_IMM_USE_FAST (use_p, imm_iter, var)
 139     {
 140       gimple *usestmt = USE_STMT (use_p);
 141       basic_block useblock;
 142
 143       if (gphi *phi = dyn_cast <gphi *> (usestmt))
 144         {
 145           int idx = PHI_ARG_INDEX_FROM_USE (use_p);
 146
 147           useblock = gimple_phi_arg_edge (phi, idx)->src;
 148         }
 149       else if (is_gimple_debug (usestmt))
 150         {
 151           *debug_stmts = true;
 152           continue;
 153         }
 154       else
 155         {
 156           useblock = gimple_bb (usestmt);
 157         }
 158
 159       /* Short circuit. Nothing dominates the entry block.  */
 160       if (useblock == ENTRY_BLOCK_PTR_FOR_FN (cfun))
 161         return NULL;
 162
 163       bitmap_set_bit (blocks, useblock->index);
 164     }
 165   commondom = BASIC_BLOCK_FOR_FN (cfun, bitmap_first_set_bit (blocks));
 166   EXECUTE_IF_SET_IN_BITMAP (blocks, 0, j, bi)
 167     commondom = nearest_common_dominator (CDI_DOMINATORS, commondom,
 168                                           BASIC_BLOCK_FOR_FN (cfun, j));
 169   return commondom;
 170 }
 171
 172 /* Given EARLY_BB and LATE_BB, two blocks in a path through the dominator
 173    tree, return the best basic block between them (inclusive) to place
 174    statements.
 175
 176    We want the most control dependent block in the shallowest loop nest.
 177
 178    If the resulting block is in a shallower loop nest, then use it.  Else
 179    only use the resulting block if it has significantly lower execution
 180    frequency than EARLY_BB to avoid gratutious statement movement.  We
 181    consider statements with VOPS more desirable to move.
 182
 183    This pass would obviously benefit from PDO as it utilizes block
 184    frequencies.  It would also benefit from recomputing frequencies
 185    if profile data is not available since frequencies often get out
 186    of sync with reality.  */
 187
 188 static basic_block
 189 select_best_block (basic_block early_bb,
 190                    basic_block late_bb,
 191                    gimple *stmt)
 192 {
 193   basic_block best_bb = late_bb;
 194   basic_block temp_bb = late_bb;
 195   int threshold;
 196
 197   while (temp_bb != early_bb)
 198     {
 199       /* If we've moved into a lower loop nest, then that becomes
 200          our best block.  */
 201       if (bb_loop_depth (temp_bb) < bb_loop_depth (best_bb))
 202         best_bb = temp_bb;
 203
 204       /* Walk up the dominator tree, hopefully we'll find a shallower
 205          loop nest.  */
 206       temp_bb = get_immediate_dominator (CDI_DOMINATORS, temp_bb);
 207     }
 208
 209   /* If we found a shallower loop nest, then we always consider that
 210      a win.  This will always give us the most control dependent block
 211      within that loop nest.  */
 212   if (bb_loop_depth (best_bb) < bb_loop_depth (early_bb))
 213     return best_bb;
 214
 215   /* Get the sinking threshold.  If the statement to be moved has memory
 216      operands, then increase the threshold by 7% as those are even more
 217      profitable to avoid, clamping at 100%.  */
 218   threshold = PARAM_VALUE (PARAM_SINK_FREQUENCY_THRESHOLD);
 219   if (gimple_vuse (stmt) || gimple_vdef (stmt))
 220     {
 221       threshold += 7;
 222       if (threshold > 100)
 223         threshold = 100;
 224     }
 225
 226   /* If BEST_BB is at the same nesting level, then require it to have
 227      significantly lower execution frequency to avoid gratutious movement.  */
 228   if (bb_loop_depth (best_bb) == bb_loop_depth (early_bb)
 229       && best_bb->count.to_frequency (cfun)
 230          < (early_bb->count.to_frequency (cfun) * threshold / 100.0))
 231     return best_bb;
 232
 233   /* No better block found, so return EARLY_BB, which happens to be the
 234      statement's original block.  */
 235   return early_bb;
 236 }
 237
 238 /* Given a statement (STMT) and the basic block it is currently in (FROMBB),
 239    determine the location to sink the statement to, if any.
