1 /* Utilities for ipa analysis.
2 Copyright (C) 2005-2014 Free Software Foundation, Inc.
3 Contributed by Kenneth Zadeck <zadeck@naturalbridge.com>
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/>. */
23 #include "coretypes.h"
26 #include "basic-block.h"
27 #include "tree-ssa-alias.h"
28 #include "internal-fn.h"
29 #include "gimple-expr.h"
32 #include "tree-inline.h"
34 #include "langhooks.h"
35 #include "splay-tree.h"
36 #include "ipa-utils.h"
37 #include "ipa-reference.h"
39 #include "diagnostic.h"
40 #include "langhooks.h"
41 #include "lto-streamer.h"
42 #include "ipa-inline.h"
44 /* Debugging function for postorder and inorder code. NOTE is a string
45 that is printed before the nodes are printed. ORDER is an array of
46 cgraph_nodes that has COUNT useful nodes in it. */
49 ipa_print_order (FILE* out
,
51 struct cgraph_node
** order
,
55 fprintf (out
, "\n\n ordered call graph: %s\n", note
);
57 for (i
= count
- 1; i
>= 0; i
--)
65 struct cgraph_node
**stack
;
67 struct cgraph_node
**result
;
69 splay_tree nodes_marked_new
;
71 bool allow_overwritable
;
75 /* This is an implementation of Tarjan's strongly connected region
76 finder as reprinted in Aho Hopcraft and Ullman's The Design and
77 Analysis of Computer Programs (1975) pages 192-193. This version
78 has been customized for cgraph_nodes. The env parameter is because
79 it is recursive and there are no nested functions here. This
80 function should only be called from itself or
81 ipa_reduced_postorder. ENV is a stack env and would be
82 unnecessary if C had nested functions. V is the node to start
86 searchc (struct searchc_env
* env
, struct cgraph_node
*v
,
87 bool (*ignore_edge
) (struct cgraph_edge
*))
89 struct cgraph_edge
*edge
;
90 struct ipa_dfs_info
*v_info
= (struct ipa_dfs_info
*) v
->aux
;
92 /* mark node as old */
93 v_info
->new_node
= false;
94 splay_tree_remove (env
->nodes_marked_new
, v
->uid
);
96 v_info
->dfn_number
= env
->count
;
97 v_info
->low_link
= env
->count
;
99 env
->stack
[(env
->stack_size
)++] = v
;
100 v_info
->on_stack
= true;
102 for (edge
= v
->callees
; edge
; edge
= edge
->next_callee
)
104 struct ipa_dfs_info
* w_info
;
105 enum availability avail
;
106 struct cgraph_node
*w
= edge
->callee
->ultimate_alias_target (&avail
);
108 if (!w
|| (ignore_edge
&& ignore_edge (edge
)))
112 && (avail
> AVAIL_INTERPOSABLE
113 || (env
->allow_overwritable
&& avail
== AVAIL_INTERPOSABLE
)))
115 w_info
= (struct ipa_dfs_info
*) w
->aux
;
116 if (w_info
->new_node
)
118 searchc (env
, w
, ignore_edge
);
120 (v_info
->low_link
< w_info
->low_link
) ?
121 v_info
->low_link
: w_info
->low_link
;
124 if ((w_info
->dfn_number
< v_info
->dfn_number
)
125 && (w_info
->on_stack
))
127 (w_info
->dfn_number
< v_info
->low_link
) ?
128 w_info
->dfn_number
: v_info
->low_link
;
133 if (v_info
->low_link
== v_info
->dfn_number
)
135 struct cgraph_node
*last
= NULL
;
136 struct cgraph_node
*x
;
137 struct ipa_dfs_info
*x_info
;
139 x
= env
->stack
[--(env
->stack_size
)];
140 x_info
= (struct ipa_dfs_info
*) x
->aux
;
141 x_info
->on_stack
= false;
142 x_info
->scc_no
= v_info
->dfn_number
;
146 x_info
->next_cycle
= last
;
150 env
->result
[env
->order_pos
++] = x
;
154 env
->result
[env
->order_pos
++] = v
;
158 /* Topsort the call graph by caller relation. Put the result in ORDER.
160 The REDUCE flag is true if you want the cycles reduced to single nodes.
161 You can use ipa_get_nodes_in_cycle to obtain a vector containing all real
162 call graph nodes in a reduced node.
