gcc/tree-vectorizer.h

   1 /* Vectorizer
   2    Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
   3    Free Software Foundation, Inc.
   4    Contributed by Dorit Naishlos <dorit@il.ibm.com>
   5
   6 This file is part of GCC.
   7
   8 GCC is free software; you can redistribute it and/or modify it under
   9 the terms of the GNU General Public License as published by the Free
  10 Software Foundation; either version 3, or (at your option) any later
  11 version.
  12
  13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  15 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  16 for more details.
  17
  18 You should have received a copy of the GNU General Public License
  19 along with GCC; see the file COPYING3.  If not see
  20 <http://www.gnu.org/licenses/>.  */
  21
  22 #ifndef GCC_TREE_VECTORIZER_H
  23 #define GCC_TREE_VECTORIZER_H
  24
  25 #include "tree-data-ref.h"
  26
  27 typedef source_location LOC;
  28 #define UNKNOWN_LOC UNKNOWN_LOCATION
  29 #define EXPR_LOC(e) EXPR_LOCATION(e)
  30 #define LOC_FILE(l) LOCATION_FILE (l)
  31 #define LOC_LINE(l) LOCATION_LINE (l)
  32
  33 /* Used for naming of new temporaries.  */
  34 enum vect_var_kind {
  35   vect_simple_var,
  36   vect_pointer_var,
  37   vect_scalar_var
  38 };
  39
  40 /* Defines type of operation.  */
  41 enum operation_type {
  42   unary_op = 1,
  43   binary_op,
  44   ternary_op
  45 };
  46
  47 /* Define type of available alignment support.  */
  48 enum dr_alignment_support {
  49   dr_unaligned_unsupported,
  50   dr_unaligned_supported,
  51   dr_explicit_realign,
  52   dr_explicit_realign_optimized,
  53   dr_aligned
  54 };
  55
  56 /* Define type of def-use cross-iteration cycle.  */
  57 enum vect_def_type {
  58   vect_uninitialized_def = 0,
  59   vect_constant_def = 1,
  60   vect_external_def,
  61   vect_internal_def,
  62   vect_induction_def,
  63   vect_reduction_def,
  64   vect_double_reduction_def,
  65   vect_nested_cycle,
  66   vect_unknown_def_type
  67 };
  68
  69 #define VECTORIZABLE_CYCLE_DEF(D) (((D) == vect_reduction_def)           \
  70                                    || ((D) == vect_double_reduction_def) \
  71                                    || ((D) == vect_nested_cycle))
  72
  73 /* Define verbosity levels.  */
  74 enum verbosity_levels {
  75   REPORT_NONE,
  76   REPORT_VECTORIZED_LOCATIONS,
  77   REPORT_UNVECTORIZED_LOCATIONS,
  78   REPORT_COST,
  79   REPORT_ALIGNMENT,
  80   REPORT_DR_DETAILS,
  81   REPORT_BAD_FORM_LOOPS,
  82   REPORT_OUTER_LOOPS,
  83   REPORT_SLP,
  84   REPORT_DETAILS,
  85   /* New verbosity levels should be added before this one.  */
  86   MAX_VERBOSITY_LEVEL
  87 };
  88
  89 /************************************************************************
  90   SLP
  91  ************************************************************************/
  92
  93 /* A computation tree of an SLP instance. Each node corresponds to a group of
  94    stmts to be packed in a SIMD stmt.  */
  95 typedef struct _slp_tree {
  96   /* Only binary and unary operations are supported. LEFT child corresponds to
  97      the first operand and RIGHT child to the second if the operation is
  98      binary.  */
  99   struct _slp_tree *left;
 100   struct _slp_tree *right;
 101   /* A group of scalar stmts to be vectorized together.  */
 102   VEC (gimple, heap) *stmts;
 103   /* Vectorized stmt/s.  */
 104   VEC (gimple, heap) *vec_stmts;
 105   /* Number of vector stmts that are created to replace the group of scalar
 106      stmts. It is calculated during the transformation phase as the number of
 107      scalar elements in one scalar iteration (GROUP_SIZE) multiplied by VF
 108      divided by vector size.  */
 109   unsigned int vec_stmts_size;
 110   /* Vectorization costs associated with SLP node.  */
 111   struct
 112   {
 113     int outside_of_loop;     /* Statements generated outside loop.  */
