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
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 provide facilities for clonning functions. I.e. creating
22 new functions based on existing functions with simple modifications,
23 such as replacement of parameters.
25 To allow whole program optimization without actual presence of function
26 bodies, an additional infrastructure is provided for so-called virtual
29 A virtual clone in the callgraph is a function that has no
30 associated body, just a description of how to create its body based
31 on a different function (which itself may be a virtual clone).
33 The description of function modifications includes adjustments to
34 the function's signature (which allows, for example, removing or
35 adding function arguments), substitutions to perform on the
36 function body, and, for inlined functions, a pointer to the
37 function that it will be inlined into.
39 It is also possible to redirect any edge of the callgraph from a
40 function to its virtual clone. This implies updating of the call
41 site to adjust for the new function signature.
43 Most of the transformations performed by inter-procedural
44 optimizations can be represented via virtual clones. For
45 instance, a constant propagation pass can produce a virtual clone
46 of the function which replaces one of its arguments by a
47 constant. The inliner can represent its decisions by producing a
48 clone of a function whose body will be later integrated into
51 Using virtual clones, the program can be easily updated
52 during the Execute stage, solving most of pass interactions
53 problems that would otherwise occur during Transform.
55 Virtual clones are later materialized in the LTRANS stage and
56 turned into real functions. Passes executed after the virtual
57 clone were introduced also perform their Transform stage
58 on new functions, so for a pass there is no significant
59 difference between operating on a real function or a virtual
60 clone introduced before its Execute stage.
62 Optimization passes then work on virtual clones introduced before
63 their Execute stage as if they were real functions. The
64 only difference is that clones are not visible during the
65 Generate Summary stage. */
69 #include "coretypes.h"
75 #include "fold-const.h"
76 #include "stringpool.h"
78 #include "internal-fn.h"
81 #include "tree-inline.h"
82 #include "langhooks.h"
87 #include "diagnostic.h"
91 #include "alloc-pool.h"
92 #include "symbol-summary.h"
94 #include "tree-iterator.h"
95 #include "tree-dump.h"
96 #include "gimple-pretty-print.h"
98 #include "ipa-inline.h"
99 #include "ipa-utils.h"
100 #include "lto-streamer.h"
103 /* Create clone of edge in the node N represented by CALL_EXPR
107 cgraph_edge::clone (cgraph_node
*n
, gcall
*call_stmt
, unsigned stmt_uid
,
108 gcov_type count_scale
, int freq_scale
, bool update_original
)
110 cgraph_edge
*new_edge
;
111 gcov_type gcov_count
= apply_probability (count
, count_scale
);
114 /* We do not want to ignore loop nest after frequency drops to 0. */
117 freq
= frequency
* (gcov_type
) freq_scale
/ CGRAPH_FREQ_BASE
;
118 if (freq
> CGRAPH_FREQ_MAX
)
119 freq
= CGRAPH_FREQ_MAX
;
121 if (indirect_unknown_callee
)
125 if (call_stmt
&& (decl
= gimple_call_fndecl (call_stmt
))
126 /* When the call is speculative, we need to resolve it
127 via cgraph_resolve_speculation and not here. */
130 cgraph_node
*callee
= cgraph_node::get (decl
);
131 gcc_checking_assert (callee
);
132 new_edge
= n
->create_edge (callee
, call_stmt
, gcov_count
, freq
);
136 new_edge
= n
->create_indirect_edge (call_stmt
,
137 indirect_info
->ecf_flags
,
139 *new_edge
->indirect_info
= *indirect_info
;
144 new_edge
= n
->create_edge (callee
, call_stmt
, gcov_count
, freq
);
147 new_edge
->indirect_info
148 = ggc_cleared_alloc
<cgraph_indirect_call_info
> ();
149 *new_edge
->indirect_info
= *indirect_info
;
153 new_edge
->inline_failed
= inline_failed
;
154 new_edge
->indirect_inlining_edge
= indirect_inlining_edge
;
155 new_edge
->lto_stmt_uid
= stmt_uid
;
156 /* Clone flags that depend on call_stmt availability manually. */
157 new_edge
->can_throw_external
= can_throw_external
;
158 new_edge
->call_stmt_cannot_inline_p
= call_stmt_cannot_inline_p
;
159 new_edge
->speculative
= speculative
;
160 new_edge
->in_polymorphic_cdtor
= in_polymorphic_cdtor
;
163 count
-= new_edge
->count
;
167 symtab
->call_edge_duplication_hooks (this, new_edge
);
171 /* Build variant of function type ORIG_TYPE skipping ARGS_TO_SKIP and the
172 return value if SKIP_RETURN is true. */
175 build_function_type_skip_args (tree orig_type
, bitmap args_to_skip
,
178 tree new_type
= NULL
;
179 tree args
, new_args
= NULL
;
183 for (args
= TYPE_ARG_TYPES (orig_type
); args
&& args
!= void_list_node
;
184 args
= TREE_CHAIN (args
), i
++)
185 if (!args_to_skip
|| !bitmap_bit_p (args_to_skip
, i
))
186 new_args
= tree_cons (NULL_TREE
, TREE_VALUE (args
), new_args
);
188 new_reversed
= nreverse (new_args
);
192 TREE_CHAIN (new_args
) = void_list_node
;
194 new_reversed
= void_list_node
;
197 /* Use copy_node to preserve as much as possible from original type
198 (debug info, attribute lists etc.)
