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_constant_def = 1,
  59   vect_invariant_def,
  60   vect_loop_def,
  61   vect_induction_def,
  62   vect_reduction_def,
  63   vect_unknown_def_type
  64 };
  65
  66 /* Define verbosity levels.  */
  67 enum verbosity_levels {
  68   REPORT_NONE,
  69   REPORT_VECTORIZED_LOOPS,
  70   REPORT_UNVECTORIZED_LOOPS,
  71   REPORT_COST,
  72   REPORT_ALIGNMENT,
  73   REPORT_DR_DETAILS,
  74   REPORT_BAD_FORM_LOOPS,
  75   REPORT_OUTER_LOOPS,
  76   REPORT_SLP,
  77   REPORT_DETAILS,
  78   /* New verbosity levels should be added before this one.  */
  79   MAX_VERBOSITY_LEVEL
  80 };
  81
  82 /************************************************************************
  83   SLP
  84  ************************************************************************/
  85
  86 /* A computation tree of an SLP instance. Each node corresponds to a group of
  87    stmts to be packed in a SIMD stmt.  */
  88 typedef struct _slp_tree {
  89   /* Only binary and unary operations are supported. LEFT child corresponds to
  90      the first operand and RIGHT child to the second if the operation is
  91      binary.  */
  92   struct _slp_tree *left;
  93   struct _slp_tree *right;
  94   /* A group of scalar stmts to be vectorized together.  */
  95   VEC (gimple, heap) *stmts;
  96   /* Vectorized stmt/s.  */
  97   VEC (gimple, heap) *vec_stmts;
  98   /* Number of vector stmts that are created to replace the group of scalar
  99      stmts. It is calculated during the transformation phase as the number of
 100      scalar elements in one scalar iteration (GROUP_SIZE) multiplied by VF
 101      divided by vector size.  */
 102   unsigned int vec_stmts_size;
 103   /* Vectorization costs associated with SLP node.  */
 104   struct
 105   {
 106     int outside_of_loop;     /* Statements generated outside loop.  */
 107     int inside_of_loop;      /* Statements generated inside loop.  */
 108   } cost;
 109 } *slp_tree;
 110
 111 DEF_VEC_P(slp_tree);
 112 DEF_VEC_ALLOC_P(slp_tree, heap);
 113
 114 /* SLP instance is a sequence of stmts in a loop that can be packed into
 115    SIMD stmts.  */
 116 typedef struct _slp_instance {
 117   /* The root of SLP tree.  */
 118   slp_tree root;
 119
 120   /* Size of groups of scalar stmts that will be replaced by SIMD stmt/s.  */
 121   unsigned int group_size;
 122
 123   /* The unrolling factor required to vectorized this SLP instance.  */
 124   unsigned int unrolling_factor;
 125
 126   /* Vectorization costs associated with SLP instance.  */
 127   struct
 128   {
 129     int outside_of_loop;     /* Statements generated outside loop.  */
 130     int inside_of_loop;      /* Statements generated inside loop.  */
 131   } cost;
 132
 133   /* Loads permutation relatively to the stores, NULL if there is no
 134      permutation.  */
 135   VEC (int, heap) *load_permutation;
 136
 137   /* The group of nodes that contain loads of this SLP instance.  */
 138   VEC (slp_tree, heap) *loads;
 139
 140   /* The first scalar load of the instance. The created vector loads will be
 141      inserted before this statement.  */
 142   gimple first_load;
 143 } *slp_instance;
 144
 145 DEF_VEC_P(slp_instance);
 146 DEF_VEC_ALLOC_P(slp_instance, heap);
 147
 148 /* Access Functions.  */
 149 #define SLP_INSTANCE_TREE(S)                     (S)->root
 150 #define SLP_INSTANCE_GROUP_SIZE(S)               (S)->group_size
 151 #define SLP_INSTANCE_UNROLLING_FACTOR(S)         (S)->unrolling_factor
 152 #define SLP_INSTANCE_OUTSIDE_OF_LOOP_COST(S)     (S)->cost.outside_of_loop
 153 #define SLP_INSTANCE_INSIDE_OF_LOOP_COST(S)      (S)->cost.inside_of_loop
 154 #define SLP_INSTANCE_LOAD_PERMUTATION(S)         (S)->load_permutation
 155 #define SLP_INSTANCE_LOADS(S)                    (S)->loads
 156 #define SLP_INSTANCE_FIRST_LOAD_STMT(S)          (S)->first_load
 157
 158 #define SLP_TREE_LEFT(S)                         (S)->left
 159 #define SLP_TREE_RIGHT(S)                        (S)->right
 160 #define SLP_TREE_SCALAR_STMTS(S)                 (S)->stmts
 161 #define SLP_TREE_VEC_STMTS(S)                    (S)->vec_stmts
 162 #define SLP_TREE_NUMBER_OF_VEC_STMTS(S)          (S)->vec_stmts_size
 163 #define SLP_TREE_OUTSIDE_OF_LOOP_COST(S)         (S)->cost.outside_of_loop
 164 #define SLP_TREE_INSIDE_OF_LOOP_COST(S)          (S)->cost.inside_of_loop
 165
 166 /*-----------------------------------------------------------------*/
 167 /* Info on vectorized loops.                                       */
 168 /*-----------------------------------------------------------------*/
 169 typedef struct _loop_vec_info {
 170
 171   /* The loop to which this info struct refers to.  */
 172   struct loop *loop;
 173
 174   /* The loop basic blocks.  */
 175   basic_block *bbs;
 176
 177   /* Number of iterations.  */
 178   tree num_iters;
 179   tree num_iters_unchanged;
 180
 181   /* Minimum number of iterations below which vectorization is expected to
 182      not be profitable (as estimated by the cost model).
