Mark as release
[official-gcc.git] / gcc / tree-ssa-pre.c
blob1e55356556fd4287913c9a2d2aba3c735b64dce9
1 /* SSA-PRE for trees.
2 Copyright (C) 2001-2014 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 "tm.h"
26 #include "tree.h"
27 #include "basic-block.h"
28 #include "gimple-pretty-print.h"
29 #include "tree-inline.h"
30 #include "hash-table.h"
31 #include "tree-ssa-alias.h"
32 #include "internal-fn.h"
33 #include "gimple-fold.h"
34 #include "tree-eh.h"
35 #include "gimple-expr.h"
36 #include "is-a.h"
37 #include "gimple.h"
38 #include "gimplify.h"
39 #include "gimple-iterator.h"
40 #include "gimplify-me.h"
41 #include "gimple-ssa.h"
42 #include "tree-cfg.h"
43 #include "tree-phinodes.h"
44 #include "ssa-iterators.h"
45 #include "stringpool.h"
46 #include "tree-ssanames.h"
47 #include "tree-ssa-loop.h"
48 #include "tree-into-ssa.h"
49 #include "expr.h"
50 #include "tree-dfa.h"
51 #include "tree-ssa.h"
52 #include "tree-iterator.h"
53 #include "alloc-pool.h"
54 #include "obstack.h"
55 #include "tree-pass.h"
56 #include "flags.h"
57 #include "langhooks.h"
58 #include "cfgloop.h"
59 #include "tree-ssa-sccvn.h"
60 #include "tree-scalar-evolution.h"
61 #include "params.h"
62 #include "dbgcnt.h"
63 #include "domwalk.h"
64 #include "ipa-prop.h"
65 #include "tree-ssa-propagate.h"
67 /* TODO:
69 1. Avail sets can be shared by making an avail_find_leader that
70 walks up the dominator tree and looks in those avail sets.
71 This might affect code optimality, it's unclear right now.
72 2. Strength reduction can be performed by anticipating expressions
73 we can repair later on.
74 3. We can do back-substitution or smarter value numbering to catch
75 commutative expressions split up over multiple statements.
78 /* For ease of terminology, "expression node" in the below refers to
79 every expression node but GIMPLE_ASSIGN, because GIMPLE_ASSIGNs
80 represent the actual statement containing the expressions we care about,
81 and we cache the value number by putting it in the expression. */
83 /* Basic algorithm
85 First we walk the statements to generate the AVAIL sets, the
86 EXP_GEN sets, and the tmp_gen sets. EXP_GEN sets represent the
87 generation of values/expressions by a given block. We use them
88 when computing the ANTIC sets. The AVAIL sets consist of
89 SSA_NAME's that represent values, so we know what values are
90 available in what blocks. AVAIL is a forward dataflow problem. In
91 SSA, values are never killed, so we don't need a kill set, or a
92 fixpoint iteration, in order to calculate the AVAIL sets. In
93 traditional parlance, AVAIL sets tell us the downsafety of the
94 expressions/values.
96 Next, we generate the ANTIC sets. These sets represent the
97 anticipatable expressions. ANTIC is a backwards dataflow
98 problem. An expression is anticipatable in a given block if it could
99 be generated in that block. This means that if we had to perform
100 an insertion in that block, of the value of that expression, we
101 could. Calculating the ANTIC sets requires phi translation of
102 expressions, because the flow goes backwards through phis. We must
103 iterate to a fixpoint of the ANTIC sets, because we have a kill
104 set. Even in SSA form, values are not live over the entire
105 function, only from their definition point onwards. So we have to
106 remove values from the ANTIC set once we go past the definition
107 point of the leaders that make them up.
108 compute_antic/compute_antic_aux performs this computation.
110 Third, we perform insertions to make partially redundant
111 expressions fully redundant.
113 An expression is partially redundant (excluding partial
114 anticipation) if:
116 1. It is AVAIL in some, but not all, of the predecessors of a
117 given block.
118 2. It is ANTIC in all the predecessors.
120 In order to make it fully redundant, we insert the expression into
121 the predecessors where it is not available, but is ANTIC.
123 For the partial anticipation case, we only perform insertion if it
124 is partially anticipated in some block, and fully available in all
125 of the predecessors.
127 insert/insert_aux/do_regular_insertion/do_partial_partial_insertion
128 performs these steps.
130 Fourth, we eliminate fully redundant expressions.
131 This is a simple statement walk that replaces redundant
132 calculations with the now available values. */
134 /* Representations of value numbers:
136 Value numbers are represented by a representative SSA_NAME. We
137 will create fake SSA_NAME's in situations where we need a
138 representative but do not have one (because it is a complex
139 expression). In order to facilitate storing the value numbers in
140 bitmaps, and keep the number of wasted SSA_NAME's down, we also
141 associate a value_id with each value number, and create full blown
142 ssa_name's only where we actually need them (IE in operands of
143 existing expressions).
145 Theoretically you could replace all the value_id's with
146 SSA_NAME_VERSION, but this would allocate a large number of
147 SSA_NAME's (which are each > 30 bytes) just to get a 4 byte number.
148 It would also require an additional indirection at each point we
149 use the value id. */
151 /* Representation of expressions on value numbers:
153 Expressions consisting of value numbers are represented the same
154 way as our VN internally represents them, with an additional
155 "pre_expr" wrapping around them in order to facilitate storing all
156 of the expressions in the same sets. */
158 /* Representation of sets:
160 The dataflow sets do not need to be sorted in any particular order
161 for the majority of their lifetime, are simply represented as two
162 bitmaps, one that keeps track of values present in the set, and one
163 that keeps track of expressions present in the set.
165 When we need them in topological order, we produce it on demand by
166 transforming the bitmap into an array and sorting it into topo
167 order. */
169 /* Type of expression, used to know which member of the PRE_EXPR union
170 is valid. */
172 enum pre_expr_kind
174 NAME,
175 NARY,
176 REFERENCE,
177 CONSTANT
180 typedef union pre_expr_union_d
182 tree name;
183 tree constant;
184 vn_nary_op_t nary;
185 vn_reference_t reference;
186 } pre_expr_union;
188 typedef struct pre_expr_d : typed_noop_remove <pre_expr_d>
190 enum pre_expr_kind kind;
191 unsigned int id;
192 pre_expr_union u;
194 /* hash_table support. */
195 typedef pre_expr_d value_type;
196 typedef pre_expr_d compare_type;
197 static inline hashval_t hash (const pre_expr_d *);
198 static inline int equal (const pre_expr_d *, const pre_expr_d *);
199 } *pre_expr;
201 #define PRE_EXPR_NAME(e) (e)->u.name
202 #define PRE_EXPR_NARY(e) (e)->u.nary
203 #define PRE_EXPR_REFERENCE(e) (e)->u.reference
204 #define PRE_EXPR_CONSTANT(e) (e)->u.constant
206 /* Compare E1 and E1 for equality. */
208 inline int
209 pre_expr_d::equal (const value_type *e1, const compare_type *e2)
211 if (e1->kind != e2->kind)
212 return false;
214 switch (e1->kind)
216 case CONSTANT:
217 return vn_constant_eq_with_type (PRE_EXPR_CONSTANT (e1),
218 PRE_EXPR_CONSTANT (e2));
219 case NAME:
220 return PRE_EXPR_NAME (e1) == PRE_EXPR_NAME (e2);
221 case NARY:
222 return vn_nary_op_eq (PRE_EXPR_NARY (e1), PRE_EXPR_NARY (e2));
223 case REFERENCE:
224 return vn_reference_eq (PRE_EXPR_REFERENCE (e1),
225 PRE_EXPR_REFERENCE (e2));
226 default:
227 gcc_unreachable ();
231 /* Hash E. */
233 inline hashval_t
234 pre_expr_d::hash (const value_type *e)
236 switch (e->kind)
238 case CONSTANT:
239 return vn_hash_constant_with_type (PRE_EXPR_CONSTANT (e));
240 case NAME:
241 return SSA_NAME_VERSION (PRE_EXPR_NAME (e));
242 case NARY:
243 return PRE_EXPR_NARY (e)->hashcode;
244 case REFERENCE:
245 return PRE_EXPR_REFERENCE (e)->hashcode;
246 default:
247 gcc_unreachable ();
251 /* Next global expression id number. */
252 static unsigned int next_expression_id;
254 /* Mapping from expression to id number we can use in bitmap sets. */
255 static vec<pre_expr> expressions;
256 static hash_table <pre_expr_d> expression_to_id;
257 static vec<unsigned> name_to_id;
259 /* Allocate an expression id for EXPR. */
261 static inline unsigned int
262 alloc_expression_id (pre_expr expr)
264 struct pre_expr_d **slot;
265 /* Make sure we won't overflow. */
266 gcc_assert (next_expression_id + 1 > next_expression_id);
267 expr->id = next_expression_id++;
268 expressions.safe_push (expr);
269 if (expr->kind == NAME)
271 unsigned version = SSA_NAME_VERSION (PRE_EXPR_NAME (expr));
272 /* vec::safe_grow_cleared allocates no headroom. Avoid frequent
273 re-allocations by using vec::reserve upfront. There is no
274 vec::quick_grow_cleared unfortunately. */
275 unsigned old_len = name_to_id.length ();
276 name_to_id.reserve (num_ssa_names - old_len);
277 name_to_id.safe_grow_cleared (num_ssa_names);
278 gcc_assert (name_to_id[version] == 0);
279 name_to_id[version] = expr->id;
281 else
283 slot = expression_to_id.find_slot (expr, INSERT);
284 gcc_assert (!*slot);
285 *slot = expr;
287 return next_expression_id - 1;
290 /* Return the expression id for tree EXPR. */
292 static inline unsigned int
293 get_expression_id (const pre_expr expr)
295 return expr->id;
298 static inline unsigned int
299 lookup_expression_id (const pre_expr expr)
301 struct pre_expr_d **slot;
303 if (expr->kind == NAME)
305 unsigned version = SSA_NAME_VERSION (PRE_EXPR_NAME (expr));
306 if (name_to_id.length () <= version)
307 return 0;
308 return name_to_id[version];
310 else
312 slot = expression_to_id.find_slot (expr, NO_INSERT);
313 if (!slot)
314 return 0;
315 return ((pre_expr)*slot)->id;
319 /* Return the existing expression id for EXPR, or create one if one
320 does not exist yet. */
322 static inline unsigned int
323 get_or_alloc_expression_id (pre_expr expr)
325 unsigned int id = lookup_expression_id (expr);
326 if (id == 0)
327 return alloc_expression_id (expr);
328 return expr->id = id;
331 /* Return the expression that has expression id ID */
333 static inline pre_expr
334 expression_for_id (unsigned int id)
336 return expressions[id];
339 /* Free the expression id field in all of our expressions,
340 and then destroy the expressions array. */
342 static void
343 clear_expression_ids (void)
345 expressions.release ();
348 static alloc_pool pre_expr_pool;
350 /* Given an SSA_NAME NAME, get or create a pre_expr to represent it. */
352 static pre_expr
353 get_or_alloc_expr_for_name (tree name)
355 struct pre_expr_d expr;
356 pre_expr result;
357 unsigned int result_id;
359 expr.kind = NAME;
360 expr.id = 0;
361 PRE_EXPR_NAME (&expr) = name;
362 result_id = lookup_expression_id (&expr);
363 if (result_id != 0)
364 return expression_for_id (result_id);
366 result = (pre_expr) pool_alloc (pre_expr_pool);
367 result->kind = NAME;
368 PRE_EXPR_NAME (result) = name;
369 alloc_expression_id (result);
370 return result;
373 /* An unordered bitmap set. One bitmap tracks values, the other,
374 expressions. */
375 typedef struct bitmap_set
377 bitmap_head expressions;
378 bitmap_head values;
379 } *bitmap_set_t;
381 #define FOR_EACH_EXPR_ID_IN_SET(set, id, bi) \
382 EXECUTE_IF_SET_IN_BITMAP (&(set)->expressions, 0, (id), (bi))
384 #define FOR_EACH_VALUE_ID_IN_SET(set, id, bi) \
385 EXECUTE_IF_SET_IN_BITMAP (&(set)->values, 0, (id), (bi))
387 /* Mapping from value id to expressions with that value_id. */
388 static vec<bitmap> value_expressions;
390 /* Sets that we need to keep track of. */
391 typedef struct bb_bitmap_sets
393 /* The EXP_GEN set, which represents expressions/values generated in
394 a basic block. */
395 bitmap_set_t exp_gen;
397 /* The PHI_GEN set, which represents PHI results generated in a
398 basic block. */
399 bitmap_set_t phi_gen;
401 /* The TMP_GEN set, which represents results/temporaries generated
402 in a basic block. IE the LHS of an expression. */
403 bitmap_set_t tmp_gen;
405 /* The AVAIL_OUT set, which represents which values are available in
406 a given basic block. */
407 bitmap_set_t avail_out;
409 /* The ANTIC_IN set, which represents which values are anticipatable
410 in a given basic block. */
411 bitmap_set_t antic_in;
413 /* The PA_IN set, which represents which values are
414 partially anticipatable in a given basic block. */
415 bitmap_set_t pa_in;
417 /* The NEW_SETS set, which is used during insertion to augment the
418 AVAIL_OUT set of blocks with the new insertions performed during
419 the current iteration. */
420 bitmap_set_t new_sets;
422 /* A cache for value_dies_in_block_x. */
423 bitmap expr_dies;
425 /* True if we have visited this block during ANTIC calculation. */
426 unsigned int visited : 1;
428 /* True we have deferred processing this block during ANTIC
429 calculation until its successor is processed. */
430 unsigned int deferred : 1;
432 /* True when the block contains a call that might not return. */
433 unsigned int contains_may_not_return_call : 1;
434 } *bb_value_sets_t;
436 #define EXP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->exp_gen
437 #define PHI_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->phi_gen
438 #define TMP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->tmp_gen
439 #define AVAIL_OUT(BB) ((bb_value_sets_t) ((BB)->aux))->avail_out
440 #define ANTIC_IN(BB) ((bb_value_sets_t) ((BB)->aux))->antic_in
441 #define PA_IN(BB) ((bb_value_sets_t) ((BB)->aux))->pa_in
442 #define NEW_SETS(BB) ((bb_value_sets_t) ((BB)->aux))->new_sets
443 #define EXPR_DIES(BB) ((bb_value_sets_t) ((BB)->aux))->expr_dies
444 #define BB_VISITED(BB) ((bb_value_sets_t) ((BB)->aux))->visited
445 #define BB_DEFERRED(BB) ((bb_value_sets_t) ((BB)->aux))->deferred
446 #define BB_MAY_NOTRETURN(BB) ((bb_value_sets_t) ((BB)->aux))->contains_may_not_return_call
449 /* Basic block list in postorder. */
450 static int *postorder;
451 static int postorder_num;
453 /* This structure is used to keep track of statistics on what
454 optimization PRE was able to perform. */
455 static struct
457 /* The number of RHS computations eliminated by PRE. */
458 int eliminations;
460 /* The number of new expressions/temporaries generated by PRE. */
461 int insertions;
463 /* The number of inserts found due to partial anticipation */
464 int pa_insert;
466 /* The number of new PHI nodes added by PRE. */
467 int phis;
468 } pre_stats;
470 static bool do_partial_partial;
471 static pre_expr bitmap_find_leader (bitmap_set_t, unsigned int);
472 static void bitmap_value_insert_into_set (bitmap_set_t, pre_expr);
473 static void bitmap_value_replace_in_set (bitmap_set_t, pre_expr);
474 static void bitmap_set_copy (bitmap_set_t, bitmap_set_t);
475 static bool bitmap_set_contains_value (bitmap_set_t, unsigned int);
476 static void bitmap_insert_into_set (bitmap_set_t, pre_expr);
477 static void bitmap_insert_into_set_1 (bitmap_set_t, pre_expr,
478 unsigned int, bool);
479 static bitmap_set_t bitmap_set_new (void);
480 static tree create_expression_by_pieces (basic_block, pre_expr, gimple_seq *,
481 tree);
482 static tree find_or_generate_expression (basic_block, tree, gimple_seq *);
483 static unsigned int get_expr_value_id (pre_expr);
485 /* We can add and remove elements and entries to and from sets
486 and hash tables, so we use alloc pools for them. */
488 static alloc_pool bitmap_set_pool;
489 static bitmap_obstack grand_bitmap_obstack;
491 /* Set of blocks with statements that have had their EH properties changed. */
492 static bitmap need_eh_cleanup;
494 /* Set of blocks with statements that have had their AB properties changed. */
495 static bitmap need_ab_cleanup;
497 /* A three tuple {e, pred, v} used to cache phi translations in the
498 phi_translate_table. */
500 typedef struct expr_pred_trans_d : typed_free_remove<expr_pred_trans_d>
502 /* The expression. */
503 pre_expr e;
505 /* The predecessor block along which we translated the expression. */
506 basic_block pred;
508 /* The value that resulted from the translation. */
509 pre_expr v;
511 /* The hashcode for the expression, pred pair. This is cached for
512 speed reasons. */
513 hashval_t hashcode;
515 /* hash_table support. */
516 typedef expr_pred_trans_d value_type;
517 typedef expr_pred_trans_d compare_type;
518 static inline hashval_t hash (const value_type *);
519 static inline int equal (const value_type *, const compare_type *);
520 } *expr_pred_trans_t;
521 typedef const struct expr_pred_trans_d *const_expr_pred_trans_t;
523 inline hashval_t
524 expr_pred_trans_d::hash (const expr_pred_trans_d *e)
526 return e->hashcode;
529 inline int
530 expr_pred_trans_d::equal (const value_type *ve1,
531 const compare_type *ve2)
533 basic_block b1 = ve1->pred;
534 basic_block b2 = ve2->pred;
536 /* If they are not translations for the same basic block, they can't
537 be equal. */
538 if (b1 != b2)
539 return false;
540 return pre_expr_d::equal (ve1->e, ve2->e);
543 /* The phi_translate_table caches phi translations for a given
544 expression and predecessor. */
545 static hash_table <expr_pred_trans_d> phi_translate_table;
547 /* Add the tuple mapping from {expression E, basic block PRED} to
548 the phi translation table and return whether it pre-existed. */
550 static inline bool
551 phi_trans_add (expr_pred_trans_t *entry, pre_expr e, basic_block pred)
553 expr_pred_trans_t *slot;
554 expr_pred_trans_d tem;
555 hashval_t hash = iterative_hash_hashval_t (pre_expr_d::hash (e),
556 pred->index);
557 tem.e = e;
558 tem.pred = pred;
559 tem.hashcode = hash;
560 slot = phi_translate_table.find_slot_with_hash (&tem, hash, INSERT);
561 if (*slot)
563 *entry = *slot;
564 return true;
567 *entry = *slot = XNEW (struct expr_pred_trans_d);
568 (*entry)->e = e;
569 (*entry)->pred = pred;
570 (*entry)->hashcode = hash;
571 return false;
575 /* Add expression E to the expression set of value id V. */
577 static void
578 add_to_value (unsigned int v, pre_expr e)
580 bitmap set;
582 gcc_checking_assert (get_expr_value_id (e) == v);
584 if (v >= value_expressions.length ())
586 value_expressions.safe_grow_cleared (v + 1);
589 set = value_expressions[v];
590 if (!set)
592 set = BITMAP_ALLOC (&grand_bitmap_obstack);
593 value_expressions[v] = set;
596 bitmap_set_bit (set, get_or_alloc_expression_id (e));
599 /* Create a new bitmap set and return it. */
601 static bitmap_set_t
602 bitmap_set_new (void)
604 bitmap_set_t ret = (bitmap_set_t) pool_alloc (bitmap_set_pool);
605 bitmap_initialize (&ret->expressions, &grand_bitmap_obstack);
606 bitmap_initialize (&ret->values, &grand_bitmap_obstack);
607 return ret;
610 /* Return the value id for a PRE expression EXPR. */
612 static unsigned int
613 get_expr_value_id (pre_expr expr)
615 unsigned int id;
616 switch (expr->kind)
618 case CONSTANT:
619 id = get_constant_value_id (PRE_EXPR_CONSTANT (expr));
620 break;
621 case NAME:
622 id = VN_INFO (PRE_EXPR_NAME (expr))->value_id;
623 break;
624 case NARY:
625 id = PRE_EXPR_NARY (expr)->value_id;
626 break;
627 case REFERENCE:
628 id = PRE_EXPR_REFERENCE (expr)->value_id;
629 break;
630 default:
631 gcc_unreachable ();
633 /* ??? We cannot assert that expr has a value-id (it can be 0), because
634 we assign value-ids only to expressions that have a result
635 in set_hashtable_value_ids. */
636 return id;
639 /* Return a SCCVN valnum (SSA name or constant) for the PRE value-id VAL. */
641 static tree
642 sccvn_valnum_from_value_id (unsigned int val)
644 bitmap_iterator bi;
645 unsigned int i;
646 bitmap exprset = value_expressions[val];
647 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
649 pre_expr vexpr = expression_for_id (i);
650 if (vexpr->kind == NAME)
651 return VN_INFO (PRE_EXPR_NAME (vexpr))->valnum;
652 else if (vexpr->kind == CONSTANT)
653 return PRE_EXPR_CONSTANT (vexpr);
655 return NULL_TREE;
658 /* Remove an expression EXPR from a bitmapped set. */
660 static void
661 bitmap_remove_from_set (bitmap_set_t set, pre_expr expr)
663 unsigned int val = get_expr_value_id (expr);
664 if (!value_id_constant_p (val))
666 bitmap_clear_bit (&set->values, val);
667 bitmap_clear_bit (&set->expressions, get_expression_id (expr));
671 static void
672 bitmap_insert_into_set_1 (bitmap_set_t set, pre_expr expr,
673 unsigned int val, bool allow_constants)
675 if (allow_constants || !value_id_constant_p (val))
677 /* We specifically expect this and only this function to be able to
678 insert constants into a set. */
679 bitmap_set_bit (&set->values, val);
680 bitmap_set_bit (&set->expressions, get_or_alloc_expression_id (expr));
684 /* Insert an expression EXPR into a bitmapped set. */
686 static void
687 bitmap_insert_into_set (bitmap_set_t set, pre_expr expr)
689 bitmap_insert_into_set_1 (set, expr, get_expr_value_id (expr), false);
692 /* Copy a bitmapped set ORIG, into bitmapped set DEST. */
694 static void
695 bitmap_set_copy (bitmap_set_t dest, bitmap_set_t orig)
697 bitmap_copy (&dest->expressions, &orig->expressions);
698 bitmap_copy (&dest->values, &orig->values);
702 /* Free memory used up by SET. */
703 static void
704 bitmap_set_free (bitmap_set_t set)
706 bitmap_clear (&set->expressions);
707 bitmap_clear (&set->values);
711 /* Generate an topological-ordered array of bitmap set SET. */
713 static vec<pre_expr>
714 sorted_array_from_bitmap_set (bitmap_set_t set)
716 unsigned int i, j;
717 bitmap_iterator bi, bj;
718 vec<pre_expr> result;
720 /* Pre-allocate roughly enough space for the array. */
721 result.create (bitmap_count_bits (&set->values));
723 FOR_EACH_VALUE_ID_IN_SET (set, i, bi)
725 /* The number of expressions having a given value is usually
726 relatively small. Thus, rather than making a vector of all
727 the expressions and sorting it by value-id, we walk the values
728 and check in the reverse mapping that tells us what expressions
729 have a given value, to filter those in our set. As a result,
730 the expressions are inserted in value-id order, which means
731 topological order.
