gcc/tree-vectorizer.h

   1 /* Vectorizer
   2    Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012
   3    Free Software Foundation, Inc.
   4    Contributed by Dorit Naishlos <dorit@il.ibm.com>
   5
   6 This file is part of GCC.
   7
   8 GCC is free software; you can redistribute it and/or modify it under
   9 the terms of the GNU General Public License as published by the Free
  10 Software Foundation; either version 3, or (at your option) any later
  11 version.
  12
  13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  15 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  16 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 #ifndef GCC_TREE_VECTORIZER_H
  23 #define GCC_TREE_VECTORIZER_H
  24
  25 #include "tree-data-ref.h"
  26 #include "target.h"
  27
  28 typedef source_location LOC;
  29 #define UNKNOWN_LOC UNKNOWN_LOCATION
  30 #define EXPR_LOC(e) EXPR_LOCATION(e)
  31 #define LOC_FILE(l) LOCATION_FILE (l)
  32 #define LOC_LINE(l) LOCATION_LINE (l)
  33
  34 /* Used for naming of new temporaries.  */
  35 enum vect_var_kind {
  36   vect_simple_var,
  37   vect_pointer_var,
  38   vect_scalar_var
  39 };
  40
  41 /* Defines type of operation.  */
  42 enum operation_type {
  43   unary_op = 1,
  44   binary_op,
  45   ternary_op
  46 };
  47
  48 /* Define type of available alignment support.  */
  49 enum dr_alignment_support {
  50   dr_unaligned_unsupported,
  51   dr_unaligned_supported,
  52   dr_explicit_realign,
  53   dr_explicit_realign_optimized,
  54   dr_aligned
  55 };
  56
  57 /* Define type of def-use cross-iteration cycle.  */
  58 enum vect_def_type {
  59   vect_uninitialized_def = 0,
  60   vect_constant_def = 1,
  61   vect_external_def,
  62   vect_internal_def,
  63   vect_induction_def,
  64   vect_reduction_def,
  65   vect_double_reduction_def,
  66   vect_nested_cycle,
  67   vect_unknown_def_type
  68 };
  69
  70 #define VECTORIZABLE_CYCLE_DEF(D) (((D) == vect_reduction_def)           \
  71                                    || ((D) == vect_double_reduction_def) \
  72                                    || ((D) == vect_nested_cycle))
  73
  74 /* Structure to encapsulate information about a group of like
  75    instructions to be presented to the target cost model.  */
  76 typedef struct _stmt_info_for_cost {
  77   int count;
  78   enum vect_cost_for_stmt kind;
  79   gimple stmt;
  80   int misalign;
  81 } stmt_info_for_cost;
  82
  83 DEF_VEC_O (stmt_info_for_cost);
  84 DEF_VEC_ALLOC_O (stmt_info_for_cost, heap);
  85
  86 typedef VEC(stmt_info_for_cost, heap) *stmt_vector_for_cost;
  87
  88 static inline void
  89 add_stmt_info_to_vec (stmt_vector_for_cost *stmt_cost_vec, int count,
  90                       enum vect_cost_for_stmt kind, gimple stmt, int misalign)
  91 {
  92   stmt_info_for_cost si;
  93   si.count = count;
  94   si.kind = kind;
  95   si.stmt = stmt;
  96   si.misalign = misalign;
  97   VEC_safe_push (stmt_info_for_cost, heap, *stmt_cost_vec, &si);
  98 }
  99
 100 /************************************************************************
 101   SLP
 102  ************************************************************************/
 103 typedef void *slp_void_p;
 104 DEF_VEC_P (slp_void_p);
 105 DEF_VEC_ALLOC_P (slp_void_p, heap);
 106
 107 /* A computation tree of an SLP instance.  Each node corresponds to a group of
 108    stmts to be packed in a SIMD stmt.  */
 109 typedef struct _slp_tree {
 110   /* Nodes that contain def-stmts of this node statements operands.  */
 111   VEC (slp_void_p, heap) *children;
 112   /* A group of scalar stmts to be vectorized together.  */
 113   VEC (gimple, heap) *stmts;
 114   /* Vectorized stmt/s.  */
 115   VEC (gimple, heap) *vec_stmts;
 116   /* Number of vector stmts that are created to replace the group of scalar
 117      stmts. It is calculated during the transformation phase as the number of
 118      scalar elements in one scalar iteration (GROUP_SIZE) multiplied by VF
 119      divided by vector size.  */
 120   unsigned int vec_stmts_size;
 121 } *slp_tree;
 122
 123 DEF_VEC_P(slp_tree);
 124 DEF_VEC_ALLOC_P(slp_tree, heap);
 125
 126 /* SLP instance is a sequence of stmts in a loop that can be packed into
 127    SIMD stmts.  */
 128 typedef struct _slp_instance {
 129   /* The root of SLP tree.  */
 130   slp_tree root;
 131
 132   /* Size of groups of scalar stmts that will be replaced by SIMD stmt/s.  */
 133   unsigned int group_size;
 134
 135   /* The unrolling factor required to vectorized this SLP instance.  */
 136   unsigned int unrolling_factor;
 137
 138   /* Vectorization costs associated with SLP instance.  */
 139   stmt_vector_for_cost body_cost_vec;
 140
 141   /* Loads permutation relatively to the stores, NULL if there is no
 142      permutation.  */
 143   VEC (int, heap) *load_permutation;
 144
 145   /* The group of nodes that contain loads of this SLP instance.  */
 146   VEC (slp_tree, heap) *loads;
 147
 148   /* The first scalar load of the instance. The created vector loads will be
