1 /* Driver of optimization process
2 Copyright (C) 2003-2022 Free Software Foundation, Inc.
3 Contributed by Jan Hubicka
4
5 This file is part of GCC.
6
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.
11
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.
16
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/>. */
20
21 /* This module implements main driver of compilation process.
22
23 The main scope of this file is to act as an interface in between
24 tree based frontends and the backend.
25
26 The front-end is supposed to use following functionality:
27
28 - finalize_function
29
30 This function is called once front-end has parsed whole body of function
31 and it is certain that the function body nor the declaration will change.
32
33 (There is one exception needed for implementing GCC extern inline
34 function.)
35
36 - varpool_finalize_decl
37
38 This function has same behavior as the above but is used for static
39 variables.
40
41 - add_asm_node
42
43 Insert new toplevel ASM statement
44
45 - finalize_compilation_unit
46
47 This function is called once (source level) compilation unit is finalized
48 and it will no longer change.
49
50 The symbol table is constructed starting from the trivially needed
51 symbols finalized by the frontend. Functions are lowered into
52 GIMPLE representation and callgraph/reference lists are constructed.
53 Those are used to discover other necessary functions and variables.
54
55 At the end the bodies of unreachable functions are removed.
56
57 The function can be called multiple times when multiple source level
58 compilation units are combined.
59
60 - compile
61
62 This passes control to the back-end. Optimizations are performed and
63 final assembler is generated. This is done in the following way. Note
64 that with link time optimization the process is split into three
65 stages (compile time, linktime analysis and parallel linktime as
66 indicated bellow).
67
68 Compile time:
69
70 1) Inter-procedural optimization.
71 (ipa_passes)
72
73 This part is further split into:
74
75 a) early optimizations. These are local passes executed in
76 the topological order on the callgraph.
77
78 The purpose of early optimizations is to optimize away simple
79 things that may otherwise confuse IP analysis. Very simple
80 propagation across the callgraph is done i.e. to discover
81 functions without side effects and simple inlining is performed.
82
83 b) early small interprocedural passes.
84
85 Those are interprocedural passes executed only at compilation
86 time. These include, for example, transactional memory lowering,
87 unreachable code removal and other simple transformations.
88
89 c) IP analysis stage. All interprocedural passes do their
90 analysis.
91
92 Interprocedural passes differ from small interprocedural
93 passes by their ability to operate across whole program
94 at linktime. Their analysis stage is performed early to
95 both reduce linking times and linktime memory usage by
96 not having to represent whole program in memory.
97
98 d) LTO streaming. When doing LTO, everything important gets
99 streamed into the object file.
100
101 Compile time and or linktime analysis stage (WPA):
102
103 At linktime units gets streamed back and symbol table is
104 merged. Function bodies are not streamed in and not
105 available.
106 e) IP propagation stage. All IP passes execute their
107 IP propagation. This is done based on the earlier analysis
108 without having function bodies at hand.
109 f) Ltrans streaming. When doing WHOPR LTO, the program
110 is partitioned and streamed into multiple object files.
111
112 Compile time and/or parallel linktime stage (ltrans)
113
114 Each of the object files is streamed back and compiled
115 separately. Now the function bodies becomes available
116 again.
117
118 2) Virtual clone materialization
119 (cgraph_materialize_clone)
120
121 IP passes can produce copies of existing functions (such
122 as versioned clones or inline clones) without actually
123 manipulating their bodies by creating virtual clones in
124 the callgraph. At this time the virtual clones are
125 turned into real functions
126 3) IP transformation
127
128 All IP passes transform function bodies based on earlier
129 decision of the IP propagation.
130
131 4) late small IP passes
132
133 Simple IP passes working within single program partition.
134
135 5) Expansion
136 (expand_all_functions)
137
138 At this stage functions that needs to be output into
139 assembler are identified and compiled in topological order
140 6) Output of variables and aliases
141 Now it is known what variable references was not optimized
142 out and thus all variables are output to the file.
143
144 Note that with -fno-toplevel-reorder passes 5 and 6
145 are combined together in cgraph_output_in_order.
146
147 Finally there are functions to manipulate the callgraph from
148 backend.
149 - cgraph_add_new_function is used to add backend produced
150 functions introduced after the unit is finalized.
151 The functions are enqueue for later processing and inserted
152 into callgraph with cgraph_process_new_functions.
153
154 - cgraph_function_versioning
155
156 produces a copy of function into new one (a version)
157 and apply simple transformations
158 */
159
160 #include "config.h"
161 #include "system.h"
162 #include "coretypes.h"
163 #include "backend.h"
164 #include "target.h"
165 #include "rtl.h"
166 #include "tree.h"
167 #include "gimple.h"
168 #include "cfghooks.h"
169 #include "regset.h" /* FIXME: For reg_obstack. */
170 #include "alloc-pool.h"
171 #include "tree-pass.h"
172 #include "stringpool.h"
173 #include "gimple-ssa.h"
174 #include "cgraph.h"
175 #include "coverage.h"
176 #include "lto-streamer.h"
177 #include "fold-const.h"
178 #include "varasm.h"
179 #include "stor-layout.h"
180 #include "output.h"
181 #include "cfgcleanup.h"
182 #include "gimple-fold.h"
183 #include "gimplify.h"
184 #include "gimple-iterator.h"
185 #include "gimplify-me.h"
186 #include "tree-cfg.h"
187 #include "tree-into-ssa.h"
188 #include "tree-ssa.h"
189 #include "langhooks.h"
190 #include "toplev.h"
191 #include "debug.h"
192 #include "symbol-summary.h"
193 #include "tree-vrp.h"
194 #include "ipa-prop.h"
195 #include "gimple-pretty-print.h"
196 #include "plugin.h"
197 #include "ipa-fnsummary.h"
198 #include "ipa-utils.h"
199 #include "except.h"
200 #include "cfgloop.h"
201 #include "context.h"
202 #include "pass_manager.h"
203 #include "tree-nested.h"
204 #include "dbgcnt.h"
205 #include "lto-section-names.h"
206 #include "stringpool.h"
207 #include "attribs.h"
208 #include "ipa-inline.h"
209 #include "omp-offload.h"
210 #include "symtab-thunks.h"
211
212 /* Queue of cgraph nodes scheduled to be added into cgraph. This is a
213 secondary queue used during optimization to accommodate passes that
214 may generate new functions that need to be optimized and expanded. */
215 vec<cgraph_node *> cgraph_new_nodes;
216
217 static void expand_all_functions (void);
218 static void mark_functions_to_output (void);
219 static void handle_alias_pairs (void);
220
221 /* Return true if this symbol is a function from the C frontend specified
222 directly in RTL form (with "__RTL"). */
223
224 bool
native_rtl_p() const225 symtab_node::native_rtl_p () const
226 {
227 if (TREE_CODE (decl) != FUNCTION_DECL)
228 return false;
229 if (!DECL_STRUCT_FUNCTION (decl))
230 return false;
231 return DECL_STRUCT_FUNCTION (decl)->curr_properties & PROP_rtl;
232 }
233
234 /* Determine if symbol declaration is needed. That is, visible to something
235 either outside this translation unit, something magic in the system
236 configury */
237 bool
needed_p(void)238 symtab_node::needed_p (void)
239 {
240 /* Double check that no one output the function into assembly file
241 early. */
242 if (!native_rtl_p ())
243 gcc_checking_assert
244 (!DECL_ASSEMBLER_NAME_SET_P (decl)
245 || !TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)));
246
247 if (!definition)
248 return false;
249
250 if (DECL_EXTERNAL (decl))
251 return false;
252
253 /* If the user told us it is used, then it must be so. */
254 if (force_output)
255 return true;
256
257 /* ABI forced symbols are needed when they are external. */
258 if (forced_by_abi && TREE_PUBLIC (decl))
259 return true;
260
261 /* Keep constructors, destructors and virtual functions. */
262 if (TREE_CODE (decl) == FUNCTION_DECL
263 && (DECL_STATIC_CONSTRUCTOR (decl) || DECL_STATIC_DESTRUCTOR (decl)))
264 return true;
265
266 /* Externally visible variables must be output. The exception is
267 COMDAT variables that must be output only when they are needed. */
268 if (TREE_PUBLIC (decl) && !DECL_COMDAT (decl))
269 return true;
270
271 return false;
272 }
273
274 /* Head and terminator of the queue of nodes to be processed while building
275 callgraph. */
276
277 static symtab_node symtab_terminator (SYMTAB_SYMBOL);
278 static symtab_node *queued_nodes = &symtab_terminator;
279
280 /* Add NODE to queue starting at QUEUED_NODES.
281 The queue is linked via AUX pointers and terminated by pointer to 1. */
282
283 static void
enqueue_node(symtab_node * node)284 enqueue_node (symtab_node *node)
285 {
286 if (node->aux)
287 return;
288 gcc_checking_assert (queued_nodes);
289 node->aux = queued_nodes;
290 queued_nodes = node;
291 }
292
293 /* Process CGRAPH_NEW_FUNCTIONS and perform actions necessary to add these
294 functions into callgraph in a way so they look like ordinary reachable
295 functions inserted into callgraph already at construction time. */
296
297 void
process_new_functions(void)298 symbol_table::process_new_functions (void)
299 {
300 tree fndecl;
301
302 if (!cgraph_new_nodes.exists ())
303 return;
304
305 handle_alias_pairs ();
306 /* Note that this queue may grow as its being processed, as the new
307 functions may generate new ones. */
308 for (unsigned i = 0; i < cgraph_new_nodes.length (); i++)
309 {
310 cgraph_node *node = cgraph_new_nodes[i];
311 fndecl = node->decl;
312 switch (state)
313 {
314 case CONSTRUCTION:
315 /* At construction time we just need to finalize function and move
316 it into reachable functions list. */
317
318 cgraph_node::finalize_function (fndecl, false);
319 call_cgraph_insertion_hooks (node);
320 enqueue_node (node);
321 break;
322
323 case IPA:
324 case IPA_SSA:
325 case IPA_SSA_AFTER_INLINING:
326 /* When IPA optimization already started, do all essential
327 transformations that has been already performed on the whole
328 cgraph but not on this function. */
329
330 gimple_register_cfg_hooks ();
331 if (!node->analyzed)
332 node->analyze ();
333 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
334 if ((state == IPA_SSA || state == IPA_SSA_AFTER_INLINING)
335 && !gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl)))
336 {
337 bool summaried_computed = ipa_fn_summaries != NULL;
338 g->get_passes ()->execute_early_local_passes ();
339 /* Early passes compute inline parameters to do inlining
340 and splitting. This is redundant for functions added late.
341 Just throw away whatever it did. */
342 if (!summaried_computed)
343 {
344 ipa_free_fn_summary ();
345 ipa_free_size_summary ();
346 }
347 }
348 else if (ipa_fn_summaries != NULL)
349 compute_fn_summary (node, true);
350 free_dominance_info (CDI_POST_DOMINATORS);
351 free_dominance_info (CDI_DOMINATORS);
352 pop_cfun ();
353 call_cgraph_insertion_hooks (node);
354 break;
355
356 case EXPANSION:
357 /* Functions created during expansion shall be compiled
358 directly. */
359 node->process = 0;
360 call_cgraph_insertion_hooks (node);
361 node->expand ();
362 break;
363
364 default:
365 gcc_unreachable ();
366 break;
367 }
368 }
369
370 cgraph_new_nodes.release ();
371 }
372
373 /* As an GCC extension we allow redefinition of the function. The
374 semantics when both copies of bodies differ is not well defined.
375 We replace the old body with new body so in unit at a time mode
376 we always use new body, while in normal mode we may end up with
377 old body inlined into some functions and new body expanded and
378 inlined in others.
