Fix typo in t-dimode
[official-gcc.git] / gcc / tree-ssa-pre.c
blob0669aaaac473c16084740029a0dd56391f485e88
1 /* Full and partial redundancy elimination and code hoisting on SSA GIMPLE.
2 Copyright (C) 2001-2021 Free Software Foundation, Inc.
3 Contributed by Daniel Berlin <dan@dberlin.org> and Steven Bosscher
4 <stevenb@suse.de>
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3, or (at your option)
11 any later version.
13 GCC is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
22 #include "config.h"
23 #include "system.h"
24 #include "coretypes.h"
25 #include "backend.h"
26 #include "rtl.h"
27 #include "tree.h"
28 #include "gimple.h"
29 #include "predict.h"
30 #include "alloc-pool.h"
31 #include "tree-pass.h"
32 #include "ssa.h"
33 #include "cgraph.h"
34 #include "gimple-pretty-print.h"
35 #include "fold-const.h"
36 #include "cfganal.h"
37 #include "gimple-fold.h"
38 #include "tree-eh.h"
39 #include "gimplify.h"
40 #include "gimple-iterator.h"
41 #include "tree-cfg.h"
42 #include "tree-into-ssa.h"
43 #include "tree-dfa.h"
44 #include "tree-ssa.h"
45 #include "cfgloop.h"
46 #include "tree-ssa-sccvn.h"
47 #include "tree-scalar-evolution.h"
48 #include "dbgcnt.h"
49 #include "domwalk.h"
50 #include "tree-ssa-propagate.h"
51 #include "tree-ssa-dce.h"
52 #include "tree-cfgcleanup.h"
53 #include "alias.h"
54 #include "gimple-range.h"
56 /* Even though this file is called tree-ssa-pre.c, we actually
57 implement a bit more than just PRE here. All of them piggy-back
58 on GVN which is implemented in tree-ssa-sccvn.c.
60 1. Full Redundancy Elimination (FRE)
61 This is the elimination phase of GVN.
63 2. Partial Redundancy Elimination (PRE)
64 This is adds computation of AVAIL_OUT and ANTIC_IN and
65 doing expression insertion to form GVN-PRE.
67 3. Code hoisting
68 This optimization uses the ANTIC_IN sets computed for PRE
69 to move expressions further up than PRE would do, to make
70 multiple computations of the same value fully redundant.
71 This pass is explained below (after the explanation of the
72 basic algorithm for PRE).
75 /* TODO:
77 1. Avail sets can be shared by making an avail_find_leader that
78 walks up the dominator tree and looks in those avail sets.
79 This might affect code optimality, it's unclear right now.
80 Currently the AVAIL_OUT sets are the remaining quadraticness in
81 memory of GVN-PRE.
82 2. Strength reduction can be performed by anticipating expressions
83 we can repair later on.
84 3. We can do back-substitution or smarter value numbering to catch
85 commutative expressions split up over multiple statements.
88 /* For ease of terminology, "expression node" in the below refers to
89 every expression node but GIMPLE_ASSIGN, because GIMPLE_ASSIGNs
90 represent the actual statement containing the expressions we care about,
91 and we cache the value number by putting it in the expression. */
93 /* Basic algorithm for Partial Redundancy Elimination:
95 First we walk the statements to generate the AVAIL sets, the
96 EXP_GEN sets, and the tmp_gen sets. EXP_GEN sets represent the
97 generation of values/expressions by a given block. We use them
98 when computing the ANTIC sets. The AVAIL sets consist of
99 SSA_NAME's that represent values, so we know what values are
100 available in what blocks. AVAIL is a forward dataflow problem. In
101 SSA, values are never killed, so we don't need a kill set, or a
102 fixpoint iteration, in order to calculate the AVAIL sets. In
103 traditional parlance, AVAIL sets tell us the downsafety of the
104 expressions/values.
106 Next, we generate the ANTIC sets. These sets represent the
107 anticipatable expressions. ANTIC is a backwards dataflow
108 problem. An expression is anticipatable in a given block if it could
109 be generated in that block. This means that if we had to perform
110 an insertion in that block, of the value of that expression, we
111 could. Calculating the ANTIC sets requires phi translation of
112 expressions, because the flow goes backwards through phis. We must
113 iterate to a fixpoint of the ANTIC sets, because we have a kill
114 set. Even in SSA form, values are not live over the entire
115 function, only from their definition point onwards. So we have to
116 remove values from the ANTIC set once we go past the definition
117 point of the leaders that make them up.
118 compute_antic/compute_antic_aux performs this computation.
120 Third, we perform insertions to make partially redundant
121 expressions fully redundant.
123 An expression is partially redundant (excluding partial
124 anticipation) if:
126 1. It is AVAIL in some, but not all, of the predecessors of a
127 given block.
128 2. It is ANTIC in all the predecessors.
130 In order to make it fully redundant, we insert the expression into
131 the predecessors where it is not available, but is ANTIC.
133 When optimizing for size, we only eliminate the partial redundancy
134 if we need to insert in only one predecessor. This avoids almost
135 completely the code size increase that PRE usually causes.
137 For the partial anticipation case, we only perform insertion if it
138 is partially anticipated in some block, and fully available in all
139 of the predecessors.
141 do_pre_regular_insertion/do_pre_partial_partial_insertion
142 performs these steps, driven by insert/insert_aux.
144 Fourth, we eliminate fully redundant expressions.
145 This is a simple statement walk that replaces redundant
146 calculations with the now available values. */
148 /* Basic algorithm for Code Hoisting:
150 Code hoisting is: Moving value computations up in the control flow
151 graph to make multiple copies redundant. Typically this is a size
152 optimization, but there are cases where it also is helpful for speed.
154 A simple code hoisting algorithm is implemented that piggy-backs on
155 the PRE infrastructure. For code hoisting, we have to know ANTIC_OUT
156 which is effectively ANTIC_IN - AVAIL_OUT. The latter two have to be
157 computed for PRE, and we can use them to perform a limited version of
158 code hoisting, too.
160 For the purpose of this implementation, a value is hoistable to a basic
161 block B if the following properties are met:
163 1. The value is in ANTIC_IN(B) -- the value will be computed on all
164 paths from B to function exit and it can be computed in B);
166 2. The value is not in AVAIL_OUT(B) -- there would be no need to
167 compute the value again and make it available twice;
169 3. All successors of B are dominated by B -- makes sure that inserting
170 a computation of the value in B will make the remaining
171 computations fully redundant;
173 4. At least one successor has the value in AVAIL_OUT -- to avoid
174 hoisting values up too far;
176 5. There are at least two successors of B -- hoisting in straight
177 line code is pointless.
179 The third condition is not strictly necessary, but it would complicate
180 the hoisting pass a lot. In fact, I don't know of any code hoisting
181 algorithm that does not have this requirement. Fortunately, experiments
182 have show that most candidate hoistable values are in regions that meet
183 this condition (e.g. diamond-shape regions).
185 The forth condition is necessary to avoid hoisting things up too far
186 away from the uses of the value. Nothing else limits the algorithm
187 from hoisting everything up as far as ANTIC_IN allows. Experiments
188 with SPEC and CSiBE have shown that hoisting up too far results in more
189 spilling, less benefits for code size, and worse benchmark scores.
190 Fortunately, in practice most of the interesting hoisting opportunities
191 are caught despite this limitation.
193 For hoistable values that meet all conditions, expressions are inserted
194 to make the calculation of the hoistable value fully redundant. We
195 perform code hoisting insertions after each round of PRE insertions,
196 because code hoisting never exposes new PRE opportunities, but PRE can
197 create new code hoisting opportunities.
199 The code hoisting algorithm is implemented in do_hoist_insert, driven
200 by insert/insert_aux. */
202 /* Representations of value numbers:
204 Value numbers are represented by a representative SSA_NAME. We
205 will create fake SSA_NAME's in situations where we need a
206 representative but do not have one (because it is a complex
207 expression). In order to facilitate storing the value numbers in
208 bitmaps, and keep the number of wasted SSA_NAME's down, we also
209 associate a value_id with each value number, and create full blown
210 ssa_name's only where we actually need them (IE in operands of
211 existing expressions).
213 Theoretically you could replace all the value_id's with
214 SSA_NAME_VERSION, but this would allocate a large number of
215 SSA_NAME's (which are each > 30 bytes) just to get a 4 byte number.
216 It would also require an additional indirection at each point we
217 use the value id. */
219 /* Representation of expressions on value numbers:
221 Expressions consisting of value numbers are represented the same
222 way as our VN internally represents them, with an additional
223 "pre_expr" wrapping around them in order to facilitate storing all
224 of the expressions in the same sets. */
226 /* Representation of sets:
228 The dataflow sets do not need to be sorted in any particular order
229 for the majority of their lifetime, are simply represented as two
230 bitmaps, one that keeps track of values present in the set, and one
231 that keeps track of expressions present in the set.
233 When we need them in topological order, we produce it on demand by
234 transforming the bitmap into an array and sorting it into topo
235 order. */
237 /* Type of expression, used to know which member of the PRE_EXPR union
238 is valid. */
240 enum pre_expr_kind
242 NAME,
243 NARY,
244 REFERENCE,
245 CONSTANT
248 union pre_expr_union
250 tree name;
251 tree constant;
252 vn_nary_op_t nary;
253 vn_reference_t reference;
256 typedef struct pre_expr_d : nofree_ptr_hash <pre_expr_d>
258 enum pre_expr_kind kind;
259 unsigned int id;
260 unsigned value_id;
261 location_t loc;
262 pre_expr_union u;
264 /* hash_table support. */
265 static inline hashval_t hash (const pre_expr_d *);
266 static inline int equal (const pre_expr_d *, const pre_expr_d *);
267 } *pre_expr;
269 #define PRE_EXPR_NAME(e) (e)->u.name
270 #define PRE_EXPR_NARY(e) (e)->u.nary
271 #define PRE_EXPR_REFERENCE(e) (e)->u.reference
272 #define PRE_EXPR_CONSTANT(e) (e)->u.constant
274 /* Compare E1 and E1 for equality. */
276 inline int
277 pre_expr_d::equal (const pre_expr_d *e1, const pre_expr_d *e2)
279 if (e1->kind != e2->kind)
280 return false;
282 switch (e1->kind)
284 case CONSTANT:
285 return vn_constant_eq_with_type (PRE_EXPR_CONSTANT (e1),
286 PRE_EXPR_CONSTANT (e2));
287 case NAME:
288 return PRE_EXPR_NAME (e1) == PRE_EXPR_NAME (e2);
289 case NARY:
290 return vn_nary_op_eq (PRE_EXPR_NARY (e1), PRE_EXPR_NARY (e2));
291 case REFERENCE:
292 return vn_reference_eq (PRE_EXPR_REFERENCE (e1),
293 PRE_EXPR_REFERENCE (e2));
294 default:
295 gcc_unreachable ();
299 /* Hash E. */
301 inline hashval_t
302 pre_expr_d::hash (const pre_expr_d *e)
304 switch (e->kind)
306 case CONSTANT:
307 return vn_hash_constant_with_type (PRE_EXPR_CONSTANT (e));
308 case NAME:
309 return SSA_NAME_VERSION (PRE_EXPR_NAME (e));
310 case NARY:
311 return PRE_EXPR_NARY (e)->hashcode;
312 case REFERENCE:
313 return PRE_EXPR_REFERENCE (e)->hashcode;
314 default:
315 gcc_unreachable ();
319 /* Next global expression id number. */
320 static unsigned int next_expression_id;
322 /* Mapping from expression to id number we can use in bitmap sets. */
323 static vec<pre_expr> expressions;
324 static hash_table<pre_expr_d> *expression_to_id;
325 static vec<unsigned> name_to_id;
327 /* Allocate an expression id for EXPR. */
329 static inline unsigned int
330 alloc_expression_id (pre_expr expr)
332 struct pre_expr_d **slot;
333 /* Make sure we won't overflow. */
334 gcc_assert (next_expression_id + 1 > next_expression_id);
335 expr->id = next_expression_id++;
336 expressions.safe_push (expr);
337 if (expr->kind == NAME)
339 unsigned version = SSA_NAME_VERSION (PRE_EXPR_NAME (expr));
340 /* vec::safe_grow_cleared allocates no headroom. Avoid frequent
341 re-allocations by using vec::reserve upfront. */
342 unsigned old_len = name_to_id.length ();
343 name_to_id.reserve (num_ssa_names - old_len);
344 name_to_id.quick_grow_cleared (num_ssa_names);
345 gcc_assert (name_to_id[version] == 0);
346 name_to_id[version] = expr->id;
348 else
350 slot = expression_to_id->find_slot (expr, INSERT);
351 gcc_assert (!*slot);
352 *slot = expr;
354 return next_expression_id - 1;
357 /* Return the expression id for tree EXPR. */
359 static inline unsigned int
360 get_expression_id (const pre_expr expr)
362 return expr->id;
365 static inline unsigned int
366 lookup_expression_id (const pre_expr expr)
368 struct pre_expr_d **slot;
370 if (expr->kind == NAME)
372 unsigned version = SSA_NAME_VERSION (PRE_EXPR_NAME (expr));
373 if (name_to_id.length () <= version)
374 return 0;
375 return name_to_id[version];
377 else
379 slot = expression_to_id->find_slot (expr, NO_INSERT);
380 if (!slot)
381 return 0;
382 return ((pre_expr)*slot)->id;
386 /* Return the existing expression id for EXPR, or create one if one
387 does not exist yet. */
389 static inline unsigned int
390 get_or_alloc_expression_id (pre_expr expr)
392 unsigned int id = lookup_expression_id (expr);
393 if (id == 0)
394 return alloc_expression_id (expr);
395 return expr->id = id;
398 /* Return the expression that has expression id ID */
400 static inline pre_expr
401 expression_for_id (unsigned int id)
403 return expressions[id];
406 static object_allocator<pre_expr_d> pre_expr_pool ("pre_expr nodes");
408 /* Given an SSA_NAME NAME, get or create a pre_expr to represent it. */
410 static pre_expr
411 get_or_alloc_expr_for_name (tree name)
413 struct pre_expr_d expr;
414 pre_expr result;
415 unsigned int result_id;
417 expr.kind = NAME;
418 expr.id = 0;
419 PRE_EXPR_NAME (&expr) = name;
420 result_id = lookup_expression_id (&expr);
421 if (result_id != 0)
422 return expression_for_id (result_id);
424 result = pre_expr_pool.allocate ();
425 result->kind = NAME;
426 result->loc = UNKNOWN_LOCATION;
427 result->value_id = VN_INFO (name)->value_id;
428 PRE_EXPR_NAME (result) = name;
429 alloc_expression_id (result);
430 return result;
433 /* Given an NARY, get or create a pre_expr to represent it. */
435 static pre_expr
436 get_or_alloc_expr_for_nary (vn_nary_op_t nary,
437 location_t loc = UNKNOWN_LOCATION)
439 struct pre_expr_d expr;
440 pre_expr result;
441 unsigned int result_id;
443 expr.kind = NARY;
444 expr.id = 0;
445 PRE_EXPR_NARY (&expr) = nary;
446 result_id = lookup_expression_id (&expr);
447 if (result_id != 0)
448 return expression_for_id (result_id);
450 result = pre_expr_pool.allocate ();
451 result->kind = NARY;
452 result->loc = loc;
453 result->value_id = nary->value_id;
454 PRE_EXPR_NARY (result) = nary;
455 alloc_expression_id (result);
456 return result;
459 /* Given an REFERENCE, get or create a pre_expr to represent it. */
461 static pre_expr
462 get_or_alloc_expr_for_reference (vn_reference_t reference,
463 location_t loc = UNKNOWN_LOCATION)
465 struct pre_expr_d expr;
466 pre_expr result;
467 unsigned int result_id;
469 expr.kind = REFERENCE;
470 expr.id = 0;
471 PRE_EXPR_REFERENCE (&expr) = reference;
472 result_id = lookup_expression_id (&expr);
473 if (result_id != 0)
474 return expression_for_id (result_id);
476 result = pre_expr_pool.allocate ();
477 result->kind = REFERENCE;
478 result->loc = loc;
479 result->value_id = reference->value_id;
480 PRE_EXPR_REFERENCE (result) = reference;
481 alloc_expression_id (result);
482 return result;
486 /* An unordered bitmap set. One bitmap tracks values, the other,
487 expressions. */
488 typedef class bitmap_set
490 public:
491 bitmap_head expressions;
492 bitmap_head values;
493 } *bitmap_set_t;
495 #define FOR_EACH_EXPR_ID_IN_SET(set, id, bi) \
496 EXECUTE_IF_SET_IN_BITMAP (&(set)->expressions, 0, (id), (bi))
498 #define FOR_EACH_VALUE_ID_IN_SET(set, id, bi) \
499 EXECUTE_IF_SET_IN_BITMAP (&(set)->values, 0, (id), (bi))
501 /* Mapping from value id to expressions with that value_id. */
502 static vec<bitmap> value_expressions;
503 /* We just record a single expression for each constant value,
504 one of kind CONSTANT. */
505 static vec<pre_expr> constant_value_expressions;
508 /* This structure is used to keep track of statistics on what
509 optimization PRE was able to perform. */
510 static struct
512 /* The number of new expressions/temporaries generated by PRE. */
513 int insertions;
515 /* The number of inserts found due to partial anticipation */
516 int pa_insert;
518 /* The number of inserts made for code hoisting. */
519 int hoist_insert;
521 /* The number of new PHI nodes added by PRE. */
522 int phis;
523 } pre_stats;
525 static bool do_partial_partial;
526 static pre_expr bitmap_find_leader (bitmap_set_t, unsigned int);
527 static void bitmap_value_insert_into_set (bitmap_set_t, pre_expr);
528 static bool bitmap_value_replace_in_set (bitmap_set_t, pre_expr);
529 static void bitmap_set_copy (bitmap_set_t, bitmap_set_t);
530 static bool bitmap_set_contains_value (bitmap_set_t, unsigned int);
531 static void bitmap_insert_into_set (bitmap_set_t, pre_expr);
532 static bitmap_set_t bitmap_set_new (void);
533 static tree create_expression_by_pieces (basic_block, pre_expr, gimple_seq *,
534 tree);
535 static tree find_or_generate_expression (basic_block, tree, gimple_seq *);
536 static unsigned int get_expr_value_id (pre_expr);
538 /* We can add and remove elements and entries to and from sets
539 and hash tables, so we use alloc pools for them. */
541 static object_allocator<bitmap_set> bitmap_set_pool ("Bitmap sets");
