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 "tm.h"
  40 #include "alias.h"
  41 #include "symtab.h"
  42 #include "tree.h"
  43 #include "fold-const.h"
  44 #include "rtl.h"
  45 #include "hard-reg-set.h"
  46 #include "profile.h"
  47 #include "predict.h"
  48 #include "function.h"
  49 #include "dominance.h"
  50 #include "cfg.h"
  51 #include "cfganal.h"
  52 #include "basic-block.h"
  53 #include "flags.h"
  54 #include "params.h"
  55 #include "coverage.h"
  56 #include "tree-pass.h"
  57 #include "tree-ssa-alias.h"
  58 #include "internal-fn.h"
  59 #include "gimple-expr.h"
  60 #include "gimple.h"
  61 #include "gimple-iterator.h"
  62 #include "tree-cfg.h"
  63 #include "tree-ssa.h"
  64 #include "tree-inline.h"
  65 #include "cfgloop.h"
  66 #include "fibonacci_heap.h"
  67
  68 static int count_insns (basic_block);
  69 static bool ignore_bb_p (const_basic_block);
  70 static bool better_p (const_edge, const_edge);
  71 static edge find_best_successor (basic_block);
  72 static edge find_best_predecessor (basic_block);
  73 static int find_trace (basic_block, basic_block *);
  74
  75 /* Minimal outgoing edge probability considered for superblock formation.  */
  76 static int probability_cutoff;
  77 static int branch_ratio_cutoff;
  78
  79 /* A bit BB->index is set if BB has already been seen, i.e. it is
  80    connected to some trace already.  */
  81 sbitmap bb_seen;
  82
  83 static inline void
  84 mark_bb_seen (basic_block bb)
  85 {
  86   unsigned int size = SBITMAP_SIZE (bb_seen);
  87
  88   if ((unsigned int)bb->index >= size)
  89     bb_seen = sbitmap_resize (bb_seen, size * 2, 0);
  90
  91   bitmap_set_bit (bb_seen, bb->index);
  92 }
  93
  94 static inline bool
  95 bb_seen_p (basic_block bb)
  96 {
  97   return bitmap_bit_p (bb_seen, bb->index);
  98 }
  99
 100 /* Return true if we should ignore the basic block for purposes of tracing.  */
 101 static bool
 102 ignore_bb_p (const_basic_block bb)
 103 {
 104   gimple g;
 105
 106   if (bb->index < NUM_FIXED_BLOCKS)
 107     return true;
 108   if (optimize_bb_for_size_p (bb))
 109     return true;
 110
 111   /* A transaction is a single entry multiple exit region.  It must be
 112      duplicated in its entirety or not at all.  */
 113   g = last_stmt (CONST_CAST_BB (bb));
 114   if (g && gimple_code (g) == GIMPLE_TRANSACTION)
 115     return true;
 116
 117   return false;
 118 }
 119
 120 /* Return number of instructions in the block.  */
 121
 122 static int
 123 count_insns (basic_block bb)
 124 {
 125   gimple_stmt_iterator gsi;
 126   gimple stmt;
 127   int n = 0;
 128
 129   for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
 130     {
 131       stmt = gsi_stmt (gsi);
 132       n += estimate_num_insns (stmt, &eni_size_weights);
 133     }
 134   return n;
 135 }
 136
 137 /* Return true if E1 is more frequent than E2.  */
 138 static bool
 139 better_p (const_edge e1, const_edge e2)
 140 {
 141   if (e1->count != e2->count)
 142     return e1->count > e2->count;
 143   if (e1->src->frequency * e1->probability !=
 144       e2->src->frequency * e2->probability)
 145     return (e1->src->frequency * e1->probability
 146             > e2->src->frequency * e2->probability);
 147   /* This is needed to avoid changes in the decision after
 148      CFG is modified.  */
 149   if (e1->src != e2->src)
 150     return e1->src->index > e2->src->index;
 151   return e1->dest->index > e2->dest->index;
 152 }
 153
 154 /* Return most frequent successor of basic block BB.  */
 155
 156 static edge
 157 find_best_successor (basic_block bb)
 158 {
 159   edge e;
 160   edge best = NULL;
 161   edge_iterator ei;
 162
 163   FOR_EACH_EDGE (e, ei, bb->succs)
 164     if (!best || better_p (e, best))
 165       best = e;
 166   if (!best || ignore_bb_p (best->dest))
 167     return NULL;
 168   if (best->probability <= probability_cutoff)
 169     return NULL;
 170   return best;
 171 }
 172
 173 /* Return most frequent predecessor of basic block BB.  */
 174
 175 static edge
 176 find_best_predecessor (basic_block bb)
 177 {
 178   edge e;
 179   edge best = NULL;
 180   edge_iterator ei;
 181
 182   FOR_EACH_EDGE (e, ei, bb->preds)
 183     if (!best || better_p (e, best))
 184       best = e;
 185   if (!best || ignore_bb_p (best->src))
 186     return NULL;
 187   if (EDGE_FREQUENCY (best) * REG_BR_PROB_BASE
 188       < bb->frequency * branch_ratio_cutoff)
 189     return NULL;
 190   return best;
 191 }
 192
 193 /* Find the trace using bb and record it in the TRACE array.
