gcc/tree-ssa-dse.c

   1 /* Dead store elimination
   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
   7 it under the terms of the GNU General Public License as published by
   8 the Free Software Foundation; either version 2, or (at your option)
   9 any later version.
  10
  11 GCC is distributed in the hope that it will be useful,
  12 but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 GNU General Public License 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
  18 the Free Software Foundation, 51 Franklin Street, Fifth Floor,
  19 Boston, MA 02110-1301, USA.  */
  20
  21 #include "config.h"
  22 #include "system.h"
  23 #include "coretypes.h"
  24 #include "tm.h"
  25 #include "ggc.h"
  26 #include "tree.h"
  27 #include "rtl.h"
  28 #include "tm_p.h"
  29 #include "basic-block.h"
  30 #include "timevar.h"
  31 #include "diagnostic.h"
  32 #include "tree-flow.h"
  33 #include "tree-pass.h"
  34 #include "tree-dump.h"
  35 #include "domwalk.h"
  36 #include "flags.h"
  37
  38 /* This file implements dead store elimination.
  39
  40    A dead store is a store into a memory location which will later be
  41    overwritten by another store without any intervening loads.  In this
  42    case the earlier store can be deleted.
  43
  44    In our SSA + virtual operand world we use immediate uses of virtual
  45    operands to detect dead stores.  If a store's virtual definition
  46    is used precisely once by a later store to the same location which
  47    post dominates the first store, then the first store is dead.
  48
  49    The single use of the store's virtual definition ensures that
  50    there are no intervening aliased loads and the requirement that
  51    the second load post dominate the first ensures that if the earlier
  52    store executes, then the later stores will execute before the function
  53    exits.
  54
  55    It may help to think of this as first moving the earlier store to
  56    the point immediately before the later store.  Again, the single
  57    use of the virtual definition and the post-dominance relationship
  58    ensure that such movement would be safe.  Clearly if there are
  59    back to back stores, then the second is redundant.
  60
  61    Reviewing section 10.7.2 in Morgan's "Building an Optimizing Compiler"
  62    may also help in understanding this code since it discusses the
  63    relationship between dead store and redundant load elimination.  In
  64    fact, they are the same transformation applied to different views of
  65    the CFG.  */
  66
  67
  68 struct dse_global_data
  69 {
  70   /* This is the global bitmap for store statements.
  71
  72      Each statement has a unique ID.  When we encounter a store statement
  73      that we want to record, set the bit corresponding to the statement's
  74      unique ID in this bitmap.  */
  75   bitmap stores;
  76 };
  77
  78 /* We allocate a bitmap-per-block for stores which are encountered
  79    during the scan of that block.  This allows us to restore the
  80    global bitmap of stores when we finish processing a block.  */
  81 struct dse_block_local_data
  82 {
  83   bitmap stores;
  84 };
  85
  86 static bool gate_dse (void);
  87 static void tree_ssa_dse (void);
  88 static void dse_initialize_block_local_data (struct dom_walk_data *,
  89                                              basic_block,
  90                                              bool);
  91 static void dse_optimize_stmt (struct dom_walk_data *,
  92                                basic_block,
  93                                block_stmt_iterator);
  94 static void dse_record_phis (struct dom_walk_data *, basic_block);
  95 static void dse_finalize_block (struct dom_walk_data *, basic_block);
  96 static void record_voperand_set (bitmap, bitmap *, unsigned int);
  97
  98 static unsigned max_stmt_uid;   /* Maximal uid of a statement.  Uids to phi
  99                                    nodes are assigned using the versions of
 100                                    ssa names they define.  */
 101
 102 /* Returns uid of statement STMT.  */
 103
 104 static unsigned
 105 get_stmt_uid (tree stmt)
 106 {
 107   if (TREE_CODE (stmt) == PHI_NODE)
 108     return SSA_NAME_VERSION (PHI_RESULT (stmt)) + max_stmt_uid;
 109
 110   return stmt_ann (stmt)->uid;
 111 }
 112
 113 /* Set bit UID in bitmaps GLOBAL and *LOCAL, creating *LOCAL as needed.  */
 114
 115 static void
 116 record_voperand_set (bitmap global, bitmap *local, unsigned int uid)
 117 {
 118   /* Lazily allocate the bitmap.  Note that we do not get a notification
