gcc/ipa-inline-transform.c

   1 /* Callgraph transformations to handle inlining
   2    Copyright (C) 2003-2014 Free Software Foundation, Inc.
   3    Contributed by Jan Hubicka
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 3, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 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 /* The inline decisions are stored in callgraph in "inline plan" and
  22    applied later.
  23
  24    To mark given call inline, use inline_call function.
  25    The function marks the edge inlinable and, if necessary, produces
  26    virtual clone in the callgraph representing the new copy of callee's
  27    function body.
  28
  29    The inline plan is applied on given function body by inline_transform.  */
  30
  31 #include "config.h"
  32 #include "system.h"
  33 #include "coretypes.h"
  34 #include "tm.h"
  35 #include "tree.h"
  36 #include "langhooks.h"
  37 #include "intl.h"
  38 #include "coverage.h"
  39 #include "ggc.h"
  40 #include "tree-cfg.h"
  41 #include "ipa-prop.h"
  42 #include "ipa-inline.h"
  43 #include "tree-inline.h"
  44 #include "tree-pass.h"
  45
  46 int ncalls_inlined;
  47 int nfunctions_inlined;
  48 bool speculation_removed;
  49
  50 /* Scale frequency of NODE edges by FREQ_SCALE.  */
  51
  52 static void
  53 update_noncloned_frequencies (struct cgraph_node *node,
  54                               int freq_scale)
  55 {
  56   struct cgraph_edge *e;
  57
  58   /* We do not want to ignore high loop nest after freq drops to 0.  */
  59   if (!freq_scale)
  60     freq_scale = 1;
  61   for (e = node->callees; e; e = e->next_callee)
  62     {
  63       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  64       if (e->frequency > CGRAPH_FREQ_MAX)
  65         e->frequency = CGRAPH_FREQ_MAX;
  66       if (!e->inline_failed)
  67         update_noncloned_frequencies (e->callee, freq_scale);
  68     }
  69   for (e = node->indirect_calls; e; e = e->next_callee)
  70     {
  71       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  72       if (e->frequency > CGRAPH_FREQ_MAX)
  73         e->frequency = CGRAPH_FREQ_MAX;
  74     }
  75 }
  76
  77 /* We removed or are going to remove the last call to NODE.
  78    Return true if we can and want proactively remove the NODE now.
  79    This is important to do, since we want inliner to know when offline
  80    copy of function was removed.  */
  81
  82 static bool
  83 can_remove_node_now_p_1 (struct cgraph_node *node)
  84 {
  85   /* FIXME: When address is taken of DECL_EXTERNAL function we still
  86      can remove its offline copy, but we would need to keep unanalyzed node in
  87      the callgraph so references can point to it.  */
  88   return (!node->address_taken
  89           && !node->has_aliases_p ()
  90           && !node->used_as_abstract_origin
  91           && node->can_remove_if_no_direct_calls_p ()
  92           /* Inlining might enable more devirtualizing, so we want to remove
  93              those only after all devirtualizable virtual calls are processed.
  94              Lacking may edges in callgraph we just preserve them post
  95              inlining.  */
  96           && !DECL_VIRTUAL_P (node->decl)
  97           /* During early inlining some unanalyzed cgraph nodes might be in the
  98              callgraph and they might reffer the function in question.  */
  99           && !cgraph_new_nodes.exists ());
 100 }
 101
 102 /* We are going to eliminate last direct call to NODE (or alias of it) via edge E.
 103    Verify that the NODE can be removed from unit and if it is contained in comdat
 104    group that the whole comdat group is removable.  */
 105
 106 static bool
 107 can_remove_node_now_p (struct cgraph_node *node, struct cgraph_edge *e)
 108 {
 109   struct cgraph_node *next;
 110   if (!can_remove_node_now_p_1 (node))
 111     return false;
 112
 113   /* When we see same comdat group, we need to be sure that all
 114      items can be removed.  */
 115   if (!node->same_comdat_group)
 116     return true;
 117   for (next = dyn_cast<cgraph_node *> (node->same_comdat_group);
 118        next != node; next = dyn_cast<cgraph_node *> (next->same_comdat_group))
 119     if ((next->callers && next->callers != e)
 120         || !can_remove_node_now_p_1 (next))
 121       return false;
 122   return true;
 123 }
 124
 125
 126 /* E is expected to be an edge being inlined.  Clone destination node of
 127    the edge and redirect it to the new clone.
