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               || (!DECL_COMDAT (node->symbol.decl)
  97                   && !DECL_EXTERNAL (node->symbol.decl)))
  98           /* During early inlining some unanalyzed cgraph nodes might be in the
  99              callgraph and they might reffer the function in question.  */
 100           && !cgraph_new_nodes);
 101 }
 102
 103 /* We are going to eliminate last direct call to NODE (or alias of it) via edge E.
 104    Verify that the NODE can be removed from unit and if it is contained in comdat
 105    group that the whole comdat group is removable.  */
 106
 107 static bool
 108 can_remove_node_now_p (struct cgraph_node *node, struct cgraph_edge *e)
 109 {
 110   struct cgraph_node *next;
 111   if (!can_remove_node_now_p_1 (node))
 112     return false;
 113
 114   /* When we see same comdat group, we need to be sure that all
 115      items can be removed.  */
 116   if (!node->symbol.same_comdat_group)
 117     return true;
 118   for (next = cgraph (node->symbol.same_comdat_group);
 119        next != node; next = cgraph (next->symbol.same_comdat_group))
 120     if ((next->callers && next->callers != e)
 121         || !can_remove_node_now_p_1 (next))
 122       return false;
 123   return true;
 124 }
 125
 126
 127 /* E is expected to be an edge being inlined.  Clone destination node of
 128    the edge and redirect it to the new clone.
 129    DUPLICATE is used for bookkeeping on whether we are actually creating new
 130    clones or re-using node originally representing out-of-line function call.
 131    */
 132
 133 void
 134 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 135                      bool update_original, int *overall_size)
 136 {
 137   if (duplicate)
 138     {
 139       /* We may eliminate the need for out-of-line copy to be output.
 140          In that case just go ahead and re-use it.  This is not just an
 141          memory optimization.  Making offline copy of fuction disappear
 142          from the program will improve future decisions on inlining.  */
 143       if (!e->callee->callers->next_caller
 144           /* Recursive inlining never wants the master clone to
 145              be overwritten.  */
 146           && update_original
 147           && can_remove_node_now_p (e->callee, e))
 148         {
 149           /* TODO: When callee is in a comdat group, we could remove all of it,
 150              including all inline clones inlined into it.  That would however
 151              need small function inlining to register edge removal hook to
 152              maintain the priority queue.
 153
 154              For now we keep the ohter functions in the group in program until
 155              cgraph_remove_unreachable_functions gets rid of them.  */
 156           gcc_assert (!e->callee->global.inlined_to);
 157           symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 158           if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->symbol.decl))
 159             {
 160               if (overall_size)
 161                 *overall_size -= inline_summary (e->callee)->size;
 162               nfunctions_inlined++;
 163             }
 164           duplicate = false;
 165           e->callee->symbol.externally_visible = false;
 166           update_noncloned_frequencies (e->callee, e->frequency);
 167         }
 168       else
 169         {
 170           struct cgraph_node *n;
 171           n = cgraph_clone_node (e->callee, e->callee->symbol.decl,
 172                                  e->count, e->frequency,
 173                                  update_original, vNULL, true);
 174           cgraph_redirect_edge_callee (e, n);
 175         }
 176     }
 177   else
 178     symtab_dissolve_same_comdat_group_list ((symtab_node) e->callee);
 179
 180   if (e->caller->global.inlined_to)
 181     e->callee->global.inlined_to = e->caller->global.inlined_to;
 182   else
 183     e->callee->global.inlined_to = e->caller;
 184
 185   /* Recursively clone all bodies.  */
 186   for (e = e->callee->callees; e; e = e->next_callee)
 187     if (!e->inline_failed)
 188       clone_inlined_nodes (e, duplicate, update_original, overall_size);
 189 }
 190
 191
 192 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 193    specify whether profile of original function should be updated.  If any new
 194    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 195    it is NULL. If UPDATE_OVERALL_SUMMARY is false, do not bother to recompute overall
 196    size of caller after inlining. Caller is required to eventually do it via
 197    inline_update_overall_summary.
