1 /* Driver of optimization process
2 Copyright (C) 2003-2014 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
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
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:
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
36 - varpool_finalize_decl
38 This function has same behavior as the above but is used for static
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.
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
70 1) Inter-procedural optimization.
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
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
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
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
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.
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
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
162 #include "coretypes.h"
166 #include "stor-layout.h"
167 #include "stringpool.h"
173 #include "hash-set.h"
174 #include "machmode.h"
175 #include "hard-reg-set.h"
177 #include "function.h"
178 #include "basic-block.h"
179 #include "tree-ssa-alias.h"
180 #include "internal-fn.h"
181 #include "gimple-fold.h"
182 #include "gimple-expr.h"
185 #include "gimplify.h"
186 #include "gimple-iterator.h"
187 #include "gimplify-me.h"
188 #include "gimple-ssa.h"
189 #include "tree-cfg.h"
190 #include "tree-into-ssa.h"
191 #include "tree-ssa.h"
192 #include "tree-inline.h"
193 #include "langhooks.h"
198 #include "diagnostic.h"
202 #include "hash-map.h"
203 #include "plugin-api.h"
206 #include "alloc-pool.h"
207 #include "ipa-prop.h"
208 #include "tree-iterator.h"
209 #include "tree-pass.h"
210 #include "tree-dump.h"
211 #include "gimple-pretty-print.h"
213 #include "coverage.h"
215 #include "ipa-inline.h"
216 #include "ipa-utils.h"
217 #include "lto-streamer.h"
220 #include "regset.h" /* FIXME: For reg_obstack. */
222 #include "pass_manager.h"
223 #include "tree-nested.h"
224 #include "gimplify.h"
227 /* Queue of cgraph nodes scheduled to be added into cgraph. This is a
228 secondary queue used during optimization to accommodate passes that
229 may generate new functions that need to be optimized and expanded. */
230 vec
<cgraph_node
*> cgraph_new_nodes
;
232 static void expand_all_functions (void);
233 static void mark_functions_to_output (void);
234 static void handle_alias_pairs (void);
236 /* Used for vtable lookup in thunk adjusting. */
237 static GTY (()) tree vtable_entry_type
;
239 /* Determine if symbol declaration is needed. That is, visible to something
240 either outside this translation unit, something magic in the system
243 symtab_node::needed_p (void)
245 /* Double check that no one output the function into assembly file
247 gcc_checking_assert (!DECL_ASSEMBLER_NAME_SET_P (decl
)
248 || !TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl
)));
253 if (DECL_EXTERNAL (decl
))
256 /* If the user told us it is used, then it must be so. */
260 /* ABI forced symbols are needed when they are external. */
261 if (forced_by_abi
&& TREE_PUBLIC (decl
))
264 /* Keep constructors, destructors and virtual functions. */
265 if (TREE_CODE (decl
) == FUNCTION_DECL
266 && (DECL_STATIC_CONSTRUCTOR (decl
) || DECL_STATIC_DESTRUCTOR (decl
)))
269 /* Externally visible variables must be output. The exception is
270 COMDAT variables that must be output only when they are needed. */
271 if (TREE_PUBLIC (decl
) && !DECL_COMDAT (decl
))
277 /* Head and terminator of the queue of nodes to be processed while building
280 static symtab_node symtab_terminator
;
281 static symtab_node
*queued_nodes
= &symtab_terminator
;
283 /* Add NODE to queue starting at QUEUED_NODES.
284 The queue is linked via AUX pointers and terminated by pointer to 1. */
287 enqueue_node (symtab_node
*node
)
291 gcc_checking_assert (queued_nodes
);
292 node
->aux
= queued_nodes
;
296 /* Process CGRAPH_NEW_FUNCTIONS and perform actions necessary to add these
297 functions into callgraph in a way so they look like ordinary reachable
298 functions inserted into callgraph already at construction time. */
301 symbol_table::process_new_functions (void)
305 if (!cgraph_new_nodes
.exists ())
308 handle_alias_pairs ();
309 /* Note that this queue may grow as its being processed, as the new
310 functions may generate new ones. */
311 for (unsigned i
= 0; i
< cgraph_new_nodes
.length (); i
++)
313 cgraph_node
*node
= cgraph_new_nodes
[i
];
318 /* At construction time we just need to finalize function and move
319 it into reachable functions list. */
321 cgraph_node::finalize_function (fndecl
, false);
322 call_cgraph_insertion_hooks (node
);
328 /* When IPA optimization already started, do all essential
329 transformations that has been already performed on the whole
330 cgraph but not on this function. */
332 gimple_register_cfg_hooks ();
335 push_cfun (DECL_STRUCT_FUNCTION (fndecl
));
337 && !gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl
)))
338 g
->get_passes ()->execute_early_local_passes ();
339 else if (inline_summary_vec
!= NULL
)
340 compute_inline_parameters (node
, true);
341 free_dominance_info (CDI_POST_DOMINATORS
);
342 free_dominance_info (CDI_DOMINATORS
);
344 call_cgraph_insertion_hooks (node
);
348 /* Functions created during expansion shall be compiled
351 call_cgraph_insertion_hooks (node
);
361 cgraph_new_nodes
.release ();
364 /* As an GCC extension we allow redefinition of the function. The
365 semantics when both copies of bodies differ is not well defined.
366 We replace the old body with new body so in unit at a time mode
367 we always use new body, while in normal mode we may end up with
368 old body inlined into some functions and new body expanded and
371 ??? It may make more sense to use one body for inlining and other
372 body for expanding the function but this is difficult to do. */
375 cgraph_node::reset (void)
377 /* If process is set, then we have already begun whole-unit analysis.
378 This is *not* testing for whether we've already emitted the function.
379 That case can be sort-of legitimately seen with real function redefinition
380 errors. I would argue that the front end should never present us with
381 such a case, but don't enforce that for now. */
382 gcc_assert (!process
);
384 /* Reset our data structures so we can analyze the function again. */
385 memset (&local
, 0, sizeof (local
));
386 memset (&global
, 0, sizeof (global
));
387 memset (&rtl
, 0, sizeof (rtl
));
392 cpp_implicit_alias
= false;
395 remove_all_references ();
398 /* Return true when there are references to the node. */
401 symtab_node::referred_to_p (void)
405 /* See if there are any references at all. */
406 if (iterate_referring (0, ref
))
408 /* For functions check also calls. */
409 cgraph_node
*cn
= dyn_cast
<cgraph_node
*> (this);
410 if (cn
&& cn
->callers
)
415 /* DECL has been parsed. Take it, queue it, compile it at the whim of the
416 logic in effect. If NO_COLLECT is true, then our caller cannot stand to have
417 the garbage collector run at the moment. We would need to either create
418 a new GC context, or just not compile right now. */
421 cgraph_node::finalize_function (tree decl
, bool no_collect
)
423 cgraph_node
*node
= cgraph_node::get_create (decl
);
425 if (node
->definition
)
427 /* Nested functions should only be defined once. */
428 gcc_assert (!DECL_CONTEXT (decl
)
429 || TREE_CODE (DECL_CONTEXT (decl
)) != FUNCTION_DECL
);
431 node
->local
.redefined_extern_inline
= true;
434 notice_global_symbol (decl
);
435 node
->definition
= true;
436 node
->lowered
= DECL_STRUCT_FUNCTION (decl
)->cfg
!= NULL
;
438 /* With -fkeep-inline-functions we are keeping all inline functions except
439 for extern inline ones. */
440 if (flag_keep_inline_functions
441 && DECL_DECLARED_INLINE_P (decl
)
442 && !DECL_EXTERNAL (decl
)
443 && !DECL_DISREGARD_INLINE_LIMITS (decl
))
444 node
->force_output
= 1;
446 /* When not optimizing, also output the static functions. (see
447 PR24561), but don't do so for always_inline functions, functions
448 declared inline and nested functions. These were optimized out
449 in the original implementation and it is unclear whether we want
450 to change the behavior here. */
452 && !node
->cpp_implicit_alias
453 && !DECL_DISREGARD_INLINE_LIMITS (decl
)
454 && !DECL_DECLARED_INLINE_P (decl
)
455 && !(DECL_CONTEXT (decl
)
456 && TREE_CODE (DECL_CONTEXT (decl
)) == FUNCTION_DECL
))
457 && !DECL_COMDAT (decl
) && !DECL_EXTERNAL (decl
))
458 node
->force_output
= 1;
460 /* If we've not yet emitted decl, tell the debug info about it. */
461 if (!TREE_ASM_WRITTEN (decl
))
462 (*debug_hooks
->deferred_inline_function
) (decl
);
464 /* Possibly warn about unused parameters. */
465 if (warn_unused_parameter
)
466 do_warn_unused_parameter (decl
);
471 if (symtab
->state
== CONSTRUCTION
472 && (node
->needed_p () || node
->referred_to_p ()))
476 /* Add the function FNDECL to the call graph.
