2007-09-24 Pranav Bhandarkar <pranav.bhandarkar@celunite.com>
[official-gcc.git] / gcc / tree-inline.c
blob9c459511b712eab5c892fa0fdfa8f15df0bae2b4
1 /* Tree inlining.
2 Copyright 2001, 2002, 2003, 2004, 2005, 2006, 2007
3 Free Software Foundation, Inc.
4 Contributed by Alexandre Oliva <aoliva@redhat.com>
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3, or (at your option)
11 any later version.
13 GCC is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
22 #include "config.h"
23 #include "system.h"
24 #include "coretypes.h"
25 #include "tm.h"
26 #include "toplev.h"
27 #include "tree.h"
28 #include "tree-inline.h"
29 #include "rtl.h"
30 #include "expr.h"
31 #include "flags.h"
32 #include "params.h"
33 #include "input.h"
34 #include "insn-config.h"
35 #include "varray.h"
36 #include "hashtab.h"
37 #include "langhooks.h"
38 #include "basic-block.h"
39 #include "tree-iterator.h"
40 #include "cgraph.h"
41 #include "intl.h"
42 #include "tree-mudflap.h"
43 #include "tree-flow.h"
44 #include "function.h"
45 #include "ggc.h"
46 #include "tree-flow.h"
47 #include "diagnostic.h"
48 #include "except.h"
49 #include "debug.h"
50 #include "pointer-set.h"
51 #include "ipa-prop.h"
52 #include "value-prof.h"
53 #include "tree-pass.h"
54 #include "target.h"
55 #include "integrate.h"
57 /* I'm not real happy about this, but we need to handle gimple and
58 non-gimple trees. */
59 #include "tree-gimple.h"
61 /* Inlining, Cloning, Versioning, Parallelization
63 Inlining: a function body is duplicated, but the PARM_DECLs are
64 remapped into VAR_DECLs, and non-void RETURN_EXPRs become
65 GIMPLE_MODIFY_STMTs that store to a dedicated returned-value variable.
66 The duplicated eh_region info of the copy will later be appended
67 to the info for the caller; the eh_region info in copied throwing
68 statements and RESX_EXPRs is adjusted accordingly.
70 Cloning: (only in C++) We have one body for a con/de/structor, and
71 multiple function decls, each with a unique parameter list.
72 Duplicate the body, using the given splay tree; some parameters
73 will become constants (like 0 or 1).
75 Versioning: a function body is duplicated and the result is a new
76 function rather than into blocks of an existing function as with
77 inlining. Some parameters will become constants.
79 Parallelization: a region of a function is duplicated resulting in
80 a new function. Variables may be replaced with complex expressions
81 to enable shared variable semantics.
83 All of these will simultaneously lookup any callgraph edges. If
84 we're going to inline the duplicated function body, and the given
85 function has some cloned callgraph nodes (one for each place this
86 function will be inlined) those callgraph edges will be duplicated.
87 If we're cloning the body, those callgraph edges will be
88 updated to point into the new body. (Note that the original
89 callgraph node and edge list will not be altered.)
91 See the CALL_EXPR handling case in copy_body_r (). */
93 /* 0 if we should not perform inlining.
94 1 if we should expand functions calls inline at the tree level.
95 2 if we should consider *all* functions to be inline
96 candidates. */
98 int flag_inline_trees = 0;
100 /* To Do:
102 o In order to make inlining-on-trees work, we pessimized
103 function-local static constants. In particular, they are now
104 always output, even when not addressed. Fix this by treating
105 function-local static constants just like global static
106 constants; the back-end already knows not to output them if they
107 are not needed.
109 o Provide heuristics to clamp inlining of recursive template
110 calls? */
113 /* Weights that estimate_num_insns uses for heuristics in inlining. */
115 eni_weights eni_inlining_weights;
117 /* Weights that estimate_num_insns uses to estimate the size of the
118 produced code. */
120 eni_weights eni_size_weights;
122 /* Weights that estimate_num_insns uses to estimate the time necessary
123 to execute the produced code. */
125 eni_weights eni_time_weights;
127 /* Prototypes. */
129 static tree declare_return_variable (copy_body_data *, tree, tree, tree *);
130 static tree copy_generic_body (copy_body_data *);
131 static bool inlinable_function_p (tree);
132 static void remap_block (tree *, copy_body_data *);
133 static tree remap_decls (tree, copy_body_data *);
134 static void copy_bind_expr (tree *, int *, copy_body_data *);
135 static tree mark_local_for_remap_r (tree *, int *, void *);
136 static void unsave_expr_1 (tree);
137 static tree unsave_r (tree *, int *, void *);
138 static void declare_inline_vars (tree, tree);
139 static void remap_save_expr (tree *, void *, int *);
140 static void add_lexical_block (tree current_block, tree new_block);
141 static tree copy_decl_to_var (tree, copy_body_data *);
142 static tree copy_result_decl_to_var (tree, copy_body_data *);
143 static tree copy_decl_no_change (tree, copy_body_data *);
144 static tree copy_decl_maybe_to_var (tree, copy_body_data *);
146 /* Insert a tree->tree mapping for ID. Despite the name suggests
147 that the trees should be variables, it is used for more than that. */
149 void
150 insert_decl_map (copy_body_data *id, tree key, tree value)
152 *pointer_map_insert (id->decl_map, key) = value;
154 /* Always insert an identity map as well. If we see this same new
155 node again, we won't want to duplicate it a second time. */
156 if (key != value)
157 *pointer_map_insert (id->decl_map, value) = value;
160 /* Construct new SSA name for old NAME. ID is the inline context. */
162 static tree
163 remap_ssa_name (tree name, copy_body_data *id)
165 tree new;
166 tree *n;
168 gcc_assert (TREE_CODE (name) == SSA_NAME);
170 n = (tree *) pointer_map_contains (id->decl_map, name);
171 if (n)
172 return *n;
174 /* Do not set DEF_STMT yet as statement is not copied yet. We do that
175 in copy_bb. */
176 new = remap_decl (SSA_NAME_VAR (name), id);
177 /* We might've substituted constant or another SSA_NAME for
178 the variable.
180 Replace the SSA name representing RESULT_DECL by variable during
181 inlining: this saves us from need to introduce PHI node in a case
182 return value is just partly initialized. */
183 if ((TREE_CODE (new) == VAR_DECL || TREE_CODE (new) == PARM_DECL)
184 && (TREE_CODE (SSA_NAME_VAR (name)) != RESULT_DECL
185 || !id->transform_return_to_modify))
187 new = make_ssa_name (new, NULL);
188 insert_decl_map (id, name, new);
189 if (IS_EMPTY_STMT (SSA_NAME_DEF_STMT (name)))
191 SSA_NAME_DEF_STMT (new) = build_empty_stmt ();
192 if (gimple_default_def (id->src_cfun, SSA_NAME_VAR (name)) == name)
193 set_default_def (SSA_NAME_VAR (new), new);
195 SSA_NAME_OCCURS_IN_ABNORMAL_PHI (new)
196 = SSA_NAME_OCCURS_IN_ABNORMAL_PHI (name);
197 TREE_TYPE (new) = TREE_TYPE (SSA_NAME_VAR (new));
199 else
200 insert_decl_map (id, name, new);
201 return new;
204 /* Remap DECL during the copying of the BLOCK tree for the function. */
206 tree
207 remap_decl (tree decl, copy_body_data *id)
209 tree *n;
210 tree fn;
212 /* We only remap local variables in the current function. */
213 fn = id->src_fn;
215 /* See if we have remapped this declaration. */
217 n = (tree *) pointer_map_contains (id->decl_map, decl);
219 /* If we didn't already have an equivalent for this declaration,
220 create one now. */
221 if (!n)
223 /* Make a copy of the variable or label. */
224 tree t = id->copy_decl (decl, id);
226 /* Remember it, so that if we encounter this local entity again
227 we can reuse this copy. Do this early because remap_type may
228 need this decl for TYPE_STUB_DECL. */
229 insert_decl_map (id, decl, t);
231 if (!DECL_P (t))
232 return t;
234 /* Remap types, if necessary. */
235 TREE_TYPE (t) = remap_type (TREE_TYPE (t), id);
236 if (TREE_CODE (t) == TYPE_DECL)
237 DECL_ORIGINAL_TYPE (t) = remap_type (DECL_ORIGINAL_TYPE (t), id);
239 /* Remap sizes as necessary. */
240 walk_tree (&DECL_SIZE (t), copy_body_r, id, NULL);
241 walk_tree (&DECL_SIZE_UNIT (t), copy_body_r, id, NULL);
243 /* If fields, do likewise for offset and qualifier. */
244 if (TREE_CODE (t) == FIELD_DECL)
246 walk_tree (&DECL_FIELD_OFFSET (t), copy_body_r, id, NULL);
247 if (TREE_CODE (DECL_CONTEXT (t)) == QUAL_UNION_TYPE)
248 walk_tree (&DECL_QUALIFIER (t), copy_body_r, id, NULL);
251 if (cfun && gimple_in_ssa_p (cfun)
252 && (TREE_CODE (t) == VAR_DECL
253 || TREE_CODE (t) == RESULT_DECL || TREE_CODE (t) == PARM_DECL))
255 tree def = gimple_default_def (id->src_cfun, decl);
256 get_var_ann (t);
257 if (TREE_CODE (decl) != PARM_DECL && def)
259 tree map = remap_ssa_name (def, id);
260 /* Watch out RESULT_DECLs whose SSA names map directly
261 to them. */
262 if (TREE_CODE (map) == SSA_NAME)
263 set_default_def (t, map);
265 add_referenced_var (t);
267 return t;
270 return unshare_expr (*n);
273 static tree
274 remap_type_1 (tree type, copy_body_data *id)
276 tree *node;
277 tree new, t;
279 if (type == NULL)
280 return type;
282 /* See if we have remapped this type. */
283 node = (tree *) pointer_map_contains (id->decl_map, type);
284 if (node)
285 return *node;
287 /* The type only needs remapping if it's variably modified. */
288 if (! variably_modified_type_p (type, id->src_fn))
290 insert_decl_map (id, type, type);
291 return type;
294 /* We do need a copy. build and register it now. If this is a pointer or
295 reference type, remap the designated type and make a new pointer or
296 reference type. */
297 if (TREE_CODE (type) == POINTER_TYPE)
299 new = build_pointer_type_for_mode (remap_type (TREE_TYPE (type), id),
300 TYPE_MODE (type),
301 TYPE_REF_CAN_ALIAS_ALL (type));
302 insert_decl_map (id, type, new);
303 return new;
305 else if (TREE_CODE (type) == REFERENCE_TYPE)
307 new = build_reference_type_for_mode (remap_type (TREE_TYPE (type), id),
308 TYPE_MODE (type),
309 TYPE_REF_CAN_ALIAS_ALL (type));
310 insert_decl_map (id, type, new);
311 return new;
313 else
314 new = copy_node (type);
316 insert_decl_map (id, type, new);
318 /* This is a new type, not a copy of an old type. Need to reassociate
319 variants. We can handle everything except the main variant lazily. */
320 t = TYPE_MAIN_VARIANT (type);
321 if (type != t)
323 t = remap_type (t, id);
324 TYPE_MAIN_VARIANT (new) = t;
325 TYPE_NEXT_VARIANT (new) = TYPE_NEXT_VARIANT (t);
326 TYPE_NEXT_VARIANT (t) = new;
328 else
330 TYPE_MAIN_VARIANT (new) = new;
331 TYPE_NEXT_VARIANT (new) = NULL;
334 if (TYPE_STUB_DECL (type))
335 TYPE_STUB_DECL (new) = remap_decl (TYPE_STUB_DECL (type), id);
337 /* Lazily create pointer and reference types. */
338 TYPE_POINTER_TO (new) = NULL;
339 TYPE_REFERENCE_TO (new) = NULL;
341 switch (TREE_CODE (new))
343 case INTEGER_TYPE:
344 case REAL_TYPE:
345 case FIXED_POINT_TYPE:
346 case ENUMERAL_TYPE:
347 case BOOLEAN_TYPE:
348 t = TYPE_MIN_VALUE (new);
349 if (t && TREE_CODE (t) != INTEGER_CST)
350 walk_tree (&TYPE_MIN_VALUE (new), copy_body_r, id, NULL);
352 t = TYPE_MAX_VALUE (new);
353 if (t && TREE_CODE (t) != INTEGER_CST)
354 walk_tree (&TYPE_MAX_VALUE (new), copy_body_r, id, NULL);
355 return new;
357 case FUNCTION_TYPE:
358 TREE_TYPE (new) = remap_type (TREE_TYPE (new), id);
359 walk_tree (&TYPE_ARG_TYPES (new), copy_body_r, id, NULL);
360 return new;
362 case ARRAY_TYPE:
363 TREE_TYPE (new) = remap_type (TREE_TYPE (new), id);
364 TYPE_DOMAIN (new) = remap_type (TYPE_DOMAIN (new), id);
365 break;
367 case RECORD_TYPE:
368 case UNION_TYPE:
369 case QUAL_UNION_TYPE:
371 tree f, nf = NULL;
373 for (f = TYPE_FIELDS (new); f ; f = TREE_CHAIN (f))
375 t = remap_decl (f, id);
376 DECL_CONTEXT (t) = new;
377 TREE_CHAIN (t) = nf;
378 nf = t;
380 TYPE_FIELDS (new) = nreverse (nf);
382 break;
384 case OFFSET_TYPE:
385 default:
386 /* Shouldn't have been thought variable sized. */
387 gcc_unreachable ();
390 walk_tree (&TYPE_SIZE (new), copy_body_r, id, NULL);
391 walk_tree (&TYPE_SIZE_UNIT (new), copy_body_r, id, NULL);
393 return new;
396 tree
397 remap_type (tree type, copy_body_data *id)
399 tree *node;
401 if (type == NULL)
402 return type;
404 /* See if we have remapped this type. */
405 node = (tree *) pointer_map_contains (id->decl_map, type);
406 if (node)
407 return *node;
409 /* The type only needs remapping if it's variably modified. */
410 if (! variably_modified_type_p (type, id->src_fn))
412 insert_decl_map (id, type, type);
413 return type;
416 return remap_type_1 (type, id);
419 static tree
420 remap_decls (tree decls, copy_body_data *id)
422 tree old_var;
423 tree new_decls = NULL_TREE;
425 /* Remap its variables. */
426 for (old_var = decls; old_var; old_var = TREE_CHAIN (old_var))
428 tree new_var;
430 /* We can not chain the local static declarations into the unexpanded_var_list
431 as we can't duplicate them or break one decl rule. Go ahead and link
432 them into unexpanded_var_list. */
433 if (!auto_var_in_fn_p (old_var, id->src_fn)
434 && !DECL_EXTERNAL (old_var))
436 cfun->unexpanded_var_list = tree_cons (NULL_TREE, old_var,
437 cfun->unexpanded_var_list);
438 continue;
441 /* Remap the variable. */
442 new_var = remap_decl (old_var, id);
444 /* If we didn't remap this variable, so we can't mess with its
445 TREE_CHAIN. If we remapped this variable to the return slot, it's
446 already declared somewhere else, so don't declare it here. */
447 if (!new_var || new_var == id->retvar)
449 else
451 gcc_assert (DECL_P (new_var));
452 TREE_CHAIN (new_var) = new_decls;
453 new_decls = new_var;
457 return nreverse (new_decls);
460 /* Copy the BLOCK to contain remapped versions of the variables
461 therein. And hook the new block into the block-tree. */
463 static void
464 remap_block (tree *block, copy_body_data *id)
466 tree old_block;
467 tree new_block;
468 tree fn;
470 /* Make the new block. */
471 old_block = *block;
472 new_block = make_node (BLOCK);
473 TREE_USED (new_block) = TREE_USED (old_block);
474 BLOCK_ABSTRACT_ORIGIN (new_block) = old_block;
475 BLOCK_SOURCE_LOCATION (new_block) = BLOCK_SOURCE_LOCATION (old_block);
476 *block = new_block;
478 /* Remap its variables. */
479 BLOCK_VARS (new_block) = remap_decls (BLOCK_VARS (old_block), id);
481 fn = id->dst_fn;
483 if (id->transform_lang_insert_block)
484 lang_hooks.decls.insert_block (new_block);
486 /* Remember the remapped block. */
487 insert_decl_map (id, old_block, new_block);
490 /* Copy the whole block tree and root it in id->block. */
491 static tree
492 remap_blocks (tree block, copy_body_data *id)
494 tree t;
495 tree new = block;
497 if (!block)
498 return NULL;
500 remap_block (&new, id);
501 gcc_assert (new != block);
502 for (t = BLOCK_SUBBLOCKS (block); t ; t = BLOCK_CHAIN (t))
503 add_lexical_block (new, remap_blocks (t, id));
504 return new;
507 static void
508 copy_statement_list (tree *tp)
510 tree_stmt_iterator oi, ni;
511 tree new;
513 new = alloc_stmt_list ();
514 ni = tsi_start (new);
515 oi = tsi_start (*tp);
516 *tp = new;
518 for (; !tsi_end_p (oi); tsi_next (&oi))
519 tsi_link_after (&ni, tsi_stmt (oi), TSI_NEW_STMT);
522 static void
523 copy_bind_expr (tree *tp, int *walk_subtrees, copy_body_data *id)
525 tree block = BIND_EXPR_BLOCK (*tp);
526 /* Copy (and replace) the statement. */
527 copy_tree_r (tp, walk_subtrees, NULL);
528 if (block)
530 remap_block (&block, id);
531 BIND_EXPR_BLOCK (*tp) = block;
534 if (BIND_EXPR_VARS (*tp))
535 /* This will remap a lot of the same decls again, but this should be
536 harmless. */
537 BIND_EXPR_VARS (*tp) = remap_decls (BIND_EXPR_VARS (*tp), id);
540 /* Called from copy_body_id via walk_tree. DATA is really an
541 `copy_body_data *'. */
543 tree
544 copy_body_r (tree *tp, int *walk_subtrees, void *data)
546 copy_body_data *id = (copy_body_data *) data;
547 tree fn = id->src_fn;
548 tree new_block;
550 /* Begin by recognizing trees that we'll completely rewrite for the
551 inlining context. Our output for these trees is completely
552 different from out input (e.g. RETURN_EXPR is deleted, and morphs
553 into an edge). Further down, we'll handle trees that get
554 duplicated and/or tweaked. */
556 /* When requested, RETURN_EXPRs should be transformed to just the
557 contained GIMPLE_MODIFY_STMT. The branch semantics of the return will
558 be handled elsewhere by manipulating the CFG rather than a statement. */
559 if (TREE_CODE (*tp) == RETURN_EXPR && id->transform_return_to_modify)
561 tree assignment = TREE_OPERAND (*tp, 0);
563 /* If we're returning something, just turn that into an
564 assignment into the equivalent of the original RESULT_DECL.