 128    DUPLICATE is used for bookkeeping on whether we are actually creating new
 129    clones or re-using node originally representing out-of-line function call.
 130    By default the offline copy is removed, when it appears dead after inlining.
 131    UPDATE_ORIGINAL prevents this transformation.
 132    If OVERALL_SIZE is non-NULL, the size is updated to reflect the
 133    transformation.
 134    FREQ_SCALE specify the scaling of frequencies of call sites.  */
 135
 136 void
 137 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 138                      bool update_original, int *overall_size, int freq_scale)
 139 {
 140   struct cgraph_node *inlining_into;
 141   struct cgraph_edge *next;
 142
 143   if (e->caller->global.inlined_to)
 144     inlining_into = e->caller->global.inlined_to;
 145   else
 146     inlining_into = e->caller;
 147
 148   if (duplicate)
 149     {
 150       /* We may eliminate the need for out-of-line copy to be output.
 151          In that case just go ahead and re-use it.  This is not just an
 152          memory optimization.  Making offline copy of fuction disappear
 153          from the program will improve future decisions on inlining.  */
 154       if (!e->callee->callers->next_caller
 155           /* Recursive inlining never wants the master clone to
 156              be overwritten.  */
 157           && update_original
 158           && can_remove_node_now_p (e->callee, e))
 159         {
 160           /* TODO: When callee is in a comdat group, we could remove all of it,
 161              including all inline clones inlined into it.  That would however
 162              need small function inlining to register edge removal hook to
 163              maintain the priority queue.
 164
 165              For now we keep the ohter functions in the group in program until
 166              cgraph_remove_unreachable_functions gets rid of them.  */
 167           gcc_assert (!e->callee->global.inlined_to);
 168           e->callee->dissolve_same_comdat_group_list ();
 169           if (e->callee->definition && !DECL_EXTERNAL (e->callee->decl))
 170             {
 171               if (overall_size)
 172                 *overall_size -= inline_summary (e->callee)->size;
 173               nfunctions_inlined++;
 174             }
 175           duplicate = false;
 176           e->callee->externally_visible = false;
 177           update_noncloned_frequencies (e->callee, e->frequency);
 178         }
 179       else
 180         {
 181           struct cgraph_node *n;
 182
 183           if (freq_scale == -1)
 184             freq_scale = e->frequency;
 185           n = e->callee->create_clone (e->callee->decl,
 186                                        MIN (e->count, e->callee->count),
 187                                        freq_scale,
 188                                        update_original, vNULL, true,
 189                                        inlining_into,
 190                                        NULL);
 191           e->redirect_callee (n);
 192         }
 193     }
 194   else
 195     e->callee->dissolve_same_comdat_group_list ();
 196
 197   e->callee->global.inlined_to = inlining_into;
 198
 199   /* Recursively clone all bodies.  */
 200   for (e = e->callee->callees; e; e = next)
 201     {
 202       next = e->next_callee;
 203       if (!e->inline_failed)
 204         clone_inlined_nodes (e, duplicate, update_original, overall_size, freq_scale);
 205       if (e->speculative && !speculation_useful_p (e, true))
 206         {
 207           e->resolve_speculation (NULL);
 208           speculation_removed = true;
 209         }
 210     }
 211 }
 212
 213
 214 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 215    specify whether profile of original function should be updated.  If any new
 216    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 217    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 218    size of caller after inlining. Caller is required to eventually do it via
 219    inline_update_overall_summary.
