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 typedef struct _vect_peel_info
 175 {
 176   int npeel;
 177   struct data_reference *dr;
 178   unsigned int count;
 179 } *vect_peel_info;
 180
 181 typedef struct _vect_peel_extended_info
 182 {
 183   struct _vect_peel_info peel_info;
 184   unsigned int inside_cost;
 185   unsigned int outside_cost;
 186 } *vect_peel_extended_info;
 187
 188 /*-----------------------------------------------------------------*/
 189 /* Info on vectorized loops.                                       */
 190 /*-----------------------------------------------------------------*/
 191 typedef struct _loop_vec_info {
 192
 193   /* The loop to which this info struct refers to.  */
 194   struct loop *loop;
 195
 196   /* The loop basic blocks.  */
 197   basic_block *bbs;
 198
 199   /* Number of iterations.  */
 200   tree num_iters;
 201   tree num_iters_unchanged;
 202
 203   /* Minimum number of iterations below which vectorization is expected to
 204      not be profitable (as estimated by the cost model).
 205      -1 indicates that vectorization will not be profitable.
 206      FORNOW: This field is an int. Will be a tree in the future, to represent
 207              values unknown at compile time.  */
 208   int min_profitable_iters;
 209
 210   /* Is the loop vectorizable? */
 211   bool vectorizable;
 212
 213   /* Unrolling factor  */
 214   int vectorization_factor;
 215
 216   /* The loop location in the source.  */
 217   LOC loop_line_number;
 218
 219   /* Unknown DRs according to which loop was peeled.  */
 220   struct data_reference *unaligned_dr;
 221
 222   /* peeling_for_alignment indicates whether peeling for alignment will take
 223      place, and what the peeling factor should be:
 224      peeling_for_alignment = X means:
 225         If X=0: Peeling for alignment will not be applied.
 226         If X>0: Peel first X iterations.
 227         If X=-1: Generate a runtime test to calculate the number of iterations
 228                  to be peeled, using the dataref recorded in the field
 229                  unaligned_dr.  */
 230   int peeling_for_alignment;
 231
 232   /* The mask used to check the alignment of pointers or arrays.  */
 233   int ptr_mask;
 234
 235   /* All data references in the loop.  */
 236   VEC (data_reference_p, heap) *datarefs;
 237
 238   /* All data dependences in the loop.  */
 239   VEC (ddr_p, heap) *ddrs;
 240
 241   /* Data Dependence Relations defining address ranges that are candidates
 242      for a run-time aliasing check.  */
 243   VEC (ddr_p, heap) *may_alias_ddrs;
 244
 245   /* Statements in the loop that have data references that are candidates for a
 246      runtime (loop versioning) misalignment check.  */
 247   VEC(gimple,heap) *may_misalign_stmts;
 248
 249   /* All interleaving chains of stores in the loop, represented by the first
 250      stmt in the chain.  */
 251   VEC(gimple, heap) *strided_stores;
 252
 253   /* All SLP instances in the loop. This is a subset of the set of STRIDED_STORES
 254      of the loop.  */
 255   VEC(slp_instance, heap) *slp_instances;
 256
 257   /* The unrolling factor needed to SLP the loop. In case of that pure SLP is
 258      applied to the loop, i.e., no unrolling is needed, this is 1.  */
 259   unsigned slp_unrolling_factor;
 260
 261   /* Reduction cycles detected in the loop. Used in loop-aware SLP.  */
 262   VEC (gimple, heap) *reductions;
 263
 264   /* Hash table used to choose the best peeling option.  */
 265   htab_t peeling_htab;
 266
 267 } *loop_vec_info;
 268
 269 /* Access Functions.  */
 270 #define LOOP_VINFO_LOOP(L)                 (L)->loop
 271 #define LOOP_VINFO_BBS(L)                  (L)->bbs
 272 #define LOOP_VINFO_NITERS(L)               (L)->num_iters
 273 /* Since LOOP_VINFO_NITERS can change after prologue peeling