542 static bitmap_obstack grand_bitmap_obstack;
544 /* A three tuple {e, pred, v} used to cache phi translations in the
545 phi_translate_table. */
547 typedef struct expr_pred_trans_d : public typed_noop_remove <expr_pred_trans_d>
549 typedef expr_pred_trans_d value_type;
550 typedef expr_pred_trans_d compare_type;
552 /* The expression ID. */
553 unsigned e;
555 /* The value expression ID that resulted from the translation. */
556 unsigned v;
558 /* hash_table support. */
559 static inline void mark_empty (expr_pred_trans_d &);
560 static inline bool is_empty (const expr_pred_trans_d &);
561 static inline void mark_deleted (expr_pred_trans_d &);
562 static inline bool is_deleted (const expr_pred_trans_d &);
563 static const bool empty_zero_p = true;
564 static inline hashval_t hash (const expr_pred_trans_d &);
565 static inline int equal (const expr_pred_trans_d &, const expr_pred_trans_d &);
566 } *expr_pred_trans_t;
567 typedef const struct expr_pred_trans_d *const_expr_pred_trans_t;
569 inline bool
570 expr_pred_trans_d::is_empty (const expr_pred_trans_d &e)
572 return e.e == 0;
575 inline bool
576 expr_pred_trans_d::is_deleted (const expr_pred_trans_d &e)
578 return e.e == -1u;
581 inline void
582 expr_pred_trans_d::mark_empty (expr_pred_trans_d &e)
584 e.e = 0;
587 inline void
588 expr_pred_trans_d::mark_deleted (expr_pred_trans_d &e)
590 e.e = -1u;
593 inline hashval_t
594 expr_pred_trans_d::hash (const expr_pred_trans_d &e)
596 return e.e;
599 inline int
600 expr_pred_trans_d::equal (const expr_pred_trans_d &ve1,
601 const expr_pred_trans_d &ve2)
603 return ve1.e == ve2.e;
606 /* Sets that we need to keep track of. */
607 typedef struct bb_bitmap_sets
609 /* The EXP_GEN set, which represents expressions/values generated in
610 a basic block. */
611 bitmap_set_t exp_gen;
613 /* The PHI_GEN set, which represents PHI results generated in a
614 basic block. */
615 bitmap_set_t phi_gen;
617 /* The TMP_GEN set, which represents results/temporaries generated
618 in a basic block. IE the LHS of an expression. */
619 bitmap_set_t tmp_gen;
621 /* The AVAIL_OUT set, which represents which values are available in
622 a given basic block. */
623 bitmap_set_t avail_out;
625 /* The ANTIC_IN set, which represents which values are anticipatable
626 in a given basic block. */
627 bitmap_set_t antic_in;
629 /* The PA_IN set, which represents which values are
630 partially anticipatable in a given basic block. */
631 bitmap_set_t pa_in;
633 /* The NEW_SETS set, which is used during insertion to augment the
634 AVAIL_OUT set of blocks with the new insertions performed during
635 the current iteration. */
636 bitmap_set_t new_sets;
638 /* A cache for value_dies_in_block_x. */
639 bitmap expr_dies;
641 /* The live virtual operand on successor edges. */
642 tree vop_on_exit;
644 /* PHI translate cache for the single successor edge. */
645 hash_table<expr_pred_trans_d> *phi_translate_table;
647 /* True if we have visited this block during ANTIC calculation. */
648 unsigned int visited : 1;
650 /* True when the block contains a call that might not return. */
651 unsigned int contains_may_not_return_call : 1;
652 } *bb_value_sets_t;
654 #define EXP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->exp_gen
655 #define PHI_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->phi_gen
656 #define TMP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->tmp_gen
657 #define AVAIL_OUT(BB) ((bb_value_sets_t) ((BB)->aux))->avail_out
658 #define ANTIC_IN(BB) ((bb_value_sets_t) ((BB)->aux))->antic_in
659 #define PA_IN(BB) ((bb_value_sets_t) ((BB)->aux))->pa_in
660 #define NEW_SETS(BB) ((bb_value_sets_t) ((BB)->aux))->new_sets
661 #define EXPR_DIES(BB) ((bb_value_sets_t) ((BB)->aux))->expr_dies
662 #define PHI_TRANS_TABLE(BB) ((bb_value_sets_t) ((BB)->aux))->phi_translate_table
663 #define BB_VISITED(BB) ((bb_value_sets_t) ((BB)->aux))->visited
664 #define BB_MAY_NOTRETURN(BB) ((bb_value_sets_t) ((BB)->aux))->contains_may_not_return_call
665 #define BB_LIVE_VOP_ON_EXIT(BB) ((bb_value_sets_t) ((BB)->aux))->vop_on_exit
668 /* Add the tuple mapping from {expression E, basic block PRED} to
669 the phi translation table and return whether it pre-existed. */
671 static inline bool
672 phi_trans_add (expr_pred_trans_t *entry, pre_expr e, basic_block pred)
674 if (!PHI_TRANS_TABLE (pred))
675 PHI_TRANS_TABLE (pred) = new hash_table<expr_pred_trans_d> (11);
677 expr_pred_trans_t slot;
678 expr_pred_trans_d tem;
679 unsigned id = get_expression_id (e);
680 tem.e = id;
681 slot = PHI_TRANS_TABLE (pred)->find_slot_with_hash (tem, id, INSERT);
682 if (slot->e)
684 *entry = slot;
685 return true;
688 *entry = slot;
689 slot->e = id;
690 return false;
694 /* Add expression E to the expression set of value id V. */
696 static void
697 add_to_value (unsigned int v, pre_expr e)
699 gcc_checking_assert (get_expr_value_id (e) == v);
701 if (value_id_constant_p (v))
703 if (e->kind != CONSTANT)
704 return;
706 if (-v >= constant_value_expressions.length ())
707 constant_value_expressions.safe_grow_cleared (-v + 1);
709 pre_expr leader = constant_value_expressions[-v];
710 if (!leader)
711 constant_value_expressions[-v] = e;
713 else
715 if (v >= value_expressions.length ())
716 value_expressions.safe_grow_cleared (v + 1);
718 bitmap set = value_expressions[v];
719 if (!set)
721 set = BITMAP_ALLOC (&grand_bitmap_obstack);
722 value_expressions[v] = set;
724 bitmap_set_bit (set, get_or_alloc_expression_id (e));
728 /* Create a new bitmap set and return it. */
730 static bitmap_set_t
731 bitmap_set_new (void)
733 bitmap_set_t ret = bitmap_set_pool.allocate ();
734 bitmap_initialize (&ret->expressions, &grand_bitmap_obstack);
735 bitmap_initialize (&ret->values, &grand_bitmap_obstack);
736 return ret;
739 /* Return the value id for a PRE expression EXPR. */
741 static unsigned int
742 get_expr_value_id (pre_expr expr)
744 /* ??? We cannot assert that expr has a value-id (it can be 0), because
745 we assign value-ids only to expressions that have a result
746 in set_hashtable_value_ids. */
747 return expr->value_id;
750 /* Return a VN valnum (SSA name or constant) for the PRE value-id VAL. */
752 static tree
753 vn_valnum_from_value_id (unsigned int val)
755 if (value_id_constant_p (val))
757 pre_expr vexpr = constant_value_expressions[-val];
758 if (vexpr)
759 return PRE_EXPR_CONSTANT (vexpr);
760 return NULL_TREE;
763 bitmap exprset = value_expressions[val];
764 bitmap_iterator bi;
765 unsigned int i;
766 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
768 pre_expr vexpr = expression_for_id (i);
769 if (vexpr->kind == NAME)
770 return VN_INFO (PRE_EXPR_NAME (vexpr))->valnum;
772 return NULL_TREE;
775 /* Insert an expression EXPR into a bitmapped set. */
777 static void
778 bitmap_insert_into_set (bitmap_set_t set, pre_expr expr)
780 unsigned int val = get_expr_value_id (expr);
781 if (! value_id_constant_p (val))
783 /* Note this is the only function causing multiple expressions
784 for the same value to appear in a set. This is needed for
785 TMP_GEN, PHI_GEN and NEW_SETs. */
786 bitmap_set_bit (&set->values, val);
787 bitmap_set_bit (&set->expressions, get_or_alloc_expression_id (expr));
791 /* Copy a bitmapped set ORIG, into bitmapped set DEST. */
793 static void
794 bitmap_set_copy (bitmap_set_t dest, bitmap_set_t orig)
796 bitmap_copy (&dest->expressions, &orig->expressions);
797 bitmap_copy (&dest->values, &orig->values);
801 /* Free memory used up by SET. */
802 static void
803 bitmap_set_free (bitmap_set_t set)
805 bitmap_clear (&set->expressions);
806 bitmap_clear (&set->values);
809 static void
810 pre_expr_DFS (pre_expr expr, bitmap_set_t set, bitmap val_visited,
811 vec<pre_expr> &post);
813 /* DFS walk leaders of VAL to their operands with leaders in SET, collecting
814 expressions in SET in postorder into POST. */
816 static void
817 pre_expr_DFS (unsigned val, bitmap_set_t set, bitmap val_visited,
818 vec<pre_expr> &post)
820 unsigned int i;
821 bitmap_iterator bi;
823 /* Iterate over all leaders and DFS recurse. Borrowed from
824 bitmap_find_leader. */
825 bitmap exprset = value_expressions[val];
826 if (!exprset->first->next)
828 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
829 if (bitmap_bit_p (&set->expressions, i))
830 pre_expr_DFS (expression_for_id (i), set, val_visited, post);
831 return;
834 EXECUTE_IF_AND_IN_BITMAP (exprset, &set->expressions, 0, i, bi)
835 pre_expr_DFS (expression_for_id (i), set, val_visited, post);
838 /* DFS walk EXPR to its operands with leaders in SET, collecting
839 expressions in SET in postorder into POST. */
841 static void
842 pre_expr_DFS (pre_expr expr, bitmap_set_t set, bitmap val_visited,
843 vec<pre_expr> &post)
845 switch (expr->kind)
847 case NARY:
849 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
850 for (unsigned i = 0; i < nary->length; i++)
852 if (TREE_CODE (nary->op[i]) != SSA_NAME)
853 continue;
854 unsigned int op_val_id = VN_INFO (nary->op[i])->value_id;
855 /* If we already found a leader for the value we've
856 recursed already. Avoid the costly bitmap_find_leader. */
857 if (bitmap_bit_p (&set->values, op_val_id)
858 && bitmap_set_bit (val_visited, op_val_id))
859 pre_expr_DFS (op_val_id, set, val_visited, post);
861 break;
863 case REFERENCE:
865 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
866 vec<vn_reference_op_s> operands = ref->operands;
867 vn_reference_op_t operand;
868 for (unsigned i = 0; operands.iterate (i, &operand); i++)
870 tree op[3];
871 op[0] = operand->op0;
872 op[1] = operand->op1;
873 op[2] = operand->op2;
874 for (unsigned n = 0; n < 3; ++n)
876 if (!op[n] || TREE_CODE (op[n]) != SSA_NAME)
877 continue;
878 unsigned op_val_id = VN_INFO (op[n])->value_id;
879 if (bitmap_bit_p (&set->values, op_val_id)
880 && bitmap_set_bit (val_visited, op_val_id))
881 pre_expr_DFS (op_val_id, set, val_visited, post);
884 break;
886 default:;
888 post.quick_push (expr);
891 /* Generate an topological-ordered array of bitmap set SET. */
893 static vec<pre_expr>
894 sorted_array_from_bitmap_set (bitmap_set_t set)
896 unsigned int i;
897 bitmap_iterator bi;
898 vec<pre_expr> result;
900 /* Pre-allocate enough space for the array. */
901 result.create (bitmap_count_bits (&set->expressions));
903 auto_bitmap val_visited (&grand_bitmap_obstack);
904 bitmap_tree_view (val_visited);
905 FOR_EACH_VALUE_ID_IN_SET (set, i, bi)
906 if (bitmap_set_bit (val_visited, i))
907 pre_expr_DFS (i, set, val_visited, result);
909 return result;
912 /* Subtract all expressions contained in ORIG from DEST. */
914 static bitmap_set_t
915 bitmap_set_subtract_expressions (bitmap_set_t dest, bitmap_set_t orig)
917 bitmap_set_t result = bitmap_set_new ();
918 bitmap_iterator bi;
919 unsigned int i;
921 bitmap_and_compl (&result->expressions, &dest->expressions,
922 &orig->expressions);
924 FOR_EACH_EXPR_ID_IN_SET (result, i, bi)
926 pre_expr expr = expression_for_id (i);
927 unsigned int value_id = get_expr_value_id (expr);
928 bitmap_set_bit (&result->values, value_id);
931 return result;
934 /* Subtract all values in bitmap set B from bitmap set A. */
936 static void
937 bitmap_set_subtract_values (bitmap_set_t a, bitmap_set_t b)
939 unsigned int i;
940 bitmap_iterator bi;
941 unsigned to_remove = -1U;
942 bitmap_and_compl_into (&a->values, &b->values);
943 FOR_EACH_EXPR_ID_IN_SET (a, i, bi)
945 if (to_remove != -1U)
947 bitmap_clear_bit (&a->expressions, to_remove);
948 to_remove = -1U;
950 pre_expr expr = expression_for_id (i);
951 if (! bitmap_bit_p (&a->values, get_expr_value_id (expr)))
952 to_remove = i;
954 if (to_remove != -1U)
955 bitmap_clear_bit (&a->expressions, to_remove);
959 /* Return true if bitmapped set SET contains the value VALUE_ID. */
961 static bool
962 bitmap_set_contains_value (bitmap_set_t set, unsigned int value_id)
964 if (value_id_constant_p (value_id))
965 return true;
967 return bitmap_bit_p (&set->values, value_id);
970 /* Return true if two bitmap sets are equal. */
972 static bool
973 bitmap_set_equal (bitmap_set_t a, bitmap_set_t b)
975 return bitmap_equal_p (&a->values, &b->values);
978 /* Replace an instance of EXPR's VALUE with EXPR in SET if it exists,
979 and add it otherwise. Return true if any changes were made. */
981 static bool
982 bitmap_value_replace_in_set (bitmap_set_t set, pre_expr expr)
984 unsigned int val = get_expr_value_id (expr);
985 if (value_id_constant_p (val))
986 return false;
988 if (bitmap_set_contains_value (set, val))
990 /* The number of expressions having a given value is usually
991 significantly less than the total number of expressions in SET.
992 Thus, rather than check, for each expression in SET, whether it
993 has the value LOOKFOR, we walk the reverse mapping that tells us
994 what expressions have a given value, and see if any of those
995 expressions are in our set. For large testcases, this is about
996 5-10x faster than walking the bitmap. If this is somehow a
997 significant lose for some cases, we can choose which set to walk
998 based on the set size. */
999 unsigned int i;
1000 bitmap_iterator bi;
1001 bitmap exprset = value_expressions[val];
1002 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
1004 if (bitmap_clear_bit (&set->expressions, i))
1006 bitmap_set_bit (&set->expressions, get_expression_id (expr));
1007 return i != get_expression_id (expr);
1010 gcc_unreachable ();
1013 bitmap_insert_into_set (set, expr);
1014 return true;
1017 /* Insert EXPR into SET if EXPR's value is not already present in
1018 SET. */
1020 static void
1021 bitmap_value_insert_into_set (bitmap_set_t set, pre_expr expr)
1023 unsigned int val = get_expr_value_id (expr);
1025 gcc_checking_assert (expr->id == get_or_alloc_expression_id (expr));
1027 /* Constant values are always considered to be part of the set. */
1028 if (value_id_constant_p (val))
1029 return;
1031 /* If the value membership changed, add the expression. */
1032 if (bitmap_set_bit (&set->values, val))
1033 bitmap_set_bit (&set->expressions, expr->id);
1036 /* Print out EXPR to outfile. */
1038 static void
1039 print_pre_expr (FILE *outfile, const pre_expr expr)
1041 if (! expr)
1043 fprintf (outfile, "NULL");
1044 return;
1046 switch (expr->kind)
1048 case CONSTANT:
1049 print_generic_expr (outfile, PRE_EXPR_CONSTANT (expr));
1050 break;
1051 case NAME:
1052 print_generic_expr (outfile, PRE_EXPR_NAME (expr));
1053 break;
1054 case NARY:
1056 unsigned int i;
1057 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1058 fprintf (outfile, "{%s,", get_tree_code_name (nary->opcode));
1059 for (i = 0; i < nary->length; i++)
1061 print_generic_expr (outfile, nary->op[i]);
1062 if (i != (unsigned) nary->length - 1)
1063 fprintf (outfile, ",");
1065 fprintf (outfile, "}");
1067 break;
1069 case REFERENCE:
1071 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1072 print_vn_reference_ops (outfile, ref->operands);
1073 if (ref->vuse)
1075 fprintf (outfile, "@");
1076 print_generic_expr (outfile, ref->vuse);
1079 break;
1082 void debug_pre_expr (pre_expr);
1084 /* Like print_pre_expr but always prints to stderr. */
1085 DEBUG_FUNCTION void
1086 debug_pre_expr (pre_expr e)
1088 print_pre_expr (stderr, e);
1089 fprintf (stderr, "\n");
1092 /* Print out SET to OUTFILE. */
1094 static void
1095 print_bitmap_set (FILE *outfile, bitmap_set_t set,
1096 const char *setname, int blockindex)
1098 fprintf (outfile, "%s[%d] := { ", setname, blockindex);
1099 if (set)
1101 bool first = true;
1102 unsigned i;
1103 bitmap_iterator bi;
1105 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
1107 const pre_expr expr = expression_for_id (i);
1109 if (!first)
1110 fprintf (outfile, ", ");
1111 first = false;
1112 print_pre_expr (outfile, expr);
1114 fprintf (outfile, " (%04d)", get_expr_value_id (expr));
1117 fprintf (outfile, " }\n");
1120 void debug_bitmap_set (bitmap_set_t);
1122 DEBUG_FUNCTION void
1123 debug_bitmap_set (bitmap_set_t set)
1125 print_bitmap_set (stderr, set, "debug", 0);
1128 void debug_bitmap_sets_for (basic_block);
1130 DEBUG_FUNCTION void
1131 debug_bitmap_sets_for (basic_block bb)
1133 print_bitmap_set (stderr, AVAIL_OUT (bb), "avail_out", bb->index);
1134 print_bitmap_set (stderr, EXP_GEN (bb), "exp_gen", bb->index);
1135 print_bitmap_set (stderr, PHI_GEN (bb), "phi_gen", bb->index);
1136 print_bitmap_set (stderr, TMP_GEN (bb), "tmp_gen", bb->index);
1137 print_bitmap_set (stderr, ANTIC_IN (bb), "antic_in", bb->index);
1138 if (do_partial_partial)
1139 print_bitmap_set (stderr, PA_IN (bb), "pa_in", bb->index);
1140 print_bitmap_set (stderr, NEW_SETS (bb), "new_sets", bb->index);
1143 /* Print out the expressions that have VAL to OUTFILE. */
1145 static void
1146 print_value_expressions (FILE *outfile, unsigned int val)
1148 bitmap set = value_expressions[val];
1149 if (set)
1151 bitmap_set x;
1152 char s[10];
1153 sprintf (s, "%04d", val);
1154 x.expressions = *set;
1155 print_bitmap_set (outfile, &x, s, 0);
1160 DEBUG_FUNCTION void
1161 debug_value_expressions (unsigned int val)
1163 print_value_expressions (stderr, val);
1166 /* Given a CONSTANT, allocate a new CONSTANT type PRE_EXPR to
1167 represent it. */
1169 static pre_expr
1170 get_or_alloc_expr_for_constant (tree constant)
1172 unsigned int result_id;
1173 struct pre_expr_d expr;
1174 pre_expr newexpr;
1176 expr.kind = CONSTANT;
1177 PRE_EXPR_CONSTANT (&expr) = constant;
1178 result_id = lookup_expression_id (&expr);
1179 if (result_id != 0)
1180 return expression_for_id (result_id);
1182 newexpr = pre_expr_pool.allocate ();
1183 newexpr->kind = CONSTANT;
1184 newexpr->loc = UNKNOWN_LOCATION;
1185 PRE_EXPR_CONSTANT (newexpr) = constant;
1186 alloc_expression_id (newexpr);
1187 newexpr->value_id = get_or_alloc_constant_value_id (constant);
1188 add_to_value (newexpr->value_id, newexpr);
1189 return newexpr;
1192 /* Get or allocate a pre_expr for a piece of GIMPLE, and return it.