 220    If callee_removed is non-NULL, set it to true if we removed callee node.
 221
 222    Return true iff any new callgraph edges were discovered as a
 223    result of inlining.  */
 224
 225 bool
 226 inline_call (struct cgraph_edge *e, bool update_original,
 227              vec<cgraph_edge *> *new_edges,
 228              int *overall_size, bool update_overall_summary,
 229              bool *callee_removed)
 230 {
 231   int old_size = 0, new_size = 0;
 232   struct cgraph_node *to = NULL;
 233   struct cgraph_edge *curr = e;
 234   struct cgraph_node *callee = e->callee->ultimate_alias_target ();
 235   bool new_edges_found = false;
 236
 237 #ifdef ENABLE_CHECKING
 238   int estimated_growth = estimate_edge_growth (e);
 239   bool predicated = inline_edge_summary (e)->predicate != NULL;
 240 #endif
 241
 242   speculation_removed = false;
 243   /* Don't inline inlined edges.  */
 244   gcc_assert (e->inline_failed);
 245   /* Don't even think of inlining inline clone.  */
 246   gcc_assert (!callee->global.inlined_to);
 247
 248   e->inline_failed = CIF_OK;
 249   DECL_POSSIBLY_INLINED (callee->decl) = true;
 250
 251   to = e->caller;
 252   if (to->global.inlined_to)
 253     to = to->global.inlined_to;
 254
 255   /* If aliases are involved, redirect edge to the actual destination and
 256      possibly remove the aliases.  */
 257   if (e->callee != callee)
 258     {
 259       struct cgraph_node *alias = e->callee, *next_alias;
 260       e->redirect_callee (callee);
 261       while (alias && alias != callee)
 262         {
 263           if (!alias->callers
 264               && can_remove_node_now_p (alias, e))
 265             {
 266               next_alias = alias->get_alias_target ();
 267               alias->remove ();
 268               if (callee_removed)
 269                 *callee_removed = true;
 270               alias = next_alias;
 271             }
 272           else
 273             break;
 274         }
 275     }
 276
 277   clone_inlined_nodes (e, true, update_original, overall_size, e->frequency);
 278
 279   gcc_assert (curr->callee->global.inlined_to == to);
 280
 281   old_size = inline_summary (to)->size;
 282   inline_merge_summary (e);
 283   if (optimize)
 284     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 285   if (update_overall_summary)
 286    inline_update_overall_summary (to);
 287   new_size = inline_summary (to)->size;
 288
 289   if (callee->calls_comdat_local)
 290     to->calls_comdat_local = true;
 291   else if (to->calls_comdat_local && callee->comdat_local_p ())
 292     {
 293       struct cgraph_edge *se = to->callees;
 294       for (; se; se = se->next_callee)
 295         if (se->inline_failed && se->callee->comdat_local_p ())
 296           break;
 297       if (se == NULL)
 298         to->calls_comdat_local = false;
 299     }
 300
 301 #ifdef ENABLE_CHECKING
 302   /* Verify that estimated growth match real growth.  Allow off-by-one
 303      error due to INLINE_SIZE_SCALE roudoff errors.  */
 304   gcc_assert (!update_overall_summary || !overall_size || new_edges_found
 305               || abs (estimated_growth - (new_size - old_size)) <= 1
 306               || speculation_removed
 307               /* FIXME: a hack.  Edges with false predicate are accounted
 308                  wrong, we should remove them from callgraph.  */
 309               || predicated);
 310 #endif
 311
 312   /* Account the change of overall unit size; external functions will be
 313      removed and are thus not accounted.  */
 314   if (overall_size
 315       && !DECL_EXTERNAL (to->decl))
 316     *overall_size += new_size - old_size;
 317   ncalls_inlined++;
 318
 319   /* This must happen after inline_merge_summary that rely on jump
 320      functions of callee to not be updated.  */
 321   return new_edges_found;
 322 }
 323
 324
 325 /* Copy function body of NODE and redirect all inline clones to it.
 326    This is done before inline plan is applied to NODE when there are
 327    still some inline clones if it.
