gcc/ipa-inline-transform.c

   1 /* Callgraph transformations to handle inlining
   2    Copyright (C) 2003, 2004, 2007, 2008, 2009, 2010, 2011
   3    Free Software Foundation, Inc.
   4    Contributed by Jan Hubicka
   5
   6 This file is part of GCC.
   7
   8 GCC is free software; you can redistribute it and/or modify it under
   9 the terms of the GNU General Public License as published by the Free
  10 Software Foundation; either version 3, or (at your option) any later
  11 version.
  12
  13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  15 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  16 for more details.
  17
  18 You should have received a copy of the GNU General Public License
  19 along with GCC; see the file COPYING3.  If not see
  20 <http://www.gnu.org/licenses/>.  */
  21
  22 /* The inline decisions are stored in callgraph in "inline plan" and
  23    applied later.
  24
  25    To mark given call inline, use inline_call function.
  26    The function marks the edge inlinable and, if necessary, produces
  27    virtual clone in the callgraph representing the new copy of callee's
  28    function body.
  29
  30    The inline plan is applied on given function body by inline_transform.  */
  31
  32 #include "config.h"
  33 #include "system.h"
  34 #include "coretypes.h"
  35 #include "tm.h"
  36 #include "tree.h"
  37 #include "langhooks.h"
  38 #include "cgraph.h"
  39 #include "timevar.h"
  40 #include "output.h"
  41 #include "intl.h"
  42 #include "coverage.h"
  43 #include "ggc.h"
  44 #include "tree-flow.h"
  45 #include "ipa-prop.h"
  46 #include "ipa-inline.h"
  47 #include "tree-inline.h"
  48 #include "tree-pass.h"
  49
  50 int ncalls_inlined;
  51 int nfunctions_inlined;
  52
  53 /* Scale frequency of NODE edges by FREQ_SCALE.  */
  54
  55 static void
  56 update_noncloned_frequencies (struct cgraph_node *node,
  57                               int freq_scale)
  58 {
  59   struct cgraph_edge *e;
  60
  61   /* We do not want to ignore high loop nest after freq drops to 0.  */
  62   if (!freq_scale)
  63     freq_scale = 1;
  64   for (e = node->callees; e; e = e->next_callee)
  65     {
  66       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  67       if (e->frequency > CGRAPH_FREQ_MAX)
  68         e->frequency = CGRAPH_FREQ_MAX;
  69       if (!e->inline_failed)
  70         update_noncloned_frequencies (e->callee, freq_scale);
  71     }
  72   for (e = node->indirect_calls; e; e = e->next_callee)
  73     {
  74       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  75       if (e->frequency > CGRAPH_FREQ_MAX)
  76         e->frequency = CGRAPH_FREQ_MAX;
  77     }
  78 }
  79
  80 /* We removed or are going to remove the last call to NODE.
  81    Return true if we can and want proactively remove the NODE now.
  82    This is important to do, since we want inliner to know when offline
  83    copy of function was removed.  */
  84
  85 static bool
  86 can_remove_node_now_p_1 (struct cgraph_node *node)
  87 {
  88   /* FIXME: When address is taken of DECL_EXTERNAL function we still
  89      can remove its offline copy, but we would need to keep unanalyzed node in
  90      the callgraph so references can point to it.  */
  91   return (!node->symbol.address_taken
  92           && !ipa_ref_has_aliases_p (&node->symbol.ref_list)
  93           && cgraph_can_remove_if_no_direct_calls_p (node)
  94           /* Inlining might enable more devirtualizing, so we want to remove
  95              those only after all devirtualizable virtual calls are processed.
  96              Lacking may edges in callgraph we just preserve them post
  97              inlining.  */
  98           && (!DECL_VIRTUAL_P (node->symbol.decl)
  99               || (!DECL_COMDAT (node->symbol.decl)
 100                   && !DECL_EXTERNAL (node->symbol.decl)))
 101           /* During early inlining some unanalyzed cgraph nodes might be in the
 102              callgraph and they might reffer the function in question.  */
 103           && !cgraph_new_nodes);
 104 }
 105
 106 /* We are going to eliminate last direct call to NODE (or alias of it) via edge E.
 107    Verify that the NODE can be removed from unit and if it is contained in comdat
 108    group that the whole comdat group is removable.  */
 109
 110 static bool
 111 can_remove_node_now_p (struct cgraph_node *node, struct cgraph_edge *e)
 112 {
 113   struct cgraph_node *next;
 114   if (!can_remove_node_now_p_1 (node))
 115     return false;
 116
 117   /* When we see same comdat group, we need to be sure that all
 118      items can be removed.  */
 119   if (!node->symbol.same_comdat_group)
 120     return true;
 121   for (next = cgraph (node->symbol.same_comdat_group);
 122        next != node; next = cgraph (next->symbol.same_comdat_group))
 123     if ((next->callers && next->callers != e)
 124         || !can_remove_node_now_p_1 (next))
 125       return false;
 126   return true;
 127 }
 128
 129
 130 /* E is expected to be an edge being inlined.  Clone destination node of
 131    the edge and redirect it to the new clone.