1193 Currently only supports constants and SSA_NAMES. */
1194 static pre_expr
1195 get_or_alloc_expr_for (tree t)
1197 if (TREE_CODE (t) == SSA_NAME)
1198 return get_or_alloc_expr_for_name (t);
1199 else if (is_gimple_min_invariant (t))
1200 return get_or_alloc_expr_for_constant (t);
1201 gcc_unreachable ();
1204 /* Return the folded version of T if T, when folded, is a gimple
1205 min_invariant or an SSA name. Otherwise, return T. */
1207 static pre_expr
1208 fully_constant_expression (pre_expr e)
1210 switch (e->kind)
1212 case CONSTANT:
1213 return e;
1214 case NARY:
1216 vn_nary_op_t nary = PRE_EXPR_NARY (e);
1217 tree res = vn_nary_simplify (nary);
1218 if (!res)
1219 return e;
1220 if (is_gimple_min_invariant (res))
1221 return get_or_alloc_expr_for_constant (res);
1222 if (TREE_CODE (res) == SSA_NAME)
1223 return get_or_alloc_expr_for_name (res);
1224 return e;
1226 case REFERENCE:
1228 vn_reference_t ref = PRE_EXPR_REFERENCE (e);
1229 tree folded;
1230 if ((folded = fully_constant_vn_reference_p (ref)))
1231 return get_or_alloc_expr_for_constant (folded);
1232 return e;
1234 default:
1235 return e;
1237 return e;
1240 /* Translate the VUSE backwards through phi nodes in E->dest, so that
1241 it has the value it would have in E->src. Set *SAME_VALID to true
1242 in case the new vuse doesn't change the value id of the OPERANDS. */
1244 static tree
1245 translate_vuse_through_block (vec<vn_reference_op_s> operands,
1246 alias_set_type set, alias_set_type base_set,
1247 tree type, tree vuse, edge e, bool *same_valid)
1249 basic_block phiblock = e->dest;
1250 gimple *phi = SSA_NAME_DEF_STMT (vuse);
1251 ao_ref ref;
1253 if (same_valid)
1254 *same_valid = true;
1256 if (gimple_bb (phi) != phiblock)
1257 return vuse;
1259 /* We have pruned expressions that are killed in PHIBLOCK via
1260 prune_clobbered_mems but we have not rewritten the VUSE to the one
1261 live at the start of the block. If there is no virtual PHI to translate
1262 through return the VUSE live at entry. Otherwise the VUSE to translate
1263 is the def of the virtual PHI node. */
1264 phi = get_virtual_phi (phiblock);
1265 if (!phi)
1266 return BB_LIVE_VOP_ON_EXIT
1267 (get_immediate_dominator (CDI_DOMINATORS, phiblock));
1269 if (same_valid
1270 && ao_ref_init_from_vn_reference (&ref, set, base_set, type, operands))
1272 bitmap visited = NULL;
1273 /* Try to find a vuse that dominates this phi node by skipping
1274 non-clobbering statements. */
1275 unsigned int cnt = param_sccvn_max_alias_queries_per_access;
1276 vuse = get_continuation_for_phi (phi, &ref, true,
1277 cnt, &visited, false, NULL, NULL);
1278 if (visited)
1279 BITMAP_FREE (visited);
1281 else
1282 vuse = NULL_TREE;
1283 /* If we didn't find any, the value ID can't stay the same. */
1284 if (!vuse && same_valid)
1285 *same_valid = false;
1287 /* ??? We would like to return vuse here as this is the canonical
1288 upmost vdef that this reference is associated with. But during
1289 insertion of the references into the hash tables we only ever
1290 directly insert with their direct gimple_vuse, hence returning
1291 something else would make us not find the other expression. */
1292 return PHI_ARG_DEF (phi, e->dest_idx);
1295 /* Like bitmap_find_leader, but checks for the value existing in SET1 *or*
1296 SET2 *or* SET3. This is used to avoid making a set consisting of the union
1297 of PA_IN and ANTIC_IN during insert and phi-translation. */
1299 static inline pre_expr
1300 find_leader_in_sets (unsigned int val, bitmap_set_t set1, bitmap_set_t set2,
1301 bitmap_set_t set3 = NULL)
1303 pre_expr result = NULL;
1305 if (set1)
1306 result = bitmap_find_leader (set1, val);
1307 if (!result && set2)
1308 result = bitmap_find_leader (set2, val);
1309 if (!result && set3)
1310 result = bitmap_find_leader (set3, val);
1311 return result;
1314 /* Get the tree type for our PRE expression e. */
1316 static tree
1317 get_expr_type (const pre_expr e)
1319 switch (e->kind)
1321 case NAME:
1322 return TREE_TYPE (PRE_EXPR_NAME (e));
1323 case CONSTANT:
1324 return TREE_TYPE (PRE_EXPR_CONSTANT (e));
1325 case REFERENCE:
1326 return PRE_EXPR_REFERENCE (e)->type;
1327 case NARY:
1328 return PRE_EXPR_NARY (e)->type;
1330 gcc_unreachable ();
1333 /* Get a representative SSA_NAME for a given expression that is available in B.
1334 Since all of our sub-expressions are treated as values, we require
1335 them to be SSA_NAME's for simplicity.
1336 Prior versions of GVNPRE used to use "value handles" here, so that
1337 an expression would be VH.11 + VH.10 instead of d_3 + e_6. In
1338 either case, the operands are really values (IE we do not expect
1339 them to be usable without finding leaders). */
1341 static tree
1342 get_representative_for (const pre_expr e, basic_block b = NULL)
1344 tree name, valnum = NULL_TREE;
1345 unsigned int value_id = get_expr_value_id (e);
1347 switch (e->kind)
1349 case NAME:
1350 return PRE_EXPR_NAME (e);
1351 case CONSTANT:
1352 return PRE_EXPR_CONSTANT (e);
1353 case NARY:
1354 case REFERENCE:
1356 /* Go through all of the expressions representing this value
1357 and pick out an SSA_NAME. */
1358 unsigned int i;
1359 bitmap_iterator bi;
1360 bitmap exprs = value_expressions[value_id];
1361 EXECUTE_IF_SET_IN_BITMAP (exprs, 0, i, bi)
1363 pre_expr rep = expression_for_id (i);
1364 if (rep->kind == NAME)
1366 tree name = PRE_EXPR_NAME (rep);
1367 valnum = VN_INFO (name)->valnum;
1368 gimple *def = SSA_NAME_DEF_STMT (name);
1369 /* We have to return either a new representative or one
1370 that can be used for expression simplification and thus
1371 is available in B. */
1372 if (! b
1373 || gimple_nop_p (def)
1374 || dominated_by_p (CDI_DOMINATORS, b, gimple_bb (def)))
1375 return name;
1377 else if (rep->kind == CONSTANT)
1378 return PRE_EXPR_CONSTANT (rep);
1381 break;
1384 /* If we reached here we couldn't find an SSA_NAME. This can
1385 happen when we've discovered a value that has never appeared in
1386 the program as set to an SSA_NAME, as the result of phi translation.
1387 Create one here.
1388 ??? We should be able to re-use this when we insert the statement
1389 to compute it. */
1390 name = make_temp_ssa_name (get_expr_type (e), gimple_build_nop (), "pretmp");
1391 vn_ssa_aux_t vn_info = VN_INFO (name);
1392 vn_info->value_id = value_id;
1393 vn_info->valnum = valnum ? valnum : name;
1394 vn_info->visited = true;
1395 /* ??? For now mark this SSA name for release by VN. */
1396 vn_info->needs_insertion = true;
1397 add_to_value (value_id, get_or_alloc_expr_for_name (name));
1398 if (dump_file && (dump_flags & TDF_DETAILS))
1400 fprintf (dump_file, "Created SSA_NAME representative ");
1401 print_generic_expr (dump_file, name);
1402 fprintf (dump_file, " for expression:");
1403 print_pre_expr (dump_file, e);
1404 fprintf (dump_file, " (%04d)\n", value_id);
1407 return name;
1411 static pre_expr
1412 phi_translate (bitmap_set_t, pre_expr, bitmap_set_t, bitmap_set_t, edge);
1414 /* Translate EXPR using phis in PHIBLOCK, so that it has the values of
1415 the phis in PRED. Return NULL if we can't find a leader for each part
1416 of the translated expression. */
1418 static pre_expr
1419 phi_translate_1 (bitmap_set_t dest,
1420 pre_expr expr, bitmap_set_t set1, bitmap_set_t set2, edge e)
1422 basic_block pred = e->src;
1423 basic_block phiblock = e->dest;
1424 location_t expr_loc = expr->loc;
1425 switch (expr->kind)
1427 case NARY:
1429 unsigned int i;
1430 bool changed = false;
1431 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1432 vn_nary_op_t newnary = XALLOCAVAR (struct vn_nary_op_s,
1433 sizeof_vn_nary_op (nary->length));
1434 memcpy (newnary, nary, sizeof_vn_nary_op (nary->length));
1436 for (i = 0; i < newnary->length; i++)
1438 if (TREE_CODE (newnary->op[i]) != SSA_NAME)
1439 continue;
1440 else
1442 pre_expr leader, result;
1443 unsigned int op_val_id = VN_INFO (newnary->op[i])->value_id;
1444 leader = find_leader_in_sets (op_val_id, set1, set2);
1445 result = phi_translate (dest, leader, set1, set2, e);
1446 if (result && result != leader)
1447 /* If op has a leader in the sets we translate make
1448 sure to use the value of the translated expression.
1449 We might need a new representative for that. */
1450 newnary->op[i] = get_representative_for (result, pred);
1451 else if (!result)
1452 return NULL;
1454 changed |= newnary->op[i] != nary->op[i];
1457 if (changed)
1459 pre_expr constant;
1460 unsigned int new_val_id;
1462 PRE_EXPR_NARY (expr) = newnary;
1463 constant = fully_constant_expression (expr);
1464 PRE_EXPR_NARY (expr) = nary;
1465 if (constant != expr)
1467 /* For non-CONSTANTs we have to make sure we can eventually
1468 insert the expression. Which means we need to have a
1469 leader for it. */
1470 if (constant->kind != CONSTANT)
1472 /* Do not allow simplifications to non-constants over
1473 backedges as this will likely result in a loop PHI node
1474 to be inserted and increased register pressure.
1475 See PR77498 - this avoids doing predcoms work in
1476 a less efficient way. */
1477 if (e->flags & EDGE_DFS_BACK)
1479 else
1481 unsigned value_id = get_expr_value_id (constant);
1482 /* We want a leader in ANTIC_OUT or AVAIL_OUT here.
1483 dest has what we computed into ANTIC_OUT sofar
1484 so pick from that - since topological sorting
1485 by sorted_array_from_bitmap_set isn't perfect
1486 we may lose some cases here. */
1487 constant = find_leader_in_sets (value_id, dest,
1488 AVAIL_OUT (pred));
1489 if (constant)
1491 if (dump_file && (dump_flags & TDF_DETAILS))
1493 fprintf (dump_file, "simplifying ");
1494 print_pre_expr (dump_file, expr);
1495 fprintf (dump_file, " translated %d -> %d to ",
1496 phiblock->index, pred->index);
1497 PRE_EXPR_NARY (expr) = newnary;
1498 print_pre_expr (dump_file, expr);
1499 PRE_EXPR_NARY (expr) = nary;
1500 fprintf (dump_file, " to ");
1501 print_pre_expr (dump_file, constant);
1502 fprintf (dump_file, "\n");
1504 return constant;
1508 else
1509 return constant;
1512 tree result = vn_nary_op_lookup_pieces (newnary->length,
1513 newnary->opcode,
1514 newnary->type,
1515 &newnary->op[0],
1516 &nary);
1517 if (result && is_gimple_min_invariant (result))
1518 return get_or_alloc_expr_for_constant (result);
1520 if (!nary || nary->predicated_values)
1522 new_val_id = get_next_value_id ();
1523 nary = vn_nary_op_insert_pieces (newnary->length,
1524 newnary->opcode,
1525 newnary->type,
1526 &newnary->op[0],
1527 result, new_val_id);
1529 expr = get_or_alloc_expr_for_nary (nary, expr_loc);
1530 add_to_value (get_expr_value_id (expr), expr);
1532 return expr;
1534 break;
1536 case REFERENCE:
1538 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1539 vec<vn_reference_op_s> operands = ref->operands;
1540 tree vuse = ref->vuse;
1541 tree newvuse = vuse;
1542 vec<vn_reference_op_s> newoperands = vNULL;
1543 bool changed = false, same_valid = true;
1544 unsigned int i, n;
1545 vn_reference_op_t operand;
1546 vn_reference_t newref;
1548 for (i = 0; operands.iterate (i, &operand); i++)
1550 pre_expr opresult;
1551 pre_expr leader;
1552 tree op[3];
1553 tree type = operand->type;
1554 vn_reference_op_s newop = *operand;
1555 op[0] = operand->op0;
1556 op[1] = operand->op1;
1557 op[2] = operand->op2;
1558 for (n = 0; n < 3; ++n)
1560 unsigned int op_val_id;
1561 if (!op[n])
1562 continue;
1563 if (TREE_CODE (op[n]) != SSA_NAME)
1565 /* We can't possibly insert these. */
1566 if (n != 0
1567 && !is_gimple_min_invariant (op[n]))
1568 break;
1569 continue;
1571 op_val_id = VN_INFO (op[n])->value_id;
1572 leader = find_leader_in_sets (op_val_id, set1, set2);
1573 opresult = phi_translate (dest, leader, set1, set2, e);
1574 if (opresult && opresult != leader)
1576 tree name = get_representative_for (opresult);
1577 changed |= name != op[n];
1578 op[n] = name;
1580 else if (!opresult)
1581 break;
1583 if (n != 3)
1585 newoperands.release ();
1586 return NULL;
1588 /* When we translate a MEM_REF across a backedge and we have
1589 restrict info that's not from our functions parameters
1590 we have to remap it since we now may deal with a different
1591 instance where the dependence info is no longer valid.
1592 See PR102970. Note instead of keeping a remapping table
1593 per backedge we simply throw away restrict info. */
1594 if ((newop.opcode == MEM_REF
1595 || newop.opcode == TARGET_MEM_REF)
1596 && newop.clique > 1
1597 && (e->flags & EDGE_DFS_BACK))
1599 newop.clique = 0;
1600 newop.base = 0;
1601 changed = true;
1603 if (!changed)
1604 continue;
1605 if (!newoperands.exists ())
1606 newoperands = operands.copy ();
1607 /* We may have changed from an SSA_NAME to a constant */
1608 if (newop.opcode == SSA_NAME && TREE_CODE (op[0]) != SSA_NAME)
1609 newop.opcode = TREE_CODE (op[0]);
1610 newop.type = type;
1611 newop.op0 = op[0];
1612 newop.op1 = op[1];
1613 newop.op2 = op[2];
1614 newoperands[i] = newop;
1616 gcc_checking_assert (i == operands.length ());
1618 if (vuse)
1620 newvuse = translate_vuse_through_block (newoperands.exists ()
1621 ? newoperands : operands,
1622 ref->set, ref->base_set,
1623 ref->type, vuse, e,
1624 changed
1625 ? NULL : &same_valid);
1626 if (newvuse == NULL_TREE)
1628 newoperands.release ();
1629 return NULL;
1633 if (changed || newvuse != vuse)
1635 unsigned int new_val_id;
1637 tree result = vn_reference_lookup_pieces (newvuse, ref->set,
1638 ref->base_set,
1639 ref->type,
1640 newoperands.exists ()
1641 ? newoperands : operands,
1642 &newref, VN_WALK);
1643 if (result)
1644 newoperands.release ();
1646 /* We can always insert constants, so if we have a partial
1647 redundant constant load of another type try to translate it
1648 to a constant of appropriate type. */
1649 if (result && is_gimple_min_invariant (result))
1651 tree tem = result;
1652 if (!useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1654 tem = fold_unary (VIEW_CONVERT_EXPR, ref->type, result);
1655 if (tem && !is_gimple_min_invariant (tem))
1656 tem = NULL_TREE;
1658 if (tem)
1659 return get_or_alloc_expr_for_constant (tem);
1662 /* If we'd have to convert things we would need to validate
1663 if we can insert the translated expression. So fail
1664 here for now - we cannot insert an alias with a different
1665 type in the VN tables either, as that would assert. */
1666 if (result
1667 && !useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1668 return NULL;
1669 else if (!result && newref
1670 && !useless_type_conversion_p (ref->type, newref->type))
1672 newoperands.release ();
1673 return NULL;
1676 if (newref)
1677 new_val_id = newref->value_id;
1678 else
1680 if (changed || !same_valid)
1681 new_val_id = get_next_value_id ();
1682 else
1683 new_val_id = ref->value_id;
1684 if (!newoperands.exists ())
1685 newoperands = operands.copy ();
1686 newref = vn_reference_insert_pieces (newvuse, ref->set,
1687 ref->base_set, ref->type,
1688 newoperands,
1689 result, new_val_id);
1690 newoperands = vNULL;
1692 expr = get_or_alloc_expr_for_reference (newref, expr_loc);
1693 add_to_value (new_val_id, expr);
1695 newoperands.release ();
1696 return expr;
1698 break;
1700 case NAME:
1702 tree name = PRE_EXPR_NAME (expr);
1703 gimple *def_stmt = SSA_NAME_DEF_STMT (name);
1704 /* If the SSA name is defined by a PHI node in this block,
1705 translate it. */
1706 if (gimple_code (def_stmt) == GIMPLE_PHI
1707 && gimple_bb (def_stmt) == phiblock)
1709 tree def = PHI_ARG_DEF (def_stmt, e->dest_idx);
1711 /* Handle constant. */
1712 if (is_gimple_min_invariant (def))
1713 return get_or_alloc_expr_for_constant (def);
1715 return get_or_alloc_expr_for_name (def);
1717 /* Otherwise return it unchanged - it will get removed if its
1718 value is not available in PREDs AVAIL_OUT set of expressions
1719 by the subtraction of TMP_GEN. */
1720 return expr;
1723 default:
1724 gcc_unreachable ();
1728 /* Wrapper around phi_translate_1 providing caching functionality. */
1730 static pre_expr
1731 phi_translate (bitmap_set_t dest, pre_expr expr,
1732 bitmap_set_t set1, bitmap_set_t set2, edge e)
1734 expr_pred_trans_t slot = NULL;
1735 pre_expr phitrans;
1737 if (!expr)
1738 return NULL;
1740 /* Constants contain no values that need translation. */
1741 if (expr->kind == CONSTANT)
1742 return expr;
1744 if (value_id_constant_p (get_expr_value_id (expr)))
1745 return expr;
1747 /* Don't add translations of NAMEs as those are cheap to translate. */
1748 if (expr->kind != NAME)
1750 if (phi_trans_add (&slot, expr, e->src))
1751 return slot->v == 0 ? NULL : expression_for_id (slot->v);
1752 /* Store NULL for the value we want to return in the case of
1753 recursing. */
1754 slot->v = 0;
1757 /* Translate. */
1758 basic_block saved_valueize_bb = vn_context_bb;
1759 vn_context_bb = e->src;
1760 phitrans = phi_translate_1 (dest, expr, set1, set2, e);
1761 vn_context_bb = saved_valueize_bb;
1763 if (slot)
1765 /* We may have reallocated. */
1766 phi_trans_add (&slot, expr, e->src);
1767 if (phitrans)
1768 slot->v = get_expression_id (phitrans);
1769 else
1770 /* Remove failed translations again, they cause insert
1771 iteration to not pick up new opportunities reliably. */
1772 PHI_TRANS_TABLE (e->src)->clear_slot (slot);
1775 return phitrans;
1779 /* For each expression in SET, translate the values through phi nodes
1780 in PHIBLOCK using edge PHIBLOCK->PRED, and store the resulting
1781 expressions in DEST. */
1783 static void
1784 phi_translate_set (bitmap_set_t dest, bitmap_set_t set, edge e)
1786 bitmap_iterator bi;
1787 unsigned int i;
1789 if (gimple_seq_empty_p (phi_nodes (e->dest)))
1791 bitmap_set_copy (dest, set);
1792 return;
1795 /* Allocate the phi-translation cache where we have an idea about
1796 its size. hash-table implementation internals tell us that
1797 allocating the table to fit twice the number of elements will
1798 make sure we do not usually re-allocate. */
1799 if (!PHI_TRANS_TABLE (e->src))
1800 PHI_TRANS_TABLE (e->src) = new hash_table<expr_pred_trans_d>
1801 (2 * bitmap_count_bits (&set->expressions));
1802 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
1804 pre_expr expr = expression_for_id (i);
1805 pre_expr translated = phi_translate (dest, expr, set, NULL, e);
1806 if (!translated)
1807 continue;
1809 bitmap_insert_into_set (dest, translated);
1813 /* Find the leader for a value (i.e., the name representing that
1814 value) in a given set, and return it. Return NULL if no leader
1815 is found. */
1817 static pre_expr
1818 bitmap_find_leader (bitmap_set_t set, unsigned int val)
1820 if (value_id_constant_p (val))
1821 return constant_value_expressions[-val];
1823 if (bitmap_set_contains_value (set, val))
1825 /* Rather than walk the entire bitmap of expressions, and see
1826 whether any of them has the value we are looking for, we look
1827 at the reverse mapping, which tells us the set of expressions
1828 that have a given value (IE value->expressions with that
1829 value) and see if any of those expressions are in our set.
