re PR target/54131 (ICE building 416.gamess, reload_cse_simplify_operands)
[official-gcc.git] / gcc / ipa-cp.c
blob332e623f41a1edce90275d2f50f08fc1bcba9a56
1 /* Interprocedural constant propagation
2 Copyright (C) 2005-2013 Free Software Foundation, Inc.
4 Contributed by Razya Ladelsky <RAZYA@il.ibm.com> and Martin Jambor
5 <mjambor@suse.cz>
7 This file is part of GCC.
9 GCC is free software; you can redistribute it and/or modify it under
10 the terms of the GNU General Public License as published by the Free
11 Software Foundation; either version 3, or (at your option) any later
12 version.
14 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
15 WARRANTY; without even the implied warranty of MERCHANTABILITY or
16 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
17 for more details.
19 You should have received a copy of the GNU General Public License
20 along with GCC; see the file COPYING3. If not see
21 <http://www.gnu.org/licenses/>. */
23 /* Interprocedural constant propagation (IPA-CP).
25 The goal of this transformation is to
27 1) discover functions which are always invoked with some arguments with the
28 same known constant values and modify the functions so that the
29 subsequent optimizations can take advantage of the knowledge, and
31 2) partial specialization - create specialized versions of functions
32 transformed in this way if some parameters are known constants only in
33 certain contexts but the estimated tradeoff between speedup and cost size
34 is deemed good.
36 The algorithm also propagates types and attempts to perform type based
37 devirtualization. Types are propagated much like constants.
39 The algorithm basically consists of three stages. In the first, functions
40 are analyzed one at a time and jump functions are constructed for all known
41 call-sites. In the second phase, the pass propagates information from the
42 jump functions across the call to reveal what values are available at what
43 call sites, performs estimations of effects of known values on functions and
44 their callees, and finally decides what specialized extra versions should be
45 created. In the third, the special versions materialize and appropriate
46 calls are redirected.
48 The algorithm used is to a certain extent based on "Interprocedural Constant
49 Propagation", by David Callahan, Keith D Cooper, Ken Kennedy, Linda Torczon,
50 Comp86, pg 152-161 and "A Methodology for Procedure Cloning" by Keith D
51 Cooper, Mary W. Hall, and Ken Kennedy.
54 First stage - intraprocedural analysis
55 =======================================
57 This phase computes jump_function and modification flags.
59 A jump function for a call-site represents the values passed as an actual
60 arguments of a given call-site. In principle, there are three types of
61 values:
63 Pass through - the caller's formal parameter is passed as an actual
64 argument, plus an operation on it can be performed.
65 Constant - a constant is passed as an actual argument.
66 Unknown - neither of the above.
68 All jump function types are described in detail in ipa-prop.h, together with
69 the data structures that represent them and methods of accessing them.
71 ipcp_generate_summary() is the main function of the first stage.
73 Second stage - interprocedural analysis
74 ========================================
76 This stage is itself divided into two phases. In the first, we propagate
77 known values over the call graph, in the second, we make cloning decisions.
78 It uses a different algorithm than the original Callahan's paper.
80 First, we traverse the functions topologically from callers to callees and,
81 for each strongly connected component (SCC), we propagate constants
82 according to previously computed jump functions. We also record what known
83 values depend on other known values and estimate local effects. Finally, we
84 propagate cumulative information about these effects from dependent values
85 to those on which they depend.
87 Second, we again traverse the call graph in the same topological order and
88 make clones for functions which we know are called with the same values in
89 all contexts and decide about extra specialized clones of functions just for
90 some contexts - these decisions are based on both local estimates and
91 cumulative estimates propagated from callees.
93 ipcp_propagate_stage() and ipcp_decision_stage() together constitute the
94 third stage.
96 Third phase - materialization of clones, call statement updates.
97 ============================================
99 This stage is currently performed by call graph code (mainly in cgraphunit.c
100 and tree-inline.c) according to instructions inserted to the call graph by
101 the second stage. */
103 #include "config.h"
104 #include "system.h"
105 #include "coretypes.h"
106 #include "tree.h"
107 #include "target.h"
108 #include "gimple.h"
109 #include "cgraph.h"
110 #include "ipa-prop.h"
111 #include "tree-flow.h"
112 #include "tree-pass.h"
113 #include "flags.h"
114 #include "diagnostic.h"
115 #include "tree-pretty-print.h"
116 #include "tree-inline.h"
117 #include "params.h"
118 #include "ipa-inline.h"
119 #include "ipa-utils.h"
121 struct ipcp_value;
123 /* Describes a particular source for an IPA-CP value. */
125 struct ipcp_value_source
127 /* Aggregate offset of the source, negative if the source is scalar value of
128 the argument itself. */
129 HOST_WIDE_INT offset;
130 /* The incoming edge that brought the value. */
131 struct cgraph_edge *cs;
132 /* If the jump function that resulted into his value was a pass-through or an
133 ancestor, this is the ipcp_value of the caller from which the described
134 value has been derived. Otherwise it is NULL. */
135 struct ipcp_value *val;
136 /* Next pointer in a linked list of sources of a value. */
137 struct ipcp_value_source *next;
138 /* If the jump function that resulted into his value was a pass-through or an
139 ancestor, this is the index of the parameter of the caller the jump
140 function references. */
141 int index;
144 /* Describes one particular value stored in struct ipcp_lattice. */
146 struct ipcp_value
148 /* The actual value for the given parameter. This is either an IPA invariant
149 or a TREE_BINFO describing a type that can be used for
150 devirtualization. */
151 tree value;
152 /* The list of sources from which this value originates. */
153 struct ipcp_value_source *sources;
154 /* Next pointers in a linked list of all values in a lattice. */
155 struct ipcp_value *next;
156 /* Next pointers in a linked list of values in a strongly connected component
157 of values. */
158 struct ipcp_value *scc_next;
159 /* Next pointers in a linked list of SCCs of values sorted topologically
160 according their sources. */
161 struct ipcp_value *topo_next;
162 /* A specialized node created for this value, NULL if none has been (so far)
163 created. */
164 struct cgraph_node *spec_node;
165 /* Depth first search number and low link for topological sorting of
166 values. */
167 int dfs, low_link;
168 /* Time benefit and size cost that specializing the function for this value
169 would bring about in this function alone. */
170 int local_time_benefit, local_size_cost;
171 /* Time benefit and size cost that specializing the function for this value
172 can bring about in it's callees (transitively). */
173 int prop_time_benefit, prop_size_cost;
174 /* True if this valye is currently on the topo-sort stack. */
175 bool on_stack;
178 /* Lattice describing potential values of a formal parameter of a function, or
179 a part of an aggreagate. TOP is represented by a lattice with zero values
180 and with contains_variable and bottom flags cleared. BOTTOM is represented
181 by a lattice with the bottom flag set. In that case, values and
182 contains_variable flag should be disregarded. */
184 struct ipcp_lattice
186 /* The list of known values and types in this lattice. Note that values are
187 not deallocated if a lattice is set to bottom because there may be value
188 sources referencing them. */
189 struct ipcp_value *values;
190 /* Number of known values and types in this lattice. */
191 int values_count;
192 /* The lattice contains a variable component (in addition to values). */
193 bool contains_variable;
194 /* The value of the lattice is bottom (i.e. variable and unusable for any
195 propagation). */
196 bool bottom;
199 /* Lattice with an offset to describe a part of an aggregate. */
201 struct ipcp_agg_lattice : public ipcp_lattice
203 /* Offset that is being described by this lattice. */
204 HOST_WIDE_INT offset;
205 /* Size so that we don't have to re-compute it every time we traverse the
206 list. Must correspond to TYPE_SIZE of all lat values. */
207 HOST_WIDE_INT size;
208 /* Next element of the linked list. */
209 struct ipcp_agg_lattice *next;
212 /* Structure containing lattices for a parameter itself and for pieces of
213 aggregates that are passed in the parameter or by a reference in a parameter
214 plus some other useful flags. */
216 struct ipcp_param_lattices
218 /* Lattice describing the value of the parameter itself. */
219 struct ipcp_lattice itself;
220 /* Lattices describing aggregate parts. */
221 struct ipcp_agg_lattice *aggs;
222 /* Number of aggregate lattices */
223 int aggs_count;
224 /* True if aggregate data were passed by reference (as opposed to by
225 value). */
226 bool aggs_by_ref;
227 /* All aggregate lattices contain a variable component (in addition to
228 values). */
229 bool aggs_contain_variable;
230 /* The value of all aggregate lattices is bottom (i.e. variable and unusable
231 for any propagation). */
232 bool aggs_bottom;
234 /* There is a virtual call based on this parameter. */
235 bool virt_call;
238 /* Allocation pools for values and their sources in ipa-cp. */
240 alloc_pool ipcp_values_pool;
241 alloc_pool ipcp_sources_pool;
242 alloc_pool ipcp_agg_lattice_pool;
244 /* Maximal count found in program. */
246 static gcov_type max_count;
248 /* Original overall size of the program. */
250 static long overall_size, max_new_size;
252 /* Head of the linked list of topologically sorted values. */
254 static struct ipcp_value *values_topo;
256 /* Return the param lattices structure corresponding to the Ith formal
257 parameter of the function described by INFO. */
258 static inline struct ipcp_param_lattices *
259 ipa_get_parm_lattices (struct ipa_node_params *info, int i)
261 gcc_assert (i >= 0 && i < ipa_get_param_count (info));
262 gcc_checking_assert (!info->ipcp_orig_node);
263 gcc_checking_assert (info->lattices);
264 return &(info->lattices[i]);
267 /* Return the lattice corresponding to the scalar value of the Ith formal
268 parameter of the function described by INFO. */
269 static inline struct ipcp_lattice *
270 ipa_get_scalar_lat (struct ipa_node_params *info, int i)
272 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
273 return &plats->itself;
276 /* Return whether LAT is a lattice with a single constant and without an
277 undefined value. */
279 static inline bool
280 ipa_lat_is_single_const (struct ipcp_lattice *lat)
282 if (lat->bottom
283 || lat->contains_variable
284 || lat->values_count != 1)
285 return false;
286 else
287 return true;
290 /* Return true iff the CS is an edge within a strongly connected component as
291 computed by ipa_reduced_postorder. */
293 static inline bool
294 edge_within_scc (struct cgraph_edge *cs)
296 struct ipa_dfs_info *caller_dfs = (struct ipa_dfs_info *) cs->caller->symbol.aux;
297 struct ipa_dfs_info *callee_dfs;
298 struct cgraph_node *callee = cgraph_function_node (cs->callee, NULL);
300 callee_dfs = (struct ipa_dfs_info *) callee->symbol.aux;
301 return (caller_dfs
302 && callee_dfs
303 && caller_dfs->scc_no == callee_dfs->scc_no);
306 /* Print V which is extracted from a value in a lattice to F. */
308 static void
309 print_ipcp_constant_value (FILE * f, tree v)
311 if (TREE_CODE (v) == TREE_BINFO)
313 fprintf (f, "BINFO ");
314 print_generic_expr (f, BINFO_TYPE (v), 0);
316 else if (TREE_CODE (v) == ADDR_EXPR
317 && TREE_CODE (TREE_OPERAND (v, 0)) == CONST_DECL)
319 fprintf (f, "& ");
320 print_generic_expr (f, DECL_INITIAL (TREE_OPERAND (v, 0)), 0);
322 else
323 print_generic_expr (f, v, 0);
326 /* Print a lattice LAT to F. */
328 static void
329 print_lattice (FILE * f, struct ipcp_lattice *lat,
330 bool dump_sources, bool dump_benefits)
332 struct ipcp_value *val;
333 bool prev = false;
335 if (lat->bottom)
337 fprintf (f, "BOTTOM\n");
338 return;
341 if (!lat->values_count && !lat->contains_variable)
343 fprintf (f, "TOP\n");
344 return;
347 if (lat->contains_variable)
349 fprintf (f, "VARIABLE");
350 prev = true;
351 if (dump_benefits)
352 fprintf (f, "\n");
355 for (val = lat->values; val; val = val->next)
357 if (dump_benefits && prev)
358 fprintf (f, " ");
359 else if (!dump_benefits && prev)
360 fprintf (f, ", ");
361 else
362 prev = true;
364 print_ipcp_constant_value (f, val->value);
366 if (dump_sources)
368 struct ipcp_value_source *s;
370 fprintf (f, " [from:");
371 for (s = val->sources; s; s = s->next)
372 fprintf (f, " %i(%i)", s->cs->caller->uid,s->cs->frequency);
373 fprintf (f, "]");
376 if (dump_benefits)
377 fprintf (f, " [loc_time: %i, loc_size: %i, "
378 "prop_time: %i, prop_size: %i]\n",
