1 /* Scalar Replacement of Aggregates (SRA) converts some structure
2 references into scalar references, exposing them to the scalar
4 Copyright (C) 2008, 2009, 2010 Free Software Foundation, Inc.
5 Contributed by Martin Jambor <mjambor@suse.cz>
7 This file is part of GCC.
9 GCC is free software; you can redistribute it and/or modify it under
10 the terms of the GNU General Public License as published by the Free
11 Software Foundation; either version 3, or (at your option) any later
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
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
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
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. */
76 #include "coretypes.h"
77 #include "alloc-pool.h"
82 #include "tree-flow.h"
84 #include "tree-pretty-print.h"
85 #include "statistics.h"
86 #include "tree-dump.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
103 static enum sra_mode sra_mode
;
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. */
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
;
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
139 /* The statement this access belongs to. */
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
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? */
173 /* Is this access an artificial one created to scalarize some record
175 unsigned total_scalarization
: 1;
177 /* Is this access an access to a non-addressable field? */
178 unsigned non_addressable
: 1;
180 /* Is this access currently in the work queue? */
181 unsigned grp_queued
: 1;
183 /* Does this group contain a write access? This flag is propagated down the
185 unsigned grp_write
: 1;
187 /* Does this group contain a read access? This flag is propagated down the
189 unsigned grp_read
: 1;
191 /* Does this group contain a read access that comes from an assignment
192 statement? This flag is propagated down the access tree. */
193 unsigned grp_assignment_read
: 1;
195 /* Does this group contain a write access that comes from an assignment
196 statement? This flag is propagated down the access tree. */
197 unsigned grp_assignment_write
: 1;
199 /* Does this group contain a read access through a scalar type? This flag is
200 not propagated in the access tree in any direction. */
201 unsigned grp_scalar_read
: 1;
203 /* Does this group contain a write access through a scalar type? This flag
204 is not propagated in the access tree in any direction. */
205 unsigned grp_scalar_write
: 1;
207 /* Other passes of the analysis use this bit to make function
208 analyze_access_subtree create scalar replacements for this group if
210 unsigned grp_hint
: 1;
212 /* Is the subtree rooted in this access fully covered by scalar
214 unsigned grp_covered
: 1;
216 /* If set to true, this access and all below it in an access tree must not be
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
223 unsigned grp_unscalarized_data
: 1;
225 /* Does this access and/or group contain a write access through a
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
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. */
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. */
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. */
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. */
334 /* Number of times sra_modify_expr or sra_modify_assign themselves changed an
338 /* Number of statements created by generate_subtree_copies. */
341 /* Number of statements created by load_assign_lhs_subreplacements. */
344 /* Number of times sra_modify_assign has deleted a statement. */
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
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
361 int aggregate_params_reduced
;
363 /* Numbber of components created when splitting aggregate parameters. */
364 int param_reductions_created
;
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);
380 fprintf (f
, ", total_scalarization = %d, grp_read = %d, grp_write = %d, "
381 "grp_assignment_read = %d, grp_assignment_write = %d, "
382 "grp_scalar_read = %d, grp_scalar_write = %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
->total_scalarization
, access
->grp_read
, access
->grp_write
,
389 access
->grp_assignment_read
, access
->grp_assignment_write
,
390 access
->grp_scalar_read
, access
->grp_scalar_write
,
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
);
397 fprintf (f
, ", write = %d, total_scalarization = %d, "
398 "grp_partial_lhs = %d\n",
399 access
->write
, access
->total_scalarization
,
400 access
->grp_partial_lhs
);
403 /* Dump a subtree rooted in ACCESS to file F, indent by LEVEL. */
406 dump_access_tree_1 (FILE *f
, struct access
*access
, int level
)
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
;
425 /* Dump all access trees for a variable, given the pointer to the first root in
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. */
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
)
451 slot
= pointer_map_contains (base_access_vec
, base
);
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
,
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
;
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
);
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
,
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
;
507 return find_access_in_subtree (access
, offset
, size
);
510 /* Add LINK to the linked list of assign links of RACC. */
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
;
522 racc
->last_link
->next
= link
;
524 racc
->last_link
= link
;
528 /* Move all link structures in their linked list in OLD_RACC to the linked list
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
);
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
;
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). */
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;
585 /* Allocate necessary structures. */
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. */
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
);
616 /* Deallocate all general structures. */
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
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
651 type_internals_preclude_sra_p (tree type
)
656 switch (TREE_CODE (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
)
667 || !DECL_FIELD_OFFSET (fld
) || !DECL_SIZE (fld
)
668 || !host_integerp (DECL_FIELD_OFFSET (fld
), 1)
669 || !host_integerp (DECL_SIZE (fld
), 1)
670 || (AGGREGATE_TYPE_P (ft
)
671 && int_bit_position (fld
) % BITS_PER_UNIT
!= 0))
674 if (AGGREGATE_TYPE_P (ft
)
675 && type_internals_preclude_sra_p (ft
))
682 et
= TREE_TYPE (type
);
684 if (AGGREGATE_TYPE_P (et
))
685 return type_internals_preclude_sra_p (et
);
694 /* If T is an SSA_NAME, return NULL if it is not a default def or return its
695 base variable if it is. Return T if it is not an SSA_NAME. */
698 get_ssa_base_param (tree t
)
700 if (TREE_CODE (t
) == SSA_NAME
)
702 if (SSA_NAME_IS_DEFAULT_DEF (t
))
703 return SSA_NAME_VAR (t
);
710 /* Mark a dereference of BASE of distance DIST in a basic block tht STMT
711 belongs to, unless the BB has already been marked as a potentially
715 mark_parm_dereference (tree base
, HOST_WIDE_INT dist
, gimple stmt
)
717 basic_block bb
= gimple_bb (stmt
);
718 int idx
, parm_index
= 0;
721 if (bitmap_bit_p (final_bbs
, bb
->index
))
724 for (parm
= DECL_ARGUMENTS (current_function_decl
);
725 parm
&& parm
!= base
;
726 parm
= DECL_CHAIN (parm
))
729 gcc_assert (parm_index
< func_param_count
);
731 idx
= bb
->index
* func_param_count
+ parm_index
;
732 if (bb_dereferences
[idx
] < dist
)
733 bb_dereferences
[idx
] = dist
;
736 /* Allocate an access structure for BASE, OFFSET and SIZE, clear it, fill in
737 the three fields. Also add it to the vector of accesses corresponding to
738 the base. Finally, return the new access. */
740 static struct access
*
741 create_access_1 (tree base
, HOST_WIDE_INT offset
, HOST_WIDE_INT size
)
743 VEC (access_p
, heap
) *vec
;
744 struct access
*access
;
747 access
= (struct access
*) pool_alloc (access_pool
);
748 memset (access
, 0, sizeof (struct access
));
750 access
->offset
= offset
;
753 slot
= pointer_map_contains (base_access_vec
, base
);
755 vec
= (VEC (access_p
, heap
) *) *slot
;
757 vec
= VEC_alloc (access_p
, heap
, 32);
759 VEC_safe_push (access_p
, heap
, vec
, access
);
761 *((struct VEC (access_p
,heap
) **)
762 pointer_map_insert (base_access_vec
, base
)) = vec
;
767 /* Create and insert access for EXPR. Return created access, or NULL if it is
770 static struct access
*
771 create_access (tree expr
, gimple stmt
, bool write
)
773 struct access
*access
;
774 HOST_WIDE_INT offset
, size
, max_size
;
776 bool ptr
, unscalarizable_region
= false;
778 base
= get_ref_base_and_extent (expr
, &offset
, &size
, &max_size
);
780 if (sra_mode
== SRA_MODE_EARLY_IPA
781 && TREE_CODE (base
) == MEM_REF
)
783 base
= get_ssa_base_param (TREE_OPERAND (base
, 0));
791 if (!DECL_P (base
) || !bitmap_bit_p (candidate_bitmap
, DECL_UID (base
)))
794 if (sra_mode
== SRA_MODE_EARLY_IPA
)
796 if (size
< 0 || size
!= max_size
)
798 disqualify_candidate (base
, "Encountered a variable sized access.");
801 if (TREE_CODE (expr
) == COMPONENT_REF
802 && DECL_BIT_FIELD (TREE_OPERAND (expr
, 1)))
804 disqualify_candidate (base
, "Encountered a bit-field access.");
807 gcc_checking_assert ((offset
% BITS_PER_UNIT
) == 0);
810 mark_parm_dereference (base
, offset
+ size
, stmt
);
814 if (size
!= max_size
)
817 unscalarizable_region
= true;
821 disqualify_candidate (base
, "Encountered an unconstrained access.");
826 access
= create_access_1 (base
, offset
, size
);
828 access
->type
= TREE_TYPE (expr
);
829 access
->write
= write
;
830 access
->grp_unscalarizable_region
= unscalarizable_region
;
833 if (TREE_CODE (expr
) == COMPONENT_REF
834 && DECL_NONADDRESSABLE_P (TREE_OPERAND (expr
, 1)))
835 access
->non_addressable
= 1;
841 /* Return true iff TYPE is a RECORD_TYPE with fields that are either of gimple
842 register types or (recursively) records with only these two kinds of fields.
843 It also returns false if any of these records contains a bit-field. */
846 type_consists_of_records_p (tree type
)
850 if (TREE_CODE (type
) != RECORD_TYPE
)
853 for (fld
= TYPE_FIELDS (type
); fld
; fld
= DECL_CHAIN (fld
))
854 if (TREE_CODE (fld
) == FIELD_DECL
)
856 tree ft
= TREE_TYPE (fld
);
858 if (DECL_BIT_FIELD (fld
))
861 if (!is_gimple_reg_type (ft
)
862 && !type_consists_of_records_p (ft
))
869 /* Create total_scalarization accesses for all scalar type fields in DECL that
870 must be of a RECORD_TYPE conforming to type_consists_of_records_p. BASE
871 must be the top-most VAR_DECL representing the variable, OFFSET must be the
872 offset of DECL within BASE. REF must be the memory reference expression for
876 completely_scalarize_record (tree base
, tree decl
, HOST_WIDE_INT offset
,
879 tree fld
, decl_type
= TREE_TYPE (decl
);
881 for (fld
= TYPE_FIELDS (decl_type
); fld
; fld
= DECL_CHAIN (fld
))
882 if (TREE_CODE (fld
) == FIELD_DECL
)
884 HOST_WIDE_INT pos
= offset
+ int_bit_position (fld
);
885 tree ft
= TREE_TYPE (fld
);
886 tree nref
= build3 (COMPONENT_REF
, TREE_TYPE (fld
), ref
, fld
,
889 if (is_gimple_reg_type (ft
))
891 struct access
*access
;
894 size
= tree_low_cst (DECL_SIZE (fld
), 1);
895 access
= create_access_1 (base
, pos
, size
);
898 access
->total_scalarization
= 1;
899 /* Accesses for intraprocedural SRA can have their stmt NULL. */
902 completely_scalarize_record (base
, fld
, pos
, nref
);
907 /* Search the given tree for a declaration by skipping handled components and
908 exclude it from the candidates. */
911 disqualify_base_of_expr (tree t
, const char *reason
)
913 t
= get_base_address (t
);
914 if (sra_mode
== SRA_MODE_EARLY_IPA
915 && TREE_CODE (t
) == MEM_REF
)
916 t
= get_ssa_base_param (TREE_OPERAND (t
, 0));
919 disqualify_candidate (t
, reason
);
922 /* Scan expression EXPR and create access structures for all accesses to
923 candidates for scalarization. Return the created access or NULL if none is
926 static struct access
*
927 build_access_from_expr_1 (tree expr
, gimple stmt
, bool write
)
929 struct access
*ret
= NULL
;
932 if (TREE_CODE (expr
) == BIT_FIELD_REF
933 || TREE_CODE (expr
) == IMAGPART_EXPR
934 || TREE_CODE (expr
) == REALPART_EXPR
)
936 expr
= TREE_OPERAND (expr
, 0);
942 /* We need to dive through V_C_Es in order to get the size of its parameter
943 and not the result type. Ada produces such statements. We are also
944 capable of handling the topmost V_C_E but not any of those buried in other
945 handled components. */
946 if (TREE_CODE (expr
) == VIEW_CONVERT_EXPR
)
947 expr
= TREE_OPERAND (expr
, 0);
949 if (contains_view_convert_expr_p (expr
))
951 disqualify_base_of_expr (expr
, "V_C_E under a different handled "
956 switch (TREE_CODE (expr
))
959 if (TREE_CODE (TREE_OPERAND (expr
, 0)) != ADDR_EXPR
960 && sra_mode
!= SRA_MODE_EARLY_IPA
)
968 case ARRAY_RANGE_REF
:
969 ret
= create_access (expr
, stmt
, write
);
976 if (write
&& partial_ref
&& ret
)
977 ret
->grp_partial_lhs
= 1;
982 /* Scan expression EXPR and create access structures for all accesses to
983 candidates for scalarization. Return true if any access has been inserted.