565 If the "assignment" is just the result decl, the result
566 decl has already been set (e.g. a recent "foo (&result_decl,
567 ...)"); just toss the entire RETURN_EXPR. */
568 if (assignment && TREE_CODE (assignment) == GIMPLE_MODIFY_STMT)
570 /* Replace the RETURN_EXPR with (a copy of) the
571 GIMPLE_MODIFY_STMT hanging underneath. */
572 *tp = copy_node (assignment);
574 else /* Else the RETURN_EXPR returns no value. */
576 *tp = NULL;
577 return (tree) (void *)1;
580 else if (TREE_CODE (*tp) == SSA_NAME)
582 *tp = remap_ssa_name (*tp, id);
583 *walk_subtrees = 0;
584 return NULL;
587 /* Local variables and labels need to be replaced by equivalent
588 variables. We don't want to copy static variables; there's only
589 one of those, no matter how many times we inline the containing
590 function. Similarly for globals from an outer function. */
591 else if (auto_var_in_fn_p (*tp, fn))
593 tree new_decl;
595 /* Remap the declaration. */
596 new_decl = remap_decl (*tp, id);
597 gcc_assert (new_decl);
598 /* Replace this variable with the copy. */
599 STRIP_TYPE_NOPS (new_decl);
600 *tp = new_decl;
601 *walk_subtrees = 0;
603 else if (TREE_CODE (*tp) == STATEMENT_LIST)
604 copy_statement_list (tp);
605 else if (TREE_CODE (*tp) == SAVE_EXPR)
606 remap_save_expr (tp, id->decl_map, walk_subtrees);
607 else if (TREE_CODE (*tp) == LABEL_DECL
608 && (! DECL_CONTEXT (*tp)
609 || decl_function_context (*tp) == id->src_fn))
610 /* These may need to be remapped for EH handling. */
611 *tp = remap_decl (*tp, id);
612 else if (TREE_CODE (*tp) == BIND_EXPR)
613 copy_bind_expr (tp, walk_subtrees, id);
614 /* Types may need remapping as well. */
615 else if (TYPE_P (*tp))
616 *tp = remap_type (*tp, id);
618 /* If this is a constant, we have to copy the node iff the type will be
619 remapped. copy_tree_r will not copy a constant. */
620 else if (CONSTANT_CLASS_P (*tp))
622 tree new_type = remap_type (TREE_TYPE (*tp), id);
624 if (new_type == TREE_TYPE (*tp))
625 *walk_subtrees = 0;
627 else if (TREE_CODE (*tp) == INTEGER_CST)
628 *tp = build_int_cst_wide (new_type, TREE_INT_CST_LOW (*tp),
629 TREE_INT_CST_HIGH (*tp));
630 else
632 *tp = copy_node (*tp);
633 TREE_TYPE (*tp) = new_type;
637 /* Otherwise, just copy the node. Note that copy_tree_r already
638 knows not to copy VAR_DECLs, etc., so this is safe. */
639 else
641 /* Here we handle trees that are not completely rewritten.
642 First we detect some inlining-induced bogosities for
643 discarding. */
644 if (TREE_CODE (*tp) == GIMPLE_MODIFY_STMT
645 && GIMPLE_STMT_OPERAND (*tp, 0) == GIMPLE_STMT_OPERAND (*tp, 1)
646 && (auto_var_in_fn_p (GIMPLE_STMT_OPERAND (*tp, 0), fn)))
648 /* Some assignments VAR = VAR; don't generate any rtl code
649 and thus don't count as variable modification. Avoid
650 keeping bogosities like 0 = 0. */
651 tree decl = GIMPLE_STMT_OPERAND (*tp, 0), value;
652 tree *n;
654 n = (tree *) pointer_map_contains (id->decl_map, decl);
655 if (n)
657 value = *n;
658 STRIP_TYPE_NOPS (value);
659 if (TREE_CONSTANT (value) || TREE_READONLY_DECL_P (value))
661 *tp = build_empty_stmt ();
662 return copy_body_r (tp, walk_subtrees, data);
666 else if (TREE_CODE (*tp) == INDIRECT_REF)
668 /* Get rid of *& from inline substitutions that can happen when a
669 pointer argument is an ADDR_EXPR. */
670 tree decl = TREE_OPERAND (*tp, 0);
671 tree *n;
673 n = (tree *) pointer_map_contains (id->decl_map, decl);
674 if (n)
676 tree new;
677 tree old;
678 /* If we happen to get an ADDR_EXPR in n->value, strip
679 it manually here as we'll eventually get ADDR_EXPRs
680 which lie about their types pointed to. In this case
681 build_fold_indirect_ref wouldn't strip the INDIRECT_REF,
682 but we absolutely rely on that. As fold_indirect_ref
683 does other useful transformations, try that first, though. */
684 tree type = TREE_TYPE (TREE_TYPE (*n));
685 new = unshare_expr (*n);
686 old = *tp;
687 *tp = fold_indirect_ref_1 (type, new);
688 if (! *tp)
690 if (TREE_CODE (new) == ADDR_EXPR)
691 *tp = TREE_OPERAND (new, 0);
692 else
694 *tp = build1 (INDIRECT_REF, type, new);
695 TREE_THIS_VOLATILE (*tp) = TREE_THIS_VOLATILE (old);
698 *walk_subtrees = 0;
699 return NULL;
703 /* Here is the "usual case". Copy this tree node, and then
704 tweak some special cases. */
705 copy_tree_r (tp, walk_subtrees, NULL);
707 /* Global variables we didn't seen yet needs to go into referenced
708 vars. */
709 if (gimple_in_ssa_p (cfun) && TREE_CODE (*tp) == VAR_DECL)
710 add_referenced_var (*tp);
712 /* If EXPR has block defined, map it to newly constructed block.
713 When inlining we want EXPRs without block appear in the block
714 of function call. */
715 if (EXPR_P (*tp) || GIMPLE_STMT_P (*tp))
717 new_block = id->block;
718 if (TREE_BLOCK (*tp))
720 tree *n;
721 n = (tree *) pointer_map_contains (id->decl_map,
722 TREE_BLOCK (*tp));
723 gcc_assert (n);
724 new_block = *n;
726 TREE_BLOCK (*tp) = new_block;
729 if (TREE_CODE (*tp) == RESX_EXPR && id->eh_region_offset)
730 TREE_OPERAND (*tp, 0) =
731 build_int_cst
732 (NULL_TREE,
733 id->eh_region_offset + TREE_INT_CST_LOW (TREE_OPERAND (*tp, 0)));
735 if (!GIMPLE_TUPLE_P (*tp) && TREE_CODE (*tp) != OMP_CLAUSE)
736 TREE_TYPE (*tp) = remap_type (TREE_TYPE (*tp), id);
738 /* The copied TARGET_EXPR has never been expanded, even if the
739 original node was expanded already. */
740 if (TREE_CODE (*tp) == TARGET_EXPR && TREE_OPERAND (*tp, 3))
742 TREE_OPERAND (*tp, 1) = TREE_OPERAND (*tp, 3);
743 TREE_OPERAND (*tp, 3) = NULL_TREE;
746 /* Variable substitution need not be simple. In particular, the
747 INDIRECT_REF substitution above. Make sure that TREE_CONSTANT
748 and friends are up-to-date. */
749 else if (TREE_CODE (*tp) == ADDR_EXPR)
751 walk_tree (&TREE_OPERAND (*tp, 0), copy_body_r, id, NULL);
752 /* Handle the case where we substituted an INDIRECT_REF
753 into the operand of the ADDR_EXPR. */
754 if (TREE_CODE (TREE_OPERAND (*tp, 0)) == INDIRECT_REF)
755 *tp = TREE_OPERAND (TREE_OPERAND (*tp, 0), 0);
756 else
757 recompute_tree_invariant_for_addr_expr (*tp);
758 *walk_subtrees = 0;
762 /* Keep iterating. */
763 return NULL_TREE;
766 /* Copy basic block, scale profile accordingly. Edges will be taken care of
767 later */
769 static basic_block
770 copy_bb (copy_body_data *id, basic_block bb, int frequency_scale, int count_scale)
772 block_stmt_iterator bsi, copy_bsi;
773 basic_block copy_basic_block;
775 /* create_basic_block() will append every new block to
776 basic_block_info automatically. */
777 copy_basic_block = create_basic_block (NULL, (void *) 0,
778 (basic_block) bb->prev_bb->aux);
779 copy_basic_block->count = bb->count * count_scale / REG_BR_PROB_BASE;
781 /* We are going to rebuild frequencies from scratch. These values have just
782 small importance to drive canonicalize_loop_headers. */
783 copy_basic_block->frequency = ((gcov_type)bb->frequency
784 * frequency_scale / REG_BR_PROB_BASE);
785 if (copy_basic_block->frequency > BB_FREQ_MAX)
786 copy_basic_block->frequency = BB_FREQ_MAX;
787 copy_bsi = bsi_start (copy_basic_block);
789 for (bsi = bsi_start (bb);
790 !bsi_end_p (bsi); bsi_next (&bsi))
792 tree stmt = bsi_stmt (bsi);
793 tree orig_stmt = stmt;
795 walk_tree (&stmt, copy_body_r, id, NULL);
797 /* RETURN_EXPR might be removed,
798 this is signalled by making stmt pointer NULL. */
799 if (stmt)
801 tree call, decl;
803 gimple_duplicate_stmt_histograms (cfun, stmt, id->src_cfun, orig_stmt);
805 /* With return slot optimization we can end up with
806 non-gimple (foo *)&this->m, fix that here. */
807 if (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT
808 && TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 1)) == NOP_EXPR
809 && !is_gimple_val (TREE_OPERAND (GIMPLE_STMT_OPERAND (stmt, 1), 0)))
810 gimplify_stmt (&stmt);
812 bsi_insert_after (&copy_bsi, stmt, BSI_NEW_STMT);
814 /* Process new statement. gimplify_stmt possibly turned statement
815 into multiple statements, we need to process all of them. */
816 while (!bsi_end_p (copy_bsi))
818 tree *stmtp = bsi_stmt_ptr (copy_bsi);
819 tree stmt = *stmtp;
820 call = get_call_expr_in (stmt);
822 if (call && CALL_EXPR_VA_ARG_PACK (call) && id->call_expr)
824 /* __builtin_va_arg_pack () should be replaced by
825 all arguments corresponding to ... in the caller. */
826 tree p, *argarray, new_call, *call_ptr;
827 int nargs = call_expr_nargs (id->call_expr);
829 for (p = DECL_ARGUMENTS (id->src_fn); p; p = TREE_CHAIN (p))
830 nargs--;
832 argarray = (tree *) alloca ((nargs + call_expr_nargs (call))
833 * sizeof (tree));
835 memcpy (argarray, CALL_EXPR_ARGP (call),
836 call_expr_nargs (call) * sizeof (*argarray));
837 memcpy (argarray + call_expr_nargs (call),
838 CALL_EXPR_ARGP (id->call_expr)
839 + (call_expr_nargs (id->call_expr) - nargs),
840 nargs * sizeof (*argarray));
842 new_call = build_call_array (TREE_TYPE (call),
843 CALL_EXPR_FN (call),
844 nargs + call_expr_nargs (call),
845 argarray);
846 /* Copy all CALL_EXPR flags, locus and block, except
847 CALL_EXPR_VA_ARG_PACK flag. */
848 CALL_EXPR_STATIC_CHAIN (new_call)
849 = CALL_EXPR_STATIC_CHAIN (call);
850 CALL_EXPR_TAILCALL (new_call) = CALL_EXPR_TAILCALL (call);
851 CALL_EXPR_RETURN_SLOT_OPT (new_call)
852 = CALL_EXPR_RETURN_SLOT_OPT (call);
853 CALL_FROM_THUNK_P (new_call) = CALL_FROM_THUNK_P (call);
854 CALL_CANNOT_INLINE_P (new_call)
855 = CALL_CANNOT_INLINE_P (call);
856 TREE_NOTHROW (new_call) = TREE_NOTHROW (call);
857 SET_EXPR_LOCUS (new_call, EXPR_LOCUS (call));
858 TREE_BLOCK (new_call) = TREE_BLOCK (call);
860 call_ptr = stmtp;
861 if (TREE_CODE (*call_ptr) == GIMPLE_MODIFY_STMT)
862 call_ptr = &GIMPLE_STMT_OPERAND (*call_ptr, 1);
863 if (TREE_CODE (*call_ptr) == WITH_SIZE_EXPR)
864 call_ptr = &TREE_OPERAND (*call_ptr, 0);
865 gcc_assert (*call_ptr == call);
866 *call_ptr = new_call;
867 stmt = *stmtp;
868 update_stmt (stmt);
870 else if (call
871 && id->call_expr
872 && (decl = get_callee_fndecl (call))
873 && DECL_BUILT_IN_CLASS (decl) == BUILT_IN_NORMAL
874 && DECL_FUNCTION_CODE (decl)
875 == BUILT_IN_VA_ARG_PACK_LEN)
877 /* __builtin_va_arg_pack_len () should be replaced by
878 the number of anonymous arguments. */
879 int nargs = call_expr_nargs (id->call_expr);
880 tree count, *call_ptr, p;
882 for (p = DECL_ARGUMENTS (id->src_fn); p; p = TREE_CHAIN (p))
883 nargs--;
885 count = build_int_cst (integer_type_node, nargs);
886 call_ptr = stmtp;
887 if (TREE_CODE (*call_ptr) == GIMPLE_MODIFY_STMT)
888 call_ptr = &GIMPLE_STMT_OPERAND (*call_ptr, 1);
889 if (TREE_CODE (*call_ptr) == WITH_SIZE_EXPR)
890 call_ptr = &TREE_OPERAND (*call_ptr, 0);
891 gcc_assert (*call_ptr == call && call_ptr != stmtp);
892 *call_ptr = count;
893 stmt = *stmtp;
894 update_stmt (stmt);
895 call = NULL_TREE;
898 /* Statements produced by inlining can be unfolded, especially
899 when we constant propagated some operands. We can't fold
900 them right now for two reasons:
901 1) folding require SSA_NAME_DEF_STMTs to be correct
902 2) we can't change function calls to builtins.