379 val->local_time_benefit, val->local_size_cost,
380 val->prop_time_benefit, val->prop_size_cost);
382 if (!dump_benefits)
383 fprintf (f, "\n");
386 /* Print all ipcp_lattices of all functions to F. */
388 static void
389 print_all_lattices (FILE * f, bool dump_sources, bool dump_benefits)
391 struct cgraph_node *node;
392 int i, count;
394 fprintf (f, "\nLattices:\n");
395 FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node)
397 struct ipa_node_params *info;
399 info = IPA_NODE_REF (node);
400 fprintf (f, " Node: %s/%i:\n", cgraph_node_name (node), node->uid);
401 count = ipa_get_param_count (info);
402 for (i = 0; i < count; i++)
404 struct ipcp_agg_lattice *aglat;
405 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
406 fprintf (f, " param [%d]: ", i);
407 print_lattice (f, &plats->itself, dump_sources, dump_benefits);
409 if (plats->virt_call)
410 fprintf (f, " virt_call flag set\n");
412 if (plats->aggs_bottom)
414 fprintf (f, " AGGS BOTTOM\n");
415 continue;
417 if (plats->aggs_contain_variable)
418 fprintf (f, " AGGS VARIABLE\n");
419 for (aglat = plats->aggs; aglat; aglat = aglat->next)
421 fprintf (f, " %soffset " HOST_WIDE_INT_PRINT_DEC ": ",
422 plats->aggs_by_ref ? "ref " : "", aglat->offset);
423 print_lattice (f, aglat, dump_sources, dump_benefits);
429 /* Determine whether it is at all technically possible to create clones of NODE
430 and store this information in the ipa_node_params structure associated
431 with NODE. */
433 static void
434 determine_versionability (struct cgraph_node *node)
436 const char *reason = NULL;
438 /* There are a number of generic reasons functions cannot be versioned. We
439 also cannot remove parameters if there are type attributes such as fnspec
440 present. */
441 if (node->alias || node->thunk.thunk_p)
442 reason = "alias or thunk";
443 else if (!node->local.versionable)
444 reason = "not a tree_versionable_function";
445 else if (cgraph_function_body_availability (node) <= AVAIL_OVERWRITABLE)
446 reason = "insufficient body availability";
448 if (reason && dump_file && !node->alias && !node->thunk.thunk_p)
449 fprintf (dump_file, "Function %s/%i is not versionable, reason: %s.\n",
450 cgraph_node_name (node), node->uid, reason);
452 node->local.versionable = (reason == NULL);
455 /* Return true if it is at all technically possible to create clones of a
456 NODE. */
458 static bool
459 ipcp_versionable_function_p (struct cgraph_node *node)
461 return node->local.versionable;
464 /* Structure holding accumulated information about callers of a node. */
466 struct caller_statistics
468 gcov_type count_sum;
469 int n_calls, n_hot_calls, freq_sum;
472 /* Initialize fields of STAT to zeroes. */
474 static inline void
475 init_caller_stats (struct caller_statistics *stats)
477 stats->count_sum = 0;
478 stats->n_calls = 0;
479 stats->n_hot_calls = 0;
480 stats->freq_sum = 0;
483 /* Worker callback of cgraph_for_node_and_aliases accumulating statistics of
484 non-thunk incoming edges to NODE. */
486 static bool
487 gather_caller_stats (struct cgraph_node *node, void *data)
489 struct caller_statistics *stats = (struct caller_statistics *) data;
490 struct cgraph_edge *cs;
492 for (cs = node->callers; cs; cs = cs->next_caller)
493 if (cs->caller->thunk.thunk_p)
494 cgraph_for_node_and_aliases (cs->caller, gather_caller_stats,
495 stats, false);
496 else
498 stats->count_sum += cs->count;
499 stats->freq_sum += cs->frequency;
500 stats->n_calls++;
501 if (cgraph_maybe_hot_edge_p (cs))
502 stats->n_hot_calls ++;
504 return false;
508 /* Return true if this NODE is viable candidate for cloning. */
510 static bool
511 ipcp_cloning_candidate_p (struct cgraph_node *node)
513 struct caller_statistics stats;
515 gcc_checking_assert (cgraph_function_with_gimple_body_p (node));
517 if (!flag_ipa_cp_clone)
519 if (dump_file)
520 fprintf (dump_file, "Not considering %s for cloning; "
521 "-fipa-cp-clone disabled.\n",
522 cgraph_node_name (node));
523 return false;
526 if (!optimize_function_for_speed_p (DECL_STRUCT_FUNCTION (node->symbol.decl)))
528 if (dump_file)
529 fprintf (dump_file, "Not considering %s for cloning; "
530 "optimizing it for size.\n",
531 cgraph_node_name (node));
532 return false;
535 init_caller_stats (&stats);
536 cgraph_for_node_and_aliases (node, gather_caller_stats, &stats, false);
538 if (inline_summary (node)->self_size < stats.n_calls)
540 if (dump_file)
541 fprintf (dump_file, "Considering %s for cloning; code might shrink.\n",
542 cgraph_node_name (node));
543 return true;
546 /* When profile is available and function is hot, propagate into it even if
547 calls seems cold; constant propagation can improve function's speed
548 significantly. */
549 if (max_count)
551 if (stats.count_sum > node->count * 90 / 100)
553 if (dump_file)
554 fprintf (dump_file, "Considering %s for cloning; "
555 "usually called directly.\n",
556 cgraph_node_name (node));
557 return true;
560 if (!stats.n_hot_calls)
562 if (dump_file)
563 fprintf (dump_file, "Not considering %s for cloning; no hot calls.\n",
564 cgraph_node_name (node));
565 return false;
567 if (dump_file)
568 fprintf (dump_file, "Considering %s for cloning.\n",
569 cgraph_node_name (node));
570 return true;
573 /* Arrays representing a topological ordering of call graph nodes and a stack
574 of noes used during constant propagation. */
576 struct topo_info
578 struct cgraph_node **order;
579 struct cgraph_node **stack;
580 int nnodes, stack_top;
583 /* Allocate the arrays in TOPO and topologically sort the nodes into order. */
585 static void
586 build_toporder_info (struct topo_info *topo)
588 topo->order = XCNEWVEC (struct cgraph_node *, cgraph_n_nodes);
589 topo->stack = XCNEWVEC (struct cgraph_node *, cgraph_n_nodes);
590 topo->stack_top = 0;
591 topo->nnodes = ipa_reduced_postorder (topo->order, true, true, NULL);
594 /* Free information about strongly connected components and the arrays in
595 TOPO. */
597 static void
598 free_toporder_info (struct topo_info *topo)
600 ipa_free_postorder_info ();
601 free (topo->order);
602 free (topo->stack);
605 /* Add NODE to the stack in TOPO, unless it is already there. */
607 static inline void
608 push_node_to_stack (struct topo_info *topo, struct cgraph_node *node)
610 struct ipa_node_params *info = IPA_NODE_REF (node);
611 if (info->node_enqueued)
612 return;
613 info->node_enqueued = 1;
614 topo->stack[topo->stack_top++] = node;
617 /* Pop a node from the stack in TOPO and return it or return NULL if the stack
618 is empty. */
620 static struct cgraph_node *
621 pop_node_from_stack (struct topo_info *topo)
623 if (topo->stack_top)
625 struct cgraph_node *node;
626 topo->stack_top--;
627 node = topo->stack[topo->stack_top];
628 IPA_NODE_REF (node)->node_enqueued = 0;
629 return node;
631 else
632 return NULL;
635 /* Set lattice LAT to bottom and return true if it previously was not set as
636 such. */
638 static inline bool
639 set_lattice_to_bottom (struct ipcp_lattice *lat)
641 bool ret = !lat->bottom;
642 lat->bottom = true;
643 return ret;
646 /* Mark lattice as containing an unknown value and return true if it previously
647 was not marked as such. */
649 static inline bool
650 set_lattice_contains_variable (struct ipcp_lattice *lat)
652 bool ret = !lat->contains_variable;
653 lat->contains_variable = true;
654 return ret;
657 /* Set all aggegate lattices in PLATS to bottom and return true if they were
658 not previously set as such. */
660 static inline bool
661 set_agg_lats_to_bottom (struct ipcp_param_lattices *plats)
663 bool ret = !plats->aggs_bottom;
664 plats->aggs_bottom = true;
665 return ret;
668 /* Mark all aggegate lattices in PLATS as containing an unknown value and
669 return true if they were not previously marked as such. */
671 static inline bool
672 set_agg_lats_contain_variable (struct ipcp_param_lattices *plats)
674 bool ret = !plats->aggs_contain_variable;
675 plats->aggs_contain_variable = true;
676 return ret;
679 /* Mark bot aggregate and scalar lattices as containing an unknown variable,
680 return true is any of them has not been marked as such so far. */
682 static inline bool
683 set_all_contains_variable (struct ipcp_param_lattices *plats)
685 bool ret = !plats->itself.contains_variable || !plats->aggs_contain_variable;
686 plats->itself.contains_variable = true;
687 plats->aggs_contain_variable = true;
688 return ret;
691 /* Initialize ipcp_lattices. */
693 static void
694 initialize_node_lattices (struct cgraph_node *node)
696 struct ipa_node_params *info = IPA_NODE_REF (node);
697 struct cgraph_edge *ie;
698 bool disable = false, variable = false;
699 int i;
701 gcc_checking_assert (cgraph_function_with_gimple_body_p (node));
702 if (!node->local.local)
704 /* When cloning is allowed, we can assume that externally visible
705 functions are not called. We will compensate this by cloning
706 later. */
707 if (ipcp_versionable_function_p (node)
708 && ipcp_cloning_candidate_p (node))
709 variable = true;
710 else
711 disable = true;
714 if (disable || variable)
716 for (i = 0; i < ipa_get_param_count (info) ; i++)
718 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
719 if (disable)
721 set_lattice_to_bottom (&plats->itself);
722 set_agg_lats_to_bottom (plats);
724 else
725 set_all_contains_variable (plats);
727 if (dump_file && (dump_flags & TDF_DETAILS)
728 && node->alias && node->thunk.thunk_p)
729 fprintf (dump_file, "Marking all lattices of %s/%i as %s\n",
730 cgraph_node_name (node), node->uid,
731 disable ? "BOTTOM" : "VARIABLE");
734 for (ie = node->indirect_calls; ie; ie = ie->next_callee)
735 if (ie->indirect_info->polymorphic)
737 gcc_checking_assert (ie->indirect_info->param_index >= 0);
738 ipa_get_parm_lattices (info,
739 ie->indirect_info->param_index)->virt_call = 1;
743 /* Return the result of a (possibly arithmetic) pass through jump function
744 JFUNC on the constant value INPUT. Return NULL_TREE if that cannot be
745 determined or itself is considered an interprocedural invariant. */
747 static tree
748 ipa_get_jf_pass_through_result (struct ipa_jump_func *jfunc, tree input)
750 tree restype, res;
752 if (ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
753 return input;
754 else if (TREE_CODE (input) == TREE_BINFO)
755 return NULL_TREE;
757 gcc_checking_assert (is_gimple_ip_invariant (input));
758 if (TREE_CODE_CLASS (ipa_get_jf_pass_through_operation (jfunc))
759 == tcc_comparison)
760 restype = boolean_type_node;
761 else
762 restype = TREE_TYPE (input);
763 res = fold_binary (ipa_get_jf_pass_through_operation (jfunc), restype,
764 input, ipa_get_jf_pass_through_operand (jfunc));
766 if (res && !is_gimple_ip_invariant (res))
767 return NULL_TREE;
769 return res;
772 /* Return the result of an ancestor jump function JFUNC on the constant value
773 INPUT. Return NULL_TREE if that cannot be determined. */
775 static tree
776 ipa_get_jf_ancestor_result (struct ipa_jump_func *jfunc, tree input)
778 if (TREE_CODE (input) == TREE_BINFO)
779 return get_binfo_at_offset (input,
780 ipa_get_jf_ancestor_offset (jfunc),
781 ipa_get_jf_ancestor_type (jfunc));
782 else if (TREE_CODE (input) == ADDR_EXPR)
784 tree t = TREE_OPERAND (input, 0);
785 t = build_ref_for_offset (EXPR_LOCATION (t), t,
786 ipa_get_jf_ancestor_offset (jfunc),
787 ipa_get_jf_ancestor_type (jfunc), NULL, false);
788 return build_fold_addr_expr (t);
790 else
791 return NULL_TREE;
794 /* Extract the acual BINFO being described by JFUNC which must be a known type
795 jump function. */
797 static tree
798 ipa_value_from_known_type_jfunc (struct ipa_jump_func *jfunc)
800 tree base_binfo = TYPE_BINFO (ipa_get_jf_known_type_base_type (jfunc));
801 if (!base_binfo)
802 return NULL_TREE;
803 return get_binfo_at_offset (base_binfo,
804 ipa_get_jf_known_type_offset (jfunc),
805 ipa_get_jf_known_type_component_type (jfunc));
808 /* Determine whether JFUNC evaluates to a known value (that is either a
809 constant or a binfo) and if so, return it. Otherwise return NULL. INFO
810 describes the caller node so that pass-through jump functions can be
811 evaluated. */
813 tree
814 ipa_value_from_jfunc (struct ipa_node_params *info, struct ipa_jump_func *jfunc)
816 if (jfunc->type == IPA_JF_CONST)
817 return ipa_get_jf_constant (jfunc);
818 else if (jfunc->type == IPA_JF_KNOWN_TYPE)
819 return ipa_value_from_known_type_jfunc (jfunc);
820 else if (jfunc->type == IPA_JF_PASS_THROUGH
821 || jfunc->type == IPA_JF_ANCESTOR)
823 tree input;
824 int idx;
826 if (jfunc->type == IPA_JF_PASS_THROUGH)
827 idx = ipa_get_jf_pass_through_formal_id (jfunc);
828 else
829 idx = ipa_get_jf_ancestor_formal_id (jfunc);
831 if (info->ipcp_orig_node)
832 input = info->known_vals[idx];
833 else
835 struct ipcp_lattice *lat;
837 if (!info->lattices)
839 gcc_checking_assert (!flag_ipa_cp);
840 return NULL_TREE;
842 lat = ipa_get_scalar_lat (info, idx);
843 if (!ipa_lat_is_single_const (lat))
844 return NULL_TREE;
845 input = lat->values->value;
848 if (!input)
849 return NULL_TREE;
851 if (jfunc->type == IPA_JF_PASS_THROUGH)
852 return ipa_get_jf_pass_through_result (jfunc, input);
853 else
854 return ipa_get_jf_ancestor_result (jfunc, input);
856 else
857 return NULL_TREE;
861 /* If checking is enabled, verify that no lattice is in the TOP state, i.e. not
