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 "gimple.h"
  50 #include "gimple-iterator.h"
  51 #include "tree-cfg.h"
  52 #include "tree-ssa.h"
  53 #include "tree-inline.h"
  54 #include "cfgloop.h"
  55
  56 static int count_insns (basic_block);
  57 static bool ignore_bb_p (const_basic_block);
  58 static bool better_p (const_edge, const_edge);
  59 static edge find_best_successor (basic_block);
  60 static edge find_best_predecessor (basic_block);
  61 static int find_trace (basic_block, basic_block *);
  62
  63 /* Minimal outgoing edge probability considered for superblock formation.  */
  64 static int probability_cutoff;
  65 static int branch_ratio_cutoff;
  66
  67 /* A bit BB->index is set if BB has already been seen, i.e. it is
  68    connected to some trace already.  */
  69 sbitmap bb_seen;
  70
  71 static inline void
  72 mark_bb_seen (basic_block bb)
  73 {
  74   unsigned int size = SBITMAP_SIZE (bb_seen);
  75
  76   if ((unsigned int)bb->index >= size)
  77     bb_seen = sbitmap_resize (bb_seen, size * 2, 0);
  78
  79   bitmap_set_bit (bb_seen, bb->index);
  80 }
  81
  82 static inline bool
  83 bb_seen_p (basic_block bb)
  84 {
  85   return bitmap_bit_p (bb_seen, bb->index);
  86 }
  87
  88 /* Return true if we should ignore the basic block for purposes of tracing.  */
  89 static bool
  90 ignore_bb_p (const_basic_block bb)
  91 {
  92   gimple g;
  93
  94   if (bb->index < NUM_FIXED_BLOCKS)
  95     return true;
  96   if (optimize_bb_for_size_p (bb))
  97     return true;
  98
  99   /* A transaction is a single entry multiple exit region.  It must be
 100      duplicated in its entirety or not at all.  */
 101   g = last_stmt (CONST_CAST_BB (bb));
 102   if (g && gimple_code (g) == GIMPLE_TRANSACTION)
 103     return true;
 104
 105   return false;
 106 }
 107
 108 /* Return number of instructions in the block.  */
 109
 110 static int
 111 count_insns (basic_block bb)
 112 {
 113   gimple_stmt_iterator gsi;
 114   gimple stmt;
 115   int n = 0;
 116
 117   for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
 118     {
 119       stmt = gsi_stmt (gsi);
 120       n += estimate_num_insns (stmt, &eni_size_weights);
 121     }
 122   return n;
 123 }
 124
 125 /* Return true if E1 is more frequent than E2.  */
 126 static bool
 127 better_p (const_edge e1, const_edge e2)
 128 {
 129   if (e1->count != e2->count)
 130     return e1->count > e2->count;
 131   if (e1->src->frequency * e1->probability !=
 132       e2->src->frequency * e2->probability)
 133     return (e1->src->frequency * e1->probability
 134             > e2->src->frequency * e2->probability);
 135   /* This is needed to avoid changes in the decision after
 136      CFG is modified.  */
 137   if (e1->src != e2->src)
 138     return e1->src->index > e2->src->index;
 139   return e1->dest->index > e2->dest->index;
 140 }
 141
 142 /* Return most frequent successor of basic block BB.  */
 143
 144 static edge
 145 find_best_successor (basic_block bb)
 146 {
 147   edge e;
 148   edge best = NULL;
 149   edge_iterator ei;
 150
 151   FOR_EACH_EDGE (e, ei, bb->succs)
 152     if (!best || better_p (e, best))
 153       best = e;
 154   if (!best || ignore_bb_p (best->dest))
 155     return NULL;
 156   if (best->probability <= probability_cutoff)
 157     return NULL;
 158   return best;
 159 }
 160
 161 /* Return most frequent predecessor of basic block BB.  */
 162
 163 static edge
 164 find_best_predecessor (basic_block bb)
 165 {
 166   edge e;
 167   edge best = NULL;
 168   edge_iterator ei;
 169
 170   FOR_EACH_EDGE (e, ei, bb->preds)
 171     if (!best || better_p (e, best))
 172       best = e;
 173   if (!best || ignore_bb_p (best->src))
 174     return NULL;
 175   if (EDGE_FREQUENCY (best) * REG_BR_PROB_BASE
 176       < bb->frequency * branch_ratio_cutoff)
 177     return NULL;
 178   return best;
 179 }
 180
 181 /* Find the trace using bb and record it in the TRACE array.
