- Remove SVN merge marker.
[official-gcc.git] / gcc / ipa-inline-analysis.c
blobc6f127ebbd238e2744ac7816d97e6bcd51058559
1 /* Inlining decision heuristics.
2 Copyright (C) 2003-2013 Free Software Foundation, Inc.
3 Contributed by Jan Hubicka
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 3, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
21 /* Analysis used by the inliner and other passes limiting code size growth.
23 We estimate for each function
24 - function body size
25 - average function execution time
26 - inlining size benefit (that is how much of function body size
27 and its call sequence is expected to disappear by inlining)
28 - inlining time benefit
29 - function frame size
30 For each call
31 - call statement size and time
33 inlinie_summary datastructures store above information locally (i.e.
34 parameters of the function itself) and globally (i.e. parameters of
35 the function created by applying all the inline decisions already
36 present in the callgraph).
38 We provide accestor to the inline_summary datastructure and
39 basic logic updating the parameters when inlining is performed.
41 The summaries are context sensitive. Context means
42 1) partial assignment of known constant values of operands
43 2) whether function is inlined into the call or not.
44 It is easy to add more variants. To represent function size and time
45 that depends on context (i.e. it is known to be optimized away when
46 context is known either by inlining or from IP-CP and clonning),
47 we use predicates. Predicates are logical formulas in
48 conjunctive-disjunctive form consisting of clauses. Clauses are bitmaps
49 specifying what conditions must be true. Conditions are simple test
50 of the form described above.
52 In order to make predicate (possibly) true, all of its clauses must
53 be (possibly) true. To make clause (possibly) true, one of conditions
54 it mentions must be (possibly) true. There are fixed bounds on
55 number of clauses and conditions and all the manipulation functions
56 are conservative in positive direction. I.e. we may lose precision
57 by thinking that predicate may be true even when it is not.
59 estimate_edge_size and estimate_edge_growth can be used to query
60 function size/time in the given context. inline_merge_summary merges
61 properties of caller and callee after inlining.
63 Finally pass_inline_parameters is exported. This is used to drive
64 computation of function parameters used by the early inliner. IPA
65 inlined performs analysis via its analyze_function method. */
67 #include "config.h"
68 #include "system.h"
69 #include "coretypes.h"
70 #include "tm.h"
71 #include "tree.h"
72 #include "tree-inline.h"
73 #include "langhooks.h"
74 #include "flags.h"
75 #include "cgraph.h"
76 #include "diagnostic.h"
77 #include "gimple-pretty-print.h"
78 #include "params.h"
79 #include "tree-pass.h"
80 #include "coverage.h"
81 #include "ggc.h"
82 #include "tree-flow.h"
83 #include "ipa-prop.h"
84 #include "lto-streamer.h"
85 #include "data-streamer.h"
86 #include "tree-streamer.h"
87 #include "ipa-inline.h"
88 #include "alloc-pool.h"
89 #include "cfgloop.h"
90 #include "cfgloop.h"
91 #include "tree-scalar-evolution.h"
93 /* Estimate runtime of function can easilly run into huge numbers with many
94 nested loops. Be sure we can compute time * INLINE_SIZE_SCALE * 2 in an
95 integer. For anything larger we use gcov_type. */
96 #define MAX_TIME 500000
98 /* Number of bits in integer, but we really want to be stable across different
99 hosts. */
100 #define NUM_CONDITIONS 32
102 enum predicate_conditions
104 predicate_false_condition = 0,
105 predicate_not_inlined_condition = 1,
106 predicate_first_dynamic_condition = 2
109 /* Special condition code we use to represent test that operand is compile time
110 constant. */
111 #define IS_NOT_CONSTANT ERROR_MARK
112 /* Special condition code we use to represent test that operand is not changed
113 across invocation of the function. When operand IS_NOT_CONSTANT it is always
114 CHANGED, however i.e. loop invariants can be NOT_CHANGED given percentage
115 of executions even when they are not compile time constants. */
116 #define CHANGED IDENTIFIER_NODE
118 /* Holders of ipa cgraph hooks: */
119 static struct cgraph_node_hook_list *function_insertion_hook_holder;
120 static struct cgraph_node_hook_list *node_removal_hook_holder;
121 static struct cgraph_2node_hook_list *node_duplication_hook_holder;
122 static struct cgraph_2edge_hook_list *edge_duplication_hook_holder;
123 static struct cgraph_edge_hook_list *edge_removal_hook_holder;
124 static void inline_node_removal_hook (struct cgraph_node *, void *);
125 static void inline_node_duplication_hook (struct cgraph_node *,
126 struct cgraph_node *, void *);
127 static void inline_edge_removal_hook (struct cgraph_edge *, void *);
128 static void inline_edge_duplication_hook (struct cgraph_edge *,
129 struct cgraph_edge *, void *);
131 /* VECtor holding inline summaries.
132 In GGC memory because conditions might point to constant trees. */
133 vec<inline_summary_t, va_gc> *inline_summary_vec;
134 vec<inline_edge_summary_t> inline_edge_summary_vec;
136 /* Cached node/edge growths. */
137 vec<int> node_growth_cache;
138 vec<edge_growth_cache_entry> edge_growth_cache;
140 /* Edge predicates goes here. */
141 static alloc_pool edge_predicate_pool;
143 /* Return true predicate (tautology).
144 We represent it by empty list of clauses. */
146 static inline struct predicate
147 true_predicate (void)
149 struct predicate p;
150 p.clause[0] = 0;
151 return p;
155 /* Return predicate testing single condition number COND. */
157 static inline struct predicate
158 single_cond_predicate (int cond)
160 struct predicate p;
161 p.clause[0] = 1 << cond;
162 p.clause[1] = 0;
163 return p;
167 /* Return false predicate. First clause require false condition. */
169 static inline struct predicate
170 false_predicate (void)
172 return single_cond_predicate (predicate_false_condition);
176 /* Return true if P is (false). */
178 static inline bool
179 true_predicate_p (struct predicate *p)
181 return !p->clause[0];
185 /* Return true if P is (false). */
187 static inline bool
188 false_predicate_p (struct predicate *p)
190 if (p->clause[0] == (1 << predicate_false_condition))
192 gcc_checking_assert (!p->clause[1]
193 && p->clause[0] == 1 << predicate_false_condition);
194 return true;
196 return false;
200 /* Return predicate that is set true when function is not inlined. */
202 static inline struct predicate
203 not_inlined_predicate (void)
205 return single_cond_predicate (predicate_not_inlined_condition);
208 /* Simple description of whether a memory load or a condition refers to a load
209 from an aggregate and if so, how and where from in the aggregate.
210 Individual fields have the same meaning like fields with the same name in
211 struct condition. */
213 struct agg_position_info
215 HOST_WIDE_INT offset;
216 bool agg_contents;
217 bool by_ref;
220 /* Add condition to condition list CONDS. AGGPOS describes whether the used
221 oprand is loaded from an aggregate and where in the aggregate it is. It can
222 be NULL, which means this not a load from an aggregate. */
224 static struct predicate
225 add_condition (struct inline_summary *summary, int operand_num,
226 struct agg_position_info *aggpos,
227 enum tree_code code, tree val)
229 int i;
230 struct condition *c;
231 struct condition new_cond;
232 HOST_WIDE_INT offset;
233 bool agg_contents, by_ref;
235 if (aggpos)
237 offset = aggpos->offset;
238 agg_contents = aggpos->agg_contents;
239 by_ref = aggpos->by_ref;
241 else
243 offset = 0;
244 agg_contents = false;
245 by_ref = false;
248 gcc_checking_assert (operand_num >= 0);
249 for (i = 0; vec_safe_iterate (summary->conds, i, &c); i++)
251 if (c->operand_num == operand_num
252 && c->code == code
253 && c->val == val
254 && c->agg_contents == agg_contents
255 && (!agg_contents || (c->offset == offset && c->by_ref == by_ref)))
256 return single_cond_predicate (i + predicate_first_dynamic_condition);
258 /* Too many conditions. Give up and return constant true. */
259 if (i == NUM_CONDITIONS - predicate_first_dynamic_condition)
260 return true_predicate ();
262 new_cond.operand_num = operand_num;
263 new_cond.code = code;
264 new_cond.val = val;
265 new_cond.agg_contents = agg_contents;
266 new_cond.by_ref = by_ref;
267 new_cond.offset = offset;
268 vec_safe_push (summary->conds, new_cond);
269 return single_cond_predicate (i + predicate_first_dynamic_condition);
273 /* Add clause CLAUSE into the predicate P. */
275 static inline void
276 add_clause (conditions conditions, struct predicate *p, clause_t clause)
278 int i;
279 int i2;
280 int insert_here = -1;
281 int c1, c2;
283 /* True clause. */
284 if (!clause)
285 return;
287 /* False clause makes the whole predicate false. Kill the other variants. */
288 if (clause == (1 << predicate_false_condition))
290 p->clause[0] = (1 << predicate_false_condition);
291 p->clause[1] = 0;
292 return;
294 if (false_predicate_p (p))
295 return;
297 /* No one should be sily enough to add false into nontrivial clauses. */
298 gcc_checking_assert (!(clause & (1 << predicate_false_condition)));
300 /* Look where to insert the clause. At the same time prune out
301 clauses of P that are implied by the new clause and thus
302 redundant. */
303 for (i = 0, i2 = 0; i <= MAX_CLAUSES; i++)
305 p->clause[i2] = p->clause[i];
307 if (!p->clause[i])
308 break;
310 /* If p->clause[i] implies clause, there is nothing to add. */
311 if ((p->clause[i] & clause) == p->clause[i])
313 /* We had nothing to add, none of clauses should've become
314 redundant. */
315 gcc_checking_assert (i == i2);
316 return;
319 if (p->clause[i] < clause && insert_here < 0)
320 insert_here = i2;
322 /* If clause implies p->clause[i], then p->clause[i] becomes redundant.
323 Otherwise the p->clause[i] has to stay. */
324 if ((p->clause[i] & clause) != clause)
325 i2++;
328 /* Look for clauses that are obviously true. I.e.
329 op0 == 5 || op0 != 5. */
330 for (c1 = predicate_first_dynamic_condition; c1 < NUM_CONDITIONS; c1++)
332 condition *cc1;
333 if (!(clause & (1 << c1)))
334 continue;
335 cc1 = &(*conditions)[c1 - predicate_first_dynamic_condition];
336 /* We have no way to represent !CHANGED and !IS_NOT_CONSTANT
337 and thus there is no point for looking for them. */
338 if (cc1->code == CHANGED || cc1->code == IS_NOT_CONSTANT)
339 continue;
340 for (c2 = c1 + 1; c2 <= NUM_CONDITIONS; c2++)
341 if (clause & (1 << c2))
343 condition *cc1 =
344 &(*conditions)[c1 - predicate_first_dynamic_condition];
345 condition *cc2 =
346 &(*conditions)[c2 - predicate_first_dynamic_condition];
347 if (cc1->operand_num == cc2->operand_num
348 && cc1->val == cc2->val
349 && cc2->code != IS_NOT_CONSTANT
350 && cc2->code != CHANGED
351 && cc1->code == invert_tree_comparison
352 (cc2->code,
353 HONOR_NANS (TYPE_MODE (TREE_TYPE (cc1->val)))))
354 return;
359 /* We run out of variants. Be conservative in positive direction. */
360 if (i2 == MAX_CLAUSES)
361 return;
362 /* Keep clauses in decreasing order. This makes equivalence testing easy. */
363 p->clause[i2 + 1] = 0;
364 if (insert_here >= 0)
365 for (; i2 > insert_here; i2--)
366 p->clause[i2] = p->clause[i2 - 1];
367 else
368 insert_here = i2;
369 p->clause[insert_here] = clause;
373 /* Return P & P2. */
375 static struct predicate
376 and_predicates (conditions conditions,
377 struct predicate *p, struct predicate *p2)
379 struct predicate out = *p;
380 int i;
382 /* Avoid busy work. */
383 if (false_predicate_p (p2) || true_predicate_p (p))
384 return *p2;
385 if (false_predicate_p (p) || true_predicate_p (p2))
386 return *p;
388 /* See how far predicates match. */
389 for (i = 0; p->clause[i] && p->clause[i] == p2->clause[i]; i++)
391 gcc_checking_assert (i < MAX_CLAUSES);
394 /* Combine the predicates rest. */
395 for (; p2->clause[i]; i++)
397 gcc_checking_assert (i < MAX_CLAUSES);
398 add_clause (conditions, &out, p2->clause[i]);
400 return out;
404 /* Return true if predicates are obviously equal. */
406 static inline bool
407 predicates_equal_p (struct predicate *p, struct predicate *p2)
409 int i;
410 for (i = 0; p->clause[i]; i++)
412 gcc_checking_assert (i < MAX_CLAUSES);
413 gcc_checking_assert (p->clause[i] > p->clause[i + 1]);
414 gcc_checking_assert (!p2->clause[i]
415 || p2->clause[i] > p2->clause[i + 1]);
416 if (p->clause[i] != p2->clause[i])
417 return false;
419 return !p2->clause[i];
423 /* Return P | P2. */
425 static struct predicate
426 or_predicates (conditions conditions,
427 struct predicate *p, struct predicate *p2)
429 struct predicate out = true_predicate ();
430 int i, j;
432 /* Avoid busy work. */
433 if (false_predicate_p (p2) || true_predicate_p (p))
434 return *p;
435 if (false_predicate_p (p) || true_predicate_p (p2))
436 return *p2;
437 if (predicates_equal_p (p, p2))
438 return *p;
440 /* OK, combine the predicates. */
441 for (i = 0; p->clause[i]; i++)
442 for (j = 0; p2->clause[j]; j++)
444 gcc_checking_assert (i < MAX_CLAUSES && j < MAX_CLAUSES);
445 add_clause (conditions, &out, p->clause[i] | p2->clause[j]);
447 return out;
451 /* Having partial truth assignment in POSSIBLE_TRUTHS, return false
452 if predicate P is known to be false. */
454 static bool
455 evaluate_predicate (struct predicate *p, clause_t possible_truths)
457 int i;
459 /* True remains true. */
460 if (true_predicate_p (p))
461 return true;
463 gcc_assert (!(possible_truths & (1 << predicate_false_condition)));
465 /* See if we can find clause we can disprove. */
466 for (i = 0; p->clause[i]; i++)
468 gcc_checking_assert (i < MAX_CLAUSES);
469 if (!(p->clause[i] & possible_truths))
470 return false;
472 return true;
475 /* Return the probability in range 0...REG_BR_PROB_BASE that the predicated
476 instruction will be recomputed per invocation of the inlined call. */
478 static int
479 predicate_probability (conditions conds,
480 struct predicate *p, clause_t possible_truths,
481 vec<inline_param_summary_t> inline_param_summary)
483 int i;
484 int combined_prob = REG_BR_PROB_BASE;
486 /* True remains true. */
487 if (true_predicate_p (p))
488 return REG_BR_PROB_BASE;
490 if (false_predicate_p (p))
491 return 0;
493 gcc_assert (!(possible_truths & (1 << predicate_false_condition)));
495 /* See if we can find clause we can disprove. */
496 for (i = 0; p->clause[i]; i++)
498 gcc_checking_assert (i < MAX_CLAUSES);
499 if (!(p->clause[i] & possible_truths))
500 return 0;
501 else
503 int this_prob = 0;
504 int i2;
505 if (!inline_param_summary.exists ())
506 return REG_BR_PROB_BASE;
507 for (i2 = 0; i2 < NUM_CONDITIONS; i2++)
508 if ((p->clause[i] & possible_truths) & (1 << i2))
510 if (i2 >= predicate_first_dynamic_condition)
512 condition *c =
513 &(*conds)[i2 - predicate_first_dynamic_condition];
514 if (c->code == CHANGED
515 && (c->operand_num <
516 (int) inline_param_summary.length ()))
518 int iprob =
519 inline_param_summary[c->operand_num].change_prob;
520 this_prob = MAX (this_prob, iprob);
522 else
523 this_prob = REG_BR_PROB_BASE;
525 else
526 this_prob = REG_BR_PROB_BASE;
528 combined_prob = MIN (this_prob, combined_prob);
529 if (!combined_prob)
530 return 0;
533 return combined_prob;
537 /* Dump conditional COND. */
539 static void
540 dump_condition (FILE *f, conditions conditions, int cond)
542 condition *c;
543 if (cond == predicate_false_condition)
544 fprintf (f, "false");
545 else if (cond == predicate_not_inlined_condition)
546 fprintf (f, "not inlined");
547 else
549 c = &(*conditions)[cond - predicate_first_dynamic_condition];
