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
  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 (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->address_taken
  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->decl)
  97               || (!DECL_COMDAT (node->decl)
  98                   && !DECL_EXTERNAL (node->decl)))
  99           /* Don't reuse if more than one function shares a comdat group.
 100              If the other function(s) are needed, we need to emit even
 101              this function out of line.  */
 102           && !node->same_comdat_group
 103           /* During early inlining some unanalyzed cgraph nodes might be in the
 104              callgraph and they might reffer the function in question.  */
 105           && !cgraph_new_nodes);
 106 }
 107
 108
 109 /* E is expected to be an edge being inlined.  Clone destination node of
 110    the edge and redirect it to the new clone.
 111    DUPLICATE is used for bookkeeping on whether we are actually creating new
 112    clones or re-using node originally representing out-of-line function call.
 113    */
 114
 115 void
 116 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 117                      bool update_original, int *overall_size)
 118 {
 119   if (duplicate)
 120     {
 121       /* We may eliminate the need for out-of-line copy to be output.
 122          In that case just go ahead and re-use it.  This is not just an
 123          memory optimization.  Making offline copy of fuction disappear
 124          from the program will improve future decisions on inlining.  */
 125       if (!e->callee->callers->next_caller
 126           /* Recursive inlining never wants the master clone to
 127              be overwritten.  */
 128           && update_original
 129           && can_remove_node_now_p (e->callee))
 130         {
 131           gcc_assert (!e->callee->global.inlined_to);
 132           if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->decl))
 133             {
 134               if (overall_size)
 135                 *overall_size -= inline_summary (e->callee)->size;
 136               nfunctions_inlined++;
 137             }
 138           duplicate = false;
 139           e->callee->local.externally_visible = false;
 140           update_noncloned_frequencies (e->callee, e->frequency);
 141         }
 142       else
 143         {
 144           struct cgraph_node *n;
 145           n = cgraph_clone_node (e->callee, e->callee->decl,
 146                                  e->count, e->frequency,
 147                                  update_original, NULL, true);
 148           cgraph_redirect_edge_callee (e, n);
 149         }
 150     }
 151
 152   if (e->caller->global.inlined_to)
 153     e->callee->global.inlined_to = e->caller->global.inlined_to;
 154   else
 155     e->callee->global.inlined_to = e->caller;
 156
 157   /* Recursively clone all bodies.  */
 158   for (e = e->callee->callees; e; e = e->next_callee)
 159     if (!e->inline_failed)
 160       clone_inlined_nodes (e, duplicate, update_original, overall_size);
 161 }
 162
 163
 164 /* Mark edge E as inlined and update callgraph accordingly.  UPDATE_ORIGINAL
 165    specify whether profile of original function should be updated.  If any new
 166    indirect edges are discovered in the process, add them to NEW_EDGES, unless
 167    it is NULL.  Return true iff any new callgraph edges were discovered as a
 168    result of inlining.  */
 169
 170 bool
 171 inline_call (struct cgraph_edge *e, bool update_original,
 172              VEC (cgraph_edge_p, heap) **new_edges,
 173              int *overall_size)
 174 {
 175   int old_size = 0, new_size = 0;
 176   struct cgraph_node *to = NULL;
 177   struct cgraph_edge *curr = e;
 178   struct cgraph_node *callee = cgraph_function_or_thunk_node (e->callee, NULL);
 179
 180   /* Don't inline inlined edges.  */
 181   gcc_assert (e->inline_failed);
 182   /* Don't even think of inlining inline clone.  */
 183   gcc_assert (!callee->global.inlined_to);
 184
 185   e->inline_failed = CIF_OK;
 186   DECL_POSSIBLY_INLINED (callee->decl) = true;
 187
 188   to = e->caller;
 189   if (to->global.inlined_to)
 190     to = to->global.inlined_to;
 191
 192   /* If aliases are involved, redirect edge to the actual destination and
 193      possibly remove the aliases.  */
 194   if (e->callee != callee)
 195     cgraph_redirect_edge_callee (e, callee);
 196
 197   clone_inlined_nodes (e, true, update_original, overall_size);
 198
 199   gcc_assert (curr->callee->global.inlined_to == to);
 200
 201   old_size = inline_summary (to)->size;
 202   inline_merge_summary (e);
 203   new_size = inline_summary (to)->size;
 204   if (overall_size)
 205     *overall_size += new_size - old_size;
 206   ncalls_inlined++;
 207
 208   if (flag_indirect_inlining && optimize)
 209     return ipa_propagate_indirect_call_infos (curr, new_edges);
 210   else
 211     return false;
 212 }
 213
 214
 215 /* Copy function body of NODE and redirect all inline clones to it.
