2015-10-01 Steven G. Kargl <kargl@gcc.gnu.org>
[official-gcc.git] / gcc / cgraphunit.c
blob04a4d3f110088adcd86b50c60618a1a4725fa233
1 /* Driver of optimization process
2 Copyright (C) 2003-2015 Free Software Foundation, Inc.
3 Contributed by Jan Hubicka
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 3, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
21 /* This module implements main driver of compilation process.
23 The main scope of this file is to act as an interface in between
24 tree based frontends and the backend.
26 The front-end is supposed to use following functionality:
28 - finalize_function
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.
33 (There is one exception needed for implementing GCC extern inline
34 function.)
36 - varpool_finalize_decl
38 This function has same behavior as the above but is used for static
39 variables.
41 - add_asm_node
43 Insert new toplevel ASM statement
45 - finalize_compilation_unit
47 This function is called once (source level) compilation unit is finalized
48 and it will no longer change.
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.
55 At the end the bodies of unreachable functions are removed.
57 The function can be called multiple times when multiple source level
58 compilation units are combined.
60 - compile
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).
68 Compile time:
70 1) Inter-procedural optimization.
71 (ipa_passes)
73 This part is further split into:
75 a) early optimizations. These are local passes executed in
76 the topological order on the callgraph.
78 The purpose of early optimiations 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.
83 b) early small interprocedural passes.
85 Those are interprocedural passes executed only at compilation
86 time. These include, for example, transational memory lowering,
87 unreachable code removal and other simple transformations.
89 c) IP analysis stage. All interprocedural passes do their
90 analysis.
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.
98 d) LTO sreaming. When doing LTO, everything important gets
99 streamed into the object file.
101 Compile time and or linktime analysis stage (WPA):
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 multple object files.
112 Compile time and/or parallel linktime stage (ltrans)
114 Each of the object files is streamed back and compiled
115 separately. Now the function bodies becomes available
116 again.
118 2) Virtual clone materialization
119 (cgraph_materialize_clone)
121 IP passes can produce copies of existing functoins (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
128 All IP passes transform function bodies based on earlier
129 decision of the IP propagation.
131 4) late small IP passes
133 Simple IP passes working within single program partition.
135 5) Expansion
136 (expand_all_functions)
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.
144 Note that with -fno-toplevel-reorder passes 5 and 6
145 are combined together in cgraph_output_in_order.
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.
154 - cgraph_function_versioning
156 produces a copy of function into new one (a version)
157 and apply simple transformations
160 #include "config.h"
161 #include "system.h"
162 #include "coretypes.h"
163 #include "backend.h"
164 #include "cfghooks.h"
165 #include "tree.h"
166 #include "gimple.h"
167 #include "rtl.h"
168 #include "alias.h"
169 #include "fold-const.h"
170 #include "varasm.h"
171 #include "stor-layout.h"
172 #include "stringpool.h"
173 #include "gimple-ssa.h"
174 #include "output.h"
175 #include "cfgcleanup.h"
176 #include "internal-fn.h"
177 #include "gimple-fold.h"
178 #include "gimplify.h"
179 #include "gimple-iterator.h"
180 #include "gimplify-me.h"
181 #include "tree-cfg.h"
182 #include "tree-into-ssa.h"
183 #include "tree-ssa.h"
184 #include "tree-inline.h"
185 #include "langhooks.h"
186 #include "toplev.h"
187 #include "flags.h"
188 #include "debug.h"
189 #include "target.h"
190 #include "diagnostic.h"
191 #include "params.h"
192 #include "intl.h"
193 #include "cgraph.h"
194 #include "alloc-pool.h"
195 #include "symbol-summary.h"
196 #include "ipa-prop.h"
197 #include "tree-iterator.h"
198 #include "tree-pass.h"
199 #include "tree-dump.h"
200 #include "gimple-pretty-print.h"
201 #include "output.h"
202 #include "coverage.h"
203 #include "plugin.h"
204 #include "ipa-inline.h"
205 #include "ipa-utils.h"
206 #include "lto-streamer.h"
207 #include "except.h"
208 #include "cfgloop.h"
209 #include "regset.h" /* FIXME: For reg_obstack. */
210 #include "context.h"
211 #include "pass_manager.h"
212 #include "tree-nested.h"
213 #include "gimplify.h"
214 #include "dbgcnt.h"
215 #include "tree-chkp.h"
216 #include "lto-section-names.h"
217 #include "omp-low.h"
218 #include "print-tree.h"
220 /* Queue of cgraph nodes scheduled to be added into cgraph. This is a
221 secondary queue used during optimization to accommodate passes that
222 may generate new functions that need to be optimized and expanded. */
223 vec<cgraph_node *> cgraph_new_nodes;
225 static void expand_all_functions (void);
226 static void mark_functions_to_output (void);
227 static void handle_alias_pairs (void);
229 /* Used for vtable lookup in thunk adjusting. */
230 static GTY (()) tree vtable_entry_type;
232 /* Determine if symbol declaration is needed. That is, visible to something
233 either outside this translation unit, something magic in the system
234 configury */
235 bool
236 symtab_node::needed_p (void)
238 /* Double check that no one output the function into assembly file
239 early. */
240 gcc_checking_assert (!DECL_ASSEMBLER_NAME_SET_P (decl)
241 || !TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)));
243 if (!definition)
244 return false;
246 if (DECL_EXTERNAL (decl))
247 return false;
249 /* If the user told us it is used, then it must be so. */
250 if (force_output)
251 return true;
253 /* ABI forced symbols are needed when they are external. */
254 if (forced_by_abi && TREE_PUBLIC (decl))
255 return true;
257 /* Keep constructors, destructors and virtual functions. */
258 if (TREE_CODE (decl) == FUNCTION_DECL
259 && (DECL_STATIC_CONSTRUCTOR (decl) || DECL_STATIC_DESTRUCTOR (decl)))
260 return true;
262 /* Externally visible variables must be output. The exception is
263 COMDAT variables that must be output only when they are needed. */
264 if (TREE_PUBLIC (decl) && !DECL_COMDAT (decl))
265 return true;
267 return false;
270 /* Head and terminator of the queue of nodes to be processed while building
271 callgraph. */
273 static symtab_node symtab_terminator;
274 static symtab_node *queued_nodes = &symtab_terminator;
276 /* Add NODE to queue starting at QUEUED_NODES.
277 The queue is linked via AUX pointers and terminated by pointer to 1. */
279 static void
280 enqueue_node (symtab_node *node)
282 if (node->aux)
283 return;
284 gcc_checking_assert (queued_nodes);
285 node->aux = queued_nodes;
286 queued_nodes = node;
289 /* Process CGRAPH_NEW_FUNCTIONS and perform actions necessary to add these
290 functions into callgraph in a way so they look like ordinary reachable
291 functions inserted into callgraph already at construction time. */
293 void
294 symbol_table::process_new_functions (void)
296 tree fndecl;
298 if (!cgraph_new_nodes.exists ())
299 return;
301 handle_alias_pairs ();
302 /* Note that this queue may grow as its being processed, as the new
303 functions may generate new ones. */
304 for (unsigned i = 0; i < cgraph_new_nodes.length (); i++)
306 cgraph_node *node = cgraph_new_nodes[i];
307 fndecl = node->decl;
308 switch (state)
310 case CONSTRUCTION:
311 /* At construction time we just need to finalize function and move
312 it into reachable functions list. */
314 cgraph_node::finalize_function (fndecl, false);
315 call_cgraph_insertion_hooks (node);
316 enqueue_node (node);
317 break;
319 case IPA:
320 case IPA_SSA:
321 case IPA_SSA_AFTER_INLINING:
322 /* When IPA optimization already started, do all essential
323 transformations that has been already performed on the whole
324 cgraph but not on this function. */
326 gimple_register_cfg_hooks ();
327 if (!node->analyzed)
328 node->analyze ();
329 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
330 if ((state == IPA_SSA || state == IPA_SSA_AFTER_INLINING)
331 && !gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl)))
332 g->get_passes ()->execute_early_local_passes ();
333 else if (inline_summaries != NULL)
334 compute_inline_parameters (node, true);
335 free_dominance_info (CDI_POST_DOMINATORS);
336 free_dominance_info (CDI_DOMINATORS);
337 pop_cfun ();
338 call_cgraph_insertion_hooks (node);
339 break;
341 case EXPANSION:
342 /* Functions created during expansion shall be compiled
343 directly. */
344 node->process = 0;
345 call_cgraph_insertion_hooks (node);
346 node->expand ();
347 break;
349 default:
350 gcc_unreachable ();
351 break;
355 cgraph_new_nodes.release ();
358 /* As an GCC extension we allow redefinition of the function. The
359 semantics when both copies of bodies differ is not well defined.
360 We replace the old body with new body so in unit at a time mode
361 we always use new body, while in normal mode we may end up with
362 old body inlined into some functions and new body expanded and
363 inlined in others.
365 ??? It may make more sense to use one body for inlining and other
366 body for expanding the function but this is difficult to do. */
368 void
369 cgraph_node::reset (void)
371 /* If process is set, then we have already begun whole-unit analysis.
