gcc/tracer.c

   1 /* The tracer pass for the GNU compiler.
   2    Contributed by Jan Hubicka, SuSE Labs.
   3    Adapted to work on GIMPLE instead of RTL by Robert Kidd, UIUC.
   4    Copyright (C) 2001-2014 Free Software Foundation, Inc.
   5
   6    This file is part of GCC.
   7
   8    GCC is free software; you can redistribute it and/or modify it
   9    under the terms of the GNU General Public License as published by
  10    the Free Software Foundation; either version 3, or (at your option)
  11    any later version.
  12
  13    GCC is distributed in the hope that it will be useful, but WITHOUT
  14    ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  15    or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
  16    License 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 /* This pass performs the tail duplication needed for superblock formation.
  23    For more information see:
  24
  25      Design and Analysis of Profile-Based Optimization in Compaq's
  26      Compilation Tools for Alpha; Journal of Instruction-Level
  27      Parallelism 3 (2000) 1-25
  28
  29    Unlike Compaq's implementation we don't do the loop peeling as most
  30    probably a better job can be done by a special pass and we don't
  31    need to worry too much about the code size implications as the tail
  32    duplicates are crossjumped again if optimizations are not
  33    performed.  */
  34
  35
  36 #include "config.h"
  37 #include "system.h"
  38 #include "coretypes.h"
  39 #include "tm.h"
  40 #include "tree.h"
  41 #include "rtl.h"
  42 #include "hard-reg-set.h"
  43 #include "profile.h"
  44 #include "basic-block.h"
  45 #include "fibheap.h"
  46 #include "flags.h"
  47 #include "params.h"
  48 #include "coverage.h"
  49 #include "tree-pass.h"
  50 #include "tree-ssa-alias.h"
  51 #include "internal-fn.h"
  52 #include "gimple-expr.h"
  53 #include "is-a.h"
  54 #include "gimple.h"
  55 #include "gimple-iterator.h"
  56 #include "tree-cfg.h"
  57 #include "tree-ssa.h"
  58 #include "tree-inline.h"
  59 #include "cfgloop.h"
  60
  61 static int count_insns (basic_block);
  62 static bool ignore_bb_p (const_basic_block);
  63 static bool better_p (const_edge, const_edge);
  64 static edge find_best_successor (basic_block);
  65 static edge find_best_predecessor (basic_block);
  66 static int find_trace (basic_block, basic_block *);
  67
  68 /* Minimal outgoing edge probability considered for superblock formation.  */
  69 static int probability_cutoff;
  70 static int branch_ratio_cutoff;
  71
  72 /* A bit BB->index is set if BB has already been seen, i.e. it is
  73    connected to some trace already.  */
  74 sbitmap bb_seen;
  75
  76 static inline void
  77 mark_bb_seen (basic_block bb)
  78 {
  79   unsigned int size = SBITMAP_SIZE (bb_seen);
  80
  81   if ((unsigned int)bb->index >= size)
  82     bb_seen = sbitmap_resize (bb_seen, size * 2, 0);
  83
  84   bitmap_set_bit (bb_seen, bb->index);
  85 }
  86
  87 static inline bool
  88 bb_seen_p (basic_block bb)
  89 {
  90   return bitmap_bit_p (bb_seen, bb->index);
  91 }
  92
  93 /* Return true if we should ignore the basic block for purposes of tracing.  */
  94 static bool
  95 ignore_bb_p (const_basic_block bb)
  96 {
  97   gimple g;
  98
  99   if (bb->index < NUM_FIXED_BLOCKS)
 100     return true;
 101   if (optimize_bb_for_size_p (bb))
 102     return true;
 103
 104   /* A transaction is a single entry multiple exit region.  It must be
 105      duplicated in its entirety or not at all.  */
 106   g = last_stmt (CONST_CAST_BB (bb));
 107   if (g && gimple_code (g) == GIMPLE_TRANSACTION)
 108     return true;
 109
 110   return false;
 111 }
 112
 113 /* Return number of instructions in the block.  */
 114
 115 static int
 116 count_insns (basic_block bb)
 117 {
 118   gimple_stmt_iterator gsi;
 119   gimple stmt;
 120   int n = 0;
 121
 122   for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
 123     {
 124       stmt = gsi_stmt (gsi);
 125       n += estimate_num_insns (stmt, &eni_size_weights);
 126     }
 127   return n;
 128 }
 129
 130 /* Return true if E1 is more frequent than E2.  */
 131 static bool
 132 better_p (const_edge e1, const_edge e2)
 133 {
 134   if (e1->count != e2->count)
 135     return e1->count > e2->count;
 136   if (e1->src->frequency * e1->probability !=
 137       e2->src->frequency * e2->probability)
 138     return (e1->src->frequency * e1->probability
 139             > e2->src->frequency * e2->probability);
 140   /* This is needed to avoid changes in the decision after
 141      CFG is modified.  */
 142   if (e1->src != e2->src)
 143     return e1->src->index > e2->src->index;
 144   return e1->dest->index > e2->dest->index;
 145 }
 146
 147 /* Return most frequent successor of basic block BB.  */
 148
 149 static edge
 150 find_best_successor (basic_block bb)
 151 {
 152   edge e;
 153   edge best = NULL;
 154   edge_iterator ei;
 155
 156   FOR_EACH_EDGE (e, ei, bb->succs)
 157     if (!best || better_p (e, best))
 158       best = e;
 159   if (!best || ignore_bb_p (best->dest))
 160     return NULL;
 161   if (best->probability <= probability_cutoff)
 162     return NULL;
 163   return best;
 164 }
 165
 166 /* Return most frequent predecessor of basic block BB.  */
 167
 168 static edge
 169 find_best_predecessor (basic_block bb)
 170 {
 171   edge e;
 172   edge best = NULL;
 173   edge_iterator ei;
 174
 175   FOR_EACH_EDGE (e, ei, bb->preds)
 176     if (!best || better_p (e, best))
 177       best = e;
 178   if (!best || ignore_bb_p (best->src))
 179     return NULL;
 180   if (EDGE_FREQUENCY (best) * REG_BR_PROB_BASE
 181       < bb->frequency * branch_ratio_cutoff)
 182     return NULL;
 183   return best;
 184 }
 185
 186 /* Find the trace using bb and record it in the TRACE array.
