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           && cgraph_can_remove_if_no_direct_calls_p (node)
  91           /* Inlining might enable more devirtualizing, so we want to remove
  92              those only after all devirtualizable virtual calls are processed.
  93              Lacking may edges in callgraph we just preserve them post
  94              inlining.  */
  95           && !DECL_VIRTUAL_P (node->symbol.decl)
  96           /* During early inlining some unanalyzed cgraph nodes might be in the
  97              callgraph and they might reffer the function in question.  */
  98           && !cgraph_new_nodes);
  99 }
 100
 101 /* We are going to eliminate last direct call to NODE (or alias of it) via edge E.
 102    Verify that the NODE can be removed from unit and if it is contained in comdat
 103    group that the whole comdat group is removable.  */
 104
 105 static bool
 106 can_remove_node_now_p (struct cgraph_node *node, struct cgraph_edge *e)
 107 {
 108   struct cgraph_node *next;
 109   if (!can_remove_node_now_p_1 (node))
 110     return false;
 111
 112   /* When we see same comdat group, we need to be sure that all
 113      items can be removed.  */
 114   if (!node->symbol.same_comdat_group)
 115     return true;
 116   for (next = cgraph (node->symbol.same_comdat_group);
 117        next != node; next = cgraph (next->symbol.same_comdat_group))
 118     if ((next->callers && next->callers != e)
 119         || !can_remove_node_now_p_1 (next))
 120       return false;
 121   return true;
 122 }
 123
 124
 125 /* E is expected to be an edge being inlined.  Clone destination node of
 126    the edge and redirect it to the new clone.
 127    DUPLICATE is used for bookkeeping on whether we are actually creating new
 128    clones or re-using node originally representing out-of-line function call.
 129    */
 130
 131 void
 132 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 133                      bool update_original, int *overall_size)
 134 {
 135   if (duplicate)
 136     {
 137       /* We may eliminate the need for out-of-line copy to be output.
 138          In that case just go ahead and re-use it.  This is not just an
 139          memory optimization.  Making offline copy of fuction disappear
 140          from the program will improve future decisions on inlining.  */
 141       if (!e->callee->callers->next_caller
 142           /* Recursive inlining never wants the master clone to
 143              be overwritten.  */
 144           && update_original
 145           && can_remove_node_now_p (e->callee, e))
 146         {
 147           /* TODO: When callee is in a comdat group, we could remove all of it,
 148              including all inline clones inlined into it.  That would however
 149              need small function inlining to register edge removal hook to
 150              maintain the priority queue.
 151
 152              For now we keep the ohter functions in the group in program until
 153              cgraph_remove_unreachable_functions gets rid of them.  */
 154           gcc_assert (!e->callee->global.inlined_to);
 155           symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 156           if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->symbol.decl))
 157             {
 158               if (overall_size)
 159                 *overall_size -= inline_summary (e->callee)->size;
 160               nfunctions_inlined++;
 161             }
 162           duplicate = false;
 163           e->callee->symbol.externally_visible = false;
 164           update_noncloned_frequencies (e->callee, e->frequency);
 165         }
 166       else
 167         {
 168           struct cgraph_node *n;
 169           n = cgraph_clone_node (e->callee, e->callee->symbol.decl,
 170                                  e->count, e->frequency,
 171                                  update_original, vNULL, true);
 172           cgraph_redirect_edge_callee (e, n);
 173         }
 174     }
 175   else
 176     symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 177
 178   if (e->caller->global.inlined_to)
 179     e->callee->global.inlined_to = e->caller->global.inlined_to;
 180   else
 181     e->callee->global.inlined_to = e->caller;
 182
 183   /* Recursively clone all bodies.  */
 184   for (e = e->callee->callees; e; e = e->next_callee)
 185     if (!e->inline_failed)
 186       clone_inlined_nodes (e, duplicate, update_original, overall_size);
 187 }
 188
 189
 190 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 191    specify whether profile of original function should be updated.  If any new
 192    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 193    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 194    size of caller after inlining. Caller is required to eventually do it via
 195    inline_update_overall_summary.
