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 "intl.h"
  40 #include "coverage.h"
  41 #include "ggc.h"
  42 #include "tree-flow.h"
  43 #include "ipa-prop.h"
  44 #include "ipa-inline.h"
  45 #include "tree-inline.h"
  46 #include "tree-pass.h"
  47
  48 int ncalls_inlined;
  49 int nfunctions_inlined;
  50
  51 /* Scale frequency of NODE edges by FREQ_SCALE.  */
  52
  53 static void
  54 update_noncloned_frequencies (struct cgraph_node *node,
  55                               int freq_scale)
  56 {
  57   struct cgraph_edge *e;
  58
  59   /* We do not want to ignore high loop nest after freq drops to 0.  */
  60   if (!freq_scale)
  61     freq_scale = 1;
  62   for (e = node->callees; e; e = e->next_callee)
  63     {
  64       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  65       if (e->frequency > CGRAPH_FREQ_MAX)
  66         e->frequency = CGRAPH_FREQ_MAX;
  67       if (!e->inline_failed)
  68         update_noncloned_frequencies (e->callee, freq_scale);
  69     }
  70   for (e = node->indirect_calls; e; e = e->next_callee)
  71     {
  72       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  73       if (e->frequency > CGRAPH_FREQ_MAX)
  74         e->frequency = CGRAPH_FREQ_MAX;
  75     }
  76 }
  77
  78 /* We removed or are going to remove the last call to NODE.
  79    Return true if we can and want proactively remove the NODE now.
  80    This is important to do, since we want inliner to know when offline
  81    copy of function was removed.  */
  82
  83 static bool
  84 can_remove_node_now_p_1 (struct cgraph_node *node)
  85 {
  86   /* FIXME: When address is taken of DECL_EXTERNAL function we still
  87      can remove its offline copy, but we would need to keep unanalyzed node in
  88      the callgraph so references can point to it.  */
  89   return (!node->symbol.address_taken
  90           && !ipa_ref_has_aliases_p (&node->symbol.ref_list)
  91           && cgraph_can_remove_if_no_direct_calls_p (node)
  92           /* Inlining might enable more devirtualizing, so we want to remove
  93              those only after all devirtualizable virtual calls are processed.
  94              Lacking may edges in callgraph we just preserve them post
  95              inlining.  */
  96           && (!DECL_VIRTUAL_P (node->symbol.decl)
  97               || (!DECL_COMDAT (node->symbol.decl)
  98                   && !DECL_EXTERNAL (node->symbol.decl)))
  99           /* During early inlining some unanalyzed cgraph nodes might be in the
 100              callgraph and they might reffer the function in question.  */
 101           && !cgraph_new_nodes);
 102 }
 103
 104 /* We are going to eliminate last direct call to NODE (or alias of it) via edge E.
 105    Verify that the NODE can be removed from unit and if it is contained in comdat
 106    group that the whole comdat group is removable.  */
 107
 108 static bool
 109 can_remove_node_now_p (struct cgraph_node *node, struct cgraph_edge *e)
 110 {
 111   struct cgraph_node *next;
 112   if (!can_remove_node_now_p_1 (node))
 113     return false;
 114
 115   /* When we see same comdat group, we need to be sure that all
 116      items can be removed.  */
 117   if (!node->symbol.same_comdat_group)
 118     return true;
 119   for (next = cgraph (node->symbol.same_comdat_group);
 120        next != node; next = cgraph (next->symbol.same_comdat_group))
 121     if ((next->callers && next->callers != e)
 122         || !can_remove_node_now_p_1 (next))
 123       return false;
 124   return true;
 125 }
 126
 127
 128 /* E is expected to be an edge being inlined.  Clone destination node of
 129    the edge and redirect it to the new clone.
 130    DUPLICATE is used for bookkeeping on whether we are actually creating new
 131    clones or re-using node originally representing out-of-line function call.
 132    */
 133
 134 void
 135 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 136                      bool update_original, int *overall_size)
 137 {
 138   if (duplicate)
 139     {
 140       /* We may eliminate the need for out-of-line copy to be output.
 141          In that case just go ahead and re-use it.  This is not just an
 142          memory optimization.  Making offline copy of fuction disappear
 143          from the program will improve future decisions on inlining.  */
 144       if (!e->callee->callers->next_caller
 145           /* Recursive inlining never wants the master clone to
 146              be overwritten.  */
 147           && update_original
 148           && can_remove_node_now_p (e->callee, e))
 149         {
 150           /* TODO: When callee is in a comdat group, we could remove all of it,
 151              including all inline clones inlined into it.  That would however
 152              need small function inlining to register edge removal hook to
 153              maintain the priority queue.
