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 "vec.h"
  42 #include "hash-map.h"
  43 #include "is-a.h"
  44 #include "plugin-api.h"
  45 #include "hashtab.h"
  46 #include "hash-set.h"
  47 #include "machmode.h"
  48 #include "hard-reg-set.h"
  49 #include "input.h"
  50 #include "function.h"
  51 #include "ipa-ref.h"
  52 #include "cgraph.h"
  53 #include "alloc-pool.h"
  54 #include "symbol-summary.h"
  55 #include "ipa-prop.h"
  56 #include "ipa-inline.h"
  57 #include "tree-inline.h"
  58 #include "tree-pass.h"
  59
  60 int ncalls_inlined;
  61 int nfunctions_inlined;
  62 bool speculation_removed;
  63
  64 /* Scale frequency of NODE edges by FREQ_SCALE.  */
  65
  66 static void
  67 update_noncloned_frequencies (struct cgraph_node *node,
  68                               int freq_scale)
  69 {
  70   struct cgraph_edge *e;
  71
  72   /* We do not want to ignore high loop nest after freq drops to 0.  */
  73   if (!freq_scale)
  74     freq_scale = 1;
  75   for (e = node->callees; e; e = e->next_callee)
  76     {
  77       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  78       if (e->frequency > CGRAPH_FREQ_MAX)
  79         e->frequency = CGRAPH_FREQ_MAX;
  80       if (!e->inline_failed)
  81         update_noncloned_frequencies (e->callee, freq_scale);
  82     }
  83   for (e = node->indirect_calls; e; e = e->next_callee)
  84     {
  85       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  86       if (e->frequency > CGRAPH_FREQ_MAX)
  87         e->frequency = CGRAPH_FREQ_MAX;
  88     }
  89 }
  90
  91 /* We removed or are going to remove the last call to NODE.
  92    Return true if we can and want proactively remove the NODE now.
  93    This is important to do, since we want inliner to know when offline
  94    copy of function was removed.  */
  95
  96 static bool
  97 can_remove_node_now_p_1 (struct cgraph_node *node, struct cgraph_edge *e)
  98 {
  99   ipa_ref *ref;
 100
 101   FOR_EACH_ALIAS (node, ref)
 102     {
 103       cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
 104       if ((alias->callers && alias->callers != e)
 105           || !can_remove_node_now_p_1 (alias, e))
 106         return false;
 107     }
 108   /* FIXME: When address is taken of DECL_EXTERNAL function we still
 109      can remove its offline copy, but we would need to keep unanalyzed node in
 110      the callgraph so references can point to it.  */
 111   return (!node->address_taken
 112           && node->can_remove_if_no_direct_calls_p ()
 113           /* Inlining might enable more devirtualizing, so we want to remove
 114              those only after all devirtualizable virtual calls are processed.
 115              Lacking may edges in callgraph we just preserve them post
 116              inlining.  */
 117           && (!DECL_VIRTUAL_P (node->decl)
 118               || !opt_for_fn (node->decl, flag_devirtualize))
 119           /* During early inlining some unanalyzed cgraph nodes might be in the
 120              callgraph and they might reffer the function in question.  */
 121           && !cgraph_new_nodes.exists ());
 122 }
 123
 124 /* We are going to eliminate last direct call to NODE (or alias of it) via edge E.
 125    Verify that the NODE can be removed from unit and if it is contained in comdat
 126    group that the whole comdat group is removable.  */
 127
 128 static bool
 129 can_remove_node_now_p (struct cgraph_node *node, struct cgraph_edge *e)
 130 {
 131   struct cgraph_node *next;
 132   if (!can_remove_node_now_p_1 (node, e))
 133     return false;
 134
 135   /* When we see same comdat group, we need to be sure that all
 136      items can be removed.  */
 137   if (!node->same_comdat_group)
 138     return true;
 139   for (next = dyn_cast<cgraph_node *> (node->same_comdat_group);
 140        next != node; next = dyn_cast<cgraph_node *> (next->same_comdat_group))
 141     {
 142       if (next->alias)
 143         continue;
 144       if ((next->callers && next->callers != e)
 145           || !can_remove_node_now_p_1 (next, e))
 146         return false;
 147     }
 148   return true;
 149 }
 150
 151 /* Return true if NODE is a master clone with non-inline clones.  */
 152
 153 static bool
 154 master_clone_with_noninline_clones_p (struct cgraph_node *node)
 155 {
 156   if (node->clone_of)
 157     return false;
 158
 159   for (struct cgraph_node *n = node->clones; n; n = n->next_sibling_clone)
 160     if (n->decl != node->decl)
 161       return true;
 162
 163   return false;
 164 }
 165
 166 /* E is expected to be an edge being inlined.  Clone destination node of
 167    the edge and redirect it to the new clone.