372 This is *not* testing for whether we've already emitted the function.
373 That case can be sort-of legitimately seen with real function redefinition
374 errors. I would argue that the front end should never present us with
375 such a case, but don't enforce that for now. */
376 gcc_assert (!process);
378 /* Reset our data structures so we can analyze the function again. */
379 memset (&local, 0, sizeof (local));
380 memset (&global, 0, sizeof (global));
381 memset (&rtl, 0, sizeof (rtl));
382 analyzed = false;
383 definition = false;
384 alias = false;
385 weakref = false;
386 cpp_implicit_alias = false;
388 remove_callees ();
389 remove_all_references ();
392 /* Return true when there are references to the node. INCLUDE_SELF is
393 true if a self reference counts as a reference. */
395 bool
396 symtab_node::referred_to_p (bool include_self)
398 ipa_ref *ref = NULL;
400 /* See if there are any references at all. */
401 if (iterate_referring (0, ref))
402 return true;
403 /* For functions check also calls. */
404 cgraph_node *cn = dyn_cast <cgraph_node *> (this);
405 if (cn && cn->callers)
407 if (include_self)
408 return true;
409 for (cgraph_edge *e = cn->callers; e; e = e->next_caller)
410 if (e->caller != this)
411 return true;
413 return false;
416 /* DECL has been parsed. Take it, queue it, compile it at the whim of the
417 logic in effect. If NO_COLLECT is true, then our caller cannot stand to have
418 the garbage collector run at the moment. We would need to either create
419 a new GC context, or just not compile right now. */
421 void
422 cgraph_node::finalize_function (tree decl, bool no_collect)
424 cgraph_node *node = cgraph_node::get_create (decl);
426 if (node->definition)
428 /* Nested functions should only be defined once. */
429 gcc_assert (!DECL_CONTEXT (decl)
430 || TREE_CODE (DECL_CONTEXT (decl)) != FUNCTION_DECL);
431 node->reset ();
432 node->local.redefined_extern_inline = true;
435 /* Set definition first before calling notice_global_symbol so that
436 it is available to notice_global_symbol. */
437 node->definition = true;
438 notice_global_symbol (decl);
439 node->lowered = DECL_STRUCT_FUNCTION (decl)->cfg != NULL;
441 /* With -fkeep-inline-functions we are keeping all inline functions except
442 for extern inline ones. */
443 if (flag_keep_inline_functions
444 && DECL_DECLARED_INLINE_P (decl)
445 && !DECL_EXTERNAL (decl)
446 && !DECL_DISREGARD_INLINE_LIMITS (decl))
447 node->force_output = 1;
449 /* When not optimizing, also output the static functions. (see
450 PR24561), but don't do so for always_inline functions, functions
451 declared inline and nested functions. These were optimized out
452 in the original implementation and it is unclear whether we want
453 to change the behavior here. */
454 if ((!opt_for_fn (decl, optimize)
455 && !node->cpp_implicit_alias
456 && !DECL_DISREGARD_INLINE_LIMITS (decl)
457 && !DECL_DECLARED_INLINE_P (decl)
458 && !(DECL_CONTEXT (decl)
459 && TREE_CODE (DECL_CONTEXT (decl)) == FUNCTION_DECL))
460 && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
461 node->force_output = 1;
463 /* If we've not yet emitted decl, tell the debug info about it. */
464 if (!TREE_ASM_WRITTEN (decl))
465 (*debug_hooks->deferred_inline_function) (decl);
467 if (!no_collect)
468 ggc_collect ();
470 if (symtab->state == CONSTRUCTION
471 && (node->needed_p () || node->referred_to_p ()))
472 enqueue_node (node);
475 /* Add the function FNDECL to the call graph.
476 Unlike finalize_function, this function is intended to be used
477 by middle end and allows insertion of new function at arbitrary point
478 of compilation. The function can be either in high, low or SSA form
479 GIMPLE.
481 The function is assumed to be reachable and have address taken (so no
482 API breaking optimizations are performed on it).
484 Main work done by this function is to enqueue the function for later
485 processing to avoid need the passes to be re-entrant. */
487 void
488 cgraph_node::add_new_function (tree fndecl, bool lowered)
490 gcc::pass_manager *passes = g->get_passes ();
491 cgraph_node *node;
493 if (dump_file)
495 struct function *fn = DECL_STRUCT_FUNCTION (fndecl);
496 const char *function_type = ((gimple_has_body_p (fndecl))
497 ? (lowered
498 ? (gimple_in_ssa_p (fn)
499 ? "ssa gimple"
500 : "low gimple")
501 : "high gimple")
502 : "to-be-gimplified");
503 fprintf (dump_file,
504 "Added new %s function %s to callgraph\n",
505 function_type,
506 fndecl_name (fndecl));
509 switch (symtab->state)
511 case PARSING:
512 cgraph_node::finalize_function (fndecl, false);
513 break;
514 case CONSTRUCTION:
515 /* Just enqueue function to be processed at nearest occurrence. */
516 node = cgraph_node::get_create (fndecl);
517 if (lowered)
518 node->lowered = true;
519 cgraph_new_nodes.safe_push (node);
520 break;
522 case IPA:
523 case IPA_SSA:
524 case IPA_SSA_AFTER_INLINING:
525 case EXPANSION:
526 /* Bring the function into finalized state and enqueue for later
527 analyzing and compilation. */
528 node = cgraph_node::get_create (fndecl);
529 node->local.local = false;
530 node->definition = true;
531 node->force_output = true;
532 if (!lowered && symtab->state == EXPANSION)
534 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
535 gimple_register_cfg_hooks ();
536 bitmap_obstack_initialize (NULL);
537 execute_pass_list (cfun, passes->all_lowering_passes);
538 passes->execute_early_local_passes ();
539 bitmap_obstack_release (NULL);
540 pop_cfun ();
542 lowered = true;
544 if (lowered)
545 node->lowered = true;
546 cgraph_new_nodes.safe_push (node);
547 break;
549 case FINISHED:
550 /* At the very end of compilation we have to do all the work up
551 to expansion. */
552 node = cgraph_node::create (fndecl);
553 if (lowered)
554 node->lowered = true;
555 node->definition = true;
556 node->analyze ();
557 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
558 gimple_register_cfg_hooks ();
559 bitmap_obstack_initialize (NULL);
560 if (!gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl)))
561 g->get_passes ()->execute_early_local_passes ();
562 bitmap_obstack_release (NULL);
563 pop_cfun ();
564 node->expand ();
565 break;
567 default:
568 gcc_unreachable ();
571 /* Set a personality if required and we already passed EH lowering. */
572 if (lowered
573 && (function_needs_eh_personality (DECL_STRUCT_FUNCTION (fndecl))
574 == eh_personality_lang))
575 DECL_FUNCTION_PERSONALITY (fndecl) = lang_hooks.eh_personality ();
578 /* Analyze the function scheduled to be output. */
579 void
580 cgraph_node::analyze (void)
582 tree decl = this->decl;
583 location_t saved_loc = input_location;
584 input_location = DECL_SOURCE_LOCATION (decl);
586 if (thunk.thunk_p)
588 cgraph_node *t = cgraph_node::get (thunk.alias);
590 create_edge (t, NULL, 0, CGRAPH_FREQ_BASE);
591 /* Target code in expand_thunk may need the thunk's target
592 to be analyzed, so recurse here. */
593 if (!t->analyzed)
594 t->analyze ();
595 if (t->alias)
597 t = t->get_alias_target ();
598 if (!t->analyzed)
599 t->analyze ();
601 if (!expand_thunk (false, false))
603 thunk.alias = NULL;
604 return;
606 thunk.alias = NULL;
608 if (alias)
609 resolve_alias (cgraph_node::get (alias_target));
610 else if (dispatcher_function)
612 /* Generate the dispatcher body of multi-versioned functions. */
613 cgraph_function_version_info *dispatcher_version_info
614 = function_version ();
615 if (dispatcher_version_info != NULL
616 && (dispatcher_version_info->dispatcher_resolver
617 == NULL_TREE))
619 tree resolver = NULL_TREE;
620 gcc_assert (targetm.generate_version_dispatcher_body);
621 resolver = targetm.generate_version_dispatcher_body (this);
622 gcc_assert (resolver != NULL_TREE);
625 else
627 push_cfun (DECL_STRUCT_FUNCTION (decl));
629 assign_assembler_name_if_neeeded (decl);
631 /* Make sure to gimplify bodies only once. During analyzing a
632 function we lower it, which will require gimplified nested
633 functions, so we can end up here with an already gimplified
634 body. */
635 if (!gimple_has_body_p (decl))
636 gimplify_function_tree (decl);
638 /* Lower the function. */
639 if (!lowered)
641 if (nested)
642 lower_nested_functions (decl);
643 gcc_assert (!nested);
645 gimple_register_cfg_hooks ();
646 bitmap_obstack_initialize (NULL);
647 execute_pass_list (cfun, g->get_passes ()->all_lowering_passes);
648 free_dominance_info (CDI_POST_DOMINATORS);
649 free_dominance_info (CDI_DOMINATORS);
650 compact_blocks ();
651 bitmap_obstack_release (NULL);
652 lowered = true;
655 pop_cfun ();
657 analyzed = true;
659 input_location = saved_loc;
662 /* C++ frontend produce same body aliases all over the place, even before PCH
663 gets streamed out. It relies on us linking the aliases with their function
664 in order to do the fixups, but ipa-ref is not PCH safe. Consequentely we
665 first produce aliases without links, but once C++ FE is sure he won't sream
666 PCH we build the links via this function. */
668 void
669 symbol_table::process_same_body_aliases (void)
671 symtab_node *node;
672 FOR_EACH_SYMBOL (node)
673 if (node->cpp_implicit_alias && !node->analyzed)
674 node->resolve_alias
675 (TREE_CODE (node->alias_target) == VAR_DECL
676 ? (symtab_node *)varpool_node::get_create (node->alias_target)
677 : (symtab_node *)cgraph_node::get_create (node->alias_target));
678 cpp_implicit_aliases_done = true;
681 /* Process attributes common for vars and functions. */
683 static void
684 process_common_attributes (symtab_node *node, tree decl)
686 tree weakref = lookup_attribute ("weakref", DECL_ATTRIBUTES (decl));
688 if (weakref && !lookup_attribute ("alias", DECL_ATTRIBUTES (decl)))
690 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
691 "%<weakref%> attribute should be accompanied with"
692 " an %<alias%> attribute");
693 DECL_WEAK (decl) = 0;
694 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
695 DECL_ATTRIBUTES (decl));
698 if (lookup_attribute ("no_reorder", DECL_ATTRIBUTES (decl)))
699 node->no_reorder = 1;
702 /* Look for externally_visible and used attributes and mark cgraph nodes
703 accordingly.
705 We cannot mark the nodes at the point the attributes are processed (in
706 handle_*_attribute) because the copy of the declarations available at that
707 point may not be canonical. For example, in:
709 void f();
710 void f() __attribute__((used));
712 the declaration we see in handle_used_attribute will be the second
713 declaration -- but the front end will subsequently merge that declaration
714 with the original declaration and discard the second declaration.
716 Furthermore, we can't mark these nodes in finalize_function because:
718 void f() {}
719 void f() __attribute__((externally_visible));
721 is valid.