 119      when the block local data structures die, so we allocate the local
 120      bitmap backed by the GC system.  */
 121   if (*local == NULL)
 122     *local = BITMAP_GGC_ALLOC ();
 123
 124   /* Set the bit in the local and global bitmaps.  */
 125   bitmap_set_bit (*local, uid);
 126   bitmap_set_bit (global, uid);
 127 }
 128
 129 /* Initialize block local data structures.  */
 130
 131 static void
 132 dse_initialize_block_local_data (struct dom_walk_data *walk_data,
 133                                  basic_block bb ATTRIBUTE_UNUSED,
 134                                  bool recycled)
 135 {
 136   struct dse_block_local_data *bd
 137     = VEC_last (void_p, walk_data->block_data_stack);
 138
 139   /* If we are given a recycled block local data structure, ensure any
 140      bitmap associated with the block is cleared.  */
 141   if (recycled)
 142     {
 143       if (bd->stores)
 144         bitmap_clear (bd->stores);
 145     }
 146 }
 147
 148 /* Attempt to eliminate dead stores in the statement referenced by BSI.
 149
 150    A dead store is a store into a memory location which will later be
 151    overwritten by another store without any intervening loads.  In this
 152    case the earlier store can be deleted.
 153
 154    In our SSA + virtual operand world we use immediate uses of virtual
 155    operands to detect dead stores.  If a store's virtual definition
 156    is used precisely once by a later store to the same location which
 157    post dominates the first store, then the first store is dead.  */
 158
 159 static void
 160 dse_optimize_stmt (struct dom_walk_data *walk_data,
 161                    basic_block bb ATTRIBUTE_UNUSED,
 162                    block_stmt_iterator bsi)
 163 {
 164   struct dse_block_local_data *bd
 165     = VEC_last (void_p, walk_data->block_data_stack);
 166   struct dse_global_data *dse_gd = walk_data->global_data;
 167   tree stmt = bsi_stmt (bsi);
 168   stmt_ann_t ann = stmt_ann (stmt);
 169
 170   /* If this statement has no virtual defs, then there is nothing
 171      to do.  */
 172   if (ZERO_SSA_OPERANDS (stmt, (SSA_OP_VMAYDEF|SSA_OP_VMUSTDEF)))
 173     return;
 174
 175   /* We know we have virtual definitions.  If this is a MODIFY_EXPR that's
 176      not also a function call, then record it into our table.  */
 177   if (get_call_expr_in (stmt))
 178     return;
 179
 180   if (ann->has_volatile_ops)
 181     return;
 182
 183   if (TREE_CODE (stmt) == MODIFY_EXPR)
 184     {
 185       use_operand_p first_use_p = NULL_USE_OPERAND_P;
 186       use_operand_p use_p = NULL;
 187       tree use, use_stmt, temp;
 188       tree defvar = NULL_TREE, usevar = NULL_TREE;
 189       bool fail = false;
 190       use_operand_p var2;
 191       def_operand_p var1;
 192       ssa_op_iter op_iter;
 193
 194       /* We want to verify that each virtual definition in STMT has
 195          precisely one use and that all the virtual definitions are
 196          used by the same single statement.  When complete, we
 197          want USE_STMT to refer to the one statement which uses
 198          all of the virtual definitions from STMT.  */
 199       use_stmt = NULL;
 200       FOR_EACH_SSA_MUST_AND_MAY_DEF_OPERAND (var1, var2, stmt, op_iter)
 201         {
 202           defvar = DEF_FROM_PTR (var1);
 203           usevar = USE_FROM_PTR (var2);
 204
 205           /* If this virtual def does not have precisely one use, then
 206              we will not be able to eliminate STMT.  */
 207           if (num_imm_uses (defvar) != 1)
 208             {
 209               fail = true;
 210               break;
 211             }
 212
 213           /* Get the one and only immediate use of DEFVAR.  */
 214           single_imm_use (defvar, &use_p, &temp);
 215           gcc_assert (use_p != NULL_USE_OPERAND_P);
 216           first_use_p = use_p;
 217           use = USE_FROM_PTR (use_p);
 218
 219           /* If the immediate use of DEF_VAR is not the same as the
 220              previously find immediate uses, then we will not be able
 221              to eliminate STMT.  */
 222           if (use_stmt == NULL)
 223             use_stmt = temp;
 224           else if (temp != use_stmt)
 225             {
 226               fail = true;
 227               break;
 228             }
 229         }
 230
 231       if (fail)
 232         {
 233           record_voperand_set (dse_gd->stores, &bd->stores, ann->uid);
 234           return;
 235         }
 236
 237       /* Skip through any PHI nodes we have already seen if the PHI
 238          represents the only use of this store.
