gcc/ipa-inline-transform.c

   1 /* Callgraph transformations to handle inlining
   2    Copyright (C) 2003-2015 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 "hash-set.h"
  36 #include "machmode.h"
  37 #include "vec.h"
  38 #include "double-int.h"
  39 #include "input.h"
  40 #include "alias.h"
  41 #include "symtab.h"
  42 #include "wide-int.h"
  43 #include "inchash.h"
  44 #include "tree.h"
  45 #include "langhooks.h"
  46 #include "intl.h"
  47 #include "coverage.h"
  48 #include "ggc.h"
  49 #include "tree-cfg.h"
  50 #include "hash-map.h"
  51 #include "is-a.h"
  52 #include "plugin-api.h"
  53 #include "hard-reg-set.h"
  54 #include "input.h"
  55 #include "function.h"
  56 #include "ipa-ref.h"
  57 #include "cgraph.h"
  58 #include "alloc-pool.h"
  59 #include "symbol-summary.h"
  60 #include "ipa-prop.h"
  61 #include "ipa-inline.h"
  62 #include "tree-inline.h"
  63 #include "tree-pass.h"
  64
  65 int ncalls_inlined;
  66 int nfunctions_inlined;
  67 bool speculation_removed;
  68
  69 /* Scale frequency of NODE edges by FREQ_SCALE.  */
  70
  71 static void
  72 update_noncloned_frequencies (struct cgraph_node *node,
  73                               int freq_scale)
  74 {
  75   struct cgraph_edge *e;
  76
  77   /* We do not want to ignore high loop nest after freq drops to 0.  */
  78   if (!freq_scale)
  79     freq_scale = 1;
  80   for (e = node->callees; e; e = e->next_callee)
  81     {
  82       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  83       if (e->frequency > CGRAPH_FREQ_MAX)
  84         e->frequency = CGRAPH_FREQ_MAX;
  85       if (!e->inline_failed)
  86         update_noncloned_frequencies (e->callee, freq_scale);
  87     }
  88   for (e = node->indirect_calls; e; e = e->next_callee)
  89     {
  90       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  91       if (e->frequency > CGRAPH_FREQ_MAX)
  92         e->frequency = CGRAPH_FREQ_MAX;
  93     }
  94 }
  95
  96 /* We removed or are going to remove the last call to NODE.
  97    Return true if we can and want proactively remove the NODE now.
  98    This is important to do, since we want inliner to know when offline
  99    copy of function was removed.  */
 100
 101 static bool
 102 can_remove_node_now_p_1 (struct cgraph_node *node, struct cgraph_edge *e)
 103 {
 104   ipa_ref *ref;
 105
 106   FOR_EACH_ALIAS (node, ref)
 107     {
 108       cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
 109       if ((alias->callers && alias->callers != e)
 110           || !can_remove_node_now_p_1 (alias, e))
 111         return false;
 112     }
 113   /* FIXME: When address is taken of DECL_EXTERNAL function we still
 114      can remove its offline copy, but we would need to keep unanalyzed node in
 115      the callgraph so references can point to it.  */
 116   return (!node->address_taken
 117           && node->can_remove_if_no_direct_calls_p ()
 118           /* Inlining might enable more devirtualizing, so we want to remove
 119              those only after all devirtualizable virtual calls are processed.
 120              Lacking may edges in callgraph we just preserve them post
 121              inlining.  */
 122           && (!DECL_VIRTUAL_P (node->decl)
 123               || !opt_for_fn (node->decl, flag_devirtualize))
 124           /* During early inlining some unanalyzed cgraph nodes might be in the
 125              callgraph and they might reffer the function in question.  */
 126           && !cgraph_new_nodes.exists ());
 127 }
 128
 129 /* We are going to eliminate last direct call to NODE (or alias of it) via edge E.
 130    Verify that the NODE can be removed from unit and if it is contained in comdat
 131    group that the whole comdat group is removable.  */
 132
 133 static bool
 134 can_remove_node_now_p (struct cgraph_node *node, struct cgraph_edge *e)
 135 {
 136   struct cgraph_node *next;
 137   if (!can_remove_node_now_p_1 (node, e))
 138     return false;
 139
 140   /* When we see same comdat group, we need to be sure that all
 141      items can be removed.  */
 142   if (!node->same_comdat_group)
 143     return true;
 144   for (next = dyn_cast<cgraph_node *> (node->same_comdat_group);
 145        next != node; next = dyn_cast<cgraph_node *> (next->same_comdat_group))
 146     {
 147       if (next->alias)
 148         continue;
 149       if ((next->callers && next->callers != e)
 150           || !can_remove_node_now_p_1 (next, e))
 151         return false;
 152     }
 153   return true;
 154 }
 155
 156 /* Return true if NODE is a master clone with non-inline clones.  */
 157
 158 static bool
 159 master_clone_with_noninline_clones_p (struct cgraph_node *node)
 160 {
 161   if (node->clone_of)
 162     return false;
 163
 164   for (struct cgraph_node *n = node->clones; n; n = n->next_sibling_clone)
 165     if (n->decl != node->decl)
 166       return true;
 167
 168   return false;
 169 }
 170
 171 /* E is expected to be an edge being inlined.  Clone destination node of
 172    the edge and redirect it to the new clone.