199 Exception is METHOD_TYPEs must have THIS argument.
200 When we are asked to remove it, we need to build new FUNCTION_TYPE
202 if (TREE_CODE (orig_type
) != METHOD_TYPE
204 || !bitmap_bit_p (args_to_skip
, 0))
206 new_type
= build_distinct_type_copy (orig_type
);
207 TYPE_ARG_TYPES (new_type
) = new_reversed
;
212 = build_distinct_type_copy (build_function_type (TREE_TYPE (orig_type
),
214 TYPE_CONTEXT (new_type
) = TYPE_CONTEXT (orig_type
);
218 TREE_TYPE (new_type
) = void_type_node
;
223 /* Build variant of function decl ORIG_DECL skipping ARGS_TO_SKIP and the
224 return value if SKIP_RETURN is true.
226 Arguments from DECL_ARGUMENTS list can't be removed now, since they are
227 linked by TREE_CHAIN directly. The caller is responsible for eliminating
228 them when they are being duplicated (i.e. copy_arguments_for_versioning). */
231 build_function_decl_skip_args (tree orig_decl
, bitmap args_to_skip
,
234 tree new_decl
= copy_node (orig_decl
);
237 new_type
= TREE_TYPE (orig_decl
);
238 if (prototype_p (new_type
)
239 || (skip_return
&& !VOID_TYPE_P (TREE_TYPE (new_type
))))
241 = build_function_type_skip_args (new_type
, args_to_skip
, skip_return
);
242 TREE_TYPE (new_decl
) = new_type
;
244 /* For declarations setting DECL_VINDEX (i.e. methods)
245 we expect first argument to be THIS pointer. */
246 if (args_to_skip
&& bitmap_bit_p (args_to_skip
, 0))
247 DECL_VINDEX (new_decl
) = NULL_TREE
;
249 /* When signature changes, we need to clear builtin info. */
250 if (DECL_BUILT_IN (new_decl
)
252 && !bitmap_empty_p (args_to_skip
))
254 DECL_BUILT_IN_CLASS (new_decl
) = NOT_BUILT_IN
;
255 DECL_FUNCTION_CODE (new_decl
) = (enum built_in_function
) 0;
257 /* The FE might have information and assumptions about the other
259 DECL_LANG_SPECIFIC (new_decl
) = NULL
;
263 /* Set flags of NEW_NODE and its decl. NEW_NODE is a newly created private
264 clone or its thunk. */
267 set_new_clone_decl_and_node_flags (cgraph_node
*new_node
)
269 DECL_EXTERNAL (new_node
->decl
) = 0;
270 TREE_PUBLIC (new_node
->decl
) = 0;
271 DECL_COMDAT (new_node
->decl
) = 0;
272 DECL_WEAK (new_node
->decl
) = 0;
273 DECL_VIRTUAL_P (new_node
->decl
) = 0;
274 DECL_STATIC_CONSTRUCTOR (new_node
->decl
) = 0;
275 DECL_STATIC_DESTRUCTOR (new_node
->decl
) = 0;
277 new_node
->externally_visible
= 0;
278 new_node
->local
.local
= 1;
279 new_node
->lowered
= true;
282 /* Duplicate thunk THUNK if necessary but make it to refer to NODE.