 183      -1 indicates that vectorization will not be profitable.
 184      FORNOW: This field is an int. Will be a tree in the future, to represent
 185              values unknown at compile time.  */
 186   int min_profitable_iters;
 187
 188   /* Is the loop vectorizable? */
 189   bool vectorizable;
 190
 191   /* Unrolling factor  */
 192   int vectorization_factor;
 193
 194   /* Unknown DRs according to which loop was peeled.  */
 195   struct data_reference *unaligned_dr;
 196
 197   /* peeling_for_alignment indicates whether peeling for alignment will take
 198      place, and what the peeling factor should be:
 199      peeling_for_alignment = X means:
 200         If X=0: Peeling for alignment will not be applied.
 201         If X>0: Peel first X iterations.
 202         If X=-1: Generate a runtime test to calculate the number of iterations
 203                  to be peeled, using the dataref recorded in the field
 204                  unaligned_dr.  */
 205   int peeling_for_alignment;
 206
 207   /* The mask used to check the alignment of pointers or arrays.  */
 208   int ptr_mask;
 209
 210   /* All data references in the loop.  */
 211   VEC (data_reference_p, heap) *datarefs;
 212
 213   /* All data dependences in the loop.  */
 214   VEC (ddr_p, heap) *ddrs;
 215
 216   /* Data Dependence Relations defining address ranges that are candidates
 217      for a run-time aliasing check.  */
 218   VEC (ddr_p, heap) *may_alias_ddrs;
 219
 220   /* Statements in the loop that have data references that are candidates for a
 221      runtime (loop versioning) misalignment check.  */
 222   VEC(gimple,heap) *may_misalign_stmts;
 223
 224   /* The loop location in the source.  */
 225   LOC loop_line_number;
 226
 227   /* All interleaving chains of stores in the loop, represented by the first
 228      stmt in the chain.  */
 229   VEC(gimple, heap) *strided_stores;
 230
 231   /* All SLP instances in the loop. This is a subset of the set of STRIDED_STORES
 232      of the loop.  */
 233   VEC(slp_instance, heap) *slp_instances;
 234
 235   /* The unrolling factor needed to SLP the loop. In case of that pure SLP is
 236      applied to the loop, i.e., no unrolling is needed, this is 1.  */
 237   unsigned slp_unrolling_factor;
 238 } *loop_vec_info;
 239
 240 /* Access Functions.  */
 241 #define LOOP_VINFO_LOOP(L)            (L)->loop
 242 #define LOOP_VINFO_BBS(L)             (L)->bbs
 243 #define LOOP_VINFO_NITERS(L)          (L)->num_iters
 244 /* Since LOOP_VINFO_NITERS can change after prologue peeling
 245    retain total unchanged scalar loop iterations for cost model.  */
 246 #define LOOP_VINFO_NITERS_UNCHANGED(L)          (L)->num_iters_unchanged
 247 #define LOOP_VINFO_COST_MODEL_MIN_ITERS(L)      (L)->min_profitable_iters
 248 #define LOOP_VINFO_VECTORIZABLE_P(L)  (L)->vectorizable
 249 #define LOOP_VINFO_VECT_FACTOR(L)     (L)->vectorization_factor
 250 #define LOOP_VINFO_PTR_MASK(L)        (L)->ptr_mask
 251 #define LOOP_VINFO_DATAREFS(L)        (L)->datarefs
 252 #define LOOP_VINFO_DDRS(L)            (L)->ddrs
 253 #define LOOP_VINFO_INT_NITERS(L)      (TREE_INT_CST_LOW ((L)->num_iters))
 254 #define LOOP_PEELING_FOR_ALIGNMENT(L) (L)->peeling_for_alignment
