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"
95 /* Enumeration of all aggregate reductions we can do. */
96 enum sra_mode
{ SRA_MODE_EARLY_IPA
, /* early call regularization */
97 SRA_MODE_EARLY_INTRA
, /* early intraprocedural SRA */
98 SRA_MODE_INTRA
}; /* late intraprocedural SRA */
100 /* Global variable describing which aggregate reduction we are performing at
102 static enum sra_mode sra_mode
;
106 /* ACCESS represents each access to an aggregate variable (as a whole or a
107 part). It can also represent a group of accesses that refer to exactly the
108 same fragment of an aggregate (i.e. those that have exactly the same offset
109 and size). Such representatives for a single aggregate, once determined,
110 are linked in a linked list and have the group fields set.
112 Moreover, when doing intraprocedural SRA, a tree is built from those
113 representatives (by the means of first_child and next_sibling pointers), in
114 which all items in a subtree are "within" the root, i.e. their offset is
115 greater or equal to offset of the root and offset+size is smaller or equal
116 to offset+size of the root. Children of an access are sorted by offset.
118 Note that accesses to parts of vector and complex number types always
119 represented by an access to the whole complex number or a vector. It is a
120 duty of the modifying functions to replace them appropriately. */
124 /* Values returned by `get_ref_base_and_extent' for each component reference
125 If EXPR isn't a component reference just set `BASE = EXPR', `OFFSET = 0',
126 `SIZE = TREE_SIZE (TREE_TYPE (expr))'. */
127 HOST_WIDE_INT offset
;
131 /* Expression. It is context dependent so do not use it to create new
132 expressions to access the original aggregate. See PR 42154 for a
138 /* The statement this access belongs to. */
141 /* Next group representative for this aggregate. */
142 struct access
*next_grp
;
144 /* Pointer to the group representative. Pointer to itself if the struct is
145 the representative. */
146 struct access
*group_representative
;
148 /* If this access has any children (in terms of the definition above), this
149 points to the first one. */
150 struct access
*first_child
;
152 /* In intraprocedural SRA, pointer to the next sibling in the access tree as
153 described above. In IPA-SRA this is a pointer to the next access
154 belonging to the same group (having the same representative). */
155 struct access
*next_sibling
;
157 /* Pointers to the first and last element in the linked list of assign
159 struct assign_link
*first_link
, *last_link
;
161 /* Pointer to the next access in the work queue. */
162 struct access
*next_queued
;
164 /* Replacement variable for this access "region." Never to be accessed
165 directly, always only by the means of get_access_replacement() and only
166 when grp_to_be_replaced flag is set. */
167 tree replacement_decl
;
169 /* Is this particular access write access? */
172 /* Is this access an artificial one created to scalarize some record
174 unsigned total_scalarization
: 1;
176 /* Is this access an access to a non-addressable field? */
177 unsigned non_addressable
: 1;
179 /* Is this access currently in the work queue? */
180 unsigned grp_queued
: 1;
182 /* Does this group contain a write access? This flag is propagated down the
184 unsigned grp_write
: 1;
186 /* Does this group contain a read access? This flag is propagated down the
188 unsigned grp_read
: 1;
190 /* Does this group contain a read access that comes from an assignment
191 statement? This flag is propagated down the access tree. */
192 unsigned grp_assignment_read
: 1;
194 /* Does this group contain a write access that comes from an assignment
195 statement? This flag is propagated down the access tree. */
196 unsigned grp_assignment_write
: 1;
198 /* Does this group contain a read access through a scalar type? This flag is
199 not propagated in the access tree in any direction. */
200 unsigned grp_scalar_read
: 1;
202 /* Does this group contain a write access through a scalar type? This flag
203 is not propagated in the access tree in any direction. */
204 unsigned grp_scalar_write
: 1;
206 /* Other passes of the analysis use this bit to make function
207 analyze_access_subtree create scalar replacements for this group if
209 unsigned grp_hint
: 1;
211 /* Is the subtree rooted in this access fully covered by scalar
213 unsigned grp_covered
: 1;
215 /* If set to true, this access and all below it in an access tree must not be
217 unsigned grp_unscalarizable_region
: 1;
219 /* Whether data have been written to parts of the aggregate covered by this
220 access which is not to be scalarized. This flag is propagated up in the
222 unsigned grp_unscalarized_data
: 1;
224 /* Does this access and/or group contain a write access through a
226 unsigned grp_partial_lhs
: 1;
228 /* Set when a scalar replacement should be created for this variable. We do
229 the decision and creation at different places because create_tmp_var
230 cannot be called from within FOR_EACH_REFERENCED_VAR. */
231 unsigned grp_to_be_replaced
: 1;
233 /* Should TREE_NO_WARNING of a replacement be set? */
234 unsigned grp_no_warning
: 1;
236 /* Is it possible that the group refers to data which might be (directly or
237 otherwise) modified? */
238 unsigned grp_maybe_modified
: 1;
240 /* Set when this is a representative of a pointer to scalar (i.e. by
241 reference) parameter which we consider for turning into a plain scalar
242 (i.e. a by value parameter). */
243 unsigned grp_scalar_ptr
: 1;
245 /* Set when we discover that this pointer is not safe to dereference in the
247 unsigned grp_not_necessarilly_dereferenced
: 1;
250 typedef struct access
*access_p
;
252 DEF_VEC_P (access_p
);
253 DEF_VEC_ALLOC_P (access_p
, heap
);
255 /* Alloc pool for allocating access structures. */
256 static alloc_pool access_pool
;
258 /* A structure linking lhs and rhs accesses from an aggregate assignment. They
259 are used to propagate subaccesses from rhs to lhs as long as they don't
260 conflict with what is already there. */
263 struct access
*lacc
, *racc
;
264 struct assign_link
*next
;
267 /* Alloc pool for allocating assign link structures. */
268 static alloc_pool link_pool
;
270 /* Base (tree) -> Vector (VEC(access_p,heap) *) map. */
271 static struct pointer_map_t
*base_access_vec
;
273 /* Bitmap of candidates. */
274 static bitmap candidate_bitmap
;
276 /* Bitmap of candidates which we should try to entirely scalarize away and
277 those which cannot be (because they are and need be used as a whole). */
278 static bitmap should_scalarize_away_bitmap
, cannot_scalarize_away_bitmap
;
280 /* Obstack for creation of fancy names. */
281 static struct obstack name_obstack
;
283 /* Head of a linked list of accesses that need to have its subaccesses
284 propagated to their assignment counterparts. */
285 static struct access
*work_queue_head
;
287 /* Number of parameters of the analyzed function when doing early ipa SRA. */
288 static int func_param_count
;
290 /* scan_function sets the following to true if it encounters a call to
291 __builtin_apply_args. */
292 static bool encountered_apply_args
;
294 /* Set by scan_function when it finds a recursive call. */
295 static bool encountered_recursive_call
;
297 /* Set by scan_function when it finds a recursive call with less actual
298 arguments than formal parameters.. */
299 static bool encountered_unchangable_recursive_call
;
301 /* This is a table in which for each basic block and parameter there is a
302 distance (offset + size) in that parameter which is dereferenced and
303 accessed in that BB. */
304 static HOST_WIDE_INT
*bb_dereferences
;
305 /* Bitmap of BBs that can cause the function to "stop" progressing by
306 returning, throwing externally, looping infinitely or calling a function
307 which might abort etc.. */
308 static bitmap final_bbs
;
310 /* Representative of no accesses at all. */
311 static struct access no_accesses_representant
;
313 /* Predicate to test the special value. */
316 no_accesses_p (struct access
*access
)
318 return access
== &no_accesses_representant
;
321 /* Dump contents of ACCESS to file F in a human friendly way. If GRP is true,
322 representative fields are dumped, otherwise those which only describe the
323 individual access are. */
327 /* Number of processed aggregates is readily available in
328 analyze_all_variable_accesses and so is not stored here. */
330 /* Number of created scalar replacements. */
333 /* Number of times sra_modify_expr or sra_modify_assign themselves changed an
337 /* Number of statements created by generate_subtree_copies. */
340 /* Number of statements created by load_assign_lhs_subreplacements. */
343 /* Number of times sra_modify_assign has deleted a statement. */
346 /* Number of times sra_modify_assign has to deal with subaccesses of LHS and
347 RHS reparately due to type conversions or nonexistent matching
349 int separate_lhs_rhs_handling
;
351 /* Number of parameters that were removed because they were unused. */
352 int deleted_unused_parameters
;
354 /* Number of scalars passed as parameters by reference that have been
355 converted to be passed by value. */
356 int scalar_by_ref_to_by_val
;
358 /* Number of aggregate parameters that were replaced by one or more of their
360 int aggregate_params_reduced
;
362 /* Numbber of components created when splitting aggregate parameters. */
363 int param_reductions_created
;
367 dump_access (FILE *f
, struct access
*access
, bool grp
)
369 fprintf (f
, "access { ");
370 fprintf (f
, "base = (%d)'", DECL_UID (access
->base
));
371 print_generic_expr (f
, access
->base
, 0);
372 fprintf (f
, "', offset = " HOST_WIDE_INT_PRINT_DEC
, access
->offset
);
373 fprintf (f
, ", size = " HOST_WIDE_INT_PRINT_DEC
, access
->size
);
374 fprintf (f
, ", expr = ");
375 print_generic_expr (f
, access
->expr
, 0);
376 fprintf (f
, ", type = ");
377 print_generic_expr (f
, access
->type
, 0);
379 fprintf (f
, ", total_scalarization = %d, grp_read = %d, grp_write = %d, "
380 "grp_assignment_read = %d, grp_assignment_write = %d, "
381 "grp_scalar_read = %d, grp_scalar_write = %d, "
382 "grp_hint = %d, grp_covered = %d, "
383 "grp_unscalarizable_region = %d, grp_unscalarized_data = %d, "
384 "grp_partial_lhs = %d, grp_to_be_replaced = %d, "
385 "grp_maybe_modified = %d, "
386 "grp_not_necessarilly_dereferenced = %d\n",
387 access
->total_scalarization
, access
->grp_read
, access
->grp_write
,
388 access
->grp_assignment_read
, access
->grp_assignment_write
,
389 access
->grp_scalar_read
, access
->grp_scalar_write
,
390 access
->grp_hint
, access
->grp_covered
,
391 access
->grp_unscalarizable_region
, access
->grp_unscalarized_data
,
392 access
->grp_partial_lhs
, access
->grp_to_be_replaced
,
393 access
->grp_maybe_modified
,
394 access
->grp_not_necessarilly_dereferenced
);
396 fprintf (f
, ", write = %d, total_scalarization = %d, "
397 "grp_partial_lhs = %d\n",
398 access
->write
, access
->total_scalarization
,
399 access
->grp_partial_lhs
);
402 /* Dump a subtree rooted in ACCESS to file F, indent by LEVEL. */
405 dump_access_tree_1 (FILE *f
, struct access
*access
, int level
)
411 for (i
= 0; i
< level
; i
++)
412 fputs ("* ", dump_file
);
414 dump_access (f
, access
, true);
416 if (access
->first_child
)
417 dump_access_tree_1 (f
, access
->first_child
, level
+ 1);
419 access
= access
->next_sibling
;
424 /* Dump all access trees for a variable, given the pointer to the first root in
428 dump_access_tree (FILE *f
, struct access
*access
)
430 for (; access
; access
= access
->next_grp
)
431 dump_access_tree_1 (f
, access
, 0);
434 /* Return true iff ACC is non-NULL and has subaccesses. */
437 access_has_children_p (struct access
*acc
)
439 return acc
&& acc
->first_child
;
442 /* Return a vector of pointers to accesses for the variable given in BASE or
443 NULL if there is none. */
445 static VEC (access_p
, heap
) *
446 get_base_access_vector (tree base
)
450 slot
= pointer_map_contains (base_access_vec
, base
);
454 return *(VEC (access_p
, heap
) **) slot
;
457 /* Find an access with required OFFSET and SIZE in a subtree of accesses rooted
458 in ACCESS. Return NULL if it cannot be found. */
460 static struct access
*
461 find_access_in_subtree (struct access
*access
, HOST_WIDE_INT offset
,
464 while (access
&& (access
->offset
!= offset
|| access
->size
!= size
))
466 struct access
*child
= access
->first_child
;
468 while (child
&& (child
->offset
+ child
->size
<= offset
))
469 child
= child
->next_sibling
;
476 /* Return the first group representative for DECL or NULL if none exists. */
478 static struct access
*
479 get_first_repr_for_decl (tree base
)
481 VEC (access_p
, heap
) *access_vec
;
483 access_vec
= get_base_access_vector (base
);
487 return VEC_index (access_p
, access_vec
, 0);
490 /* Find an access representative for the variable BASE and given OFFSET and
491 SIZE. Requires that access trees have already been built. Return NULL if
492 it cannot be found. */
494 static struct access
*
495 get_var_base_offset_size_access (tree base
, HOST_WIDE_INT offset
,
498 struct access
*access
;
500 access
= get_first_repr_for_decl (base
);
501 while (access
&& (access
->offset
+ access
->size
<= offset
))
502 access
= access
->next_grp
;
506 return find_access_in_subtree (access
, offset
, size
);
509 /* Add LINK to the linked list of assign links of RACC. */
511 add_link_to_rhs (struct access
*racc
, struct assign_link
*link
)
513 gcc_assert (link
->racc
== racc
);
515 if (!racc
->first_link
)
517 gcc_assert (!racc
->last_link
);
518 racc
->first_link
= link
;
521 racc
->last_link
->next
= link
;
523 racc
->last_link
= link
;
527 /* Move all link structures in their linked list in OLD_RACC to the linked list
530 relink_to_new_repr (struct access
*new_racc
, struct access
*old_racc
)
532 if (!old_racc
->first_link
)
534 gcc_assert (!old_racc
->last_link
);
538 if (new_racc
->first_link
)
540 gcc_assert (!new_racc
->last_link
->next
);
541 gcc_assert (!old_racc
->last_link
|| !old_racc
->last_link
->next
);
543 new_racc
->last_link
->next
= old_racc
->first_link
;
544 new_racc
->last_link
= old_racc
->last_link
;
548 gcc_assert (!new_racc
->last_link
);
550 new_racc
->first_link
= old_racc
->first_link
;
551 new_racc
->last_link
= old_racc
->last_link
;
553 old_racc
->first_link
= old_racc
->last_link
= NULL
;
556 /* Add ACCESS to the work queue (which is actually a stack). */
559 add_access_to_work_queue (struct access
*access
)
561 if (!access
->grp_queued
)
563 gcc_assert (!access
->next_queued
);
564 access
->next_queued
= work_queue_head
;
565 access
->grp_queued
= 1;
566 work_queue_head
= access
;
570 /* Pop an access from the work queue, and return it, assuming there is one. */
572 static struct access
*
573 pop_access_from_work_queue (void)
575 struct access
*access
= work_queue_head
;
577 work_queue_head
= access
->next_queued
;
578 access
->next_queued
= NULL
;
579 access
->grp_queued
= 0;
584 /* Allocate necessary structures. */
587 sra_initialize (void)
589 candidate_bitmap
= BITMAP_ALLOC (NULL
);
590 should_scalarize_away_bitmap
= BITMAP_ALLOC (NULL
);
591 cannot_scalarize_away_bitmap
= BITMAP_ALLOC (NULL
);
592 gcc_obstack_init (&name_obstack
);
593 access_pool
= create_alloc_pool ("SRA accesses", sizeof (struct access
), 16);
594 link_pool
= create_alloc_pool ("SRA links", sizeof (struct assign_link
), 16);
595 base_access_vec
= pointer_map_create ();
596 memset (&sra_stats
, 0, sizeof (sra_stats
));
597 encountered_apply_args
= false;
598 encountered_recursive_call
= false;
599 encountered_unchangable_recursive_call
= false;
602 /* Hook fed to pointer_map_traverse, deallocate stored vectors. */
605 delete_base_accesses (const void *key ATTRIBUTE_UNUSED
, void **value
,
606 void *data ATTRIBUTE_UNUSED
)
608 VEC (access_p
, heap
) *access_vec
;
609 access_vec
= (VEC (access_p
, heap
) *) *value
;
610 VEC_free (access_p
, heap
, access_vec
);
615 /* Deallocate all general structures. */
618 sra_deinitialize (void)
620 BITMAP_FREE (candidate_bitmap
);
621 BITMAP_FREE (should_scalarize_away_bitmap
);
622 BITMAP_FREE (cannot_scalarize_away_bitmap
);
623 free_alloc_pool (access_pool
);
624 free_alloc_pool (link_pool
);
625 obstack_free (&name_obstack
, NULL
);
627 pointer_map_traverse (base_access_vec
, delete_base_accesses
, NULL
);
628 pointer_map_destroy (base_access_vec
);
631 /* Remove DECL from candidates for SRA and write REASON to the dump file if
634 disqualify_candidate (tree decl
, const char *reason
)
636 bitmap_clear_bit (candidate_bitmap
, DECL_UID (decl
));
638 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
640 fprintf (dump_file
, "! Disqualifying ");
641 print_generic_expr (dump_file
, decl
, 0);
642 fprintf (dump_file
, " - %s\n", reason
);
646 /* Return true iff the type contains a field or an element which does not allow
650 type_internals_preclude_sra_p (tree type
)
655 switch (TREE_CODE (type
))
659 case QUAL_UNION_TYPE
:
660 for (fld
= TYPE_FIELDS (type
); fld
; fld
= DECL_CHAIN (fld
))
661 if (TREE_CODE (fld
) == FIELD_DECL
)
663 tree ft
= TREE_TYPE (fld
);
665 if (TREE_THIS_VOLATILE (fld
)
666 || !DECL_FIELD_OFFSET (fld
) || !DECL_SIZE (fld
)
667 || !host_integerp (DECL_FIELD_OFFSET (fld
), 1)
668 || !host_integerp (DECL_SIZE (fld
), 1)
669 || (AGGREGATE_TYPE_P (ft
)
670 && int_bit_position (fld
) % BITS_PER_UNIT
!= 0))
673 if (AGGREGATE_TYPE_P (ft
)
674 && type_internals_preclude_sra_p (ft
))
681 et
= TREE_TYPE (type
);
683 if (AGGREGATE_TYPE_P (et
))
684 return type_internals_preclude_sra_p (et
);
693 /* If T is an SSA_NAME, return NULL if it is not a default def or return its
694 base variable if it is. Return T if it is not an SSA_NAME. */
697 get_ssa_base_param (tree t
)
699 if (TREE_CODE (t
) == SSA_NAME
)
701 if (SSA_NAME_IS_DEFAULT_DEF (t
))
702 return SSA_NAME_VAR (t
);
709 /* Mark a dereference of BASE of distance DIST in a basic block tht STMT
710 belongs to, unless the BB has already been marked as a potentially
714 mark_parm_dereference (tree base
, HOST_WIDE_INT dist
, gimple stmt
)
716 basic_block bb
= gimple_bb (stmt
);
717 int idx
, parm_index
= 0;
720 if (bitmap_bit_p (final_bbs
, bb
->index
))
723 for (parm
= DECL_ARGUMENTS (current_function_decl
);
724 parm
&& parm
!= base
;
725 parm
= DECL_CHAIN (parm
))
728 gcc_assert (parm_index
< func_param_count
);
730 idx
= bb
->index
* func_param_count
+ parm_index
;
731 if (bb_dereferences
[idx
] < dist
)
732 bb_dereferences
[idx
] = dist
;
735 /* Allocate an access structure for BASE, OFFSET and SIZE, clear it, fill in
736 the three fields. Also add it to the vector of accesses corresponding to
737 the base. Finally, return the new access. */
739 static struct access
*
740 create_access_1 (tree base
, HOST_WIDE_INT offset
, HOST_WIDE_INT size
)
742 VEC (access_p
, heap
) *vec
;
743 struct access
*access
;
746 access
= (struct access
*) pool_alloc (access_pool
);
747 memset (access
, 0, sizeof (struct access
));
749 access
->offset
= offset
;
752 slot
= pointer_map_contains (base_access_vec
, base
);
754 vec
= (VEC (access_p
, heap
) *) *slot
;
756 vec
= VEC_alloc (access_p
, heap
, 32);
758 VEC_safe_push (access_p
, heap
, vec
, access
);
760 *((struct VEC (access_p
,heap
) **)
761 pointer_map_insert (base_access_vec
, base
)) = vec
;
766 /* Create and insert access for EXPR. Return created access, or NULL if it is
769 static struct access
*
770 create_access (tree expr
, gimple stmt
, bool write
)
772 struct access
*access
;
773 HOST_WIDE_INT offset
, size
, max_size
;
775 bool ptr
, unscalarizable_region
= false;
777 base
= get_ref_base_and_extent (expr
, &offset
, &size
, &max_size
);
779 if (sra_mode
== SRA_MODE_EARLY_IPA
780 && TREE_CODE (base
) == MEM_REF
)
782 base
= get_ssa_base_param (TREE_OPERAND (base
, 0));
790 if (!DECL_P (base
) || !bitmap_bit_p (candidate_bitmap
, DECL_UID (base
)))
793 if (sra_mode
== SRA_MODE_EARLY_IPA
)
795 if (size
< 0 || size
!= max_size
)
797 disqualify_candidate (base
, "Encountered a variable sized access.");
800 if (TREE_CODE (expr
) == COMPONENT_REF
801 && DECL_BIT_FIELD (TREE_OPERAND (expr
, 1)))
803 disqualify_candidate (base
, "Encountered a bit-field access.");
806 gcc_checking_assert ((offset
% BITS_PER_UNIT
) == 0);
809 mark_parm_dereference (base
, offset
+ size
, stmt
);
813 if (size
!= max_size
)
816 unscalarizable_region
= true;
820 disqualify_candidate (base
, "Encountered an unconstrained access.");
825 access
= create_access_1 (base
, offset
, size
);
827 access
->type
= TREE_TYPE (expr
);
828 access
->write
= write
;
829 access
->grp_unscalarizable_region
= unscalarizable_region
;
832 if (TREE_CODE (expr
) == COMPONENT_REF
833 && DECL_NONADDRESSABLE_P (TREE_OPERAND (expr
, 1)))
834 access
->non_addressable
= 1;
840 /* Return true iff TYPE is a RECORD_TYPE with fields that are either of gimple
841 register types or (recursively) records with only these two kinds of fields.
