gcc/ipa-profile.c

   1 /* Basic IPA optimizations based on profile.
   2    Copyright (C) 2003-2014 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 under
   7 the terms of the GNU General Public License as published by the Free
   8 Software Foundation; either version 3, or (at your option) any later
   9 version.
  10
  11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  12 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 COPYING3.  If not see
  18 <http://www.gnu.org/licenses/>.  */
  19
  20 /* ipa-profile pass implements the following analysis propagating profille
  21    inter-procedurally.
  22
  23    - Count histogram construction.  This is a histogram analyzing how much
  24      time is spent executing statements with a given execution count read
  25      from profile feedback. This histogram is complette only with LTO,
  26      otherwise it contains information only about the current unit.
  27
  28      Similar histogram is also estimated by coverage runtime.  This histogram
  29      is not dependent on LTO, but it suffers from various defects; first
  30      gcov runtime is not weighting individual basic block by estimated execution
  31      time and second the merging of multiple runs makes assumption that the
  32      histogram distribution did not change.  Consequentely histogram constructed
  33      here may be more precise.
  34
  35      The information is used to set hot/cold thresholds.
  36    - Next speculative indirect call resolution is performed:  the local
  37      profile pass assigns profile-id to each function and provide us with a
  38      histogram specifying the most common target.  We look up the callgraph
  39      node corresponding to the target and produce a speculative call.
  40
  41      This call may or may not survive through IPA optimization based on decision
  42      of inliner.
  43    - Finally we propagate the following flags: unlikely executed, executed
  44      once, executed at startup and executed at exit.  These flags are used to
  45      control code size/performance threshold and and code placement (by producing
  46      .text.unlikely/.text.hot/.text.startup/.text.exit subsections).  */
  47 #include "config.h"
  48 #include "system.h"
  49 #include "coretypes.h"
  50 #include "tm.h"
  51 #include "tree.h"
  52 #include "cgraph.h"
  53 #include "tree-pass.h"
  54 #include "tree-ssa-alias.h"
  55 #include "internal-fn.h"
  56 #include "gimple-expr.h"
  57 #include "gimple.h"
  58 #include "gimple-iterator.h"
  59 #include "flags.h"
  60 #include "target.h"
  61 #include "tree-iterator.h"
  62 #include "ipa-utils.h"
  63 #include "profile.h"
  64 #include "params.h"
  65 #include "value-prof.h"
  66 #include "alloc-pool.h"
  67 #include "tree-inline.h"
  68 #include "lto-streamer.h"
  69 #include "data-streamer.h"
  70 #include "ipa-inline.h"
  71
  72 /* Entry in the histogram.  */
  73
  74 struct histogram_entry
  75 {
  76   gcov_type count;
  77   int time;
  78   int size;
  79 };
  80
  81 /* Histogram of profile values.
  82    The histogram is represented as an ordered vector of entries allocated via
  83    histogram_pool. During construction a separate hashtable is kept to lookup
  84    duplicate entries.  */
  85
  86 vec<histogram_entry *> histogram;
  87 static alloc_pool histogram_pool;
  88
  89 /* Hashtable support for storing SSA names hashed by their SSA_NAME_VAR.  */
  90
  91 struct histogram_hash : typed_noop_remove <histogram_entry>
  92 {
  93   typedef histogram_entry value_type;
  94   typedef histogram_entry compare_type;
  95   static inline hashval_t hash (const value_type *);
  96   static inline int equal (const value_type *, const compare_type *);
  97 };
  98
  99 inline hashval_t
 100 histogram_hash::hash (const histogram_entry *val)
 101 {
 102   return val->count;
 103 }
 104
 105 inline int
 106 histogram_hash::equal (const histogram_entry *val, const histogram_entry *val2)
 107 {
 108   return val->count == val2->count;
 109 }
 110
 111 /* Account TIME and SIZE executed COUNT times into HISTOGRAM.