 194    Return number of basic blocks recorded.  */
 195
 196 static int
 197 find_trace (basic_block bb, basic_block *trace)
 198 {
 199   int i = 0;
 200   edge e;
 201
 202   if (dump_file)
 203     fprintf (dump_file, "Trace seed %i [%i]", bb->index, bb->frequency);
 204
 205   while ((e = find_best_predecessor (bb)) != NULL)
 206     {
 207       basic_block bb2 = e->src;
 208       if (bb_seen_p (bb2) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 209           || find_best_successor (bb2) != e)
 210         break;
 211       if (dump_file)
 212         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 213       bb = bb2;
 214     }
 215   if (dump_file)
 216     fprintf (dump_file, " forward %i [%i]", bb->index, bb->frequency);
 217   trace[i++] = bb;
 218
 219   /* Follow the trace in forward direction.  */
 220   while ((e = find_best_successor (bb)) != NULL)
 221     {
 222       bb = e->dest;
 223       if (bb_seen_p (bb) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
 224           || find_best_predecessor (bb) != e)
 225         break;
 226       if (dump_file)
 227         fprintf (dump_file, ",%i [%i]", bb->index, bb->frequency);
 228       trace[i++] = bb;
 229     }
 230   if (dump_file)
 231     fprintf (dump_file, "\n");
 232   return i;
 233 }
 234
 235 /* Look for basic blocks in frequency order, construct traces and tail duplicate
 236    if profitable.  */
 237
 238 static bool
 239 tail_duplicate (void)
 240 {
 241   auto_vec<fibonacci_node<long, basic_block_def>*> blocks;
 242   blocks.safe_grow_cleared (last_basic_block_for_fn (cfun));
 243
 244   basic_block *trace = XNEWVEC (basic_block, n_basic_blocks_for_fn (cfun));
 245   int *counts = XNEWVEC (int, last_basic_block_for_fn (cfun));
 246   int ninsns = 0, nduplicated = 0;
 247   gcov_type weighted_insns = 0, traced_insns = 0;
 248   fibonacci_heap<long, basic_block_def> heap (LONG_MIN);
 249   gcov_type cover_insns;
 250   int max_dup_insns;
 251   basic_block bb;
 252   bool changed = false;
 253
 254   /* Create an oversized sbitmap to reduce the chance that we need to
 255      resize it.  */
 256   bb_seen = sbitmap_alloc (last_basic_block_for_fn (cfun) * 2);
 257   bitmap_clear (bb_seen);
 258   initialize_original_copy_tables ();
 259
 260   if (profile_info && flag_branch_probabilities)
 261     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
 262   else
 263     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
 264   probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
 265
 266   branch_ratio_cutoff =
 267     (REG_BR_PROB_BASE / 100 * PARAM_VALUE (TRACER_MIN_BRANCH_RATIO));
 268
 269   FOR_EACH_BB_FN (bb, cfun)
 270     {
 271       int n = count_insns (bb);
 272       if (!ignore_bb_p (bb))
 273         blocks[bb->index] = heap.insert (-bb->frequency, bb);
 274
 275       counts [bb->index] = n;
 276       ninsns += n;
 277       weighted_insns += n * bb->frequency;
 278     }
 279
 280   if (profile_info && flag_branch_probabilities)
 281     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE_FEEDBACK);
 282   else
 283     cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE);
 284   cover_insns = (weighted_insns * cover_insns + 50) / 100;
 285   max_dup_insns = (ninsns * PARAM_VALUE (TRACER_MAX_CODE_GROWTH) + 50) / 100;
 286
 287   while (traced_insns < cover_insns && nduplicated < max_dup_insns
 288          && !heap.empty ())
 289     {
 290       basic_block bb = heap.extract_min ();
 291       int n, pos;
 292
 293       if (!bb)
 294         break;
 295
 296       blocks[bb->index] = NULL;
 297
 298       if (ignore_bb_p (bb))
 299         continue;
 300       gcc_assert (!bb_seen_p (bb));
 301
 302       n = find_trace (bb, trace);
 303
 304       bb = trace[0];
 305       traced_insns += bb->frequency * counts [bb->index];
 306       if (blocks[bb->index])
 307         {
 308           heap.delete_node (blocks[bb->index]);
 309           blocks[bb->index] = NULL;
 310         }
 311
 312       for (pos = 1; pos < n; pos++)
 313         {
 314           basic_block bb2 = trace[pos];
 315
 316           if (blocks[bb2->index])
 317             {
 318               heap.delete_node (blocks[bb2->index]);
 319               blocks[bb2->index] = NULL;
 320             }
 321           traced_insns += bb2->frequency * counts [bb2->index];
 322           if (EDGE_COUNT (bb2->preds) > 1
 323               && can_duplicate_block_p (bb2)
 324               /* We have the tendency to duplicate the loop header
 325                  of all do { } while loops.  Do not do that - it is
 326                  not profitable and it might create a loop with multiple
 327                  entries or at least rotate the loop.  */
 328               && bb2->loop_father->header != bb2)
 329             {
 330               edge e;
 331               basic_block copy;
 332
 333               nduplicated += counts [bb2->index];
 334
 335               e = find_edge (bb, bb2);
 336
 337               copy = duplicate_block (bb2, e, bb);
 338               flush_pending_stmts (e);
 339
 340               add_phi_args_after_copy (&copy, 1, NULL);
 341
 342               /* Reconsider the original copy of block we've duplicated.