283 ARGS_TO_SKIP, if non-NULL, determines which parameters should be omitted.
284 Function can return NODE if no thunk is necessary, which can happen when
285 thunk is this_adjusting but we are removing this parameter. */
288 duplicate_thunk_for_node (cgraph_node
*thunk
, cgraph_node
*node
)
290 cgraph_node
*new_thunk
, *thunk_of
;
291 thunk_of
= thunk
->callees
->callee
->ultimate_alias_target ();
293 if (thunk_of
->thunk
.thunk_p
)
294 node
= duplicate_thunk_for_node (thunk_of
, node
);
296 if (!DECL_ARGUMENTS (thunk
->decl
))
297 thunk
->get_untransformed_body ();
300 for (cs
= node
->callers
; cs
; cs
= cs
->next_caller
)
301 if (cs
->caller
->thunk
.thunk_p
302 && cs
->caller
->thunk
.this_adjusting
== thunk
->thunk
.this_adjusting
303 && cs
->caller
->thunk
.fixed_offset
== thunk
->thunk
.fixed_offset
304 && cs
->caller
->thunk
.virtual_offset_p
== thunk
->thunk
.virtual_offset_p
305 && cs
->caller
->thunk
.virtual_value
== thunk
->thunk
.virtual_value
)
309 if (!node
->clone
.args_to_skip
)
310 new_decl
= copy_node (thunk
->decl
);
313 /* We do not need to duplicate this_adjusting thunks if we have removed
315 if (thunk
->thunk
.this_adjusting
316 && bitmap_bit_p (node
->clone
.args_to_skip
, 0))
319 new_decl
= build_function_decl_skip_args (thunk
->decl
,
320 node
->clone
.args_to_skip
,
324 tree
*link
= &DECL_ARGUMENTS (new_decl
);
326 for (tree pd
= DECL_ARGUMENTS (thunk
->decl
); pd
; pd
= DECL_CHAIN (pd
), i
++)
328 if (!node
->clone
.args_to_skip
329 || !bitmap_bit_p (node
->clone
.args_to_skip
, i
))
331 tree nd
= copy_node (pd
);
332 DECL_CONTEXT (nd
) = new_decl
;
334 link
= &DECL_CHAIN (nd
);
339 gcc_checking_assert (!DECL_STRUCT_FUNCTION (new_decl
));
340 gcc_checking_assert (!DECL_INITIAL (new_decl
));
341 gcc_checking_assert (!DECL_RESULT (new_decl
));
342 gcc_checking_assert (!DECL_RTL_SET_P (new_decl
));
344 DECL_NAME (new_decl
) = clone_function_name (thunk
->decl
, "artificial_thunk");
345 SET_DECL_ASSEMBLER_NAME (new_decl
, DECL_NAME (new_decl
));
347 new_thunk
= cgraph_node::create (new_decl
);
348 set_new_clone_decl_and_node_flags (new_thunk
);
349 new_thunk
->definition
= true;
350 new_thunk
->thunk
= thunk
->thunk
;
351 new_thunk
->unique_name
= in_lto_p
;
352 new_thunk
->former_clone_of
= thunk
->decl
;
353 new_thunk
->clone
.args_to_skip
= node
->clone
.args_to_skip
;
354 new_thunk
->clone
.combined_args_to_skip
= node
->clone
.combined_args_to_skip
;
356 cgraph_edge
*e
= new_thunk
->create_edge (node
, NULL
, 0,
358 e
->call_stmt_cannot_inline_p
= true;
359 symtab
->call_edge_duplication_hooks (thunk
->callees
, e
);
360 symtab
->call_cgraph_duplication_hooks (thunk
, new_thunk
);
364 /* If E does not lead to a thunk, simply redirect it to N. Otherwise create
365 one or more equivalent thunks for N and redirect E to the first in the
366 chain. Note that it is then necessary to call
367 n->expand_all_artificial_thunks once all callers are redirected. */
370 cgraph_edge::redirect_callee_duplicating_thunks (cgraph_node
*n
)
372 cgraph_node
*orig_to
= callee
->ultimate_alias_target ();
373 if (orig_to
->thunk
.thunk_p
)
374 n
= duplicate_thunk_for_node (orig_to
, n
);
379 /* Call expand_thunk on all callers that are thunks and if analyze those nodes
380 that were expanded. */
383 cgraph_node::expand_all_artificial_thunks ()
386 for (e
= callers
; e
;)
387 if (e
->caller
->thunk
.thunk_p
)
389 cgraph_node
*thunk
= e
->caller
;
392 if (thunk
->expand_thunk (false, false))
394 thunk
->thunk
.thunk_p
= false;
397 thunk
->expand_all_artificial_thunks ();
403 /* Create node representing clone of N executed COUNT times. Decrease
404 the execution counts from original node too.
405 The new clone will have decl set to DECL that may or may not be the same
408 When UPDATE_ORIGINAL is true, the counts are subtracted from the original
409 function's profile to reflect the fact that part of execution is handled
411 When CALL_DUPLICATOIN_HOOK is true, the ipa passes are acknowledged about
412 the new clone. Otherwise the caller is responsible for doing so later.
