PR target/49824
[official-gcc.git] / gcc / tree-sra.c
blobc83f480643c0be818ef3ca0f08624d5e5877d995
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, 2011 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"
94 #include "ipa-inline.h"
96 /* Enumeration of all aggregate reductions we can do. */
97 enum sra_mode { SRA_MODE_EARLY_IPA, /* early call regularization */
98 SRA_MODE_EARLY_INTRA, /* early intraprocedural SRA */
99 SRA_MODE_INTRA }; /* late intraprocedural SRA */
101 /* Global variable describing which aggregate reduction we are performing at
102 the moment. */
103 static enum sra_mode sra_mode;
105 struct assign_link;
107 /* ACCESS represents each access to an aggregate variable (as a whole or a
108 part). It can also represent a group of accesses that refer to exactly the
109 same fragment of an aggregate (i.e. those that have exactly the same offset
110 and size). Such representatives for a single aggregate, once determined,
111 are linked in a linked list and have the group fields set.
113 Moreover, when doing intraprocedural SRA, a tree is built from those
114 representatives (by the means of first_child and next_sibling pointers), in
115 which all items in a subtree are "within" the root, i.e. their offset is
116 greater or equal to offset of the root and offset+size is smaller or equal
117 to offset+size of the root. Children of an access are sorted by offset.
119 Note that accesses to parts of vector and complex number types always
120 represented by an access to the whole complex number or a vector. It is a
121 duty of the modifying functions to replace them appropriately. */
123 struct access
125 /* Values returned by `get_ref_base_and_extent' for each component reference
126 If EXPR isn't a component reference just set `BASE = EXPR', `OFFSET = 0',
127 `SIZE = TREE_SIZE (TREE_TYPE (expr))'. */
128 HOST_WIDE_INT offset;
129 HOST_WIDE_INT size;
130 tree base;
132 /* Expression. It is context dependent so do not use it to create new
133 expressions to access the original aggregate. See PR 42154 for a
134 testcase. */
135 tree expr;
136 /* Type. */
137 tree type;
139 /* The statement this access belongs to. */
140 gimple stmt;
142 /* Next group representative for this aggregate. */
143 struct access *next_grp;
145 /* Pointer to the group representative. Pointer to itself if the struct is
146 the representative. */
147 struct access *group_representative;
149 /* If this access has any children (in terms of the definition above), this
150 points to the first one. */
151 struct access *first_child;
153 /* In intraprocedural SRA, pointer to the next sibling in the access tree as
154 described above. In IPA-SRA this is a pointer to the next access
155 belonging to the same group (having the same representative). */
156 struct access *next_sibling;
158 /* Pointers to the first and last element in the linked list of assign
159 links. */
160 struct assign_link *first_link, *last_link;
162 /* Pointer to the next access in the work queue. */
163 struct access *next_queued;
165 /* Replacement variable for this access "region." Never to be accessed
166 directly, always only by the means of get_access_replacement() and only
167 when grp_to_be_replaced flag is set. */
168 tree replacement_decl;
170 /* Is this particular access write access? */
171 unsigned write : 1;
173 /* Is this access an access to a non-addressable field? */
174 unsigned non_addressable : 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 /* Does this group contain a write access that comes from an assignment
192 statement? This flag is propagated down the access tree. */
193 unsigned grp_assignment_write : 1;
195 /* Does this group contain a read access through a scalar type? This flag is
196 not propagated in the access tree in any direction. */
197 unsigned grp_scalar_read : 1;
199 /* Does this group contain a write access through a scalar type? This flag
200 is not propagated in the access tree in any direction. */
201 unsigned grp_scalar_write : 1;
203 /* Is this access an artificial one created to scalarize some record
204 entirely? */
205 unsigned grp_total_scalarization : 1;
207 /* Other passes of the analysis use this bit to make function
208 analyze_access_subtree create scalar replacements for this group if
209 possible. */
210 unsigned grp_hint : 1;
212 /* Is the subtree rooted in this access fully covered by scalar
213 replacements? */
214 unsigned grp_covered : 1;
216 /* If set to true, this access and all below it in an access tree must not be
217 scalarized. */
218 unsigned grp_unscalarizable_region : 1;
220 /* Whether data have been written to parts of the aggregate covered by this
221 access which is not to be scalarized. This flag is propagated up in the
222 access tree. */
223 unsigned grp_unscalarized_data : 1;
225 /* Does this access and/or group contain a write access through a
226 BIT_FIELD_REF? */
227 unsigned grp_partial_lhs : 1;
229 /* Set when a scalar replacement should be created for this variable. We do
230 the decision and creation at different places because create_tmp_var
231 cannot be called from within FOR_EACH_REFERENCED_VAR. */
232 unsigned grp_to_be_replaced : 1;
234 /* Should TREE_NO_WARNING of a replacement be set? */
235 unsigned grp_no_warning : 1;
237 /* Is it possible that the group refers to data which might be (directly or
238 otherwise) modified? */
239 unsigned grp_maybe_modified : 1;
241 /* Set when this is a representative of a pointer to scalar (i.e. by
242 reference) parameter which we consider for turning into a plain scalar
243 (i.e. a by value parameter). */
244 unsigned grp_scalar_ptr : 1;
246 /* Set when we discover that this pointer is not safe to dereference in the
247 caller. */
248 unsigned grp_not_necessarilly_dereferenced : 1;
251 typedef struct access *access_p;
253 DEF_VEC_P (access_p);
254 DEF_VEC_ALLOC_P (access_p, heap);
256 /* Alloc pool for allocating access structures. */
257 static alloc_pool access_pool;
259 /* A structure linking lhs and rhs accesses from an aggregate assignment. They
260 are used to propagate subaccesses from rhs to lhs as long as they don't
261 conflict with what is already there. */
262 struct assign_link
264 struct access *lacc, *racc;
265 struct assign_link *next;
268 /* Alloc pool for allocating assign link structures. */
269 static alloc_pool link_pool;
271 /* Base (tree) -> Vector (VEC(access_p,heap) *) map. */
272 static struct pointer_map_t *base_access_vec;
274 /* Bitmap of candidates. */
275 static bitmap candidate_bitmap;
277 /* Bitmap of candidates which we should try to entirely scalarize away and
278 those which cannot be (because they are and need be used as a whole). */
279 static bitmap should_scalarize_away_bitmap, cannot_scalarize_away_bitmap;
281 /* Obstack for creation of fancy names. */
282 static struct obstack name_obstack;
284 /* Head of a linked list of accesses that need to have its subaccesses
285 propagated to their assignment counterparts. */
286 static struct access *work_queue_head;
288 /* Number of parameters of the analyzed function when doing early ipa SRA. */
289 static int func_param_count;
291 /* scan_function sets the following to true if it encounters a call to
292 __builtin_apply_args. */
293 static bool encountered_apply_args;
295 /* Set by scan_function when it finds a recursive call. */
296 static bool encountered_recursive_call;
298 /* Set by scan_function when it finds a recursive call with less actual
299 arguments than formal parameters.. */
300 static bool encountered_unchangable_recursive_call;
302 /* This is a table in which for each basic block and parameter there is a
303 distance (offset + size) in that parameter which is dereferenced and
304 accessed in that BB. */
305 static HOST_WIDE_INT *bb_dereferences;
306 /* Bitmap of BBs that can cause the function to "stop" progressing by
307 returning, throwing externally, looping infinitely or calling a function
308 which might abort etc.. */
309 static bitmap final_bbs;
311 /* Representative of no accesses at all. */
312 static struct access no_accesses_representant;
314 /* Predicate to test the special value. */
316 static inline bool
317 no_accesses_p (struct access *access)
319 return access == &no_accesses_representant;
322 /* Dump contents of ACCESS to file F in a human friendly way. If GRP is true,
323 representative fields are dumped, otherwise those which only describe the
324 individual access are. */
326 static struct
328 /* Number of processed aggregates is readily available in
329 analyze_all_variable_accesses and so is not stored here. */
331 /* Number of created scalar replacements. */
332 int replacements;
334 /* Number of times sra_modify_expr or sra_modify_assign themselves changed an
335 expression. */
336 int exprs;
338 /* Number of statements created by generate_subtree_copies. */
339 int subtree_copies;
341 /* Number of statements created by load_assign_lhs_subreplacements. */
342 int subreplacements;
344 /* Number of times sra_modify_assign has deleted a statement. */
345 int deleted;
347 /* Number of times sra_modify_assign has to deal with subaccesses of LHS and
348 RHS reparately due to type conversions or nonexistent matching
349 references. */
350 int separate_lhs_rhs_handling;
352 /* Number of parameters that were removed because they were unused. */
353 int deleted_unused_parameters;
355 /* Number of scalars passed as parameters by reference that have been
356 converted to be passed by value. */
357 int scalar_by_ref_to_by_val;
359 /* Number of aggregate parameters that were replaced by one or more of their
360 components. */
361 int aggregate_params_reduced;
363 /* Numbber of components created when splitting aggregate parameters. */
364 int param_reductions_created;
365 } sra_stats;
367 static void
368 dump_access (FILE *f, struct access *access, bool grp)
370 fprintf (f, "access { ");
371 fprintf (f, "base = (%d)'", DECL_UID (access->base));
372 print_generic_expr (f, access->base, 0);
373 fprintf (f, "', offset = " HOST_WIDE_INT_PRINT_DEC, access->offset);
374 fprintf (f, ", size = " HOST_WIDE_INT_PRINT_DEC, access->size);
375 fprintf (f, ", expr = ");
376 print_generic_expr (f, access->expr, 0);
377 fprintf (f, ", type = ");
378 print_generic_expr (f, access->type, 0);
379 if (grp)
380 fprintf (f, ", grp_read = %d, grp_write = %d, grp_assignment_read = %d, "
381 "grp_assignment_write = %d, grp_scalar_read = %d, "
382 "grp_scalar_write = %d, grp_total_scalarization = %d, "
383 "grp_hint = %d, grp_covered = %d, "
384 "grp_unscalarizable_region = %d, grp_unscalarized_data = %d, "
385 "grp_partial_lhs = %d, grp_to_be_replaced = %d, "
386 "grp_maybe_modified = %d, "
387 "grp_not_necessarilly_dereferenced = %d\n",
388 access->grp_read, access->grp_write, access->grp_assignment_read,
389 access->grp_assignment_write, access->grp_scalar_read,
390 access->grp_scalar_write, access->grp_total_scalarization,
391 access->grp_hint, access->grp_covered,
392 access->grp_unscalarizable_region, access->grp_unscalarized_data,
393 access->grp_partial_lhs, access->grp_to_be_replaced,
394 access->grp_maybe_modified,
395 access->grp_not_necessarilly_dereferenced);
396 else
397 fprintf (f, ", write = %d, grp_total_scalarization = %d, "
398 "grp_partial_lhs = %d\n",
399 access->write, access->grp_total_scalarization,
400 access->grp_partial_lhs);
403 /* Dump a subtree rooted in ACCESS to file F, indent by LEVEL. */
405 static void
406 dump_access_tree_1 (FILE *f, struct access *access, int level)
410 int i;
412 for (i = 0; i < level; i++)
413 fputs ("* ", dump_file);
415 dump_access (f, access, true);
417 if (access->first_child)
418 dump_access_tree_1 (f, access->first_child, level + 1);
420 access = access->next_sibling;
422 while (access);
425 /* Dump all access trees for a variable, given the pointer to the first root in
426 ACCESS. */
428 static void
429 dump_access_tree (FILE *f, struct access *access)
431 for (; access; access = access->next_grp)
432 dump_access_tree_1 (f, access, 0);
435 /* Return true iff ACC is non-NULL and has subaccesses. */
437 static inline bool
438 access_has_children_p (struct access *acc)
440 return acc && acc->first_child;
443 /* Return a vector of pointers to accesses for the variable given in BASE or
444 NULL if there is none. */
446 static VEC (access_p, heap) *
447 get_base_access_vector (tree base)
449 void **slot;
451 slot = pointer_map_contains (base_access_vec, base);
452 if (!slot)
453 return NULL;
454 else
455 return *(VEC (access_p, heap) **) slot;
458 /* Find an access with required OFFSET and SIZE in a subtree of accesses rooted
459 in ACCESS. Return NULL if it cannot be found. */
461 static struct access *
462 find_access_in_subtree (struct access *access, HOST_WIDE_INT offset,
463 HOST_WIDE_INT size)
465 while (access && (access->offset != offset || access->size != size))
467 struct access *child = access->first_child;
469 while (child && (child->offset + child->size <= offset))
470 child = child->next_sibling;
471 access = child;
474 return access;
477 /* Return the first group representative for DECL or NULL if none exists. */
479 static struct access *
480 get_first_repr_for_decl (tree base)
482 VEC (access_p, heap) *access_vec;
484 access_vec = get_base_access_vector (base);
485 if (!access_vec)
486 return NULL;
488 return VEC_index (access_p, access_vec, 0);
491 /* Find an access representative for the variable BASE and given OFFSET and
492 SIZE. Requires that access trees have already been built. Return NULL if
493 it cannot be found. */
495 static struct access *
496 get_var_base_offset_size_access (tree base, HOST_WIDE_INT offset,
497 HOST_WIDE_INT size)
499 struct access *access;
501 access = get_first_repr_for_decl (base);
502 while (access && (access->offset + access->size <= offset))
503 access = access->next_grp;
504 if (!access)
505 return NULL;
507 return find_access_in_subtree (access, offset, size);
510 /* Add LINK to the linked list of assign links of RACC. */
511 static void
512 add_link_to_rhs (struct access *racc, struct assign_link *link)
514 gcc_assert (link->racc == racc);
516 if (!racc->first_link)
518 gcc_assert (!racc->last_link);
519 racc->first_link = link;
521 else
522 racc->last_link->next = link;
524 racc->last_link = link;
525 link->next = NULL;
528 /* Move all link structures in their linked list in OLD_RACC to the linked list
529 in NEW_RACC. */
530 static void
531 relink_to_new_repr (struct access *new_racc, struct access *old_racc)
533 if (!old_racc->first_link)
535 gcc_assert (!old_racc->last_link);
536 return;
539 if (new_racc->first_link)
541 gcc_assert (!new_racc->last_link->next);
542 gcc_assert (!old_racc->last_link || !old_racc->last_link->next);
544 new_racc->last_link->next = old_racc->first_link;
545 new_racc->last_link = old_racc->last_link;
547 else
549 gcc_assert (!new_racc->last_link);
551 new_racc->first_link = old_racc->first_link;
552 new_racc->last_link = old_racc->last_link;
554 old_racc->first_link = old_racc->last_link = NULL;
557 /* Add ACCESS to the work queue (which is actually a stack). */
559 static void
560 add_access_to_work_queue (struct access *access)
562 if (!access->grp_queued)
564 gcc_assert (!access->next_queued);
565 access->next_queued = work_queue_head;
566 access->grp_queued = 1;
567 work_queue_head = access;
571 /* Pop an access from the work queue, and return it, assuming there is one. */
573 static struct access *
574 pop_access_from_work_queue (void)
576 struct access *access = work_queue_head;
578 work_queue_head = access->next_queued;
579 access->next_queued = NULL;
580 access->grp_queued = 0;
581 return access;
585 /* Allocate necessary structures. */
587 static void
588 sra_initialize (void)
590 candidate_bitmap = BITMAP_ALLOC (NULL);
591 should_scalarize_away_bitmap = BITMAP_ALLOC (NULL);
592 cannot_scalarize_away_bitmap = BITMAP_ALLOC (NULL);
593 gcc_obstack_init (&name_obstack);
594 access_pool = create_alloc_pool ("SRA accesses", sizeof (struct access), 16);
595 link_pool = create_alloc_pool ("SRA links", sizeof (struct assign_link), 16);
596 base_access_vec = pointer_map_create ();
597 memset (&sra_stats, 0, sizeof (sra_stats));
598 encountered_apply_args = false;
599 encountered_recursive_call = false;
600 encountered_unchangable_recursive_call = false;
603 /* Hook fed to pointer_map_traverse, deallocate stored vectors. */
605 static bool
606 delete_base_accesses (const void *key ATTRIBUTE_UNUSED, void **value,
607 void *data ATTRIBUTE_UNUSED)
609 VEC (access_p, heap) *access_vec;
610 access_vec = (VEC (access_p, heap) *) *value;
611 VEC_free (access_p, heap, access_vec);
613 return true;
616 /* Deallocate all general structures. */
618 static void
619 sra_deinitialize (void)
621 BITMAP_FREE (candidate_bitmap);
622 BITMAP_FREE (should_scalarize_away_bitmap);
623 BITMAP_FREE (cannot_scalarize_away_bitmap);
624 free_alloc_pool (access_pool);
625 free_alloc_pool (link_pool);
626 obstack_free (&name_obstack, NULL);
628 pointer_map_traverse (base_access_vec, delete_base_accesses, NULL);
629 pointer_map_destroy (base_access_vec);
632 /* Remove DECL from candidates for SRA and write REASON to the dump file if
633 there is one. */
634 static void
635 disqualify_candidate (tree decl, const char *reason)
637 bitmap_clear_bit (candidate_bitmap, DECL_UID (decl));
639 if (dump_file && (dump_flags & TDF_DETAILS))
641 fprintf (dump_file, "! Disqualifying ");
642 print_generic_expr (dump_file, decl, 0);
643 fprintf (dump_file, " - %s\n", reason);
647 /* Return true iff the type contains a field or an element which does not allow
648 scalarization. */
650 static bool
651 type_internals_preclude_sra_p (tree type, const char **msg)
653 tree fld;
654 tree et;
656 switch (TREE_CODE (type))
658 case RECORD_TYPE:
659 case UNION_TYPE:
660 case QUAL_UNION_TYPE:
661 for (fld = TYPE_FIELDS (type); fld; fld = DECL_CHAIN (fld))
662 if (TREE_CODE (fld) == FIELD_DECL)
664 tree ft = TREE_TYPE (fld);
666 if (TREE_THIS_VOLATILE (fld))
668 *msg = "volatile structure field";
669 return true;
671 if (!DECL_FIELD_OFFSET (fld))
673 *msg = "no structure field offset";
674 return true;
676 if (!DECL_SIZE (fld))
678 *msg = "zero structure field size";
679 return true;
681 if (!host_integerp (DECL_FIELD_OFFSET (fld), 1))
683 *msg = "structure field offset not fixed";
684 return true;
686 if (!host_integerp (DECL_SIZE (fld), 1))
688 *msg = "structure field size not fixed";
689 return true;
691 if (AGGREGATE_TYPE_P (ft)
692 && int_bit_position (fld) % BITS_PER_UNIT != 0)
694 *msg = "structure field is bit field";
695 return true;
698 if (AGGREGATE_TYPE_P (ft) && type_internals_preclude_sra_p (ft, msg))
699 return true;
702 return false;
704 case ARRAY_TYPE:
705 et = TREE_TYPE (type);
707 if (TYPE_VOLATILE (et))
709 *msg = "element type is volatile";
710 return true;
713 if (AGGREGATE_TYPE_P (et) && type_internals_preclude_sra_p (et, msg))
714 return true;
716 return false;
718 default:
719 return false;
723 /* If T is an SSA_NAME, return NULL if it is not a default def or return its
724 base variable if it is. Return T if it is not an SSA_NAME. */
726 static tree
727 get_ssa_base_param (tree t)
729 if (TREE_CODE (t) == SSA_NAME)
731 if (SSA_NAME_IS_DEFAULT_DEF (t))
732 return SSA_NAME_VAR (t);
733 else
734 return NULL_TREE;
736 return t;
739 /* Mark a dereference of BASE of distance DIST in a basic block tht STMT
740 belongs to, unless the BB has already been marked as a potentially
741 final. */
743 static void
744 mark_parm_dereference (tree base, HOST_WIDE_INT dist, gimple stmt)
746 basic_block bb = gimple_bb (stmt);
747 int idx, parm_index = 0;
748 tree parm;
750 if (bitmap_bit_p (final_bbs, bb->index))
751 return;
753 for (parm = DECL_ARGUMENTS (current_function_decl);
754 parm && parm != base;
755 parm = DECL_CHAIN (parm))
756 parm_index++;
758 gcc_assert (parm_index < func_param_count);
760 idx = bb->index * func_param_count + parm_index;
761 if (bb_dereferences[idx] < dist)
762 bb_dereferences[idx] = dist;
765 /* Allocate an access structure for BASE, OFFSET and SIZE, clear it, fill in
766 the three fields. Also add it to the vector of accesses corresponding to
767 the base. Finally, return the new access. */
769 static struct access *
770 create_access_1 (tree base, HOST_WIDE_INT offset, HOST_WIDE_INT size)
772 VEC (access_p, heap) *vec;
773 struct access *access;
774 void **slot;
776 access = (struct access *) pool_alloc (access_pool);
777 memset (access, 0, sizeof (struct access));
778 access->base = base;
779 access->offset = offset;
780 access->size = size;
782 slot = pointer_map_contains (base_access_vec, base);
783 if (slot)
784 vec = (VEC (access_p, heap) *) *slot;
785 else
786 vec = VEC_alloc (access_p, heap, 32);
788 VEC_safe_push (access_p, heap, vec, access);
790 *((struct VEC (access_p,heap) **)
791 pointer_map_insert (base_access_vec, base)) = vec;
793 return access;
796 /* Create and insert access for EXPR. Return created access, or NULL if it is
797 not possible. */
799 static struct access *
800 create_access (tree expr, gimple stmt, bool write)
802 struct access *access;
803 HOST_WIDE_INT offset, size, max_size;
804 tree base = expr;
805 bool ptr, unscalarizable_region = false;
807 base = get_ref_base_and_extent (expr, &offset, &size, &max_size);
809 if (sra_mode == SRA_MODE_EARLY_IPA
810 && TREE_CODE (base) == MEM_REF)
812 base = get_ssa_base_param (TREE_OPERAND (base, 0));
813 if (!base)
814 return NULL;
815 ptr = true;
817 else
818 ptr = false;
820 if (!DECL_P (base) || !bitmap_bit_p (candidate_bitmap, DECL_UID (base)))
821 return NULL;
823 if (sra_mode == SRA_MODE_EARLY_IPA)
825 if (size < 0 || size != max_size)
827 disqualify_candidate (base, "Encountered a variable sized access.");
828 return NULL;
830 if (TREE_CODE (expr) == COMPONENT_REF
831 && DECL_BIT_FIELD (TREE_OPERAND (expr, 1)))
833 disqualify_candidate (base, "Encountered a bit-field access.");
834 return NULL;
836 gcc_checking_assert ((offset % BITS_PER_UNIT) == 0);
838 if (ptr)
839 mark_parm_dereference (base, offset + size, stmt);
841 else
843 if (size != max_size)
845 size = max_size;
846 unscalarizable_region = true;
848 if (size < 0)
850 disqualify_candidate (base, "Encountered an unconstrained access.");
851 return NULL;
855 access = create_access_1 (base, offset, size);
856 access->expr = expr;
857 access->type = TREE_TYPE (expr);
858 access->write = write;
859 access->grp_unscalarizable_region = unscalarizable_region;
860 access->stmt = stmt;
862 if (TREE_CODE (expr) == COMPONENT_REF
863 && DECL_NONADDRESSABLE_P (TREE_OPERAND (expr, 1)))
864 access->non_addressable = 1;
866 return access;
870 /* Return true iff TYPE is a RECORD_TYPE with fields that are either of gimple
871 register types or (recursively) records with only these two kinds of fields.
