re PR preprocessor/36674 (#include location is offset by one row in errors from prepr...
[gcc.git] / gcc / tree-vectorizer.h
1 /* Vectorizer
2 Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009 Free
3 Software Foundation, Inc.
4 Contributed by Dorit Naishlos <dorit@il.ibm.com>
5
6 This file is part of GCC.
7
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
11 version.
12
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
21
22 #ifndef GCC_TREE_VECTORIZER_H
23 #define GCC_TREE_VECTORIZER_H
24
25 #include "tree-data-ref.h"
26
27 typedef source_location LOC;
28 #define UNKNOWN_LOC UNKNOWN_LOCATION
29 #define EXPR_LOC(e) EXPR_LOCATION(e)
30 #define LOC_FILE(l) LOCATION_FILE (l)
31 #define LOC_LINE(l) LOCATION_LINE (l)
32
33 /* Used for naming of new temporaries. */
34 enum vect_var_kind {
35 vect_simple_var,
36 vect_pointer_var,
37 vect_scalar_var
38 };
39
40 /* Defines type of operation. */
41 enum operation_type {
42 unary_op = 1,
43 binary_op,
44 ternary_op
45 };
46
47 /* Define type of available alignment support. */
48 enum dr_alignment_support {
49 dr_unaligned_unsupported,
50 dr_unaligned_supported,
51 dr_explicit_realign,
52 dr_explicit_realign_optimized,
53 dr_aligned
54 };
55
56 /* Define type of def-use cross-iteration cycle. */
57 enum vect_def_type {
58 vect_uninitialized_def = 0,
59 vect_constant_def = 1,
60 vect_external_def,
61 vect_internal_def,
62 vect_induction_def,
63 vect_reduction_def,
64 vect_unknown_def_type
65 };
66
67 /* Define verbosity levels. */
68 enum verbosity_levels {
69 REPORT_NONE,
70 REPORT_VECTORIZED_LOCATIONS,
71 REPORT_UNVECTORIZED_LOCATIONS,
72 REPORT_COST,
73 REPORT_ALIGNMENT,
74 REPORT_DR_DETAILS,
75 REPORT_BAD_FORM_LOOPS,
76 REPORT_OUTER_LOOPS,
77 REPORT_SLP,
78 REPORT_DETAILS,
79 /* New verbosity levels should be added before this one. */
80 MAX_VERBOSITY_LEVEL
81 };
82
83 /************************************************************************
84 SLP
85 ************************************************************************/
86
87 /* A computation tree of an SLP instance. Each node corresponds to a group of
88 stmts to be packed in a SIMD stmt. */
89 typedef struct _slp_tree {
90 /* Only binary and unary operations are supported. LEFT child corresponds to
91 the first operand and RIGHT child to the second if the operation is
92 binary. */
93 struct _slp_tree *left;
94 struct _slp_tree *right;
95 /* A group of scalar stmts to be vectorized together. */
96 VEC (gimple, heap) *stmts;
97 /* Vectorized stmt/s. */
98 VEC (gimple, heap) *vec_stmts;
99 /* Number of vector stmts that are created to replace the group of scalar
100 stmts. It is calculated during the transformation phase as the number of
101 scalar elements in one scalar iteration (GROUP_SIZE) multiplied by VF
102 divided by vector size. */
103 unsigned int vec_stmts_size;
104 /* Vectorization costs associated with SLP node. */
105 struct
106 {
107 int outside_of_loop; /* Statements generated outside loop. */
108 int inside_of_loop; /* Statements generated inside loop. */
109 } cost;
110 } *slp_tree;
111
112 DEF_VEC_P(slp_tree);
113 DEF_VEC_ALLOC_P(slp_tree, heap);
114
115 /* SLP instance is a sequence of stmts in a loop that can be packed into
116 SIMD stmts. */
117 typedef struct _slp_instance {
118 /* The root of SLP tree. */
119 slp_tree root;
120
121 /* Size of groups of scalar stmts that will be replaced by SIMD stmt/s. */
122 unsigned int group_size;
123
124 /* The unrolling factor required to vectorized this SLP instance. */
125 unsigned int unrolling_factor;
126
127 /* Vectorization costs associated with SLP instance. */
128 struct
129 {
130 int outside_of_loop; /* Statements generated outside loop. */
131 int inside_of_loop; /* Statements generated inside loop. */
132 } cost;
133
134 /* Loads permutation relatively to the stores, NULL if there is no
135 permutation. */