 196
 197    Return true iff any new callgraph edges were discovered as a
 198    result of inlining.  */
 199
 200 bool
 201 inline_call (struct cgraph_edge *e, bool update_original,
 202              vec<cgraph_edge_p> *new_edges,
 203              int *overall_size, bool update_overall_summary)
 204 {
 205   int old_size = 0, new_size = 0;
 206   struct cgraph_node *to = NULL;
 207   struct cgraph_edge *curr = e;
 208   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 209   bool new_edges_found = false;
 210
 211 #ifdef ENABLE_CHECKING
 212   int estimated_growth = estimate_edge_growth (e);
 213   bool predicated = inline_edge_summary (e)->predicate != NULL;
 214 #endif
 215
 216   /* Don't inline inlined edges.  */
 217   gcc_assert (e->inline_failed);
 218   /* Don't even think of inlining inline clone.  */
 219   gcc_assert (!callee->global.inlined_to);
 220
 221   e->inline_failed = CIF_OK;
 222   DECL_POSSIBLY_INLINED (callee->symbol.decl) = true;
 223
 224   to = e->caller;
 225   if (to->global.inlined_to)
 226     to = to->global.inlined_to;
 227
 228   /* If aliases are involved, redirect edge to the actual destination and
 229      possibly remove the aliases.  */
 230   if (e->callee != callee)
 231     {
 232       struct cgraph_node *alias = e->callee, *next_alias;
 233       cgraph_redirect_edge_callee (e, callee);
 234       while (alias && alias != callee)
 235         {
 236           if (!alias->callers
 237               && can_remove_node_now_p (alias, e))
 238             {
 239               next_alias = cgraph_alias_aliased_node (alias);
 240               cgraph_remove_node (alias);
 241               alias = next_alias;
 242             }
 243           else
 244             break;
 245         }
 246     }
 247
 248   clone_inlined_nodes (e, true, update_original, overall_size);
 249
 250   gcc_assert (curr->callee->global.inlined_to == to);
 251
 252   old_size = inline_summary (to)->size;
 253   inline_merge_summary (e);
 254   if (optimize)
 255     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 256   if (update_overall_summary)
 257    inline_update_overall_summary (to);
 258   new_size = inline_summary (to)->size;
 259
 260 #ifdef ENABLE_CHECKING
 261   /* Verify that estimated growth match real growth.  Allow off-by-one
 262      error due to INLINE_SIZE_SCALE roudoff errors.  */
 263   gcc_assert (!update_overall_summary || !overall_size
 264               || abs (estimated_growth - (new_size - old_size)) <= 1
 265               /* FIXME: a hack.  Edges with false predicate are accounted
 266                  wrong, we should remove them from callgraph.  */
 267               || predicated);
 268 #endif
 269
 270   /* Account the change of overall unit size; external functions will be
 271      removed and are thus not accounted.  */
 272   if (overall_size
 273       && !DECL_EXTERNAL (to->symbol.decl))
 274     *overall_size += new_size - old_size;
 275   ncalls_inlined++;
 276
 277   /* This must happen after inline_merge_summary that rely on jump
 278      functions of callee to not be updated.  */
 279   return new_edges_found;
 280 }
 281
 282
 283 /* Copy function body of NODE and redirect all inline clones to it.
 284    This is done before inline plan is applied to NODE when there are
 285    still some inline clones if it.