723 So, we walk the nodes at the end of the translation unit, applying the
724 attributes at that point. */
726 static void
727 process_function_and_variable_attributes (cgraph_node *first,
728 varpool_node *first_var)
730 cgraph_node *node;
731 varpool_node *vnode;
733 for (node = symtab->first_function (); node != first;
734 node = symtab->next_function (node))
736 tree decl = node->decl;
737 if (DECL_PRESERVE_P (decl))
738 node->mark_force_output ();
739 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
741 if (! TREE_PUBLIC (node->decl))
742 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes,
743 "%<externally_visible%>"
744 " attribute have effect only on public objects");
746 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl))
747 && (node->definition && !node->alias))
749 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes,
750 "%<weakref%> attribute ignored"
751 " because function is defined");
752 DECL_WEAK (decl) = 0;
753 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
754 DECL_ATTRIBUTES (decl));
757 if (lookup_attribute ("always_inline", DECL_ATTRIBUTES (decl))
758 && !DECL_DECLARED_INLINE_P (decl)
759 /* redefining extern inline function makes it DECL_UNINLINABLE. */
760 && !DECL_UNINLINABLE (decl))
761 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
762 "always_inline function might not be inlinable");
764 process_common_attributes (node, decl);
766 for (vnode = symtab->first_variable (); vnode != first_var;
767 vnode = symtab->next_variable (vnode))
769 tree decl = vnode->decl;
770 if (DECL_EXTERNAL (decl)
771 && DECL_INITIAL (decl))
772 varpool_node::finalize_decl (decl);
773 if (DECL_PRESERVE_P (decl))
774 vnode->force_output = true;
775 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
777 if (! TREE_PUBLIC (vnode->decl))
778 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes,
779 "%<externally_visible%>"
780 " attribute have effect only on public objects");
782 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl))
783 && vnode->definition
784 && DECL_INITIAL (decl))
786 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes,
787 "%<weakref%> attribute ignored"
788 " because variable is initialized");
789 DECL_WEAK (decl) = 0;
790 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
791 DECL_ATTRIBUTES (decl));
793 process_common_attributes (vnode, decl);
797 /* Mark DECL as finalized. By finalizing the declaration, frontend instruct the
798 middle end to output the variable to asm file, if needed or externally
799 visible. */
801 void
802 varpool_node::finalize_decl (tree decl)
804 varpool_node *node = varpool_node::get_create (decl);
806 gcc_assert (TREE_STATIC (decl) || DECL_EXTERNAL (decl));
808 if (node->definition)
809 return;
810 /* Set definition first before calling notice_global_symbol so that
811 it is available to notice_global_symbol. */
812 node->definition = true;
813 notice_global_symbol (decl);
814 if (TREE_THIS_VOLATILE (decl) || DECL_PRESERVE_P (decl)
815 /* Traditionally we do not eliminate static variables when not
816 optimizing and when not doing toplevel reoder. */
817 || node->no_reorder
818 || ((!flag_toplevel_reorder
819 && !DECL_COMDAT (node->decl)
820 && !DECL_ARTIFICIAL (node->decl))))
821 node->force_output = true;
823 if (symtab->state == CONSTRUCTION
824 && (node->needed_p () || node->referred_to_p ()))
825 enqueue_node (node);
826 if (symtab->state >= IPA_SSA)
827 node->analyze ();
828 /* Some frontends produce various interface variables after compilation
829 finished. */
830 if (symtab->state == FINISHED
831 || (!flag_toplevel_reorder
832 && symtab->state == EXPANSION))
833 node->assemble_decl ();
835 if (DECL_INITIAL (decl))
836 chkp_register_var_initializer (decl);
839 /* EDGE is an polymorphic call. Mark all possible targets as reachable
840 and if there is only one target, perform trivial devirtualization.
841 REACHABLE_CALL_TARGETS collects target lists we already walked to
842 avoid udplicate work. */
844 static void
845 walk_polymorphic_call_targets (hash_set<void *> *reachable_call_targets,
846 cgraph_edge *edge)
848 unsigned int i;
849 void *cache_token;
850 bool final;
851 vec <cgraph_node *>targets
852 = possible_polymorphic_call_targets
853 (edge, &final, &cache_token);
855 if (!reachable_call_targets->add (cache_token))
857 if (symtab->dump_file)
858 dump_possible_polymorphic_call_targets
859 (symtab->dump_file, edge);
861 for (i = 0; i < targets.length (); i++)
863 /* Do not bother to mark virtual methods in anonymous namespace;
864 either we will find use of virtual table defining it, or it is
865 unused. */
866 if (targets[i]->definition
867 && TREE_CODE
868 (TREE_TYPE (targets[i]->decl))
869 == METHOD_TYPE
870 && !type_in_anonymous_namespace_p
871 (TYPE_METHOD_BASETYPE (TREE_TYPE (targets[i]->decl))))
872 enqueue_node (targets[i]);
876 /* Very trivial devirtualization; when the type is
877 final or anonymous (so we know all its derivation)
878 and there is only one possible virtual call target,
879 make the edge direct. */
880 if (final)
882 if (targets.length () <= 1 && dbg_cnt (devirt))
884 cgraph_node *target;
885 if (targets.length () == 1)
886 target = targets[0];
887 else
888 target = cgraph_node::create
889 (builtin_decl_implicit (BUILT_IN_UNREACHABLE));
891 if (symtab->dump_file)
893 fprintf (symtab->dump_file,
894 "Devirtualizing call: ");
895 print_gimple_stmt (symtab->dump_file,
896 edge->call_stmt, 0,
897 TDF_SLIM);
899 if (dump_enabled_p ())
901 location_t locus = gimple_location_safe (edge->call_stmt);
902 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, locus,
903 "devirtualizing call in %s to %s\n",
904 edge->caller->name (), target->name ());
907 edge->make_direct (target);
908 edge->redirect_call_stmt_to_callee ();
910 /* Call to __builtin_unreachable shouldn't be instrumented. */
911 if (!targets.length ())
912 gimple_call_set_with_bounds (edge->call_stmt, false);
914 if (symtab->dump_file)
916 fprintf (symtab->dump_file,
917 "Devirtualized as: ");
918 print_gimple_stmt (symtab->dump_file,
919 edge->call_stmt, 0,
920 TDF_SLIM);
926 /* Issue appropriate warnings for the global declaration DECL. */
928 static void
929 check_global_declaration (symtab_node *snode)
931 tree decl = snode->decl;
933 /* Warn about any function declared static but not defined. We don't
934 warn about variables, because many programs have static variables
935 that exist only to get some text into the object file. */
936 if (TREE_CODE (decl) == FUNCTION_DECL
937 && DECL_INITIAL (decl) == 0
938 && DECL_EXTERNAL (decl)
939 && ! DECL_ARTIFICIAL (decl)
940 && ! TREE_NO_WARNING (decl)
941 && ! TREE_PUBLIC (decl)
942 && (warn_unused_function
943 || snode->referred_to_p (/*include_self=*/false)))
945 if (snode->referred_to_p (/*include_self=*/false))
946 pedwarn (input_location, 0, "%q+F used but never defined", decl);
947 else
948 warning (OPT_Wunused_function, "%q+F declared %<static%> but never defined", decl);
949 /* This symbol is effectively an "extern" declaration now. */
950 TREE_PUBLIC (decl) = 1;
953 /* Warn about static fns or vars defined but not used. */
954 if (((warn_unused_function && TREE_CODE (decl) == FUNCTION_DECL)
955 || (((warn_unused_variable && ! TREE_READONLY (decl))
956 || (warn_unused_const_variable && TREE_READONLY (decl)))
957 && TREE_CODE (decl) == VAR_DECL))
958 && ! DECL_IN_SYSTEM_HEADER (decl)
959 && ! snode->referred_to_p (/*include_self=*/false)
960 /* This TREE_USED check is needed in addition to referred_to_p
961 above, because the `__unused__' attribute is not being
962 considered for referred_to_p. */
963 && ! TREE_USED (decl)
964 /* The TREE_USED bit for file-scope decls is kept in the identifier,
965 to handle multiple external decls in different scopes. */
966 && ! (DECL_NAME (decl) && TREE_USED (DECL_NAME (decl)))
967 && ! DECL_EXTERNAL (decl)
968 && ! DECL_ARTIFICIAL (decl)
969 && ! DECL_ABSTRACT_ORIGIN (decl)
970 && ! TREE_PUBLIC (decl)
971 /* A volatile variable might be used in some non-obvious way. */
972 && ! TREE_THIS_VOLATILE (decl)
973 /* Global register variables must be declared to reserve them. */
974 && ! (TREE_CODE (decl) == VAR_DECL && DECL_REGISTER (decl))
975 /* Global ctors and dtors are called by the runtime. */
976 && (TREE_CODE (decl) != FUNCTION_DECL
977 || (!DECL_STATIC_CONSTRUCTOR (decl)
978 && !DECL_STATIC_DESTRUCTOR (decl)))
979 /* Otherwise, ask the language. */
980 && lang_hooks.decls.warn_unused_global (decl))
981 warning_at (DECL_SOURCE_LOCATION (decl),
982 (TREE_CODE (decl) == FUNCTION_DECL)
983 ? OPT_Wunused_function
984 : (TREE_READONLY (decl)
985 ? OPT_Wunused_const_variable
986 : OPT_Wunused_variable),
987 "%qD defined but not used", decl);
990 /* Discover all functions and variables that are trivially needed, analyze
991 them as well as all functions and variables referred by them */
992 static cgraph_node *first_analyzed;
993 static varpool_node *first_analyzed_var;
995 /* FIRST_TIME is set to TRUE for the first time we are called for a
996 translation unit from finalize_compilation_unit() or false
997 otherwise. */
999 static void
1000 analyze_functions (bool first_time)
1002 /* Keep track of already processed nodes when called multiple times for
1003 intermodule optimization. */
1004 cgraph_node *first_handled = first_analyzed;
1005 varpool_node *first_handled_var = first_analyzed_var;
1006 hash_set<void *> reachable_call_targets;
1008 symtab_node *node;
1009 symtab_node *next;
1010 int i;
1011 ipa_ref *ref;
1012 bool changed = true;
1013 location_t saved_loc = input_location;
1015 bitmap_obstack_initialize (NULL);
1016 symtab->state = CONSTRUCTION;
1017 input_location = UNKNOWN_LOCATION;
1019 /* Ugly, but the fixup can not happen at a time same body alias is created;
1020 C++ FE is confused about the COMDAT groups being right. */
1021 if (symtab->cpp_implicit_aliases_done)
1022 FOR_EACH_SYMBOL (node)
1023 if (node->cpp_implicit_alias)
1024 node->fixup_same_cpp_alias_visibility (node->get_alias_target ());
1025 build_type_inheritance_graph ();
1027 /* Analysis adds static variables that in turn adds references to new functions.
1028 So we need to iterate the process until it stabilize. */
1029 while (changed)
1031 changed = false;
1032 process_function_and_variable_attributes (first_analyzed,
1033 first_analyzed_var);
1035 /* First identify the trivially needed symbols. */
1036 for (node = symtab->first_symbol ();
1037 node != first_analyzed
1038 && node != first_analyzed_var; node = node->next)
1040 /* Convert COMDAT group designators to IDENTIFIER_NODEs. */
1041 node->get_comdat_group_id ();
1042 if (node->needed_p ())
1044 enqueue_node (node);
1045 if (!changed && symtab->dump_file)
1046 fprintf (symtab->dump_file, "Trivially needed symbols:");
1047 changed = true;
1048 if (symtab->dump_file)
1049 fprintf (symtab->dump_file, " %s", node->asm_name ());
1050 if (!changed && symtab->dump_file)
1051 fprintf (symtab->dump_file, "\n");
1053 if (node == first_analyzed
1054 || node == first_analyzed_var)
1055 break;
1057 symtab->process_new_functions ();
1058 first_analyzed_var = symtab->first_variable ();
1059 first_analyzed = symtab->first_function ();
1061 if (changed && symtab->dump_file)
1062 fprintf (symtab->dump_file, "\n");
1064 /* Lower representation, build callgraph edges and references for all trivially
1065 needed symbols and all symbols referred by them. */
1066 while (queued_nodes != &symtab_terminator)
1068 changed = true;
1069 node = queued_nodes;
1070 queued_nodes = (symtab_node *)queued_nodes->aux;
1071 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
1072 if (cnode && cnode->definition)
1074 cgraph_edge *edge;
1075 tree decl = cnode->decl;
1077 /* ??? It is possible to create extern inline function
1078 and later using weak alias attribute to kill its body.