414 If the new node is being inlined into another one, NEW_INLINED_TO should be
415 the outline function the new one is (even indirectly) inlined to. All hooks
416 will see this in node's global.inlined_to, when invoked. Can be NULL if the
417 node is not inlined. */
420 cgraph_node::create_clone (tree new_decl
, gcov_type gcov_count
, int freq
,
421 bool update_original
,
422 vec
<cgraph_edge
*> redirect_callers
,
423 bool call_duplication_hook
,
424 cgraph_node
*new_inlined_to
,
427 cgraph_node
*new_node
= symtab
->create_empty ();
429 gcov_type count_scale
;
432 new_node
->decl
= new_decl
;
433 new_node
->register_symbol ();
434 new_node
->origin
= origin
;
435 new_node
->lto_file_data
= lto_file_data
;
436 if (new_node
->origin
)
438 new_node
->next_nested
= new_node
->origin
->nested
;
439 new_node
->origin
->nested
= new_node
;
441 new_node
->analyzed
= analyzed
;
442 new_node
->definition
= definition
;
443 new_node
->local
= local
;
444 new_node
->externally_visible
= false;
445 new_node
->no_reorder
= no_reorder
;
446 new_node
->local
.local
= true;
447 new_node
->global
= global
;
448 new_node
->global
.inlined_to
= new_inlined_to
;
450 new_node
->count
= count
;
451 new_node
->frequency
= frequency
;
452 new_node
->tp_first_run
= tp_first_run
;
453 new_node
->tm_clone
= tm_clone
;
454 new_node
->icf_merged
= icf_merged
;
455 new_node
->merged
= merged
;
457 new_node
->clone
.tree_map
= NULL
;
458 new_node
->clone
.args_to_skip
= args_to_skip
;
459 new_node
->split_part
= split_part
;
461 new_node
->clone
.combined_args_to_skip
= clone
.combined_args_to_skip
;
462 else if (clone
.combined_args_to_skip
)
464 new_node
->clone
.combined_args_to_skip
= BITMAP_GGC_ALLOC ();
465 bitmap_ior (new_node
->clone
.combined_args_to_skip
,
466 clone
.combined_args_to_skip
, args_to_skip
);
469 new_node
->clone
.combined_args_to_skip
= args_to_skip
;
473 if (new_node
->count
> count
)
474 count_scale
= REG_BR_PROB_BASE
;
476 count_scale
= GCOV_COMPUTE_SCALE (new_node
->count
, count
);
487 FOR_EACH_VEC_ELT (redirect_callers
, i
, e
)
489 /* Redirect calls to the old version node to point to its new
490 version. The only exception is when the edge was proved to
491 be unreachable during the clonning procedure. */
493 || DECL_BUILT_IN_CLASS (e
->callee
->decl
) != BUILT_IN_NORMAL
494 || DECL_FUNCTION_CODE (e
->callee
->decl
) != BUILT_IN_UNREACHABLE
)
495 e
->redirect_callee_duplicating_thunks (new_node
);
497 new_node
->expand_all_artificial_thunks ();
499 for (e
= callees
;e
; e
=e
->next_callee
)
500 e
->clone (new_node
, e
->call_stmt
, e
->lto_stmt_uid
, count_scale
,
501 freq
, update_original
);
503 for (e
= indirect_calls
; e
; e
= e
->next_callee
)
504 e
->clone (new_node
, e
->call_stmt
, e
->lto_stmt_uid
,
505 count_scale
, freq
, update_original
);
506 new_node
->clone_references (this);
508 new_node
->next_sibling_clone
= clones
;
510 clones
->prev_sibling_clone
= new_node
;
512 new_node
->clone_of
= this;
514 if (call_duplication_hook
)
515 symtab
->call_cgraph_duplication_hooks (this, new_node
);
519 static GTY(()) unsigned int clone_fn_id_num
;
521 /* Return a new assembler name for a clone with SUFFIX of a decl named
525 clone_function_name_1 (const char *name
, const char *suffix
)
527 size_t len
= strlen (name
);
528 char *tmp_name
, *prefix
;
530 prefix
= XALLOCAVEC (char, len
+ strlen (suffix
) + 2);
531 memcpy (prefix
, name
, len
);
532 strcpy (prefix
+ len
+ 1, suffix
);
533 #ifndef NO_DOT_IN_LABEL
535 #elif !defined NO_DOLLAR_IN_LABEL
540 ASM_FORMAT_PRIVATE_NAME (tmp_name
, prefix
, clone_fn_id_num
++);
541 return get_identifier (tmp_name
);
544 /* Return a new assembler name for a clone of DECL with SUFFIX. */
547 clone_function_name (tree decl
, const char *suffix
)
549 tree name
= DECL_ASSEMBLER_NAME (decl
);
550 return clone_function_name_1 (IDENTIFIER_POINTER (name
), suffix
);
554 /* Create callgraph node clone with new declaration. The actual body will
555 be copied later at compilation stage.
