2011-08-19 Vladimir Makarov <vmakarov@redhat.com>
[official-gcc.git] / gcc / tree-sra.c
blob2d97845ffe672185f73601a0c1f6ab06c073cfcd
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, model->type, 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 get_var_ann (repl);
1829 add_referenced_var (repl);
1830 if (rename)
1831 mark_sym_for_renaming (repl);
1833 if (!access->grp_partial_lhs
1834 && (TREE_CODE (access->type) == COMPLEX_TYPE
1835 || TREE_CODE (access->type) == VECTOR_TYPE))
1836 DECL_GIMPLE_REG_P (repl) = 1;
1838 DECL_SOURCE_LOCATION (repl) = DECL_SOURCE_LOCATION (access->base);
1839 DECL_ARTIFICIAL (repl) = 1;
1840 DECL_IGNORED_P (repl) = DECL_IGNORED_P (access->base);
1842 if (DECL_NAME (access->base)
1843 && !DECL_IGNORED_P (access->base)
1844 && !DECL_ARTIFICIAL (access->base))
1846 char *pretty_name = make_fancy_name (access->expr);
1847 tree debug_expr = unshare_expr (access->expr), d;
1849 DECL_NAME (repl) = get_identifier (pretty_name);
1850 obstack_free (&name_obstack, pretty_name);
1852 /* Get rid of any SSA_NAMEs embedded in debug_expr,
1853 as DECL_DEBUG_EXPR isn't considered when looking for still
1854 used SSA_NAMEs and thus they could be freed. All debug info
1855 generation cares is whether something is constant or variable
1856 and that get_ref_base_and_extent works properly on the
1857 expression. */
1858 for (d = debug_expr; handled_component_p (d); d = TREE_OPERAND (d, 0))
1859 switch (TREE_CODE (d))
1861 case ARRAY_REF:
1862 case ARRAY_RANGE_REF:
1863 if (TREE_OPERAND (d, 1)
1864 && TREE_CODE (TREE_OPERAND (d, 1)) == SSA_NAME)
1865 TREE_OPERAND (d, 1) = SSA_NAME_VAR (TREE_OPERAND (d, 1));
1866 if (TREE_OPERAND (d, 3)
1867 && TREE_CODE (TREE_OPERAND (d, 3)) == SSA_NAME)
1868 TREE_OPERAND (d, 3) = SSA_NAME_VAR (TREE_OPERAND (d, 3));
1869 /* FALLTHRU */
1870 case COMPONENT_REF:
1871 if (TREE_OPERAND (d, 2)
1872 && TREE_CODE (TREE_OPERAND (d, 2)) == SSA_NAME)
1873 TREE_OPERAND (d, 2) = SSA_NAME_VAR (TREE_OPERAND (d, 2));
1874 break;
1875 default:
1876 break;
1878 SET_DECL_DEBUG_EXPR (repl, debug_expr);
1879 DECL_DEBUG_EXPR_IS_FROM (repl) = 1;
1880 if (access->grp_no_warning)
1881 TREE_NO_WARNING (repl) = 1;
1882 else
1883 TREE_NO_WARNING (repl) = TREE_NO_WARNING (access->base);
1885 else
1886 TREE_NO_WARNING (repl) = 1;
1888 if (dump_file)
1890 fprintf (dump_file, "Created a replacement for ");
1891 print_generic_expr (dump_file, access->base, 0);
1892 fprintf (dump_file, " offset: %u, size: %u: ",
1893 (unsigned) access->offset, (unsigned) access->size);
1894 print_generic_expr (dump_file, repl, 0);
1895 fprintf (dump_file, "\n");
1897 sra_stats.replacements++;
1899 return repl;
1902 /* Return ACCESS scalar replacement, create it if it does not exist yet. */
1904 static inline tree
1905 get_access_replacement (struct access *access)
1907 gcc_assert (access->grp_to_be_replaced);
1909 if (!access->replacement_decl)
1910 access->replacement_decl = create_access_replacement (access, true);
1911 return access->replacement_decl;
1914 /* Return ACCESS scalar replacement, create it if it does not exist yet but do
1915 not mark it for renaming. */
1917 static inline tree
1918 get_unrenamed_access_replacement (struct access *access)
1920 gcc_assert (!access->grp_to_be_replaced);
1922 if (!access->replacement_decl)
1923 access->replacement_decl = create_access_replacement (access, false);
1924 return access->replacement_decl;
1928 /* Build a subtree of accesses rooted in *ACCESS, and move the pointer in the
1929 linked list along the way. Stop when *ACCESS is NULL or the access pointed
1930 to it is not "within" the root. Return false iff some accesses partially
1931 overlap. */
1933 static bool
1934 build_access_subtree (struct access **access)
1936 struct access *root = *access, *last_child = NULL;
1937 HOST_WIDE_INT limit = root->offset + root->size;
1939 *access = (*access)->next_grp;
1940 while (*access && (*access)->offset + (*access)->size <= limit)
1942 if (!last_child)
1943 root->first_child = *access;
1944 else
1945 last_child->next_sibling = *access;
1946 last_child = *access;
1948 if (!build_access_subtree (access))
1949 return false;
1952 if (*access && (*access)->offset < limit)
1953 return false;
1955 return true;
1958 /* Build a tree of access representatives, ACCESS is the pointer to the first
1959 one, others are linked in a list by the next_grp field. Return false iff
1960 some accesses partially overlap. */
1962 static bool
1963 build_access_trees (struct access *access)
1965 while (access)
1967 struct access *root = access;
1969 if (!build_access_subtree (&access))
1970 return false;
1971 root->next_grp = access;
1973 return true;
1976 /* Return true if expr contains some ARRAY_REFs into a variable bounded
1977 array. */
1979 static bool
1980 expr_with_var_bounded_array_refs_p (tree expr)
1982 while (handled_component_p (expr))
1984 if (TREE_CODE (expr) == ARRAY_REF
1985 && !host_integerp (array_ref_low_bound (expr), 0))
1986 return true;
1987 expr = TREE_OPERAND (expr, 0);
1989 return false;
1992 /* Analyze the subtree of accesses rooted in ROOT, scheduling replacements when
1993 both seeming beneficial and when ALLOW_REPLACEMENTS allows it. Also set all
1994 sorts of access flags appropriately along the way, notably always set
1995 grp_read and grp_assign_read according to MARK_READ and grp_write when
1996 MARK_WRITE is true.
1998 Creating a replacement for a scalar access is considered beneficial if its
1999 grp_hint is set (this means we are either attempting total scalarization or
2000 there is more than one direct read access) or according to the following
2001 table:
2003 Access written to through a scalar type (once or more times)
2005 | Written to in an assignment statement
2007 | | Access read as scalar _once_
2008 | | |
2009 | | | Read in an assignment statement
2010 | | | |
2011 | | | | Scalarize Comment
2012 -----------------------------------------------------------------------------
2013 0 0 0 0 No access for the scalar
2014 0 0 0 1 No access for the scalar
2015 0 0 1 0 No Single read - won't help
2016 0 0 1 1 No The same case
2017 0 1 0 0 No access for the scalar
2018 0 1 0 1 No access for the scalar
2019 0 1 1 0 Yes s = *g; return s.i;
2020 0 1 1 1 Yes The same case as above
2021 1 0 0 0 No Won't help
2022 1 0 0 1 Yes s.i = 1; *g = s;
2023 1 0 1 0 Yes s.i = 5; g = s.i;
2024 1 0 1 1 Yes The same case as above
2025 1 1 0 0 No Won't help.
2026 1 1 0 1 Yes s.i = 1; *g = s;
2027 1 1 1 0 Yes s = *g; return s.i;
2028 1 1 1 1 Yes Any of the above yeses */
2030 static bool
2031 analyze_access_subtree (struct access *root, struct access *parent,
2032 bool allow_replacements)
2034 struct access *child;
2035 HOST_WIDE_INT limit = root->offset + root->size;
2036 HOST_WIDE_INT covered_to = root->offset;
2037 bool scalar = is_gimple_reg_type (root->type);
2038 bool hole = false, sth_created = false;
2040 if (parent)
2042 if (parent->grp_read)
2043 root->grp_read = 1;
2044 if (parent->grp_assignment_read)
2045 root->grp_assignment_read = 1;
2046 if (parent->grp_write)
2047 root->grp_write = 1;
2048 if (parent->grp_assignment_write)
2049 root->grp_assignment_write = 1;
2050 if (parent->grp_total_scalarization)
2051 root->grp_total_scalarization = 1;
2054 if (root->grp_unscalarizable_region)
2055 allow_replacements = false;
2057 if (allow_replacements && expr_with_var_bounded_array_refs_p (root->expr))
2058 allow_replacements = false;
2060 for (child = root->first_child; child; child = child->next_sibling)
2062 hole |= covered_to < child->offset;
2063 sth_created |= analyze_access_subtree (child, root,
2064 allow_replacements && !scalar);
2066 root->grp_unscalarized_data |= child->grp_unscalarized_data;
2067 root->grp_total_scalarization &= child->grp_total_scalarization;
2068 if (child->grp_covered)
2069 covered_to += child->size;
2070 else
2071 hole = true;
2074 if (allow_replacements && scalar && !root->first_child
2075 && (root->grp_hint
2076 || ((root->grp_scalar_read || root->grp_assignment_read)
2077 && (root->grp_scalar_write || root->grp_assignment_write))))
2079 if (dump_file && (dump_flags & TDF_DETAILS))
2081 fprintf (dump_file, "Marking ");
2082 print_generic_expr (dump_file, root->base, 0);
2083 fprintf (dump_file, " offset: %u, size: %u: ",
2084 (unsigned) root->offset, (unsigned) root->size);
2085 fprintf (dump_file, " to be replaced.\n");
2088 root->grp_to_be_replaced = 1;
2089 sth_created = true;
2090 hole = false;
2092 else
2094 if (covered_to < limit)
2095 hole = true;
2096 if (scalar)
2097 root->grp_total_scalarization = 0;
2100 if (sth_created
2101 && (!hole || root->grp_total_scalarization))
2103 root->grp_covered = 1;
2104 return true;
2106 if (root->grp_write || TREE_CODE (root->base) == PARM_DECL)
2107 root->grp_unscalarized_data = 1; /* not covered and written to */
2108 if (sth_created)
2109 return true;
2110 return false;
2113 /* Analyze all access trees linked by next_grp by the means of
2114 analyze_access_subtree. */
2115 static bool
2116 analyze_access_trees (struct access *access)
2118 bool ret = false;
2120 while (access)
2122 if (analyze_access_subtree (access, NULL, true))
2123 ret = true;
2124 access = access->next_grp;
2127 return ret;
2130 /* Return true iff a potential new child of LACC at offset OFFSET and with size
2131 SIZE would conflict with an already existing one. If exactly such a child
2132 already exists in LACC, store a pointer to it in EXACT_MATCH. */
2134 static bool
2135 child_would_conflict_in_lacc (struct access *lacc, HOST_WIDE_INT norm_offset,
2136 HOST_WIDE_INT size, struct access **exact_match)
2138 struct access *child;
2140 for (child = lacc->first_child; child; child = child->next_sibling)
2142 if (child->offset == norm_offset && child->size == size)
2144 *exact_match = child;
2145 return true;
2148 if (child->offset < norm_offset + size
2149 && child->offset + child->size > norm_offset)
2150 return true;
2153 return false;
2156 /* Create a new child access of PARENT, with all properties just like MODEL
2157 except for its offset and with its grp_write false and grp_read true.