 239
 240          Note this does not handle the case where the store has
 241          multiple V_{MAY,MUST}_DEFs which all reach a set of PHI nodes in the
 242          same block.  */
 243       while (use_p != NULL_USE_OPERAND_P
 244              && TREE_CODE (use_stmt) == PHI_NODE
 245              && bitmap_bit_p (dse_gd->stores, get_stmt_uid (use_stmt)))
 246         {
 247           /* Skip past this PHI and loop again in case we had a PHI
 248              chain.  */
 249           if (single_imm_use (PHI_RESULT (use_stmt), &use_p, &use_stmt))
 250             use = USE_FROM_PTR (use_p);
 251         }
 252
 253       /* If we have precisely one immediate use at this point, then we may
 254          have found redundant store.  */
 255       if (use_p != NULL_USE_OPERAND_P
 256           && bitmap_bit_p (dse_gd->stores, get_stmt_uid (use_stmt))
 257           && operand_equal_p (TREE_OPERAND (stmt, 0),
 258                               TREE_OPERAND (use_stmt, 0), 0))
 259         {
 260           tree def;
 261           ssa_op_iter iter;
 262
 263           /* Make sure we propagate the ABNORMAL bit setting.  */
 264           if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (USE_FROM_PTR (first_use_p)))
 265             SSA_NAME_OCCURS_IN_ABNORMAL_PHI (usevar) = 1;
 266           /* Then we need to fix the operand of the consuming stmt.  */
 267           SET_USE (first_use_p, usevar);
 268
 269           if (dump_file && (dump_flags & TDF_DETAILS))
 270             {
 271               fprintf (dump_file, "  Deleted dead store '");
 272               print_generic_expr (dump_file, bsi_stmt (bsi), dump_flags);
 273               fprintf (dump_file, "'\n");
 274             }
 275
 276           /* Remove the dead store.  */
 277           bsi_remove (&bsi);
 278
 279           /* The virtual defs for the dead statement will need to be
 280              updated.  Since these names are going to disappear,
 281              FUD chains for uses downstream need to be updated.  */
 282           FOR_EACH_SSA_TREE_OPERAND (def, stmt, iter, SSA_OP_VIRTUAL_DEFS)
 283             mark_sym_for_renaming (SSA_NAME_VAR (def));
 284
 285           /* And release any SSA_NAMEs set in this statement back to the
 286              SSA_NAME manager.  */
 287           release_defs (stmt);
 288         }
 289
 290       record_voperand_set (dse_gd->stores, &bd->stores, ann->uid);
 291     }
 292 }
 293
 294 /* Record that we have seen the PHIs at the start of BB which correspond
 295    to virtual operands.  */
 296 static void
 297 dse_record_phis (struct dom_walk_data *walk_data, basic_block bb)
 298 {
 299   struct dse_block_local_data *bd
 300     = VEC_last (void_p, walk_data->block_data_stack);
 301   struct dse_global_data *dse_gd = walk_data->global_data;
 302   tree phi;
 303
 304   for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
 305     if (!is_gimple_reg (PHI_RESULT (phi)))
 306       record_voperand_set (dse_gd->stores,
 307                            &bd->stores,
 308                            get_stmt_uid (phi));
 309 }
 310
 311 static void
 312 dse_finalize_block (struct dom_walk_data *walk_data,
 313                     basic_block bb ATTRIBUTE_UNUSED)
 314 {
 315   struct dse_block_local_data *bd
 316     = VEC_last (void_p, walk_data->block_data_stack);
 317   struct dse_global_data *dse_gd = walk_data->global_data;
 318   bitmap stores = dse_gd->stores;
 319   unsigned int i;
 320   bitmap_iterator bi;
 321