 240    Returns true if there is such location; in that case, TOGSI points to the
 241    statement before that STMT should be moved.  */
 242
 243 static bool
 244 statement_sink_location (gimple *stmt, basic_block frombb,
 245                          gimple_stmt_iterator *togsi, bool *zero_uses_p)
 246 {
 247   gimple *use;
 248   use_operand_p one_use = NULL_USE_OPERAND_P;
 249   basic_block sinkbb;
 250   use_operand_p use_p;
 251   def_operand_p def_p;
 252   ssa_op_iter iter;
 253   imm_use_iterator imm_iter;
 254
 255   *zero_uses_p = false;
 256
 257   /* We only can sink assignments and non-looping const/pure calls.  */
 258   int cf;
 259   if (!is_gimple_assign (stmt)
 260       && (!is_gimple_call (stmt)
 261           || !((cf = gimple_call_flags (stmt)) & (ECF_CONST|ECF_PURE))
 262           || (cf & ECF_LOOPING_CONST_OR_PURE)))
 263     return false;
 264
 265   /* We only can sink stmts with a single definition.  */
 266   def_p = single_ssa_def_operand (stmt, SSA_OP_ALL_DEFS);
 267   if (def_p == NULL_DEF_OPERAND_P)
 268     return false;
 269
 270   /* There are a few classes of things we can't or don't move, some because we
 271      don't have code to handle it, some because it's not profitable and some
 272      because it's not legal.
 273
 274      We can't sink things that may be global stores, at least not without
 275      calculating a lot more information, because we may cause it to no longer
 276      be seen by an external routine that needs it depending on where it gets
 277      moved to.
 278
 279      We can't sink statements that end basic blocks without splitting the
 280      incoming edge for the sink location to place it there.
 281
 282      We can't sink statements that have volatile operands.
 283
 284      We don't want to sink dead code, so anything with 0 immediate uses is not
 285      sunk.
 286
 287      Don't sink BLKmode assignments if current function has any local explicit
 288      register variables, as BLKmode assignments may involve memcpy or memset
 289      calls or, on some targets, inline expansion thereof that sometimes need
 290      to use specific hard registers.
 291
 292   */
 293   if (stmt_ends_bb_p (stmt)
 294       || gimple_has_side_effects (stmt)
 295       || (cfun->has_local_explicit_reg_vars
 296           && TYPE_MODE (TREE_TYPE (gimple_get_lhs (stmt))) == BLKmode))
 297     return false;
 298
 299   /* Return if there are no immediate uses of this stmt.  */
 300   if (has_zero_uses (DEF_FROM_PTR (def_p)))
 301     {
 302       *zero_uses_p = true;
 303       return false;
 304     }
 305
 306   if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (DEF_FROM_PTR (def_p)))
 307     return false;
 308
 309   FOR_EACH_SSA_USE_OPERAND (use_p, stmt, iter, SSA_OP_ALL_USES)
 310     {
 311       tree use = USE_FROM_PTR (use_p);
 312       if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (use))
 313         return false;
 314     }
 315
 316   use = NULL;
 317
 318   /* If stmt is a store the one and only use needs to be the VOP
 319      merging PHI node.  */
 320   if (virtual_operand_p (DEF_FROM_PTR (def_p)))
 321     {
 322       FOR_EACH_IMM_USE_FAST (use_p, imm_iter, DEF_FROM_PTR (def_p))
 323         {
 324           gimple *use_stmt = USE_STMT (use_p);
 325
 326           /* A killing definition is not a use.  */
 327           if ((gimple_has_lhs (use_stmt)
 328                && operand_equal_p (gimple_get_lhs (stmt),
 329                                    gimple_get_lhs (use_stmt), 0))
 330               || stmt_kills_ref_p (use_stmt, gimple_get_lhs (stmt)))
 331             {
 332               /* If use_stmt is or might be a nop assignment then USE_STMT
 333                  acts as a use as well as definition.  */
 334               if (stmt != use_stmt
 335                   && ref_maybe_used_by_stmt_p (use_stmt,
 336                                                gimple_get_lhs (stmt)))
 337                 return false;
 338               continue;