2158 Return the new access or NULL if it cannot be created. Note that this access
2159 is created long after all splicing and sorting, it's not located in any
2160 access vector and is automatically a representative of its group. */
2162 static struct access *
2163 create_artificial_child_access (struct access *parent, struct access *model,
2164 HOST_WIDE_INT new_offset)
2166 struct access *access;
2167 struct access **child;
2168 tree expr = parent->base;
2170 gcc_assert (!model->grp_unscalarizable_region);
2172 access = (struct access *) pool_alloc (access_pool);
2173 memset (access, 0, sizeof (struct access));
2174 if (!build_user_friendly_ref_for_offset (&expr, TREE_TYPE (expr), new_offset,
2175 model->type))
2177 access->grp_no_warning = true;
2178 expr = build_ref_for_model (EXPR_LOCATION (parent->base), parent->base,
2179 new_offset, model, NULL, false);
2182 access->base = parent->base;
2183 access->expr = expr;
2184 access->offset = new_offset;
2185 access->size = model->size;
2186 access->type = model->type;
2187 access->grp_write = true;
2188 access->grp_read = false;
2190 child = &parent->first_child;
2191 while (*child && (*child)->offset < new_offset)
2192 child = &(*child)->next_sibling;
2194 access->next_sibling = *child;
2195 *child = access;
2197 return access;
2201 /* Propagate all subaccesses of RACC across an assignment link to LACC. Return
2202 true if any new subaccess was created. Additionally, if RACC is a scalar
2203 access but LACC is not, change the type of the latter, if possible. */
2205 static bool
2206 propagate_subaccesses_across_link (struct access *lacc, struct access *racc)
2208 struct access *rchild;
2209 HOST_WIDE_INT norm_delta = lacc->offset - racc->offset;
2210 bool ret = false;
2212 if (is_gimple_reg_type (lacc->type)
2213 || lacc->grp_unscalarizable_region
2214 || racc->grp_unscalarizable_region)
2215 return false;
2217 if (!lacc->first_child && !racc->first_child
2218 && is_gimple_reg_type (racc->type))
2220 tree t = lacc->base;
2222 lacc->type = racc->type;
2223 if (build_user_friendly_ref_for_offset (&t, TREE_TYPE (t), lacc->offset,
2224 racc->type))
2225 lacc->expr = t;
2226 else
2228 lacc->expr = build_ref_for_model (EXPR_LOCATION (lacc->base),
2229 lacc->base, lacc->offset,
2230 racc, NULL, false);
2231 lacc->grp_no_warning = true;
2233 return false;
2236 for (rchild = racc->first_child; rchild; rchild = rchild->next_sibling)
2238 struct access *new_acc = NULL;
2239 HOST_WIDE_INT norm_offset = rchild->offset + norm_delta;
2241 if (rchild->grp_unscalarizable_region)
2242 continue;
2244 if (child_would_conflict_in_lacc (lacc, norm_offset, rchild->size,
2245 &new_acc))
2247 if (new_acc)
2249 rchild->grp_hint = 1;
2250 new_acc->grp_hint |= new_acc->grp_read;
2251 if (rchild->first_child)
2252 ret |= propagate_subaccesses_across_link (new_acc, rchild);
2254 continue;
2257 rchild->grp_hint = 1;
2258 new_acc = create_artificial_child_access (lacc, rchild, norm_offset);
2259 if (new_acc)
2261 ret = true;
2262 if (racc->first_child)
2263 propagate_subaccesses_across_link (new_acc, rchild);
2267 return ret;
2270 /* Propagate all subaccesses across assignment links. */
2272 static void
2273 propagate_all_subaccesses (void)
2275 while (work_queue_head)
2277 struct access *racc = pop_access_from_work_queue ();
2278 struct assign_link *link;
2280 gcc_assert (racc->first_link);
2282 for (link = racc->first_link; link; link = link->next)
2284 struct access *lacc = link->lacc;
2286 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (lacc->base)))
2287 continue;
2288 lacc = lacc->group_representative;
2289 if (propagate_subaccesses_across_link (lacc, racc)
2290 && lacc->first_link)
2291 add_access_to_work_queue (lacc);
2296 /* Go through all accesses collected throughout the (intraprocedural) analysis
2297 stage, exclude overlapping ones, identify representatives and build trees
2298 out of them, making decisions about scalarization on the way. Return true
2299 iff there are any to-be-scalarized variables after this stage. */
2301 static bool
2302 analyze_all_variable_accesses (void)
2304 int res = 0;
2305 bitmap tmp = BITMAP_ALLOC (NULL);
2306 bitmap_iterator bi;
2307 unsigned i, max_total_scalarization_size;
2309 max_total_scalarization_size = UNITS_PER_WORD * BITS_PER_UNIT
2310 * MOVE_RATIO (optimize_function_for_speed_p (cfun));
2312 EXECUTE_IF_SET_IN_BITMAP (candidate_bitmap, 0, i, bi)
2313 if (bitmap_bit_p (should_scalarize_away_bitmap, i)
2314 && !bitmap_bit_p (cannot_scalarize_away_bitmap, i))
2316 tree var = referenced_var (i);
2318 if (TREE_CODE (var) == VAR_DECL
2319 && type_consists_of_records_p (TREE_TYPE (var)))
2321 if ((unsigned) tree_low_cst (TYPE_SIZE (TREE_TYPE (var)), 1)
2322 <= max_total_scalarization_size)
2324 completely_scalarize_var (var);
2325 if (dump_file && (dump_flags & TDF_DETAILS))
2327 fprintf (dump_file, "Will attempt to totally scalarize ");
2328 print_generic_expr (dump_file, var, 0);
2329 fprintf (dump_file, " (UID: %u): \n", DECL_UID (var));
2332 else if (dump_file && (dump_flags & TDF_DETAILS))
2334 fprintf (dump_file, "Too big to totally scalarize: ");
2335 print_generic_expr (dump_file, var, 0);
2336 fprintf (dump_file, " (UID: %u)\n", DECL_UID (var));
2341 bitmap_copy (tmp, candidate_bitmap);
2342 EXECUTE_IF_SET_IN_BITMAP (tmp, 0, i, bi)
2344 tree var = referenced_var (i);
2345 struct access *access;
2347 access = sort_and_splice_var_accesses (var);
2348 if (!access || !build_access_trees (access))
2349 disqualify_candidate (var,
2350 "No or inhibitingly overlapping accesses.");
2353 propagate_all_subaccesses ();
2355 bitmap_copy (tmp, candidate_bitmap);
2356 EXECUTE_IF_SET_IN_BITMAP (tmp, 0, i, bi)
2358 tree var = referenced_var (i);
2359 struct access *access = get_first_repr_for_decl (var);
2361 if (analyze_access_trees (access))
2363 res++;
2364 if (dump_file && (dump_flags & TDF_DETAILS))
2366 fprintf (dump_file, "\nAccess trees for ");
2367 print_generic_expr (dump_file, var, 0);
2368 fprintf (dump_file, " (UID: %u): \n", DECL_UID (var));
2369 dump_access_tree (dump_file, access);
2370 fprintf (dump_file, "\n");
2373 else
2374 disqualify_candidate (var, "No scalar replacements to be created.");
2377 BITMAP_FREE (tmp);
2379 if (res)
2381 statistics_counter_event (cfun, "Scalarized aggregates", res);
2382 return true;
2384 else
2385 return false;
2388 /* Generate statements copying scalar replacements of accesses within a subtree
2389 into or out of AGG. ACCESS, all its children, siblings and their children
2390 are to be processed. AGG is an aggregate type expression (can be a
2391 declaration but does not have to be, it can for example also be a mem_ref or
2392 a series of handled components). TOP_OFFSET is the offset of the processed
2393 subtree which has to be subtracted from offsets of individual accesses to
2394 get corresponding offsets for AGG. If CHUNK_SIZE is non-null, copy only
2395 replacements in the interval <start_offset, start_offset + chunk_size>,
2396 otherwise copy all. GSI is a statement iterator used to place the new
2397 statements. WRITE should be true when the statements should write from AGG
2398 to the replacement and false if vice versa. if INSERT_AFTER is true, new
2399 statements will be added after the current statement in GSI, they will be
2400 added before the statement otherwise. */
2402 static void
2403 generate_subtree_copies (struct access *access, tree agg,
2404 HOST_WIDE_INT top_offset,
2405 HOST_WIDE_INT start_offset, HOST_WIDE_INT chunk_size,
2406 gimple_stmt_iterator *gsi, bool write,
2407 bool insert_after, location_t loc)
2411 if (chunk_size && access->offset >= start_offset + chunk_size)
2412 return;
2414 if (access->grp_to_be_replaced
2415 && (chunk_size == 0
2416 || access->offset + access->size > start_offset))
2418 tree expr, repl = get_access_replacement (access);
2419 gimple stmt;
2421 expr = build_ref_for_model (loc, agg, access->offset - top_offset,
2422 access, gsi, insert_after);
2424 if (write)
2426 if (access->grp_partial_lhs)
2427 expr = force_gimple_operand_gsi (gsi, expr, true, NULL_TREE,
2428 !insert_after,
2429 insert_after ? GSI_NEW_STMT
2430 : GSI_SAME_STMT);
2431 stmt = gimple_build_assign (repl, expr);
2433 else
2435 TREE_NO_WARNING (repl) = 1;
2436 if (access->grp_partial_lhs)
2437 repl = force_gimple_operand_gsi (gsi, repl, true, NULL_TREE,
2438 !insert_after,
2439 insert_after ? GSI_NEW_STMT
2440 : GSI_SAME_STMT);
2441 stmt = gimple_build_assign (expr, repl);
2443 gimple_set_location (stmt, loc);
2445 if (insert_after)
2446 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2447 else
2448 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2449 update_stmt (stmt);
2450 sra_stats.subtree_copies++;
2453 if (access->first_child)
2454 generate_subtree_copies (access->first_child, agg, top_offset,
2455 start_offset, chunk_size, gsi,
2456 write, insert_after, loc);
2458 access = access->next_sibling;
2460 while (access);
2463 /* Assign zero to all scalar replacements in an access subtree. ACCESS is the
2464 the root of the subtree to be processed. GSI is the statement iterator used
2465 for inserting statements which are added after the current statement if
2466 INSERT_AFTER is true or before it otherwise. */
2468 static void
2469 init_subtree_with_zero (struct access *access, gimple_stmt_iterator *gsi,
2470 bool insert_after, location_t loc)
2473 struct access *child;
2475 if (access->grp_to_be_replaced)
2477 gimple stmt;
2479 stmt = gimple_build_assign (get_access_replacement (access),
2480 build_zero_cst (access->type));
2481 if (insert_after)
2482 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2483 else
2484 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2485 update_stmt (stmt);
2486 gimple_set_location (stmt, loc);
2489 for (child = access->first_child; child; child = child->next_sibling)
2490 init_subtree_with_zero (child, gsi, insert_after, loc);
2493 /* Search for an access representative for the given expression EXPR and
2494 return it or NULL if it cannot be found. */
2496 static struct access *
2497 get_access_for_expr (tree expr)
2499 HOST_WIDE_INT offset, size, max_size;
2500 tree base;
2502 /* FIXME: This should not be necessary but Ada produces V_C_Es with a type of
2503 a different size than the size of its argument and we need the latter
2504 one. */
2505 if (TREE_CODE (expr) == VIEW_CONVERT_EXPR)
2506 expr = TREE_OPERAND (expr, 0);
2508 base = get_ref_base_and_extent (expr, &offset, &size, &max_size);
2509 if (max_size == -1 || !DECL_P (base))
2510 return NULL;
2512 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (base)))
2513 return NULL;
2515 return get_var_base_offset_size_access (base, offset, max_size);
2518 /* Replace the expression EXPR with a scalar replacement if there is one and
2519 generate other statements to do type conversion or subtree copying if
2520 necessary. GSI is used to place newly created statements, WRITE is true if
2521 the expression is being written to (it is on a LHS of a statement or output
2522 in an assembly statement). */
2524 static bool
2525 sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write)
2527 location_t loc;
2528 struct access *access;
2529 tree type, bfr;
2531 if (TREE_CODE (*expr) == BIT_FIELD_REF)
2533 bfr = *expr;
2534 expr = &TREE_OPERAND (*expr, 0);
2536 else
2537 bfr = NULL_TREE;
2539 if (TREE_CODE (*expr) == REALPART_EXPR || TREE_CODE (*expr) == IMAGPART_EXPR)
2540 expr = &TREE_OPERAND (*expr, 0);
2541 access = get_access_for_expr (*expr);
2542 if (!access)
2543 return false;
2544 type = TREE_TYPE (*expr);
2546 loc = gimple_location (gsi_stmt (*gsi));
2547 if (access->grp_to_be_replaced)
2549 tree repl = get_access_replacement (access);
2550 /* If we replace a non-register typed access simply use the original
2551 access expression to extract the scalar component afterwards.
2552 This happens if scalarizing a function return value or parameter
2553 like in gcc.c-torture/execute/20041124-1.c, 20050316-1.c and
2554 gcc.c-torture/compile/20011217-1.c.