842 It also returns false if any of these records contains a bit-field. */
845 type_consists_of_records_p (tree type
)
849 if (TREE_CODE (type
) != RECORD_TYPE
)
852 for (fld
= TYPE_FIELDS (type
); fld
; fld
= DECL_CHAIN (fld
))
853 if (TREE_CODE (fld
) == FIELD_DECL
)
855 tree ft
= TREE_TYPE (fld
);
857 if (DECL_BIT_FIELD (fld
))
860 if (!is_gimple_reg_type (ft
)
861 && !type_consists_of_records_p (ft
))
868 /* Create total_scalarization accesses for all scalar type fields in DECL that
869 must be of a RECORD_TYPE conforming to type_consists_of_records_p. BASE
870 must be the top-most VAR_DECL representing the variable, OFFSET must be the
871 offset of DECL within BASE. REF must be the memory reference expression for
875 completely_scalarize_record (tree base
, tree decl
, HOST_WIDE_INT offset
,
878 tree fld
, decl_type
= TREE_TYPE (decl
);
880 for (fld
= TYPE_FIELDS (decl_type
); fld
; fld
= DECL_CHAIN (fld
))
881 if (TREE_CODE (fld
) == FIELD_DECL
)
883 HOST_WIDE_INT pos
= offset
+ int_bit_position (fld
);
884 tree ft
= TREE_TYPE (fld
);
885 tree nref
= build3 (COMPONENT_REF
, TREE_TYPE (fld
), ref
, fld
,
888 if (is_gimple_reg_type (ft
))
890 struct access
*access
;
893 size
= tree_low_cst (DECL_SIZE (fld
), 1);
894 access
= create_access_1 (base
, pos
, size
);
897 access
->total_scalarization
= 1;
898 /* Accesses for intraprocedural SRA can have their stmt NULL. */
901 completely_scalarize_record (base
, fld
, pos
, nref
);
906 /* Search the given tree for a declaration by skipping handled components and
907 exclude it from the candidates. */
910 disqualify_base_of_expr (tree t
, const char *reason
)
912 t
= get_base_address (t
);
913 if (sra_mode
== SRA_MODE_EARLY_IPA
914 && TREE_CODE (t
) == MEM_REF
)
915 t
= get_ssa_base_param (TREE_OPERAND (t
, 0));
918 disqualify_candidate (t
, reason
);
921 /* Scan expression EXPR and create access structures for all accesses to
922 candidates for scalarization. Return the created access or NULL if none is
925 static struct access
*
926 build_access_from_expr_1 (tree expr
, gimple stmt
, bool write
)
928 struct access
*ret
= NULL
;
931 if (TREE_CODE (expr
) == BIT_FIELD_REF
932 || TREE_CODE (expr
) == IMAGPART_EXPR
933 || TREE_CODE (expr
) == REALPART_EXPR
)
935 expr
= TREE_OPERAND (expr
, 0);
941 /* We need to dive through V_C_Es in order to get the size of its parameter
942 and not the result type. Ada produces such statements. We are also
943 capable of handling the topmost V_C_E but not any of those buried in other
944 handled components. */
945 if (TREE_CODE (expr
) == VIEW_CONVERT_EXPR
)
946 expr
= TREE_OPERAND (expr
, 0);
948 if (contains_view_convert_expr_p (expr
))
950 disqualify_base_of_expr (expr
, "V_C_E under a different handled "
955 switch (TREE_CODE (expr
))
958 if (TREE_CODE (TREE_OPERAND (expr
, 0)) != ADDR_EXPR
959 && sra_mode
!= SRA_MODE_EARLY_IPA
)
967 case ARRAY_RANGE_REF
:
968 ret
= create_access (expr
, stmt
, write
);
975 if (write
&& partial_ref
&& ret
)
976 ret
->grp_partial_lhs
= 1;
981 /* Scan expression EXPR and create access structures for all accesses to
982 candidates for scalarization. Return true if any access has been inserted.
983 STMT must be the statement from which the expression is taken, WRITE must be
984 true if the expression is a store and false otherwise. */
987 build_access_from_expr (tree expr
, gimple stmt
, bool write
)
989 struct access
*access
;
991 access
= build_access_from_expr_1 (expr
, stmt
, write
);
994 /* This means the aggregate is accesses as a whole in a way other than an
995 assign statement and thus cannot be removed even if we had a scalar
996 replacement for everything. */
997 if (cannot_scalarize_away_bitmap
)
998 bitmap_set_bit (cannot_scalarize_away_bitmap
, DECL_UID (access
->base
));
1004 /* Disqualify LHS and RHS for scalarization if STMT must end its basic block in
1005 modes in which it matters, return true iff they have been disqualified. RHS
1006 may be NULL, in that case ignore it. If we scalarize an aggregate in
1007 intra-SRA we may need to add statements after each statement. This is not
1008 possible if a statement unconditionally has to end the basic block. */
1010 disqualify_ops_if_throwing_stmt (gimple stmt
, tree lhs
, tree rhs
)
1012 if ((sra_mode
== SRA_MODE_EARLY_INTRA
|| sra_mode
== SRA_MODE_INTRA
)
1013 && (stmt_can_throw_internal (stmt
) || stmt_ends_bb_p (stmt
)))
1015 disqualify_base_of_expr (lhs
, "LHS of a throwing stmt.");
1017 disqualify_base_of_expr (rhs
, "RHS of a throwing stmt.");
1023 /* Scan expressions occuring in STMT, create access structures for all accesses
1024 to candidates for scalarization and remove those candidates which occur in
1025 statements or expressions that prevent them from being split apart. Return
1026 true if any access has been inserted. */
1029 build_accesses_from_assign (gimple stmt
)
1032 struct access
*lacc
, *racc
;
1034 if (!gimple_assign_single_p (stmt
))
1037 lhs
= gimple_assign_lhs (stmt
);
1038 rhs
= gimple_assign_rhs1 (stmt
);
1040 if (disqualify_ops_if_throwing_stmt (stmt
, lhs
, rhs
))
1043 racc
= build_access_from_expr_1 (rhs
, stmt
, false);
1044 lacc
= build_access_from_expr_1 (lhs
, stmt
, true);
1047 lacc
->grp_assignment_write
= 1;
1051 racc
->grp_assignment_read
= 1;
1052 if (should_scalarize_away_bitmap
&& !gimple_has_volatile_ops (stmt
)
1053 && !is_gimple_reg_type (racc
->type
))
1054 bitmap_set_bit (should_scalarize_away_bitmap
, DECL_UID (racc
->base
));
1058 && (sra_mode
== SRA_MODE_EARLY_INTRA
|| sra_mode
== SRA_MODE_INTRA
)
1059 && !lacc
->grp_unscalarizable_region
1060 && !racc
->grp_unscalarizable_region
1061 && AGGREGATE_TYPE_P (TREE_TYPE (lhs
))
1062 /* FIXME: Turn the following line into an assert after PR 40058 is
1064 && lacc
->size
== racc
->size
1065 && useless_type_conversion_p (lacc
->type
, racc
->type
))
1067 struct assign_link
*link
;
1069 link
= (struct assign_link
*) pool_alloc (link_pool
);
1070 memset (link
, 0, sizeof (struct assign_link
));
1075 add_link_to_rhs (racc
, link
);
1078 return lacc
|| racc
;
1081 /* Callback of walk_stmt_load_store_addr_ops visit_addr used to determine
1082 GIMPLE_ASM operands with memory constrains which cannot be scalarized. */
1085 asm_visit_addr (gimple stmt ATTRIBUTE_UNUSED
, tree op
,
1086 void *data ATTRIBUTE_UNUSED
)
1088 op
= get_base_address (op
);
1091 disqualify_candidate (op
, "Non-scalarizable GIMPLE_ASM operand.");
1096 /* Return true iff callsite CALL has at least as many actual arguments as there
1097 are formal parameters of the function currently processed by IPA-SRA. */
1100 callsite_has_enough_arguments_p (gimple call
)
1102 return gimple_call_num_args (call
) >= (unsigned) func_param_count
;
1105 /* Scan function and look for interesting expressions and create access
1106 structures for them. Return true iff any access is created. */
1109 scan_function (void)
1116 gimple_stmt_iterator gsi
;
1117 for (gsi
= gsi_start_bb (bb
); !gsi_end_p (gsi
); gsi_next (&gsi
))
1119 gimple stmt
= gsi_stmt (gsi
);
1123 if (final_bbs
&& stmt_can_throw_external (stmt
))
1124 bitmap_set_bit (final_bbs
, bb
->index
);
1125 switch (gimple_code (stmt
))
1128 t
= gimple_return_retval (stmt
);
1130 ret
|= build_access_from_expr (t
, stmt
, false);
1132 bitmap_set_bit (final_bbs
, bb
->index
);
1136 ret
|= build_accesses_from_assign (stmt
);
1140 for (i
= 0; i
< gimple_call_num_args (stmt
); i
++)
1141 ret
|= build_access_from_expr (gimple_call_arg (stmt
, i
),
1144 if (sra_mode
== SRA_MODE_EARLY_IPA
)
1146 tree dest
= gimple_call_fndecl (stmt
);
1147 int flags
= gimple_call_flags (stmt
);
1151 if (DECL_BUILT_IN_CLASS (dest
) == BUILT_IN_NORMAL
1152 && DECL_FUNCTION_CODE (dest
) == BUILT_IN_APPLY_ARGS
)
1153 encountered_apply_args
= true;
1154 if (cgraph_get_node (dest
)
1155 == cgraph_get_node (current_function_decl
))
1157 encountered_recursive_call
= true;
1158 if (!callsite_has_enough_arguments_p (stmt
))
1159 encountered_unchangable_recursive_call
= true;
1164 && (flags
& (ECF_CONST
| ECF_PURE
)) == 0)
1165 bitmap_set_bit (final_bbs
, bb
->index
);
1168 t
= gimple_call_lhs (stmt
);
1169 if (t
&& !disqualify_ops_if_throwing_stmt (stmt
, t
, NULL
))
1170 ret
|= build_access_from_expr (t
, stmt
, true);
1174 walk_stmt_load_store_addr_ops (stmt
, NULL
, NULL
, NULL
,
1177 bitmap_set_bit (final_bbs
, bb
->index
);
1179 for (i
= 0; i
< gimple_asm_ninputs (stmt
); i
++)
1181 t
= TREE_VALUE (gimple_asm_input_op (stmt
, i
));
1182 ret
|= build_access_from_expr (t
, stmt
, false);
1184 for (i
= 0; i
< gimple_asm_noutputs (stmt
); i
++)
1186 t
= TREE_VALUE (gimple_asm_output_op (stmt
, i
));
1187 ret
|= build_access_from_expr (t
, stmt
, true);
1200 /* Helper of QSORT function. There are pointers to accesses in the array. An
1201 access is considered smaller than another if it has smaller offset or if the
1202 offsets are the same but is size is bigger. */
1205 compare_access_positions (const void *a
, const void *b
)
1207 const access_p
*fp1
= (const access_p
*) a
;
1208 const access_p
*fp2
= (const access_p
*) b
;
1209 const access_p f1
= *fp1
;
1210 const access_p f2
= *fp2
;
1212 if (f1
->offset
!= f2
->offset
)
1213 return f1
->offset
< f2
->offset
? -1 : 1;
1215 if (f1
->size
== f2
->size
)
1217 if (f1
->type
== f2
->type
)
1219 /* Put any non-aggregate type before any aggregate type. */
1220 else if (!is_gimple_reg_type (f1
->type
)
1221 && is_gimple_reg_type (f2
->type
))
1223 else if (is_gimple_reg_type (f1
->type
)
1224 && !is_gimple_reg_type (f2
->type
))
1226 /* Put any complex or vector type before any other scalar type. */
1227 else if (TREE_CODE (f1
->type
) != COMPLEX_TYPE
1228 && TREE_CODE (f1
->type
) != VECTOR_TYPE
1229 && (TREE_CODE (f2
->type
) == COMPLEX_TYPE
1230 || TREE_CODE (f2
->type
) == VECTOR_TYPE
))
1232 else if ((TREE_CODE (f1
->type
) == COMPLEX_TYPE
1233 || TREE_CODE (f1
->type
) == VECTOR_TYPE
)
1234 && TREE_CODE (f2
->type
) != COMPLEX_TYPE
1235 && TREE_CODE (f2
->type
) != VECTOR_TYPE
)
1237 /* Put the integral type with the bigger precision first. */
1238 else if (INTEGRAL_TYPE_P (f1
->type
)
1239 && INTEGRAL_TYPE_P (f2
->type
))
1240 return TYPE_PRECISION (f2
->type
) - TYPE_PRECISION (f1
->type
);
1241 /* Put any integral type with non-full precision last. */
1242 else if (INTEGRAL_TYPE_P (f1
->type
)
1243 && (TREE_INT_CST_LOW (TYPE_SIZE (f1
->type
))
1244 != TYPE_PRECISION (f1
->type
)))
1246 else if (INTEGRAL_TYPE_P (f2
->type
)
1247 && (TREE_INT_CST_LOW (TYPE_SIZE (f2
->type
))
1248 != TYPE_PRECISION (f2
->type
)))
1250 /* Stabilize the sort. */
1251 return TYPE_UID (f1
->type
) - TYPE_UID (f2
->type
);
1254 /* We want the bigger accesses first, thus the opposite operator in the next
1256 return f1
->size
> f2
->size
? -1 : 1;
1260 /* Append a name of the declaration to the name obstack. A helper function for
1264 make_fancy_decl_name (tree decl
)
1268 tree name
= DECL_NAME (decl
);
1270 obstack_grow (&name_obstack
, IDENTIFIER_POINTER (name
),
1271 IDENTIFIER_LENGTH (name
));
1274 sprintf (buffer
, "D%u", DECL_UID (decl
));
1275 obstack_grow (&name_obstack
, buffer
, strlen (buffer
));
1279 /* Helper for make_fancy_name. */
1282 make_fancy_name_1 (tree expr
)
1289 make_fancy_decl_name (expr
);
1293 switch (TREE_CODE (expr
))
1296 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1297 obstack_1grow (&name_obstack
, '$');
1298 make_fancy_decl_name (TREE_OPERAND (expr
, 1));
1302 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1303 obstack_1grow (&name_obstack
, '$');
1304 /* Arrays with only one element may not have a constant as their
1306 index
= TREE_OPERAND (expr
, 1);
1307 if (TREE_CODE (index
) != INTEGER_CST
)
1309 sprintf (buffer
, HOST_WIDE_INT_PRINT_DEC
, TREE_INT_CST_LOW (index
));
1310 obstack_grow (&name_obstack
, buffer
, strlen (buffer
));
1314 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1318 make_fancy_name_1 (TREE_OPERAND (expr
, 0));
1319 if (!integer_zerop (TREE_OPERAND (expr
, 1)))
1321 obstack_1grow (&name_obstack
, '$');
1322 sprintf (buffer
, HOST_WIDE_INT_PRINT_DEC
,
1323 TREE_INT_CST_LOW (TREE_OPERAND (expr
, 1)));
1324 obstack_grow (&name_obstack
, buffer
, strlen (buffer
));
1331 gcc_unreachable (); /* we treat these as scalars. */
1338 /* Create a human readable name for replacement variable of ACCESS. */
1341 make_fancy_name (tree expr
)
1343 make_fancy_name_1 (expr
);
1344 obstack_1grow (&name_obstack
, '\0');
1345 return XOBFINISH (&name_obstack
, char *);
1348 /* Construct a MEM_REF that would reference a part of aggregate BASE of type
1349 EXP_TYPE at the given OFFSET. If BASE is something for which
1350 get_addr_base_and_unit_offset returns NULL, gsi must be non-NULL and is used
1351 to insert new statements either before or below the current one as specified
1352 by INSERT_AFTER. This function is not capable of handling bitfields. */
1355 build_ref_for_offset (location_t loc
, tree base
, HOST_WIDE_INT offset
,
1356 tree exp_type
, gimple_stmt_iterator
*gsi
,
1359 tree prev_base
= base
;
1361 HOST_WIDE_INT base_offset
;
1363 gcc_checking_assert (offset
% BITS_PER_UNIT
== 0);
1365 base
= get_addr_base_and_unit_offset (base
, &base_offset
);
1367 /* get_addr_base_and_unit_offset returns NULL for references with a variable
1368 offset such as array[var_index]. */
1374 gcc_checking_assert (gsi
);
1375 tmp
= create_tmp_reg (build_pointer_type (TREE_TYPE (prev_base
)), NULL
);
1376 add_referenced_var (tmp
);
1377 tmp
= make_ssa_name (tmp
, NULL
);
1378 addr
= build_fold_addr_expr (unshare_expr (prev_base
));
1379 stmt
= gimple_build_assign (tmp
, addr
);
1380 gimple_set_location (stmt
, loc
);
1381 SSA_NAME_DEF_STMT (tmp
) = stmt
;
1383 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
1385 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
1388 off
= build_int_cst (reference_alias_ptr_type (prev_base
),
1389 offset
/ BITS_PER_UNIT
);
1392 else if (TREE_CODE (base
) == MEM_REF
)
1394 off
= build_int_cst (TREE_TYPE (TREE_OPERAND (base
, 1)),
1395 base_offset
+ offset
/ BITS_PER_UNIT
);
1396 off
= int_const_binop (PLUS_EXPR
, TREE_OPERAND (base
, 1), off
, 0);
1397 base
= unshare_expr (TREE_OPERAND (base
, 0));
1401 off
= build_int_cst (reference_alias_ptr_type (base
