gcc/tree-vectorizer.h

   1 /* Vectorizer
   2    Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
   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
  27 typedef source_location LOC;
  28 #define UNKNOWN_LOC UNKNOWN_LOCATION
  29 #define EXPR_LOC(e) EXPR_LOCATION(e)
  30 #define LOC_FILE(l) LOCATION_FILE (l)
  31 #define LOC_LINE(l) LOCATION_LINE (l)
  32
  33 /* Used for naming of new temporaries.  */
  34 enum vect_var_kind {
  35   vect_simple_var,
  36   vect_pointer_var,
  37   vect_scalar_var
  38 };
  39
  40 /* Defines type of operation.  */
  41 enum operation_type {
  42   unary_op = 1,
  43   binary_op,
  44   ternary_op
  45 };
  46
  47 /* Define type of available alignment support.  */
  48 enum dr_alignment_support {
  49   dr_unaligned_unsupported,
  50   dr_unaligned_supported,
  51   dr_explicit_realign,
  52   dr_explicit_realign_optimized,
  53   dr_aligned
  54 };
  55
  56 /* Define type of def-use cross-iteration cycle.  */
  57 enum vect_def_type {
  58   vect_uninitialized_def = 0,
  59   vect_constant_def = 1,
  60   vect_external_def,
  61   vect_internal_def,
  62   vect_induction_def,
  63   vect_reduction_def,
  64   vect_double_reduction_def,
  65   vect_nested_cycle,
  66   vect_unknown_def_type
  67 };
  68
  69 #define VECTORIZABLE_CYCLE_DEF(D) (((D) == vect_reduction_def)           \
  70                                    || ((D) == vect_double_reduction_def) \
  71                                    || ((D) == vect_nested_cycle))
  72
  73 /* Define verbosity levels.  */
  74 enum verbosity_levels {
  75   REPORT_NONE,
  76   REPORT_VECTORIZED_LOCATIONS,
  77   REPORT_UNVECTORIZED_LOCATIONS,
  78   REPORT_COST,
  79   REPORT_ALIGNMENT,
  80   REPORT_DR_DETAILS,
  81   REPORT_BAD_FORM_LOOPS,
  82   REPORT_OUTER_LOOPS,
  83   REPORT_SLP,
  84   REPORT_DETAILS,
  85   /* New verbosity levels should be added before this one.  */
  86   MAX_VERBOSITY_LEVEL
  87 };
  88
  89 /************************************************************************
  90   SLP
  91  ************************************************************************/
  92
  93 /* A computation tree of an SLP instance. Each node corresponds to a group of
  94    stmts to be packed in a SIMD stmt.  */
  95 typedef struct _slp_tree {
  96   /* Only binary and unary operations are supported. LEFT child corresponds to
  97      the first operand and RIGHT child to the second if the operation is
  98      binary.  */
  99   struct _slp_tree *left;
 100   struct _slp_tree *right;
 101   /* A group of scalar stmts to be vectorized together.  */
 102   VEC (gimple, heap) *stmts;
 103   /* Vectorized stmt/s.  */
 104   VEC (gimple, heap) *vec_stmts;
 105   /* Number of vector stmts that are created to replace the group of scalar
 106      stmts. It is calculated during the transformation phase as the number of
 107      scalar elements in one scalar iteration (GROUP_SIZE) multiplied by VF
 108      divided by vector size.  */
 109   unsigned int vec_stmts_size;
 110   /* Vectorization costs associated with SLP node.  */
 111   struct
 112   {
 113     int outside_of_loop;     /* Statements generated outside loop.  */
 114     int inside_of_loop;      /* Statements generated inside loop.  */
 115   } cost;
 116 } *slp_tree;
 117
 118 DEF_VEC_P(slp_tree);
 119 DEF_VEC_ALLOC_P(slp_tree, heap);
 120
 121 /* SLP instance is a sequence of stmts in a loop that can be packed into
 122    SIMD stmts.  */
 123 typedef struct _slp_instance {
 124   /* The root of SLP tree.  */
 125   slp_tree root;
 126
 127   /* Size of groups of scalar stmts that will be replaced by SIMD stmt/s.  */
 128   unsigned int group_size;
 129
 130   /* The unrolling factor required to vectorized this SLP instance.  */
 131   unsigned int unrolling_factor;
 132
 133   /* Vectorization costs associated with SLP instance.  */
 134   struct
 135   {
 136     int outside_of_loop;     /* Statements generated outside loop.  */
 137     int inside_of_loop;      /* Statements generated inside loop.  */
 138   } cost;
 139
 140   /* Loads permutation relatively to the stores, NULL if there is no
 141      permutation.  */
 142   VEC (int, heap) *load_permutation;
 143
 144   /* The group of nodes that contain loads of this SLP instance.  */
 145   VEC (slp_tree, heap) *loads;
 146
 147   /* The first scalar load of the instance. The created vector loads will be
 148      inserted before this statement.  */
 149   gimple first_load;
 150 } *slp_instance;
 151
 152 DEF_VEC_P(slp_instance);
 153 DEF_VEC_ALLOC_P(slp_instance, heap);
 154
 155 /* Access Functions.  */
 156 #define SLP_INSTANCE_TREE(S)                     (S)->root
