testsuite: -mbig/-mlittle only is valid for powerpc-linux.
[official-gcc.git] / gcc / tree-ssa-pre.cc
blob7c2a52e18506b68696b0ad91219efdf823c7d189
1 /* Full and partial redundancy elimination and code hoisting on SSA GIMPLE.
2 Copyright (C) 2001-2022 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.cc, 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.cc.
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;
1239 /* Translate the VUSE backwards through phi nodes in E->dest, so that
1240 it has the value it would have in E->src. Set *SAME_VALID to true
1241 in case the new vuse doesn't change the value id of the OPERANDS. */
1243 static tree
1244 translate_vuse_through_block (vec<vn_reference_op_s> operands,
1245 alias_set_type set, alias_set_type base_set,
1246 tree type, tree vuse, edge e, bool *same_valid)
1248 basic_block phiblock = e->dest;
1249 gimple *phi = SSA_NAME_DEF_STMT (vuse);
1250 ao_ref ref;
1252 if (same_valid)
1253 *same_valid = true;
1255 if (gimple_bb (phi) != phiblock)
1256 return vuse;
1258 /* We have pruned expressions that are killed in PHIBLOCK via
1259 prune_clobbered_mems but we have not rewritten the VUSE to the one
1260 live at the start of the block. If there is no virtual PHI to translate
1261 through return the VUSE live at entry. Otherwise the VUSE to translate
1262 is the def of the virtual PHI node. */
1263 phi = get_virtual_phi (phiblock);
1264 if (!phi)
1265 return BB_LIVE_VOP_ON_EXIT
1266 (get_immediate_dominator (CDI_DOMINATORS, phiblock));
1268 if (same_valid
1269 && ao_ref_init_from_vn_reference (&ref, set, base_set, type, operands))
1271 bitmap visited = NULL;
1272 /* Try to find a vuse that dominates this phi node by skipping
1273 non-clobbering statements. */
1274 unsigned int cnt = param_sccvn_max_alias_queries_per_access;
1275 vuse = get_continuation_for_phi (phi, &ref, true,
1276 cnt, &visited, false, NULL, NULL);
1277 if (visited)
1278 BITMAP_FREE (visited);
1280 else
1281 vuse = NULL_TREE;
1282 /* If we didn't find any, the value ID can't stay the same. */
1283 if (!vuse && same_valid)
1284 *same_valid = false;
1286 /* ??? We would like to return vuse here as this is the canonical
1287 upmost vdef that this reference is associated with. But during
1288 insertion of the references into the hash tables we only ever
1289 directly insert with their direct gimple_vuse, hence returning
1290 something else would make us not find the other expression. */
1291 return PHI_ARG_DEF (phi, e->dest_idx);
1294 /* Like bitmap_find_leader, but checks for the value existing in SET1 *or*
1295 SET2 *or* SET3. This is used to avoid making a set consisting of the union
1296 of PA_IN and ANTIC_IN during insert and phi-translation. */
1298 static inline pre_expr
1299 find_leader_in_sets (unsigned int val, bitmap_set_t set1, bitmap_set_t set2,
1300 bitmap_set_t set3 = NULL)
1302 pre_expr result = NULL;
1304 if (set1)
1305 result = bitmap_find_leader (set1, val);
1306 if (!result && set2)
1307 result = bitmap_find_leader (set2, val);
1308 if (!result && set3)
1309 result = bitmap_find_leader (set3, val);
1310 return result;
1313 /* Get the tree type for our PRE expression e. */
1315 static tree
1316 get_expr_type (const pre_expr e)
1318 switch (e->kind)
1320 case NAME:
1321 return TREE_TYPE (PRE_EXPR_NAME (e));
1322 case CONSTANT:
1323 return TREE_TYPE (PRE_EXPR_CONSTANT (e));
1324 case REFERENCE:
1325 return PRE_EXPR_REFERENCE (e)->type;
1326 case NARY:
1327 return PRE_EXPR_NARY (e)->type;
1329 gcc_unreachable ();
1332 /* Get a representative SSA_NAME for a given expression that is available in B.
1333 Since all of our sub-expressions are treated as values, we require
1334 them to be SSA_NAME's for simplicity.
1335 Prior versions of GVNPRE used to use "value handles" here, so that
1336 an expression would be VH.11 + VH.10 instead of d_3 + e_6. In
1337 either case, the operands are really values (IE we do not expect
1338 them to be usable without finding leaders). */
1340 static tree
1341 get_representative_for (const pre_expr e, basic_block b = NULL)
1343 tree name, valnum = NULL_TREE;
1344 unsigned int value_id = get_expr_value_id (e);
1346 switch (e->kind)
1348 case NAME:
1349 return PRE_EXPR_NAME (e);
1350 case CONSTANT:
1351 return PRE_EXPR_CONSTANT (e);
1352 case NARY:
1353 case REFERENCE:
1355 /* Go through all of the expressions representing this value
1356 and pick out an SSA_NAME. */
1357 unsigned int i;
1358 bitmap_iterator bi;
1359 bitmap exprs = value_expressions[value_id];
1360 EXECUTE_IF_SET_IN_BITMAP (exprs, 0, i, bi)
1362 pre_expr rep = expression_for_id (i);
1363 if (rep->kind == NAME)
1365 tree name = PRE_EXPR_NAME (rep);
1366 valnum = VN_INFO (name)->valnum;
1367 gimple *def = SSA_NAME_DEF_STMT (name);
1368 /* We have to return either a new representative or one
1369 that can be used for expression simplification and thus
1370 is available in B. */
1371 if (! b
1372 || gimple_nop_p (def)
1373 || dominated_by_p (CDI_DOMINATORS, b, gimple_bb (def)))
1374 return name;
1376 else if (rep->kind == CONSTANT)
1377 return PRE_EXPR_CONSTANT (rep);
1380 break;
1383 /* If we reached here we couldn't find an SSA_NAME. This can
1384 happen when we've discovered a value that has never appeared in
1385 the program as set to an SSA_NAME, as the result of phi translation.
1386 Create one here.
1387 ??? We should be able to re-use this when we insert the statement
1388 to compute it. */
1389 name = make_temp_ssa_name (get_expr_type (e), gimple_build_nop (), "pretmp");
1390 vn_ssa_aux_t vn_info = VN_INFO (name);
1391 vn_info->value_id = value_id;
1392 vn_info->valnum = valnum ? valnum : name;
1393 vn_info->visited = true;
1394 /* ??? For now mark this SSA name for release by VN. */
1395 vn_info->needs_insertion = true;
1396 add_to_value (value_id, get_or_alloc_expr_for_name (name));
1397 if (dump_file && (dump_flags & TDF_DETAILS))
1399 fprintf (dump_file, "Created SSA_NAME representative ");
1400 print_generic_expr (dump_file, name);
1401 fprintf (dump_file, " for expression:");
1402 print_pre_expr (dump_file, e);
1403 fprintf (dump_file, " (%04d)\n", value_id);
1406 return name;
1410 static pre_expr
1411 phi_translate (bitmap_set_t, pre_expr, bitmap_set_t, bitmap_set_t, edge);
1413 /* Translate EXPR using phis in PHIBLOCK, so that it has the values of
1414 the phis in PRED. Return NULL if we can't find a leader for each part
1415 of the translated expression. */
1417 static pre_expr
1418 phi_translate_1 (bitmap_set_t dest,
1419 pre_expr expr, bitmap_set_t set1, bitmap_set_t set2, edge e)
1421 basic_block pred = e->src;
1422 basic_block phiblock = e->dest;
1423 location_t expr_loc = expr->loc;
1424 switch (expr->kind)
1426 case NARY:
1428 unsigned int i;
1429 bool changed = false;
1430 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1431 vn_nary_op_t newnary = XALLOCAVAR (struct vn_nary_op_s,
1432 sizeof_vn_nary_op (nary->length));
1433 memcpy (newnary, nary, sizeof_vn_nary_op (nary->length));
1435 for (i = 0; i < newnary->length; i++)
1437 if (TREE_CODE (newnary->op[i]) != SSA_NAME)
1438 continue;
1439 else
1441 pre_expr leader, result;
1442 unsigned int op_val_id = VN_INFO (newnary->op[i])->value_id;
1443 leader = find_leader_in_sets (op_val_id, set1, set2);
1444 result = phi_translate (dest, leader, set1, set2, e);
1445 if (result && result != leader)
1446 /* If op has a leader in the sets we translate make
1447 sure to use the value of the translated expression.
1448 We might need a new representative for that. */
1449 newnary->op[i] = get_representative_for (result, pred);
1450 else if (!result)
1451 return NULL;
1453 changed |= newnary->op[i] != nary->op[i];
1456 if (changed)
1458 pre_expr constant;
1459 unsigned int new_val_id;
1461 PRE_EXPR_NARY (expr) = newnary;
1462 constant = fully_constant_expression (expr);
1463 PRE_EXPR_NARY (expr) = nary;
1464 if (constant != expr)
1466 /* For non-CONSTANTs we have to make sure we can eventually
1467 insert the expression. Which means we need to have a
1468 leader for it. */
1469 if (constant->kind != CONSTANT)
1471 /* Do not allow simplifications to non-constants over
1472 backedges as this will likely result in a loop PHI node
1473 to be inserted and increased register pressure.
1474 See PR77498 - this avoids doing predcoms work in
1475 a less efficient way. */
1476 if (e->flags & EDGE_DFS_BACK)
1478 else
1480 unsigned value_id = get_expr_value_id (constant);
1481 /* We want a leader in ANTIC_OUT or AVAIL_OUT here.
1482 dest has what we computed into ANTIC_OUT sofar
1483 so pick from that - since topological sorting
1484 by sorted_array_from_bitmap_set isn't perfect
1485 we may lose some cases here. */
1486 constant = find_leader_in_sets (value_id, dest,
1487 AVAIL_OUT (pred));
1488 if (constant)
1490 if (dump_file && (dump_flags & TDF_DETAILS))
1492 fprintf (dump_file, "simplifying ");
1493 print_pre_expr (dump_file, expr);
1494 fprintf (dump_file, " translated %d -> %d to ",
1495 phiblock->index, pred->index);
1496 PRE_EXPR_NARY (expr) = newnary;
1497 print_pre_expr (dump_file, expr);
1498 PRE_EXPR_NARY (expr) = nary;
1499 fprintf (dump_file, " to ");
1500 print_pre_expr (dump_file, constant);
1501 fprintf (dump_file, "\n");
1503 return constant;
1507 else
1508 return constant;
1511 tree result = vn_nary_op_lookup_pieces (newnary->length,
1512 newnary->opcode,
1513 newnary->type,
1514 &newnary->op[0],
1515 &nary);
1516 if (result && is_gimple_min_invariant (result))
1517 return get_or_alloc_expr_for_constant (result);
1519 if (!nary || nary->predicated_values)
1521 new_val_id = get_next_value_id ();
1522 nary = vn_nary_op_insert_pieces (newnary->length,
1523 newnary->opcode,
1524 newnary->type,
1525 &newnary->op[0],
1526 result, new_val_id);
1528 expr = get_or_alloc_expr_for_nary (nary, expr_loc);
1529 add_to_value (get_expr_value_id (expr), expr);
1531 return expr;
1533 break;
1535 case REFERENCE:
1537 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1538 vec<vn_reference_op_s> operands = ref->operands;
1539 tree vuse = ref->vuse;
1540 tree newvuse = vuse;
1541 vec<vn_reference_op_s> newoperands = vNULL;
1542 bool changed = false, same_valid = true;
1543 unsigned int i, n;
1544 vn_reference_op_t operand;
1545 vn_reference_t newref;
1547 for (i = 0; operands.iterate (i, &operand); i++)
1549 pre_expr opresult;
1550 pre_expr leader;
1551 tree op[3];
1552 tree type = operand->type;
1553 vn_reference_op_s newop = *operand;
1554 op[0] = operand->op0;
1555 op[1] = operand->op1;
1556 op[2] = operand->op2;
1557 for (n = 0; n < 3; ++n)
1559 unsigned int op_val_id;
1560 if (!op[n])
1561 continue;
1562 if (TREE_CODE (op[n]) != SSA_NAME)
1564 /* We can't possibly insert these. */
1565 if (n != 0
1566 && !is_gimple_min_invariant (op[n]))
1567 break;
1568 continue;
1570 op_val_id = VN_INFO (op[n])->value_id;
1571 leader = find_leader_in_sets (op_val_id, set1, set2);
1572 opresult = phi_translate (dest, leader, set1, set2, e);
1573 if (opresult && opresult != leader)
1575 tree name = get_representative_for (opresult);
1576 changed |= name != op[n];
1577 op[n] = name;
1579 else if (!opresult)
1580 break;
1582 if (n != 3)
1584 newoperands.release ();
1585 return NULL;
1587 /* When we translate a MEM_REF across a backedge and we have
1588 restrict info that's not from our functions parameters
1589 we have to remap it since we now may deal with a different
1590 instance where the dependence info is no longer valid.