 187    Return number of basic blocks recorded.  */
 188
 189 static int
 190 find_trace (basic_block bb, basic_block *trace)
 191 {
 192   int i = 0;
 193   edge e;
 194
 195   if (dump_file)
 196     fprintf (dump_file, "Trace seed %i [%i]", bb->index, bb->frequency);
 197
 198   while ((e = find_best_predecessor (bb)) != NULL)
 199     {
 200       basic_block bb2 = e->src;
 201       if (bb_seen_p (bb2) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 202           || find_best_successor (bb2) != e)
 203         break;
 204       if (dump_file)
 205         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 206       bb = bb2;
 207     }
 208   if (dump_file)
 209     fprintf (dump_file, " forward %i [%i]", bb->index, bb->frequency);
 210   trace[i++] = bb;
 211
 212   /* Follow the trace in forward direction.  */
 213   while ((e = find_best_successor (bb)) != NULL)
 214     {
 215       bb = e->dest;
 216       if (bb_seen_p (bb) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 217           || find_best_predecessor (bb) != e)
 218         break;
 219       if (dump_file)
 220         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 221       trace[i++] = bb;
 222     }
 223   if (dump_file)
 224     fprintf (dump_file, "\n");
 225   return i;
 226 }
 227
 228 /* Look for basic blocks in frequency order, construct traces and tail duplicate
 229    if profitable.  */
 230
 231 static bool
 232 tail_duplicate (void)
 233 {
 234   fibnode_t *blocks = XCNEWVEC (fibnode_t, last_basic_block_for_fn (cfun));
 235   basic_block *trace = XNEWVEC (basic_block, n_basic_blocks_for_fn (cfun));
 236   int *counts = XNEWVEC (int, last_basic_block_for_fn (cfun));
 237   int ninsns = 0, nduplicated = 0;
 238   gcov_type weighted_insns = 0, traced_insns = 0;
 239   fibheap_t heap = fibheap_new ();
 240   gcov_type cover_insns;
 241   int max_dup_insns;
 242   basic_block bb;
 243   bool changed = false;
 244
 245   /* Create an oversized sbitmap to reduce the chance that we need to
 246      resize it.  */
 247   bb_seen = sbitmap_alloc (last_basic_block_for_fn (cfun) * 2);
 248   bitmap_clear (bb_seen);
 249   initialize_original_copy_tables ();
 250
 251   if (profile_info && flag_branch_probabilities)
 252     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
 253   else
 254     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
 255   probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
 256
 257   branch_ratio_cutoff =
 258     (REG_BR_PROB_BASE / 100 * PARAM_VALUE (TRACER_MIN_BRANCH_RATIO));
 259
 260   FOR_EACH_BB_FN (bb, cfun)
 261     {
 262       int n = count_insns (bb);
 263       if (!ignore_bb_p (bb))
 264         blocks[bb->index] = fibheap_insert (heap, -bb->frequency,
 265                                             bb);
 266
 267       counts [bb->index] = n;
 268       ninsns += n;
 269       weighted_insns += n * bb->frequency;
 270     }
 271
 272   if (profile_info && flag_branch_probabilities)
 273     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE_FEEDBACK);
 274   else
 275     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE);
 276   cover_insns = (weighted_insns * cover_insns + 50) / 100;
 277   max_dup_insns = (ninsns * PARAM_VALUE (TRACER_MAX_CODE_GROWTH) + 50) / 100;
 278
 279   while (traced_insns < cover_insns && nduplicated < max_dup_insns
 280          && !fibheap_empty (heap))
 281     {
 282       basic_block bb = (basic_block) fibheap_extract_min (heap);
 283       int n, pos;
 284
 285       if (!bb)
 286         break;
 287
 288       blocks[bb->index] = NULL;
 289
 290       if (ignore_bb_p (bb))
 291         continue;
 292       gcc_assert (!bb_seen_p (bb));
 293
 294       n = find_trace (bb, trace);
 295
 296       bb = trace[0];
 297       traced_insns += bb->frequency * counts [bb->index];
 298       if (blocks[bb->index])
 299         {
 300           fibheap_delete_node (heap, blocks[bb->index]);
 301           blocks[bb->index] = NULL;
 302         }
 303
 304       for (pos = 1; pos < n; pos++)
 305         {
 306           basic_block bb2 = trace[pos];