 198
 199    Return true iff any new callgraph edges were discovered as a
 200    result of inlining.  */
 201
 202 bool
 203 inline_call (struct cgraph_edge *e, bool update_original,
 204              vec<cgraph_edge_p> *new_edges,
 205              int *overall_size, bool update_overall_summary)
 206 {
 207   int old_size = 0, new_size = 0;
 208   struct cgraph_node *to = NULL;
 209   struct cgraph_edge *curr = e;
 210   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 211   bool new_edges_found = false;
 212
 213 #ifdef ENABLE_CHECKING
 214   int estimated_growth = estimate_edge_growth (e);
 215   bool predicated = inline_edge_summary (e)->predicate != NULL;
 216 #endif
 217
 218   /* Don't inline inlined edges.  */
 219   gcc_assert (e->inline_failed);
 220   /* Don't even think of inlining inline clone.  */
 221   gcc_assert (!callee->global.inlined_to);
 222
 223   e->inline_failed = CIF_OK;
 224   DECL_POSSIBLY_INLINED (callee->symbol.decl) = true;
 225
 226   to = e->caller;
 227   if (to->global.inlined_to)
 228     to = to->global.inlined_to;
 229
 230   /* If aliases are involved, redirect edge to the actual destination and
 231      possibly remove the aliases.  */
 232   if (e->callee != callee)
 233     {
 234       struct cgraph_node *alias = e->callee, *next_alias;
 235       cgraph_redirect_edge_callee (e, callee);
 236       while (alias && alias != callee)
 237         {
 238           if (!alias->callers
 239               && can_remove_node_now_p (alias, e))
 240             {
 241               next_alias = cgraph_alias_aliased_node (alias);
 242               cgraph_remove_node (alias);
 243               alias = next_alias;
 244             }
 245           else
 246             break;
 247         }
 248     }
 249
 250   clone_inlined_nodes (e, true, update_original, overall_size);
 251
 252   gcc_assert (curr->callee->global.inlined_to == to);
 253
 254   old_size = inline_summary (to)->size;
 255   inline_merge_summary (e);
 256   if (optimize)
 257     new_edges_found = ipa_propagate_indirect_call_infos (curr, new_edges);
 258   if (update_overall_summary)
 259    inline_update_overall_summary (to);
 260   new_size = inline_summary (to)->size;
 261
 262 #ifdef ENABLE_CHECKING
 263   /* Verify that estimated growth match real growth.  Allow off-by-one
 264      error due to INLINE_SIZE_SCALE roudoff errors.  */
 265   gcc_assert (!update_overall_summary || !overall_size
 266               || abs (estimated_growth - (new_size - old_size)) <= 1
 267               /* FIXME: a hack.  Edges with false predicate are accounted
 268                  wrong, we should remove them from callgraph.  */
 269               || predicated);
 270 #endif
 271
 272   /* Account the change of overall unit size; external functions will be
 273      removed and are thus not accounted.  */
 274   if (overall_size
 275       && !DECL_EXTERNAL (to->symbol.decl))
 276     *overall_size += new_size - old_size;
 277   ncalls_inlined++;
 278
 279   /* This must happen after inline_merge_summary that rely on jump
 280      functions of callee to not be updated.  */
 281   return new_edges_found;
 282 }
 283
 284
 285 /* Copy function body of NODE and redirect all inline clones to it.
 286    This is done before inline plan is applied to NODE when there are
 287    still some inline clones if it.