903 So we just mark statement for later folding. We mark
904 all new statements, instead just statements that has changed
905 by some nontrivial substitution so even statements made
906 foldable indirectly are updated. If this turns out to be
907 expensive, copy_body can be told to watch for nontrivial
908 changes. */
909 if (id->statements_to_fold)
910 pointer_set_insert (id->statements_to_fold, stmt);
911 /* We're duplicating a CALL_EXPR. Find any corresponding
912 callgraph edges and update or duplicate them. */
913 if (call && (decl = get_callee_fndecl (call)))
915 struct cgraph_node *node;
916 struct cgraph_edge *edge;
918 switch (id->transform_call_graph_edges)
920 case CB_CGE_DUPLICATE:
921 edge = cgraph_edge (id->src_node, orig_stmt);
922 if (edge)
923 cgraph_clone_edge (edge, id->dst_node, stmt,
924 REG_BR_PROB_BASE, 1, edge->frequency, true);
925 break;
927 case CB_CGE_MOVE_CLONES:
928 for (node = id->dst_node->next_clone;
929 node;
930 node = node->next_clone)
932 edge = cgraph_edge (node, orig_stmt);
933 gcc_assert (edge);
934 cgraph_set_call_stmt (edge, stmt);
936 /* FALLTHRU */
938 case CB_CGE_MOVE:
939 edge = cgraph_edge (id->dst_node, orig_stmt);
940 if (edge)
941 cgraph_set_call_stmt (edge, stmt);
942 break;
944 default:
945 gcc_unreachable ();
948 /* If you think we can abort here, you are wrong.
949 There is no region 0 in tree land. */
950 gcc_assert (lookup_stmt_eh_region_fn (id->src_cfun, orig_stmt)
951 != 0);
953 if (tree_could_throw_p (stmt)
954 /* When we are cloning for inlining, we are supposed to
955 construct a clone that calls precisely the same functions
956 as original. However IPA optimizers might've proved
957 earlier some function calls as non-trapping that might
958 render some basic blocks dead that might become
959 unreachable.
961 We can't update SSA with unreachable blocks in CFG and thus
962 we prevent the scenario by preserving even the "dead" eh
963 edges until the point they are later removed by
964 fixup_cfg pass. */
965 || (id->transform_call_graph_edges == CB_CGE_MOVE_CLONES
966 && lookup_stmt_eh_region_fn (id->src_cfun, orig_stmt) > 0))
968 int region = lookup_stmt_eh_region_fn (id->src_cfun, orig_stmt);
969 /* Add an entry for the copied tree in the EH hashtable.
970 When cloning or versioning, use the hashtable in
971 cfun, and just copy the EH number. When inlining, use the
972 hashtable in the caller, and adjust the region number. */
973 if (region > 0)
974 add_stmt_to_eh_region (stmt, region + id->eh_region_offset);
976 /* If this tree doesn't have a region associated with it,
977 and there is a "current region,"
978 then associate this tree with the current region
979 and add edges associated with this region. */
980 if ((lookup_stmt_eh_region_fn (id->src_cfun,
981 orig_stmt) <= 0
982 && id->eh_region > 0)
983 && tree_could_throw_p (stmt))
984 add_stmt_to_eh_region (stmt, id->eh_region);
986 if (gimple_in_ssa_p (cfun))
988 ssa_op_iter i;
989 tree def;
991 find_new_referenced_vars (bsi_stmt_ptr (copy_bsi));
992 FOR_EACH_SSA_TREE_OPERAND (def, stmt, i, SSA_OP_DEF)
993 if (TREE_CODE (def) == SSA_NAME)
994 SSA_NAME_DEF_STMT (def) = stmt;
996 bsi_next (&copy_bsi);
998 copy_bsi = bsi_last (copy_basic_block);
1001 return copy_basic_block;
1004 /* Inserting Single Entry Multiple Exit region in SSA form into code in SSA
1005 form is quite easy, since dominator relationship for old basic blocks does
1006 not change.
1008 There is however exception where inlining might change dominator relation
1009 across EH edges from basic block within inlined functions destinating
1010 to landing pads in function we inline into.
1012 The function mark PHI_RESULT of such PHI nodes for renaming; it is
1013 safe the EH edges are abnormal and SSA_NAME_OCCURS_IN_ABNORMAL_PHI
1014 must be set. This means, that there will be no overlapping live ranges
1015 for the underlying symbol.
1017 This might change in future if we allow redirecting of EH edges and
1018 we might want to change way build CFG pre-inlining to include
1019 all the possible edges then. */
1020 static void
1021 update_ssa_across_eh_edges (basic_block bb)
1023 edge e;
1024 edge_iterator ei;
1026 FOR_EACH_EDGE (e, ei, bb->succs)
1027 if (!e->dest->aux
1028 || ((basic_block)e->dest->aux)->index == ENTRY_BLOCK)
1030 tree phi;
1032 gcc_assert (e->flags & EDGE_EH);
1033 for (phi = phi_nodes (e->dest); phi; phi = PHI_CHAIN (phi))
1035 gcc_assert (SSA_NAME_OCCURS_IN_ABNORMAL_PHI
1036 (PHI_RESULT (phi)));
1037 mark_sym_for_renaming
1038 (SSA_NAME_VAR (PHI_RESULT (phi)));
1043 /* Copy edges from BB into its copy constructed earlier, scale profile
1044 accordingly. Edges will be taken care of later. Assume aux
1045 pointers to point to the copies of each BB. */
1046 static void
1047 copy_edges_for_bb (basic_block bb, int count_scale)
1049 basic_block new_bb = (basic_block) bb->aux;
1050 edge_iterator ei;
1051 edge old_edge;
1052 block_stmt_iterator bsi;
1053 int flags;
1055 /* Use the indices from the original blocks to create edges for the
1056 new ones. */
1057 FOR_EACH_EDGE (old_edge, ei, bb->succs)
1058 if (!(old_edge->flags & EDGE_EH))
1060 edge new;
1062 flags = old_edge->flags;
1064 /* Return edges do get a FALLTHRU flag when the get inlined. */
1065 if (old_edge->dest->index == EXIT_BLOCK && !old_edge->flags
1066 && old_edge->dest->aux != EXIT_BLOCK_PTR)
1067 flags |= EDGE_FALLTHRU;
1068 new = make_edge (new_bb, (basic_block) old_edge->dest->aux, flags);
1069 new->count = old_edge->count * count_scale / REG_BR_PROB_BASE;
1070 new->probability = old_edge->probability;
1073 if (bb->index == ENTRY_BLOCK || bb->index == EXIT_BLOCK)
1074 return;
1076 for (bsi = bsi_start (new_bb); !bsi_end_p (bsi);)
1078 tree copy_stmt;
1080 copy_stmt = bsi_stmt (bsi);
1081 update_stmt (copy_stmt);
1082 if (gimple_in_ssa_p (cfun))
1083 mark_symbols_for_renaming (copy_stmt);
1084 /* Do this before the possible split_block. */
1085 bsi_next (&bsi);
1087 /* If this tree could throw an exception, there are two
1088 cases where we need to add abnormal edge(s): the
1089 tree wasn't in a region and there is a "current
1090 region" in the caller; or the original tree had
1091 EH edges. In both cases split the block after the tree,
1092 and add abnormal edge(s) as needed; we need both
1093 those from the callee and the caller.
1094 We check whether the copy can throw, because the const
1095 propagation can change an INDIRECT_REF which throws
1096 into a COMPONENT_REF which doesn't. If the copy
1097 can throw, the original could also throw. */
1099 if (tree_can_throw_internal (copy_stmt))
1101 if (!bsi_end_p (bsi))
1102 /* Note that bb's predecessor edges aren't necessarily
1103 right at this point; split_block doesn't care. */
1105 edge e = split_block (new_bb, copy_stmt);
1107 new_bb = e->dest;
1108 new_bb->aux = e->src->aux;
1109 bsi = bsi_start (new_bb);
1112 make_eh_edges (copy_stmt);
1114 if (gimple_in_ssa_p (cfun))
1115 update_ssa_across_eh_edges (bb_for_stmt (copy_stmt));
1120 /* Copy the PHIs. All blocks and edges are copied, some blocks
1121 was possibly split and new outgoing EH edges inserted.
1122 BB points to the block of original function and AUX pointers links
1123 the original and newly copied blocks. */
1125 static void
1126 copy_phis_for_bb (basic_block bb, copy_body_data *id)
1128 basic_block new_bb = bb->aux;
1129 edge_iterator ei;
1130 tree phi;
1132 for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
1134 tree res = PHI_RESULT (phi);
1135 tree new_res = res;
1136 tree new_phi;
1137 edge new_edge;
1139 if (is_gimple_reg (res))
1141 walk_tree (&new_res, copy_body_r, id, NULL);
1142 SSA_NAME_DEF_STMT (new_res)
1143 = new_phi = create_phi_node (new_res, new_bb);
1144 FOR_EACH_EDGE (new_edge, ei, new_bb->preds)
1146 edge old_edge = find_edge (new_edge->src->aux, bb);
1147 tree arg = PHI_ARG_DEF_FROM_EDGE (phi, old_edge);
1148 tree new_arg = arg;
1150 walk_tree (&new_arg, copy_body_r, id, NULL);
1151 gcc_assert (new_arg);
1152 add_phi_arg (new_phi, new_arg, new_edge);
1158 /* Wrapper for remap_decl so it can be used as a callback. */
1159 static tree
1160 remap_decl_1 (tree decl, void *data)
1162 return remap_decl (decl, (copy_body_data *) data);
1165 /* Build struct function and associated datastructures for the new clone
1166 NEW_FNDECL to be build. CALLEE_FNDECL is the original */
1168 static void
1169 initialize_cfun (tree new_fndecl, tree callee_fndecl, gcov_type count,
1170 int frequency)
1172 struct function *new_cfun
1173 = (struct function *) ggc_alloc_cleared (sizeof (struct function));
1174 struct function *src_cfun = DECL_STRUCT_FUNCTION (callee_fndecl);
1175 int count_scale, frequency_scale;
1177 if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count)
1178 count_scale = (REG_BR_PROB_BASE * count
1179 / ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count);
1180 else
1181 count_scale = 1;
1183 if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency)
1184 frequency_scale = (REG_BR_PROB_BASE * frequency
1186 ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency);
1187 else
1188 frequency_scale = count_scale;
1190 /* Register specific tree functions. */
1191 tree_register_cfg_hooks ();
1192 *new_cfun = *DECL_STRUCT_FUNCTION (callee_fndecl);
1193 new_cfun->funcdef_no = get_next_funcdef_no ();
1194 VALUE_HISTOGRAMS (new_cfun) = NULL;
1195 new_cfun->unexpanded_var_list = NULL;
1196 new_cfun->cfg = NULL;
1197 new_cfun->decl = new_fndecl /*= copy_node (callee_fndecl)*/;
1198 DECL_STRUCT_FUNCTION (new_fndecl) = new_cfun;
1199 push_cfun (new_cfun);
1200 init_empty_tree_cfg ();
1202 ENTRY_BLOCK_PTR->count =
1203 (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count * count_scale /
1204 REG_BR_PROB_BASE);
1205 ENTRY_BLOCK_PTR->frequency =
1206 (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency *
1207 frequency_scale / REG_BR_PROB_BASE);
1208 EXIT_BLOCK_PTR->count =
1209 (EXIT_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count * count_scale /
1210 REG_BR_PROB_BASE);
1211 EXIT_BLOCK_PTR->frequency =
1212 (EXIT_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency *
1213 frequency_scale / REG_BR_PROB_BASE);
1214 if (src_cfun->eh)
1215 init_eh_for_function ();
1217 if (src_cfun->gimple_df)
1219 init_tree_ssa ();
1220 cfun->gimple_df->in_ssa_p = true;
1221 init_ssa_operands ();
1223 pop_cfun ();
1226 /* Make a copy of the body of FN so that it can be inserted inline in
1227 another function. Walks FN via CFG, returns new fndecl. */
1229 static tree
1230 copy_cfg_body (copy_body_data * id, gcov_type count, int frequency,
1231 basic_block entry_block_map, basic_block exit_block_map)
1233 tree callee_fndecl = id->src_fn;
1234 /* Original cfun for the callee, doesn't change. */
1235 struct function *src_cfun = DECL_STRUCT_FUNCTION (callee_fndecl);
1236 struct function *cfun_to_copy;
1237 basic_block bb;
1238 tree new_fndecl = NULL;
1239 int count_scale, frequency_scale;
1240 int last;
1242 if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count)
1243 count_scale = (REG_BR_PROB_BASE * count
1244 / ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count);
1245 else
1246 count_scale = 1;
1248 if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency)
1249 frequency_scale = (REG_BR_PROB_BASE * frequency
1251 ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency);
1252 else
1253 frequency_scale = count_scale;
1255 /* Register specific tree functions. */
1256 tree_register_cfg_hooks ();
1258 /* Must have a CFG here at this point. */
1259 gcc_assert (ENTRY_BLOCK_PTR_FOR_FUNCTION
1260 (DECL_STRUCT_FUNCTION (callee_fndecl)));
1262 cfun_to_copy = id->src_cfun = DECL_STRUCT_FUNCTION (callee_fndecl);
1265 ENTRY_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy)->aux = entry_block_map;
1266 EXIT_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy)->aux = exit_block_map;
1267 entry_block_map->aux = ENTRY_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy);
1268 exit_block_map->aux = EXIT_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy);
1270 /* Duplicate any exception-handling regions. */
1271 if (cfun->eh)
1273 id->eh_region_offset
1274 = duplicate_eh_regions (cfun_to_copy, remap_decl_1, id,
1275 0, id->eh_region);
1277 /* Use aux pointers to map the original blocks to copy. */
1278 FOR_EACH_BB_FN (bb, cfun_to_copy)
1280 basic_block new = copy_bb (id, bb, frequency_scale, count_scale);
1281 bb->aux = new;
1282 new->aux = bb;
1285 last = last_basic_block;
1286 /* Now that we've duplicated the blocks, duplicate their edges. */
1287 FOR_ALL_BB_FN (bb, cfun_to_copy)
1288 copy_edges_for_bb (bb, count_scale);
1289 if (gimple_in_ssa_p (cfun))
1290 FOR_ALL_BB_FN (bb, cfun_to_copy)
1291 copy_phis_for_bb (bb, id);
1292 FOR_ALL_BB_FN (bb, cfun_to_copy)
1294 ((basic_block)bb->aux)->aux = NULL;
1295 bb->aux = NULL;
1297 /* Zero out AUX fields of newly created block during EH edge
1298 insertion. */
1299 for (; last < last_basic_block; last++)
1300 BASIC_BLOCK (last)->aux = NULL;
1301 entry_block_map->aux = NULL;
1302 exit_block_map->aux = NULL;
1304 return new_fndecl;
1307 /* Make a copy of the body of FN so that it can be inserted inline in
1308 another function. */
1310 static tree
1311 copy_generic_body (copy_body_data *id)
1313 tree body;
1314 tree fndecl = id->src_fn;
1316 body = DECL_SAVED_TREE (fndecl);
1317 walk_tree (&body, copy_body_r, id, NULL);
1319 return body;
1322 static tree
1323 copy_body (copy_body_data *id, gcov_type count, int frequency,
1324 basic_block entry_block_map, basic_block exit_block_map)
1326 tree fndecl = id->src_fn;
1327 tree body;
1329 /* If this body has a CFG, walk CFG and copy. */
1330 gcc_assert (ENTRY_BLOCK_PTR_FOR_FUNCTION (DECL_STRUCT_FUNCTION (fndecl)));
1331 body = copy_cfg_body (id, count, frequency, entry_block_map, exit_block_map);
1333 return body;
1336 /* Return true if VALUE is an ADDR_EXPR of an automatic variable
1337 defined in function FN, or of a data member thereof. */
1339 static bool
1340 self_inlining_addr_expr (tree value, tree fn)
1342 tree var;
1344 if (TREE_CODE (value) != ADDR_EXPR)
1345 return false;
1347 var = get_base_address (TREE_OPERAND (value, 0));
1349 return var && auto_var_in_fn_p (var, fn);
1352 static void
1353 setup_one_parameter (copy_body_data *id, tree p, tree value, tree fn,
1354 basic_block bb, tree *vars)
1356 tree init_stmt;
1357 tree var;
1358 tree var_sub;
1359 tree rhs = value;
1360 tree def = (gimple_in_ssa_p (cfun)
1361 ? gimple_default_def (id->src_cfun, p) : NULL);
1363 if (value
1364 && value != error_mark_node
1365 && !useless_type_conversion_p (TREE_TYPE (p), TREE_TYPE (value)))
1366 rhs = fold_build1 (NOP_EXPR, TREE_TYPE (p), value);
1368 /* If the parameter is never assigned to, has no SSA_NAMEs created,
1369 we may not need to create a new variable here at all. Instead, we may
1370 be able to just use the argument value. */
1371 if (TREE_READONLY (p)
1372 && !TREE_ADDRESSABLE (p)
1373 && value && !TREE_SIDE_EFFECTS (value)
1374 && !def)
1376 /* We may produce non-gimple trees by adding NOPs or introduce
1377 invalid sharing when operand is not really constant.