 274    retain total unchanged scalar loop iterations for cost model.  */
 275 #define LOOP_VINFO_NITERS_UNCHANGED(L)     (L)->num_iters_unchanged
 276 #define LOOP_VINFO_COST_MODEL_MIN_ITERS(L) (L)->min_profitable_iters
 277 #define LOOP_VINFO_VECTORIZABLE_P(L)       (L)->vectorizable
 278 #define LOOP_VINFO_VECT_FACTOR(L)          (L)->vectorization_factor
 279 #define LOOP_VINFO_PTR_MASK(L)             (L)->ptr_mask
 280 #define LOOP_VINFO_DATAREFS(L)             (L)->datarefs
 281 #define LOOP_VINFO_DDRS(L)                 (L)->ddrs
 282 #define LOOP_VINFO_INT_NITERS(L)           (TREE_INT_CST_LOW ((L)->num_iters))
 283 #define LOOP_PEELING_FOR_ALIGNMENT(L)      (L)->peeling_for_alignment
 284 #define LOOP_VINFO_UNALIGNED_DR(L)         (L)->unaligned_dr
 285 #define LOOP_VINFO_MAY_MISALIGN_STMTS(L)   (L)->may_misalign_stmts
 286 #define LOOP_VINFO_LOC(L)                  (L)->loop_line_number
 287 #define LOOP_VINFO_MAY_ALIAS_DDRS(L)       (L)->may_alias_ddrs
 288 #define LOOP_VINFO_STRIDED_STORES(L)       (L)->strided_stores
 289 #define LOOP_VINFO_SLP_INSTANCES(L)        (L)->slp_instances
 290 #define LOOP_VINFO_SLP_UNROLLING_FACTOR(L) (L)->slp_unrolling_factor
 291 #define LOOP_VINFO_REDUCTIONS(L)           (L)->reductions
 292 #define LOOP_VINFO_PEELING_HTAB(L)         (L)->peeling_htab
 293
 294 #define LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT(L) \
 295 VEC_length (gimple, (L)->may_misalign_stmts) > 0
 296 #define LOOP_REQUIRES_VERSIONING_FOR_ALIAS(L)     \
 297 VEC_length (ddr_p, (L)->may_alias_ddrs) > 0
 298
 299 #define NITERS_KNOWN_P(n)                     \
 300 (host_integerp ((n),0)                        \
 301 && TREE_INT_CST_LOW ((n)) > 0)
 302
 303 #define LOOP_VINFO_NITERS_KNOWN_P(L)          \
 304 NITERS_KNOWN_P((L)->num_iters)
 305
 306 static inline loop_vec_info
 307 loop_vec_info_for_loop (struct loop *loop)
 308 {
 309   return (loop_vec_info) loop->aux;
 310 }
 311
 312 static inline bool
 313 nested_in_vect_loop_p (struct loop *loop, gimple stmt)
 314 {
 315   return (loop->inner
 316           && (loop->inner == (gimple_bb (stmt))->loop_father));
 317 }
 318
 319 typedef struct _bb_vec_info {
 320
 321   basic_block bb;
 322   /* All interleaving chains of stores in the basic block, represented by the
 323      first stmt in the chain.  */
 324   VEC(gimple, heap) *strided_stores;
 325
 326   /* All SLP instances in the basic block. This is a subset of the set of
 327      STRIDED_STORES of the basic block.  */
 328   VEC(slp_instance, heap) *slp_instances;
 329
 330   /* All data references in the basic block.  */
 331   VEC (data_reference_p, heap) *datarefs;
 332
 333   /* All data dependences in the basic block.  */
 334   VEC (ddr_p, heap) *ddrs;
 335 } *bb_vec_info;
 336
 337 #define BB_VINFO_BB(B)              (B)->bb
 338 #define BB_VINFO_STRIDED_STORES(B)  (B)->strided_stores
 339 #define BB_VINFO_SLP_INSTANCES(B)   (B)->slp_instances
 340 #define BB_VINFO_DATAREFS(B)        (B)->datarefs
 341 #define BB_VINFO_DDRS(B)            (B)->ddrs
 342
 343 static inline bb_vec_info
 344 vec_info_for_bb (basic_block bb)
 345 {
 346   return (bb_vec_info) bb->aux;
 347 }
 348
 349 /*-----------------------------------------------------------------*/
 350 /* Info on vectorized defs.                                        */
 351 /*-----------------------------------------------------------------*/
 352 enum stmt_vec_info_type {
 353   undef_vec_info_type = 0,
 354   load_vec_info_type,
 355   store_vec_info_type,
 356   op_vec_info_type,
 357   call_vec_info_type,
 358   assignment_vec_info_type,
 359   condition_vec_info_type,
 360   reduc_vec_info_type,
 361   induc_vec_info_type,
 362   type_promotion_vec_info_type,
 363   type_demotion_vec_info_type,
 364   type_conversion_vec_info_type,
 365   loop_exit_ctrl_vec_info_type
 366 };
 367
 368 /* Indicates whether/how a variable is used in the scope of loop/basic