 216    This is done before inline plan is applied to NODE when there are
 217    still some inline clones if it.
 218
 219    This is neccesary because inline decisions are not really transitive
 220    and the other inline clones may have different bodies.  */
 221
 222 static struct cgraph_node *
 223 save_inline_function_body (struct cgraph_node *node)
 224 {
 225   struct cgraph_node *first_clone, *n;
 226
 227   if (dump_file)
 228     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 229              cgraph_node_name (node));
 230
 231   gcc_assert (node == cgraph_get_node (node->decl));
 232
 233   /* first_clone will be turned into real function.  */
 234   first_clone = node->clones;
 235   first_clone->decl = copy_node (node->decl);
 236   cgraph_insert_node_to_hashtable (first_clone);
 237   gcc_assert (first_clone == cgraph_get_node (first_clone->decl));
 238
 239   /* Now reshape the clone tree, so all other clones descends from
 240      first_clone.  */
 241   if (first_clone->next_sibling_clone)
 242     {
 243       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 244         n->clone_of = first_clone;
 245       n->clone_of = first_clone;
 246       n->next_sibling_clone = first_clone->clones;
 247       if (first_clone->clones)
 248         first_clone->clones->prev_sibling_clone = n;
 249       first_clone->clones = first_clone->next_sibling_clone;
 250       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 251       first_clone->next_sibling_clone = NULL;
 252       gcc_assert (!first_clone->prev_sibling_clone);
 253     }
 254   first_clone->clone_of = NULL;
 255
 256   /* Now node in question has no clones.  */
 257   node->clones = NULL;
 258
 259   /* Inline clones share decl with the function they are cloned
 260      from.  Walk the whole clone tree and redirect them all to the
 261      new decl.  */
 262   if (first_clone->clones)
 263     for (n = first_clone->clones; n != first_clone;)
 264       {
 265         gcc_assert (n->decl == node->decl);
 266         n->decl = first_clone->decl;
 267         if (n->clones)
 268           n = n->clones;
 269         else if (n->next_sibling_clone)
 270           n = n->next_sibling_clone;
 271         else
 272           {
 273             while (n != first_clone && !n->next_sibling_clone)
 274               n = n->clone_of;
 275             if (n != first_clone)
 276               n = n->next_sibling_clone;
 277           }
 278       }
 279
 280   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 281   tree_function_versioning (node->decl, first_clone->decl, NULL, true, NULL,
 282                             NULL, NULL);
 283
 284   /* The function will be short lived and removed after we inline all the clones,
 285      but make it internal so we won't confuse ourself.  */
 286   DECL_EXTERNAL (first_clone->decl) = 0;
 287   DECL_COMDAT_GROUP (first_clone->decl) = NULL_TREE;
 288   TREE_PUBLIC (first_clone->decl) = 0;
 289   DECL_COMDAT (first_clone->decl) = 0;
 290   VEC_free (ipa_opt_pass, heap,
 291             first_clone->ipa_transforms_to_apply);
 292   first_clone->ipa_transforms_to_apply = NULL;
 293
 294 #ifdef ENABLE_CHECKING
 295   verify_cgraph_node (first_clone);
 296 #endif
 297   return first_clone;
 298 }
 299
 300
 301 /* Apply inline plan to function.  */
 302
 303 unsigned int
 304 inline_transform (struct cgraph_node *node)
 305 {
 306   unsigned int todo = 0;
 307   struct cgraph_edge *e;
 308   bool inline_p = false;
 309
 310   /* FIXME: Currently the pass manager is adding inline transform more than
 311      once to some clones.  This needs revisiting after WPA cleanups.  */
 312   if (cfun->after_inlining)
 313     return 0;
 314
 315   /* We might need the body of this function so that we can expand
 316      it inline somewhere else.  */
 317   if (cgraph_preserve_function_body_p (node))
 318     save_inline_function_body (node);
 319
 320   for (e = node->callees; e; e = e->next_callee)
 321     {
 322       cgraph_redirect_edge_call_stmt_to_callee (e);
 323       if (!e->inline_failed || warn_inline)
 324         inline_p = true;
 325     }
 326
 327   if (inline_p)
 328     {
 329       timevar_push (TV_INTEGRATION);
 330       todo = optimize_inline_calls (current_function_decl);
 331       timevar_pop (TV_INTEGRATION);
 332     }
 333   cfun->always_inline_functions_inlined = true;
 334   cfun->after_inlining = true;
 335   return todo | execute_fixup_cfg ();
 336 }