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) 2001-2013 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 "basic-block.h"
  44 #include "fibheap.h"
  45 #include "flags.h"
  46 #include "params.h"
  47 #include "coverage.h"
  48 #include "tree-pass.h"
  49 #include "tree-flow.h"
  50 #include "tree-inline.h"
  51 #include "cfgloop.h"
  52
  53 static int count_insns (basic_block);
  54 static bool ignore_bb_p (const_basic_block);
  55 static bool better_p (const_edge, const_edge);
  56 static edge find_best_successor (basic_block);
  57 static edge find_best_predecessor (basic_block);
  58 static int find_trace (basic_block, basic_block *);
  59
  60 /* Minimal outgoing edge probability considered for superblock formation.  */
  61 static int probability_cutoff;
  62 static int branch_ratio_cutoff;
  63
  64 /* A bit BB->index is set if BB has already been seen, i.e. it is
  65    connected to some trace already.  */
  66 sbitmap bb_seen;
  67
  68 static inline void
  69 mark_bb_seen (basic_block bb)
  70 {
  71   unsigned int size = SBITMAP_SIZE (bb_seen);
  72
  73   if ((unsigned int)bb->index >= size)
  74     bb_seen = sbitmap_resize (bb_seen, size * 2, 0);
  75
  76   bitmap_set_bit (bb_seen, bb->index);
  77 }
  78
  79 static inline bool
  80 bb_seen_p (basic_block bb)
  81 {
  82   return bitmap_bit_p (bb_seen, bb->index);
  83 }
  84
  85 /* Return true if we should ignore the basic block for purposes of tracing.  */
  86 static bool
  87 ignore_bb_p (const_basic_block bb)
  88 {
  89   gimple g;
  90
  91   if (bb->index < NUM_FIXED_BLOCKS)
  92     return true;
  93   if (optimize_bb_for_size_p (bb))
  94     return true;
  95
  96   /* A transaction is a single entry multiple exit region.  It must be
  97      duplicated in its entirety or not at all.  */
  98   g = last_stmt (CONST_CAST_BB (bb));
  99   if (g && gimple_code (g) == GIMPLE_TRANSACTION)
 100     return true;
 101
 102   return false;
 103 }
 104
 105 /* Return number of instructions in the block.  */
 106
 107 static int
 108 count_insns (basic_block bb)
 109 {
 110   gimple_stmt_iterator gsi;
 111   gimple stmt;
 112   int n = 0;
 113
 114   for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
 115     {
 116       stmt = gsi_stmt (gsi);
 117       n += estimate_num_insns (stmt, &eni_size_weights);
 118     }
 119   return n;
 120 }
 121
 122 /* Return true if E1 is more frequent than E2.  */
 123 static bool
 124 better_p (const_edge e1, const_edge e2)
 125 {
 126   if (e1->count != e2->count)
 127     return e1->count > e2->count;
 128   if (e1->src->frequency * e1->probability !=
 129       e2->src->frequency * e2->probability)
 130     return (e1->src->frequency * e1->probability
 131             > e2->src->frequency * e2->probability);
 132   /* This is needed to avoid changes in the decision after
 133      CFG is modified.  */
 134   if (e1->src != e2->src)
 135     return e1->src->index > e2->src->index;
 136   return e1->dest->index > e2->dest->index;
 137 }
 138
 139 /* Return most frequent successor of basic block BB.  */
 140
 141 static edge
 142 find_best_successor (basic_block bb)
 143 {
 144   edge e;
 145   edge best = NULL;
 146   edge_iterator ei;
 147
 148   FOR_EACH_EDGE (e, ei, bb->succs)
 149     if (!best || better_p (e, best))
 150       best = e;
 151   if (!best || ignore_bb_p (best->dest))
 152     return NULL;
 153   if (best->probability <= probability_cutoff)
 154     return NULL;
 155   return best;
 156 }
 157
 158 /* Return most frequent predecessor of basic block BB.  */
 159
 160 static edge
 161 find_best_predecessor (basic_block bb)
 162 {
 163   edge e;
 164   edge best = NULL;
 165   edge_iterator ei;
 166
 167   FOR_EACH_EDGE (e, ei, bb->preds)
 168     if (!best || better_p (e, best))
 169       best = e;
 170   if (!best || ignore_bb_p (best->src))
 171     return NULL;
 172   if (EDGE_FREQUENCY (best) * REG_BR_PROB_BASE
 173       < bb->frequency * branch_ratio_cutoff)
 174     return NULL;
 175   return best;
 176 }
 177
 178 /* Find the trace using bb and record it in the TRACE array.
