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           && !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    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           symtab_dissolve_same_comdat_group_list (e->callee);
 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 = cgraph_clone_node (e->callee, e->callee->decl,
 186                                  e->count, freq_scale, update_original,
 187                                  vNULL, true, inlining_into);
 188           cgraph_redirect_edge_callee (e, n);
 189         }
 190     }
 191   else
 192     symtab_dissolve_same_comdat_group_list (e->callee);
 193
 194   e->callee->global.inlined_to = inlining_into;
 195
 196   /* Recursively clone all bodies.  */
 197   for (e = e->callee->callees; e; e = next)
 198     {
 199       next = e->next_callee;
 200       if (!e->inline_failed)
 201         clone_inlined_nodes (e, duplicate, update_original, overall_size, freq_scale);
 202       if (e->speculative && !speculation_useful_p (e, true))
 203         {
 204           cgraph_resolve_speculation (e, NULL);
 205           speculation_removed = true;
 206         }
 207     }
 208 }
 209
 210
 211 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 212    specify whether profile of original function should be updated.  If any new
 213    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 214    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 215    size of caller after inlining. Caller is required to eventually do it via
 216    inline_update_overall_summary.
 217
 218    Return true iff any new callgraph edges were discovered as a
 219    result of inlining.  */
 220
 221 bool
 222 inline_call (struct cgraph_edge *e, bool update_original,
 223              vec<cgraph_edge_p> *new_edges,
 224              int *overall_size, bool update_overall_summary)
 225 {
 226   int old_size = 0, new_size = 0;
 227   struct cgraph_node *to = NULL;
 228   struct cgraph_edge *curr = e;
 229   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 230   bool new_edges_found = false;
 231
 232 #ifdef ENABLE_CHECKING
 233   int estimated_growth = estimate_edge_growth (e);
 234   bool predicated = inline_edge_summary (e)->predicate != NULL;
 235 #endif
 236
 237   speculation_removed = false;
 238   /* Don't inline inlined edges.  */
 239   gcc_assert (e->inline_failed);
 240   /* Don't even think of inlining inline clone.  */
 241   gcc_assert (!callee->global.inlined_to);
 242
 243   e->inline_failed = CIF_OK;
 244   DECL_POSSIBLY_INLINED (callee->decl) = true;
 245
 246   to = e->caller;
 247   if (to->global.inlined_to)
 248     to = to->global.inlined_to;
 249
 250   /* If aliases are involved, redirect edge to the actual destination and
 251      possibly remove the aliases.  */
 252   if (e->callee != callee)
 253     {
 254       struct cgraph_node *alias = e->callee, *next_alias;
 255       cgraph_redirect_edge_callee (e, callee);
 256       while (alias && alias != callee)
 257         {
 258           if (!alias->callers
 259               && can_remove_node_now_p (alias, e))
 260             {
 261               next_alias = cgraph_alias_target (alias);
 262               cgraph_remove_node (alias);
 263               alias = next_alias;
 264             }
 265           else
 266             break;
 267         }
 268     }
 269
 270   clone_inlined_nodes (e, true, update_original, overall_size, e->frequency);
 271
 272   gcc_assert (curr->callee->global.inlined_to == to);
 273
 274   old_size = inline_summary (to)->size;
 275   inline_merge_summary (e);
 276   if (optimize)
 277     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 278   if (update_overall_summary)
 279    inline_update_overall_summary (to);
 280   new_size = inline_summary (to)->size;
 281
 282   if (callee->calls_comdat_local)
 283     to->calls_comdat_local = true;
 284   else if (to->calls_comdat_local && symtab_comdat_local_p (callee))
 285     {
 286       struct cgraph_edge *se = to->callees;
 287       for (; se; se = se->next_callee)
 288         if (se->inline_failed && symtab_comdat_local_p (se->callee))
 289           break;
 290       if (se == NULL)
 291         to->calls_comdat_local = false;
 292     }
 293
 294 #ifdef ENABLE_CHECKING
 295   /* Verify that estimated growth match real growth.  Allow off-by-one
 296      error due to INLINE_SIZE_SCALE roudoff errors.  */
 297   gcc_assert (!update_overall_summary || !overall_size || new_edges_found
 298               || abs (estimated_growth - (new_size - old_size)) <= 1
 299               || speculation_removed
 300               /* FIXME: a hack.  Edges with false predicate are accounted
 301                  wrong, we should remove them from callgraph.  */
 302               || predicated);
 303 #endif
 304
 305   /* Account the change of overall unit size; external functions will be
 306      removed and are thus not accounted.  */
 307   if (overall_size
 308       && !DECL_EXTERNAL (to->decl))
 309     *overall_size += new_size - old_size;
 310   ncalls_inlined++;
 311
 312   /* This must happen after inline_merge_summary that rely on jump
 313      functions of callee to not be updated.  */
 314   return new_edges_found;
 315 }
 316
 317
 318 /* Copy function body of NODE and redirect all inline clones to it.
 319    This is done before inline plan is applied to NODE when there are
 320    still some inline clones if it.