1830 The number of expressions per value is usually significantly
1831 less than the number of expressions in the set. In fact, for
1832 large testcases, doing it this way is roughly 5-10x faster
1833 than walking the bitmap.
1834 If this is somehow a significant lose for some cases, we can
1835 choose which set to walk based on which set is smaller. */
1836 unsigned int i;
1837 bitmap_iterator bi;
1838 bitmap exprset = value_expressions[val];
1840 if (!exprset->first->next)
1841 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
1842 if (bitmap_bit_p (&set->expressions, i))
1843 return expression_for_id (i);
1845 EXECUTE_IF_AND_IN_BITMAP (exprset, &set->expressions, 0, i, bi)
1846 return expression_for_id (i);
1848 return NULL;
1851 /* Determine if EXPR, a memory expression, is ANTIC_IN at the top of
1852 BLOCK by seeing if it is not killed in the block. Note that we are
1853 only determining whether there is a store that kills it. Because
1854 of the order in which clean iterates over values, we are guaranteed
1855 that altered operands will have caused us to be eliminated from the
1856 ANTIC_IN set already. */
1858 static bool
1859 value_dies_in_block_x (pre_expr expr, basic_block block)
1861 tree vuse = PRE_EXPR_REFERENCE (expr)->vuse;
1862 vn_reference_t refx = PRE_EXPR_REFERENCE (expr);
1863 gimple *def;
1864 gimple_stmt_iterator gsi;
1865 unsigned id = get_expression_id (expr);
1866 bool res = false;
1867 ao_ref ref;
1869 if (!vuse)
1870 return false;
1872 /* Lookup a previously calculated result. */
1873 if (EXPR_DIES (block)
1874 && bitmap_bit_p (EXPR_DIES (block), id * 2))
1875 return bitmap_bit_p (EXPR_DIES (block), id * 2 + 1);
1877 /* A memory expression {e, VUSE} dies in the block if there is a
1878 statement that may clobber e. If, starting statement walk from the
1879 top of the basic block, a statement uses VUSE there can be no kill
1880 inbetween that use and the original statement that loaded {e, VUSE},
1881 so we can stop walking. */
1882 ref.base = NULL_TREE;
1883 for (gsi = gsi_start_bb (block); !gsi_end_p (gsi); gsi_next (&gsi))
1885 tree def_vuse, def_vdef;
1886 def = gsi_stmt (gsi);
1887 def_vuse = gimple_vuse (def);
1888 def_vdef = gimple_vdef (def);
1890 /* Not a memory statement. */
1891 if (!def_vuse)
1892 continue;
1894 /* Not a may-def. */
1895 if (!def_vdef)
1897 /* A load with the same VUSE, we're done. */
1898 if (def_vuse == vuse)
1899 break;
1901 continue;
1904 /* Init ref only if we really need it. */
1905 if (ref.base == NULL_TREE
1906 && !ao_ref_init_from_vn_reference (&ref, refx->set, refx->base_set,
1907 refx->type, refx->operands))
1909 res = true;
1910 break;
1912 /* If the statement may clobber expr, it dies. */
1913 if (stmt_may_clobber_ref_p_1 (def, &ref))
1915 res = true;
1916 break;
1920 /* Remember the result. */
1921 if (!EXPR_DIES (block))
1922 EXPR_DIES (block) = BITMAP_ALLOC (&grand_bitmap_obstack);
1923 bitmap_set_bit (EXPR_DIES (block), id * 2);
1924 if (res)
1925 bitmap_set_bit (EXPR_DIES (block), id * 2 + 1);
1927 return res;
1931 /* Determine if OP is valid in SET1 U SET2, which it is when the union
1932 contains its value-id. */
1934 static bool
1935 op_valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, tree op)
1937 if (op && TREE_CODE (op) == SSA_NAME)
1939 unsigned int value_id = VN_INFO (op)->value_id;
1940 if (!(bitmap_set_contains_value (set1, value_id)
1941 || (set2 && bitmap_set_contains_value (set2, value_id))))
1942 return false;
1944 return true;
1947 /* Determine if the expression EXPR is valid in SET1 U SET2.
1948 ONLY SET2 CAN BE NULL.
1949 This means that we have a leader for each part of the expression
1950 (if it consists of values), or the expression is an SSA_NAME.
1951 For loads/calls, we also see if the vuse is killed in this block. */
1953 static bool
1954 valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, pre_expr expr)
1956 switch (expr->kind)
1958 case NAME:
1959 /* By construction all NAMEs are available. Non-available
1960 NAMEs are removed by subtracting TMP_GEN from the sets. */
1961 return true;
1962 case NARY:
1964 unsigned int i;
1965 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1966 for (i = 0; i < nary->length; i++)
1967 if (!op_valid_in_sets (set1, set2, nary->op[i]))
1968 return false;
1969 return true;
1971 break;
1972 case REFERENCE:
1974 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1975 vn_reference_op_t vro;
1976 unsigned int i;
1978 FOR_EACH_VEC_ELT (ref->operands, i, vro)
1980 if (!op_valid_in_sets (set1, set2, vro->op0)
1981 || !op_valid_in_sets (set1, set2, vro->op1)
1982 || !op_valid_in_sets (set1, set2, vro->op2))
1983 return false;
1985 return true;
1987 default:
1988 gcc_unreachable ();
1992 /* Clean the set of expressions SET1 that are no longer valid in SET1 or SET2.
1993 This means expressions that are made up of values we have no leaders for
1994 in SET1 or SET2. */
1996 static void
1997 clean (bitmap_set_t set1, bitmap_set_t set2 = NULL)
1999 vec<pre_expr> exprs = sorted_array_from_bitmap_set (set1);
2000 pre_expr expr;
2001 int i;
2003 FOR_EACH_VEC_ELT (exprs, i, expr)
2005 if (!valid_in_sets (set1, set2, expr))
2007 unsigned int val = get_expr_value_id (expr);
2008 bitmap_clear_bit (&set1->expressions, get_expression_id (expr));
2009 /* We are entered with possibly multiple expressions for a value
2010 so before removing a value from the set see if there's an
2011 expression for it left. */
2012 if (! bitmap_find_leader (set1, val))
2013 bitmap_clear_bit (&set1->values, val);
2016 exprs.release ();
2018 if (flag_checking)
2020 unsigned j;
2021 bitmap_iterator bi;
2022 FOR_EACH_EXPR_ID_IN_SET (set1, j, bi)
2023 gcc_assert (valid_in_sets (set1, set2, expression_for_id (j)));
2027 /* Clean the set of expressions that are no longer valid in SET because
2028 they are clobbered in BLOCK or because they trap and may not be executed. */
2030 static void
2031 prune_clobbered_mems (bitmap_set_t set, basic_block block)
2033 bitmap_iterator bi;
2034 unsigned i;
2035 unsigned to_remove = -1U;
2036 bool any_removed = false;
2038 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
2040 /* Remove queued expr. */
2041 if (to_remove != -1U)
2043 bitmap_clear_bit (&set->expressions, to_remove);
2044 any_removed = true;
2045 to_remove = -1U;
2048 pre_expr expr = expression_for_id (i);
2049 if (expr->kind == REFERENCE)
2051 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2052 if (ref->vuse)
2054 gimple *def_stmt = SSA_NAME_DEF_STMT (ref->vuse);
2055 if (!gimple_nop_p (def_stmt)
2056 && ((gimple_bb (def_stmt) != block
2057 && !dominated_by_p (CDI_DOMINATORS,
2058 block, gimple_bb (def_stmt)))
2059 || (gimple_bb (def_stmt) == block
2060 && value_dies_in_block_x (expr, block))))
2061 to_remove = i;
2063 /* If the REFERENCE may trap make sure the block does not contain
2064 a possible exit point.
2065 ??? This is overly conservative if we translate AVAIL_OUT
2066 as the available expression might be after the exit point. */
2067 if (BB_MAY_NOTRETURN (block)
2068 && vn_reference_may_trap (ref))
2069 to_remove = i;
2071 else if (expr->kind == NARY)
2073 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2074 /* If the NARY may trap make sure the block does not contain
2075 a possible exit point.
2076 ??? This is overly conservative if we translate AVAIL_OUT
2077 as the available expression might be after the exit point. */
2078 if (BB_MAY_NOTRETURN (block)
2079 && vn_nary_may_trap (nary))
2080 to_remove = i;
2084 /* Remove queued expr. */
2085 if (to_remove != -1U)
2087 bitmap_clear_bit (&set->expressions, to_remove);
2088 any_removed = true;
2091 /* Above we only removed expressions, now clean the set of values
2092 which no longer have any corresponding expression. We cannot
2093 clear the value at the time we remove an expression since there
2094 may be multiple expressions per value.
2095 If we'd queue possibly to be removed values we could use
2096 the bitmap_find_leader way to see if there's still an expression
2097 for it. For some ratio of to be removed values and number of
2098 values/expressions in the set this might be faster than rebuilding
2099 the value-set. */
2100 if (any_removed)
2102 bitmap_clear (&set->values);
2103 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
2105 pre_expr expr = expression_for_id (i);
2106 unsigned int value_id = get_expr_value_id (expr);
2107 bitmap_set_bit (&set->values, value_id);
2112 /* Compute the ANTIC set for BLOCK.
2114 If succs(BLOCK) > 1 then
2115 ANTIC_OUT[BLOCK] = intersection of ANTIC_IN[b] for all succ(BLOCK)
2116 else if succs(BLOCK) == 1 then
2117 ANTIC_OUT[BLOCK] = phi_translate (ANTIC_IN[succ(BLOCK)])
2119 ANTIC_IN[BLOCK] = clean(ANTIC_OUT[BLOCK] U EXP_GEN[BLOCK] - TMP_GEN[BLOCK])
2121 Note that clean() is deferred until after the iteration. */
2123 static bool
2124 compute_antic_aux (basic_block block, bool block_has_abnormal_pred_edge)
2126 bitmap_set_t S, old, ANTIC_OUT;
2127 edge e;
2128 edge_iterator ei;
2130 bool was_visited = BB_VISITED (block);
2131 bool changed = ! BB_VISITED (block);
2132 BB_VISITED (block) = 1;
2133 old = ANTIC_OUT = S = NULL;
2135 /* If any edges from predecessors are abnormal, antic_in is empty,
2136 so do nothing. */
2137 if (block_has_abnormal_pred_edge)
2138 goto maybe_dump_sets;
2140 old = ANTIC_IN (block);
2141 ANTIC_OUT = bitmap_set_new ();
2143 /* If the block has no successors, ANTIC_OUT is empty. */
2144 if (EDGE_COUNT (block->succs) == 0)
2146 /* If we have one successor, we could have some phi nodes to
2147 translate through. */
2148 else if (single_succ_p (block))
2150 e = single_succ_edge (block);
2151 gcc_assert (BB_VISITED (e->dest));
2152 phi_translate_set (ANTIC_OUT, ANTIC_IN (e->dest), e);
2154 /* If we have multiple successors, we take the intersection of all of
2155 them. Note that in the case of loop exit phi nodes, we may have
2156 phis to translate through. */
2157 else
2159 size_t i;
2160 edge first = NULL;
2162 auto_vec<edge> worklist (EDGE_COUNT (block->succs));
2163 FOR_EACH_EDGE (e, ei, block->succs)
2165 if (!first
2166 && BB_VISITED (e->dest))
2167 first = e;
2168 else if (BB_VISITED (e->dest))
2169 worklist.quick_push (e);
2170 else
2172 /* Unvisited successors get their ANTIC_IN replaced by the
2173 maximal set to arrive at a maximum ANTIC_IN solution.
2174 We can ignore them in the intersection operation and thus
2175 need not explicitely represent that maximum solution. */
2176 if (dump_file && (dump_flags & TDF_DETAILS))
2177 fprintf (dump_file, "ANTIC_IN is MAX on %d->%d\n",
2178 e->src->index, e->dest->index);
2182 /* Of multiple successors we have to have visited one already
2183 which is guaranteed by iteration order. */
2184 gcc_assert (first != NULL);
2186 phi_translate_set (ANTIC_OUT, ANTIC_IN (first->dest), first);
2188 /* If we have multiple successors we need to intersect the ANTIC_OUT
2189 sets. For values that's a simple intersection but for
2190 expressions it is a union. Given we want to have a single
2191 expression per value in our sets we have to canonicalize.
2192 Avoid randomness and running into cycles like for PR82129 and
2193 canonicalize the expression we choose to the one with the
2194 lowest id. This requires we actually compute the union first. */
2195 FOR_EACH_VEC_ELT (worklist, i, e)
2197 if (!gimple_seq_empty_p (phi_nodes (e->dest)))
2199 bitmap_set_t tmp = bitmap_set_new ();
2200 phi_translate_set (tmp, ANTIC_IN (e->dest), e);
2201 bitmap_and_into (&ANTIC_OUT->values, &tmp->values);
2202 bitmap_ior_into (&ANTIC_OUT->expressions, &tmp->expressions);
2203 bitmap_set_free (tmp);
2205 else
2207 bitmap_and_into (&ANTIC_OUT->values, &ANTIC_IN (e->dest)->values);
2208 bitmap_ior_into (&ANTIC_OUT->expressions,
2209 &ANTIC_IN (e->dest)->expressions);
2212 if (! worklist.is_empty ())
2214 /* Prune expressions not in the value set. */
2215 bitmap_iterator bi;
2216 unsigned int i;
2217 unsigned int to_clear = -1U;
2218 FOR_EACH_EXPR_ID_IN_SET (ANTIC_OUT, i, bi)
2220 if (to_clear != -1U)
2222 bitmap_clear_bit (&ANTIC_OUT->expressions, to_clear);
2223 to_clear = -1U;
2225 pre_expr expr = expression_for_id (i);
2226 unsigned int value_id = get_expr_value_id (expr);
2227 if (!bitmap_bit_p (&ANTIC_OUT->values, value_id))
2228 to_clear = i;
2230 if (to_clear != -1U)
2231 bitmap_clear_bit (&ANTIC_OUT->expressions, to_clear);
2235 /* Prune expressions that are clobbered in block and thus become
2236 invalid if translated from ANTIC_OUT to ANTIC_IN. */
2237 prune_clobbered_mems (ANTIC_OUT, block);
2239 /* Generate ANTIC_OUT - TMP_GEN. */
2240 S = bitmap_set_subtract_expressions (ANTIC_OUT, TMP_GEN (block));
2242 /* Start ANTIC_IN with EXP_GEN - TMP_GEN. */
2243 ANTIC_IN (block) = bitmap_set_subtract_expressions (EXP_GEN (block),
2244 TMP_GEN (block));
2246 /* Then union in the ANTIC_OUT - TMP_GEN values,
2247 to get ANTIC_OUT U EXP_GEN - TMP_GEN */
2248 bitmap_ior_into (&ANTIC_IN (block)->values, &S->values);
2249 bitmap_ior_into (&ANTIC_IN (block)->expressions, &S->expressions);
2251 /* clean (ANTIC_IN (block)) is defered to after the iteration converged
2252 because it can cause non-convergence, see for example PR81181. */
2254 /* Intersect ANTIC_IN with the old ANTIC_IN. This is required until
2255 we properly represent the maximum expression set, thus not prune
2256 values without expressions during the iteration. */
2257 if (was_visited
2258 && bitmap_and_into (&ANTIC_IN (block)->values, &old->values))
2260 if (dump_file && (dump_flags & TDF_DETAILS))
2261 fprintf (dump_file, "warning: intersecting with old ANTIC_IN "
2262 "shrinks the set\n");
2263 /* Prune expressions not in the value set. */
2264 bitmap_iterator bi;
2265 unsigned int i;
2266 unsigned int to_clear = -1U;
2267 FOR_EACH_EXPR_ID_IN_SET (ANTIC_IN (block), i, bi)
2269 if (to_clear != -1U)
2271 bitmap_clear_bit (&ANTIC_IN (block)->expressions, to_clear);
2272 to_clear = -1U;
2274 pre_expr expr = expression_for_id (i);
2275 unsigned int value_id = get_expr_value_id (expr);
2276 if (!bitmap_bit_p (&ANTIC_IN (block)->values, value_id))
2277 to_clear = i;
2279 if (to_clear != -1U)
2280 bitmap_clear_bit (&ANTIC_IN (block)->expressions, to_clear);
2283 if (!bitmap_set_equal (old, ANTIC_IN (block)))
2284 changed = true;
2286 maybe_dump_sets:
2287 if (dump_file && (dump_flags & TDF_DETAILS))
2289 if (ANTIC_OUT)
2290 print_bitmap_set (dump_file, ANTIC_OUT, "ANTIC_OUT", block->index);
2292 if (changed)
2293 fprintf (dump_file, "[changed] ");
2294 print_bitmap_set (dump_file, ANTIC_IN (block), "ANTIC_IN",
2295 block->index);
2297 if (S)
2298 print_bitmap_set (dump_file, S, "S", block->index);
2300 if (old)
2301 bitmap_set_free (old);
2302 if (S)
2303 bitmap_set_free (S);
2304 if (ANTIC_OUT)
2305 bitmap_set_free (ANTIC_OUT);
2306 return changed;
2309 /* Compute PARTIAL_ANTIC for BLOCK.