1591 See PR102970. Note instead of keeping a remapping table
1592 per backedge we simply throw away restrict info. */
1593 if ((newop.opcode == MEM_REF
1594 || newop.opcode == TARGET_MEM_REF)
1595 && newop.clique > 1
1596 && (e->flags & EDGE_DFS_BACK))
1598 newop.clique = 0;
1599 newop.base = 0;
1600 changed = true;
1602 if (!changed)
1603 continue;
1604 if (!newoperands.exists ())
1605 newoperands = operands.copy ();
1606 /* We may have changed from an SSA_NAME to a constant */
1607 if (newop.opcode == SSA_NAME && TREE_CODE (op[0]) != SSA_NAME)
1608 newop.opcode = TREE_CODE (op[0]);
1609 newop.type = type;
1610 newop.op0 = op[0];
1611 newop.op1 = op[1];
1612 newop.op2 = op[2];
1613 newoperands[i] = newop;
1615 gcc_checking_assert (i == operands.length ());
1617 if (vuse)
1619 newvuse = translate_vuse_through_block (newoperands.exists ()
1620 ? newoperands : operands,
1621 ref->set, ref->base_set,
1622 ref->type, vuse, e,
1623 changed
1624 ? NULL : &same_valid);
1625 if (newvuse == NULL_TREE)
1627 newoperands.release ();
1628 return NULL;
1632 if (changed || newvuse != vuse)
1634 unsigned int new_val_id;
1636 tree result = vn_reference_lookup_pieces (newvuse, ref->set,
1637 ref->base_set,
1638 ref->type,
1639 newoperands.exists ()
1640 ? newoperands : operands,
1641 &newref, VN_WALK);
1642 if (result)
1643 newoperands.release ();
1645 /* We can always insert constants, so if we have a partial
1646 redundant constant load of another type try to translate it
1647 to a constant of appropriate type. */
1648 if (result && is_gimple_min_invariant (result))
1650 tree tem = result;
1651 if (!useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1653 tem = fold_unary (VIEW_CONVERT_EXPR, ref->type, result);
1654 if (tem && !is_gimple_min_invariant (tem))
1655 tem = NULL_TREE;
1657 if (tem)
1658 return get_or_alloc_expr_for_constant (tem);
1661 /* If we'd have to convert things we would need to validate
1662 if we can insert the translated expression. So fail
1663 here for now - we cannot insert an alias with a different
1664 type in the VN tables either, as that would assert. */
1665 if (result
1666 && !useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1667 return NULL;
1668 else if (!result && newref
1669 && !useless_type_conversion_p (ref->type, newref->type))
1671 newoperands.release ();
1672 return NULL;
1675 if (newref)
1676 new_val_id = newref->value_id;
1677 else
1679 if (changed || !same_valid)
1680 new_val_id = get_next_value_id ();
1681 else
1682 new_val_id = ref->value_id;
1683 if (!newoperands.exists ())
1684 newoperands = operands.copy ();
1685 newref = vn_reference_insert_pieces (newvuse, ref->set,
1686 ref->base_set, ref->type,
1687 newoperands,
1688 result, new_val_id);
1689 newoperands = vNULL;
1691 expr = get_or_alloc_expr_for_reference (newref, expr_loc);
1692 add_to_value (new_val_id, expr);
1694 newoperands.release ();
1695 return expr;
1697 break;
1699 case NAME:
1701 tree name = PRE_EXPR_NAME (expr);
1702 gimple *def_stmt = SSA_NAME_DEF_STMT (name);
1703 /* If the SSA name is defined by a PHI node in this block,
1704 translate it. */
1705 if (gimple_code (def_stmt) == GIMPLE_PHI
1706 && gimple_bb (def_stmt) == phiblock)
1708 tree def = PHI_ARG_DEF (def_stmt, e->dest_idx);
1710 /* Handle constant. */
1711 if (is_gimple_min_invariant (def))
1712 return get_or_alloc_expr_for_constant (def);
1714 return get_or_alloc_expr_for_name (def);
1716 /* Otherwise return it unchanged - it will get removed if its
1717 value is not available in PREDs AVAIL_OUT set of expressions
1718 by the subtraction of TMP_GEN. */
1719 return expr;
1722 default:
1723 gcc_unreachable ();
1727 /* Wrapper around phi_translate_1 providing caching functionality. */
1729 static pre_expr
1730 phi_translate (bitmap_set_t dest, pre_expr expr,
1731 bitmap_set_t set1, bitmap_set_t set2, edge e)
1733 expr_pred_trans_t slot = NULL;
1734 pre_expr phitrans;
1736 if (!expr)
1737 return NULL;
1739 /* Constants contain no values that need translation. */
1740 if (expr->kind == CONSTANT)
1741 return expr;
1743 if (value_id_constant_p (get_expr_value_id (expr)))
1744 return expr;
1746 /* Don't add translations of NAMEs as those are cheap to translate. */
1747 if (expr->kind != NAME)
1749 if (phi_trans_add (&slot, expr, e->src))
1750 return slot->v == 0 ? NULL : expression_for_id (slot->v);
1751 /* Store NULL for the value we want to return in the case of
1752 recursing. */
1753 slot->v = 0;
1756 /* Translate. */
1757 basic_block saved_valueize_bb = vn_context_bb;
1758 vn_context_bb = e->src;
1759 phitrans = phi_translate_1 (dest, expr, set1, set2, e);
1760 vn_context_bb = saved_valueize_bb;
1762 if (slot)
1764 /* We may have reallocated. */
1765 phi_trans_add (&slot, expr, e->src);
1766 if (phitrans)
1767 slot->v = get_expression_id (phitrans);
1768 else
1769 /* Remove failed translations again, they cause insert
1770 iteration to not pick up new opportunities reliably. */
1771 PHI_TRANS_TABLE (e->src)->clear_slot (slot);
1774 return phitrans;
1778 /* For each expression in SET, translate the values through phi nodes
1779 in PHIBLOCK using edge PHIBLOCK->PRED, and store the resulting
1780 expressions in DEST. */
1782 static void
1783 phi_translate_set (bitmap_set_t dest, bitmap_set_t set, edge e)
1785 bitmap_iterator bi;
1786 unsigned int i;
1788 if (gimple_seq_empty_p (phi_nodes (e->dest)))
1790 bitmap_set_copy (dest, set);
1791 return;
1794 /* Allocate the phi-translation cache where we have an idea about
1795 its size. hash-table implementation internals tell us that
1796 allocating the table to fit twice the number of elements will
1797 make sure we do not usually re-allocate. */
1798 if (!PHI_TRANS_TABLE (e->src))
1799 PHI_TRANS_TABLE (e->src) = new hash_table<expr_pred_trans_d>
1800 (2 * bitmap_count_bits (&set->expressions));
1801 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
1803 pre_expr expr = expression_for_id (i);
1804 pre_expr translated = phi_translate (dest, expr, set, NULL, e);
1805 if (!translated)
1806 continue;
1808 bitmap_insert_into_set (dest, translated);
1812 /* Find the leader for a value (i.e., the name representing that
1813 value) in a given set, and return it. Return NULL if no leader
1814 is found. */
1816 static pre_expr
1817 bitmap_find_leader (bitmap_set_t set, unsigned int val)
1819 if (value_id_constant_p (val))
1820 return constant_value_expressions[-val];
1822 if (bitmap_set_contains_value (set, val))
1824 /* Rather than walk the entire bitmap of expressions, and see
1825 whether any of them has the value we are looking for, we look
1826 at the reverse mapping, which tells us the set of expressions
1827 that have a given value (IE value->expressions with that
1828 value) and see if any of those expressions are in our set.
1829 The number of expressions per value is usually significantly
1830 less than the number of expressions in the set. In fact, for
1831 large testcases, doing it this way is roughly 5-10x faster
1832 than walking the bitmap.
1833 If this is somehow a significant lose for some cases, we can
1834 choose which set to walk based on which set is smaller. */
1835 unsigned int i;
1836 bitmap_iterator bi;
1837 bitmap exprset = value_expressions[val];
1839 if (!exprset->first->next)
1840 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
1841 if (bitmap_bit_p (&set->expressions, i))
1842 return expression_for_id (i);
1844 EXECUTE_IF_AND_IN_BITMAP (exprset, &set->expressions, 0, i, bi)
1845 return expression_for_id (i);
1847 return NULL;
1850 /* Determine if EXPR, a memory expression, is ANTIC_IN at the top of
1851 BLOCK by seeing if it is not killed in the block. Note that we are
1852 only determining whether there is a store that kills it. Because
1853 of the order in which clean iterates over values, we are guaranteed
1854 that altered operands will have caused us to be eliminated from the
1855 ANTIC_IN set already. */
1857 static bool
1858 value_dies_in_block_x (pre_expr expr, basic_block block)
1860 tree vuse = PRE_EXPR_REFERENCE (expr)->vuse;
1861 vn_reference_t refx = PRE_EXPR_REFERENCE (expr);
1862 gimple *def;
1863 gimple_stmt_iterator gsi;
1864 unsigned id = get_expression_id (expr);
1865 bool res = false;
1866 ao_ref ref;
1868 if (!vuse)
1869 return false;
1871 /* Lookup a previously calculated result. */
1872 if (EXPR_DIES (block)
1873 && bitmap_bit_p (EXPR_DIES (block), id * 2))
1874 return bitmap_bit_p (EXPR_DIES (block), id * 2 + 1);
1876 /* A memory expression {e, VUSE} dies in the block if there is a
1877 statement that may clobber e. If, starting statement walk from the
1878 top of the basic block, a statement uses VUSE there can be no kill
1879 inbetween that use and the original statement that loaded {e, VUSE},
1880 so we can stop walking. */
1881 ref.base = NULL_TREE;
1882 for (gsi = gsi_start_bb (block); !gsi_end_p (gsi); gsi_next (&gsi))
1884 tree def_vuse, def_vdef;
1885 def = gsi_stmt (gsi);
1886 def_vuse = gimple_vuse (def);
1887 def_vdef = gimple_vdef (def);
1889 /* Not a memory statement. */
1890 if (!def_vuse)
1891 continue;
1893 /* Not a may-def. */
1894 if (!def_vdef)
1896 /* A load with the same VUSE, we're done. */
1897 if (def_vuse == vuse)
1898 break;
1900 continue;
1903 /* Init ref only if we really need it. */
1904 if (ref.base == NULL_TREE
1905 && !ao_ref_init_from_vn_reference (&ref, refx->set, refx->base_set,
1906 refx->type, refx->operands))
1908 res = true;
1909 break;
1911 /* If the statement may clobber expr, it dies. */
1912 if (stmt_may_clobber_ref_p_1 (def, &ref))
1914 res = true;
1915 break;
1919 /* Remember the result. */
1920 if (!EXPR_DIES (block))
1921 EXPR_DIES (block) = BITMAP_ALLOC (&grand_bitmap_obstack);
1922 bitmap_set_bit (EXPR_DIES (block), id * 2);
1923 if (res)
1924 bitmap_set_bit (EXPR_DIES (block), id * 2 + 1);
1926 return res;
1930 /* Determine if OP is valid in SET1 U SET2, which it is when the union
1931 contains its value-id. */
1933 static bool
1934 op_valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, tree op)
1936 if (op && TREE_CODE (op) == SSA_NAME)
1938 unsigned int value_id = VN_INFO (op)->value_id;
1939 if (!(bitmap_set_contains_value (set1, value_id)
1940 || (set2 && bitmap_set_contains_value (set2, value_id))))
1941 return false;
1943 return true;
1946 /* Determine if the expression EXPR is valid in SET1 U SET2.
1947 ONLY SET2 CAN BE NULL.
1948 This means that we have a leader for each part of the expression
1949 (if it consists of values), or the expression is an SSA_NAME.
1950 For loads/calls, we also see if the vuse is killed in this block. */
1952 static bool
1953 valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, pre_expr expr)
1955 switch (expr->kind)
1957 case NAME:
1958 /* By construction all NAMEs are available. Non-available
1959 NAMEs are removed by subtracting TMP_GEN from the sets. */
1960 return true;
1961 case NARY:
1963 unsigned int i;
1964 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1965 for (i = 0; i < nary->length; i++)
1966 if (!op_valid_in_sets (set1, set2, nary->op[i]))
1967 return false;
1968 return true;
1970 break;
1971 case REFERENCE:
1973 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1974 vn_reference_op_t vro;
1975 unsigned int i;
1977 FOR_EACH_VEC_ELT (ref->operands, i, vro)
1979 if (!op_valid_in_sets (set1, set2, vro->op0)
1980 || !op_valid_in_sets (set1, set2, vro->op1)
1981 || !op_valid_in_sets (set1, set2, vro->op2))
1982 return false;
1984 return true;
1986 default:
1987 gcc_unreachable ();
1991 /* Clean the set of expressions SET1 that are no longer valid in SET1 or SET2.
1992 This means expressions that are made up of values we have no leaders for
1993 in SET1 or SET2. */
1995 static void
1996 clean (bitmap_set_t set1, bitmap_set_t set2 = NULL)
1998 vec<pre_expr> exprs = sorted_array_from_bitmap_set (set1);
1999 pre_expr expr;
2000 int i;
2002 FOR_EACH_VEC_ELT (exprs, i, expr)
2004 if (!valid_in_sets (set1, set2, expr))
2006 unsigned int val = get_expr_value_id (expr);
2007 bitmap_clear_bit (&set1->expressions, get_expression_id (expr));
2008 /* We are entered with possibly multiple expressions for a value
2009 so before removing a value from the set see if there's an
2010 expression for it left. */
2011 if (! bitmap_find_leader (set1, val))
2012 bitmap_clear_bit (&set1->values, val);
2015 exprs.release ();
2017 if (flag_checking)
2019 unsigned j;
2020 bitmap_iterator bi;
2021 FOR_EACH_EXPR_ID_IN_SET (set1, j, bi)
2022 gcc_assert (valid_in_sets (set1, set2, expression_for_id (j)));
2026 /* Clean the set of expressions that are no longer valid in SET because
2027 they are clobbered in BLOCK or because they trap and may not be executed. */
2029 static void
2030 prune_clobbered_mems (bitmap_set_t set, basic_block block)
2032 bitmap_iterator bi;
2033 unsigned i;
2034 unsigned to_remove = -1U;
2035 bool any_removed = false;
2037 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
2039 /* Remove queued expr. */
2040 if (to_remove != -1U)
2042 bitmap_clear_bit (&set->expressions, to_remove);
2043 any_removed = true;
2044 to_remove = -1U;
2047 pre_expr expr = expression_for_id (i);
2048 if (expr->kind == REFERENCE)
2050 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2051 if (ref->vuse)
2053 gimple *def_stmt = SSA_NAME_DEF_STMT (ref->vuse);
2054 if (!gimple_nop_p (def_stmt)
2055 && ((gimple_bb (def_stmt) != block
2056 && !dominated_by_p (CDI_DOMINATORS,
2057 block, gimple_bb (def_stmt)))
2058 || (gimple_bb (def_stmt) == block
2059 && value_dies_in_block_x (expr, block))))
2060 to_remove = i;
2062 /* If the REFERENCE may trap make sure the block does not contain
2063 a possible exit point.
2064 ??? This is overly conservative if we translate AVAIL_OUT
2065 as the available expression might be after the exit point. */
2066 if (BB_MAY_NOTRETURN (block)
2067 && vn_reference_may_trap (ref))
2068 to_remove = i;
2070 else if (expr->kind == NARY)
2072 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2073 /* If the NARY may trap make sure the block does not contain
2074 a possible exit point.
2075 ??? This is overly conservative if we translate AVAIL_OUT
2076 as the available expression might be after the exit point. */
2077 if (BB_MAY_NOTRETURN (block)
2078 && vn_nary_may_trap (nary))
2079 to_remove = i;
2083 /* Remove queued expr. */
2084 if (to_remove != -1U)
2086 bitmap_clear_bit (&set->expressions, to_remove);
2087 any_removed = true;
2090 /* Above we only removed expressions, now clean the set of values
2091 which no longer have any corresponding expression. We cannot
2092 clear the value at the time we remove an expression since there
2093 may be multiple expressions per value.