136 VEC (int, heap) *load_permutation;
137
138 /* The group of nodes that contain loads of this SLP instance. */
139 VEC (slp_tree, heap) *loads;
140
141 /* The first scalar load of the instance. The created vector loads will be
142 inserted before this statement. */
143 gimple first_load;
144 } *slp_instance;
145
146 DEF_VEC_P(slp_instance);
147 DEF_VEC_ALLOC_P(slp_instance, heap);
148
149 /* Access Functions. */
150 #define SLP_INSTANCE_TREE(S) (S)->root
151 #define SLP_INSTANCE_GROUP_SIZE(S) (S)->group_size
152 #define SLP_INSTANCE_UNROLLING_FACTOR(S) (S)->unrolling_factor
153 #define SLP_INSTANCE_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
154 #define SLP_INSTANCE_INSIDE_OF_LOOP_COST(S) (S)->cost.inside_of_loop
155 #define SLP_INSTANCE_LOAD_PERMUTATION(S) (S)->load_permutation
156 #define SLP_INSTANCE_LOADS(S) (S)->loads
157 #define SLP_INSTANCE_FIRST_LOAD_STMT(S) (S)->first_load
158
159 #define SLP_TREE_LEFT(S) (S)->left
160 #define SLP_TREE_RIGHT(S) (S)->right
161 #define SLP_TREE_SCALAR_STMTS(S) (S)->stmts
162 #define SLP_TREE_VEC_STMTS(S) (S)->vec_stmts
163 #define SLP_TREE_NUMBER_OF_VEC_STMTS(S) (S)->vec_stmts_size
164 #define SLP_TREE_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
165 #define SLP_TREE_INSIDE_OF_LOOP_COST(S) (S)->cost.inside_of_loop
166
167 /*-----------------------------------------------------------------*/
168 /* Info on vectorized loops. */
169 /*-----------------------------------------------------------------*/
170 typedef struct _loop_vec_info {
171
172 /* The loop to which this info struct refers to. */
173 struct loop *loop;
174
175 /* The loop basic blocks. */
176 basic_block *bbs;
177
178 /* Number of iterations. */
179 tree num_iters;
180 tree num_iters_unchanged;
181
182 /* Minimum number of iterations below which vectorization is expected to
183 not be profitable (as estimated by the cost model).
184 -1 indicates that vectorization will not be profitable.
185 FORNOW: This field is an int. Will be a tree in the future, to represent
186 values unknown at compile time. */
187 int min_profitable_iters;
188
189 /* Is the loop vectorizable? */
190 bool vectorizable;
191
192 /* Unrolling factor */
193 int vectorization_factor;
194
195 /* Unknown DRs according to which loop was peeled. */
196 struct data_reference *unaligned_dr;
197
198 /* peeling_for_alignment indicates whether peeling for alignment will take
199 place, and what the peeling factor should be:
200 peeling_for_alignment = X means:
201 If X=0: Peeling for alignment will not be applied.
202 If X>0: Peel first X iterations.
203 If X=-1: Generate a runtime test to calculate the number of iterations
204 to be peeled, using the dataref recorded in the field
205 unaligned_dr. */
206 int peeling_for_alignment;
207
208 /* The mask used to check the alignment of pointers or arrays. */
209 int ptr_mask;
210
211 /* All data references in the loop. */
212 VEC (data_reference_p, heap) *datarefs;
213
214 /* All data dependences in the loop. */
215 VEC (ddr_p, heap) *ddrs;
216
217 /* Data Dependence Relations defining address ranges that are candidates
218 for a run-time aliasing check. */
219 VEC (ddr_p, heap) *may_alias_ddrs;
220
221 /* Statements in the loop that have data references that are candidates for a
222 runtime (loop versioning) misalignment check. */
223 VEC(gimple,heap) *may_misalign_stmts;
224
225 /* The loop location in the source. */
226 LOC loop_line_number;
227
228 /* All interleaving chains of stores in the loop, represented by the first
229 stmt in the chain. */
230 VEC(gimple, heap) *strided_stores;
231
232 /* All SLP instances in the loop. This is a subset of the set of STRIDED_STORES
233 of the loop. */
234 VEC(slp_instance, heap) *slp_instances;
235
236 /* The unrolling factor needed to SLP the loop. In case of that pure SLP is
237 applied to the loop, i.e., no unrolling is needed, this is 1. */
