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>
6 This file is part of GCC.
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
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
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/>. */
22 #ifndef GCC_TREE_VECTORIZER_H
23 #define GCC_TREE_VECTORIZER_H
25 #include "tree-data-ref.h"
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)
33 /* Used for naming of new temporaries. */
40 /* Defines type of operation. */
47 /* Define type of available alignment support. */
48 enum dr_alignment_support
{
49 dr_unaligned_unsupported
,
50 dr_unaligned_supported
,
52 dr_explicit_realign_optimized
,
56 /* Define type of def-use cross-iteration cycle. */
58 vect_uninitialized_def
= 0,
59 vect_constant_def
= 1,
64 vect_double_reduction_def
,
69 #define VECTORIZABLE_CYCLE_DEF(D) (((D) == vect_reduction_def) \
70 || ((D) == vect_double_reduction_def) \
71 || ((D) == vect_nested_cycle))
73 /* Define verbosity levels. */
74 enum verbosity_levels
{
76 REPORT_VECTORIZED_LOCATIONS
,
77 REPORT_UNVECTORIZED_LOCATIONS
,
81 REPORT_BAD_FORM_LOOPS
,
85 /* New verbosity levels should be added before this one. */
89 /************************************************************************
91 ************************************************************************/
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
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. */
113 int outside_of_loop
; /* Statements generated outside loop. */
114 int inside_of_loop
; /* Statements generated inside loop. */
119 DEF_VEC_ALLOC_P(slp_tree
, heap
);
121 /* SLP instance is a sequence of stmts in a loop that can be packed into
123 typedef struct _slp_instance
{
124 /* The root of SLP tree. */
127 /* Size of groups of scalar stmts that will be replaced by SIMD stmt/s. */
128 unsigned int group_size
;
130 /* The unrolling factor required to vectorized this SLP instance. */
131 unsigned int unrolling_factor
;
133 /* Vectorization costs associated with SLP instance. */
136 int outside_of_loop
; /* Statements generated outside loop. */
137 int inside_of_loop
; /* Statements generated inside loop. */
140 /* Loads permutation relatively to the stores, NULL if there is no
142 VEC (int, heap
) *load_permutation
;
144 /* The group of nodes that contain loads of this SLP instance. */
145 VEC (slp_tree
, heap
) *loads
;
147 /* The first scalar load of the instance. The created vector loads will be
148 inserted before this statement. */
152 DEF_VEC_P(slp_instance
);
153 DEF_VEC_ALLOC_P(slp_instance
, heap
);
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
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
173 /*-----------------------------------------------------------------*/
174 /* Info on vectorized loops. */
175 /*-----------------------------------------------------------------*/
176 typedef struct _loop_vec_info
{
178 /* The loop to which this info struct refers to. */
181 /* The loop basic blocks. */
184 /* Number of iterations. */
186 tree num_iters_unchanged
;
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
;
195 /* Is the loop vectorizable? */
198 /* Unrolling factor */
199 int vectorization_factor
;
201 /* Unknown DRs according to which loop was peeled. */
202 struct data_reference
*unaligned_dr
;
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
212 int peeling_for_alignment
;
214 /* The mask used to check the alignment of pointers or arrays. */
217 /* All data references in the loop. */
218 VEC (data_reference_p
, heap
) *datarefs
;
220 /* All data dependences in the loop. */
221 VEC (ddr_p
, heap
) *ddrs
;