1378 It is not big deal to prohibit constant propagation here as
1379 we will constant propagate in DOM1 pass anyway. */
1380 if (is_gimple_min_invariant (value)
1381 && useless_type_conversion_p (TREE_TYPE (p),
1382 TREE_TYPE (value))
1383 /* We have to be very careful about ADDR_EXPR. Make sure
1384 the base variable isn't a local variable of the inlined
1385 function, e.g., when doing recursive inlining, direct or
1386 mutually-recursive or whatever, which is why we don't
1387 just test whether fn == current_function_decl. */
1388 && ! self_inlining_addr_expr (value, fn))
1390 insert_decl_map (id, p, value);
1391 return;
1395 /* Make an equivalent VAR_DECL. Note that we must NOT remap the type
1396 here since the type of this decl must be visible to the calling
1397 function. */
1398 var = copy_decl_to_var (p, id);
1399 if (gimple_in_ssa_p (cfun) && TREE_CODE (var) == VAR_DECL)
1401 get_var_ann (var);
1402 add_referenced_var (var);
1405 /* See if the frontend wants to pass this by invisible reference. If
1406 so, our new VAR_DECL will have REFERENCE_TYPE, and we need to
1407 replace uses of the PARM_DECL with dereferences. */
1408 if (TREE_TYPE (var) != TREE_TYPE (p)
1409 && POINTER_TYPE_P (TREE_TYPE (var))
1410 && TREE_TYPE (TREE_TYPE (var)) == TREE_TYPE (p))
1412 insert_decl_map (id, var, var);
1413 var_sub = build_fold_indirect_ref (var);
1415 else
1416 var_sub = var;
1418 /* Register the VAR_DECL as the equivalent for the PARM_DECL;
1419 that way, when the PARM_DECL is encountered, it will be
1420 automatically replaced by the VAR_DECL. */
1421 insert_decl_map (id, p, var_sub);
1423 /* Declare this new variable. */
1424 TREE_CHAIN (var) = *vars;
1425 *vars = var;
1427 /* Make gimplifier happy about this variable. */
1428 DECL_SEEN_IN_BIND_EXPR_P (var) = 1;
1430 /* Even if P was TREE_READONLY, the new VAR should not be.
1431 In the original code, we would have constructed a
1432 temporary, and then the function body would have never
1433 changed the value of P. However, now, we will be
1434 constructing VAR directly. The constructor body may
1435 change its value multiple times as it is being
1436 constructed. Therefore, it must not be TREE_READONLY;
1437 the back-end assumes that TREE_READONLY variable is
1438 assigned to only once. */
1439 if (TYPE_NEEDS_CONSTRUCTING (TREE_TYPE (p)))
1440 TREE_READONLY (var) = 0;
1442 /* If there is no setup required and we are in SSA, take the easy route
1443 replacing all SSA names representing the function parameter by the
1444 SSA name passed to function.
1446 We need to construct map for the variable anyway as it might be used
1447 in different SSA names when parameter is set in function.
1449 FIXME: This usually kills the last connection in between inlined
1450 function parameter and the actual value in debug info. Can we do
1451 better here? If we just inserted the statement, copy propagation
1452 would kill it anyway as it always did in older versions of GCC.
1454 We might want to introduce a notion that single SSA_NAME might
1455 represent multiple variables for purposes of debugging. */
1456 if (gimple_in_ssa_p (cfun) && rhs && def && is_gimple_reg (p)
1457 && (TREE_CODE (rhs) == SSA_NAME
1458 || is_gimple_min_invariant (rhs))
1459 && !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (def))
1461 insert_decl_map (id, def, rhs);
1462 return;
1465 /* Initialize this VAR_DECL from the equivalent argument. Convert
1466 the argument to the proper type in case it was promoted. */
1467 if (value)
1469 block_stmt_iterator bsi = bsi_last (bb);
1471 if (rhs == error_mark_node)
1473 insert_decl_map (id, p, var_sub);
1474 return;
1477 STRIP_USELESS_TYPE_CONVERSION (rhs);
1479 /* We want to use GIMPLE_MODIFY_STMT, not INIT_EXPR here so that we
1480 keep our trees in gimple form. */
1481 if (def && gimple_in_ssa_p (cfun) && is_gimple_reg (p))
1483 def = remap_ssa_name (def, id);
1484 init_stmt = build_gimple_modify_stmt (def, rhs);
1485 SSA_NAME_DEF_STMT (def) = init_stmt;
1486 SSA_NAME_IS_DEFAULT_DEF (def) = 0;
1487 set_default_def (var, NULL);
1489 else
1490 init_stmt = build_gimple_modify_stmt (var, rhs);
1492 /* If we did not create a gimple value and we did not create a gimple
1493 cast of a gimple value, then we will need to gimplify INIT_STMTS
1494 at the end. Note that is_gimple_cast only checks the outer
1495 tree code, not its operand. Thus the explicit check that its
1496 operand is a gimple value. */
1497 if ((!is_gimple_val (rhs)
1498 && (!is_gimple_cast (rhs)
1499 || !is_gimple_val (TREE_OPERAND (rhs, 0))))
1500 || !is_gimple_reg (var))
1502 tree_stmt_iterator i;
1504 push_gimplify_context ();
1505 gimplify_stmt (&init_stmt);
1506 if (gimple_in_ssa_p (cfun)
1507 && init_stmt && TREE_CODE (init_stmt) == STATEMENT_LIST)
1509 /* The replacement can expose previously unreferenced
1510 variables. */
1511 for (i = tsi_start (init_stmt); !tsi_end_p (i); tsi_next (&i))
1512 find_new_referenced_vars (tsi_stmt_ptr (i));
1514 pop_gimplify_context (NULL);
1517 /* If VAR represents a zero-sized variable, it's possible that the
1518 assignment statment may result in no gimple statements. */
1519 if (init_stmt)
1520 bsi_insert_after (&bsi, init_stmt, BSI_NEW_STMT);
1521 if (gimple_in_ssa_p (cfun))
1522 for (;!bsi_end_p (bsi); bsi_next (&bsi))
1523 mark_symbols_for_renaming (bsi_stmt (bsi));
1527 /* Generate code to initialize the parameters of the function at the
1528 top of the stack in ID from the CALL_EXPR EXP. */
1530 static void
1531 initialize_inlined_parameters (copy_body_data *id, tree exp,
1532 tree fn, basic_block bb)
1534 tree parms;
1535 tree a;
1536 tree p;
1537 tree vars = NULL_TREE;
1538 call_expr_arg_iterator iter;
1539 tree static_chain = CALL_EXPR_STATIC_CHAIN (exp);
1541 /* Figure out what the parameters are. */
1542 parms = DECL_ARGUMENTS (fn);
1544 /* Loop through the parameter declarations, replacing each with an
1545 equivalent VAR_DECL, appropriately initialized. */
1546 for (p = parms, a = first_call_expr_arg (exp, &iter); p;
1547 a = next_call_expr_arg (&iter), p = TREE_CHAIN (p))
1548 setup_one_parameter (id, p, a, fn, bb, &vars);
1550 /* Initialize the static chain. */
1551 p = DECL_STRUCT_FUNCTION (fn)->static_chain_decl;
1552 gcc_assert (fn != current_function_decl);
1553 if (p)
1555 /* No static chain? Seems like a bug in tree-nested.c. */
1556 gcc_assert (static_chain);
1558 setup_one_parameter (id, p, static_chain, fn, bb, &vars);
1561 declare_inline_vars (id->block, vars);
1564 /* Declare a return variable to replace the RESULT_DECL for the
1565 function we are calling. An appropriate DECL_STMT is returned.
1566 The USE_STMT is filled to contain a use of the declaration to
1567 indicate the return value of the function.
1569 RETURN_SLOT, if non-null is place where to store the result. It
1570 is set only for CALL_EXPR_RETURN_SLOT_OPT. MODIFY_DEST, if non-null,
1571 was the LHS of the GIMPLE_MODIFY_STMT to which this call is the RHS.
1573 The return value is a (possibly null) value that is the result of the
1574 function as seen by the callee. *USE_P is a (possibly null) value that
1575 holds the result as seen by the caller. */
1577 static tree
1578 declare_return_variable (copy_body_data *id, tree return_slot, tree modify_dest,
1579 tree *use_p)
1581 tree callee = id->src_fn;
1582 tree caller = id->dst_fn;
1583 tree result = DECL_RESULT (callee);
1584 tree callee_type = TREE_TYPE (result);
1585 tree caller_type = TREE_TYPE (TREE_TYPE (callee));
1586 tree var, use;
1588 /* We don't need to do anything for functions that don't return
1589 anything. */
1590 if (!result || VOID_TYPE_P (callee_type))
1592 *use_p = NULL_TREE;
1593 return NULL_TREE;
1596 /* If there was a return slot, then the return value is the
1597 dereferenced address of that object. */
1598 if (return_slot)
1600 /* The front end shouldn't have used both return_slot and
1601 a modify expression. */
1602 gcc_assert (!modify_dest);
1603 if (DECL_BY_REFERENCE (result))
1605 tree return_slot_addr = build_fold_addr_expr (return_slot);
1606 STRIP_USELESS_TYPE_CONVERSION (return_slot_addr);
1608 /* We are going to construct *&return_slot and we can't do that
1609 for variables believed to be not addressable.
1611 FIXME: This check possibly can match, because values returned
1612 via return slot optimization are not believed to have address
1613 taken by alias analysis. */
1614 gcc_assert (TREE_CODE (return_slot) != SSA_NAME);
1615 if (gimple_in_ssa_p (cfun))
1617 HOST_WIDE_INT bitsize;
1618 HOST_WIDE_INT bitpos;
1619 tree offset;
1620 enum machine_mode mode;
1621 int unsignedp;
1622 int volatilep;
1623 tree base;
1624 base = get_inner_reference (return_slot, &bitsize, &bitpos,
1625 &offset,
1626 &mode, &unsignedp, &volatilep,
1627 false);
1628 if (TREE_CODE (base) == INDIRECT_REF)
1629 base = TREE_OPERAND (base, 0);
1630 if (TREE_CODE (base) == SSA_NAME)
1631 base = SSA_NAME_VAR (base);
1632 mark_sym_for_renaming (base);
1634 var = return_slot_addr;
1636 else
1638 var = return_slot;
1639 gcc_assert (TREE_CODE (var) != SSA_NAME);
1641 if ((TREE_CODE (TREE_TYPE (result)) == COMPLEX_TYPE
1642 || TREE_CODE (TREE_TYPE (result)) == VECTOR_TYPE)
1643 && !DECL_GIMPLE_REG_P (result)
1644 && DECL_P (var))
1645 DECL_GIMPLE_REG_P (var) = 0;
1646 use = NULL;
1647 goto done;
1650 /* All types requiring non-trivial constructors should have been handled. */
1651 gcc_assert (!TREE_ADDRESSABLE (callee_type));
1653 /* Attempt to avoid creating a new temporary variable. */
1654 if (modify_dest
1655 && TREE_CODE (modify_dest) != SSA_NAME)
1657 bool use_it = false;
1659 /* We can't use MODIFY_DEST if there's type promotion involved. */
1660 if (!useless_type_conversion_p (callee_type, caller_type))
1661 use_it = false;
1663 /* ??? If we're assigning to a variable sized type, then we must
1664 reuse the destination variable, because we've no good way to
1665 create variable sized temporaries at this point. */
1666 else if (TREE_CODE (TYPE_SIZE_UNIT (caller_type)) != INTEGER_CST)
1667 use_it = true;
1669 /* If the callee cannot possibly modify MODIFY_DEST, then we can
1670 reuse it as the result of the call directly. Don't do this if
1671 it would promote MODIFY_DEST to addressable. */
1672 else if (TREE_ADDRESSABLE (result))
1673 use_it = false;
1674 else
1676 tree base_m = get_base_address (modify_dest);
1678 /* If the base isn't a decl, then it's a pointer, and we don't
1679 know where that's going to go. */
1680 if (!DECL_P (base_m))
1681 use_it = false;
1682 else if (is_global_var (base_m))
1683 use_it = false;
1684 else if ((TREE_CODE (TREE_TYPE (result)) == COMPLEX_TYPE
1685 || TREE_CODE (TREE_TYPE (result)) == VECTOR_TYPE)
1686 && !DECL_GIMPLE_REG_P (result)
1687 && DECL_GIMPLE_REG_P (base_m))
1688 use_it = false;
1689 else if (!TREE_ADDRESSABLE (base_m))
1690 use_it = true;
1693 if (use_it)
1695 var = modify_dest;
1696 use = NULL;
1697 goto done;
1701 gcc_assert (TREE_CODE (TYPE_SIZE_UNIT (callee_type)) == INTEGER_CST);
1703 var = copy_result_decl_to_var (result, id);
1704 if (gimple_in_ssa_p (cfun))
1706 get_var_ann (var);
1707 add_referenced_var (var);
1710 DECL_SEEN_IN_BIND_EXPR_P (var) = 1;
1711 DECL_STRUCT_FUNCTION (caller)->unexpanded_var_list
1712 = tree_cons (NULL_TREE, var,
1713 DECL_STRUCT_FUNCTION (caller)->unexpanded_var_list);
1715 /* Do not have the rest of GCC warn about this variable as it should
1716 not be visible to the user. */
1717 TREE_NO_WARNING (var) = 1;
1719 declare_inline_vars (id->block, var);
1721 /* Build the use expr. If the return type of the function was
1722 promoted, convert it back to the expected type. */
1723 use = var;
1724 if (!useless_type_conversion_p (caller_type, TREE_TYPE (var)))
1725 use = fold_convert (caller_type, var);
1727 STRIP_USELESS_TYPE_CONVERSION (use);
1729 if (DECL_BY_REFERENCE (result))
1730 var = build_fold_addr_expr (var);
1732 done:
1733 /* Register the VAR_DECL as the equivalent for the RESULT_DECL; that
1734 way, when the RESULT_DECL is encountered, it will be
1735 automatically replaced by the VAR_DECL. */
1736 insert_decl_map (id, result, var);
1738 /* Remember this so we can ignore it in remap_decls. */
1739 id->retvar = var;
1741 *use_p = use;
1742 return var;
1745 /* Returns nonzero if a function can be inlined as a tree. */
1747 bool
1748 tree_inlinable_function_p (tree fn)
1750 return inlinable_function_p (fn);
1753 static const char *inline_forbidden_reason;
1755 static tree
1756 inline_forbidden_p_1 (tree *nodep, int *walk_subtrees ATTRIBUTE_UNUSED,
1757 void *fnp)
1759 tree node = *nodep;
1760 tree fn = (tree) fnp;
1761 tree t;
1763 switch (TREE_CODE (node))
1765 case CALL_EXPR:
1766 /* Refuse to inline alloca call unless user explicitly forced so as
1767 this may change program's memory overhead drastically when the
1768 function using alloca is called in loop. In GCC present in
1769 SPEC2000 inlining into schedule_block cause it to require 2GB of
1770 RAM instead of 256MB. */
1771 if (alloca_call_p (node)
1772 && !lookup_attribute ("always_inline", DECL_ATTRIBUTES (fn)))
1774 inline_forbidden_reason
1775 = G_("function %q+F can never be inlined because it uses "
1776 "alloca (override using the always_inline attribute)");
1777 return node;
1779 t = get_callee_fndecl (node);
1780 if (! t)
1781 break;
1783 /* We cannot inline functions that call setjmp. */
1784 if (setjmp_call_p (t))
1786 inline_forbidden_reason
1787 = G_("function %q+F can never be inlined because it uses setjmp");
1788 return node;
1791 if (DECL_BUILT_IN_CLASS (t) == BUILT_IN_NORMAL)
1792 switch (DECL_FUNCTION_CODE (t))
1794 /* We cannot inline functions that take a variable number of
1795 arguments. */
1796 case BUILT_IN_VA_START:
1797 case BUILT_IN_STDARG_START:
1798 case BUILT_IN_NEXT_ARG:
1799 case BUILT_IN_VA_END:
1800 inline_forbidden_reason
1801 = G_("function %q+F can never be inlined because it "
1802 "uses variable argument lists");
1803 return node;
1805 case BUILT_IN_LONGJMP:
1806 /* We can't inline functions that call __builtin_longjmp at
1807 all. The non-local goto machinery really requires the
1808 destination be in a different function. If we allow the
1809 function calling __builtin_longjmp to be inlined into the
1810 function calling __builtin_setjmp, Things will Go Awry. */
1811 inline_forbidden_reason
1812 = G_("function %q+F can never be inlined because "
1813 "it uses setjmp-longjmp exception handling");
1814 return node;
1816 case BUILT_IN_NONLOCAL_GOTO:
1817 /* Similarly. */
1818 inline_forbidden_reason
1819 = G_("function %q+F can never be inlined because "
1820 "it uses non-local goto");
1821 return node;
1823 case BUILT_IN_RETURN:
1824 case BUILT_IN_APPLY_ARGS:
1825 /* If a __builtin_apply_args caller would be inlined,
1826 it would be saving arguments of the function it has
1827 been inlined into. Similarly __builtin_return would
1828 return from the function the inline has been inlined into. */
1829 inline_forbidden_reason
1830 = G_("function %q+F can never be inlined because "
1831 "it uses __builtin_return or __builtin_apply_args");
1832 return node;
1834 default:
1835 break;
1837 break;
1839 case GOTO_EXPR:
1840 t = TREE_OPERAND (node, 0);
1842 /* We will not inline a function which uses computed goto. The
1843 addresses of its local labels, which may be tucked into
1844 global storage, are of course not constant across
1845 instantiations, which causes unexpected behavior. */
1846 if (TREE_CODE (t) != LABEL_DECL)
1848 inline_forbidden_reason
1849 = G_("function %q+F can never be inlined "
1850 "because it contains a computed goto");
1851 return node;
1853 break;
1855 case LABEL_EXPR:
1856 t = TREE_OPERAND (node, 0);
1857 if (DECL_NONLOCAL (t))
1859 /* We cannot inline a function that receives a non-local goto
1860 because we cannot remap the destination label used in the
1861 function that is performing the non-local goto. */
1862 inline_forbidden_reason
1863 = G_("function %q+F can never be inlined "
1864 "because it receives a non-local goto");
1865 return node;
1867 break;
1869 case RECORD_TYPE:
1870 case UNION_TYPE:
1871 /* We cannot inline a function of the form
1873 void F (int i) { struct S { int ar[i]; } s; }
1875 Attempting to do so produces a catch-22.
1876 If walk_tree examines the TYPE_FIELDS chain of RECORD_TYPE/
1877 UNION_TYPE nodes, then it goes into infinite recursion on a
1878 structure containing a pointer to its own type. If it doesn't,
1879 then the type node for S doesn't get adjusted properly when
1880 F is inlined.
1882 ??? This is likely no longer true, but it's too late in the 4.0
1883 cycle to try to find out. This should be checked for 4.1. */
1884 for (t = TYPE_FIELDS (node); t; t = TREE_CHAIN (t))
1885 if (variably_modified_type_p (TREE_TYPE (t), NULL))
1887 inline_forbidden_reason
1888 = G_("function %q+F can never be inlined "
1889 "because it uses variable sized variables");
1890 return node;
1893 default:
1894 break;
1897 return NULL_TREE;
1900 /* Return subexpression representing possible alloca call, if any. */
1901 static tree
1902 inline_forbidden_p (tree fndecl)
1904 location_t saved_loc = input_location;
1905 block_stmt_iterator bsi;
1906 basic_block bb;
1907 tree ret = NULL_TREE;
1909 FOR_EACH_BB_FN (bb, DECL_STRUCT_FUNCTION (fndecl))
1910 for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
1912 ret = walk_tree_without_duplicates (bsi_stmt_ptr (bsi),
1913 inline_forbidden_p_1, fndecl);
1914 if (ret)
1915 goto egress;
1918 egress:
1919 input_location = saved_loc;
1920 return ret;
1923 /* Returns nonzero if FN is a function that does not have any
1924 fundamental inline blocking properties. */
1926 static bool
1927 inlinable_function_p (tree fn)
1929 bool inlinable = true;
1930 bool do_warning;
1931 tree always_inline;
1933 /* If we've already decided this function shouldn't be inlined,
1934 there's no need to check again. */
1935 if (DECL_UNINLINABLE (fn))
1936 return false;
1938 /* We only warn for functions declared `inline' by the user. */
1939 do_warning = (warn_inline
1940 && DECL_INLINE (fn)
1941 && DECL_DECLARED_INLINE_P (fn)
1942 && !DECL_IN_SYSTEM_HEADER (fn));
1944 always_inline = lookup_attribute ("always_inline", DECL_ATTRIBUTES (fn));
1946 if (flag_really_no_inline
1947 && always_inline == NULL)
1949 if (do_warning)
1950 warning (OPT_Winline, "function %q+F can never be inlined because it "
1951 "is suppressed using -fno-inline", fn);
1952 inlinable = false;
1955 /* Don't auto-inline anything that might not be bound within
1956 this unit of translation. */
1957 else if (!DECL_DECLARED_INLINE_P (fn)
1958 && DECL_REPLACEABLE_P (fn))
1959 inlinable = false;
1961 else if (!function_attribute_inlinable_p (fn))
1963 if (do_warning)
1964 warning (OPT_Winline, "function %q+F can never be inlined because it "
1965 "uses attributes conflicting with inlining", fn);
1966 inlinable = false;
1969 /* If we don't have the function body available, we can't inline it.
1970 However, this should not be recorded since we also get here for
1971 forward declared inline functions. Therefore, return at once. */
1972 if (!DECL_SAVED_TREE (fn))
1973 return false;
1975 /* If we're not inlining at all, then we cannot inline this function. */
1976 else if (!flag_inline_trees)
1977 inlinable = false;
1979 /* Only try to inline functions if DECL_INLINE is set. This should be
1980 true for all functions declared `inline', and for all other functions
1981 as well with -finline-functions.
1983 Don't think of disregarding DECL_INLINE when flag_inline_trees == 2;
1984 it's the front-end that must set DECL_INLINE in this case, because
1985 dwarf2out loses if a function that does not have DECL_INLINE set is
1986 inlined anyway. That is why we have both DECL_INLINE and
1987 DECL_DECLARED_INLINE_P. */
1988 /* FIXME: When flag_inline_trees dies, the check for flag_unit_at_a_time
1989 here should be redundant. */
1990 else if (!DECL_INLINE (fn) && !flag_unit_at_a_time)
1991 inlinable = false;
1993 else if (inline_forbidden_p (fn))
1995 /* See if we should warn about uninlinable functions. Previously,
1996 some of these warnings would be issued while trying to expand
1997 the function inline, but that would cause multiple warnings
1998 about functions that would for example call alloca. But since
1999 this a property of the function, just one warning is enough.
2000 As a bonus we can now give more details about the reason why a
2001 function is not inlinable. */
2002 if (always_inline)
2003 sorry (inline_forbidden_reason, fn);
2004 else if (do_warning)
2005 warning (OPT_Winline, inline_forbidden_reason, fn);
2007 inlinable = false;
2010 /* Squirrel away the result so that we don't have to check again. */
2011 DECL_UNINLINABLE (fn) = !inlinable;
2013 return inlinable;
2016 /* Estimate the cost of a memory move. Use machine dependent
2017 word size and take possible memcpy call into account. */
2020 estimate_move_cost (tree type)
2022 HOST_WIDE_INT size;
2024 size = int_size_in_bytes (type);
2026 if (size < 0 || size > MOVE_MAX_PIECES * MOVE_RATIO)
2027 /* Cost of a memcpy call, 3 arguments and the call. */
2028 return 4;
2029 else
2030 return ((size + MOVE_MAX_PIECES - 1) / MOVE_MAX_PIECES);
2033 /* Arguments for estimate_num_insns_1. */
2035 struct eni_data
2037 /* Used to return the number of insns. */
2038 int count;
2040 /* Weights of various constructs. */
2041 eni_weights *weights;
2044 /* Used by estimate_num_insns. Estimate number of instructions seen
2045 by given statement. */
2047 static tree
2048 estimate_num_insns_1 (tree *tp, int *walk_subtrees, void *data)
2050 struct eni_data *d = data;
2051 tree x = *tp;
2052 unsigned cost;
2054 if (IS_TYPE_OR_DECL_P (x))
2056 *walk_subtrees = 0;
2057 return NULL;
2059 /* Assume that constants and references counts nothing. These should
2060 be majorized by amount of operations among them we count later
2061 and are common target of CSE and similar optimizations. */
2062 else if (CONSTANT_CLASS_P (x) || REFERENCE_CLASS_P (x))
2063 return NULL;
2065 switch (TREE_CODE (x))
2067 /* Containers have no cost. */
2068 case TREE_LIST:
2069 case TREE_VEC:
2070 case BLOCK:
2071 case COMPONENT_REF:
2072 case BIT_FIELD_REF:
2073 case INDIRECT_REF:
2074 case ALIGN_INDIRECT_REF:
2075 case MISALIGNED_INDIRECT_REF:
2076 case ARRAY_REF:
2077 case ARRAY_RANGE_REF:
2078 case OBJ_TYPE_REF:
2079 case EXC_PTR_EXPR: /* ??? */
2080 case FILTER_EXPR: /* ??? */
2081 case COMPOUND_EXPR:
2082 case BIND_EXPR:
2083 case WITH_CLEANUP_EXPR:
2084 case NOP_EXPR:
2085 case CONVERT_EXPR:
2086 case VIEW_CONVERT_EXPR:
2087 case SAVE_EXPR:
2088 case ADDR_EXPR:
2089 case COMPLEX_EXPR:
2090 case RANGE_EXPR:
2091 case CASE_LABEL_EXPR:
2092 case SSA_NAME:
2093 case CATCH_EXPR:
2094 case EH_FILTER_EXPR:
2095 case STATEMENT_LIST:
2096 case ERROR_MARK:
2097 case NON_LVALUE_EXPR:
2098 case FDESC_EXPR:
2099 case VA_ARG_EXPR:
2100 case TRY_CATCH_EXPR:
2101 case TRY_FINALLY_EXPR:
2102 case LABEL_EXPR:
2103 case GOTO_EXPR:
2104 case RETURN_EXPR:
2105 case EXIT_EXPR:
2106 case LOOP_EXPR:
2107 case PHI_NODE:
2108 case WITH_SIZE_EXPR:
2109 case OMP_CLAUSE:
2110 case OMP_RETURN:
2111 case OMP_CONTINUE:
2112 case OMP_SECTIONS_SWITCH:
2113 break;
2115 /* We don't account constants for now. Assume that the cost is amortized
2116 by operations that do use them. We may re-consider this decision once
2117 we are able to optimize the tree before estimating its size and break
2118 out static initializers. */
2119 case IDENTIFIER_NODE:
2120 case INTEGER_CST:
2121 case REAL_CST:
2122 case FIXED_CST:
2123 case COMPLEX_CST:
2124 case VECTOR_CST:
2125 case STRING_CST:
2126 *walk_subtrees = 0;
2127 return NULL;
2129 /* CHANGE_DYNAMIC_TYPE_EXPR explicitly expands to nothing. */
2130 case CHANGE_DYNAMIC_TYPE_EXPR:
2131 *walk_subtrees = 0;
2132 return NULL;
2134 /* Try to estimate the cost of assignments. We have three cases to
2135 deal with:
2136 1) Simple assignments to registers;
2137 2) Stores to things that must live in memory. This includes
2138 "normal" stores to scalars, but also assignments of large
2139 structures, or constructors of big arrays;
2140 3) TARGET_EXPRs.
2142 Let us look at the first two cases, assuming we have "a = b + C":
2143 <GIMPLE_MODIFY_STMT <var_decl "a">
2144 <plus_expr <var_decl "b"> <constant C>>
2145 If "a" is a GIMPLE register, the assignment to it is free on almost
2146 any target, because "a" usually ends up in a real register. Hence
2147 the only cost of this expression comes from the PLUS_EXPR, and we
2148 can ignore the GIMPLE_MODIFY_STMT.
2149 If "a" is not a GIMPLE register, the assignment to "a" will most
2150 likely be a real store, so the cost of the GIMPLE_MODIFY_STMT is the cost
2151 of moving something into "a", which we compute using the function
2152 estimate_move_cost.
2154 The third case deals with TARGET_EXPRs, for which the semantics are
2155 that a temporary is assigned, unless the TARGET_EXPR itself is being
2156 assigned to something else. In the latter case we do not need the
2157 temporary. E.g. in:
2158 <GIMPLE_MODIFY_STMT <var_decl "a"> <target_expr>>, the
2159 GIMPLE_MODIFY_STMT is free. */
2160 case INIT_EXPR:
2161 case GIMPLE_MODIFY_STMT:
2162 /* Is the right and side a TARGET_EXPR? */
2163 if (TREE_CODE (GENERIC_TREE_OPERAND (x, 1)) == TARGET_EXPR)
2164 break;
2165 /* ... fall through ... */
2167 case TARGET_EXPR:
2168 x = GENERIC_TREE_OPERAND (x, 0);
2169 /* Is this an assignments to a register? */
2170 if (is_gimple_reg (x))
2171 break;
2172 /* Otherwise it's a store, so fall through to compute the move cost. */
2174 case CONSTRUCTOR:
2175 d->count += estimate_move_cost (TREE_TYPE (x));
2176 break;
2178 /* Assign cost of 1 to usual operations.