 288
 289    This is necessary because inline decisions are not really transitive
 290    and the other inline clones may have different bodies.  */
 291
 292 static struct cgraph_node *
 293 save_inline_function_body (struct cgraph_node *node)
 294 {
 295   struct cgraph_node *first_clone, *n;
 296
 297   if (dump_file)
 298     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 299              cgraph_node_name (node));
 300
 301   gcc_assert (node == cgraph_get_node (node->symbol.decl));
 302
 303   /* first_clone will be turned into real function.  */
 304   first_clone = node->clones;
 305   first_clone->symbol.decl = copy_node (node->symbol.decl);
 306   symtab_insert_node_to_hashtable ((symtab_node) first_clone);
 307   gcc_assert (first_clone == cgraph_get_node (first_clone->symbol.decl));
 308
 309   /* Now reshape the clone tree, so all other clones descends from
 310      first_clone.  */
 311   if (first_clone->next_sibling_clone)
 312     {
 313       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 314         n->clone_of = first_clone;
 315       n->clone_of = first_clone;
 316       n->next_sibling_clone = first_clone->clones;
 317       if (first_clone->clones)
 318         first_clone->clones->prev_sibling_clone = n;
 319       first_clone->clones = first_clone->next_sibling_clone;
 320       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 321       first_clone->next_sibling_clone = NULL;
 322       gcc_assert (!first_clone->prev_sibling_clone);
 323     }
 324   first_clone->clone_of = NULL;
 325
 326   /* Now node in question has no clones.  */
 327   node->clones = NULL;
 328
 329   /* Inline clones share decl with the function they are cloned
 330      from.  Walk the whole clone tree and redirect them all to the
 331      new decl.  */
 332   if (first_clone->clones)
 333     for (n = first_clone->clones; n != first_clone;)
 334       {
 335         gcc_assert (n->symbol.decl == node->symbol.decl);
 336         n->symbol.decl = first_clone->symbol.decl;
 337         if (n->clones)
 338           n = n->clones;
 339         else if (n->next_sibling_clone)
 340           n = n->next_sibling_clone;
 341         else
 342           {
 343             while (n != first_clone && !n->next_sibling_clone)
 344               n = n->clone_of;
 345             if (n != first_clone)
 346               n = n->next_sibling_clone;
 347           }
 348       }
 349
 350   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 351   tree_function_versioning (node->symbol.decl, first_clone->symbol.decl,
 352                             NULL, true, NULL, false,
 353                             NULL, NULL);
 354
 355   /* The function will be short lived and removed after we inline all the clones,
 356      but make it internal so we won't confuse ourself.  */
 357   DECL_EXTERNAL (first_clone->symbol.decl) = 0;
 358   DECL_COMDAT_GROUP (first_clone->symbol.decl) = NULL_TREE;
 359   TREE_PUBLIC (first_clone->symbol.decl) = 0;
 360   DECL_COMDAT (first_clone->symbol.decl) = 0;
 361   first_clone->ipa_transforms_to_apply.release ();
 362
 363   /* When doing recursive inlining, the clone may become unnecessary.
 364      This is possible i.e. in the case when the recursive function is proved to be
 365      non-throwing and the recursion happens only in the EH landing pad.
 366      We can not remove the clone until we are done with saving the body.
 367      Remove it now.  */
 368   if (!first_clone->callers)
 369     {
 370       cgraph_remove_node_and_inline_clones (first_clone, NULL);
 371       first_clone = NULL;
 372     }
 373 #ifdef ENABLE_CHECKING
 374   else
 375     verify_cgraph_node (first_clone);
 376 #endif
 377   return first_clone;
 378 }
 379
 380 /* Return true when function body of DECL still needs to be kept around
 381    for later re-use.  */
 382 static bool
 383 preserve_function_body_p (struct cgraph_node *node)
 384 {
 385   gcc_assert (cgraph_global_info_ready);
 386   gcc_assert (!node->alias && !node->thunk.thunk_p);
 387
 388   /* Look if there is any clone around.  */
 389   if (node->clones)
 390     return true;
 391   return false;
 392 }
 393
 394 /* Apply inline plan to function.  */
 395
 396 unsigned int
 397 inline_transform (struct cgraph_node *node)
 398 {
 399   unsigned int todo = 0;
 400   struct cgraph_edge *e;
 401
 402   /* FIXME: Currently the pass manager is adding inline transform more than
 403      once to some clones.  This needs revisiting after WPA cleanups.  */
 404   if (cfun->after_inlining)
 405     return 0;
 406
 407   /* We might need the body of this function so that we can expand
 408      it inline somewhere else.  */
 409   if (preserve_function_body_p (node))
 410     save_inline_function_body (node);
 411
 412   for (e = node->callees; e; e = e->next_callee)
 413     cgraph_redirect_edge_call_stmt_to_callee (e);
 414
 415   timevar_push (TV_INTEGRATION);
 416   if (node->callees)
 417     todo = optimize_inline_calls (current_function_decl);
 418   timevar_pop (TV_INTEGRATION);
 419
 420   cfun->always_inline_functions_inlined = true;
 421   cfun->after_inlining = true;
 422   todo |= execute_fixup_cfg ();
 423
 424   if (!(todo & TODO_update_ssa_any))
 425     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 426     todo |= TODO_update_ssa_only_virtuals;
 427
 428   return todo;
 429 }