238 unsigned slp_unrolling_factor;
239 } *loop_vec_info;
240
241 /* Access Functions. */
242 #define LOOP_VINFO_LOOP(L) (L)->loop
243 #define LOOP_VINFO_BBS(L) (L)->bbs
244 #define LOOP_VINFO_NITERS(L) (L)->num_iters
245 /* Since LOOP_VINFO_NITERS can change after prologue peeling
246 retain total unchanged scalar loop iterations for cost model. */
247 #define LOOP_VINFO_NITERS_UNCHANGED(L) (L)->num_iters_unchanged
248 #define LOOP_VINFO_COST_MODEL_MIN_ITERS(L) (L)->min_profitable_iters
249 #define LOOP_VINFO_VECTORIZABLE_P(L) (L)->vectorizable
250 #define LOOP_VINFO_VECT_FACTOR(L) (L)->vectorization_factor
251 #define LOOP_VINFO_PTR_MASK(L) (L)->ptr_mask
252 #define LOOP_VINFO_DATAREFS(L) (L)->datarefs
253 #define LOOP_VINFO_DDRS(L) (L)->ddrs
254 #define LOOP_VINFO_INT_NITERS(L) (TREE_INT_CST_LOW ((L)->num_iters))
255 #define LOOP_PEELING_FOR_ALIGNMENT(L) (L)->peeling_for_alignment
256 #define LOOP_VINFO_UNALIGNED_DR(L) (L)->unaligned_dr
257 #define LOOP_VINFO_MAY_MISALIGN_STMTS(L) (L)->may_misalign_stmts
258 #define LOOP_VINFO_LOC(L) (L)->loop_line_number
259 #define LOOP_VINFO_MAY_ALIAS_DDRS(L) (L)->may_alias_ddrs
260 #define LOOP_VINFO_STRIDED_STORES(L) (L)->strided_stores
261 #define LOOP_VINFO_SLP_INSTANCES(L) (L)->slp_instances
262 #define LOOP_VINFO_SLP_UNROLLING_FACTOR(L) (L)->slp_unrolling_factor
263
264 #define NITERS_KNOWN_P(n) \
265 (host_integerp ((n),0) \
266 && TREE_INT_CST_LOW ((n)) > 0)
267
268 #define LOOP_VINFO_NITERS_KNOWN_P(L) \
269 NITERS_KNOWN_P((L)->num_iters)
270
271 static inline loop_vec_info
272 loop_vec_info_for_loop (struct loop *loop)
273 {
274 return (loop_vec_info) loop->aux;
275 }
276
277 static inline bool
278 nested_in_vect_loop_p (struct loop *loop, gimple stmt)
279 {
280 return (loop->inner
281 && (loop->inner == (gimple_bb (stmt))->loop_father));
282 }
283
284 /*-----------------------------------------------------------------*/
285 /* Info on vectorized defs. */
286 /*-----------------------------------------------------------------*/
287 enum stmt_vec_info_type {
288 undef_vec_info_type = 0,
289 load_vec_info_type,
290 store_vec_info_type,
291 op_vec_info_type,
292 call_vec_info_type,
293 assignment_vec_info_type,
294 condition_vec_info_type,
295 reduc_vec_info_type,
296 induc_vec_info_type,
297 type_promotion_vec_info_type,
298 type_demotion_vec_info_type,
299 type_conversion_vec_info_type,
300 loop_exit_ctrl_vec_info_type
301 };
302
303 /* Indicates whether/how a variable is used in the scope of loop/basic
304 block. */
305 enum vect_relevant {
306 vect_unused_in_scope = 0,
307 vect_used_in_outer_by_reduction,
308 vect_used_in_outer,
309
310 /* defs that feed computations that end up (only) in a reduction. These
311 defs may be used by non-reduction stmts, but eventually, any
312 computations/values that are affected by these defs are used to compute
313 a reduction (i.e. don't get stored to memory, for example). We use this
314 to identify computations that we can change the order in which they are
315 computed. */
316 vect_used_by_reduction,
317
318 vect_used_in_scope
319 };
320
321 /* The type of vectorization that can be applied to the stmt: regular loop-based
322 vectorization; pure SLP - the stmt is a part of SLP instances and does not
323 have uses outside SLP instances; or hybrid SLP and loop-based - the stmt is
324 a part of SLP instance and also must be loop-based vectorized, since it has
325 uses outside SLP sequences.
326
327 In the loop context the meanings of pure and hybrid SLP are slightly
328 different. By saying that pure SLP is applied to the loop, we mean that we
329 exploit only intra-iteration parallelism in the loop; i.e., the loop can be
330 vectorized without doing any conceptual unrolling, cause we don't pack
331 together stmts from different iterations, only within a single iteration.