164 Set ALLOW_OVERWRITABLE if nodes with such availability should be included.
165 IGNORE_EDGE, if non-NULL is a hook that may make some edges insignificant
166 for the topological sort. */
169 ipa_reduced_postorder (struct cgraph_node
**order
,
170 bool reduce
, bool allow_overwritable
,
171 bool (*ignore_edge
) (struct cgraph_edge
*))
173 struct cgraph_node
*node
;
174 struct searchc_env env
;
175 splay_tree_node result
;
176 env
.stack
= XCNEWVEC (struct cgraph_node
*, cgraph_n_nodes
);
180 env
.nodes_marked_new
= splay_tree_new (splay_tree_compare_ints
, 0, 0);
183 env
.allow_overwritable
= allow_overwritable
;
185 FOR_EACH_DEFINED_FUNCTION (node
)
187 enum availability avail
= node
->get_availability ();
189 if (avail
> AVAIL_INTERPOSABLE
190 || (allow_overwritable
191 && (avail
== AVAIL_INTERPOSABLE
)))
193 /* Reuse the info if it is already there. */
194 struct ipa_dfs_info
*info
= (struct ipa_dfs_info
*) node
->aux
;
196 info
= XCNEW (struct ipa_dfs_info
);
197 info
->new_node
= true;
198 info
->on_stack
= false;
199 info
->next_cycle
= NULL
;
202 splay_tree_insert (env
.nodes_marked_new
,
203 (splay_tree_key
)node
->uid
,
204 (splay_tree_value
)node
);
209 result
= splay_tree_min (env
.nodes_marked_new
);
212 node
= (struct cgraph_node
*)result
->value
;
213 searchc (&env
, node
, ignore_edge
);
214 result
= splay_tree_min (env
.nodes_marked_new
);
216 splay_tree_delete (env
.nodes_marked_new
);
219 return env
.order_pos
;
222 /* Deallocate all ipa_dfs_info structures pointed to by the aux pointer of call
226 ipa_free_postorder_info (void)
228 struct cgraph_node
*node
;
229 FOR_EACH_DEFINED_FUNCTION (node
)
231 /* Get rid of the aux information. */
240 /* Get the set of nodes for the cycle in the reduced call graph starting
244 ipa_get_nodes_in_cycle (struct cgraph_node
*node
)
246 vec
<cgraph_node
*> v
= vNULL
;
247 struct ipa_dfs_info
*node_dfs_info
;
251 node_dfs_info
= (struct ipa_dfs_info
*) node
->aux
;
252 node
= node_dfs_info
->next_cycle
;
257 /* Return true iff the CS is an edge within a strongly connected component as
258 computed by ipa_reduced_postorder. */
261 ipa_edge_within_scc (struct cgraph_edge
*cs
)
263 struct ipa_dfs_info
*caller_dfs
= (struct ipa_dfs_info
*) cs
->caller
->aux
;
264 struct ipa_dfs_info
*callee_dfs
;
265 struct cgraph_node
*callee
= cs
->callee
->function_symbol ();
267 callee_dfs
= (struct ipa_dfs_info
*) callee
->aux
;
270 && caller_dfs
->scc_no
== callee_dfs
->scc_no
);
273 struct postorder_stack
275 struct cgraph_node
*node
;
276 struct cgraph_edge
*edge
;
280 /* Fill array order with all nodes with output flag set in the reverse
281 topological order. Return the number of elements in the array.
282 FIXME: While walking, consider aliases, too. */
285 ipa_reverse_postorder (struct cgraph_node
**order
)
287 struct cgraph_node
*node
, *node2
;
290 struct cgraph_edge
*edge
;
292 struct ipa_ref
*ref
= NULL
;
294 struct postorder_stack
*stack
=
295 XCNEWVEC (struct postorder_stack
, cgraph_n_nodes
);
297 /* We have to deal with cycles nicely, so use a depth first traversal
298 output algorithm. Ignore the fact that some functions won't need
299 to be output and put them into order as well, so we get dependencies
300 right through inline functions. */
301 FOR_EACH_FUNCTION (node
)
303 for (pass
= 0; pass
< 2; pass
++)
304 FOR_EACH_FUNCTION (node
)
307 || (!node
->address_taken
308 && !node
->global
.inlined_to
309 && !node
->alias
&& !node
->thunk
.thunk_p
310 && !node
->only_called_directly_p ())))
313 stack
[stack_size
].node
= node
;
314 stack
[stack_size
].edge
= node
->callers
;
315 stack
[stack_size
].ref
= 0;
316 node
->aux
= (void *)(size_t)1;
317 while (stack_size
>= 0)
322 while (stack
[stack_size
].edge
&& !node2
)
324 edge