984 STMT must be the statement from which the expression is taken, WRITE must be
985 true if the expression is a store and false otherwise. */
988 build_access_from_expr (tree expr
, gimple stmt
, bool write
)
990 struct access
*access
;
992 access
= build_access_from_expr_1 (expr
, stmt
, write
);
995 /* This means the aggregate is accesses as a whole in a way other than an
996 assign statement and thus cannot be removed even if we had a scalar
997 replacement for everything. */
998 if (cannot_scalarize_away_bitmap
)
999 bitmap_set_bit (cannot_scalarize_away_bitmap
, DECL_UID (access
->base
));
1005 /* Disqualify LHS and RHS for scalarization if STMT must end its basic block in
1006 modes in which it matters, return true iff they have been disqualified. RHS
1007 may be NULL, in that case ignore it. If we scalarize an aggregate in
1008 intra-SRA we may need to add statements after each statement. This is not
1009 possible if a statement unconditionally has to end the basic block. */
1011 disqualify_ops_if_throwing_stmt (gimple stmt
, tree lhs
, tree rhs
)
1013 if ((sra_mode
== SRA_MODE_EARLY_INTRA
|| sra_mode
== SRA_MODE_INTRA
)
1014 && (stmt_can_throw_internal (stmt
) || stmt_ends_bb_p (stmt
)))
1016 disqualify_base_of_expr (lhs
, "LHS of a throwing stmt.");
1018 disqualify_base_of_expr (rhs
, "RHS of a throwing stmt.");
1024 /* Scan expressions occuring in STMT, create access structures for all accesses
1025 to candidates for scalarization and remove those candidates which occur in
1026 statements or expressions that prevent them from being split apart. Return
1027 true if any access has been inserted. */
1030 build_accesses_from_assign (gimple stmt
)
1033 struct access
*lacc
, *racc
;
1035 if (!gimple_assign_single_p (stmt
))
1038 lhs
= gimple_assign_lhs (stmt
);
1039 rhs
= gimple_assign_rhs1 (stmt
);
1041 if (disqualify_ops_if_throwing_stmt (stmt
, lhs
, rhs
))
1044 racc
= build_access_from_expr_1 (rhs
, stmt
, false);
1045 lacc
= build_access_from_expr_1 (lhs
, stmt
, true);
1048 lacc
->grp_assignment_write
= 1;
1052 racc
->grp_assignment_read
= 1;
1053 if (should_scalarize_away_bitmap
&& !gimple_has_volatile_ops (stmt
)
1054 && !is_gimple_reg_type (racc
->type
))
1055 bitmap_set_bit (should_scalarize_away_bitmap
, DECL_UID (racc
->base
));
1059 && (sra_mode
== SRA_MODE_EARLY_INTRA
|| sra_mode
== SRA_MODE_INTRA
)
1060 && !lacc
->grp_unscalarizable_region
1061 && !racc
->grp_unscalarizable_region
1062 && AGGREGATE_TYPE_P (TREE_TYPE (lhs
))
1063 /* FIXME: Turn the following line into an assert after PR 40058 is
1065 && lacc
->size
== racc
->size
1066 && useless_type_conversion_p (lacc
->type
, racc
->type
))
1068 struct assign_link
*link
;
1070 link
= (struct assign_link
*) pool_alloc (link_pool
);
1071 memset (link
, 0, sizeof (struct assign_link
));
1076 add_link_to_rhs (racc
, link
);
1079 return lacc
|| racc
;
1082 /* Callback of walk_stmt_load_store_addr_ops visit_addr used to determine
1083 GIMPLE_ASM operands with memory constrains which cannot be scalarized. */
1086 asm_visit_addr (gimple stmt ATTRIBUTE_UNUSED
, tree op
,
1087 void *data ATTRIBUTE_UNUSED
)
1089 op
= get_base_address (op
);
1092 disqualify_candidate (op
, "Non-scalarizable GIMPLE_ASM operand.");
1097 /* Return true iff callsite CALL has at least as many actual arguments as there
1098 are formal parameters of the function currently processed by IPA-SRA. */
1101 callsite_has_enough_arguments_p (gimple call
)
1103 return gimple_call_num_args (call
) >= (unsigned) func_param_count
;
1106 /* Scan function and look for interesting expressions and create access
1107 structures for them. Return true iff any access is created. */
1110 scan_function (void)
1117 gimple_stmt_iterator gsi
;
1118 for (gsi
= gsi_start_bb (bb
); !gsi_end_p (gsi
); gsi_next (&gsi
))
1120 gimple stmt
= gsi_stmt (gsi
);
1124 if (final_bbs
&& stmt_can_throw_external (stmt
))
1125 bitmap_set_bit (final_bbs
, bb
->index
);
1126 switch (gimple_code (stmt
))
1129 t
= gimple_return_retval (stmt
);
1131 ret
|= build_access_from_expr (t
, stmt
, false);
1133 bitmap_set_bit (final_bbs
, bb
->index
);
1137 ret
|= build_accesses_from_assign (stmt
);
1141 for (i
= 0; i
< gimple_call_num_args (stmt
); i
++)
1142 ret
|= build_access_from_expr (gimple_call_arg (stmt
, i
),
1145 if (sra_mode
== SRA_MODE_EARLY_IPA
)
1147 tree dest
= gimple_call_fndecl (stmt
);
1148 int flags
= gimple_call_flags (stmt
);
1152 if (DECL_BUILT_IN_CLASS (dest
) == BUILT_IN_NORMAL
1153 && DECL_FUNCTION_CODE (dest
) == BUILT_IN_APPLY_ARGS
)
1154 encountered_apply_args
= true;
1155 if (cgraph_get_node (dest
)
1156 == cgraph_get_node (current_function_decl
))
1158 encountered_recursive_call
= true;
1159 if (!callsite_has_enough_arguments_p (stmt
))
1160 encountered_unchangable_recursive_call
= true;
1165 && (flags
& (ECF_CONST
| ECF_PURE
)) == 0)
1166 bitmap_set_bit (final_bbs
, bb
->index
);
1169 t
= gimple_call_lhs (stmt
);
1170 if (t
&& !disqualify_ops_if_throwing_stmt (stmt
, t
, NULL
))
1171 ret
|= build_access_from_expr (t
, stmt
, true);
1175 walk_stmt_load_store_addr_ops (stmt
, NULL
, NULL
, NULL
,
1178 bitmap_set_bit (final_bbs
, bb
->index
);
1180 for (i
= 0; i
< gimple_asm_ninputs (stmt
); i
++)
1182 t
= TREE_VALUE (gimple_asm_input_op (stmt
, i
));
1183 ret
|= build_access_from_expr (t
, stmt
, false);
1185 for (i
= 0; i
< gimple_asm_noutputs (stmt
); i
++)
1187 t
= TREE_VALUE (gimple_asm_output_op (stmt
, i
));
1188 ret
|= build_access_from_expr (t
, stmt
, true);
1201 /* Helper of QSORT function. There are pointers to accesses in the array. An
1202 access is considered smaller than another if it has smaller offset or if the
1203 offsets are the same but is size is bigger. */
1206 compare_access_positions (const void *a
, const void *b
)
1208 const access_p
*fp1
= (const access_p
*) a
;
1209 const access_p
*fp2
= (const access_p
*) b
;
1210 const access_p f1
= *fp1
;
1211 const access_p f2
= *fp2
;
1213 if (f1
->offset
!= f2
->offset
)
1214 return f1
->offset
< f2
->offset
? -1 : 1;
1216 if (f1
->size
== f2
->size
)
1218 if (f1
->type
== f2
->type
)
1220 /* Put any non-aggregate type before any aggregate type. */
1221 else if (!is_gimple_reg_type (f1
->type
)
1222 && is_gimple_reg_type (f2
->type
))
1224 else if (is_gimple_reg_type (f1
->type
)
1225 && !is_gimple_reg_type (f2
->type
))
1227 /* Put any complex or vector type before any other scalar type. */
1228 else if (TREE_CODE (f1
->type
) != COMPLEX_TYPE
1229 && TREE_CODE (f1
->type
) != VECTOR_TYPE
1230 && (TREE_CODE (f2
->type
) == COMPLEX_TYPE
1231 || TREE_CODE (f2
->type
) == VECTOR_TYPE
))
1233 else if ((TREE_CODE (f1
->type
) == COMPLEX_TYPE
1234 || TREE_CODE (f1
->type
) == VECTOR_TYPE
)
1235 && TREE_CODE (f2
->type
) != COMPLEX_TYPE
1236 && TREE_CODE (f2
->type
) != VECTOR_TYPE
)
1238 /* Put the integral type with the bigger precision first. */
1239 else if (INTEGRAL_TYPE_P (f1
->type
)
1240 && INTEGRAL_TYPE_P (f2
->type
))
1241 return TYPE_PRECISION (f2
->type
) - TYPE_PRECISION (f1
->type
);
1242 /* Put any integral type with non-full precision last. */
1243 else if (INTEGRAL_TYPE_P (f1
->type
)
1244 && (TREE_INT_CST_LOW (TYPE_SIZE (f1
->type
))
1245 != TYPE_PRECISION (f1
->type
)))
1247 else if (INTEGRAL_TYPE_P (f2
->type
)
1248 && (TREE_INT_CST_LOW (TYPE_SIZE (f2
->type
))
1249 != TYPE_PRECISION (f2
->type
)))
1251 /* Stabilize the sort. */
1252 return TYPE_UID (f1
->type
) - TYPE_UID (f2
->type
);
1255 /* We want the bigger accesses first, thus the opposite operator in the next
1257 return f1
->size
> f2
->size
? -1 : 1;
1261 /* Append a name of the declaration to the name obstack. A helper function for
1265 make_fancy_decl_name (tree decl
)
1269 tree name
= DECL_NAME (decl
);
1271 obstack_grow (&name_obstack
, IDENTIFIER_POINTER (name
),
1272 IDENTIFIER_LENGTH (name
));
1275 sprintf (buffer
, "D%u", DECL_UID (decl
));
1276 obstack_grow (&name_obstack
, buffer
, strlen (buffer
));
1280 /* Helper for make_fancy_name. */
1283 make_fancy_name_1 (tree expr
)
1290 make_fancy_decl_name (expr
);
1294 switch (TREE_CODE (expr
))
1297 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1298 obstack_1grow (&name_obstack
, '$');
1299 make_fancy_decl_name (TREE_OPERAND (expr
, 1));
1303 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1304 obstack_1grow (&name_obstack
, '$');
1305 /* Arrays with only one element may not have a constant as their
1307 index
= TREE_OPERAND (expr
, 1);
1308 if (TREE_CODE (index
) != INTEGER_CST
)
1310 sprintf (buffer
, HOST_WIDE_INT_PRINT_DEC
, TREE_INT_CST_LOW (index
));
1311 obstack_grow (&name_obstack
, buffer
, strlen (buffer
));
1315 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1319 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1320 if (!integer_zerop (TREE_OPERAND (expr
, 1)))
1322 obstack_1grow (&name_obstack
, '$');
1323 sprintf (buffer
, HOST_WIDE_INT_PRINT_DEC
,
1324 TREE_INT_CST_LOW (TREE_OPERAND (expr
, 1)));
1325 obstack_grow (&name_obstack
, buffer
, strlen (buffer
));
1332 gcc_unreachable (); /* we treat these as scalars. */
1339 /* Create a human readable name for replacement variable of ACCESS. */
1342 make_fancy_name (tree expr
)
1344 make_fancy_name_1 (expr
);
1345 obstack_1grow (&name_obstack
, '\0');
1346 return XOBFINISH (&name_obstack
, char *);
1349 /* Construct a MEM_REF that would reference a part of aggregate BASE of type
1350 EXP_TYPE at the given OFFSET. If BASE is something for which
1351 get_addr_base_and_unit_offset returns NULL, gsi must be non-NULL and is used
1352 to insert new statements either before or below the current one as specified
1353 by INSERT_AFTER. This function is not capable of handling bitfields. */
1356 build_ref_for_offset (location_t loc
, tree base
, HOST_WIDE_INT offset
,
1357 tree exp_type
, gimple_stmt_iterator
*gsi
,
1360 tree prev_base
= base
;
1362 HOST_WIDE_INT base_offset
;
1364 gcc_checking_assert (offset
% BITS_PER_UNIT
== 0);
1366 base
= get_addr_base_and_unit_offset (base
, &base_offset
);
1368 /* get_addr_base_and_unit_offset returns NULL for references with a variable
1369 offset such as array[var_index]. */
1375 gcc_checking_assert (gsi
);
1376 tmp
= create_tmp_reg (build_pointer_type (TREE_TYPE (prev_base
)), NULL
);
1377 add_referenced_var (tmp
);
1378 tmp
= make_ssa_name (tmp
, NULL
);
1379 addr
= build_fold_addr_expr (unshare_expr (prev_base
));
1380 STRIP_USELESS_TYPE_CONVERSION (addr
);
1381 stmt
= gimple_build_assign (tmp
, addr
);
1382 gimple_set_location (stmt
, loc
);
1383 SSA_NAME_DEF_STMT (tmp
) = stmt
;
1385 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
1387 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
1390 off
= build_int_cst (reference_alias_ptr_type (prev_base
),
1391 offset
/ BITS_PER_UNIT
);
1394 else if (TREE_CODE (base
) == MEM_REF
)
1396 off
= build_int_cst (TREE_TYPE (TREE_OPERAND (base
, 1)),
1397 base_offset
+ offset
/ BITS_PER_UNIT
);
1398 off
= int_const_binop (PLUS_EXPR
, TREE_OPERAND (base
, 1), off
);
1399 base
= unshare_expr (TREE_OPERAND (base
, 0));
1403 off
= build_int_cst (reference_alias_ptr_type (base
),
1404 base_offset
+ offset
/ BITS_PER_UNIT
);
1405 base
= build_fold_addr_expr (unshare_expr (base
));
1408 return fold_build2_loc (loc
, MEM_REF
, exp_type
, base
, off
);
1411 /* Construct a memory reference to a part of an aggregate BASE at the given
1412 OFFSET and of the same type as MODEL. In case this is a reference to a
1413 component, the function will replicate the last COMPONENT_REF of model's
1414 expr to access it. GSI and INSERT_AFTER have the same meaning as in
1415 build_ref_for_offset. */
1418 build_ref_for_model (location_t loc
, tree base
, HOST_WIDE_INT offset
,
1419 struct access
*model
, gimple_stmt_iterator
*gsi
,
1422 if (TREE_CODE (model
->expr
) == COMPONENT_REF
)
1425 offset
-= int_bit_position (TREE_OPERAND (model
->expr
, 1));
1426 exp_type
= TREE_TYPE (TREE_OPERAND (model
->expr
, 0));
1427 t
= build_ref_for_offset (loc
, base
, offset
, exp_type
, gsi
, insert_after
);
1428 return fold_build3_loc (loc
, COMPONENT_REF
, model
->type
, t
,
1429 TREE_OPERAND (model
->expr
, 1), NULL_TREE
);
1432 return build_ref_for_offset (loc
, base
, offset
, model
->type
,
1436 /* Construct a memory reference consisting of component_refs and array_refs to
1437 a part of an aggregate *RES (which is of type TYPE). The requested part
1438 should have type EXP_TYPE at be the given OFFSET. This function might not
1439 succeed, it returns true when it does and only then *RES points to something
1440 meaningful. This function should be used only to build expressions that we
1441 might need to present to user (e.g. in warnings). In all other situations,
1442 build_ref_for_model or build_ref_for_offset should be used instead. */
1445 build_user_friendly_ref_for_offset (tree
*res
, tree type
, HOST_WIDE_INT offset
,
1451 tree tr_size
, index
, minidx
;
1452 HOST_WIDE_INT el_size
;
1454 if (offset
== 0 && exp_type
1455 && types_compatible_p (exp_type
, type
))
1458 switch (TREE_CODE (type
))
1461 case QUAL_UNION_TYPE
:
1463 for (fld
= TYPE_FIELDS (type
); fld
; fld
= DECL_CHAIN (fld
))
1465 HOST_WIDE_INT pos
, size
;
1466 tree expr
, *expr_ptr
;
1468 if (TREE_CODE (fld
) != FIELD_DECL
)
1471 pos
= int_bit_position (fld
);
1472 gcc_assert (TREE_CODE (type
) == RECORD_TYPE
|| pos
== 0);
1473 tr_size
= DECL_SIZE (fld
);
1474 if (!tr_size
|| !host_integerp (tr_size
, 1))
1476 size
= tree_low_cst (tr_size
, 1);
1482 else if (pos
> offset
|| (pos
+ size
) <= offset
)
1485 expr
= build3 (COMPONENT_REF
, TREE_TYPE (fld
), *res
, fld
,
1488 if (build_user_friendly_ref_for_offset (expr_ptr
, TREE_TYPE (fld
),
1489 offset
- pos
, exp_type
))
1498 tr_size
= TYPE_SIZE (TREE_TYPE (type
));
1499 if (!tr_size
|| !host_integerp (tr_size
, 1))
1501 el_size
= tree_low_cst (tr_size
, 1);
1503 minidx
= TYPE_MIN_VALUE (TYPE_DOMAIN (type
));
1504 if (TREE_CODE (minidx
) != INTEGER_CST
|| el_size
== 0)
1506 index
= build_int_cst (TYPE_DOMAIN (type
), offset
/ el_size
);
1507 if (!integer_zerop (minidx
))
1508 index
= int_const_binop (PLUS_EXPR
, index
, minidx
);
1509 *res
= build4 (ARRAY_REF
, TREE_TYPE (type
), *res
, index
,
1510 NULL_TREE
, NULL_TREE
);
1511 offset
= offset
% el_size
;
1512 type
= TREE_TYPE (type
);
1527 /* Return true iff TYPE is stdarg va_list type. */
1530 is_va_list_type (tree type
)
1532 return TYPE_MAIN_VARIANT (type
) == TYPE_MAIN_VARIANT (va_list_type_node
);
1535 /* The very first phase of intraprocedural SRA. It marks in candidate_bitmap
1536 those with type which is suitable for scalarization. */
1539 find_var_candidates (void)
1542 referenced_var_iterator rvi
;
1545 FOR_EACH_REFERENCED_VAR (cfun
, var
, rvi
)
1547 if (TREE_CODE (var
) != VAR_DECL
&& TREE_CODE (var
) != PARM_DECL
)
1549 type
= TREE_TYPE (var
);
1551 if (!AGGREGATE_TYPE_P (type
)
1552 || needs_to_live_in_memory (var
)
1553 || TREE_THIS_VOLATILE (var
)
1554 || !COMPLETE_TYPE_P (type
)
1555 || !host_integerp (TYPE_SIZE (type
), 1)
1556 || tree_low_cst (TYPE_SIZE (type
), 1) == 0
1557 || type_internals_preclude_sra_p (type
)
1558 /* Fix for PR 41089. tree-stdarg.c needs to have va_lists intact but
1559 we also want to schedule it rather late. Thus we ignore it in
1561 || (sra_mode
== SRA_MODE_EARLY_INTRA
1562 && is_va_list_type (type
)))
1565 bitmap_set_bit (candidate_bitmap
, DECL_UID (var
));
1567 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1569 fprintf (dump_file
, "Candidate (%d): ", DECL_UID (var
));
1570 print_generic_expr (dump_file
, var
, 0);
1571 fprintf (dump_file
, "\n");
1579 /* Sort all accesses for the given variable, check for partial overlaps and
1580 return NULL if there are any. If there are none, pick a representative for
1581 each combination of offset and size and create a linked list out of them.
