2015-05-05 Yvan Roux <yvan.roux@linaro.org>
[official-gcc.git] / gcc / cgraphunit.c
blob0a50f61a384e2a939f024a66562c3373e4074de1
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 "tm.h"
164 #include "hash-set.h"
165 #include "machmode.h"
166 #include "vec.h"
167 #include "double-int.h"
168 #include "input.h"
169 #include "alias.h"
170 #include "symtab.h"
171 #include "wide-int.h"
172 #include "inchash.h"
173 #include "tree.h"
174 #include "fold-const.h"
175 #include "varasm.h"
176 #include "stor-layout.h"
177 #include "stringpool.h"
178 #include "output.h"
179 #include "rtl.h"
180 #include "predict.h"
181 #include "hard-reg-set.h"
182 #include "input.h"
183 #include "function.h"
184 #include "basic-block.h"
185 #include "tree-ssa-alias.h"
186 #include "internal-fn.h"
187 #include "gimple-fold.h"
188 #include "gimple-expr.h"
189 #include "is-a.h"
190 #include "gimple.h"
191 #include "gimplify.h"
192 #include "gimple-iterator.h"
193 #include "gimplify-me.h"
194 #include "gimple-ssa.h"
195 #include "tree-cfg.h"
196 #include "tree-into-ssa.h"
197 #include "tree-ssa.h"
198 #include "tree-inline.h"
199 #include "langhooks.h"
200 #include "toplev.h"
201 #include "flags.h"
202 #include "debug.h"
203 #include "target.h"
204 #include "diagnostic.h"
205 #include "params.h"
206 #include "intl.h"
207 #include "hash-map.h"
208 #include "plugin-api.h"
209 #include "ipa-ref.h"
210 #include "cgraph.h"
211 #include "alloc-pool.h"
212 #include "symbol-summary.h"
213 #include "ipa-prop.h"
214 #include "tree-iterator.h"
215 #include "tree-pass.h"
216 #include "tree-dump.h"
217 #include "gimple-pretty-print.h"
218 #include "output.h"
219 #include "coverage.h"
220 #include "plugin.h"
221 #include "ipa-inline.h"
222 #include "ipa-utils.h"
223 #include "lto-streamer.h"
224 #include "except.h"
225 #include "cfgloop.h"
226 #include "regset.h" /* FIXME: For reg_obstack. */
227 #include "context.h"
228 #include "pass_manager.h"
229 #include "tree-nested.h"
230 #include "gimplify.h"
231 #include "dbgcnt.h"
232 #include "tree-chkp.h"
233 #include "lto-section-names.h"
234 #include "omp-low.h"
235 #include "print-tree.h"
237 /* Queue of cgraph nodes scheduled to be added into cgraph. This is a
238 secondary queue used during optimization to accommodate passes that
239 may generate new functions that need to be optimized and expanded. */
240 vec<cgraph_node *> cgraph_new_nodes;
242 static void expand_all_functions (void);
243 static void mark_functions_to_output (void);
244 static void handle_alias_pairs (void);
246 /* Used for vtable lookup in thunk adjusting. */
247 static GTY (()) tree vtable_entry_type;
249 /* Determine if symbol declaration is needed. That is, visible to something
250 either outside this translation unit, something magic in the system
251 configury */
252 bool
253 symtab_node::needed_p (void)
255 /* Double check that no one output the function into assembly file
256 early. */
257 gcc_checking_assert (!DECL_ASSEMBLER_NAME_SET_P (decl)
258 || !TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)));
260 if (!definition)
261 return false;
263 if (DECL_EXTERNAL (decl))
264 return false;
266 /* If the user told us it is used, then it must be so. */
267 if (force_output)
268 return true;
270 /* ABI forced symbols are needed when they are external. */
271 if (forced_by_abi && TREE_PUBLIC (decl))
272 return true;
274 /* Keep constructors, destructors and virtual functions. */
275 if (TREE_CODE (decl) == FUNCTION_DECL
276 && (DECL_STATIC_CONSTRUCTOR (decl) || DECL_STATIC_DESTRUCTOR (decl)))
277 return true;
279 /* Externally visible variables must be output. The exception is
280 COMDAT variables that must be output only when they are needed. */
281 if (TREE_PUBLIC (decl) && !DECL_COMDAT (decl))
282 return true;
284 return false;
287 /* Head and terminator of the queue of nodes to be processed while building
288 callgraph. */
290 static symtab_node symtab_terminator;
291 static symtab_node *queued_nodes = &symtab_terminator;
293 /* Add NODE to queue starting at QUEUED_NODES.
294 The queue is linked via AUX pointers and terminated by pointer to 1. */
296 static void
297 enqueue_node (symtab_node *node)
299 if (node->aux)
300 return;
301 gcc_checking_assert (queued_nodes);
302 node->aux = queued_nodes;
303 queued_nodes = node;
306 /* Process CGRAPH_NEW_FUNCTIONS and perform actions necessary to add these
307 functions into callgraph in a way so they look like ordinary reachable
308 functions inserted into callgraph already at construction time. */
310 void
311 symbol_table::process_new_functions (void)
313 tree fndecl;
315 if (!cgraph_new_nodes.exists ())
316 return;
318 handle_alias_pairs ();
319 /* Note that this queue may grow as its being processed, as the new
320 functions may generate new ones. */
321 for (unsigned i = 0; i < cgraph_new_nodes.length (); i++)
323 cgraph_node *node = cgraph_new_nodes[i];
324 fndecl = node->decl;
325 switch (state)
327 case CONSTRUCTION:
328 /* At construction time we just need to finalize function and move
329 it into reachable functions list. */
331 cgraph_node::finalize_function (fndecl, false);
332 call_cgraph_insertion_hooks (node);
333 enqueue_node (node);
334 break;
336 case IPA:
337 case IPA_SSA:
338 case IPA_SSA_AFTER_INLINING:
339 /* When IPA optimization already started, do all essential
340 transformations that has been already performed on the whole
341 cgraph but not on this function. */
343 gimple_register_cfg_hooks ();
344 if (!node->analyzed)
345 node->analyze ();
346 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
347 if ((state == IPA_SSA || state == IPA_SSA_AFTER_INLINING)
348 && !gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl)))
349 g->get_passes ()->execute_early_local_passes ();
350 else if (inline_summaries != NULL)
351 compute_inline_parameters (node, true);
352 free_dominance_info (CDI_POST_DOMINATORS);
353 free_dominance_info (CDI_DOMINATORS);
354 pop_cfun ();
355 call_cgraph_insertion_hooks (node);
356 break;
358 case EXPANSION:
359 /* Functions created during expansion shall be compiled
360 directly. */
361 node->process = 0;
362 call_cgraph_insertion_hooks (node);
363 node->expand ();
364 break;
366 default:
367 gcc_unreachable ();
368 break;
372 cgraph_new_nodes.release ();
375 /* As an GCC extension we allow redefinition of the function. The
376 semantics when both copies of bodies differ is not well defined.
377 We replace the old body with new body so in unit at a time mode
378 we always use new body, while in normal mode we may end up with
379 old body inlined into some functions and new body expanded and
380 inlined in others.
382 ??? It may make more sense to use one body for inlining and other
383 body for expanding the function but this is difficult to do. */
385 void
386 cgraph_node::reset (void)
388 /* If process is set, then we have already begun whole-unit analysis.
389 This is *not* testing for whether we've already emitted the function.
390 That case can be sort-of legitimately seen with real function redefinition
391 errors. I would argue that the front end should never present us with
392 such a case, but don't enforce that for now. */
393 gcc_assert (!process);
395 /* Reset our data structures so we can analyze the function again. */
396 memset (&local, 0, sizeof (local));
397 memset (&global, 0, sizeof (global));
398 memset (&rtl, 0, sizeof (rtl));
399 analyzed = false;
400 definition = false;
401 alias = false;
402 weakref = false;
403 cpp_implicit_alias = false;
405 remove_callees ();
406 remove_all_references ();
409 /* Return true when there are references to the node. */
411 bool
412 symtab_node::referred_to_p (void)
414 ipa_ref *ref = NULL;
416 /* See if there are any references at all. */
417 if (iterate_referring (0, ref))
418 return true;
419 /* For functions check also calls. */
420 cgraph_node *cn = dyn_cast <cgraph_node *> (this);
421 if (cn && cn->callers)
422 return true;
423 return false;
426 /* DECL has been parsed. Take it, queue it, compile it at the whim of the
427 logic in effect. If NO_COLLECT is true, then our caller cannot stand to have
428 the garbage collector run at the moment. We would need to either create
429 a new GC context, or just not compile right now. */
431 void
432 cgraph_node::finalize_function (tree decl, bool no_collect)
434 cgraph_node *node = cgraph_node::get_create (decl);
436 if (node->definition)
438 /* Nested functions should only be defined once. */
439 gcc_assert (!DECL_CONTEXT (decl)
440 || TREE_CODE (DECL_CONTEXT (decl)) != FUNCTION_DECL);
441 node->reset ();
442 node->local.redefined_extern_inline = true;
445 /* Set definition first before calling notice_global_symbol so that
446 it is available to notice_global_symbol. */
447 node->definition = true;
448 notice_global_symbol (decl);
449 node->lowered = DECL_STRUCT_FUNCTION (decl)->cfg != NULL;
451 /* With -fkeep-inline-functions we are keeping all inline functions except
452 for extern inline ones. */
453 if (flag_keep_inline_functions
454 && DECL_DECLARED_INLINE_P (decl)
455 && !DECL_EXTERNAL (decl)
456 && !DECL_DISREGARD_INLINE_LIMITS (decl))
457 node->force_output = 1;
459 /* When not optimizing, also output the static functions. (see
460 PR24561), but don't do so for always_inline functions, functions
461 declared inline and nested functions. These were optimized out
462 in the original implementation and it is unclear whether we want
463 to change the behavior here. */
464 if ((!opt_for_fn (decl, optimize)
465 && !node->cpp_implicit_alias
466 && !DECL_DISREGARD_INLINE_LIMITS (decl)
467 && !DECL_DECLARED_INLINE_P (decl)
468 && !(DECL_CONTEXT (decl)
469 && TREE_CODE (DECL_CONTEXT (decl)) == FUNCTION_DECL))
470 && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
471 node->force_output = 1;
473 /* If we've not yet emitted decl, tell the debug info about it. */
474 if (!TREE_ASM_WRITTEN (decl))
475 (*debug_hooks->deferred_inline_function) (decl);
477 /* Possibly warn about unused parameters. */
478 if (warn_unused_parameter)
479 do_warn_unused_parameter (decl);
481 if (!no_collect)
482 ggc_collect ();
484 if (symtab->state == CONSTRUCTION
485 && (node->needed_p () || node->referred_to_p ()))
486 enqueue_node (node);
489 /* Add the function FNDECL to the call graph.