862 bottom, not containing a variable component and without any known value at
863 the same time. */
865 DEBUG_FUNCTION void
866 ipcp_verify_propagated_values (void)
868 struct cgraph_node *node;
870 FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node)
872 struct ipa_node_params *info = IPA_NODE_REF (node);
873 int i, count = ipa_get_param_count (info);
875 for (i = 0; i < count; i++)
877 struct ipcp_lattice *lat = ipa_get_scalar_lat (info, i);
879 if (!lat->bottom
880 && !lat->contains_variable
881 && lat->values_count == 0)
883 if (dump_file)
885 fprintf (dump_file, "\nIPA lattices after constant "
886 "propagation:\n");
887 print_all_lattices (dump_file, true, false);
890 gcc_unreachable ();
896 /* Return true iff X and Y should be considered equal values by IPA-CP. */
898 static bool
899 values_equal_for_ipcp_p (tree x, tree y)
901 gcc_checking_assert (x != NULL_TREE && y != NULL_TREE);
903 if (x == y)
904 return true;
906 if (TREE_CODE (x) == TREE_BINFO || TREE_CODE (y) == TREE_BINFO)
907 return false;
909 if (TREE_CODE (x) == ADDR_EXPR
910 && TREE_CODE (y) == ADDR_EXPR
911 && TREE_CODE (TREE_OPERAND (x, 0)) == CONST_DECL
912 && TREE_CODE (TREE_OPERAND (y, 0)) == CONST_DECL)
913 return operand_equal_p (DECL_INITIAL (TREE_OPERAND (x, 0)),
914 DECL_INITIAL (TREE_OPERAND (y, 0)), 0);
915 else
916 return operand_equal_p (x, y, 0);
919 /* Add a new value source to VAL, marking that a value comes from edge CS and
920 (if the underlying jump function is a pass-through or an ancestor one) from
921 a caller value SRC_VAL of a caller parameter described by SRC_INDEX. OFFSET
922 is negative if the source was the scalar value of the parameter itself or
923 the offset within an aggregate. */
925 static void
926 add_value_source (struct ipcp_value *val, struct cgraph_edge *cs,
927 struct ipcp_value *src_val, int src_idx, HOST_WIDE_INT offset)
929 struct ipcp_value_source *src;
931 src = (struct ipcp_value_source *) pool_alloc (ipcp_sources_pool);
932 src->offset = offset;
933 src->cs = cs;
934 src->val = src_val;
935 src->index = src_idx;
937 src->next = val->sources;
938 val->sources = src;
941 /* Try to add NEWVAL to LAT, potentially creating a new struct ipcp_value for
942 it. CS, SRC_VAL SRC_INDEX and OFFSET are meant for add_value_source and
943 have the same meaning. */
945 static bool
946 add_value_to_lattice (struct ipcp_lattice *lat, tree newval,
947 struct cgraph_edge *cs, struct ipcp_value *src_val,
948 int src_idx, HOST_WIDE_INT offset)
950 struct ipcp_value *val;
952 if (lat->bottom)
953 return false;
955 for (val = lat->values; val; val = val->next)
956 if (values_equal_for_ipcp_p (val->value, newval))
958 if (edge_within_scc (cs))
960 struct ipcp_value_source *s;
961 for (s = val->sources; s ; s = s->next)
962 if (s->cs == cs)
963 break;
964 if (s)
965 return false;
968 add_value_source (val, cs, src_val, src_idx, offset);
969 return false;
972 if (lat->values_count == PARAM_VALUE (PARAM_IPA_CP_VALUE_LIST_SIZE))
974 /* We can only free sources, not the values themselves, because sources
975 of other values in this this SCC might point to them. */
976 for (val = lat->values; val; val = val->next)
978 while (val->sources)
980 struct ipcp_value_source *src = val->sources;
981 val->sources = src->next;
982 pool_free (ipcp_sources_pool, src);
986 lat->values = NULL;
987 return set_lattice_to_bottom (lat);
990 lat->values_count++;
991 val = (struct ipcp_value *) pool_alloc (ipcp_values_pool);
992 memset (val, 0, sizeof (*val));
994 add_value_source (val, cs, src_val, src_idx, offset);
995 val->value = newval;
996 val->next = lat->values;
997 lat->values = val;
998 return true;
1001 /* Like above but passes a special value of offset to distinguish that the
1002 origin is the scalar value of the parameter rather than a part of an
1003 aggregate. */
1005 static inline bool
1006 add_scalar_value_to_lattice (struct ipcp_lattice *lat, tree newval,
1007 struct cgraph_edge *cs,
1008 struct ipcp_value *src_val, int src_idx)
1010 return add_value_to_lattice (lat, newval, cs, src_val, src_idx, -1);
1013 /* Propagate values through a pass-through jump function JFUNC associated with
1014 edge CS, taking values from SRC_LAT and putting them into DEST_LAT. SRC_IDX
1015 is the index of the source parameter. */
1017 static bool
1018 propagate_vals_accross_pass_through (struct cgraph_edge *cs,
1019 struct ipa_jump_func *jfunc,
1020 struct ipcp_lattice *src_lat,
1021 struct ipcp_lattice *dest_lat,
1022 int src_idx)
1024 struct ipcp_value *src_val;
1025 bool ret = false;
1027 if (ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
1028 for (src_val = src_lat->values; src_val; src_val = src_val->next)
1029 ret |= add_scalar_value_to_lattice (dest_lat, src_val->value, cs,
1030 src_val, src_idx);
1031 /* Do not create new values when propagating within an SCC because if there
1032 are arithmetic functions with circular dependencies, there is infinite
1033 number of them and we would just make lattices bottom. */
1034 else if (edge_within_scc (cs))
1035 ret = set_lattice_contains_variable (dest_lat);
1036 else
1037 for (src_val = src_lat->values; src_val; src_val = src_val->next)
1039 tree cstval = src_val->value;
1041 if (TREE_CODE (cstval) == TREE_BINFO)
1043 ret |= set_lattice_contains_variable (dest_lat);
1044 continue;
1046 cstval = ipa_get_jf_pass_through_result (jfunc, cstval);
1048 if (cstval)
1049 ret |= add_scalar_value_to_lattice (dest_lat, cstval, cs, src_val,
1050 src_idx);
1051 else
1052 ret |= set_lattice_contains_variable (dest_lat);
1055 return ret;
1058 /* Propagate values through an ancestor jump function JFUNC associated with
1059 edge CS, taking values from SRC_LAT and putting them into DEST_LAT. SRC_IDX
1060 is the index of the source parameter. */
1062 static bool
1063 propagate_vals_accross_ancestor (struct cgraph_edge *cs,
1064 struct ipa_jump_func *jfunc,
1065 struct ipcp_lattice *src_lat,
1066 struct ipcp_lattice *dest_lat,
1067 int src_idx)
1069 struct ipcp_value *src_val;
1070 bool ret = false;
1072 if (edge_within_scc (cs))
1073 return set_lattice_contains_variable (dest_lat);
1075 for (src_val = src_lat->values; src_val; src_val = src_val->next)
1077 tree t = ipa_get_jf_ancestor_result (jfunc, src_val->value);
1079 if (t)
1080 ret |= add_scalar_value_to_lattice (dest_lat, t, cs, src_val, src_idx);
1081 else
1082 ret |= set_lattice_contains_variable (dest_lat);
1085 return ret;
1088 /* Propagate scalar values across jump function JFUNC that is associated with
1089 edge CS and put the values into DEST_LAT. */
1091 static bool
1092 propagate_scalar_accross_jump_function (struct cgraph_edge *cs,
1093 struct ipa_jump_func *jfunc,
1094 struct ipcp_lattice *dest_lat)
1096 if (dest_lat->bottom)
1097 return false;
1099 if (jfunc->type == IPA_JF_CONST
1100 || jfunc->type == IPA_JF_KNOWN_TYPE)
1102 tree val;
1104 if (jfunc->type == IPA_JF_KNOWN_TYPE)
1106 val = ipa_value_from_known_type_jfunc (jfunc);
1107 if (!val)
1108 return set_lattice_contains_variable (dest_lat);
1110 else
1111 val = ipa_get_jf_constant (jfunc);
1112 return add_scalar_value_to_lattice (dest_lat, val, cs, NULL, 0);
1114 else if (jfunc->type == IPA_JF_PASS_THROUGH
1115 || jfunc->type == IPA_JF_ANCESTOR)
1117 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
1118 struct ipcp_lattice *src_lat;
1119 int src_idx;
1120 bool ret;
1122 if (jfunc->type == IPA_JF_PASS_THROUGH)
1123 src_idx = ipa_get_jf_pass_through_formal_id (jfunc);
1124 else
1125 src_idx = ipa_get_jf_ancestor_formal_id (jfunc);
1127 src_lat = ipa_get_scalar_lat (caller_info, src_idx);
1128 if (src_lat->bottom)
1129 return set_lattice_contains_variable (dest_lat);
1131 /* If we would need to clone the caller and cannot, do not propagate. */
1132 if (!ipcp_versionable_function_p (cs->caller)
1133 && (src_lat->contains_variable
1134 || (src_lat->values_count > 1)))
1135 return set_lattice_contains_variable (dest_lat);
1137 if (jfunc->type == IPA_JF_PASS_THROUGH)
1138 ret = propagate_vals_accross_pass_through (cs, jfunc, src_lat,
1139 dest_lat, src_idx);
1140 else
1141 ret = propagate_vals_accross_ancestor (cs, jfunc, src_lat, dest_lat,
1142 src_idx);
1144 if (src_lat->contains_variable)
1145 ret |= set_lattice_contains_variable (dest_lat);
1147 return ret;
1150 /* TODO: We currently do not handle member method pointers in IPA-CP (we only
1151 use it for indirect inlining), we should propagate them too. */
1152 return set_lattice_contains_variable (dest_lat);
1155 /* If DEST_PLATS already has aggregate items, check that aggs_by_ref matches
1156 NEW_AGGS_BY_REF and if not, mark all aggs as bottoms and return true (in all
1157 other cases, return false). If there are no aggregate items, set
1158 aggs_by_ref to NEW_AGGS_BY_REF. */
1160 static bool
1161 set_check_aggs_by_ref (struct ipcp_param_lattices *dest_plats,
1162 bool new_aggs_by_ref)
1164 if (dest_plats->aggs)
1166 if (dest_plats->aggs_by_ref != new_aggs_by_ref)
1168 set_agg_lats_to_bottom (dest_plats);
1169 return true;
1172 else
1173 dest_plats->aggs_by_ref = new_aggs_by_ref;
1174 return false;
1177 /* Walk aggregate lattices in DEST_PLATS from ***AGLAT on, until ***aglat is an
1178 already existing lattice for the given OFFSET and SIZE, marking all skipped
1179 lattices as containing variable and checking for overlaps. If there is no
1180 already existing lattice for the OFFSET and VAL_SIZE, create one, initialize
1181 it with offset, size and contains_variable to PRE_EXISTING, and return true,
1182 unless there are too many already. If there are two many, return false. If
1183 there are overlaps turn whole DEST_PLATS to bottom and return false. If any
1184 skipped lattices were newly marked as containing variable, set *CHANGE to
1185 true. */
1187 static bool
1188 merge_agg_lats_step (struct ipcp_param_lattices *dest_plats,
1189 HOST_WIDE_INT offset, HOST_WIDE_INT val_size,
1190 struct ipcp_agg_lattice ***aglat,
1191 bool pre_existing, bool *change)
1193 gcc_checking_assert (offset >= 0);
1195 while (**aglat && (**aglat)->offset < offset)
1197 if ((**aglat)->offset + (**aglat)->size > offset)
1199 set_agg_lats_to_bottom (dest_plats);
1200 return false;
1202 *change |= set_lattice_contains_variable (**aglat);
1203 *aglat = &(**aglat)->next;
1206 if (**aglat && (**aglat)->offset == offset)
1208 if ((**aglat)->size != val_size
1209 || ((**aglat)->next
1210 && (**aglat)->next->offset < offset + val_size))
1212 set_agg_lats_to_bottom (dest_plats);
1213 return false;
1215 gcc_checking_assert (!(**aglat)->next
1216 || (**aglat)->next->offset >= offset + val_size);
1217 return true;
1219 else
1221 struct ipcp_agg_lattice *new_al;
1223 if (**aglat && (**aglat)->offset < offset + val_size)
1225 set_agg_lats_to_bottom (dest_plats);
1226 return false;
1228 if (dest_plats->aggs_count == PARAM_VALUE (PARAM_IPA_MAX_AGG_ITEMS))
1229 return false;
1230 dest_plats->aggs_count++;
1231 new_al = (struct ipcp_agg_lattice *) pool_alloc (ipcp_agg_lattice_pool);
1232 memset (new_al, 0, sizeof (*new_al));
1234 new_al->offset = offset;
1235 new_al->size = val_size;
1236 new_al->contains_variable = pre_existing;
1238 new_al->next = **aglat;
1239 **aglat = new_al;
1240 return true;
1244 /* Set all AGLAT and all other aggregate lattices reachable by next pointers as
1245 containing an unknown value. */
1247 static bool
1248 set_chain_of_aglats_contains_variable (struct ipcp_agg_lattice *aglat)
1250 bool ret = false;
1251 while (aglat)
1253 ret |= set_lattice_contains_variable (aglat);
1254 aglat = aglat->next;
1256 return ret;
1259 /* Merge existing aggregate lattices in SRC_PLATS to DEST_PLATS, subtracting
1260 DELTA_OFFSET. CS is the call graph edge and SRC_IDX the index of the source
1261 parameter used for lattice value sources. Return true if DEST_PLATS changed
1262 in any way. */
1264 static bool
1265 merge_aggregate_lattices (struct cgraph_edge *cs,
1266 struct ipcp_param_lattices *dest_plats,
1267 struct ipcp_param_lattices *src_plats,
1268 int src_idx, HOST_WIDE_INT offset_delta)
1270 bool pre_existing = dest_plats->aggs != NULL;
1271 struct ipcp_agg_lattice **dst_aglat;
1272 bool ret = false;
1274 if (set_check_aggs_by_ref (dest_plats, src_plats->aggs_by_ref))
1275 return true;
1276 if (src_plats->aggs_bottom)
1277 return set_agg_lats_contain_variable (dest_plats);
1278 if (src_plats->aggs_contain_variable)
1279 ret |= set_agg_lats_contain_variable (dest_plats);
1280 dst_aglat = &dest_plats->aggs;
1282 for (struct ipcp_agg_lattice *src_aglat = src_plats->aggs;
1283 src_aglat;
1284 src_aglat = src_aglat->next)
1286 HOST_WIDE_INT new_offset = src_aglat->offset - offset_delta;
1288 if (new_offset < 0)
1289 continue;
1290 if (merge_agg_lats_step (dest_plats, new_offset, src_aglat->size,
1291 &dst_aglat, pre_existing, &ret))
1293 struct ipcp_agg_lattice *new_al = *dst_aglat;
1295 dst_aglat = &(*dst_aglat)->next;
1296 if (src_aglat->bottom)
1298 ret |= set_lattice_contains_variable (new_al);
1299 continue;
1301 if (src_aglat->contains_variable)
1302 ret |= set_lattice_contains_variable (new_al);
1303 for (struct ipcp_value *val = src_aglat->values;
1304 val;
1305 val = val->next)
1306 ret |= add_value_to_lattice (new_al, val->value, cs, val, src_idx,
1307 src_aglat->offset);
1309 else if (dest_plats->aggs_bottom)
1310 return true;