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
;
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
;
231 /* The loop location in the source. */
232 LOC loop_line_number
;
234 /* All interleaving chains of stores in the loop, represented by the first
235 stmt in the chain. */
236 VEC(gimple
, heap
) *strided_stores
;
238 /* All SLP instances in the loop. This is a subset of the set of STRIDED_STORES
240 VEC(slp_instance
, heap
) *slp_instances
;
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
;
246 /* Reduction cycles detected in the loop. Used in loop-aware SLP. */
247 VEC (gimple
, heap
) *reductions
;
250 /* Access Functions. */
251 #define LOOP_VINFO_LOOP(L) (L)->loop
252 #define LOOP_VINFO_BBS(L) (L)->bbs
253 #define LOOP_VINFO_NITERS(L) (L)->num_iters
254 /* Since LOOP_VINFO_NITERS can change after prologue peeling
255 retain total unchanged scalar loop iterations for cost model. */
256 #define LOOP_VINFO_NITERS_UNCHANGED(L) (L)->num_iters_unchanged
257 #define LOOP_VINFO_COST_MODEL_MIN_ITERS(L) (L)->min_profitable_iters
258 #define LOOP_VINFO_VECTORIZABLE_P(L) (L)->vectorizable
259 #define LOOP_VINFO_VECT_FACTOR(L) (L)->vectorization_factor
260 #define LOOP_VINFO_PTR_MASK(L) (L)->ptr_mask
261 #define LOOP_VINFO_DATAREFS(L) (L)->datarefs
262 #define LOOP_VINFO_DDRS(L) (L)->ddrs
263 #define LOOP_VINFO_INT_NITERS(L) (TREE_INT_CST_LOW ((L)->num_iters))
264 #define LOOP_PEELING_FOR_ALIGNMENT(L) (L)->peeling_for_alignment
265 #define LOOP_VINFO_UNALIGNED_DR(L) (L)->unaligned_dr
266 #define LOOP_VINFO_MAY_MISALIGN_STMTS(L) (L)->may_misalign_stmts
267 #define LOOP_VINFO_LOC(L) (L)->loop_line_number
268 #define LOOP_VINFO_MAY_ALIAS_DDRS(L) (L)->may_alias_ddrs
269 #define LOOP_VINFO_STRIDED_STORES(L) (L)->strided_stores
270 #define LOOP_VINFO_SLP_INSTANCES(L) (L)->slp_instances
271 #define LOOP_VINFO_SLP_UNROLLING_FACTOR(L) (L)->slp_unrolling_factor
272 #define LOOP_VINFO_REDUCTIONS(L) (L)->reductions
274 #define LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT(L) \
275 VEC_length (gimple, (L)->may_misalign_stmts) > 0
276 #define LOOP_REQUIRES_VERSIONING_FOR_ALIAS(L) \
277 VEC_length (ddr_p, (L)->may_alias_ddrs) > 0
279 #define NITERS_KNOWN_P(n) \
280 (host_integerp ((n),0) \
281 && TREE_INT_CST_LOW ((n)) > 0)
283 #define LOOP_VINFO_NITERS_KNOWN_P(L) \
284 NITERS_KNOWN_P((L)->num_iters)
286 static inline loop_vec_info
287 loop_vec_info_for_loop (struct loop
*loop
)
289 return (loop_vec_info
) loop
->aux
;
293 nested_in_vect_loop_p (struct loop
*loop
, gimple stmt
)
296 && (loop
->inner
== (gimple_bb (stmt
))->loop_father
));
299 typedef struct _bb_vec_info
{
302 /* All interleaving chains of stores in the basic block, represented by the
303 first stmt in the chain. */
304 VEC(gimple
, heap
) *strided_stores
;
306 /* All SLP instances in the basic block. This is a subset of the set of
307 STRIDED_STORES of the basic block. */
308 VEC(slp_instance
, heap
) *slp_instances
;
310 /* All data references in the basic block. */
311 VEC (data_reference_p
, heap
) *datarefs
;
313 /* All data dependences in the basic block. */
314 VEC (ddr_p
, heap
) *ddrs
;
317 #define BB_VINFO_BB(B) (B)->bb
318 #define BB_VINFO_STRIDED_STORES(B) (B)->strided_stores
319 #define BB_VINFO_SLP_INSTANCES(B) (B)->slp_instances
320 #define BB_VINFO_DATAREFS(B) (B)->datarefs
321 #define BB_VINFO_DDRS(B) (B)->ddrs
323 static inline bb_vec_info
324 vec_info_for_bb (basic_block bb
)
326 return (bb_vec_info
) bb
->aux
;
329 /*-----------------------------------------------------------------*/
330 /* Info on vectorized defs. */
331 /*-----------------------------------------------------------------*/