477 Unlike finalize_function, this function is intended to be used
478 by middle end and allows insertion of new function at arbitrary point
479 of compilation. The function can be either in high, low or SSA form
482 The function is assumed to be reachable and have address taken (so no
483 API breaking optimizations are performed on it).
485 Main work done by this function is to enqueue the function for later
486 processing to avoid need the passes to be re-entrant. */
489 cgraph_node::add_new_function (tree fndecl
, bool lowered
)
491 gcc::pass_manager
*passes
= g
->get_passes ();
493 switch (symtab
->state
)
496 cgraph_node::finalize_function (fndecl
, false);
499 /* Just enqueue function to be processed at nearest occurrence. */
500 node
= cgraph_node::get_create (fndecl
);
502 node
->lowered
= true;
503 cgraph_new_nodes
.safe_push (node
);
509 /* Bring the function into finalized state and enqueue for later
510 analyzing and compilation. */
511 node
= cgraph_node::get_create (fndecl
);
512 node
->local
.local
= false;
513 node
->definition
= true;
514 node
->force_output
= true;
515 if (!lowered
&& symtab
->state
== EXPANSION
)
517 push_cfun (DECL_STRUCT_FUNCTION (fndecl
));
518 gimple_register_cfg_hooks ();
519 bitmap_obstack_initialize (NULL
);
520 execute_pass_list (cfun
, passes
->all_lowering_passes
);
521 passes
->execute_early_local_passes ();
522 bitmap_obstack_release (NULL
);
528 node
->lowered
= true;
529 cgraph_new_nodes
.safe_push (node
);
533 /* At the very end of compilation we have to do all the work up
535 node
= cgraph_node::create (fndecl
);
537 node
->lowered
= true;
538 node
->definition
= true;
540 push_cfun (DECL_STRUCT_FUNCTION (fndecl
));
541 gimple_register_cfg_hooks ();
542 bitmap_obstack_initialize (NULL
);
543 if (!gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl
)))
544 g
->get_passes ()->execute_early_local_passes ();
545 bitmap_obstack_release (NULL
);
554 /* Set a personality if required and we already passed EH lowering. */
556 && (function_needs_eh_personality (DECL_STRUCT_FUNCTION (fndecl
))
557 == eh_personality_lang
))
558 DECL_FUNCTION_PERSONALITY (fndecl
) = lang_hooks
.eh_personality ();
561 /* Analyze the function scheduled to be output. */
563 cgraph_node::analyze (void)
565 tree decl
= this->decl
;
566 location_t saved_loc
= input_location
;
567 input_location
= DECL_SOURCE_LOCATION (decl
);
571 create_edge (cgraph_node::get (thunk
.alias
),
572 NULL
, 0, CGRAPH_FREQ_BASE
);
573 if (!expand_thunk (false, false))
581 resolve_alias (cgraph_node::get (alias_target
));
582 else if (dispatcher_function
)
584 /* Generate the dispatcher body of multi-versioned functions. */
585 cgraph_function_version_info
*dispatcher_version_info
586 = function_version ();
587 if (dispatcher_version_info
!= NULL
588 && (dispatcher_version_info
->dispatcher_resolver
591 tree resolver
= NULL_TREE
;
592 gcc_assert (targetm
.generate_version_dispatcher_body
);
593 resolver
= targetm
.generate_version_dispatcher_body (this);
594 gcc_assert (resolver
!= NULL_TREE
);
599 push_cfun (DECL_STRUCT_FUNCTION (decl
));
601 assign_assembler_name_if_neeeded (decl
);
603 /* Make sure to gimplify bodies only once. During analyzing a
604 function we lower it, which will require gimplified nested
605 functions, so we can end up here with an already gimplified
607 if (!gimple_has_body_p (decl
))
608 gimplify_function_tree (decl
);
609 dump_function (TDI_generic
, decl
);
611 /* Lower the function. */
615 lower_nested_functions (decl
);
616 gcc_assert (!nested
);
618 gimple_register_cfg_hooks ();
619 bitmap_obstack_initialize (NULL
);
620 execute_pass_list (cfun
, g
->get_passes ()->all_lowering_passes
);
621 free_dominance_info (CDI_POST_DOMINATORS
);
622 free_dominance_info (CDI_DOMINATORS
);
624 bitmap_obstack_release (NULL
);
632 input_location
= saved_loc
;
635 /* C++ frontend produce same body aliases all over the place, even before PCH
636 gets streamed out. It relies on us linking the aliases with their function
637 in order to do the fixups, but ipa-ref is not PCH safe. Consequentely we
638 first produce aliases without links, but once C++ FE is sure he won't sream
639 PCH we build the links via this function. */
642 symbol_table::process_same_body_aliases (void)
645 FOR_EACH_SYMBOL (node
)
646 if (node
->cpp_implicit_alias
&& !node
->analyzed
)
648 (TREE_CODE (node
->alias_target
) == VAR_DECL
649 ? (symtab_node
*)varpool_node::get_create (node
->alias_target
)
650 : (symtab_node
*)cgraph_node::get_create (node
->alias_target
));
651 cpp_implicit_aliases_done
= true;
654 /* Process attributes common for vars and functions. */
657 process_common_attributes (symtab_node
*node
, tree decl
)
659 tree weakref
= lookup_attribute ("weakref", DECL_ATTRIBUTES (decl
));
661 if (weakref
&& !lookup_attribute ("alias", DECL_ATTRIBUTES (decl
)))
663 warning_at (DECL_SOURCE_LOCATION (decl
), OPT_Wattributes
,
664 "%<weakref%> attribute should be accompanied with"
665 " an %<alias%> attribute");
666 DECL_WEAK (decl
) = 0;
667 DECL_ATTRIBUTES (decl
) = remove_attribute ("weakref",
668 DECL_ATTRIBUTES (decl
));
671 if (lookup_attribute ("no_reorder", DECL_ATTRIBUTES (decl
)))
672 node
->no_reorder
= 1;
675 /* Look for externally_visible and used attributes and mark cgraph nodes
678 We cannot mark the nodes at the point the attributes are processed (in
679 handle_*_attribute) because the copy of the declarations available at that
680 point may not be canonical. For example, in:
683 void f() __attribute__((used));
685 the declaration we see in handle_used_attribute will be the second
686 declaration -- but the front end will subsequently merge that declaration
687 with the original declaration and discard the second declaration.