 114     int inside_of_loop;      /* Statements generated inside loop.  */
 115   } cost;
 116 } *slp_tree;
 117
 118 DEF_VEC_P(slp_tree);
 119 DEF_VEC_ALLOC_P(slp_tree, heap);
 120
 121 /* SLP instance is a sequence of stmts in a loop that can be packed into
 122    SIMD stmts.  */
 123 typedef struct _slp_instance {
 124   /* The root of SLP tree.  */
 125   slp_tree root;
 126
 127   /* Size of groups of scalar stmts that will be replaced by SIMD stmt/s.  */
 128   unsigned int group_size;
 129
 130   /* The unrolling factor required to vectorized this SLP instance.  */
 131   unsigned int unrolling_factor;
 132
 133   /* Vectorization costs associated with SLP instance.  */
 134   struct
 135   {
 136     int outside_of_loop;     /* Statements generated outside loop.  */
 137     int inside_of_loop;      /* Statements generated inside loop.  */
 138   } cost;
 139
 140   /* Loads permutation relatively to the stores, NULL if there is no
 141      permutation.  */
 142   VEC (int, heap) *load_permutation;
 143
 144   /* The group of nodes that contain loads of this SLP instance.  */
 145   VEC (slp_tree, heap) *loads;
 146
 147   /* The first scalar load of the instance. The created vector loads will be
 148      inserted before this statement.  */
 149   gimple first_load;
 150 } *slp_instance;
 151
 152 DEF_VEC_P(slp_instance);
 153 DEF_VEC_ALLOC_P(slp_instance, heap);
 154
 155 /* Access Functions.  */
 156 #define SLP_INSTANCE_TREE(S)                     (S)->root
 157 #define SLP_INSTANCE_GROUP_SIZE(S)               (S)->group_size
 158 #define SLP_INSTANCE_UNROLLING_FACTOR(S)         (S)->unrolling_factor
 159 #define SLP_INSTANCE_OUTSIDE_OF_LOOP_COST(S)     (S)->cost.outside_of_loop
 160 #define SLP_INSTANCE_INSIDE_OF_LOOP_COST(S)      (S)->cost.inside_of_loop
 161 #define SLP_INSTANCE_LOAD_PERMUTATION(S)         (S)->load_permutation
 162 #define SLP_INSTANCE_LOADS(S)                    (S)->loads
 163 #define SLP_INSTANCE_FIRST_LOAD_STMT(S)          (S)->first_load
 164
 165 #define SLP_TREE_LEFT(S)                         (S)->left
 166 #define SLP_TREE_RIGHT(S)                        (S)->right
 167 #define SLP_TREE_SCALAR_STMTS(S)                 (S)->stmts
 168 #define SLP_TREE_VEC_STMTS(S)                    (S)->vec_stmts
 169 #define SLP_TREE_NUMBER_OF_VEC_STMTS(S)          (S)->vec_stmts_size
 170 #define SLP_TREE_OUTSIDE_OF_LOOP_COST(S)         (S)->cost.outside_of_loop
 171 #define SLP_TREE_INSIDE_OF_LOOP_COST(S)          (S)->cost.inside_of_loop
 172
 173 /*-----------------------------------------------------------------*/
 174 /* Info on vectorized loops.                                       */
 175 /*-----------------------------------------------------------------*/
 176 typedef struct _loop_vec_info {
 177
 178   /* The loop to which this info struct refers to.  */
 179   struct loop *loop;
 180
 181   /* The loop basic blocks.  */
 182   basic_block *bbs;
 183
 184   /* Number of iterations.  */
 185   tree num_iters;
 186   tree num_iters_unchanged;
 187
 188   /* Minimum number of iterations below which vectorization is expected to
 189      not be profitable (as estimated by the cost model).
 190      -1 indicates that vectorization will not be profitable.
 191      FORNOW: This field is an int. Will be a tree in the future, to represent
 192              values unknown at compile time.  */
 193   int min_profitable_iters;
 194
 195   /* Is the loop vectorizable? */
 196   bool vectorizable;
 197
 198   /* Unrolling factor  */
 199   int vectorization_factor;
 200
 201   /* Unknown DRs according to which loop was peeled.  */
 202   struct data_reference *unaligned_dr;
 203
 204   /* peeling_for_alignment indicates whether peeling for alignment will take
 205      place, and what the peeling factor should be:
 206      peeling_for_alignment = X means:
 207         If X=0: Peeling for alignment will not be applied.
 208         If X>0: Peel first X iterations.
 209         If X=-1: Generate a runtime test to calculate the number of iterations
 210                  to be peeled, using the dataref recorded in the field
 211                  unaligned_dr.  */
 212   int peeling_for_alignment;
 213
 214   /* The mask used to check the alignment of pointers or arrays.  */
 215   int ptr_mask;
 216
 217   /* All data references in the loop.  */
 218   VEC (data_reference_p, heap) *datarefs;
 219
 220   /* All data dependences in the loop.  */
 221   VEC (ddr_p, heap) *ddrs;
 222
 223   /* Data Dependence Relations defining address ranges that are candidates
 224      for a run-time aliasing check.  */
 225   VEC (ddr_p, heap) *may_alias_ddrs;
 226
 227   /* Statements in the loop that have data references that are candidates for a
 228      runtime (loop versioning) misalignment check.  */
 229   VEC(gimple,heap) *may_misalign_stmts;
 230
 231   /* The loop location in the source.  */
 232   LOC loop_line_number;
 233
 234   /* All interleaving chains of stores in the loop, represented by the first
 235      stmt in the chain.  */
 236   VEC(gimple, heap) *strided_stores;
 237
 238   /* All SLP instances in the loop. This is a subset of the set of STRIDED_STORES
 239      of the loop.  */
 240   VEC(slp_instance, heap) *slp_instances;
 241
 242   /* The unrolling factor needed to SLP the loop. In case of that pure SLP is
 243      applied to the loop, i.e., no unrolling is needed, this is 1.  */
 244   unsigned slp_unrolling_factor;
 245 } *loop_vec_info;
 246
 247 /* Access Functions.  */
 248 #define LOOP_VINFO_LOOP(L)                 (L)->loop
 249 #define LOOP_VINFO_BBS(L)                  (L)->bbs
 250 #define LOOP_VINFO_NITERS(L)               (L)->num_iters
 251 /* Since LOOP_VINFO_NITERS can change after prologue peeling
 252    retain total unchanged scalar loop iterations for cost model.  */
 253 #define LOOP_VINFO_NITERS_UNCHANGED(L)     (L)->num_iters_unchanged
 254 #define LOOP_VINFO_COST_MODEL_MIN_ITERS(L) (L)->min_profitable_iters
 255 #define LOOP_VINFO_VECTORIZABLE_P(L)       (L)->vectorizable
 256 #define LOOP_VINFO_VECT_FACTOR(L)          (L)->vectorization_factor
 257 #define LOOP_VINFO_PTR_MASK(L)             (L)->ptr_mask
 258 #define LOOP_VINFO_DATAREFS(L)             (L)->datarefs
 259 #define LOOP_VINFO_DDRS(L)                 (L)->ddrs
 260 #define LOOP_VINFO_INT_NITERS(L)           (TREE_INT_CST_LOW ((L)->num_iters))
 261 #define LOOP_PEELING_FOR_ALIGNMENT(L)      (L)->peeling_for_alignment
 262 #define LOOP_VINFO_UNALIGNED_DR(L)         (L)->unaligned_dr
 263 #define LOOP_VINFO_MAY_MISALIGN_STMTS(L)   (L)->may_misalign_stmts
 264 #define LOOP_VINFO_LOC(L)                  (L)->loop_line_number
 265 #define LOOP_VINFO_MAY_ALIAS_DDRS(L)       (L)->may_alias_ddrs
 266 #define LOOP_VINFO_STRIDED_STORES(L)       (L)->strided_stores
 267 #define LOOP_VINFO_SLP_INSTANCES(L)        (L)->slp_instances
 268 #define LOOP_VINFO_SLP_UNROLLING_FACTOR(L) (L)->slp_unrolling_factor
 269
 270 #define LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT(L) \
 271 VEC_length (gimple, (L)->may_misalign_stmts) > 0
 272 #define LOOP_REQUIRES_VERSIONING_FOR_ALIAS(L)     \
 273 VEC_length (ddr_p, (L)->may_alias_ddrs) > 0
 274
 275 #define NITERS_KNOWN_P(n)                     \
 276 (host_integerp ((n),0)                        \
 277 && TREE_INT_CST_LOW ((n)) > 0)
 278
 279 #define LOOP_VINFO_NITERS_KNOWN_P(L)          \
 280 NITERS_KNOWN_P((L)->num_iters)
 281
 282 static inline loop_vec_info
 283 loop_vec_info_for_loop (struct loop *loop)
 284 {
 285   return (loop_vec_info) loop->aux;
 286 }
 287
 288 static inline bool
 289 nested_in_vect_loop_p (struct loop *loop, gimple stmt)
 290 {
 291   return (loop->inner
 292           && (loop->inner == (gimple_bb (stmt))->loop_father));
 293 }
 294
 295 typedef struct _bb_vec_info {
 296
 297   basic_block bb;
 298   /* All interleaving chains of stores in the basic block, represented by the
 299      first stmt in the chain.  */
 300   VEC(gimple, heap) *strided_stores;
 301
 302   /* All SLP instances in the basic block. This is a subset of the set of
 303      STRIDED_STORES of the basic block.  */
 304   VEC(slp_instance, heap) *slp_instances;
 305
 306   /* All data references in the basic block.  */
 307   VEC (data_reference_p, heap) *datarefs;
 308
 309   /* All data dependences in the basic block.  */
 310   VEC (ddr_p, heap) *ddrs;
 311 } *bb_vec_info;
 312
 313 #define BB_VINFO_BB(B)              (B)->bb
 314 #define BB_VINFO_STRIDED_STORES(B)  (B)->strided_stores
 315 #define BB_VINFO_SLP_INSTANCES(B)   (B)->slp_instances
 316 #define BB_VINFO_DATAREFS(B)        (B)->datarefs
 317 #define BB_VINFO_DDRS(B)            (B)->ddrs
 318
 319 static inline bb_vec_info
 320 vec_info_for_bb (basic_block bb)
 321 {
 322   return (bb_vec_info) bb->aux;
 323 }
 324
 325 /*-----------------------------------------------------------------*/
 326 /* Info on vectorized defs.                                        */
 327 /*-----------------------------------------------------------------*/
 328 enum stmt_vec_info_type {
 329   undef_vec_info_type = 0,
 330   load_vec_info_type,
 331   store_vec_info_type,
 332   op_vec_info_type,
 333   call_vec_info_type,
 334   assignment_vec_info_type,
 335   condition_vec_info_type,
 336   reduc_vec_info_type,
 337   induc_vec_info_type,
 338   type_promotion_vec_info_type,
 339   type_demotion_vec_info_type,
 340   type_conversion_vec_info_type,
 341   loop_exit_ctrl_vec_info_type
 342 };
 343
 344 /* Indicates whether/how a variable is used in the scope of loop/basic
 345    block.  */
 346 enum vect_relevant {
 347   vect_unused_in_scope = 0,
 348   /* The def is in the inner loop, and the use is in the outer loop, and the
 349      use is a reduction stmt.  */
 350   vect_used_in_outer_by_reduction,
 351   /* The def is in the inner loop, and the use is in the outer loop (and is
 352      not part of reduction).  */
 353   vect_used_in_outer,
 354
 355   /* defs that feed computations that end up (only) in a reduction. These
 356      defs may be used by non-reduction stmts, but eventually, any
 357      computations/values that are affected by these defs are used to compute
 358      a reduction (i.e. don't get stored to memory, for example). We use this
 359      to identify computations that we can change the order in which they are
 360      computed.  */
 361   vect_used_by_reduction,
 362
 363   vect_used_in_scope
 364 };
 365
 366 /* The type of vectorization that can be applied to the stmt: regular loop-based
 367    vectorization; pure SLP - the stmt is a part of SLP instances and does not
 368    have uses outside SLP instances; or hybrid SLP and loop-based - the stmt is
 369    a part of SLP instance and also must be loop-based vectorized, since it has
 370    uses outside SLP sequences.