332 Loop hybrid SLP means that we exploit both intra-iteration and
333 inter-iteration parallelism (e.g., number of elements in the vector is 4
334 and the slp-group-size is 2, in which case we don't have enough parallelism
335 within an iteration, so we obtain the rest of the parallelism from subsequent
336 iterations by unrolling the loop by 2). */
337 enum slp_vect_type {
338 loop_vect = 0,
339 pure_slp,
340 hybrid
341 };
342
343
344 typedef struct data_reference *dr_p;
345 DEF_VEC_P(dr_p);
346 DEF_VEC_ALLOC_P(dr_p,heap);
347
348 typedef struct _stmt_vec_info {
349
350 enum stmt_vec_info_type type;
351
352 /* The stmt to which this info struct refers to. */
353 gimple stmt;
354
355 /* The loop_vec_info with respect to which STMT is vectorized. */
356 loop_vec_info loop_vinfo;
357
358 /* Not all stmts in the loop need to be vectorized. e.g, the increment
359 of the loop induction variable and computation of array indexes. relevant
360 indicates whether the stmt needs to be vectorized. */
361 enum vect_relevant relevant;
362
363 /* Indicates whether this stmts is part of a computation whose result is
364 used outside the loop. */
365 bool live;
366
367 /* The vector type to be used. */
368 tree vectype;
369
370 /* The vectorized version of the stmt. */
371 gimple vectorized_stmt;
372
373
374 /** The following is relevant only for stmts that contain a non-scalar
375 data-ref (array/pointer/struct access). A GIMPLE stmt is expected to have
376 at most one such data-ref. **/
377
378 /* Information about the data-ref (access function, etc),
379 relative to the inner-most containing loop. */
380 struct data_reference *data_ref_info;
381
382 /* Information about the data-ref relative to this loop
383 nest (the loop that is being considered for vectorization). */
384 tree dr_base_address;
385 tree dr_init;
386 tree dr_offset;
387 tree dr_step;
388 tree dr_aligned_to;
389
390 /* Stmt is part of some pattern (computation idiom) */
391 bool in_pattern_p;
392
393 /* Used for various bookkeeping purposes, generally holding a pointer to
394 some other stmt S that is in some way "related" to this stmt.
395 Current use of this field is:
396 If this stmt is part of a pattern (i.e. the field 'in_pattern_p' is
397 true): S is the "pattern stmt" that represents (and replaces) the
398 sequence of stmts that constitutes the pattern. Similarly, the
399 related_stmt of the "pattern stmt" points back to this stmt (which is
400 the last stmt in the original sequence of stmts that constitutes the
401 pattern). */
402 gimple related_stmt;
403
404 /* List of datarefs that are known to have the same alignment as the dataref
405 of this stmt. */
406 VEC(dr_p,heap) *same_align_refs;
407
408 /* Classify the def of this stmt. */
409 enum vect_def_type def_type;
410
411 /* Interleaving info. */
412 /* First data-ref in the interleaving group. */
413 gimple first_dr;
414 /* Pointer to the next data-ref in the group. */
415 gimple next_dr;
416 /* The size of the interleaving group. */
417 unsigned int size;
418 /* For stores, number of stores from this group seen. We vectorize the last
419 one. */
420 unsigned int store_count;
421 /* For loads only, the gap from the previous load. For consecutive loads, GAP
422 is 1. */
423 unsigned int gap;
424 /* In case that two or more stmts share data-ref, this is the pointer to the
425 previously detected stmt with the same dr. */
426 gimple same_dr_stmt;
427 /* For loads only, if there is a store with the same location, this field is
428 TRUE. */
429 bool read_write_dep;
430
431 /* Vectorization costs associated with statement. */