733 If this is somehow a significant lose for some cases, we can
734 choose which set to walk based on the set size. */
735 bitmap exprset = value_expressions[i];
736 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, j, bj)
738 if (bitmap_bit_p (&set->expressions, j))
739 result.safe_push (expression_for_id (j));
743 return result;
746 /* Perform bitmapped set operation DEST &= ORIG. */
748 static void
749 bitmap_set_and (bitmap_set_t dest, bitmap_set_t orig)
751 bitmap_iterator bi;
752 unsigned int i;
754 if (dest != orig)
756 bitmap_head temp;
757 bitmap_initialize (&temp, &grand_bitmap_obstack);
759 bitmap_and_into (&dest->values, &orig->values);
760 bitmap_copy (&temp, &dest->expressions);
761 EXECUTE_IF_SET_IN_BITMAP (&temp, 0, i, bi)
763 pre_expr expr = expression_for_id (i);
764 unsigned int value_id = get_expr_value_id (expr);
765 if (!bitmap_bit_p (&dest->values, value_id))
766 bitmap_clear_bit (&dest->expressions, i);
768 bitmap_clear (&temp);
772 /* Subtract all values and expressions contained in ORIG from DEST. */
774 static bitmap_set_t
775 bitmap_set_subtract (bitmap_set_t dest, bitmap_set_t orig)
777 bitmap_set_t result = bitmap_set_new ();
778 bitmap_iterator bi;
779 unsigned int i;
781 bitmap_and_compl (&result->expressions, &dest->expressions,
782 &orig->expressions);
784 FOR_EACH_EXPR_ID_IN_SET (result, i, bi)
786 pre_expr expr = expression_for_id (i);
787 unsigned int value_id = get_expr_value_id (expr);
788 bitmap_set_bit (&result->values, value_id);
791 return result;
794 /* Subtract all the values in bitmap set B from bitmap set A. */
796 static void
797 bitmap_set_subtract_values (bitmap_set_t a, bitmap_set_t b)
799 unsigned int i;
800 bitmap_iterator bi;
801 bitmap_head temp;
803 bitmap_initialize (&temp, &grand_bitmap_obstack);
805 bitmap_copy (&temp, &a->expressions);
806 EXECUTE_IF_SET_IN_BITMAP (&temp, 0, i, bi)
808 pre_expr expr = expression_for_id (i);
809 if (bitmap_set_contains_value (b, get_expr_value_id (expr)))
810 bitmap_remove_from_set (a, expr);
812 bitmap_clear (&temp);
816 /* Return true if bitmapped set SET contains the value VALUE_ID. */
818 static bool
819 bitmap_set_contains_value (bitmap_set_t set, unsigned int value_id)
821 if (value_id_constant_p (value_id))
822 return true;
824 if (!set || bitmap_empty_p (&set->expressions))
825 return false;
827 return bitmap_bit_p (&set->values, value_id);
830 static inline bool
831 bitmap_set_contains_expr (bitmap_set_t set, const pre_expr expr)
833 return bitmap_bit_p (&set->expressions, get_expression_id (expr));
836 /* Replace an instance of value LOOKFOR with expression EXPR in SET. */
838 static void
839 bitmap_set_replace_value (bitmap_set_t set, unsigned int lookfor,
840 const pre_expr expr)
842 bitmap exprset;
843 unsigned int i;
844 bitmap_iterator bi;
846 if (value_id_constant_p (lookfor))
847 return;
849 if (!bitmap_set_contains_value (set, lookfor))
850 return;
852 /* The number of expressions having a given value is usually
853 significantly less than the total number of expressions in SET.
854 Thus, rather than check, for each expression in SET, whether it
855 has the value LOOKFOR, we walk the reverse mapping that tells us
856 what expressions have a given value, and see if any of those
857 expressions are in our set. For large testcases, this is about
858 5-10x faster than walking the bitmap. If this is somehow a
859 significant lose for some cases, we can choose which set to walk
860 based on the set size. */
861 exprset = value_expressions[lookfor];
862 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
864 if (bitmap_clear_bit (&set->expressions, i))
866 bitmap_set_bit (&set->expressions, get_expression_id (expr));
867 return;
871 gcc_unreachable ();
874 /* Return true if two bitmap sets are equal. */
876 static bool
877 bitmap_set_equal (bitmap_set_t a, bitmap_set_t b)
879 return bitmap_equal_p (&a->values, &b->values);
882 /* Replace an instance of EXPR's VALUE with EXPR in SET if it exists,
883 and add it otherwise. */
885 static void
886 bitmap_value_replace_in_set (bitmap_set_t set, pre_expr expr)
888 unsigned int val = get_expr_value_id (expr);
890 if (bitmap_set_contains_value (set, val))
891 bitmap_set_replace_value (set, val, expr);
892 else
893 bitmap_insert_into_set (set, expr);
896 /* Insert EXPR into SET if EXPR's value is not already present in
897 SET. */
899 static void
900 bitmap_value_insert_into_set (bitmap_set_t set, pre_expr expr)
902 unsigned int val = get_expr_value_id (expr);
904 gcc_checking_assert (expr->id == get_or_alloc_expression_id (expr));
906 /* Constant values are always considered to be part of the set. */
907 if (value_id_constant_p (val))
908 return;
910 /* If the value membership changed, add the expression. */
911 if (bitmap_set_bit (&set->values, val))
912 bitmap_set_bit (&set->expressions, expr->id);
915 /* Print out EXPR to outfile. */
917 static void
918 print_pre_expr (FILE *outfile, const pre_expr expr)
920 switch (expr->kind)
922 case CONSTANT:
923 print_generic_expr (outfile, PRE_EXPR_CONSTANT (expr), 0);
924 break;
925 case NAME:
926 print_generic_expr (outfile, PRE_EXPR_NAME (expr), 0);
927 break;
928 case NARY:
930 unsigned int i;
931 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
932 fprintf (outfile, "{%s,", get_tree_code_name (nary->opcode));
933 for (i = 0; i < nary->length; i++)
935 print_generic_expr (outfile, nary->op[i], 0);
936 if (i != (unsigned) nary->length - 1)
937 fprintf (outfile, ",");
939 fprintf (outfile, "}");
941 break;
943 case REFERENCE:
945 vn_reference_op_t vro;
946 unsigned int i;
947 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
948 fprintf (outfile, "{");
949 for (i = 0;
950 ref->operands.iterate (i, &vro);
951 i++)
953 bool closebrace = false;
954 if (vro->opcode != SSA_NAME
955 && TREE_CODE_CLASS (vro->opcode) != tcc_declaration)
957 fprintf (outfile, "%s", get_tree_code_name (vro->opcode));
958 if (vro->op0)
960 fprintf (outfile, "<");
961 closebrace = true;
964 if (vro->op0)
966 print_generic_expr (outfile, vro->op0, 0);
967 if (vro->op1)
969 fprintf (outfile, ",");
970 print_generic_expr (outfile, vro->op1, 0);
972 if (vro->op2)
974 fprintf (outfile, ",");
975 print_generic_expr (outfile, vro->op2, 0);
978 if (closebrace)
979 fprintf (outfile, ">");
980 if (i != ref->operands.length () - 1)
981 fprintf (outfile, ",");
983 fprintf (outfile, "}");
984 if (ref->vuse)
986 fprintf (outfile, "@");
987 print_generic_expr (outfile, ref->vuse, 0);
990 break;
993 void debug_pre_expr (pre_expr);
995 /* Like print_pre_expr but always prints to stderr. */
996 DEBUG_FUNCTION void
997 debug_pre_expr (pre_expr e)
999 print_pre_expr (stderr, e);
1000 fprintf (stderr, "\n");
1003 /* Print out SET to OUTFILE. */
1005 static void
1006 print_bitmap_set (FILE *outfile, bitmap_set_t set,
1007 const char *setname, int blockindex)
1009 fprintf (outfile, "%s[%d] := { ", setname, blockindex);
1010 if (set)
1012 bool first = true;
1013 unsigned i;
1014 bitmap_iterator bi;
1016 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
1018 const pre_expr expr = expression_for_id (i);
1020 if (!first)
1021 fprintf (outfile, ", ");
1022 first = false;
1023 print_pre_expr (outfile, expr);
1025 fprintf (outfile, " (%04d)", get_expr_value_id (expr));
1028 fprintf (outfile, " }\n");
1031 void debug_bitmap_set (bitmap_set_t);
1033 DEBUG_FUNCTION void
1034 debug_bitmap_set (bitmap_set_t set)
1036 print_bitmap_set (stderr, set, "debug", 0);
1039 void debug_bitmap_sets_for (basic_block);
1041 DEBUG_FUNCTION void
1042 debug_bitmap_sets_for (basic_block bb)
1044 print_bitmap_set (stderr, AVAIL_OUT (bb), "avail_out", bb->index);
1045 print_bitmap_set (stderr, EXP_GEN (bb), "exp_gen", bb->index);
1046 print_bitmap_set (stderr, PHI_GEN (bb), "phi_gen", bb->index);
1047 print_bitmap_set (stderr, TMP_GEN (bb), "tmp_gen", bb->index);
1048 print_bitmap_set (stderr, ANTIC_IN (bb), "antic_in", bb->index);
1049 if (do_partial_partial)
1050 print_bitmap_set (stderr, PA_IN (bb), "pa_in", bb->index);
1051 print_bitmap_set (stderr, NEW_SETS (bb), "new_sets", bb->index);
1054 /* Print out the expressions that have VAL to OUTFILE. */
1056 static void
1057 print_value_expressions (FILE *outfile, unsigned int val)
1059 bitmap set = value_expressions[val];
1060 if (set)
1062 bitmap_set x;
1063 char s[10];
1064 sprintf (s, "%04d", val);
1065 x.expressions = *set;
1066 print_bitmap_set (outfile, &x, s, 0);
1071 DEBUG_FUNCTION void
1072 debug_value_expressions (unsigned int val)
1074 print_value_expressions (stderr, val);
1077 /* Given a CONSTANT, allocate a new CONSTANT type PRE_EXPR to
1078 represent it. */
1080 static pre_expr
1081 get_or_alloc_expr_for_constant (tree constant)
1083 unsigned int result_id;
1084 unsigned int value_id;
1085 struct pre_expr_d expr;
1086 pre_expr newexpr;
1088 expr.kind = CONSTANT;
1089 PRE_EXPR_CONSTANT (&expr) = constant;
1090 result_id = lookup_expression_id (&expr);
1091 if (result_id != 0)
1092 return expression_for_id (result_id);
1094 newexpr = (pre_expr) pool_alloc (pre_expr_pool);
1095 newexpr->kind = CONSTANT;
1096 PRE_EXPR_CONSTANT (newexpr) = constant;
1097 alloc_expression_id (newexpr);
1098 value_id = get_or_alloc_constant_value_id (constant);
1099 add_to_value (value_id, newexpr);
1100 return newexpr;
1103 /* Given a value id V, find the actual tree representing the constant
1104 value if there is one, and return it. Return NULL if we can't find
1105 a constant. */
1107 static tree
1108 get_constant_for_value_id (unsigned int v)
1110 if (value_id_constant_p (v))
1112 unsigned int i;
1113 bitmap_iterator bi;
1114 bitmap exprset = value_expressions[v];
1116 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
1118 pre_expr expr = expression_for_id (i);
1119 if (expr->kind == CONSTANT)
1120 return PRE_EXPR_CONSTANT (expr);
1123 return NULL;
1126 /* Get or allocate a pre_expr for a piece of GIMPLE, and return it.
1127 Currently only supports constants and SSA_NAMES. */
1128 static pre_expr
1129 get_or_alloc_expr_for (tree t)
1131 if (TREE_CODE (t) == SSA_NAME)
1132 return get_or_alloc_expr_for_name (t);
1133 else if (is_gimple_min_invariant (t))
1134 return get_or_alloc_expr_for_constant (t);
1135 else
1137 /* More complex expressions can result from SCCVN expression
1138 simplification that inserts values for them. As they all
1139 do not have VOPs the get handled by the nary ops struct. */
1140 vn_nary_op_t result;
1141 unsigned int result_id;
1142 vn_nary_op_lookup (t, &result);
1143 if (result != NULL)
1145 pre_expr e = (pre_expr) pool_alloc (pre_expr_pool);
1146 e->kind = NARY;
1147 PRE_EXPR_NARY (e) = result;
1148 result_id = lookup_expression_id (e);
1149 if (result_id != 0)
1151 pool_free (pre_expr_pool, e);
1152 e = expression_for_id (result_id);
1153 return e;
1155 alloc_expression_id (e);
1156 return e;
1159 return NULL;
1162 /* Return the folded version of T if T, when folded, is a gimple
1163 min_invariant. Otherwise, return T. */
1165 static pre_expr
1166 fully_constant_expression (pre_expr e)
1168 switch (e->kind)
1170 case CONSTANT:
1171 return e;
1172 case NARY:
1174 vn_nary_op_t nary = PRE_EXPR_NARY (e);
1175 switch (TREE_CODE_CLASS (nary->opcode))
1177 case tcc_binary:
1178 case tcc_comparison:
1180 /* We have to go from trees to pre exprs to value ids to
1181 constants. */
1182 tree naryop0 = nary->op[0];
1183 tree naryop1 = nary->op[1];
1184 tree result;
1185 if (!is_gimple_min_invariant (naryop0))
1187 pre_expr rep0 = get_or_alloc_expr_for (naryop0);
1188 unsigned int vrep0 = get_expr_value_id (rep0);
1189 tree const0 = get_constant_for_value_id (vrep0);
1190 if (const0)
1191 naryop0 = fold_convert (TREE_TYPE (naryop0), const0);
1193 if (!is_gimple_min_invariant (naryop1))
1195 pre_expr rep1 = get_or_alloc_expr_for (naryop1);
1196 unsigned int vrep1 = get_expr_value_id (rep1);
1197 tree const1 = get_constant_for_value_id (vrep1);
1198 if (const1)
1199 naryop1 = fold_convert (TREE_TYPE (naryop1), const1);
1201 result = fold_binary (nary->opcode, nary->type,
1202 naryop0, naryop1);
1203 if (result && is_gimple_min_invariant (result))
1204 return get_or_alloc_expr_for_constant (result);
1205 /* We might have simplified the expression to a
1206 SSA_NAME for example from x_1 * 1. But we cannot
1207 insert a PHI for x_1 unconditionally as x_1 might
1208 not be available readily. */
1209 return e;
1211 case tcc_reference:
1212 if (nary->opcode != REALPART_EXPR
1213 && nary->opcode != IMAGPART_EXPR
1214 && nary->opcode != VIEW_CONVERT_EXPR)
1215 return e;
1216 /* Fallthrough. */
1217 case tcc_unary:
1219 /* We have to go from trees to pre exprs to value ids to
1220 constants. */
1221 tree naryop0 = nary->op[0];
1222 tree const0, result;
1223 if (is_gimple_min_invariant (naryop0))
1224 const0 = naryop0;
1225 else
1227 pre_expr rep0 = get_or_alloc_expr_for (naryop0);
1228 unsigned int vrep0 = get_expr_value_id (rep0);
1229 const0 = get_constant_for_value_id (vrep0);
1231 result = NULL;
1232 if (const0)
1234 tree type1 = TREE_TYPE (nary->op[0]);
1235 const0 = fold_convert (type1, const0);
1236 result = fold_unary (nary->opcode, nary->type, const0);
1238 if (result && is_gimple_min_invariant (result))
1239 return get_or_alloc_expr_for_constant (result);
1240 return e;
1242 default:
1243 return e;
1246 case REFERENCE:
1248 vn_reference_t ref = PRE_EXPR_REFERENCE (e);
1249 tree folded;
1250 if ((folded = fully_constant_vn_reference_p (ref)))
1251 return get_or_alloc_expr_for_constant (folded);
1252 return e;
1254 default:
1255 return e;
1257 return e;
1260 /* Translate the VUSE backwards through phi nodes in PHIBLOCK, so that
1261 it has the value it would have in BLOCK. Set *SAME_VALID to true
1262 in case the new vuse doesn't change the value id of the OPERANDS. */
1264 static tree
1265 translate_vuse_through_block (vec<vn_reference_op_s> operands,
1266 alias_set_type set, tree type, tree vuse,
1267 basic_block phiblock,
1268 basic_block block, bool *same_valid)
1270 gimple phi = SSA_NAME_DEF_STMT (vuse);
1271 ao_ref ref;
1272 edge e = NULL;
1273 bool use_oracle;
1275 *same_valid = true;
1277 if (gimple_bb (phi) != phiblock)
1278 return vuse;
1280 use_oracle = ao_ref_init_from_vn_reference (&ref, set, type, operands);
1282 /* Use the alias-oracle to find either the PHI node in this block,
1283 the first VUSE used in this block that is equivalent to vuse or
1284 the first VUSE which definition in this block kills the value. */
1285 if (gimple_code (phi) == GIMPLE_PHI)
1286 e = find_edge (block, phiblock);
1287 else if (use_oracle)
1288 while (!stmt_may_clobber_ref_p_1 (phi, &ref))
1290 vuse = gimple_vuse (phi);
1291 phi = SSA_NAME_DEF_STMT (vuse);
1292 if (gimple_bb (phi) != phiblock)
1293 return vuse;
1294 if (gimple_code (phi) == GIMPLE_PHI)
1296 e = find_edge (block, phiblock);
1297 break;
1300 else
1301 return NULL_TREE;
1303 if (e)
1305 if (use_oracle)
1307 bitmap visited = NULL;
1308 unsigned int cnt;
1309 /* Try to find a vuse that dominates this phi node by skipping
1310 non-clobbering statements. */
1311 vuse = get_continuation_for_phi (phi, &ref, &cnt, &visited, false);
1312 if (visited)
1313 BITMAP_FREE (visited);
1315 else
1316 vuse = NULL_TREE;
1317 if (!vuse)
1319 /* If we didn't find any, the value ID can't stay the same,
1320 but return the translated vuse. */
1321 *same_valid = false;
1322 vuse = PHI_ARG_DEF (phi, e->dest_idx);
1324 /* ??? We would like to return vuse here as this is the canonical
1325 upmost vdef that this reference is associated with. But during
1326 insertion of the references into the hash tables we only ever
1327 directly insert with their direct gimple_vuse, hence returning
1328 something else would make us not find the other expression. */
1329 return PHI_ARG_DEF (phi, e->dest_idx);
1332 return NULL_TREE;
1335 /* Like bitmap_find_leader, but checks for the value existing in SET1 *or*
1336 SET2. This is used to avoid making a set consisting of the union
1337 of PA_IN and ANTIC_IN during insert. */
1339 static inline pre_expr
1340 find_leader_in_sets (unsigned int val, bitmap_set_t set1, bitmap_set_t set2)
1342 pre_expr result;
1344 result = bitmap_find_leader (set1, val);
1345 if (!result && set2)
1346 result = bitmap_find_leader (set2, val);
1347 return result;
1350 /* Get the tree type for our PRE expression e. */
1352 static tree
1353 get_expr_type (const pre_expr e)
1355 switch (e->kind)
1357 case NAME:
1358 return TREE_TYPE (PRE_EXPR_NAME (e));
1359 case CONSTANT:
1360 return TREE_TYPE (PRE_EXPR_CONSTANT (e));
1361 case REFERENCE:
1362 return PRE_EXPR_REFERENCE (e)->type;
1363 case NARY:
1364 return PRE_EXPR_NARY (e)->type;
1366 gcc_unreachable ();
1369 /* Get a representative SSA_NAME for a given expression.