 371
 372    In the loop context the meanings of pure and hybrid SLP are slightly
 373    different. By saying that pure SLP is applied to the loop, we mean that we
 374    exploit only intra-iteration parallelism in the loop; i.e., the loop can be
 375    vectorized without doing any conceptual unrolling, cause we don't pack
 376    together stmts from different iterations, only within a single iteration.
 377    Loop hybrid SLP means that we exploit both intra-iteration and
 378    inter-iteration parallelism (e.g., number of elements in the vector is 4
 379    and the slp-group-size is 2, in which case we don't have enough parallelism
 380    within an iteration, so we obtain the rest of the parallelism from subsequent
 381    iterations by unrolling the loop by 2).  */
 382 enum slp_vect_type {
 383   loop_vect = 0,
 384   pure_slp,
 385   hybrid
 386 };
 387
 388
 389 typedef struct data_reference *dr_p;
 390 DEF_VEC_P(dr_p);
 391 DEF_VEC_ALLOC_P(dr_p,heap);
 392
 393 typedef struct _stmt_vec_info {
 394
 395   enum stmt_vec_info_type type;
 396
 397   /* The stmt to which this info struct refers to.  */
 398   gimple stmt;
 399
 400   /* The loop_vec_info with respect to which STMT is vectorized.  */
 401   loop_vec_info loop_vinfo;
 402
 403   /* Not all stmts in the loop need to be vectorized. e.g, the increment
 404      of the loop induction variable and computation of array indexes. relevant
 405      indicates whether the stmt needs to be vectorized.  */
 406   enum vect_relevant relevant;
 407
 408   /* Indicates whether this stmts is part of a computation whose result is
 409      used outside the loop.  */
 410   bool live;
 411
 412   /* The vector type to be used.  */
 413   tree vectype;
 414
 415   /* The vectorized version of the stmt.  */
 416   gimple vectorized_stmt;
 417
 418
 419   /** The following is relevant only for stmts that contain a non-scalar
 420      data-ref (array/pointer/struct access). A GIMPLE stmt is expected to have
 421      at most one such data-ref.  **/
 422
 423   /* Information about the data-ref (access function, etc),
 424      relative to the inner-most containing loop.  */
 425   struct data_reference *data_ref_info;
 426
 427   /* Information about the data-ref relative to this loop
 428      nest (the loop that is being considered for vectorization).  */
 429   tree dr_base_address;
 430   tree dr_init;
 431   tree dr_offset;
 432   tree dr_step;
 433   tree dr_aligned_to;
 434
 435   /* Stmt is part of some pattern (computation idiom)  */
 436   bool in_pattern_p;
 437
 438   /* Used for various bookkeeping purposes, generally holding a pointer to
 439      some other stmt S that is in some way "related" to this stmt.