 255 #define LOOP_VINFO_UNALIGNED_DR(L)    (L)->unaligned_dr
 256 #define LOOP_VINFO_MAY_MISALIGN_STMTS(L) (L)->may_misalign_stmts
 257 #define LOOP_VINFO_LOC(L)             (L)->loop_line_number
 258 #define LOOP_VINFO_MAY_ALIAS_DDRS(L)  (L)->may_alias_ddrs
 259 #define LOOP_VINFO_STRIDED_STORES(L)  (L)->strided_stores
 260 #define LOOP_VINFO_SLP_INSTANCES(L)   (L)->slp_instances
 261 #define LOOP_VINFO_SLP_UNROLLING_FACTOR(L) (L)->slp_unrolling_factor
 262
 263 #define NITERS_KNOWN_P(n)                     \
 264 (host_integerp ((n),0)                        \
 265 && TREE_INT_CST_LOW ((n)) > 0)
 266
 267 #define LOOP_VINFO_NITERS_KNOWN_P(L)                     \
 268 NITERS_KNOWN_P((L)->num_iters)
 269
 270 static inline loop_vec_info
 271 loop_vec_info_for_loop (struct loop *loop)
 272 {
 273   return (loop_vec_info) loop->aux;
 274 }
 275
 276 static inline bool
 277 nested_in_vect_loop_p (struct loop *loop, gimple stmt)
 278 {
 279   return (loop->inner
 280           && (loop->inner == (gimple_bb (stmt))->loop_father));
 281 }
 282
 283 /*-----------------------------------------------------------------*/
 284 /* Info on vectorized defs.                                        */
 285 /*-----------------------------------------------------------------*/
 286 enum stmt_vec_info_type {
 287   undef_vec_info_type = 0,
 288   load_vec_info_type,
 289   store_vec_info_type,
 290   op_vec_info_type,
 291   call_vec_info_type,
 292   assignment_vec_info_type,
 293   condition_vec_info_type,
 294   reduc_vec_info_type,
 295   induc_vec_info_type,
 296   type_promotion_vec_info_type,
 297   type_demotion_vec_info_type,
 298   type_conversion_vec_info_type,
 299   loop_exit_ctrl_vec_info_type
 300 };
 301
 302 /* Indicates whether/how a variable is used in the loop.  */
 303 enum vect_relevant {
 304   vect_unused_in_loop = 0,
 305   vect_used_in_outer_by_reduction,
 306   vect_used_in_outer,
 307
 308   /* defs that feed computations that end up (only) in a reduction. These
 309      defs may be used by non-reduction stmts, but eventually, any
 310      computations/values that are affected by these defs are used to compute
 311      a reduction (i.e. don't get stored to memory, for example). We use this
 312      to identify computations that we can change the order in which they are
 313      computed.  */
 314   vect_used_by_reduction,
 315
 316   vect_used_in_loop
 317 };
 318
 319 /* The type of vectorization that can be applied to the stmt: regular loop-based
 320    vectorization; pure SLP - the stmt is a part of SLP instances and does not
 321    have uses outside SLP instances; or hybrid SLP and loop-based - the stmt is
 322    a part of SLP instance and also must be loop-based vectorized, since it has
 323    uses outside SLP sequences.
 324
 325    In the loop context the meanings of pure and hybrid SLP are slightly
 326    different. By saying that pure SLP is applied to the loop, we mean that we
 327    exploit only intra-iteration parallelism in the loop; i.e., the loop can be
 328    vectorized without doing any conceptual unrolling, cause we don't pack
 329    together stmts from different iterations, only within a single iteration.
