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) 2001-2014 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 "tree-ssa-alias.h"
  50 #include "internal-fn.h"
  51 #include "gimple-expr.h"
  52 #include "is-a.h"
  53 #include "gimple.h"
  54 #include "gimple-iterator.h"
  55 #include "tree-cfg.h"
  56 #include "tree-ssa.h"
  57 #include "tree-inline.h"
  58 #include "cfgloop.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   fibnode_t *blocks = XCNEWVEC (fibnode_t, last_basic_block_for_fn (cfun));
 234   basic_block *trace = XNEWVEC (basic_block, n_basic_blocks_for_fn (cfun));
 235   int *counts = XNEWVEC (int, last_basic_block_for_fn (cfun));
 236   int ninsns = 0, nduplicated = 0;
 237   gcov_type weighted_insns = 0, traced_insns = 0;
 238   fibheap_t heap = fibheap_new ();
 239   gcov_type cover_insns;
 240   int max_dup_insns;
 241   basic_block bb;
 242   bool changed = false;
 243
 244   /* Create an oversized sbitmap to reduce the chance that we need to
 245      resize it.  */
 246   bb_seen = sbitmap_alloc (last_basic_block_for_fn (cfun) * 2);
 247   bitmap_clear (bb_seen);
 248   initialize_original_copy_tables ();
 249
 250   if (profile_info && flag_branch_probabilities)
 251     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
 252   else
 253     probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
 254   probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
 255
 256   branch_ratio_cutoff =
 257     (REG_BR_PROB_BASE / 100 * PARAM_VALUE (TRACER_MIN_BRANCH_RATIO));
 258
 259   FOR_EACH_BB_FN (bb, cfun)
 260     {
 261       int n = count_insns (bb);
 262       if (!ignore_bb_p (bb))
 263         blocks[bb->index] = fibheap_insert (heap, -bb->frequency,
 264                                             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          && !fibheap_empty (heap))
 280     {
 281       basic_block bb = (basic_block) fibheap_extract_min (heap);
 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           fibheap_delete_node (heap, 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               fibheap_delete_node (heap, 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] =
 337                 fibheap_insert (heap, -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 (blocks);
 363   free (trace);
 364   free (counts);
 365   fibheap_delete (heap);
 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 }