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