 369    block.  */
 370 enum vect_relevant {
 371   vect_unused_in_scope = 0,
 372   /* The def is in the inner loop, and the use is in the outer loop, and the
 373      use is a reduction stmt.  */
 374   vect_used_in_outer_by_reduction,
 375   /* The def is in the inner loop, and the use is in the outer loop (and is
 376      not part of reduction).  */
 377   vect_used_in_outer,
 378
 379   /* defs that feed computations that end up (only) in a reduction. These
 380      defs may be used by non-reduction stmts, but eventually, any
 381      computations/values that are affected by these defs are used to compute
 382      a reduction (i.e. don't get stored to memory, for example). We use this
 383      to identify computations that we can change the order in which they are
 384      computed.  */
 385   vect_used_by_reduction,
 386
 387   vect_used_in_scope
 388 };
 389
 390 /* The type of vectorization that can be applied to the stmt: regular loop-based
 391    vectorization; pure SLP - the stmt is a part of SLP instances and does not
 392    have uses outside SLP instances; or hybrid SLP and loop-based - the stmt is
 393    a part of SLP instance and also must be loop-based vectorized, since it has
 394    uses outside SLP sequences.
 395
 396    In the loop context the meanings of pure and hybrid SLP are slightly
 397    different. By saying that pure SLP is applied to the loop, we mean that we
 398    exploit only intra-iteration parallelism in the loop; i.e., the loop can be
 399    vectorized without doing any conceptual unrolling, cause we don't pack
 400    together stmts from different iterations, only within a single iteration.
 401    Loop hybrid SLP means that we exploit both intra-iteration and
 402    inter-iteration parallelism (e.g., number of elements in the vector is 4
 403    and the slp-group-size is 2, in which case we don't have enough parallelism
 404    within an iteration, so we obtain the rest of the parallelism from subsequent
 405    iterations by unrolling the loop by 2).  */
 406 enum slp_vect_type {
 407   loop_vect = 0,
 408   pure_slp,
 409   hybrid
 410 };
 411
 412
 413 typedef struct data_reference *dr_p;
 414 DEF_VEC_P(dr_p);
 415 DEF_VEC_ALLOC_P(dr_p,heap);
 416
 417 typedef struct _stmt_vec_info {
 418
 419   enum stmt_vec_info_type type;
 420
 421   /* Indicates whether this stmts is part of a computation whose result is
 422      used outside the loop.  */
 423   bool live;
 424
 425   /* Stmt is part of some pattern (computation idiom)  */
 426   bool in_pattern_p;
 427
 428   /* For loads only, if there is a store with the same location, this field is
 429      TRUE.  */
 430   bool read_write_dep;
 431
 432   /* The stmt to which this info struct refers to.  */
 433   gimple stmt;
 434
 435   /* The loop_vec_info with respect to which STMT is vectorized.  */
 436   loop_vec_info loop_vinfo;
 437
 438   /* The vector type to be used for the LHS of this statement.  */
 439   tree vectype;
 440
 441   /* The vectorized version of the stmt.  */
 442   gimple vectorized_stmt;
 443
 444
 445   /** The following is relevant only for stmts that contain a non-scalar
 446      data-ref (array/pointer/struct access). A GIMPLE stmt is expected to have
 447      at most one such data-ref.  **/
 448
 449   /* Information about the data-ref (access function, etc),
 450      relative to the inner-most containing loop.  */
 451   struct data_reference *data_ref_info;
 452
 453   /* Information about the data-ref relative to this loop
 454      nest (the loop that is being considered for vectorization).  */
 455   tree dr_base_address;
 456   tree dr_init;
 457   tree dr_offset;
 458   tree dr_step;
 459   tree dr_aligned_to;
 460
 461   /* Used for various bookkeeping purposes, generally holding a pointer to
 462      some other stmt S that is in some way "related" to this stmt.