 173    DUPLICATE is used for bookkeeping on whether we are actually creating new
 174    clones or re-using node originally representing out-of-line function call.
 175    By default the offline copy is removed, when it appears dead after inlining.
 176    UPDATE_ORIGINAL prevents this transformation.
 177    If OVERALL_SIZE is non-NULL, the size is updated to reflect the
 178    transformation.
 179    FREQ_SCALE specify the scaling of frequencies of call sites.  */
 180
 181 void
 182 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 183                      bool update_original, int *overall_size, int freq_scale)
 184 {
 185   struct cgraph_node *inlining_into;
 186   struct cgraph_edge *next;
 187
 188   if (e->caller->global.inlined_to)
 189     inlining_into = e->caller->global.inlined_to;
 190   else
 191     inlining_into = e->caller;
 192
 193   if (duplicate)
 194     {
 195       /* We may eliminate the need for out-of-line copy to be output.
 196          In that case just go ahead and re-use it.  This is not just an
 197          memory optimization.  Making offline copy of fuction disappear
 198          from the program will improve future decisions on inlining.  */
 199       if (!e->callee->callers->next_caller
 200           /* Recursive inlining never wants the master clone to
 201              be overwritten.  */
 202           && update_original
 203           && can_remove_node_now_p (e->callee, e)
 204           /* We cannot overwrite a master clone with non-inline clones
 205              until after these clones are materialized.  */
 206           && !master_clone_with_noninline_clones_p (e->callee))
 207         {
 208           /* TODO: When callee is in a comdat group, we could remove all of it,
 209              including all inline clones inlined into it.  That would however
 210              need small function inlining to register edge removal hook to
 211              maintain the priority queue.
 212
 213              For now we keep the ohter functions in the group in program until
 214              cgraph_remove_unreachable_functions gets rid of them.  */
 215           gcc_assert (!e->callee->global.inlined_to);
 216           e->callee->dissolve_same_comdat_group_list ();
 217           if (e->callee->definition && !DECL_EXTERNAL (e->callee->decl))
 218             {
 219               if (overall_size)
 220                 *overall_size -= inline_summaries->get (e->callee)->size;
 221               nfunctions_inlined++;
 222             }
 223           duplicate = false;
 224           e->callee->externally_visible = false;
 225           update_noncloned_frequencies (e->callee, e->frequency);
 226         }
 227       else
 228         {
 229           struct cgraph_node *n;
 230
 231           if (freq_scale == -1)
 232             freq_scale = e->frequency;
 233           n = e->callee->create_clone (e->callee->decl,
 234                                        MIN (e->count, e->callee->count),
 235                                        freq_scale,
 236                                        update_original, vNULL, true,
 237                                        inlining_into,
 238                                        NULL);
 239           n->used_as_abstract_origin = e->callee->used_as_abstract_origin;
 240           e->redirect_callee (n);
 241         }
 242     }
 243   else
 244     e->callee->dissolve_same_comdat_group_list ();
 245
 246   e->callee->global.inlined_to = inlining_into;
 247
 248   /* Recursively clone all bodies.  */
 249   for (e = e->callee->callees; e; e = next)
 250     {
 251       next = e->next_callee;
 252       if (!e->inline_failed)
 253         clone_inlined_nodes (e, duplicate, update_original, overall_size, freq_scale);
 254       if (e->speculative && !speculation_useful_p (e, true))
 255         {
 256           e->resolve_speculation (NULL);
 257           speculation_removed = true;
 258         }
 259     }
 260 }
 261
 262
 263 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 264    specify whether profile of original function should be updated.  If any new
 265    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 266    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 267    size of caller after inlining. Caller is required to eventually do it via
 268    inline_update_overall_summary.