1582 Return the pointer to the first representative and make sure it is the first
1583 one in the vector of accesses. */
1585 static struct access
*
1586 sort_and_splice_var_accesses (tree var
)
1588 int i
, j
, access_count
;
1589 struct access
*res
, **prev_acc_ptr
= &res
;
1590 VEC (access_p
, heap
) *access_vec
;
1592 HOST_WIDE_INT low
= -1, high
= 0;
1594 access_vec
= get_base_access_vector (var
);
1597 access_count
= VEC_length (access_p
, access_vec
);
1599 /* Sort by <OFFSET, SIZE>. */
1600 VEC_qsort (access_p
, access_vec
, compare_access_positions
);
1603 while (i
< access_count
)
1605 struct access
*access
= VEC_index (access_p
, access_vec
, i
);
1606 bool grp_write
= access
->write
;
1607 bool grp_read
= !access
->write
;
1608 bool grp_scalar_write
= access
->write
1609 && is_gimple_reg_type (access
->type
);
1610 bool grp_scalar_read
= !access
->write
1611 && is_gimple_reg_type (access
->type
);
1612 bool grp_assignment_read
= access
->grp_assignment_read
;
1613 bool grp_assignment_write
= access
->grp_assignment_write
;
1614 bool multiple_scalar_reads
= false;
1615 bool total_scalarization
= access
->total_scalarization
;
1616 bool grp_partial_lhs
= access
->grp_partial_lhs
;
1617 bool first_scalar
= is_gimple_reg_type (access
->type
);
1618 bool unscalarizable_region
= access
->grp_unscalarizable_region
;
1620 if (first
|| access
->offset
>= high
)
1623 low
= access
->offset
;
1624 high
= access
->offset
+ access
->size
;
1626 else if (access
->offset
> low
&& access
->offset
+ access
->size
> high
)
1629 gcc_assert (access
->offset
>= low
1630 && access
->offset
+ access
->size
<= high
);
1633 while (j
< access_count
)
1635 struct access
*ac2
= VEC_index (access_p
, access_vec
, j
);
1636 if (ac2
->offset
!= access
->offset
|| ac2
->size
!= access
->size
)
1641 grp_scalar_write
= (grp_scalar_write
1642 || is_gimple_reg_type (ac2
->type
));
1647 if (is_gimple_reg_type (ac2
->type
))
1649 if (grp_scalar_read
)
1650 multiple_scalar_reads
= true;
1652 grp_scalar_read
= true;
1655 grp_assignment_read
|= ac2
->grp_assignment_read
;
1656 grp_assignment_write
|= ac2
->grp_assignment_write
;
1657 grp_partial_lhs
|= ac2
->grp_partial_lhs
;
1658 unscalarizable_region
|= ac2
->grp_unscalarizable_region
;
1659 total_scalarization
|= ac2
->total_scalarization
;
1660 relink_to_new_repr (access
, ac2
);
1662 /* If there are both aggregate-type and scalar-type accesses with
1663 this combination of size and offset, the comparison function
1664 should have put the scalars first. */
1665 gcc_assert (first_scalar
|| !is_gimple_reg_type (ac2
->type
));
1666 ac2
->group_representative
= access
;
1672 access
->group_representative
= access
;
1673 access
->grp_write
= grp_write
;
1674 access
->grp_read
= grp_read
;
1675 access
->grp_scalar_read
= grp_scalar_read
;
1676 access
->grp_scalar_write
= grp_scalar_write
;
1677 access
->grp_assignment_read
= grp_assignment_read
;
1678 access
->grp_assignment_write
= grp_assignment_write
;
1679 access
->grp_hint
= multiple_scalar_reads
|| total_scalarization
;
1680 access
->grp_partial_lhs
= grp_partial_lhs
;
1681 access
->grp_unscalarizable_region
= unscalarizable_region
;
1682 if (access
->first_link
)
1683 add_access_to_work_queue (access
);
1685 *prev_acc_ptr
= access
;
1686 prev_acc_ptr
= &access
->next_grp
;
1689 gcc_assert (res
== VEC_index (access_p
, access_vec
, 0));
1693 /* Create a variable for the given ACCESS which determines the type, name and a
1694 few other properties. Return the variable declaration and store it also to
1695 ACCESS->replacement. */
1698 create_access_replacement (struct access
*access
, bool rename
)
1702 repl
= create_tmp_var (access
->type
, "SR");
1704 add_referenced_var (repl
);
1706 mark_sym_for_renaming (repl
);
1708 if (!access
->grp_partial_lhs
1709 && (TREE_CODE (access
->type
) == COMPLEX_TYPE
1710 || TREE_CODE (access
->type
) == VECTOR_TYPE
))
1711 DECL_GIMPLE_REG_P (repl
) = 1;
1713 DECL_SOURCE_LOCATION (repl
) = DECL_SOURCE_LOCATION (access
->base
);
1714 DECL_ARTIFICIAL (repl
) = 1;
1715 DECL_IGNORED_P (repl
) = DECL_IGNORED_P (access
->base
);
1717 if (DECL_NAME (access
->base
)
1718 && !DECL_IGNORED_P (access
->base
)
1719 && !DECL_ARTIFICIAL (access
->base
))
1721 char *pretty_name
= make_fancy_name (access
->expr
);
1722 tree debug_expr
= unshare_expr (access
->expr
), d
;
1724 DECL_NAME (repl
) = get_identifier (pretty_name
);
1725 obstack_free (&name_obstack
, pretty_name
);
1727 /* Get rid of any SSA_NAMEs embedded in debug_expr,
1728 as DECL_DEBUG_EXPR isn't considered when looking for still
1729 used SSA_NAMEs and thus they could be freed. All debug info
1730 generation cares is whether something is constant or variable
1731 and that get_ref_base_and_extent works properly on the
1733 for (d
= debug_expr
; handled_component_p (d
); d
= TREE_OPERAND (d
, 0))
1734 switch (TREE_CODE (d
))
1737 case ARRAY_RANGE_REF
:
1738 if (TREE_OPERAND (d
, 1)
1739 && TREE_CODE (TREE_OPERAND (d
, 1)) == SSA_NAME
)
1740 TREE_OPERAND (d
, 1) = SSA_NAME_VAR (TREE_OPERAND (d
, 1));
1741 if (TREE_OPERAND (d
, 3)
1742 && TREE_CODE (TREE_OPERAND (d
, 3)) == SSA_NAME
)
1743 TREE_OPERAND (d
, 3) = SSA_NAME_VAR (TREE_OPERAND (d
, 3));
1746 if (TREE_OPERAND (d
, 2)
1747 && TREE_CODE (TREE_OPERAND (d
, 2)) == SSA_NAME
)
1748 TREE_OPERAND (d
, 2) = SSA_NAME_VAR (TREE_OPERAND (d
, 2));
1753 SET_DECL_DEBUG_EXPR (repl
, debug_expr
);
1754 DECL_DEBUG_EXPR_IS_FROM (repl
) = 1;
1755 if (access
->grp_no_warning
)
1756 TREE_NO_WARNING (repl
) = 1;
1758 TREE_NO_WARNING (repl
) = TREE_NO_WARNING (access
->base
);
1761 TREE_NO_WARNING (repl
) = 1;
1765 fprintf (dump_file
, "Created a replacement for ");
1766 print_generic_expr (dump_file
, access
->base
, 0);
1767 fprintf (dump_file
, " offset: %u, size: %u: ",
1768 (unsigned) access
->offset
, (unsigned) access
->size
);
1769 print_generic_expr (dump_file
, repl
, 0);
1770 fprintf (dump_file
, "\n");
1772 sra_stats
.replacements
++;
1777 /* Return ACCESS scalar replacement, create it if it does not exist yet. */
1780 get_access_replacement (struct access
*access
)
1782 gcc_assert (access
->grp_to_be_replaced
);
1784 if (!access
->replacement_decl
)
1785 access
->replacement_decl
= create_access_replacement (access
, true);
1786 return access
->replacement_decl
;
1789 /* Return ACCESS scalar replacement, create it if it does not exist yet but do
1790 not mark it for renaming. */
1793 get_unrenamed_access_replacement (struct access
*access
)
1795 gcc_assert (!access
->grp_to_be_replaced
);
1797 if (!access
->replacement_decl
)
1798 access
->replacement_decl
= create_access_replacement (access
, false);
1799 return access
->replacement_decl
;
1803 /* Build a subtree of accesses rooted in *ACCESS, and move the pointer in the
1804 linked list along the way. Stop when *ACCESS is NULL or the access pointed
1805 to it is not "within" the root. Return false iff some accesses partially
1809 build_access_subtree (struct access
**access
)
1811 struct access
*root
= *access
, *last_child
= NULL
;
1812 HOST_WIDE_INT limit
= root
->offset
+ root
->size
;
1814 *access
= (*access
)->next_grp
;
1815 while (*access
&& (*access
)->offset
+ (*access
)->size
<= limit
)
1818 root
->first_child
= *access
;
1820 last_child
->next_sibling
= *access
;
1821 last_child
= *access
;
1823 if (!build_access_subtree (access
))
1827 if (*access
&& (*access
)->offset
< limit
)
1833 /* Build a tree of access representatives, ACCESS is the pointer to the first
1834 one, others are linked in a list by the next_grp field. Return false iff
1835 some accesses partially overlap. */
1838 build_access_trees (struct access
*access
)
1842 struct access
*root
= access
;
1844 if (!build_access_subtree (&access
))
1846 root
->next_grp
= access
;
1851 /* Return true if expr contains some ARRAY_REFs into a variable bounded
1855 expr_with_var_bounded_array_refs_p (tree expr
)
1857 while (handled_component_p (expr
))
1859 if (TREE_CODE (expr
) == ARRAY_REF
1860 && !host_integerp (array_ref_low_bound (expr
), 0))
1862 expr
= TREE_OPERAND (expr
, 0);
1867 enum mark_rw_status
{ SRA_MRRW_NOTHING
, SRA_MRRW_DIRECT
, SRA_MRRW_ASSIGN
};
1869 /* Analyze the subtree of accesses rooted in ROOT, scheduling replacements when
1870 both seeming beneficial and when ALLOW_REPLACEMENTS allows it. Also set all
1871 sorts of access flags appropriately along the way, notably always set
1872 grp_read and grp_assign_read according to MARK_READ and grp_write when
1875 Creating a replacement for a scalar access is considered beneficial if its
1876 grp_hint is set (this means we are either attempting total scalarization or
1877 there is more than one direct read access) or according to the following
1880 Access written to through a scalar type (once or more times)
1882 | Written to in an assignment statement
1884 | | Access read as scalar _once_
1886 | | | Read in an assignment statement
1888 | | | | Scalarize Comment
1889 -----------------------------------------------------------------------------
1890 0 0 0 0 No access for the scalar
1891 0 0 0 1 No access for the scalar
1892 0 0 1 0 No Single read - won't help
1893 0 0 1 1 No The same case
1894 0 1 0 0 No access for the scalar
1895 0 1 0 1 No access for the scalar
1896 0 1 1 0 Yes s = *g; return s.i;
1897 0 1 1 1 Yes The same case as above
1898 1 0 0 0 No Won't help
1899 1 0 0 1 Yes s.i = 1; *g = s;
1900 1 0 1 0 Yes s.i = 5; g = s.i;
1901 1 0 1 1 Yes The same case as above
1902 1 1 0 0 No Won't help.
1903 1 1 0 1 Yes s.i = 1; *g = s;
1904 1 1 1 0 Yes s = *g; return s.i;
1905 1 1 1 1 Yes Any of the above yeses */
1908 analyze_access_subtree (struct access
*root
, bool allow_replacements
,
1909 enum mark_rw_status mark_read
,
1910 enum mark_rw_status mark_write
)
1912 struct access
*child
;
1913 HOST_WIDE_INT limit
= root
->offset
+ root
->size
;
1914 HOST_WIDE_INT covered_to
= root
->offset
;
1915 bool scalar
= is_gimple_reg_type (root
->type
);
1916 bool hole
= false, sth_created
= false;
1918 if (root
->grp_assignment_read
)
1919 mark_read
= SRA_MRRW_ASSIGN
;
1920 else if (mark_read
== SRA_MRRW_ASSIGN
)
1923 root
->grp_assignment_read
= 1;
1925 else if (mark_read
== SRA_MRRW_DIRECT
)
1927 else if (root
->grp_read
)
1928 mark_read
= SRA_MRRW_DIRECT
;
1930 if (root
->grp_assignment_write
)
1931 mark_write
= SRA_MRRW_ASSIGN
;
1932 else if (mark_write
== SRA_MRRW_ASSIGN
)
1934 root
->grp_write
= 1;
1935 root
->grp_assignment_write
= 1;
1937 else if (mark_write
== SRA_MRRW_DIRECT
)
1938 root
->grp_write
= 1;
1939 else if (root
->grp_write
)
1940 mark_write
= SRA_MRRW_DIRECT
;
1942 if (root
->grp_unscalarizable_region
)
1943 allow_replacements
= false;
1945 if (allow_replacements
&& expr_with_var_bounded_array_refs_p (root
->expr
))
1946 allow_replacements
= false;
1948 for (child
= root
->first_child
; child
; child
= child
->next_sibling
)
1950 if (!hole
&& child
->offset
< covered_to
)
1953 covered_to
+= child
->size
;
1955 sth_created
|= analyze_access_subtree (child
,
1956 allow_replacements
&& !scalar
,
1957 mark_read
, mark_write
);
1959 root
->grp_unscalarized_data
|= child
->grp_unscalarized_data
;
1960 hole
|= !child
->grp_covered
;
1963 if (allow_replacements
&& scalar
&& !root
->first_child
1965 || ((root
->grp_scalar_read
|| root
->grp_assignment_read
)
1966 && (root
->grp_scalar_write
|| root
->grp_assignment_write
))))
1968 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1970 fprintf (dump_file
, "Marking ");
1971 print_generic_expr (dump_file
, root
->base
, 0);
1972 fprintf (dump_file
, " offset: %u, size: %u: ",
1973 (unsigned) root
->offset
, (unsigned) root
->size
);
1974 fprintf (dump_file
, " to be replaced.\n");
1977 root
->grp_to_be_replaced
= 1;
1981 else if (covered_to
< limit
)
1984 if (sth_created
&& !hole
)
1986 root
->grp_covered
= 1;
1989 if (root
->grp_write
|| TREE_CODE (root
->base
) == PARM_DECL
)
1990 root
->grp_unscalarized_data
= 1; /* not covered and written to */
1996 /* Analyze all access trees linked by next_grp by the means of
1997 analyze_access_subtree. */
1999 analyze_access_trees (struct access
*access
)
2005 if (analyze_access_subtree (access
, true,
2006 SRA_MRRW_NOTHING
, SRA_MRRW_NOTHING
))
2008 access
= access
->next_grp
;
2014 /* Return true iff a potential new child of LACC at offset OFFSET and with size
2015 SIZE would conflict with an already existing one. If exactly such a child
2016 already exists in LACC, store a pointer to it in EXACT_MATCH. */
2019 child_would_conflict_in_lacc (struct access
*lacc
, HOST_WIDE_INT norm_offset
,
2020 HOST_WIDE_INT size
, struct access
**exact_match
)
2022 struct access
*child
;
2024 for (child
= lacc
->first_child
; child
; child
= child
->next_sibling
)
2026 if (child
->offset
== norm_offset
&& child
->size
== size
)
2028 *exact_match
= child
;
2032 if (child
->offset
< norm_offset
+ size
2033 && child
->offset
+ child
->size
> norm_offset
)
2040 /* Create a new child access of PARENT, with all properties just like MODEL
2041 except for its offset and with its grp_write false and grp_read true.