490 Unlike finalize_function, this function is intended to be used
491 by middle end and allows insertion of new function at arbitrary point
492 of compilation. The function can be either in high, low or SSA form
493 GIMPLE.
495 The function is assumed to be reachable and have address taken (so no
496 API breaking optimizations are performed on it).
498 Main work done by this function is to enqueue the function for later
499 processing to avoid need the passes to be re-entrant. */
501 void
502 cgraph_node::add_new_function (tree fndecl, bool lowered)
504 gcc::pass_manager *passes = g->get_passes ();
505 cgraph_node *node;
506 switch (symtab->state)
508 case PARSING:
509 cgraph_node::finalize_function (fndecl, false);
510 break;
511 case CONSTRUCTION:
512 /* Just enqueue function to be processed at nearest occurrence. */
513 node = cgraph_node::get_create (fndecl);
514 if (lowered)
515 node->lowered = true;
516 cgraph_new_nodes.safe_push (node);
517 break;
519 case IPA:
520 case IPA_SSA:
521 case IPA_SSA_AFTER_INLINING:
522 case EXPANSION:
523 /* Bring the function into finalized state and enqueue for later
524 analyzing and compilation. */
525 node = cgraph_node::get_create (fndecl);
526 node->local.local = false;
527 node->definition = true;
528 node->force_output = true;
529 if (!lowered && symtab->state == EXPANSION)
531 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
532 gimple_register_cfg_hooks ();
533 bitmap_obstack_initialize (NULL);
534 execute_pass_list (cfun, passes->all_lowering_passes);
535 passes->execute_early_local_passes ();
536 bitmap_obstack_release (NULL);
537 pop_cfun ();
539 lowered = true;
541 if (lowered)
542 node->lowered = true;
543 cgraph_new_nodes.safe_push (node);
544 break;
546 case FINISHED:
547 /* At the very end of compilation we have to do all the work up
548 to expansion. */
549 node = cgraph_node::create (fndecl);
550 if (lowered)
551 node->lowered = true;
552 node->definition = true;
553 node->analyze ();
554 push_cfun (DECL_STRUCT_FUNCTION (fndecl));
555 gimple_register_cfg_hooks ();
556 bitmap_obstack_initialize (NULL);
557 if (!gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl)))
558 g->get_passes ()->execute_early_local_passes ();
559 bitmap_obstack_release (NULL);
560 pop_cfun ();
561 node->expand ();
562 break;
564 default:
565 gcc_unreachable ();
568 /* Set a personality if required and we already passed EH lowering. */
569 if (lowered
570 && (function_needs_eh_personality (DECL_STRUCT_FUNCTION (fndecl))
571 == eh_personality_lang))
572 DECL_FUNCTION_PERSONALITY (fndecl) = lang_hooks.eh_personality ();
575 /* Analyze the function scheduled to be output. */
576 void
577 cgraph_node::analyze (void)
579 tree decl = this->decl;
580 location_t saved_loc = input_location;
581 input_location = DECL_SOURCE_LOCATION (decl);
583 if (thunk.thunk_p)
585 cgraph_node *t = cgraph_node::get (thunk.alias);
587 create_edge (t, NULL, 0, CGRAPH_FREQ_BASE);
588 /* Target code in expand_thunk may need the thunk's target
589 to be analyzed, so recurse here. */
590 if (!t->analyzed)
591 t->analyze ();
592 if (t->alias)
594 t = t->get_alias_target ();
595 if (!t->analyzed)
596 t->analyze ();
598 if (!expand_thunk (false, false))
600 thunk.alias = NULL;
601 return;
603 thunk.alias = NULL;
605 if (alias)
606 resolve_alias (cgraph_node::get (alias_target));
607 else if (dispatcher_function)
609 /* Generate the dispatcher body of multi-versioned functions. */
610 cgraph_function_version_info *dispatcher_version_info
611 = function_version ();
612 if (dispatcher_version_info != NULL
613 && (dispatcher_version_info->dispatcher_resolver
614 == NULL_TREE))
616 tree resolver = NULL_TREE;
617 gcc_assert (targetm.generate_version_dispatcher_body);
618 resolver = targetm.generate_version_dispatcher_body (this);
619 gcc_assert (resolver != NULL_TREE);
622 else
624 push_cfun (DECL_STRUCT_FUNCTION (decl));
626 assign_assembler_name_if_neeeded (decl);
628 /* Make sure to gimplify bodies only once. During analyzing a
629 function we lower it, which will require gimplified nested
630 functions, so we can end up here with an already gimplified
631 body. */
632 if (!gimple_has_body_p (decl))
633 gimplify_function_tree (decl);
634 dump_function (TDI_generic, decl);
636 /* Lower the function. */
637 if (!lowered)
639 if (nested)
640 lower_nested_functions (decl);
641 gcc_assert (!nested);
643 gimple_register_cfg_hooks ();
644 bitmap_obstack_initialize (NULL);
645 execute_pass_list (cfun, g->get_passes ()->all_lowering_passes);
646 free_dominance_info (CDI_POST_DOMINATORS);
647 free_dominance_info (CDI_DOMINATORS);
648 compact_blocks ();
649 bitmap_obstack_release (NULL);
650 lowered = true;
653 pop_cfun ();
655 analyzed = true;
657 input_location = saved_loc;
660 /* C++ frontend produce same body aliases all over the place, even before PCH
661 gets streamed out. It relies on us linking the aliases with their function
662 in order to do the fixups, but ipa-ref is not PCH safe. Consequentely we
663 first produce aliases without links, but once C++ FE is sure he won't sream
664 PCH we build the links via this function. */
666 void
667 symbol_table::process_same_body_aliases (void)
669 symtab_node *node;
670 FOR_EACH_SYMBOL (node)
671 if (node->cpp_implicit_alias && !node->analyzed)
672 node->resolve_alias
673 (TREE_CODE (node->alias_target) == VAR_DECL
674 ? (symtab_node *)varpool_node::get_create (node->alias_target)
675 : (symtab_node *)cgraph_node::get_create (node->alias_target));
676 cpp_implicit_aliases_done = true;
679 /* Process attributes common for vars and functions. */
681 static void
682 process_common_attributes (symtab_node *node, tree decl)
684 tree weakref = lookup_attribute ("weakref", DECL_ATTRIBUTES (decl));
686 if (weakref && !lookup_attribute ("alias", DECL_ATTRIBUTES (decl)))
688 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
689 "%<weakref%> attribute should be accompanied with"
690 " an %<alias%> attribute");
691 DECL_WEAK (decl) = 0;
692 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
693 DECL_ATTRIBUTES (decl));
696 if (lookup_attribute ("no_reorder", DECL_ATTRIBUTES (decl)))
697 node->no_reorder = 1;
700 /* Look for externally_visible and used attributes and mark cgraph nodes
701 accordingly.
703 We cannot mark the nodes at the point the attributes are processed (in
704 handle_*_attribute) because the copy of the declarations available at that
705 point may not be canonical. For example, in:
707 void f();
708 void f() __attribute__((used));
710 the declaration we see in handle_used_attribute will be the second
711 declaration -- but the front end will subsequently merge that declaration
712 with the original declaration and discard the second declaration.
714 Furthermore, we can't mark these nodes in finalize_function because:
716 void f() {}
717 void f() __attribute__((externally_visible));
719 is valid.