550 fprintf (f, "op%i", c->operand_num);
551 if (c->agg_contents)
552 fprintf (f, "[%soffset: " HOST_WIDE_INT_PRINT_DEC "]",
553 c->by_ref ? "ref " : "", c->offset);
554 if (c->code == IS_NOT_CONSTANT)
556 fprintf (f, " not constant");
557 return;
559 if (c->code == CHANGED)
561 fprintf (f, " changed");
562 return;
564 fprintf (f, " %s ", op_symbol_code (c->code));
565 print_generic_expr (f, c->val, 1);
570 /* Dump clause CLAUSE. */
572 static void
573 dump_clause (FILE *f, conditions conds, clause_t clause)
575 int i;
576 bool found = false;
577 fprintf (f, "(");
578 if (!clause)
579 fprintf (f, "true");
580 for (i = 0; i < NUM_CONDITIONS; i++)
581 if (clause & (1 << i))
583 if (found)
584 fprintf (f, " || ");
585 found = true;
586 dump_condition (f, conds, i);
588 fprintf (f, ")");
592 /* Dump predicate PREDICATE. */
594 static void
595 dump_predicate (FILE *f, conditions conds, struct predicate *pred)
597 int i;
598 if (true_predicate_p (pred))
599 dump_clause (f, conds, 0);
600 else
601 for (i = 0; pred->clause[i]; i++)
603 if (i)
604 fprintf (f, " && ");
605 dump_clause (f, conds, pred->clause[i]);
607 fprintf (f, "\n");
611 /* Dump inline hints. */
612 void
613 dump_inline_hints (FILE *f, inline_hints hints)
615 if (!hints)
616 return;
617 fprintf (f, "inline hints:");
618 if (hints & INLINE_HINT_indirect_call)
620 hints &= ~INLINE_HINT_indirect_call;
621 fprintf (f, " indirect_call");
623 if (hints & INLINE_HINT_loop_iterations)
625 hints &= ~INLINE_HINT_loop_iterations;
626 fprintf (f, " loop_iterations");
628 if (hints & INLINE_HINT_loop_stride)
630 hints &= ~INLINE_HINT_loop_stride;
631 fprintf (f, " loop_stride");
633 if (hints & INLINE_HINT_same_scc)
635 hints &= ~INLINE_HINT_same_scc;
636 fprintf (f, " same_scc");
638 if (hints & INLINE_HINT_in_scc)
640 hints &= ~INLINE_HINT_in_scc;
641 fprintf (f, " in_scc");
643 if (hints & INLINE_HINT_cross_module)
645 hints &= ~INLINE_HINT_cross_module;
646 fprintf (f, " cross_module");
648 if (hints & INLINE_HINT_declared_inline)
650 hints &= ~INLINE_HINT_declared_inline;
651 fprintf (f, " declared_inline");
653 if (hints & INLINE_HINT_array_index)
655 hints &= ~INLINE_HINT_array_index;
656 fprintf (f, " array_index");
658 gcc_assert (!hints);
662 /* Record SIZE and TIME under condition PRED into the inline summary. */
664 static void
665 account_size_time (struct inline_summary *summary, int size, int time,
666 struct predicate *pred)
668 size_time_entry *e;
669 bool found = false;
670 int i;
672 if (false_predicate_p (pred))
673 return;
675 /* We need to create initial empty unconitional clause, but otherwie
676 we don't need to account empty times and sizes. */
677 if (!size && !time && summary->entry)
678 return;
680 /* Watch overflow that might result from insane profiles. */
681 if (time > MAX_TIME * INLINE_TIME_SCALE)
682 time = MAX_TIME * INLINE_TIME_SCALE;
683 gcc_assert (time >= 0);
685 for (i = 0; vec_safe_iterate (summary->entry, i, &e); i++)
686 if (predicates_equal_p (&e->predicate, pred))
688 found = true;
689 break;
691 if (i == 256)
693 i = 0;
694 found = true;
695 e = &(*summary->entry)[0];
696 gcc_assert (!e->predicate.clause[0]);
697 if (dump_file && (dump_flags & TDF_DETAILS))
698 fprintf (dump_file,
699 "\t\tReached limit on number of entries, "
700 "ignoring the predicate.");
702 if (dump_file && (dump_flags & TDF_DETAILS) && (time || size))
704 fprintf (dump_file,
705 "\t\tAccounting size:%3.2f, time:%3.2f on %spredicate:",
706 ((double) size) / INLINE_SIZE_SCALE,
707 ((double) time) / INLINE_TIME_SCALE, found ? "" : "new ");
708 dump_predicate (dump_file, summary->conds, pred);
710 if (!found)
712 struct size_time_entry new_entry;
713 new_entry.size = size;
714 new_entry.time = time;
715 new_entry.predicate = *pred;
716 vec_safe_push (summary->entry, new_entry);
718 else
720 e->size += size;
721 e->time += time;
722 if (e->time > MAX_TIME * INLINE_TIME_SCALE)
723 e->time = MAX_TIME * INLINE_TIME_SCALE;
727 /* Set predicate for edge E. */
729 static void
730 edge_set_predicate (struct cgraph_edge *e, struct predicate *predicate)
732 struct inline_edge_summary *es = inline_edge_summary (e);
733 if (predicate && !true_predicate_p (predicate))
735 if (!es->predicate)
736 es->predicate = (struct predicate *) pool_alloc (edge_predicate_pool);
737 *es->predicate = *predicate;
739 else
741 if (es->predicate)
742 pool_free (edge_predicate_pool, es->predicate);
743 es->predicate = NULL;
747 /* Set predicate for hint *P. */
749 static void
750 set_hint_predicate (struct predicate **p, struct predicate new_predicate)
752 if (false_predicate_p (&new_predicate) || true_predicate_p (&new_predicate))
754 if (*p)
755 pool_free (edge_predicate_pool, *p);
756 *p = NULL;
758 else
760 if (!*p)
761 *p = (struct predicate *) pool_alloc (edge_predicate_pool);
762 **p = new_predicate;
767 /* KNOWN_VALS is partial mapping of parameters of NODE to constant values.
768 KNOWN_AGGS is a vector of aggreggate jump functions for each parameter.
769 Return clause of possible truths. When INLINE_P is true, assume that we are
770 inlining.
772 ERROR_MARK means compile time invariant. */
774 static clause_t
775 evaluate_conditions_for_known_args (struct cgraph_node *node,
776 bool inline_p,
777 vec<tree> known_vals,
778 vec<ipa_agg_jump_function_p>
779 known_aggs)
781 clause_t clause = inline_p ? 0 : 1 << predicate_not_inlined_condition;
782 struct inline_summary *info = inline_summary (node);
783 int i;
784 struct condition *c;
786 for (i = 0; vec_safe_iterate (info->conds, i, &c); i++)
788 tree val;
789 tree res;
791 /* We allow call stmt to have fewer arguments than the callee function
792 (especially for K&R style programs). So bound check here (we assume
793 known_aggs vector, if non-NULL, has the same length as
794 known_vals). */
795 gcc_checking_assert (!known_aggs.exists ()
796 || (known_vals.length () == known_aggs.length ()));
797 if (c->operand_num >= (int) known_vals.length ())
799 clause |= 1 << (i + predicate_first_dynamic_condition);
800 continue;
803 if (c->agg_contents)
805 struct ipa_agg_jump_function *agg;
807 if (c->code == CHANGED
808 && !c->by_ref
809 && (known_vals[c->operand_num] == error_mark_node))
810 continue;
812 if (known_aggs.exists ())
814 agg = known_aggs[c->operand_num];
815 val = ipa_find_agg_cst_for_param (agg, c->offset, c->by_ref);
817 else
818 val = NULL_TREE;
820 else
822 val = known_vals[c->operand_num];
823 if (val == error_mark_node && c->code != CHANGED)
824 val = NULL_TREE;
827 if (!val)
829 clause |= 1 << (i + predicate_first_dynamic_condition);
830 continue;
832 if (c->code == IS_NOT_CONSTANT || c->code == CHANGED)
833 continue;
834 res = fold_binary_to_constant (c->code, boolean_type_node, val, c->val);
835 if (res && integer_zerop (res))
836 continue;
837 clause |= 1 << (i + predicate_first_dynamic_condition);
839 return clause;
843 /* Work out what conditions might be true at invocation of E. */
845 static void
846 evaluate_properties_for_edge (struct cgraph_edge *e, bool inline_p,
847 clause_t *clause_ptr,
848 vec<tree> *known_vals_ptr,
849 vec<tree> *known_binfos_ptr,
850 vec<ipa_agg_jump_function_p> *known_aggs_ptr)
852 struct cgraph_node *callee =
853 cgraph_function_or_thunk_node (e->callee, NULL);
854 struct inline_summary *info = inline_summary (callee);
855 vec<tree> known_vals = vNULL;
856 vec<ipa_agg_jump_function_p> known_aggs = vNULL;
858 if (clause_ptr)
859 *clause_ptr = inline_p ? 0 : 1 << predicate_not_inlined_condition;
860 if (known_vals_ptr)
861 known_vals_ptr->create (0);
862 if (known_binfos_ptr)
863 known_binfos_ptr->create (0);
865 if (ipa_node_params_vector.exists ()
866 && !e->call_stmt_cannot_inline_p
867 && ((clause_ptr && info->conds) || known_vals_ptr || known_binfos_ptr))
869 struct ipa_node_params *parms_info;
870 struct ipa_edge_args *args = IPA_EDGE_REF (e);
871 struct inline_edge_summary *es = inline_edge_summary (e);
872 int i, count = ipa_get_cs_argument_count (args);
874 if (e->caller->global.inlined_to)
875 parms_info = IPA_NODE_REF (e->caller->global.inlined_to);
876 else
877 parms_info = IPA_NODE_REF (e->caller);
879 if (count && (info->conds || known_vals_ptr))
880 known_vals.safe_grow_cleared (count);
881 if (count && (info->conds || known_aggs_ptr))
882 known_aggs.safe_grow_cleared (count);
883 if (count && known_binfos_ptr)
884 known_binfos_ptr->safe_grow_cleared (count);
886 for (i = 0; i < count; i++)
888 struct ipa_jump_func *jf = ipa_get_ith_jump_func (args, i);
889 tree cst = ipa_value_from_jfunc (parms_info, jf);
890 if (cst)
892 if (known_vals.exists () && TREE_CODE (cst) != TREE_BINFO)
893 known_vals[i] = cst;
894 else if (known_binfos_ptr != NULL
895 && TREE_CODE (cst) == TREE_BINFO)
896 (*known_binfos_ptr)[i] = cst;
898 else if (inline_p && !es->param[i].change_prob)
899 known_vals[i] = error_mark_node;
900 /* TODO: When IPA-CP starts propagating and merging aggregate jump
901 functions, use its knowledge of the caller too, just like the
902 scalar case above. */
903 known_aggs[i] = &jf->agg;
907 if (clause_ptr)
908 *clause_ptr = evaluate_conditions_for_known_args (callee, inline_p,
909 known_vals, known_aggs);
911 if (known_vals_ptr)
912 *known_vals_ptr = known_vals;
913 else
914 known_vals.release ();
916 if (known_aggs_ptr)
917 *known_aggs_ptr = known_aggs;
918 else
919 known_aggs.release ();
923 /* Allocate the inline summary vector or resize it to cover all cgraph nodes. */
925 static void
926 inline_summary_alloc (void)
928 if (!node_removal_hook_holder)
929 node_removal_hook_holder =
930 cgraph_add_node_removal_hook (&inline_node_removal_hook, NULL);
931 if (!edge_removal_hook_holder)
932 edge_removal_hook_holder =
933 cgraph_add_edge_removal_hook (&inline_edge_removal_hook, NULL);
934 if (!node_duplication_hook_holder)
935 node_duplication_hook_holder =
936 cgraph_add_node_duplication_hook (&inline_node_duplication_hook, NULL);
937 if (!edge_duplication_hook_holder)
938 edge_duplication_hook_holder =
939 cgraph_add_edge_duplication_hook (&inline_edge_duplication_hook, NULL);
941 if (vec_safe_length (inline_summary_vec) <= (unsigned) cgraph_max_uid)
942 vec_safe_grow_cleared (inline_summary_vec, cgraph_max_uid + 1);
943 if (inline_edge_summary_vec.length () <= (unsigned) cgraph_edge_max_uid)
944 inline_edge_summary_vec.safe_grow_cleared (cgraph_edge_max_uid + 1);
945 if (!edge_predicate_pool)
946 edge_predicate_pool = create_alloc_pool ("edge predicates",
947 sizeof (struct predicate), 10);
950 /* We are called multiple time for given function; clear
951 data from previous run so they are not cumulated. */
953 static void
954 reset_inline_edge_summary (struct cgraph_edge *e)
956 if (e->uid < (int) inline_edge_summary_vec.length ())
958 struct inline_edge_summary *es = inline_edge_summary (e);
960 es->call_stmt_size = es->call_stmt_time = 0;
961 if (es->predicate)
962 pool_free (edge_predicate_pool, es->predicate);
963 es->predicate = NULL;
964 es->param.release ();
968 /* We are called multiple time for given function; clear
969 data from previous run so they are not cumulated. */
971 static void
972 reset_inline_summary (struct cgraph_node *node)
974 struct inline_summary *info = inline_summary (node);
975 struct cgraph_edge *e;
977 info->self_size = info->self_time = 0;
978 info->estimated_stack_size = 0;
979 info->estimated_self_stack_size = 0;
980 info->stack_frame_offset = 0;
981 info->size = 0;
982 info->time = 0;
983 info->growth = 0;
984 info->scc_no = 0;
985 if (info->loop_iterations)
987 pool_free (edge_predicate_pool, info->loop_iterations);
988 info->loop_iterations = NULL;
990 if (info->loop_stride)
992 pool_free (edge_predicate_pool, info->loop_stride);
993 info->loop_stride = NULL;
995 if (info->array_index)
997 pool_free (edge_predicate_pool, info->array_index);
998 info->array_index = NULL;
1000 vec_free (info->conds);
1001 vec_free (info->entry);
1002 for (e = node->callees; e; e = e->next_callee)
1003 reset_inline_edge_summary (e);
1004 for (e = node->indirect_calls; e; e = e->next_callee)
1005 reset_inline_edge_summary (e);
1008 /* Hook that is called by cgraph.c when a node is removed. */
1010 static void
1011 inline_node_removal_hook (struct cgraph_node *node,
1012 void *data ATTRIBUTE_UNUSED)
1014 struct inline_summary *info;
1015 if (vec_safe_length (inline_summary_vec) <= (unsigned) node->uid)
1016 return;
1017 info = inline_summary (node);
1018 reset_inline_summary (node);
1019 memset (info, 0, sizeof (inline_summary_t));
1022 /* Remap predicate P of former function to be predicate of duplicated functoin.
1023 POSSIBLE_TRUTHS is clause of possible truths in the duplicated node,
1024 INFO is inline summary of the duplicated node. */
1026 static struct predicate
1027 remap_predicate_after_duplication (struct predicate *p,
1028 clause_t possible_truths,
1029 struct inline_summary *info)
1031 struct predicate new_predicate = true_predicate ();
1032 int j;
1033 for (j = 0; p->clause[j]; j++)
1034 if (!(possible_truths & p->clause[j]))
1036 new_predicate = false_predicate ();
1037 break;
1039 else
1040 add_clause (info->conds, &new_predicate,
1041 possible_truths & p->clause[j]);
1042 return new_predicate;
1045 /* Same as remap_predicate_after_duplication but handle hint predicate *P.
1046 Additionally care about allocating new memory slot for updated predicate
1047 and set it to NULL when it becomes true or false (and thus uninteresting).
1050 static void
1051 remap_hint_predicate_after_duplication (struct predicate **p,
1052 clause_t possible_truths,
1053 struct inline_summary *info)
1055 struct predicate new_predicate;
1057 if (!*p)
1058 return;
1060 new_predicate = remap_predicate_after_duplication (*p,
1061 possible_truths, info);
1062 /* We do not want to free previous predicate; it is used by node origin. */
1063 *p = NULL;
1064 set_hint_predicate (p, new_predicate);
1068 /* Hook that is called by cgraph.c when a node is duplicated. */
1070 static void
1071 inline_node_duplication_hook (struct cgraph_node *src,
1072 struct cgraph_node *dst,
1073 ATTRIBUTE_UNUSED void *data)
1075 struct inline_summary *info;
1076 inline_summary_alloc ();
1077 info = inline_summary (dst);
1078 memcpy (info, inline_summary (src), sizeof (struct inline_summary));
1079 /* TODO: as an optimization, we may avoid copying conditions
1080 that are known to be false or true. */
1081 info->conds = vec_safe_copy (info->conds);
1083 /* When there are any replacements in the function body, see if we can figure
1084 out that something was optimized out. */
1085 if (ipa_node_params_vector.exists () && dst->clone.tree_map)
1087 vec<size_time_entry, va_gc> *entry = info->entry;
1088 /* Use SRC parm info since it may not be copied yet. */
1089 struct ipa_node_params *parms_info = IPA_NODE_REF (src);
1090 vec<tree> known_vals = vNULL;
1091 int count = ipa_get_param_count (parms_info);
1092 int i, j;
1093 clause_t possible_truths;
1094 struct predicate true_pred = true_predicate ();
1095 size_time_entry *e;
1096 int optimized_out_size = 0;
1097 bool inlined_to_p = false;
1098 struct cgraph_edge *edge;
1100 info->entry = 0;
1101 known_vals.safe_grow_cleared (count);
1102 for (i = 0; i < count; i++)
1104 tree t = ipa_get_param (parms_info, i);
1105 struct ipa_replace_map *r;
1107 for (j = 0; vec_safe_iterate (dst->clone.tree_map, j, &r); j++)
1109 if (r->old_tree == t && r->replace_p && !r->ref_p)
1111 known_vals[i] = r->new_tree;
1112 break;
1116 possible_truths = evaluate_conditions_for_known_args (dst, false,
1117 known_vals,
1118 vNULL);
1119 known_vals.release ();
1121 account_size_time (info, 0, 0, &true_pred);
1123 /* Remap size_time vectors.