2179 ??? We may consider mapping RTL costs to this. */
2180 case COND_EXPR:
2181 case VEC_COND_EXPR:
2183 case PLUS_EXPR:
2184 case POINTER_PLUS_EXPR:
2185 case MINUS_EXPR:
2186 case MULT_EXPR:
2188 case FIXED_CONVERT_EXPR:
2189 case FIX_TRUNC_EXPR:
2191 case NEGATE_EXPR:
2192 case FLOAT_EXPR:
2193 case MIN_EXPR:
2194 case MAX_EXPR:
2195 case ABS_EXPR:
2197 case LSHIFT_EXPR:
2198 case RSHIFT_EXPR:
2199 case LROTATE_EXPR:
2200 case RROTATE_EXPR:
2201 case VEC_LSHIFT_EXPR:
2202 case VEC_RSHIFT_EXPR:
2204 case BIT_IOR_EXPR:
2205 case BIT_XOR_EXPR:
2206 case BIT_AND_EXPR:
2207 case BIT_NOT_EXPR:
2209 case TRUTH_ANDIF_EXPR:
2210 case TRUTH_ORIF_EXPR:
2211 case TRUTH_AND_EXPR:
2212 case TRUTH_OR_EXPR:
2213 case TRUTH_XOR_EXPR:
2214 case TRUTH_NOT_EXPR:
2216 case LT_EXPR:
2217 case LE_EXPR:
2218 case GT_EXPR:
2219 case GE_EXPR:
2220 case EQ_EXPR:
2221 case NE_EXPR:
2222 case ORDERED_EXPR:
2223 case UNORDERED_EXPR:
2225 case UNLT_EXPR:
2226 case UNLE_EXPR:
2227 case UNGT_EXPR:
2228 case UNGE_EXPR:
2229 case UNEQ_EXPR:
2230 case LTGT_EXPR:
2232 case CONJ_EXPR:
2234 case PREDECREMENT_EXPR:
2235 case PREINCREMENT_EXPR:
2236 case POSTDECREMENT_EXPR:
2237 case POSTINCREMENT_EXPR:
2239 case ASM_EXPR:
2241 case REALIGN_LOAD_EXPR:
2243 case REDUC_MAX_EXPR:
2244 case REDUC_MIN_EXPR:
2245 case REDUC_PLUS_EXPR:
2246 case WIDEN_SUM_EXPR:
2247 case DOT_PROD_EXPR:
2248 case VEC_WIDEN_MULT_HI_EXPR:
2249 case VEC_WIDEN_MULT_LO_EXPR:
2250 case VEC_UNPACK_HI_EXPR:
2251 case VEC_UNPACK_LO_EXPR:
2252 case VEC_UNPACK_FLOAT_HI_EXPR:
2253 case VEC_UNPACK_FLOAT_LO_EXPR:
2254 case VEC_PACK_TRUNC_EXPR:
2255 case VEC_PACK_SAT_EXPR:
2256 case VEC_PACK_FIX_TRUNC_EXPR:
2258 case WIDEN_MULT_EXPR:
2260 case VEC_EXTRACT_EVEN_EXPR:
2261 case VEC_EXTRACT_ODD_EXPR:
2262 case VEC_INTERLEAVE_HIGH_EXPR:
2263 case VEC_INTERLEAVE_LOW_EXPR:
2265 case RESX_EXPR:
2266 d->count += 1;
2267 break;
2269 case SWITCH_EXPR:
2270 /* TODO: Cost of a switch should be derived from the number of
2271 branches. */
2272 d->count += d->weights->switch_cost;
2273 break;
2275 /* Few special cases of expensive operations. This is useful
2276 to avoid inlining on functions having too many of these. */
2277 case TRUNC_DIV_EXPR:
2278 case CEIL_DIV_EXPR:
2279 case FLOOR_DIV_EXPR:
2280 case ROUND_DIV_EXPR:
2281 case EXACT_DIV_EXPR:
2282 case TRUNC_MOD_EXPR:
2283 case CEIL_MOD_EXPR:
2284 case FLOOR_MOD_EXPR:
2285 case ROUND_MOD_EXPR:
2286 case RDIV_EXPR:
2287 d->count += d->weights->div_mod_cost;
2288 break;
2289 case CALL_EXPR:
2291 tree decl = get_callee_fndecl (x);
2293 if (decl && DECL_BUILT_IN_CLASS (decl) == BUILT_IN_MD)
2294 cost = d->weights->target_builtin_call_cost;
2295 else
2296 cost = d->weights->call_cost;
2298 if (decl && DECL_BUILT_IN_CLASS (decl) == BUILT_IN_NORMAL)
2299 switch (DECL_FUNCTION_CODE (decl))
2301 case BUILT_IN_CONSTANT_P:
2302 *walk_subtrees = 0;
2303 return NULL_TREE;
2304 case BUILT_IN_EXPECT:
2305 return NULL_TREE;
2306 /* Prefetch instruction is not expensive. */
2307 case BUILT_IN_PREFETCH:
2308 cost = 1;
2309 break;
2310 default:
2311 break;
2314 /* Our cost must be kept in sync with cgraph_estimate_size_after_inlining
2315 that does use function declaration to figure out the arguments. */
2316 if (!decl)
2318 tree a;
2319 call_expr_arg_iterator iter;
2320 FOR_EACH_CALL_EXPR_ARG (a, iter, x)
2321 d->count += estimate_move_cost (TREE_TYPE (a));
2323 else
2325 tree arg;
2326 for (arg = DECL_ARGUMENTS (decl); arg; arg = TREE_CHAIN (arg))
2327 d->count += estimate_move_cost (TREE_TYPE (arg));
2330 d->count += cost;
2331 break;
2334 case OMP_PARALLEL:
2335 case OMP_FOR:
2336 case OMP_SECTIONS:
2337 case OMP_SINGLE:
2338 case OMP_SECTION:
2339 case OMP_MASTER:
2340 case OMP_ORDERED:
2341 case OMP_CRITICAL:
2342 case OMP_ATOMIC:
2343 /* OpenMP directives are generally very expensive. */
2344 d->count += d->weights->omp_cost;
2345 break;
2347 default:
2348 gcc_unreachable ();
2350 return NULL;
2353 /* Estimate number of instructions that will be created by expanding EXPR.
2354 WEIGHTS contains weights attributed to various constructs. */
2357 estimate_num_insns (tree expr, eni_weights *weights)
2359 struct pointer_set_t *visited_nodes;
2360 basic_block bb;
2361 block_stmt_iterator bsi;
2362 struct function *my_function;
2363 struct eni_data data;
2365 data.count = 0;
2366 data.weights = weights;
2368 /* If we're given an entire function, walk the CFG. */
2369 if (TREE_CODE (expr) == FUNCTION_DECL)
2371 my_function = DECL_STRUCT_FUNCTION (expr);
2372 gcc_assert (my_function && my_function->cfg);
2373 visited_nodes = pointer_set_create ();
2374 FOR_EACH_BB_FN (bb, my_function)
2376 for (bsi = bsi_start (bb);
2377 !bsi_end_p (bsi);
2378 bsi_next (&bsi))
2380 walk_tree (bsi_stmt_ptr (bsi), estimate_num_insns_1,
2381 &data, visited_nodes);
2384 pointer_set_destroy (visited_nodes);
2386 else
2387 walk_tree_without_duplicates (&expr, estimate_num_insns_1, &data);
2389 return data.count;
2392 /* Initializes weights used by estimate_num_insns. */
2394 void
2395 init_inline_once (void)
2397 eni_inlining_weights.call_cost = PARAM_VALUE (PARAM_INLINE_CALL_COST);
2398 eni_inlining_weights.target_builtin_call_cost = 1;
2399 eni_inlining_weights.div_mod_cost = 10;
2400 eni_inlining_weights.switch_cost = 1;
2401 eni_inlining_weights.omp_cost = 40;
2403 eni_size_weights.call_cost = 1;
2404 eni_size_weights.target_builtin_call_cost = 1;
2405 eni_size_weights.div_mod_cost = 1;
2406 eni_size_weights.switch_cost = 10;
2407 eni_size_weights.omp_cost = 40;
2409 /* Estimating time for call is difficult, since we have no idea what the
2410 called function does. In the current uses of eni_time_weights,
2411 underestimating the cost does less harm than overestimating it, so
2412 we choose a rather small value here. */
2413 eni_time_weights.call_cost = 10;
2414 eni_time_weights.target_builtin_call_cost = 10;
2415 eni_time_weights.div_mod_cost = 10;
2416 eni_time_weights.switch_cost = 4;
2417 eni_time_weights.omp_cost = 40;
2420 /* Install new lexical TREE_BLOCK underneath 'current_block'. */
2421 static void
2422 add_lexical_block (tree current_block, tree new_block)
2424 tree *blk_p;
2426 /* Walk to the last sub-block. */
2427 for (blk_p = &BLOCK_SUBBLOCKS (current_block);
2428 *blk_p;
2429 blk_p = &TREE_CHAIN (*blk_p))
2431 *blk_p = new_block;
2432 BLOCK_SUPERCONTEXT (new_block) = current_block;
2435 /* If *TP is a CALL_EXPR, replace it with its inline expansion. */
2437 static bool
2438 expand_call_inline (basic_block bb, tree stmt, tree *tp, void *data)
2440 copy_body_data *id;
2441 tree t;
2442 tree use_retvar;
2443 tree fn;
2444 struct pointer_map_t *st;
2445 tree return_slot;
2446 tree modify_dest;
2447 location_t saved_location;
2448 struct cgraph_edge *cg_edge;
2449 const char *reason;
2450 basic_block return_block;
2451 edge e;
2452 block_stmt_iterator bsi, stmt_bsi;
2453 bool successfully_inlined = FALSE;
2454 bool purge_dead_abnormal_edges;
2455 tree t_step;
2456 tree var;
2458 /* See what we've got. */
2459 id = (copy_body_data *) data;
2460 t = *tp;
2462 /* Set input_location here so we get the right instantiation context
2463 if we call instantiate_decl from inlinable_function_p. */
2464 saved_location = input_location;
2465 if (EXPR_HAS_LOCATION (t))
2466 input_location = EXPR_LOCATION (t);
2468 /* From here on, we're only interested in CALL_EXPRs. */
2469 if (TREE_CODE (t) != CALL_EXPR)
2470 goto egress;
2472 /* First, see if we can figure out what function is being called.
2473 If we cannot, then there is no hope of inlining the function. */
2474 fn = get_callee_fndecl (t);
2475 if (!fn)
2476 goto egress;
2478 /* Turn forward declarations into real ones. */
2479 fn = cgraph_node (fn)->decl;
2481 /* If fn is a declaration of a function in a nested scope that was
2482 globally declared inline, we don't set its DECL_INITIAL.
2483 However, we can't blindly follow DECL_ABSTRACT_ORIGIN because the
2484 C++ front-end uses it for cdtors to refer to their internal
2485 declarations, that are not real functions. Fortunately those
2486 don't have trees to be saved, so we can tell by checking their
2487 DECL_SAVED_TREE. */
2488 if (! DECL_INITIAL (fn)
2489 && DECL_ABSTRACT_ORIGIN (fn)
2490 && DECL_SAVED_TREE (DECL_ABSTRACT_ORIGIN (fn)))
2491 fn = DECL_ABSTRACT_ORIGIN (fn);
2493 /* Objective C and fortran still calls tree_rest_of_compilation directly.
2494 Kill this check once this is fixed. */
2495 if (!id->dst_node->analyzed)
2496 goto egress;
2498 cg_edge = cgraph_edge (id->dst_node, stmt);
2500 /* Constant propagation on argument done during previous inlining
2501 may create new direct call. Produce an edge for it. */
2502 if (!cg_edge)
2504 struct cgraph_node *dest = cgraph_node (fn);
2506 /* We have missing edge in the callgraph. This can happen in one case
2507 where previous inlining turned indirect call into direct call by
2508 constant propagating arguments. In all other cases we hit a bug
2509 (incorrect node sharing is most common reason for missing edges. */
2510 gcc_assert (dest->needed || !flag_unit_at_a_time);
2511 cgraph_create_edge (id->dst_node, dest, stmt,
2512 bb->count, CGRAPH_FREQ_BASE,
2513 bb->loop_depth)->inline_failed
2514 = N_("originally indirect function call not considered for inlining");
2515 if (dump_file)
2517 fprintf (dump_file, "Created new direct edge to %s",
2518 cgraph_node_name (dest));
2520 goto egress;
2523 /* Don't try to inline functions that are not well-suited to
2524 inlining. */
2525 if (!cgraph_inline_p (cg_edge, &reason))
2527 if (lookup_attribute ("always_inline", DECL_ATTRIBUTES (fn))
2528 /* Avoid warnings during early inline pass. */
2529 && (!flag_unit_at_a_time || cgraph_global_info_ready))
2531 sorry ("inlining failed in call to %q+F: %s", fn, reason);
2532 sorry ("called from here");
2534 else if (warn_inline && DECL_DECLARED_INLINE_P (fn)
2535 && !DECL_IN_SYSTEM_HEADER (fn)
2536 && strlen (reason)
2537 && !lookup_attribute ("noinline", DECL_ATTRIBUTES (fn))
2538 /* Avoid warnings during early inline pass. */
2539 && (!flag_unit_at_a_time || cgraph_global_info_ready))
2541 warning (OPT_Winline, "inlining failed in call to %q+F: %s",
2542 fn, reason);
2543 warning (OPT_Winline, "called from here");
2545 goto egress;
2547 fn = cg_edge->callee->decl;
2549 #ifdef ENABLE_CHECKING
2550 if (cg_edge->callee->decl != id->dst_node->decl)
2551 verify_cgraph_node (cg_edge->callee);
2552 #endif
2554 /* We will be inlining this callee. */
2555 id->eh_region = lookup_stmt_eh_region (stmt);
2557 /* Split the block holding the CALL_EXPR. */
2558 e = split_block (bb, stmt);
2559 bb = e->src;
2560 return_block = e->dest;
2561 remove_edge (e);
2563 /* split_block splits after the statement; work around this by
2564 moving the call into the second block manually. Not pretty,
2565 but seems easier than doing the CFG manipulation by hand
2566 when the CALL_EXPR is in the last statement of BB. */
2567 stmt_bsi = bsi_last (bb);
2568 bsi_remove (&stmt_bsi, false);
2570 /* If the CALL_EXPR was in the last statement of BB, it may have
2571 been the source of abnormal edges. In this case, schedule
2572 the removal of dead abnormal edges. */
2573 bsi = bsi_start (return_block);
2574 if (bsi_end_p (bsi))
2576 bsi_insert_after (&bsi, stmt, BSI_NEW_STMT);
2577 purge_dead_abnormal_edges = true;
2579 else
2581 bsi_insert_before (&bsi, stmt, BSI_NEW_STMT);
2582 purge_dead_abnormal_edges = false;
2585 stmt_bsi = bsi_start (return_block);
2587 /* Build a block containing code to initialize the arguments, the
2588 actual inline expansion of the body, and a label for the return
2589 statements within the function to jump to. The type of the
2590 statement expression is the return type of the function call. */
2591 id->block = make_node (BLOCK);
2592 BLOCK_ABSTRACT_ORIGIN (id->block) = fn;
2593 BLOCK_SOURCE_LOCATION (id->block) = input_location;
2594 add_lexical_block (TREE_BLOCK (stmt), id->block);
2596 /* Local declarations will be replaced by their equivalents in this
2597 map. */
2598 st = id->decl_map;
2599 id->decl_map = pointer_map_create ();
2601 /* Record the function we are about to inline. */
2602 id->src_fn = fn;
2603 id->src_node = cg_edge->callee;
2604 id->src_cfun = DECL_STRUCT_FUNCTION (fn);
2605 id->call_expr = t;
2607 initialize_inlined_parameters (id, t, fn, bb);
2609 if (DECL_INITIAL (fn))
2610 add_lexical_block (id->block, remap_blocks (DECL_INITIAL (fn), id));
2612 /* Return statements in the function body will be replaced by jumps
2613 to the RET_LABEL. */
2615 gcc_assert (DECL_INITIAL (fn));
2616 gcc_assert (TREE_CODE (DECL_INITIAL (fn)) == BLOCK);
2618 /* Find the lhs to which the result of this call is assigned. */
2619 return_slot = NULL;
2620 if (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT)
2622 modify_dest = GIMPLE_STMT_OPERAND (stmt, 0);
2624 /* The function which we are inlining might not return a value,
2625 in which case we should issue a warning that the function
2626 does not return a value. In that case the optimizers will
2627 see that the variable to which the value is assigned was not
2628 initialized. We do not want to issue a warning about that
2629 uninitialized variable. */
2630 if (DECL_P (modify_dest))
2631 TREE_NO_WARNING (modify_dest) = 1;
2632 if (CALL_EXPR_RETURN_SLOT_OPT (t))
2634 return_slot = modify_dest;
2635 modify_dest = NULL;
2638 else
2639 modify_dest = NULL;
2641 /* Declare the return variable for the function. */
2642 declare_return_variable (id, return_slot,
2643 modify_dest, &use_retvar);
2645 /* This is it. Duplicate the callee body. Assume callee is
2646 pre-gimplified. Note that we must not alter the caller
2647 function in any way before this point, as this CALL_EXPR may be
2648 a self-referential call; if we're calling ourselves, we need to
2649 duplicate our body before altering anything. */
2650 copy_body (id, bb->count, bb->frequency, bb, return_block);
2652 /* Add local vars in this inlined callee to caller. */
2653 t_step = id->src_cfun->unexpanded_var_list;
2654 for (; t_step; t_step = TREE_CHAIN (t_step))
2656 var = TREE_VALUE (t_step);
2657 if (TREE_STATIC (var) && !TREE_ASM_WRITTEN (var))
2658 cfun->unexpanded_var_list = tree_cons (NULL_TREE, var,
2659 cfun->unexpanded_var_list);
2660 else
2661 cfun->unexpanded_var_list = tree_cons (NULL_TREE, remap_decl (var, id),
2662 cfun->unexpanded_var_list);
2665 /* Clean up. */
2666 pointer_map_destroy (id->decl_map);
2667 id->decl_map = st;
2669 /* If the inlined function returns a result that we care about,
2670 clobber the CALL_EXPR with a reference to the return variable. */
2671 if (use_retvar && (TREE_CODE (bsi_stmt (stmt_bsi)) != CALL_EXPR))
2673 *tp = use_retvar;
2674 if (gimple_in_ssa_p (cfun))
2676 update_stmt (stmt);
2677 mark_symbols_for_renaming (stmt);
2679 maybe_clean_or_replace_eh_stmt (stmt, stmt);
2681 else
2682 /* We're modifying a TSI owned by gimple_expand_calls_inline();
2683 tsi_delink() will leave the iterator in a sane state. */
2685 /* Handle case of inlining function that miss return statement so
2686 return value becomes undefined. */
2687 if (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT
2688 && TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) == SSA_NAME)
2690 tree name = TREE_OPERAND (stmt, 0);
2691 tree var = SSA_NAME_VAR (TREE_OPERAND (stmt, 0));
2692 tree def = gimple_default_def (cfun, var);
2694 /* If the variable is used undefined, make this name undefined via
2695 move. */
2696 if (def)
2698 TREE_OPERAND (stmt, 1) = def;
2699 update_stmt (stmt);
2701 /* Otherwise make this variable undefined. */
2702 else
2704 bsi_remove (&stmt_bsi, true);
2705 set_default_def (var, name);
2706 SSA_NAME_DEF_STMT (name) = build_empty_stmt ();
2709 else
2710 bsi_remove (&stmt_bsi, true);
2713 if (purge_dead_abnormal_edges)
2714 tree_purge_dead_abnormal_call_edges (return_block);
2716 /* If the value of the new expression is ignored, that's OK. We
2717 don't warn about this for CALL_EXPRs, so we shouldn't warn about
2718 the equivalent inlined version either. */
2719 TREE_USED (*tp) = 1;
2721 /* Output the inlining info for this abstract function, since it has been
2722 inlined. If we don't do this now, we can lose the information about the
2723 variables in the function when the blocks get blown away as soon as we
2724 remove the cgraph node. */
2725 (*debug_hooks->outlining_inline_function) (cg_edge->callee->decl);
2727 /* Update callgraph if needed. */
2728 cgraph_remove_node (cg_edge->callee);
2730 id->block = NULL_TREE;
2731 successfully_inlined = TRUE;
2733 egress:
2734 input_location = saved_location;
2735 return successfully_inlined;
2738 /* Expand call statements reachable from STMT_P.