432 struct
433 {
434 int outside_of_loop; /* Statements generated outside loop. */
435 int inside_of_loop; /* Statements generated inside loop. */
436 } cost;
437
438 /* Whether the stmt is SLPed, loop-based vectorized, or both. */
439 enum slp_vect_type slp_type;
440 } *stmt_vec_info;
441
442 /* Access Functions. */
443 #define STMT_VINFO_TYPE(S) (S)->type
444 #define STMT_VINFO_STMT(S) (S)->stmt
445 #define STMT_VINFO_LOOP_VINFO(S) (S)->loop_vinfo
446 #define STMT_VINFO_RELEVANT(S) (S)->relevant
447 #define STMT_VINFO_LIVE_P(S) (S)->live
448 #define STMT_VINFO_VECTYPE(S) (S)->vectype
449 #define STMT_VINFO_VEC_STMT(S) (S)->vectorized_stmt
450 #define STMT_VINFO_DATA_REF(S) (S)->data_ref_info
451
452 #define STMT_VINFO_DR_BASE_ADDRESS(S) (S)->dr_base_address
453 #define STMT_VINFO_DR_INIT(S) (S)->dr_init
454 #define STMT_VINFO_DR_OFFSET(S) (S)->dr_offset
455 #define STMT_VINFO_DR_STEP(S) (S)->dr_step
456 #define STMT_VINFO_DR_ALIGNED_TO(S) (S)->dr_aligned_to
457
458 #define STMT_VINFO_IN_PATTERN_P(S) (S)->in_pattern_p
459 #define STMT_VINFO_RELATED_STMT(S) (S)->related_stmt
460 #define STMT_VINFO_SAME_ALIGN_REFS(S) (S)->same_align_refs
461 #define STMT_VINFO_DEF_TYPE(S) (S)->def_type
462 #define STMT_VINFO_DR_GROUP_FIRST_DR(S) (S)->first_dr
463 #define STMT_VINFO_DR_GROUP_NEXT_DR(S) (S)->next_dr
464 #define STMT_VINFO_DR_GROUP_SIZE(S) (S)->size
465 #define STMT_VINFO_DR_GROUP_STORE_COUNT(S) (S)->store_count
466 #define STMT_VINFO_DR_GROUP_GAP(S) (S)->gap
467 #define STMT_VINFO_DR_GROUP_SAME_DR_STMT(S)(S)->same_dr_stmt
468 #define STMT_VINFO_DR_GROUP_READ_WRITE_DEPENDENCE(S) (S)->read_write_dep
469 #define STMT_VINFO_STRIDED_ACCESS(S) ((S)->first_dr != NULL)
470
471 #define DR_GROUP_FIRST_DR(S) (S)->first_dr
472 #define DR_GROUP_NEXT_DR(S) (S)->next_dr
473 #define DR_GROUP_SIZE(S) (S)->size
474 #define DR_GROUP_STORE_COUNT(S) (S)->store_count
475 #define DR_GROUP_GAP(S) (S)->gap
476 #define DR_GROUP_SAME_DR_STMT(S) (S)->same_dr_stmt
477 #define DR_GROUP_READ_WRITE_DEPENDENCE(S) (S)->read_write_dep
478
479 #define STMT_VINFO_RELEVANT_P(S) ((S)->relevant != vect_unused_in_scope)
480 #define STMT_VINFO_OUTSIDE_OF_LOOP_COST(S) (S)->cost.outside_of_loop
481 #define STMT_VINFO_INSIDE_OF_LOOP_COST(S) (S)->cost.inside_of_loop
482
483 #define HYBRID_SLP_STMT(S) ((S)->slp_type == hybrid)
484 #define PURE_SLP_STMT(S) ((S)->slp_type == pure_slp)
485 #define STMT_SLP_TYPE(S) (S)->slp_type
486
487 /* These are some defines for the initial implementation of the vectorizer's
488 cost model. These will later be target specific hooks. */
489
490 /* Cost of conditional taken branch. */
491 #ifndef TARG_COND_TAKEN_BRANCH_COST
492 #define TARG_COND_TAKEN_BRANCH_COST 3
493 #endif
494
495 /* Cost of conditional not taken branch. */
496 #ifndef TARG_COND_NOT_TAKEN_BRANCH_COST
497 #define TARG_COND_NOT_TAKEN_BRANCH_COST 1
498 #endif
499
500 /* Cost of any scalar operation, excluding load and store. */
501 #ifndef TARG_SCALAR_STMT_COST
502 #define TARG_SCALAR_STMT_COST 1
503 #endif
504
505 /* Cost of scalar load. */
506 #ifndef TARG_SCALAR_LOAD_COST
507 #define TARG_SCALAR_LOAD_COST 1
508 #endif
509
510 /* Cost of scalar store. */
511 #ifndef TARG_SCALAR_STORE_COST
512 #define TARG_SCALAR_STORE_COST 1
513 #endif
514
515 /* Cost of any vector operation, excluding load, store or vector to scalar
516 operation. */
517 #ifndef TARG_VEC_STMT_COST
518 #define TARG_VEC_STMT_COST 1
519 #endif
520
521 /* Cost of vector to scalar operation. */
522 #ifndef TARG_VEC_TO_SCALAR_COST
523 #define TARG_VEC_TO_SCALAR_COST 1
524 #endif
525
526 /* Cost of scalar to vector operation. */
