[AArch64] Rewrite v<max,min><nm><q><v>_<sfu><8, 16, 32, 64> intrinsics using builtins.
[official-gcc.git] / gcc / ipa-cp.c
blob48521cfffbf380c4115c010b2c58ae86601dd410
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 /* Determine whether JFUNC evaluates to a known value (that is either a
795 constant or a binfo) and if so, return it. Otherwise return NULL. INFO
796 describes the caller node so that pass-through jump functions can be
797 evaluated. */
799 tree
800 ipa_value_from_jfunc (struct ipa_node_params *info, struct ipa_jump_func *jfunc)
802 if (jfunc->type == IPA_JF_CONST)
803 return ipa_get_jf_constant (jfunc);
804 else if (jfunc->type == IPA_JF_KNOWN_TYPE)
805 return ipa_binfo_from_known_type_jfunc (jfunc);
806 else if (jfunc->type == IPA_JF_PASS_THROUGH
807 || jfunc->type == IPA_JF_ANCESTOR)
809 tree input;
810 int idx;
812 if (jfunc->type == IPA_JF_PASS_THROUGH)
813 idx = ipa_get_jf_pass_through_formal_id (jfunc);
814 else
815 idx = ipa_get_jf_ancestor_formal_id (jfunc);
817 if (info->ipcp_orig_node)
818 input = info->known_vals[idx];
819 else
821 struct ipcp_lattice *lat;
823 if (!info->lattices)
825 gcc_checking_assert (!flag_ipa_cp);
826 return NULL_TREE;
828 lat = ipa_get_scalar_lat (info, idx);
829 if (!ipa_lat_is_single_const (lat))
830 return NULL_TREE;
831 input = lat->values->value;
834 if (!input)
835 return NULL_TREE;
837 if (jfunc->type == IPA_JF_PASS_THROUGH)
838 return ipa_get_jf_pass_through_result (jfunc, input);
839 else
840 return ipa_get_jf_ancestor_result (jfunc, input);
842 else
843 return NULL_TREE;
847 /* If checking is enabled, verify that no lattice is in the TOP state, i.e. not
848 bottom, not containing a variable component and without any known value at
849 the same time. */
851 DEBUG_FUNCTION void
852 ipcp_verify_propagated_values (void)
854 struct cgraph_node *node;
856 FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node)
858 struct ipa_node_params *info = IPA_NODE_REF (node);
859 int i, count = ipa_get_param_count (info);
861 for (i = 0; i < count; i++)
863 struct ipcp_lattice *lat = ipa_get_scalar_lat (info, i);
865 if (!lat->bottom
866 && !lat->contains_variable
867 && lat->values_count == 0)
869 if (dump_file)
871 fprintf (dump_file, "\nIPA lattices after constant "
872 "propagation:\n");
873 print_all_lattices (dump_file, true, false);
876 gcc_unreachable ();
882 /* Return true iff X and Y should be considered equal values by IPA-CP. */
884 static bool
885 values_equal_for_ipcp_p (tree x, tree y)
887 gcc_checking_assert (x != NULL_TREE && y != NULL_TREE);
889 if (x == y)
890 return true;
892 if (TREE_CODE (x) == TREE_BINFO || TREE_CODE (y) == TREE_BINFO)
893 return false;
895 if (TREE_CODE (x) == ADDR_EXPR
896 && TREE_CODE (y) == ADDR_EXPR
897 && TREE_CODE (TREE_OPERAND (x, 0)) == CONST_DECL
898 && TREE_CODE (TREE_OPERAND (y, 0)) == CONST_DECL)
899 return operand_equal_p (DECL_INITIAL (TREE_OPERAND (x, 0)),
900 DECL_INITIAL (TREE_OPERAND (y, 0)), 0);
901 else
902 return operand_equal_p (x, y, 0);
905 /* Add a new value source to VAL, marking that a value comes from edge CS and
906 (if the underlying jump function is a pass-through or an ancestor one) from
907 a caller value SRC_VAL of a caller parameter described by SRC_INDEX. OFFSET
908 is negative if the source was the scalar value of the parameter itself or
909 the offset within an aggregate. */
911 static void
912 add_value_source (struct ipcp_value *val, struct cgraph_edge *cs,
913 struct ipcp_value *src_val, int src_idx, HOST_WIDE_INT offset)
915 struct ipcp_value_source *src;
917 src = (struct ipcp_value_source *) pool_alloc (ipcp_sources_pool);
918 src->offset = offset;
919 src->cs = cs;
920 src->val = src_val;
921 src->index = src_idx;
923 src->next = val->sources;
924 val->sources = src;
927 /* Try to add NEWVAL to LAT, potentially creating a new struct ipcp_value for
928 it. CS, SRC_VAL SRC_INDEX and OFFSET are meant for add_value_source and