 149      inserted before this statement.  */
 150   gimple first_load;
 151 } *slp_instance;
 152
 153 DEF_VEC_P(slp_instance);
 154 DEF_VEC_ALLOC_P(slp_instance, heap);
 155
 156 /* Access Functions.  */
 157 #define SLP_INSTANCE_TREE(S)                     (S)->root
 158 #define SLP_INSTANCE_GROUP_SIZE(S)               (S)->group_size
 159 #define SLP_INSTANCE_UNROLLING_FACTOR(S)         (S)->unrolling_factor
 160 #define SLP_INSTANCE_BODY_COST_VEC(S)            (S)->body_cost_vec
 161 #define SLP_INSTANCE_LOAD_PERMUTATION(S)         (S)->load_permutation
 162 #define SLP_INSTANCE_LOADS(S)                    (S)->loads
 163 #define SLP_INSTANCE_FIRST_LOAD_STMT(S)          (S)->first_load
 164
 165 #define SLP_TREE_CHILDREN(S)                     (S)->children
 166 #define SLP_TREE_SCALAR_STMTS(S)                 (S)->stmts
 167 #define SLP_TREE_VEC_STMTS(S)                    (S)->vec_stmts
 168 #define SLP_TREE_NUMBER_OF_VEC_STMTS(S)          (S)->vec_stmts_size
 169
 170 /* This structure is used in creation of an SLP tree.  Each instance
 171    corresponds to the same operand in a group of scalar stmts in an SLP
 172    node.  */
 173 typedef struct _slp_oprnd_info
 174 {
 175   /* Def-stmts for the operands.  */
 176   VEC (gimple, heap) *def_stmts;
 177   /* Information about the first statement, its vector def-type, type, the
 178      operand itself in case it's constant, and an indication if it's a pattern
 179      stmt.  */
 180   enum vect_def_type first_dt;
 181   tree first_def_type;
 182   tree first_const_oprnd;
 183   bool first_pattern;
 184 } *slp_oprnd_info;
 185
 186 DEF_VEC_P(slp_oprnd_info);
 187 DEF_VEC_ALLOC_P(slp_oprnd_info, heap);
 188
 189
 190 typedef struct _vect_peel_info
 191 {
 192   int npeel;
 193   struct data_reference *dr;
 194   unsigned int count;
 195 } *vect_peel_info;
 196
 197 typedef struct _vect_peel_extended_info
 198 {
 199   struct _vect_peel_info peel_info;
 200   unsigned int inside_cost;
 201   unsigned int outside_cost;
 202   stmt_vector_for_cost body_cost_vec;
 203 } *vect_peel_extended_info;
 204
 205 /*-----------------------------------------------------------------*/
 206 /* Info on vectorized loops.                                       */
 207 /*-----------------------------------------------------------------*/
 208 typedef struct _loop_vec_info {
 209
 210   /* The loop to which this info struct refers to.  */
 211   struct loop *loop;
 212
 213   /* The loop basic blocks.  */
 214   basic_block *bbs;
 215
 216   /* Number of iterations.  */
 217   tree num_iters;
 218   tree num_iters_unchanged;
 219
 220   /* Minimum number of iterations below which vectorization is expected to
 221      not be profitable (as estimated by the cost model).
 222      -1 indicates that vectorization will not be profitable.
 223      FORNOW: This field is an int. Will be a tree in the future, to represent
 224              values unknown at compile time.  */
 225   int min_profitable_iters;
 226
 227   /* Is the loop vectorizable? */
 228   bool vectorizable;
 229
 230   /* Unrolling factor  */
 231   int vectorization_factor;
 232
 233   /* The loop location in the source.  */
 234   LOC loop_line_number;
 235
 236   /* Unknown DRs according to which loop was peeled.  */
 237   struct data_reference *unaligned_dr;
 238
 239   /* peeling_for_alignment indicates whether peeling for alignment will take
 240      place, and what the peeling factor should be:
 241      peeling_for_alignment = X means:
 242         If X=0: Peeling for alignment will not be applied.
 243         If X>0: Peel first X iterations.
 244         If X=-1: Generate a runtime test to calculate the number of iterations
 245                  to be peeled, using the dataref recorded in the field
 246                  unaligned_dr.  */
 247   int peeling_for_alignment;
 248
 249   /* The mask used to check the alignment of pointers or arrays.  */
 250   int ptr_mask;
 251
 252   /* The loop nest in which the data dependences are computed.  */
 253   VEC (loop_p, heap) *loop_nest;
 254
 255   /* All data references in the loop.  */
 256   VEC (data_reference_p, heap) *datarefs;
 257
 258   /* All data dependences in the loop.  */
 259   VEC (ddr_p, heap) *ddrs;
 260
 261   /* Data Dependence Relations defining address ranges that are candidates
 262      for a run-time aliasing check.  */
 263   VEC (ddr_p, heap) *may_alias_ddrs;
 264
 265   /* Statements in the loop that have data references that are candidates for a
 266      runtime (loop versioning) misalignment check.  */
 267   VEC(gimple,heap) *may_misalign_stmts;
 268
 269   /* All interleaving chains of stores in the loop, represented by the first
 270      stmt in the chain.  */
 271   VEC(gimple, heap) *grouped_stores;
 272
 273   /* All SLP instances in the loop. This is a subset of the set of GROUP_STORES
 274      of the loop.  */
 275   VEC(slp_instance, heap) *slp_instances;
 276
 277   /* The unrolling factor needed to SLP the loop. In case of that pure SLP is
