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 and increase loop nest
  53    by NEST.  */
  54
  55 static void
  56 update_noncloned_frequencies (struct cgraph_node *node,
  57                               int freq_scale, int nest)
  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->loop_nest += nest;
  67       e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
  68       if (e->frequency > CGRAPH_FREQ_MAX)
  69         e->frequency = CGRAPH_FREQ_MAX;
  70       if (!e->inline_failed)
  71         update_noncloned_frequencies (e->callee, freq_scale, nest);
  72     }
  73 }
  74
  75
  76 /* E is expected to be an edge being inlined.  Clone destination node of
  77    the edge and redirect it to the new clone.
  78    DUPLICATE is used for bookkeeping on whether we are actually creating new
  79    clones or re-using node originally representing out-of-line function call.
  80    */
  81
  82 void
  83 clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
  84                      bool update_original, int *overall_size)
  85 {
  86   HOST_WIDE_INT peak;
  87   struct inline_summary *caller_info, *callee_info;
  88
  89   if (duplicate)
  90     {
  91       /* We may eliminate the need for out-of-line copy to be output.
  92          In that case just go ahead and re-use it.  This is not just an
  93          memory optimization.  Making offline copy of fuction disappear
  94          from the program will improve future decisions on inlining.  */
  95       if (!e->callee->callers->next_caller
  96           /* Recursive inlining never wants the master clone to
  97              be overwritten.  */
  98           && update_original
  99           /* FIXME: When address is taken of DECL_EXTERNAL function we still
 100              can remove its offline copy, but we would need to keep unanalyzed
 101              node in the callgraph so references can point to it.  */
 102           && !e->callee->address_taken
 103           && cgraph_can_remove_if_no_direct_calls_p (e->callee)
 104           /* Inlining might enable more devirtualizing, so we want to remove
 105              those only after all devirtualizable virtual calls are processed.
 106              Lacking may edges in callgraph we just preserve them post
 107              inlining.  */
 108           && (!DECL_VIRTUAL_P (e->callee->decl)
 109               || (!DECL_COMDAT (e->callee->decl)
 110                   && !DECL_EXTERNAL (e->callee->decl)))
 111           /* Don't reuse if more than one function shares a comdat group.
 112              If the other function(s) are needed, we need to emit even
 113              this function out of line.  */
 114           && !e->callee->same_comdat_group
 115           /* During early inlining some unanalyzed cgraph nodes might be in the
 116              callgraph and they might reffer the function in question.  */
 117           && !cgraph_new_nodes)
 118         {
 119           gcc_assert (!e->callee->global.inlined_to);
 120           if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->decl))
 121             {
 122               if (overall_size)
 123                 *overall_size -= inline_summary (e->callee)->size;
 124               nfunctions_inlined++;
 125             }
 126           duplicate = false;
 127           e->callee->local.externally_visible = false;
 128           update_noncloned_frequencies (e->callee, e->frequency, e->loop_nest);
 129         }
 130       else
 131         {
 132           struct cgraph_node *n;
 133           n = cgraph_clone_node (e->callee, e->callee->decl,
 134                                  e->count, e->frequency, e->loop_nest,
 135                                  update_original, NULL);
 136           cgraph_redirect_edge_callee (e, n);
 137         }
 138     }
 139
 140   callee_info = inline_summary (e->callee);
 141   caller_info = inline_summary (e->caller);
 142
 143   if (e->caller->global.inlined_to)
 144     e->callee->global.inlined_to = e->caller->global.inlined_to;
 145   else
 146     e->callee->global.inlined_to = e->caller;
 147   callee_info->stack_frame_offset
 148     = caller_info->stack_frame_offset
 149       + caller_info->estimated_self_stack_size;
 150   peak = callee_info->stack_frame_offset
 151       + callee_info->estimated_self_stack_size;
 152   if (inline_summary (e->callee->global.inlined_to)->estimated_stack_size
 153       < peak)
 154     inline_summary (e->callee->global.inlined_to)->estimated_stack_size = peak;
 155   cgraph_propagate_frequency (e->callee);
 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 inline_summary *info;
 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 (!e->callee->global.inlined_to);
 184
 185   e->inline_failed = CIF_OK;
 186   DECL_POSSIBLY_INLINED (e->callee->decl) = true;
 187
 188   clone_inlined_nodes (e, true, update_original, overall_size);
 189
 190   /* Now update size of caller and all functions caller is inlined into.  */
 191   for (;e && !e->inline_failed; e = e->caller->callers)
 192     {
 193       to = e->caller;
 194       info = inline_summary (to);
 195       old_size = info->size;
 196       new_size = estimate_size_after_inlining (to, curr);
 197       info->size = new_size;
 198       info->time = estimate_time_after_inlining (to, curr);
 199     }
 200   gcc_assert (curr->callee->global.inlined_to == to);
 201   if (overall_size && new_size > old_size)
 202     *overall_size += new_size - old_size;
 203   ncalls_inlined++;
 204
 205   if (flag_indirect_inlining && optimize)
 206     return ipa_propagate_indirect_call_infos (curr, new_edges);
 207   else
 208     return false;
 209 }
 210
 211
 212 /* Copy function body of NODE and redirect all inline clones to it.