2739 We can only have CALL_EXPRs as the "toplevel" tree code or nested
2740 in a GIMPLE_MODIFY_STMT. See tree-gimple.c:get_call_expr_in(). We can
2741 unfortunately not use that function here because we need a pointer
2742 to the CALL_EXPR, not the tree itself. */
2744 static bool
2745 gimple_expand_calls_inline (basic_block bb, copy_body_data *id)
2747 block_stmt_iterator bsi;
2749 /* Register specific tree functions. */
2750 tree_register_cfg_hooks ();
2751 for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
2753 tree *expr_p = bsi_stmt_ptr (bsi);
2754 tree stmt = *expr_p;
2756 if (TREE_CODE (*expr_p) == GIMPLE_MODIFY_STMT)
2757 expr_p = &GIMPLE_STMT_OPERAND (*expr_p, 1);
2758 if (TREE_CODE (*expr_p) == WITH_SIZE_EXPR)
2759 expr_p = &TREE_OPERAND (*expr_p, 0);
2760 if (TREE_CODE (*expr_p) == CALL_EXPR)
2761 if (expand_call_inline (bb, stmt, expr_p, id))
2762 return true;
2764 return false;
2767 /* Walk all basic blocks created after FIRST and try to fold every statement
2768 in the STATEMENTS pointer set. */
2769 static void
2770 fold_marked_statements (int first, struct pointer_set_t *statements)
2772 for (;first < n_basic_blocks;first++)
2773 if (BASIC_BLOCK (first))
2775 block_stmt_iterator bsi;
2776 for (bsi = bsi_start (BASIC_BLOCK (first));
2777 !bsi_end_p (bsi); bsi_next (&bsi))
2778 if (pointer_set_contains (statements, bsi_stmt (bsi)))
2780 tree old_stmt = bsi_stmt (bsi);
2781 if (fold_stmt (bsi_stmt_ptr (bsi)))
2783 update_stmt (bsi_stmt (bsi));
2784 if (maybe_clean_or_replace_eh_stmt (old_stmt, bsi_stmt (bsi)))
2785 tree_purge_dead_eh_edges (BASIC_BLOCK (first));
2791 /* Return true if BB has at least one abnormal outgoing edge. */
2793 static inline bool
2794 has_abnormal_outgoing_edge_p (basic_block bb)
2796 edge e;
2797 edge_iterator ei;
2799 FOR_EACH_EDGE (e, ei, bb->succs)
2800 if (e->flags & EDGE_ABNORMAL)
2801 return true;
2803 return false;
2806 /* When a block from the inlined function contains a call with side-effects
2807 in the middle gets inlined in a function with non-locals labels, the call
2808 becomes a potential non-local goto so we need to add appropriate edge. */
2810 static void
2811 make_nonlocal_label_edges (void)
2813 block_stmt_iterator bsi;
2814 basic_block bb;
2816 FOR_EACH_BB (bb)
2818 for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
2820 tree stmt = bsi_stmt (bsi);
2821 if (tree_can_make_abnormal_goto (stmt))
2823 if (stmt == bsi_stmt (bsi_last (bb)))
2825 if (!has_abnormal_outgoing_edge_p (bb))
2826 make_abnormal_goto_edges (bb, true);
2828 else
2830 edge e = split_block (bb, stmt);
2831 bb = e->src;
2832 make_abnormal_goto_edges (bb, true);
2834 break;
2837 /* Update PHIs on nonlocal goto receivers we (possibly)
2838 just created new edges into. */
2839 if (TREE_CODE (stmt) == LABEL_EXPR
2840 && gimple_in_ssa_p (cfun))
2842 tree target = LABEL_EXPR_LABEL (stmt);
2843 if (DECL_NONLOCAL (target))
2845 tree phi;
2847 for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
2849 gcc_assert (SSA_NAME_OCCURS_IN_ABNORMAL_PHI
2850 (PHI_RESULT (phi)));
2851 mark_sym_for_renaming
2852 (SSA_NAME_VAR (PHI_RESULT (phi)));
2860 /* Expand calls to inline functions in the body of FN. */
2862 unsigned int
2863 optimize_inline_calls (tree fn)
2865 copy_body_data id;
2866 tree prev_fn;
2867 basic_block bb;
2868 int last = n_basic_blocks;
2869 /* There is no point in performing inlining if errors have already
2870 occurred -- and we might crash if we try to inline invalid
2871 code. */
2872 if (errorcount || sorrycount)
2873 return 0;
2875 /* Clear out ID. */
2876 memset (&id, 0, sizeof (id));
2878 id.src_node = id.dst_node = cgraph_node (fn);
2879 id.dst_fn = fn;
2880 /* Or any functions that aren't finished yet. */
2881 prev_fn = NULL_TREE;
2882 if (current_function_decl)
2884 id.dst_fn = current_function_decl;
2885 prev_fn = current_function_decl;
2888 id.copy_decl = copy_decl_maybe_to_var;
2889 id.transform_call_graph_edges = CB_CGE_DUPLICATE;
2890 id.transform_new_cfg = false;
2891 id.transform_return_to_modify = true;
2892 id.transform_lang_insert_block = false;
2893 id.statements_to_fold = pointer_set_create ();
2895 push_gimplify_context ();
2897 /* We make no attempts to keep dominance info up-to-date. */
2898 free_dominance_info (CDI_DOMINATORS);
2899 free_dominance_info (CDI_POST_DOMINATORS);
2901 /* Reach the trees by walking over the CFG, and note the
2902 enclosing basic-blocks in the call edges. */
2903 /* We walk the blocks going forward, because inlined function bodies
2904 will split id->current_basic_block, and the new blocks will
2905 follow it; we'll trudge through them, processing their CALL_EXPRs
2906 along the way. */
2907 FOR_EACH_BB (bb)
2908 gimple_expand_calls_inline (bb, &id);
2910 pop_gimplify_context (NULL);
2912 #ifdef ENABLE_CHECKING
2914 struct cgraph_edge *e;
2916 verify_cgraph_node (id.dst_node);
2918 /* Double check that we inlined everything we are supposed to inline. */
2919 for (e = id.dst_node->callees; e; e = e->next_callee)
2920 gcc_assert (e->inline_failed);
2922 #endif
2924 /* Fold the statements before compacting/renumbering the basic blocks. */
2925 fold_marked_statements (last, id.statements_to_fold);
2926 pointer_set_destroy (id.statements_to_fold);
2928 /* Renumber the (code) basic_blocks consecutively. */
2929 compact_blocks ();
2930 /* Renumber the lexical scoping (non-code) blocks consecutively. */
2931 number_blocks (fn);
2933 /* We are not going to maintain the cgraph edges up to date.
2934 Kill it so it won't confuse us. */
2935 cgraph_node_remove_callees (id.dst_node);
2937 fold_cond_expr_cond ();
2938 if (current_function_has_nonlocal_label)
2939 make_nonlocal_label_edges ();
2940 /* It would be nice to check SSA/CFG/statement consistency here, but it is
2941 not possible yet - the IPA passes might make various functions to not
2942 throw and they don't care to proactively update local EH info. This is
2943 done later in fixup_cfg pass that also execute the verification. */
2944 return (TODO_update_ssa | TODO_cleanup_cfg
2945 | (gimple_in_ssa_p (cfun) ? TODO_remove_unused_locals : 0)
2946 | (profile_status != PROFILE_ABSENT ? TODO_rebuild_frequencies : 0));
2949 /* FN is a function that has a complete body, and CLONE is a function whose
2950 body is to be set to a copy of FN, mapping argument declarations according
2951 to the ARG_MAP splay_tree. */
2953 void
2954 clone_body (tree clone, tree fn, void *arg_map)
2956 copy_body_data id;
2958 /* Clone the body, as if we were making an inline call. But, remap the
2959 parameters in the callee to the parameters of caller. */
2960 memset (&id, 0, sizeof (id));
2961 id.src_fn = fn;
2962 id.dst_fn = clone;
2963 id.src_cfun = DECL_STRUCT_FUNCTION (fn);
2964 id.decl_map = (struct pointer_map_t *)arg_map;
2966 id.copy_decl = copy_decl_no_change;
2967 id.transform_call_graph_edges = CB_CGE_DUPLICATE;
2968 id.transform_new_cfg = true;
2969 id.transform_return_to_modify = false;
2970 id.transform_lang_insert_block = true;
2972 /* We're not inside any EH region. */
2973 id.eh_region = -1;
2975 /* Actually copy the body. */
2976 append_to_statement_list_force (copy_generic_body (&id), &DECL_SAVED_TREE (clone));
2979 /* Passed to walk_tree. Copies the node pointed to, if appropriate. */
2981 tree
2982 copy_tree_r (tree *tp, int *walk_subtrees, void *data ATTRIBUTE_UNUSED)
2984 enum tree_code code = TREE_CODE (*tp);
2985 enum tree_code_class cl = TREE_CODE_CLASS (code);
2987 /* We make copies of most nodes. */
2988 if (IS_EXPR_CODE_CLASS (cl)
2989 || IS_GIMPLE_STMT_CODE_CLASS (cl)
2990 || code == TREE_LIST
2991 || code == TREE_VEC
2992 || code == TYPE_DECL
2993 || code == OMP_CLAUSE)
2995 /* Because the chain gets clobbered when we make a copy, we save it
2996 here. */
2997 tree chain = NULL_TREE, new;
2999 if (!GIMPLE_TUPLE_P (*tp))
3000 chain = TREE_CHAIN (*tp);
3002 /* Copy the node. */
3003 new = copy_node (*tp);
3005 /* Propagate mudflap marked-ness. */
3006 if (flag_mudflap && mf_marked_p (*tp))
3007 mf_mark (new);
3009 *tp = new;
3011 /* Now, restore the chain, if appropriate. That will cause
3012 walk_tree to walk into the chain as well. */
3013 if (code == PARM_DECL
3014 || code == TREE_LIST
3015 || code == OMP_CLAUSE)
3016 TREE_CHAIN (*tp) = chain;
3018 /* For now, we don't update BLOCKs when we make copies. So, we
3019 have to nullify all BIND_EXPRs. */
3020 if (TREE_CODE (*tp) == BIND_EXPR)
3021 BIND_EXPR_BLOCK (*tp) = NULL_TREE;
3023 else if (code == CONSTRUCTOR)
3025 /* CONSTRUCTOR nodes need special handling because
3026 we need to duplicate the vector of elements. */
3027 tree new;
3029 new = copy_node (*tp);
3031 /* Propagate mudflap marked-ness. */
3032 if (flag_mudflap && mf_marked_p (*tp))
3033 mf_mark (new);
3035 CONSTRUCTOR_ELTS (new) = VEC_copy (constructor_elt, gc,
3036 CONSTRUCTOR_ELTS (*tp));
3037 *tp = new;
3039 else if (TREE_CODE_CLASS (code) == tcc_type)
3040 *walk_subtrees = 0;
3041 else if (TREE_CODE_CLASS (code) == tcc_declaration)
3042 *walk_subtrees = 0;
3043 else if (TREE_CODE_CLASS (code) == tcc_constant)
3044 *walk_subtrees = 0;
3045 else
3046 gcc_assert (code != STATEMENT_LIST);
3047 return NULL_TREE;
3050 /* The SAVE_EXPR pointed to by TP is being copied. If ST contains
3051 information indicating to what new SAVE_EXPR this one should be mapped,
3052 use that one. Otherwise, create a new node and enter it in ST. FN is
3053 the function into which the copy will be placed. */
3055 static void
3056 remap_save_expr (tree *tp, void *st_, int *walk_subtrees)
3058 struct pointer_map_t *st = (struct pointer_map_t *) st_;
3059 tree *n;
3060 tree t;
3062 /* See if we already encountered this SAVE_EXPR. */
3063 n = (tree *) pointer_map_contains (st, *tp);
3065 /* If we didn't already remap this SAVE_EXPR, do so now. */
3066 if (!n)
3068 t = copy_node (*tp);
3070 /* Remember this SAVE_EXPR. */
3071 *pointer_map_insert (st, *tp) = t;
3072 /* Make sure we don't remap an already-remapped SAVE_EXPR. */
3073 *pointer_map_insert (st, t) = t;
3075 else
3077 /* We've already walked into this SAVE_EXPR; don't do it again. */
3078 *walk_subtrees = 0;
3079 t = *n;
3082 /* Replace this SAVE_EXPR with the copy. */
3083 *tp = t;
3086 /* Called via walk_tree. If *TP points to a DECL_STMT for a local label,
3087 copies the declaration and enters it in the splay_tree in DATA (which is
3088 really an `copy_body_data *'). */
3090 static tree
3091 mark_local_for_remap_r (tree *tp, int *walk_subtrees ATTRIBUTE_UNUSED,
3092 void *data)
3094 copy_body_data *id = (copy_body_data *) data;
3096 /* Don't walk into types. */
3097 if (TYPE_P (*tp))
3098 *walk_subtrees = 0;
3100 else if (TREE_CODE (*tp) == LABEL_EXPR)
3102 tree decl = TREE_OPERAND (*tp, 0);
3104 /* Copy the decl and remember the copy. */
3105 insert_decl_map (id, decl, id->copy_decl (decl, id));
3108 return NULL_TREE;
3111 /* Perform any modifications to EXPR required when it is unsaved. Does
3112 not recurse into EXPR's subtrees. */
3114 static void
3115 unsave_expr_1 (tree expr)
3117 switch (TREE_CODE (expr))
3119 case TARGET_EXPR:
3120 /* Don't mess with a TARGET_EXPR that hasn't been expanded.