872 It also returns false if any of these records contains a bit-field. */
874 static bool
875 type_consists_of_records_p (tree type)
877 tree fld;
879 if (TREE_CODE (type) != RECORD_TYPE)
880 return false;
882 for (fld = TYPE_FIELDS (type); fld; fld = DECL_CHAIN (fld))
883 if (TREE_CODE (fld) == FIELD_DECL)
885 tree ft = TREE_TYPE (fld);
887 if (DECL_BIT_FIELD (fld))
888 return false;
890 if (!is_gimple_reg_type (ft)
891 && !type_consists_of_records_p (ft))
892 return false;
895 return true;
898 /* Create total_scalarization accesses for all scalar type fields in DECL that
899 must be of a RECORD_TYPE conforming to type_consists_of_records_p. BASE
900 must be the top-most VAR_DECL representing the variable, OFFSET must be the
901 offset of DECL within BASE. REF must be the memory reference expression for
902 the given decl. */
904 static void
905 completely_scalarize_record (tree base, tree decl, HOST_WIDE_INT offset,
906 tree ref)
908 tree fld, decl_type = TREE_TYPE (decl);
910 for (fld = TYPE_FIELDS (decl_type); fld; fld = DECL_CHAIN (fld))
911 if (TREE_CODE (fld) == FIELD_DECL)
913 HOST_WIDE_INT pos = offset + int_bit_position (fld);
914 tree ft = TREE_TYPE (fld);
915 tree nref = build3 (COMPONENT_REF, TREE_TYPE (fld), ref, fld,
916 NULL_TREE);
918 if (is_gimple_reg_type (ft))
920 struct access *access;
921 HOST_WIDE_INT size;
923 size = tree_low_cst (DECL_SIZE (fld), 1);
924 access = create_access_1 (base, pos, size);
925 access->expr = nref;
926 access->type = ft;
927 access->grp_total_scalarization = 1;
928 /* Accesses for intraprocedural SRA can have their stmt NULL. */
930 else
931 completely_scalarize_record (base, fld, pos, nref);
935 /* Create total_scalarization accesses for all scalar type fields in VAR and
936 for VAR a a whole. VAR must be of a RECORD_TYPE conforming to
937 type_consists_of_records_p. */
939 static void
940 completely_scalarize_var (tree var)
942 HOST_WIDE_INT size = tree_low_cst (DECL_SIZE (var), 1);
943 struct access *access;
945 access = create_access_1 (var, 0, size);
946 access->expr = var;
947 access->type = TREE_TYPE (var);
948 access->grp_total_scalarization = 1;
950 completely_scalarize_record (var, var, 0, var);
953 /* Search the given tree for a declaration by skipping handled components and
954 exclude it from the candidates. */
956 static void
957 disqualify_base_of_expr (tree t, const char *reason)
959 t = get_base_address (t);
960 if (sra_mode == SRA_MODE_EARLY_IPA
961 && TREE_CODE (t) == MEM_REF)
962 t = get_ssa_base_param (TREE_OPERAND (t, 0));
964 if (t && DECL_P (t))
965 disqualify_candidate (t, reason);
968 /* Scan expression EXPR and create access structures for all accesses to
969 candidates for scalarization. Return the created access or NULL if none is
970 created. */
972 static struct access *
973 build_access_from_expr_1 (tree expr, gimple stmt, bool write)
975 struct access *ret = NULL;
976 bool partial_ref;
978 if (TREE_CODE (expr) == BIT_FIELD_REF
979 || TREE_CODE (expr) == IMAGPART_EXPR
980 || TREE_CODE (expr) == REALPART_EXPR)
982 expr = TREE_OPERAND (expr, 0);
983 partial_ref = true;
985 else
986 partial_ref = false;
988 /* We need to dive through V_C_Es in order to get the size of its parameter
989 and not the result type. Ada produces such statements. We are also
990 capable of handling the topmost V_C_E but not any of those buried in other
991 handled components. */
992 if (TREE_CODE (expr) == VIEW_CONVERT_EXPR)
993 expr = TREE_OPERAND (expr, 0);
995 if (contains_view_convert_expr_p (expr))
997 disqualify_base_of_expr (expr, "V_C_E under a different handled "
998 "component.");
999 return NULL;
1002 switch (TREE_CODE (expr))
1004 case MEM_REF:
1005 if (TREE_CODE (TREE_OPERAND (expr, 0)) != ADDR_EXPR
1006 && sra_mode != SRA_MODE_EARLY_IPA)
1007 return NULL;
1008 /* fall through */
1009 case VAR_DECL:
1010 case PARM_DECL:
1011 case RESULT_DECL:
1012 case COMPONENT_REF:
1013 case ARRAY_REF:
1014 case ARRAY_RANGE_REF:
1015 ret = create_access (expr, stmt, write);
1016 break;
1018 default:
1019 break;
1022 if (write && partial_ref && ret)
1023 ret->grp_partial_lhs = 1;
1025 return ret;
1028 /* Scan expression EXPR and create access structures for all accesses to
1029 candidates for scalarization. Return true if any access has been inserted.
1030 STMT must be the statement from which the expression is taken, WRITE must be
1031 true if the expression is a store and false otherwise. */
1033 static bool
1034 build_access_from_expr (tree expr, gimple stmt, bool write)
1036 struct access *access;
1038 access = build_access_from_expr_1 (expr, stmt, write);
1039 if (access)
1041 /* This means the aggregate is accesses as a whole in a way other than an
1042 assign statement and thus cannot be removed even if we had a scalar
1043 replacement for everything. */
1044 if (cannot_scalarize_away_bitmap)
1045 bitmap_set_bit (cannot_scalarize_away_bitmap, DECL_UID (access->base));
1046 return true;
1048 return false;
1051 /* Disqualify LHS and RHS for scalarization if STMT must end its basic block in
1052 modes in which it matters, return true iff they have been disqualified. RHS
1053 may be NULL, in that case ignore it. If we scalarize an aggregate in
1054 intra-SRA we may need to add statements after each statement. This is not
1055 possible if a statement unconditionally has to end the basic block. */
1056 static bool
1057 disqualify_ops_if_throwing_stmt (gimple stmt, tree lhs, tree rhs)
1059 if ((sra_mode == SRA_MODE_EARLY_INTRA || sra_mode == SRA_MODE_INTRA)
1060 && (stmt_can_throw_internal (stmt) || stmt_ends_bb_p (stmt)))
1062 disqualify_base_of_expr (lhs, "LHS of a throwing stmt.");
1063 if (rhs)
1064 disqualify_base_of_expr (rhs, "RHS of a throwing stmt.");
1065 return true;
1067 return false;
1070 /* Return true iff type of EXP is not sufficiently aligned. */
1072 static bool
1073 tree_non_mode_aligned_mem_p (tree exp)
1075 enum machine_mode mode = TYPE_MODE (TREE_TYPE (exp));
1076 unsigned int align;
1078 if (TREE_CODE (exp) == VIEW_CONVERT_EXPR)
1079 exp = TREE_OPERAND (exp, 0);
1081 if (TREE_CODE (exp) == SSA_NAME
1082 || TREE_CODE (exp) == MEM_REF
1083 || mode == BLKmode
1084 || is_gimple_min_invariant (exp)
1085 || !STRICT_ALIGNMENT)
1086 return false;
1088 align = get_object_alignment (exp);
1089 if (GET_MODE_ALIGNMENT (mode) > align)
1090 return true;
1092 return false;
1095 /* Scan expressions occuring in STMT, create access structures for all accesses
1096 to candidates for scalarization and remove those candidates which occur in
1097 statements or expressions that prevent them from being split apart. Return
1098 true if any access has been inserted. */
1100 static bool
1101 build_accesses_from_assign (gimple stmt)
1103 tree lhs, rhs;
1104 struct access *lacc, *racc;
1106 if (!gimple_assign_single_p (stmt))
1107 return false;
1109 lhs = gimple_assign_lhs (stmt);
1110 rhs = gimple_assign_rhs1 (stmt);
1112 if (disqualify_ops_if_throwing_stmt (stmt, lhs, rhs))
1113 return false;
1115 racc = build_access_from_expr_1 (rhs, stmt, false);
1116 lacc = build_access_from_expr_1 (lhs, stmt, true);
1118 if (lacc)
1120 lacc->grp_assignment_write = 1;
1121 lacc->grp_unscalarizable_region |= tree_non_mode_aligned_mem_p (rhs);
1124 if (racc)
1126 racc->grp_assignment_read = 1;
1127 if (should_scalarize_away_bitmap && !gimple_has_volatile_ops (stmt)
1128 && !is_gimple_reg_type (racc->type))
1129 bitmap_set_bit (should_scalarize_away_bitmap, DECL_UID (racc->base));
1130 racc->grp_unscalarizable_region |= tree_non_mode_aligned_mem_p (lhs);
1133 if (lacc && racc
1134 && (sra_mode == SRA_MODE_EARLY_INTRA || sra_mode == SRA_MODE_INTRA)
1135 && !lacc->grp_unscalarizable_region
1136 && !racc->grp_unscalarizable_region
1137 && AGGREGATE_TYPE_P (TREE_TYPE (lhs))
1138 /* FIXME: Turn the following line into an assert after PR 40058 is
1139 fixed. */
1140 && lacc->size == racc->size
1141 && useless_type_conversion_p (lacc->type, racc->type))
1143 struct assign_link *link;
1145 link = (struct assign_link *) pool_alloc (link_pool);
1146 memset (link, 0, sizeof (struct assign_link));
1148 link->lacc = lacc;
1149 link->racc = racc;
1151 add_link_to_rhs (racc, link);
1154 return lacc || racc;
1157 /* Callback of walk_stmt_load_store_addr_ops visit_addr used to determine
1158 GIMPLE_ASM operands with memory constrains which cannot be scalarized. */
1160 static bool
1161 asm_visit_addr (gimple stmt ATTRIBUTE_UNUSED, tree op,
1162 void *data ATTRIBUTE_UNUSED)
1164 op = get_base_address (op);
1165 if (op
1166 && DECL_P (op))
1167 disqualify_candidate (op, "Non-scalarizable GIMPLE_ASM operand.");
1169 return false;
1172 /* Return true iff callsite CALL has at least as many actual arguments as there
1173 are formal parameters of the function currently processed by IPA-SRA. */
1175 static inline bool
1176 callsite_has_enough_arguments_p (gimple call)
1178 return gimple_call_num_args (call) >= (unsigned) func_param_count;
1181 /* Scan function and look for interesting expressions and create access
1182 structures for them. Return true iff any access is created. */
1184 static bool
1185 scan_function (void)
1187 basic_block bb;
1188 bool ret = false;
1190 FOR_EACH_BB (bb)
1192 gimple_stmt_iterator gsi;
1193 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
1195 gimple stmt = gsi_stmt (gsi);
1196 tree t;
1197 unsigned i;
1199 if (final_bbs && stmt_can_throw_external (stmt))
1200 bitmap_set_bit (final_bbs, bb->index);
1201 switch (gimple_code (stmt))
1203 case GIMPLE_RETURN:
1204 t = gimple_return_retval (stmt);
1205 if (t != NULL_TREE)
1206 ret |= build_access_from_expr (t, stmt, false);
1207 if (final_bbs)
1208 bitmap_set_bit (final_bbs, bb->index);
1209 break;
1211 case GIMPLE_ASSIGN:
1212 ret |= build_accesses_from_assign (stmt);
1213 break;
1215 case GIMPLE_CALL:
1216 for (i = 0; i < gimple_call_num_args (stmt); i++)
1217 ret |= build_access_from_expr (gimple_call_arg (stmt, i),
1218 stmt, false);
1220 if (sra_mode == SRA_MODE_EARLY_IPA)
1222 tree dest = gimple_call_fndecl (stmt);
1223 int flags = gimple_call_flags (stmt);
1225 if (dest)
1227 if (DECL_BUILT_IN_CLASS (dest) == BUILT_IN_NORMAL
1228 && DECL_FUNCTION_CODE (dest) == BUILT_IN_APPLY_ARGS)
1229 encountered_apply_args = true;
1230 if (cgraph_get_node (dest)
1231 == cgraph_get_node (current_function_decl))
1233 encountered_recursive_call = true;
1234 if (!callsite_has_enough_arguments_p (stmt))
1235 encountered_unchangable_recursive_call = true;
1239 if (final_bbs
1240 && (flags & (ECF_CONST | ECF_PURE)) == 0)
1241 bitmap_set_bit (final_bbs, bb->index);
1244 t = gimple_call_lhs (stmt);
1245 if (t && !disqualify_ops_if_throwing_stmt (stmt, t, NULL))
1246 ret |= build_access_from_expr (t, stmt, true);
1247 break;
1249 case GIMPLE_ASM:
1250 walk_stmt_load_store_addr_ops (stmt, NULL, NULL, NULL,
1251 asm_visit_addr);
1252 if (final_bbs)
1253 bitmap_set_bit (final_bbs, bb->index);
1255 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
1257 t = TREE_VALUE (gimple_asm_input_op (stmt, i));
1258 ret |= build_access_from_expr (t, stmt, false);
1260 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
1262 t = TREE_VALUE (gimple_asm_output_op (stmt, i));
1263 ret |= build_access_from_expr (t, stmt, true);
1265 break;
1267 default:
1268 break;
1273 return ret;
1276 /* Helper of QSORT function. There are pointers to accesses in the array. An
1277 access is considered smaller than another if it has smaller offset or if the
1278 offsets are the same but is size is bigger. */
1280 static int
1281 compare_access_positions (const void *a, const void *b)
1283 const access_p *fp1 = (const access_p *) a;
1284 const access_p *fp2 = (const access_p *) b;
1285 const access_p f1 = *fp1;
1286 const access_p f2 = *fp2;
1288 if (f1->offset != f2->offset)
1289 return f1->offset < f2->offset ? -1 : 1;
1291 if (f1->size == f2->size)
1293 if (f1->type == f2->type)
1294 return 0;
1295 /* Put any non-aggregate type before any aggregate type. */
1296 else if (!is_gimple_reg_type (f1->type)
1297 && is_gimple_reg_type (f2->type))
1298 return 1;
1299 else if (is_gimple_reg_type (f1->type)
1300 && !is_gimple_reg_type (f2->type))
1301 return -1;
1302 /* Put any complex or vector type before any other scalar type. */
1303 else if (TREE_CODE (f1->type) != COMPLEX_TYPE
1304 && TREE_CODE (f1->type) != VECTOR_TYPE
1305 && (TREE_CODE (f2->type) == COMPLEX_TYPE
1306 || TREE_CODE (f2->type) == VECTOR_TYPE))
1307 return 1;
1308 else if ((TREE_CODE (f1->type) == COMPLEX_TYPE
1309 || TREE_CODE (f1->type) == VECTOR_TYPE)
1310 && TREE_CODE (f2->type) != COMPLEX_TYPE
1311 && TREE_CODE (f2->type) != VECTOR_TYPE)
1312 return -1;
1313 /* Put the integral type with the bigger precision first. */
1314 else if (INTEGRAL_TYPE_P (f1->type)
1315 && INTEGRAL_TYPE_P (f2->type))
1316 return TYPE_PRECISION (f2->type) - TYPE_PRECISION (f1->type);
1317 /* Put any integral type with non-full precision last. */
1318 else if (INTEGRAL_TYPE_P (f1->type)
1319 && (TREE_INT_CST_LOW (TYPE_SIZE (f1->type))
1320 != TYPE_PRECISION (f1->type)))
1321 return 1;
1322 else if (INTEGRAL_TYPE_P (f2->type)
1323 && (TREE_INT_CST_LOW (TYPE_SIZE (f2->type))
1324 != TYPE_PRECISION (f2->type)))
1325 return -1;
1326 /* Stabilize the sort. */
1327 return TYPE_UID (f1->type) - TYPE_UID (f2->type);
1330 /* We want the bigger accesses first, thus the opposite operator in the next
1331 line: */
1332 return f1->size > f2->size ? -1 : 1;
1336 /* Append a name of the declaration to the name obstack. A helper function for
1337 make_fancy_name. */
1339 static void
1340 make_fancy_decl_name (tree decl)
1342 char buffer[32];
1344 tree name = DECL_NAME (decl);
1345 if (name)
1346 obstack_grow (&name_obstack, IDENTIFIER_POINTER (name),
1347 IDENTIFIER_LENGTH (name));
1348 else
1350 sprintf (buffer, "D%u", DECL_UID (decl));
1351 obstack_grow (&name_obstack, buffer, strlen (buffer));
1355 /* Helper for make_fancy_name. */
1357 static void
1358 make_fancy_name_1 (tree expr)
1360 char buffer[32];
1361 tree index;
1363 if (DECL_P (expr))
1365 make_fancy_decl_name (expr);
1366 return;
1369 switch (TREE_CODE (expr))
1371 case COMPONENT_REF:
1372 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1373 obstack_1grow (&name_obstack, '$');
1374 make_fancy_decl_name (TREE_OPERAND (expr, 1));
1375 break;
1377 case ARRAY_REF:
1378 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1379 obstack_1grow (&name_obstack, '$');
1380 /* Arrays with only one element may not have a constant as their
1381 index. */
1382 index = TREE_OPERAND (expr, 1);
1383 if (TREE_CODE (index) != INTEGER_CST)
1384 break;
1385 sprintf (buffer, HOST_WIDE_INT_PRINT_DEC, TREE_INT_CST_LOW (index));
1386 obstack_grow (&name_obstack, buffer, strlen (buffer));
1387 break;
1389 case ADDR_EXPR:
1390 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1391 break;
1393 case MEM_REF:
1394 make_fancy_name_1 (TREE_OPERAND (expr, 0));
1395 if (!integer_zerop (TREE_OPERAND (expr, 1)))
1397 obstack_1grow (&name_obstack, '$');
1398 sprintf (buffer, HOST_WIDE_INT_PRINT_DEC,
1399 TREE_INT_CST_LOW (TREE_OPERAND (expr, 1)));
1400 obstack_grow (&name_obstack, buffer, strlen (buffer));
1402 break;
1404 case BIT_FIELD_REF:
1405 case REALPART_EXPR:
1406 case IMAGPART_EXPR:
1407 gcc_unreachable (); /* we treat these as scalars. */
1408 break;
1409 default:
1410 break;
1414 /* Create a human readable name for replacement variable of ACCESS. */
1416 static char *
1417 make_fancy_name (tree expr)
1419 make_fancy_name_1 (expr);
1420 obstack_1grow (&name_obstack, '\0');
1421 return XOBFINISH (&name_obstack, char *);
1424 /* Construct a MEM_REF that would reference a part of aggregate BASE of type
1425 EXP_TYPE at the given OFFSET. If BASE is something for which
1426 get_addr_base_and_unit_offset returns NULL, gsi must be non-NULL and is used
1427 to insert new statements either before or below the current one as specified
1428 by INSERT_AFTER. This function is not capable of handling bitfields. */
1430 tree
1431 build_ref_for_offset (location_t loc, tree base, HOST_WIDE_INT offset,
1432 tree exp_type, gimple_stmt_iterator *gsi,
1433 bool insert_after)
1435 tree prev_base = base;
1436 tree off;
1437 HOST_WIDE_INT base_offset;
1439 gcc_checking_assert (offset % BITS_PER_UNIT == 0);
1441 base = get_addr_base_and_unit_offset (base, &base_offset);