2311 If succs(BLOCK) > 1 then
2312 PA_OUT[BLOCK] = value wise union of PA_IN[b] + all ANTIC_IN not
2313 in ANTIC_OUT for all succ(BLOCK)
2314 else if succs(BLOCK) == 1 then
2315 PA_OUT[BLOCK] = phi_translate (PA_IN[succ(BLOCK)])
2317 PA_IN[BLOCK] = clean(PA_OUT[BLOCK] - TMP_GEN[BLOCK] - ANTIC_IN[BLOCK])
2320 static void
2321 compute_partial_antic_aux (basic_block block,
2322 bool block_has_abnormal_pred_edge)
2324 bitmap_set_t old_PA_IN;
2325 bitmap_set_t PA_OUT;
2326 edge e;
2327 edge_iterator ei;
2328 unsigned long max_pa = param_max_partial_antic_length;
2330 old_PA_IN = PA_OUT = NULL;
2332 /* If any edges from predecessors are abnormal, antic_in is empty,
2333 so do nothing. */
2334 if (block_has_abnormal_pred_edge)
2335 goto maybe_dump_sets;
2337 /* If there are too many partially anticipatable values in the
2338 block, phi_translate_set can take an exponential time: stop
2339 before the translation starts. */
2340 if (max_pa
2341 && single_succ_p (block)
2342 && bitmap_count_bits (&PA_IN (single_succ (block))->values) > max_pa)
2343 goto maybe_dump_sets;
2345 old_PA_IN = PA_IN (block);
2346 PA_OUT = bitmap_set_new ();
2348 /* If the block has no successors, ANTIC_OUT is empty. */
2349 if (EDGE_COUNT (block->succs) == 0)
2351 /* If we have one successor, we could have some phi nodes to
2352 translate through. Note that we can't phi translate across DFS
2353 back edges in partial antic, because it uses a union operation on
2354 the successors. For recurrences like IV's, we will end up
2355 generating a new value in the set on each go around (i + 3 (VH.1)
2356 VH.1 + 1 (VH.2), VH.2 + 1 (VH.3), etc), forever. */
2357 else if (single_succ_p (block))
2359 e = single_succ_edge (block);
2360 if (!(e->flags & EDGE_DFS_BACK))
2361 phi_translate_set (PA_OUT, PA_IN (e->dest), e);
2363 /* If we have multiple successors, we take the union of all of
2364 them. */
2365 else
2367 size_t i;
2369 auto_vec<edge> worklist (EDGE_COUNT (block->succs));
2370 FOR_EACH_EDGE (e, ei, block->succs)
2372 if (e->flags & EDGE_DFS_BACK)
2373 continue;
2374 worklist.quick_push (e);
2376 if (worklist.length () > 0)
2378 FOR_EACH_VEC_ELT (worklist, i, e)
2380 unsigned int i;
2381 bitmap_iterator bi;
2383 FOR_EACH_EXPR_ID_IN_SET (ANTIC_IN (e->dest), i, bi)
2384 bitmap_value_insert_into_set (PA_OUT,
2385 expression_for_id (i));
2386 if (!gimple_seq_empty_p (phi_nodes (e->dest)))
2388 bitmap_set_t pa_in = bitmap_set_new ();
2389 phi_translate_set (pa_in, PA_IN (e->dest), e);
2390 FOR_EACH_EXPR_ID_IN_SET (pa_in, i, bi)
2391 bitmap_value_insert_into_set (PA_OUT,
2392 expression_for_id (i));
2393 bitmap_set_free (pa_in);
2395 else
2396 FOR_EACH_EXPR_ID_IN_SET (PA_IN (e->dest), i, bi)
2397 bitmap_value_insert_into_set (PA_OUT,
2398 expression_for_id (i));
2403 /* Prune expressions that are clobbered in block and thus become
2404 invalid if translated from PA_OUT to PA_IN. */
2405 prune_clobbered_mems (PA_OUT, block);
2407 /* PA_IN starts with PA_OUT - TMP_GEN.
2408 Then we subtract things from ANTIC_IN. */
2409 PA_IN (block) = bitmap_set_subtract_expressions (PA_OUT, TMP_GEN (block));
2411 /* For partial antic, we want to put back in the phi results, since
2412 we will properly avoid making them partially antic over backedges. */
2413 bitmap_ior_into (&PA_IN (block)->values, &PHI_GEN (block)->values);
2414 bitmap_ior_into (&PA_IN (block)->expressions, &PHI_GEN (block)->expressions);
2416 /* PA_IN[block] = PA_IN[block] - ANTIC_IN[block] */
2417 bitmap_set_subtract_values (PA_IN (block), ANTIC_IN (block));
2419 clean (PA_IN (block), ANTIC_IN (block));
2421 maybe_dump_sets:
2422 if (dump_file && (dump_flags & TDF_DETAILS))
2424 if (PA_OUT)
2425 print_bitmap_set (dump_file, PA_OUT, "PA_OUT", block->index);
2427 print_bitmap_set (dump_file, PA_IN (block), "PA_IN", block->index);
2429 if (old_PA_IN)
2430 bitmap_set_free (old_PA_IN);
2431 if (PA_OUT)
2432 bitmap_set_free (PA_OUT);
2435 /* Compute ANTIC and partial ANTIC sets. */
2437 static void
2438 compute_antic (void)
2440 bool changed = true;
2441 int num_iterations = 0;
2442 basic_block block;
2443 int i;
2444 edge_iterator ei;
2445 edge e;
2447 /* If any predecessor edges are abnormal, we punt, so antic_in is empty.
2448 We pre-build the map of blocks with incoming abnormal edges here. */
2449 auto_sbitmap has_abnormal_preds (last_basic_block_for_fn (cfun));
2450 bitmap_clear (has_abnormal_preds);
2452 FOR_ALL_BB_FN (block, cfun)
2454 BB_VISITED (block) = 0;
2456 FOR_EACH_EDGE (e, ei, block->preds)
2457 if (e->flags & EDGE_ABNORMAL)
2459 bitmap_set_bit (has_abnormal_preds, block->index);
2460 break;
2463 /* While we are here, give empty ANTIC_IN sets to each block. */
2464 ANTIC_IN (block) = bitmap_set_new ();
2465 if (do_partial_partial)
2466 PA_IN (block) = bitmap_set_new ();
2469 /* At the exit block we anticipate nothing. */
2470 BB_VISITED (EXIT_BLOCK_PTR_FOR_FN (cfun)) = 1;
2472 /* For ANTIC computation we need a postorder that also guarantees that
2473 a block with a single successor is visited after its successor.
2474 RPO on the inverted CFG has this property. */
2475 auto_vec<int, 20> postorder;
2476 inverted_post_order_compute (&postorder);
2478 auto_sbitmap worklist (last_basic_block_for_fn (cfun) + 1);
2479 bitmap_clear (worklist);
2480 FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR_FOR_FN (cfun)->preds)
2481 bitmap_set_bit (worklist, e->src->index);
2482 while (changed)
2484 if (dump_file && (dump_flags & TDF_DETAILS))
2485 fprintf (dump_file, "Starting iteration %d\n", num_iterations);
2486 /* ??? We need to clear our PHI translation cache here as the
2487 ANTIC sets shrink and we restrict valid translations to
2488 those having operands with leaders in ANTIC. Same below
2489 for PA ANTIC computation. */
2490 num_iterations++;
2491 changed = false;
2492 for (i = postorder.length () - 1; i >= 0; i--)
2494 if (bitmap_bit_p (worklist, postorder[i]))
2496 basic_block block = BASIC_BLOCK_FOR_FN (cfun, postorder[i]);
2497 bitmap_clear_bit (worklist, block->index);
2498 if (compute_antic_aux (block,
2499 bitmap_bit_p (has_abnormal_preds,
2500 block->index)))
2502 FOR_EACH_EDGE (e, ei, block->preds)
2503 bitmap_set_bit (worklist, e->src->index);
2504 changed = true;
2508 /* Theoretically possible, but *highly* unlikely. */
2509 gcc_checking_assert (num_iterations < 500);
2512 /* We have to clean after the dataflow problem converged as cleaning
2513 can cause non-convergence because it is based on expressions
2514 rather than values. */
2515 FOR_EACH_BB_FN (block, cfun)
2516 clean (ANTIC_IN (block));
2518 statistics_histogram_event (cfun, "compute_antic iterations",
2519 num_iterations);
2521 if (do_partial_partial)
2523 /* For partial antic we ignore backedges and thus we do not need
2524 to perform any iteration when we process blocks in postorder. */
2525 for (i = postorder.length () - 1; i >= 0; i--)
2527 basic_block block = BASIC_BLOCK_FOR_FN (cfun, postorder[i]);
2528 compute_partial_antic_aux (block,
2529 bitmap_bit_p (has_abnormal_preds,
2530 block->index));
2536 /* Inserted expressions are placed onto this worklist, which is used
2537 for performing quick dead code elimination of insertions we made
2538 that didn't turn out to be necessary. */
2539 static bitmap inserted_exprs;
2541 /* The actual worker for create_component_ref_by_pieces. */
2543 static tree
2544 create_component_ref_by_pieces_1 (basic_block block, vn_reference_t ref,
2545 unsigned int *operand, gimple_seq *stmts)
2547 vn_reference_op_t currop = &ref->operands[*operand];
2548 tree genop;
2549 ++*operand;
2550 switch (currop->opcode)
2552 case CALL_EXPR:
2553 gcc_unreachable ();
2555 case MEM_REF:
2557 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2558 stmts);
2559 if (!baseop)
2560 return NULL_TREE;
2561 tree offset = currop->op0;
2562 if (TREE_CODE (baseop) == ADDR_EXPR
2563 && handled_component_p (TREE_OPERAND (baseop, 0)))
2565 poly_int64 off;
2566 tree base;
2567 base = get_addr_base_and_unit_offset (TREE_OPERAND (baseop, 0),
2568 &off);
2569 gcc_assert (base);
2570 offset = int_const_binop (PLUS_EXPR, offset,
2571 build_int_cst (TREE_TYPE (offset),
2572 off));
2573 baseop = build_fold_addr_expr (base);
2575 genop = build2 (MEM_REF, currop->type, baseop, offset);
2576 MR_DEPENDENCE_CLIQUE (genop) = currop->clique;
2577 MR_DEPENDENCE_BASE (genop) = currop->base;
2578 REF_REVERSE_STORAGE_ORDER (genop) = currop->reverse;
2579 return genop;
2582 case TARGET_MEM_REF:
2584 tree genop0 = NULL_TREE, genop1 = NULL_TREE;
2585 vn_reference_op_t nextop = &ref->operands[(*operand)++];
2586 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2587 stmts);
2588 if (!baseop)
2589 return NULL_TREE;
2590 if (currop->op0)
2592 genop0 = find_or_generate_expression (block, currop->op0, stmts);
2593 if (!genop0)
2594 return NULL_TREE;
2596 if (nextop->op0)
2598 genop1 = find_or_generate_expression (block, nextop->op0, stmts);
2599 if (!genop1)
2600 return NULL_TREE;
2602 genop = build5 (TARGET_MEM_REF, currop->type,
2603 baseop, currop->op2, genop0, currop->op1, genop1);
2605 MR_DEPENDENCE_CLIQUE (genop) = currop->clique;
2606 MR_DEPENDENCE_BASE (genop) = currop->base;
2607 return genop;
2610 case ADDR_EXPR:
2611 if (currop->op0)
2613 gcc_assert (is_gimple_min_invariant (currop->op0));
2614 return currop->op0;
2616 /* Fallthrough. */
2617 case REALPART_EXPR:
2618 case IMAGPART_EXPR:
2619 case VIEW_CONVERT_EXPR:
2621 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2622 stmts);
2623 if (!genop0)
2624 return NULL_TREE;
2625 return fold_build1 (currop->opcode, currop->type, genop0);
2628 case WITH_SIZE_EXPR:
2630 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2631 stmts);
2632 if (!genop0)
2633 return NULL_TREE;
2634 tree genop1 = find_or_generate_expression (block, currop->op0, stmts);
2635 if (!genop1)
2636 return NULL_TREE;
2637 return fold_build2 (currop->opcode, currop->type, genop0, genop1);
2640 case BIT_FIELD_REF:
2642 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2643 stmts);
2644 if (!genop0)
2645 return NULL_TREE;
2646 tree op1 = currop->op0;
2647 tree op2 = currop->op1;
2648 tree t = build3 (BIT_FIELD_REF, currop->type, genop0, op1, op2);
2649 REF_REVERSE_STORAGE_ORDER (t) = currop->reverse;
2650 return fold (t);
2653 /* For array ref vn_reference_op's, operand 1 of the array ref
2654 is op0 of the reference op and operand 3 of the array ref is
2655 op1. */
2656 case ARRAY_RANGE_REF:
2657 case ARRAY_REF:
2659 tree genop0;
2660 tree genop1 = currop->op0;
2661 tree genop2 = currop->op1;
2662 tree genop3 = currop->op2;
2663 genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2664 stmts);
2665 if (!genop0)
2666 return NULL_TREE;
2667 genop1 = find_or_generate_expression (block, genop1, stmts);
2668 if (!genop1)
2669 return NULL_TREE;
2670 if (genop2)
2672 tree domain_type = TYPE_DOMAIN (TREE_TYPE (genop0));
2673 /* Drop zero minimum index if redundant. */
2674 if (integer_zerop (genop2)
2675 && (!domain_type
2676 || integer_zerop (TYPE_MIN_VALUE (domain_type))))
2677 genop2 = NULL_TREE;
2678 else
2680 genop2 = find_or_generate_expression (block, genop2, stmts);
2681 if (!genop2)
2682 return NULL_TREE;
2685 if (genop3)
2687 tree elmt_type = TREE_TYPE (TREE_TYPE (genop0));
2688 /* We can't always put a size in units of the element alignment
2689 here as the element alignment may be not visible. See
2690 PR43783. Simply drop the element size for constant
2691 sizes. */
2692 if (TREE_CODE (genop3) == INTEGER_CST
2693 && TREE_CODE (TYPE_SIZE_UNIT (elmt_type)) == INTEGER_CST
2694 && wi::eq_p (wi::to_offset (TYPE_SIZE_UNIT (elmt_type)),
2695 (wi::to_offset (genop3)
2696 * vn_ref_op_align_unit (currop))))
2697 genop3 = NULL_TREE;
2698 else
2700 genop3 = find_or_generate_expression (block, genop3, stmts);
2701 if (!genop3)
2702 return NULL_TREE;
2705 return build4 (currop->opcode, currop->type, genop0, genop1,
2706 genop2, genop3);
2708 case COMPONENT_REF:
2710 tree op0;
2711 tree op1;
2712 tree genop2 = currop->op1;
2713 op0 = create_component_ref_by_pieces_1 (block, ref, operand, stmts);
2714 if (!op0)
2715 return NULL_TREE;
2716 /* op1 should be a FIELD_DECL, which are represented by themselves. */
2717 op1 = currop->op0;
2718 if (genop2)
2720 genop2 = find_or_generate_expression (block, genop2, stmts);
2721 if (!genop2)
2722 return NULL_TREE;
2724 return fold_build3 (COMPONENT_REF, TREE_TYPE (op1), op0, op1, genop2);
2727 case SSA_NAME:
2729 genop = find_or_generate_expression (block, currop->op0, stmts);
2730 return genop;
2732 case STRING_CST:
2733 case INTEGER_CST:
2734 case POLY_INT_CST:
2735 case COMPLEX_CST:
2736 case VECTOR_CST:
2737 case REAL_CST:
2738 case CONSTRUCTOR:
2739 case VAR_DECL:
2740 case PARM_DECL:
2741 case CONST_DECL:
2742 case RESULT_DECL:
2743 case FUNCTION_DECL:
2744 return currop->op0;
2746 default:
2747 gcc_unreachable ();
2751 /* For COMPONENT_REF's and ARRAY_REF's, we can't have any intermediates for the
2752 COMPONENT_REF or MEM_REF or ARRAY_REF portion, because we'd end up with
2753 trying to rename aggregates into ssa form directly, which is a no no.
2755 Thus, this routine doesn't create temporaries, it just builds a
2756 single access expression for the array, calling
2757 find_or_generate_expression to build the innermost pieces.
2759 This function is a subroutine of create_expression_by_pieces, and
2760 should not be called on it's own unless you really know what you
2761 are doing. */
2763 static tree
2764 create_component_ref_by_pieces (basic_block block, vn_reference_t ref,
2765 gimple_seq *stmts)
2767 unsigned int op = 0;
2768 return create_component_ref_by_pieces_1 (block, ref, &op, stmts);
2771 /* Find a simple leader for an expression, or generate one using
2772 create_expression_by_pieces from a NARY expression for the value.
2773 BLOCK is the basic_block we are looking for leaders in.
2774 OP is the tree expression to find a leader for or generate.
2775 Returns the leader or NULL_TREE on failure. */
2777 static tree
2778 find_or_generate_expression (basic_block block, tree op, gimple_seq *stmts)
2780 pre_expr expr = get_or_alloc_expr_for (op);
2781 unsigned int lookfor = get_expr_value_id (expr);
2782 pre_expr leader = bitmap_find_leader (AVAIL_OUT (block), lookfor);
2783 if (leader)
2785 if (leader->kind == NAME)
2786 return PRE_EXPR_NAME (leader);
2787 else if (leader->kind == CONSTANT)
2788 return PRE_EXPR_CONSTANT (leader);
2790 /* Defer. */
2791 return NULL_TREE;
2794 /* It must be a complex expression, so generate it recursively. Note
2795 that this is only necessary to handle gcc.dg/tree-ssa/ssa-pre28.c
2796 where the insert algorithm fails to insert a required expression. */
2797 bitmap exprset = value_expressions[lookfor];
2798 bitmap_iterator bi;
2799 unsigned int i;
2800 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
2802 pre_expr temp = expression_for_id (i);
2803 /* We cannot insert random REFERENCE expressions at arbitrary
2804 places. We can insert NARYs which eventually re-materializes
2805 its operand values. */
2806 if (temp->kind == NARY)
2807 return create_expression_by_pieces (block, temp, stmts,
2808 get_expr_type (expr));
2811 /* Defer. */
2812 return NULL_TREE;
2815 /* Create an expression in pieces, so that we can handle very complex
2816 expressions that may be ANTIC, but not necessary GIMPLE.
2817 BLOCK is the basic block the expression will be inserted into,
2818 EXPR is the expression to insert (in value form)
2819 STMTS is a statement list to append the necessary insertions into.
2821 This function will die if we hit some value that shouldn't be
2822 ANTIC but is (IE there is no leader for it, or its components).
2823 The function returns NULL_TREE in case a different antic expression
2824 has to be inserted first.