 278      applied to the loop, i.e., no unrolling is needed, this is 1.  */
 279   unsigned slp_unrolling_factor;
 280
 281   /* Reduction cycles detected in the loop. Used in loop-aware SLP.  */
 282   VEC (gimple, heap) *reductions;
 283
 284   /* All reduction chains in the loop, represented by the first
 285      stmt in the chain.  */
 286   VEC (gimple, heap) *reduction_chains;
 287
 288   /* Hash table used to choose the best peeling option.  */
 289   htab_t peeling_htab;
 290
 291   /* Cost data used by the target cost model.  */
 292   void *target_cost_data;
 293
 294   /* When we have grouped data accesses with gaps, we may introduce invalid
 295      memory accesses.  We peel the last iteration of the loop to prevent
 296      this.  */
 297   bool peeling_for_gaps;
 298
 299   /* Reductions are canonicalized so that the last operand is the reduction
 300      operand.  If this places a constant into RHS1, this decanonicalizes
 301      GIMPLE for other phases, so we must track when this has occurred and
 302      fix it up.  */
 303   bool operands_swapped;
 304
 305 } *loop_vec_info;
 306
 307 /* Access Functions.  */
 308 #define LOOP_VINFO_LOOP(L)                 (L)->loop
 309 #define LOOP_VINFO_BBS(L)                  (L)->bbs
 310 #define LOOP_VINFO_NITERS(L)               (L)->num_iters
 311 /* Since LOOP_VINFO_NITERS can change after prologue peeling
 312    retain total unchanged scalar loop iterations for cost model.  */
 313 #define LOOP_VINFO_NITERS_UNCHANGED(L)     (L)->num_iters_unchanged
 314 #define LOOP_VINFO_COST_MODEL_MIN_ITERS(L) (L)->min_profitable_iters
 315 #define LOOP_VINFO_VECTORIZABLE_P(L)       (L)->vectorizable
 316 #define LOOP_VINFO_VECT_FACTOR(L)          (L)->vectorization_factor
 317 #define LOOP_VINFO_PTR_MASK(L)             (L)->ptr_mask
 318 #define LOOP_VINFO_LOOP_NEST(L)            (L)->loop_nest
 319 #define LOOP_VINFO_DATAREFS(L)             (L)->datarefs
 320 #define LOOP_VINFO_DDRS(L)                 (L)->ddrs
 321 #define LOOP_VINFO_INT_NITERS(L)           (TREE_INT_CST_LOW ((L)->num_iters))
 322 #define LOOP_PEELING_FOR_ALIGNMENT(L)      (L)->peeling_for_alignment
 323 #define LOOP_VINFO_UNALIGNED_DR(L)         (L)->unaligned_dr
 324 #define LOOP_VINFO_MAY_MISALIGN_STMTS(L)   (L)->may_misalign_stmts
 325 #define LOOP_VINFO_LOC(L)                  (L)->loop_line_number
 326 #define LOOP_VINFO_MAY_ALIAS_DDRS(L)       (L)->may_alias_ddrs
 327 #define LOOP_VINFO_GROUPED_STORES(L)       (L)->grouped_stores
 328 #define LOOP_VINFO_SLP_INSTANCES(L)        (L)->slp_instances
 329 #define LOOP_VINFO_SLP_UNROLLING_FACTOR(L) (L)->slp_unrolling_factor
 330 #define LOOP_VINFO_REDUCTIONS(L)           (L)->reductions
 331 #define LOOP_VINFO_REDUCTION_CHAINS(L)     (L)->reduction_chains
 332 #define LOOP_VINFO_PEELING_HTAB(L)         (L)->peeling_htab
 333 #define LOOP_VINFO_TARGET_COST_DATA(L)     (L)->target_cost_data
 334 #define LOOP_VINFO_PEELING_FOR_GAPS(L)     (L)->peeling_for_gaps
 335 #define LOOP_VINFO_OPERANDS_SWAPPED(L)     (L)->operands_swapped
 336
 337 #define LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT(L) \
 338 VEC_length (gimple, (L)->may_misalign_stmts) > 0
 339 #define LOOP_REQUIRES_VERSIONING_FOR_ALIAS(L)     \
 340 VEC_length (ddr_p, (L)->may_alias_ddrs) > 0
 341
 342 #define NITERS_KNOWN_P(n)                     \
 343 (host_integerp ((n),0)                        \
 344 && TREE_INT_CST_LOW ((n)) > 0)
 345
 346 #define LOOP_VINFO_NITERS_KNOWN_P(L)          \
 347 NITERS_KNOWN_P((L)->num_iters)
 348
 349 static inline loop_vec_info
 350 loop_vec_info_for_loop (struct loop *loop)
 351 {
 352   return (loop_vec_info) loop->aux;
 353 }
 354
 355 static inline bool
 356 nested_in_vect_loop_p (struct loop *loop, gimple stmt)
 357 {
 358   return (loop->inner
 359           && (loop->inner == (gimple_bb (stmt))->loop_father));
 360 }
 361
 362 typedef struct _bb_vec_info {
 363
 364   basic_block bb;
 365   /* All interleaving chains of stores in the basic block, represented by the
 366      first stmt in the chain.  */
 367   VEC(gimple, heap) *grouped_stores;
 368
 369   /* All SLP instances in the basic block. This is a subset of the set of
 370      GROUP_STORES of the basic block.  */
 371   VEC(slp_instance, heap) *slp_instances;
 372
 373   /* All data references in the basic block.  */
 374   VEC (data_reference_p, heap) *datarefs;
 375
 376   /* All data dependences in the basic block.  */
 377   VEC (ddr_p, heap) *ddrs;
 378
 379   /* Cost data used by the target cost model.  */
 380   void *target_cost_data;
 381
 382 } *bb_vec_info;
 383
 384 #define BB_VINFO_BB(B)               (B)->bb
 385 #define BB_VINFO_GROUPED_STORES(B)   (B)->grouped_stores
 386 #define BB_VINFO_SLP_INSTANCES(B)    (B)->slp_instances
 387 #define BB_VINFO_DATAREFS(B)         (B)->datarefs
 388 #define BB_VINFO_DDRS(B)             (B)->ddrs
 389 #define BB_VINFO_TARGET_COST_DATA(B) (B)->target_cost_data
 390
 391 static inline bb_vec_info
 392 vec_info_for_bb (basic_block bb)
 393 {
 394   return (bb_vec_info) bb->aux;
 395 }
 396