 330    Loop hybrid SLP means that we exploit both intra-iteration and
 331    inter-iteration parallelism (e.g., number of elements in the vector is 4
 332    and the slp-group-size is 2, in which case we don't have enough parallelism
 333    within an iteration, so we obtain the rest of the parallelism from subsequent
 334    iterations by unrolling the loop by 2).  */
 335 enum slp_vect_type {
 336   loop_vect = 0,
 337   pure_slp,
 338   hybrid
 339 };
 340
 341
 342 typedef struct data_reference *dr_p;
 343 DEF_VEC_P(dr_p);
 344 DEF_VEC_ALLOC_P(dr_p,heap);
 345
 346 typedef struct _stmt_vec_info {
 347
 348   enum stmt_vec_info_type type;
 349
 350   /* The stmt to which this info struct refers to.  */
 351   gimple stmt;
 352
 353   /* The loop_vec_info with respect to which STMT is vectorized.  */
 354   loop_vec_info loop_vinfo;
 355
 356   /* Not all stmts in the loop need to be vectorized. e.g, the increment
 357      of the loop induction variable and computation of array indexes. relevant
 358      indicates whether the stmt needs to be vectorized.  */
 359   enum vect_relevant relevant;
 360
 361   /* Indicates whether this stmts is part of a computation whose result is
 362      used outside the loop.  */
 363   bool live;
 364
 365   /* The vector type to be used.  */
 366   tree vectype;
 367
 368   /* The vectorized version of the stmt.  */
 369   gimple vectorized_stmt;
 370
 371
 372   /** The following is relevant only for stmts that contain a non-scalar
 373      data-ref (array/pointer/struct access). A GIMPLE stmt is expected to have
 374      at most one such data-ref.  **/
 375
 376   /* Information about the data-ref (access function, etc),
 377      relative to the inner-most containing loop.  */
 378   struct data_reference *data_ref_info;
 379
 380   /* Information about the data-ref relative to this loop
 381      nest (the loop that is being considered for vectorization).  */
 382   tree dr_base_address;
 383   tree dr_init;
 384   tree dr_offset;
 385   tree dr_step;
 386   tree dr_aligned_to;
 387
 388   /* Stmt is part of some pattern (computation idiom)  */
 389   bool in_pattern_p;
 390
 391   /* Used for various bookkeeping purposes, generally holding a pointer to
 392      some other stmt S that is in some way "related" to this stmt.
 393      Current use of this field is:
 394         If this stmt is part of a pattern (i.e. the field 'in_pattern_p' is
 395         true): S is the "pattern stmt" that represents (and replaces) the
 396         sequence of stmts that constitutes the pattern.  Similarly, the
 397         related_stmt of the "pattern stmt" points back to this stmt (which is
 398         the last stmt in the original sequence of stmts that constitutes the
 399         pattern).  */
 400   gimple related_stmt;
 401
 402   /* List of datarefs that are known to have the same alignment as the dataref
 403      of this stmt.  */
 404   VEC(dr_p,heap) *same_align_refs;
 405
 406   /* Classify the def of this stmt.  */
 407   enum vect_def_type def_type;
 408
 409   /* Interleaving info.  */
 410   /* First data-ref in the interleaving group.  */
 411   gimple first_dr;
 412   /* Pointer to the next data-ref in the group.  */
 413   gimple next_dr;
 414   /* The size of the interleaving group.  */
 415   unsigned int size;
 416   /* For stores, number of stores from this group seen. We vectorize the last
 417      one.  */
 418   unsigned int store_count;
 419   /* For loads only, the gap from the previous load. For consecutive loads, GAP
 420      is 1.  */
 421   unsigned int gap;
 422   /* In case that two or more stmts share data-ref, this is the pointer to the
 423      previously detected stmt with the same dr.  */
 424   gimple same_dr_stmt;
 425   /* For loads only, if there is a store with the same location, this field is
 426      TRUE.  */
 427   bool read_write_dep;
 428
 429   /* Vectorization costs associated with statement.  */
 430   struct
 431   {
 432     int outside_of_loop;     /* Statements generated outside loop.  */
 433     int inside_of_loop;      /* Statements generated inside loop.  */
