Merge from mainline (160224:163495).
[official-gcc/graphite-test-results.git] / gcc / tree-sra.c
blob83d2d277c5be8f2cab113e83e8a1bd2b31aac652
1 /* Scalar Replacement of Aggregates (SRA) converts some structure
2 references into scalar references, exposing them to the scalar
3 optimizers.
4 Copyright (C) 2008, 2009, 2010 Free Software Foundation, Inc.
5 Contributed by Martin Jambor <mjambor@suse.cz>
7 This file is part of GCC.
9 GCC is free software; you can redistribute it and/or modify it under
10 the terms of the GNU General Public License as published by the Free
11 Software Foundation; either version 3, or (at your option) any later
12 version.
14 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
15 WARRANTY; without even the implied warranty of MERCHANTABILITY or
16 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
17 for more details.
19 You should have received a copy of the GNU General Public License
20 along with GCC; see the file COPYING3. If not see
21 <http://www.gnu.org/licenses/>. */
23 /* This file implements Scalar Reduction of Aggregates (SRA). SRA is run
24 twice, once in the early stages of compilation (early SRA) and once in the
25 late stages (late SRA). The aim of both is to turn references to scalar
26 parts of aggregates into uses of independent scalar variables.
28 The two passes are nearly identical, the only difference is that early SRA
29 does not scalarize unions which are used as the result in a GIMPLE_RETURN
30 statement because together with inlining this can lead to weird type
31 conversions.
33 Both passes operate in four stages:
35 1. The declarations that have properties which make them candidates for
36 scalarization are identified in function find_var_candidates(). The
37 candidates are stored in candidate_bitmap.
39 2. The function body is scanned. In the process, declarations which are
40 used in a manner that prevent their scalarization are removed from the
41 candidate bitmap. More importantly, for every access into an aggregate,
42 an access structure (struct access) is created by create_access() and
43 stored in a vector associated with the aggregate. Among other
44 information, the aggregate declaration, the offset and size of the access
45 and its type are stored in the structure.
47 On a related note, assign_link structures are created for every assign
48 statement between candidate aggregates and attached to the related
49 accesses.
51 3. The vectors of accesses are analyzed. They are first sorted according to
52 their offset and size and then scanned for partially overlapping accesses
53 (i.e. those which overlap but one is not entirely within another). Such
54 an access disqualifies the whole aggregate from being scalarized.
56 If there is no such inhibiting overlap, a representative access structure
57 is chosen for every unique combination of offset and size. Afterwards,
58 the pass builds a set of trees from these structures, in which children
59 of an access are within their parent (in terms of offset and size).
61 Then accesses are propagated whenever possible (i.e. in cases when it
62 does not create a partially overlapping access) across assign_links from
63 the right hand side to the left hand side.
65 Then the set of trees for each declaration is traversed again and those
66 accesses which should be replaced by a scalar are identified.
68 4. The function is traversed again, and for every reference into an
69 aggregate that has some component which is about to be scalarized,
70 statements are amended and new statements are created as necessary.
71 Finally, if a parameter got scalarized, the scalar replacements are
72 initialized with values from respective parameter aggregates. */
74 #include "config.h"
75 #include "system.h"
76 #include "coretypes.h"
77 #include "alloc-pool.h"
78 #include "tm.h"
79 #include "tree.h"
80 #include "gimple.h"
81 #include "cgraph.h"
82 #include "tree-flow.h"
83 #include "ipa-prop.h"
84 #include "tree-pretty-print.h"
85 #include "statistics.h"
86 #include "tree-dump.h"
87 #include "timevar.h"
88 #include "params.h"
89 #include "target.h"
90 #include "flags.h"
91 #include "dbgcnt.h"
92 #include "tree-inline.h"
93 #include "gimple-pretty-print.h"
95 /* Enumeration of all aggregate reductions we can do. */
96 enum sra_mode { SRA_MODE_EARLY_IPA, /* early call regularization */
97 SRA_MODE_EARLY_INTRA, /* early intraprocedural SRA */
98 SRA_MODE_INTRA }; /* late intraprocedural SRA */
100 /* Global variable describing which aggregate reduction we are performing at
101 the moment. */
102 static enum sra_mode sra_mode;
104 struct assign_link;
106 /* ACCESS represents each access to an aggregate variable (as a whole or a
107 part). It can also represent a group of accesses that refer to exactly the
108 same fragment of an aggregate (i.e. those that have exactly the same offset
109 and size). Such representatives for a single aggregate, once determined,
110 are linked in a linked list and have the group fields set.
112 Moreover, when doing intraprocedural SRA, a tree is built from those
113 representatives (by the means of first_child and next_sibling pointers), in
114 which all items in a subtree are "within" the root, i.e. their offset is
115 greater or equal to offset of the root and offset+size is smaller or equal
116 to offset+size of the root. Children of an access are sorted by offset.
118 Note that accesses to parts of vector and complex number types always
119 represented by an access to the whole complex number or a vector. It is a
120 duty of the modifying functions to replace them appropriately. */
122 struct access
124 /* Values returned by `get_ref_base_and_extent' for each component reference
125 If EXPR isn't a component reference just set `BASE = EXPR', `OFFSET = 0',
126 `SIZE = TREE_SIZE (TREE_TYPE (expr))'. */
127 HOST_WIDE_INT offset;
128 HOST_WIDE_INT size;
129 tree base;
131 /* Expression. It is context dependent so do not use it to create new
132 expressions to access the original aggregate. See PR 42154 for a
133 testcase. */
134 tree expr;
135 /* Type. */
136 tree type;
138 /* The statement this access belongs to. */
139 gimple stmt;
141 /* Next group representative for this aggregate. */
142 struct access *next_grp;
144 /* Pointer to the group representative. Pointer to itself if the struct is
145 the representative. */
146 struct access *group_representative;
148 /* If this access has any children (in terms of the definition above), this
149 points to the first one. */
150 struct access *first_child;
152 /* In intraprocedural SRA, pointer to the next sibling in the access tree as
153 described above. In IPA-SRA this is a pointer to the next access
154 belonging to the same group (having the same representative). */
155 struct access *next_sibling;
157 /* Pointers to the first and last element in the linked list of assign
158 links. */
159 struct assign_link *first_link, *last_link;
161 /* Pointer to the next access in the work queue. */
162 struct access *next_queued;
164 /* Replacement variable for this access "region." Never to be accessed
165 directly, always only by the means of get_access_replacement() and only
166 when grp_to_be_replaced flag is set. */
167 tree replacement_decl;
169 /* Is this particular access write access? */
170 unsigned write : 1;
172 /* Is this access an artificial one created to scalarize some record
173 entirely? */
174 unsigned total_scalarization : 1;
176 /* Is this access currently in the work queue? */
177 unsigned grp_queued : 1;
179 /* Does this group contain a write access? This flag is propagated down the
180 access tree. */
181 unsigned grp_write : 1;
183 /* Does this group contain a read access? This flag is propagated down the
184 access tree. */
185 unsigned grp_read : 1;
187 /* Does this group contain a read access that comes from an assignment
188 statement? This flag is propagated down the access tree. */
189 unsigned grp_assignment_read : 1;
191 /* Other passes of the analysis use this bit to make function
192 analyze_access_subtree create scalar replacements for this group if
193 possible. */
194 unsigned grp_hint : 1;
196 /* Is the subtree rooted in this access fully covered by scalar
197 replacements? */
198 unsigned grp_covered : 1;
200 /* If set to true, this access and all below it in an access tree must not be
201 scalarized. */
202 unsigned grp_unscalarizable_region : 1;
204 /* Whether data have been written to parts of the aggregate covered by this
205 access which is not to be scalarized. This flag is propagated up in the
206 access tree. */
207 unsigned grp_unscalarized_data : 1;
209 /* Does this access and/or group contain a write access through a
210 BIT_FIELD_REF? */
211 unsigned grp_partial_lhs : 1;
213 /* Set when a scalar replacement should be created for this variable. We do
214 the decision and creation at different places because create_tmp_var
215 cannot be called from within FOR_EACH_REFERENCED_VAR. */
216 unsigned grp_to_be_replaced : 1;
218 /* Is it possible that the group refers to data which might be (directly or
219 otherwise) modified? */
220 unsigned grp_maybe_modified : 1;
222 /* Set when this is a representative of a pointer to scalar (i.e. by
223 reference) parameter which we consider for turning into a plain scalar
224 (i.e. a by value parameter). */
225 unsigned grp_scalar_ptr : 1;
227 /* Set when we discover that this pointer is not safe to dereference in the
228 caller. */
229 unsigned grp_not_necessarilly_dereferenced : 1;
232 typedef struct access *access_p;
234 DEF_VEC_P (access_p);
235 DEF_VEC_ALLOC_P (access_p, heap);
237 /* Alloc pool for allocating access structures. */
238 static alloc_pool access_pool;
240 /* A structure linking lhs and rhs accesses from an aggregate assignment. They
241 are used to propagate subaccesses from rhs to lhs as long as they don't
242 conflict with what is already there. */
243 struct assign_link
245 struct access *lacc, *racc;
246 struct assign_link *next;
249 /* Alloc pool for allocating assign link structures. */
250 static alloc_pool link_pool;
252 /* Base (tree) -> Vector (VEC(access_p,heap) *) map. */
253 static struct pointer_map_t *base_access_vec;
255 /* Bitmap of candidates. */
256 static bitmap candidate_bitmap;
258 /* Bitmap of candidates which we should try to entirely scalarize away and
259 those which cannot be (because they are and need be used as a whole). */
260 static bitmap should_scalarize_away_bitmap, cannot_scalarize_away_bitmap;
262 /* Obstack for creation of fancy names. */
263 static struct obstack name_obstack;
265 /* Head of a linked list of accesses that need to have its subaccesses
266 propagated to their assignment counterparts. */
267 static struct access *work_queue_head;
269 /* Number of parameters of the analyzed function when doing early ipa SRA. */
270 static int func_param_count;
272 /* scan_function sets the following to true if it encounters a call to
273 __builtin_apply_args. */
274 static bool encountered_apply_args;
276 /* Set by scan_function when it finds a recursive call. */
277 static bool encountered_recursive_call;
279 /* Set by scan_function when it finds a recursive call with less actual
280 arguments than formal parameters.. */
281 static bool encountered_unchangable_recursive_call;
283 /* This is a table in which for each basic block and parameter there is a
284 distance (offset + size) in that parameter which is dereferenced and
285 accessed in that BB. */
286 static HOST_WIDE_INT *bb_dereferences;
287 /* Bitmap of BBs that can cause the function to "stop" progressing by
288 returning, throwing externally, looping infinitely or calling a function
289 which might abort etc.. */
290 static bitmap final_bbs;
292 /* Representative of no accesses at all. */
293 static struct access no_accesses_representant;
295 /* Predicate to test the special value. */
297 static inline bool
298 no_accesses_p (struct access *access)
300 return access == &no_accesses_representant;
303 /* Dump contents of ACCESS to file F in a human friendly way. If GRP is true,
304 representative fields are dumped, otherwise those which only describe the
305 individual access are. */
307 static struct
309 /* Number of processed aggregates is readily available in
310 analyze_all_variable_accesses and so is not stored here. */
312 /* Number of created scalar replacements. */
313 int replacements;
315 /* Number of times sra_modify_expr or sra_modify_assign themselves changed an
316 expression. */
317 int exprs;
319 /* Number of statements created by generate_subtree_copies. */
320 int subtree_copies;
322 /* Number of statements created by load_assign_lhs_subreplacements. */
323 int subreplacements;
325 /* Number of times sra_modify_assign has deleted a statement. */
326 int deleted;
328 /* Number of times sra_modify_assign has to deal with subaccesses of LHS and
329 RHS reparately due to type conversions or nonexistent matching
330 references. */
331 int separate_lhs_rhs_handling;
333 /* Number of parameters that were removed because they were unused. */
334 int deleted_unused_parameters;
336 /* Number of scalars passed as parameters by reference that have been
337 converted to be passed by value. */
338 int scalar_by_ref_to_by_val;
340 /* Number of aggregate parameters that were replaced by one or more of their
341 components. */
342 int aggregate_params_reduced;
344 /* Numbber of components created when splitting aggregate parameters. */
345 int param_reductions_created;
346 } sra_stats;
348 static void
349 dump_access (FILE *f, struct access *access, bool grp)
351 fprintf (f, "access { ");
352 fprintf (f, "base = (%d)'", DECL_UID (access->base));
353 print_generic_expr (f, access->base, 0);
354 fprintf (f, "', offset = " HOST_WIDE_INT_PRINT_DEC, access->offset);
355 fprintf (f, ", size = " HOST_WIDE_INT_PRINT_DEC, access->size);
356 fprintf (f, ", expr = ");
357 print_generic_expr (f, access->expr, 0);
358 fprintf (f, ", type = ");
359 print_generic_expr (f, access->type, 0);
360 if (grp)
361 fprintf (f, ", grp_write = %d, total_scalarization = %d, "
362 "grp_read = %d, grp_hint = %d, grp_assignment_read = %d,"
363 "grp_covered = %d, grp_unscalarizable_region = %d, "
364 "grp_unscalarized_data = %d, grp_partial_lhs = %d, "
365 "grp_to_be_replaced = %d, grp_maybe_modified = %d, "
366 "grp_not_necessarilly_dereferenced = %d\n",
367 access->grp_write, access->total_scalarization,
368 access->grp_read, access->grp_hint, access->grp_assignment_read,
369 access->grp_covered, access->grp_unscalarizable_region,
370 access->grp_unscalarized_data, access->grp_partial_lhs,
371 access->grp_to_be_replaced, access->grp_maybe_modified,
372 access->grp_not_necessarilly_dereferenced);
373 else
374 fprintf (f, ", write = %d, total_scalarization = %d, "
375 "grp_partial_lhs = %d\n",
376 access->write, access->total_scalarization,
377 access->grp_partial_lhs);
380 /* Dump a subtree rooted in ACCESS to file F, indent by LEVEL. */
382 static void
383 dump_access_tree_1 (FILE *f, struct access *access, int level)
387 int i;
389 for (i = 0; i < level; i++)
390 fputs ("* ", dump_file);
392 dump_access (f, access, true);
394 if (access->first_child)
395 dump_access_tree_1 (f, access->first_child, level + 1);
397 access = access->next_sibling;
399 while (access);
402 /* Dump all access trees for a variable, given the pointer to the first root in
403 ACCESS. */
405 static void
406 dump_access_tree (FILE *f, struct access *access)
408 for (; access; access = access->next_grp)
409 dump_access_tree_1 (f, access, 0);
412 /* Return true iff ACC is non-NULL and has subaccesses. */
414 static inline bool
415 access_has_children_p (struct access *acc)
417 return acc && acc->first_child;
420 /* Return a vector of pointers to accesses for the variable given in BASE or
421 NULL if there is none. */
423 static VEC (access_p, heap) *
424 get_base_access_vector (tree base)
426 void **slot;
428 slot = pointer_map_contains (base_access_vec, base);
429 if (!slot)
430 return NULL;
431 else
432 return *(VEC (access_p, heap) **) slot;
435 /* Find an access with required OFFSET and SIZE in a subtree of accesses rooted
436 in ACCESS. Return NULL if it cannot be found. */
438 static struct access *
439 find_access_in_subtree (struct access *access, HOST_WIDE_INT offset,
440 HOST_WIDE_INT size)
442 while (access && (access->offset != offset || access->size != size))
444 struct access *child = access->first_child;
446 while (child && (child->offset + child->size <= offset))
447 child = child->next_sibling;
448 access = child;
451 return access;
454 /* Return the first group representative for DECL or NULL if none exists. */
456 static struct access *
457 get_first_repr_for_decl (tree base)
459 VEC (access_p, heap) *access_vec;
461 access_vec = get_base_access_vector (base);
462 if (!access_vec)
463 return NULL;
465 return VEC_index (access_p, access_vec, 0);
468 /* Find an access representative for the variable BASE and given OFFSET and
469 SIZE. Requires that access trees have already been built. Return NULL if
470 it cannot be found. */
472 static struct access *
473 get_var_base_offset_size_access (tree base, HOST_WIDE_INT offset,
474 HOST_WIDE_INT size)
476 struct access *access;
478 access = get_first_repr_for_decl (base);
479 while (access && (access->offset + access->size <= offset))