 397 /*-----------------------------------------------------------------*/
 398 /* Info on vectorized defs.                                        */
 399 /*-----------------------------------------------------------------*/
 400 enum stmt_vec_info_type {
 401   undef_vec_info_type = 0,
 402   load_vec_info_type,
 403   store_vec_info_type,
 404   shift_vec_info_type,
 405   op_vec_info_type,
 406   call_vec_info_type,
 407   assignment_vec_info_type,
 408   condition_vec_info_type,
 409   reduc_vec_info_type,
 410   induc_vec_info_type,
 411   type_promotion_vec_info_type,
 412   type_demotion_vec_info_type,
 413   type_conversion_vec_info_type,
 414   loop_exit_ctrl_vec_info_type
 415 };
 416
 417 /* Indicates whether/how a variable is used in the scope of loop/basic
 418    block.  */
 419 enum vect_relevant {
 420   vect_unused_in_scope = 0,
 421   /* The def is in the inner loop, and the use is in the outer loop, and the
 422      use is a reduction stmt.  */
 423   vect_used_in_outer_by_reduction,
 424   /* The def is in the inner loop, and the use is in the outer loop (and is
 425      not part of reduction).  */
 426   vect_used_in_outer,
 427
 428   /* defs that feed computations that end up (only) in a reduction. These
 429      defs may be used by non-reduction stmts, but eventually, any
 430      computations/values that are affected by these defs are used to compute
 431      a reduction (i.e. don't get stored to memory, for example). We use this
 432      to identify computations that we can change the order in which they are
 433      computed.  */
 434   vect_used_by_reduction,
 435
 436   vect_used_in_scope
 437 };
 438
 439 /* The type of vectorization that can be applied to the stmt: regular loop-based
 440    vectorization; pure SLP - the stmt is a part of SLP instances and does not
 441    have uses outside SLP instances; or hybrid SLP and loop-based - the stmt is
 442    a part of SLP instance and also must be loop-based vectorized, since it has
 443    uses outside SLP sequences.
 444
 445    In the loop context the meanings of pure and hybrid SLP are slightly
 446    different. By saying that pure SLP is applied to the loop, we mean that we
 447    exploit only intra-iteration parallelism in the loop; i.e., the loop can be
 448    vectorized without doing any conceptual unrolling, cause we don't pack
 449    together stmts from different iterations, only within a single iteration.
 450    Loop hybrid SLP means that we exploit both intra-iteration and
 451    inter-iteration parallelism (e.g., number of elements in the vector is 4
 452    and the slp-group-size is 2, in which case we don't have enough parallelism
 453    within an iteration, so we obtain the rest of the parallelism from subsequent
 454    iterations by unrolling the loop by 2).  */
 455 enum slp_vect_type {
 456   loop_vect = 0,
 457   pure_slp,
 458   hybrid
 459 };
 460
 461
 462 typedef struct data_reference *dr_p;
 463 DEF_VEC_P(dr_p);
 464 DEF_VEC_ALLOC_P(dr_p,heap);
 465
 466 typedef struct _stmt_vec_info {
 467
 468   enum stmt_vec_info_type type;
 469
 470   /* Indicates whether this stmts is part of a computation whose result is
 471      used outside the loop.  */
 472   bool live;
 473
 474   /* Stmt is part of some pattern (computation idiom)  */
 475   bool in_pattern_p;
 476
 477   /* For loads only, if there is a store with the same location, this field is
 478      TRUE.  */
 479   bool read_write_dep;
 480
 481   /* The stmt to which this info struct refers to.  */
 482   gimple stmt;
 483
 484   /* The loop_vec_info with respect to which STMT is vectorized.  */
 485   loop_vec_info loop_vinfo;
 486
 487   /* The vector type to be used for the LHS of this statement.  */
 488   tree vectype;
 489
 490   /* The vectorized version of the stmt.  */
 491   gimple vectorized_stmt;
 492
 493
 494   /** The following is relevant only for stmts that contain a non-scalar
 495      data-ref (array/pointer/struct access). A GIMPLE stmt is expected to have
 496      at most one such data-ref.  **/
 497
 498   /* Information about the data-ref (access function, etc),
 499      relative to the inner-most containing loop.  */
 500   struct data_reference *data_ref_info;
 501
 502   /* Information about the data-ref relative to this loop
 503      nest (the loop that is being considered for vectorization).  */
 504   tree dr_base_address;
 505   tree dr_init;
 506   tree dr_offset;
 507   tree dr_step;
 508   tree dr_aligned_to;
 509
 510   /* For loop PHI nodes, the evolution part of it.  This makes sure
 511      this information is still available in vect_update_ivs_after_vectorizer
 512      where we may not be able to re-analyze the PHI nodes evolution as
 513      peeling for the prologue loop can make it unanalyzable.  The evolution
 514      part is still correct though.  */
 515   tree loop_phi_evolution_part;
 516
 517   /* Used for various bookkeeping purposes, generally holding a pointer to
 518      some other stmt S that is in some way "related" to this stmt.