 434   } cost;
 435
 436   /*  Whether the stmt is SLPed, loop-based vectorized, or both.  */
 437   enum slp_vect_type slp_type;
 438 } *stmt_vec_info;
 439
 440 /* Access Functions.  */
 441 #define STMT_VINFO_TYPE(S)                 (S)->type
 442 #define STMT_VINFO_STMT(S)                 (S)->stmt
 443 #define STMT_VINFO_LOOP_VINFO(S)           (S)->loop_vinfo
 444 #define STMT_VINFO_RELEVANT(S)             (S)->relevant
 445 #define STMT_VINFO_LIVE_P(S)               (S)->live
 446 #define STMT_VINFO_VECTYPE(S)              (S)->vectype
 447 #define STMT_VINFO_VEC_STMT(S)             (S)->vectorized_stmt
 448 #define STMT_VINFO_DATA_REF(S)             (S)->data_ref_info
 449
 450 #define STMT_VINFO_DR_BASE_ADDRESS(S)      (S)->dr_base_address
 451 #define STMT_VINFO_DR_INIT(S)              (S)->dr_init
 452 #define STMT_VINFO_DR_OFFSET(S)            (S)->dr_offset
 453 #define STMT_VINFO_DR_STEP(S)              (S)->dr_step
 454 #define STMT_VINFO_DR_ALIGNED_TO(S)        (S)->dr_aligned_to
 455
 456 #define STMT_VINFO_IN_PATTERN_P(S)         (S)->in_pattern_p
 457 #define STMT_VINFO_RELATED_STMT(S)         (S)->related_stmt
 458 #define STMT_VINFO_SAME_ALIGN_REFS(S)      (S)->same_align_refs
 459 #define STMT_VINFO_DEF_TYPE(S)             (S)->def_type
 460 #define STMT_VINFO_DR_GROUP_FIRST_DR(S)    (S)->first_dr
 461 #define STMT_VINFO_DR_GROUP_NEXT_DR(S)     (S)->next_dr
 462 #define STMT_VINFO_DR_GROUP_SIZE(S)        (S)->size
 463 #define STMT_VINFO_DR_GROUP_STORE_COUNT(S) (S)->store_count
 464 #define STMT_VINFO_DR_GROUP_GAP(S)         (S)->gap
 465 #define STMT_VINFO_DR_GROUP_SAME_DR_STMT(S)(S)->same_dr_stmt
 466 #define STMT_VINFO_DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 467 #define STMT_VINFO_STRIDED_ACCESS(S)      ((S)->first_dr != NULL)
 468
 469 #define DR_GROUP_FIRST_DR(S)               (S)->first_dr
 470 #define DR_GROUP_NEXT_DR(S)                (S)->next_dr
 471 #define DR_GROUP_SIZE(S)                   (S)->size
 472 #define DR_GROUP_STORE_COUNT(S)            (S)->store_count
 473 #define DR_GROUP_GAP(S)                    (S)->gap
 474 #define DR_GROUP_SAME_DR_STMT(S)           (S)->same_dr_stmt
 475 #define DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 476
 477 #define STMT_VINFO_RELEVANT_P(S)          ((S)->relevant != vect_unused_in_loop)
 478 #define STMT_VINFO_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
 479 #define STMT_VINFO_INSIDE_OF_LOOP_COST(S)  (S)->cost.inside_of_loop
 480
 481 #define HYBRID_SLP_STMT(S)                ((S)->slp_type == hybrid)
 482 #define PURE_SLP_STMT(S)                  ((S)->slp_type == pure_slp)
 483 #define STMT_SLP_TYPE(S)                   (S)->slp_type
 484
 485 /* These are some defines for the initial implementation of the vectorizer's
 486    cost model.  These will later be target specific hooks.  */
 487
 488 /* Cost of conditional taken branch.  */
 489 #ifndef TARG_COND_TAKEN_BRANCH_COST
 490 #define TARG_COND_TAKEN_BRANCH_COST        3
 491 #endif
 492
 493 /* Cost of conditional not taken branch.  */
 494 #ifndef TARG_COND_NOT_TAKEN_BRANCH_COST
 495 #define TARG_COND_NOT_TAKEN_BRANCH_COST        1
 496 #endif
 497
 498 /* Cost of any scalar operation, excluding load and store.  */
 499 #ifndef TARG_SCALAR_STMT_COST
 500 #define TARG_SCALAR_STMT_COST           1
 501 #endif
 502
 503 /* Cost of scalar load.  */
 504 #ifndef TARG_SCALAR_LOAD_COST
 505 #define TARG_SCALAR_LOAD_COST           1
 506 #endif
 507
 508 /* Cost of scalar store.  */
 509 #ifndef TARG_SCALAR_STORE_COST
 510 #define TARG_SCALAR_STORE_COST           1
 511 #endif
 512
 513 /* Cost of any vector operation, excluding load, store or vector to scalar
 514    operation.  */
 515 #ifndef TARG_VEC_STMT_COST
 516 #define TARG_VEC_STMT_COST           1
 517 #endif
 518
 519 /* Cost of vector to scalar operation.  */
 520 #ifndef TARG_VEC_TO_SCALAR_COST
 521 #define TARG_VEC_TO_SCALAR_COST      1
 522 #endif
 523
 524 /* Cost of scalar to vector operation.  */
 525 #ifndef TARG_SCALAR_TO_VEC_COST
 526 #define TARG_SCALAR_TO_VEC_COST      1
 527 #endif
 528
 529 /* Cost of aligned vector load.  */