2042 Return the new access or NULL if it cannot be created. Note that this access
2043 is created long after all splicing and sorting, it's not located in any
2044 access vector and is automatically a representative of its group. */
2046 static struct access
*
2047 create_artificial_child_access (struct access
*parent
, struct access
*model
,
2048 HOST_WIDE_INT new_offset
)
2050 struct access
*access
;
2051 struct access
**child
;
2052 tree expr
= parent
->base
;
2054 gcc_assert (!model
->grp_unscalarizable_region
);
2056 access
= (struct access
*) pool_alloc (access_pool
);
2057 memset (access
, 0, sizeof (struct access
));
2058 if (!build_user_friendly_ref_for_offset (&expr
, TREE_TYPE (expr
), new_offset
,
2061 access
->grp_no_warning
= true;
2062 expr
= build_ref_for_model (EXPR_LOCATION (parent
->base
), parent
->base
,
2063 new_offset
, model
, NULL
, false);
2066 access
->base
= parent
->base
;
2067 access
->expr
= expr
;
2068 access
->offset
= new_offset
;
2069 access
->size
= model
->size
;
2070 access
->type
= model
->type
;
2071 access
->grp_write
= true;
2072 access
->grp_read
= false;
2074 child
= &parent
->first_child
;
2075 while (*child
&& (*child
)->offset
< new_offset
)
2076 child
= &(*child
)->next_sibling
;
2078 access
->next_sibling
= *child
;
2085 /* Propagate all subaccesses of RACC across an assignment link to LACC. Return
2086 true if any new subaccess was created. Additionally, if RACC is a scalar
2087 access but LACC is not, change the type of the latter, if possible. */
2090 propagate_subaccesses_across_link (struct access
*lacc
, struct access
*racc
)
2092 struct access
*rchild
;
2093 HOST_WIDE_INT norm_delta
= lacc
->offset
- racc
->offset
;
2096 if (is_gimple_reg_type (lacc
->type
)
2097 || lacc
->grp_unscalarizable_region
2098 || racc
->grp_unscalarizable_region
)
2101 if (!lacc
->first_child
&& !racc
->first_child
2102 && is_gimple_reg_type (racc
->type
))
2104 tree t
= lacc
->base
;
2106 lacc
->type
= racc
->type
;
2107 if (build_user_friendly_ref_for_offset (&t
, TREE_TYPE (t
), lacc
->offset
,
2112 lacc
->expr
= build_ref_for_model (EXPR_LOCATION (lacc
->base
),
2113 lacc
->base
, lacc
->offset
,
2115 lacc
->grp_no_warning
= true;
2120 for (rchild
= racc
->first_child
; rchild
; rchild
= rchild
->next_sibling
)
2122 struct access
*new_acc
= NULL
;
2123 HOST_WIDE_INT norm_offset
= rchild
->offset
+ norm_delta
;
2125 if (rchild
->grp_unscalarizable_region
)
2128 if (child_would_conflict_in_lacc (lacc
, norm_offset
, rchild
->size
,
2133 rchild
->grp_hint
= 1;
2134 new_acc
->grp_hint
|= new_acc
->grp_read
;
2135 if (rchild
->first_child
)
2136 ret
|= propagate_subaccesses_across_link (new_acc
, rchild
);
2141 rchild
->grp_hint
= 1;
2142 new_acc
= create_artificial_child_access (lacc
, rchild
, norm_offset
);
2146 if (racc
->first_child
)
2147 propagate_subaccesses_across_link (new_acc
, rchild
);
2154 /* Propagate all subaccesses across assignment links. */
2157 propagate_all_subaccesses (void)
2159 while (work_queue_head
)
2161 struct access
*racc
= pop_access_from_work_queue ();
2162 struct assign_link
*link
;
2164 gcc_assert (racc
->first_link
);
2166 for (link
= racc
->first_link
; link
; link
= link
->next
)
2168 struct access
*lacc
= link
->lacc
;
2170 if (!bitmap_bit_p (candidate_bitmap
, DECL_UID (lacc
->base
)))
2172 lacc
= lacc
->group_representative
;
2173 if (propagate_subaccesses_across_link (lacc
, racc
)
2174 && lacc
->first_link
)
2175 add_access_to_work_queue (lacc
);
2180 /* Go through all accesses collected throughout the (intraprocedural) analysis
2181 stage, exclude overlapping ones, identify representatives and build trees
2182 out of them, making decisions about scalarization on the way. Return true
2183 iff there are any to-be-scalarized variables after this stage. */
2186 analyze_all_variable_accesses (void)
2189 bitmap tmp
= BITMAP_ALLOC (NULL
);
2191 unsigned i
, max_total_scalarization_size
;
2193 max_total_scalarization_size
= UNITS_PER_WORD
* BITS_PER_UNIT
2194 * MOVE_RATIO (optimize_function_for_speed_p (cfun
));
2196 EXECUTE_IF_SET_IN_BITMAP (candidate_bitmap
, 0, i
, bi
)
2197 if (bitmap_bit_p (should_scalarize_away_bitmap
, i
)
2198 && !bitmap_bit_p (cannot_scalarize_away_bitmap
, i
))
2200 tree var
= referenced_var (i
);
2202 if (TREE_CODE (var
) == VAR_DECL
2203 && ((unsigned) tree_low_cst (TYPE_SIZE (TREE_TYPE (var
)), 1)
2204 <= max_total_scalarization_size
)
2205 && type_consists_of_records_p (TREE_TYPE (var
)))
2207 completely_scalarize_record (var
, var
, 0, var
);
2208 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2210 fprintf (dump_file
, "Will attempt to totally scalarize ");
2211 print_generic_expr (dump_file
, var
, 0);
2212 fprintf (dump_file
, " (UID: %u): \n", DECL_UID (var
));
2217 bitmap_copy (tmp
, candidate_bitmap
);
2218 EXECUTE_IF_SET_IN_BITMAP (tmp
, 0, i
, bi
)
2220 tree var
= referenced_var (i
);
2221 struct access
*access
;
2223 access
= sort_and_splice_var_accesses (var
);
2224 if (!access
|| !build_access_trees (access
))
2225 disqualify_candidate (var
,
2226 "No or inhibitingly overlapping accesses.");
2229 propagate_all_subaccesses ();
2231 bitmap_copy (tmp
, candidate_bitmap
);
2232 EXECUTE_IF_SET_IN_BITMAP (tmp
, 0, i
, bi
)
2234 tree var
= referenced_var (i
);
2235 struct access
*access
= get_first_repr_for_decl (var
);
2237 if (analyze_access_trees (access
))
2240 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2242 fprintf (dump_file
, "\nAccess trees for ");
2243 print_generic_expr (dump_file
, var
, 0);
2244 fprintf (dump_file
, " (UID: %u): \n", DECL_UID (var
));
2245 dump_access_tree (dump_file
, access
);
2246 fprintf (dump_file
, "\n");
2250 disqualify_candidate (var
, "No scalar replacements to be created.");
2257 statistics_counter_event (cfun
, "Scalarized aggregates", res
);
2264 /* Generate statements copying scalar replacements of accesses within a subtree
2265 into or out of AGG. ACCESS, all its children, siblings and their children
2266 are to be processed. AGG is an aggregate type expression (can be a
2267 declaration but does not have to be, it can for example also be a mem_ref or
2268 a series of handled components). TOP_OFFSET is the offset of the processed
2269 subtree which has to be subtracted from offsets of individual accesses to
2270 get corresponding offsets for AGG. If CHUNK_SIZE is non-null, copy only
2271 replacements in the interval <start_offset, start_offset + chunk_size>,
2272 otherwise copy all. GSI is a statement iterator used to place the new
2273 statements. WRITE should be true when the statements should write from AGG
2274 to the replacement and false if vice versa. if INSERT_AFTER is true, new
2275 statements will be added after the current statement in GSI, they will be
2276 added before the statement otherwise. */
2279 generate_subtree_copies (struct access
*access
, tree agg
,
2280 HOST_WIDE_INT top_offset
,
2281 HOST_WIDE_INT start_offset
, HOST_WIDE_INT chunk_size
,
2282 gimple_stmt_iterator
*gsi
, bool write
,
2283 bool insert_after
, location_t loc
)
2287 if (chunk_size
&& access
->offset
>= start_offset
+ chunk_size
)
2290 if (access
->grp_to_be_replaced
2292 || access
->offset
+ access
->size
> start_offset
))
2294 tree expr
, repl
= get_access_replacement (access
);
2297 expr
= build_ref_for_model (loc
, agg
, access
->offset
- top_offset
,
2298 access
, gsi
, insert_after
);
2302 if (access
->grp_partial_lhs
)
2303 expr
= force_gimple_operand_gsi (gsi
, expr
, true, NULL_TREE
,
2305 insert_after
? GSI_NEW_STMT
2307 stmt
= gimple_build_assign (repl
, expr
);
2311 TREE_NO_WARNING (repl
) = 1;
2312 if (access
->grp_partial_lhs
)
2313 repl
= force_gimple_operand_gsi (gsi
, repl
, true, NULL_TREE
,
2315 insert_after
? GSI_NEW_STMT
2317 stmt
= gimple_build_assign (expr
, repl
);
2319 gimple_set_location (stmt
, loc
);
2322 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
2324 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
2326 sra_stats
.subtree_copies
++;
2329 if (access
->first_child
)
2330 generate_subtree_copies (access
->first_child
, agg
, top_offset
,
2331 start_offset
, chunk_size
, gsi
,
2332 write
, insert_after
, loc
);
2334 access
= access
->next_sibling
;
2339 /* Assign zero to all scalar replacements in an access subtree. ACCESS is the
2340 the root of the subtree to be processed. GSI is the statement iterator used
2341 for inserting statements which are added after the current statement if
2342 INSERT_AFTER is true or before it otherwise. */
2345 init_subtree_with_zero (struct access
*access
, gimple_stmt_iterator
*gsi
,
2346 bool insert_after
, location_t loc
)
2349 struct access
*child
;
2351 if (access
->grp_to_be_replaced
)
2355 stmt
= gimple_build_assign (get_access_replacement (access
),
2356 build_zero_cst (access
->type
));
2358 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
2360 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
2362 gimple_set_location (stmt
, loc
);
2365 for (child
= access
->first_child
; child
; child
= child
->next_sibling
)
2366 init_subtree_with_zero (child
, gsi
, insert_after
, loc
);
2369 /* Search for an access representative for the given expression EXPR and
2370 return it or NULL if it cannot be found. */
2372 static struct access
*
2373 get_access_for_expr (tree expr
)
2375 HOST_WIDE_INT offset
, size
, max_size
;
2378 /* FIXME: This should not be necessary but Ada produces V_C_Es with a type of
2379 a different size than the size of its argument and we need the latter
2381 if (TREE_CODE (expr
) == VIEW_CONVERT_EXPR
)
2382 expr
= TREE_OPERAND (expr
, 0);
2384 base
= get_ref_base_and_extent (expr
, &offset
, &size
, &max_size
);
2385 if (max_size
== -1 || !DECL_P (base
))
2388 if (!bitmap_bit_p (candidate_bitmap
, DECL_UID (base
)))
2391 return get_var_base_offset_size_access (base
, offset
, max_size
);
2394 /* Replace the expression EXPR with a scalar replacement if there is one and
2395 generate other statements to do type conversion or subtree copying if
2396 necessary. GSI is used to place newly created statements, WRITE is true if
2397 the expression is being written to (it is on a LHS of a statement or output
2398 in an assembly statement). */
2401 sra_modify_expr (tree
*expr
, gimple_stmt_iterator
*gsi
, bool write
)
2404 struct access
*access
;
2407 if (TREE_CODE (*expr
) == BIT_FIELD_REF
)
2410 expr
= &TREE_OPERAND (*expr
, 0);
2415 if (TREE_CODE (*expr
) == REALPART_EXPR
|| TREE_CODE (*expr
) == IMAGPART_EXPR
)
2416 expr
= &TREE_OPERAND (*expr
, 0);
2417 access
= get_access_for_expr (*expr
);
2420 type
= TREE_TYPE (*expr
);
2422 loc
= gimple_location (gsi_stmt (*gsi
));
2423 if (access
->grp_to_be_replaced
)
2425 tree repl
= get_access_replacement (access
);
2426 /* If we replace a non-register typed access simply use the original
2427 access expression to extract the scalar component afterwards.
2428 This happens if scalarizing a function return value or parameter
2429 like in gcc.c-torture/execute/20041124-1.c, 20050316-1.c and
2430 gcc.c-torture/compile/20011217-1.c.
2432 We also want to use this when accessing a complex or vector which can
2433 be accessed as a different type too, potentially creating a need for
2434 type conversion (see PR42196) and when scalarized unions are involved
2435 in assembler statements (see PR42398). */
2436 if (!useless_type_conversion_p (type
, access
->type
))
2440 ref
= build_ref_for_model (loc
, access
->base
, access
->offset
, access
,
2447 if (access
->grp_partial_lhs
)
2448 ref
= force_gimple_operand_gsi (gsi
, ref
, true, NULL_TREE
,
2449 false, GSI_NEW_STMT
);
2450 stmt
= gimple_build_assign (repl
, ref
);
2451 gimple_set_location (stmt
, loc
);
2452 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
2458 if (access
->grp_partial_lhs
)
2459 repl
= force_gimple_operand_gsi (gsi
, repl
, true, NULL_TREE
,
2460 true, GSI_SAME_STMT
);
2461 stmt
= gimple_build_assign (ref
, repl
);
2462 gimple_set_location (stmt
, loc
);
2463 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
2471 if (access
->first_child
)
2473 HOST_WIDE_INT start_offset
, chunk_size
;
2475 && host_integerp (TREE_OPERAND (bfr
, 1), 1)
2476 && host_integerp (TREE_OPERAND (bfr
, 2), 1))
2478 chunk_size
= tree_low_cst (TREE_OPERAND (bfr
, 1), 1);
2479 start_offset
= access
->offset
2480 + tree_low_cst (TREE_OPERAND (bfr
, 2), 1);
2483 start_offset
= chunk_size
= 0;
2485 generate_subtree_copies (access
->first_child
, access
->base
, 0,
2486 start_offset
, chunk_size
, gsi
, write
, write
,
2492 /* Where scalar replacements of the RHS have been written to when a replacement
2493 of a LHS of an assigments cannot be direclty loaded from a replacement of
2495 enum unscalarized_data_handling
{ SRA_UDH_NONE
, /* Nothing done so far. */
2496 SRA_UDH_RIGHT
, /* Data flushed to the RHS. */
2497 SRA_UDH_LEFT
}; /* Data flushed to the LHS. */
2499 /* Store all replacements in the access tree rooted in TOP_RACC either to their
2500 base aggregate if there are unscalarized data or directly to LHS of the
2501 statement that is pointed to by GSI otherwise. */
2503 static enum unscalarized_data_handling
2504 handle_unscalarized_data_in_subtree (struct access
*top_racc
,
2505 gimple_stmt_iterator
*gsi
)
2507 if (top_racc
->grp_unscalarized_data
)
2509 generate_subtree_copies (top_racc
->first_child
, top_racc
->base
, 0, 0, 0,
2511 gimple_location (gsi_stmt (*gsi
)));
2512 return SRA_UDH_RIGHT
;
2516 tree lhs
= gimple_assign_lhs (gsi_stmt (*gsi
));
2517 generate_subtree_copies (top_racc
->first_child
, lhs
, top_racc
->offset
,
2518 0, 0, gsi
, false, false,
2519 gimple_location (gsi_stmt (*gsi
)));
2520 return SRA_UDH_LEFT
;
2525 /* Try to generate statements to load all sub-replacements in an access subtree
2526 formed by children of LACC from scalar replacements in the TOP_RACC subtree.