929 have the same meaning. */
931 static bool
932 add_value_to_lattice (struct ipcp_lattice *lat, tree newval,
933 struct cgraph_edge *cs, struct ipcp_value *src_val,
934 int src_idx, HOST_WIDE_INT offset)
936 struct ipcp_value *val;
938 if (lat->bottom)
939 return false;
941 for (val = lat->values; val; val = val->next)
942 if (values_equal_for_ipcp_p (val->value, newval))
944 if (edge_within_scc (cs))
946 struct ipcp_value_source *s;
947 for (s = val->sources; s ; s = s->next)
948 if (s->cs == cs)
949 break;
950 if (s)
951 return false;
954 add_value_source (val, cs, src_val, src_idx, offset);
955 return false;
958 if (lat->values_count == PARAM_VALUE (PARAM_IPA_CP_VALUE_LIST_SIZE))
960 /* We can only free sources, not the values themselves, because sources
961 of other values in this this SCC might point to them. */
962 for (val = lat->values; val; val = val->next)
964 while (val->sources)
966 struct ipcp_value_source *src = val->sources;
967 val->sources = src->next;
968 pool_free (ipcp_sources_pool, src);
972 lat->values = NULL;
973 return set_lattice_to_bottom (lat);
976 lat->values_count++;
977 val = (struct ipcp_value *) pool_alloc (ipcp_values_pool);
978 memset (val, 0, sizeof (*val));
980 add_value_source (val, cs, src_val, src_idx, offset);
981 val->value = newval;
982 val->next = lat->values;
983 lat->values = val;
984 return true;
987 /* Like above but passes a special value of offset to distinguish that the
988 origin is the scalar value of the parameter rather than a part of an
989 aggregate. */
991 static inline bool
992 add_scalar_value_to_lattice (struct ipcp_lattice *lat, tree newval,
993 struct cgraph_edge *cs,
994 struct ipcp_value *src_val, int src_idx)
996 return add_value_to_lattice (lat, newval, cs, src_val, src_idx, -1);
999 /* Propagate values through a pass-through jump function JFUNC associated with
1000 edge CS, taking values from SRC_LAT and putting them into DEST_LAT. SRC_IDX
1001 is the index of the source parameter. */
1003 static bool
1004 propagate_vals_accross_pass_through (struct cgraph_edge *cs,
1005 struct ipa_jump_func *jfunc,
1006 struct ipcp_lattice *src_lat,
1007 struct ipcp_lattice *dest_lat,
1008 int src_idx)
1010 struct ipcp_value *src_val;
1011 bool ret = false;
1013 if (ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
1014 for (src_val = src_lat->values; src_val; src_val = src_val->next)
1015 ret |= add_scalar_value_to_lattice (dest_lat, src_val->value, cs,
1016 src_val, src_idx);
1017 /* Do not create new values when propagating within an SCC because if there
1018 are arithmetic functions with circular dependencies, there is infinite
1019 number of them and we would just make lattices bottom. */
1020 else if (edge_within_scc (cs))
1021 ret = set_lattice_contains_variable (dest_lat);
1022 else
1023 for (src_val = src_lat->values; src_val; src_val = src_val->next)
1025 tree cstval = src_val->value;
1027 if (TREE_CODE (cstval) == TREE_BINFO)
1029 ret |= set_lattice_contains_variable (dest_lat);
1030 continue;
1032 cstval = ipa_get_jf_pass_through_result (jfunc, cstval);
1034 if (cstval)
1035 ret |= add_scalar_value_to_lattice (dest_lat, cstval, cs, src_val,
1036 src_idx);
1037 else
1038 ret |= set_lattice_contains_variable (dest_lat);
1041 return ret;
1044 /* Propagate values through an ancestor jump function JFUNC associated with
1045 edge CS, taking values from SRC_LAT and putting them into DEST_LAT. SRC_IDX
1046 is the index of the source parameter. */
1048 static bool
1049 propagate_vals_accross_ancestor (struct cgraph_edge *cs,
1050 struct ipa_jump_func *jfunc,
1051 struct ipcp_lattice *src_lat,
1052 struct ipcp_lattice *dest_lat,
1053 int src_idx)
1055 struct ipcp_value *src_val;
1056 bool ret = false;
1058 if (edge_within_scc (cs))
1059 return set_lattice_contains_variable (dest_lat);
1061 for (src_val = src_lat->values; src_val; src_val = src_val->next)