= stack
[stack_size
].edge
;
325 node2
= edge
->caller
;
326 stack
[stack_size
].edge
= edge
->next_caller
;
327 /* Break possible cycles involving always-inline
328 functions by ignoring edges from always-inline
329 functions to non-always-inline functions. */
330 if (DECL_DISREGARD_INLINE_LIMITS (edge
->caller
->decl
)
331 && !DECL_DISREGARD_INLINE_LIMITS
332 (edge
->callee
->function_symbol ()->decl
))
335 for (; stack
[stack_size
].node
->iterate_referring (
336 stack
[stack_size
].ref
,
338 stack
[stack_size
].ref
++)
340 if (ref
->use
== IPA_REF_ALIAS
)
341 node2
= dyn_cast
<cgraph_node
*> (ref
->referring
);
347 stack
[++stack_size
].node
= node2
;
348 stack
[stack_size
].edge
= node2
->callers
;
349 stack
[stack_size
].ref
= 0;
350 node2
->aux
= (void *)(size_t)1;
353 order
[order_pos
++] = stack
[stack_size
--].node
;
357 FOR_EACH_FUNCTION (node
)
364 /* Given a memory reference T, will return the variable at the bottom
365 of the access. Unlike get_base_address, this will recurse through
369 get_base_var (tree t
)
371 while (!SSA_VAR_P (t
)
372 && (!CONSTANT_CLASS_P (t
))
373 && TREE_CODE (t
) != LABEL_DECL
374 && TREE_CODE (t
) != FUNCTION_DECL
375 && TREE_CODE (t
) != CONST_DECL
376 && TREE_CODE (t
) != CONSTRUCTOR
)
378 t
= TREE_OPERAND (t
, 0);
384 /* Create a new cgraph node set. */
387 cgraph_node_set_new (void)
389 cgraph_node_set new_node_set
;
391 new_node_set
= XCNEW (struct cgraph_node_set_def
);
392 new_node_set
->map
= pointer_map_create ();
393 new_node_set
->nodes
.create (0);
398 /* Add cgraph_node NODE to cgraph_node_set SET. */
401 cgraph_node_set_add (cgraph_node_set set
, struct cgraph_node
*node
)
405 slot
= pointer_map_insert (set
->map
, node
);
409 int index
= (size_t) *slot
- 1;
410 gcc_checking_assert ((set
->nodes
[index
]
415 *slot
= (void *)(size_t) (set
->nodes
.length () + 1);
417 /* Insert into node vector. */
418 set
->nodes
.safe_push (node
);
422 /* Remove cgraph_node NODE from cgraph_node_set SET. */
425 cgraph_node_set_remove (cgraph_node_set set
, struct cgraph_node
*node
)
427 void **slot
, **last_slot
;
429 struct cgraph_node
*last_node
;
431 slot
= pointer_map_contains (set
->map
, node
);
432 if (slot
== NULL
|| !*slot
)
435 index
= (size_t) *slot
- 1;
436 gcc_checking_assert (set
->nodes
[index
]
439 /* Remove from vector. We do this by swapping node with the last element
441 last_node
= set
->nodes
.pop ();
442 if (last_node
!= node
)
444 last_slot
= pointer_map_contains (set
->map
, last_node
);
445 gcc_checking_assert (last_slot
&& *last_slot
);
446 *last_slot
= (void *)(size_t) (index
+ 1);
448 /* Move the last element to the original spot of NODE. */
449 set
->nodes
[index
] = last_node
;
452 /* Remove element from hash table. */
457 /* Find NODE in SET and return an iterator to it if found. A null iterator
458 is returned if NODE is not in SET. */
460 cgraph_node_set_iterator
461 cgraph_node_set_find (cgraph_node_set set
, struct cgraph_node
*node
)
464 cgraph_node_set_iterator csi
;
466 slot
= pointer_map_contains (set
->map
, node
);
467 if (slot
== NULL
|| !*slot
)
468 csi
.index
= (unsigned) ~0;
470 csi
.index
= (size_t)*slot
- 1;
477 /* Dump content of SET to file F. */
480 dump_cgraph_node_set (FILE *f
, cgraph_node_set set
)
482 cgraph_node_set_iterator iter
;
484 for (iter
= csi_start (set
); !csi_end_p (iter
); csi_next (&iter
))
486 struct cgraph_node
*node
= csi_node (iter
);
487 fprintf (f
, " %s/%i", node
->name (), node
->order
);
493 /* Dump content of SET to stderr. */
496 debug_cgraph_node_set (cgraph_node_set set
)
498 dump_cgraph_node_set (stderr
, set
);
502 /* Free varpool node set. */
505 free_cgraph_node_set (cgraph_node_set set
)
507 set
->nodes
.release ();
508 pointer_map_destroy (set
->map
);