 440      Current use of this field is:
 441         If this stmt is part of a pattern (i.e. the field 'in_pattern_p' is
 442         true): S is the "pattern stmt" that represents (and replaces) the
 443         sequence of stmts that constitutes the pattern.  Similarly, the
 444         related_stmt of the "pattern stmt" points back to this stmt (which is
 445         the last stmt in the original sequence of stmts that constitutes the
 446         pattern).  */
 447   gimple related_stmt;
 448
 449   /* List of datarefs that are known to have the same alignment as the dataref
 450      of this stmt.  */
 451   VEC(dr_p,heap) *same_align_refs;
 452
 453   /* Classify the def of this stmt.  */
 454   enum vect_def_type def_type;
 455
 456   /* Interleaving info.  */
 457   /* First data-ref in the interleaving group.  */
 458   gimple first_dr;
 459   /* Pointer to the next data-ref in the group.  */
 460   gimple next_dr;
 461   /* The size of the interleaving group.  */
 462   unsigned int size;
 463   /* For stores, number of stores from this group seen. We vectorize the last
 464      one.  */
 465   unsigned int store_count;
 466   /* For loads only, the gap from the previous load. For consecutive loads, GAP
 467      is 1.  */
 468   unsigned int gap;
 469   /* In case that two or more stmts share data-ref, this is the pointer to the
 470      previously detected stmt with the same dr.  */
 471   gimple same_dr_stmt;
 472   /* For loads only, if there is a store with the same location, this field is
 473      TRUE.  */
 474   bool read_write_dep;
 475
 476   /* Vectorization costs associated with statement.  */
 477   struct
 478   {
 479     int outside_of_loop;     /* Statements generated outside loop.  */
 480     int inside_of_loop;      /* Statements generated inside loop.  */
 481   } cost;
 482
 483   /*  Whether the stmt is SLPed, loop-based vectorized, or both.  */
 484   enum slp_vect_type slp_type;
 485
 486   /* The bb_vec_info with respect to which STMT is vectorized.  */
 487   bb_vec_info bb_vinfo;
 488 } *stmt_vec_info;
 489
 490 /* Access Functions.  */
 491 #define STMT_VINFO_TYPE(S)                 (S)->type
 492 #define STMT_VINFO_STMT(S)                 (S)->stmt
 493 #define STMT_VINFO_LOOP_VINFO(S)           (S)->loop_vinfo
 494 #define STMT_VINFO_BB_VINFO(S)             (S)->bb_vinfo
 495 #define STMT_VINFO_RELEVANT(S)             (S)->relevant
 496 #define STMT_VINFO_LIVE_P(S)               (S)->live
 497 #define STMT_VINFO_VECTYPE(S)              (S)->vectype
 498 #define STMT_VINFO_VEC_STMT(S)             (S)->vectorized_stmt
 499 #define STMT_VINFO_DATA_REF(S)             (S)->data_ref_info
 500
 501 #define STMT_VINFO_DR_BASE_ADDRESS(S)      (S)->dr_base_address
 502 #define STMT_VINFO_DR_INIT(S)              (S)->dr_init
 503 #define STMT_VINFO_DR_OFFSET(S)            (S)->dr_offset
 504 #define STMT_VINFO_DR_STEP(S)              (S)->dr_step
 505 #define STMT_VINFO_DR_ALIGNED_TO(S)        (S)->dr_aligned_to
 506
 507 #define STMT_VINFO_IN_PATTERN_P(S)         (S)->in_pattern_p
 508 #define STMT_VINFO_RELATED_STMT(S)         (S)->related_stmt
 509 #define STMT_VINFO_SAME_ALIGN_REFS(S)      (S)->same_align_refs
 510 #define STMT_VINFO_DEF_TYPE(S)             (S)->def_type
 511 #define STMT_VINFO_DR_GROUP_FIRST_DR(S)    (S)->first_dr
 512 #define STMT_VINFO_DR_GROUP_NEXT_DR(S)     (S)->next_dr
 513 #define STMT_VINFO_DR_GROUP_SIZE(S)        (S)->size
 514 #define STMT_VINFO_DR_GROUP_STORE_COUNT(S) (S)->store_count
 515 #define STMT_VINFO_DR_GROUP_GAP(S)         (S)->gap
 516 #define STMT_VINFO_DR_GROUP_SAME_DR_STMT(S)(S)->same_dr_stmt
 517 #define STMT_VINFO_DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 518 #define STMT_VINFO_STRIDED_ACCESS(S)      ((S)->first_dr != NULL)
 519
 520 #define DR_GROUP_FIRST_DR(S)               (S)->first_dr
 521 #define DR_GROUP_NEXT_DR(S)                (S)->next_dr
 522 #define DR_GROUP_SIZE(S)                   (S)->size
 523 #define DR_GROUP_STORE_COUNT(S)            (S)->store_count
 524 #define DR_GROUP_GAP(S)                    (S)->gap
 525 #define DR_GROUP_SAME_DR_STMT(S)           (S)->same_dr_stmt
 526 #define DR_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
 527
 528 #define STMT_VINFO_RELEVANT_P(S)          ((S)->relevant != vect_unused_in_scope)
 529 #define STMT_VINFO_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
 530 #define STMT_VINFO_INSIDE_OF_LOOP_COST(S)  (S)->cost.inside_of_loop
 531
 532 #define HYBRID_SLP_STMT(S)                ((S)->slp_type == hybrid)
 533 #define PURE_SLP_STMT(S)                  ((S)->slp_type == pure_slp)
 534 #define STMT_SLP_TYPE(S)                   (S)->slp_type
 535