2825 This function may also generate expressions that are themselves
2826 partially or fully redundant. Those that are will be either made
2827 fully redundant during the next iteration of insert (for partially
2828 redundant ones), or eliminated by eliminate (for fully redundant
2829 ones). */
2831 static tree
2832 create_expression_by_pieces (basic_block block, pre_expr expr,
2833 gimple_seq *stmts, tree type)
2835 tree name;
2836 tree folded;
2837 gimple_seq forced_stmts = NULL;
2838 unsigned int value_id;
2839 gimple_stmt_iterator gsi;
2840 tree exprtype = type ? type : get_expr_type (expr);
2841 pre_expr nameexpr;
2842 gassign *newstmt;
2844 switch (expr->kind)
2846 /* We may hit the NAME/CONSTANT case if we have to convert types
2847 that value numbering saw through. */
2848 case NAME:
2849 folded = PRE_EXPR_NAME (expr);
2850 if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (folded))
2851 return NULL_TREE;
2852 if (useless_type_conversion_p (exprtype, TREE_TYPE (folded)))
2853 return folded;
2854 break;
2855 case CONSTANT:
2857 folded = PRE_EXPR_CONSTANT (expr);
2858 tree tem = fold_convert (exprtype, folded);
2859 if (is_gimple_min_invariant (tem))
2860 return tem;
2861 break;
2863 case REFERENCE:
2864 if (PRE_EXPR_REFERENCE (expr)->operands[0].opcode == CALL_EXPR)
2866 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2867 unsigned int operand = 1;
2868 vn_reference_op_t currop = &ref->operands[0];
2869 tree sc = NULL_TREE;
2870 tree fn = NULL_TREE;
2871 if (currop->op0)
2873 fn = find_or_generate_expression (block, currop->op0, stmts);
2874 if (!fn)
2875 return NULL_TREE;
2877 if (currop->op1)
2879 sc = find_or_generate_expression (block, currop->op1, stmts);
2880 if (!sc)
2881 return NULL_TREE;
2883 auto_vec<tree> args (ref->operands.length () - 1);
2884 while (operand < ref->operands.length ())
2886 tree arg = create_component_ref_by_pieces_1 (block, ref,
2887 &operand, stmts);
2888 if (!arg)
2889 return NULL_TREE;
2890 args.quick_push (arg);
2892 gcall *call;
2893 if (currop->op0)
2895 call = gimple_build_call_vec (fn, args);
2896 gimple_call_set_fntype (call, currop->type);
2898 else
2899 call = gimple_build_call_internal_vec ((internal_fn)currop->clique,
2900 args);
2901 gimple_set_location (call, expr->loc);
2902 if (sc)
2903 gimple_call_set_chain (call, sc);
2904 tree forcedname = make_ssa_name (ref->type);
2905 gimple_call_set_lhs (call, forcedname);
2906 /* There's no CCP pass after PRE which would re-compute alignment
2907 information so make sure we re-materialize this here. */
2908 if (gimple_call_builtin_p (call, BUILT_IN_ASSUME_ALIGNED)
2909 && args.length () - 2 <= 1
2910 && tree_fits_uhwi_p (args[1])
2911 && (args.length () != 3 || tree_fits_uhwi_p (args[2])))
2913 unsigned HOST_WIDE_INT halign = tree_to_uhwi (args[1]);
2914 unsigned HOST_WIDE_INT hmisalign
2915 = args.length () == 3 ? tree_to_uhwi (args[2]) : 0;
2916 if ((halign & (halign - 1)) == 0
2917 && (hmisalign & ~(halign - 1)) == 0
2918 && (unsigned int)halign != 0)
2919 set_ptr_info_alignment (get_ptr_info (forcedname),
2920 halign, hmisalign);
2922 gimple_set_vuse (call, BB_LIVE_VOP_ON_EXIT (block));
2923 gimple_seq_add_stmt_without_update (&forced_stmts, call);
2924 folded = forcedname;
2926 else
2928 folded = create_component_ref_by_pieces (block,
2929 PRE_EXPR_REFERENCE (expr),
2930 stmts);
2931 if (!folded)
2932 return NULL_TREE;
2933 name = make_temp_ssa_name (exprtype, NULL, "pretmp");
2934 newstmt = gimple_build_assign (name, folded);
2935 gimple_set_location (newstmt, expr->loc);
2936 gimple_seq_add_stmt_without_update (&forced_stmts, newstmt);
2937 gimple_set_vuse (newstmt, BB_LIVE_VOP_ON_EXIT (block));
2938 folded = name;
2940 break;
2941 case NARY:
2943 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2944 tree *genop = XALLOCAVEC (tree, nary->length);
2945 unsigned i;
2946 for (i = 0; i < nary->length; ++i)
2948 genop[i] = find_or_generate_expression (block, nary->op[i], stmts);
2949 if (!genop[i])
2950 return NULL_TREE;
2951 /* Ensure genop[] is properly typed for POINTER_PLUS_EXPR. It
2952 may have conversions stripped. */
2953 if (nary->opcode == POINTER_PLUS_EXPR)
2955 if (i == 0)
2956 genop[i] = gimple_convert (&forced_stmts,
2957 nary->type, genop[i]);
2958 else if (i == 1)
2959 genop[i] = gimple_convert (&forced_stmts,
2960 sizetype, genop[i]);
2962 else
2963 genop[i] = gimple_convert (&forced_stmts,
2964 TREE_TYPE (nary->op[i]), genop[i]);
2966 if (nary->opcode == CONSTRUCTOR)
2968 vec<constructor_elt, va_gc> *elts = NULL;
2969 for (i = 0; i < nary->length; ++i)
2970 CONSTRUCTOR_APPEND_ELT (elts, NULL_TREE, genop[i]);
2971 folded = build_constructor (nary->type, elts);
2972 name = make_temp_ssa_name (exprtype, NULL, "pretmp");
2973 newstmt = gimple_build_assign (name, folded);
2974 gimple_set_location (newstmt, expr->loc);
2975 gimple_seq_add_stmt_without_update (&forced_stmts, newstmt);
2976 folded = name;
2978 else
2980 switch (nary->length)
2982 case 1:
2983 folded = gimple_build (&forced_stmts, expr->loc,
2984 nary->opcode, nary->type, genop[0]);
2985 break;
2986 case 2:
2987 folded = gimple_build (&forced_stmts, expr->loc, nary->opcode,
2988 nary->type, genop[0], genop[1]);
2989 break;
2990 case 3:
2991 folded = gimple_build (&forced_stmts, expr->loc, nary->opcode,
2992 nary->type, genop[0], genop[1],
2993 genop[2]);
2994 break;
2995 default:
2996 gcc_unreachable ();
3000 break;
3001 default:
3002 gcc_unreachable ();
3005 folded = gimple_convert (&forced_stmts, exprtype, folded);
3007 /* If there is nothing to insert, return the simplified result. */
3008 if (gimple_seq_empty_p (forced_stmts))
3009 return folded;
3010 /* If we simplified to a constant return it and discard eventually
3011 built stmts. */
3012 if (is_gimple_min_invariant (folded))
3014 gimple_seq_discard (forced_stmts);
3015 return folded;
3017 /* Likewise if we simplified to sth not queued for insertion. */
3018 bool found = false;
3019 gsi = gsi_last (forced_stmts);
3020 for (; !gsi_end_p (gsi); gsi_prev (&gsi))
3022 gimple *stmt = gsi_stmt (gsi);
3023 tree forcedname = gimple_get_lhs (stmt);
3024 if (forcedname == folded)
3026 found = true;
3027 break;
3030 if (! found)
3032 gimple_seq_discard (forced_stmts);
3033 return folded;
3035 gcc_assert (TREE_CODE (folded) == SSA_NAME);
3037 /* If we have any intermediate expressions to the value sets, add them
3038 to the value sets and chain them in the instruction stream. */
3039 if (forced_stmts)
3041 gsi = gsi_start (forced_stmts);
3042 for (; !gsi_end_p (gsi); gsi_next (&gsi))
3044 gimple *stmt = gsi_stmt (gsi);
3045 tree forcedname = gimple_get_lhs (stmt);
3046 pre_expr nameexpr;
3048 if (forcedname != folded)
3050 vn_ssa_aux_t vn_info = VN_INFO (forcedname);
3051 vn_info->valnum = forcedname;
3052 vn_info->value_id = get_next_value_id ();
3053 nameexpr = get_or_alloc_expr_for_name (forcedname);
3054 add_to_value (vn_info->value_id, nameexpr);
3055 if (NEW_SETS (block))
3056 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
3057 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
3060 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (forcedname));
3062 gimple_seq_add_seq (stmts, forced_stmts);
3065 name = folded;
3067 /* Fold the last statement. */
3068 gsi = gsi_last (*stmts);
3069 if (fold_stmt_inplace (&gsi))
3070 update_stmt (gsi_stmt (gsi));
3072 /* Add a value number to the temporary.
3073 The value may already exist in either NEW_SETS, or AVAIL_OUT, because
3074 we are creating the expression by pieces, and this particular piece of
3075 the expression may have been represented. There is no harm in replacing
3076 here. */
3077 value_id = get_expr_value_id (expr);
3078 vn_ssa_aux_t vn_info = VN_INFO (name);
3079 vn_info->value_id = value_id;
3080 vn_info->valnum = vn_valnum_from_value_id (value_id);
3081 if (vn_info->valnum == NULL_TREE)
3082 vn_info->valnum = name;
3083 gcc_assert (vn_info->valnum != NULL_TREE);
3084 nameexpr = get_or_alloc_expr_for_name (name);
3085 add_to_value (value_id, nameexpr);
3086 if (NEW_SETS (block))
3087 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
3088 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
3090 pre_stats.insertions++;
3091 if (dump_file && (dump_flags & TDF_DETAILS))
3093 fprintf (dump_file, "Inserted ");
3094 print_gimple_stmt (dump_file, gsi_stmt (gsi_last (*stmts)), 0);
3095 fprintf (dump_file, " in predecessor %d (%04d)\n",
3096 block->index, value_id);
3099 return name;
3103 /* Insert the to-be-made-available values of expression EXPRNUM for each
3104 predecessor, stored in AVAIL, into the predecessors of BLOCK, and
3105 merge the result with a phi node, given the same value number as
3106 NODE. Return true if we have inserted new stuff. */
3108 static bool
3109 insert_into_preds_of_block (basic_block block, unsigned int exprnum,
3110 vec<pre_expr> &avail)
3112 pre_expr expr = expression_for_id (exprnum);
3113 pre_expr newphi;
3114 unsigned int val = get_expr_value_id (expr);
3115 edge pred;
3116 bool insertions = false;
3117 bool nophi = false;
3118 basic_block bprime;
3119 pre_expr eprime;
3120 edge_iterator ei;
3121 tree type = get_expr_type (expr);
3122 tree temp;
3123 gphi *phi;
3125 /* Make sure we aren't creating an induction variable. */
3126 if (bb_loop_depth (block) > 0 && EDGE_COUNT (block->preds) == 2)
3128 bool firstinsideloop = false;
3129 bool secondinsideloop = false;
3130 firstinsideloop = flow_bb_inside_loop_p (block->loop_father,
3131 EDGE_PRED (block, 0)->src);
3132 secondinsideloop = flow_bb_inside_loop_p (block->loop_father,
3133 EDGE_PRED (block, 1)->src);
3134 /* Induction variables only have one edge inside the loop. */
3135 if ((firstinsideloop ^ secondinsideloop)
3136 && expr->kind != REFERENCE)
3138 if (dump_file && (dump_flags & TDF_DETAILS))
3139 fprintf (dump_file, "Skipping insertion of phi for partial "
3140 "redundancy: Looks like an induction variable\n");
3141 nophi = true;
3145 /* Make the necessary insertions. */
3146 FOR_EACH_EDGE (pred, ei, block->preds)
3148 /* When we are not inserting a PHI node do not bother inserting
3149 into places that do not dominate the anticipated computations. */
3150 if (nophi && !dominated_by_p (CDI_DOMINATORS, block, pred->src))
3151 continue;
3152 gimple_seq stmts = NULL;
3153 tree builtexpr;
3154 bprime = pred->src;
3155 eprime = avail[pred->dest_idx];
3156 builtexpr = create_expression_by_pieces (bprime, eprime,
3157 &stmts, type);
3158 gcc_assert (!(pred->flags & EDGE_ABNORMAL));
3159 if (!gimple_seq_empty_p (stmts))
3161 basic_block new_bb = gsi_insert_seq_on_edge_immediate (pred, stmts);
3162 gcc_assert (! new_bb);
3163 insertions = true;
3165 if (!builtexpr)
3167 /* We cannot insert a PHI node if we failed to insert
3168 on one edge. */
3169 nophi = true;
3170 continue;
3172 if (is_gimple_min_invariant (builtexpr))
3173 avail[pred->dest_idx] = get_or_alloc_expr_for_constant (builtexpr);
3174 else
3175 avail[pred->dest_idx] = get_or_alloc_expr_for_name (builtexpr);
3177 /* If we didn't want a phi node, and we made insertions, we still have
3178 inserted new stuff, and thus return true. If we didn't want a phi node,
3179 and didn't make insertions, we haven't added anything new, so return
3180 false. */
3181 if (nophi && insertions)
3182 return true;
3183 else if (nophi && !insertions)
3184 return false;
3186 /* Now build a phi for the new variable. */
3187 temp = make_temp_ssa_name (type, NULL, "prephitmp");
3188 phi = create_phi_node (temp, block);
3190 vn_ssa_aux_t vn_info = VN_INFO (temp);
3191 vn_info->value_id = val;
3192 vn_info->valnum = vn_valnum_from_value_id (val);
3193 if (vn_info->valnum == NULL_TREE)
3194 vn_info->valnum = temp;
3195 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3196 FOR_EACH_EDGE (pred, ei, block->preds)
3198 pre_expr ae = avail[pred->dest_idx];
3199 gcc_assert (get_expr_type (ae) == type
3200 || useless_type_conversion_p (type, get_expr_type (ae)));
3201 if (ae->kind == CONSTANT)
3202 add_phi_arg (phi, unshare_expr (PRE_EXPR_CONSTANT (ae)),
3203 pred, UNKNOWN_LOCATION);
3204 else
3205 add_phi_arg (phi, PRE_EXPR_NAME (ae), pred, UNKNOWN_LOCATION);
3208 newphi = get_or_alloc_expr_for_name (temp);
3209 add_to_value (val, newphi);
3211 /* The value should *not* exist in PHI_GEN, or else we wouldn't be doing
3212 this insertion, since we test for the existence of this value in PHI_GEN
3213 before proceeding with the partial redundancy checks in insert_aux.
3215 The value may exist in AVAIL_OUT, in particular, it could be represented
3216 by the expression we are trying to eliminate, in which case we want the
3217 replacement to occur. If it's not existing in AVAIL_OUT, we want it
3218 inserted there.
3220 Similarly, to the PHI_GEN case, the value should not exist in NEW_SETS of
3221 this block, because if it did, it would have existed in our dominator's
3222 AVAIL_OUT, and would have been skipped due to the full redundancy check.
3225 bitmap_insert_into_set (PHI_GEN (block), newphi);
3226 bitmap_value_replace_in_set (AVAIL_OUT (block),
3227 newphi);
3228 if (NEW_SETS (block))
3229 bitmap_insert_into_set (NEW_SETS (block), newphi);
3231 /* If we insert a PHI node for a conversion of another PHI node
3232 in the same basic-block try to preserve range information.
3233 This is important so that followup loop passes receive optimal
3234 number of iteration analysis results. See PR61743. */
3235 if (expr->kind == NARY
3236 && CONVERT_EXPR_CODE_P (expr->u.nary->opcode)
3237 && TREE_CODE (expr->u.nary->op[0]) == SSA_NAME
3238 && gimple_bb (SSA_NAME_DEF_STMT (expr->u.nary->op[0])) == block
3239 && INTEGRAL_TYPE_P (type)
3240 && INTEGRAL_TYPE_P (TREE_TYPE (expr->u.nary->op[0]))
3241 && (TYPE_PRECISION (type)
3242 >= TYPE_PRECISION (TREE_TYPE (expr->u.nary->op[0])))
3243 && SSA_NAME_RANGE_INFO (expr->u.nary->op[0]))
3245 value_range r;
3246 if (get_range_query (cfun)->range_of_expr (r, expr->u.nary->op[0])
3247 && r.kind () == VR_RANGE
3248 && !wi::neg_p (r.lower_bound (), SIGNED)
3249 && !wi::neg_p (r.upper_bound (), SIGNED))
3250 /* Just handle extension and sign-changes of all-positive ranges. */
3251 set_range_info (temp, VR_RANGE,
3252 wide_int_storage::from (r.lower_bound (),
3253 TYPE_PRECISION (type),
3254 TYPE_SIGN (type)),
3255 wide_int_storage::from (r.upper_bound (),
3256 TYPE_PRECISION (type),
3257 TYPE_SIGN (type)));
3260 if (dump_file && (dump_flags & TDF_DETAILS))
3262 fprintf (dump_file, "Created phi ");
3263 print_gimple_stmt (dump_file, phi, 0);
3264 fprintf (dump_file, " in block %d (%04d)\n", block->index, val);
3266 pre_stats.phis++;
3267 return true;
3272 /* Perform insertion of partially redundant or hoistable values.
3273 For BLOCK, do the following:
3274 1. Propagate the NEW_SETS of the dominator into the current block.
3275 If the block has multiple predecessors,
3276 2a. Iterate over the ANTIC expressions for the block to see if
3277 any of them are partially redundant.
3278 2b. If so, insert them into the necessary predecessors to make
3279 the expression fully redundant.
3280 2c. Insert a new PHI merging the values of the predecessors.
3281 2d. Insert the new PHI, and the new expressions, into the
3282 NEW_SETS set.
3283 If the block has multiple successors,
3284 3a. Iterate over the ANTIC values for the block to see if
3285 any of them are good candidates for hoisting.
3286 3b. If so, insert expressions computing the values in BLOCK,
3287 and add the new expressions into the NEW_SETS set.
3288 4. Recursively call ourselves on the dominator children of BLOCK.
3290 Steps 1, 2a, and 4 are done by insert_aux. 2b, 2c and 2d are done by
3291 do_pre_regular_insertion and do_partial_insertion. 3a and 3b are
3292 done in do_hoist_insertion.