557 TODO: after merging in ipa-sra use function call notes instead of args_to_skip
561 cgraph_node::create_virtual_clone (vec
<cgraph_edge
*> redirect_callers
,
562 vec
<ipa_replace_map
*, va_gc
> *tree_map
,
563 bitmap args_to_skip
, const char * suffix
)
565 tree old_decl
= decl
;
566 cgraph_node
*new_node
= NULL
;
569 ipa_replace_map
*map
;
573 gcc_checking_assert (tree_versionable_function_p (old_decl
));
575 gcc_assert (local
.can_change_signature
|| !args_to_skip
);
577 /* Make a new FUNCTION_DECL tree node */
579 new_decl
= copy_node (old_decl
);
581 new_decl
= build_function_decl_skip_args (old_decl
, args_to_skip
, false);
583 /* These pointers represent function body and will be populated only when clone
585 gcc_assert (new_decl
!= old_decl
);
586 DECL_STRUCT_FUNCTION (new_decl
) = NULL
;
587 DECL_ARGUMENTS (new_decl
) = NULL
;
588 DECL_INITIAL (new_decl
) = NULL
;
589 DECL_RESULT (new_decl
) = NULL
;
590 /* We can not do DECL_RESULT (new_decl) = NULL; here because of LTO partitioning
591 sometimes storing only clone decl instead of original. */
593 /* Generate a new name for the new version. */
594 len
= IDENTIFIER_LENGTH (DECL_NAME (old_decl
));
595 name
= XALLOCAVEC (char, len
+ strlen (suffix
) + 2);
596 memcpy (name
, IDENTIFIER_POINTER (DECL_NAME (old_decl
)), len
);
597 strcpy (name
+ len
+ 1, suffix
);
599 DECL_NAME (new_decl
) = get_identifier (name
);
600 SET_DECL_ASSEMBLER_NAME (new_decl
, clone_function_name (old_decl
, suffix
));
601 SET_DECL_RTL (new_decl
, NULL
);
603 new_node
= create_clone (new_decl
, count
, CGRAPH_FREQ_BASE
, false,
604 redirect_callers
, false, NULL
, args_to_skip
);
606 /* Update the properties.
607 Make clone visible only within this translation unit. Make sure
608 that is not weak also.
609 ??? We cannot use COMDAT linkage because there is no
610 ABI support for this. */
611 set_new_clone_decl_and_node_flags (new_node
);
612 new_node
->clone
.tree_map
= tree_map
;
613 if (!implicit_section
)
614 new_node
->set_section (get_section ());
616 /* Clones of global symbols or symbols with unique names are unique. */
617 if ((TREE_PUBLIC (old_decl
)
618 && !DECL_EXTERNAL (old_decl
)
619 && !DECL_WEAK (old_decl
)
620 && !DECL_COMDAT (old_decl
))
622 new_node
->unique_name
= true;
623 FOR_EACH_VEC_SAFE_ELT (tree_map
, i
, map
)
624 new_node
->maybe_create_reference (map
->new_tree
, IPA_REF_ADDR
, NULL
);
626 if (ipa_transforms_to_apply
.exists ())
627 new_node
->ipa_transforms_to_apply
628 = ipa_transforms_to_apply
.copy ();
630 symtab
->call_cgraph_duplication_hooks (this, new_node
);
635 /* callgraph node being removed from symbol table; see if its entry can be
636 replaced by other inline clone. */
638 cgraph_node::find_replacement (void)
640 cgraph_node
*next_inline_clone
, *replacement
;
642 for (next_inline_clone
= clones
;
644 && next_inline_clone
->decl
!= decl
;
645 next_inline_clone
= next_inline_clone
->next_sibling_clone
)
648 /* If there is inline clone of the node being removed, we need
649 to put it into the position of removed node and reorganize all
650 other clones to be based on it. */
651 if (next_inline_clone
)
654 cgraph_node
*new_clones
;
656 replacement
= next_inline_clone
;
658 /* Unlink inline clone from the list of clones of removed node. */
659 if (next_inline_clone
->next_sibling_clone
)
660 next_inline_clone
->next_sibling_clone
->prev_sibling_clone
661 = next_inline_clone
->prev_sibling_clone
;
662 if (next_inline_clone
->prev_sibling_clone
)
664 gcc_assert (clones
!= next_inline_clone
);
665 next_inline_clone
->prev_sibling_clone
->next_sibling_clone
666 = next_inline_clone
->next_sibling_clone
;
670 gcc_assert (clones
== next_inline_clone
);
671 clones
= next_inline_clone
->next_sibling_clone
;
677 /* Copy clone info. */
678 next_inline_clone
->clone
= clone
;
680 /* Now place it into clone tree at same level at NODE. */
681 next_inline_clone
->clone_of
= clone_of
;
682 next_inline_clone
->prev_sibling_clone
= NULL
;
683 next_inline_clone
->next_sibling_clone
= NULL
;
686 if (clone_of
->clones
)
687 clone_of
->clones
->prev_sibling_clone
= next_inline_clone
;
688 next_inline_clone
->next_sibling_clone
= clone_of
->clones
;
689 clone_of
->clones
= next_inline_clone
;
692 /* Merge the clone list. */
695 if (!next_inline_clone
->clones
)
696 next_inline_clone
->clones
= new_clones
;
699 n
= next_inline_clone
->clones
;
700 while (n
->next_sibling_clone
)
701 n
= n
->next_sibling_clone
;
702 n
->next_sibling_clone
= new_clones
;
703 new_clones
->prev_sibling_clone
= n
;
707 /* Update clone_of pointers. */
711 n
->clone_of
= next_inline_clone
;
712 n
= n
->next_sibling_clone
;
720 /* Like cgraph_set_call_stmt but walk the clone tree and update all
721 clones sharing the same function body.