 132    DUPLICATE is used for bookkeeping on whether we are actually creating new
 133    clones or re-using node originally representing out-of-line function call.
 134    */
 135
 136 void
 137 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 138                      bool update_original, int *overall_size)
 139 {
 140   if (duplicate)
 141     {
 142       /* We may eliminate the need for out-of-line copy to be output.
 143          In that case just go ahead and re-use it.  This is not just an
 144          memory optimization.  Making offline copy of fuction disappear
 145          from the program will improve future decisions on inlining.  */
 146       if (!e->callee->callers->next_caller
 147           /* Recursive inlining never wants the master clone to
 148              be overwritten.  */
 149           && update_original
 150           && can_remove_node_now_p (e->callee, e))
 151         {
 152           /* TODO: When callee is in a comdat group, we could remove all of it,
 153              including all inline clones inlined into it.  That would however
 154              need small function inlining to register edge removal hook to
 155              maintain the priority queue.
 156
 157              For now we keep the ohter functions in the group in program until
 158              cgraph_remove_unreachable_functions gets rid of them.  */
 159           gcc_assert (!e->callee->global.inlined_to);
 160           symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 161           if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->symbol.decl))
 162             {
 163               if (overall_size)
 164                 *overall_size -= inline_summary (e->callee)->size;
 165               nfunctions_inlined++;
 166             }
 167           duplicate = false;
 168           e->callee->symbol.externally_visible = false;
 169           update_noncloned_frequencies (e->callee, e->frequency);
 170         }
 171       else
 172         {
 173           struct cgraph_node *n;
 174           n = cgraph_clone_node (e->callee, e->callee->symbol.decl,
 175                                  e->count, e->frequency,
 176                                  update_original, NULL, true);
 177           cgraph_redirect_edge_callee (e, n);
 178         }
 179     }
 180   else
 181     symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 182
 183   if (e->caller->global.inlined_to)
 184     e->callee->global.inlined_to = e->caller->global.inlined_to;
 185   else
 186     e->callee->global.inlined_to = e->caller;
 187
 188   /* Recursively clone all bodies.  */
 189   for (e = e->callee->callees; e; e = e->next_callee)
 190     if (!e->inline_failed)
 191       clone_inlined_nodes (e, duplicate, update_original, overall_size);
 192 }
 193
 194
 195 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 196    specify whether profile of original function should be updated.  If any new
 197    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 198    it is NULL.  Return true iff any new callgraph edges were discovered as a
 199    result of inlining.  */
 200
 201 bool
 202 inline_call (struct cgraph_edge *e, bool update_original,
 203              VEC (cgraph_edge_p, heap) **new_edges,
 204              int *overall_size)
 205 {
 206   int old_size = 0, new_size = 0;
 207   struct cgraph_node *to = NULL;
 208   struct cgraph_edge *curr = e;
 209   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 210
 211   /* Don't inline inlined edges.  */
 212   gcc_assert (e->inline_failed);
 213   /* Don't even think of inlining inline clone.  */
 214   gcc_assert (!callee->global.inlined_to);
 215
 216   e->inline_failed = CIF_OK;
 217   DECL_POSSIBLY_INLINED (callee->symbol.decl) = true;
 218
 219   to = e->caller;
 220   if (to->global.inlined_to)
 221     to = to->global.inlined_to;
 222
 223   /* If aliases are involved, redirect edge to the actual destination and
 224      possibly remove the aliases.  */
 225   if (e->callee != callee)
 226     {
 227       struct cgraph_node *alias = e->callee, *next_alias;
 228       cgraph_redirect_edge_callee (e, callee);
 229       while (alias && alias != callee)
 230         {
 231           if (!alias->callers
 232               && can_remove_node_now_p (alias, e))
 233             {
 234               next_alias = cgraph_alias_aliased_node (alias);
 235               cgraph_remove_node (alias);
 236               alias = next_alias;
 237             }
 238           else
 239             break;
 240         }
 241     }
 242
 243   clone_inlined_nodes (e, true, update_original, overall_size);
 244
 245   gcc_assert (curr->callee->global.inlined_to == to);
 246
 247   old_size = inline_summary (to)->size;
 248   inline_merge_summary (e);
 249   new_size = inline_summary (to)->size;
 250   if (overall_size)
 251     *overall_size += new_size - old_size;
 252   ncalls_inlined++;
 253
 254   /* This must happen after inline_merge_summary that rely on jump
 255      functions of callee to not be updated.  */
 256   if (optimize)
 257     return ipa_propagate_indirect_call_infos (curr, new_edges);
 258   else
 259     return false;
 260 }
 261
 262
 263 /* Copy function body of NODE and redirect all inline clones to it.
 264    This is done before inline plan is applied to NODE when there are
 265    still some inline clones if it.