1312 ret |= set_chain_of_aglats_contains_variable (*dst_aglat);
1313 return ret;
1316 /* Determine whether there is anything to propagate FROM SRC_PLATS through a
1317 pass-through JFUNC and if so, whether it has conform and conforms to the
1318 rules about propagating values passed by reference. */
1320 static bool
1321 agg_pass_through_permissible_p (struct ipcp_param_lattices *src_plats,
1322 struct ipa_jump_func *jfunc)
1324 return src_plats->aggs
1325 && (!src_plats->aggs_by_ref
1326 || ipa_get_jf_pass_through_agg_preserved (jfunc));
1329 /* Propagate scalar values across jump function JFUNC that is associated with
1330 edge CS and put the values into DEST_LAT. */
1332 static bool
1333 propagate_aggs_accross_jump_function (struct cgraph_edge *cs,
1334 struct ipa_jump_func *jfunc,
1335 struct ipcp_param_lattices *dest_plats)
1337 bool ret = false;
1339 if (dest_plats->aggs_bottom)
1340 return false;
1342 if (jfunc->type == IPA_JF_PASS_THROUGH
1343 && ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
1345 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
1346 int src_idx = ipa_get_jf_pass_through_formal_id (jfunc);
1347 struct ipcp_param_lattices *src_plats;
1349 src_plats = ipa_get_parm_lattices (caller_info, src_idx);
1350 if (agg_pass_through_permissible_p (src_plats, jfunc))
1352 /* Currently we do not produce clobber aggregate jump
1353 functions, replace with merging when we do. */
1354 gcc_assert (!jfunc->agg.items);
1355 ret |= merge_aggregate_lattices (cs, dest_plats, src_plats,
1356 src_idx, 0);
1358 else
1359 ret |= set_agg_lats_contain_variable (dest_plats);
1361 else if (jfunc->type == IPA_JF_ANCESTOR
1362 && ipa_get_jf_ancestor_agg_preserved (jfunc))
1364 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
1365 int src_idx = ipa_get_jf_ancestor_formal_id (jfunc);
1366 struct ipcp_param_lattices *src_plats;
1368 src_plats = ipa_get_parm_lattices (caller_info, src_idx);
1369 if (src_plats->aggs && src_plats->aggs_by_ref)
1371 /* Currently we do not produce clobber aggregate jump
1372 functions, replace with merging when we do. */
1373 gcc_assert (!jfunc->agg.items);
1374 ret |= merge_aggregate_lattices (cs, dest_plats, src_plats, src_idx,
1375 ipa_get_jf_ancestor_offset (jfunc));
1377 else if (!src_plats->aggs_by_ref)
1378 ret |= set_agg_lats_to_bottom (dest_plats);
1379 else
1380 ret |= set_agg_lats_contain_variable (dest_plats);
1382 else if (jfunc->agg.items)
1384 bool pre_existing = dest_plats->aggs != NULL;
1385 struct ipcp_agg_lattice **aglat = &dest_plats->aggs;
1386 struct ipa_agg_jf_item *item;
1387 int i;
1389 if (set_check_aggs_by_ref (dest_plats, jfunc->agg.by_ref))
1390 return true;
1392 FOR_EACH_VEC_ELT (*jfunc->agg.items, i, item)
1394 HOST_WIDE_INT val_size;
1396 if (item->offset < 0)
1397 continue;
1398 gcc_checking_assert (is_gimple_ip_invariant (item->value));
1399 val_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (item->value)), 1);
1401 if (merge_agg_lats_step (dest_plats, item->offset, val_size,
1402 &aglat, pre_existing, &ret))
1404 ret |= add_value_to_lattice (*aglat, item->value, cs, NULL, 0, 0);
1405 aglat = &(*aglat)->next;
1407 else if (dest_plats->aggs_bottom)
1408 return true;
1411 ret |= set_chain_of_aglats_contains_variable (*aglat);
1413 else
1414 ret |= set_agg_lats_contain_variable (dest_plats);
1416 return ret;
1419 /* Propagate constants from the caller to the callee of CS. INFO describes the
1420 caller. */
1422 static bool
1423 propagate_constants_accross_call (struct cgraph_edge *cs)
1425 struct ipa_node_params *callee_info;
1426 enum availability availability;
1427 struct cgraph_node *callee, *alias_or_thunk;
1428 struct ipa_edge_args *args;
1429 bool ret = false;
1430 int i, args_count, parms_count;
1432 callee = cgraph_function_node (cs->callee, &availability);
1433 if (!callee->analyzed)
1434 return false;
1435 gcc_checking_assert (cgraph_function_with_gimple_body_p (callee));
1436 callee_info = IPA_NODE_REF (callee);
1438 args = IPA_EDGE_REF (cs);
1439 args_count = ipa_get_cs_argument_count (args);
1440 parms_count = ipa_get_param_count (callee_info);
1442 /* If this call goes through a thunk we must not propagate to the first (0th)
1443 parameter. However, we might need to uncover a thunk from below a series
1444 of aliases first. */
1445 alias_or_thunk = cs->callee;
1446 while (alias_or_thunk->alias)
1447 alias_or_thunk = cgraph_alias_aliased_node (alias_or_thunk);
1448 if (alias_or_thunk->thunk.thunk_p)
1450 ret |= set_all_contains_variable (ipa_get_parm_lattices (callee_info,
1451 0));
1452 i = 1;
1454 else
1455 i = 0;
1457 for (; (i < args_count) && (i < parms_count); i++)
1459 struct ipa_jump_func *jump_func = ipa_get_ith_jump_func (args, i);
1460 struct ipcp_param_lattices *dest_plats;
1462 dest_plats = ipa_get_parm_lattices (callee_info, i);
1463 if (availability == AVAIL_OVERWRITABLE)
1464 ret |= set_all_contains_variable (dest_plats);
1465 else
1467 ret |= propagate_scalar_accross_jump_function (cs, jump_func,
1468 &dest_plats->itself);
1469 ret |= propagate_aggs_accross_jump_function (cs, jump_func,
1470 dest_plats);
1473 for (; i < parms_count; i++)
1474 ret |= set_all_contains_variable (ipa_get_parm_lattices (callee_info, i));
1476 return ret;
1479 /* If an indirect edge IE can be turned into a direct one based on KNOWN_VALS
1480 (which can contain both constants and binfos) or KNOWN_BINFOS (which can be
1481 NULL) return the destination. */
1483 tree
1484 ipa_get_indirect_edge_target (struct cgraph_edge *ie,
1485 vec<tree> known_vals,
1486 vec<tree> known_binfos,
1487 vec<ipa_agg_jump_function_p> known_aggs)
1489 int param_index = ie->indirect_info->param_index;
1490 HOST_WIDE_INT token, anc_offset;
1491 tree otr_type;
1492 tree t;
1494 if (param_index == -1)
1495 return NULL_TREE;
1497 if (!ie->indirect_info->polymorphic)
1499 tree t;
1501 if (ie->indirect_info->agg_contents)
1503 if (known_aggs.length ()
1504 > (unsigned int) param_index)
1506 struct ipa_agg_jump_function *agg;
1507 agg = known_aggs[param_index];
1508 t = ipa_find_agg_cst_for_param (agg, ie->indirect_info->offset,
1509 ie->indirect_info->by_ref);
1511 else
1512 t = NULL;
1514 else
1515 t = (known_vals.length () > (unsigned int) param_index
1516 ? known_vals[param_index] : NULL);
1518 if (t &&
1519 TREE_CODE (t) == ADDR_EXPR
1520 && TREE_CODE (TREE_OPERAND (t, 0)) == FUNCTION_DECL)
1521 return TREE_OPERAND (t, 0);
1522 else
1523 return NULL_TREE;
1526 gcc_assert (!ie->indirect_info->agg_contents);
1527 token = ie->indirect_info->otr_token;
1528 anc_offset = ie->indirect_info->offset;
1529 otr_type = ie->indirect_info->otr_type;
1531 t = known_vals[param_index];
1532 if (!t && known_binfos.length () > (unsigned int) param_index)
1533 t = known_binfos[param_index];
1534 if (!t)
1535 return NULL_TREE;
1537 if (TREE_CODE (t) != TREE_BINFO)
1539 tree binfo;
1540 binfo = gimple_extract_devirt_binfo_from_cst (t);
1541 if (!binfo)
1542 return NULL_TREE;
1543 binfo = get_binfo_at_offset (binfo, anc_offset, otr_type);
1544 if (!binfo)
1545 return NULL_TREE;
1546 return gimple_get_virt_method_for_binfo (token, binfo);
1548 else
1550 tree binfo;
1552 binfo = get_binfo_at_offset (t, anc_offset, otr_type);
1553 if (!binfo)
1554 return NULL_TREE;
1555 return gimple_get_virt_method_for_binfo (token, binfo);
1559 /* Calculate devirtualization time bonus for NODE, assuming we know KNOWN_CSTS
1560 and KNOWN_BINFOS. */
1562 static int
1563 devirtualization_time_bonus (struct cgraph_node *node,
1564 vec<tree> known_csts,
1565 vec<tree> known_binfos)
1567 struct cgraph_edge *ie;
1568 int res = 0;
1570 for (ie = node->indirect_calls; ie; ie = ie->next_callee)
1572 struct cgraph_node *callee;
1573 struct inline_summary *isummary;
1574 tree target;
1576 target = ipa_get_indirect_edge_target (ie, known_csts, known_binfos,
1577 vNULL);
1578 if (!target)
1579 continue;
1581 /* Only bare minimum benefit for clearly un-inlineable targets. */
1582 res += 1;
1583 callee = cgraph_get_node (target);
1584 if (!callee || !callee->analyzed)
1585 continue;
1586 isummary = inline_summary (callee);
1587 if (!isummary->inlinable)
1588 continue;
1590 /* FIXME: The values below need re-considering and perhaps also
1591 integrating into the cost metrics, at lest in some very basic way. */
1592 if (isummary->size <= MAX_INLINE_INSNS_AUTO / 4)
1593 res += 31;
1594 else if (isummary->size <= MAX_INLINE_INSNS_AUTO / 2)
1595 res += 15;
1596 else if (isummary->size <= MAX_INLINE_INSNS_AUTO
1597 || DECL_DECLARED_INLINE_P (callee->symbol.decl))
1598 res += 7;
1601 return res;
1604 /* Return time bonus incurred because of HINTS. */
1606 static int
1607 hint_time_bonus (inline_hints hints)
1609 if (hints & (INLINE_HINT_loop_iterations | INLINE_HINT_loop_stride))
1610 return PARAM_VALUE (PARAM_IPA_CP_LOOP_HINT_BONUS);
1611 return 0;
1614 /* Return true if cloning NODE is a good idea, given the estimated TIME_BENEFIT
1615 and SIZE_COST and with the sum of frequencies of incoming edges to the
1616 potential new clone in FREQUENCIES. */
1618 static bool
1619 good_cloning_opportunity_p (struct cgraph_node *node, int time_benefit,
1620 int freq_sum, gcov_type count_sum, int size_cost)
1622 if (time_benefit == 0
1623 || !flag_ipa_cp_clone
1624 || !optimize_function_for_speed_p (DECL_STRUCT_FUNCTION (node->symbol.decl)))
1625 return false;
1627 gcc_assert (size_cost > 0);
1629 if (max_count)
1631 int factor = (count_sum * 1000) / max_count;
1632 HOST_WIDEST_INT evaluation = (((HOST_WIDEST_INT) time_benefit * factor)
1633 / size_cost);
1635 if (dump_file && (dump_flags & TDF_DETAILS))
1636 fprintf (dump_file, " good_cloning_opportunity_p (time: %i, "
1637 "size: %i, count_sum: " HOST_WIDE_INT_PRINT_DEC
1638 ") -> evaluation: " HOST_WIDEST_INT_PRINT_DEC
1639 ", threshold: %i\n",
1640 time_benefit, size_cost, (HOST_WIDE_INT) count_sum,
1641 evaluation, 500);
1643 return evaluation >= PARAM_VALUE (PARAM_IPA_CP_EVAL_THRESHOLD);
1645 else
1647 HOST_WIDEST_INT evaluation = (((HOST_WIDEST_INT) time_benefit * freq_sum)
1648 / size_cost);
1650 if (dump_file && (dump_flags & TDF_DETAILS))
1651 fprintf (dump_file, " good_cloning_opportunity_p (time: %i, "
1652 "size: %i, freq_sum: %i) -> evaluation: "
1653 HOST_WIDEST_INT_PRINT_DEC ", threshold: %i\n",
1654 time_benefit, size_cost, freq_sum, evaluation,
1655 CGRAPH_FREQ_BASE /2);
1657 return evaluation >= PARAM_VALUE (PARAM_IPA_CP_EVAL_THRESHOLD);
1661 /* Return all context independent values from aggregate lattices in PLATS in a
1662 vector. Return NULL if there are none. */
1664 static vec<ipa_agg_jf_item_t, va_gc> *
1665 context_independent_aggregate_values (struct ipcp_param_lattices *plats)
1667 vec<ipa_agg_jf_item_t, va_gc> *res = NULL;
1669 if (plats->aggs_bottom
1670 || plats->aggs_contain_variable
1671 || plats->aggs_count == 0)
1672 return NULL;
1674 for (struct ipcp_agg_lattice *aglat = plats->aggs;
1675 aglat;
1676 aglat = aglat->next)
1677 if (ipa_lat_is_single_const (aglat))
1679 struct ipa_agg_jf_item item;
1680 item.offset = aglat->offset;
1681 item.value = aglat->values->value;
1682 vec_safe_push (res, item);
1684 return res;
1687 /* Allocate KNOWN_CSTS, KNOWN_BINFOS and, if non-NULL, KNOWN_AGGS and populate
1688 them with values of parameters that are known independent of the context.
1689 INFO describes the function. If REMOVABLE_PARAMS_COST is non-NULL, the
1690 movement cost of all removable parameters will be stored in it. */
1692 static bool
1693 gather_context_independent_values (struct ipa_node_params *info,
1694 vec<tree> *known_csts,
1695 vec<tree> *known_binfos,
1696 vec<ipa_agg_jump_function_t> *known_aggs,
1697 int *removable_params_cost)
1699 int i, count = ipa_get_param_count (info);
1700 bool ret = false;
1702 known_csts->create (0);
1703 known_binfos->create (0);
1704 known_csts->safe_grow_cleared (count);
1705 known_binfos->safe_grow_cleared (count);
1706 if (known_aggs)
1708 known_aggs->create (0);
1709 known_aggs->safe_grow_cleared (count);
1712 if (removable_params_cost)
1713 *removable_params_cost = 0;
1715 for (i = 0; i < count ; i++)
1717 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
1718 struct ipcp_lattice *lat = &plats->itself;
1720 if (ipa_lat_is_single_const (lat))
1722 struct ipcp_value *val = lat->values;
1723 if (TREE_CODE (val->value) != TREE_BINFO)
1725 (*known_csts)[i] = val->value;
1726 if (removable_params_cost)
1727 *removable_params_cost
1728 += estimate_move_cost (TREE_TYPE (val->value));
1729 ret = true;
1731 else if (plats->virt_call)
1733 (*known_binfos)[i] = val->value;
1734 ret = true;
1736 else if (removable_params_cost
1737 && !ipa_is_param_used (info, i))
1738 *removable_params_cost
1739 += estimate_move_cost (TREE_TYPE (ipa_get_param (info, i)));
1741 else if (removable_params_cost
1742 && !ipa_is_param_used (info, i))
1743 *removable_params_cost
1744 += estimate_move_cost (TREE_TYPE (ipa_get_param (info, i)));
1746 if (known_aggs)
1748 vec<ipa_agg_jf_item_t, va_gc> *agg_items;
1749 struct ipa_agg_jump_function *ajf;
1751 agg_items = context_independent_aggregate_values (plats);
1752 ajf = &(*known_aggs)[i];
1753 ajf->items = agg_items;
1754 ajf->by_ref = plats->aggs_by_ref;
1755 ret |= agg_items != NULL;
1759 return ret;
1762 /* The current interface in ipa-inline-analysis requires a pointer vector.