 307
 308           if (blocks[bb2->index])
 309             {
 310               fibheap_delete_node (heap, blocks[bb2->index]);
 311               blocks[bb2->index] = NULL;
 312             }
 313           traced_insns += bb2->frequency * counts [bb2->index];
 314           if (EDGE_COUNT (bb2->preds) > 1
 315               && can_duplicate_block_p (bb2)
 316               /* We have the tendency to duplicate the loop header
 317                  of all do { } while loops.  Do not do that - it is
 318                  not profitable and it might create a loop with multiple
 319                  entries or at least rotate the loop.  */
 320               && bb2->loop_father->header != bb2)
 321             {
 322               edge e;
 323               basic_block copy;
 324
 325               nduplicated += counts [bb2->index];
 326
 327               e = find_edge (bb, bb2);
 328
 329               copy = duplicate_block (bb2, e, bb);
 330               flush_pending_stmts (e);
 331
 332               add_phi_args_after_copy (&copy, 1, NULL);
 333
 334               /* Reconsider the original copy of block we've duplicated.
 335                  Removing the most common predecessor may make it to be
 336                  head.  */
 337               blocks[bb2->index] =
 338                 fibheap_insert (heap, -bb2->frequency, bb2);
 339
 340               if (dump_file)
 341                 fprintf (dump_file, "Duplicated %i as %i [%i]\n",
 342                          bb2->index, copy->index, copy->frequency);
 343
 344               bb2 = copy;
 345               changed = true;
 346             }
 347           mark_bb_seen (bb2);
 348           bb = bb2;
 349           /* In case the trace became infrequent, stop duplicating.  */
 350           if (ignore_bb_p (bb))
 351             break;
 352         }
 353       if (dump_file)
 354         fprintf (dump_file, " covered now %.1f\n\n",
 355                  traced_insns * 100.0 / weighted_insns);
 356     }
 357   if (dump_file)
 358     fprintf (dump_file, "Duplicated %i insns (%i%%)\n", nduplicated,
 359              nduplicated * 100 / ninsns);
 360
 361   free_original_copy_tables ();
 362   sbitmap_free (bb_seen);
 363   free (blocks);
 364   free (trace);
 365   free (counts);
 366   fibheap_delete (heap);
 367
 368   return changed;
 369 }
 370 \f
 371 namespace {
 372
 373 const pass_data pass_data_tracer =
 374 {
 375   GIMPLE_PASS, /* type */
 376   "tracer", /* name */
 377   OPTGROUP_NONE, /* optinfo_flags */
 378   TV_TRACER, /* tv_id */
 379   0, /* properties_required */
 380   0, /* properties_provided */
 381   0, /* properties_destroyed */
 382   0, /* todo_flags_start */
 383   TODO_update_ssa, /* todo_flags_finish */
 384 };
 385
 386 class pass_tracer : public gimple_opt_pass
 387 {
 388 public:
 389   pass_tracer (gcc::context *ctxt)
 390     : gimple_opt_pass (pass_data_tracer, ctxt)
 391   {}
 392
 393   /* opt_pass methods: */
 394   virtual bool gate (function *)
 395     {
 396       return (optimize > 0 && flag_tracer && flag_reorder_blocks);
 397     }
 398
 399   virtual unsigned int execute (function *);
 400
 401 }; // class pass_tracer
 402
 403 unsigned int
 404 pass_tracer::execute (function *fun)
 405 {
 406   bool changed;
 407
 408   if (n_basic_blocks_for_fn (fun) <= NUM_FIXED_BLOCKS + 1)
 409     return 0;
 410
 411   mark_dfs_back_edges ();
 412   if (dump_file)
 413     brief_dump_cfg (dump_file, dump_flags);
 414
 415   /* Trace formation is done on the fly inside tail_duplicate */
 416   changed = tail_duplicate ();
 417   if (changed)
 418     {
 419       free_dominance_info (CDI_DOMINATORS);
 420       /* If we changed the CFG schedule loops for fixup by cleanup_cfg.  */
 421       loops_state_set (LOOPS_NEED_FIXUP);
 422     }
 423
 424   if (dump_file)
 425     brief_dump_cfg (dump_file, dump_flags);
 426
 427   return changed ? TODO_cleanup_cfg : 0;
 428 }
 429 } // anon namespace
 430
 431 gimple_opt_pass *
 432 make_pass_tracer (gcc::context *ctxt)
 433 {
 434   return new pass_tracer (ctxt);
 435 }