/cp
[official-gcc.git] / gcc / ipa-icf.c
blobfb7bd482d643d27cf8e9737f3a7e3f052d797ef8
1 /* Interprocedural Identical Code Folding pass
2 Copyright (C) 2014-2015 Free Software Foundation, Inc.
4 Contributed by Jan Hubicka <hubicka@ucw.cz> and Martin Liska <mliska@suse.cz>
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
11 version.
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 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 /* Interprocedural Identical Code Folding for functions and
23 read-only variables.
25 The goal of this transformation is to discover functions and read-only
26 variables which do have exactly the same semantics.
28 In case of functions,
29 we could either create a virtual clone or do a simple function wrapper
30 that will call equivalent function. If the function is just locally visible,
31 all function calls can be redirected. For read-only variables, we create
32 aliases if possible.
34 Optimization pass arranges as follows:
35 1) All functions and read-only variables are visited and internal
36 data structure, either sem_function or sem_variables is created.
37 2) For every symbol from the previous step, VAR_DECL and FUNCTION_DECL are
38 saved and matched to corresponding sem_items.
39 3) These declaration are ignored for equality check and are solved
40 by Value Numbering algorithm published by Alpert, Zadeck in 1992.
41 4) We compute hash value for each symbol.
42 5) Congruence classes are created based on hash value. If hash value are
43 equal, equals function is called and symbols are deeply compared.
44 We must prove that all SSA names, declarations and other items
45 correspond.
46 6) Value Numbering is executed for these classes. At the end of the process
47 all symbol members in remaining classes can be merged.
48 7) Merge operation creates alias in case of read-only variables. For
49 callgraph node, we must decide if we can redirect local calls,
50 create an alias or a thunk.
54 #include "config.h"
55 #include "system.h"
56 #include <list>
57 #include "coretypes.h"
58 #include "alias.h"
59 #include "backend.h"
60 #include "tree.h"
61 #include "gimple.h"
62 #include "rtl.h"
63 #include "ssa.h"
64 #include "options.h"
65 #include "fold-const.h"
66 #include "internal-fn.h"
67 #include "flags.h"
68 #include "insn-config.h"
69 #include "expmed.h"
70 #include "dojump.h"
71 #include "explow.h"
72 #include "calls.h"
73 #include "emit-rtl.h"
74 #include "varasm.h"
75 #include "stmt.h"
76 #include "expr.h"
77 #include "gimple-iterator.h"
78 #include "tree-cfg.h"
79 #include "tree-dfa.h"
80 #include "tree-pass.h"
81 #include "gimple-pretty-print.h"
82 #include "cgraph.h"
83 #include "alloc-pool.h"
84 #include "symbol-summary.h"
85 #include "ipa-prop.h"
86 #include "ipa-inline.h"
87 #include "cfgloop.h"
88 #include "except.h"
89 #include "coverage.h"
90 #include "attribs.h"
91 #include "print-tree.h"
92 #include "target.h"
93 #include "lto-streamer.h"
94 #include "data-streamer.h"
95 #include "ipa-utils.h"
96 #include "ipa-icf-gimple.h"
97 #include "ipa-icf.h"
98 #include "stor-layout.h"
99 #include "dbgcnt.h"
101 using namespace ipa_icf_gimple;
103 namespace ipa_icf {
105 /* Initialization and computation of symtab node hash, there data
106 are propagated later on. */
108 static sem_item_optimizer *optimizer = NULL;
110 /* Constructor. */
112 symbol_compare_collection::symbol_compare_collection (symtab_node *node)
114 m_references.create (0);
115 m_interposables.create (0);
117 ipa_ref *ref;
119 if (is_a <varpool_node *> (node) && DECL_VIRTUAL_P (node->decl))
120 return;
122 for (unsigned i = 0; node->iterate_reference (i, ref); i++)
124 if (ref->address_matters_p ())
125 m_references.safe_push (ref->referred);
127 if (ref->referred->get_availability () <= AVAIL_INTERPOSABLE)
129 if (ref->address_matters_p ())
130 m_references.safe_push (ref->referred);
131 else
132 m_interposables.safe_push (ref->referred);
136 if (is_a <cgraph_node *> (node))
138 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
140 for (cgraph_edge *e = cnode->callees; e; e = e->next_callee)
141 if (e->callee->get_availability () <= AVAIL_INTERPOSABLE)
142 m_interposables.safe_push (e->callee);
146 /* Constructor for key value pair, where _ITEM is key and _INDEX is a target. */
148 sem_usage_pair::sem_usage_pair (sem_item *_item, unsigned int _index):
149 item (_item), index (_index)
153 /* Semantic item constructor for a node of _TYPE, where STACK is used
154 for bitmap memory allocation. */
156 sem_item::sem_item (sem_item_type _type,
157 bitmap_obstack *stack): type(_type), hash(0)
159 setup (stack);
162 /* Semantic item constructor for a node of _TYPE, where STACK is used
163 for bitmap memory allocation. The item is based on symtab node _NODE
164 with computed _HASH. */
166 sem_item::sem_item (sem_item_type _type, symtab_node *_node,
167 hashval_t _hash, bitmap_obstack *stack): type(_type),
168 node (_node), hash (_hash)
170 decl = node->decl;
171 setup (stack);
174 /* Add reference to a semantic TARGET. */
176 void
177 sem_item::add_reference (sem_item *target)
179 refs.safe_push (target);
180 unsigned index = refs.length ();
181 target->usages.safe_push (new sem_usage_pair(this, index));
182 bitmap_set_bit (target->usage_index_bitmap, index);
183 refs_set.add (target->node);
186 /* Initialize internal data structures. Bitmap STACK is used for
187 bitmap memory allocation process. */
189 void
190 sem_item::setup (bitmap_obstack *stack)
192 gcc_checking_assert (node);
194 refs.create (0);
195 tree_refs.create (0);
196 usages.create (0);
197 usage_index_bitmap = BITMAP_ALLOC (stack);
200 sem_item::~sem_item ()
202 for (unsigned i = 0; i < usages.length (); i++)
203 delete usages[i];
205 refs.release ();
206 tree_refs.release ();
207 usages.release ();
209 BITMAP_FREE (usage_index_bitmap);
212 /* Dump function for debugging purpose. */
214 DEBUG_FUNCTION void
215 sem_item::dump (void)
217 if (dump_file)
219 fprintf (dump_file, "[%s] %s (%u) (tree:%p)\n", type == FUNC ? "func" : "var",
220 node->name(), node->order, (void *) node->decl);
221 fprintf (dump_file, " hash: %u\n", get_hash ());
222 fprintf (dump_file, " references: ");
224 for (unsigned i = 0; i < refs.length (); i++)
225 fprintf (dump_file, "%s%s ", refs[i]->node->name (),
226 i < refs.length() - 1 ? "," : "");
228 fprintf (dump_file, "\n");
232 /* Return true if target supports alias symbols. */
234 bool
235 sem_item::target_supports_symbol_aliases_p (void)
237 #if !defined (ASM_OUTPUT_DEF) || (!defined(ASM_OUTPUT_WEAK_ALIAS) && !defined (ASM_WEAKEN_DECL))
238 return false;
239 #else
240 return true;
241 #endif
244 /* Semantic function constructor that uses STACK as bitmap memory stack. */
246 sem_function::sem_function (bitmap_obstack *stack): sem_item (FUNC, stack),
247 m_checker (NULL), m_compared_func (NULL)
249 bb_sizes.create (0);
250 bb_sorted.create (0);
253 /* Constructor based on callgraph node _NODE with computed hash _HASH.
254 Bitmap STACK is used for memory allocation. */
255 sem_function::sem_function (cgraph_node *node, hashval_t hash,
256 bitmap_obstack *stack):
257 sem_item (FUNC, node, hash, stack),
258 m_checker (NULL), m_compared_func (NULL)
260 bb_sizes.create (0);
261 bb_sorted.create (0);
264 sem_function::~sem_function ()
266 for (unsigned i = 0; i < bb_sorted.length (); i++)
267 delete (bb_sorted[i]);
269 bb_sizes.release ();
270 bb_sorted.release ();
273 /* Calculates hash value based on a BASIC_BLOCK. */
275 hashval_t
276 sem_function::get_bb_hash (const sem_bb *basic_block)
278 inchash::hash hstate;
280 hstate.add_int (basic_block->nondbg_stmt_count);
281 hstate.add_int (basic_block->edge_count);
283 return hstate.end ();
286 /* References independent hash function. */
288 hashval_t
289 sem_function::get_hash (void)
291 if(!hash)
293 inchash::hash hstate;
294 hstate.add_int (177454); /* Random number for function type. */
296 hstate.add_int (arg_count);
297 hstate.add_int (cfg_checksum);
298 hstate.add_int (gcode_hash);
300 for (unsigned i = 0; i < bb_sorted.length (); i++)
301 hstate.merge_hash (get_bb_hash (bb_sorted[i]));
303 for (unsigned i = 0; i < bb_sizes.length (); i++)
304 hstate.add_int (bb_sizes[i]);
307 /* Add common features of declaration itself. */
308 if (DECL_FUNCTION_SPECIFIC_TARGET (decl))
309 hstate.add_wide_int
310 (cl_target_option_hash
311 (TREE_TARGET_OPTION (DECL_FUNCTION_SPECIFIC_TARGET (decl))));
312 if (DECL_FUNCTION_SPECIFIC_OPTIMIZATION (decl))
313 (cl_optimization_hash
314 (TREE_OPTIMIZATION (DECL_FUNCTION_SPECIFIC_OPTIMIZATION (decl))));
315 hstate.add_flag (DECL_CXX_CONSTRUCTOR_P (decl));
316 hstate.add_flag (DECL_CXX_DESTRUCTOR_P (decl));
318 hash = hstate.end ();
321 return hash;
324 /* Return ture if A1 and A2 represent equivalent function attribute lists.
325 Based on comp_type_attributes. */
327 bool
328 sem_item::compare_attributes (const_tree a1, const_tree a2)
330 const_tree a;
331 if (a1 == a2)
332 return true;
333 for (a = a1; a != NULL_TREE; a = TREE_CHAIN (a))
335 const struct attribute_spec *as;
336 const_tree attr;
338 as = lookup_attribute_spec (get_attribute_name (a));
339 /* TODO: We can introduce as->affects_decl_identity
340 and as->affects_decl_reference_identity if attribute mismatch
341 gets a common reason to give up on merging. It may not be worth
342 the effort.
343 For example returns_nonnull affects only references, while
344 optimize attribute can be ignored because it is already lowered
345 into flags representation and compared separately. */
346 if (!as)
347 continue;
349 attr = lookup_attribute (as->name, CONST_CAST_TREE (a2));
350 if (!attr || !attribute_value_equal (a, attr))
351 break;
353 if (!a)
355 for (a = a2; a != NULL_TREE; a = TREE_CHAIN (a))
357 const struct attribute_spec *as;
359 as = lookup_attribute_spec (get_attribute_name (a));
360 if (!as)
361 continue;
363 if (!lookup_attribute (as->name, CONST_CAST_TREE (a1)))
364 break;
365 /* We don't need to compare trees again, as we did this
366 already in first loop. */
368 if (!a)
369 return true;
371 /* TODO: As in comp_type_attributes we may want to introduce target hook. */
372 return false;
375 /* Compare properties of symbols N1 and N2 that does not affect semantics of
376 symbol itself but affects semantics of its references from USED_BY (which
377 may be NULL if it is unknown). If comparsion is false, symbols
378 can still be merged but any symbols referring them can't.
380 If ADDRESS is true, do extra checking needed for IPA_REF_ADDR.
382 TODO: We can also split attributes to those that determine codegen of
383 a function body/variable constructor itself and those that are used when
384 referring to it. */
386 bool
387 sem_item::compare_referenced_symbol_properties (symtab_node *used_by,
388 symtab_node *n1,
389 symtab_node *n2,
390 bool address)
392 if (is_a <cgraph_node *> (n1))
394 /* Inline properties matters: we do now want to merge uses of inline
395 function to uses of normal function because inline hint would be lost.
396 We however can merge inline function to noinline because the alias
397 will keep its DECL_DECLARED_INLINE flag.
399 Also ignore inline flag when optimizing for size or when function
400 is known to not be inlinable.