332 enum stmt_vec_info_type
{
333 undef_vec_info_type
= 0,
338 assignment_vec_info_type
,
339 condition_vec_info_type
,
342 type_promotion_vec_info_type
,
343 type_demotion_vec_info_type
,
344 type_conversion_vec_info_type
,
345 loop_exit_ctrl_vec_info_type
348 /* Indicates whether/how a variable is used in the scope of loop/basic
351 vect_unused_in_scope
= 0,
352 /* The def is in the inner loop, and the use is in the outer loop, and the
353 use is a reduction stmt. */
354 vect_used_in_outer_by_reduction
,
355 /* The def is in the inner loop, and the use is in the outer loop (and is
356 not part of reduction). */
359 /* defs that feed computations that end up (only) in a reduction. These
360 defs may be used by non-reduction stmts, but eventually, any
361 computations/values that are affected by these defs are used to compute
362 a reduction (i.e. don't get stored to memory, for example). We use this
363 to identify computations that we can change the order in which they are
365 vect_used_by_reduction
,
370 /* The type of vectorization that can be applied to the stmt: regular loop-based
371 vectorization; pure SLP - the stmt is a part of SLP instances and does not
372 have uses outside SLP instances; or hybrid SLP and loop-based - the stmt is
373 a part of SLP instance and also must be loop-based vectorized, since it has
374 uses outside SLP sequences.
376 In the loop context the meanings of pure and hybrid SLP are slightly
377 different. By saying that pure SLP is applied to the loop, we mean that we
378 exploit only intra-iteration parallelism in the loop; i.e., the loop can be
379 vectorized without doing any conceptual unrolling, cause we don't pack
380 together stmts from different iterations, only within a single iteration.
381 Loop hybrid SLP means that we exploit both intra-iteration and
382 inter-iteration parallelism (e.g., number of elements in the vector is 4
383 and the slp-group-size is 2, in which case we don't have enough parallelism
384 within an iteration, so we obtain the rest of the parallelism from subsequent
385 iterations by unrolling the loop by 2). */
393 typedef struct data_reference
*dr_p
;
395 DEF_VEC_ALLOC_P(dr_p
,heap
);
397 typedef struct _stmt_vec_info
{
399 enum stmt_vec_info_type type
;
401 /* The stmt to which this info struct refers to. */
404 /* The loop_vec_info with respect to which STMT is vectorized. */
405 loop_vec_info loop_vinfo
;
407 /* Not all stmts in the loop need to be vectorized. e.g, the increment
408 of the loop induction variable and computation of array indexes. relevant
409 indicates whether the stmt needs to be vectorized. */
410 enum vect_relevant relevant
;
412 /* Indicates whether this stmts is part of a computation whose result is
413 used outside the loop. */
416 /* The vector type to be used for the LHS of this statement. */
419 /* The vectorized version of the stmt. */
420 gimple vectorized_stmt
;
423 /** The following is relevant only for stmts that contain a non-scalar
424 data-ref (array/pointer/struct access). A GIMPLE stmt is expected to have
425 at most one such data-ref. **/
427 /* Information about the data-ref (access function, etc),
428 relative to the inner-most containing loop. */
429 struct data_reference
*data_ref_info
;
431 /* Information about the data-ref relative to this loop
432 nest (the loop that is being considered for vectorization). */
433 tree dr_base_address
;
439 /* Stmt is part of some pattern (computation idiom) */
442 /* Used for various bookkeeping purposes, generally holding a pointer to
443 some other stmt S that is in some way "related" to this stmt.