 343                  Removing the most common predecessor may make it to be
 344                  head.  */
 345               blocks[bb2->index] = heap.insert (-bb2->frequency, bb2);
 346
 347               if (dump_file)
 348                 fprintf (dump_file, "Duplicated %i as %i [%i]\n",
 349                          bb2->index, copy->index, copy->frequency);
 350
 351               bb2 = copy;
 352               changed = true;
 353             }
 354           mark_bb_seen (bb2);
 355           bb = bb2;
 356           /* In case the trace became infrequent, stop duplicating.  */
 357           if (ignore_bb_p (bb))
 358             break;
 359         }
 360       if (dump_file)
 361         fprintf (dump_file, " covered now %.1f\n\n",
 362                  traced_insns * 100.0 / weighted_insns);
 363     }
 364   if (dump_file)
 365     fprintf (dump_file, "Duplicated %i insns (%i%%)\n", nduplicated,
 366              nduplicated * 100 / ninsns);
 367
 368   free_original_copy_tables ();
 369   sbitmap_free (bb_seen);
 370   free (trace);
 371   free (counts);
 372
 373   return changed;
 374 }
 375 \f
 376 namespace {
 377
 378 const pass_data pass_data_tracer =
 379 {
 380   GIMPLE_PASS, /* type */
 381   "tracer", /* name */
 382   OPTGROUP_NONE, /* optinfo_flags */
 383   TV_TRACER, /* tv_id */
 384   0, /* properties_required */
 385   0, /* properties_provided */
 386   0, /* properties_destroyed */
 387   0, /* todo_flags_start */
 388   TODO_update_ssa, /* todo_flags_finish */
 389 };
 390
 391 class pass_tracer : public gimple_opt_pass
 392 {
 393 public:
 394   pass_tracer (gcc::context *ctxt)
 395     : gimple_opt_pass (pass_data_tracer, ctxt)
 396   {}
 397
 398   /* opt_pass methods: */
 399   virtual bool gate (function *)
 400     {
 401       return (optimize > 0 && flag_tracer && flag_reorder_blocks);
 402     }
 403
 404   virtual unsigned int execute (function *);
 405
 406 }; // class pass_tracer
 407
 408 unsigned int
 409 pass_tracer::execute (function *fun)
 410 {
 411   bool changed;
 412
 413   if (n_basic_blocks_for_fn (fun) <= NUM_FIXED_BLOCKS + 1)
 414     return 0;
 415
 416   mark_dfs_back_edges ();
 417   if (dump_file)
 418     brief_dump_cfg (dump_file, dump_flags);
 419
 420   /* Trace formation is done on the fly inside tail_duplicate */
 421   changed = tail_duplicate ();
 422   if (changed)
 423     {
 424       free_dominance_info (CDI_DOMINATORS);
 425       /* If we changed the CFG schedule loops for fixup by cleanup_cfg.  */
 426       loops_state_set (LOOPS_NEED_FIXUP);
 427     }
 428
 429   if (dump_file)
 430     brief_dump_cfg (dump_file, dump_flags);
 431
 432   return changed ? TODO_cleanup_cfg : 0;
 433 }
 434 } // anon namespace
 435
 436 gimple_opt_pass *
 437 make_pass_tracer (gcc::context *ctxt)
 438 {
 439   return new pass_tracer (ctxt);
 440 }