480 access = access->next_grp;
481 if (!access)
482 return NULL;
484 return find_access_in_subtree (access, offset, size);
487 /* Add LINK to the linked list of assign links of RACC. */
488 static void
489 add_link_to_rhs (struct access *racc, struct assign_link *link)
491 gcc_assert (link->racc == racc);
493 if (!racc->first_link)
495 gcc_assert (!racc->last_link);
496 racc->first_link = link;
498 else
499 racc->last_link->next = link;
501 racc->last_link = link;
502 link->next = NULL;
505 /* Move all link structures in their linked list in OLD_RACC to the linked list
506 in NEW_RACC. */
507 static void
508 relink_to_new_repr (struct access *new_racc, struct access *old_racc)
510 if (!old_racc->first_link)
512 gcc_assert (!old_racc->last_link);
513 return;
516 if (new_racc->first_link)
518 gcc_assert (!new_racc->last_link->next);
519 gcc_assert (!old_racc->last_link || !old_racc->last_link->next);
521 new_racc->last_link->next = old_racc->first_link;
522 new_racc->last_link = old_racc->last_link;
524 else
526 gcc_assert (!new_racc->last_link);
528 new_racc->first_link = old_racc->first_link;
529 new_racc->last_link = old_racc->last_link;
531 old_racc->first_link = old_racc->last_link = NULL;
534 /* Add ACCESS to the work queue (which is actually a stack). */
536 static void
537 add_access_to_work_queue (struct access *access)
539 if (!access->grp_queued)
541 gcc_assert (!access->next_queued);
542 access->next_queued = work_queue_head;
543 access->grp_queued = 1;
544 work_queue_head = access;
548 /* Pop an access from the work queue, and return it, assuming there is one. */
550 static struct access *
551 pop_access_from_work_queue (void)
553 struct access *access = work_queue_head;
555 work_queue_head = access->next_queued;
556 access->next_queued = NULL;
557 access->grp_queued = 0;
558 return access;
562 /* Allocate necessary structures. */
564 static void
565 sra_initialize (void)
567 candidate_bitmap = BITMAP_ALLOC (NULL);
568 should_scalarize_away_bitmap = BITMAP_ALLOC (NULL);
569 cannot_scalarize_away_bitmap = BITMAP_ALLOC (NULL);
570 gcc_obstack_init (&name_obstack);
571 access_pool = create_alloc_pool ("SRA accesses", sizeof (struct access), 16);
572 link_pool = create_alloc_pool ("SRA links", sizeof (struct assign_link), 16);
573 base_access_vec = pointer_map_create ();
574 memset (&sra_stats, 0, sizeof (sra_stats));
575 encountered_apply_args = false;
576 encountered_recursive_call = false;
577 encountered_unchangable_recursive_call = false;
580 /* Hook fed to pointer_map_traverse, deallocate stored vectors. */
582 static bool
583 delete_base_accesses (const void *key ATTRIBUTE_UNUSED, void **value,
584 void *data ATTRIBUTE_UNUSED)
586 VEC (access_p, heap) *access_vec;
587 access_vec = (VEC (access_p, heap) *) *value;
588 VEC_free (access_p, heap, access_vec);
590 return true;
593 /* Deallocate all general structures. */
595 static void
596 sra_deinitialize (void)
598 BITMAP_FREE (candidate_bitmap);
599 BITMAP_FREE (should_scalarize_away_bitmap);
600 BITMAP_FREE (cannot_scalarize_away_bitmap);
601 free_alloc_pool (access_pool);
602 free_alloc_pool (link_pool);
603 obstack_free (&name_obstack, NULL);
605 pointer_map_traverse (base_access_vec, delete_base_accesses, NULL);
606 pointer_map_destroy (base_access_vec);
609 /* Remove DECL from candidates for SRA and write REASON to the dump file if
610 there is one. */
611 static void
612 disqualify_candidate (tree decl, const char *reason)
614 bitmap_clear_bit (candidate_bitmap, DECL_UID (decl));
616 if (dump_file && (dump_flags & TDF_DETAILS))
618 fprintf (dump_file, "! Disqualifying ");
619 print_generic_expr (dump_file, decl, 0);
620 fprintf (dump_file, " - %s\n", reason);
624 /* Return true iff the type contains a field or an element which does not allow
625 scalarization. */
627 static bool
628 type_internals_preclude_sra_p (tree type)
630 tree fld;
631 tree et;
633 switch (TREE_CODE (type))
635 case RECORD_TYPE:
636 case UNION_TYPE:
637 case QUAL_UNION_TYPE:
638 for (fld = TYPE_FIELDS (type); fld; fld = DECL_CHAIN (fld))
639 if (TREE_CODE (fld) == FIELD_DECL)
641 tree ft = TREE_TYPE (fld);
643 if (TREE_THIS_VOLATILE (fld)
644 || !DECL_FIELD_OFFSET (fld) || !DECL_SIZE (fld)
645 || !host_integerp (DECL_FIELD_OFFSET (fld), 1)
646 || !host_integerp (DECL_SIZE (fld), 1))
647 return true;
649 if (AGGREGATE_TYPE_P (ft)
650 && type_internals_preclude_sra_p (ft))
651 return true;
654 return false;
656 case ARRAY_TYPE:
657 et = TREE_TYPE (type);
659 if (AGGREGATE_TYPE_P (et))
660 return type_internals_preclude_sra_p (et);
661 else
662 return false;
664 default:
665 return false;
669 /* If T is an SSA_NAME, return NULL if it is not a default def or return its
670 base variable if it is. Return T if it is not an SSA_NAME. */
672 static tree
673 get_ssa_base_param (tree t)
675 if (TREE_CODE (t) == SSA_NAME)
677 if (SSA_NAME_IS_DEFAULT_DEF (t))
678 return SSA_NAME_VAR (t);
679 else
680 return NULL_TREE;
682 return t;
685 /* Mark a dereference of BASE of distance DIST in a basic block tht STMT
686 belongs to, unless the BB has already been marked as a potentially
687 final. */
689 static void
690 mark_parm_dereference (tree base, HOST_WIDE_INT dist, gimple stmt)
692 basic_block bb = gimple_bb (stmt);
693 int idx, parm_index = 0;
694 tree parm;
696 if (bitmap_bit_p (final_bbs, bb->index))
697 return;
699 for (parm = DECL_ARGUMENTS (current_function_decl);
700 parm && parm != base;
701 parm = DECL_CHAIN (parm))
702 parm_index++;
704 gcc_assert (parm_index < func_param_count);
706 idx = bb->index * func_param_count + parm_index;
707 if (bb_dereferences[idx] < dist)
708 bb_dereferences[idx] = dist;
711 /* Allocate an access structure for BASE, OFFSET and SIZE, clear it, fill in
712 the three fields. Also add it to the vector of accesses corresponding to
713 the base. Finally, return the new access. */
715 static struct access *
716 create_access_1 (tree base, HOST_WIDE_INT offset, HOST_WIDE_INT size)
718 VEC (access_p, heap) *vec;
719 struct access *access;
720 void **slot;
722 access = (struct access *) pool_alloc (access_pool);
723 memset (access, 0, sizeof (struct access));
724 access->base = base;
725 access->offset = offset;
726 access->size = size;
728 slot = pointer_map_contains (base_access_vec, base);
729 if (slot)
730 vec = (VEC (access_p, heap) *) *slot;
731 else
732 vec = VEC_alloc (access_p, heap, 32);
734 VEC_safe_push (access_p, heap, vec, access);
736 *((struct VEC (access_p,heap) **)
737 pointer_map_insert (base_access_vec, base)) = vec;
739 return access;
742 /* Create and insert access for EXPR. Return created access, or NULL if it is
743 not possible. */
745 static struct access *
746 create_access (tree expr, gimple stmt, bool write)
748 struct access *access;
749 HOST_WIDE_INT offset, size, max_size;
750 tree base = expr;
751 bool ptr, unscalarizable_region = false;
753 base = get_ref_base_and_extent (expr, &offset, &size, &max_size);
755 if (sra_mode == SRA_MODE_EARLY_IPA
756 && TREE_CODE (base) == MEM_REF)
758 base = get_ssa_base_param (TREE_OPERAND (base, 0));
759 if (!base)
760 return NULL;
761 ptr = true;
763 else
764 ptr = false;
766 if (!DECL_P (base) || !bitmap_bit_p (candidate_bitmap, DECL_UID (base)))
767 return NULL;
769 if (sra_mode == SRA_MODE_EARLY_IPA)
771 if (size < 0 || size != max_size)
773 disqualify_candidate (base, "Encountered a variable sized access.");
774 return NULL;
776 if ((offset % BITS_PER_UNIT) != 0 || (size % BITS_PER_UNIT) != 0)
778 disqualify_candidate (base,
779 "Encountered an acces not aligned to a byte.");
780 return NULL;
783 if (ptr)
784 mark_parm_dereference (base, offset + size, stmt);
786 else
788 if (size != max_size)
790 size = max_size;
791 unscalarizable_region = true;
793 if (size < 0)
795 disqualify_candidate (base, "Encountered an unconstrained access.");
796 return NULL;
800 access = create_access_1 (base, offset, size);
801 access->expr = expr;
802 access->type = TREE_TYPE (expr);
803 access->write = write;
804 access->grp_unscalarizable_region = unscalarizable_region;
805 access->stmt = stmt;
807 return access;
811 /* Return true iff TYPE is a RECORD_TYPE with fields that are either of gimple
812 register types or (recursively) records with only these two kinds of fields.
813 It also returns false if any of these records has a zero-size field as its
814 last field or has a bit-field. */
816 static bool
817 type_consists_of_records_p (tree type)
819 tree fld;
820 bool last_fld_has_zero_size = false;
822 if (TREE_CODE (type) != RECORD_TYPE)
823 return false;
825 for (fld = TYPE_FIELDS (type); fld; fld = DECL_CHAIN (fld))
826 if (TREE_CODE (fld) == FIELD_DECL)
828 tree ft = TREE_TYPE (fld);
830 if (DECL_BIT_FIELD (fld))
831 return false;
833 if (!is_gimple_reg_type (ft)
834 && !type_consists_of_records_p (ft))
835 return false;
837 last_fld_has_zero_size = tree_low_cst (DECL_SIZE (fld), 1) == 0;
840 if (last_fld_has_zero_size)
841 return false;
843 return true;
846 /* Create total_scalarization accesses for all scalar type fields in DECL that
847 must be of a RECORD_TYPE conforming to type_consists_of_records_p. BASE
848 must be the top-most VAR_DECL representing the variable, OFFSET must be the
849 offset of DECL within BASE. REF must be the memory reference expression for
850 the given decl. */
852 static void
853 completely_scalarize_record (tree base, tree decl, HOST_WIDE_INT offset,
854 tree ref)
856 tree fld, decl_type = TREE_TYPE (decl);
858 for (fld = TYPE_FIELDS (decl_type); fld; fld = DECL_CHAIN (fld))
859 if (TREE_CODE (fld) == FIELD_DECL)
861 HOST_WIDE_INT pos = offset + int_bit_position (fld);
862 tree ft = TREE_TYPE (fld);
863 tree nref = build3 (COMPONENT_REF, TREE_TYPE (fld), ref, fld,
864 NULL_TREE);
866 if (is_gimple_reg_type (ft))
868 struct access *access;
869 HOST_WIDE_INT size;
871 size = tree_low_cst (DECL_SIZE (fld), 1);
872 access = create_access_1 (base, pos, size);
873 access->expr = nref;
874 access->type = ft;
875 access->total_scalarization = 1;
876 /* Accesses for intraprocedural SRA can have their stmt NULL. */
878 else
879 completely_scalarize_record (base, fld, pos, nref);
884 /* Search the given tree for a declaration by skipping handled components and
885 exclude it from the candidates. */
887 static void
888 disqualify_base_of_expr (tree t, const char *reason)
890 t = get_base_address (t);
891 if (sra_mode == SRA_MODE_EARLY_IPA
892 && TREE_CODE (t) == MEM_REF)
893 t = get_ssa_base_param (TREE_OPERAND (t, 0));
895 if (t && DECL_P (t))
896 disqualify_candidate (t, reason);
899 /* Scan expression EXPR and create access structures for all accesses to
900 candidates for scalarization. Return the created access or NULL if none is
901 created. */
903 static struct access *
904 build_access_from_expr_1 (tree expr, gimple stmt, bool write)
906 struct access *ret = NULL;
907 bool partial_ref;
909 if (TREE_CODE (expr) == BIT_FIELD_REF
910 || TREE_CODE (expr) == IMAGPART_EXPR
911 || TREE_CODE (expr) == REALPART_EXPR)
913 expr = TREE_OPERAND (expr, 0);
914 partial_ref = true;
916 else
917 partial_ref = false;
919 /* We need to dive through V_C_Es in order to get the size of its parameter
920 and not the result type. Ada produces such statements. We are also
921 capable of handling the topmost V_C_E but not any of those buried in other
922 handled components. */
923 if (TREE_CODE (expr) == VIEW_CONVERT_EXPR)
924 expr = TREE_OPERAND (expr, 0);
926 if (contains_view_convert_expr_p (expr))
928 disqualify_base_of_expr (expr, "V_C_E under a different handled "
929 "component.");
930 return NULL;
933 switch (TREE_CODE (expr))
935 case MEM_REF:
936 if (TREE_CODE (TREE_OPERAND (expr, 0)) != ADDR_EXPR
937 && sra_mode != SRA_MODE_EARLY_IPA)
938 return NULL;
939 /* fall through */
940 case VAR_DECL:
941 case PARM_DECL:
942 case RESULT_DECL:
943 case COMPONENT_REF:
944 case ARRAY_REF:
945 case ARRAY_RANGE_REF:
946 ret = create_access (expr, stmt, write);
947 break;
949 default:
950 break;
953 if (write && partial_ref && ret)
954 ret->grp_partial_lhs = 1;
956 return ret;
959 /* Scan expression EXPR and create access structures for all accesses to
960 candidates for scalarization. Return true if any access has been inserted.
961 STMT must be the statement from which the expression is taken, WRITE must be
962 true if the expression is a store and false otherwise. */
964 static bool
965 build_access_from_expr (tree expr, gimple stmt, bool write)
967 struct access *access;
969 access = build_access_from_expr_1 (expr, stmt, write);
970 if (access)
972 /* This means the aggregate is accesses as a whole in a way other than an
973 assign statement and thus cannot be removed even if we had a scalar
974 replacement for everything. */
975 if (cannot_scalarize_away_bitmap)
976 bitmap_set_bit (cannot_scalarize_away_bitmap, DECL_UID (access->base));
977 return true;
979 return false;
982 /* Disqualify LHS and RHS for scalarization if STMT must end its basic block in
983 modes in which it matters, return true iff they have been disqualified. RHS
984 may be NULL, in that case ignore it. If we scalarize an aggregate in
985 intra-SRA we may need to add statements after each statement. This is not
986 possible if a statement unconditionally has to end the basic block. */
987 static bool
988 disqualify_ops_if_throwing_stmt (gimple stmt, tree lhs, tree rhs)
990 if ((sra_mode == SRA_MODE_EARLY_INTRA || sra_mode == SRA_MODE_INTRA)
991 && (stmt_can_throw_internal (stmt) || stmt_ends_bb_p (stmt)))
993 disqualify_base_of_expr (lhs, "LHS of a throwing stmt.");
994 if (rhs)
995 disqualify_base_of_expr (rhs, "RHS of a throwing stmt.");
996 return true;
998 return false;
1001 /* Scan expressions occuring in STMT, create access structures for all accesses
1002 to candidates for scalarization and remove those candidates which occur in
1003 statements or expressions that prevent them from being split apart. Return
1004 true if any access has been inserted. */
1006 static bool
1007 build_accesses_from_assign (gimple stmt)
1009 tree lhs, rhs;
1010 struct access *lacc, *racc;
1012 if (!gimple_assign_single_p (stmt))
1013 return false;
1015 lhs = gimple_assign_lhs (stmt);
1016 rhs = gimple_assign_rhs1 (stmt);
1018 if (disqualify_ops_if_throwing_stmt (stmt, lhs, rhs))
1019 return false;
1021 racc = build_access_from_expr_1 (rhs, stmt, false);
1022 lacc = build_access_from_expr_1 (lhs, stmt, true);
1024 if (racc)
1026 racc->grp_assignment_read = 1;
1027 if (should_scalarize_away_bitmap && !gimple_has_volatile_ops (stmt)
1028 && !is_gimple_reg_type (racc->type))
1029 bitmap_set_bit (should_scalarize_away_bitmap, DECL_UID (racc->base));
1032 if (lacc && racc
1033 && (sra_mode == SRA_MODE_EARLY_INTRA || sra_mode == SRA_MODE_INTRA)
1034 && !lacc->grp_unscalarizable_region
1035 && !racc->grp_unscalarizable_region
1036 && AGGREGATE_TYPE_P (TREE_TYPE (lhs))
1037 /* FIXME: Turn the following line into an assert after PR 40058 is
1038 fixed. */
1039 && lacc->size == racc->size
1040 && useless_type_conversion_p (lacc->type, racc->type))
1042 struct assign_link *link;
1044 link = (struct assign_link *) pool_alloc (link_pool);
1045 memset (link, 0, sizeof (struct assign_link));
1047 link->lacc = lacc;
1048 link->racc = racc;
1050 add_link_to_rhs (racc, link);
1053 return lacc || racc;
1056 /* Callback of walk_stmt_load_store_addr_ops visit_addr used to determine
1057 GIMPLE_ASM operands with memory constrains which cannot be scalarized. */
1059 static bool
1060 asm_visit_addr (gimple stmt ATTRIBUTE_UNUSED, tree op,
1061 void *data ATTRIBUTE_UNUSED)
1063 op = get_base_address (op);
1064 if (op
1065 && DECL_P (op))
1066 disqualify_candidate (op, "Non-scalarizable GIMPLE_ASM operand.");
1068 return false;
1071 /* Return true iff callsite CALL has at least as many actual arguments as there
1072 are formal parameters of the function currently processed by IPA-SRA. */
1074 static inline bool
1075 callsite_has_enough_arguments_p (gimple call)
1077 return gimple_call_num_args (call) >= (unsigned) func_param_count;
1080 /* Scan function and look for interesting expressions and create access
1081 structures for them. Return true iff any access is created. */
1083 static bool
1084 scan_function (void)
1086 basic_block bb;
1087 bool ret = false;
1089 FOR_EACH_BB (bb)
1091 gimple_stmt_iterator gsi;
1092 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
1094 gimple stmt = gsi_stmt (gsi);
1095 tree t;
1096 unsigned i;
1098 if (final_bbs && stmt_can_throw_external (stmt))
1099 bitmap_set_bit (final_bbs, bb->index);
1100 switch (gimple_code (stmt))
1102 case GIMPLE_RETURN:
1103 t = gimple_return_retval (stmt);
1104 if (t != NULL_TREE)
1105 ret |= build_access_from_expr (t, stmt, false);
1106 if (final_bbs)
1107 bitmap_set_bit (final_bbs, bb->index);
1108 break;
1110 case GIMPLE_ASSIGN:
1111 ret |= build_accesses_from_assign (stmt);
1112 break;
1114 case GIMPLE_CALL:
1115 for (i = 0; i < gimple_call_num_args (stmt); i++)
1116 ret |= build_access_from_expr (gimple_call_arg (stmt, i),
1117 stmt, false);
1119 if (sra_mode == SRA_MODE_EARLY_IPA)
1121 tree dest = gimple_call_fndecl (stmt);
1122 int flags = gimple_call_flags (stmt);
1124 if (dest)
1126 if (DECL_BUILT_IN_CLASS (dest) == BUILT_IN_NORMAL
1127 && DECL_FUNCTION_CODE (dest) == BUILT_IN_APPLY_ARGS)
1128 encountered_apply_args = true;
1129 if (cgraph_get_node (dest)
