Emit .note.GNU-stack for hard-float linux targets.
[official-gcc.git] / gcc / ipa.c
blob554819316682a3f2cd5950baba84b858edc03b16
1 /* Basic IPA optimizations and utilities.
2 Copyright (C) 2003-2020 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 #include "config.h"
21 #include "system.h"
22 #include "coretypes.h"
23 #include "backend.h"
24 #include "target.h"
25 #include "tree.h"
26 #include "gimple.h"
27 #include "alloc-pool.h"
28 #include "tree-pass.h"
29 #include "stringpool.h"
30 #include "cgraph.h"
31 #include "gimplify.h"
32 #include "tree-iterator.h"
33 #include "ipa-utils.h"
34 #include "symbol-summary.h"
35 #include "tree-vrp.h"
36 #include "ipa-prop.h"
37 #include "ipa-fnsummary.h"
38 #include "dbgcnt.h"
39 #include "debug.h"
40 #include "stringpool.h"
41 #include "attribs.h"
43 /* Return true when NODE has ADDR reference. */
45 static bool
46 has_addr_references_p (struct cgraph_node *node,
47 void *)
49 int i;
50 struct ipa_ref *ref = NULL;
52 for (i = 0; node->iterate_referring (i, ref); i++)
53 if (ref->use == IPA_REF_ADDR)
54 return true;
55 return false;
58 /* Return true when NODE can be target of an indirect call. */
60 static bool
61 is_indirect_call_target_p (struct cgraph_node *node, void *)
63 return node->indirect_call_target;
66 /* Look for all functions inlined to NODE and update their inlined_to pointers
67 to INLINED_TO. */
69 static void
70 update_inlined_to_pointer (struct cgraph_node *node, struct cgraph_node *inlined_to)
72 struct cgraph_edge *e;
73 for (e = node->callees; e; e = e->next_callee)
74 if (e->callee->inlined_to)
76 e->callee->inlined_to = inlined_to;
77 update_inlined_to_pointer (e->callee, inlined_to);
81 /* Add symtab NODE to queue starting at FIRST.
83 The queue is linked via AUX pointers and terminated by pointer to 1.
84 We enqueue nodes at two occasions: when we find them reachable or when we find
85 their bodies needed for further clonning. In the second case we mark them
86 by pointer to 2 after processing so they are re-queue when they become
87 reachable. */
89 static void
90 enqueue_node (symtab_node *node, symtab_node **first,
91 hash_set<symtab_node *> *reachable)
93 /* Node is still in queue; do nothing. */
94 if (node->aux && node->aux != (void *) 2)
95 return;
96 /* Node was already processed as unreachable, re-enqueue
97 only if it became reachable now. */
98 if (node->aux == (void *)2 && !reachable->contains (node))
99 return;
100 node->aux = *first;
101 *first = node;
104 /* Return true if NODE may get inlined later.
105 This is used to keep DECL_EXTERNAL function bodies around long enough
106 so inliner can proces them. */
108 static bool
109 possible_inline_candidate_p (symtab_node *node)
111 if (symtab->state >= IPA_SSA_AFTER_INLINING)
112 return false;
113 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
114 if (!cnode)
115 return false;
116 if (DECL_UNINLINABLE (cnode->decl))
117 return false;
118 if (opt_for_fn (cnode->decl, optimize))
119 return true;
120 if (symtab->state >= IPA_SSA)
121 return false;
122 return lookup_attribute ("always_inline", DECL_ATTRIBUTES (node->decl));
125 /* Process references. */
127 static void
128 process_references (symtab_node *snode,
129 symtab_node **first,
130 hash_set<symtab_node *> *reachable)
132 int i;
133 struct ipa_ref *ref = NULL;
134 for (i = 0; snode->iterate_reference (i, ref); i++)
136 symtab_node *node = ref->referred;
137 symtab_node *body = node->ultimate_alias_target ();
139 if (node->definition && !node->in_other_partition
140 && ((!DECL_EXTERNAL (node->decl) || node->alias)
141 || (possible_inline_candidate_p (node)
142 /* We use variable constructors during late compilation for
143 constant folding. Keep references alive so partitioning
144 knows about potential references. */
145 || (VAR_P (node->decl)
146 && (flag_wpa
147 || flag_incremental_link
148 == INCREMENTAL_LINK_LTO)
149 && dyn_cast <varpool_node *> (node)
150 ->ctor_useable_for_folding_p ()))))
152 /* Be sure that we will not optimize out alias target
153 body. */
154 if (DECL_EXTERNAL (node->decl)
155 && node->alias
156 && symtab->state < IPA_SSA_AFTER_INLINING)
157 reachable->add (body);
158 reachable->add (node);
160 enqueue_node (node, first, reachable);
164 /* EDGE is an polymorphic call. If BEFORE_INLINING_P is set, mark
165 all its potential targets as reachable to permit later inlining if
166 devirtualization happens. After inlining still keep their declarations
167 around, so we can devirtualize to a direct call.