1763 Create it.
1765 FIXME: That interface should be re-worked, this is slightly silly. Still,
1766 I'd like to discuss how to change it first and this demonstrates the
1767 issue. */
1769 static vec<ipa_agg_jump_function_p>
1770 agg_jmp_p_vec_for_t_vec (vec<ipa_agg_jump_function_t> known_aggs)
1772 vec<ipa_agg_jump_function_p> ret;
1773 struct ipa_agg_jump_function *ajf;
1774 int i;
1776 ret.create (known_aggs.length ());
1777 FOR_EACH_VEC_ELT (known_aggs, i, ajf)
1778 ret.quick_push (ajf);
1779 return ret;
1782 /* Iterate over known values of parameters of NODE and estimate the local
1783 effects in terms of time and size they have. */
1785 static void
1786 estimate_local_effects (struct cgraph_node *node)
1788 struct ipa_node_params *info = IPA_NODE_REF (node);
1789 int i, count = ipa_get_param_count (info);
1790 vec<tree> known_csts, known_binfos;
1791 vec<ipa_agg_jump_function_t> known_aggs;
1792 vec<ipa_agg_jump_function_p> known_aggs_ptrs;
1793 bool always_const;
1794 int base_time = inline_summary (node)->time;
1795 int removable_params_cost;
1797 if (!count || !ipcp_versionable_function_p (node))
1798 return;
1800 if (dump_file && (dump_flags & TDF_DETAILS))
1801 fprintf (dump_file, "\nEstimating effects for %s/%i, base_time: %i.\n",
1802 cgraph_node_name (node), node->uid, base_time);
1804 always_const = gather_context_independent_values (info, &known_csts,
1805 &known_binfos, &known_aggs,
1806 &removable_params_cost);
1807 known_aggs_ptrs = agg_jmp_p_vec_for_t_vec (known_aggs);
1808 if (always_const)
1810 struct caller_statistics stats;
1811 inline_hints hints;
1812 int time, size;
1814 init_caller_stats (&stats);
1815 cgraph_for_node_and_aliases (node, gather_caller_stats, &stats, false);
1816 estimate_ipcp_clone_size_and_time (node, known_csts, known_binfos,
1817 known_aggs_ptrs, &size, &time, &hints);
1818 time -= devirtualization_time_bonus (node, known_csts, known_binfos);
1819 time -= hint_time_bonus (hints);
1820 time -= removable_params_cost;
1821 size -= stats.n_calls * removable_params_cost;
1823 if (dump_file)
1824 fprintf (dump_file, " - context independent values, size: %i, "
1825 "time_benefit: %i\n", size, base_time - time);
1827 if (size <= 0
1828 || cgraph_will_be_removed_from_program_if_no_direct_calls (node))
1830 info->do_clone_for_all_contexts = true;
1831 base_time = time;
1833 if (dump_file)
1834 fprintf (dump_file, " Decided to specialize for all "
1835 "known contexts, code not going to grow.\n");
1837 else if (good_cloning_opportunity_p (node, base_time - time,
1838 stats.freq_sum, stats.count_sum,
1839 size))
1841 if (size + overall_size <= max_new_size)
1843 info->do_clone_for_all_contexts = true;
1844 base_time = time;
1845 overall_size += size;
1847 if (dump_file)
1848 fprintf (dump_file, " Decided to specialize for all "
1849 "known contexts, growth deemed beneficial.\n");
1851 else if (dump_file && (dump_flags & TDF_DETAILS))
1852 fprintf (dump_file, " Not cloning for all contexts because "
1853 "max_new_size would be reached with %li.\n",
1854 size + overall_size);
1858 for (i = 0; i < count ; i++)
1860 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
1861 struct ipcp_lattice *lat = &plats->itself;
1862 struct ipcp_value *val;
1863 int emc;
1865 if (lat->bottom
1866 || !lat->values
1867 || known_csts[i]
1868 || known_binfos[i])
1869 continue;
1871 for (val = lat->values; val; val = val->next)
1873 int time, size, time_benefit;
1874 inline_hints hints;
1876 if (TREE_CODE (val->value) != TREE_BINFO)
1878 known_csts[i] = val->value;
1879 known_binfos[i] = NULL_TREE;
1880 emc = estimate_move_cost (TREE_TYPE (val->value));
1882 else if (plats->virt_call)
1884 known_csts[i] = NULL_TREE;
1885 known_binfos[i] = val->value;
1886 emc = 0;
1888 else
1889 continue;
1891 estimate_ipcp_clone_size_and_time (node, known_csts, known_binfos,
1892 known_aggs_ptrs, &size, &time,
1893 &hints);
1894 time_benefit = base_time - time
1895 + devirtualization_time_bonus (node, known_csts, known_binfos)
1896 + hint_time_bonus (hints)
1897 + removable_params_cost + emc;
1899 gcc_checking_assert (size >=0);
1900 /* The inliner-heuristics based estimates may think that in certain
1901 contexts some functions do not have any size at all but we want
1902 all specializations to have at least a tiny cost, not least not to
1903 divide by zero. */
1904 if (size == 0)
1905 size = 1;
1907 if (dump_file && (dump_flags & TDF_DETAILS))
1909 fprintf (dump_file, " - estimates for value ");
1910 print_ipcp_constant_value (dump_file, val->value);
1911 fprintf (dump_file, " for parameter ");
1912 print_generic_expr (dump_file, ipa_get_param (info, i), 0);
1913 fprintf (dump_file, ": time_benefit: %i, size: %i\n",
1914 time_benefit, size);
1917 val->local_time_benefit = time_benefit;
1918 val->local_size_cost = size;
1920 known_binfos[i] = NULL_TREE;
1921 known_csts[i] = NULL_TREE;
1924 for (i = 0; i < count ; i++)
1926 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
1927 struct ipa_agg_jump_function *ajf;
1928 struct ipcp_agg_lattice *aglat;
1930 if (plats->aggs_bottom || !plats->aggs)
1931 continue;
1933 ajf = &known_aggs[i];
1934 for (aglat = plats->aggs; aglat; aglat = aglat->next)
1936 struct ipcp_value *val;
1937 if (aglat->bottom || !aglat->values
1938 /* If the following is true, the one value is in known_aggs. */
1939 || (!plats->aggs_contain_variable
1940 && ipa_lat_is_single_const (aglat)))
1941 continue;
1943 for (val = aglat->values; val; val = val->next)
1945 int time, size, time_benefit;
1946 struct ipa_agg_jf_item item;
1947 inline_hints hints;
1949 item.offset = aglat->offset;
1950 item.value = val->value;
1951 vec_safe_push (ajf->items, item);
1953 estimate_ipcp_clone_size_and_time (node, known_csts, known_binfos,
1954 known_aggs_ptrs, &size, &time,
1955 &hints);
1956 time_benefit = base_time - time
1957 + devirtualization_time_bonus (node, known_csts, known_binfos)
1958 + hint_time_bonus (hints);
1959 gcc_checking_assert (size >=0);
1960 if (size == 0)
1961 size = 1;
1963 if (dump_file && (dump_flags & TDF_DETAILS))
1965 fprintf (dump_file, " - estimates for value ");
1966 print_ipcp_constant_value (dump_file, val->value);
1967 fprintf (dump_file, " for parameter ");
1968 print_generic_expr (dump_file, ipa_get_param (info, i), 0);
1969 fprintf (dump_file, "[%soffset: " HOST_WIDE_INT_PRINT_DEC
1970 "]: time_benefit: %i, size: %i\n",
1971 plats->aggs_by_ref ? "ref " : "",
1972 aglat->offset, time_benefit, size);
1975 val->local_time_benefit = time_benefit;
1976 val->local_size_cost = size;
1977 ajf->items->pop ();
1982 for (i = 0; i < count ; i++)
1983 vec_free (known_aggs[i].items);
1985 known_csts.release ();
1986 known_binfos.release ();
1987 known_aggs.release ();
1988 known_aggs_ptrs.release ();
1992 /* Add value CUR_VAL and all yet-unsorted values it is dependent on to the
1993 topological sort of values. */
1995 static void
1996 add_val_to_toposort (struct ipcp_value *cur_val)
1998 static int dfs_counter = 0;
1999 static struct ipcp_value *stack;
2000 struct ipcp_value_source *src;
2002 if (cur_val->dfs)
2003 return;
2005 dfs_counter++;
2006 cur_val->dfs = dfs_counter;
2007 cur_val->low_link = dfs_counter;
2009 cur_val->topo_next = stack;
2010 stack = cur_val;
2011 cur_val->on_stack = true;
2013 for (src = cur_val->sources; src; src = src->next)
2014 if (src->val)
2016 if (src->val->dfs == 0)
2018 add_val_to_toposort (src->val);
2019 if (src->val->low_link < cur_val->low_link)
2020 cur_val->low_link = src->val->low_link;
2022 else if (src->val->on_stack
2023 && src->val->dfs < cur_val->low_link)
2024 cur_val->low_link = src->val->dfs;
2027 if (cur_val->dfs == cur_val->low_link)
2029 struct ipcp_value *v, *scc_list = NULL;
2033 v = stack;
2034 stack = v->topo_next;
2035 v->on_stack = false;
2037 v->scc_next = scc_list;
2038 scc_list = v;
2040 while (v != cur_val);
2042 cur_val->topo_next = values_topo;
2043 values_topo = cur_val;
2047 /* Add all values in lattices associated with NODE to the topological sort if
2048 they are not there yet. */
2050 static void
2051 add_all_node_vals_to_toposort (struct cgraph_node *node)
2053 struct ipa_node_params *info = IPA_NODE_REF (node);
2054 int i, count = ipa_get_param_count (info);
2056 for (i = 0; i < count ; i++)
2058 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
2059 struct ipcp_lattice *lat = &plats->itself;
2060 struct ipcp_agg_lattice *aglat;
2061 struct ipcp_value *val;
2063 if (!lat->bottom)
2064 for (val = lat->values; val; val = val->next)
2065 add_val_to_toposort (val);
2067 if (!plats->aggs_bottom)
2068 for (aglat = plats->aggs; aglat; aglat = aglat->next)
2069 if (!aglat->bottom)
2070 for (val = aglat->values; val; val = val->next)
2071 add_val_to_toposort (val);
2075 /* One pass of constants propagation along the call graph edges, from callers
2076 to callees (requires topological ordering in TOPO), iterate over strongly
2077 connected components. */
2079 static void
2080 propagate_constants_topo (struct topo_info *topo)
2082 int i;
2084 for (i = topo->nnodes - 1; i >= 0; i--)
2086 struct cgraph_node *v, *node = topo->order[i];
2087 struct ipa_dfs_info *node_dfs_info;
2089 if (!cgraph_function_with_gimple_body_p (node))
2090 continue;
2092 node_dfs_info = (struct ipa_dfs_info *) node->symbol.aux;
2093 /* First, iteratively propagate within the strongly connected component
2094 until all lattices stabilize. */
2095 v = node_dfs_info->next_cycle;
2096 while (v)
2098 push_node_to_stack (topo, v);
2099 v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle;
2102 v = node;
2103 while (v)
2105 struct cgraph_edge *cs;
2107 for (cs = v->callees; cs; cs = cs->next_callee)
2108 if (edge_within_scc (cs)
2109 && propagate_constants_accross_call (cs))
2110 push_node_to_stack (topo, cs->callee);
2111 v = pop_node_from_stack (topo);
2114 /* Afterwards, propagate along edges leading out of the SCC, calculates
2115 the local effects of the discovered constants and all valid values to
2116 their topological sort. */
2117 v = node;
2118 while (v)
2120 struct cgraph_edge *cs;
2122 estimate_local_effects (v);
2123 add_all_node_vals_to_toposort (v);
2124 for (cs = v->callees; cs; cs = cs->next_callee)
2125 if (!edge_within_scc (cs))
2126 propagate_constants_accross_call (cs);
2128 v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle;
2134 /* Return the sum of A and B if none of them is bigger than INT_MAX/2, return
2135 the bigger one if otherwise. */
2137 static int
2138 safe_add (int a, int b)
2140 if (a > INT_MAX/2 || b > INT_MAX/2)
2141 return a > b ? a : b;
2142 else
2143 return a + b;
2147 /* Propagate the estimated effects of individual values along the topological
2148 from the dependent values to those they depend on. */
2150 static void
2151 propagate_effects (void)
2153 struct ipcp_value *base;
2155 for (base = values_topo; base; base = base->topo_next)
2157 struct ipcp_value_source *src;
2158 struct ipcp_value *val;
2159 int time = 0, size = 0;
2161 for (val = base; val; val = val->scc_next)
2163 time = safe_add (time,
2164 val->local_time_benefit + val->prop_time_benefit);
2165 size = safe_add (size, val->local_size_cost + val->prop_size_cost);
2168 for (val = base; val; val = val->scc_next)
2169 for (src = val->sources; src; src = src->next)
2170 if (src->val
2171 && cgraph_maybe_hot_edge_p (src->cs))
2173 src->val->prop_time_benefit = safe_add (time,
2174 src->val->prop_time_benefit);
2175 src->val->prop_size_cost = safe_add (size,
2176 src->val->prop_size_cost);
2182 /* Propagate constants, binfos and their effects from the summaries
2183 interprocedurally. */
2185 static void
2186 ipcp_propagate_stage (struct topo_info *topo)
2188 struct cgraph_node *node;
2190 if (dump_file)
2191 fprintf (dump_file, "\n Propagating constants:\n\n");
2193 if (in_lto_p)
2194 ipa_update_after_lto_read ();
2197 FOR_EACH_DEFINED_FUNCTION (node)
2199 struct ipa_node_params *info = IPA_NODE_REF (node);
2201 determine_versionability (node);
2202 if (cgraph_function_with_gimple_body_p (node))
2204 info->lattices = XCNEWVEC (struct ipcp_param_lattices,
2205 ipa_get_param_count (info));
2206 initialize_node_lattices (node);
2208 if (node->count > max_count)
2209 max_count = node->count;
2210 overall_size += inline_summary (node)->self_size;
2213 max_new_size = overall_size;
2214 if (max_new_size < PARAM_VALUE (PARAM_LARGE_UNIT_INSNS))
2215 max_new_size = PARAM_VALUE (PARAM_LARGE_UNIT_INSNS);
2216 max_new_size += max_new_size * PARAM_VALUE (PARAM_IPCP_UNIT_GROWTH) / 100 + 1;
2218 if (dump_file)
2219 fprintf (dump_file, "\noverall_size: %li, max_new_size: %li\n",
2220 overall_size, max_new_size);
2222 propagate_constants_topo (topo);
2223 #ifdef ENABLE_CHECKING
2224 ipcp_verify_propagated_values ();
2225 #endif
2226 propagate_effects ();
2228 if (dump_file)
2230 fprintf (dump_file, "\nIPA lattices after all propagation:\n");
2231 print_all_lattices (dump_file, (dump_flags & TDF_DETAILS), true);
2235 /* Discover newly direct outgoing edges from NODE which is a new clone with
2236 known KNOWN_VALS and make them direct. */
2238 static void
2239 ipcp_discover_new_direct_edges (struct cgraph_node *node,
2240 vec<tree> known_vals)
2242 struct cgraph_edge *ie, *next_ie;
2243 bool found = false;
2245 for (ie = node->indirect_calls; ie; ie = next_ie)
2247 tree target;
2249 next_ie = ie->next_callee;
2250 target = ipa_get_indirect_edge_target (ie, known_vals, vNULL, vNULL);
2251 if (target)
2253 ipa_make_edge_direct_to_target (ie, target);
2254 found = true;
2257 /* Turning calls to direct calls will improve overall summary. */
2258 if (found)
2259 inline_update_overall_summary (node);
2262 /* Vector of pointers which for linked lists of clones of an original crgaph
2263 edge. */
2265 static vec<cgraph_edge_p> next_edge_clone;
2267 static inline void
2268 grow_next_edge_clone_vector (void)
2270 if (next_edge_clone.length ()
2271 <= (unsigned) cgraph_edge_max_uid)
2272 next_edge_clone.safe_grow_cleared (cgraph_edge_max_uid + 1);
2275 /* Edge duplication hook to grow the appropriate linked list in
2276 next_edge_clone. */
2278 static void
2279 ipcp_edge_duplication_hook (struct cgraph_edge *src, struct cgraph_edge *dst,
2280 __attribute__((unused)) void *data)
2282 grow_next_edge_clone_vector ();
2283 next_edge_clone[dst->uid] = next_edge_clone[src->uid];
2284 next_edge_clone[src->uid] = dst;
2287 /* See if NODE is a clone with a known aggregate value at a given OFFSET of a
2288 parameter with the given INDEX. */
2290 static tree
2291 get_clone_agg_value (struct cgraph_node *node, HOST_WIDEST_INT offset,
2292 int index)
2294 struct ipa_agg_replacement_value *aggval;
2296 aggval = ipa_get_agg_replacements_for_node (node);
2297 while (aggval)
2299 if (aggval->offset == offset
2300 && aggval->index == index)
2301 return aggval->value;
2302 aggval = aggval->next;
2304 return NULL_TREE;
2307 /* Return true if edge CS does bring about the value described by SRC. */
2309 static bool
2310 cgraph_edge_brings_value_p (struct cgraph_edge *cs,
2311 struct ipcp_value_source *src)
2313 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
2314 struct ipa_node_params *dst_info = IPA_NODE_REF (cs->callee);
2316 if ((dst_info->ipcp_orig_node && !dst_info->is_all_contexts_clone)
2317 || caller_info->node_dead)
2318 return false;
2319 if (!src->val)
2320 return true;
2322 if (caller_info->ipcp_orig_node)
2324 tree t;
2325 if (src->offset == -1)
2326 t = caller_info->known_vals[src->index];
2327 else
2328 t = get_clone_agg_value (cs->caller, src->offset, src->index);
2329 return (t != NULL_TREE
2330 && values_equal_for_ipcp_p (src->val->value, t));
2332 else
2334 struct ipcp_agg_lattice *aglat;
2335 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (caller_info,
2336 src->index);
2337 if (src->offset == -1)
2338 return (ipa_lat_is_single_const (&plats->itself)
2339 && values_equal_for_ipcp_p (src->val->value,
2340 plats->itself.values->value));
2341 else
2343 if (plats->aggs_bottom || plats->aggs_contain_variable)
2344 return false;
2345 for (aglat = plats->aggs; aglat; aglat = aglat->next)
2346 if (aglat->offset == src->offset)
2347 return (ipa_lat_is_single_const (aglat)
2348 && values_equal_for_ipcp_p (src->val->value,
2349 aglat->values->value));
2351 return false;
2355 /* Get the next clone in the linked list of clones of an edge. */
2357 static inline struct cgraph_edge *
2358 get_next_cgraph_edge_clone (struct cgraph_edge *cs)
2360 return next_edge_clone[cs->uid];
2363 /* Given VAL, iterate over all its sources and if they still hold, add their
2364 edge frequency and their number into *FREQUENCY and *CALLER_COUNT
2365 respectively. */
2367 static bool
2368 get_info_about_necessary_edges (struct ipcp_value *val, int *freq_sum,
2369 gcov_type *count_sum, int *caller_count)
2371 struct ipcp_value_source *src;
2372 int freq = 0, count = 0;
2373 gcov_type cnt = 0;
2374 bool hot = false;
2376 for (src = val->sources; src; src = src->next)
2378 struct cgraph_edge *cs = src->cs;
2379 while (cs)
2381 if (cgraph_edge_brings_value_p (cs, src))
2383 count++;
2384 freq += cs->frequency;
2385 cnt += cs->count;
2386 hot |= cgraph_maybe_hot_edge_p (cs);
2388 cs = get_next_cgraph_edge_clone (cs);
2392 *freq_sum = freq;
2393 *count_sum = cnt;
2394 *caller_count = count;
2395 return hot;
2398 /* Return a vector of incoming edges that do bring value VAL. It is assumed
2399 their number is known and equal to CALLER_COUNT. */
2401 static vec<cgraph_edge_p>
2402 gather_edges_for_value (struct ipcp_value *val, int caller_count)
2404 struct ipcp_value_source *src;
2405 vec<cgraph_edge_p> ret;
2407 ret.create (caller_count);
2408 for (src = val->sources; src; src = src->next)
2410 struct cgraph_edge *cs = src->cs;
2411 while (cs)
2413 if (cgraph_edge_brings_value_p (cs, src))
2414 ret.quick_push (cs);
2415 cs = get_next_cgraph_edge_clone (cs);
2419 return ret;
2422 /* Construct a replacement map for a know VALUE for a formal parameter PARAM.