1063 tree t = ipa_get_jf_ancestor_result (jfunc, src_val->value);
1065 if (t)
1066 ret |= add_scalar_value_to_lattice (dest_lat, t, cs, src_val, src_idx);
1067 else
1068 ret |= set_lattice_contains_variable (dest_lat);
1071 return ret;
1074 /* Propagate scalar values across jump function JFUNC that is associated with
1075 edge CS and put the values into DEST_LAT. */
1077 static bool
1078 propagate_scalar_accross_jump_function (struct cgraph_edge *cs,
1079 struct ipa_jump_func *jfunc,
1080 struct ipcp_lattice *dest_lat)
1082 if (dest_lat->bottom)
1083 return false;
1085 if (jfunc->type == IPA_JF_CONST
1086 || jfunc->type == IPA_JF_KNOWN_TYPE)
1088 tree val;
1090 if (jfunc->type == IPA_JF_KNOWN_TYPE)
1092 val = ipa_binfo_from_known_type_jfunc (jfunc);
1093 if (!val)
1094 return set_lattice_contains_variable (dest_lat);
1096 else
1097 val = ipa_get_jf_constant (jfunc);
1098 return add_scalar_value_to_lattice (dest_lat, val, cs, NULL, 0);
1100 else if (jfunc->type == IPA_JF_PASS_THROUGH
1101 || jfunc->type == IPA_JF_ANCESTOR)
1103 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
1104 struct ipcp_lattice *src_lat;
1105 int src_idx;
1106 bool ret;
1108 if (jfunc->type == IPA_JF_PASS_THROUGH)
1109 src_idx = ipa_get_jf_pass_through_formal_id (jfunc);
1110 else
1111 src_idx = ipa_get_jf_ancestor_formal_id (jfunc);
1113 src_lat = ipa_get_scalar_lat (caller_info, src_idx);
1114 if (src_lat->bottom)
1115 return set_lattice_contains_variable (dest_lat);
1117 /* If we would need to clone the caller and cannot, do not propagate. */
1118 if (!ipcp_versionable_function_p (cs->caller)
1119 && (src_lat->contains_variable
1120 || (src_lat->values_count > 1)))
1121 return set_lattice_contains_variable (dest_lat);
1123 if (jfunc->type == IPA_JF_PASS_THROUGH)
1124 ret = propagate_vals_accross_pass_through (cs, jfunc, src_lat,
1125 dest_lat, src_idx);
1126 else
1127 ret = propagate_vals_accross_ancestor (cs, jfunc, src_lat, dest_lat,
1128 src_idx);
1130 if (src_lat->contains_variable)
1131 ret |= set_lattice_contains_variable (dest_lat);
1133 return ret;
1136 /* TODO: We currently do not handle member method pointers in IPA-CP (we only
1137 use it for indirect inlining), we should propagate them too. */
1138 return set_lattice_contains_variable (dest_lat);
1141 /* If DEST_PLATS already has aggregate items, check that aggs_by_ref matches
1142 NEW_AGGS_BY_REF and if not, mark all aggs as bottoms and return true (in all
1143 other cases, return false). If there are no aggregate items, set
1144 aggs_by_ref to NEW_AGGS_BY_REF. */
1146 static bool
1147 set_check_aggs_by_ref (struct ipcp_param_lattices *dest_plats,
1148 bool new_aggs_by_ref)
1150 if (dest_plats->aggs)
1152 if (dest_plats->aggs_by_ref != new_aggs_by_ref)
1154 set_agg_lats_to_bottom (dest_plats);
1155 return true;
1158 else
1159 dest_plats->aggs_by_ref = new_aggs_by_ref;
1160 return false;
1163 /* Walk aggregate lattices in DEST_PLATS from ***AGLAT on, until ***aglat is an
1164 already existing lattice for the given OFFSET and SIZE, marking all skipped
1165 lattices as containing variable and checking for overlaps. If there is no
1166 already existing lattice for the OFFSET and VAL_SIZE, create one, initialize
1167 it with offset, size and contains_variable to PRE_EXISTING, and return true,
1168 unless there are too many already. If there are two many, return false. If
1169 there are overlaps turn whole DEST_PLATS to bottom and return false. If any
1170 skipped lattices were newly marked as containing variable, set *CHANGE to
1171 true. */
1173 static bool
1174 merge_agg_lats_step (struct ipcp_param_lattices *dest_plats,
1175 HOST_WIDE_INT offset, HOST_WIDE_INT val_size,
1176 struct ipcp_agg_lattice ***aglat,
1177 bool pre_existing, bool *change)
1179 gcc_checking_assert (offset >= 0);
1181 while (**aglat && (**aglat)->offset < offset)