1124 Simplify the predicate by prunning out alternatives that are known
1125 to be false.
1126 TODO: as on optimization, we can also eliminate conditions known
1127 to be true. */
1128 for (i = 0; vec_safe_iterate (entry, i, &e); i++)
1130 struct predicate new_predicate;
1131 new_predicate = remap_predicate_after_duplication (&e->predicate,
1132 possible_truths,
1133 info);
1134 if (false_predicate_p (&new_predicate))
1135 optimized_out_size += e->size;
1136 else
1137 account_size_time (info, e->size, e->time, &new_predicate);
1140 /* Remap edge predicates with the same simplification as above.
1141 Also copy constantness arrays. */
1142 for (edge = dst->callees; edge; edge = edge->next_callee)
1144 struct predicate new_predicate;
1145 struct inline_edge_summary *es = inline_edge_summary (edge);
1147 if (!edge->inline_failed)
1148 inlined_to_p = true;
1149 if (!es->predicate)
1150 continue;
1151 new_predicate = remap_predicate_after_duplication (es->predicate,
1152 possible_truths,
1153 info);
1154 if (false_predicate_p (&new_predicate)
1155 && !false_predicate_p (es->predicate))
1157 optimized_out_size += es->call_stmt_size * INLINE_SIZE_SCALE;
1158 edge->frequency = 0;
1160 edge_set_predicate (edge, &new_predicate);
1163 /* Remap indirect edge predicates with the same simplificaiton as above.
1164 Also copy constantness arrays. */
1165 for (edge = dst->indirect_calls; edge; edge = edge->next_callee)
1167 struct predicate new_predicate;
1168 struct inline_edge_summary *es = inline_edge_summary (edge);
1170 gcc_checking_assert (edge->inline_failed);
1171 if (!es->predicate)
1172 continue;
1173 new_predicate = remap_predicate_after_duplication (es->predicate,
1174 possible_truths,
1175 info);
1176 if (false_predicate_p (&new_predicate)
1177 && !false_predicate_p (es->predicate))
1179 optimized_out_size += es->call_stmt_size * INLINE_SIZE_SCALE;
1180 edge->frequency = 0;
1182 edge_set_predicate (edge, &new_predicate);
1184 remap_hint_predicate_after_duplication (&info->loop_iterations,
1185 possible_truths, info);
1186 remap_hint_predicate_after_duplication (&info->loop_stride,
1187 possible_truths, info);
1188 remap_hint_predicate_after_duplication (&info->array_index,
1189 possible_truths, info);
1191 /* If inliner or someone after inliner will ever start producing
1192 non-trivial clones, we will get trouble with lack of information
1193 about updating self sizes, because size vectors already contains
1194 sizes of the calees. */
1195 gcc_assert (!inlined_to_p || !optimized_out_size);
1197 else
1199 info->entry = vec_safe_copy (info->entry);
1200 if (info->loop_iterations)
1202 predicate p = *info->loop_iterations;
1203 info->loop_iterations = NULL;
1204 set_hint_predicate (&info->loop_iterations, p);
1206 if (info->loop_stride)
1208 predicate p = *info->loop_stride;
1209 info->loop_stride = NULL;
1210 set_hint_predicate (&info->loop_stride, p);
1212 if (info->array_index)
1214 predicate p = *info->array_index;
1215 info->array_index = NULL;
1216 set_hint_predicate (&info->array_index, p);
1219 inline_update_overall_summary (dst);
1223 /* Hook that is called by cgraph.c when a node is duplicated. */
1225 static void
1226 inline_edge_duplication_hook (struct cgraph_edge *src,
1227 struct cgraph_edge *dst,
1228 ATTRIBUTE_UNUSED void *data)
1230 struct inline_edge_summary *info;
1231 struct inline_edge_summary *srcinfo;
1232 inline_summary_alloc ();
1233 info = inline_edge_summary (dst);
1234 srcinfo = inline_edge_summary (src);
1235 memcpy (info, srcinfo, sizeof (struct inline_edge_summary));
1236 info->predicate = NULL;
1237 edge_set_predicate (dst, srcinfo->predicate);
1238 info->param = srcinfo->param.copy ();
1242 /* Keep edge cache consistent across edge removal. */
1244 static void
1245 inline_edge_removal_hook (struct cgraph_edge *edge,
1246 void *data ATTRIBUTE_UNUSED)
1248 if (edge_growth_cache.exists ())
1249 reset_edge_growth_cache (edge);
1250 reset_inline_edge_summary (edge);
1254 /* Initialize growth caches. */
1256 void
1257 initialize_growth_caches (void)
1259 if (cgraph_edge_max_uid)
1260 edge_growth_cache.safe_grow_cleared (cgraph_edge_max_uid);
1261 if (cgraph_max_uid)
1262 node_growth_cache.safe_grow_cleared (cgraph_max_uid);
1266 /* Free growth caches. */
1268 void
1269 free_growth_caches (void)
1271 edge_growth_cache.release ();
1272 node_growth_cache.release ();
1276 /* Dump edge summaries associated to NODE and recursively to all clones.
1277 Indent by INDENT. */
1279 static void
1280 dump_inline_edge_summary (FILE *f, int indent, struct cgraph_node *node,
1281 struct inline_summary *info)
1283 struct cgraph_edge *edge;
1284 for (edge = node->callees; edge; edge = edge->next_callee)
1286 struct inline_edge_summary *es = inline_edge_summary (edge);
1287 struct cgraph_node *callee =
1288 cgraph_function_or_thunk_node (edge->callee, NULL);
1289 int i;
1291 fprintf (f,
1292 "%*s%s/%i %s\n%*s loop depth:%2i freq:%4i size:%2i"
1293 " time: %2i callee size:%2i stack:%2i",
1294 indent, "", cgraph_node_name (callee), callee->symbol.order,
1295 !edge->inline_failed
1296 ? "inlined" : cgraph_inline_failed_string (edge-> inline_failed),
1297 indent, "", es->loop_depth, edge->frequency,
1298 es->call_stmt_size, es->call_stmt_time,
1299 (int) inline_summary (callee)->size / INLINE_SIZE_SCALE,
1300 (int) inline_summary (callee)->estimated_stack_size);
1302 if (es->predicate)
1304 fprintf (f, " predicate: ");
1305 dump_predicate (f, info->conds, es->predicate);
1307 else
1308 fprintf (f, "\n");
1309 if (es->param.exists ())
1310 for (i = 0; i < (int) es->param.length (); i++)
1312 int prob = es->param[i].change_prob;
1314 if (!prob)
1315 fprintf (f, "%*s op%i is compile time invariant\n",
1316 indent + 2, "", i);
1317 else if (prob != REG_BR_PROB_BASE)
1318 fprintf (f, "%*s op%i change %f%% of time\n", indent + 2, "", i,
1319 prob * 100.0 / REG_BR_PROB_BASE);
1321 if (!edge->inline_failed)
1323 fprintf (f, "%*sStack frame offset %i, callee self size %i,"
1324 " callee size %i\n",
1325 indent + 2, "",
1326 (int) inline_summary (callee)->stack_frame_offset,
1327 (int) inline_summary (callee)->estimated_self_stack_size,
1328 (int) inline_summary (callee)->estimated_stack_size);
1329 dump_inline_edge_summary (f, indent + 2, callee, info);
1332 for (edge = node->indirect_calls; edge; edge = edge->next_callee)
1334 struct inline_edge_summary *es = inline_edge_summary (edge);
1335 fprintf (f, "%*sindirect call loop depth:%2i freq:%4i size:%2i"
1336 " time: %2i",
1337 indent, "",
1338 es->loop_depth,
1339 edge->frequency, es->call_stmt_size, es->call_stmt_time);
1340 if (es->predicate)
1342 fprintf (f, "predicate: ");
1343 dump_predicate (f, info->conds, es->predicate);
1345 else
1346 fprintf (f, "\n");
1351 void
1352 dump_inline_summary (FILE *f, struct cgraph_node *node)
1354 if (node->analyzed)
1356 struct inline_summary *s = inline_summary (node);
1357 size_time_entry *e;
1358 int i;
1359 fprintf (f, "Inline summary for %s/%i", cgraph_node_name (node),
1360 node->symbol.order);
1361 if (DECL_DISREGARD_INLINE_LIMITS (node->symbol.decl))
1362 fprintf (f, " always_inline");
1363 if (s->inlinable)
1364 fprintf (f, " inlinable");
1365 fprintf (f, "\n self time: %i\n", s->self_time);
1366 fprintf (f, " global time: %i\n", s->time);
1367 fprintf (f, " self size: %i\n", s->self_size);
1368 fprintf (f, " global size: %i\n", s->size);
1369 fprintf (f, " self stack: %i\n",
1370 (int) s->estimated_self_stack_size);
1371 fprintf (f, " global stack: %i\n", (int) s->estimated_stack_size);
1372 if (s->growth)
1373 fprintf (f, " estimated growth:%i\n", (int) s->growth);
1374 if (s->scc_no)
1375 fprintf (f, " In SCC: %i\n", (int) s->scc_no);
1376 for (i = 0; vec_safe_iterate (s->entry, i, &e); i++)
1378 fprintf (f, " size:%f, time:%f, predicate:",
1379 (double) e->size / INLINE_SIZE_SCALE,
1380 (double) e->time / INLINE_TIME_SCALE);
1381 dump_predicate (f, s->conds, &e->predicate);
1383 if (s->loop_iterations)
1385 fprintf (f, " loop iterations:");
1386 dump_predicate (f, s->conds, s->loop_iterations);
1388 if (s->loop_stride)
1390 fprintf (f, " loop stride:");
1391 dump_predicate (f, s->conds, s->loop_stride);
1393 if (s->array_index)
1395 fprintf (f, " array index:");
1396 dump_predicate (f, s->conds, s->array_index);
1398 fprintf (f, " calls:\n");
1399 dump_inline_edge_summary (f, 4, node, s);
1400 fprintf (f, "\n");
1404 DEBUG_FUNCTION void
1405 debug_inline_summary (struct cgraph_node *node)
1407 dump_inline_summary (stderr, node);
1410 void
1411 dump_inline_summaries (FILE *f)
1413 struct cgraph_node *node;
1415 FOR_EACH_DEFINED_FUNCTION (node)
1416 if (!node->global.inlined_to)
1417 dump_inline_summary (f, node);
1420 /* Give initial reasons why inlining would fail on EDGE. This gets either
1421 nullified or usually overwritten by more precise reasons later. */
1423 void
1424 initialize_inline_failed (struct cgraph_edge *e)
1426 struct cgraph_node *callee = e->callee;
1428 if (e->indirect_unknown_callee)
1429 e->inline_failed = CIF_INDIRECT_UNKNOWN_CALL;
1430 else if (!callee->analyzed)
1431 e->inline_failed = CIF_BODY_NOT_AVAILABLE;
1432 else if (callee->local.redefined_extern_inline)
1433 e->inline_failed = CIF_REDEFINED_EXTERN_INLINE;
1434 else if (e->call_stmt_cannot_inline_p)
1435 e->inline_failed = CIF_MISMATCHED_ARGUMENTS;
1436 else
1437 e->inline_failed = CIF_FUNCTION_NOT_CONSIDERED;
1440 /* Callback of walk_aliased_vdefs. Flags that it has been invoked to the
1441 boolean variable pointed to by DATA. */
1443 static bool
1444 mark_modified (ao_ref *ao ATTRIBUTE_UNUSED, tree vdef ATTRIBUTE_UNUSED,
1445 void *data)
1447 bool *b = (bool *) data;
1448 *b = true;
1449 return true;
1452 /* If OP refers to value of function parameter, return the corresponding
1453 parameter. */
1455 static tree
1456 unmodified_parm_1 (gimple stmt, tree op)
1458 /* SSA_NAME referring to parm default def? */
1459 if (TREE_CODE (op) == SSA_NAME
1460 && SSA_NAME_IS_DEFAULT_DEF (op)
1461 && TREE_CODE (SSA_NAME_VAR (op)) == PARM_DECL)
1462 return SSA_NAME_VAR (op);
1463 /* Non-SSA parm reference? */
1464 if (TREE_CODE (op) == PARM_DECL)
1466 bool modified = false;
1468 ao_ref refd;
1469 ao_ref_init (&refd, op);
1470 walk_aliased_vdefs (&refd, gimple_vuse (stmt), mark_modified, &modified,
1471 NULL);
1472 if (!modified)
1473 return op;
1475 return NULL_TREE;
1478 /* If OP refers to value of function parameter, return the corresponding
1479 parameter. Also traverse chains of SSA register assignments. */
1481 static tree
1482 unmodified_parm (gimple stmt, tree op)
1484 tree res = unmodified_parm_1 (stmt, op);
1485 if (res)
1486 return res;
1488 if (TREE_CODE (op) == SSA_NAME
1489 && !SSA_NAME_IS_DEFAULT_DEF (op)
1490 && gimple_assign_single_p (SSA_NAME_DEF_STMT (op)))
1491 return unmodified_parm (SSA_NAME_DEF_STMT (op),
1492 gimple_assign_rhs1 (SSA_NAME_DEF_STMT (op)));
1493 return NULL_TREE;
1496 /* If OP refers to a value of a function parameter or value loaded from an
1497 aggregate passed to a parameter (either by value or reference), return TRUE
1498 and store the number of the parameter to *INDEX_P and information whether
1499 and how it has been loaded from an aggregate into *AGGPOS. INFO describes
1500 the function parameters, STMT is the statement in which OP is used or
1501 loaded. */
1503 static bool
1504 unmodified_parm_or_parm_agg_item (struct ipa_node_params *info,
1505 gimple stmt, tree op, int *index_p,
1506 struct agg_position_info *aggpos)
1508 tree res = unmodified_parm_1 (stmt, op);
1510 gcc_checking_assert (aggpos);
1511 if (res)
1513 *index_p = ipa_get_param_decl_index (info, res);
1514 if (*index_p < 0)
1515 return false;
1516 aggpos->agg_contents = false;
1517 aggpos->by_ref = false;
1518 return true;
1521 if (TREE_CODE (op) == SSA_NAME)
1523 if (SSA_NAME_IS_DEFAULT_DEF (op)
1524 || !gimple_assign_single_p (SSA_NAME_DEF_STMT (op)))
1525 return false;
1526 stmt = SSA_NAME_DEF_STMT (op);
1527 op = gimple_assign_rhs1 (stmt);
1528 if (!REFERENCE_CLASS_P (op))
1529 return unmodified_parm_or_parm_agg_item (info, stmt, op, index_p,
1530 aggpos);
1533 aggpos->agg_contents = true;
1534 return ipa_load_from_parm_agg (info, stmt, op, index_p, &aggpos->offset,
1535 &aggpos->by_ref);
1538 /* See if statement might disappear after inlining.
1539 0 - means not eliminated
1540 1 - half of statements goes away
1541 2 - for sure it is eliminated.
1542 We are not terribly sophisticated, basically looking for simple abstraction
1543 penalty wrappers. */
1545 static int
1546 eliminated_by_inlining_prob (gimple stmt)
1548 enum gimple_code code = gimple_code (stmt);
1549 enum tree_code rhs_code;
1551 if (!optimize)
1552 return 0;
1554 switch (code)
1556 case GIMPLE_RETURN:
1557 return 2;
1558 case GIMPLE_ASSIGN:
1559 if (gimple_num_ops (stmt) != 2)
1560 return 0;
1562 rhs_code = gimple_assign_rhs_code (stmt);
1564 /* Casts of parameters, loads from parameters passed by reference
1565 and stores to return value or parameters are often free after
1566 inlining dua to SRA and further combining.