721 So, we walk the nodes at the end of the translation unit, applying the
722 attributes at that point. */
724 static void
725 process_function_and_variable_attributes (cgraph_node *first,
726 varpool_node *first_var)
728 cgraph_node *node;
729 varpool_node *vnode;
731 for (node = symtab->first_function (); node != first;
732 node = symtab->next_function (node))
734 tree decl = node->decl;
735 if (DECL_PRESERVE_P (decl))
736 node->mark_force_output ();
737 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
739 if (! TREE_PUBLIC (node->decl))
740 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes,
741 "%<externally_visible%>"
742 " attribute have effect only on public objects");
744 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl))
745 && (node->definition && !node->alias))
747 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes,
748 "%<weakref%> attribute ignored"
749 " because function is defined");
750 DECL_WEAK (decl) = 0;
751 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
752 DECL_ATTRIBUTES (decl));
755 if (lookup_attribute ("always_inline", DECL_ATTRIBUTES (decl))
756 && !DECL_DECLARED_INLINE_P (decl)
757 /* redefining extern inline function makes it DECL_UNINLINABLE. */
758 && !DECL_UNINLINABLE (decl))
759 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes,
760 "always_inline function might not be inlinable");
762 process_common_attributes (node, decl);
764 for (vnode = symtab->first_variable (); vnode != first_var;
765 vnode = symtab->next_variable (vnode))
767 tree decl = vnode->decl;
768 if (DECL_EXTERNAL (decl)
769 && DECL_INITIAL (decl))
770 varpool_node::finalize_decl (decl);
771 if (DECL_PRESERVE_P (decl))
772 vnode->force_output = true;
773 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
775 if (! TREE_PUBLIC (vnode->decl))
776 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes,
777 "%<externally_visible%>"
778 " attribute have effect only on public objects");
780 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl))
781 && vnode->definition
782 && DECL_INITIAL (decl))
784 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes,
785 "%<weakref%> attribute ignored"
786 " because variable is initialized");
787 DECL_WEAK (decl) = 0;
788 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref",
789 DECL_ATTRIBUTES (decl));
791 process_common_attributes (vnode, decl);
795 /* Mark DECL as finalized. By finalizing the declaration, frontend instruct the
796 middle end to output the variable to asm file, if needed or externally
797 visible. */
799 void
800 varpool_node::finalize_decl (tree decl)
802 varpool_node *node = varpool_node::get_create (decl);
804 gcc_assert (TREE_STATIC (decl) || DECL_EXTERNAL (decl));
806 if (node->definition)
807 return;
808 /* Set definition first before calling notice_global_symbol so that
809 it is available to notice_global_symbol. */
810 node->definition = true;
811 notice_global_symbol (decl);
812 if (TREE_THIS_VOLATILE (decl) || DECL_PRESERVE_P (decl)
813 /* Traditionally we do not eliminate static variables when not
814 optimizing and when not doing toplevel reoder. */
815 || node->no_reorder
816 || ((!flag_toplevel_reorder
817 && !DECL_COMDAT (node->decl)
818 && !DECL_ARTIFICIAL (node->decl))))
819 node->force_output = true;
821 if (symtab->state == CONSTRUCTION
822 && (node->needed_p () || node->referred_to_p ()))
823 enqueue_node (node);
824 if (symtab->state >= IPA_SSA)
825 node->analyze ();
826 /* Some frontends produce various interface variables after compilation
827 finished. */
828 if (symtab->state == FINISHED
829 || (!flag_toplevel_reorder
830 && symtab->state == EXPANSION))
831 node->assemble_decl ();
833 if (DECL_INITIAL (decl))
834 chkp_register_var_initializer (decl);
837 /* EDGE is an polymorphic call. Mark all possible targets as reachable
838 and if there is only one target, perform trivial devirtualization.
839 REACHABLE_CALL_TARGETS collects target lists we already walked to
840 avoid udplicate work. */
842 static void
843 walk_polymorphic_call_targets (hash_set<void *> *reachable_call_targets,
844 cgraph_edge *edge)
846 unsigned int i;
847 void *cache_token;
848 bool final;
849 vec <cgraph_node *>targets
850 = possible_polymorphic_call_targets
851 (edge, &final, &cache_token);
853 if (!reachable_call_targets->add (cache_token))
855 if (symtab->dump_file)
856 dump_possible_polymorphic_call_targets
857 (symtab->dump_file, edge);
859 for (i = 0; i < targets.length (); i++)
861 /* Do not bother to mark virtual methods in anonymous namespace;
862 either we will find use of virtual table defining it, or it is
863 unused. */
864 if (targets[i]->definition
865 && TREE_CODE
866 (TREE_TYPE (targets[i]->decl))
867 == METHOD_TYPE
868 && !type_in_anonymous_namespace_p
869 (method_class_type
870 (TREE_TYPE (targets[i]->decl))))
871 enqueue_node (targets[i]);
875 /* Very trivial devirtualization; when the type is
876 final or anonymous (so we know all its derivation)
877 and there is only one possible virtual call target,
878 make the edge direct. */
879 if (final)
881 if (targets.length () <= 1 && dbg_cnt (devirt))
883 cgraph_node *target;
884 if (targets.length () == 1)
885 target = targets[0];
886 else
887 target = cgraph_node::create
888 (builtin_decl_implicit (BUILT_IN_UNREACHABLE));
890 if (symtab->dump_file)
892 fprintf (symtab->dump_file,
893 "Devirtualizing call: ");
894 print_gimple_stmt (symtab->dump_file,
895 edge->call_stmt, 0,
896 TDF_SLIM);
898 if (dump_enabled_p ())
900 location_t locus = gimple_location_safe (edge->call_stmt);
901 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, locus,
902 "devirtualizing call in %s to %s\n",
903 edge->caller->name (), target->name ());
906 edge->make_direct (target);
907 edge->redirect_call_stmt_to_callee ();
909 /* Call to __builtin_unreachable shouldn't be instrumented. */
910 if (!targets.length ())
911 gimple_call_set_with_bounds (edge->call_stmt, false);
913 if (symtab->dump_file)
915 fprintf (symtab->dump_file,
916 "Devirtualized as: ");
917 print_gimple_stmt (symtab->dump_file,
918 edge->call_stmt, 0,
919 TDF_SLIM);
926 /* Discover all functions and variables that are trivially needed, analyze
927 them as well as all functions and variables referred by them */
928 static cgraph_node *first_analyzed;
929 static varpool_node *first_analyzed_var;
931 static void
932 analyze_functions (void)
934 /* Keep track of already processed nodes when called multiple times for
935 intermodule optimization. */
936 cgraph_node *first_handled = first_analyzed;
937 varpool_node *first_handled_var = first_analyzed_var;
938 hash_set<void *> reachable_call_targets;
940 symtab_node *node;
941 symtab_node *next;
942 int i;
943 ipa_ref *ref;
944 bool changed = true;
945 location_t saved_loc = input_location;
947 bitmap_obstack_initialize (NULL);
948 symtab->state = CONSTRUCTION;
949 input_location = UNKNOWN_LOCATION;
951 /* Ugly, but the fixup can not happen at a time same body alias is created;
952 C++ FE is confused about the COMDAT groups being right. */
953 if (symtab->cpp_implicit_aliases_done)
954 FOR_EACH_SYMBOL (node)
955 if (node->cpp_implicit_alias)
956 node->fixup_same_cpp_alias_visibility (node->get_alias_target ());
957 build_type_inheritance_graph ();
959 /* Analysis adds static variables that in turn adds references to new functions.
960 So we need to iterate the process until it stabilize. */
961 while (changed)
963 changed = false;
964 process_function_and_variable_attributes (first_analyzed,
965 first_analyzed_var);
967 /* First identify the trivially needed symbols. */
968 for (node = symtab->first_symbol ();
969 node != first_analyzed
970 && node != first_analyzed_var; node = node->next)
972 /* Convert COMDAT group designators to IDENTIFIER_NODEs. */
973 node->get_comdat_group_id ();
974 if (node->needed_p ())
976 enqueue_node (node);
977 if (!changed && symtab->dump_file)
978 fprintf (symtab->dump_file, "Trivially needed symbols:");
979 changed = true;
980 if (symtab->dump_file)
981 fprintf (symtab->dump_file, " %s", node->asm_name ());
982 if (!changed && symtab->dump_file)
983 fprintf (symtab->dump_file, "\n");
985 if (node == first_analyzed
986 || node == first_analyzed_var)
987 break;
989 symtab->process_new_functions ();
990 first_analyzed_var = symtab->first_variable ();
991 first_analyzed = symtab->first_function ();
993 if (changed && symtab->dump_file)
994 fprintf (symtab->dump_file, "\n");
996 /* Lower representation, build callgraph edges and references for all trivially
997 needed symbols and all symbols referred by them. */
998 while (queued_nodes != &symtab_terminator)
1000 changed = true;
1001 node = queued_nodes;
1002 queued_nodes = (symtab_node *)queued_nodes->aux;
1003 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
1004 if (cnode && cnode->definition)
1006 cgraph_edge *edge;
1007 tree decl = cnode->decl;
1009 /* ??? It is possible to create extern inline function
1010 and later using weak alias attribute to kill its body.
1011 See gcc.c-torture/compile/20011119-1.c */
1012 if (!DECL_STRUCT_FUNCTION (decl)
1013 && !cnode->alias
1014 && !cnode->thunk.thunk_p
1015 && !cnode->dispatcher_function)
1017 cnode->reset ();
1018 cnode->local.redefined_extern_inline = true;
1019 continue;
1022 if (!cnode->analyzed)
1023 cnode->analyze ();
1025 for (edge = cnode->callees; edge; edge = edge->next_callee)
1026 if (edge->callee->definition
1027 && (!DECL_EXTERNAL (edge->callee->decl)
1028 /* When not optimizing, do not try to analyze extern
1029 inline functions. Doing so is pointless. */
1030 || opt_for_fn (edge->callee->decl, optimize)
1031 /* Weakrefs needs to be preserved. */
1032 || edge->callee->alias
1033 /* always_inline functions are inlined aven at -O0. */
1034 || lookup_attribute
1035 ("always_inline",
1036 DECL_ATTRIBUTES (edge->callee->decl))
1037 /* Multiversioned functions needs the dispatcher to
1038 be produced locally even for extern functions. */
1039 || edge->callee->function_version ()))
1040 enqueue_node (edge->callee);
1041 if (opt_for_fn (cnode->decl, optimize)
1042 && opt_for_fn (cnode->decl, flag_devirtualize))
1044 cgraph_edge *next;
1046 for (edge = cnode->indirect_calls; edge; edge = next)
1048 next = edge->next_callee;
1049 if (edge->indirect_info->polymorphic)
1050 walk_polymorphic_call_targets (&reachable_call_targets,
1051 edge);
1055 /* If decl is a clone of an abstract function,
1056 mark that abstract function so that we don't release its body.