444 Current use of this field is:
445 If this stmt is part of a pattern (i.e. the field 'in_pattern_p' is
446 true): S is the "pattern stmt" that represents (and replaces) the
447 sequence of stmts that constitutes the pattern. Similarly, the
448 related_stmt of the "pattern stmt" points back to this stmt (which is
449 the last stmt in the original sequence of stmts that constitutes the
453 /* List of datarefs that are known to have the same alignment as the dataref
455 VEC(dr_p
,heap
) *same_align_refs
;
457 /* Classify the def of this stmt. */
458 enum vect_def_type def_type
;
460 /* Interleaving info. */
461 /* First data-ref in the interleaving group. */
463 /* Pointer to the next data-ref in the group. */
465 /* The size of the interleaving group. */
467 /* For stores, number of stores from this group seen. We vectorize the last
469 unsigned int store_count
;
470 /* For loads only, the gap from the previous load. For consecutive loads, GAP
473 /* In case that two or more stmts share data-ref, this is the pointer to the
474 previously detected stmt with the same dr. */
476 /* For loads only, if there is a store with the same location, this field is
480 /* Vectorization costs associated with statement. */
483 int outside_of_loop
; /* Statements generated outside loop. */
484 int inside_of_loop
; /* Statements generated inside loop. */
487 /* Whether the stmt is SLPed, loop-based vectorized, or both. */
488 enum slp_vect_type slp_type
;
490 /* The bb_vec_info with respect to which STMT is vectorized. */
491 bb_vec_info bb_vinfo
;
494 /* Access Functions. */
495 #define STMT_VINFO_TYPE(S) (S)->type
496 #define STMT_VINFO_STMT(S) (S)->stmt
497 #define STMT_VINFO_LOOP_VINFO(S) (S)->loop_vinfo
498 #define STMT_VINFO_BB_VINFO(S) (S)->bb_vinfo
499 #define STMT_VINFO_RELEVANT(S) (S)->relevant
500 #define STMT_VINFO_LIVE_P(S) (S)->live
501 #define STMT_VINFO_VECTYPE(S) (S)->vectype
502 #define STMT_VINFO_VEC_STMT(S) (S)->vectorized_stmt
503 #define STMT_VINFO_DATA_REF(S) (S)->data_ref_info
505 #define STMT_VINFO_DR_BASE_ADDRESS(S) (S)->dr_base_address
506 #define STMT_VINFO_DR_INIT(S) (S)->dr_init
507 #define STMT_VINFO_DR_OFFSET(S) (S)->dr_offset
508 #define STMT_VINFO_DR_STEP(S) (S)->dr_step
509 #define STMT_VINFO_DR_ALIGNED_TO(S) (S)->dr_aligned_to
511 #define STMT_VINFO_IN_PATTERN_P(S) (S)->in_pattern_p
512 #define STMT_VINFO_RELATED_STMT(S) (S)->related_stmt
513 #define STMT_VINFO_SAME_ALIGN_REFS(S) (S)->same_align_refs
514 #define STMT_VINFO_DEF_TYPE(S) (S)->def_type
515 #define STMT_VINFO_DR_GROUP_FIRST_DR(S) (S)->first_dr
516 #define STMT_VINFO_DR_GROUP_NEXT_DR(S) (S)->next_dr
517 #define STMT_VINFO_DR_GROUP_SIZE(S) (S)->size
518 #define STMT_VINFO_DR_GROUP_STORE_COUNT(S) (S)->store_count
519 #define STMT_VINFO_DR_GROUP_GAP(S) (S)->gap
520 #define STMT_VINFO_DR_GROUP_SAME_DR_STMT(S)(S)->same_dr_stmt
521 #define STMT_VINFO_DR_GROUP_READ_WRITE_DEPENDENCE(S) (S)->read_write_dep
522 #define STMT_VINFO_STRIDED_ACCESS(S) ((S)->first_dr != NULL)
524 #define DR_GROUP_FIRST_DR(S) (S)->first_dr