1130 == cgraph_get_node (current_function_decl))
1132 encountered_recursive_call = true;
1133 if (!callsite_has_enough_arguments_p (stmt))
1134 encountered_unchangable_recursive_call = true;
1138 if (final_bbs
1139 && (flags & (ECF_CONST | ECF_PURE)) == 0)
1140 bitmap_set_bit (final_bbs, bb->index);
1143 t = gimple_call_lhs (stmt);
1144 if (t && !disqualify_ops_if_throwing_stmt (stmt, t, NULL))
1145 ret |= build_access_from_expr (t, stmt, true);
1146 break;
1148 case GIMPLE_ASM:
1149 walk_stmt_load_store_addr_ops (stmt, NULL, NULL, NULL,
1150 asm_visit_addr);
1151 if (final_bbs)
1152 bitmap_set_bit (final_bbs, bb->index);
1154 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
1156 t = TREE_VALUE (gimple_asm_input_op (stmt, i));
1157 ret |= build_access_from_expr (t, stmt, false);
1159 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
1161 t = TREE_VALUE (gimple_asm_output_op (stmt, i));
1162 ret |= build_access_from_expr (t, stmt, true);
1164 break;
1166 default:
1167 break;
1172 return ret;
1175 /* Helper of QSORT function. There are pointers to accesses in the array. An
1176 access is considered smaller than another if it has smaller offset or if the
1177 offsets are the same but is size is bigger. */
1179 static int
1180 compare_access_positions (const void *a, const void *b)
1182 const access_p *fp1 = (const access_p *) a;
1183 const access_p *fp2 = (const access_p *) b;
1184 const access_p f1 = *fp1;
1185 const access_p f2 = *fp2;
1187 if (f1->offset != f2->offset)
1188 return f1->offset < f2->offset ? -1 : 1;
1190 if (f1->size == f2->size)
1192 if (f1->type == f2->type)
1193 return 0;
1194 /* Put any non-aggregate type before any aggregate type. */
1195 else if (!is_gimple_reg_type (f1->type)
1196 && is_gimple_reg_type (f2->type))
1197 return 1;
1198 else if (is_gimple_reg_type (f1->type)
1199 && !is_gimple_reg_type (f2->type))
1200 return -1;
1201 /* Put any complex or vector type before any other scalar type. */
1202 else if (TREE_CODE (f1->type) != COMPLEX_TYPE
1203 && TREE_CODE (f1->type) != VECTOR_TYPE
1204 && (TREE_CODE (f2->type) == COMPLEX_TYPE
1205 || TREE_CODE (f2->type) == VECTOR_TYPE))
1206 return 1;
1207 else if ((TREE_CODE (f1->type) == COMPLEX_TYPE
1208 || TREE_CODE (f1->type) == VECTOR_TYPE)
1209 && TREE_CODE (f2->type) != COMPLEX_TYPE
1210 && TREE_CODE (f2->type) != VECTOR_TYPE)
1211 return -1;
1212 /* Put the integral type with the bigger precision first. */
1213 else if (INTEGRAL_TYPE_P (f1->type)
1214 && INTEGRAL_TYPE_P (f2->type))
1215 return TYPE_PRECISION (f2->type) - TYPE_PRECISION (f1->type);
1216 /* Put any integral type with non-full precision last. */
1217 else if (INTEGRAL_TYPE_P (f1->type)
1218 && (TREE_INT_CST_LOW (TYPE_SIZE (f1->type))
1219 != TYPE_PRECISION (f1->type)))
1220 return 1;
1221 else if (INTEGRAL_TYPE_P (f2->type)
1222 && (TREE_INT_CST_LOW (TYPE_SIZE (f2->type))
1223 != TYPE_PRECISION (f2->type)))
1224 return -1;
1225 /* Stabilize the sort. */
1226 return TYPE_UID (f1->type) - TYPE_UID (f2->type);
1229 /* We want the bigger accesses first, thus the opposite operator in the next
1230 line: */
1231 return f1->size > f2->size ? -1 : 1;
1235 /* Append a name of the declaration to the name obstack. A helper function for
1236 make_fancy_name. */
1238 static void
1239 make_fancy_decl_name (tree decl)
1241 char buffer[32];
1243 tree name = DECL_NAME (decl);
1244 if (name)
1245 obstack_grow (&name_obstack, IDENTIFIER_POINTER (name),
1246 IDENTIFIER_LENGTH (name));
1247 else
1249 sprintf (buffer, "D%u", DECL_UID (decl));
1250 obstack_grow (&name_obstack, buffer, strlen (buffer));
1254 /* Helper for make_fancy_name. */
1256 static void
1257 make_fancy_name_1 (tree expr)
1259 char buffer[32];
1260 tree index;
1262 if (DECL_P (expr))
1264 make_fancy_decl_name (expr);
1265 return;
1268 switch (TREE_CODE (expr))
1270 case COMPONENT_REF:
1271 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1272 obstack_1grow (&name_obstack, '$');
1273 make_fancy_decl_name (TREE_OPERAND (expr, 1));
1274 break;
1276 case ARRAY_REF:
1277 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1278 obstack_1grow (&name_obstack, '$');
1279 /* Arrays with only one element may not have a constant as their
1280 index. */
1281 index = TREE_OPERAND (expr, 1);
1282 if (TREE_CODE (index) != INTEGER_CST)
1283 break;
1284 sprintf (buffer, HOST_WIDE_INT_PRINT_DEC, TREE_INT_CST_LOW (index));
1285 obstack_grow (&name_obstack, buffer, strlen (buffer));
1286 break;
1288 case ADDR_EXPR:
1289 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1290 break;
1292 case MEM_REF:
1293 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1294 if (!integer_zerop (TREE_OPERAND (expr, 1)))
1296 obstack_1grow (&name_obstack, '$');
1297 sprintf (buffer, HOST_WIDE_INT_PRINT_DEC,
1298 TREE_INT_CST_LOW (TREE_OPERAND (expr, 1)));
1299 obstack_grow (&name_obstack, buffer, strlen (buffer));
1301 break;
1303 case BIT_FIELD_REF:
1304 case REALPART_EXPR:
1305 case IMAGPART_EXPR:
1306 gcc_unreachable (); /* we treat these as scalars. */
1307 break;
1308 default:
1309 break;
1313 /* Create a human readable name for replacement variable of ACCESS. */
1315 static char *
1316 make_fancy_name (tree expr)
1318 make_fancy_name_1 (expr);
1319 obstack_1grow (&name_obstack, '\0');
1320 return XOBFINISH (&name_obstack, char *);
1323 /* Helper function for build_ref_for_offset.
1325 FIXME: Eventually this should be rewritten to either re-use the
1326 original access expression unshared (which is good for alias
1327 analysis) or to build a MEM_REF expression. */
1329 static bool
1330 build_ref_for_offset_1 (tree *res, tree type, HOST_WIDE_INT offset,
1331 tree exp_type)
1333 while (1)
1335 tree fld;
1336 tree tr_size, index, minidx;
1337 HOST_WIDE_INT el_size;
1339 if (offset == 0 && exp_type
1340 && types_compatible_p (exp_type, type))
1341 return true;
1343 switch (TREE_CODE (type))
1345 case UNION_TYPE:
1346 case QUAL_UNION_TYPE:
1347 case RECORD_TYPE:
1348 for (fld = TYPE_FIELDS (type); fld; fld = DECL_CHAIN (fld))
1350 HOST_WIDE_INT pos, size;
1351 tree expr, *expr_ptr;
1353 if (TREE_CODE (fld) != FIELD_DECL)
1354 continue;
1356 pos = int_bit_position (fld);
1357 gcc_assert (TREE_CODE (type) == RECORD_TYPE || pos == 0);
1358 tr_size = DECL_SIZE (fld);
1359 if (!tr_size || !host_integerp (tr_size, 1))
1360 continue;
1361 size = tree_low_cst (tr_size, 1);
1362 if (size == 0)
1364 if (pos != offset)
1365 continue;
1367 else if (pos > offset || (pos + size) <= offset)
1368 continue;
1370 if (res)
1372 expr = build3 (COMPONENT_REF, TREE_TYPE (fld), *res, fld,
1373 NULL_TREE);
1374 expr_ptr = &expr;
1376 else
1377 expr_ptr = NULL;
1378 if (build_ref_for_offset_1 (expr_ptr, TREE_TYPE (fld),
1379 offset - pos, exp_type))
1381 if (res)
1382 *res = expr;
1383 return true;
1386 return false;
1388 case ARRAY_TYPE:
1389 tr_size = TYPE_SIZE (TREE_TYPE (type));
1390 if (!tr_size || !host_integerp (tr_size, 1))
1391 return false;
1392 el_size = tree_low_cst (tr_size, 1);
1394 minidx = TYPE_MIN_VALUE (TYPE_DOMAIN (type));
1395 if (TREE_CODE (minidx) != INTEGER_CST || el_size == 0)
1396 return false;
1397 if (res)
1399 index = build_int_cst (TYPE_DOMAIN (type), offset / el_size);
1400 if (!integer_zerop (minidx))
1401 index = int_const_binop (PLUS_EXPR, index, minidx, 0);
1402 *res = build4 (ARRAY_REF, TREE_TYPE (type), *res, index,
1403 NULL_TREE, NULL_TREE);
1405 offset = offset % el_size;
1406 type = TREE_TYPE (type);
1407 break;
1409 default:
1410 if (offset != 0)
1411 return false;
1413 if (exp_type)
1414 return false;
1415 else
1416 return true;
1421 /* Construct an expression that would reference a part of aggregate *EXPR of
1422 type TYPE at the given OFFSET of the type EXP_TYPE. If EXPR is NULL, the
1423 function only determines whether it can build such a reference without
1424 actually doing it, otherwise, the tree it points to is unshared first and
1425 then used as a base for furhter sub-references. */
1427 bool
1428 build_ref_for_offset (tree *expr, tree type, HOST_WIDE_INT offset,
1429 tree exp_type, bool allow_ptr)
1431 location_t loc = expr ? EXPR_LOCATION (*expr) : UNKNOWN_LOCATION;
1433 if (expr)
1434 *expr = unshare_expr (*expr);
1436 if (allow_ptr && POINTER_TYPE_P (type))
1438 type = TREE_TYPE (type);
1439 if (expr)
1440 *expr = build_simple_mem_ref_loc (loc, *expr);
1443 return build_ref_for_offset_1 (expr, type, offset, exp_type);
1446 /* Return true iff TYPE is stdarg va_list type. */
1448 static inline bool
1449 is_va_list_type (tree type)
1451 return TYPE_MAIN_VARIANT (type) == TYPE_MAIN_VARIANT (va_list_type_node);
1454 /* The very first phase of intraprocedural SRA. It marks in candidate_bitmap
1455 those with type which is suitable for scalarization. */
1457 static bool
1458 find_var_candidates (void)
1460 tree var, type;
1461 referenced_var_iterator rvi;
1462 bool ret = false;
1464 FOR_EACH_REFERENCED_VAR (var, rvi)
1466 if (TREE_CODE (var) != VAR_DECL && TREE_CODE (var) != PARM_DECL)
1467 continue;
1468 type = TREE_TYPE (var);
1470 if (!AGGREGATE_TYPE_P (type)
1471 || needs_to_live_in_memory (var)
1472 || TREE_THIS_VOLATILE (var)
1473 || !COMPLETE_TYPE_P (type)
1474 || !host_integerp (TYPE_SIZE (type), 1)
1475 || tree_low_cst (TYPE_SIZE (type), 1) == 0
1476 || type_internals_preclude_sra_p (type)
1477 /* Fix for PR 41089. tree-stdarg.c needs to have va_lists intact but
1478 we also want to schedule it rather late. Thus we ignore it in
1479 the early pass. */
1480 || (sra_mode == SRA_MODE_EARLY_INTRA
1481 && is_va_list_type (type)))
1482 continue;
1484 bitmap_set_bit (candidate_bitmap, DECL_UID (var));
1486 if (dump_file && (dump_flags & TDF_DETAILS))
1488 fprintf (dump_file, "Candidate (%d): ", DECL_UID (var));
1489 print_generic_expr (dump_file, var, 0);
1490 fprintf (dump_file, "\n");
1492 ret = true;
1495 return ret;
1498 /* Sort all accesses for the given variable, check for partial overlaps and
1499 return NULL if there are any. If there are none, pick a representative for
1500 each combination of offset and size and create a linked list out of them.
1501 Return the pointer to the first representative and make sure it is the first
1502 one in the vector of accesses. */
1504 static struct access *
1505 sort_and_splice_var_accesses (tree var)
1507 int i, j, access_count;
1508 struct access *res, **prev_acc_ptr = &res;
1509 VEC (access_p, heap) *access_vec;
1510 bool first = true;
1511 HOST_WIDE_INT low = -1, high = 0;
1513 access_vec = get_base_access_vector (var);
1514 if (!access_vec)
1515 return NULL;
1516 access_count = VEC_length (access_p, access_vec);
1518 /* Sort by <OFFSET, SIZE>. */
1519 qsort (VEC_address (access_p, access_vec), access_count, sizeof (access_p),
1520 compare_access_positions);
1522 i = 0;
1523 while (i < access_count)
1525 struct access *access = VEC_index (access_p, access_vec, i);
1526 bool grp_write = access->write;
1527 bool grp_read = !access->write;
1528 bool grp_assignment_read = access->grp_assignment_read;
1529 bool multiple_reads = false;
1530 bool total_scalarization = access->total_scalarization;
1531 bool grp_partial_lhs = access->grp_partial_lhs;
1532 bool first_scalar = is_gimple_reg_type (access->type);
1533 bool unscalarizable_region = access->grp_unscalarizable_region;
1535 if (first || access->offset >= high)
1537 first = false;
1538 low = access->offset;
1539 high = access->offset + access->size;
1541 else if (access->offset > low && access->offset + access->size > high)
1542 return NULL;
1543 else
1544 gcc_assert (access->offset >= low
1545 && access->offset + access->size <= high);
1547 j = i + 1;
1548 while (j < access_count)
1550 struct access *ac2 = VEC_index (access_p, access_vec, j);
1551 if (ac2->offset != access->offset || ac2->size != access->size)
1552 break;
1553 if (ac2->write)
1554 grp_write = true;
1555 else
1557 if (grp_read)
1558 multiple_reads = true;
1559 else
1560 grp_read = true;
1562 grp_assignment_read |= ac2->grp_assignment_read;
1563 grp_partial_lhs |= ac2->grp_partial_lhs;
1564 unscalarizable_region |= ac2->grp_unscalarizable_region;
1565 total_scalarization |= ac2->total_scalarization;
1566 relink_to_new_repr (access, ac2);
1568 /* If there are both aggregate-type and scalar-type accesses with
1569 this combination of size and offset, the comparison function
1570 should have put the scalars first. */
1571 gcc_assert (first_scalar || !is_gimple_reg_type (ac2->type));
1572 ac2->group_representative = access;
1573 j++;
1576 i = j;
1578 access->group_representative = access;
1579 access->grp_write = grp_write;
1580 access->grp_read = grp_read;
1581 access->grp_assignment_read = grp_assignment_read;
1582 access->grp_hint = multiple_reads || total_scalarization;
1583 access->grp_partial_lhs = grp_partial_lhs;
1584 access->grp_unscalarizable_region = unscalarizable_region;
1585 if (access->first_link)
1586 add_access_to_work_queue (access);
1588 *prev_acc_ptr = access;
1589 prev_acc_ptr = &access->next_grp;
1592 gcc_assert (res == VEC_index (access_p, access_vec, 0));
1593 return res;
1596 /* Create a variable for the given ACCESS which determines the type, name and a
1597 few other properties. Return the variable declaration and store it also to
1598 ACCESS->replacement. */
1600 static tree
1601 create_access_replacement (struct access *access, bool rename)
1603 tree repl;
1605 repl = create_tmp_var (access->type, "SR");
1606 get_var_ann (repl);
1607 add_referenced_var (repl);
1608 if (rename)
1609 mark_sym_for_renaming (repl);
1611 if (!access->grp_partial_lhs
1612 && (TREE_CODE (access->type) == COMPLEX_TYPE
1613 || TREE_CODE (access->type) == VECTOR_TYPE))
1614 DECL_GIMPLE_REG_P (repl) = 1;
1616 DECL_SOURCE_LOCATION (repl) = DECL_SOURCE_LOCATION (access->base);
1617 DECL_ARTIFICIAL (repl) = 1;
1618 DECL_IGNORED_P (repl) = DECL_IGNORED_P (access->base);
1620 if (DECL_NAME (access->base)
1621 && !DECL_IGNORED_P (access->base)
1622 && !DECL_ARTIFICIAL (access->base))
1624 char *pretty_name = make_fancy_name (access->expr);
1625 tree debug_expr = unshare_expr (access->expr), d;
1627 DECL_NAME (repl) = get_identifier (pretty_name);
1628 obstack_free (&name_obstack, pretty_name);
1630 /* Get rid of any SSA_NAMEs embedded in debug_expr,
1631 as DECL_DEBUG_EXPR isn't considered when looking for still
1632 used SSA_NAMEs and thus they could be freed. All debug info
1633 generation cares is whether something is constant or variable
1634 and that get_ref_base_and_extent works properly on the
1635 expression. */
1636 for (d = debug_expr; handled_component_p (d); d = TREE_OPERAND (d, 0))
1637 switch (TREE_CODE (d))
1639 case ARRAY_REF:
1640 case ARRAY_RANGE_REF:
1641 if (TREE_OPERAND (d, 1)
1642 && TREE_CODE (TREE_OPERAND (d, 1)) == SSA_NAME)
1643 TREE_OPERAND (d, 1) = SSA_NAME_VAR (TREE_OPERAND (d, 1));
1644 if (TREE_OPERAND (d, 3)
1645 && TREE_CODE (TREE_OPERAND (d, 3)) == SSA_NAME)
1646 TREE_OPERAND (d, 3) = SSA_NAME_VAR (TREE_OPERAND (d, 3));
1647 /* FALLTHRU */
1648 case COMPONENT_REF:
1649 if (TREE_OPERAND (d, 2)
1650 && TREE_CODE (TREE_OPERAND (d, 2)) == SSA_NAME)
1651 TREE_OPERAND (d, 2) = SSA_NAME_VAR (TREE_OPERAND (d, 2));
1652 break;
1653 default:
1654 break;
1656 SET_DECL_DEBUG_EXPR (repl, debug_expr);
1657 DECL_DEBUG_EXPR_IS_FROM (repl) = 1;
1658 TREE_NO_WARNING (repl) = TREE_NO_WARNING (access->base);
1660 else
1661 TREE_NO_WARNING (repl) = 1;
1663 if (dump_file)
1665 fprintf (dump_file, "Created a replacement for ");
1666 print_generic_expr (dump_file, access->base, 0);
1667 fprintf (dump_file, " offset: %u, size: %u: ",
1668 (unsigned) access->offset, (unsigned) access->size);
1669 print_generic_expr (dump_file, repl, 0);
1670 fprintf (dump_file, "\n");
1672 sra_stats.replacements++;
1674 return repl;
1677 /* Return ACCESS scalar replacement, create it if it does not exist yet. */
1679 static inline tree
1680 get_access_replacement (struct access *access)
1682 gcc_assert (access->grp_to_be_replaced);
1684 if (!access->replacement_decl)
1685 access->replacement_decl = create_access_replacement (access, true);
1686 return access->replacement_decl;
1689 /* Return ACCESS scalar replacement, create it if it does not exist yet but do
1690 not mark it for renaming. */
1692 static inline tree
1693 get_unrenamed_access_replacement (struct access *access)
1695 gcc_assert (!access->grp_to_be_replaced);
1697 if (!access->replacement_decl)
1698 access->replacement_decl = create_access_replacement (access, false);
1699 return access->replacement_decl;
1703 /* Build a subtree of accesses rooted in *ACCESS, and move the pointer in the
1704 linked list along the way. Stop when *ACCESS is NULL or the access pointed
1705 to it is not "within" the root. Return false iff some accesses partially
1706 overlap. */
1708 static bool
1709 build_access_subtree (struct access **access)
1711 struct access *root = *access, *last_child = NULL;
1712 HOST_WIDE_INT limit = root->offset + root->size;
1714 *access = (*access)->next_grp;
1715 while (*access && (*access)->offset + (*access)->size <= limit)
1717 if (!last_child)
1718 root->first_child = *access;
1719 else
1720 last_child->next_sibling = *access;
1721 last_child = *access;
1723 if (!build_access_subtree (access))
1724 return false;
1727 if (*access && (*access)->offset < limit)
1728 return false;
1730 return true;
1733 /* Build a tree of access representatives, ACCESS is the pointer to the first