1183 if ((**aglat)->offset + (**aglat)->size > offset)
1185 set_agg_lats_to_bottom (dest_plats);
1186 return false;
1188 *change |= set_lattice_contains_variable (**aglat);
1189 *aglat = &(**aglat)->next;
1192 if (**aglat && (**aglat)->offset == offset)
1194 if ((**aglat)->size != val_size
1195 || ((**aglat)->next
1196 && (**aglat)->next->offset < offset + val_size))
1198 set_agg_lats_to_bottom (dest_plats);
1199 return false;
1201 gcc_checking_assert (!(**aglat)->next
1202 || (**aglat)->next->offset >= offset + val_size);
1203 return true;
1205 else
1207 struct ipcp_agg_lattice *new_al;
1209 if (**aglat && (**aglat)->offset < offset + val_size)
1211 set_agg_lats_to_bottom (dest_plats);
1212 return false;
1214 if (dest_plats->aggs_count == PARAM_VALUE (PARAM_IPA_MAX_AGG_ITEMS))
1215 return false;
1216 dest_plats->aggs_count++;
1217 new_al = (struct ipcp_agg_lattice *) pool_alloc (ipcp_agg_lattice_pool);
1218 memset (new_al, 0, sizeof (*new_al));
1220 new_al->offset = offset;
1221 new_al->size = val_size;
1222 new_al->contains_variable = pre_existing;
1224 new_al->next = **aglat;
1225 **aglat = new_al;
1226 return true;
1230 /* Set all AGLAT and all other aggregate lattices reachable by next pointers as
1231 containing an unknown value. */
1233 static bool
1234 set_chain_of_aglats_contains_variable (struct ipcp_agg_lattice *aglat)
1236 bool ret = false;
1237 while (aglat)
1239 ret |= set_lattice_contains_variable (aglat);
1240 aglat = aglat->next;
1242 return ret;
1245 /* Merge existing aggregate lattices in SRC_PLATS to DEST_PLATS, subtracting
1246 DELTA_OFFSET. CS is the call graph edge and SRC_IDX the index of the source
1247 parameter used for lattice value sources. Return true if DEST_PLATS changed
1248 in any way. */
1250 static bool
1251 merge_aggregate_lattices (struct cgraph_edge *cs,
1252 struct ipcp_param_lattices *dest_plats,
1253 struct ipcp_param_lattices *src_plats,
1254 int src_idx, HOST_WIDE_INT offset_delta)
1256 bool pre_existing = dest_plats->aggs != NULL;
1257 struct ipcp_agg_lattice **dst_aglat;
1258 bool ret = false;
1260 if (set_check_aggs_by_ref (dest_plats, src_plats->aggs_by_ref))
1261 return true;
1262 if (src_plats->aggs_bottom)
1263 return set_agg_lats_contain_variable (dest_plats);
1264 if (src_plats->aggs_contain_variable)
1265 ret |= set_agg_lats_contain_variable (dest_plats);
1266 dst_aglat = &dest_plats->aggs;
1268 for (struct ipcp_agg_lattice *src_aglat = src_plats->aggs;
1269 src_aglat;
1270 src_aglat = src_aglat->next)
1272 HOST_WIDE_INT new_offset = src_aglat->offset - offset_delta;
1274 if (new_offset < 0)
1275 continue;
1276 if (merge_agg_lats_step (dest_plats, new_offset, src_aglat->size,
1277 &dst_aglat, pre_existing, &ret))
1279 struct ipcp_agg_lattice *new_al = *dst_aglat;
1281 dst_aglat = &(*dst_aglat)->next;
1282 if (src_aglat->bottom)
1284 ret |= set_lattice_contains_variable (new_al);
1285 continue;
1287 if (src_aglat->contains_variable)
1288 ret |= set_lattice_contains_variable (new_al);
1289 for (struct ipcp_value *val = src_aglat->values;
1290 val;
1291 val = val->next)
1292 ret |= add_value_to_lattice (new_al, val->value, cs, val, src_idx,
1293 src_aglat->offset);
1295 else if (dest_plats->aggs_bottom)
1296 return true;
1298 ret |= set_chain_of_aglats_contains_variable (*dst_aglat);
1299 return ret;
1302 /* Determine whether there is anything to propagate FROM SRC_PLATS through a
1303 pass-through JFUNC and if so, whether it has conform and conforms to the
1304 rules about propagating values passed by reference. */
1306 static bool
1307 agg_pass_through_permissible_p (struct ipcp_param_lattices *src_plats,
1308 struct ipa_jump_func *jfunc)
1310 return src_plats->aggs
1311 && (!src_plats->aggs_by_ref
1312 || ipa_get_jf_pass_through_agg_preserved (jfunc));
1315 /* Propagate scalar values across jump function JFUNC that is associated with
1316 edge CS and put the values into DEST_LAT. */