 154
 155              For now we keep the ohter functions in the group in program until
 156              cgraph_remove_unreachable_functions gets rid of them.  */
 157           gcc_assert (!e->callee->global.inlined_to);
 158           symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 159           if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->symbol.decl))
 160             {
 161               if (overall_size)
 162                 *overall_size -= inline_summary (e->callee)->size;
 163               nfunctions_inlined++;
 164             }
 165           duplicate = false;
 166           e->callee->symbol.externally_visible = false;
 167           update_noncloned_frequencies (e->callee, e->frequency);
 168         }
 169       else
 170         {
 171           struct cgraph_node *n;
 172           n = cgraph_clone_node (e->callee, e->callee->symbol.decl,
 173                                  e->count, e->frequency,
 174                                  update_original, NULL, true);
 175           cgraph_redirect_edge_callee (e, n);
 176         }
 177     }
 178   else
 179     symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 180
 181   if (e->caller->global.inlined_to)
 182     e->callee->global.inlined_to = e->caller->global.inlined_to;
 183   else
 184     e->callee->global.inlined_to = e->caller;
 185
 186   /* Recursively clone all bodies.  */
 187   for (e = e->callee->callees; e; e = e->next_callee)
 188     if (!e->inline_failed)
 189       clone_inlined_nodes (e, duplicate, update_original, overall_size);
 190 }
 191
 192
 193 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 194    specify whether profile of original function should be updated.  If any new
 195    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 196    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 197    size of caller after inlining. Caller is required to eventually do it via
 198    inline_update_overall_summary.
 199
 200    Return true iff any new callgraph edges were discovered as a
 201    result of inlining.  */
 202
 203 bool
 204 inline_call (struct cgraph_edge *e, bool update_original,
 205              VEC (cgraph_edge_p, heap) **new_edges,
 206              int *overall_size, bool update_overall_summary)
 207 {
 208   int old_size = 0, new_size = 0;
 209   struct cgraph_node *to = NULL;
 210   struct cgraph_edge *curr = e;
 211   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 212
 213   /* Don't inline inlined edges.  */
 214   gcc_assert (e->inline_failed);
 215   /* Don't even think of inlining inline clone.  */
 216   gcc_assert (!callee->global.inlined_to);
 217
 218   e->inline_failed = CIF_OK;
 219   DECL_POSSIBLY_INLINED (callee->symbol.decl) = true;
 220
 221   to = e->caller;
 222   if (to->global.inlined_to)
 223     to = to->global.inlined_to;
 224
 225   /* If aliases are involved, redirect edge to the actual destination and
 226      possibly remove the aliases.  */
 227   if (e->callee != callee)
 228     {
 229       struct cgraph_node *alias = e->callee, *next_alias;
 230       cgraph_redirect_edge_callee (e, callee);
 231       while (alias && alias != callee)
 232         {
 233           if (!alias->callers
 234               && can_remove_node_now_p (alias, e))
 235             {
 236               next_alias = cgraph_alias_aliased_node (alias);
 237               cgraph_remove_node (alias);
 238               alias = next_alias;
 239             }
 240           else
 241             break;
 242         }
 243     }
 244
 245   clone_inlined_nodes (e, true, update_original, overall_size);
 246
 247   gcc_assert (curr->callee->global.inlined_to == to);
 248
 249   old_size = inline_summary (to)->size;
 250   inline_merge_summary (e);
 251   if (update_overall_summary)
 252    inline_update_overall_summary (to);
 253   new_size = inline_summary (to)->size;
 254   if (overall_size)
 255     *overall_size += new_size - old_size;
 256   ncalls_inlined++;
 257
 258   /* This must happen after inline_merge_summary that rely on jump
 259      functions of callee to not be updated.  */
 260   if (optimize)
 261     return ipa_propagate_indirect_call_infos (curr, new_edges);
 262   else
 263     return false;
 264 }
 265
 266
 267 /* Copy function body of NODE and redirect all inline clones to it.
 268    This is done before inline plan is applied to NODE when there are
 269    still some inline clones if it.