525 #define DR_GROUP_NEXT_DR(S) (S)->next_dr
526 #define DR_GROUP_SIZE(S) (S)->size
527 #define DR_GROUP_STORE_COUNT(S) (S)->store_count
528 #define DR_GROUP_GAP(S) (S)->gap
529 #define DR_GROUP_SAME_DR_STMT(S) (S)->same_dr_stmt
530 #define DR_GROUP_READ_WRITE_DEPENDENCE(S) (S)->read_write_dep
532 #define STMT_VINFO_RELEVANT_P(S) ((S)->relevant != vect_unused_in_scope)
533 #define STMT_VINFO_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
534 #define STMT_VINFO_INSIDE_OF_LOOP_COST(S) (S)->cost.inside_of_loop
536 #define HYBRID_SLP_STMT(S) ((S)->slp_type == hybrid)
537 #define PURE_SLP_STMT(S) ((S)->slp_type == pure_slp)
538 #define STMT_SLP_TYPE(S) (S)->slp_type
540 /* These are some defines for the initial implementation of the vectorizer's
541 cost model. These will later be target specific hooks. */
543 /* Cost of conditional taken branch. */
544 #ifndef TARG_COND_TAKEN_BRANCH_COST
545 #define TARG_COND_TAKEN_BRANCH_COST 3
548 /* Cost of conditional not taken branch. */
549 #ifndef TARG_COND_NOT_TAKEN_BRANCH_COST
550 #define TARG_COND_NOT_TAKEN_BRANCH_COST 1
553 /* Cost of any scalar operation, excluding load and store. */
554 #ifndef TARG_SCALAR_STMT_COST
555 #define TARG_SCALAR_STMT_COST 1
558 /* Cost of scalar load. */
559 #ifndef TARG_SCALAR_LOAD_COST
560 #define TARG_SCALAR_LOAD_COST 1
563 /* Cost of scalar store. */
564 #ifndef TARG_SCALAR_STORE_COST
565 #define TARG_SCALAR_STORE_COST 1
568 /* Cost of any vector operation, excluding load, store or vector to scalar
570 #ifndef TARG_VEC_STMT_COST
571 #define TARG_VEC_STMT_COST 1
574 /* Cost of vector to scalar operation. */
575 #ifndef TARG_VEC_TO_SCALAR_COST
576 #define TARG_VEC_TO_SCALAR_COST 1
579 /* Cost of scalar to vector operation. */
580 #ifndef TARG_SCALAR_TO_VEC_COST
581 #define TARG_SCALAR_TO_VEC_COST 1
584 /* Cost of aligned vector load. */
585 #ifndef TARG_VEC_LOAD_COST
586 #define TARG_VEC_LOAD_COST 1
589 /* Cost of misaligned vector load. */
590 #ifndef TARG_VEC_UNALIGNED_LOAD_COST
591 #define TARG_VEC_UNALIGNED_LOAD_COST 2
594 /* Cost of vector store. */
595 #ifndef TARG_VEC_STORE_COST
596 #define TARG_VEC_STORE_COST 1
599 /* Cost of vector permutation. */
600 #ifndef TARG_VEC_PERMUTE_COST
601 #define TARG_VEC_PERMUTE_COST 1
604 /* The maximum number of intermediate steps required in multi-step type
606 #define MAX_INTERM_CVT_STEPS 3
608 /* The maximum vectorization factor supported by any target (V32QI). */
609 #define MAX_VECTORIZATION_FACTOR 32
611 /* Avoid GTY(()) on stmt_vec_info. */
612 typedef void *vec_void_p
;
613 DEF_VEC_P (vec_void_p
);
614 DEF_VEC_ALLOC_P (vec_void_p
, heap
);
616 extern VEC(vec_void_p
,heap
) *stmt_vec_info_vec
;
618 void init_stmt_vec_info_vec (void);
619 void free_stmt_vec_info_vec (void);
621 static inline stmt_vec_info
622 vinfo_for_stmt (gimple stmt
)
624 unsigned int uid
= gimple_uid (stmt
);
628 gcc_assert (uid
<= VEC_length (vec_void_p
, stmt_vec_info_vec
));
629 return (stmt_vec_info
) VEC_index (vec_void_p
, stmt_vec_info_vec
, uid
- 1);