2556 We also want to use this when accessing a complex or vector which can
2557 be accessed as a different type too, potentially creating a need for
2558 type conversion (see PR42196) and when scalarized unions are involved
2559 in assembler statements (see PR42398). */
2560 if (!useless_type_conversion_p (type, access->type))
2562 tree ref;
2564 ref = build_ref_for_model (loc, access->base, access->offset, access,
2565 NULL, false);
2567 if (write)
2569 gimple stmt;
2571 if (access->grp_partial_lhs)
2572 ref = force_gimple_operand_gsi (gsi, ref, true, NULL_TREE,
2573 false, GSI_NEW_STMT);
2574 stmt = gimple_build_assign (repl, ref);
2575 gimple_set_location (stmt, loc);
2576 gsi_insert_after (gsi, stmt, GSI_NEW_STMT);
2578 else
2580 gimple stmt;
2582 if (access->grp_partial_lhs)
2583 repl = force_gimple_operand_gsi (gsi, repl, true, NULL_TREE,
2584 true, GSI_SAME_STMT);
2585 stmt = gimple_build_assign (ref, repl);
2586 gimple_set_location (stmt, loc);
2587 gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
2590 else
2591 *expr = repl;
2592 sra_stats.exprs++;
2595 if (access->first_child)
2597 HOST_WIDE_INT start_offset, chunk_size;
2598 if (bfr
2599 && host_integerp (TREE_OPERAND (bfr, 1), 1)
2600 && host_integerp (TREE_OPERAND (bfr, 2), 1))
2602 chunk_size = tree_low_cst (TREE_OPERAND (bfr, 1), 1);
2603 start_offset = access->offset
2604 + tree_low_cst (TREE_OPERAND (bfr, 2), 1);
2606 else
2607 start_offset = chunk_size = 0;
2609 generate_subtree_copies (access->first_child, access->base, 0,
2610 start_offset, chunk_size, gsi, write, write,
2611 loc);
2613 return true;
2616 /* Where scalar replacements of the RHS have been written to when a replacement
2617 of a LHS of an assigments cannot be direclty loaded from a replacement of
2618 the RHS. */
2619 enum unscalarized_data_handling { SRA_UDH_NONE, /* Nothing done so far. */
2620 SRA_UDH_RIGHT, /* Data flushed to the RHS. */
2621 SRA_UDH_LEFT }; /* Data flushed to the LHS. */
2623 /* Store all replacements in the access tree rooted in TOP_RACC either to their
2624 base aggregate if there are unscalarized data or directly to LHS of the
2625 statement that is pointed to by GSI otherwise. */
2627 static enum unscalarized_data_handling
2628 handle_unscalarized_data_in_subtree (struct access *top_racc,
2629 gimple_stmt_iterator *gsi)
2631 if (top_racc->grp_unscalarized_data)
2633 generate_subtree_copies (top_racc->first_child, top_racc->base, 0, 0, 0,
2634 gsi, false, false,
2635 gimple_location (gsi_stmt (*gsi)));
2636 return SRA_UDH_RIGHT;
2638 else
2640 tree lhs = gimple_assign_lhs (gsi_stmt (*gsi));
2641 generate_subtree_copies (top_racc->first_child, lhs, top_racc->offset,
2642 0, 0, gsi, false, false,
2643 gimple_location (gsi_stmt (*gsi)));
2644 return SRA_UDH_LEFT;
2649 /* Try to generate statements to load all sub-replacements in an access subtree
2650 formed by children of LACC from scalar replacements in the TOP_RACC subtree.
2651 If that is not possible, refresh the TOP_RACC base aggregate and load the
2652 accesses from it. LEFT_OFFSET is the offset of the left whole subtree being
2653 copied. NEW_GSI is stmt iterator used for statement insertions after the
2654 original assignment, OLD_GSI is used to insert statements before the
2655 assignment. *REFRESHED keeps the information whether we have needed to
2656 refresh replacements of the LHS and from which side of the assignments this
2657 takes place. */
2659 static void
2660 load_assign_lhs_subreplacements (struct access *lacc, struct access *top_racc,
2661 HOST_WIDE_INT left_offset,
2662 gimple_stmt_iterator *old_gsi,
2663 gimple_stmt_iterator *new_gsi,
2664 enum unscalarized_data_handling *refreshed)
2666 location_t loc = gimple_location (gsi_stmt (*old_gsi));
2667 for (lacc = lacc->first_child; lacc; lacc = lacc->next_sibling)
2669 if (lacc->grp_to_be_replaced)
2671 struct access *racc;
2672 HOST_WIDE_INT offset = lacc->offset - left_offset + top_racc->offset;
2673 gimple stmt;
2674 tree rhs;
2676 racc = find_access_in_subtree (top_racc, offset, lacc->size);
2677 if (racc && racc->grp_to_be_replaced)
2679 rhs = get_access_replacement (racc);
2680 if (!useless_type_conversion_p (lacc->type, racc->type))
2681 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR, lacc->type, rhs);
2683 else
2685 /* No suitable access on the right hand side, need to load from
2686 the aggregate. See if we have to update it first... */
2687 if (*refreshed == SRA_UDH_NONE)
2688 *refreshed = handle_unscalarized_data_in_subtree (top_racc,
2689 old_gsi);
2691 if (*refreshed == SRA_UDH_LEFT)
2692 rhs = build_ref_for_model (loc, lacc->base, lacc->offset, lacc,
2693 new_gsi, true);
2694 else
2695 rhs = build_ref_for_model (loc, top_racc->base, offset, lacc,
2696 new_gsi, true);
2699 stmt = gimple_build_assign (get_access_replacement (lacc), rhs);
2700 gsi_insert_after (new_gsi, stmt, GSI_NEW_STMT);
2701 gimple_set_location (stmt, loc);
2702 update_stmt (stmt);
2703 sra_stats.subreplacements++;
2705 else if (*refreshed == SRA_UDH_NONE
2706 && lacc->grp_read && !lacc->grp_covered)
2707 *refreshed = handle_unscalarized_data_in_subtree (top_racc,
2708 old_gsi);
2710 if (lacc->first_child)
2711 load_assign_lhs_subreplacements (lacc, top_racc, left_offset,
2712 old_gsi, new_gsi, refreshed);
2716 /* Result code for SRA assignment modification. */
2717 enum assignment_mod_result { SRA_AM_NONE, /* nothing done for the stmt */
2718 SRA_AM_MODIFIED, /* stmt changed but not
2719 removed */
2720 SRA_AM_REMOVED }; /* stmt eliminated */
2722 /* Modify assignments with a CONSTRUCTOR on their RHS. STMT contains a pointer
2723 to the assignment and GSI is the statement iterator pointing at it. Returns
2724 the same values as sra_modify_assign. */
2726 static enum assignment_mod_result
2727 sra_modify_constructor_assign (gimple *stmt, gimple_stmt_iterator *gsi)
2729 tree lhs = gimple_assign_lhs (*stmt);
2730 struct access *acc;
2731 location_t loc;
2733 acc = get_access_for_expr (lhs);
2734 if (!acc)
2735 return SRA_AM_NONE;
2737 loc = gimple_location (*stmt);
2738 if (VEC_length (constructor_elt,
2739 CONSTRUCTOR_ELTS (gimple_assign_rhs1 (*stmt))) > 0)
2741 /* I have never seen this code path trigger but if it can happen the
2742 following should handle it gracefully. */
2743 if (access_has_children_p (acc))
2744 generate_subtree_copies (acc->first_child, acc->base, 0, 0, 0, gsi,
2745 true, true, loc);
2746 return SRA_AM_MODIFIED;
2749 if (acc->grp_covered)
2751 init_subtree_with_zero (acc, gsi, false, loc);
2752 unlink_stmt_vdef (*stmt);
2753 gsi_remove (gsi, true);
2754 return SRA_AM_REMOVED;
2756 else
2758 init_subtree_with_zero (acc, gsi, true, loc);
2759 return SRA_AM_MODIFIED;
2763 /* Create and return a new suitable default definition SSA_NAME for RACC which
2764 is an access describing an uninitialized part of an aggregate that is being
2765 loaded. */
2767 static tree
2768 get_repl_default_def_ssa_name (struct access *racc)
2770 tree repl, decl;
2772 decl = get_unrenamed_access_replacement (racc);
2774 repl = gimple_default_def (cfun, decl);
2775 if (!repl)
2777 repl = make_ssa_name (decl, gimple_build_nop ());
2778 set_default_def (decl, repl);
2781 return repl;
2784 /* Return true if REF has a COMPONENT_REF with a bit-field field declaration
2785 somewhere in it. */
2787 static inline bool
2788 contains_bitfld_comp_ref_p (const_tree ref)
2790 while (handled_component_p (ref))
2792 if (TREE_CODE (ref) == COMPONENT_REF
2793 && DECL_BIT_FIELD (TREE_OPERAND (ref, 1)))
2794 return true;
2795 ref = TREE_OPERAND (ref, 0);
2798 return false;
2801 /* Return true if REF has an VIEW_CONVERT_EXPR or a COMPONENT_REF with a
2802 bit-field field declaration somewhere in it. */
2804 static inline bool
2805 contains_vce_or_bfcref_p (const_tree ref)
2807 while (handled_component_p (ref))
2809 if (TREE_CODE (ref) == VIEW_CONVERT_EXPR
2810 || (TREE_CODE (ref) == COMPONENT_REF
2811 && DECL_BIT_FIELD (TREE_OPERAND (ref, 1))))
2812 return true;
2813 ref = TREE_OPERAND (ref, 0);
2816 return false;
2819 /* Examine both sides of the assignment statement pointed to by STMT, replace
2820 them with a scalare replacement if there is one and generate copying of
2821 replacements if scalarized aggregates have been used in the assignment. GSI
2822 is used to hold generated statements for type conversions and subtree
2823 copying. */
2825 static enum assignment_mod_result
2826 sra_modify_assign (gimple *stmt, gimple_stmt_iterator *gsi)
2828 struct access *lacc, *racc;
2829 tree lhs, rhs;
2830 bool modify_this_stmt = false;
2831 bool force_gimple_rhs = false;
2832 location_t loc;
2833 gimple_stmt_iterator orig_gsi = *gsi;
2835 if (!gimple_assign_single_p (*stmt))
2836 return SRA_AM_NONE;
2837 lhs = gimple_assign_lhs (*stmt);
2838 rhs = gimple_assign_rhs1 (*stmt);
2840 if (TREE_CODE (rhs) == CONSTRUCTOR)
2841 return sra_modify_constructor_assign (stmt, gsi);
2843 if (TREE_CODE (rhs) == REALPART_EXPR || TREE_CODE (lhs) == REALPART_EXPR
2844 || TREE_CODE (rhs) == IMAGPART_EXPR || TREE_CODE (lhs) == IMAGPART_EXPR
2845 || TREE_CODE (rhs) == BIT_FIELD_REF || TREE_CODE (lhs) == BIT_FIELD_REF)
2847 modify_this_stmt = sra_modify_expr (gimple_assign_rhs1_ptr (*stmt),
2848 gsi, false);
2849 modify_this_stmt |= sra_modify_expr (gimple_assign_lhs_ptr (*stmt),
2850 gsi, true);
2851 return modify_this_stmt ? SRA_AM_MODIFIED : SRA_AM_NONE;
2854 lacc = get_access_for_expr (lhs);
2855 racc = get_access_for_expr (rhs);
2856 if (!lacc && !racc)
2857 return SRA_AM_NONE;
2859 loc = gimple_location (*stmt);
2860 if (lacc && lacc->grp_to_be_replaced)
2862 lhs = get_access_replacement (lacc);
2863 gimple_assign_set_lhs (*stmt, lhs);
2864 modify_this_stmt = true;
2865 if (lacc->grp_partial_lhs)
2866 force_gimple_rhs = true;
2867 sra_stats.exprs++;
2870 if (racc && racc->grp_to_be_replaced)
2872 rhs = get_access_replacement (racc);
2873 modify_this_stmt = true;
2874 if (racc->grp_partial_lhs)
2875 force_gimple_rhs = true;
2876 sra_stats.exprs++;
2879 if (modify_this_stmt)
2881 if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
2883 /* If we can avoid creating a VIEW_CONVERT_EXPR do so.
2884 ??? This should move to fold_stmt which we simply should
2885 call after building a VIEW_CONVERT_EXPR here. */
2886 if (AGGREGATE_TYPE_P (TREE_TYPE (lhs))
2887 && !contains_bitfld_comp_ref_p (lhs)
2888 && !access_has_children_p (lacc))
2890 lhs = build_ref_for_model (loc, lhs, 0, racc, gsi, false);
2891 gimple_assign_set_lhs (*stmt, lhs);
2893 else if (AGGREGATE_TYPE_P (TREE_TYPE (rhs))
2894 && !contains_vce_or_bfcref_p (rhs)
2895 && !access_has_children_p (racc))
2896 rhs = build_ref_for_model (loc, rhs, 0, lacc, gsi, false);
2898 if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
2900 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR, TREE_TYPE (lhs),
2901 rhs);
2902 if (is_gimple_reg_type (TREE_TYPE (lhs))
2903 && TREE_CODE (lhs) != SSA_NAME)
2904 force_gimple_rhs = true;
2909 /* From this point on, the function deals with assignments in between
2910 aggregates when at least one has scalar reductions of some of its
2911 components. There are three possible scenarios: Both the LHS and RHS have
2912 to-be-scalarized components, 2) only the RHS has or 3) only the LHS has.