 463      Current use of this field is:
 464         If this stmt is part of a pattern (i.e. the field 'in_pattern_p' is
 465         true): S is the "pattern stmt" that represents (and replaces) the
 466         sequence of stmts that constitutes the pattern.  Similarly, the
 467         related_stmt of the "pattern stmt" points back to this stmt (which is
 468         the last stmt in the original sequence of stmts that constitutes the
 469         pattern).  */
 470   gimple related_stmt;
 471
 472   /* List of datarefs that are known to have the same alignment as the dataref
 473      of this stmt.  */
 474   VEC(dr_p,heap) *same_align_refs;
 475
 476   /* Classify the def of this stmt.  */
 477   enum vect_def_type def_type;
 478
 479   /*  Whether the stmt is SLPed, loop-based vectorized, or both.  */
 480   enum slp_vect_type slp_type;
 481
 482   /* Interleaving info.  */
 483   /* First data-ref in the interleaving group.  */
 484   gimple first_dr;
 485   /* Pointer to the next data-ref in the group.  */
 486   gimple next_dr;
 487   /* In case that two or more stmts share data-ref, this is the pointer to the
 488      previously detected stmt with the same dr.  */
 489   gimple same_dr_stmt;
 490   /* The size of the interleaving group.  */
 491   unsigned int size;
 492   /* For stores, number of stores from this group seen. We vectorize the last
 493      one.  */
 494   unsigned int store_count;
 495   /* For loads only, the gap from the previous load. For consecutive loads, GAP
 496      is 1.  */
 497   unsigned int gap;
 498
 499   /* Not all stmts in the loop need to be vectorized. e.g, the increment
 500      of the loop induction variable and computation of array indexes. relevant
 501      indicates whether the stmt needs to be vectorized.  */
 502   enum vect_relevant relevant;
 503
 504   /* Vectorization costs associated with statement.  */
 505   struct
 506   {
 507     int outside_of_loop;     /* Statements generated outside loop.  */
 508     int inside_of_loop;      /* Statements generated inside loop.  */
 509   } cost;
 510
 511   /* The bb_vec_info with respect to which STMT is vectorized.  */
 512   bb_vec_info bb_vinfo;
 513
 514   /* Is this statement vectorizable or should it be skipped in (partial)
 515      vectorization.  */
 516   bool vectorizable;
 517 } *stmt_vec_info;
 518
 519 /* Access Functions.  */
 520 #define STMT_VINFO_TYPE(S)                 (S)->type
 521 #define STMT_VINFO_STMT(S)                 (S)->stmt
 522 #define STMT_VINFO_LOOP_VINFO(S)           (S)->loop_vinfo
 523 #define STMT_VINFO_BB_VINFO(S)             (S)->bb_vinfo
 524 #define STMT_VINFO_RELEVANT(S)             (S)->relevant
 525 #define STMT_VINFO_LIVE_P(S)               (S)->live
 526 #define STMT_VINFO_VECTYPE(S)              (S)->vectype
 527 #define STMT_VINFO_VEC_STMT(S)             (S)->vectorized_stmt
 528 #define STMT_VINFO_VECTORIZABLE(S)         (S)->vectorizable
 529 #define STMT_VINFO_DATA_REF(S)             (S)->data_ref_info
 530
 531 #define STMT_VINFO_DR_BASE_ADDRESS(S)      (S)->dr_base_address
 532 #define STMT_VINFO_DR_INIT(S)              (S)->dr_init
 533 #define STMT_VINFO_DR_OFFSET(S)            (S)->dr_offset
 534 #define STMT_VINFO_DR_STEP(S)              (S)->dr_step
 535 #define STMT_VINFO_DR_ALIGNED_TO(S)        (S)->dr_aligned_to
 536
 537 #define STMT_VINFO_IN_PATTERN_P(S)         (S)->in_pattern_p
 538 #define STMT_VINFO_RELATED_STMT(S)         (S)->related_stmt
 539 #define STMT_VINFO_SAME_ALIGN_REFS(S)      (S)->same_align_refs
 540 #define STMT_VINFO_DEF_TYPE(S)             (S)->def_type