633 set_vinfo_for_stmt (gimple stmt
, stmt_vec_info info
)
635 unsigned int uid
= gimple_uid (stmt
);
639 uid
= VEC_length (vec_void_p
, stmt_vec_info_vec
) + 1;
640 gimple_set_uid (stmt
, uid
);
641 VEC_safe_push (vec_void_p
, heap
, stmt_vec_info_vec
, (vec_void_p
) info
);
644 VEC_replace (vec_void_p
, stmt_vec_info_vec
, uid
- 1, (vec_void_p
) info
);
648 get_earlier_stmt (gimple stmt1
, gimple stmt2
)
650 unsigned int uid1
, uid2
;
658 uid1
= gimple_uid (stmt1
);
659 uid2
= gimple_uid (stmt2
);
661 if (uid1
== 0 || uid2
== 0)
664 gcc_assert (uid1
<= VEC_length (vec_void_p
, stmt_vec_info_vec
));
665 gcc_assert (uid2
<= VEC_length (vec_void_p
, stmt_vec_info_vec
));
674 is_pattern_stmt_p (stmt_vec_info stmt_info
)
677 stmt_vec_info related_stmt_info
;
679 related_stmt
= STMT_VINFO_RELATED_STMT (stmt_info
);
681 && (related_stmt_info
= vinfo_for_stmt (related_stmt
))
682 && STMT_VINFO_IN_PATTERN_P (related_stmt_info
))
689 is_loop_header_bb_p (basic_block bb
)
691 if (bb
== (bb
->loop_father
)->header
)
693 gcc_assert (EDGE_COUNT (bb
->preds
) == 1);
698 stmt_vinfo_set_inside_of_loop_cost (stmt_vec_info stmt_info
, slp_tree slp_node
,
702 SLP_TREE_INSIDE_OF_LOOP_COST (slp_node
) = cost
;
704 STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info
) = cost
;
708 stmt_vinfo_set_outside_of_loop_cost (stmt_vec_info stmt_info
, slp_tree slp_node
,
712 SLP_TREE_OUTSIDE_OF_LOOP_COST (slp_node
) = cost
;
714 STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info
) = cost
;
722 for (i
= 0; i
< x
; i
++)
728 /*-----------------------------------------------------------------*/
729 /* Info on data references alignment. */
730 /*-----------------------------------------------------------------*/
732 /* Reflects actual alignment of first access in the vectorized loop,
733 taking into account peeling/versioning if applied. */
734 #define DR_MISALIGNMENT(DR) ((int) (size_t) (DR)->aux)
735 #define SET_DR_MISALIGNMENT(DR, VAL) ((DR)->aux = (void *) (size_t) (VAL))
738 aligned_access_p (struct data_reference
*data_ref_info
)
740 return (DR_MISALIGNMENT (data_ref_info
) == 0);
744 known_alignment_for_access_p (struct data_reference
*data_ref_info
)
746 return (DR_MISALIGNMENT (data_ref_info
) != -1);
749 /* vect_dump will be set to stderr or dump_file if exist. */
750 extern FILE *vect_dump
;
751 extern LOC vect_loop_location
;
753 /*-----------------------------------------------------------------*/
754 /* Function prototypes. */
755 /*-----------------------------------------------------------------*/
757 /* Simple loop peeling and versioning utilities for vectorizer's purposes -
758 in tree-vect-loop-manip.c. */
759 extern void slpeel_make_loop_iterate_ntimes (struct loop
*, tree
);
760 extern bool slpeel_can_duplicate_loop_p (const struct loop
*, const_edge
);
761 extern void vect_loop_versioning (loop_vec_info
, bool, tree
*, gimple_seq
*);
762 extern void vect_do_peeling_for_loop_bound (loop_vec_info
, tree
*,
764 extern void vect_do_peeling_for_alignment (loop_vec_info
);
765 extern LOC