379
380 ??? It may make more sense to use one body for inlining and other
381 body for expanding the function but this is difficult to do. */
382
383 void
reset(void)384 cgraph_node::reset (void)
385 {
386 /* If process is set, then we have already begun whole-unit analysis.
387 This is *not* testing for whether we've already emitted the function.
388 That case can be sort-of legitimately seen with real function redefinition
389 errors. I would argue that the front end should never present us with
390 such a case, but don't enforce that for now. */
391 gcc_assert (!process);
392
393 /* Reset our data structures so we can analyze the function again. */
394 inlined_to = NULL;
395 memset (&rtl, 0, sizeof (rtl));
396 analyzed = false;
397 definition = false;
398 alias = false;
399 transparent_alias = false;
400 weakref = false;
401 cpp_implicit_alias = false;
402
403 remove_callees ();
404 remove_all_references ();
405 }
406
407 /* Return true when there are references to the node. INCLUDE_SELF is
408 true if a self reference counts as a reference. */
409
410 bool
referred_to_p(bool include_self)411 symtab_node::referred_to_p (bool include_self)
412 {
413 ipa_ref *ref = NULL;
414
415 /* See if there are any references at all. */
416 if (iterate_referring (0, ref))
417 return true;
418 /* For functions check also calls. */
419 cgraph_node *cn = dyn_cast <cgraph_node *> (this);
420 if (cn && cn->callers)
421 {
422 if (include_self)
423 return true;
424 for (cgraph_edge *e = cn->callers; e; e = e->next_caller)
425 if (e->caller != this)
426 return true;
427 }
428 return false;
429 }
430
431 /* DECL has been parsed. Take it, queue it, compile it at the whim of the
432 logic in effect. If NO_COLLECT is true, then our caller cannot stand to have
433 the garbage collector run at the moment. We would need to either create
434 a new GC context, or just not compile right now. */
435
436 void
finalize_function(tree decl,bool no_collect)437 cgraph_node::finalize_function (tree decl, bool no_collect)
438 {
439 cgraph_node *node = cgraph_node::get_create (decl);
440
441 if (node->definition)
442 {
443 /* Nested functions should only be defined once. */
444 gcc_assert (!DECL_CONTEXT (decl)
445 || TREE_CODE (DECL_CONTEXT (decl)) != FUNCTION_DECL);
446 node->reset ();
447 node->redefined_extern_inline = true;
448 }
449
450 /* Set definition first before calling notice_global_symbol so that
451 it is available to notice_global_symbol. */
452 node->definition = true;
453 notice_global_symbol (decl);
454 node->lowered = DECL_STRUCT_FUNCTION (decl)->cfg != NULL;
455 node->semantic_interposition = opt_for_fn (decl, flag_semantic_interposition);
456 if (!flag_toplevel_reorder)
457 node->no_reorder = true;
458
459 /* With -fkeep-inline-functions we are keeping all inline functions except
460 for extern inline ones. */
461 if (flag_keep_inline_functions
462 && DECL_DECLARED_INLINE_P (decl)
463 && !DECL_EXTERNAL (decl)
464 && !DECL_DISREGARD_INLINE_LIMITS (decl))
465 node->force_output = 1;
466
467 /* __RTL functions were already output as soon as they were parsed (due
468 to the large amount of global state in the backend).
469 Mark such functions as "force_output" to reflect the fact that they
470 will be in the asm file when considering the symbols they reference.
471 The attempt to output them later on will bail out immediately. */
472 if (node->native_rtl_p ())
473 node->force_output = 1;
474
475 /* When not optimizing, also output the static functions. (see
476 PR24561), but don't do so for always_inline functions, functions
477 declared inline and nested functions. These were optimized out
478 in the original implementation and it is unclear whether we want
479 to change the behavior here. */
480 if (((!opt_for_fn (decl, optimize) || flag_keep_static_functions
481 || node->no_reorder)
482 && !node->cpp_implicit_alias
483 && !DECL_DISREGARD_INLINE_LIMITS (decl)
484 && !DECL_DECLARED_INLINE_P (decl)
485 && !(DECL_CONTEXT (decl)
486 && TREE_CODE (DECL_CONTEXT (decl)) == FUNCTION_DECL))
487 && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
488 node->force_output = 1;
489
490 /* If we've not yet emitted decl, tell the debug info about it. */
491 if (!TREE_ASM_WRITTEN (decl))
492 (*debug_hooks->deferred_inline_function) (decl);
493
494 if (!no_collect)
495 ggc_collect ();
496
497 if (symtab->state == CONSTRUCTION
498 && (node->needed_p () || node->referred_to_p ()))
499 enqueue_node (node);
500 }
501
502 /* Add the function FNDECL to the call graph.
503 Unlike finalize_function, this function is intended to be used
504 by middle end and allows insertion of new function at arbitrary point
505 of compilation. The function can be either in high, low or SSA form
506 GIMPLE.
507
508 The function is assumed to be reachable and have address taken (so no
509 API breaking optimizations are performed on it).
510
511 Main work done by this function is to enqueue the function for later
512 processing to avoid need the passes to be re-entrant. */
513
514 void
add_new_function(tree fndecl,bool lowered)515 cgraph_node::add_new_function (tree fndecl, bool lowered)
516 {
517 gcc::pass_manager *passes = g->get_passes ();
518 cgraph_node *node;
519
520 if (dump_file)
521 {
522 struct function *fn = DECL_STRUCT_FUNCTION (fndecl);
523 const char *function_type = ((gimple_has_body_p (fndecl))
524 ? (lowered
525 ? (gimple_in_ssa_p (fn)
526 ? "ssa gimple"
527 : "low gimple")
528 : "high gimple")
529 : "to-be-gimplified");
530 fprintf (dump_file,
531 "Added new %s function %s to callgraph\n",
532 function_type,
533 fndecl_name (fndecl));
534 }
535
536 switch (symtab->state)
537 {
538 case PARSING:
539 cgraph_node::finalize_function (fndecl, false);
540 break;
541 case CONSTRUCTION:
542 /* Just enqueue function to be processed at nearest occurrence. */
543 node = cgraph_node::get_create (fndecl);
544 if (lowered)
545 node->lowered = true;
546 cgraph_new_nodes.safe_push (node);
547 break;
548
549 case IPA:
550 case IPA_SSA:
551 case IPA_SSA_AFTER_INLINING:
552 case EXPANSION:
553 /* Bring the function into finalized state and enqueue for later
554 analyzing and compilation. */
555 node = cgraph_node::get_create (fndecl);
556 node->local = false;
557 node->definition = true;
558 node->semantic_interposition = opt_for_fn (fndecl,
559 flag_semantic_interposition);
560 node->force_output = true;
561 if (TREE_PUBLIC (fndecl))
562 node->externally_visible = true;
563 if (!lowered && symtab->state == EXPANSION)
564 {
565 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
566 gimple_register_cfg_hooks ();
567 bitmap_obstack_initialize (NULL);
568 execute_pass_list (cfun, passes->all_lowering_passes);
569 passes->execute_early_local_passes ();
570 bitmap_obstack_release (NULL);
571 pop_cfun ();
572
573 lowered = true;
574 }
575 if (lowered)
576 node->lowered = true;
577 cgraph_new_nodes.safe_push (node);
578 break;
579
580 case FINISHED:
581 /* At the very end of compilation we have to do all the work up
582 to expansion. */
583 node = cgraph_node::create (fndecl);
584 if (lowered)
585 node->lowered = true;
586 node->definition = true;
587 node->semantic_interposition = opt_for_fn (fndecl,
588 flag_semantic_interposition);
589 node->analyze ();
590 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
591 gimple_register_cfg_hooks ();
592 bitmap_obstack_initialize (NULL);
593 if (!gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl)))
594 g->get_passes ()->execute_early_local_passes ();
595 bitmap_obstack_release (NULL);
596 pop_cfun ();
597 node->expand ();
598 break;
599
600 default:
601 gcc_unreachable ();
602 }
603
604 /* Set a personality if required and we already passed EH lowering. */
605 if (lowered
606 && (function_needs_eh_personality (DECL_STRUCT_FUNCTION (fndecl))
607 == eh_personality_lang))
608 DECL_FUNCTION_PERSONALITY (fndecl) = lang_hooks.eh_personality ();
609 }
610
611 /* Analyze the function scheduled to be output. */
612 void
analyze(void)613 cgraph_node::analyze (void)
614 {
615 if (native_rtl_p ())
616 {
617 analyzed = true;
618 return;
619 }
620
621 tree decl = this->decl;
622 location_t saved_loc = input_location;
623 input_location = DECL_SOURCE_LOCATION (decl);
624 semantic_interposition = opt_for_fn (decl, flag_semantic_interposition);
625
626 if (thunk)
627 {
628 thunk_info *info = thunk_info::get (this);
629 cgraph_node *t = cgraph_node::get (info->alias);
630
631 create_edge (t, NULL, t->count);
632 callees->can_throw_external = !TREE_NOTHROW (t->decl);
633 /* Target code in expand_thunk may need the thunk's target
634 to be analyzed, so recurse here. */
635 if (!t->analyzed && t->definition)
636 t->analyze ();
637 if (t->alias)
638 {
639 t = t->get_alias_target ();
640 if (!t->analyzed && t->definition)
641 t->analyze ();
642 }
643 bool ret = expand_thunk (this, false, false);
644 thunk_info::get (this)->alias = NULL;
645 if (!ret)
646 return;
647 }
648 if (alias)
649 resolve_alias (cgraph_node::get (alias_target), transparent_alias);
650 else if (dispatcher_function)
651 {
652 /* Generate the dispatcher body of multi-versioned functions. */
653 cgraph_function_version_info *dispatcher_version_info
654 = function_version ();
655 if (dispatcher_version_info != NULL
656 && (dispatcher_version_info->dispatcher_resolver
657 == NULL_TREE))
658 {
659 tree resolver = NULL_TREE;
660 gcc_assert (targetm.generate_version_dispatcher_body);
661 resolver = targetm.generate_version_dispatcher_body (this);
662 gcc_assert (resolver != NULL_TREE);
663 }
664 }
665 else
666 {
667 push_cfun (DECL_STRUCT_FUNCTION (decl));
668
669 assign_assembler_name_if_needed (decl);
670
671 /* Make sure to gimplify bodies only once. During analyzing a
672 function we lower it, which will require gimplified nested
673 functions, so we can end up here with an already gimplified
674 body. */
675 if (!gimple_has_body_p (decl))
676 gimplify_function_tree (decl);
677
678 /* Lower the function. */
679 if (!lowered)
680 {
681 if (first_nested_function (this))
682 lower_nested_functions (decl);
683
684 gimple_register_cfg_hooks ();
685 bitmap_obstack_initialize (NULL);
686 execute_pass_list (cfun, g->get_passes ()->all_lowering_passes);
687 compact_blocks ();
688 bitmap_obstack_release (NULL);
689 lowered = true;
690 }
691
692 pop_cfun ();
693 }
694 analyzed = true;
695
696 input_location = saved_loc;
697 }
698
699 /* C++ frontend produce same body aliases all over the place, even before PCH
700 gets streamed out. It relies on us linking the aliases with their function
701 in order to do the fixups, but ipa-ref is not PCH safe. Consequently we
702 first produce aliases without links, but once C++ FE is sure he won't stream
703 PCH we build the links via this function. */
704
705 void
process_same_body_aliases(void)706 symbol_table::process_same_body_aliases (void)
707 {
708 symtab_node *node;
709 FOR_EACH_SYMBOL (node)
710 if (node->cpp_implicit_alias && !node->analyzed)
711 node->resolve_alias
712 (VAR_P (node->alias_target)
713 ? (symtab_node *)varpool_node::get_create (node->alias_target)
714 : (symtab_node *)cgraph_node::get_create (node->alias_target));
715 cpp_implicit_aliases_done = true;
716 }
717
718 /* Process a symver attribute. */
719
720 static void
process_symver_attribute(symtab_node * n)721 process_symver_attribute (symtab_node *n)
722 {
723 tree value = lookup_attribute ("symver", DECL_ATTRIBUTES (n->decl));
724
725 for (; value != NULL; value = TREE_CHAIN (value))
726 {
727 /* Starting from bintuils 2.35 gas supports:
728 # Assign foo to bar@V1 and baz@V2.