169 Also try to make trivial devirutalization when no or only one target is
170 possible. */
172 static void
173 walk_polymorphic_call_targets (hash_set<void *> *reachable_call_targets,
174 struct cgraph_edge *edge,
175 symtab_node **first,
176 hash_set<symtab_node *> *reachable)
178 unsigned int i;
179 void *cache_token;
180 bool final;
181 vec <cgraph_node *>targets
182 = possible_polymorphic_call_targets
183 (edge, &final, &cache_token);
185 if (!reachable_call_targets->add (cache_token))
187 for (i = 0; i < targets.length (); i++)
189 struct cgraph_node *n = targets[i];
191 /* Do not bother to mark virtual methods in anonymous namespace;
192 either we will find use of virtual table defining it, or it is
193 unused. */
194 if (TREE_CODE (TREE_TYPE (n->decl)) == METHOD_TYPE
195 && type_in_anonymous_namespace_p
196 (TYPE_METHOD_BASETYPE (TREE_TYPE (n->decl))))
197 continue;
199 n->indirect_call_target = true;
200 symtab_node *body = n->function_symbol ();
202 /* Prior inlining, keep alive bodies of possible targets for
203 devirtualization. */
204 if (n->definition
205 && (possible_inline_candidate_p (body)
206 && opt_for_fn (body->decl, flag_devirtualize)))
208 /* Be sure that we will not optimize out alias target
209 body. */
210 if (DECL_EXTERNAL (n->decl)
211 && n->alias
212 && symtab->state < IPA_SSA_AFTER_INLINING)
213 reachable->add (body);
214 reachable->add (n);
216 /* Even after inlining we want to keep the possible targets in the
217 boundary, so late passes can still produce direct call even if
218 the chance for inlining is lost. */
219 enqueue_node (n, first, reachable);
223 /* Very trivial devirtualization; when the type is
224 final or anonymous (so we know all its derivation)
225 and there is only one possible virtual call target,
226 make the edge direct. */
227 if (final)
229 if (targets.length () <= 1 && dbg_cnt (devirt))
231 cgraph_node *target, *node = edge->caller;
232 if (targets.length () == 1)
233 target = targets[0];
234 else
235 target = cgraph_node::get_create
236 (builtin_decl_implicit (BUILT_IN_UNREACHABLE));
238 if (dump_enabled_p ())
240 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, edge->call_stmt,
241 "devirtualizing call in %s to %s\n",
242 edge->caller->dump_name (),
243 target->dump_name ());
245 edge = cgraph_edge::make_direct (edge, target);
246 if (ipa_fn_summaries)
247 ipa_update_overall_fn_summary (node->inlined_to
248 ? node->inlined_to : node);
249 else if (edge->call_stmt)
250 cgraph_edge::redirect_call_stmt_to_callee (edge);
255 /* Perform reachability analysis and reclaim all unreachable nodes.
257 The algorithm is basically mark&sweep but with some extra refinements:
259 - reachable extern inline functions needs special handling; the bodies needs
260 to stay in memory until inlining in hope that they will be inlined.
261 After inlining we release their bodies and turn them into unanalyzed
262 nodes even when they are reachable.
264 - virtual functions are kept in callgraph even if they seem unreachable in
265 hope calls to them will be devirtualized.
267 Again we remove them after inlining. In late optimization some
268 devirtualization may happen, but it is not important since we won't inline
269 the call. In theory early opts and IPA should work out all important cases.
271 - virtual clones needs bodies of their origins for later materialization;
272 this means that we want to keep the body even if the origin is unreachable
273 otherwise. To avoid origin from sitting in the callgraph and being
274 walked by IPA passes, we turn them into unanalyzed nodes with body
275 defined.
277 We maintain set of function declaration where body needs to stay in
278 body_needed_for_clonning
280 Inline clones represent special case: their declaration match the
281 declaration of origin and cgraph_remove_node already knows how to
282 reshape callgraph and preserve body when offline copy of function or
283 inline clone is being removed.
285 - C++ virtual tables keyed to other unit are represented as DECL_EXTERNAL
286 variables with DECL_INITIAL set. We finalize these and keep reachable
287 ones around for constant folding purposes. After inlining we however
288 stop walking their references to let everything static referneced by them
289 to be removed when it is otherwise unreachable.
291 We maintain queue of both reachable symbols (i.e. defined symbols that needs
292 to stay) and symbols that are in boundary (i.e. external symbols referenced
293 by reachable symbols or origins of clones). The queue is represented
294 as linked list by AUX pointer terminated by 1.
296 At the end we keep all reachable symbols. For symbols in boundary we always
297 turn definition into a declaration, but we may keep function body around
298 based on body_needed_for_clonning
300 All symbols that enter the queue have AUX pointer non-zero and are in the
301 boundary. Pointer set REACHABLE is used to track reachable symbols.
303 Every symbol can be visited twice - once as part of boundary and once
304 as real reachable symbol. enqueue_node needs to decide whether the
305 node needs to be re-queued for second processing. For this purpose
306 we set AUX pointer of processed symbols in the boundary to constant 2. */
308 bool
309 symbol_table::remove_unreachable_nodes (FILE *file)
311 symtab_node *first = (symtab_node *) (void *) 1;
312 struct cgraph_node *node, *next;
313 varpool_node *vnode, *vnext;
314 bool changed = false;
315 hash_set<symtab_node *> reachable;
316 hash_set<tree> body_needed_for_clonning;
317 hash_set<void *> reachable_call_targets;
319 timevar_push (TV_IPA_UNREACHABLE);
320 build_type_inheritance_graph ();
321 if (file)
322 fprintf (file, "\nReclaiming functions:");
323 if (flag_checking)
325 FOR_EACH_FUNCTION (node)
326 gcc_assert (!node->aux);
327 FOR_EACH_VARIABLE (vnode)
328 gcc_assert (!vnode->aux);
330 /* Mark functions whose bodies are obviously needed.