1370 Since all of our sub-expressions are treated as values, we require
1371 them to be SSA_NAME's for simplicity.
1372 Prior versions of GVNPRE used to use "value handles" here, so that
1373 an expression would be VH.11 + VH.10 instead of d_3 + e_6. In
1374 either case, the operands are really values (IE we do not expect
1375 them to be usable without finding leaders). */
1377 static tree
1378 get_representative_for (const pre_expr e)
1380 tree name;
1381 unsigned int value_id = get_expr_value_id (e);
1383 switch (e->kind)
1385 case NAME:
1386 return PRE_EXPR_NAME (e);
1387 case CONSTANT:
1388 return PRE_EXPR_CONSTANT (e);
1389 case NARY:
1390 case REFERENCE:
1392 /* Go through all of the expressions representing this value
1393 and pick out an SSA_NAME. */
1394 unsigned int i;
1395 bitmap_iterator bi;
1396 bitmap exprs = value_expressions[value_id];
1397 EXECUTE_IF_SET_IN_BITMAP (exprs, 0, i, bi)
1399 pre_expr rep = expression_for_id (i);
1400 if (rep->kind == NAME)
1401 return PRE_EXPR_NAME (rep);
1402 else if (rep->kind == CONSTANT)
1403 return PRE_EXPR_CONSTANT (rep);
1406 break;
1409 /* If we reached here we couldn't find an SSA_NAME. This can
1410 happen when we've discovered a value that has never appeared in
1411 the program as set to an SSA_NAME, as the result of phi translation.
1412 Create one here.
1413 ??? We should be able to re-use this when we insert the statement
1414 to compute it. */
1415 name = make_temp_ssa_name (get_expr_type (e), gimple_build_nop (), "pretmp");
1416 VN_INFO_GET (name)->value_id = value_id;
1417 VN_INFO (name)->valnum = name;
1418 /* ??? For now mark this SSA name for release by SCCVN. */
1419 VN_INFO (name)->needs_insertion = true;
1420 add_to_value (value_id, get_or_alloc_expr_for_name (name));
1421 if (dump_file && (dump_flags & TDF_DETAILS))
1423 fprintf (dump_file, "Created SSA_NAME representative ");
1424 print_generic_expr (dump_file, name, 0);
1425 fprintf (dump_file, " for expression:");
1426 print_pre_expr (dump_file, e);
1427 fprintf (dump_file, " (%04d)\n", value_id);
1430 return name;
1435 static pre_expr
1436 phi_translate (pre_expr expr, bitmap_set_t set1, bitmap_set_t set2,
1437 basic_block pred, basic_block phiblock);
1439 /* Translate EXPR using phis in PHIBLOCK, so that it has the values of
1440 the phis in PRED. Return NULL if we can't find a leader for each part
1441 of the translated expression. */
1443 static pre_expr
1444 phi_translate_1 (pre_expr expr, bitmap_set_t set1, bitmap_set_t set2,
1445 basic_block pred, basic_block phiblock)
1447 switch (expr->kind)
1449 case NARY:
1451 unsigned int i;
1452 bool changed = false;
1453 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1454 vn_nary_op_t newnary = XALLOCAVAR (struct vn_nary_op_s,
1455 sizeof_vn_nary_op (nary->length));
1456 memcpy (newnary, nary, sizeof_vn_nary_op (nary->length));
1458 for (i = 0; i < newnary->length; i++)
1460 if (TREE_CODE (newnary->op[i]) != SSA_NAME)
1461 continue;
1462 else
1464 pre_expr leader, result;
1465 unsigned int op_val_id = VN_INFO (newnary->op[i])->value_id;
1466 leader = find_leader_in_sets (op_val_id, set1, set2);
1467 result = phi_translate (leader, set1, set2, pred, phiblock);
1468 if (result && result != leader)
1470 tree name = get_representative_for (result);
1471 if (!name)
1472 return NULL;
1473 newnary->op[i] = name;
1475 else if (!result)
1476 return NULL;
1478 changed |= newnary->op[i] != nary->op[i];
1481 if (changed)
1483 pre_expr constant;
1484 unsigned int new_val_id;
1486 tree result = vn_nary_op_lookup_pieces (newnary->length,
1487 newnary->opcode,
1488 newnary->type,
1489 &newnary->op[0],
1490 &nary);
1491 if (result && is_gimple_min_invariant (result))
1492 return get_or_alloc_expr_for_constant (result);
1494 expr = (pre_expr) pool_alloc (pre_expr_pool);
1495 expr->kind = NARY;
1496 expr->id = 0;
1497 if (nary)
1499 PRE_EXPR_NARY (expr) = nary;
1500 constant = fully_constant_expression (expr);
1501 if (constant != expr)
1502 return constant;
1504 new_val_id = nary->value_id;
1505 get_or_alloc_expression_id (expr);
1507 else
1509 new_val_id = get_next_value_id ();
1510 value_expressions.safe_grow_cleared (get_max_value_id () + 1);
1511 nary = vn_nary_op_insert_pieces (newnary->length,
1512 newnary->opcode,
1513 newnary->type,
1514 &newnary->op[0],
1515 result, new_val_id);
1516 PRE_EXPR_NARY (expr) = nary;
1517 constant = fully_constant_expression (expr);
1518 if (constant != expr)
1519 return constant;
1520 get_or_alloc_expression_id (expr);
1522 add_to_value (new_val_id, expr);
1524 return expr;
1526 break;
1528 case REFERENCE:
1530 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1531 vec<vn_reference_op_s> operands = ref->operands;
1532 tree vuse = ref->vuse;
1533 tree newvuse = vuse;
1534 vec<vn_reference_op_s> newoperands = vNULL;
1535 bool changed = false, same_valid = true;
1536 unsigned int i, j, n;
1537 vn_reference_op_t operand;
1538 vn_reference_t newref;
1540 for (i = 0, j = 0;
1541 operands.iterate (i, &operand); i++, j++)
1543 pre_expr opresult;
1544 pre_expr leader;
1545 tree op[3];
1546 tree type = operand->type;
1547 vn_reference_op_s newop = *operand;
1548 op[0] = operand->op0;
1549 op[1] = operand->op1;
1550 op[2] = operand->op2;
1551 for (n = 0; n < 3; ++n)
1553 unsigned int op_val_id;
1554 if (!op[n])
1555 continue;
1556 if (TREE_CODE (op[n]) != SSA_NAME)
1558 /* We can't possibly insert these. */
1559 if (n != 0
1560 && !is_gimple_min_invariant (op[n]))
1561 break;
1562 continue;
1564 op_val_id = VN_INFO (op[n])->value_id;
1565 leader = find_leader_in_sets (op_val_id, set1, set2);
1566 if (!leader)
1567 break;
1568 opresult = phi_translate (leader, set1, set2, pred, phiblock);
1569 if (!opresult)
1570 break;
1571 if (opresult != leader)
1573 tree name = get_representative_for (opresult);
1574 if (!name)
1575 break;
1576 changed |= name != op[n];
1577 op[n] = name;
1580 if (n != 3)
1582 newoperands.release ();
1583 return NULL;
1585 if (!newoperands.exists ())
1586 newoperands = operands.copy ();
1587 /* We may have changed from an SSA_NAME to a constant */
1588 if (newop.opcode == SSA_NAME && TREE_CODE (op[0]) != SSA_NAME)
1589 newop.opcode = TREE_CODE (op[0]);
1590 newop.type = type;
1591 newop.op0 = op[0];
1592 newop.op1 = op[1];
1593 newop.op2 = op[2];
1594 /* If it transforms a non-constant ARRAY_REF into a constant
1595 one, adjust the constant offset. */
1596 if (newop.opcode == ARRAY_REF
1597 && newop.off == -1
1598 && TREE_CODE (op[0]) == INTEGER_CST
1599 && TREE_CODE (op[1]) == INTEGER_CST
1600 && TREE_CODE (op[2]) == INTEGER_CST)
1602 double_int off = tree_to_double_int (op[0]);
1603 off += -tree_to_double_int (op[1]);
1604 off *= tree_to_double_int (op[2]);
1605 if (off.fits_shwi ())
1606 newop.off = off.low;
1608 newoperands[j] = newop;
1609 /* If it transforms from an SSA_NAME to an address, fold with
1610 a preceding indirect reference. */
1611 if (j > 0 && op[0] && TREE_CODE (op[0]) == ADDR_EXPR
1612 && newoperands[j - 1].opcode == MEM_REF)
1613 vn_reference_fold_indirect (&newoperands, &j);
1615 if (i != operands.length ())
1617 newoperands.release ();
1618 return NULL;
1621 if (vuse)
1623 newvuse = translate_vuse_through_block (newoperands,
1624 ref->set, ref->type,
1625 vuse, phiblock, pred,
1626 &same_valid);
1627 if (newvuse == NULL_TREE)
1629 newoperands.release ();
1630 return NULL;
1634 if (changed || newvuse != vuse)
1636 unsigned int new_val_id;
1637 pre_expr constant;
1639 tree result = vn_reference_lookup_pieces (newvuse, ref->set,
1640 ref->type,
1641 newoperands,
1642 &newref, VN_WALK);
1643 if (result)
1644 newoperands.release ();
1646 /* We can always insert constants, so if we have a partial
1647 redundant constant load of another type try to translate it
1648 to a constant of appropriate type. */
1649 if (result && is_gimple_min_invariant (result))
1651 tree tem = result;
1652 if (!useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1654 tem = fold_unary (VIEW_CONVERT_EXPR, ref->type, result);
1655 if (tem && !is_gimple_min_invariant (tem))
1656 tem = NULL_TREE;
1658 if (tem)
1659 return get_or_alloc_expr_for_constant (tem);
1662 /* If we'd have to convert things we would need to validate
1663 if we can insert the translated expression. So fail
1664 here for now - we cannot insert an alias with a different
1665 type in the VN tables either, as that would assert. */
1666 if (result
1667 && !useless_type_conversion_p (ref->type, TREE_TYPE (result)))
1668 return NULL;
1669 else if (!result && newref
1670 && !useless_type_conversion_p (ref->type, newref->type))
1672 newoperands.release ();
1673 return NULL;
1676 expr = (pre_expr) pool_alloc (pre_expr_pool);
1677 expr->kind = REFERENCE;
1678 expr->id = 0;
1680 if (newref)
1682 PRE_EXPR_REFERENCE (expr) = newref;
1683 constant = fully_constant_expression (expr);
1684 if (constant != expr)
1685 return constant;
1687 new_val_id = newref->value_id;
1688 get_or_alloc_expression_id (expr);
1690 else
1692 if (changed || !same_valid)
1694 new_val_id = get_next_value_id ();
1695 value_expressions.safe_grow_cleared
1696 (get_max_value_id () + 1);
1698 else
1699 new_val_id = ref->value_id;
1700 newref = vn_reference_insert_pieces (newvuse, ref->set,
1701 ref->type,
1702 newoperands,
1703 result, new_val_id);
1704 newoperands.create (0);
1705 PRE_EXPR_REFERENCE (expr) = newref;
1706 constant = fully_constant_expression (expr);
1707 if (constant != expr)
1708 return constant;
1709 get_or_alloc_expression_id (expr);
1711 add_to_value (new_val_id, expr);
1713 newoperands.release ();
1714 return expr;
1716 break;
1718 case NAME:
1720 tree name = PRE_EXPR_NAME (expr);
1721 gimple def_stmt = SSA_NAME_DEF_STMT (name);
1722 /* If the SSA name is defined by a PHI node in this block,
1723 translate it. */
1724 if (gimple_code (def_stmt) == GIMPLE_PHI
1725 && gimple_bb (def_stmt) == phiblock)
1727 edge e = find_edge (pred, gimple_bb (def_stmt));
1728 tree def = PHI_ARG_DEF (def_stmt, e->dest_idx);
1730 /* Handle constant. */
1731 if (is_gimple_min_invariant (def))
1732 return get_or_alloc_expr_for_constant (def);
1734 return get_or_alloc_expr_for_name (def);
1736 /* Otherwise return it unchanged - it will get cleaned if its
1737 value is not available in PREDs AVAIL_OUT set of expressions. */
1738 return expr;
1741 default:
1742 gcc_unreachable ();
1746 /* Wrapper around phi_translate_1 providing caching functionality. */
1748 static pre_expr
1749 phi_translate (pre_expr expr, bitmap_set_t set1, bitmap_set_t set2,
1750 basic_block pred, basic_block phiblock)
1752 expr_pred_trans_t slot = NULL;
1753 pre_expr phitrans;
1755 if (!expr)
1756 return NULL;
1758 /* Constants contain no values that need translation. */
1759 if (expr->kind == CONSTANT)
1760 return expr;
1762 if (value_id_constant_p (get_expr_value_id (expr)))
1763 return expr;
1765 /* Don't add translations of NAMEs as those are cheap to translate. */
1766 if (expr->kind != NAME)
1768 if (phi_trans_add (&slot, expr, pred))
1769 return slot->v;
1770 /* Store NULL for the value we want to return in the case of
1771 recursing. */
1772 slot->v = NULL;
1775 /* Translate. */
1776 phitrans = phi_translate_1 (expr, set1, set2, pred, phiblock);
1778 if (slot)
1780 if (phitrans)
1781 slot->v = phitrans;
1782 else
1783 /* Remove failed translations again, they cause insert
1784 iteration to not pick up new opportunities reliably. */
1785 phi_translate_table.remove_elt_with_hash (slot, slot->hashcode);
1788 return phitrans;
1792 /* For each expression in SET, translate the values through phi nodes
1793 in PHIBLOCK using edge PHIBLOCK->PRED, and store the resulting
1794 expressions in DEST. */
1796 static void
1797 phi_translate_set (bitmap_set_t dest, bitmap_set_t set, basic_block pred,
1798 basic_block phiblock)
1800 vec<pre_expr> exprs;
1801 pre_expr expr;
1802 int i;
1804 if (gimple_seq_empty_p (phi_nodes (phiblock)))
1806 bitmap_set_copy (dest, set);
1807 return;
1810 exprs = sorted_array_from_bitmap_set (set);
1811 FOR_EACH_VEC_ELT (exprs, i, expr)
1813 pre_expr translated;
1814 translated = phi_translate (expr, set, NULL, pred, phiblock);
1815 if (!translated)
1816 continue;
1818 /* We might end up with multiple expressions from SET being
1819 translated to the same value. In this case we do not want
1820 to retain the NARY or REFERENCE expression but prefer a NAME
1821 which would be the leader. */
1822 if (translated->kind == NAME)
1823 bitmap_value_replace_in_set (dest, translated);
1824 else
1825 bitmap_value_insert_into_set (dest, translated);
1827 exprs.release ();
1830 /* Find the leader for a value (i.e., the name representing that
1831 value) in a given set, and return it. Return NULL if no leader
1832 is found. */
1834 static pre_expr
1835 bitmap_find_leader (bitmap_set_t set, unsigned int val)
1837 if (value_id_constant_p (val))
1839 unsigned int i;
1840 bitmap_iterator bi;
1841 bitmap exprset = value_expressions[val];
1843 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
1845 pre_expr expr = expression_for_id (i);
1846 if (expr->kind == CONSTANT)
1847 return expr;
1850 if (bitmap_set_contains_value (set, val))
1852 /* Rather than walk the entire bitmap of expressions, and see
1853 whether any of them has the value we are looking for, we look
1854 at the reverse mapping, which tells us the set of expressions
1855 that have a given value (IE value->expressions with that
1856 value) and see if any of those expressions are in our set.
1857 The number of expressions per value is usually significantly
1858 less than the number of expressions in the set. In fact, for
1859 large testcases, doing it this way is roughly 5-10x faster
1860 than walking the bitmap.
1861 If this is somehow a significant lose for some cases, we can
1862 choose which set to walk based on which set is smaller. */
1863 unsigned int i;
1864 bitmap_iterator bi;
1865 bitmap exprset = value_expressions[val];
1867 EXECUTE_IF_AND_IN_BITMAP (exprset, &set->expressions, 0, i, bi)
1868 return expression_for_id (i);
1870 return NULL;
1873 /* Determine if EXPR, a memory expression, is ANTIC_IN at the top of
1874 BLOCK by seeing if it is not killed in the block. Note that we are
1875 only determining whether there is a store that kills it. Because
1876 of the order in which clean iterates over values, we are guaranteed
1877 that altered operands will have caused us to be eliminated from the
1878 ANTIC_IN set already. */
1880 static bool
1881 value_dies_in_block_x (pre_expr expr, basic_block block)
1883 tree vuse = PRE_EXPR_REFERENCE (expr)->vuse;
1884 vn_reference_t refx = PRE_EXPR_REFERENCE (expr);
1885 gimple def;
1886 gimple_stmt_iterator gsi;
1887 unsigned id = get_expression_id (expr);
1888 bool res = false;
1889 ao_ref ref;
1891 if (!vuse)
1892 return false;
1894 /* Lookup a previously calculated result. */
1895 if (EXPR_DIES (block)
1896 && bitmap_bit_p (EXPR_DIES (block), id * 2))
1897 return bitmap_bit_p (EXPR_DIES (block), id * 2 + 1);
1899 /* A memory expression {e, VUSE} dies in the block if there is a
1900 statement that may clobber e. If, starting statement walk from the
1901 top of the basic block, a statement uses VUSE there can be no kill
1902 inbetween that use and the original statement that loaded {e, VUSE},
1903 so we can stop walking. */
1904 ref.base = NULL_TREE;
1905 for (gsi = gsi_start_bb (block); !gsi_end_p (gsi); gsi_next (&gsi))
1907 tree def_vuse, def_vdef;
1908 def = gsi_stmt (gsi);
1909 def_vuse = gimple_vuse (def);
1910 def_vdef = gimple_vdef (def);
1912 /* Not a memory statement. */
1913 if (!def_vuse)
1914 continue;
1916 /* Not a may-def. */
1917 if (!def_vdef)
1919 /* A load with the same VUSE, we're done. */
1920 if (def_vuse == vuse)
1921 break;
1923 continue;
1926 /* Init ref only if we really need it. */
1927 if (ref.base == NULL_TREE
1928 && !ao_ref_init_from_vn_reference (&ref, refx->set, refx->type,
1929 refx->operands))
1931 res = true;
1932 break;
1934 /* If the statement may clobber expr, it dies. */
1935 if (stmt_may_clobber_ref_p_1 (def, &ref))
1937 res = true;
1938 break;
1942 /* Remember the result. */
1943 if (!EXPR_DIES (block))
1944 EXPR_DIES (block) = BITMAP_ALLOC (&grand_bitmap_obstack);
1945 bitmap_set_bit (EXPR_DIES (block), id * 2);
1946 if (res)
1947 bitmap_set_bit (EXPR_DIES (block), id * 2 + 1);
1949 return res;
1953 /* Determine if OP is valid in SET1 U SET2, which it is when the union
1954 contains its value-id. */
1956 static bool
1957 op_valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, tree op)
1959 if (op && TREE_CODE (op) == SSA_NAME)
1961 unsigned int value_id = VN_INFO (op)->value_id;
1962 if (!(bitmap_set_contains_value (set1, value_id)
1963 || (set2 && bitmap_set_contains_value (set2, value_id))))
1964 return false;
1966 return true;
1969 /* Determine if the expression EXPR is valid in SET1 U SET2.