find_loop_location (struct loop
*);
766 extern bool vect_can_advance_ivs_p (loop_vec_info
);
768 /* In tree-vect-stmts.c. */
769 extern tree
get_vectype_for_scalar_type (tree
);
770 extern tree
get_same_sized_vectype (tree
, tree
);
771 extern bool vect_is_simple_use (tree
, loop_vec_info
, bb_vec_info
, gimple
*,
772 tree
*, enum vect_def_type
*);
773 extern bool vect_is_simple_use_1 (tree
, loop_vec_info
, bb_vec_info
, gimple
*,
774 tree
*, enum vect_def_type
*, tree
*);
775 extern bool supportable_widening_operation (enum tree_code
, gimple
, tree
, tree
,
776 tree
*, tree
*, enum tree_code
*,
777 enum tree_code
*, int *,
778 VEC (tree
, heap
) **);
779 extern bool supportable_narrowing_operation (enum tree_code
, tree
, tree
,
781 int *, VEC (tree
, heap
) **);
782 extern stmt_vec_info
new_stmt_vec_info (gimple stmt
, loop_vec_info
,
784 extern void free_stmt_vec_info (gimple stmt
);
785 extern tree
vectorizable_function (gimple
, tree
, tree
);
786 extern void vect_model_simple_cost (stmt_vec_info
, int, enum vect_def_type
*,
788 extern void vect_model_store_cost (stmt_vec_info
, int, enum vect_def_type
,
790 extern void vect_model_load_cost (stmt_vec_info
, int, slp_tree
);
791 extern void vect_finish_stmt_generation (gimple
, gimple
,
792 gimple_stmt_iterator
*);
793 extern bool vect_mark_stmts_to_be_vectorized (loop_vec_info
);
794 extern int cost_for_stmt (gimple
);
795 extern tree
vect_get_vec_def_for_operand (tree
, gimple
, tree
*);
796 extern tree
vect_init_vector (gimple
, tree
, tree
,
797 gimple_stmt_iterator
*);
798 extern tree
vect_get_vec_def_for_stmt_copy (enum vect_def_type
, tree
);
799 extern bool vect_transform_stmt (gimple
, gimple_stmt_iterator
*,
800 bool *, slp_tree
, slp_instance
);
801 extern void vect_remove_stores (gimple
);
802 extern bool vect_analyze_stmt (gimple
, bool *, slp_tree
);
803 extern bool vectorizable_condition (gimple
, gimple_stmt_iterator
*, gimple
*,
806 /* In tree-vect-data-refs.c. */
807 extern bool vect_can_force_dr_alignment_p (const_tree
, unsigned int);
808 extern enum dr_alignment_support vect_supportable_dr_alignment
809 (struct data_reference
*);
810 extern tree
vect_get_smallest_scalar_type (gimple
, HOST_WIDE_INT
*,
812 extern bool vect_analyze_data_ref_dependences (loop_vec_info
, bb_vec_info
,
814 extern bool vect_enhance_data_refs_alignment (loop_vec_info
);
815 extern bool vect_analyze_data_refs_alignment (loop_vec_info
, bb_vec_info
);
816 extern bool vect_verify_datarefs_alignment (loop_vec_info
, bb_vec_info
);
817 extern bool vect_analyze_data_ref_accesses (loop_vec_info
, bb_vec_info
);
818 extern bool vect_prune_runtime_alias_test_list (loop_vec_info
);
819 extern bool vect_analyze_data_refs (loop_vec_info
, bb_vec_info
, int *);
820 extern tree
vect_create_data_ref_ptr (gimple
, struct loop
*, tree
, tree
*,
821 gimple
*, bool, bool *);
822 extern tree
bump_vector_ptr (tree
, gimple
, gimple_stmt_iterator
*, gimple
, tree
);
823 extern tree
vect_create_destination_var (tree
, tree
);
824 extern bool vect_strided_store_supported (tree