402 TODO: the optimize_size checks can also be assumed to be true if
403 unit has no !optimize_size functions. */
405 if ((!used_by || address || !is_a <cgraph_node *> (used_by)
406 || !opt_for_fn (used_by->decl, optimize_size))
407 && !opt_for_fn (n1->decl, optimize_size)
408 && n1->get_availability () > AVAIL_INTERPOSABLE
409 && (!DECL_UNINLINABLE (n1->decl) || !DECL_UNINLINABLE (n2->decl)))
411 if (DECL_DISREGARD_INLINE_LIMITS (n1->decl)
412 != DECL_DISREGARD_INLINE_LIMITS (n2->decl))
413 return return_false_with_msg
414 ("DECL_DISREGARD_INLINE_LIMITS are different");
416 if (DECL_DECLARED_INLINE_P (n1->decl)
417 != DECL_DECLARED_INLINE_P (n2->decl))
418 return return_false_with_msg ("inline attributes are different");
421 if (DECL_IS_OPERATOR_NEW (n1->decl)
422 != DECL_IS_OPERATOR_NEW (n2->decl))
423 return return_false_with_msg ("operator new flags are different");
426 /* Merging two definitions with a reference to equivalent vtables, but
427 belonging to a different type may result in ipa-polymorphic-call analysis
428 giving a wrong answer about the dynamic type of instance. */
429 if (is_a <varpool_node *> (n1))
431 if ((DECL_VIRTUAL_P (n1->decl) || DECL_VIRTUAL_P (n2->decl))
432 && (DECL_VIRTUAL_P (n1->decl) != DECL_VIRTUAL_P (n2->decl)
433 || !types_must_be_same_for_odr (DECL_CONTEXT (n1->decl),
434 DECL_CONTEXT (n2->decl)))
435 && (!used_by || !is_a <cgraph_node *> (used_by) || address
436 || opt_for_fn (used_by->decl, flag_devirtualize)))
437 return return_false_with_msg
438 ("references to virtual tables can not be merged");
440 if (address && DECL_ALIGN (n1->decl) != DECL_ALIGN (n2->decl))
441 return return_false_with_msg ("alignment mismatch");
443 /* For functions we compare attributes in equals_wpa, because we do
444 not know what attributes may cause codegen differences, but for
445 variables just compare attributes for references - the codegen
446 for constructors is affected only by those attributes that we lower
447 to explicit representation (such as DECL_ALIGN or DECL_SECTION). */
448 if (!compare_attributes (DECL_ATTRIBUTES (n1->decl),
449 DECL_ATTRIBUTES (n2->decl)))
450 return return_false_with_msg ("different var decl attributes");
451 if (comp_type_attributes (TREE_TYPE (n1->decl),
452 TREE_TYPE (n2->decl)) != 1)
453 return return_false_with_msg ("different var type attributes");
456 /* When matching virtual tables, be sure to also match information
457 relevant for polymorphic call analysis. */
458 if (used_by && is_a <varpool_node *> (used_by)
459 && DECL_VIRTUAL_P (used_by->decl))
461 if (DECL_VIRTUAL_P (n1->decl) != DECL_VIRTUAL_P (n2->decl))
462 return return_false_with_msg ("virtual flag mismatch");
463 if (DECL_VIRTUAL_P (n1->decl) && is_a <cgraph_node *> (n1)
464 && (DECL_FINAL_P (n1->decl) != DECL_FINAL_P (n2->decl)))
465 return return_false_with_msg ("final flag mismatch");
467 return true;
470 /* Hash properties that are compared by compare_referenced_symbol_properties. */
472 void
473 sem_item::hash_referenced_symbol_properties (symtab_node *ref,
474 inchash::hash &hstate,
475 bool address)
477 if (is_a <cgraph_node *> (ref))
479 if ((type != FUNC || address || !opt_for_fn (decl, optimize_size))
480 && !opt_for_fn (ref->decl, optimize_size)
481 && !DECL_UNINLINABLE (ref->decl))
483 hstate.add_flag (DECL_DISREGARD_INLINE_LIMITS (ref->decl));
484 hstate.add_flag (DECL_DECLARED_INLINE_P (ref->decl));
486 hstate.add_flag (DECL_IS_OPERATOR_NEW (ref->decl));
488 else if (is_a <varpool_node *> (ref))
490 hstate.add_flag (DECL_VIRTUAL_P (ref->decl));
491 if (address)
492 hstate.add_int (DECL_ALIGN (ref->decl));
497 /* For a given symbol table nodes N1 and N2, we check that FUNCTION_DECLs
498 point to a same function. Comparison can be skipped if IGNORED_NODES
499 contains these nodes. ADDRESS indicate if address is taken. */
501 bool
502 sem_item::compare_symbol_references (
503 hash_map <symtab_node *, sem_item *> &ignored_nodes,
504 symtab_node *n1, symtab_node *n2, bool address)
506 enum availability avail1, avail2;
508 if (n1 == n2)
509 return true;
511 /* Never match variable and function. */
512 if (is_a <varpool_node *> (n1) != is_a <varpool_node *> (n2))
513 return false;
515 if (!compare_referenced_symbol_properties (node, n1, n2, address))
516 return false;
517 if (address && n1->equal_address_to (n2) == 1)
518 return true;
519 if (!address && n1->semantically_equivalent_p (n2))
520 return true;
522 n1 = n1->ultimate_alias_target (&avail1);
523 n2 = n2->ultimate_alias_target (&avail2);
525 if (avail1 >= AVAIL_INTERPOSABLE && ignored_nodes.get (n1)
526 && avail2 >= AVAIL_INTERPOSABLE && ignored_nodes.get (n2))
527 return true;
529 return return_false_with_msg ("different references");
532 /* If cgraph edges E1 and E2 are indirect calls, verify that
533 ECF flags are the same. */
535 bool sem_function::compare_edge_flags (cgraph_edge *e1, cgraph_edge *e2)
537 if (e1->indirect_info && e2->indirect_info)
539 int e1_flags = e1->indirect_info->ecf_flags;
540 int e2_flags = e2->indirect_info->ecf_flags;
542 if (e1_flags != e2_flags)
543 return return_false_with_msg ("ICF flags are different");
545 else if (e1->indirect_info || e2->indirect_info)
546 return false;
548 return true;
551 /* Return true if parameter I may be used. */
553 bool
554 sem_function::param_used_p (unsigned int i)
556 if (ipa_node_params_sum == NULL)
557 return false;
559 struct ipa_node_params *parms_info = IPA_NODE_REF (get_node ());
561 if (parms_info->descriptors.is_empty ()
562 || parms_info->descriptors.length () <= i)
563 return true;
565 return ipa_is_param_used (IPA_NODE_REF (get_node ()), i);
568 /* Perform additional check needed to match types function parameters that are
569 used. Unlike for normal decls it matters if type is TYPE_RESTRICT and we
570 make an assumption that REFERENCE_TYPE parameters are always non-NULL. */
572 bool
573 sem_function::compatible_parm_types_p (tree parm1, tree parm2)
575 /* Be sure that parameters are TBAA compatible. */
576 if (!func_checker::compatible_types_p (parm1, parm2))
577 return return_false_with_msg ("parameter type is not compatible");
579 if (POINTER_TYPE_P (parm1)
580 && (TYPE_RESTRICT (parm1) != TYPE_RESTRICT (parm2)))
581 return return_false_with_msg ("argument restrict flag mismatch");
583 /* nonnull_arg_p implies non-zero range to REFERENCE types. */
584 if (POINTER_TYPE_P (parm1)
585 && TREE_CODE (parm1) != TREE_CODE (parm2)
586 && opt_for_fn (decl, flag_delete_null_pointer_checks))
587 return return_false_with_msg ("pointer wrt reference mismatch");
589 return true;
592 /* Fast equality function based on knowledge known in WPA. */
594 bool
595 sem_function::equals_wpa (sem_item *item,
596 hash_map <symtab_node *, sem_item *> &ignored_nodes)
598 gcc_assert (item->type == FUNC);
599 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
600 cgraph_node *cnode2 = dyn_cast <cgraph_node *> (item->node);
602 m_compared_func = static_cast<sem_function *> (item);
604 if (cnode->thunk.thunk_p != cnode2->thunk.thunk_p)
605 return return_false_with_msg ("thunk_p mismatch");
607 if (cnode->thunk.thunk_p)
609 if (cnode->thunk.fixed_offset != cnode2->thunk.fixed_offset)
610 return return_false_with_msg ("thunk fixed_offset mismatch");
611 if (cnode->thunk.virtual_value != cnode2->thunk.virtual_value)
612 return return_false_with_msg ("thunk virtual_value mismatch");
613 if (cnode->thunk.this_adjusting != cnode2->thunk.this_adjusting)
614 return return_false_with_msg ("thunk this_adjusting mismatch");
615 if (cnode->thunk.virtual_offset_p != cnode2->thunk.virtual_offset_p)
616 return return_false_with_msg ("thunk virtual_offset_p mismatch");
617 if (cnode->thunk.add_pointer_bounds_args
618 != cnode2->thunk.add_pointer_bounds_args)
619 return return_false_with_msg ("thunk add_pointer_bounds_args mismatch");
622 /* Compare special function DECL attributes. */
623 if (DECL_FUNCTION_PERSONALITY (decl)
624 != DECL_FUNCTION_PERSONALITY (item->decl))
625 return return_false_with_msg ("function personalities are different");
627 if (DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT (decl)
628 != DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT (item->decl))
629 return return_false_with_msg ("intrument function entry exit "
630 "attributes are different");
632 if (DECL_NO_LIMIT_STACK (decl) != DECL_NO_LIMIT_STACK (item->decl))
633 return return_false_with_msg ("no stack limit attributes are different");
635 if (DECL_CXX_CONSTRUCTOR_P (decl) != DECL_CXX_CONSTRUCTOR_P (item->decl))
636 return return_false_with_msg ("DECL_CXX_CONSTRUCTOR mismatch");
638 if (DECL_CXX_DESTRUCTOR_P (decl) != DECL_CXX_DESTRUCTOR_P (item->decl))
639 return return_false_with_msg ("DECL_CXX_DESTRUCTOR mismatch");
641 /* TODO: pure/const flags mostly matters only for references, except for
642 the fact that codegen takes LOOPING flag as a hint that loops are
643 finite. We may arrange the code to always pick leader that has least
644 specified flags and then this can go into comparing symbol properties. */
645 if (flags_from_decl_or_type (decl) != flags_from_decl_or_type (item->decl))
646 return return_false_with_msg ("decl_or_type flags are different");
648 /* Do not match polymorphic constructors of different types. They calls
649 type memory location for ipa-polymorphic-call and we do not want
650 it to get confused by wrong type. */
651 if (DECL_CXX_CONSTRUCTOR_P (decl)
652 && TREE_CODE (TREE_TYPE (decl)) == METHOD_TYPE)
654 if (TREE_CODE (TREE_TYPE (item->decl)) != METHOD_TYPE)
655 return return_false_with_msg ("DECL_CXX_CONSTURCTOR type mismatch");
656 else if (!func_checker::compatible_polymorphic_types_p
657 (TYPE_METHOD_BASETYPE (TREE_TYPE (decl)),
658 TYPE_METHOD_BASETYPE (TREE_TYPE (item->decl)), false))
659 return return_false_with_msg ("ctor polymorphic type mismatch");
662 /* Checking function TARGET and OPTIMIZATION flags. */
663 cl_target_option *tar1 = target_opts_for_fn (decl);
664 cl_target_option *tar2 = target_opts_for_fn (item->decl);
666 if (tar1 != tar2 && !cl_target_option_eq (tar1, tar2))
668 if (dump_file && (dump_flags & TDF_DETAILS))
670 fprintf (dump_file, "target flags difference");
671 cl_target_option_print_diff (dump_file, 2, tar1, tar2);
674 return return_false_with_msg ("Target flags are different");
677 cl_optimization *opt1 = opts_for_fn (decl);
678 cl_optimization *opt2 = opts_for_fn (item->decl);
680 if (opt1 != opt2 && memcmp (opt1, opt2, sizeof(cl_optimization)))
682 if (dump_file && (dump_flags & TDF_DETAILS))
684 fprintf (dump_file, "optimization flags difference");
685 cl_optimization_print_diff (dump_file, 2, opt1, opt2);
688 return return_false_with_msg ("optimization flags are different");
691 /* Result type checking. */
692 if (!func_checker::compatible_types_p
693 (TREE_TYPE (TREE_TYPE (decl)),
694 TREE_TYPE (TREE_TYPE (m_compared_func->decl))))
695 return return_false_with_msg ("result types are different");
697 /* Checking types of arguments. */
698 tree list1 = TYPE_ARG_TYPES (TREE_TYPE (decl)),
699 list2 = TYPE_ARG_TYPES (TREE_TYPE (m_compared_func->decl));
700 for (unsigned i = 0; list1 && list2;
701 list1 = TREE_CHAIN (list1), list2 = TREE_CHAIN (list2), i++)
703 tree parm1 = TREE_VALUE (list1);
704 tree parm2 = TREE_VALUE (list2);
706 /* This guard is here for function pointer with attributes (pr59927.c). */
707 if (!parm1 || !parm2)
708 return return_false_with_msg ("NULL argument type");
710 /* Verify that types are compatible to ensure that both functions
711 have same calling conventions. */
712 if (!types_compatible_p (parm1, parm2))
713 return return_false_with_msg ("parameter types are not compatible");
715 if (!param_used_p (i))
716 continue;
718 /* Perform additional checks for used parameters. */
719 if (!compatible_parm_types_p (parm1, parm2))
720 return false;
723 if (list1 || list2)
724 return return_false_with_msg ("Mismatched number of parameters");
726 if (node->num_references () != item->node->num_references ())
727 return return_false_with_msg ("different number of references");
729 /* Checking function attributes.
730 This is quadratic in number of attributes */
731 if (comp_type_attributes (TREE_TYPE (decl),
732 TREE_TYPE (item->decl)) != 1)
733 return return_false_with_msg ("different type attributes");
734 if (!compare_attributes (DECL_ATTRIBUTES (decl),
735 DECL_ATTRIBUTES (item->decl)))
736 return return_false_with_msg ("different decl attributes");
738 /* The type of THIS pointer type memory location for
739 ipa-polymorphic-call-analysis. */
740 if (opt_for_fn (decl, flag_devirtualize)
741 && (TREE_CODE (TREE_TYPE (decl)) == METHOD_TYPE
742 || TREE_CODE (TREE_TYPE (item->decl)) == METHOD_TYPE)
743 && param_used_p (0)
744 && compare_polymorphic_p ())
746 if (TREE_CODE (TREE_TYPE (decl)) != TREE_CODE (TREE_TYPE (item->decl)))
747 return return_false_with_msg ("METHOD_TYPE and FUNCTION_TYPE mismatch");
748 if (!func_checker::compatible_polymorphic_types_p
749 (TYPE_METHOD_BASETYPE (TREE_TYPE (decl)),
750 TYPE_METHOD_BASETYPE (TREE_TYPE (item->decl)), false))
751 return return_false_with_msg ("THIS pointer ODR type mismatch");
754 ipa_ref *ref = NULL, *ref2 = NULL;
755 for (unsigned i = 0; node->iterate_reference (i, ref); i++)
757 item->node->iterate_reference (i, ref2);
759 if (ref->use != ref2->use)
760 return return_false_with_msg ("reference use mismatch");
762 if (!compare_symbol_references (ignored_nodes, ref->referred,
763 ref2->referred,
764 ref->address_matters_p ()))
765 return false;
768 cgraph_edge *e1 = dyn_cast <cgraph_node *> (node)->callees;
769 cgraph_edge *e2 = dyn_cast <cgraph_node *> (item->node)->callees;
771 while (e1 && e2)
773 if (!compare_symbol_references (ignored_nodes, e1->callee,
774 e2->callee, false))
775 return false;
776 if (!compare_edge_flags (e1, e2))
777 return false;
779 e1 = e1->next_callee;
780 e2 = e2->next_callee;
783 if (e1 || e2)
784 return return_false_with_msg ("different number of calls");
786 e1 = dyn_cast <cgraph_node *> (node)->indirect_calls;
787 e2 = dyn_cast <cgraph_node *> (item->node)->indirect_calls;
789 while (e1 && e2)
791 if (!compare_edge_flags (e1, e2))
792 return false;
794 e1 = e1->next_callee;
795 e2 = e2->next_callee;
798 if (e1 || e2)
799 return return_false_with_msg ("different number of indirect calls");
801 return true;
804 /* Update hash by address sensitive references. We iterate over all
805 sensitive references (address_matters_p) and we hash ultime alias
806 target of these nodes, which can improve a semantic item hash.
808 Also hash in referenced symbols properties. This can be done at any time
809 (as the properties should not change), but it is convenient to do it here
810 while we walk the references anyway. */
812 void
813 sem_item::update_hash_by_addr_refs (hash_map <symtab_node *,
814 sem_item *> &m_symtab_node_map)
816 ipa_ref* ref;
817 inchash::hash hstate (hash);
819 for (unsigned i = 0; node->iterate_reference (i, ref); i++)
821 hstate.add_int (ref->use);
822 hash_referenced_symbol_properties (ref->referred, hstate,
823 ref->use == IPA_REF_ADDR);
824 if (ref->address_matters_p () || !m_symtab_node_map.get (ref->referred))
825 hstate.add_int (ref->referred->ultimate_alias_target ()->order);
828 if (is_a <cgraph_node *> (node))
830 for (cgraph_edge *e = dyn_cast <cgraph_node *> (node)->callers; e;
831 e = e->next_caller)
833 sem_item **result = m_symtab_node_map.get (e->callee);
834 hash_referenced_symbol_properties (e->callee, hstate, false);
835 if (!result)
836 hstate.add_int (e->callee->ultimate_alias_target ()->order);
840 hash = hstate.end ();
843 /* Update hash by computed local hash values taken from different
844 semantic items.