),
1402 base_offset
+ offset
/ BITS_PER_UNIT
);
1403 base
= build_fold_addr_expr (unshare_expr (base
));
1406 return fold_build2_loc (loc
, MEM_REF
, exp_type
, base
, off
);
1409 /* Construct a memory reference to a part of an aggregate BASE at the given
1410 OFFSET and of the same type as MODEL. In case this is a reference to a
1411 component, the function will replicate the last COMPONENT_REF of model's
1412 expr to access it. GSI and INSERT_AFTER have the same meaning as in
1413 build_ref_for_offset. */
1416 build_ref_for_model (location_t loc
, tree base
, HOST_WIDE_INT offset
,
1417 struct access
*model
, gimple_stmt_iterator
*gsi
,
1420 if (TREE_CODE (model
->expr
) == COMPONENT_REF
)
1423 offset
-= int_bit_position (TREE_OPERAND (model
->expr
, 1));
1424 exp_type
= TREE_TYPE (TREE_OPERAND (model
->expr
, 0));
1425 t
= build_ref_for_offset (loc
, base
, offset
, exp_type
, gsi
, insert_after
);
1426 return fold_build3_loc (loc
, COMPONENT_REF
, model
->type
, t
,
1427 TREE_OPERAND (model
->expr
, 1), NULL_TREE
);
1430 return build_ref_for_offset (loc
, base
, offset
, model
->type
,
1434 /* Construct a memory reference consisting of component_refs and array_refs to
1435 a part of an aggregate *RES (which is of type TYPE). The requested part
1436 should have type EXP_TYPE at be the given OFFSET. This function might not
1437 succeed, it returns true when it does and only then *RES points to something
1438 meaningful. This function should be used only to build expressions that we
1439 might need to present to user (e.g. in warnings). In all other situations,
1440 build_ref_for_model or build_ref_for_offset should be used instead. */
1443 build_user_friendly_ref_for_offset (tree
*res
, tree type
, HOST_WIDE_INT offset
,
1449 tree tr_size
, index
, minidx
;
1450 HOST_WIDE_INT el_size
;
1452 if (offset
== 0 && exp_type
1453 && types_compatible_p (exp_type
, type
))
1456 switch (TREE_CODE (type
))
1459 case QUAL_UNION_TYPE
:
1461 for (fld
= TYPE_FIELDS (type
); fld
; fld
= DECL_CHAIN (fld
))
1463 HOST_WIDE_INT pos
, size
;
1464 tree expr
, *expr_ptr
;
1466 if (TREE_CODE (fld
) != FIELD_DECL
)
1469 pos
= int_bit_position (fld
);
1470 gcc_assert (TREE_CODE (type
) == RECORD_TYPE
|| pos
== 0);
1471 tr_size
= DECL_SIZE (fld
);
1472 if (!tr_size
|| !host_integerp (tr_size
, 1))
1474 size
= tree_low_cst (tr_size
, 1);
1480 else if (pos
> offset
|| (pos
+ size
) <= offset
)
1483 expr
= build3 (COMPONENT_REF
, TREE_TYPE (fld
), *res
, fld
,
1486 if (build_user_friendly_ref_for_offset (expr_ptr
, TREE_TYPE (fld
),
1487 offset
- pos
, exp_type
))
1496 tr_size
= TYPE_SIZE (TREE_TYPE (type
));
1497 if (!tr_size
|| !host_integerp (tr_size
, 1))
1499 el_size
= tree_low_cst (tr_size
, 1);
1501 minidx
= TYPE_MIN_VALUE (TYPE_DOMAIN (type
));
1502 if (TREE_CODE (minidx
) != INTEGER_CST
|| el_size
== 0)
1504 index
= build_int_cst (TYPE_DOMAIN (type
), offset
/ el_size
);
1505 if (!integer_zerop (minidx
))
1506 index
= int_const_binop (PLUS_EXPR
, index
, minidx
, 0);
1507 *res
= build4 (ARRAY_REF
, TREE_TYPE (type
), *res
, index
,
1508 NULL_TREE
, NULL_TREE
);
1509 offset
= offset
% el_size
;
1510 type
= TREE_TYPE (type
);
1525 /* Return true iff TYPE is stdarg va_list type. */
1528 is_va_list_type (tree type
)
1530 return TYPE_MAIN_VARIANT (type
) == TYPE_MAIN_VARIANT (va_list_type_node
);
1533 /* The very first phase of intraprocedural SRA. It marks in candidate_bitmap
1534 those with type which is suitable for scalarization. */
1537 find_var_candidates (void)
1540 referenced_var_iterator rvi
;
1543 FOR_EACH_REFERENCED_VAR (var
, rvi
)
1545 if (TREE_CODE (var
) != VAR_DECL
&& TREE_CODE (var
) != PARM_DECL
)
1547 type
= TREE_TYPE (var
);
1549 if (!AGGREGATE_TYPE_P (type
)
1550 || needs_to_live_in_memory (var
)
1551 || TREE_THIS_VOLATILE (var
)
1552 || !COMPLETE_TYPE_P (type
)
1553 || !host_integerp (TYPE_SIZE (type
), 1)
1554 || tree_low_cst (TYPE_SIZE (type
), 1) == 0
1555 || type_internals_preclude_sra_p (type
)
1556 /* Fix for PR 41089. tree-stdarg.c needs to have va_lists intact but
1557 we also want to schedule it rather late. Thus we ignore it in
1559 || (sra_mode
== SRA_MODE_EARLY_INTRA
1560 && is_va_list_type (type
)))
1563 bitmap_set_bit (candidate_bitmap
, DECL_UID (var
));
1565 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1567 fprintf (dump_file
, "Candidate (%d): ", DECL_UID (var
));
1568 print_generic_expr (dump_file
, var
, 0);
1569 fprintf (dump_file
, "\n");
1577 /* Sort all accesses for the given variable, check for partial overlaps and
1578 return NULL if there are any. If there are none, pick a representative for
1579 each combination of offset and size and create a linked list out of them.
1580 Return the pointer to the first representative and make sure it is the first
1581 one in the vector of accesses. */
1583 static struct access
*
1584 sort_and_splice_var_accesses (tree var
)
1586 int i
, j
, access_count
;
1587 struct access
*res
, **prev_acc_ptr
= &res
;
1588 VEC (access_p
, heap
) *access_vec
;
1590 HOST_WIDE_INT low
= -1, high
= 0;
1592 access_vec
= get_base_access_vector (var
);
1595 access_count
= VEC_length (access_p
, access_vec
);
1597 /* Sort by <OFFSET, SIZE>. */
1598 VEC_qsort (access_p
, access_vec
, compare_access_positions
);
1601 while (i
< access_count
)
1603 struct access
*access
= VEC_index (access_p
, access_vec
, i
);
1604 bool grp_write
= access
->write
;
1605 bool grp_read
= !access
->write
;
1606 bool grp_scalar_write
= access
->write
1607 && is_gimple_reg_type (access
->type
);
1608 bool grp_scalar_read
= !access
->write
1609 && is_gimple_reg_type (access
->type
);
1610 bool grp_assignment_read
= access
->grp_assignment_read
;
1611 bool grp_assignment_write
= access
->grp_assignment_write
;
1612 bool multiple_scalar_reads
= false;
1613 bool total_scalarization
= access
->total_scalarization
;
1614 bool grp_partial_lhs
= access
->grp_partial_lhs
;
1615 bool first_scalar
= is_gimple_reg_type (access
->type
);
1616 bool unscalarizable_region
= access
->grp_unscalarizable_region
;
1618 if (first
|| access
->offset
>= high
)
1621 low
= access
->offset
;
1622 high
= access
->offset
+ access
->size
;
1624 else if (access
->offset
> low
&& access
->offset
+ access
->size
> high
)
1627 gcc_assert (access
->offset
>= low
1628 && access
->offset
+ access
->size
<= high
);
1631 while (j
< access_count
)
1633 struct access
*ac2
= VEC_index (access_p
, access_vec
, j
);
1634 if (ac2
->offset
!= access
->offset
|| ac2
->size
!= access
->size
)
1639 grp_scalar_write
= (grp_scalar_write
1640 || is_gimple_reg_type (ac2
->type
));
1645 if (is_gimple_reg_type (ac2
->type
))
1647 if (grp_scalar_read
)
1648 multiple_scalar_reads
= true;
1650 grp_scalar_read
= true;
1653 grp_assignment_read
|= ac2
->grp_assignment_read
;
1654 grp_assignment_write
|= ac2
->grp_assignment_write
;
1655 grp_partial_lhs
|= ac2
->grp_partial_lhs
;
1656 unscalarizable_region
|= ac2
->grp_unscalarizable_region
;
1657 total_scalarization
|= ac2
->total_scalarization
;
1658 relink_to_new_repr (access
, ac2
);
1660 /* If there are both aggregate-type and scalar-type accesses with
1661 this combination of size and offset, the comparison function
1662 should have put the scalars first. */
1663 gcc_assert (first_scalar
|| !is_gimple_reg_type (ac2
->type
));
1664 ac2
->group_representative
= access
;
1670 access
->group_representative
= access
;
1671 access
->grp_write
= grp_write
;
1672 access
->grp_read
= grp_read
;
1673 access
->grp_scalar_read
= grp_scalar_read
;
1674 access
->grp_scalar_write
= grp_scalar_write
;
1675 access
->grp_assignment_read
= grp_assignment_read
;
1676 access
->grp_assignment_write
= grp_assignment_write
;
1677 access
->grp_hint
= multiple_scalar_reads
|| total_scalarization
;
1678 access
->grp_partial_lhs
= grp_partial_lhs
;
1679 access
->grp_unscalarizable_region
= unscalarizable_region
;
1680 if (access
->first_link
)
1681 add_access_to_work_queue (access
);
1683 *prev_acc_ptr
= access
;
1684 prev_acc_ptr
= &access
->next_grp
;
1687 gcc_assert (res
== VEC_index (access_p
, access_vec
, 0));
1691 /* Create a variable for the given ACCESS which determines the type, name and a
1692 few other properties. Return the variable declaration and store it also to
1693 ACCESS->replacement. */
1696 create_access_replacement (struct access
*access
, bool rename
)
1700 repl
= create_tmp_var (access
->type
, "SR");
1702 add_referenced_var (repl
);
1704 mark_sym_for_renaming (repl
);
1706 if (!access
->grp_partial_lhs
1707 && (TREE_CODE (access
->type
) == COMPLEX_TYPE
1708 || TREE_CODE (access
->type
) == VECTOR_TYPE
))
1709 DECL_GIMPLE_REG_P (repl
) = 1;
1711 DECL_SOURCE_LOCATION (repl
) = DECL_SOURCE_LOCATION (access
->base
);
1712 DECL_ARTIFICIAL (repl
) = 1;
1713 DECL_IGNORED_P (repl
) = DECL_IGNORED_P (access
->base
);
1715 if (DECL_NAME (access
->base
)
1716 && !DECL_IGNORED_P (access
->base
)
1717 && !DECL_ARTIFICIAL (access
->base
))
1719 char *pretty_name
= make_fancy_name (access
->expr
);
1720 tree debug_expr
= unshare_expr (access
->expr
), d
;
1722 DECL_NAME (repl
) = get_identifier (pretty_name
);
1723 obstack_free (&name_obstack
, pretty_name
);
1725 /* Get rid of any SSA_NAMEs embedded in debug_expr,
1726 as DECL_DEBUG_EXPR isn't considered when looking for still
1727 used SSA_NAMEs and thus they could be freed. All debug info
1728 generation cares is whether something is constant or variable
1729 and that get_ref_base_and_extent works properly on the
1731 for (d
= debug_expr
; handled_component_p (d
); d
= TREE_OPERAND (d
, 0))
1732 switch (TREE_CODE (d
))
1735 case ARRAY_RANGE_REF
:
1736 if (TREE_OPERAND (d
, 1)
1737 && TREE_CODE (TREE_OPERAND (d
, 1)) == SSA_NAME
)
1738 TREE_OPERAND (d
, 1) = SSA_NAME_VAR (TREE_OPERAND (d
, 1));
1739 if (TREE_OPERAND (d
, 3)
1740 && TREE_CODE (TREE_OPERAND (d
, 3)) == SSA_NAME
)
1741 TREE_OPERAND (d
, 3) = SSA_NAME_VAR (TREE_OPERAND (d
, 3));
1744 if (TREE_OPERAND (d
, 2)
1745 && TREE_CODE (TREE_OPERAND (d
, 2)) == SSA_NAME
)
1746 TREE_OPERAND (d
, 2) = SSA_NAME_VAR (TREE_OPERAND (d
, 2));
1751 SET_DECL_DEBUG_EXPR (repl
, debug_expr
);
1752 DECL_DEBUG_EXPR_IS_FROM (repl
) = 1;
1753 if (access
->grp_no_warning
)
1754 TREE_NO_WARNING (repl
) = 1;
1756 TREE_NO_WARNING (repl
) = TREE_NO_WARNING (access
->base
);
1759 TREE_NO_WARNING (repl
) = 1;
1763 fprintf (dump_file
, "Created a replacement for ");
1764 print_generic_expr (dump_file
, access
->base
, 0);
1765 fprintf (dump_file
, " offset: %u, size: %u: ",
1766 (unsigned) access
->offset
, (unsigned) access
->size
);
1767 print_generic_expr (dump_file
, repl
, 0);
1768 fprintf (dump_file
, "\n");
1770 sra_stats
.replacements
++;
1775 /* Return ACCESS scalar replacement, create it if it does not exist yet. */
1778 get_access_replacement (struct access
*access
)
1780 gcc_assert (access
->grp_to_be_replaced
);
1782 if (!access
->replacement_decl
)
1783 access
->replacement_decl
= create_access_replacement (access
, true);
1784 return access
->replacement_decl
;
1787 /* Return ACCESS scalar replacement, create it if it does not exist yet but do
1788 not mark it for renaming. */
1791 get_unrenamed_access_replacement (struct access
*access
)
1793 gcc_assert (!access
->grp_to_be_replaced
);
1795 if (!access
->replacement_decl
)
1796 access
->replacement_decl
= create_access_replacement (access
, false);
1797 return access
->replacement_decl
;
1801 /* Build a subtree of accesses rooted in *ACCESS, and move the pointer in the
1802 linked list along the way. Stop when *ACCESS is NULL or the access pointed
1803 to it is not "within" the root. Return false iff some accesses partially
1807 build_access_subtree (struct access
**access
)
1809 struct access
*root
= *access
, *last_child
= NULL
;
1810 HOST_WIDE_INT limit
= root
->offset
+ root
->size
;
1812 *access
= (*access
)->next_grp
;
1813 while (*access
&& (*access
)->offset
+ (*access
)->size
<= limit
)
1816 root
->first_child
= *access
;
1818 last_child
->next_sibling
= *access
;
1819 last_child
= *access
;
1821 if (!build_access_subtree (access
))
1825 if (*access
&& (*access
)->offset
< limit
)
1831 /* Build a tree of access representatives, ACCESS is the pointer to the first
1832 one, others are linked in a list by the next_grp field. Return false iff
1833 some accesses partially overlap. */
1836 build_access_trees (struct access
*access
)
1840 struct access
*root
= access
;
1842 if (!build_access_subtree (&access
))
1844 root
->next_grp
= access
;
1849 /* Return true if expr contains some ARRAY_REFs into a variable bounded
1853 expr_with_var_bounded_array_refs_p (tree expr
)
1855 while (handled_component_p (expr
))
1857 if (TREE_CODE (expr
) == ARRAY_REF
1858 && !host_integerp (array_ref_low_bound (expr
), 0))
1860 expr
= TREE_OPERAND (expr
, 0);
1865 enum mark_rw_status
{ SRA_MRRW_NOTHING
, SRA_MRRW_DIRECT
, SRA_MRRW_ASSIGN
};
1867 /* Analyze the subtree of accesses rooted in ROOT, scheduling replacements when
1868 both seeming beneficial and when ALLOW_REPLACEMENTS allows it. Also set all
1869 sorts of access flags appropriately along the way, notably always set
1870 grp_read and grp_assign_read according to MARK_READ and grp_write when
1873 Creating a replacement for a scalar access is considered beneficial if its
1874 grp_hint is set (this means we are either attempting total scalarization or
1875 there is more than one direct read access) or according to the following
1878 Access written to through a scalar type (once or more times)
1880 | Written to in an assignment statement
1882 | | Access read as scalar _once_
1884 | | | Read in an assignment statement
1886 | | | | Scalarize Comment
1887 -----------------------------------------------------------------------------
1888 0 0 0 0 No access for the scalar
1889 0 0 0 1 No access for the scalar
1890 0 0 1 0 No Single read - won't help
1891 0 0 1 1 No The same case
1892 0 1 0 0 No access for the scalar
1893 0 1 0 1 No access for the scalar
1894 0 1 1 0 Yes s = *g; return s.i;
1895 0 1 1 1 Yes The same case as above
1896 1 0 0 0 No Won't help
1897 1 0 0 1 Yes s.i = 1; *g = s;
1898 1 0 1 0 Yes s.i = 5; g = s.i;
1899 1 0 1 1 Yes The same case as above
1900 1 1 0 0 No Won't help.