 269    If callee_removed is non-NULL, set it to true if we removed callee node.
 270
 271    Return true iff any new callgraph edges were discovered as a
 272    result of inlining.  */
 273
 274 bool
 275 inline_call (struct cgraph_edge *e, bool update_original,
 276              vec<cgraph_edge *> *new_edges,
 277              int *overall_size, bool update_overall_summary,
 278              bool *callee_removed)
 279 {
 280   int old_size = 0, new_size = 0;
 281   struct cgraph_node *to = NULL;
 282   struct cgraph_edge *curr = e;
 283   struct cgraph_node *callee = e->callee->ultimate_alias_target ();
 284   bool new_edges_found = false;
 285
 286 #ifdef ENABLE_CHECKING
 287   int estimated_growth = estimate_edge_growth (e);
 288   bool predicated = inline_edge_summary (e)->predicate != NULL;
 289 #endif
 290
 291   speculation_removed = false;
 292   /* Don't inline inlined edges.  */
 293   gcc_assert (e->inline_failed);
 294   /* Don't even think of inlining inline clone.  */
 295   gcc_assert (!callee->global.inlined_to);
 296
 297   e->inline_failed = CIF_OK;
 298   DECL_POSSIBLY_INLINED (callee->decl) = true;
 299
 300   to = e->caller;
 301   if (to->global.inlined_to)
 302     to = to->global.inlined_to;
 303
 304   /* If aliases are involved, redirect edge to the actual destination and
 305      possibly remove the aliases.  */
 306   if (e->callee != callee)
 307     {
 308       struct cgraph_node *alias = e->callee, *next_alias;
 309       e->redirect_callee (callee);
 310       while (alias && alias != callee)
 311         {
 312           if (!alias->callers
 313               && can_remove_node_now_p (alias, e))
 314             {
 315               next_alias = alias->get_alias_target ();
 316               alias->remove ();
 317               if (callee_removed)
 318                 *callee_removed = true;
 319               alias = next_alias;
 320             }
 321           else
 322             break;
 323         }
 324     }
 325
 326   clone_inlined_nodes (e, true, update_original, overall_size, e->frequency);
 327
 328   gcc_assert (curr->callee->global.inlined_to == to);
 329
 330   old_size = inline_summaries->get (to)->size;
 331   inline_merge_summary (e);
 332   if (optimize)
 333     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 334   if (update_overall_summary)
 335    inline_update_overall_summary (to);
 336   new_size = inline_summaries->get (to)->size;
 337
 338   if (callee->calls_comdat_local)
 339     to->calls_comdat_local = true;
 340   else if (to->calls_comdat_local && callee->comdat_local_p ())
 341     {
 342       struct cgraph_edge *se = to->callees;
 343       for (; se; se = se->next_callee)
 344         if (se->inline_failed && se->callee->comdat_local_p ())
 345           break;
 346       if (se == NULL)
 347         to->calls_comdat_local = false;
 348     }
 349
 350 #ifdef ENABLE_CHECKING
 351   /* Verify that estimated growth match real growth.  Allow off-by-one
 352      error due to INLINE_SIZE_SCALE roudoff errors.  */
 353   gcc_assert (!update_overall_summary || !overall_size || new_edges_found
 354               || abs (estimated_growth - (new_size - old_size)) <= 1
 355               || speculation_removed
 356               /* FIXME: a hack.  Edges with false predicate are accounted
 357                  wrong, we should remove them from callgraph.  */
 358               || predicated);
 359 #endif
 360
 361   /* Account the change of overall unit size; external functions will be
 362      removed and are thus not accounted.  */
 363   if (overall_size
 364       && !DECL_EXTERNAL (to->decl))
 365     *overall_size += new_size - old_size;
 366   ncalls_inlined++;
 367
 368   /* This must happen after inline_merge_summary that rely on jump
 369      functions of callee to not be updated.  */
 370   return new_edges_found;
 371 }
 372
 373
 374 /* Copy function body of NODE and redirect all inline clones to it.
 375    This is done before inline plan is applied to NODE when there are
 376    still some inline clones if it.