 157 #define SLP_INSTANCE_GROUP_SIZE(S)               (S)->group_size
 158 #define SLP_INSTANCE_UNROLLING_FACTOR(S)         (S)->unrolling_factor
 159 #define SLP_INSTANCE_OUTSIDE_OF_LOOP_COST(S)     (S)->cost.outside_of_loop
 160 #define SLP_INSTANCE_INSIDE_OF_LOOP_COST(S)      (S)->cost.inside_of_loop
 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_LEFT(S)                         (S)->left
 166 #define SLP_TREE_RIGHT(S)                        (S)->right
 167 #define SLP_TREE_SCALAR_STMTS(S)                 (S)->stmts
 168 #define SLP_TREE_VEC_STMTS(S)                    (S)->vec_stmts
 169 #define SLP_TREE_NUMBER_OF_VEC_STMTS(S)          (S)->vec_stmts_size
 170 #define SLP_TREE_OUTSIDE_OF_LOOP_COST(S)         (S)->cost.outside_of_loop
 171 #define SLP_TREE_INSIDE_OF_LOOP_COST(S)          (S)->cost.inside_of_loop
 172
 173 /*-----------------------------------------------------------------*/
 174 /* Info on vectorized loops.                                       */
 175 /*-----------------------------------------------------------------*/
 176 typedef struct _loop_vec_info {
 177
 178   /* The loop to which this info struct refers to.  */
 179   struct loop *loop;
 180
 181   /* The loop basic blocks.  */
 182   basic_block *bbs;
 183
 184   /* Number of iterations.  */
 185   tree num_iters;
 186   tree num_iters_unchanged;
 187
 188   /* Minimum number of iterations below which vectorization is expected to
 189      not be profitable (as estimated by the cost model).
 190      -1 indicates that vectorization will not be profitable.
 191      FORNOW: This field is an int. Will be a tree in the future, to represent
 192              values unknown at compile time.  */
 193   int min_profitable_iters;
 194
 195   /* Is the loop vectorizable? */
 196   bool vectorizable;
 197
 198   /* Unrolling factor  */
 199   int vectorization_factor;
 200
 201   /* Unknown DRs according to which loop was peeled.  */
 202   struct data_reference *unaligned_dr;
 203
 204   /* peeling_for_alignment indicates whether peeling for alignment will take
 205      place, and what the peeling factor should be:
 206      peeling_for_alignment = X means:
 207         If X=0: Peeling for alignment will not be applied.
 208         If X>0: Peel first X iterations.
 209         If X=-1: Generate a runtime test to calculate the number of iterations
 210                  to be peeled, using the dataref recorded in the field
 211                  unaligned_dr.  */
 212   int peeling_for_alignment;
 213
 214   /* The mask used to check the alignment of pointers or arrays.  */
 215   int ptr_mask;
 216
 217   /* All data references in the loop.  */
 218   VEC (data_reference_p, heap) *datarefs;
 219
 220   /* All data dependences in the loop.  */
 221   VEC (ddr_p, heap) *ddrs;
 222
 223   /* Data Dependence Relations defining address ranges that are candidates
 224      for a run-time aliasing check.  */
 225   VEC (ddr_p, heap) *may_alias_ddrs;
 226
 227   /* Statements in the loop that have data references that are candidates for a
 228      runtime (loop versioning) misalignment check.  */
 229   VEC(gimple,heap) *may_misalign_stmts;
 230
 231   /* The loop location in the source.  */
 232   LOC loop_line_number;
 233
 234   /* All interleaving chains of stores in the loop, represented by the first
 235      stmt in the chain.  */
 236   VEC(gimple, heap) *strided_stores;
 237
 238   /* All SLP instances in the loop. This is a subset of the set of STRIDED_STORES
 239      of the loop.  */
 240   VEC(slp_instance, heap) *slp_instances;
 241
 242   /* The unrolling factor needed to SLP the loop. In case of that pure SLP is
 243      applied to the loop, i.e., no unrolling is needed, this is 1.  */
 244   unsigned slp_unrolling_factor;
 245
 246   /* Reduction cycles detected in the loop. Used in loop-aware SLP.  */
 247   VEC (gimple, heap) *reductions;
 248 } *loop_vec_info;
 249
 250 /* Access Functions.  */
 251 #define LOOP_VINFO_LOOP(L)                 (L)->loop
 252 #define LOOP_VINFO_BBS(L)                  (L)->bbs
 253 #define LOOP_VINFO_NITERS(L)               (L)->num_iters
 254 /* Since LOOP_VINFO_NITERS can change after prologue peeling
 255    retain total unchanged scalar loop iterations for cost model.  */
 256 #define LOOP_VINFO_NITERS_UNCHANGED(L)     (L)->num_iters_unchanged
 257 #define LOOP_VINFO_COST_MODEL_MIN_ITERS(L) (L)->min_profitable_iters
 258 #define LOOP_VINFO_VECTORIZABLE_P(L)       (L)->vectorizable
 259 #define LOOP_VINFO_VECT_FACTOR(L)          (L)->vectorization_factor
 260 #define LOOP_VINFO_PTR_MASK(L)             (L)->ptr_mask
 261 #define LOOP_VINFO_DATAREFS(L)             (L)->datarefs