513 /* Create a new varpool node set. */
516 varpool_node_set_new (void)
518 varpool_node_set new_node_set
;
520 new_node_set
= XCNEW (struct varpool_node_set_def
);
521 new_node_set
->map
= pointer_map_create ();
522 new_node_set
->nodes
.create (0);
527 /* Add varpool_node NODE to varpool_node_set SET. */
530 varpool_node_set_add (varpool_node_set set
, varpool_node
*node
)
534 slot
= pointer_map_insert (set
->map
, node
);
538 int index
= (size_t) *slot
- 1;
539 gcc_checking_assert ((set
->nodes
[index
]
544 *slot
= (void *)(size_t) (set
->nodes
.length () + 1);
546 /* Insert into node vector. */
547 set
->nodes
.safe_push (node
);
551 /* Remove varpool_node NODE from varpool_node_set SET. */
554 varpool_node_set_remove (varpool_node_set set
, varpool_node
*node
)
556 void **slot
, **last_slot
;
558 varpool_node
*last_node
;
560 slot
= pointer_map_contains (set
->map
, node
);
561 if (slot
== NULL
|| !*slot
)
564 index
= (size_t) *slot
- 1;
565 gcc_checking_assert (set
->nodes
[index
]
568 /* Remove from vector. We do this by swapping node with the last element
570 last_node
= set
->nodes
.pop ();
571 if (last_node
!= node
)
573 last_slot
= pointer_map_contains (set
->map
, last_node
);
574 gcc_checking_assert (last_slot
&& *last_slot
);
575 *last_slot
= (void *)(size_t) (index
+ 1);
577 /* Move the last element to the original spot of NODE. */
578 set
->nodes
[index
] = last_node
;
581 /* Remove element from hash table. */
586 /* Find NODE in SET and return an iterator to it if found. A null iterator
587 is returned if NODE is not in SET. */
589 varpool_node_set_iterator
590 varpool_node_set_find (varpool_node_set set
, varpool_node
*node
)
593 varpool_node_set_iterator vsi
;
595 slot
= pointer_map_contains (set
->map
, node
);
596 if (slot
== NULL
|| !*slot
)
597 vsi
.index
= (unsigned) ~0;
599 vsi
.index
= (size_t)*slot
- 1;
606 /* Dump content of SET to file F. */
609 dump_varpool_node_set (FILE *f
, varpool_node_set set
)
611 varpool_node_set_iterator iter
;
613 for (iter
= vsi_start (set
); !vsi_end_p (iter
); vsi_next (&iter
))
615 varpool_node
*node
= vsi_node (iter
);
616 fprintf (f
, " %s", node
->name ());
622 /* Free varpool node set. */
625 free_varpool_node_set (varpool_node_set set
)
627 set
->nodes
.release ();
628 pointer_map_destroy (set
->map
);
633 /* Dump content of SET to stderr. */
636 debug_varpool_node_set (varpool_node_set set
)
638 dump_varpool_node_set (stderr
, set
);
642 /* SRC and DST are going to be merged. Take SRC's profile and merge it into
643 DST so it is not going to be lost. Destroy SRC's body on the way. */
646 ipa_merge_profiles (struct cgraph_node
*dst
,
647 struct cgraph_node
*src
)
649 tree oldsrcdecl
= src
->decl
;
650 struct function
*srccfun
, *dstcfun
;
656 if (src
->frequency
< dst
->frequency
)
657 src
->frequency
= dst
->frequency
;
659 /* Time profiles are merged. */
660 if (dst
->tp_first_run
> src
->tp_first_run
&& src
->tp_first_run
)
661 dst
->tp_first_run
= src
->tp_first_run
;
665 if (!dst
->profile_id
)
666 dst
->profile_id
= src
->profile_id
;
668 gcc_assert (src
->profile_id
== dst
->profile_id
);
673 if (cgraph_dump_file
)
675 fprintf (cgraph_dump_file
, "Merging profiles of %s/%i to %s/%i\n",
676 xstrdup (src
->name ()), src
->order
,
677 xstrdup (dst
->name ()), dst
->order
);
679 dst
->count
+= src
->count
;
681 /* This is ugly. We need to get both function bodies into memory.
682 If declaration is merged, we need to duplicate it to be able
683 to load body that is being replaced. This makes symbol table
684 temporarily inconsistent. */
685 if (src
->decl
== dst
->decl
)
688 struct lto_in_decl_state temp
;
689 struct lto_in_decl_state
*state
;
691 /* We are going to move the decl, we want to remove its file decl data.