 536 /* These are some defines for the initial implementation of the vectorizer's
 537    cost model.  These will later be target specific hooks.  */
 538
 539 /* Cost of conditional taken branch.  */
 540 #ifndef TARG_COND_TAKEN_BRANCH_COST
 541 #define TARG_COND_TAKEN_BRANCH_COST        3
 542 #endif
 543
 544 /* Cost of conditional not taken branch.  */
 545 #ifndef TARG_COND_NOT_TAKEN_BRANCH_COST
 546 #define TARG_COND_NOT_TAKEN_BRANCH_COST        1
 547 #endif
 548
 549 /* Cost of any scalar operation, excluding load and store.  */
 550 #ifndef TARG_SCALAR_STMT_COST
 551 #define TARG_SCALAR_STMT_COST           1
 552 #endif
 553
 554 /* Cost of scalar load.  */
 555 #ifndef TARG_SCALAR_LOAD_COST
 556 #define TARG_SCALAR_LOAD_COST           1
 557 #endif
 558
 559 /* Cost of scalar store.  */
 560 #ifndef TARG_SCALAR_STORE_COST
 561 #define TARG_SCALAR_STORE_COST           1
 562 #endif
 563
 564 /* Cost of any vector operation, excluding load, store or vector to scalar
 565    operation.  */
 566 #ifndef TARG_VEC_STMT_COST
 567 #define TARG_VEC_STMT_COST           1
 568 #endif
 569
 570 /* Cost of vector to scalar operation.  */
 571 #ifndef TARG_VEC_TO_SCALAR_COST
 572 #define TARG_VEC_TO_SCALAR_COST      1
 573 #endif
 574
 575 /* Cost of scalar to vector operation.  */
 576 #ifndef TARG_SCALAR_TO_VEC_COST
 577 #define TARG_SCALAR_TO_VEC_COST      1
 578 #endif
 579
 580 /* Cost of aligned vector load.  */
 581 #ifndef TARG_VEC_LOAD_COST
 582 #define TARG_VEC_LOAD_COST           1
 583 #endif
 584
 585 /* Cost of misaligned vector load.  */
 586 #ifndef TARG_VEC_UNALIGNED_LOAD_COST
 587 #define TARG_VEC_UNALIGNED_LOAD_COST 2
 588 #endif
 589
 590 /* Cost of vector store.  */
 591 #ifndef TARG_VEC_STORE_COST
 592 #define TARG_VEC_STORE_COST          1
 593 #endif
 594
 595 /* Cost of vector permutation.  */
 596 #ifndef TARG_VEC_PERMUTE_COST
 597 #define TARG_VEC_PERMUTE_COST          1
 598 #endif
 599
 600 /* The maximum number of intermediate steps required in multi-step type
 601    conversion.  */
 602 #define MAX_INTERM_CVT_STEPS         3
 603
 604 /* Avoid GTY(()) on stmt_vec_info.  */
 605 typedef void *vec_void_p;
 606 DEF_VEC_P (vec_void_p);
 607 DEF_VEC_ALLOC_P (vec_void_p, heap);
 608
 609 extern VEC(vec_void_p,heap) *stmt_vec_info_vec;
 610
 611 void init_stmt_vec_info_vec (void);
 612 void free_stmt_vec_info_vec (void);
 613
 614 static inline stmt_vec_info
 615 vinfo_for_stmt (gimple stmt)
 616 {
 617   unsigned int uid = gimple_uid (stmt);
 618   if (uid == 0)
 619     return NULL;
 620
 621   gcc_assert (uid <= VEC_length (vec_void_p, stmt_vec_info_vec));
 622   return (stmt_vec_info) VEC_index (vec_void_p, stmt_vec_info_vec, uid - 1);
 623 }
 624
 625 static inline void
 626 set_vinfo_for_stmt (gimple stmt, stmt_vec_info info)
 627 {
 628   unsigned int uid = gimple_uid (stmt);
 629   if (uid == 0)
 630     {
 631       gcc_assert (info);
 632       uid = VEC_length (vec_void_p, stmt_vec_info_vec) + 1;
 633       gimple_set_uid (stmt, uid);
 634       VEC_safe_push (vec_void_p, heap, stmt_vec_info_vec, (vec_void_p) info);
 635     }
 636   else
 637     VEC_replace (vec_void_p, stmt_vec_info_vec, uid - 1, (vec_void_p) info);
 638 }
 639
 640 static inline gimple
 641 get_earlier_stmt (gimple stmt1, gimple stmt2)
 642 {
 643   unsigned int uid1, uid2;
 644
 645   if (stmt1 == NULL)
 646     return stmt2;
 647
 648   if (stmt2 == NULL)
 649     return stmt1;
 650
 651   uid1 = gimple_uid (stmt1);
 652   uid2 = gimple_uid (stmt2);
 653
 654   if (uid1 == 0 || uid2 == 0)
 655     return NULL;
 656
 657   gcc_assert (uid1 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 658   gcc_assert (uid2 <= VEC_length (vec_void_p, stmt_vec_info_vec));
 659
 660   if (uid1 < uid2)
 661     return stmt1;
 662   else
 663     return stmt2;
 664 }
 665
 666 static inline bool
 667 is_pattern_stmt_p (stmt_vec_info stmt_info)
 668 {
 669   gimple related_stmt;
 670   stmt_vec_info related_stmt_info;
 671
 672   related_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
 673   if (related_stmt
 674       && (related_stmt_info = vinfo_for_stmt (related_stmt))
 675       && STMT_VINFO_IN_PATTERN_P (related_stmt_info))
 676     return true;
 677
 678   return false;
 679 }
 680
 681 static inline bool
 682 is_loop_header_bb_p (basic_block bb)
 683 {
 684   if (bb == (bb->loop_father)->header)
 685     return true;
 686   gcc_assert (EDGE_COUNT (bb->preds) == 1);
 687   return false;
 688 }
 689
 690 static inline void
 691 stmt_vinfo_set_inside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 692                                     int cost)
 693 {
 694   if (slp_node)
 695     SLP_TREE_INSIDE_OF_LOOP_COST (slp_node) = cost;
 696   else