1734 one, others are linked in a list by the next_grp field. Return false iff
1735 some accesses partially overlap. */
1737 static bool
1738 build_access_trees (struct access *access)
1740 while (access)
1742 struct access *root = access;
1744 if (!build_access_subtree (&access))
1745 return false;
1746 root->next_grp = access;
1748 return true;
1751 /* Return true if expr contains some ARRAY_REFs into a variable bounded
1752 array. */
1754 static bool
1755 expr_with_var_bounded_array_refs_p (tree expr)
1757 while (handled_component_p (expr))
1759 if (TREE_CODE (expr) == ARRAY_REF
1760 && !host_integerp (array_ref_low_bound (expr), 0))
1761 return true;
1762 expr = TREE_OPERAND (expr, 0);
1764 return false;
1767 enum mark_read_status { SRA_MR_NOT_READ, SRA_MR_READ, SRA_MR_ASSIGN_READ};
1769 /* Analyze the subtree of accesses rooted in ROOT, scheduling replacements when
1770 both seeming beneficial and when ALLOW_REPLACEMENTS allows it. Also set all
1771 sorts of access flags appropriately along the way, notably always set
1772 grp_read and grp_assign_read according to MARK_READ and grp_write when
1773 MARK_WRITE is true. */
1775 static bool
1776 analyze_access_subtree (struct access *root, bool allow_replacements,
1777 enum mark_read_status mark_read, bool mark_write)
1779 struct access *child;
1780 HOST_WIDE_INT limit = root->offset + root->size;
1781 HOST_WIDE_INT covered_to = root->offset;
1782 bool scalar = is_gimple_reg_type (root->type);
1783 bool hole = false, sth_created = false;
1784 bool direct_read = root->grp_read;
1786 if (mark_read == SRA_MR_ASSIGN_READ)
1788 root->grp_read = 1;
1789 root->grp_assignment_read = 1;
1791 if (mark_read == SRA_MR_READ)
1792 root->grp_read = 1;
1793 else if (root->grp_assignment_read)
1794 mark_read = SRA_MR_ASSIGN_READ;
1795 else if (root->grp_read)
1796 mark_read = SRA_MR_READ;
1798 if (mark_write)
1799 root->grp_write = true;
1800 else if (root->grp_write)
1801 mark_write = true;
1803 if (root->grp_unscalarizable_region)
1804 allow_replacements = false;
1806 if (allow_replacements && expr_with_var_bounded_array_refs_p (root->expr))
1807 allow_replacements = false;
1809 for (child = root->first_child; child; child = child->next_sibling)
1811 if (!hole && child->offset < covered_to)
1812 hole = true;
1813 else
1814 covered_to += child->size;
1816 sth_created |= analyze_access_subtree (child,
1817 allow_replacements && !scalar,
1818 mark_read, mark_write);
1820 root->grp_unscalarized_data |= child->grp_unscalarized_data;
1821 hole |= !child->grp_covered;
1824 if (allow_replacements && scalar && !root->first_child
1825 && (root->grp_hint
1826 || (root->grp_write && (direct_read || root->grp_assignment_read)))
1827 /* We must not ICE later on when trying to build an access to the
1828 original data within the aggregate even when it is impossible to do in
1829 a defined way like in the PR 42703 testcase. Therefore we check
1830 pre-emptively here that we will be able to do that. */
1831 && build_ref_for_offset (NULL, TREE_TYPE (root->base), root->offset,
1832 root->type, false))
1834 if (dump_file && (dump_flags & TDF_DETAILS))
1836 fprintf (dump_file, "Marking ");
1837 print_generic_expr (dump_file, root->base, 0);
1838 fprintf (dump_file, " offset: %u, size: %u: ",
1839 (unsigned) root->offset, (unsigned) root->size);
1840 fprintf (dump_file, " to be replaced.\n");
1843 root->grp_to_be_replaced = 1;
1844 sth_created = true;
1845 hole = false;
1847 else if (covered_to < limit)
1848 hole = true;
1850 if (sth_created && !hole)
1852 root->grp_covered = 1;
1853 return true;
1855 if (root->grp_write || TREE_CODE (root->base) == PARM_DECL)
1856 root->grp_unscalarized_data = 1; /* not covered and written to */
1857 if (sth_created)
1858 return true;
1859 return false;
1862 /* Analyze all access trees linked by next_grp by the means of
1863 analyze_access_subtree. */
1864 static bool
1865 analyze_access_trees (struct access *access)
1867 bool ret = false;
1869 while (access)
1871 if (analyze_access_subtree (access, true, SRA_MR_NOT_READ, false))
1872 ret = true;
1873 access = access->next_grp;
1876 return ret;
1879 /* Return true iff a potential new child of LACC at offset OFFSET and with size
1880 SIZE would conflict with an already existing one. If exactly such a child
1881 already exists in LACC, store a pointer to it in EXACT_MATCH. */
1883 static bool
1884 child_would_conflict_in_lacc (struct access *lacc, HOST_WIDE_INT norm_offset,
1885 HOST_WIDE_INT size, struct access **exact_match)
1887 struct access *child;
1889 for (child = lacc->first_child; child; child = child->next_sibling)
1891 if (child->offset == norm_offset && child->size == size)
1893 *exact_match = child;
1894 return true;
1897 if (child->offset < norm_offset + size
1898 && child->offset + child->size > norm_offset)
1899 return true;
1902 return false;
1905 /* Create a new child access of PARENT, with all properties just like MODEL
1906 except for its offset and with its grp_write false and grp_read true.
1907 Return the new access or NULL if it cannot be created. Note that this access
1908 is created long after all splicing and sorting, it's not located in any
1909 access vector and is automatically a representative of its group. */
1911 static struct access *
1912 create_artificial_child_access (struct access *parent, struct access *model,
1913 HOST_WIDE_INT new_offset)
1915 struct access *access;
1916 struct access **child;
1917 tree expr = parent->base;;
1919 gcc_assert (!model->grp_unscalarizable_region);
1921 if (!build_ref_for_offset (&expr, TREE_TYPE (expr), new_offset,
1922 model->type, false))
1923 return NULL;
1925 access = (struct access *) pool_alloc (access_pool);
1926 memset (access, 0, sizeof (struct access));
1927 access->base = parent->base;
1928 access->expr = expr;
1929 access->offset = new_offset;
1930 access->size = model->size;
1931 access->type = model->type;
1932 access->grp_write = true;
1933 access->grp_read = false;
1935 child = &parent->first_child;
1936 while (*child && (*child)->offset < new_offset)
1937 child = &(*child)->next_sibling;
1939 access->next_sibling = *child;
1940 *child = access;
1942 return access;
1946 /* Propagate all subaccesses of RACC across an assignment link to LACC. Return
1947 true if any new subaccess was created. Additionally, if RACC is a scalar
1948 access but LACC is not, change the type of the latter, if possible. */
1950 static bool
1951 propagate_subaccesses_across_link (struct access *lacc, struct access *racc)
1953 struct access *rchild;
1954 HOST_WIDE_INT norm_delta = lacc->offset - racc->offset;
1955 bool ret = false;
1957 if (is_gimple_reg_type (lacc->type)
1958 || lacc->grp_unscalarizable_region
1959 || racc->grp_unscalarizable_region)
1960 return false;
1962 if (!lacc->first_child && !racc->first_child
1963 && is_gimple_reg_type (racc->type))
1965 tree t = lacc->base;
1967 if (build_ref_for_offset (&t, TREE_TYPE (t), lacc->offset, racc->type,
1968 false))
1970 lacc->expr = t;
1971 lacc->type = racc->type;
1973 return false;
1976 for (rchild = racc->first_child; rchild; rchild = rchild->next_sibling)
1978 struct access *new_acc = NULL;
1979 HOST_WIDE_INT norm_offset = rchild->offset + norm_delta;
1981 if (rchild->grp_unscalarizable_region)
1982 continue;
1984 if (child_would_conflict_in_lacc (lacc, norm_offset, rchild->size,
1985 &new_acc))
1987 if (new_acc)
1989 rchild->grp_hint = 1;
1990 new_acc->grp_hint |= new_acc->grp_read;
1991 if (rchild->first_child)
1992 ret |= propagate_subaccesses_across_link (new_acc, rchild);
1994 continue;
1997 /* If a (part of) a union field is on the RHS of an assignment, it can
1998 have sub-accesses which do not make sense on the LHS (PR 40351).
1999 Check that this is not the case. */
2000 if (!build_ref_for_offset (NULL, TREE_TYPE (lacc->base), norm_offset,
2001 rchild->type, false))
2002 continue;
2004 rchild->grp_hint = 1;
2005 new_acc = create_artificial_child_access (lacc, rchild, norm_offset);
2006 if (new_acc)
2008 ret = true;
2009 if (racc->first_child)
2010 propagate_subaccesses_across_link (new_acc, rchild);
2014 return ret;
2017 /* Propagate all subaccesses across assignment links. */
2019 static void
2020 propagate_all_subaccesses (void)
2022 while (work_queue_head)
2024 struct access *racc = pop_access_from_work_queue ();
2025 struct assign_link *link;
2027 gcc_assert (racc->first_link);
2029 for (link = racc->first_link; link; link = link->next)
2031 struct access *lacc = link->lacc;
2033 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (lacc->base)))
2034 continue;
2035 lacc = lacc->group_representative;
2036 if (propagate_subaccesses_across_link (lacc, racc)
2037 && lacc->first_link)
2038 add_access_to_work_queue (lacc);
2043 /* Go through all accesses collected throughout the (intraprocedural) analysis
2044 stage, exclude overlapping ones, identify representatives and build trees
2045 out of them, making decisions about scalarization on the way. Return true
2046 iff there are any to-be-scalarized variables after this stage. */
2048 static bool
2049 analyze_all_variable_accesses (void)
2051 int res = 0;
2052 bitmap tmp = BITMAP_ALLOC (NULL);
2053 bitmap_iterator bi;
2054 unsigned i, max_total_scalarization_size;
2056 max_total_scalarization_size = UNITS_PER_WORD * BITS_PER_UNIT
2057 * MOVE_RATIO (optimize_function_for_speed_p (cfun));
2059 EXECUTE_IF_SET_IN_BITMAP (candidate_bitmap, 0, i, bi)
2060 if (bitmap_bit_p (should_scalarize_away_bitmap, i)
2061 && !bitmap_bit_p (cannot_scalarize_away_bitmap, i))
2063 tree var = referenced_var (i);
2065 if (TREE_CODE (var) == VAR_DECL
2066 && ((unsigned) tree_low_cst (TYPE_SIZE (TREE_TYPE (var)), 1)
2067 <= max_total_scalarization_size)
2068 && type_consists_of_records_p (TREE_TYPE (var)))
2070 completely_scalarize_record (var, var, 0, var);
2071 if (dump_file && (dump_flags & TDF_DETAILS))
2073 fprintf (dump_file, "Will attempt to totally scalarize ");
2074 print_generic_expr (dump_file, var, 0);
2075 fprintf (dump_file, " (UID: %u): \n", DECL_UID (var));
2080 bitmap_copy (tmp, candidate_bitmap);
2081 EXECUTE_IF_SET_IN_BITMAP (tmp, 0, i, bi)
2083 tree var = referenced_var (i);
2084 struct access *access;
2086 access = sort_and_splice_var_accesses (var);
2087 if (!access || !build_access_trees (access))
2088 disqualify_candidate (var,
2089 "No or inhibitingly overlapping accesses.");
2092 propagate_all_subaccesses ();
2094 bitmap_copy (tmp, candidate_bitmap);
2095 EXECUTE_IF_SET_IN_BITMAP (tmp, 0, i, bi)
2097 tree var = referenced_var (i);
2098 struct access *access = get_first_repr_for_decl (var);
2100 if (analyze_access_trees (access))
2102 res++;
2103 if (dump_file && (dump_flags & TDF_DETAILS))
2105 fprintf (dump_file, "\nAccess trees for ");
2106 print_generic_expr (dump_file, var, 0);
2107 fprintf (dump_file, " (UID: %u): \n", DECL_UID (var));
2108 dump_access_tree (dump_file, access);
2109 fprintf (dump_file, "\n");
2112 else
2113 disqualify_candidate (var, "No scalar replacements to be created.");
2116 BITMAP_FREE (tmp);
2118 if (res)
2120 statistics_counter_event (cfun, "Scalarized aggregates", res);
2121 return true;
2123 else
2124 return false;
2127 /* Return true iff a reference statement into aggregate AGG can be built for
2128 every single to-be-replaced accesses that is a child of ACCESS, its sibling
2129 or a child of its sibling. TOP_OFFSET is the offset from the processed
2130 access subtree that has to be subtracted from offset of each access. */
2132 static bool
2133 ref_expr_for_all_replacements_p (struct access *access, tree agg,
2134 HOST_WIDE_INT top_offset)
2138 if (access->grp_to_be_replaced
2139 && !build_ref_for_offset (NULL, TREE_TYPE (agg),
2140 access->offset - top_offset,
2141 access->type, false))
2142 return false;
2144 if (access->first_child
2145 && !ref_expr_for_all_replacements_p (access->first_child, agg,
2146 top_offset))
2147 return false;
2149 access = access->next_sibling;
2151 while (access);
2153 return true;
2156 /* Generate statements copying scalar replacements of accesses within a subtree
2157 into or out of AGG. ACCESS is the first child of the root of the subtree to
2158 be processed. AGG is an aggregate type expression (can be a declaration but
2159 does not have to be, it can for example also be an indirect_ref).
2160 TOP_OFFSET is the offset of the processed subtree which has to be subtracted
2161 from offsets of individual accesses to get corresponding offsets for AGG.
2162 If CHUNK_SIZE is non-null, copy only replacements in the interval
2163 <start_offset, start_offset + chunk_size>, otherwise copy all. GSI is a
2164 statement iterator used to place the new statements. WRITE should be true
2165 when the statements should write from AGG to the replacement and false if
2166 vice versa. if INSERT_AFTER is true, new statements will be added after the
2167 current statement in GSI, they will be added before the statement
2168 otherwise. */
2170 static void
2171 generate_subtree_copies (struct access *access, tree agg,
2172 HOST_WIDE_INT top_offset,
2173 HOST_WIDE_INT start_offset, HOST_WIDE_INT chunk_size,
2174 gimple_stmt_iterator *gsi, bool write,
2175 bool insert_after)
2179 tree expr = agg;
2181 if (chunk_size && access->offset >= start_offset + chunk_size)
2182 return;
2184 if (access->grp_to_be_replaced
2185 && (chunk_size == 0
2186 || access->offset + access->size > start_offset))
2188 tree repl = get_access_replacement (access);
2189 bool ref_found;
2190 gimple stmt;
2192 ref_found = build_ref_for_offset (&expr, TREE_TYPE (agg),
2193 access->offset - top_offset,
2194 access->type, false);
2195 gcc_assert (ref_found);
2197 if (write)
2199 if (access->grp_partial_lhs)
2200 expr = force_gimple_operand_gsi (gsi, expr, true, NULL_TREE,
2201 !insert_after,
2202 insert_after ? GSI_NEW_STMT
2203 : GSI_SAME_STMT);
2204 stmt = gimple_build_assign (repl, expr);
2206 else
2208 TREE_NO_WARNING (repl) = 1;
2209 if (access->grp_partial_lhs)
2210 repl = force_gimple_operand_gsi (gsi, repl, true, NULL_TREE,
2211 !insert_after,
2212 insert_after ? GSI_NEW_STMT
2213 : GSI_SAME_STMT);
2214 stmt = gimple_build_assign (expr, repl);
2217 if (insert_after)
2218 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2219 else
2220 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2221 update_stmt (stmt);
2222 sra_stats.subtree_copies++;
2225 if (access->first_child)
2226 generate_subtree_copies (access->first_child, agg, top_offset,
2227 start_offset, chunk_size, gsi,
2228 write, insert_after);
2230 access = access->next_sibling;
2232 while (access);
2235 /* Assign zero to all scalar replacements in an access subtree. ACCESS is the
2236 the root of the subtree to be processed. GSI is the statement iterator used
2237 for inserting statements which are added after the current statement if
2238 INSERT_AFTER is true or before it otherwise. */
2240 static void
2241 init_subtree_with_zero (struct access *access, gimple_stmt_iterator *gsi,
2242 bool insert_after)
2245 struct access *child;
2247 if (access->grp_to_be_replaced)
2249 gimple stmt;
2251 stmt = gimple_build_assign (get_access_replacement (access),
2252 fold_convert (access->type,
2253 integer_zero_node));
2254 if (insert_after)
2255 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2256 else
2257 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2258 update_stmt (stmt);
2261 for (child = access->first_child; child; child = child->next_sibling)
2262 init_subtree_with_zero (child, gsi, insert_after);
2265 /* Search for an access representative for the given expression EXPR and
2266 return it or NULL if it cannot be found. */
2268 static struct access *
2269 get_access_for_expr (tree expr)
2271 HOST_WIDE_INT offset, size, max_size;
2272 tree base;
2274 /* FIXME: This should not be necessary but Ada produces V_C_Es with a type of
2275 a different size than the size of its argument and we need the latter
2276 one. */
2277 if (TREE_CODE (expr) == VIEW_CONVERT_EXPR)
2278 expr = TREE_OPERAND (expr, 0);
2280 base = get_ref_base_and_extent (expr, &offset, &size, &max_size);
2281 if (max_size == -1 || !DECL_P (base))
2282 return NULL;
2284 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (base)))
2285 return NULL;
2287 return get_var_base_offset_size_access (base, offset, max_size);
2290 /* Replace the expression EXPR with a scalar replacement if there is one and
2291 generate other statements to do type conversion or subtree copying if
2292 necessary. GSI is used to place newly created statements, WRITE is true if
2293 the expression is being written to (it is on a LHS of a statement or output
2294 in an assembly statement). */
2296 static bool
2297 sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write)
2299 struct access *access;
2300 tree type, bfr;
2302 if (TREE_CODE (*expr) == BIT_FIELD_REF)
2304 bfr = *expr;
2305 expr = &TREE_OPERAND (*expr, 0);
2307 else
2308 bfr = NULL_TREE;
2310 if (TREE_CODE (*expr) == REALPART_EXPR || TREE_CODE (*expr) == IMAGPART_EXPR)
2311 expr = &TREE_OPERAND (*expr, 0);
2312 access = get_access_for_expr (*expr);
2313 if (!access)
2314 return false;
2315 type = TREE_TYPE (*expr);
2317 if (access->grp_to_be_replaced)
2319 tree repl = get_access_replacement (access);
2320 /* If we replace a non-register typed access simply use the original
2321 access expression to extract the scalar component afterwards.