2527 If that is not possible, refresh the TOP_RACC base aggregate and load the
2528 accesses from it. LEFT_OFFSET is the offset of the left whole subtree being
2529 copied. NEW_GSI is stmt iterator used for statement insertions after the
2530 original assignment, OLD_GSI is used to insert statements before the
2531 assignment. *REFRESHED keeps the information whether we have needed to
2532 refresh replacements of the LHS and from which side of the assignments this
2536 load_assign_lhs_subreplacements (struct access
*lacc
, struct access
*top_racc
,
2537 HOST_WIDE_INT left_offset
,
2538 gimple_stmt_iterator
*old_gsi
,
2539 gimple_stmt_iterator
*new_gsi
,
2540 enum unscalarized_data_handling
*refreshed
)
2542 location_t loc
= gimple_location (gsi_stmt (*old_gsi
));
2543 for (lacc
= lacc
->first_child
; lacc
; lacc
= lacc
->next_sibling
)
2545 if (lacc
->grp_to_be_replaced
)
2547 struct access
*racc
;
2548 HOST_WIDE_INT offset
= lacc
->offset
- left_offset
+ top_racc
->offset
;
2552 racc
= find_access_in_subtree (top_racc
, offset
, lacc
->size
);
2553 if (racc
&& racc
->grp_to_be_replaced
)
2555 rhs
= get_access_replacement (racc
);
2556 if (!useless_type_conversion_p (lacc
->type
, racc
->type
))
2557 rhs
= fold_build1_loc (loc
, VIEW_CONVERT_EXPR
, lacc
->type
, rhs
);
2561 /* No suitable access on the right hand side, need to load from
2562 the aggregate. See if we have to update it first... */
2563 if (*refreshed
== SRA_UDH_NONE
)
2564 *refreshed
= handle_unscalarized_data_in_subtree (top_racc
,
2567 if (*refreshed
== SRA_UDH_LEFT
)
2568 rhs
= build_ref_for_model (loc
, lacc
->base
, lacc
->offset
, lacc
,
2571 rhs
= build_ref_for_model (loc
, top_racc
->base
, offset
, lacc
,
2575 stmt
= gimple_build_assign (get_access_replacement (lacc
), rhs
);
2576 gsi_insert_after (new_gsi
, stmt
, GSI_NEW_STMT
);
2577 gimple_set_location (stmt
, loc
);
2579 sra_stats
.subreplacements
++;
2581 else if (*refreshed
== SRA_UDH_NONE
2582 && lacc
->grp_read
&& !lacc
->grp_covered
)
2583 *refreshed
= handle_unscalarized_data_in_subtree (top_racc
,
2586 if (lacc
->first_child
)
2587 load_assign_lhs_subreplacements (lacc
, top_racc
, left_offset
,
2588 old_gsi
, new_gsi
, refreshed
);
2592 /* Result code for SRA assignment modification. */
2593 enum assignment_mod_result
{ SRA_AM_NONE
, /* nothing done for the stmt */
2594 SRA_AM_MODIFIED
, /* stmt changed but not
2596 SRA_AM_REMOVED
}; /* stmt eliminated */
2598 /* Modify assignments with a CONSTRUCTOR on their RHS. STMT contains a pointer
2599 to the assignment and GSI is the statement iterator pointing at it. Returns
2600 the same values as sra_modify_assign. */
2602 static enum assignment_mod_result
2603 sra_modify_constructor_assign (gimple
*stmt
, gimple_stmt_iterator
*gsi
)
2605 tree lhs
= gimple_assign_lhs (*stmt
);
2609 acc
= get_access_for_expr (lhs
);
2613 loc
= gimple_location (*stmt
);
2614 if (VEC_length (constructor_elt
,
2615 CONSTRUCTOR_ELTS (gimple_assign_rhs1 (*stmt
))) > 0)
2617 /* I have never seen this code path trigger but if it can happen the
2618 following should handle it gracefully. */
2619 if (access_has_children_p (acc
))
2620 generate_subtree_copies (acc
->first_child
, acc
->base
, 0, 0, 0, gsi
,
2622 return SRA_AM_MODIFIED
;
2625 if (acc
->grp_covered
)
2627 init_subtree_with_zero (acc
, gsi
, false, loc
);
2628 unlink_stmt_vdef (*stmt
);
2629 gsi_remove (gsi
, true);
2630 return SRA_AM_REMOVED
;
2634 init_subtree_with_zero (acc
, gsi
, true, loc
);
2635 return SRA_AM_MODIFIED
;
2639 /* Create and return a new suitable default definition SSA_NAME for RACC which
2640 is an access describing an uninitialized part of an aggregate that is being
2644 get_repl_default_def_ssa_name (struct access
*racc
)
2648 decl
= get_unrenamed_access_replacement (racc
);
2650 repl
= gimple_default_def (cfun
, decl
);
2653 repl
= make_ssa_name (decl
, gimple_build_nop ());
2654 set_default_def (decl
, repl
);
2660 /* Return true if REF has a COMPONENT_REF with a bit-field field declaration
2664 contains_bitfld_comp_ref_p (const_tree ref
)
2666 while (handled_component_p (ref
))
2668 if (TREE_CODE (ref
) == COMPONENT_REF
2669 && DECL_BIT_FIELD (TREE_OPERAND (ref
, 1)))
2671 ref
= TREE_OPERAND (ref
, 0);
2677 /* Return true if REF has an VIEW_CONVERT_EXPR or a COMPONENT_REF with a
2678 bit-field field declaration somewhere in it. */
2681 contains_vce_or_bfcref_p (const_tree ref
)
2683 while (handled_component_p (ref
))
2685 if (TREE_CODE (ref
) == VIEW_CONVERT_EXPR
2686 || (TREE_CODE (ref
) == COMPONENT_REF
2687 && DECL_BIT_FIELD (TREE_OPERAND (ref
, 1))))
2689 ref
= TREE_OPERAND (ref
, 0);
2695 /* Examine both sides of the assignment statement pointed to by STMT, replace
2696 them with a scalare replacement if there is one and generate copying of
2697 replacements if scalarized aggregates have been used in the assignment. GSI
2698 is used to hold generated statements for type conversions and subtree
2701 static enum assignment_mod_result
2702 sra_modify_assign (gimple
*stmt
, gimple_stmt_iterator
*gsi
)
2704 struct access
*lacc
, *racc
;
2706 bool modify_this_stmt
= false;
2707 bool force_gimple_rhs
= false;
2709 gimple_stmt_iterator orig_gsi
= *gsi
;
2711 if (!gimple_assign_single_p (*stmt
))
2713 lhs
= gimple_assign_lhs (*stmt
);
2714 rhs
= gimple_assign_rhs1 (*stmt
);
2716 if (TREE_CODE (rhs
) == CONSTRUCTOR
)
2717 return sra_modify_constructor_assign (stmt
, gsi
);
2719 if (TREE_CODE (rhs
) == REALPART_EXPR
|| TREE_CODE (lhs
) == REALPART_EXPR
2720 || TREE_CODE (rhs
) == IMAGPART_EXPR
|| TREE_CODE (lhs
) == IMAGPART_EXPR
2721 || TREE_CODE (rhs
) == BIT_FIELD_REF
|| TREE_CODE (lhs
) == BIT_FIELD_REF
)
2723 modify_this_stmt
= sra_modify_expr (gimple_assign_rhs1_ptr (*stmt
),
2725 modify_this_stmt
|= sra_modify_expr (gimple_assign_lhs_ptr (*stmt
),
2727 return modify_this_stmt
? SRA_AM_MODIFIED
: SRA_AM_NONE
;
2730 lacc
= get_access_for_expr (lhs
);
2731 racc
= get_access_for_expr (rhs
);
2735 loc
= gimple_location (*stmt
);
2736 if (lacc
&& lacc
->grp_to_be_replaced
)
2738 lhs
= get_access_replacement (lacc
);
2739 gimple_assign_set_lhs (*stmt
, lhs
);
2740 modify_this_stmt
= true;
2741 if (lacc
->grp_partial_lhs
)
2742 force_gimple_rhs
= true;
2746 if (racc
&& racc
->grp_to_be_replaced
)
2748 rhs
= get_access_replacement (racc
);
2749 modify_this_stmt
= true;
2750 if (racc
->grp_partial_lhs
)
2751 force_gimple_rhs
= true;
2755 if (modify_this_stmt
)
2757 if (!useless_type_conversion_p (TREE_TYPE (lhs
), TREE_TYPE (rhs
)))
2759 /* If we can avoid creating a VIEW_CONVERT_EXPR do so.
2760 ??? This should move to fold_stmt which we simply should
2761 call after building a VIEW_CONVERT_EXPR here. */
2762 if (AGGREGATE_TYPE_P (TREE_TYPE (lhs
))
2763 && !contains_bitfld_comp_ref_p (lhs
)
2764 && !access_has_children_p (lacc
))
2766 lhs
= build_ref_for_model (loc
, lhs
, 0, racc
, gsi
, false);
2767 gimple_assign_set_lhs (*stmt
, lhs
);
2769 else if (AGGREGATE_TYPE_P (TREE_TYPE (rhs
))
2770 && !contains_vce_or_bfcref_p (rhs
)
2771 && !access_has_children_p (racc
))
2772 rhs
= build_ref_for_model (loc
, rhs
, 0, lacc
, gsi
, false);
2774 if (!useless_type_conversion_p (TREE_TYPE (lhs
), TREE_TYPE (rhs
)))
2776 rhs
= fold_build1_loc (loc
, VIEW_CONVERT_EXPR
, TREE_TYPE (lhs
),
2778 if (is_gimple_reg_type (TREE_TYPE (lhs
))
2779 && TREE_CODE (lhs
) != SSA_NAME
)
2780 force_gimple_rhs
= true;
2785 /* From this point on, the function deals with assignments in between
2786 aggregates when at least one has scalar reductions of some of its
2787 components. There are three possible scenarios: Both the LHS and RHS have
2788 to-be-scalarized components, 2) only the RHS has or 3) only the LHS has.
2790 In the first case, we would like to load the LHS components from RHS
2791 components whenever possible. If that is not possible, we would like to
2792 read it directly from the RHS (after updating it by storing in it its own
2793 components). If there are some necessary unscalarized data in the LHS,
2794 those will be loaded by the original assignment too. If neither of these
2795 cases happen, the original statement can be removed. Most of this is done
2796 by load_assign_lhs_subreplacements.
2798 In the second case, we would like to store all RHS scalarized components
2799 directly into LHS and if they cover the aggregate completely, remove the
2800 statement too. In the third case, we want the LHS components to be loaded
2801 directly from the RHS (DSE will remove the original statement if it
2804 This is a bit complex but manageable when types match and when unions do
2805 not cause confusion in a way that we cannot really load a component of LHS
2806 from the RHS or vice versa (the access representing this level can have
2807 subaccesses that are accessible only through a different union field at a
2808 higher level - different from the one used in the examined expression).
2811 Therefore, I specially handle a fourth case, happening when there is a
2812 specific type cast or it is impossible to locate a scalarized subaccess on
2813 the other side of the expression. If that happens, I simply "refresh" the
2814 RHS by storing in it is scalarized components leave the original statement
2815 there to do the copying and then load the scalar replacements of the LHS.
2816 This is what the first branch does. */
2818 if (gimple_has_volatile_ops (*stmt
)
2819 || contains_vce_or_bfcref_p (rhs
)
2820 || contains_vce_or_bfcref_p (lhs
))
2822 if (access_has_children_p (racc
))
2823 generate_subtree_copies (racc
->first_child
, racc
->base
, 0, 0, 0,
2824 gsi
, false, false, loc
);
2825 if (access_has_children_p (lacc
))
2826 generate_subtree_copies (lacc
->first_child
, lacc
->base
, 0, 0, 0,
2827 gsi
, true, true, loc
);
2828 sra_stats
.separate_lhs_rhs_handling
++;
2832 if (access_has_children_p (lacc
) && access_has_children_p (racc
))
2834 gimple_stmt_iterator orig_gsi
= *gsi
;
2835 enum unscalarized_data_handling refreshed
;
2837 if (lacc
->grp_read
&& !lacc
->grp_covered
)
2838 refreshed
= handle_unscalarized_data_in_subtree (racc
, gsi
);
2840 refreshed
= SRA_UDH_NONE
;
2842 load_assign_lhs_subreplacements (lacc
, racc
, lacc
->offset
,
2843 &orig_gsi
, gsi
, &refreshed
);
2844 if (refreshed
!= SRA_UDH_RIGHT
)
2847 unlink_stmt_vdef (*stmt
);
2848 gsi_remove (&orig_gsi
, true);
2849 sra_stats
.deleted
++;
2850 return SRA_AM_REMOVED
;
2857 if (!racc
->grp_to_be_replaced
&& !racc
->grp_unscalarized_data
)
2861 fprintf (dump_file
, "Removing load: ");
2862 print_gimple_stmt (dump_file
, *stmt
, 0, 0);
2865 if (TREE_CODE (lhs
) == SSA_NAME
)
2867 rhs
= get_repl_default_def_ssa_name (racc
);
2868 if (!useless_type_conversion_p (TREE_TYPE (lhs
),
2870 rhs
= fold_build1_loc (loc
, VIEW_CONVERT_EXPR
,
2871 TREE_TYPE (lhs
), rhs
);
2875 if (racc
->first_child
)
2876 generate_subtree_copies (racc
->first_child
, lhs
,
2877 racc
->offset
, 0, 0, gsi
,
2880 gcc_assert (*stmt
== gsi_stmt (*gsi
));
2881 unlink_stmt_vdef (*stmt
);
2882 gsi_remove (gsi
, true);
2883 sra_stats
.deleted
++;
2884 return SRA_AM_REMOVED
;
2887 else if (racc
->first_child
)
2888 generate_subtree_copies (racc
->first_child
, lhs
, racc
->offset
,
2889 0, 0, gsi
, false, true, loc
);
2891 if (access_has_children_p (lacc
))
2892 generate_subtree_copies (lacc
->first_child
, rhs
, lacc
->offset
,
2893 0, 0, gsi
, true, true, loc
);
2897 /* This gimplification must be done after generate_subtree_copies, lest we
2898 insert the subtree copies in the middle of the gimplified sequence. */
2899 if (force_gimple_rhs
)
2900 rhs
= force_gimple_operand_gsi (&orig_gsi
, rhs
, true, NULL_TREE
,
2901 true, GSI_SAME_STMT
);
2902 if (gimple_assign_rhs1 (*stmt
) != rhs
)
2904 modify_this_stmt
= true;
2905 gimple_assign_set_rhs_from_tree (&orig_gsi
, rhs
);
2906 gcc_assert (*stmt
== gsi_stmt (orig_gsi
));
2909 return modify_this_stmt