729 .symver foo, bar@V1
730 .symver foo, baz@V2
731 */
732 const char *purpose = IDENTIFIER_POINTER (TREE_PURPOSE (value));
733 if (strcmp (purpose, "symver") != 0)
734 continue;
735
736 tree symver = get_identifier_with_length
737 (TREE_STRING_POINTER (TREE_VALUE (TREE_VALUE (value))),
738 TREE_STRING_LENGTH (TREE_VALUE (TREE_VALUE (value))));
739 symtab_node *def = symtab_node::get_for_asmname (symver);
740
741 if (def)
742 {
743 error_at (DECL_SOURCE_LOCATION (n->decl),
744 "duplicate definition of a symbol version");
745 inform (DECL_SOURCE_LOCATION (def->decl),
746 "same version was previously defined here");
747 return;
748 }
749 if (!n->definition)
750 {
751 error_at (DECL_SOURCE_LOCATION (n->decl),
752 "symbol needs to be defined to have a version");
753 return;
754 }
755 if (DECL_COMMON (n->decl))
756 {
757 error_at (DECL_SOURCE_LOCATION (n->decl),
758 "common symbol cannot be versioned");
759 return;
760 }
761 if (DECL_COMDAT (n->decl))
762 {
763 error_at (DECL_SOURCE_LOCATION (n->decl),
764 "comdat symbol cannot be versioned");
765 return;
766 }
767 if (n->weakref)
768 {
769 error_at (DECL_SOURCE_LOCATION (n->decl),
770 "%<weakref%> cannot be versioned");
771 return;
772 }
773 if (!TREE_PUBLIC (n->decl))
774 {
775 error_at (DECL_SOURCE_LOCATION (n->decl),
776 "versioned symbol must be public");
777 return;
778 }
779 if (DECL_VISIBILITY (n->decl) != VISIBILITY_DEFAULT)
780 {
781 error_at (DECL_SOURCE_LOCATION (n->decl),
782 "versioned symbol must have default visibility");
783 return;
784 }
785
786 /* Create new symbol table entry representing the version. */
787 tree new_decl = copy_node (n->decl);
788
789 DECL_INITIAL (new_decl) = NULL_TREE;
790 if (TREE_CODE (new_decl) == FUNCTION_DECL)
791 DECL_STRUCT_FUNCTION (new_decl) = NULL;
792 SET_DECL_ASSEMBLER_NAME (new_decl, symver);
793 TREE_PUBLIC (new_decl) = 1;
794 DECL_ATTRIBUTES (new_decl) = NULL;
795
796 symtab_node *symver_node = symtab_node::get_create (new_decl);
797 symver_node->alias = true;
798 symver_node->definition = true;
799 symver_node->symver = true;
800 symver_node->create_reference (n, IPA_REF_ALIAS, NULL);
801 symver_node->analyzed = true;
802 }
803 }
804
805 /* Process attributes common for vars and functions. */
806
807 static void
process_common_attributes(symtab_node * node,tree decl)808 process_common_attributes (symtab_node *node, tree decl)
809 {
810 tree weakref = lookup_attribute ("weakref", DECL_ATTRIBUTES (decl));
811
812 if (weakref && !lookup_attribute ("alias", DECL_ATTRIBUTES (decl)))
813 {
814 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
815 "%<weakref%> attribute should be accompanied with"
816 " an %<alias%> attribute");
817 DECL_WEAK (decl) = 0;
818 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
819 DECL_ATTRIBUTES (decl));
820 }
821
822 if (lookup_attribute ("no_reorder", DECL_ATTRIBUTES (decl)))
823 node->no_reorder = 1;
824 process_symver_attribute (node);
825 }
826
827 /* Look for externally_visible and used attributes and mark cgraph nodes
828 accordingly.
829
830 We cannot mark the nodes at the point the attributes are processed (in
831 handle_*_attribute) because the copy of the declarations available at that
832 point may not be canonical. For example, in:
833
834 void f();
835 void f() __attribute__((used));
836
837 the declaration we see in handle_used_attribute will be the second
838 declaration -- but the front end will subsequently merge that declaration
839 with the original declaration and discard the second declaration.
840
841 Furthermore, we can't mark these nodes in finalize_function because:
842
843 void f() {}
844 void f() __attribute__((externally_visible));
845
846 is valid.
847
848 So, we walk the nodes at the end of the translation unit, applying the
849 attributes at that point. */
850
851 static void
process_function_and_variable_attributes(cgraph_node * first,varpool_node * first_var)852 process_function_and_variable_attributes (cgraph_node *first,
853 varpool_node *first_var)
854 {
855 cgraph_node *node;
856 varpool_node *vnode;
857
858 for (node = symtab->first_function (); node != first;
859 node = symtab->next_function (node))
860 {
861 tree decl = node->decl;
862
863 if (node->alias
864 && lookup_attribute ("flatten", DECL_ATTRIBUTES (decl)))
865 {
866 tree tdecl = node->get_alias_target_tree ();
867 if (!tdecl || !DECL_P (tdecl)
868 || !lookup_attribute ("flatten", DECL_ATTRIBUTES (tdecl)))
869 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
870 "%<flatten%> attribute is ignored on aliases");
871 }
872 if (DECL_PRESERVE_P (decl))
873 node->mark_force_output ();
874 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
875 {
876 if (! TREE_PUBLIC (node->decl))
877 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes,
878 "%<externally_visible%>"
879 " attribute have effect only on public objects");
880 }
881 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl))
882 && node->definition
883 && (!node->alias || DECL_INITIAL (decl) != error_mark_node))
884 {
885 /* NODE->DEFINITION && NODE->ALIAS is nonzero for valid weakref
886 function declarations; DECL_INITIAL is non-null for invalid
887 weakref functions that are also defined. */
888 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
889 "%<weakref%> attribute ignored"
890 " because function is defined");
891 DECL_WEAK (decl) = 0;
892 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
893 DECL_ATTRIBUTES (decl));
894 DECL_ATTRIBUTES (decl) = remove_attribute ("alias",
895 DECL_ATTRIBUTES (decl));
896 node->alias = false;
897 node->weakref = false;
898 node->transparent_alias = false;
899 }
900 else if (lookup_attribute ("alias", DECL_ATTRIBUTES (decl))
901 && node->definition
902 && !node->alias)
903 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes,
904 "%<alias%> attribute ignored"
905 " because function is defined");
906
907 if (lookup_attribute ("always_inline", DECL_ATTRIBUTES (decl))
908 && !DECL_DECLARED_INLINE_P (decl)
909 /* redefining extern inline function makes it DECL_UNINLINABLE. */
910 && !DECL_UNINLINABLE (decl))
911 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
912 "%<always_inline%> function might not be inlinable");
913
914 process_common_attributes (node, decl);
915 }
916 for (vnode = symtab->first_variable (); vnode != first_var;
917 vnode = symtab->next_variable (vnode))
918 {
919 tree decl = vnode->decl;
920 if (DECL_EXTERNAL (decl)
921 && DECL_INITIAL (decl))
922 varpool_node::finalize_decl (decl);
923 if (DECL_PRESERVE_P (decl))
924 vnode->force_output = true;
925 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
926 {
927 if (! TREE_PUBLIC (vnode->decl))
928 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes,
929 "%<externally_visible%>"
930 " attribute have effect only on public objects");
931 }
932 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl))
933 && vnode->definition
934 && DECL_INITIAL (decl))
935 {
936 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes,
937 "%<weakref%> attribute ignored"
938 " because variable is initialized");
939 DECL_WEAK (decl) = 0;
940 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
941 DECL_ATTRIBUTES (decl));
942 }
943 process_common_attributes (vnode, decl);
944 }
945 }
946
947 /* Mark DECL as finalized. By finalizing the declaration, frontend instruct the
948 middle end to output the variable to asm file, if needed or externally
949 visible. */
950
951 void
finalize_decl(tree decl)952 varpool_node::finalize_decl (tree decl)
953 {
954 varpool_node *node = varpool_node::get_create (decl);
955
956 gcc_assert (TREE_STATIC (decl) || DECL_EXTERNAL (decl));
957
958 if (node->definition)
959 return;
960 /* Set definition first before calling notice_global_symbol so that
961 it is available to notice_global_symbol. */
962 node->definition = true;
963 node->semantic_interposition = flag_semantic_interposition;
964 notice_global_symbol (decl);
965 if (!flag_toplevel_reorder)
966 node->no_reorder = true;
967 if (TREE_THIS_VOLATILE (decl) || DECL_PRESERVE_P (decl)
968 /* Traditionally we do not eliminate static variables when not
969 optimizing and when not doing toplevel reorder. */
970 || (node->no_reorder && !DECL_COMDAT (node->decl)
971 && !DECL_ARTIFICIAL (node->decl)))
972 node->force_output = true;
973
974 if (symtab->state == CONSTRUCTION
975 && (node->needed_p () || node->referred_to_p ()))
976 enqueue_node (node);
977 if (symtab->state >= IPA_SSA)
978 node->analyze ();
979 /* Some frontends produce various interface variables after compilation
980 finished. */
981 if (symtab->state == FINISHED
982 || (node->no_reorder
983 && symtab->state == EXPANSION))
984 node->assemble_decl ();
985 }
986
987 /* EDGE is an polymorphic call. Mark all possible targets as reachable
988 and if there is only one target, perform trivial devirtualization.