2322 This happens if scalarizing a function return value or parameter
2323 like in gcc.c-torture/execute/20041124-1.c, 20050316-1.c and
2324 gcc.c-torture/compile/20011217-1.c.
2326 We also want to use this when accessing a complex or vector which can
2327 be accessed as a different type too, potentially creating a need for
2328 type conversion (see PR42196) and when scalarized unions are involved
2329 in assembler statements (see PR42398). */
2330 if (!useless_type_conversion_p (type, access->type))
2332 tree ref = access->base;
2333 bool ok;
2335 ok = build_ref_for_offset (&ref, TREE_TYPE (ref),
2336 access->offset, access->type, false);
2337 gcc_assert (ok);
2339 if (write)
2341 gimple stmt;
2343 if (access->grp_partial_lhs)
2344 ref = force_gimple_operand_gsi (gsi, ref, true, NULL_TREE,
2345 false, GSI_NEW_STMT);
2346 stmt = gimple_build_assign (repl, ref);
2347 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2349 else
2351 gimple stmt;
2353 if (access->grp_partial_lhs)
2354 repl = force_gimple_operand_gsi (gsi, repl, true, NULL_TREE,
2355 true, GSI_SAME_STMT);
2356 stmt = gimple_build_assign (ref, repl);
2357 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2360 else
2361 *expr = repl;
2362 sra_stats.exprs++;
2365 if (access->first_child)
2367 HOST_WIDE_INT start_offset, chunk_size;
2368 if (bfr
2369 && host_integerp (TREE_OPERAND (bfr, 1), 1)
2370 && host_integerp (TREE_OPERAND (bfr, 2), 1))
2372 chunk_size = tree_low_cst (TREE_OPERAND (bfr, 1), 1);
2373 start_offset = access->offset
2374 + tree_low_cst (TREE_OPERAND (bfr, 2), 1);
2376 else
2377 start_offset = chunk_size = 0;
2379 generate_subtree_copies (access->first_child, access->base, 0,
2380 start_offset, chunk_size, gsi, write, write);
2382 return true;
2385 /* Where scalar replacements of the RHS have been written to when a replacement
2386 of a LHS of an assigments cannot be direclty loaded from a replacement of
2387 the RHS. */
2388 enum unscalarized_data_handling { SRA_UDH_NONE, /* Nothing done so far. */
2389 SRA_UDH_RIGHT, /* Data flushed to the RHS. */
2390 SRA_UDH_LEFT }; /* Data flushed to the LHS. */
2392 /* Store all replacements in the access tree rooted in TOP_RACC either to their
2393 base aggregate if there are unscalarized data or directly to LHS
2394 otherwise. */
2396 static enum unscalarized_data_handling
2397 handle_unscalarized_data_in_subtree (struct access *top_racc, tree lhs,
2398 gimple_stmt_iterator *gsi)
2400 if (top_racc->grp_unscalarized_data)
2402 generate_subtree_copies (top_racc->first_child, top_racc->base, 0, 0, 0,
2403 gsi, false, false);
2404 return SRA_UDH_RIGHT;
2406 else
2408 generate_subtree_copies (top_racc->first_child, lhs, top_racc->offset,
2409 0, 0, gsi, false, false);
2410 return SRA_UDH_LEFT;
2415 /* Try to generate statements to load all sub-replacements in an access
2416 (sub)tree (LACC is the first child) from scalar replacements in the TOP_RACC
2417 (sub)tree. If that is not possible, refresh the TOP_RACC base aggregate and
2418 load the accesses from it. LEFT_OFFSET is the offset of the left whole
2419 subtree being copied, RIGHT_OFFSET is the same thing for the right subtree.
2420 NEW_GSI is stmt iterator used for statement insertions after the original
2421 assignment, OLD_GSI is used to insert statements before the assignment.
2422 *REFRESHED keeps the information whether we have needed to refresh
2423 replacements of the LHS and from which side of the assignments this takes
2424 place. */
2426 static void
2427 load_assign_lhs_subreplacements (struct access *lacc, struct access *top_racc,
2428 HOST_WIDE_INT left_offset,
2429 HOST_WIDE_INT right_offset,
2430 gimple_stmt_iterator *old_gsi,
2431 gimple_stmt_iterator *new_gsi,
2432 enum unscalarized_data_handling *refreshed,
2433 tree lhs)
2435 location_t loc = EXPR_LOCATION (lacc->expr);
2438 if (lacc->grp_to_be_replaced)
2440 struct access *racc;
2441 HOST_WIDE_INT offset = lacc->offset - left_offset + right_offset;
2442 gimple stmt;
2443 tree rhs;
2445 racc = find_access_in_subtree (top_racc, offset, lacc->size);
2446 if (racc && racc->grp_to_be_replaced)
2448 rhs = get_access_replacement (racc);
2449 if (!useless_type_conversion_p (lacc->type, racc->type))
2450 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR, lacc->type, rhs);
2452 else
2454 /* No suitable access on the right hand side, need to load from
2455 the aggregate. See if we have to update it first... */
2456 if (*refreshed == SRA_UDH_NONE)
2457 *refreshed = handle_unscalarized_data_in_subtree (top_racc,
2458 lhs, old_gsi);
2460 if (*refreshed == SRA_UDH_LEFT)
2462 bool repl_found;
2464 rhs = lacc->base;
2465 repl_found = build_ref_for_offset (&rhs, TREE_TYPE (rhs),
2466 lacc->offset, lacc->type,
2467 false);
2468 gcc_assert (repl_found);
2470 else
2472 bool repl_found;
2474 rhs = top_racc->base;
2475 repl_found = build_ref_for_offset (&rhs,
2476 TREE_TYPE (top_racc->base),
2477 offset, lacc->type, false);
2478 gcc_assert (repl_found);
2482 stmt = gimple_build_assign (get_access_replacement (lacc), rhs);
2483 gsi_insert_after (new_gsi, stmt, GSI_NEW_STMT);
2484 update_stmt (stmt);
2485 sra_stats.subreplacements++;
2487 else if (*refreshed == SRA_UDH_NONE
2488 && lacc->grp_read && !lacc->grp_covered)
2489 *refreshed = handle_unscalarized_data_in_subtree (top_racc, lhs,
2490 old_gsi);
2492 if (lacc->first_child)
2493 load_assign_lhs_subreplacements (lacc->first_child, top_racc,
2494 left_offset, right_offset,
2495 old_gsi, new_gsi, refreshed, lhs);
2496 lacc = lacc->next_sibling;
2498 while (lacc);
2501 /* Result code for SRA assignment modification. */
2502 enum assignment_mod_result { SRA_AM_NONE, /* nothing done for the stmt */
2503 SRA_AM_MODIFIED, /* stmt changed but not
2504 removed */
2505 SRA_AM_REMOVED }; /* stmt eliminated */
2507 /* Modify assignments with a CONSTRUCTOR on their RHS. STMT contains a pointer
2508 to the assignment and GSI is the statement iterator pointing at it. Returns
2509 the same values as sra_modify_assign. */
2511 static enum assignment_mod_result
2512 sra_modify_constructor_assign (gimple *stmt, gimple_stmt_iterator *gsi)
2514 tree lhs = gimple_assign_lhs (*stmt);
2515 struct access *acc;
2517 acc = get_access_for_expr (lhs);
2518 if (!acc)
2519 return SRA_AM_NONE;
2521 if (VEC_length (constructor_elt,
2522 CONSTRUCTOR_ELTS (gimple_assign_rhs1 (*stmt))) > 0)
2524 /* I have never seen this code path trigger but if it can happen the
2525 following should handle it gracefully. */
2526 if (access_has_children_p (acc))
2527 generate_subtree_copies (acc->first_child, acc->base, 0, 0, 0, gsi,
2528 true, true);
2529 return SRA_AM_MODIFIED;
2532 if (acc->grp_covered)
2534 init_subtree_with_zero (acc, gsi, false);
2535 unlink_stmt_vdef (*stmt);
2536 gsi_remove (gsi, true);
2537 return SRA_AM_REMOVED;
2539 else
2541 init_subtree_with_zero (acc, gsi, true);
2542 return SRA_AM_MODIFIED;
2546 /* Create and return a new suitable default definition SSA_NAME for RACC which
2547 is an access describing an uninitialized part of an aggregate that is being
2548 loaded. */
2550 static tree
2551 get_repl_default_def_ssa_name (struct access *racc)
2553 tree repl, decl;
2555 decl = get_unrenamed_access_replacement (racc);
2557 repl = gimple_default_def (cfun, decl);
2558 if (!repl)
2560 repl = make_ssa_name (decl, gimple_build_nop ());
2561 set_default_def (decl, repl);
2564 return repl;
2567 /* Examine both sides of the assignment statement pointed to by STMT, replace
2568 them with a scalare replacement if there is one and generate copying of
2569 replacements if scalarized aggregates have been used in the assignment. GSI
2570 is used to hold generated statements for type conversions and subtree
2571 copying. */
2573 static enum assignment_mod_result
2574 sra_modify_assign (gimple *stmt, gimple_stmt_iterator *gsi)
2576 struct access *lacc, *racc;
2577 tree lhs, rhs;
2578 bool modify_this_stmt = false;
2579 bool force_gimple_rhs = false;
2580 location_t loc = gimple_location (*stmt);
2581 gimple_stmt_iterator orig_gsi = *gsi;
2583 if (!gimple_assign_single_p (*stmt))
2584 return SRA_AM_NONE;
2585 lhs = gimple_assign_lhs (*stmt);
2586 rhs = gimple_assign_rhs1 (*stmt);
2588 if (TREE_CODE (rhs) == CONSTRUCTOR)
2589 return sra_modify_constructor_assign (stmt, gsi);
2591 if (TREE_CODE (rhs) == REALPART_EXPR || TREE_CODE (lhs) == REALPART_EXPR
2592 || TREE_CODE (rhs) == IMAGPART_EXPR || TREE_CODE (lhs) == IMAGPART_EXPR
2593 || TREE_CODE (rhs) == BIT_FIELD_REF || TREE_CODE (lhs) == BIT_FIELD_REF)
2595 modify_this_stmt = sra_modify_expr (gimple_assign_rhs1_ptr (*stmt),
2596 gsi, false);
2597 modify_this_stmt |= sra_modify_expr (gimple_assign_lhs_ptr (*stmt),
2598 gsi, true);
2599 return modify_this_stmt ? SRA_AM_MODIFIED : SRA_AM_NONE;
2602 lacc = get_access_for_expr (lhs);
2603 racc = get_access_for_expr (rhs);
2604 if (!lacc && !racc)
2605 return SRA_AM_NONE;
2607 if (lacc && lacc->grp_to_be_replaced)
2609 lhs = get_access_replacement (lacc);
2610 gimple_assign_set_lhs (*stmt, lhs);
2611 modify_this_stmt = true;
2612 if (lacc->grp_partial_lhs)
2613 force_gimple_rhs = true;
2614 sra_stats.exprs++;
2617 if (racc && racc->grp_to_be_replaced)
2619 rhs = get_access_replacement (racc);
2620 modify_this_stmt = true;
2621 if (racc->grp_partial_lhs)
2622 force_gimple_rhs = true;
2623 sra_stats.exprs++;
2626 if (modify_this_stmt)
2628 if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
2630 /* If we can avoid creating a VIEW_CONVERT_EXPR do so.
2631 ??? This should move to fold_stmt which we simply should
2632 call after building a VIEW_CONVERT_EXPR here. */
2633 if (AGGREGATE_TYPE_P (TREE_TYPE (lhs))
2634 && !access_has_children_p (lacc))
2636 tree expr = lhs;
2637 if (build_ref_for_offset (&expr, TREE_TYPE (lhs), 0,
2638 TREE_TYPE (rhs), false))
2640 lhs = expr;
2641 gimple_assign_set_lhs (*stmt, expr);
2644 else if (AGGREGATE_TYPE_P (TREE_TYPE (rhs))
2645 && !access_has_children_p (racc))
2647 tree expr = rhs;
2648 if (build_ref_for_offset (&expr, TREE_TYPE (rhs), 0,
2649 TREE_TYPE (lhs), false))
2650 rhs = expr;
2652 if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
2654 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR, TREE_TYPE (lhs), rhs);
2655 if (is_gimple_reg_type (TREE_TYPE (lhs))
2656 && TREE_CODE (lhs) != SSA_NAME)
2657 force_gimple_rhs = true;
2662 /* From this point on, the function deals with assignments in between
2663 aggregates when at least one has scalar reductions of some of its
2664 components. There are three possible scenarios: Both the LHS and RHS have
2665 to-be-scalarized components, 2) only the RHS has or 3) only the LHS has.
2667 In the first case, we would like to load the LHS components from RHS
2668 components whenever possible. If that is not possible, we would like to
2669 read it directly from the RHS (after updating it by storing in it its own
2670 components). If there are some necessary unscalarized data in the LHS,
2671 those will be loaded by the original assignment too. If neither of these
2672 cases happen, the original statement can be removed. Most of this is done
2673 by load_assign_lhs_subreplacements.
2675 In the second case, we would like to store all RHS scalarized components
2676 directly into LHS and if they cover the aggregate completely, remove the
2677 statement too. In the third case, we want the LHS components to be loaded
2678 directly from the RHS (DSE will remove the original statement if it
2679 becomes redundant).
2681 This is a bit complex but manageable when types match and when unions do
2682 not cause confusion in a way that we cannot really load a component of LHS
2683 from the RHS or vice versa (the access representing this level can have
2684 subaccesses that are accessible only through a different union field at a
2685 higher level - different from the one used in the examined expression).
2686 Unions are fun.
2688 Therefore, I specially handle a fourth case, happening when there is a
2689 specific type cast or it is impossible to locate a scalarized subaccess on
2690 the other side of the expression. If that happens, I simply "refresh" the
2691 RHS by storing in it is scalarized components leave the original statement
2692 there to do the copying and then load the scalar replacements of the LHS.