331 This is mostly when they can be referenced externally. Inline clones
332 are special since their declarations are shared with master clone and thus
333 cgraph_can_remove_if_no_direct_calls_and_refs_p should not be called on them. */
334 FOR_EACH_FUNCTION (node)
336 node->used_as_abstract_origin = false;
337 node->indirect_call_target = false;
338 if (node->definition
339 && !node->inlined_to
340 && !node->in_other_partition
341 && !node->can_remove_if_no_direct_calls_and_refs_p ())
343 gcc_assert (!node->inlined_to);
344 reachable.add (node);
345 enqueue_node (node, &first, &reachable);
347 else
348 gcc_assert (!node->aux);
351 /* Mark variables that are obviously needed. */
352 FOR_EACH_DEFINED_VARIABLE (vnode)
353 if (!vnode->can_remove_if_no_refs_p()
354 && !vnode->in_other_partition)
356 reachable.add (vnode);
357 enqueue_node (vnode, &first, &reachable);
360 /* Perform reachability analysis. */
361 while (first != (symtab_node *) (void *) 1)
363 bool in_boundary_p = !reachable.contains (first);
364 symtab_node *node = first;
366 first = (symtab_node *)first->aux;
368 /* If we are processing symbol in boundary, mark its AUX pointer for
369 possible later re-processing in enqueue_node. */
370 if (in_boundary_p)
372 node->aux = (void *)2;
373 if (node->alias && node->analyzed)
374 enqueue_node (node->get_alias_target (), &first, &reachable);
376 else
378 if (TREE_CODE (node->decl) == FUNCTION_DECL
379 && DECL_ABSTRACT_ORIGIN (node->decl))
381 struct cgraph_node *origin_node
382 = cgraph_node::get (DECL_ABSTRACT_ORIGIN (node->decl));
383 if (origin_node && !origin_node->used_as_abstract_origin)
385 origin_node->used_as_abstract_origin = true;
386 gcc_assert (!origin_node->prev_sibling_clone);
387 gcc_assert (!origin_node->next_sibling_clone);
388 for (cgraph_node *n = origin_node->clones; n;
389 n = n->next_sibling_clone)
390 if (n->decl == DECL_ABSTRACT_ORIGIN (node->decl))
391 n->used_as_abstract_origin = true;
394 /* If any non-external and non-local symbol in a comdat group is
395 reachable, force all externally visible symbols in the same comdat
396 group to be reachable as well. Comdat-local symbols
397 can be discarded if all uses were inlined. */
398 if (node->same_comdat_group
399 && node->externally_visible
400 && !DECL_EXTERNAL (node->decl))
402 symtab_node *next;
403 for (next = node->same_comdat_group;
404 next != node;
405 next = next->same_comdat_group)
406 if (!next->comdat_local_p ()
407 && !DECL_EXTERNAL (next->decl)
408 && !reachable.add (next))
409 enqueue_node (next, &first, &reachable);
411 /* Mark references as reachable. */
412 process_references (node, &first, &reachable);
415 if (cgraph_node *cnode = dyn_cast <cgraph_node *> (node))
417 /* Mark the callees reachable unless they are direct calls to extern
418 inline functions we decided to not inline. */
419 if (!in_boundary_p)
421 struct cgraph_edge *e;
422 /* Keep alive possible targets for devirtualization. */
423 if (opt_for_fn (cnode->decl, optimize)
424 && opt_for_fn (cnode->decl, flag_devirtualize))
426 struct cgraph_edge *next;
427 for (e = cnode->indirect_calls; e; e = next)
429 next = e->next_callee;
430 if (e->indirect_info->polymorphic)
431 walk_polymorphic_call_targets (&reachable_call_targets,
432 e, &first, &reachable);
435 for (e = cnode->callees; e; e = e->next_callee)
437 symtab_node *body = e->callee->function_symbol ();
438 if (e->callee->definition
439 && !e->callee->in_other_partition
440 && (!e->inline_failed
441 || !DECL_EXTERNAL (e->callee->decl)
442 || e->callee->alias
443 || possible_inline_candidate_p (e->callee)))
445 /* Be sure that we will not optimize out alias target
446 body. */
447 if (DECL_EXTERNAL (e->callee->decl)
448 && e->callee->alias
449 && symtab->state < IPA_SSA_AFTER_INLINING)
450 reachable.add (body);
451 reachable.add (e->callee);
453 enqueue_node (e->callee, &first, &reachable);
456 /* When inline clone exists, mark body to be preserved so when removing
457 offline copy of the function we don't kill it. */
458 if (cnode->inlined_to)
459 body_needed_for_clonning.add (cnode->decl);
461 /* For non-inline clones, force their origins to the boundary and ensure
462 that body is not removed. */
463 while (cnode->clone_of)
465 bool noninline = cnode->clone_of->decl != cnode->decl;
466 cnode = cnode->clone_of;
467 if (noninline)
469 body_needed_for_clonning.add (cnode->decl);
470 enqueue_node (cnode, &first, &reachable);
475 else if (cnode->thunk.thunk_p)
476 enqueue_node (cnode->callees->callee, &first, &reachable);
478 /* If any reachable function has simd clones, mark them as
479 reachable as well. */
480 if (cnode->simd_clones)
482 cgraph_node *next;
483 for (next = cnode->simd_clones;
484 next;
485 next = next->simdclone->next_clone)
486 if (in_boundary_p
487 || !reachable.add (next))
488 enqueue_node (next, &first, &reachable);
491 /* When we see constructor of external variable, keep referred nodes in the
492 boundary. This will also hold initializers of the external vars NODE
493 refers to. */
494 varpool_node *vnode = dyn_cast <varpool_node *> (node);
495 if (vnode
496 && DECL_EXTERNAL (node->decl)
497 && !vnode->alias
498 && in_boundary_p)
500 struct ipa_ref *ref = NULL;
501 for (int i = 0; node->iterate_reference (i, ref); i++)
502 enqueue_node (ref->referred, &first, &reachable);
506 /* Remove unreachable functions. */
507 for (node = first_function (); node; node = next)
509 next = next_function (node);
511 /* If node is not needed at all, remove it. */
512 if (!node->aux)
514 if (file)
515 fprintf (file, " %s", node->dump_name ());
516 node->remove ();
517 changed = true;
519 /* If node is unreachable, remove its body. */
520 else if (!reachable.contains (node))
522 /* We keep definitions of thunks and aliases in the boundary so
523 we can walk to the ultimate alias targets and function symbols
524 reliably. */
525 if (node->alias || node->thunk.thunk_p)
527 else if (!body_needed_for_clonning.contains (node->decl))
529 /* Make the node a non-clone so that we do not attempt to
530 materialize it later. */
531 if (node->clone_of)
532 node->remove_from_clone_tree ();
533 node->release_body ();
535 else if (!node->clone_of)
