gcc/tree-vectorizer.c

   1 /* Vectorizer
   2    Copyright (C) 2003-2013 Free Software Foundation, Inc.
   3    Contributed by Dorit Naishlos <dorit@il.ibm.com>
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 3, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING3.  If not see
  19 <http://www.gnu.org/licenses/>.  */
  20
  21 /* Loop and basic block vectorizer.
  22
  23   This file contains drivers for the three vectorizers:
  24   (1) loop vectorizer (inter-iteration parallelism),
  25   (2) loop-aware SLP (intra-iteration parallelism) (invoked by the loop
  26       vectorizer)
  27   (3) BB vectorizer (out-of-loops), aka SLP
  28
  29   The rest of the vectorizer's code is organized as follows:
  30   - tree-vect-loop.c - loop specific parts such as reductions, etc. These are
  31     used by drivers (1) and (2).
  32   - tree-vect-loop-manip.c - vectorizer's loop control-flow utilities, used by
  33     drivers (1) and (2).
  34   - tree-vect-slp.c - BB vectorization specific analysis and transformation,
  35     used by drivers (2) and (3).
  36   - tree-vect-stmts.c - statements analysis and transformation (used by all).
  37   - tree-vect-data-refs.c - vectorizer specific data-refs analysis and
  38     manipulations (used by all).
  39   - tree-vect-patterns.c - vectorizable code patterns detector (used by all)
  40
  41   Here's a poor attempt at illustrating that:
  42
  43      tree-vectorizer.c:
  44      loop_vect()  loop_aware_slp()  slp_vect()
  45           |        /           \          /
  46           |       /             \        /
  47           tree-vect-loop.c  tree-vect-slp.c
  48                 | \      \  /      /   |
  49                 |  \      \/      /    |
  50                 |   \     /\     /     |
  51                 |    \   /  \   /      |
  52          tree-vect-stmts.c  tree-vect-data-refs.c
  53                        \      /
  54                     tree-vect-patterns.c
  55 */
  56
  57 #include "config.h"
  58 #include "system.h"
  59 #include "coretypes.h"
  60 #include "dumpfile.h"
  61 #include "tm.h"
  62 #include "ggc.h"
  63 #include "tree.h"
  64 #include "tree-pretty-print.h"
  65 #include "tree-flow.h"
  66 #include "cfgloop.h"
  67 #include "tree-vectorizer.h"
  68 #include "tree-pass.h"
  69 #include "hash-table.h"
  70 #include "tree-ssa-propagate.h"
  71 #include "dbgcnt.h"
  72
  73 /* Loop or bb location.  */
  74 LOC vect_location;
  75
  76 /* Vector mapping GIMPLE stmt to stmt_vec_info. */
  77 vec<vec_void_p> stmt_vec_info_vec;
  78 \f
  79 /* For mapping simduid to vectorization factor.  */
  80
  81 struct simduid_to_vf : typed_free_remove<simduid_to_vf>
  82 {
  83   unsigned int simduid;
  84   int vf;
  85
  86   /* hash_table support.  */
  87   typedef simduid_to_vf value_type;
  88   typedef simduid_to_vf compare_type;
  89   static inline hashval_t hash (const value_type *);
  90   static inline int equal (const value_type *, const compare_type *);
  91 };
  92
  93 inline hashval_t
  94 simduid_to_vf::hash (const value_type *p)
  95 {
  96   return p->simduid;
  97 }
  98
  99 inline int
 100 simduid_to_vf::equal (const value_type *p1, const value_type *p2)
 101 {
 102   return p1->simduid == p2->simduid;
 103 }
 104
 105 /* This hash maps the OMP simd array to the corresponding simduid used
 106    to index into it.  Like thus,
 107
 108         _7 = GOMP_SIMD_LANE (simduid.0)
 109         ...
 110         ...