1901 1 1 0 1 Yes s.i = 1; *g = s;
1902 1 1 1 0 Yes s = *g; return s.i;
1903 1 1 1 1 Yes Any of the above yeses */
1906 analyze_access_subtree (struct access
*root
, bool allow_replacements
,
1907 enum mark_rw_status mark_read
,
1908 enum mark_rw_status mark_write
)
1910 struct access
*child
;
1911 HOST_WIDE_INT limit
= root
->offset
+ root
->size
;
1912 HOST_WIDE_INT covered_to
= root
->offset
;
1913 bool scalar
= is_gimple_reg_type (root
->type
);
1914 bool hole
= false, sth_created
= false;
1916 if (root
->grp_assignment_read
)
1917 mark_read
= SRA_MRRW_ASSIGN
;
1918 else if (mark_read
== SRA_MRRW_ASSIGN
)
1921 root
->grp_assignment_read
= 1;
1923 else if (mark_read
== SRA_MRRW_DIRECT
)
1925 else if (root
->grp_read
)
1926 mark_read
= SRA_MRRW_DIRECT
;
1928 if (root
->grp_assignment_write
)
1929 mark_write
= SRA_MRRW_ASSIGN
;
1930 else if (mark_write
== SRA_MRRW_ASSIGN
)
1932 root
->grp_write
= 1;
1933 root
->grp_assignment_write
= 1;
1935 else if (mark_write
== SRA_MRRW_DIRECT
)
1936 root
->grp_write
= 1;
1937 else if (root
->grp_write
)
1938 mark_write
= SRA_MRRW_DIRECT
;
1940 if (root
->grp_unscalarizable_region
)
1941 allow_replacements
= false;
1943 if (allow_replacements
&& expr_with_var_bounded_array_refs_p (root
->expr
))
1944 allow_replacements
= false;
1946 for (child
= root
->first_child
; child
; child
= child
->next_sibling
)
1948 if (!hole
&& child
->offset
< covered_to
)
1951 covered_to
+= child
->size
;
1953 sth_created
|= analyze_access_subtree (child
,
1954 allow_replacements
&& !scalar
,
1955 mark_read
, mark_write
);
1957 root
->grp_unscalarized_data
|= child
->grp_unscalarized_data
;
1958 hole
|= !child
->grp_covered
;
1961 if (allow_replacements
&& scalar
&& !root
->first_child
1963 || ((root
->grp_scalar_read
|| root
->grp_assignment_read
)
1964 && (root
->grp_scalar_write
|| root
->grp_assignment_write
))))
1966 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1968 fprintf (dump_file
, "Marking ");
1969 print_generic_expr (dump_file
, root
->base
, 0);
1970 fprintf (dump_file
, " offset: %u, size: %u: ",
1971 (unsigned) root
->offset
, (unsigned) root
->size
);
1972 fprintf (dump_file
, " to be replaced.\n");
1975 root
->grp_to_be_replaced
= 1;
1979 else if (covered_to
< limit
)
1982 if (sth_created
&& !hole
)
1984 root
->grp_covered
= 1;
1987 if (root
->grp_write
|| TREE_CODE (root
->base
) == PARM_DECL
)
1988 root
->grp_unscalarized_data
= 1; /* not covered and written to */
1994 /* Analyze all access trees linked by next_grp by the means of
1995 analyze_access_subtree. */
1997 analyze_access_trees (struct access
*access
)
2003 if (analyze_access_subtree (access
, true,
2004 SRA_MRRW_NOTHING
, SRA_MRRW_NOTHING
))
2006 access
= access
->next_grp
;
2012 /* Return true iff a potential new child of LACC at offset OFFSET and with size
2013 SIZE would conflict with an already existing one. If exactly such a child
2014 already exists in LACC, store a pointer to it in EXACT_MATCH. */
2017 child_would_conflict_in_lacc (struct access
*lacc
, HOST_WIDE_INT norm_offset
,
2018 HOST_WIDE_INT size
, struct access
**exact_match
)
2020 struct access
*child
;
2022 for (child
= lacc
->first_child
; child
; child
= child
->next_sibling
)
2024 if (child
->offset
== norm_offset
&& child
->size
== size
)
2026 *exact_match
= child
;
2030 if (child
->offset
< norm_offset
+ size
2031 && child
->offset
+ child
->size
> norm_offset
)
2038 /* Create a new child access of PARENT, with all properties just like MODEL
2039 except for its offset and with its grp_write false and grp_read true.
2040 Return the new access or NULL if it cannot be created. Note that this access
2041 is created long after all splicing and sorting, it's not located in any
2042 access vector and is automatically a representative of its group. */
2044 static struct access
*
2045 create_artificial_child_access (struct access
*parent
, struct access
*model
,
2046 HOST_WIDE_INT new_offset
)
2048 struct access
*access
;
2049 struct access
**child
;
2050 tree expr
= parent
->base
;
2052 gcc_assert (!model
->grp_unscalarizable_region
);
2054 access
= (struct access
*) pool_alloc (access_pool
);
2055 memset (access
, 0, sizeof (struct access
));
2056 if (!build_user_friendly_ref_for_offset (&expr
, TREE_TYPE (expr
), new_offset
,
2059 access
->grp_no_warning
= true;
2060 expr
= build_ref_for_model (EXPR_LOCATION (parent
->base
), parent
->base
,
2061 new_offset
, model
, NULL
, false);
2064 access
->base
= parent
->base
;
2065 access
->expr
= expr
;
2066 access
->offset
= new_offset
;
2067 access
->size
= model
->size
;
2068 access
->type
= model
->type
;
2069 access
->grp_write
= true;
2070 access
->grp_read
= false;
2072 child
= &parent
->first_child
;
2073 while (*child
&& (*child
)->offset
< new_offset
)
2074 child
= &(*child
)->next_sibling
;
2076 access
->next_sibling
= *child
;
2083 /* Propagate all subaccesses of RACC across an assignment link to LACC. Return
2084 true if any new subaccess was created. Additionally, if RACC is a scalar
2085 access but LACC is not, change the type of the latter, if possible. */
2088 propagate_subaccesses_across_link (struct access
*lacc
, struct access
*racc
)
2090 struct access
*rchild
;
2091 HOST_WIDE_INT norm_delta
= lacc
->offset
- racc
->offset
;
2094 if (is_gimple_reg_type (lacc
->type
)
2095 || lacc
->grp_unscalarizable_region
2096 || racc
->grp_unscalarizable_region
)
2099 if (!lacc
->first_child
&& !racc
->first_child
2100 && is_gimple_reg_type (racc
->type
))
2102 tree t
= lacc
->base
;
2104 lacc
->type
= racc
->type
;
2105 if (build_user_friendly_ref_for_offset (&t
, TREE_TYPE (t
), lacc
->offset
,
2110 lacc
->expr
= build_ref_for_model (EXPR_LOCATION (lacc
->base
),
2111 lacc
->base
, lacc
->offset
,
2113 lacc
->grp_no_warning
= true;
2118 for (rchild
= racc
->first_child
; rchild
; rchild
= rchild
->next_sibling
)
2120 struct access
*new_acc
= NULL
;
2121 HOST_WIDE_INT norm_offset
= rchild
->offset
+ norm_delta
;
2123 if (rchild
->grp_unscalarizable_region
)
2126 if (child_would_conflict_in_lacc (lacc
, norm_offset
, rchild
->size
,
2131 rchild
->grp_hint
= 1;
2132 new_acc
->grp_hint
|= new_acc
->grp_read
;
2133 if (rchild
->first_child
)
2134 ret
|= propagate_subaccesses_across_link (new_acc
, rchild
);
2139 rchild
->grp_hint
= 1;
2140 new_acc
= create_artificial_child_access (lacc
, rchild
, norm_offset
);
2144 if (racc
->first_child
)
2145 propagate_subaccesses_across_link (new_acc
, rchild
);
2152 /* Propagate all subaccesses across assignment links. */
2155 propagate_all_subaccesses (void)
2157 while (work_queue_head
)
2159 struct access
*racc
= pop_access_from_work_queue ();
2160 struct assign_link
*link
;
2162 gcc_assert (racc
->first_link
);
2164 for (link
= racc
->first_link
; link
; link
= link
->next
)
2166 struct access
*lacc
= link
->lacc
;
2168 if (!bitmap_bit_p (candidate_bitmap
, DECL_UID (lacc
->base
)))
2170 lacc
= lacc
->group_representative
;
2171 if (propagate_subaccesses_across_link (lacc
, racc
)
2172 && lacc
->first_link
)
2173 add_access_to_work_queue (lacc
);
2178 /* Go through all accesses collected throughout the (intraprocedural) analysis
2179 stage, exclude overlapping ones, identify representatives and build trees
2180 out of them, making decisions about scalarization on the way. Return true
2181 iff there are any to-be-scalarized variables after this stage. */
2184 analyze_all_variable_accesses (void)
2187 bitmap tmp
= BITMAP_ALLOC (NULL
);
2189 unsigned i
, max_total_scalarization_size
;
2191 max_total_scalarization_size
= UNITS_PER_WORD
* BITS_PER_UNIT
2192 * MOVE_RATIO (optimize_function_for_speed_p (cfun
));
2194 EXECUTE_IF_SET_IN_BITMAP (candidate_bitmap
, 0, i
, bi
)
2195 if (bitmap_bit_p (should_scalarize_away_bitmap
, i
)
2196 && !bitmap_bit_p (cannot_scalarize_away_bitmap
, i
))
2198 tree var
= referenced_var (i
);
2200 if (TREE_CODE (var
) == VAR_DECL
2201 && ((unsigned) tree_low_cst (TYPE_SIZE (TREE_TYPE (var
)), 1)
2202 <= max_total_scalarization_size
)
2203 && type_consists_of_records_p (TREE_TYPE (var
)))
2205 completely_scalarize_record (var
, var
, 0, var
);
2206 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2208 fprintf (dump_file
, "Will attempt to totally scalarize ");
2209 print_generic_expr (dump_file
, var
, 0);
2210 fprintf (dump_file
, " (UID: %u): \n", DECL_UID (var
));
2215 bitmap_copy (tmp
, candidate_bitmap
);
2216 EXECUTE_IF_SET_IN_BITMAP (tmp
, 0, i
, bi
)
2218 tree var
= referenced_var (i
);
2219 struct access
*access
;
2221 access
= sort_and_splice_var_accesses (var
);
2222 if (!access
|| !build_access_trees (access
))
2223 disqualify_candidate (var
,
2224 "No or inhibitingly overlapping accesses.");
2227 propagate_all_subaccesses ();
2229 bitmap_copy (tmp
, candidate_bitmap
);
2230 EXECUTE_IF_SET_IN_BITMAP (tmp
, 0, i
, bi
)
2232 tree var
= referenced_var (i
);
2233 struct access
*access
= get_first_repr_for_decl (var
);
2235 if (analyze_access_trees (access
))
2238 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2240 fprintf (dump_file
, "\nAccess trees for ");
2241 print_generic_expr (dump_file
, var
, 0);
2242 fprintf (dump_file
, " (UID: %u): \n", DECL_UID (var
));
2243 dump_access_tree (dump_file
, access
);
2244 fprintf (dump_file
, "\n");
2248 disqualify_candidate (var
, "No scalar replacements to be created.");
2255 statistics_counter_event (cfun
, "Scalarized aggregates", res
);
2262 /* Generate statements copying scalar replacements of accesses within a subtree
2263 into or out of AGG. ACCESS, all its children, siblings and their children
2264 are to be processed. AGG is an aggregate type expression (can be a
2265 declaration but does not have to be, it can for example also be a mem_ref or
2266 a series of handled components). TOP_OFFSET is the offset of the processed
2267 subtree which has to be subtracted from offsets of individual accesses to
2268 get corresponding offsets for AGG. If CHUNK_SIZE is non-null, copy only
2269 replacements in the interval <start_offset, start_offset + chunk_size>,
2270 otherwise copy all. GSI is a statement iterator used to place the new
2271 statements. WRITE should be true when the statements should write from AGG
2272 to the replacement and false if vice versa. if INSERT_AFTER is true, new
2273 statements will be added after the current statement in GSI, they will be
2274 added before the statement otherwise. */
2277 generate_subtree_copies (struct access
*access
, tree agg
,
2278 HOST_WIDE_INT top_offset
,
2279 HOST_WIDE_INT start_offset
, HOST_WIDE_INT chunk_size
,
2280 gimple_stmt_iterator
*gsi
, bool write
,
2281 bool insert_after
, location_t loc
)
2285 if (chunk_size
&& access
->offset
>= start_offset
+ chunk_size
)
2288 if (access
->grp_to_be_replaced
2290 || access
->offset
+ access
->size
> start_offset
))
2292 tree expr
, repl
= get_access_replacement (access
);
2295 expr
= build_ref_for_model (loc
, agg
, access
->offset
- top_offset
,
2296 access
, gsi
, insert_after
);
2300 if (access
->grp_partial_lhs
)
2301 expr
= force_gimple_operand_gsi (gsi
, expr
, true, NULL_TREE
,
2303 insert_after
? GSI_NEW_STMT
2305 stmt
= gimple_build_assign (repl
, expr
);
2309 TREE_NO_WARNING (repl
) = 1;
2310 if (access
->grp_partial_lhs
)
2311 repl
= force_gimple_operand_gsi (gsi
, repl
, true, NULL_TREE
,
2313 insert_after
? GSI_NEW_STMT
2315 stmt
= gimple_build_assign (expr
, repl
);
2317 gimple_set_location (stmt
, loc
);
2320 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
2322 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
2324 sra_stats
.subtree_copies
++;
2327 if (access
->first_child
)
2328 generate_subtree_copies (access
->first_child
, agg
, top_offset
,
2329 start_offset
, chunk_size
, gsi
,
2330 write
, insert_after
, loc
);
2332 access
= access
->next_sibling
;
2337 /* Assign zero to all scalar replacements in an access subtree. ACCESS is the
2338 the root of the subtree to be processed. GSI is the statement iterator used
2339 for inserting statements which are added after the current statement if
2340 INSERT_AFTER is true or before it otherwise. */
2343 init_subtree_with_zero (struct access
*access
, gimple_stmt_iterator
*gsi
,
2344 bool insert_after
, location_t loc
)
2347 struct access
*child
;
2349 if (access
->grp_to_be_replaced
)
2353 stmt
= gimple_build_assign (get_access_replacement (access
),
2354 build_zero_cst (access
->type
));
2356 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
2358 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
2360 gimple_set_location (stmt
, loc
);
2363 for (child
= access
->first_child
; child
; child
= child
->next_sibling
)
2364 init_subtree_with_zero (child
, gsi
, insert_after
, loc
);
2367 /* Search for an access representative for the given expression EXPR and
2368 return it or NULL if it cannot be found. */
2370 static struct access
*
2371 get_access_for_expr (tree expr
)
2373 HOST_WIDE_INT offset
, size
, max_size
;
2376 /* FIXME: This should not be necessary but Ada produces V_C_Es with a type of
2377 a different size than the size of its argument and we need the latter
2379 if (TREE_CODE (expr
) == VIEW_CONVERT_EXPR
)
2380 expr
= TREE_OPERAND (expr
, 0);
2382 base
= get_ref_base_and_extent (expr
, &offset
, &size
, &max_size
);
2383 if (max_size
== -1 || !DECL_P (base
))
2386 if (!bitmap_bit_p (candidate_bitmap
, DECL_UID (base
)))
2389 return get_var_base_offset_size_access (base
, offset
, max_size
);
2392 /* Replace the expression EXPR with a scalar replacement if there is one and
2393 generate other statements to do type conversion or subtree copying if
2394 necessary. GSI is used to place newly created statements, WRITE is true if
2395 the expression is being written to (it is on a LHS of a statement or output
2396 in an assembly statement). */
2399 sra_modify_expr (tree
*expr
, gimple_stmt_iterator
*gsi
, bool write
)
2402 struct access
*access
;
2405 if (TREE_CODE (*expr
) == BIT_FIELD_REF
)
2408 expr
= &TREE_OPERAND (*expr
, 0);
2413 if (TREE_CODE (*expr
) == REALPART_EXPR
|| TREE_CODE (*expr
) == IMAGPART_EXPR
)
2414 expr
= &TREE_OPERAND (*expr
, 0);
2415 access
= get_access_for_expr (*expr
);
2418 type
= TREE_TYPE (*expr
);
2420 loc
= gimple_location (gsi_stmt (*gsi
));
2421 if (access
->grp_to_be_replaced
)
2423 tree repl
= get_access_replacement (access
);
2424 /* If we replace a non-register typed access simply use the original
2425 access expression to extract the scalar component afterwards.