2914 In the first case, we would like to load the LHS components from RHS
2915 components whenever possible. If that is not possible, we would like to
2916 read it directly from the RHS (after updating it by storing in it its own
2917 components). If there are some necessary unscalarized data in the LHS,
2918 those will be loaded by the original assignment too. If neither of these
2919 cases happen, the original statement can be removed. Most of this is done
2920 by load_assign_lhs_subreplacements.
2922 In the second case, we would like to store all RHS scalarized components
2923 directly into LHS and if they cover the aggregate completely, remove the
2924 statement too. In the third case, we want the LHS components to be loaded
2925 directly from the RHS (DSE will remove the original statement if it
2926 becomes redundant).
2928 This is a bit complex but manageable when types match and when unions do
2929 not cause confusion in a way that we cannot really load a component of LHS
2930 from the RHS or vice versa (the access representing this level can have
2931 subaccesses that are accessible only through a different union field at a
2932 higher level - different from the one used in the examined expression).
2933 Unions are fun.
2935 Therefore, I specially handle a fourth case, happening when there is a
2936 specific type cast or it is impossible to locate a scalarized subaccess on
2937 the other side of the expression. If that happens, I simply "refresh" the
2938 RHS by storing in it is scalarized components leave the original statement
2939 there to do the copying and then load the scalar replacements of the LHS.
2940 This is what the first branch does. */
2942 if (modify_this_stmt
2943 || gimple_has_volatile_ops (*stmt)
2944 || contains_vce_or_bfcref_p (rhs)
2945 || contains_vce_or_bfcref_p (lhs))
2947 if (access_has_children_p (racc))
2948 generate_subtree_copies (racc->first_child, racc->base, 0, 0, 0,
2949 gsi, false, false, loc);
2950 if (access_has_children_p (lacc))
2951 generate_subtree_copies (lacc->first_child, lacc->base, 0, 0, 0,
2952 gsi, true, true, loc);
2953 sra_stats.separate_lhs_rhs_handling++;
2955 else
2957 if (access_has_children_p (lacc) && access_has_children_p (racc))
2959 gimple_stmt_iterator orig_gsi = *gsi;
2960 enum unscalarized_data_handling refreshed;
2962 if (lacc->grp_read && !lacc->grp_covered)
2963 refreshed = handle_unscalarized_data_in_subtree (racc, gsi);
2964 else
2965 refreshed = SRA_UDH_NONE;
2967 load_assign_lhs_subreplacements (lacc, racc, lacc->offset,
2968 &orig_gsi, gsi, &refreshed);
2969 if (refreshed != SRA_UDH_RIGHT)
2971 gsi_next (gsi);
2972 unlink_stmt_vdef (*stmt);
2973 gsi_remove (&orig_gsi, true);
2974 sra_stats.deleted++;
2975 return SRA_AM_REMOVED;
2978 else
2980 if (racc)
2982 if (!racc->grp_to_be_replaced && !racc->grp_unscalarized_data)
2984 if (dump_file)
2986 fprintf (dump_file, "Removing load: ");
2987 print_gimple_stmt (dump_file, *stmt, 0, 0);
2990 if (TREE_CODE (lhs) == SSA_NAME)
2992 rhs = get_repl_default_def_ssa_name (racc);
2993 if (!useless_type_conversion_p (TREE_TYPE (lhs),
2994 TREE_TYPE (rhs)))
2995 rhs = fold_build1_loc (loc, VIEW_CONVERT_EXPR,
2996 TREE_TYPE (lhs), rhs);
2998 else
3000 if (racc->first_child)
3001 generate_subtree_copies (racc->first_child, lhs,
3002 racc->offset, 0, 0, gsi,
3003 false, false, loc);
3005 gcc_assert (*stmt == gsi_stmt (*gsi));
3006 unlink_stmt_vdef (*stmt);
3007 gsi_remove (gsi, true);
3008 sra_stats.deleted++;
3009 return SRA_AM_REMOVED;
3012 else if (racc->first_child)
3013 generate_subtree_copies (racc->first_child, lhs, racc->offset,
3014 0, 0, gsi, false, true, loc);
3016 if (access_has_children_p (lacc))
3017 generate_subtree_copies (lacc->first_child, rhs, lacc->offset,
3018 0, 0, gsi, true, true, loc);
3022 /* This gimplification must be done after generate_subtree_copies, lest we
3023 insert the subtree copies in the middle of the gimplified sequence. */
3024 if (force_gimple_rhs)
3025 rhs = force_gimple_operand_gsi (&orig_gsi, rhs, true, NULL_TREE,
3026 true, GSI_SAME_STMT);
3027 if (gimple_assign_rhs1 (*stmt) != rhs)
3029 modify_this_stmt = true;
3030 gimple_assign_set_rhs_from_tree (&orig_gsi, rhs);
3031 gcc_assert (*stmt == gsi_stmt (orig_gsi));
3034 return modify_this_stmt ? SRA_AM_MODIFIED : SRA_AM_NONE;
3037 /* Traverse the function body and all modifications as decided in
3038 analyze_all_variable_accesses. Return true iff the CFG has been
3039 changed. */
3041 static bool
3042 sra_modify_function_body (void)
3044 bool cfg_changed = false;
3045 basic_block bb;
3047 FOR_EACH_BB (bb)
3049 gimple_stmt_iterator gsi = gsi_start_bb (bb);
3050 while (!gsi_end_p (gsi))
3052 gimple stmt = gsi_stmt (gsi);
3053 enum assignment_mod_result assign_result;
3054 bool modified = false, deleted = false;
3055 tree *t;
3056 unsigned i;
3058 switch (gimple_code (stmt))
3060 case GIMPLE_RETURN:
3061 t = gimple_return_retval_ptr (stmt);
3062 if (*t != NULL_TREE)
3063 modified |= sra_modify_expr (t, &gsi, false);
3064 break;
3066 case GIMPLE_ASSIGN:
3067 assign_result = sra_modify_assign (&stmt, &gsi);
3068 modified |= assign_result == SRA_AM_MODIFIED;
3069 deleted = assign_result == SRA_AM_REMOVED;
3070 break;
3072 case GIMPLE_CALL:
3073 /* Operands must be processed before the lhs. */
3074 for (i = 0; i < gimple_call_num_args (stmt); i++)
3076 t = gimple_call_arg_ptr (stmt, i);
3077 modified |= sra_modify_expr (t, &gsi, false);
3080 if (gimple_call_lhs (stmt))
3082 t = gimple_call_lhs_ptr (stmt);
3083 modified |= sra_modify_expr (t, &gsi, true);
3085 break;
3087 case GIMPLE_ASM:
3088 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
3090 t = &TREE_VALUE (gimple_asm_input_op (stmt, i));
3091 modified |= sra_modify_expr (t, &gsi, false);
3093 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
3095 t = &TREE_VALUE (gimple_asm_output_op (stmt, i));
3096 modified |= sra_modify_expr (t, &gsi, true);
3098 break;
3100 default:
3101 break;
3104 if (modified)
3106 update_stmt (stmt);
3107 if (maybe_clean_eh_stmt (stmt)
3108 && gimple_purge_dead_eh_edges (gimple_bb (stmt)))
3109 cfg_changed = true;
3111 if (!deleted)
3112 gsi_next (&gsi);
3116 return cfg_changed;
3119 /* Generate statements initializing scalar replacements of parts of function
3120 parameters. */
3122 static void
3123 initialize_parameter_reductions (void)
3125 gimple_stmt_iterator gsi;
3126 gimple_seq seq = NULL;
3127 tree parm;
3129 for (parm = DECL_ARGUMENTS (current_function_decl);
3130 parm;
3131 parm = DECL_CHAIN (parm))
3133 VEC (access_p, heap) *access_vec;
3134 struct access *access;
3136 if (!bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3137 continue;
3138 access_vec = get_base_access_vector (parm);
3139 if (!access_vec)
3140 continue;
3142 if (!seq)
3144 seq = gimple_seq_alloc ();
3145 gsi = gsi_start (seq);
3148 for (access = VEC_index (access_p, access_vec, 0);
3149 access;
3150 access = access->next_grp)
3151 generate_subtree_copies (access, parm, 0, 0, 0, &gsi, true, true,
3152 EXPR_LOCATION (parm));
3155 if (seq)
3156 gsi_insert_seq_on_edge_immediate (single_succ_edge (ENTRY_BLOCK_PTR), seq);
3159 /* The "main" function of intraprocedural SRA passes. Runs the analysis and if
3160 it reveals there are components of some aggregates to be scalarized, it runs
3161 the required transformations. */
3162 static unsigned int
3163 perform_intra_sra (void)
3165 int ret = 0;
3166 sra_initialize ();
3168 if (!find_var_candidates ())
3169 goto out;
3171 if (!scan_function ())
3172 goto out;
3174 if (!analyze_all_variable_accesses ())
3175 goto out;
3177 if (sra_modify_function_body ())
3178 ret = TODO_update_ssa | TODO_cleanup_cfg;
3179 else
3180 ret = TODO_update_ssa;
3181 initialize_parameter_reductions ();
3183 statistics_counter_event (cfun, "Scalar replacements created",
3184 sra_stats.replacements);
3185 statistics_counter_event (cfun, "Modified expressions", sra_stats.exprs);
3186 statistics_counter_event (cfun, "Subtree copy stmts",
3187 sra_stats.subtree_copies);
3188 statistics_counter_event (cfun, "Subreplacement stmts",
3189 sra_stats.subreplacements);
3190 statistics_counter_event (cfun, "Deleted stmts", sra_stats.deleted);
3191 statistics_counter_event (cfun, "Separate LHS and RHS handling",
3192 sra_stats.separate_lhs_rhs_handling);
3194 out:
3195 sra_deinitialize ();
3196 return ret;
3199 /* Perform early intraprocedural SRA. */
3200 static unsigned int
3201 early_intra_sra (void)
3203 sra_mode = SRA_MODE_EARLY_INTRA;
3204 return perform_intra_sra ();
3207 /* Perform "late" intraprocedural SRA. */
3208 static unsigned int
3209 late_intra_sra (void)
3211 sra_mode = SRA_MODE_INTRA;
3212 return perform_intra_sra ();
3216 static bool
3217 gate_intra_sra (void)
3219 return flag_tree_sra != 0 && dbg_cnt (tree_sra);
3223 struct gimple_opt_pass pass_sra_early =
3226 GIMPLE_PASS,
3227 "esra", /* name */
3228 gate_intra_sra, /* gate */
3229 early_intra_sra, /* execute */
3230 NULL, /* sub */
3231 NULL, /* next */
3232 0, /* static_pass_number */
3233 TV_TREE_SRA, /* tv_id */
3234 PROP_cfg | PROP_ssa, /* properties_required */
3235 0, /* properties_provided */
3236 0, /* properties_destroyed */
3237 0, /* todo_flags_start */
3238 TODO_update_ssa
3239 | TODO_ggc_collect