1443 /* get_addr_base_and_unit_offset returns NULL for references with a variable
1444 offset such as array[var_index]. */
1445 if (!base)
1447 gimple stmt;
1448 tree tmp, addr;
1450 gcc_checking_assert (gsi);
1451 tmp = create_tmp_reg (build_pointer_type (TREE_TYPE (prev_base)), NULL);
1452 add_referenced_var (tmp);
1453 tmp = make_ssa_name (tmp, NULL);
1454 addr = build_fold_addr_expr (unshare_expr (prev_base));
1455 STRIP_USELESS_TYPE_CONVERSION (addr);
1456 stmt = gimple_build_assign (tmp, addr);
1457 gimple_set_location (stmt, loc);
1458 SSA_NAME_DEF_STMT (tmp) = stmt;
1459 if (insert_after)
1460 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
1461 else
1462 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
1463 update_stmt (stmt);
1465 off = build_int_cst (reference_alias_ptr_type (prev_base),
1466 offset / BITS_PER_UNIT);
1467 base = tmp;
1469 else if (TREE_CODE (base) == MEM_REF)
1471 off = build_int_cst (TREE_TYPE (TREE_OPERAND (base, 1)),
1472 base_offset + offset / BITS_PER_UNIT);
1473 off = int_const_binop (PLUS_EXPR, TREE_OPERAND (base, 1), off);
1474 base = unshare_expr (TREE_OPERAND (base, 0));
1476 else
1478 off = build_int_cst (reference_alias_ptr_type (base),
1479 base_offset + offset / BITS_PER_UNIT);
1480 base = build_fold_addr_expr (unshare_expr (base));
1483 return fold_build2_loc (loc, MEM_REF, exp_type, base, off);
1486 /* Construct a memory reference to a part of an aggregate BASE at the given
1487 OFFSET and of the same type as MODEL. In case this is a reference to a
1488 component, the function will replicate the last COMPONENT_REF of model's
1489 expr to access it. GSI and INSERT_AFTER have the same meaning as in
1490 build_ref_for_offset. */
1492 static tree
1493 build_ref_for_model (location_t loc, tree base, HOST_WIDE_INT offset,
1494 struct access *model, gimple_stmt_iterator *gsi,
1495 bool insert_after)
1497 if (TREE_CODE (model->expr) == COMPONENT_REF)
1499 tree t, exp_type, fld = TREE_OPERAND (model->expr, 1);
1500 tree cr_offset = component_ref_field_offset (model->expr);
1502 gcc_assert (cr_offset && host_integerp (cr_offset, 1));
1503 offset -= TREE_INT_CST_LOW (cr_offset) * BITS_PER_UNIT;
1504 offset -= TREE_INT_CST_LOW (DECL_FIELD_BIT_OFFSET (fld));
1505 exp_type = TREE_TYPE (TREE_OPERAND (model->expr, 0));
1506 t = build_ref_for_offset (loc, base, offset, exp_type, gsi, insert_after);
1507 return fold_build3_loc (loc, COMPONENT_REF, TREE_TYPE (fld), t, fld,
1508 TREE_OPERAND (model->expr, 2));
1510 else
1511 return build_ref_for_offset (loc, base, offset, model->type,
1512 gsi, insert_after);
1515 /* Construct a memory reference consisting of component_refs and array_refs to
1516 a part of an aggregate *RES (which is of type TYPE). The requested part
1517 should have type EXP_TYPE at be the given OFFSET. This function might not
1518 succeed, it returns true when it does and only then *RES points to something
1519 meaningful. This function should be used only to build expressions that we
1520 might need to present to user (e.g. in warnings). In all other situations,
1521 build_ref_for_model or build_ref_for_offset should be used instead. */
1523 static bool
1524 build_user_friendly_ref_for_offset (tree *res, tree type, HOST_WIDE_INT offset,
1525 tree exp_type)
1527 while (1)
1529 tree fld;
1530 tree tr_size, index, minidx;
1531 HOST_WIDE_INT el_size;
1533 if (offset == 0 && exp_type
1534 && types_compatible_p (exp_type, type))
1535 return true;
1537 switch (TREE_CODE (type))
1539 case UNION_TYPE:
1540 case QUAL_UNION_TYPE:
1541 case RECORD_TYPE:
1542 for (fld = TYPE_FIELDS (type); fld; fld = DECL_CHAIN (fld))
1544 HOST_WIDE_INT pos, size;
1545 tree expr, *expr_ptr;
1547 if (TREE_CODE (fld) != FIELD_DECL)
1548 continue;
1550 pos = int_bit_position (fld);
1551 gcc_assert (TREE_CODE (type) == RECORD_TYPE || pos == 0);
1552 tr_size = DECL_SIZE (fld);
1553 if (!tr_size || !host_integerp (tr_size, 1))
1554 continue;
1555 size = tree_low_cst (tr_size, 1);
1556 if (size == 0)
1558 if (pos != offset)
1559 continue;
1561 else if (pos > offset || (pos + size) <= offset)
1562 continue;
1564 expr = build3 (COMPONENT_REF, TREE_TYPE (fld), *res, fld,
1565 NULL_TREE);
1566 expr_ptr = &expr;
1567 if (build_user_friendly_ref_for_offset (expr_ptr, TREE_TYPE (fld),
1568 offset - pos, exp_type))
1570 *res = expr;
1571 return true;
1574 return false;
1576 case ARRAY_TYPE:
1577 tr_size = TYPE_SIZE (TREE_TYPE (type));
1578 if (!tr_size || !host_integerp (tr_size, 1))
1579 return false;
1580 el_size = tree_low_cst (tr_size, 1);
1582 minidx = TYPE_MIN_VALUE (TYPE_DOMAIN (type));
1583 if (TREE_CODE (minidx) != INTEGER_CST || el_size == 0)
1584 return false;
1585 index = build_int_cst (TYPE_DOMAIN (type), offset / el_size);
1586 if (!integer_zerop (minidx))
1587 index = int_const_binop (PLUS_EXPR, index, minidx);
1588 *res = build4 (ARRAY_REF, TREE_TYPE (type), *res, index,
1589 NULL_TREE, NULL_TREE);
1590 offset = offset % el_size;
1591 type = TREE_TYPE (type);
1592 break;
1594 default:
1595 if (offset != 0)
1596 return false;
1598 if (exp_type)
1599 return false;
1600 else
1601 return true;
1606 /* Return true iff TYPE is stdarg va_list type. */
1608 static inline bool
1609 is_va_list_type (tree type)
1611 return TYPE_MAIN_VARIANT (type) == TYPE_MAIN_VARIANT (va_list_type_node);
1614 /* Print message to dump file why a variable was rejected. */
1616 static void
1617 reject (tree var, const char *msg)
1619 if (dump_file && (dump_flags & TDF_DETAILS))
1621 fprintf (dump_file, "Rejected (%d): %s: ", DECL_UID (var), msg);
1622 print_generic_expr (dump_file, var, 0);
1623 fprintf (dump_file, "\n");
1627 /* The very first phase of intraprocedural SRA. It marks in candidate_bitmap
1628 those with type which is suitable for scalarization. */
1630 static bool
1631 find_var_candidates (void)
1633 tree var, type;
1634 referenced_var_iterator rvi;
1635 bool ret = false;
1636 const char *msg;
1638 FOR_EACH_REFERENCED_VAR (cfun, var, rvi)
1640 if (TREE_CODE (var) != VAR_DECL && TREE_CODE (var) != PARM_DECL)
1641 continue;
1642 type = TREE_TYPE (var);
1644 if (!AGGREGATE_TYPE_P (type))
1646 reject (var, "not aggregate");
1647 continue;
1649 if (needs_to_live_in_memory (var))
1651 reject (var, "needs to live in memory");
1652 continue;
1654 if (TREE_THIS_VOLATILE (var))
1656 reject (var, "is volatile");
1657 continue;
1659 if (!COMPLETE_TYPE_P (type))
1661 reject (var, "has incomplete type");
1662 continue;
1664 if (!host_integerp (TYPE_SIZE (type), 1))
1666 reject (var, "type size not fixed");
1667 continue;
1669 if (tree_low_cst (TYPE_SIZE (type), 1) == 0)
1671 reject (var, "type size is zero");
1672 continue;
1674 if (type_internals_preclude_sra_p (type, &msg))
1676 reject (var, msg);
1677 continue;
1679 if (/* Fix for PR 41089. tree-stdarg.c needs to have va_lists intact but
1680 we also want to schedule it rather late. Thus we ignore it in
1681 the early pass. */
1682 (sra_mode == SRA_MODE_EARLY_INTRA
1683 && is_va_list_type (type)))
1685 reject (var, "is va_list");
1686 continue;
1689 bitmap_set_bit (candidate_bitmap, DECL_UID (var));
1691 if (dump_file && (dump_flags & TDF_DETAILS))
1693 fprintf (dump_file, "Candidate (%d): ", DECL_UID (var));
1694 print_generic_expr (dump_file, var, 0);
1695 fprintf (dump_file, "\n");
1697 ret = true;
1700 return ret;
1703 /* Sort all accesses for the given variable, check for partial overlaps and
1704 return NULL if there are any. If there are none, pick a representative for
1705 each combination of offset and size and create a linked list out of them.
1706 Return the pointer to the first representative and make sure it is the first
1707 one in the vector of accesses. */
1709 static struct access *
1710 sort_and_splice_var_accesses (tree var)
1712 int i, j, access_count;
1713 struct access *res, **prev_acc_ptr = &res;
1714 VEC (access_p, heap) *access_vec;
1715 bool first = true;
1716 HOST_WIDE_INT low = -1, high = 0;
1718 access_vec = get_base_access_vector (var);
1719 if (!access_vec)
1720 return NULL;
1721 access_count = VEC_length (access_p, access_vec);
1723 /* Sort by <OFFSET, SIZE>. */
1724 VEC_qsort (access_p, access_vec, compare_access_positions);
1726 i = 0;
1727 while (i < access_count)
1729 struct access *access = VEC_index (access_p, access_vec, i);
1730 bool grp_write = access->write;
1731 bool grp_read = !access->write;
1732 bool grp_scalar_write = access->write
1733 && is_gimple_reg_type (access->type);
1734 bool grp_scalar_read = !access->write
1735 && is_gimple_reg_type (access->type);
1736 bool grp_assignment_read = access->grp_assignment_read;
1737 bool grp_assignment_write = access->grp_assignment_write;
1738 bool multiple_scalar_reads = false;
1739 bool total_scalarization = access->grp_total_scalarization;
1740 bool grp_partial_lhs = access->grp_partial_lhs;
1741 bool first_scalar = is_gimple_reg_type (access->type);
1742 bool unscalarizable_region = access->grp_unscalarizable_region;
1744 if (first || access->offset >= high)
1746 first = false;
1747 low = access->offset;
1748 high = access->offset + access->size;
1750 else if (access->offset > low && access->offset + access->size > high)
1751 return NULL;
1752 else
1753 gcc_assert (access->offset >= low
1754 && access->offset + access->size <= high);
1756 j = i + 1;
1757 while (j < access_count)
1759 struct access *ac2 = VEC_index (access_p, access_vec, j);
1760 if (ac2->offset != access->offset || ac2->size != access->size)
1761 break;
1762 if (ac2->write)
1764 grp_write = true;
1765 grp_scalar_write = (grp_scalar_write
1766 || is_gimple_reg_type (ac2->type));
1768 else
1770 grp_read = true;
1771 if (is_gimple_reg_type (ac2->type))
1773 if (grp_scalar_read)
1774 multiple_scalar_reads = true;
1775 else
1776 grp_scalar_read = true;
1779 grp_assignment_read |= ac2->grp_assignment_read;
1780 grp_assignment_write |= ac2->grp_assignment_write;
1781 grp_partial_lhs |= ac2->grp_partial_lhs;
1782 unscalarizable_region |= ac2->grp_unscalarizable_region;
1783 total_scalarization |= ac2->grp_total_scalarization;
1784 relink_to_new_repr (access, ac2);
1786 /* If there are both aggregate-type and scalar-type accesses with
1787 this combination of size and offset, the comparison function
1788 should have put the scalars first. */
1789 gcc_assert (first_scalar || !is_gimple_reg_type (ac2->type));
1790 ac2->group_representative = access;
1791 j++;
1794 i = j;
1796 access->group_representative = access;
1797 access->grp_write = grp_write;
1798 access->grp_read = grp_read;
1799 access->grp_scalar_read = grp_scalar_read;
1800 access->grp_scalar_write = grp_scalar_write;
1801 access->grp_assignment_read = grp_assignment_read;
1802 access->grp_assignment_write = grp_assignment_write;
1803 access->grp_hint = multiple_scalar_reads || total_scalarization;
1804 access->grp_total_scalarization = total_scalarization;
1805 access->grp_partial_lhs = grp_partial_lhs;
1806 access->grp_unscalarizable_region = unscalarizable_region;
1807 if (access->first_link)
1808 add_access_to_work_queue (access);
1810 *prev_acc_ptr = access;
1811 prev_acc_ptr = &access->next_grp;
1814 gcc_assert (res == VEC_index (access_p, access_vec, 0));
1815 return res;
1818 /* Create a variable for the given ACCESS which determines the type, name and a
1819 few other properties. Return the variable declaration and store it also to
1820 ACCESS->replacement. */
1822 static tree
1823 create_access_replacement (struct access *access, bool rename)
1825 tree repl;
1827 repl = create_tmp_var (access->type, "SR");
1828 add_referenced_var (repl);
1829 if (rename)
1830 mark_sym_for_renaming (repl);
1832 if (!access->grp_partial_lhs
1833 && (TREE_CODE (access->type) == COMPLEX_TYPE
1834 || TREE_CODE (access->type) == VECTOR_TYPE))
1835 DECL_GIMPLE_REG_P (repl) = 1;
1837 DECL_SOURCE_LOCATION (repl) = DECL_SOURCE_LOCATION (access->base);
1838 DECL_ARTIFICIAL (repl) = 1;
1839 DECL_IGNORED_P (repl) = DECL_IGNORED_P (access->base);
1841 if (DECL_NAME (access->base)
1842 && !DECL_IGNORED_P (access->base)
1843 && !DECL_ARTIFICIAL (access->base))
1845 char *pretty_name = make_fancy_name (access->expr);
1846 tree debug_expr = unshare_expr (access->expr), d;
1848 DECL_NAME (repl) = get_identifier (pretty_name);
1849 obstack_free (&name_obstack, pretty_name);
1851 /* Get rid of any SSA_NAMEs embedded in debug_expr,
1852 as DECL_DEBUG_EXPR isn't considered when looking for still
1853 used SSA_NAMEs and thus they could be freed. All debug info
1854 generation cares is whether something is constant or variable
1855 and that get_ref_base_and_extent works properly on the
1856 expression. */
1857 for (d = debug_expr; handled_component_p (d); d = TREE_OPERAND (d, 0))
1858 switch (TREE_CODE (d))
1860 case ARRAY_REF:
1861 case ARRAY_RANGE_REF:
1862 if (TREE_OPERAND (d, 1)
1863 && TREE_CODE (TREE_OPERAND (d, 1)) == SSA_NAME)
1864 TREE_OPERAND (d, 1) = SSA_NAME_VAR (TREE_OPERAND (d, 1));
1865 if (TREE_OPERAND (d, 3)
1866 && TREE_CODE (TREE_OPERAND (d, 3)) == SSA_NAME)
1867 TREE_OPERAND (d, 3) = SSA_NAME_VAR (TREE_OPERAND (d, 3));
1868 /* FALLTHRU */
1869 case COMPONENT_REF:
1870 if (TREE_OPERAND (d, 2)
1871 && TREE_CODE (TREE_OPERAND (d, 2)) == SSA_NAME)
1872 TREE_OPERAND (d, 2) = SSA_NAME_VAR (TREE_OPERAND (d, 2));
1873 break;
1874 default:
1875 break;
1877 SET_DECL_DEBUG_EXPR (repl, debug_expr);
1878 DECL_DEBUG_EXPR_IS_FROM (repl) = 1;
1879 if (access->grp_no_warning)
1880 TREE_NO_WARNING (repl) = 1;
1881 else
1882 TREE_NO_WARNING (repl) = TREE_NO_WARNING (access->base);
1884 else
1885 TREE_NO_WARNING (repl) = 1;
1887 if (dump_file)
1889 fprintf (dump_file, "Created a replacement for ");
1890 print_generic_expr (dump_file, access->base, 0);
1891 fprintf (dump_file, " offset: %u, size: %u: ",
1892 (unsigned) access->offset, (unsigned) access->size);
1893 print_generic_expr (dump_file, repl, 0);
1894 fprintf (dump_file, "\n");
1896 sra_stats.replacements++;
1898 return repl;
1901 /* Return ACCESS scalar replacement, create it if it does not exist yet. */
1903 static inline tree
1904 get_access_replacement (struct access *access)
1906 gcc_assert (access->grp_to_be_replaced);
1908 if (!access->replacement_decl)
1909 access->replacement_decl = create_access_replacement (access, true);
1910 return access->replacement_decl;
1913 /* Return ACCESS scalar replacement, create it if it does not exist yet but do
1914 not mark it for renaming. */
1916 static inline tree
1917 get_unrenamed_access_replacement (struct access *access)
1919 gcc_assert (!access->grp_to_be_replaced);
1921 if (!access->replacement_decl)
1922 access->replacement_decl = create_access_replacement (access, false);
1923 return access->replacement_decl;
1927 /* Build a subtree of accesses rooted in *ACCESS, and move the pointer in the
1928 linked list along the way. Stop when *ACCESS is NULL or the access pointed
1929 to it is not "within" the root. Return false iff some accesses partially
1930 overlap. */
1932 static bool
1933 build_access_subtree (struct access **access)
1935 struct access *root = *access, *last_child = NULL;
1936 HOST_WIDE_INT limit = root->offset + root->size;
1938 *access = (*access)->next_grp;
1939 while (*access && (*access)->offset + (*access)->size <= limit)
1941 if (!last_child)
1942 root->first_child = *access;
1943 else
1944 last_child->next_sibling = *access;
1945 last_child = *access;
1947 if (!build_access_subtree (access))
1948 return false;
1951 if (*access && (*access)->offset < limit)
1952 return false;
1954 return true;
1957 /* Build a tree of access representatives, ACCESS is the pointer to the first
1958 one, others are linked in a list by the next_grp field. Return false iff
1959 some accesses partially overlap. */
1961 static bool
1962 build_access_trees (struct access *access)
1964 while (access)
1966 struct access *root = access;
1968 if (!build_access_subtree (&access))
1969 return false;
1970 root->next_grp = access;
1972 return true;
1975 /* Return true if expr contains some ARRAY_REFs into a variable bounded
1976 array. */
1978 static bool
1979 expr_with_var_bounded_array_refs_p (tree expr)
1981 while (handled_component_p (expr))
1983 if (TREE_CODE (expr) == ARRAY_REF
1984 && !host_integerp (array_ref_low_bound (expr), 0))
1985 return true;
1986 expr = TREE_OPERAND (expr, 0);
1988 return false;
1991 /* Analyze the subtree of accesses rooted in ROOT, scheduling replacements when