1057 The DECL_INITIAL() of that abstract function declaration
1058 will be later needed to output debug info. */
1059 if (DECL_ABSTRACT_ORIGIN (decl))
1061 cgraph_node *origin_node
1062 = cgraph_node::get_create (DECL_ABSTRACT_ORIGIN (decl));
1063 origin_node->used_as_abstract_origin = true;
1066 else
1068 varpool_node *vnode = dyn_cast <varpool_node *> (node);
1069 if (vnode && vnode->definition && !vnode->analyzed)
1070 vnode->analyze ();
1073 if (node->same_comdat_group)
1075 symtab_node *next;
1076 for (next = node->same_comdat_group;
1077 next != node;
1078 next = next->same_comdat_group)
1079 if (!next->comdat_local_p ())
1080 enqueue_node (next);
1082 for (i = 0; node->iterate_reference (i, ref); i++)
1083 if (ref->referred->definition
1084 && (!DECL_EXTERNAL (ref->referred->decl)
1085 || ((TREE_CODE (ref->referred->decl) != FUNCTION_DECL
1086 && optimize)
1087 || (TREE_CODE (ref->referred->decl) == FUNCTION_DECL
1088 && opt_for_fn (ref->referred->decl, optimize))
1089 || node->alias
1090 || ref->referred->alias)))
1091 enqueue_node (ref->referred);
1092 symtab->process_new_functions ();
1095 update_type_inheritance_graph ();
1097 /* Collect entry points to the unit. */
1098 if (symtab->dump_file)
1100 fprintf (symtab->dump_file, "\n\nInitial ");
1101 symtab_node::dump_table (symtab->dump_file);
1104 if (symtab->dump_file)
1105 fprintf (symtab->dump_file, "\nRemoving unused symbols:");
1107 for (node = symtab->first_symbol ();
1108 node != first_handled
1109 && node != first_handled_var; node = next)
1111 next = node->next;
1112 if (!node->aux && !node->referred_to_p ())
1114 if (symtab->dump_file)
1115 fprintf (symtab->dump_file, " %s", node->name ());
1116 node->remove ();
1117 continue;
1119 if (cgraph_node *cnode = dyn_cast <cgraph_node *> (node))
1121 tree decl = node->decl;
1123 if (cnode->definition && !gimple_has_body_p (decl)
1124 && !cnode->alias
1125 && !cnode->thunk.thunk_p)
1126 cnode->reset ();
1128 gcc_assert (!cnode->definition || cnode->thunk.thunk_p
1129 || cnode->alias
1130 || gimple_has_body_p (decl));
1131 gcc_assert (cnode->analyzed == cnode->definition);
1133 node->aux = NULL;
1135 for (;node; node = node->next)
1136 node->aux = NULL;
1137 first_analyzed = symtab->first_function ();
1138 first_analyzed_var = symtab->first_variable ();
1139 if (symtab->dump_file)
1141 fprintf (symtab->dump_file, "\n\nReclaimed ");
1142 symtab_node::dump_table (symtab->dump_file);
1144 bitmap_obstack_release (NULL);
1145 ggc_collect ();
1146 /* Initialize assembler name hash, in particular we want to trigger C++
1147 mangling and same body alias creation before we free DECL_ARGUMENTS
1148 used by it. */
1149 if (!seen_error ())
1150 symtab->symtab_initialize_asm_name_hash ();
1152 input_location = saved_loc;
1155 /* Translate the ugly representation of aliases as alias pairs into nice
1156 representation in callgraph. We don't handle all cases yet,
1157 unfortunately. */
1159 static void
1160 handle_alias_pairs (void)
1162 alias_pair *p;
1163 unsigned i;
1165 for (i = 0; alias_pairs && alias_pairs->iterate (i, &p);)
1167 symtab_node *target_node = symtab_node::get_for_asmname (p->target);
1169 /* Weakrefs with target not defined in current unit are easy to handle:
1170 they behave just as external variables except we need to note the
1171 alias flag to later output the weakref pseudo op into asm file. */
1172 if (!target_node
1173 && lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl)) != NULL)
1175 symtab_node *node = symtab_node::get (p->decl);
1176 if (node)
1178 node->alias_target = p->target;
1179 node->weakref = true;
1180 node->alias = true;
1182 alias_pairs->unordered_remove (i);
1183 continue;
1185 else if (!target_node)
1187 error ("%q+D aliased to undefined symbol %qE", p->decl, p->target);
1188 symtab_node *node = symtab_node::get (p->decl);
1189 if (node)
1190 node->alias = false;
1191 alias_pairs->unordered_remove (i);
1192 continue;
1195 if (DECL_EXTERNAL (target_node->decl)
1196 /* We use local aliases for C++ thunks to force the tailcall
1197 to bind locally. This is a hack - to keep it working do
1198 the following (which is not strictly correct). */
1199 && (TREE_CODE (target_node->decl) != FUNCTION_DECL
1200 || ! DECL_VIRTUAL_P (target_node->decl))
1201 && ! lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl)))
1203 error ("%q+D aliased to external symbol %qE",
1204 p->decl, p->target);
1207 if (TREE_CODE (p->decl) == FUNCTION_DECL
1208 && target_node && is_a <cgraph_node *> (target_node))
1210 cgraph_node *src_node = cgraph_node::get (p->decl);
1211 if (src_node && src_node->definition)
1212 src_node->reset ();
1213 cgraph_node::create_alias (p->decl, target_node->decl);
1214 alias_pairs->unordered_remove (i);
1216 else if (TREE_CODE (p->decl) == VAR_DECL
1217 && target_node && is_a <varpool_node *> (target_node))
1219 varpool_node::create_alias (p->decl, target_node->decl);
1220 alias_pairs->unordered_remove (i);
1222 else
1224 error ("%q+D alias in between function and variable is not supported",
1225 p->decl);
1226 warning (0, "%q+D aliased declaration",
1227 target_node->decl);
1228 alias_pairs->unordered_remove (i);
1231 vec_free (alias_pairs);
1235 /* Figure out what functions we want to assemble. */
1237 static void
1238 mark_functions_to_output (void)
1240 cgraph_node *node;
1241 #ifdef ENABLE_CHECKING
1242 bool check_same_comdat_groups = false;
1244 FOR_EACH_FUNCTION (node)
1245 gcc_assert (!node->process);
1246 #endif
1248 FOR_EACH_FUNCTION (node)
1250 tree decl = node->decl;
1252 gcc_assert (!node->process || node->same_comdat_group);
1253 if (node->process)
1254 continue;
1256 /* We need to output all local functions that are used and not
1257 always inlined, as well as those that are reachable from
1258 outside the current compilation unit. */
1259 if (node->analyzed
1260 && !node->thunk.thunk_p
1261 && !node->alias
1262 && !node->global.inlined_to
1263 && !TREE_ASM_WRITTEN (decl)
1264 && !DECL_EXTERNAL (decl))
1266 node->process = 1;
1267 if (node->same_comdat_group)
1269 cgraph_node *next;
1270 for (next = dyn_cast<cgraph_node *> (node->same_comdat_group);
1271 next != node;
1272 next = dyn_cast<cgraph_node *> (next->same_comdat_group))
1273 if (!next->thunk.thunk_p && !next->alias
1274 && !next->comdat_local_p ())
1275 next->process = 1;
1278 else if (node->same_comdat_group)
1280 #ifdef ENABLE_CHECKING
1281 check_same_comdat_groups = true;
1282 #endif
1284 else
1286 /* We should've reclaimed all functions that are not needed. */
1287 #ifdef ENABLE_CHECKING
1288 if (!node->global.inlined_to
1289 && gimple_has_body_p (decl)
1290 /* FIXME: in ltrans unit when offline copy is outside partition but inline copies
1291 are inside partition, we can end up not removing the body since we no longer
1292 have analyzed node pointing to it. */
1293 && !node->in_other_partition
1294 && !node->alias
1295 && !node->clones
1296 && !DECL_EXTERNAL (decl))
1298 node->debug ();
1299 internal_error ("failed to reclaim unneeded function");
1301 #endif
1302 gcc_assert (node->global.inlined_to
1303 || !gimple_has_body_p (decl)
1304 || node->in_other_partition
1305 || node->clones
1306 || DECL_ARTIFICIAL (decl)
1307 || DECL_EXTERNAL (decl));
1312 #ifdef ENABLE_CHECKING
1313 if (check_same_comdat_groups)
1314 FOR_EACH_FUNCTION (node)
1315 if (node->same_comdat_group && !node->process)
1317 tree decl = node->decl;
1318 if (!node->global.inlined_to
1319 && gimple_has_body_p (decl)
1320 /* FIXME: in an ltrans unit when the offline copy is outside a
1321 partition but inline copies are inside a partition, we can
1322 end up not removing the body since we no longer have an
1323 analyzed node pointing to it. */
1324 && !node->in_other_partition
1325 && !node->clones
1326 && !DECL_EXTERNAL (decl))
1328 node->debug ();
1329 internal_error ("failed to reclaim unneeded function in same "
1330 "comdat group");
1333 #endif
1336 /* DECL is FUNCTION_DECL. Initialize datastructures so DECL is a function
1337 in lowered gimple form. IN_SSA is true if the gimple is in SSA.