2423 Return it or NULL if for some reason it cannot be created. */
2425 static struct ipa_replace_map *
2426 get_replacement_map (tree value, tree parm)
2428 tree req_type = TREE_TYPE (parm);
2429 struct ipa_replace_map *replace_map;
2431 if (!useless_type_conversion_p (req_type, TREE_TYPE (value)))
2433 if (fold_convertible_p (req_type, value))
2434 value = fold_build1 (NOP_EXPR, req_type, value);
2435 else if (TYPE_SIZE (req_type) == TYPE_SIZE (TREE_TYPE (value)))
2436 value = fold_build1 (VIEW_CONVERT_EXPR, req_type, value);
2437 else
2439 if (dump_file)
2441 fprintf (dump_file, " const ");
2442 print_generic_expr (dump_file, value, 0);
2443 fprintf (dump_file, " can't be converted to param ");
2444 print_generic_expr (dump_file, parm, 0);
2445 fprintf (dump_file, "\n");
2447 return NULL;
2451 replace_map = ggc_alloc_ipa_replace_map ();
2452 if (dump_file)
2454 fprintf (dump_file, " replacing param ");
2455 print_generic_expr (dump_file, parm, 0);
2456 fprintf (dump_file, " with const ");
2457 print_generic_expr (dump_file, value, 0);
2458 fprintf (dump_file, "\n");
2460 replace_map->old_tree = parm;
2461 replace_map->new_tree = value;
2462 replace_map->replace_p = true;
2463 replace_map->ref_p = false;
2465 return replace_map;
2468 /* Dump new profiling counts */
2470 static void
2471 dump_profile_updates (struct cgraph_node *orig_node,
2472 struct cgraph_node *new_node)
2474 struct cgraph_edge *cs;
2476 fprintf (dump_file, " setting count of the specialized node to "
2477 HOST_WIDE_INT_PRINT_DEC "\n", (HOST_WIDE_INT) new_node->count);
2478 for (cs = new_node->callees; cs ; cs = cs->next_callee)
2479 fprintf (dump_file, " edge to %s has count "
2480 HOST_WIDE_INT_PRINT_DEC "\n",
2481 cgraph_node_name (cs->callee), (HOST_WIDE_INT) cs->count);
2483 fprintf (dump_file, " setting count of the original node to "
2484 HOST_WIDE_INT_PRINT_DEC "\n", (HOST_WIDE_INT) orig_node->count);
2485 for (cs = orig_node->callees; cs ; cs = cs->next_callee)
2486 fprintf (dump_file, " edge to %s is left with "
2487 HOST_WIDE_INT_PRINT_DEC "\n",
2488 cgraph_node_name (cs->callee), (HOST_WIDE_INT) cs->count);
2491 /* After a specialized NEW_NODE version of ORIG_NODE has been created, update
2492 their profile information to reflect this. */
2494 static void
2495 update_profiling_info (struct cgraph_node *orig_node,
2496 struct cgraph_node *new_node)
2498 struct cgraph_edge *cs;
2499 struct caller_statistics stats;
2500 gcov_type new_sum, orig_sum;
2501 gcov_type remainder, orig_node_count = orig_node->count;
2503 if (orig_node_count == 0)
2504 return;
2506 init_caller_stats (&stats);
2507 cgraph_for_node_and_aliases (orig_node, gather_caller_stats, &stats, false);
2508 orig_sum = stats.count_sum;
2509 init_caller_stats (&stats);
2510 cgraph_for_node_and_aliases (new_node, gather_caller_stats, &stats, false);
2511 new_sum = stats.count_sum;
2513 if (orig_node_count < orig_sum + new_sum)
2515 if (dump_file)
2516 fprintf (dump_file, " Problem: node %s/%i has too low count "
2517 HOST_WIDE_INT_PRINT_DEC " while the sum of incoming "
2518 "counts is " HOST_WIDE_INT_PRINT_DEC "\n",
2519 cgraph_node_name (orig_node), orig_node->uid,
2520 (HOST_WIDE_INT) orig_node_count,
2521 (HOST_WIDE_INT) (orig_sum + new_sum));
2523 orig_node_count = (orig_sum + new_sum) * 12 / 10;
2524 if (dump_file)
2525 fprintf (dump_file, " proceeding by pretending it was "
2526 HOST_WIDE_INT_PRINT_DEC "\n",
2527 (HOST_WIDE_INT) orig_node_count);
2530 new_node->count = new_sum;
2531 remainder = orig_node_count - new_sum;
2532 orig_node->count = remainder;
2534 for (cs = new_node->callees; cs ; cs = cs->next_callee)
2535 if (cs->frequency)
2536 cs->count = cs->count * (new_sum * REG_BR_PROB_BASE
2537 / orig_node_count) / REG_BR_PROB_BASE;
2538 else
2539 cs->count = 0;
2541 for (cs = orig_node->callees; cs ; cs = cs->next_callee)
2542 cs->count = cs->count * (remainder * REG_BR_PROB_BASE
2543 / orig_node_count) / REG_BR_PROB_BASE;
2545 if (dump_file)
2546 dump_profile_updates (orig_node, new_node);
2549 /* Update the respective profile of specialized NEW_NODE and the original
2550 ORIG_NODE after additional edges with cumulative count sum REDIRECTED_SUM
2551 have been redirected to the specialized version. */
2553 static void
2554 update_specialized_profile (struct cgraph_node *new_node,
2555 struct cgraph_node *orig_node,
2556 gcov_type redirected_sum)
2558 struct cgraph_edge *cs;
2559 gcov_type new_node_count, orig_node_count = orig_node->count;
2561 if (dump_file)
2562 fprintf (dump_file, " the sum of counts of redirected edges is "
2563 HOST_WIDE_INT_PRINT_DEC "\n", (HOST_WIDE_INT) redirected_sum);
2564 if (orig_node_count == 0)
2565 return;
2567 gcc_assert (orig_node_count >= redirected_sum);
2569 new_node_count = new_node->count;
2570 new_node->count += redirected_sum;
2571 orig_node->count -= redirected_sum;
2573 for (cs = new_node->callees; cs ; cs = cs->next_callee)
2574 if (cs->frequency)
2575 cs->count += cs->count * redirected_sum / new_node_count;
2576 else
2577 cs->count = 0;
2579 for (cs = orig_node->callees; cs ; cs = cs->next_callee)
2581 gcov_type dec = cs->count * (redirected_sum * REG_BR_PROB_BASE
2582 / orig_node_count) / REG_BR_PROB_BASE;
2583 if (dec < cs->count)
2584 cs->count -= dec;
2585 else
2586 cs->count = 0;
2589 if (dump_file)
2590 dump_profile_updates (orig_node, new_node);
2593 /* Create a specialized version of NODE with known constants and types of
2594 parameters in KNOWN_VALS and redirect all edges in CALLERS to it. */
2596 static struct cgraph_node *
2597 create_specialized_node (struct cgraph_node *node,
2598 vec<tree> known_vals,
2599 struct ipa_agg_replacement_value *aggvals,
2600 vec<cgraph_edge_p> callers)
2602 struct ipa_node_params *new_info, *info = IPA_NODE_REF (node);
2603 vec<ipa_replace_map_p, va_gc> *replace_trees = NULL;
2604 struct cgraph_node *new_node;
2605 int i, count = ipa_get_param_count (info);
2606 bitmap args_to_skip;
2608 gcc_assert (!info->ipcp_orig_node);
2610 if (node->local.can_change_signature)
2612 args_to_skip = BITMAP_GGC_ALLOC ();
2613 for (i = 0; i < count; i++)
2615 tree t = known_vals[i];
2617 if ((t && TREE_CODE (t) != TREE_BINFO)
2618 || !ipa_is_param_used (info, i))
2619 bitmap_set_bit (args_to_skip, i);
2622 else
2624 args_to_skip = NULL;
2625 if (dump_file && (dump_flags & TDF_DETAILS))
2626 fprintf (dump_file, " cannot change function signature\n");
2629 for (i = 0; i < count ; i++)
2631 tree t = known_vals[i];
2632 if (t && TREE_CODE (t) != TREE_BINFO)
2634 struct ipa_replace_map *replace_map;
2636 replace_map = get_replacement_map (t, ipa_get_param (info, i));
2637 if (replace_map)
2638 vec_safe_push (replace_trees, replace_map);
2642 new_node = cgraph_create_virtual_clone (node, callers, replace_trees,
2643 args_to_skip, "constprop");
2644 ipa_set_node_agg_value_chain (new_node, aggvals);
2645 if (dump_file && (dump_flags & TDF_DETAILS))
2647 fprintf (dump_file, " the new node is %s/%i.\n",
2648 cgraph_node_name (new_node), new_node->uid);
2649 if (aggvals)
2650 ipa_dump_agg_replacement_values (dump_file, aggvals);
2652 gcc_checking_assert (ipa_node_params_vector.exists ()
2653 && (ipa_node_params_vector.length ()
2654 > (unsigned) cgraph_max_uid));
2655 update_profiling_info (node, new_node);
2656 new_info = IPA_NODE_REF (new_node);
2657 new_info->ipcp_orig_node = node;
2658 new_info->known_vals = known_vals;
2660 ipcp_discover_new_direct_edges (new_node, known_vals);
2662 callers.release ();
2663 return new_node;
2666 /* Given a NODE, and a subset of its CALLERS, try to populate blanks slots in
2667 KNOWN_VALS with constants and types that are also known for all of the
2668 CALLERS. */
2670 static void
2671 find_more_scalar_values_for_callers_subset (struct cgraph_node *node,
2672 vec<tree> known_vals,
2673 vec<cgraph_edge_p> callers)
2675 struct ipa_node_params *info = IPA_NODE_REF (node);
2676 int i, count = ipa_get_param_count (info);
2678 for (i = 0; i < count ; i++)
2680 struct cgraph_edge *cs;
2681 tree newval = NULL_TREE;
2682 int j;
2684 if (ipa_get_scalar_lat (info, i)->bottom || known_vals[i])
2685 continue;
2687 FOR_EACH_VEC_ELT (callers, j, cs)
2689 struct ipa_jump_func *jump_func;
2690 tree t;
2692 if (i >= ipa_get_cs_argument_count (IPA_EDGE_REF (cs)))
2694 newval = NULL_TREE;
2695 break;
2697 jump_func = ipa_get_ith_jump_func (IPA_EDGE_REF (cs), i);
2698 t = ipa_value_from_jfunc (IPA_NODE_REF (cs->caller), jump_func);
2699 if (!t
2700 || (newval
2701 && !values_equal_for_ipcp_p (t, newval)))
2703 newval = NULL_TREE;
2704 break;
2706 else
2707 newval = t;
2710 if (newval)
2712 if (dump_file && (dump_flags & TDF_DETAILS))
2714 fprintf (dump_file, " adding an extra known scalar value ");
2715 print_ipcp_constant_value (dump_file, newval);
2716 fprintf (dump_file, " for parameter ");
2717 print_generic_expr (dump_file, ipa_get_param (info, i), 0);
2718 fprintf (dump_file, "\n");
2721 known_vals[i] = newval;
2726 /* Go through PLATS and create a vector of values consisting of values and
2727 offsets (minus OFFSET) of lattices that contain only a single value. */
2729 static vec<ipa_agg_jf_item_t>
2730 copy_plats_to_inter (struct ipcp_param_lattices *plats, HOST_WIDE_INT offset)
2732 vec<ipa_agg_jf_item_t> res = vNULL;
2734 if (!plats->aggs || plats->aggs_contain_variable || plats->aggs_bottom)
2735 return vNULL;
2737 for (struct ipcp_agg_lattice *aglat = plats->aggs; aglat; aglat = aglat->next)
2738 if (ipa_lat_is_single_const (aglat))
2740 struct ipa_agg_jf_item ti;
2741 ti.offset = aglat->offset - offset;
2742 ti.value = aglat->values->value;
2743 res.safe_push (ti);
2745 return res;
2748 /* Intersect all values in INTER with single value lattices in PLATS (while
2749 subtracting OFFSET). */
2751 static void
2752 intersect_with_plats (struct ipcp_param_lattices *plats,
2753 vec<ipa_agg_jf_item_t> *inter,
2754 HOST_WIDE_INT offset)
2756 struct ipcp_agg_lattice *aglat;
2757 struct ipa_agg_jf_item *item;
2758 int k;
2760 if (!plats->aggs || plats->aggs_contain_variable || plats->aggs_bottom)
2762 inter->release ();
2763 return;
2766 aglat = plats->aggs;
2767 FOR_EACH_VEC_ELT (*inter, k, item)
2769 bool found = false;
2770 if (!item->value)
2771 continue;
2772 while (aglat)
2774 if (aglat->offset - offset > item->offset)
2775 break;
2776 if (aglat->offset - offset == item->offset)
2778 gcc_checking_assert (item->value);