 530 #ifndef TARG_VEC_LOAD_COST
 531 #define TARG_VEC_LOAD_COST           1
 532 #endif
 533
 534 /* Cost of misaligned vector load.  */
 535 #ifndef TARG_VEC_UNALIGNED_LOAD_COST
 536 #define TARG_VEC_UNALIGNED_LOAD_COST 2
 537 #endif
 538
 539 /* Cost of vector store.  */
 540 #ifndef TARG_VEC_STORE_COST
 541 #define TARG_VEC_STORE_COST          1
 542 #endif
 543
 544 /* Cost of vector permutation.  */
 545 #ifndef TARG_VEC_PERMUTE_COST
 546 #define TARG_VEC_PERMUTE_COST          1
 547 #endif
 548
 549 /* The maximum number of intermediate steps required in multi-step type
 550    conversion.  */
 551 #define MAX_INTERM_CVT_STEPS         3
 552
 553 /* Avoid GTY(()) on stmt_vec_info.  */
 554 typedef void *vec_void_p;
 555 DEF_VEC_P (vec_void_p);
 556 DEF_VEC_ALLOC_P (vec_void_p, heap);
 557
 558 extern VEC(vec_void_p,heap) *stmt_vec_info_vec;
 559
 560 void init_stmt_vec_info_vec (void);
 561 void free_stmt_vec_info_vec (void);
 562
 563 static inline stmt_vec_info
 564 vinfo_for_stmt (gimple stmt)
 565 {
 566   unsigned int uid = gimple_uid (stmt);
 567   if (uid == 0)
 568     return NULL;
 569
 570   gcc_assert (uid <= VEC_length (vec_void_p, stmt_vec_info_vec));
 571   return (stmt_vec_info) VEC_index (vec_void_p, stmt_vec_info_vec, uid - 1);
 572 }
 573
 574 static inline void
 575 set_vinfo_for_stmt (gimple stmt, stmt_vec_info info)
 576 {
 577   unsigned int uid = gimple_uid (stmt);
 578   if (uid == 0)
 579     {
 580       gcc_assert (info);
 581       uid = VEC_length (vec_void_p, stmt_vec_info_vec) + 1;
 582       gimple_set_uid (stmt, uid);
 583       VEC_safe_push (vec_void_p, heap, stmt_vec_info_vec, (vec_void_p) info);
 584     }
 585   else
 586     VEC_replace (vec_void_p, stmt_vec_info_vec, uid - 1, (vec_void_p) info);
 587 }
 588
 589 static inline gimple
 590 get_earlier_stmt (gimple stmt1, gimple stmt2)
 591 {
 592   unsigned int uid1, uid2;
 593
 594   if (stmt1 == NULL)
 595     return stmt2;
 596
 597   if (stmt2 == NULL)
 598     return stmt1;
 599
 600   uid1 = gimple_uid (stmt1);
 601   uid2 = gimple_uid (stmt2);
 602
 603   if (uid1 == 0 || uid2 == 0)
 604     return NULL;
 605
 606   gcc_assert (uid1 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 607   gcc_assert (uid2 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 608
 609   if (uid1 < uid2)
 610     return stmt1;
 611   else
 612     return stmt2;
 613 }
 614
 615 static inline bool
 616 is_pattern_stmt_p (stmt_vec_info stmt_info)
 617 {
 618   gimple related_stmt;
 619   stmt_vec_info related_stmt_info;
 620
 621   related_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
 622   if (related_stmt
 623       && (related_stmt_info = vinfo_for_stmt (related_stmt))
 624       && STMT_VINFO_IN_PATTERN_P (related_stmt_info))
 625     return true;
 626
 627   return false;
 628 }
 629
 630 static inline bool
 631 is_loop_header_bb_p (basic_block bb)
 632 {
 633   if (bb == (bb->loop_father)->header)
 634     return true;
 635   gcc_assert (EDGE_COUNT (bb->preds) == 1);
 636   return false;
 637 }
 638
 639 static inline void
 640 stmt_vinfo_set_inside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 641                                     int cost)
 642 {
 643   if (slp_node)
 644     SLP_TREE_INSIDE_OF_LOOP_COST (slp_node) = cost;
 645   else
 646     STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) = cost;
 647 }
 648
 649 static inline void
 650 stmt_vinfo_set_outside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 651                                      int cost)
 652 {
 653   if (slp_node)
 654     SLP_TREE_OUTSIDE_OF_LOOP_COST (slp_node) = cost;
 655   else
 656     STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = cost;
 657 }
 658
 659 static inline int
 660 vect_pow2 (int x)
 661 {
 662   int i, res = 1;
 663
 664   for (i = 0; i < x; i++)
 665     res *= 2;
 666
 667   return res;
 668 }
 669
 670 /*-----------------------------------------------------------------*/
 671 /* Info on data references alignment.                              */
 672 /*-----------------------------------------------------------------*/
 673