 168    DUPLICATE is used for bookkeeping on whether we are actually creating new
 169    clones or re-using node originally representing out-of-line function call.
 170    By default the offline copy is removed, when it appears dead after inlining.
 171    UPDATE_ORIGINAL prevents this transformation.
 172    If OVERALL_SIZE is non-NULL, the size is updated to reflect the
 173    transformation.
 174    FREQ_SCALE specify the scaling of frequencies of call sites.  */
 175
 176 void
 177 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 178                      bool update_original, int *overall_size, int freq_scale)
 179 {
 180   struct cgraph_node *inlining_into;
 181   struct cgraph_edge *next;
 182
 183   if (e->caller->global.inlined_to)
 184     inlining_into = e->caller->global.inlined_to;
 185   else
 186     inlining_into = e->caller;
 187
 188   if (duplicate)
 189     {
 190       /* We may eliminate the need for out-of-line copy to be output.
 191          In that case just go ahead and re-use it.  This is not just an
 192          memory optimization.  Making offline copy of fuction disappear
 193          from the program will improve future decisions on inlining.  */
 194       if (!e->callee->callers->next_caller
 195           /* Recursive inlining never wants the master clone to
 196              be overwritten.  */
 197           && update_original
 198           && can_remove_node_now_p (e->callee, e)
 199           /* We cannot overwrite a master clone with non-inline clones
 200              until after these clones are materialized.  */
 201           && !master_clone_with_noninline_clones_p (e->callee))
 202         {
 203           /* TODO: When callee is in a comdat group, we could remove all of it,
 204              including all inline clones inlined into it.  That would however
 205              need small function inlining to register edge removal hook to
 206              maintain the priority queue.
 207
 208              For now we keep the ohter functions in the group in program until
 209              cgraph_remove_unreachable_functions gets rid of them.  */
 210           gcc_assert (!e->callee->global.inlined_to);
 211           e->callee->dissolve_same_comdat_group_list ();
 212           if (e->callee->definition && !DECL_EXTERNAL (e->callee->decl))
 213             {
 214               if (overall_size)
 215                 *overall_size -= inline_summaries->get (e->callee)->size;
 216               nfunctions_inlined++;
 217             }
 218           duplicate = false;
 219           e->callee->externally_visible = false;
 220           update_noncloned_frequencies (e->callee, e->frequency);
 221         }
 222       else
 223         {
 224           struct cgraph_node *n;
 225
 226           if (freq_scale == -1)
 227             freq_scale = e->frequency;
 228           n = e->callee->create_clone (e->callee->decl,
 229                                        MIN (e->count, e->callee->count),
 230                                        freq_scale,
 231                                        update_original, vNULL, true,
 232                                        inlining_into,
 233                                        NULL);
 234           n->used_as_abstract_origin = e->callee->used_as_abstract_origin;
 235           e->redirect_callee (n);
 236         }
 237     }
 238   else
 239     e->callee->dissolve_same_comdat_group_list ();
 240
 241   e->callee->global.inlined_to = inlining_into;
 242
 243   /* Recursively clone all bodies.  */
 244   for (e = e->callee->callees; e; e = next)
 245     {
 246       next = e->next_callee;
 247       if (!e->inline_failed)
 248         clone_inlined_nodes (e, duplicate, update_original, overall_size, freq_scale);
 249       if (e->speculative && !speculation_useful_p (e, true))
 250         {
 251           e->resolve_speculation (NULL);
 252           speculation_removed = true;
 253         }
 254     }
 255 }
 256
 257
 258 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 259    specify whether profile of original function should be updated.  If any new
 260    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 261    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 262    size of caller after inlining. Caller is required to eventually do it via
 263    inline_update_overall_summary.