536 gcc_assert (in_lto_p || DECL_RESULT (node->decl));
537 if (node->definition && !node->alias && !node->thunk.thunk_p)
539 if (file)
540 fprintf (file, " %s", node->dump_name ());
541 node->body_removed = true;
542 node->analyzed = false;
543 node->definition = false;
544 node->cpp_implicit_alias = false;
545 node->alias = false;
546 node->transparent_alias = false;
547 node->thunk.thunk_p = false;
548 node->weakref = false;
549 /* After early inlining we drop always_inline attributes on
550 bodies of functions that are still referenced (have their
551 address taken). */
552 DECL_ATTRIBUTES (node->decl)
553 = remove_attribute ("always_inline",
554 DECL_ATTRIBUTES (node->decl));
555 if (!node->in_other_partition)
556 node->local = false;
557 node->remove_callees ();
558 node->remove_all_references ();
559 changed = true;
562 else
563 gcc_assert (node->clone_of || !node->has_gimple_body_p ()
564 || in_lto_p || DECL_RESULT (node->decl));
567 /* Inline clones might be kept around so their materializing allows further
568 cloning. If the function the clone is inlined into is removed, we need
569 to turn it into normal cone. */
570 FOR_EACH_FUNCTION (node)
572 if (node->inlined_to
573 && !node->callers)
575 gcc_assert (node->clones);
576 node->inlined_to = NULL;
577 update_inlined_to_pointer (node, node);
579 node->aux = NULL;
582 /* Remove unreachable variables. */
583 if (file)
584 fprintf (file, "\nReclaiming variables:");
585 for (vnode = first_variable (); vnode; vnode = vnext)
587 vnext = next_variable (vnode);
588 if (!vnode->aux
589 /* For can_refer_decl_in_current_unit_p we want to track for
590 all external variables if they are defined in other partition
591 or not. */
592 && (!flag_ltrans || !DECL_EXTERNAL (vnode->decl)))
594 struct ipa_ref *ref = NULL;
596 /* First remove the aliases, so varpool::remove can possibly lookup
597 the constructor and save it for future use. */
598 while (vnode->iterate_direct_aliases (0, ref))
600 if (file)
601 fprintf (file, " %s", ref->referred->dump_name ());
602 ref->referring->remove ();
604 if (file)
605 fprintf (file, " %s", vnode->dump_name ());
606 vnext = next_variable (vnode);
607 /* Signal removal to the debug machinery. */
608 if (! flag_wpa || flag_incremental_link == INCREMENTAL_LINK_LTO)
610 vnode->definition = false;
611 (*debug_hooks->late_global_decl) (vnode->decl);
613 vnode->remove ();
614 changed = true;
616 else if (!reachable.contains (vnode) && !vnode->alias)
618 tree init;
619 if (vnode->definition)
621 if (file)
622 fprintf (file, " %s", vnode->dump_name ());
623 changed = true;
625 /* Keep body if it may be useful for constant folding. */
626 if ((flag_wpa || flag_incremental_link == INCREMENTAL_LINK_LTO)
627 || ((init = ctor_for_folding (vnode->decl)) == error_mark_node))
628 vnode->remove_initializer ();
629 else
630 DECL_INITIAL (vnode->decl) = init;
631 vnode->body_removed = true;
632 vnode->definition = false;
633 vnode->analyzed = false;
634 vnode->aux = NULL;
636 vnode->remove_from_same_comdat_group ();
638 vnode->remove_all_references ();
640 else
641 vnode->aux = NULL;
644 /* Now update address_taken flags and try to promote functions to be local. */
645 if (file)
646 fprintf (file, "\nClearing address taken flags:");
647 FOR_EACH_DEFINED_FUNCTION (node)
648 if (node->address_taken
649 && !node->used_from_other_partition)
651 if (!node->call_for_symbol_and_aliases
652 (has_addr_references_p, NULL, true))
654 if (file)
655 fprintf (file, " %s", node->dump_name ());
656 node->address_taken = false;
657 changed = true;
658 if (node->local_p ()
659 /* Virtual functions may be kept in cgraph just because
660 of possible later devirtualization. Do not mark them as
661 local too early so we won't optimize them out before
662 we are done with polymorphic call analysis. */
663 && (symtab->state >= IPA_SSA_AFTER_INLINING
664 || !node->call_for_symbol_and_aliases
665 (is_indirect_call_target_p, NULL, true)))
667 node->local = true;
668 if (file)
669 fprintf (file, " (local)");
673 if (file)
674 fprintf (file, "\n");
676 symtab_node::checking_verify_symtab_nodes ();
678 /* If we removed something, perhaps profile could be improved. */
679 if (changed && (optimize || in_lto_p) && ipa_call_summaries)
680 FOR_EACH_DEFINED_FUNCTION (node)
681 ipa_propagate_frequency (node);
683 timevar_pop (TV_IPA_UNREACHABLE);
684 return changed;
687 /* Process references to VNODE and set flags WRITTEN, ADDRESS_TAKEN, READ
688 as needed, also clear EXPLICIT_REFS if the references to given variable
689 do not need to be explicit. */
691 void
692 process_references (varpool_node *vnode,
693 bool *written, bool *address_taken,
694 bool *read, bool *explicit_refs)
696 int i;
697 struct ipa_ref *ref;
699 if (!vnode->all_refs_explicit_p ()
700 || TREE_THIS_VOLATILE (vnode->decl))
701 *explicit_refs = false;
703 for (i = 0; vnode->iterate_referring (i, ref)
704 && *explicit_refs && (!*written || !*address_taken || !*read); i++)
705 switch (ref->use)
707 case IPA_REF_ADDR:
708 *address_taken = true;
709 break;
710 case IPA_REF_LOAD:
711 *read = true;
712 break;
713 case IPA_REF_STORE:
714 *written = true;
715 break;
716 case IPA_REF_ALIAS:
717 process_references (dyn_cast<varpool_node *> (ref->referring), written,
718 address_taken, read, explicit_refs);
719 break;
723 /* Set TREE_READONLY bit. */
725 bool
726 set_readonly_bit (varpool_node *vnode, void *data ATTRIBUTE_UNUSED)
728 TREE_READONLY (vnode->decl) = true;
729 return false;
732 /* Set writeonly bit and clear the initalizer, since it will not be needed. */
734 bool
735 set_writeonly_bit (varpool_node *vnode, void *data)
737 vnode->writeonly = true;
738 if (optimize || in_lto_p)
740 DECL_INITIAL (vnode->decl) = NULL;
741 if (!vnode->alias)
743 if (vnode->num_references ())
744 *(bool *)data = true;
745 vnode->remove_all_references ();
748 return false;
751 /* Clear addressale bit of VNODE. */
753 bool
754 clear_addressable_bit (varpool_node *vnode, void *data ATTRIBUTE_UNUSED)
756 vnode->address_taken = false;
757 TREE_ADDRESSABLE (vnode->decl) = 0;
758 return false;
761 /* Discover variables that have no longer address taken, are read-only or
762 write-only and update their flags.