845 TODO: stronger SCC based hashing would be desirable here. */
847 void
848 sem_item::update_hash_by_local_refs (hash_map <symtab_node *,
849 sem_item *> &m_symtab_node_map)
851 ipa_ref* ref;
852 inchash::hash state (hash);
854 for (unsigned j = 0; node->iterate_reference (j, ref); j++)
856 sem_item **result = m_symtab_node_map.get (ref->referring);
857 if (result)
858 state.merge_hash ((*result)->hash);
861 if (type == FUNC)
863 for (cgraph_edge *e = dyn_cast <cgraph_node *> (node)->callees; e;
864 e = e->next_callee)
866 sem_item **result = m_symtab_node_map.get (e->caller);
867 if (result)
868 state.merge_hash ((*result)->hash);
872 global_hash = state.end ();
875 /* Returns true if the item equals to ITEM given as argument. */
877 bool
878 sem_function::equals (sem_item *item,
879 hash_map <symtab_node *, sem_item *> &)
881 gcc_assert (item->type == FUNC);
882 bool eq = equals_private (item);
884 if (m_checker != NULL)
886 delete m_checker;
887 m_checker = NULL;
890 if (dump_file && (dump_flags & TDF_DETAILS))
891 fprintf (dump_file,
892 "Equals called for:%s:%s (%u:%u) (%s:%s) with result: %s\n\n",
893 xstrdup_for_dump (node->name()),
894 xstrdup_for_dump (item->node->name ()),
895 node->order,
896 item->node->order,
897 xstrdup_for_dump (node->asm_name ()),
898 xstrdup_for_dump (item->node->asm_name ()),
899 eq ? "true" : "false");
901 return eq;
904 /* Processes function equality comparison. */
906 bool
907 sem_function::equals_private (sem_item *item)
909 if (item->type != FUNC)
910 return false;
912 basic_block bb1, bb2;
913 edge e1, e2;
914 edge_iterator ei1, ei2;
915 bool result = true;
916 tree arg1, arg2;
918 m_compared_func = static_cast<sem_function *> (item);
920 gcc_assert (decl != item->decl);
922 if (bb_sorted.length () != m_compared_func->bb_sorted.length ()
923 || edge_count != m_compared_func->edge_count
924 || cfg_checksum != m_compared_func->cfg_checksum)
925 return return_false ();
927 m_checker = new func_checker (decl, m_compared_func->decl,
928 compare_polymorphic_p (),
929 false,
930 &refs_set,
931 &m_compared_func->refs_set);
932 arg1 = DECL_ARGUMENTS (decl);
933 arg2 = DECL_ARGUMENTS (m_compared_func->decl);
934 for (unsigned i = 0;
935 arg1 && arg2; arg1 = DECL_CHAIN (arg1), arg2 = DECL_CHAIN (arg2), i++)
937 if (!types_compatible_p (TREE_TYPE (arg1), TREE_TYPE (arg2)))
938 return return_false_with_msg ("argument types are not compatible");
939 if (!param_used_p (i))
940 continue;
941 /* Perform additional checks for used parameters. */
942 if (!compatible_parm_types_p (TREE_TYPE (arg1), TREE_TYPE (arg2)))
943 return false;
944 if (!m_checker->compare_decl (arg1, arg2))
945 return return_false ();
947 if (arg1 || arg2)
948 return return_false_with_msg ("Mismatched number of arguments");
950 if (!dyn_cast <cgraph_node *> (node)->has_gimple_body_p ())
951 return true;
953 /* Fill-up label dictionary. */
954 for (unsigned i = 0; i < bb_sorted.length (); ++i)
956 m_checker->parse_labels (bb_sorted[i]);
957 m_checker->parse_labels (m_compared_func->bb_sorted[i]);
960 /* Checking all basic blocks. */
961 for (unsigned i = 0; i < bb_sorted.length (); ++i)
962 if(!m_checker->compare_bb (bb_sorted[i], m_compared_func->bb_sorted[i]))
963 return return_false();
965 dump_message ("All BBs are equal\n");
967 auto_vec <int> bb_dict;
969 /* Basic block edges check. */
970 for (unsigned i = 0; i < bb_sorted.length (); ++i)
972 bb1 = bb_sorted[i]->bb;
973 bb2 = m_compared_func->bb_sorted[i]->bb;
975 ei2 = ei_start (bb2->preds);
977 for (ei1 = ei_start (bb1->preds); ei_cond (ei1, &e1); ei_next (&ei1))
979 ei_cond (ei2, &e2);
981 if (e1->flags != e2->flags)
982 return return_false_with_msg ("flags comparison returns false");
984 if (!bb_dict_test (&bb_dict, e1->src->index, e2->src->index))
985 return return_false_with_msg ("edge comparison returns false");
987 if (!bb_dict_test (&bb_dict, e1->dest->index, e2->dest->index))
988 return return_false_with_msg ("BB comparison returns false");
990 if (!m_checker->compare_edge (e1, e2))
991 return return_false_with_msg ("edge comparison returns false");
993 ei_next (&ei2);
997 /* Basic block PHI nodes comparison. */
998 for (unsigned i = 0; i < bb_sorted.length (); i++)
999 if (!compare_phi_node (bb_sorted[i]->bb, m_compared_func->bb_sorted[i]->bb))
1000 return return_false_with_msg ("PHI node comparison returns false");
1002 return result;
1005 /* Set LOCAL_P of NODE to true if DATA is non-NULL.
1006 Helper for call_for_symbol_thunks_and_aliases. */
1008 static bool
1009 set_local (cgraph_node *node, void *data)
1011 node->local.local = data != NULL;
1012 return false;
1015 /* TREE_ADDRESSABLE of NODE to true.
1016 Helper for call_for_symbol_thunks_and_aliases. */
1018 static bool
1019 set_addressable (varpool_node *node, void *)
1021 TREE_ADDRESSABLE (node->decl) = 1;
1022 return false;
1025 /* Clear DECL_RTL of NODE.
1026 Helper for call_for_symbol_thunks_and_aliases. */
1028 static bool
1029 clear_decl_rtl (symtab_node *node, void *)
1031 SET_DECL_RTL (node->decl, NULL);
1032 return false;
1035 /* Redirect all callers of N and its aliases to TO. Remove aliases if
1036 possible. Return number of redirections made. */
1038 static int
1039 redirect_all_callers (cgraph_node *n, cgraph_node *to)
1041 int nredirected = 0;
1042 ipa_ref *ref;
1043 cgraph_edge *e = n->callers;
1045 while (e)
1047 /* Redirecting thunks to interposable symbols or symbols in other sections
1048 may not be supported by target output code. Play safe for now and
1049 punt on redirection. */
1050 if (!e->caller->thunk.thunk_p)
1052 struct cgraph_edge *nexte = e->next_caller;
1053 e->redirect_callee (to);
1054 e = nexte;
1055 nredirected++;
1057 else
1058 e = e->next_callee;
1060 for (unsigned i = 0; n->iterate_direct_aliases (i, ref);)
1062 bool removed = false;
1063 cgraph_node *n_alias = dyn_cast <cgraph_node *> (ref->referring);
1065 if ((DECL_COMDAT_GROUP (n->decl)
1066 && (DECL_COMDAT_GROUP (n->decl)
1067 == DECL_COMDAT_GROUP (n_alias->decl)))
1068 || (n_alias->get_availability () > AVAIL_INTERPOSABLE
1069 && n->get_availability () > AVAIL_INTERPOSABLE))
1071 nredirected += redirect_all_callers (n_alias, to);
1072 if (n_alias->can_remove_if_no_direct_calls_p ()
1073 && !n_alias->call_for_symbol_and_aliases (cgraph_node::has_thunk_p,
1074 NULL, true)
1075 && !n_alias->has_aliases_p ())
1076 n_alias->remove ();
1078 if (!removed)
1079 i++;
1081 return nredirected;
1084 /* Merges instance with an ALIAS_ITEM, where alias, thunk or redirection can
1085 be applied. */
1087 bool
1088 sem_function::merge (sem_item *alias_item)
1090 gcc_assert (alias_item->type == FUNC);
1092 sem_function *alias_func = static_cast<sem_function *> (alias_item);
1094 cgraph_node *original = get_node ();
1095 cgraph_node *local_original = NULL;
1096 cgraph_node *alias = alias_func->get_node ();
1098 bool create_wrapper = false;
1099 bool create_alias = false;
1100 bool redirect_callers = false;
1101 bool remove = false;
1103 bool original_discardable = false;
1104 bool original_discarded = false;
1106 bool original_address_matters = original->address_matters_p ();
1107 bool alias_address_matters = alias->address_matters_p ();
1109 if (DECL_EXTERNAL (alias->decl))
1111 if (dump_file)
1112 fprintf (dump_file, "Not unifying; alias is external.\n\n");
1113 return false;
1116 if (DECL_NO_INLINE_WARNING_P (original->decl)
1117 != DECL_NO_INLINE_WARNING_P (alias->decl))
1119 if (dump_file)
1120 fprintf (dump_file,
1121 "Not unifying; "
1122 "DECL_NO_INLINE_WARNING mismatch.\n\n");
1123 return false;
1126 /* Do not attempt to mix functions from different user sections;
1127 we do not know what user intends with those. */
1128 if (((DECL_SECTION_NAME (original->decl) && !original->implicit_section)
1129 || (DECL_SECTION_NAME (alias->decl) && !alias->implicit_section))
1130 && DECL_SECTION_NAME (original->decl) != DECL_SECTION_NAME (alias->decl))
1132 if (dump_file)
1133 fprintf (dump_file,
1134 "Not unifying; "
1135 "original and alias are in different sections.\n\n");
1136 return false;
1139 /* See if original is in a section that can be discarded if the main
1140 symbol is not used. */
1142 if (original->can_be_discarded_p ())
1143 original_discardable = true;
1144 /* Also consider case where we have resolution info and we know that
1145 original's definition is not going to be used. In this case we can not
1146 create alias to original. */
1147 if (node->resolution != LDPR_UNKNOWN
1148 && !decl_binds_to_current_def_p (node->decl))
1149 original_discardable = original_discarded = true;
1151 /* Creating a symtab alias is the optimal way to merge.
1152 It however can not be used in the following cases:
1154 1) if ORIGINAL and ALIAS may be possibly compared for address equality.
1155 2) if ORIGINAL is in a section that may be discarded by linker or if
1156 it is an external functions where we can not create an alias
1157 (ORIGINAL_DISCARDABLE)
1158 3) if target do not support symbol aliases.
1159 4) original and alias lie in different comdat groups.
1161 If we can not produce alias, we will turn ALIAS into WRAPPER of ORIGINAL
1162 and/or redirect all callers from ALIAS to ORIGINAL. */
1163 if ((original_address_matters && alias_address_matters)
1164 || (original_discardable
1165 && (!DECL_COMDAT_GROUP (alias->decl)
1166 || (DECL_COMDAT_GROUP (alias->decl)
1167 != DECL_COMDAT_GROUP (original->decl))))
1168 || original_discarded
1169 || !sem_item::target_supports_symbol_aliases_p ()
1170 || DECL_COMDAT_GROUP (alias->decl) != DECL_COMDAT_GROUP (original->decl))
1172 /* First see if we can produce wrapper. */
1174 /* Symbol properties that matter for references must be preserved.
1175 TODO: We can produce wrapper, but we need to produce alias of ORIGINAL
1176 with proper properties. */
1177 if (!sem_item::compare_referenced_symbol_properties (NULL, original, alias,
1178 alias->address_taken))
1180 if (dump_file)
1181 fprintf (dump_file,
1182 "Wrapper cannot be created because referenced symbol "
1183 "properties mismatch\n");
1185 /* Do not turn function in one comdat group into wrapper to another
1186 comdat group. Other compiler producing the body of the
1187 another comdat group may make opossite decision and with unfortunate
1188 linker choices this may close a loop. */
1189 else if (DECL_COMDAT_GROUP (original->decl)
1190 && DECL_COMDAT_GROUP (alias->decl)
1191 && (DECL_COMDAT_GROUP (alias->decl)
1192 != DECL_COMDAT_GROUP (original->decl)))
1194 if (dump_file)
1195 fprintf (dump_file,
1196 "Wrapper cannot be created because of COMDAT\n");
1198 else if (DECL_STATIC_CHAIN (alias->decl))
1200 if (dump_file)
1201 fprintf (dump_file,
1202 "Can not create wrapper of nested functions.\n");
1204 /* TODO: We can also deal with variadic functions never calling
1205 VA_START. */
1206 else if (stdarg_p (TREE_TYPE (alias->decl)))
1208 if (dump_file)
1209 fprintf (dump_file,
1210 "can not create wrapper of stdarg function.\n");
1212 else if (inline_summaries
1213 && inline_summaries->get (alias)->self_size <= 2)
1215 if (dump_file)
1216 fprintf (dump_file, "Wrapper creation is not "
1217 "profitable (function is too small).\n");
1219 /* If user paid attention to mark function noinline, assume it is
1220 somewhat special and do not try to turn it into a wrapper that can
1221 not be undone by inliner. */
1222 else if (lookup_attribute ("noinline", DECL_ATTRIBUTES (alias->decl)))
1224 if (dump_file)
1225 fprintf (dump_file, "Wrappers are not created for noinline.\n");
1227 else
1228 create_wrapper = true;
1230 /* We can redirect local calls in the case both alias and orignal
1231 are not interposable. */
1232 redirect_callers
1233 = alias->get_availability () > AVAIL_INTERPOSABLE
1234 && original->get_availability () > AVAIL_INTERPOSABLE
1235 && !alias->instrumented_version;
1236 /* TODO: We can redirect, but we need to produce alias of ORIGINAL
1237 with proper properties. */
1238 if (!sem_item::compare_referenced_symbol_properties (NULL, original, alias,
1239 alias->address_taken))
1240 redirect_callers = false;
1242 if (!redirect_callers && !create_wrapper)
1244 if (dump_file)
1245 fprintf (dump_file, "Not unifying; can not redirect callers nor "
1246 "produce wrapper\n\n");
1247 return false;
1250 /* Work out the symbol the wrapper should call.