1079 See gcc.c-torture/compile/20011119-1.c */
1080 if (!DECL_STRUCT_FUNCTION (decl)
1081 && !cnode->alias
1082 && !cnode->thunk.thunk_p
1083 && !cnode->dispatcher_function)
1085 cnode->reset ();
1086 cnode->local.redefined_extern_inline = true;
1087 continue;
1090 if (!cnode->analyzed)
1091 cnode->analyze ();
1093 for (edge = cnode->callees; edge; edge = edge->next_callee)
1094 if (edge->callee->definition
1095 && (!DECL_EXTERNAL (edge->callee->decl)
1096 /* When not optimizing, do not try to analyze extern
1097 inline functions. Doing so is pointless. */
1098 || opt_for_fn (edge->callee->decl, optimize)
1099 /* Weakrefs needs to be preserved. */
1100 || edge->callee->alias
1101 /* always_inline functions are inlined aven at -O0. */
1102 || lookup_attribute
1103 ("always_inline",
1104 DECL_ATTRIBUTES (edge->callee->decl))
1105 /* Multiversioned functions needs the dispatcher to
1106 be produced locally even for extern functions. */
1107 || edge->callee->function_version ()))
1108 enqueue_node (edge->callee);
1109 if (opt_for_fn (cnode->decl, optimize)
1110 && opt_for_fn (cnode->decl, flag_devirtualize))
1112 cgraph_edge *next;
1114 for (edge = cnode->indirect_calls; edge; edge = next)
1116 next = edge->next_callee;
1117 if (edge->indirect_info->polymorphic)
1118 walk_polymorphic_call_targets (&reachable_call_targets,
1119 edge);
1123 /* If decl is a clone of an abstract function,
1124 mark that abstract function so that we don't release its body.
1125 The DECL_INITIAL() of that abstract function declaration
1126 will be later needed to output debug info. */
1127 if (DECL_ABSTRACT_ORIGIN (decl))
1129 cgraph_node *origin_node
1130 = cgraph_node::get_create (DECL_ABSTRACT_ORIGIN (decl));
1131 origin_node->used_as_abstract_origin = true;
1134 else
1136 varpool_node *vnode = dyn_cast <varpool_node *> (node);
1137 if (vnode && vnode->definition && !vnode->analyzed)
1138 vnode->analyze ();
1141 if (node->same_comdat_group)
1143 symtab_node *next;
1144 for (next = node->same_comdat_group;
1145 next != node;
1146 next = next->same_comdat_group)
1147 if (!next->comdat_local_p ())
1148 enqueue_node (next);
1150 for (i = 0; node->iterate_reference (i, ref); i++)
1151 if (ref->referred->definition
1152 && (!DECL_EXTERNAL (ref->referred->decl)
1153 || ((TREE_CODE (ref->referred->decl) != FUNCTION_DECL
1154 && optimize)
1155 || (TREE_CODE (ref->referred->decl) == FUNCTION_DECL
1156 && opt_for_fn (ref->referred->decl, optimize))
1157 || node->alias
1158 || ref->referred->alias)))
1159 enqueue_node (ref->referred);
1160 symtab->process_new_functions ();
1163 update_type_inheritance_graph ();
1165 /* Collect entry points to the unit. */
1166 if (symtab->dump_file)
1168 fprintf (symtab->dump_file, "\n\nInitial ");
1169 symtab_node::dump_table (symtab->dump_file);
1172 if (first_time)
1174 symtab_node *snode;
1175 FOR_EACH_SYMBOL (snode)
1176 check_global_declaration (snode);
1179 if (symtab->dump_file)
1180 fprintf (symtab->dump_file, "\nRemoving unused symbols:");
1182 for (node = symtab->first_symbol ();
1183 node != first_handled
1184 && node != first_handled_var; node = next)
1186 next = node->next;
1187 if (!node->aux && !node->referred_to_p ())
1189 if (symtab->dump_file)
1190 fprintf (symtab->dump_file, " %s", node->name ());
1192 /* See if the debugger can use anything before the DECL
1193 passes away. Perhaps it can notice a DECL that is now a
1194 constant and can tag the early DIE with an appropriate
1195 attribute.
1197 Otherwise, this is the last chance the debug_hooks have
1198 at looking at optimized away DECLs, since
1199 late_global_decl will subsequently be called from the
1200 contents of the now pruned symbol table. */
1201 if (!decl_function_context (node->decl))
1202 (*debug_hooks->late_global_decl) (node->decl);
1204 node->remove ();
1205 continue;
1207 if (cgraph_node *cnode = dyn_cast <cgraph_node *> (node))
1209 tree decl = node->decl;
1211 if (cnode->definition && !gimple_has_body_p (decl)
1212 && !cnode->alias
1213 && !cnode->thunk.thunk_p)
1214 cnode->reset ();
1216 gcc_assert (!cnode->definition || cnode->thunk.thunk_p
1217 || cnode->alias
1218 || gimple_has_body_p (decl));
1219 gcc_assert (cnode->analyzed == cnode->definition);
1221 node->aux = NULL;
1223 for (;node; node = node->next)
1224 node->aux = NULL;
1225 first_analyzed = symtab->first_function ();
1226 first_analyzed_var = symtab->first_variable ();
1227 if (symtab->dump_file)
1229 fprintf (symtab->dump_file, "\n\nReclaimed ");
1230 symtab_node::dump_table (symtab->dump_file);
1232 bitmap_obstack_release (NULL);
1233 ggc_collect ();
1234 /* Initialize assembler name hash, in particular we want to trigger C++
1235 mangling and same body alias creation before we free DECL_ARGUMENTS
1236 used by it. */
1237 if (!seen_error ())
1238 symtab->symtab_initialize_asm_name_hash ();
1240 input_location = saved_loc;
1243 /* Translate the ugly representation of aliases as alias pairs into nice
1244 representation in callgraph. We don't handle all cases yet,
1245 unfortunately. */
1247 static void
1248 handle_alias_pairs (void)
1250 alias_pair *p;
1251 unsigned i;
1253 for (i = 0; alias_pairs && alias_pairs->iterate (i, &p);)
1255 symtab_node *target_node = symtab_node::get_for_asmname (p->target);
1257 /* Weakrefs with target not defined in current unit are easy to handle:
1258 they behave just as external variables except we need to note the
1259 alias flag to later output the weakref pseudo op into asm file. */
1260 if (!target_node
1261 && lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl)) != NULL)
1263 symtab_node *node = symtab_node::get (p->decl);
1264 if (node)
1266 node->alias_target = p->target;
1267 node->weakref = true;
1268 node->alias = true;
1270 alias_pairs->unordered_remove (i);
1271 continue;
1273 else if (!target_node)
1275 error ("%q+D aliased to undefined symbol %qE", p->decl, p->target);
1276 symtab_node *node = symtab_node::get (p->decl);
1277 if (node)
1278 node->alias = false;
1279 alias_pairs->unordered_remove (i);
1280 continue;
1283 if (DECL_EXTERNAL (target_node->decl)
1284 /* We use local aliases for C++ thunks to force the tailcall
1285 to bind locally. This is a hack - to keep it working do
1286 the following (which is not strictly correct). */
1287 && (TREE_CODE (target_node->decl) != FUNCTION_DECL
1288 || ! DECL_VIRTUAL_P (target_node->decl))
1289 && ! lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl)))
1291 error ("%q+D aliased to external symbol %qE",
1292 p->decl, p->target);
1295 if (TREE_CODE (p->decl) == FUNCTION_DECL
1296 && target_node && is_a <cgraph_node *> (target_node))
1298 cgraph_node *src_node = cgraph_node::get (p->decl);
1299 if (src_node && src_node->definition)
1300 src_node->reset ();
1301 cgraph_node::create_alias (p->decl, target_node->decl);
1302 alias_pairs->unordered_remove (i);
1304 else if (TREE_CODE (p->decl) == VAR_DECL
1305 && target_node && is_a <varpool_node *> (target_node))
1307 varpool_node::create_alias (p->decl, target_node->decl);
1308 alias_pairs->unordered_remove (i);
1310 else
1312 error ("%q+D alias in between function and variable is not supported",
1313 p->decl);
1314 warning (0, "%q+D aliased declaration",
1315 target_node->decl);
1316 alias_pairs->unordered_remove (i);
1319 vec_free (alias_pairs);
1323 /* Figure out what functions we want to assemble. */
1325 static void
1326 mark_functions_to_output (void)
1328 cgraph_node *node;
1329 #ifdef ENABLE_CHECKING
1330 bool check_same_comdat_groups = false;
1332 FOR_EACH_FUNCTION (node)
1333 gcc_assert (!node->process);
1334 #endif
1336 FOR_EACH_FUNCTION (node)
1338 tree decl = node->decl;
1340 gcc_assert (!node->process || node->same_comdat_group);
1341 if (node->process)
1342 continue;
1344 /* We need to output all local functions that are used and not
1345 always inlined, as well as those that are reachable from
1346 outside the current compilation unit. */
1347 if (node->analyzed
1348 && !node->thunk.thunk_p
1349 && !node->alias
1350 && !node->global.inlined_to
1351 && !TREE_ASM_WRITTEN (decl)
1352 && !DECL_EXTERNAL (decl))
1354 node->process = 1;
1355 if (node->same_comdat_group)
1357 cgraph_node *next;
1358 for (next = dyn_cast<cgraph_node *> (node->same_comdat_group);
1359 next != node;
1360 next = dyn_cast<cgraph_node *> (next->same_comdat_group))
1361 if (!next->thunk.thunk_p && !next->alias
1362 && !next->comdat_local_p ())
1363 next->process = 1;
1366 else if (node->same_comdat_group)
1368 #ifdef ENABLE_CHECKING
1369 check_same_comdat_groups = true;
1370 #endif
1372 else
1374 /* We should've reclaimed all functions that are not needed. */
1375 #ifdef ENABLE_CHECKING
1376 if (!node->global.inlined_to
1377 && gimple_has_body_p (decl)
1378 /* FIXME: in ltrans unit when offline copy is outside partition but inline copies
1379 are inside partition, we can end up not removing the body since we no longer
1380 have analyzed node pointing to it. */
1381 && !node->in_other_partition
1382 && !node->alias
1383 && !node->clones
1384 && !DECL_EXTERNAL (decl))
1386 node->debug ();
1387 internal_error ("failed to reclaim unneeded function");
1389 #endif
1390 gcc_assert (node->global.inlined_to
1391 || !gimple_has_body_p (decl)
1392 || node->in_other_partition
1393 || node->clones
1394 || DECL_ARTIFICIAL (decl)
1395 || DECL_EXTERNAL (decl));
1400 #ifdef ENABLE_CHECKING
1401 if (check_same_comdat_groups)
1402 FOR_EACH_FUNCTION (node)
1403 if (node->same_comdat_group && !node->process)
1405 tree decl = node->decl;
1406 if (!node->global.inlined_to
1407 && gimple_has_body_p (decl)
1408 /* FIXME: in an ltrans unit when the offline copy is outside a
1409 partition but inline copies are inside a partition, we can
1410 end up not removing the body since we no longer have an
1411 analyzed node pointing to it. */
1412 && !node->in_other_partition
1413 && !node->clones
1414 && !DECL_EXTERNAL (decl))
1416 node->debug ();
1417 internal_error ("failed to reclaim unneeded function in same "
1418 "comdat group");
1421 #endif
1424 /* DECL is FUNCTION_DECL. Initialize datastructures so DECL is a function
1425 in lowered gimple form. IN_SSA is true if the gimple is in SSA.
