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