1567 Assume that half of statements goes away. */
1568 if (rhs_code == CONVERT_EXPR
1569 || rhs_code == NOP_EXPR
1570 || rhs_code == VIEW_CONVERT_EXPR
1571 || rhs_code == ADDR_EXPR
1572 || gimple_assign_rhs_class (stmt) == GIMPLE_SINGLE_RHS)
1574 tree rhs = gimple_assign_rhs1 (stmt);
1575 tree lhs = gimple_assign_lhs (stmt);
1576 tree inner_rhs = get_base_address (rhs);
1577 tree inner_lhs = get_base_address (lhs);
1578 bool rhs_free = false;
1579 bool lhs_free = false;
1581 if (!inner_rhs)
1582 inner_rhs = rhs;
1583 if (!inner_lhs)
1584 inner_lhs = lhs;
1586 /* Reads of parameter are expected to be free. */
1587 if (unmodified_parm (stmt, inner_rhs))
1588 rhs_free = true;
1589 /* Match expressions of form &this->field. Those will most likely
1590 combine with something upstream after inlining. */
1591 else if (TREE_CODE (inner_rhs) == ADDR_EXPR)
1593 tree op = get_base_address (TREE_OPERAND (inner_rhs, 0));
1594 if (TREE_CODE (op) == PARM_DECL)
1595 rhs_free = true;
1596 else if (TREE_CODE (op) == MEM_REF
1597 && unmodified_parm (stmt, TREE_OPERAND (op, 0)))
1598 rhs_free = true;
1601 /* When parameter is not SSA register because its address is taken
1602 and it is just copied into one, the statement will be completely
1603 free after inlining (we will copy propagate backward). */
1604 if (rhs_free && is_gimple_reg (lhs))
1605 return 2;
1607 /* Reads of parameters passed by reference
1608 expected to be free (i.e. optimized out after inlining). */
1609 if (TREE_CODE (inner_rhs) == MEM_REF
1610 && unmodified_parm (stmt, TREE_OPERAND (inner_rhs, 0)))
1611 rhs_free = true;
1613 /* Copying parameter passed by reference into gimple register is
1614 probably also going to copy propagate, but we can't be quite
1615 sure. */
1616 if (rhs_free && is_gimple_reg (lhs))
1617 lhs_free = true;
1619 /* Writes to parameters, parameters passed by value and return value
1620 (either dirrectly or passed via invisible reference) are free.
1622 TODO: We ought to handle testcase like
1623 struct a {int a,b;};
1624 struct a
1625 retrurnsturct (void)
1627 struct a a ={1,2};
1628 return a;
1631 This translate into:
1633 retrurnsturct ()
1635 int a$b;
1636 int a$a;
1637 struct a a;
1638 struct a D.2739;
1640 <bb 2>:
1641 D.2739.a = 1;
1642 D.2739.b = 2;
1643 return D.2739;
1646 For that we either need to copy ipa-split logic detecting writes
1647 to return value. */
1648 if (TREE_CODE (inner_lhs) == PARM_DECL
1649 || TREE_CODE (inner_lhs) == RESULT_DECL
1650 || (TREE_CODE (inner_lhs) == MEM_REF
1651 && (unmodified_parm (stmt, TREE_OPERAND (inner_lhs, 0))
1652 || (TREE_CODE (TREE_OPERAND (inner_lhs, 0)) == SSA_NAME
1653 && SSA_NAME_VAR (TREE_OPERAND (inner_lhs, 0))
1654 && TREE_CODE (SSA_NAME_VAR (TREE_OPERAND
1655 (inner_lhs,
1656 0))) == RESULT_DECL))))
1657 lhs_free = true;
1658 if (lhs_free
1659 && (is_gimple_reg (rhs) || is_gimple_min_invariant (rhs)))
1660 rhs_free = true;
1661 if (lhs_free && rhs_free)
1662 return 1;
1664 return 0;
1665 default:
1666 return 0;
1671 /* If BB ends by a conditional we can turn into predicates, attach corresponding
1672 predicates to the CFG edges. */
1674 static void
1675 set_cond_stmt_execution_predicate (struct ipa_node_params *info,
1676 struct inline_summary *summary,
1677 basic_block bb)
1679 gimple last;
1680 tree op;
1681 int index;
1682 struct agg_position_info aggpos;
1683 enum tree_code code, inverted_code;
1684 edge e;
1685 edge_iterator ei;
1686 gimple set_stmt;
1687 tree op2;
1689 last = last_stmt (bb);
1690 if (!last || gimple_code (last) != GIMPLE_COND)
1691 return;
1692 if (!is_gimple_ip_invariant (gimple_cond_rhs (last)))
1693 return;
1694 op = gimple_cond_lhs (last);
1695 /* TODO: handle conditionals like
1696 var = op0 < 4;
1697 if (var != 0). */
1698 if (unmodified_parm_or_parm_agg_item (info, last, op, &index, &aggpos))
1700 code = gimple_cond_code (last);
1701 inverted_code
1702 = invert_tree_comparison (code,
1703 HONOR_NANS (TYPE_MODE (TREE_TYPE (op))));
1705 FOR_EACH_EDGE (e, ei, bb->succs)
1707 struct predicate p = add_condition (summary, index, &aggpos,
1708 e->flags & EDGE_TRUE_VALUE
1709 ? code : inverted_code,
1710 gimple_cond_rhs (last));
1711 e->aux = pool_alloc (edge_predicate_pool);
1712 *(struct predicate *) e->aux = p;
1716 if (TREE_CODE (op) != SSA_NAME)
1717 return;
1718 /* Special case
1719 if (builtin_constant_p (op))
1720 constant_code
1721 else
1722 nonconstant_code.
1723 Here we can predicate nonconstant_code. We can't
1724 really handle constant_code since we have no predicate
1725 for this and also the constant code is not known to be
1726 optimized away when inliner doen't see operand is constant.
1727 Other optimizers might think otherwise. */
1728 if (gimple_cond_code (last) != NE_EXPR
1729 || !integer_zerop (gimple_cond_rhs (last)))
1730 return;
1731 set_stmt = SSA_NAME_DEF_STMT (op);
1732 if (!gimple_call_builtin_p (set_stmt, BUILT_IN_CONSTANT_P)
1733 || gimple_call_num_args (set_stmt) != 1)
1734 return;
1735 op2 = gimple_call_arg (set_stmt, 0);
1736 if (!unmodified_parm_or_parm_agg_item
1737 (info, set_stmt, op2, &index, &aggpos))
1738 return;
1739 FOR_EACH_EDGE (e, ei, bb->succs) if (e->flags & EDGE_FALSE_VALUE)
1741 struct predicate p = add_condition (summary, index, &aggpos,
1742 IS_NOT_CONSTANT, NULL_TREE);
1743 e->aux = pool_alloc (edge_predicate_pool);
1744 *(struct predicate *) e->aux = p;
1749 /* If BB ends by a switch we can turn into predicates, attach corresponding
1750 predicates to the CFG edges. */
1752 static void
1753 set_switch_stmt_execution_predicate (struct ipa_node_params *info,
1754 struct inline_summary *summary,
1755 basic_block bb)
1757 gimple last;
1758 tree op;
1759 int index;
1760 struct agg_position_info aggpos;
1761 edge e;
1762 edge_iterator ei;
1763 size_t n;
1764 size_t case_idx;
1766 last = last_stmt (bb);
1767 if (!last || gimple_code (last) != GIMPLE_SWITCH)
1768 return;
1769 op = gimple_switch_index (last);
1770 if (!unmodified_parm_or_parm_agg_item (info, last, op, &index, &aggpos))
1771 return;
1773 FOR_EACH_EDGE (e, ei, bb->succs)
1775 e->aux = pool_alloc (edge_predicate_pool);
1776 *(struct predicate *) e->aux = false_predicate ();
1778 n = gimple_switch_num_labels (last);
1779 for (case_idx = 0; case_idx < n; ++case_idx)
1781 tree cl = gimple_switch_label (last, case_idx);
1782 tree min, max;
1783 struct predicate p;
1785 e = find_edge (bb, label_to_block (CASE_LABEL (cl)));
1786 min = CASE_LOW (cl);
1787 max = CASE_HIGH (cl);
1789 /* For default we might want to construct predicate that none
1790 of cases is met, but it is bit hard to do not having negations
1791 of conditionals handy. */
1792 if (!min && !max)
1793 p = true_predicate ();
1794 else if (!max)
1795 p = add_condition (summary, index, &aggpos, EQ_EXPR, min);
1796 else
1798 struct predicate p1, p2;
1799 p1 = add_condition (summary, index, &aggpos, GE_EXPR, min);
1800 p2 = add_condition (summary, index, &aggpos, LE_EXPR, max);
1801 p = and_predicates (summary->conds, &p1, &p2);
1803 *(struct predicate *) e->aux
1804 = or_predicates (summary->conds, &p, (struct predicate *) e->aux);
1809 /* For each BB in NODE attach to its AUX pointer predicate under
1810 which it is executable. */
1812 static void
1813 compute_bb_predicates (struct cgraph_node *node,
1814 struct ipa_node_params *parms_info,
1815 struct inline_summary *summary)
1817 struct function *my_function = DECL_STRUCT_FUNCTION (node->symbol.decl);
1818 bool done = false;
1819 basic_block bb;
1821 FOR_EACH_BB_FN (bb, my_function)
1823 set_cond_stmt_execution_predicate (parms_info, summary, bb);
1824 set_switch_stmt_execution_predicate (parms_info, summary, bb);
1827 /* Entry block is always executable. */
1828 ENTRY_BLOCK_PTR_FOR_FUNCTION (my_function)->aux
1829 = pool_alloc (edge_predicate_pool);
1830 *(struct predicate *) ENTRY_BLOCK_PTR_FOR_FUNCTION (my_function)->aux
1831 = true_predicate ();
1833 /* A simple dataflow propagation of predicates forward in the CFG.
1834 TODO: work in reverse postorder. */
1835 while (!done)
1837 done = true;
1838 FOR_EACH_BB_FN (bb, my_function)
1840 struct predicate p = false_predicate ();
1841 edge e;
1842 edge_iterator ei;
1843 FOR_EACH_EDGE (e, ei, bb->preds)
1845 if (e->src->aux)
1847 struct predicate this_bb_predicate
1848 = *(struct predicate *) e->src->aux;
1849 if (e->aux)
1850 this_bb_predicate
1851 = and_predicates (summary->conds, &this_bb_predicate,
1852 (struct predicate *) e->aux);
1853 p = or_predicates (summary->conds, &p, &this_bb_predicate);
1854 if (true_predicate_p (&p))
1855 break;
1858 if (false_predicate_p (&p))
1859 gcc_assert (!bb->aux);
1860 else
1862 if (!bb->aux)
1864 done = false;
1865 bb->aux = pool_alloc (edge_predicate_pool);
1866 *((struct predicate *) bb->aux) = p;
1868 else if (!predicates_equal_p (&p, (struct predicate *) bb->aux))
1870 done = false;
1871 *((struct predicate *) bb->aux) = p;
1879 /* We keep info about constantness of SSA names. */
1881 typedef struct predicate predicate_t;
1882 /* Return predicate specifying when the STMT might have result that is not
1883 a compile time constant. */
1885 static struct predicate
1886 will_be_nonconstant_expr_predicate (struct ipa_node_params *info,
1887 struct inline_summary *summary,
1888 tree expr,
1889 vec<predicate_t> nonconstant_names)
1891 tree parm;
1892 int index;
1894 while (UNARY_CLASS_P (expr))
1895 expr = TREE_OPERAND (expr, 0);
1897 parm = unmodified_parm (NULL, expr);
1898 if (parm && (index = ipa_get_param_decl_index (info, parm)) >= 0)
1899 return add_condition (summary, index, NULL, CHANGED, NULL_TREE);
1900 if (is_gimple_min_invariant (expr))
1901 return false_predicate ();
1902 if (TREE_CODE (expr) == SSA_NAME)
1903 return nonconstant_names[SSA_NAME_VERSION (expr)];
1904 if (BINARY_CLASS_P (expr) || COMPARISON_CLASS_P (expr))
1906 struct predicate p1 = will_be_nonconstant_expr_predicate
1907 (info, summary, TREE_OPERAND (expr, 0),
1908 nonconstant_names);
1909 struct predicate p2;
1910 if (true_predicate_p (&p1))
1911 return p1;
1912 p2 = will_be_nonconstant_expr_predicate (info, summary,
1913 TREE_OPERAND (expr, 1),
1914 nonconstant_names);
1915 return or_predicates (summary->conds, &p1, &p2);
1917 else if (TREE_CODE (expr) == COND_EXPR)
1919 struct predicate p1 = will_be_nonconstant_expr_predicate
1920 (info, summary, TREE_OPERAND (expr, 0),
1921 nonconstant_names);
1922 struct predicate p2;
1923 if (true_predicate_p (&p1))
1924 return p1;
1925 p2 = will_be_nonconstant_expr_predicate (info, summary,
1926 TREE_OPERAND (expr, 1),
1927 nonconstant_names);
1928 if (true_predicate_p (&p2))
1929 return p2;
1930 p1 = or_predicates (summary->conds, &p1, &p2);
1931 p2 = will_be_nonconstant_expr_predicate (info, summary,
1932 TREE_OPERAND (expr, 2),
1933 nonconstant_names);
1934 return or_predicates (summary->conds, &p1, &p2);
1936 else
1938 debug_tree (expr);
1939 gcc_unreachable ();
1941 return false_predicate ();
1945 /* Return predicate specifying when the STMT might have result that is not
1946 a compile time constant. */
1948 static struct predicate
1949 will_be_nonconstant_predicate (struct ipa_node_params *info,
1950 struct inline_summary *summary,
1951 gimple stmt,
1952 vec<predicate_t> nonconstant_names)
1954 struct predicate p = true_predicate ();
1955 ssa_op_iter iter;
1956 tree use;
1957 struct predicate op_non_const;
1958 bool is_load;
1959 int base_index;
1960 struct agg_position_info aggpos;
1962 /* What statments might be optimized away
1963 when their arguments are constant
1964 TODO: also trivial builtins.
1965 builtin_constant_p is already handled later. */
1966 if (gimple_code (stmt) != GIMPLE_ASSIGN
1967 && gimple_code (stmt) != GIMPLE_COND
1968 && gimple_code (stmt) != GIMPLE_SWITCH)
1969 return p;
1971 /* Stores will stay anyway. */
1972 if (gimple_store_p (stmt))
1973 return p;
1975 is_load = gimple_assign_load_p (stmt);
1977 /* Loads can be optimized when the value is known. */
1978 if (is_load)
1980 tree op;
1981 gcc_assert (gimple_assign_single_p (stmt));
1982 op = gimple_assign_rhs1 (stmt);
1983 if (!unmodified_parm_or_parm_agg_item (info, stmt, op, &base_index,
1984 &aggpos))
1985 return p;
1987 else
1988 base_index = -1;
1990 /* See if we understand all operands before we start
1991 adding conditionals. */
1992 FOR_EACH_SSA_TREE_OPERAND (use, stmt, iter, SSA_OP_USE)
1994 tree parm = unmodified_parm (stmt, use);
1995 /* For arguments we can build a condition. */
1996 if (parm && ipa_get_param_decl_index (info, parm) >= 0)
1997 continue;
1998 if (TREE_CODE (use) != SSA_NAME)
1999 return p;
2000 /* If we know when operand is constant,
2001 we still can say something useful. */
2002 if (!true_predicate_p (&nonconstant_names[SSA_NAME_VERSION (use)]))
2003 continue;
2004 return p;
2007 if (is_load)
2008 op_non_const =
2009 add_condition (summary, base_index, &aggpos, CHANGED, NULL);
2010 else
2011 op_non_const = false_predicate ();
2012 FOR_EACH_SSA_TREE_OPERAND (use, stmt, iter, SSA_OP_USE)
2014 tree parm = unmodified_parm (stmt, use);
2015 int index;
2017 if (parm && (index = ipa_get_param_decl_index (info, parm)) >= 0)
2019 if (index != base_index)
2020 p = add_condition (summary, index, NULL, CHANGED, NULL_TREE);
2021 else
2022 continue;
2024 else
2025 p = nonconstant_names[SSA_NAME_VERSION (use)];
2026 op_non_const = or_predicates (summary->conds, &p, &op_non_const);
2028 if (gimple_code (stmt) == GIMPLE_ASSIGN
2029 && TREE_CODE (gimple_assign_lhs (stmt)) == SSA_NAME)
2030 nonconstant_names[SSA_NAME_VERSION (gimple_assign_lhs (stmt))]
2031 = op_non_const;
2032 return op_non_const;
2035 struct record_modified_bb_info
2037 bitmap bb_set;
2038 gimple stmt;
2041 /* Callback of walk_aliased_vdefs. Records basic blocks where the value may be
2042 set except for info->stmt. */
2044 static bool
2045 record_modified (ao_ref *ao ATTRIBUTE_UNUSED, tree vdef, void *data)
2047 struct record_modified_bb_info *info =
2048 (struct record_modified_bb_info *) data;
2049 if (SSA_NAME_DEF_STMT (vdef) == info->stmt)
2050 return false;
2051 bitmap_set_bit (info->bb_set,
2052 SSA_NAME_IS_DEFAULT_DEF (vdef)
2053 ? ENTRY_BLOCK_PTR->index
2054 : gimple_bb (SSA_NAME_DEF_STMT (vdef))->index);
2055 return false;
2058 /* Return probability (based on REG_BR_PROB_BASE) that I-th parameter of STMT
2059 will change since last invocation of STMT.
2061 Value 0 is reserved for compile time invariants.
2062 For common parameters it is REG_BR_PROB_BASE. For loop invariants it
2063 ought to be REG_BR_PROB_BASE / estimated_iters. */
2065 static int
2066 param_change_prob (gimple stmt, int i)
2068 tree op = gimple_call_arg (stmt, i);
2069 basic_block bb = gimple_bb (stmt);
2070 tree base;
2072 /* Global invariants neve change. */
2073 if (is_gimple_min_invariant (op))
2074 return 0;
2075 /* We would have to do non-trivial analysis to really work out what
2076 is the probability of value to change (i.e. when init statement
2077 is in a sibling loop of the call).