 321
 322    This is necessary because inline decisions are not really transitive
 323    and the other inline clones may have different bodies.  */
 324
 325 static struct cgraph_node *
 326 save_inline_function_body (struct cgraph_node *node)
 327 {
 328   struct cgraph_node *first_clone, *n;
 329
 330   if (dump_file)
 331     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 332              node->name ());
 333
 334   gcc_assert (node == cgraph_get_node (node->decl));
 335
 336   /* first_clone will be turned into real function.  */
 337   first_clone = node->clones;
 338   first_clone->decl = copy_node (node->decl);
 339   symtab_insert_node_to_hashtable (first_clone);
 340   gcc_assert (first_clone == cgraph_get_node (first_clone->decl));
 341
 342   /* Now reshape the clone tree, so all other clones descends from
 343      first_clone.  */
 344   if (first_clone->next_sibling_clone)
 345     {
 346       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 347         n->clone_of = first_clone;
 348       n->clone_of = first_clone;
 349       n->next_sibling_clone = first_clone->clones;
 350       if (first_clone->clones)
 351         first_clone->clones->prev_sibling_clone = n;
 352       first_clone->clones = first_clone->next_sibling_clone;
 353       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 354       first_clone->next_sibling_clone = NULL;
 355       gcc_assert (!first_clone->prev_sibling_clone);
 356     }
 357   first_clone->clone_of = NULL;
 358
 359   /* Now node in question has no clones.  */
 360   node->clones = NULL;
 361
 362   /* Inline clones share decl with the function they are cloned
 363      from.  Walk the whole clone tree and redirect them all to the
 364      new decl.  */
 365   if (first_clone->clones)
 366     for (n = first_clone->clones; n != first_clone;)
 367       {
 368         gcc_assert (n->decl == node->decl);
 369         n->decl = first_clone->decl;
 370         if (n->clones)
 371           n = n->clones;
 372         else if (n->next_sibling_clone)
 373           n = n->next_sibling_clone;
 374         else
 375           {
 376             while (n != first_clone && !n->next_sibling_clone)
 377               n = n->clone_of;
 378             if (n != first_clone)
 379               n = n->next_sibling_clone;
 380           }
 381       }
 382
 383   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 384   tree_function_versioning (node->decl, first_clone->decl,
 385                             NULL, true, NULL, false,
 386                             NULL, NULL);
 387
 388   /* The function will be short lived and removed after we inline all the clones,
 389      but make it internal so we won't confuse ourself.  */
 390   DECL_EXTERNAL (first_clone->decl) = 0;
 391   TREE_PUBLIC (first_clone->decl) = 0;
 392   DECL_COMDAT (first_clone->decl) = 0;
 393   first_clone->ipa_transforms_to_apply.release ();
 394
 395   /* When doing recursive inlining, the clone may become unnecessary.
 396      This is possible i.e. in the case when the recursive function is proved to be
 397      non-throwing and the recursion happens only in the EH landing pad.
 398      We can not remove the clone until we are done with saving the body.
 399      Remove it now.  */
 400   if (!first_clone->callers)
 401     {
 402       cgraph_remove_node_and_inline_clones (first_clone, NULL);
 403       first_clone = NULL;
 404     }
 405 #ifdef ENABLE_CHECKING
 406   else
 407     verify_cgraph_node (first_clone);
 408 #endif
 409   return first_clone;
 410 }
 411
 412 /* Return true when function body of DECL still needs to be kept around
 413    for later re-use.  */
 414 static bool
 415 preserve_function_body_p (struct cgraph_node *node)
 416 {
 417   gcc_assert (cgraph_global_info_ready);
 418   gcc_assert (!node->alias && !node->thunk.thunk_p);
 419
 420   /* Look if there is any clone around.  */
 421   if (node->clones)
 422     return true;
 423   return false;
 424 }
 425
 426 /* Apply inline plan to function.  */
 427
 428 unsigned int
 429 inline_transform (struct cgraph_node *node)
 430 {
 431   unsigned int todo = 0;
 432   struct cgraph_edge *e, *next;
 433
 434   /* FIXME: Currently the pass manager is adding inline transform more than
 435      once to some clones.  This needs revisiting after WPA cleanups.  */
 436   if (cfun->after_inlining)
 437     return 0;
 438
 439   /* We might need the body of this function so that we can expand
 440      it inline somewhere else.  */
 441   if (preserve_function_body_p (node))
 442     save_inline_function_body (node);
 443
 444   for (e = node->callees; e; e = next)
 445     {
 446       next = e->next_callee;
 447       cgraph_redirect_edge_call_stmt_to_callee (e);
 448     }
 449   ipa_remove_all_references (&node->ref_list);
 450
 451   timevar_push (TV_INTEGRATION);
 452   if (node->callees && optimize)
 453     todo = optimize_inline_calls (current_function_decl);
 454   timevar_pop (TV_INTEGRATION);
 455
 456   cfun->always_inline_functions_inlined = true;
 457   cfun->after_inlining = true;
 458   todo |= execute_fixup_cfg ();
 459
 460   if (!(todo & TODO_update_ssa_any))
 461     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 462     todo |= TODO_update_ssa_only_virtuals;
 463
 464   return todo;
 465 }