 111         D.1737[_7] = stuff;
 112
 113
 114    This hash maps from the simduid.0 to OMP simd array (D.1737[]).  */
 115
 116 struct simd_array_to_simduid : typed_free_remove<simd_array_to_simduid>
 117 {
 118   tree decl;
 119   unsigned int simduid;
 120
 121   /* hash_table support.  */
 122   typedef simd_array_to_simduid value_type;
 123   typedef simd_array_to_simduid compare_type;
 124   static inline hashval_t hash (const value_type *);
 125   static inline int equal (const value_type *, const compare_type *);
 126 };
 127
 128 inline hashval_t
 129 simd_array_to_simduid::hash (const value_type *p)
 130 {
 131   return DECL_UID (p->decl);
 132 }
 133
 134 inline int
 135 simd_array_to_simduid::equal (const value_type *p1, const value_type *p2)
 136 {
 137   return p1->decl == p2->decl;
 138 }
 139
 140 /* Fold IFN_GOMP_SIMD_LANE, IFN_GOMP_SIMD_VF and IFN_GOMP_SIMD_LAST_LANE
 141    into their corresponding constants.  */
 142
 143 static void
 144 adjust_simduid_builtins (hash_table <simduid_to_vf> &htab)
 145 {
 146   basic_block bb;
 147
 148   FOR_EACH_BB (bb)
 149     {
 150       gimple_stmt_iterator i;
 151
 152       for (i = gsi_start_bb (bb); !gsi_end_p (i); gsi_next (&i))
 153         {
 154           unsigned int vf = 1;
 155           enum internal_fn ifn;
 156           gimple stmt = gsi_stmt (i);
 157           tree t;
 158           if (!is_gimple_call (stmt)
 159               || !gimple_call_internal_p (stmt))
 160             continue;
 161           ifn = gimple_call_internal_fn (stmt);
 162           switch (ifn)
 163             {
 164             case IFN_GOMP_SIMD_LANE:
 165             case IFN_GOMP_SIMD_VF:
 166             case IFN_GOMP_SIMD_LAST_LANE:
 167               break;
 168             default:
 169               continue;
 170             }
 171           tree arg = gimple_call_arg (stmt, 0);
 172           gcc_assert (arg != NULL_TREE);
 173           gcc_assert (TREE_CODE (arg) == SSA_NAME);
 174           simduid_to_vf *p = NULL, data;
 175           data.simduid = DECL_UID (SSA_NAME_VAR (arg));
 176           if (htab.is_created ())
 177             p = htab.find (&data);
 178           if (p)
 179             vf = p->vf;
 180           switch (ifn)
 181             {
 182             case IFN_GOMP_SIMD_VF:
 183               t = build_int_cst (unsigned_type_node, vf);
 184               break;
 185             case IFN_GOMP_SIMD_LANE:
 186               t = build_int_cst (unsigned_type_node, 0);
 187               break;
 188             case IFN_GOMP_SIMD_LAST_LANE:
 189               t = gimple_call_arg (stmt, 1);
 190               break;
 191             default:
 192               gcc_unreachable ();
 193             }
 194           update_call_from_tree (&i, t);
 195         }
 196     }
 197 }
 198
 199 /* Helper structure for note_simd_array_uses.  */
 200
 201 struct note_simd_array_uses_struct
 202 {
 203   hash_table <simd_array_to_simduid> *htab;
 204   unsigned int simduid;
 205 };
 206
 207 /* Callback for note_simd_array_uses, called through walk_gimple_op.  */
 208
 209 static tree
 210 note_simd_array_uses_cb (tree *tp, int *walk_subtrees, void *data)
 211 {
 212   struct walk_stmt_info *wi = (struct walk_stmt_info *) data;
 213   struct note_simd_array_uses_struct *ns
 214     = (struct note_simd_array_uses_struct *) wi->info;
 215
 216   if (TYPE_P (*tp))
 217     *walk_subtrees = 0;
 218   else if (VAR_P (*tp)
 219            && lookup_attribute ("omp simd array", DECL_ATTRIBUTES (*tp))
 220            && DECL_CONTEXT (*tp) == current_function_decl)
 221     {
 222       simd_array_to_simduid data;
 223       if (!ns->htab->is_created ())
 224         ns->htab->create (15);
 225       data.decl = *tp;
 226       data.simduid = ns->simduid;
 227       simd_array_to_simduid **slot = ns->htab->find_slot (&data, INSERT);
 228       if (*slot == NULL)
 229         {
 230           simd_array_to_simduid *p = XNEW (simd_array_to_simduid);
 231           *p = data;
 232           *slot = p;
 233         }
 234       else if ((*slot)->simduid != ns->simduid)
 235         (*slot)->simduid = -1U;
 236       *walk_subtrees = 0;