 697     STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) = cost;
 698 }
 699
 700 static inline void
 701 stmt_vinfo_set_outside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
 702                                      int cost)
 703 {
 704   if (slp_node)
 705     SLP_TREE_OUTSIDE_OF_LOOP_COST (slp_node) = cost;
 706   else
 707     STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = cost;
 708 }
 709
 710 static inline int
 711 vect_pow2 (int x)
 712 {
 713   int i, res = 1;
 714
 715   for (i = 0; i < x; i++)
 716     res *= 2;
 717
 718   return res;
 719 }
 720
 721 /*-----------------------------------------------------------------*/
 722 /* Info on data references alignment.                              */
 723 /*-----------------------------------------------------------------*/
 724
 725 /* Reflects actual alignment of first access in the vectorized loop,
 726    taking into account peeling/versioning if applied.  */
 727 #define DR_MISALIGNMENT(DR)   ((int) (size_t) (DR)->aux)
 728 #define SET_DR_MISALIGNMENT(DR, VAL)   ((DR)->aux = (void *) (size_t) (VAL))
 729
 730 static inline bool
 731 aligned_access_p (struct data_reference *data_ref_info)
 732 {
 733   return (DR_MISALIGNMENT (data_ref_info) == 0);
 734 }
 735
 736 static inline bool
 737 known_alignment_for_access_p (struct data_reference *data_ref_info)
 738 {
 739   return (DR_MISALIGNMENT (data_ref_info) != -1);
 740 }
 741
 742 /* vect_dump will be set to stderr or dump_file if exist.  */
 743 extern FILE *vect_dump;
 744 extern LOC vect_loop_location;
 745
 746 /*-----------------------------------------------------------------*/
 747 /* Function prototypes.                                            */
 748 /*-----------------------------------------------------------------*/
 749
 750 /* Simple loop peeling and versioning utilities for vectorizer's purposes -
 751    in tree-vect-loop-manip.c.  */
 752 extern void slpeel_make_loop_iterate_ntimes (struct loop *, tree);
 753 extern bool slpeel_can_duplicate_loop_p (const struct loop *, const_edge);
 754 extern void vect_loop_versioning (loop_vec_info, bool, tree *, gimple_seq *);
 755 extern void vect_do_peeling_for_loop_bound (loop_vec_info, tree *,
 756                                             tree, gimple_seq);
 757 extern void vect_do_peeling_for_alignment (loop_vec_info);
 758 extern LOC find_loop_location (struct loop *);
 759 extern bool vect_can_advance_ivs_p (loop_vec_info);
 760
 761 /* In tree-vect-stmts.c.  */
 762 extern tree get_vectype_for_scalar_type (tree);
 763 extern bool vect_is_simple_use (tree, loop_vec_info, bb_vec_info, gimple *,
 764                                 tree *,  enum vect_def_type *);
 765 extern bool supportable_widening_operation (enum tree_code, gimple, tree,
 766                                             tree *, tree *, enum tree_code *,
 767                                             enum tree_code *, int *,
 768                                             VEC (tree, heap) **);
 769 extern bool supportable_narrowing_operation (enum tree_code, const_gimple,
 770                                              tree, enum tree_code *, int *,
 771                                              VEC (tree, heap) **);
 772 extern stmt_vec_info new_stmt_vec_info (gimple stmt, loop_vec_info,
 773                                         bb_vec_info);
 774 extern void free_stmt_vec_info (gimple stmt);
 775 extern tree vectorizable_function (gimple, tree, tree);
 776 extern void vect_model_simple_cost (stmt_vec_info, int, enum vect_def_type *,
 777                                     slp_tree);
 778 extern void vect_model_store_cost (stmt_vec_info, int, enum vect_def_type,
 779                                    slp_tree);
 780 extern void vect_model_load_cost (stmt_vec_info, int, slp_tree);
 781 extern void vect_finish_stmt_generation (gimple, gimple,
 782                                          gimple_stmt_iterator *);
 783 extern bool vect_mark_stmts_to_be_vectorized (loop_vec_info);
 784 extern int cost_for_stmt (gimple);
 785 extern tree vect_get_vec_def_for_operand (tree, gimple, tree *);
 786 extern tree vect_init_vector (gimple, tree, tree,
 787                               gimple_stmt_iterator *);
 788 extern tree vect_get_vec_def_for_stmt_copy (enum vect_def_type, tree);
 789 extern bool vect_transform_stmt (gimple, gimple_stmt_iterator *,
 790                                  bool *, slp_tree, slp_instance);
 791 extern void vect_remove_stores (gimple);
 792 extern bool vect_analyze_stmt (gimple, bool *, slp_tree);
 793 extern bool vectorizable_condition (gimple, gimple_stmt_iterator *, gimple *,
 794                                     tree, int);
 795
 796 /* In tree-vect-data-refs.c.  */