 264    If callee_removed is non-NULL, set it to true if we removed callee node.
 265
 266    Return true iff any new callgraph edges were discovered as a
 267    result of inlining.  */
 268
 269 bool
 270 inline_call (struct cgraph_edge *e, bool update_original,
 271              vec<cgraph_edge *> *new_edges,
 272              int *overall_size, bool update_overall_summary,
 273              bool *callee_removed)
 274 {
 275   int old_size = 0, new_size = 0;
 276   struct cgraph_node *to = NULL;
 277   struct cgraph_edge *curr = e;
 278   struct cgraph_node *callee = e->callee->ultimate_alias_target ();
 279   bool new_edges_found = false;
 280
 281 #ifdef ENABLE_CHECKING
 282   int estimated_growth = estimate_edge_growth (e);
 283   bool predicated = inline_edge_summary (e)->predicate != NULL;
 284 #endif
 285
 286   speculation_removed = false;
 287   /* Don't inline inlined edges.  */
 288   gcc_assert (e->inline_failed);
 289   /* Don't even think of inlining inline clone.  */
 290   gcc_assert (!callee->global.inlined_to);
 291
 292   e->inline_failed = CIF_OK;
 293   DECL_POSSIBLY_INLINED (callee->decl) = true;
 294
 295   to = e->caller;
 296   if (to->global.inlined_to)
 297     to = to->global.inlined_to;
 298
 299   /* If aliases are involved, redirect edge to the actual destination and
 300      possibly remove the aliases.  */
 301   if (e->callee != callee)
 302     {
 303       struct cgraph_node *alias = e->callee, *next_alias;
 304       e->redirect_callee (callee);
 305       while (alias && alias != callee)
 306         {
 307           if (!alias->callers
 308               && can_remove_node_now_p (alias, e))
 309             {
 310               next_alias = alias->get_alias_target ();
 311               alias->remove ();
 312               if (callee_removed)
 313                 *callee_removed = true;
 314               alias = next_alias;
 315             }
 316           else
 317             break;
 318         }
 319     }
 320
 321   clone_inlined_nodes (e, true, update_original, overall_size, e->frequency);
 322
 323   gcc_assert (curr->callee->global.inlined_to == to);
 324
 325   old_size = inline_summaries->get (to)->size;
 326   inline_merge_summary (e);
 327   if (optimize)
 328     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 329   if (update_overall_summary)
 330    inline_update_overall_summary (to);
 331   new_size = inline_summaries->get (to)->size;
 332
 333   if (callee->calls_comdat_local)
 334     to->calls_comdat_local = true;
 335   else if (to->calls_comdat_local && callee->comdat_local_p ())
 336     {
 337       struct cgraph_edge *se = to->callees;
 338       for (; se; se = se->next_callee)
 339         if (se->inline_failed && se->callee->comdat_local_p ())
 340           break;
 341       if (se == NULL)
 342         to->calls_comdat_local = false;
 343     }
 344
 345 #ifdef ENABLE_CHECKING
 346   /* Verify that estimated growth match real growth.  Allow off-by-one
 347      error due to INLINE_SIZE_SCALE roudoff errors.  */
 348   gcc_assert (!update_overall_summary || !overall_size || new_edges_found
 349               || abs (estimated_growth - (new_size - old_size)) <= 1
 350               || speculation_removed
 351               /* FIXME: a hack.  Edges with false predicate are accounted
 352                  wrong, we should remove them from callgraph.  */
 353               || predicated);
 354 #endif
 355
 356   /* Account the change of overall unit size; external functions will be
 357      removed and are thus not accounted.  */
 358   if (overall_size
 359       && !DECL_EXTERNAL (to->decl))
 360     *overall_size += new_size - old_size;
 361   ncalls_inlined++;
 362
 363   /* This must happen after inline_merge_summary that rely on jump
 364      functions of callee to not be updated.  */
 365   return new_edges_found;
 366 }
 367
 368
 369 /* Copy function body of NODE and redirect all inline clones to it.
 370    This is done before inline plan is applied to NODE when there are
 371    still some inline clones if it.
