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->address_taken
  92           && !ipa_ref_has_aliases_p (&node->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->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->same_comdat_group)
 118     return true;
 119   for (next = node->same_comdat_group;
 120        next != node; next = next->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           if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->decl))
 159             {
 160               if (overall_size)
 161                 *overall_size -= inline_summary (e->callee)->size;
 162               nfunctions_inlined++;
 163             }
 164           duplicate = false;
 165           e->callee->local.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->decl,
 172                                  e->count, e->frequency,
 173                                  update_original, NULL, true);
 174           cgraph_redirect_edge_callee (e, n);
 175         }
 176     }
 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.  Return true iff any new callgraph edges were discovered as a
 194    result of inlining.  */
 195
 196 bool
 197 inline_call (struct cgraph_edge *e, bool update_original,
 198              VEC (cgraph_edge_p, heap) **new_edges,
 199              int *overall_size)
 200 {
 201   int old_size = 0, new_size = 0;
 202   struct cgraph_node *to = NULL;
 203   struct cgraph_edge *curr = e;
 204   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 205
 206   /* Don't inline inlined edges.  */
 207   gcc_assert (e->inline_failed);
 208   /* Don't even think of inlining inline clone.  */
 209   gcc_assert (!callee->global.inlined_to);
 210
 211   e->inline_failed = CIF_OK;
 212   DECL_POSSIBLY_INLINED (callee->decl) = true;
 213
 214   to = e->caller;
 215   if (to->global.inlined_to)
 216     to = to->global.inlined_to;
 217
 218   /* If aliases are involved, redirect edge to the actual destination and
 219      possibly remove the aliases.  */
 220   if (e->callee != callee)
 221     {
 222       struct cgraph_node *alias = e->callee, *next_alias;
 223       cgraph_redirect_edge_callee (e, callee);
 224       while (alias && alias != callee)
 225         {
 226           if (!alias->callers
 227               && can_remove_node_now_p (alias, e))
 228             {
 229               next_alias = cgraph_alias_aliased_node (alias);
 230               cgraph_remove_node (alias);
 231               alias = next_alias;
 232             }
 233           else
 234             break;
 235         }
 236     }
 237
 238   clone_inlined_nodes (e, true, update_original, overall_size);
 239
 240   gcc_assert (curr->callee->global.inlined_to == to);
 241
 242   old_size = inline_summary (to)->size;
 243   inline_merge_summary (e);
 244   new_size = inline_summary (to)->size;
 245   if (overall_size)
 246     *overall_size += new_size - old_size;
 247   ncalls_inlined++;
 248
 249   /* This must happen after inline_merge_summary that rely on jump
 250      functions of callee to not be updated.  */
 251   if (optimize)
 252     return ipa_propagate_indirect_call_infos (curr, new_edges);
 253   else
 254     return false;
 255 }
 256
 257
 258 /* Copy function body of NODE and redirect all inline clones to it.
 259    This is done before inline plan is applied to NODE when there are
 260    still some inline clones if it.
 261
 262    This is neccesary because inline decisions are not really transitive
 263    and the other inline clones may have different bodies.  */
 264
 265 static struct cgraph_node *
 266 save_inline_function_body (struct cgraph_node *node)
 267 {
 268   struct cgraph_node *first_clone, *n;
 269
 270   if (dump_file)
 271     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 272              cgraph_node_name (node));
 273
 274   gcc_assert (node == cgraph_get_node (node->decl));
 275
 276   /* first_clone will be turned into real function.  */
 277   first_clone = node->clones;
 278   first_clone->decl = copy_node (node->decl);
 279   cgraph_insert_node_to_hashtable (first_clone);
 280   gcc_assert (first_clone == cgraph_get_node (first_clone->decl));
 281
 282   /* Now reshape the clone tree, so all other clones descends from
 283      first_clone.  */
 284   if (first_clone->next_sibling_clone)
 285     {
 286       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 287         n->clone_of = first_clone;
 288       n->clone_of = first_clone;
 289       n->next_sibling_clone = first_clone->clones;
 290       if (first_clone->clones)
 291         first_clone->clones->prev_sibling_clone = n;
 292       first_clone->clones = first_clone->next_sibling_clone;
 293       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 294       first_clone->next_sibling_clone = NULL;
 295       gcc_assert (!first_clone->prev_sibling_clone);
 296     }
 297   first_clone->clone_of = NULL;
 298
 299   /* Now node in question has no clones.  */
 300   node->clones = NULL;
 301
 302   /* Inline clones share decl with the function they are cloned
 303      from.  Walk the whole clone tree and redirect them all to the
 304      new decl.  */
 305   if (first_clone->clones)
 306     for (n = first_clone->clones; n != first_clone;)
 307       {
 308         gcc_assert (n->decl == node->decl);
 309         n->decl = first_clone->decl;
 310         if (n->clones)
 311           n = n->clones;
 312         else if (n->next_sibling_clone)
 313           n = n->next_sibling_clone;
 314         else
 315           {
 316             while (n != first_clone && !n->next_sibling_clone)
 317               n = n->clone_of;
 318             if (n != first_clone)
 319               n = n->next_sibling_clone;
 320           }
 321       }
 322
 323   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 324   tree_function_versioning (node->decl, first_clone->decl, NULL, true, NULL,
 325                             false, NULL, NULL);
 326
 327   /* The function will be short lived and removed after we inline all the clones,
 328      but make it internal so we won't confuse ourself.  */
 329   DECL_EXTERNAL (first_clone->decl) = 0;
 330   DECL_COMDAT_GROUP (first_clone->decl) = NULL_TREE;
 331   TREE_PUBLIC (first_clone->decl) = 0;
 332   DECL_COMDAT (first_clone->decl) = 0;
 333   VEC_free (ipa_opt_pass, heap,
 334             first_clone->ipa_transforms_to_apply);
 335   first_clone->ipa_transforms_to_apply = NULL;
 336
 337 #ifdef ENABLE_CHECKING
 338   verify_cgraph_node (first_clone);
 339 #endif
 340   return first_clone;
 341 }
 342
 343
 344 /* Apply inline plan to function.  */
 345
 346 unsigned int
 347 inline_transform (struct cgraph_node *node)
 348 {
 349   unsigned int todo = 0;
 350   struct cgraph_edge *e;
 351
 352   /* FIXME: Currently the pass manager is adding inline transform more than
 353      once to some clones.  This needs revisiting after WPA cleanups.  */
 354   if (cfun->after_inlining)
 355     return 0;
 356
 357   /* We might need the body of this function so that we can expand
 358      it inline somewhere else.  */
 359   if (cgraph_preserve_function_body_p (node))
 360     save_inline_function_body (node);
 361
 362   for (e = node->callees; e; e = e->next_callee)
 363     cgraph_redirect_edge_call_stmt_to_callee (e);
 364
 365   timevar_push (TV_INTEGRATION);
 366   if (node->callees)
 367     todo = optimize_inline_calls (current_function_decl);
 368   timevar_pop (TV_INTEGRATION);
 369
 370   cfun->always_inline_functions_inlined = true;
 371   cfun->after_inlining = true;
 372   todo |= execute_fixup_cfg ();
 373
 374   if (!(todo & TODO_update_ssa_any))
 375     /* Redirecting edges might lead to a need for vops to be recomputed.  */
 376     todo |= TODO_update_ssa_only_virtuals;
 377
 378   return todo;
 379 }