764 Return true when unreachable symbol removal should be done.
766 FIXME: This cannot be done in between gimplify and omp_expand since
767 readonly flag plays role on what is shared and what is not. Currently we do
768 this transformation as part of whole program visibility and re-do at
769 ipa-reference pass (to take into account clonning), but it would
770 make sense to do it before early optimizations. */
772 bool
773 ipa_discover_variable_flags (void)
775 if (!flag_ipa_reference_addressable)
776 return false;
778 bool remove_p = false;
779 varpool_node *vnode;
780 if (dump_file)
781 fprintf (dump_file, "Clearing variable flags:");
782 FOR_EACH_VARIABLE (vnode)
783 if (!vnode->alias
784 && (TREE_ADDRESSABLE (vnode->decl)
785 || !vnode->writeonly
786 || !TREE_READONLY (vnode->decl)))
788 bool written = false;
789 bool address_taken = false;
790 bool read = false;
791 bool explicit_refs = true;
793 process_references (vnode, &written, &address_taken, &read,
794 &explicit_refs);
795 if (!explicit_refs)
796 continue;
797 if (!address_taken)
799 if (TREE_ADDRESSABLE (vnode->decl) && dump_file)
800 fprintf (dump_file, " %s (non-addressable)",
801 vnode->dump_name ());
802 vnode->call_for_symbol_and_aliases (clear_addressable_bit, NULL,
803 true);
805 if (!address_taken && !written
806 /* Making variable in explicit section readonly can cause section
807 type conflict.
808 See e.g. gcc.c-torture/compile/pr23237.c */
809 && vnode->get_section () == NULL)
811 if (!TREE_READONLY (vnode->decl) && dump_file)
812 fprintf (dump_file, " %s (read-only)", vnode->dump_name ());
813 vnode->call_for_symbol_and_aliases (set_readonly_bit, NULL, true);
815 if (!vnode->writeonly && !read && !address_taken && written)
817 if (dump_file)
818 fprintf (dump_file, " %s (write-only)", vnode->dump_name ());
819 vnode->call_for_symbol_and_aliases (set_writeonly_bit, &remove_p,
820 true);
823 if (dump_file)
824 fprintf (dump_file, "\n");
825 return remove_p;
828 /* Generate and emit a static constructor or destructor. WHICH must
829 be one of 'I' (for a constructor), 'D' (for a destructor).
830 BODY is a STATEMENT_LIST containing GENERIC
831 statements. PRIORITY is the initialization priority for this
832 constructor or destructor.
834 FINAL specify whether the externally visible name for collect2 should
835 be produced. */
837 static void
838 cgraph_build_static_cdtor_1 (char which, tree body, int priority, bool final,
839 tree optimization,
840 tree target)
842 static int counter = 0;
843 char which_buf[16];
844 tree decl, name, resdecl;
846 /* The priority is encoded in the constructor or destructor name.
847 collect2 will sort the names and arrange that they are called at
848 program startup. */
849 if (!targetm.have_ctors_dtors && final)
851 sprintf (which_buf, "%c_%.5d_%d", which, priority, counter++);
852 name = get_file_function_name (which_buf);
854 else
856 /* Proudce sane name but one not recognizable by collect2, just for the
857 case we fail to inline the function. */
858 sprintf (which_buf, "_sub_%c_%.5d_%d", which, priority, counter++);
859 name = get_identifier (which_buf);
862 decl = build_decl (input_location, FUNCTION_DECL, name,
863 build_function_type_list (void_type_node, NULL_TREE));
864 current_function_decl = decl;
866 resdecl = build_decl (input_location,
867 RESULT_DECL, NULL_TREE, void_type_node);
868 DECL_ARTIFICIAL (resdecl) = 1;
869 DECL_RESULT (decl) = resdecl;
870 DECL_CONTEXT (resdecl) = decl;
872 allocate_struct_function (decl, false);
874 TREE_STATIC (decl) = 1;
875 TREE_USED (decl) = 1;
876 DECL_FUNCTION_SPECIFIC_OPTIMIZATION (decl) = optimization;
877 DECL_FUNCTION_SPECIFIC_TARGET (decl) = target;
878 DECL_ARTIFICIAL (decl) = 1;
879 DECL_IGNORED_P (decl) = 1;
880 DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT (decl) = 1;
881 DECL_SAVED_TREE (decl) = body;
882 if (!targetm.have_ctors_dtors && final)
884 TREE_PUBLIC (decl) = 1;
885 DECL_PRESERVE_P (decl) = 1;
887 DECL_UNINLINABLE (decl) = 1;
889 DECL_INITIAL (decl) = make_node (BLOCK);
890 BLOCK_SUPERCONTEXT (DECL_INITIAL (decl)) = decl;
891 TREE_USED (DECL_INITIAL (decl)) = 1;
893 DECL_SOURCE_LOCATION (decl) = input_location;
894 cfun->function_end_locus = input_location;
896 switch (which)
898 case 'I':
899 DECL_STATIC_CONSTRUCTOR (decl) = 1;
900 decl_init_priority_insert (decl, priority);
901 break;
902 case 'D':
903 DECL_STATIC_DESTRUCTOR (decl) = 1;
904 decl_fini_priority_insert (decl, priority);
905 break;
906 default:
907 gcc_unreachable ();
910 gimplify_function_tree (decl);
912 cgraph_node::add_new_function (decl, false);
914 set_cfun (NULL);
915 current_function_decl = NULL;
918 /* Generate and emit a static constructor or destructor. WHICH must
919 be one of 'I' (for a constructor) or 'D' (for a destructor).