2094 If we'd queue possibly to be removed values we could use
2095 the bitmap_find_leader way to see if there's still an expression
2096 for it. For some ratio of to be removed values and number of
2097 values/expressions in the set this might be faster than rebuilding
2098 the value-set. */
2099 if (any_removed)
2101 bitmap_clear (&set->values);
2102 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
2104 pre_expr expr = expression_for_id (i);
2105 unsigned int value_id = get_expr_value_id (expr);
2106 bitmap_set_bit (&set->values, value_id);
2111 /* Compute the ANTIC set for BLOCK.
2113 If succs(BLOCK) > 1 then
2114 ANTIC_OUT[BLOCK] = intersection of ANTIC_IN[b] for all succ(BLOCK)
2115 else if succs(BLOCK) == 1 then
2116 ANTIC_OUT[BLOCK] = phi_translate (ANTIC_IN[succ(BLOCK)])
2118 ANTIC_IN[BLOCK] = clean(ANTIC_OUT[BLOCK] U EXP_GEN[BLOCK] - TMP_GEN[BLOCK])
2120 Note that clean() is deferred until after the iteration. */
2122 static bool
2123 compute_antic_aux (basic_block block, bool block_has_abnormal_pred_edge)
2125 bitmap_set_t S, old, ANTIC_OUT;
2126 edge e;
2127 edge_iterator ei;
2129 bool was_visited = BB_VISITED (block);
2130 bool changed = ! BB_VISITED (block);
2131 BB_VISITED (block) = 1;
2132 old = ANTIC_OUT = S = NULL;
2134 /* If any edges from predecessors are abnormal, antic_in is empty,
2135 so do nothing. */
2136 if (block_has_abnormal_pred_edge)
2137 goto maybe_dump_sets;
2139 old = ANTIC_IN (block);
2140 ANTIC_OUT = bitmap_set_new ();
2142 /* If the block has no successors, ANTIC_OUT is empty. */
2143 if (EDGE_COUNT (block->succs) == 0)
2145 /* If we have one successor, we could have some phi nodes to
2146 translate through. */
2147 else if (single_succ_p (block))
2149 e = single_succ_edge (block);
2150 gcc_assert (BB_VISITED (e->dest));
2151 phi_translate_set (ANTIC_OUT, ANTIC_IN (e->dest), e);
2153 /* If we have multiple successors, we take the intersection of all of
2154 them. Note that in the case of loop exit phi nodes, we may have
2155 phis to translate through. */
2156 else
2158 size_t i;
2159 edge first = NULL;
2161 auto_vec<edge> worklist (EDGE_COUNT (block->succs));
2162 FOR_EACH_EDGE (e, ei, block->succs)
2164 if (!first
2165 && BB_VISITED (e->dest))
2166 first = e;
2167 else if (BB_VISITED (e->dest))
2168 worklist.quick_push (e);
2169 else
2171 /* Unvisited successors get their ANTIC_IN replaced by the
2172 maximal set to arrive at a maximum ANTIC_IN solution.
2173 We can ignore them in the intersection operation and thus
2174 need not explicitely represent that maximum solution. */
2175 if (dump_file && (dump_flags & TDF_DETAILS))
2176 fprintf (dump_file, "ANTIC_IN is MAX on %d->%d\n",
2177 e->src->index, e->dest->index);
2181 /* Of multiple successors we have to have visited one already
2182 which is guaranteed by iteration order. */
2183 gcc_assert (first != NULL);
2185 phi_translate_set (ANTIC_OUT, ANTIC_IN (first->dest), first);
2187 /* If we have multiple successors we need to intersect the ANTIC_OUT
2188 sets. For values that's a simple intersection but for
2189 expressions it is a union. Given we want to have a single
2190 expression per value in our sets we have to canonicalize.
2191 Avoid randomness and running into cycles like for PR82129 and
2192 canonicalize the expression we choose to the one with the
2193 lowest id. This requires we actually compute the union first. */
2194 FOR_EACH_VEC_ELT (worklist, i, e)
2196 if (!gimple_seq_empty_p (phi_nodes (e->dest)))
2198 bitmap_set_t tmp = bitmap_set_new ();
2199 phi_translate_set (tmp, ANTIC_IN (e->dest), e);
2200 bitmap_and_into (&ANTIC_OUT->values, &tmp->values);
2201 bitmap_ior_into (&ANTIC_OUT->expressions, &tmp->expressions);
2202 bitmap_set_free (tmp);
2204 else
2206 bitmap_and_into (&ANTIC_OUT->values, &ANTIC_IN (e->dest)->values);
2207 bitmap_ior_into (&ANTIC_OUT->expressions,
2208 &ANTIC_IN (e->dest)->expressions);
2211 if (! worklist.is_empty ())
2213 /* Prune expressions not in the value set. */
2214 bitmap_iterator bi;
2215 unsigned int i;
2216 unsigned int to_clear = -1U;
2217 FOR_EACH_EXPR_ID_IN_SET (ANTIC_OUT, i, bi)
2219 if (to_clear != -1U)
2221 bitmap_clear_bit (&ANTIC_OUT->expressions, to_clear);
2222 to_clear = -1U;
2224 pre_expr expr = expression_for_id (i);
2225 unsigned int value_id = get_expr_value_id (expr);
2226 if (!bitmap_bit_p (&ANTIC_OUT->values, value_id))
2227 to_clear = i;
2229 if (to_clear != -1U)
2230 bitmap_clear_bit (&ANTIC_OUT->expressions, to_clear);
2234 /* Prune expressions that are clobbered in block and thus become
2235 invalid if translated from ANTIC_OUT to ANTIC_IN. */
2236 prune_clobbered_mems (ANTIC_OUT, block);
2238 /* Generate ANTIC_OUT - TMP_GEN. */
2239 S = bitmap_set_subtract_expressions (ANTIC_OUT, TMP_GEN (block));
2241 /* Start ANTIC_IN with EXP_GEN - TMP_GEN. */
2242 ANTIC_IN (block) = bitmap_set_subtract_expressions (EXP_GEN (block),
2243 TMP_GEN (block));
2245 /* Then union in the ANTIC_OUT - TMP_GEN values,
2246 to get ANTIC_OUT U EXP_GEN - TMP_GEN */
2247 bitmap_ior_into (&ANTIC_IN (block)->values, &S->values);
2248 bitmap_ior_into (&ANTIC_IN (block)->expressions, &S->expressions);
2250 /* clean (ANTIC_IN (block)) is defered to after the iteration converged
2251 because it can cause non-convergence, see for example PR81181. */
2253 /* Intersect ANTIC_IN with the old ANTIC_IN. This is required until
2254 we properly represent the maximum expression set, thus not prune
2255 values without expressions during the iteration. */
2256 if (was_visited
2257 && bitmap_and_into (&ANTIC_IN (block)->values, &old->values))
2259 if (dump_file && (dump_flags & TDF_DETAILS))
2260 fprintf (dump_file, "warning: intersecting with old ANTIC_IN "
2261 "shrinks the set\n");
2262 /* Prune expressions not in the value set. */
2263 bitmap_iterator bi;
2264 unsigned int i;
2265 unsigned int to_clear = -1U;
2266 FOR_EACH_EXPR_ID_IN_SET (ANTIC_IN (block), i, bi)
2268 if (to_clear != -1U)
2270 bitmap_clear_bit (&ANTIC_IN (block)->expressions, to_clear);
2271 to_clear = -1U;
2273 pre_expr expr = expression_for_id (i);
2274 unsigned int value_id = get_expr_value_id (expr);
2275 if (!bitmap_bit_p (&ANTIC_IN (block)->values, value_id))
2276 to_clear = i;
2278 if (to_clear != -1U)
2279 bitmap_clear_bit (&ANTIC_IN (block)->expressions, to_clear);
2282 if (!bitmap_set_equal (old, ANTIC_IN (block)))
2283 changed = true;
2285 maybe_dump_sets:
2286 if (dump_file && (dump_flags & TDF_DETAILS))
2288 if (ANTIC_OUT)
2289 print_bitmap_set (dump_file, ANTIC_OUT, "ANTIC_OUT", block->index);
2291 if (changed)
2292 fprintf (dump_file, "[changed] ");
2293 print_bitmap_set (dump_file, ANTIC_IN (block), "ANTIC_IN",
2294 block->index);
2296 if (S)
2297 print_bitmap_set (dump_file, S, "S", block->index);
2299 if (old)
2300 bitmap_set_free (old);
2301 if (S)
2302 bitmap_set_free (S);
2303 if (ANTIC_OUT)
2304 bitmap_set_free (ANTIC_OUT);
2305 return changed;
2308 /* Compute PARTIAL_ANTIC for BLOCK.
2310 If succs(BLOCK) > 1 then
2311 PA_OUT[BLOCK] = value wise union of PA_IN[b] + all ANTIC_IN not
2312 in ANTIC_OUT for all succ(BLOCK)
2313 else if succs(BLOCK) == 1 then
2314 PA_OUT[BLOCK] = phi_translate (PA_IN[succ(BLOCK)])
2316 PA_IN[BLOCK] = clean(PA_OUT[BLOCK] - TMP_GEN[BLOCK] - ANTIC_IN[BLOCK])
2319 static void
2320 compute_partial_antic_aux (basic_block block,
2321 bool block_has_abnormal_pred_edge)
2323 bitmap_set_t old_PA_IN;
2324 bitmap_set_t PA_OUT;
2325 edge e;
2326 edge_iterator ei;
2327 unsigned long max_pa = param_max_partial_antic_length;
2329 old_PA_IN = PA_OUT = NULL;
2331 /* If any edges from predecessors are abnormal, antic_in is empty,
2332 so do nothing. */
2333 if (block_has_abnormal_pred_edge)
2334 goto maybe_dump_sets;
2336 /* If there are too many partially anticipatable values in the
2337 block, phi_translate_set can take an exponential time: stop
2338 before the translation starts. */
2339 if (max_pa
2340 && single_succ_p (block)
2341 && bitmap_count_bits (&PA_IN (single_succ (block))->values) > max_pa)
2342 goto maybe_dump_sets;
2344 old_PA_IN = PA_IN (block);
2345 PA_OUT = bitmap_set_new ();
2347 /* If the block has no successors, ANTIC_OUT is empty. */
2348 if (EDGE_COUNT (block->succs) == 0)
2350 /* If we have one successor, we could have some phi nodes to
2351 translate through. Note that we can't phi translate across DFS
2352 back edges in partial antic, because it uses a union operation on
2353 the successors. For recurrences like IV's, we will end up
2354 generating a new value in the set on each go around (i + 3 (VH.1)
2355 VH.1 + 1 (VH.2), VH.2 + 1 (VH.3), etc), forever. */
2356 else if (single_succ_p (block))
2358 e = single_succ_edge (block);
2359 if (!(e->flags & EDGE_DFS_BACK))
2360 phi_translate_set (PA_OUT, PA_IN (e->dest), e);
2362 /* If we have multiple successors, we take the union of all of
2363 them. */
2364 else
2366 size_t i;
2368 auto_vec<edge> worklist (EDGE_COUNT (block->succs));
2369 FOR_EACH_EDGE (e, ei, block->succs)
2371 if (e->flags & EDGE_DFS_BACK)
2372 continue;
2373 worklist.quick_push (e);
2375 if (worklist.length () > 0)
2377 FOR_EACH_VEC_ELT (worklist, i, e)
2379 unsigned int i;
2380 bitmap_iterator bi;
2382 FOR_EACH_EXPR_ID_IN_SET (ANTIC_IN (e->dest), i, bi)
2383 bitmap_value_insert_into_set (PA_OUT,
2384 expression_for_id (i));
2385 if (!gimple_seq_empty_p (phi_nodes (e->dest)))
2387 bitmap_set_t pa_in = bitmap_set_new ();
2388 phi_translate_set (pa_in, PA_IN (e->dest), e);
2389 FOR_EACH_EXPR_ID_IN_SET (pa_in, i, bi)
2390 bitmap_value_insert_into_set (PA_OUT,
2391 expression_for_id (i));
2392 bitmap_set_free (pa_in);
2394 else
2395 FOR_EACH_EXPR_ID_IN_SET (PA_IN (e->dest), i, bi)
2396 bitmap_value_insert_into_set (PA_OUT,
2397 expression_for_id (i));
2402 /* Prune expressions that are clobbered in block and thus become
2403 invalid if translated from PA_OUT to PA_IN. */
2404 prune_clobbered_mems (PA_OUT, block);
2406 /* PA_IN starts with PA_OUT - TMP_GEN.
2407 Then we subtract things from ANTIC_IN. */
2408 PA_IN (block) = bitmap_set_subtract_expressions (PA_OUT, TMP_GEN (block));
2410 /* For partial antic, we want to put back in the phi results, since
2411 we will properly avoid making them partially antic over backedges. */
2412 bitmap_ior_into (&PA_IN (block)->values, &PHI_GEN (block)->values);
2413 bitmap_ior_into (&PA_IN (block)->expressions, &PHI_GEN (block)->expressions);
2415 /* PA_IN[block] = PA_IN[block] - ANTIC_IN[block] */
2416 bitmap_set_subtract_values (PA_IN (block), ANTIC_IN (block));
2418 clean (PA_IN (block), ANTIC_IN (block));
2420 maybe_dump_sets:
2421 if (dump_file && (dump_flags & TDF_DETAILS))
2423 if (PA_OUT)
2424 print_bitmap_set (dump_file, PA_OUT, "PA_OUT", block->index);
2426 print_bitmap_set (dump_file, PA_IN (block), "PA_IN", block->index);
2428 if (old_PA_IN)
2429 bitmap_set_free (old_PA_IN);
2430 if (PA_OUT)
2431 bitmap_set_free (PA_OUT);
2434 /* Compute ANTIC and partial ANTIC sets. */
2436 static void
2437 compute_antic (void)
2439 bool changed = true;
2440 int num_iterations = 0;
2441 basic_block block;
2442 int i;
2443 edge_iterator ei;
2444 edge e;
2446 /* If any predecessor edges are abnormal, we punt, so antic_in is empty.
