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) 2001-2015 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 "backend.h"
  40 #include "tree.h"
  41 #include "gimple.h"
  42 #include "rtl.h"
  43 #include "alias.h"
  44 #include "fold-const.h"
  45 #include "profile.h"
  46 #include "cfganal.h"
  47 #include "flags.h"
  48 #include "params.h"
  49 #include "coverage.h"
  50 #include "tree-pass.h"
  51 #include "internal-fn.h"
  52 #include "gimple-iterator.h"
  53 #include "tree-cfg.h"
  54 #include "tree-ssa.h"
  55 #include "tree-inline.h"
  56 #include "cfgloop.h"
  57 #include "fibonacci_heap.h"
  58
  59 static int count_insns (basic_block);
  60 static bool ignore_bb_p (const_basic_block);
  61 static bool better_p (const_edge, const_edge);
  62 static edge find_best_successor (basic_block);
  63 static edge find_best_predecessor (basic_block);
  64 static int find_trace (basic_block, basic_block *);
  65
  66 /* Minimal outgoing edge probability considered for superblock formation.  */
  67 static int probability_cutoff;
  68 static int branch_ratio_cutoff;
  69
  70 /* A bit BB->index is set if BB has already been seen, i.e. it is
  71    connected to some trace already.  */
  72 sbitmap bb_seen;
  73
  74 static inline void
  75 mark_bb_seen (basic_block bb)
  76 {
  77   unsigned int size = SBITMAP_SIZE (bb_seen);
  78
  79   if ((unsigned int)bb->index >= size)
  80     bb_seen = sbitmap_resize (bb_seen, size * 2, 0);
  81
  82   bitmap_set_bit (bb_seen, bb->index);
  83 }
  84
  85 static inline bool
  86 bb_seen_p (basic_block bb)
  87 {
  88   return bitmap_bit_p (bb_seen, bb->index);
  89 }
  90
  91 /* Return true if we should ignore the basic block for purposes of tracing.  */
  92 static bool
  93 ignore_bb_p (const_basic_block bb)
  94 {
  95   gimple g;
  96
  97   if (bb->index < NUM_FIXED_BLOCKS)
  98     return true;
  99   if (optimize_bb_for_size_p (bb))
 100     return true;
 101
 102   /* A transaction is a single entry multiple exit region.  It must be
 103      duplicated in its entirety or not at all.  */
 104   g = last_stmt (CONST_CAST_BB (bb));
 105   if (g && gimple_code (g) == GIMPLE_TRANSACTION)
 106     return true;
 107
 108   return false;
 109 }
 110
 111 /* Return number of instructions in the block.  */
 112
 113 static int
 114 count_insns (basic_block bb)
 115 {
 116   gimple_stmt_iterator gsi;
 117   gimple stmt;
 118   int n = 0;
 119
 120   for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
 121     {
 122       stmt = gsi_stmt (gsi);
 123       n += estimate_num_insns (stmt, &eni_size_weights);
 124     }
 125   return n;
 126 }
 127
 128 /* Return true if E1 is more frequent than E2.  */
 129 static bool
 130 better_p (const_edge e1, const_edge e2)
 131 {
 132   if (e1->count != e2->count)
 133     return e1->count > e2->count;
 134   if (e1->src->frequency * e1->probability !=
 135       e2->src->frequency * e2->probability)
 136     return (e1->src->frequency * e1->probability
 137             > e2->src->frequency * e2->probability);
 138   /* This is needed to avoid changes in the decision after
 139      CFG is modified.  */
 140   if (e1->src != e2->src)
 141     return e1->src->index > e2->src->index;
 142   return e1->dest->index > e2->dest->index;
 143 }
 144
 145 /* Return most frequent successor of basic block BB.  */
 146
 147 static edge
 148 find_best_successor (basic_block bb)
 149 {
 150   edge e;
 151   edge best = NULL;
 152   edge_iterator ei;
 153
 154   FOR_EACH_EDGE (e, ei, bb->succs)
 155     if (!best || better_p (e, best))
 156       best = e;
 157   if (!best || ignore_bb_p (best->dest))
 158     return NULL;
 159   if (best->probability <= probability_cutoff)
 160     return NULL;
 161   return best;
 162 }
 163
 164 /* Return most frequent predecessor of basic block BB.  */
 165
 166 static edge
 167 find_best_predecessor (basic_block bb)
 168 {
 169   edge e;
 170   edge best = NULL;
 171   edge_iterator ei;
 172
 173   FOR_EACH_EDGE (e, ei, bb->preds)
 174     if (!best || better_p (e, best))
 175       best = e;
 176   if (!best || ignore_bb_p (best->src))
 177     return NULL;
 178   if (EDGE_FREQUENCY (best) * REG_BR_PROB_BASE
 179       < bb->frequency * branch_ratio_cutoff)
 180     return NULL;
 181   return best;
 182 }
 183
 184 /* Find the trace using bb and record it in the TRACE array.