1318 static bool
1319 propagate_aggs_accross_jump_function (struct cgraph_edge *cs,
1320 struct ipa_jump_func *jfunc,
1321 struct ipcp_param_lattices *dest_plats)
1323 bool ret = false;
1325 if (dest_plats->aggs_bottom)
1326 return false;
1328 if (jfunc->type == IPA_JF_PASS_THROUGH
1329 && ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
1331 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
1332 int src_idx = ipa_get_jf_pass_through_formal_id (jfunc);
1333 struct ipcp_param_lattices *src_plats;
1335 src_plats = ipa_get_parm_lattices (caller_info, src_idx);
1336 if (agg_pass_through_permissible_p (src_plats, jfunc))
1338 /* Currently we do not produce clobber aggregate jump
1339 functions, replace with merging when we do. */
1340 gcc_assert (!jfunc->agg.items);
1341 ret |= merge_aggregate_lattices (cs, dest_plats, src_plats,
1342 src_idx, 0);
1344 else
1345 ret |= set_agg_lats_contain_variable (dest_plats);
1347 else if (jfunc->type == IPA_JF_ANCESTOR
1348 && ipa_get_jf_ancestor_agg_preserved (jfunc))
1350 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
1351 int src_idx = ipa_get_jf_ancestor_formal_id (jfunc);
1352 struct ipcp_param_lattices *src_plats;
1354 src_plats = ipa_get_parm_lattices (caller_info, src_idx);
1355 if (src_plats->aggs && src_plats->aggs_by_ref)
1357 /* Currently we do not produce clobber aggregate jump
1358 functions, replace with merging when we do. */
1359 gcc_assert (!jfunc->agg.items);
1360 ret |= merge_aggregate_lattices (cs, dest_plats, src_plats, src_idx,
1361 ipa_get_jf_ancestor_offset (jfunc));
1363 else if (!src_plats->aggs_by_ref)
1364 ret |= set_agg_lats_to_bottom (dest_plats);
1365 else
1366 ret |= set_agg_lats_contain_variable (dest_plats);
1368 else if (jfunc->agg.items)
1370 bool pre_existing = dest_plats->aggs != NULL;
1371 struct ipcp_agg_lattice **aglat = &dest_plats->aggs;
1372 struct ipa_agg_jf_item *item;
1373 int i;
1375 if (set_check_aggs_by_ref (dest_plats, jfunc->agg.by_ref))
1376 return true;
1378 FOR_EACH_VEC_ELT (*jfunc->agg.items, i, item)
1380 HOST_WIDE_INT val_size;
1382 if (item->offset < 0)
1383 continue;
1384 gcc_checking_assert (is_gimple_ip_invariant (item->value));
1385 val_size = tree_low_cst (TYPE_SIZE (TREE_TYPE (item->value)), 1);
1387 if (merge_agg_lats_step (dest_plats, item->offset, val_size,
1388 &aglat, pre_existing, &ret))
1390 ret |= add_value_to_lattice (*aglat, item->value, cs, NULL, 0, 0);
1391 aglat = &(*aglat)->next;
1393 else if (dest_plats->aggs_bottom)
1394 return true;
1397 ret |= set_chain_of_aglats_contains_variable (*aglat);
1399 else
1400 ret |= set_agg_lats_contain_variable (dest_plats);
1402 return ret;
1405 /* Propagate constants from the caller to the callee of CS. INFO describes the
1406 caller. */
1408 static bool
1409 propagate_constants_accross_call (struct cgraph_edge *cs)
1411 struct ipa_node_params *callee_info;
1412 enum availability availability;
1413 struct cgraph_node *callee, *alias_or_thunk;
1414 struct ipa_edge_args *args;
1415 bool ret = false;
1416 int i, args_count, parms_count;
1418 callee = cgraph_function_node (cs->callee, &availability);
1419 if (!callee->analyzed)
1420 return false;
1421 gcc_checking_assert (cgraph_function_with_gimple_body_p (callee));
1422 callee_info = IPA_NODE_REF (callee);
1424 args = IPA_EDGE_REF (cs);
1425 args_count = ipa_get_cs_argument_count (args);
1426 parms_count = ipa_get_param_count (callee_info);
1428 /* If this call goes through a thunk we must not propagate to the first (0th)
1429 parameter. However, we might need to uncover a thunk from below a series
1430 of aliases first. */
1431 alias_or_thunk = cs->callee;
1432 while (alias_or_thunk->alias)
1433 alias_or_thunk = cgraph_alias_aliased_node (alias_or_thunk);
1434 if (alias_or_thunk->thunk.thunk_p)
1436 ret |= set_all_contains_variable (ipa_get_parm_lattices (callee_info,
1437 0));
1438 i = 1;
1440 else
1441 i = 0;