1339 Set current_function_decl and cfun to newly constructed empty function body.
1340 return basic block in the function body. */
1342 basic_block
1343 init_lowered_empty_function (tree decl, bool in_ssa, gcov_type count)
1345 basic_block bb;
1346 edge e;
1348 current_function_decl = decl;
1349 allocate_struct_function (decl, false);
1350 gimple_register_cfg_hooks ();
1351 init_empty_tree_cfg ();
1353 if (in_ssa)
1355 init_tree_ssa (cfun);
1356 init_ssa_operands (cfun);
1357 cfun->gimple_df->in_ssa_p = true;
1358 cfun->curr_properties |= PROP_ssa;
1361 DECL_INITIAL (decl) = make_node (BLOCK);
1363 DECL_SAVED_TREE (decl) = error_mark_node;
1364 cfun->curr_properties |= (PROP_gimple_lcf | PROP_gimple_leh | PROP_gimple_any
1365 | PROP_cfg | PROP_loops);
1367 set_loops_for_fn (cfun, ggc_cleared_alloc<loops> ());
1368 init_loops_structure (cfun, loops_for_fn (cfun), 1);
1369 loops_for_fn (cfun)->state |= LOOPS_MAY_HAVE_MULTIPLE_LATCHES;
1371 /* Create BB for body of the function and connect it properly. */
1372 ENTRY_BLOCK_PTR_FOR_FN (cfun)->count = count;
1373 ENTRY_BLOCK_PTR_FOR_FN (cfun)->frequency = REG_BR_PROB_BASE;
1374 EXIT_BLOCK_PTR_FOR_FN (cfun)->count = count;
1375 EXIT_BLOCK_PTR_FOR_FN (cfun)->frequency = REG_BR_PROB_BASE;
1376 bb = create_basic_block (NULL, ENTRY_BLOCK_PTR_FOR_FN (cfun));
1377 bb->count = count;
1378 bb->frequency = BB_FREQ_MAX;
1379 e = make_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun), bb, EDGE_FALLTHRU);
1380 e->count = count;
1381 e->probability = REG_BR_PROB_BASE;
1382 e = make_edge (bb, EXIT_BLOCK_PTR_FOR_FN (cfun), 0);
1383 e->count = count;
1384 e->probability = REG_BR_PROB_BASE;
1385 add_bb_to_loop (bb, ENTRY_BLOCK_PTR_FOR_FN (cfun)->loop_father);
1387 return bb;
1390 /* Adjust PTR by the constant FIXED_OFFSET, and by the vtable
1391 offset indicated by VIRTUAL_OFFSET, if that is
1392 non-null. THIS_ADJUSTING is nonzero for a this adjusting thunk and
1393 zero for a result adjusting thunk. */
1395 static tree
1396 thunk_adjust (gimple_stmt_iterator * bsi,
1397 tree ptr, bool this_adjusting,
1398 HOST_WIDE_INT fixed_offset, tree virtual_offset)
1400 gassign *stmt;
1401 tree ret;
1403 if (this_adjusting
1404 && fixed_offset != 0)
1406 stmt = gimple_build_assign
1407 (ptr, fold_build_pointer_plus_hwi_loc (input_location,
1408 ptr,
1409 fixed_offset));
1410 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1413 /* If there's a virtual offset, look up that value in the vtable and
1414 adjust the pointer again. */
1415 if (virtual_offset)
1417 tree vtabletmp;
1418 tree vtabletmp2;
1419 tree vtabletmp3;
1421 if (!vtable_entry_type)
1423 tree vfunc_type = make_node (FUNCTION_TYPE);
1424 TREE_TYPE (vfunc_type) = integer_type_node;
1425 TYPE_ARG_TYPES (vfunc_type) = NULL_TREE;
1426 layout_type (vfunc_type);
1428 vtable_entry_type = build_pointer_type (vfunc_type);
1431 vtabletmp =
1432 create_tmp_reg (build_pointer_type
1433 (build_pointer_type (vtable_entry_type)), "vptr");
1435 /* The vptr is always at offset zero in the object. */
1436 stmt = gimple_build_assign (vtabletmp,
1437 build1 (NOP_EXPR, TREE_TYPE (vtabletmp),
1438 ptr));
1439 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1441 /* Form the vtable address. */
1442 vtabletmp2 = create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp)),
1443 "vtableaddr");
1444 stmt = gimple_build_assign (vtabletmp2,
1445 build_simple_mem_ref (vtabletmp));
1446 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1448 /* Find the entry with the vcall offset. */
1449 stmt = gimple_build_assign (vtabletmp2,
1450 fold_build_pointer_plus_loc (input_location,
1451 vtabletmp2,
1452 virtual_offset));
1453 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1455 /* Get the offset itself. */
1456 vtabletmp3 = create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp2)),
1457 "vcalloffset");
1458 stmt = gimple_build_assign (vtabletmp3,
1459 build_simple_mem_ref (vtabletmp2));
1460 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1462 /* Adjust the `this' pointer. */
1463 ptr = fold_build_pointer_plus_loc (input_location, ptr, vtabletmp3);
1464 ptr = force_gimple_operand_gsi (bsi, ptr, true, NULL_TREE, false,
1465 GSI_CONTINUE_LINKING);
1468 if (!this_adjusting
1469 && fixed_offset != 0)
1470 /* Adjust the pointer by the constant. */
1472 tree ptrtmp;
1474 if (TREE_CODE (ptr) == VAR_DECL)
1475 ptrtmp = ptr;
1476 else
1478 ptrtmp = create_tmp_reg (TREE_TYPE (ptr), "ptr");
1479 stmt = gimple_build_assign (ptrtmp, ptr);
1480 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1482 ptr = fold_build_pointer_plus_hwi_loc (input_location,
1483 ptrtmp, fixed_offset);
1486 /* Emit the statement and gimplify the adjustment expression. */
1487 ret = create_tmp_reg (TREE_TYPE (ptr), "adjusted_this");
1488 stmt = gimple_build_assign (ret, ptr);
1489 gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
1491 return ret;
1494 /* Expand thunk NODE to gimple if possible.
1495 When FORCE_GIMPLE_THUNK is true, gimple thunk is created and
1496 no assembler is produced.
1497 When OUTPUT_ASM_THUNK is true, also produce assembler for
1498 thunks that are not lowered. */
1500 bool
1501 cgraph_node::expand_thunk (bool output_asm_thunks, bool force_gimple_thunk)
1503 bool this_adjusting = thunk.this_adjusting;
1504 HOST_WIDE_INT fixed_offset = thunk.fixed_offset;
1505 HOST_WIDE_INT virtual_value = thunk.virtual_value;
1506 tree virtual_offset = NULL;
1507 tree alias = callees->callee->decl;
1508 tree thunk_fndecl = decl;
1509 tree a;
1511 /* Instrumentation thunk is the same function with
1512 a different signature. Never need to expand it. */
1513 if (thunk.add_pointer_bounds_args)
1514 return false;
1516 if (!force_gimple_thunk && this_adjusting
1517 && targetm.asm_out.can_output_mi_thunk (thunk_fndecl, fixed_offset,
1518 virtual_value, alias))
1520 const char *fnname;
1521 tree fn_block;
1522 tree restype = TREE_TYPE (TREE_TYPE (thunk_fndecl));
1524 if (!output_asm_thunks)
1526 analyzed = true;
1527 return false;
1530 if (in_lto_p)
1531 get_untransformed_body ();
1532 a = DECL_ARGUMENTS (thunk_fndecl);
1534 current_function_decl = thunk_fndecl;
1536 /* Ensure thunks are emitted in their correct sections. */
1537 resolve_unique_section (thunk_fndecl, 0,
1538 flag_function_sections);
1540 DECL_RESULT (thunk_fndecl)
1541 = build_decl (DECL_SOURCE_LOCATION (thunk_fndecl),
1542 RESULT_DECL, 0, restype);
1543 DECL_CONTEXT (DECL_RESULT (thunk_fndecl)) = thunk_fndecl;
1544 fnname = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (thunk_fndecl));
1546 /* The back end expects DECL_INITIAL to contain a BLOCK, so we
1547 create one. */
1548 fn_block = make_node (BLOCK);
1549 BLOCK_VARS (fn_block) = a;
1550 DECL_INITIAL (thunk_fndecl) = fn_block;
1551 init_function_start (thunk_fndecl);
1552 cfun->is_thunk = 1;
1553 insn_locations_init ();
1554 set_curr_insn_location (DECL_SOURCE_LOCATION (thunk_fndecl));
1555 prologue_location = curr_insn_location ();
1556 assemble_start_function (thunk_fndecl, fnname);
1558 targetm.asm_out.output_mi_thunk (asm_out_file, thunk_fndecl,
1559 fixed_offset, virtual_value, alias);
1561 assemble_end_function (thunk_fndecl, fnname);
1562 insn_locations_finalize ();
1563 init_insn_lengths ();
1564 free_after_compilation (cfun);
1565 set_cfun (NULL);
1566 TREE_ASM_WRITTEN (thunk_fndecl) = 1;
1567 thunk.thunk_p = false;
1568 analyzed = false;
1570 else if (stdarg_p (TREE_TYPE (thunk_fndecl)))
1572 error ("generic thunk code fails for method %qD which uses %<...%>",
1573 thunk_fndecl);
1574 TREE_ASM_WRITTEN (thunk_fndecl) = 1;
1575 analyzed = true;
1576 return false;
1578 else
1580 tree restype;
1581 basic_block bb, then_bb, else_bb, return_bb;
1582 gimple_stmt_iterator bsi;
1583 int nargs = 0;
1584 tree arg;
1585 int i;
1586 tree resdecl;
1587 tree restmp = NULL;
1588 tree resbnd = NULL;
1590 gcall *call;
1591 greturn *ret;
1592 bool alias_is_noreturn = TREE_THIS_VOLATILE (alias);
1594 if (in_lto_p)
1595 get_untransformed_body ();
1596 a = DECL_ARGUMENTS (thunk_fndecl);
1598 current_function_decl = thunk_fndecl;
1600 /* Ensure thunks are emitted in their correct sections. */
1601 resolve_unique_section (thunk_fndecl, 0,
1602 flag_function_sections);
1604 DECL_IGNORED_P (thunk_fndecl) = 1;
1605 bitmap_obstack_initialize (NULL);
1607 if (thunk.virtual_offset_p)
1608 virtual_offset = size_int (virtual_value);
1610 /* Build the return declaration for the function. */
1611 restype = TREE_TYPE (TREE_TYPE (thunk_fndecl));
1612 if (DECL_RESULT (thunk_fndecl) == NULL_TREE)
1614 resdecl = build_decl (input_location, RESULT_DECL, 0, restype);
1615 DECL_ARTIFICIAL (resdecl) = 1;
1616 DECL_IGNORED_P (resdecl) = 1;
1617 DECL_RESULT (thunk_fndecl) = resdecl;
1618 DECL_CONTEXT (DECL_RESULT (thunk_fndecl)) = thunk_fndecl;
1620 else
1621 resdecl = DECL_RESULT (thunk_fndecl);
1623 bb = then_bb = else_bb = return_bb
1624 = init_lowered_empty_function (thunk_fndecl, true, count);
1626 bsi = gsi_start_bb (bb);
1628 /* Build call to the function being thunked. */
1629 if (!VOID_TYPE_P (restype) && !alias_is_noreturn)