 112    HASHTABLE is the on-side hash kept to avoid duplicates.  */
 113
 114 static void
 115 account_time_size (hash_table<histogram_hash> *hashtable,
 116                    vec<histogram_entry *> &histogram,
 117                    gcov_type count, int time, int size)
 118 {
 119   histogram_entry key = {count, 0, 0};
 120   histogram_entry **val = hashtable->find_slot (&key, INSERT);
 121
 122   if (!*val)
 123     {
 124       *val = (histogram_entry *) pool_alloc (histogram_pool);
 125       **val = key;
 126       histogram.safe_push (*val);
 127     }
 128   (*val)->time += time;
 129   (*val)->size += size;
 130 }
 131
 132 int
 133 cmp_counts (const void *v1, const void *v2)
 134 {
 135   const histogram_entry *h1 = *(const histogram_entry * const *)v1;
 136   const histogram_entry *h2 = *(const histogram_entry * const *)v2;
 137   if (h1->count < h2->count)
 138     return 1;
 139   if (h1->count > h2->count)
 140     return -1;
 141   return 0;
 142 }
 143
 144 /* Dump HISTOGRAM to FILE.  */
 145
 146 static void
 147 dump_histogram (FILE *file, vec<histogram_entry *> histogram)
 148 {
 149   unsigned int i;
 150   gcov_type overall_time = 0, cumulated_time = 0, cumulated_size = 0, overall_size = 0;
 151
 152   fprintf (dump_file, "Histogram:\n");
 153   for (i = 0; i < histogram.length (); i++)
 154     {
 155       overall_time += histogram[i]->count * histogram[i]->time;
 156       overall_size += histogram[i]->size;
 157     }
 158   if (!overall_time)
 159     overall_time = 1;
 160   if (!overall_size)
 161     overall_size = 1;
 162   for (i = 0; i < histogram.length (); i++)
 163     {
 164       cumulated_time += histogram[i]->count * histogram[i]->time;
 165       cumulated_size += histogram[i]->size;
 166       fprintf (file, "  %"PRId64": time:%i (%2.2f) size:%i (%2.2f)\n",
 167                (int64_t) histogram[i]->count,
 168                histogram[i]->time,
 169                cumulated_time * 100.0 / overall_time,
 170                histogram[i]->size,
 171                cumulated_size * 100.0 / overall_size);
 172    }
 173 }
 174
 175 /* Collect histogram from CFG profiles.  */
 176
 177 static void
 178 ipa_profile_generate_summary (void)
 179 {
 180   struct cgraph_node *node;
 181   gimple_stmt_iterator gsi;
 182   basic_block bb;
 183
 184   hash_table<histogram_hash> hashtable (10);
 185   histogram_pool = create_alloc_pool ("IPA histogram", sizeof (struct histogram_entry),
 186                                       10);
 187
 188   FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node)
 189     FOR_EACH_BB_FN (bb, DECL_STRUCT_FUNCTION (node->decl))
 190       {
 191         int time = 0;
 192         int size = 0;
 193         for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
 194           {
 195             gimple stmt = gsi_stmt (gsi);
 196             if (gimple_code (stmt) == GIMPLE_CALL
 197                 && !gimple_call_fndecl (stmt))
 198               {
 199                 histogram_value h;
 200                 h = gimple_histogram_value_of_type
 201                       (DECL_STRUCT_FUNCTION (node->decl),
 202                        stmt, HIST_TYPE_INDIR_CALL);
 203                 /* No need to do sanity check: gimple_ic_transform already
 204                    takes away bad histograms.  */
 205                 if (h)
 206                   {
 207                     /* counter 0 is target, counter 1 is number of execution we called target,
 208                        counter 2 is total number of executions.  */
 209                     if (h->hvalue.counters[2])
 210                       {
 211                         struct cgraph_edge * e = cgraph_edge (node, stmt);
 212                         if (e && !e->indirect_unknown_callee)
 213                           continue;
 214                         e->indirect_info->common_target_id
 215                           = h->hvalue.counters [0];
 216                         e->indirect_info->common_target_probability
 217                           = GCOV_COMPUTE_SCALE (h->hvalue.counters [1], h->hvalue.counters [2]);