1443 for (; (i < args_count) && (i < parms_count); i++)
1445 struct ipa_jump_func *jump_func = ipa_get_ith_jump_func (args, i);
1446 struct ipcp_param_lattices *dest_plats;
1448 dest_plats = ipa_get_parm_lattices (callee_info, i);
1449 if (availability == AVAIL_OVERWRITABLE)
1450 ret |= set_all_contains_variable (dest_plats);
1451 else
1453 ret |= propagate_scalar_accross_jump_function (cs, jump_func,
1454 &dest_plats->itself);
1455 ret |= propagate_aggs_accross_jump_function (cs, jump_func,
1456 dest_plats);
1459 for (; i < parms_count; i++)
1460 ret |= set_all_contains_variable (ipa_get_parm_lattices (callee_info, i));
1462 return ret;
1465 /* If an indirect edge IE can be turned into a direct one based on KNOWN_VALS
1466 (which can contain both constants and binfos), KNOWN_BINFOS, KNOWN_AGGS or
1467 AGG_REPS return the destination. The latter three can be NULL. If AGG_REPS
1468 is not NULL, KNOWN_AGGS is ignored. */
1470 static tree
1471 ipa_get_indirect_edge_target_1 (struct cgraph_edge *ie,
1472 vec<tree> known_vals,
1473 vec<tree> known_binfos,
1474 vec<ipa_agg_jump_function_p> known_aggs,
1475 struct ipa_agg_replacement_value *agg_reps)
1477 int param_index = ie->indirect_info->param_index;
1478 HOST_WIDE_INT token, anc_offset;
1479 tree otr_type;
1480 tree t;
1482 if (param_index == -1)
1483 return NULL_TREE;
1485 if (!ie->indirect_info->polymorphic)
1487 tree t;
1489 if (ie->indirect_info->agg_contents)
1491 if (agg_reps)
1493 t = NULL;
1494 while (agg_reps)
1496 if (agg_reps->index == param_index
1497 && agg_reps->offset == ie->indirect_info->offset)
1499 t = agg_reps->value;
1500 break;
1502 agg_reps = agg_reps->next;
1505 else if (known_aggs.length () > (unsigned int) param_index)
1507 struct ipa_agg_jump_function *agg;
1508 agg = known_aggs[param_index];
1509 t = ipa_find_agg_cst_for_param (agg, ie->indirect_info->offset,
1510 ie->indirect_info->by_ref);
1512 else
1513 t = NULL;
1515 else
1516 t = (known_vals.length () > (unsigned int) param_index
1517 ? known_vals[param_index] : NULL);
1519 if (t &&
1520 TREE_CODE (t) == ADDR_EXPR
1521 && TREE_CODE (TREE_OPERAND (t, 0)) == FUNCTION_DECL)
1522 return TREE_OPERAND (t, 0);
1523 else
1524 return NULL_TREE;
1527 gcc_assert (!ie->indirect_info->agg_contents);
1528 token = ie->indirect_info->otr_token;
1529 anc_offset = ie->indirect_info->offset;
1530 otr_type = ie->indirect_info->otr_type;
1532 t = known_vals[param_index];
1533 if (!t && known_binfos.length () > (unsigned int) param_index)
1534 t = known_binfos[param_index];
1535 if (!t)
1536 return NULL_TREE;
1538 if (TREE_CODE (t) != TREE_BINFO)
1540 tree binfo;
1541 binfo = gimple_extract_devirt_binfo_from_cst (t);
1542 if (!binfo)
1543 return NULL_TREE;
1544 binfo = get_binfo_at_offset (binfo, anc_offset, otr_type);
1545 if (!binfo)
1546 return NULL_TREE;
1547 return gimple_get_virt_method_for_binfo (token, binfo);
1549 else
1551 tree binfo;
1553 binfo = get_binfo_at_offset (t, anc_offset, otr_type);
1554 if (!binfo)
1555 return NULL_TREE;
1556 return gimple_get_virt_method_for_binfo (token, binfo);
1561 /* If an indirect edge IE can be turned into a direct one based on KNOWN_VALS
1562 (which can contain both constants and binfos), KNOWN_BINFOS (which can be
1563 NULL) or KNOWN_AGGS (which also can be NULL) return the destination. */
1565 tree
1566 ipa_get_indirect_edge_target (struct cgraph_edge *ie,
1567 vec<tree> known_vals,
1568 vec<tree> known_binfos,
1569 vec<ipa_agg_jump_function_p> known_aggs)
1571 return ipa_get_indirect_edge_target_1 (ie, known_vals, known_binfos,
1572 known_aggs, NULL);
1575 /* Calculate devirtualization time bonus for NODE, assuming we know KNOWN_CSTS
1576 and KNOWN_BINFOS. */
1578 static int
1579 devirtualization_time_bonus (struct cgraph_node *node,
1580 vec<tree> known_csts,
1581 vec<tree> known_binfos,
1582 vec<ipa_agg_jump_function_p> known_aggs)
1584 struct cgraph_edge *ie;