3295 static bool
3296 do_pre_regular_insertion (basic_block block, basic_block dom,
3297 vec<pre_expr> exprs)
3299 bool new_stuff = false;
3300 pre_expr expr;
3301 auto_vec<pre_expr, 2> avail;
3302 int i;
3304 avail.safe_grow (EDGE_COUNT (block->preds), true);
3306 FOR_EACH_VEC_ELT (exprs, i, expr)
3308 if (expr->kind == NARY
3309 || expr->kind == REFERENCE)
3311 unsigned int val;
3312 bool by_some = false;
3313 bool cant_insert = false;
3314 bool all_same = true;
3315 pre_expr first_s = NULL;
3316 edge pred;
3317 basic_block bprime;
3318 pre_expr eprime = NULL;
3319 edge_iterator ei;
3320 pre_expr edoubleprime = NULL;
3321 bool do_insertion = false;
3323 val = get_expr_value_id (expr);
3324 if (bitmap_set_contains_value (PHI_GEN (block), val))
3325 continue;
3326 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3328 if (dump_file && (dump_flags & TDF_DETAILS))
3330 fprintf (dump_file, "Found fully redundant value: ");
3331 print_pre_expr (dump_file, expr);
3332 fprintf (dump_file, "\n");
3334 continue;
3337 FOR_EACH_EDGE (pred, ei, block->preds)
3339 unsigned int vprime;
3341 /* We should never run insertion for the exit block
3342 and so not come across fake pred edges. */
3343 gcc_assert (!(pred->flags & EDGE_FAKE));
3344 bprime = pred->src;
3345 /* We are looking at ANTIC_OUT of bprime. */
3346 eprime = phi_translate (NULL, expr, ANTIC_IN (block), NULL, pred);
3348 /* eprime will generally only be NULL if the
3349 value of the expression, translated
3350 through the PHI for this predecessor, is
3351 undefined. If that is the case, we can't
3352 make the expression fully redundant,
3353 because its value is undefined along a
3354 predecessor path. We can thus break out
3355 early because it doesn't matter what the
3356 rest of the results are. */
3357 if (eprime == NULL)
3359 avail[pred->dest_idx] = NULL;
3360 cant_insert = true;
3361 break;
3364 vprime = get_expr_value_id (eprime);
3365 edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime),
3366 vprime);
3367 if (edoubleprime == NULL)
3369 avail[pred->dest_idx] = eprime;
3370 all_same = false;
3372 else
3374 avail[pred->dest_idx] = edoubleprime;
3375 by_some = true;
3376 /* We want to perform insertions to remove a redundancy on
3377 a path in the CFG we want to optimize for speed. */
3378 if (optimize_edge_for_speed_p (pred))
3379 do_insertion = true;
3380 if (first_s == NULL)
3381 first_s = edoubleprime;
3382 else if (!pre_expr_d::equal (first_s, edoubleprime))
3383 all_same = false;
3386 /* If we can insert it, it's not the same value
3387 already existing along every predecessor, and
3388 it's defined by some predecessor, it is
3389 partially redundant. */
3390 if (!cant_insert && !all_same && by_some)
3392 if (!do_insertion)
3394 if (dump_file && (dump_flags & TDF_DETAILS))
3396 fprintf (dump_file, "Skipping partial redundancy for "
3397 "expression ");
3398 print_pre_expr (dump_file, expr);
3399 fprintf (dump_file, " (%04d), no redundancy on to be "
3400 "optimized for speed edge\n", val);
3403 else if (dbg_cnt (treepre_insert))
3405 if (dump_file && (dump_flags & TDF_DETAILS))
3407 fprintf (dump_file, "Found partial redundancy for "
3408 "expression ");
3409 print_pre_expr (dump_file, expr);
3410 fprintf (dump_file, " (%04d)\n",
3411 get_expr_value_id (expr));
3413 if (insert_into_preds_of_block (block,
3414 get_expression_id (expr),
3415 avail))
3416 new_stuff = true;
3419 /* If all edges produce the same value and that value is
3420 an invariant, then the PHI has the same value on all
3421 edges. Note this. */
3422 else if (!cant_insert
3423 && all_same
3424 && (edoubleprime->kind != NAME
3425 || !SSA_NAME_OCCURS_IN_ABNORMAL_PHI
3426 (PRE_EXPR_NAME (edoubleprime))))
3428 gcc_assert (edoubleprime->kind == CONSTANT
3429 || edoubleprime->kind == NAME);
3431 tree temp = make_temp_ssa_name (get_expr_type (expr),
3432 NULL, "pretmp");
3433 gassign *assign
3434 = gimple_build_assign (temp,
3435 edoubleprime->kind == CONSTANT ?
3436 PRE_EXPR_CONSTANT (edoubleprime) :
3437 PRE_EXPR_NAME (edoubleprime));
3438 gimple_stmt_iterator gsi = gsi_after_labels (block);
3439 gsi_insert_before (&gsi, assign, GSI_NEW_STMT);
3441 vn_ssa_aux_t vn_info = VN_INFO (temp);
3442 vn_info->value_id = val;
3443 vn_info->valnum = vn_valnum_from_value_id (val);
3444 if (vn_info->valnum == NULL_TREE)
3445 vn_info->valnum = temp;
3446 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3447 pre_expr newe = get_or_alloc_expr_for_name (temp);
3448 add_to_value (val, newe);
3449 bitmap_value_replace_in_set (AVAIL_OUT (block), newe);
3450 bitmap_insert_into_set (NEW_SETS (block), newe);
3451 bitmap_insert_into_set (PHI_GEN (block), newe);
3456 return new_stuff;
3460 /* Perform insertion for partially anticipatable expressions. There
3461 is only one case we will perform insertion for these. This case is
3462 if the expression is partially anticipatable, and fully available.
3463 In this case, we know that putting it earlier will enable us to
3464 remove the later computation. */
3466 static bool
3467 do_pre_partial_partial_insertion (basic_block block, basic_block dom,
3468 vec<pre_expr> exprs)
3470 bool new_stuff = false;
3471 pre_expr expr;
3472 auto_vec<pre_expr, 2> avail;
3473 int i;
3475 avail.safe_grow (EDGE_COUNT (block->preds), true);
3477 FOR_EACH_VEC_ELT (exprs, i, expr)
3479 if (expr->kind == NARY
3480 || expr->kind == REFERENCE)
3482 unsigned int val;
3483 bool by_all = true;
3484 bool cant_insert = false;
3485 edge pred;
3486 basic_block bprime;
3487 pre_expr eprime = NULL;
3488 edge_iterator ei;
3490 val = get_expr_value_id (expr);
3491 if (bitmap_set_contains_value (PHI_GEN (block), val))
3492 continue;
3493 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3494 continue;
3496 FOR_EACH_EDGE (pred, ei, block->preds)
3498 unsigned int vprime;
3499 pre_expr edoubleprime;
3501 /* We should never run insertion for the exit block
3502 and so not come across fake pred edges. */
3503 gcc_assert (!(pred->flags & EDGE_FAKE));
3504 bprime = pred->src;
3505 eprime = phi_translate (NULL, expr, ANTIC_IN (block),
3506 PA_IN (block), pred);
3508 /* eprime will generally only be NULL if the
3509 value of the expression, translated
3510 through the PHI for this predecessor, is
3511 undefined. If that is the case, we can't
3512 make the expression fully redundant,
3513 because its value is undefined along a
3514 predecessor path. We can thus break out
3515 early because it doesn't matter what the
3516 rest of the results are. */
3517 if (eprime == NULL)
3519 avail[pred->dest_idx] = NULL;
3520 cant_insert = true;
3521 break;
3524 vprime = get_expr_value_id (eprime);
3525 edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime), vprime);
3526 avail[pred->dest_idx] = edoubleprime;
3527 if (edoubleprime == NULL)
3529 by_all = false;
3530 break;
3534 /* If we can insert it, it's not the same value
3535 already existing along every predecessor, and
3536 it's defined by some predecessor, it is
3537 partially redundant. */
3538 if (!cant_insert && by_all)
3540 edge succ;
3541 bool do_insertion = false;
3543 /* Insert only if we can remove a later expression on a path
3544 that we want to optimize for speed.
3545 The phi node that we will be inserting in BLOCK is not free,
3546 and inserting it for the sake of !optimize_for_speed successor
3547 may cause regressions on the speed path. */
3548 FOR_EACH_EDGE (succ, ei, block->succs)
3550 if (bitmap_set_contains_value (PA_IN (succ->dest), val)
3551 || bitmap_set_contains_value (ANTIC_IN (succ->dest), val))
3553 if (optimize_edge_for_speed_p (succ))
3554 do_insertion = true;
3558 if (!do_insertion)
3560 if (dump_file && (dump_flags & TDF_DETAILS))
3562 fprintf (dump_file, "Skipping partial partial redundancy "
3563 "for expression ");
3564 print_pre_expr (dump_file, expr);
3565 fprintf (dump_file, " (%04d), not (partially) anticipated "
3566 "on any to be optimized for speed edges\n", val);
3569 else if (dbg_cnt (treepre_insert))
3571 pre_stats.pa_insert++;
3572 if (dump_file && (dump_flags & TDF_DETAILS))
3574 fprintf (dump_file, "Found partial partial redundancy "
3575 "for expression ");
3576 print_pre_expr (dump_file, expr);
3577 fprintf (dump_file, " (%04d)\n",
3578 get_expr_value_id (expr));
3580 if (insert_into_preds_of_block (block,
3581 get_expression_id (expr),
3582 avail))
3583 new_stuff = true;
3589 return new_stuff;
3592 /* Insert expressions in BLOCK to compute hoistable values up.
3593 Return TRUE if something was inserted, otherwise return FALSE.
3594 The caller has to make sure that BLOCK has at least two successors. */
3596 static bool
3597 do_hoist_insertion (basic_block block)
3599 edge e;
3600 edge_iterator ei;
3601 bool new_stuff = false;
3602 unsigned i;
3603 gimple_stmt_iterator last;
3605 /* At least two successors, or else... */
3606 gcc_assert (EDGE_COUNT (block->succs) >= 2);
3608 /* Check that all successors of BLOCK are dominated by block.
3609 We could use dominated_by_p() for this, but actually there is a much
3610 quicker check: any successor that is dominated by BLOCK can't have
3611 more than one predecessor edge. */
3612 FOR_EACH_EDGE (e, ei, block->succs)
3613 if (! single_pred_p (e->dest))
3614 return false;
3616 /* Determine the insertion point. If we cannot safely insert before
3617 the last stmt if we'd have to, bail out. */
3618 last = gsi_last_bb (block);
3619 if (!gsi_end_p (last)
3620 && !is_ctrl_stmt (gsi_stmt (last))
3621 && stmt_ends_bb_p (gsi_stmt (last)))
3622 return false;
3624 /* Compute the set of hoistable expressions from ANTIC_IN. First compute
3625 hoistable values. */
3626 bitmap_set hoistable_set;
3628 /* A hoistable value must be in ANTIC_IN(block)
3629 but not in AVAIL_OUT(BLOCK). */
3630 bitmap_initialize (&hoistable_set.values, &grand_bitmap_obstack);
3631 bitmap_and_compl (&hoistable_set.values,
3632 &ANTIC_IN (block)->values, &AVAIL_OUT (block)->values);
3634 /* Short-cut for a common case: hoistable_set is empty. */
3635 if (bitmap_empty_p (&hoistable_set.values))
3636 return false;
3638 /* Compute which of the hoistable values is in AVAIL_OUT of
3639 at least one of the successors of BLOCK. */
3640 bitmap_head availout_in_some;
3641 bitmap_initialize (&availout_in_some, &grand_bitmap_obstack);
3642 FOR_EACH_EDGE (e, ei, block->succs)
3643 /* Do not consider expressions solely because their availability
3644 on loop exits. They'd be ANTIC-IN throughout the whole loop
3645 and thus effectively hoisted across loops by combination of
3646 PRE and hoisting. */
3647 if (! loop_exit_edge_p (block->loop_father, e))
3648 bitmap_ior_and_into (&availout_in_some, &hoistable_set.values,
3649 &AVAIL_OUT (e->dest)->values);
3650 bitmap_clear (&hoistable_set.values);
3652 /* Short-cut for a common case: availout_in_some is empty. */
3653 if (bitmap_empty_p (&availout_in_some))
3654 return false;
3656 /* Hack hoitable_set in-place so we can use sorted_array_from_bitmap_set. */
3657 bitmap_move (&hoistable_set.values, &availout_in_some);
3658 hoistable_set.expressions = ANTIC_IN (block)->expressions;
3660 /* Now finally construct the topological-ordered expression set. */
3661 vec<pre_expr> exprs = sorted_array_from_bitmap_set (&hoistable_set);
3663 bitmap_clear (&hoistable_set.values);
3665 /* If there are candidate values for hoisting, insert expressions
3666 strategically to make the hoistable expressions fully redundant. */
3667 pre_expr expr;
3668 FOR_EACH_VEC_ELT (exprs, i, expr)
3670 /* While we try to sort expressions topologically above the
3671 sorting doesn't work out perfectly. Catch expressions we
3672 already inserted. */
3673 unsigned int value_id = get_expr_value_id (expr);
3674 if (bitmap_set_contains_value (AVAIL_OUT (block), value_id))
3676 if (dump_file && (dump_flags & TDF_DETAILS))
3678 fprintf (dump_file,
3679 "Already inserted expression for ");
3680 print_pre_expr (dump_file, expr);
3681 fprintf (dump_file, " (%04d)\n", value_id);
3683 continue;
3686 /* If we end up with a punned expression representation and this
3687 happens to be a float typed one give up - we can't know for
3688 sure whether all paths perform the floating-point load we are
3689 about to insert and on some targets this can cause correctness
3690 issues. See PR88240. */
3691 if (expr->kind == REFERENCE
3692 && PRE_EXPR_REFERENCE (expr)->punned
3693 && FLOAT_TYPE_P (get_expr_type (expr)))
3694 continue;
3696 /* OK, we should hoist this value. Perform the transformation. */
3697 pre_stats.hoist_insert++;
3698 if (dump_file && (dump_flags & TDF_DETAILS))
3700 fprintf (dump_file,
3701 "Inserting expression in block %d for code hoisting: ",
3702 block->index);
3703 print_pre_expr (dump_file, expr);
3704 fprintf (dump_file, " (%04d)\n", value_id);
3707 gimple_seq stmts = NULL;
3708 tree res = create_expression_by_pieces (block, expr, &stmts,
3709 get_expr_type (expr));
3711 /* Do not return true if expression creation ultimately
3712 did not insert any statements. */
3713 if (gimple_seq_empty_p (stmts))
3714 res = NULL_TREE;
3715 else
3717 if (gsi_end_p (last) || is_ctrl_stmt (gsi_stmt (last)))
3718 gsi_insert_seq_before (&last, stmts, GSI_SAME_STMT);
3719 else
3720 gsi_insert_seq_after (&last, stmts, GSI_NEW_STMT);
3723 /* Make sure to not return true if expression creation ultimately
3724 failed but also make sure to insert any stmts produced as they
3725 are tracked in inserted_exprs. */
3726 if (! res)
3727 continue;
3729 new_stuff = true;
3732 exprs.release ();
3734 return new_stuff;
3737 /* Perform insertion of partially redundant and hoistable values. */
3739 static void
3740 insert (void)
3742 basic_block bb;
3744 FOR_ALL_BB_FN (bb, cfun)
3745 NEW_SETS (bb) = bitmap_set_new ();
3747 int *rpo = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
3748 int *bb_rpo = XNEWVEC (int, last_basic_block_for_fn (cfun) + 1);
3749 int rpo_num = pre_and_rev_post_order_compute (NULL, rpo, false);
3750 for (int i = 0; i < rpo_num; ++i)
3751 bb_rpo[rpo[i]] = i;
3753 int num_iterations = 0;
3754 bool changed;
3757 num_iterations++;
3758 if (dump_file && dump_flags & TDF_DETAILS)
3759 fprintf (dump_file, "Starting insert iteration %d\n", num_iterations);
3761 changed = false;
3762 for (int idx = 0; idx < rpo_num; ++idx)
3764 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[idx]);
3765 basic_block dom = get_immediate_dominator (CDI_DOMINATORS, block);
3766 if (dom)
3768 unsigned i;
3769 bitmap_iterator bi;
3770 bitmap_set_t newset;
3772 /* First, update the AVAIL_OUT set with anything we may have
3773 inserted higher up in the dominator tree. */
3774 newset = NEW_SETS (dom);
3776 /* Note that we need to value_replace both NEW_SETS, and
3777 AVAIL_OUT. For both the case of NEW_SETS, the value may be
3778 represented by some non-simple expression here that we want
3779 to replace it with. */
3780 bool avail_out_changed = false;
3781 FOR_EACH_EXPR_ID_IN_SET (newset, i, bi)
3783 pre_expr expr = expression_for_id (i);
3784 bitmap_value_replace_in_set (NEW_SETS (block), expr);
3785 avail_out_changed
3786 |= bitmap_value_replace_in_set (AVAIL_OUT (block), expr);
3788 /* We need to iterate if AVAIL_OUT of an already processed
3789 block source changed. */
3790 if (avail_out_changed && !changed)
3792 edge_iterator ei;
3793 edge e;
3794 FOR_EACH_EDGE (e, ei, block->succs)
3795 if (e->dest->index != EXIT_BLOCK
3796 && bb_rpo[e->dest->index] < idx)
3797 changed = true;
3800 /* Insert expressions for partial redundancies. */
3801 if (flag_tree_pre && !single_pred_p (block))
3803 vec<pre_expr> exprs
3804 = sorted_array_from_bitmap_set (ANTIC_IN (block));
3805 /* Sorting is not perfect, iterate locally. */
3806 while (do_pre_regular_insertion (block, dom, exprs))
3808 exprs.release ();
3809 if (do_partial_partial)
3811 exprs = sorted_array_from_bitmap_set (PA_IN (block));
3812 while (do_pre_partial_partial_insertion (block, dom,
3813 exprs))
3815 exprs.release ();
3821 /* Clear the NEW sets before the next iteration. We have already
3822 fully propagated its contents. */
3823 if (changed)
3824 FOR_ALL_BB_FN (bb, cfun)
3825 bitmap_set_free (NEW_SETS (bb));
3827 while (changed);
3829 statistics_histogram_event (cfun, "insert iterations", num_iterations);
3831 /* AVAIL_OUT is not needed after insertion so we don't have to
3832 propagate NEW_SETS from hoist insertion. */
3833 FOR_ALL_BB_FN (bb, cfun)
3835 bitmap_set_free (NEW_SETS (bb));
3836 bitmap_set_pool.remove (NEW_SETS (bb));
3837 NEW_SETS (bb) = NULL;
3840 /* Insert expressions for hoisting. Do a backward walk here since
3841 inserting into BLOCK exposes new opportunities in its predecessors.
3842 Since PRE and hoist insertions can cause back-to-back iteration
3843 and we are interested in PRE insertion exposed hoisting opportunities
3844 but not in hoisting exposed PRE ones do hoist insertion only after
3845 PRE insertion iteration finished and do not iterate it. */
3846 if (flag_code_hoisting)
3847 for (int idx = rpo_num - 1; idx >= 0; --idx)
3849 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[idx]);
3850 if (EDGE_COUNT (block->succs) >= 2)
3851 changed |= do_hoist_insertion (block);
3854 free (rpo);
3855 free (bb_rpo);
3859 /* Compute the AVAIL set for all basic blocks.
3861 This function performs value numbering of the statements in each basic
3862 block. The AVAIL sets are built from information we glean while doing
3863 this value numbering, since the AVAIL sets contain only one entry per
3864 value.
3866 AVAIL_IN[BLOCK] = AVAIL_OUT[dom(BLOCK)].
3867 AVAIL_OUT[BLOCK] = AVAIL_IN[BLOCK] U PHI_GEN[BLOCK] U TMP_GEN[BLOCK]. */
3869 static void
3870 compute_avail (function *fun)
3873 basic_block block, son;
3874 basic_block *worklist;
3875 size_t sp = 0;
3876 unsigned i;
3877 tree name;
3879 /* We pretend that default definitions are defined in the entry block.