 797 extern bool vect_can_force_dr_alignment_p (const_tree, unsigned int);
 798 extern enum dr_alignment_support vect_supportable_dr_alignment
 799                                            (struct data_reference *);
 800 extern tree vect_get_smallest_scalar_type (gimple, HOST_WIDE_INT *,
 801                                            HOST_WIDE_INT *);
 802 extern bool vect_analyze_data_ref_dependences (loop_vec_info, bb_vec_info);
 803 extern bool vect_enhance_data_refs_alignment (loop_vec_info);
 804 extern bool vect_analyze_data_refs_alignment (loop_vec_info, bb_vec_info);
 805 extern bool vect_verify_datarefs_alignment (loop_vec_info, bb_vec_info);
 806 extern bool vect_analyze_data_ref_accesses (loop_vec_info, bb_vec_info);
 807 extern bool vect_prune_runtime_alias_test_list (loop_vec_info);
 808 extern bool vect_analyze_data_refs (loop_vec_info, bb_vec_info);
 809 extern tree vect_create_data_ref_ptr (gimple, struct loop *, tree, tree *,
 810                                       gimple *, bool, bool *);
 811 extern tree bump_vector_ptr (tree, gimple, gimple_stmt_iterator *, gimple, tree);
 812 extern tree vect_create_destination_var (tree, tree);
 813 extern bool vect_strided_store_supported (tree);
 814 extern bool vect_strided_load_supported (tree);
 815 extern bool vect_permute_store_chain (VEC(tree,heap) *,unsigned int, gimple,
 816                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 817 extern tree vect_setup_realignment (gimple, gimple_stmt_iterator *, tree *,
 818                                     enum dr_alignment_support, tree,
 819                                     struct loop **);
 820 extern bool vect_permute_load_chain (VEC(tree,heap) *,unsigned int, gimple,
 821                                     gimple_stmt_iterator *, VEC(tree,heap) **);
 822 extern bool vect_transform_strided_load (gimple, VEC(tree,heap) *, int,
 823                                          gimple_stmt_iterator *);
 824 extern int vect_get_place_in_interleaving_chain (gimple, gimple);
 825 extern tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
 826 extern tree vect_create_addr_base_for_vector_ref (gimple, gimple_seq *,
 827                                                   tree, struct loop *);
 828
 829 /* In tree-vect-loop.c.  */
 830 /* FORNOW: Used in tree-parloops.c.  */
 831 extern void destroy_loop_vec_info (loop_vec_info, bool);
 832 extern gimple vect_is_simple_reduction (loop_vec_info, gimple, bool, bool *);
 833 /* Drive for loop analysis stage.  */
 834 extern loop_vec_info vect_analyze_loop (struct loop *);
 835 /* Drive for loop transformation stage.  */
 836 extern void vect_transform_loop (loop_vec_info);
 837 extern loop_vec_info vect_analyze_loop_form (struct loop *);
 838 extern bool vectorizable_live_operation (gimple, gimple_stmt_iterator *,
 839                                          gimple *);
 840 extern bool vectorizable_reduction (gimple, gimple_stmt_iterator *, gimple *);
 841 extern bool vectorizable_induction (gimple, gimple_stmt_iterator *, gimple *);
 842 extern int vect_estimate_min_profitable_iters (loop_vec_info);
 843 extern tree get_initial_def_for_reduction (gimple, tree, tree *);
 844 extern int vect_min_worthwhile_factor (enum tree_code);
 845
 846
 847 /* In tree-vect-slp.c.  */
 848 extern void vect_free_slp_instance (slp_instance);
 849 extern bool vect_transform_slp_perm_load (gimple, VEC (tree, heap) *,
 850                                           gimple_stmt_iterator *, int,
 851                                           slp_instance, bool);
 852 extern bool vect_schedule_slp (loop_vec_info, bb_vec_info);
 853 extern void vect_update_slp_costs_according_to_vf (loop_vec_info);
 854 extern bool vect_analyze_slp (loop_vec_info, bb_vec_info);
 855 extern void vect_make_slp_decision (loop_vec_info);
 856 extern void vect_detect_hybrid_slp (loop_vec_info);
 857 extern void vect_get_slp_defs (slp_tree, VEC (tree,heap) **,
 858                                VEC (tree,heap) **);
 859 extern LOC find_bb_location (basic_block);
 860 extern bb_vec_info vect_slp_analyze_bb (basic_block);
 861 extern void vect_slp_transform_bb (basic_block);
 862
 863 /* In tree-vect-patterns.c.  */
 864 /* Pattern recognition functions.
 865    Additional pattern recognition functions can (and will) be added
 866    in the future.  */
 867 typedef gimple (* vect_recog_func_ptr) (gimple, tree *, tree *);
 868 #define NUM_PATTERNS 4
 869 void vect_pattern_recog (loop_vec_info);
 870
 871 /* In tree-vectorizer.c.  */
 872 unsigned vectorize_loops (void);
 873 /* Vectorization debug information */
 874 extern bool vect_print_dump_info (enum verbosity_levels);
 875
 876 #endif  /* GCC_TREE_VECTORIZER_H  */