 519      Current use of this field is:
 520         If this stmt is part of a pattern (i.e. the field 'in_pattern_p' is
 521         true): S is the "pattern stmt" that represents (and replaces) the
 522         sequence of stmts that constitutes the pattern.  Similarly, the
 523         related_stmt of the "pattern stmt" points back to this stmt (which is
 524         the last stmt in the original sequence of stmts that constitutes the
 525         pattern).  */
 526   gimple related_stmt;
 527
 528   /* Used to keep a sequence of def stmts of a pattern stmt if such exists.  */
 529   gimple_seq pattern_def_seq;
 530
 531   /* List of datarefs that are known to have the same alignment as the dataref
 532      of this stmt.  */
 533   VEC(dr_p,heap) *same_align_refs;
 534
 535   /* Classify the def of this stmt.  */
 536   enum vect_def_type def_type;
 537
 538   /*  Whether the stmt is SLPed, loop-based vectorized, or both.  */
 539   enum slp_vect_type slp_type;
 540
 541   /* Interleaving and reduction chains info.  */
 542   /* First element in the group.  */
 543   gimple first_element;
 544   /* Pointer to the next element in the group.  */
 545   gimple next_element;
 546   /* For data-refs, in case that two or more stmts share data-ref, this is the
 547      pointer to the previously detected stmt with the same dr.  */
 548   gimple same_dr_stmt;
 549   /* The size of the group.  */
 550   unsigned int size;
 551   /* For stores, number of stores from this group seen. We vectorize the last
 552      one.  */
 553   unsigned int store_count;
 554   /* For loads only, the gap from the previous load. For consecutive loads, GAP
 555      is 1.  */
 556   unsigned int gap;
 557
 558   /* Not all stmts in the loop need to be vectorized. e.g, the increment
 559      of the loop induction variable and computation of array indexes. relevant
 560      indicates whether the stmt needs to be vectorized.  */
 561   enum vect_relevant relevant;
 562
 563   /* The bb_vec_info with respect to which STMT is vectorized.  */
 564   bb_vec_info bb_vinfo;
 565
 566   /* Is this statement vectorizable or should it be skipped in (partial)
 567      vectorization.  */
 568   bool vectorizable;
 569
 570   /* For loads only, true if this is a gather load.  */
 571   bool gather_p;
 572   bool stride_load_p;
 573 } *stmt_vec_info;
 574
 575 /* Access Functions.  */
 576 #define STMT_VINFO_TYPE(S)                 (S)->type
 577 #define STMT_VINFO_STMT(S)                 (S)->stmt
 578 #define STMT_VINFO_LOOP_VINFO(S)           (S)->loop_vinfo
 579 #define STMT_VINFO_BB_VINFO(S)             (S)->bb_vinfo
 580 #define STMT_VINFO_RELEVANT(S)             (S)->relevant
 581 #define STMT_VINFO_LIVE_P(S)               (S)->live
 582 #define STMT_VINFO_VECTYPE(S)              (S)->vectype
 583 #define STMT_VINFO_VEC_STMT(S)             (S)->vectorized_stmt
 584 #define STMT_VINFO_VECTORIZABLE(S)         (S)->vectorizable
 585 #define STMT_VINFO_DATA_REF(S)             (S)->data_ref_info
 586 #define STMT_VINFO_GATHER_P(S)             (S)->gather_p
 587 #define STMT_VINFO_STRIDE_LOAD_P(S)        (S)->stride_load_p
 588
 589 #define STMT_VINFO_DR_BASE_ADDRESS(S)      (S)->dr_base_address
 590 #define STMT_VINFO_DR_INIT(S)              (S)->dr_init
 591 #define STMT_VINFO_DR_OFFSET(S)            (S)->dr_offset
 592 #define STMT_VINFO_DR_STEP(S)              (S)->dr_step
 593 #define STMT_VINFO_DR_ALIGNED_TO(S)        (S)->dr_aligned_to
 594
 595 #define STMT_VINFO_IN_PATTERN_P(S)         (S)->in_pattern_p
 596 #define STMT_VINFO_RELATED_STMT(S)         (S)->related_stmt
 597 #define STMT_VINFO_PATTERN_DEF_SEQ(S)      (S)->pattern_def_seq
 598 #define STMT_VINFO_SAME_ALIGN_REFS(S)      (S)->same_align_refs
 599 #define STMT_VINFO_DEF_TYPE(S)             (S)->def_type
 600 #define STMT_VINFO_GROUP_FIRST_ELEMENT(S)  (S)->first_element
 601 #define STMT_VINFO_GROUP_NEXT_ELEMENT(S)   (S)->next_element
 602 #define STMT_VINFO_GROUP_SIZE(S)           (S)->size
 603 #define STMT_VINFO_GROUP_STORE_COUNT(S)    (S)->store_count
 604 #define STMT_VINFO_GROUP_GAP(S)            (S)->gap
 605 #define STMT_VINFO_GROUP_SAME_DR_STMT(S)   (S)->same_dr_stmt
 606 #define STMT_VINFO_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 607 #define STMT_VINFO_GROUPED_ACCESS(S)      ((S)->first_element != NULL && (S)->data_ref_info)
 608 #define STMT_VINFO_LOOP_PHI_EVOLUTION_PART(S) (S)->loop_phi_evolution_part
 609
 610 #define GROUP_FIRST_ELEMENT(S)          (S)->first_element
 611 #define GROUP_NEXT_ELEMENT(S)           (S)->next_element
 612 #define GROUP_SIZE(S)                   (S)->size
 613 #define GROUP_STORE_COUNT(S)            (S)->store_count
 614 #define GROUP_GAP(S)                    (S)->gap
 615 #define GROUP_SAME_DR_STMT(S)           (S)->same_dr_stmt
 616 #define GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 617
 618 #define STMT_VINFO_RELEVANT_P(S)          ((S)->relevant != vect_unused_in_scope)
 619
 620 #define HYBRID_SLP_STMT(S)                ((S)->slp_type == hybrid)
 621 #define PURE_SLP_STMT(S)                  ((S)->slp_type == pure_slp)
 622 #define STMT_SLP_TYPE(S)                   (S)->slp_type
 623
 624 #define VECT_MAX_COST 1000
 625
 626 /* The maximum number of intermediate steps required in multi-step type
 627    conversion.  */
 628 #define MAX_INTERM_CVT_STEPS         3
 629
 630 /* The maximum vectorization factor supported by any target (V32QI).  */
 631 #define MAX_VECTORIZATION_FACTOR 32
 632
 633 /* Avoid GTY(()) on stmt_vec_info.  */
 634 typedef void *vec_void_p;