 262 #define LOOP_VINFO_DDRS(L)                 (L)->ddrs
 263 #define LOOP_VINFO_INT_NITERS(L)           (TREE_INT_CST_LOW ((L)->num_iters))
 264 #define LOOP_PEELING_FOR_ALIGNMENT(L)      (L)->peeling_for_alignment
 265 #define LOOP_VINFO_UNALIGNED_DR(L)         (L)->unaligned_dr
 266 #define LOOP_VINFO_MAY_MISALIGN_STMTS(L)   (L)->may_misalign_stmts
 267 #define LOOP_VINFO_LOC(L)                  (L)->loop_line_number
 268 #define LOOP_VINFO_MAY_ALIAS_DDRS(L)       (L)->may_alias_ddrs
 269 #define LOOP_VINFO_STRIDED_STORES(L)       (L)->strided_stores
 270 #define LOOP_VINFO_SLP_INSTANCES(L)        (L)->slp_instances
 271 #define LOOP_VINFO_SLP_UNROLLING_FACTOR(L) (L)->slp_unrolling_factor
 272 #define LOOP_VINFO_REDUCTIONS(L)           (L)->reductions
 273
 274 #define LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT(L) \
 275 VEC_length (gimple, (L)->may_misalign_stmts) > 0
 276 #define LOOP_REQUIRES_VERSIONING_FOR_ALIAS(L)     \
 277 VEC_length (ddr_p, (L)->may_alias_ddrs) > 0
 278
 279 #define NITERS_KNOWN_P(n)                     \
 280 (host_integerp ((n),0)                        \
 281 && TREE_INT_CST_LOW ((n)) > 0)
 282
 283 #define LOOP_VINFO_NITERS_KNOWN_P(L)          \
 284 NITERS_KNOWN_P((L)->num_iters)
 285
 286 static inline loop_vec_info
 287 loop_vec_info_for_loop (struct loop *loop)
 288 {
 289   return (loop_vec_info) loop->aux;
 290 }
 291
 292 static inline bool
 293 nested_in_vect_loop_p (struct loop *loop, gimple stmt)
 294 {
 295   return (loop->inner
 296           && (loop->inner == (gimple_bb (stmt))->loop_father));
 297 }
 298
 299 typedef struct _bb_vec_info {
 300
 301   basic_block bb;
 302   /* All interleaving chains of stores in the basic block, represented by the
 303      first stmt in the chain.  */
 304   VEC(gimple, heap) *strided_stores;
 305
 306   /* All SLP instances in the basic block. This is a subset of the set of
 307      STRIDED_STORES of the basic block.  */
 308   VEC(slp_instance, heap) *slp_instances;
 309
 310   /* All data references in the basic block.  */
 311   VEC (data_reference_p, heap) *datarefs;
 312
 313   /* All data dependences in the basic block.  */
 314   VEC (ddr_p, heap) *ddrs;
 315 } *bb_vec_info;
 316
 317 #define BB_VINFO_BB(B)              (B)->bb
 318 #define BB_VINFO_STRIDED_STORES(B)  (B)->strided_stores
 319 #define BB_VINFO_SLP_INSTANCES(B)   (B)->slp_instances
 320 #define BB_VINFO_DATAREFS(B)        (B)->datarefs
 321 #define BB_VINFO_DDRS(B)            (B)->ddrs
 322
 323 static inline bb_vec_info
 324 vec_info_for_bb (basic_block bb)
 325 {
 326   return (bb_vec_info) bb->aux;
 327 }
 328
 329 /*-----------------------------------------------------------------*/
 330 /* Info on vectorized defs.                                        */
 331 /*-----------------------------------------------------------------*/
 332 enum stmt_vec_info_type {
 333   undef_vec_info_type = 0,
 334   load_vec_info_type,
 335   store_vec_info_type,
 336   op_vec_info_type,
 337   call_vec_info_type,
 338   assignment_vec_info_type,
 339   condition_vec_info_type,
 340   reduc_vec_info_type,
 341   induc_vec_info_type,
 342   type_promotion_vec_info_type,
 343   type_demotion_vec_info_type,
 344   type_conversion_vec_info_type,
 345   loop_exit_ctrl_vec_info_type
 346 };
 347
 348 /* Indicates whether/how a variable is used in the scope of loop/basic
 349    block.  */
 350 enum vect_relevant {
 351   vect_unused_in_scope = 0,
 352   /* The def is in the inner loop, and the use is in the outer loop, and the
 353      use is a reduction stmt.  */
 354   vect_used_in_outer_by_reduction,
 355   /* The def is in the inner loop, and the use is in the outer loop (and is
 356      not part of reduction).  */
 357   vect_used_in_outer,
 358
 359   /* defs that feed computations that end up (only) in a reduction. These
 360      defs may be used by non-reduction stmts, but eventually, any
 361      computations/values that are affected by these defs are used to compute
 362      a reduction (i.e. don't get stored to memory, for example). We use this
 363      to identify computations that we can change the order in which they are
 364      computed.  */
 365   vect_used_by_reduction,
 366
 367   vect_used_in_scope
 368 };
 369
 370 /* The type of vectorization that can be applied to the stmt: regular loop-based
 371    vectorization; pure SLP - the stmt is a part of SLP instances and does not
 372    have uses outside SLP instances; or hybrid SLP and loop-based - the stmt is
 373    a part of SLP instance and also must be loop-based vectorized, since it has
 374    uses outside SLP sequences.