1585 int res = 0;
1587 for (ie = node->indirect_calls; ie; ie = ie->next_callee)
1589 struct cgraph_node *callee;
1590 struct inline_summary *isummary;
1591 tree target;
1593 target = ipa_get_indirect_edge_target (ie, known_csts, known_binfos,
1594 known_aggs);
1595 if (!target)
1596 continue;
1598 /* Only bare minimum benefit for clearly un-inlineable targets. */
1599 res += 1;
1600 callee = cgraph_get_node (target);
1601 if (!callee || !callee->analyzed)
1602 continue;
1603 isummary = inline_summary (callee);
1604 if (!isummary->inlinable)
1605 continue;
1607 /* FIXME: The values below need re-considering and perhaps also
1608 integrating into the cost metrics, at lest in some very basic way. */
1609 if (isummary->size <= MAX_INLINE_INSNS_AUTO / 4)
1610 res += 31;
1611 else if (isummary->size <= MAX_INLINE_INSNS_AUTO / 2)
1612 res += 15;
1613 else if (isummary->size <= MAX_INLINE_INSNS_AUTO
1614 || DECL_DECLARED_INLINE_P (callee->symbol.decl))
1615 res += 7;
1618 return res;
1621 /* Return time bonus incurred because of HINTS. */
1623 static int
1624 hint_time_bonus (inline_hints hints)
1626 int result = 0;
1627 if (hints & (INLINE_HINT_loop_iterations | INLINE_HINT_loop_stride))
1628 result += PARAM_VALUE (PARAM_IPA_CP_LOOP_HINT_BONUS);
1629 if (hints & INLINE_HINT_array_index)
1630 result += PARAM_VALUE (PARAM_IPA_CP_ARRAY_INDEX_HINT_BONUS);
1631 return result;
1634 /* Return true if cloning NODE is a good idea, given the estimated TIME_BENEFIT
1635 and SIZE_COST and with the sum of frequencies of incoming edges to the
1636 potential new clone in FREQUENCIES. */
1638 static bool
1639 good_cloning_opportunity_p (struct cgraph_node *node, int time_benefit,
1640 int freq_sum, gcov_type count_sum, int size_cost)
1642 if (time_benefit == 0
1643 || !flag_ipa_cp_clone
1644 || !optimize_function_for_speed_p (DECL_STRUCT_FUNCTION (node->symbol.decl)))
1645 return false;
1647 gcc_assert (size_cost > 0);
1649 if (max_count)
1651 int factor = (count_sum * 1000) / max_count;
1652 HOST_WIDEST_INT evaluation = (((HOST_WIDEST_INT) time_benefit * factor)
1653 / size_cost);
1655 if (dump_file && (dump_flags & TDF_DETAILS))
1656 fprintf (dump_file, " good_cloning_opportunity_p (time: %i, "
1657 "size: %i, count_sum: " HOST_WIDE_INT_PRINT_DEC
1658 ") -> evaluation: " HOST_WIDEST_INT_PRINT_DEC
1659 ", threshold: %i\n",
1660 time_benefit, size_cost, (HOST_WIDE_INT) count_sum,
1661 evaluation, PARAM_VALUE (PARAM_IPA_CP_EVAL_THRESHOLD));
1663 return evaluation >= PARAM_VALUE (PARAM_IPA_CP_EVAL_THRESHOLD);
1665 else
1667 HOST_WIDEST_INT evaluation = (((HOST_WIDEST_INT) time_benefit * freq_sum)
1668 / size_cost);
1670 if (dump_file && (dump_flags & TDF_DETAILS))
1671 fprintf (dump_file, " good_cloning_opportunity_p (time: %i, "
1672 "size: %i, freq_sum: %i) -> evaluation: "
1673 HOST_WIDEST_INT_PRINT_DEC ", threshold: %i\n",
1674 time_benefit, size_cost, freq_sum, evaluation,
1675 PARAM_VALUE (PARAM_IPA_CP_EVAL_THRESHOLD));
1677 return evaluation >= PARAM_VALUE (PARAM_IPA_CP_EVAL_THRESHOLD);
1681 /* Return all context independent values from aggregate lattices in PLATS in a
1682 vector. Return NULL if there are none. */
1684 static vec<ipa_agg_jf_item_t, va_gc> *
1685 context_independent_aggregate_values (struct ipcp_param_lattices *plats)
1687 vec<ipa_agg_jf_item_t, va_gc> *res = NULL;
1689 if (plats->aggs_bottom
1690 || plats->aggs_contain_variable
1691 || plats->aggs_count == 0)
1692 return NULL;
1694 for (struct ipcp_agg_lattice *aglat = plats->aggs;
1695 aglat;
1696 aglat = aglat->next)
1697 if (ipa_lat_is_single_const (aglat))
1699 struct ipa_agg_jf_item item;
1700 item.offset = aglat->offset;
1701 item.value = aglat->values->value;
1702 vec_safe_push (res, item);
1704 return res;
1707 /* Allocate KNOWN_CSTS, KNOWN_BINFOS and, if non-NULL, KNOWN_AGGS and populate
1708 them with values of parameters that are known independent of the context.