 377
 378    This is necessary because inline decisions are not really transitive
 379    and the other inline clones may have different bodies.  */
 380
 381 static struct cgraph_node *
 382 save_inline_function_body (struct cgraph_node *node)
 383 {
 384   struct cgraph_node *first_clone, *n;
 385
 386   if (dump_file)
 387     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 388              node->name ());
 389
 390   gcc_assert (node == cgraph_node::get (node->decl));
 391
 392   /* first_clone will be turned into real function.  */
 393   first_clone = node->clones;
 394   first_clone->decl = copy_node (node->decl);
 395   first_clone->decl->decl_with_vis.symtab_node = first_clone;
 396   gcc_assert (first_clone == cgraph_node::get (first_clone->decl));
 397
 398   /* Now reshape the clone tree, so all other clones descends from
 399      first_clone.  */
 400   if (first_clone->next_sibling_clone)
 401     {
 402       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 403         n->clone_of = first_clone;
 404       n->clone_of = first_clone;
 405       n->next_sibling_clone = first_clone->clones;
 406       if (first_clone->clones)
 407         first_clone->clones->prev_sibling_clone = n;
 408       first_clone->clones = first_clone->next_sibling_clone;
 409       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 410       first_clone->next_sibling_clone = NULL;
 411       gcc_assert (!first_clone->prev_sibling_clone);
 412     }
 413   first_clone->clone_of = NULL;
 414
 415   /* Now node in question has no clones.  */
 416   node->clones = NULL;
 417
 418   /* Inline clones share decl with the function they are cloned
 419      from.  Walk the whole clone tree and redirect them all to the
 420      new decl.  */
 421   if (first_clone->clones)
 422     for (n = first_clone->clones; n != first_clone;)
 423       {
 424         gcc_assert (n->decl == node->decl);
 425         n->decl = first_clone->decl;
 426         if (n->clones)
 427           n = n->clones;
 428         else if (n->next_sibling_clone)
 429           n = n->next_sibling_clone;
 430         else
 431           {
 432             while (n != first_clone && !n->next_sibling_clone)
 433               n = n->clone_of;
 434             if (n != first_clone)
 435               n = n->next_sibling_clone;
 436           }
 437       }
 438
 439   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 440   tree_function_versioning (node->decl, first_clone->decl,
 441                             NULL, true, NULL, false,
 442                             NULL, NULL);
 443
 444   /* The function will be short lived and removed after we inline all the clones,
 445      but make it internal so we won't confuse ourself.  */
 446   DECL_EXTERNAL (first_clone->decl) = 0;
 447   TREE_PUBLIC (first_clone->decl) = 0;
 448   DECL_COMDAT (first_clone->decl) = 0;
 449   first_clone->ipa_transforms_to_apply.release ();
 450
 451   /* When doing recursive inlining, the clone may become unnecessary.
 452      This is possible i.e. in the case when the recursive function is proved to be
 453      non-throwing and the recursion happens only in the EH landing pad.
 454      We can not remove the clone until we are done with saving the body.
 455      Remove it now.  */
 456   if (!first_clone->callers)
 457     {
 458       first_clone->remove_symbol_and_inline_clones ();
 459       first_clone = NULL;
 460     }
 461 #ifdef ENABLE_CHECKING
 462   else
 463     first_clone->verify ();
 464 #endif
 465   return first_clone;
 466 }
 467
 468 /* Return true when function body of DECL still needs to be kept around
 469    for later re-use.  */
 470 static bool
 471 preserve_function_body_p (struct cgraph_node *node)
 472 {
 473   gcc_assert (symtab->global_info_ready);
 474   gcc_assert (!node->alias && !node->thunk.thunk_p);
 475
 476   /* Look if there is any clone around.  */
 477   if (node->clones)
 478     return true;
 479   return false;
 480 }
 481
 482 /* Apply inline plan to function.  */
 483
 484 unsigned int
 485 inline_transform (struct cgraph_node *node)
 486 {
 487   unsigned int todo = 0;
 488   struct cgraph_edge *e, *next;
 489   bool has_inline = false;
 490
 491   /* FIXME: Currently the pass manager is adding inline transform more than
 492      once to some clones.  This needs revisiting after WPA cleanups.  */
 493   if (cfun->after_inlining)
 494     return 0;
 495
 496   /* We might need the body of this function so that we can expand
 497      it inline somewhere else.  */
 498   if (preserve_function_body_p (node))
 499     save_inline_function_body (node);
 500
 501   for (e = node->callees; e; e = next)
 502     {
 503       if (!e->inline_failed)
 504         has_inline = true;
 505       next = e->next_callee;
 506       e->redirect_call_stmt_to_callee ();
 507     }
 508   node->remove_all_references ();
 509
 510   timevar_push (TV_INTEGRATION);
 511   if (node->callees && (optimize || has_inline))
 512     todo = optimize_inline_calls (current_function_decl);
 513   timevar_pop (TV_INTEGRATION);
 514
 515   cfun->always_inline_functions_inlined = true;
 516   cfun->after_inlining = true;
 517   todo |= execute_fixup_cfg ();
 518
 519   if (!(todo & TODO_update_ssa_any))
 520     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 521     todo |= TODO_update_ssa_only_virtuals;
 522
 523   return todo;
 524 }