fix single argument static_assert
[official-gcc.git] / gcc / tree-ssa-pre.cc
blob75217f5cde1208f8390f5455d94ba9683f5e914f
1 /* Full and partial redundancy elimination and code hoisting on SSA GIMPLE.
2 Copyright (C) 2001-2024 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-iterator.h"
38 #include "gimple-fold.h"
39 #include "tree-eh.h"
40 #include "gimplify.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;
326 static obstack pre_expr_obstack;
328 /* Allocate an expression id for EXPR. */
330 static inline unsigned int
331 alloc_expression_id (pre_expr expr)
333 struct pre_expr_d **slot;
334 /* Make sure we won't overflow. */
335 gcc_assert (next_expression_id + 1 > next_expression_id);
336 expr->id = next_expression_id++;
337 expressions.safe_push (expr);
338 if (expr->kind == NAME)
340 unsigned version = SSA_NAME_VERSION (PRE_EXPR_NAME (expr));
341 /* vec::safe_grow_cleared allocates no headroom. Avoid frequent
342 re-allocations by using vec::reserve upfront. */
343 unsigned old_len = name_to_id.length ();
344 name_to_id.reserve (num_ssa_names - old_len);
345 name_to_id.quick_grow_cleared (num_ssa_names);
346 gcc_assert (name_to_id[version] == 0);
347 name_to_id[version] = expr->id;
349 else
351 slot = expression_to_id->find_slot (expr, INSERT);
352 gcc_assert (!*slot);
353 *slot = expr;
355 return next_expression_id - 1;
358 /* Return the expression id for tree EXPR. */
360 static inline unsigned int
361 get_expression_id (const pre_expr expr)
363 return expr->id;
366 static inline unsigned int
367 lookup_expression_id (const pre_expr expr)
369 struct pre_expr_d **slot;
371 if (expr->kind == NAME)
373 unsigned version = SSA_NAME_VERSION (PRE_EXPR_NAME (expr));
374 if (name_to_id.length () <= version)
375 return 0;
376 return name_to_id[version];
378 else
380 slot = expression_to_id->find_slot (expr, NO_INSERT);
381 if (!slot)
382 return 0;
383 return ((pre_expr)*slot)->id;
387 /* Return the expression that has expression id ID */
389 static inline pre_expr
390 expression_for_id (unsigned int id)
392 return expressions[id];
395 static object_allocator<pre_expr_d> pre_expr_pool ("pre_expr nodes");
397 /* Given an SSA_NAME NAME, get or create a pre_expr to represent it. */
399 static pre_expr
400 get_or_alloc_expr_for_name (tree name)
402 struct pre_expr_d expr;
403 pre_expr result;
404 unsigned int result_id;
406 expr.kind = NAME;
407 expr.id = 0;
408 PRE_EXPR_NAME (&expr) = name;
409 result_id = lookup_expression_id (&expr);
410 if (result_id != 0)
411 return expression_for_id (result_id);
413 result = pre_expr_pool.allocate ();
414 result->kind = NAME;
415 result->loc = UNKNOWN_LOCATION;
416 result->value_id = VN_INFO (name)->value_id;
417 PRE_EXPR_NAME (result) = name;
418 alloc_expression_id (result);
419 return result;
422 /* Given an NARY, get or create a pre_expr to represent it. Assign
423 VALUE_ID to it or allocate a new value-id if it is zero. Record
424 LOC as the original location of the expression. */
426 static pre_expr
427 get_or_alloc_expr_for_nary (vn_nary_op_t nary, unsigned value_id,
428 location_t loc = UNKNOWN_LOCATION)
430 struct pre_expr_d expr;
431 pre_expr result;
432 unsigned int result_id;
434 gcc_assert (value_id == 0 || !value_id_constant_p (value_id));
436 expr.kind = NARY;
437 expr.id = 0;
438 nary->hashcode = vn_nary_op_compute_hash (nary);
439 PRE_EXPR_NARY (&expr) = nary;
440 result_id = lookup_expression_id (&expr);
441 if (result_id != 0)
442 return expression_for_id (result_id);
444 result = pre_expr_pool.allocate ();
445 result->kind = NARY;
446 result->loc = loc;
447 result->value_id = value_id ? value_id : get_next_value_id ();
448 PRE_EXPR_NARY (result)
449 = alloc_vn_nary_op_noinit (nary->length, &pre_expr_obstack);
450 memcpy (PRE_EXPR_NARY (result), nary, sizeof_vn_nary_op (nary->length));
451 alloc_expression_id (result);
452 return result;
455 /* Given an REFERENCE, get or create a pre_expr to represent it. */
457 static pre_expr
458 get_or_alloc_expr_for_reference (vn_reference_t reference,
459 location_t loc = UNKNOWN_LOCATION)
461 struct pre_expr_d expr;
462 pre_expr result;
463 unsigned int result_id;
465 expr.kind = REFERENCE;
466 expr.id = 0;
467 PRE_EXPR_REFERENCE (&expr) = reference;
468 result_id = lookup_expression_id (&expr);
469 if (result_id != 0)
470 return expression_for_id (result_id);
472 result = pre_expr_pool.allocate ();
473 result->kind = REFERENCE;
474 result->loc = loc;
475 result->value_id = reference->value_id;
476 PRE_EXPR_REFERENCE (result) = reference;
477 alloc_expression_id (result);
478 return result;
482 /* An unordered bitmap set. One bitmap tracks values, the other,
483 expressions. */
484 typedef class bitmap_set
486 public:
487 bitmap_head expressions;
488 bitmap_head values;
489 } *bitmap_set_t;
491 #define FOR_EACH_EXPR_ID_IN_SET(set, id, bi) \
492 EXECUTE_IF_SET_IN_BITMAP (&(set)->expressions, 0, (id), (bi))
494 #define FOR_EACH_VALUE_ID_IN_SET(set, id, bi) \
495 EXECUTE_IF_SET_IN_BITMAP (&(set)->values, 0, (id), (bi))
497 /* Mapping from value id to expressions with that value_id. */
498 static vec<bitmap> value_expressions;
499 /* We just record a single expression for each constant value,
500 one of kind CONSTANT. */
501 static vec<pre_expr> constant_value_expressions;
504 /* This structure is used to keep track of statistics on what
505 optimization PRE was able to perform. */
506 static struct
508 /* The number of new expressions/temporaries generated by PRE. */
509 int insertions;
511 /* The number of inserts found due to partial anticipation */
512 int pa_insert;
514 /* The number of inserts made for code hoisting. */
515 int hoist_insert;
517 /* The number of new PHI nodes added by PRE. */
518 int phis;
519 } pre_stats;
521 static bool do_partial_partial;
522 static pre_expr bitmap_find_leader (bitmap_set_t, unsigned int);
523 static void bitmap_value_insert_into_set (bitmap_set_t, pre_expr);
524 static bool bitmap_value_replace_in_set (bitmap_set_t, pre_expr);
525 static void bitmap_set_copy (bitmap_set_t, bitmap_set_t);
526 static bool bitmap_set_contains_value (bitmap_set_t, unsigned int);
527 static void bitmap_insert_into_set (bitmap_set_t, pre_expr);
528 static bitmap_set_t bitmap_set_new (void);
529 static tree create_expression_by_pieces (basic_block, pre_expr, gimple_seq *,
530 tree);
531 static tree find_or_generate_expression (basic_block, tree, gimple_seq *);
532 static unsigned int get_expr_value_id (pre_expr);
534 /* We can add and remove elements and entries to and from sets
535 and hash tables, so we use alloc pools for them. */
537 static object_allocator<bitmap_set> bitmap_set_pool ("Bitmap sets");
538 static bitmap_obstack grand_bitmap_obstack;
540 /* A three tuple {e, pred, v} used to cache phi translations in the
541 phi_translate_table. */
543 typedef struct expr_pred_trans_d : public typed_noop_remove <expr_pred_trans_d>
545 typedef expr_pred_trans_d value_type;
546 typedef expr_pred_trans_d compare_type;
548 /* The expression ID. */
549 unsigned e;
551 /* The value expression ID that resulted from the translation. */
552 unsigned v;
554 /* hash_table support. */
555 static inline void mark_empty (expr_pred_trans_d &);
556 static inline bool is_empty (const expr_pred_trans_d &);
557 static inline void mark_deleted (expr_pred_trans_d &);
558 static inline bool is_deleted (const expr_pred_trans_d &);
559 static const bool empty_zero_p = true;
560 static inline hashval_t hash (const expr_pred_trans_d &);
561 static inline int equal (const expr_pred_trans_d &, const expr_pred_trans_d &);
562 } *expr_pred_trans_t;
563 typedef const struct expr_pred_trans_d *const_expr_pred_trans_t;
565 inline bool
566 expr_pred_trans_d::is_empty (const expr_pred_trans_d &e)
568 return e.e == 0;
571 inline bool
572 expr_pred_trans_d::is_deleted (const expr_pred_trans_d &e)
574 return e.e == -1u;
577 inline void
578 expr_pred_trans_d::mark_empty (expr_pred_trans_d &e)
580 e.e = 0;
583 inline void
584 expr_pred_trans_d::mark_deleted (expr_pred_trans_d &e)
586 e.e = -1u;
589 inline hashval_t
590 expr_pred_trans_d::hash (const expr_pred_trans_d &e)
592 return e.e;
595 inline int
596 expr_pred_trans_d::equal (const expr_pred_trans_d &ve1,
597 const expr_pred_trans_d &ve2)
599 return ve1.e == ve2.e;
602 /* Sets that we need to keep track of. */
603 typedef struct bb_bitmap_sets
605 /* The EXP_GEN set, which represents expressions/values generated in
606 a basic block. */
607 bitmap_set_t exp_gen;
609 /* The PHI_GEN set, which represents PHI results generated in a
610 basic block. */
611 bitmap_set_t phi_gen;
613 /* The TMP_GEN set, which represents results/temporaries generated
614 in a basic block. IE the LHS of an expression. */
615 bitmap_set_t tmp_gen;
617 /* The AVAIL_OUT set, which represents which values are available in
618 a given basic block. */
619 bitmap_set_t avail_out;
621 /* The ANTIC_IN set, which represents which values are anticipatable
622 in a given basic block. */
623 bitmap_set_t antic_in;
625 /* The PA_IN set, which represents which values are
626 partially anticipatable in a given basic block. */
627 bitmap_set_t pa_in;
629 /* The NEW_SETS set, which is used during insertion to augment the
630 AVAIL_OUT set of blocks with the new insertions performed during
631 the current iteration. */
632 bitmap_set_t new_sets;
634 /* A cache for value_dies_in_block_x. */
635 bitmap expr_dies;
637 /* The live virtual operand on successor edges. */
638 tree vop_on_exit;
640 /* PHI translate cache for the single successor edge. */
641 hash_table<expr_pred_trans_d> *phi_translate_table;
643 /* True if we have visited this block during ANTIC calculation. */
644 unsigned int visited : 1;
646 /* True when the block contains a call that might not return. */
647 unsigned int contains_may_not_return_call : 1;
648 } *bb_value_sets_t;
650 #define EXP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->exp_gen
651 #define PHI_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->phi_gen
652 #define TMP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->tmp_gen
653 #define AVAIL_OUT(BB) ((bb_value_sets_t) ((BB)->aux))->avail_out
654 #define ANTIC_IN(BB) ((bb_value_sets_t) ((BB)->aux))->antic_in
655 #define PA_IN(BB) ((bb_value_sets_t) ((BB)->aux))->pa_in
656 #define NEW_SETS(BB) ((bb_value_sets_t) ((BB)->aux))->new_sets
657 #define EXPR_DIES(BB) ((bb_value_sets_t) ((BB)->aux))->expr_dies
658 #define PHI_TRANS_TABLE(BB) ((bb_value_sets_t) ((BB)->aux))->phi_translate_table
659 #define BB_VISITED(BB) ((bb_value_sets_t) ((BB)->aux))->visited
660 #define BB_MAY_NOTRETURN(BB) ((bb_value_sets_t) ((BB)->aux))->contains_may_not_return_call
661 #define BB_LIVE_VOP_ON_EXIT(BB) ((bb_value_sets_t) ((BB)->aux))->vop_on_exit
664 /* Add the tuple mapping from {expression E, basic block PRED} to
665 the phi translation table and return whether it pre-existed. */
667 static inline bool
668 phi_trans_add (expr_pred_trans_t *entry, pre_expr e, basic_block pred)
670 if (!PHI_TRANS_TABLE (pred))
671 PHI_TRANS_TABLE (pred) = new hash_table<expr_pred_trans_d> (11);
673 expr_pred_trans_t slot;
674 expr_pred_trans_d tem;
675 unsigned id = get_expression_id (e);
676 tem.e = id;
677 slot = PHI_TRANS_TABLE (pred)->find_slot_with_hash (tem, id, INSERT);
678 if (slot->e)
680 *entry = slot;
681 return true;
684 *entry = slot;
685 slot->e = id;
686 return false;
690 /* Add expression E to the expression set of value id V. */
692 static void
693 add_to_value (unsigned int v, pre_expr e)
695 gcc_checking_assert (get_expr_value_id (e) == v);
697 if (value_id_constant_p (v))
699 if (e->kind != CONSTANT)
700 return;
702 if (-v >= constant_value_expressions.length ())
703 constant_value_expressions.safe_grow_cleared (-v + 1);
705 pre_expr leader = constant_value_expressions[-v];
706 if (!leader)
707 constant_value_expressions[-v] = e;
709 else
711 if (v >= value_expressions.length ())
712 value_expressions.safe_grow_cleared (v + 1);
714 bitmap set = value_expressions[v];
715 if (!set)
717 set = BITMAP_ALLOC (&grand_bitmap_obstack);
718 value_expressions[v] = set;
720 bitmap_set_bit (set, get_expression_id (e));
724 /* Create a new bitmap set and return it. */
726 static bitmap_set_t
727 bitmap_set_new (void)
729 bitmap_set_t ret = bitmap_set_pool.allocate ();
730 bitmap_initialize (&ret->expressions, &grand_bitmap_obstack);
731 bitmap_initialize (&ret->values, &grand_bitmap_obstack);
732 return ret;
735 /* Return the value id for a PRE expression EXPR. */
737 static unsigned int
738 get_expr_value_id (pre_expr expr)
740 /* ??? We cannot assert that expr has a value-id (it can be 0), because
741 we assign value-ids only to expressions that have a result
742 in set_hashtable_value_ids. */
743 return expr->value_id;
746 /* Return a VN valnum (SSA name or constant) for the PRE value-id VAL. */
748 static tree
749 vn_valnum_from_value_id (unsigned int val)
751 if (value_id_constant_p (val))
753 pre_expr vexpr = constant_value_expressions[-val];
754 if (vexpr)
755 return PRE_EXPR_CONSTANT (vexpr);
756 return NULL_TREE;
759 bitmap exprset = value_expressions[val];
760 bitmap_iterator bi;
761 unsigned int i;
762 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
764 pre_expr vexpr = expression_for_id (i);
765 if (vexpr->kind == NAME)
766 return VN_INFO (PRE_EXPR_NAME (vexpr))->valnum;
768 return NULL_TREE;
771 /* Insert an expression EXPR into a bitmapped set. */
773 static void
774 bitmap_insert_into_set (bitmap_set_t set, pre_expr expr)
776 unsigned int val = get_expr_value_id (expr);
777 if (! value_id_constant_p (val))
779 /* Note this is the only function causing multiple expressions
780 for the same value to appear in a set. This is needed for
781 TMP_GEN, PHI_GEN and NEW_SETs. */
782 bitmap_set_bit (&set->values, val);
783 bitmap_set_bit (&set->expressions, get_expression_id (expr));
787 /* Copy a bitmapped set ORIG, into bitmapped set DEST. */
789 static void
790 bitmap_set_copy (bitmap_set_t dest, bitmap_set_t orig)
792 bitmap_copy (&dest->expressions, &orig->expressions);
793 bitmap_copy (&dest->values, &orig->values);
797 /* Free memory used up by SET. */
798 static void
799 bitmap_set_free (bitmap_set_t set)
801 bitmap_clear (&set->expressions);
802 bitmap_clear (&set->values);
805 static void
806 pre_expr_DFS (pre_expr expr, bitmap_set_t set, bitmap val_visited,
807 vec<pre_expr> &post);
809 /* DFS walk leaders of VAL to their operands with leaders in SET, collecting
810 expressions in SET in postorder into POST. */
812 static void
813 pre_expr_DFS (unsigned val, bitmap_set_t set, bitmap val_visited,
814 vec<pre_expr> &post)
816 unsigned int i;
817 bitmap_iterator bi;
819 /* Iterate over all leaders and DFS recurse. Borrowed from
820 bitmap_find_leader. */
821 bitmap exprset = value_expressions[val];
822 if (!exprset->first->next)
824 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
825 if (bitmap_bit_p (&set->expressions, i))
826 pre_expr_DFS (expression_for_id (i), set, val_visited, post);
827 return;
830 EXECUTE_IF_AND_IN_BITMAP (exprset, &set->expressions, 0, i, bi)
831 pre_expr_DFS (expression_for_id (i), set, val_visited, post);
834 /* DFS walk EXPR to its operands with leaders in SET, collecting
835 expressions in SET in postorder into POST. */
837 static void
838 pre_expr_DFS (pre_expr expr, bitmap_set_t set, bitmap val_visited,
839 vec<pre_expr> &post)
841 switch (expr->kind)
843 case NARY:
845 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
846 for (unsigned i = 0; i < nary->length; i++)
848 if (TREE_CODE (nary->op[i]) != SSA_NAME)
849 continue;
850 unsigned int op_val_id = VN_INFO (nary->op[i])->value_id;
851 /* If we already found a leader for the value we've
852 recursed already. Avoid the costly bitmap_find_leader. */
853 if (bitmap_bit_p (&set->values, op_val_id)
854 && bitmap_set_bit (val_visited, op_val_id))
855 pre_expr_DFS (op_val_id, set, val_visited, post);
857 break;
859 case REFERENCE:
861 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
862 vec<vn_reference_op_s> operands = ref->operands;
863 vn_reference_op_t operand;
864 for (unsigned i = 0; operands.iterate (i, &operand); i++)
866 tree op[3];
867 op[0] = operand->op0;
868 op[1] = operand->op1;
869 op[2] = operand->op2;
870 for (unsigned n = 0; n < 3; ++n)
872 if (!op[n] || TREE_CODE (op[n]) != SSA_NAME)
873 continue;
874 unsigned op_val_id = VN_INFO (op[n])->value_id;
875 if (bitmap_bit_p (&set->values, op_val_id)
876 && bitmap_set_bit (val_visited, op_val_id))
877 pre_expr_DFS (op_val_id, set, val_visited, post);
880 break;
882 default:;
884 post.quick_push (expr);
887 /* Generate an topological-ordered array of bitmap set SET. */
889 static vec<pre_expr>
890 sorted_array_from_bitmap_set (bitmap_set_t set)
892 unsigned int i;
893 bitmap_iterator bi;
894 vec<pre_expr> result;
896 /* Pre-allocate enough space for the array. */
897 result.create (bitmap_count_bits (&set->expressions));
899 auto_bitmap val_visited (&grand_bitmap_obstack);
900 bitmap_tree_view (val_visited);
901 FOR_EACH_VALUE_ID_IN_SET (set, i, bi)
902 if (bitmap_set_bit (val_visited, i))
903 pre_expr_DFS (i, set, val_visited, result);
905 return result;
908 /* Subtract all expressions contained in ORIG from DEST. */
910 static bitmap_set_t
911 bitmap_set_subtract_expressions (bitmap_set_t dest, bitmap_set_t orig)
913 bitmap_set_t result = bitmap_set_new ();
914 bitmap_iterator bi;
915 unsigned int i;
917 bitmap_and_compl (&result->expressions, &dest->expressions,
918 &orig->expressions);
920 FOR_EACH_EXPR_ID_IN_SET (result, i, bi)
922 pre_expr expr = expression_for_id (i);
923 unsigned int value_id = get_expr_value_id (expr);
924 bitmap_set_bit (&result->values, value_id);
927 return result;
930 /* Subtract all values in bitmap set B from bitmap set A. */
932 static void
933 bitmap_set_subtract_values (bitmap_set_t a, bitmap_set_t b)
935 unsigned int i;
936 bitmap_iterator bi;
937 unsigned to_remove = -1U;
938 bitmap_and_compl_into (&a->values, &b->values);
939 FOR_EACH_EXPR_ID_IN_SET (a, i, bi)
941 if (to_remove != -1U)
943 bitmap_clear_bit (&a->expressions, to_remove);
944 to_remove = -1U;
946 pre_expr expr = expression_for_id (i);
947 if (! bitmap_bit_p (&a->values, get_expr_value_id (expr)))
948 to_remove = i;
950 if (to_remove != -1U)
951 bitmap_clear_bit (&a->expressions, to_remove);
955 /* Return true if bitmapped set SET contains the value VALUE_ID. */
957 static bool
958 bitmap_set_contains_value (bitmap_set_t set, unsigned int value_id)
960 if (value_id_constant_p (value_id))
961 return true;
963 return bitmap_bit_p (&set->values, value_id);
966 /* Return true if two bitmap sets are equal. */
968 static bool
969 bitmap_set_equal (bitmap_set_t a, bitmap_set_t b)
971 return bitmap_equal_p (&a->values, &b->values);
974 /* Replace an instance of EXPR's VALUE with EXPR in SET if it exists,
975 and add it otherwise. Return true if any changes were made. */
977 static bool
978 bitmap_value_replace_in_set (bitmap_set_t set, pre_expr expr)
980 unsigned int val = get_expr_value_id (expr);
981 if (value_id_constant_p (val))
982 return false;
984 if (bitmap_set_contains_value (set, val))
986 /* The number of expressions having a given value is usually
987 significantly less than the total number of expressions in SET.