689 Furthermore, we can't mark these nodes in finalize_function because:
692 void f() __attribute__((externally_visible));
696 So, we walk the nodes at the end of the translation unit, applying the
697 attributes at that point. */
700 process_function_and_variable_attributes (cgraph_node
*first
,
701 varpool_node
*first_var
)
706 for (node
= symtab
->first_function (); node
!= first
;
707 node
= symtab
->next_function (node
))
709 tree decl
= node
->decl
;
710 if (DECL_PRESERVE_P (decl
))
711 node
->mark_force_output ();
712 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl
)))
714 if (! TREE_PUBLIC (node
->decl
))
715 warning_at (DECL_SOURCE_LOCATION (node
->decl
), OPT_Wattributes
,
716 "%<externally_visible%>"
717 " attribute have effect only on public objects");
719 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl
))
720 && (node
->definition
&& !node
->alias
))
722 warning_at (DECL_SOURCE_LOCATION (node
->decl
), OPT_Wattributes
,
723 "%<weakref%> attribute ignored"
724 " because function is defined");
725 DECL_WEAK (decl
) = 0;
726 DECL_ATTRIBUTES (decl
) = remove_attribute ("weakref",
727 DECL_ATTRIBUTES (decl
));
730 if (lookup_attribute ("always_inline", DECL_ATTRIBUTES (decl
))
731 && !DECL_DECLARED_INLINE_P (decl
)
732 /* redefining extern inline function makes it DECL_UNINLINABLE. */
733 && !DECL_UNINLINABLE (decl
))
734 warning_at (DECL_SOURCE_LOCATION (decl
), OPT_Wattributes
,
735 "always_inline function might not be inlinable");
737 process_common_attributes (node
, decl
);
739 for (vnode
= symtab
->first_variable (); vnode
!= first_var
;
740 vnode
= symtab
->next_variable (vnode
))
742 tree decl
= vnode
->decl
;
743 if (DECL_EXTERNAL (decl
)
744 && DECL_INITIAL (decl
))
745 varpool_node::finalize_decl (decl
);
746 if (DECL_PRESERVE_P (decl
))
747 vnode
->force_output
= true;
748 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl
)))
750 if (! TREE_PUBLIC (vnode
->decl
))
751 warning_at (DECL_SOURCE_LOCATION (vnode
->decl
), OPT_Wattributes
,
752 "%<externally_visible%>"
753 " attribute have effect only on public objects");
755 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl
))
757 && DECL_INITIAL (decl
))
759 warning_at (DECL_SOURCE_LOCATION (vnode
->decl
), OPT_Wattributes
,
760 "%<weakref%> attribute ignored"
761 " because variable is initialized");
762 DECL_WEAK (decl
) = 0;
763 DECL_ATTRIBUTES (decl
) = remove_attribute ("weakref",
764 DECL_ATTRIBUTES (decl
));
766 process_common_attributes (vnode
, decl
);
770 /* Mark DECL as finalized. By finalizing the declaration, frontend instruct the
771 middle end to output the variable to asm file, if needed or externally
775 varpool_node::finalize_decl (tree decl
)
777 varpool_node
*node
= varpool_node::get_create (decl
);
779 gcc_assert (TREE_STATIC (decl
) || DECL_EXTERNAL (decl
));
781 if (node
->definition
)
783 notice_global_symbol (decl
);
784 node
->definition
= true;
785 if (TREE_THIS_VOLATILE (decl
) || DECL_PRESERVE_P (decl
)
786 /* Traditionally we do not eliminate static variables when not
787 optimizing and when not doing toplevel reoder. */
789 || ((!flag_toplevel_reorder
790 && !DECL_COMDAT (node
->decl
)
791 && !DECL_ARTIFICIAL (node
->decl
))))
792 node
->force_output
= true;
794 if (symtab
->state
== CONSTRUCTION
795 && (node
->needed_p () || node
->referred_to_p ()))
797 if (symtab
->state
>= IPA_SSA
)
799 /* Some frontends produce various interface variables after compilation
801 if (symtab
->state
== FINISHED
802 || (!flag_toplevel_reorder
803 && symtab
->state
== EXPANSION
))
804 node
->assemble_decl ();
807 /* EDGE is an polymorphic call. Mark all possible targets as reachable
808 and if there is only one target, perform trivial devirtualization.
809 REACHABLE_CALL_TARGETS collects target lists we already walked to
810 avoid udplicate work. */
813 walk_polymorphic_call_targets (hash_set
<void *> *reachable_call_targets
,
819 vec
<cgraph_node
*>targets
820 = possible_polymorphic_call_targets
821 (edge
, &final
, &cache_token
);
823 if (!reachable_call_targets
->add (cache_token
))
825 if (symtab
->dump_file
)
826 dump_possible_polymorphic_call_targets
827 (symtab
->dump_file
, edge
);
829 for (i
= 0; i
< targets
.length (); i
++)
831 /* Do not bother to mark virtual methods in anonymous namespace;
832 either we will find use of virtual table defining it, or it is
834 if (targets
[i
]->definition
836 (TREE_TYPE (targets
[i
]->decl
))
838 && !type_in_anonymous_namespace_p
840 (TREE_TYPE (targets
[i
]->decl
))))
841 enqueue_node (targets
[i
]);
845 /* Very trivial devirtualization; when the type is
846 final or anonymous (so we know all its derivation)
847 and there is only one possible virtual call target,
848 make the edge direct. */
851 if (targets
.length () <= 1 && dbg_cnt (devirt
))
854 if (targets
.length () == 1)
857 target
= cgraph_node::create
858 (builtin_decl_implicit (BUILT_IN_UNREACHABLE
));
860 if (symtab
->dump_file
)
862 fprintf (symtab
->dump_file
,
863 "Devirtualizing call: ");
864 print_gimple_stmt (symtab
->dump_file
,
868 if (dump_enabled_p ())
870 location_t locus
= gimple_location_safe (edge
->call_stmt
);
871 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS
, locus
,
872 "devirtualizing call in %s to %s\n",
873 edge
->caller
->name (), target
->name ());
876 edge
->make_direct (target
);
877 edge
->redirect_call_stmt_to_callee ();
878 if (symtab
->dump_file
)
880 fprintf (symtab
->dump_file
,
881 "Devirtualized as: ");
882 print_gimple_stmt (symtab
->dump_file
,
891 /* Discover all functions and variables that are trivially needed, analyze
892 them as well as all functions and variables referred by them */
893 static cgraph_node
*first_analyzed
;
894 static varpool_node
*first_analyzed_var
;
897 analyze_functions (void)
899 /* Keep track of already processed nodes when called multiple times for
900 intermodule optimization. */
901 cgraph_node
*first_handled
= first_analyzed
;
902 varpool_node
*first_handled_var
= first_analyzed_var
;
903 hash_set
<void *> reachable_call_targets
;
910 location_t saved_loc
= input_location
;
912 bitmap_obstack_initialize (NULL
);
913 symtab
->state
= CONSTRUCTION
;
914 input_location
= UNKNOWN_LOCATION
;
916 /* Ugly, but the fixup can not happen at a time same body alias is created;
917 C++ FE is confused about the COMDAT groups being right. */
918 if (symtab
->cpp_implicit_aliases_done
)
919 FOR_EACH_SYMBOL (node
)
920 if (node
->cpp_implicit_alias
)
921 node
->fixup_same_cpp_alias_visibility (node
->get_alias_target ());
922 if (optimize
&& flag_devirtualize
)
923 build_type_inheritance_graph ();
925 /* Analysis adds static variables that in turn adds references to new functions.
926 So we need to iterate the process until it stabilize. */
930 process_function_and_variable_attributes (first_analyzed
,
933 /* First identify the trivially needed symbols. */
934 for (node
= symtab
->first_symbol ();
935 node
!= first_analyzed
936 && node
!= first_analyzed_var
; node
= node
->next
)
938 /* Convert COMDAT group designators to IDENTIFIER_NODEs. */
939 node
->get_comdat_group_id ();
940 if (node
->needed_p ())
943 if (!changed
&& symtab
->dump_file
)
944 fprintf (symtab
->dump_file
, "Trivially needed symbols:");
946 if (symtab
->dump_file
)
947 fprintf (symtab
->dump_file
, " %s", node
->asm_name ());
948 if (!changed
&& symtab
->dump_file
)
949 fprintf (symtab
->dump_file
, "\n");
951 if (node
== first_analyzed
952 || node
== first_analyzed_var
)
955 symtab
->process_new_functions ();
956 first_analyzed_var
= symtab
->first_variable ();
957 first_analyzed
= symtab
->first_function ();
959 if (changed
&& symtab
->dump_file
)
960 fprintf (symtab
->dump_file
, "\n");
962 /* Lower representation, build callgraph edges and references for all trivially
963 needed symbols and all symbols referred by them. */
964 while (queued_nodes
!= &symtab_terminator
)
968 queued_nodes
= (symtab_node
*)queued_nodes
->aux
;
969 cgraph_node
*cnode
= dyn_cast
<cgraph_node
*> (node
);
970 if (cnode
&& cnode
->definition
)
973 tree decl
= cnode
->decl
;
975 /* ??? It is possible to create extern inline function
976 and later using weak alias attribute to kill its body.
977 See gcc.c-torture/compile/20011119-1.c */
978 if (!DECL_STRUCT_FUNCTION (decl
)
980 && !cnode
->thunk
.thunk_p
981 && !cnode
->dispatcher_function
)
984 cnode
->local
.redefined_extern_inline
= true;
988 if (!cnode
->analyzed
)
991 for (edge
= cnode
->callees
; edge
; edge
= edge
->next_callee
)
992 if (edge
->callee
->definition
)
993 enqueue_node (edge
->callee
);
994 if (optimize
&& flag_devirtualize
)
998 for (edge
= cnode
->indirect_calls
; edge
; edge
= next
)
1000 next
= edge
->next_callee
;
1001 if (edge
->indirect_info
->polymorphic
)
1002 walk_polymorphic_call_targets (&reachable_call_targets
,
1007 /* If decl is a clone of an abstract function,
1008 mark that abstract function so that we don't release its body.