2079 We do an conservative estimate: when call is executed N times more often
2080 than the statement defining value, we take the frequency 1/N. */
2081 if (TREE_CODE (op) == SSA_NAME)
2083 int init_freq;
2085 if (!bb->frequency)
2086 return REG_BR_PROB_BASE;
2088 if (SSA_NAME_IS_DEFAULT_DEF (op))
2089 init_freq = ENTRY_BLOCK_PTR->frequency;
2090 else
2091 init_freq = gimple_bb (SSA_NAME_DEF_STMT (op))->frequency;
2093 if (!init_freq)
2094 init_freq = 1;
2095 if (init_freq < bb->frequency)
2096 return MAX (GCOV_COMPUTE_SCALE (init_freq, bb->frequency), 1);
2097 else
2098 return REG_BR_PROB_BASE;
2101 base = get_base_address (op);
2102 if (base)
2104 ao_ref refd;
2105 int max;
2106 struct record_modified_bb_info info;
2107 bitmap_iterator bi;
2108 unsigned index;
2110 if (const_value_known_p (base))
2111 return 0;
2112 if (!bb->frequency)
2113 return REG_BR_PROB_BASE;
2114 ao_ref_init (&refd, op);
2115 info.stmt = stmt;
2116 info.bb_set = BITMAP_ALLOC (NULL);
2117 walk_aliased_vdefs (&refd, gimple_vuse (stmt), record_modified, &info,
2118 NULL);
2119 if (bitmap_bit_p (info.bb_set, bb->index))
2121 BITMAP_FREE (info.bb_set);
2122 return REG_BR_PROB_BASE;
2125 /* Assume that every memory is initialized at entry.
2126 TODO: Can we easilly determine if value is always defined
2127 and thus we may skip entry block? */
2128 if (ENTRY_BLOCK_PTR->frequency)
2129 max = ENTRY_BLOCK_PTR->frequency;
2130 else
2131 max = 1;
2133 EXECUTE_IF_SET_IN_BITMAP (info.bb_set, 0, index, bi)
2134 max = MIN (max, BASIC_BLOCK (index)->frequency);
2136 BITMAP_FREE (info.bb_set);
2137 if (max < bb->frequency)
2138 return MAX (GCOV_COMPUTE_SCALE (max, bb->frequency), 1);
2139 else
2140 return REG_BR_PROB_BASE;
2142 return REG_BR_PROB_BASE;
2145 /* Find whether a basic block BB is the final block of a (half) diamond CFG
2146 sub-graph and if the predicate the condition depends on is known. If so,
2147 return true and store the pointer the predicate in *P. */
2149 static bool
2150 phi_result_unknown_predicate (struct ipa_node_params *info,
2151 struct inline_summary *summary, basic_block bb,
2152 struct predicate *p,
2153 vec<predicate_t> nonconstant_names)
2155 edge e;
2156 edge_iterator ei;
2157 basic_block first_bb = NULL;
2158 gimple stmt;
2160 if (single_pred_p (bb))
2162 *p = false_predicate ();
2163 return true;
2166 FOR_EACH_EDGE (e, ei, bb->preds)
2168 if (single_succ_p (e->src))
2170 if (!single_pred_p (e->src))
2171 return false;
2172 if (!first_bb)
2173 first_bb = single_pred (e->src);
2174 else if (single_pred (e->src) != first_bb)
2175 return false;
2177 else
2179 if (!first_bb)
2180 first_bb = e->src;
2181 else if (e->src != first_bb)
2182 return false;
2186 if (!first_bb)
2187 return false;
2189 stmt = last_stmt (first_bb);
2190 if (!stmt
2191 || gimple_code (stmt) != GIMPLE_COND
2192 || !is_gimple_ip_invariant (gimple_cond_rhs (stmt)))
2193 return false;
2195 *p = will_be_nonconstant_expr_predicate (info, summary,
2196 gimple_cond_lhs (stmt),
2197 nonconstant_names);
2198 if (true_predicate_p (p))
2199 return false;
2200 else
2201 return true;
2204 /* Given a PHI statement in a function described by inline properties SUMMARY
2205 and *P being the predicate describing whether the selected PHI argument is
2206 known, store a predicate for the result of the PHI statement into
2207 NONCONSTANT_NAMES, if possible. */
2209 static void
2210 predicate_for_phi_result (struct inline_summary *summary, gimple phi,
2211 struct predicate *p,
2212 vec<predicate_t> nonconstant_names)
2214 unsigned i;
2216 for (i = 0; i < gimple_phi_num_args (phi); i++)
2218 tree arg = gimple_phi_arg (phi, i)->def;
2219 if (!is_gimple_min_invariant (arg))
2221 gcc_assert (TREE_CODE (arg) == SSA_NAME);
2222 *p = or_predicates (summary->conds, p,
2223 &nonconstant_names[SSA_NAME_VERSION (arg)]);
2224 if (true_predicate_p (p))
2225 return;
2229 if (dump_file && (dump_flags & TDF_DETAILS))
2231 fprintf (dump_file, "\t\tphi predicate: ");
2232 dump_predicate (dump_file, summary->conds, p);
2234 nonconstant_names[SSA_NAME_VERSION (gimple_phi_result (phi))] = *p;
2237 /* Return predicate specifying when array index in access OP becomes non-constant. */
2239 static struct predicate
2240 array_index_predicate (struct inline_summary *info,
2241 vec< predicate_t> nonconstant_names, tree op)
2243 struct predicate p = false_predicate ();
2244 while (handled_component_p (op))
2246 if (TREE_CODE (op) == ARRAY_REF || TREE_CODE (op) == ARRAY_RANGE_REF)
2248 if (TREE_CODE (TREE_OPERAND (op, 1)) == SSA_NAME)
2249 p = or_predicates (info->conds, &p,
2250 &nonconstant_names[SSA_NAME_VERSION
2251 (TREE_OPERAND (op, 1))]);
2253 op = TREE_OPERAND (op, 0);
2255 return p;
2258 /* Compute function body size parameters for NODE.
2259 When EARLY is true, we compute only simple summaries without
2260 non-trivial predicates to drive the early inliner. */
2262 static void
2263 estimate_function_body_sizes (struct cgraph_node *node, bool early)
2265 gcov_type time = 0;
2266 /* Estimate static overhead for function prologue/epilogue and alignment. */
2267 int size = 2;
2268 /* Benefits are scaled by probability of elimination that is in range
2269 <0,2>. */
2270 basic_block bb;
2271 gimple_stmt_iterator bsi;
2272 struct function *my_function = DECL_STRUCT_FUNCTION (node->symbol.decl);
2273 int freq;
2274 struct inline_summary *info = inline_summary (node);
2275 struct predicate bb_predicate;
2276 struct ipa_node_params *parms_info = NULL;
2277 vec<predicate_t> nonconstant_names = vNULL;
2278 int nblocks, n;
2279 int *order;
2280 predicate array_index = true_predicate ();
2282 info->conds = NULL;
2283 info->entry = NULL;
2285 if (optimize && !early)
2287 calculate_dominance_info (CDI_DOMINATORS);
2288 loop_optimizer_init (LOOPS_NORMAL | LOOPS_HAVE_RECORDED_EXITS);
2290 if (ipa_node_params_vector.exists ())
2292 parms_info = IPA_NODE_REF (node);
2293 nonconstant_names.safe_grow_cleared
2294 (SSANAMES (my_function)->length ());
2298 if (dump_file)
2299 fprintf (dump_file, "\nAnalyzing function body size: %s\n",
2300 cgraph_node_name (node));
2302 /* When we run into maximal number of entries, we assign everything to the
2303 constant truth case. Be sure to have it in list. */
2304 bb_predicate = true_predicate ();
2305 account_size_time (info, 0, 0, &bb_predicate);
2307 bb_predicate = not_inlined_predicate ();
2308 account_size_time (info, 2 * INLINE_SIZE_SCALE, 0, &bb_predicate);
2310 gcc_assert (my_function && my_function->cfg);
2311 if (parms_info)
2312 compute_bb_predicates (node, parms_info, info);
2313 gcc_assert (cfun == my_function);
2314 order = XNEWVEC (int, n_basic_blocks);
2315 nblocks = pre_and_rev_post_order_compute (NULL, order, false);
2316 for (n = 0; n < nblocks; n++)
2318 bb = BASIC_BLOCK (order[n]);
2319 freq = compute_call_stmt_bb_frequency (node->symbol.decl, bb);
2321 /* TODO: Obviously predicates can be propagated down across CFG. */
2322 if (parms_info)
2324 if (bb->aux)
2325 bb_predicate = *(struct predicate *) bb->aux;
2326 else
2327 bb_predicate = false_predicate ();
2329 else
2330 bb_predicate = true_predicate ();
2332 if (dump_file && (dump_flags & TDF_DETAILS))
2334 fprintf (dump_file, "\n BB %i predicate:", bb->index);
2335 dump_predicate (dump_file, info->conds, &bb_predicate);
2338 if (parms_info && nonconstant_names.exists ())
2340 struct predicate phi_predicate;
2341 bool first_phi = true;
2343 for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
2345 if (first_phi
2346 && !phi_result_unknown_predicate (parms_info, info, bb,
2347 &phi_predicate,
2348 nonconstant_names))
2349 break;
2350 first_phi = false;
2351 if (dump_file && (dump_flags & TDF_DETAILS))
2353 fprintf (dump_file, " ");
2354 print_gimple_stmt (dump_file, gsi_stmt (bsi), 0, 0);
2356 predicate_for_phi_result (info, gsi_stmt (bsi), &phi_predicate,
2357 nonconstant_names);
2361 for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
2363 gimple stmt = gsi_stmt (bsi);
2364 int this_size = estimate_num_insns (stmt, &eni_size_weights);
2365 int this_time = estimate_num_insns (stmt, &eni_time_weights);
2366 int prob;
2367 struct predicate will_be_nonconstant;
2369 if (dump_file && (dump_flags & TDF_DETAILS))
2371 fprintf (dump_file, " ");
2372 print_gimple_stmt (dump_file, stmt, 0, 0);
2373 fprintf (dump_file, "\t\tfreq:%3.2f size:%3i time:%3i\n",
2374 ((double) freq) / CGRAPH_FREQ_BASE, this_size,
2375 this_time);
2378 if (gimple_assign_load_p (stmt) && nonconstant_names.exists ())
2380 struct predicate this_array_index;
2381 this_array_index =
2382 array_index_predicate (info, nonconstant_names,
2383 gimple_assign_rhs1 (stmt));
2384 if (!false_predicate_p (&this_array_index))
2385 array_index =
2386 and_predicates (info->conds, &array_index,
2387 &this_array_index);
2389 if (gimple_store_p (stmt) && nonconstant_names.exists ())
2391 struct predicate this_array_index;
2392 this_array_index =
2393 array_index_predicate (info, nonconstant_names,
2394 gimple_get_lhs (stmt));
2395 if (!false_predicate_p (&this_array_index))
2396 array_index =
2397 and_predicates (info->conds, &array_index,
2398 &this_array_index);
2402 if (is_gimple_call (stmt))
2404 struct cgraph_edge *edge = cgraph_edge (node, stmt);
2405 struct inline_edge_summary *es = inline_edge_summary (edge);
2407 /* Special case: results of BUILT_IN_CONSTANT_P will be always
2408 resolved as constant. We however don't want to optimize
2409 out the cgraph edges. */
2410 if (nonconstant_names.exists ()
2411 && gimple_call_builtin_p (stmt, BUILT_IN_CONSTANT_P)
2412 && gimple_call_lhs (stmt)
2413 && TREE_CODE (gimple_call_lhs (stmt)) == SSA_NAME)
2415 struct predicate false_p = false_predicate ();
2416 nonconstant_names[SSA_NAME_VERSION (gimple_call_lhs (stmt))]
2417 = false_p;
2419 if (ipa_node_params_vector.exists ())
2421 int count = gimple_call_num_args (stmt);
2422 int i;
2424 if (count)
2425 es->param.safe_grow_cleared (count);
2426 for (i = 0; i < count; i++)
2428 int prob = param_change_prob (stmt, i);
2429 gcc_assert (prob >= 0 && prob <= REG_BR_PROB_BASE);
2430 es->param[i].change_prob = prob;
2434 es->call_stmt_size = this_size;
2435 es->call_stmt_time = this_time;
2436 es->loop_depth = bb_loop_depth (bb);
2437 edge_set_predicate (edge, &bb_predicate);
2440 /* TODO: When conditional jump or swithc is known to be constant, but
2441 we did not translate it into the predicates, we really can account
2442 just maximum of the possible paths. */
2443 if (parms_info)
2444 will_be_nonconstant
2445 = will_be_nonconstant_predicate (parms_info, info,
2446 stmt, nonconstant_names);
2447 if (this_time || this_size)
2449 struct predicate p;
2451 this_time *= freq;
2453 prob = eliminated_by_inlining_prob (stmt);
2454 if (prob == 1 && dump_file && (dump_flags & TDF_DETAILS))
2455 fprintf (dump_file,
2456 "\t\t50%% will be eliminated by inlining\n");
2457 if (prob == 2 && dump_file && (dump_flags & TDF_DETAILS))
2458 fprintf (dump_file, "\t\tWill be eliminated by inlining\n");
2460 if (parms_info)
2461 p = and_predicates (info->conds, &bb_predicate,
2462 &will_be_nonconstant);
2463 else
2464 p = true_predicate ();
2466 if (!false_predicate_p (&p))
2468 time += this_time;
2469 size += this_size;
2470 if (time > MAX_TIME * INLINE_TIME_SCALE)
2471 time = MAX_TIME * INLINE_TIME_SCALE;
2474 /* We account everything but the calls. Calls have their own
2475 size/time info attached to cgraph edges. This is necessary
2476 in order to make the cost disappear after inlining. */
2477 if (!is_gimple_call (stmt))
2479 if (prob)
2481 struct predicate ip = not_inlined_predicate ();
2482 ip = and_predicates (info->conds, &ip, &p);
2483 account_size_time (info, this_size * prob,
2484 this_time * prob, &ip);
2486 if (prob != 2)
2487 account_size_time (info, this_size * (2 - prob),
2488 this_time * (2 - prob), &p);
2491 gcc_assert (time >= 0);
2492 gcc_assert (size >= 0);
2496 set_hint_predicate (&inline_summary (node)->array_index, array_index);
2497 time = (time + CGRAPH_FREQ_BASE / 2) / CGRAPH_FREQ_BASE;
2498 if (time > MAX_TIME)
2499 time = MAX_TIME;
2500 free (order);
2502 if (!early && nonconstant_names.exists ())
2504 struct loop *loop;
2505 loop_iterator li;
2506 predicate loop_iterations = true_predicate ();
2507 predicate loop_stride = true_predicate ();
2509 if (dump_file && (dump_flags & TDF_DETAILS))
2510 flow_loops_dump (dump_file, NULL, 0);
2511 scev_initialize ();
2512 FOR_EACH_LOOP (li, loop, 0)
2514 vec<edge> exits;
2515 edge ex;
2516 unsigned int j, i;
2517 struct tree_niter_desc niter_desc;
2518 basic_block *body = get_loop_body (loop);
2519 bb_predicate = *(struct predicate *) loop->header->aux;
2521 exits = get_loop_exit_edges (loop);
2522 FOR_EACH_VEC_ELT (exits, j, ex)
2523 if (number_of_iterations_exit (loop, ex, &niter_desc, false)
2524 && !is_gimple_min_invariant (niter_desc.niter))
2526 predicate will_be_nonconstant
2527 = will_be_nonconstant_expr_predicate (parms_info, info,
2528 niter_desc.niter,
2529 nonconstant_names);
2530 if (!true_predicate_p (&will_be_nonconstant))
2531 will_be_nonconstant = and_predicates (info->conds,
2532 &bb_predicate,
2533 &will_be_nonconstant);
2534 if (!true_predicate_p (&will_be_nonconstant)
2535 && !false_predicate_p (&will_be_nonconstant))
2536 /* This is slightly inprecise. We may want to represent each
2537 loop with independent predicate. */
2538 loop_iterations =
2539 and_predicates (info->conds, &loop_iterations,
2540 &will_be_nonconstant);
2542 exits.release ();
2544 for (i = 0; i < loop->num_nodes; i++)
2546 gimple_stmt_iterator gsi;
2547 bb_predicate = *(struct predicate *) body[i]->aux;
2548 for (gsi = gsi_start_bb (body[i]); !gsi_end_p (gsi);
2549 gsi_next (&gsi))
2551 gimple stmt = gsi_stmt (gsi);
2552 affine_iv iv;
2553 ssa_op_iter iter;
2554 tree use;
2556 FOR_EACH_SSA_TREE_OPERAND (use, stmt, iter, SSA_OP_USE)
2558 predicate will_be_nonconstant;
2560 if (!simple_iv
2561 (loop, loop_containing_stmt (stmt), use, &iv, true)
2562 || is_gimple_min_invariant (iv.step))
2563 continue;
2564 will_be_nonconstant
2565 = will_be_nonconstant_expr_predicate (parms_info, info,
2566 iv.step,
2567 nonconstant_names);
2568 if (!true_predicate_p (&will_be_nonconstant))
2569 will_be_nonconstant
2570 = and_predicates (info->conds,
2571 &bb_predicate,
2572 &will_be_nonconstant);
2573 if (!true_predicate_p (&will_be_nonconstant)
2574 && !false_predicate_p (&will_be_nonconstant))
2575 /* This is slightly inprecise. We may want to represent
2576 each loop with independent predicate. */
2577 loop_stride =
2578 and_predicates (info->conds, &loop_stride,
2579 &will_be_nonconstant);
2583 free (body);
2585 set_hint_predicate (&inline_summary (node)->loop_iterations,
2586 loop_iterations);
2587 set_hint_predicate (&inline_summary (node)->loop_stride, loop_stride);
2588 scev_finalize ();
2590 FOR_ALL_BB_FN (bb, my_function)
2592 edge e;
2593 edge_iterator ei;
2595 if (bb->aux)
2596 pool_free (edge_predicate_pool, bb->aux);
2597 bb->aux = NULL;
2598 FOR_EACH_EDGE (e, ei, bb->succs)
2600 if (e->aux)
2601 pool_free (edge_predicate_pool, e->aux);
2602 e->aux = NULL;
2605 inline_summary (node)->self_time = time;
2606 inline_summary (node)->self_size = size;
2607 nonconstant_names.release ();
2608 if (optimize && !early)
2610 loop_optimizer_finalize ();
2611 free_dominance_info (CDI_DOMINATORS);
2613 if (dump_file)
2615 fprintf (dump_file, "\n");
2616 dump_inline_summary (dump_file, node);
2621 /* Compute parameters of functions used by inliner.