1427 Set current_function_decl and cfun to newly constructed empty function body.
1428 return basic block in the function body. */
1430 basic_block
1431 init_lowered_empty_function (tree decl, bool in_ssa, gcov_type count)
1433 basic_block bb;
1434 edge e;
1436 current_function_decl = decl;
1437 allocate_struct_function (decl, false);
1438 gimple_register_cfg_hooks ();
1439 init_empty_tree_cfg ();
1441 if (in_ssa)
1443 init_tree_ssa (cfun);
1444 init_ssa_operands (cfun);
1445 cfun->gimple_df->in_ssa_p = true;
1446 cfun->curr_properties |= PROP_ssa;
1449 DECL_INITIAL (decl) = make_node (BLOCK);
1451 DECL_SAVED_TREE (decl) = error_mark_node;
1452 cfun->curr_properties |= (PROP_gimple_lcf | PROP_gimple_leh | PROP_gimple_any
1453 | PROP_cfg | PROP_loops);
1455 set_loops_for_fn (cfun, ggc_cleared_alloc<loops> ());
1456 init_loops_structure (cfun, loops_for_fn (cfun), 1);
1457 loops_for_fn (cfun)->state |= LOOPS_MAY_HAVE_MULTIPLE_LATCHES;
1459 /* Create BB for body of the function and connect it properly. */
1460 ENTRY_BLOCK_PTR_FOR_FN (cfun)->count = count;
1461 ENTRY_BLOCK_PTR_FOR_FN (cfun)->frequency = REG_BR_PROB_BASE;
1462 EXIT_BLOCK_PTR_FOR_FN (cfun)->count = count;
1463 EXIT_BLOCK_PTR_FOR_FN (cfun)->frequency = REG_BR_PROB_BASE;
1464 bb = create_basic_block (NULL, ENTRY_BLOCK_PTR_FOR_FN (cfun));
1465 bb->count = count;
1466 bb->frequency = BB_FREQ_MAX;
1467 e = make_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun), bb, EDGE_FALLTHRU);
1468 e->count = count;
1469 e->probability = REG_BR_PROB_BASE;
1470 e = make_edge (bb, EXIT_BLOCK_PTR_FOR_FN (cfun), 0);
1471 e->count = count;
1472 e->probability = REG_BR_PROB_BASE;
1473 add_bb_to_loop (bb, ENTRY_BLOCK_PTR_FOR_FN (cfun)->loop_father);
1475 return bb;
1478 /* Adjust PTR by the constant FIXED_OFFSET, and by the vtable
1479 offset indicated by VIRTUAL_OFFSET, if that is
1480 non-null. THIS_ADJUSTING is nonzero for a this adjusting thunk and
1481 zero for a result adjusting thunk. */
1483 static tree
1484 thunk_adjust (gimple_stmt_iterator * bsi,
1485 tree ptr, bool this_adjusting,
1486 HOST_WIDE_INT fixed_offset, tree virtual_offset)
1488 gassign *stmt;
1489 tree ret;
1491 if (this_adjusting
1492 && fixed_offset != 0)
1494 stmt = gimple_build_assign
1495 (ptr, fold_build_pointer_plus_hwi_loc (input_location,
1496 ptr,
1497 fixed_offset));
1498 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1501 /* If there's a virtual offset, look up that value in the vtable and
1502 adjust the pointer again. */
1503 if (virtual_offset)
1505 tree vtabletmp;
1506 tree vtabletmp2;
1507 tree vtabletmp3;
1509 if (!vtable_entry_type)
1511 tree vfunc_type = make_node (FUNCTION_TYPE);
1512 TREE_TYPE (vfunc_type) = integer_type_node;
1513 TYPE_ARG_TYPES (vfunc_type) = NULL_TREE;
1514 layout_type (vfunc_type);
1516 vtable_entry_type = build_pointer_type (vfunc_type);
1519 vtabletmp =
1520 create_tmp_reg (build_pointer_type
1521 (build_pointer_type (vtable_entry_type)), "vptr");
1523 /* The vptr is always at offset zero in the object. */
1524 stmt = gimple_build_assign (vtabletmp,
1525 build1 (NOP_EXPR, TREE_TYPE (vtabletmp),
1526 ptr));
1527 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1529 /* Form the vtable address. */
1530 vtabletmp2 = create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp)),
1531 "vtableaddr");
1532 stmt = gimple_build_assign (vtabletmp2,
1533 build_simple_mem_ref (vtabletmp));
1534 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1536 /* Find the entry with the vcall offset. */
1537 stmt = gimple_build_assign (vtabletmp2,
1538 fold_build_pointer_plus_loc (input_location,
1539 vtabletmp2,
1540 virtual_offset));
1541 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1543 /* Get the offset itself. */
1544 vtabletmp3 = create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp2)),
1545 "vcalloffset");
1546 stmt = gimple_build_assign (vtabletmp3,
1547 build_simple_mem_ref (vtabletmp2));
1548 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1550 /* Adjust the `this' pointer. */
1551 ptr = fold_build_pointer_plus_loc (input_location, ptr, vtabletmp3);
1552 ptr = force_gimple_operand_gsi (bsi, ptr, true, NULL_TREE, false,
1553 GSI_CONTINUE_LINKING);
1556 if (!this_adjusting
1557 && fixed_offset != 0)
1558 /* Adjust the pointer by the constant. */
1560 tree ptrtmp;
1562 if (TREE_CODE (ptr) == VAR_DECL)
1563 ptrtmp = ptr;
1564 else
1566 ptrtmp = create_tmp_reg (TREE_TYPE (ptr), "ptr");
1567 stmt = gimple_build_assign (ptrtmp, ptr);
1568 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1570 ptr = fold_build_pointer_plus_hwi_loc (input_location,
1571 ptrtmp, fixed_offset);
1574 /* Emit the statement and gimplify the adjustment expression. */
1575 ret = create_tmp_reg (TREE_TYPE (ptr), "adjusted_this");
1576 stmt = gimple_build_assign (ret, ptr);
1577 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1579 return ret;
1582 /* Expand thunk NODE to gimple if possible.
1583 When FORCE_GIMPLE_THUNK is true, gimple thunk is created and
1584 no assembler is produced.
1585 When OUTPUT_ASM_THUNK is true, also produce assembler for
1586 thunks that are not lowered. */
1588 bool
1589 cgraph_node::expand_thunk (bool output_asm_thunks, bool force_gimple_thunk)
1591 bool this_adjusting = thunk.this_adjusting;
1592 HOST_WIDE_INT fixed_offset = thunk.fixed_offset;
1593 HOST_WIDE_INT virtual_value = thunk.virtual_value;
1594 tree virtual_offset = NULL;
1595 tree alias = callees->callee->decl;
1596 tree thunk_fndecl = decl;
1597 tree a;
1599 /* Instrumentation thunk is the same function with
1600 a different signature. Never need to expand it. */
1601 if (thunk.add_pointer_bounds_args)
1602 return false;
1604 if (!force_gimple_thunk && this_adjusting
1605 && targetm.asm_out.can_output_mi_thunk (thunk_fndecl, fixed_offset,
1606 virtual_value, alias))
1608 const char *fnname;
1609 tree fn_block;
1610 tree restype = TREE_TYPE (TREE_TYPE (thunk_fndecl));
1612 if (!output_asm_thunks)
1614 analyzed = true;
1615 return false;
1618 if (in_lto_p)
1619 get_untransformed_body ();
1620 a = DECL_ARGUMENTS (thunk_fndecl);
1622 current_function_decl = thunk_fndecl;
1624 /* Ensure thunks are emitted in their correct sections. */
1625 resolve_unique_section (thunk_fndecl, 0,
1626 flag_function_sections);
1628 DECL_RESULT (thunk_fndecl)
1629 = build_decl (DECL_SOURCE_LOCATION (thunk_fndecl),
1630 RESULT_DECL, 0, restype);
1631 DECL_CONTEXT (DECL_RESULT (thunk_fndecl)) = thunk_fndecl;
1632 fnname = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (thunk_fndecl));
1634 /* The back end expects DECL_INITIAL to contain a BLOCK, so we
1635 create one. */
1636 fn_block = make_node (BLOCK);
1637 BLOCK_VARS (fn_block) = a;
1638 DECL_INITIAL (thunk_fndecl) = fn_block;
1639 init_function_start (thunk_fndecl);
1640 cfun->is_thunk = 1;
1641 insn_locations_init ();
1642 set_curr_insn_location (DECL_SOURCE_LOCATION (thunk_fndecl));
1643 prologue_location = curr_insn_location ();
1644 assemble_start_function (thunk_fndecl, fnname);
1646 targetm.asm_out.output_mi_thunk (asm_out_file, thunk_fndecl,
1647 fixed_offset, virtual_value, alias);
1649 assemble_end_function (thunk_fndecl, fnname);
1650 insn_locations_finalize ();
1651 init_insn_lengths ();
1652 free_after_compilation (cfun);
1653 set_cfun (NULL);
1654 TREE_ASM_WRITTEN (thunk_fndecl) = 1;
1655 thunk.thunk_p = false;
1656 analyzed = false;
1658 else if (stdarg_p (TREE_TYPE (thunk_fndecl)))
1660 error ("generic thunk code fails for method %qD which uses %<...%>",
1661 thunk_fndecl);
1662 TREE_ASM_WRITTEN (thunk_fndecl) = 1;
1663 analyzed = true;
1664 return false;
1666 else
1668 tree restype;
1669 basic_block bb, then_bb, else_bb, return_bb;
1670 gimple_stmt_iterator bsi;
1671 int nargs = 0;
1672 tree arg;
1673 int i;
1674 tree resdecl;
1675 tree restmp = NULL;
1676 tree resbnd = NULL;
1678 gcall *call;
1679 greturn *ret;
1680 bool alias_is_noreturn = TREE_THIS_VOLATILE (alias);
1682 if (in_lto_p)
1683 get_untransformed_body ();
1684 a = DECL_ARGUMENTS (thunk_fndecl);
1686 current_function_decl = thunk_fndecl;
1688 /* Ensure thunks are emitted in their correct sections. */
1689 resolve_unique_section (thunk_fndecl, 0,
1690 flag_function_sections);
1692 DECL_IGNORED_P (thunk_fndecl) = 1;
1693 bitmap_obstack_initialize (NULL);
1695 if (thunk.virtual_offset_p)
1696 virtual_offset = size_int (virtual_value);
1698 /* Build the return declaration for the function. */
1699 restype = TREE_TYPE (TREE_TYPE (thunk_fndecl));
1700 if (DECL_RESULT (thunk_fndecl) == NULL_TREE)
1702 resdecl = build_decl (input_location, RESULT_DECL, 0, restype);
1703 DECL_ARTIFICIAL (resdecl) = 1;
1704 DECL_IGNORED_P (resdecl) = 1;
1705 DECL_RESULT (thunk_fndecl) = resdecl;
1706 DECL_CONTEXT (DECL_RESULT (thunk_fndecl)) = thunk_fndecl;
1708 else
1709 resdecl = DECL_RESULT (thunk_fndecl);
1711 bb = then_bb = else_bb = return_bb
1712 = init_lowered_empty_function (thunk_fndecl, true, count);
1714 bsi = gsi_start_bb (bb);
1716 /* Build call to the function being thunked. */
1717 if (!VOID_TYPE_P (restype) && !alias_is_noreturn)