? SRA_AM_MODIFIED
: SRA_AM_NONE
;
2912 /* Traverse the function body and all modifications as decided in
2913 analyze_all_variable_accesses. Return true iff the CFG has been
2917 sra_modify_function_body (void)
2919 bool cfg_changed
= false;
2924 gimple_stmt_iterator gsi
= gsi_start_bb (bb
);
2925 while (!gsi_end_p (gsi
))
2927 gimple stmt
= gsi_stmt (gsi
);
2928 enum assignment_mod_result assign_result
;
2929 bool modified
= false, deleted
= false;
2933 switch (gimple_code (stmt
))
2936 t
= gimple_return_retval_ptr (stmt
);
2937 if (*t
!= NULL_TREE
)
2938 modified
|= sra_modify_expr (t
, &gsi
, false);
2942 assign_result
= sra_modify_assign (&stmt
, &gsi
);
2943 modified
|= assign_result
== SRA_AM_MODIFIED
;
2944 deleted
= assign_result
== SRA_AM_REMOVED
;
2948 /* Operands must be processed before the lhs. */
2949 for (i
= 0; i
< gimple_call_num_args (stmt
); i
++)
2951 t
= gimple_call_arg_ptr (stmt
, i
);
2952 modified
|= sra_modify_expr (t
, &gsi
, false);
2955 if (gimple_call_lhs (stmt
))
2957 t
= gimple_call_lhs_ptr (stmt
);
2958 modified
|= sra_modify_expr (t
, &gsi
, true);
2963 for (i
= 0; i
< gimple_asm_ninputs (stmt
); i
++)
2965 t
= &TREE_VALUE (gimple_asm_input_op (stmt
, i
));
2966 modified
|= sra_modify_expr (t
, &gsi
, false);
2968 for (i
= 0; i
< gimple_asm_noutputs (stmt
); i
++)
2970 t
= &TREE_VALUE (gimple_asm_output_op (stmt
, i
));
2971 modified
|= sra_modify_expr (t
, &gsi
, true);
2982 if (maybe_clean_eh_stmt (stmt
)
2983 && gimple_purge_dead_eh_edges (gimple_bb (stmt
)))
2994 /* Generate statements initializing scalar replacements of parts of function
2998 initialize_parameter_reductions (void)
3000 gimple_stmt_iterator gsi
;
3001 gimple_seq seq
= NULL
;
3004 for (parm
= DECL_ARGUMENTS (current_function_decl
);
3006 parm
= DECL_CHAIN (parm
))
3008 VEC (access_p
, heap
) *access_vec
;
3009 struct access
*access
;
3011 if (!bitmap_bit_p (candidate_bitmap
, DECL_UID (parm
)))
3013 access_vec
= get_base_access_vector (parm
);
3019 seq
= gimple_seq_alloc ();
3020 gsi
= gsi_start (seq
);
3023 for (access
= VEC_index (access_p
, access_vec
, 0);
3025 access
= access
->next_grp
)
3026 generate_subtree_copies (access
, parm
, 0, 0, 0, &gsi
, true, true,
3027 EXPR_LOCATION (parm
));
3031 gsi_insert_seq_on_edge_immediate (single_succ_edge (ENTRY_BLOCK_PTR
), seq
);
3034 /* The "main" function of intraprocedural SRA passes. Runs the analysis and if
3035 it reveals there are components of some aggregates to be scalarized, it runs
3036 the required transformations. */
3038 perform_intra_sra (void)
3043 if (!find_var_candidates ())
3046 if (!scan_function ())
3049 if (!analyze_all_variable_accesses ())
3052 if (sra_modify_function_body ())
3053 ret
= TODO_update_ssa
| TODO_cleanup_cfg
;
3055 ret
= TODO_update_ssa
;
3056 initialize_parameter_reductions ();
3058 statistics_counter_event (cfun
, "Scalar replacements created",
3059 sra_stats
.replacements
);
3060 statistics_counter_event (cfun
, "Modified expressions", sra_stats
.exprs
);
3061 statistics_counter_event (cfun
, "Subtree copy stmts",
3062 sra_stats
.subtree_copies
);
3063 statistics_counter_event (cfun
, "Subreplacement stmts",
3064 sra_stats
.subreplacements
);
3065 statistics_counter_event (cfun
, "Deleted stmts", sra_stats
.deleted
);
3066 statistics_counter_event (cfun
, "Separate LHS and RHS handling",
3067 sra_stats
.separate_lhs_rhs_handling
);
3070 sra_deinitialize ();
3074 /* Perform early intraprocedural SRA. */
3076 early_intra_sra (void)
3078 sra_mode
= SRA_MODE_EARLY_INTRA
;
3079 return perform_intra_sra ();
3082 /* Perform "late" intraprocedural SRA. */
3084 late_intra_sra (void)
3086 sra_mode
= SRA_MODE_INTRA
;
3087 return perform_intra_sra ();
3092 gate_intra_sra (void)
3094 return flag_tree_sra
!= 0 && dbg_cnt (tree_sra
);
3098 struct gimple_opt_pass pass_sra_early
=
3103 gate_intra_sra
, /* gate */
3104 early_intra_sra
, /* execute */
3107 0, /* static_pass_number */
3108 TV_TREE_SRA
, /* tv_id */
3109 PROP_cfg
| PROP_ssa
, /* properties_required */
3110 0, /* properties_provided */
3111 0, /* properties_destroyed */
3112 0, /* todo_flags_start */
3116 | TODO_verify_ssa
/* todo_flags_finish */
3120 struct gimple_opt_pass pass_sra
=
3125 gate_intra_sra
, /* gate */
3126 late_intra_sra
, /* execute */
3129 0, /* static_pass_number */
3130 TV_TREE_SRA
, /* tv_id */
3131 PROP_cfg
| PROP_ssa
, /* properties_required */
3132 0, /* properties_provided */
3133 0, /* properties_destroyed */
3134 TODO_update_address_taken
, /* todo_flags_start */
3138 | TODO_verify_ssa
/* todo_flags_finish */
3143 /* Return true iff PARM (which must be a parm_decl) is an unused scalar
3147 is_unused_scalar_param (tree parm
)
3150 return (is_gimple_reg (parm
)
3151 && (!(name
= gimple_default_def (cfun
, parm
))
3152 || has_zero_uses (name
)));
3155 /* Scan immediate uses of a default definition SSA name of a parameter PARM and
3156 examine whether there are any direct or otherwise infeasible ones. If so,
3157 return true, otherwise return false. PARM must be a gimple register with a
3158 non-NULL default definition. */
3161 ptr_parm_has_direct_uses (tree parm
)
3163 imm_use_iterator ui
;
3165 tree name
= gimple_default_def (cfun
, parm
);
3168 FOR_EACH_IMM_USE_STMT (stmt
, ui
, name
)
3171 use_operand_p use_p
;
3173 if (is_gimple_debug (stmt
))
3176 /* Valid uses include dereferences on the lhs and the rhs. */
3177 if (gimple_has_lhs (stmt
))
3179 tree lhs
= gimple_get_lhs (stmt
);
3180 while (handled_component_p (lhs
))
3181 lhs
= TREE_OPERAND (lhs
, 0);
3182 if (TREE_CODE (lhs
) == MEM_REF
3183 && TREE_OPERAND (lhs
, 0) == name
3184 && integer_zerop (TREE_OPERAND (lhs
, 1))
3185 && types_compatible_p (TREE_TYPE (lhs
),
3186 TREE_TYPE (TREE_TYPE (name
))))
3189 if (gimple_assign_single_p (stmt
))
3191 tree rhs
= gimple_assign_rhs1 (stmt
);
3192 while (handled_component_p (rhs
))
3193 rhs
= TREE_OPERAND (rhs
, 0);
3194 if (TREE_CODE (rhs
) == MEM_REF
3195 && TREE_OPERAND (rhs
, 0) == name
3196 && integer_zerop (TREE_OPERAND (rhs
, 1))
3197 && types_compatible_p (TREE_TYPE (rhs
),
3198 TREE_TYPE (TREE_TYPE (name
))))
3201 else if (is_gimple_call (stmt
))
3204 for (i
= 0; i
< gimple_call_num_args (stmt
); ++i
)
3206 tree arg
= gimple_call_arg (stmt
, i
);
3207 while (handled_component_p (arg
))
3208 arg
= TREE_OPERAND (arg
, 0);
3209 if (TREE_CODE (arg
) == MEM_REF
3210 && TREE_OPERAND (arg
, 0) == name
3211 && integer_zerop (TREE_OPERAND (arg
, 1))
3212 && types_compatible_p (TREE_TYPE (arg
),
3213 TREE_TYPE (TREE_TYPE (name
))))
3218 /* If the number of valid uses does not match the number of
3219 uses in this stmt there is an unhandled use. */
3220 FOR_EACH_IMM_USE_ON_STMT (use_p
, ui
)
3227 BREAK_FROM_IMM_USE_STMT (ui
);
3233 /* Identify candidates for reduction for IPA-SRA based on their type and mark
3234 them in candidate_bitmap. Note that these do not necessarily include
3235 parameter which are unused and thus can be removed. Return true iff any
3236 such candidate has been found. */
3239 find_param_candidates (void)
3245 for (parm
= DECL_ARGUMENTS (current_function_decl
);
3247 parm
= DECL_CHAIN (parm
))
3249 tree type
= TREE_TYPE (parm
);
3253 if (TREE_THIS_VOLATILE (parm
)
3254 || TREE_ADDRESSABLE (parm
)
3255 || (!is_gimple_reg_type (type
) && is_va_list_type (type
)))
3258 if (is_unused_scalar_param (parm
))
3264 if (POINTER_TYPE_P (type
))
3266 type
= TREE_TYPE (type
);
3268 if (TREE_CODE (type
) == FUNCTION_TYPE
3269 || TYPE_VOLATILE (type
)
3270 || (TREE_CODE (type
) == ARRAY_TYPE
3271 && TYPE_NONALIASED_COMPONENT (type
))
3272 || !is_gimple_reg (parm
)
3273 || is_va_list_type (type
)
3274 || ptr_parm_has_direct_uses (parm
))
3277 else if (!AGGREGATE_TYPE_P (type
))
3280 if (!COMPLETE_TYPE_P (type
)
3281 || !host_integerp (TYPE_SIZE (type
), 1)
3282 || tree_low_cst (TYPE_SIZE (type
), 1) == 0
3283 || (AGGREGATE_TYPE_P (type
)
3284 && type_internals_preclude_sra_p (type
)))
3287 bitmap_set_bit (candidate_bitmap
, DECL_UID (parm
));
3289 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
3291 fprintf (dump_file
, "Candidate (%d): ", DECL_UID (parm
));
3292 print_generic_expr (dump_file
, parm
, 0);
3293 fprintf (dump_file
, "\n");
3297 func_param_count
= count
;
3301 /* Callback of walk_aliased_vdefs, marks the access passed as DATA as
3305 mark_maybe_modified (ao_ref
*ao ATTRIBUTE_UNUSED
, tree vdef ATTRIBUTE_UNUSED
,
3308 struct access
*repr
= (struct access
*) data
;
3310 repr
->grp_maybe_modified
= 1;
3314 /* Analyze what representatives (in linked lists accessible from
3315 REPRESENTATIVES) can be modified by side effects of statements in the
3316 current function. */
3319 analyze_modified_params (VEC (access_p
, heap
) *representatives
)
3323 for (i
= 0; i
< func_param_count
; i
++)
3325 struct access
*repr
;
3327 for (repr
= VEC_index (access_p
, representatives
, i
);
3329 repr
= repr
->next_grp
)
3331 struct access
*access
;
3335 if (no_accesses_p (repr
))
3337 if (!POINTER_TYPE_P (TREE_TYPE (repr
->base
))
3338 || repr
->grp_maybe_modified
)
3341 ao_ref_init (&ar
, repr
->expr
);
3342 visited
= BITMAP_ALLOC (NULL
);
3343 for (access
= repr
; access
; access
= access
->next_sibling
)
3345 /* All accesses are read ones, otherwise grp_maybe_modified would
3346 be trivially set. */
3347 walk_aliased_vdefs (&ar
, gimple_vuse (access
->stmt
),
3348 mark_maybe_modified
, repr
, &visited
);
3349 if (repr
->grp_maybe_modified
)
3352 BITMAP_FREE (visited
);
3357 /* Propagate distances in bb_dereferences in the opposite direction than the
3358 control flow edges, in each step storing the maximum of the current value
3359 and the minimum of all successors. These steps are repeated until the table
3360 stabilizes. Note that BBs which might terminate the functions (according to
3361 final_bbs bitmap) never updated in this way. */
3364 propagate_dereference_distances (void)
3366 VEC (basic_block
, heap
) *queue
;
3369 queue
= VEC_alloc (basic_block
, heap
, last_basic_block_for_function (cfun
));
3370 VEC_quick_push (basic_block
, queue
, ENTRY_BLOCK_PTR
);
3373 VEC_quick_push (basic_block
, queue
, bb
);
3377 while (!VEC_empty (basic_block
, queue
))
3381 bool change
= false;
3384 bb
= VEC_pop (basic_block
, queue
);
3387 if (bitmap_bit_p (final_bbs
, bb
->index
))
3390 for (i
= 0; i
< func_param_count
; i
++)
3392 int idx
= bb
->index
* func_param_count
+ i
;
3394 HOST_WIDE_INT inh
= 0;
3396 FOR_EACH_EDGE (e
, ei
, bb
->succs
)
3398 int succ_idx
= e
->dest
->index
* func_param_count
+ i
;
3400 if (e
->src
== EXIT_BLOCK_PTR
)
3406 inh
= bb_dereferences
[succ_idx
];
3408 else if (bb_dereferences
[succ_idx
] < inh
)
3409 inh
= bb_dereferences
[succ_idx
];
3412 if (!first
&& bb_dereferences
[idx
] < inh
)
3414 bb_dereferences
[idx
] = inh
;
3419 if (change
&& !bitmap_bit_p (final_bbs
, bb
->index
))
3420 FOR_EACH_EDGE (e
, ei
, bb
->preds
)
3425 e
->src
->aux
= e
->src
;
3426 VEC_quick_push (basic_block
, queue
, e
->src
);
3430 VEC_free (basic_block
, heap
, queue
);
3433 /* Dump a dereferences TABLE with heading STR to file F. */
3436 dump_dereferences_table (FILE *f
, const char *str
, HOST_WIDE_INT
*table
)
3440 fprintf (dump_file
, str
);
3441 FOR_BB_BETWEEN (bb
, ENTRY_BLOCK_PTR
, EXIT_BLOCK_PTR
, next_bb
)
3443 fprintf (f
, "%4i %i ", bb
->index
, bitmap_bit_p (final_bbs
, bb
->index
));
3444 if (bb
!= EXIT_BLOCK_PTR
)
3447 for (i
= 0; i
< func_param_count
; i
++)
3449 int idx
= bb
->index
* func_param_count
+ i
;
3450 fprintf (f
, " %4" HOST_WIDE_INT_PRINT
"d", table
[idx
]);
3455 fprintf (dump_file
, "\n");
3458 /* Determine what (parts of) parameters passed by reference that are not
3459 assigned to are not certainly dereferenced in this function and thus the
3460 dereferencing cannot be safely moved to the caller without potentially
3461 introducing a segfault. Mark such REPRESENTATIVES as
3462 grp_not_necessarilly_dereferenced.