 185    Return number of basic blocks recorded.  */
 186
 187 static int
 188 find_trace (basic_block bb, basic_block *trace)
 189 {
 190   int i = 0;
 191   edge e;
 192
 193   if (dump_file)
 194     fprintf (dump_file, "Trace seed %i [%i]", bb->index, bb->frequency);
 195
 196   while ((e = find_best_predecessor (bb)) != NULL)
 197     {
 198       basic_block bb2 = e->src;
 199       if (bb_seen_p (bb2) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 200           || find_best_successor (bb2) != e)
 201         break;
 202       if (dump_file)
 203         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 204       bb = bb2;
 205     }
 206   if (dump_file)
 207     fprintf (dump_file, " forward %i [%i]", bb->index, bb->frequency);
 208   trace[i++] = bb;
 209
 210   /* Follow the trace in forward direction.  */
 211   while ((e = find_best_successor (bb)) != NULL)
 212     {
 213       bb = e->dest;
 214       if (bb_seen_p (bb) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 215           || find_best_predecessor (bb) != e)
 216         break;
 217       if (dump_file)
 218         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 219       trace[i++] = bb;
 220     }
 221   if (dump_file)
 222     fprintf (dump_file, "\n");
 223   return i;
 224 }
 225
 226 /* Look for basic blocks in frequency order, construct traces and tail duplicate
 227    if profitable.  */
 228
 229 static bool
 230 tail_duplicate (void)
 231 {
 232   auto_vec<fibonacci_node<long, basic_block_def>*> blocks;
 233   blocks.safe_grow_cleared (last_basic_block_for_fn (cfun));
 234
 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   fibonacci_heap<long, basic_block_def> heap (LONG_MIN);
 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] = heap.insert (-bb->frequency, bb);
 265
 266       counts [bb->index] = n;
 267       ninsns += n;
 268       weighted_insns += n * bb->frequency;
 269     }
 270
 271   if (profile_info && flag_branch_probabilities)
 272     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE_FEEDBACK);
 273   else
 274     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE);
 275   cover_insns = (weighted_insns * cover_insns + 50) / 100;
 276   max_dup_insns = (ninsns * PARAM_VALUE (TRACER_MAX_CODE_GROWTH) + 50) / 100;
 277
 278   while (traced_insns < cover_insns && nduplicated < max_dup_insns
 279          && !heap.empty ())
 280     {
 281       basic_block bb = heap.extract_min ();
 282       int n, pos;
 283
 284       if (!bb)
 285         break;
 286
 287       blocks[bb->index] = NULL;
 288
 289       if (ignore_bb_p (bb))
 290         continue;
 291       gcc_assert (!bb_seen_p (bb));
 292
 293       n = find_trace (bb, trace);
 294
 295       bb = trace[0];
 296       traced_insns += bb->frequency * counts [bb->index];
 297       if (blocks[bb->index])
 298         {
 299           heap.delete_node (blocks[bb->index]);
 300           blocks[bb->index] = NULL;
 301         }
 302
 303       for (pos = 1; pos < n; pos++)
 304         {
 305           basic_block bb2 = trace[pos];
 306
 307           if (blocks[bb2->index])
 308             {
 309               heap.delete_node (blocks[bb2->index]);
 310               blocks[bb2->index] = NULL;
 311             }
 312           traced_insns += bb2->frequency * counts [bb2->index];
 313           if (EDGE_COUNT (bb2->preds) > 1
 314               && can_duplicate_block_p (bb2)
 315               /* We have the tendency to duplicate the loop header
 316                  of all do { } while loops.  Do not do that - it is
 317                  not profitable and it might create a loop with multiple
 318                  entries or at least rotate the loop.  */
 319               && bb2->loop_father->header != bb2)
 320             {
 321               edge e;
 322               basic_block copy;
 323
 324               nduplicated += counts [bb2->index];
 325
 326               e = find_edge (bb, bb2);
 327
 328               copy = duplicate_block (bb2, e, bb);
 329               flush_pending_stmts (e);
 330
 331               add_phi_args_after_copy (&copy, 1, NULL);
 332
 333               /* Reconsider the original copy of block we've duplicated.