 375
 376    In the loop context the meanings of pure and hybrid SLP are slightly
 377    different. By saying that pure SLP is applied to the loop, we mean that we
 378    exploit only intra-iteration parallelism in the loop; i.e., the loop can be
 379    vectorized without doing any conceptual unrolling, cause we don't pack
 380    together stmts from different iterations, only within a single iteration.
 381    Loop hybrid SLP means that we exploit both intra-iteration and
 382    inter-iteration parallelism (e.g., number of elements in the vector is 4
 383    and the slp-group-size is 2, in which case we don't have enough parallelism
 384    within an iteration, so we obtain the rest of the parallelism from subsequent
 385    iterations by unrolling the loop by 2).  */
 386 enum slp_vect_type {
 387   loop_vect = 0,
 388   pure_slp,
 389   hybrid
 390 };
 391
 392
 393 typedef struct data_reference *dr_p;
 394 DEF_VEC_P(dr_p);
 395 DEF_VEC_ALLOC_P(dr_p,heap);
 396
 397 typedef struct _stmt_vec_info {
 398
 399   enum stmt_vec_info_type type;
 400
 401   /* The stmt to which this info struct refers to.  */
 402   gimple stmt;
 403
 404   /* The loop_vec_info with respect to which STMT is vectorized.  */
 405   loop_vec_info loop_vinfo;
 406
 407   /* Not all stmts in the loop need to be vectorized. e.g, the increment
 408      of the loop induction variable and computation of array indexes. relevant
 409      indicates whether the stmt needs to be vectorized.  */
 410   enum vect_relevant relevant;
 411
 412   /* Indicates whether this stmts is part of a computation whose result is
 413      used outside the loop.  */
 414   bool live;
 415
 416   /* The vector type to be used for the LHS of this statement.  */
 417   tree vectype;
 418
 419   /* The vectorized version of the stmt.  */
 420   gimple vectorized_stmt;
 421
 422
 423   /** The following is relevant only for stmts that contain a non-scalar
 424      data-ref (array/pointer/struct access). A GIMPLE stmt is expected to have
 425      at most one such data-ref.  **/
 426
 427   /* Information about the data-ref (access function, etc),
 428      relative to the inner-most containing loop.  */
 429   struct data_reference *data_ref_info;
 430
 431   /* Information about the data-ref relative to this loop
 432      nest (the loop that is being considered for vectorization).  */
 433   tree dr_base_address;
 434   tree dr_init;
 435   tree dr_offset;
 436   tree dr_step;
 437   tree dr_aligned_to;
 438
 439   /* Stmt is part of some pattern (computation idiom)  */
 440   bool in_pattern_p;
 441
 442   /* Used for various bookkeeping purposes, generally holding a pointer to
 443      some other stmt S that is in some way "related" to this stmt.
 444      Current use of this field is:
 445         If this stmt is part of a pattern (i.e. the field 'in_pattern_p' is
 446         true): S is the "pattern stmt" that represents (and replaces) the
 447         sequence of stmts that constitutes the pattern.  Similarly, the
 448         related_stmt of the "pattern stmt" points back to this stmt (which is
 449         the last stmt in the original sequence of stmts that constitutes the
 450         pattern).  */
 451   gimple related_stmt;
 452
 453   /* List of datarefs that are known to have the same alignment as the dataref
 454      of this stmt.  */
 455   VEC(dr_p,heap) *same_align_refs;
 456
 457   /* Classify the def of this stmt.  */
 458   enum vect_def_type def_type;
 459
 460   /* Interleaving info.  */
 461   /* First data-ref in the interleaving group.  */
 462   gimple first_dr;
 463   /* Pointer to the next data-ref in the group.  */
 464   gimple next_dr;
 465   /* The size of the interleaving group.  */
 466   unsigned int size;
 467   /* For stores, number of stores from this group seen. We vectorize the last
 468      one.  */
 469   unsigned int store_count;
 470   /* For loads only, the gap from the previous load. For consecutive loads, GAP
 471      is 1.  */
 472   unsigned int gap;
 473   /* In case that two or more stmts share data-ref, this is the pointer to the
 474      previously detected stmt with the same dr.  */
 475   gimple same_dr_stmt;
 476   /* For loads only, if there is a store with the same location, this field is
 477      TRUE.  */
 478   bool read_write_dep;
 479
 480   /* Vectorization costs associated with statement.  */
 481   struct
 482   {
 483     int outside_of_loop;     /* Statements generated outside loop.  */