2426 This happens if scalarizing a function return value or parameter
2427 like in gcc.c-torture/execute/20041124-1.c, 20050316-1.c and
2428 gcc.c-torture/compile/20011217-1.c.
2430 We also want to use this when accessing a complex or vector which can
2431 be accessed as a different type too, potentially creating a need for
2432 type conversion (see PR42196) and when scalarized unions are involved
2433 in assembler statements (see PR42398). */
2434 if (!useless_type_conversion_p (type
, access
->type
))
2438 ref
= build_ref_for_model (loc
, access
->base
, access
->offset
, access
,
2445 if (access
->grp_partial_lhs
)
2446 ref
= force_gimple_operand_gsi (gsi
, ref
, true, NULL_TREE
,
2447 false, GSI_NEW_STMT
);
2448 stmt
= gimple_build_assign (repl
, ref
);
2449 gimple_set_location (stmt
, loc
);
2450 gsi_insert_after (gsi
, stmt
, GSI_NEW_STMT
);
2456 if (access
->grp_partial_lhs
)
2457 repl
= force_gimple_operand_gsi (gsi
, repl
, true, NULL_TREE
,
2458 true, GSI_SAME_STMT
);
2459 stmt
= gimple_build_assign (ref
, repl
);
2460 gimple_set_location (stmt
, loc
);
2461 gsi_insert_before (gsi
, stmt
, GSI_SAME_STMT
);
2469 if (access
->first_child
)
2471 HOST_WIDE_INT start_offset
, chunk_size
;
2473 && host_integerp (TREE_OPERAND (bfr
, 1), 1)
2474 && host_integerp (TREE_OPERAND (bfr
, 2), 1))
2476 chunk_size
= tree_low_cst (TREE_OPERAND (bfr
, 1), 1);
2477 start_offset
= access
->offset
2478 + tree_low_cst (TREE_OPERAND (bfr
, 2), 1);
2481 start_offset
= chunk_size
= 0;
2483 generate_subtree_copies (access
->first_child
, access
->base
, 0,
2484 start_offset
, chunk_size
, gsi
, write
, write
,
2490 /* Where scalar replacements of the RHS have been written to when a replacement
2491 of a LHS of an assigments cannot be direclty loaded from a replacement of
2493 enum unscalarized_data_handling
{ SRA_UDH_NONE
, /* Nothing done so far. */
2494 SRA_UDH_RIGHT
, /* Data flushed to the RHS. */
2495 SRA_UDH_LEFT
}; /* Data flushed to the LHS. */
2497 /* Store all replacements in the access tree rooted in TOP_RACC either to their
2498 base aggregate if there are unscalarized data or directly to LHS of the
2499 statement that is pointed to by GSI otherwise. */
2501 static enum unscalarized_data_handling
2502 handle_unscalarized_data_in_subtree (struct access
*top_racc
,
2503 gimple_stmt_iterator
*gsi
)
2505 if (top_racc
->grp_unscalarized_data
)
2507 generate_subtree_copies (top_racc
->first_child
, top_racc
->base
, 0, 0, 0,
2509 gimple_location (gsi_stmt (*gsi
)));
2510 return SRA_UDH_RIGHT
;
2514 tree lhs
= gimple_assign_lhs (gsi_stmt (*gsi
));
2515 generate_subtree_copies (top_racc
->first_child
, lhs
, top_racc
->offset
,
2516 0, 0, gsi
, false, false,
2517 gimple_location (gsi_stmt (*gsi
)));
2518 return SRA_UDH_LEFT
;
2523 /* Try to generate statements to load all sub-replacements in an access subtree
2524 formed by children of LACC from scalar replacements in the TOP_RACC subtree.
2525 If that is not possible, refresh the TOP_RACC base aggregate and load the
2526 accesses from it. LEFT_OFFSET is the offset of the left whole subtree being
2527 copied. NEW_GSI is stmt iterator used for statement insertions after the
2528 original assignment, OLD_GSI is used to insert statements before the
2529 assignment. *REFRESHED keeps the information whether we have needed to
2530 refresh replacements of the LHS and from which side of the assignments this
2534 load_assign_lhs_subreplacements (struct access
*lacc
, struct access
*top_racc
,
2535 HOST_WIDE_INT left_offset
,
2536 gimple_stmt_iterator
*old_gsi
,
2537 gimple_stmt_iterator
*new_gsi
,
2538 enum unscalarized_data_handling
*refreshed
)
2540 location_t loc
= gimple_location (gsi_stmt (*old_gsi
));
2541 for (lacc
= lacc
->first_child
; lacc
; lacc
= lacc
->next_sibling
)
2543 if (lacc
->grp_to_be_replaced
)
2545 struct access
*racc
;
2546 HOST_WIDE_INT offset
= lacc
->offset
- left_offset
+ top_racc
->offset
;
2550 racc
= find_access_in_subtree (top_racc
, offset
, lacc
->size
);
2551 if (racc
&& racc
->grp_to_be_replaced
)
2553 rhs
= get_access_replacement (racc
);
2554 if (!useless_type_conversion_p (lacc
->type
, racc
->type
))
2555 rhs
= fold_build1_loc (loc
, VIEW_CONVERT_EXPR
, lacc
->type
, rhs
);
2559 /* No suitable access on the right hand side, need to load from
2560 the aggregate. See if we have to update it first... */
2561 if (*refreshed
== SRA_UDH_NONE
)
2562 *refreshed
= handle_unscalarized_data_in_subtree (top_racc
,
2565 if (*refreshed
== SRA_UDH_LEFT
)
2566 rhs
= build_ref_for_model (loc
, lacc
->base
, lacc
->offset
, lacc
,
2569 rhs
= build_ref_for_model (loc
, top_racc
->base
, offset
, lacc
,
2573 stmt
= gimple_build_assign (get_access_replacement (lacc
), rhs
);
2574 gsi_insert_after (new_gsi
, stmt
, GSI_NEW_STMT
);
2575 gimple_set_location (stmt
, loc
);
2577 sra_stats
.subreplacements
++;
2579 else if (*refreshed
== SRA_UDH_NONE
2580 && lacc
->grp_read
&& !lacc
->grp_covered
)
2581 *refreshed
= handle_unscalarized_data_in_subtree (top_racc
,
2584 if (lacc
->first_child
)
2585 load_assign_lhs_subreplacements (lacc
, top_racc
, left_offset
,
2586 old_gsi
, new_gsi
, refreshed
);
2590 /* Result code for SRA assignment modification. */
2591 enum assignment_mod_result
{ SRA_AM_NONE
, /* nothing done for the stmt */
2592 SRA_AM_MODIFIED
, /* stmt changed but not
2594 SRA_AM_REMOVED
}; /* stmt eliminated */
2596 /* Modify assignments with a CONSTRUCTOR on their RHS. STMT contains a pointer
2597 to the assignment and GSI is the statement iterator pointing at it. Returns
2598 the same values as sra_modify_assign. */
2600 static enum assignment_mod_result
2601 sra_modify_constructor_assign (gimple
*stmt
, gimple_stmt_iterator
*gsi
)
2603 tree lhs
= gimple_assign_lhs (*stmt
);
2607 acc
= get_access_for_expr (lhs
);
2611 loc
= gimple_location (*stmt
);
2612 if (VEC_length (constructor_elt
,
2613 CONSTRUCTOR_ELTS (gimple_assign_rhs1 (*stmt
))) > 0)
2615 /* I have never seen this code path trigger but if it can happen the
2616 following should handle it gracefully. */
2617 if (access_has_children_p (acc
))
2618 generate_subtree_copies (acc
->first_child
, acc
->base
, 0, 0, 0, gsi
,
2620 return SRA_AM_MODIFIED
;
2623 if (acc
->grp_covered
)
2625 init_subtree_with_zero (acc
, gsi
, false, loc
);
2626 unlink_stmt_vdef (*stmt
);
2627 gsi_remove (gsi
, true);
2628 return SRA_AM_REMOVED
;
2632 init_subtree_with_zero (acc
, gsi
, true, loc
);
2633 return SRA_AM_MODIFIED
;
2637 /* Create and return a new suitable default definition SSA_NAME for RACC which
2638 is an access describing an uninitialized part of an aggregate that is being
2642 get_repl_default_def_ssa_name (struct access
*racc
)
2646 decl
= get_unrenamed_access_replacement (racc
);
2648 repl
= gimple_default_def (cfun
, decl
);
2651 repl
= make_ssa_name (decl
, gimple_build_nop ());
2652 set_default_def (decl
, repl
);
2658 /* Return true if REF has a COMPONENT_REF with a bit-field field declaration
2662 contains_bitfld_comp_ref_p (const_tree ref
)
2664 while (handled_component_p (ref
))
2666 if (TREE_CODE (ref
) == COMPONENT_REF
2667 && DECL_BIT_FIELD (TREE_OPERAND (ref
, 1)))
2669 ref
= TREE_OPERAND (ref
, 0);
2675 /* Return true if REF has an VIEW_CONVERT_EXPR or a COMPONENT_REF with a
2676 bit-field field declaration somewhere in it. */
2679 contains_vce_or_bfcref_p (const_tree ref
)
2681 while (handled_component_p (ref
))
2683 if (TREE_CODE (ref
) == VIEW_CONVERT_EXPR
2684 || (TREE_CODE (ref
) == COMPONENT_REF
2685 && DECL_BIT_FIELD (TREE_OPERAND (ref
, 1))))
2687 ref
= TREE_OPERAND (ref
, 0);
2693 /* Examine both sides of the assignment statement pointed to by STMT, replace
2694 them with a scalare replacement if there is one and generate copying of
2695 replacements if scalarized aggregates have been used in the assignment. GSI
2696 is used to hold generated statements for type conversions and subtree
2699 static enum assignment_mod_result
2700 sra_modify_assign (gimple
*stmt
, gimple_stmt_iterator
*gsi
)
2702 struct access
*lacc
, *racc
;
2704 bool modify_this_stmt
= false;
2705 bool force_gimple_rhs
= false;
2707 gimple_stmt_iterator orig_gsi
= *gsi
;
2709 if (!gimple_assign_single_p (*stmt
))
2711 lhs
= gimple_assign_lhs (*stmt
);
2712 rhs
= gimple_assign_rhs1 (*stmt
);
2714 if (TREE_CODE (rhs
) == CONSTRUCTOR
)
2715 return sra_modify_constructor_assign (stmt
, gsi
);
2717 if (TREE_CODE (rhs
) == REALPART_EXPR
|| TREE_CODE (lhs
) == REALPART_EXPR
2718 || TREE_CODE (rhs
) == IMAGPART_EXPR
|| TREE_CODE (lhs
) == IMAGPART_EXPR
2719 || TREE_CODE (rhs
) == BIT_FIELD_REF
|| TREE_CODE (lhs
) == BIT_FIELD_REF
)
2721 modify_this_stmt
= sra_modify_expr (gimple_assign_rhs1_ptr (*stmt
),
2723 modify_this_stmt
|= sra_modify_expr (gimple_assign_lhs_ptr (*stmt
),
2725 return modify_this_stmt
? SRA_AM_MODIFIED
: SRA_AM_NONE
;
2728 lacc
= get_access_for_expr (lhs
);
2729 racc
= get_access_for_expr (rhs
);
2733 loc
= gimple_location (*stmt
);
2734 if (lacc
&& lacc
->grp_to_be_replaced
)
2736 lhs
= get_access_replacement (lacc
);
2737 gimple_assign_set_lhs (*stmt
, lhs
);
2738 modify_this_stmt
= true;
2739 if (lacc
->grp_partial_lhs
)
2740 force_gimple_rhs
= true;
2744 if (racc
&& racc
->grp_to_be_replaced
)
2746 rhs
= get_access_replacement (racc
);
2747 modify_this_stmt
= true;
2748 if (racc
->grp_partial_lhs
)
2749 force_gimple_rhs
= true;
2753 if (modify_this_stmt
)
2755 if (!useless_type_conversion_p (TREE_TYPE (lhs
), TREE_TYPE (rhs
)))
2757 /* If we can avoid creating a VIEW_CONVERT_EXPR do so.
2758 ??? This should move to fold_stmt which we simply should
2759 call after building a VIEW_CONVERT_EXPR here. */
2760 if (AGGREGATE_TYPE_P (TREE_TYPE (lhs
))
2761 && !contains_bitfld_comp_ref_p (lhs
)
2762 && !access_has_children_p (lacc
))
2764 lhs
= build_ref_for_model (loc
, lhs
, 0, racc
, gsi
, false);
2765 gimple_assign_set_lhs (*stmt
, lhs
);
2767 else if (AGGREGATE_TYPE_P (TREE_TYPE (rhs
))
2768 && !contains_vce_or_bfcref_p (rhs
)
2769 && !access_has_children_p (racc
))
2770 rhs
= build_ref_for_model (loc
, rhs
, 0, lacc
, gsi
, false);
2772 if (!useless_type_conversion_p (TREE_TYPE (lhs
), TREE_TYPE (rhs
)))
2774 rhs
= fold_build1_loc (loc
, VIEW_CONVERT_EXPR
, TREE_TYPE (lhs
),
2776 if (is_gimple_reg_type (TREE_TYPE (lhs
))
2777 && TREE_CODE (lhs
) != SSA_NAME
)
2778 force_gimple_rhs
= true;
2783 /* From this point on, the function deals with assignments in between
2784 aggregates when at least one has scalar reductions of some of its
2785 components. There are three possible scenarios: Both the LHS and RHS have
2786 to-be-scalarized components, 2) only the RHS has or 3) only the LHS has.
2788 In the first case, we would like to load the LHS components from RHS
2789 components whenever possible. If that is not possible, we would like to
2790 read it directly from the RHS (after updating it by storing in it its own
2791 components). If there are some necessary unscalarized data in the LHS,
2792 those will be loaded by the original assignment too. If neither of these
2793 cases happen, the original statement can be removed. Most of this is done
2794 by load_assign_lhs_subreplacements.
2796 In the second case, we would like to store all RHS scalarized components
2797 directly into LHS and if they cover the aggregate completely, remove the
2798 statement too. In the third case, we want the LHS components to be loaded
2799 directly from the RHS (DSE will remove the original statement if it
2802 This is a bit complex but manageable when types match and when unions do
2803 not cause confusion in a way that we cannot really load a component of LHS
2804 from the RHS or vice versa (the access representing this level can have
2805 subaccesses that are accessible only through a different union field at a
2806 higher level - different from the one used in the examined expression).
2809 Therefore, I specially handle a fourth case, happening when there is a
2810 specific type cast or it is impossible to locate a scalarized subaccess on
2811 the other side of the expression. If that happens, I simply "refresh" the
2812 RHS by storing in it is scalarized components leave the original statement
2813 there to do the copying and then load the scalar replacements of the LHS.