1992 both seeming beneficial and when ALLOW_REPLACEMENTS allows it. Also set all
1993 sorts of access flags appropriately along the way, notably always set
1994 grp_read and grp_assign_read according to MARK_READ and grp_write when
1995 MARK_WRITE is true.
1997 Creating a replacement for a scalar access is considered beneficial if its
1998 grp_hint is set (this means we are either attempting total scalarization or
1999 there is more than one direct read access) or according to the following
2000 table:
2002 Access written to through a scalar type (once or more times)
2004 | Written to in an assignment statement
2006 | | Access read as scalar _once_
2007 | | |
2008 | | | Read in an assignment statement
2009 | | | |
2010 | | | | Scalarize Comment
2011 -----------------------------------------------------------------------------
2012 0 0 0 0 No access for the scalar
2013 0 0 0 1 No access for the scalar
2014 0 0 1 0 No Single read - won't help
2015 0 0 1 1 No The same case
2016 0 1 0 0 No access for the scalar
2017 0 1 0 1 No access for the scalar
2018 0 1 1 0 Yes s = *g; return s.i;
2019 0 1 1 1 Yes The same case as above
2020 1 0 0 0 No Won't help
2021 1 0 0 1 Yes s.i = 1; *g = s;
2022 1 0 1 0 Yes s.i = 5; g = s.i;
2023 1 0 1 1 Yes The same case as above
2024 1 1 0 0 No Won't help.
2025 1 1 0 1 Yes s.i = 1; *g = s;
2026 1 1 1 0 Yes s = *g; return s.i;
2027 1 1 1 1 Yes Any of the above yeses */
2029 static bool
2030 analyze_access_subtree (struct access *root, struct access *parent,
2031 bool allow_replacements)
2033 struct access *child;
2034 HOST_WIDE_INT limit = root->offset + root->size;
2035 HOST_WIDE_INT covered_to = root->offset;
2036 bool scalar = is_gimple_reg_type (root->type);
2037 bool hole = false, sth_created = false;
2039 if (parent)
2041 if (parent->grp_read)
2042 root->grp_read = 1;
2043 if (parent->grp_assignment_read)
2044 root->grp_assignment_read = 1;
2045 if (parent->grp_write)
2046 root->grp_write = 1;
2047 if (parent->grp_assignment_write)
2048 root->grp_assignment_write = 1;
2049 if (parent->grp_total_scalarization)
2050 root->grp_total_scalarization = 1;
2053 if (root->grp_unscalarizable_region)
2054 allow_replacements = false;
2056 if (allow_replacements && expr_with_var_bounded_array_refs_p (root->expr))
2057 allow_replacements = false;
2059 for (child = root->first_child; child; child = child->next_sibling)
2061 hole |= covered_to < child->offset;
2062 sth_created |= analyze_access_subtree (child, root,
2063 allow_replacements && !scalar);
2065 root->grp_unscalarized_data |= child->grp_unscalarized_data;
2066 root->grp_total_scalarization &= child->grp_total_scalarization;
2067 if (child->grp_covered)
2068 covered_to += child->size;
2069 else
2070 hole = true;
2073 if (allow_replacements && scalar && !root->first_child
2074 && (root->grp_hint
2075 || ((root->grp_scalar_read || root->grp_assignment_read)
2076 && (root->grp_scalar_write || root->grp_assignment_write))))
2078 bool new_integer_type;
2079 if (TREE_CODE (root->type) == ENUMERAL_TYPE)
2081 tree rt = root->type;
2082 root->type = build_nonstandard_integer_type (TYPE_PRECISION (rt),
2083 TYPE_UNSIGNED (rt));
2084 new_integer_type = true;
2086 else
2087 new_integer_type = false;
2089 if (dump_file && (dump_flags & TDF_DETAILS))
2091 fprintf (dump_file, "Marking ");
2092 print_generic_expr (dump_file, root->base, 0);
2093 fprintf (dump_file, " offset: %u, size: %u ",
2094 (unsigned) root->offset, (unsigned) root->size);
2095 fprintf (dump_file, " to be replaced%s.\n",
2096 new_integer_type ? " with an integer": "");
2099 root->grp_to_be_replaced = 1;
2100 sth_created = true;
2101 hole = false;
2103 else
2105 if (covered_to < limit)
2106 hole = true;
2107 if (scalar)
2108 root->grp_total_scalarization = 0;
2111 if (sth_created
2112 && (!hole || root->grp_total_scalarization))
2114 root->grp_covered = 1;
2115 return true;
2117 if (root->grp_write || TREE_CODE (root->base) == PARM_DECL)
2118 root->grp_unscalarized_data = 1; /* not covered and written to */
2119 if (sth_created)
2120 return true;
2121 return false;
2124 /* Analyze all access trees linked by next_grp by the means of
2125 analyze_access_subtree. */
2126 static bool
2127 analyze_access_trees (struct access *access)
2129 bool ret = false;
2131 while (access)
2133 if (analyze_access_subtree (access, NULL, true))
2134 ret = true;
2135 access = access->next_grp;
2138 return ret;
2141 /* Return true iff a potential new child of LACC at offset OFFSET and with size
2142 SIZE would conflict with an already existing one. If exactly such a child
2143 already exists in LACC, store a pointer to it in EXACT_MATCH. */
2145 static bool
2146 child_would_conflict_in_lacc (struct access *lacc, HOST_WIDE_INT norm_offset,
2147 HOST_WIDE_INT size, struct access **exact_match)
2149 struct access *child;
2151 for (child = lacc->first_child; child; child = child->next_sibling)
2153 if (child->offset == norm_offset && child->size == size)
2155 *exact_match = child;
2156 return true;
2159 if (child->offset < norm_offset + size
2160 && child->offset + child->size > norm_offset)
2161 return true;
2164 return false;
2167 /* Create a new child access of PARENT, with all properties just like MODEL
2168 except for its offset and with its grp_write false and grp_read true.
2169 Return the new access or NULL if it cannot be created. Note that this access
2170 is created long after all splicing and sorting, it's not located in any
2171 access vector and is automatically a representative of its group. */
2173 static struct access *
2174 create_artificial_child_access (struct access *parent, struct access *model,
2175 HOST_WIDE_INT new_offset)
2177 struct access *access;
2178 struct access **child;
2179 tree expr = parent->base;
2181 gcc_assert (!model->grp_unscalarizable_region);
2183 access = (struct access *) pool_alloc (access_pool);
2184 memset (access, 0, sizeof (struct access));
2185 if (!build_user_friendly_ref_for_offset (&expr, TREE_TYPE (expr), new_offset,
2186 model->type))
2188 access->grp_no_warning = true;
2189 expr = build_ref_for_model (EXPR_LOCATION (parent->base), parent->base,
2190 new_offset, model, NULL, false);
2193 access->base = parent->base;
2194 access->expr = expr;
2195 access->offset = new_offset;
2196 access->size = model->size;
2197 access->type = model->type;
2198 access->grp_write = true;
2199 access->grp_read = false;
2201 child = &parent->first_child;
2202 while (*child && (*child)->offset < new_offset)
2203 child = &(*child)->next_sibling;
2205 access->next_sibling = *child;
2206 *child = access;
2208 return access;
2212 /* Propagate all subaccesses of RACC across an assignment link to LACC. Return
2213 true if any new subaccess was created. Additionally, if RACC is a scalar
2214 access but LACC is not, change the type of the latter, if possible. */
2216 static bool
2217 propagate_subaccesses_across_link (struct access *lacc, struct access *racc)
2219 struct access *rchild;
2220 HOST_WIDE_INT norm_delta = lacc->offset - racc->offset;
2221 bool ret = false;
2223 if (is_gimple_reg_type (lacc->type)
2224 || lacc->grp_unscalarizable_region
2225 || racc->grp_unscalarizable_region)
2226 return false;
2228 if (!lacc->first_child && !racc->first_child
2229 && is_gimple_reg_type (racc->type))
2231 tree t = lacc->base;
2233 lacc->type = racc->type;
2234 if (build_user_friendly_ref_for_offset (&t, TREE_TYPE (t), lacc->offset,
2235 racc->type))
2236 lacc->expr = t;
2237 else
2239 lacc->expr = build_ref_for_model (EXPR_LOCATION (lacc->base),
2240 lacc->base, lacc->offset,
2241 racc, NULL, false);
2242 lacc->grp_no_warning = true;
2244 return false;
2247 for (rchild = racc->first_child; rchild; rchild = rchild->next_sibling)
2249 struct access *new_acc = NULL;
2250 HOST_WIDE_INT norm_offset = rchild->offset + norm_delta;
2252 if (rchild->grp_unscalarizable_region)
2253 continue;
2255 if (child_would_conflict_in_lacc (lacc, norm_offset, rchild->size,
2256 &new_acc))
2258 if (new_acc)
2260 rchild->grp_hint = 1;
2261 new_acc->grp_hint |= new_acc->grp_read;
2262 if (rchild->first_child)
2263 ret |= propagate_subaccesses_across_link (new_acc, rchild);
2265 continue;
2268 rchild->grp_hint = 1;
2269 new_acc = create_artificial_child_access (lacc, rchild, norm_offset);
2270 if (new_acc)
2272 ret = true;
2273 if (racc->first_child)
2274 propagate_subaccesses_across_link (new_acc, rchild);
2278 return ret;
2281 /* Propagate all subaccesses across assignment links. */
2283 static void
2284 propagate_all_subaccesses (void)
2286 while (work_queue_head)
2288 struct access *racc = pop_access_from_work_queue ();
2289 struct assign_link *link;
2291 gcc_assert (racc->first_link);
2293 for (link = racc->first_link; link; link = link->next)
2295 struct access *lacc = link->lacc;
2297 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (lacc->base)))
2298 continue;
2299 lacc = lacc->group_representative;
2300 if (propagate_subaccesses_across_link (lacc, racc)
2301 && lacc->first_link)
2302 add_access_to_work_queue (lacc);
2307 /* Go through all accesses collected throughout the (intraprocedural) analysis
2308 stage, exclude overlapping ones, identify representatives and build trees
2309 out of them, making decisions about scalarization on the way. Return true
2310 iff there are any to-be-scalarized variables after this stage. */
2312 static bool
2313 analyze_all_variable_accesses (void)
2315 int res = 0;
2316 bitmap tmp = BITMAP_ALLOC (NULL);
2317 bitmap_iterator bi;
2318 unsigned i, max_total_scalarization_size;
2320 max_total_scalarization_size = UNITS_PER_WORD * BITS_PER_UNIT
2321 * MOVE_RATIO (optimize_function_for_speed_p (cfun));
2323 EXECUTE_IF_SET_IN_BITMAP (candidate_bitmap, 0, i, bi)
2324 if (bitmap_bit_p (should_scalarize_away_bitmap, i)
2325 && !bitmap_bit_p (cannot_scalarize_away_bitmap, i))
2327 tree var = referenced_var (i);
2329 if (TREE_CODE (var) == VAR_DECL
2330 && type_consists_of_records_p (TREE_TYPE (var)))
2332 if ((unsigned) tree_low_cst (TYPE_SIZE (TREE_TYPE (var)), 1)
2333 <= max_total_scalarization_size)
2335 completely_scalarize_var (var);
2336 if (dump_file && (dump_flags & TDF_DETAILS))
2338 fprintf (dump_file, "Will attempt to totally scalarize ");
2339 print_generic_expr (dump_file, var, 0);
2340 fprintf (dump_file, " (UID: %u): \n", DECL_UID (var));
2343 else if (dump_file && (dump_flags & TDF_DETAILS))
2345 fprintf (dump_file, "Too big to totally scalarize: ");
2346 print_generic_expr (dump_file, var, 0);
2347 fprintf (dump_file, " (UID: %u)\n", DECL_UID (var));
2352 bitmap_copy (tmp, candidate_bitmap);
2353 EXECUTE_IF_SET_IN_BITMAP (tmp, 0, i, bi)
2355 tree var = referenced_var (i);
2356 struct access *access;
2358 access = sort_and_splice_var_accesses (var);
2359 if (!access || !build_access_trees (access))
2360 disqualify_candidate (var,
2361 "No or inhibitingly overlapping accesses.");
2364 propagate_all_subaccesses ();
2366 bitmap_copy (tmp, candidate_bitmap);
2367 EXECUTE_IF_SET_IN_BITMAP (tmp, 0, i, bi)
2369 tree var = referenced_var (i);
2370 struct access *access = get_first_repr_for_decl (var);
2372 if (analyze_access_trees (access))
2374 res++;
2375 if (dump_file && (dump_flags & TDF_DETAILS))
2377 fprintf (dump_file, "\nAccess trees for ");
2378 print_generic_expr (dump_file, var, 0);
2379 fprintf (dump_file, " (UID: %u): \n", DECL_UID (var));
2380 dump_access_tree (dump_file, access);
2381 fprintf (dump_file, "\n");
2384 else
2385 disqualify_candidate (var, "No scalar replacements to be created.");
2388 BITMAP_FREE (tmp);
2390 if (res)
2392 statistics_counter_event (cfun, "Scalarized aggregates", res);
2393 return true;
2395 else
2396 return false;
2399 /* Generate statements copying scalar replacements of accesses within a subtree
2400 into or out of AGG. ACCESS, all its children, siblings and their children
2401 are to be processed. AGG is an aggregate type expression (can be a
2402 declaration but does not have to be, it can for example also be a mem_ref or
2403 a series of handled components). TOP_OFFSET is the offset of the processed
2404 subtree which has to be subtracted from offsets of individual accesses to
2405 get corresponding offsets for AGG. If CHUNK_SIZE is non-null, copy only
2406 replacements in the interval <start_offset, start_offset + chunk_size>,
2407 otherwise copy all. GSI is a statement iterator used to place the new
2408 statements. WRITE should be true when the statements should write from AGG
2409 to the replacement and false if vice versa. if INSERT_AFTER is true, new
2410 statements will be added after the current statement in GSI, they will be
2411 added before the statement otherwise. */
2413 static void
2414 generate_subtree_copies (struct access *access, tree agg,
2415 HOST_WIDE_INT top_offset,
2416 HOST_WIDE_INT start_offset, HOST_WIDE_INT chunk_size,
2417 gimple_stmt_iterator *gsi, bool write,
2418 bool insert_after, location_t loc)
2422 if (chunk_size && access->offset >= start_offset + chunk_size)
2423 return;
2425 if (access->grp_to_be_replaced
2426 && (chunk_size == 0
2427 || access->offset + access->size > start_offset))
2429 tree expr, repl = get_access_replacement (access);
2430 gimple stmt;
2432 expr = build_ref_for_model (loc, agg, access->offset - top_offset,
2433 access, gsi, insert_after);
2435 if (write)
2437 if (access->grp_partial_lhs)
2438 expr = force_gimple_operand_gsi (gsi, expr, true, NULL_TREE,
2439 !insert_after,
2440 insert_after ? GSI_NEW_STMT
2441 : GSI_SAME_STMT);
2442 stmt = gimple_build_assign (repl, expr);
2444 else
2446 TREE_NO_WARNING (repl) = 1;
2447 if (access->grp_partial_lhs)
2448 repl = force_gimple_operand_gsi (gsi, repl, true, NULL_TREE,
2449 !insert_after,
2450 insert_after ? GSI_NEW_STMT
2451 : GSI_SAME_STMT);
2452 stmt = gimple_build_assign (expr, repl);
2454 gimple_set_location (stmt, loc);
2456 if (insert_after)
2457 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2458 else
2459 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2460 update_stmt (stmt);
2461 sra_stats.subtree_copies++;
2464 if (access->first_child)
2465 generate_subtree_copies (access->first_child, agg, top_offset,
2466 start_offset, chunk_size, gsi,
2467 write, insert_after, loc);
2469 access = access->next_sibling;
2471 while (access);
2474 /* Assign zero to all scalar replacements in an access subtree. ACCESS is the
2475 the root of the subtree to be processed. GSI is the statement iterator used
2476 for inserting statements which are added after the current statement if
2477 INSERT_AFTER is true or before it otherwise. */
2479 static void
2480 init_subtree_with_zero (struct access *access, gimple_stmt_iterator *gsi,
2481 bool insert_after, location_t loc)
2484 struct access *child;
2486 if (access->grp_to_be_replaced)
2488 gimple stmt;
2490 stmt = gimple_build_assign (get_access_replacement (access),
2491 build_zero_cst (access->type));
2492 if (insert_after)
2493 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2494 else
2495 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2496 update_stmt (stmt);
2497 gimple_set_location (stmt, loc);
2500 for (child = access->first_child; child; child = child->next_sibling)
2501 init_subtree_with_zero (child, gsi, insert_after, loc);
2504 /* Search for an access representative for the given expression EXPR and
2505 return it or NULL if it cannot be found. */
2507 static struct access *
2508 get_access_for_expr (tree expr)
2510 HOST_WIDE_INT offset, size, max_size;
2511 tree base;
2513 /* FIXME: This should not be necessary but Ada produces V_C_Es with a type of
2514 a different size than the size of its argument and we need the latter
2515 one. */
2516 if (TREE_CODE (expr) == VIEW_CONVERT_EXPR)
2517 expr = TREE_OPERAND (expr, 0);
2519 base = get_ref_base_and_extent (expr, &offset, &size, &max_size);
2520 if (max_size == -1 || !DECL_P (base))
2521 return NULL;
2523 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (base)))
2524 return NULL;
2526 return get_var_base_offset_size_access (base, offset, max_size);
2529 /* Replace the expression EXPR with a scalar replacement if there is one and
2530 generate other statements to do type conversion or subtree copying if
2531 necessary. GSI is used to place newly created statements, WRITE is true if
2532 the expression is being written to (it is on a LHS of a statement or output
2533 in an assembly statement). */
2535 static bool
2536 sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write)
2538 location_t loc;
2539 struct access *access;
2540 tree type, bfr;
2542 if (TREE_CODE (*expr) == BIT_FIELD_REF)
2544 bfr = *expr;
2545 expr = &TREE_OPERAND (*expr, 0);
2547 else
2548 bfr = NULL_TREE;
2550 if (TREE_CODE (*expr) == REALPART_EXPR || TREE_CODE (*expr) == IMAGPART_EXPR)
2551 expr = &TREE_OPERAND (*expr, 0);
2552 access = get_access_for_expr (*expr);
2553 if (!access)
2554 return false;
2555 type = TREE_TYPE (*expr);
2557 loc = gimple_location (gsi_stmt (*gsi));
2558 if (access->grp_to_be_replaced)
2560 tree repl = get_access_replacement (access);
2561 /* If we replace a non-register typed access simply use the original
2562 access expression to extract the scalar component afterwards.