 635 DEF_VEC_P (vec_void_p);
 636 DEF_VEC_ALLOC_P (vec_void_p, heap);
 637
 638 extern VEC(vec_void_p,heap) *stmt_vec_info_vec;
 639
 640 void init_stmt_vec_info_vec (void);
 641 void free_stmt_vec_info_vec (void);
 642
 643 /* Return a stmt_vec_info corresponding to STMT.  */
 644
 645 static inline stmt_vec_info
 646 vinfo_for_stmt (gimple stmt)
 647 {
 648   unsigned int uid = gimple_uid (stmt);
 649   if (uid == 0)
 650     return NULL;
 651
 652   return (stmt_vec_info) VEC_index (vec_void_p, stmt_vec_info_vec, uid - 1);
 653 }
 654
 655 /* Set vectorizer information INFO for STMT.  */
 656
 657 static inline void
 658 set_vinfo_for_stmt (gimple stmt, stmt_vec_info info)
 659 {
 660   unsigned int uid = gimple_uid (stmt);
 661   if (uid == 0)
 662     {
 663       gcc_checking_assert (info);
 664       uid = VEC_length (vec_void_p, stmt_vec_info_vec) + 1;
 665       gimple_set_uid (stmt, uid);
 666       VEC_safe_push (vec_void_p, heap, stmt_vec_info_vec, (vec_void_p) info);
 667     }
 668   else
 669     VEC_replace (vec_void_p, stmt_vec_info_vec, uid - 1, (vec_void_p) info);
 670 }
 671
 672 /* Return the earlier statement between STMT1 and STMT2.  */
 673
 674 static inline gimple
 675 get_earlier_stmt (gimple stmt1, gimple stmt2)
 676 {
 677   unsigned int uid1, uid2;
 678
 679   if (stmt1 == NULL)
 680     return stmt2;
 681
 682   if (stmt2 == NULL)
 683     return stmt1;
 684
 685   uid1 = gimple_uid (stmt1);
 686   uid2 = gimple_uid (stmt2);
 687
 688   if (uid1 == 0 || uid2 == 0)
 689     return NULL;
 690
 691   gcc_checking_assert (uid1 <= VEC_length (vec_void_p, stmt_vec_info_vec)
 692                        && uid2 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 693
 694   if (uid1 < uid2)
 695     return stmt1;
 696   else
 697     return stmt2;
 698 }
 699
 700 /* Return the later statement between STMT1 and STMT2.  */
 701
 702 static inline gimple
 703 get_later_stmt (gimple stmt1, gimple stmt2)
 704 {
 705   unsigned int uid1, uid2;
 706
 707   if (stmt1 == NULL)
 708     return stmt2;
 709
 710   if (stmt2 == NULL)
 711     return stmt1;
 712
 713   uid1 = gimple_uid (stmt1);
 714   uid2 = gimple_uid (stmt2);
 715
 716   if (uid1 == 0 || uid2 == 0)
 717     return NULL;
 718
 719   gcc_assert (uid1 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 720   gcc_assert (uid2 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 721
 722   if (uid1 > uid2)
 723     return stmt1;
 724   else
 725     return stmt2;
 726 }
 727
 728 /* Return TRUE if a statement represented by STMT_INFO is a part of a
 729    pattern.  */
 730
 731 static inline bool
 732 is_pattern_stmt_p (stmt_vec_info stmt_info)
 733 {
 734   gimple related_stmt;
 735   stmt_vec_info related_stmt_info;
 736
 737   related_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
 738   if (related_stmt
 739       && (related_stmt_info = vinfo_for_stmt (related_stmt))
 740       && STMT_VINFO_IN_PATTERN_P (related_stmt_info))
 741     return true;
 742
 743   return false;
 744 }
 745
 746 /* Return true if BB is a loop header.  */
 747
 748 static inline bool
 749 is_loop_header_bb_p (basic_block bb)
 750 {
 751   if (bb == (bb->loop_father)->header)
 752     return true;
 753   gcc_checking_assert (EDGE_COUNT (bb->preds) == 1);
 754   return false;
 755 }
 756
 757 /* Return pow2 (X).  */
 758
 759 static inline int
 760 vect_pow2 (int x)
 761 {
 762   int i, res = 1;
 763
 764   for (i = 0; i < x; i++)
 765     res *= 2;
 766
 767   return res;
 768 }
 769
 770 /* Alias targetm.vectorize.builtin_vectorization_cost.  */
 771
 772 static inline int
 773 builtin_vectorization_cost (enum vect_cost_for_stmt type_of_cost,
 774                             tree vectype, int misalign)
 775 {
 776   return targetm.vectorize.builtin_vectorization_cost (type_of_cost,
 777                                                        vectype, misalign);
 778 }
 779
 780 /* Get cost by calling cost target builtin.  */
 781
 782 static inline
 783 int vect_get_stmt_cost (enum vect_cost_for_stmt type_of_cost)
 784 {
 785   return builtin_vectorization_cost (type_of_cost, NULL, 0);
 786 }
 787
 788 /* Alias targetm.vectorize.init_cost.  */
 789
 790 static inline void *
 791 init_cost (struct loop *loop_info)
 792 {
 793   return targetm.vectorize.init_cost (loop_info);
 794 }
 795
 796 /* Alias targetm.vectorize.add_stmt_cost.  */
 797
 798 static inline unsigned
 799 add_stmt_cost (void *data, int count, enum vect_cost_for_stmt kind,
 800                stmt_vec_info stmt_info, int misalign,
 801                enum vect_cost_model_location where)
 802 {
 803   return targetm.vectorize.add_stmt_cost (data, count, kind,
 804                                           stmt_info, misalign, where);
 805 }
 806
 807 /* Alias targetm.vectorize.finish_cost.  */
 808
 809 static inline void
 810 finish_cost (void *data, unsigned *prologue_cost,
 811              unsigned *body_cost, unsigned *epilogue_cost)
 812 {
 813   targetm.vectorize.finish_cost (data, prologue_cost, body_cost, epilogue_cost);
 814 }
 815
 816 /* Alias targetm.vectorize.destroy_cost_data.  */
 817
 818 static inline void
 819 destroy_cost_data (void *data)
 820 {
 821   targetm.vectorize.destroy_cost_data (data);
 822 }
 823
 824
 825 /*-----------------------------------------------------------------*/
 826 /* Info on data references alignment.                              */
 827 /*-----------------------------------------------------------------*/
 828
 829 /* Reflects actual alignment of first access in the vectorized loop,