2622 EARLY is true when we compute parameters for the early inliner */
2624 void
2625 compute_inline_parameters (struct cgraph_node *node, bool early)
2627 HOST_WIDE_INT self_stack_size;
2628 struct cgraph_edge *e;
2629 struct inline_summary *info;
2631 gcc_assert (!node->global.inlined_to);
2633 inline_summary_alloc ();
2635 info = inline_summary (node);
2636 reset_inline_summary (node);
2638 /* FIXME: Thunks are inlinable, but tree-inline don't know how to do that.
2639 Once this happen, we will need to more curefully predict call
2640 statement size. */
2641 if (node->thunk.thunk_p)
2643 struct inline_edge_summary *es = inline_edge_summary (node->callees);
2644 struct predicate t = true_predicate ();
2646 info->inlinable = 0;
2647 node->callees->call_stmt_cannot_inline_p = true;
2648 node->local.can_change_signature = false;
2649 es->call_stmt_time = 1;
2650 es->call_stmt_size = 1;
2651 account_size_time (info, 0, 0, &t);
2652 return;
2655 /* Even is_gimple_min_invariant rely on current_function_decl. */
2656 push_cfun (DECL_STRUCT_FUNCTION (node->symbol.decl));
2658 /* Estimate the stack size for the function if we're optimizing. */
2659 self_stack_size = optimize ? estimated_stack_frame_size (node) : 0;
2660 info->estimated_self_stack_size = self_stack_size;
2661 info->estimated_stack_size = self_stack_size;
2662 info->stack_frame_offset = 0;
2664 /* Can this function be inlined at all? */
2665 info->inlinable = tree_inlinable_function_p (node->symbol.decl);
2667 /* Type attributes can use parameter indices to describe them. */
2668 if (TYPE_ATTRIBUTES (TREE_TYPE (node->symbol.decl)))
2669 node->local.can_change_signature = false;
2670 else
2672 /* Otherwise, inlinable functions always can change signature. */
2673 if (info->inlinable)
2674 node->local.can_change_signature = true;
2675 else
2677 /* Functions calling builtin_apply can not change signature. */
2678 for (e = node->callees; e; e = e->next_callee)
2680 tree cdecl = e->callee->symbol.decl;
2681 if (DECL_BUILT_IN (cdecl)
2682 && DECL_BUILT_IN_CLASS (cdecl) == BUILT_IN_NORMAL
2683 && (DECL_FUNCTION_CODE (cdecl) == BUILT_IN_APPLY_ARGS
2684 || DECL_FUNCTION_CODE (cdecl) == BUILT_IN_VA_START))
2685 break;
2687 node->local.can_change_signature = !e;
2690 estimate_function_body_sizes (node, early);
2692 /* Inlining characteristics are maintained by the cgraph_mark_inline. */
2693 info->time = info->self_time;
2694 info->size = info->self_size;
2695 info->stack_frame_offset = 0;
2696 info->estimated_stack_size = info->estimated_self_stack_size;
2697 #ifdef ENABLE_CHECKING
2698 inline_update_overall_summary (node);
2699 gcc_assert (info->time == info->self_time && info->size == info->self_size);
2700 #endif
2702 pop_cfun ();
2706 /* Compute parameters of functions used by inliner using
2707 current_function_decl. */
2709 static unsigned int
2710 compute_inline_parameters_for_current (void)
2712 compute_inline_parameters (cgraph_get_node (current_function_decl), true);
2713 return 0;
2716 struct gimple_opt_pass pass_inline_parameters =
2719 GIMPLE_PASS,
2720 "inline_param", /* name */
2721 OPTGROUP_INLINE, /* optinfo_flags */
2722 NULL, /* gate */
2723 compute_inline_parameters_for_current, /* execute */
2724 NULL, /* sub */
2725 NULL, /* next */
2726 0, /* static_pass_number */
2727 TV_INLINE_PARAMETERS, /* tv_id */
2728 0, /* properties_required */
2729 0, /* properties_provided */
2730 0, /* properties_destroyed */
2731 0, /* todo_flags_start */
2732 0 /* todo_flags_finish */
2737 /* Estimate benefit devirtualizing indirect edge IE, provided KNOWN_VALS and
2738 KNOWN_BINFOS. */
2740 static bool
2741 estimate_edge_devirt_benefit (struct cgraph_edge *ie,
2742 int *size, int *time,
2743 vec<tree> known_vals,
2744 vec<tree> known_binfos,
2745 vec<ipa_agg_jump_function_p> known_aggs)
2747 tree target;
2748 struct cgraph_node *callee;
2749 struct inline_summary *isummary;
2751 if (!known_vals.exists () && !known_binfos.exists ())
2752 return false;
2753 if (!flag_indirect_inlining)
2754 return false;
2756 target = ipa_get_indirect_edge_target (ie, known_vals, known_binfos,
2757 known_aggs);
2758 if (!target)
2759 return false;
2761 /* Account for difference in cost between indirect and direct calls. */
2762 *size -= (eni_size_weights.indirect_call_cost - eni_size_weights.call_cost);
2763 *time -= (eni_time_weights.indirect_call_cost - eni_time_weights.call_cost);
2764 gcc_checking_assert (*time >= 0);
2765 gcc_checking_assert (*size >= 0);
2767 callee = cgraph_get_node (target);
2768 if (!callee || !callee->analyzed)
2769 return false;
2770 isummary = inline_summary (callee);
2771 return isummary->inlinable;
2774 /* Increase SIZE and TIME for size and time needed to handle edge E. */
2776 static inline void
2777 estimate_edge_size_and_time (struct cgraph_edge *e, int *size, int *time,
2778 int prob,
2779 vec<tree> known_vals,
2780 vec<tree> known_binfos,
2781 vec<ipa_agg_jump_function_p> known_aggs,
2782 inline_hints *hints)
2784 struct inline_edge_summary *es = inline_edge_summary (e);
2785 int call_size = es->call_stmt_size;
2786 int call_time = es->call_stmt_time;
2787 if (!e->callee
2788 && estimate_edge_devirt_benefit (e, &call_size, &call_time,
2789 known_vals, known_binfos, known_aggs)
2790 && hints && cgraph_maybe_hot_edge_p (e))
2791 *hints |= INLINE_HINT_indirect_call;
2792 *size += call_size * INLINE_SIZE_SCALE;
2793 *time += apply_probability ((gcov_type) call_time, prob)
2794 * e->frequency * (INLINE_TIME_SCALE / CGRAPH_FREQ_BASE);
2795 if (*time > MAX_TIME * INLINE_TIME_SCALE)
2796 *time = MAX_TIME * INLINE_TIME_SCALE;
2801 /* Increase SIZE and TIME for size and time needed to handle all calls in NODE.
2802 POSSIBLE_TRUTHS, KNOWN_VALS and KNOWN_BINFOS describe context of the call
2803 site. */
2805 static void
2806 estimate_calls_size_and_time (struct cgraph_node *node, int *size, int *time,
2807 inline_hints *hints,
2808 clause_t possible_truths,
2809 vec<tree> known_vals,
2810 vec<tree> known_binfos,
2811 vec<ipa_agg_jump_function_p> known_aggs)
2813 struct cgraph_edge *e;
2814 for (e = node->callees; e; e = e->next_callee)
2816 struct inline_edge_summary *es = inline_edge_summary (e);
2817 if (!es->predicate
2818 || evaluate_predicate (es->predicate, possible_truths))
2820 if (e->inline_failed)
2822 /* Predicates of calls shall not use NOT_CHANGED codes,
2823 sowe do not need to compute probabilities. */
2824 estimate_edge_size_and_time (e, size, time, REG_BR_PROB_BASE,
2825 known_vals, known_binfos,
2826 known_aggs, hints);
2828 else
2829 estimate_calls_size_and_time (e->callee, size, time, hints,
2830 possible_truths,
2831 known_vals, known_binfos,
2832 known_aggs);
2835 for (e = node->indirect_calls; e; e = e->next_callee)
2837 struct inline_edge_summary *es = inline_edge_summary (e);
2838 if (!es->predicate
2839 || evaluate_predicate (es->predicate, possible_truths))
2840 estimate_edge_size_and_time (e, size, time, REG_BR_PROB_BASE,
2841 known_vals, known_binfos, known_aggs,
2842 hints);
2847 /* Estimate size and time needed to execute NODE assuming
2848 POSSIBLE_TRUTHS clause, and KNOWN_VALS and KNOWN_BINFOS information
2849 about NODE's arguments. */
2851 static void
2852 estimate_node_size_and_time (struct cgraph_node *node,
2853 clause_t possible_truths,
2854 vec<tree> known_vals,
2855 vec<tree> known_binfos,
2856 vec<ipa_agg_jump_function_p> known_aggs,
2857 int *ret_size, int *ret_time,
2858 inline_hints *ret_hints,
2859 vec<inline_param_summary_t>
2860 inline_param_summary)
2862 struct inline_summary *info = inline_summary (node);
2863 size_time_entry *e;
2864 int size = 0;
2865 int time = 0;
2866 inline_hints hints = 0;
2867 int i;
2869 if (dump_file && (dump_flags & TDF_DETAILS))
2871 bool found = false;
2872 fprintf (dump_file, " Estimating body: %s/%i\n"
2873 " Known to be false: ", cgraph_node_name (node),
2874 node->symbol.order);
2876 for (i = predicate_not_inlined_condition;
2877 i < (predicate_first_dynamic_condition
2878 + (int) vec_safe_length (info->conds)); i++)
2879 if (!(possible_truths & (1 << i)))
2881 if (found)
2882 fprintf (dump_file, ", ");
2883 found = true;
2884 dump_condition (dump_file, info->conds, i);
2888 for (i = 0; vec_safe_iterate (info->entry, i, &e); i++)
2889 if (evaluate_predicate (&e->predicate, possible_truths))
2891 size += e->size;
2892 gcc_checking_assert (e->time >= 0);
2893 gcc_checking_assert (time >= 0);
2894 if (!inline_param_summary.exists ())
2895 time += e->time;
2896 else
2898 int prob = predicate_probability (info->conds,
2899 &e->predicate,
2900 possible_truths,
2901 inline_param_summary);
2902 gcc_checking_assert (prob >= 0);
2903 gcc_checking_assert (prob <= REG_BR_PROB_BASE);
2904 time += apply_probability ((gcov_type) e->time, prob);
2906 if (time > MAX_TIME * INLINE_TIME_SCALE)
2907 time = MAX_TIME * INLINE_TIME_SCALE;
2908 gcc_checking_assert (time >= 0);
2911 gcc_checking_assert (size >= 0);
2912 gcc_checking_assert (time >= 0);
2914 if (info->loop_iterations
2915 && !evaluate_predicate (info->loop_iterations, possible_truths))
2916 hints |= INLINE_HINT_loop_iterations;
2917 if (info->loop_stride
2918 && !evaluate_predicate (info->loop_stride, possible_truths))
2919 hints |= INLINE_HINT_loop_stride;
2920 if (info->array_index
2921 && !evaluate_predicate (info->array_index, possible_truths))
2922 hints |= INLINE_HINT_array_index;
2923 if (info->scc_no)
2924 hints |= INLINE_HINT_in_scc;
2925 if (DECL_DECLARED_INLINE_P (node->symbol.decl))
2926 hints |= INLINE_HINT_declared_inline;
2928 estimate_calls_size_and_time (node, &size, &time, &hints, possible_truths,
2929 known_vals, known_binfos, known_aggs);
2930 gcc_checking_assert (size >= 0);
2931 gcc_checking_assert (time >= 0);
2932 time = RDIV (time, INLINE_TIME_SCALE);
2933 size = RDIV (size, INLINE_SIZE_SCALE);
2935 if (dump_file && (dump_flags & TDF_DETAILS))
2936 fprintf (dump_file, "\n size:%i time:%i\n", (int) size, (int) time);
2937 if (ret_time)
2938 *ret_time = time;
2939 if (ret_size)
2940 *ret_size = size;
2941 if (ret_hints)
2942 *ret_hints = hints;
2943 return;
2947 /* Estimate size and time needed to execute callee of EDGE assuming that
2948 parameters known to be constant at caller of EDGE are propagated.
2949 KNOWN_VALS and KNOWN_BINFOS are vectors of assumed known constant values
2950 and types for parameters. */
2952 void
2953 estimate_ipcp_clone_size_and_time (struct cgraph_node *node,
2954 vec<tree> known_vals,
2955 vec<tree> known_binfos,
2956 vec<ipa_agg_jump_function_p> known_aggs,
2957 int *ret_size, int *ret_time,
2958 inline_hints *hints)
2960 clause_t clause;
2962 clause = evaluate_conditions_for_known_args (node, false, known_vals,
2963 known_aggs);
2964 estimate_node_size_and_time (node, clause, known_vals, known_binfos,
2965 known_aggs, ret_size, ret_time, hints, vNULL);
2968 /* Translate all conditions from callee representation into caller
2969 representation and symbolically evaluate predicate P into new predicate.
2971 INFO is inline_summary of function we are adding predicate into, CALLEE_INFO
2972 is summary of function predicate P is from. OPERAND_MAP is array giving
2973 callee formal IDs the caller formal IDs. POSSSIBLE_TRUTHS is clausule of all
2974 callee conditions that may be true in caller context. TOPLEV_PREDICATE is
2975 predicate under which callee is executed. OFFSET_MAP is an array of of
2976 offsets that need to be added to conditions, negative offset means that
2977 conditions relying on values passed by reference have to be discarded
2978 because they might not be preserved (and should be considered offset zero
2979 for other purposes). */
2981 static struct predicate
2982 remap_predicate (struct inline_summary *info,
2983 struct inline_summary *callee_info,
2984 struct predicate *p,
2985 vec<int> operand_map,
2986 vec<int> offset_map,
2987 clause_t possible_truths, struct predicate *toplev_predicate)
2989 int i;
2990 struct predicate out = true_predicate ();
2992 /* True predicate is easy. */
2993 if (true_predicate_p (p))
2994 return *toplev_predicate;
2995 for (i = 0; p->clause[i]; i++)
2997 clause_t clause = p->clause[i];
2998 int cond;
2999 struct predicate clause_predicate = false_predicate ();
3001 gcc_assert (i < MAX_CLAUSES);
3003 for (cond = 0; cond < NUM_CONDITIONS; cond++)
3004 /* Do we have condition we can't disprove? */
3005 if (clause & possible_truths & (1 << cond))
3007 struct predicate cond_predicate;
3008 /* Work out if the condition can translate to predicate in the
3009 inlined function. */
3010 if (cond >= predicate_first_dynamic_condition)
3012 struct condition *c;
3014 c = &(*callee_info->conds)[cond
3016 predicate_first_dynamic_condition];
3017 /* See if we can remap condition operand to caller's operand.