2447 We pre-build the map of blocks with incoming abnormal edges here. */
2448 auto_sbitmap has_abnormal_preds (last_basic_block_for_fn (cfun));
2449 bitmap_clear (has_abnormal_preds);
2451 FOR_ALL_BB_FN (block, cfun)
2453 BB_VISITED (block) = 0;
2455 FOR_EACH_EDGE (e, ei, block->preds)
2456 if (e->flags & EDGE_ABNORMAL)
2458 bitmap_set_bit (has_abnormal_preds, block->index);
2459 break;
2462 /* While we are here, give empty ANTIC_IN sets to each block. */
2463 ANTIC_IN (block) = bitmap_set_new ();
2464 if (do_partial_partial)
2465 PA_IN (block) = bitmap_set_new ();
2468 /* At the exit block we anticipate nothing. */
2469 BB_VISITED (EXIT_BLOCK_PTR_FOR_FN (cfun)) = 1;
2471 /* For ANTIC computation we need a postorder that also guarantees that
2472 a block with a single successor is visited after its successor.
2473 RPO on the inverted CFG has this property. */
2474 auto_vec<int, 20> postorder;
2475 inverted_post_order_compute (&postorder);
2477 auto_sbitmap worklist (last_basic_block_for_fn (cfun) + 1);
2478 bitmap_clear (worklist);
2479 FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR_FOR_FN (cfun)->preds)
2480 bitmap_set_bit (worklist, e->src->index);
2481 while (changed)
2483 if (dump_file && (dump_flags & TDF_DETAILS))
2484 fprintf (dump_file, "Starting iteration %d\n", num_iterations);
2485 /* ??? We need to clear our PHI translation cache here as the
2486 ANTIC sets shrink and we restrict valid translations to
2487 those having operands with leaders in ANTIC. Same below
2488 for PA ANTIC computation. */
2489 num_iterations++;
2490 changed = false;
2491 for (i = postorder.length () - 1; i >= 0; i--)
2493 if (bitmap_bit_p (worklist, postorder[i]))
2495 basic_block block = BASIC_BLOCK_FOR_FN (cfun, postorder[i]);
2496 bitmap_clear_bit (worklist, block->index);
2497 if (compute_antic_aux (block,
2498 bitmap_bit_p (has_abnormal_preds,
2499 block->index)))
2501 FOR_EACH_EDGE (e, ei, block->preds)
2502 bitmap_set_bit (worklist, e->src->index);
2503 changed = true;
2507 /* Theoretically possible, but *highly* unlikely. */
2508 gcc_checking_assert (num_iterations < 500);
2511 /* We have to clean after the dataflow problem converged as cleaning
2512 can cause non-convergence because it is based on expressions
2513 rather than values. */
2514 FOR_EACH_BB_FN (block, cfun)
2515 clean (ANTIC_IN (block));
2517 statistics_histogram_event (cfun, "compute_antic iterations",
2518 num_iterations);
2520 if (do_partial_partial)
2522 /* For partial antic we ignore backedges and thus we do not need
2523 to perform any iteration when we process blocks in postorder. */
2524 for (i = postorder.length () - 1; i >= 0; i--)
2526 basic_block block = BASIC_BLOCK_FOR_FN (cfun, postorder[i]);
2527 compute_partial_antic_aux (block,
2528 bitmap_bit_p (has_abnormal_preds,
2529 block->index));
2535 /* Inserted expressions are placed onto this worklist, which is used
2536 for performing quick dead code elimination of insertions we made
2537 that didn't turn out to be necessary. */
2538 static bitmap inserted_exprs;
2540 /* The actual worker for create_component_ref_by_pieces. */
2542 static tree
2543 create_component_ref_by_pieces_1 (basic_block block, vn_reference_t ref,
2544 unsigned int *operand, gimple_seq *stmts)
2546 vn_reference_op_t currop = &ref->operands[*operand];
2547 tree genop;
2548 ++*operand;
2549 switch (currop->opcode)
2551 case CALL_EXPR:
2552 gcc_unreachable ();
2554 case MEM_REF:
2556 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2557 stmts);
2558 if (!baseop)
2559 return NULL_TREE;
2560 tree offset = currop->op0;
2561 if (TREE_CODE (baseop) == ADDR_EXPR
2562 && handled_component_p (TREE_OPERAND (baseop, 0)))
2564 poly_int64 off;
2565 tree base;
2566 base = get_addr_base_and_unit_offset (TREE_OPERAND (baseop, 0),
2567 &off);
2568 gcc_assert (base);
2569 offset = int_const_binop (PLUS_EXPR, offset,
2570 build_int_cst (TREE_TYPE (offset),
2571 off));
2572 baseop = build_fold_addr_expr (base);
2574 genop = build2 (MEM_REF, currop->type, baseop, offset);
2575 MR_DEPENDENCE_CLIQUE (genop) = currop->clique;
2576 MR_DEPENDENCE_BASE (genop) = currop->base;
2577 REF_REVERSE_STORAGE_ORDER (genop) = currop->reverse;
2578 return genop;
2581 case TARGET_MEM_REF:
2583 tree genop0 = NULL_TREE, genop1 = NULL_TREE;
2584 vn_reference_op_t nextop = &ref->operands[(*operand)++];
2585 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2586 stmts);
2587 if (!baseop)
2588 return NULL_TREE;
2589 if (currop->op0)
2591 genop0 = find_or_generate_expression (block, currop->op0, stmts);
2592 if (!genop0)
2593 return NULL_TREE;
2595 if (nextop->op0)
2597 genop1 = find_or_generate_expression (block, nextop->op0, stmts);
2598 if (!genop1)
2599 return NULL_TREE;
2601 genop = build5 (TARGET_MEM_REF, currop->type,
2602 baseop, currop->op2, genop0, currop->op1, genop1);
2604 MR_DEPENDENCE_CLIQUE (genop) = currop->clique;
2605 MR_DEPENDENCE_BASE (genop) = currop->base;
2606 return genop;
2609 case ADDR_EXPR:
2610 if (currop->op0)
2612 gcc_assert (is_gimple_min_invariant (currop->op0));
2613 return currop->op0;
2615 /* Fallthrough. */
2616 case REALPART_EXPR:
2617 case IMAGPART_EXPR:
2618 case VIEW_CONVERT_EXPR:
2620 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2621 stmts);
2622 if (!genop0)
2623 return NULL_TREE;
2624 return fold_build1 (currop->opcode, currop->type, genop0);
2627 case WITH_SIZE_EXPR:
2629 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2630 stmts);
2631 if (!genop0)
2632 return NULL_TREE;
2633 tree genop1 = find_or_generate_expression (block, currop->op0, stmts);
2634 if (!genop1)
2635 return NULL_TREE;
2636 return fold_build2 (currop->opcode, currop->type, genop0, genop1);
2639 case BIT_FIELD_REF:
2641 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2642 stmts);
2643 if (!genop0)
2644 return NULL_TREE;
2645 tree op1 = currop->op0;
2646 tree op2 = currop->op1;
2647 tree t = build3 (BIT_FIELD_REF, currop->type, genop0, op1, op2);
2648 REF_REVERSE_STORAGE_ORDER (t) = currop->reverse;
2649 return fold (t);
2652 /* For array ref vn_reference_op's, operand 1 of the array ref
2653 is op0 of the reference op and operand 3 of the array ref is
2654 op1. */
2655 case ARRAY_RANGE_REF:
2656 case ARRAY_REF:
2658 tree genop0;
2659 tree genop1 = currop->op0;
2660 tree genop2 = currop->op1;
2661 tree genop3 = currop->op2;
2662 genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2663 stmts);
2664 if (!genop0)
2665 return NULL_TREE;
2666 genop1 = find_or_generate_expression (block, genop1, stmts);
2667 if (!genop1)
2668 return NULL_TREE;
2669 if (genop2)
2671 tree domain_type = TYPE_DOMAIN (TREE_TYPE (genop0));
2672 /* Drop zero minimum index if redundant. */
2673 if (integer_zerop (genop2)
2674 && (!domain_type
2675 || integer_zerop (TYPE_MIN_VALUE (domain_type))))
2676 genop2 = NULL_TREE;
2677 else
2679 genop2 = find_or_generate_expression (block, genop2, stmts);
2680 if (!genop2)
2681 return NULL_TREE;
2684 if (genop3)
2686 tree elmt_type = TREE_TYPE (TREE_TYPE (genop0));
2687 /* We can't always put a size in units of the element alignment
2688 here as the element alignment may be not visible. See
2689 PR43783. Simply drop the element size for constant
2690 sizes. */
2691 if (TREE_CODE (genop3) == INTEGER_CST
2692 && TREE_CODE (TYPE_SIZE_UNIT (elmt_type)) == INTEGER_CST
2693 && wi::eq_p (wi::to_offset (TYPE_SIZE_UNIT (elmt_type)),
2694 (wi::to_offset (genop3)
2695 * vn_ref_op_align_unit (currop))))
2696 genop3 = NULL_TREE;
2697 else
2699 genop3 = find_or_generate_expression (block, genop3, stmts);
2700 if (!genop3)
2701 return NULL_TREE;
2704 return build4 (currop->opcode, currop->type, genop0, genop1,
2705 genop2, genop3);
2707 case COMPONENT_REF:
2709 tree op0;
2710 tree op1;
2711 tree genop2 = currop->op1;
2712 op0 = create_component_ref_by_pieces_1 (block, ref, operand, stmts);
2713 if (!op0)
2714 return NULL_TREE;
2715 /* op1 should be a FIELD_DECL, which are represented by themselves. */
2716 op1 = currop->op0;
2717 if (genop2)
2719 genop2 = find_or_generate_expression (block, genop2, stmts);
2720 if (!genop2)
2721 return NULL_TREE;
2723 return fold_build3 (COMPONENT_REF, TREE_TYPE (op1), op0, op1, genop2);
2726 case SSA_NAME:
2728 genop = find_or_generate_expression (block, currop->op0, stmts);
2729 return genop;
2731 case STRING_CST:
2732 case INTEGER_CST:
2733 case POLY_INT_CST:
2734 case COMPLEX_CST:
2735 case VECTOR_CST:
2736 case REAL_CST:
2737 case CONSTRUCTOR:
2738 case VAR_DECL:
2739 case PARM_DECL:
2740 case CONST_DECL:
2741 case RESULT_DECL:
2742 case FUNCTION_DECL:
2743 return currop->op0;
2745 default:
2746 gcc_unreachable ();
2750 /* For COMPONENT_REF's and ARRAY_REF's, we can't have any intermediates for the
2751 COMPONENT_REF or MEM_REF or ARRAY_REF portion, because we'd end up with
2752 trying to rename aggregates into ssa form directly, which is a no no.
2754 Thus, this routine doesn't create temporaries, it just builds a
2755 single access expression for the array, calling
2756 find_or_generate_expression to build the innermost pieces.
2758 This function is a subroutine of create_expression_by_pieces, and
2759 should not be called on it's own unless you really know what you
2760 are doing. */
2762 static tree
2763 create_component_ref_by_pieces (basic_block block, vn_reference_t ref,
2764 gimple_seq *stmts)
2766 unsigned int op = 0;
2767 return create_component_ref_by_pieces_1 (block, ref, &op, stmts);
2770 /* Find a simple leader for an expression, or generate one using
2771 create_expression_by_pieces from a NARY expression for the value.
2772 BLOCK is the basic_block we are looking for leaders in.
2773 OP is the tree expression to find a leader for or generate.
2774 Returns the leader or NULL_TREE on failure. */
2776 static tree
2777 find_or_generate_expression (basic_block block, tree op, gimple_seq *stmts)
2779 pre_expr expr = get_or_alloc_expr_for (op);
2780 unsigned int lookfor = get_expr_value_id (expr);
2781 pre_expr leader = bitmap_find_leader (AVAIL_OUT (block), lookfor);
2782 if (leader)
2784 if (leader->kind == NAME)
2785 return PRE_EXPR_NAME (leader);
2786 else if (leader->kind == CONSTANT)
2787 return PRE_EXPR_CONSTANT (leader);
2789 /* Defer. */
2790 return NULL_TREE;
2793 /* It must be a complex expression, so generate it recursively. Note
2794 that this is only necessary to handle gcc.dg/tree-ssa/ssa-pre28.c
2795 where the insert algorithm fails to insert a required expression. */
2796 bitmap exprset = value_expressions[lookfor];
2797 bitmap_iterator bi;
2798 unsigned int i;
2799 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
2801 pre_expr temp = expression_for_id (i);
2802 /* We cannot insert random REFERENCE expressions at arbitrary
2803 places. We can insert NARYs which eventually re-materializes
2804 its operand values. */
2805 if (temp->kind == NARY)
2806 return create_expression_by_pieces (block, temp, stmts,
2807 get_expr_type (expr));
2810 /* Defer. */
2811 return NULL_TREE;
2814 /* Create an expression in pieces, so that we can handle very complex
2815 expressions that may be ANTIC, but not necessary GIMPLE.
2816 BLOCK is the basic block the expression will be inserted into,
2817 EXPR is the expression to insert (in value form)
2818 STMTS is a statement list to append the necessary insertions into.
2820 This function will die if we hit some value that shouldn't be
2821 ANTIC but is (IE there is no leader for it, or its components).
2822 The function returns NULL_TREE in case a different antic expression
2823 has to be inserted first.