3240 | TODO_verify_ssa /* todo_flags_finish */
3244 struct gimple_opt_pass pass_sra =
3247 GIMPLE_PASS,
3248 "sra", /* name */
3249 gate_intra_sra, /* gate */
3250 late_intra_sra, /* execute */
3251 NULL, /* sub */
3252 NULL, /* next */
3253 0, /* static_pass_number */
3254 TV_TREE_SRA, /* tv_id */
3255 PROP_cfg | PROP_ssa, /* properties_required */
3256 0, /* properties_provided */
3257 0, /* properties_destroyed */
3258 TODO_update_address_taken, /* todo_flags_start */
3259 TODO_update_ssa
3260 | TODO_ggc_collect
3261 | TODO_verify_ssa /* todo_flags_finish */
3266 /* Return true iff PARM (which must be a parm_decl) is an unused scalar
3267 parameter. */
3269 static bool
3270 is_unused_scalar_param (tree parm)
3272 tree name;
3273 return (is_gimple_reg (parm)
3274 && (!(name = gimple_default_def (cfun, parm))
3275 || has_zero_uses (name)));
3278 /* Scan immediate uses of a default definition SSA name of a parameter PARM and
3279 examine whether there are any direct or otherwise infeasible ones. If so,
3280 return true, otherwise return false. PARM must be a gimple register with a
3281 non-NULL default definition. */
3283 static bool
3284 ptr_parm_has_direct_uses (tree parm)
3286 imm_use_iterator ui;
3287 gimple stmt;
3288 tree name = gimple_default_def (cfun, parm);
3289 bool ret = false;
3291 FOR_EACH_IMM_USE_STMT (stmt, ui, name)
3293 int uses_ok = 0;
3294 use_operand_p use_p;
3296 if (is_gimple_debug (stmt))
3297 continue;
3299 /* Valid uses include dereferences on the lhs and the rhs. */
3300 if (gimple_has_lhs (stmt))
3302 tree lhs = gimple_get_lhs (stmt);
3303 while (handled_component_p (lhs))
3304 lhs = TREE_OPERAND (lhs, 0);
3305 if (TREE_CODE (lhs) == MEM_REF
3306 && TREE_OPERAND (lhs, 0) == name
3307 && integer_zerop (TREE_OPERAND (lhs, 1))
3308 && types_compatible_p (TREE_TYPE (lhs),
3309 TREE_TYPE (TREE_TYPE (name)))
3310 && !TREE_THIS_VOLATILE (lhs))
3311 uses_ok++;
3313 if (gimple_assign_single_p (stmt))
3315 tree rhs = gimple_assign_rhs1 (stmt);
3316 while (handled_component_p (rhs))
3317 rhs = TREE_OPERAND (rhs, 0);
3318 if (TREE_CODE (rhs) == MEM_REF
3319 && TREE_OPERAND (rhs, 0) == name
3320 && integer_zerop (TREE_OPERAND (rhs, 1))
3321 && types_compatible_p (TREE_TYPE (rhs),
3322 TREE_TYPE (TREE_TYPE (name)))
3323 && !TREE_THIS_VOLATILE (rhs))
3324 uses_ok++;
3326 else if (is_gimple_call (stmt))
3328 unsigned i;
3329 for (i = 0; i < gimple_call_num_args (stmt); ++i)
3331 tree arg = gimple_call_arg (stmt, i);
3332 while (handled_component_p (arg))
3333 arg = TREE_OPERAND (arg, 0);
3334 if (TREE_CODE (arg) == MEM_REF
3335 && TREE_OPERAND (arg, 0) == name
3336 && integer_zerop (TREE_OPERAND (arg, 1))
3337 && types_compatible_p (TREE_TYPE (arg),
3338 TREE_TYPE (TREE_TYPE (name)))
3339 && !TREE_THIS_VOLATILE (arg))
3340 uses_ok++;
3344 /* If the number of valid uses does not match the number of
3345 uses in this stmt there is an unhandled use. */
3346 FOR_EACH_IMM_USE_ON_STMT (use_p, ui)
3347 --uses_ok;
3349 if (uses_ok != 0)
3350 ret = true;
3352 if (ret)
3353 BREAK_FROM_IMM_USE_STMT (ui);
3356 return ret;
3359 /* Identify candidates for reduction for IPA-SRA based on their type and mark
3360 them in candidate_bitmap. Note that these do not necessarily include
3361 parameter which are unused and thus can be removed. Return true iff any
3362 such candidate has been found. */
3364 static bool
3365 find_param_candidates (void)
3367 tree parm;
3368 int count = 0;
3369 bool ret = false;
3370 const char *msg;
3372 for (parm = DECL_ARGUMENTS (current_function_decl);
3373 parm;
3374 parm = DECL_CHAIN (parm))
3376 tree type = TREE_TYPE (parm);
3378 count++;
3380 if (TREE_THIS_VOLATILE (parm)
3381 || TREE_ADDRESSABLE (parm)
3382 || (!is_gimple_reg_type (type) && is_va_list_type (type)))
3383 continue;
3385 if (is_unused_scalar_param (parm))
3387 ret = true;
3388 continue;
3391 if (POINTER_TYPE_P (type))
3393 type = TREE_TYPE (type);
3395 if (TREE_CODE (type) == FUNCTION_TYPE
3396 || TYPE_VOLATILE (type)
3397 || (TREE_CODE (type) == ARRAY_TYPE
3398 && TYPE_NONALIASED_COMPONENT (type))
3399 || !is_gimple_reg (parm)
3400 || is_va_list_type (type)
3401 || ptr_parm_has_direct_uses (parm))
3402 continue;
3404 else if (!AGGREGATE_TYPE_P (type))
3405 continue;
3407 if (!COMPLETE_TYPE_P (type)
3408 || !host_integerp (TYPE_SIZE (type), 1)
3409 || tree_low_cst (TYPE_SIZE (type), 1) == 0
3410 || (AGGREGATE_TYPE_P (type)
3411 && type_internals_preclude_sra_p (type, &msg)))
3412 continue;
3414 bitmap_set_bit (candidate_bitmap, DECL_UID (parm));
3415 ret = true;
3416 if (dump_file && (dump_flags & TDF_DETAILS))
3418 fprintf (dump_file, "Candidate (%d): ", DECL_UID (parm));
3419 print_generic_expr (dump_file, parm, 0);
3420 fprintf (dump_file, "\n");
3424 func_param_count = count;
3425 return ret;
3428 /* Callback of walk_aliased_vdefs, marks the access passed as DATA as
3429 maybe_modified. */
3431 static bool
3432 mark_maybe_modified (ao_ref *ao ATTRIBUTE_UNUSED, tree vdef ATTRIBUTE_UNUSED,
3433 void *data)
3435 struct access *repr = (struct access *) data;
3437 repr->grp_maybe_modified = 1;
3438 return true;
3441 /* Analyze what representatives (in linked lists accessible from
3442 REPRESENTATIVES) can be modified by side effects of statements in the
3443 current function. */
3445 static void
3446 analyze_modified_params (VEC (access_p, heap) *representatives)
3448 int i;
3450 for (i = 0; i < func_param_count; i++)
3452 struct access *repr;
3454 for (repr = VEC_index (access_p, representatives, i);
3455 repr;
3456 repr = repr->next_grp)
3458 struct access *access;
3459 bitmap visited;
3460 ao_ref ar;
3462 if (no_accesses_p (repr))
3463 continue;
3464 if (!POINTER_TYPE_P (TREE_TYPE (repr->base))
3465 || repr->grp_maybe_modified)
3466 continue;
3468 ao_ref_init (&ar, repr->expr);
3469 visited = BITMAP_ALLOC (NULL);
3470 for (access = repr; access; access = access->next_sibling)
3472 /* All accesses are read ones, otherwise grp_maybe_modified would
3473 be trivially set. */
3474 walk_aliased_vdefs (&ar, gimple_vuse (access->stmt),
3475 mark_maybe_modified, repr, &visited);
3476 if (repr->grp_maybe_modified)
3477 break;
3479 BITMAP_FREE (visited);
3484 /* Propagate distances in bb_dereferences in the opposite direction than the
3485 control flow edges, in each step storing the maximum of the current value
3486 and the minimum of all successors. These steps are repeated until the table
3487 stabilizes. Note that BBs which might terminate the functions (according to
3488 final_bbs bitmap) never updated in this way. */
3490 static void
3491 propagate_dereference_distances (void)
3493 VEC (basic_block, heap) *queue;
3494 basic_block bb;
3496 queue = VEC_alloc (basic_block, heap, last_basic_block_for_function (cfun));
3497 VEC_quick_push (basic_block, queue, ENTRY_BLOCK_PTR);
3498 FOR_EACH_BB (bb)
3500 VEC_quick_push (basic_block, queue, bb);
3501 bb->aux = bb;
3504 while (!VEC_empty (basic_block, queue))
3506 edge_iterator ei;
3507 edge e;
3508 bool change = false;
3509 int i;
3511 bb = VEC_pop (basic_block, queue);
3512 bb->aux = NULL;
3514 if (bitmap_bit_p (final_bbs, bb->index))
3515 continue;
3517 for (i = 0; i < func_param_count; i++)
3519 int idx = bb->index * func_param_count + i;
3520 bool first = true;
3521 HOST_WIDE_INT inh = 0;
3523 FOR_EACH_EDGE (e, ei, bb->succs)
3525 int succ_idx = e->dest->index * func_param_count + i;
3527 if (e->src == EXIT_BLOCK_PTR)
3528 continue;
3530 if (first)
3532 first = false;
3533 inh = bb_dereferences [succ_idx];
3535 else if (bb_dereferences [succ_idx] < inh)
3536 inh = bb_dereferences [succ_idx];
3539 if (!first && bb_dereferences[idx] < inh)
3541 bb_dereferences[idx] = inh;
3542 change = true;
3546 if (change && !bitmap_bit_p (final_bbs, bb->index))
3547 FOR_EACH_EDGE (e, ei, bb->preds)
3549 if (e->src->aux)
3550 continue;
3552 e->src->aux = e->src;
3553 VEC_quick_push (basic_block, queue, e->src);
3557 VEC_free (basic_block, heap, queue);
3560 /* Dump a dereferences TABLE with heading STR to file F. */
3562 static void
3563 dump_dereferences_table (FILE *f, const char *str, HOST_WIDE_INT *table)
3565 basic_block bb;
3567 fprintf (dump_file, str);
3568 FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, EXIT_BLOCK_PTR, next_bb)
3570 fprintf (f, "%4i %i ", bb->index, bitmap_bit_p (final_bbs, bb->index));
3571 if (bb != EXIT_BLOCK_PTR)
3573 int i;
3574 for (i = 0; i < func_param_count; i++)
3576 int idx = bb->index * func_param_count + i;
3577 fprintf (f, " %4" HOST_WIDE_INT_PRINT "d", table[idx]);
3580 fprintf (f, "\n");
3582 fprintf (dump_file, "\n");
3585 /* Determine what (parts of) parameters passed by reference that are not
3586 assigned to are not certainly dereferenced in this function and thus the
3587 dereferencing cannot be safely moved to the caller without potentially
3588 introducing a segfault. Mark such REPRESENTATIVES as
3589 grp_not_necessarilly_dereferenced.
3591 The dereferenced maximum "distance," i.e. the offset + size of the accessed
3592 part is calculated rather than simple booleans are calculated for each
3593 pointer parameter to handle cases when only a fraction of the whole
3594 aggregate is allocated (see testsuite/gcc.c-torture/execute/ipa-sra-2.c for
3595 an example).
