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 /* Return true if NODE is a master clone with non-inline clones.  */
 126
 127 static bool
 128 master_clone_with_noninline_clones_p (struct cgraph_node *node)
 129 {
 130   if (node->clone_of)
 131     return false;
 132
 133   for (struct cgraph_node *n = node->clones; n; n = n->next_sibling_clone)
 134     if (n->decl != node->decl)
 135       return true;
 136
 137   return false;
 138 }
 139
 140 /* E is expected to be an edge being inlined.  Clone destination node of
 141    the edge and redirect it to the new clone.
 142    DUPLICATE is used for bookkeeping on whether we are actually creating new
 143    clones or re-using node originally representing out-of-line function call.
 144    By default the offline copy is removed, when it appears dead after inlining.
 145    UPDATE_ORIGINAL prevents this transformation.
 146    If OVERALL_SIZE is non-NULL, the size is updated to reflect the
 147    transformation.
 148    FREQ_SCALE specify the scaling of frequencies of call sites.  */
 149
 150 void
 151 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 152                      bool update_original, int *overall_size, int freq_scale)
 153 {
 154   struct cgraph_node *inlining_into;
 155   struct cgraph_edge *next;
 156
 157   if (e->caller->global.inlined_to)
 158     inlining_into = e->caller->global.inlined_to;
 159   else
 160     inlining_into = e->caller;
 161
 162   if (duplicate)
 163     {
 164       /* We may eliminate the need for out-of-line copy to be output.
 165          In that case just go ahead and re-use it.  This is not just an
 166          memory optimization.  Making offline copy of fuction disappear
 167          from the program will improve future decisions on inlining.  */
 168       if (!e->callee->callers->next_caller
 169           /* Recursive inlining never wants the master clone to
 170              be overwritten.  */
 171           && update_original
 172           && can_remove_node_now_p (e->callee, e)
 173           /* We cannot overwrite a master clone with non-inline clones
 174              until after these clones are materialized.  */
 175           && !master_clone_with_noninline_clones_p (e->callee))
 176         {
 177           /* TODO: When callee is in a comdat group, we could remove all of it,
 178              including all inline clones inlined into it.  That would however
 179              need small function inlining to register edge removal hook to
 180              maintain the priority queue.
 181
 182              For now we keep the ohter functions in the group in program until
 183              cgraph_remove_unreachable_functions gets rid of them.  */
 184           gcc_assert (!e->callee->global.inlined_to);
 185           e->callee->dissolve_same_comdat_group_list ();
 186           if (e->callee->definition && !DECL_EXTERNAL (e->callee->decl))
 187             {
 188               if (overall_size)
 189                 *overall_size -= inline_summary (e->callee)->size;
 190               nfunctions_inlined++;
 191             }
 192           duplicate = false;
 193           e->callee->externally_visible = false;
 194           update_noncloned_frequencies (e->callee, e->frequency);
 195         }
 196       else
 197         {
 198           struct cgraph_node *n;
 199
 200           if (freq_scale == -1)
 201             freq_scale = e->frequency;
 202           n = e->callee->create_clone (e->callee->decl,
 203                                        MIN (e->count, e->callee->count),
 204                                        freq_scale,
 205                                        update_original, vNULL, true,
 206                                        inlining_into,
 207                                        NULL);
 208           e->redirect_callee (n);
 209         }
 210     }
 211   else
 212     e->callee->dissolve_same_comdat_group_list ();
 213
 214   e->callee->global.inlined_to = inlining_into;
 215
 216   /* Recursively clone all bodies.  */
 217   for (e = e->callee->callees; e; e = next)
 218     {
 219       next = e->next_callee;
 220       if (!e->inline_failed)
 221         clone_inlined_nodes (e, duplicate, update_original, overall_size, freq_scale);
 222       if (e->speculative && !speculation_useful_p (e, true))
 223         {
 224           e->resolve_speculation (NULL);
 225           speculation_removed = true;
 226         }
 227     }
 228 }
 229
 230
 231 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 232    specify whether profile of original function should be updated.  If any new
 233    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 234    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 235    size of caller after inlining. Caller is required to eventually do it via
 236    inline_update_overall_summary.