2824 This function may also generate expressions that are themselves
2825 partially or fully redundant. Those that are will be either made
2826 fully redundant during the next iteration of insert (for partially
2827 redundant ones), or eliminated by eliminate (for fully redundant
2828 ones). */
2830 static tree
2831 create_expression_by_pieces (basic_block block, pre_expr expr,
2832 gimple_seq *stmts, tree type)
2834 tree name;
2835 tree folded;
2836 gimple_seq forced_stmts = NULL;
2837 unsigned int value_id;
2838 gimple_stmt_iterator gsi;
2839 tree exprtype = type ? type : get_expr_type (expr);
2840 pre_expr nameexpr;
2841 gassign *newstmt;
2843 switch (expr->kind)
2845 /* We may hit the NAME/CONSTANT case if we have to convert types
2846 that value numbering saw through. */
2847 case NAME:
2848 folded = PRE_EXPR_NAME (expr);
2849 if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (folded))
2850 return NULL_TREE;
2851 if (useless_type_conversion_p (exprtype, TREE_TYPE (folded)))
2852 return folded;
2853 break;
2854 case CONSTANT:
2856 folded = PRE_EXPR_CONSTANT (expr);
2857 tree tem = fold_convert (exprtype, folded);
2858 if (is_gimple_min_invariant (tem))
2859 return tem;
2860 break;
2862 case REFERENCE:
2863 if (PRE_EXPR_REFERENCE (expr)->operands[0].opcode == CALL_EXPR)
2865 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2866 unsigned int operand = 1;
2867 vn_reference_op_t currop = &ref->operands[0];
2868 tree sc = NULL_TREE;
2869 tree fn = NULL_TREE;
2870 if (currop->op0)
2872 fn = find_or_generate_expression (block, currop->op0, stmts);
2873 if (!fn)
2874 return NULL_TREE;
2876 if (currop->op1)
2878 sc = find_or_generate_expression (block, currop->op1, stmts);
2879 if (!sc)
2880 return NULL_TREE;
2882 auto_vec<tree> args (ref->operands.length () - 1);
2883 while (operand < ref->operands.length ())
2885 tree arg = create_component_ref_by_pieces_1 (block, ref,
2886 &operand, stmts);
2887 if (!arg)
2888 return NULL_TREE;
2889 args.quick_push (arg);
2891 gcall *call;
2892 if (currop->op0)
2894 call = gimple_build_call_vec (fn, args);
2895 gimple_call_set_fntype (call, currop->type);
2897 else
2898 call = gimple_build_call_internal_vec ((internal_fn)currop->clique,
2899 args);
2900 gimple_set_location (call, expr->loc);
2901 if (sc)
2902 gimple_call_set_chain (call, sc);
2903 tree forcedname = make_ssa_name (ref->type);
2904 gimple_call_set_lhs (call, forcedname);
2905 /* There's no CCP pass after PRE which would re-compute alignment
2906 information so make sure we re-materialize this here. */
2907 if (gimple_call_builtin_p (call, BUILT_IN_ASSUME_ALIGNED)
2908 && args.length () - 2 <= 1
2909 && tree_fits_uhwi_p (args[1])
2910 && (args.length () != 3 || tree_fits_uhwi_p (args[2])))
2912 unsigned HOST_WIDE_INT halign = tree_to_uhwi (args[1]);
2913 unsigned HOST_WIDE_INT hmisalign
2914 = args.length () == 3 ? tree_to_uhwi (args[2]) : 0;
2915 if ((halign & (halign - 1)) == 0
2916 && (hmisalign & ~(halign - 1)) == 0
2917 && (unsigned int)halign != 0)
2918 set_ptr_info_alignment (get_ptr_info (forcedname),
2919 halign, hmisalign);
2921 gimple_set_vuse (call, BB_LIVE_VOP_ON_EXIT (block));
2922 gimple_seq_add_stmt_without_update (&forced_stmts, call);
2923 folded = forcedname;
2925 else
2927 folded = create_component_ref_by_pieces (block,
2928 PRE_EXPR_REFERENCE (expr),
2929 stmts);
2930 if (!folded)
2931 return NULL_TREE;
2932 name = make_temp_ssa_name (exprtype, NULL, "pretmp");
2933 newstmt = gimple_build_assign (name, folded);
2934 gimple_set_location (newstmt, expr->loc);
2935 gimple_seq_add_stmt_without_update (&forced_stmts, newstmt);
2936 gimple_set_vuse (newstmt, BB_LIVE_VOP_ON_EXIT (block));
2937 folded = name;
2939 break;
2940 case NARY:
2942 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2943 tree *genop = XALLOCAVEC (tree, nary->length);
2944 unsigned i;
2945 for (i = 0; i < nary->length; ++i)
2947 genop[i] = find_or_generate_expression (block, nary->op[i], stmts);
2948 if (!genop[i])
2949 return NULL_TREE;
2950 /* Ensure genop[] is properly typed for POINTER_PLUS_EXPR. It
2951 may have conversions stripped. */
2952 if (nary->opcode == POINTER_PLUS_EXPR)
2954 if (i == 0)
2955 genop[i] = gimple_convert (&forced_stmts,
2956 nary->type, genop[i]);
2957 else if (i == 1)
2958 genop[i] = gimple_convert (&forced_stmts,
2959 sizetype, genop[i]);
2961 else
2962 genop[i] = gimple_convert (&forced_stmts,
2963 TREE_TYPE (nary->op[i]), genop[i]);
2965 if (nary->opcode == CONSTRUCTOR)
2967 vec<constructor_elt, va_gc> *elts = NULL;
2968 for (i = 0; i < nary->length; ++i)
2969 CONSTRUCTOR_APPEND_ELT (elts, NULL_TREE, genop[i]);
2970 folded = build_constructor (nary->type, elts);
2971 name = make_temp_ssa_name (exprtype, NULL, "pretmp");
2972 newstmt = gimple_build_assign (name, folded);
2973 gimple_set_location (newstmt, expr->loc);
2974 gimple_seq_add_stmt_without_update (&forced_stmts, newstmt);
2975 folded = name;
2977 else
2979 switch (nary->length)
2981 case 1:
2982 folded = gimple_build (&forced_stmts, expr->loc,
2983 nary->opcode, nary->type, genop[0]);
2984 break;
2985 case 2:
2986 folded = gimple_build (&forced_stmts, expr->loc, nary->opcode,
2987 nary->type, genop[0], genop[1]);
2988 break;
2989 case 3:
2990 folded = gimple_build (&forced_stmts, expr->loc, nary->opcode,
2991 nary->type, genop[0], genop[1],
2992 genop[2]);
2993 break;
2994 default:
2995 gcc_unreachable ();
2999 break;
3000 default:
3001 gcc_unreachable ();
3004 folded = gimple_convert (&forced_stmts, exprtype, folded);
3006 /* If there is nothing to insert, return the simplified result. */
3007 if (gimple_seq_empty_p (forced_stmts))
3008 return folded;
3009 /* If we simplified to a constant return it and discard eventually
3010 built stmts. */
3011 if (is_gimple_min_invariant (folded))
3013 gimple_seq_discard (forced_stmts);
3014 return folded;
3016 /* Likewise if we simplified to sth not queued for insertion. */
3017 bool found = false;
3018 gsi = gsi_last (forced_stmts);
3019 for (; !gsi_end_p (gsi); gsi_prev (&gsi))
3021 gimple *stmt = gsi_stmt (gsi);
3022 tree forcedname = gimple_get_lhs (stmt);
3023 if (forcedname == folded)
3025 found = true;
3026 break;
3029 if (! found)
3031 gimple_seq_discard (forced_stmts);
3032 return folded;
3034 gcc_assert (TREE_CODE (folded) == SSA_NAME);
3036 /* If we have any intermediate expressions to the value sets, add them
3037 to the value sets and chain them in the instruction stream. */
3038 if (forced_stmts)
3040 gsi = gsi_start (forced_stmts);
3041 for (; !gsi_end_p (gsi); gsi_next (&gsi))
3043 gimple *stmt = gsi_stmt (gsi);
3044 tree forcedname = gimple_get_lhs (stmt);
3045 pre_expr nameexpr;
3047 if (forcedname != folded)
3049 vn_ssa_aux_t vn_info = VN_INFO (forcedname);
3050 vn_info->valnum = forcedname;
3051 vn_info->value_id = get_next_value_id ();
3052 nameexpr = get_or_alloc_expr_for_name (forcedname);
3053 add_to_value (vn_info->value_id, nameexpr);
3054 if (NEW_SETS (block))
3055 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
3056 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
3059 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (forcedname));
3061 gimple_seq_add_seq (stmts, forced_stmts);
3064 name = folded;
3066 /* Fold the last statement. */
3067 gsi = gsi_last (*stmts);
3068 if (fold_stmt_inplace (&gsi))
3069 update_stmt (gsi_stmt (gsi));
3071 /* Add a value number to the temporary.
3072 The value may already exist in either NEW_SETS, or AVAIL_OUT, because
3073 we are creating the expression by pieces, and this particular piece of
3074 the expression may have been represented. There is no harm in replacing
3075 here. */
3076 value_id = get_expr_value_id (expr);
3077 vn_ssa_aux_t vn_info = VN_INFO (name);
3078 vn_info->value_id = value_id;
3079 vn_info->valnum = vn_valnum_from_value_id (value_id);
3080 if (vn_info->valnum == NULL_TREE)
3081 vn_info->valnum = name;
3082 gcc_assert (vn_info->valnum != NULL_TREE);
3083 nameexpr = get_or_alloc_expr_for_name (name);
3084 add_to_value (value_id, nameexpr);
3085 if (NEW_SETS (block))
3086 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
3087 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
3089 pre_stats.insertions++;
3090 if (dump_file && (dump_flags & TDF_DETAILS))
3092 fprintf (dump_file, "Inserted ");
3093 print_gimple_stmt (dump_file, gsi_stmt (gsi_last (*stmts)), 0);
3094 fprintf (dump_file, " in predecessor %d (%04d)\n",
3095 block->index, value_id);
3098 return name;
3102 /* Insert the to-be-made-available values of expression EXPRNUM for each
3103 predecessor, stored in AVAIL, into the predecessors of BLOCK, and
3104 merge the result with a phi node, given the same value number as
3105 NODE. Return true if we have inserted new stuff. */
3107 static bool
3108 insert_into_preds_of_block (basic_block block, unsigned int exprnum,
3109 vec<pre_expr> &avail)
3111 pre_expr expr = expression_for_id (exprnum);
3112 pre_expr newphi;
3113 unsigned int val = get_expr_value_id (expr);
3114 edge pred;
3115 bool insertions = false;
3116 bool nophi = false;
3117 basic_block bprime;
3118 pre_expr eprime;
3119 edge_iterator ei;
3120 tree type = get_expr_type (expr);
3121 tree temp;
3122 gphi *phi;
3124 /* Make sure we aren't creating an induction variable. */
3125 if (bb_loop_depth (block) > 0 && EDGE_COUNT (block->preds) == 2)
3127 bool firstinsideloop = false;
3128 bool secondinsideloop = false;
3129 firstinsideloop = flow_bb_inside_loop_p (block->loop_father,
3130 EDGE_PRED (block, 0)->src);
3131 secondinsideloop = flow_bb_inside_loop_p (block->loop_father,
3132 EDGE_PRED (block, 1)->src);
3133 /* Induction variables only have one edge inside the loop. */
3134 if ((firstinsideloop ^ secondinsideloop)
3135 && expr->kind != REFERENCE)
3137 if (dump_file && (dump_flags & TDF_DETAILS))
3138 fprintf (dump_file, "Skipping insertion of phi for partial "
3139 "redundancy: Looks like an induction variable\n");
3140 nophi = true;
3144 /* Make the necessary insertions. */
3145 FOR_EACH_EDGE (pred, ei, block->preds)
3147 /* When we are not inserting a PHI node do not bother inserting
3148 into places that do not dominate the anticipated computations. */
3149 if (nophi && !dominated_by_p (CDI_DOMINATORS, block, pred->src))
3150 continue;
3151 gimple_seq stmts = NULL;
3152 tree builtexpr;
3153 bprime = pred->src;
3154 eprime = avail[pred->dest_idx];
3155 builtexpr = create_expression_by_pieces (bprime, eprime,
3156 &stmts, type);
3157 gcc_assert (!(pred->flags & EDGE_ABNORMAL));
3158 if (!gimple_seq_empty_p (stmts))
3160 basic_block new_bb = gsi_insert_seq_on_edge_immediate (pred, stmts);
3161 gcc_assert (! new_bb);
3162 insertions = true;
3164 if (!builtexpr)
3166 /* We cannot insert a PHI node if we failed to insert
3167 on one edge. */
3168 nophi = true;
3169 continue;
3171 if (is_gimple_min_invariant (builtexpr))
3172 avail[pred->dest_idx] = get_or_alloc_expr_for_constant (builtexpr);
3173 else
3174 avail[pred->dest_idx] = get_or_alloc_expr_for_name (builtexpr);
3176 /* If we didn't want a phi node, and we made insertions, we still have
3177 inserted new stuff, and thus return true. If we didn't want a phi node,
3178 and didn't make insertions, we haven't added anything new, so return
3179 false. */
3180 if (nophi && insertions)
3181 return true;
3182 else if (nophi && !insertions)
3183 return false;
3185 /* Now build a phi for the new variable. */
3186 temp = make_temp_ssa_name (type, NULL, "prephitmp");
3187 phi = create_phi_node (temp, block);
3189 vn_ssa_aux_t vn_info = VN_INFO (temp);
3190 vn_info->value_id = val;
3191 vn_info->valnum = vn_valnum_from_value_id (val);
3192 if (vn_info->valnum == NULL_TREE)
3193 vn_info->valnum = temp;
3194 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3195 FOR_EACH_EDGE (pred, ei, block->preds)
3197 pre_expr ae = avail[pred->dest_idx];
3198 gcc_assert (get_expr_type (ae) == type
3199 || useless_type_conversion_p (type, get_expr_type (ae)));
3200 if (ae->kind == CONSTANT)
3201 add_phi_arg (phi, unshare_expr (PRE_EXPR_CONSTANT (ae)),
3202 pred, UNKNOWN_LOCATION);
3203 else
3204 add_phi_arg (phi, PRE_EXPR_NAME (ae), pred, UNKNOWN_LOCATION);
3207 newphi = get_or_alloc_expr_for_name (temp);
3208 add_to_value (val, newphi);
3210 /* The value should *not* exist in PHI_GEN, or else we wouldn't be doing
3211 this insertion, since we test for the existence of this value in PHI_GEN
3212 before proceeding with the partial redundancy checks in insert_aux.
3214 The value may exist in AVAIL_OUT, in particular, it could be represented
3215 by the expression we are trying to eliminate, in which case we want the
3216 replacement to occur. If it's not existing in AVAIL_OUT, we want it
3217 inserted there.
3219 Similarly, to the PHI_GEN case, the value should not exist in NEW_SETS of
3220 this block, because if it did, it would have existed in our dominator's
3221 AVAIL_OUT, and would have been skipped due to the full redundancy check.
3224 bitmap_insert_into_set (PHI_GEN (block), newphi);
3225 bitmap_value_replace_in_set (AVAIL_OUT (block),
3226 newphi);
3227 if (NEW_SETS (block))
3228 bitmap_insert_into_set (NEW_SETS (block), newphi);
3230 /* If we insert a PHI node for a conversion of another PHI node
3231 in the same basic-block try to preserve range information.