1009 The DECL_INITIAL() of that abstract function declaration
1010 will be later needed to output debug info. */
1011 if (DECL_ABSTRACT_ORIGIN (decl
))
1013 cgraph_node
*origin_node
1014 = cgraph_node::get_create (DECL_ABSTRACT_ORIGIN (decl
));
1015 origin_node
->used_as_abstract_origin
= true;
1020 varpool_node
*vnode
= dyn_cast
<varpool_node
*> (node
);
1021 if (vnode
&& vnode
->definition
&& !vnode
->analyzed
)
1025 if (node
->same_comdat_group
)
1028 for (next
= node
->same_comdat_group
;
1030 next
= next
->same_comdat_group
)
1031 enqueue_node (next
);
1033 for (i
= 0; node
->iterate_reference (i
, ref
); i
++)
1034 if (ref
->referred
->definition
)
1035 enqueue_node (ref
->referred
);
1036 symtab
->process_new_functions ();
1039 if (optimize
&& flag_devirtualize
)
1040 update_type_inheritance_graph ();
1042 /* Collect entry points to the unit. */
1043 if (symtab
->dump_file
)
1045 fprintf (symtab
->dump_file
, "\n\nInitial ");
1046 symtab_node::dump_table (symtab
->dump_file
);
1049 if (symtab
->dump_file
)
1050 fprintf (symtab
->dump_file
, "\nRemoving unused symbols:");
1052 for (node
= symtab
->first_symbol ();
1053 node
!= first_handled
1054 && node
!= first_handled_var
; node
= next
)
1057 if (!node
->aux
&& !node
->referred_to_p ())
1059 if (symtab
->dump_file
)
1060 fprintf (symtab
->dump_file
, " %s", node
->name ());
1064 if (cgraph_node
*cnode
= dyn_cast
<cgraph_node
*> (node
))
1066 tree decl
= node
->decl
;
1068 if (cnode
->definition
&& !gimple_has_body_p (decl
)
1070 && !cnode
->thunk
.thunk_p
)
1073 gcc_assert (!cnode
->definition
|| cnode
->thunk
.thunk_p
1075 || gimple_has_body_p (decl
));
1076 gcc_assert (cnode
->analyzed
== cnode
->definition
);
1080 for (;node
; node
= node
->next
)
1082 first_analyzed
= symtab
->first_function ();
1083 first_analyzed_var
= symtab
->first_variable ();
1084 if (symtab
->dump_file
)
1086 fprintf (symtab
->dump_file
, "\n\nReclaimed ");
1087 symtab_node::dump_table (symtab
->dump_file
);
1089 bitmap_obstack_release (NULL
);
1091 /* Initialize assembler name hash, in particular we want to trigger C++
1092 mangling and same body alias creation before we free DECL_ARGUMENTS
1095 symtab
->symtab_initialize_asm_name_hash ();
1097 input_location
= saved_loc
;
1100 /* Translate the ugly representation of aliases as alias pairs into nice
1101 representation in callgraph. We don't handle all cases yet,
1105 handle_alias_pairs (void)
1110 for (i
= 0; alias_pairs
&& alias_pairs
->iterate (i
, &p
);)
1112 symtab_node
*target_node
= symtab_node::get_for_asmname (p
->target
);
1114 /* Weakrefs with target not defined in current unit are easy to handle:
1115 they behave just as external variables except we need to note the
1116 alias flag to later output the weakref pseudo op into asm file. */
1118 && lookup_attribute ("weakref", DECL_ATTRIBUTES (p
->decl
)) != NULL
)
1120 symtab_node
*node
= symtab_node::get (p
->decl
);
1123 node
->alias_target
= p
->target
;
1124 node
->weakref
= true;
1127 alias_pairs
->unordered_remove (i
);
1130 else if (!target_node
)
1132 error ("%q+D aliased to undefined symbol %qE", p
->decl
, p
->target
);
1133 symtab_node
*node
= symtab_node::get (p
->decl
);
1135 node
->alias
= false;
1136 alias_pairs
->unordered_remove (i
);
1140 if (DECL_EXTERNAL (target_node
->decl
)
1141 /* We use local aliases for C++ thunks to force the tailcall
1142 to bind locally. This is a hack - to keep it working do
1143 the following (which is not strictly correct). */
1144 && (TREE_CODE (target_node
->decl
) != FUNCTION_DECL
1145 || ! DECL_VIRTUAL_P (target_node
->decl
))
1146 && ! lookup_attribute ("weakref", DECL_ATTRIBUTES (p
->decl
)))
1148 error ("%q+D aliased to external symbol %qE",
1149 p
->decl
, p
->target
);
1152 if (TREE_CODE (p
->decl
) == FUNCTION_DECL
1153 && target_node
&& is_a
<cgraph_node
*> (target_node
))
1155 cgraph_node
*src_node
= cgraph_node::get (p
->decl
);
1156 if (src_node
&& src_node
->definition
)
1158 cgraph_node::create_alias (p
->decl
, target_node
->decl
);
1159 alias_pairs
->unordered_remove (i
);
1161 else if (TREE_CODE (p
->decl
) == VAR_DECL
1162 && target_node
&& is_a
<varpool_node
*> (target_node
))
1164 varpool_node::create_alias (p
->decl
, target_node
->decl
);
1165 alias_pairs
->unordered_remove (i
);
1169 error ("%q+D alias in between function and variable is not supported",
1171 warning (0, "%q+D aliased declaration",
1173 alias_pairs
->unordered_remove (i
);
1176 vec_free (alias_pairs
);
1180 /* Figure out what functions we want to assemble. */
1183 mark_functions_to_output (void)
1186 #ifdef ENABLE_CHECKING
1187 bool check_same_comdat_groups
= false;
1189 FOR_EACH_FUNCTION (node
)
1190 gcc_assert (!node
->process
);
1193 FOR_EACH_FUNCTION (node
)
1195 tree decl
= node
->decl
;
1197 gcc_assert (!node
->process
|| node
->same_comdat_group
);
1201 /* We need to output all local functions that are used and not
1202 always inlined, as well as those that are reachable from
1203 outside the current compilation unit. */
1205 && !node
->thunk
.thunk_p
1207 && !node
->global
.inlined_to
1208 && !TREE_ASM_WRITTEN (decl
)
1209 && !DECL_EXTERNAL (decl
))
1212 if (node
->same_comdat_group
)
1215 for (next
= dyn_cast
<cgraph_node
*> (node
->same_comdat_group
);
1217 next
= dyn_cast
<cgraph_node
*> (next
->same_comdat_group
))
1218 if (!next
->thunk
.thunk_p
&& !next
->alias
1219 && !next
->comdat_local_p ())
1223 else if (node
->same_comdat_group
)
1225 #ifdef ENABLE_CHECKING
1226 check_same_comdat_groups
= true;
1231 /* We should've reclaimed all functions that are not needed. */
1232 #ifdef ENABLE_CHECKING
1233 if (!node
->global
.inlined_to
1234 && gimple_has_body_p (decl
)
1235 /* FIXME: in ltrans unit when offline copy is outside partition but inline copies
1236 are inside partition, we can end up not removing the body since we no longer
1237 have analyzed node pointing to it. */
1238 && !node
->in_other_partition
1241 && !DECL_EXTERNAL (decl
))
1244 internal_error ("failed to reclaim unneeded function");
1247 gcc_assert (node
->global
.inlined_to
1248 || !gimple_has_body_p (decl
)
1249 || node
->in_other_partition
1251 || DECL_ARTIFICIAL (decl
)
1252 || DECL_EXTERNAL (decl
));
1257 #ifdef ENABLE_CHECKING
1258 if (check_same_comdat_groups
)
1259 FOR_EACH_FUNCTION (node
)
1260 if (node
->same_comdat_group
&& !node
->process
)
1262 tree decl
= node
->decl
;
1263 if (!node
->global
.inlined_to
1264 && gimple_has_body_p (decl
)
1265 /* FIXME: in an ltrans unit when the offline copy is outside a
1266 partition but inline copies are inside a partition, we can
1267 end up not removing the body since we no longer have an
1268 analyzed node pointing to it. */
1269 && !node
->in_other_partition
1271 && !DECL_EXTERNAL (decl
))
1274 internal_error ("failed to reclaim unneeded function in same "
1281 /* DECL is FUNCTION_DECL. Initialize datastructures so DECL is a function
1282 in lowered gimple form. IN_SSA is true if the gimple is in SSA.