2693 This is what the first branch does. */
2695 if (gimple_has_volatile_ops (*stmt)
2696 || contains_view_convert_expr_p (rhs)
2697 || contains_view_convert_expr_p (lhs)
2698 || (access_has_children_p (racc)
2699 && !ref_expr_for_all_replacements_p (racc, lhs, racc->offset))
2700 || (access_has_children_p (lacc)
2701 && !ref_expr_for_all_replacements_p (lacc, rhs, lacc->offset)))
2703 if (access_has_children_p (racc))
2704 generate_subtree_copies (racc->first_child, racc->base, 0, 0, 0,
2705 gsi, false, false);
2706 if (access_has_children_p (lacc))
2707 generate_subtree_copies (lacc->first_child, lacc->base, 0, 0, 0,
2708 gsi, true, true);
2709 sra_stats.separate_lhs_rhs_handling++;
2711 else
2713 if (access_has_children_p (lacc) && access_has_children_p (racc))
2715 gimple_stmt_iterator orig_gsi = *gsi;
2716 enum unscalarized_data_handling refreshed;
2718 if (lacc->grp_read && !lacc->grp_covered)
2719 refreshed = handle_unscalarized_data_in_subtree (racc, lhs, gsi);
2720 else
2721 refreshed = SRA_UDH_NONE;
2723 load_assign_lhs_subreplacements (lacc->first_child, racc,
2724 lacc->offset, racc->offset,
2725 &orig_gsi, gsi, &refreshed, lhs);
2726 if (refreshed != SRA_UDH_RIGHT)
2728 gsi_next (gsi);
2729 unlink_stmt_vdef (*stmt);
2730 gsi_remove (&orig_gsi, true);
2731 sra_stats.deleted++;
2732 return SRA_AM_REMOVED;
2735 else
2737 if (racc)
2739 if (!racc->grp_to_be_replaced && !racc->grp_unscalarized_data)
2741 if (dump_file)
2743 fprintf (dump_file, "Removing load: ");
2744 print_gimple_stmt (dump_file, *stmt, 0, 0);
2747 if (TREE_CODE (lhs) == SSA_NAME)
2749 rhs = get_repl_default_def_ssa_name (racc);
2750 if (!useless_type_conversion_p (TREE_TYPE (lhs),
2751 TREE_TYPE (rhs)))
2752 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR,
2753 TREE_TYPE (lhs), rhs);
2755 else
2757 if (racc->first_child)
2758 generate_subtree_copies (racc->first_child, lhs,
2759 racc->offset, 0, 0, gsi,
2760 false, false);
2762 gcc_assert (*stmt == gsi_stmt (*gsi));
2763 unlink_stmt_vdef (*stmt);
2764 gsi_remove (gsi, true);
2765 sra_stats.deleted++;
2766 return SRA_AM_REMOVED;
2769 else if (racc->first_child)
2770 generate_subtree_copies (racc->first_child, lhs,
2771 racc->offset, 0, 0, gsi, false, true);
2773 if (access_has_children_p (lacc))
2774 generate_subtree_copies (lacc->first_child, rhs, lacc->offset,
2775 0, 0, gsi, true, true);
2779 /* This gimplification must be done after generate_subtree_copies, lest we
2780 insert the subtree copies in the middle of the gimplified sequence. */
2781 if (force_gimple_rhs)
2782 rhs = force_gimple_operand_gsi (&orig_gsi, rhs, true, NULL_TREE,
2783 true, GSI_SAME_STMT);
2784 if (gimple_assign_rhs1 (*stmt) != rhs)
2786 gimple_assign_set_rhs_from_tree (&orig_gsi, rhs);
2787 gcc_assert (*stmt == gsi_stmt (orig_gsi));
2790 return modify_this_stmt ? SRA_AM_MODIFIED : SRA_AM_NONE;
2793 /* Traverse the function body and all modifications as decided in
2794 analyze_all_variable_accesses. Return true iff the CFG has been
2795 changed. */
2797 static bool
2798 sra_modify_function_body (void)
2800 bool cfg_changed = false;
2801 basic_block bb;
2803 FOR_EACH_BB (bb)
2805 gimple_stmt_iterator gsi = gsi_start_bb (bb);
2806 while (!gsi_end_p (gsi))
2808 gimple stmt = gsi_stmt (gsi);
2809 enum assignment_mod_result assign_result;
2810 bool modified = false, deleted = false;
2811 tree *t;
2812 unsigned i;
2814 switch (gimple_code (stmt))
2816 case GIMPLE_RETURN:
2817 t = gimple_return_retval_ptr (stmt);
2818 if (*t != NULL_TREE)
2819 modified |= sra_modify_expr (t, &gsi, false);
2820 break;
2822 case GIMPLE_ASSIGN:
2823 assign_result = sra_modify_assign (&stmt, &gsi);
2824 modified |= assign_result == SRA_AM_MODIFIED;
2825 deleted = assign_result == SRA_AM_REMOVED;
2826 break;
2828 case GIMPLE_CALL:
2829 /* Operands must be processed before the lhs. */
2830 for (i = 0; i < gimple_call_num_args (stmt); i++)
2832 t = gimple_call_arg_ptr (stmt, i);
2833 modified |= sra_modify_expr (t, &gsi, false);
2836 if (gimple_call_lhs (stmt))
2838 t = gimple_call_lhs_ptr (stmt);
2839 modified |= sra_modify_expr (t, &gsi, true);
2841 break;
2843 case GIMPLE_ASM:
2844 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
2846 t = &TREE_VALUE (gimple_asm_input_op (stmt, i));
2847 modified |= sra_modify_expr (t, &gsi, false);
2849 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
2851 t = &TREE_VALUE (gimple_asm_output_op (stmt, i));
2852 modified |= sra_modify_expr (t, &gsi, true);
2854 break;
2856 default:
2857 break;
2860 if (modified)
2862 update_stmt (stmt);
2863 if (maybe_clean_eh_stmt (stmt)
2864 && gimple_purge_dead_eh_edges (gimple_bb (stmt)))
2865 cfg_changed = true;
2867 if (!deleted)
2868 gsi_next (&gsi);
2872 return cfg_changed;
2875 /* Generate statements initializing scalar replacements of parts of function
2876 parameters. */
2878 static void
2879 initialize_parameter_reductions (void)
2881 gimple_stmt_iterator gsi;
2882 gimple_seq seq = NULL;
2883 tree parm;
2885 for (parm = DECL_ARGUMENTS (current_function_decl);
2886 parm;
2887 parm = DECL_CHAIN (parm))
2889 VEC (access_p, heap) *access_vec;
2890 struct access *access;
2892 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
2893 continue;
2894 access_vec = get_base_access_vector (parm);
2895 if (!access_vec)
2896 continue;
2898 if (!seq)
2900 seq = gimple_seq_alloc ();
2901 gsi = gsi_start (seq);
2904 for (access = VEC_index (access_p, access_vec, 0);
2905 access;
2906 access = access->next_grp)
2907 generate_subtree_copies (access, parm, 0, 0, 0, &gsi, true, true);
2910 if (seq)
2911 gsi_insert_seq_on_edge_immediate (single_succ_edge (ENTRY_BLOCK_PTR), seq);
2914 /* The "main" function of intraprocedural SRA passes. Runs the analysis and if
2915 it reveals there are components of some aggregates to be scalarized, it runs
2916 the required transformations. */
2917 static unsigned int
2918 perform_intra_sra (void)
2920 int ret = 0;
2921 sra_initialize ();
2923 if (!find_var_candidates ())
2924 goto out;
2926 if (!scan_function ())
2927 goto out;
2929 if (!analyze_all_variable_accesses ())
2930 goto out;
2932 if (sra_modify_function_body ())
2933 ret = TODO_update_ssa | TODO_cleanup_cfg;
2934 else
2935 ret = TODO_update_ssa;
2936 initialize_parameter_reductions ();
2938 statistics_counter_event (cfun, "Scalar replacements created",
2939 sra_stats.replacements);
2940 statistics_counter_event (cfun, "Modified expressions", sra_stats.exprs);
2941 statistics_counter_event (cfun, "Subtree copy stmts",
2942 sra_stats.subtree_copies);
2943 statistics_counter_event (cfun, "Subreplacement stmts",
2944 sra_stats.subreplacements);
2945 statistics_counter_event (cfun, "Deleted stmts", sra_stats.deleted);
2946 statistics_counter_event (cfun, "Separate LHS and RHS handling",
2947 sra_stats.separate_lhs_rhs_handling);
2949 out:
2950 sra_deinitialize ();
2951 return ret;
2954 /* Perform early intraprocedural SRA. */
2955 static unsigned int
2956 early_intra_sra (void)
2958 sra_mode = SRA_MODE_EARLY_INTRA;
2959 return perform_intra_sra ();
2962 /* Perform "late" intraprocedural SRA. */
2963 static unsigned int
2964 late_intra_sra (void)
2966 sra_mode = SRA_MODE_INTRA;
2967 return perform_intra_sra ();
2971 static bool
2972 gate_intra_sra (void)
2974 return flag_tree_sra != 0 && dbg_cnt (tree_sra);
2978 struct gimple_opt_pass pass_sra_early =
2981 GIMPLE_PASS,
2982 "esra", /* name */
2983 gate_intra_sra, /* gate */
2984 early_intra_sra, /* execute */
2985 NULL, /* sub */
2986 NULL, /* next */
2987 0, /* static_pass_number */
2988 TV_TREE_SRA, /* tv_id */
2989 PROP_cfg | PROP_ssa, /* properties_required */
2990 0, /* properties_provided */
2991 0, /* properties_destroyed */
2992 0, /* todo_flags_start */
2993 TODO_dump_func
2994 | TODO_update_ssa
2995 | TODO_ggc_collect
2996 | TODO_verify_ssa /* todo_flags_finish */
3000 struct gimple_opt_pass pass_sra =
3003 GIMPLE_PASS,
3004 "sra", /* name */
3005 gate_intra_sra, /* gate */
3006 late_intra_sra, /* execute */
3007 NULL, /* sub */
3008 NULL, /* next */
3009 0, /* static_pass_number */
3010 TV_TREE_SRA, /* tv_id */
3011 PROP_cfg | PROP_ssa, /* properties_required */
3012 0, /* properties_provided */
3013 0, /* properties_destroyed */
3014 TODO_update_address_taken, /* todo_flags_start */
3015 TODO_dump_func
3016 | TODO_update_ssa
3017 | TODO_ggc_collect
3018 | TODO_verify_ssa /* todo_flags_finish */
3023 /* Return true iff PARM (which must be a parm_decl) is an unused scalar
3024 parameter. */
3026 static bool
3027 is_unused_scalar_param (tree parm)
3029 tree name;
3030 return (is_gimple_reg (parm)
3031 && (!(name = gimple_default_def (cfun, parm))
3032 || has_zero_uses (name)));
3035 /* Scan immediate uses of a default definition SSA name of a parameter PARM and
3036 examine whether there are any direct or otherwise infeasible ones. If so,
3037 return true, otherwise return false. PARM must be a gimple register with a
3038 non-NULL default definition. */
3040 static bool
3041 ptr_parm_has_direct_uses (tree parm)
3043 imm_use_iterator ui;
3044 gimple stmt;
3045 tree name = gimple_default_def (cfun, parm);
3046 bool ret = false;
3048 FOR_EACH_IMM_USE_STMT (stmt, ui, name)
3050 int uses_ok = 0;
3051 use_operand_p use_p;
3053 if (is_gimple_debug (stmt))
3054 continue;
3056 /* Valid uses include dereferences on the lhs and the rhs. */
3057 if (gimple_has_lhs (stmt))
3059 tree lhs = gimple_get_lhs (stmt);
3060 while (handled_component_p (lhs))
3061 lhs = TREE_OPERAND (lhs, 0);
3062 if (TREE_CODE (lhs) == MEM_REF
3063 && TREE_OPERAND (lhs, 0) == name
3064 && integer_zerop (TREE_OPERAND (lhs, 1))
3065 && types_compatible_p (TREE_TYPE (lhs),
3066 TREE_TYPE (TREE_TYPE (name))))
3067 uses_ok++;
3069 if (gimple_assign_single_p (stmt))
3071 tree rhs = gimple_assign_rhs1 (stmt);
3072 while (handled_component_p (rhs))
3073 rhs = TREE_OPERAND (rhs, 0);
3074 if (TREE_CODE (rhs) == MEM_REF
3075 && TREE_OPERAND (rhs, 0) == name
3076 && integer_zerop (TREE_OPERAND (rhs, 1))
3077 && types_compatible_p (TREE_TYPE (rhs),
3078 TREE_TYPE (TREE_TYPE (name))))
3079 uses_ok++;
3081 else if (is_gimple_call (stmt))
3083 unsigned i;
3084 for (i = 0; i < gimple_call_num_args (stmt); ++i)
3086 tree arg = gimple_call_arg (stmt, i);
3087 while (handled_component_p (arg))
3088 arg = TREE_OPERAND (arg, 0);
3089 if (TREE_CODE (arg) == MEM_REF
3090 && TREE_OPERAND (arg, 0) == name
3091 && integer_zerop (TREE_OPERAND (arg, 1))
3092 && types_compatible_p (TREE_TYPE (arg),
3093 TREE_TYPE (TREE_TYPE (name))))
3094 uses_ok++;
3098 /* If the number of valid uses does not match the number of
3099 uses in this stmt there is an unhandled use. */
3100 FOR_EACH_IMM_USE_ON_STMT (use_p, ui)
3101 --uses_ok;
3103 if (uses_ok != 0)
3104 ret = true;
3106 if (ret)
3107 BREAK_FROM_IMM_USE_STMT (ui);
3110 return ret;
3113 /* Identify candidates for reduction for IPA-SRA based on their type and mark
3114 them in candidate_bitmap. Note that these do not necessarily include
3115 parameter which are unused and thus can be removed. Return true iff any
3116 such candidate has been found. */
3118 static bool
3119 find_param_candidates (void)
3121 tree parm;
3122 int count = 0;
3123 bool ret = false;
3125 for (parm = DECL_ARGUMENTS (current_function_decl);
3126 parm;
3127 parm = DECL_CHAIN (parm))
3129 tree type = TREE_TYPE (parm);
3131 count++;
3133 if (TREE_THIS_VOLATILE (parm)
3134 || TREE_ADDRESSABLE (parm)
3135 || (!is_gimple_reg_type (type) && is_va_list_type (type)))
3136 continue;
3138 if (is_unused_scalar_param (parm))
3140 ret = true;
3141 continue;
3144 if (POINTER_TYPE_P (type))
3146 type = TREE_TYPE (type);
3148 if (TREE_CODE (type) == FUNCTION_TYPE
3149 || TYPE_VOLATILE (type)
3150 || (TREE_CODE (type) == ARRAY_TYPE
3151 && TYPE_NONALIASED_COMPONENT (type))
3152 || !is_gimple_reg (parm)
3153 || is_va_list_type (type)
3154 || ptr_parm_has_direct_uses (parm))
3155 continue;
3157 else if (!AGGREGATE_TYPE_P (type))
3158 continue;
3160 if (!COMPLETE_TYPE_P (type)
3161 || !host_integerp (TYPE_SIZE (type), 1)
3162 || tree_low_cst (TYPE_SIZE (type), 1) == 0
3163 || (AGGREGATE_TYPE_P (type)
3164 && type_internals_preclude_sra_p (type)))
3165 continue;
3167 bitmap_set_bit (candidate_bitmap, DECL_UID (parm));
3168 ret = true;
3169 if (dump_file && (dump_flags & TDF_DETAILS))
3171 fprintf (dump_file, "Candidate (%d): ", DECL_UID (parm));
3172 print_generic_expr (dump_file, parm, 0);
3173 fprintf (dump_file, "\n");
3177 func_param_count = count;
3178 return ret;
3181 /* Callback of walk_aliased_vdefs, marks the access passed as DATA as
3182 maybe_modified. */
3184 static bool
3185 mark_maybe_modified (ao_ref *ao ATTRIBUTE_UNUSED, tree vdef ATTRIBUTE_UNUSED,
3186 void *data)
3188 struct access *repr = (struct access *) data;
3190 repr->grp_maybe_modified = 1;
3191 return true;
3194 /* Analyze what representatives (in linked lists accessible from
3195 REPRESENTATIVES) can be modified by side effects of statements in the
3196 current function. */
3198 static void
3199 analyze_modified_params (VEC (access_p, heap) *representatives)
3201 int i;
3203 for (i = 0; i < func_param_count; i++)
3205 struct access *repr;
3207 for (repr = VEC_index (access_p, representatives, i);
3208 repr;
3209 repr = repr->next_grp)
3211 struct access *access;
3212 bitmap visited;
3213 ao_ref ar;
3215 if (no_accesses_p (repr))
3216 continue;
3217 if (!POINTER_TYPE_P (TREE_TYPE (repr->base))
3218 || repr->grp_maybe_modified)
3219 continue;
3221 ao_ref_init (&ar, repr->expr);
3222 visited = BITMAP_ALLOC (NULL);
3223 for (access = repr; access; access = access->next_sibling)
3225 /* All accesses are read ones, otherwise grp_maybe_modified would
3226 be trivially set. */
3227 walk_aliased_vdefs (&ar, gimple_vuse (access->stmt),
3228 mark_maybe_modified, repr, &visited);
3229 if (repr->grp_maybe_modified)
3230 break;
3232 BITMAP_FREE (visited);
3237 /* Propagate distances in bb_dereferences in the opposite direction than the
3238 control flow edges, in each step storing the maximum of the current value
3239 and the minimum of all successors. These steps are repeated until the table
3240 stabilizes. Note that BBs which might terminate the functions (according to
3241 final_bbs bitmap) never updated in this way. */
3243 static void
3244 propagate_dereference_distances (void)
3246 VEC (basic_block, heap) *queue;
3247 basic_block bb;
3249 queue = VEC_alloc (basic_block, heap, last_basic_block_for_function (cfun));
3250 VEC_quick_push (basic_block, queue, ENTRY_BLOCK_PTR);
3251 FOR_EACH_BB (bb)
3253 VEC_quick_push (basic_block, queue, bb);
3254 bb->aux = bb;
3257 while (!VEC_empty (basic_block, queue))
3259 edge_iterator ei;
3260 edge e;
3261 bool change = false;
3262 int i;
3264 bb = VEC_pop (basic_block, queue);
3265 bb->aux = NULL;
3267 if (bitmap_bit_p (final_bbs, bb->index))
3268 continue;
3270 for (i = 0; i < func_param_count; i++)
3272 int idx = bb->index * func_param_count + i;
3273 bool first = true;
3274 HOST_WIDE_INT inh = 0;
3276 FOR_EACH_EDGE (e, ei, bb->succs)
3278 int succ_idx = e->dest->index * func_param_count + i;
3280 if (e->src == EXIT_BLOCK_PTR)
3281 continue;
3283 if (first)
3285 first = false;
3286 inh = bb_dereferences [succ_idx];
3288 else if (bb_dereferences [succ_idx] < inh)
3289 inh = bb_dereferences [succ_idx];
3292 if (!first && bb_dereferences[idx] < inh)
3294 bb_dereferences[idx] = inh;
3295 change = true;
3299 if (change && !bitmap_bit_p (final_bbs, bb->index))
3300 FOR_EACH_EDGE (e, ei, bb->preds)
3302 if (e->src->aux)
3303 continue;
3305 e->src->aux = e->src;
3306 VEC_quick_push (basic_block, queue, e->src);
3310 VEC_free (basic_block, heap, queue);
3313 /* Dump a dereferences TABLE with heading STR to file F. */
3315 static void
3316 dump_dereferences_table (FILE *f, const char *str, HOST_WIDE_INT *table)
3318 basic_block bb;
3320 fprintf (dump_file, str);
3321 FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, EXIT_BLOCK_PTR, next_bb)
3323 fprintf (f, "%4i %i ", bb->index, bitmap_bit_p (final_bbs, bb->index));
3324 if (bb != EXIT_BLOCK_PTR)
3326 int i;
3327 for (i = 0; i < func_param_count; i++)
3329 int idx = bb->index * func_param_count + i;
3330 fprintf (f, " %4" HOST_WIDE_INT_PRINT "d", table[idx]);
3333 fprintf (f, "\n");
3335 fprintf (dump_file, "\n");
3338 /* Determine what (parts of) parameters passed by reference that are not
3339 assigned to are not certainly dereferenced in this function and thus the
3340 dereferencing cannot be safely moved to the caller without potentially
3341 introducing a segfault. Mark such REPRESENTATIVES as
3342 grp_not_necessarilly_dereferenced.