1251 If ORIGINAL is interposable, we need to call a local alias.
1252 Also produce local alias (if possible) as an optimization.
1254 Local aliases can not be created inside comdat groups because that
1255 prevents inlining. */
1256 if (!original_discardable && !original->get_comdat_group ())
1258 local_original
1259 = dyn_cast <cgraph_node *> (original->noninterposable_alias ());
1260 if (!local_original
1261 && original->get_availability () > AVAIL_INTERPOSABLE)
1262 local_original = original;
1264 /* If we can not use local alias, fallback to the original
1265 when possible. */
1266 else if (original->get_availability () > AVAIL_INTERPOSABLE)
1267 local_original = original;
1269 /* If original is COMDAT local, we can not really redirect calls outside
1270 of its comdat group to it. */
1271 if (original->comdat_local_p ())
1272 redirect_callers = false;
1273 if (!local_original)
1275 if (dump_file)
1276 fprintf (dump_file, "Not unifying; "
1277 "can not produce local alias.\n\n");
1278 return false;
1281 if (!redirect_callers && !create_wrapper)
1283 if (dump_file)
1284 fprintf (dump_file, "Not unifying; "
1285 "can not redirect callers nor produce a wrapper\n\n");
1286 return false;
1288 if (!create_wrapper
1289 && !alias->call_for_symbol_and_aliases (cgraph_node::has_thunk_p,
1290 NULL, true)
1291 && !alias->can_remove_if_no_direct_calls_p ())
1293 if (dump_file)
1294 fprintf (dump_file, "Not unifying; can not make wrapper and "
1295 "function has other uses than direct calls\n\n");
1296 return false;
1299 else
1300 create_alias = true;
1302 if (redirect_callers)
1304 int nredirected = redirect_all_callers (alias, local_original);
1306 if (nredirected)
1308 alias->icf_merged = true;
1309 local_original->icf_merged = true;
1311 if (dump_file && nredirected)
1312 fprintf (dump_file, "%i local calls have been "
1313 "redirected.\n", nredirected);
1316 /* If all callers was redirected, do not produce wrapper. */
1317 if (alias->can_remove_if_no_direct_calls_p ()
1318 && !alias->has_aliases_p ())
1320 create_wrapper = false;
1321 remove = true;
1323 gcc_assert (!create_alias);
1325 else if (create_alias)
1327 alias->icf_merged = true;
1329 /* Remove the function's body. */
1330 ipa_merge_profiles (original, alias);
1331 alias->release_body (true);
1332 alias->reset ();
1333 /* Notice global symbol possibly produced RTL. */
1334 ((symtab_node *)alias)->call_for_symbol_and_aliases (clear_decl_rtl,
1335 NULL, true);
1337 /* Create the alias. */
1338 cgraph_node::create_alias (alias_func->decl, decl);
1339 alias->resolve_alias (original);
1341 original->call_for_symbol_thunks_and_aliases
1342 (set_local, (void *)(size_t) original->local_p (), true);
1344 if (dump_file)
1345 fprintf (dump_file, "Unified; Function alias has been created.\n\n");
1347 if (create_wrapper)
1349 gcc_assert (!create_alias);
1350 alias->icf_merged = true;
1351 local_original->icf_merged = true;
1353 ipa_merge_profiles (local_original, alias, true);
1354 alias->create_wrapper (local_original);
1356 if (dump_file)
1357 fprintf (dump_file, "Unified; Wrapper has been created.\n\n");
1360 /* It's possible that redirection can hit thunks that block
1361 redirection opportunities. */
1362 gcc_assert (alias->icf_merged || remove || redirect_callers);
1363 original->icf_merged = true;
1365 /* Inform the inliner about cross-module merging. */
1366 if ((original->lto_file_data || alias->lto_file_data)
1367 && original->lto_file_data != alias->lto_file_data)
1368 local_original->merged = original->merged = true;
1370 if (remove)
1372 ipa_merge_profiles (original, alias);
1373 alias->release_body ();
1374 alias->reset ();
1375 alias->body_removed = true;
1376 alias->icf_merged = true;
1377 if (dump_file)
1378 fprintf (dump_file, "Unified; Function body was removed.\n");
1381 return true;
1384 /* Semantic item initialization function. */
1386 void
1387 sem_function::init (void)
1389 if (in_lto_p)
1390 get_node ()->get_untransformed_body ();
1392 tree fndecl = node->decl;
1393 function *func = DECL_STRUCT_FUNCTION (fndecl);
1395 gcc_assert (func);
1396 gcc_assert (SSANAMES (func));
1398 ssa_names_size = SSANAMES (func)->length ();
1399 node = node;
1401 decl = fndecl;
1402 region_tree = func->eh->region_tree;
1404 /* iterating all function arguments. */
1405 arg_count = count_formal_params (fndecl);
1407 edge_count = n_edges_for_fn (func);
1408 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
1409 if (!cnode->thunk.thunk_p)
1411 cfg_checksum = coverage_compute_cfg_checksum (func);
1413 inchash::hash hstate;
1415 basic_block bb;
1416 FOR_EACH_BB_FN (bb, func)
1418 unsigned nondbg_stmt_count = 0;
1420 edge e;
1421 for (edge_iterator ei = ei_start (bb->preds); ei_cond (ei, &e);
1422 ei_next (&ei))
1423 cfg_checksum = iterative_hash_host_wide_int (e->flags,
1424 cfg_checksum);
1426 for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p (gsi);
1427 gsi_next (&gsi))
1429 gimple stmt = gsi_stmt (gsi);
1431 if (gimple_code (stmt) != GIMPLE_DEBUG
1432 && gimple_code (stmt) != GIMPLE_PREDICT)
1434 hash_stmt (stmt, hstate);
1435 nondbg_stmt_count++;
1439 gcode_hash = hstate.end ();
1440 bb_sizes.safe_push (nondbg_stmt_count);
1442 /* Inserting basic block to hash table. */
1443 sem_bb *semantic_bb = new sem_bb (bb, nondbg_stmt_count,
1444 EDGE_COUNT (bb->preds)
1445 + EDGE_COUNT (bb->succs));
1447 bb_sorted.safe_push (semantic_bb);
1450 else
1452 cfg_checksum = 0;
1453 inchash::hash hstate;
1454 hstate.add_wide_int (cnode->thunk.fixed_offset);
1455 hstate.add_wide_int (cnode->thunk.virtual_value);
1456 hstate.add_flag (cnode->thunk.this_adjusting);
1457 hstate.add_flag (cnode->thunk.virtual_offset_p);
1458 hstate.add_flag (cnode->thunk.add_pointer_bounds_args);
1459 gcode_hash = hstate.end ();
1463 /* Accumulate to HSTATE a hash of expression EXP.
1464 Identical to inchash::add_expr, but guaranteed to be stable across LTO
1465 and DECL equality classes. */
1467 void
1468 sem_item::add_expr (const_tree exp, inchash::hash &hstate)
1470 if (exp == NULL_TREE)
1472 hstate.merge_hash (0);
1473 return;
1476 /* Handled component can be matched in a cureful way proving equivalence
1477 even if they syntactically differ. Just skip them. */
1478 STRIP_NOPS (exp);
1479 while (handled_component_p (exp))
1480 exp = TREE_OPERAND (exp, 0);
1482 enum tree_code code = TREE_CODE (exp);
1483 hstate.add_int (code);
1485 switch (code)
1487 /* Use inchash::add_expr for everything that is LTO stable. */
1488 case VOID_CST:
1489 case INTEGER_CST:
1490 case REAL_CST:
1491 case FIXED_CST:
1492 case STRING_CST:
1493 case COMPLEX_CST:
1494 case VECTOR_CST:
1495 inchash::add_expr (exp, hstate);
1496 break;
1497 case CONSTRUCTOR:
1499 unsigned HOST_WIDE_INT idx;
1500 tree value;
1502 hstate.add_wide_int (int_size_in_bytes (TREE_TYPE (exp)));
1504 FOR_EACH_CONSTRUCTOR_VALUE (CONSTRUCTOR_ELTS (exp), idx, value)
1505 if (value)
1506 add_expr (value, hstate);
1507 break;
1509 case ADDR_EXPR:
1510 case FDESC_EXPR:
1511 add_expr (get_base_address (TREE_OPERAND (exp, 0)), hstate);
1512 break;
1513 case SSA_NAME:
1514 case VAR_DECL:
1515 case CONST_DECL:
1516 case PARM_DECL:
1517 hstate.add_wide_int (int_size_in_bytes (TREE_TYPE (exp)));
1518 break;
1519 case MEM_REF:
1520 case POINTER_PLUS_EXPR:
1521 case MINUS_EXPR:
1522 case RANGE_EXPR:
1523 add_expr (TREE_OPERAND (exp, 0), hstate);
1524 add_expr (TREE_OPERAND (exp, 1), hstate);
1525 break;
1526 case PLUS_EXPR:
1528 inchash::hash one, two;
1529 add_expr (TREE_OPERAND (exp, 0), one);
1530 add_expr (TREE_OPERAND (exp, 1), two);
1531 hstate.add_commutative (one, two);
1533 break;
1534 CASE_CONVERT:
1535 hstate.add_wide_int (int_size_in_bytes (TREE_TYPE (exp)));
1536 return add_expr (TREE_OPERAND (exp, 0), hstate);
1537 default:
1538 break;
1542 /* Accumulate to HSTATE a hash of type t.
1543 TYpes that may end up being compatible after LTO type merging needs to have
1544 the same hash. */
1546 void
1547 sem_item::add_type (const_tree type, inchash::hash &hstate)
1549 if (type == NULL_TREE)
1551 hstate.merge_hash (0);
1552 return;
1555 type = TYPE_MAIN_VARIANT (type);
1556 if (TYPE_CANONICAL (type))
1557 type = TYPE_CANONICAL (type);
1559 if (!AGGREGATE_TYPE_P (type))
1560 hstate.add_int (TYPE_MODE (type));
1562 if (TREE_CODE (type) == COMPLEX_TYPE)
1564 hstate.add_int (COMPLEX_TYPE);
1565 sem_item::add_type (TREE_TYPE (type), hstate);
1567 else if (INTEGRAL_TYPE_P (type))
1569 hstate.add_int (INTEGER_TYPE);
1570 hstate.add_flag (TYPE_UNSIGNED (type));
1571 hstate.add_int (TYPE_PRECISION (type));
1573 else if (VECTOR_TYPE_P (type))
1575 hstate.add_int (VECTOR_TYPE);
1576 hstate.add_int (TYPE_PRECISION (type));
1577 sem_item::add_type (TREE_TYPE (type), hstate);
1579 else if (TREE_CODE (type) == ARRAY_TYPE)
1581 hstate.add_int (ARRAY_TYPE);
1582 /* Do not hash size, so complete and incomplete types can match. */
1583 sem_item::add_type (TREE_TYPE (type), hstate);
1585 else if (RECORD_OR_UNION_TYPE_P (type))
1587 hashval_t *val = optimizer->m_type_hash_cache.get (type);
1589 if (!val)
1591 inchash::hash hstate2;
1592 unsigned nf;
1593 tree f;
1594 hashval_t hash;
1596 hstate2.add_int (RECORD_TYPE);
1597 gcc_assert (COMPLETE_TYPE_P (type));
1599 for (f = TYPE_FIELDS (type), nf = 0; f; f = TREE_CHAIN (f))
1600 if (TREE_CODE (f) == FIELD_DECL)
1602 add_type (TREE_TYPE (f), hstate2);
1603 nf++;
1606 hstate2.add_int (nf);
1607 hash = hstate2.end ();
1608 hstate.add_wide_int (hash);
1609 optimizer->m_type_hash_cache.put (type, hash);
1611 else
1612 hstate.add_wide_int (*val);
1616 /* Improve accumulated hash for HSTATE based on a gimple statement STMT. */
1618 void
1619 sem_function::hash_stmt (gimple stmt, inchash::hash &hstate)
1621 enum gimple_code code = gimple_code (stmt);
1623 hstate.add_int (code);
1625 switch (code)
1627 case GIMPLE_SWITCH:
1628 add_expr (gimple_switch_index (as_a <gswitch *> (stmt)), hstate);
1629 break;
1630 case GIMPLE_ASSIGN:
1631 hstate.add_int (gimple_assign_rhs_code (stmt));
1632 if (commutative_tree_code (gimple_assign_rhs_code (stmt))
1633 || commutative_ternary_tree_code (gimple_assign_rhs_code (stmt)))
1635 inchash::hash one, two;
1637 add_expr (gimple_assign_rhs1 (stmt), one);
1638 add_type (TREE_TYPE (gimple_assign_rhs1 (stmt)), one);
1639 add_expr (gimple_assign_rhs2 (stmt), two);
1640 hstate.add_commutative (one, two);
1641 if (commutative_ternary_tree_code (gimple_assign_rhs_code (stmt)))
1643 add_expr (gimple_assign_rhs3 (stmt), hstate);
1644 add_type (TREE_TYPE (gimple_assign_rhs3 (stmt)), hstate);
1646 add_expr (gimple_assign_lhs (stmt), hstate);
1647 add_type (TREE_TYPE (gimple_assign_lhs (stmt)), two);
1648 break;
1650 /* ... fall through ... */
1651 case GIMPLE_CALL:
1652 case GIMPLE_ASM:
1653 case GIMPLE_COND:
1654 case GIMPLE_GOTO:
1655 case GIMPLE_RETURN:
1656 /* All these statements are equivalent if their operands are. */
1657 for (unsigned i = 0; i < gimple_num_ops (stmt); ++i)
1659 add_expr (gimple_op (stmt, i), hstate);
1660 if (gimple_op (stmt, i))
1661 add_type (TREE_TYPE (gimple_op (stmt, i)), hstate);
1663 default:
1664 break;
1669 /* Return true if polymorphic comparison must be processed. */
1671 bool
1672 sem_function::compare_polymorphic_p (void)
1674 struct cgraph_edge *e;
1676 if (!opt_for_fn (get_node ()->decl, flag_devirtualize))
1677 return false;
1678 if (get_node ()->indirect_calls != NULL)
1679 return true;
1680 /* TODO: We can do simple propagation determining what calls may lead to
1681 a polymorphic call. */
1682 for (e = get_node ()->callees; e; e = e->next_callee)
1683 if (e->callee->definition
1684 && opt_for_fn (e->callee->decl, flag_devirtualize))
1685 return true;
1686 return false;
1689 /* For a given call graph NODE, the function constructs new
1690 semantic function item. */
1692 sem_function *
1693 sem_function::parse (cgraph_node *node, bitmap_obstack *stack)
1695 tree fndecl = node->decl;
1696 function *func = DECL_STRUCT_FUNCTION (fndecl);
1698 if (!func || (!node->has_gimple_body_p () && !node->thunk.thunk_p))
1699 return NULL;
1701 if (lookup_attribute_by_prefix ("omp ", DECL_ATTRIBUTES (node->decl)) != NULL)
1702 return NULL;
1704 sem_function *f = new sem_function (node, 0, stack);
1706 f->init ();
1708 return f;
1711 /* For given basic blocks BB1 and BB2 (from functions FUNC1 and FUNC),
1712 return true if phi nodes are semantically equivalent in these blocks . */
1714 bool
1715 sem_function::compare_phi_node (basic_block bb1, basic_block bb2)
1717 gphi_iterator si1, si2;
1718 gphi *phi1, *phi2;
1719 unsigned size1, size2, i;
1720 tree t1, t2;
1721 edge e1, e2;
1723 gcc_assert (bb1 != NULL);
1724 gcc_assert (bb2 != NULL);
1726 si2 = gsi_start_phis (bb2);
1727 for (si1 = gsi_start_phis (bb1); !gsi_end_p (si1);
1728 gsi_next (&si1))
1730 gsi_next_nonvirtual_phi (&si1);
1731 gsi_next_nonvirtual_phi (&si2);
1733 if (gsi_end_p (si1) && gsi_end_p (si2))
1734 break;
1736 if (gsi_end_p (si1) || gsi_end_p (si2))
1737 return return_false();
1739 phi1 = si1.phi ();
1740 phi2 = si2.phi ();
1742 tree phi_result1 = gimple_phi_result (phi1);
1743 tree phi_result2 = gimple_phi_result (phi2);
1745 if (!m_checker->compare_operand (phi_result1, phi_result2))
1746 return return_false_with_msg ("PHI results are different");
1748 size1 = gimple_phi_num_args (phi1);
1749 size2 = gimple_phi_num_args (phi2);
1751 if (size1 != size2)
1752 return return_false ();
1754 for (i = 0; i < size1; ++i)
1756 t1 = gimple_phi_arg (phi1, i)->def;
1757 t2 = gimple_phi_arg (phi2, i)->def;
1759 if (!m_checker->compare_operand (t1, t2))
1760 return return_false ();
1762 e1 = gimple_phi_arg_edge (phi1, i);
1763 e2 = gimple_phi_arg_edge (phi2, i);
1765 if (!m_checker->compare_edge (e1, e2))
1766 return return_false ();
1769 gsi_next (&si2);
1772 return true;
1775 /* Returns true if tree T can be compared as a handled component. */
1777 bool
1778 sem_function::icf_handled_component_p (tree t)
1780 tree_code tc = TREE_CODE (t);
1782 return (handled_component_p (t)
1783 || tc == ADDR_EXPR || tc == MEM_REF || tc == OBJ_TYPE_REF);
1786 /* Basic blocks dictionary BB_DICT returns true if SOURCE index BB
1787 corresponds to TARGET. */
1789 bool
1790 sem_function::bb_dict_test (vec<int> *bb_dict, int source, int target)
1792 source++;
1793 target++;
1795 if (bb_dict->length () <= (unsigned)source)
1796 bb_dict->safe_grow_cleared (source + 1);
1798 if ((*bb_dict)[source] == 0)
1800 (*bb_dict)[source] = target;
1801 return true;
1803 else
1804 return (*bb_dict)[source] == target;
1808 /* Semantic variable constructor that uses STACK as bitmap memory stack. */
1810 sem_variable::sem_variable (bitmap_obstack *stack): sem_item (VAR, stack)
1814 /* Constructor based on varpool node _NODE with computed hash _HASH.