 218                         if (e->indirect_info->common_target_probability > REG_BR_PROB_BASE)
 219                           {
 220                             if (dump_file)
 221                               fprintf (dump_file, "Probability capped to 1\n");
 222                             e->indirect_info->common_target_probability = REG_BR_PROB_BASE;
 223                           }
 224                       }
 225                     gimple_remove_histogram_value (DECL_STRUCT_FUNCTION (node->decl),
 226                                                     stmt, h);
 227                   }
 228               }
 229             time += estimate_num_insns (stmt, &eni_time_weights);
 230             size += estimate_num_insns (stmt, &eni_size_weights);
 231           }
 232         account_time_size (&hashtable, histogram, bb->count, time, size);
 233       }
 234   histogram.qsort (cmp_counts);
 235 }
 236
 237 /* Serialize the ipa info for lto.  */
 238
 239 static void
 240 ipa_profile_write_summary (void)
 241 {
 242   struct lto_simple_output_block *ob
 243     = lto_create_simple_output_block (LTO_section_ipa_profile);
 244   unsigned int i;
 245
 246   streamer_write_uhwi_stream (ob->main_stream, histogram.length ());
 247   for (i = 0; i < histogram.length (); i++)
 248     {
 249       streamer_write_gcov_count_stream (ob->main_stream, histogram[i]->count);
 250       streamer_write_uhwi_stream (ob->main_stream, histogram[i]->time);
 251       streamer_write_uhwi_stream (ob->main_stream, histogram[i]->size);
 252     }
 253   lto_destroy_simple_output_block (ob);
 254 }
 255
 256 /* Deserialize the ipa info for lto.  */
 257
 258 static void
 259 ipa_profile_read_summary (void)
 260 {
 261   struct lto_file_decl_data ** file_data_vec
 262     = lto_get_file_decl_data ();
 263   struct lto_file_decl_data * file_data;
 264   int j = 0;
 265
 266   hash_table<histogram_hash> hashtable (10);
 267   histogram_pool = create_alloc_pool ("IPA histogram", sizeof (struct histogram_entry),
 268                                       10);
 269
 270   while ((file_data = file_data_vec[j++]))
 271     {
 272       const char *data;
 273       size_t len;
 274       struct lto_input_block *ib
 275         = lto_create_simple_input_block (file_data,
 276                                          LTO_section_ipa_profile,
 277                                          &data, &len);
 278       if (ib)
 279         {
 280           unsigned int num = streamer_read_uhwi (ib);
 281           unsigned int n;
 282           for (n = 0; n < num; n++)
 283             {
 284               gcov_type count = streamer_read_gcov_count (ib);
 285               int time = streamer_read_uhwi (ib);
 286               int size = streamer_read_uhwi (ib);
 287               account_time_size (&hashtable, histogram,
 288                                  count, time, size);
 289             }
 290           lto_destroy_simple_input_block (file_data,
 291                                           LTO_section_ipa_profile,
 292                                           ib, data, len);
 293         }
 294     }
 295   histogram.qsort (cmp_counts);
 296 }
 297
 298 /* Data used by ipa_propagate_frequency.  */
 299
 300 struct ipa_propagate_frequency_data
 301 {
 302   bool maybe_unlikely_executed;
 303   bool maybe_executed_once;
 304   bool only_called_at_startup;
 305   bool only_called_at_exit;
 306 };
 307
 308 /* Worker for ipa_propagate_frequency_1.  */
 309
 310 static bool
 311 ipa_propagate_frequency_1 (struct cgraph_node *node, void *data)
 312 {
 313   struct ipa_propagate_frequency_data *d;
 314   struct cgraph_edge *edge;
 315
 316   d = (struct ipa_propagate_frequency_data *)data;
 317   for (edge = node->callers;
 318        edge && (d->maybe_unlikely_executed || d->maybe_executed_once
 319                 || d->only_called_at_startup || d->only_called_at_exit);
 320        edge = edge->next_caller)
 321     {
 322       if (edge->caller != node)
 323         {
 324           d->only_called_at_startup &= edge->caller->only_called_at_startup;
 325           /* It makes sense to put main() together with the static constructors.