2563 This happens if scalarizing a function return value or parameter
2564 like in gcc.c-torture/execute/20041124-1.c, 20050316-1.c and
2565 gcc.c-torture/compile/20011217-1.c.
2567 We also want to use this when accessing a complex or vector which can
2568 be accessed as a different type too, potentially creating a need for
2569 type conversion (see PR42196) and when scalarized unions are involved
2570 in assembler statements (see PR42398). */
2571 if (!useless_type_conversion_p (type, access->type))
2573 tree ref;
2575 ref = build_ref_for_model (loc, access->base, access->offset, access,
2576 NULL, false);
2578 if (write)
2580 gimple stmt;
2582 if (access->grp_partial_lhs)
2583 ref = force_gimple_operand_gsi (gsi, ref, true, NULL_TREE,
2584 false, GSI_NEW_STMT);
2585 stmt = gimple_build_assign (repl, ref);
2586 gimple_set_location (stmt, loc);
2587 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2589 else
2591 gimple stmt;
2593 if (access->grp_partial_lhs)
2594 repl = force_gimple_operand_gsi (gsi, repl, true, NULL_TREE,
2595 true, GSI_SAME_STMT);
2596 stmt = gimple_build_assign (ref, repl);
2597 gimple_set_location (stmt, loc);
2598 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2601 else
2602 *expr = repl;
2603 sra_stats.exprs++;
2606 if (access->first_child)
2608 HOST_WIDE_INT start_offset, chunk_size;
2609 if (bfr
2610 && host_integerp (TREE_OPERAND (bfr, 1), 1)
2611 && host_integerp (TREE_OPERAND (bfr, 2), 1))
2613 chunk_size = tree_low_cst (TREE_OPERAND (bfr, 1), 1);
2614 start_offset = access->offset
2615 + tree_low_cst (TREE_OPERAND (bfr, 2), 1);
2617 else
2618 start_offset = chunk_size = 0;
2620 generate_subtree_copies (access->first_child, access->base, 0,
2621 start_offset, chunk_size, gsi, write, write,
2622 loc);
2624 return true;
2627 /* Where scalar replacements of the RHS have been written to when a replacement
2628 of a LHS of an assigments cannot be direclty loaded from a replacement of
2629 the RHS. */
2630 enum unscalarized_data_handling { SRA_UDH_NONE, /* Nothing done so far. */
2631 SRA_UDH_RIGHT, /* Data flushed to the RHS. */
2632 SRA_UDH_LEFT }; /* Data flushed to the LHS. */
2634 /* Store all replacements in the access tree rooted in TOP_RACC either to their
2635 base aggregate if there are unscalarized data or directly to LHS of the
2636 statement that is pointed to by GSI otherwise. */
2638 static enum unscalarized_data_handling
2639 handle_unscalarized_data_in_subtree (struct access *top_racc,
2640 gimple_stmt_iterator *gsi)
2642 if (top_racc->grp_unscalarized_data)
2644 generate_subtree_copies (top_racc->first_child, top_racc->base, 0, 0, 0,
2645 gsi, false, false,
2646 gimple_location (gsi_stmt (*gsi)));
2647 return SRA_UDH_RIGHT;
2649 else
2651 tree lhs = gimple_assign_lhs (gsi_stmt (*gsi));
2652 generate_subtree_copies (top_racc->first_child, lhs, top_racc->offset,
2653 0, 0, gsi, false, false,
2654 gimple_location (gsi_stmt (*gsi)));
2655 return SRA_UDH_LEFT;
2660 /* Try to generate statements to load all sub-replacements in an access subtree
2661 formed by children of LACC from scalar replacements in the TOP_RACC subtree.
2662 If that is not possible, refresh the TOP_RACC base aggregate and load the
2663 accesses from it. LEFT_OFFSET is the offset of the left whole subtree being
2664 copied. NEW_GSI is stmt iterator used for statement insertions after the
2665 original assignment, OLD_GSI is used to insert statements before the
2666 assignment. *REFRESHED keeps the information whether we have needed to
2667 refresh replacements of the LHS and from which side of the assignments this
2668 takes place. */
2670 static void
2671 load_assign_lhs_subreplacements (struct access *lacc, struct access *top_racc,
2672 HOST_WIDE_INT left_offset,
2673 gimple_stmt_iterator *old_gsi,
2674 gimple_stmt_iterator *new_gsi,
2675 enum unscalarized_data_handling *refreshed)
2677 location_t loc = gimple_location (gsi_stmt (*old_gsi));
2678 for (lacc = lacc->first_child; lacc; lacc = lacc->next_sibling)
2680 if (lacc->grp_to_be_replaced)
2682 struct access *racc;
2683 HOST_WIDE_INT offset = lacc->offset - left_offset + top_racc->offset;
2684 gimple stmt;
2685 tree rhs;
2687 racc = find_access_in_subtree (top_racc, offset, lacc->size);
2688 if (racc && racc->grp_to_be_replaced)
2690 rhs = get_access_replacement (racc);
2691 if (!useless_type_conversion_p (lacc->type, racc->type))
2692 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR, lacc->type, rhs);
2694 else
2696 /* No suitable access on the right hand side, need to load from
2697 the aggregate. See if we have to update it first... */
2698 if (*refreshed == SRA_UDH_NONE)
2699 *refreshed = handle_unscalarized_data_in_subtree (top_racc,
2700 old_gsi);
2702 if (*refreshed == SRA_UDH_LEFT)
2703 rhs = build_ref_for_model (loc, lacc->base, lacc->offset, lacc,
2704 new_gsi, true);
2705 else
2706 rhs = build_ref_for_model (loc, top_racc->base, offset, lacc,
2707 new_gsi, true);
2710 stmt = gimple_build_assign (get_access_replacement (lacc), rhs);
2711 gsi_insert_after (new_gsi, stmt, GSI_NEW_STMT);
2712 gimple_set_location (stmt, loc);
2713 update_stmt (stmt);
2714 sra_stats.subreplacements++;
2716 else if (*refreshed == SRA_UDH_NONE
2717 && lacc->grp_read && !lacc->grp_covered)
2718 *refreshed = handle_unscalarized_data_in_subtree (top_racc,
2719 old_gsi);
2721 if (lacc->first_child)
2722 load_assign_lhs_subreplacements (lacc, top_racc, left_offset,
2723 old_gsi, new_gsi, refreshed);
2727 /* Result code for SRA assignment modification. */
2728 enum assignment_mod_result { SRA_AM_NONE, /* nothing done for the stmt */
2729 SRA_AM_MODIFIED, /* stmt changed but not
2730 removed */
2731 SRA_AM_REMOVED }; /* stmt eliminated */
2733 /* Modify assignments with a CONSTRUCTOR on their RHS. STMT contains a pointer
2734 to the assignment and GSI is the statement iterator pointing at it. Returns
2735 the same values as sra_modify_assign. */
2737 static enum assignment_mod_result
2738 sra_modify_constructor_assign (gimple *stmt, gimple_stmt_iterator *gsi)
2740 tree lhs = gimple_assign_lhs (*stmt);
2741 struct access *acc;
2742 location_t loc;
2744 acc = get_access_for_expr (lhs);
2745 if (!acc)
2746 return SRA_AM_NONE;
2748 loc = gimple_location (*stmt);
2749 if (VEC_length (constructor_elt,
2750 CONSTRUCTOR_ELTS (gimple_assign_rhs1 (*stmt))) > 0)
2752 /* I have never seen this code path trigger but if it can happen the
2753 following should handle it gracefully. */
2754 if (access_has_children_p (acc))
2755 generate_subtree_copies (acc->first_child, acc->base, 0, 0, 0, gsi,
2756 true, true, loc);
2757 return SRA_AM_MODIFIED;
2760 if (acc->grp_covered)
2762 init_subtree_with_zero (acc, gsi, false, loc);
2763 unlink_stmt_vdef (*stmt);
2764 gsi_remove (gsi, true);
2765 return SRA_AM_REMOVED;
2767 else
2769 init_subtree_with_zero (acc, gsi, true, loc);
2770 return SRA_AM_MODIFIED;
2774 /* Create and return a new suitable default definition SSA_NAME for RACC which
2775 is an access describing an uninitialized part of an aggregate that is being
2776 loaded. */
2778 static tree
2779 get_repl_default_def_ssa_name (struct access *racc)
2781 tree repl, decl;
2783 decl = get_unrenamed_access_replacement (racc);
2785 repl = gimple_default_def (cfun, decl);
2786 if (!repl)
2788 repl = make_ssa_name (decl, gimple_build_nop ());
2789 set_default_def (decl, repl);
2792 return repl;
2795 /* Return true if REF has a COMPONENT_REF with a bit-field field declaration
2796 somewhere in it. */
2798 static inline bool
2799 contains_bitfld_comp_ref_p (const_tree ref)
2801 while (handled_component_p (ref))
2803 if (TREE_CODE (ref) == COMPONENT_REF
2804 && DECL_BIT_FIELD (TREE_OPERAND (ref, 1)))
2805 return true;
2806 ref = TREE_OPERAND (ref, 0);
2809 return false;
2812 /* Return true if REF has an VIEW_CONVERT_EXPR or a COMPONENT_REF with a
2813 bit-field field declaration somewhere in it. */
2815 static inline bool
2816 contains_vce_or_bfcref_p (const_tree ref)
2818 while (handled_component_p (ref))
2820 if (TREE_CODE (ref) == VIEW_CONVERT_EXPR
2821 || (TREE_CODE (ref) == COMPONENT_REF
2822 && DECL_BIT_FIELD (TREE_OPERAND (ref, 1))))
2823 return true;
2824 ref = TREE_OPERAND (ref, 0);
2827 return false;
2830 /* Examine both sides of the assignment statement pointed to by STMT, replace
2831 them with a scalare replacement if there is one and generate copying of
2832 replacements if scalarized aggregates have been used in the assignment. GSI
2833 is used to hold generated statements for type conversions and subtree
2834 copying. */
2836 static enum assignment_mod_result
2837 sra_modify_assign (gimple *stmt, gimple_stmt_iterator *gsi)
2839 struct access *lacc, *racc;
2840 tree lhs, rhs;
2841 bool modify_this_stmt = false;
2842 bool force_gimple_rhs = false;
2843 location_t loc;
2844 gimple_stmt_iterator orig_gsi = *gsi;
2846 if (!gimple_assign_single_p (*stmt))
2847 return SRA_AM_NONE;
2848 lhs = gimple_assign_lhs (*stmt);
2849 rhs = gimple_assign_rhs1 (*stmt);
2851 if (TREE_CODE (rhs) == CONSTRUCTOR)
2852 return sra_modify_constructor_assign (stmt, gsi);
2854 if (TREE_CODE (rhs) == REALPART_EXPR || TREE_CODE (lhs) == REALPART_EXPR
2855 || TREE_CODE (rhs) == IMAGPART_EXPR || TREE_CODE (lhs) == IMAGPART_EXPR
2856 || TREE_CODE (rhs) == BIT_FIELD_REF || TREE_CODE (lhs) == BIT_FIELD_REF)
2858 modify_this_stmt = sra_modify_expr (gimple_assign_rhs1_ptr (*stmt),
2859 gsi, false);
2860 modify_this_stmt |= sra_modify_expr (gimple_assign_lhs_ptr (*stmt),
2861 gsi, true);
2862 return modify_this_stmt ? SRA_AM_MODIFIED : SRA_AM_NONE;
2865 lacc = get_access_for_expr (lhs);
2866 racc = get_access_for_expr (rhs);
2867 if (!lacc && !racc)
2868 return SRA_AM_NONE;
2870 loc = gimple_location (*stmt);
2871 if (lacc && lacc->grp_to_be_replaced)
2873 lhs = get_access_replacement (lacc);
2874 gimple_assign_set_lhs (*stmt, lhs);
2875 modify_this_stmt = true;
2876 if (lacc->grp_partial_lhs)
2877 force_gimple_rhs = true;
2878 sra_stats.exprs++;
2881 if (racc && racc->grp_to_be_replaced)
2883 rhs = get_access_replacement (racc);
2884 modify_this_stmt = true;
2885 if (racc->grp_partial_lhs)
2886 force_gimple_rhs = true;
2887 sra_stats.exprs++;
2890 if (modify_this_stmt)
2892 if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
2894 /* If we can avoid creating a VIEW_CONVERT_EXPR do so.
2895 ??? This should move to fold_stmt which we simply should
2896 call after building a VIEW_CONVERT_EXPR here. */
2897 if (AGGREGATE_TYPE_P (TREE_TYPE (lhs))
2898 && !contains_bitfld_comp_ref_p (lhs)
2899 && !access_has_children_p (lacc))
2901 lhs = build_ref_for_model (loc, lhs, 0, racc, gsi, false);
2902 gimple_assign_set_lhs (*stmt, lhs);
2904 else if (AGGREGATE_TYPE_P (TREE_TYPE (rhs))
2905 && !contains_vce_or_bfcref_p (rhs)
2906 && !access_has_children_p (racc))
2907 rhs = build_ref_for_model (loc, rhs, 0, lacc, gsi, false);
2909 if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
2911 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR, TREE_TYPE (lhs),
2912 rhs);
2913 if (is_gimple_reg_type (TREE_TYPE (lhs))
2914 && TREE_CODE (lhs) != SSA_NAME)
2915 force_gimple_rhs = true;
2920 /* From this point on, the function deals with assignments in between
2921 aggregates when at least one has scalar reductions of some of its
2922 components. There are three possible scenarios: Both the LHS and RHS have
2923 to-be-scalarized components, 2) only the RHS has or 3) only the LHS has.
2925 In the first case, we would like to load the LHS components from RHS
2926 components whenever possible. If that is not possible, we would like to
2927 read it directly from the RHS (after updating it by storing in it its own
2928 components). If there are some necessary unscalarized data in the LHS,
2929 those will be loaded by the original assignment too. If neither of these
2930 cases happen, the original statement can be removed. Most of this is done
2931 by load_assign_lhs_subreplacements.
2933 In the second case, we would like to store all RHS scalarized components
2934 directly into LHS and if they cover the aggregate completely, remove the
2935 statement too. In the third case, we want the LHS components to be loaded
2936 directly from the RHS (DSE will remove the original statement if it
2937 becomes redundant).
2939 This is a bit complex but manageable when types match and when unions do
2940 not cause confusion in a way that we cannot really load a component of LHS
2941 from the RHS or vice versa (the access representing this level can have
2942 subaccesses that are accessible only through a different union field at a
2943 higher level - different from the one used in the examined expression).
2944 Unions are fun.
2946 Therefore, I specially handle a fourth case, happening when there is a
2947 specific type cast or it is impossible to locate a scalarized subaccess on
2948 the other side of the expression. If that happens, I simply "refresh" the
2949 RHS by storing in it is scalarized components leave the original statement
2950 there to do the copying and then load the scalar replacements of the LHS.