1284 Set current_function_decl and cfun to newly constructed empty function body.
1285 return basic block in the function body. */
1288 init_lowered_empty_function (tree decl
, bool in_ssa
)
1292 current_function_decl
= decl
;
1293 allocate_struct_function (decl
, false);
1294 gimple_register_cfg_hooks ();
1295 init_empty_tree_cfg ();
1299 init_tree_ssa (cfun
);
1300 init_ssa_operands (cfun
);
1301 cfun
->gimple_df
->in_ssa_p
= true;
1302 cfun
->curr_properties
|= PROP_ssa
;
1305 DECL_INITIAL (decl
) = make_node (BLOCK
);
1307 DECL_SAVED_TREE (decl
) = error_mark_node
;
1308 cfun
->curr_properties
|= (PROP_gimple_lcf
| PROP_gimple_leh
| PROP_gimple_any
1309 | PROP_cfg
| PROP_loops
);
1311 set_loops_for_fn (cfun
, ggc_cleared_alloc
<loops
> ());
1312 init_loops_structure (cfun
, loops_for_fn (cfun
), 1);
1313 loops_for_fn (cfun
)->state
|= LOOPS_MAY_HAVE_MULTIPLE_LATCHES
;
1315 /* Create BB for body of the function and connect it properly. */
1316 bb
= create_basic_block (NULL
, (void *) 0, ENTRY_BLOCK_PTR_FOR_FN (cfun
));
1317 make_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun
), bb
, EDGE_FALLTHRU
);
1318 make_edge (bb
, EXIT_BLOCK_PTR_FOR_FN (cfun
), 0);
1319 add_bb_to_loop (bb
, ENTRY_BLOCK_PTR_FOR_FN (cfun
)->loop_father
);
1324 /* Adjust PTR by the constant FIXED_OFFSET, and by the vtable
1325 offset indicated by VIRTUAL_OFFSET, if that is
1326 non-null. THIS_ADJUSTING is nonzero for a this adjusting thunk and
1327 zero for a result adjusting thunk. */
1330 thunk_adjust (gimple_stmt_iterator
* bsi
,
1331 tree ptr
, bool this_adjusting
,
1332 HOST_WIDE_INT fixed_offset
, tree virtual_offset
)
1338 && fixed_offset
!= 0)
1340 stmt
= gimple_build_assign
1341 (ptr
, fold_build_pointer_plus_hwi_loc (input_location
,
1344 gsi_insert_after (bsi
, stmt
, GSI_NEW_STMT
);
1347 /* If there's a virtual offset, look up that value in the vtable and
1348 adjust the pointer again. */
1355 if (!vtable_entry_type
)
1357 tree vfunc_type
= make_node (FUNCTION_TYPE
);
1358 TREE_TYPE (vfunc_type
) = integer_type_node
;
1359 TYPE_ARG_TYPES (vfunc_type
) = NULL_TREE
;
1360 layout_type (vfunc_type
);
1362 vtable_entry_type
= build_pointer_type (vfunc_type
);
1366 create_tmp_reg (build_pointer_type
1367 (build_pointer_type (vtable_entry_type
)), "vptr");
1369 /* The vptr is always at offset zero in the object. */
1370 stmt
= gimple_build_assign (vtabletmp
,
1371 build1 (NOP_EXPR
, TREE_TYPE (vtabletmp
),
1373 gsi_insert_after (bsi
, stmt
, GSI_NEW_STMT
);
1375 /* Form the vtable address. */
1376 vtabletmp2
= create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp
)),
1378 stmt
= gimple_build_assign (vtabletmp2
,
1379 build_simple_mem_ref (vtabletmp
));
1380 gsi_insert_after (bsi
, stmt
, GSI_NEW_STMT
);
1382 /* Find the entry with the vcall offset. */
1383 stmt
= gimple_build_assign (vtabletmp2
,
1384 fold_build_pointer_plus_loc (input_location
,
1387 gsi_insert_after (bsi
, stmt
, GSI_NEW_STMT
);
1389 /* Get the offset itself. */
1390 vtabletmp3
= create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp2
)),
1392 stmt
= gimple_build_assign (vtabletmp3
,
1393 build_simple_mem_ref (vtabletmp2
));
1394 gsi_insert_after (bsi
, stmt
, GSI_NEW_STMT
);
1396 /* Adjust the `this' pointer. */
1397 ptr
= fold_build_pointer_plus_loc (input_location
, ptr
, vtabletmp3
);
1398 ptr
= force_gimple_operand_gsi (bsi
, ptr
, true, NULL_TREE
, false,
1399 GSI_CONTINUE_LINKING
);
1403 && fixed_offset
!= 0)
1404 /* Adjust the pointer by the constant. */
1408 if (TREE_CODE (ptr
) == VAR_DECL
)
1412 ptrtmp
= create_tmp_reg (TREE_TYPE (ptr
), "ptr");
1413 stmt
= gimple_build_assign (ptrtmp
, ptr
);
1414 gsi_insert_after (bsi
, stmt
, GSI_NEW_STMT
);
1416 ptr
= fold_build_pointer_plus_hwi_loc (input_location
,
1417 ptrtmp
, fixed_offset
);
1420 /* Emit the statement and gimplify the adjustment expression. */
1421 ret
= create_tmp_reg (TREE_TYPE (ptr
), "adjusted_this");
1422 stmt
= gimple_build_assign (ret
, ptr
);
1423 gsi_insert_after (bsi
, stmt
, GSI_NEW_STMT
);
1428 /* Expand thunk NODE to gimple if possible.
1429 When FORCE_GIMPLE_THUNK is true, gimple thunk is created and
1430 no assembler is produced.
1431 When OUTPUT_ASM_THUNK is true, also produce assembler for
1432 thunks that are not lowered. */
1435 cgraph_node::expand_thunk (bool output_asm_thunks
, bool force_gimple_thunk
)
1437 bool this_adjusting
= thunk
.this_adjusting
;
1438 HOST_WIDE_INT fixed_offset
= thunk
.fixed_offset
;
1439 HOST_WIDE_INT virtual_value
= thunk
.virtual_value
;
1440 tree virtual_offset
= NULL
;
1441 tree alias
= callees
->callee
->decl
;
1442 tree thunk_fndecl
= decl
;
1446 if (!force_gimple_thunk
&& this_adjusting
1447 && targetm
.asm_out
.can_output_mi_thunk (thunk_fndecl
, fixed_offset
,
1448 virtual_value
, alias
))
1452 tree restype
= TREE_TYPE (TREE_TYPE (thunk_fndecl
));
1454 if (!output_asm_thunks
)
1462 a
= DECL_ARGUMENTS (thunk_fndecl
);
1464 current_function_decl
= thunk_fndecl
;
1466 /* Ensure thunks are emitted in their correct sections. */
1467 resolve_unique_section (thunk_fndecl
, 0, flag_function_sections
);
1469 DECL_RESULT (thunk_fndecl
)
1470 = build_decl (DECL_SOURCE_LOCATION (thunk_fndecl
),
1471 RESULT_DECL
, 0, restype
);
1472 DECL_CONTEXT (DECL_RESULT (thunk_fndecl
)) = thunk_fndecl
;
1473 fnname
= IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (thunk_fndecl
));
1475 /* The back end expects DECL_INITIAL to contain a BLOCK, so we
1477 fn_block
= make_node (BLOCK
);
1478 BLOCK_VARS (fn_block
) = a
;
1479 DECL_INITIAL (thunk_fndecl
) = fn_block
;
1480 init_function_start (thunk_fndecl
);
1482 insn_locations_init ();
1483 set_curr_insn_location (DECL_SOURCE_LOCATION (thunk_fndecl
));
1484 prologue_location
= curr_insn_location ();
1485 assemble_start_function (thunk_fndecl
, fnname
);
1487 targetm
.asm_out
.output_mi_thunk (asm_out_file
, thunk_fndecl
,
1488 fixed_offset
, virtual_value
, alias
);
1490 assemble_end_function (thunk_fndecl
, fnname
);
1491 insn_locations_finalize ();
1492 init_insn_lengths ();
1493 free_after_compilation (cfun
);
1495 TREE_ASM_WRITTEN (thunk_fndecl
) = 1;
1496 thunk
.thunk_p
= false;
1502 basic_block bb
, then_bb
, else_bb
, return_bb
;
1503 gimple_stmt_iterator bsi
;
1515 a
= DECL_ARGUMENTS (thunk_fndecl
);
1517 current_function_decl
= thunk_fndecl
;
1519 /* Ensure thunks are emitted in their correct sections. */
1520 resolve_unique_section (thunk_fndecl
, 0, flag_function_sections
);
1522 DECL_IGNORED_P (thunk_fndecl
) = 1;
1523 bitmap_obstack_initialize (NULL
);
1525 if (thunk
.virtual_offset_p
)
1526 virtual_offset
= size_int (virtual_value
);
1528 /* Build the return declaration for the function. */
1529 restype
= TREE_TYPE (TREE_TYPE (thunk_fndecl
));
1530 if (DECL_RESULT (thunk_fndecl
) == NULL_TREE
)
1532 resdecl
= build_decl (input_location
, RESULT_DECL
, 0, restype
);
1533 DECL_ARTIFICIAL (resdecl
) = 1;
1534 DECL_IGNORED_P (resdecl
) = 1;
1535 DECL_RESULT (thunk_fndecl
) = resdecl
;
1536 DECL_CONTEXT (DECL_RESULT (thunk_fndecl
)) = thunk_fndecl