 237    If callee_removed is non-NULL, set it to true if we removed callee node.
 238
 239    Return true iff any new callgraph edges were discovered as a
 240    result of inlining.  */
 241
 242 bool
 243 inline_call (struct cgraph_edge *e, bool update_original,
 244              vec<cgraph_edge *> *new_edges,
 245              int *overall_size, bool update_overall_summary,
 246              bool *callee_removed)
 247 {
 248   int old_size = 0, new_size = 0;
 249   struct cgraph_node *to = NULL;
 250   struct cgraph_edge *curr = e;
 251   struct cgraph_node *callee = e->callee->ultimate_alias_target ();
 252   bool new_edges_found = false;
 253
 254 #ifdef ENABLE_CHECKING
 255   int estimated_growth = estimate_edge_growth (e);
 256   bool predicated = inline_edge_summary (e)->predicate != NULL;
 257 #endif
 258
 259   speculation_removed = false;
 260   /* Don't inline inlined edges.  */
 261   gcc_assert (e->inline_failed);
 262   /* Don't even think of inlining inline clone.  */
 263   gcc_assert (!callee->global.inlined_to);
 264
 265   e->inline_failed = CIF_OK;
 266   DECL_POSSIBLY_INLINED (callee->decl) = true;
 267
 268   to = e->caller;
 269   if (to->global.inlined_to)
 270     to = to->global.inlined_to;
 271
 272   /* If aliases are involved, redirect edge to the actual destination and
 273      possibly remove the aliases.  */
 274   if (e->callee != callee)
 275     {
 276       struct cgraph_node *alias = e->callee, *next_alias;
 277       e->redirect_callee (callee);
 278       while (alias && alias != callee)
 279         {
 280           if (!alias->callers
 281               && can_remove_node_now_p (alias, e))
 282             {
 283               next_alias = alias->get_alias_target ();
 284               alias->remove ();
 285               if (callee_removed)
 286                 *callee_removed = true;
 287               alias = next_alias;
 288             }
 289           else
 290             break;
 291         }
 292     }
 293
 294   clone_inlined_nodes (e, true, update_original, overall_size, e->frequency);
 295
 296   gcc_assert (curr->callee->global.inlined_to == to);
 297
 298   old_size = inline_summary (to)->size;
 299   inline_merge_summary (e);
 300   if (optimize)
 301     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 302   if (update_overall_summary)
 303    inline_update_overall_summary (to);
 304   new_size = inline_summary (to)->size;
 305
 306   if (callee->calls_comdat_local)
 307     to->calls_comdat_local = true;
 308   else if (to->calls_comdat_local && callee->comdat_local_p ())
 309     {
 310       struct cgraph_edge *se = to->callees;
 311       for (; se; se = se->next_callee)
 312         if (se->inline_failed && se->callee->comdat_local_p ())
 313           break;
 314       if (se == NULL)
 315         to->calls_comdat_local = false;
 316     }
 317
 318 #ifdef ENABLE_CHECKING
 319   /* Verify that estimated growth match real growth.  Allow off-by-one
 320      error due to INLINE_SIZE_SCALE roudoff errors.  */
 321   gcc_assert (!update_overall_summary || !overall_size || new_edges_found
 322               || abs (estimated_growth - (new_size - old_size)) <= 1
 323               || speculation_removed
 324               /* FIXME: a hack.  Edges with false predicate are accounted
 325                  wrong, we should remove them from callgraph.  */
 326               || predicated);
 327 #endif
 328
 329   /* Account the change of overall unit size; external functions will be
 330      removed and are thus not accounted.  */
 331   if (overall_size
 332       && !DECL_EXTERNAL (to->decl))
 333     *overall_size += new_size - old_size;
 334   ncalls_inlined++;
 335
 336   /* This must happen after inline_merge_summary that rely on jump
 337      functions of callee to not be updated.  */
 338   return new_edges_found;
 339 }
 340
 341
 342 /* Copy function body of NODE and redirect all inline clones to it.
 343    This is done before inline plan is applied to NODE when there are
 344    still some inline clones if it.