3232 This is important so that followup loop passes receive optimal
3233 number of iteration analysis results. See PR61743. */
3234 if (expr->kind == NARY
3235 && CONVERT_EXPR_CODE_P (expr->u.nary->opcode)
3236 && TREE_CODE (expr->u.nary->op[0]) == SSA_NAME
3237 && gimple_bb (SSA_NAME_DEF_STMT (expr->u.nary->op[0])) == block
3238 && INTEGRAL_TYPE_P (type)
3239 && INTEGRAL_TYPE_P (TREE_TYPE (expr->u.nary->op[0]))
3240 && (TYPE_PRECISION (type)
3241 >= TYPE_PRECISION (TREE_TYPE (expr->u.nary->op[0])))
3242 && SSA_NAME_RANGE_INFO (expr->u.nary->op[0]))
3244 value_range r;
3245 if (get_range_query (cfun)->range_of_expr (r, expr->u.nary->op[0])
3246 && r.kind () == VR_RANGE
3247 && !wi::neg_p (r.lower_bound (), SIGNED)
3248 && !wi::neg_p (r.upper_bound (), SIGNED))
3249 /* Just handle extension and sign-changes of all-positive ranges. */
3250 set_range_info (temp, VR_RANGE,
3251 wide_int_storage::from (r.lower_bound (),
3252 TYPE_PRECISION (type),
3253 TYPE_SIGN (type)),
3254 wide_int_storage::from (r.upper_bound (),
3255 TYPE_PRECISION (type),
3256 TYPE_SIGN (type)));
3259 if (dump_file && (dump_flags & TDF_DETAILS))
3261 fprintf (dump_file, "Created phi ");
3262 print_gimple_stmt (dump_file, phi, 0);
3263 fprintf (dump_file, " in block %d (%04d)\n", block->index, val);
3265 pre_stats.phis++;
3266 return true;
3271 /* Perform insertion of partially redundant or hoistable values.
3272 For BLOCK, do the following:
3273 1. Propagate the NEW_SETS of the dominator into the current block.
3274 If the block has multiple predecessors,
3275 2a. Iterate over the ANTIC expressions for the block to see if
3276 any of them are partially redundant.
3277 2b. If so, insert them into the necessary predecessors to make
3278 the expression fully redundant.
3279 2c. Insert a new PHI merging the values of the predecessors.
3280 2d. Insert the new PHI, and the new expressions, into the
3281 NEW_SETS set.
3282 If the block has multiple successors,
3283 3a. Iterate over the ANTIC values for the block to see if
3284 any of them are good candidates for hoisting.
3285 3b. If so, insert expressions computing the values in BLOCK,
3286 and add the new expressions into the NEW_SETS set.
3287 4. Recursively call ourselves on the dominator children of BLOCK.
3289 Steps 1, 2a, and 4 are done by insert_aux. 2b, 2c and 2d are done by
3290 do_pre_regular_insertion and do_partial_insertion. 3a and 3b are
3291 done in do_hoist_insertion.
3294 static bool
3295 do_pre_regular_insertion (basic_block block, basic_block dom,
3296 vec<pre_expr> exprs)
3298 bool new_stuff = false;
3299 pre_expr expr;
3300 auto_vec<pre_expr, 2> avail;
3301 int i;
3303 avail.safe_grow (EDGE_COUNT (block->preds), true);
3305 FOR_EACH_VEC_ELT (exprs, i, expr)
3307 if (expr->kind == NARY
3308 || expr->kind == REFERENCE)
3310 unsigned int val;
3311 bool by_some = false;
3312 bool cant_insert = false;
3313 bool all_same = true;
3314 pre_expr first_s = NULL;
3315 edge pred;
3316 basic_block bprime;
3317 pre_expr eprime = NULL;
3318 edge_iterator ei;
3319 pre_expr edoubleprime = NULL;
3320 bool do_insertion = false;
3322 val = get_expr_value_id (expr);
3323 if (bitmap_set_contains_value (PHI_GEN (block), val))
3324 continue;
3325 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3327 if (dump_file && (dump_flags & TDF_DETAILS))
3329 fprintf (dump_file, "Found fully redundant value: ");
3330 print_pre_expr (dump_file, expr);
3331 fprintf (dump_file, "\n");
3333 continue;
3336 FOR_EACH_EDGE (pred, ei, block->preds)
3338 unsigned int vprime;
3340 /* We should never run insertion for the exit block
3341 and so not come across fake pred edges. */
3342 gcc_assert (!(pred->flags & EDGE_FAKE));
3343 bprime = pred->src;
3344 /* We are looking at ANTIC_OUT of bprime. */
3345 eprime = phi_translate (NULL, expr, ANTIC_IN (block), NULL, pred);
3347 /* eprime will generally only be NULL if the
3348 value of the expression, translated
3349 through the PHI for this predecessor, is
3350 undefined. If that is the case, we can't
3351 make the expression fully redundant,
3352 because its value is undefined along a
3353 predecessor path. We can thus break out
3354 early because it doesn't matter what the
3355 rest of the results are. */
3356 if (eprime == NULL)
3358 avail[pred->dest_idx] = NULL;
3359 cant_insert = true;
3360 break;
3363 vprime = get_expr_value_id (eprime);
3364 edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime),
3365 vprime);
3366 if (edoubleprime == NULL)
3368 avail[pred->dest_idx] = eprime;
3369 all_same = false;
3371 else
3373 avail[pred->dest_idx] = edoubleprime;
3374 by_some = true;
3375 /* We want to perform insertions to remove a redundancy on
3376 a path in the CFG we want to optimize for speed. */
3377 if (optimize_edge_for_speed_p (pred))
3378 do_insertion = true;
3379 if (first_s == NULL)
3380 first_s = edoubleprime;
3381 else if (!pre_expr_d::equal (first_s, edoubleprime))
3382 all_same = false;
3385 /* If we can insert it, it's not the same value
3386 already existing along every predecessor, and
3387 it's defined by some predecessor, it is
3388 partially redundant. */
3389 if (!cant_insert && !all_same && by_some)
3391 if (!do_insertion)
3393 if (dump_file && (dump_flags & TDF_DETAILS))
3395 fprintf (dump_file, "Skipping partial redundancy for "
3396 "expression ");
3397 print_pre_expr (dump_file, expr);
3398 fprintf (dump_file, " (%04d), no redundancy on to be "
3399 "optimized for speed edge\n", val);
3402 else if (dbg_cnt (treepre_insert))
3404 if (dump_file && (dump_flags & TDF_DETAILS))
3406 fprintf (dump_file, "Found partial redundancy for "
3407 "expression ");
3408 print_pre_expr (dump_file, expr);
3409 fprintf (dump_file, " (%04d)\n",
3410 get_expr_value_id (expr));
3412 if (insert_into_preds_of_block (block,
3413 get_expression_id (expr),
3414 avail))
3415 new_stuff = true;
3418 /* If all edges produce the same value and that value is
3419 an invariant, then the PHI has the same value on all
3420 edges. Note this. */
3421 else if (!cant_insert
3422 && all_same
3423 && (edoubleprime->kind != NAME
3424 || !SSA_NAME_OCCURS_IN_ABNORMAL_PHI
3425 (PRE_EXPR_NAME (edoubleprime))))
3427 gcc_assert (edoubleprime->kind == CONSTANT
3428 || edoubleprime->kind == NAME);
3430 tree temp = make_temp_ssa_name (get_expr_type (expr),
3431 NULL, "pretmp");
3432 gassign *assign
3433 = gimple_build_assign (temp,
3434 edoubleprime->kind == CONSTANT ?
3435 PRE_EXPR_CONSTANT (edoubleprime) :
3436 PRE_EXPR_NAME (edoubleprime));
3437 gimple_stmt_iterator gsi = gsi_after_labels (block);
3438 gsi_insert_before (&gsi, assign, GSI_NEW_STMT);
3440 vn_ssa_aux_t vn_info = VN_INFO (temp);
3441 vn_info->value_id = val;
3442 vn_info->valnum = vn_valnum_from_value_id (val);
3443 if (vn_info->valnum == NULL_TREE)
3444 vn_info->valnum = temp;
3445 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3446 pre_expr newe = get_or_alloc_expr_for_name (temp);
3447 add_to_value (val, newe);
3448 bitmap_value_replace_in_set (AVAIL_OUT (block), newe);
3449 bitmap_insert_into_set (NEW_SETS (block), newe);
3450 bitmap_insert_into_set (PHI_GEN (block), newe);
3455 return new_stuff;
3459 /* Perform insertion for partially anticipatable expressions. There
3460 is only one case we will perform insertion for these. This case is
3461 if the expression is partially anticipatable, and fully available.
3462 In this case, we know that putting it earlier will enable us to
3463 remove the later computation. */
3465 static bool
3466 do_pre_partial_partial_insertion (basic_block block, basic_block dom,
3467 vec<pre_expr> exprs)
3469 bool new_stuff = false;
3470 pre_expr expr;
3471 auto_vec<pre_expr, 2> avail;
3472 int i;
3474 avail.safe_grow (EDGE_COUNT (block->preds), true);
3476 FOR_EACH_VEC_ELT (exprs, i, expr)
3478 if (expr->kind == NARY
3479 || expr->kind == REFERENCE)
3481 unsigned int val;
3482 bool by_all = true;
3483 bool cant_insert = false;
3484 edge pred;
3485 basic_block bprime;
3486 pre_expr eprime = NULL;
3487 edge_iterator ei;
3489 val = get_expr_value_id (expr);
3490 if (bitmap_set_contains_value (PHI_GEN (block), val))
3491 continue;
3492 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3493 continue;
3495 FOR_EACH_EDGE (pred, ei, block->preds)
3497 unsigned int vprime;
3498 pre_expr edoubleprime;
3500 /* We should never run insertion for the exit block
3501 and so not come across fake pred edges. */
3502 gcc_assert (!(pred->flags & EDGE_FAKE));
3503 bprime = pred->src;
3504 eprime = phi_translate (NULL, expr, ANTIC_IN (block),
3505 PA_IN (block), pred);
3507 /* eprime will generally only be NULL if the
3508 value of the expression, translated
3509 through the PHI for this predecessor, is
3510 undefined. If that is the case, we can't
3511 make the expression fully redundant,
3512 because its value is undefined along a
3513 predecessor path. We can thus break out
3514 early because it doesn't matter what the
3515 rest of the results are. */
3516 if (eprime == NULL)
3518 avail[pred->dest_idx] = NULL;
3519 cant_insert = true;
3520 break;
3523 vprime = get_expr_value_id (eprime);
3524 edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime), vprime);
3525 avail[pred->dest_idx] = edoubleprime;
3526 if (edoubleprime == NULL)
3528 by_all = false;
3529 break;
3533 /* If we can insert it, it's not the same value
3534 already existing along every predecessor, and
3535 it's defined by some predecessor, it is
3536 partially redundant. */
3537 if (!cant_insert && by_all)
3539 edge succ;
3540 bool do_insertion = false;
3542 /* Insert only if we can remove a later expression on a path
3543 that we want to optimize for speed.
3544 The phi node that we will be inserting in BLOCK is not free,
3545 and inserting it for the sake of !optimize_for_speed successor
3546 may cause regressions on the speed path. */
3547 FOR_EACH_EDGE (succ, ei, block->succs)
3549 if (bitmap_set_contains_value (PA_IN (succ->dest), val)
3550 || bitmap_set_contains_value (ANTIC_IN (succ->dest), val))
3552 if (optimize_edge_for_speed_p (succ))
3553 do_insertion = true;
3557 if (!do_insertion)
3559 if (dump_file && (dump_flags & TDF_DETAILS))
3561 fprintf (dump_file, "Skipping partial partial redundancy "
3562 "for expression ");
3563 print_pre_expr (dump_file, expr);
3564 fprintf (dump_file, " (%04d), not (partially) anticipated "
3565 "on any to be optimized for speed edges\n", val);
3568 else if (dbg_cnt (treepre_insert))
3570 pre_stats.pa_insert++;
3571 if (dump_file && (dump_flags & TDF_DETAILS))
3573 fprintf (dump_file, "Found partial partial redundancy "
3574 "for expression ");
3575 print_pre_expr (dump_file, expr);
3576 fprintf (dump_file, " (%04d)\n",
3577 get_expr_value_id (expr));
3579 if (insert_into_preds_of_block (block,
3580 get_expression_id (expr),
3581 avail))
3582 new_stuff = true;
3588 return new_stuff;
3591 /* Insert expressions in BLOCK to compute hoistable values up.
3592 Return TRUE if something was inserted, otherwise return FALSE.
3593 The caller has to make sure that BLOCK has at least two successors. */
3595 static bool
3596 do_hoist_insertion (basic_block block)
3598 edge e;
3599 edge_iterator ei;
3600 bool new_stuff = false;
3601 unsigned i;
3602 gimple_stmt_iterator last;
3604 /* At least two successors, or else... */
3605 gcc_assert (EDGE_COUNT (block->succs) >= 2);
3607 /* Check that all successors of BLOCK are dominated by block.