 237     }
 238   return NULL_TREE;
 239 }
 240
 241 /* Find "omp simd array" temporaries and map them to corresponding
 242    simduid.  */
 243
 244 static void
 245 note_simd_array_uses (hash_table <simd_array_to_simduid> *htab)
 246 {
 247   basic_block bb;
 248   gimple_stmt_iterator gsi;
 249   struct walk_stmt_info wi;
 250   struct note_simd_array_uses_struct ns;
 251
 252   memset (&wi, 0, sizeof (wi));
 253   wi.info = &ns;
 254   ns.htab = htab;
 255
 256   FOR_EACH_BB (bb)
 257     for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
 258       {
 259         gimple stmt = gsi_stmt (gsi);
 260         if (!is_gimple_call (stmt) || !gimple_call_internal_p (stmt))
 261           continue;
 262         switch (gimple_call_internal_fn (stmt))
 263           {
 264           case IFN_GOMP_SIMD_LANE:
 265           case IFN_GOMP_SIMD_VF:
 266           case IFN_GOMP_SIMD_LAST_LANE:
 267             break;
 268           default:
 269             continue;
 270           }
 271         tree lhs = gimple_call_lhs (stmt);
 272         if (lhs == NULL_TREE)
 273           continue;
 274         imm_use_iterator use_iter;
 275         gimple use_stmt;
 276         ns.simduid = DECL_UID (SSA_NAME_VAR (gimple_call_arg (stmt, 0)));
 277         FOR_EACH_IMM_USE_STMT (use_stmt, use_iter, lhs)
 278           if (!is_gimple_debug (use_stmt))
 279             walk_gimple_op (use_stmt, note_simd_array_uses_cb, &wi);
 280       }
 281 }
 282 \f
 283 /* A helper function to free data refs.  */
 284
 285 void
 286 vect_destroy_datarefs (loop_vec_info loop_vinfo, bb_vec_info bb_vinfo)
 287 {
 288   vec<data_reference_p> datarefs;
 289   struct data_reference *dr;
 290   unsigned int i;
 291
 292  if (loop_vinfo)
 293     datarefs = LOOP_VINFO_DATAREFS (loop_vinfo);
 294   else
 295     datarefs = BB_VINFO_DATAREFS (bb_vinfo);
 296
 297   FOR_EACH_VEC_ELT (datarefs, i, dr)
 298     if (dr->aux)
 299       {
 300         free (dr->aux);
 301         dr->aux = NULL;
 302       }
 303
 304   free_data_refs (datarefs);
 305 }
 306
 307
 308 /* Function vectorize_loops.
 309
 310    Entry point to loop vectorization phase.  */
 311
 312 unsigned
 313 vectorize_loops (void)
 314 {
 315   unsigned int i;
 316   unsigned int num_vectorized_loops = 0;
 317   unsigned int vect_loops_num;
 318   loop_iterator li;
 319   struct loop *loop;
 320   hash_table <simduid_to_vf> simduid_to_vf_htab;
 321   hash_table <simd_array_to_simduid> simd_array_to_simduid_htab;
 322
 323   vect_loops_num = number_of_loops (cfun);
 324
 325   /* Bail out if there are no loops.  */
 326   if (vect_loops_num <= 1)
 327     {
 328       if (cfun->has_simduid_loops)
 329         adjust_simduid_builtins (simduid_to_vf_htab);
 330       return 0;
 331     }
 332
 333   if (cfun->has_simduid_loops)
 334     note_simd_array_uses (&simd_array_to_simduid_htab);
 335
 336   init_stmt_vec_info_vec ();
 337
 338   /*  ----------- Analyze loops. -----------  */
 339
 340   /* If some loop was duplicated, it gets bigger number
 341      than all previously defined loops.  This fact allows us to run
 342      only over initial loops skipping newly generated ones.  */
 343   FOR_EACH_LOOP (li, loop, 0)
 344     if ((flag_tree_vectorize && optimize_loop_nest_for_speed_p (loop))
 345         || loop->force_vect)
 346       {
 347         loop_vec_info loop_vinfo;
 348         vect_location = find_loop_location (loop);
 349         if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOC
 350             && dump_enabled_p ())
 351           dump_printf (MSG_NOTE, "\nAnalyzing loop at %s:%d\n",
 352                        LOC_FILE (vect_location), LOC_LINE (vect_location));
 353
 354         loop_vinfo = vect_analyze_loop (loop);
 355         loop->aux = loop_vinfo;
 356
 357         if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
 358           continue;
 359
 360         if (!dbg_cnt (vect_loop))
 361           break;
 362
 363         if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOC
 364             && dump_enabled_p ())
 365           dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, vect_location,