1631 if (DECL_BY_REFERENCE (resdecl))
1633 restmp = gimple_fold_indirect_ref (resdecl);
1634 if (!restmp)
1635 restmp = build2 (MEM_REF,
1636 TREE_TYPE (TREE_TYPE (DECL_RESULT (alias))),
1637 resdecl,
1638 build_int_cst (TREE_TYPE
1639 (DECL_RESULT (alias)), 0));
1641 else if (!is_gimple_reg_type (restype))
1643 if (aggregate_value_p (resdecl, TREE_TYPE (thunk_fndecl)))
1645 restmp = resdecl;
1647 if (TREE_CODE (restmp) == VAR_DECL)
1648 add_local_decl (cfun, restmp);
1649 BLOCK_VARS (DECL_INITIAL (current_function_decl)) = restmp;
1651 else
1652 restmp = create_tmp_var (restype, "retval");
1654 else
1655 restmp = create_tmp_reg (restype, "retval");
1658 for (arg = a; arg; arg = DECL_CHAIN (arg))
1659 nargs++;
1660 auto_vec<tree> vargs (nargs);
1661 i = 0;
1662 arg = a;
1663 if (this_adjusting)
1665 vargs.quick_push (thunk_adjust (&bsi, a, 1, fixed_offset,
1666 virtual_offset));
1667 arg = DECL_CHAIN (a);
1668 i = 1;
1671 if (nargs)
1672 for (; i < nargs; i++, arg = DECL_CHAIN (arg))
1674 tree tmp = arg;
1675 if (!is_gimple_val (arg))
1677 tmp = create_tmp_reg (TYPE_MAIN_VARIANT
1678 (TREE_TYPE (arg)), "arg");
1679 gimple stmt = gimple_build_assign (tmp, arg);
1680 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT);
1682 vargs.quick_push (tmp);
1684 call = gimple_build_call_vec (build_fold_addr_expr_loc (0, alias), vargs);
1685 callees->call_stmt = call;
1686 gimple_call_set_from_thunk (call, true);
1687 gimple_call_set_with_bounds (call, instrumentation_clone);
1689 /* Return slot optimization is always possible and in fact requred to
1690 return values with DECL_BY_REFERENCE. */
1691 if (aggregate_value_p (resdecl, TREE_TYPE (thunk_fndecl))
1692 && (!is_gimple_reg_type (TREE_TYPE (resdecl))
1693 || DECL_BY_REFERENCE (resdecl)))
1694 gimple_call_set_return_slot_opt (call, true);
1696 if (restmp && !alias_is_noreturn)
1698 gimple_call_set_lhs (call, restmp);
1699 gcc_assert (useless_type_conversion_p (TREE_TYPE (restmp),
1700 TREE_TYPE (TREE_TYPE (alias))));
1702 gsi_insert_after (&bsi, call, GSI_NEW_STMT);
1703 if (!alias_is_noreturn)
1705 if (instrumentation_clone
1706 && !DECL_BY_REFERENCE (resdecl)
1707 && restmp
1708 && BOUNDED_P (restmp))
1710 resbnd = chkp_insert_retbnd_call (NULL, restmp, &bsi);
1711 create_edge (get_create (gimple_call_fndecl (gsi_stmt (bsi))),
1712 as_a <gcall *> (gsi_stmt (bsi)),
1713 callees->count, callees->frequency);
1716 if (restmp && !this_adjusting
1717 && (fixed_offset || virtual_offset))
1719 tree true_label = NULL_TREE;
1721 if (TREE_CODE (TREE_TYPE (restmp)) == POINTER_TYPE)
1723 gimple stmt;
1724 edge e;
1725 /* If the return type is a pointer, we need to
1726 protect against NULL. We know there will be an
1727 adjustment, because that's why we're emitting a
1728 thunk. */
1729 then_bb = create_basic_block (NULL, bb);
1730 then_bb->count = count - count / 16;
1731 then_bb->frequency = BB_FREQ_MAX - BB_FREQ_MAX / 16;
1732 return_bb = create_basic_block (NULL, then_bb);
1733 return_bb->count = count;
1734 return_bb->frequency = BB_FREQ_MAX;
1735 else_bb = create_basic_block (NULL, else_bb);
1736 then_bb->count = count / 16;
1737 then_bb->frequency = BB_FREQ_MAX / 16;
1738 add_bb_to_loop (then_bb, bb->loop_father);
1739 add_bb_to_loop (return_bb, bb->loop_father);
1740 add_bb_to_loop (else_bb, bb->loop_father);
1741 remove_edge (single_succ_edge (bb));
1742 true_label = gimple_block_label (then_bb);
1743 stmt = gimple_build_cond (NE_EXPR, restmp,
1744 build_zero_cst (TREE_TYPE (restmp)),
1745 NULL_TREE, NULL_TREE);
1746 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT);
1747 e = make_edge (bb, then_bb, EDGE_TRUE_VALUE);
1748 e->probability = REG_BR_PROB_BASE - REG_BR_PROB_BASE / 16;
1749 e->count = count - count / 16;
1750 e = make_edge (bb, else_bb, EDGE_FALSE_VALUE);
1751 e->probability = REG_BR_PROB_BASE / 16;
1752 e->count = count / 16;
1753 e = make_edge (return_bb, EXIT_BLOCK_PTR_FOR_FN (cfun), 0);
1754 e->probability = REG_BR_PROB_BASE;
1755 e->count = count;
1756 e = make_edge (then_bb, return_bb, EDGE_FALLTHRU);
1757 e->probability = REG_BR_PROB_BASE;
1758 e->count = count - count / 16;
1759 e = make_edge (else_bb, return_bb, EDGE_FALLTHRU);
1760 e->probability = REG_BR_PROB_BASE;
1761 e->count = count / 16;
1762 bsi = gsi_last_bb (then_bb);
1765 restmp = thunk_adjust (&bsi, restmp, /*this_adjusting=*/0,
1766 fixed_offset, virtual_offset);
1767 if (true_label)
1769 gimple stmt;
1770 bsi = gsi_last_bb (else_bb);
1771 stmt = gimple_build_assign (restmp,
1772 build_zero_cst (TREE_TYPE (restmp)));
1773 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT);
1774 bsi = gsi_last_bb (return_bb);
1777 else
1778 gimple_call_set_tail (call, true);
1780 /* Build return value. */
1781 if (!DECL_BY_REFERENCE (resdecl))
1782 ret = gimple_build_return (restmp);
1783 else
1784 ret = gimple_build_return (resdecl);
1785 gimple_return_set_retbnd (ret, resbnd);
1787 gsi_insert_after (&bsi, ret, GSI_NEW_STMT);
1789 else
1791 gimple_call_set_tail (call, true);
1792 remove_edge (single_succ_edge (bb));
1795 cfun->gimple_df->in_ssa_p = true;
1796 profile_status_for_fn (cfun)
1797 = count ? PROFILE_READ : PROFILE_GUESSED;
1798 /* FIXME: C++ FE should stop setting TREE_ASM_WRITTEN on thunks. */
1799 TREE_ASM_WRITTEN (thunk_fndecl) = false;
1800 delete_unreachable_blocks ();
1801 update_ssa (TODO_update_ssa);
1802 #ifdef ENABLE_CHECKING
1803 verify_flow_info ();
1804 #endif
1805 free_dominance_info (CDI_DOMINATORS);
1807 /* Since we want to emit the thunk, we explicitly mark its name as
1808 referenced. */
1809 thunk.thunk_p = false;
1810 lowered = true;
1811 bitmap_obstack_release (NULL);
1813 current_function_decl = NULL;
1814 set_cfun (NULL);
1815 return true;
1818 /* Assemble thunks and aliases associated to node. */
1820 void
1821 cgraph_node::assemble_thunks_and_aliases (void)
1823 cgraph_edge *e;
1824 ipa_ref *ref;
1826 for (e = callers; e;)
1827 if (e->caller->thunk.thunk_p
1828 && !e->caller->thunk.add_pointer_bounds_args)
1830 cgraph_node *thunk = e->caller;
1832 e = e->next_caller;
1833 thunk->expand_thunk (true, false);
1834 thunk->assemble_thunks_and_aliases ();
1836 else
1837 e = e->next_caller;
1839 FOR_EACH_ALIAS (this, ref)
1841 cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
1842 bool saved_written = TREE_ASM_WRITTEN (decl);
1844 /* Force assemble_alias to really output the alias this time instead
1845 of buffering it in same alias pairs. */
1846 TREE_ASM_WRITTEN (decl) = 1;
1847 do_assemble_alias (alias->decl,
1848 DECL_ASSEMBLER_NAME (decl));
1849 alias->assemble_thunks_and_aliases ();
1850 TREE_ASM_WRITTEN (decl) = saved_written;
1854 /* Expand function specified by node. */
1856 void
1857 cgraph_node::expand (void)
1859 location_t saved_loc;
1861 /* We ought to not compile any inline clones. */
1862 gcc_assert (!global.inlined_to);
1864 announce_function (decl);
1865 process = 0;
1866 gcc_assert (lowered);
1867 get_untransformed_body ();
1869 /* Generate RTL for the body of DECL. */
1871 timevar_push (TV_REST_OF_COMPILATION);
1873 gcc_assert (symtab->global_info_ready);
1875 /* Initialize the default bitmap obstack. */
1876 bitmap_obstack_initialize (NULL);
1878 /* Initialize the RTL code for the function. */
1879 current_function_decl = decl;
1880 saved_loc = input_location;
1881 input_location = DECL_SOURCE_LOCATION (decl);
1882 init_function_start (decl);
1884 gimple_register_cfg_hooks ();
1886 bitmap_obstack_initialize (&reg_obstack); /* FIXME, only at RTL generation*/
1888 execute_all_ipa_transforms ();
1890 /* Perform all tree transforms and optimizations. */
1892 /* Signal the start of passes. */
1893 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_START, NULL);
1895 execute_pass_list (cfun, g->get_passes ()->all_passes);
1897 /* Signal the end of passes. */
1898 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_END, NULL);
1900 bitmap_obstack_release (&reg_obstack);
1902 /* Release the default bitmap obstack. */
1903 bitmap_obstack_release (NULL);
1905 /* If requested, warn about function definitions where the function will
1906 return a value (usually of some struct or union type) which itself will
1907 take up a lot of stack space. */
1908 if (warn_larger_than && !DECL_EXTERNAL (decl) && TREE_TYPE (decl))
1910 tree ret_type = TREE_TYPE (TREE_TYPE (decl));
1912 if (ret_type && TYPE_SIZE_UNIT (ret_type)
1913 && TREE_CODE (TYPE_SIZE_UNIT (ret_type)) == INTEGER_CST
1914 && 0 < compare_tree_int (TYPE_SIZE_UNIT (ret_type),
1915 larger_than_size))
1917 unsigned int size_as_int
1918 = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (ret_type));
1920 if (compare_tree_int (TYPE_SIZE_UNIT (ret_type), size_as_int) == 0)
1921 warning (OPT_Wlarger_than_, "size of return value of %q+D is %u bytes",
1922 decl, size_as_int);
1923 else
1924 warning (OPT_Wlarger_than_, "size of return value of %q+D is larger than %wd bytes",
1925 decl, larger_than_size);
1929 gimple_set_body (decl, NULL);
1930 if (DECL_STRUCT_FUNCTION (decl) == 0
1931 && !cgraph_node::get (decl)->origin)
1933 /* Stop pointing to the local nodes about to be freed.