1970 ONLY SET2 CAN BE NULL.
1971 This means that we have a leader for each part of the expression
1972 (if it consists of values), or the expression is an SSA_NAME.
1973 For loads/calls, we also see if the vuse is killed in this block. */
1975 static bool
1976 valid_in_sets (bitmap_set_t set1, bitmap_set_t set2, pre_expr expr,
1977 basic_block block)
1979 switch (expr->kind)
1981 case NAME:
1982 return bitmap_find_leader (AVAIL_OUT (block),
1983 get_expr_value_id (expr)) != NULL;
1984 case NARY:
1986 unsigned int i;
1987 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
1988 for (i = 0; i < nary->length; i++)
1989 if (!op_valid_in_sets (set1, set2, nary->op[i]))
1990 return false;
1991 return true;
1993 break;
1994 case REFERENCE:
1996 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
1997 vn_reference_op_t vro;
1998 unsigned int i;
2000 FOR_EACH_VEC_ELT (ref->operands, i, vro)
2002 if (!op_valid_in_sets (set1, set2, vro->op0)
2003 || !op_valid_in_sets (set1, set2, vro->op1)
2004 || !op_valid_in_sets (set1, set2, vro->op2))
2005 return false;
2007 return true;
2009 default:
2010 gcc_unreachable ();
2014 /* Clean the set of expressions that are no longer valid in SET1 or
2015 SET2. This means expressions that are made up of values we have no
2016 leaders for in SET1 or SET2. This version is used for partial
2017 anticipation, which means it is not valid in either ANTIC_IN or
2018 PA_IN. */
2020 static void
2021 dependent_clean (bitmap_set_t set1, bitmap_set_t set2, basic_block block)
2023 vec<pre_expr> exprs = sorted_array_from_bitmap_set (set1);
2024 pre_expr expr;
2025 int i;
2027 FOR_EACH_VEC_ELT (exprs, i, expr)
2029 if (!valid_in_sets (set1, set2, expr, block))
2030 bitmap_remove_from_set (set1, expr);
2032 exprs.release ();
2035 /* Clean the set of expressions that are no longer valid in SET. This
2036 means expressions that are made up of values we have no leaders for
2037 in SET. */
2039 static void
2040 clean (bitmap_set_t set, basic_block block)
2042 vec<pre_expr> exprs = sorted_array_from_bitmap_set (set);
2043 pre_expr expr;
2044 int i;
2046 FOR_EACH_VEC_ELT (exprs, i, expr)
2048 if (!valid_in_sets (set, NULL, expr, block))
2049 bitmap_remove_from_set (set, expr);
2051 exprs.release ();
2054 /* Clean the set of expressions that are no longer valid in SET because
2055 they are clobbered in BLOCK or because they trap and may not be executed. */
2057 static void
2058 prune_clobbered_mems (bitmap_set_t set, basic_block block)
2060 bitmap_iterator bi;
2061 unsigned i;
2063 FOR_EACH_EXPR_ID_IN_SET (set, i, bi)
2065 pre_expr expr = expression_for_id (i);
2066 if (expr->kind == REFERENCE)
2068 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2069 if (ref->vuse)
2071 gimple def_stmt = SSA_NAME_DEF_STMT (ref->vuse);
2072 if (!gimple_nop_p (def_stmt)
2073 && ((gimple_bb (def_stmt) != block
2074 && !dominated_by_p (CDI_DOMINATORS,
2075 block, gimple_bb (def_stmt)))
2076 || (gimple_bb (def_stmt) == block
2077 && value_dies_in_block_x (expr, block))))
2078 bitmap_remove_from_set (set, expr);
2081 else if (expr->kind == NARY)
2083 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2084 /* If the NARY may trap make sure the block does not contain
2085 a possible exit point.
2086 ??? This is overly conservative if we translate AVAIL_OUT
2087 as the available expression might be after the exit point. */
2088 if (BB_MAY_NOTRETURN (block)
2089 && vn_nary_may_trap (nary))
2090 bitmap_remove_from_set (set, expr);
2095 static sbitmap has_abnormal_preds;
2097 /* List of blocks that may have changed during ANTIC computation and
2098 thus need to be iterated over. */
2100 static sbitmap changed_blocks;
2102 /* Decide whether to defer a block for a later iteration, or PHI
2103 translate SOURCE to DEST using phis in PHIBLOCK. Return false if we
2104 should defer the block, and true if we processed it. */
2106 static bool
2107 defer_or_phi_translate_block (bitmap_set_t dest, bitmap_set_t source,
2108 basic_block block, basic_block phiblock)
2110 if (!BB_VISITED (phiblock))
2112 bitmap_set_bit (changed_blocks, block->index);
2113 BB_VISITED (block) = 0;
2114 BB_DEFERRED (block) = 1;
2115 return false;
2117 else
2118 phi_translate_set (dest, source, block, phiblock);
2119 return true;
2122 /* Compute the ANTIC set for BLOCK.
2124 If succs(BLOCK) > 1 then
2125 ANTIC_OUT[BLOCK] = intersection of ANTIC_IN[b] for all succ(BLOCK)
2126 else if succs(BLOCK) == 1 then
2127 ANTIC_OUT[BLOCK] = phi_translate (ANTIC_IN[succ(BLOCK)])
2129 ANTIC_IN[BLOCK] = clean(ANTIC_OUT[BLOCK] U EXP_GEN[BLOCK] - TMP_GEN[BLOCK])
2132 static bool
2133 compute_antic_aux (basic_block block, bool block_has_abnormal_pred_edge)
2135 bool changed = false;
2136 bitmap_set_t S, old, ANTIC_OUT;
2137 bitmap_iterator bi;
2138 unsigned int bii;
2139 edge e;
2140 edge_iterator ei;
2142 old = ANTIC_OUT = S = NULL;
2143 BB_VISITED (block) = 1;
2145 /* If any edges from predecessors are abnormal, antic_in is empty,
2146 so do nothing. */
2147 if (block_has_abnormal_pred_edge)
2148 goto maybe_dump_sets;
2150 old = ANTIC_IN (block);
2151 ANTIC_OUT = bitmap_set_new ();
2153 /* If the block has no successors, ANTIC_OUT is empty. */
2154 if (EDGE_COUNT (block->succs) == 0)
2156 /* If we have one successor, we could have some phi nodes to
2157 translate through. */
2158 else if (single_succ_p (block))
2160 basic_block succ_bb = single_succ (block);
2162 /* We trade iterations of the dataflow equations for having to
2163 phi translate the maximal set, which is incredibly slow
2164 (since the maximal set often has 300+ members, even when you
2165 have a small number of blocks).
2166 Basically, we defer the computation of ANTIC for this block
2167 until we have processed it's successor, which will inevitably
2168 have a *much* smaller set of values to phi translate once
2169 clean has been run on it.
2170 The cost of doing this is that we technically perform more
2171 iterations, however, they are lower cost iterations.
2173 Timings for PRE on tramp3d-v4:
2174 without maximal set fix: 11 seconds
2175 with maximal set fix/without deferring: 26 seconds
2176 with maximal set fix/with deferring: 11 seconds
2179 if (!defer_or_phi_translate_block (ANTIC_OUT, ANTIC_IN (succ_bb),
2180 block, succ_bb))
2182 changed = true;
2183 goto maybe_dump_sets;
2186 /* If we have multiple successors, we take the intersection of all of
2187 them. Note that in the case of loop exit phi nodes, we may have
2188 phis to translate through. */
2189 else
2191 size_t i;
2192 basic_block bprime, first = NULL;
2194 auto_vec<basic_block> worklist (EDGE_COUNT (block->succs));
2195 FOR_EACH_EDGE (e, ei, block->succs)
2197 if (!first
2198 && BB_VISITED (e->dest))
2199 first = e->dest;
2200 else if (BB_VISITED (e->dest))
2201 worklist.quick_push (e->dest);
2204 /* Of multiple successors we have to have visited one already. */
2205 if (!first)
2207 bitmap_set_bit (changed_blocks, block->index);
2208 BB_VISITED (block) = 0;
2209 BB_DEFERRED (block) = 1;
2210 changed = true;
2211 goto maybe_dump_sets;
2214 if (!gimple_seq_empty_p (phi_nodes (first)))
2215 phi_translate_set (ANTIC_OUT, ANTIC_IN (first), block, first);
2216 else
2217 bitmap_set_copy (ANTIC_OUT, ANTIC_IN (first));
2219 FOR_EACH_VEC_ELT (worklist, i, bprime)
2221 if (!gimple_seq_empty_p (phi_nodes (bprime)))
2223 bitmap_set_t tmp = bitmap_set_new ();
2224 phi_translate_set (tmp, ANTIC_IN (bprime), block, bprime);
2225 bitmap_set_and (ANTIC_OUT, tmp);
2226 bitmap_set_free (tmp);
2228 else
2229 bitmap_set_and (ANTIC_OUT, ANTIC_IN (bprime));
2233 /* Prune expressions that are clobbered in block and thus become
2234 invalid if translated from ANTIC_OUT to ANTIC_IN. */
2235 prune_clobbered_mems (ANTIC_OUT, block);
2237 /* Generate ANTIC_OUT - TMP_GEN. */
2238 S = bitmap_set_subtract (ANTIC_OUT, TMP_GEN (block));
2240 /* Start ANTIC_IN with EXP_GEN - TMP_GEN. */
2241 ANTIC_IN (block) = bitmap_set_subtract (EXP_GEN (block),
2242 TMP_GEN (block));
2244 /* Then union in the ANTIC_OUT - TMP_GEN values,
2245 to get ANTIC_OUT U EXP_GEN - TMP_GEN */
2246 FOR_EACH_EXPR_ID_IN_SET (S, bii, bi)
2247 bitmap_value_insert_into_set (ANTIC_IN (block),
2248 expression_for_id (bii));
2250 clean (ANTIC_IN (block), block);
2252 if (!bitmap_set_equal (old, ANTIC_IN (block)))
2254 changed = true;
2255 bitmap_set_bit (changed_blocks, block->index);
2256 FOR_EACH_EDGE (e, ei, block->preds)
2257 bitmap_set_bit (changed_blocks, e->src->index);
2259 else
2260 bitmap_clear_bit (changed_blocks, block->index);
2262 maybe_dump_sets:
2263 if (dump_file && (dump_flags & TDF_DETAILS))
2265 if (!BB_DEFERRED (block) || BB_VISITED (block))
2267 if (ANTIC_OUT)
2268 print_bitmap_set (dump_file, ANTIC_OUT, "ANTIC_OUT", block->index);
2270 print_bitmap_set (dump_file, ANTIC_IN (block), "ANTIC_IN",
2271 block->index);
2273 if (S)
2274 print_bitmap_set (dump_file, S, "S", block->index);
2276 else
2278 fprintf (dump_file,
2279 "Block %d was deferred for a future iteration.\n",
2280 block->index);
2283 if (old)
2284 bitmap_set_free (old);
2285 if (S)
2286 bitmap_set_free (S);
2287 if (ANTIC_OUT)
2288 bitmap_set_free (ANTIC_OUT);
2289 return changed;
2292 /* Compute PARTIAL_ANTIC for BLOCK.
2294 If succs(BLOCK) > 1 then
2295 PA_OUT[BLOCK] = value wise union of PA_IN[b] + all ANTIC_IN not
2296 in ANTIC_OUT for all succ(BLOCK)
2297 else if succs(BLOCK) == 1 then
2298 PA_OUT[BLOCK] = phi_translate (PA_IN[succ(BLOCK)])
2300 PA_IN[BLOCK] = dependent_clean(PA_OUT[BLOCK] - TMP_GEN[BLOCK]
2301 - ANTIC_IN[BLOCK])
2304 static bool
2305 compute_partial_antic_aux (basic_block block,
2306 bool block_has_abnormal_pred_edge)
2308 bool changed = false;
2309 bitmap_set_t old_PA_IN;
2310 bitmap_set_t PA_OUT;
2311 edge e;
2312 edge_iterator ei;
2313 unsigned long max_pa = PARAM_VALUE (PARAM_MAX_PARTIAL_ANTIC_LENGTH);
2315 old_PA_IN = PA_OUT = NULL;
2317 /* If any edges from predecessors are abnormal, antic_in is empty,
2318 so do nothing. */
2319 if (block_has_abnormal_pred_edge)
2320 goto maybe_dump_sets;
2322 /* If there are too many partially anticipatable values in the
2323 block, phi_translate_set can take an exponential time: stop
2324 before the translation starts. */
2325 if (max_pa
2326 && single_succ_p (block)
2327 && bitmap_count_bits (&PA_IN (single_succ (block))->values) > max_pa)
2328 goto maybe_dump_sets;
2330 old_PA_IN = PA_IN (block);
2331 PA_OUT = bitmap_set_new ();
2333 /* If the block has no successors, ANTIC_OUT is empty. */
2334 if (EDGE_COUNT (block->succs) == 0)
2336 /* If we have one successor, we could have some phi nodes to
2337 translate through. Note that we can't phi translate across DFS
2338 back edges in partial antic, because it uses a union operation on
2339 the successors. For recurrences like IV's, we will end up
2340 generating a new value in the set on each go around (i + 3 (VH.1)
2341 VH.1 + 1 (VH.2), VH.2 + 1 (VH.3), etc), forever. */
2342 else if (single_succ_p (block))
2344 basic_block succ = single_succ (block);
2345 if (!(single_succ_edge (block)->flags & EDGE_DFS_BACK))
2346 phi_translate_set (PA_OUT, PA_IN (succ), block, succ);
2348 /* If we have multiple successors, we take the union of all of
2349 them. */
2350 else
2352 size_t i;
2353 basic_block bprime;
2355 auto_vec<basic_block> 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->dest);
2362 if (worklist.length () > 0)
2364 FOR_EACH_VEC_ELT (worklist, i, bprime)
2366 unsigned int i;
2367 bitmap_iterator bi;
2369 FOR_EACH_EXPR_ID_IN_SET (ANTIC_IN (bprime), i, bi)
2370 bitmap_value_insert_into_set (PA_OUT,
2371 expression_for_id (i));
2372 if (!gimple_seq_empty_p (phi_nodes (bprime)))
2374 bitmap_set_t pa_in = bitmap_set_new ();
2375 phi_translate_set (pa_in, PA_IN (bprime), block, bprime);
2376 FOR_EACH_EXPR_ID_IN_SET (pa_in, i, bi)
2377 bitmap_value_insert_into_set (PA_OUT,
2378 expression_for_id (i));
2379 bitmap_set_free (pa_in);
2381 else
2382 FOR_EACH_EXPR_ID_IN_SET (PA_IN (bprime), i, bi)
2383 bitmap_value_insert_into_set (PA_OUT,
2384 expression_for_id (i));
2389 /* Prune expressions that are clobbered in block and thus become
2390 invalid if translated from PA_OUT to PA_IN. */
2391 prune_clobbered_mems (PA_OUT, block);
2393 /* PA_IN starts with PA_OUT - TMP_GEN.
2394 Then we subtract things from ANTIC_IN. */
2395 PA_IN (block) = bitmap_set_subtract (PA_OUT, TMP_GEN (block));
2397 /* For partial antic, we want to put back in the phi results, since
2398 we will properly avoid making them partially antic over backedges. */
2399 bitmap_ior_into (&PA_IN (block)->values, &PHI_GEN (block)->values);
2400 bitmap_ior_into (&PA_IN (block)->expressions, &PHI_GEN (block)->expressions);
2402 /* PA_IN[block] = PA_IN[block] - ANTIC_IN[block] */
2403 bitmap_set_subtract_values (PA_IN (block), ANTIC_IN (block));
2405 dependent_clean (PA_IN (block), ANTIC_IN (block), block);
2407 if (!bitmap_set_equal (old_PA_IN, PA_IN (block)))
2409 changed = true;
2410 bitmap_set_bit (changed_blocks, block->index);
2411 FOR_EACH_EDGE (e, ei, block->preds)
2412 bitmap_set_bit (changed_blocks, e->src->index);
2414 else
2415 bitmap_clear_bit (changed_blocks, block->index);
2417 maybe_dump_sets:
2418 if (dump_file && (dump_flags & TDF_DETAILS))
2420 if (PA_OUT)
2421 print_bitmap_set (dump_file, PA_OUT, "PA_OUT", block->index);
2423 print_bitmap_set (dump_file, PA_IN (block), "PA_IN", block->index);
2425 if (old_PA_IN)
2426 bitmap_set_free (old_PA_IN);
2427 if (PA_OUT)
2428 bitmap_set_free (PA_OUT);
2429 return changed;
2432 /* Compute ANTIC and partial ANTIC sets. */
2434 static void
2435 compute_antic (void)
2437 bool changed = true;
2438 int num_iterations = 0;
2439 basic_block block;
2440 int i;
2442 /* If any predecessor edges are abnormal, we punt, so antic_in is empty.