988 Thus, rather than check, for each expression in SET, whether it
989 has the value LOOKFOR, we walk the reverse mapping that tells us
990 what expressions have a given value, and see if any of those
991 expressions are in our set. For large testcases, this is about
992 5-10x faster than walking the bitmap. If this is somehow a
993 significant lose for some cases, we can choose which set to walk
994 based on the set size. */
995 unsigned int i;
996 bitmap_iterator bi;
997 bitmap exprset = value_expressions[val];
998 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
1000 if (bitmap_clear_bit (&set->expressions, i))
1002 bitmap_set_bit (&set->expressions, get_expression_id (expr));
1003 return i != get_expression_id (expr);
1006 gcc_unreachable ();
1009 bitmap_insert_into_set (set, expr);
1010 return true;
1013 /* Insert EXPR into SET if EXPR's value is not already present in
1014 SET. */
1016 static void
1017 bitmap_value_insert_into_set (bitmap_set_t set, pre_expr expr)
1019 unsigned int val = get_expr_value_id (expr);
1021 gcc_checking_assert (expr->id == get_expression_id (expr));
1023 /* Constant values are always considered to be part of the set. */
1024 if (value_id_constant_p (val))
1025 return;
1027 /* If the value membership changed, add the expression. */
1028 if (bitmap_set_bit (&set->values, val))
1029 bitmap_set_bit (&set->expressions, expr->id);
1032 /* Print out EXPR to outfile. */
1034 static void
1035 print_pre_expr (FILE *outfile, const pre_expr expr)
1037 if (! expr)
1039 fprintf (outfile, "NULL");
1040 return;
1042 switch (expr->kind)
1044 case CONSTANT:
1045 print_generic_expr (outfile, PRE_EXPR_CONSTANT (expr));
1046 break;
1047 case NAME:
1048 print_generic_expr (outfile, PRE_EXPR_NAME (expr));
1049 break;
1050 case NARY:
1052 unsigned int i;
1053 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1054 fprintf (outfile, "{%s,", get_tree_code_name (nary->opcode));
1055 for (i = 0; i < nary->length; i++)
1057 print_generic_expr (outfile, nary->op[i]);
1058 if (i != (unsigned) nary->length - 1)
1059 fprintf (outfile, ",");
1061 fprintf (outfile, "}");
1063 break;
1065 case REFERENCE:
1067 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1068 print_vn_reference_ops (outfile, ref->operands);
1069 if (ref->vuse)
1071 fprintf (outfile, "@");
1072 print_generic_expr (outfile, ref->vuse);
1075 break;
1078 void debug_pre_expr (pre_expr);
1080 /* Like print_pre_expr but always prints to stderr. */
1081 DEBUG_FUNCTION void
1082 debug_pre_expr (pre_expr e)
1084 print_pre_expr (stderr, e);
1085 fprintf (stderr, "\n");
1088 /* Print out SET to OUTFILE. */
1090 static void
1091 print_bitmap_set (FILE *outfile, bitmap_set_t set,
1092 const char *setname, int blockindex)
1094 fprintf (outfile, "%s[%d] := { ", setname, blockindex);
1095 if (set)
1097 bool first = true;
1098 unsigned i;
1099 bitmap_iterator bi;
1101 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
1103 const pre_expr expr = expression_for_id (i);
1105 if (!first)
1106 fprintf (outfile, ", ");
1107 first = false;
1108 print_pre_expr (outfile, expr);
1110 fprintf (outfile, " (%04d)", get_expr_value_id (expr));
1113 fprintf (outfile, " }\n");
1116 void debug_bitmap_set (bitmap_set_t);
1118 DEBUG_FUNCTION void
1119 debug_bitmap_set (bitmap_set_t set)
1121 print_bitmap_set (stderr, set, "debug", 0);
1124 void debug_bitmap_sets_for (basic_block);
1126 DEBUG_FUNCTION void
1127 debug_bitmap_sets_for (basic_block bb)
1129 print_bitmap_set (stderr, AVAIL_OUT (bb), "avail_out", bb->index);
1130 print_bitmap_set (stderr, EXP_GEN (bb), "exp_gen", bb->index);
1131 print_bitmap_set (stderr, PHI_GEN (bb), "phi_gen", bb->index);
1132 print_bitmap_set (stderr, TMP_GEN (bb), "tmp_gen", bb->index);
1133 print_bitmap_set (stderr, ANTIC_IN (bb), "antic_in", bb->index);
1134 if (do_partial_partial)
1135 print_bitmap_set (stderr, PA_IN (bb), "pa_in", bb->index);
1136 print_bitmap_set (stderr, NEW_SETS (bb), "new_sets", bb->index);
1139 /* Print out the expressions that have VAL to OUTFILE. */
1141 static void
1142 print_value_expressions (FILE *outfile, unsigned int val)
1144 bitmap set = value_expressions[val];
1145 if (set)
1147 bitmap_set x;
1148 char s[10];
1149 sprintf (s, "%04d", val);
1150 x.expressions = *set;
1151 print_bitmap_set (outfile, &x, s, 0);
1156 DEBUG_FUNCTION void
1157 debug_value_expressions (unsigned int val)
1159 print_value_expressions (stderr, val);
1162 /* Given a CONSTANT, allocate a new CONSTANT type PRE_EXPR to
1163 represent it. */
1165 static pre_expr
1166 get_or_alloc_expr_for_constant (tree constant)
1168 unsigned int result_id;
1169 struct pre_expr_d expr;
1170 pre_expr newexpr;
1172 expr.kind = CONSTANT;
1173 PRE_EXPR_CONSTANT (&expr) = constant;
1174 result_id = lookup_expression_id (&expr);
1175 if (result_id != 0)
1176 return expression_for_id (result_id);
1178 newexpr = pre_expr_pool.allocate ();
1179 newexpr->kind = CONSTANT;
1180 newexpr->loc = UNKNOWN_LOCATION;
1181 PRE_EXPR_CONSTANT (newexpr) = constant;
1182 alloc_expression_id (newexpr);
1183 newexpr->value_id = get_or_alloc_constant_value_id (constant);
1184 add_to_value (newexpr->value_id, newexpr);
1185 return newexpr;
1188 /* Return the folded version of T if T, when folded, is a gimple
1189 min_invariant or an SSA name. Otherwise, return T. */
1191 static pre_expr
1192 fully_constant_expression (pre_expr e)
1194 switch (e->kind)
1196 case CONSTANT:
1197 return e;
1198 case NARY:
1200 vn_nary_op_t nary = PRE_EXPR_NARY (e);
1201 tree res = vn_nary_simplify (nary);
1202 if (!res)
1203 return e;
1204 if (is_gimple_min_invariant (res))
1205 return get_or_alloc_expr_for_constant (res);
1206 if (TREE_CODE (res) == SSA_NAME)
1207 return get_or_alloc_expr_for_name (res);
1208 return e;
1210 case REFERENCE:
1212 vn_reference_t ref = PRE_EXPR_REFERENCE (e);
1213 tree folded;
1214 if ((folded = fully_constant_vn_reference_p (ref)))
1215 return get_or_alloc_expr_for_constant (folded);
1216 return e;
1218 default:
1219 return e;
1223 /* Translate the VUSE backwards through phi nodes in E->dest, so that
1224 it has the value it would have in E->src. Set *SAME_VALID to true
1225 in case the new vuse doesn't change the value id of the OPERANDS. */
1227 static tree
1228 translate_vuse_through_block (vec<vn_reference_op_s> operands,
1229 alias_set_type set, alias_set_type base_set,
1230 tree type, tree vuse, edge e, bool *same_valid)
1232 basic_block phiblock = e->dest;
1233 gimple *phi = SSA_NAME_DEF_STMT (vuse);
1234 ao_ref ref;
1236 if (same_valid)
1237 *same_valid = true;
1239 /* If value-numbering provided a memory state for this
1240 that dominates PHIBLOCK we can just use that. */
1241 if (gimple_nop_p (phi)
1242 || (gimple_bb (phi) != phiblock
1243 && dominated_by_p (CDI_DOMINATORS, phiblock, gimple_bb (phi))))
1244 return vuse;
1246 /* We have pruned expressions that are killed in PHIBLOCK via
1247 prune_clobbered_mems but we have not rewritten the VUSE to the one
1248 live at the start of the block. If there is no virtual PHI to translate
1249 through return the VUSE live at entry. Otherwise the VUSE to translate
1250 is the def of the virtual PHI node. */
1251 phi = get_virtual_phi (phiblock);
1252 if (!phi)
1253 return BB_LIVE_VOP_ON_EXIT
1254 (get_immediate_dominator (CDI_DOMINATORS, phiblock));
1256 if (same_valid
1257 && ao_ref_init_from_vn_reference (&ref, set, base_set, type, operands))
1259 bitmap visited = NULL;
1260 /* Try to find a vuse that dominates this phi node by skipping
1261 non-clobbering statements. */
1262 unsigned int cnt = param_sccvn_max_alias_queries_per_access;
1263 vuse = get_continuation_for_phi (phi, &ref, true,
1264 cnt, &visited, false, NULL, NULL);
1265 if (visited)
1266 BITMAP_FREE (visited);
1268 else
1269 vuse = NULL_TREE;
1270 /* If we didn't find any, the value ID can't stay the same. */
1271 if (!vuse && same_valid)
1272 *same_valid = false;
1274 /* ??? We would like to return vuse here as this is the canonical
1275 upmost vdef that this reference is associated with. But during
1276 insertion of the references into the hash tables we only ever
1277 directly insert with their direct gimple_vuse, hence returning
1278 something else would make us not find the other expression. */
1279 return PHI_ARG_DEF (phi, e->dest_idx);
1282 /* Like bitmap_find_leader, but checks for the value existing in SET1 *or*
1283 SET2 *or* SET3. This is used to avoid making a set consisting of the union
1284 of PA_IN and ANTIC_IN during insert and phi-translation. */
1286 static inline pre_expr
1287 find_leader_in_sets (unsigned int val, bitmap_set_t set1, bitmap_set_t set2,
1288 bitmap_set_t set3 = NULL)
1290 pre_expr result = NULL;
1292 if (set1)
1293 result = bitmap_find_leader (set1, val);
1294 if (!result && set2)
1295 result = bitmap_find_leader (set2, val);
1296 if (!result && set3)
1297 result = bitmap_find_leader (set3, val);
1298 return result;
1301 /* Get the tree type for our PRE expression e. */
1303 static tree
1304 get_expr_type (const pre_expr e)
1306 switch (e->kind)
1308 case NAME:
1309 return TREE_TYPE (PRE_EXPR_NAME (e));
1310 case CONSTANT:
1311 return TREE_TYPE (PRE_EXPR_CONSTANT (e));
1312 case REFERENCE:
1313 return PRE_EXPR_REFERENCE (e)->type;
1314 case NARY:
1315 return PRE_EXPR_NARY (e)->type;
1317 gcc_unreachable ();
1320 /* Get a representative SSA_NAME for a given expression that is available in B.
1321 Since all of our sub-expressions are treated as values, we require
1322 them to be SSA_NAME's for simplicity.
1323 Prior versions of GVNPRE used to use "value handles" here, so that
1324 an expression would be VH.11 + VH.10 instead of d_3 + e_6. In
1325 either case, the operands are really values (IE we do not expect
1326 them to be usable without finding leaders). */
1328 static tree
1329 get_representative_for (const pre_expr e, basic_block b = NULL)
1331 tree name, valnum = NULL_TREE;
1332 unsigned int value_id = get_expr_value_id (e);
1334 switch (e->kind)
1336 case NAME:
1337 return PRE_EXPR_NAME (e);
1338 case CONSTANT:
1339 return PRE_EXPR_CONSTANT (e);
1340 case NARY:
1341 case REFERENCE:
1343 /* Go through all of the expressions representing this value
1344 and pick out an SSA_NAME. */
1345 unsigned int i;
1346 bitmap_iterator bi;
1347 bitmap exprs = value_expressions[value_id];
1348 EXECUTE_IF_SET_IN_BITMAP (exprs, 0, i, bi)
1350 pre_expr rep = expression_for_id (i);
1351 if (rep->kind == NAME)
1353 tree name = PRE_EXPR_NAME (rep);
1354 valnum = VN_INFO (name)->valnum;
1355 gimple *def = SSA_NAME_DEF_STMT (name);
1356 /* We have to return either a new representative or one
1357 that can be used for expression simplification and thus
1358 is available in B. */
1359 if (! b
1360 || gimple_nop_p (def)
1361 || dominated_by_p (CDI_DOMINATORS, b, gimple_bb (def)))
1362 return name;
1364 else if (rep->kind == CONSTANT)
1365 return PRE_EXPR_CONSTANT (rep);
1368 break;
1371 /* If we reached here we couldn't find an SSA_NAME. This can
1372 happen when we've discovered a value that has never appeared in
1373 the program as set to an SSA_NAME, as the result of phi translation.
1374 Create one here.
1375 ??? We should be able to re-use this when we insert the statement
1376 to compute it. */
1377 name = make_temp_ssa_name (get_expr_type (e), gimple_build_nop (), "pretmp");
1378 vn_ssa_aux_t vn_info = VN_INFO (name);
1379 vn_info->value_id = value_id;
1380 vn_info->valnum = valnum ? valnum : name;
1381 vn_info->visited = true;
1382 /* ??? For now mark this SSA name for release by VN. */
1383 vn_info->needs_insertion = true;
1384 add_to_value (value_id, get_or_alloc_expr_for_name (name));
1385 if (dump_file && (dump_flags & TDF_DETAILS))
1387 fprintf (dump_file, "Created SSA_NAME representative ");
1388 print_generic_expr (dump_file, name);
1389 fprintf (dump_file, " for expression:");
1390 print_pre_expr (dump_file, e);
1391 fprintf (dump_file, " (%04d)\n", value_id);
1394 return name;
1398 static pre_expr
1399 phi_translate (bitmap_set_t, pre_expr, bitmap_set_t, bitmap_set_t, edge);
1401 /* Translate EXPR using phis in PHIBLOCK, so that it has the values of
1402 the phis in PRED. Return NULL if we can't find a leader for each part
1403 of the translated expression. */
1405 static pre_expr
1406 phi_translate_1 (bitmap_set_t dest,
1407 pre_expr expr, bitmap_set_t set1, bitmap_set_t set2, edge e)
1409 basic_block pred = e->src;
1410 basic_block phiblock = e->dest;
1411 location_t expr_loc = expr->loc;
1412 switch (expr->kind)
1414 case NARY:
1416 unsigned int i;
1417 bool changed = false;
1418 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1419 vn_nary_op_t newnary = XALLOCAVAR (struct vn_nary_op_s,
1420 sizeof_vn_nary_op (nary->length));
1421 memcpy (newnary, nary, sizeof_vn_nary_op (nary->length));
1423 for (i = 0; i < newnary->length; i++)
1425 if (TREE_CODE (newnary->op[i]) != SSA_NAME)
1426 continue;
1427 else
1429 pre_expr leader, result;
1430 unsigned int op_val_id = VN_INFO (newnary->op[i])->value_id;
1431 leader = find_leader_in_sets (op_val_id, set1, set2);
1432 result = phi_translate (dest, leader, set1, set2, e);
1433 if (result && result != leader)
1434 /* If op has a leader in the sets we translate make
1435 sure to use the value of the translated expression.