;
1539 resdecl
= DECL_RESULT (thunk_fndecl
);
1541 bb
= then_bb
= else_bb
= return_bb
= init_lowered_empty_function (thunk_fndecl
, true);
1543 bsi
= gsi_start_bb (bb
);
1545 /* Build call to the function being thunked. */
1546 if (!VOID_TYPE_P (restype
))
1548 if (DECL_BY_REFERENCE (resdecl
))
1549 restmp
= gimple_fold_indirect_ref (resdecl
);
1550 else if (!is_gimple_reg_type (restype
))
1554 if (TREE_CODE (restmp
) == VAR_DECL
)
1555 add_local_decl (cfun
, restmp
);
1556 BLOCK_VARS (DECL_INITIAL (current_function_decl
)) = restmp
;
1559 restmp
= create_tmp_reg (restype
, "retval");
1562 for (arg
= a
; arg
; arg
= DECL_CHAIN (arg
))
1564 auto_vec
<tree
> vargs (nargs
);
1566 vargs
.quick_push (thunk_adjust (&bsi
, a
, 1, fixed_offset
,
1569 vargs
.quick_push (a
);
1572 for (i
= 1, arg
= DECL_CHAIN (a
); i
< nargs
; i
++, arg
= DECL_CHAIN (arg
))
1575 if (!is_gimple_val (arg
))
1577 tmp
= create_tmp_reg (TYPE_MAIN_VARIANT
1578 (TREE_TYPE (arg
)), "arg");
1579 gimple stmt
= gimple_build_assign (tmp
, arg
);
1580 gsi_insert_after (&bsi
, stmt
, GSI_NEW_STMT
);
1582 vargs
.quick_push (tmp
);
1584 call
= gimple_build_call_vec (build_fold_addr_expr_loc (0, alias
), vargs
);
1585 callees
->call_stmt
= call
;
1586 gimple_call_set_from_thunk (call
, true);
1589 gimple_call_set_lhs (call
, restmp
);
1590 gcc_assert (useless_type_conversion_p (TREE_TYPE (restmp
),
1591 TREE_TYPE (TREE_TYPE (alias
))));
1593 gsi_insert_after (&bsi
, call
, GSI_NEW_STMT
);
1594 if (!(gimple_call_flags (call
) & ECF_NORETURN
))
1596 if (restmp
&& !this_adjusting
1597 && (fixed_offset
|| virtual_offset
))
1599 tree true_label
= NULL_TREE
;
1601 if (TREE_CODE (TREE_TYPE (restmp
)) == POINTER_TYPE
)
1604 /* If the return type is a pointer, we need to
1605 protect against NULL. We know there will be an
1606 adjustment, because that's why we're emitting a
1608 then_bb
= create_basic_block (NULL
, (void *) 0, bb
);
1609 return_bb
= create_basic_block (NULL
, (void *) 0, then_bb
);
1610 else_bb
= create_basic_block (NULL
, (void *) 0, else_bb
);
1611 add_bb_to_loop (then_bb
, bb
->loop_father
);
1612 add_bb_to_loop (return_bb
, bb
->loop_father
);
1613 add_bb_to_loop (else_bb
, bb
->loop_father
);
1614 remove_edge (single_succ_edge (bb
));
1615 true_label
= gimple_block_label (then_bb
);
1616 stmt
= gimple_build_cond (NE_EXPR
, restmp
,
1617 build_zero_cst (TREE_TYPE (restmp
)),
1618 NULL_TREE
, NULL_TREE
);
1619 gsi_insert_after (&bsi
, stmt
, GSI_NEW_STMT
);
1620 make_edge (bb
, then_bb
, EDGE_TRUE_VALUE
);
1621 make_edge (bb
, else_bb
, EDGE_FALSE_VALUE
);
1622 make_edge (return_bb
, EXIT_BLOCK_PTR_FOR_FN (cfun
), 0);
1623 make_edge (then_bb
, return_bb
, EDGE_FALLTHRU
);
1624 make_edge (else_bb
, return_bb
, EDGE_FALLTHRU
);
1625 bsi
= gsi_last_bb (then_bb
);
1628 restmp
= thunk_adjust (&bsi
, restmp
, /*this_adjusting=*/0,
1629 fixed_offset
, virtual_offset
);
1633 bsi
= gsi_last_bb (else_bb
);
1634 stmt
= gimple_build_assign (restmp
,
1635 build_zero_cst (TREE_TYPE (restmp
)));
1636 gsi_insert_after (&bsi
, stmt
, GSI_NEW_STMT
);
1637 bsi
= gsi_last_bb (return_bb
);
1641 gimple_call_set_tail (call
, true);
1643 /* Build return value. */
1644 ret
= gimple_build_return (restmp
);
1645 gsi_insert_after (&bsi
, ret
, GSI_NEW_STMT
);
1649 gimple_call_set_tail (call
, true);
1650 remove_edge (single_succ_edge (bb
));
1653 cfun
->gimple_df
->in_ssa_p
= true;
1654 /* FIXME: C++ FE should stop setting TREE_ASM_WRITTEN on thunks. */
1655 TREE_ASM_WRITTEN (thunk_fndecl
) = false;
1656 delete_unreachable_blocks ();
1657 update_ssa (TODO_update_ssa
);
1658 #ifdef ENABLE_CHECKING
1659 verify_flow_info ();
1661 free_dominance_info (CDI_DOMINATORS
);
1663 /* Since we want to emit the thunk, we explicitly mark its name as
1665 thunk
.thunk_p
= false;
1667 bitmap_obstack_release (NULL
);
1669 current_function_decl
= NULL
;
1674 /* Assemble thunks and aliases associated to node. */
1677 cgraph_node::assemble_thunks_and_aliases (void)
1682 for (e
= callers
; e
;)
1683 if (e
->caller
->thunk
.thunk_p
)
1685 cgraph_node
*thunk
= e
->caller
;
1688 thunk
->expand_thunk (true, false);
1689 thunk
->assemble_thunks_and_aliases ();
1694 FOR_EACH_ALIAS (this, ref
)
1696 cgraph_node
*alias
= dyn_cast
<cgraph_node
*> (ref
->referring
);
1697 bool saved_written
= TREE_ASM_WRITTEN (decl
);
1699 /* Force assemble_alias to really output the alias this time instead
1700 of buffering it in same alias pairs. */
1701 TREE_ASM_WRITTEN (decl
) = 1;
1702 do_assemble_alias (alias
->decl
,
1703 DECL_ASSEMBLER_NAME (decl
));
1704 alias
->assemble_thunks_and_aliases ();
1705 TREE_ASM_WRITTEN (decl
) = saved_written
;
1709 /* Expand function specified by node. */
1712 cgraph_node::expand (void)
1714 location_t saved_loc
;
1716 /* We ought to not compile any inline clones. */
1717 gcc_assert (!global
.inlined_to
);
1719 announce_function (decl
);
1721 gcc_assert (lowered
);
1724 /* Generate RTL for the body of DECL. */
1726 timevar_push (TV_REST_OF_COMPILATION
);
1728 gcc_assert (symtab
->global_info_ready
);
1730 /* Initialize the default bitmap obstack. */
1731 bitmap_obstack_initialize (NULL
);
1733 /* Initialize the RTL code for the function. */
1734 current_function_decl
= decl
;
1735 saved_loc
= input_location
;
1736 input_location
= DECL_SOURCE_LOCATION (decl
);
1737 init_function_start (decl
);
1739 gimple_register_cfg_hooks ();
1741 bitmap_obstack_initialize (®_obstack
); /* FIXME, only at RTL generation*/
1743 execute_all_ipa_transforms ();
1745 /* Perform all tree transforms and optimizations. */
1747 /* Signal the start of passes. */
1748 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_START
, NULL
);
1750 execute_pass_list (cfun
, g
->get_passes ()->all_passes
);
1752 /* Signal the end of passes. */
1753 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_END
, NULL
);
1755 bitmap_obstack_release (®_obstack
);
1757 /* Release the default bitmap obstack. */
1758 bitmap_obstack_release (NULL
);
1760 /* If requested, warn about function definitions where the function will
1761 return a value (usually of some struct or union type) which itself will
1762 take up a lot of stack space. */
1763 if (warn_larger_than
&& !DECL_EXTERNAL (decl
) && TREE_TYPE (decl
))
1765 tree ret_type
= TREE_TYPE (TREE_TYPE (decl
));
1767 if (ret_type
&& TYPE_SIZE_UNIT (ret_type
)
1768 && TREE_CODE (TYPE_SIZE_UNIT (ret_type
)) == INTEGER_CST
1769 && 0 < compare_tree_int (TYPE_SIZE_UNIT (ret_type
),
1772 unsigned int size_as_int
1773 = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (ret_type
));
1775 if (compare_tree_int (TYPE_SIZE_UNIT (ret_type
), size_as_int
) == 0)
1776 warning (OPT_Wlarger_than_
, "size of return value of %q+D is %u bytes",
1779 warning (OPT_Wlarger_than_
, "size of return value of %q+D is larger than %wd bytes",
1780 decl
, larger_than_size
);
1784 gimple_set_body (decl
, NULL
);
1785 if (DECL_STRUCT_FUNCTION (decl
) == 0
1786 && !cgraph_node::get (decl
)->origin
)
1788 /* Stop pointing to the local nodes about to be freed.