 674 /* Reflects actual alignment of first access in the vectorized loop,
 675    taking into account peeling/versioning if applied.  */
 676 #define DR_MISALIGNMENT(DR)   ((int) (size_t) (DR)->aux)
 677 #define SET_DR_MISALIGNMENT(DR, VAL)   ((DR)->aux = (void *) (size_t) (VAL))
 678
 679 static inline bool
 680 aligned_access_p (struct data_reference *data_ref_info)
 681 {
 682   return (DR_MISALIGNMENT (data_ref_info) == 0);
 683 }
 684
 685 static inline bool
 686 known_alignment_for_access_p (struct data_reference *data_ref_info)
 687 {
 688   return (DR_MISALIGNMENT (data_ref_info) != -1);
 689 }
 690
 691 /* vect_dump will be set to stderr or dump_file if exist.  */
 692 extern FILE *vect_dump;
 693 extern LOC vect_loop_location;
 694
 695 extern enum verbosity_levels vect_verbosity_level;
 696
 697 /* Bitmap of virtual variables to be renamed.  */
 698 extern bitmap vect_memsyms_to_rename;
 699
 700
 701 /*-----------------------------------------------------------------*/
 702 /* Function prototypes.                                            */
 703 /*-----------------------------------------------------------------*/
 704
 705 /* Simple loop peeling and versioning utilities for vectorizer's purposes -
 706    in tree-vect-loop-manip.c.  */
 707 extern void slpeel_make_loop_iterate_ntimes (struct loop *, tree);
 708 extern bool slpeel_can_duplicate_loop_p (const struct loop *, const_edge);
 709 extern void vect_loop_versioning (loop_vec_info, bool, tree *, gimple_seq *);
 710 extern void vect_do_peeling_for_loop_bound (loop_vec_info, tree *,
 711                                             tree, gimple_seq);
 712 extern void vect_do_peeling_for_alignment (loop_vec_info);
 713 extern LOC find_loop_location (struct loop *);
 714 extern bool vect_can_advance_ivs_p (loop_vec_info);
 715
 716 /* In tree-vect-stmts.c.  */
 717 extern tree get_vectype_for_scalar_type (tree);
 718 extern bool vect_is_simple_use (tree, loop_vec_info, gimple *, tree *,
 719                                 enum vect_def_type *);
 720 extern bool supportable_widening_operation (enum tree_code, gimple, tree,
 721                                             tree *, tree *, enum tree_code *,
 722                                             enum tree_code *, int *,
 723                                             VEC (tree, heap) **);
 724 extern bool supportable_narrowing_operation (enum tree_code, const_gimple,
 725                                              tree, enum tree_code *, int *,
 726                                              VEC (tree, heap) **);
 727 extern stmt_vec_info new_stmt_vec_info (gimple stmt, loop_vec_info);
 728 extern void free_stmt_vec_info (gimple stmt);
 729 extern tree vectorizable_function (gimple, tree, tree);
 730 extern void vect_model_simple_cost (stmt_vec_info, int, enum vect_def_type *,
 731                                     slp_tree);
 732 extern void vect_model_store_cost (stmt_vec_info, int, enum vect_def_type,
 733                                    slp_tree);
 734 extern void vect_model_load_cost (stmt_vec_info, int, slp_tree);
 735 extern void vect_finish_stmt_generation (gimple, gimple,
 736                                          gimple_stmt_iterator *);
 737 extern bool vect_mark_stmts_to_be_vectorized (loop_vec_info);
 738 extern int cost_for_stmt (gimple);
 739 extern tree vect_get_vec_def_for_operand (tree, gimple, tree *);
 740 extern tree vect_init_vector (gimple, tree, tree,
 741                               gimple_stmt_iterator *);
 742 extern tree vect_get_vec_def_for_stmt_copy (enum vect_def_type, tree);
 743 extern bool vect_transform_stmt (gimple, gimple_stmt_iterator *,
 744                                  bool *, slp_tree, slp_instance);
 745 extern void vect_remove_stores (gimple);
 746 extern bool vect_analyze_operations (loop_vec_info);
 747
 748 /* In tree-vect-data-refs.c.  */
 749 extern bool vect_can_force_dr_alignment_p (const_tree, unsigned int);
 750 extern enum dr_alignment_support vect_supportable_dr_alignment
 751                                            (struct data_reference *);
 752 extern tree vect_get_smallest_scalar_type (gimple, HOST_WIDE_INT *,