3464 The dereferenced maximum "distance," i.e. the offset + size of the accessed
3465 part is calculated rather than simple booleans are calculated for each
3466 pointer parameter to handle cases when only a fraction of the whole
3467 aggregate is allocated (see testsuite/gcc.c-torture/execute/ipa-sra-2.c for
3470 The maximum dereference distances for each pointer parameter and BB are
3471 already stored in bb_dereference. This routine simply propagates these
3472 values upwards by propagate_dereference_distances and then compares the
3473 distances of individual parameters in the ENTRY BB to the equivalent
3474 distances of each representative of a (fraction of a) parameter. */
3477 analyze_caller_dereference_legality (VEC (access_p
, heap
) *representatives
)
3481 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
3482 dump_dereferences_table (dump_file
,
3483 "Dereference table before propagation:\n",
3486 propagate_dereference_distances ();
3488 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
3489 dump_dereferences_table (dump_file
,
3490 "Dereference table after propagation:\n",
3493 for (i
= 0; i
< func_param_count
; i
++)
3495 struct access
*repr
= VEC_index (access_p
, representatives
, i
);
3496 int idx
= ENTRY_BLOCK_PTR
->index
* func_param_count
+ i
;
3498 if (!repr
|| no_accesses_p (repr
))
3503 if ((repr
->offset
+ repr
->size
) > bb_dereferences
[idx
])
3504 repr
->grp_not_necessarilly_dereferenced
= 1;
3505 repr
= repr
->next_grp
;
3511 /* Return the representative access for the parameter declaration PARM if it is
3512 a scalar passed by reference which is not written to and the pointer value
3513 is not used directly. Thus, if it is legal to dereference it in the caller
3514 and we can rule out modifications through aliases, such parameter should be
3515 turned into one passed by value. Return NULL otherwise. */
3517 static struct access
*
3518 unmodified_by_ref_scalar_representative (tree parm
)
3520 int i
, access_count
;
3521 struct access
*repr
;
3522 VEC (access_p
, heap
) *access_vec
;
3524 access_vec
= get_base_access_vector (parm
);
3525 gcc_assert (access_vec
);
3526 repr
= VEC_index (access_p
, access_vec
, 0);
3529 repr
->group_representative
= repr
;
3531 access_count
= VEC_length (access_p
, access_vec
);
3532 for (i
= 1; i
< access_count
; i
++)
3534 struct access
*access
= VEC_index (access_p
, access_vec
, i
);
3537 access
->group_representative
= repr
;
3538 access
->next_sibling
= repr
->next_sibling
;
3539 repr
->next_sibling
= access
;
3543 repr
->grp_scalar_ptr
= 1;
3547 /* Return true iff this access precludes IPA-SRA of the parameter it is
3551 access_precludes_ipa_sra_p (struct access
*access
)
3553 /* Avoid issues such as the second simple testcase in PR 42025. The problem
3554 is incompatible assign in a call statement (and possibly even in asm
3555 statements). This can be relaxed by using a new temporary but only for
3556 non-TREE_ADDRESSABLE types and is probably not worth the complexity. (In
3557 intraprocedural SRA we deal with this by keeping the old aggregate around,
3558 something we cannot do in IPA-SRA.) */
3560 && (is_gimple_call (access
->stmt
)
3561 || gimple_code (access
->stmt
) == GIMPLE_ASM
))
3568 /* Sort collected accesses for parameter PARM, identify representatives for
3569 each accessed region and link them together. Return NULL if there are
3570 different but overlapping accesses, return the special ptr value meaning
3571 there are no accesses for this parameter if that is the case and return the
3572 first representative otherwise. Set *RO_GRP if there is a group of accesses
3573 with only read (i.e. no write) accesses. */
3575 static struct access
*
3576 splice_param_accesses (tree parm
, bool *ro_grp
)
3578 int i
, j
, access_count
, group_count
;
3579 int agg_size
, total_size
= 0;
3580 struct access
*access
, *res
, **prev_acc_ptr
= &res
;
3581 VEC (access_p
, heap
) *access_vec
;
3583 access_vec
= get_base_access_vector (parm
);
3585 return &no_accesses_representant
;
3586 access_count
= VEC_length (access_p
, access_vec
);
3588 VEC_qsort (access_p
, access_vec
, compare_access_positions
);
3593 while (i
< access_count
)
3597 access
= VEC_index (access_p
, access_vec
, i
);
3598 modification
= access
->write
;
3599 if (access_precludes_ipa_sra_p (access
))
3601 a1_alias_type
= reference_alias_ptr_type (access
->expr
);
3603 /* Access is about to become group representative unless we find some
3604 nasty overlap which would preclude us from breaking this parameter
3608 while (j
< access_count
)
3610 struct access
*ac2
= VEC_index (access_p
, access_vec
, j
);
3611 if (ac2
->offset
!= access
->offset
)
3613 /* All or nothing law for parameters. */
3614 if (access
->offset
+ access
->size
> ac2
->offset
)
3619 else if (ac2
->size
!= access
->size
)
3622 if (access_precludes_ipa_sra_p (ac2
)
3623 || (ac2
->type
!= access
->type
3624 && (TREE_ADDRESSABLE (ac2
->type
)
3625 || TREE_ADDRESSABLE (access
->type
)))
3626 || (reference_alias_ptr_type (ac2
->expr
) != a1_alias_type
))
3629 modification
|= ac2
->write
;
3630 ac2
->group_representative
= access
;
3631 ac2
->next_sibling
= access
->next_sibling
;
3632 access
->next_sibling
= ac2
;
3637 access
->grp_maybe_modified
= modification
;
3640 *prev_acc_ptr
= access
;
3641 prev_acc_ptr
= &access
->next_grp
;
3642 total_size
+= access
->size
;
3646 if (POINTER_TYPE_P (TREE_TYPE (parm
)))
3647 agg_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm
))), 1);
3649 agg_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (parm
)), 1);
3650 if (total_size
>= agg_size
)
3653 gcc_assert (group_count
> 0);
3657 /* Decide whether parameters with representative accesses given by REPR should
3658 be reduced into components. */
3661 decide_one_param_reduction (struct access
*repr
)
3663 int total_size
, cur_parm_size
, agg_size
, new_param_count
, parm_size_limit
;
3668 cur_parm_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (parm
)), 1);
3669 gcc_assert (cur_parm_size
> 0);
3671 if (POINTER_TYPE_P (TREE_TYPE (parm
)))
3674 agg_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm
))), 1);
3679 agg_size
= cur_parm_size
;
3685 fprintf (dump_file
, "Evaluating PARAM group sizes for ");
3686 print_generic_expr (dump_file
, parm
, 0);
3687 fprintf (dump_file
, " (UID: %u): \n", DECL_UID (parm
));
3688 for (acc
= repr
; acc
; acc
= acc
->next_grp
)
3689 dump_access (dump_file
, acc
, true);
3693 new_param_count
= 0;
3695 for (; repr
; repr
= repr
->next_grp
)
3697 gcc_assert (parm
== repr
->base
);
3699 /* Taking the address of a non-addressable field is verboten. */
3700 if (by_ref
&& repr
->non_addressable
)
3703 if (!by_ref
|| (!repr
->grp_maybe_modified
3704 && !repr
->grp_not_necessarilly_dereferenced
))
3705 total_size
+= repr
->size
;
3707 total_size
+= cur_parm_size
;
3712 gcc_assert (new_param_count
> 0);
3714 if (optimize_function_for_size_p (cfun
))
3715 parm_size_limit
= cur_parm_size
;
3717 parm_size_limit
= (PARAM_VALUE (PARAM_IPA_SRA_PTR_GROWTH_FACTOR
)
3720 if (total_size
< agg_size
3721 && total_size
<= parm_size_limit
)
3724 fprintf (dump_file
, " ....will be split into %i components\n",
3726 return new_param_count
;
3732 /* The order of the following enums is important, we need to do extra work for
3733 UNUSED_PARAMS, BY_VAL_ACCESSES and UNMODIF_BY_REF_ACCESSES. */
3734 enum ipa_splicing_result
{ NO_GOOD_ACCESS
, UNUSED_PARAMS
, BY_VAL_ACCESSES
,
3735 MODIF_BY_REF_ACCESSES
, UNMODIF_BY_REF_ACCESSES
};
3737 /* Identify representatives of all accesses to all candidate parameters for
3738 IPA-SRA. Return result based on what representatives have been found. */
3740 static enum ipa_splicing_result
3741 splice_all_param_accesses (VEC (access_p
, heap
) **representatives
)
3743 enum ipa_splicing_result result
= NO_GOOD_ACCESS
;
3745 struct access
*repr
;
3747 *representatives
= VEC_alloc (access_p
, heap
, func_param_count
);
3749 for (parm
= DECL_ARGUMENTS (current_function_decl
);
3751 parm
= DECL_CHAIN (parm
))
3753 if (is_unused_scalar_param (parm
))
3755 VEC_quick_push (access_p
, *representatives
,
3756 &no_accesses_representant
);
3757 if (result
== NO_GOOD_ACCESS
)
3758 result
= UNUSED_PARAMS
;
3760 else if (POINTER_TYPE_P (TREE_TYPE (parm
))
3761 && is_gimple_reg_type (TREE_TYPE (TREE_TYPE (parm
)))
3762 && bitmap_bit_p (candidate_bitmap
, DECL_UID (parm
)))
3764 repr
= unmodified_by_ref_scalar_representative (parm
);
3765 VEC_quick_push (access_p
, *representatives
, repr
);
3767 result
= UNMODIF_BY_REF_ACCESSES
;
3769 else if (bitmap_bit_p (candidate_bitmap
, DECL_UID (parm
)))
3771 bool ro_grp
= false;
3772 repr
= splice_param_accesses (parm
, &ro_grp
);
3773 VEC_quick_push (access_p
, *representatives
, repr
);
3775 if (repr
&& !no_accesses_p (repr
))
3777 if (POINTER_TYPE_P (TREE_TYPE (parm
)))
3780 result
= UNMODIF_BY_REF_ACCESSES
;
3781 else if (result
< MODIF_BY_REF_ACCESSES
)
3782 result
= MODIF_BY_REF_ACCESSES
;
3784 else if (result
< BY_VAL_ACCESSES
)
3785 result
= BY_VAL_ACCESSES
;
3787 else if (no_accesses_p (repr
) && (result
== NO_GOOD_ACCESS
))
3788 result
= UNUSED_PARAMS
;
3791 VEC_quick_push (access_p
, *representatives
, NULL
);
3794 if (result
== NO_GOOD_ACCESS
)
3796 VEC_free (access_p
, heap
, *representatives
);
3797 *representatives
= NULL
;
3798 return NO_GOOD_ACCESS
;
3804 /* Return the index of BASE in PARMS. Abort if it is not found. */
3807 get_param_index (tree base
, VEC(tree
, heap
) *parms
)
3811 len
= VEC_length (tree
, parms
);
3812 for (i
= 0; i
< len
; i
++)
3813 if (VEC_index (tree
, parms
, i
) == base
)
3818 /* Convert the decisions made at the representative level into compact
3819 parameter adjustments. REPRESENTATIVES are pointers to first
3820 representatives of each param accesses, ADJUSTMENTS_COUNT is the expected
3821 final number of adjustments. */
3823 static ipa_parm_adjustment_vec
3824 turn_representatives_into_adjustments (VEC (access_p
, heap
) *representatives
,
3825 int adjustments_count
)
3827 VEC (tree
, heap
) *parms
;
3828 ipa_parm_adjustment_vec adjustments
;
3832 gcc_assert (adjustments_count
> 0);
3833 parms
= ipa_get_vector_of_formal_parms (current_function_decl
);
3834 adjustments
= VEC_alloc (ipa_parm_adjustment_t
, heap
, adjustments_count
);
3835 parm
= DECL_ARGUMENTS (current_function_decl
);
3836 for (i
= 0; i
< func_param_count
; i
++, parm
= DECL_CHAIN (parm
))
3838 struct access
*repr
= VEC_index (access_p
, representatives
, i
);
3840 if (!repr
|| no_accesses_p (repr
))
3842 struct ipa_parm_adjustment
*adj
;
3844 adj
= VEC_quick_push (ipa_parm_adjustment_t
, adjustments
, NULL
);
3845 memset (adj
, 0, sizeof (*adj
));
3846 adj
->base_index
= get_param_index (parm
, parms
);
3849 adj
->copy_param
= 1;
3851 adj
->remove_param
= 1;
3855 struct ipa_parm_adjustment
*adj
;
3856 int index
= get_param_index (parm
, parms
);
3858 for (; repr
; repr
= repr
->next_grp
)
3860 adj
= VEC_quick_push (ipa_parm_adjustment_t
, adjustments
, NULL
);
3861 memset (adj
, 0, sizeof (*adj
));
3862 gcc_assert (repr
->base
== parm
);
3863 adj
->base_index
= index
;
3864 adj
->base
= repr
->base
;
3865 adj
->type
= repr
->type
;
3866 adj
->alias_ptr_type
= reference_alias_ptr_type (repr
->expr
);
3867 adj
->offset
= repr
->offset
;
3868 adj
->by_ref
= (POINTER_TYPE_P (TREE_TYPE (repr
->base
))
3869 && (repr
->grp_maybe_modified
3870 || repr
->grp_not_necessarilly_dereferenced
));
3875 VEC_free (tree
, heap
, parms
);
3879 /* Analyze the collected accesses and produce a plan what to do with the
3880 parameters in the form of adjustments, NULL meaning nothing. */
3882 static ipa_parm_adjustment_vec
3883 analyze_all_param_acesses (void)
3885 enum ipa_splicing_result repr_state
;
3886 bool proceed
= false;
3887 int i
, adjustments_count
= 0;
3888 VEC (access_p
, heap
) *representatives
;
3889 ipa_parm_adjustment_vec adjustments
;
3891 repr_state
= splice_all_param_accesses (&representatives
);
3892 if (repr_state
== NO_GOOD_ACCESS
)
3895 /* If there are any parameters passed by reference which are not modified
3896 directly, we need to check whether they can be modified indirectly. */
3897 if (repr_state
== UNMODIF_BY_REF_ACCESSES
)
3899 analyze_caller_dereference_legality (representatives
);
3900 analyze_modified_params (representatives
);
3903 for (i
= 0; i
< func_param_count
; i
++)
3905 struct access
*repr
= VEC_index (access_p
, representatives
, i
);
3907 if (repr
&& !no_accesses_p (repr
))
3909 if (repr
->grp_scalar_ptr
)
3911 adjustments_count
++;
3912 if (repr
->grp_not_necessarilly_dereferenced
3913 || repr
->grp_maybe_modified
)
3914 VEC_replace (access_p
, representatives
, i
, NULL
);
3918 sra_stats
.scalar_by_ref_to_by_val
++;
3923 int new_components
= decide_one_param_reduction (repr
);
3925 if (new_components
== 0)
3927 VEC_replace (access_p
, representatives
, i
, NULL
);
3928 adjustments_count
++;
3932 adjustments_count
+= new_components
;
3933 sra_stats
.aggregate_params_reduced
++;
3934 sra_stats
.param_reductions_created
+= new_components
;
3941 if (no_accesses_p (repr
))
3944 sra_stats
.deleted_unused_parameters
++;
3946 adjustments_count
++;
3950 if (!proceed
&& dump_file
)
3951 fprintf (dump_file
, "NOT proceeding to change params.\n");
3954 adjustments
= turn_representatives_into_adjustments (representatives
,
3959 VEC_free (access_p
, heap
, representatives
);
3963 /* If a parameter replacement identified by ADJ does not yet exist in the form
3964 of declaration, create it and record it, otherwise return the previously
3968 get_replaced_param_substitute (struct ipa_parm_adjustment
*adj
)
3971 if (!adj
->new_ssa_base
)
3973 char *pretty_name
= make_fancy_name (adj
->base
);
3975 repl
= create_tmp_reg (TREE_TYPE (adj
->base
), "ISR");
3976 DECL_NAME (repl
) = get_identifier (pretty_name
);
3977 obstack_free (&name_obstack
, pretty_name
);
3980 add_referenced_var (repl
);
3981 adj
->new_ssa_base
= repl
;
3984 repl
= adj
->new_ssa_base
;
3988 /* Find the first adjustment for a particular parameter BASE in a vector of
3989 ADJUSTMENTS which is not a copy_param. Return NULL if there is no such
3992 static struct ipa_parm_adjustment
*
3993 get_adjustment_for_base (ipa_parm_adjustment_vec adjustments
, tree base
)
3997 len
= VEC_length (ipa_parm_adjustment_t
, adjustments
);
3998 for (i
= 0; i
< len
; i
++)
4000 struct ipa_parm_adjustment
*adj
;
4002 adj
= VEC_index (ipa_parm_adjustment_t
, adjustments
, i
);
4003 if (!adj
->copy_param
&& adj
->base
== base
)
4010 /* If the statement STMT defines an SSA_NAME of a parameter which is to be
4011 removed because its value is not used, replace the SSA_NAME with a one
4012 relating to a created VAR_DECL together all of its uses and return true.