 339             }
 340
 341           if (gimple_code (use_stmt) != GIMPLE_PHI)
 342             return false;
 343
 344           if (use
 345               && use != use_stmt)
 346             return false;
 347
 348           use = use_stmt;
 349         }
 350       if (!use)
 351         return false;
 352     }
 353   /* If all the immediate uses are not in the same place, find the nearest
 354      common dominator of all the immediate uses.  For PHI nodes, we have to
 355      find the nearest common dominator of all of the predecessor blocks, since
 356      that is where insertion would have to take place.  */
 357   else if (gimple_vuse (stmt)
 358            || !all_immediate_uses_same_place (def_p))
 359     {
 360       bool debug_stmts = false;
 361       basic_block commondom = nearest_common_dominator_of_uses (def_p,
 362                                                                 &debug_stmts);
 363
 364       if (commondom == frombb)
 365         return false;
 366
 367       /* If this is a load then do not sink past any stores.
 368          ???  This is overly simple but cheap.  We basically look
 369          for an existing load with the same VUSE in the path to one
 370          of the sink candidate blocks and we adjust commondom to the
 371          nearest to commondom.  */
 372       if (gimple_vuse (stmt))
 373         {
 374           /* Do not sink loads from hard registers.  */
 375           if (gimple_assign_single_p (stmt)
 376               && TREE_CODE (gimple_assign_rhs1 (stmt)) == VAR_DECL
 377               && DECL_HARD_REGISTER (gimple_assign_rhs1 (stmt)))
 378             return false;
 379
 380           imm_use_iterator imm_iter;
 381           use_operand_p use_p;
 382           basic_block found = NULL;
 383           FOR_EACH_IMM_USE_FAST (use_p, imm_iter, gimple_vuse (stmt))
 384             {
 385               gimple *use_stmt = USE_STMT (use_p);
 386               basic_block bb = gimple_bb (use_stmt);
 387               /* For PHI nodes the block we know sth about
 388                  is the incoming block with the use.  */
 389               if (gimple_code (use_stmt) == GIMPLE_PHI)
 390                 bb = EDGE_PRED (bb, PHI_ARG_INDEX_FROM_USE (use_p))->src;
 391               /* Any dominator of commondom would be ok with
 392                  adjusting commondom to that block.  */
 393               bb = nearest_common_dominator (CDI_DOMINATORS, bb, commondom);
 394               if (!found)
 395                 found = bb;
 396               else if (dominated_by_p (CDI_DOMINATORS, bb, found))
 397                 found = bb;
 398               /* If we can't improve, stop.  */
 399               if (found == commondom)
 400                 break;
 401             }
 402           commondom = found;
 403           if (commondom == frombb)
 404             return false;
 405         }
 406
 407       /* Our common dominator has to be dominated by frombb in order to be a
 408          trivially safe place to put this statement, since it has multiple
 409          uses.  */
 410       if (!dominated_by_p (CDI_DOMINATORS, commondom, frombb))
 411         return false;
 412
 413       commondom = select_best_block (frombb, commondom, stmt);
 414
 415       if (commondom == frombb)
 416         return false;
 417
 418       *togsi = gsi_after_labels (commondom);
 419
 420       return true;
 421     }
 422   else
 423     {
 424       FOR_EACH_IMM_USE_FAST (one_use, imm_iter, DEF_FROM_PTR (def_p))
 425         {
 426           if (is_gimple_debug (USE_STMT (one_use)))
 427             continue;
 428           break;
 429         }
 430       use = USE_STMT (one_use);
 431
 432       if (gimple_code (use) != GIMPLE_PHI)
 433         {
 434           sinkbb = gimple_bb (use);
 435           sinkbb = select_best_block (frombb, gimple_bb (use), stmt);
 436
 437           if (sinkbb == frombb)
 438             return false;
 439
 440           *togsi = gsi_for_stmt (use);
 441
 442           return true;