920 BODY is a STATEMENT_LIST containing GENERIC
921 statements. PRIORITY is the initialization priority for this
922 constructor or destructor. */
924 void
925 cgraph_build_static_cdtor (char which, tree body, int priority)
927 /* FIXME: We should be able to
928 gcc_assert (!in_lto_p);
929 because at LTO time the global options are not safe to use.
930 Unfortunately ASAN finish_file will produce constructors late and they
931 may lead to surprises. */
932 cgraph_build_static_cdtor_1 (which, body, priority, false,
933 optimization_default_node,
934 target_option_default_node);
937 /* When target does not have ctors and dtors, we call all constructor
938 and destructor by special initialization/destruction function
939 recognized by collect2.
941 When we are going to build this function, collect all constructors and
942 destructors and turn them into normal functions. */
944 static void
945 record_cdtor_fn (struct cgraph_node *node, vec<tree> *ctors, vec<tree> *dtors)
947 if (DECL_STATIC_CONSTRUCTOR (node->decl))
948 ctors->safe_push (node->decl);
949 if (DECL_STATIC_DESTRUCTOR (node->decl))
950 dtors->safe_push (node->decl);
951 node = cgraph_node::get (node->decl);
952 DECL_DISREGARD_INLINE_LIMITS (node->decl) = 1;
955 /* Define global constructors/destructor functions for the CDTORS, of
956 which they are LEN. The CDTORS are sorted by initialization
957 priority. If CTOR_P is true, these are constructors; otherwise,
958 they are destructors. */
960 static void
961 build_cdtor (bool ctor_p, const vec<tree> &cdtors)
963 size_t i,j;
964 size_t len = cdtors.length ();
966 i = 0;
967 while (i < len)
969 tree body;
970 tree fn;
971 priority_type priority;
973 priority = 0;
974 body = NULL_TREE;
975 j = i;
978 priority_type p;
979 fn = cdtors[j];
980 p = ctor_p ? DECL_INIT_PRIORITY (fn) : DECL_FINI_PRIORITY (fn);
981 if (j == i)
982 priority = p;
983 else if (p != priority)
984 break;
985 j++;
987 while (j < len);
989 /* When there is only one cdtor and target supports them, do nothing. */
990 if (j == i + 1
991 && targetm.have_ctors_dtors)
993 i++;
994 continue;
996 /* Find the next batch of constructors/destructors with the same
997 initialization priority. */
998 for (;i < j; i++)
1000 tree call;
1001 fn = cdtors[i];
1002 call = build_call_expr (fn, 0);
1003 if (ctor_p)
1004 DECL_STATIC_CONSTRUCTOR (fn) = 0;
1005 else
1006 DECL_STATIC_DESTRUCTOR (fn) = 0;
1007 /* We do not want to optimize away pure/const calls here.
1008 When optimizing, these should be already removed, when not
1009 optimizing, we want user to be able to breakpoint in them. */
1010 TREE_SIDE_EFFECTS (call) = 1;
1011 append_to_statement_list (call, &body);
1013 gcc_assert (body != NULL_TREE);
1014 /* Generate a function to call all the function of like
1015 priority. */
1016 cgraph_build_static_cdtor_1 (ctor_p ? 'I' : 'D', body, priority, true,
1017 DECL_FUNCTION_SPECIFIC_OPTIMIZATION (cdtors[0]),
1018 DECL_FUNCTION_SPECIFIC_TARGET (cdtors[0]));
1022 /* Comparison function for qsort. P1 and P2 are actually of type
1023 "tree *" and point to static constructors. DECL_INIT_PRIORITY is
1024 used to determine the sort order. */
1026 static int
1027 compare_ctor (const void *p1, const void *p2)
1029 tree f1;
1030 tree f2;
1031 int priority1;
1032 int priority2;
1034 f1 = *(const tree *)p1;
1035 f2 = *(const tree *)p2;
1036 priority1 = DECL_INIT_PRIORITY (f1);
1037 priority2 = DECL_INIT_PRIORITY (f2);
1039 if (priority1 < priority2)
1040 return -1;
1041 else if (priority1 > priority2)
1042 return 1;
1043 else
1044 /* Ensure a stable sort. Constructors are executed in backwarding
1045 order to make LTO initialize braries first. */
1046 return DECL_UID (f2) - DECL_UID (f1);
1049 /* Comparison function for qsort. P1 and P2 are actually of type
1050 "tree *" and point to static destructors. DECL_FINI_PRIORITY is
1051 used to determine the sort order. */
1053 static int
1054 compare_dtor (const void *p1, const void *p2)
1056 tree f1;
1057 tree f2;
1058 int priority1;
1059 int priority2;
1061 f1 = *(const tree *)p1;
1062 f2 = *(const tree *)p2;
1063 priority1 = DECL_FINI_PRIORITY (f1);
1064 priority2 = DECL_FINI_PRIORITY (f2);
1066 if (priority1 < priority2)
1067 return -1;
1068 else if (priority1 > priority2)
1069 return 1;
1070 else
1071 /* Ensure a stable sort. */
1072 return DECL_UID (f1) - DECL_UID (f2);
1075 /* Generate functions to call static constructors and destructors
1076 for targets that do not support .ctors/.dtors sections. These
1077 functions have magic names which are detected by collect2. */
1079 static void
1080 build_cdtor_fns (vec<tree> *ctors, vec<tree> *dtors)
1082 if (!ctors->is_empty ())
1084 gcc_assert (!targetm.have_ctors_dtors || in_lto_p);
1085 ctors->qsort (compare_ctor);
1086 build_cdtor (/*ctor_p=*/true, *ctors);
1089 if (!dtors->is_empty ())
1091 gcc_assert (!targetm.have_ctors_dtors || in_lto_p);
1092 dtors->qsort (compare_dtor);
1093 build_cdtor (/*ctor_p=*/false, *dtors);
1097 /* Look for constructors and destructors and produce function calling them.