 328
 329    This is necessary because inline decisions are not really transitive
 330    and the other inline clones may have different bodies.  */
 331
 332 static struct cgraph_node *
 333 save_inline_function_body (struct cgraph_node *node)
 334 {
 335   struct cgraph_node *first_clone, *n;
 336
 337   if (dump_file)
 338     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 339              node->name ());
 340
 341   gcc_assert (node == cgraph_node::get (node->decl));
 342
 343   /* first_clone will be turned into real function.  */
 344   first_clone = node->clones;
 345   first_clone->decl = copy_node (node->decl);
 346   first_clone->decl->decl_with_vis.symtab_node = first_clone;
 347   gcc_assert (first_clone == cgraph_node::get (first_clone->decl));
 348
 349   /* Now reshape the clone tree, so all other clones descends from
 350      first_clone.  */
 351   if (first_clone->next_sibling_clone)
 352     {
 353       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 354         n->clone_of = first_clone;
 355       n->clone_of = first_clone;
 356       n->next_sibling_clone = first_clone->clones;
 357       if (first_clone->clones)
 358         first_clone->clones->prev_sibling_clone = n;
 359       first_clone->clones = first_clone->next_sibling_clone;
 360       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 361       first_clone->next_sibling_clone = NULL;
 362       gcc_assert (!first_clone->prev_sibling_clone);
 363     }
 364   first_clone->clone_of = NULL;
 365
 366   /* Now node in question has no clones.  */
 367   node->clones = NULL;
 368
 369   /* Inline clones share decl with the function they are cloned
 370      from.  Walk the whole clone tree and redirect them all to the
 371      new decl.  */
 372   if (first_clone->clones)
 373     for (n = first_clone->clones; n != first_clone;)
 374       {
 375         gcc_assert (n->decl == node->decl);
 376         n->decl = first_clone->decl;
 377         if (n->clones)
 378           n = n->clones;
 379         else if (n->next_sibling_clone)
 380           n = n->next_sibling_clone;
 381         else
 382           {
 383             while (n != first_clone && !n->next_sibling_clone)
 384               n = n->clone_of;
 385             if (n != first_clone)
 386               n = n->next_sibling_clone;
 387           }
 388       }
 389
 390   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 391   tree_function_versioning (node->decl, first_clone->decl,
 392                             NULL, true, NULL, false,
 393                             NULL, NULL);
 394
 395   /* The function will be short lived and removed after we inline all the clones,
 396      but make it internal so we won't confuse ourself.  */
 397   DECL_EXTERNAL (first_clone->decl) = 0;
 398   TREE_PUBLIC (first_clone->decl) = 0;
 399   DECL_COMDAT (first_clone->decl) = 0;
 400   first_clone->ipa_transforms_to_apply.release ();
 401
 402   /* When doing recursive inlining, the clone may become unnecessary.
 403      This is possible i.e. in the case when the recursive function is proved to be
 404      non-throwing and the recursion happens only in the EH landing pad.
 405      We can not remove the clone until we are done with saving the body.
 406      Remove it now.  */
 407   if (!first_clone->callers)
 408     {
 409       first_clone->remove_symbol_and_inline_clones ();
 410       first_clone = NULL;
 411     }
 412 #ifdef ENABLE_CHECKING
 413   else
 414     first_clone->verify ();
 415 #endif
 416   return first_clone;
 417 }
 418
 419 /* Return true when function body of DECL still needs to be kept around
 420    for later re-use.  */
 421 static bool
 422 preserve_function_body_p (struct cgraph_node *node)
 423 {
 424   gcc_assert (symtab->global_info_ready);
 425   gcc_assert (!node->alias && !node->thunk.thunk_p);
 426
 427   /* Look if there is any clone around.  */
 428   if (node->clones)
 429     return true;
 430   return false;
 431 }
 432
 433 /* Apply inline plan to function.  */
 434
 435 unsigned int
 436 inline_transform (struct cgraph_node *node)
 437 {
 438   unsigned int todo = 0;
 439   struct cgraph_edge *e, *next;
 440
 441   /* FIXME: Currently the pass manager is adding inline transform more than
 442      once to some clones.  This needs revisiting after WPA cleanups.  */
 443   if (cfun->after_inlining)
 444     return 0;
 445
 446   /* We might need the body of this function so that we can expand
 447      it inline somewhere else.  */
 448   if (preserve_function_body_p (node))
 449     save_inline_function_body (node);
 450
 451   for (e = node->callees; e; e = next)
 452     {
 453       next = e->next_callee;
 454       e->redirect_call_stmt_to_callee ();
 455     }
 456   node->remove_all_references ();
 457
 458   timevar_push (TV_INTEGRATION);
 459   if (node->callees && optimize)
 460     todo = optimize_inline_calls (current_function_decl);
 461   timevar_pop (TV_INTEGRATION);
 462
 463   cfun->always_inline_functions_inlined = true;
 464   cfun->after_inlining = true;
 465   todo |= execute_fixup_cfg ();
 466
 467   if (!(todo & TODO_update_ssa_any))
 468     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 469     todo |= TODO_update_ssa_only_virtuals;
 470
 471   return todo;
 472 }