 541 #define STMT_VINFO_DR_GROUP_FIRST_DR(S)    (S)->first_dr
 542 #define STMT_VINFO_DR_GROUP_NEXT_DR(S)     (S)->next_dr
 543 #define STMT_VINFO_DR_GROUP_SIZE(S)        (S)->size
 544 #define STMT_VINFO_DR_GROUP_STORE_COUNT(S) (S)->store_count
 545 #define STMT_VINFO_DR_GROUP_GAP(S)         (S)->gap
 546 #define STMT_VINFO_DR_GROUP_SAME_DR_STMT(S)(S)->same_dr_stmt
 547 #define STMT_VINFO_DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 548 #define STMT_VINFO_STRIDED_ACCESS(S)      ((S)->first_dr != NULL)
 549
 550 #define DR_GROUP_FIRST_DR(S)               (S)->first_dr
 551 #define DR_GROUP_NEXT_DR(S)                (S)->next_dr
 552 #define DR_GROUP_SIZE(S)                   (S)->size
 553 #define DR_GROUP_STORE_COUNT(S)            (S)->store_count
 554 #define DR_GROUP_GAP(S)                    (S)->gap
 555 #define DR_GROUP_SAME_DR_STMT(S)           (S)->same_dr_stmt
 556 #define DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 557
 558 #define STMT_VINFO_RELEVANT_P(S)          ((S)->relevant != vect_unused_in_scope)
 559 #define STMT_VINFO_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
 560 #define STMT_VINFO_INSIDE_OF_LOOP_COST(S)  (S)->cost.inside_of_loop
 561
 562 #define HYBRID_SLP_STMT(S)                ((S)->slp_type == hybrid)
 563 #define PURE_SLP_STMT(S)                  ((S)->slp_type == pure_slp)
 564 #define STMT_SLP_TYPE(S)                   (S)->slp_type
 565
 566 #define VECT_MAX_COST 1000
 567
 568 /* The maximum number of intermediate steps required in multi-step type
 569    conversion.  */
 570 #define MAX_INTERM_CVT_STEPS         3
 571
 572 /* The maximum vectorization factor supported by any target (V32QI).  */
 573 #define MAX_VECTORIZATION_FACTOR 32
 574
 575 /* Avoid GTY(()) on stmt_vec_info.  */
 576 typedef void *vec_void_p;
 577 DEF_VEC_P (vec_void_p);
 578 DEF_VEC_ALLOC_P (vec_void_p, heap);
 579
 580 extern VEC(vec_void_p,heap) *stmt_vec_info_vec;
 581
 582 void init_stmt_vec_info_vec (void);
 583 void free_stmt_vec_info_vec (void);
 584
 585 static inline stmt_vec_info
 586 vinfo_for_stmt (gimple stmt)
 587 {
 588   unsigned int uid = gimple_uid (stmt);
 589   if (uid == 0)
 590     return NULL;
 591
 592   return (stmt_vec_info) VEC_index (vec_void_p, stmt_vec_info_vec, uid - 1);
 593 }
 594
 595 static inline void
 596 set_vinfo_for_stmt (gimple stmt, stmt_vec_info info)
 597 {
 598   unsigned int uid = gimple_uid (stmt);
 599   if (uid == 0)
 600     {
 601       gcc_checking_assert (info);
 602       uid = VEC_length (vec_void_p, stmt_vec_info_vec) + 1;
 603       gimple_set_uid (stmt, uid);
 604       VEC_safe_push (vec_void_p, heap, stmt_vec_info_vec, (vec_void_p) info);
 605     }
 606   else
 607     VEC_replace (vec_void_p, stmt_vec_info_vec, uid - 1, (vec_void_p) info);
 608 }
 609
 610 static inline gimple
 611 get_earlier_stmt (gimple stmt1, gimple stmt2)
 612 {
 613   unsigned int uid1, uid2;
 614
 615   if (stmt1 == NULL)
 616     return stmt2;
 617
 618   if (stmt2 == NULL)
 619     return stmt1;
 620
 621   uid1 = gimple_uid (stmt1);
 622   uid2 = gimple_uid (stmt2);
 623
 624   if (uid1 == 0 || uid2 == 0)
 625     return NULL;
 626
 627   gcc_checking_assert (uid1 <= VEC_length (vec_void_p, stmt_vec_info_vec)
 628                        && uid2 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 629
 630   if (uid1 < uid2)
 631     return stmt1;
 632   else
 633     return stmt2;
 634 }
 635
 636 static inline bool
 637 is_pattern_stmt_p (stmt_vec_info stmt_info)
 638 {
 639   gimple related_stmt;
 640   stmt_vec_info related_stmt_info;
 641
 642   related_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