 366                            "loop vectorized\n");
 367         vect_transform_loop (loop_vinfo);
 368         num_vectorized_loops++;
 369         /* Now that the loop has been vectorized, allow it to be unrolled
 370            etc.  */
 371         loop->force_vect = false;
 372
 373         if (loop->simduid)
 374           {
 375             simduid_to_vf *simduid_to_vf_data = XNEW (simduid_to_vf);
 376             if (!simduid_to_vf_htab.is_created ())
 377               simduid_to_vf_htab.create (15);
 378             simduid_to_vf_data->simduid = DECL_UID (loop->simduid);
 379             simduid_to_vf_data->vf = loop_vinfo->vectorization_factor;
 380             *simduid_to_vf_htab.find_slot (simduid_to_vf_data, INSERT)
 381               = simduid_to_vf_data;
 382           }
 383       }
 384
 385   vect_location = UNKNOWN_LOC;
 386
 387   statistics_counter_event (cfun, "Vectorized loops", num_vectorized_loops);
 388   if (dump_enabled_p ()
 389       || (num_vectorized_loops > 0 && dump_enabled_p ()))
 390     dump_printf_loc (MSG_NOTE, vect_location,
 391                      "vectorized %u loops in function.\n",
 392                      num_vectorized_loops);
 393
 394   /*  ----------- Finalize. -----------  */
 395
 396   for (i = 1; i < vect_loops_num; i++)
 397     {
 398       loop_vec_info loop_vinfo;
 399
 400       loop = get_loop (cfun, i);
 401       if (!loop)
 402         continue;
 403       loop_vinfo = (loop_vec_info) loop->aux;
 404       destroy_loop_vec_info (loop_vinfo, true);
 405       loop->aux = NULL;
 406     }
 407
 408   free_stmt_vec_info_vec ();
 409
 410   /* Fold IFN_GOMP_SIMD_{VF,LANE,LAST_LANE} builtins.  */
 411   if (cfun->has_simduid_loops)
 412     adjust_simduid_builtins (simduid_to_vf_htab);
 413
 414   /* Shrink any "omp array simd" temporary arrays to the
 415      actual vectorization factors.  */
 416   if (simd_array_to_simduid_htab.is_created ())
 417     {
 418       for (hash_table <simd_array_to_simduid>::iterator iter
 419            = simd_array_to_simduid_htab.begin ();
 420            iter != simd_array_to_simduid_htab.end (); ++iter)
 421         if ((*iter).simduid != -1U)
 422           {
 423             tree decl = (*iter).decl;
 424             int vf = 1;
 425             if (simduid_to_vf_htab.is_created ())
 426               {
 427                 simduid_to_vf *p = NULL, data;
 428                 data.simduid = (*iter).simduid;
 429                 p = simduid_to_vf_htab.find (&data);
 430                 if (p)
 431                   vf = p->vf;
 432               }
 433             tree atype
 434               = build_array_type_nelts (TREE_TYPE (TREE_TYPE (decl)), vf);
 435             TREE_TYPE (decl) = atype;
 436             relayout_decl (decl);
 437           }
 438
 439       simd_array_to_simduid_htab.dispose ();
 440     }
 441   if (simduid_to_vf_htab.is_created ())
 442     simduid_to_vf_htab.dispose ();
 443
 444   if (num_vectorized_loops > 0)
 445     {
 446       /* If we vectorized any loop only virtual SSA form needs to be updated.
 447          ???  Also while we try hard to update loop-closed SSA form we fail
 448          to properly do this in some corner-cases (see PR56286).  */
 449       rewrite_into_loop_closed_ssa (NULL, TODO_update_ssa_only_virtuals);
 450       return TODO_cleanup_cfg;
 451     }
 452
 453   return 0;
 454 }
 455
 456
 457 /*  Entry point to basic block SLP phase.  */
 458
 459 static unsigned int
 460 execute_vect_slp (void)
 461 {
 462   basic_block bb;
 463
 464   init_stmt_vec_info_vec ();
 465
 466   FOR_EACH_BB (bb)
 467     {
 468       vect_location = find_bb_location (bb);
 469
 470       if (vect_slp_analyze_bb (bb))
 471         {
 472           if (!dbg_cnt (vect_slp))
 473             break;
 474
 475           vect_slp_transform_bb (bb);
 476           if (dump_enabled_p ())
 477             dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, vect_location,
 478                              "basic block vectorized\n");
 479         }
 480     }
 481
 482   free_stmt_vec_info_vec ();
 483   return 0;
 484 }
 485
 486 static bool
 487 gate_vect_slp (void)
 488 {