 326              It will be executed for sure, but rest of functions called from
 327              main are definitely not at startup only.  */
 328           if (MAIN_NAME_P (DECL_NAME (edge->caller->decl)))
 329             d->only_called_at_startup = 0;
 330           d->only_called_at_exit &= edge->caller->only_called_at_exit;
 331         }
 332
 333       /* When profile feedback is available, do not try to propagate too hard;
 334          counts are already good guide on function frequencies and roundoff
 335          errors can make us to push function into unlikely section even when
 336          it is executed by the train run.  Transfer the function only if all
 337          callers are unlikely executed.  */
 338       if (profile_info && flag_branch_probabilities
 339           && (edge->caller->frequency != NODE_FREQUENCY_UNLIKELY_EXECUTED
 340               || (edge->caller->global.inlined_to
 341                   && edge->caller->global.inlined_to->frequency
 342                      != NODE_FREQUENCY_UNLIKELY_EXECUTED)))
 343           d->maybe_unlikely_executed = false;
 344       if (!edge->frequency)
 345         continue;
 346       switch (edge->caller->frequency)
 347         {
 348         case NODE_FREQUENCY_UNLIKELY_EXECUTED:
 349           break;
 350         case NODE_FREQUENCY_EXECUTED_ONCE:
 351           if (dump_file && (dump_flags & TDF_DETAILS))
 352             fprintf (dump_file, "  Called by %s that is executed once\n",
 353                      edge->caller->name ());
 354           d->maybe_unlikely_executed = false;
 355           if (inline_edge_summary (edge)->loop_depth)
 356             {
 357               d->maybe_executed_once = false;
 358               if (dump_file && (dump_flags & TDF_DETAILS))
 359                 fprintf (dump_file, "  Called in loop\n");
 360             }
 361           break;
 362         case NODE_FREQUENCY_HOT:
 363         case NODE_FREQUENCY_NORMAL:
 364           if (dump_file && (dump_flags & TDF_DETAILS))
 365             fprintf (dump_file, "  Called by %s that is normal or hot\n",
 366                      edge->caller->name ());
 367           d->maybe_unlikely_executed = false;
 368           d->maybe_executed_once = false;
 369           break;
 370         }
 371     }
 372   return edge != NULL;
 373 }
 374
 375 /* Return ture if NODE contains hot calls.  */
 376
 377 bool
 378 contains_hot_call_p (struct cgraph_node *node)
 379 {
 380   struct cgraph_edge *e;
 381   for (e = node->callees; e; e = e->next_callee)
 382     if (cgraph_maybe_hot_edge_p (e))
 383       return true;
 384     else if (!e->inline_failed
 385              && contains_hot_call_p (e->callee))
 386       return true;
 387   for (e = node->indirect_calls; e; e = e->next_callee)
 388     if (cgraph_maybe_hot_edge_p (e))
 389       return true;
 390   return false;
 391 }
 392
 393 /* See if the frequency of NODE can be updated based on frequencies of its
 394    callers.  */
 395 bool
 396 ipa_propagate_frequency (struct cgraph_node *node)
 397 {
 398   struct ipa_propagate_frequency_data d = {true, true, true, true};
 399   bool changed = false;
 400
 401   /* We can not propagate anything useful about externally visible functions
 402      nor about virtuals.  */
 403   if (!node->local.local
 404       || node->alias
 405       || (flag_devirtualize && DECL_VIRTUAL_P (node->decl)))
 406     return false;
 407   gcc_assert (node->analyzed);
 408   if (dump_file && (dump_flags & TDF_DETAILS))
 409     fprintf (dump_file, "Processing frequency %s\n", node->name ());
 410
 411   cgraph_for_node_and_aliases (node, ipa_propagate_frequency_1, &d, true);
 412
 413   if ((d.only_called_at_startup && !d.only_called_at_exit)
 414       && !node->only_called_at_startup)
 415     {
 416        node->only_called_at_startup = true;
 417        if (dump_file)
 418          fprintf (dump_file, "Node %s promoted to only called at startup.\n",
 419                   node->name ());
 420        changed = true;
 421     }
 422   if ((d.only_called_at_exit && !d.only_called_at_startup)
 423       && !node->only_called_at_exit)