 179    Return number of basic blocks recorded.  */
 180
 181 static int
 182 find_trace (basic_block bb, basic_block *trace)
 183 {
 184   int i = 0;
 185   edge e;
 186
 187   if (dump_file)
 188     fprintf (dump_file, "Trace seed %i [%i]", bb->index, bb->frequency);
 189
 190   while ((e = find_best_predecessor (bb)) != NULL)
 191     {
 192       basic_block bb2 = e->src;
 193       if (bb_seen_p (bb2) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 194           || find_best_successor (bb2) != e)
 195         break;
 196       if (dump_file)
 197         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 198       bb = bb2;
 199     }
 200   if (dump_file)
 201     fprintf (dump_file, " forward %i [%i]", bb->index, bb->frequency);
 202   trace[i++] = bb;
 203
 204   /* Follow the trace in forward direction.  */
 205   while ((e = find_best_successor (bb)) != NULL)
 206     {
 207       bb = e->dest;
 208       if (bb_seen_p (bb) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 209           || find_best_predecessor (bb) != e)
 210         break;
 211       if (dump_file)
 212         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 213       trace[i++] = bb;
 214     }
 215   if (dump_file)
 216     fprintf (dump_file, "\n");
 217   return i;
 218 }
 219
 220 /* Look for basic blocks in frequency order, construct traces and tail duplicate
 221    if profitable.  */
 222
 223 static bool
 224 tail_duplicate (void)
 225 {
 226   fibnode_t *blocks = XCNEWVEC (fibnode_t, last_basic_block);
 227   basic_block *trace = XNEWVEC (basic_block, n_basic_blocks);
 228   int *counts = XNEWVEC (int, last_basic_block);
 229   int ninsns = 0, nduplicated = 0;
 230   gcov_type weighted_insns = 0, traced_insns = 0;
 231   fibheap_t heap = fibheap_new ();
 232   gcov_type cover_insns;
 233   int max_dup_insns;
 234   basic_block bb;
 235   bool changed = false;
 236
 237   /* Create an oversized sbitmap to reduce the chance that we need to
 238      resize it.  */
 239   bb_seen = sbitmap_alloc (last_basic_block * 2);
 240   bitmap_clear (bb_seen);
 241   initialize_original_copy_tables ();
 242
 243   if (profile_info && flag_branch_probabilities)
 244     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
 245   else
 246     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
 247   probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
 248
 249   branch_ratio_cutoff =
 250     (REG_BR_PROB_BASE / 100 * PARAM_VALUE (TRACER_MIN_BRANCH_RATIO));
 251
 252   FOR_EACH_BB (bb)
 253     {
 254       int n = count_insns (bb);
 255       if (!ignore_bb_p (bb))
 256         blocks[bb->index] = fibheap_insert (heap, -bb->frequency,
 257                                             bb);
 258
 259       counts [bb->index] = n;
 260       ninsns += n;
 261       weighted_insns += n * bb->frequency;
 262     }
 263
 264   if (profile_info && flag_branch_probabilities)
 265     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE_FEEDBACK);
 266   else
 267     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE);
 268   cover_insns = (weighted_insns * cover_insns + 50) / 100;
 269   max_dup_insns = (ninsns * PARAM_VALUE (TRACER_MAX_CODE_GROWTH) + 50) / 100;
 270
 271   while (traced_insns < cover_insns && nduplicated < max_dup_insns
 272          && !fibheap_empty (heap))
 273     {
 274       basic_block bb = (basic_block) fibheap_extract_min (heap);
 275       int n, pos;
 276
 277       if (!bb)
 278         break;
 279
 280       blocks[bb->index] = NULL;
 281
 282       if (ignore_bb_p (bb))
 283         continue;
 284       gcc_assert (!bb_seen_p (bb));
 285
 286       n = find_trace (bb, trace);
 287
 288       bb = trace[0];
 289       traced_insns += bb->frequency * counts [bb->index];
 290       if (blocks[bb->index])
 291         {
 292           fibheap_delete_node (heap, blocks[bb->index]);
 293           blocks[bb->index] = NULL;
 294         }
 295
 296       for (pos = 1; pos < n; pos++)
 297         {
 298           basic_block bb2 = trace[pos];
 299
 300           if (blocks[bb2->index])
 301             {
 302               fibheap_delete_node (heap, blocks[bb2->index]);
 303               blocks[bb2->index] = NULL;
 304             }
 305           traced_insns += bb2->frequency * counts [bb2->index];
 306           if (EDGE_COUNT (bb2->preds) > 1
 307               && can_duplicate_block_p (bb2)
 308               /* We have the tendency to duplicate the loop header
 309                  of all do { } while loops.  Do not do that - it is
 310                  not profitable and it might create a loop with multiple
 311                  entries or at least rotate the loop.  */
 312               && (!current_loops
 313                   || bb2->loop_father->header != bb2))
 314             {
 315               edge e;
 316               basic_block copy;
 317
 318               nduplicated += counts [bb2->index];
 319
 320               e = find_edge (bb, bb2);
 321
 322               copy = duplicate_block (bb2, e, bb);
 323               flush_pending_stmts (e);
 324
 325               add_phi_args_after_copy (&copy, 1, NULL);
 326
 327               /* Reconsider the original copy of block we've duplicated.