692 and link these with the new decl. */
693 temp
.fn_decl
= src
->decl
;
694 slot
= htab_find_slot (src
->lto_file_data
->function_decl_states
,
696 state
= (lto_in_decl_state
*)*slot
;
697 htab_clear_slot (src
->lto_file_data
->function_decl_states
, slot
);
700 /* Duplicate the decl and be sure it does not link into body of DST. */
701 src
->decl
= copy_node (src
->decl
);
702 DECL_STRUCT_FUNCTION (src
->decl
) = NULL
;
703 DECL_ARGUMENTS (src
->decl
) = NULL
;
704 DECL_INITIAL (src
->decl
) = NULL
;
705 DECL_RESULT (src
->decl
) = NULL
;
707 /* Associate the decl state with new declaration, so LTO streamer
709 state
->fn_decl
= src
->decl
;
710 slot
= htab_find_slot (src
->lto_file_data
->function_decl_states
,
717 srccfun
= DECL_STRUCT_FUNCTION (src
->decl
);
718 dstcfun
= DECL_STRUCT_FUNCTION (dst
->decl
);
719 if (n_basic_blocks_for_fn (srccfun
)
720 != n_basic_blocks_for_fn (dstcfun
))
722 if (cgraph_dump_file
)
723 fprintf (cgraph_dump_file
,
724 "Giving up; number of basic block mismatch.\n");
727 else if (last_basic_block_for_fn (srccfun
)
728 != last_basic_block_for_fn (dstcfun
))
730 if (cgraph_dump_file
)
731 fprintf (cgraph_dump_file
,
732 "Giving up; last block mismatch.\n");
737 basic_block srcbb
, dstbb
;
739 FOR_ALL_BB_FN (srcbb
, srccfun
)
743 dstbb
= BASIC_BLOCK_FOR_FN (dstcfun
, srcbb
->index
);
746 if (cgraph_dump_file
)
747 fprintf (cgraph_dump_file
,
748 "No matching block for bb %i.\n",
753 if (EDGE_COUNT (srcbb
->succs
) != EDGE_COUNT (dstbb
->succs
))
755 if (cgraph_dump_file
)
756 fprintf (cgraph_dump_file
,
757 "Edge count mistmatch for bb %i.\n",
762 for (i
= 0; i
< EDGE_COUNT (srcbb
->succs
); i
++)
764 edge srce
= EDGE_SUCC (srcbb
, i
);
765 edge dste
= EDGE_SUCC (dstbb
, i
);
766 if (srce
->dest
->index
!= dste
->dest
->index
)
768 if (cgraph_dump_file
)
769 fprintf (cgraph_dump_file
,
770 "Succ edge mistmatch for bb %i.\n",
780 struct cgraph_edge
*e
;
781 basic_block srcbb
, dstbb
;
783 /* TODO: merge also statement histograms. */
784 FOR_ALL_BB_FN (srcbb
, srccfun
)
788 dstbb
= BASIC_BLOCK_FOR_FN (dstcfun
, srcbb
->index
);
789 dstbb
->count
+= srcbb
->count
;
790 for (i
= 0; i
< EDGE_COUNT (srcbb
->succs
); i
++)
792 edge srce
= EDGE_SUCC (srcbb
, i
);
793 edge dste
= EDGE_SUCC (dstbb
, i
);
794 dste
->count
+= srce
->count
;
799 compute_function_frequency ();
801 for (e
= dst
->callees
; e
; e
= e
->next_callee
)
803 gcc_assert (!e
->speculative
);
804 e
->count
= gimple_bb (e
->call_stmt
)->count
;
805 e
->frequency
= compute_call_stmt_bb_frequency
807 gimple_bb (e
->call_stmt
));
809 for (e
= dst
->indirect_calls
; e
; e
= e
->next_callee
)
811 gcc_assert (!e
->speculative
);
812 e
->count
= gimple_bb (e
->call_stmt
)->count
;
813 e
->frequency
= compute_call_stmt_bb_frequency
815 gimple_bb (e
->call_stmt
));
817 src
->release_body ();
818 inline_update_overall_summary (dst
);
820 /* TODO: if there is no match, we can scale up. */
821 src
->decl
= oldsrcdecl
;
824 /* Return true if call to DEST is known to be self-recusive call withing FUNC. */
827 recursive_call_p (tree func
, tree dest
)
829 struct cgraph_node
*dest_node
= cgraph_node::get_create (dest
);
830 struct cgraph_node
*cnode
= cgraph_node::get_create (func
);
832 return dest_node
->semantically_equivalent_p (cnode
);