989 REACHABLE_CALL_TARGETS collects target lists we already walked to
990 avoid duplicate work. */
991
992 static void
walk_polymorphic_call_targets(hash_set<void * > * reachable_call_targets,cgraph_edge * edge)993 walk_polymorphic_call_targets (hash_set<void *> *reachable_call_targets,
994 cgraph_edge *edge)
995 {
996 unsigned int i;
997 void *cache_token;
998 bool final;
999 vec <cgraph_node *>targets
1000 = possible_polymorphic_call_targets
1001 (edge, &final, &cache_token);
1002
1003 if (!reachable_call_targets->add (cache_token))
1004 {
1005 if (symtab->dump_file)
1006 dump_possible_polymorphic_call_targets
1007 (symtab->dump_file, edge);
1008
1009 for (i = 0; i < targets.length (); i++)
1010 {
1011 /* Do not bother to mark virtual methods in anonymous namespace;
1012 either we will find use of virtual table defining it, or it is
1013 unused. */
1014 if (targets[i]->definition
1015 && TREE_CODE
1016 (TREE_TYPE (targets[i]->decl))
1017 == METHOD_TYPE
1018 && !type_in_anonymous_namespace_p
1019 (TYPE_METHOD_BASETYPE (TREE_TYPE (targets[i]->decl))))
1020 enqueue_node (targets[i]);
1021 }
1022 }
1023
1024 /* Very trivial devirtualization; when the type is
1025 final or anonymous (so we know all its derivation)
1026 and there is only one possible virtual call target,
1027 make the edge direct. */
1028 if (final)
1029 {
1030 if (targets.length () <= 1 && dbg_cnt (devirt))
1031 {
1032 cgraph_node *target;
1033 if (targets.length () == 1)
1034 target = targets[0];
1035 else
1036 target = cgraph_node::create
1037 (builtin_decl_implicit (BUILT_IN_UNREACHABLE));
1038
1039 if (symtab->dump_file)
1040 {
1041 fprintf (symtab->dump_file,
1042 "Devirtualizing call: ");
1043 print_gimple_stmt (symtab->dump_file,
1044 edge->call_stmt, 0,
1045 TDF_SLIM);
1046 }
1047 if (dump_enabled_p ())
1048 {
1049 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, edge->call_stmt,
1050 "devirtualizing call in %s to %s\n",
1051 edge->caller->dump_name (),
1052 target->dump_name ());
1053 }
1054
1055 edge = cgraph_edge::make_direct (edge, target);
1056 gimple *new_call = cgraph_edge::redirect_call_stmt_to_callee (edge);
1057
1058 if (symtab->dump_file)
1059 {
1060 fprintf (symtab->dump_file, "Devirtualized as: ");
1061 print_gimple_stmt (symtab->dump_file, new_call, 0, TDF_SLIM);
1062 }
1063 }
1064 }
1065 }
1066
1067 /* Issue appropriate warnings for the global declaration DECL. */
1068
1069 static void
check_global_declaration(symtab_node * snode)1070 check_global_declaration (symtab_node *snode)
1071 {
1072 const char *decl_file;
1073 tree decl = snode->decl;
1074
1075 /* Warn about any function declared static but not defined. We don't
1076 warn about variables, because many programs have static variables
1077 that exist only to get some text into the object file. */
1078 if (TREE_CODE (decl) == FUNCTION_DECL
1079 && DECL_INITIAL (decl) == 0
1080 && DECL_EXTERNAL (decl)
1081 && ! DECL_ARTIFICIAL (decl)
1082 && ! TREE_PUBLIC (decl))
1083 {
1084 if (warning_suppressed_p (decl, OPT_Wunused))
1085 ;
1086 else if (snode->referred_to_p (/*include_self=*/false))
1087 pedwarn (input_location, 0, "%q+F used but never defined", decl);
1088 else
1089 warning (OPT_Wunused_function, "%q+F declared %<static%> but never "
1090 "defined", decl);
1091 /* This symbol is effectively an "extern" declaration now. */
1092 TREE_PUBLIC (decl) = 1;
1093 }
1094
1095 /* Warn about static fns or vars defined but not used. */
1096 if (((warn_unused_function && TREE_CODE (decl) == FUNCTION_DECL)
1097 || (((warn_unused_variable && ! TREE_READONLY (decl))
1098 || (warn_unused_const_variable > 0 && TREE_READONLY (decl)
1099 && (warn_unused_const_variable == 2
1100 || (main_input_filename != NULL
1101 && (decl_file = DECL_SOURCE_FILE (decl)) != NULL
1102 && filename_cmp (main_input_filename,
1103 decl_file) == 0))))
1104 && VAR_P (decl)))
1105 && ! DECL_IN_SYSTEM_HEADER (decl)
1106 && ! snode->referred_to_p (/*include_self=*/false)
1107 /* This TREE_USED check is needed in addition to referred_to_p
1108 above, because the `__unused__' attribute is not being
1109 considered for referred_to_p. */
1110 && ! TREE_USED (decl)
1111 /* The TREE_USED bit for file-scope decls is kept in the identifier,
1112 to handle multiple external decls in different scopes. */
1113 && ! (DECL_NAME (decl) && TREE_USED (DECL_NAME (decl)))
1114 && ! DECL_EXTERNAL (decl)
1115 && ! DECL_ARTIFICIAL (decl)
1116 && ! DECL_ABSTRACT_ORIGIN (decl)
1117 && ! TREE_PUBLIC (decl)
1118 /* A volatile variable might be used in some non-obvious way. */
1119 && (! VAR_P (decl) || ! TREE_THIS_VOLATILE (decl))
1120 /* Global register variables must be declared to reserve them. */
1121 && ! (VAR_P (decl) && DECL_REGISTER (decl))
1122 /* Global ctors and dtors are called by the runtime. */
1123 && (TREE_CODE (decl) != FUNCTION_DECL
1124 || (!DECL_STATIC_CONSTRUCTOR (decl)
1125 && !DECL_STATIC_DESTRUCTOR (decl)))
1126 && (! VAR_P (decl) || !warning_suppressed_p (decl, OPT_Wunused_variable))
1127 /* Otherwise, ask the language. */
1128 && lang_hooks.decls.warn_unused_global (decl))
1129 warning_at (DECL_SOURCE_LOCATION (decl),
1130 (TREE_CODE (decl) == FUNCTION_DECL)
1131 ? OPT_Wunused_function
1132 : (TREE_READONLY (decl)
1133 ? OPT_Wunused_const_variable_
1134 : OPT_Wunused_variable),
1135 "%qD defined but not used", decl);
1136 }
1137
1138 /* Discover all functions and variables that are trivially needed, analyze
1139 them as well as all functions and variables referred by them */
1140 static cgraph_node *first_analyzed;
1141 static varpool_node *first_analyzed_var;
1142
1143 /* FIRST_TIME is set to TRUE for the first time we are called for a
1144 translation unit from finalize_compilation_unit() or false
1145 otherwise. */
1146
1147 static void
analyze_functions(bool first_time)1148 analyze_functions (bool first_time)
1149 {
1150 /* Keep track of already processed nodes when called multiple times for
1151 intermodule optimization. */
1152 cgraph_node *first_handled = first_analyzed;
1153 varpool_node *first_handled_var = first_analyzed_var;
1154 hash_set<void *> reachable_call_targets;
1155
1156 symtab_node *node;
1157 symtab_node *next;
1158 int i;
1159 ipa_ref *ref;
1160 bool changed = true;
1161 location_t saved_loc = input_location;
1162
1163 bitmap_obstack_initialize (NULL);
1164 symtab->state = CONSTRUCTION;
1165 input_location = UNKNOWN_LOCATION;
1166
1167 thunk_info::process_early_thunks ();
1168
1169 /* Ugly, but the fixup cannot happen at a time same body alias is created;
1170 C++ FE is confused about the COMDAT groups being right. */
1171 if (symtab->cpp_implicit_aliases_done)
1172 FOR_EACH_SYMBOL (node)
1173 if (node->cpp_implicit_alias)
1174 node->fixup_same_cpp_alias_visibility (node->get_alias_target ());
1175 build_type_inheritance_graph ();
1176
1177 if (flag_openmp && first_time)
1178 omp_discover_implicit_declare_target ();
1179
1180 /* Analysis adds static variables that in turn adds references to new functions.
1181 So we need to iterate the process until it stabilize. */
1182 while (changed)
1183 {
1184 changed = false;
1185 process_function_and_variable_attributes (first_analyzed,
1186 first_analyzed_var);
1187
1188 /* First identify the trivially needed symbols. */
1189 for (node = symtab->first_symbol ();
1190 node != first_analyzed
1191 && node != first_analyzed_var; node = node->next)
1192 {
1193 /* Convert COMDAT group designators to IDENTIFIER_NODEs. */
1194 node->get_comdat_group_id ();
1195 if (node->needed_p ())
1196 {
1197 enqueue_node (node);
1198 if (!changed && symtab->dump_file)
1199 fprintf (symtab->dump_file, "Trivially needed symbols:");
1200 changed = true;
1201 if (symtab->dump_file)
1202 fprintf (symtab->dump_file, " %s", node->dump_asm_name ());
1203 }
1204 if (node == first_analyzed
1205 || node == first_analyzed_var)
1206 break;
1207 }
1208 symtab->process_new_functions ();
1209 first_analyzed_var = symtab->first_variable ();
1210 first_analyzed = symtab->first_function ();
1211
1212 if (changed && symtab->dump_file)
1213 fprintf (symtab->dump_file, "\n");
1214
1215 /* Lower representation, build callgraph edges and references for all trivially
1216 needed symbols and all symbols referred by them. */
1217 while (queued_nodes != &symtab_terminator)
1218 {
1219 changed = true;
1220 node = queued_nodes;
1221 queued_nodes = (symtab_node *)queued_nodes->aux;
1222 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
1223 if (cnode && cnode->definition)
1224 {
1225 cgraph_edge *edge;
1226 tree decl = cnode->decl;
1227
1228 /* ??? It is possible to create extern inline function
1229 and later using weak alias attribute to kill its body.
1230 See gcc.c-torture/compile/20011119-1.c */
1231 if (!DECL_STRUCT_FUNCTION (decl)
1232 && !cnode->alias
1233 && !cnode->thunk
1234 && !cnode->dispatcher_function)
1235 {
1236 cnode->reset ();
1237 cnode->redefined_extern_inline = true;
1238 continue;
1239 }
1240
1241 if (!cnode->analyzed)
1242 cnode->analyze ();
1243
1244 for (edge = cnode->callees; edge; edge = edge->next_callee)
1245 if (edge->callee->definition
1246 && (!DECL_EXTERNAL (edge->callee->decl)
1247 /* When not optimizing, do not try to analyze extern
1248 inline functions. Doing so is pointless. */
1249 || opt_for_fn (edge->callee->decl, optimize)
1250 /* Weakrefs needs to be preserved. */
1251 || edge->callee->alias
1252 /* always_inline functions are inlined even at -O0. */
1253 || lookup_attribute
1254 ("always_inline",
1255 DECL_ATTRIBUTES (edge->callee->decl))
1256 /* Multiversioned functions needs the dispatcher to
1257 be produced locally even for extern functions. */
1258 || edge->callee->function_version ()))
1259 enqueue_node (edge->callee);
1260 if (opt_for_fn (cnode->decl, optimize)
1261 && opt_for_fn (cnode->decl, flag_devirtualize))
1262 {
1263 cgraph_edge *next;
1264
1265 for (edge = cnode->indirect_calls; edge; edge = next)
1266 {
1267 next = edge->next_callee;
1268 if (edge->indirect_info->polymorphic)
1269 walk_polymorphic_call_targets (&reachable_call_targets,
1270 edge);
1271 }
1272 }
1273
1274 /* If decl is a clone of an abstract function,
1275 mark that abstract function so that we don't release its body.
1276 The DECL_INITIAL() of that abstract function declaration
1277 will be later needed to output debug info. */
1278 if (DECL_ABSTRACT_ORIGIN (decl))
1279 {
1280 cgraph_node *origin_node
1281 = cgraph_node::get_create (DECL_ABSTRACT_ORIGIN (decl));
1282 origin_node->used_as_abstract_origin = true;
1283 }
1284 /* Preserve a functions function context node. It will
1285 later be needed to output debug info. */
1286 if (tree fn = decl_function_context (decl))
1287 {
1288 cgraph_node *origin_node = cgraph_node::get_create (fn);
1289 enqueue_node (origin_node);
1290 }
1291 }
1292 else
1293 {
1294 varpool_node *vnode = dyn_cast <varpool_node *> (node);
1295 if (vnode && vnode->definition && !vnode->analyzed)
1296 vnode->analyze ();
1297 }
1298
1299 if (node->same_comdat_group)
1300 {
1301 symtab_node *next;
1302 for (next = node->same_comdat_group;
1303 next != node;
1304 next = next->same_comdat_group)
1305 if (!next->comdat_local_p ())
1306 enqueue_node (next);
1307 }
1308 for (i = 0; node->iterate_reference (i, ref); i++)
1309 if (ref->referred->definition
1310 && (!DECL_EXTERNAL (ref->referred->decl)
1311 || ((TREE_CODE (ref->referred->decl) != FUNCTION_DECL
1312 && optimize)
1313 || (TREE_CODE (ref->referred->decl) == FUNCTION_DECL
1314 && opt_for_fn (ref->referred->decl, optimize))
1315 || node->alias
1316 || ref->referred->alias)))
1317 enqueue_node (ref->referred);
1318 symtab->process_new_functions ();
1319 }
1320 }
1321 update_type_inheritance_graph ();
1322
1323 /* Collect entry points to the unit. */
1324 if (symtab->dump_file)
1325 {
1326 fprintf (symtab->dump_file, "\n\nInitial ");
1327 symtab->dump (symtab->dump_file);
1328 }
1329
1330 if (first_time)
1331 {
1332 symtab_node *snode;
1333 FOR_EACH_SYMBOL (snode)
1334 check_global_declaration (snode);
1335 }
1336
1337 if (symtab->dump_file)
1338 fprintf (symtab->dump_file, "\nRemoving unused symbols:");
1339
1340 for (node = symtab->first_symbol ();
1341 node != first_handled
1342 && node != first_handled_var; node = next)
1343 {
1344 next = node->next;
1345 /* For symbols declared locally we clear TREE_READONLY when emitting
1346 the constructor (if one is needed). For external declarations we can
1347 not safely assume that the type is readonly because we may be called
1348 during its construction. */
1349 if (TREE_CODE (node->decl) == VAR_DECL
1350 && TYPE_P (TREE_TYPE (node->decl))
1351 && TYPE_NEEDS_CONSTRUCTING (TREE_TYPE (node->decl))
1352 && DECL_EXTERNAL (node->decl))
1353 TREE_READONLY (node->decl) = 0;
1354 if (!node->aux && !node->referred_to_p ())
1355 {
1356 if (symtab->dump_file)
1357 fprintf (symtab->dump_file, " %s", node->dump_name ());
1358
1359 /* See if the debugger can use anything before the DECL
1360 passes away. Perhaps it can notice a DECL that is now a
1361 constant and can tag the early DIE with an appropriate
1362 attribute.