3608 We could use dominated_by_p() for this, but actually there is a much
3609 quicker check: any successor that is dominated by BLOCK can't have
3610 more than one predecessor edge. */
3611 FOR_EACH_EDGE (e, ei, block->succs)
3612 if (! single_pred_p (e->dest))
3613 return false;
3615 /* Determine the insertion point. If we cannot safely insert before
3616 the last stmt if we'd have to, bail out. */
3617 last = gsi_last_bb (block);
3618 if (!gsi_end_p (last)
3619 && !is_ctrl_stmt (gsi_stmt (last))
3620 && stmt_ends_bb_p (gsi_stmt (last)))
3621 return false;
3623 /* Compute the set of hoistable expressions from ANTIC_IN. First compute
3624 hoistable values. */
3625 bitmap_set hoistable_set;
3627 /* A hoistable value must be in ANTIC_IN(block)
3628 but not in AVAIL_OUT(BLOCK). */
3629 bitmap_initialize (&hoistable_set.values, &grand_bitmap_obstack);
3630 bitmap_and_compl (&hoistable_set.values,
3631 &ANTIC_IN (block)->values, &AVAIL_OUT (block)->values);
3633 /* Short-cut for a common case: hoistable_set is empty. */
3634 if (bitmap_empty_p (&hoistable_set.values))
3635 return false;
3637 /* Compute which of the hoistable values is in AVAIL_OUT of
3638 at least one of the successors of BLOCK. */
3639 bitmap_head availout_in_some;
3640 bitmap_initialize (&availout_in_some, &grand_bitmap_obstack);
3641 FOR_EACH_EDGE (e, ei, block->succs)
3642 /* Do not consider expressions solely because their availability
3643 on loop exits. They'd be ANTIC-IN throughout the whole loop
3644 and thus effectively hoisted across loops by combination of
3645 PRE and hoisting. */
3646 if (! loop_exit_edge_p (block->loop_father, e))
3647 bitmap_ior_and_into (&availout_in_some, &hoistable_set.values,
3648 &AVAIL_OUT (e->dest)->values);
3649 bitmap_clear (&hoistable_set.values);
3651 /* Short-cut for a common case: availout_in_some is empty. */
3652 if (bitmap_empty_p (&availout_in_some))
3653 return false;
3655 /* Hack hoitable_set in-place so we can use sorted_array_from_bitmap_set. */
3656 bitmap_move (&hoistable_set.values, &availout_in_some);
3657 hoistable_set.expressions = ANTIC_IN (block)->expressions;
3659 /* Now finally construct the topological-ordered expression set. */
3660 vec<pre_expr> exprs = sorted_array_from_bitmap_set (&hoistable_set);
3662 bitmap_clear (&hoistable_set.values);
3664 /* If there are candidate values for hoisting, insert expressions
3665 strategically to make the hoistable expressions fully redundant. */
3666 pre_expr expr;
3667 FOR_EACH_VEC_ELT (exprs, i, expr)
3669 /* While we try to sort expressions topologically above the
3670 sorting doesn't work out perfectly. Catch expressions we
3671 already inserted. */
3672 unsigned int value_id = get_expr_value_id (expr);
3673 if (bitmap_set_contains_value (AVAIL_OUT (block), value_id))
3675 if (dump_file && (dump_flags & TDF_DETAILS))
3677 fprintf (dump_file,
3678 "Already inserted expression for ");
3679 print_pre_expr (dump_file, expr);
3680 fprintf (dump_file, " (%04d)\n", value_id);
3682 continue;
3685 /* If we end up with a punned expression representation and this
3686 happens to be a float typed one give up - we can't know for
3687 sure whether all paths perform the floating-point load we are
3688 about to insert and on some targets this can cause correctness
3689 issues. See PR88240. */
3690 if (expr->kind == REFERENCE
3691 && PRE_EXPR_REFERENCE (expr)->punned
3692 && FLOAT_TYPE_P (get_expr_type (expr)))
3693 continue;
3695 /* OK, we should hoist this value. Perform the transformation. */
3696 pre_stats.hoist_insert++;
3697 if (dump_file && (dump_flags & TDF_DETAILS))
3699 fprintf (dump_file,
3700 "Inserting expression in block %d for code hoisting: ",
3701 block->index);
3702 print_pre_expr (dump_file, expr);
3703 fprintf (dump_file, " (%04d)\n", value_id);
3706 gimple_seq stmts = NULL;
3707 tree res = create_expression_by_pieces (block, expr, &stmts,
3708 get_expr_type (expr));
3710 /* Do not return true if expression creation ultimately
3711 did not insert any statements. */
3712 if (gimple_seq_empty_p (stmts))
3713 res = NULL_TREE;
3714 else
3716 if (gsi_end_p (last) || is_ctrl_stmt (gsi_stmt (last)))
3717 gsi_insert_seq_before (&last, stmts, GSI_SAME_STMT);
3718 else
3719 gsi_insert_seq_after (&last, stmts, GSI_NEW_STMT);
3722 /* Make sure to not return true if expression creation ultimately
3723 failed but also make sure to insert any stmts produced as they
3724 are tracked in inserted_exprs. */
3725 if (! res)
3726 continue;
3728 new_stuff = true;
3731 exprs.release ();
3733 return new_stuff;
3736 /* Perform insertion of partially redundant and hoistable values. */
3738 static void
3739 insert (void)
3741 basic_block bb;
3743 FOR_ALL_BB_FN (bb, cfun)
3744 NEW_SETS (bb) = bitmap_set_new ();
3746 int *rpo = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
3747 int *bb_rpo = XNEWVEC (int, last_basic_block_for_fn (cfun) + 1);
3748 int rpo_num = pre_and_rev_post_order_compute (NULL, rpo, false);
3749 for (int i = 0; i < rpo_num; ++i)
3750 bb_rpo[rpo[i]] = i;
3752 int num_iterations = 0;
3753 bool changed;
3756 num_iterations++;
3757 if (dump_file && dump_flags & TDF_DETAILS)
3758 fprintf (dump_file, "Starting insert iteration %d\n", num_iterations);
3760 changed = false;
3761 for (int idx = 0; idx < rpo_num; ++idx)
3763 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[idx]);
3764 basic_block dom = get_immediate_dominator (CDI_DOMINATORS, block);
3765 if (dom)
3767 unsigned i;
3768 bitmap_iterator bi;
3769 bitmap_set_t newset;
3771 /* First, update the AVAIL_OUT set with anything we may have
3772 inserted higher up in the dominator tree. */
3773 newset = NEW_SETS (dom);
3775 /* Note that we need to value_replace both NEW_SETS, and
3776 AVAIL_OUT. For both the case of NEW_SETS, the value may be
3777 represented by some non-simple expression here that we want
3778 to replace it with. */
3779 bool avail_out_changed = false;
3780 FOR_EACH_EXPR_ID_IN_SET (newset, i, bi)
3782 pre_expr expr = expression_for_id (i);
3783 bitmap_value_replace_in_set (NEW_SETS (block), expr);
3784 avail_out_changed
3785 |= bitmap_value_replace_in_set (AVAIL_OUT (block), expr);
3787 /* We need to iterate if AVAIL_OUT of an already processed
3788 block source changed. */
3789 if (avail_out_changed && !changed)
3791 edge_iterator ei;
3792 edge e;
3793 FOR_EACH_EDGE (e, ei, block->succs)
3794 if (e->dest->index != EXIT_BLOCK
3795 && bb_rpo[e->dest->index] < idx)
3796 changed = true;
3799 /* Insert expressions for partial redundancies. */
3800 if (flag_tree_pre && !single_pred_p (block))
3802 vec<pre_expr> exprs
3803 = sorted_array_from_bitmap_set (ANTIC_IN (block));
3804 /* Sorting is not perfect, iterate locally. */
3805 while (do_pre_regular_insertion (block, dom, exprs))
3807 exprs.release ();
3808 if (do_partial_partial)
3810 exprs = sorted_array_from_bitmap_set (PA_IN (block));
3811 while (do_pre_partial_partial_insertion (block, dom,
3812 exprs))
3814 exprs.release ();
3820 /* Clear the NEW sets before the next iteration. We have already
3821 fully propagated its contents. */
3822 if (changed)
3823 FOR_ALL_BB_FN (bb, cfun)
3824 bitmap_set_free (NEW_SETS (bb));
3826 while (changed);
3828 statistics_histogram_event (cfun, "insert iterations", num_iterations);
3830 /* AVAIL_OUT is not needed after insertion so we don't have to
3831 propagate NEW_SETS from hoist insertion. */
3832 FOR_ALL_BB_FN (bb, cfun)
3834 bitmap_set_free (NEW_SETS (bb));
3835 bitmap_set_pool.remove (NEW_SETS (bb));
3836 NEW_SETS (bb) = NULL;
3839 /* Insert expressions for hoisting. Do a backward walk here since
3840 inserting into BLOCK exposes new opportunities in its predecessors.
3841 Since PRE and hoist insertions can cause back-to-back iteration
3842 and we are interested in PRE insertion exposed hoisting opportunities
3843 but not in hoisting exposed PRE ones do hoist insertion only after
3844 PRE insertion iteration finished and do not iterate it. */
3845 if (flag_code_hoisting)
3846 for (int idx = rpo_num - 1; idx >= 0; --idx)
3848 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[idx]);
3849 if (EDGE_COUNT (block->succs) >= 2)
3850 changed |= do_hoist_insertion (block);
3853 free (rpo);
3854 free (bb_rpo);
3858 /* Compute the AVAIL set for all basic blocks.
3860 This function performs value numbering of the statements in each basic
3861 block. The AVAIL sets are built from information we glean while doing
3862 this value numbering, since the AVAIL sets contain only one entry per
3863 value.
3865 AVAIL_IN[BLOCK] = AVAIL_OUT[dom(BLOCK)].
3866 AVAIL_OUT[BLOCK] = AVAIL_IN[BLOCK] U PHI_GEN[BLOCK] U TMP_GEN[BLOCK]. */
3868 static void
3869 compute_avail (function *fun)
3872 basic_block block, son;
3873 basic_block *worklist;
3874 size_t sp = 0;
3875 unsigned i;
3876 tree name;
3878 /* We pretend that default definitions are defined in the entry block.
3879 This includes function arguments and the static chain decl. */
3880 FOR_EACH_SSA_NAME (i, name, fun)
3882 pre_expr e;
3883 if (!SSA_NAME_IS_DEFAULT_DEF (name)
3884 || has_zero_uses (name)
3885 || virtual_operand_p (name))
3886 continue;
3888 e = get_or_alloc_expr_for_name (name);
3889 add_to_value (get_expr_value_id (e), e);
3890 bitmap_insert_into_set (TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (fun)), e);
3891 bitmap_value_insert_into_set (AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3895 if (dump_file && (dump_flags & TDF_DETAILS))
3897 print_bitmap_set (dump_file, TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3898 "tmp_gen", ENTRY_BLOCK);
3899 print_bitmap_set (dump_file, AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3900 "avail_out", ENTRY_BLOCK);
3903 /* Allocate the worklist. */
3904 worklist = XNEWVEC (basic_block, n_basic_blocks_for_fn (fun));
3906 /* Seed the algorithm by putting the dominator children of the entry
3907 block on the worklist. */
3908 for (son = first_dom_son (CDI_DOMINATORS, ENTRY_BLOCK_PTR_FOR_FN (fun));
3909 son;
3910 son = next_dom_son (CDI_DOMINATORS, son))
3911 worklist[sp++] = son;
3913 BB_LIVE_VOP_ON_EXIT (ENTRY_BLOCK_PTR_FOR_FN (fun))
3914 = ssa_default_def (fun, gimple_vop (fun));
3916 /* Loop until the worklist is empty. */
3917 while (sp)
3919 gimple *stmt;
3920 basic_block dom;
3922 /* Pick a block from the worklist. */
3923 block = worklist[--sp];
3924 vn_context_bb = block;
3926 /* Initially, the set of available values in BLOCK is that of
3927 its immediate dominator. */
3928 dom = get_immediate_dominator (CDI_DOMINATORS, block);
3929 if (dom)
3931 bitmap_set_copy (AVAIL_OUT (block), AVAIL_OUT (dom));
3932 BB_LIVE_VOP_ON_EXIT (block) = BB_LIVE_VOP_ON_EXIT (dom);
3935 /* Generate values for PHI nodes. */
3936 for (gphi_iterator gsi = gsi_start_phis (block); !gsi_end_p (gsi);
3937 gsi_next (&gsi))
3939 tree result = gimple_phi_result (gsi.phi ());
3941 /* We have no need for virtual phis, as they don't represent
3942 actual computations. */
3943 if (virtual_operand_p (result))
3945 BB_LIVE_VOP_ON_EXIT (block) = result;
3946 continue;
3949 pre_expr e = get_or_alloc_expr_for_name (result);
3950 add_to_value (get_expr_value_id (e), e);
3951 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
3952 bitmap_insert_into_set (PHI_GEN (block), e);
3955 BB_MAY_NOTRETURN (block) = 0;
3957 /* Now compute value numbers and populate value sets with all
3958 the expressions computed in BLOCK. */
3959 bool set_bb_may_notreturn = false;
3960 for (gimple_stmt_iterator gsi = gsi_start_bb (block); !gsi_end_p (gsi);
3961 gsi_next (&gsi))
3963 ssa_op_iter iter;
3964 tree op;
3966 stmt = gsi_stmt (gsi);
3968 if (set_bb_may_notreturn)
3970 BB_MAY_NOTRETURN (block) = 1;
3971 set_bb_may_notreturn = false;
3974 /* Cache whether the basic-block has any non-visible side-effect
3975 or control flow.
3976 If this isn't a call or it is the last stmt in the
3977 basic-block then the CFG represents things correctly. */
3978 if (is_gimple_call (stmt) && !stmt_ends_bb_p (stmt))
3980 /* Non-looping const functions always return normally.
3981 Otherwise the call might not return or have side-effects
3982 that forbids hoisting possibly trapping expressions
3983 before it. */
3984 int flags = gimple_call_flags (stmt);
3985 if (!(flags & (ECF_CONST|ECF_PURE))
3986 || (flags & ECF_LOOPING_CONST_OR_PURE)
3987 || stmt_can_throw_external (fun, stmt))
3988 /* Defer setting of BB_MAY_NOTRETURN to avoid it
3989 influencing the processing of the call itself. */
3990 set_bb_may_notreturn = true;
3993 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_DEF)
3995 pre_expr e = get_or_alloc_expr_for_name (op);
3997 add_to_value (get_expr_value_id (e), e);
3998 bitmap_insert_into_set (TMP_GEN (block), e);
3999 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
4002 if (gimple_vdef (stmt))
4003 BB_LIVE_VOP_ON_EXIT (block) = gimple_vdef (stmt);
4005 if (gimple_has_side_effects (stmt)
4006 || stmt_could_throw_p (fun, stmt)
4007 || is_gimple_debug (stmt))
4008 continue;
4010 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_USE)
4012 if (ssa_undefined_value_p (op))
4013 continue;
4014 pre_expr e = get_or_alloc_expr_for_name (op);
4015 bitmap_value_insert_into_set (EXP_GEN (block), e);
4018 switch (gimple_code (stmt))
4020 case GIMPLE_RETURN:
4021 continue;
4023 case GIMPLE_CALL:
4025 vn_reference_t ref;
4026 vn_reference_s ref1;
4027 pre_expr result = NULL;
4029 vn_reference_lookup_call (as_a <gcall *> (stmt), &ref, &ref1);
4030 /* There is no point to PRE a call without a value. */
4031 if (!ref || !ref->result)
4032 continue;
4034 /* If the value of the call is not invalidated in
4035 this block until it is computed, add the expression
4036 to EXP_GEN. */
4037 if ((!gimple_vuse (stmt)
4038 || gimple_code
4039 (SSA_NAME_DEF_STMT (gimple_vuse (stmt))) == GIMPLE_PHI
4040 || gimple_bb (SSA_NAME_DEF_STMT
4041 (gimple_vuse (stmt))) != block)
4042 /* If the REFERENCE traps and there was a preceding
4043 point in the block that might not return avoid
4044 adding the reference to EXP_GEN. */
4045 && (!BB_MAY_NOTRETURN (block)
4046 || !vn_reference_may_trap (ref)))
4048 result = get_or_alloc_expr_for_reference
4049 (ref, gimple_location (stmt));
4050 add_to_value (get_expr_value_id (result), result);
4051 bitmap_value_insert_into_set (EXP_GEN (block), result);
4053 continue;
4056 case GIMPLE_ASSIGN:
4058 pre_expr result = NULL;
4059 switch (vn_get_stmt_kind (stmt))
4061 case VN_NARY:
4063 enum tree_code code = gimple_assign_rhs_code (stmt);
4064 vn_nary_op_t nary;
4066 /* COND_EXPR is awkward in that it contains an
4067 embedded complex expression.
