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