527 #ifndef TARG_SCALAR_TO_VEC_COST
528 #define TARG_SCALAR_TO_VEC_COST 1
529 #endif
530
531 /* Cost of aligned vector load. */
532 #ifndef TARG_VEC_LOAD_COST
533 #define TARG_VEC_LOAD_COST 1
534 #endif
535
536 /* Cost of misaligned vector load. */
537 #ifndef TARG_VEC_UNALIGNED_LOAD_COST
538 #define TARG_VEC_UNALIGNED_LOAD_COST 2
539 #endif
540
541 /* Cost of vector store. */
542 #ifndef TARG_VEC_STORE_COST
543 #define TARG_VEC_STORE_COST 1
544 #endif
545
546 /* Cost of vector permutation. */
547 #ifndef TARG_VEC_PERMUTE_COST
548 #define TARG_VEC_PERMUTE_COST 1
549 #endif
550
551 /* The maximum number of intermediate steps required in multi-step type
552 conversion. */
553 #define MAX_INTERM_CVT_STEPS 3
554
555 /* Avoid GTY(()) on stmt_vec_info. */
556 typedef void *vec_void_p;
557 DEF_VEC_P (vec_void_p);
558 DEF_VEC_ALLOC_P (vec_void_p, heap);
559
560 extern VEC(vec_void_p,heap) *stmt_vec_info_vec;
561
562 void init_stmt_vec_info_vec (void);
563 void free_stmt_vec_info_vec (void);
564
565 static inline stmt_vec_info
566 vinfo_for_stmt (gimple stmt)
567 {
568 unsigned int uid = gimple_uid (stmt);
569 if (uid == 0)
570 return NULL;
571
572 gcc_assert (uid <= VEC_length (vec_void_p, stmt_vec_info_vec));
573 return (stmt_vec_info) VEC_index (vec_void_p, stmt_vec_info_vec, uid - 1);
574 }
575
576 static inline void
577 set_vinfo_for_stmt (gimple stmt, stmt_vec_info info)
578 {
579 unsigned int uid = gimple_uid (stmt);
580 if (uid == 0)
581 {
582 gcc_assert (info);
583 uid = VEC_length (vec_void_p, stmt_vec_info_vec) + 1;
584 gimple_set_uid (stmt, uid);
585 VEC_safe_push (vec_void_p, heap, stmt_vec_info_vec, (vec_void_p) info);
586 }
587 else
588 VEC_replace (vec_void_p, stmt_vec_info_vec, uid - 1, (vec_void_p) info);
589 }
590
591 static inline gimple
592 get_earlier_stmt (gimple stmt1, gimple stmt2)
593 {
594 unsigned int uid1, uid2;
595
596 if (stmt1 == NULL)
597 return stmt2;
598
599 if (stmt2 == NULL)
600 return stmt1;
601
602 uid1 = gimple_uid (stmt1);
603 uid2 = gimple_uid (stmt2);
604
605 if (uid1 == 0 || uid2 == 0)
606 return NULL;
607
608 gcc_assert (uid1 <= VEC_length (vec_void_p, stmt_vec_info_vec));
609 gcc_assert (uid2 <= VEC_length (vec_void_p, stmt_vec_info_vec));
610
611 if (uid1 < uid2)
612 return stmt1;
613 else
614 return stmt2;
615 }
616
617 static inline bool
618 is_pattern_stmt_p (stmt_vec_info stmt_info)
619 {
620 gimple related_stmt;
621 stmt_vec_info related_stmt_info;
622
623 related_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
624 if (related_stmt
625 && (related_stmt_info = vinfo_for_stmt (related_stmt))
626 && STMT_VINFO_IN_PATTERN_P (related_stmt_info))
627 return true;
628
629 return false;
630 }
631
632 static inline bool
633 is_loop_header_bb_p (basic_block bb)
634 {
635 if (bb == (bb->loop_father)->header)
636 return true;
637 gcc_assert (EDGE_COUNT (bb->preds) == 1);
638 return false;
639 }
640
641 static inline void
642 stmt_vinfo_set_inside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
643 int cost)
644 {
645 if (slp_node)
646 SLP_TREE_INSIDE_OF_LOOP_COST (slp_node) = cost;
647 else
648 STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) = cost;
649 }
650
651 static inline void
652 stmt_vinfo_set_outside_of_loop_cost (stmt_vec_info stmt_info, slp_tree slp_node,
653 int cost)
654 {
655 if (slp_node)
656 SLP_TREE_OUTSIDE_OF_LOOP_COST (slp_node) = cost;
657 else
658 STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = cost;
659 }
660
661 static inline int
662 vect_pow2 (int x)
663 {
664 int i, res = 1;
665
666 for (i = 0; i < x; i++)
667 res *= 2;
668
669 return res;
670 }
671