 266
 267    This is neccesary because inline decisions are not really transitive
 268    and the other inline clones may have different bodies.  */
 269
 270 static struct cgraph_node *
 271 save_inline_function_body (struct cgraph_node *node)
 272 {
 273   struct cgraph_node *first_clone, *n;
 274
 275   if (dump_file)
 276     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 277              cgraph_node_name (node));
 278
 279   gcc_assert (node == cgraph_get_node (node->symbol.decl));
 280
 281   /* first_clone will be turned into real function.  */
 282   first_clone = node->clones;
 283   first_clone->symbol.decl = copy_node (node->symbol.decl);
 284   symtab_insert_node_to_hashtable ((symtab_node) first_clone);
 285   gcc_assert (first_clone == cgraph_get_node (first_clone->symbol.decl));
 286
 287   /* Now reshape the clone tree, so all other clones descends from
 288      first_clone.  */
 289   if (first_clone->next_sibling_clone)
 290     {
 291       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 292         n->clone_of = first_clone;
 293       n->clone_of = first_clone;
 294       n->next_sibling_clone = first_clone->clones;
 295       if (first_clone->clones)
 296         first_clone->clones->prev_sibling_clone = n;
 297       first_clone->clones = first_clone->next_sibling_clone;
 298       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 299       first_clone->next_sibling_clone = NULL;
 300       gcc_assert (!first_clone->prev_sibling_clone);
 301     }
 302   first_clone->clone_of = NULL;
 303
 304   /* Now node in question has no clones.  */
 305   node->clones = NULL;
 306
 307   /* Inline clones share decl with the function they are cloned
 308      from.  Walk the whole clone tree and redirect them all to the
 309      new decl.  */
 310   if (first_clone->clones)
 311     for (n = first_clone->clones; n != first_clone;)
 312       {
 313         gcc_assert (n->symbol.decl == node->symbol.decl);
 314         n->symbol.decl = first_clone->symbol.decl;
 315         if (n->clones)
 316           n = n->clones;
 317         else if (n->next_sibling_clone)
 318           n = n->next_sibling_clone;
 319         else
 320           {
 321             while (n != first_clone && !n->next_sibling_clone)
 322               n = n->clone_of;
 323             if (n != first_clone)
 324               n = n->next_sibling_clone;
 325           }
 326       }
 327
 328   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 329   tree_function_versioning (node->symbol.decl, first_clone->symbol.decl,
 330                             NULL, true, NULL, false, NULL, NULL);
 331
 332   /* The function will be short lived and removed after we inline all the clones,
 333      but make it internal so we won't confuse ourself.  */
 334   DECL_EXTERNAL (first_clone->symbol.decl) = 0;
 335   DECL_COMDAT_GROUP (first_clone->symbol.decl) = NULL_TREE;
 336   TREE_PUBLIC (first_clone->symbol.decl) = 0;
 337   DECL_COMDAT (first_clone->symbol.decl) = 0;
 338   VEC_free (ipa_opt_pass, heap,
 339             first_clone->ipa_transforms_to_apply);
 340   first_clone->ipa_transforms_to_apply = NULL;
 341
 342   /* When doing recursive inlining, the clone may become unnecessary.
 343      This is possible i.e. in the case when the recursive function is proved to be
 344      non-throwing and the recursion happens only in the EH landing pad.
 345      We can not remove the clone until we are done with saving the body.
 346      Remove it now.  */
 347   if (!first_clone->callers)
 348     {
 349       cgraph_remove_node_and_inline_clones (first_clone, NULL);
 350       first_clone = NULL;
 351     }
 352 #ifdef ENABLE_CHECKING
 353   else
 354     verify_cgraph_node (first_clone);
 355 #endif
 356   return first_clone;
 357 }
 358
 359 /* Return true when function body of DECL still needs to be kept around
 360    for later re-use.  */
 361 static bool
 362 preserve_function_body_p (struct cgraph_node *node)
 363 {
 364   gcc_assert (cgraph_global_info_ready);
 365   gcc_assert (!node->alias && !node->thunk.thunk_p);
 366
 367   /* Look if there is any clone around.  */
 368   if (node->clones)
 369     return true;
 370   return false;
 371 }
 372
 373 /* Apply inline plan to function.  */
 374
 375 unsigned int
 376 inline_transform (struct cgraph_node *node)
 377 {
 378   unsigned int todo = 0;
 379   struct cgraph_edge *e;
 380
 381   /* FIXME: Currently the pass manager is adding inline transform more than
 382      once to some clones.  This needs revisiting after WPA cleanups.  */
 383   if (cfun->after_inlining)
 384     return 0;
 385
 386   /* We might need the body of this function so that we can expand
 387      it inline somewhere else.  */
 388   if (preserve_function_body_p (node))
 389     save_inline_function_body (node);
 390
 391   for (e = node->callees; e; e = e->next_callee)
 392     cgraph_redirect_edge_call_stmt_to_callee (e);
 393
 394   timevar_push (TV_INTEGRATION);
 395   if (node->callees)
 396     todo = optimize_inline_calls (current_function_decl);
 397   timevar_pop (TV_INTEGRATION);
 398
 399   cfun->always_inline_functions_inlined = true;
 400   cfun->after_inlining = true;
 401   todo |= execute_fixup_cfg ();
 402
 403   if (!(todo & TODO_update_ssa_any))
 404     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 405     todo |= TODO_update_ssa_only_virtuals;
 406
 407   return todo;
 408 }