gcc/ipa-inline-transform.c

   1 /* Callgraph transformations to handle inlining
   2    Copyright (C) 2003-2013 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 "cgraph.h"
  38 #include "intl.h"
  39 #include "coverage.h"
  40 #include "ggc.h"
  41 #include "tree-flow.h"
  42 #include "ipa-prop.h"
  43 #include "ipa-inline.h"
  44 #include "tree-inline.h"
  45 #include "tree-pass.h"
  46
  47 int ncalls_inlined;
  48 int nfunctions_inlined;
  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->symbol.address_taken
  89           && !ipa_ref_has_aliases_p (&node->symbol.ref_list)
  90           && !node->used_as_abstract_origin
  91           && cgraph_can_remove_if_no_direct_calls_p (node)
  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->symbol.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);
 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->symbol.same_comdat_group)
 116     return true;
 117   for (next = cgraph (node->symbol.same_comdat_group);
 118        next != node; next = cgraph (next->symbol.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    */
 131
 132 void
 133 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 134                      bool update_original, int *overall_size)
 135 {
 136   struct cgraph_node *inlining_into;
 137
 138   if (e->caller->global.inlined_to)
 139     inlining_into = e->caller->global.inlined_to;
 140   else
 141     inlining_into = e->caller;
 142
 143   if (duplicate)
 144     {
 145       /* We may eliminate the need for out-of-line copy to be output.
 146          In that case just go ahead and re-use it.  This is not just an
 147          memory optimization.  Making offline copy of fuction disappear
 148          from the program will improve future decisions on inlining.  */
 149       if (!e->callee->callers->next_caller
 150           /* Recursive inlining never wants the master clone to
 151              be overwritten.  */
 152           && update_original
 153           && can_remove_node_now_p (e->callee, e))
 154         {
 155           /* TODO: When callee is in a comdat group, we could remove all of it,
 156              including all inline clones inlined into it.  That would however
 157              need small function inlining to register edge removal hook to
 158              maintain the priority queue.
 159
 160              For now we keep the ohter functions in the group in program until
 161              cgraph_remove_unreachable_functions gets rid of them.  */
 162           gcc_assert (!e->callee->global.inlined_to);
 163           symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 164           if (e->callee->symbol.definition && !DECL_EXTERNAL (e->callee->symbol.decl))
 165             {
 166               if (overall_size)
 167                 *overall_size -= inline_summary (e->callee)->size;
 168               nfunctions_inlined++;
 169             }
 170           duplicate = false;
 171           e->callee->symbol.externally_visible = false;
 172           update_noncloned_frequencies (e->callee, e->frequency);
 173         }
 174       else
 175         {
 176           struct cgraph_node *n;
 177           n = cgraph_clone_node (e->callee, e->callee->symbol.decl,
 178                                  e->count, e->frequency, update_original,
 179                                  vNULL, true, inlining_into);
 180           cgraph_redirect_edge_callee (e, n);
 181         }
 182     }
 183   else
 184     symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 185
 186   e->callee->global.inlined_to = inlining_into;
 187
 188   /* Recursively clone all bodies.  */
 189   for (e = e->callee->callees; e; e = e->next_callee)
 190     if (!e->inline_failed)
 191       clone_inlined_nodes (e, duplicate, update_original, overall_size);
 192 }
 193
 194
 195 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 196    specify whether profile of original function should be updated.  If any new
 197    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 198    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 199    size of caller after inlining. Caller is required to eventually do it via
 200    inline_update_overall_summary.