2951 This is what the first branch does. */
2953 if (modify_this_stmt
2954 || gimple_has_volatile_ops (*stmt)
2955 || contains_vce_or_bfcref_p (rhs)
2956 || contains_vce_or_bfcref_p (lhs))
2958 if (access_has_children_p (racc))
2959 generate_subtree_copies (racc->first_child, racc->base, 0, 0, 0,
2960 gsi, false, false, loc);
2961 if (access_has_children_p (lacc))
2962 generate_subtree_copies (lacc->first_child, lacc->base, 0, 0, 0,
2963 gsi, true, true, loc);
2964 sra_stats.separate_lhs_rhs_handling++;
2966 else
2968 if (access_has_children_p (lacc) && access_has_children_p (racc))
2970 gimple_stmt_iterator orig_gsi = *gsi;
2971 enum unscalarized_data_handling refreshed;
2973 if (lacc->grp_read && !lacc->grp_covered)
2974 refreshed = handle_unscalarized_data_in_subtree (racc, gsi);
2975 else
2976 refreshed = SRA_UDH_NONE;
2978 load_assign_lhs_subreplacements (lacc, racc, lacc->offset,
2979 &orig_gsi, gsi, &refreshed);
2980 if (refreshed != SRA_UDH_RIGHT)
2982 gsi_next (gsi);
2983 unlink_stmt_vdef (*stmt);
2984 gsi_remove (&orig_gsi, true);
2985 sra_stats.deleted++;
2986 return SRA_AM_REMOVED;
2989 else
2991 if (racc)
2993 if (!racc->grp_to_be_replaced && !racc->grp_unscalarized_data)
2995 if (dump_file)
2997 fprintf (dump_file, "Removing load: ");
2998 print_gimple_stmt (dump_file, *stmt, 0, 0);
3001 if (TREE_CODE (lhs) == SSA_NAME)
3003 rhs = get_repl_default_def_ssa_name (racc);
3004 if (!useless_type_conversion_p (TREE_TYPE (lhs),
3005 TREE_TYPE (rhs)))
3006 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR,
3007 TREE_TYPE (lhs), rhs);
3009 else
3011 if (racc->first_child)
3012 generate_subtree_copies (racc->first_child, lhs,
3013 racc->offset, 0, 0, gsi,
3014 false, false, loc);
3016 gcc_assert (*stmt == gsi_stmt (*gsi));
3017 unlink_stmt_vdef (*stmt);
3018 gsi_remove (gsi, true);
3019 sra_stats.deleted++;
3020 return SRA_AM_REMOVED;
3023 else if (racc->first_child)
3024 generate_subtree_copies (racc->first_child, lhs, racc->offset,
3025 0, 0, gsi, false, true, loc);
3027 if (access_has_children_p (lacc))
3028 generate_subtree_copies (lacc->first_child, rhs, lacc->offset,
3029 0, 0, gsi, true, true, loc);
3033 /* This gimplification must be done after generate_subtree_copies, lest we
3034 insert the subtree copies in the middle of the gimplified sequence. */
3035 if (force_gimple_rhs)
3036 rhs = force_gimple_operand_gsi (&orig_gsi, rhs, true, NULL_TREE,
3037 true, GSI_SAME_STMT);
3038 if (gimple_assign_rhs1 (*stmt) != rhs)
3040 modify_this_stmt = true;
3041 gimple_assign_set_rhs_from_tree (&orig_gsi, rhs);
3042 gcc_assert (*stmt == gsi_stmt (orig_gsi));
3045 return modify_this_stmt ? SRA_AM_MODIFIED : SRA_AM_NONE;
3048 /* Traverse the function body and all modifications as decided in
3049 analyze_all_variable_accesses. Return true iff the CFG has been
3050 changed. */
3052 static bool
3053 sra_modify_function_body (void)
3055 bool cfg_changed = false;
3056 basic_block bb;
3058 FOR_EACH_BB (bb)
3060 gimple_stmt_iterator gsi = gsi_start_bb (bb);
3061 while (!gsi_end_p (gsi))
3063 gimple stmt = gsi_stmt (gsi);
3064 enum assignment_mod_result assign_result;
3065 bool modified = false, deleted = false;
3066 tree *t;
3067 unsigned i;
3069 switch (gimple_code (stmt))
3071 case GIMPLE_RETURN:
3072 t = gimple_return_retval_ptr (stmt);
3073 if (*t != NULL_TREE)
3074 modified |= sra_modify_expr (t, &gsi, false);
3075 break;
3077 case GIMPLE_ASSIGN:
3078 assign_result = sra_modify_assign (&stmt, &gsi);
3079 modified |= assign_result == SRA_AM_MODIFIED;
3080 deleted = assign_result == SRA_AM_REMOVED;
3081 break;
3083 case GIMPLE_CALL:
3084 /* Operands must be processed before the lhs. */
3085 for (i = 0; i < gimple_call_num_args (stmt); i++)
3087 t = gimple_call_arg_ptr (stmt, i);
3088 modified |= sra_modify_expr (t, &gsi, false);
3091 if (gimple_call_lhs (stmt))
3093 t = gimple_call_lhs_ptr (stmt);
3094 modified |= sra_modify_expr (t, &gsi, true);
3096 break;
3098 case GIMPLE_ASM:
3099 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
3101 t = &TREE_VALUE (gimple_asm_input_op (stmt, i));
3102 modified |= sra_modify_expr (t, &gsi, false);
3104 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
3106 t = &TREE_VALUE (gimple_asm_output_op (stmt, i));
3107 modified |= sra_modify_expr (t, &gsi, true);
3109 break;
3111 default:
3112 break;
3115 if (modified)
3117 update_stmt (stmt);
3118 if (maybe_clean_eh_stmt (stmt)
3119 && gimple_purge_dead_eh_edges (gimple_bb (stmt)))
3120 cfg_changed = true;
3122 if (!deleted)
3123 gsi_next (&gsi);
3127 return cfg_changed;
3130 /* Generate statements initializing scalar replacements of parts of function
3131 parameters. */
3133 static void
3134 initialize_parameter_reductions (void)
3136 gimple_stmt_iterator gsi;
3137 gimple_seq seq = NULL;
3138 tree parm;
3140 for (parm = DECL_ARGUMENTS (current_function_decl);
3141 parm;
3142 parm = DECL_CHAIN (parm))
3144 VEC (access_p, heap) *access_vec;
3145 struct access *access;
3147 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3148 continue;
3149 access_vec = get_base_access_vector (parm);
3150 if (!access_vec)
3151 continue;
3153 if (!seq)
3155 seq = gimple_seq_alloc ();
3156 gsi = gsi_start (seq);
3159 for (access = VEC_index (access_p, access_vec, 0);
3160 access;
3161 access = access->next_grp)
3162 generate_subtree_copies (access, parm, 0, 0, 0, &gsi, true, true,
3163 EXPR_LOCATION (parm));
3166 if (seq)
3167 gsi_insert_seq_on_edge_immediate (single_succ_edge (ENTRY_BLOCK_PTR), seq);
3170 /* The "main" function of intraprocedural SRA passes. Runs the analysis and if
3171 it reveals there are components of some aggregates to be scalarized, it runs
3172 the required transformations. */
3173 static unsigned int
3174 perform_intra_sra (void)
3176 int ret = 0;
3177 sra_initialize ();
3179 if (!find_var_candidates ())
3180 goto out;
3182 if (!scan_function ())
3183 goto out;
3185 if (!analyze_all_variable_accesses ())
3186 goto out;
3188 if (sra_modify_function_body ())
3189 ret = TODO_update_ssa | TODO_cleanup_cfg;
3190 else
3191 ret = TODO_update_ssa;
3192 initialize_parameter_reductions ();
3194 statistics_counter_event (cfun, "Scalar replacements created",
3195 sra_stats.replacements);
3196 statistics_counter_event (cfun, "Modified expressions", sra_stats.exprs);
3197 statistics_counter_event (cfun, "Subtree copy stmts",
3198 sra_stats.subtree_copies);
3199 statistics_counter_event (cfun, "Subreplacement stmts",
3200 sra_stats.subreplacements);
3201 statistics_counter_event (cfun, "Deleted stmts", sra_stats.deleted);
3202 statistics_counter_event (cfun, "Separate LHS and RHS handling",
3203 sra_stats.separate_lhs_rhs_handling);
3205 out:
3206 sra_deinitialize ();
3207 return ret;
3210 /* Perform early intraprocedural SRA. */
3211 static unsigned int
3212 early_intra_sra (void)
3214 sra_mode = SRA_MODE_EARLY_INTRA;
3215 return perform_intra_sra ();
3218 /* Perform "late" intraprocedural SRA. */
3219 static unsigned int
3220 late_intra_sra (void)
3222 sra_mode = SRA_MODE_INTRA;
3223 return perform_intra_sra ();
3227 static bool
3228 gate_intra_sra (void)
3230 return flag_tree_sra != 0 && dbg_cnt (tree_sra);
3234 struct gimple_opt_pass pass_sra_early =
3237 GIMPLE_PASS,
3238 "esra", /* name */
3239 gate_intra_sra, /* gate */
3240 early_intra_sra, /* execute */
3241 NULL, /* sub */
3242 NULL, /* next */
3243 0, /* static_pass_number */
3244 TV_TREE_SRA, /* tv_id */
3245 PROP_cfg | PROP_ssa, /* properties_required */
3246 0, /* properties_provided */
3247 0, /* properties_destroyed */
3248 0, /* todo_flags_start */
3249 TODO_update_ssa
3250 | TODO_ggc_collect
3251 | TODO_verify_ssa /* todo_flags_finish */
3255 struct gimple_opt_pass pass_sra =
3258 GIMPLE_PASS,
3259 "sra", /* name */
3260 gate_intra_sra, /* gate */
3261 late_intra_sra, /* execute */
3262 NULL, /* sub */
3263 NULL, /* next */
3264 0, /* static_pass_number */
3265 TV_TREE_SRA, /* tv_id */
3266 PROP_cfg | PROP_ssa, /* properties_required */
3267 0, /* properties_provided */
3268 0, /* properties_destroyed */
3269 TODO_update_address_taken, /* todo_flags_start */
3270 TODO_update_ssa
3271 | TODO_ggc_collect
3272 | TODO_verify_ssa /* todo_flags_finish */
3277 /* Return true iff PARM (which must be a parm_decl) is an unused scalar
3278 parameter. */
3280 static bool
3281 is_unused_scalar_param (tree parm)
3283 tree name;
3284 return (is_gimple_reg (parm)
3285 && (!(name = gimple_default_def (cfun, parm))
3286 || has_zero_uses (name)));
3289 /* Scan immediate uses of a default definition SSA name of a parameter PARM and
3290 examine whether there are any direct or otherwise infeasible ones. If so,
3291 return true, otherwise return false. PARM must be a gimple register with a
3292 non-NULL default definition. */
3294 static bool
3295 ptr_parm_has_direct_uses (tree parm)
3297 imm_use_iterator ui;
3298 gimple stmt;
3299 tree name = gimple_default_def (cfun, parm);
3300 bool ret = false;
3302 FOR_EACH_IMM_USE_STMT (stmt, ui, name)
3304 int uses_ok = 0;
3305 use_operand_p use_p;
3307 if (is_gimple_debug (stmt))
3308 continue;
3310 /* Valid uses include dereferences on the lhs and the rhs. */
3311 if (gimple_has_lhs (stmt))
3313 tree lhs = gimple_get_lhs (stmt);
3314 while (handled_component_p (lhs))
3315 lhs = TREE_OPERAND (lhs, 0);
3316 if (TREE_CODE (lhs) == MEM_REF
3317 && TREE_OPERAND (lhs, 0) == name
3318 && integer_zerop (TREE_OPERAND (lhs, 1))
3319 && types_compatible_p (TREE_TYPE (lhs),
3320 TREE_TYPE (TREE_TYPE (name)))
3321 && !TREE_THIS_VOLATILE (lhs))
3322 uses_ok++;
3324 if (gimple_assign_single_p (stmt))
3326 tree rhs = gimple_assign_rhs1 (stmt);
3327 while (handled_component_p (rhs))
3328 rhs = TREE_OPERAND (rhs, 0);
3329 if (TREE_CODE (rhs) == MEM_REF
3330 && TREE_OPERAND (rhs, 0) == name
3331 && integer_zerop (TREE_OPERAND (rhs, 1))
3332 && types_compatible_p (TREE_TYPE (rhs),
3333 TREE_TYPE (TREE_TYPE (name)))
3334 && !TREE_THIS_VOLATILE (rhs))
3335 uses_ok++;
3337 else if (is_gimple_call (stmt))
3339 unsigned i;
3340 for (i = 0; i < gimple_call_num_args (stmt); ++i)
3342 tree arg = gimple_call_arg (stmt, i);
3343 while (handled_component_p (arg))
3344 arg = TREE_OPERAND (arg, 0);
3345 if (TREE_CODE (arg) == MEM_REF
3346 && TREE_OPERAND (arg, 0) == name
3347 && integer_zerop (TREE_OPERAND (arg, 1))
3348 && types_compatible_p (TREE_TYPE (arg),
3349 TREE_TYPE (TREE_TYPE (name)))
3350 && !TREE_THIS_VOLATILE (arg))
3351 uses_ok++;
3355 /* If the number of valid uses does not match the number of
3356 uses in this stmt there is an unhandled use. */
3357 FOR_EACH_IMM_USE_ON_STMT (use_p, ui)
3358 --uses_ok;
3360 if (uses_ok != 0)
3361 ret = true;
3363 if (ret)
3364 BREAK_FROM_IMM_USE_STMT (ui);
3367 return ret;
3370 /* Identify candidates for reduction for IPA-SRA based on their type and mark
3371 them in candidate_bitmap. Note that these do not necessarily include
3372 parameter which are unused and thus can be removed. Return true iff any
3373 such candidate has been found. */
3375 static bool
3376 find_param_candidates (void)
3378 tree parm;
3379 int count = 0;
3380 bool ret = false;
3381 const char *msg;
3383 for (parm = DECL_ARGUMENTS (current_function_decl);
3384 parm;
3385 parm = DECL_CHAIN (parm))
3387 tree type = TREE_TYPE (parm);
3389 count++;
3391 if (TREE_THIS_VOLATILE (parm)
3392 || TREE_ADDRESSABLE (parm)
3393 || (!is_gimple_reg_type (type) && is_va_list_type (type)))
3394 continue;
3396 if (is_unused_scalar_param (parm))
3398 ret = true;
3399 continue;
3402 if (POINTER_TYPE_P (type))
3404 type = TREE_TYPE (type);
3406 if (TREE_CODE (type) == FUNCTION_TYPE
3407 || TYPE_VOLATILE (type)
3408 || (TREE_CODE (type) == ARRAY_TYPE
3409 && TYPE_NONALIASED_COMPONENT (type))
3410 || !is_gimple_reg (parm)
3411 || is_va_list_type (type)
3412 || ptr_parm_has_direct_uses (parm))
3413 continue;
3415 else if (!AGGREGATE_TYPE_P (type))
3416 continue;
3418 if (!COMPLETE_TYPE_P (type)
3419 || !host_integerp (TYPE_SIZE (type), 1)
3420 || tree_low_cst (TYPE_SIZE (type), 1) == 0
3421 || (AGGREGATE_TYPE_P (type)
3422 && type_internals_preclude_sra_p (type, &msg)))
3423 continue;
3425 bitmap_set_bit (candidate_bitmap, DECL_UID (parm));
3426 ret = true;
3427 if (dump_file && (dump_flags & TDF_DETAILS))
3429 fprintf (dump_file, "Candidate (%d): ", DECL_UID (parm));
3430 print_generic_expr (dump_file, parm, 0);
3431 fprintf (dump_file, "\n");
3435 func_param_count = count;
3436 return ret;
3439 /* Callback of walk_aliased_vdefs, marks the access passed as DATA as
3440 maybe_modified. */
3442 static bool
3443 mark_maybe_modified (ao_ref *ao ATTRIBUTE_UNUSED, tree vdef ATTRIBUTE_UNUSED,
3444 void *data)
3446 struct access *repr = (struct access *) data;
3448 repr->grp_maybe_modified = 1;
3449 return true;
3452 /* Analyze what representatives (in linked lists accessible from
3453 REPRESENTATIVES) can be modified by side effects of statements in the
3454 current function. */
3456 static void
3457 analyze_modified_params (VEC (access_p, heap) *representatives)
3459 int i;
3461 for (i = 0; i < func_param_count; i++)
3463 struct access *repr;
3465 for (repr = VEC_index (access_p, representatives, i);
3466 repr;
3467 repr = repr->next_grp)
3469 struct access *access;
3470 bitmap visited;
3471 ao_ref ar;
3473 if (no_accesses_p (repr))
3474 continue;
3475 if (!POINTER_TYPE_P (TREE_TYPE (repr->base))
3476 || repr->grp_maybe_modified)
3477 continue;
3479 ao_ref_init (&ar, repr->expr);
3480 visited = BITMAP_ALLOC (NULL);
3481 for (access = repr; access; access = access->next_sibling)
3483 /* All accesses are read ones, otherwise grp_maybe_modified would
3484 be trivially set. */
3485 walk_aliased_vdefs (&ar, gimple_vuse (access->stmt),
3486 mark_maybe_modified, repr, &visited);
3487 if (repr->grp_maybe_modified)
3488 break;
3490 BITMAP_FREE (visited);
3495 /* Propagate distances in bb_dereferences in the opposite direction than the
3496 control flow edges, in each step storing the maximum of the current value
3497 and the minimum of all successors. These steps are repeated until the table
3498 stabilizes. Note that BBs which might terminate the functions (according to
3499 final_bbs bitmap) never updated in this way. */
3501 static void
3502 propagate_dereference_distances (void)
3504 VEC (basic_block, heap) *queue;
3505 basic_block bb;
3507 queue = VEC_alloc (basic_block, heap, last_basic_block_for_function (cfun));
3508 VEC_quick_push (basic_block, queue, ENTRY_BLOCK_PTR);
3509 FOR_EACH_BB (bb)
3511 VEC_quick_push (basic_block, queue, bb);
3512 bb->aux = bb;
3515 while (!VEC_empty (basic_block, queue))
3517 edge_iterator ei;
3518 edge e;
3519 bool change = false;
3520 int i;
3522 bb = VEC_pop (basic_block, queue);
3523 bb->aux = NULL;
3525 if (bitmap_bit_p (final_bbs, bb->index))
3526 continue;
3528 for (i = 0; i < func_param_count; i++)
3530 int idx = bb->index * func_param_count + i;
3531 bool first = true;
3532 HOST_WIDE_INT inh = 0;
3534 FOR_EACH_EDGE (e, ei, bb->succs)
3536 int succ_idx = e->dest->index * func_param_count + i;
3538 if (e->src == EXIT_BLOCK_PTR)
3539 continue;
3541 if (first)
3543 first = false;
3544 inh = bb_dereferences [succ_idx];
3546 else if (bb_dereferences [succ_idx] < inh)
3547 inh = bb_dereferences [succ_idx];
3550 if (!first && bb_dereferences[idx] < inh)
3552 bb_dereferences[idx] = inh;
3553 change = true;
3557 if (change && !bitmap_bit_p (final_bbs, bb->index))
3558 FOR_EACH_EDGE (e, ei, bb->preds)
3560 if (e->src->aux)
3561 continue;
3563 e->src->aux = e->src;
3564 VEC_quick_push (basic_block, queue, e->src);
3568 VEC_free (basic_block, heap, queue);
3571 /* Dump a dereferences TABLE with heading STR to file F. */
3573 static void
3574 dump_dereferences_table (FILE *f, const char *str, HOST_WIDE_INT *table)
3576 basic_block bb;
3578 fprintf (dump_file, str);
3579 FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, EXIT_BLOCK_PTR, next_bb)
3581 fprintf (f, "%4i %i ", bb->index, bitmap_bit_p (final_bbs, bb->index));
3582 if (bb != EXIT_BLOCK_PTR)
3584 int i;
3585 for (i = 0; i < func_param_count; i++)
3587 int idx = bb->index * func_param_count + i;
3588 fprintf (f, " %4" HOST_WIDE_INT_PRINT "d", table[idx]);
3591 fprintf (f, "\n");
3593 fprintf (dump_file, "\n");
3596 /* Determine what (parts of) parameters passed by reference that are not
3597 assigned to are not certainly dereferenced in this function and thus the
3598 dereferencing cannot be safely moved to the caller without potentially
3599 introducing a segfault. Mark such REPRESENTATIVES as
3600 grp_not_necessarilly_dereferenced.
3602 The dereferenced maximum "distance," i.e. the offset + size of the accessed
3603 part is calculated rather than simple booleans are calculated for each
3604 pointer parameter to handle cases when only a fraction of the whole
3605 aggregate is allocated (see testsuite/gcc.c-torture/execute/ipa-sra-2.c for
3606 an example).