1815 Bitmap STACK is used for memory allocation. */
1817 sem_variable::sem_variable (varpool_node *node, hashval_t _hash,
1818 bitmap_obstack *stack): sem_item(VAR,
1819 node, _hash, stack)
1821 gcc_checking_assert (node);
1822 gcc_checking_assert (get_node ());
1825 /* Fast equality function based on knowledge known in WPA. */
1827 bool
1828 sem_variable::equals_wpa (sem_item *item,
1829 hash_map <symtab_node *, sem_item *> &ignored_nodes)
1831 gcc_assert (item->type == VAR);
1833 if (node->num_references () != item->node->num_references ())
1834 return return_false_with_msg ("different number of references");
1836 if (DECL_TLS_MODEL (decl) || DECL_TLS_MODEL (item->decl))
1837 return return_false_with_msg ("TLS model");
1839 /* DECL_ALIGN is safe to merge, because we will always chose the largest
1840 alignment out of all aliases. */
1842 if (DECL_VIRTUAL_P (decl) != DECL_VIRTUAL_P (item->decl))
1843 return return_false_with_msg ("Virtual flag mismatch");
1845 if (DECL_SIZE (decl) != DECL_SIZE (item->decl)
1846 && ((!DECL_SIZE (decl) || !DECL_SIZE (item->decl))
1847 || !operand_equal_p (DECL_SIZE (decl),
1848 DECL_SIZE (item->decl), OEP_ONLY_CONST)))
1849 return return_false_with_msg ("size mismatch");
1851 /* Do not attempt to mix data from different user sections;
1852 we do not know what user intends with those. */
1853 if (((DECL_SECTION_NAME (decl) && !node->implicit_section)
1854 || (DECL_SECTION_NAME (item->decl) && !item->node->implicit_section))
1855 && DECL_SECTION_NAME (decl) != DECL_SECTION_NAME (item->decl))
1856 return return_false_with_msg ("user section mismatch");
1858 if (DECL_IN_TEXT_SECTION (decl) != DECL_IN_TEXT_SECTION (item->decl))
1859 return return_false_with_msg ("text section");
1861 ipa_ref *ref = NULL, *ref2 = NULL;
1862 for (unsigned i = 0; node->iterate_reference (i, ref); i++)
1864 item->node->iterate_reference (i, ref2);
1866 if (ref->use != ref2->use)
1867 return return_false_with_msg ("reference use mismatch");
1869 if (!compare_symbol_references (ignored_nodes,
1870 ref->referred, ref2->referred,
1871 ref->address_matters_p ()))
1872 return false;
1875 return true;
1878 /* Returns true if the item equals to ITEM given as argument. */
1880 bool
1881 sem_variable::equals (sem_item *item,
1882 hash_map <symtab_node *, sem_item *> &)
1884 gcc_assert (item->type == VAR);
1885 bool ret;
1887 if (DECL_INITIAL (decl) == error_mark_node && in_lto_p)
1888 dyn_cast <varpool_node *>(node)->get_constructor ();
1889 if (DECL_INITIAL (item->decl) == error_mark_node && in_lto_p)
1890 dyn_cast <varpool_node *>(item->node)->get_constructor ();
1892 /* As seen in PR ipa/65303 we have to compare variables types. */
1893 if (!func_checker::compatible_types_p (TREE_TYPE (decl),
1894 TREE_TYPE (item->decl)))
1895 return return_false_with_msg ("variables types are different");
1897 ret = sem_variable::equals (DECL_INITIAL (decl),
1898 DECL_INITIAL (item->node->decl));
1899 if (dump_file && (dump_flags & TDF_DETAILS))
1900 fprintf (dump_file,
1901 "Equals called for vars:%s:%s (%u:%u) (%s:%s) with result: %s\n\n",
1902 xstrdup_for_dump (node->name()),
1903 xstrdup_for_dump (item->node->name ()),
1904 node->order, item->node->order,
1905 xstrdup_for_dump (node->asm_name ()),
1906 xstrdup_for_dump (item->node->asm_name ()), ret ? "true" : "false");
1908 return ret;
1911 /* Compares trees T1 and T2 for semantic equality. */
1913 bool
1914 sem_variable::equals (tree t1, tree t2)
1916 if (!t1 || !t2)
1917 return return_with_debug (t1 == t2);
1918 if (t1 == t2)
1919 return true;
1920 tree_code tc1 = TREE_CODE (t1);
1921 tree_code tc2 = TREE_CODE (t2);
1923 if (tc1 != tc2)
1924 return return_false_with_msg ("TREE_CODE mismatch");
1926 switch (tc1)
1928 case CONSTRUCTOR:
1930 vec<constructor_elt, va_gc> *v1, *v2;
1931 unsigned HOST_WIDE_INT idx;
1933 enum tree_code typecode = TREE_CODE (TREE_TYPE (t1));
1934 if (typecode != TREE_CODE (TREE_TYPE (t2)))
1935 return return_false_with_msg ("constructor type mismatch");
1937 if (typecode == ARRAY_TYPE)
1939 HOST_WIDE_INT size_1 = int_size_in_bytes (TREE_TYPE (t1));
1940 /* For arrays, check that the sizes all match. */
1941 if (TYPE_MODE (TREE_TYPE (t1)) != TYPE_MODE (TREE_TYPE (t2))
1942 || size_1 == -1
1943 || size_1 != int_size_in_bytes (TREE_TYPE (t2)))
1944 return return_false_with_msg ("constructor array size mismatch");
1946 else if (!func_checker::compatible_types_p (TREE_TYPE (t1),
1947 TREE_TYPE (t2)))
1948 return return_false_with_msg ("constructor type incompatible");
1950 v1 = CONSTRUCTOR_ELTS (t1);
1951 v2 = CONSTRUCTOR_ELTS (t2);
1952 if (vec_safe_length (v1) != vec_safe_length (v2))
1953 return return_false_with_msg ("constructor number of elts mismatch");
1955 for (idx = 0; idx < vec_safe_length (v1); ++idx)
1957 constructor_elt *c1 = &(*v1)[idx];
1958 constructor_elt *c2 = &(*v2)[idx];
1960 /* Check that each value is the same... */
1961 if (!sem_variable::equals (c1->value, c2->value))
1962 return false;
1963 /* ... and that they apply to the same fields! */
1964 if (!sem_variable::equals (c1->index, c2->index))
1965 return false;
1967 return true;
1969 case MEM_REF:
1971 tree x1 = TREE_OPERAND (t1, 0);
1972 tree x2 = TREE_OPERAND (t2, 0);
1973 tree y1 = TREE_OPERAND (t1, 1);
1974 tree y2 = TREE_OPERAND (t2, 1);
1976 if (!func_checker::compatible_types_p (TREE_TYPE (x1), TREE_TYPE (x2)))
1977 return return_false ();
1979 /* Type of the offset on MEM_REF does not matter. */
1980 return return_with_debug (sem_variable::equals (x1, x2)
1981 && wi::to_offset (y1)
1982 == wi::to_offset (y2));
1984 case ADDR_EXPR:
1985 case FDESC_EXPR:
1987 tree op1 = TREE_OPERAND (t1, 0);
1988 tree op2 = TREE_OPERAND (t2, 0);
1989 return sem_variable::equals (op1, op2);
1991 /* References to other vars/decls are compared using ipa-ref. */
1992 case FUNCTION_DECL:
1993 case VAR_DECL:
1994 if (decl_in_symtab_p (t1) && decl_in_symtab_p (t2))
1995 return true;
1996 return return_false_with_msg ("Declaration mismatch");
1997 case CONST_DECL:
1998 /* TODO: We can check CONST_DECL by its DECL_INITIAL, but for that we
1999 need to process its VAR/FUNCTION references without relying on ipa-ref
2000 compare. */
2001 case FIELD_DECL:
2002 case LABEL_DECL:
2003 return return_false_with_msg ("Declaration mismatch");
2004 case INTEGER_CST:
2005 /* Integer constants are the same only if the same width of type. */
2006 if (TYPE_PRECISION (TREE_TYPE (t1)) != TYPE_PRECISION (TREE_TYPE (t2)))
2007 return return_false_with_msg ("INTEGER_CST precision mismatch");
2008 if (TYPE_MODE (TREE_TYPE (t1)) != TYPE_MODE (TREE_TYPE (t2)))
2009 return return_false_with_msg ("INTEGER_CST mode mismatch");
2010 return return_with_debug (tree_int_cst_equal (t1, t2));
2011 case STRING_CST:
2012 if (TYPE_MODE (TREE_TYPE (t1)) != TYPE_MODE (TREE_TYPE (t2)))
2013 return return_false_with_msg ("STRING_CST mode mismatch");
2014 if (TREE_STRING_LENGTH (t1) != TREE_STRING_LENGTH (t2))
2015 return return_false_with_msg ("STRING_CST length mismatch");
2016 if (memcmp (TREE_STRING_POINTER (t1), TREE_STRING_POINTER (t2),
2017 TREE_STRING_LENGTH (t1)))
2018 return return_false_with_msg ("STRING_CST mismatch");
2019 return true;
2020 case FIXED_CST:
2021 /* Fixed constants are the same only if the same width of type. */
2022 if (TYPE_PRECISION (TREE_TYPE (t1)) != TYPE_PRECISION (TREE_TYPE (t2)))
2023 return return_false_with_msg ("FIXED_CST precision mismatch");
2025 return return_with_debug (FIXED_VALUES_IDENTICAL (TREE_FIXED_CST (t1),
2026 TREE_FIXED_CST (t2)));
2027 case COMPLEX_CST:
2028 return (sem_variable::equals (TREE_REALPART (t1), TREE_REALPART (t2))
2029 && sem_variable::equals (TREE_IMAGPART (t1), TREE_IMAGPART (t2)));
2030 case REAL_CST:
2031 /* Real constants are the same only if the same width of type. */
2032 if (TYPE_PRECISION (TREE_TYPE (t1)) != TYPE_PRECISION (TREE_TYPE (t2)))
2033 return return_false_with_msg ("REAL_CST precision mismatch");
2034 return return_with_debug (REAL_VALUES_IDENTICAL (TREE_REAL_CST (t1),
2035 TREE_REAL_CST (t2)));
2036 case VECTOR_CST:
2038 unsigned i;
2040 if (VECTOR_CST_NELTS (t1) != VECTOR_CST_NELTS (t2))
2041 return return_false_with_msg ("VECTOR_CST nelts mismatch");
2043 for (i = 0; i < VECTOR_CST_NELTS (t1); ++i)
2044 if (!sem_variable::equals (VECTOR_CST_ELT (t1, i),
2045 VECTOR_CST_ELT (t2, i)))
2046 return 0;
2048 return 1;
2050 case ARRAY_REF:
2051 case ARRAY_RANGE_REF:
2053 tree x1 = TREE_OPERAND (t1, 0);
2054 tree x2 = TREE_OPERAND (t2, 0);
2055 tree y1 = TREE_OPERAND (t1, 1);
2056 tree y2 = TREE_OPERAND (t2, 1);
2058 if (!sem_variable::equals (x1, x2) || !sem_variable::equals (y1, y2))
2059 return false;
2060 if (!sem_variable::equals (array_ref_low_bound (t1),
2061 array_ref_low_bound (t2)))
2062 return false;
2063 if (!sem_variable::equals (array_ref_element_size (t1),
2064 array_ref_element_size (t2)))
2065 return false;
2066 return true;
2069 case COMPONENT_REF:
2070 case POINTER_PLUS_EXPR:
2071 case PLUS_EXPR:
2072 case MINUS_EXPR:
2073 case RANGE_EXPR:
2075 tree x1 = TREE_OPERAND (t1, 0);
2076 tree x2 = TREE_OPERAND (t2, 0);
2077 tree y1 = TREE_OPERAND (t1, 1);
2078 tree y2 = TREE_OPERAND (t2, 1);
2080 return sem_variable::equals (x1, x2) && sem_variable::equals (y1, y2);
2083 CASE_CONVERT:
2084 case VIEW_CONVERT_EXPR:
2085 if (!func_checker::compatible_types_p (TREE_TYPE (t1), TREE_TYPE (t2)))
2086 return return_false ();
2087 return sem_variable::equals (TREE_OPERAND (t1, 0), TREE_OPERAND (t2, 0));
2088 case ERROR_MARK:
2089 return return_false_with_msg ("ERROR_MARK");
2090 default:
2091 return return_false_with_msg ("Unknown TREE code reached");
2095 /* Parser function that visits a varpool NODE. */
2097 sem_variable *
2098 sem_variable::parse (varpool_node *node, bitmap_obstack *stack)
2100 if (TREE_THIS_VOLATILE (node->decl) || DECL_HARD_REGISTER (node->decl)
2101 || node->alias)
2102 return NULL;
2104 sem_variable *v = new sem_variable (node, 0, stack);
2106 v->init ();
2108 return v;
2111 /* References independent hash function. */
2113 hashval_t
2114 sem_variable::get_hash (void)
2116 if (hash)
2117 return hash;
2119 /* All WPA streamed in symbols should have their hashes computed at compile
2120 time. At this point, the constructor may not be in memory at all.