4068 Don't even try to shove it through PRE. */
4069 if (code == COND_EXPR)
4070 continue;
4072 vn_nary_op_lookup_stmt (stmt, &nary);
4073 if (!nary || nary->predicated_values)
4074 continue;
4076 /* If the NARY traps and there was a preceding
4077 point in the block that might not return avoid
4078 adding the nary to EXP_GEN. */
4079 if (BB_MAY_NOTRETURN (block)
4080 && vn_nary_may_trap (nary))
4081 continue;
4083 result = get_or_alloc_expr_for_nary
4084 (nary, gimple_location (stmt));
4085 break;
4088 case VN_REFERENCE:
4090 tree rhs1 = gimple_assign_rhs1 (stmt);
4091 ao_ref rhs1_ref;
4092 ao_ref_init (&rhs1_ref, rhs1);
4093 alias_set_type set = ao_ref_alias_set (&rhs1_ref);
4094 alias_set_type base_set
4095 = ao_ref_base_alias_set (&rhs1_ref);
4096 vec<vn_reference_op_s> operands
4097 = vn_reference_operands_for_lookup (rhs1);
4098 vn_reference_t ref;
4099 vn_reference_lookup_pieces (gimple_vuse (stmt), set,
4100 base_set, TREE_TYPE (rhs1),
4101 operands, &ref, VN_WALK);
4102 if (!ref)
4104 operands.release ();
4105 continue;
4108 /* If the REFERENCE traps and there was a preceding
4109 point in the block that might not return avoid
4110 adding the reference to EXP_GEN. */
4111 if (BB_MAY_NOTRETURN (block)
4112 && vn_reference_may_trap (ref))
4114 operands.release ();
4115 continue;
4118 /* If the value of the reference is not invalidated in
4119 this block until it is computed, add the expression
4120 to EXP_GEN. */
4121 if (gimple_vuse (stmt))
4123 gimple *def_stmt;
4124 bool ok = true;
4125 def_stmt = SSA_NAME_DEF_STMT (gimple_vuse (stmt));
4126 while (!gimple_nop_p (def_stmt)
4127 && gimple_code (def_stmt) != GIMPLE_PHI
4128 && gimple_bb (def_stmt) == block)
4130 if (stmt_may_clobber_ref_p
4131 (def_stmt, gimple_assign_rhs1 (stmt)))
4133 ok = false;
4134 break;
4136 def_stmt
4137 = SSA_NAME_DEF_STMT (gimple_vuse (def_stmt));
4139 if (!ok)
4141 operands.release ();
4142 continue;
4146 /* If the load was value-numbered to another
4147 load make sure we do not use its expression
4148 for insertion if it wouldn't be a valid
4149 replacement. */
4150 /* At the momemt we have a testcase
4151 for hoist insertion of aligned vs. misaligned
4152 variants in gcc.dg/torture/pr65270-1.c thus
4153 with just alignment to be considered we can
4154 simply replace the expression in the hashtable
4155 with the most conservative one. */
4156 vn_reference_op_t ref1 = &ref->operands.last ();
4157 while (ref1->opcode != TARGET_MEM_REF
4158 && ref1->opcode != MEM_REF
4159 && ref1 != &ref->operands[0])
4160 --ref1;
4161 vn_reference_op_t ref2 = &operands.last ();
4162 while (ref2->opcode != TARGET_MEM_REF
4163 && ref2->opcode != MEM_REF
4164 && ref2 != &operands[0])
4165 --ref2;
4166 if ((ref1->opcode == TARGET_MEM_REF
4167 || ref1->opcode == MEM_REF)
4168 && (TYPE_ALIGN (ref1->type)
4169 > TYPE_ALIGN (ref2->type)))
4170 ref1->type
4171 = build_aligned_type (ref1->type,
4172 TYPE_ALIGN (ref2->type));
4173 /* TBAA behavior is an obvious part so make sure
4174 that the hashtable one covers this as well
4175 by adjusting the ref alias set and its base. */
4176 if (ref->set == set
4177 || alias_set_subset_of (set, ref->set))
4179 else if (ref1->opcode != ref2->opcode
4180 || (ref1->opcode != MEM_REF
4181 && ref1->opcode != TARGET_MEM_REF))
4183 /* With mismatching base opcodes or bases
4184 other than MEM_REF or TARGET_MEM_REF we
4185 can't do any easy TBAA adjustment. */
4186 operands.release ();
4187 continue;
4189 else if (alias_set_subset_of (ref->set, set))
4191 ref->set = set;
4192 if (ref1->opcode == MEM_REF)
4193 ref1->op0
4194 = wide_int_to_tree (TREE_TYPE (ref2->op0),
4195 wi::to_wide (ref1->op0));
4196 else
4197 ref1->op2
4198 = wide_int_to_tree (TREE_TYPE (ref2->op2),
4199 wi::to_wide (ref1->op2));
4201 else
4203 ref->set = 0;
4204 if (ref1->opcode == MEM_REF)
4205 ref1->op0
4206 = wide_int_to_tree (ptr_type_node,
4207 wi::to_wide (ref1->op0));
4208 else
4209 ref1->op2
4210 = wide_int_to_tree (ptr_type_node,
4211 wi::to_wide (ref1->op2));
4213 operands.release ();
4215 result = get_or_alloc_expr_for_reference
4216 (ref, gimple_location (stmt));
4217 break;
4220 default:
4221 continue;
4224 add_to_value (get_expr_value_id (result), result);
4225 bitmap_value_insert_into_set (EXP_GEN (block), result);
4226 continue;
4228 default:
4229 break;
4232 if (set_bb_may_notreturn)
4234 BB_MAY_NOTRETURN (block) = 1;
4235 set_bb_may_notreturn = false;
4238 if (dump_file && (dump_flags & TDF_DETAILS))
4240 print_bitmap_set (dump_file, EXP_GEN (block),
4241 "exp_gen", block->index);
4242 print_bitmap_set (dump_file, PHI_GEN (block),
4243 "phi_gen", block->index);
4244 print_bitmap_set (dump_file, TMP_GEN (block),
4245 "tmp_gen", block->index);
4246 print_bitmap_set (dump_file, AVAIL_OUT (block),
4247 "avail_out", block->index);
4250 /* Put the dominator children of BLOCK on the worklist of blocks
4251 to compute available sets for. */
4252 for (son = first_dom_son (CDI_DOMINATORS, block);
4253 son;
4254 son = next_dom_son (CDI_DOMINATORS, son))
4255 worklist[sp++] = son;
4257 vn_context_bb = NULL;
4259 free (worklist);
4263 /* Initialize data structures used by PRE. */
4265 static void
4266 init_pre (void)
4268 basic_block bb;
4270 next_expression_id = 1;
4271 expressions.create (0);
4272 expressions.safe_push (NULL);
4273 value_expressions.create (get_max_value_id () + 1);
4274 value_expressions.quick_grow_cleared (get_max_value_id () + 1);
4275 constant_value_expressions.create (get_max_constant_value_id () + 1);
4276 constant_value_expressions.quick_grow_cleared (get_max_constant_value_id () + 1);
4277 name_to_id.create (0);
4279 inserted_exprs = BITMAP_ALLOC (NULL);
4281 connect_infinite_loops_to_exit ();
4282 memset (&pre_stats, 0, sizeof (pre_stats));
4284 alloc_aux_for_blocks (sizeof (struct bb_bitmap_sets));
4286 calculate_dominance_info (CDI_DOMINATORS);
4288 bitmap_obstack_initialize (&grand_bitmap_obstack);
4289 expression_to_id = new hash_table<pre_expr_d> (num_ssa_names * 3);
4290 FOR_ALL_BB_FN (bb, cfun)
4292 EXP_GEN (bb) = bitmap_set_new ();
4293 PHI_GEN (bb) = bitmap_set_new ();
4294 TMP_GEN (bb) = bitmap_set_new ();
4295 AVAIL_OUT (bb) = bitmap_set_new ();
4296 PHI_TRANS_TABLE (bb) = NULL;
4301 /* Deallocate data structures used by PRE. */
4303 static void
4304 fini_pre ()
4306 value_expressions.release ();
4307 constant_value_expressions.release ();
4308 expressions.release ();
4309 bitmap_obstack_release (&grand_bitmap_obstack);
4310 bitmap_set_pool.release ();
4311 pre_expr_pool.release ();
4312 delete expression_to_id;
4313 expression_to_id = NULL;
4314 name_to_id.release ();
4316 basic_block bb;
4317 FOR_ALL_BB_FN (bb, cfun)
4318 if (bb->aux && PHI_TRANS_TABLE (bb))
4319 delete PHI_TRANS_TABLE (bb);
4320 free_aux_for_blocks ();
4323 namespace {
4325 const pass_data pass_data_pre =
4327 GIMPLE_PASS, /* type */
4328 "pre", /* name */
4329 OPTGROUP_NONE, /* optinfo_flags */
4330 TV_TREE_PRE, /* tv_id */
4331 ( PROP_cfg | PROP_ssa ), /* properties_required */
4332 0, /* properties_provided */
4333 0, /* properties_destroyed */
4334 TODO_rebuild_alias, /* todo_flags_start */
4335 0, /* todo_flags_finish */
4338 class pass_pre : public gimple_opt_pass
4340 public:
4341 pass_pre (gcc::context *ctxt)
4342 : gimple_opt_pass (pass_data_pre, ctxt)
4345 /* opt_pass methods: */
4346 virtual bool gate (function *)
4347 { return flag_tree_pre != 0 || flag_code_hoisting != 0; }
4348 virtual unsigned int execute (function *);
4350 }; // class pass_pre
4352 /* Valueization hook for RPO VN when we are calling back to it
4353 at ANTIC compute time. */
4355 static tree
4356 pre_valueize (tree name)
4358 if (TREE_CODE (name) == SSA_NAME)
4360 tree tem = VN_INFO (name)->valnum;
4361 if (tem != VN_TOP && tem != name)
4363 if (TREE_CODE (tem) != SSA_NAME
4364 || SSA_NAME_IS_DEFAULT_DEF (tem))
4365 return tem;
4366 /* We create temporary SSA names for representatives that
4367 do not have a definition (yet) but are not default defs either
4368 assume they are fine to use. */
4369 basic_block def_bb = gimple_bb (SSA_NAME_DEF_STMT (tem));
4370 if (! def_bb
4371 || dominated_by_p (CDI_DOMINATORS, vn_context_bb, def_bb))
4372 return tem;
4373 /* ??? Now we could look for a leader. Ideally we'd somehow
4374 expose RPO VN leaders and get rid of AVAIL_OUT as well... */
4377 return name;
4380 unsigned int
4381 pass_pre::execute (function *fun)
4383 unsigned int todo = 0;
4385 do_partial_partial =
4386 flag_tree_partial_pre && optimize_function_for_speed_p (fun);
4388 /* This has to happen before VN runs because
4389 loop_optimizer_init may create new phis, etc. */
4390 loop_optimizer_init (LOOPS_NORMAL);
4391 split_edges_for_insertion ();
4392 scev_initialize ();
4393 calculate_dominance_info (CDI_DOMINATORS);
4395 run_rpo_vn (VN_WALK);
4397 init_pre ();
4399 vn_valueize = pre_valueize;
4401 /* Insert can get quite slow on an incredibly large number of basic
4402 blocks due to some quadratic behavior. Until this behavior is
4403 fixed, don't run it when he have an incredibly large number of
4404 bb's. If we aren't going to run insert, there is no point in
4405 computing ANTIC, either, even though it's plenty fast nor do
4406 we require AVAIL. */
4407 if (n_basic_blocks_for_fn (fun) < 4000)
4409 compute_avail (fun);
4410 compute_antic ();
4411 insert ();
4414 /* Make sure to remove fake edges before committing our inserts.
4415 This makes sure we don't end up with extra critical edges that
4416 we would need to split. */
4417 remove_fake_exit_edges ();
4418 gsi_commit_edge_inserts ();
4420 /* Eliminate folds statements which might (should not...) end up
4421 not keeping virtual operands up-to-date. */
4422 gcc_assert (!need_ssa_update_p (fun));
4424 statistics_counter_event (fun, "Insertions", pre_stats.insertions);
4425 statistics_counter_event (fun, "PA inserted", pre_stats.pa_insert);
4426 statistics_counter_event (fun, "HOIST inserted", pre_stats.hoist_insert);
4427 statistics_counter_event (fun, "New PHIs", pre_stats.phis);
4429 todo |= eliminate_with_rpo_vn (inserted_exprs);
4431 vn_valueize = NULL;
4433 fini_pre ();
4435 scev_finalize ();
4436 loop_optimizer_finalize ();
4438 /* Perform a CFG cleanup before we run simple_dce_from_worklist since
4439 unreachable code regions will have not up-to-date SSA form which
4440 confuses it. */
4441 bool need_crit_edge_split = false;
4442 if (todo & TODO_cleanup_cfg)
4444 cleanup_tree_cfg ();
4445 need_crit_edge_split = true;
4448 /* Because we don't follow exactly the standard PRE algorithm, and decide not
4449 to insert PHI nodes sometimes, and because value numbering of casts isn't
4450 perfect, we sometimes end up inserting dead code. This simple DCE-like
4451 pass removes any insertions we made that weren't actually used. */
4452 simple_dce_from_worklist (inserted_exprs);
4453 BITMAP_FREE (inserted_exprs);
4455 /* TODO: tail_merge_optimize may merge all predecessors of a block, in which
4456 case we can merge the block with the remaining predecessor of the block.
4457 It should either:
4458 - call merge_blocks after each tail merge iteration
4459 - call merge_blocks after all tail merge iterations
4460 - mark TODO_cleanup_cfg when necessary. */
4461 todo |= tail_merge_optimize (need_crit_edge_split);
4463 free_rpo_vn ();
4465 /* Tail merging invalidates the virtual SSA web, together with
4466 cfg-cleanup opportunities exposed by PRE this will wreck the
4467 SSA updating machinery. So make sure to run update-ssa
4468 manually, before eventually scheduling cfg-cleanup as part of
4469 the todo. */
4470 update_ssa (TODO_update_ssa_only_virtuals);
4472 return todo;
4475 } // anon namespace
4477 gimple_opt_pass *
4478 make_pass_pre (gcc::context *ctxt)
4480 return new pass_pre (ctxt);