gcc/tree-vectorizer.h

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