1098 This is needed for targets not supporting ctors or dtors, but we perform the
1099 transformation also at linktime to merge possibly numerous
1100 constructors/destructors into single function to improve code locality and
1101 reduce size. */
1103 static unsigned int
1104 ipa_cdtor_merge (void)
1106 /* A vector of FUNCTION_DECLs declared as static constructors. */
1107 auto_vec<tree, 20> ctors;
1108 /* A vector of FUNCTION_DECLs declared as static destructors. */
1109 auto_vec<tree, 20> dtors;
1110 struct cgraph_node *node;
1111 FOR_EACH_DEFINED_FUNCTION (node)
1112 if (DECL_STATIC_CONSTRUCTOR (node->decl)
1113 || DECL_STATIC_DESTRUCTOR (node->decl))
1114 record_cdtor_fn (node, &ctors, &dtors);
1115 build_cdtor_fns (&ctors, &dtors);
1116 return 0;
1119 namespace {
1121 const pass_data pass_data_ipa_cdtor_merge =
1123 IPA_PASS, /* type */
1124 "cdtor", /* name */
1125 OPTGROUP_NONE, /* optinfo_flags */
1126 TV_CGRAPHOPT, /* tv_id */
1127 0, /* properties_required */
1128 0, /* properties_provided */
1129 0, /* properties_destroyed */
1130 0, /* todo_flags_start */
1131 0, /* todo_flags_finish */
1134 class pass_ipa_cdtor_merge : public ipa_opt_pass_d
1136 public:
1137 pass_ipa_cdtor_merge (gcc::context *ctxt)
1138 : ipa_opt_pass_d (pass_data_ipa_cdtor_merge, ctxt,
1139 NULL, /* generate_summary */
1140 NULL, /* write_summary */
1141 NULL, /* read_summary */
1142 NULL, /* write_optimization_summary */
1143 NULL, /* read_optimization_summary */
1144 NULL, /* stmt_fixup */
1145 0, /* function_transform_todo_flags_start */
1146 NULL, /* function_transform */
1147 NULL) /* variable_transform */
1150 /* opt_pass methods: */
1151 virtual bool gate (function *);
1152 virtual unsigned int execute (function *) { return ipa_cdtor_merge (); }
1154 }; // class pass_ipa_cdtor_merge
1156 bool
1157 pass_ipa_cdtor_merge::gate (function *)
1159 /* Perform the pass when we have no ctors/dtors support
1160 or at LTO time to merge multiple constructors into single
1161 function. */
1162 return !targetm.have_ctors_dtors || in_lto_p;
1165 } // anon namespace
1167 ipa_opt_pass_d *
1168 make_pass_ipa_cdtor_merge (gcc::context *ctxt)
1170 return new pass_ipa_cdtor_merge (ctxt);
1173 /* Invalid pointer representing BOTTOM for single user dataflow. */
1174 #define BOTTOM ((cgraph_node *)(size_t) 2)
1176 /* Meet operation for single user dataflow.
1177 Here we want to associate variables with sigle function that may access it.
1179 FUNCTION is current single user of a variable, VAR is variable that uses it.
1180 Latttice is stored in SINGLE_USER_MAP.
1182 We represent:
1183 - TOP by no entry in SIGNLE_USER_MAP
1184 - BOTTOM by BOTTOM in AUX pointer (to save lookups)
1185 - known single user by cgraph pointer in SINGLE_USER_MAP. */
1187 cgraph_node *
1188 meet (cgraph_node *function, varpool_node *var,
1189 hash_map<varpool_node *, cgraph_node *> &single_user_map)
1191 struct cgraph_node *user, **f;
1193 if (var->aux == BOTTOM)
1194 return BOTTOM;
1196 f = single_user_map.get (var);
1197 if (!f)
1198 return function;
1199 user = *f;
1200 if (!function)
1201 return user;
1202 else if (function != user)
1203 return BOTTOM;
1204 else
1205 return function;
1208 /* Propagation step of single-use dataflow.
1210 Check all uses of VNODE and see if they are used by single function FUNCTION.
1211 SINGLE_USER_MAP represents the dataflow lattice. */
1213 cgraph_node *
1214 propagate_single_user (varpool_node *vnode, cgraph_node *function,
1215 hash_map<varpool_node *, cgraph_node *> &single_user_map)
1217 int i;
1218 struct ipa_ref *ref;
1220 gcc_assert (!vnode->externally_visible);
1222 /* If node is an alias, first meet with its target. */
1223 if (vnode->alias)
1224 function = meet (function, vnode->get_alias_target (), single_user_map);
1226 /* Check all users and see if they correspond to a single function. */
1227 for (i = 0; vnode->iterate_referring (i, ref) && function != BOTTOM; i++)
1229 struct cgraph_node *cnode = dyn_cast <cgraph_node *> (ref->referring);
1230 if (cnode)
1232 if (cnode->inlined_to)
1233 cnode = cnode->inlined_to;
1234 if (!function)
1235 function = cnode;
1236 else if (function != cnode)
1237 function = BOTTOM;
1239 else
1240 function = meet (function, dyn_cast <varpool_node *> (ref->referring),
1241 single_user_map);
1243 return function;
1246 /* Pass setting used_by_single_function flag.