2779 if (values_equal_for_ipcp_p (item->value, aglat->values->value))
2780 found = true;
2781 break;
2783 aglat = aglat->next;
2785 if (!found)
2786 item->value = NULL_TREE;
2790 /* Copy agggregate replacement values of NODE (which is an IPA-CP clone) to the
2791 vector result while subtracting OFFSET from the individual value offsets. */
2793 static vec<ipa_agg_jf_item_t>
2794 agg_replacements_to_vector (struct cgraph_node *node, HOST_WIDE_INT offset)
2796 struct ipa_agg_replacement_value *av;
2797 vec<ipa_agg_jf_item_t> res = vNULL;
2799 for (av = ipa_get_agg_replacements_for_node (node); av; av = av->next)
2801 struct ipa_agg_jf_item item;
2802 gcc_checking_assert (av->value);
2803 item.offset = av->offset - offset;
2804 item.value = av->value;
2805 res.safe_push (item);
2808 return res;
2811 /* Intersect all values in INTER with those that we have already scheduled to
2812 be replaced in parameter number INDEX of NODE, which is an IPA-CP clone
2813 (while subtracting OFFSET). */
2815 static void
2816 intersect_with_agg_replacements (struct cgraph_node *node, int index,
2817 vec<ipa_agg_jf_item_t> *inter,
2818 HOST_WIDE_INT offset)
2820 struct ipa_agg_replacement_value *srcvals;
2821 struct ipa_agg_jf_item *item;
2822 int i;
2824 srcvals = ipa_get_agg_replacements_for_node (node);
2825 if (!srcvals)
2827 inter->release ();
2828 return;
2831 FOR_EACH_VEC_ELT (*inter, i, item)
2833 struct ipa_agg_replacement_value *av;
2834 bool found = false;
2835 if (!item->value)
2836 continue;
2837 for (av = srcvals; av; av = av->next)
2839 gcc_checking_assert (av->value);
2840 if (av->index == index
2841 && av->offset - offset == item->offset)
2843 if (values_equal_for_ipcp_p (item->value, av->value))
2844 found = true;
2845 break;
2848 if (!found)
2849 item->value = NULL_TREE;
2853 /* Intersect values in INTER with aggregate values that come along edge CS to
2854 parameter number INDEX and return it. If INTER does not actually exist yet,
2855 copy all incoming values to it. If we determine we ended up with no values
2856 whatsoever, return a released vector. */
2858 static vec<ipa_agg_jf_item_t>
2859 intersect_aggregates_with_edge (struct cgraph_edge *cs, int index,
2860 vec<ipa_agg_jf_item_t> inter)
2862 struct ipa_jump_func *jfunc;
2863 jfunc = ipa_get_ith_jump_func (IPA_EDGE_REF (cs), index);
2864 if (jfunc->type == IPA_JF_PASS_THROUGH
2865 && ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
2867 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
2868 int src_idx = ipa_get_jf_pass_through_formal_id (jfunc);
2870 if (caller_info->ipcp_orig_node)
2872 struct cgraph_node *orig_node = caller_info->ipcp_orig_node;
2873 struct ipcp_param_lattices *orig_plats;
2874 orig_plats = ipa_get_parm_lattices (IPA_NODE_REF (orig_node),
2875 src_idx);
2876 if (agg_pass_through_permissible_p (orig_plats, jfunc))
2878 if (!inter.exists ())
2879 inter = agg_replacements_to_vector (cs->caller, 0);
2880 else
2881 intersect_with_agg_replacements (cs->caller, src_idx,
2882 &inter, 0);
2885 else
2887 struct ipcp_param_lattices *src_plats;
2888 src_plats = ipa_get_parm_lattices (caller_info, src_idx);
2889 if (agg_pass_through_permissible_p (src_plats, jfunc))
2891 /* Currently we do not produce clobber aggregate jump
2892 functions, adjust when we do. */
2893 gcc_checking_assert (!jfunc->agg.items);
2894 if (!inter.exists ())
2895 inter = copy_plats_to_inter (src_plats, 0);
2896 else
2897 intersect_with_plats (src_plats, &inter, 0);
2901 else if (jfunc->type == IPA_JF_ANCESTOR
2902 && ipa_get_jf_ancestor_agg_preserved (jfunc))
2904 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
2905 int src_idx = ipa_get_jf_ancestor_formal_id (jfunc);
2906 struct ipcp_param_lattices *src_plats;
2907 HOST_WIDE_INT delta = ipa_get_jf_ancestor_offset (jfunc);
2909 if (caller_info->ipcp_orig_node)
2911 if (!inter.exists ())
2912 inter = agg_replacements_to_vector (cs->caller, delta);
2913 else
2914 intersect_with_agg_replacements (cs->caller, index, &inter,
2915 delta);
2917 else
2919 src_plats = ipa_get_parm_lattices (caller_info, src_idx);;
2920 /* Currently we do not produce clobber aggregate jump
2921 functions, adjust when we do. */
2922 gcc_checking_assert (!src_plats->aggs || !jfunc->agg.items);
2923 if (!inter.exists ())
2924 inter = copy_plats_to_inter (src_plats, delta);
2925 else
2926 intersect_with_plats (src_plats, &inter, delta);
2929 else if (jfunc->agg.items)
2931 struct ipa_agg_jf_item *item;
2932 int k;
2934 if (!inter.exists ())
2935 for (unsigned i = 0; i < jfunc->agg.items->length (); i++)
2936 inter.safe_push ((*jfunc->agg.items)[i]);
2937 else
2938 FOR_EACH_VEC_ELT (inter, k, item)
2940 int l = 0;
2941 bool found = false;;
2943 if (!item->value)
2944 continue;
2946 while ((unsigned) l < jfunc->agg.items->length ())
2948 struct ipa_agg_jf_item *ti;
2949 ti = &(*jfunc->agg.items)[l];
2950 if (ti->offset > item->offset)
2951 break;
2952 if (ti->offset == item->offset)
2954 gcc_checking_assert (ti->value);
2955 if (values_equal_for_ipcp_p (item->value,
2956 ti->value))
2957 found = true;
2958 break;
2960 l++;
2962 if (!found)
2963 item->value = NULL;
2966 else
2968 inter.release();
2969 return vec<ipa_agg_jf_item_t>();
2971 return inter;
2974 /* Look at edges in CALLERS and collect all known aggregate values that arrive
2975 from all of them. */
2977 static struct ipa_agg_replacement_value *
2978 find_aggregate_values_for_callers_subset (struct cgraph_node *node,
2979 vec<cgraph_edge_p> callers)
2981 struct ipa_node_params *dest_info = IPA_NODE_REF (node);
2982 struct ipa_agg_replacement_value *res = NULL;
2983 struct cgraph_edge *cs;
2984 int i, j, count = ipa_get_param_count (dest_info);
2986 FOR_EACH_VEC_ELT (callers, j, cs)
2988 int c = ipa_get_cs_argument_count (IPA_EDGE_REF (cs));
2989 if (c < count)
2990 count = c;
2993 for (i = 0; i < count ; i++)
2995 struct cgraph_edge *cs;
2996 vec<ipa_agg_jf_item_t> inter = vNULL;
2997 struct ipa_agg_jf_item *item;
2998 int j;
3000 /* Among other things, the following check should deal with all by_ref
3001 mismatches. */
3002 if (ipa_get_parm_lattices (dest_info, i)->aggs_bottom)
3003 continue;
3005 FOR_EACH_VEC_ELT (callers, j, cs)
3007 inter = intersect_aggregates_with_edge (cs, i, inter);
3009 if (!inter.exists ())
3010 goto next_param;
3013 FOR_EACH_VEC_ELT (inter, j, item)
3015 struct ipa_agg_replacement_value *v;
3017 if (!item->value)
3018 continue;
3020 v = ggc_alloc_ipa_agg_replacement_value ();
3021 v->index = i;
3022 v->offset = item->offset;
3023 v->value = item->value;
3024 v->next = res;
3025 res = v;
3028 next_param:
3029 if (inter.exists ())
3030 inter.release ();
3032 return res;
3035 /* Turn KNOWN_AGGS into a list of aggreate replacement values. */
3037 static struct ipa_agg_replacement_value *
3038 known_aggs_to_agg_replacement_list (vec<ipa_agg_jump_function_t> known_aggs)
3040 struct ipa_agg_replacement_value *res = NULL;
3041 struct ipa_agg_jump_function *aggjf;
3042 struct ipa_agg_jf_item *item;
3043 int i, j;
3045 FOR_EACH_VEC_ELT (known_aggs, i, aggjf)
3046 FOR_EACH_VEC_SAFE_ELT (aggjf->items, j, item)
3048 struct ipa_agg_replacement_value *v;
3049 v = ggc_alloc_ipa_agg_replacement_value ();
3050 v->index = i;
3051 v->offset = item->offset;
3052 v->value = item->value;
3053 v->next = res;
3054 res = v;
3056 return res;
3059 /* Determine whether CS also brings all scalar values that the NODE is
3060 specialized for. */
3062 static bool
3063 cgraph_edge_brings_all_scalars_for_node (struct cgraph_edge *cs,
3064 struct cgraph_node *node)
3066 struct ipa_node_params *dest_info = IPA_NODE_REF (node);
3067 int count = ipa_get_param_count (dest_info);
3068 struct ipa_node_params *caller_info;
3069 struct ipa_edge_args *args;
3070 int i;
3072 caller_info = IPA_NODE_REF (cs->caller);
3073 args = IPA_EDGE_REF (cs);
3074 for (i = 0; i < count; i++)
3076 struct ipa_jump_func *jump_func;
3077 tree val, t;
3079 val = dest_info->known_vals[i];
3080 if (!val)
3081 continue;
3083 if (i >= ipa_get_cs_argument_count (args))
3084 return false;
3085 jump_func = ipa_get_ith_jump_func (args, i);
3086 t = ipa_value_from_jfunc (caller_info, jump_func);
3087 if (!t || !values_equal_for_ipcp_p (val, t))
3088 return false;
3090 return true;
3093 /* Determine whether CS also brings all aggregate values that NODE is
3094 specialized for. */
3095 static bool
3096 cgraph_edge_brings_all_agg_vals_for_node (struct cgraph_edge *cs,
3097 struct cgraph_node *node)
3099 struct ipa_node_params *orig_caller_info = IPA_NODE_REF (cs->caller);
3100 struct ipa_agg_replacement_value *aggval;
3101 int i, ec, count;
3103 aggval = ipa_get_agg_replacements_for_node (node);
3104 if (!aggval)
3105 return true;
3107 count = ipa_get_param_count (IPA_NODE_REF (node));
3108 ec = ipa_get_cs_argument_count (IPA_EDGE_REF (cs));
3109 if (ec < count)
3110 for (struct ipa_agg_replacement_value *av = aggval; av; av = av->next)
3111 if (aggval->index >= ec)
3112 return false;
3114 if (orig_caller_info->ipcp_orig_node)
3115 orig_caller_info = IPA_NODE_REF (orig_caller_info->ipcp_orig_node);
3117 for (i = 0; i < count; i++)
3119 static vec<ipa_agg_jf_item_t> values = vec<ipa_agg_jf_item_t>();
3120 struct ipcp_param_lattices *plats;
3121 bool interesting = false;
3122 for (struct ipa_agg_replacement_value *av = aggval; av; av = av->next)
3123 if (aggval->index == i)
3125 interesting = true;
3126 break;
3128 if (!interesting)
3129 continue;
3131 plats = ipa_get_parm_lattices (orig_caller_info, aggval->index);
3132 if (plats->aggs_bottom)
3133 return false;
3135 values = intersect_aggregates_with_edge (cs, i, values);
3136 if (!values.exists())
3137 return false;
3139 for (struct ipa_agg_replacement_value *av = aggval; av; av = av->next)
3140 if (aggval->index == i)
3142 struct ipa_agg_jf_item *item;
3143 int j;
3144 bool found = false;
3145 FOR_EACH_VEC_ELT (values, j, item)
3146 if (item->value
3147 && item->offset == av->offset
3148 && values_equal_for_ipcp_p (item->value, av->value))
3149 found = true;
3150 if (!found)
3152 values.release();
3153 return false;
3157 return true;
3160 /* Given an original NODE and a VAL for which we have already created a
3161 specialized clone, look whether there are incoming edges that still lead
3162 into the old node but now also bring the requested value and also conform to
3163 all other criteria such that they can be redirected the the special node.