1363
1364 Otherwise, this is the last chance the debug_hooks have
1365 at looking at optimized away DECLs, since
1366 late_global_decl will subsequently be called from the
1367 contents of the now pruned symbol table. */
1368 if (VAR_P (node->decl)
1369 && !decl_function_context (node->decl))
1370 {
1371 /* We are reclaiming totally unreachable code and variables
1372 so they effectively appear as readonly. Show that to
1373 the debug machinery. */
1374 TREE_READONLY (node->decl) = 1;
1375 node->definition = false;
1376 (*debug_hooks->late_global_decl) (node->decl);
1377 }
1378
1379 node->remove ();
1380 continue;
1381 }
1382 if (cgraph_node *cnode = dyn_cast <cgraph_node *> (node))
1383 {
1384 tree decl = node->decl;
1385
1386 if (cnode->definition && !gimple_has_body_p (decl)
1387 && !cnode->alias
1388 && !cnode->thunk)
1389 cnode->reset ();
1390
1391 gcc_assert (!cnode->definition || cnode->thunk
1392 || cnode->alias
1393 || gimple_has_body_p (decl)
1394 || cnode->native_rtl_p ());
1395 gcc_assert (cnode->analyzed == cnode->definition);
1396 }
1397 node->aux = NULL;
1398 }
1399 for (;node; node = node->next)
1400 node->aux = NULL;
1401 first_analyzed = symtab->first_function ();
1402 first_analyzed_var = symtab->first_variable ();
1403 if (symtab->dump_file)
1404 {
1405 fprintf (symtab->dump_file, "\n\nReclaimed ");
1406 symtab->dump (symtab->dump_file);
1407 }
1408 bitmap_obstack_release (NULL);
1409 ggc_collect ();
1410 /* Initialize assembler name hash, in particular we want to trigger C++
1411 mangling and same body alias creation before we free DECL_ARGUMENTS
1412 used by it. */
1413 if (!seen_error ())
1414 symtab->symtab_initialize_asm_name_hash ();
1415
1416 input_location = saved_loc;
1417 }
1418
1419 /* Check declaration of the type of ALIAS for compatibility with its TARGET
1420 (which may be an ifunc resolver) and issue a diagnostic when they are
1421 not compatible according to language rules (plus a C++ extension for
1422 non-static member functions). */
1423
1424 static void
maybe_diag_incompatible_alias(tree alias,tree target)1425 maybe_diag_incompatible_alias (tree alias, tree target)
1426 {
1427 tree altype = TREE_TYPE (alias);
1428 tree targtype = TREE_TYPE (target);
1429
1430 bool ifunc = cgraph_node::get (alias)->ifunc_resolver;
1431 tree funcptr = altype;
1432
1433 if (ifunc)
1434 {
1435 /* Handle attribute ifunc first. */
1436 if (TREE_CODE (altype) == METHOD_TYPE)
1437 {
1438 /* Set FUNCPTR to the type of the alias target. If the type
1439 is a non-static member function of class C, construct a type
1440 of an ordinary function taking C* as the first argument,
1441 followed by the member function argument list, and use it
1442 instead to check for incompatibility. This conversion is
1443 not defined by the language but an extension provided by
1444 G++. */
1445
1446 tree rettype = TREE_TYPE (altype);
1447 tree args = TYPE_ARG_TYPES (altype);
1448 altype = build_function_type (rettype, args);
1449 funcptr = altype;
1450 }
1451
1452 targtype = TREE_TYPE (targtype);
1453
1454 if (POINTER_TYPE_P (targtype))
1455 {
1456 targtype = TREE_TYPE (targtype);
1457
1458 /* Only issue Wattribute-alias for conversions to void* with
1459 -Wextra. */
1460 if (VOID_TYPE_P (targtype) && !extra_warnings)
1461 return;
1462
1463 /* Proceed to handle incompatible ifunc resolvers below. */
1464 }
1465 else
1466 {
1467 funcptr = build_pointer_type (funcptr);
1468
1469 error_at (DECL_SOURCE_LOCATION (target),
1470 "%<ifunc%> resolver for %qD must return %qT",
1471 alias, funcptr);
1472 inform (DECL_SOURCE_LOCATION (alias),
1473 "resolver indirect function declared here");
1474 return;
1475 }
1476 }
1477
1478 if ((!FUNC_OR_METHOD_TYPE_P (targtype)
1479 || (prototype_p (altype)
1480 && prototype_p (targtype)
1481 && !types_compatible_p (altype, targtype))))
1482 {
1483 /* Warn for incompatibilities. Avoid warning for functions
1484 without a prototype to make it possible to declare aliases
1485 without knowing the exact type, as libstdc++ does. */
1486 if (ifunc)
1487 {
1488 funcptr = build_pointer_type (funcptr);
1489
1490 auto_diagnostic_group d;
1491 if (warning_at (DECL_SOURCE_LOCATION (target),
1492 OPT_Wattribute_alias_,
1493 "%<ifunc%> resolver for %qD should return %qT",
1494 alias, funcptr))
1495 inform (DECL_SOURCE_LOCATION (alias),
1496 "resolver indirect function declared here");
1497 }
1498 else
1499 {
1500 auto_diagnostic_group d;
1501 if (warning_at (DECL_SOURCE_LOCATION (alias),
1502 OPT_Wattribute_alias_,
1503 "%qD alias between functions of incompatible "
1504 "types %qT and %qT", alias, altype, targtype))
1505 inform (DECL_SOURCE_LOCATION (target),
1506 "aliased declaration here");
1507 }
1508 }
1509 }
1510
1511 /* Translate the ugly representation of aliases as alias pairs into nice
1512 representation in callgraph. We don't handle all cases yet,
1513 unfortunately. */
1514
1515 static void
handle_alias_pairs(void)1516 handle_alias_pairs (void)
1517 {
1518 alias_pair *p;
1519 unsigned i;
1520
1521 for (i = 0; alias_pairs && alias_pairs->iterate (i, &p);)
1522 {
1523 symtab_node *target_node = symtab_node::get_for_asmname (p->target);
1524
1525 /* Weakrefs with target not defined in current unit are easy to handle:
1526 they behave just as external variables except we need to note the
1527 alias flag to later output the weakref pseudo op into asm file. */
1528 if (!target_node
1529 && lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl)) != NULL)
1530 {
1531 symtab_node *node = symtab_node::get (p->decl);
1532 if (node)
1533 {
1534 node->alias_target = p->target;
1535 node->weakref = true;
1536 node->alias = true;
1537 node->transparent_alias = true;
1538 }
1539 alias_pairs->unordered_remove (i);
1540 continue;
1541 }
1542 else if (!target_node)
1543 {
1544 error ("%q+D aliased to undefined symbol %qE", p->decl, p->target);
1545 symtab_node *node = symtab_node::get (p->decl);
1546 if (node)
1547 node->alias = false;
1548 alias_pairs->unordered_remove (i);
1549 continue;
1550 }
1551
1552 if (DECL_EXTERNAL (target_node->decl)
1553 /* We use local aliases for C++ thunks to force the tailcall
1554 to bind locally. This is a hack - to keep it working do
1555 the following (which is not strictly correct). */
1556 && (TREE_CODE (target_node->decl) != FUNCTION_DECL
1557 || ! DECL_VIRTUAL_P (target_node->decl))
1558 && ! lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl)))
1559 {
1560 error ("%q+D aliased to external symbol %qE",
1561 p->decl, p->target);
1562 }
1563
1564 if (TREE_CODE (p->decl) == FUNCTION_DECL
1565 && target_node && is_a <cgraph_node *> (target_node))
1566 {
1567 maybe_diag_incompatible_alias (p->decl, target_node->decl);
1568
1569 maybe_diag_alias_attributes (p->decl, target_node->decl);
1570
1571 cgraph_node *src_node = cgraph_node::get (p->decl);
1572 if (src_node && src_node->definition)
1573 src_node->reset ();
1574 cgraph_node::create_alias (p->decl, target_node->decl);
1575 alias_pairs->unordered_remove (i);
1576 }
1577 else if (VAR_P (p->decl)
1578 && target_node && is_a <varpool_node *> (target_node))
1579 {
1580 varpool_node::create_alias (p->decl, target_node->decl);
1581 alias_pairs->unordered_remove (i);
1582 }
1583 else
1584 {
1585 error ("%q+D alias between function and variable is not supported",
1586 p->decl);
1587 inform (DECL_SOURCE_LOCATION (target_node->decl),
1588 "aliased declaration here");
1589
1590 alias_pairs->unordered_remove (i);
1591 }
1592 }
1593 vec_free (alias_pairs);
1594 }
1595
1596
1597 /* Figure out what functions we want to assemble. */
1598
1599 static void
mark_functions_to_output(void)1600 mark_functions_to_output (void)
1601 {
1602 bool check_same_comdat_groups = false;
1603 cgraph_node *node;
1604
1605 if (flag_checking)
1606 FOR_EACH_FUNCTION (node)
1607 gcc_assert (!node->process);
1608
1609 FOR_EACH_FUNCTION (node)
1610 {
1611 tree decl = node->decl;
1612
1613 gcc_assert (!node->process || node->same_comdat_group);
1614 if (node->process)
1615 continue;
1616
1617 /* We need to output all local functions that are used and not
1618 always inlined, as well as those that are reachable from
1619 outside the current compilation unit. */
1620 if (node->analyzed
1621 && !node->thunk
1622 && !node->alias
1623 && !node->inlined_to
1624 && !TREE_ASM_WRITTEN (decl)
1625 && !DECL_EXTERNAL (decl))
1626 {
1627 node->process = 1;
1628 if (node->same_comdat_group)
1629 {
1630 cgraph_node *next;
1631 for (next = dyn_cast<cgraph_node *> (node->same_comdat_group);
1632 next != node;
1633 next = dyn_cast<cgraph_node *> (next->same_comdat_group))
1634 if (!next->thunk && !next->alias
1635 && !next->comdat_local_p ())
1636 next->process = 1;
1637 }
1638 }
1639 else if (node->same_comdat_group)
1640 {
1641 if (flag_checking)
1642 check_same_comdat_groups = true;
1643 }
1644 else
1645 {
1646 /* We should've reclaimed all functions that are not needed. */
1647 if (flag_checking
1648 && !node->inlined_to
1649 && gimple_has_body_p (decl)
1650 /* FIXME: in ltrans unit when offline copy is outside partition but inline copies
1651 are inside partition, we can end up not removing the body since we no longer
1652 have analyzed node pointing to it. */
1653 && !node->in_other_partition
1654 && !node->alias
1655 && !node->clones
1656 && !DECL_EXTERNAL (decl))
1657 {
1658 node->debug ();
1659 internal_error ("failed to reclaim unneeded function");
1660 }
1661 gcc_assert (node->inlined_to
1662 || !gimple_has_body_p (decl)
1663 || node->in_other_partition
1664 || node->clones
1665 || DECL_ARTIFICIAL (decl)
1666 || DECL_EXTERNAL (decl));
1667
1668 }
1669
1670 }
1671 if (flag_checking && check_same_comdat_groups)
1672 FOR_EACH_FUNCTION (node)
1673 if (node->same_comdat_group && !node->process)
1674 {
1675 tree decl = node->decl;
1676 if (!node->inlined_to
1677 && gimple_has_body_p (decl)
1678 /* FIXME: in an ltrans unit when the offline copy is outside a
1679 partition but inline copies are inside a partition, we can
1680 end up not removing the body since we no longer have an
1681 analyzed node pointing to it. */
1682 && !node->in_other_partition
1683 && !node->clones
1684 && !DECL_EXTERNAL (decl))
1685 {
1686 node->debug ();
1687 internal_error ("failed to reclaim unneeded function in same "
1688 "comdat group");
1689 }
1690 }
1691 }
1692
1693 /* DECL is FUNCTION_DECL. Initialize datastructures so DECL is a function
1694 in lowered gimple form. IN_SSA is true if the gimple is in SSA.