2443 We pre-build the map of blocks with incoming abnormal edges here. */
2444 has_abnormal_preds = sbitmap_alloc (last_basic_block_for_fn (cfun));
2445 bitmap_clear (has_abnormal_preds);
2447 FOR_ALL_BB_FN (block, cfun)
2449 edge_iterator ei;
2450 edge e;
2452 FOR_EACH_EDGE (e, ei, block->preds)
2454 e->flags &= ~EDGE_DFS_BACK;
2455 if (e->flags & EDGE_ABNORMAL)
2457 bitmap_set_bit (has_abnormal_preds, block->index);
2458 break;
2462 BB_VISITED (block) = 0;
2463 BB_DEFERRED (block) = 0;
2465 /* While we are here, give empty ANTIC_IN sets to each block. */
2466 ANTIC_IN (block) = bitmap_set_new ();
2467 PA_IN (block) = bitmap_set_new ();
2470 /* At the exit block we anticipate nothing. */
2471 BB_VISITED (EXIT_BLOCK_PTR_FOR_FN (cfun)) = 1;
2473 changed_blocks = sbitmap_alloc (last_basic_block_for_fn (cfun) + 1);
2474 bitmap_ones (changed_blocks);
2475 while (changed)
2477 if (dump_file && (dump_flags & TDF_DETAILS))
2478 fprintf (dump_file, "Starting iteration %d\n", num_iterations);
2479 /* ??? We need to clear our PHI translation cache here as the
2480 ANTIC sets shrink and we restrict valid translations to
2481 those having operands with leaders in ANTIC. Same below
2482 for PA ANTIC computation. */
2483 num_iterations++;
2484 changed = false;
2485 for (i = postorder_num - 1; i >= 0; i--)
2487 if (bitmap_bit_p (changed_blocks, postorder[i]))
2489 basic_block block = BASIC_BLOCK_FOR_FN (cfun, postorder[i]);
2490 changed |= compute_antic_aux (block,
2491 bitmap_bit_p (has_abnormal_preds,
2492 block->index));
2495 /* Theoretically possible, but *highly* unlikely. */
2496 gcc_checking_assert (num_iterations < 500);
2499 statistics_histogram_event (cfun, "compute_antic iterations",
2500 num_iterations);
2502 if (do_partial_partial)
2504 bitmap_ones (changed_blocks);
2505 mark_dfs_back_edges ();
2506 num_iterations = 0;
2507 changed = true;
2508 while (changed)
2510 if (dump_file && (dump_flags & TDF_DETAILS))
2511 fprintf (dump_file, "Starting iteration %d\n", num_iterations);
2512 num_iterations++;
2513 changed = false;
2514 for (i = postorder_num - 1 ; i >= 0; i--)
2516 if (bitmap_bit_p (changed_blocks, postorder[i]))
2518 basic_block block = BASIC_BLOCK_FOR_FN (cfun, postorder[i]);
2519 changed
2520 |= compute_partial_antic_aux (block,
2521 bitmap_bit_p (has_abnormal_preds,
2522 block->index));
2525 /* Theoretically possible, but *highly* unlikely. */
2526 gcc_checking_assert (num_iterations < 500);
2528 statistics_histogram_event (cfun, "compute_partial_antic iterations",
2529 num_iterations);
2531 sbitmap_free (has_abnormal_preds);
2532 sbitmap_free (changed_blocks);
2536 /* Inserted expressions are placed onto this worklist, which is used
2537 for performing quick dead code elimination of insertions we made
2538 that didn't turn out to be necessary. */
2539 static bitmap inserted_exprs;
2541 /* The actual worker for create_component_ref_by_pieces. */
2543 static tree
2544 create_component_ref_by_pieces_1 (basic_block block, vn_reference_t ref,
2545 unsigned int *operand, gimple_seq *stmts)
2547 vn_reference_op_t currop = &ref->operands[*operand];
2548 tree genop;
2549 ++*operand;
2550 switch (currop->opcode)
2552 case CALL_EXPR:
2554 tree folded, sc = NULL_TREE;
2555 unsigned int nargs = 0;
2556 tree fn, *args;
2557 if (TREE_CODE (currop->op0) == FUNCTION_DECL)
2558 fn = currop->op0;
2559 else
2560 fn = find_or_generate_expression (block, currop->op0, stmts);
2561 if (!fn)
2562 return NULL_TREE;
2563 if (currop->op1)
2565 sc = find_or_generate_expression (block, currop->op1, stmts);
2566 if (!sc)
2567 return NULL_TREE;
2569 args = XNEWVEC (tree, ref->operands.length () - 1);
2570 while (*operand < ref->operands.length ())
2572 args[nargs] = create_component_ref_by_pieces_1 (block, ref,
2573 operand, stmts);
2574 if (!args[nargs])
2575 return NULL_TREE;
2576 nargs++;
2578 folded = build_call_array (currop->type,
2579 (TREE_CODE (fn) == FUNCTION_DECL
2580 ? build_fold_addr_expr (fn) : fn),
2581 nargs, args);
2582 free (args);
2583 if (sc)
2584 CALL_EXPR_STATIC_CHAIN (folded) = sc;
2585 return folded;
2588 case MEM_REF:
2590 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2591 stmts);
2592 if (!baseop)
2593 return NULL_TREE;
2594 tree offset = currop->op0;
2595 if (TREE_CODE (baseop) == ADDR_EXPR
2596 && handled_component_p (TREE_OPERAND (baseop, 0)))
2598 HOST_WIDE_INT off;
2599 tree base;
2600 base = get_addr_base_and_unit_offset (TREE_OPERAND (baseop, 0),
2601 &off);
2602 gcc_assert (base);
2603 offset = int_const_binop (PLUS_EXPR, offset,
2604 build_int_cst (TREE_TYPE (offset),
2605 off));
2606 baseop = build_fold_addr_expr (base);
2608 return fold_build2 (MEM_REF, currop->type, baseop, offset);
2611 case TARGET_MEM_REF:
2613 tree genop0 = NULL_TREE, genop1 = NULL_TREE;
2614 vn_reference_op_t nextop = &ref->operands[++*operand];
2615 tree baseop = create_component_ref_by_pieces_1 (block, ref, operand,
2616 stmts);
2617 if (!baseop)
2618 return NULL_TREE;
2619 if (currop->op0)
2621 genop0 = find_or_generate_expression (block, currop->op0, stmts);
2622 if (!genop0)
2623 return NULL_TREE;
2625 if (nextop->op0)
2627 genop1 = find_or_generate_expression (block, nextop->op0, stmts);
2628 if (!genop1)
2629 return NULL_TREE;
2631 return build5 (TARGET_MEM_REF, currop->type,
2632 baseop, currop->op2, genop0, currop->op1, genop1);
2635 case ADDR_EXPR:
2636 if (currop->op0)
2638 gcc_assert (is_gimple_min_invariant (currop->op0));
2639 return currop->op0;
2641 /* Fallthrough. */
2642 case REALPART_EXPR:
2643 case IMAGPART_EXPR:
2644 case VIEW_CONVERT_EXPR:
2646 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2647 stmts);
2648 if (!genop0)
2649 return NULL_TREE;
2650 return fold_build1 (currop->opcode, currop->type, genop0);
2653 case WITH_SIZE_EXPR:
2655 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2656 stmts);
2657 if (!genop0)
2658 return NULL_TREE;
2659 tree genop1 = find_or_generate_expression (block, currop->op0, stmts);
2660 if (!genop1)
2661 return NULL_TREE;
2662 return fold_build2 (currop->opcode, currop->type, genop0, genop1);
2665 case BIT_FIELD_REF:
2667 tree genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2668 stmts);
2669 if (!genop0)
2670 return NULL_TREE;
2671 tree op1 = currop->op0;
2672 tree op2 = currop->op1;
2673 return fold_build3 (BIT_FIELD_REF, currop->type, genop0, op1, op2);
2676 /* For array ref vn_reference_op's, operand 1 of the array ref
2677 is op0 of the reference op and operand 3 of the array ref is
2678 op1. */
2679 case ARRAY_RANGE_REF:
2680 case ARRAY_REF:
2682 tree genop0;
2683 tree genop1 = currop->op0;
2684 tree genop2 = currop->op1;
2685 tree genop3 = currop->op2;
2686 genop0 = create_component_ref_by_pieces_1 (block, ref, operand,
2687 stmts);
2688 if (!genop0)
2689 return NULL_TREE;
2690 genop1 = find_or_generate_expression (block, genop1, stmts);
2691 if (!genop1)
2692 return NULL_TREE;
2693 if (genop2)
2695 tree domain_type = TYPE_DOMAIN (TREE_TYPE (genop0));
2696 /* Drop zero minimum index if redundant. */
2697 if (integer_zerop (genop2)
2698 && (!domain_type
2699 || integer_zerop (TYPE_MIN_VALUE (domain_type))))
2700 genop2 = NULL_TREE;
2701 else
2703 genop2 = find_or_generate_expression (block, genop2, stmts);
2704 if (!genop2)
2705 return NULL_TREE;
2708 if (genop3)
2710 tree elmt_type = TREE_TYPE (TREE_TYPE (genop0));
2711 /* We can't always put a size in units of the element alignment
2712 here as the element alignment may be not visible. See
2713 PR43783. Simply drop the element size for constant
2714 sizes. */
2715 if (tree_int_cst_equal (genop3, TYPE_SIZE_UNIT (elmt_type)))
2716 genop3 = NULL_TREE;
2717 else
2719 genop3 = size_binop (EXACT_DIV_EXPR, genop3,
2720 size_int (TYPE_ALIGN_UNIT (elmt_type)));
2721 genop3 = find_or_generate_expression (block, genop3, stmts);
2722 if (!genop3)
2723 return NULL_TREE;
2726 return build4 (currop->opcode, currop->type, genop0, genop1,
2727 genop2, genop3);
2729 case COMPONENT_REF:
2731 tree op0;
2732 tree op1;
2733 tree genop2 = currop->op1;
2734 op0 = create_component_ref_by_pieces_1 (block, ref, operand, stmts);
2735 if (!op0)
2736 return NULL_TREE;
2737 /* op1 should be a FIELD_DECL, which are represented by themselves. */
2738 op1 = currop->op0;
2739 if (genop2)
2741 genop2 = find_or_generate_expression (block, genop2, stmts);
2742 if (!genop2)
2743 return NULL_TREE;
2745 return fold_build3 (COMPONENT_REF, TREE_TYPE (op1), op0, op1, genop2);
2748 case SSA_NAME:
2750 genop = find_or_generate_expression (block, currop->op0, stmts);
2751 return genop;
2753 case STRING_CST:
2754 case INTEGER_CST:
2755 case COMPLEX_CST:
2756 case VECTOR_CST:
2757 case REAL_CST:
2758 case CONSTRUCTOR:
2759 case VAR_DECL:
2760 case PARM_DECL:
2761 case CONST_DECL:
2762 case RESULT_DECL:
2763 case FUNCTION_DECL:
2764 return currop->op0;
2766 default:
2767 gcc_unreachable ();
2771 /* For COMPONENT_REF's and ARRAY_REF's, we can't have any intermediates for the
2772 COMPONENT_REF or MEM_REF or ARRAY_REF portion, because we'd end up with
2773 trying to rename aggregates into ssa form directly, which is a no no.
2775 Thus, this routine doesn't create temporaries, it just builds a
2776 single access expression for the array, calling
2777 find_or_generate_expression to build the innermost pieces.
2779 This function is a subroutine of create_expression_by_pieces, and
2780 should not be called on it's own unless you really know what you
2781 are doing. */
2783 static tree
2784 create_component_ref_by_pieces (basic_block block, vn_reference_t ref,
2785 gimple_seq *stmts)
2787 unsigned int op = 0;
2788 return create_component_ref_by_pieces_1 (block, ref, &op, stmts);
2791 /* Find a simple leader for an expression, or generate one using
2792 create_expression_by_pieces from a NARY expression for the value.
2793 BLOCK is the basic_block we are looking for leaders in.
2794 OP is the tree expression to find a leader for or generate.
2795 Returns the leader or NULL_TREE on failure. */
2797 static tree
2798 find_or_generate_expression (basic_block block, tree op, gimple_seq *stmts)
2800 pre_expr expr = get_or_alloc_expr_for (op);
2801 unsigned int lookfor = get_expr_value_id (expr);
2802 pre_expr leader = bitmap_find_leader (AVAIL_OUT (block), lookfor);
2803 if (leader)
2805 if (leader->kind == NAME)
2806 return PRE_EXPR_NAME (leader);
2807 else if (leader->kind == CONSTANT)
2808 return PRE_EXPR_CONSTANT (leader);
2810 /* Defer. */
2811 return NULL_TREE;
2814 /* It must be a complex expression, so generate it recursively. Note
2815 that this is only necessary to handle gcc.dg/tree-ssa/ssa-pre28.c
2816 where the insert algorithm fails to insert a required expression. */
2817 bitmap exprset = value_expressions[lookfor];
2818 bitmap_iterator bi;
2819 unsigned int i;
2820 EXECUTE_IF_SET_IN_BITMAP (exprset, 0, i, bi)
2822 pre_expr temp = expression_for_id (i);
2823 /* We cannot insert random REFERENCE expressions at arbitrary
2824 places. We can insert NARYs which eventually re-materializes
2825 its operand values. */
2826 if (temp->kind == NARY)
2827 return create_expression_by_pieces (block, temp, stmts,
2828 get_expr_type (expr));
2831 /* Defer. */
2832 return NULL_TREE;
2835 #define NECESSARY GF_PLF_1
2837 /* Create an expression in pieces, so that we can handle very complex
2838 expressions that may be ANTIC, but not necessary GIMPLE.
2839 BLOCK is the basic block the expression will be inserted into,
2840 EXPR is the expression to insert (in value form)
2841 STMTS is a statement list to append the necessary insertions into.
2843 This function will die if we hit some value that shouldn't be
2844 ANTIC but is (IE there is no leader for it, or its components).
2845 The function returns NULL_TREE in case a different antic expression
2846 has to be inserted first.
2847 This function may also generate expressions that are themselves
2848 partially or fully redundant. Those that are will be either made
2849 fully redundant during the next iteration of insert (for partially
2850 redundant ones), or eliminated by eliminate (for fully redundant
2851 ones). */
2853 static tree
2854 create_expression_by_pieces (basic_block block, pre_expr expr,
2855 gimple_seq *stmts, tree type)
2857 tree name;
2858 tree folded;
2859 gimple_seq forced_stmts = NULL;
2860 unsigned int value_id;
2861 gimple_stmt_iterator gsi;
2862 tree exprtype = type ? type : get_expr_type (expr);
2863 pre_expr nameexpr;
2864 gimple newstmt;
2866 switch (expr->kind)
2868 /* We may hit the NAME/CONSTANT case if we have to convert types
2869 that value numbering saw through. */
2870 case NAME:
2871 folded = PRE_EXPR_NAME (expr);
2872 break;
2873 case CONSTANT:
2874 folded = PRE_EXPR_CONSTANT (expr);
2875 break;
2876 case REFERENCE:
2878 vn_reference_t ref = PRE_EXPR_REFERENCE (expr);
2879 folded = create_component_ref_by_pieces (block, ref, stmts);
2880 if (!folded)
2881 return NULL_TREE;
2883 break;
2884 case NARY:
2886 vn_nary_op_t nary = PRE_EXPR_NARY (expr);
2887 tree *genop = XALLOCAVEC (tree, nary->length);
2888 unsigned i;
2889 for (i = 0; i < nary->length; ++i)
2891 genop[i] = find_or_generate_expression (block, nary->op[i], stmts);
2892 if (!genop[i])
2893 return NULL_TREE;
2894 /* Ensure genop[] is properly typed for POINTER_PLUS_EXPR. It
2895 may have conversions stripped. */
2896 if (nary->opcode == POINTER_PLUS_EXPR)
2898 if (i == 0)
2899 genop[i] = fold_convert (nary->type, genop[i]);
2900 else if (i == 1)
2901 genop[i] = convert_to_ptrofftype (genop[i]);
2903 else
2904 genop[i] = fold_convert (TREE_TYPE (nary->op[i]), genop[i]);
2906 if (nary->opcode == CONSTRUCTOR)
2908 vec<constructor_elt, va_gc> *elts = NULL;
2909 for (i = 0; i < nary->length; ++i)
2910 CONSTRUCTOR_APPEND_ELT (elts, NULL_TREE, genop[i]);
2911 folded = build_constructor (nary->type, elts);
2913 else
2915 switch (nary->length)
2917 case 1:
2918 folded = fold_build1 (nary->opcode, nary->type,
2919 genop[0]);
2920 break;
2921 case 2:
2922 folded = fold_build2 (nary->opcode, nary->type,
2923 genop[0], genop[1]);
2924 break;
2925 case 3:
2926 folded = fold_build3 (nary->opcode, nary->type,
2927 genop[0], genop[1], genop[2]);
2928 break;
2929 default:
2930 gcc_unreachable ();
2934 break;
2935 default:
2936 gcc_unreachable ();
2939 if (!useless_type_conversion_p (exprtype, TREE_TYPE (folded)))
2940 folded = fold_convert (exprtype, folded);
2942 /* Force the generated expression to be a sequence of GIMPLE
2943 statements.
2944 We have to call unshare_expr because force_gimple_operand may
2945 modify the tree we pass to it. */
2946 folded = force_gimple_operand (unshare_expr (folded), &forced_stmts,
2947 false, NULL);
2949 /* If we have any intermediate expressions to the value sets, add them
2950 to the value sets and chain them in the instruction stream. */
2951 if (forced_stmts)
2953 gsi = gsi_start (forced_stmts);
2954 for (; !gsi_end_p (gsi); gsi_next (&gsi))
2956 gimple stmt = gsi_stmt (gsi);
2957 tree forcedname = gimple_get_lhs (stmt);
2958 pre_expr nameexpr;
2960 if (TREE_CODE (forcedname) == SSA_NAME)
2962 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (forcedname));
2963 VN_INFO_GET (forcedname)->valnum = forcedname;
2964 VN_INFO (forcedname)->value_id = get_next_value_id ();
2965 nameexpr = get_or_alloc_expr_for_name (forcedname);
2966 add_to_value (VN_INFO (forcedname)->value_id, nameexpr);
2967 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
2968 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
2971 gimple_seq_add_seq (stmts, forced_stmts);
2974 name = make_temp_ssa_name (exprtype, NULL, "pretmp");
2975 newstmt = gimple_build_assign (name, folded);
2976 gimple_set_plf (newstmt, NECESSARY, false);
2978 gimple_seq_add_stmt (stmts, newstmt);
2979 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (name));
2981 /* Fold the last statement. */
2982 gsi = gsi_last (*stmts);
2983 if (fold_stmt_inplace (&gsi))
2984 update_stmt (gsi_stmt (gsi));
2986 /* Add a value number to the temporary.
2987 The value may already exist in either NEW_SETS, or AVAIL_OUT, because
2988 we are creating the expression by pieces, and this particular piece of
2989 the expression may have been represented. There is no harm in replacing
2990 here. */
2991 value_id = get_expr_value_id (expr);
2992 VN_INFO_GET (name)->value_id = value_id;
2993 VN_INFO (name)->valnum = sccvn_valnum_from_value_id (value_id);
2994 if (VN_INFO (name)->valnum == NULL_TREE)
2995 VN_INFO (name)->valnum = name;
2996 gcc_assert (VN_INFO (name)->valnum != NULL_TREE);
2997 nameexpr = get_or_alloc_expr_for_name (name);
2998 add_to_value (value_id, nameexpr);
2999 if (NEW_SETS (block))
3000 bitmap_value_replace_in_set (NEW_SETS (block), nameexpr);
3001 bitmap_value_replace_in_set (AVAIL_OUT (block), nameexpr);
3003 pre_stats.insertions++;
3004 if (dump_file && (dump_flags & TDF_DETAILS))
3006 fprintf (dump_file, "Inserted ");
3007 print_gimple_stmt (dump_file, newstmt, 0, 0);
3008 fprintf (dump_file, " in predecessor %d (%04d)\n",
3009 block->index, value_id);
3012 return name;
3016 /* Returns true if we want to inhibit the insertions of PHI nodes
3017 for the given EXPR for basic block BB (a member of a loop).