3880 This includes function arguments and the static chain decl. */
3881 FOR_EACH_SSA_NAME (i, name, fun)
3883 pre_expr e;
3884 if (!SSA_NAME_IS_DEFAULT_DEF (name)
3885 || has_zero_uses (name)
3886 || virtual_operand_p (name))
3887 continue;
3889 e = get_or_alloc_expr_for_name (name);
3890 add_to_value (get_expr_value_id (e), e);
3891 bitmap_insert_into_set (TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (fun)), e);
3892 bitmap_value_insert_into_set (AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3896 if (dump_file && (dump_flags & TDF_DETAILS))
3898 print_bitmap_set (dump_file, TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3899 "tmp_gen", ENTRY_BLOCK);
3900 print_bitmap_set (dump_file, AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3901 "avail_out", ENTRY_BLOCK);
3904 /* Allocate the worklist. */
3905 worklist = XNEWVEC (basic_block, n_basic_blocks_for_fn (fun));
3907 /* Seed the algorithm by putting the dominator children of the entry
3908 block on the worklist. */
3909 for (son = first_dom_son (CDI_DOMINATORS, ENTRY_BLOCK_PTR_FOR_FN (fun));
3910 son;
3911 son = next_dom_son (CDI_DOMINATORS, son))
3912 worklist[sp++] = son;
3914 BB_LIVE_VOP_ON_EXIT (ENTRY_BLOCK_PTR_FOR_FN (fun))
3915 = ssa_default_def (fun, gimple_vop (fun));
3917 /* Loop until the worklist is empty. */
3918 while (sp)
3920 gimple *stmt;
3921 basic_block dom;
3923 /* Pick a block from the worklist. */
3924 block = worklist[--sp];
3925 vn_context_bb = block;
3927 /* Initially, the set of available values in BLOCK is that of
3928 its immediate dominator. */
3929 dom = get_immediate_dominator (CDI_DOMINATORS, block);
3930 if (dom)
3932 bitmap_set_copy (AVAIL_OUT (block), AVAIL_OUT (dom));
3933 BB_LIVE_VOP_ON_EXIT (block) = BB_LIVE_VOP_ON_EXIT (dom);
3936 /* Generate values for PHI nodes. */
3937 for (gphi_iterator gsi = gsi_start_phis (block); !gsi_end_p (gsi);
3938 gsi_next (&gsi))
3940 tree result = gimple_phi_result (gsi.phi ());
3942 /* We have no need for virtual phis, as they don't represent
3943 actual computations. */
3944 if (virtual_operand_p (result))
3946 BB_LIVE_VOP_ON_EXIT (block) = result;
3947 continue;
3950 pre_expr e = get_or_alloc_expr_for_name (result);
3951 add_to_value (get_expr_value_id (e), e);
3952 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
3953 bitmap_insert_into_set (PHI_GEN (block), e);
3956 BB_MAY_NOTRETURN (block) = 0;
3958 /* Now compute value numbers and populate value sets with all
3959 the expressions computed in BLOCK. */
3960 bool set_bb_may_notreturn = false;
3961 for (gimple_stmt_iterator gsi = gsi_start_bb (block); !gsi_end_p (gsi);
3962 gsi_next (&gsi))
3964 ssa_op_iter iter;
3965 tree op;
3967 stmt = gsi_stmt (gsi);
3969 if (set_bb_may_notreturn)
3971 BB_MAY_NOTRETURN (block) = 1;
3972 set_bb_may_notreturn = false;
3975 /* Cache whether the basic-block has any non-visible side-effect
3976 or control flow.
3977 If this isn't a call or it is the last stmt in the
3978 basic-block then the CFG represents things correctly. */
3979 if (is_gimple_call (stmt) && !stmt_ends_bb_p (stmt))
3981 /* Non-looping const functions always return normally.
3982 Otherwise the call might not return or have side-effects
3983 that forbids hoisting possibly trapping expressions
3984 before it. */
3985 int flags = gimple_call_flags (stmt);
3986 if (!(flags & (ECF_CONST|ECF_PURE))
3987 || (flags & ECF_LOOPING_CONST_OR_PURE)
3988 || stmt_can_throw_external (fun, stmt))
3989 /* Defer setting of BB_MAY_NOTRETURN to avoid it
3990 influencing the processing of the call itself. */
3991 set_bb_may_notreturn = true;
3994 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_DEF)
3996 pre_expr e = get_or_alloc_expr_for_name (op);
3998 add_to_value (get_expr_value_id (e), e);
3999 bitmap_insert_into_set (TMP_GEN (block), e);
4000 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
4003 if (gimple_vdef (stmt))
4004 BB_LIVE_VOP_ON_EXIT (block) = gimple_vdef (stmt);
4006 if (gimple_has_side_effects (stmt)
4007 || stmt_could_throw_p (fun, stmt)
4008 || is_gimple_debug (stmt))
4009 continue;
4011 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_USE)
4013 if (ssa_undefined_value_p (op))
4014 continue;
4015 pre_expr e = get_or_alloc_expr_for_name (op);
4016 bitmap_value_insert_into_set (EXP_GEN (block), e);
4019 switch (gimple_code (stmt))
4021 case GIMPLE_RETURN:
4022 continue;
4024 case GIMPLE_CALL:
4026 vn_reference_t ref;
4027 vn_reference_s ref1;
4028 pre_expr result = NULL;
4030 vn_reference_lookup_call (as_a <gcall *> (stmt), &ref, &ref1);
4031 /* There is no point to PRE a call without a value. */
4032 if (!ref || !ref->result)
4033 continue;
4035 /* If the value of the call is not invalidated in
4036 this block until it is computed, add the expression
4037 to EXP_GEN. */
4038 if ((!gimple_vuse (stmt)
4039 || gimple_code
4040 (SSA_NAME_DEF_STMT (gimple_vuse (stmt))) == GIMPLE_PHI
4041 || gimple_bb (SSA_NAME_DEF_STMT
4042 (gimple_vuse (stmt))) != block)
4043 /* If the REFERENCE traps and there was a preceding
4044 point in the block that might not return avoid
4045 adding the reference to EXP_GEN. */
4046 && (!BB_MAY_NOTRETURN (block)
4047 || !vn_reference_may_trap (ref)))
4049 result = get_or_alloc_expr_for_reference
4050 (ref, gimple_location (stmt));
4051 add_to_value (get_expr_value_id (result), result);
4052 bitmap_value_insert_into_set (EXP_GEN (block), result);
4054 continue;
4057 case GIMPLE_ASSIGN:
4059 pre_expr result = NULL;
4060 switch (vn_get_stmt_kind (stmt))
4062 case VN_NARY:
4064 enum tree_code code = gimple_assign_rhs_code (stmt);
4065 vn_nary_op_t nary;
4067 /* COND_EXPR is awkward in that it contains an
4068 embedded complex expression.
4069 Don't even try to shove it through PRE. */
4070 if (code == COND_EXPR)
4071 continue;
4073 vn_nary_op_lookup_stmt (stmt, &nary);
4074 if (!nary || nary->predicated_values)
4075 continue;
4077 /* If the NARY traps and there was a preceding
4078 point in the block that might not return avoid
4079 adding the nary to EXP_GEN. */
4080 if (BB_MAY_NOTRETURN (block)
4081 && vn_nary_may_trap (nary))
4082 continue;
4084 result = get_or_alloc_expr_for_nary
4085 (nary, gimple_location (stmt));
4086 break;
4089 case VN_REFERENCE:
4091 tree rhs1 = gimple_assign_rhs1 (stmt);
4092 ao_ref rhs1_ref;
4093 ao_ref_init (&rhs1_ref, rhs1);
4094 alias_set_type set = ao_ref_alias_set (&rhs1_ref);
4095 alias_set_type base_set
4096 = ao_ref_base_alias_set (&rhs1_ref);
4097 vec<vn_reference_op_s> operands
4098 = vn_reference_operands_for_lookup (rhs1);
4099 vn_reference_t ref;
4100 vn_reference_lookup_pieces (gimple_vuse (stmt), set,
4101 base_set, TREE_TYPE (rhs1),
4102 operands, &ref, VN_WALK);
4103 if (!ref)
4105 operands.release ();
4106 continue;
4109 /* If the REFERENCE traps and there was a preceding
4110 point in the block that might not return avoid
4111 adding the reference to EXP_GEN. */
4112 if (BB_MAY_NOTRETURN (block)
4113 && vn_reference_may_trap (ref))
4115 operands.release ();
4116 continue;
4119 /* If the value of the reference is not invalidated in
4120 this block until it is computed, add the expression
4121 to EXP_GEN. */
4122 if (gimple_vuse (stmt))
4124 gimple *def_stmt;
4125 bool ok = true;
4126 def_stmt = SSA_NAME_DEF_STMT (gimple_vuse (stmt));
4127 while (!gimple_nop_p (def_stmt)
4128 && gimple_code (def_stmt) != GIMPLE_PHI
4129 && gimple_bb (def_stmt) == block)
4131 if (stmt_may_clobber_ref_p
4132 (def_stmt, gimple_assign_rhs1 (stmt)))
4134 ok = false;
4135 break;
4137 def_stmt
4138 = SSA_NAME_DEF_STMT (gimple_vuse (def_stmt));
4140 if (!ok)
4142 operands.release ();
4143 continue;
4147 /* If the load was value-numbered to another
4148 load make sure we do not use its expression
4149 for insertion if it wouldn't be a valid
4150 replacement. */
4151 /* At the momemt we have a testcase
4152 for hoist insertion of aligned vs. misaligned
4153 variants in gcc.dg/torture/pr65270-1.c thus
4154 with just alignment to be considered we can
4155 simply replace the expression in the hashtable
4156 with the most conservative one. */
4157 vn_reference_op_t ref1 = &ref->operands.last ();
4158 while (ref1->opcode != TARGET_MEM_REF
4159 && ref1->opcode != MEM_REF
4160 && ref1 != &ref->operands[0])
4161 --ref1;
4162 vn_reference_op_t ref2 = &operands.last ();
4163 while (ref2->opcode != TARGET_MEM_REF
4164 && ref2->opcode != MEM_REF
4165 && ref2 != &operands[0])
4166 --ref2;
4167 if ((ref1->opcode == TARGET_MEM_REF
4168 || ref1->opcode == MEM_REF)
4169 && (TYPE_ALIGN (ref1->type)
4170 > TYPE_ALIGN (ref2->type)))
4171 ref1->type
4172 = build_aligned_type (ref1->type,
4173 TYPE_ALIGN (ref2->type));
4174 /* TBAA behavior is an obvious part so make sure
4175 that the hashtable one covers this as well
4176 by adjusting the ref alias set and its base. */
4177 if (ref->set == set
4178 || alias_set_subset_of (set, ref->set))
4180 else if (ref1->opcode != ref2->opcode
4181 || (ref1->opcode != MEM_REF
4182 && ref1->opcode != TARGET_MEM_REF))
4184 /* With mismatching base opcodes or bases
4185 other than MEM_REF or TARGET_MEM_REF we
4186 can't do any easy TBAA adjustment. */
4187 operands.release ();
4188 continue;
4190 else if (alias_set_subset_of (ref->set, set))
4192 ref->set = set;
4193 if (ref1->opcode == MEM_REF)
4194 ref1->op0
4195 = wide_int_to_tree (TREE_TYPE (ref2->op0),
4196 wi::to_wide (ref1->op0));
4197 else
4198 ref1->op2
4199 = wide_int_to_tree (TREE_TYPE (ref2->op2),
4200 wi::to_wide (ref1->op2));
4202 else
4204 ref->set = 0;
4205 if (ref1->opcode == MEM_REF)
4206 ref1->op0
4207 = wide_int_to_tree (ptr_type_node,
4208 wi::to_wide (ref1->op0));
4209 else
4210 ref1->op2
4211 = wide_int_to_tree (ptr_type_node,
4212 wi::to_wide (ref1->op2));
4214 operands.release ();
4216 result = get_or_alloc_expr_for_reference
4217 (ref, gimple_location (stmt));
4218 break;
4221 default:
4222 continue;
4225 add_to_value (get_expr_value_id (result), result);
4226 bitmap_value_insert_into_set (EXP_GEN (block), result);
4227 continue;
4229 default:
4230 break;
4233 if (set_bb_may_notreturn)
4235 BB_MAY_NOTRETURN (block) = 1;
4236 set_bb_may_notreturn = false;
4239 if (dump_file && (dump_flags & TDF_DETAILS))
4241 print_bitmap_set (dump_file, EXP_GEN (block),
4242 "exp_gen", block->index);
4243 print_bitmap_set (dump_file, PHI_GEN (block),
4244 "phi_gen", block->index);
4245 print_bitmap_set (dump_file, TMP_GEN (block),
4246 "tmp_gen", block->index);
4247 print_bitmap_set (dump_file, AVAIL_OUT (block),
4248 "avail_out", block->index);
4251 /* Put the dominator children of BLOCK on the worklist of blocks
4252 to compute available sets for. */
4253 for (son = first_dom_son (CDI_DOMINATORS, block);
4254 son;
4255 son = next_dom_son (CDI_DOMINATORS, son))
4256 worklist[sp++] = son;
4258 vn_context_bb = NULL;
4260 free (worklist);
4264 /* Initialize data structures used by PRE. */
4266 static void
4267 init_pre (void)
4269 basic_block bb;
4271 next_expression_id = 1;
4272 expressions.create (0);
4273 expressions.safe_push (NULL);
4274 value_expressions.create (get_max_value_id () + 1);
4275 value_expressions.quick_grow_cleared (get_max_value_id () + 1);
4276 constant_value_expressions.create (get_max_constant_value_id () + 1);
4277 constant_value_expressions.quick_grow_cleared (get_max_constant_value_id () + 1);
4278 name_to_id.create (0);
4280 inserted_exprs = BITMAP_ALLOC (NULL);
4282 connect_infinite_loops_to_exit ();
4283 memset (&pre_stats, 0, sizeof (pre_stats));
4285 alloc_aux_for_blocks (sizeof (struct bb_bitmap_sets));
4287 calculate_dominance_info (CDI_DOMINATORS);
4289 bitmap_obstack_initialize (&grand_bitmap_obstack);
4290 expression_to_id = new hash_table<pre_expr_d> (num_ssa_names * 3);
4291 FOR_ALL_BB_FN (bb, cfun)
4293 EXP_GEN (bb) = bitmap_set_new ();
4294 PHI_GEN (bb) = bitmap_set_new ();
4295 TMP_GEN (bb) = bitmap_set_new ();
4296 AVAIL_OUT (bb) = bitmap_set_new ();
4297 PHI_TRANS_TABLE (bb) = NULL;
4302 /* Deallocate data structures used by PRE. */
4304 static void
4305 fini_pre ()
4307 value_expressions.release ();
4308 constant_value_expressions.release ();
4309 expressions.release ();
4310 BITMAP_FREE (inserted_exprs);
4311 bitmap_obstack_release (&grand_bitmap_obstack);
4312 bitmap_set_pool.release ();
4313 pre_expr_pool.release ();
4314 delete expression_to_id;
4315 expression_to_id = NULL;
4316 name_to_id.release ();
4318 basic_block bb;
4319 FOR_ALL_BB_FN (bb, cfun)
4320 if (bb->aux && PHI_TRANS_TABLE (bb))
4321 delete PHI_TRANS_TABLE (bb);
4322 free_aux_for_blocks ();
4325 namespace {
4327 const pass_data pass_data_pre =
4329 GIMPLE_PASS, /* type */
4330 "pre", /* name */
4331 OPTGROUP_NONE, /* optinfo_flags */
4332 TV_TREE_PRE, /* tv_id */
4333 ( PROP_cfg | PROP_ssa ), /* properties_required */
4334 0, /* properties_provided */
4335 0, /* properties_destroyed */
4336 TODO_rebuild_alias, /* todo_flags_start */
4337 0, /* todo_flags_finish */
4340 class pass_pre : public gimple_opt_pass
4342 public:
4343 pass_pre (gcc::context *ctxt)
4344 : gimple_opt_pass (pass_data_pre, ctxt)
4347 /* opt_pass methods: */
4348 virtual bool gate (function *)
4349 { return flag_tree_pre != 0 || flag_code_hoisting != 0; }
4350 virtual unsigned int execute (function *);
4352 }; // class pass_pre
4354 /* Valueization hook for RPO VN when we are calling back to it
4355 at ANTIC compute time. */
4357 static tree
4358 pre_valueize (tree name)
4360 if (TREE_CODE (name) == SSA_NAME)
4362 tree tem = VN_INFO (name)->valnum;
4363 if (tem != VN_TOP && tem != name)
4365 if (TREE_CODE (tem) != SSA_NAME
4366 || SSA_NAME_IS_DEFAULT_DEF (tem))
4367 return tem;
4368 /* We create temporary SSA names for representatives that
4369 do not have a definition (yet) but are not default defs either
4370 assume they are fine to use. */
4371 basic_block def_bb = gimple_bb (SSA_NAME_DEF_STMT (tem));
4372 if (! def_bb
4373 || dominated_by_p (CDI_DOMINATORS, vn_context_bb, def_bb))
4374 return tem;
4375 /* ??? Now we could look for a leader. Ideally we'd somehow
4376 expose RPO VN leaders and get rid of AVAIL_OUT as well... */
4379 return name;
4382 unsigned int
4383 pass_pre::execute (function *fun)
4385 unsigned int todo = 0;
4387 do_partial_partial =
4388 flag_tree_partial_pre && optimize_function_for_speed_p (fun);
4390 /* This has to happen before VN runs because
4391 loop_optimizer_init may create new phis, etc. */
4392 loop_optimizer_init (LOOPS_NORMAL);
4393 split_edges_for_insertion ();
4394 scev_initialize ();
4395 calculate_dominance_info (CDI_DOMINATORS);
4397 run_rpo_vn (VN_WALK);
4399 init_pre ();
4401 vn_valueize = pre_valueize;
4403 /* Insert can get quite slow on an incredibly large number of basic
4404 blocks due to some quadratic behavior. Until this behavior is
4405 fixed, don't run it when he have an incredibly large number of
4406 bb's. If we aren't going to run insert, there is no point in
4407 computing ANTIC, either, even though it's plenty fast nor do
4408 we require AVAIL. */
4409 if (n_basic_blocks_for_fn (fun) < 4000)
4411 compute_avail (fun);
4412 compute_antic ();
4413 insert ();
4416 /* Make sure to remove fake edges before committing our inserts.
4417 This makes sure we don't end up with extra critical edges that
4418 we would need to split. */
4419 remove_fake_exit_edges ();
4420 gsi_commit_edge_inserts ();
4422 /* Eliminate folds statements which might (should not...) end up
4423 not keeping virtual operands up-to-date. */
4424 gcc_assert (!need_ssa_update_p (fun));
4426 statistics_counter_event (fun, "Insertions", pre_stats.insertions);
4427 statistics_counter_event (fun, "PA inserted", pre_stats.pa_insert);
4428 statistics_counter_event (fun, "HOIST inserted", pre_stats.hoist_insert);
4429 statistics_counter_event (fun, "New PHIs", pre_stats.phis);
4431 todo |= eliminate_with_rpo_vn (inserted_exprs);
4433 vn_valueize = NULL;
4435 /* Because we don't follow exactly the standard PRE algorithm, and decide not
4436 to insert PHI nodes sometimes, and because value numbering of casts isn't
4437 perfect, we sometimes end up inserting dead code. This simple DCE-like
4438 pass removes any insertions we made that weren't actually used. */
4439 simple_dce_from_worklist (inserted_exprs);
4441 fini_pre ();
4443 scev_finalize ();
4444 loop_optimizer_finalize ();
4446 /* TODO: tail_merge_optimize may merge all predecessors of a block, in which
4447 case we can merge the block with the remaining predecessor of the block.
4448 It should either:
4449 - call merge_blocks after each tail merge iteration
4450 - call merge_blocks after all tail merge iterations
4451 - mark TODO_cleanup_cfg when necessary
4452 - share the cfg cleanup with fini_pre. */
4453 todo |= tail_merge_optimize (todo);
4455 free_rpo_vn ();
4457 /* Tail merging invalidates the virtual SSA web, together with
4458 cfg-cleanup opportunities exposed by PRE this will wreck the
4459 SSA updating machinery. So make sure to run update-ssa
4460 manually, before eventually scheduling cfg-cleanup as part of
4461 the todo. */
4462 update_ssa (TODO_update_ssa_only_virtuals);
4464 return todo;
4467 } // anon namespace
4469 gimple_opt_pass *
4470 make_pass_pre (gcc::context *ctxt)
4472 return new pass_pre (ctxt);