3608 The maximum dereference distances for each pointer parameter and BB are
3609 already stored in bb_dereference. This routine simply propagates these
3610 values upwards by propagate_dereference_distances and then compares the
3611 distances of individual parameters in the ENTRY BB to the equivalent
3612 distances of each representative of a (fraction of a) parameter. */
3614 static void
3615 analyze_caller_dereference_legality (VEC (access_p, heap) *representatives)
3617 int i;
3619 if (dump_file && (dump_flags & TDF_DETAILS))
3620 dump_dereferences_table (dump_file,
3621 "Dereference table before propagation:\n",
3622 bb_dereferences);
3624 propagate_dereference_distances ();
3626 if (dump_file && (dump_flags & TDF_DETAILS))
3627 dump_dereferences_table (dump_file,
3628 "Dereference table after propagation:\n",
3629 bb_dereferences);
3631 for (i = 0; i < func_param_count; i++)
3633 struct access *repr = VEC_index (access_p, representatives, i);
3634 int idx = ENTRY_BLOCK_PTR->index * func_param_count + i;
3636 if (!repr || no_accesses_p (repr))
3637 continue;
3641 if ((repr->offset + repr->size) > bb_dereferences[idx])
3642 repr->grp_not_necessarilly_dereferenced = 1;
3643 repr = repr->next_grp;
3645 while (repr);
3649 /* Return the representative access for the parameter declaration PARM if it is
3650 a scalar passed by reference which is not written to and the pointer value
3651 is not used directly. Thus, if it is legal to dereference it in the caller
3652 and we can rule out modifications through aliases, such parameter should be
3653 turned into one passed by value. Return NULL otherwise. */
3655 static struct access *
3656 unmodified_by_ref_scalar_representative (tree parm)
3658 int i, access_count;
3659 struct access *repr;
3660 VEC (access_p, heap) *access_vec;
3662 access_vec = get_base_access_vector (parm);
3663 gcc_assert (access_vec);
3664 repr = VEC_index (access_p, access_vec, 0);
3665 if (repr->write)
3666 return NULL;
3667 repr->group_representative = repr;
3669 access_count = VEC_length (access_p, access_vec);
3670 for (i = 1; i < access_count; i++)
3672 struct access *access = VEC_index (access_p, access_vec, i);
3673 if (access->write)
3674 return NULL;
3675 access->group_representative = repr;
3676 access->next_sibling = repr->next_sibling;
3677 repr->next_sibling = access;
3680 repr->grp_read = 1;
3681 repr->grp_scalar_ptr = 1;
3682 return repr;
3685 /* Return true iff this access precludes IPA-SRA of the parameter it is
3686 associated with. */
3688 static bool
3689 access_precludes_ipa_sra_p (struct access *access)
3691 /* Avoid issues such as the second simple testcase in PR 42025. The problem
3692 is incompatible assign in a call statement (and possibly even in asm
3693 statements). This can be relaxed by using a new temporary but only for
3694 non-TREE_ADDRESSABLE types and is probably not worth the complexity. (In
3695 intraprocedural SRA we deal with this by keeping the old aggregate around,
3696 something we cannot do in IPA-SRA.) */
3697 if (access->write
3698 && (is_gimple_call (access->stmt)
3699 || gimple_code (access->stmt) == GIMPLE_ASM))
3700 return true;
3702 if (tree_non_mode_aligned_mem_p (access->expr))
3703 return true;
3705 return false;
3709 /* Sort collected accesses for parameter PARM, identify representatives for
3710 each accessed region and link them together. Return NULL if there are
3711 different but overlapping accesses, return the special ptr value meaning
3712 there are no accesses for this parameter if that is the case and return the
3713 first representative otherwise. Set *RO_GRP if there is a group of accesses
3714 with only read (i.e. no write) accesses. */
3716 static struct access *
3717 splice_param_accesses (tree parm, bool *ro_grp)
3719 int i, j, access_count, group_count;
3720 int agg_size, total_size = 0;
3721 struct access *access, *res, **prev_acc_ptr = &res;
3722 VEC (access_p, heap) *access_vec;
3724 access_vec = get_base_access_vector (parm);
3725 if (!access_vec)
3726 return &no_accesses_representant;
3727 access_count = VEC_length (access_p, access_vec);
3729 VEC_qsort (access_p, access_vec, compare_access_positions);
3731 i = 0;
3732 total_size = 0;
3733 group_count = 0;
3734 while (i < access_count)
3736 bool modification;
3737 tree a1_alias_type;
3738 access = VEC_index (access_p, access_vec, i);
3739 modification = access->write;
3740 if (access_precludes_ipa_sra_p (access))
3741 return NULL;
3742 a1_alias_type = reference_alias_ptr_type (access->expr);
3744 /* Access is about to become group representative unless we find some
3745 nasty overlap which would preclude us from breaking this parameter
3746 apart. */
3748 j = i + 1;
3749 while (j < access_count)
3751 struct access *ac2 = VEC_index (access_p, access_vec, j);
3752 if (ac2->offset != access->offset)
3754 /* All or nothing law for parameters. */
3755 if (access->offset + access->size > ac2->offset)
3756 return NULL;
3757 else
3758 break;
3760 else if (ac2->size != access->size)
3761 return NULL;
3763 if (access_precludes_ipa_sra_p (ac2)
3764 || (ac2->type != access->type
3765 && (TREE_ADDRESSABLE (ac2->type)
3766 || TREE_ADDRESSABLE (access->type)))
3767 || (reference_alias_ptr_type (ac2->expr) != a1_alias_type))
3768 return NULL;
3770 modification |= ac2->write;
3771 ac2->group_representative = access;
3772 ac2->next_sibling = access->next_sibling;
3773 access->next_sibling = ac2;
3774 j++;
3777 group_count++;
3778 access->grp_maybe_modified = modification;
3779 if (!modification)
3780 *ro_grp = true;
3781 *prev_acc_ptr = access;
3782 prev_acc_ptr = &access->next_grp;
3783 total_size += access->size;
3784 i = j;
3787 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3788 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm))), 1);
3789 else
3790 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (parm)), 1);
3791 if (total_size >= agg_size)
3792 return NULL;
3794 gcc_assert (group_count > 0);
3795 return res;
3798 /* Decide whether parameters with representative accesses given by REPR should
3799 be reduced into components. */
3801 static int
3802 decide_one_param_reduction (struct access *repr)
3804 int total_size, cur_parm_size, agg_size, new_param_count, parm_size_limit;
3805 bool by_ref;
3806 tree parm;
3808 parm = repr->base;
3809 cur_parm_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (parm)), 1);
3810 gcc_assert (cur_parm_size > 0);
3812 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3814 by_ref = true;
3815 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm))), 1);
3817 else
3819 by_ref = false;
3820 agg_size = cur_parm_size;
3823 if (dump_file)
3825 struct access *acc;
3826 fprintf (dump_file, "Evaluating PARAM group sizes for ");
3827 print_generic_expr (dump_file, parm, 0);
3828 fprintf (dump_file, " (UID: %u): \n", DECL_UID (parm));
3829 for (acc = repr; acc; acc = acc->next_grp)
3830 dump_access (dump_file, acc, true);
3833 total_size = 0;
3834 new_param_count = 0;
3836 for (; repr; repr = repr->next_grp)
3838 gcc_assert (parm == repr->base);
3840 /* Taking the address of a non-addressable field is verboten. */
3841 if (by_ref && repr->non_addressable)
3842 return 0;
3844 if (!by_ref || (!repr->grp_maybe_modified
3845 && !repr->grp_not_necessarilly_dereferenced))
3846 total_size += repr->size;
3847 else
3848 total_size += cur_parm_size;
3850 new_param_count++;
3853 gcc_assert (new_param_count > 0);
3855 if (optimize_function_for_size_p (cfun))
3856 parm_size_limit = cur_parm_size;
3857 else
3858 parm_size_limit = (PARAM_VALUE (PARAM_IPA_SRA_PTR_GROWTH_FACTOR)
3859 * cur_parm_size);
3861 if (total_size < agg_size
3862 && total_size <= parm_size_limit)
3864 if (dump_file)
3865 fprintf (dump_file, " ....will be split into %i components\n",
3866 new_param_count);
3867 return new_param_count;
3869 else
3870 return 0;
3873 /* The order of the following enums is important, we need to do extra work for
3874 UNUSED_PARAMS, BY_VAL_ACCESSES and UNMODIF_BY_REF_ACCESSES. */
3875 enum ipa_splicing_result { NO_GOOD_ACCESS, UNUSED_PARAMS, BY_VAL_ACCESSES,
3876 MODIF_BY_REF_ACCESSES, UNMODIF_BY_REF_ACCESSES };
3878 /* Identify representatives of all accesses to all candidate parameters for
3879 IPA-SRA. Return result based on what representatives have been found. */
3881 static enum ipa_splicing_result
3882 splice_all_param_accesses (VEC (access_p, heap) **representatives)
3884 enum ipa_splicing_result result = NO_GOOD_ACCESS;
3885 tree parm;
3886 struct access *repr;
3888 *representatives = VEC_alloc (access_p, heap, func_param_count);
3890 for (parm = DECL_ARGUMENTS (current_function_decl);
3891 parm;
3892 parm = DECL_CHAIN (parm))
3894 if (is_unused_scalar_param (parm))
3896 VEC_quick_push (access_p, *representatives,
3897 &no_accesses_representant);
3898 if (result == NO_GOOD_ACCESS)
3899 result = UNUSED_PARAMS;
3901 else if (POINTER_TYPE_P (TREE_TYPE (parm))
3902 && is_gimple_reg_type (TREE_TYPE (TREE_TYPE (parm)))
3903 && bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3905 repr = unmodified_by_ref_scalar_representative (parm);
3906 VEC_quick_push (access_p, *representatives, repr);
3907 if (repr)
3908 result = UNMODIF_BY_REF_ACCESSES;
3910 else if (bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3912 bool ro_grp = false;
3913 repr = splice_param_accesses (parm, &ro_grp);
3914 VEC_quick_push (access_p, *representatives, repr);
3916 if (repr && !no_accesses_p (repr))
3918 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3920 if (ro_grp)
3921 result = UNMODIF_BY_REF_ACCESSES;
3922 else if (result < MODIF_BY_REF_ACCESSES)
3923 result = MODIF_BY_REF_ACCESSES;
3925 else if (result < BY_VAL_ACCESSES)
3926 result = BY_VAL_ACCESSES;
3928 else if (no_accesses_p (repr) && (result == NO_GOOD_ACCESS))
3929 result = UNUSED_PARAMS;
3931 else
3932 VEC_quick_push (access_p, *representatives, NULL);
3935 if (result == NO_GOOD_ACCESS)
3937 VEC_free (access_p, heap, *representatives);
3938 *representatives = NULL;
3939 return NO_GOOD_ACCESS;
3942 return result;
3945 /* Return the index of BASE in PARMS. Abort if it is not found. */
3947 static inline int
3948 get_param_index (tree base, VEC(tree, heap) *parms)
3950 int i, len;
3952 len = VEC_length (tree, parms);
3953 for (i = 0; i < len; i++)
3954 if (VEC_index (tree, parms, i) == base)
3955 return i;
3956 gcc_unreachable ();
3959 /* Convert the decisions made at the representative level into compact
3960 parameter adjustments. REPRESENTATIVES are pointers to first
3961 representatives of each param accesses, ADJUSTMENTS_COUNT is the expected
3962 final number of adjustments. */
3964 static ipa_parm_adjustment_vec
3965 turn_representatives_into_adjustments (VEC (access_p, heap) *representatives,
3966 int adjustments_count)
3968 VEC (tree, heap) *parms;
3969 ipa_parm_adjustment_vec adjustments;
3970 tree parm;
3971 int i;
3973 gcc_assert (adjustments_count > 0);
3974 parms = ipa_get_vector_of_formal_parms (current_function_decl);
3975 adjustments = VEC_alloc (ipa_parm_adjustment_t, heap, adjustments_count);
3976 parm = DECL_ARGUMENTS (current_function_decl);
3977 for (i = 0; i < func_param_count; i++, parm = DECL_CHAIN (parm))
3979 struct access *repr = VEC_index (access_p, representatives, i);
3981 if (!repr || no_accesses_p (repr))
3983 struct ipa_parm_adjustment *adj;
3985 adj = VEC_quick_push (ipa_parm_adjustment_t, adjustments, NULL);
3986 memset (adj, 0, sizeof (*adj));
3987 adj->base_index = get_param_index (parm, parms);
3988 adj->base = parm;
3989 if (!repr)
3990 adj->copy_param = 1;
3991 else
3992 adj->remove_param = 1;
3994 else
3996 struct ipa_parm_adjustment *adj;
3997 int index = get_param_index (parm, parms);
3999 for (; repr; repr = repr->next_grp)
4001 adj = VEC_quick_push (ipa_parm_adjustment_t, adjustments, NULL);
4002 memset (adj, 0, sizeof (*adj));
4003 gcc_assert (repr->base == parm);
4004 adj->base_index = index;
4005 adj->base = repr->base;
4006 adj->type = repr->type;
4007 adj->alias_ptr_type = reference_alias_ptr_type (repr->expr);
4008 adj->offset = repr->offset;
4009 adj->by_ref = (POINTER_TYPE_P (TREE_TYPE (repr->base))
4010 && (repr->grp_maybe_modified
4011 || repr->grp_not_necessarilly_dereferenced));
4016 VEC_free (tree, heap, parms);
4017 return adjustments;
4020 /* Analyze the collected accesses and produce a plan what to do with the
4021 parameters in the form of adjustments, NULL meaning nothing. */
4023 static ipa_parm_adjustment_vec
4024 analyze_all_param_acesses (void)
4026 enum ipa_splicing_result repr_state;
4027 bool proceed = false;
4028 int i, adjustments_count = 0;
4029 VEC (access_p, heap) *representatives;
4030 ipa_parm_adjustment_vec adjustments;
4032 repr_state = splice_all_param_accesses (&representatives);
4033 if (repr_state == NO_GOOD_ACCESS)
4034 return NULL;
4036 /* If there are any parameters passed by reference which are not modified
4037 directly, we need to check whether they can be modified indirectly. */
4038 if (repr_state == UNMODIF_BY_REF_ACCESSES)
4040 analyze_caller_dereference_legality (representatives);
4041 analyze_modified_params (representatives);
4044 for (i = 0; i < func_param_count; i++)
4046 struct access *repr = VEC_index (access_p, representatives, i);
4048 if (repr && !no_accesses_p (repr))
4050 if (repr->grp_scalar_ptr)
4052 adjustments_count++;
4053 if (repr->grp_not_necessarilly_dereferenced
4054 || repr->grp_maybe_modified)
4055 VEC_replace (access_p, representatives, i, NULL);
4056 else
4058 proceed = true;
4059 sra_stats.scalar_by_ref_to_by_val++;
4062 else
4064 int new_components = decide_one_param_reduction (repr);
4066 if (new_components == 0)
4068 VEC_replace (access_p, representatives, i, NULL);
4069 adjustments_count++;
4071 else
4073 adjustments_count += new_components;
4074 sra_stats.aggregate_params_reduced++;
4075 sra_stats.param_reductions_created += new_components;
4076 proceed = true;
4080 else
4082 if (no_accesses_p (repr))
4084 proceed = true;
4085 sra_stats.deleted_unused_parameters++;
4087 adjustments_count++;
4091 if (!proceed && dump_file)
4092 fprintf (dump_file, "NOT proceeding to change params.\n");
4094 if (proceed)
4095 adjustments = turn_representatives_into_adjustments (representatives,
4096 adjustments_count);
4097 else
4098 adjustments = NULL;
4100 VEC_free (access_p, heap, representatives);
4101 return adjustments;
4104 /* If a parameter replacement identified by ADJ does not yet exist in the form
4105 of declaration, create it and record it, otherwise return the previously
4106 created one. */
4108 static tree
4109 get_replaced_param_substitute (struct ipa_parm_adjustment *adj)
4111 tree repl;
4112 if (!adj->new_ssa_base)
4114 char *pretty_name = make_fancy_name (adj->base);
4116 repl = create_tmp_reg (TREE_TYPE (adj->base), "ISR");
4117 DECL_NAME (repl) = get_identifier (pretty_name);
4118 obstack_free (&name_obstack, pretty_name);
4120 add_referenced_var (repl);
4121 adj->new_ssa_base = repl;
4123 else
4124 repl = adj->new_ssa_base;
4125 return repl;
4128 /* Find the first adjustment for a particular parameter BASE in a vector of
4129 ADJUSTMENTS which is not a copy_param. Return NULL if there is no such
4130 adjustment. */
4132 static struct ipa_parm_adjustment *
4133 get_adjustment_for_base (ipa_parm_adjustment_vec adjustments, tree base)
4135 int i, len;
4137 len = VEC_length (ipa_parm_adjustment_t, adjustments);
4138 for (i = 0; i < len; i++)
4140 struct ipa_parm_adjustment *adj;
4142 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
4143 if (!adj->copy_param && adj->base == base)
4144 return adj;
4147 return NULL;
4150 /* If the statement STMT defines an SSA_NAME of a parameter which is to be
4151 removed because its value is not used, replace the SSA_NAME with a one
4152 relating to a created VAR_DECL together all of its uses and return true.