3164 This function can therefore redirect the final edge in a SCC. */
3166 static void
3167 perhaps_add_new_callers (struct cgraph_node *node, struct ipcp_value *val)
3169 struct ipcp_value_source *src;
3170 gcov_type redirected_sum = 0;
3172 for (src = val->sources; src; src = src->next)
3174 struct cgraph_edge *cs = src->cs;
3175 while (cs)
3177 enum availability availability;
3178 struct cgraph_node *dst = cgraph_function_node (cs->callee,
3179 &availability);
3180 if ((dst == node || IPA_NODE_REF (dst)->is_all_contexts_clone)
3181 && availability > AVAIL_OVERWRITABLE
3182 && cgraph_edge_brings_value_p (cs, src))
3184 if (cgraph_edge_brings_all_scalars_for_node (cs, val->spec_node)
3185 && cgraph_edge_brings_all_agg_vals_for_node (cs,
3186 val->spec_node))
3188 if (dump_file)
3189 fprintf (dump_file, " - adding an extra caller %s/%i"
3190 " of %s/%i\n",
3191 xstrdup (cgraph_node_name (cs->caller)),
3192 cs->caller->uid,
3193 xstrdup (cgraph_node_name (val->spec_node)),
3194 val->spec_node->uid);
3196 cgraph_redirect_edge_callee (cs, val->spec_node);
3197 redirected_sum += cs->count;
3200 cs = get_next_cgraph_edge_clone (cs);
3204 if (redirected_sum)
3205 update_specialized_profile (val->spec_node, node, redirected_sum);
3209 /* Copy KNOWN_BINFOS to KNOWN_VALS. */
3211 static void
3212 move_binfos_to_values (vec<tree> known_vals,
3213 vec<tree> known_binfos)
3215 tree t;
3216 int i;
3218 for (i = 0; known_binfos.iterate (i, &t); i++)
3219 if (t)
3220 known_vals[i] = t;
3223 /* Return true if there is a replacement equivalent to VALUE, INDEX and OFFSET
3224 among those in the AGGVALS list. */
3226 DEBUG_FUNCTION bool
3227 ipcp_val_in_agg_replacements_p (struct ipa_agg_replacement_value *aggvals,
3228 int index, HOST_WIDE_INT offset, tree value)
3230 while (aggvals)
3232 if (aggvals->index == index
3233 && aggvals->offset == offset
3234 && values_equal_for_ipcp_p (aggvals->value, value))
3235 return true;
3236 aggvals = aggvals->next;
3238 return false;
3241 /* Decide wheter to create a special version of NODE for value VAL of parameter
3242 at the given INDEX. If OFFSET is -1, the value is for the parameter itself,
3243 otherwise it is stored at the given OFFSET of the parameter. KNOWN_CSTS,
3244 KNOWN_BINFOS and KNOWN_AGGS describe the other already known values. */
3246 static bool
3247 decide_about_value (struct cgraph_node *node, int index, HOST_WIDE_INT offset,
3248 struct ipcp_value *val, vec<tree> known_csts,
3249 vec<tree> known_binfos)
3251 struct ipa_agg_replacement_value *aggvals;
3252 int freq_sum, caller_count;
3253 gcov_type count_sum;
3254 vec<cgraph_edge_p> callers;
3255 vec<tree> kv;
3257 if (val->spec_node)
3259 perhaps_add_new_callers (node, val);
3260 return false;
3262 else if (val->local_size_cost + overall_size > max_new_size)
3264 if (dump_file && (dump_flags & TDF_DETAILS))
3265 fprintf (dump_file, " Ignoring candidate value because "
3266 "max_new_size would be reached with %li.\n",
3267 val->local_size_cost + overall_size);
3268 return false;
3270 else if (!get_info_about_necessary_edges (val, &freq_sum, &count_sum,
3271 &caller_count))
3272 return false;
3274 if (dump_file && (dump_flags & TDF_DETAILS))
3276 fprintf (dump_file, " - considering value ");
3277 print_ipcp_constant_value (dump_file, val->value);
3278 fprintf (dump_file, " for parameter ");
3279 print_generic_expr (dump_file, ipa_get_param (IPA_NODE_REF (node),
3280 index), 0);
3281 if (offset != -1)
3282 fprintf (dump_file, ", offset: " HOST_WIDE_INT_PRINT_DEC, offset);
3283 fprintf (dump_file, " (caller_count: %i)\n", caller_count);
3286 if (!good_cloning_opportunity_p (node, val->local_time_benefit,
3287 freq_sum, count_sum,
3288 val->local_size_cost)
3289 && !good_cloning_opportunity_p (node,
3290 val->local_time_benefit
3291 + val->prop_time_benefit,
3292 freq_sum, count_sum,
3293 val->local_size_cost
3294 + val->prop_size_cost))
3295 return false;
3297 if (dump_file)
3298 fprintf (dump_file, " Creating a specialized node of %s/%i.\n",
3299 cgraph_node_name (node), node->uid);
3301 callers = gather_edges_for_value (val, caller_count);
3302 kv = known_csts.copy ();
3303 move_binfos_to_values (kv, known_binfos);
3304 if (offset == -1)
3305 kv[index] = val->value;
3306 find_more_scalar_values_for_callers_subset (node, kv, callers);
3307 aggvals = find_aggregate_values_for_callers_subset (node, callers);
3308 gcc_checking_assert (offset == -1
3309 || ipcp_val_in_agg_replacements_p (aggvals, index,
3310 offset, val->value));
3311 val->spec_node = create_specialized_node (node, kv, aggvals, callers);
3312 overall_size += val->local_size_cost;
3314 /* TODO: If for some lattice there is only one other known value
3315 left, make a special node for it too. */
3317 return true;
3320 /* Decide whether and what specialized clones of NODE should be created. */
3322 static bool
3323 decide_whether_version_node (struct cgraph_node *node)
3325 struct ipa_node_params *info = IPA_NODE_REF (node);
3326 int i, count = ipa_get_param_count (info);
3327 vec<tree> known_csts, known_binfos;
3328 vec<ipa_agg_jump_function_t> known_aggs = vNULL;
3329 bool ret = false;
3331 if (count == 0)
3332 return false;
3334 if (dump_file && (dump_flags & TDF_DETAILS))
3335 fprintf (dump_file, "\nEvaluating opportunities for %s/%i.\n",
3336 cgraph_node_name (node), node->uid);
3338 gather_context_independent_values (info, &known_csts, &known_binfos,
3339 info->do_clone_for_all_contexts ? &known_aggs
3340 : NULL, NULL);
3342 for (i = 0; i < count ;i++)
3344 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
3345 struct ipcp_lattice *lat = &plats->itself;
3346 struct ipcp_value *val;
3348 if (!lat->bottom
3349 && !known_csts[i]
3350 && !known_binfos[i])
3351 for (val = lat->values; val; val = val->next)
3352 ret |= decide_about_value (node, i, -1, val, known_csts,
3353 known_binfos);
3355 if (!plats->aggs_bottom)
3357 struct ipcp_agg_lattice *aglat;
3358 struct ipcp_value *val;
3359 for (aglat = plats->aggs; aglat; aglat = aglat->next)
3360 if (!aglat->bottom && aglat->values
3361 /* If the following is false, the one value is in
3362 known_aggs. */
3363 && (plats->aggs_contain_variable
3364 || !ipa_lat_is_single_const (aglat)))
3365 for (val = aglat->values; val; val = val->next)
3366 ret |= decide_about_value (node, i, aglat->offset, val,
3367 known_csts, known_binfos);
3369 info = IPA_NODE_REF (node);
3372 if (info->do_clone_for_all_contexts)
3374 struct cgraph_node *clone;
3375 vec<cgraph_edge_p> callers;
3377 if (dump_file)
3378 fprintf (dump_file, " - Creating a specialized node of %s/%i "
3379 "for all known contexts.\n", cgraph_node_name (node),
3380 node->uid);
3382 callers = collect_callers_of_node (node);
3383 move_binfos_to_values (known_csts, known_binfos);
3384 clone = create_specialized_node (node, known_csts,
3385 known_aggs_to_agg_replacement_list (known_aggs),
3386 callers);
3387 info = IPA_NODE_REF (node);
3388 info->do_clone_for_all_contexts = false;
3389 IPA_NODE_REF (clone)->is_all_contexts_clone = true;
3390 ret = true;
3392 else
3393 known_csts.release ();
3395 known_binfos.release ();
3396 return ret;
3399 /* Transitively mark all callees of NODE within the same SCC as not dead. */
3401 static void
3402 spread_undeadness (struct cgraph_node *node)
3404 struct cgraph_edge *cs;
3406 for (cs = node->callees; cs; cs = cs->next_callee)
3407 if (edge_within_scc (cs))
3409 struct cgraph_node *callee;
3410 struct ipa_node_params *info;
3412 callee = cgraph_function_node (cs->callee, NULL);
3413 info = IPA_NODE_REF (callee);
3415 if (info->node_dead)
3417 info->node_dead = 0;
3418 spread_undeadness (callee);
3423 /* Return true if NODE has a caller from outside of its SCC that is not
3424 dead. Worker callback for cgraph_for_node_and_aliases. */
3426 static bool
3427 has_undead_caller_from_outside_scc_p (struct cgraph_node *node,
3428 void *data ATTRIBUTE_UNUSED)
3430 struct cgraph_edge *cs;
3432 for (cs = node->callers; cs; cs = cs->next_caller)
3433 if (cs->caller->thunk.thunk_p
3434 && cgraph_for_node_and_aliases (cs->caller,
3435 has_undead_caller_from_outside_scc_p,
3436 NULL, true))
3437 return true;
3438 else if (!edge_within_scc (cs)
3439 && !IPA_NODE_REF (cs->caller)->node_dead)
3440 return true;
3441 return false;
3445 /* Identify nodes within the same SCC as NODE which are no longer needed
3446 because of new clones and will be removed as unreachable. */
3448 static void
3449 identify_dead_nodes (struct cgraph_node *node)
3451 struct cgraph_node *v;
3452 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3453 if (cgraph_will_be_removed_from_program_if_no_direct_calls (v)
3454 && !cgraph_for_node_and_aliases (v,
3455 has_undead_caller_from_outside_scc_p,
3456 NULL, true))
3457 IPA_NODE_REF (v)->node_dead = 1;
3459 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3460 if (!IPA_NODE_REF (v)->node_dead)
3461 spread_undeadness (v);
3463 if (dump_file && (dump_flags & TDF_DETAILS))
3465 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3466 if (IPA_NODE_REF (v)->node_dead)
3467 fprintf (dump_file, " Marking node as dead: %s/%i.\n",
3468 cgraph_node_name (v), v->uid);
3472 /* The decision stage. Iterate over the topological order of call graph nodes
3473 TOPO and make specialized clones if deemed beneficial. */
3475 static void
3476 ipcp_decision_stage (struct topo_info *topo)
3478 int i;
3480 if (dump_file)
3481 fprintf (dump_file, "\nIPA decision stage:\n\n");
3483 for (i = topo->nnodes - 1; i >= 0; i--)
3485 struct cgraph_node *node = topo->order[i];
3486 bool change = false, iterate = true;
3488 while (iterate)
3490 struct cgraph_node *v;
3491 iterate = false;
3492 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3493 if (cgraph_function_with_gimple_body_p (v)
3494 && ipcp_versionable_function_p (v))
3495 iterate |= decide_whether_version_node (v);
3497 change |= iterate;
3499 if (change)
3500 identify_dead_nodes (node);
3504 /* The IPCP driver. */
3506 static unsigned int
3507 ipcp_driver (void)
3509 struct cgraph_2edge_hook_list *edge_duplication_hook_holder;
3510 struct topo_info topo;
3512 ipa_check_create_node_params ();
3513 ipa_check_create_edge_args ();
3514 grow_next_edge_clone_vector ();
3515 edge_duplication_hook_holder =
3516 cgraph_add_edge_duplication_hook (&ipcp_edge_duplication_hook, NULL);
3517 ipcp_values_pool = create_alloc_pool ("IPA-CP values",
3518 sizeof (struct ipcp_value), 32);
3519 ipcp_sources_pool = create_alloc_pool ("IPA-CP value sources",
3520 sizeof (struct ipcp_value_source), 64);
3521 ipcp_agg_lattice_pool = create_alloc_pool ("IPA_CP aggregate lattices",
3522 sizeof (struct ipcp_agg_lattice),
3523 32);
3524 if (dump_file)
3526 fprintf (dump_file, "\nIPA structures before propagation:\n");
3527 if (dump_flags & TDF_DETAILS)
3528 ipa_print_all_params (dump_file);
3529 ipa_print_all_jump_functions (dump_file);
3532 /* Topological sort. */
3533 build_toporder_info (&topo);
3534 /* Do the interprocedural propagation. */
3535 ipcp_propagate_stage (&topo);
3536 /* Decide what constant propagation and cloning should be performed. */
3537 ipcp_decision_stage (&topo);
3539 /* Free all IPCP structures. */
3540 free_toporder_info (&topo);
3541 next_edge_clone.release ();
3542 cgraph_remove_edge_duplication_hook (edge_duplication_hook_holder);
3543 ipa_free_all_structures_after_ipa_cp ();
3544 if (dump_file)
3545 fprintf (dump_file, "\nIPA constant propagation end\n");
3546 return 0;
3549 /* Initialization and computation of IPCP data structures. This is the initial
3550 intraprocedural analysis of functions, which gathers information to be
3551 propagated later on. */
3553 static void
3554 ipcp_generate_summary (void)
3556 struct cgraph_node *node;
3558 if (dump_file)
3559 fprintf (dump_file, "\nIPA constant propagation start:\n");
3560 ipa_register_cgraph_hooks ();
3562 FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node)
3564 node->local.versionable
3565 = tree_versionable_function_p (node->symbol.decl);
3566 ipa_analyze_node (node);
3570 /* Write ipcp summary for nodes in SET. */
3572 static void
3573 ipcp_write_summary (void)
3575 ipa_prop_write_jump_functions ();
3578 /* Read ipcp summary. */
3580 static void
3581 ipcp_read_summary (void)
3583 ipa_prop_read_jump_functions ();
3586 /* Gate for IPCP optimization. */
3588 static bool
3589 cgraph_gate_cp (void)
3591 /* FIXME: We should remove the optimize check after we ensure we never run
3592 IPA passes when not optimizing. */
3593 return flag_ipa_cp && optimize;
3596 struct ipa_opt_pass_d pass_ipa_cp =
3599 IPA_PASS,
3600 "cp", /* name */
3601 OPTGROUP_NONE, /* optinfo_flags */
3602 cgraph_gate_cp, /* gate */
3603 ipcp_driver, /* execute */
3604 NULL, /* sub */
3605 NULL, /* next */
3606 0, /* static_pass_number */
3607 TV_IPA_CONSTANT_PROP, /* tv_id */
3608 0, /* properties_required */
3609 0, /* properties_provided */
3610 0, /* properties_destroyed */
3611 0, /* todo_flags_start */
3612 TODO_dump_symtab |
3613 TODO_remove_functions | TODO_ggc_collect /* todo_flags_finish */
3615 ipcp_generate_summary, /* generate_summary */
3616 ipcp_write_summary, /* write_summary */
3617 ipcp_read_summary, /* read_summary */
3618 ipa_prop_write_all_agg_replacement, /* write_optimization_summary */
3619 ipa_prop_read_all_agg_replacement, /* read_optimization_summary */
3620 NULL, /* stmt_fixup */
3621 0, /* TODOs */
3622 ipcp_transform_function, /* function_transform */
3623 NULL, /* variable_transform */