3597 The maximum dereference distances for each pointer parameter and BB are
3598 already stored in bb_dereference. This routine simply propagates these
3599 values upwards by propagate_dereference_distances and then compares the
3600 distances of individual parameters in the ENTRY BB to the equivalent
3601 distances of each representative of a (fraction of a) parameter. */
3603 static void
3604 analyze_caller_dereference_legality (VEC (access_p, heap) *representatives)
3606 int i;
3608 if (dump_file && (dump_flags & TDF_DETAILS))
3609 dump_dereferences_table (dump_file,
3610 "Dereference table before propagation:\n",
3611 bb_dereferences);
3613 propagate_dereference_distances ();
3615 if (dump_file && (dump_flags & TDF_DETAILS))
3616 dump_dereferences_table (dump_file,
3617 "Dereference table after propagation:\n",
3618 bb_dereferences);
3620 for (i = 0; i < func_param_count; i++)
3622 struct access *repr = VEC_index (access_p, representatives, i);
3623 int idx = ENTRY_BLOCK_PTR->index * func_param_count + i;
3625 if (!repr || no_accesses_p (repr))
3626 continue;
3630 if ((repr->offset + repr->size) > bb_dereferences[idx])
3631 repr->grp_not_necessarilly_dereferenced = 1;
3632 repr = repr->next_grp;
3634 while (repr);
3638 /* Return the representative access for the parameter declaration PARM if it is
3639 a scalar passed by reference which is not written to and the pointer value
3640 is not used directly. Thus, if it is legal to dereference it in the caller
3641 and we can rule out modifications through aliases, such parameter should be
3642 turned into one passed by value. Return NULL otherwise. */
3644 static struct access *
3645 unmodified_by_ref_scalar_representative (tree parm)
3647 int i, access_count;
3648 struct access *repr;
3649 VEC (access_p, heap) *access_vec;
3651 access_vec = get_base_access_vector (parm);
3652 gcc_assert (access_vec);
3653 repr = VEC_index (access_p, access_vec, 0);
3654 if (repr->write)
3655 return NULL;
3656 repr->group_representative = repr;
3658 access_count = VEC_length (access_p, access_vec);
3659 for (i = 1; i < access_count; i++)
3661 struct access *access = VEC_index (access_p, access_vec, i);
3662 if (access->write)
3663 return NULL;
3664 access->group_representative = repr;
3665 access->next_sibling = repr->next_sibling;
3666 repr->next_sibling = access;
3669 repr->grp_read = 1;
3670 repr->grp_scalar_ptr = 1;
3671 return repr;
3674 /* Return true iff this access precludes IPA-SRA of the parameter it is
3675 associated with. */
3677 static bool
3678 access_precludes_ipa_sra_p (struct access *access)
3680 /* Avoid issues such as the second simple testcase in PR 42025. The problem
3681 is incompatible assign in a call statement (and possibly even in asm
3682 statements). This can be relaxed by using a new temporary but only for
3683 non-TREE_ADDRESSABLE types and is probably not worth the complexity. (In
3684 intraprocedural SRA we deal with this by keeping the old aggregate around,
3685 something we cannot do in IPA-SRA.) */
3686 if (access->write
3687 && (is_gimple_call (access->stmt)
3688 || gimple_code (access->stmt) == GIMPLE_ASM))
3689 return true;
3691 if (tree_non_mode_aligned_mem_p (access->expr))
3692 return true;
3694 return false;
3698 /* Sort collected accesses for parameter PARM, identify representatives for
3699 each accessed region and link them together. Return NULL if there are
3700 different but overlapping accesses, return the special ptr value meaning
3701 there are no accesses for this parameter if that is the case and return the
3702 first representative otherwise. Set *RO_GRP if there is a group of accesses
3703 with only read (i.e. no write) accesses. */
3705 static struct access *
3706 splice_param_accesses (tree parm, bool *ro_grp)
3708 int i, j, access_count, group_count;
3709 int agg_size, total_size = 0;
3710 struct access *access, *res, **prev_acc_ptr = &res;
3711 VEC (access_p, heap) *access_vec;
3713 access_vec = get_base_access_vector (parm);
3714 if (!access_vec)
3715 return &no_accesses_representant;
3716 access_count = VEC_length (access_p, access_vec);
3718 VEC_qsort (access_p, access_vec, compare_access_positions);
3720 i = 0;
3721 total_size = 0;
3722 group_count = 0;
3723 while (i < access_count)
3725 bool modification;
3726 tree a1_alias_type;
3727 access = VEC_index (access_p, access_vec, i);
3728 modification = access->write;
3729 if (access_precludes_ipa_sra_p (access))
3730 return NULL;
3731 a1_alias_type = reference_alias_ptr_type (access->expr);
3733 /* Access is about to become group representative unless we find some
3734 nasty overlap which would preclude us from breaking this parameter
3735 apart. */
3737 j = i + 1;
3738 while (j < access_count)
3740 struct access *ac2 = VEC_index (access_p, access_vec, j);
3741 if (ac2->offset != access->offset)
3743 /* All or nothing law for parameters. */
3744 if (access->offset + access->size > ac2->offset)
3745 return NULL;
3746 else
3747 break;
3749 else if (ac2->size != access->size)
3750 return NULL;
3752 if (access_precludes_ipa_sra_p (ac2)
3753 || (ac2->type != access->type
3754 && (TREE_ADDRESSABLE (ac2->type)
3755 || TREE_ADDRESSABLE (access->type)))
3756 || (reference_alias_ptr_type (ac2->expr) != a1_alias_type))
3757 return NULL;
3759 modification |= ac2->write;
3760 ac2->group_representative = access;
3761 ac2->next_sibling = access->next_sibling;
3762 access->next_sibling = ac2;
3763 j++;
3766 group_count++;
3767 access->grp_maybe_modified = modification;
3768 if (!modification)
3769 *ro_grp = true;
3770 *prev_acc_ptr = access;
3771 prev_acc_ptr = &access->next_grp;
3772 total_size += access->size;
3773 i = j;
3776 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3777 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm))), 1);
3778 else
3779 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (parm)), 1);
3780 if (total_size >= agg_size)
3781 return NULL;
3783 gcc_assert (group_count > 0);
3784 return res;
3787 /* Decide whether parameters with representative accesses given by REPR should
3788 be reduced into components. */
3790 static int
3791 decide_one_param_reduction (struct access *repr)
3793 int total_size, cur_parm_size, agg_size, new_param_count, parm_size_limit;
3794 bool by_ref;
3795 tree parm;
3797 parm = repr->base;
3798 cur_parm_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (parm)), 1);
3799 gcc_assert (cur_parm_size > 0);
3801 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3803 by_ref = true;
3804 agg_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm))), 1);
3806 else
3808 by_ref = false;
3809 agg_size = cur_parm_size;
3812 if (dump_file)
3814 struct access *acc;
3815 fprintf (dump_file, "Evaluating PARAM group sizes for ");
3816 print_generic_expr (dump_file, parm, 0);
3817 fprintf (dump_file, " (UID: %u): \n", DECL_UID (parm));
3818 for (acc = repr; acc; acc = acc->next_grp)
3819 dump_access (dump_file, acc, true);
3822 total_size = 0;
3823 new_param_count = 0;
3825 for (; repr; repr = repr->next_grp)
3827 gcc_assert (parm == repr->base);
3829 /* Taking the address of a non-addressable field is verboten. */
3830 if (by_ref && repr->non_addressable)
3831 return 0;
3833 if (!by_ref || (!repr->grp_maybe_modified
3834 && !repr->grp_not_necessarilly_dereferenced))
3835 total_size += repr->size;
3836 else
3837 total_size += cur_parm_size;
3839 new_param_count++;
3842 gcc_assert (new_param_count > 0);
3844 if (optimize_function_for_size_p (cfun))
3845 parm_size_limit = cur_parm_size;
3846 else
3847 parm_size_limit = (PARAM_VALUE (PARAM_IPA_SRA_PTR_GROWTH_FACTOR)
3848 * cur_parm_size);
3850 if (total_size < agg_size
3851 && total_size <= parm_size_limit)
3853 if (dump_file)
3854 fprintf (dump_file, " ....will be split into %i components\n",
3855 new_param_count);
3856 return new_param_count;
3858 else
3859 return 0;
3862 /* The order of the following enums is important, we need to do extra work for
3863 UNUSED_PARAMS, BY_VAL_ACCESSES and UNMODIF_BY_REF_ACCESSES. */
3864 enum ipa_splicing_result { NO_GOOD_ACCESS, UNUSED_PARAMS, BY_VAL_ACCESSES,
3865 MODIF_BY_REF_ACCESSES, UNMODIF_BY_REF_ACCESSES };
3867 /* Identify representatives of all accesses to all candidate parameters for
3868 IPA-SRA. Return result based on what representatives have been found. */
3870 static enum ipa_splicing_result
3871 splice_all_param_accesses (VEC (access_p, heap) **representatives)
3873 enum ipa_splicing_result result = NO_GOOD_ACCESS;
3874 tree parm;
3875 struct access *repr;
3877 *representatives = VEC_alloc (access_p, heap, func_param_count);
3879 for (parm = DECL_ARGUMENTS (current_function_decl);
3880 parm;
3881 parm = DECL_CHAIN (parm))
3883 if (is_unused_scalar_param (parm))
3885 VEC_quick_push (access_p, *representatives,
3886 &no_accesses_representant);
3887 if (result == NO_GOOD_ACCESS)
3888 result = UNUSED_PARAMS;
3890 else if (POINTER_TYPE_P (TREE_TYPE (parm))
3891 && is_gimple_reg_type (TREE_TYPE (TREE_TYPE (parm)))
3892 && bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3894 repr = unmodified_by_ref_scalar_representative (parm);
3895 VEC_quick_push (access_p, *representatives, repr);
3896 if (repr)
3897 result = UNMODIF_BY_REF_ACCESSES;
3899 else if (bitmap_bit_p (candidate_bitmap, DECL_UID (parm)))
3901 bool ro_grp = false;
3902 repr = splice_param_accesses (parm, &ro_grp);
3903 VEC_quick_push (access_p, *representatives, repr);
3905 if (repr && !no_accesses_p (repr))
3907 if (POINTER_TYPE_P (TREE_TYPE (parm)))
3909 if (ro_grp)
3910 result = UNMODIF_BY_REF_ACCESSES;
3911 else if (result < MODIF_BY_REF_ACCESSES)
3912 result = MODIF_BY_REF_ACCESSES;
3914 else if (result < BY_VAL_ACCESSES)
3915 result = BY_VAL_ACCESSES;
3917 else if (no_accesses_p (repr) && (result == NO_GOOD_ACCESS))
3918 result = UNUSED_PARAMS;
3920 else
3921 VEC_quick_push (access_p, *representatives, NULL);
3924 if (result == NO_GOOD_ACCESS)
3926 VEC_free (access_p, heap, *representatives);
3927 *representatives = NULL;
3928 return NO_GOOD_ACCESS;
3931 return result;
3934 /* Return the index of BASE in PARMS. Abort if it is not found. */
3936 static inline int
3937 get_param_index (tree base, VEC(tree, heap) *parms)
3939 int i, len;
3941 len = VEC_length (tree, parms);
3942 for (i = 0; i < len; i++)
3943 if (VEC_index (tree, parms, i) == base)
3944 return i;
3945 gcc_unreachable ();
3948 /* Convert the decisions made at the representative level into compact
3949 parameter adjustments. REPRESENTATIVES are pointers to first
3950 representatives of each param accesses, ADJUSTMENTS_COUNT is the expected
3951 final number of adjustments. */
3953 static ipa_parm_adjustment_vec
3954 turn_representatives_into_adjustments (VEC (access_p, heap) *representatives,
3955 int adjustments_count)
3957 VEC (tree, heap) *parms;
3958 ipa_parm_adjustment_vec adjustments;
3959 tree parm;
3960 int i;
3962 gcc_assert (adjustments_count > 0);
3963 parms = ipa_get_vector_of_formal_parms (current_function_decl);
3964 adjustments = VEC_alloc (ipa_parm_adjustment_t, heap, adjustments_count);
3965 parm = DECL_ARGUMENTS (current_function_decl);
3966 for (i = 0; i < func_param_count; i++, parm = DECL_CHAIN (parm))
3968 struct access *repr = VEC_index (access_p, representatives, i);
3970 if (!repr || no_accesses_p (repr))
3972 struct ipa_parm_adjustment *adj;
3974 adj = VEC_quick_push (ipa_parm_adjustment_t, adjustments, NULL);
3975 memset (adj, 0, sizeof (*adj));
3976 adj->base_index = get_param_index (parm, parms);
3977 adj->base = parm;
3978 if (!repr)
3979 adj->copy_param = 1;
3980 else
3981 adj->remove_param = 1;
3983 else
3985 struct ipa_parm_adjustment *adj;
3986 int index = get_param_index (parm, parms);
3988 for (; repr; repr = repr->next_grp)
3990 adj = VEC_quick_push (ipa_parm_adjustment_t, adjustments, NULL);
3991 memset (adj, 0, sizeof (*adj));
3992 gcc_assert (repr->base == parm);
3993 adj->base_index = index;
3994 adj->base = repr->base;
3995 adj->type = repr->type;
3996 adj->alias_ptr_type = reference_alias_ptr_type (repr->expr);
3997 adj->offset = repr->offset;
3998 adj->by_ref = (POINTER_TYPE_P (TREE_TYPE (repr->base))
3999 && (repr->grp_maybe_modified
4000 || repr->grp_not_necessarilly_dereferenced));
4005 VEC_free (tree, heap, parms);
4006 return adjustments;
4009 /* Analyze the collected accesses and produce a plan what to do with the
4010 parameters in the form of adjustments, NULL meaning nothing. */
4012 static ipa_parm_adjustment_vec
4013 analyze_all_param_acesses (void)
4015 enum ipa_splicing_result repr_state;
4016 bool proceed = false;
4017 int i, adjustments_count = 0;
4018 VEC (access_p, heap) *representatives;
4019 ipa_parm_adjustment_vec adjustments;
4021 repr_state = splice_all_param_accesses (&representatives);
4022 if (repr_state == NO_GOOD_ACCESS)
4023 return NULL;
4025 /* If there are any parameters passed by reference which are not modified
4026 directly, we need to check whether they can be modified indirectly. */
4027 if (repr_state == UNMODIF_BY_REF_ACCESSES)
4029 analyze_caller_dereference_legality (representatives);
4030 analyze_modified_params (representatives);
4033 for (i = 0; i < func_param_count; i++)
4035 struct access *repr = VEC_index (access_p, representatives, i);
4037 if (repr && !no_accesses_p (repr))
4039 if (repr->grp_scalar_ptr)
4041 adjustments_count++;
4042 if (repr->grp_not_necessarilly_dereferenced
4043 || repr->grp_maybe_modified)
4044 VEC_replace (access_p, representatives, i, NULL);
4045 else
4047 proceed = true;
4048 sra_stats.scalar_by_ref_to_by_val++;
4051 else
4053 int new_components = decide_one_param_reduction (repr);
4055 if (new_components == 0)
4057 VEC_replace (access_p, representatives, i, NULL);
4058 adjustments_count++;
4060 else
4062 adjustments_count += new_components;
4063 sra_stats.aggregate_params_reduced++;
4064 sra_stats.param_reductions_created += new_components;
4065 proceed = true;
4069 else
4071 if (no_accesses_p (repr))
4073 proceed = true;
4074 sra_stats.deleted_unused_parameters++;
4076 adjustments_count++;
4080 if (!proceed && dump_file)
4081 fprintf (dump_file, "NOT proceeding to change params.\n");
4083 if (proceed)
4084 adjustments = turn_representatives_into_adjustments (representatives,
4085 adjustments_count);
4086 else
4087 adjustments = NULL;
4089 VEC_free (access_p, heap, representatives);
4090 return adjustments;
4093 /* If a parameter replacement identified by ADJ does not yet exist in the form
4094 of declaration, create it and record it, otherwise return the previously
4095 created one. */
4097 static tree
4098 get_replaced_param_substitute (struct ipa_parm_adjustment *adj)
4100 tree repl;
4101 if (!adj->new_ssa_base)
4103 char *pretty_name = make_fancy_name (adj->base);
4105 repl = create_tmp_reg (TREE_TYPE (adj->base), "ISR");
4106 DECL_NAME (repl) = get_identifier (pretty_name);
4107 obstack_free (&name_obstack, pretty_name);
4109 get_var_ann (repl);
4110 add_referenced_var (repl);
4111 adj->new_ssa_base = repl;
4113 else
4114 repl = adj->new_ssa_base;
4115 return repl;
4118 /* Find the first adjustment for a particular parameter BASE in a vector of
4119 ADJUSTMENTS which is not a copy_param. Return NULL if there is no such
4120 adjustment. */
4122 static struct ipa_parm_adjustment *
4123 get_adjustment_for_base (ipa_parm_adjustment_vec adjustments, tree base)
4125 int i, len;
4127 len = VEC_length (ipa_parm_adjustment_t, adjustments);
4128 for (i = 0; i < len; i++)
4130 struct ipa_parm_adjustment *adj;
4132 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
4133 if (!adj->copy_param && adj->base == base)
4134 return adj;
4137 return NULL;
4140 /* If the statement STMT defines an SSA_NAME of a parameter which is to be
4141 removed because its value is not used, replace the SSA_NAME with a one
4142 relating to a created VAR_DECL together all of its uses and return true.