3121 It's OK for this to happen if it was part of a subtree that
3122 isn't immediately expanded, such as operand 2 of another
3123 TARGET_EXPR. */
3124 if (TREE_OPERAND (expr, 1))
3125 break;
3127 TREE_OPERAND (expr, 1) = TREE_OPERAND (expr, 3);
3128 TREE_OPERAND (expr, 3) = NULL_TREE;
3129 break;
3131 default:
3132 break;
3136 /* Called via walk_tree when an expression is unsaved. Using the
3137 splay_tree pointed to by ST (which is really a `splay_tree'),
3138 remaps all local declarations to appropriate replacements. */
3140 static tree
3141 unsave_r (tree *tp, int *walk_subtrees, void *data)
3143 copy_body_data *id = (copy_body_data *) data;
3144 struct pointer_map_t *st = id->decl_map;
3145 tree *n;
3147 /* Only a local declaration (variable or label). */
3148 if ((TREE_CODE (*tp) == VAR_DECL && !TREE_STATIC (*tp))
3149 || TREE_CODE (*tp) == LABEL_DECL)
3151 /* Lookup the declaration. */
3152 n = (tree *) pointer_map_contains (st, *tp);
3154 /* If it's there, remap it. */
3155 if (n)
3156 *tp = *n;
3159 else if (TREE_CODE (*tp) == STATEMENT_LIST)
3160 copy_statement_list (tp);
3161 else if (TREE_CODE (*tp) == BIND_EXPR)
3162 copy_bind_expr (tp, walk_subtrees, id);
3163 else if (TREE_CODE (*tp) == SAVE_EXPR)
3164 remap_save_expr (tp, st, walk_subtrees);
3165 else
3167 copy_tree_r (tp, walk_subtrees, NULL);
3169 /* Do whatever unsaving is required. */
3170 unsave_expr_1 (*tp);
3173 /* Keep iterating. */
3174 return NULL_TREE;
3177 /* Copies everything in EXPR and replaces variables, labels
3178 and SAVE_EXPRs local to EXPR. */
3180 tree
3181 unsave_expr_now (tree expr)
3183 copy_body_data id;
3185 /* There's nothing to do for NULL_TREE. */
3186 if (expr == 0)
3187 return expr;
3189 /* Set up ID. */
3190 memset (&id, 0, sizeof (id));
3191 id.src_fn = current_function_decl;
3192 id.dst_fn = current_function_decl;
3193 id.decl_map = pointer_map_create ();
3195 id.copy_decl = copy_decl_no_change;
3196 id.transform_call_graph_edges = CB_CGE_DUPLICATE;
3197 id.transform_new_cfg = false;
3198 id.transform_return_to_modify = false;
3199 id.transform_lang_insert_block = false;
3201 /* Walk the tree once to find local labels. */
3202 walk_tree_without_duplicates (&expr, mark_local_for_remap_r, &id);
3204 /* Walk the tree again, copying, remapping, and unsaving. */
3205 walk_tree (&expr, unsave_r, &id, NULL);
3207 /* Clean up. */
3208 pointer_map_destroy (id.decl_map);
3210 return expr;
3213 /* Allow someone to determine if SEARCH is a child of TOP from gdb. */
3215 static tree
3216 debug_find_tree_1 (tree *tp, int *walk_subtrees ATTRIBUTE_UNUSED, void *data)
3218 if (*tp == data)
3219 return (tree) data;
3220 else
3221 return NULL;
3224 bool
3225 debug_find_tree (tree top, tree search)
3227 return walk_tree_without_duplicates (&top, debug_find_tree_1, search) != 0;
3231 /* Declare the variables created by the inliner. Add all the variables in
3232 VARS to BIND_EXPR. */
3234 static void
3235 declare_inline_vars (tree block, tree vars)
3237 tree t;
3238 for (t = vars; t; t = TREE_CHAIN (t))
3240 DECL_SEEN_IN_BIND_EXPR_P (t) = 1;
3241 gcc_assert (!TREE_STATIC (t) && !TREE_ASM_WRITTEN (t));
3242 cfun->unexpanded_var_list =
3243 tree_cons (NULL_TREE, t,
3244 cfun->unexpanded_var_list);
3247 if (block)
3248 BLOCK_VARS (block) = chainon (BLOCK_VARS (block), vars);
3252 /* Copy NODE (which must be a DECL). The DECL originally was in the FROM_FN,
3253 but now it will be in the TO_FN. PARM_TO_VAR means enable PARM_DECL to
3254 VAR_DECL translation. */
3256 static tree
3257 copy_decl_for_dup_finish (copy_body_data *id, tree decl, tree copy)
3259 /* Don't generate debug information for the copy if we wouldn't have
3260 generated it for the copy either. */
3261 DECL_ARTIFICIAL (copy) = DECL_ARTIFICIAL (decl);
3262 DECL_IGNORED_P (copy) = DECL_IGNORED_P (decl);
3264 /* Set the DECL_ABSTRACT_ORIGIN so the debugging routines know what
3265 declaration inspired this copy. */
3266 DECL_ABSTRACT_ORIGIN (copy) = DECL_ORIGIN (decl);
3268 /* The new variable/label has no RTL, yet. */
3269 if (CODE_CONTAINS_STRUCT (TREE_CODE (copy), TS_DECL_WRTL)
3270 && !TREE_STATIC (copy) && !DECL_EXTERNAL (copy))
3271 SET_DECL_RTL (copy, NULL_RTX);
3273 /* These args would always appear unused, if not for this. */
3274 TREE_USED (copy) = 1;
3276 /* Set the context for the new declaration. */
3277 if (!DECL_CONTEXT (decl))
3278 /* Globals stay global. */
3280 else if (DECL_CONTEXT (decl) != id->src_fn)
3281 /* Things that weren't in the scope of the function we're inlining
3282 from aren't in the scope we're inlining to, either. */
3284 else if (TREE_STATIC (decl))
3285 /* Function-scoped static variables should stay in the original
3286 function. */
3288 else
3289 /* Ordinary automatic local variables are now in the scope of the
3290 new function. */
3291 DECL_CONTEXT (copy) = id->dst_fn;
3293 return copy;
3296 static tree
3297 copy_decl_to_var (tree decl, copy_body_data *id)
3299 tree copy, type;
3301 gcc_assert (TREE_CODE (decl) == PARM_DECL
3302 || TREE_CODE (decl) == RESULT_DECL);
3304 type = TREE_TYPE (decl);
3306 copy = build_decl (VAR_DECL, DECL_NAME (decl), type);
3307 TREE_ADDRESSABLE (copy) = TREE_ADDRESSABLE (decl);
3308 TREE_READONLY (copy) = TREE_READONLY (decl);
3309 TREE_THIS_VOLATILE (copy) = TREE_THIS_VOLATILE (decl);
3310 DECL_GIMPLE_REG_P (copy) = DECL_GIMPLE_REG_P (decl);
3311 DECL_NO_TBAA_P (copy) = DECL_NO_TBAA_P (decl);
3313 return copy_decl_for_dup_finish (id, decl, copy);
3316 /* Like copy_decl_to_var, but create a return slot object instead of a
3317 pointer variable for return by invisible reference. */
3319 static tree
3320 copy_result_decl_to_var (tree decl, copy_body_data *id)
3322 tree copy, type;
3324 gcc_assert (TREE_CODE (decl) == PARM_DECL
3325 || TREE_CODE (decl) == RESULT_DECL);
3327 type = TREE_TYPE (decl);
3328 if (DECL_BY_REFERENCE (decl))
3329 type = TREE_TYPE (type);
3331 copy = build_decl (VAR_DECL, DECL_NAME (decl), type);
3332 TREE_READONLY (copy) = TREE_READONLY (decl);
3333 TREE_THIS_VOLATILE (copy) = TREE_THIS_VOLATILE (decl);
3334 if (!DECL_BY_REFERENCE (decl))
3336 TREE_ADDRESSABLE (copy) = TREE_ADDRESSABLE (decl);
3337 DECL_GIMPLE_REG_P (copy) = DECL_GIMPLE_REG_P (decl);
3338 DECL_NO_TBAA_P (copy) = DECL_NO_TBAA_P (decl);
3341 return copy_decl_for_dup_finish (id, decl, copy);
3345 static tree
3346 copy_decl_no_change (tree decl, copy_body_data *id)
3348 tree copy;
3350 copy = copy_node (decl);
3352 /* The COPY is not abstract; it will be generated in DST_FN. */
3353 DECL_ABSTRACT (copy) = 0;
3354 lang_hooks.dup_lang_specific_decl (copy);
3356 /* TREE_ADDRESSABLE isn't used to indicate that a label's address has
3357 been taken; it's for internal bookkeeping in expand_goto_internal. */
3358 if (TREE_CODE (copy) == LABEL_DECL)
3360 TREE_ADDRESSABLE (copy) = 0;
3361 LABEL_DECL_UID (copy) = -1;
3364 return copy_decl_for_dup_finish (id, decl, copy);
3367 static tree
3368 copy_decl_maybe_to_var (tree decl, copy_body_data *id)
3370 if (TREE_CODE (decl) == PARM_DECL || TREE_CODE (decl) == RESULT_DECL)
3371 return copy_decl_to_var (decl, id);
3372 else
3373 return copy_decl_no_change (decl, id);
3376 /* Return a copy of the function's argument tree. */
3377 static tree
3378 copy_arguments_for_versioning (tree orig_parm, copy_body_data * id)
3380 tree *arg_copy, *parg;
3382 arg_copy = &orig_parm;
3383 for (parg = arg_copy; *parg; parg = &TREE_CHAIN (*parg))
3385 tree new = remap_decl (*parg, id);
3386 lang_hooks.dup_lang_specific_decl (new);
3387 TREE_CHAIN (new) = TREE_CHAIN (*parg);
3388 *parg = new;
3390 return orig_parm;
3393 /* Return a copy of the function's static chain. */
3394 static tree
3395 copy_static_chain (tree static_chain, copy_body_data * id)
3397 tree *chain_copy, *pvar;
3399 chain_copy = &static_chain;
3400 for (pvar = chain_copy; *pvar; pvar = &TREE_CHAIN (*pvar))
3402 tree new = remap_decl (*pvar, id);
3403 lang_hooks.dup_lang_specific_decl (new);
3404 TREE_CHAIN (new) = TREE_CHAIN (*pvar);
3405 *pvar = new;
3407 return static_chain;
3410 /* Return true if the function is allowed to be versioned.
3411 This is a guard for the versioning functionality. */
3412 bool
3413 tree_versionable_function_p (tree fndecl)
3415 if (fndecl == NULL_TREE)
3416 return false;
3417 /* ??? There are cases where a function is
3418 uninlinable but can be versioned. */
3419 if (!tree_inlinable_function_p (fndecl))
3420 return false;
3422 return true;
3425 /* Create a copy of a function's tree.
3426 OLD_DECL and NEW_DECL are FUNCTION_DECL tree nodes
3427 of the original function and the new copied function
3428 respectively. In case we want to replace a DECL
3429 tree with another tree while duplicating the function's
3430 body, TREE_MAP represents the mapping between these
3431 trees. If UPDATE_CLONES is set, the call_stmt fields
3432 of edges of clones of the function will be updated. */
3433 void
3434 tree_function_versioning (tree old_decl, tree new_decl, varray_type tree_map,
3435 bool update_clones)
3437 struct cgraph_node *old_version_node;
3438 struct cgraph_node *new_version_node;
3439 copy_body_data id;
3440 tree p;
3441 unsigned i;
3442 struct ipa_replace_map *replace_info;
3443 basic_block old_entry_block;
3444 tree t_step;
3445 tree old_current_function_decl = current_function_decl;
3447 gcc_assert (TREE_CODE (old_decl) == FUNCTION_DECL
3448 && TREE_CODE (new_decl) == FUNCTION_DECL);
3449 DECL_POSSIBLY_INLINED (old_decl) = 1;
3451 old_version_node = cgraph_node (old_decl);
3452 new_version_node = cgraph_node (new_decl);
3454 DECL_ARTIFICIAL (new_decl) = 1;
3455 DECL_ABSTRACT_ORIGIN (new_decl) = DECL_ORIGIN (old_decl);
3457 /* Prepare the data structures for the tree copy. */
3458 memset (&id, 0, sizeof (id));
3460 /* Generate a new name for the new version. */
3461 if (!update_clones)
3463 DECL_NAME (new_decl) = create_tmp_var_name (NULL);
3464 SET_DECL_ASSEMBLER_NAME (new_decl, DECL_NAME (new_decl));
3465 SET_DECL_RTL (new_decl, NULL_RTX);
3466 id.statements_to_fold = pointer_set_create ();
3469 id.decl_map = pointer_map_create ();
3470 id.src_fn = old_decl;
3471 id.dst_fn = new_decl;
3472 id.src_node = old_version_node;
3473 id.dst_node = new_version_node;
3474 id.src_cfun = DECL_STRUCT_FUNCTION (old_decl);
3476 id.copy_decl = copy_decl_no_change;
3477 id.transform_call_graph_edges
3478 = update_clones ? CB_CGE_MOVE_CLONES : CB_CGE_MOVE;
3479 id.transform_new_cfg = true;
3480 id.transform_return_to_modify = false;
3481 id.transform_lang_insert_block = false;
3483 current_function_decl = new_decl;
3484 old_entry_block = ENTRY_BLOCK_PTR_FOR_FUNCTION
3485 (DECL_STRUCT_FUNCTION (old_decl));
3486 initialize_cfun (new_decl, old_decl,
3487 old_entry_block->count,
3488 old_entry_block->frequency);
3489 push_cfun (DECL_STRUCT_FUNCTION (new_decl));
3491 /* Copy the function's static chain. */
3492 p = DECL_STRUCT_FUNCTION (old_decl)->static_chain_decl;
3493 if (p)
3494 DECL_STRUCT_FUNCTION (new_decl)->static_chain_decl =
3495 copy_static_chain (DECL_STRUCT_FUNCTION (old_decl)->static_chain_decl,
3496 &id);
3497 /* Copy the function's arguments. */
3498 if (DECL_ARGUMENTS (old_decl) != NULL_TREE)
3499 DECL_ARGUMENTS (new_decl) =
3500 copy_arguments_for_versioning (DECL_ARGUMENTS (old_decl), &id);
3502 /* If there's a tree_map, prepare for substitution. */
3503 if (tree_map)
3504 for (i = 0; i < VARRAY_ACTIVE_SIZE (tree_map); i++)
3506 replace_info = VARRAY_GENERIC_PTR (tree_map, i);
3507 if (replace_info->replace_p)
3508 insert_decl_map (&id, replace_info->old_tree,
3509 replace_info->new_tree);
3512 DECL_INITIAL (new_decl) = remap_blocks (DECL_INITIAL (id.src_fn), &id);
3514 /* Renumber the lexical scoping (non-code) blocks consecutively. */
3515 number_blocks (id.dst_fn);
3517 if (DECL_STRUCT_FUNCTION (old_decl)->unexpanded_var_list != NULL_TREE)
3518 /* Add local vars. */
3519 for (t_step = DECL_STRUCT_FUNCTION (old_decl)->unexpanded_var_list;
3520 t_step; t_step = TREE_CHAIN (t_step))
3522 tree var = TREE_VALUE (t_step);
3523 if (TREE_STATIC (var) && !TREE_ASM_WRITTEN (var))
3524 cfun->unexpanded_var_list = tree_cons (NULL_TREE, var,
3525 cfun->unexpanded_var_list);
3526 else
3527 cfun->unexpanded_var_list =
3528 tree_cons (NULL_TREE, remap_decl (var, &id),
3529 cfun->unexpanded_var_list);
3532 /* Copy the Function's body. */
3533 copy_body (&id, old_entry_block->count, old_entry_block->frequency, ENTRY_BLOCK_PTR, EXIT_BLOCK_PTR);
3535 if (DECL_RESULT (old_decl) != NULL_TREE)
3537 tree *res_decl = &DECL_RESULT (old_decl);
3538 DECL_RESULT (new_decl) = remap_decl (*res_decl, &id);
3539 lang_hooks.dup_lang_specific_decl (DECL_RESULT (new_decl));
3542 /* Renumber the lexical scoping (non-code) blocks consecutively. */
3543 number_blocks (new_decl);
3545 /* Clean up. */
3546 pointer_map_destroy (id.decl_map);
3547 if (!update_clones)
3549 fold_marked_statements (0, id.statements_to_fold);
3550 pointer_set_destroy (id.statements_to_fold);
3551 fold_cond_expr_cond ();
3553 if (gimple_in_ssa_p (cfun))
3555 free_dominance_info (CDI_DOMINATORS);
3556 free_dominance_info (CDI_POST_DOMINATORS);
3557 if (!update_clones)
3558 delete_unreachable_blocks ();
3559 update_ssa (TODO_update_ssa);
3560 if (!update_clones)
3562 fold_cond_expr_cond ();
3563 if (need_ssa_update_p ())
3564 update_ssa (TODO_update_ssa);
3567 free_dominance_info (CDI_DOMINATORS);
3568 free_dominance_info (CDI_POST_DOMINATORS);
3569 pop_cfun ();
3570 current_function_decl = old_current_function_decl;
3571 gcc_assert (!current_function_decl
3572 || DECL_STRUCT_FUNCTION (current_function_decl) == cfun);
3573 return;
3576 /* Duplicate a type, fields and all. */
3578 tree
3579 build_duplicate_type (tree type)
3581 struct copy_body_data id;
3583 memset (&id, 0, sizeof (id));
3584 id.src_fn = current_function_decl;
3585 id.dst_fn = current_function_decl;
3586 id.src_cfun = cfun;
3587 id.decl_map = pointer_map_create ();
3589 type = remap_type_1 (type, &id);
3591 pointer_map_destroy (id.decl_map);
3593 return type;