 484     int inside_of_loop;      /* Statements generated inside loop.  */
 485   } cost;
 486
 487   /*  Whether the stmt is SLPed, loop-based vectorized, or both.  */
 488   enum slp_vect_type slp_type;
 489
 490   /* The bb_vec_info with respect to which STMT is vectorized.  */
 491   bb_vec_info bb_vinfo;
 492
 493   /* Is this statement vectorizable or should it be skipped in (partial)
 494      vectorization.  */
 495   bool vectorizable;
 496 } *stmt_vec_info;
 497
 498 /* Access Functions.  */
 499 #define STMT_VINFO_TYPE(S)                 (S)->type
 500 #define STMT_VINFO_STMT(S)                 (S)->stmt
 501 #define STMT_VINFO_LOOP_VINFO(S)           (S)->loop_vinfo
 502 #define STMT_VINFO_BB_VINFO(S)             (S)->bb_vinfo
 503 #define STMT_VINFO_RELEVANT(S)             (S)->relevant
 504 #define STMT_VINFO_LIVE_P(S)               (S)->live
 505 #define STMT_VINFO_VECTYPE(S)              (S)->vectype
 506 #define STMT_VINFO_VEC_STMT(S)             (S)->vectorized_stmt
 507 #define STMT_VINFO_VECTORIZABLE(S)         (S)->vectorizable
 508 #define STMT_VINFO_DATA_REF(S)             (S)->data_ref_info
 509
 510 #define STMT_VINFO_DR_BASE_ADDRESS(S)      (S)->dr_base_address
 511 #define STMT_VINFO_DR_INIT(S)              (S)->dr_init
 512 #define STMT_VINFO_DR_OFFSET(S)            (S)->dr_offset
 513 #define STMT_VINFO_DR_STEP(S)              (S)->dr_step
 514 #define STMT_VINFO_DR_ALIGNED_TO(S)        (S)->dr_aligned_to
 515
 516 #define STMT_VINFO_IN_PATTERN_P(S)         (S)->in_pattern_p
 517 #define STMT_VINFO_RELATED_STMT(S)         (S)->related_stmt
 518 #define STMT_VINFO_SAME_ALIGN_REFS(S)      (S)->same_align_refs
 519 #define STMT_VINFO_DEF_TYPE(S)             (S)->def_type
 520 #define STMT_VINFO_DR_GROUP_FIRST_DR(S)    (S)->first_dr
 521 #define STMT_VINFO_DR_GROUP_NEXT_DR(S)     (S)->next_dr
 522 #define STMT_VINFO_DR_GROUP_SIZE(S)        (S)->size
 523 #define STMT_VINFO_DR_GROUP_STORE_COUNT(S) (S)->store_count
 524 #define STMT_VINFO_DR_GROUP_GAP(S)         (S)->gap
 525 #define STMT_VINFO_DR_GROUP_SAME_DR_STMT(S)(S)->same_dr_stmt
 526 #define STMT_VINFO_DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 527 #define STMT_VINFO_STRIDED_ACCESS(S)      ((S)->first_dr != NULL)
 528
 529 #define DR_GROUP_FIRST_DR(S)               (S)->first_dr
 530 #define DR_GROUP_NEXT_DR(S)                (S)->next_dr
 531 #define DR_GROUP_SIZE(S)                   (S)->size
 532 #define DR_GROUP_STORE_COUNT(S)            (S)->store_count
 533 #define DR_GROUP_GAP(S)                    (S)->gap
 534 #define DR_GROUP_SAME_DR_STMT(S)           (S)->same_dr_stmt
 535 #define DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 536
 537 #define STMT_VINFO_RELEVANT_P(S)          ((S)->relevant != vect_unused_in_scope)
 538 #define STMT_VINFO_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
 539 #define STMT_VINFO_INSIDE_OF_LOOP_COST(S)  (S)->cost.inside_of_loop
 540
 541 #define HYBRID_SLP_STMT(S)                ((S)->slp_type == hybrid)
 542 #define PURE_SLP_STMT(S)                  ((S)->slp_type == pure_slp)
 543 #define STMT_SLP_TYPE(S)                   (S)->slp_type
 544
 545 /* These are some defines for the initial implementation of the vectorizer's
 546    cost model.  These will later be target specific hooks.  */
 547
 548 /* Cost of conditional taken branch.  */
 549 #ifndef TARG_COND_TAKEN_BRANCH_COST
 550 #define TARG_COND_TAKEN_BRANCH_COST        3
 551 #endif
 552
 553 /* Cost of conditional not taken branch.  */
 554 #ifndef TARG_COND_NOT_TAKEN_BRANCH_COST
 555 #define TARG_COND_NOT_TAKEN_BRANCH_COST        1
 556 #endif
 557
 558 /* Cost of any scalar operation, excluding load and store.  */
 559 #ifndef TARG_SCALAR_STMT_COST
 560 #define TARG_SCALAR_STMT_COST           1
 561 #endif
 562
 563 /* Cost of scalar load.  */
 564 #ifndef TARG_SCALAR_LOAD_COST
 565 #define TARG_SCALAR_LOAD_COST           1
 566 #endif
 567
 568 /* Cost of scalar store.  */
 569 #ifndef TARG_SCALAR_STORE_COST
 570 #define TARG_SCALAR_STORE_COST           1
 571 #endif
 572
 573 /* Cost of any vector operation, excluding load, store or vector to scalar
 574    operation.  */
 575 #ifndef TARG_VEC_STMT_COST
 576 #define TARG_VEC_STMT_COST           1
 577 #endif
 578
 579 /* Cost of vector to scalar operation.  */
 580 #ifndef TARG_VEC_TO_SCALAR_COST
 581 #define TARG_VEC_TO_SCALAR_COST      1
 582 #endif
 583
 584 /* Cost of scalar to vector operation.  */