672 /*-----------------------------------------------------------------*/
673 /* Info on data references alignment. */
674 /*-----------------------------------------------------------------*/
675
676 /* Reflects actual alignment of first access in the vectorized loop,
677 taking into account peeling/versioning if applied. */
678 #define DR_MISALIGNMENT(DR) ((int) (size_t) (DR)->aux)
679 #define SET_DR_MISALIGNMENT(DR, VAL) ((DR)->aux = (void *) (size_t) (VAL))
680
681 static inline bool
682 aligned_access_p (struct data_reference *data_ref_info)
683 {
684 return (DR_MISALIGNMENT (data_ref_info) == 0);
685 }
686
687 static inline bool
688 known_alignment_for_access_p (struct data_reference *data_ref_info)
689 {
690 return (DR_MISALIGNMENT (data_ref_info) != -1);
691 }
692
693 /* vect_dump will be set to stderr or dump_file if exist. */
694 extern FILE *vect_dump;
695 extern LOC vect_loop_location;
696
697 /* Bitmap of virtual variables to be renamed. */
698 extern bitmap vect_memsyms_to_rename;
699
700 /*-----------------------------------------------------------------*/
701 /* Function prototypes. */
702 /*-----------------------------------------------------------------*/
703
704 /* Simple loop peeling and versioning utilities for vectorizer's purposes -
705 in tree-vect-loop-manip.c. */
706 extern void slpeel_make_loop_iterate_ntimes (struct loop *, tree);
707 extern bool slpeel_can_duplicate_loop_p (const struct loop *, const_edge);
708 extern void vect_loop_versioning (loop_vec_info, bool, tree *, gimple_seq *);
709 extern void vect_do_peeling_for_loop_bound (loop_vec_info, tree *,
710 tree, gimple_seq);
711 extern void vect_do_peeling_for_alignment (loop_vec_info);
712 extern LOC find_loop_location (struct loop *);
713 extern bool vect_can_advance_ivs_p (loop_vec_info);
714
715 /* In tree-vect-stmts.c. */
716 extern tree get_vectype_for_scalar_type (tree);
717 extern bool vect_is_simple_use (tree, loop_vec_info, gimple *, tree *,
718 enum vect_def_type *);
719 extern bool supportable_widening_operation (enum tree_code, gimple, tree,
720 tree *, tree *, enum tree_code *,
721 enum tree_code *, int *,
722 VEC (tree, heap) **);
723 extern bool supportable_narrowing_operation (enum tree_code, const_gimple,
724 tree, enum tree_code *, int *,
725 VEC (tree, heap) **);
726 extern stmt_vec_info new_stmt_vec_info (gimple stmt, loop_vec_info);
727 extern void free_stmt_vec_info (gimple stmt);
728 extern tree vectorizable_function (gimple, tree, tree);
729 extern void vect_model_simple_cost (stmt_vec_info, int, enum vect_def_type *,
730 slp_tree);
731 extern void vect_model_store_cost (stmt_vec_info, int, enum vect_def_type,
732 slp_tree);
733 extern void vect_model_load_cost (stmt_vec_info, int, slp_tree);
734 extern void vect_finish_stmt_generation (gimple, gimple,
735 gimple_stmt_iterator *);
736 extern bool vect_mark_stmts_to_be_vectorized (loop_vec_info);
737 extern int cost_for_stmt (gimple);
738 extern tree vect_get_vec_def_for_operand (tree, gimple, tree *);
739 extern tree vect_init_vector (gimple, tree, tree,
740 gimple_stmt_iterator *);
741 extern tree vect_get_vec_def_for_stmt_copy (enum vect_def_type, tree);
742 extern bool vect_transform_stmt (gimple, gimple_stmt_iterator *,
743 bool *, slp_tree, slp_instance);
744 extern void vect_remove_stores (gimple);
745 extern bool vect_analyze_stmt (gimple, bool *);
746
747 /* In tree-vect-data-refs.c. */
748 extern bool vect_can_force_dr_alignment_p (const_tree, unsigned int);
749 extern enum dr_alignment_support vect_supportable_dr_alignment
750 (struct data_reference *);