 643   if (related_stmt
 644       && (related_stmt_info = vinfo_for_stmt (related_stmt))
 645       && STMT_VINFO_IN_PATTERN_P (related_stmt_info))
 646     return true;
 647
 648   return false;
 649 }
 650
 651 static inline bool
 652 is_loop_header_bb_p (basic_block bb)
 653 {
 654   if (bb == (bb->loop_father)->header)
 655     return true;
 656   gcc_checking_assert (EDGE_COUNT (bb->preds) == 1);
 657   return false;
 658 }
 659
 660 static inline void
 661 stmt_vinfo_set_inside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 662                                     int cost)
 663 {
 664   if (slp_node)
 665     SLP_TREE_INSIDE_OF_LOOP_COST (slp_node) = cost;
 666   else
 667     STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) = cost;
 668 }
 669
 670 static inline void
 671 stmt_vinfo_set_outside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 672                                      int cost)
 673 {
 674   if (slp_node)
 675     SLP_TREE_OUTSIDE_OF_LOOP_COST (slp_node) = cost;
 676   else
 677     STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = cost;
 678 }
 679
 680 static inline int
 681 vect_pow2 (int x)
 682 {
 683   int i, res = 1;
 684
 685   for (i = 0; i < x; i++)
 686     res *= 2;
 687
 688   return res;
 689 }
 690
 691 /*-----------------------------------------------------------------*/
 692 /* Info on data references alignment.                              */
 693 /*-----------------------------------------------------------------*/
 694
 695 /* Reflects actual alignment of first access in the vectorized loop,
 696    taking into account peeling/versioning if applied.  */
 697 #define DR_MISALIGNMENT(DR)   ((int) (size_t) (DR)->aux)
 698 #define SET_DR_MISALIGNMENT(DR, VAL)   ((DR)->aux = (void *) (size_t) (VAL))
 699
 700 static inline bool
 701 aligned_access_p (struct data_reference *data_ref_info)
 702 {
 703   return (DR_MISALIGNMENT (data_ref_info) == 0);
 704 }
 705
 706 static inline bool
 707 known_alignment_for_access_p (struct data_reference *data_ref_info)
 708 {
 709   return (DR_MISALIGNMENT (data_ref_info) != -1);
 710 }
 711
 712 /* vect_dump will be set to stderr or dump_file if exist.  */
 713 extern FILE *vect_dump;
 714 extern LOC vect_loop_location;
 715
 716 /*-----------------------------------------------------------------*/
 717 /* Function prototypes.                                            */
 718 /*-----------------------------------------------------------------*/
 719
 720 /* Simple loop peeling and versioning utilities for vectorizer's purposes -
 721    in tree-vect-loop-manip.c.  */
 722 extern void slpeel_make_loop_iterate_ntimes (struct loop *, tree);
 723 extern bool slpeel_can_duplicate_loop_p (const struct loop *, const_edge);
 724 extern void vect_loop_versioning (loop_vec_info, bool, tree *, gimple_seq *);
 725 extern void vect_do_peeling_for_loop_bound (loop_vec_info, tree *,
 726                                             tree, gimple_seq);
 727 extern void vect_do_peeling_for_alignment (loop_vec_info);
 728 extern LOC find_loop_location (struct loop *);
 729 extern bool vect_can_advance_ivs_p (loop_vec_info);
 730
 731 /* In tree-vect-stmts.c.  */
 732 extern tree get_vectype_for_scalar_type (tree);
 733 extern tree get_same_sized_vectype (tree, tree);
 734 extern bool vect_is_simple_use (tree, loop_vec_info, bb_vec_info, gimple *,
 735                                 tree *,  enum vect_def_type *);
 736 extern bool vect_is_simple_use_1 (tree, loop_vec_info, bb_vec_info, gimple *,
 737                                   tree *,  enum vect_def_type *, tree *);