2121 DECL_INITIAL (decl) would be error_mark_node in that case. */
2122 gcc_assert (!node->lto_file_data);
2123 tree ctor = DECL_INITIAL (decl);
2124 inchash::hash hstate;
2126 hstate.add_int (456346417);
2127 if (DECL_SIZE (decl) && tree_fits_shwi_p (DECL_SIZE (decl)))
2128 hstate.add_wide_int (tree_to_shwi (DECL_SIZE (decl)));
2129 add_expr (ctor, hstate);
2130 hash = hstate.end ();
2132 return hash;
2135 /* Merges instance with an ALIAS_ITEM, where alias, thunk or redirection can
2136 be applied. */
2138 bool
2139 sem_variable::merge (sem_item *alias_item)
2141 gcc_assert (alias_item->type == VAR);
2143 if (!sem_item::target_supports_symbol_aliases_p ())
2145 if (dump_file)
2146 fprintf (dump_file, "Not unifying; "
2147 "Symbol aliases are not supported by target\n\n");
2148 return false;
2151 if (DECL_EXTERNAL (alias_item->decl))
2153 if (dump_file)
2154 fprintf (dump_file, "Not unifying; alias is external.\n\n");
2155 return false;
2158 sem_variable *alias_var = static_cast<sem_variable *> (alias_item);
2160 varpool_node *original = get_node ();
2161 varpool_node *alias = alias_var->get_node ();
2162 bool original_discardable = false;
2164 bool original_address_matters = original->address_matters_p ();
2165 bool alias_address_matters = alias->address_matters_p ();
2167 /* See if original is in a section that can be discarded if the main
2168 symbol is not used.
2169 Also consider case where we have resolution info and we know that
2170 original's definition is not going to be used. In this case we can not
2171 create alias to original. */
2172 if (original->can_be_discarded_p ()
2173 || (node->resolution != LDPR_UNKNOWN
2174 && !decl_binds_to_current_def_p (node->decl)))
2175 original_discardable = true;
2177 gcc_assert (!TREE_ASM_WRITTEN (alias->decl));
2179 /* Constant pool machinery is not quite ready for aliases.
2180 TODO: varasm code contains logic for merging DECL_IN_CONSTANT_POOL.
2181 For LTO merging does not happen that is an important missing feature.
2182 We can enable merging with LTO if the DECL_IN_CONSTANT_POOL
2183 flag is dropped and non-local symbol name is assigned. */
2184 if (DECL_IN_CONSTANT_POOL (alias->decl)
2185 || DECL_IN_CONSTANT_POOL (original->decl))
2187 if (dump_file)
2188 fprintf (dump_file,
2189 "Not unifying; constant pool variables.\n\n");
2190 return false;
2193 /* Do not attempt to mix functions from different user sections;
2194 we do not know what user intends with those. */
2195 if (((DECL_SECTION_NAME (original->decl) && !original->implicit_section)
2196 || (DECL_SECTION_NAME (alias->decl) && !alias->implicit_section))
2197 && DECL_SECTION_NAME (original->decl) != DECL_SECTION_NAME (alias->decl))
2199 if (dump_file)
2200 fprintf (dump_file,
2201 "Not unifying; "
2202 "original and alias are in different sections.\n\n");
2203 return false;
2206 /* We can not merge if address comparsion metters. */
2207 if (original_address_matters && alias_address_matters
2208 && flag_merge_constants < 2)
2210 if (dump_file)
2211 fprintf (dump_file,
2212 "Not unifying; "
2213 "adress of original and alias may be compared.\n\n");
2214 return false;
2216 if (DECL_COMDAT_GROUP (original->decl) != DECL_COMDAT_GROUP (alias->decl))
2218 if (dump_file)
2219 fprintf (dump_file, "Not unifying; alias cannot be created; "
2220 "across comdat group boundary\n\n");
2222 return false;
2225 if (original_discardable)
2227 if (dump_file)
2228 fprintf (dump_file, "Not unifying; alias cannot be created; "
2229 "target is discardable\n\n");
2231 return false;
2233 else
2235 gcc_assert (!original->alias);
2236 gcc_assert (!alias->alias);
2238 alias->analyzed = false;
2240 DECL_INITIAL (alias->decl) = NULL;
2241 ((symtab_node *)alias)->call_for_symbol_and_aliases (clear_decl_rtl,
2242 NULL, true);
2243 alias->need_bounds_init = false;
2244 alias->remove_all_references ();
2245 if (TREE_ADDRESSABLE (alias->decl))
2246 original->call_for_symbol_and_aliases (set_addressable, NULL, true);
2248 varpool_node::create_alias (alias_var->decl, decl);
2249 alias->resolve_alias (original);
2251 if (dump_file)
2252 fprintf (dump_file, "Unified; Variable alias has been created.\n\n");
2254 return true;
2258 /* Dump symbol to FILE. */
2260 void
2261 sem_variable::dump_to_file (FILE *file)
2263 gcc_assert (file);
2265 print_node (file, "", decl, 0);
2266 fprintf (file, "\n\n");
2269 unsigned int sem_item_optimizer::class_id = 0;
2271 sem_item_optimizer::sem_item_optimizer (): worklist (0), m_classes (0),
2272 m_classes_count (0), m_cgraph_node_hooks (NULL), m_varpool_node_hooks (NULL)
2274 m_items.create (0);
2275 bitmap_obstack_initialize (&m_bmstack);
2278 sem_item_optimizer::~sem_item_optimizer ()
2280 for (unsigned int i = 0; i < m_items.length (); i++)
2281 delete m_items[i];
2283 for (hash_table<congruence_class_group_hash>::iterator it = m_classes.begin ();
2284 it != m_classes.end (); ++it)
2286 for (unsigned int i = 0; i < (*it)->classes.length (); i++)
2287 delete (*it)->classes[i];
2289 (*it)->classes.release ();
2290 free (*it);
2293 m_items.release ();
2295 bitmap_obstack_release (&m_bmstack);
2298 /* Write IPA ICF summary for symbols. */
2300 void
2301 sem_item_optimizer::write_summary (void)
2303 unsigned int count = 0;
2305 output_block *ob = create_output_block (LTO_section_ipa_icf);
2306 lto_symtab_encoder_t encoder = ob->decl_state->symtab_node_encoder;
2307 ob->symbol = NULL;
2309 /* Calculate number of symbols to be serialized. */
2310 for (lto_symtab_encoder_iterator lsei = lsei_start_in_partition (encoder);
2311 !lsei_end_p (lsei);
2312 lsei_next_in_partition (&lsei))
2314 symtab_node *node = lsei_node (lsei);
2316 if (m_symtab_node_map.get (node))
2317 count++;
2320 streamer_write_uhwi (ob, count);
2322 /* Process all of the symbols. */
2323 for (lto_symtab_encoder_iterator lsei = lsei_start_in_partition (encoder);
2324 !lsei_end_p (lsei);
2325 lsei_next_in_partition (&lsei))
2327 symtab_node *node = lsei_node (lsei);
2329 sem_item **item = m_symtab_node_map.get (node);
2331 if (item && *item)
2333 int node_ref = lto_symtab_encoder_encode (encoder, node);
2334 streamer_write_uhwi_stream (ob->main_stream, node_ref);
2336 streamer_write_uhwi (ob, (*item)->get_hash ());
2340 streamer_write_char_stream (ob->main_stream, 0);
2341 produce_asm (ob, NULL);
2342 destroy_output_block (ob);
2345 /* Reads a section from LTO stream file FILE_DATA. Input block for DATA
2346 contains LEN bytes. */
2348 void
2349 sem_item_optimizer::read_section (lto_file_decl_data *file_data,
2350 const char *data, size_t len)
2352 const lto_function_header *header =
2353 (const lto_function_header *) data;
2354 const int cfg_offset = sizeof (lto_function_header);
2355 const int main_offset = cfg_offset + header->cfg_size;
2356 const int string_offset = main_offset + header->main_size;
2357 data_in *data_in;
2358 unsigned int i;
2359 unsigned int count;
2361 lto_input_block ib_main ((const char *) data + main_offset, 0,
2362 header->main_size, file_data->mode_table);
2364 data_in =
2365 lto_data_in_create (file_data, (const char *) data + string_offset,
2366 header->string_size, vNULL);
2368 count = streamer_read_uhwi (&ib_main);
2370 for (i = 0; i < count; i++)
2372 unsigned int index;
2373 symtab_node *node;
2374 lto_symtab_encoder_t encoder;
2376 index = streamer_read_uhwi (&ib_main);
2377 encoder = file_data->symtab_node_encoder;
2378 node = lto_symtab_encoder_deref (encoder, index);
2380 hashval_t hash = streamer_read_uhwi (&ib_main);
2382 gcc_assert (node->definition);
2384 if (dump_file)
2385 fprintf (dump_file, "Symbol added:%s (tree: %p, uid:%u)\n",
2386 node->asm_name (), (void *) node->decl, node->order);
2388 if (is_a<cgraph_node *> (node))
2390 cgraph_node *cnode = dyn_cast <cgraph_node *> (node);
2392 m_items.safe_push (new sem_function (cnode, hash, &m_bmstack));
2394 else
2396 varpool_node *vnode = dyn_cast <varpool_node *> (node);
2398 m_items.safe_push (new sem_variable (vnode, hash, &m_bmstack));
2402 lto_free_section_data (file_data, LTO_section_ipa_icf, NULL, data,
2403 len);
2404 lto_data_in_delete (data_in);
2407 /* Read IPA IPA ICF summary for symbols. */
2409 void
2410 sem_item_optimizer::read_summary (void)
2412 lto_file_decl_data **file_data_vec = lto_get_file_decl_data ();
2413 lto_file_decl_data *file_data;
2414 unsigned int j = 0;
2416 while ((file_data = file_data_vec[j++]))
2418 size_t len;
2419 const char *data = lto_get_section_data (file_data,
2420 LTO_section_ipa_icf, NULL, &len);
2422 if (data)
2423 read_section (file_data, data, len);
2427 /* Register callgraph and varpool hooks. */
2429 void
2430 sem_item_optimizer::register_hooks (void)
2432 if (!m_cgraph_node_hooks)
2433 m_cgraph_node_hooks = symtab->add_cgraph_removal_hook
2434 (&sem_item_optimizer::cgraph_removal_hook, this);
2436 if (!m_varpool_node_hooks)
2437 m_varpool_node_hooks = symtab->add_varpool_removal_hook
2438 (&sem_item_optimizer::varpool_removal_hook, this);
2441 /* Unregister callgraph and varpool hooks. */
2443 void
2444 sem_item_optimizer::unregister_hooks (void)
2446 if (m_cgraph_node_hooks)
2447 symtab->remove_cgraph_removal_hook (m_cgraph_node_hooks);
2449 if (m_varpool_node_hooks)
2450 symtab->remove_varpool_removal_hook (m_varpool_node_hooks);
2453 /* Adds a CLS to hashtable associated by hash value. */
2455 void
2456 sem_item_optimizer::add_class (congruence_class *cls)
2458 gcc_assert (cls->members.length ());
2460 congruence_class_group *group = get_group_by_hash (
2461 cls->members[0]->get_hash (),
2462 cls->members[0]->type);
2463 group->classes.safe_push (cls);
2466 /* Gets a congruence class group based on given HASH value and TYPE. */
2468 congruence_class_group *
2469 sem_item_optimizer::get_group_by_hash (hashval_t hash, sem_item_type type)
2471 congruence_class_group *item = XNEW (congruence_class_group);
2472 item->hash = hash;
2473 item->type = type;
2475 congruence_class_group **slot = m_classes.find_slot (item, INSERT);
2477 if (*slot)
2478 free (item);
2479 else
2481 item->classes.create (1);
2482 *slot = item;
2485 return *slot;
2488 /* Callgraph removal hook called for a NODE with a custom DATA. */
2490 void
2491 sem_item_optimizer::cgraph_removal_hook (cgraph_node *node, void *data)
2493 sem_item_optimizer *optimizer = (sem_item_optimizer *) data;
2494 optimizer->remove_symtab_node (node);
2497 /* Varpool removal hook called for a NODE with a custom DATA. */
2499 void
2500 sem_item_optimizer::varpool_removal_hook (varpool_node *node, void *data)
2502 sem_item_optimizer *optimizer = (sem_item_optimizer *) data;
2503 optimizer->remove_symtab_node (node);
2506 /* Remove symtab NODE triggered by symtab removal hooks. */
2508 void
2509 sem_item_optimizer::remove_symtab_node (symtab_node *node)
2511 gcc_assert (!m_classes.elements());
2513 m_removed_items_set.add (node);
2516 void
2517 sem_item_optimizer::remove_item (sem_item *item)
2519 if (m_symtab_node_map.get (item->node))
2520 m_symtab_node_map.remove (item->node);
2521 delete item;
2524 /* Removes all callgraph and varpool nodes that are marked by symtab
2525 as deleted. */
2527 void
2528 sem_item_optimizer::filter_removed_items (void)
2530 auto_vec <sem_item *> filtered;
2532 for (unsigned int i = 0; i < m_items.length(); i++)
2534 sem_item *item = m_items[i];
2536 if (m_removed_items_set.contains (item->node))
2538 remove_item (item);
2539 continue;
2542 if (item->type == FUNC)
2544 cgraph_node *cnode = static_cast <sem_function *>(item)->get_node ();
2546 if (in_lto_p && (cnode->alias || cnode->body_removed))
2547 remove_item (item);
2548 else
2549 filtered.safe_push (item);
2551 else /* VAR. */
2553 if (!flag_ipa_icf_variables)
2554 remove_item (item);
2555 else
2557 /* Filter out non-readonly variables. */
2558 tree decl = item->decl;
2559 if (TREE_READONLY (decl))
2560 filtered.safe_push (item);
2561 else
2562 remove_item (item);
2567 /* Clean-up of released semantic items. */
2569 m_items.release ();
2570 for (unsigned int i = 0; i < filtered.length(); i++)
2571 m_items.safe_push (filtered[i]);
2574 /* Optimizer entry point which returns true in case it processes
2575 a merge operation. True is returned if there's a merge operation
2576 processed. */
2578 bool
2579 sem_item_optimizer::execute (void)
2581 filter_removed_items ();
2582 unregister_hooks ();
2584 build_graph ();
2585 update_hash_by_addr_refs ();
2586 build_hash_based_classes ();
2588 if (dump_file)
2589 fprintf (dump_file, "Dump after hash based groups\n");
2590 dump_cong_classes ();
2592 for (unsigned int i = 0; i < m_items.length(); i++)
2593 m_items[i]->init_wpa ();
2595 subdivide_classes_by_equality (true);
2597 if (dump_file)
2598 fprintf (dump_file, "Dump after WPA based types groups\n");
2600 dump_cong_classes ();
2602 process_cong_reduction ();
2603 verify_classes ();
2605 if (dump_file)
2606 fprintf (dump_file, "Dump after callgraph-based congruence reduction\n");