 443         }
 444     }
 445
 446   sinkbb = find_bb_for_arg (as_a <gphi *> (use), DEF_FROM_PTR (def_p));
 447
 448   /* This can happen if there are multiple uses in a PHI.  */
 449   if (!sinkbb)
 450     return false;
 451
 452   sinkbb = select_best_block (frombb, sinkbb, stmt);
 453   if (!sinkbb || sinkbb == frombb)
 454     return false;
 455
 456   /* If the latch block is empty, don't make it non-empty by sinking
 457      something into it.  */
 458   if (sinkbb == frombb->loop_father->latch
 459       && empty_block_p (sinkbb))
 460     return false;
 461
 462   *togsi = gsi_after_labels (sinkbb);
 463
 464   return true;
 465 }
 466
 467 /* Perform code sinking on BB */
 468
 469 static void
 470 sink_code_in_bb (basic_block bb)
 471 {
 472   basic_block son;
 473   gimple_stmt_iterator gsi;
 474   edge_iterator ei;
 475   edge e;
 476   bool last = true;
 477
 478   /* If this block doesn't dominate anything, there can't be any place to sink
 479      the statements to.  */
 480   if (first_dom_son (CDI_DOMINATORS, bb) == NULL)
 481     goto earlyout;
 482
 483   /* We can't move things across abnormal edges, so don't try.  */
 484   FOR_EACH_EDGE (e, ei, bb->succs)
 485     if (e->flags & EDGE_ABNORMAL)
 486       goto earlyout;
 487
 488   for (gsi = gsi_last_bb (bb); !gsi_end_p (gsi);)
 489     {
 490       gimple *stmt = gsi_stmt (gsi);
 491       gimple_stmt_iterator togsi;
 492       bool zero_uses_p;
 493
 494       if (!statement_sink_location (stmt, bb, &togsi, &zero_uses_p))
 495         {
 496           gimple_stmt_iterator saved = gsi;
 497           if (!gsi_end_p (gsi))
 498             gsi_prev (&gsi);
 499           /* If we face a dead stmt remove it as it possibly blocks
 500              sinking of uses.  */
 501           if (zero_uses_p
 502               && ! gimple_vdef (stmt))
 503             {
 504               gsi_remove (&saved, true);
 505               release_defs (stmt);
 506             }
 507           else
 508             last = false;
 509           continue;
 510         }
 511       if (dump_file)
 512         {
 513           fprintf (dump_file, "Sinking ");
 514           print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS);
 515           fprintf (dump_file, " from bb %d to bb %d\n",
 516                    bb->index, (gsi_bb (togsi))->index);
 517         }
 518
 519       /* Update virtual operands of statements in the path we
 520          do not sink to.  */
 521       if (gimple_vdef (stmt))
 522         {
 523           imm_use_iterator iter;
 524           use_operand_p use_p;
 525           gimple *vuse_stmt;
 526
 527           FOR_EACH_IMM_USE_STMT (vuse_stmt, iter, gimple_vdef (stmt))
 528             if (gimple_code (vuse_stmt) != GIMPLE_PHI)
 529               FOR_EACH_IMM_USE_ON_STMT (use_p, iter)
 530                 SET_USE (use_p, gimple_vuse (stmt));
 531         }
 532
 533       /* If this is the end of the basic block, we need to insert at the end
 534          of the basic block.  */
 535       if (gsi_end_p (togsi))
 536         gsi_move_to_bb_end (&gsi, gsi_bb (togsi));
 537       else
 538         gsi_move_before (&gsi, &togsi);
 539
 540       sink_stats.sunk++;
 541
 542       /* If we've just removed the last statement of the BB, the
 543          gsi_end_p() test below would fail, but gsi_prev() would have
 544          succeeded, and we want it to succeed.  So we keep track of
 545          whether we're at the last statement and pick up the new last
 546          statement.  */
 547       if (last)
 548         {
 549           gsi = gsi_last_bb (bb);
 550           continue;
 551         }
 552
 553       last = false;
 554       if (!gsi_end_p (gsi))
 555         gsi_prev (&gsi);
 556
 557     }
 558  earlyout:
 559   for (son = first_dom_son (CDI_POST_DOMINATORS, bb);
 560        son;
 561        son = next_dom_son (CDI_POST_DOMINATORS, son))
 562     {
 563       sink_code_in_bb (son);
 564     }
 565 }
 566
 567 /* Perform code sinking.