 738 extern bool supportable_widening_operation (enum tree_code, gimple, tree, tree,
 739                                             tree *, tree *, enum tree_code *,
 740                                             enum tree_code *, int *,
 741                                             VEC (tree, heap) **);
 742 extern bool supportable_narrowing_operation (enum tree_code, tree, tree,
 743                                              enum tree_code *,
 744                                              int *, VEC (tree, heap) **);
 745 extern stmt_vec_info new_stmt_vec_info (gimple stmt, loop_vec_info,
 746                                         bb_vec_info);
 747 extern void free_stmt_vec_info (gimple stmt);
 748 extern tree vectorizable_function (gimple, tree, tree);
 749 extern void vect_model_simple_cost (stmt_vec_info, int, enum vect_def_type *,
 750                                     slp_tree);
 751 extern void vect_model_store_cost (stmt_vec_info, int, enum vect_def_type,
 752                                    slp_tree);
 753 extern void vect_model_load_cost (stmt_vec_info, int, slp_tree);
 754 extern void vect_finish_stmt_generation (gimple, gimple,
 755                                          gimple_stmt_iterator *);
 756 extern bool vect_mark_stmts_to_be_vectorized (loop_vec_info);
 757 extern int cost_for_stmt (gimple);
 758 extern tree vect_get_vec_def_for_operand (tree, gimple, tree *);
 759 extern tree vect_init_vector (gimple, tree, tree,
 760                               gimple_stmt_iterator *);
 761 extern tree vect_get_vec_def_for_stmt_copy (enum vect_def_type, tree);
 762 extern bool vect_transform_stmt (gimple, gimple_stmt_iterator *,
 763                                  bool *, slp_tree, slp_instance);
 764 extern void vect_remove_stores (gimple);
 765 extern bool vect_analyze_stmt (gimple, bool *, slp_tree);
 766 extern bool vectorizable_condition (gimple, gimple_stmt_iterator *, gimple *,
 767                                     tree, int);
 768 extern void vect_get_load_cost (struct data_reference *, int, bool,
 769                                 unsigned int *, unsigned int *);
 770 extern void vect_get_store_cost (struct data_reference *, int, unsigned int *);
 771
 772 /* In tree-vect-data-refs.c.  */
 773 extern bool vect_can_force_dr_alignment_p (const_tree, unsigned int);
 774 extern enum dr_alignment_support vect_supportable_dr_alignment
 775                                            (struct data_reference *, bool);
 776 extern tree vect_get_smallest_scalar_type (gimple, HOST_WIDE_INT *,
 777                                            HOST_WIDE_INT *);
 778 extern bool vect_analyze_data_ref_dependences (loop_vec_info, bb_vec_info,
 779                                                int *);
 780 extern bool vect_enhance_data_refs_alignment (loop_vec_info);
 781 extern bool vect_analyze_data_refs_alignment (loop_vec_info, bb_vec_info);
 782 extern bool vect_verify_datarefs_alignment (loop_vec_info, bb_vec_info);
 783 extern bool vect_analyze_data_ref_accesses (loop_vec_info, bb_vec_info);
 784 extern bool vect_prune_runtime_alias_test_list (loop_vec_info);
 785 extern bool vect_analyze_data_refs (loop_vec_info, bb_vec_info, int *);
 786 extern tree vect_create_data_ref_ptr (gimple, struct loop *, tree, tree *,
 787                                       gimple *, bool, bool *);
 788 extern tree bump_vector_ptr (tree, gimple, gimple_stmt_iterator *, gimple, tree);
 789 extern tree vect_create_destination_var (tree, tree);
 790 extern bool vect_strided_store_supported (tree);
 791 extern bool vect_strided_load_supported (tree);