1719 if (DECL_BY_REFERENCE (resdecl))
1721 restmp = gimple_fold_indirect_ref (resdecl);
1722 if (!restmp)
1723 restmp = build2 (MEM_REF,
1724 TREE_TYPE (TREE_TYPE (DECL_RESULT (alias))),
1725 resdecl,
1726 build_int_cst (TREE_TYPE
1727 (DECL_RESULT (alias)), 0));
1729 else if (!is_gimple_reg_type (restype))
1731 if (aggregate_value_p (resdecl, TREE_TYPE (thunk_fndecl)))
1733 restmp = resdecl;
1735 if (TREE_CODE (restmp) == VAR_DECL)
1736 add_local_decl (cfun, restmp);
1737 BLOCK_VARS (DECL_INITIAL (current_function_decl)) = restmp;
1739 else
1740 restmp = create_tmp_var (restype, "retval");
1742 else
1743 restmp = create_tmp_reg (restype, "retval");
1746 for (arg = a; arg; arg = DECL_CHAIN (arg))
1747 nargs++;
1748 auto_vec<tree> vargs (nargs);
1749 i = 0;
1750 arg = a;
1751 if (this_adjusting)
1753 vargs.quick_push (thunk_adjust (&bsi, a, 1, fixed_offset,
1754 virtual_offset));
1755 arg = DECL_CHAIN (a);
1756 i = 1;
1759 if (nargs)
1760 for (; i < nargs; i++, arg = DECL_CHAIN (arg))
1762 tree tmp = arg;
1763 if (!is_gimple_val (arg))
1765 tmp = create_tmp_reg (TYPE_MAIN_VARIANT
1766 (TREE_TYPE (arg)), "arg");
1767 gimple *stmt = gimple_build_assign (tmp, arg);
1768 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT);
1770 vargs.quick_push (tmp);
1772 call = gimple_build_call_vec (build_fold_addr_expr_loc (0, alias), vargs);
1773 callees->call_stmt = call;
1774 gimple_call_set_from_thunk (call, true);
1775 gimple_call_set_with_bounds (call, instrumentation_clone);
1777 /* Return slot optimization is always possible and in fact requred to
1778 return values with DECL_BY_REFERENCE. */
1779 if (aggregate_value_p (resdecl, TREE_TYPE (thunk_fndecl))
1780 && (!is_gimple_reg_type (TREE_TYPE (resdecl))
1781 || DECL_BY_REFERENCE (resdecl)))
1782 gimple_call_set_return_slot_opt (call, true);
1784 if (restmp && !alias_is_noreturn)
1786 gimple_call_set_lhs (call, restmp);
1787 gcc_assert (useless_type_conversion_p (TREE_TYPE (restmp),
1788 TREE_TYPE (TREE_TYPE (alias))));
1790 gsi_insert_after (&bsi, call, GSI_NEW_STMT);
1791 if (!alias_is_noreturn)
1793 if (instrumentation_clone
1794 && !DECL_BY_REFERENCE (resdecl)
1795 && restmp
1796 && BOUNDED_P (restmp))
1798 resbnd = chkp_insert_retbnd_call (NULL, restmp, &bsi);
1799 create_edge (get_create (gimple_call_fndecl (gsi_stmt (bsi))),
1800 as_a <gcall *> (gsi_stmt (bsi)),
1801 callees->count, callees->frequency);
1804 if (restmp && !this_adjusting
1805 && (fixed_offset || virtual_offset))
1807 tree true_label = NULL_TREE;
1809 if (TREE_CODE (TREE_TYPE (restmp)) == POINTER_TYPE)
1811 gimple *stmt;
1812 edge e;
1813 /* If the return type is a pointer, we need to
1814 protect against NULL. We know there will be an
1815 adjustment, because that's why we're emitting a
1816 thunk. */
1817 then_bb = create_basic_block (NULL, bb);
1818 then_bb->count = count - count / 16;
1819 then_bb->frequency = BB_FREQ_MAX - BB_FREQ_MAX / 16;
1820 return_bb = create_basic_block (NULL, then_bb);
1821 return_bb->count = count;
1822 return_bb->frequency = BB_FREQ_MAX;
1823 else_bb = create_basic_block (NULL, else_bb);
1824 then_bb->count = count / 16;
1825 then_bb->frequency = BB_FREQ_MAX / 16;
1826 add_bb_to_loop (then_bb, bb->loop_father);
1827 add_bb_to_loop (return_bb, bb->loop_father);
1828 add_bb_to_loop (else_bb, bb->loop_father);
1829 remove_edge (single_succ_edge (bb));
1830 true_label = gimple_block_label (then_bb);
1831 stmt = gimple_build_cond (NE_EXPR, restmp,
1832 build_zero_cst (TREE_TYPE (restmp)),
1833 NULL_TREE, NULL_TREE);
1834 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT);
1835 e = make_edge (bb, then_bb, EDGE_TRUE_VALUE);
1836 e->probability = REG_BR_PROB_BASE - REG_BR_PROB_BASE / 16;
1837 e->count = count - count / 16;
1838 e = make_edge (bb, else_bb, EDGE_FALSE_VALUE);
1839 e->probability = REG_BR_PROB_BASE / 16;
1840 e->count = count / 16;
1841 e = make_edge (return_bb, EXIT_BLOCK_PTR_FOR_FN (cfun), 0);
1842 e->probability = REG_BR_PROB_BASE;
1843 e->count = count;
1844 e = make_edge (then_bb, return_bb, EDGE_FALLTHRU);
1845 e->probability = REG_BR_PROB_BASE;
1846 e->count = count - count / 16;
1847 e = make_edge (else_bb, return_bb, EDGE_FALLTHRU);
1848 e->probability = REG_BR_PROB_BASE;
1849 e->count = count / 16;
1850 bsi = gsi_last_bb (then_bb);
1853 restmp = thunk_adjust (&bsi, restmp, /*this_adjusting=*/0,
1854 fixed_offset, virtual_offset);
1855 if (true_label)
1857 gimple *stmt;
1858 bsi = gsi_last_bb (else_bb);
1859 stmt = gimple_build_assign (restmp,
1860 build_zero_cst (TREE_TYPE (restmp)));
1861 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT);
1862 bsi = gsi_last_bb (return_bb);
1865 else
1866 gimple_call_set_tail (call, true);
1868 /* Build return value. */
1869 if (!DECL_BY_REFERENCE (resdecl))
1870 ret = gimple_build_return (restmp);
1871 else
1872 ret = gimple_build_return (resdecl);
1873 gimple_return_set_retbnd (ret, resbnd);
1875 gsi_insert_after (&bsi, ret, GSI_NEW_STMT);
1877 else
1879 gimple_call_set_tail (call, true);
1880 remove_edge (single_succ_edge (bb));
1883 cfun->gimple_df->in_ssa_p = true;
1884 profile_status_for_fn (cfun)
1885 = count ? PROFILE_READ : PROFILE_GUESSED;
1886 /* FIXME: C++ FE should stop setting TREE_ASM_WRITTEN on thunks. */
1887 TREE_ASM_WRITTEN (thunk_fndecl) = false;
1888 delete_unreachable_blocks ();
1889 update_ssa (TODO_update_ssa);
1890 #ifdef ENABLE_CHECKING
1891 verify_flow_info ();
1892 #endif
1893 free_dominance_info (CDI_DOMINATORS);
1895 /* Since we want to emit the thunk, we explicitly mark its name as
1896 referenced. */
1897 thunk.thunk_p = false;
1898 lowered = true;
1899 bitmap_obstack_release (NULL);
1901 current_function_decl = NULL;
1902 set_cfun (NULL);
1903 return true;
1906 /* Assemble thunks and aliases associated to node. */
1908 void
1909 cgraph_node::assemble_thunks_and_aliases (void)
1911 cgraph_edge *e;
1912 ipa_ref *ref;
1914 for (e = callers; e;)
1915 if (e->caller->thunk.thunk_p
1916 && !e->caller->thunk.add_pointer_bounds_args)
1918 cgraph_node *thunk = e->caller;
1920 e = e->next_caller;
1921 thunk->expand_thunk (true, false);
1922 thunk->assemble_thunks_and_aliases ();
1924 else
1925 e = e->next_caller;
1927 FOR_EACH_ALIAS (this, ref)
1929 cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
1930 bool saved_written = TREE_ASM_WRITTEN (decl);
1932 /* Force assemble_alias to really output the alias this time instead
1933 of buffering it in same alias pairs. */
1934 TREE_ASM_WRITTEN (decl) = 1;
1935 do_assemble_alias (alias->decl,
1936 DECL_ASSEMBLER_NAME (decl));
1937 alias->assemble_thunks_and_aliases ();
1938 TREE_ASM_WRITTEN (decl) = saved_written;
1942 /* Expand function specified by node. */
1944 void
1945 cgraph_node::expand (void)
1947 location_t saved_loc;
1949 /* We ought to not compile any inline clones. */
1950 gcc_assert (!global.inlined_to);
1952 announce_function (decl);
1953 process = 0;
1954 gcc_assert (lowered);
1955 get_untransformed_body ();
1957 /* Generate RTL for the body of DECL. */
1959 timevar_push (TV_REST_OF_COMPILATION);
1961 gcc_assert (symtab->global_info_ready);
1963 /* Initialize the default bitmap obstack. */
1964 bitmap_obstack_initialize (NULL);
1966 /* Initialize the RTL code for the function. */
1967 current_function_decl = decl;
1968 saved_loc = input_location;
1969 input_location = DECL_SOURCE_LOCATION (decl);
1970 init_function_start (decl);
1972 gimple_register_cfg_hooks ();
1974 bitmap_obstack_initialize (&reg_obstack); /* FIXME, only at RTL generation*/
1976 execute_all_ipa_transforms ();
1978 /* Perform all tree transforms and optimizations. */
1980 /* Signal the start of passes. */
1981 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_START, NULL);
1983 execute_pass_list (cfun, g->get_passes ()->all_passes);
1985 /* Signal the end of passes. */
1986 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_END, NULL);
1988 bitmap_obstack_release (&reg_obstack);
1990 /* Release the default bitmap obstack. */
1991 bitmap_obstack_release (NULL);
1993 /* If requested, warn about function definitions where the function will
1994 return a value (usually of some struct or union type) which itself will
1995 take up a lot of stack space. */
1996 if (warn_larger_than && !DECL_EXTERNAL (decl) && TREE_TYPE (decl))
1998 tree ret_type = TREE_TYPE (TREE_TYPE (decl));
2000 if (ret_type && TYPE_SIZE_UNIT (ret_type)
2001 && TREE_CODE (TYPE_SIZE_UNIT (ret_type)) == INTEGER_CST
2002 && 0 < compare_tree_int (TYPE_SIZE_UNIT (ret_type),
2003 larger_than_size))
2005 unsigned int size_as_int
2006 = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (ret_type));
2008 if (compare_tree_int (TYPE_SIZE_UNIT (ret_type), size_as_int) == 0)
2009 warning (OPT_Wlarger_than_, "size of return value of %q+D is %u bytes",
2010 decl, size_as_int);
2011 else
2012 warning (OPT_Wlarger_than_, "size of return value of %q+D is larger than %wd bytes",
2013 decl, larger_than_size);
2017 gimple_set_body (decl, NULL);
2018 if (DECL_STRUCT_FUNCTION (decl) == 0
2019 && !cgraph_node::get (decl)->origin)
2021 /* Stop pointing to the local nodes about to be freed.