751 extern tree vect_get_smallest_scalar_type (gimple, HOST_WIDE_INT *,
752 HOST_WIDE_INT *);
753 extern bool vect_analyze_data_ref_dependences (loop_vec_info);
754 extern bool vect_enhance_data_refs_alignment (loop_vec_info);
755 extern bool vect_analyze_data_refs_alignment (loop_vec_info);
756 extern bool vect_analyze_data_ref_accesses (loop_vec_info);
757 extern bool vect_prune_runtime_alias_test_list (loop_vec_info);
758 extern bool vect_analyze_data_refs (loop_vec_info);
759 extern tree vect_create_data_ref_ptr (gimple, struct loop *, tree, tree *,
760 gimple *, bool, bool *);
761 extern tree bump_vector_ptr (tree, gimple, gimple_stmt_iterator *, gimple, tree);
762 extern tree vect_create_destination_var (tree, tree);
763 extern bool vect_strided_store_supported (tree);
764 extern bool vect_strided_load_supported (tree);
765 extern bool vect_permute_store_chain (VEC(tree,heap) *,unsigned int, gimple,
766 gimple_stmt_iterator *, VEC(tree,heap) **);
767 extern tree vect_setup_realignment (gimple, gimple_stmt_iterator *, tree *,
768 enum dr_alignment_support, tree,
769 struct loop **);
770 extern bool vect_permute_load_chain (VEC(tree,heap) *,unsigned int, gimple,
771 gimple_stmt_iterator *, VEC(tree,heap) **);
772 extern bool vect_transform_strided_load (gimple, VEC(tree,heap) *, int,
773 gimple_stmt_iterator *);
774 extern int vect_get_place_in_interleaving_chain (gimple, gimple);
775 extern tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
776 extern tree vect_create_addr_base_for_vector_ref (gimple, gimple_seq *,
777 tree, struct loop *);
778
779 /* In tree-vect-loop.c. */
780 /* FORNOW: Used in tree-parloops.c. */
781 extern void destroy_loop_vec_info (loop_vec_info, bool);
782 extern gimple vect_is_simple_reduction (loop_vec_info, gimple);
783 /* Drive for loop analysis stage. */
784 extern loop_vec_info vect_analyze_loop (struct loop *);
785 /* Drive for loop transformation stage. */
786 extern void vect_transform_loop (loop_vec_info);
787 extern loop_vec_info vect_analyze_loop_form (struct loop *);
788 extern bool vectorizable_live_operation (gimple, gimple_stmt_iterator *,
789 gimple *);
790 extern bool vectorizable_reduction (gimple, gimple_stmt_iterator *, gimple *);
791 extern bool vectorizable_induction (gimple, gimple_stmt_iterator *, gimple *);
792 extern int vect_estimate_min_profitable_iters (loop_vec_info);
793 extern tree get_initial_def_for_reduction (gimple, tree, tree *);
794 extern int vect_min_worthwhile_factor (enum tree_code);
795
796
797 /* In tree-vect-slp.c. */
798 extern void vect_free_slp_instance (slp_instance);
799 extern bool vect_transform_slp_perm_load (gimple, VEC (tree, heap) *,
800 gimple_stmt_iterator *, int,
801 slp_instance, bool);
802 extern bool vect_schedule_slp (loop_vec_info);
803 extern void vect_update_slp_costs_according_to_vf (loop_vec_info);
804 extern bool vect_analyze_slp (loop_vec_info);
805 extern void vect_make_slp_decision (loop_vec_info);
806 extern void vect_detect_hybrid_slp (loop_vec_info);
807 extern void vect_get_slp_defs (slp_tree, VEC (tree,heap) **,
808 VEC (tree,heap) **);
809
810 /* In tree-vect-patterns.c. */
811 /* Pattern recognition functions.
812 Additional pattern recognition functions can (and will) be added
813 in the future. */
814 typedef gimple (* vect_recog_func_ptr) (gimple, tree *, tree *);
815 #define NUM_PATTERNS 4
816 void vect_pattern_recog (loop_vec_info);
817
818 /* Vectorization debug information - in tree-vectorizer.c. */
819 extern bool vect_print_dump_info (enum verbosity_levels);
820 extern void vect_set_verbosity_level (const char *);
821
822 #endif /* GCC_TREE_VECTORIZER_H */