 345
 346    This is necessary because inline decisions are not really transitive
 347    and the other inline clones may have different bodies.  */
 348
 349 static struct cgraph_node *
 350 save_inline_function_body (struct cgraph_node *node)
 351 {
 352   struct cgraph_node *first_clone, *n;
 353
 354   if (dump_file)
 355     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 356              node->name ());
 357
 358   gcc_assert (node == cgraph_node::get (node->decl));
 359
 360   /* first_clone will be turned into real function.  */
 361   first_clone = node->clones;
 362   first_clone->decl = copy_node (node->decl);
 363   first_clone->decl->decl_with_vis.symtab_node = first_clone;
 364   gcc_assert (first_clone == cgraph_node::get (first_clone->decl));
 365
 366   /* Now reshape the clone tree, so all other clones descends from
 367      first_clone.  */
 368   if (first_clone->next_sibling_clone)
 369     {
 370       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 371         n->clone_of = first_clone;
 372       n->clone_of = first_clone;
 373       n->next_sibling_clone = first_clone->clones;
 374       if (first_clone->clones)
 375         first_clone->clones->prev_sibling_clone = n;
 376       first_clone->clones = first_clone->next_sibling_clone;
 377       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 378       first_clone->next_sibling_clone = NULL;
 379       gcc_assert (!first_clone->prev_sibling_clone);
 380     }
 381   first_clone->clone_of = NULL;
 382
 383   /* Now node in question has no clones.  */
 384   node->clones = NULL;
 385
 386   /* Inline clones share decl with the function they are cloned
 387      from.  Walk the whole clone tree and redirect them all to the
 388      new decl.  */
 389   if (first_clone->clones)
 390     for (n = first_clone->clones; n != first_clone;)
 391       {
 392         gcc_assert (n->decl == node->decl);
 393         n->decl = first_clone->decl;
 394         if (n->clones)
 395           n = n->clones;
 396         else if (n->next_sibling_clone)
 397           n = n->next_sibling_clone;
 398         else
 399           {
 400             while (n != first_clone && !n->next_sibling_clone)
 401               n = n->clone_of;
 402             if (n != first_clone)
 403               n = n->next_sibling_clone;
 404           }
 405       }
 406
 407   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 408   tree_function_versioning (node->decl, first_clone->decl,
 409                             NULL, true, NULL, false,
 410                             NULL, NULL);
 411
 412   /* The function will be short lived and removed after we inline all the clones,
 413      but make it internal so we won't confuse ourself.  */
 414   DECL_EXTERNAL (first_clone->decl) = 0;
 415   TREE_PUBLIC (first_clone->decl) = 0;
 416   DECL_COMDAT (first_clone->decl) = 0;
 417   first_clone->ipa_transforms_to_apply.release ();
 418
 419   /* When doing recursive inlining, the clone may become unnecessary.
 420      This is possible i.e. in the case when the recursive function is proved to be
 421      non-throwing and the recursion happens only in the EH landing pad.
 422      We can not remove the clone until we are done with saving the body.
 423      Remove it now.  */
 424   if (!first_clone->callers)
 425     {
 426       first_clone->remove_symbol_and_inline_clones ();
 427       first_clone = NULL;
 428     }
 429 #ifdef ENABLE_CHECKING
 430   else
 431     first_clone->verify ();
 432 #endif
 433   return first_clone;
 434 }
 435
 436 /* Return true when function body of DECL still needs to be kept around
 437    for later re-use.  */
 438 static bool
 439 preserve_function_body_p (struct cgraph_node *node)
 440 {
 441   gcc_assert (symtab->global_info_ready);
 442   gcc_assert (!node->alias && !node->thunk.thunk_p);
 443
 444   /* Look if there is any clone around.  */
 445   if (node->clones)
 446     return true;
 447   return false;
 448 }
 449
 450 /* Apply inline plan to function.  */
 451
 452 unsigned int
 453 inline_transform (struct cgraph_node *node)
 454 {
 455   unsigned int todo = 0;
 456   struct cgraph_edge *e, *next;
 457
 458   /* FIXME: Currently the pass manager is adding inline transform more than
 459      once to some clones.  This needs revisiting after WPA cleanups.  */
 460   if (cfun->after_inlining)
 461     return 0;
 462
 463   /* We might need the body of this function so that we can expand
 464      it inline somewhere else.  */
 465   if (preserve_function_body_p (node))
 466     save_inline_function_body (node);
 467
 468   for (e = node->callees; e; e = next)
 469     {
 470       next = e->next_callee;
 471       e->redirect_call_stmt_to_callee ();
 472     }
 473   node->remove_all_references ();
 474
 475   timevar_push (TV_INTEGRATION);
 476   if (node->callees && optimize)
 477     todo = optimize_inline_calls (current_function_decl);
 478   timevar_pop (TV_INTEGRATION);
 479
 480   cfun->always_inline_functions_inlined = true;
 481   cfun->after_inlining = true;
 482   todo |= execute_fixup_cfg ();
 483
 484   if (!(todo & TODO_update_ssa_any))
 485     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 486     todo |= TODO_update_ssa_only_virtuals;
 487
 488   return todo;
 489 }