722 When WHOLE_SPECULATIVE_EDGES is true, all three components of
723 speculative edge gets updated. Otherwise we update only direct
727 cgraph_node::set_call_stmt_including_clones (gimple old_stmt
,
729 bool update_speculative
)
732 cgraph_edge
*edge
= get_edge (old_stmt
);
735 edge
->set_call_stmt (new_stmt
, update_speculative
);
741 cgraph_edge
*edge
= node
->get_edge (old_stmt
);
744 edge
->set_call_stmt (new_stmt
, update_speculative
);
745 /* If UPDATE_SPECULATIVE is false, it means that we are turning
746 speculative call into a real code sequence. Update the
748 if (edge
->speculative
&& !update_speculative
)
750 cgraph_edge
*direct
, *indirect
;
753 gcc_assert (!edge
->indirect_unknown_callee
);
754 edge
->speculative_call_info (direct
, indirect
, ref
);
755 direct
->speculative
= false;
756 indirect
->speculative
= false;
757 ref
->speculative
= false;
762 else if (node
->next_sibling_clone
)
763 node
= node
->next_sibling_clone
;
766 while (node
!= this && !node
->next_sibling_clone
)
767 node
= node
->clone_of
;
769 node
= node
->next_sibling_clone
;
774 /* Like cgraph_create_edge walk the clone tree and update all clones sharing
775 same function body. If clones already have edge for OLD_STMT; only
776 update the edge same way as cgraph_set_call_stmt_including_clones does.
778 TODO: COUNT and LOOP_DEPTH should be properly distributed based on relative
779 frequencies of the clones. */
782 cgraph_node::create_edge_including_clones (cgraph_node
*callee
,
783 gimple old_stmt
, gcall
*stmt
,
786 cgraph_inline_failed_t reason
)
791 if (!get_edge (stmt
))
793 edge
= create_edge (callee
, stmt
, count
, freq
);
794 edge
->inline_failed
= reason
;
801 cgraph_edge
*edge
= node
->get_edge (old_stmt
);
803 /* It is possible that clones already contain the edge while
804 master didn't. Either we promoted indirect call into direct
805 call in the clone or we are processing clones of unreachable
806 master where edges has been removed. */
808 edge
->set_call_stmt (stmt
);
809 else if (! node
->get_edge (stmt
))
811 edge
= node
->create_edge (callee
, stmt
, count
, freq
);
812 edge
->inline_failed
= reason
;
817 else if (node
->next_sibling_clone
)
818 node
= node
->next_sibling_clone
;
821 while (node
!= this && !node
->next_sibling_clone
)
822 node
= node
->clone_of
;
824 node
= node
->next_sibling_clone
;
829 /* Remove the node from cgraph and all inline clones inlined into it.
830 Skip however removal of FORBIDDEN_NODE and return true if it needs to be
831 removed. This allows to call the function from outer loop walking clone
835 cgraph_node::remove_symbol_and_inline_clones (cgraph_node
*forbidden_node
)
837 cgraph_edge
*e
, *next
;
840 if (this == forbidden_node
)
845 for (e
= callees
; e
; e
= next
)
847 next
= e
->next_callee
;
848 if (!e
->inline_failed
)
849 found
|= e
->callee
->remove_symbol_and_inline_clones (forbidden_node
);
855 /* The edges representing the callers of the NEW_VERSION node were
856 fixed by cgraph_function_versioning (), now the call_expr in their
857 respective tree code should be updated to call the NEW_VERSION. */
860 update_call_expr (cgraph_node
*new_version
)
864 gcc_assert (new_version
);
866 /* Update the call expr on the edges to call the new version. */
867 for (e
= new_version
->callers
; e
; e
= e
->next_caller
)
869 function
*inner_function
= DECL_STRUCT_FUNCTION (e
->caller
->decl
);
870 gimple_call_set_fndecl (e
->call_stmt
, new_version
->decl
);
871 maybe_clean_eh_stmt_fn (inner_function
, e
->call_stmt
);
876 /* Create a new cgraph node which is the new version of
877 callgraph node. REDIRECT_CALLERS holds the callers
878 edges which should be redirected to point to
879 NEW_VERSION. ALL the callees edges of the node
880 are cloned to the new version node. Return the new
883 If non-NULL BLOCK_TO_COPY determine what basic blocks
884 was copied to prevent duplications of calls that are dead
888 cgraph_node::create_version_clone (tree new_decl
,
889 vec
<cgraph_edge
*> redirect_callers
,
892 cgraph_node
*new_version
;
896 new_version
= cgraph_node::create (new_decl
);
898 new_version
->analyzed
= analyzed
;
899 new_version
->definition
= definition
;
900 new_version
->local
= local
;
901 new_version
->externally_visible
= false;
902 new_version
->no_reorder
= no_reorder
;
903 new_version
->local
.local
= new_version
->definition
;
904 new_version
->global
= global
;
905 new_version
->rtl
= rtl
;
906 new_version
->count
= count
;
908 for (e
= callees
; e
; e
=e
->next_callee
)
910 || bitmap_bit_p (bbs_to_copy
, gimple_bb (e
->call_stmt
)->index
))
911 e
->clone (new_version
, e
->call_stmt
,
912 e
->lto_stmt_uid
, REG_BR_PROB_BASE
,
915 for (e
= indirect_calls
; e
; e
=e
->next_callee
)
917 || bitmap_bit_p (bbs_to_copy
, gimple_bb (e
->call_stmt
)->index
))
918 e
->clone (new_version
, e
->call_stmt
,
919 e
->lto_stmt_uid
, REG_BR_PROB_BASE
,
922 FOR_EACH_VEC_ELT (redirect_callers
, i
, e
)
924 /* Redirect calls to the old version node to point to its new
926 e
->redirect_callee (new_version
);
929 symtab
->call_cgraph_duplication_hooks (this, new_version
);
934 /* Perform function versioning.