2814 This is what the first branch does. */
2816 if (gimple_has_volatile_ops (*stmt
)
2817 || contains_vce_or_bfcref_p (rhs
)
2818 || contains_vce_or_bfcref_p (lhs
))
2820 if (access_has_children_p (racc
))
2821 generate_subtree_copies (racc
->first_child
, racc
->base
, 0, 0, 0,
2822 gsi
, false, false, loc
);
2823 if (access_has_children_p (lacc
))
2824 generate_subtree_copies (lacc
->first_child
, lacc
->base
, 0, 0, 0,
2825 gsi
, true, true, loc
);
2826 sra_stats
.separate_lhs_rhs_handling
++;
2830 if (access_has_children_p (lacc
) && access_has_children_p (racc
))
2832 gimple_stmt_iterator orig_gsi
= *gsi
;
2833 enum unscalarized_data_handling refreshed
;
2835 if (lacc
->grp_read
&& !lacc
->grp_covered
)
2836 refreshed
= handle_unscalarized_data_in_subtree (racc
, gsi
);
2838 refreshed
= SRA_UDH_NONE
;
2840 load_assign_lhs_subreplacements (lacc
, racc
, lacc
->offset
,
2841 &orig_gsi
, gsi
, &refreshed
);
2842 if (refreshed
!= SRA_UDH_RIGHT
)
2845 unlink_stmt_vdef (*stmt
);
2846 gsi_remove (&orig_gsi
, true);
2847 sra_stats
.deleted
++;
2848 return SRA_AM_REMOVED
;
2855 if (!racc
->grp_to_be_replaced
&& !racc
->grp_unscalarized_data
)
2859 fprintf (dump_file
, "Removing load: ");
2860 print_gimple_stmt (dump_file
, *stmt
, 0, 0);
2863 if (TREE_CODE (lhs
) == SSA_NAME
)
2865 rhs
= get_repl_default_def_ssa_name (racc
);
2866 if (!useless_type_conversion_p (TREE_TYPE (lhs
),
2868 rhs
= fold_build1_loc (loc
, VIEW_CONVERT_EXPR
,
2869 TREE_TYPE (lhs
), rhs
);
2873 if (racc
->first_child
)
2874 generate_subtree_copies (racc
->first_child
, lhs
,
2875 racc
->offset
, 0, 0, gsi
,
2878 gcc_assert (*stmt
== gsi_stmt (*gsi
));
2879 unlink_stmt_vdef (*stmt
);
2880 gsi_remove (gsi
, true);
2881 sra_stats
.deleted
++;
2882 return SRA_AM_REMOVED
;
2885 else if (racc
->first_child
)
2886 generate_subtree_copies (racc
->first_child
, lhs
, racc
->offset
,
2887 0, 0, gsi
, false, true, loc
);
2889 if (access_has_children_p (lacc
))
2890 generate_subtree_copies (lacc
->first_child
, rhs
, lacc
->offset
,
2891 0, 0, gsi
, true, true, loc
);
2895 /* This gimplification must be done after generate_subtree_copies, lest we
2896 insert the subtree copies in the middle of the gimplified sequence. */
2897 if (force_gimple_rhs
)
2898 rhs
= force_gimple_operand_gsi (&orig_gsi
, rhs
, true, NULL_TREE
,
2899 true, GSI_SAME_STMT
);
2900 if (gimple_assign_rhs1 (*stmt
) != rhs
)
2902 modify_this_stmt
= true;
2903 gimple_assign_set_rhs_from_tree (&orig_gsi
, rhs
);
2904 gcc_assert (*stmt
== gsi_stmt (orig_gsi
));
2907 return modify_this_stmt
? SRA_AM_MODIFIED
: SRA_AM_NONE
;
2910 /* Traverse the function body and all modifications as decided in
2911 analyze_all_variable_accesses. Return true iff the CFG has been
2915 sra_modify_function_body (void)
2917 bool cfg_changed
= false;
2922 gimple_stmt_iterator gsi
= gsi_start_bb (bb
);
2923 while (!gsi_end_p (gsi
))
2925 gimple stmt
= gsi_stmt (gsi
);
2926 enum assignment_mod_result assign_result
;
2927 bool modified
= false, deleted
= false;
2931 switch (gimple_code (stmt
))
2934 t
= gimple_return_retval_ptr (stmt
);
2935 if (*t
!= NULL_TREE
)
2936 modified
|= sra_modify_expr (t
, &gsi
, false);
2940 assign_result
= sra_modify_assign (&stmt
, &gsi
);
2941 modified
|= assign_result
== SRA_AM_MODIFIED
;
2942 deleted
= assign_result
== SRA_AM_REMOVED
;
2946 /* Operands must be processed before the lhs. */
2947 for (i
= 0; i
< gimple_call_num_args (stmt
); i
++)
2949 t
= gimple_call_arg_ptr (stmt
, i
);
2950 modified
|= sra_modify_expr (t
, &gsi
, false);
2953 if (gimple_call_lhs (stmt
))
2955 t
= gimple_call_lhs_ptr (stmt
);
2956 modified
|= sra_modify_expr (t
, &gsi
, true);
2961 for (i
= 0; i
< gimple_asm_ninputs (stmt
); i
++)
2963 t
= &TREE_VALUE (gimple_asm_input_op (stmt
, i
));
2964 modified
|= sra_modify_expr (t
, &gsi
, false);
2966 for (i
= 0; i
< gimple_asm_noutputs (stmt
); i
++)
2968 t
= &TREE_VALUE (gimple_asm_output_op (stmt
, i
));
2969 modified
|= sra_modify_expr (t
, &gsi
, true);
2980 if (maybe_clean_eh_stmt (stmt
)
2981 && gimple_purge_dead_eh_edges (gimple_bb (stmt
)))
2992 /* Generate statements initializing scalar replacements of parts of function
2996 initialize_parameter_reductions (void)
2998 gimple_stmt_iterator gsi
;
2999 gimple_seq seq
= NULL
;
3002 for (parm
= DECL_ARGUMENTS (current_function_decl
);
3004 parm
= DECL_CHAIN (parm
))
3006 VEC (access_p
, heap
) *access_vec
;
3007 struct access
*access
;
3009 if (!bitmap_bit_p (candidate_bitmap
, DECL_UID (parm
)))
3011 access_vec
= get_base_access_vector (parm
);
3017 seq
= gimple_seq_alloc ();
3018 gsi
= gsi_start (seq
);
3021 for (access
= VEC_index (access_p
, access_vec
, 0);
3023 access
= access
->next_grp
)
3024 generate_subtree_copies (access
, parm
, 0, 0, 0, &gsi
, true, true,
3025 EXPR_LOCATION (parm
));
3029 gsi_insert_seq_on_edge_immediate (single_succ_edge (ENTRY_BLOCK_PTR
), seq
);
3032 /* The "main" function of intraprocedural SRA passes. Runs the analysis and if
3033 it reveals there are components of some aggregates to be scalarized, it runs
3034 the required transformations. */
3036 perform_intra_sra (void)
3041 if (!find_var_candidates ())
3044 if (!scan_function ())
3047 if (!analyze_all_variable_accesses ())
3050 if (sra_modify_function_body ())
3051 ret
= TODO_update_ssa
| TODO_cleanup_cfg
;
3053 ret
= TODO_update_ssa
;
3054 initialize_parameter_reductions ();
3056 statistics_counter_event (cfun
, "Scalar replacements created",
3057 sra_stats
.replacements
);
3058 statistics_counter_event (cfun
, "Modified expressions", sra_stats
.exprs
);
3059 statistics_counter_event (cfun
, "Subtree copy stmts",
3060 sra_stats
.subtree_copies
);
3061 statistics_counter_event (cfun
, "Subreplacement stmts",
3062 sra_stats
.subreplacements
);
3063 statistics_counter_event (cfun
, "Deleted stmts", sra_stats
.deleted
);
3064 statistics_counter_event (cfun
, "Separate LHS and RHS handling",
3065 sra_stats
.separate_lhs_rhs_handling
);
3068 sra_deinitialize ();
3072 /* Perform early intraprocedural SRA. */
3074 early_intra_sra (void)
3076 sra_mode
= SRA_MODE_EARLY_INTRA
;
3077 return perform_intra_sra ();
3080 /* Perform "late" intraprocedural SRA. */
3082 late_intra_sra (void)
3084 sra_mode
= SRA_MODE_INTRA
;
3085 return perform_intra_sra ();
3090 gate_intra_sra (void)
3092 return flag_tree_sra
!= 0 && dbg_cnt (tree_sra
);
3096 struct gimple_opt_pass pass_sra_early
=
3101 gate_intra_sra
, /* gate */
3102 early_intra_sra
, /* execute */
3105 0, /* static_pass_number */
3106 TV_TREE_SRA
, /* tv_id */
3107 PROP_cfg
| PROP_ssa
, /* properties_required */
3108 0, /* properties_provided */
3109 0, /* properties_destroyed */
3110 0, /* todo_flags_start */
3114 | TODO_verify_ssa
/* todo_flags_finish */
3118 struct gimple_opt_pass pass_sra
=
3123 gate_intra_sra
, /* gate */
3124 late_intra_sra
, /* execute */
3127 0, /* static_pass_number */
3128 TV_TREE_SRA
, /* tv_id */
3129 PROP_cfg
| PROP_ssa
, /* properties_required */
3130 0, /* properties_provided */
3131 0, /* properties_destroyed */
3132 TODO_update_address_taken
, /* todo_flags_start */
3136 | TODO_verify_ssa
/* todo_flags_finish */
3141 /* Return true iff PARM (which must be a parm_decl) is an unused scalar
3145 is_unused_scalar_param (tree parm
)
3148 return (is_gimple_reg (parm
)
3149 && (!(name
= gimple_default_def (cfun
, parm
))
3150 || has_zero_uses (name
)));
3153 /* Scan immediate uses of a default definition SSA name of a parameter PARM and
3154 examine whether there are any direct or otherwise infeasible ones. If so,
3155 return true, otherwise return false. PARM must be a gimple register with a
3156 non-NULL default definition. */
3159 ptr_parm_has_direct_uses (tree parm
)
3161 imm_use_iterator ui
;
3163 tree name
= gimple_default_def (cfun
, parm
);
3166 FOR_EACH_IMM_USE_STMT (stmt
, ui
, name
)
3169 use_operand_p use_p
;
3171 if (is_gimple_debug (stmt
))
3174 /* Valid uses include dereferences on the lhs and the rhs. */
3175 if (gimple_has_lhs (stmt
))
3177 tree lhs
= gimple_get_lhs (stmt
);
3178 while (handled_component_p (lhs
))
3179 lhs
= TREE_OPERAND (lhs
, 0);
3180 if (TREE_CODE (lhs
) == MEM_REF
3181 && TREE_OPERAND (lhs
, 0) == name
3182 && integer_zerop (TREE_OPERAND (lhs
, 1))
3183 && types_compatible_p (TREE_TYPE (lhs
),
3184 TREE_TYPE (TREE_TYPE (name
))))
3187 if (gimple_assign_single_p (stmt
))
3189 tree rhs
= gimple_assign_rhs1 (stmt
);
3190 while (handled_component_p (rhs
))
3191 rhs
= TREE_OPERAND (rhs
, 0);
3192 if (TREE_CODE (rhs
) == MEM_REF
3193 && TREE_OPERAND (rhs
, 0) == name
3194 && integer_zerop (TREE_OPERAND (rhs
, 1))
3195 && types_compatible_p (TREE_TYPE (rhs
),
3196 TREE_TYPE (TREE_TYPE (name
))))
3199 else if (is_gimple_call (stmt
))
3202 for (i
= 0; i
< gimple_call_num_args (stmt
); ++i
)
3204 tree arg
= gimple_call_arg (stmt
, i
);
3205 while (handled_component_p (arg
))
3206 arg
= TREE_OPERAND (arg
, 0);
3207 if (TREE_CODE (arg
) == MEM_REF
3208 && TREE_OPERAND (arg
, 0) == name
3209 && integer_zerop (TREE_OPERAND (arg
, 1))
3210 && types_compatible_p (TREE_TYPE (arg
),
3211 TREE_TYPE (TREE_TYPE (name
))))
3216 /* If the number of valid uses does not match the number of
3217 uses in this stmt there is an unhandled use. */
3218 FOR_EACH_IMM_USE_ON_STMT (use_p
, ui
)
3225 BREAK_FROM_IMM_USE_STMT (ui
);
3231 /* Identify candidates for reduction for IPA-SRA based on their type and mark
3232 them in candidate_bitmap. Note that these do not necessarily include
3233 parameter which are unused and thus can be removed. Return true iff any
3234 such candidate has been found. */
3237 find_param_candidates (void)
3243 for (parm
= DECL_ARGUMENTS (current_function_decl
);
3245 parm
= DECL_CHAIN (parm
))
3247 tree type
= TREE_TYPE (parm
);
3251 if (TREE_THIS_VOLATILE (parm
)
3252 || TREE_ADDRESSABLE (parm
)
3253 || (!is_gimple_reg_type (type
) && is_va_list_type (type
)))
3256 if (is_unused_scalar_param (parm
))
3262 if (POINTER_TYPE_P (type
))
3264 type
= TREE_TYPE (type
);
3266 if (TREE_CODE (type
) == FUNCTION_TYPE
3267 || TYPE_VOLATILE (type
)
3268 || (TREE_CODE (type
) == ARRAY_TYPE
3269 && TYPE_NONALIASED_COMPONENT (type
))
3270 || !is_gimple_reg (parm
)
3271 || is_va_list_type (type
)
3272 || ptr_parm_has_direct_uses (parm
))
3275 else if (!AGGREGATE_TYPE_P (type
))
3278 if (!COMPLETE_TYPE_P (type
)
3279 || !host_integerp (TYPE_SIZE (type
), 1)
3280 || tree_low_cst (TYPE_SIZE (type
), 1) == 0
3281 || (AGGREGATE_TYPE_P (type
)
3282 && type_internals_preclude_sra_p (type
)))
3285 bitmap_set_bit (candidate_bitmap
, DECL_UID (parm
));
3287 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
3289 fprintf (dump_file
, "Candidate (%d): ", DECL_UID (parm
));
3290 print_generic_expr (dump_file
, parm
, 0);
3291 fprintf (dump_file
, "\n");
3295 func_param_count
= count
;
3299 /* Callback of walk_aliased_vdefs, marks the access passed as DATA as
3303 mark_maybe_modified (ao_ref
*ao ATTRIBUTE_UNUSED
, tree vdef ATTRIBUTE_UNUSED
,
3306 struct access
*repr
= (struct access
*) data
;
3308 repr
->grp_maybe_modified
= 1;
3312 /* Analyze what representatives (in linked lists accessible from
3313 REPRESENTATIVES) can be modified by side effects of statements in the
3314 current function. */
3317 analyze_modified_params (VEC (access_p
, heap
) *representatives
)
3321 for (i
= 0; i
< func_param_count
; i
++)
3323 struct access
*repr
;
3325 for (repr
= VEC_index (access_p
, representatives
, i
);
3327 repr
= repr
->next_grp
)
3329 struct access
*access
;
3333 if (no_accesses_p (repr
))
3335 if (!POINTER_TYPE_P (TREE_TYPE (repr
->base
))
3336 || repr
->grp_maybe_modified
)
3339 ao_ref_init (&ar
, repr
->expr
);
3340 visited
= BITMAP_ALLOC (NULL
);
3341 for (access
= repr
; access
; access
= access
->next_sibling
)
3343 /* All accesses are read ones, otherwise grp_maybe_modified would
3344 be trivially set. */
3345 walk_aliased_vdefs (&ar
, gimple_vuse (access
->stmt
),
3346 mark_maybe_modified
, repr
, &visited
);
3347 if (repr
->grp_maybe_modified
)
3350 BITMAP_FREE (visited
);
3355 /* Propagate distances in bb_dereferences in the opposite direction than the
3356 control flow edges, in each step storing the maximum of the current value
3357 and the minimum of all successors. These steps are repeated until the table
3358 stabilizes. Note that BBs which might terminate the functions (according to
3359 final_bbs bitmap) never updated in this way. */
3362 propagate_dereference_distances (void)
3364 VEC (basic_block
, heap
) *queue
;
3367 queue
= VEC_alloc (basic_block
, heap
, last_basic_block_for_function (cfun
));
3368 VEC_quick_push (basic_block
, queue
, ENTRY_BLOCK_PTR
);
3371 VEC_quick_push (basic_block
, queue
, bb
);
3375 while (!VEC_empty (basic_block
, queue
))
3379 bool change
= false;
3382 bb
= VEC_pop (basic_block
, queue
);
3385 if (bitmap_bit_p (final_bbs
, bb
->index
))
3388 for (i
= 0; i
< func_param_count
; i
++)
3390 int idx
= bb
->index
* func_param_count
+ i
;
3392 HOST_WIDE_INT inh
= 0;
3394 FOR_EACH_EDGE (e
, ei
, bb
->succs
)
3396 int succ_idx
= e
->dest
->index
* func_param_count
+ i
;
3398 if (e
->src
== EXIT_BLOCK_PTR
)
3404 inh
= bb_dereferences
[succ_idx
];
3406 else if (bb_dereferences
[succ_idx
] < inh
)
3407 inh
= bb_dereferences
[succ_idx
];
3410 if (!first
&& bb_dereferences
[idx
] < inh
)
3412 bb_dereferences
[idx
] = inh
;
3417 if (change
&& !bitmap_bit_p (final_bbs
, bb
->index
))
3418 FOR_EACH_EDGE (e
, ei
, bb
->preds
)
3423 e
->src
->aux
= e
->src
;
3424 VEC_quick_push (basic_block
, queue
, e
->src
);
3428 VEC_free (basic_block
, heap
, queue
);
3431 /* Dump a dereferences TABLE with heading STR to file F. */
3434 dump_dereferences_table (FILE *f
, const char *str
, HOST_WIDE_INT
*table
)
3438 fprintf (dump_file
, str
);
3439 FOR_BB_BETWEEN (bb
, ENTRY_BLOCK_PTR
, EXIT_BLOCK_PTR
, next_bb
)
3441 fprintf (f
, "%4i %i ", bb
->index
, bitmap_bit_p (final_bbs
, bb
->index
));
3442 if (bb
!= EXIT_BLOCK_PTR
)
3445 for (i
= 0; i
< func_param_count
; i
++)
3447 int idx
= bb
->index
* func_param_count
+ i
;
3448 fprintf (f
, " %4" HOST_WIDE_INT_PRINT
"d", table
[idx
]);
3453 fprintf (dump_file
, "\n");
3456 /* Determine what (parts of) parameters passed by reference that are not
3457 assigned to are not certainly dereferenced in this function and thus the
3458 dereferencing cannot be safely moved to the caller without potentially
3459 introducing a segfault. Mark such REPRESENTATIVES as
3460 grp_not_necessarilly_dereferenced.