1789 But DECL_INITIAL must remain nonzero so we know this
1790 was an actual function definition.
1791 For a nested function, this is done in c_pop_function_context.
1792 If rest_of_compilation set this to 0, leave it 0. */
1793 if (DECL_INITIAL (decl
) != 0)
1794 DECL_INITIAL (decl
) = error_mark_node
;
1797 input_location
= saved_loc
;
1800 timevar_pop (TV_REST_OF_COMPILATION
);
1802 /* Make sure that BE didn't give up on compiling. */
1803 gcc_assert (TREE_ASM_WRITTEN (decl
));
1805 current_function_decl
= NULL
;
1807 /* It would make a lot more sense to output thunks before function body to get more
1808 forward and lest backwarding jumps. This however would need solving problem
1809 with comdats. See PR48668. Also aliases must come after function itself to
1810 make one pass assemblers, like one on AIX, happy. See PR 50689.
1811 FIXME: Perhaps thunks should be move before function IFF they are not in comdat
1813 assemble_thunks_and_aliases ();
1815 /* Eliminate all call edges. This is important so the GIMPLE_CALL no longer
1816 points to the dead function body. */
1818 remove_all_references ();
1821 /* Node comparer that is responsible for the order that corresponds
1822 to time when a function was launched for the first time. */
1825 node_cmp (const void *pa
, const void *pb
)
1827 const cgraph_node
*a
= *(const cgraph_node
* const *) pa
;
1828 const cgraph_node
*b
= *(const cgraph_node
* const *) pb
;
1830 /* Functions with time profile must be before these without profile. */
1831 if (!a
->tp_first_run
|| !b
->tp_first_run
)
1832 return a
->tp_first_run
- b
->tp_first_run
;
1834 return a
->tp_first_run
!= b
->tp_first_run
1835 ? b
->tp_first_run
- a
->tp_first_run
1836 : b
->order
- a
->order
;
1839 /* Expand all functions that must be output.
1841 Attempt to topologically sort the nodes so function is output when
1842 all called functions are already assembled to allow data to be
1843 propagated across the callgraph. Use a stack to get smaller distance
1844 between a function and its callees (later we may choose to use a more
1845 sophisticated algorithm for function reordering; we will likely want
1846 to use subsections to make the output functions appear in top-down
1850 expand_all_functions (void)
1853 cgraph_node
**order
= XCNEWVEC (cgraph_node
*,
1854 symtab
->cgraph_count
);
1855 unsigned int expanded_func_count
= 0, profiled_func_count
= 0;
1856 int order_pos
, new_order_pos
= 0;
1859 order_pos
= ipa_reverse_postorder (order
);
1860 gcc_assert (order_pos
== symtab
->cgraph_count
);
1862 /* Garbage collector may remove inline clones we eliminate during
1863 optimization. So we must be sure to not reference them. */
1864 for (i
= 0; i
< order_pos
; i
++)
1865 if (order
[i
]->process
)
1866 order
[new_order_pos
++] = order
[i
];
1868 if (flag_profile_reorder_functions
)
1869 qsort (order
, new_order_pos
, sizeof (cgraph_node
*), node_cmp
);
1871 for (i
= new_order_pos
- 1; i
>= 0; i
--)
1877 expanded_func_count
++;
1878 if(node
->tp_first_run
)
1879 profiled_func_count
++;
1881 if (symtab
->dump_file
)
1882 fprintf (symtab
->dump_file
,
1883 "Time profile order in expand_all_functions:%s:%d\n",
1884 node
->asm_name (), node
->tp_first_run
);
1891 fprintf (dump_file
, "Expanded functions with time profile (%s):%u/%u\n",
1892 main_input_filename
, profiled_func_count
, expanded_func_count
);
1894 if (symtab
->dump_file
&& flag_profile_reorder_functions
)
1895 fprintf (symtab
->dump_file
, "Expanded functions with time profile:%u/%u\n",
1896 profiled_func_count
, expanded_func_count
);
1898 symtab
->process_new_functions ();
1899 free_gimplify_stack ();
1904 /* This is used to sort the node types by the cgraph order number. */
1906 enum cgraph_order_sort_kind
1908 ORDER_UNDEFINED
= 0,
1914 struct cgraph_order_sort
1916 enum cgraph_order_sort_kind kind
;
1925 /* Output all functions, variables, and asm statements in the order
1926 according to their order fields, which is the order in which they
1927 appeared in the file. This implements -fno-toplevel-reorder. In
1928 this mode we may output functions and variables which don't really
1930 When NO_REORDER is true only do this for symbols marked no reorder. */
1933 output_in_order (bool no_reorder
)
1936 cgraph_order_sort
*nodes
;
1941 max
= symtab
->order
;
1942 nodes
= XCNEWVEC (cgraph_order_sort
, max
);
1944 FOR_EACH_DEFINED_FUNCTION (pf
)
1946 if (pf
->process
&& !pf
->thunk
.thunk_p
&& !pf
->alias
)
1948 if (no_reorder
&& !pf
->no_reorder
)
1951 gcc_assert (nodes
[i
].kind
== ORDER_UNDEFINED
);
1952 nodes
[i
].kind
= ORDER_FUNCTION
;
1957 FOR_EACH_DEFINED_VARIABLE (pv
)
1958 if (!DECL_EXTERNAL (pv
->decl
))
1960 if (no_reorder
&& !pv
->no_reorder
)
1963 gcc_assert (nodes
[i
].kind
== ORDER_UNDEFINED
);
1964 nodes
[i
].kind
= ORDER_VAR
;
1968 for (pa
= symtab
->first_asm_symbol (); pa
; pa
= pa
->next
)
1971 gcc_assert (nodes
[i
].kind
== ORDER_UNDEFINED
);
1972 nodes
[i
].kind
= ORDER_ASM
;
1976 /* In toplevel reorder mode we output all statics; mark them as needed. */
1978 for (i
= 0; i
< max
; ++i
)
1979 if (nodes
[i
].kind
== ORDER_VAR
)
1980 nodes
[i
].u
.v
->finalize_named_section_flags ();
1982 for (i
= 0; i
< max
; ++i
)
1984 switch (nodes
[i
].kind
)
1986 case ORDER_FUNCTION
:
1987 nodes
[i
].u
.f
->process
= 0;
1988 nodes
[i
].u
.f
->expand ();
1992 nodes
[i
].u
.v
->assemble_decl ();
1996 assemble_asm (nodes
[i
].u
.a
->asm_str
);
1999 case ORDER_UNDEFINED
:
2007 symtab
->clear_asm_symbols ();
2015 gcc::pass_manager
*passes
= g
->get_passes ();
2018 current_function_decl
= NULL
;
2019 gimple_register_cfg_hooks ();
2020 bitmap_obstack_initialize (NULL
);
2022 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_START
, NULL
);
2026 execute_ipa_pass_list (passes
->all_small_ipa_passes
);
2031 /* This extra symtab_remove_unreachable_nodes pass tends to catch some
2032 devirtualization and other changes where removal iterate. */
2033 symtab
->remove_unreachable_nodes (true, symtab
->dump_file
);
2035 /* If pass_all_early_optimizations was not scheduled, the state of
2036 the cgraph will not be properly updated. Update it now. */
2037 if (symtab
->state
< IPA_SSA
)
2038 symtab
->state
= IPA_SSA
;
2042 /* Generate coverage variables and constructors. */
2045 /* Process new functions added. */
2047 current_function_decl
= NULL
;
2048 symtab
->process_new_functions ();
2050 execute_ipa_summary_passes
2051 ((ipa_opt_pass_d
*) passes
->all_regular_ipa_passes
);
2054 /* Some targets need to handle LTO assembler output specially. */
2055 if (flag_generate_lto
)
2056 targetm
.asm_out
.lto_start ();
2059 ipa_write_summaries ();
2061 if (flag_generate_lto
)
2062 targetm
.asm_out
.lto_end ();
2064 if (!flag_ltrans
&& (in_lto_p
|| !flag_lto
|| flag_fat_lto_objects
))
2065 execute_ipa_pass_list (passes
->all_regular_ipa_passes
);
2066 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_END
, NULL
);