1934 But DECL_INITIAL must remain nonzero so we know this
1935 was an actual function definition.
1936 For a nested function, this is done in c_pop_function_context.
1937 If rest_of_compilation set this to 0, leave it 0. */
1938 if (DECL_INITIAL (decl) != 0)
1939 DECL_INITIAL (decl) = error_mark_node;
1942 input_location = saved_loc;
1944 ggc_collect ();
1945 timevar_pop (TV_REST_OF_COMPILATION);
1947 /* Make sure that BE didn't give up on compiling. */
1948 gcc_assert (TREE_ASM_WRITTEN (decl));
1949 set_cfun (NULL);
1950 current_function_decl = NULL;
1952 /* It would make a lot more sense to output thunks before function body to get more
1953 forward and lest backwarding jumps. This however would need solving problem
1954 with comdats. See PR48668. Also aliases must come after function itself to
1955 make one pass assemblers, like one on AIX, happy. See PR 50689.
1956 FIXME: Perhaps thunks should be move before function IFF they are not in comdat
1957 groups. */
1958 assemble_thunks_and_aliases ();
1959 release_body ();
1960 /* Eliminate all call edges. This is important so the GIMPLE_CALL no longer
1961 points to the dead function body. */
1962 remove_callees ();
1963 remove_all_references ();
1966 /* Node comparer that is responsible for the order that corresponds
1967 to time when a function was launched for the first time. */
1969 static int
1970 node_cmp (const void *pa, const void *pb)
1972 const cgraph_node *a = *(const cgraph_node * const *) pa;
1973 const cgraph_node *b = *(const cgraph_node * const *) pb;
1975 /* Functions with time profile must be before these without profile. */
1976 if (!a->tp_first_run || !b->tp_first_run)
1977 return a->tp_first_run - b->tp_first_run;
1979 return a->tp_first_run != b->tp_first_run
1980 ? b->tp_first_run - a->tp_first_run
1981 : b->order - a->order;
1984 /* Expand all functions that must be output.
1986 Attempt to topologically sort the nodes so function is output when
1987 all called functions are already assembled to allow data to be
1988 propagated across the callgraph. Use a stack to get smaller distance
1989 between a function and its callees (later we may choose to use a more
1990 sophisticated algorithm for function reordering; we will likely want
1991 to use subsections to make the output functions appear in top-down
1992 order). */
1994 static void
1995 expand_all_functions (void)
1997 cgraph_node *node;
1998 cgraph_node **order = XCNEWVEC (cgraph_node *,
1999 symtab->cgraph_count);
2000 unsigned int expanded_func_count = 0, profiled_func_count = 0;
2001 int order_pos, new_order_pos = 0;
2002 int i;
2004 order_pos = ipa_reverse_postorder (order);
2005 gcc_assert (order_pos == symtab->cgraph_count);
2007 /* Garbage collector may remove inline clones we eliminate during
2008 optimization. So we must be sure to not reference them. */
2009 for (i = 0; i < order_pos; i++)
2010 if (order[i]->process)
2011 order[new_order_pos++] = order[i];
2013 if (flag_profile_reorder_functions)
2014 qsort (order, new_order_pos, sizeof (cgraph_node *), node_cmp);
2016 for (i = new_order_pos - 1; i >= 0; i--)
2018 node = order[i];
2020 if (node->process)
2022 expanded_func_count++;
2023 if(node->tp_first_run)
2024 profiled_func_count++;
2026 if (symtab->dump_file)
2027 fprintf (symtab->dump_file,
2028 "Time profile order in expand_all_functions:%s:%d\n",
2029 node->asm_name (), node->tp_first_run);
2030 node->process = 0;
2031 node->expand ();
2035 if (dump_file)
2036 fprintf (dump_file, "Expanded functions with time profile (%s):%u/%u\n",
2037 main_input_filename, profiled_func_count, expanded_func_count);
2039 if (symtab->dump_file && flag_profile_reorder_functions)
2040 fprintf (symtab->dump_file, "Expanded functions with time profile:%u/%u\n",
2041 profiled_func_count, expanded_func_count);
2043 symtab->process_new_functions ();
2044 free_gimplify_stack ();
2046 free (order);
2049 /* This is used to sort the node types by the cgraph order number. */
2051 enum cgraph_order_sort_kind
2053 ORDER_UNDEFINED = 0,
2054 ORDER_FUNCTION,
2055 ORDER_VAR,
2056 ORDER_ASM
2059 struct cgraph_order_sort
2061 enum cgraph_order_sort_kind kind;
2062 union
2064 cgraph_node *f;
2065 varpool_node *v;
2066 asm_node *a;
2067 } u;
2070 /* Output all functions, variables, and asm statements in the order
2071 according to their order fields, which is the order in which they
2072 appeared in the file. This implements -fno-toplevel-reorder. In
2073 this mode we may output functions and variables which don't really
2074 need to be output.
2075 When NO_REORDER is true only do this for symbols marked no reorder. */
2077 static void
2078 output_in_order (bool no_reorder)
2080 int max;
2081 cgraph_order_sort *nodes;
2082 int i;
2083 cgraph_node *pf;
2084 varpool_node *pv;
2085 asm_node *pa;
2086 max = symtab->order;
2087 nodes = XCNEWVEC (cgraph_order_sort, max);
2089 FOR_EACH_DEFINED_FUNCTION (pf)
2091 if (pf->process && !pf->thunk.thunk_p && !pf->alias)
2093 if (no_reorder && !pf->no_reorder)
2094 continue;
2095 i = pf->order;
2096 gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
2097 nodes[i].kind = ORDER_FUNCTION;
2098 nodes[i].u.f = pf;
2102 FOR_EACH_DEFINED_VARIABLE (pv)
2103 if (!DECL_EXTERNAL (pv->decl))
2105 if (no_reorder && !pv->no_reorder)
2106 continue;
2107 i = pv->order;
2108 gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
2109 nodes[i].kind = ORDER_VAR;
2110 nodes[i].u.v = pv;
2113 for (pa = symtab->first_asm_symbol (); pa; pa = pa->next)
2115 i = pa->order;
2116 gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
2117 nodes[i].kind = ORDER_ASM;
2118 nodes[i].u.a = pa;
2121 /* In toplevel reorder mode we output all statics; mark them as needed. */
2123 for (i = 0; i < max; ++i)
2124 if (nodes[i].kind == ORDER_VAR)
2125 nodes[i].u.v->finalize_named_section_flags ();
2127 for (i = 0; i < max; ++i)
2129 switch (nodes[i].kind)
2131 case ORDER_FUNCTION:
2132 nodes[i].u.f->process = 0;
2133 nodes[i].u.f->expand ();
2134 break;
2136 case ORDER_VAR:
2137 nodes[i].u.v->assemble_decl ();
2138 break;
2140 case ORDER_ASM:
2141 assemble_asm (nodes[i].u.a->asm_str);
2142 break;
2144 case ORDER_UNDEFINED:
2145 break;
2147 default:
2148 gcc_unreachable ();
2152 symtab->clear_asm_symbols ();
2154 free (nodes);
2157 static void
2158 ipa_passes (void)
2160 gcc::pass_manager *passes = g->get_passes ();
2162 set_cfun (NULL);
2163 current_function_decl = NULL;
2164 gimple_register_cfg_hooks ();
2165 bitmap_obstack_initialize (NULL);
2167 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_START, NULL);
2169 if (!in_lto_p)
2171 execute_ipa_pass_list (passes->all_small_ipa_passes);
2172 if (seen_error ())
2173 return;
2176 /* This extra symtab_remove_unreachable_nodes pass tends to catch some
2177 devirtualization and other changes where removal iterate. */
2178 symtab->remove_unreachable_nodes (symtab->dump_file);
2180 /* If pass_all_early_optimizations was not scheduled, the state of
2181 the cgraph will not be properly updated. Update it now. */
2182 if (symtab->state < IPA_SSA)
2183 symtab->state = IPA_SSA;
2185 if (!in_lto_p)
2187 /* Generate coverage variables and constructors. */
2188 coverage_finish ();
2190 /* Process new functions added. */
2191 set_cfun (NULL);
2192 current_function_decl = NULL;
2193 symtab->process_new_functions ();
2195 execute_ipa_summary_passes
2196 ((ipa_opt_pass_d *) passes->all_regular_ipa_passes);
2199 /* Some targets need to handle LTO assembler output specially. */
2200 if (flag_generate_lto || flag_generate_offload)
2201 targetm.asm_out.lto_start ();
2203 if (!in_lto_p)
2205 if (g->have_offload)
2207 section_name_prefix = OFFLOAD_SECTION_NAME_PREFIX;
2208 lto_stream_offload_p = true;
2209 ipa_write_summaries ();
2210 lto_stream_offload_p = false;
2212 if (flag_lto)
2214 section_name_prefix = LTO_SECTION_NAME_PREFIX;
2215 lto_stream_offload_p = false;
2216 ipa_write_summaries ();
2220 if (flag_generate_lto || flag_generate_offload)
2221 targetm.asm_out.lto_end ();
2223 if (!flag_ltrans && (in_lto_p || !flag_lto || flag_fat_lto_objects))
2224 execute_ipa_pass_list (passes->all_regular_ipa_passes);
2225 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_END, NULL);
2227 bitmap_obstack_release (NULL);
2231 /* Return string alias is alias of. */
2233 static tree
2234 get_alias_symbol (tree decl)
2236 tree alias = lookup_attribute ("alias", DECL_ATTRIBUTES (decl));
2237 return get_identifier (TREE_STRING_POINTER
2238 (TREE_VALUE (TREE_VALUE (alias))));
2242 /* Weakrefs may be associated to external decls and thus not output
2243 at expansion time. Emit all necessary aliases. */
2245 void
2246 symbol_table::output_weakrefs (void)
2248 symtab_node *node;
2249 cgraph_node *cnode;
2250 FOR_EACH_SYMBOL (node)
2251 if (node->alias
2252 && !TREE_ASM_WRITTEN (node->decl)
2253 && (!(cnode = dyn_cast <cgraph_node *> (node))
2254 || !cnode->instrumented_version
2255 || !TREE_ASM_WRITTEN (cnode->instrumented_version->decl))
2256 && node->weakref)
2258 tree target;
2260 /* Weakrefs are special by not requiring target definition in current
2261 compilation unit. It is thus bit hard to work out what we want to
2262 alias.