3018 We want to do this, when we fear that the induction variable we
3019 create might inhibit vectorization. */
3021 static bool
3022 inhibit_phi_insertion (basic_block bb, pre_expr expr)
3024 vn_reference_t vr = PRE_EXPR_REFERENCE (expr);
3025 vec<vn_reference_op_s> ops = vr->operands;
3026 vn_reference_op_t op;
3027 unsigned i;
3029 /* If we aren't going to vectorize we don't inhibit anything. */
3030 if (!flag_tree_loop_vectorize)
3031 return false;
3033 /* Otherwise we inhibit the insertion when the address of the
3034 memory reference is a simple induction variable. In other
3035 cases the vectorizer won't do anything anyway (either it's
3036 loop invariant or a complicated expression). */
3037 FOR_EACH_VEC_ELT (ops, i, op)
3039 switch (op->opcode)
3041 case CALL_EXPR:
3042 /* Calls are not a problem. */
3043 return false;
3045 case ARRAY_REF:
3046 case ARRAY_RANGE_REF:
3047 if (TREE_CODE (op->op0) != SSA_NAME)
3048 break;
3049 /* Fallthru. */
3050 case SSA_NAME:
3052 basic_block defbb = gimple_bb (SSA_NAME_DEF_STMT (op->op0));
3053 affine_iv iv;
3054 /* Default defs are loop invariant. */
3055 if (!defbb)
3056 break;
3057 /* Defined outside this loop, also loop invariant. */
3058 if (!flow_bb_inside_loop_p (bb->loop_father, defbb))
3059 break;
3060 /* If it's a simple induction variable inhibit insertion,
3061 the vectorizer might be interested in this one. */
3062 if (simple_iv (bb->loop_father, bb->loop_father,
3063 op->op0, &iv, true))
3064 return true;
3065 /* No simple IV, vectorizer can't do anything, hence no
3066 reason to inhibit the transformation for this operand. */
3067 break;
3069 default:
3070 break;
3073 return false;
3076 /* Insert the to-be-made-available values of expression EXPRNUM for each
3077 predecessor, stored in AVAIL, into the predecessors of BLOCK, and
3078 merge the result with a phi node, given the same value number as
3079 NODE. Return true if we have inserted new stuff. */
3081 static bool
3082 insert_into_preds_of_block (basic_block block, unsigned int exprnum,
3083 vec<pre_expr> avail)
3085 pre_expr expr = expression_for_id (exprnum);
3086 pre_expr newphi;
3087 unsigned int val = get_expr_value_id (expr);
3088 edge pred;
3089 bool insertions = false;
3090 bool nophi = false;
3091 basic_block bprime;
3092 pre_expr eprime;
3093 edge_iterator ei;
3094 tree type = get_expr_type (expr);
3095 tree temp;
3096 gimple phi;
3098 /* Make sure we aren't creating an induction variable. */
3099 if (bb_loop_depth (block) > 0 && EDGE_COUNT (block->preds) == 2)
3101 bool firstinsideloop = false;
3102 bool secondinsideloop = false;
3103 firstinsideloop = flow_bb_inside_loop_p (block->loop_father,
3104 EDGE_PRED (block, 0)->src);
3105 secondinsideloop = flow_bb_inside_loop_p (block->loop_father,
3106 EDGE_PRED (block, 1)->src);
3107 /* Induction variables only have one edge inside the loop. */
3108 if ((firstinsideloop ^ secondinsideloop)
3109 && (expr->kind != REFERENCE
3110 || inhibit_phi_insertion (block, expr)))
3112 if (dump_file && (dump_flags & TDF_DETAILS))
3113 fprintf (dump_file, "Skipping insertion of phi for partial redundancy: Looks like an induction variable\n");
3114 nophi = true;
3118 /* Make the necessary insertions. */
3119 FOR_EACH_EDGE (pred, ei, block->preds)
3121 gimple_seq stmts = NULL;
3122 tree builtexpr;
3123 bprime = pred->src;
3124 eprime = avail[pred->dest_idx];
3126 if (eprime->kind != NAME && eprime->kind != CONSTANT)
3128 builtexpr = create_expression_by_pieces (bprime, eprime,
3129 &stmts, type);
3130 gcc_assert (!(pred->flags & EDGE_ABNORMAL));
3131 gsi_insert_seq_on_edge (pred, stmts);
3132 if (!builtexpr)
3134 /* We cannot insert a PHI node if we failed to insert
3135 on one edge. */
3136 nophi = true;
3137 continue;
3139 avail[pred->dest_idx] = get_or_alloc_expr_for_name (builtexpr);
3140 insertions = true;
3142 else if (eprime->kind == CONSTANT)
3144 /* Constants may not have the right type, fold_convert
3145 should give us back a constant with the right type. */
3146 tree constant = PRE_EXPR_CONSTANT (eprime);
3147 if (!useless_type_conversion_p (type, TREE_TYPE (constant)))
3149 tree builtexpr = fold_convert (type, constant);
3150 if (!is_gimple_min_invariant (builtexpr))
3152 tree forcedexpr = force_gimple_operand (builtexpr,
3153 &stmts, true,
3154 NULL);
3155 if (!is_gimple_min_invariant (forcedexpr))
3157 if (forcedexpr != builtexpr)
3159 VN_INFO_GET (forcedexpr)->valnum = PRE_EXPR_CONSTANT (eprime);
3160 VN_INFO (forcedexpr)->value_id = get_expr_value_id (eprime);
3162 if (stmts)
3164 gimple_stmt_iterator gsi;
3165 gsi = gsi_start (stmts);
3166 for (; !gsi_end_p (gsi); gsi_next (&gsi))
3168 gimple stmt = gsi_stmt (gsi);
3169 tree lhs = gimple_get_lhs (stmt);
3170 if (TREE_CODE (lhs) == SSA_NAME)
3171 bitmap_set_bit (inserted_exprs,
3172 SSA_NAME_VERSION (lhs));
3173 gimple_set_plf (stmt, NECESSARY, false);
3175 gsi_insert_seq_on_edge (pred, stmts);
3177 avail[pred->dest_idx]
3178 = get_or_alloc_expr_for_name (forcedexpr);
3181 else
3182 avail[pred->dest_idx]
3183 = get_or_alloc_expr_for_constant (builtexpr);
3186 else if (eprime->kind == NAME)
3188 /* We may have to do a conversion because our value
3189 numbering can look through types in certain cases, but
3190 our IL requires all operands of a phi node have the same
3191 type. */
3192 tree name = PRE_EXPR_NAME (eprime);
3193 if (!useless_type_conversion_p (type, TREE_TYPE (name)))
3195 tree builtexpr;
3196 tree forcedexpr;
3197 builtexpr = fold_convert (type, name);
3198 forcedexpr = force_gimple_operand (builtexpr,
3199 &stmts, true,
3200 NULL);
3202 if (forcedexpr != name)
3204 VN_INFO_GET (forcedexpr)->valnum = VN_INFO (name)->valnum;
3205 VN_INFO (forcedexpr)->value_id = VN_INFO (name)->value_id;
3208 if (stmts)
3210 gimple_stmt_iterator gsi;
3211 gsi = gsi_start (stmts);
3212 for (; !gsi_end_p (gsi); gsi_next (&gsi))
3214 gimple stmt = gsi_stmt (gsi);
3215 tree lhs = gimple_get_lhs (stmt);
3216 if (TREE_CODE (lhs) == SSA_NAME)
3217 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (lhs));
3218 gimple_set_plf (stmt, NECESSARY, false);
3220 gsi_insert_seq_on_edge (pred, stmts);
3222 avail[pred->dest_idx] = get_or_alloc_expr_for_name (forcedexpr);
3226 /* If we didn't want a phi node, and we made insertions, we still have
3227 inserted new stuff, and thus return true. If we didn't want a phi node,
3228 and didn't make insertions, we haven't added anything new, so return
3229 false. */
3230 if (nophi && insertions)
3231 return true;
3232 else if (nophi && !insertions)
3233 return false;
3235 /* Now build a phi for the new variable. */
3236 temp = make_temp_ssa_name (type, NULL, "prephitmp");
3237 phi = create_phi_node (temp, block);
3239 gimple_set_plf (phi, NECESSARY, false);
3240 VN_INFO_GET (temp)->value_id = val;
3241 VN_INFO (temp)->valnum = sccvn_valnum_from_value_id (val);
3242 if (VN_INFO (temp)->valnum == NULL_TREE)
3243 VN_INFO (temp)->valnum = temp;
3244 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3245 FOR_EACH_EDGE (pred, ei, block->preds)
3247 pre_expr ae = avail[pred->dest_idx];
3248 gcc_assert (get_expr_type (ae) == type
3249 || useless_type_conversion_p (type, get_expr_type (ae)));
3250 if (ae->kind == CONSTANT)
3251 add_phi_arg (phi, unshare_expr (PRE_EXPR_CONSTANT (ae)),
3252 pred, UNKNOWN_LOCATION);
3253 else
3254 add_phi_arg (phi, PRE_EXPR_NAME (ae), pred, UNKNOWN_LOCATION);
3257 newphi = get_or_alloc_expr_for_name (temp);
3258 add_to_value (val, newphi);
3260 /* The value should *not* exist in PHI_GEN, or else we wouldn't be doing
3261 this insertion, since we test for the existence of this value in PHI_GEN
3262 before proceeding with the partial redundancy checks in insert_aux.
3264 The value may exist in AVAIL_OUT, in particular, it could be represented
3265 by the expression we are trying to eliminate, in which case we want the
3266 replacement to occur. If it's not existing in AVAIL_OUT, we want it
3267 inserted there.
3269 Similarly, to the PHI_GEN case, the value should not exist in NEW_SETS of
3270 this block, because if it did, it would have existed in our dominator's
3271 AVAIL_OUT, and would have been skipped due to the full redundancy check.
3274 bitmap_insert_into_set (PHI_GEN (block), newphi);
3275 bitmap_value_replace_in_set (AVAIL_OUT (block),
3276 newphi);
3277 bitmap_insert_into_set (NEW_SETS (block),
3278 newphi);
3280 if (dump_file && (dump_flags & TDF_DETAILS))
3282 fprintf (dump_file, "Created phi ");
3283 print_gimple_stmt (dump_file, phi, 0, 0);
3284 fprintf (dump_file, " in block %d (%04d)\n", block->index, val);
3286 pre_stats.phis++;
3287 return true;
3292 /* Perform insertion of partially redundant values.
3293 For BLOCK, do the following:
3294 1. Propagate the NEW_SETS of the dominator into the current block.
3295 If the block has multiple predecessors,
3296 2a. Iterate over the ANTIC expressions for the block to see if
3297 any of them are partially redundant.
3298 2b. If so, insert them into the necessary predecessors to make
3299 the expression fully redundant.
3300 2c. Insert a new PHI merging the values of the predecessors.
3301 2d. Insert the new PHI, and the new expressions, into the
3302 NEW_SETS set.
3303 3. Recursively call ourselves on the dominator children of BLOCK.
3305 Steps 1, 2a, and 3 are done by insert_aux. 2b, 2c and 2d are done by
3306 do_regular_insertion and do_partial_insertion.
3310 static bool
3311 do_regular_insertion (basic_block block, basic_block dom)
3313 bool new_stuff = false;
3314 vec<pre_expr> exprs;
3315 pre_expr expr;
3316 vec<pre_expr> avail = vNULL;
3317 int i;
3319 exprs = sorted_array_from_bitmap_set (ANTIC_IN (block));
3320 avail.safe_grow (EDGE_COUNT (block->preds));
3322 FOR_EACH_VEC_ELT (exprs, i, expr)
3324 if (expr->kind == NARY
3325 || expr->kind == REFERENCE)
3327 unsigned int val;
3328 bool by_some = false;
3329 bool cant_insert = false;
3330 bool all_same = true;
3331 pre_expr first_s = NULL;
3332 edge pred;
3333 basic_block bprime;
3334 pre_expr eprime = NULL;
3335 edge_iterator ei;
3336 pre_expr edoubleprime = NULL;
3337 bool do_insertion = false;
3339 val = get_expr_value_id (expr);
3340 if (bitmap_set_contains_value (PHI_GEN (block), val))
3341 continue;
3342 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3344 if (dump_file && (dump_flags & TDF_DETAILS))
3346 fprintf (dump_file, "Found fully redundant value: ");
3347 print_pre_expr (dump_file, expr);
3348 fprintf (dump_file, "\n");
3350 continue;
3353 FOR_EACH_EDGE (pred, ei, block->preds)
3355 unsigned int vprime;
3357 /* We should never run insertion for the exit block
3358 and so not come across fake pred edges. */
3359 gcc_assert (!(pred->flags & EDGE_FAKE));
3360 bprime = pred->src;
3361 eprime = phi_translate (expr, ANTIC_IN (block), NULL,
3362 bprime, block);
3364 /* eprime will generally only be NULL if the
3365 value of the expression, translated
3366 through the PHI for this predecessor, is
3367 undefined. If that is the case, we can't
3368 make the expression fully redundant,
3369 because its value is undefined along a
3370 predecessor path. We can thus break out
3371 early because it doesn't matter what the
3372 rest of the results are. */
3373 if (eprime == NULL)
3375 avail[pred->dest_idx] = NULL;
3376 cant_insert = true;
3377 break;
3380 eprime = fully_constant_expression (eprime);
3381 vprime = get_expr_value_id (eprime);
3382 edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime),
3383 vprime);
3384 if (edoubleprime == NULL)
3386 avail[pred->dest_idx] = eprime;
3387 all_same = false;
3389 else
3391 avail[pred->dest_idx] = edoubleprime;
3392 by_some = true;
3393 /* We want to perform insertions to remove a redundancy on
3394 a path in the CFG we want to optimize for speed. */
3395 if (optimize_edge_for_speed_p (pred))
3396 do_insertion = true;
3397 if (first_s == NULL)
3398 first_s = edoubleprime;
3399 else if (!pre_expr_d::equal (first_s, edoubleprime))
3400 all_same = false;
3403 /* If we can insert it, it's not the same value
3404 already existing along every predecessor, and
3405 it's defined by some predecessor, it is
3406 partially redundant. */
3407 if (!cant_insert && !all_same && by_some)
3409 if (!do_insertion)
3411 if (dump_file && (dump_flags & TDF_DETAILS))
3413 fprintf (dump_file, "Skipping partial redundancy for "
3414 "expression ");
3415 print_pre_expr (dump_file, expr);
3416 fprintf (dump_file, " (%04d), no redundancy on to be "
3417 "optimized for speed edge\n", val);
3420 else if (dbg_cnt (treepre_insert))
3422 if (dump_file && (dump_flags & TDF_DETAILS))
3424 fprintf (dump_file, "Found partial redundancy for "
3425 "expression ");
3426 print_pre_expr (dump_file, expr);
3427 fprintf (dump_file, " (%04d)\n",
3428 get_expr_value_id (expr));
3430 if (insert_into_preds_of_block (block,
3431 get_expression_id (expr),
3432 avail))
3433 new_stuff = true;
3436 /* If all edges produce the same value and that value is
3437 an invariant, then the PHI has the same value on all
3438 edges. Note this. */
3439 else if (!cant_insert && all_same)
3441 gcc_assert (edoubleprime->kind == CONSTANT
3442 || edoubleprime->kind == NAME);
3444 tree temp = make_temp_ssa_name (get_expr_type (expr),
3445 NULL, "pretmp");
3446 gimple assign = gimple_build_assign (temp,
3447 edoubleprime->kind == CONSTANT ? PRE_EXPR_CONSTANT (edoubleprime) : PRE_EXPR_NAME (edoubleprime));
3448 gimple_stmt_iterator gsi = gsi_after_labels (block);
3449 gsi_insert_before (&gsi, assign, GSI_NEW_STMT);
3451 gimple_set_plf (assign, NECESSARY, false);
3452 VN_INFO_GET (temp)->value_id = val;
3453 VN_INFO (temp)->valnum = sccvn_valnum_from_value_id (val);
3454 if (VN_INFO (temp)->valnum == NULL_TREE)
3455 VN_INFO (temp)->valnum = temp;
3456 bitmap_set_bit (inserted_exprs, SSA_NAME_VERSION (temp));
3457 pre_expr newe = get_or_alloc_expr_for_name (temp);
3458 add_to_value (val, newe);
3459 bitmap_value_replace_in_set (AVAIL_OUT (block), newe);
3460 bitmap_insert_into_set (NEW_SETS (block), newe);
3465 exprs.release ();
3466 return new_stuff;
3470 /* Perform insertion for partially anticipatable expressions. There
3471 is only one case we will perform insertion for these. This case is
3472 if the expression is partially anticipatable, and fully available.
3473 In this case, we know that putting it earlier will enable us to
3474 remove the later computation. */
3477 static bool
3478 do_partial_partial_insertion (basic_block block, basic_block dom)
3480 bool new_stuff = false;
3481 vec<pre_expr> exprs;
3482 pre_expr expr;
3483 auto_vec<pre_expr> avail;
3484 int i;
3486 exprs = sorted_array_from_bitmap_set (PA_IN (block));
3487 avail.safe_grow (EDGE_COUNT (block->preds));
3489 FOR_EACH_VEC_ELT (exprs, i, expr)
3491 if (expr->kind == NARY
3492 || expr->kind == REFERENCE)
3494 unsigned int val;
3495 bool by_all = true;
3496 bool cant_insert = false;
3497 edge pred;
3498 basic_block bprime;
3499 pre_expr eprime = NULL;
3500 edge_iterator ei;
3502 val = get_expr_value_id (expr);
3503 if (bitmap_set_contains_value (PHI_GEN (block), val))
3504 continue;
3505 if (bitmap_set_contains_value (AVAIL_OUT (dom), val))
3506 continue;
3508 FOR_EACH_EDGE (pred, ei, block->preds)
3510 unsigned int vprime;
3511 pre_expr edoubleprime;
3513 /* We should never run insertion for the exit block
3514 and so not come across fake pred edges. */
3515 gcc_assert (!(pred->flags & EDGE_FAKE));
3516 bprime = pred->src;
3517 eprime = phi_translate (expr, ANTIC_IN (block),
3518 PA_IN (block),
3519 bprime, block);
3521 /* eprime will generally only be NULL if the
3522 value of the expression, translated
3523 through the PHI for this predecessor, is
3524 undefined. If that is the case, we can't
3525 make the expression fully redundant,
3526 because its value is undefined along a
3527 predecessor path. We can thus break out
3528 early because it doesn't matter what the
3529 rest of the results are. */
3530 if (eprime == NULL)
3532 avail[pred->dest_idx] = NULL;
3533 cant_insert = true;
3534 break;
3537 eprime = fully_constant_expression (eprime);
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 exprs.release ();
3604 return new_stuff;
3607 static bool
3608 insert_aux (basic_block block)
3610 basic_block son;
3611 bool new_stuff = false;
3613 if (block)
3615 basic_block dom;
3616 dom = get_immediate_dominator (CDI_DOMINATORS, block);
3617 if (dom)
3619 unsigned i;
3620 bitmap_iterator bi;
3621 bitmap_set_t newset = NEW_SETS (dom);
3622 if (newset)
3624 /* Note that we need to value_replace both NEW_SETS, and
3625 AVAIL_OUT. For both the case of NEW_SETS, the value may be
3626 represented by some non-simple expression here that we want
3627 to replace it with. */
3628 FOR_EACH_EXPR_ID_IN_SET (newset, i, bi)
3630 pre_expr expr = expression_for_id (i);
3631 bitmap_value_replace_in_set (NEW_SETS (block), expr);
3632 bitmap_value_replace_in_set (AVAIL_OUT (block), expr);
3635 if (!single_pred_p (block))
3637 new_stuff |= do_regular_insertion (block, dom);
3638 if (do_partial_partial)
3639 new_stuff |= do_partial_partial_insertion (block, dom);
3643 for (son = first_dom_son (CDI_DOMINATORS, block);
3644 son;
3645 son = next_dom_son (CDI_DOMINATORS, son))
3647 new_stuff |= insert_aux (son);
3650 return new_stuff;
3653 /* Perform insertion of partially redundant values. */
3655 static void
3656 insert (void)
3658 bool new_stuff = true;
3659 basic_block bb;
3660 int num_iterations = 0;
3662 FOR_ALL_BB_FN (bb, cfun)
3663 NEW_SETS (bb) = bitmap_set_new ();
3665 while (new_stuff)
3667 num_iterations++;
3668 if (dump_file && dump_flags & TDF_DETAILS)
3669 fprintf (dump_file, "Starting insert iteration %d\n", num_iterations);
3670 new_stuff = insert_aux (ENTRY_BLOCK_PTR_FOR_FN (cfun));
3672 /* Clear the NEW sets before the next iteration. We have already
3673 fully propagated its contents. */
3674 if (new_stuff)
3675 FOR_ALL_BB_FN (bb, cfun)
3676 bitmap_set_free (NEW_SETS (bb));
3678 statistics_histogram_event (cfun, "insert iterations", num_iterations);
3682 /* Compute the AVAIL set for all basic blocks.
3684 This function performs value numbering of the statements in each basic
3685 block. The AVAIL sets are built from information we glean while doing
3686 this value numbering, since the AVAIL sets contain only one entry per
3687 value.
3689 AVAIL_IN[BLOCK] = AVAIL_OUT[dom(BLOCK)].
3690 AVAIL_OUT[BLOCK] = AVAIL_IN[BLOCK] U PHI_GEN[BLOCK] U TMP_GEN[BLOCK]. */
3692 static void
3693 compute_avail (void)
3696 basic_block block, son;
3697 basic_block *worklist;
3698 size_t sp = 0;
3699 unsigned i;
3701 /* We pretend that default definitions are defined in the entry block.