 322   /* Unwind the stores noted in this basic block.  */
 323   if (bd->stores)
 324     EXECUTE_IF_SET_IN_BITMAP (bd->stores, 0, i, bi)
 325       {
 326         bitmap_clear_bit (stores, i);
 327       }
 328 }
 329
 330 static void
 331 tree_ssa_dse (void)
 332 {
 333   struct dom_walk_data walk_data;
 334   struct dse_global_data dse_gd;
 335   basic_block bb;
 336
 337   /* Create a UID for each statement in the function.  Ordering of the
 338      UIDs is not important for this pass.  */
 339   max_stmt_uid = 0;
 340   FOR_EACH_BB (bb)
 341     {
 342       block_stmt_iterator bsi;
 343
 344       for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
 345         stmt_ann (bsi_stmt (bsi))->uid = max_stmt_uid++;
 346     }
 347
 348   /* We might consider making this a property of each pass so that it
 349      can be [re]computed on an as-needed basis.  Particularly since
 350      this pass could be seen as an extension of DCE which needs post
 351      dominators.  */
 352   calculate_dominance_info (CDI_POST_DOMINATORS);
 353
 354   /* Dead store elimination is fundamentally a walk of the post-dominator
 355      tree and a backwards walk of statements within each block.  */
 356   walk_data.walk_stmts_backward = true;
 357   walk_data.dom_direction = CDI_POST_DOMINATORS;
 358   walk_data.initialize_block_local_data = dse_initialize_block_local_data;
 359   walk_data.before_dom_children_before_stmts = NULL;
 360   walk_data.before_dom_children_walk_stmts = dse_optimize_stmt;
 361   walk_data.before_dom_children_after_stmts = dse_record_phis;
 362   walk_data.after_dom_children_before_stmts = NULL;
 363   walk_data.after_dom_children_walk_stmts = NULL;
 364   walk_data.after_dom_children_after_stmts = dse_finalize_block;
 365   walk_data.interesting_blocks = NULL;
 366
 367   walk_data.block_local_data_size = sizeof (struct dse_block_local_data);
 368
 369   /* This is the main hash table for the dead store elimination pass.  */
 370   dse_gd.stores = BITMAP_ALLOC (NULL);
 371   walk_data.global_data = &dse_gd;
 372
 373   /* Initialize the dominator walker.  */
 374   init_walk_dominator_tree (&walk_data);
 375
 376   /* Recursively walk the dominator tree.  */
 377   walk_dominator_tree (&walk_data, EXIT_BLOCK_PTR);
 378
 379   /* Finalize the dominator walker.  */
 380   fini_walk_dominator_tree (&walk_data);
 381
 382   /* Release the main bitmap.  */
 383   BITMAP_FREE (dse_gd.stores);
 384
 385   /* For now, just wipe the post-dominator information.  */
 386   free_dominance_info (CDI_POST_DOMINATORS);
 387 }
 388
 389 static bool
 390 gate_dse (void)
 391 {
 392   return flag_tree_dse != 0;
 393 }
 394
 395 struct tree_opt_pass pass_dse = {
 396   "dse",                        /* name */
 397   gate_dse,                     /* gate */
 398   tree_ssa_dse,                 /* execute */
 399   NULL,                         /* sub */
 400   NULL,                         /* next */
 401   0,                            /* static_pass_number */
 402   TV_TREE_DSE,                  /* tv_id */
 403   PROP_cfg
 404     | PROP_ssa
 405     | PROP_alias,               /* properties_required */
 406   0,                            /* properties_provided */
 407   0,                            /* properties_destroyed */
 408   0,                            /* todo_flags_start */
 409   TODO_dump_func
 410     | TODO_ggc_collect
 411     | TODO_update_ssa
 412     | TODO_verify_ssa,          /* todo_flags_finish */
 413   0                             /* letter */
 414 };