 213    This is done before inline plan is applied to NODE when there are
 214    still some inline clones if it.
 215
 216    This is neccesary because inline decisions are not really transitive
 217    and the other inline clones may have different bodies.  */
 218
 219 static struct cgraph_node *
 220 save_inline_function_body (struct cgraph_node *node)
 221 {
 222   struct cgraph_node *first_clone, *n;
 223
 224   if (dump_file)
 225     fprintf (dump_file, "\nSaving body of %s for later reuse\n",
 226              cgraph_node_name (node));
 227
 228   gcc_assert (node == cgraph_get_node (node->decl));
 229
 230   /* first_clone will be turned into real function.  */
 231   first_clone = node->clones;
 232   first_clone->decl = copy_node (node->decl);
 233   cgraph_insert_node_to_hashtable (first_clone);
 234   gcc_assert (first_clone == cgraph_get_node (first_clone->decl));
 235
 236   /* Now reshape the clone tree, so all other clones descends from
 237      first_clone.  */
 238   if (first_clone->next_sibling_clone)
 239     {
 240       for (n = first_clone->next_sibling_clone; n->next_sibling_clone; n = n->next_sibling_clone)
 241         n->clone_of = first_clone;
 242       n->clone_of = first_clone;
 243       n->next_sibling_clone = first_clone->clones;
 244       if (first_clone->clones)
 245         first_clone->clones->prev_sibling_clone = n;
 246       first_clone->clones = first_clone->next_sibling_clone;
 247       first_clone->next_sibling_clone->prev_sibling_clone = NULL;
 248       first_clone->next_sibling_clone = NULL;
 249       gcc_assert (!first_clone->prev_sibling_clone);
 250     }
 251   first_clone->clone_of = NULL;
 252
 253   /* Now node in question has no clones.  */
 254   node->clones = NULL;
 255
 256   /* Inline clones share decl with the function they are cloned
 257      from.  Walk the whole clone tree and redirect them all to the
 258      new decl.  */
 259   if (first_clone->clones)
 260     for (n = first_clone->clones; n != first_clone;)
 261       {
 262         gcc_assert (n->decl == node->decl);
 263         n->decl = first_clone->decl;
 264         if (n->clones)
 265           n = n->clones;
 266         else if (n->next_sibling_clone)
 267           n = n->next_sibling_clone;
 268         else
 269           {
 270             while (n != first_clone && !n->next_sibling_clone)
 271               n = n->clone_of;
 272             if (n != first_clone)
 273               n = n->next_sibling_clone;
 274           }
 275       }
 276
 277   /* Copy the OLD_VERSION_NODE function tree to the new version.  */
 278   tree_function_versioning (node->decl, first_clone->decl, NULL, true, NULL,
 279                             NULL, NULL);
 280
 281   /* The function will be short lived and removed after we inline all the clones,
 282      but make it internal so we won't confuse ourself.  */
 283   DECL_EXTERNAL (first_clone->decl) = 0;
 284   DECL_COMDAT_GROUP (first_clone->decl) = NULL_TREE;
 285   TREE_PUBLIC (first_clone->decl) = 0;
 286   DECL_COMDAT (first_clone->decl) = 0;
 287   VEC_free (ipa_opt_pass, heap,
 288             first_clone->ipa_transforms_to_apply);
 289   first_clone->ipa_transforms_to_apply = NULL;
 290
 291 #ifdef ENABLE_CHECKING
 292   verify_cgraph_node (first_clone);
 293 #endif
 294   return first_clone;
 295 }
 296
 297
 298 /* Apply inline plan to function.  */
 299
 300 unsigned int
 301 inline_transform (struct cgraph_node *node)
 302 {
 303   unsigned int todo = 0;
 304   struct cgraph_edge *e;
 305   bool inline_p = false;
 306
 307   /* FIXME: Currently the pass manager is adding inline transform more than
 308      once to some clones.  This needs revisiting after WPA cleanups.  */
 309   if (cfun->after_inlining)
 310     return 0;
 311
 312   /* We might need the body of this function so that we can expand
 313      it inline somewhere else.  */
 314   if (cgraph_preserve_function_body_p (node->decl))
 315     save_inline_function_body (node);
 316
 317   for (e = node->callees; e; e = e->next_callee)
 318     {
 319       cgraph_redirect_edge_call_stmt_to_callee (e);
 320       if (!e->inline_failed || warn_inline)
 321         inline_p = true;
 322     }
 323
 324   if (inline_p)
 325     {
 326       timevar_push (TV_INTEGRATION);
 327       todo = optimize_inline_calls (current_function_decl);
 328       timevar_pop (TV_INTEGRATION);
 329     }
 330   cfun->always_inline_functions_inlined = true;
 331   cfun->after_inlining = true;
 332   return todo | execute_fixup_cfg ();
 333 }