4153 ADJUSTMENTS is a pointer to an adjustments vector. */
4155 static bool
4156 replace_removed_params_ssa_names (gimple stmt,
4157 ipa_parm_adjustment_vec adjustments)
4159 struct ipa_parm_adjustment *adj;
4160 tree lhs, decl, repl, name;
4162 if (gimple_code (stmt) == GIMPLE_PHI)
4163 lhs = gimple_phi_result (stmt);
4164 else if (is_gimple_assign (stmt))
4165 lhs = gimple_assign_lhs (stmt);
4166 else if (is_gimple_call (stmt))
4167 lhs = gimple_call_lhs (stmt);
4168 else
4169 gcc_unreachable ();
4171 if (TREE_CODE (lhs) != SSA_NAME)
4172 return false;
4173 decl = SSA_NAME_VAR (lhs);
4174 if (TREE_CODE (decl) != PARM_DECL)
4175 return false;
4177 adj = get_adjustment_for_base (adjustments, decl);
4178 if (!adj)
4179 return false;
4181 repl = get_replaced_param_substitute (adj);
4182 name = make_ssa_name (repl, stmt);
4184 if (dump_file)
4186 fprintf (dump_file, "replacing an SSA name of a removed param ");
4187 print_generic_expr (dump_file, lhs, 0);
4188 fprintf (dump_file, " with ");
4189 print_generic_expr (dump_file, name, 0);
4190 fprintf (dump_file, "\n");
4193 if (is_gimple_assign (stmt))
4194 gimple_assign_set_lhs (stmt, name);
4195 else if (is_gimple_call (stmt))
4196 gimple_call_set_lhs (stmt, name);
4197 else
4198 gimple_phi_set_result (stmt, name);
4200 replace_uses_by (lhs, name);
4201 release_ssa_name (lhs);
4202 return true;
4205 /* If the expression *EXPR should be replaced by a reduction of a parameter, do
4206 so. ADJUSTMENTS is a pointer to a vector of adjustments. CONVERT
4207 specifies whether the function should care about type incompatibility the
4208 current and new expressions. If it is false, the function will leave
4209 incompatibility issues to the caller. Return true iff the expression
4210 was modified. */
4212 static bool
4213 sra_ipa_modify_expr (tree *expr, bool convert,
4214 ipa_parm_adjustment_vec adjustments)
4216 int i, len;
4217 struct ipa_parm_adjustment *adj, *cand = NULL;
4218 HOST_WIDE_INT offset, size, max_size;
4219 tree base, src;
4221 len = VEC_length (ipa_parm_adjustment_t, adjustments);
4223 if (TREE_CODE (*expr) == BIT_FIELD_REF
4224 || TREE_CODE (*expr) == IMAGPART_EXPR
4225 || TREE_CODE (*expr) == REALPART_EXPR)
4227 expr = &TREE_OPERAND (*expr, 0);
4228 convert = true;
4231 base = get_ref_base_and_extent (*expr, &offset, &size, &max_size);
4232 if (!base || size == -1 || max_size == -1)
4233 return false;
4235 if (TREE_CODE (base) == MEM_REF)
4237 offset += mem_ref_offset (base).low * BITS_PER_UNIT;
4238 base = TREE_OPERAND (base, 0);
4241 base = get_ssa_base_param (base);
4242 if (!base || TREE_CODE (base) != PARM_DECL)
4243 return false;
4245 for (i = 0; i < len; i++)
4247 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
4249 if (adj->base == base &&
4250 (adj->offset == offset || adj->remove_param))
4252 cand = adj;
4253 break;
4256 if (!cand || cand->copy_param || cand->remove_param)
4257 return false;
4259 if (cand->by_ref)
4260 src = build_simple_mem_ref (cand->reduction);
4261 else
4262 src = cand->reduction;
4264 if (dump_file && (dump_flags & TDF_DETAILS))
4266 fprintf (dump_file, "About to replace expr ");
4267 print_generic_expr (dump_file, *expr, 0);
4268 fprintf (dump_file, " with ");
4269 print_generic_expr (dump_file, src, 0);
4270 fprintf (dump_file, "\n");
4273 if (convert && !useless_type_conversion_p (TREE_TYPE (*expr), cand->type))
4275 tree vce = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (*expr), src);
4276 *expr = vce;
4278 else
4279 *expr = src;
4280 return true;
4283 /* If the statement pointed to by STMT_PTR contains any expressions that need
4284 to replaced with a different one as noted by ADJUSTMENTS, do so. Handle any
4285 potential type incompatibilities (GSI is used to accommodate conversion
4286 statements and must point to the statement). Return true iff the statement
4287 was modified. */
4289 static bool
4290 sra_ipa_modify_assign (gimple *stmt_ptr, gimple_stmt_iterator *gsi,
4291 ipa_parm_adjustment_vec adjustments)
4293 gimple stmt = *stmt_ptr;
4294 tree *lhs_p, *rhs_p;
4295 bool any;
4297 if (!gimple_assign_single_p (stmt))
4298 return false;
4300 rhs_p = gimple_assign_rhs1_ptr (stmt);
4301 lhs_p = gimple_assign_lhs_ptr (stmt);
4303 any = sra_ipa_modify_expr (rhs_p, false, adjustments);
4304 any |= sra_ipa_modify_expr (lhs_p, false, adjustments);
4305 if (any)
4307 tree new_rhs = NULL_TREE;
4309 if (!useless_type_conversion_p (TREE_TYPE (*lhs_p), TREE_TYPE (*rhs_p)))
4311 if (TREE_CODE (*rhs_p) == CONSTRUCTOR)
4313 /* V_C_Es of constructors can cause trouble (PR 42714). */
4314 if (is_gimple_reg_type (TREE_TYPE (*lhs_p)))
4315 *rhs_p = build_zero_cst (TREE_TYPE (*lhs_p));
4316 else
4317 *rhs_p = build_constructor (TREE_TYPE (*lhs_p), 0);
4319 else
4320 new_rhs = fold_build1_loc (gimple_location (stmt),
4321 VIEW_CONVERT_EXPR, TREE_TYPE (*lhs_p),
4322 *rhs_p);
4324 else if (REFERENCE_CLASS_P (*rhs_p)
4325 && is_gimple_reg_type (TREE_TYPE (*lhs_p))
4326 && !is_gimple_reg (*lhs_p))
4327 /* This can happen when an assignment in between two single field
4328 structures is turned into an assignment in between two pointers to
4329 scalars (PR 42237). */
4330 new_rhs = *rhs_p;
4332 if (new_rhs)
4334 tree tmp = force_gimple_operand_gsi (gsi, new_rhs, true, NULL_TREE,
4335 true, GSI_SAME_STMT);
4337 gimple_assign_set_rhs_from_tree (gsi, tmp);
4340 return true;
4343 return false;
4346 /* Traverse the function body and all modifications as described in
4347 ADJUSTMENTS. Return true iff the CFG has been changed. */
4349 static bool
4350 ipa_sra_modify_function_body (ipa_parm_adjustment_vec adjustments)
4352 bool cfg_changed = false;
4353 basic_block bb;
4355 FOR_EACH_BB (bb)
4357 gimple_stmt_iterator gsi;
4359 for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi))
4360 replace_removed_params_ssa_names (gsi_stmt (gsi), adjustments);
4362 gsi = gsi_start_bb (bb);
4363 while (!gsi_end_p (gsi))
4365 gimple stmt = gsi_stmt (gsi);
4366 bool modified = false;
4367 tree *t;
4368 unsigned i;
4370 switch (gimple_code (stmt))
4372 case GIMPLE_RETURN:
4373 t = gimple_return_retval_ptr (stmt);
4374 if (*t != NULL_TREE)
4375 modified |= sra_ipa_modify_expr (t, true, adjustments);
4376 break;
4378 case GIMPLE_ASSIGN:
4379 modified |= sra_ipa_modify_assign (&stmt, &gsi, adjustments);
4380 modified |= replace_removed_params_ssa_names (stmt, adjustments);
4381 break;
4383 case GIMPLE_CALL:
4384 /* Operands must be processed before the lhs. */
4385 for (i = 0; i < gimple_call_num_args (stmt); i++)
4387 t = gimple_call_arg_ptr (stmt, i);
4388 modified |= sra_ipa_modify_expr (t, true, adjustments);
4391 if (gimple_call_lhs (stmt))
4393 t = gimple_call_lhs_ptr (stmt);
4394 modified |= sra_ipa_modify_expr (t, false, adjustments);
4395 modified |= replace_removed_params_ssa_names (stmt,
4396 adjustments);
4398 break;
4400 case GIMPLE_ASM:
4401 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
4403 t = &TREE_VALUE (gimple_asm_input_op (stmt, i));
4404 modified |= sra_ipa_modify_expr (t, true, adjustments);
4406 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
4408 t = &TREE_VALUE (gimple_asm_output_op (stmt, i));
4409 modified |= sra_ipa_modify_expr (t, false, adjustments);
4411 break;
4413 default:
4414 break;
4417 if (modified)
4419 update_stmt (stmt);
4420 if (maybe_clean_eh_stmt (stmt)
4421 && gimple_purge_dead_eh_edges (gimple_bb (stmt)))
4422 cfg_changed = true;
4424 gsi_next (&gsi);
4428 return cfg_changed;
4431 /* Call gimple_debug_bind_reset_value on all debug statements describing
4432 gimple register parameters that are being removed or replaced. */
4434 static void
4435 sra_ipa_reset_debug_stmts (ipa_parm_adjustment_vec adjustments)
4437 int i, len;
4438 gimple_stmt_iterator *gsip = NULL, gsi;
4440 if (MAY_HAVE_DEBUG_STMTS && single_succ_p (ENTRY_BLOCK_PTR))
4442 gsi = gsi_after_labels (single_succ (ENTRY_BLOCK_PTR));
4443 gsip = &gsi;
4445 len = VEC_length (ipa_parm_adjustment_t, adjustments);
4446 for (i = 0; i < len; i++)
4448 struct ipa_parm_adjustment *adj;
4449 imm_use_iterator ui;
4450 gimple stmt, def_temp;
4451 tree name, vexpr, copy = NULL_TREE;
4452 use_operand_p use_p;
4454 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
4455 if (adj->copy_param || !is_gimple_reg (adj->base))
4456 continue;
4457 name = gimple_default_def (cfun, adj->base);
4458 vexpr = NULL;
4459 if (name)
4460 FOR_EACH_IMM_USE_STMT (stmt, ui, name)
4462 /* All other users must have been removed by
4463 ipa_sra_modify_function_body. */
4464 gcc_assert (is_gimple_debug (stmt));
4465 if (vexpr == NULL && gsip != NULL)
4467 gcc_assert (TREE_CODE (adj->base) == PARM_DECL);
4468 vexpr = make_node (DEBUG_EXPR_DECL);
4469 def_temp = gimple_build_debug_source_bind (vexpr, adj->base,
4470 NULL);
4471 DECL_ARTIFICIAL (vexpr) = 1;
4472 TREE_TYPE (vexpr) = TREE_TYPE (name);
4473 DECL_MODE (vexpr) = DECL_MODE (adj->base);
4474 gsi_insert_before (gsip, def_temp, GSI_SAME_STMT);
4476 if (vexpr)
4478 FOR_EACH_IMM_USE_ON_STMT (use_p, ui)
4479 SET_USE (use_p, vexpr);
4481 else
4482 gimple_debug_bind_reset_value (stmt);
4483 update_stmt (stmt);
4485 /* Create a VAR_DECL for debug info purposes. */
4486 if (!DECL_IGNORED_P (adj->base))
4488 copy = build_decl (DECL_SOURCE_LOCATION (current_function_decl),
4489 VAR_DECL, DECL_NAME (adj->base),
4490 TREE_TYPE (adj->base));
4491 if (DECL_PT_UID_SET_P (adj->base))
4492 SET_DECL_PT_UID (copy, DECL_PT_UID (adj->base));
4493 TREE_ADDRESSABLE (copy) = TREE_ADDRESSABLE (adj->base);
4494 TREE_READONLY (copy) = TREE_READONLY (adj->base);
4495 TREE_THIS_VOLATILE (copy) = TREE_THIS_VOLATILE (adj->base);
4496 DECL_GIMPLE_REG_P (copy) = DECL_GIMPLE_REG_P (adj->base);
4497 DECL_ARTIFICIAL (copy) = DECL_ARTIFICIAL (adj->base);
4498 DECL_IGNORED_P (copy) = DECL_IGNORED_P (adj->base);
4499 DECL_ABSTRACT_ORIGIN (copy) = DECL_ORIGIN (adj->base);
4500 DECL_SEEN_IN_BIND_EXPR_P (copy) = 1;
4501 SET_DECL_RTL (copy, 0);
4502 TREE_USED (copy) = 1;
4503 DECL_CONTEXT (copy) = current_function_decl;
4504 add_referenced_var (copy);
4505 add_local_decl (cfun, copy);
4506 DECL_CHAIN (copy) =
4507 BLOCK_VARS (DECL_INITIAL (current_function_decl));
4508 BLOCK_VARS (DECL_INITIAL (current_function_decl)) = copy;
4510 if (gsip != NULL && copy && target_for_debug_bind (adj->base))
4512 gcc_assert (TREE_CODE (adj->base) == PARM_DECL);
4513 if (vexpr)
4514 def_temp = gimple_build_debug_bind (copy, vexpr, NULL);
4515 else
4516 def_temp = gimple_build_debug_source_bind (copy, adj->base,
4517 NULL);
4518 gsi_insert_before (gsip, def_temp, GSI_SAME_STMT);
4523 /* Return false iff all callers have at least as many actual arguments as there
4524 are formal parameters in the current function. */
4526 static bool
4527 not_all_callers_have_enough_arguments_p (struct cgraph_node *node,
4528 void *data ATTRIBUTE_UNUSED)
4530 struct cgraph_edge *cs;
4531 for (cs = node->callers; cs; cs = cs->next_caller)
4532 if (!callsite_has_enough_arguments_p (cs->call_stmt))
4533 return true;
4535 return false;
4538 /* Convert all callers of NODE. */
4540 static bool
4541 convert_callers_for_node (struct cgraph_node *node,
4542 void *data)
4544 ipa_parm_adjustment_vec adjustments = (ipa_parm_adjustment_vec)data;
4545 bitmap recomputed_callers = BITMAP_ALLOC (NULL);
4546 struct cgraph_edge *cs;
4548 for (cs = node->callers; cs; cs = cs->next_caller)
4550 current_function_decl = cs->caller->decl;
4551 push_cfun (DECL_STRUCT_FUNCTION (cs->caller->decl));
4553 if (dump_file)
4554 fprintf (dump_file, "Adjusting call (%i -> %i) %s -> %s\n",
4555 cs->caller->uid, cs->callee->uid,
4556 cgraph_node_name (cs->caller),
4557 cgraph_node_name (cs->callee));
4559 ipa_modify_call_arguments (cs, cs->call_stmt, adjustments);
4561 pop_cfun ();
4564 for (cs = node->callers; cs; cs = cs->next_caller)
4565 if (bitmap_set_bit (recomputed_callers, cs->caller->uid)
4566 && gimple_in_ssa_p (DECL_STRUCT_FUNCTION (cs->caller->decl)))
4567 compute_inline_parameters (cs->caller, true);
4568 BITMAP_FREE (recomputed_callers);
4570 return true;
4573 /* Convert all callers of NODE to pass parameters as given in ADJUSTMENTS. */
4575 static void
4576 convert_callers (struct cgraph_node *node, tree old_decl,
4577 ipa_parm_adjustment_vec adjustments)
4579 tree old_cur_fndecl = current_function_decl;
4580 basic_block this_block;
4582 cgraph_for_node_and_aliases (node, convert_callers_for_node,
4583 adjustments, false);
4585 current_function_decl = old_cur_fndecl;
4587 if (!encountered_recursive_call)
4588 return;
4590 FOR_EACH_BB (this_block)
4592 gimple_stmt_iterator gsi;
4594 for (gsi = gsi_start_bb (this_block); !gsi_end_p (gsi); gsi_next (&gsi))
4596 gimple stmt = gsi_stmt (gsi);
4597 tree call_fndecl;
4598 if (gimple_code (stmt) != GIMPLE_CALL)
4599 continue;
4600 call_fndecl = gimple_call_fndecl (stmt);
4601 if (call_fndecl == old_decl)
4603 if (dump_file)
4604 fprintf (dump_file, "Adjusting recursive call");
4605 gimple_call_set_fndecl (stmt, node->decl);
4606 ipa_modify_call_arguments (NULL, stmt, adjustments);
4611 return;
4614 /* Perform all the modification required in IPA-SRA for NODE to have parameters
4615 as given in ADJUSTMENTS. Return true iff the CFG has been changed. */
4617 static bool
4618 modify_function (struct cgraph_node *node, ipa_parm_adjustment_vec adjustments)
4620 struct cgraph_node *new_node;
4621 bool cfg_changed;
4622 VEC (cgraph_edge_p, heap) * redirect_callers = collect_callers_of_node (node);
4624 rebuild_cgraph_edges ();
4625 free_dominance_info (CDI_DOMINATORS);
4626 pop_cfun ();
4627 current_function_decl = NULL_TREE;
4629 new_node = cgraph_function_versioning (node, redirect_callers, NULL, NULL,
4630 NULL, NULL, "isra");
4631 current_function_decl = new_node->decl;
4632 push_cfun (DECL_STRUCT_FUNCTION (new_node->decl));
4634 ipa_modify_formal_parameters (current_function_decl, adjustments, "ISRA");
4635 cfg_changed = ipa_sra_modify_function_body (adjustments);
4636 sra_ipa_reset_debug_stmts (adjustments);
4637 convert_callers (new_node, node->decl, adjustments);
4638 cgraph_make_node_local (new_node);
4639 return cfg_changed;
4642 /* Return false the function is apparently unsuitable for IPA-SRA based on it's
4643 attributes, return true otherwise. NODE is the cgraph node of the current
4644 function. */
4646 static bool
4647 ipa_sra_preliminary_function_checks (struct cgraph_node *node)
4649 if (!cgraph_node_can_be_local_p (node))
4651 if (dump_file)
4652 fprintf (dump_file, "Function not local to this compilation unit.\n");
4653 return false;
4656 if (!node->local.can_change_signature)
4658 if (dump_file)
4659 fprintf (dump_file, "Function can not change signature.\n");
4660 return false;
4663 if (!tree_versionable_function_p (node->decl))
4665 if (dump_file)
4666 fprintf (dump_file, "Function is not versionable.\n");
4667 return false;
4670 if (DECL_VIRTUAL_P (current_function_decl))
4672 if (dump_file)
4673 fprintf (dump_file, "Function is a virtual method.\n");
4674 return false;
4677 if ((DECL_COMDAT (node->decl) || DECL_EXTERNAL (node->decl))
4678 && inline_summary(node)->size >= MAX_INLINE_INSNS_AUTO)
4680 if (dump_file)
4681 fprintf (dump_file, "Function too big to be made truly local.\n");
4682 return false;
4685 if (!node->callers)
4687 if (dump_file)
4688 fprintf (dump_file,
4689 "Function has no callers in this compilation unit.\n");
4690 return false;
4693 if (cfun->stdarg)
4695 if (dump_file)
4696 fprintf (dump_file, "Function uses stdarg. \n");
4697 return false;
4700 if (TYPE_ATTRIBUTES (TREE_TYPE (node->decl)))
4701 return false;
4703 return true;
4706 /* Perform early interprocedural SRA. */
4708 static unsigned int
4709 ipa_early_sra (void)
4711 struct cgraph_node *node = cgraph_get_node (current_function_decl);
4712 ipa_parm_adjustment_vec adjustments;
4713 int ret = 0;
4715 if (!ipa_sra_preliminary_function_checks (node))
4716 return 0;
4718 sra_initialize ();
4719 sra_mode = SRA_MODE_EARLY_IPA;
4721 if (!find_param_candidates ())
4723 if (dump_file)
4724 fprintf (dump_file, "Function has no IPA-SRA candidates.\n");
4725 goto simple_out;
4728 if (cgraph_for_node_and_aliases (node, not_all_callers_have_enough_arguments_p,
4729 NULL, true))
4731 if (dump_file)
4732 fprintf (dump_file, "There are callers with insufficient number of "
4733 "arguments.\n");
4734 goto simple_out;
4737 bb_dereferences = XCNEWVEC (HOST_WIDE_INT,
4738 func_param_count
4739 * last_basic_block_for_function (cfun));
4740 final_bbs = BITMAP_ALLOC (NULL);
4742 scan_function ();
4743 if (encountered_apply_args)
4745 if (dump_file)
4746 fprintf (dump_file, "Function calls __builtin_apply_args().\n");
4747 goto out;
4750 if (encountered_unchangable_recursive_call)
4752 if (dump_file)
4753 fprintf (dump_file, "Function calls itself with insufficient "
4754 "number of arguments.\n");
4755 goto out;
4758 adjustments = analyze_all_param_acesses ();
4759 if (!adjustments)
4760 goto out;
4761 if (dump_file)
4762 ipa_dump_param_adjustments (dump_file, adjustments, current_function_decl);
4764 if (modify_function (node, adjustments))
4765 ret = TODO_update_ssa | TODO_cleanup_cfg;
4766 else
4767 ret = TODO_update_ssa;
4768 VEC_free (ipa_parm_adjustment_t, heap, adjustments);
4770 statistics_counter_event (cfun, "Unused parameters deleted",
4771 sra_stats.deleted_unused_parameters);
4772 statistics_counter_event (cfun, "Scalar parameters converted to by-value",
4773 sra_stats.scalar_by_ref_to_by_val);
4774 statistics_counter_event (cfun, "Aggregate parameters broken up",
4775 sra_stats.aggregate_params_reduced);
4776 statistics_counter_event (cfun, "Aggregate parameter components created",
4777 sra_stats.param_reductions_created);
4779 out:
4780 BITMAP_FREE (final_bbs);
4781 free (bb_dereferences);
4782 simple_out:
4783 sra_deinitialize ();
4784 return ret;
4787 /* Return if early ipa sra shall be performed. */
4788 static bool
4789 ipa_early_sra_gate (void)
4791 return flag_ipa_sra && dbg_cnt (eipa_sra);
4794 struct gimple_opt_pass pass_early_ipa_sra =
4797 GIMPLE_PASS,
4798 "eipa_sra", /* name */
4799 ipa_early_sra_gate, /* gate */
4800 ipa_early_sra, /* execute */
4801 NULL, /* sub */
4802 NULL, /* next */
4803 0, /* static_pass_number */
4804 TV_IPA_SRA, /* tv_id */
4805 0, /* properties_required */
4806 0, /* properties_provided */
4807 0, /* properties_destroyed */
4808 0, /* todo_flags_start */
4809 TODO_dump_cgraph /* todo_flags_finish */