935 Function versioning includes copying of the tree and
936 a callgraph update (creating a new cgraph node and updating
937 its callees and callers).
939 REDIRECT_CALLERS varray includes the edges to be redirected
942 TREE_MAP is a mapping of tree nodes we want to replace with
943 new ones (according to results of prior analysis).
945 If non-NULL ARGS_TO_SKIP determine function parameters to remove
947 If SKIP_RETURN is true, the new version will return void.
948 If non-NULL BLOCK_TO_COPY determine what basic blocks to copy.
949 If non_NULL NEW_ENTRY determine new entry BB of the clone.
951 Return the new version's cgraph node. */
954 cgraph_node::create_version_clone_with_body
955 (vec
<cgraph_edge
*> redirect_callers
,
956 vec
<ipa_replace_map
*, va_gc
> *tree_map
, bitmap args_to_skip
,
957 bool skip_return
, bitmap bbs_to_copy
, basic_block new_entry_block
,
958 const char *clone_name
)
960 tree old_decl
= decl
;
961 cgraph_node
*new_version_node
= NULL
;
964 if (!tree_versionable_function_p (old_decl
))
967 gcc_assert (local
.can_change_signature
|| !args_to_skip
);
969 /* Make a new FUNCTION_DECL tree node for the new version. */
970 if (!args_to_skip
&& !skip_return
)
971 new_decl
= copy_node (old_decl
);
974 = build_function_decl_skip_args (old_decl
, args_to_skip
, skip_return
);
976 /* Generate a new name for the new version. */
977 DECL_NAME (new_decl
) = clone_function_name (old_decl
, clone_name
);
978 SET_DECL_ASSEMBLER_NAME (new_decl
, DECL_NAME (new_decl
));
979 SET_DECL_RTL (new_decl
, NULL
);
981 /* When the old decl was a con-/destructor make sure the clone isn't. */
982 DECL_STATIC_CONSTRUCTOR (new_decl
) = 0;
983 DECL_STATIC_DESTRUCTOR (new_decl
) = 0;
985 /* Create the new version's call-graph node.
986 and update the edges of the new node. */
987 new_version_node
= create_version_clone (new_decl
, redirect_callers
,
990 if (ipa_transforms_to_apply
.exists ())
991 new_version_node
->ipa_transforms_to_apply
992 = ipa_transforms_to_apply
.copy ();
993 /* Copy the OLD_VERSION_NODE function tree to the new version. */
994 tree_function_versioning (old_decl
, new_decl
, tree_map
, false, args_to_skip
,
995 skip_return
, bbs_to_copy
, new_entry_block
);
997 /* Update the new version's properties.
998 Make The new version visible only within this translation unit. Make sure
999 that is not weak also.