 182    Return number of basic blocks recorded.  */
 183
 184 static int
 185 find_trace (basic_block bb, basic_block *trace)
 186 {
 187   int i = 0;
 188   edge e;
 189
 190   if (dump_file)
 191     fprintf (dump_file, "Trace seed %i [%i]", bb->index, bb->frequency);
 192
 193   while ((e = find_best_predecessor (bb)) != NULL)
 194     {
 195       basic_block bb2 = e->src;
 196       if (bb_seen_p (bb2) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 197           || find_best_successor (bb2) != e)
 198         break;
 199       if (dump_file)
 200         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 201       bb = bb2;
 202     }
 203   if (dump_file)
 204     fprintf (dump_file, " forward %i [%i]", bb->index, bb->frequency);
 205   trace[i++] = bb;
 206
 207   /* Follow the trace in forward direction.  */
 208   while ((e = find_best_successor (bb)) != NULL)
 209     {
 210       bb = e->dest;
 211       if (bb_seen_p (bb) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 212           || find_best_predecessor (bb) != e)
 213         break;
 214       if (dump_file)
 215         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 216       trace[i++] = bb;
 217     }
 218   if (dump_file)
 219     fprintf (dump_file, "\n");
 220   return i;
 221 }
 222
 223 /* Look for basic blocks in frequency order, construct traces and tail duplicate
 224    if profitable.  */
 225
 226 static bool
 227 tail_duplicate (void)
 228 {
 229   fibnode_t *blocks = XCNEWVEC (fibnode_t, last_basic_block);
 230   basic_block *trace = XNEWVEC (basic_block, n_basic_blocks);
 231   int *counts = XNEWVEC (int, last_basic_block);
 232   int ninsns = 0, nduplicated = 0;
 233   gcov_type weighted_insns = 0, traced_insns = 0;
 234   fibheap_t heap = fibheap_new ();
 235   gcov_type cover_insns;
 236   int max_dup_insns;
 237   basic_block bb;
 238   bool changed = false;
 239
 240   /* Create an oversized sbitmap to reduce the chance that we need to
 241      resize it.  */
 242   bb_seen = sbitmap_alloc (last_basic_block * 2);
 243   bitmap_clear (bb_seen);
 244   initialize_original_copy_tables ();
 245
 246   if (profile_info && flag_branch_probabilities)
 247     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
 248   else
 249     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
 250   probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
 251
 252   branch_ratio_cutoff =
 253     (REG_BR_PROB_BASE / 100 * PARAM_VALUE (TRACER_MIN_BRANCH_RATIO));
 254
 255   FOR_EACH_BB (bb)
 256     {
 257       int n = count_insns (bb);
 258       if (!ignore_bb_p (bb))
 259         blocks[bb->index] = fibheap_insert (heap, -bb->frequency,
 260                                             bb);
 261
 262       counts [bb->index] = n;
 263       ninsns += n;
 264       weighted_insns += n * bb->frequency;
 265     }
 266
 267   if (profile_info && flag_branch_probabilities)
 268     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE_FEEDBACK);
 269   else
 270     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE);
 271   cover_insns = (weighted_insns * cover_insns + 50) / 100;
 272   max_dup_insns = (ninsns * PARAM_VALUE (TRACER_MAX_CODE_GROWTH) + 50) / 100;
 273
 274   while (traced_insns < cover_insns && nduplicated < max_dup_insns
 275          && !fibheap_empty (heap))
 276     {
 277       basic_block bb = (basic_block) fibheap_extract_min (heap);
 278       int n, pos;
 279
 280       if (!bb)
 281         break;
 282
 283       blocks[bb->index] = NULL;
 284
 285       if (ignore_bb_p (bb))
 286         continue;
 287       gcc_assert (!bb_seen_p (bb));
 288
 289       n = find_trace (bb, trace);
 290
 291       bb = trace[0];
 292       traced_insns += bb->frequency * counts [bb->index];
 293       if (blocks[bb->index])
 294         {
 295           fibheap_delete_node (heap, blocks[bb->index]);
 296           blocks[bb->index] = NULL;
 297         }
 298
 299       for (pos = 1; pos < n; pos++)
 300         {
 301           basic_block bb2 = trace[pos];
 302
 303           if (blocks[bb2->index])
 304             {
 305               fibheap_delete_node (heap, blocks[bb2->index]);
 306               blocks[bb2->index] = NULL;
 307             }
 308           traced_insns += bb2->frequency * counts [bb2->index];
 309           if (EDGE_COUNT (bb2->preds) > 1
 310               && can_duplicate_block_p (bb2)
 311               /* We have the tendency to duplicate the loop header
 312                  of all do { } while loops.  Do not do that - it is
 313                  not profitable and it might create a loop with multiple
 314                  entries or at least rotate the loop.  */
 315               && (!current_loops
 316                   || bb2->loop_father->header != bb2))
 317             {
 318               edge e;
 319               basic_block copy;
 320
 321               nduplicated += counts [bb2->index];
 322
 323               e = find_edge (bb, bb2);
 324
 325               copy = duplicate_block (bb2, e, bb);
 326               flush_pending_stmts (e);
 327
 328               add_phi_args_after_copy (&copy, 1, NULL);
 329
 330               /* Reconsider the original copy of block we've duplicated.