1436 We might need a new representative for that. */
1437 newnary->op[i] = get_representative_for (result, pred);
1438 else if (!result)
1439 return NULL;
1441 changed |= newnary->op[i] != nary->op[i];
1444 if (changed)
1446 pre_expr constant;
1447 unsigned int new_val_id;
1449 PRE_EXPR_NARY (expr) = newnary;
1450 constant = fully_constant_expression (expr);
1451 PRE_EXPR_NARY (expr) = nary;
1452 if (constant != expr)
1454 /* For non-CONSTANTs we have to make sure we can eventually
1455 insert the expression. Which means we need to have a
1456 leader for it. */
1457 if (constant->kind != CONSTANT)
1459 /* Do not allow simplifications to non-constants over
1460 backedges as this will likely result in a loop PHI node
1461 to be inserted and increased register pressure.
1462 See PR77498 - this avoids doing predcoms work in
1463 a less efficient way. */
1464 if (e->flags & EDGE_DFS_BACK)
1466 else
1468 unsigned value_id = get_expr_value_id (constant);
1469 /* We want a leader in ANTIC_OUT or AVAIL_OUT here.
1470 dest has what we computed into ANTIC_OUT sofar
1471 so pick from that - since topological sorting
1472 by sorted_array_from_bitmap_set isn't perfect
1473 we may lose some cases here. */
1474 constant = find_leader_in_sets (value_id, dest,
1475 AVAIL_OUT (pred));
1476 if (constant)
1478 if (dump_file && (dump_flags & TDF_DETAILS))
1480 fprintf (dump_file, "simplifying ");
1481 print_pre_expr (dump_file, expr);
1482 fprintf (dump_file, " translated %d -> %d to ",
1483 phiblock->index, pred->index);
1484 PRE_EXPR_NARY (expr) = newnary;
1485 print_pre_expr (dump_file, expr);
1486 PRE_EXPR_NARY (expr) = nary;
1487 fprintf (dump_file, " to ");
1488 print_pre_expr (dump_file, constant);
1489 fprintf (dump_file, "\n");
1491 return constant;
1495 else
1496 return constant;
1499 tree result = vn_nary_op_lookup_pieces (newnary->length,
1500 newnary->opcode,
1501 newnary->type,
1502 &newnary->op[0],
1503 &nary);
1504 if (result && is_gimple_min_invariant (result))
1505 return get_or_alloc_expr_for_constant (result);
1507 if (!nary || nary->predicated_values)
1508 new_val_id = 0;
1509 else
1510 new_val_id = nary->value_id;
1511 expr = get_or_alloc_expr_for_nary (newnary, new_val_id, expr_loc);
1512 add_to_value (get_expr_value_id (expr), expr);
1514 return expr;
1516 break;
1518 case REFERENCE:
1520 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1521 vec<vn_reference_op_s> operands = ref->operands;
1522 tree vuse = ref->vuse;
1523 tree newvuse = vuse;
1524 vec<vn_reference_op_s> newoperands = vNULL;
1525 bool changed = false, same_valid = true;
1526 unsigned int i, n;
1527 vn_reference_op_t operand;
1528 vn_reference_t newref;
1530 for (i = 0; operands.iterate (i, &operand); i++)
1532 pre_expr opresult;
1533 pre_expr leader;
1534 tree op[3];
1535 tree type = operand->type;
1536 vn_reference_op_s newop = *operand;
1537 op[0] = operand->op0;
1538 op[1] = operand->op1;
1539 op[2] = operand->op2;
1540 for (n = 0; n < 3; ++n)
1542 unsigned int op_val_id;
1543 if (!op[n])
1544 continue;
1545 if (TREE_CODE (op[n]) != SSA_NAME)
1547 /* We can't possibly insert these. */
1548 if (n != 0
1549 && !is_gimple_min_invariant (op[n]))
1550 break;
1551 continue;
1553 op_val_id = VN_INFO (op[n])->value_id;
1554 leader = find_leader_in_sets (op_val_id, set1, set2);
1555 opresult = phi_translate (dest, leader, set1, set2, e);
1556 if (opresult && opresult != leader)
1558 tree name = get_representative_for (opresult);
1559 changed |= name != op[n];
1560 op[n] = name;
1562 else if (!opresult)
1563 break;
1565 if (n != 3)
1567 newoperands.release ();
1568 return NULL;
1570 /* When we translate a MEM_REF across a backedge and we have
1571 restrict info that's not from our functions parameters
1572 we have to remap it since we now may deal with a different
1573 instance where the dependence info is no longer valid.
1574 See PR102970. Note instead of keeping a remapping table
1575 per backedge we simply throw away restrict info. */
1576 if ((newop.opcode == MEM_REF
1577 || newop.opcode == TARGET_MEM_REF)
1578 && newop.clique > 1
1579 && (e->flags & EDGE_DFS_BACK))
1581 newop.clique = 0;
1582 newop.base = 0;
1583 changed = true;
1585 if (!changed)
1586 continue;
1587 if (!newoperands.exists ())
1588 newoperands = operands.copy ();
1589 /* We may have changed from an SSA_NAME to a constant */
1590 if (newop.opcode == SSA_NAME && TREE_CODE (op[0]) != SSA_NAME)
1591 newop.opcode = TREE_CODE (op[0]);
1592 newop.type = type;
1593 newop.op0 = op[0];
1594 newop.op1 = op[1];
1595 newop.op2 = op[2];
1596 newoperands[i] = newop;
1598 gcc_checking_assert (i == operands.length ());
1600 if (vuse)
1602 newvuse = translate_vuse_through_block (newoperands.exists ()
1603 ? newoperands : operands,
1604 ref->set, ref->base_set,
1605 ref->type, vuse, e,
1606 changed
1607 ? NULL : &same_valid);
1608 if (newvuse == NULL_TREE)
1610 newoperands.release ();
1611 return NULL;
1615 if (changed || newvuse != vuse)
1617 unsigned int new_val_id;
1619 tree result = vn_reference_lookup_pieces (newvuse, ref->set,
1620 ref->base_set,
1621 ref->type,
1622 newoperands.exists ()
1623 ? newoperands : operands,
1624 &newref, VN_WALK);
1625 if (result)
1626 newoperands.release ();
1628 /* We can always insert constants, so if we have a partial
1629 redundant constant load of another type try to translate it
1630 to a constant of appropriate type. */
1631 if (result && is_gimple_min_invariant (result))
1633 tree tem = result;
1634 if (!useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1636 tem = fold_unary (VIEW_CONVERT_EXPR, ref->type, result);
1637 if (tem && !is_gimple_min_invariant (tem))
1638 tem = NULL_TREE;
1640 if (tem)
1641 return get_or_alloc_expr_for_constant (tem);
1644 /* If we'd have to convert things we would need to validate
1645 if we can insert the translated expression. So fail
1646 here for now - we cannot insert an alias with a different
1647 type in the VN tables either, as that would assert. */
1648 if (result
1649 && !useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1650 return NULL;
1651 else if (!result && newref
1652 && !useless_type_conversion_p (ref->type, newref->type))
1654 newoperands.release ();
1655 return NULL;
1658 if (newref)
1659 new_val_id = newref->value_id;
1660 else
1662 if (changed || !same_valid)
1663 new_val_id = get_next_value_id ();
1664 else
1665 new_val_id = ref->value_id;
1666 if (!newoperands.exists ())
1667 newoperands = operands.copy ();
1668 newref = vn_reference_insert_pieces (newvuse, ref->set,
1669 ref->base_set,
1670 ref->offset, ref->max_size,
1671 ref->type, newoperands,
1672 result, new_val_id);
1673 newoperands = vNULL;
1675 expr = get_or_alloc_expr_for_reference (newref, expr_loc);
1676 add_to_value (new_val_id, expr);
1678 newoperands.release ();
1679 return expr;
1681 break;
1683 case NAME:
1685 tree name = PRE_EXPR_NAME (expr);
1686 gimple *def_stmt = SSA_NAME_DEF_STMT (name);
1687 /* If the SSA name is defined by a PHI node in this block,
1688 translate it. */
1689 if (gimple_code (def_stmt) == GIMPLE_PHI
1690 && gimple_bb (def_stmt) == phiblock)
1692 tree def = PHI_ARG_DEF (def_stmt, e->dest_idx);
1694 /* Handle constant. */
1695 if (is_gimple_min_invariant (def))
1696 return get_or_alloc_expr_for_constant (def);
1698 return get_or_alloc_expr_for_name (def);
1700 /* Otherwise return it unchanged - it will get removed if its
1701 value is not available in PREDs AVAIL_OUT set of expressions
1702 by the subtraction of TMP_GEN. */
1703 return expr;
1706 default:
1707 gcc_unreachable ();
1711 /* Wrapper around phi_translate_1 providing caching functionality. */
1713 static pre_expr
1714 phi_translate (bitmap_set_t dest, pre_expr expr,
1715 bitmap_set_t set1, bitmap_set_t set2, edge e)
1717 expr_pred_trans_t slot = NULL;
1718 pre_expr phitrans;
1720 if (!expr)
1721 return NULL;
1723 /* Constants contain no values that need translation. */
1724 if (expr->kind == CONSTANT)
1725 return expr;
1727 if (value_id_constant_p (get_expr_value_id (expr)))
1728 return expr;
1730 /* Don't add translations of NAMEs as those are cheap to translate. */
1731 if (expr->kind != NAME)
1733 if (phi_trans_add (&slot, expr, e->src))
1734 return slot->v == 0 ? NULL : expression_for_id (slot->v);
1735 /* Store NULL for the value we want to return in the case of
1736 recursing. */
1737 slot->v = 0;
1740 /* Translate. */
1741 basic_block saved_valueize_bb = vn_context_bb;
1742 vn_context_bb = e->src;
1743 phitrans = phi_translate_1 (dest, expr, set1, set2, e);
1744 vn_context_bb = saved_valueize_bb;
1746 if (slot)
1748 /* We may have reallocated. */
1749 phi_trans_add (&slot, expr, e->src);
1750 if (phitrans)
1751 slot->v = get_expression_id (phitrans);
1752 else
1753 /* Remove failed translations again, they cause insert
1754 iteration to not pick up new opportunities reliably. */
1755 PHI_TRANS_TABLE (e->src)->clear_slot (slot);
1758 return phitrans;
1762 /* For each expression in SET, translate the values through phi nodes
1763 in PHIBLOCK using edge PHIBLOCK->PRED, and store the resulting
1764 expressions in DEST. */
1766 static void
1767 phi_translate_set (bitmap_set_t dest, bitmap_set_t set, edge e)
1769 bitmap_iterator bi;
1770 unsigned int i;
1772 if (gimple_seq_empty_p (phi_nodes (e->dest)))
1774 bitmap_set_copy (dest, set);
1775 return;
1778 /* Allocate the phi-translation cache where we have an idea about
1779 its size. hash-table implementation internals tell us that
1780 allocating the table to fit twice the number of elements will
1781 make sure we do not usually re-allocate. */
1782 if (!PHI_TRANS_TABLE (e->src))
1783 PHI_TRANS_TABLE (e->src) = new hash_table<expr_pred_trans_d>
1784 (2 * bitmap_count_bits (&set->expressions));
1785 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
1787 pre_expr expr = expression_for_id (i);
1788 pre_expr translated = phi_translate (dest, expr, set, NULL, e);
1789 if (!translated)
1790 continue;
1792 bitmap_insert_into_set (dest, translated);
1796 /* Find the leader for a value (i.e., the name representing that
1797 value) in a given set, and return it. Return NULL if no leader
1798 is found. */
1800 static pre_expr
1801 bitmap_find_leader (bitmap_set_t set, unsigned int val)
1803 if (value_id_constant_p (val))
1804 return constant_value_expressions[-val];
1806 if (bitmap_set_contains_value (set, val))
1808 /* Rather than walk the entire bitmap of expressions, and see
1809 whether any of them has the value we are looking for, we look
1810 at the reverse mapping, which tells us the set of expressions
1811 that have a given value (IE value->expressions with that
1812 value) and see if any of those expressions are in our set.
1813 The number of expressions per value is usually significantly
1814 less than the number of expressions in the set. In fact, for
1815 large testcases, doing it this way is roughly 5-10x faster
1816 than walking the bitmap.
1817 If this is somehow a significant lose for some cases, we can
1818 choose which set to walk based on which set is smaller. */
1819 unsigned int i;
1820 bitmap_iterator bi;
1821 bitmap exprset = value_expressions[val];
1823 if (!exprset->first->next)
1824 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
1825 if (bitmap_bit_p (&set->expressions, i))
1826 return expression_for_id (i);
1828 EXECUTE_IF_AND_IN_BITMAP (exprset, &set->expressions, 0, i, bi)
1829 return expression_for_id (i);
1831 return NULL;
1834 /* Determine if EXPR, a memory expression, is ANTIC_IN at the top of
1835 BLOCK by seeing if it is not killed in the block. Note that we are
1836 only determining whether there is a store that kills it. Because
1837 of the order in which clean iterates over values, we are guaranteed
1838 that altered operands will have caused us to be eliminated from the
1839 ANTIC_IN set already. */
1841 static bool
1842 value_dies_in_block_x (pre_expr expr, basic_block block)
1844 tree vuse = PRE_EXPR_REFERENCE (expr)->vuse;
1845 vn_reference_t refx = PRE_EXPR_REFERENCE (expr);
1846 gimple *def;
1847 gimple_stmt_iterator gsi;
1848 unsigned id = get_expression_id (expr);
1849 bool res = false;
1850 ao_ref ref;
1852 if (!vuse)
1853 return false;
1855 /* Lookup a previously calculated result. */
1856 if (EXPR_DIES (block)
1857 && bitmap_bit_p (EXPR_DIES (block), id * 2))
1858 return bitmap_bit_p (EXPR_DIES (block), id * 2 + 1);
1860 /* A memory expression {e, VUSE} dies in the block if there is a
1861 statement that may clobber e. If, starting statement walk from the
1862 top of the basic block, a statement uses VUSE there can be no kill
1863 inbetween that use and the original statement that loaded {e, VUSE},
1864 so we can stop walking. */
1865 ref.base = NULL_TREE;
1866 for (gsi = gsi_start_bb (block); !gsi_end_p (gsi); gsi_next (&gsi))
1868 tree def_vuse, def_vdef;
1869 def = gsi_stmt (gsi);
1870 def_vuse = gimple_vuse (def);
1871 def_vdef = gimple_vdef (def);
1873 /* Not a memory statement. */
1874 if (!def_vuse)
1875 continue;
1877 /* Not a may-def. */
1878 if (!def_vdef)
1880 /* A load with the same VUSE, we're done. */
1881 if (def_vuse == vuse)
1882 break;
1884 continue;
1887 /* Init ref only if we really need it. */
1888 if (ref.base == NULL_TREE
1889 && !ao_ref_init_from_vn_reference (&ref, refx->set, refx->base_set,
1890 refx->type, refx->operands))
1892 res = true;
1893 break;
1895 /* If the statement may clobber expr, it dies. */
1896 if (stmt_may_clobber_ref_p_1 (def, &ref))
1898 res = true;
1899 break;
1903 /* Remember the result. */
1904 if (!EXPR_DIES (block))
1905 EXPR_DIES (block) = BITMAP_ALLOC (&grand_bitmap_obstack);
1906 bitmap_set_bit (EXPR_DIES (block), id * 2);
1907 if (res)
1908 bitmap_set_bit (EXPR_DIES (block), id * 2 + 1);
1910 return res;
1914 /* Determine if OP is valid in SET1 U SET2, which it is when the union
1915 contains its value-id. */
1917 static bool
1918 op_valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, tree op)
1920 if (op && TREE_CODE (op) == SSA_NAME)
1922 unsigned int value_id = VN_INFO (op)->value_id;
1923 if (!(bitmap_set_contains_value (set1, value_id)
1924 || (set2 && bitmap_set_contains_value (set2, value_id))))
1925 return false;
1927 return true;
1930 /* Determine if the expression EXPR is valid in SET1 U SET2.
1931 ONLY SET2 CAN BE NULL.
1932 This means that we have a leader for each part of the expression
1933 (if it consists of values), or the expression is an SSA_NAME.
1934 For loads/calls, we also see if the vuse is killed in this block. */
1936 static bool
1937 valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, pre_expr expr)
1939 switch (expr->kind)
1941 case NAME:
1942 /* By construction all NAMEs are available. Non-available
1943 NAMEs are removed by subtracting TMP_GEN from the sets. */
1944 return true;
1945 case NARY:
1947 unsigned int i;
1948 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1949 for (i = 0; i < nary->length; i++)
1950 if (!op_valid_in_sets (set1, set2, nary->op[i]))
1951 return false;
1952 return true;
1954 break;
1955 case REFERENCE:
1957 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1958 vn_reference_op_t vro;
1959 unsigned int i;
1961 FOR_EACH_VEC_ELT (ref->operands, i, vro)
1963 if (!op_valid_in_sets (set1, set2, vro->op0)
1964 || !op_valid_in_sets (set1, set2, vro->op1)
1965 || !op_valid_in_sets (set1, set2, vro->op2))
1966 return false;
1968 return true;
1970 default:
1971 gcc_unreachable ();
1975 /* Clean the set of expressions SET1 that are no longer valid in SET1 or SET2.
1976 This means expressions that are made up of values we have no leaders for
1977 in SET1 or SET2. */
1979 static void
1980 clean (bitmap_set_t set1, bitmap_set_t set2 = NULL)
1982 vec<pre_expr> exprs = sorted_array_from_bitmap_set (set1);
1983 pre_expr expr;
1984 int i;
1986 FOR_EACH_VEC_ELT (exprs, i, expr)
1988 if (!valid_in_sets (set1, set2, expr))
1990 unsigned int val = get_expr_value_id (expr);
1991 bitmap_clear_bit (&set1->expressions, get_expression_id (expr));
1992 /* We are entered with possibly multiple expressions for a value
1993 so before removing a value from the set see if there's an
1994 expression for it left. */
1995 if (! bitmap_find_leader (set1, val))
1996 bitmap_clear_bit (&set1->values, val);
1999 exprs.release ();
2001 if (flag_checking)
2003 unsigned j;
2004 bitmap_iterator bi;
2005 FOR_EACH_EXPR_ID_IN_SET (set1, j, bi)
2006 gcc_assert (valid_in_sets (set1, set2, expression_for_id (j)));
2010 /* Clean the set of expressions that are no longer valid in SET because
2011 they are clobbered in BLOCK or because they trap and may not be executed. */
2013 static void
2014 prune_clobbered_mems (bitmap_set_t set, basic_block block)
2016 bitmap_iterator bi;
2017 unsigned i;
2018 unsigned to_remove = -1U;
2019 bool any_removed = false;
2021 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
2023 /* Remove queued expr. */
2024 if (to_remove != -1U)
2026 bitmap_clear_bit (&set->expressions, to_remove);
2027 any_removed = true;
2028 to_remove = -1U;
2031 pre_expr expr = expression_for_id (i);
2032 if (expr->kind == REFERENCE)
2034 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2035 if (ref->vuse)
2037 gimple *def_stmt = SSA_NAME_DEF_STMT (ref->vuse);
2038 if (!gimple_nop_p (def_stmt)
2039 /* If value-numbering provided a memory state for this
2040 that dominates BLOCK we're done, otherwise we have
2041 to check if the value dies in BLOCK. */
2042 && !(gimple_bb (def_stmt) != block
2043 && dominated_by_p (CDI_DOMINATORS,
2044 block, gimple_bb (def_stmt)))
2045 && value_dies_in_block_x (expr, block))
2046 to_remove = i;
2048 /* If the REFERENCE may trap make sure the block does not contain
2049 a possible exit point.