 830    taking into account peeling/versioning if applied.  */
 831 #define DR_MISALIGNMENT(DR)   ((int) (size_t) (DR)->aux)
 832 #define SET_DR_MISALIGNMENT(DR, VAL)   ((DR)->aux = (void *) (size_t) (VAL))
 833
 834 /* Return TRUE if the data access is aligned, and FALSE otherwise.  */
 835
 836 static inline bool
 837 aligned_access_p (struct data_reference *data_ref_info)
 838 {
 839   return (DR_MISALIGNMENT (data_ref_info) == 0);
 840 }
 841
 842 /* Return TRUE if the alignment of the data access is known, and FALSE
 843    otherwise.  */
 844
 845 static inline bool
 846 known_alignment_for_access_p (struct data_reference *data_ref_info)
 847 {
 848   return (DR_MISALIGNMENT (data_ref_info) != -1);
 849 }
 850
 851 /* vect_dump will be set to stderr or dump_file if exist.  */
 852 extern FILE *vect_dump;
 853 extern LOC vect_loop_location;
 854
 855 /*-----------------------------------------------------------------*/
 856 /* Function prototypes.                                            */
 857 /*-----------------------------------------------------------------*/
 858
 859 /* Simple loop peeling and versioning utilities for vectorizer's purposes -
 860    in tree-vect-loop-manip.c.  */
 861 extern void slpeel_make_loop_iterate_ntimes (struct loop *, tree);
 862 extern bool slpeel_can_duplicate_loop_p (const struct loop *, const_edge);
 863 extern void vect_loop_versioning (loop_vec_info, unsigned int, bool);
 864 extern void vect_do_peeling_for_loop_bound (loop_vec_info, tree *,
 865                                             unsigned int, bool);
 866 extern void vect_do_peeling_for_alignment (loop_vec_info, unsigned int, bool);
 867 extern LOC find_loop_location (struct loop *);
 868 extern bool vect_can_advance_ivs_p (loop_vec_info);
 869
 870 /* In tree-vect-stmts.c.  */
 871 extern unsigned int current_vector_size;
 872 extern tree get_vectype_for_scalar_type (tree);
 873 extern tree get_same_sized_vectype (tree, tree);
 874 extern bool vect_is_simple_use (tree, gimple, loop_vec_info,
 875                                 bb_vec_info, gimple *,
 876                                 tree *,  enum vect_def_type *);
 877 extern bool vect_is_simple_use_1 (tree, gimple, loop_vec_info,
 878                                   bb_vec_info, gimple *,
 879                                   tree *,  enum vect_def_type *, tree *);
 880 extern bool supportable_widening_operation (enum tree_code, gimple, tree, tree,
 881                                             enum tree_code *, enum tree_code *,
 882                                             int *, VEC (tree, heap) **);
 883 extern bool supportable_narrowing_operation (enum tree_code, tree, tree,
 884                                              enum tree_code *,
 885                                              int *, VEC (tree, heap) **);
 886 extern stmt_vec_info new_stmt_vec_info (gimple stmt, loop_vec_info,
 887                                         bb_vec_info);
 888 extern void free_stmt_vec_info (gimple stmt);
 889 extern tree vectorizable_function (gimple, tree, tree);
 890 extern void vect_model_simple_cost (stmt_vec_info, int, enum vect_def_type *,
 891                                     stmt_vector_for_cost *,
 892                                     stmt_vector_for_cost *);
 893 extern void vect_model_store_cost (stmt_vec_info, int, bool,
 894                                    enum vect_def_type, slp_tree,
 895                                    stmt_vector_for_cost *,
 896                                    stmt_vector_for_cost *);
 897 extern void vect_model_load_cost (stmt_vec_info, int, bool, slp_tree,
 898                                   stmt_vector_for_cost *,
 899                                   stmt_vector_for_cost *);
 900 extern unsigned record_stmt_cost (stmt_vector_for_cost *, int,
 901                                   enum vect_cost_for_stmt, stmt_vec_info,
 902                                   int, enum vect_cost_model_location);
 903 extern void vect_finish_stmt_generation (gimple, gimple,
 904                                          gimple_stmt_iterator *);
 905 extern bool vect_mark_stmts_to_be_vectorized (loop_vec_info);
 906 extern tree vect_get_vec_def_for_operand (tree, gimple, tree *);
 907 extern tree vect_init_vector (gimple, tree, tree,
 908                               gimple_stmt_iterator *);
 909 extern tree vect_get_vec_def_for_stmt_copy (enum vect_def_type, tree);
 910 extern bool vect_transform_stmt (gimple, gimple_stmt_iterator *,
 911                                  bool *, slp_tree, slp_instance);
 912 extern void vect_remove_stores (gimple);
 913 extern bool vect_analyze_stmt (gimple, bool *, slp_tree);
 914 extern bool vectorizable_condition (gimple, gimple_stmt_iterator *, gimple *,
 915                                     tree, int, slp_tree);
 916 extern void vect_get_load_cost (struct data_reference *, int, bool,
 917                                 unsigned int *, unsigned int *,
 918                                 stmt_vector_for_cost *,
 919                                 stmt_vector_for_cost *, bool);
 920 extern void vect_get_store_cost (struct data_reference *, int,
 921                                  unsigned int *, stmt_vector_for_cost *);
 922 extern bool vect_supportable_shift (enum tree_code, tree);
 923 extern void vect_get_vec_defs (tree, tree, gimple, VEC (tree, heap) **,
 924                                VEC (tree, heap) **, slp_tree, int);
 925 extern tree vect_gen_perm_mask (tree, unsigned char *);
 926
 927 /* In tree-vect-data-refs.c.  */