 568    This moves code down the flowgraph when we know it would be
 569    profitable to do so, or it wouldn't increase the number of
 570    executions of the statement.
 571
 572    IE given
 573
 574    a_1 = b + c;
 575    if (<something>)
 576    {
 577    }
 578    else
 579    {
 580      foo (&b, &c);
 581      a_5 = b + c;
 582    }
 583    a_6 = PHI (a_5, a_1);
 584    USE a_6.
 585
 586    we'll transform this into:
 587
 588    if (<something>)
 589    {
 590       a_1 = b + c;
 591    }
 592    else
 593    {
 594       foo (&b, &c);
 595       a_5 = b + c;
 596    }
 597    a_6 = PHI (a_5, a_1);
 598    USE a_6.
 599
 600    Note that this reduces the number of computations of a = b + c to 1
 601    when we take the else edge, instead of 2.
 602 */
 603 namespace {
 604
 605 const pass_data pass_data_sink_code =
 606 {
 607   GIMPLE_PASS, /* type */
 608   "sink", /* name */
 609   OPTGROUP_NONE, /* optinfo_flags */
 610   TV_TREE_SINK, /* tv_id */
 611   /* PROP_no_crit_edges is ensured by running split_critical_edges in
 612      pass_data_sink_code::execute ().  */
 613   ( PROP_cfg | PROP_ssa ), /* properties_required */
 614   0, /* properties_provided */
 615   0, /* properties_destroyed */
 616   0, /* todo_flags_start */
 617   TODO_update_ssa, /* todo_flags_finish */
 618 };
 619
 620 class pass_sink_code : public gimple_opt_pass
 621 {
 622 public:
 623   pass_sink_code (gcc::context *ctxt)
 624     : gimple_opt_pass (pass_data_sink_code, ctxt)
 625   {}
 626
 627   /* opt_pass methods: */
 628   virtual bool gate (function *) { return flag_tree_sink != 0; }
 629   virtual unsigned int execute (function *);
 630
 631 }; // class pass_sink_code
 632
 633 unsigned int
 634 pass_sink_code::execute (function *fun)
 635 {
 636   loop_optimizer_init (LOOPS_NORMAL);
 637   split_critical_edges ();
 638   connect_infinite_loops_to_exit ();
 639   memset (&sink_stats, 0, sizeof (sink_stats));
 640   calculate_dominance_info (CDI_DOMINATORS);
 641   calculate_dominance_info (CDI_POST_DOMINATORS);
 642   sink_code_in_bb (EXIT_BLOCK_PTR_FOR_FN (fun));
 643   statistics_counter_event (fun, "Sunk statements", sink_stats.sunk);
 644   free_dominance_info (CDI_POST_DOMINATORS);
 645   remove_fake_exit_edges ();
 646   loop_optimizer_finalize ();
 647
 648   return 0;
 649 }
 650
 651 } // anon namespace
 652
 653 gimple_opt_pass *
 654 make_pass_sink_code (gcc::context *ctxt)
 655 {
 656   return new pass_sink_code (ctxt);
 657 }