3018 Otherwise give up. */
3019 if (!operand_map.exists ()
3020 || (int) operand_map.length () <= c->operand_num
3021 || operand_map[c->operand_num] == -1
3022 /* TODO: For non-aggregate conditions, adding an offset is
3023 basically an arithmetic jump function processing which
3024 we should support in future. */
3025 || ((!c->agg_contents || !c->by_ref)
3026 && offset_map[c->operand_num] > 0)
3027 || (c->agg_contents && c->by_ref
3028 && offset_map[c->operand_num] < 0))
3029 cond_predicate = true_predicate ();
3030 else
3032 struct agg_position_info ap;
3033 HOST_WIDE_INT offset_delta = offset_map[c->operand_num];
3034 if (offset_delta < 0)
3036 gcc_checking_assert (!c->agg_contents || !c->by_ref);
3037 offset_delta = 0;
3039 gcc_assert (!c->agg_contents
3040 || c->by_ref || offset_delta == 0);
3041 ap.offset = c->offset + offset_delta;
3042 ap.agg_contents = c->agg_contents;
3043 ap.by_ref = c->by_ref;
3044 cond_predicate = add_condition (info,
3045 operand_map[c->operand_num],
3046 &ap, c->code, c->val);
3049 /* Fixed conditions remains same, construct single
3050 condition predicate. */
3051 else
3053 cond_predicate.clause[0] = 1 << cond;
3054 cond_predicate.clause[1] = 0;
3056 clause_predicate = or_predicates (info->conds, &clause_predicate,
3057 &cond_predicate);
3059 out = and_predicates (info->conds, &out, &clause_predicate);
3061 return and_predicates (info->conds, &out, toplev_predicate);
3065 /* Update summary information of inline clones after inlining.
3066 Compute peak stack usage. */
3068 static void
3069 inline_update_callee_summaries (struct cgraph_node *node, int depth)
3071 struct cgraph_edge *e;
3072 struct inline_summary *callee_info = inline_summary (node);
3073 struct inline_summary *caller_info = inline_summary (node->callers->caller);
3074 HOST_WIDE_INT peak;
3076 callee_info->stack_frame_offset
3077 = caller_info->stack_frame_offset
3078 + caller_info->estimated_self_stack_size;
3079 peak = callee_info->stack_frame_offset
3080 + callee_info->estimated_self_stack_size;
3081 if (inline_summary (node->global.inlined_to)->estimated_stack_size < peak)
3082 inline_summary (node->global.inlined_to)->estimated_stack_size = peak;
3083 cgraph_propagate_frequency (node);
3084 for (e = node->callees; e; e = e->next_callee)
3086 if (!e->inline_failed)
3087 inline_update_callee_summaries (e->callee, depth);
3088 inline_edge_summary (e)->loop_depth += depth;
3090 for (e = node->indirect_calls; e; e = e->next_callee)
3091 inline_edge_summary (e)->loop_depth += depth;
3094 /* Update change_prob of EDGE after INLINED_EDGE has been inlined.
3095 When functoin A is inlined in B and A calls C with parameter that
3096 changes with probability PROB1 and C is known to be passthroug
3097 of argument if B that change with probability PROB2, the probability
3098 of change is now PROB1*PROB2. */
3100 static void
3101 remap_edge_change_prob (struct cgraph_edge *inlined_edge,
3102 struct cgraph_edge *edge)
3104 if (ipa_node_params_vector.exists ())
3106 int i;
3107 struct ipa_edge_args *args = IPA_EDGE_REF (edge);
3108 struct inline_edge_summary *es = inline_edge_summary (edge);
3109 struct inline_edge_summary *inlined_es
3110 = inline_edge_summary (inlined_edge);
3112 for (i = 0; i < ipa_get_cs_argument_count (args); i++)
3114 struct ipa_jump_func *jfunc = ipa_get_ith_jump_func (args, i);
3115 if (jfunc->type == IPA_JF_PASS_THROUGH
3116 && (ipa_get_jf_pass_through_formal_id (jfunc)
3117 < (int) inlined_es->param.length ()))
3119 int jf_formal_id = ipa_get_jf_pass_through_formal_id (jfunc);
3120 int prob1 = es->param[i].change_prob;
3121 int prob2 = inlined_es->param[jf_formal_id].change_prob;
3122 int prob = combine_probabilities (prob1, prob2);
3124 if (prob1 && prob2 && !prob)
3125 prob = 1;
3127 es->param[i].change_prob = prob;
3133 /* Update edge summaries of NODE after INLINED_EDGE has been inlined.
3135 Remap predicates of callees of NODE. Rest of arguments match
3136 remap_predicate.
3138 Also update change probabilities. */
3140 static void
3141 remap_edge_summaries (struct cgraph_edge *inlined_edge,
3142 struct cgraph_node *node,
3143 struct inline_summary *info,
3144 struct inline_summary *callee_info,
3145 vec<int> operand_map,
3146 vec<int> offset_map,
3147 clause_t possible_truths,
3148 struct predicate *toplev_predicate)
3150 struct cgraph_edge *e;
3151 for (e = node->callees; e; e = e->next_callee)
3153 struct inline_edge_summary *es = inline_edge_summary (e);
3154 struct predicate p;
3156 if (e->inline_failed)
3158 remap_edge_change_prob (inlined_edge, e);
3160 if (es->predicate)
3162 p = remap_predicate (info, callee_info,
3163 es->predicate, operand_map, offset_map,
3164 possible_truths, toplev_predicate);
3165 edge_set_predicate (e, &p);
3166 /* TODO: We should remove the edge for code that will be
3167 optimized out, but we need to keep verifiers and tree-inline
3168 happy. Make it cold for now. */
3169 if (false_predicate_p (&p))
3171 e->count = 0;
3172 e->frequency = 0;
3175 else
3176 edge_set_predicate (e, toplev_predicate);
3178 else
3179 remap_edge_summaries (inlined_edge, e->callee, info, callee_info,
3180 operand_map, offset_map, possible_truths,
3181 toplev_predicate);
3183 for (e = node->indirect_calls; e; e = e->next_callee)
3185 struct inline_edge_summary *es = inline_edge_summary (e);
3186 struct predicate p;
3188 remap_edge_change_prob (inlined_edge, e);
3189 if (es->predicate)
3191 p = remap_predicate (info, callee_info,
3192 es->predicate, operand_map, offset_map,
3193 possible_truths, toplev_predicate);
3194 edge_set_predicate (e, &p);
3195 /* TODO: We should remove the edge for code that will be optimized
3196 out, but we need to keep verifiers and tree-inline happy.
3197 Make it cold for now. */
3198 if (false_predicate_p (&p))
3200 e->count = 0;
3201 e->frequency = 0;
3204 else
3205 edge_set_predicate (e, toplev_predicate);
3209 /* Same as remap_predicate, but set result into hint *HINT. */
3211 static void
3212 remap_hint_predicate (struct inline_summary *info,
3213 struct inline_summary *callee_info,
3214 struct predicate **hint,
3215 vec<int> operand_map,
3216 vec<int> offset_map,
3217 clause_t possible_truths,
3218 struct predicate *toplev_predicate)
3220 predicate p;
3222 if (!*hint)
3223 return;
3224 p = remap_predicate (info, callee_info,
3225 *hint,
3226 operand_map, offset_map,
3227 possible_truths, toplev_predicate);
3228 if (!false_predicate_p (&p) && !true_predicate_p (&p))
3230 if (!*hint)
3231 set_hint_predicate (hint, p);
3232 else
3233 **hint = and_predicates (info->conds, *hint, &p);
3237 /* We inlined EDGE. Update summary of the function we inlined into. */
3239 void
3240 inline_merge_summary (struct cgraph_edge *edge)
3242 struct inline_summary *callee_info = inline_summary (edge->callee);
3243 struct cgraph_node *to = (edge->caller->global.inlined_to
3244 ? edge->caller->global.inlined_to : edge->caller);
3245 struct inline_summary *info = inline_summary (to);
3246 clause_t clause = 0; /* not_inline is known to be false. */
3247 size_time_entry *e;
3248 vec<int> operand_map = vNULL;
3249 vec<int> offset_map = vNULL;
3250 int i;
3251 struct predicate toplev_predicate;
3252 struct predicate true_p = true_predicate ();
3253 struct inline_edge_summary *es = inline_edge_summary (edge);
3255 if (es->predicate)
3256 toplev_predicate = *es->predicate;
3257 else
3258 toplev_predicate = true_predicate ();
3260 if (ipa_node_params_vector.exists () && callee_info->conds)
3262 struct ipa_edge_args *args = IPA_EDGE_REF (edge);
3263 int count = ipa_get_cs_argument_count (args);
3264 int i;
3266 evaluate_properties_for_edge (edge, true, &clause, NULL, NULL, NULL);
3267 if (count)
3269 operand_map.safe_grow_cleared (count);
3270 offset_map.safe_grow_cleared (count);
3272 for (i = 0; i < count; i++)
3274 struct ipa_jump_func *jfunc = ipa_get_ith_jump_func (args, i);
3275 int map = -1;
3277 /* TODO: handle non-NOPs when merging. */
3278 if (jfunc->type == IPA_JF_PASS_THROUGH)
3280 if (ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
3281 map = ipa_get_jf_pass_through_formal_id (jfunc);
3282 if (!ipa_get_jf_pass_through_agg_preserved (jfunc))
3283 offset_map[i] = -1;
3285 else if (jfunc->type == IPA_JF_ANCESTOR)
3287 HOST_WIDE_INT offset = ipa_get_jf_ancestor_offset (jfunc);
3288 if (offset >= 0 && offset < INT_MAX)
3290 map = ipa_get_jf_ancestor_formal_id (jfunc);
3291 if (!ipa_get_jf_ancestor_agg_preserved (jfunc))
3292 offset = -1;
3293 offset_map[i] = offset;
3296 operand_map[i] = map;
3297 gcc_assert (map < ipa_get_param_count (IPA_NODE_REF (to)));
3300 for (i = 0; vec_safe_iterate (callee_info->entry, i, &e); i++)
3302 struct predicate p = remap_predicate (info, callee_info,
3303 &e->predicate, operand_map,
3304 offset_map, clause,
3305 &toplev_predicate);
3306 if (!false_predicate_p (&p))
3308 gcov_type add_time = ((gcov_type) e->time * edge->frequency
3309 + CGRAPH_FREQ_BASE / 2) / CGRAPH_FREQ_BASE;
3310 int prob = predicate_probability (callee_info->conds,
3311 &e->predicate,
3312 clause, es->param);
3313 add_time = apply_probability ((gcov_type) add_time, prob);
3314 if (add_time > MAX_TIME * INLINE_TIME_SCALE)
3315 add_time = MAX_TIME * INLINE_TIME_SCALE;
3316 if (prob != REG_BR_PROB_BASE
3317 && dump_file && (dump_flags & TDF_DETAILS))
3319 fprintf (dump_file, "\t\tScaling time by probability:%f\n",
3320 (double) prob / REG_BR_PROB_BASE);
3322 account_size_time (info, e->size, add_time, &p);
3325 remap_edge_summaries (edge, edge->callee, info, callee_info, operand_map,
3326 offset_map, clause, &toplev_predicate);
3327 remap_hint_predicate (info, callee_info,
3328 &callee_info->loop_iterations,
3329 operand_map, offset_map, clause, &toplev_predicate);
3330 remap_hint_predicate (info, callee_info,
3331 &callee_info->loop_stride,
3332 operand_map, offset_map, clause, &toplev_predicate);
3333 remap_hint_predicate (info, callee_info,
3334 &callee_info->array_index,
3335 operand_map, offset_map, clause, &toplev_predicate);
3337 inline_update_callee_summaries (edge->callee,
3338 inline_edge_summary (edge)->loop_depth);
3340 /* We do not maintain predicates of inlined edges, free it. */
3341 edge_set_predicate (edge, &true_p);
3342 /* Similarly remove param summaries. */
3343 es->param.release ();
3344 operand_map.release ();
3345 offset_map.release ();
3348 /* For performance reasons inline_merge_summary is not updating overall size
3349 and time. Recompute it. */
3351 void
3352 inline_update_overall_summary (struct cgraph_node *node)
3354 struct inline_summary *info = inline_summary (node);
3355 size_time_entry *e;
3356 int i;
3358 info->size = 0;
3359 info->time = 0;
3360 for (i = 0; vec_safe_iterate (info->entry, i, &e); i++)
3362 info->size += e->size, info->time += e->time;
3363 if (info->time > MAX_TIME * INLINE_TIME_SCALE)
3364 info->time = MAX_TIME * INLINE_TIME_SCALE;
3366 estimate_calls_size_and_time (node, &info->size, &info->time, NULL,
3367 ~(clause_t) (1 << predicate_false_condition),
3368 vNULL, vNULL, vNULL);
3369 info->time = (info->time + INLINE_TIME_SCALE / 2) / INLINE_TIME_SCALE;
3370 info->size = (info->size + INLINE_SIZE_SCALE / 2) / INLINE_SIZE_SCALE;
3373 /* Return hints derrived from EDGE. */
3375 simple_edge_hints (struct cgraph_edge *edge)
3377 int hints = 0;
3378 struct cgraph_node *to = (edge->caller->global.inlined_to
3379 ? edge->caller->global.inlined_to : edge->caller);
3380 if (inline_summary (to)->scc_no
3381 && inline_summary (to)->scc_no == inline_summary (edge->callee)->scc_no
3382 && !cgraph_edge_recursive_p (edge))
3383 hints |= INLINE_HINT_same_scc;
3385 if (to->symbol.lto_file_data && edge->callee->symbol.lto_file_data
3386 && to->symbol.lto_file_data != edge->callee->symbol.lto_file_data)
3387 hints |= INLINE_HINT_cross_module;
3389 return hints;
3392 /* Estimate the time cost for the caller when inlining EDGE.
3393 Only to be called via estimate_edge_time, that handles the
3394 caching mechanism.
3396 When caching, also update the cache entry. Compute both time and
3397 size, since we always need both metrics eventually. */
3400 do_estimate_edge_time (struct cgraph_edge *edge)
3402 int time;
3403 int size;
3404 inline_hints hints;
3405 struct cgraph_node *callee;
3406 clause_t clause;
3407 vec<tree> known_vals;
3408 vec<tree> known_binfos;
3409 vec<ipa_agg_jump_function_p> known_aggs;
3410 struct inline_edge_summary *es = inline_edge_summary (edge);
3412 callee = cgraph_function_or_thunk_node (edge->callee, NULL);
3414 gcc_checking_assert (edge->inline_failed);
3415 evaluate_properties_for_edge (edge, true,
3416 &clause, &known_vals, &known_binfos,
3417 &known_aggs);
3418 estimate_node_size_and_time (callee, clause, known_vals, known_binfos,
3419 known_aggs, &size, &time, &hints, es->param);
3420 known_vals.release ();
3421 known_binfos.release ();
3422 known_aggs.release ();
3423 gcc_checking_assert (size >= 0);
3424 gcc_checking_assert (time >= 0);
3426 /* When caching, update the cache entry. */
3427 if (edge_growth_cache.exists ())
3429 if ((int) edge_growth_cache.length () <= edge->uid)
3430 edge_growth_cache.safe_grow_cleared (cgraph_edge_max_uid);
3431 edge_growth_cache[edge->uid].time = time + (time >= 0);
3433 edge_growth_cache[edge->uid].size = size + (size >= 0);
3434 hints |= simple_edge_hints (edge);
3435 edge_growth_cache[edge->uid].hints = hints + 1;
3437 return time;
3441 /* Return estimated callee growth after inlining EDGE.
3442 Only to be called via estimate_edge_size. */
3445 do_estimate_edge_size (struct cgraph_edge *edge)
3447 int size;
3448 struct cgraph_node *callee;
3449 clause_t clause;
3450 vec<tree> known_vals;
3451 vec<tree> known_binfos;
3452 vec<ipa_agg_jump_function_p> known_aggs;
3454 /* When we do caching, use do_estimate_edge_time to populate the entry. */
3456 if (edge_growth_cache.exists ())
3458 do_estimate_edge_time (edge);
3459 size = edge_growth_cache[edge->uid].size;
3460 gcc_checking_assert (size);
3461 return size - (size > 0);
3464 callee = cgraph_function_or_thunk_node (edge->callee, NULL);
3466 /* Early inliner runs without caching, go ahead and do the dirty work. */
3467 gcc_checking_assert (edge->inline_failed);
3468 evaluate_properties_for_edge (edge, true,
3469 &clause, &known_vals, &known_binfos,
3470 &known_aggs);
3471 estimate_node_size_and_time (callee, clause, known_vals, known_binfos,
3472 known_aggs, &size, NULL, NULL, vNULL);
3473 known_vals.release ();
3474 known_binfos.release ();
3475 known_aggs.release ();
3476 return size;
3480 /* Estimate the growth of the caller when inlining EDGE.