 372
 373    This is necessary because inline decisions are not really transitive
 374    and the other inline clones may have different bodies.  */
 375
 376 static struct cgraph_node *
 377 save_inline_function_body (struct cgraph_node *node)
 378 {
 379   struct cgraph_node *first_clone, *n;
 380
 381   if (dump_file)
 382     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 383              node->name ());
 384
 385   gcc_assert (node == cgraph_node::get (node->decl));
 386
 387   /* first_clone will be turned into real function.  */
 388   first_clone = node->clones;
 389   first_clone->decl = copy_node (node->decl);
 390   first_clone->decl->decl_with_vis.symtab_node = first_clone;
 391   gcc_assert (first_clone == cgraph_node::get (first_clone->decl));
 392
 393   /* Now reshape the clone tree, so all other clones descends from
 394      first_clone.  */
 395   if (first_clone->next_sibling_clone)
 396     {
 397       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 398         n->clone_of = first_clone;
 399       n->clone_of = first_clone;
 400       n->next_sibling_clone = first_clone->clones;
 401       if (first_clone->clones)
 402         first_clone->clones->prev_sibling_clone = n;
 403       first_clone->clones = first_clone->next_sibling_clone;
 404       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 405       first_clone->next_sibling_clone = NULL;
 406       gcc_assert (!first_clone->prev_sibling_clone);
 407     }
 408   first_clone->clone_of = NULL;
 409
 410   /* Now node in question has no clones.  */
 411   node->clones = NULL;
 412
 413   /* Inline clones share decl with the function they are cloned
 414      from.  Walk the whole clone tree and redirect them all to the
 415      new decl.  */
 416   if (first_clone->clones)
 417     for (n = first_clone->clones; n != first_clone;)
 418       {
 419         gcc_assert (n->decl == node->decl);
 420         n->decl = first_clone->decl;
 421         if (n->clones)
 422           n = n->clones;
 423         else if (n->next_sibling_clone)
 424           n = n->next_sibling_clone;
 425         else
 426           {
 427             while (n != first_clone && !n->next_sibling_clone)
 428               n = n->clone_of;
 429             if (n != first_clone)
 430               n = n->next_sibling_clone;
 431           }
 432       }
 433
 434   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 435   tree_function_versioning (node->decl, first_clone->decl,
 436                             NULL, true, NULL, false,
 437                             NULL, NULL);
 438
 439   /* The function will be short lived and removed after we inline all the clones,
 440      but make it internal so we won't confuse ourself.  */
 441   DECL_EXTERNAL (first_clone->decl) = 0;
 442   TREE_PUBLIC (first_clone->decl) = 0;
 443   DECL_COMDAT (first_clone->decl) = 0;
 444   first_clone->ipa_transforms_to_apply.release ();
 445
 446   /* When doing recursive inlining, the clone may become unnecessary.
 447      This is possible i.e. in the case when the recursive function is proved to be
 448      non-throwing and the recursion happens only in the EH landing pad.
 449      We can not remove the clone until we are done with saving the body.
 450      Remove it now.  */
 451   if (!first_clone->callers)
 452     {
 453       first_clone->remove_symbol_and_inline_clones ();
 454       first_clone = NULL;
 455     }
 456 #ifdef ENABLE_CHECKING
 457   else
 458     first_clone->verify ();
 459 #endif
 460   return first_clone;
 461 }
 462
 463 /* Return true when function body of DECL still needs to be kept around
 464    for later re-use.  */
 465 static bool
 466 preserve_function_body_p (struct cgraph_node *node)
 467 {
 468   gcc_assert (symtab->global_info_ready);
 469   gcc_assert (!node->alias && !node->thunk.thunk_p);
 470
 471   /* Look if there is any clone around.  */
 472   if (node->clones)
 473     return true;
 474   return false;
 475 }
 476
 477 /* Apply inline plan to function.  */
 478
 479 unsigned int
 480 inline_transform (struct cgraph_node *node)
 481 {
 482   unsigned int todo = 0;
 483   struct cgraph_edge *e, *next;
 484   bool has_inline = false;
 485
 486   /* FIXME: Currently the pass manager is adding inline transform more than
 487      once to some clones.  This needs revisiting after WPA cleanups.  */
 488   if (cfun->after_inlining)
 489     return 0;
 490
 491   /* We might need the body of this function so that we can expand
 492      it inline somewhere else.  */
 493   if (preserve_function_body_p (node))
 494     save_inline_function_body (node);
 495
 496   for (e = node->callees; e; e = next)
 497     {
 498       if (!e->inline_failed)
 499         has_inline = true;
 500       next = e->next_callee;
 501       e->redirect_call_stmt_to_callee ();
 502     }
 503   node->remove_all_references ();
 504
 505   timevar_push (TV_INTEGRATION);
 506   if (node->callees && (optimize || has_inline))
 507     todo = optimize_inline_calls (current_function_decl);
 508   timevar_pop (TV_INTEGRATION);
 509
 510   cfun->always_inline_functions_inlined = true;
 511   cfun->after_inlining = true;
 512   todo |= execute_fixup_cfg ();
 513
 514   if (!(todo & TODO_update_ssa_any))
 515     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 516     todo |= TODO_update_ssa_only_virtuals;
 517
 518   return todo;
 519 }