3344 The dereferenced maximum "distance," i.e. the offset + size of the accessed
3345 part is calculated rather than simple booleans are calculated for each
3346 pointer parameter to handle cases when only a fraction of the whole
3347 aggregate is allocated (see testsuite/gcc.c-torture/execute/ipa-sra-2.c for
3348 an example).
3350 The maximum dereference distances for each pointer parameter and BB are
3351 already stored in bb_dereference. This routine simply propagates these
3352 values upwards by propagate_dereference_distances and then compares the
3353 distances of individual parameters in the ENTRY BB to the equivalent
3354 distances of each representative of a (fraction of a) parameter. */
3356 static void
3357 analyze_caller_dereference_legality (VEC (access_p, heap) *representatives)
3359 int i;
3361 if (dump_file && (dump_flags & TDF_DETAILS))
3362 dump_dereferences_table (dump_file,
3363 "Dereference table before propagation:\n",
3364 bb_dereferences);
3366 propagate_dereference_distances ();
3368 if (dump_file && (dump_flags & TDF_DETAILS))
3369 dump_dereferences_table (dump_file,
3370 "Dereference table after propagation:\n",
3371 bb_dereferences);
3373 for (i = 0; i < func_param_count; i++)
3375 struct access *repr = VEC_index (access_p, representatives, i);
3376 int idx = ENTRY_BLOCK_PTR->index * func_param_count + i;
3378 if (!repr || no_accesses_p (repr))
3379 continue;
3383 if ((repr->offset + repr->size) > bb_dereferences[idx])
3384 repr->grp_not_necessarilly_dereferenced = 1;
3385 repr = repr->next_grp;
3387 while (repr);
3391 /* Return the representative access for the parameter declaration PARM if it is
3392 a scalar passed by reference which is not written to and the pointer value
3393 is not used directly. Thus, if it is legal to dereference it in the caller
3394 and we can rule out modifications through aliases, such parameter should be
3395 turned into one passed by value. Return NULL otherwise. */
3397 static struct access *
3398 unmodified_by_ref_scalar_representative (tree parm)
3400 int i, access_count;
3401 struct access *repr;
3402 VEC (access_p, heap) *access_vec;
3404 access_vec = get_base_access_vector (parm);
3405 gcc_assert (access_vec);
3406 repr = VEC_index (access_p, access_vec, 0);
3407 if (repr->write)
3408 return NULL;
3409 repr->group_representative = repr;
3411 access_count = VEC_length (access_p, access_vec);
3412 for (i = 1; i < access_count; i++)
3414 struct access *access = VEC_index (access_p, access_vec, i);
3415 if (access->write)
3416 return NULL;
3417 access->group_representative = repr;
3418 access->next_sibling = repr->next_sibling;
3419 repr->next_sibling = access;
3422 repr->grp_read = 1;
3423 repr->grp_scalar_ptr = 1;
3424 return repr;
3427 /* Return true iff this access precludes IPA-SRA of the parameter it is
3428 associated with. */
3430 static bool
3431 access_precludes_ipa_sra_p (struct access *access)
3433 /* Avoid issues such as the second simple testcase in PR 42025. The problem
3434 is incompatible assign in a call statement (and possibly even in asm
3435 statements). This can be relaxed by using a new temporary but only for
3436 non-TREE_ADDRESSABLE types and is probably not worth the complexity. (In
3437 intraprocedural SRA we deal with this by keeping the old aggregate around,
3438 something we cannot do in IPA-SRA.) */
3439 if (access->write
3440 && (is_gimple_call (access->stmt)
3441 || gimple_code (access->stmt) == GIMPLE_ASM))
3442 return true;
3444 return false;
3448 /* Sort collected accesses for parameter PARM, identify representatives for
3449 each accessed region and link them together. Return NULL if there are
3450 different but overlapping accesses, return the special ptr value meaning
3451 there are no accesses for this parameter if that is the case and return the
3452 first representative otherwise. Set *RO_GRP if there is a group of accesses
3453 with only read (i.e. no write) accesses. */
3455 static struct access *
3456 splice_param_accesses (tree parm, bool *ro_grp)
3458 int i, j, access_count, group_count;
3459 int agg_size, total_size = 0;
3460 struct access *access, *res, **prev_acc_ptr = &res;
3461 VEC (access_p, heap) *access_vec;
3463 access_vec = get_base_access_vector (parm);
3464 if (!access_vec)
3465 return &no_accesses_representant;
3466 access_count = VEC_length (access_p, access_vec);
3468 qsort (VEC_address (access_p, access_vec), access_count, sizeof (access_p),
3469 compare_access_positions);
3471 i = 0;
3472 total_size = 0;
3473 group_count = 0;
3474 while (i < access_count)
3476 bool modification;
3477 access = VEC_index (access_p, access_vec, i);
3478 modification = access->write;
3479 if (access_precludes_ipa_sra_p (access))
3480 return NULL;
3482 /* Access is about to become group representative unless we find some
3483 nasty overlap which would preclude us from breaking this parameter
3484 apart. */
3486 j = i + 1;
3487 while (j < access_count)
3489 struct access *ac2 = VEC_index (access_p, access_vec, j);
3490 if (ac2->offset != access->offset)
3492 /* All or nothing law for parameters. */
3493 if (access->offset + access->size > ac2->offset)
3494 return NULL;
3495 else
3496 break;
3498 else if (ac2->size != access->size)
3499 return NULL;
3501 if (access_precludes_ipa_sra_p (ac2))
3502 return NULL;
3504 modification |= ac2->write;
3505 ac2->group_representative = access;
3506 ac2->next_sibling = access->next_sibling;
3507 access->next_sibling = ac2;
3508 j++;
3511 group_count++;
3512 access->grp_maybe_modified = modification;
3513 if (!modification)
3514 *ro_grp = true;
3515 *prev_acc_ptr = access;
3516 prev_acc_ptr = &access->next_grp;
3517 total_size += access->size;
3518 i = j;
3521 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3522 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm))), 1);
3523 else
3524 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (parm)), 1);
3525 if (total_size >= agg_size)
3526 return NULL;
3528 gcc_assert (group_count > 0);
3529 return res;
3532 /* Decide whether parameters with representative accesses given by REPR should
3533 be reduced into components. */
3535 static int
3536 decide_one_param_reduction (struct access *repr)
3538 int total_size, cur_parm_size, agg_size, new_param_count, parm_size_limit;
3539 bool by_ref;
3540 tree parm;
3542 parm = repr->base;
3543 cur_parm_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (parm)), 1);
3544 gcc_assert (cur_parm_size > 0);
3546 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3548 by_ref = true;
3549 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm))), 1);
3551 else
3553 by_ref = false;
3554 agg_size = cur_parm_size;
3557 if (dump_file)
3559 struct access *acc;
3560 fprintf (dump_file, "Evaluating PARAM group sizes for ");
3561 print_generic_expr (dump_file, parm, 0);
3562 fprintf (dump_file, " (UID: %u): \n", DECL_UID (parm));
3563 for (acc = repr; acc; acc = acc->next_grp)
3564 dump_access (dump_file, acc, true);
3567 total_size = 0;
3568 new_param_count = 0;
3570 for (; repr; repr = repr->next_grp)
3572 gcc_assert (parm == repr->base);
3573 new_param_count++;
3575 if (!by_ref || (!repr->grp_maybe_modified
3576 && !repr->grp_not_necessarilly_dereferenced))
3577 total_size += repr->size;
3578 else
3579 total_size += cur_parm_size;
3582 gcc_assert (new_param_count > 0);
3584 if (optimize_function_for_size_p (cfun))
3585 parm_size_limit = cur_parm_size;
3586 else
3587 parm_size_limit = (PARAM_VALUE (PARAM_IPA_SRA_PTR_GROWTH_FACTOR)
3588 * cur_parm_size);
3590 if (total_size < agg_size
3591 && total_size <= parm_size_limit)
3593 if (dump_file)
3594 fprintf (dump_file, " ....will be split into %i components\n",
3595 new_param_count);
3596 return new_param_count;
3598 else
3599 return 0;
3602 /* The order of the following enums is important, we need to do extra work for
3603 UNUSED_PARAMS, BY_VAL_ACCESSES and UNMODIF_BY_REF_ACCESSES. */
3604 enum ipa_splicing_result { NO_GOOD_ACCESS, UNUSED_PARAMS, BY_VAL_ACCESSES,
3605 MODIF_BY_REF_ACCESSES, UNMODIF_BY_REF_ACCESSES };
3607 /* Identify representatives of all accesses to all candidate parameters for
3608 IPA-SRA. Return result based on what representatives have been found. */
3610 static enum ipa_splicing_result
3611 splice_all_param_accesses (VEC (access_p, heap) **representatives)
3613 enum ipa_splicing_result result = NO_GOOD_ACCESS;
3614 tree parm;
3615 struct access *repr;
3617 *representatives = VEC_alloc (access_p, heap, func_param_count);
3619 for (parm = DECL_ARGUMENTS (current_function_decl);
3620 parm;
3621 parm = DECL_CHAIN (parm))
3623 if (is_unused_scalar_param (parm))
3625 VEC_quick_push (access_p, *representatives,
3626 &no_accesses_representant);
3627 if (result == NO_GOOD_ACCESS)
3628 result = UNUSED_PARAMS;
3630 else if (POINTER_TYPE_P (TREE_TYPE (parm))
3631 && is_gimple_reg_type (TREE_TYPE (TREE_TYPE (parm)))
3632 && bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3634 repr = unmodified_by_ref_scalar_representative (parm);
3635 VEC_quick_push (access_p, *representatives, repr);
3636 if (repr)
3637 result = UNMODIF_BY_REF_ACCESSES;
3639 else if (bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3641 bool ro_grp = false;
3642 repr = splice_param_accesses (parm, &ro_grp);
3643 VEC_quick_push (access_p, *representatives, repr);
3645 if (repr && !no_accesses_p (repr))
3647 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3649 if (ro_grp)
3650 result = UNMODIF_BY_REF_ACCESSES;
3651 else if (result < MODIF_BY_REF_ACCESSES)
3652 result = MODIF_BY_REF_ACCESSES;
3654 else if (result < BY_VAL_ACCESSES)
3655 result = BY_VAL_ACCESSES;
3657 else if (no_accesses_p (repr) && (result == NO_GOOD_ACCESS))
3658 result = UNUSED_PARAMS;
3660 else
3661 VEC_quick_push (access_p, *representatives, NULL);
3664 if (result == NO_GOOD_ACCESS)
3666 VEC_free (access_p, heap, *representatives);
3667 *representatives = NULL;
3668 return NO_GOOD_ACCESS;
3671 return result;
3674 /* Return the index of BASE in PARMS. Abort if it is not found. */
3676 static inline int
3677 get_param_index (tree base, VEC(tree, heap) *parms)
3679 int i, len;
3681 len = VEC_length (tree, parms);
3682 for (i = 0; i < len; i++)
3683 if (VEC_index (tree, parms, i) == base)
3684 return i;
3685 gcc_unreachable ();
3688 /* Convert the decisions made at the representative level into compact
3689 parameter adjustments. REPRESENTATIVES are pointers to first
3690 representatives of each param accesses, ADJUSTMENTS_COUNT is the expected
3691 final number of adjustments. */
3693 static ipa_parm_adjustment_vec
3694 turn_representatives_into_adjustments (VEC (access_p, heap) *representatives,
3695 int adjustments_count)
3697 VEC (tree, heap) *parms;
3698 ipa_parm_adjustment_vec adjustments;
3699 tree parm;
3700 int i;
3702 gcc_assert (adjustments_count > 0);
3703 parms = ipa_get_vector_of_formal_parms (current_function_decl);
3704 adjustments = VEC_alloc (ipa_parm_adjustment_t, heap, adjustments_count);
3705 parm = DECL_ARGUMENTS (current_function_decl);
3706 for (i = 0; i < func_param_count; i++, parm = DECL_CHAIN (parm))
3708 struct access *repr = VEC_index (access_p, representatives, i);
3710 if (!repr || no_accesses_p (repr))
3712 struct ipa_parm_adjustment *adj;
3714 adj = VEC_quick_push (ipa_parm_adjustment_t, adjustments, NULL);
3715 memset (adj, 0, sizeof (*adj));
3716 adj->base_index = get_param_index (parm, parms);
3717 adj->base = parm;
3718 if (!repr)
3719 adj->copy_param = 1;
3720 else
3721 adj->remove_param = 1;
3723 else
3725 struct ipa_parm_adjustment *adj;
3726 int index = get_param_index (parm, parms);
3728 for (; repr; repr = repr->next_grp)
3730 adj = VEC_quick_push (ipa_parm_adjustment_t, adjustments, NULL);
3731 memset (adj, 0, sizeof (*adj));
3732 gcc_assert (repr->base == parm);
3733 adj->base_index = index;
3734 adj->base = repr->base;
3735 adj->type = repr->type;
3736 adj->offset = repr->offset;
3737 adj->by_ref = (POINTER_TYPE_P (TREE_TYPE (repr->base))
3738 && (repr->grp_maybe_modified
3739 || repr->grp_not_necessarilly_dereferenced));
3744 VEC_free (tree, heap, parms);
3745 return adjustments;
3748 /* Analyze the collected accesses and produce a plan what to do with the
3749 parameters in the form of adjustments, NULL meaning nothing. */
3751 static ipa_parm_adjustment_vec
3752 analyze_all_param_acesses (void)
3754 enum ipa_splicing_result repr_state;
3755 bool proceed = false;
3756 int i, adjustments_count = 0;
3757 VEC (access_p, heap) *representatives;
3758 ipa_parm_adjustment_vec adjustments;
3760 repr_state = splice_all_param_accesses (&representatives);
3761 if (repr_state == NO_GOOD_ACCESS)
3762 return NULL;
3764 /* If there are any parameters passed by reference which are not modified
3765 directly, we need to check whether they can be modified indirectly. */
3766 if (repr_state == UNMODIF_BY_REF_ACCESSES)
3768 analyze_caller_dereference_legality (representatives);
3769 analyze_modified_params (representatives);
3772 for (i = 0; i < func_param_count; i++)
3774 struct access *repr = VEC_index (access_p, representatives, i);
3776 if (repr && !no_accesses_p (repr))
3778 if (repr->grp_scalar_ptr)
3780 adjustments_count++;
3781 if (repr->grp_not_necessarilly_dereferenced
3782 || repr->grp_maybe_modified)
3783 VEC_replace (access_p, representatives, i, NULL);
3784 else
3786 proceed = true;
3787 sra_stats.scalar_by_ref_to_by_val++;
3790 else
3792 int new_components = decide_one_param_reduction (repr);
3794 if (new_components == 0)
3796 VEC_replace (access_p, representatives, i, NULL);
3797 adjustments_count++;
3799 else
3801 adjustments_count += new_components;
3802 sra_stats.aggregate_params_reduced++;
3803 sra_stats.param_reductions_created += new_components;
3804 proceed = true;
3808 else
3810 if (no_accesses_p (repr))
3812 proceed = true;
3813 sra_stats.deleted_unused_parameters++;
3815 adjustments_count++;
3819 if (!proceed && dump_file)
3820 fprintf (dump_file, "NOT proceeding to change params.\n");
3822 if (proceed)
3823 adjustments = turn_representatives_into_adjustments (representatives,
3824 adjustments_count);
3825 else
3826 adjustments = NULL;
3828 VEC_free (access_p, heap, representatives);
3829 return adjustments;
3832 /* If a parameter replacement identified by ADJ does not yet exist in the form
3833 of declaration, create it and record it, otherwise return the previously
3834 created one. */
3836 static tree
3837 get_replaced_param_substitute (struct ipa_parm_adjustment *adj)
3839 tree repl;
3840 if (!adj->new_ssa_base)
3842 char *pretty_name = make_fancy_name (adj->base);
3844 repl = create_tmp_reg (TREE_TYPE (adj->base), "ISR");
3845 DECL_NAME (repl) = get_identifier (pretty_name);
3846 obstack_free (&name_obstack, pretty_name);
3848 get_var_ann (repl);
3849 add_referenced_var (repl);
3850 adj->new_ssa_base = repl;
3852 else
3853 repl = adj->new_ssa_base;
3854 return repl;
3857 /* Find the first adjustment for a particular parameter BASE in a vector of
3858 ADJUSTMENTS which is not a copy_param. Return NULL if there is no such
3859 adjustment. */
3861 static struct ipa_parm_adjustment *
3862 get_adjustment_for_base (ipa_parm_adjustment_vec adjustments, tree base)
3864 int i, len;
3866 len = VEC_length (ipa_parm_adjustment_t, adjustments);
3867 for (i = 0; i < len; i++)
3869 struct ipa_parm_adjustment *adj;
3871 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
3872 if (!adj->copy_param && adj->base == base)
3873 return adj;
3876 return NULL;
3879 /* If the statement STMT defines an SSA_NAME of a parameter which is to be
3880 removed because its value is not used, replace the SSA_NAME with a one
3881 relating to a created VAR_DECL together all of its uses and return true.