2022 But DECL_INITIAL must remain nonzero so we know this
2023 was an actual function definition.
2024 For a nested function, this is done in c_pop_function_context.
2025 If rest_of_compilation set this to 0, leave it 0. */
2026 if (DECL_INITIAL (decl) != 0)
2027 DECL_INITIAL (decl) = error_mark_node;
2030 input_location = saved_loc;
2032 ggc_collect ();
2033 timevar_pop (TV_REST_OF_COMPILATION);
2035 /* Make sure that BE didn't give up on compiling. */
2036 gcc_assert (TREE_ASM_WRITTEN (decl));
2037 set_cfun (NULL);
2038 current_function_decl = NULL;
2040 /* It would make a lot more sense to output thunks before function body to get more
2041 forward and lest backwarding jumps. This however would need solving problem
2042 with comdats. See PR48668. Also aliases must come after function itself to
2043 make one pass assemblers, like one on AIX, happy. See PR 50689.
2044 FIXME: Perhaps thunks should be move before function IFF they are not in comdat
2045 groups. */
2046 assemble_thunks_and_aliases ();
2047 release_body ();
2048 /* Eliminate all call edges. This is important so the GIMPLE_CALL no longer
2049 points to the dead function body. */
2050 remove_callees ();
2051 remove_all_references ();
2054 /* Node comparer that is responsible for the order that corresponds
2055 to time when a function was launched for the first time. */
2057 static int
2058 node_cmp (const void *pa, const void *pb)
2060 const cgraph_node *a = *(const cgraph_node * const *) pa;
2061 const cgraph_node *b = *(const cgraph_node * const *) pb;
2063 /* Functions with time profile must be before these without profile. */
2064 if (!a->tp_first_run || !b->tp_first_run)
2065 return a->tp_first_run - b->tp_first_run;
2067 return a->tp_first_run != b->tp_first_run
2068 ? b->tp_first_run - a->tp_first_run
2069 : b->order - a->order;
2072 /* Expand all functions that must be output.
2074 Attempt to topologically sort the nodes so function is output when
2075 all called functions are already assembled to allow data to be
2076 propagated across the callgraph. Use a stack to get smaller distance
2077 between a function and its callees (later we may choose to use a more
2078 sophisticated algorithm for function reordering; we will likely want
2079 to use subsections to make the output functions appear in top-down
2080 order). */
2082 static void
2083 expand_all_functions (void)
2085 cgraph_node *node;
2086 cgraph_node **order = XCNEWVEC (cgraph_node *,
2087 symtab->cgraph_count);
2088 unsigned int expanded_func_count = 0, profiled_func_count = 0;
2089 int order_pos, new_order_pos = 0;
2090 int i;
2092 order_pos = ipa_reverse_postorder (order);
2093 gcc_assert (order_pos == symtab->cgraph_count);
2095 /* Garbage collector may remove inline clones we eliminate during
2096 optimization. So we must be sure to not reference them. */
2097 for (i = 0; i < order_pos; i++)
2098 if (order[i]->process)
2099 order[new_order_pos++] = order[i];
2101 if (flag_profile_reorder_functions)
2102 qsort (order, new_order_pos, sizeof (cgraph_node *), node_cmp);
2104 for (i = new_order_pos - 1; i >= 0; i--)
2106 node = order[i];
2108 if (node->process)
2110 expanded_func_count++;
2111 if(node->tp_first_run)
2112 profiled_func_count++;
2114 if (symtab->dump_file)
2115 fprintf (symtab->dump_file,
2116 "Time profile order in expand_all_functions:%s:%d\n",
2117 node->asm_name (), node->tp_first_run);
2118 node->process = 0;
2119 node->expand ();
2123 if (dump_file)
2124 fprintf (dump_file, "Expanded functions with time profile (%s):%u/%u\n",
2125 main_input_filename, profiled_func_count, expanded_func_count);
2127 if (symtab->dump_file && flag_profile_reorder_functions)
2128 fprintf (symtab->dump_file, "Expanded functions with time profile:%u/%u\n",
2129 profiled_func_count, expanded_func_count);
2131 symtab->process_new_functions ();
2132 free_gimplify_stack ();
2134 free (order);
2137 /* This is used to sort the node types by the cgraph order number. */
2139 enum cgraph_order_sort_kind
2141 ORDER_UNDEFINED = 0,
2142 ORDER_FUNCTION,
2143 ORDER_VAR,
2144 ORDER_ASM
2147 struct cgraph_order_sort
2149 enum cgraph_order_sort_kind kind;
2150 union
2152 cgraph_node *f;
2153 varpool_node *v;
2154 asm_node *a;
2155 } u;
2158 /* Output all functions, variables, and asm statements in the order
2159 according to their order fields, which is the order in which they
2160 appeared in the file. This implements -fno-toplevel-reorder. In
2161 this mode we may output functions and variables which don't really
2162 need to be output.
2163 When NO_REORDER is true only do this for symbols marked no reorder. */
2165 static void
2166 output_in_order (bool no_reorder)
2168 int max;
2169 cgraph_order_sort *nodes;
2170 int i;
2171 cgraph_node *pf;
2172 varpool_node *pv;
2173 asm_node *pa;
2174 max = symtab->order;
2175 nodes = XCNEWVEC (cgraph_order_sort, max);
2177 FOR_EACH_DEFINED_FUNCTION (pf)
2179 if (pf->process && !pf->thunk.thunk_p && !pf->alias)
2181 if (no_reorder && !pf->no_reorder)
2182 continue;
2183 i = pf->order;
2184 gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
2185 nodes[i].kind = ORDER_FUNCTION;
2186 nodes[i].u.f = pf;
2190 FOR_EACH_DEFINED_VARIABLE (pv)
2191 if (!DECL_EXTERNAL (pv->decl))
2193 if (no_reorder && !pv->no_reorder)
2194 continue;
2195 i = pv->order;
2196 gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
2197 nodes[i].kind = ORDER_VAR;
2198 nodes[i].u.v = pv;
2201 for (pa = symtab->first_asm_symbol (); pa; pa = pa->next)
2203 i = pa->order;
2204 gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
2205 nodes[i].kind = ORDER_ASM;
2206 nodes[i].u.a = pa;
2209 /* In toplevel reorder mode we output all statics; mark them as needed. */
2211 for (i = 0; i < max; ++i)
2212 if (nodes[i].kind == ORDER_VAR)
2213 nodes[i].u.v->finalize_named_section_flags ();
2215 for (i = 0; i < max; ++i)
2217 switch (nodes[i].kind)
2219 case ORDER_FUNCTION:
2220 nodes[i].u.f->process = 0;
2221 nodes[i].u.f->expand ();
2222 break;
2224 case ORDER_VAR:
2225 nodes[i].u.v->assemble_decl ();
2226 break;
2228 case ORDER_ASM:
2229 assemble_asm (nodes[i].u.a->asm_str);
2230 break;
2232 case ORDER_UNDEFINED:
2233 break;
2235 default:
2236 gcc_unreachable ();
2240 symtab->clear_asm_symbols ();
2242 free (nodes);
2245 static void
2246 ipa_passes (void)
2248 gcc::pass_manager *passes = g->get_passes ();
2250 set_cfun (NULL);
2251 current_function_decl = NULL;
2252 gimple_register_cfg_hooks ();
2253 bitmap_obstack_initialize (NULL);
2255 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_START, NULL);
2257 if (!in_lto_p)
2259 execute_ipa_pass_list (passes->all_small_ipa_passes);
2260 if (seen_error ())
2261 return;
2264 /* This extra symtab_remove_unreachable_nodes pass tends to catch some
2265 devirtualization and other changes where removal iterate. */
2266 symtab->remove_unreachable_nodes (symtab->dump_file);
2268 /* If pass_all_early_optimizations was not scheduled, the state of
2269 the cgraph will not be properly updated. Update it now. */
2270 if (symtab->state < IPA_SSA)
2271 symtab->state = IPA_SSA;
2273 if (!in_lto_p)
2275 /* Generate coverage variables and constructors. */
2276 coverage_finish ();
2278 /* Process new functions added. */
2279 set_cfun (NULL);
2280 current_function_decl = NULL;
2281 symtab->process_new_functions ();
2283 execute_ipa_summary_passes
2284 ((ipa_opt_pass_d *) passes->all_regular_ipa_passes);
2287 /* Some targets need to handle LTO assembler output specially. */
2288 if (flag_generate_lto || flag_generate_offload)
2289 targetm.asm_out.lto_start ();
2291 if (!in_lto_p)
2293 if (g->have_offload)
2295 section_name_prefix = OFFLOAD_SECTION_NAME_PREFIX;
2296 lto_stream_offload_p = true;
2297 ipa_write_summaries ();
2298 lto_stream_offload_p = false;
2300 if (flag_lto)
2302 section_name_prefix = LTO_SECTION_NAME_PREFIX;
2303 lto_stream_offload_p = false;
2304 ipa_write_summaries ();
2308 if (flag_generate_lto || flag_generate_offload)
2309 targetm.asm_out.lto_end ();
2311 if (!flag_ltrans && (in_lto_p || !flag_lto || flag_fat_lto_objects))
2312 execute_ipa_pass_list (passes->all_regular_ipa_passes);
2313 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_END, NULL);
2315 bitmap_obstack_release (NULL);
2319 /* Return string alias is alias of. */
2321 static tree
2322 get_alias_symbol (tree decl)
2324 tree alias = lookup_attribute ("alias", DECL_ATTRIBUTES (decl));
2325 return get_identifier (TREE_STRING_POINTER
2326 (TREE_VALUE (TREE_VALUE (alias))));
2330 /* Weakrefs may be associated to external decls and thus not output
2331 at expansion time. Emit all necessary aliases. */
2333 void
2334 symbol_table::output_weakrefs (void)
2336 symtab_node *node;
2337 cgraph_node *cnode;
2338 FOR_EACH_SYMBOL (node)
2339 if (node->alias
2340 && !TREE_ASM_WRITTEN (node->decl)
2341 && (!(cnode = dyn_cast <cgraph_node *> (node))
2342 || !cnode->instrumented_version
2343 || !TREE_ASM_WRITTEN (cnode->instrumented_version->decl))
2344 && node->weakref)
2346 tree target;
2348 /* Weakrefs are special by not requiring target definition in current
2349 compilation unit. It is thus bit hard to work out what we want to
2350 alias.