 286
 287    This is necessary because inline decisions are not really transitive
 288    and the other inline clones may have different bodies.  */
 289
 290 static struct cgraph_node *
 291 save_inline_function_body (struct cgraph_node *node)
 292 {
 293   struct cgraph_node *first_clone, *n;
 294
 295   if (dump_file)
 296     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 297              cgraph_node_name (node));
 298
 299   gcc_assert (node == cgraph_get_node (node->symbol.decl));
 300
 301   /* first_clone will be turned into real function.  */
 302   first_clone = node->clones;
 303   first_clone->symbol.decl = copy_node (node->symbol.decl);
 304   symtab_insert_node_to_hashtable ((symtab_node) first_clone);
 305   gcc_assert (first_clone == cgraph_get_node (first_clone->symbol.decl));
 306
 307   /* Now reshape the clone tree, so all other clones descends from
 308      first_clone.  */
 309   if (first_clone->next_sibling_clone)
 310     {
 311       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 312         n->clone_of = first_clone;
 313       n->clone_of = first_clone;
 314       n->next_sibling_clone = first_clone->clones;
 315       if (first_clone->clones)
 316         first_clone->clones->prev_sibling_clone = n;
 317       first_clone->clones = first_clone->next_sibling_clone;
 318       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 319       first_clone->next_sibling_clone = NULL;
 320       gcc_assert (!first_clone->prev_sibling_clone);
 321     }
 322   first_clone->clone_of = NULL;
 323
 324   /* Now node in question has no clones.  */
 325   node->clones = NULL;
 326
 327   /* Inline clones share decl with the function they are cloned
 328      from.  Walk the whole clone tree and redirect them all to the
 329      new decl.  */
 330   if (first_clone->clones)
 331     for (n = first_clone->clones; n != first_clone;)
 332       {
 333         gcc_assert (n->symbol.decl == node->symbol.decl);
 334         n->symbol.decl = first_clone->symbol.decl;
 335         if (n->clones)
 336           n = n->clones;
 337         else if (n->next_sibling_clone)
 338           n = n->next_sibling_clone;
 339         else
 340           {
 341             while (n != first_clone && !n->next_sibling_clone)
 342               n = n->clone_of;
 343             if (n != first_clone)
 344               n = n->next_sibling_clone;
 345           }
 346       }
 347
 348   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 349   tree_function_versioning (node->symbol.decl, first_clone->symbol.decl,
 350                             NULL, true, NULL, false,
 351                             NULL, NULL);
 352
 353   /* The function will be short lived and removed after we inline all the clones,
 354      but make it internal so we won't confuse ourself.  */
 355   DECL_EXTERNAL (first_clone->symbol.decl) = 0;
 356   DECL_COMDAT_GROUP (first_clone->symbol.decl) = NULL_TREE;
 357   TREE_PUBLIC (first_clone->symbol.decl) = 0;
 358   DECL_COMDAT (first_clone->symbol.decl) = 0;
 359   first_clone->ipa_transforms_to_apply.release ();
 360
 361   /* When doing recursive inlining, the clone may become unnecessary.
 362      This is possible i.e. in the case when the recursive function is proved to be
 363      non-throwing and the recursion happens only in the EH landing pad.
 364      We can not remove the clone until we are done with saving the body.
 365      Remove it now.  */
 366   if (!first_clone->callers)
 367     {
 368       cgraph_remove_node_and_inline_clones (first_clone, NULL);
 369       first_clone = NULL;
 370     }
 371 #ifdef ENABLE_CHECKING
 372   else
 373     verify_cgraph_node (first_clone);
 374 #endif
 375   return first_clone;
 376 }
 377
 378 /* Return true when function body of DECL still needs to be kept around
 379    for later re-use.  */
 380 static bool
 381 preserve_function_body_p (struct cgraph_node *node)
 382 {
 383   gcc_assert (cgraph_global_info_ready);
 384   gcc_assert (!node->alias && !node->thunk.thunk_p);
 385
 386   /* Look if there is any clone around.  */
 387   if (node->clones)
 388     return true;
 389   return false;
 390 }
 391
 392 /* Apply inline plan to function.  */
 393
 394 unsigned int
 395 inline_transform (struct cgraph_node *node)
 396 {
 397   unsigned int todo = 0;
 398   struct cgraph_edge *e;
 399
 400   /* FIXME: Currently the pass manager is adding inline transform more than
 401      once to some clones.  This needs revisiting after WPA cleanups.  */
 402   if (cfun->after_inlining)
 403     return 0;
 404
 405   /* We might need the body of this function so that we can expand
 406      it inline somewhere else.  */
 407   if (preserve_function_body_p (node))
 408     save_inline_function_body (node);
 409
 410   for (e = node->callees; e; e = e->next_callee)
 411     cgraph_redirect_edge_call_stmt_to_callee (e);
 412
 413   timevar_push (TV_INTEGRATION);
 414   if (node->callees)
 415     todo = optimize_inline_calls (current_function_decl);
 416   timevar_pop (TV_INTEGRATION);
 417
 418   cfun->always_inline_functions_inlined = true;
 419   cfun->after_inlining = true;
 420   todo |= execute_fixup_cfg ();
 421
 422   if (!(todo & TODO_update_ssa_any))
 423     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 424     todo |= TODO_update_ssa_only_virtuals;
 425
 426   return todo;
 427 }