 792 extern bool vect_permute_store_chain (VEC(tree,heap) *,unsigned int, gimple,
 793                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 794 extern tree vect_setup_realignment (gimple, gimple_stmt_iterator *, tree *,
 795                                     enum dr_alignment_support, tree,
 796                                     struct loop **);
 797 extern bool vect_permute_load_chain (VEC(tree,heap) *,unsigned int, gimple,
 798                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 799 extern bool vect_transform_strided_load (gimple, VEC(tree,heap) *, int,
 800                                          gimple_stmt_iterator *);
 801 extern int vect_get_place_in_interleaving_chain (gimple, gimple);
 802 extern tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
 803 extern tree vect_create_addr_base_for_vector_ref (gimple, gimple_seq *,
 804                                                   tree, struct loop *);
 805
 806 /* In tree-vect-loop.c.  */
 807 /* FORNOW: Used in tree-parloops.c.  */
 808 extern void destroy_loop_vec_info (loop_vec_info, bool);
 809 extern gimple vect_force_simple_reduction (loop_vec_info, gimple, bool, bool *);
 810 /* Drive for loop analysis stage.  */
 811 extern loop_vec_info vect_analyze_loop (struct loop *);
 812 /* Drive for loop transformation stage.  */
 813 extern void vect_transform_loop (loop_vec_info);
 814 extern loop_vec_info vect_analyze_loop_form (struct loop *);
 815 extern bool vectorizable_live_operation (gimple, gimple_stmt_iterator *,
 816                                          gimple *);
 817 extern bool vectorizable_reduction (gimple, gimple_stmt_iterator *, gimple *,
 818                                     slp_tree);
 819 extern bool vectorizable_induction (gimple, gimple_stmt_iterator *, gimple *);
 820 extern int vect_estimate_min_profitable_iters (loop_vec_info);
 821 extern tree get_initial_def_for_reduction (gimple, tree, tree *);
 822 extern int vect_min_worthwhile_factor (enum tree_code);
 823 extern int vect_get_known_peeling_cost (loop_vec_info, int, int *, int);
 824 extern int vect_get_single_scalar_iteraion_cost (loop_vec_info);
 825
 826 /* In tree-vect-slp.c.  */
 827 extern void vect_free_slp_instance (slp_instance);
 828 extern bool vect_transform_slp_perm_load (gimple, VEC (tree, heap) *,
 829                                           gimple_stmt_iterator *, int,
 830                                           slp_instance, bool);
 831 extern bool vect_schedule_slp (loop_vec_info, bb_vec_info);
 832 extern void vect_update_slp_costs_according_to_vf (loop_vec_info);
 833 extern bool vect_analyze_slp (loop_vec_info, bb_vec_info);
 834 extern void vect_make_slp_decision (loop_vec_info);
 835 extern void vect_detect_hybrid_slp (loop_vec_info);
 836 extern void vect_get_slp_defs (slp_tree, VEC (tree,heap) **,
 837                                VEC (tree,heap) **, int);
 838 extern LOC find_bb_location (basic_block);
 839 extern bb_vec_info vect_slp_analyze_bb (basic_block);
 840 extern void vect_slp_transform_bb (basic_block);
 841
 842 /* In tree-vect-patterns.c.  */
 843 /* Pattern recognition functions.
 844    Additional pattern recognition functions can (and will) be added
 845    in the future.  */
 846 typedef gimple (* vect_recog_func_ptr) (gimple, tree *, tree *);
 847 #define NUM_PATTERNS 4
 848 void vect_pattern_recog (loop_vec_info);
 849
 850 /* In tree-vectorizer.c.  */
 851 unsigned vectorize_loops (void);
 852 /* Vectorization debug information */
 853 extern bool vect_print_dump_info (enum verbosity_levels);
 854
 855 #endif  /* GCC_TREE_VECTORIZER_H  */