);
825 extern bool vect_strided_load_supported (tree
);
826 extern bool vect_permute_store_chain (VEC(tree
,heap
) *,unsigned int, gimple
,
827 gimple_stmt_iterator
*, VEC(tree
,heap
) **);
828 extern tree
vect_setup_realignment (gimple
, gimple_stmt_iterator
*, tree
*,
829 enum dr_alignment_support
, tree
,
831 extern bool vect_permute_load_chain (VEC(tree
,heap
) *,unsigned int, gimple
,
832 gimple_stmt_iterator
*, VEC(tree
,heap
) **);
833 extern bool vect_transform_strided_load (gimple
, VEC(tree
,heap
) *, int,
834 gimple_stmt_iterator
*);
835 extern int vect_get_place_in_interleaving_chain (gimple
, gimple
);
836 extern tree
vect_get_new_vect_var (tree
, enum vect_var_kind
, const char *);
837 extern tree
vect_create_addr_base_for_vector_ref (gimple
, gimple_seq
*,
838 tree
, struct loop
*);
840 /* In tree-vect-loop.c. */
841 /* FORNOW: Used in tree-parloops.c. */
842 extern void destroy_loop_vec_info (loop_vec_info
, bool);
843 extern gimple
vect_is_simple_reduction (loop_vec_info
, gimple
, bool, bool *);
844 /* Drive for loop analysis stage. */
845 extern loop_vec_info
vect_analyze_loop (struct loop
*);
846 /* Drive for loop transformation stage. */
847 extern void vect_transform_loop (loop_vec_info
);
848 extern loop_vec_info
vect_analyze_loop_form (struct loop
*);
849 extern bool vectorizable_live_operation (gimple
, gimple_stmt_iterator
*,
851 extern bool vectorizable_reduction (gimple
, gimple_stmt_iterator
*, gimple
*,
853 extern bool vectorizable_induction (gimple
, gimple_stmt_iterator
*, gimple
*);
854 extern int vect_estimate_min_profitable_iters (loop_vec_info
);
855 extern tree
get_initial_def_for_reduction (gimple
, tree
, tree
*);
856 extern int vect_min_worthwhile_factor (enum tree_code
);
859 /* In tree-vect-slp.c. */
860 extern void vect_free_slp_instance (slp_instance
);
861 extern bool vect_transform_slp_perm_load (gimple
, VEC (tree
, heap
) *,
862 gimple_stmt_iterator
*, int,
864 extern bool vect_schedule_slp (loop_vec_info
, bb_vec_info
);
865 extern void vect_update_slp_costs_according_to_vf (loop_vec_info
);
866 extern bool vect_analyze_slp (loop_vec_info
, bb_vec_info
);
867 extern void vect_make_slp_decision (loop_vec_info
);
868 extern void vect_detect_hybrid_slp (loop_vec_info
);
869 extern void vect_get_slp_defs (slp_tree
, VEC (tree
,heap
) **,
870 VEC (tree
,heap
) **, int);
871 extern LOC
find_bb_location (basic_block
);
872 extern bb_vec_info
vect_slp_analyze_bb (basic_block
);
873 extern void vect_slp_transform_bb (basic_block
);
875 /* In tree-vect-patterns.c. */
876 /* Pattern recognition functions.
877 Additional pattern recognition functions can (and will) be added
879 typedef gimple (* vect_recog_func_ptr
) (gimple
, tree
*, tree
*);
880 #define NUM_PATTERNS 4
881 void vect_pattern_recog (loop_vec_info
);
883 /* In tree-vectorizer.c. */
884 unsigned vectorize_loops (void);
885 /* Vectorization debug information */
886 extern bool vect_print_dump_info (enum verbosity_levels
);
888 #endif /* GCC_TREE_VECTORIZER_H */