 585 #ifndef TARG_SCALAR_TO_VEC_COST
 586 #define TARG_SCALAR_TO_VEC_COST      1
 587 #endif
 588
 589 /* Cost of aligned vector load.  */
 590 #ifndef TARG_VEC_LOAD_COST
 591 #define TARG_VEC_LOAD_COST           1
 592 #endif
 593
 594 /* Cost of misaligned vector load.  */
 595 #ifndef TARG_VEC_UNALIGNED_LOAD_COST
 596 #define TARG_VEC_UNALIGNED_LOAD_COST 2
 597 #endif
 598
 599 /* Cost of vector store.  */
 600 #ifndef TARG_VEC_STORE_COST
 601 #define TARG_VEC_STORE_COST          1
 602 #endif
 603
 604 /* Cost of vector permutation.  */
 605 #ifndef TARG_VEC_PERMUTE_COST
 606 #define TARG_VEC_PERMUTE_COST          1
 607 #endif
 608
 609 /* The maximum number of intermediate steps required in multi-step type
 610    conversion.  */
 611 #define MAX_INTERM_CVT_STEPS         3
 612
 613 /* The maximum vectorization factor supported by any target (V32QI).  */
 614 #define MAX_VECTORIZATION_FACTOR 32
 615
 616 /* Avoid GTY(()) on stmt_vec_info.  */
 617 typedef void *vec_void_p;
 618 DEF_VEC_P (vec_void_p);
 619 DEF_VEC_ALLOC_P (vec_void_p, heap);
 620
 621 extern VEC(vec_void_p,heap) *stmt_vec_info_vec;
 622
 623 void init_stmt_vec_info_vec (void);
 624 void free_stmt_vec_info_vec (void);
 625
 626 static inline stmt_vec_info
 627 vinfo_for_stmt (gimple stmt)
 628 {
 629   unsigned int uid = gimple_uid (stmt);
 630   if (uid == 0)
 631     return NULL;
 632
 633   gcc_assert (uid <= VEC_length (vec_void_p, stmt_vec_info_vec));
 634   return (stmt_vec_info) VEC_index (vec_void_p, stmt_vec_info_vec, uid - 1);
 635 }
 636
 637 static inline void
 638 set_vinfo_for_stmt (gimple stmt, stmt_vec_info info)
 639 {
 640   unsigned int uid = gimple_uid (stmt);
 641   if (uid == 0)
 642     {
 643       gcc_assert (info);
 644       uid = VEC_length (vec_void_p, stmt_vec_info_vec) + 1;
 645       gimple_set_uid (stmt, uid);
 646       VEC_safe_push (vec_void_p, heap, stmt_vec_info_vec, (vec_void_p) info);
 647     }
 648   else
 649     VEC_replace (vec_void_p, stmt_vec_info_vec, uid - 1, (vec_void_p) info);
 650 }
 651
 652 static inline gimple
 653 get_earlier_stmt (gimple stmt1, gimple stmt2)
 654 {
 655   unsigned int uid1, uid2;
 656
 657   if (stmt1 == NULL)
 658     return stmt2;
 659
 660   if (stmt2 == NULL)
 661     return stmt1;
 662
 663   uid1 = gimple_uid (stmt1);
 664   uid2 = gimple_uid (stmt2);
 665
 666   if (uid1 == 0 || uid2 == 0)
 667     return NULL;
 668
 669   gcc_assert (uid1 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 670   gcc_assert (uid2 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 671
 672   if (uid1 < uid2)
 673     return stmt1;
 674   else
 675     return stmt2;
 676 }
 677
 678 static inline bool
 679 is_pattern_stmt_p (stmt_vec_info stmt_info)
 680 {
 681   gimple related_stmt;
 682   stmt_vec_info related_stmt_info;
 683
 684   related_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
 685   if (related_stmt
 686       && (related_stmt_info = vinfo_for_stmt (related_stmt))
 687       && STMT_VINFO_IN_PATTERN_P (related_stmt_info))
 688     return true;
 689
 690   return false;
 691 }
 692
 693 static inline bool
 694 is_loop_header_bb_p (basic_block bb)
 695 {
 696   if (bb == (bb->loop_father)->header)
 697     return true;
 698   gcc_assert (EDGE_COUNT (bb->preds) == 1);
 699   return false;
 700 }
 701
 702 static inline void
 703 stmt_vinfo_set_inside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 704                                     int cost)
 705 {
 706   if (slp_node)
 707     SLP_TREE_INSIDE_OF_LOOP_COST (slp_node) = cost;
 708   else
 709     STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) = cost;
 710 }
 711
 712 static inline void
 713 stmt_vinfo_set_outside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 714                                      int cost)
 715 {
 716   if (slp_node)
 717     SLP_TREE_OUTSIDE_OF_LOOP_COST (slp_node) = cost;
 718   else
 719     STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = cost;
 720 }
 721
 722 static inline int
 723 vect_pow2 (int x)
 724 {
 725   int i, res = 1;
 726
 727   for (i = 0; i < x; i++)
 728     res *= 2;
 729
 730   return res;
 731 }
 732
 733 /*-----------------------------------------------------------------*/
 734 /* Info on data references alignment.                              */
 735 /*-----------------------------------------------------------------*/
 736
 737 /* Reflects actual alignment of first access in the vectorized loop,