2608 dump_cong_classes ();
2610 parse_nonsingleton_classes ();
2611 subdivide_classes_by_equality ();
2613 if (dump_file)
2614 fprintf (dump_file, "Dump after full equality comparison of groups\n");
2616 dump_cong_classes ();
2618 unsigned int prev_class_count = m_classes_count;
2620 process_cong_reduction ();
2621 dump_cong_classes ();
2622 verify_classes ();
2623 bool merged_p = merge_classes (prev_class_count);
2625 if (dump_file && (dump_flags & TDF_DETAILS))
2626 symtab_node::dump_table (dump_file);
2628 return merged_p;
2631 /* Function responsible for visiting all potential functions and
2632 read-only variables that can be merged. */
2634 void
2635 sem_item_optimizer::parse_funcs_and_vars (void)
2637 cgraph_node *cnode;
2639 if (flag_ipa_icf_functions)
2640 FOR_EACH_DEFINED_FUNCTION (cnode)
2642 sem_function *f = sem_function::parse (cnode, &m_bmstack);
2643 if (f)
2645 m_items.safe_push (f);
2646 m_symtab_node_map.put (cnode, f);
2648 if (dump_file)
2649 fprintf (dump_file, "Parsed function:%s\n", f->node->asm_name ());
2651 if (dump_file && (dump_flags & TDF_DETAILS))
2652 f->dump_to_file (dump_file);
2654 else if (dump_file)
2655 fprintf (dump_file, "Not parsed function:%s\n", cnode->asm_name ());
2658 varpool_node *vnode;
2660 if (flag_ipa_icf_variables)
2661 FOR_EACH_DEFINED_VARIABLE (vnode)
2663 sem_variable *v = sem_variable::parse (vnode, &m_bmstack);
2665 if (v)
2667 m_items.safe_push (v);
2668 m_symtab_node_map.put (vnode, v);
2673 /* Makes pairing between a congruence class CLS and semantic ITEM. */
2675 void
2676 sem_item_optimizer::add_item_to_class (congruence_class *cls, sem_item *item)
2678 item->index_in_class = cls->members.length ();
2679 cls->members.safe_push (item);
2680 item->cls = cls;
2683 /* For each semantic item, append hash values of references. */
2685 void
2686 sem_item_optimizer::update_hash_by_addr_refs ()
2688 /* First, append to hash sensitive references and class type if it need to
2689 be matched for ODR. */
2690 for (unsigned i = 0; i < m_items.length (); i++)
2692 m_items[i]->update_hash_by_addr_refs (m_symtab_node_map);
2693 if (m_items[i]->type == FUNC)
2695 if (TREE_CODE (TREE_TYPE (m_items[i]->decl)) == METHOD_TYPE
2696 && contains_polymorphic_type_p
2697 (TYPE_METHOD_BASETYPE (TREE_TYPE (m_items[i]->decl)))
2698 && (DECL_CXX_CONSTRUCTOR_P (m_items[i]->decl)
2699 || (static_cast<sem_function *> (m_items[i])->param_used_p (0)
2700 && static_cast<sem_function *> (m_items[i])
2701 ->compare_polymorphic_p ())))
2703 tree class_type
2704 = TYPE_METHOD_BASETYPE (TREE_TYPE (m_items[i]->decl));
2705 inchash::hash hstate (m_items[i]->hash);
2707 if (TYPE_NAME (class_type)
2708 && DECL_ASSEMBLER_NAME_SET_P (TYPE_NAME (class_type)))
2709 hstate.add_wide_int
2710 (IDENTIFIER_HASH_VALUE
2711 (DECL_ASSEMBLER_NAME (TYPE_NAME (class_type))));
2713 m_items[i]->hash = hstate.end ();
2718 /* Once all symbols have enhanced hash value, we can append
2719 hash values of symbols that are seen by IPA ICF and are
2720 references by a semantic item. Newly computed values
2721 are saved to global_hash member variable. */
2722 for (unsigned i = 0; i < m_items.length (); i++)
2723 m_items[i]->update_hash_by_local_refs (m_symtab_node_map);
2725 /* Global hash value replace current hash values. */
2726 for (unsigned i = 0; i < m_items.length (); i++)
2727 m_items[i]->hash = m_items[i]->global_hash;
2730 /* Congruence classes are built by hash value. */
2732 void
2733 sem_item_optimizer::build_hash_based_classes (void)
2735 for (unsigned i = 0; i < m_items.length (); i++)
2737 sem_item *item = m_items[i];
2739 congruence_class_group *group = get_group_by_hash (item->hash,
2740 item->type);
2742 if (!group->classes.length ())
2744 m_classes_count++;
2745 group->classes.safe_push (new congruence_class (class_id++));
2748 add_item_to_class (group->classes[0], item);
2752 /* Build references according to call graph. */
2754 void
2755 sem_item_optimizer::build_graph (void)
2757 for (unsigned i = 0; i < m_items.length (); i++)
2759 sem_item *item = m_items[i];
2760 m_symtab_node_map.put (item->node, item);
2763 for (unsigned i = 0; i < m_items.length (); i++)
2765 sem_item *item = m_items[i];
2767 if (item->type == FUNC)
2769 cgraph_node *cnode = dyn_cast <cgraph_node *> (item->node);
2771 cgraph_edge *e = cnode->callees;
2772 while (e)
2774 sem_item **slot = m_symtab_node_map.get
2775 (e->callee->ultimate_alias_target ());
2776 if (slot)
2777 item->add_reference (*slot);
2779 e = e->next_callee;
2783 ipa_ref *ref = NULL;
2784 for (unsigned i = 0; item->node->iterate_reference (i, ref); i++)
2786 sem_item **slot = m_symtab_node_map.get
2787 (ref->referred->ultimate_alias_target ());
2788 if (slot)
2789 item->add_reference (*slot);
2794 /* Semantic items in classes having more than one element and initialized.
2795 In case of WPA, we load function body. */
2797 void
2798 sem_item_optimizer::parse_nonsingleton_classes (void)
2800 unsigned int init_called_count = 0;
2802 for (unsigned i = 0; i < m_items.length (); i++)
2803 if (m_items[i]->cls->members.length () > 1)
2805 m_items[i]->init ();
2806 init_called_count++;
2809 if (dump_file)
2810 fprintf (dump_file, "Init called for %u items (%.2f%%).\n", init_called_count,
2811 m_items.length () ? 100.0f * init_called_count / m_items.length (): 0.0f);
2814 /* Equality function for semantic items is used to subdivide existing
2815 classes. If IN_WPA, fast equality function is invoked. */
2817 void
2818 sem_item_optimizer::subdivide_classes_by_equality (bool in_wpa)
2820 for (hash_table <congruence_class_group_hash>::iterator it = m_classes.begin ();
2821 it != m_classes.end (); ++it)
2823 unsigned int class_count = (*it)->classes.length ();
2825 for (unsigned i = 0; i < class_count; i++)
2827 congruence_class *c = (*it)->classes [i];
2829 if (c->members.length() > 1)
2831 auto_vec <sem_item *> new_vector;
2833 sem_item *first = c->members[0];
2834 new_vector.safe_push (first);
2836 unsigned class_split_first = (*it)->classes.length ();
2838 for (unsigned j = 1; j < c->members.length (); j++)
2840 sem_item *item = c->members[j];
2842 bool equals = in_wpa ? first->equals_wpa (item,
2843 m_symtab_node_map) : first->equals (item, m_symtab_node_map);
2845 if (equals)
2846 new_vector.safe_push (item);
2847 else
2849 bool integrated = false;
2851 for (unsigned k = class_split_first; k < (*it)->classes.length (); k++)
2853 sem_item *x = (*it)->classes[k]->members[0];
2854 bool equals = in_wpa ? x->equals_wpa (item,
2855 m_symtab_node_map) : x->equals (item, m_symtab_node_map);
2857 if (equals)
2859 integrated = true;
2860 add_item_to_class ((*it)->classes[k], item);
2862 break;
2866 if (!integrated)
2868 congruence_class *c = new congruence_class (class_id++);
2869 m_classes_count++;
2870 add_item_to_class (c, item);
2872 (*it)->classes.safe_push (c);
2877 // we replace newly created new_vector for the class we've just splitted
2878 c->members.release ();
2879 c->members.create (new_vector.length ());
2881 for (unsigned int j = 0; j < new_vector.length (); j++)
2882 add_item_to_class (c, new_vector[j]);
2887 verify_classes ();
2890 /* Subdivide classes by address references that members of the class
2891 reference. Example can be a pair of functions that have an address
2892 taken from a function. If these addresses are different the class
2893 is split. */
2895 unsigned
2896 sem_item_optimizer::subdivide_classes_by_sensitive_refs ()
2898 typedef hash_map <symbol_compare_hash, vec <sem_item *> > subdivide_hash_map;
2900 unsigned newly_created_classes = 0;
2902 for (hash_table <congruence_class_group_hash>::iterator it = m_classes.begin ();
2903 it != m_classes.end (); ++it)
2905 unsigned int class_count = (*it)->classes.length ();
2906 auto_vec<congruence_class *> new_classes;
2908 for (unsigned i = 0; i < class_count; i++)
2910 congruence_class *c = (*it)->classes [i];
2912 if (c->members.length() > 1)
2914 subdivide_hash_map split_map;
2916 for (unsigned j = 0; j < c->members.length (); j++)
2918 sem_item *source_node = c->members[j];
2920 symbol_compare_collection *collection = new symbol_compare_collection (source_node->node);
2922 bool existed;
2923 vec <sem_item *> *slot = &split_map.get_or_insert (collection,
2924 &existed);
2925 gcc_checking_assert (slot);
2927 slot->safe_push (source_node);
2929 if (existed)
2930 delete collection;
2933 /* If the map contains more than one key, we have to split the map
2934 appropriately. */
2935 if (split_map.elements () != 1)
2937 bool first_class = true;
2939 for (subdivide_hash_map::iterator it2 = split_map.begin ();
2940 it2 != split_map.end (); ++it2)
2942 congruence_class *new_cls;
2943 new_cls = new congruence_class (class_id++);
2945 for (unsigned k = 0; k < (*it2).second.length (); k++)
2946 add_item_to_class (new_cls, (*it2).second[k]);
2948 worklist_push (new_cls);
2949 newly_created_classes++;
2951 if (first_class)
2953 (*it)->classes[i] = new_cls;
2954 first_class = false;
2956 else
2958 new_classes.safe_push (new_cls);
2959 m_classes_count++;
2964 /* Release memory. */
2965 for (subdivide_hash_map::iterator it2 = split_map.begin ();
2966 it2 != split_map.end (); ++it2)
2968 delete (*it2).first;
2969 (*it2).second.release ();
2974 for (unsigned i = 0; i < new_classes.length (); i++)
2975 (*it)->classes.safe_push (new_classes[i]);
2978 return newly_created_classes;
2981 /* Verify congruence classes if checking is enabled. */
2983 void
2984 sem_item_optimizer::verify_classes (void)
2986 #if ENABLE_CHECKING
2987 for (hash_table <congruence_class_group_hash>::iterator it = m_classes.begin ();
2988 it != m_classes.end (); ++it)
2990 for (unsigned int i = 0; i < (*it)->classes.length (); i++)
2992 congruence_class *cls = (*it)->classes[i];
2994 gcc_checking_assert (cls);
2995 gcc_checking_assert (cls->members.length () > 0);
2997 for (unsigned int j = 0; j < cls->members.length (); j++)
2999 sem_item *item = cls->members[j];
3001 gcc_checking_assert (item);
3002 gcc_checking_assert (item->cls == cls);
3004 for (unsigned k = 0; k < item->usages.length (); k++)
3006 sem_usage_pair *usage = item->usages[k];
3007 gcc_checking_assert (usage->item->index_in_class <
3008 usage->item->cls->members.length ());
3013 #endif
3016 /* Disposes split map traverse function. CLS_PTR is pointer to congruence
3017 class, BSLOT is bitmap slot we want to release. DATA is mandatory,
3018 but unused argument. */
3020 bool
3021 sem_item_optimizer::release_split_map (congruence_class * const &,
3022 bitmap const &b, traverse_split_pair *)
3024 bitmap bmp = b;
3026 BITMAP_FREE (bmp);
3028 return true;
3031 /* Process split operation for a class given as pointer CLS_PTR,
3032 where bitmap B splits congruence class members. DATA is used
3033 as argument of split pair. */
3035 bool
3036 sem_item_optimizer::traverse_congruence_split (congruence_class * const &cls,
3037 bitmap const &b, traverse_split_pair *pair)
3039 sem_item_optimizer *optimizer = pair->optimizer;
3040 const congruence_class *splitter_cls = pair->cls;
3042 /* If counted bits are greater than zero and less than the number of members
3043 a group will be splitted. */
3044 unsigned popcount = bitmap_count_bits (b);
3046 if (popcount > 0 && popcount < cls->members.length ())
3048 congruence_class* newclasses[2] = { new congruence_class (class_id++), new congruence_class (class_id++) };
3050 for (unsigned int i = 0; i < cls->members.length (); i++)
3052 int target = bitmap_bit_p (b, i);
3053 congruence_class *tc = newclasses[target];
3055 add_item_to_class (tc, cls->members[i]);
3058 #ifdef ENABLE_CHECKING
3059 for (unsigned int i = 0; i < 2; i++)
3060 gcc_checking_assert (newclasses[i]->members.length ());
3061 #endif
3063 if (splitter_cls == cls)
3064 optimizer->splitter_class_removed = true;
3066 /* Remove old class from worklist if presented. */
3067 bool in_worklist = cls->in_worklist;
3069 if (in_worklist)
3070 cls->in_worklist = false;
3072 congruence_class_group g;
3073 g.hash = cls->members[0]->get_hash ();
3074 g.type = cls->members[0]->type;
3076 congruence_class_group *slot = optimizer->m_classes.find(&g);
3078 for (unsigned int i = 0; i < slot->classes.length (); i++)
3079 if (slot->classes[i] == cls)
3081 slot->classes.ordered_remove (i);
3082 break;
3085 /* New class will be inserted and integrated to work list. */
3086 for (unsigned int i = 0; i < 2; i++)
3087 optimizer->add_class (newclasses[i]);
3089 /* Two classes replace one, so that increment just by one. */
3090 optimizer->m_classes_count++;
3092 /* If OLD class was presented in the worklist, we remove the class
3093 and replace it will both newly created classes. */
3094 if (in_worklist)
3095 for (unsigned int i = 0; i < 2; i++)
3096 optimizer->worklist_push (newclasses[i]);
3097 else /* Just smaller class is inserted. */
3099 unsigned int smaller_index = newclasses[0]->members.length () <
3100 newclasses[1]->members.length () ?