 328                  Removing the most common predecessor may make it to be
 329                  head.  */
 330               blocks[bb2->index] =
 331                 fibheap_insert (heap, -bb2->frequency, bb2);
 332
 333               if (dump_file)
 334                 fprintf (dump_file, "Duplicated %i as %i [%i]\n",
 335                          bb2->index, copy->index, copy->frequency);
 336
 337               bb2 = copy;
 338               changed = true;
 339             }
 340           mark_bb_seen (bb2);
 341           bb = bb2;
 342           /* In case the trace became infrequent, stop duplicating.  */
 343           if (ignore_bb_p (bb))
 344             break;
 345         }
 346       if (dump_file)
 347         fprintf (dump_file, " covered now %.1f\n\n",
 348                  traced_insns * 100.0 / weighted_insns);
 349     }
 350   if (dump_file)
 351     fprintf (dump_file, "Duplicated %i insns (%i%%)\n", nduplicated,
 352              nduplicated * 100 / ninsns);
 353
 354   free_original_copy_tables ();
 355   sbitmap_free (bb_seen);
 356   free (blocks);
 357   free (trace);
 358   free (counts);
 359   fibheap_delete (heap);
 360
 361   return changed;
 362 }
 363
 364 /* Main entry point to this file.  */
 365
 366 static unsigned int
 367 tracer (void)
 368 {
 369   bool changed;
 370
 371   if (n_basic_blocks <= NUM_FIXED_BLOCKS + 1)
 372     return 0;
 373
 374   mark_dfs_back_edges ();
 375   if (dump_file)
 376     brief_dump_cfg (dump_file, dump_flags);
 377
 378   /* Trace formation is done on the fly inside tail_duplicate */
 379   changed = tail_duplicate ();
 380   if (changed)
 381     {
 382       free_dominance_info (CDI_DOMINATORS);
 383       /* If we changed the CFG schedule loops for fixup by cleanup_cfg.  */
 384       if (current_loops)
 385         loops_state_set (LOOPS_NEED_FIXUP);
 386     }
 387
 388   if (dump_file)
 389     brief_dump_cfg (dump_file, dump_flags);
 390
 391   return changed ? TODO_cleanup_cfg : 0;
 392 }
 393 \f
 394 static bool
 395 gate_tracer (void)
 396 {
 397   return (optimize > 0 && flag_tracer && flag_reorder_blocks);
 398 }
 399
 400 struct gimple_opt_pass pass_tracer =
 401 {
 402  {
 403   GIMPLE_PASS,
 404   "tracer",                             /* name */
 405   OPTGROUP_NONE,                        /* optinfo_flags */
 406   gate_tracer,                          /* gate */
 407   tracer,                               /* execute */
 408   NULL,                                 /* sub */
 409   NULL,                                 /* next */
 410   0,                                    /* static_pass_number */
 411   TV_TRACER,                            /* tv_id */
 412   0,                                    /* properties_required */
 413   0,                                    /* properties_provided */
 414   0,                                    /* properties_destroyed */
 415   0,                                    /* todo_flags_start */
 416   TODO_update_ssa
 417     | TODO_verify_ssa                   /* todo_flags_finish */
 418  }
 419 };