2263 When alias target is defined, we need to fetch it from symtab reference,
2264 otherwise it is pointed to by alias_target. */
2265 if (node->alias_target)
2266 target = (DECL_P (node->alias_target)
2267 ? DECL_ASSEMBLER_NAME (node->alias_target)
2268 : node->alias_target);
2269 else if (node->analyzed)
2270 target = DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl);
2271 else
2273 gcc_unreachable ();
2274 target = get_alias_symbol (node->decl);
2276 do_assemble_alias (node->decl, target);
2280 /* Perform simple optimizations based on callgraph. */
2282 void
2283 symbol_table::compile (void)
2285 if (seen_error ())
2286 return;
2288 #ifdef ENABLE_CHECKING
2289 symtab_node::verify_symtab_nodes ();
2290 #endif
2292 timevar_push (TV_CGRAPHOPT);
2293 if (pre_ipa_mem_report)
2295 fprintf (stderr, "Memory consumption before IPA\n");
2296 dump_memory_report (false);
2298 if (!quiet_flag)
2299 fprintf (stderr, "Performing interprocedural optimizations\n");
2300 state = IPA;
2302 /* Offloading requires LTO infrastructure. */
2303 if (!in_lto_p && g->have_offload)
2304 flag_generate_offload = 1;
2306 /* If LTO is enabled, initialize the streamer hooks needed by GIMPLE. */
2307 if (flag_generate_lto || flag_generate_offload)
2308 lto_streamer_hooks_init ();
2310 /* Don't run the IPA passes if there was any error or sorry messages. */
2311 if (!seen_error ())
2312 ipa_passes ();
2314 /* Do nothing else if any IPA pass found errors or if we are just streaming LTO. */
2315 if (seen_error ()
2316 || (!in_lto_p && flag_lto && !flag_fat_lto_objects))
2318 timevar_pop (TV_CGRAPHOPT);
2319 return;
2322 global_info_ready = true;
2323 if (dump_file)
2325 fprintf (dump_file, "Optimized ");
2326 symtab_node:: dump_table (dump_file);
2328 if (post_ipa_mem_report)
2330 fprintf (stderr, "Memory consumption after IPA\n");
2331 dump_memory_report (false);
2333 timevar_pop (TV_CGRAPHOPT);
2335 /* Output everything. */
2336 (*debug_hooks->assembly_start) ();
2337 if (!quiet_flag)
2338 fprintf (stderr, "Assembling functions:\n");
2339 #ifdef ENABLE_CHECKING
2340 symtab_node::verify_symtab_nodes ();
2341 #endif
2343 materialize_all_clones ();
2344 bitmap_obstack_initialize (NULL);
2345 execute_ipa_pass_list (g->get_passes ()->all_late_ipa_passes);
2346 bitmap_obstack_release (NULL);
2347 mark_functions_to_output ();
2349 /* When weakref support is missing, we autmatically translate all
2350 references to NODE to references to its ultimate alias target.
2351 The renaming mechanizm uses flag IDENTIFIER_TRANSPARENT_ALIAS and
2352 TREE_CHAIN.
2354 Set up this mapping before we output any assembler but once we are sure
2355 that all symbol renaming is done.
2357 FIXME: All this uglyness can go away if we just do renaming at gimple
2358 level by physically rewritting the IL. At the moment we can only redirect
2359 calls, so we need infrastructure for renaming references as well. */
2360 #ifndef ASM_OUTPUT_WEAKREF
2361 symtab_node *node;
2363 FOR_EACH_SYMBOL (node)
2364 if (node->alias
2365 && lookup_attribute ("weakref", DECL_ATTRIBUTES (node->decl)))
2367 IDENTIFIER_TRANSPARENT_ALIAS
2368 (DECL_ASSEMBLER_NAME (node->decl)) = 1;
2369 TREE_CHAIN (DECL_ASSEMBLER_NAME (node->decl))
2370 = (node->alias_target ? node->alias_target
2371 : DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl));
2373 #endif
2375 state = EXPANSION;
2377 if (!flag_toplevel_reorder)
2378 output_in_order (false);
2379 else
2381 /* Output first asm statements and anything ordered. The process
2382 flag is cleared for these nodes, so we skip them later. */
2383 output_in_order (true);
2384 expand_all_functions ();
2385 output_variables ();
2388 process_new_functions ();
2389 state = FINISHED;
2390 output_weakrefs ();
2392 if (dump_file)
2394 fprintf (dump_file, "\nFinal ");
2395 symtab_node::dump_table (dump_file);
2397 #ifdef ENABLE_CHECKING
2398 symtab_node::verify_symtab_nodes ();
2399 /* Double check that all inline clones are gone and that all
2400 function bodies have been released from memory. */
2401 if (!seen_error ())
2403 cgraph_node *node;
2404 bool error_found = false;
2406 FOR_EACH_DEFINED_FUNCTION (node)
2407 if (node->global.inlined_to
2408 || gimple_has_body_p (node->decl))
2410 error_found = true;
2411 node->debug ();
2413 if (error_found)
2414 internal_error ("nodes with unreleased memory found");
2416 #endif
2420 /* Analyze the whole compilation unit once it is parsed completely. */
2422 void
2423 symbol_table::finalize_compilation_unit (void)
2425 timevar_push (TV_CGRAPH);
2427 /* If we're here there's no current function anymore. Some frontends
2428 are lazy in clearing these. */
2429 current_function_decl = NULL;
2430 set_cfun (NULL);
2432 /* Do not skip analyzing the functions if there were errors, we
2433 miss diagnostics for following functions otherwise. */
2435 /* Emit size functions we didn't inline. */
2436 finalize_size_functions ();
2438 /* Mark alias targets necessary and emit diagnostics. */
2439 handle_alias_pairs ();
2441 if (!quiet_flag)
2443 fprintf (stderr, "\nAnalyzing compilation unit\n");
2444 fflush (stderr);
2447 if (flag_dump_passes)
2448 dump_passes ();
2450 /* Gimplify and lower all functions, compute reachability and
2451 remove unreachable nodes. */
2452 analyze_functions ();
2454 /* Mark alias targets necessary and emit diagnostics. */
2455 handle_alias_pairs ();
2457 /* Gimplify and lower thunks. */
2458 analyze_functions ();
2460 /* Finally drive the pass manager. */
2461 compile ();
2463 timevar_pop (TV_CGRAPH);
2466 /* Reset all state within cgraphunit.c so that we can rerun the compiler
2467 within the same process. For use by toplev::finalize. */
2469 void
2470 cgraphunit_c_finalize (void)
2472 gcc_assert (cgraph_new_nodes.length () == 0);
2473 cgraph_new_nodes.truncate (0);
2475 vtable_entry_type = NULL;
2476 queued_nodes = &symtab_terminator;
2478 first_analyzed = NULL;
2479 first_analyzed_var = NULL;
2482 /* Creates a wrapper from cgraph_node to TARGET node. Thunk is used for this
2483 kind of wrapper method. */
2485 void
2486 cgraph_node::create_wrapper (cgraph_node *target)
2488 /* Preserve DECL_RESULT so we get right by reference flag. */
2489 tree decl_result = DECL_RESULT (decl);
2491 /* Remove the function's body but keep arguments to be reused
2492 for thunk. */
2493 release_body (true);
2494 reset ();
2496 DECL_UNINLINABLE (decl) = false;
2497 DECL_RESULT (decl) = decl_result;
2498 DECL_INITIAL (decl) = NULL;
2499 allocate_struct_function (decl, false);
2500 set_cfun (NULL);
2502 /* Turn alias into thunk and expand it into GIMPLE representation. */
2503 definition = true;
2505 memset (&thunk, 0, sizeof (cgraph_thunk_info));
2506 thunk.thunk_p = true;
2507 create_edge (target, NULL, count, CGRAPH_FREQ_BASE);
2509 tree arguments = DECL_ARGUMENTS (decl);
2511 while (arguments)
2513 TREE_ADDRESSABLE (arguments) = false;
2514 arguments = TREE_CHAIN (arguments);
2517 expand_thunk (false, true);
2519 /* Inline summary set-up. */
2520 analyze ();
2521 inline_analyze_function (this);
2524 #include "gt-cgraphunit.h"