3702 This includes function arguments and the static chain decl. */
3703 for (i = 1; i < num_ssa_names; ++i)
3705 tree name = ssa_name (i);
3706 pre_expr e;
3707 if (!name
3708 || !SSA_NAME_IS_DEFAULT_DEF (name)
3709 || has_zero_uses (name)
3710 || virtual_operand_p (name))
3711 continue;
3713 e = get_or_alloc_expr_for_name (name);
3714 add_to_value (get_expr_value_id (e), e);
3715 bitmap_insert_into_set (TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (cfun)), e);
3716 bitmap_value_insert_into_set (AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (cfun)),
3720 if (dump_file && (dump_flags & TDF_DETAILS))
3722 print_bitmap_set (dump_file, TMP_GEN (ENTRY_BLOCK_PTR_FOR_FN (cfun)),
3723 "tmp_gen", ENTRY_BLOCK);
3724 print_bitmap_set (dump_file, AVAIL_OUT (ENTRY_BLOCK_PTR_FOR_FN (cfun)),
3725 "avail_out", ENTRY_BLOCK);
3728 /* Allocate the worklist. */
3729 worklist = XNEWVEC (basic_block, n_basic_blocks_for_fn (cfun));
3731 /* Seed the algorithm by putting the dominator children of the entry
3732 block on the worklist. */
3733 for (son = first_dom_son (CDI_DOMINATORS, ENTRY_BLOCK_PTR_FOR_FN (cfun));
3734 son;
3735 son = next_dom_son (CDI_DOMINATORS, son))
3736 worklist[sp++] = son;
3738 /* Loop until the worklist is empty. */
3739 while (sp)
3741 gimple_stmt_iterator gsi;
3742 gimple stmt;
3743 basic_block dom;
3745 /* Pick a block from the worklist. */
3746 block = worklist[--sp];
3748 /* Initially, the set of available values in BLOCK is that of
3749 its immediate dominator. */
3750 dom = get_immediate_dominator (CDI_DOMINATORS, block);
3751 if (dom)
3752 bitmap_set_copy (AVAIL_OUT (block), AVAIL_OUT (dom));
3754 /* Generate values for PHI nodes. */
3755 for (gsi = gsi_start_phis (block); !gsi_end_p (gsi); gsi_next (&gsi))
3757 tree result = gimple_phi_result (gsi_stmt (gsi));
3759 /* We have no need for virtual phis, as they don't represent
3760 actual computations. */
3761 if (virtual_operand_p (result))
3762 continue;
3764 pre_expr e = get_or_alloc_expr_for_name (result);
3765 add_to_value (get_expr_value_id (e), e);
3766 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
3767 bitmap_insert_into_set (PHI_GEN (block), e);
3770 BB_MAY_NOTRETURN (block) = 0;
3772 /* Now compute value numbers and populate value sets with all
3773 the expressions computed in BLOCK. */
3774 for (gsi = gsi_start_bb (block); !gsi_end_p (gsi); gsi_next (&gsi))
3776 ssa_op_iter iter;
3777 tree op;
3779 stmt = gsi_stmt (gsi);
3781 /* Cache whether the basic-block has any non-visible side-effect
3782 or control flow.
3783 If this isn't a call or it is the last stmt in the
3784 basic-block then the CFG represents things correctly. */
3785 if (is_gimple_call (stmt) && !stmt_ends_bb_p (stmt))
3787 /* Non-looping const functions always return normally.
3788 Otherwise the call might not return or have side-effects
3789 that forbids hoisting possibly trapping expressions
3790 before it. */
3791 int flags = gimple_call_flags (stmt);
3792 if (!(flags & ECF_CONST)
3793 || (flags & ECF_LOOPING_CONST_OR_PURE))
3794 BB_MAY_NOTRETURN (block) = 1;
3797 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_DEF)
3799 pre_expr e = get_or_alloc_expr_for_name (op);
3801 add_to_value (get_expr_value_id (e), e);
3802 bitmap_insert_into_set (TMP_GEN (block), e);
3803 bitmap_value_insert_into_set (AVAIL_OUT (block), e);
3806 if (gimple_has_side_effects (stmt)
3807 || stmt_could_throw_p (stmt)
3808 || is_gimple_debug (stmt))
3809 continue;
3811 FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_USE)
3813 if (ssa_undefined_value_p (op))
3814 continue;
3815 pre_expr e = get_or_alloc_expr_for_name (op);
3816 bitmap_value_insert_into_set (EXP_GEN (block), e);
3819 switch (gimple_code (stmt))
3821 case GIMPLE_RETURN:
3822 continue;
3824 case GIMPLE_CALL:
3826 vn_reference_t ref;
3827 pre_expr result = NULL;
3828 auto_vec<vn_reference_op_s> ops;
3830 /* We can value number only calls to real functions. */
3831 if (gimple_call_internal_p (stmt))
3832 continue;
3834 copy_reference_ops_from_call (stmt, &ops);
3835 vn_reference_lookup_pieces (gimple_vuse (stmt), 0,
3836 gimple_expr_type (stmt),
3837 ops, &ref, VN_NOWALK);
3838 if (!ref)
3839 continue;
3841 /* If the value of the call is not invalidated in
3842 this block until it is computed, add the expression
3843 to EXP_GEN. */
3844 if (!gimple_vuse (stmt)
3845 || gimple_code
3846 (SSA_NAME_DEF_STMT (gimple_vuse (stmt))) == GIMPLE_PHI
3847 || gimple_bb (SSA_NAME_DEF_STMT
3848 (gimple_vuse (stmt))) != block)
3850 result = (pre_expr) pool_alloc (pre_expr_pool);
3851 result->kind = REFERENCE;
3852 result->id = 0;
3853 PRE_EXPR_REFERENCE (result) = ref;
3855 get_or_alloc_expression_id (result);
3856 add_to_value (get_expr_value_id (result), result);
3857 bitmap_value_insert_into_set (EXP_GEN (block), result);
3859 continue;
3862 case GIMPLE_ASSIGN:
3864 pre_expr result = NULL;
3865 switch (vn_get_stmt_kind (stmt))
3867 case VN_NARY:
3869 enum tree_code code = gimple_assign_rhs_code (stmt);
3870 vn_nary_op_t nary;
3872 /* COND_EXPR and VEC_COND_EXPR are awkward in
3873 that they contain an embedded complex expression.
3874 Don't even try to shove those through PRE. */
3875 if (code == COND_EXPR
3876 || code == VEC_COND_EXPR)
3877 continue;
3879 vn_nary_op_lookup_stmt (stmt, &nary);
3880 if (!nary)
3881 continue;
3883 /* If the NARY traps and there was a preceding
3884 point in the block that might not return avoid
3885 adding the nary to EXP_GEN. */
3886 if (BB_MAY_NOTRETURN (block)
3887 && vn_nary_may_trap (nary))
3888 continue;
3890 result = (pre_expr) pool_alloc (pre_expr_pool);
3891 result->kind = NARY;
3892 result->id = 0;
3893 PRE_EXPR_NARY (result) = nary;
3894 break;
3897 case VN_REFERENCE:
3899 vn_reference_t ref;
3900 vn_reference_lookup (gimple_assign_rhs1 (stmt),
3901 gimple_vuse (stmt),
3902 VN_WALK, &ref);
3903 if (!ref)
3904 continue;
3906 /* If the value of the reference is not invalidated in
3907 this block until it is computed, add the expression
3908 to EXP_GEN. */
3909 if (gimple_vuse (stmt))
3911 gimple def_stmt;
3912 bool ok = true;
3913 def_stmt = SSA_NAME_DEF_STMT (gimple_vuse (stmt));
3914 while (!gimple_nop_p (def_stmt)
3915 && gimple_code (def_stmt) != GIMPLE_PHI
3916 && gimple_bb (def_stmt) == block)
3918 if (stmt_may_clobber_ref_p
3919 (def_stmt, gimple_assign_rhs1 (stmt)))
3921 ok = false;
3922 break;
3924 def_stmt
3925 = SSA_NAME_DEF_STMT (gimple_vuse (def_stmt));
3927 if (!ok)
3928 continue;
3931 result = (pre_expr) pool_alloc (pre_expr_pool);
3932 result->kind = REFERENCE;
3933 result->id = 0;
3934 PRE_EXPR_REFERENCE (result) = ref;
3935 break;
3938 default:
3939 continue;
3942 get_or_alloc_expression_id (result);
3943 add_to_value (get_expr_value_id (result), result);
3944 bitmap_value_insert_into_set (EXP_GEN (block), result);
3945 continue;
3947 default:
3948 break;
3952 if (dump_file && (dump_flags & TDF_DETAILS))
3954 print_bitmap_set (dump_file, EXP_GEN (block),
3955 "exp_gen", block->index);
3956 print_bitmap_set (dump_file, PHI_GEN (block),
3957 "phi_gen", block->index);
3958 print_bitmap_set (dump_file, TMP_GEN (block),
3959 "tmp_gen", block->index);
3960 print_bitmap_set (dump_file, AVAIL_OUT (block),
3961 "avail_out", block->index);
3964 /* Put the dominator children of BLOCK on the worklist of blocks
3965 to compute available sets for. */
3966 for (son = first_dom_son (CDI_DOMINATORS, block);
3967 son;
3968 son = next_dom_son (CDI_DOMINATORS, son))
3969 worklist[sp++] = son;
3972 free (worklist);
3976 /* Local state for the eliminate domwalk. */
3977 static vec<gimple> el_to_remove;
3978 static vec<gimple> el_to_update;
3979 static unsigned int el_todo;
3980 static vec<tree> el_avail;
3981 static vec<tree> el_avail_stack;
3983 /* Return a leader for OP that is available at the current point of the
3984 eliminate domwalk. */
3986 static tree
3987 eliminate_avail (tree op)
3989 tree valnum = VN_INFO (op)->valnum;
3990 if (TREE_CODE (valnum) == SSA_NAME)
3992 if (SSA_NAME_IS_DEFAULT_DEF (valnum))
3993 return valnum;
3994 if (el_avail.length () > SSA_NAME_VERSION (valnum))
3995 return el_avail[SSA_NAME_VERSION (valnum)];
3997 else if (is_gimple_min_invariant (valnum))
3998 return valnum;
3999 return NULL_TREE;
4002 /* At the current point of the eliminate domwalk make OP available. */
4004 static void
4005 eliminate_push_avail (tree op)
4007 tree valnum = VN_INFO (op)->valnum;
4008 if (TREE_CODE (valnum) == SSA_NAME)
4010 if (el_avail.length () <= SSA_NAME_VERSION (valnum))
4011 el_avail.safe_grow_cleared (SSA_NAME_VERSION (valnum) + 1);
4012 el_avail[SSA_NAME_VERSION (valnum)] = op;
4013 el_avail_stack.safe_push (op);
4017 /* Insert the expression recorded by SCCVN for VAL at *GSI. Returns
4018 the leader for the expression if insertion was successful. */
4020 static tree
4021 eliminate_insert (gimple_stmt_iterator *gsi, tree val)
4023 tree expr = vn_get_expr_for (val);
4024 if (!CONVERT_EXPR_P (expr)
4025 && TREE_CODE (expr) != VIEW_CONVERT_EXPR)
4026 return NULL_TREE;
4028 tree op = TREE_OPERAND (expr, 0);
4029 tree leader = TREE_CODE (op) == SSA_NAME ? eliminate_avail (op) : op;
4030 if (!leader)
4031 return NULL_TREE;
4033 tree res = make_temp_ssa_name (TREE_TYPE (val), NULL, "pretmp");
4034 gimple tem = gimple_build_assign (res,
4035 fold_build1 (TREE_CODE (expr),
4036 TREE_TYPE (expr), leader));
4037 gsi_insert_before (gsi, tem, GSI_SAME_STMT);
4038 VN_INFO_GET (res)->valnum = val;
4040 if (TREE_CODE (leader) == SSA_NAME)
4041 gimple_set_plf (SSA_NAME_DEF_STMT (leader), NECESSARY, true);
4043 pre_stats.insertions++;
4044 if (dump_file && (dump_flags & TDF_DETAILS))
4046 fprintf (dump_file, "Inserted ");
4047 print_gimple_stmt (dump_file, tem, 0, 0);
4050 return res;
4053 class eliminate_dom_walker : public dom_walker
4055 public:
4056 eliminate_dom_walker (cdi_direction direction) : dom_walker (direction) {}
4058 virtual void before_dom_children (basic_block);
4059 virtual void after_dom_children (basic_block);
4062 /* Perform elimination for the basic-block B during the domwalk. */
4064 void
4065 eliminate_dom_walker::before_dom_children (basic_block b)
4067 gimple_stmt_iterator gsi;
4068 gimple stmt;
4070 /* Mark new bb. */
4071 el_avail_stack.safe_push (NULL_TREE);
4073 for (gsi = gsi_start_phis (b); !gsi_end_p (gsi);)
4075 gimple stmt, phi = gsi_stmt (gsi);
4076 tree sprime = NULL_TREE, res = PHI_RESULT (phi);
4077 gimple_stmt_iterator gsi2;
4079 /* We want to perform redundant PHI elimination. Do so by
4080 replacing the PHI with a single copy if possible.
4081 Do not touch inserted, single-argument or virtual PHIs. */
4082 if (gimple_phi_num_args (phi) == 1
4083 || virtual_operand_p (res))
4085 gsi_next (&gsi);
4086 continue;
4089 sprime = eliminate_avail (res);
4090 if (!sprime
4091 || sprime == res)
4093 eliminate_push_avail (res);
4094 gsi_next (&gsi);
4095 continue;
4097 else if (is_gimple_min_invariant (sprime))
4099 if (!useless_type_conversion_p (TREE_TYPE (res),
4100 TREE_TYPE (sprime)))
4101 sprime = fold_convert (TREE_TYPE (res), sprime);
4104 if (dump_file && (dump_flags & TDF_DETAILS))
4106 fprintf (dump_file, "Replaced redundant PHI node defining ");
4107 print_generic_expr (dump_file, res, 0);
4108 fprintf (dump_file, " with ");
4109 print_generic_expr (dump_file, sprime, 0);
4110 fprintf (dump_file, "\n");
4113 remove_phi_node (&gsi, false);
4115 if (inserted_exprs
4116 && !bitmap_bit_p (inserted_exprs, SSA_NAME_VERSION (res))
4117 && TREE_CODE (sprime) == SSA_NAME)
4118 gimple_set_plf (SSA_NAME_DEF_STMT (sprime), NECESSARY, true);
4120 if (!useless_type_conversion_p (TREE_TYPE (res), TREE_TYPE (sprime)))
4121 sprime = fold_convert (TREE_TYPE (res), sprime);
4122 stmt = gimple_build_assign (res, sprime);
4123 gimple_set_plf (stmt, NECESSARY, gimple_plf (phi, NECESSARY));
4125 gsi2 = gsi_after_labels (b);
4126 gsi_insert_before (&gsi2, stmt, GSI_NEW_STMT);
4127 /* Queue the copy for eventual removal. */
4128 el_to_remove.safe_push (stmt);
4129 /* If we inserted this PHI node ourself, it's not an elimination. */
4130 if (inserted_exprs
4131 && bitmap_bit_p (inserted_exprs, SSA_NAME_VERSION (res)))
4132 pre_stats.phis--;
4133 else
4134 pre_stats.eliminations++;
4137 for (gsi = gsi_start_bb (b); !gsi_end_p (gsi); gsi_next (&gsi))
4139 tree lhs = NULL_TREE;
4140 tree rhs = NULL_TREE;
4142 stmt = gsi_stmt (gsi);
4144 if (gimple_has_lhs (stmt))
4145 lhs = gimple_get_lhs (stmt);
4147 if (gimple_assign_single_p (stmt))
4148 rhs = gimple_assign_rhs1 (stmt);
4150 /* Lookup the RHS of the expression, see if we have an
4151 available computation for it. If so, replace the RHS with
4152 the available computation. */
4153 if (gimple_has_lhs (stmt)
4154 && TREE_CODE (lhs) == SSA_NAME
4155 && !gimple_has_volatile_ops (stmt))
4157 tree sprime;
4158 gimple orig_stmt = stmt;
4160 sprime = eliminate_avail (lhs);
4161 /* If there is no usable leader mark lhs as leader for its value. */
4162 if (!sprime)
4163 eliminate_push_avail (lhs);
4165 /* See PR43491. Do not replace a global register variable when
4166 it is a the RHS of an assignment. Do replace local register
4167 variables since gcc does not guarantee a local variable will
4168 be allocated in register.