 489   /* Apply SLP either if the vectorizer is on and the user didn't specify
 490      whether to run SLP or not, or if the SLP flag was set by the user.  */
 491   return ((flag_tree_vectorize != 0 && flag_tree_slp_vectorize != 0)
 492           || flag_tree_slp_vectorize == 1);
 493 }
 494
 495 namespace {
 496
 497 const pass_data pass_data_slp_vectorize =
 498 {
 499   GIMPLE_PASS, /* type */
 500   "slp", /* name */
 501   OPTGROUP_LOOP | OPTGROUP_VEC, /* optinfo_flags */
 502   true, /* has_gate */
 503   true, /* has_execute */
 504   TV_TREE_SLP_VECTORIZATION, /* tv_id */
 505   ( PROP_ssa | PROP_cfg ), /* properties_required */
 506   0, /* properties_provided */
 507   0, /* properties_destroyed */
 508   0, /* todo_flags_start */
 509   ( TODO_verify_ssa | TODO_update_ssa
 510     | TODO_verify_stmts ), /* todo_flags_finish */
 511 };
 512
 513 class pass_slp_vectorize : public gimple_opt_pass
 514 {
 515 public:
 516   pass_slp_vectorize(gcc::context *ctxt)
 517     : gimple_opt_pass(pass_data_slp_vectorize, ctxt)
 518   {}
 519
 520   /* opt_pass methods: */
 521   bool gate () { return gate_vect_slp (); }
 522   unsigned int execute () { return execute_vect_slp (); }
 523
 524 }; // class pass_slp_vectorize
 525
 526 } // anon namespace
 527
 528 gimple_opt_pass *
 529 make_pass_slp_vectorize (gcc::context *ctxt)
 530 {
 531   return new pass_slp_vectorize (ctxt);
 532 }
 533
 534
 535 /* Increase alignment of global arrays to improve vectorization potential.
 536    TODO:
 537    - Consider also structs that have an array field.
 538    - Use ipa analysis to prune arrays that can't be vectorized?
 539      This should involve global alignment analysis and in the future also
 540      array padding.  */
 541
 542 static unsigned int
 543 increase_alignment (void)
 544 {
 545   struct varpool_node *vnode;
 546
 547   vect_location = UNKNOWN_LOC;
 548
 549   /* Increase the alignment of all global arrays for vectorization.  */
 550   FOR_EACH_DEFINED_VARIABLE (vnode)
 551     {
 552       tree vectype, decl = vnode->symbol.decl;
 553       tree t;
 554       unsigned int alignment;
 555
 556       t = TREE_TYPE(decl);
 557       if (TREE_CODE (t) != ARRAY_TYPE)
 558         continue;
 559       vectype = get_vectype_for_scalar_type (strip_array_types (t));
 560       if (!vectype)
 561         continue;
 562       alignment = TYPE_ALIGN (vectype);
 563       if (DECL_ALIGN (decl) >= alignment)
 564         continue;
 565
 566       if (vect_can_force_dr_alignment_p (decl, alignment))
 567         {
 568           DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
 569           DECL_USER_ALIGN (decl) = 1;
 570           dump_printf (MSG_NOTE, "Increasing alignment of decl: ");
 571           dump_generic_expr (MSG_NOTE, TDF_SLIM, decl);
 572           dump_printf (MSG_NOTE, "\n");
 573         }
 574     }
 575   return 0;
 576 }
 577
 578
 579 static bool
 580 gate_increase_alignment (void)
 581 {
 582   return flag_section_anchors && flag_tree_vectorize;
 583 }
 584
 585
 586 namespace {
 587
 588 const pass_data pass_data_ipa_increase_alignment =
 589 {
 590   SIMPLE_IPA_PASS, /* type */
 591   "increase_alignment", /* name */
 592   OPTGROUP_LOOP | OPTGROUP_VEC, /* optinfo_flags */
 593   true, /* has_gate */
 594   true, /* has_execute */
 595   TV_IPA_OPT, /* tv_id */
 596   0, /* properties_required */
 597   0, /* properties_provided */
 598   0, /* properties_destroyed */
 599   0, /* todo_flags_start */
 600   0, /* todo_flags_finish */
 601 };
 602
 603 class pass_ipa_increase_alignment : public simple_ipa_opt_pass
 604 {
 605 public:
 606   pass_ipa_increase_alignment(gcc::context *ctxt)
 607     : simple_ipa_opt_pass(pass_data_ipa_increase_alignment, ctxt)
 608   {}
 609
 610   /* opt_pass methods: */
 611   bool gate () { return gate_increase_alignment (); }
 612   unsigned int execute () { return increase_alignment (); }
 613
 614 }; // class pass_ipa_increase_alignment
 615
 616 } // anon namespace
 617
 618 simple_ipa_opt_pass *
 619 make_pass_ipa_increase_alignment (gcc::context *ctxt)
 620 {
 621   return new pass_ipa_increase_alignment (ctxt);
 622 }