 270
 271    This is necessary because inline decisions are not really transitive
 272    and the other inline clones may have different bodies.  */
 273
 274 static struct cgraph_node *
 275 save_inline_function_body (struct cgraph_node *node)
 276 {
 277   struct cgraph_node *first_clone, *n;
 278
 279   if (dump_file)
 280     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 281              cgraph_node_name (node));
 282
 283   gcc_assert (node == cgraph_get_node (node->symbol.decl));
 284
 285   /* first_clone will be turned into real function.  */
 286   first_clone = node->clones;
 287   first_clone->symbol.decl = copy_node (node->symbol.decl);
 288   symtab_insert_node_to_hashtable ((symtab_node) first_clone);
 289   gcc_assert (first_clone == cgraph_get_node (first_clone->symbol.decl));
 290
 291   /* Now reshape the clone tree, so all other clones descends from
 292      first_clone.  */
 293   if (first_clone->next_sibling_clone)
 294     {
 295       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 296         n->clone_of = first_clone;
 297       n->clone_of = first_clone;
 298       n->next_sibling_clone = first_clone->clones;
 299       if (first_clone->clones)
 300         first_clone->clones->prev_sibling_clone = n;
 301       first_clone->clones = first_clone->next_sibling_clone;
 302       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 303       first_clone->next_sibling_clone = NULL;
 304       gcc_assert (!first_clone->prev_sibling_clone);
 305     }
 306   first_clone->clone_of = NULL;
 307
 308   /* Now node in question has no clones.  */
 309   node->clones = NULL;
 310
 311   /* Inline clones share decl with the function they are cloned
 312      from.  Walk the whole clone tree and redirect them all to the
 313      new decl.  */
 314   if (first_clone->clones)
 315     for (n = first_clone->clones; n != first_clone;)
 316       {
 317         gcc_assert (n->symbol.decl == node->symbol.decl);
 318         n->symbol.decl = first_clone->symbol.decl;
 319         if (n->clones)
 320           n = n->clones;
 321         else if (n->next_sibling_clone)
 322           n = n->next_sibling_clone;
 323         else
 324           {
 325             while (n != first_clone && !n->next_sibling_clone)
 326               n = n->clone_of;
 327             if (n != first_clone)
 328               n = n->next_sibling_clone;
 329           }
 330       }
 331
 332   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 333   tree_function_versioning (node->symbol.decl, first_clone->symbol.decl,
 334                             NULL, true, NULL, false, NULL, NULL);
 335
 336   /* The function will be short lived and removed after we inline all the clones,
 337      but make it internal so we won't confuse ourself.  */
 338   DECL_EXTERNAL (first_clone->symbol.decl) = 0;
 339   DECL_COMDAT_GROUP (first_clone->symbol.decl) = NULL_TREE;
 340   TREE_PUBLIC (first_clone->symbol.decl) = 0;
 341   DECL_COMDAT (first_clone->symbol.decl) = 0;
 342   VEC_free (ipa_opt_pass, heap,
 343             first_clone->ipa_transforms_to_apply);
 344   first_clone->ipa_transforms_to_apply = NULL;
 345
 346   /* When doing recursive inlining, the clone may become unnecessary.
 347      This is possible i.e. in the case when the recursive function is proved to be
 348      non-throwing and the recursion happens only in the EH landing pad.
 349      We can not remove the clone until we are done with saving the body.
 350      Remove it now.  */
 351   if (!first_clone->callers)
 352     {
 353       cgraph_remove_node_and_inline_clones (first_clone, NULL);
 354       first_clone = NULL;
 355     }
 356 #ifdef ENABLE_CHECKING
 357   else
 358     verify_cgraph_node (first_clone);
 359 #endif
 360   return first_clone;
 361 }
 362
 363 /* Return true when function body of DECL still needs to be kept around
 364    for later re-use.  */
 365 static bool
 366 preserve_function_body_p (struct cgraph_node *node)
 367 {
 368   gcc_assert (cgraph_global_info_ready);
 369   gcc_assert (!node->alias && !node->thunk.thunk_p);
 370
 371   /* Look if there is any clone around.  */
 372   if (node->clones)
 373     return true;
 374   return false;
 375 }
 376
 377 /* Apply inline plan to function.  */
 378
 379 unsigned int
 380 inline_transform (struct cgraph_node *node)
 381 {
 382   unsigned int todo = 0;
 383   struct cgraph_edge *e;
 384
 385   /* FIXME: Currently the pass manager is adding inline transform more than
 386      once to some clones.  This needs revisiting after WPA cleanups.  */
 387   if (cfun->after_inlining)
 388     return 0;
 389
 390   /* We might need the body of this function so that we can expand
 391      it inline somewhere else.  */
 392   if (preserve_function_body_p (node))
 393     save_inline_function_body (node);
 394
 395   for (e = node->callees; e; e = e->next_callee)
 396     cgraph_redirect_edge_call_stmt_to_callee (e);
 397
 398   timevar_push (TV_INTEGRATION);
 399   if (node->callees)
 400     todo = optimize_inline_calls (current_function_decl);
 401   timevar_pop (TV_INTEGRATION);
 402
 403   cfun->always_inline_functions_inlined = true;
 404   cfun->after_inlining = true;
 405   todo |= execute_fixup_cfg ();
 406
 407   if (!(todo & TODO_update_ssa_any))
 408     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 409     todo |= TODO_update_ssa_only_virtuals;
 410
 411   return todo;
 412 }