2068 bitmap_obstack_release (NULL
);
2072 /* Return string alias is alias of. */
2075 get_alias_symbol (tree decl
)
2077 tree alias
= lookup_attribute ("alias", DECL_ATTRIBUTES (decl
));
2078 return get_identifier (TREE_STRING_POINTER
2079 (TREE_VALUE (TREE_VALUE (alias
))));
2083 /* Weakrefs may be associated to external decls and thus not output
2084 at expansion time. Emit all necessary aliases. */
2087 symbol_table::output_weakrefs (void)
2090 FOR_EACH_SYMBOL (node
)
2092 && !TREE_ASM_WRITTEN (node
->decl
)
2097 /* Weakrefs are special by not requiring target definition in current
2098 compilation unit. It is thus bit hard to work out what we want to
2100 When alias target is defined, we need to fetch it from symtab reference,
2101 otherwise it is pointed to by alias_target. */
2102 if (node
->alias_target
)
2103 target
= (DECL_P (node
->alias_target
)
2104 ? DECL_ASSEMBLER_NAME (node
->alias_target
)
2105 : node
->alias_target
);
2106 else if (node
->analyzed
)
2107 target
= DECL_ASSEMBLER_NAME (node
->get_alias_target ()->decl
);
2111 target
= get_alias_symbol (node
->decl
);
2113 do_assemble_alias (node
->decl
, target
);
2117 /* Perform simple optimizations based on callgraph. */
2120 symbol_table::compile (void)
2125 #ifdef ENABLE_CHECKING
2126 symtab_node::verify_symtab_nodes ();
2129 timevar_push (TV_CGRAPHOPT
);
2130 if (pre_ipa_mem_report
)
2132 fprintf (stderr
, "Memory consumption before IPA\n");
2133 dump_memory_report (false);
2136 fprintf (stderr
, "Performing interprocedural optimizations\n");
2139 /* If LTO is enabled, initialize the streamer hooks needed by GIMPLE. */
2141 lto_streamer_hooks_init ();
2143 /* Don't run the IPA passes if there was any error or sorry messages. */
2147 /* Do nothing else if any IPA pass found errors or if we are just streaming LTO. */
2149 || (!in_lto_p
&& flag_lto
&& !flag_fat_lto_objects
))
2151 timevar_pop (TV_CGRAPHOPT
);
2155 /* This pass remove bodies of extern inline functions we never inlined.
2156 Do this later so other IPA passes see what is really going on.
2157 FIXME: This should be run just after inlining by pasmanager. */
2158 remove_unreachable_nodes (false, dump_file
);
2159 global_info_ready
= true;
2162 fprintf (dump_file
, "Optimized ");
2163 symtab_node:: dump_table (dump_file
);
2165 if (post_ipa_mem_report
)
2167 fprintf (stderr
, "Memory consumption after IPA\n");
2168 dump_memory_report (false);
2170 timevar_pop (TV_CGRAPHOPT
);
2172 /* Output everything. */
2173 (*debug_hooks
->assembly_start
) ();
2175 fprintf (stderr
, "Assembling functions:\n");
2176 #ifdef ENABLE_CHECKING
2177 symtab_node::verify_symtab_nodes ();
2180 materialize_all_clones ();
2181 bitmap_obstack_initialize (NULL
);
2182 execute_ipa_pass_list (g
->get_passes ()->all_late_ipa_passes
);
2183 bitmap_obstack_release (NULL
);
2184 mark_functions_to_output ();
2186 /* When weakref support is missing, we autmatically translate all
2187 references to NODE to references to its ultimate alias target.
2188 The renaming mechanizm uses flag IDENTIFIER_TRANSPARENT_ALIAS and
2191 Set up this mapping before we output any assembler but once we are sure
2192 that all symbol renaming is done.
2194 FIXME: All this uglyness can go away if we just do renaming at gimple
2195 level by physically rewritting the IL. At the moment we can only redirect
2196 calls, so we need infrastructure for renaming references as well. */
2197 #ifndef ASM_OUTPUT_WEAKREF
2200 FOR_EACH_SYMBOL (node
)
2202 && lookup_attribute ("weakref", DECL_ATTRIBUTES (node
->decl
)))
2204 IDENTIFIER_TRANSPARENT_ALIAS
2205 (DECL_ASSEMBLER_NAME (node
->decl
)) = 1;
2206 TREE_CHAIN (DECL_ASSEMBLER_NAME (node
->decl
))
2207 = (node
->alias_target
? node
->alias_target
2208 : DECL_ASSEMBLER_NAME (node
->get_alias_target ()->decl
));
2214 if (!flag_toplevel_reorder
)
2215 output_in_order (false);
2218 /* Output first asm statements and anything ordered. The process
2219 flag is cleared for these nodes, so we skip them later. */
2220 output_in_order (true);
2221 expand_all_functions ();
2222 output_variables ();
2225 process_new_functions ();
2231 fprintf (dump_file
, "\nFinal ");
2232 symtab_node::dump_table (dump_file
);
2234 #ifdef ENABLE_CHECKING
2235 symtab_node::verify_symtab_nodes ();
2236 /* Double check that all inline clones are gone and that all
2237 function bodies have been released from memory. */
2241 bool error_found
= false;
2243 FOR_EACH_DEFINED_FUNCTION (node
)
2244 if (node
->global
.inlined_to
2245 || gimple_has_body_p (node
->decl
))
2251 internal_error ("nodes with unreleased memory found");
2257 /* Analyze the whole compilation unit once it is parsed completely. */
2260 symbol_table::finalize_compilation_unit (void)
2262 timevar_push (TV_CGRAPH
);
2264 /* If we're here there's no current function anymore. Some frontends
2265 are lazy in clearing these. */
2266 current_function_decl
= NULL
;
2269 /* Do not skip analyzing the functions if there were errors, we
2270 miss diagnostics for following functions otherwise. */
2272 /* Emit size functions we didn't inline. */
2273 finalize_size_functions ();
2275 /* Mark alias targets necessary and emit diagnostics. */
2276 handle_alias_pairs ();
2280 fprintf (stderr
, "\nAnalyzing compilation unit\n");
2284 if (flag_dump_passes
)
2287 /* Gimplify and lower all functions, compute reachability and
2288 remove unreachable nodes. */
2289 analyze_functions ();
2291 /* Mark alias targets necessary and emit diagnostics. */
2292 handle_alias_pairs ();
2294 /* Gimplify and lower thunks. */
2295 analyze_functions ();
2297 /* Finally drive the pass manager. */
2300 timevar_pop (TV_CGRAPH
);
2303 /* Reset all state within cgraphunit.c so that we can rerun the compiler
2304 within the same process. For use by toplev::finalize. */
2307 cgraphunit_c_finalize (void)
2309 gcc_assert (cgraph_new_nodes
.length () == 0);
2310 cgraph_new_nodes
.truncate (0);
2312 vtable_entry_type
= NULL
;
2313 queued_nodes
= &symtab_terminator
;
2315 first_analyzed
= NULL
;
2316 first_analyzed_var
= NULL
;
2319 /* Creates a wrapper from cgraph_node to TARGET node. Thunk is used for this
2320 kind of wrapper method. */
2323 cgraph_node::create_wrapper (cgraph_node
*target
)
2325 /* Preserve DECL_RESULT so we get right by reference flag. */
2326 tree decl_result
= DECL_RESULT (decl
);
2328 /* Remove the function's body but keep arguments to be reused
2330 release_body (true);
2333 DECL_RESULT (decl
) = decl_result
;
2334 DECL_INITIAL (decl
) = NULL
;
2335 allocate_struct_function (decl
, false);
2338 /* Turn alias into thunk and expand it into GIMPLE representation. */
2340 thunk
.thunk_p
= true;
2341 thunk
.this_adjusting
= false;
2343 cgraph_edge
*e
= create_edge (target
, NULL
, 0, CGRAPH_FREQ_BASE
);
2345 expand_thunk (false, true);
2346 e
->call_stmt_cannot_inline_p
= true;
2348 /* Inline summary set-up. */
2350 inline_analyze_function (this);
2353 #include "gt-cgraphunit.h"