1695
1696 Set current_function_decl and cfun to newly constructed empty function body.
1697 return basic block in the function body. */
1698
1699 basic_block
init_lowered_empty_function(tree decl,bool in_ssa,profile_count count)1700 init_lowered_empty_function (tree decl, bool in_ssa, profile_count count)
1701 {
1702 basic_block bb;
1703 edge e;
1704
1705 current_function_decl = decl;
1706 allocate_struct_function (decl, false);
1707 gimple_register_cfg_hooks ();
1708 init_empty_tree_cfg ();
1709 init_tree_ssa (cfun);
1710
1711 if (in_ssa)
1712 {
1713 init_ssa_operands (cfun);
1714 cfun->gimple_df->in_ssa_p = true;
1715 cfun->curr_properties |= PROP_ssa;
1716 }
1717
1718 DECL_INITIAL (decl) = make_node (BLOCK);
1719 BLOCK_SUPERCONTEXT (DECL_INITIAL (decl)) = decl;
1720
1721 DECL_SAVED_TREE (decl) = error_mark_node;
1722 cfun->curr_properties |= (PROP_gimple_lcf | PROP_gimple_leh | PROP_gimple_any
1723 | PROP_cfg | PROP_loops);
1724
1725 set_loops_for_fn (cfun, ggc_cleared_alloc<loops> ());
1726 init_loops_structure (cfun, loops_for_fn (cfun), 1);
1727 loops_for_fn (cfun)->state |= LOOPS_MAY_HAVE_MULTIPLE_LATCHES;
1728
1729 /* Create BB for body of the function and connect it properly. */
1730 ENTRY_BLOCK_PTR_FOR_FN (cfun)->count = count;
1731 EXIT_BLOCK_PTR_FOR_FN (cfun)->count = count;
1732 bb = create_basic_block (NULL, ENTRY_BLOCK_PTR_FOR_FN (cfun));
1733 bb->count = count;
1734 e = make_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun), bb, EDGE_FALLTHRU);
1735 e->probability = profile_probability::always ();
1736 e = make_edge (bb, EXIT_BLOCK_PTR_FOR_FN (cfun), 0);
1737 e->probability = profile_probability::always ();
1738 add_bb_to_loop (bb, ENTRY_BLOCK_PTR_FOR_FN (cfun)->loop_father);
1739
1740 return bb;
1741 }
1742
1743 /* Assemble thunks and aliases associated to node. */
1744
1745 void
assemble_thunks_and_aliases(void)1746 cgraph_node::assemble_thunks_and_aliases (void)
1747 {
1748 cgraph_edge *e;
1749 ipa_ref *ref;
1750
1751 for (e = callers; e;)
1752 if (e->caller->thunk
1753 && !e->caller->inlined_to)
1754 {
1755 cgraph_node *thunk = e->caller;
1756
1757 e = e->next_caller;
1758 expand_thunk (thunk, !rtl_dump_and_exit, false);
1759 thunk->assemble_thunks_and_aliases ();
1760 }
1761 else
1762 e = e->next_caller;
1763
1764 FOR_EACH_ALIAS (this, ref)
1765 {
1766 cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
1767 if (!alias->transparent_alias)
1768 {
1769 bool saved_written = TREE_ASM_WRITTEN (decl);
1770
1771 /* Force assemble_alias to really output the alias this time instead
1772 of buffering it in same alias pairs. */
1773 TREE_ASM_WRITTEN (decl) = 1;
1774 if (alias->symver)
1775 do_assemble_symver (alias->decl,
1776 DECL_ASSEMBLER_NAME (decl));
1777 else
1778 do_assemble_alias (alias->decl,
1779 DECL_ASSEMBLER_NAME (decl));
1780 alias->assemble_thunks_and_aliases ();
1781 TREE_ASM_WRITTEN (decl) = saved_written;
1782 }
1783 }
1784 }
1785
1786 /* Expand function specified by node. */
1787
1788 void
expand(void)1789 cgraph_node::expand (void)
1790 {
1791 location_t saved_loc;
1792
1793 /* We ought to not compile any inline clones. */
1794 gcc_assert (!inlined_to);
1795
1796 /* __RTL functions are compiled as soon as they are parsed, so don't
1797 do it again. */
1798 if (native_rtl_p ())
1799 return;
1800
1801 announce_function (decl);
1802 process = 0;
1803 gcc_assert (lowered);
1804
1805 /* Initialize the default bitmap obstack. */
1806 bitmap_obstack_initialize (NULL);
1807 get_untransformed_body ();
1808
1809 /* Generate RTL for the body of DECL. */
1810
1811 timevar_push (TV_REST_OF_COMPILATION);
1812
1813 gcc_assert (symtab->global_info_ready);
1814
1815 /* Initialize the RTL code for the function. */
1816 saved_loc = input_location;
1817 input_location = DECL_SOURCE_LOCATION (decl);
1818
1819 gcc_assert (DECL_STRUCT_FUNCTION (decl));
1820 push_cfun (DECL_STRUCT_FUNCTION (decl));
1821 init_function_start (decl);
1822
1823 gimple_register_cfg_hooks ();
1824
1825 bitmap_obstack_initialize (®_obstack); /* FIXME, only at RTL generation*/
1826
1827 update_ssa (TODO_update_ssa_only_virtuals);
1828 if (ipa_transforms_to_apply.exists ())
1829 execute_all_ipa_transforms (false);
1830
1831 /* Perform all tree transforms and optimizations. */
1832
1833 /* Signal the start of passes. */
1834 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_START, NULL);
1835
1836 execute_pass_list (cfun, g->get_passes ()->all_passes);
1837
1838 /* Signal the end of passes. */
1839 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_END, NULL);
1840
1841 bitmap_obstack_release (®_obstack);
1842
1843 /* Release the default bitmap obstack. */
1844 bitmap_obstack_release (NULL);
1845
1846 /* If requested, warn about function definitions where the function will
1847 return a value (usually of some struct or union type) which itself will
1848 take up a lot of stack space. */
1849 if (!DECL_EXTERNAL (decl) && TREE_TYPE (decl))
1850 {
1851 tree ret_type = TREE_TYPE (TREE_TYPE (decl));
1852
1853 if (ret_type && TYPE_SIZE_UNIT (ret_type)
1854 && TREE_CODE (TYPE_SIZE_UNIT (ret_type)) == INTEGER_CST
1855 && compare_tree_int (TYPE_SIZE_UNIT (ret_type),
1856 warn_larger_than_size) > 0)
1857 {
1858 unsigned int size_as_int
1859 = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (ret_type));
1860
1861 if (compare_tree_int (TYPE_SIZE_UNIT (ret_type), size_as_int) == 0)
1862 warning (OPT_Wlarger_than_,
1863 "size of return value of %q+D is %u bytes",
1864 decl, size_as_int);
1865 else
1866 warning (OPT_Wlarger_than_,
1867 "size of return value of %q+D is larger than %wu bytes",
1868 decl, warn_larger_than_size);
1869 }
1870 }
1871
1872 gimple_set_body (decl, NULL);
1873 if (DECL_STRUCT_FUNCTION (decl) == 0)
1874 {
1875 /* Stop pointing to the local nodes about to be freed.
1876 But DECL_INITIAL must remain nonzero so we know this
1877 was an actual function definition. */
1878 if (DECL_INITIAL (decl) != 0)
1879 DECL_INITIAL (decl) = error_mark_node;
1880 }
1881
1882 input_location = saved_loc;
1883
1884 ggc_collect ();
1885 timevar_pop (TV_REST_OF_COMPILATION);
1886
1887 /* Make sure that BE didn't give up on compiling. */
1888 gcc_assert (TREE_ASM_WRITTEN (decl));
1889 if (cfun)
1890 pop_cfun ();
1891
1892 /* It would make a lot more sense to output thunks before function body to
1893 get more forward and fewer backward jumps. This however would need
1894 solving problem with comdats. See PR48668. Also aliases must come after
1895 function itself to make one pass assemblers, like one on AIX, happy.
1896 See PR 50689.
1897 FIXME: Perhaps thunks should be move before function IFF they are not in
1898 comdat groups. */
1899 assemble_thunks_and_aliases ();
1900 release_body ();
1901 }
1902
1903 /* Node comparator that is responsible for the order that corresponds
1904 to time when a function was launched for the first time. */
1905
1906 int
tp_first_run_node_cmp(const void * pa,const void * pb)1907 tp_first_run_node_cmp (const void *pa, const void *pb)
1908 {
1909 const cgraph_node *a = *(const cgraph_node * const *) pa;
1910 const cgraph_node *b = *(const cgraph_node * const *) pb;
1911 unsigned int tp_first_run_a = a->tp_first_run;
1912 unsigned int tp_first_run_b = b->tp_first_run;
1913
1914 if (!opt_for_fn (a->decl, flag_profile_reorder_functions)
1915 || a->no_reorder)
1916 tp_first_run_a = 0;
1917 if (!opt_for_fn (b->decl, flag_profile_reorder_functions)
1918 || b->no_reorder)
1919 tp_first_run_b = 0;
1920
1921 if (tp_first_run_a == tp_first_run_b)
1922 return a->order - b->order;
1923
1924 /* Functions with time profile must be before these without profile. */
1925 tp_first_run_a = (tp_first_run_a - 1) & INT_MAX;
1926 tp_first_run_b = (tp_first_run_b - 1) & INT_MAX;
1927
1928 return tp_first_run_a - tp_first_run_b;
1929 }
1930
1931 /* Expand all functions that must be output.