3462 The dereferenced maximum "distance," i.e. the offset + size of the accessed
3463 part is calculated rather than simple booleans are calculated for each
3464 pointer parameter to handle cases when only a fraction of the whole
3465 aggregate is allocated (see testsuite/gcc.c-torture/execute/ipa-sra-2.c for
3468 The maximum dereference distances for each pointer parameter and BB are
3469 already stored in bb_dereference. This routine simply propagates these
3470 values upwards by propagate_dereference_distances and then compares the
3471 distances of individual parameters in the ENTRY BB to the equivalent
3472 distances of each representative of a (fraction of a) parameter. */
3475 analyze_caller_dereference_legality (VEC (access_p
, heap
) *representatives
)
3479 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
3480 dump_dereferences_table (dump_file
,
3481 "Dereference table before propagation:\n",
3484 propagate_dereference_distances ();
3486 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
3487 dump_dereferences_table (dump_file
,
3488 "Dereference table after propagation:\n",
3491 for (i
= 0; i
< func_param_count
; i
++)
3493 struct access
*repr
= VEC_index (access_p
, representatives
, i
);
3494 int idx
= ENTRY_BLOCK_PTR
->index
* func_param_count
+ i
;
3496 if (!repr
|| no_accesses_p (repr
))
3501 if ((repr
->offset
+ repr
->size
) > bb_dereferences
[idx
])
3502 repr
->grp_not_necessarilly_dereferenced
= 1;
3503 repr
= repr
->next_grp
;
3509 /* Return the representative access for the parameter declaration PARM if it is
3510 a scalar passed by reference which is not written to and the pointer value
3511 is not used directly. Thus, if it is legal to dereference it in the caller
3512 and we can rule out modifications through aliases, such parameter should be
3513 turned into one passed by value. Return NULL otherwise. */
3515 static struct access
*
3516 unmodified_by_ref_scalar_representative (tree parm
)
3518 int i
, access_count
;
3519 struct access
*repr
;
3520 VEC (access_p
, heap
) *access_vec
;
3522 access_vec
= get_base_access_vector (parm
);
3523 gcc_assert (access_vec
);
3524 repr
= VEC_index (access_p
, access_vec
, 0);
3527 repr
->group_representative
= repr
;
3529 access_count
= VEC_length (access_p
, access_vec
);
3530 for (i
= 1; i
< access_count
; i
++)
3532 struct access
*access
= VEC_index (access_p
, access_vec
, i
);
3535 access
->group_representative
= repr
;
3536 access
->next_sibling
= repr
->next_sibling
;
3537 repr
->next_sibling
= access
;
3541 repr
->grp_scalar_ptr
= 1;
3545 /* Return true iff this access precludes IPA-SRA of the parameter it is
3549 access_precludes_ipa_sra_p (struct access
*access
)
3551 /* Avoid issues such as the second simple testcase in PR 42025. The problem
3552 is incompatible assign in a call statement (and possibly even in asm
3553 statements). This can be relaxed by using a new temporary but only for
3554 non-TREE_ADDRESSABLE types and is probably not worth the complexity. (In
3555 intraprocedural SRA we deal with this by keeping the old aggregate around,
3556 something we cannot do in IPA-SRA.) */
3558 && (is_gimple_call (access
->stmt
)
3559 || gimple_code (access
->stmt
) == GIMPLE_ASM
))
3566 /* Sort collected accesses for parameter PARM, identify representatives for
3567 each accessed region and link them together. Return NULL if there are
3568 different but overlapping accesses, return the special ptr value meaning
3569 there are no accesses for this parameter if that is the case and return the
3570 first representative otherwise. Set *RO_GRP if there is a group of accesses
3571 with only read (i.e. no write) accesses. */
3573 static struct access
*
3574 splice_param_accesses (tree parm
, bool *ro_grp
)
3576 int i
, j
, access_count
, group_count
;
3577 int agg_size
, total_size
= 0;
3578 struct access
*access
, *res
, **prev_acc_ptr
= &res
;
3579 VEC (access_p
, heap
) *access_vec
;
3581 access_vec
= get_base_access_vector (parm
);
3583 return &no_accesses_representant
;
3584 access_count
= VEC_length (access_p
, access_vec
);
3586 VEC_qsort (access_p
, access_vec
, compare_access_positions
);
3591 while (i
< access_count
)
3595 access
= VEC_index (access_p
, access_vec
, i
);
3596 modification
= access
->write
;
3597 if (access_precludes_ipa_sra_p (access
))
3599 a1_alias_type
= reference_alias_ptr_type (access
->expr
);
3601 /* Access is about to become group representative unless we find some
3602 nasty overlap which would preclude us from breaking this parameter
3606 while (j
< access_count
)
3608 struct access
*ac2
= VEC_index (access_p
, access_vec
, j
);
3609 if (ac2
->offset
!= access
->offset
)
3611 /* All or nothing law for parameters. */
3612 if (access
->offset
+ access
->size
> ac2
->offset
)
3617 else if (ac2
->size
!= access
->size
)
3620 if (access_precludes_ipa_sra_p (ac2
)
3621 || (ac2
->type
!= access
->type
3622 && (TREE_ADDRESSABLE (ac2
->type
)
3623 || TREE_ADDRESSABLE (access
->type
)))
3624 || (reference_alias_ptr_type (ac2
->expr
) != a1_alias_type
))
3627 modification
|= ac2
->write
;
3628 ac2
->group_representative
= access
;
3629 ac2
->next_sibling
= access
->next_sibling
;
3630 access
->next_sibling
= ac2
;
3635 access
->grp_maybe_modified
= modification
;
3638 *prev_acc_ptr
= access
;
3639 prev_acc_ptr
= &access
->next_grp
;
3640 total_size
+= access
->size
;
3644 if (POINTER_TYPE_P (TREE_TYPE (parm
)))
3645 agg_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm
))), 1);
3647 agg_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (parm
)), 1);
3648 if (total_size
>= agg_size
)
3651 gcc_assert (group_count
> 0);
3655 /* Decide whether parameters with representative accesses given by REPR should
3656 be reduced into components. */
3659 decide_one_param_reduction (struct access
*repr
)
3661 int total_size
, cur_parm_size
, agg_size
, new_param_count
, parm_size_limit
;
3666 cur_parm_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (parm
)), 1);
3667 gcc_assert (cur_parm_size
> 0);
3669 if (POINTER_TYPE_P (TREE_TYPE (parm
)))
3672 agg_size
= tree_low_cst (TYPE_SIZE (TREE_TYPE (TREE_TYPE (parm
))), 1);
3677 agg_size
= cur_parm_size
;
3683 fprintf (dump_file
, "Evaluating PARAM group sizes for ");
3684 print_generic_expr (dump_file
, parm
, 0);
3685 fprintf (dump_file
, " (UID: %u): \n", DECL_UID (parm
));
3686 for (acc
= repr
; acc
; acc
= acc
->next_grp
)
3687 dump_access (dump_file
, acc
, true);
3691 new_param_count
= 0;
3693 for (; repr
; repr
= repr
->next_grp
)
3695 gcc_assert (parm
== repr
->base
);
3697 /* Taking the address of a non-addressable field is verboten. */
3698 if (by_ref
&& repr
->non_addressable
)
3701 if (!by_ref
|| (!repr
->grp_maybe_modified
3702 && !repr
->grp_not_necessarilly_dereferenced
))
3703 total_size
+= repr
->size
;
3705 total_size
+= cur_parm_size
;
3710 gcc_assert (new_param_count
> 0);
3712 if (optimize_function_for_size_p (cfun
))
3713 parm_size_limit
= cur_parm_size
;
3715 parm_size_limit
= (PARAM_VALUE (PARAM_IPA_SRA_PTR_GROWTH_FACTOR
)
3718 if (total_size
< agg_size
3719 && total_size
<= parm_size_limit
)
3722 fprintf (dump_file
, " ....will be split into %i components\n",
3724 return new_param_count
;
3730 /* The order of the following enums is important, we need to do extra work for
3731 UNUSED_PARAMS, BY_VAL_ACCESSES and UNMODIF_BY_REF_ACCESSES. */
3732 enum ipa_splicing_result
{ NO_GOOD_ACCESS
, UNUSED_PARAMS
, BY_VAL_ACCESSES
,
3733 MODIF_BY_REF_ACCESSES
, UNMODIF_BY_REF_ACCESSES
};
3735 /* Identify representatives of all accesses to all candidate parameters for
3736 IPA-SRA. Return result based on what representatives have been found. */
3738 static enum ipa_splicing_result
3739 splice_all_param_accesses (VEC (access_p
, heap
) **representatives
)
3741 enum ipa_splicing_result result
= NO_GOOD_ACCESS
;
3743 struct access
*repr
;
3745 *representatives
= VEC_alloc (access_p
, heap
, func_param_count
);
3747 for (parm
= DECL_ARGUMENTS (current_function_decl
);
3749 parm
= DECL_CHAIN (parm
))
3751 if (is_unused_scalar_param (parm
))
3753 VEC_quick_push (access_p
, *representatives
,
3754 &no_accesses_representant
);
3755 if (result
== NO_GOOD_ACCESS
)
3756 result
= UNUSED_PARAMS
;
3758 else if (POINTER_TYPE_P (TREE_TYPE (parm
))
3759 && is_gimple_reg_type (TREE_TYPE (TREE_TYPE (parm
)))
3760 && bitmap_bit_p (candidate_bitmap
, DECL_UID (parm
)))
3762 repr
= unmodified_by_ref_scalar_representative (parm
);
3763 VEC_quick_push (access_p
, *representatives
, repr
);
3765 result
= UNMODIF_BY_REF_ACCESSES
;
3767 else if (bitmap_bit_p (candidate_bitmap
, DECL_UID (parm
)))
3769 bool ro_grp
= false;
3770 repr
= splice_param_accesses (parm
, &ro_grp
);
3771 VEC_quick_push (access_p
, *representatives
, repr
);
3773 if (repr
&& !no_accesses_p (repr
))
3775 if (POINTER_TYPE_P (TREE_TYPE (parm
)))
3778 result
= UNMODIF_BY_REF_ACCESSES
;
3779 else if (result
< MODIF_BY_REF_ACCESSES
)
3780 result
= MODIF_BY_REF_ACCESSES
;
3782 else if (result
< BY_VAL_ACCESSES
)
3783 result
= BY_VAL_ACCESSES
;
3785 else if (no_accesses_p (repr
) && (result
== NO_GOOD_ACCESS
))
3786 result
= UNUSED_PARAMS
;
3789 VEC_quick_push (access_p
, *representatives
, NULL
);
3792 if (result
== NO_GOOD_ACCESS
)
3794 VEC_free (access_p
, heap
, *representatives
);
3795 *representatives
= NULL
;
3796 return NO_GOOD_ACCESS
;
3802 /* Return the index of BASE in PARMS. Abort if it is not found. */
3805 get_param_index (tree base
, VEC(tree
, heap
) *parms
)
3809 len
= VEC_length (tree
, parms
);
3810 for (i
= 0; i
< len
; i
++)
3811 if (VEC_index (tree
, parms
, i
) == base
)
3816 /* Convert the decisions made at the representative level into compact
3817 parameter adjustments. REPRESENTATIVES are pointers to first
3818 representatives of each param accesses, ADJUSTMENTS_COUNT is the expected
3819 final number of adjustments. */
3821 static ipa_parm_adjustment_vec
3822 turn_representatives_into_adjustments (VEC (access_p
, heap
) *representatives
,
3823 int adjustments_count
)
3825 VEC (tree
, heap
) *parms
;
3826 ipa_parm_adjustment_vec adjustments
;
3830 gcc_assert (adjustments_count
> 0);
3831 parms
= ipa_get_vector_of_formal_parms (current_function_decl
);
3832 adjustments
= VEC_alloc (ipa_parm_adjustment_t
, heap
, adjustments_count
);
3833 parm
= DECL_ARGUMENTS (current_function_decl
);
3834 for (i
= 0; i
< func_param_count
; i
++, parm
= DECL_CHAIN (parm
))
3836 struct access
*repr
= VEC_index (access_p
, representatives
, i
);
3838 if (!repr
|| no_accesses_p (repr
))
3840 struct ipa_parm_adjustment
*adj
;
3842 adj
= VEC_quick_push (ipa_parm_adjustment_t
, adjustments
, NULL
);
3843 memset (adj
, 0, sizeof (*adj
));
3844 adj
->base_index
= get_param_index (parm
, parms
);
3847 adj
->copy_param
= 1;
3849 adj
->remove_param
= 1;
3853 struct ipa_parm_adjustment
*adj
;
3854 int index
= get_param_index (parm
, parms
);
3856 for (; repr
; repr
= repr
->next_grp
)
3858 adj
= VEC_quick_push (ipa_parm_adjustment_t
, adjustments
, NULL
);
3859 memset (adj
, 0, sizeof (*adj
));
3860 gcc_assert (repr
->base
== parm
);
3861 adj
->base_index
= index
;
3862 adj
->base
= repr
->base
;
3863 adj
->type
= repr
->type
;
3864 adj
->alias_ptr_type
= reference_alias_ptr_type (repr
->expr
);
3865 adj
->offset
= repr
->offset
;
3866 adj
->by_ref
= (POINTER_TYPE_P (TREE_TYPE (repr
->base
))
3867 && (repr
->grp_maybe_modified
3868 || repr
->grp_not_necessarilly_dereferenced
));
3873 VEC_free (tree
, heap
, parms
);
3877 /* Analyze the collected accesses and produce a plan what to do with the
3878 parameters in the form of adjustments, NULL meaning nothing. */
3880 static ipa_parm_adjustment_vec
3881 analyze_all_param_acesses (void)
3883 enum ipa_splicing_result repr_state
;
3884 bool proceed
= false;
3885 int i
, adjustments_count
= 0;
3886 VEC (access_p
, heap
) *representatives
;
3887 ipa_parm_adjustment_vec adjustments
;
3889 repr_state
= splice_all_param_accesses (&representatives
);
3890 if (repr_state
== NO_GOOD_ACCESS
)
3893 /* If there are any parameters passed by reference which are not modified
3894 directly, we need to check whether they can be modified indirectly. */
3895 if (repr_state
== UNMODIF_BY_REF_ACCESSES
)
3897 analyze_caller_dereference_legality (representatives
);
3898 analyze_modified_params (representatives
);
3901 for (i
= 0; i
< func_param_count
; i
++)
3903 struct access
*repr
= VEC_index (access_p
, representatives
, i
);
3905 if (repr
&& !no_accesses_p (repr
))
3907 if (repr
->grp_scalar_ptr
)
3909 adjustments_count
++;
3910 if (repr
->grp_not_necessarilly_dereferenced
3911 || repr
->grp_maybe_modified
)
3912 VEC_replace (access_p
, representatives
, i
, NULL
);
3916 sra_stats
.scalar_by_ref_to_by_val
++;
3921 int new_components
= decide_one_param_reduction (repr
);
3923 if (new_components
== 0)
3925 VEC_replace (access_p
, representatives
, i
, NULL
);
3926 adjustments_count
++;
3930 adjustments_count
+= new_components
;
3931 sra_stats
.aggregate_params_reduced
++;
3932 sra_stats
.param_reductions_created
+= new_components
;
3939 if (no_accesses_p (repr
))
3942 sra_stats
.deleted_unused_parameters
++;
3944 adjustments_count
++;
3948 if (!proceed
&& dump_file
)
3949 fprintf (dump_file
, "NOT proceeding to change params.\n");
3952 adjustments
= turn_representatives_into_adjustments (representatives
,
3957 VEC_free (access_p
, heap
, representatives
);
3961 /* If a parameter replacement identified by ADJ does not yet exist in the form
3962 of declaration, create it and record it, otherwise return the previously
3966 get_replaced_param_substitute (struct ipa_parm_adjustment
*adj
)
3969 if (!adj
->new_ssa_base
)
3971 char *pretty_name
= make_fancy_name (adj
->base
);
3973 repl
= create_tmp_reg (TREE_TYPE (adj
->base
), "ISR");
3974 DECL_NAME (repl
) = get_identifier (pretty_name
);
3975 obstack_free (&name_obstack
, pretty_name
);
3978 add_referenced_var (repl
);
3979 adj
->new_ssa_base
= repl
;
3982 repl
= adj
->new_ssa_base
;
3986 /* Find the first adjustment for a particular parameter BASE in a vector of
3987 ADJUSTMENTS which is not a copy_param. Return NULL if there is no such
3990 static struct ipa_parm_adjustment
*
3991 get_adjustment_for_base (ipa_parm_adjustment_vec adjustments
, tree base
)
3995 len
= VEC_length (ipa_parm_adjustment_t
, adjustments
);
3996 for (i
= 0; i
< len
; i
++)
3998 struct ipa_parm_adjustment
*adj
;
4000 adj
= VEC_index (ipa_parm_adjustment_t
, adjustments
, i
);
4001 if (!adj
->copy_param
&& adj
->base
== base
)
4008 /* If the statement STMT defines an SSA_NAME of a parameter which is to be
4009 removed because its value is not used, replace the SSA_NAME with a one
4010 relating to a created VAR_DECL together all of its uses and return true.
