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