2050 ??? This is overly conservative if we translate AVAIL_OUT
2051 as the available expression might be after the exit point. */
2052 if (BB_MAY_NOTRETURN (block)
2053 && vn_reference_may_trap (ref))
2054 to_remove = i;
2056 else if (expr->kind == NARY)
2058 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2059 /* If the NARY may trap make sure the block does not contain
2060 a possible exit point.
2061 ??? This is overly conservative if we translate AVAIL_OUT
2062 as the available expression might be after the exit point. */
2063 if (BB_MAY_NOTRETURN (block)
2064 && vn_nary_may_trap (nary))
2065 to_remove = i;
2069 /* Remove queued expr. */
2070 if (to_remove != -1U)
2072 bitmap_clear_bit (&set->expressions, to_remove);
2073 any_removed = true;
2076 /* Above we only removed expressions, now clean the set of values
2077 which no longer have any corresponding expression. We cannot
2078 clear the value at the time we remove an expression since there
2079 may be multiple expressions per value.
2080 If we'd queue possibly to be removed values we could use
2081 the bitmap_find_leader way to see if there's still an expression
2082 for it. For some ratio of to be removed values and number of
2083 values/expressions in the set this might be faster than rebuilding
2084 the value-set. */
2085 if (any_removed)
2087 bitmap_clear (&set->values);
2088 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
2090 pre_expr expr = expression_for_id (i);
2091 unsigned int value_id = get_expr_value_id (expr);
2092 bitmap_set_bit (&set->values, value_id);
2097 /* Compute the ANTIC set for BLOCK.
2099 If succs(BLOCK) > 1 then
2100 ANTIC_OUT[BLOCK] = intersection of ANTIC_IN[b] for all succ(BLOCK)
2101 else if succs(BLOCK) == 1 then
2102 ANTIC_OUT[BLOCK] = phi_translate (ANTIC_IN[succ(BLOCK)])
2104 ANTIC_IN[BLOCK] = clean(ANTIC_OUT[BLOCK] U EXP_GEN[BLOCK] - TMP_GEN[BLOCK])
2106 Note that clean() is deferred until after the iteration. */
2108 static bool
2109 compute_antic_aux (basic_block block, bool block_has_abnormal_pred_edge)
2111 bitmap_set_t S, old, ANTIC_OUT;
2112 edge e;
2113 edge_iterator ei;
2115 bool was_visited = BB_VISITED (block);
2116 bool changed = ! BB_VISITED (block);
2117 BB_VISITED (block) = 1;
2118 old = ANTIC_OUT = S = NULL;
2120 /* If any edges from predecessors are abnormal, antic_in is empty,
2121 so do nothing. */
2122 if (block_has_abnormal_pred_edge)
2123 goto maybe_dump_sets;
2125 old = ANTIC_IN (block);
2126 ANTIC_OUT = bitmap_set_new ();
2128 /* If the block has no successors, ANTIC_OUT is empty. */
2129 if (EDGE_COUNT (block->succs) == 0)
2131 /* If we have one successor, we could have some phi nodes to
2132 translate through. */
2133 else if (single_succ_p (block))
2135 e = single_succ_edge (block);
2136 gcc_assert (BB_VISITED (e->dest));
2137 phi_translate_set (ANTIC_OUT, ANTIC_IN (e->dest), e);
2139 /* If we have multiple successors, we take the intersection of all of
2140 them. Note that in the case of loop exit phi nodes, we may have
2141 phis to translate through. */
2142 else
2144 size_t i;
2145 edge first = NULL;
2147 auto_vec<edge> worklist (EDGE_COUNT (block->succs));
2148 FOR_EACH_EDGE (e, ei, block->succs)
2150 if (!first
2151 && BB_VISITED (e->dest))
2152 first = e;
2153 else if (BB_VISITED (e->dest))
2154 worklist.quick_push (e);
2155 else
2157 /* Unvisited successors get their ANTIC_IN replaced by the
2158 maximal set to arrive at a maximum ANTIC_IN solution.
2159 We can ignore them in the intersection operation and thus
2160 need not explicitely represent that maximum solution. */
2161 if (dump_file && (dump_flags & TDF_DETAILS))
2162 fprintf (dump_file, "ANTIC_IN is MAX on %d->%d\n",
2163 e->src->index, e->dest->index);
2167 /* Of multiple successors we have to have visited one already
2168 which is guaranteed by iteration order. */
2169 gcc_assert (first != NULL);
2171 phi_translate_set (ANTIC_OUT, ANTIC_IN (first->dest), first);
2173 /* If we have multiple successors we need to intersect the ANTIC_OUT
2174 sets. For values that's a simple intersection but for
2175 expressions it is a union. Given we want to have a single
2176 expression per value in our sets we have to canonicalize.
2177 Avoid randomness and running into cycles like for PR82129 and
2178 canonicalize the expression we choose to the one with the
2179 lowest id. This requires we actually compute the union first. */
2180 FOR_EACH_VEC_ELT (worklist, i, e)
2182 if (!gimple_seq_empty_p (phi_nodes (e->dest)))
2184 bitmap_set_t tmp = bitmap_set_new ();
2185 phi_translate_set (tmp, ANTIC_IN (e->dest), e);
2186 bitmap_and_into (&ANTIC_OUT->values, &tmp->values);
2187 bitmap_ior_into (&ANTIC_OUT->expressions, &tmp->expressions);
2188 bitmap_set_free (tmp);
2190 else
2192 bitmap_and_into (&ANTIC_OUT->values, &ANTIC_IN (e->dest)->values);
2193 bitmap_ior_into (&ANTIC_OUT->expressions,
2194 &ANTIC_IN (e->dest)->expressions);
2197 if (! worklist.is_empty ())
2199 /* Prune expressions not in the value set. */
2200 bitmap_iterator bi;
2201 unsigned int i;
2202 unsigned int to_clear = -1U;
2203 FOR_EACH_EXPR_ID_IN_SET (ANTIC_OUT, i, bi)
2205 if (to_clear != -1U)
2207 bitmap_clear_bit (&ANTIC_OUT->expressions, to_clear);
2208 to_clear = -1U;
2210 pre_expr expr = expression_for_id (i);
2211 unsigned int value_id = get_expr_value_id (expr);
2212 if (!bitmap_bit_p (&ANTIC_OUT->values, value_id))
2213 to_clear = i;
2215 if (to_clear != -1U)
2216 bitmap_clear_bit (&ANTIC_OUT->expressions, to_clear);
2220 /* Dump ANTIC_OUT before it's pruned. */
2221 if (dump_file && (dump_flags & TDF_DETAILS))
2222 print_bitmap_set (dump_file, ANTIC_OUT, "ANTIC_OUT", block->index);
2224 /* Prune expressions that are clobbered in block and thus become
2225 invalid if translated from ANTIC_OUT to ANTIC_IN. */
2226 prune_clobbered_mems (ANTIC_OUT, block);
2228 /* Generate ANTIC_OUT - TMP_GEN. */
2229 S = bitmap_set_subtract_expressions (ANTIC_OUT, TMP_GEN (block));
2231 /* Start ANTIC_IN with EXP_GEN - TMP_GEN. */
2232 ANTIC_IN (block) = bitmap_set_subtract_expressions (EXP_GEN (block),
2233 TMP_GEN (block));
2235 /* Then union in the ANTIC_OUT - TMP_GEN values,
2236 to get ANTIC_OUT U EXP_GEN - TMP_GEN */
2237 bitmap_ior_into (&ANTIC_IN (block)->values, &S->values);
2238 bitmap_ior_into (&ANTIC_IN (block)->expressions, &S->expressions);
2240 /* clean (ANTIC_IN (block)) is defered to after the iteration converged
2241 because it can cause non-convergence, see for example PR81181. */
2243 /* Intersect ANTIC_IN with the old ANTIC_IN. This is required until
2244 we properly represent the maximum expression set, thus not prune
2245 values without expressions during the iteration. */
2246 if (was_visited
2247 && bitmap_and_into (&ANTIC_IN (block)->values, &old->values))
2249 if (dump_file && (dump_flags & TDF_DETAILS))
2250 fprintf (dump_file, "warning: intersecting with old ANTIC_IN "
2251 "shrinks the set\n");
2252 /* Prune expressions not in the value set. */
2253 bitmap_iterator bi;
2254 unsigned int i;
2255 unsigned int to_clear = -1U;
2256 FOR_EACH_EXPR_ID_IN_SET (ANTIC_IN (block), i, bi)
2258 if (to_clear != -1U)
2260 bitmap_clear_bit (&ANTIC_IN (block)->expressions, to_clear);
2261 to_clear = -1U;
2263 pre_expr expr = expression_for_id (i);
2264 unsigned int value_id = get_expr_value_id (expr);
2265 if (!bitmap_bit_p (&ANTIC_IN (block)->values, value_id))
2266 to_clear = i;
2268 if (to_clear != -1U)
2269 bitmap_clear_bit (&ANTIC_IN (block)->expressions, to_clear);
2272 if (!bitmap_set_equal (old, ANTIC_IN (block)))
2273 changed = true;
2275 maybe_dump_sets:
2276 if (dump_file && (dump_flags & TDF_DETAILS))
2278 if (changed)
2279 fprintf (dump_file, "[changed] ");
2280 print_bitmap_set (dump_file, ANTIC_IN (block), "ANTIC_IN",
2281 block->index);
2283 if (S)
2284 print_bitmap_set (dump_file, S, "S", block->index);
2286 if (old)
2287 bitmap_set_free (old);
2288 if (S)
2289 bitmap_set_free (S);
2290 if (ANTIC_OUT)
2291 bitmap_set_free (ANTIC_OUT);
2292 return changed;
2295 /* Compute PARTIAL_ANTIC for BLOCK.
2297 If succs(BLOCK) > 1 then
2298 PA_OUT[BLOCK] = value wise union of PA_IN[b] + all ANTIC_IN not
2299 in ANTIC_OUT for all succ(BLOCK)
2300 else if succs(BLOCK) == 1 then
2301 PA_OUT[BLOCK] = phi_translate (PA_IN[succ(BLOCK)])
2303 PA_IN[BLOCK] = clean(PA_OUT[BLOCK] - TMP_GEN[BLOCK] - ANTIC_IN[BLOCK])
2306 static void
2307 compute_partial_antic_aux (basic_block block,
2308 bool block_has_abnormal_pred_edge)
2310 bitmap_set_t old_PA_IN;
2311 bitmap_set_t PA_OUT;
2312 edge e;
2313 edge_iterator ei;
2314 unsigned long max_pa = param_max_partial_antic_length;
2316 old_PA_IN = PA_OUT = NULL;
2318 /* If any edges from predecessors are abnormal, antic_in is empty,
2319 so do nothing. */
2320 if (block_has_abnormal_pred_edge)
2321 goto maybe_dump_sets;
2323 /* If there are too many partially anticipatable values in the
2324 block, phi_translate_set can take an exponential time: stop
2325 before the translation starts. */
2326 if (max_pa
2327 && single_succ_p (block)
2328 && bitmap_count_bits (&PA_IN (single_succ (block))->values) > max_pa)
2329 goto maybe_dump_sets;
2331 old_PA_IN = PA_IN (block);
2332 PA_OUT = bitmap_set_new ();
2334 /* If the block has no successors, ANTIC_OUT is empty. */
2335 if (EDGE_COUNT (block->succs) == 0)
2337 /* If we have one successor, we could have some phi nodes to
2338 translate through. Note that we can't phi translate across DFS
2339 back edges in partial antic, because it uses a union operation on
2340 the successors. For recurrences like IV's, we will end up
2341 generating a new value in the set on each go around (i + 3 (VH.1)
2342 VH.1 + 1 (VH.2), VH.2 + 1 (VH.3), etc), forever. */
2343 else if (single_succ_p (block))
2345 e = single_succ_edge (block);
2346 if (!(e->flags & EDGE_DFS_BACK))
2347 phi_translate_set (PA_OUT, PA_IN (e->dest), e);
2349 /* If we have multiple successors, we take the union of all of
2350 them. */
2351 else
2353 size_t i;
2355 auto_vec<edge> worklist (EDGE_COUNT (block->succs));
2356 FOR_EACH_EDGE (e, ei, block->succs)
2358 if (e->flags & EDGE_DFS_BACK)
2359 continue;
2360 worklist.quick_push (e);
2362 if (worklist.length () > 0)
2364 FOR_EACH_VEC_ELT (worklist, i, e)
2366 unsigned int i;
2367 bitmap_iterator bi;
2369 if (!gimple_seq_empty_p (phi_nodes (e->dest)))
2371 bitmap_set_t antic_in = bitmap_set_new ();
2372 phi_translate_set (antic_in, ANTIC_IN (e->dest), e);
2373 FOR_EACH_EXPR_ID_IN_SET (antic_in, i, bi)
2374 bitmap_value_insert_into_set (PA_OUT,
2375 expression_for_id (i));
2376 bitmap_set_free (antic_in);
2377 bitmap_set_t pa_in = bitmap_set_new ();
2378 phi_translate_set (pa_in, PA_IN (e->dest), e);
2379 FOR_EACH_EXPR_ID_IN_SET (pa_in, i, bi)
2380 bitmap_value_insert_into_set (PA_OUT,
2381 expression_for_id (i));
2382 bitmap_set_free (pa_in);
2384 else
2386 FOR_EACH_EXPR_ID_IN_SET (ANTIC_IN (e->dest), i, bi)
2387 bitmap_value_insert_into_set (PA_OUT,
2388 expression_for_id (i));
2389 FOR_EACH_EXPR_ID_IN_SET (PA_IN (e->dest), i, bi)
2390 bitmap_value_insert_into_set (PA_OUT,
2391 expression_for_id (i));
2397 /* Prune expressions that are clobbered in block and thus become
2398 invalid if translated from PA_OUT to PA_IN. */
2399 prune_clobbered_mems (PA_OUT, block);
2401 /* PA_IN starts with PA_OUT - TMP_GEN.
2402 Then we subtract things from ANTIC_IN. */
2403 PA_IN (block) = bitmap_set_subtract_expressions (PA_OUT, TMP_GEN (block));
2405 /* For partial antic, we want to put back in the phi results, since
2406 we will properly avoid making them partially antic over backedges. */
2407 bitmap_ior_into (&PA_IN (block)->values, &PHI_GEN (block)->values);
2408 bitmap_ior_into (&PA_IN (block)->expressions, &PHI_GEN (block)->expressions);
2410 /* PA_IN[block] = PA_IN[block] - ANTIC_IN[block] */
2411 bitmap_set_subtract_values (PA_IN (block), ANTIC_IN (block));
2413 clean (PA_IN (block), ANTIC_IN (block));
2415 maybe_dump_sets:
2416 if (dump_file && (dump_flags & TDF_DETAILS))
2418 if (PA_OUT)
2419 print_bitmap_set (dump_file, PA_OUT, "PA_OUT", block->index);
2421 print_bitmap_set (dump_file, PA_IN (block), "PA_IN", block->index);
2423 if (old_PA_IN)
2424 bitmap_set_free (old_PA_IN);
2425 if (PA_OUT)
2426 bitmap_set_free (PA_OUT);
2429 /* Compute ANTIC and partial ANTIC sets. */
2431 static void
2432 compute_antic (void)
2434 bool changed = true;
2435 int num_iterations = 0;
2436 basic_block block;
2437 int i;
2438 edge_iterator ei;
2439 edge e;
2441 /* If any predecessor edges are abnormal, we punt, so antic_in is empty.
2442 We pre-build the map of blocks with incoming abnormal edges here. */
2443 auto_sbitmap has_abnormal_preds (last_basic_block_for_fn (cfun));
2444 bitmap_clear (has_abnormal_preds);
2446 FOR_ALL_BB_FN (block, cfun)
2448 BB_VISITED (block) = 0;
2450 FOR_EACH_EDGE (e, ei, block->preds)
2451 if (e->flags & EDGE_ABNORMAL)
2453 bitmap_set_bit (has_abnormal_preds, block->index);
2454 break;
2457 /* While we are here, give empty ANTIC_IN sets to each block. */
2458 ANTIC_IN (block) = bitmap_set_new ();
2459 if (do_partial_partial)
2460 PA_IN (block) = bitmap_set_new ();
2463 /* At the exit block we anticipate nothing. */
2464 BB_VISITED (EXIT_BLOCK_PTR_FOR_FN (cfun)) = 1;
2466 /* For ANTIC computation we need a postorder that also guarantees that
2467 a block with a single successor is visited after its successor.
