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