 738    taking into account peeling/versioning if applied.  */
 739 #define DR_MISALIGNMENT(DR)   ((int) (size_t) (DR)->aux)
 740 #define SET_DR_MISALIGNMENT(DR, VAL)   ((DR)->aux = (void *) (size_t) (VAL))
 741
 742 static inline bool
 743 aligned_access_p (struct data_reference *data_ref_info)
 744 {
 745   return (DR_MISALIGNMENT (data_ref_info) == 0);
 746 }
 747
 748 static inline bool
 749 known_alignment_for_access_p (struct data_reference *data_ref_info)
 750 {
 751   return (DR_MISALIGNMENT (data_ref_info) != -1);
 752 }
 753
 754 /* vect_dump will be set to stderr or dump_file if exist.  */
 755 extern FILE *vect_dump;
 756 extern LOC vect_loop_location;
 757
 758 /*-----------------------------------------------------------------*/
 759 /* Function prototypes.                                            */
 760 /*-----------------------------------------------------------------*/
 761
 762 /* Simple loop peeling and versioning utilities for vectorizer's purposes -
 763    in tree-vect-loop-manip.c.  */
 764 extern void slpeel_make_loop_iterate_ntimes (struct loop *, tree);
 765 extern bool slpeel_can_duplicate_loop_p (const struct loop *, const_edge);
 766 extern void vect_loop_versioning (loop_vec_info, bool, tree *, gimple_seq *);
 767 extern void vect_do_peeling_for_loop_bound (loop_vec_info, tree *,
 768                                             tree, gimple_seq);
 769 extern void vect_do_peeling_for_alignment (loop_vec_info);
 770 extern LOC find_loop_location (struct loop *);
 771 extern bool vect_can_advance_ivs_p (loop_vec_info);
 772
 773 /* In tree-vect-stmts.c.  */
 774 extern tree get_vectype_for_scalar_type (tree);
 775 extern tree get_same_sized_vectype (tree, tree);
 776 extern bool vect_is_simple_use (tree, loop_vec_info, bb_vec_info, gimple *,
 777                                 tree *,  enum vect_def_type *);
 778 extern bool vect_is_simple_use_1 (tree, loop_vec_info, bb_vec_info, gimple *,
 779                                   tree *,  enum vect_def_type *, tree *);
 780 extern bool supportable_widening_operation (enum tree_code, gimple, tree, tree,
 781                                             tree *, tree *, enum tree_code *,
 782                                             enum tree_code *, int *,
 783                                             VEC (tree, heap) **);
 784 extern bool supportable_narrowing_operation (enum tree_code, tree, tree,
 785                                              enum tree_code *,
 786                                              int *, VEC (tree, heap) **);
 787 extern stmt_vec_info new_stmt_vec_info (gimple stmt, loop_vec_info,
 788                                         bb_vec_info);
 789 extern void free_stmt_vec_info (gimple stmt);
 790 extern tree vectorizable_function (gimple, tree, tree);
 791 extern void vect_model_simple_cost (stmt_vec_info, int, enum vect_def_type *,
 792                                     slp_tree);
 793 extern void vect_model_store_cost (stmt_vec_info, int, enum vect_def_type,
 794                                    slp_tree);
 795 extern void vect_model_load_cost (stmt_vec_info, int, slp_tree);
 796 extern void vect_finish_stmt_generation (gimple, gimple,
 797                                          gimple_stmt_iterator *);
 798 extern bool vect_mark_stmts_to_be_vectorized (loop_vec_info);
 799 extern int cost_for_stmt (gimple);
 800 extern tree vect_get_vec_def_for_operand (tree, gimple, tree *);
 801 extern tree vect_init_vector (gimple, tree, tree,
 802                               gimple_stmt_iterator *);
 803 extern tree vect_get_vec_def_for_stmt_copy (enum vect_def_type, tree);
 804 extern bool vect_transform_stmt (gimple, gimple_stmt_iterator *,
 805                                  bool *, slp_tree, slp_instance);
 806 extern void vect_remove_stores (gimple);
 807 extern bool vect_analyze_stmt (gimple, bool *, slp_tree);
 808 extern bool vectorizable_condition (gimple, gimple_stmt_iterator *, gimple *,
 809                                     tree, int);
 810
 811 /* In tree-vect-data-refs.c.  */
 812 extern bool vect_can_force_dr_alignment_p (const_tree, unsigned int);
 813 extern enum dr_alignment_support vect_supportable_dr_alignment
 814                                            (struct data_reference *);
 815 extern tree vect_get_smallest_scalar_type (gimple, HOST_WIDE_INT *,
 816                                            HOST_WIDE_INT *);