2468 RPO on the inverted CFG has this property. */
2469 int *rpo = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
2470 int n = inverted_rev_post_order_compute (cfun, rpo);
2472 auto_sbitmap worklist (last_basic_block_for_fn (cfun) + 1);
2473 bitmap_clear (worklist);
2474 FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR_FOR_FN (cfun)->preds)
2475 bitmap_set_bit (worklist, e->src->index);
2476 while (changed)
2478 if (dump_file && (dump_flags & TDF_DETAILS))
2479 fprintf (dump_file, "Starting iteration %d\n", num_iterations);
2480 /* ??? We need to clear our PHI translation cache here as the
2481 ANTIC sets shrink and we restrict valid translations to
2482 those having operands with leaders in ANTIC. Same below
2483 for PA ANTIC computation. */
2484 num_iterations++;
2485 changed = false;
2486 for (i = 0; i < n; ++i)
2488 if (bitmap_bit_p (worklist, rpo[i]))
2490 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[i]);
2491 bitmap_clear_bit (worklist, block->index);
2492 if (compute_antic_aux (block,
2493 bitmap_bit_p (has_abnormal_preds,
2494 block->index)))
2496 FOR_EACH_EDGE (e, ei, block->preds)
2497 bitmap_set_bit (worklist, e->src->index);
2498 changed = true;
2502 /* Theoretically possible, but *highly* unlikely. */
2503 gcc_checking_assert (num_iterations < 500);
2506 /* We have to clean after the dataflow problem converged as cleaning
2507 can cause non-convergence because it is based on expressions
2508 rather than values. */
2509 FOR_EACH_BB_FN (block, cfun)
2510 clean (ANTIC_IN (block));
2512 statistics_histogram_event (cfun, "compute_antic iterations",
2513 num_iterations);
2515 if (do_partial_partial)
2517 /* For partial antic we ignore backedges and thus we do not need
2518 to perform any iteration when we process blocks in rpo. */
2519 for (i = 0; i < n; ++i)
2521 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[i]);
2522 compute_partial_antic_aux (block,
2523 bitmap_bit_p (has_abnormal_preds,
2524 block->index));
2528 free (rpo);
2532 /* Inserted expressions are placed onto this worklist, which is used
2533 for performing quick dead code elimination of insertions we made
2534 that didn't turn out to be necessary. */
2535 static bitmap inserted_exprs;
2537 /* The actual worker for create_component_ref_by_pieces. */
2539 static tree
2540 create_component_ref_by_pieces_1 (basic_block block, vn_reference_t ref,
2541 unsigned int *operand, gimple_seq *stmts)
2543 vn_reference_op_t currop = &ref->operands[*operand];
2544 tree genop;
2545 ++*operand;
2546 switch (currop->opcode)
2548 case CALL_EXPR:
2549 gcc_unreachable ();
2551 case MEM_REF:
2553 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2554 stmts);
2555 if (!baseop)
2556 return NULL_TREE;
2557 tree offset = currop->op0;
2558 if (TREE_CODE (baseop) == ADDR_EXPR
2559 && handled_component_p (TREE_OPERAND (baseop, 0)))
2561 poly_int64 off;
2562 tree base;
2563 base = get_addr_base_and_unit_offset (TREE_OPERAND (baseop, 0),
2564 &off);
2565 gcc_assert (base);
2566 offset = int_const_binop (PLUS_EXPR, offset,
2567 build_int_cst (TREE_TYPE (offset),
2568 off));
2569 baseop = build_fold_addr_expr (base);
2571 genop = build2 (MEM_REF, currop->type, baseop, offset);
2572 MR_DEPENDENCE_CLIQUE (genop) = currop->clique;
2573 MR_DEPENDENCE_BASE (genop) = currop->base;
2574 REF_REVERSE_STORAGE_ORDER (genop) = currop->reverse;
2575 return genop;
2578 case TARGET_MEM_REF:
2580 tree genop0 = NULL_TREE, genop1 = NULL_TREE;
2581 vn_reference_op_t nextop = &ref->operands[(*operand)++];
2582 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2583 stmts);
2584 if (!baseop)
2585 return NULL_TREE;
2586 if (currop->op0)
2588 genop0 = find_or_generate_expression (block, currop->op0, stmts);
2589 if (!genop0)
2590 return NULL_TREE;
2592 if (nextop->op0)
2594 genop1 = find_or_generate_expression (block, nextop->op0, stmts);
2595 if (!genop1)
2596 return NULL_TREE;
2598 genop = build5 (TARGET_MEM_REF, currop->type,
2599 baseop, currop->op2, genop0, currop->op1, genop1);
2601 MR_DEPENDENCE_CLIQUE (genop) = currop->clique;
2602 MR_DEPENDENCE_BASE (genop) = currop->base;
2603 return genop;
2606 case ADDR_EXPR:
2607 if (currop->op0)
2609 gcc_assert (is_gimple_min_invariant (currop->op0));
2610 return currop->op0;
2612 /* Fallthrough. */
2613 case REALPART_EXPR:
2614 case IMAGPART_EXPR:
2615 case VIEW_CONVERT_EXPR:
2617 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2618 stmts);
2619 if (!genop0)
2620 return NULL_TREE;
2621 return fold_build1 (currop->opcode, currop->type, genop0);
2624 case WITH_SIZE_EXPR:
2626 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2627 stmts);
2628 if (!genop0)
2629 return NULL_TREE;
2630 tree genop1 = find_or_generate_expression (block, currop->op0, stmts);
2631 if (!genop1)
2632 return NULL_TREE;
2633 return fold_build2 (currop->opcode, currop->type, genop0, genop1);
2636 case BIT_FIELD_REF:
2638 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2639 stmts);
2640 if (!genop0)
2641 return NULL_TREE;
2642 tree op1 = currop->op0;
2643 tree op2 = currop->op1;
2644 tree t = build3 (BIT_FIELD_REF, currop->type, genop0, op1, op2);
2645 REF_REVERSE_STORAGE_ORDER (t) = currop->reverse;
2646 return fold (t);
2649 /* For array ref vn_reference_op's, operand 1 of the array ref
2650 is op0 of the reference op and operand 3 of the array ref is
2651 op1. */
2652 case ARRAY_RANGE_REF:
2653 case ARRAY_REF:
2655 tree genop0;
2656 tree genop1 = currop->op0;
2657 tree genop2 = currop->op1;
2658 tree genop3 = currop->op2;
2659 genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2660 stmts);
2661 if (!genop0)
2662 return NULL_TREE;
2663 genop1 = find_or_generate_expression (block, genop1, stmts);
2664 if (!genop1)
2665 return NULL_TREE;
2666 if (genop2)
2668 tree domain_type = TYPE_DOMAIN (TREE_TYPE (genop0));
2669 /* Drop zero minimum index if redundant. */
2670 if (integer_zerop (genop2)
2671 && (!domain_type
2672 || integer_zerop (TYPE_MIN_VALUE (domain_type))))
2673 genop2 = NULL_TREE;
2674 else
2676 genop2 = find_or_generate_expression (block, genop2, stmts);
2677 if (!genop2)
2678 return NULL_TREE;
2681 if (genop3)
2683 tree elmt_type = TREE_TYPE (TREE_TYPE (genop0));
2684 /* We can't always put a size in units of the element alignment
2685 here as the element alignment may be not visible. See
2686 PR43783. Simply drop the element size for constant
2687 sizes. */
2688 if (TREE_CODE (genop3) == INTEGER_CST
2689 && TREE_CODE (TYPE_SIZE_UNIT (elmt_type)) == INTEGER_CST
2690 && wi::eq_p (wi::to_offset (TYPE_SIZE_UNIT (elmt_type)),
2691 (wi::to_offset (genop3)
2692 * vn_ref_op_align_unit (currop))))
2693 genop3 = NULL_TREE;
2694 else
2696 genop3 = find_or_generate_expression (block, genop3, stmts);
2697 if (!genop3)
2698 return NULL_TREE;
2701 return build4 (currop->opcode, currop->type, genop0, genop1,
2702 genop2, genop3);
2704 case COMPONENT_REF:
2706 tree op0;
2707 tree op1;
2708 tree genop2 = currop->op1;
2709 op0 = create_component_ref_by_pieces_1 (block, ref, operand, stmts);
2710 if (!op0)
2711 return NULL_TREE;
2712 /* op1 should be a FIELD_DECL, which are represented by themselves. */
2713 op1 = currop->op0;
2714 if (genop2)
2716 genop2 = find_or_generate_expression (block, genop2, stmts);
2717 if (!genop2)
2718 return NULL_TREE;
2720 return fold_build3 (COMPONENT_REF, TREE_TYPE (op1), op0, op1, genop2);
2723 case SSA_NAME:
2725 genop = find_or_generate_expression (block, currop->op0, stmts);
2726 return genop;
2728 case STRING_CST:
2729 case INTEGER_CST:
2730 case POLY_INT_CST:
2731 case COMPLEX_CST:
2732 case VECTOR_CST:
2733 case REAL_CST:
2734 case CONSTRUCTOR:
2735 case VAR_DECL:
2736 case PARM_DECL:
2737 case CONST_DECL:
2738 case RESULT_DECL:
2739 case FUNCTION_DECL:
2740 return currop->op0;
2742 default:
2743 gcc_unreachable ();
2747 /* For COMPONENT_REF's and ARRAY_REF's, we can't have any intermediates for the
2748 COMPONENT_REF or MEM_REF or ARRAY_REF portion, because we'd end up with
2749 trying to rename aggregates into ssa form directly, which is a no no.
2751 Thus, this routine doesn't create temporaries, it just builds a
2752 single access expression for the array, calling
2753 find_or_generate_expression to build the innermost pieces.
2755 This function is a subroutine of create_expression_by_pieces, and
2756 should not be called on it's own unless you really know what you
2757 are doing. */
2759 static tree
2760 create_component_ref_by_pieces (basic_block block, vn_reference_t ref,
2761 gimple_seq *stmts)
2763 unsigned int op = 0;
2764 return create_component_ref_by_pieces_1 (block, ref, &op, stmts);
2767 /* Find a simple leader for an expression, or generate one using
2768 create_expression_by_pieces from a NARY expression for the value.
2769 BLOCK is the basic_block we are looking for leaders in.
2770 OP is the tree expression to find a leader for or generate.
2771 Returns the leader or NULL_TREE on failure. */
2773 static tree
2774 find_or_generate_expression (basic_block block, tree op, gimple_seq *stmts)
2776 /* Constants are always leaders. */
2777 if (is_gimple_min_invariant (op))
2778 return op;
2780 gcc_assert (TREE_CODE (op) == SSA_NAME);
2781 vn_ssa_aux_t info = VN_INFO (op);
2782 unsigned int lookfor = info->value_id;
2783 if (value_id_constant_p (lookfor))
2784 return info->valnum;
2786 pre_expr leader = bitmap_find_leader (AVAIL_OUT (block), lookfor);
2787 if (leader)
2789 if (leader->kind == NAME)
2790 return PRE_EXPR_NAME (leader);
2791 else if (leader->kind == CONSTANT)
2792 return PRE_EXPR_CONSTANT (leader);
2794 /* Defer. */
2795 return NULL_TREE;
2797 gcc_assert (!value_id_constant_p (lookfor));
2799 /* It must be a complex expression, so generate it recursively. Note
2800 that this is only necessary to handle gcc.dg/tree-ssa/ssa-pre28.c
2801 where the insert algorithm fails to insert a required expression. */
2802 bitmap exprset = value_expressions[lookfor];
2803 bitmap_iterator bi;
2804 unsigned int i;
2805 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
2807 pre_expr temp = expression_for_id (i);
2808 /* We cannot insert random REFERENCE expressions at arbitrary
2809 places. We can insert NARYs which eventually re-materializes
2810 its operand values. */
2811 if (temp->kind == NARY)
2812 return create_expression_by_pieces (block, temp, stmts,
2813 TREE_TYPE (op));
2816 /* Defer. */
2817 return NULL_TREE;
2820 /* Create an expression in pieces, so that we can handle very complex
2821 expressions that may be ANTIC, but not necessary GIMPLE.
2822 BLOCK is the basic block the expression will be inserted into,
2823 EXPR is the expression to insert (in value form)
2824 STMTS is a statement list to append the necessary insertions into.
2826 This function will die if we hit some value that shouldn't be
2827 ANTIC but is (IE there is no leader for it, or its components).
2828 The function returns NULL_TREE in case a different antic expression
2829 has to be inserted first.
2830 This function may also generate expressions that are themselves
2831 partially or fully redundant. Those that are will be either made
2832 fully redundant during the next iteration of insert (for partially
2833 redundant ones), or eliminated by eliminate (for fully redundant
2834 ones). */
2836 static tree
2837 create_expression_by_pieces (basic_block block, pre_expr expr,
2838 gimple_seq *stmts, tree type)
2840 tree name;
2841 tree folded;
2842 gimple_seq forced_stmts = NULL;
2843 unsigned int value_id;
2844 gimple_stmt_iterator gsi;
2845 tree exprtype = type ? type : get_expr_type (expr);
2846 pre_expr nameexpr;
2847 gassign *newstmt;
2849 switch (expr->kind)
2851 /* We may hit the NAME/CONSTANT case if we have to convert types
2852 that value numbering saw through. */
2853 case NAME:
2854 folded = PRE_EXPR_NAME (expr);
2855 if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (folded))
2856 return NULL_TREE;
2857 if (useless_type_conversion_p (exprtype, TREE_TYPE (folded)))
2858 return folded;
2859 break;
2860 case CONSTANT:
2862 folded = PRE_EXPR_CONSTANT (expr);
2863 tree tem = fold_convert (exprtype, folded);
2864 if (is_gimple_min_invariant (tem))
2865 return tem;
2866 break;
2868 case REFERENCE:
2869 if (PRE_EXPR_REFERENCE (expr)->operands[0].opcode == CALL_EXPR)
2871 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2872 unsigned int operand = 1;
2873 vn_reference_op_t currop = &ref->operands[0];
2874 tree sc = NULL_TREE;
2875 tree fn = NULL_TREE;
2876 if (currop->op0)
2878 fn = find_or_generate_expression (block, currop->op0, stmts);
2879 if (!fn)
2880 return NULL_TREE;
2882 if (currop->op1)
2884 sc = find_or_generate_expression (block, currop->op1, stmts);
2885 if (!sc)
2886 return NULL_TREE;
2888 auto_vec<tree> args (ref->operands.length () - 1);
2889 while (operand < ref->operands.length ())
2891 tree arg = create_component_ref_by_pieces_1 (block, ref,
2892 &operand, stmts);
2893 if (!arg)
2894 return NULL_TREE;
2895 args.quick_push (arg);
2897 gcall *call;
2898 if (currop->op0)
2900 call = gimple_build_call_vec (fn, args);
2901 gimple_call_set_fntype (call, currop->type);
2903 else
2904 call = gimple_build_call_internal_vec ((internal_fn)currop->clique,
2905 args);
2906 gimple_set_location (call, expr->loc);
2907 if (sc)
2908 gimple_call_set_chain (call, sc);
2909 tree forcedname = make_ssa_name (ref->type);
2910 gimple_call_set_lhs (call, forcedname);
2911 /* There's no CCP pass after PRE which would re-compute alignment
2912 information so make sure we re-materialize this here. */
2913 if (gimple_call_builtin_p (call, BUILT_IN_ASSUME_ALIGNED)
2914 && args.length () - 2 <= 1
2915 && tree_fits_uhwi_p (args[1])
2916 && (args.length () != 3 || tree_fits_uhwi_p (args[2])))
2918 unsigned HOST_WIDE_INT halign = tree_to_uhwi (args[1]);
2919 unsigned HOST_WIDE_INT hmisalign
2920 = args.length () == 3 ? tree_to_uhwi (args[2]) : 0;
2921 if ((halign & (halign - 1)) == 0
2922 && (hmisalign & ~(halign - 1)) == 0
2923 && (unsigned int)halign != 0)
2924 set_ptr_info_alignment (get_ptr_info (forcedname),
2925 halign, hmisalign);
2927 gimple_set_vuse (call, BB_LIVE_VOP_ON_EXIT (block));
2928 gimple_seq_add_stmt_without_update (&forced_stmts, call);
2929 folded = forcedname;
2931 else
2933 folded = create_component_ref_by_pieces (block,
2934 PRE_EXPR_REFERENCE (expr),
2935 stmts);
2936 if (!folded)
2937 return NULL_TREE;
2938 name = make_temp_ssa_name (exprtype, NULL, "pretmp");
2939 newstmt = gimple_build_assign (name, folded);
2940 gimple_set_location (newstmt, expr->loc);
2941 gimple_seq_add_stmt_without_update (&forced_stmts, newstmt);
2942 gimple_set_vuse (newstmt, BB_LIVE_VOP_ON_EXIT (block));
2943 folded = name;
2945 break;
2946 case NARY:
2948 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2949 tree *genop = XALLOCAVEC (tree, nary->length);
2950 unsigned i;
2951 for (i = 0; i < nary->length; ++i)
2953 genop[i] = find_or_generate_expression (block, nary->op[i], stmts);
2954 if (!genop[i])
2955 return NULL_TREE;
2956 /* Ensure genop[] is properly typed for POINTER_PLUS_EXPR. It
2957 may have conversions stripped. */
2958 if (nary->opcode == POINTER_PLUS_EXPR)
2960 if (i == 0)
2961 genop[i] = gimple_convert (&forced_stmts,
2962 nary->type, genop[i]);
2963 else if (i == 1)
2964 genop[i] = gimple_convert (&forced_stmts,
2965 sizetype, genop[i]);
2967 else
2968 genop[i] = gimple_convert (&forced_stmts,
2969 TREE_TYPE (nary->op[i]), genop[i]);
2971 if (nary->opcode == CONSTRUCTOR)
2973 vec<constructor_elt, va_gc> *elts = NULL;
2974 for (i = 0; i < nary->length; ++i)
2975 CONSTRUCTOR_APPEND_ELT (elts, NULL_TREE, genop[i]);
2976 folded = build_constructor (nary->type, elts);
2977 name = make_temp_ssa_name (exprtype, NULL, "pretmp");
2978 newstmt = gimple_build_assign (name, folded);
2979 gimple_set_location (newstmt, expr->loc);
2980 gimple_seq_add_stmt_without_update (&forced_stmts, newstmt);
2981 folded = name;
2983 else
2985 switch (nary->length)
2987 case 1:
2988 folded = gimple_build (&forced_stmts, expr->loc,
2989 nary->opcode, nary->type, genop[0]);
2990 break;
2991 case 2:
2992 folded = gimple_build (&forced_stmts, expr->loc, nary->opcode,
2993 nary->type, genop[0], genop[1]);
2994 break;
2995 case 3:
2996 folded = gimple_build (&forced_stmts, expr->loc, nary->opcode,
2997 nary->type, genop[0], genop[1],
2998 genop[2]);
2999 break;
3000 default:
3001 gcc_unreachable ();
3005 break;
3006 default:
3007 gcc_unreachable ();
3010 folded = gimple_convert (&forced_stmts, exprtype, folded);
3012 /* If there is nothing to insert, return the simplified result. */
3013 if (gimple_seq_empty_p (forced_stmts))
3014 return folded;
3015 /* If we simplified to a constant return it and discard eventually
3016 built stmts. */
3017 if (is_gimple_min_invariant (folded))
3019 gimple_seq_discard (forced_stmts);
3020 return folded;
3022 /* Likewise if we simplified to sth not queued for insertion. */
3023 bool found = false;
3024 gsi = gsi_last (forced_stmts);
3025 for (; !gsi_end_p (gsi); gsi_prev (&gsi))
3027 gimple *stmt = gsi_stmt (gsi);
3028 tree forcedname = gimple_get_lhs (stmt);
3029 if (forcedname == folded)
3031 found = true;
3032 break;
3035 if (! found)
3037 gimple_seq_discard (forced_stmts);
3038 return folded;
3040 gcc_assert (TREE_CODE (folded) == SSA_NAME);
3042 /* If we have any intermediate expressions to the value sets, add them
3043 to the value sets and chain them in the instruction stream. */
3044 if (forced_stmts)
3046 gsi = gsi_start (forced_stmts);
3047 for (; !gsi_end_p (gsi); gsi_next (&gsi))
3049 gimple *stmt = gsi_stmt (gsi);
3050 tree forcedname = gimple_get_lhs (stmt);
3051 pre_expr nameexpr;
3053 if (forcedname != folded)
3055 vn_ssa_aux_t vn_info = VN_INFO (forcedname);
3056 vn_info->valnum = forcedname;
3057 vn_info->value_id = get_next_value_id ();
3058 nameexpr = get_or_alloc_expr_for_name (forcedname);
3059 add_to_value (vn_info->value_id, nameexpr);
3060 if (NEW_SETS (block))
3061 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
3062 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
3065 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (forcedname));
3067 gimple_seq_add_seq (stmts, forced_stmts);
3070 name = folded;
3072 /* Fold the last statement. */
3073 gsi = gsi_last (*stmts);
3074 if (fold_stmt_inplace (&gsi))
3075 update_stmt (gsi_stmt (gsi));
3077 /* Add a value number to the temporary.