 753                                            HOST_WIDE_INT *);
 754 extern bool vect_analyze_data_ref_dependences (loop_vec_info);
 755 extern bool vect_enhance_data_refs_alignment (loop_vec_info);
 756 extern bool vect_analyze_data_refs_alignment (loop_vec_info);
 757 extern bool vect_analyze_data_ref_accesses (loop_vec_info);
 758 extern bool vect_prune_runtime_alias_test_list (loop_vec_info);
 759 extern bool vect_analyze_data_refs (loop_vec_info);
 760 extern tree vect_create_data_ref_ptr (gimple, struct loop *, tree, tree *,
 761                                       gimple *, bool, bool *);
 762 extern tree bump_vector_ptr (tree, gimple, gimple_stmt_iterator *, gimple, tree);
 763 extern tree vect_create_destination_var (tree, tree);
 764 extern bool vect_strided_store_supported (tree);
 765 extern bool vect_strided_load_supported (tree);
 766 extern bool vect_permute_store_chain (VEC(tree,heap) *,unsigned int, gimple,
 767                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 768 extern tree vect_setup_realignment (gimple, gimple_stmt_iterator *, tree *,
 769                                     enum dr_alignment_support, tree,
 770                                     struct loop **);
 771 extern bool vect_permute_load_chain (VEC(tree,heap) *,unsigned int, gimple,
 772                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 773 extern bool vect_transform_strided_load (gimple, VEC(tree,heap) *, int,
 774                                          gimple_stmt_iterator *);
 775 extern int vect_get_place_in_interleaving_chain (gimple, gimple);
 776 extern tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
 777 extern tree vect_create_addr_base_for_vector_ref (gimple, gimple_seq *,
 778                                                   tree, struct loop *);
 779
 780 /* In tree-vect-loop.c.  */
 781 /* FORNOW: Used in tree-parloops.c.  */
 782 extern void destroy_loop_vec_info (loop_vec_info, bool);
 783 extern gimple vect_is_simple_reduction (loop_vec_info, gimple);
 784 /* Drive for loop analysis stage.  */
 785 extern loop_vec_info vect_analyze_loop (struct loop *);
 786 /* Drive for loop transformation stage.  */
 787 extern void vect_transform_loop (loop_vec_info);
 788 extern loop_vec_info vect_analyze_loop_form (struct loop *);
 789 extern bool vectorizable_live_operation (gimple, gimple_stmt_iterator *,
 790                                          gimple *);
 791 extern bool vectorizable_reduction (gimple, gimple_stmt_iterator *, gimple *);
 792 extern bool vectorizable_induction (gimple, gimple_stmt_iterator *, gimple *);
 793 extern int vect_estimate_min_profitable_iters (loop_vec_info);
 794 extern tree get_initial_def_for_reduction (gimple, tree, tree *);
 795 extern int vect_min_worthwhile_factor (enum tree_code);
 796
 797
 798 /* In tree-vect-slp.c.  */
 799 extern void vect_free_slp_instance (slp_instance);
 800 extern bool vect_transform_slp_perm_load (gimple, VEC (tree, heap) *,
 801                                           gimple_stmt_iterator *, int,
 802                                           slp_instance, bool);
 803 extern bool vect_schedule_slp (loop_vec_info);
 804 extern void vect_update_slp_costs_according_to_vf (loop_vec_info);
 805 extern bool vect_analyze_slp (loop_vec_info);
 806 extern void vect_make_slp_decision (loop_vec_info);
 807 extern void vect_detect_hybrid_slp (loop_vec_info);
 808 extern void vect_get_slp_defs (slp_tree, VEC (tree,heap) **,
 809                                VEC (tree,heap) **);
 810
 811 /* In tree-vect-patterns.c.  */
 812 /* Pattern recognition functions.
 813    Additional pattern recognition functions can (and will) be added
 814    in the future.  */
 815 typedef gimple (* vect_recog_func_ptr) (gimple, tree *, tree *);
 816 #define NUM_PATTERNS 4
 817 void vect_pattern_recog (loop_vec_info);
 818
 819 /*  Vectorization debug information - in tree-vectorizer.c.  */
 820 extern bool vect_print_dump_info (enum verbosity_levels);
 821 extern void vect_set_verbosity_level (const char *);
 822
 823 #endif  /* GCC_TREE_VECTORIZER_H  */