1932
1933 Attempt to topologically sort the nodes so function is output when
1934 all called functions are already assembled to allow data to be
1935 propagated across the callgraph. Use a stack to get smaller distance
1936 between a function and its callees (later we may choose to use a more
1937 sophisticated algorithm for function reordering; we will likely want
1938 to use subsections to make the output functions appear in top-down
1939 order). */
1940
1941 static void
expand_all_functions(void)1942 expand_all_functions (void)
1943 {
1944 cgraph_node *node;
1945 cgraph_node **order = XCNEWVEC (cgraph_node *,
1946 symtab->cgraph_count);
1947 cgraph_node **tp_first_run_order = XCNEWVEC (cgraph_node *,
1948 symtab->cgraph_count);
1949 unsigned int expanded_func_count = 0, profiled_func_count = 0;
1950 int order_pos, tp_first_run_order_pos = 0, new_order_pos = 0;
1951 int i;
1952
1953 order_pos = ipa_reverse_postorder (order);
1954 gcc_assert (order_pos == symtab->cgraph_count);
1955
1956 /* Garbage collector may remove inline clones we eliminate during
1957 optimization. So we must be sure to not reference them. */
1958 for (i = 0; i < order_pos; i++)
1959 if (order[i]->process)
1960 {
1961 if (order[i]->tp_first_run
1962 && opt_for_fn (order[i]->decl, flag_profile_reorder_functions))
1963 tp_first_run_order[tp_first_run_order_pos++] = order[i];
1964 else
1965 order[new_order_pos++] = order[i];
1966 }
1967
1968 /* First output functions with time profile in specified order. */
1969 qsort (tp_first_run_order, tp_first_run_order_pos,
1970 sizeof (cgraph_node *), tp_first_run_node_cmp);
1971 for (i = 0; i < tp_first_run_order_pos; i++)
1972 {
1973 node = tp_first_run_order[i];
1974
1975 if (node->process)
1976 {
1977 expanded_func_count++;
1978 profiled_func_count++;
1979
1980 if (symtab->dump_file)
1981 fprintf (symtab->dump_file,
1982 "Time profile order in expand_all_functions:%s:%d\n",
1983 node->dump_asm_name (), node->tp_first_run);
1984 node->process = 0;
1985 node->expand ();
1986 }
1987 }
1988
1989 /* Output functions in RPO so callees get optimized before callers. This
1990 makes ipa-ra and other propagators to work.
1991 FIXME: This is far from optimal code layout. */
1992 for (i = new_order_pos - 1; i >= 0; i--)
1993 {
1994 node = order[i];
1995
1996 if (node->process)
1997 {
1998 expanded_func_count++;
1999 node->process = 0;
2000 node->expand ();
2001 }
2002 }
2003
2004 if (dump_file)
2005 fprintf (dump_file, "Expanded functions with time profile (%s):%u/%u\n",
2006 main_input_filename, profiled_func_count, expanded_func_count);
2007
2008 if (symtab->dump_file && tp_first_run_order_pos)
2009 fprintf (symtab->dump_file, "Expanded functions with time profile:%u/%u\n",
2010 profiled_func_count, expanded_func_count);
2011
2012 symtab->process_new_functions ();
2013 free_gimplify_stack ();
2014 delete ipa_saved_clone_sources;
2015 ipa_saved_clone_sources = NULL;
2016 free (order);
2017 free (tp_first_run_order);
2018 }
2019
2020 /* This is used to sort the node types by the cgraph order number. */
2021
2022 enum cgraph_order_sort_kind
2023 {
2024 ORDER_FUNCTION,
2025 ORDER_VAR,
2026 ORDER_VAR_UNDEF,
2027 ORDER_ASM
2028 };
2029
2030 struct cgraph_order_sort
2031 {
2032 /* Construct from a cgraph_node. */
cgraph_order_sortcgraph_order_sort2033 cgraph_order_sort (cgraph_node *node)
2034 : kind (ORDER_FUNCTION), order (node->order)
2035 {
2036 u.f = node;
2037 }
2038
2039 /* Construct from a varpool_node. */
cgraph_order_sortcgraph_order_sort2040 cgraph_order_sort (varpool_node *node)
2041 : kind (node->definition ? ORDER_VAR : ORDER_VAR_UNDEF), order (node->order)
2042 {
2043 u.v = node;
2044 }
2045
2046 /* Construct from a asm_node. */
cgraph_order_sortcgraph_order_sort2047 cgraph_order_sort (asm_node *node)
2048 : kind (ORDER_ASM), order (node->order)
2049 {
2050 u.a = node;
2051 }
2052
2053 /* Assembly cgraph_order_sort based on its type. */
2054 void process ();
2055
2056 enum cgraph_order_sort_kind kind;
2057 union
2058 {
2059 cgraph_node *f;
2060 varpool_node *v;
2061 asm_node *a;
2062 } u;
2063 int order;
2064 };
2065
2066 /* Assembly cgraph_order_sort based on its type. */
2067
2068 void
process()2069 cgraph_order_sort::process ()
2070 {
2071 switch (kind)
2072 {
2073 case ORDER_FUNCTION:
2074 u.f->process = 0;
2075 u.f->expand ();
2076 break;
2077 case ORDER_VAR:
2078 u.v->assemble_decl ();
2079 break;
2080 case ORDER_VAR_UNDEF:
2081 assemble_undefined_decl (u.v->decl);
2082 break;
2083 case ORDER_ASM:
2084 assemble_asm (u.a->asm_str);
2085 break;
2086 default:
2087 gcc_unreachable ();
2088 }
2089 }
2090
2091 /* Compare cgraph_order_sort by order. */
2092
2093 static int
cgraph_order_cmp(const void * a_p,const void * b_p)2094 cgraph_order_cmp (const void *a_p, const void *b_p)
2095 {
2096 const cgraph_order_sort *nodea = (const cgraph_order_sort *)a_p;
2097 const cgraph_order_sort *nodeb = (const cgraph_order_sort *)b_p;
2098
2099 return nodea->order - nodeb->order;
2100 }
2101
2102 /* Output all functions, variables, and asm statements in the order
2103 according to their order fields, which is the order in which they
2104 appeared in the file. This implements -fno-toplevel-reorder. In
2105 this mode we may output functions and variables which don't really
2106 need to be output. */
2107
2108 static void
output_in_order(void)2109 output_in_order (void)
2110 {
2111 int i;
2112 cgraph_node *cnode;
2113 varpool_node *vnode;
2114 asm_node *anode;
2115 auto_vec<cgraph_order_sort> nodes;
2116 cgraph_order_sort *node;
2117
2118 FOR_EACH_DEFINED_FUNCTION (cnode)
2119 if (cnode->process && !cnode->thunk
2120 && !cnode->alias && cnode->no_reorder)
2121 nodes.safe_push (cgraph_order_sort (cnode));
2122
2123 /* There is a similar loop in symbol_table::output_variables.
2124 Please keep them in sync. */
2125 FOR_EACH_VARIABLE (vnode)
2126 if (vnode->no_reorder
2127 && !DECL_HARD_REGISTER (vnode->decl)
2128 && !DECL_HAS_VALUE_EXPR_P (vnode->decl))
2129 nodes.safe_push (cgraph_order_sort (vnode));
2130
2131 for (anode = symtab->first_asm_symbol (); anode; anode = anode->next)
2132 nodes.safe_push (cgraph_order_sort (anode));
2133
2134 /* Sort nodes by order. */
2135 nodes.qsort (cgraph_order_cmp);
2136
2137 /* In toplevel reorder mode we output all statics; mark them as needed. */
2138 FOR_EACH_VEC_ELT (nodes, i, node)
2139 if (node->kind == ORDER_VAR)
2140 node->u.v->finalize_named_section_flags ();
2141
2142 FOR_EACH_VEC_ELT (nodes, i, node)
2143 node->process ();
2144
2145 symtab->clear_asm_symbols ();
2146 }
2147
2148 static void
ipa_passes(void)2149 ipa_passes (void)
2150 {
2151 gcc::pass_manager *passes = g->get_passes ();
2152
2153 set_cfun (NULL);
2154 current_function_decl = NULL;
2155 gimple_register_cfg_hooks ();
2156 bitmap_obstack_initialize (NULL);
2157
2158 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_START, NULL);
2159
2160 if (!in_lto_p)
2161 {
2162 execute_ipa_pass_list (passes->all_small_ipa_passes);
2163 if (seen_error ())
2164 return;
2165 }
2166
2167 /* This extra symtab_remove_unreachable_nodes pass tends to catch some
2168 devirtualization and other changes where removal iterate. */
2169 symtab->remove_unreachable_nodes (symtab->dump_file);
2170
2171 /* If pass_all_early_optimizations was not scheduled, the state of
2172 the cgraph will not be properly updated. Update it now. */
2173 if (symtab->state < IPA_SSA)
2174 symtab->state = IPA_SSA;
2175
2176 if (!in_lto_p)
2177 {
2178 /* Generate coverage variables and constructors. */
2179 coverage_finish ();
2180
2181 /* Process new functions added. */
2182 set_cfun (NULL);
2183 current_function_decl = NULL;
2184 symtab->process_new_functions ();
2185
2186 execute_ipa_summary_passes
2187 ((ipa_opt_pass_d *) passes->all_regular_ipa_passes);
2188 }
2189
2190 /* Some targets need to handle LTO assembler output specially. */
2191 if (flag_generate_lto || flag_generate_offload)
2192 targetm.asm_out.lto_start ();
2193
2194 if (!in_lto_p
2195 || flag_incremental_link == INCREMENTAL_LINK_LTO)
2196 {
2197 if (!quiet_flag)
2198 fprintf (stderr, "Streaming LTO\n");
2199 if (g->have_offload)
2200 {
2201 section_name_prefix = OFFLOAD_SECTION_NAME_PREFIX;
2202 lto_stream_offload_p = true;
2203 ipa_write_summaries ();
2204 lto_stream_offload_p = false;
2205 }
2206 if (flag_lto)
2207 {
2208 section_name_prefix = LTO_SECTION_NAME_PREFIX;
2209 lto_stream_offload_p = false;
2210 ipa_write_summaries ();
2211 }
2212 }
2213
2214 if (flag_generate_lto || flag_generate_offload)
2215 targetm.asm_out.lto_end ();
2216
2217 if (!flag_ltrans
2218 && ((in_lto_p && flag_incremental_link != INCREMENTAL_LINK_LTO)
2219 || !flag_lto || flag_fat_lto_objects))
2220 execute_ipa_pass_list (passes->all_regular_ipa_passes);
2221 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_END, NULL);
2222
2223 bitmap_obstack_release (NULL);
2224 }
2225
2226
2227 /* Weakrefs may be associated to external decls and thus not output
2228 at expansion time. Emit all necessary aliases. */
2229
2230 void
output_weakrefs(void)2231 symbol_table::output_weakrefs (void)
2232 {
2233 symtab_node *node;
2234 FOR_EACH_SYMBOL (node)
2235 if (node->alias
2236 && !TREE_ASM_WRITTEN (node->decl)
2237 && node->weakref)
2238 {
2239 tree target;
2240
2241 /* Weakrefs are special by not requiring target definition in current
2242 compilation unit. It is thus bit hard to work out what we want to
2243 alias.