 817 extern bool vect_analyze_data_ref_dependences (loop_vec_info, bb_vec_info,
 818                                                int *);
 819 extern bool vect_enhance_data_refs_alignment (loop_vec_info);
 820 extern bool vect_analyze_data_refs_alignment (loop_vec_info, bb_vec_info);
 821 extern bool vect_verify_datarefs_alignment (loop_vec_info, bb_vec_info);
 822 extern bool vect_analyze_data_ref_accesses (loop_vec_info, bb_vec_info);
 823 extern bool vect_prune_runtime_alias_test_list (loop_vec_info);
 824 extern bool vect_analyze_data_refs (loop_vec_info, bb_vec_info, int *);
 825 extern tree vect_create_data_ref_ptr (gimple, struct loop *, tree, tree *,
 826                                       gimple *, bool, bool *);
 827 extern tree bump_vector_ptr (tree, gimple, gimple_stmt_iterator *, gimple, tree);
 828 extern tree vect_create_destination_var (tree, tree);
 829 extern bool vect_strided_store_supported (tree);
 830 extern bool vect_strided_load_supported (tree);
 831 extern bool vect_permute_store_chain (VEC(tree,heap) *,unsigned int, gimple,
 832                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 833 extern tree vect_setup_realignment (gimple, gimple_stmt_iterator *, tree *,
 834                                     enum dr_alignment_support, tree,
 835                                     struct loop **);
 836 extern bool vect_permute_load_chain (VEC(tree,heap) *,unsigned int, gimple,
 837                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 838 extern bool vect_transform_strided_load (gimple, VEC(tree,heap) *, int,
 839                                          gimple_stmt_iterator *);
 840 extern int vect_get_place_in_interleaving_chain (gimple, gimple);
 841 extern tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
 842 extern tree vect_create_addr_base_for_vector_ref (gimple, gimple_seq *,
 843                                                   tree, struct loop *);
 844
 845 /* In tree-vect-loop.c.  */
 846 /* FORNOW: Used in tree-parloops.c.  */
 847 extern void destroy_loop_vec_info (loop_vec_info, bool);
 848 extern gimple vect_is_simple_reduction (loop_vec_info, gimple, bool, bool *);
 849 /* Drive for loop analysis stage.  */
 850 extern loop_vec_info vect_analyze_loop (struct loop *);
 851 /* Drive for loop transformation stage.  */
 852 extern void vect_transform_loop (loop_vec_info);
 853 extern loop_vec_info vect_analyze_loop_form (struct loop *);
 854 extern bool vectorizable_live_operation (gimple, gimple_stmt_iterator *,
 855                                          gimple *);
 856 extern bool vectorizable_reduction (gimple, gimple_stmt_iterator *, gimple *,
 857                                     slp_tree);
 858 extern bool vectorizable_induction (gimple, gimple_stmt_iterator *, gimple *);
 859 extern int vect_estimate_min_profitable_iters (loop_vec_info);
 860 extern tree get_initial_def_for_reduction (gimple, tree, tree *);
 861 extern int vect_min_worthwhile_factor (enum tree_code);
 862
 863
 864 /* In tree-vect-slp.c.  */
 865 extern void vect_free_slp_instance (slp_instance);
 866 extern bool vect_transform_slp_perm_load (gimple, VEC (tree, heap) *,
 867                                           gimple_stmt_iterator *, int,
 868                                           slp_instance, bool);
 869 extern bool vect_schedule_slp (loop_vec_info, bb_vec_info);
 870 extern void vect_update_slp_costs_according_to_vf (loop_vec_info);
 871 extern bool vect_analyze_slp (loop_vec_info, bb_vec_info);
 872 extern void vect_make_slp_decision (loop_vec_info);
 873 extern void vect_detect_hybrid_slp (loop_vec_info);
 874 extern void vect_get_slp_defs (slp_tree, VEC (tree,heap) **,
 875                                VEC (tree,heap) **, int);
 876 extern LOC find_bb_location (basic_block);
 877 extern bb_vec_info vect_slp_analyze_bb (basic_block);
 878 extern void vect_slp_transform_bb (basic_block);
 879
 880 /* In tree-vect-patterns.c.  */
 881 /* Pattern recognition functions.
 882    Additional pattern recognition functions can (and will) be added
 883    in the future.  */
 884 typedef gimple (* vect_recog_func_ptr) (gimple, tree *, tree *);
 885 #define NUM_PATTERNS 4
 886 void vect_pattern_recog (loop_vec_info);
 887
 888 /* In tree-vectorizer.c.  */
 889 unsigned vectorize_loops (void);
 890 /* Vectorization debug information */
 891 extern bool vect_print_dump_info (enum verbosity_levels);
 892
 893 #endif  /* GCC_TREE_VECTORIZER_H  */