3078 The value may already exist in either NEW_SETS, or AVAIL_OUT, because
3079 we are creating the expression by pieces, and this particular piece of
3080 the expression may have been represented. There is no harm in replacing
3081 here. */
3082 value_id = get_expr_value_id (expr);
3083 vn_ssa_aux_t vn_info = VN_INFO (name);
3084 vn_info->value_id = value_id;
3085 vn_info->valnum = vn_valnum_from_value_id (value_id);
3086 if (vn_info->valnum == NULL_TREE)
3087 vn_info->valnum = name;
3088 gcc_assert (vn_info->valnum != NULL_TREE);
3089 nameexpr = get_or_alloc_expr_for_name (name);
3090 add_to_value (value_id, nameexpr);
3091 if (NEW_SETS (block))
3092 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
3093 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
3095 pre_stats.insertions++;
3096 if (dump_file && (dump_flags & TDF_DETAILS))
3098 fprintf (dump_file, "Inserted ");
3099 print_gimple_stmt (dump_file, gsi_stmt (gsi_last (*stmts)), 0);
3100 fprintf (dump_file, " in predecessor %d (%04d)\n",
3101 block->index, value_id);
3104 return name;
3108 /* Insert the to-be-made-available values of expression EXPRNUM for each
3109 predecessor, stored in AVAIL, into the predecessors of BLOCK, and
3110 merge the result with a phi node, given the same value number as
3111 NODE. Return true if we have inserted new stuff. */
3113 static bool
3114 insert_into_preds_of_block (basic_block block, unsigned int exprnum,
3115 vec<pre_expr> &avail)
3117 pre_expr expr = expression_for_id (exprnum);
3118 pre_expr newphi;
3119 unsigned int val = get_expr_value_id (expr);
3120 edge pred;
3121 bool insertions = false;
3122 bool nophi = false;
3123 basic_block bprime;
3124 pre_expr eprime;
3125 edge_iterator ei;
3126 tree type = get_expr_type (expr);
3127 tree temp;
3128 gphi *phi;
3130 /* Make sure we aren't creating an induction variable. */
3131 if (bb_loop_depth (block) > 0 && EDGE_COUNT (block->preds) == 2)
3133 bool firstinsideloop = false;
3134 bool secondinsideloop = false;
3135 firstinsideloop = flow_bb_inside_loop_p (block->loop_father,
3136 EDGE_PRED (block, 0)->src);
3137 secondinsideloop = flow_bb_inside_loop_p (block->loop_father,
3138 EDGE_PRED (block, 1)->src);
3139 /* Induction variables only have one edge inside the loop. */
3140 if ((firstinsideloop ^ secondinsideloop)
3141 && expr->kind != REFERENCE)
3143 if (dump_file && (dump_flags & TDF_DETAILS))
3144 fprintf (dump_file, "Skipping insertion of phi for partial "
3145 "redundancy: Looks like an induction variable\n");
3146 nophi = true;
3150 /* Make the necessary insertions. */
3151 FOR_EACH_EDGE (pred, ei, block->preds)
3153 /* When we are not inserting a PHI node do not bother inserting
3154 into places that do not dominate the anticipated computations. */
3155 if (nophi && !dominated_by_p (CDI_DOMINATORS, block, pred->src))
3156 continue;
3157 gimple_seq stmts = NULL;
3158 tree builtexpr;
3159 bprime = pred->src;
3160 eprime = avail[pred->dest_idx];
3161 builtexpr = create_expression_by_pieces (bprime, eprime,
3162 &stmts, type);
3163 gcc_assert (!(pred->flags & EDGE_ABNORMAL));
3164 if (!gimple_seq_empty_p (stmts))
3166 basic_block new_bb = gsi_insert_seq_on_edge_immediate (pred, stmts);
3167 gcc_assert (! new_bb);
3168 insertions = true;
3170 if (!builtexpr)
3172 /* We cannot insert a PHI node if we failed to insert
3173 on one edge. */
3174 nophi = true;
3175 continue;
3177 if (is_gimple_min_invariant (builtexpr))
3178 avail[pred->dest_idx] = get_or_alloc_expr_for_constant (builtexpr);
3179 else
3180 avail[pred->dest_idx] = get_or_alloc_expr_for_name (builtexpr);
3182 /* If we didn't want a phi node, and we made insertions, we still have
3183 inserted new stuff, and thus return true. If we didn't want a phi node,
3184 and didn't make insertions, we haven't added anything new, so return
3185 false. */
3186 if (nophi && insertions)
3187 return true;
3188 else if (nophi && !insertions)
3189 return false;
3191 /* Now build a phi for the new variable. */
3192 temp = make_temp_ssa_name (type, NULL, "prephitmp");
3193 phi = create_phi_node (temp, block);
3195 vn_ssa_aux_t vn_info = VN_INFO (temp);
3196 vn_info->value_id = val;
3197 vn_info->valnum = vn_valnum_from_value_id (val);
3198 if (vn_info->valnum == NULL_TREE)
3199 vn_info->valnum = temp;
3200 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3201 FOR_EACH_EDGE (pred, ei, block->preds)
3203 pre_expr ae = avail[pred->dest_idx];
3204 gcc_assert (get_expr_type (ae) == type
3205 || useless_type_conversion_p (type, get_expr_type (ae)));
3206 if (ae->kind == CONSTANT)
3207 add_phi_arg (phi, unshare_expr (PRE_EXPR_CONSTANT (ae)),
3208 pred, UNKNOWN_LOCATION);
3209 else
3210 add_phi_arg (phi, PRE_EXPR_NAME (ae), pred, UNKNOWN_LOCATION);
3213 newphi = get_or_alloc_expr_for_name (temp);
3214 add_to_value (val, newphi);
3216 /* The value should *not* exist in PHI_GEN, or else we wouldn't be doing
3217 this insertion, since we test for the existence of this value in PHI_GEN
3218 before proceeding with the partial redundancy checks in insert_aux.
3220 The value may exist in AVAIL_OUT, in particular, it could be represented
3221 by the expression we are trying to eliminate, in which case we want the
3222 replacement to occur. If it's not existing in AVAIL_OUT, we want it
3223 inserted there.
3225 Similarly, to the PHI_GEN case, the value should not exist in NEW_SETS of
3226 this block, because if it did, it would have existed in our dominator's
3227 AVAIL_OUT, and would have been skipped due to the full redundancy check.
3230 bitmap_insert_into_set (PHI_GEN (block), newphi);
3231 bitmap_value_replace_in_set (AVAIL_OUT (block),
3232 newphi);
3233 if (NEW_SETS (block))
3234 bitmap_insert_into_set (NEW_SETS (block), newphi);
3236 /* If we insert a PHI node for a conversion of another PHI node
3237 in the same basic-block try to preserve range information.
3238 This is important so that followup loop passes receive optimal
3239 number of iteration analysis results. See PR61743. */
3240 if (expr->kind == NARY
3241 && CONVERT_EXPR_CODE_P (expr->u.nary->opcode)
3242 && TREE_CODE (expr->u.nary->op[0]) == SSA_NAME
3243 && gimple_bb (SSA_NAME_DEF_STMT (expr->u.nary->op[0])) == block
3244 && INTEGRAL_TYPE_P (type)
3245 && INTEGRAL_TYPE_P (TREE_TYPE (expr->u.nary->op[0]))
3246 && (TYPE_PRECISION (type)
3247 >= TYPE_PRECISION (TREE_TYPE (expr->u.nary->op[0])))
3248 && SSA_NAME_RANGE_INFO (expr->u.nary->op[0]))
3250 value_range r;
3251 if (get_range_query (cfun)->range_of_expr (r, expr->u.nary->op[0])
3252 && !r.undefined_p ()
3253 && !r.varying_p ()
3254 && !wi::neg_p (r.lower_bound (), SIGNED)
3255 && !wi::neg_p (r.upper_bound (), SIGNED))
3257 /* Just handle extension and sign-changes of all-positive ranges. */
3258 range_cast (r, type);
3259 set_range_info (temp, r);
3263 if (dump_file && (dump_flags & TDF_DETAILS))
3265 fprintf (dump_file, "Created phi ");
3266 print_gimple_stmt (dump_file, phi, 0);
3267 fprintf (dump_file, " in block %d (%04d)\n", block->index, val);
3269 pre_stats.phis++;
3270 return true;
3275 /* Perform insertion of partially redundant or hoistable values.
3276 For BLOCK, do the following:
3277 1. Propagate the NEW_SETS of the dominator into the current block.
3278 If the block has multiple predecessors,
3279 2a. Iterate over the ANTIC expressions for the block to see if
3280 any of them are partially redundant.
3281 2b. If so, insert them into the necessary predecessors to make
3282 the expression fully redundant.
3283 2c. Insert a new PHI merging the values of the predecessors.
3284 2d. Insert the new PHI, and the new expressions, into the
3285 NEW_SETS set.
3286 If the block has multiple successors,
3287 3a. Iterate over the ANTIC values for the block to see if
3288 any of them are good candidates for hoisting.
3289 3b. If so, insert expressions computing the values in BLOCK,
3290 and add the new expressions into the NEW_SETS set.
3291 4. Recursively call ourselves on the dominator children of BLOCK.
3293 Steps 1, 2a, and 4 are done by insert_aux. 2b, 2c and 2d are done by
3294 do_pre_regular_insertion and do_partial_insertion. 3a and 3b are
3295 done in do_hoist_insertion.
3298 static bool
3299 do_pre_regular_insertion (basic_block block, basic_block dom,
3300 vec<pre_expr> exprs)
3302 bool new_stuff = false;
3303 pre_expr expr;
3304 auto_vec<pre_expr, 2> avail;
3305 int i;
3307 avail.safe_grow (EDGE_COUNT (block->preds), true);
3309 FOR_EACH_VEC_ELT (exprs, i, expr)
3311 if (expr->kind == NARY
3312 || expr->kind == REFERENCE)
3314 unsigned int val;
3315 bool by_some = false;
3316 bool cant_insert = false;
3317 bool all_same = true;
3318 unsigned num_inserts = 0;
3319 unsigned num_const = 0;
3320 pre_expr first_s = NULL;
3321 edge pred;
3322 basic_block bprime;
3323 pre_expr eprime = NULL;
3324 edge_iterator ei;
3325 pre_expr edoubleprime = NULL;
3326 bool do_insertion = false;
3328 val = get_expr_value_id (expr);
3329 if (bitmap_set_contains_value (PHI_GEN (block), val))
3330 continue;
3331 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3333 if (dump_file && (dump_flags & TDF_DETAILS))
3335 fprintf (dump_file, "Found fully redundant value: ");
3336 print_pre_expr (dump_file, expr);
3337 fprintf (dump_file, "\n");
3339 continue;
3342 FOR_EACH_EDGE (pred, ei, block->preds)
3344 unsigned int vprime;
3346 /* We should never run insertion for the exit block
3347 and so not come across fake pred edges. */
3348 gcc_assert (!(pred->flags & EDGE_FAKE));
3349 bprime = pred->src;
3350 /* We are looking at ANTIC_OUT of bprime. */
3351 eprime = phi_translate (NULL, expr, ANTIC_IN (block), NULL, pred);
3353 /* eprime will generally only be NULL if the
3354 value of the expression, translated
3355 through the PHI for this predecessor, is
3356 undefined. If that is the case, we can't
3357 make the expression fully redundant,
3358 because its value is undefined along a
3359 predecessor path. We can thus break out
3360 early because it doesn't matter what the
3361 rest of the results are. */
3362 if (eprime == NULL)
3364 avail[pred->dest_idx] = NULL;
3365 cant_insert = true;
3366 break;
3369 vprime = get_expr_value_id (eprime);
3370 edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime),
3371 vprime);
3372 if (edoubleprime == NULL)
3374 avail[pred->dest_idx] = eprime;
3375 all_same = false;
3376 num_inserts++;
3378 else
3380 avail[pred->dest_idx] = edoubleprime;
3381 by_some = true;
3382 if (edoubleprime->kind == CONSTANT)
3383 num_const++;
3384 /* We want to perform insertions to remove a redundancy on
3385 a path in the CFG we want to optimize for speed. */
3386 if (optimize_edge_for_speed_p (pred))
3387 do_insertion = true;
3388 if (first_s == NULL)
3389 first_s = edoubleprime;
3390 else if (!pre_expr_d::equal (first_s, edoubleprime))
3391 all_same = false;
3394 /* If we can insert it, it's not the same value
3395 already existing along every predecessor, and
3396 it's defined by some predecessor, it is
3397 partially redundant. */
3398 if (!cant_insert && !all_same && by_some)
3400 /* If the expression is redundant on all edges and we need
3401 to at most insert one copy from a constant do the PHI
3402 insertion even when not optimizing a path that's to be
3403 optimized for speed. */
3404 if (num_inserts == 0 && num_const <= 1)
3405 do_insertion = true;
3406 if (!do_insertion)
3408 if (dump_file && (dump_flags & TDF_DETAILS))
3410 fprintf (dump_file, "Skipping partial redundancy for "
3411 "expression ");
3412 print_pre_expr (dump_file, expr);
3413 fprintf (dump_file, " (%04d), no redundancy on to be "
3414 "optimized for speed edge\n", val);
3417 else if (dbg_cnt (treepre_insert))
3419 if (dump_file && (dump_flags & TDF_DETAILS))
3421 fprintf (dump_file, "Found partial redundancy for "
3422 "expression ");
3423 print_pre_expr (dump_file, expr);
3424 fprintf (dump_file, " (%04d)\n",
3425 get_expr_value_id (expr));
3427 if (insert_into_preds_of_block (block,
3428 get_expression_id (expr),
3429 avail))
3430 new_stuff = true;
3433 /* If all edges produce the same value and that value is
3434 an invariant, then the PHI has the same value on all
3435 edges. Note this. */
3436 else if (!cant_insert
3437 && all_same
3438 && (edoubleprime->kind != NAME
3439 || !SSA_NAME_OCCURS_IN_ABNORMAL_PHI
3440 (PRE_EXPR_NAME (edoubleprime))))
3442 gcc_assert (edoubleprime->kind == CONSTANT
3443 || edoubleprime->kind == NAME);
3445 tree temp = make_temp_ssa_name (get_expr_type (expr),
3446 NULL, "pretmp");
3447 gassign *assign
3448 = gimple_build_assign (temp,
3449 edoubleprime->kind == CONSTANT ?
3450 PRE_EXPR_CONSTANT (edoubleprime) :
3451 PRE_EXPR_NAME (edoubleprime));
3452 gimple_stmt_iterator gsi = gsi_after_labels (block);
3453 gsi_insert_before (&gsi, assign, GSI_NEW_STMT);
3455 vn_ssa_aux_t vn_info = VN_INFO (temp);
3456 vn_info->value_id = val;
3457 vn_info->valnum = vn_valnum_from_value_id (val);
3458 if (vn_info->valnum == NULL_TREE)
3459 vn_info->valnum = temp;
3460 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3461 pre_expr newe = get_or_alloc_expr_for_name (temp);
3462 add_to_value (val, newe);
3463 bitmap_value_replace_in_set (AVAIL_OUT (block), newe);
3464 bitmap_insert_into_set (NEW_SETS (block), newe);
3465 bitmap_insert_into_set (PHI_GEN (block), newe);
3470 return new_stuff;
3474 /* Perform insertion for partially anticipatable expressions. There
3475 is only one case we will perform insertion for these. This case is
3476 if the expression is partially anticipatable, and fully available.
3477 In this case, we know that putting it earlier will enable us to
3478 remove the later computation. */
3480 static bool
3481 do_pre_partial_partial_insertion (basic_block block, basic_block dom,
3482 vec<pre_expr> exprs)
3484 bool new_stuff = false;
3485 pre_expr expr;
3486 auto_vec<pre_expr, 2> avail;
3487 int i;
3489 avail.safe_grow (EDGE_COUNT (block->preds), true);
3491 FOR_EACH_VEC_ELT (exprs, i, expr)
3493 if (expr->kind == NARY
3494 || expr->kind == REFERENCE)
3496 unsigned int val;
3497 bool by_all = true;
3498 bool cant_insert = false;
3499 edge pred;
3500 basic_block bprime;
3501 pre_expr eprime = NULL;
3502 edge_iterator ei;
3504 val = get_expr_value_id (expr);
3505 if (bitmap_set_contains_value (PHI_GEN (block), val))
3506 continue;
3507 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3508 continue;
3510 FOR_EACH_EDGE (pred, ei, block->preds)
3512 unsigned int vprime;
3513 pre_expr edoubleprime;
3515 /* We should never run insertion for the exit block
3516 and so not come across fake pred edges. */
3517 gcc_assert (!(pred->flags & EDGE_FAKE));
3518 bprime = pred->src;
3519 eprime = phi_translate (NULL, expr, ANTIC_IN (block),
3520 PA_IN (block), pred);
3522 /* eprime will generally only be NULL if the
3523 value of the expression, translated
3524 through the PHI for this predecessor, is
3525 undefined. If that is the case, we can't
3526 make the expression fully redundant,
3527 because its value is undefined along a
3528 predecessor path. We can thus break out
3529 early because it doesn't matter what the
3530 rest of the results are. */
3531 if (eprime == NULL)
3533 avail[pred->dest_idx] = NULL;
3534 cant_insert = true;
3535 break;
3538 vprime = get_expr_value_id (eprime);
3539 edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime), vprime);
3540 avail[pred->dest_idx] = edoubleprime;
3541 if (edoubleprime == NULL)
3543 by_all = false;
3544 break;
3548 /* If we can insert it, it's not the same value
3549 already existing along every predecessor, and
3550 it's defined by some predecessor, it is
3551 partially redundant. */
3552 if (!cant_insert && by_all)
3554 edge succ;
3555 bool do_insertion = false;
3557 /* Insert only if we can remove a later expression on a path
3558 that we want to optimize for speed.