1247 This flag is set on variable when there is only one function that may
1248 possibly referr to it. */
1250 static unsigned int
1251 ipa_single_use (void)
1253 varpool_node *first = (varpool_node *) (void *) 1;
1254 varpool_node *var;
1255 hash_map<varpool_node *, cgraph_node *> single_user_map;
1257 FOR_EACH_DEFINED_VARIABLE (var)
1258 if (!var->all_refs_explicit_p ())
1259 var->aux = BOTTOM;
1260 else
1262 /* Enqueue symbol for dataflow. */
1263 var->aux = first;
1264 first = var;
1267 /* The actual dataflow. */
1269 while (first != (void *) 1)
1271 cgraph_node *user, *orig_user, **f;
1273 var = first;
1274 first = (varpool_node *)first->aux;
1276 f = single_user_map.get (var);
1277 if (f)
1278 orig_user = *f;
1279 else
1280 orig_user = NULL;
1281 user = propagate_single_user (var, orig_user, single_user_map);
1283 gcc_checking_assert (var->aux != BOTTOM);
1285 /* If user differs, enqueue all references. */
1286 if (user != orig_user)
1288 unsigned int i;
1289 ipa_ref *ref;
1291 single_user_map.put (var, user);
1293 /* Enqueue all aliases for re-processing. */
1294 for (i = 0; var->iterate_direct_aliases (i, ref); i++)
1295 if (!ref->referring->aux)
1297 ref->referring->aux = first;
1298 first = dyn_cast <varpool_node *> (ref->referring);
1300 /* Enqueue all users for re-processing. */
1301 for (i = 0; var->iterate_reference (i, ref); i++)
1302 if (!ref->referred->aux
1303 && ref->referred->definition
1304 && is_a <varpool_node *> (ref->referred))
1306 ref->referred->aux = first;
1307 first = dyn_cast <varpool_node *> (ref->referred);
1310 /* If user is BOTTOM, just punt on this var. */
1311 if (user == BOTTOM)
1312 var->aux = BOTTOM;
1313 else
1314 var->aux = NULL;
1316 else
1317 var->aux = NULL;
1320 FOR_EACH_DEFINED_VARIABLE (var)
1322 if (var->aux != BOTTOM)
1324 /* Not having the single user known means that the VAR is
1325 unreachable. Either someone forgot to remove unreachable
1326 variables or the reachability here is wrong. */
1328 gcc_checking_assert (single_user_map.get (var));
1330 if (dump_file)
1332 fprintf (dump_file, "Variable %s is used by single function\n",
1333 var->dump_name ());
1335 var->used_by_single_function = true;
1337 var->aux = NULL;
1339 return 0;
1342 namespace {
1344 const pass_data pass_data_ipa_single_use =
1346 IPA_PASS, /* type */
1347 "single-use", /* name */
1348 OPTGROUP_NONE, /* optinfo_flags */
1349 TV_CGRAPHOPT, /* tv_id */
1350 0, /* properties_required */
1351 0, /* properties_provided */
1352 0, /* properties_destroyed */
1353 0, /* todo_flags_start */
1354 0, /* todo_flags_finish */
1357 class pass_ipa_single_use : public ipa_opt_pass_d
1359 public:
1360 pass_ipa_single_use (gcc::context *ctxt)
1361 : ipa_opt_pass_d (pass_data_ipa_single_use, ctxt,
1362 NULL, /* generate_summary */
1363 NULL, /* write_summary */
1364 NULL, /* read_summary */
1365 NULL, /* write_optimization_summary */
1366 NULL, /* read_optimization_summary */
1367 NULL, /* stmt_fixup */
1368 0, /* function_transform_todo_flags_start */
1369 NULL, /* function_transform */
1370 NULL) /* variable_transform */
1373 /* opt_pass methods: */
1374 virtual unsigned int execute (function *) { return ipa_single_use (); }
1376 }; // class pass_ipa_single_use
1378 } // anon namespace
1380 ipa_opt_pass_d *
1381 make_pass_ipa_single_use (gcc::context *ctxt)
1383 return new pass_ipa_single_use (ctxt);
1386 /* Materialize all clones. */
1388 namespace {
1390 const pass_data pass_data_materialize_all_clones =
1392 SIMPLE_IPA_PASS, /* type */
1393 "materialize-all-clones", /* name */
1394 OPTGROUP_NONE, /* optinfo_flags */
1395 TV_IPA_OPT, /* tv_id */
1396 0, /* properties_required */
1397 0, /* properties_provided */
1398 0, /* properties_destroyed */
1399 0, /* todo_flags_start */
1400 0, /* todo_flags_finish */
1403 class pass_materialize_all_clones : public simple_ipa_opt_pass
1405 public:
1406 pass_materialize_all_clones (gcc::context *ctxt)
1407 : simple_ipa_opt_pass (pass_data_materialize_all_clones, ctxt)
1410 /* opt_pass methods: */
1411 virtual unsigned int execute (function *)
1413 symtab->materialize_all_clones ();
1414 return 0;
1417 }; // class pass_materialize_all_clones
1419 } // anon namespace
1421 simple_ipa_opt_pass *
1422 make_pass_materialize_all_clones (gcc::context *ctxt)
1424 return new pass_materialize_all_clones (ctxt);