3882 ADJUSTMENTS is a pointer to an adjustments vector. */
3884 static bool
3885 replace_removed_params_ssa_names (gimple stmt,
3886 ipa_parm_adjustment_vec adjustments)
3888 struct ipa_parm_adjustment *adj;
3889 tree lhs, decl, repl, name;
3891 if (gimple_code (stmt) == GIMPLE_PHI)
3892 lhs = gimple_phi_result (stmt);
3893 else if (is_gimple_assign (stmt))
3894 lhs = gimple_assign_lhs (stmt);
3895 else if (is_gimple_call (stmt))
3896 lhs = gimple_call_lhs (stmt);
3897 else
3898 gcc_unreachable ();
3900 if (TREE_CODE (lhs) != SSA_NAME)
3901 return false;
3902 decl = SSA_NAME_VAR (lhs);
3903 if (TREE_CODE (decl) != PARM_DECL)
3904 return false;
3906 adj = get_adjustment_for_base (adjustments, decl);
3907 if (!adj)
3908 return false;
3910 repl = get_replaced_param_substitute (adj);
3911 name = make_ssa_name (repl, stmt);
3913 if (dump_file)
3915 fprintf (dump_file, "replacing an SSA name of a removed param ");
3916 print_generic_expr (dump_file, lhs, 0);
3917 fprintf (dump_file, " with ");
3918 print_generic_expr (dump_file, name, 0);
3919 fprintf (dump_file, "\n");
3922 if (is_gimple_assign (stmt))
3923 gimple_assign_set_lhs (stmt, name);
3924 else if (is_gimple_call (stmt))
3925 gimple_call_set_lhs (stmt, name);
3926 else
3927 gimple_phi_set_result (stmt, name);
3929 replace_uses_by (lhs, name);
3930 release_ssa_name (lhs);
3931 return true;
3934 /* If the expression *EXPR should be replaced by a reduction of a parameter, do
3935 so. ADJUSTMENTS is a pointer to a vector of adjustments. CONVERT
3936 specifies whether the function should care about type incompatibility the
3937 current and new expressions. If it is false, the function will leave
3938 incompatibility issues to the caller. Return true iff the expression
3939 was modified. */
3941 static bool
3942 sra_ipa_modify_expr (tree *expr, bool convert,
3943 ipa_parm_adjustment_vec adjustments)
3945 int i, len;
3946 struct ipa_parm_adjustment *adj, *cand = NULL;
3947 HOST_WIDE_INT offset, size, max_size;
3948 tree base, src;
3950 len = VEC_length (ipa_parm_adjustment_t, adjustments);
3952 if (TREE_CODE (*expr) == BIT_FIELD_REF
3953 || TREE_CODE (*expr) == IMAGPART_EXPR
3954 || TREE_CODE (*expr) == REALPART_EXPR)
3956 expr = &TREE_OPERAND (*expr, 0);
3957 convert = true;
3960 base = get_ref_base_and_extent (*expr, &offset, &size, &max_size);
3961 if (!base || size == -1 || max_size == -1)
3962 return false;
3964 if (TREE_CODE (base) == MEM_REF)
3966 offset += mem_ref_offset (base).low * BITS_PER_UNIT;
3967 base = TREE_OPERAND (base, 0);
3970 base = get_ssa_base_param (base);
3971 if (!base || TREE_CODE (base) != PARM_DECL)
3972 return false;
3974 for (i = 0; i < len; i++)
3976 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
3978 if (adj->base == base &&
3979 (adj->offset == offset || adj->remove_param))
3981 cand = adj;
3982 break;
3985 if (!cand || cand->copy_param || cand->remove_param)
3986 return false;
3988 if (cand->by_ref)
3989 src = build_simple_mem_ref (cand->reduction);
3990 else
3991 src = cand->reduction;
3993 if (dump_file && (dump_flags & TDF_DETAILS))
3995 fprintf (dump_file, "About to replace expr ");
3996 print_generic_expr (dump_file, *expr, 0);
3997 fprintf (dump_file, " with ");
3998 print_generic_expr (dump_file, src, 0);
3999 fprintf (dump_file, "\n");
4002 if (convert && !useless_type_conversion_p (TREE_TYPE (*expr), cand->type))
4004 tree vce = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (*expr), src);
4005 *expr = vce;
4007 else
4008 *expr = src;
4009 return true;
4012 /* If the statement pointed to by STMT_PTR contains any expressions that need
4013 to replaced with a different one as noted by ADJUSTMENTS, do so. Handle any
4014 potential type incompatibilities (GSI is used to accommodate conversion
4015 statements and must point to the statement). Return true iff the statement
4016 was modified. */
4018 static bool
4019 sra_ipa_modify_assign (gimple *stmt_ptr, gimple_stmt_iterator *gsi,
4020 ipa_parm_adjustment_vec adjustments)
4022 gimple stmt = *stmt_ptr;
4023 tree *lhs_p, *rhs_p;
4024 bool any;
4026 if (!gimple_assign_single_p (stmt))
4027 return false;
4029 rhs_p = gimple_assign_rhs1_ptr (stmt);
4030 lhs_p = gimple_assign_lhs_ptr (stmt);
4032 any = sra_ipa_modify_expr (rhs_p, false, adjustments);
4033 any |= sra_ipa_modify_expr (lhs_p, false, adjustments);
4034 if (any)
4036 tree new_rhs = NULL_TREE;
4038 if (!useless_type_conversion_p (TREE_TYPE (*lhs_p), TREE_TYPE (*rhs_p)))
4040 if (TREE_CODE (*rhs_p) == CONSTRUCTOR)
4042 /* V_C_Es of constructors can cause trouble (PR 42714). */
4043 if (is_gimple_reg_type (TREE_TYPE (*lhs_p)))
4044 *rhs_p = fold_convert (TREE_TYPE (*lhs_p), integer_zero_node);
4045 else
4046 *rhs_p = build_constructor (TREE_TYPE (*lhs_p), 0);
4048 else
4049 new_rhs = fold_build1_loc (gimple_location (stmt),
4050 VIEW_CONVERT_EXPR, TREE_TYPE (*lhs_p),
4051 *rhs_p);
4053 else if (REFERENCE_CLASS_P (*rhs_p)
4054 && is_gimple_reg_type (TREE_TYPE (*lhs_p))
4055 && !is_gimple_reg (*lhs_p))
4056 /* This can happen when an assignment in between two single field
4057 structures is turned into an assignment in between two pointers to
4058 scalars (PR 42237). */
4059 new_rhs = *rhs_p;
4061 if (new_rhs)
4063 tree tmp = force_gimple_operand_gsi (gsi, new_rhs, true, NULL_TREE,
4064 true, GSI_SAME_STMT);
4066 gimple_assign_set_rhs_from_tree (gsi, tmp);
4069 return true;
4072 return false;
4075 /* Traverse the function body and all modifications as described in
4076 ADJUSTMENTS. Return true iff the CFG has been changed. */
4078 static bool
4079 ipa_sra_modify_function_body (ipa_parm_adjustment_vec adjustments)
4081 bool cfg_changed = false;
4082 basic_block bb;
4084 FOR_EACH_BB (bb)
4086 gimple_stmt_iterator gsi;
4088 for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi))
4089 replace_removed_params_ssa_names (gsi_stmt (gsi), adjustments);
4091 gsi = gsi_start_bb (bb);
4092 while (!gsi_end_p (gsi))
4094 gimple stmt = gsi_stmt (gsi);
4095 bool modified = false;
4096 tree *t;
4097 unsigned i;
4099 switch (gimple_code (stmt))
4101 case GIMPLE_RETURN:
4102 t = gimple_return_retval_ptr (stmt);
4103 if (*t != NULL_TREE)
4104 modified |= sra_ipa_modify_expr (t, true, adjustments);
4105 break;
4107 case GIMPLE_ASSIGN:
4108 modified |= sra_ipa_modify_assign (&stmt, &gsi, adjustments);
4109 modified |= replace_removed_params_ssa_names (stmt, adjustments);
4110 break;
4112 case GIMPLE_CALL:
4113 /* Operands must be processed before the lhs. */
4114 for (i = 0; i < gimple_call_num_args (stmt); i++)
4116 t = gimple_call_arg_ptr (stmt, i);
4117 modified |= sra_ipa_modify_expr (t, true, adjustments);
4120 if (gimple_call_lhs (stmt))
4122 t = gimple_call_lhs_ptr (stmt);
4123 modified |= sra_ipa_modify_expr (t, false, adjustments);
4124 modified |= replace_removed_params_ssa_names (stmt,
4125 adjustments);
4127 break;
4129 case GIMPLE_ASM:
4130 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
4132 t = &TREE_VALUE (gimple_asm_input_op (stmt, i));
4133 modified |= sra_ipa_modify_expr (t, true, adjustments);
4135 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
4137 t = &TREE_VALUE (gimple_asm_output_op (stmt, i));
4138 modified |= sra_ipa_modify_expr (t, false, adjustments);
4140 break;
4142 default:
4143 break;
4146 if (modified)
4148 update_stmt (stmt);
4149 if (maybe_clean_eh_stmt (stmt)
4150 && gimple_purge_dead_eh_edges (gimple_bb (stmt)))
4151 cfg_changed = true;
4153 gsi_next (&gsi);
4157 return cfg_changed;
4160 /* Call gimple_debug_bind_reset_value on all debug statements describing
4161 gimple register parameters that are being removed or replaced. */
4163 static void
4164 sra_ipa_reset_debug_stmts (ipa_parm_adjustment_vec adjustments)
4166 int i, len;
4168 len = VEC_length (ipa_parm_adjustment_t, adjustments);
4169 for (i = 0; i < len; i++)
4171 struct ipa_parm_adjustment *adj;
4172 imm_use_iterator ui;
4173 gimple stmt;
4174 tree name;
4176 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
4177 if (adj->copy_param || !is_gimple_reg (adj->base))
4178 continue;
4179 name = gimple_default_def (cfun, adj->base);
4180 if (!name)
4181 continue;
4182 FOR_EACH_IMM_USE_STMT (stmt, ui, name)
4184 /* All other users must have been removed by
4185 ipa_sra_modify_function_body. */
4186 gcc_assert (is_gimple_debug (stmt));
4187 gimple_debug_bind_reset_value (stmt);
4188 update_stmt (stmt);
4193 /* Return true iff all callers have at least as many actual arguments as there
4194 are formal parameters in the current function. */
4196 static bool
4197 all_callers_have_enough_arguments_p (struct cgraph_node *node)
4199 struct cgraph_edge *cs;
4200 for (cs = node->callers; cs; cs = cs->next_caller)
4201 if (!callsite_has_enough_arguments_p (cs->call_stmt))
4202 return false;
4204 return true;
4208 /* Convert all callers of NODE to pass parameters as given in ADJUSTMENTS. */
4210 static void
4211 convert_callers (struct cgraph_node *node, tree old_decl,
4212 ipa_parm_adjustment_vec adjustments)
4214 tree old_cur_fndecl = current_function_decl;
4215 struct cgraph_edge *cs;
4216 basic_block this_block;
4217 bitmap recomputed_callers = BITMAP_ALLOC (NULL);
4219 for (cs = node->callers; cs; cs = cs->next_caller)
4221 current_function_decl = cs->caller->decl;
4222 push_cfun (DECL_STRUCT_FUNCTION (cs->caller->decl));
4224 if (dump_file)
4225 fprintf (dump_file, "Adjusting call (%i -> %i) %s -> %s\n",
4226 cs->caller->uid, cs->callee->uid,
4227 cgraph_node_name (cs->caller),
4228 cgraph_node_name (cs->callee));
4230 ipa_modify_call_arguments (cs, cs->call_stmt, adjustments);
4232 pop_cfun ();
4235 for (cs = node->callers; cs; cs = cs->next_caller)
4236 if (bitmap_set_bit (recomputed_callers, cs->caller->uid))
4237 compute_inline_parameters (cs->caller);
4238 BITMAP_FREE (recomputed_callers);
4240 current_function_decl = old_cur_fndecl;
4242 if (!encountered_recursive_call)
4243 return;
4245 FOR_EACH_BB (this_block)
4247 gimple_stmt_iterator gsi;
4249 for (gsi = gsi_start_bb (this_block); !gsi_end_p (gsi); gsi_next (&gsi))
4251 gimple stmt = gsi_stmt (gsi);
4252 tree call_fndecl;
4253 if (gimple_code (stmt) != GIMPLE_CALL)
4254 continue;
4255 call_fndecl = gimple_call_fndecl (stmt);
4256 if (call_fndecl == old_decl)
4258 if (dump_file)
4259 fprintf (dump_file, "Adjusting recursive call");
4260 gimple_call_set_fndecl (stmt, node->decl);
4261 ipa_modify_call_arguments (NULL, stmt, adjustments);
4266 return;
4269 /* Perform all the modification required in IPA-SRA for NODE to have parameters
4270 as given in ADJUSTMENTS. Return true iff the CFG has been changed. */
4272 static bool
4273 modify_function (struct cgraph_node *node, ipa_parm_adjustment_vec adjustments)
4275 struct cgraph_node *new_node;
4276 struct cgraph_edge *cs;
4277 bool cfg_changed;
4278 VEC (cgraph_edge_p, heap) * redirect_callers;
4279 int node_callers;
4281 node_callers = 0;
4282 for (cs = node->callers; cs != NULL; cs = cs->next_caller)
4283 node_callers++;
4284 redirect_callers = VEC_alloc (cgraph_edge_p, heap, node_callers);
4285 for (cs = node->callers; cs != NULL; cs = cs->next_caller)
4286 VEC_quick_push (cgraph_edge_p, redirect_callers, cs);
4288 rebuild_cgraph_edges ();
4289 pop_cfun ();
4290 current_function_decl = NULL_TREE;
4292 new_node = cgraph_function_versioning (node, redirect_callers, NULL, NULL,
4293 NULL, NULL, "isra");
4294 current_function_decl = new_node->decl;
4295 push_cfun (DECL_STRUCT_FUNCTION (new_node->decl));
4297 ipa_modify_formal_parameters (current_function_decl, adjustments, "ISRA");
4298 cfg_changed = ipa_sra_modify_function_body (adjustments);
4299 sra_ipa_reset_debug_stmts (adjustments);
4300 convert_callers (new_node, node->decl, adjustments);
4301 cgraph_make_node_local (new_node);
4302 return cfg_changed;
4305 /* Return false the function is apparently unsuitable for IPA-SRA based on it's
4306 attributes, return true otherwise. NODE is the cgraph node of the current
4307 function. */
4309 static bool
4310 ipa_sra_preliminary_function_checks (struct cgraph_node *node)
4312 if (!cgraph_node_can_be_local_p (node))
4314 if (dump_file)
4315 fprintf (dump_file, "Function not local to this compilation unit.\n");
4316 return false;
4319 if (!tree_versionable_function_p (node->decl))
4321 if (dump_file)
4322 fprintf (dump_file, "Function is not versionable.\n");
4323 return false;
4326 if (DECL_VIRTUAL_P (current_function_decl))
4328 if (dump_file)
4329 fprintf (dump_file, "Function is a virtual method.\n");
4330 return false;
4333 if ((DECL_COMDAT (node->decl) || DECL_EXTERNAL (node->decl))
4334 && node->global.size >= MAX_INLINE_INSNS_AUTO)
4336 if (dump_file)
4337 fprintf (dump_file, "Function too big to be made truly local.\n");
4338 return false;
4341 if (!node->callers)
4343 if (dump_file)
4344 fprintf (dump_file,
4345 "Function has no callers in this compilation unit.\n");
4346 return false;
4349 if (cfun->stdarg)
4351 if (dump_file)
4352 fprintf (dump_file, "Function uses stdarg. \n");
4353 return false;
4356 if (TYPE_ATTRIBUTES (TREE_TYPE (node->decl)))
4357 return false;
4359 return true;
4362 /* Perform early interprocedural SRA. */
4364 static unsigned int
4365 ipa_early_sra (void)
4367 struct cgraph_node *node = cgraph_node (current_function_decl);
4368 ipa_parm_adjustment_vec adjustments;
4369 int ret = 0;
4371 if (!ipa_sra_preliminary_function_checks (node))
4372 return 0;
4374 sra_initialize ();
4375 sra_mode = SRA_MODE_EARLY_IPA;
4377 if (!find_param_candidates ())
4379 if (dump_file)
4380 fprintf (dump_file, "Function has no IPA-SRA candidates.\n");
4381 goto simple_out;
4384 if (!all_callers_have_enough_arguments_p (node))
4386 if (dump_file)
4387 fprintf (dump_file, "There are callers with insufficient number of "
4388 "arguments.\n");
4389 goto simple_out;
4392 bb_dereferences = XCNEWVEC (HOST_WIDE_INT,
4393 func_param_count
4394 * last_basic_block_for_function (cfun));
4395 final_bbs = BITMAP_ALLOC (NULL);
4397 scan_function ();
4398 if (encountered_apply_args)
4400 if (dump_file)
4401 fprintf (dump_file, "Function calls __builtin_apply_args().\n");
4402 goto out;
4405 if (encountered_unchangable_recursive_call)
4407 if (dump_file)
4408 fprintf (dump_file, "Function calls itself with insufficient "
4409 "number of arguments.\n");
4410 goto out;
4413 adjustments = analyze_all_param_acesses ();
4414 if (!adjustments)
4415 goto out;
4416 if (dump_file)
4417 ipa_dump_param_adjustments (dump_file, adjustments, current_function_decl);
4419 if (modify_function (node, adjustments))
4420 ret = TODO_update_ssa | TODO_cleanup_cfg;
4421 else
4422 ret = TODO_update_ssa;
4423 VEC_free (ipa_parm_adjustment_t, heap, adjustments);
4425 statistics_counter_event (cfun, "Unused parameters deleted",
4426 sra_stats.deleted_unused_parameters);
4427 statistics_counter_event (cfun, "Scalar parameters converted to by-value",
4428 sra_stats.scalar_by_ref_to_by_val);
4429 statistics_counter_event (cfun, "Aggregate parameters broken up",
4430 sra_stats.aggregate_params_reduced);
4431 statistics_counter_event (cfun, "Aggregate parameter components created",
4432 sra_stats.param_reductions_created);
4434 out:
4435 BITMAP_FREE (final_bbs);
4436 free (bb_dereferences);
4437 simple_out:
4438 sra_deinitialize ();
4439 return ret;
4442 /* Return if early ipa sra shall be performed. */
4443 static bool
4444 ipa_early_sra_gate (void)
4446 return flag_ipa_sra && dbg_cnt (eipa_sra);
4449 struct gimple_opt_pass pass_early_ipa_sra =
4452 GIMPLE_PASS,
4453 "eipa_sra", /* name */
4454 ipa_early_sra_gate, /* gate */
4455 ipa_early_sra, /* execute */
4456 NULL, /* sub */
4457 NULL, /* next */
4458 0, /* static_pass_number */
4459 TV_IPA_SRA, /* tv_id */
4460 0, /* properties_required */
4461 0, /* properties_provided */
4462 0, /* properties_destroyed */
4463 0, /* todo_flags_start */
4464 TODO_dump_func | TODO_dump_cgraph /* todo_flags_finish */