3559 The phi node that we will be inserting in BLOCK is not free,
3560 and inserting it for the sake of !optimize_for_speed successor
3561 may cause regressions on the speed path. */
3562 FOR_EACH_EDGE (succ, ei, block->succs)
3564 if (bitmap_set_contains_value (PA_IN (succ->dest), val)
3565 || bitmap_set_contains_value (ANTIC_IN (succ->dest), val))
3567 if (optimize_edge_for_speed_p (succ))
3568 do_insertion = true;
3572 if (!do_insertion)
3574 if (dump_file && (dump_flags & TDF_DETAILS))
3576 fprintf (dump_file, "Skipping partial partial redundancy "
3577 "for expression ");
3578 print_pre_expr (dump_file, expr);
3579 fprintf (dump_file, " (%04d), not (partially) anticipated "
3580 "on any to be optimized for speed edges\n", val);
3583 else if (dbg_cnt (treepre_insert))
3585 pre_stats.pa_insert++;
3586 if (dump_file && (dump_flags & TDF_DETAILS))
3588 fprintf (dump_file, "Found partial partial redundancy "
3589 "for expression ");
3590 print_pre_expr (dump_file, expr);
3591 fprintf (dump_file, " (%04d)\n",
3592 get_expr_value_id (expr));
3594 if (insert_into_preds_of_block (block,
3595 get_expression_id (expr),
3596 avail))
3597 new_stuff = true;
3603 return new_stuff;
3606 /* Insert expressions in BLOCK to compute hoistable values up.
3607 Return TRUE if something was inserted, otherwise return FALSE.
3608 The caller has to make sure that BLOCK has at least two successors. */
3610 static bool
3611 do_hoist_insertion (basic_block block)
3613 edge e;
3614 edge_iterator ei;
3615 bool new_stuff = false;
3616 unsigned i;
3617 gimple_stmt_iterator last;
3619 /* At least two successors, or else... */
3620 gcc_assert (EDGE_COUNT (block->succs) >= 2);
3622 /* Check that all successors of BLOCK are dominated by block.
3623 We could use dominated_by_p() for this, but actually there is a much
3624 quicker check: any successor that is dominated by BLOCK can't have
3625 more than one predecessor edge. */
3626 FOR_EACH_EDGE (e, ei, block->succs)
3627 if (! single_pred_p (e->dest))
3628 return false;
3630 /* Determine the insertion point. If we cannot safely insert before
3631 the last stmt if we'd have to, bail out. */
3632 last = gsi_last_bb (block);
3633 if (!gsi_end_p (last)
3634 && !is_ctrl_stmt (gsi_stmt (last))
3635 && stmt_ends_bb_p (gsi_stmt (last)))
3636 return false;
3638 /* We have multiple successors, compute ANTIC_OUT by taking the intersection
3639 of all of ANTIC_IN translating through PHI nodes. Track the union
3640 of the expression sets so we can pick a representative that is
3641 fully generatable out of hoistable expressions. */
3642 bitmap_set_t ANTIC_OUT = bitmap_set_new ();
3643 bool first = true;
3644 FOR_EACH_EDGE (e, ei, block->succs)
3646 if (first)
3648 phi_translate_set (ANTIC_OUT, ANTIC_IN (e->dest), e);
3649 first = false;
3651 else if (!gimple_seq_empty_p (phi_nodes (e->dest)))
3653 bitmap_set_t tmp = bitmap_set_new ();
3654 phi_translate_set (tmp, ANTIC_IN (e->dest), e);
3655 bitmap_and_into (&ANTIC_OUT->values, &tmp->values);
3656 bitmap_ior_into (&ANTIC_OUT->expressions, &tmp->expressions);
3657 bitmap_set_free (tmp);
3659 else
3661 bitmap_and_into (&ANTIC_OUT->values, &ANTIC_IN (e->dest)->values);
3662 bitmap_ior_into (&ANTIC_OUT->expressions,
3663 &ANTIC_IN (e->dest)->expressions);
3667 /* Compute the set of hoistable expressions from ANTIC_OUT. First compute
3668 hoistable values. */
3669 bitmap_set hoistable_set;
3671 /* A hoistable value must be in ANTIC_OUT(block)
3672 but not in AVAIL_OUT(BLOCK). */
3673 bitmap_initialize (&hoistable_set.values, &grand_bitmap_obstack);
3674 bitmap_and_compl (&hoistable_set.values,
3675 &ANTIC_OUT->values, &AVAIL_OUT (block)->values);
3677 /* Short-cut for a common case: hoistable_set is empty. */
3678 if (bitmap_empty_p (&hoistable_set.values))
3680 bitmap_set_free (ANTIC_OUT);
3681 return false;
3684 /* Compute which of the hoistable values is in AVAIL_OUT of
3685 at least one of the successors of BLOCK. */
3686 bitmap_head availout_in_some;
3687 bitmap_initialize (&availout_in_some, &grand_bitmap_obstack);
3688 FOR_EACH_EDGE (e, ei, block->succs)
3689 /* Do not consider expressions solely because their availability
3690 on loop exits. They'd be ANTIC-IN throughout the whole loop
3691 and thus effectively hoisted across loops by combination of
3692 PRE and hoisting. */
3693 if (! loop_exit_edge_p (block->loop_father, e))
3694 bitmap_ior_and_into (&availout_in_some, &hoistable_set.values,
3695 &AVAIL_OUT (e->dest)->values);
3696 bitmap_clear (&hoistable_set.values);
3698 /* Short-cut for a common case: availout_in_some is empty. */
3699 if (bitmap_empty_p (&availout_in_some))
3701 bitmap_set_free (ANTIC_OUT);
3702 return false;
3705 /* Hack hoistable_set in-place so we can use sorted_array_from_bitmap_set. */
3706 bitmap_move (&hoistable_set.values, &availout_in_some);
3707 hoistable_set.expressions = ANTIC_OUT->expressions;
3709 /* Now finally construct the topological-ordered expression set. */
3710 vec<pre_expr> exprs = sorted_array_from_bitmap_set (&hoistable_set);
3712 /* If there are candidate values for hoisting, insert expressions
3713 strategically to make the hoistable expressions fully redundant. */
3714 pre_expr expr;
3715 FOR_EACH_VEC_ELT (exprs, i, expr)
3717 /* While we try to sort expressions topologically above the
3718 sorting doesn't work out perfectly. Catch expressions we
3719 already inserted. */
3720 unsigned int value_id = get_expr_value_id (expr);
3721 if (bitmap_set_contains_value (AVAIL_OUT (block), value_id))
3723 if (dump_file && (dump_flags & TDF_DETAILS))
3725 fprintf (dump_file,
3726 "Already inserted expression for ");
3727 print_pre_expr (dump_file, expr);
3728 fprintf (dump_file, " (%04d)\n", value_id);
3730 continue;
3733 /* If we end up with a punned expression representation and this
3734 happens to be a float typed one give up - we can't know for
3735 sure whether all paths perform the floating-point load we are
3736 about to insert and on some targets this can cause correctness
3737 issues. See PR88240. */
3738 if (expr->kind == REFERENCE
3739 && PRE_EXPR_REFERENCE (expr)->punned
3740 && FLOAT_TYPE_P (get_expr_type (expr)))
3741 continue;
3743 /* Only hoist if the full expression is available for hoisting.
3744 This avoids hoisting values that are not common and for
3745 example evaluate an expression that's not valid to evaluate
3746 unconditionally (PR112310). */
3747 if (!valid_in_sets (&hoistable_set, AVAIL_OUT (block), expr))
3748 continue;
3750 /* OK, we should hoist this value. Perform the transformation. */
3751 pre_stats.hoist_insert++;
3752 if (dump_file && (dump_flags & TDF_DETAILS))
3754 fprintf (dump_file,
3755 "Inserting expression in block %d for code hoisting: ",
3756 block->index);
3757 print_pre_expr (dump_file, expr);
3758 fprintf (dump_file, " (%04d)\n", value_id);
3761 gimple_seq stmts = NULL;
3762 tree res = create_expression_by_pieces (block, expr, &stmts,
3763 get_expr_type (expr));
3765 /* Do not return true if expression creation ultimately
3766 did not insert any statements. */
3767 if (gimple_seq_empty_p (stmts))
3768 res = NULL_TREE;
3769 else
3771 if (gsi_end_p (last) || is_ctrl_stmt (gsi_stmt (last)))
3772 gsi_insert_seq_before (&last, stmts, GSI_SAME_STMT);
3773 else
3774 gsi_insert_seq_after (&last, stmts, GSI_NEW_STMT);
3777 /* Make sure to not return true if expression creation ultimately
3778 failed but also make sure to insert any stmts produced as they
3779 are tracked in inserted_exprs. */
3780 if (! res)
3781 continue;
3783 new_stuff = true;
3786 exprs.release ();
3787 bitmap_clear (&hoistable_set.values);
3788 bitmap_set_free (ANTIC_OUT);
3790 return new_stuff;
3793 /* Perform insertion of partially redundant and hoistable values. */
3795 static void
3796 insert (void)
3798 basic_block bb;
3800 FOR_ALL_BB_FN (bb, cfun)
3801 NEW_SETS (bb) = bitmap_set_new ();
3803 int *rpo = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
3804 int *bb_rpo = XNEWVEC (int, last_basic_block_for_fn (cfun) + 1);
3805 int rpo_num = pre_and_rev_post_order_compute (NULL, rpo, false);
3806 for (int i = 0; i < rpo_num; ++i)
3807 bb_rpo[rpo[i]] = i;
3809 int num_iterations = 0;
3810 bool changed;
3813 num_iterations++;
3814 if (dump_file && dump_flags & TDF_DETAILS)
3815 fprintf (dump_file, "Starting insert iteration %d\n", num_iterations);
3817 changed = false;
3818 for (int idx = 0; idx < rpo_num; ++idx)
3820 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[idx]);
3821 basic_block dom = get_immediate_dominator (CDI_DOMINATORS, block);
3822 if (dom)
3824 unsigned i;
3825 bitmap_iterator bi;
3826 bitmap_set_t newset;
3828 /* First, update the AVAIL_OUT set with anything we may have
3829 inserted higher up in the dominator tree. */
3830 newset = NEW_SETS (dom);
3832 /* Note that we need to value_replace both NEW_SETS, and
3833 AVAIL_OUT. For both the case of NEW_SETS, the value may be
3834 represented by some non-simple expression here that we want
3835 to replace it with. */
3836 bool avail_out_changed = false;
3837 FOR_EACH_EXPR_ID_IN_SET (newset, i, bi)
3839 pre_expr expr = expression_for_id (i);
3840 bitmap_value_replace_in_set (NEW_SETS (block), expr);
3841 avail_out_changed
3842 |= bitmap_value_replace_in_set (AVAIL_OUT (block), expr);
3844 /* We need to iterate if AVAIL_OUT of an already processed
3845 block source changed. */
3846 if (avail_out_changed && !changed)
3848 edge_iterator ei;
3849 edge e;
3850 FOR_EACH_EDGE (e, ei, block->succs)
3851 if (e->dest->index != EXIT_BLOCK
3852 && bb_rpo[e->dest->index] < idx)
3853 changed = true;
3856 /* Insert expressions for partial redundancies. */
3857 if (flag_tree_pre && !single_pred_p (block))
3859 vec<pre_expr> exprs
3860 = sorted_array_from_bitmap_set (ANTIC_IN (block));
3861 /* Sorting is not perfect, iterate locally. */
3862 while (do_pre_regular_insertion (block, dom, exprs))
3864 exprs.release ();
3865 if (do_partial_partial)
3867 exprs = sorted_array_from_bitmap_set (PA_IN (block));
3868 while (do_pre_partial_partial_insertion (block, dom,
3869 exprs))
3871 exprs.release ();
3877 /* Clear the NEW sets before the next iteration. We have already
3878 fully propagated its contents. */
3879 if (changed)
3880 FOR_ALL_BB_FN (bb, cfun)
3881 bitmap_set_free (NEW_SETS (bb));
3883 while (changed);
3885 statistics_histogram_event (cfun, "insert iterations", num_iterations);
3887 /* AVAIL_OUT is not needed after insertion so we don't have to
3888 propagate NEW_SETS from hoist insertion. */
3889 FOR_ALL_BB_FN (bb, cfun)
3891 bitmap_set_free (NEW_SETS (bb));
3892 bitmap_set_pool.remove (NEW_SETS (bb));
3893 NEW_SETS (bb) = NULL;
3896 /* Insert expressions for hoisting. Do a backward walk here since
3897 inserting into BLOCK exposes new opportunities in its predecessors.
3898 Since PRE and hoist insertions can cause back-to-back iteration
3899 and we are interested in PRE insertion exposed hoisting opportunities
3900 but not in hoisting exposed PRE ones do hoist insertion only after
3901 PRE insertion iteration finished and do not iterate it. */
3902 if (flag_code_hoisting)
3903 for (int idx = rpo_num - 1; idx >= 0; --idx)
3905 basic_block block = BASIC_BLOCK_FOR_FN (cfun, rpo[idx]);
3906 if (EDGE_COUNT (block->succs) >= 2)
3907 changed |= do_hoist_insertion (block);
3910 free (rpo);
3911 free (bb_rpo);
3915 /* Compute the AVAIL set for all basic blocks.
3917 This function performs value numbering of the statements in each basic
3918 block. The AVAIL sets are built from information we glean while doing
3919 this value numbering, since the AVAIL sets contain only one entry per
3920 value.
3922 AVAIL_IN[BLOCK] = AVAIL_OUT[dom(BLOCK)].
3923 AVAIL_OUT[BLOCK] = AVAIL_IN[BLOCK] U PHI_GEN[BLOCK] U TMP_GEN[BLOCK]. */
3925 static void
3926 compute_avail (function *fun)
3929 basic_block block, son;
3930 basic_block *worklist;
3931 size_t sp = 0;
3932 unsigned i;
3933 tree name;
3935 /* We pretend that default definitions are defined in the entry block.
3936 This includes function arguments and the static chain decl. */
3937 FOR_EACH_SSA_NAME (i, name, fun)
3939 pre_expr e;
3940 if (!SSA_NAME_IS_DEFAULT_DEF (name)
3941 || has_zero_uses (name)
3942 || virtual_operand_p (name))
3943 continue;
3945 e = get_or_alloc_expr_for_name (name);
3946 add_to_value (get_expr_value_id (e), e);
3947 bitmap_insert_into_set (TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (fun)), e);
3948 bitmap_value_insert_into_set (AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3952 if (dump_file && (dump_flags & TDF_DETAILS))
3954 print_bitmap_set (dump_file, TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3955 "tmp_gen", ENTRY_BLOCK);
3956 print_bitmap_set (dump_file, AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (fun)),
3957 "avail_out", ENTRY_BLOCK);
3960 /* Allocate the worklist. */
3961 worklist = XNEWVEC (basic_block, n_basic_blocks_for_fn (fun));
3963 /* Seed the algorithm by putting the dominator children of the entry
3964 block on the worklist. */
3965 for (son = first_dom_son (CDI_DOMINATORS, ENTRY_BLOCK_PTR_FOR_FN (fun));
3966 son;
3967 son = next_dom_son (CDI_DOMINATORS, son))
3968 worklist[sp++] = son;
3970 BB_LIVE_VOP_ON_EXIT (ENTRY_BLOCK_PTR_FOR_FN (fun))
3971 = ssa_default_def (fun, gimple_vop (fun));
3973 /* Loop until the worklist is empty. */
3974 while (sp)
3976 gimple *stmt;
3977 basic_block dom;
3979 /* Pick a block from the worklist. */
3980 block = worklist[--sp];
3981 vn_context_bb = block;
3983 /* Initially, the set of available values in BLOCK is that of
3984 its immediate dominator. */
3985 dom = get_immediate_dominator (CDI_DOMINATORS, block);
3986 if (dom)
3988 bitmap_set_copy (AVAIL_OUT (block), AVAIL_OUT (dom));
3989 BB_LIVE_VOP_ON_EXIT (block) = BB_LIVE_VOP_ON_EXIT (dom);
3992 /* Generate values for PHI nodes. */
3993 for (gphi_iterator gsi = gsi_start_phis (block); !gsi_end_p (gsi);
3994 gsi_next (&gsi))
3996 tree result = gimple_phi_result (gsi.phi ());
3998 /* We have no need for virtual phis, as they don't represent
3999 actual computations. */
4000 if (virtual_operand_p (result))
4002 BB_LIVE_VOP_ON_EXIT (block) = result;
4003 continue;
4006 pre_expr e = get_or_alloc_expr_for_name (result);
4007 add_to_value (get_expr_value_id (e), e);
4008 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
4009 bitmap_insert_into_set (PHI_GEN (block), e);
4012 BB_MAY_NOTRETURN (block) = 0;
4014 /* Now compute value numbers and populate value sets with all
4015 the expressions computed in BLOCK. */
4016 bool set_bb_may_notreturn = false;
4017 for (gimple_stmt_iterator gsi = gsi_start_bb (block); !gsi_end_p (gsi);
4018 gsi_next (&gsi))
4020 ssa_op_iter iter;
4021 tree op;
4023 stmt = gsi_stmt (gsi);
4025 if (set_bb_may_notreturn)
4027 BB_MAY_NOTRETURN (block) = 1;
4028 set_bb_may_notreturn = false;
4031 /* Cache whether the basic-block has any non-visible side-effect
4032 or control flow.
4033 If this isn't a call or it is the last stmt in the
4034 basic-block then the CFG represents things correctly. */
4035 if (is_gimple_call (stmt) && !stmt_ends_bb_p (stmt))
4037 /* Non-looping const functions always return normally.