3481 Only to be called via estimate_edge_size. */
3483 inline_hints
3484 do_estimate_edge_hints (struct cgraph_edge *edge)
3486 inline_hints hints;
3487 struct cgraph_node *callee;
3488 clause_t clause;
3489 vec<tree> known_vals;
3490 vec<tree> known_binfos;
3491 vec<ipa_agg_jump_function_p> known_aggs;
3493 /* When we do caching, use do_estimate_edge_time to populate the entry. */
3495 if (edge_growth_cache.exists ())
3497 do_estimate_edge_time (edge);
3498 hints = edge_growth_cache[edge->uid].hints;
3499 gcc_checking_assert (hints);
3500 return hints - 1;
3503 callee = cgraph_function_or_thunk_node (edge->callee, NULL);
3505 /* Early inliner runs without caching, go ahead and do the dirty work. */
3506 gcc_checking_assert (edge->inline_failed);
3507 evaluate_properties_for_edge (edge, true,
3508 &clause, &known_vals, &known_binfos,
3509 &known_aggs);
3510 estimate_node_size_and_time (callee, clause, known_vals, known_binfos,
3511 known_aggs, NULL, NULL, &hints, vNULL);
3512 known_vals.release ();
3513 known_binfos.release ();
3514 known_aggs.release ();
3515 hints |= simple_edge_hints (edge);
3516 return hints;
3520 /* Estimate self time of the function NODE after inlining EDGE. */
3523 estimate_time_after_inlining (struct cgraph_node *node,
3524 struct cgraph_edge *edge)
3526 struct inline_edge_summary *es = inline_edge_summary (edge);
3527 if (!es->predicate || !false_predicate_p (es->predicate))
3529 gcov_type time =
3530 inline_summary (node)->time + estimate_edge_time (edge);
3531 if (time < 0)
3532 time = 0;
3533 if (time > MAX_TIME)
3534 time = MAX_TIME;
3535 return time;
3537 return inline_summary (node)->time;
3541 /* Estimate the size of NODE after inlining EDGE which should be an
3542 edge to either NODE or a call inlined into NODE. */
3545 estimate_size_after_inlining (struct cgraph_node *node,
3546 struct cgraph_edge *edge)
3548 struct inline_edge_summary *es = inline_edge_summary (edge);
3549 if (!es->predicate || !false_predicate_p (es->predicate))
3551 int size = inline_summary (node)->size + estimate_edge_growth (edge);
3552 gcc_assert (size >= 0);
3553 return size;
3555 return inline_summary (node)->size;
3559 struct growth_data
3561 bool self_recursive;
3562 int growth;
3566 /* Worker for do_estimate_growth. Collect growth for all callers. */
3568 static bool
3569 do_estimate_growth_1 (struct cgraph_node *node, void *data)
3571 struct cgraph_edge *e;
3572 struct growth_data *d = (struct growth_data *) data;
3574 for (e = node->callers; e; e = e->next_caller)
3576 gcc_checking_assert (e->inline_failed);
3578 if (e->caller == node
3579 || (e->caller->global.inlined_to
3580 && e->caller->global.inlined_to == node))
3581 d->self_recursive = true;
3582 d->growth += estimate_edge_growth (e);
3584 return false;
3588 /* Estimate the growth caused by inlining NODE into all callees. */
3591 do_estimate_growth (struct cgraph_node *node)
3593 struct growth_data d = { 0, false };
3594 struct inline_summary *info = inline_summary (node);
3596 cgraph_for_node_and_aliases (node, do_estimate_growth_1, &d, true);
3598 /* For self recursive functions the growth estimation really should be
3599 infinity. We don't want to return very large values because the growth
3600 plays various roles in badness computation fractions. Be sure to not
3601 return zero or negative growths. */
3602 if (d.self_recursive)
3603 d.growth = d.growth < info->size ? info->size : d.growth;
3604 else if (DECL_EXTERNAL (node->symbol.decl))
3606 else
3608 if (cgraph_will_be_removed_from_program_if_no_direct_calls (node))
3609 d.growth -= info->size;
3610 /* COMDAT functions are very often not shared across multiple units
3611 since they come from various template instantiations.
3612 Take this into account. */
3613 else if (DECL_COMDAT (node->symbol.decl)
3614 && cgraph_can_remove_if_no_direct_calls_p (node))
3615 d.growth -= (info->size
3616 * (100 - PARAM_VALUE (PARAM_COMDAT_SHARING_PROBABILITY))
3617 + 50) / 100;
3620 if (node_growth_cache.exists ())
3622 if ((int) node_growth_cache.length () <= node->uid)
3623 node_growth_cache.safe_grow_cleared (cgraph_max_uid);
3624 node_growth_cache[node->uid] = d.growth + (d.growth >= 0);
3626 return d.growth;
3630 /* This function performs intraprocedural analysis in NODE that is required to
3631 inline indirect calls. */
3633 static void
3634 inline_indirect_intraprocedural_analysis (struct cgraph_node *node)
3636 ipa_analyze_node (node);
3637 if (dump_file && (dump_flags & TDF_DETAILS))
3639 ipa_print_node_params (dump_file, node);
3640 ipa_print_node_jump_functions (dump_file, node);
3645 /* Note function body size. */
3647 static void
3648 inline_analyze_function (struct cgraph_node *node)
3650 push_cfun (DECL_STRUCT_FUNCTION (node->symbol.decl));
3652 if (dump_file)
3653 fprintf (dump_file, "\nAnalyzing function: %s/%u\n",
3654 cgraph_node_name (node), node->symbol.order);
3655 if (optimize && !node->thunk.thunk_p)
3656 inline_indirect_intraprocedural_analysis (node);
3657 compute_inline_parameters (node, false);
3659 pop_cfun ();
3663 /* Called when new function is inserted to callgraph late. */
3665 static void
3666 add_new_function (struct cgraph_node *node, void *data ATTRIBUTE_UNUSED)
3668 inline_analyze_function (node);
3672 /* Note function body size. */
3674 void
3675 inline_generate_summary (void)
3677 struct cgraph_node *node;
3679 function_insertion_hook_holder =
3680 cgraph_add_function_insertion_hook (&add_new_function, NULL);
3682 ipa_register_cgraph_hooks ();
3683 inline_free_summary ();
3685 FOR_EACH_DEFINED_FUNCTION (node)
3686 if (!node->alias)
3687 inline_analyze_function (node);
3691 /* Read predicate from IB. */
3693 static struct predicate
3694 read_predicate (struct lto_input_block *ib)
3696 struct predicate out;
3697 clause_t clause;
3698 int k = 0;
3702 gcc_assert (k <= MAX_CLAUSES);
3703 clause = out.clause[k++] = streamer_read_uhwi (ib);
3705 while (clause);
3707 /* Zero-initialize the remaining clauses in OUT. */
3708 while (k <= MAX_CLAUSES)
3709 out.clause[k++] = 0;
3711 return out;
3715 /* Write inline summary for edge E to OB. */
3717 static void
3718 read_inline_edge_summary (struct lto_input_block *ib, struct cgraph_edge *e)
3720 struct inline_edge_summary *es = inline_edge_summary (e);
3721 struct predicate p;
3722 int length, i;
3724 es->call_stmt_size = streamer_read_uhwi (ib);
3725 es->call_stmt_time = streamer_read_uhwi (ib);
3726 es->loop_depth = streamer_read_uhwi (ib);
3727 p = read_predicate (ib);
3728 edge_set_predicate (e, &p);
3729 length = streamer_read_uhwi (ib);
3730 if (length)
3732 es->param.safe_grow_cleared (length);
3733 for (i = 0; i < length; i++)
3734 es->param[i].change_prob = streamer_read_uhwi (ib);
3739 /* Stream in inline summaries from the section. */
3741 static void
3742 inline_read_section (struct lto_file_decl_data *file_data, const char *data,
3743 size_t len)
3745 const struct lto_function_header *header =
3746 (const struct lto_function_header *) data;
3747 const int cfg_offset = sizeof (struct lto_function_header);
3748 const int main_offset = cfg_offset + header->cfg_size;
3749 const int string_offset = main_offset + header->main_size;
3750 struct data_in *data_in;
3751 struct lto_input_block ib;
3752 unsigned int i, count2, j;
3753 unsigned int f_count;
3755 LTO_INIT_INPUT_BLOCK (ib, (const char *) data + main_offset, 0,
3756 header->main_size);
3758 data_in =
3759 lto_data_in_create (file_data, (const char *) data + string_offset,
3760 header->string_size, vNULL);
3761 f_count = streamer_read_uhwi (&ib);
3762 for (i = 0; i < f_count; i++)
3764 unsigned int index;
3765 struct cgraph_node *node;
3766 struct inline_summary *info;
3767 lto_symtab_encoder_t encoder;
3768 struct bitpack_d bp;
3769 struct cgraph_edge *e;
3770 predicate p;
3772 index = streamer_read_uhwi (&ib);
3773 encoder = file_data->symtab_node_encoder;
3774 node = cgraph (lto_symtab_encoder_deref (encoder, index));
3775 info = inline_summary (node);
3777 info->estimated_stack_size
3778 = info->estimated_self_stack_size = streamer_read_uhwi (&ib);
3779 info->size = info->self_size = streamer_read_uhwi (&ib);
3780 info->time = info->self_time = streamer_read_uhwi (&ib);
3782 bp = streamer_read_bitpack (&ib);
3783 info->inlinable = bp_unpack_value (&bp, 1);
3785 count2 = streamer_read_uhwi (&ib);
3786 gcc_assert (!info->conds);
3787 for (j = 0; j < count2; j++)
3789 struct condition c;
3790 c.operand_num = streamer_read_uhwi (&ib);
3791 c.code = (enum tree_code) streamer_read_uhwi (&ib);
3792 c.val = stream_read_tree (&ib, data_in);
3793 bp = streamer_read_bitpack (&ib);
3794 c.agg_contents = bp_unpack_value (&bp, 1);
3795 c.by_ref = bp_unpack_value (&bp, 1);
3796 if (c.agg_contents)
3797 c.offset = streamer_read_uhwi (&ib);
3798 vec_safe_push (info->conds, c);
3800 count2 = streamer_read_uhwi (&ib);
3801 gcc_assert (!info->entry);
3802 for (j = 0; j < count2; j++)
3804 struct size_time_entry e;
3806 e.size = streamer_read_uhwi (&ib);
3807 e.time = streamer_read_uhwi (&ib);
3808 e.predicate = read_predicate (&ib);
3810 vec_safe_push (info->entry, e);
3813 p = read_predicate (&ib);
3814 set_hint_predicate (&info->loop_iterations, p);
3815 p = read_predicate (&ib);
3816 set_hint_predicate (&info->loop_stride, p);
3817 p = read_predicate (&ib);
3818 set_hint_predicate (&info->array_index, p);
3819 for (e = node->callees; e; e = e->next_callee)
3820 read_inline_edge_summary (&ib, e);
3821 for (e = node->indirect_calls; e; e = e->next_callee)
3822 read_inline_edge_summary (&ib, e);
3825 lto_free_section_data (file_data, LTO_section_inline_summary, NULL, data,
3826 len);
3827 lto_data_in_delete (data_in);
3831 /* Read inline summary. Jump functions are shared among ipa-cp
3832 and inliner, so when ipa-cp is active, we don't need to write them
3833 twice. */
3835 void
3836 inline_read_summary (void)
3838 struct lto_file_decl_data **file_data_vec = lto_get_file_decl_data ();
3839 struct lto_file_decl_data *file_data;
3840 unsigned int j = 0;
3842 inline_summary_alloc ();
3844 while ((file_data = file_data_vec[j++]))
3846 size_t len;
3847 const char *data = lto_get_section_data (file_data,
3848 LTO_section_inline_summary,
3849 NULL, &len);
3850 if (data)
3851 inline_read_section (file_data, data, len);
3852 else
3853 /* Fatal error here. We do not want to support compiling ltrans units
3854 with different version of compiler or different flags than the WPA
3855 unit, so this should never happen. */
3856 fatal_error ("ipa inline summary is missing in input file");
3858 if (optimize)
3860 ipa_register_cgraph_hooks ();
3861 if (!flag_ipa_cp)
3862 ipa_prop_read_jump_functions ();
3864 function_insertion_hook_holder =
3865 cgraph_add_function_insertion_hook (&add_new_function, NULL);
3869 /* Write predicate P to OB. */
3871 static void
3872 write_predicate (struct output_block *ob, struct predicate *p)
3874 int j;
3875 if (p)
3876 for (j = 0; p->clause[j]; j++)
3878 gcc_assert (j < MAX_CLAUSES);
3879 streamer_write_uhwi (ob, p->clause[j]);
3881 streamer_write_uhwi (ob, 0);
3885 /* Write inline summary for edge E to OB. */
3887 static void
3888 write_inline_edge_summary (struct output_block *ob, struct cgraph_edge *e)
3890 struct inline_edge_summary *es = inline_edge_summary (e);
3891 int i;
3893 streamer_write_uhwi (ob, es->call_stmt_size);
3894 streamer_write_uhwi (ob, es->call_stmt_time);
3895 streamer_write_uhwi (ob, es->loop_depth);
3896 write_predicate (ob, es->predicate);
3897 streamer_write_uhwi (ob, es->param.length ());
3898 for (i = 0; i < (int) es->param.length (); i++)
3899 streamer_write_uhwi (ob, es->param[i].change_prob);
3903 /* Write inline summary for node in SET.
3904 Jump functions are shared among ipa-cp and inliner, so when ipa-cp is
3905 active, we don't need to write them twice. */
3907 void
3908 inline_write_summary (void)
3910 struct cgraph_node *node;
3911 struct output_block *ob = create_output_block (LTO_section_inline_summary);
3912 lto_symtab_encoder_t encoder = ob->decl_state->symtab_node_encoder;
3913 unsigned int count = 0;
3914 int i;
3916 for (i = 0; i < lto_symtab_encoder_size (encoder); i++)
3918 symtab_node snode = lto_symtab_encoder_deref (encoder, i);
3919 cgraph_node *cnode = dyn_cast <cgraph_node> (snode);
3920 if (cnode && cnode->analyzed)
3921 count++;
3923 streamer_write_uhwi (ob, count);
3925 for (i = 0; i < lto_symtab_encoder_size (encoder); i++)
3927 symtab_node snode = lto_symtab_encoder_deref (encoder, i);
3928 cgraph_node *cnode = dyn_cast <cgraph_node> (snode);
3929 if (cnode && (node = cnode)->analyzed)
3931 struct inline_summary *info = inline_summary (node);
3932 struct bitpack_d bp;
3933 struct cgraph_edge *edge;
3934 int i;
3935 size_time_entry *e;
3936 struct condition *c;
3938 streamer_write_uhwi (ob,
3939 lto_symtab_encoder_encode (encoder,
3940 (symtab_node)
3941 node));
3942 streamer_write_hwi (ob, info->estimated_self_stack_size);
3943 streamer_write_hwi (ob, info->self_size);
3944 streamer_write_hwi (ob, info->self_time);
3945 bp = bitpack_create (ob->main_stream);
3946 bp_pack_value (&bp, info->inlinable, 1);
3947 streamer_write_bitpack (&bp);
3948 streamer_write_uhwi (ob, vec_safe_length (info->conds));
3949 for (i = 0; vec_safe_iterate (info->conds, i, &c); i++)
3951 streamer_write_uhwi (ob, c->operand_num);
3952 streamer_write_uhwi (ob, c->code);
3953 stream_write_tree (ob, c->val, true);
3954 bp = bitpack_create (ob->main_stream);
3955 bp_pack_value (&bp, c->agg_contents, 1);
3956 bp_pack_value (&bp, c->by_ref, 1);
3957 streamer_write_bitpack (&bp);
3958 if (c->agg_contents)
3959 streamer_write_uhwi (ob, c->offset);
3961 streamer_write_uhwi (ob, vec_safe_length (info->entry));
3962 for (i = 0; vec_safe_iterate (info->entry, i, &e); i++)
3964 streamer_write_uhwi (ob, e->size);
3965 streamer_write_uhwi (ob, e->time);
3966 write_predicate (ob, &e->predicate);
3968 write_predicate (ob, info->loop_iterations);
3969 write_predicate (ob, info->loop_stride);
3970 write_predicate (ob, info->array_index);
3971 for (edge = node->callees; edge; edge = edge->next_callee)
3972 write_inline_edge_summary (ob, edge);
3973 for (edge = node->indirect_calls; edge; edge = edge->next_callee)
3974 write_inline_edge_summary (ob, edge);
3977 streamer_write_char_stream (ob->main_stream, 0);
3978 produce_asm (ob, NULL);
3979 destroy_output_block (ob);
3981 if (optimize && !flag_ipa_cp)
3982 ipa_prop_write_jump_functions ();
3986 /* Release inline summary. */
3988 void
3989 inline_free_summary (void)
3991 struct cgraph_node *node;
3992 if (!inline_edge_summary_vec.exists ())
3993 return;
3994 FOR_EACH_DEFINED_FUNCTION (node)
3995 reset_inline_summary (node);
3996 if (function_insertion_hook_holder)
3997 cgraph_remove_function_insertion_hook (function_insertion_hook_holder);
3998 function_insertion_hook_holder = NULL;
3999 if (node_removal_hook_holder)
4000 cgraph_remove_node_removal_hook (node_removal_hook_holder);
4001 node_removal_hook_holder = NULL;
4002 if (edge_removal_hook_holder)
4003 cgraph_remove_edge_removal_hook (edge_removal_hook_holder);
4004 edge_removal_hook_holder = NULL;
4005 if (node_duplication_hook_holder)
4006 cgraph_remove_node_duplication_hook (node_duplication_hook_holder);
4007 node_duplication_hook_holder = NULL;
4008 if (edge_duplication_hook_holder)
4009 cgraph_remove_edge_duplication_hook (edge_duplication_hook_holder);
4010 edge_duplication_hook_holder = NULL;
4011 vec_free (inline_summary_vec);
4012 inline_edge_summary_vec.release ();
4013 if (edge_predicate_pool)
4014 free_alloc_pool (edge_predicate_pool);
4015 edge_predicate_pool = 0;