 424     {
 425        node->only_called_at_exit = true;
 426        if (dump_file)
 427          fprintf (dump_file, "Node %s promoted to only called at exit.\n",
 428                   node->name ());
 429        changed = true;
 430     }
 431
 432   /* With profile we can decide on hot/normal based on count.  */
 433   if (node->count)
 434     {
 435       bool hot = false;
 436       if (node->count >= get_hot_bb_threshold ())
 437         hot = true;
 438       if (!hot)
 439         hot |= contains_hot_call_p (node);
 440       if (hot)
 441         {
 442           if (node->frequency != NODE_FREQUENCY_HOT)
 443             {
 444               if (dump_file)
 445                 fprintf (dump_file, "Node %s promoted to hot.\n",
 446                          node->name ());
 447               node->frequency = NODE_FREQUENCY_HOT;
 448               return true;
 449             }
 450           return false;
 451         }
 452       else if (node->frequency == NODE_FREQUENCY_HOT)
 453         {
 454           if (dump_file)
 455             fprintf (dump_file, "Node %s reduced to normal.\n",
 456                      node->name ());
 457           node->frequency = NODE_FREQUENCY_NORMAL;
 458           changed = true;
 459         }
 460     }
 461   /* These come either from profile or user hints; never update them.  */
 462   if (node->frequency == NODE_FREQUENCY_HOT
 463       || node->frequency == NODE_FREQUENCY_UNLIKELY_EXECUTED)
 464     return changed;
 465   if (d.maybe_unlikely_executed)
 466     {
 467       node->frequency = NODE_FREQUENCY_UNLIKELY_EXECUTED;
 468       if (dump_file)
 469         fprintf (dump_file, "Node %s promoted to unlikely executed.\n",
 470                  node->name ());
 471       changed = true;
 472     }
 473   else if (d.maybe_executed_once && node->frequency != NODE_FREQUENCY_EXECUTED_ONCE)
 474     {
 475       node->frequency = NODE_FREQUENCY_EXECUTED_ONCE;
 476       if (dump_file)
 477         fprintf (dump_file, "Node %s promoted to executed once.\n",
 478                  node->name ());
 479       changed = true;
 480     }
 481   return changed;
 482 }
 483
 484 /* Simple ipa profile pass propagating frequencies across the callgraph.  */
 485
 486 static unsigned int
 487 ipa_profile (void)
 488 {
 489   struct cgraph_node **order;
 490   struct cgraph_edge *e;
 491   int order_pos;
 492   bool something_changed = false;
 493   int i;
 494   gcov_type overall_time = 0, cutoff = 0, cumulated = 0, overall_size = 0;
 495   struct cgraph_node *n,*n2;
 496   int nindirect = 0, ncommon = 0, nunknown = 0, nuseless = 0, nconverted = 0;
 497   bool node_map_initialized = false;
 498
 499   if (dump_file)
 500     dump_histogram (dump_file, histogram);
 501   for (i = 0; i < (int)histogram.length (); i++)
 502     {
 503       overall_time += histogram[i]->count * histogram[i]->time;
 504       overall_size += histogram[i]->size;
 505     }
 506   if (overall_time)
 507     {
 508       gcov_type threshold;
 509
 510       gcc_assert (overall_size);
 511       if (dump_file)
 512         {
 513           gcov_type min, cumulated_time = 0, cumulated_size = 0;
 514
 515           fprintf (dump_file, "Overall time: %"PRId64"\n",
 516                    (int64_t)overall_time);
 517           min = get_hot_bb_threshold ();
 518           for (i = 0; i < (int)histogram.length () && histogram[i]->count >= min;
 519                i++)
 520             {
 521               cumulated_time += histogram[i]->count * histogram[i]->time;
 522               cumulated_size += histogram[i]->size;
 523             }
 524           fprintf (dump_file, "GCOV min count: %"PRId64
 525                    " Time:%3.2f%% Size:%3.2f%%\n",
 526                    (int64_t)min,
 527                    cumulated_time * 100.0 / overall_time,
 528                    cumulated_size * 100.0 / overall_size);
 529         }
 530       cutoff = (overall_time * PARAM_VALUE (HOT_BB_COUNT_WS_PERMILLE) + 500) / 1000;
 531       threshold = 0;
 532       for (i = 0; cumulated < cutoff; i++)
 533         {