 334                  Removing the most common predecessor may make it to be
 335                  head.  */
 336               blocks[bb2->index] = heap.insert (-bb2->frequency, bb2);
 337
 338               if (dump_file)
 339                 fprintf (dump_file, "Duplicated %i as %i [%i]\n",
 340                          bb2->index, copy->index, copy->frequency);
 341
 342               bb2 = copy;
 343               changed = true;
 344             }
 345           mark_bb_seen (bb2);
 346           bb = bb2;
 347           /* In case the trace became infrequent, stop duplicating.  */
 348           if (ignore_bb_p (bb))
 349             break;
 350         }
 351       if (dump_file)
 352         fprintf (dump_file, " covered now %.1f\n\n",
 353                  traced_insns * 100.0 / weighted_insns);
 354     }
 355   if (dump_file)
 356     fprintf (dump_file, "Duplicated %i insns (%i%%)\n", nduplicated,
 357              nduplicated * 100 / ninsns);
 358
 359   free_original_copy_tables ();
 360   sbitmap_free (bb_seen);
 361   free (trace);
 362   free (counts);
 363
 364   return changed;
 365 }
 366 \f
 367 namespace {
 368
 369 const pass_data pass_data_tracer =
 370 {
 371   GIMPLE_PASS, /* type */
 372   "tracer", /* name */
 373   OPTGROUP_NONE, /* optinfo_flags */
 374   TV_TRACER, /* tv_id */
 375   0, /* properties_required */
 376   0, /* properties_provided */
 377   0, /* properties_destroyed */
 378   0, /* todo_flags_start */
 379   TODO_update_ssa, /* todo_flags_finish */
 380 };
 381
 382 class pass_tracer : public gimple_opt_pass
 383 {
 384 public:
 385   pass_tracer (gcc::context *ctxt)
 386     : gimple_opt_pass (pass_data_tracer, ctxt)
 387   {}
 388
 389   /* opt_pass methods: */
 390   virtual bool gate (function *)
 391     {
 392       return (optimize > 0 && flag_tracer && flag_reorder_blocks);
 393     }
 394
 395   virtual unsigned int execute (function *);
 396
 397 }; // class pass_tracer
 398
 399 unsigned int
 400 pass_tracer::execute (function *fun)
 401 {
 402   bool changed;
 403
 404   if (n_basic_blocks_for_fn (fun) <= NUM_FIXED_BLOCKS + 1)
 405     return 0;
 406
 407   mark_dfs_back_edges ();
 408   if (dump_file)
 409     brief_dump_cfg (dump_file, dump_flags);
 410
 411   /* Trace formation is done on the fly inside tail_duplicate */
 412   changed = tail_duplicate ();
 413   if (changed)
 414     {
 415       free_dominance_info (CDI_DOMINATORS);
 416       /* If we changed the CFG schedule loops for fixup by cleanup_cfg.  */
 417       loops_state_set (LOOPS_NEED_FIXUP);
 418     }
 419
 420   if (dump_file)
 421     brief_dump_cfg (dump_file, dump_flags);
 422
 423   return changed ? TODO_cleanup_cfg : 0;
 424 }
 425 } // anon namespace
 426
 427 gimple_opt_pass *
 428 make_pass_tracer (gcc::context *ctxt)
 429 {
 430   return new pass_tracer (ctxt);
 431 }