1000 ??? We cannot use COMDAT linkage because there is no
1001 ABI support for this. */
1002 new_version_node
->make_decl_local ();
1003 DECL_VIRTUAL_P (new_version_node
->decl
) = 0;
1004 new_version_node
->externally_visible
= 0;
1005 new_version_node
->local
.local
= 1;
1006 new_version_node
->lowered
= true;
1007 if (!implicit_section
)
1008 new_version_node
->set_section (get_section ());
1009 /* Clones of global symbols or symbols with unique names are unique. */
1010 if ((TREE_PUBLIC (old_decl
)
1011 && !DECL_EXTERNAL (old_decl
)
1012 && !DECL_WEAK (old_decl
)
1013 && !DECL_COMDAT (old_decl
))
1015 new_version_node
->unique_name
= true;
1017 /* Update the call_expr on the edges to call the new version node. */
1018 update_call_expr (new_version_node
);
1020 symtab
->call_cgraph_insertion_hooks (this);
1021 return new_version_node
;
1024 /* Given virtual clone, turn it into actual clone. */
1027 cgraph_materialize_clone (cgraph_node
*node
)
1029 bitmap_obstack_initialize (NULL
);
1030 node
->former_clone_of
= node
->clone_of
->decl
;
1031 if (node
->clone_of
->former_clone_of
)
1032 node
->former_clone_of
= node
->clone_of
->former_clone_of
;
1033 /* Copy the OLD_VERSION_NODE function tree to the new version. */
1034 tree_function_versioning (node
->clone_of
->decl
, node
->decl
,
1035 node
->clone
.tree_map
, true,
1036 node
->clone
.args_to_skip
, false,
1038 if (symtab
->dump_file
)
1040 dump_function_to_file (node
->clone_of
->decl
, symtab
->dump_file
,
1042 dump_function_to_file (node
->decl
, symtab
->dump_file
, dump_flags
);
1045 /* Function is no longer clone. */
1046 if (node
->next_sibling_clone
)
1047 node
->next_sibling_clone
->prev_sibling_clone
= node
->prev_sibling_clone
;
1048 if (node
->prev_sibling_clone
)
1049 node
->prev_sibling_clone
->next_sibling_clone
= node
->next_sibling_clone
;
1051 node
->clone_of
->clones
= node
->next_sibling_clone
;
1052 node
->next_sibling_clone
= NULL
;
1053 node
->prev_sibling_clone
= NULL
;
1054 if (!node
->clone_of
->analyzed
&& !node
->clone_of
->clones
)
1056 node
->clone_of
->release_body ();
1057 node
->clone_of
->remove_callees ();
1058 node
->clone_of
->remove_all_references ();
1060 node
->clone_of
= NULL
;
1061 bitmap_obstack_release (NULL
);
1064 /* Once all functions from compilation unit are in memory, produce all clones
1065 and update all calls. We might also do this on demand if we don't want to
1066 bring all functions to memory prior compilation, but current WHOPR
1067 implementation does that and it is a bit easier to keep everything right in
1071 symbol_table::materialize_all_clones (void)
1074 bool stabilized
= false;
1077 if (symtab
->dump_file
)
1078 fprintf (symtab
->dump_file
, "Materializing clones\n");
1079 #ifdef ENABLE_CHECKING
1080 cgraph_node::verify_cgraph_nodes ();
1083 /* We can also do topological order, but number of iterations should be
1084 bounded by number of IPA passes since single IPA pass is probably not
1085 going to create clones of clones it created itself. */
1089 FOR_EACH_FUNCTION (node
)
1091 if (node
->clone_of
&& node
->decl
!= node
->clone_of
->decl
1092 && !gimple_has_body_p (node
->decl
))
1094 if (!node
->clone_of
->clone_of
)
1095 node
->clone_of
->get_untransformed_body ();
1096 if (gimple_has_body_p (node
->clone_of
->decl
))
1098 if (symtab
->dump_file
)
1100 fprintf (symtab
->dump_file
, "cloning %s to %s\n",
1101 xstrdup_for_dump (node
->clone_of
->name ()),
1102 xstrdup_for_dump (node
->name ()));
1103 if (node
->clone
.tree_map
)
1106 fprintf (symtab
->dump_file
, " replace map: ");
1108 i
< vec_safe_length (node
->clone
.tree_map
);
1111 ipa_replace_map
*replace_info
;
1112 replace_info
= (*node
->clone
.tree_map
)[i
];
1113 print_generic_expr (symtab
->dump_file
, replace_info
->old_tree
, 0);
1114 fprintf (symtab
->dump_file
, " -> ");
1115 print_generic_expr (symtab
->dump_file
, replace_info
->new_tree
, 0);
1116 fprintf (symtab
->dump_file
, "%s%s;",
1117 replace_info
->replace_p
? "(replace)":"",
1118 replace_info
->ref_p
? "(ref)":"");
1120 fprintf (symtab
->dump_file
, "\n");
1122 if (node
->clone
.args_to_skip
)
1124 fprintf (symtab
->dump_file
, " args_to_skip: ");
1125 dump_bitmap (symtab
->dump_file
,
1126 node
->clone
.args_to_skip
);
1128 if (node
->clone
.args_to_skip
)
1130 fprintf (symtab
->dump_file
, " combined_args_to_skip:");
1131 dump_bitmap (symtab
->dump_file
, node
->clone
.combined_args_to_skip
);
1134 cgraph_materialize_clone (node
);
1140 FOR_EACH_FUNCTION (node
)
1141 if (!node
->analyzed
&& node
->callees
)
1143 node
->remove_callees ();
1144 node
->remove_all_references ();
1147 node
->clear_stmts_in_references ();
1148 if (symtab
->dump_file
)
1149 fprintf (symtab
->dump_file
, "Materialization Call site updates done.\n");
1150 #ifdef ENABLE_CHECKING
1151 cgraph_node::verify_cgraph_nodes ();
1153 symtab
->remove_unreachable_nodes (symtab
->dump_file
);
1156 #include "gt-cgraphclones.h"