 534           cumulated += histogram[i]->count * histogram[i]->time;
 535           threshold = histogram[i]->count;
 536         }
 537       if (!threshold)
 538         threshold = 1;
 539       if (dump_file)
 540         {
 541           gcov_type cumulated_time = 0, cumulated_size = 0;
 542
 543           for (i = 0;
 544                i < (int)histogram.length () && histogram[i]->count >= threshold;
 545                i++)
 546             {
 547               cumulated_time += histogram[i]->count * histogram[i]->time;
 548               cumulated_size += histogram[i]->size;
 549             }
 550           fprintf (dump_file, "Determined min count: %"PRId64
 551                    " Time:%3.2f%% Size:%3.2f%%\n",
 552                    (int64_t)threshold,
 553                    cumulated_time * 100.0 / overall_time,
 554                    cumulated_size * 100.0 / overall_size);
 555         }
 556       if (threshold > get_hot_bb_threshold ()
 557           || in_lto_p)
 558         {
 559           if (dump_file)
 560             fprintf (dump_file, "Threshold updated.\n");
 561           set_hot_bb_threshold (threshold);
 562         }
 563     }
 564   histogram.release ();
 565   free_alloc_pool (histogram_pool);
 566
 567   /* Produce speculative calls: we saved common traget from porfiling into
 568      e->common_target_id.  Now, at link time, we can look up corresponding
 569      function node and produce speculative call.  */
 570
 571   FOR_EACH_DEFINED_FUNCTION (n)
 572     {
 573       bool update = false;
 574
 575       for (e = n->indirect_calls; e; e = e->next_callee)
 576         {
 577           if (n->count)
 578             nindirect++;
 579           if (e->indirect_info->common_target_id)
 580             {
 581               if (!node_map_initialized)
 582                 init_node_map (false);
 583               node_map_initialized = true;
 584               ncommon++;
 585               n2 = find_func_by_profile_id (e->indirect_info->common_target_id);
 586               if (n2)
 587                 {
 588                   if (dump_file)
 589                     {
 590                       fprintf (dump_file, "Indirect call -> direct call from"
 591                                " other module %s/%i => %s/%i, prob %3.2f\n",
 592                                xstrdup (n->name ()), n->order,
 593                                xstrdup (n2->name ()), n2->order,
 594                                e->indirect_info->common_target_probability
 595                                / (float)REG_BR_PROB_BASE);
 596                     }
 597                   if (e->indirect_info->common_target_probability
 598                       < REG_BR_PROB_BASE / 2)
 599                     {
 600                       nuseless++;
 601                       if (dump_file)
 602                         fprintf (dump_file,
 603                                  "Not speculating: probability is too low.\n");
 604                     }
 605                   else if (!cgraph_maybe_hot_edge_p (e))
 606                     {
 607                       nuseless++;
 608                       if (dump_file)
 609                         fprintf (dump_file,
 610                                  "Not speculating: call is cold.\n");
 611                     }
 612                   else if (cgraph_function_body_availability (n2)
 613                            <= AVAIL_OVERWRITABLE
 614                            && symtab_can_be_discarded (n2))
 615                     {
 616                       nuseless++;
 617                       if (dump_file)
 618                         fprintf (dump_file,
 619                                  "Not speculating: target is overwritable "
 620                                  "and can be discarded.\n");
 621                     }
 622                   else
 623                     {
 624                       /* Target may be overwritable, but profile says that
 625                          control flow goes to this particular implementation
 626                          of N2.  Speculate on the local alias to allow inlining.