 201
 202    Return true iff any new callgraph edges were discovered as a
 203    result of inlining.  */
 204
 205 bool
 206 inline_call (struct cgraph_edge *e, bool update_original,
 207              vec<cgraph_edge_p> *new_edges,
 208              int *overall_size, bool update_overall_summary)
 209 {
 210   int old_size = 0, new_size = 0;
 211   struct cgraph_node *to = NULL;
 212   struct cgraph_edge *curr = e;
 213   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 214   bool new_edges_found = false;
 215
 216 #ifdef ENABLE_CHECKING
 217   int estimated_growth = estimate_edge_growth (e);
 218   bool predicated = inline_edge_summary (e)->predicate != NULL;
 219 #endif
 220
 221   /* Don't inline inlined edges.  */
 222   gcc_assert (e->inline_failed);
 223   /* Don't even think of inlining inline clone.  */
 224   gcc_assert (!callee->global.inlined_to);
 225
 226   e->inline_failed = CIF_OK;
 227   DECL_POSSIBLY_INLINED (callee->symbol.decl) = true;
 228
 229   to = e->caller;
 230   if (to->global.inlined_to)
 231     to = to->global.inlined_to;
 232
 233   /* If aliases are involved, redirect edge to the actual destination and
 234      possibly remove the aliases.  */
 235   if (e->callee != callee)
 236     {
 237       struct cgraph_node *alias = e->callee, *next_alias;
 238       cgraph_redirect_edge_callee (e, callee);
 239       while (alias && alias != callee)
 240         {
 241           if (!alias->callers
 242               && can_remove_node_now_p (alias, e))
 243             {
 244               next_alias = cgraph_alias_target (alias);
 245               cgraph_remove_node (alias);
 246               alias = next_alias;
 247             }
 248           else
 249             break;
 250         }
 251     }
 252
 253   clone_inlined_nodes (e, true, update_original, overall_size);
 254
 255   gcc_assert (curr->callee->global.inlined_to == to);
 256
 257   old_size = inline_summary (to)->size;
 258   inline_merge_summary (e);
 259   if (optimize)
 260     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 261   if (update_overall_summary)
 262    inline_update_overall_summary (to);
 263   new_size = inline_summary (to)->size;
 264
 265 #ifdef ENABLE_CHECKING
 266   /* Verify that estimated growth match real growth.  Allow off-by-one
 267      error due to INLINE_SIZE_SCALE roudoff errors.  */
 268   gcc_assert (!update_overall_summary || !overall_size || new_edges_found
 269               || abs (estimated_growth - (new_size - old_size)) <= 1
 270               /* FIXME: a hack.  Edges with false predicate are accounted
 271                  wrong, we should remove them from callgraph.  */
 272               || predicated);
 273 #endif
 274
 275   /* Account the change of overall unit size; external functions will be
 276      removed and are thus not accounted.  */
 277   if (overall_size
 278       && !DECL_EXTERNAL (to->symbol.decl))
 279     *overall_size += new_size - old_size;
 280   ncalls_inlined++;
 281
 282   /* This must happen after inline_merge_summary that rely on jump
 283      functions of callee to not be updated.  */
 284   return new_edges_found;
 285 }
 286
 287
 288 /* Copy function body of NODE and redirect all inline clones to it.
 289    This is done before inline plan is applied to NODE when there are
 290    still some inline clones if it.
 291
 292    This is necessary because inline decisions are not really transitive
 293    and the other inline clones may have different bodies.  */
 294
 295 static struct cgraph_node *
 296 save_inline_function_body (struct cgraph_node *node)
 297 {
 298   struct cgraph_node *first_clone, *n;
 299
 300   if (dump_file)
 301     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 302              cgraph_node_name (node));
 303
 304   gcc_assert (node == cgraph_get_node (node->symbol.decl));
 305
 306   /* first_clone will be turned into real function.  */
 307   first_clone = node->clones;
 308   first_clone->symbol.decl = copy_node (node->symbol.decl);
 309   symtab_insert_node_to_hashtable ((symtab_node) first_clone);
 310   gcc_assert (first_clone == cgraph_get_node (first_clone->symbol.decl));
 311
 312   /* Now reshape the clone tree, so all other clones descends from
 313      first_clone.  */
 314   if (first_clone->next_sibling_clone)
 315     {
 316       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 317         n->clone_of = first_clone;