2351 When alias target is defined, we need to fetch it from symtab reference,
2352 otherwise it is pointed to by alias_target. */
2353 if (node->alias_target)
2354 target = (DECL_P (node->alias_target)
2355 ? DECL_ASSEMBLER_NAME (node->alias_target)
2356 : node->alias_target);
2357 else if (node->analyzed)
2358 target = DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl);
2359 else
2361 gcc_unreachable ();
2362 target = get_alias_symbol (node->decl);
2364 do_assemble_alias (node->decl, target);
2368 /* Perform simple optimizations based on callgraph. */
2370 void
2371 symbol_table::compile (void)
2373 if (seen_error ())
2374 return;
2376 #ifdef ENABLE_CHECKING
2377 symtab_node::verify_symtab_nodes ();
2378 #endif
2380 timevar_push (TV_CGRAPHOPT);
2381 if (pre_ipa_mem_report)
2383 fprintf (stderr, "Memory consumption before IPA\n");
2384 dump_memory_report (false);
2386 if (!quiet_flag)
2387 fprintf (stderr, "Performing interprocedural optimizations\n");
2388 state = IPA;
2390 /* Offloading requires LTO infrastructure. */
2391 if (!in_lto_p && g->have_offload)
2392 flag_generate_offload = 1;
2394 /* If LTO is enabled, initialize the streamer hooks needed by GIMPLE. */
2395 if (flag_generate_lto || flag_generate_offload)
2396 lto_streamer_hooks_init ();
2398 /* Don't run the IPA passes if there was any error or sorry messages. */
2399 if (!seen_error ())
2400 ipa_passes ();
2402 /* Do nothing else if any IPA pass found errors or if we are just streaming LTO. */
2403 if (seen_error ()
2404 || (!in_lto_p && flag_lto && !flag_fat_lto_objects))
2406 timevar_pop (TV_CGRAPHOPT);
2407 return;
2410 global_info_ready = true;
2411 if (dump_file)
2413 fprintf (dump_file, "Optimized ");
2414 symtab_node:: dump_table (dump_file);
2416 if (post_ipa_mem_report)
2418 fprintf (stderr, "Memory consumption after IPA\n");
2419 dump_memory_report (false);
2421 timevar_pop (TV_CGRAPHOPT);
2423 /* Output everything. */
2424 (*debug_hooks->assembly_start) ();
2425 if (!quiet_flag)
2426 fprintf (stderr, "Assembling functions:\n");
2427 #ifdef ENABLE_CHECKING
2428 symtab_node::verify_symtab_nodes ();
2429 #endif
2431 materialize_all_clones ();
2432 bitmap_obstack_initialize (NULL);
2433 execute_ipa_pass_list (g->get_passes ()->all_late_ipa_passes);
2434 bitmap_obstack_release (NULL);
2435 mark_functions_to_output ();
2437 /* When weakref support is missing, we autmatically translate all
2438 references to NODE to references to its ultimate alias target.
2439 The renaming mechanizm uses flag IDENTIFIER_TRANSPARENT_ALIAS and
2440 TREE_CHAIN.
2442 Set up this mapping before we output any assembler but once we are sure
2443 that all symbol renaming is done.
2445 FIXME: All this uglyness can go away if we just do renaming at gimple
2446 level by physically rewritting the IL. At the moment we can only redirect
2447 calls, so we need infrastructure for renaming references as well. */
2448 #ifndef ASM_OUTPUT_WEAKREF
2449 symtab_node *node;
2451 FOR_EACH_SYMBOL (node)
2452 if (node->alias
2453 && lookup_attribute ("weakref", DECL_ATTRIBUTES (node->decl)))
2455 IDENTIFIER_TRANSPARENT_ALIAS
2456 (DECL_ASSEMBLER_NAME (node->decl)) = 1;
2457 TREE_CHAIN (DECL_ASSEMBLER_NAME (node->decl))
2458 = (node->alias_target ? node->alias_target
2459 : DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl));
2461 #endif
2463 state = EXPANSION;
2465 if (!flag_toplevel_reorder)
2466 output_in_order (false);
2467 else
2469 /* Output first asm statements and anything ordered. The process
2470 flag is cleared for these nodes, so we skip them later. */
2471 output_in_order (true);
2472 expand_all_functions ();
2473 output_variables ();
2476 process_new_functions ();
2477 state = FINISHED;
2478 output_weakrefs ();
2480 if (dump_file)
2482 fprintf (dump_file, "\nFinal ");
2483 symtab_node::dump_table (dump_file);
2485 #ifdef ENABLE_CHECKING
2486 symtab_node::verify_symtab_nodes ();
2487 /* Double check that all inline clones are gone and that all
2488 function bodies have been released from memory. */
2489 if (!seen_error ())
2491 cgraph_node *node;
2492 bool error_found = false;
2494 FOR_EACH_DEFINED_FUNCTION (node)
2495 if (node->global.inlined_to
2496 || gimple_has_body_p (node->decl))
2498 error_found = true;
2499 node->debug ();
2501 if (error_found)
2502 internal_error ("nodes with unreleased memory found");
2504 #endif
2508 /* Analyze the whole compilation unit once it is parsed completely. */
2510 void
2511 symbol_table::finalize_compilation_unit (void)
2513 timevar_push (TV_CGRAPH);
2515 /* If we're here there's no current function anymore. Some frontends
2516 are lazy in clearing these. */
2517 current_function_decl = NULL;
2518 set_cfun (NULL);
2520 /* Do not skip analyzing the functions if there were errors, we
2521 miss diagnostics for following functions otherwise. */
2523 /* Emit size functions we didn't inline. */
2524 finalize_size_functions ();
2526 /* Mark alias targets necessary and emit diagnostics. */
2527 handle_alias_pairs ();
2529 if (!quiet_flag)
2531 fprintf (stderr, "\nAnalyzing compilation unit\n");
2532 fflush (stderr);
2535 if (flag_dump_passes)
2536 dump_passes ();
2538 /* Gimplify and lower all functions, compute reachability and
2539 remove unreachable nodes. */
2540 analyze_functions (/*first_time=*/true);
2542 /* Mark alias targets necessary and emit diagnostics. */
2543 handle_alias_pairs ();
2545 /* Gimplify and lower thunks. */
2546 analyze_functions (/*first_time=*/false);
2548 if (!seen_error ())
2550 /* Emit early debug for reachable functions, and by consequence,
2551 locally scoped symbols. */
2552 struct cgraph_node *cnode;
2553 FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (cnode)
2554 (*debug_hooks->early_global_decl) (cnode->decl);
2556 /* Clean up anything that needs cleaning up after initial debug
2557 generation. */
2558 (*debug_hooks->early_finish) ();
2561 /* Finally drive the pass manager. */
2562 compile ();
2564 timevar_pop (TV_CGRAPH);
2567 /* Reset all state within cgraphunit.c so that we can rerun the compiler
2568 within the same process. For use by toplev::finalize. */
2570 void
2571 cgraphunit_c_finalize (void)
2573 gcc_assert (cgraph_new_nodes.length () == 0);
2574 cgraph_new_nodes.truncate (0);
2576 vtable_entry_type = NULL;
2577 queued_nodes = &symtab_terminator;
2579 first_analyzed = NULL;
2580 first_analyzed_var = NULL;
2583 /* Creates a wrapper from cgraph_node to TARGET node. Thunk is used for this
2584 kind of wrapper method. */
2586 void
2587 cgraph_node::create_wrapper (cgraph_node *target)
2589 /* Preserve DECL_RESULT so we get right by reference flag. */
2590 tree decl_result = DECL_RESULT (decl);
2592 /* Remove the function's body but keep arguments to be reused
2593 for thunk. */
2594 release_body (true);
2595 reset ();
2597 DECL_UNINLINABLE (decl) = false;
2598 DECL_RESULT (decl) = decl_result;
2599 DECL_INITIAL (decl) = NULL;
2600 allocate_struct_function (decl, false);
2601 set_cfun (NULL);
2603 /* Turn alias into thunk and expand it into GIMPLE representation. */
2604 definition = true;
2606 memset (&thunk, 0, sizeof (cgraph_thunk_info));
2607 thunk.thunk_p = true;
2608 create_edge (target, NULL, count, CGRAPH_FREQ_BASE);
2609 callees->can_throw_external = !TREE_NOTHROW (target->decl);
2611 tree arguments = DECL_ARGUMENTS (decl);
2613 while (arguments)
2615 TREE_ADDRESSABLE (arguments) = false;
2616 arguments = TREE_CHAIN (arguments);
2619 expand_thunk (false, true);
2621 /* Inline summary set-up. */
2622 analyze ();
2623 inline_analyze_function (this);
2626 #include "gt-cgraphunit.h"