 627                        */
 628                       if (!symtab_can_be_discarded (n2))
 629                         {
 630                           cgraph_node *alias;
 631                           alias = cgraph (symtab_nonoverwritable_alias
 632                                            (n2));
 633                           if (alias)
 634                             n2 = alias;
 635                         }
 636                       nconverted++;
 637                       cgraph_turn_edge_to_speculative
 638                         (e, n2,
 639                          apply_scale (e->count,
 640                                       e->indirect_info->common_target_probability),
 641                          apply_scale (e->frequency,
 642                                       e->indirect_info->common_target_probability));
 643                       update = true;
 644                     }
 645                 }
 646               else
 647                 {
 648                   if (dump_file)
 649                     fprintf (dump_file, "Function with profile-id %i not found.\n",
 650                              e->indirect_info->common_target_id);
 651                   nunknown++;
 652                 }
 653             }
 654          }
 655        if (update)
 656          inline_update_overall_summary (n);
 657      }
 658   if (node_map_initialized)
 659     del_node_map ();
 660   if (dump_file && nindirect)
 661     fprintf (dump_file,
 662              "%i indirect calls trained.\n"
 663              "%i (%3.2f%%) have common target.\n"
 664              "%i (%3.2f%%) targets was not found.\n"
 665              "%i (%3.2f%%) speculations seems useless.\n"
 666              "%i (%3.2f%%) speculations produced.\n",
 667              nindirect,
 668              ncommon, ncommon * 100.0 / nindirect,
 669              nunknown, nunknown * 100.0 / nindirect,
 670              nuseless, nuseless * 100.0 / nindirect,
 671              nconverted, nconverted * 100.0 / nindirect);
 672
 673   order = XCNEWVEC (struct cgraph_node *, cgraph_n_nodes);
 674   order_pos = ipa_reverse_postorder (order);
 675   for (i = order_pos - 1; i >= 0; i--)
 676     {
 677       if (order[i]->local.local && ipa_propagate_frequency (order[i]))
 678         {
 679           for (e = order[i]->callees; e; e = e->next_callee)
 680             if (e->callee->local.local && !e->callee->aux)
 681               {
 682                 something_changed = true;
 683                 e->callee->aux = (void *)1;
 684               }
 685         }
 686       order[i]->aux = NULL;
 687     }
 688
 689   while (something_changed)
 690     {
 691       something_changed = false;
 692       for (i = order_pos - 1; i >= 0; i--)
 693         {
 694           if (order[i]->aux && ipa_propagate_frequency (order[i]))
 695             {
 696               for (e = order[i]->callees; e; e = e->next_callee)
 697                 if (e->callee->local.local && !e->callee->aux)
 698                   {
 699                     something_changed = true;
 700                     e->callee->aux = (void *)1;
 701                   }
 702             }
 703           order[i]->aux = NULL;
 704         }
 705     }
 706   free (order);
 707   return 0;
 708 }
 709
 710 namespace {
 711
 712 const pass_data pass_data_ipa_profile =
 713 {
 714   IPA_PASS, /* type */
 715   "profile_estimate", /* name */
 716   OPTGROUP_NONE, /* optinfo_flags */
 717   true, /* has_execute */
 718   TV_IPA_PROFILE, /* tv_id */
 719   0, /* properties_required */
 720   0, /* properties_provided */
 721   0, /* properties_destroyed */
 722   0, /* todo_flags_start */
 723   0, /* todo_flags_finish */
 724 };
 725
 726 class pass_ipa_profile : public ipa_opt_pass_d
 727 {
 728 public:
 729   pass_ipa_profile (gcc::context *ctxt)
 730     : ipa_opt_pass_d (pass_data_ipa_profile, ctxt,
 731                       ipa_profile_generate_summary, /* generate_summary */
 732                       ipa_profile_write_summary, /* write_summary */
 733                       ipa_profile_read_summary, /* read_summary */
 734                       NULL, /* write_optimization_summary */
 735                       NULL, /* read_optimization_summary */
 736                       NULL, /* stmt_fixup */
 737                       0, /* function_transform_todo_flags_start */
 738                       NULL, /* function_transform */
 739                       NULL) /* variable_transform */
 740   {}
 741
 742   /* opt_pass methods: */
 743   virtual bool gate (function *) { return flag_ipa_profile; }
 744   virtual unsigned int execute (function *) { return ipa_profile (); }
 745
 746 }; // class pass_ipa_profile
 747
 748 } // anon namespace
 749
 750 ipa_opt_pass_d *
 751 make_pass_ipa_profile (gcc::context *ctxt)
 752 {
 753   return new pass_ipa_profile (ctxt);
 754 }