4143 ADJUSTMENTS is a pointer to an adjustments vector. */
4145 static bool
4146 replace_removed_params_ssa_names (gimple stmt,
4147 ipa_parm_adjustment_vec adjustments)
4149 struct ipa_parm_adjustment *adj;
4150 tree lhs, decl, repl, name;
4152 if (gimple_code (stmt) == GIMPLE_PHI)
4153 lhs = gimple_phi_result (stmt);
4154 else if (is_gimple_assign (stmt))
4155 lhs = gimple_assign_lhs (stmt);
4156 else if (is_gimple_call (stmt))
4157 lhs = gimple_call_lhs (stmt);
4158 else
4159 gcc_unreachable ();
4161 if (TREE_CODE (lhs) != SSA_NAME)
4162 return false;
4163 decl = SSA_NAME_VAR (lhs);
4164 if (TREE_CODE (decl) != PARM_DECL)
4165 return false;
4167 adj = get_adjustment_for_base (adjustments, decl);
4168 if (!adj)
4169 return false;
4171 repl = get_replaced_param_substitute (adj);
4172 name = make_ssa_name (repl, stmt);
4174 if (dump_file)
4176 fprintf (dump_file, "replacing an SSA name of a removed param ");
4177 print_generic_expr (dump_file, lhs, 0);
4178 fprintf (dump_file, " with ");
4179 print_generic_expr (dump_file, name, 0);
4180 fprintf (dump_file, "\n");
4183 if (is_gimple_assign (stmt))
4184 gimple_assign_set_lhs (stmt, name);
4185 else if (is_gimple_call (stmt))
4186 gimple_call_set_lhs (stmt, name);
4187 else
4188 gimple_phi_set_result (stmt, name);
4190 replace_uses_by (lhs, name);
4191 release_ssa_name (lhs);
4192 return true;
4195 /* If the expression *EXPR should be replaced by a reduction of a parameter, do
4196 so. ADJUSTMENTS is a pointer to a vector of adjustments. CONVERT
4197 specifies whether the function should care about type incompatibility the
4198 current and new expressions. If it is false, the function will leave
4199 incompatibility issues to the caller. Return true iff the expression
4200 was modified. */
4202 static bool
4203 sra_ipa_modify_expr (tree *expr, bool convert,
4204 ipa_parm_adjustment_vec adjustments)
4206 int i, len;
4207 struct ipa_parm_adjustment *adj, *cand = NULL;
4208 HOST_WIDE_INT offset, size, max_size;
4209 tree base, src;
4211 len = VEC_length (ipa_parm_adjustment_t, adjustments);
4213 if (TREE_CODE (*expr) == BIT_FIELD_REF
4214 || TREE_CODE (*expr) == IMAGPART_EXPR
4215 || TREE_CODE (*expr) == REALPART_EXPR)
4217 expr = &TREE_OPERAND (*expr, 0);
4218 convert = true;
4221 base = get_ref_base_and_extent (*expr, &offset, &size, &max_size);
4222 if (!base || size == -1 || max_size == -1)
4223 return false;
4225 if (TREE_CODE (base) == MEM_REF)
4227 offset += mem_ref_offset (base).low * BITS_PER_UNIT;
4228 base = TREE_OPERAND (base, 0);
4231 base = get_ssa_base_param (base);
4232 if (!base || TREE_CODE (base) != PARM_DECL)
4233 return false;
4235 for (i = 0; i < len; i++)
4237 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
4239 if (adj->base == base &&
4240 (adj->offset == offset || adj->remove_param))
4242 cand = adj;
4243 break;
4246 if (!cand || cand->copy_param || cand->remove_param)
4247 return false;
4249 if (cand->by_ref)
4250 src = build_simple_mem_ref (cand->reduction);
4251 else
4252 src = cand->reduction;
4254 if (dump_file && (dump_flags & TDF_DETAILS))
4256 fprintf (dump_file, "About to replace expr ");
4257 print_generic_expr (dump_file, *expr, 0);
4258 fprintf (dump_file, " with ");
4259 print_generic_expr (dump_file, src, 0);
4260 fprintf (dump_file, "\n");
4263 if (convert && !useless_type_conversion_p (TREE_TYPE (*expr), cand->type))
4265 tree vce = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (*expr), src);
4266 *expr = vce;
4268 else
4269 *expr = src;
4270 return true;
4273 /* If the statement pointed to by STMT_PTR contains any expressions that need
4274 to replaced with a different one as noted by ADJUSTMENTS, do so. Handle any
4275 potential type incompatibilities (GSI is used to accommodate conversion
4276 statements and must point to the statement). Return true iff the statement
4277 was modified. */
4279 static bool
4280 sra_ipa_modify_assign (gimple *stmt_ptr, gimple_stmt_iterator *gsi,
4281 ipa_parm_adjustment_vec adjustments)
4283 gimple stmt = *stmt_ptr;
4284 tree *lhs_p, *rhs_p;
4285 bool any;
4287 if (!gimple_assign_single_p (stmt))
4288 return false;
4290 rhs_p = gimple_assign_rhs1_ptr (stmt);
4291 lhs_p = gimple_assign_lhs_ptr (stmt);
4293 any = sra_ipa_modify_expr (rhs_p, false, adjustments);
4294 any |= sra_ipa_modify_expr (lhs_p, false, adjustments);
4295 if (any)
4297 tree new_rhs = NULL_TREE;
4299 if (!useless_type_conversion_p (TREE_TYPE (*lhs_p), TREE_TYPE (*rhs_p)))
4301 if (TREE_CODE (*rhs_p) == CONSTRUCTOR)
4303 /* V_C_Es of constructors can cause trouble (PR 42714). */
4304 if (is_gimple_reg_type (TREE_TYPE (*lhs_p)))
4305 *rhs_p = build_zero_cst (TREE_TYPE (*lhs_p));
4306 else
4307 *rhs_p = build_constructor (TREE_TYPE (*lhs_p), 0);
4309 else
4310 new_rhs = fold_build1_loc (gimple_location (stmt),
4311 VIEW_CONVERT_EXPR, TREE_TYPE (*lhs_p),
4312 *rhs_p);
4314 else if (REFERENCE_CLASS_P (*rhs_p)
4315 && is_gimple_reg_type (TREE_TYPE (*lhs_p))
4316 && !is_gimple_reg (*lhs_p))
4317 /* This can happen when an assignment in between two single field
4318 structures is turned into an assignment in between two pointers to
4319 scalars (PR 42237). */
4320 new_rhs = *rhs_p;
4322 if (new_rhs)
4324 tree tmp = force_gimple_operand_gsi (gsi, new_rhs, true, NULL_TREE,
4325 true, GSI_SAME_STMT);
4327 gimple_assign_set_rhs_from_tree (gsi, tmp);
4330 return true;
4333 return false;
4336 /* Traverse the function body and all modifications as described in
4337 ADJUSTMENTS. Return true iff the CFG has been changed. */
4339 static bool
4340 ipa_sra_modify_function_body (ipa_parm_adjustment_vec adjustments)
4342 bool cfg_changed = false;
4343 basic_block bb;
4345 FOR_EACH_BB (bb)
4347 gimple_stmt_iterator gsi;
4349 for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi))
4350 replace_removed_params_ssa_names (gsi_stmt (gsi), adjustments);
4352 gsi = gsi_start_bb (bb);
4353 while (!gsi_end_p (gsi))
4355 gimple stmt = gsi_stmt (gsi);
4356 bool modified = false;
4357 tree *t;
4358 unsigned i;
4360 switch (gimple_code (stmt))
4362 case GIMPLE_RETURN:
4363 t = gimple_return_retval_ptr (stmt);
4364 if (*t != NULL_TREE)
4365 modified |= sra_ipa_modify_expr (t, true, adjustments);
4366 break;
4368 case GIMPLE_ASSIGN:
4369 modified |= sra_ipa_modify_assign (&stmt, &gsi, adjustments);
4370 modified |= replace_removed_params_ssa_names (stmt, adjustments);
4371 break;
4373 case GIMPLE_CALL:
4374 /* Operands must be processed before the lhs. */
4375 for (i = 0; i < gimple_call_num_args (stmt); i++)
4377 t = gimple_call_arg_ptr (stmt, i);
4378 modified |= sra_ipa_modify_expr (t, true, adjustments);
4381 if (gimple_call_lhs (stmt))
4383 t = gimple_call_lhs_ptr (stmt);
4384 modified |= sra_ipa_modify_expr (t, false, adjustments);
4385 modified |= replace_removed_params_ssa_names (stmt,
4386 adjustments);
4388 break;
4390 case GIMPLE_ASM:
4391 for (i = 0; i < gimple_asm_ninputs (stmt); i++)
4393 t = &TREE_VALUE (gimple_asm_input_op (stmt, i));
4394 modified |= sra_ipa_modify_expr (t, true, adjustments);
4396 for (i = 0; i < gimple_asm_noutputs (stmt); i++)
4398 t = &TREE_VALUE (gimple_asm_output_op (stmt, i));
4399 modified |= sra_ipa_modify_expr (t, false, adjustments);
4401 break;
4403 default:
4404 break;
4407 if (modified)
4409 update_stmt (stmt);
4410 if (maybe_clean_eh_stmt (stmt)
4411 && gimple_purge_dead_eh_edges (gimple_bb (stmt)))
4412 cfg_changed = true;
4414 gsi_next (&gsi);
4418 return cfg_changed;
4421 /* Call gimple_debug_bind_reset_value on all debug statements describing
4422 gimple register parameters that are being removed or replaced. */
4424 static void
4425 sra_ipa_reset_debug_stmts (ipa_parm_adjustment_vec adjustments)
4427 int i, len;
4428 gimple_stmt_iterator *gsip = NULL, gsi;
4430 if (MAY_HAVE_DEBUG_STMTS && single_succ_p (ENTRY_BLOCK_PTR))
4432 gsi = gsi_after_labels (single_succ (ENTRY_BLOCK_PTR));
4433 gsip = &gsi;
4435 len = VEC_length (ipa_parm_adjustment_t, adjustments);
4436 for (i = 0; i < len; i++)
4438 struct ipa_parm_adjustment *adj;
4439 imm_use_iterator ui;
4440 gimple stmt, def_temp;
4441 tree name, vexpr, copy = NULL_TREE;
4442 use_operand_p use_p;
4444 adj = VEC_index (ipa_parm_adjustment_t, adjustments, i);
4445 if (adj->copy_param || !is_gimple_reg (adj->base))
4446 continue;
4447 name = gimple_default_def (cfun, adj->base);
4448 vexpr = NULL;
4449 if (name)
4450 FOR_EACH_IMM_USE_STMT (stmt, ui, name)
4452 /* All other users must have been removed by
4453 ipa_sra_modify_function_body. */
4454 gcc_assert (is_gimple_debug (stmt));
4455 if (vexpr == NULL && gsip != NULL)
4457 gcc_assert (TREE_CODE (adj->base) == PARM_DECL);
4458 vexpr = make_node (DEBUG_EXPR_DECL);
4459 def_temp = gimple_build_debug_source_bind (vexpr, adj->base,
4460 NULL);
4461 DECL_ARTIFICIAL (vexpr) = 1;
4462 TREE_TYPE (vexpr) = TREE_TYPE (name);
4463 DECL_MODE (vexpr) = DECL_MODE (adj->base);
4464 gsi_insert_before (gsip, def_temp, GSI_SAME_STMT);
4466 if (vexpr)
4468 FOR_EACH_IMM_USE_ON_STMT (use_p, ui)
4469 SET_USE (use_p, vexpr);
4471 else
4472 gimple_debug_bind_reset_value (stmt);
4473 update_stmt (stmt);
4475 /* Create a VAR_DECL for debug info purposes. */