 331                  Removing the most common predecessor may make it to be
 332                  head.  */
 333               blocks[bb2->index] =
 334                 fibheap_insert (heap, -bb2->frequency, bb2);
 335
 336               if (dump_file)
 337                 fprintf (dump_file, "Duplicated %i as %i [%i]\n",
 338                          bb2->index, copy->index, copy->frequency);
 339
 340               bb2 = copy;
 341               changed = true;
 342             }
 343           mark_bb_seen (bb2);
 344           bb = bb2;
 345           /* In case the trace became infrequent, stop duplicating.  */
 346           if (ignore_bb_p (bb))
 347             break;
 348         }
 349       if (dump_file)
 350         fprintf (dump_file, " covered now %.1f\n\n",
 351                  traced_insns * 100.0 / weighted_insns);
 352     }
 353   if (dump_file)
 354     fprintf (dump_file, "Duplicated %i insns (%i%%)\n", nduplicated,
 355              nduplicated * 100 / ninsns);
 356
 357   free_original_copy_tables ();
 358   sbitmap_free (bb_seen);
 359   free (blocks);
 360   free (trace);
 361   free (counts);
 362   fibheap_delete (heap);
 363
 364   return changed;
 365 }
 366
 367 /* Main entry point to this file.  */
 368
 369 static unsigned int
 370 tracer (void)
 371 {
 372   bool changed;
 373
 374   if (n_basic_blocks <= NUM_FIXED_BLOCKS + 1)
 375     return 0;
 376
 377   mark_dfs_back_edges ();
 378   if (dump_file)
 379     brief_dump_cfg (dump_file, dump_flags);
 380
 381   /* Trace formation is done on the fly inside tail_duplicate */
 382   changed = tail_duplicate ();
 383   if (changed)
 384     {
 385       free_dominance_info (CDI_DOMINATORS);
 386       /* If we changed the CFG schedule loops for fixup by cleanup_cfg.  */
 387       if (current_loops)
 388         loops_state_set (LOOPS_NEED_FIXUP);
 389     }
 390
 391   if (dump_file)
 392     brief_dump_cfg (dump_file, dump_flags);
 393
 394   return changed ? TODO_cleanup_cfg : 0;
 395 }
 396 \f
 397 static bool
 398 gate_tracer (void)
 399 {
 400   return (optimize > 0 && flag_tracer && flag_reorder_blocks);
 401 }
 402
 403 namespace {
 404
 405 const pass_data pass_data_tracer =
 406 {
 407   GIMPLE_PASS, /* type */
 408   "tracer", /* name */
 409   OPTGROUP_NONE, /* optinfo_flags */
 410   true, /* has_gate */
 411   true, /* has_execute */
 412   TV_TRACER, /* tv_id */
 413   0, /* properties_required */
 414   0, /* properties_provided */
 415   0, /* properties_destroyed */
 416   0, /* todo_flags_start */
 417   ( TODO_update_ssa | TODO_verify_ssa ), /* todo_flags_finish */
 418 };
 419
 420 class pass_tracer : public gimple_opt_pass
 421 {
 422 public:
 423   pass_tracer (gcc::context *ctxt)
 424     : gimple_opt_pass (pass_data_tracer, ctxt)
 425   {}
 426
 427   /* opt_pass methods: */
 428   bool gate () { return gate_tracer (); }
 429   unsigned int execute () { return tracer (); }
 430
 431 }; // class pass_tracer
 432
 433 } // anon namespace
 434
 435 gimple_opt_pass *
 436 make_pass_tracer (gcc::context *ctxt)
 437 {
 438   return new pass_tracer (ctxt);
 439 }