4013 ADJUSTMENTS is a pointer to an adjustments vector. */
4016 replace_removed_params_ssa_names (gimple stmt
,
4017 ipa_parm_adjustment_vec adjustments
)
4019 struct ipa_parm_adjustment
*adj
;
4020 tree lhs
, decl
, repl
, name
;
4022 if (gimple_code (stmt
) == GIMPLE_PHI
)
4023 lhs
= gimple_phi_result (stmt
);
4024 else if (is_gimple_assign (stmt
))
4025 lhs
= gimple_assign_lhs (stmt
);
4026 else if (is_gimple_call (stmt
))
4027 lhs
= gimple_call_lhs (stmt
);
4031 if (TREE_CODE (lhs
) != SSA_NAME
)
4033 decl
= SSA_NAME_VAR (lhs
);
4034 if (TREE_CODE (decl
) != PARM_DECL
)
4037 adj
= get_adjustment_for_base (adjustments
, decl
);
4041 repl
= get_replaced_param_substitute (adj
);
4042 name
= make_ssa_name (repl
, stmt
);
4046 fprintf (dump_file
, "replacing an SSA name of a removed param ");
4047 print_generic_expr (dump_file
, lhs
, 0);
4048 fprintf (dump_file
, " with ");
4049 print_generic_expr (dump_file
, name
, 0);
4050 fprintf (dump_file
, "\n");
4053 if (is_gimple_assign (stmt
))
4054 gimple_assign_set_lhs (stmt
, name
);
4055 else if (is_gimple_call (stmt
))
4056 gimple_call_set_lhs (stmt
, name
);
4058 gimple_phi_set_result (stmt
, name
);
4060 replace_uses_by (lhs
, name
);
4061 release_ssa_name (lhs
);
4065 /* If the expression *EXPR should be replaced by a reduction of a parameter, do
4066 so. ADJUSTMENTS is a pointer to a vector of adjustments. CONVERT
4067 specifies whether the function should care about type incompatibility the
4068 current and new expressions. If it is false, the function will leave
4069 incompatibility issues to the caller. Return true iff the expression
4073 sra_ipa_modify_expr (tree
*expr
, bool convert
,
4074 ipa_parm_adjustment_vec adjustments
)
4077 struct ipa_parm_adjustment
*adj
, *cand
= NULL
;
4078 HOST_WIDE_INT offset
, size
, max_size
;
4081 len
= VEC_length (ipa_parm_adjustment_t
, adjustments
);
4083 if (TREE_CODE (*expr
) == BIT_FIELD_REF
4084 || TREE_CODE (*expr
) == IMAGPART_EXPR
4085 || TREE_CODE (*expr
) == REALPART_EXPR
)
4087 expr
= &TREE_OPERAND (*expr
, 0);
4091 base
= get_ref_base_and_extent (*expr
, &offset
, &size
, &max_size
);
4092 if (!base
|| size
== -1 || max_size
== -1)
4095 if (TREE_CODE (base
) == MEM_REF
)
4097 offset
+= mem_ref_offset (base
).low
* BITS_PER_UNIT
;
4098 base
= TREE_OPERAND (base
, 0);
4101 base
= get_ssa_base_param (base
);
4102 if (!base
|| TREE_CODE (base
) != PARM_DECL
)
4105 for (i
= 0; i
< len
; i
++)
4107 adj
= VEC_index (ipa_parm_adjustment_t
, adjustments
, i
);
4109 if (adj
->base
== base
&&
4110 (adj
->offset
== offset
|| adj
->remove_param
))
4116 if (!cand
|| cand
->copy_param
|| cand
->remove_param
)
4120 src
= build_simple_mem_ref (cand
->reduction
);
4122 src
= cand
->reduction
;
4124 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
4126 fprintf (dump_file
, "About to replace expr ");
4127 print_generic_expr (dump_file
, *expr
, 0);
4128 fprintf (dump_file
, " with ");
4129 print_generic_expr (dump_file
, src
, 0);
4130 fprintf (dump_file
, "\n");
4133 if (convert
&& !useless_type_conversion_p (TREE_TYPE (*expr
), cand
->type
))
4135 tree vce
= build1 (VIEW_CONVERT_EXPR
, TREE_TYPE (*expr
), src
);
4143 /* If the statement pointed to by STMT_PTR contains any expressions that need
4144 to replaced with a different one as noted by ADJUSTMENTS, do so. Handle any
4145 potential type incompatibilities (GSI is used to accommodate conversion
4146 statements and must point to the statement). Return true iff the statement
4150 sra_ipa_modify_assign (gimple
*stmt_ptr
, gimple_stmt_iterator
*gsi
,
4151 ipa_parm_adjustment_vec adjustments
)
4153 gimple stmt
= *stmt_ptr
;
4154 tree
*lhs_p
, *rhs_p
;
4157 if (!gimple_assign_single_p (stmt
))
4160 rhs_p
= gimple_assign_rhs1_ptr (stmt
);
4161 lhs_p
= gimple_assign_lhs_ptr (stmt
);
4163 any
= sra_ipa_modify_expr (rhs_p
, false, adjustments
);
4164 any
|= sra_ipa_modify_expr (lhs_p
, false, adjustments
);
4167 tree new_rhs
= NULL_TREE
;
4169 if (!useless_type_conversion_p (TREE_TYPE (*lhs_p
), TREE_TYPE (*rhs_p
)))
4171 if (TREE_CODE (*rhs_p
) == CONSTRUCTOR
)
4173 /* V_C_Es of constructors can cause trouble (PR 42714). */
4174 if (is_gimple_reg_type (TREE_TYPE (*lhs_p
)))
4175 *rhs_p
= build_zero_cst (TREE_TYPE (*lhs_p
));
4177 *rhs_p
= build_constructor (TREE_TYPE (*lhs_p
), 0);
4180 new_rhs
= fold_build1_loc (gimple_location (stmt
),
4181 VIEW_CONVERT_EXPR
, TREE_TYPE (*lhs_p
),
4184 else if (REFERENCE_CLASS_P (*rhs_p
)
4185 && is_gimple_reg_type (TREE_TYPE (*lhs_p
))
4186 && !is_gimple_reg (*lhs_p
))
4187 /* This can happen when an assignment in between two single field
4188 structures is turned into an assignment in between two pointers to
4189 scalars (PR 42237). */
4194 tree tmp
= force_gimple_operand_gsi (gsi
, new_rhs
, true, NULL_TREE
,
4195 true, GSI_SAME_STMT
);
4197 gimple_assign_set_rhs_from_tree (gsi
, tmp
);
4206 /* Traverse the function body and all modifications as described in
4207 ADJUSTMENTS. Return true iff the CFG has been changed. */
4210 ipa_sra_modify_function_body (ipa_parm_adjustment_vec adjustments
)
4212 bool cfg_changed
= false;
4217 gimple_stmt_iterator gsi
;
4219 for (gsi
= gsi_start_phis (bb
); !gsi_end_p (gsi
); gsi_next (&gsi
))
4220 replace_removed_params_ssa_names (gsi_stmt (gsi
), adjustments
);
4222 gsi
= gsi_start_bb (bb
);
4223 while (!gsi_end_p (gsi
))
4225 gimple stmt
= gsi_stmt (gsi
);
4226 bool modified
= false;
4230 switch (gimple_code (stmt
))
4233 t
= gimple_return_retval_ptr (stmt
);
4234 if (*t
!= NULL_TREE
)
4235 modified
|= sra_ipa_modify_expr (t
, true, adjustments
);
4239 modified
|= sra_ipa_modify_assign (&stmt
, &gsi
, adjustments
);
4240 modified
|= replace_removed_params_ssa_names (stmt
, adjustments
);
4244 /* Operands must be processed before the lhs. */
4245 for (i
= 0; i
< gimple_call_num_args (stmt
); i
++)
4247 t
= gimple_call_arg_ptr (stmt
, i
);
4248 modified
|= sra_ipa_modify_expr (t
, true, adjustments
);
4251 if (gimple_call_lhs (stmt
))
4253 t
= gimple_call_lhs_ptr (stmt
);
4254 modified
|= sra_ipa_modify_expr (t
, false, adjustments
);
4255 modified
|= replace_removed_params_ssa_names (stmt
,
4261 for (i
= 0; i
< gimple_asm_ninputs (stmt
); i
++)
4263 t
= &TREE_VALUE (gimple_asm_input_op (stmt
, i
));
4264 modified
|= sra_ipa_modify_expr (t
, true, adjustments
);
4266 for (i
= 0; i
< gimple_asm_noutputs (stmt
); i
++)
4268 t
= &TREE_VALUE (gimple_asm_output_op (stmt
, i
));
4269 modified
|= sra_ipa_modify_expr (t
, false, adjustments
);
4280 if (maybe_clean_eh_stmt (stmt
)
4281 && gimple_purge_dead_eh_edges (gimple_bb (stmt
)))
4291 /* Call gimple_debug_bind_reset_value on all debug statements describing
4292 gimple register parameters that are being removed or replaced. */
4295 sra_ipa_reset_debug_stmts (ipa_parm_adjustment_vec adjustments
)
4299 len
= VEC_length (ipa_parm_adjustment_t
, adjustments
);
4300 for (i
= 0; i
< len
; i
++)
4302 struct ipa_parm_adjustment
*adj
;
4303 imm_use_iterator ui
;
4307 adj
= VEC_index (ipa_parm_adjustment_t
, adjustments
, i
);
4308 if (adj
->copy_param
|| !is_gimple_reg (adj
->base
))
4310 name
= gimple_default_def (cfun
, adj
->base
);
4313 FOR_EACH_IMM_USE_STMT (stmt
, ui
, name
)
4315 /* All other users must have been removed by
4316 ipa_sra_modify_function_body. */
4317 gcc_assert (is_gimple_debug (stmt
));
4318 gimple_debug_bind_reset_value (stmt
);
4324 /* Return true iff all callers have at least as many actual arguments as there
4325 are formal parameters in the current function. */
4328 all_callers_have_enough_arguments_p (struct cgraph_node
*node
)
4330 struct cgraph_edge
*cs
;
4331 for (cs
= node
->callers
; cs
; cs
= cs
->next_caller
)
4332 if (!callsite_has_enough_arguments_p (cs
->call_stmt
))
4339 /* Convert all callers of NODE to pass parameters as given in ADJUSTMENTS. */
4342 convert_callers (struct cgraph_node
*node
, tree old_decl
,
4343 ipa_parm_adjustment_vec adjustments
)
4345 tree old_cur_fndecl
= current_function_decl
;
4346 struct cgraph_edge
*cs
;
4347 basic_block this_block
;
4348 bitmap recomputed_callers
= BITMAP_ALLOC (NULL
);
4350 for (cs
= node
->callers
; cs
; cs
= cs
->next_caller
)
4352 current_function_decl
= cs
->caller
->decl
;
4353 push_cfun (DECL_STRUCT_FUNCTION (cs
->caller
->decl
));
4356 fprintf (dump_file
, "Adjusting call (%i -> %i) %s -> %s\n",
4357 cs
->caller
->uid
, cs
->callee
->uid
,
4358 cgraph_node_name (cs
->caller
),
4359 cgraph_node_name (cs
->callee
));
4361 ipa_modify_call_arguments (cs
, cs
->call_stmt
, adjustments
);
4366 for (cs
= node
->callers
; cs
; cs
= cs
->next_caller
)
4367 if (bitmap_set_bit (recomputed_callers
, cs
->caller
->uid
)
4368 && gimple_in_ssa_p (DECL_STRUCT_FUNCTION (cs
->caller
->decl
)))
4369 compute_inline_parameters (cs
->caller
, true);
4370 BITMAP_FREE (recomputed_callers
);
4372 current_function_decl
= old_cur_fndecl
;
4374 if (!encountered_recursive_call
)
4377 FOR_EACH_BB (this_block
)
4379 gimple_stmt_iterator gsi
;
4381 for (gsi
= gsi_start_bb (this_block
); !gsi_end_p (gsi
); gsi_next (&gsi
))
4383 gimple stmt
= gsi_stmt (gsi
);
4385 if (gimple_code (stmt
) != GIMPLE_CALL
)
4387 call_fndecl
= gimple_call_fndecl (stmt
);
4388 if (call_fndecl
== old_decl
)
4391 fprintf (dump_file
, "Adjusting recursive call");
4392 gimple_call_set_fndecl (stmt
, node
->decl
);
4393 ipa_modify_call_arguments (NULL
, stmt
, adjustments
);
4401 /* Perform all the modification required in IPA-SRA for NODE to have parameters
4402 as given in ADJUSTMENTS. Return true iff the CFG has been changed. */
4405 modify_function (struct cgraph_node
*node
, ipa_parm_adjustment_vec adjustments
)
4407 struct cgraph_node
*new_node
;
4408 struct cgraph_edge
*cs
;
4410 VEC (cgraph_edge_p
, heap
) * redirect_callers
;
4414 for (cs
= node
->callers
; cs
!= NULL
; cs
= cs
->next_caller
)
4416 redirect_callers
= VEC_alloc (cgraph_edge_p
, heap
, node_callers
);
4417 for (cs
= node
->callers
; cs
!= NULL
; cs
= cs
->next_caller
)
4418 VEC_quick_push (cgraph_edge_p
, redirect_callers
, cs
);
4420 rebuild_cgraph_edges ();
4422 current_function_decl
= NULL_TREE
;
4424 new_node
= cgraph_function_versioning (node
, redirect_callers
, NULL
, NULL
,
4425 NULL
, NULL
, "isra");
4426 current_function_decl
= new_node
->decl
;
4427 push_cfun (DECL_STRUCT_FUNCTION (new_node
->decl
));
4429 ipa_modify_formal_parameters (current_function_decl
, adjustments
, "ISRA");
4430 cfg_changed
= ipa_sra_modify_function_body (adjustments
);
4431 sra_ipa_reset_debug_stmts (adjustments
);
4432 convert_callers (new_node
, node
->decl
, adjustments
);
4433 cgraph_make_node_local (new_node
);
4437 /* Return false the function is apparently unsuitable for IPA-SRA based on it's
4438 attributes, return true otherwise. NODE is the cgraph node of the current
4442 ipa_sra_preliminary_function_checks (struct cgraph_node
*node
)
4444 if (!cgraph_node_can_be_local_p (node
))
4447 fprintf (dump_file
, "Function not local to this compilation unit.\n");
4451 if (!node
->local
.can_change_signature
)
4454 fprintf (dump_file
, "Function can not change signature.\n");
4458 if (!tree_versionable_function_p (node
->decl
))
4461 fprintf (dump_file
, "Function is not versionable.\n");
4465 if (DECL_VIRTUAL_P (current_function_decl
))
4468 fprintf (dump_file
, "Function is a virtual method.\n");
4472 if ((DECL_COMDAT (node
->decl
) || DECL_EXTERNAL (node
->decl
))
4473 && inline_summary(node
)->size
>= MAX_INLINE_INSNS_AUTO
)
4476 fprintf (dump_file
, "Function too big to be made truly local.\n");
4484 "Function has no callers in this compilation unit.\n");
4491 fprintf (dump_file
, "Function uses stdarg. \n");
4495 if (TYPE_ATTRIBUTES (TREE_TYPE (node
->decl
)))
4501 /* Perform early interprocedural SRA. */
4504 ipa_early_sra (void)
4506 struct cgraph_node
*node
= cgraph_get_node (current_function_decl
);
4507 ipa_parm_adjustment_vec adjustments
;
4510 if (!ipa_sra_preliminary_function_checks (node
))
4514 sra_mode
= SRA_MODE_EARLY_IPA
;
4516 if (!find_param_candidates ())
4519 fprintf (dump_file
, "Function has no IPA-SRA candidates.\n");
4523 if (!all_callers_have_enough_arguments_p (node
))
4526 fprintf (dump_file
, "There are callers with insufficient number of "
4531 bb_dereferences
= XCNEWVEC (HOST_WIDE_INT
,
4533 * last_basic_block_for_function (cfun
));
4534 final_bbs
= BITMAP_ALLOC (NULL
);
4537 if (encountered_apply_args
)
4540 fprintf (dump_file
, "Function calls __builtin_apply_args().\n");
4544 if (encountered_unchangable_recursive_call
)
4547 fprintf (dump_file
, "Function calls itself with insufficient "
4548 "number of arguments.\n");
4552 adjustments
= analyze_all_param_acesses ();
4556 ipa_dump_param_adjustments (dump_file
, adjustments
, current_function_decl
);
4558 if (modify_function (node
, adjustments
))
4559 ret
= TODO_update_ssa
| TODO_cleanup_cfg
;
4561 ret
= TODO_update_ssa
;
4562 VEC_free (ipa_parm_adjustment_t
, heap
, adjustments
);
4564 statistics_counter_event (cfun
, "Unused parameters deleted",
4565 sra_stats
.deleted_unused_parameters
);
4566 statistics_counter_event (cfun
, "Scalar parameters converted to by-value",
4567 sra_stats
.scalar_by_ref_to_by_val
);
4568 statistics_counter_event (cfun
, "Aggregate parameters broken up",
4569 sra_stats
.aggregate_params_reduced
);
4570 statistics_counter_event (cfun
, "Aggregate parameter components created",
4571 sra_stats
.param_reductions_created
);
4574 BITMAP_FREE (final_bbs
);
4575 free (bb_dereferences
);
4577 sra_deinitialize ();
4581 /* Return if early ipa sra shall be performed. */
4583 ipa_early_sra_gate (void)
4585 return flag_ipa_sra
&& dbg_cnt (eipa_sra
);
4588 struct gimple_opt_pass pass_early_ipa_sra
=
4592 "eipa_sra", /* name */
4593 ipa_early_sra_gate
, /* gate */
4594 ipa_early_sra
, /* execute */
4597 0, /* static_pass_number */
4598 TV_IPA_SRA
, /* tv_id */
4599 0, /* properties_required */
4600 0, /* properties_provided */
4601 0, /* properties_destroyed */
4602 0, /* todo_flags_start */
4603 TODO_dump_func
| TODO_dump_cgraph
/* todo_flags_finish */