4011 ADJUSTMENTS is a pointer to an adjustments vector. */
4014 replace_removed_params_ssa_names (gimple stmt
,
4015 ipa_parm_adjustment_vec adjustments
)
4017 struct ipa_parm_adjustment
*adj
;
4018 tree lhs
, decl
, repl
, name
;
4020 if (gimple_code (stmt
) == GIMPLE_PHI
)
4021 lhs
= gimple_phi_result (stmt
);
4022 else if (is_gimple_assign (stmt
))
4023 lhs
= gimple_assign_lhs (stmt
);
4024 else if (is_gimple_call (stmt
))
4025 lhs
= gimple_call_lhs (stmt
);
4029 if (TREE_CODE (lhs
) != SSA_NAME
)
4031 decl
= SSA_NAME_VAR (lhs
);
4032 if (TREE_CODE (decl
) != PARM_DECL
)
4035 adj
= get_adjustment_for_base (adjustments
, decl
);
4039 repl
= get_replaced_param_substitute (adj
);
4040 name
= make_ssa_name (repl
, stmt
);
4044 fprintf (dump_file
, "replacing an SSA name of a removed param ");
4045 print_generic_expr (dump_file
, lhs
, 0);
4046 fprintf (dump_file
, " with ");
4047 print_generic_expr (dump_file
, name
, 0);
4048 fprintf (dump_file
, "\n");
4051 if (is_gimple_assign (stmt
))
4052 gimple_assign_set_lhs (stmt
, name
);
4053 else if (is_gimple_call (stmt
))
4054 gimple_call_set_lhs (stmt
, name
);
4056 gimple_phi_set_result (stmt
, name
);
4058 replace_uses_by (lhs
, name
);
4059 release_ssa_name (lhs
);
4063 /* If the expression *EXPR should be replaced by a reduction of a parameter, do
4064 so. ADJUSTMENTS is a pointer to a vector of adjustments. CONVERT
4065 specifies whether the function should care about type incompatibility the
4066 current and new expressions. If it is false, the function will leave
4067 incompatibility issues to the caller. Return true iff the expression
4071 sra_ipa_modify_expr (tree
*expr
, bool convert
,
4072 ipa_parm_adjustment_vec adjustments
)
4075 struct ipa_parm_adjustment
*adj
, *cand
= NULL
;
4076 HOST_WIDE_INT offset
, size
, max_size
;
4079 len
= VEC_length (ipa_parm_adjustment_t
, adjustments
);
4081 if (TREE_CODE (*expr
) == BIT_FIELD_REF
4082 || TREE_CODE (*expr
) == IMAGPART_EXPR
4083 || TREE_CODE (*expr
) == REALPART_EXPR
)
4085 expr
= &TREE_OPERAND (*expr
, 0);
4089 base
= get_ref_base_and_extent (*expr
, &offset
, &size
, &max_size
);
4090 if (!base
|| size
== -1 || max_size
== -1)
4093 if (TREE_CODE (base
) == MEM_REF
)
4095 offset
+= mem_ref_offset (base
).low
* BITS_PER_UNIT
;
4096 base
= TREE_OPERAND (base
, 0);
4099 base
= get_ssa_base_param (base
);
4100 if (!base
|| TREE_CODE (base
) != PARM_DECL
)
4103 for (i
= 0; i
< len
; i
++)
4105 adj
= VEC_index (ipa_parm_adjustment_t
, adjustments
, i
);
4107 if (adj
->base
== base
&&
4108 (adj
->offset
== offset
|| adj
->remove_param
))
4114 if (!cand
|| cand
->copy_param
|| cand
->remove_param
)
4118 src
= build_simple_mem_ref (cand
->reduction
);
4120 src
= cand
->reduction
;
4122 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
4124 fprintf (dump_file
, "About to replace expr ");
4125 print_generic_expr (dump_file
, *expr
, 0);
4126 fprintf (dump_file
, " with ");
4127 print_generic_expr (dump_file
, src
, 0);
4128 fprintf (dump_file
, "\n");
4131 if (convert
&& !useless_type_conversion_p (TREE_TYPE (*expr
), cand
->type
))
4133 tree vce
= build1 (VIEW_CONVERT_EXPR
, TREE_TYPE (*expr
), src
);
4141 /* If the statement pointed to by STMT_PTR contains any expressions that need
4142 to replaced with a different one as noted by ADJUSTMENTS, do so. Handle any
4143 potential type incompatibilities (GSI is used to accommodate conversion
4144 statements and must point to the statement). Return true iff the statement
4148 sra_ipa_modify_assign (gimple
*stmt_ptr
, gimple_stmt_iterator
*gsi
,
4149 ipa_parm_adjustment_vec adjustments
)
4151 gimple stmt
= *stmt_ptr
;
4152 tree
*lhs_p
, *rhs_p
;
4155 if (!gimple_assign_single_p (stmt
))
4158 rhs_p
= gimple_assign_rhs1_ptr (stmt
);
4159 lhs_p
= gimple_assign_lhs_ptr (stmt
);
4161 any
= sra_ipa_modify_expr (rhs_p
, false, adjustments
);
4162 any
|= sra_ipa_modify_expr (lhs_p
, false, adjustments
);
4165 tree new_rhs
= NULL_TREE
;
4167 if (!useless_type_conversion_p (TREE_TYPE (*lhs_p
), TREE_TYPE (*rhs_p
)))
4169 if (TREE_CODE (*rhs_p
) == CONSTRUCTOR
)
4171 /* V_C_Es of constructors can cause trouble (PR 42714). */
4172 if (is_gimple_reg_type (TREE_TYPE (*lhs_p
)))
4173 *rhs_p
= build_zero_cst (TREE_TYPE (*lhs_p
));
4175 *rhs_p
= build_constructor (TREE_TYPE (*lhs_p
), 0);
4178 new_rhs
= fold_build1_loc (gimple_location (stmt
),
4179 VIEW_CONVERT_EXPR
, TREE_TYPE (*lhs_p
),
4182 else if (REFERENCE_CLASS_P (*rhs_p
)
4183 && is_gimple_reg_type (TREE_TYPE (*lhs_p
))
4184 && !is_gimple_reg (*lhs_p
))
4185 /* This can happen when an assignment in between two single field
4186 structures is turned into an assignment in between two pointers to
4187 scalars (PR 42237). */
4192 tree tmp
= force_gimple_operand_gsi (gsi
, new_rhs
, true, NULL_TREE
,
4193 true, GSI_SAME_STMT
);
4195 gimple_assign_set_rhs_from_tree (gsi
, tmp
);
4204 /* Traverse the function body and all modifications as described in
4205 ADJUSTMENTS. Return true iff the CFG has been changed. */
4208 ipa_sra_modify_function_body (ipa_parm_adjustment_vec adjustments
)
4210 bool cfg_changed
= false;
4215 gimple_stmt_iterator gsi
;
4217 for (gsi
= gsi_start_phis (bb
); !gsi_end_p (gsi
); gsi_next (&gsi
))
4218 replace_removed_params_ssa_names (gsi_stmt (gsi
), adjustments
);
4220 gsi
= gsi_start_bb (bb
);
4221 while (!gsi_end_p (gsi
))
4223 gimple stmt
= gsi_stmt (gsi
);
4224 bool modified
= false;
4228 switch (gimple_code (stmt
))
4231 t
= gimple_return_retval_ptr (stmt
);
4232 if (*t
!= NULL_TREE
)
4233 modified
|= sra_ipa_modify_expr (t
, true, adjustments
);
4237 modified
|= sra_ipa_modify_assign (&stmt
, &gsi
, adjustments
);
4238 modified
|= replace_removed_params_ssa_names (stmt
, adjustments
);
4242 /* Operands must be processed before the lhs. */
4243 for (i
= 0; i
< gimple_call_num_args (stmt
); i
++)
4245 t
= gimple_call_arg_ptr (stmt
, i
);
4246 modified
|= sra_ipa_modify_expr (t
, true, adjustments
);
4249 if (gimple_call_lhs (stmt
))
4251 t
= gimple_call_lhs_ptr (stmt
);
4252 modified
|= sra_ipa_modify_expr (t
, false, adjustments
);
4253 modified
|= replace_removed_params_ssa_names (stmt
,
4259 for (i
= 0; i
< gimple_asm_ninputs (stmt
); i
++)
4261 t
= &TREE_VALUE (gimple_asm_input_op (stmt
, i
));
4262 modified
|= sra_ipa_modify_expr (t
, true, adjustments
);
4264 for (i
= 0; i
< gimple_asm_noutputs (stmt
); i
++)
4266 t
= &TREE_VALUE (gimple_asm_output_op (stmt
, i
));
4267 modified
|= sra_ipa_modify_expr (t
, false, adjustments
);
4278 if (maybe_clean_eh_stmt (stmt
)
4279 && gimple_purge_dead_eh_edges (gimple_bb (stmt
)))
4289 /* Call gimple_debug_bind_reset_value on all debug statements describing
4290 gimple register parameters that are being removed or replaced. */
4293 sra_ipa_reset_debug_stmts (ipa_parm_adjustment_vec adjustments
)
4297 len
= VEC_length (ipa_parm_adjustment_t
, adjustments
);
4298 for (i
= 0; i
< len
; i
++)
4300 struct ipa_parm_adjustment
*adj
;
4301 imm_use_iterator ui
;
4305 adj
= VEC_index (ipa_parm_adjustment_t
, adjustments
, i
);
4306 if (adj
->copy_param
|| !is_gimple_reg (adj
->base
))
4308 name
= gimple_default_def (cfun
, adj
->base
);
4311 FOR_EACH_IMM_USE_STMT (stmt
, ui
, name
)
4313 /* All other users must have been removed by
4314 ipa_sra_modify_function_body. */
4315 gcc_assert (is_gimple_debug (stmt
));
4316 gimple_debug_bind_reset_value (stmt
);
4322 /* Return true iff all callers have at least as many actual arguments as there
4323 are formal parameters in the current function. */
4326 all_callers_have_enough_arguments_p (struct cgraph_node
*node
)
4328 struct cgraph_edge
*cs
;
4329 for (cs
= node
->callers
; cs
; cs
= cs
->next_caller
)
4330 if (!callsite_has_enough_arguments_p (cs
->call_stmt
))
4337 /* Convert all callers of NODE to pass parameters as given in ADJUSTMENTS. */
4340 convert_callers (struct cgraph_node
*node
, tree old_decl
,
4341 ipa_parm_adjustment_vec adjustments
)
4343 tree old_cur_fndecl
= current_function_decl
;
4344 struct cgraph_edge
*cs
;
4345 basic_block this_block
;
4346 bitmap recomputed_callers
= BITMAP_ALLOC (NULL
);
4348 for (cs
= node
->callers
; cs
; cs
= cs
->next_caller
)
4350 current_function_decl
= cs
->caller
->decl
;
4351 push_cfun (DECL_STRUCT_FUNCTION (cs
->caller
->decl
));
4354 fprintf (dump_file
, "Adjusting call (%i -> %i) %s -> %s\n",
4355 cs
->caller
->uid
, cs
->callee
->uid
,
4356 cgraph_node_name (cs
->caller
),
4357 cgraph_node_name (cs
->callee
));
4359 ipa_modify_call_arguments (cs
, cs
->call_stmt
, adjustments
);
4364 for (cs
= node
->callers
; cs
; cs
= cs
->next_caller
)
4365 if (bitmap_set_bit (recomputed_callers
, cs
->caller
->uid
))
4366 compute_inline_parameters (cs
->caller
);
4367 BITMAP_FREE (recomputed_callers
);
4369 current_function_decl
= old_cur_fndecl
;
4371 if (!encountered_recursive_call
)
4374 FOR_EACH_BB (this_block
)
4376 gimple_stmt_iterator gsi
;
4378 for (gsi
= gsi_start_bb (this_block
); !gsi_end_p (gsi
); gsi_next (&gsi
))
4380 gimple stmt
= gsi_stmt (gsi
);
4382 if (gimple_code (stmt
) != GIMPLE_CALL
)
4384 call_fndecl
= gimple_call_fndecl (stmt
);
4385 if (call_fndecl
== old_decl
)
4388 fprintf (dump_file
, "Adjusting recursive call");
4389 gimple_call_set_fndecl (stmt
, node
->decl
);
4390 ipa_modify_call_arguments (NULL
, stmt
, adjustments
);
4398 /* Perform all the modification required in IPA-SRA for NODE to have parameters
4399 as given in ADJUSTMENTS. Return true iff the CFG has been changed. */
4402 modify_function (struct cgraph_node
*node
, ipa_parm_adjustment_vec adjustments
)
4404 struct cgraph_node
*new_node
;
4405 struct cgraph_edge
*cs
;
4407 VEC (cgraph_edge_p
, heap
) * redirect_callers
;
4411 for (cs
= node
->callers
; cs
!= NULL
; cs
= cs
->next_caller
)
4413 redirect_callers
= VEC_alloc (cgraph_edge_p
, heap
, node_callers
);
4414 for (cs
= node
->callers
; cs
!= NULL
; cs
= cs
->next_caller
)
4415 VEC_quick_push (cgraph_edge_p
, redirect_callers
, cs
);
4417 rebuild_cgraph_edges ();
4419 current_function_decl
= NULL_TREE
;
4421 new_node
= cgraph_function_versioning (node
, redirect_callers
, NULL
, NULL
,
4422 NULL
, NULL
, "isra");
4423 current_function_decl
= new_node
->decl
;
4424 push_cfun (DECL_STRUCT_FUNCTION (new_node
->decl
));
4426 ipa_modify_formal_parameters (current_function_decl
, adjustments
, "ISRA");
4427 cfg_changed
= ipa_sra_modify_function_body (adjustments
);
4428 sra_ipa_reset_debug_stmts (adjustments
);
4429 convert_callers (new_node
, node
->decl
, adjustments
);
4430 cgraph_make_node_local (new_node
);
4434 /* Return false the function is apparently unsuitable for IPA-SRA based on it's
4435 attributes, return true otherwise. NODE is the cgraph node of the current
4439 ipa_sra_preliminary_function_checks (struct cgraph_node
*node
)
4441 if (!cgraph_node_can_be_local_p (node
))
4444 fprintf (dump_file
, "Function not local to this compilation unit.\n");
4448 if (!node
->local
.can_change_signature
)
4451 fprintf (dump_file
, "Function can not change signature.\n");
4455 if (!tree_versionable_function_p (node
->decl
))
4458 fprintf (dump_file
, "Function is not versionable.\n");
4462 if (DECL_VIRTUAL_P (current_function_decl
))
4465 fprintf (dump_file
, "Function is a virtual method.\n");
4469 if ((DECL_COMDAT (node
->decl
) || DECL_EXTERNAL (node
->decl
))
4470 && node
->global
.size
>= MAX_INLINE_INSNS_AUTO
)
4473 fprintf (dump_file
, "Function too big to be made truly local.\n");
4481 "Function has no callers in this compilation unit.\n");
4488 fprintf (dump_file
, "Function uses stdarg. \n");
4492 if (TYPE_ATTRIBUTES (TREE_TYPE (node
->decl
)))
4498 /* Perform early interprocedural SRA. */
4501 ipa_early_sra (void)
4503 struct cgraph_node
*node
= cgraph_node (current_function_decl
);
4504 ipa_parm_adjustment_vec adjustments
;
4507 if (!ipa_sra_preliminary_function_checks (node
))
4511 sra_mode
= SRA_MODE_EARLY_IPA
;
4513 if (!find_param_candidates ())
4516 fprintf (dump_file
, "Function has no IPA-SRA candidates.\n");
4520 if (!all_callers_have_enough_arguments_p (node
))
4523 fprintf (dump_file
, "There are callers with insufficient number of "
4528 bb_dereferences
= XCNEWVEC (HOST_WIDE_INT
,
4530 * last_basic_block_for_function (cfun
));
4531 final_bbs
= BITMAP_ALLOC (NULL
);
4534 if (encountered_apply_args
)
4537 fprintf (dump_file
, "Function calls __builtin_apply_args().\n");
4541 if (encountered_unchangable_recursive_call
)
4544 fprintf (dump_file
, "Function calls itself with insufficient "
4545 "number of arguments.\n");
4549 adjustments
= analyze_all_param_acesses ();
4553 ipa_dump_param_adjustments (dump_file
, adjustments
, current_function_decl
);
4555 if (modify_function (node
, adjustments
))
4556 ret
= TODO_update_ssa
| TODO_cleanup_cfg
;
4558 ret
= TODO_update_ssa
;
4559 VEC_free (ipa_parm_adjustment_t
, heap
, adjustments
);
4561 statistics_counter_event (cfun
, "Unused parameters deleted",
4562 sra_stats
.deleted_unused_parameters
);
4563 statistics_counter_event (cfun
, "Scalar parameters converted to by-value",
4564 sra_stats
.scalar_by_ref_to_by_val
);
4565 statistics_counter_event (cfun
, "Aggregate parameters broken up",
4566 sra_stats
.aggregate_params_reduced
);
4567 statistics_counter_event (cfun
, "Aggregate parameter components created",
4568 sra_stats
.param_reductions_created
);
4571 BITMAP_FREE (final_bbs
);
4572 free (bb_dereferences
);
4574 sra_deinitialize ();
4578 /* Return if early ipa sra shall be performed. */
4580 ipa_early_sra_gate (void)
4582 return flag_ipa_sra
&& dbg_cnt (eipa_sra
);
4585 struct gimple_opt_pass pass_early_ipa_sra
=
4589 "eipa_sra", /* name */
4590 ipa_early_sra_gate
, /* gate */
4591 ipa_early_sra
, /* execute */
4594 0, /* static_pass_number */
4595 TV_IPA_SRA
, /* tv_id */
4596 0, /* properties_required */
4597 0, /* properties_provided */
4598 0, /* properties_destroyed */
4599 0, /* todo_flags_start */
4600 TODO_dump_func
| TODO_dump_cgraph
/* todo_flags_finish */