4476 if (!DECL_IGNORED_P (adj->base))
4478 copy = build_decl (DECL_SOURCE_LOCATION (current_function_decl),
4479 VAR_DECL, DECL_NAME (adj->base),
4480 TREE_TYPE (adj->base));
4481 if (DECL_PT_UID_SET_P (adj->base))
4482 SET_DECL_PT_UID (copy, DECL_PT_UID (adj->base));
4483 TREE_ADDRESSABLE (copy) = TREE_ADDRESSABLE (adj->base);
4484 TREE_READONLY (copy) = TREE_READONLY (adj->base);
4485 TREE_THIS_VOLATILE (copy) = TREE_THIS_VOLATILE (adj->base);
4486 DECL_GIMPLE_REG_P (copy) = DECL_GIMPLE_REG_P (adj->base);
4487 DECL_ARTIFICIAL (copy) = DECL_ARTIFICIAL (adj->base);
4488 DECL_IGNORED_P (copy) = DECL_IGNORED_P (adj->base);
4489 DECL_ABSTRACT_ORIGIN (copy) = DECL_ORIGIN (adj->base);
4490 DECL_SEEN_IN_BIND_EXPR_P (copy) = 1;
4491 SET_DECL_RTL (copy, 0);
4492 TREE_USED (copy) = 1;
4493 DECL_CONTEXT (copy) = current_function_decl;
4494 add_referenced_var (copy);
4495 add_local_decl (cfun, copy);
4496 DECL_CHAIN (copy) =
4497 BLOCK_VARS (DECL_INITIAL (current_function_decl));
4498 BLOCK_VARS (DECL_INITIAL (current_function_decl)) = copy;
4500 if (gsip != NULL && copy && target_for_debug_bind (adj->base))
4502 gcc_assert (TREE_CODE (adj->base) == PARM_DECL);
4503 if (vexpr)
4504 def_temp = gimple_build_debug_bind (copy, vexpr, NULL);
4505 else
4506 def_temp = gimple_build_debug_source_bind (copy, adj->base,
4507 NULL);
4508 gsi_insert_before (gsip, def_temp, GSI_SAME_STMT);
4513 /* Return false iff all callers have at least as many actual arguments as there
4514 are formal parameters in the current function. */
4516 static bool
4517 not_all_callers_have_enough_arguments_p (struct cgraph_node *node,
4518 void *data ATTRIBUTE_UNUSED)
4520 struct cgraph_edge *cs;
4521 for (cs = node->callers; cs; cs = cs->next_caller)
4522 if (!callsite_has_enough_arguments_p (cs->call_stmt))
4523 return true;
4525 return false;
4528 /* Convert all callers of NODE. */
4530 static bool
4531 convert_callers_for_node (struct cgraph_node *node,
4532 void *data)
4534 ipa_parm_adjustment_vec adjustments = (ipa_parm_adjustment_vec)data;
4535 bitmap recomputed_callers = BITMAP_ALLOC (NULL);
4536 struct cgraph_edge *cs;
4538 for (cs = node->callers; cs; cs = cs->next_caller)
4540 current_function_decl = cs->caller->decl;
4541 push_cfun (DECL_STRUCT_FUNCTION (cs->caller->decl));
4543 if (dump_file)
4544 fprintf (dump_file, "Adjusting call (%i -> %i) %s -> %s\n",
4545 cs->caller->uid, cs->callee->uid,
4546 cgraph_node_name (cs->caller),
4547 cgraph_node_name (cs->callee));
4549 ipa_modify_call_arguments (cs, cs->call_stmt, adjustments);
4551 pop_cfun ();
4554 for (cs = node->callers; cs; cs = cs->next_caller)
4555 if (bitmap_set_bit (recomputed_callers, cs->caller->uid)
4556 && gimple_in_ssa_p (DECL_STRUCT_FUNCTION (cs->caller->decl)))
4557 compute_inline_parameters (cs->caller, true);
4558 BITMAP_FREE (recomputed_callers);
4560 return true;
4563 /* Convert all callers of NODE to pass parameters as given in ADJUSTMENTS. */
4565 static void
4566 convert_callers (struct cgraph_node *node, tree old_decl,
4567 ipa_parm_adjustment_vec adjustments)
4569 tree old_cur_fndecl = current_function_decl;
4570 basic_block this_block;
4572 cgraph_for_node_and_aliases (node, convert_callers_for_node,
4573 adjustments, false);
4575 current_function_decl = old_cur_fndecl;
4577 if (!encountered_recursive_call)
4578 return;
4580 FOR_EACH_BB (this_block)
4582 gimple_stmt_iterator gsi;
4584 for (gsi = gsi_start_bb (this_block); !gsi_end_p (gsi); gsi_next (&gsi))
4586 gimple stmt = gsi_stmt (gsi);
4587 tree call_fndecl;
4588 if (gimple_code (stmt) != GIMPLE_CALL)
4589 continue;
4590 call_fndecl = gimple_call_fndecl (stmt);
4591 if (call_fndecl == old_decl)
4593 if (dump_file)
4594 fprintf (dump_file, "Adjusting recursive call");
4595 gimple_call_set_fndecl (stmt, node->decl);
4596 ipa_modify_call_arguments (NULL, stmt, adjustments);
4601 return;
4604 /* Perform all the modification required in IPA-SRA for NODE to have parameters
4605 as given in ADJUSTMENTS. Return true iff the CFG has been changed. */
4607 static bool
4608 modify_function (struct cgraph_node *node, ipa_parm_adjustment_vec adjustments)
4610 struct cgraph_node *new_node;
4611 bool cfg_changed;
4612 VEC (cgraph_edge_p, heap) * redirect_callers = collect_callers_of_node (node);
4614 rebuild_cgraph_edges ();
4615 pop_cfun ();
4616 current_function_decl = NULL_TREE;
4618 new_node = cgraph_function_versioning (node, redirect_callers, NULL, NULL,
4619 NULL, NULL, "isra");
4620 current_function_decl = new_node->decl;
4621 push_cfun (DECL_STRUCT_FUNCTION (new_node->decl));
4623 ipa_modify_formal_parameters (current_function_decl, adjustments, "ISRA");
4624 cfg_changed = ipa_sra_modify_function_body (adjustments);
4625 sra_ipa_reset_debug_stmts (adjustments);
4626 convert_callers (new_node, node->decl, adjustments);
4627 cgraph_make_node_local (new_node);
4628 return cfg_changed;
4631 /* Return false the function is apparently unsuitable for IPA-SRA based on it's
4632 attributes, return true otherwise. NODE is the cgraph node of the current
4633 function. */
4635 static bool
4636 ipa_sra_preliminary_function_checks (struct cgraph_node *node)
4638 if (!cgraph_node_can_be_local_p (node))
4640 if (dump_file)
4641 fprintf (dump_file, "Function not local to this compilation unit.\n");
4642 return false;
4645 if (!node->local.can_change_signature)
4647 if (dump_file)
4648 fprintf (dump_file, "Function can not change signature.\n");
4649 return false;
4652 if (!tree_versionable_function_p (node->decl))
4654 if (dump_file)
4655 fprintf (dump_file, "Function is not versionable.\n");
4656 return false;
4659 if (DECL_VIRTUAL_P (current_function_decl))
4661 if (dump_file)
4662 fprintf (dump_file, "Function is a virtual method.\n");
4663 return false;
4666 if ((DECL_COMDAT (node->decl) || DECL_EXTERNAL (node->decl))
4667 && inline_summary(node)->size >= MAX_INLINE_INSNS_AUTO)
4669 if (dump_file)
4670 fprintf (dump_file, "Function too big to be made truly local.\n");
4671 return false;
4674 if (!node->callers)
4676 if (dump_file)
4677 fprintf (dump_file,
4678 "Function has no callers in this compilation unit.\n");
4679 return false;
4682 if (cfun->stdarg)
4684 if (dump_file)
4685 fprintf (dump_file, "Function uses stdarg. \n");
4686 return false;
4689 if (TYPE_ATTRIBUTES (TREE_TYPE (node->decl)))
4690 return false;
4692 return true;
4695 /* Perform early interprocedural SRA. */
4697 static unsigned int
4698 ipa_early_sra (void)
4700 struct cgraph_node *node = cgraph_get_node (current_function_decl);
4701 ipa_parm_adjustment_vec adjustments;
4702 int ret = 0;
4704 if (!ipa_sra_preliminary_function_checks (node))
4705 return 0;
4707 sra_initialize ();
4708 sra_mode = SRA_MODE_EARLY_IPA;
4710 if (!find_param_candidates ())
4712 if (dump_file)
4713 fprintf (dump_file, "Function has no IPA-SRA candidates.\n");
4714 goto simple_out;
4717 if (cgraph_for_node_and_aliases (node, not_all_callers_have_enough_arguments_p,
4718 NULL, true))
4720 if (dump_file)
4721 fprintf (dump_file, "There are callers with insufficient number of "
4722 "arguments.\n");
4723 goto simple_out;
4726 bb_dereferences = XCNEWVEC (HOST_WIDE_INT,
4727 func_param_count
4728 * last_basic_block_for_function (cfun));
4729 final_bbs = BITMAP_ALLOC (NULL);
4731 scan_function ();
4732 if (encountered_apply_args)
4734 if (dump_file)
4735 fprintf (dump_file, "Function calls __builtin_apply_args().\n");
4736 goto out;
4739 if (encountered_unchangable_recursive_call)
4741 if (dump_file)
4742 fprintf (dump_file, "Function calls itself with insufficient "
4743 "number of arguments.\n");
4744 goto out;
4747 adjustments = analyze_all_param_acesses ();
4748 if (!adjustments)
4749 goto out;
4750 if (dump_file)
4751 ipa_dump_param_adjustments (dump_file, adjustments, current_function_decl);
4753 if (modify_function (node, adjustments))
4754 ret = TODO_update_ssa | TODO_cleanup_cfg;
4755 else
4756 ret = TODO_update_ssa;
4757 VEC_free (ipa_parm_adjustment_t, heap, adjustments);
4759 statistics_counter_event (cfun, "Unused parameters deleted",
4760 sra_stats.deleted_unused_parameters);
4761 statistics_counter_event (cfun, "Scalar parameters converted to by-value",
4762 sra_stats.scalar_by_ref_to_by_val);
4763 statistics_counter_event (cfun, "Aggregate parameters broken up",
4764 sra_stats.aggregate_params_reduced);
4765 statistics_counter_event (cfun, "Aggregate parameter components created",
4766 sra_stats.param_reductions_created);
4768 out:
4769 BITMAP_FREE (final_bbs);
4770 free (bb_dereferences);
4771 simple_out:
4772 sra_deinitialize ();
4773 return ret;
4776 /* Return if early ipa sra shall be performed. */
4777 static bool
4778 ipa_early_sra_gate (void)
4780 return flag_ipa_sra && dbg_cnt (eipa_sra);
4783 struct gimple_opt_pass pass_early_ipa_sra =
4786 GIMPLE_PASS,
4787 "eipa_sra", /* name */
4788 ipa_early_sra_gate, /* gate */
4789 ipa_early_sra, /* execute */
4790 NULL, /* sub */
4791 NULL, /* next */
4792 0, /* static_pass_number */
4793 TV_IPA_SRA, /* tv_id */
4794 0, /* properties_required */
4795 0, /* properties_provided */
4796 0, /* properties_destroyed */
4797 0, /* todo_flags_start */
4798 TODO_dump_cgraph /* todo_flags_finish */