3101 0 : 1;
3102 optimizer->worklist_push (newclasses[smaller_index]);
3105 if (dump_file && (dump_flags & TDF_DETAILS))
3107 fprintf (dump_file, " congruence class splitted:\n");
3108 cls->dump (dump_file, 4);
3110 fprintf (dump_file, " newly created groups:\n");
3111 for (unsigned int i = 0; i < 2; i++)
3112 newclasses[i]->dump (dump_file, 4);
3115 /* Release class if not presented in work list. */
3116 if (!in_worklist)
3117 delete cls;
3121 return true;
3124 /* Tests if a class CLS used as INDEXth splits any congruence classes.
3125 Bitmap stack BMSTACK is used for bitmap allocation. */
3127 void
3128 sem_item_optimizer::do_congruence_step_for_index (congruence_class *cls,
3129 unsigned int index)
3131 hash_map <congruence_class *, bitmap> split_map;
3133 for (unsigned int i = 0; i < cls->members.length (); i++)
3135 sem_item *item = cls->members[i];
3137 /* Iterate all usages that have INDEX as usage of the item. */
3138 for (unsigned int j = 0; j < item->usages.length (); j++)
3140 sem_usage_pair *usage = item->usages[j];
3142 if (usage->index != index)
3143 continue;
3145 bitmap *slot = split_map.get (usage->item->cls);
3146 bitmap b;
3148 if(!slot)
3150 b = BITMAP_ALLOC (&m_bmstack);
3151 split_map.put (usage->item->cls, b);
3153 else
3154 b = *slot;
3156 #if ENABLE_CHECKING
3157 gcc_checking_assert (usage->item->cls);
3158 gcc_checking_assert (usage->item->index_in_class <
3159 usage->item->cls->members.length ());
3160 #endif
3162 bitmap_set_bit (b, usage->item->index_in_class);
3166 traverse_split_pair pair;
3167 pair.optimizer = this;
3168 pair.cls = cls;
3170 splitter_class_removed = false;
3171 split_map.traverse
3172 <traverse_split_pair *, sem_item_optimizer::traverse_congruence_split> (&pair);
3174 /* Bitmap clean-up. */
3175 split_map.traverse
3176 <traverse_split_pair *, sem_item_optimizer::release_split_map> (NULL);
3179 /* Every usage of a congruence class CLS is a candidate that can split the
3180 collection of classes. Bitmap stack BMSTACK is used for bitmap
3181 allocation. */
3183 void
3184 sem_item_optimizer::do_congruence_step (congruence_class *cls)
3186 bitmap_iterator bi;
3187 unsigned int i;
3189 bitmap usage = BITMAP_ALLOC (&m_bmstack);
3191 for (unsigned int i = 0; i < cls->members.length (); i++)
3192 bitmap_ior_into (usage, cls->members[i]->usage_index_bitmap);
3194 EXECUTE_IF_SET_IN_BITMAP (usage, 0, i, bi)
3196 if (dump_file && (dump_flags & TDF_DETAILS))
3197 fprintf (dump_file, " processing congruence step for class: %u, index: %u\n",
3198 cls->id, i);
3200 do_congruence_step_for_index (cls, i);
3202 if (splitter_class_removed)
3203 break;
3206 BITMAP_FREE (usage);
3209 /* Adds a newly created congruence class CLS to worklist. */
3211 void
3212 sem_item_optimizer::worklist_push (congruence_class *cls)
3214 /* Return if the class CLS is already presented in work list. */
3215 if (cls->in_worklist)
3216 return;
3218 cls->in_worklist = true;
3219 worklist.push_back (cls);
3222 /* Pops a class from worklist. */
3224 congruence_class *
3225 sem_item_optimizer::worklist_pop (void)
3227 congruence_class *cls;
3229 while (!worklist.empty ())
3231 cls = worklist.front ();
3232 worklist.pop_front ();
3233 if (cls->in_worklist)
3235 cls->in_worklist = false;
3237 return cls;
3239 else
3241 /* Work list item was already intended to be removed.
3242 The only reason for doing it is to split a class.
3243 Thus, the class CLS is deleted. */
3244 delete cls;
3248 return NULL;
3251 /* Iterative congruence reduction function. */
3253 void
3254 sem_item_optimizer::process_cong_reduction (void)
3256 for (hash_table<congruence_class_group_hash>::iterator it = m_classes.begin ();
3257 it != m_classes.end (); ++it)
3258 for (unsigned i = 0; i < (*it)->classes.length (); i++)
3259 if ((*it)->classes[i]->is_class_used ())
3260 worklist_push ((*it)->classes[i]);
3262 if (dump_file)
3263 fprintf (dump_file, "Worklist has been filled with: %lu\n",
3264 (unsigned long) worklist.size ());
3266 if (dump_file && (dump_flags & TDF_DETAILS))
3267 fprintf (dump_file, "Congruence class reduction\n");
3269 congruence_class *cls;
3271 /* Process complete congruence reduction. */
3272 while ((cls = worklist_pop ()) != NULL)
3273 do_congruence_step (cls);
3275 /* Subdivide newly created classes according to references. */
3276 unsigned new_classes = subdivide_classes_by_sensitive_refs ();
3278 if (dump_file)
3279 fprintf (dump_file, "Address reference subdivision created: %u "
3280 "new classes.\n", new_classes);
3283 /* Debug function prints all informations about congruence classes. */
3285 void
3286 sem_item_optimizer::dump_cong_classes (void)
3288 if (!dump_file)
3289 return;
3291 fprintf (dump_file,
3292 "Congruence classes: %u (unique hash values: %lu), with total: %u items\n",
3293 m_classes_count, (unsigned long) m_classes.elements(), m_items.length ());
3295 /* Histogram calculation. */
3296 unsigned int max_index = 0;
3297 unsigned int* histogram = XCNEWVEC (unsigned int, m_items.length () + 1);
3299 for (hash_table<congruence_class_group_hash>::iterator it = m_classes.begin ();
3300 it != m_classes.end (); ++it)
3302 for (unsigned i = 0; i < (*it)->classes.length (); i++)
3304 unsigned int c = (*it)->classes[i]->members.length ();
3305 histogram[c]++;
3307 if (c > max_index)
3308 max_index = c;
3311 fprintf (dump_file,
3312 "Class size histogram [num of members]: number of classe number of classess\n");
3314 for (unsigned int i = 0; i <= max_index; i++)
3315 if (histogram[i])
3316 fprintf (dump_file, "[%u]: %u classes\n", i, histogram[i]);
3318 fprintf (dump_file, "\n\n");
3321 if (dump_flags & TDF_DETAILS)
3322 for (hash_table<congruence_class_group_hash>::iterator it = m_classes.begin ();
3323 it != m_classes.end (); ++it)
3325 fprintf (dump_file, " group: with %u classes:\n", (*it)->classes.length ());
3327 for (unsigned i = 0; i < (*it)->classes.length (); i++)
3329 (*it)->classes[i]->dump (dump_file, 4);
3331 if(i < (*it)->classes.length () - 1)
3332 fprintf (dump_file, " ");
3336 free (histogram);
3339 /* After reduction is done, we can declare all items in a group
3340 to be equal. PREV_CLASS_COUNT is start number of classes
3341 before reduction. True is returned if there's a merge operation
3342 processed. */
3344 bool
3345 sem_item_optimizer::merge_classes (unsigned int prev_class_count)
3347 unsigned int item_count = m_items.length ();
3348 unsigned int class_count = m_classes_count;
3349 unsigned int equal_items = item_count - class_count;
3351 unsigned int non_singular_classes_count = 0;
3352 unsigned int non_singular_classes_sum = 0;
3354 bool merged_p = false;
3356 for (hash_table<congruence_class_group_hash>::iterator it = m_classes.begin ();
3357 it != m_classes.end (); ++it)
3358 for (unsigned int i = 0; i < (*it)->classes.length (); i++)
3360 congruence_class *c = (*it)->classes[i];
3361 if (c->members.length () > 1)
3363 non_singular_classes_count++;
3364 non_singular_classes_sum += c->members.length ();
3368 if (dump_file)
3370 fprintf (dump_file, "\nItem count: %u\n", item_count);
3371 fprintf (dump_file, "Congruent classes before: %u, after: %u\n",
3372 prev_class_count, class_count);
3373 fprintf (dump_file, "Average class size before: %.2f, after: %.2f\n",
3374 prev_class_count ? 1.0f * item_count / prev_class_count : 0.0f,
3375 class_count ? 1.0f * item_count / class_count : 0.0f);
3376 fprintf (dump_file, "Average non-singular class size: %.2f, count: %u\n",
3377 non_singular_classes_count ? 1.0f * non_singular_classes_sum /
3378 non_singular_classes_count : 0.0f,
3379 non_singular_classes_count);
3380 fprintf (dump_file, "Equal symbols: %u\n", equal_items);
3381 fprintf (dump_file, "Fraction of visited symbols: %.2f%%\n\n",
3382 item_count ? 100.0f * equal_items / item_count : 0.0f);
3385 for (hash_table<congruence_class_group_hash>::iterator it = m_classes.begin ();
3386 it != m_classes.end (); ++it)
3387 for (unsigned int i = 0; i < (*it)->classes.length (); i++)
3389 congruence_class *c = (*it)->classes[i];
3391 if (c->members.length () == 1)
3392 continue;
3394 gcc_assert (c->members.length ());
3396 sem_item *source = c->members[0];
3398 for (unsigned int j = 1; j < c->members.length (); j++)
3400 sem_item *alias = c->members[j];
3402 if (dump_file)
3404 fprintf (dump_file, "Semantic equality hit:%s->%s\n",
3405 xstrdup_for_dump (source->node->name ()),
3406 xstrdup_for_dump (alias->node->name ()));
3407 fprintf (dump_file, "Assembler symbol names:%s->%s\n",
3408 xstrdup_for_dump (source->node->asm_name ()),
3409 xstrdup_for_dump (alias->node->asm_name ()));
3412 if (lookup_attribute ("no_icf", DECL_ATTRIBUTES (alias->decl)))
3414 if (dump_file)
3415 fprintf (dump_file,
3416 "Merge operation is skipped due to no_icf "
3417 "attribute.\n\n");
3419 continue;
3422 if (dump_file && (dump_flags & TDF_DETAILS))
3424 source->dump_to_file (dump_file);
3425 alias->dump_to_file (dump_file);
3428 if (dbg_cnt (merged_ipa_icf))
3429 merged_p |= source->merge (alias);
3433 return merged_p;
3436 /* Dump function prints all class members to a FILE with an INDENT. */
3438 void
3439 congruence_class::dump (FILE *file, unsigned int indent) const
3441 FPRINTF_SPACES (file, indent, "class with id: %u, hash: %u, items: %u\n",
3442 id, members[0]->get_hash (), members.length ());
3444 FPUTS_SPACES (file, indent + 2, "");
3445 for (unsigned i = 0; i < members.length (); i++)
3446 fprintf (file, "%s(%p/%u) ", members[i]->node->asm_name (),
3447 (void *) members[i]->decl,
3448 members[i]->node->order);
3450 fprintf (file, "\n");
3453 /* Returns true if there's a member that is used from another group. */
3455 bool
3456 congruence_class::is_class_used (void)
3458 for (unsigned int i = 0; i < members.length (); i++)
3459 if (members[i]->usages.length ())
3460 return true;
3462 return false;
3465 /* Generate pass summary for IPA ICF pass. */
3467 static void
3468 ipa_icf_generate_summary (void)
3470 if (!optimizer)
3471 optimizer = new sem_item_optimizer ();
3473 optimizer->register_hooks ();
3474 optimizer->parse_funcs_and_vars ();
3477 /* Write pass summary for IPA ICF pass. */
3479 static void
3480 ipa_icf_write_summary (void)
3482 gcc_assert (optimizer);
3484 optimizer->write_summary ();
3487 /* Read pass summary for IPA ICF pass. */
3489 static void
3490 ipa_icf_read_summary (void)
3492 if (!optimizer)
3493 optimizer = new sem_item_optimizer ();
3495 optimizer->read_summary ();
3496 optimizer->register_hooks ();
3499 /* Semantic equality exection function. */
3501 static unsigned int
3502 ipa_icf_driver (void)
3504 gcc_assert (optimizer);
3506 bool merged_p = optimizer->execute ();
3508 delete optimizer;
3509 optimizer = NULL;
3511 return merged_p ? TODO_remove_functions : 0;
3514 const pass_data pass_data_ipa_icf =
3516 IPA_PASS, /* type */
3517 "icf", /* name */
3518 OPTGROUP_IPA, /* optinfo_flags */
3519 TV_IPA_ICF, /* tv_id */
3520 0, /* properties_required */
3521 0, /* properties_provided */
3522 0, /* properties_destroyed */
3523 0, /* todo_flags_start */
3524 0, /* todo_flags_finish */
3527 class pass_ipa_icf : public ipa_opt_pass_d
3529 public:
3530 pass_ipa_icf (gcc::context *ctxt)
3531 : ipa_opt_pass_d (pass_data_ipa_icf, ctxt,
3532 ipa_icf_generate_summary, /* generate_summary */
3533 ipa_icf_write_summary, /* write_summary */
3534 ipa_icf_read_summary, /* read_summary */
3535 NULL, /*
3536 write_optimization_summary */
3537 NULL, /*
3538 read_optimization_summary */
3539 NULL, /* stmt_fixup */
3540 0, /* function_transform_todo_flags_start */
3541 NULL, /* function_transform */
3542 NULL) /* variable_transform */
3545 /* opt_pass methods: */
3546 virtual bool gate (function *)
3548 return in_lto_p || flag_ipa_icf_variables || flag_ipa_icf_functions;
3551 virtual unsigned int execute (function *)
3553 return ipa_icf_driver();
3555 }; // class pass_ipa_icf
3557 } // ipa_icf namespace
3559 ipa_opt_pass_d *
3560 make_pass_ipa_icf (gcc::context *ctxt)
3562 return new ipa_icf::pass_ipa_icf (ctxt);