4038 Otherwise the call might not return or have side-effects
4039 that forbids hoisting possibly trapping expressions
4040 before it. */
4041 int flags = gimple_call_flags (stmt);
4042 if (!(flags & (ECF_CONST|ECF_PURE))
4043 || (flags & ECF_LOOPING_CONST_OR_PURE)
4044 || stmt_can_throw_external (fun, stmt))
4045 /* Defer setting of BB_MAY_NOTRETURN to avoid it
4046 influencing the processing of the call itself. */
4047 set_bb_may_notreturn = true;
4050 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_DEF)
4052 pre_expr e = get_or_alloc_expr_for_name (op);
4053 add_to_value (get_expr_value_id (e), e);
4054 bitmap_insert_into_set (TMP_GEN (block), e);
4055 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
4058 if (gimple_vdef (stmt))
4059 BB_LIVE_VOP_ON_EXIT (block) = gimple_vdef (stmt);
4061 if (gimple_has_side_effects (stmt)
4062 || stmt_could_throw_p (fun, stmt)
4063 || is_gimple_debug (stmt))
4064 continue;
4066 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_USE)
4068 if (ssa_undefined_value_p (op))
4069 continue;
4070 pre_expr e = get_or_alloc_expr_for_name (op);
4071 bitmap_value_insert_into_set (EXP_GEN (block), e);
4074 switch (gimple_code (stmt))
4076 case GIMPLE_RETURN:
4077 continue;
4079 case GIMPLE_CALL:
4081 vn_reference_t ref;
4082 vn_reference_s ref1;
4083 pre_expr result = NULL;
4085 vn_reference_lookup_call (as_a <gcall *> (stmt), &ref, &ref1);
4086 /* There is no point to PRE a call without a value. */
4087 if (!ref || !ref->result)
4088 continue;
4090 /* If the value of the call is not invalidated in
4091 this block until it is computed, add the expression
4092 to EXP_GEN. */
4093 if ((!gimple_vuse (stmt)
4094 || gimple_code
4095 (SSA_NAME_DEF_STMT (gimple_vuse (stmt))) == GIMPLE_PHI
4096 || gimple_bb (SSA_NAME_DEF_STMT
4097 (gimple_vuse (stmt))) != block)
4098 /* If the REFERENCE traps and there was a preceding
4099 point in the block that might not return avoid
4100 adding the reference to EXP_GEN. */
4101 && (!BB_MAY_NOTRETURN (block)
4102 || !vn_reference_may_trap (ref)))
4104 result = get_or_alloc_expr_for_reference
4105 (ref, gimple_location (stmt));
4106 add_to_value (get_expr_value_id (result), result);
4107 bitmap_value_insert_into_set (EXP_GEN (block), result);
4109 continue;
4112 case GIMPLE_ASSIGN:
4114 pre_expr result = NULL;
4115 switch (vn_get_stmt_kind (stmt))
4117 case VN_NARY:
4119 enum tree_code code = gimple_assign_rhs_code (stmt);
4120 vn_nary_op_t nary;
4122 /* COND_EXPR is awkward in that it contains an
4123 embedded complex expression.
4124 Don't even try to shove it through PRE. */
4125 if (code == COND_EXPR)
4126 continue;
4128 vn_nary_op_lookup_stmt (stmt, &nary);
4129 if (!nary || nary->predicated_values)
4130 continue;
4132 unsigned value_id = nary->value_id;
4133 if (value_id_constant_p (value_id))
4134 continue;
4136 /* Record the un-valueized expression for EXP_GEN. */
4137 nary = XALLOCAVAR (struct vn_nary_op_s,
4138 sizeof_vn_nary_op
4139 (vn_nary_length_from_stmt (stmt)));
4140 init_vn_nary_op_from_stmt (nary, as_a <gassign *> (stmt));
4142 /* If the NARY traps and there was a preceding
4143 point in the block that might not return avoid
4144 adding the nary to EXP_GEN. */
4145 if (BB_MAY_NOTRETURN (block)
4146 && vn_nary_may_trap (nary))
4147 continue;
4149 result = get_or_alloc_expr_for_nary
4150 (nary, value_id, gimple_location (stmt));
4151 break;
4154 case VN_REFERENCE:
4156 tree rhs1 = gimple_assign_rhs1 (stmt);
4157 ao_ref rhs1_ref;
4158 ao_ref_init (&rhs1_ref, rhs1);
4159 alias_set_type set = ao_ref_alias_set (&rhs1_ref);
4160 alias_set_type base_set
4161 = ao_ref_base_alias_set (&rhs1_ref);
4162 vec<vn_reference_op_s> operands
4163 = vn_reference_operands_for_lookup (rhs1);
4164 vn_reference_t ref;
4165 vn_reference_lookup_pieces (gimple_vuse (stmt), set,
4166 base_set, TREE_TYPE (rhs1),
4167 operands, &ref, VN_WALK);
4168 if (!ref)
4170 operands.release ();
4171 continue;
4174 /* If the REFERENCE traps and there was a preceding
4175 point in the block that might not return avoid
4176 adding the reference to EXP_GEN. */
4177 if (BB_MAY_NOTRETURN (block)
4178 && vn_reference_may_trap (ref))
4180 operands.release ();
4181 continue;
4184 /* If the value of the reference is not invalidated in
4185 this block until it is computed, add the expression
4186 to EXP_GEN. */
4187 if (gimple_vuse (stmt))
4189 gimple *def_stmt;
4190 bool ok = true;
4191 def_stmt = SSA_NAME_DEF_STMT (gimple_vuse (stmt));
4192 while (!gimple_nop_p (def_stmt)
4193 && gimple_code (def_stmt) != GIMPLE_PHI
4194 && gimple_bb (def_stmt) == block)
4196 if (stmt_may_clobber_ref_p
4197 (def_stmt, gimple_assign_rhs1 (stmt)))
4199 ok = false;
4200 break;
4202 def_stmt
4203 = SSA_NAME_DEF_STMT (gimple_vuse (def_stmt));
4205 if (!ok)
4207 operands.release ();
4208 continue;
4212 /* If the load was value-numbered to another
4213 load make sure we do not use its expression
4214 for insertion if it wouldn't be a valid
4215 replacement. */
4216 /* At the momemt we have a testcase
4217 for hoist insertion of aligned vs. misaligned
4218 variants in gcc.dg/torture/pr65270-1.c thus
4219 with just alignment to be considered we can
4220 simply replace the expression in the hashtable
4221 with the most conservative one. */
4222 vn_reference_op_t ref1 = &ref->operands.last ();
4223 while (ref1->opcode != TARGET_MEM_REF
4224 && ref1->opcode != MEM_REF
4225 && ref1 != &ref->operands[0])
4226 --ref1;
4227 vn_reference_op_t ref2 = &operands.last ();
4228 while (ref2->opcode != TARGET_MEM_REF
4229 && ref2->opcode != MEM_REF
4230 && ref2 != &operands[0])
4231 --ref2;
4232 if ((ref1->opcode == TARGET_MEM_REF
4233 || ref1->opcode == MEM_REF)
4234 && (TYPE_ALIGN (ref1->type)
4235 > TYPE_ALIGN (ref2->type)))
4236 ref1->type
4237 = build_aligned_type (ref1->type,
4238 TYPE_ALIGN (ref2->type));
4239 /* TBAA behavior is an obvious part so make sure
4240 that the hashtable one covers this as well
4241 by adjusting the ref alias set and its base. */
4242 if ((ref->set == set
4243 || alias_set_subset_of (set, ref->set))
4244 && (ref->base_set == base_set
4245 || alias_set_subset_of (base_set, ref->base_set)))
4247 else if (ref1->opcode != ref2->opcode
4248 || (ref1->opcode != MEM_REF
4249 && ref1->opcode != TARGET_MEM_REF))
4251 /* With mismatching base opcodes or bases
4252 other than MEM_REF or TARGET_MEM_REF we
4253 can't do any easy TBAA adjustment. */
4254 operands.release ();
4255 continue;
4257 else if (ref->set == set
4258 || alias_set_subset_of (ref->set, set))
4260 tree reft = reference_alias_ptr_type (rhs1);
4261 ref->set = set;
4262 ref->base_set = set;
4263 if (ref1->opcode == MEM_REF)
4264 ref1->op0
4265 = wide_int_to_tree (reft,
4266 wi::to_wide (ref1->op0));
4267 else
4268 ref1->op2
4269 = wide_int_to_tree (reft,
4270 wi::to_wide (ref1->op2));
4272 else
4274 ref->set = 0;
4275 ref->base_set = 0;
4276 if (ref1->opcode == MEM_REF)
4277 ref1->op0
4278 = wide_int_to_tree (ptr_type_node,
4279 wi::to_wide (ref1->op0));
4280 else
4281 ref1->op2
4282 = wide_int_to_tree (ptr_type_node,
4283 wi::to_wide (ref1->op2));
4285 /* We also need to make sure that the access path
4286 ends in an access of the same size as otherwise
4287 we might assume an access may not trap while in
4288 fact it might. That's independent of whether
4289 TBAA is in effect. */
4290 if (TYPE_SIZE (ref1->type) != TYPE_SIZE (ref2->type)
4291 && (! TYPE_SIZE (ref1->type)
4292 || ! TYPE_SIZE (ref2->type)
4293 || ! operand_equal_p (TYPE_SIZE (ref1->type),
4294 TYPE_SIZE (ref2->type))))
4296 operands.release ();
4297 continue;
4299 operands.release ();
4301 result = get_or_alloc_expr_for_reference
4302 (ref, gimple_location (stmt));
4303 break;
4306 default:
4307 continue;
4310 add_to_value (get_expr_value_id (result), result);
4311 bitmap_value_insert_into_set (EXP_GEN (block), result);
4312 continue;
4314 default:
4315 break;
4318 if (set_bb_may_notreturn)
4320 BB_MAY_NOTRETURN (block) = 1;
4321 set_bb_may_notreturn = false;
4324 if (dump_file && (dump_flags & TDF_DETAILS))
4326 print_bitmap_set (dump_file, EXP_GEN (block),
4327 "exp_gen", block->index);
4328 print_bitmap_set (dump_file, PHI_GEN (block),
4329 "phi_gen", block->index);
4330 print_bitmap_set (dump_file, TMP_GEN (block),
4331 "tmp_gen", block->index);
4332 print_bitmap_set (dump_file, AVAIL_OUT (block),
4333 "avail_out", block->index);
4336 /* Put the dominator children of BLOCK on the worklist of blocks
4337 to compute available sets for. */
4338 for (son = first_dom_son (CDI_DOMINATORS, block);
4339 son;
4340 son = next_dom_son (CDI_DOMINATORS, son))
4341 worklist[sp++] = son;
4343 vn_context_bb = NULL;
4345 free (worklist);
4349 /* Initialize data structures used by PRE. */
4351 static void
4352 init_pre (void)
4354 basic_block bb;
4356 next_expression_id = 1;
4357 expressions.create (0);
4358 expressions.safe_push (NULL);
4359 value_expressions.create (get_max_value_id () + 1);
4360 value_expressions.quick_grow_cleared (get_max_value_id () + 1);
4361 constant_value_expressions.create (get_max_constant_value_id () + 1);
4362 constant_value_expressions.quick_grow_cleared (get_max_constant_value_id () + 1);
4363 name_to_id.create (0);
4364 gcc_obstack_init (&pre_expr_obstack);
4366 inserted_exprs = BITMAP_ALLOC (NULL);
4368 connect_infinite_loops_to_exit ();
4369 memset (&pre_stats, 0, sizeof (pre_stats));
4371 alloc_aux_for_blocks (sizeof (struct bb_bitmap_sets));
4373 calculate_dominance_info (CDI_DOMINATORS);
4375 bitmap_obstack_initialize (&grand_bitmap_obstack);
4376 expression_to_id = new hash_table<pre_expr_d> (num_ssa_names * 3);
4377 FOR_ALL_BB_FN (bb, cfun)
4379 EXP_GEN (bb) = bitmap_set_new ();
4380 PHI_GEN (bb) = bitmap_set_new ();
4381 TMP_GEN (bb) = bitmap_set_new ();
4382 AVAIL_OUT (bb) = bitmap_set_new ();
4383 PHI_TRANS_TABLE (bb) = NULL;
4388 /* Deallocate data structures used by PRE. */
4390 static void
4391 fini_pre ()
4393 value_expressions.release ();
4394 constant_value_expressions.release ();
4395 expressions.release ();
4396 bitmap_obstack_release (&grand_bitmap_obstack);
4397 bitmap_set_pool.release ();
4398 pre_expr_pool.release ();
4399 delete expression_to_id;
4400 expression_to_id = NULL;
4401 name_to_id.release ();
4402 obstack_free (&pre_expr_obstack, NULL);
4404 basic_block bb;
4405 FOR_ALL_BB_FN (bb, cfun)
4406 if (bb->aux && PHI_TRANS_TABLE (bb))
4407 delete PHI_TRANS_TABLE (bb);
4408 free_aux_for_blocks ();
4411 namespace {
4413 const pass_data pass_data_pre =
4415 GIMPLE_PASS, /* type */
4416 "pre", /* name */
4417 OPTGROUP_NONE, /* optinfo_flags */
4418 TV_TREE_PRE, /* tv_id */
4419 ( PROP_cfg | PROP_ssa ), /* properties_required */
4420 0, /* properties_provided */
4421 0, /* properties_destroyed */
4422 TODO_rebuild_alias, /* todo_flags_start */
4423 0, /* todo_flags_finish */
4426 class pass_pre : public gimple_opt_pass
4428 public:
4429 pass_pre (gcc::context *ctxt)
4430 : gimple_opt_pass (pass_data_pre, ctxt)
4433 /* opt_pass methods: */
4434 bool gate (function *) final override
4435 { return flag_tree_pre != 0 || flag_code_hoisting != 0; }
4436 unsigned int execute (function *) final override;
4438 }; // class pass_pre
4440 /* Valueization hook for RPO VN when we are calling back to it
4441 at ANTIC compute time. */
4443 static tree
4444 pre_valueize (tree name)
4446 if (TREE_CODE (name) == SSA_NAME)
4448 tree tem = VN_INFO (name)->valnum;
4449 if (tem != VN_TOP && tem != name)
4451 if (TREE_CODE (tem) != SSA_NAME
4452 || SSA_NAME_IS_DEFAULT_DEF (tem))
4453 return tem;
4454 /* We create temporary SSA names for representatives that
4455 do not have a definition (yet) but are not default defs either
4456 assume they are fine to use. */
4457 basic_block def_bb = gimple_bb (SSA_NAME_DEF_STMT (tem));
4458 if (! def_bb
4459 || dominated_by_p (CDI_DOMINATORS, vn_context_bb, def_bb))
4460 return tem;
4461 /* ??? Now we could look for a leader. Ideally we'd somehow
4462 expose RPO VN leaders and get rid of AVAIL_OUT as well... */
4465 return name;
4468 unsigned int
4469 pass_pre::execute (function *fun)
4471 unsigned int todo = 0;
4473 do_partial_partial =
4474 flag_tree_partial_pre && optimize_function_for_speed_p (fun);
4476 /* This has to happen before VN runs because
4477 loop_optimizer_init may create new phis, etc. */
4478 loop_optimizer_init (LOOPS_NORMAL);
4479 split_edges_for_insertion ();
4480 scev_initialize ();
4481 calculate_dominance_info (CDI_DOMINATORS);
4483 run_rpo_vn (VN_WALK);
4485 init_pre ();
4487 vn_valueize = pre_valueize;
4489 /* Insert can get quite slow on an incredibly large number of basic
4490 blocks due to some quadratic behavior. Until this behavior is
4491 fixed, don't run it when he have an incredibly large number of
4492 bb's. If we aren't going to run insert, there is no point in
4493 computing ANTIC, either, even though it's plenty fast nor do
4494 we require AVAIL. */
4495 if (n_basic_blocks_for_fn (fun) < 4000)
4497 compute_avail (fun);
4498 compute_antic ();
4499 insert ();
4502 /* Make sure to remove fake edges before committing our inserts.
4503 This makes sure we don't end up with extra critical edges that
4504 we would need to split. */
4505 remove_fake_exit_edges ();
4506 gsi_commit_edge_inserts ();
4508 /* Eliminate folds statements which might (should not...) end up
4509 not keeping virtual operands up-to-date. */
4510 gcc_assert (!need_ssa_update_p (fun));
4512 statistics_counter_event (fun, "Insertions", pre_stats.insertions);
4513 statistics_counter_event (fun, "PA inserted", pre_stats.pa_insert);
4514 statistics_counter_event (fun, "HOIST inserted", pre_stats.hoist_insert);
4515 statistics_counter_event (fun, "New PHIs", pre_stats.phis);
4517 todo |= eliminate_with_rpo_vn (inserted_exprs);
4519 vn_valueize = NULL;
4521 fini_pre ();
4523 scev_finalize ();
4524 loop_optimizer_finalize ();
4526 /* Perform a CFG cleanup before we run simple_dce_from_worklist since
4527 unreachable code regions will have not up-to-date SSA form which
4528 confuses it. */
4529 bool need_crit_edge_split = false;
4530 if (todo & TODO_cleanup_cfg)
4532 cleanup_tree_cfg ();
4533 need_crit_edge_split = true;
4536 /* Because we don't follow exactly the standard PRE algorithm, and decide not
4537 to insert PHI nodes sometimes, and because value numbering of casts isn't
4538 perfect, we sometimes end up inserting dead code. This simple DCE-like
4539 pass removes any insertions we made that weren't actually used. */
4540 simple_dce_from_worklist (inserted_exprs);
4541 BITMAP_FREE (inserted_exprs);
4543 /* TODO: tail_merge_optimize may merge all predecessors of a block, in which
4544 case we can merge the block with the remaining predecessor of the block.
4545 It should either:
4546 - call merge_blocks after each tail merge iteration
4547 - call merge_blocks after all tail merge iterations
4548 - mark TODO_cleanup_cfg when necessary. */
4549 todo |= tail_merge_optimize (need_crit_edge_split);
4551 free_rpo_vn ();
4553 /* Tail merging invalidates the virtual SSA web, together with
4554 cfg-cleanup opportunities exposed by PRE this will wreck the
4555 SSA updating machinery. So make sure to run update-ssa
4556 manually, before eventually scheduling cfg-cleanup as part of
4557 the todo. */
4558 update_ssa (TODO_update_ssa_only_virtuals);
4560 return todo;
4563 } // anon namespace
4565 gimple_opt_pass *
4566 make_pass_pre (gcc::context *ctxt)
4568 return new pass_pre (ctxt);