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