 318       n->clone_of = first_clone;
 319       n->next_sibling_clone = first_clone->clones;
 320       if (first_clone->clones)
 321         first_clone->clones->prev_sibling_clone = n;
 322       first_clone->clones = first_clone->next_sibling_clone;
 323       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 324       first_clone->next_sibling_clone = NULL;
 325       gcc_assert (!first_clone->prev_sibling_clone);
 326     }
 327   first_clone->clone_of = NULL;
 328
 329   /* Now node in question has no clones.  */
 330   node->clones = NULL;
 331
 332   /* Inline clones share decl with the function they are cloned
 333      from.  Walk the whole clone tree and redirect them all to the
 334      new decl.  */
 335   if (first_clone->clones)
 336     for (n = first_clone->clones; n != first_clone;)
 337       {
 338         gcc_assert (n->symbol.decl == node->symbol.decl);
 339         n->symbol.decl = first_clone->symbol.decl;
 340         if (n->clones)
 341           n = n->clones;
 342         else if (n->next_sibling_clone)
 343           n = n->next_sibling_clone;
 344         else
 345           {
 346             while (n != first_clone && !n->next_sibling_clone)
 347               n = n->clone_of;
 348             if (n != first_clone)
 349               n = n->next_sibling_clone;
 350           }
 351       }
 352
 353   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 354   tree_function_versioning (node->symbol.decl, first_clone->symbol.decl,
 355                             NULL, true, NULL, false,
 356                             NULL, NULL);
 357
 358   /* The function will be short lived and removed after we inline all the clones,
 359      but make it internal so we won't confuse ourself.  */
 360   DECL_EXTERNAL (first_clone->symbol.decl) = 0;
 361   DECL_COMDAT_GROUP (first_clone->symbol.decl) = NULL_TREE;
 362   TREE_PUBLIC (first_clone->symbol.decl) = 0;
 363   DECL_COMDAT (first_clone->symbol.decl) = 0;
 364   first_clone->ipa_transforms_to_apply.release ();
 365
 366   /* When doing recursive inlining, the clone may become unnecessary.
 367      This is possible i.e. in the case when the recursive function is proved to be
 368      non-throwing and the recursion happens only in the EH landing pad.
 369      We can not remove the clone until we are done with saving the body.
 370      Remove it now.  */
 371   if (!first_clone->callers)
 372     {
 373       cgraph_remove_node_and_inline_clones (first_clone, NULL);
 374       first_clone = NULL;
 375     }
 376 #ifdef ENABLE_CHECKING
 377   else
 378     verify_cgraph_node (first_clone);
 379 #endif
 380   return first_clone;
 381 }
 382
 383 /* Return true when function body of DECL still needs to be kept around
 384    for later re-use.  */
 385 static bool
 386 preserve_function_body_p (struct cgraph_node *node)
 387 {
 388   gcc_assert (cgraph_global_info_ready);
 389   gcc_assert (!node->symbol.alias && !node->thunk.thunk_p);
 390
 391   /* Look if there is any clone around.  */
 392   if (node->clones)
 393     return true;
 394   return false;
 395 }
 396
 397 /* Apply inline plan to function.  */
 398
 399 unsigned int
 400 inline_transform (struct cgraph_node *node)
 401 {
 402   unsigned int todo = 0;
 403   struct cgraph_edge *e;
 404
 405   /* FIXME: Currently the pass manager is adding inline transform more than
 406      once to some clones.  This needs revisiting after WPA cleanups.  */
 407   if (cfun->after_inlining)
 408     return 0;
 409
 410   /* We might need the body of this function so that we can expand
 411      it inline somewhere else.  */
 412   if (preserve_function_body_p (node))
 413     save_inline_function_body (node);
 414
 415   for (e = node->callees; e; e = e->next_callee)
 416     cgraph_redirect_edge_call_stmt_to_callee (e);
 417   ipa_remove_all_references (&node->symbol.ref_list);
 418
 419   timevar_push (TV_INTEGRATION);
 420   if (node->callees)
 421     todo = optimize_inline_calls (current_function_decl);
 422   timevar_pop (TV_INTEGRATION);
 423
 424   cfun->always_inline_functions_inlined = true;
 425   cfun->after_inlining = true;
 426   todo |= execute_fixup_cfg ();
 427
 428   if (!(todo & TODO_update_ssa_any))
 429     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 430     todo |= TODO_update_ssa_only_virtuals;
 431
 432   return todo;
 433 }