4169 Do not perform copy propagation or undo constant propagation. */
4170 if (gimple_assign_single_p (stmt)
4171 && (TREE_CODE (rhs) == SSA_NAME
4172 || is_gimple_min_invariant (rhs)
4173 || (TREE_CODE (rhs) == VAR_DECL
4174 && is_global_var (rhs)
4175 && DECL_HARD_REGISTER (rhs))))
4176 continue;
4178 if (!sprime)
4180 /* If there is no existing usable leader but SCCVN thinks
4181 it has an expression it wants to use as replacement,
4182 insert that. */
4183 tree val = VN_INFO (lhs)->valnum;
4184 if (val != VN_TOP
4185 && TREE_CODE (val) == SSA_NAME
4186 && VN_INFO (val)->needs_insertion
4187 && VN_INFO (val)->expr != NULL_TREE
4188 && (sprime = eliminate_insert (&gsi, val)) != NULL_TREE)
4189 eliminate_push_avail (sprime);
4191 else if (is_gimple_min_invariant (sprime))
4193 /* If there is no existing leader but SCCVN knows this
4194 value is constant, use that constant. */
4195 if (!useless_type_conversion_p (TREE_TYPE (lhs),
4196 TREE_TYPE (sprime)))
4197 sprime = fold_convert (TREE_TYPE (lhs), sprime);
4199 if (dump_file && (dump_flags & TDF_DETAILS))
4201 fprintf (dump_file, "Replaced ");
4202 print_gimple_expr (dump_file, stmt, 0, 0);
4203 fprintf (dump_file, " with ");
4204 print_generic_expr (dump_file, sprime, 0);
4205 fprintf (dump_file, " in ");
4206 print_gimple_stmt (dump_file, stmt, 0, 0);
4208 pre_stats.eliminations++;
4209 propagate_tree_value_into_stmt (&gsi, sprime);
4210 stmt = gsi_stmt (gsi);
4211 update_stmt (stmt);
4213 /* If we removed EH side-effects from the statement, clean
4214 its EH information. */
4215 if (maybe_clean_or_replace_eh_stmt (orig_stmt, stmt))
4217 bitmap_set_bit (need_eh_cleanup,
4218 gimple_bb (stmt)->index);
4219 if (dump_file && (dump_flags & TDF_DETAILS))
4220 fprintf (dump_file, " Removed EH side-effects.\n");
4222 continue;
4225 if (sprime
4226 && sprime != lhs
4227 && (rhs == NULL_TREE
4228 || TREE_CODE (rhs) != SSA_NAME
4229 || may_propagate_copy (rhs, sprime)))
4231 bool can_make_abnormal_goto
4232 = is_gimple_call (stmt)
4233 && stmt_can_make_abnormal_goto (stmt);
4235 gcc_assert (sprime != rhs);
4237 if (dump_file && (dump_flags & TDF_DETAILS))
4239 fprintf (dump_file, "Replaced ");
4240 print_gimple_expr (dump_file, stmt, 0, 0);
4241 fprintf (dump_file, " with ");
4242 print_generic_expr (dump_file, sprime, 0);
4243 fprintf (dump_file, " in ");
4244 print_gimple_stmt (dump_file, stmt, 0, 0);
4247 if (TREE_CODE (sprime) == SSA_NAME)
4248 gimple_set_plf (SSA_NAME_DEF_STMT (sprime),
4249 NECESSARY, true);
4250 /* We need to make sure the new and old types actually match,
4251 which may require adding a simple cast, which fold_convert
4252 will do for us. */
4253 if ((!rhs || TREE_CODE (rhs) != SSA_NAME)
4254 && !useless_type_conversion_p (gimple_expr_type (stmt),
4255 TREE_TYPE (sprime)))
4256 sprime = fold_convert (gimple_expr_type (stmt), sprime);
4258 pre_stats.eliminations++;
4259 propagate_tree_value_into_stmt (&gsi, sprime);
4260 stmt = gsi_stmt (gsi);
4261 update_stmt (stmt);
4263 /* If we removed EH side-effects from the statement, clean
4264 its EH information. */
4265 if (maybe_clean_or_replace_eh_stmt (orig_stmt, stmt))
4267 bitmap_set_bit (need_eh_cleanup,
4268 gimple_bb (stmt)->index);
4269 if (dump_file && (dump_flags & TDF_DETAILS))
4270 fprintf (dump_file, " Removed EH side-effects.\n");
4273 /* Likewise for AB side-effects. */
4274 if (can_make_abnormal_goto
4275 && !stmt_can_make_abnormal_goto (stmt))
4277 bitmap_set_bit (need_ab_cleanup,
4278 gimple_bb (stmt)->index);
4279 if (dump_file && (dump_flags & TDF_DETAILS))
4280 fprintf (dump_file, " Removed AB side-effects.\n");
4284 /* If the statement is a scalar store, see if the expression
4285 has the same value number as its rhs. If so, the store is
4286 dead. */
4287 else if (gimple_assign_single_p (stmt)
4288 && !gimple_has_volatile_ops (stmt)
4289 && !is_gimple_reg (gimple_assign_lhs (stmt))
4290 && (TREE_CODE (rhs) == SSA_NAME
4291 || is_gimple_min_invariant (rhs)))
4293 tree val;
4294 val = vn_reference_lookup (gimple_assign_lhs (stmt),
4295 gimple_vuse (stmt), VN_WALK, NULL);
4296 if (TREE_CODE (rhs) == SSA_NAME)
4297 rhs = VN_INFO (rhs)->valnum;
4298 if (val
4299 && operand_equal_p (val, rhs, 0))
4301 if (dump_file && (dump_flags & TDF_DETAILS))
4303 fprintf (dump_file, "Deleted redundant store ");
4304 print_gimple_stmt (dump_file, stmt, 0, 0);
4307 /* Queue stmt for removal. */
4308 el_to_remove.safe_push (stmt);
4311 /* Visit COND_EXPRs and fold the comparison with the
4312 available value-numbers. */
4313 else if (gimple_code (stmt) == GIMPLE_COND)
4315 tree op0 = gimple_cond_lhs (stmt);
4316 tree op1 = gimple_cond_rhs (stmt);
4317 tree result;
4319 if (TREE_CODE (op0) == SSA_NAME)
4320 op0 = VN_INFO (op0)->valnum;
4321 if (TREE_CODE (op1) == SSA_NAME)
4322 op1 = VN_INFO (op1)->valnum;
4323 result = fold_binary (gimple_cond_code (stmt), boolean_type_node,
4324 op0, op1);
4325 if (result && TREE_CODE (result) == INTEGER_CST)
4327 if (integer_zerop (result))
4328 gimple_cond_make_false (stmt);
4329 else
4330 gimple_cond_make_true (stmt);
4331 update_stmt (stmt);
4332 el_todo = TODO_cleanup_cfg;
4335 /* Visit indirect calls and turn them into direct calls if
4336 possible. */
4337 if (is_gimple_call (stmt))
4339 tree orig_fn = gimple_call_fn (stmt);
4340 tree fn;
4341 if (!orig_fn)
4342 continue;
4343 if (TREE_CODE (orig_fn) == SSA_NAME)
4344 fn = VN_INFO (orig_fn)->valnum;
4345 else if (TREE_CODE (orig_fn) == OBJ_TYPE_REF
4346 && TREE_CODE (OBJ_TYPE_REF_EXPR (orig_fn)) == SSA_NAME)
4348 fn = VN_INFO (OBJ_TYPE_REF_EXPR (orig_fn))->valnum;
4349 if (!gimple_call_addr_fndecl (fn))
4351 fn = ipa_intraprocedural_devirtualization (stmt);
4352 if (fn)
4353 fn = build_fold_addr_expr (fn);
4356 else
4357 continue;
4358 if (gimple_call_addr_fndecl (fn) != NULL_TREE
4359 && useless_type_conversion_p (TREE_TYPE (orig_fn),
4360 TREE_TYPE (fn)))
4362 bool can_make_abnormal_goto
4363 = stmt_can_make_abnormal_goto (stmt);
4364 bool was_noreturn = gimple_call_noreturn_p (stmt);
4366 if (dump_file && (dump_flags & TDF_DETAILS))
4368 fprintf (dump_file, "Replacing call target with ");
4369 print_generic_expr (dump_file, fn, 0);
4370 fprintf (dump_file, " in ");
4371 print_gimple_stmt (dump_file, stmt, 0, 0);
4374 gimple_call_set_fn (stmt, fn);
4375 el_to_update.safe_push (stmt);
4377 /* When changing a call into a noreturn call, cfg cleanup
4378 is needed to fix up the noreturn call. */
4379 if (!was_noreturn && gimple_call_noreturn_p (stmt))
4380 el_todo |= TODO_cleanup_cfg;
4382 /* If we removed EH side-effects from the statement, clean
4383 its EH information. */
4384 if (maybe_clean_or_replace_eh_stmt (stmt, stmt))
4386 bitmap_set_bit (need_eh_cleanup,
4387 gimple_bb (stmt)->index);
4388 if (dump_file && (dump_flags & TDF_DETAILS))
4389 fprintf (dump_file, " Removed EH side-effects.\n");
4392 /* Likewise for AB side-effects. */
4393 if (can_make_abnormal_goto
4394 && !stmt_can_make_abnormal_goto (stmt))
4396 bitmap_set_bit (need_ab_cleanup,
4397 gimple_bb (stmt)->index);
4398 if (dump_file && (dump_flags & TDF_DETAILS))
4399 fprintf (dump_file, " Removed AB side-effects.\n");
4402 /* Changing an indirect call to a direct call may
4403 have exposed different semantics. This may
4404 require an SSA update. */
4405 el_todo |= TODO_update_ssa_only_virtuals;
4411 /* Make no longer available leaders no longer available. */
4413 void
4414 eliminate_dom_walker::after_dom_children (basic_block)
4416 tree entry;
4417 while ((entry = el_avail_stack.pop ()) != NULL_TREE)
4418 el_avail[SSA_NAME_VERSION (VN_INFO (entry)->valnum)] = NULL_TREE;
4421 /* Eliminate fully redundant computations. */
4423 static unsigned int
4424 eliminate (void)
4426 gimple_stmt_iterator gsi;
4427 gimple stmt;
4428 unsigned i;
4430 need_eh_cleanup = BITMAP_ALLOC (NULL);
4431 need_ab_cleanup = BITMAP_ALLOC (NULL);
4433 el_to_remove.create (0);
4434 el_to_update.create (0);
4435 el_todo = 0;
4436 el_avail.create (0);
4437 el_avail_stack.create (0);
4439 eliminate_dom_walker (CDI_DOMINATORS).walk (cfun->cfg->x_entry_block_ptr);
4441 el_avail.release ();
4442 el_avail_stack.release ();
4444 /* We cannot remove stmts during BB walk, especially not release SSA
4445 names there as this confuses the VN machinery. The stmts ending
4446 up in el_to_remove are either stores or simple copies. */
4447 FOR_EACH_VEC_ELT (el_to_remove, i, stmt)
4449 tree lhs = gimple_assign_lhs (stmt);
4450 tree rhs = gimple_assign_rhs1 (stmt);
4451 use_operand_p use_p;
4452 gimple use_stmt;
4454 /* If there is a single use only, propagate the equivalency
4455 instead of keeping the copy. */
4456 if (TREE_CODE (lhs) == SSA_NAME
4457 && TREE_CODE (rhs) == SSA_NAME
4458 && single_imm_use (lhs, &use_p, &use_stmt)
4459 && may_propagate_copy (USE_FROM_PTR (use_p), rhs))
4461 SET_USE (use_p, rhs);
4462 update_stmt (use_stmt);
4463 if (inserted_exprs
4464 && bitmap_bit_p (inserted_exprs, SSA_NAME_VERSION (lhs))
4465 && TREE_CODE (rhs) == SSA_NAME)
4466 gimple_set_plf (SSA_NAME_DEF_STMT (rhs), NECESSARY, true);
4469 /* If this is a store or a now unused copy, remove it. */
4470 if (TREE_CODE (lhs) != SSA_NAME
4471 || has_zero_uses (lhs))
4473 basic_block bb = gimple_bb (stmt);
4474 gsi = gsi_for_stmt (stmt);
4475 unlink_stmt_vdef (stmt);
4476 if (gsi_remove (&gsi, true))
4477 bitmap_set_bit (need_eh_cleanup, bb->index);
4478 if (inserted_exprs
4479 && TREE_CODE (lhs) == SSA_NAME)
4480 bitmap_clear_bit (inserted_exprs, SSA_NAME_VERSION (lhs));
4481 release_defs (stmt);
4484 el_to_remove.release ();
4486 /* We cannot update call statements with virtual operands during
4487 SSA walk. This might remove them which in turn makes our
4488 VN lattice invalid. */
4489 FOR_EACH_VEC_ELT (el_to_update, i, stmt)
4490 update_stmt (stmt);
4491 el_to_update.release ();
4493 return el_todo;
4496 /* Perform CFG cleanups made necessary by elimination. */
4498 static unsigned
4499 fini_eliminate (void)
4501 bool do_eh_cleanup = !bitmap_empty_p (need_eh_cleanup);
4502 bool do_ab_cleanup = !bitmap_empty_p (need_ab_cleanup);
4504 if (do_eh_cleanup)
4505 gimple_purge_all_dead_eh_edges (need_eh_cleanup);
4507 if (do_ab_cleanup)
4508 gimple_purge_all_dead_abnormal_call_edges (need_ab_cleanup);
4510 BITMAP_FREE (need_eh_cleanup);
4511 BITMAP_FREE (need_ab_cleanup);
4513 if (do_eh_cleanup || do_ab_cleanup)
4514 return TODO_cleanup_cfg;
4515 return 0;
4518 /* Borrow a bit of tree-ssa-dce.c for the moment.
4519 XXX: In 4.1, we should be able to just run a DCE pass after PRE, though
4520 this may be a bit faster, and we may want critical edges kept split. */
4522 /* If OP's defining statement has not already been determined to be necessary,
4523 mark that statement necessary. Return the stmt, if it is newly
4524 necessary. */
4526 static inline gimple
4527 mark_operand_necessary (tree op)
4529 gimple stmt;
4531 gcc_assert (op);
4533 if (TREE_CODE (op) != SSA_NAME)
4534 return NULL;
4536 stmt = SSA_NAME_DEF_STMT (op);
4537 gcc_assert (stmt);
4539 if (gimple_plf (stmt, NECESSARY)
4540 || gimple_nop_p (stmt))
4541 return NULL;
4543 gimple_set_plf (stmt, NECESSARY, true);
4544 return stmt;
4547 /* Because we don't follow exactly the standard PRE algorithm, and decide not
4548 to insert PHI nodes sometimes, and because value numbering of casts isn't
4549 perfect, we sometimes end up inserting dead code. This simple DCE-like
4550 pass removes any insertions we made that weren't actually used. */
4552 static void
4553 remove_dead_inserted_code (void)
4555 bitmap worklist;
4556 unsigned i;
4557 bitmap_iterator bi;
4558 gimple t;
4560 worklist = BITMAP_ALLOC (NULL);
4561 EXECUTE_IF_SET_IN_BITMAP (inserted_exprs, 0, i, bi)
4563 t = SSA_NAME_DEF_STMT (ssa_name (i));
4564 if (gimple_plf (t, NECESSARY))
4565 bitmap_set_bit (worklist, i);
4567 while (!bitmap_empty_p (worklist))
4569 i = bitmap_first_set_bit (worklist);
4570 bitmap_clear_bit (worklist, i);
4571 t = SSA_NAME_DEF_STMT (ssa_name (i));
4573 /* PHI nodes are somewhat special in that each PHI alternative has
4574 data and control dependencies. All the statements feeding the
4575 PHI node's arguments are always necessary. */
4576 if (gimple_code (t) == GIMPLE_PHI)
4578 unsigned k;
4580 for (k = 0; k < gimple_phi_num_args (t); k++)
4582 tree arg = PHI_ARG_DEF (t, k);
4583 if (TREE_CODE (arg) == SSA_NAME)
4585 gimple n = mark_operand_necessary (arg);
4586 if (n)
4587 bitmap_set_bit (worklist, SSA_NAME_VERSION (arg));
4591 else
4593 /* Propagate through the operands. Examine all the USE, VUSE and
4594 VDEF operands in this statement. Mark all the statements
4595 which feed this statement's uses as necessary. */
4596 ssa_op_iter iter;
4597 tree use;
4599 /* The operands of VDEF expressions are also needed as they
4600 represent potential definitions that may reach this
4601 statement (VDEF operands allow us to follow def-def
4602 links). */
4604 FOR_EACH_SSA_TREE_OPERAND (use, t, iter, SSA_OP_ALL_USES)
4606 gimple n = mark_operand_necessary (use);
4607 if (n)
4608 bitmap_set_bit (worklist, SSA_NAME_VERSION (use));
4613 EXECUTE_IF_SET_IN_BITMAP (inserted_exprs, 0, i, bi)
4615 t = SSA_NAME_DEF_STMT (ssa_name (i));
4616 if (!gimple_plf (t, NECESSARY))
4618 gimple_stmt_iterator gsi;
4620 if (dump_file && (dump_flags & TDF_DETAILS))
4622 fprintf (dump_file, "Removing unnecessary insertion:");
4623 print_gimple_stmt (dump_file, t, 0, 0);
4626 gsi = gsi_for_stmt (t);
4627 if (gimple_code (t) == GIMPLE_PHI)
4628 remove_phi_node (&gsi, true);
4629 else
4631 gsi_remove (&gsi, true);
4632 release_defs (t);
4636 BITMAP_FREE (worklist);
4640 /* Initialize data structures used by PRE. */
4642 static void
4643 init_pre (void)
4645 basic_block bb;
4647 next_expression_id = 1;
4648 expressions.create (0);
4649 expressions.safe_push (NULL);
4650 value_expressions.create (get_max_value_id () + 1);
4651 value_expressions.safe_grow_cleared (get_max_value_id () + 1);
4652 name_to_id.create (0);
4654 inserted_exprs = BITMAP_ALLOC (NULL);
4656 connect_infinite_loops_to_exit ();
4657 memset (&pre_stats, 0, sizeof (pre_stats));
4659 postorder = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
4660 postorder_num = inverted_post_order_compute (postorder);
4662 alloc_aux_for_blocks (sizeof (struct bb_bitmap_sets));
4664 calculate_dominance_info (CDI_POST_DOMINATORS);
4665 calculate_dominance_info (CDI_DOMINATORS);
4667 bitmap_obstack_initialize (&grand_bitmap_obstack);
4668 phi_translate_table.create (5110);
4669 expression_to_id.create (num_ssa_names * 3);
4670 bitmap_set_pool = create_alloc_pool ("Bitmap sets",
4671 sizeof (struct bitmap_set), 30);
4672 pre_expr_pool = create_alloc_pool ("pre_expr nodes",
4673 sizeof (struct pre_expr_d), 30);
4674 FOR_ALL_BB_FN (bb, cfun)
4676 EXP_GEN (bb) = bitmap_set_new ();
4677 PHI_GEN (bb) = bitmap_set_new ();
4678 TMP_GEN (bb) = bitmap_set_new ();
4679 AVAIL_OUT (bb) = bitmap_set_new ();
4684 /* Deallocate data structures used by PRE. */
4686 static void
4687 fini_pre ()
4689 free (postorder);
4690 value_expressions.release ();
4691 BITMAP_FREE (inserted_exprs);
4692 bitmap_obstack_release (&grand_bitmap_obstack);
4693 free_alloc_pool (bitmap_set_pool);
4694 free_alloc_pool (pre_expr_pool);
4695 phi_translate_table.dispose ();
4696 expression_to_id.dispose ();
4697 name_to_id.release ();
4699 free_aux_for_blocks ();
4701 free_dominance_info (CDI_POST_DOMINATORS);
4704 /* Gate and execute functions for PRE. */
4706 static unsigned int
4707 do_pre (void)
4709 unsigned int todo = 0;
4711 do_partial_partial =
4712 flag_tree_partial_pre && optimize_function_for_speed_p (cfun);
4714 /* This has to happen before SCCVN runs because
4715 loop_optimizer_init may create new phis, etc. */
4716 loop_optimizer_init (LOOPS_NORMAL);
4718 if (!run_scc_vn (VN_WALK))
4720 loop_optimizer_finalize ();
4721 return 0;
4724 init_pre ();
4725 scev_initialize ();
4727 /* Collect and value number expressions computed in each basic block. */
4728 compute_avail ();
4730 /* Insert can get quite slow on an incredibly large number of basic
4731 blocks due to some quadratic behavior. Until this behavior is
4732 fixed, don't run it when he have an incredibly large number of
4733 bb's. If we aren't going to run insert, there is no point in
4734 computing ANTIC, either, even though it's plenty fast. */
4735 if (n_basic_blocks_for_fn (cfun) < 4000)
4737 compute_antic ();
4738 insert ();
4741 /* Make sure to remove fake edges before committing our inserts.
4742 This makes sure we don't end up with extra critical edges that
4743 we would need to split. */
4744 remove_fake_exit_edges ();
4745 gsi_commit_edge_inserts ();
4747 /* Remove all the redundant expressions. */
4748 todo |= eliminate ();
4750 statistics_counter_event (cfun, "Insertions", pre_stats.insertions);
4751 statistics_counter_event (cfun, "PA inserted", pre_stats.pa_insert);
4752 statistics_counter_event (cfun, "New PHIs", pre_stats.phis);
4753 statistics_counter_event (cfun, "Eliminated", pre_stats.eliminations);
4755 clear_expression_ids ();
4756 remove_dead_inserted_code ();
4757 todo |= TODO_verify_flow;
4759 scev_finalize ();
4760 fini_pre ();
4761 todo |= fini_eliminate ();
4762 loop_optimizer_finalize ();
4764 /* TODO: tail_merge_optimize may merge all predecessors of a block, in which
4765 case we can merge the block with the remaining predecessor of the block.
4766 It should either:
4767 - call merge_blocks after each tail merge iteration
4768 - call merge_blocks after all tail merge iterations
4769 - mark TODO_cleanup_cfg when necessary
4770 - share the cfg cleanup with fini_pre. */
4771 todo |= tail_merge_optimize (todo);
4773 free_scc_vn ();
4775 /* Tail merging invalidates the virtual SSA web, together with
4776 cfg-cleanup opportunities exposed by PRE this will wreck the
4777 SSA updating machinery. So make sure to run update-ssa
4778 manually, before eventually scheduling cfg-cleanup as part of
4779 the todo. */
4780 update_ssa (TODO_update_ssa_only_virtuals);
4782 return todo;
4785 static bool
4786 gate_pre (void)
4788 return flag_tree_pre != 0;
4791 namespace {
4793 const pass_data pass_data_pre =
4795 GIMPLE_PASS, /* type */
4796 "pre", /* name */
4797 OPTGROUP_NONE, /* optinfo_flags */
4798 true, /* has_gate */
4799 true, /* has_execute */
4800 TV_TREE_PRE, /* tv_id */
4801 /* PROP_no_crit_edges is ensured by placing pass_split_crit_edges before
4802 pass_pre. */
4803 ( PROP_no_crit_edges | PROP_cfg | PROP_ssa ), /* properties_required */
4804 0, /* properties_provided */
4805 PROP_no_crit_edges, /* properties_destroyed */
4806 TODO_rebuild_alias, /* todo_flags_start */
4807 TODO_verify_ssa, /* todo_flags_finish */
4810 class pass_pre : public gimple_opt_pass
4812 public:
4813 pass_pre (gcc::context *ctxt)
4814 : gimple_opt_pass (pass_data_pre, ctxt)
4817 /* opt_pass methods: */
4818 bool gate () { return gate_pre (); }
4819 unsigned int execute () { return do_pre (); }
4821 }; // class pass_pre
4823 } // anon namespace
4825 gimple_opt_pass *
4826 make_pass_pre (gcc::context *ctxt)
4828 return new pass_pre (ctxt);
4832 /* Gate and execute functions for FRE. */
4834 static unsigned int
4835 execute_fre (void)
4837 unsigned int todo = 0;
4839 if (!run_scc_vn (VN_WALKREWRITE))
4840 return 0;
4842 memset (&pre_stats, 0, sizeof (pre_stats));
4844 /* Remove all the redundant expressions. */
4845 todo |= eliminate ();
4847 todo |= fini_eliminate ();
4849 free_scc_vn ();
4851 statistics_counter_event (cfun, "Insertions", pre_stats.insertions);
4852 statistics_counter_event (cfun, "Eliminated", pre_stats.eliminations);
4854 return todo;
4857 static bool
4858 gate_fre (void)
4860 return flag_tree_fre != 0;
4863 namespace {
4865 const pass_data pass_data_fre =
4867 GIMPLE_PASS, /* type */
4868 "fre", /* name */
4869 OPTGROUP_NONE, /* optinfo_flags */
4870 true, /* has_gate */
4871 true, /* has_execute */
4872 TV_TREE_FRE, /* tv_id */
4873 ( PROP_cfg | PROP_ssa ), /* properties_required */
4874 0, /* properties_provided */
4875 0, /* properties_destroyed */
4876 0, /* todo_flags_start */
4877 TODO_verify_ssa, /* todo_flags_finish */
4880 class pass_fre : public gimple_opt_pass
4882 public:
4883 pass_fre (gcc::context *ctxt)
4884 : gimple_opt_pass (pass_data_fre, ctxt)
4887 /* opt_pass methods: */
4888 opt_pass * clone () { return new pass_fre (m_ctxt); }
4889 bool gate () { return gate_fre (); }
4890 unsigned int execute () { return execute_fre (); }
4892 }; // class pass_fre
4894 } // anon namespace
4896 gimple_opt_pass *
4897 make_pass_fre (gcc::context *ctxt)
4899 return new pass_fre (ctxt);