 928 extern bool vect_can_force_dr_alignment_p (const_tree, unsigned int);
 929 extern enum dr_alignment_support vect_supportable_dr_alignment
 930                                            (struct data_reference *, bool);
 931 extern tree vect_get_smallest_scalar_type (gimple, HOST_WIDE_INT *,
 932                                            HOST_WIDE_INT *);
 933 extern bool vect_analyze_data_ref_dependences (loop_vec_info, bb_vec_info,
 934                                                int *);
 935 extern bool vect_enhance_data_refs_alignment (loop_vec_info);
 936 extern bool vect_analyze_data_refs_alignment (loop_vec_info, bb_vec_info);
 937 extern bool vect_verify_datarefs_alignment (loop_vec_info, bb_vec_info);
 938 extern bool vect_analyze_data_ref_accesses (loop_vec_info, bb_vec_info);
 939 extern bool vect_prune_runtime_alias_test_list (loop_vec_info);
 940 extern tree vect_check_gather (gimple, loop_vec_info, tree *, tree *,
 941                                int *);
 942 extern bool vect_check_strided_load (gimple, loop_vec_info, tree *, tree *);
 943 extern bool vect_analyze_data_refs (loop_vec_info, bb_vec_info, int *);
 944 extern tree vect_create_data_ref_ptr (gimple, tree, struct loop *, tree,
 945                                       tree *, gimple_stmt_iterator *,
 946                                       gimple *, bool, bool *);
 947 extern tree bump_vector_ptr (tree, gimple, gimple_stmt_iterator *, gimple, tree);
 948 extern tree vect_create_destination_var (tree, tree);
 949 extern bool vect_grouped_store_supported (tree, unsigned HOST_WIDE_INT);
 950 extern bool vect_store_lanes_supported (tree, unsigned HOST_WIDE_INT);
 951 extern bool vect_grouped_load_supported (tree, unsigned HOST_WIDE_INT);
 952 extern bool vect_load_lanes_supported (tree, unsigned HOST_WIDE_INT);
 953 extern void vect_permute_store_chain (VEC(tree,heap) *,unsigned int, gimple,
 954                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 955 extern tree vect_setup_realignment (gimple, gimple_stmt_iterator *, tree *,
 956                                     enum dr_alignment_support, tree,
 957                                     struct loop **);
 958 extern void vect_transform_grouped_load (gimple, VEC(tree,heap) *, int,
 959                                          gimple_stmt_iterator *);
 960 extern void vect_record_grouped_load_vectors (gimple, VEC(tree,heap) *);
 961 extern int vect_get_place_in_interleaving_chain (gimple, gimple);
 962 extern tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
 963 extern tree vect_create_addr_base_for_vector_ref (gimple, gimple_seq *,
 964                                                   tree, struct loop *);
 965
 966 /* In tree-vect-loop.c.  */
 967 /* FORNOW: Used in tree-parloops.c.  */
 968 extern void destroy_loop_vec_info (loop_vec_info, bool);
 969 extern gimple vect_force_simple_reduction (loop_vec_info, gimple, bool, bool *);
 970 /* Drive for loop analysis stage.  */
 971 extern loop_vec_info vect_analyze_loop (struct loop *);
 972 /* Drive for loop transformation stage.  */
 973 extern void vect_transform_loop (loop_vec_info);
 974 extern loop_vec_info vect_analyze_loop_form (struct loop *);
 975 extern bool vectorizable_live_operation (gimple, gimple_stmt_iterator *,
 976                                          gimple *);
 977 extern bool vectorizable_reduction (gimple, gimple_stmt_iterator *, gimple *,
 978                                     slp_tree);
 979 extern bool vectorizable_induction (gimple, gimple_stmt_iterator *, gimple *);
 980 extern int vect_estimate_min_profitable_iters (loop_vec_info);
 981 extern tree get_initial_def_for_reduction (gimple, tree, tree *);
 982 extern int vect_min_worthwhile_factor (enum tree_code);
 983 extern int vect_get_known_peeling_cost (loop_vec_info, int, int *, int,
 984                                         stmt_vector_for_cost *,
 985                                         stmt_vector_for_cost *);
 986 extern int vect_get_single_scalar_iteration_cost (loop_vec_info);
 987
 988 /* In tree-vect-slp.c.  */
 989 extern void vect_free_slp_instance (slp_instance);
 990 extern bool vect_transform_slp_perm_load (gimple, VEC (tree, heap) *,
 991                                           gimple_stmt_iterator *, int,
 992                                           slp_instance, bool);
 993 extern bool vect_schedule_slp (loop_vec_info, bb_vec_info);
 994 extern void vect_update_slp_costs_according_to_vf (loop_vec_info);
 995 extern bool vect_analyze_slp (loop_vec_info, bb_vec_info);
 996 extern bool vect_make_slp_decision (loop_vec_info);
 997 extern void vect_detect_hybrid_slp (loop_vec_info);
 998 extern void vect_get_slp_defs (VEC (tree, heap) *, slp_tree,
 999                                VEC (slp_void_p, heap) **, int);
1000
1001 extern LOC find_bb_location (basic_block);
1002 extern bb_vec_info vect_slp_analyze_bb (basic_block);
1003 extern void vect_slp_transform_bb (basic_block);
1004
1005 /* In tree-vect-patterns.c.  */
1006 /* Pattern recognition functions.
1007    Additional pattern recognition functions can (and will) be added
1008    in the future.  */
1009 typedef gimple (* vect_recog_func_ptr) (VEC (gimple, heap) **, tree *, tree *);
1010 #define NUM_PATTERNS 10
1011 void vect_pattern_recog (loop_vec_info, bb_vec_info);
1012
1013 /* In tree-vectorizer.c.  */
1014 unsigned vectorize_loops (void);
1015 /* Vectorization debug information */
1016 extern bool vect_print_dump_info (enum vect_verbosity_levels);
1017
1018 #endif  /* GCC_TREE_VECTORIZER_H  */