2244 When alias target is defined, we need to fetch it from symtab reference,
2245 otherwise it is pointed to by alias_target. */
2246 if (node->alias_target)
2247 target = (DECL_P (node->alias_target)
2248 ? DECL_ASSEMBLER_NAME (node->alias_target)
2249 : node->alias_target);
2250 else if (node->analyzed)
2251 target = DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl);
2252 else
2253 gcc_unreachable ();
2254 do_assemble_alias (node->decl, target);
2255 }
2256 }
2257
2258 /* Perform simple optimizations based on callgraph. */
2259
2260 void
compile(void)2261 symbol_table::compile (void)
2262 {
2263 if (seen_error ())
2264 return;
2265
2266 symtab_node::checking_verify_symtab_nodes ();
2267
2268 timevar_push (TV_CGRAPHOPT);
2269 if (pre_ipa_mem_report)
2270 dump_memory_report ("Memory consumption before IPA");
2271 if (!quiet_flag)
2272 fprintf (stderr, "Performing interprocedural optimizations\n");
2273 state = IPA;
2274
2275 /* If LTO is enabled, initialize the streamer hooks needed by GIMPLE. */
2276 if (flag_generate_lto || flag_generate_offload)
2277 lto_streamer_hooks_init ();
2278
2279 /* Don't run the IPA passes if there was any error or sorry messages. */
2280 if (!seen_error ())
2281 {
2282 timevar_start (TV_CGRAPH_IPA_PASSES);
2283 ipa_passes ();
2284 timevar_stop (TV_CGRAPH_IPA_PASSES);
2285 }
2286 /* Do nothing else if any IPA pass found errors or if we are just streaming LTO. */
2287 if (seen_error ()
2288 || ((!in_lto_p || flag_incremental_link == INCREMENTAL_LINK_LTO)
2289 && flag_lto && !flag_fat_lto_objects))
2290 {
2291 timevar_pop (TV_CGRAPHOPT);
2292 return;
2293 }
2294
2295 global_info_ready = true;
2296 if (dump_file)
2297 {
2298 fprintf (dump_file, "Optimized ");
2299 symtab->dump (dump_file);
2300 }
2301 if (post_ipa_mem_report)
2302 dump_memory_report ("Memory consumption after IPA");
2303 timevar_pop (TV_CGRAPHOPT);
2304
2305 /* Output everything. */
2306 switch_to_section (text_section);
2307 (*debug_hooks->assembly_start) ();
2308 if (!quiet_flag)
2309 fprintf (stderr, "Assembling functions:\n");
2310 symtab_node::checking_verify_symtab_nodes ();
2311
2312 bitmap_obstack_initialize (NULL);
2313 execute_ipa_pass_list (g->get_passes ()->all_late_ipa_passes);
2314 bitmap_obstack_release (NULL);
2315 mark_functions_to_output ();
2316
2317 /* When weakref support is missing, we automatically translate all
2318 references to NODE to references to its ultimate alias target.
2319 The renaming mechanism uses flag IDENTIFIER_TRANSPARENT_ALIAS and
2320 TREE_CHAIN.
2321
2322 Set up this mapping before we output any assembler but once we are sure
2323 that all symbol renaming is done.
2324
2325 FIXME: All this ugliness can go away if we just do renaming at gimple
2326 level by physically rewriting the IL. At the moment we can only redirect
2327 calls, so we need infrastructure for renaming references as well. */
2328 #ifndef ASM_OUTPUT_WEAKREF
2329 symtab_node *node;
2330
2331 FOR_EACH_SYMBOL (node)
2332 if (node->alias
2333 && lookup_attribute ("weakref", DECL_ATTRIBUTES (node->decl)))
2334 {
2335 IDENTIFIER_TRANSPARENT_ALIAS
2336 (DECL_ASSEMBLER_NAME (node->decl)) = 1;
2337 TREE_CHAIN (DECL_ASSEMBLER_NAME (node->decl))
2338 = (node->alias_target ? node->alias_target
2339 : DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl));
2340 }
2341 #endif
2342
2343 state = EXPANSION;
2344
2345 /* Output first asm statements and anything ordered. The process
2346 flag is cleared for these nodes, so we skip them later. */
2347 output_in_order ();
2348
2349 timevar_start (TV_CGRAPH_FUNC_EXPANSION);
2350 expand_all_functions ();
2351 timevar_stop (TV_CGRAPH_FUNC_EXPANSION);
2352
2353 output_variables ();
2354
2355 process_new_functions ();
2356 state = FINISHED;
2357 output_weakrefs ();
2358
2359 if (dump_file)
2360 {
2361 fprintf (dump_file, "\nFinal ");
2362 symtab->dump (dump_file);
2363 }
2364 if (!flag_checking)
2365 return;
2366 symtab_node::verify_symtab_nodes ();
2367 /* Double check that all inline clones are gone and that all
2368 function bodies have been released from memory. */
2369 if (!seen_error ())
2370 {
2371 cgraph_node *node;
2372 bool error_found = false;
2373
2374 FOR_EACH_DEFINED_FUNCTION (node)
2375 if (node->inlined_to
2376 || gimple_has_body_p (node->decl))
2377 {
2378 error_found = true;
2379 node->debug ();
2380 }
2381 if (error_found)
2382 internal_error ("nodes with unreleased memory found");
2383 }
2384 }
2385
2386 /* Earlydebug dump file, flags, and number. */
2387
2388 static int debuginfo_early_dump_nr;
2389 static FILE *debuginfo_early_dump_file;
2390 static dump_flags_t debuginfo_early_dump_flags;
2391
2392 /* Debug dump file, flags, and number. */
2393
2394 static int debuginfo_dump_nr;
2395 static FILE *debuginfo_dump_file;
2396 static dump_flags_t debuginfo_dump_flags;
2397
2398 /* Register the debug and earlydebug dump files. */
2399
2400 void
debuginfo_early_init(void)2401 debuginfo_early_init (void)
2402 {
2403 gcc::dump_manager *dumps = g->get_dumps ();
2404 debuginfo_early_dump_nr = dumps->dump_register (".earlydebug", "earlydebug",
2405 "earlydebug", DK_tree,
2406 OPTGROUP_NONE,
2407 false);
2408 debuginfo_dump_nr = dumps->dump_register (".debug", "debug",
2409 "debug", DK_tree,
2410 OPTGROUP_NONE,
2411 false);
2412 }
2413
2414 /* Initialize the debug and earlydebug dump files. */
2415
2416 void
debuginfo_init(void)2417 debuginfo_init (void)
2418 {
2419 gcc::dump_manager *dumps = g->get_dumps ();
2420 debuginfo_dump_file = dump_begin (debuginfo_dump_nr, NULL);
2421 debuginfo_dump_flags = dumps->get_dump_file_info (debuginfo_dump_nr)->pflags;
2422 debuginfo_early_dump_file = dump_begin (debuginfo_early_dump_nr, NULL);
2423 debuginfo_early_dump_flags
2424 = dumps->get_dump_file_info (debuginfo_early_dump_nr)->pflags;
2425 }
2426
2427 /* Finalize the debug and earlydebug dump files. */
2428
2429 void
debuginfo_fini(void)2430 debuginfo_fini (void)
2431 {
2432 if (debuginfo_dump_file)
2433 dump_end (debuginfo_dump_nr, debuginfo_dump_file);
2434 if (debuginfo_early_dump_file)
2435 dump_end (debuginfo_early_dump_nr, debuginfo_early_dump_file);
2436 }
2437
2438 /* Set dump_file to the debug dump file. */
2439
2440 void
debuginfo_start(void)2441 debuginfo_start (void)
2442 {
2443 set_dump_file (debuginfo_dump_file);
2444 }
2445
2446 /* Undo setting dump_file to the debug dump file. */
2447
2448 void
debuginfo_stop(void)2449 debuginfo_stop (void)
2450 {
2451 set_dump_file (NULL);
2452 }
2453
2454 /* Set dump_file to the earlydebug dump file. */
2455
2456 void
debuginfo_early_start(void)2457 debuginfo_early_start (void)
2458 {
2459 set_dump_file (debuginfo_early_dump_file);
2460 }
2461
2462 /* Undo setting dump_file to the earlydebug dump file. */
2463
2464 void
debuginfo_early_stop(void)2465 debuginfo_early_stop (void)
2466 {
2467 set_dump_file (NULL);
2468 }
2469
2470 /* Analyze the whole compilation unit once it is parsed completely. */
2471
2472 void
finalize_compilation_unit(void)2473 symbol_table::finalize_compilation_unit (void)
2474 {
2475 timevar_push (TV_CGRAPH);
2476
2477 /* If we're here there's no current function anymore. Some frontends
2478 are lazy in clearing these. */
2479 current_function_decl = NULL;
2480 set_cfun (NULL);
2481
2482 /* Do not skip analyzing the functions if there were errors, we
2483 miss diagnostics for following functions otherwise. */
2484
2485 /* Emit size functions we didn't inline. */
2486 finalize_size_functions ();
2487
2488 /* Mark alias targets necessary and emit diagnostics. */
2489 handle_alias_pairs ();
2490
2491 if (!quiet_flag)
2492 {
2493 fprintf (stderr, "\nAnalyzing compilation unit\n");
2494 fflush (stderr);
2495 }
2496
2497 if (flag_dump_passes)
2498 dump_passes ();
2499
2500 /* Gimplify and lower all functions, compute reachability and
2501 remove unreachable nodes. */
2502 analyze_functions (/*first_time=*/true);
2503
2504 /* Mark alias targets necessary and emit diagnostics. */
2505 handle_alias_pairs ();
2506
2507 /* Gimplify and lower thunks. */
2508 analyze_functions (/*first_time=*/false);
2509
2510 /* All nested functions should be lowered now. */
2511 nested_function_info::release ();
2512
2513 /* Offloading requires LTO infrastructure. */
2514 if (!in_lto_p && g->have_offload)
2515 flag_generate_offload = 1;
2516
2517 if (!seen_error ())
2518 {
2519 /* Give the frontends the chance to emit early debug based on
2520 what is still reachable in the TU. */
2521 (*lang_hooks.finalize_early_debug) ();
2522
2523 /* Clean up anything that needs cleaning up after initial debug
2524 generation. */
2525 debuginfo_early_start ();
2526 (*debug_hooks->early_finish) (main_input_filename);
2527 debuginfo_early_stop ();
2528 }
2529
2530 /* Finally drive the pass manager. */
2531 compile ();
2532
2533 timevar_pop (TV_CGRAPH);
2534 }
2535
2536 /* Reset all state within cgraphunit.cc so that we can rerun the compiler
2537 within the same process. For use by toplev::finalize. */
2538
2539 void
cgraphunit_cc_finalize(void)2540 cgraphunit_cc_finalize (void)
2541 {
2542 gcc_assert (cgraph_new_nodes.length () == 0);
2543 cgraph_new_nodes.truncate (0);
2544
2545 queued_nodes = &symtab_terminator;
2546
2547 first_analyzed = NULL;
2548 first_analyzed_var = NULL;
2549 }
2550
2551 /* Creates a wrapper from cgraph_node to TARGET node. Thunk is used for this
2552 kind of wrapper method. */
2553
2554 void
create_wrapper(cgraph_node * target)2555 cgraph_node::create_wrapper (cgraph_node *target)
2556 {
2557 /* Preserve DECL_RESULT so we get right by reference flag. */
2558 tree decl_result = DECL_RESULT (decl);
2559
2560 /* Remove the function's body but keep arguments to be reused
2561 for thunk. */
2562 release_body (true);
2563 reset ();
2564
2565 DECL_UNINLINABLE (decl) = false;
2566 DECL_RESULT (decl) = decl_result;
2567 DECL_INITIAL (decl) = NULL;
2568 allocate_struct_function (decl, false);
2569 set_cfun (NULL);
2570
2571 /* Turn alias into thunk and expand it into GIMPLE representation. */
2572 definition = true;
2573 semantic_interposition = opt_for_fn (decl, flag_semantic_interposition);
2574
2575 /* Create empty thunk, but be sure we did not keep former thunk around.
2576 In that case we would need to preserve the info. */
2577 gcc_checking_assert (!thunk_info::get (this));
2578 thunk_info::get_create (this);
2579 thunk = true;
2580 create_edge (target, NULL, count);
2581 callees->can_throw_external = !TREE_NOTHROW (target->decl);
2582
2583 tree arguments = DECL_ARGUMENTS (decl);
2584
2585 while (arguments)
2586 {
2587 TREE_ADDRESSABLE (arguments) = false;
2588 arguments = TREE_CHAIN (arguments);
2589 }
2590
2591 expand_thunk (this, false, true);
2592 thunk_info::remove (this);
2593
2594 /* Inline summary set-up. */
2595 analyze ();
2596 inline_analyze_function (this);
2597 }
2598