1709 INFO describes the function. If REMOVABLE_PARAMS_COST is non-NULL, the
1710 movement cost of all removable parameters will be stored in it. */
1712 static bool
1713 gather_context_independent_values (struct ipa_node_params *info,
1714 vec<tree> *known_csts,
1715 vec<tree> *known_binfos,
1716 vec<ipa_agg_jump_function_t> *known_aggs,
1717 int *removable_params_cost)
1719 int i, count = ipa_get_param_count (info);
1720 bool ret = false;
1722 known_csts->create (0);
1723 known_binfos->create (0);
1724 known_csts->safe_grow_cleared (count);
1725 known_binfos->safe_grow_cleared (count);
1726 if (known_aggs)
1728 known_aggs->create (0);
1729 known_aggs->safe_grow_cleared (count);
1732 if (removable_params_cost)
1733 *removable_params_cost = 0;
1735 for (i = 0; i < count ; i++)
1737 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
1738 struct ipcp_lattice *lat = &plats->itself;
1740 if (ipa_lat_is_single_const (lat))
1742 struct ipcp_value *val = lat->values;
1743 if (TREE_CODE (val->value) != TREE_BINFO)
1745 (*known_csts)[i] = val->value;
1746 if (removable_params_cost)
1747 *removable_params_cost
1748 += estimate_move_cost (TREE_TYPE (val->value));
1749 ret = true;
1751 else if (plats->virt_call)
1753 (*known_binfos)[i] = val->value;
1754 ret = true;
1756 else if (removable_params_cost
1757 && !ipa_is_param_used (info, i))
1758 *removable_params_cost
1759 += estimate_move_cost (TREE_TYPE (ipa_get_param (info, i)));
1761 else if (removable_params_cost
1762 && !ipa_is_param_used (info, i))
1763 *removable_params_cost
1764 += estimate_move_cost (TREE_TYPE (ipa_get_param (info, i)));
1766 if (known_aggs)
1768 vec<ipa_agg_jf_item_t, va_gc> *agg_items;
1769 struct ipa_agg_jump_function *ajf;
1771 agg_items = context_independent_aggregate_values (plats);
1772 ajf = &(*known_aggs)[i];
1773 ajf->items = agg_items;
1774 ajf->by_ref = plats->aggs_by_ref;
1775 ret |= agg_items != NULL;
1779 return ret;
1782 /* The current interface in ipa-inline-analysis requires a pointer vector.
1783 Create it.
1785 FIXME: That interface should be re-worked, this is slightly silly. Still,
1786 I'd like to discuss how to change it first and this demonstrates the
1787 issue. */
1789 static vec<ipa_agg_jump_function_p>
1790 agg_jmp_p_vec_for_t_vec (vec<ipa_agg_jump_function_t> known_aggs)
1792 vec<ipa_agg_jump_function_p> ret;
1793 struct ipa_agg_jump_function *ajf;
1794 int i;
1796 ret.create (known_aggs.length ());
1797 FOR_EACH_VEC_ELT (known_aggs, i, ajf)
1798 ret.quick_push (ajf);
1799 return ret;
1802 /* Iterate over known values of parameters of NODE and estimate the local
1803 effects in terms of time and size they have. */
1805 static void
1806 estimate_local_effects (struct cgraph_node *node)
1808 struct ipa_node_params *info = IPA_NODE_REF (node);
1809 int i, count = ipa_get_param_count (info);
1810 vec<tree> known_csts, known_binfos;
1811 vec<ipa_agg_jump_function_t> known_aggs;
1812 vec<ipa_agg_jump_function_p> known_aggs_ptrs;
1813 bool always_const;
1814 int base_time = inline_summary (node)->time;
1815 int removable_params_cost;
1817 if (!count || !ipcp_versionable_function_p (node))
1818 return;
1820 if (dump_file && (dump_flags & TDF_DETAILS))
1821 fprintf (dump_file, "\nEstimating effects for %s/%i, base_time: %i.\n",
1822 cgraph_node_name (node), node->uid, base_time);
1824 always_const = gather_context_independent_values (info, &known_csts,
1825 &known_binfos, &known_aggs,
1826 &removable_params_cost);
1827 known_aggs_ptrs = agg_jmp_p_vec_for_t_vec (known_aggs);
1828 if (always_const)
1830 struct caller_statistics stats;
1831 inline_hints hints;
1832 int time, size;
1834 init_caller_stats (&stats);
1835 cgraph_for_node_and_aliases (node, gather_caller_stats, &stats, false);
1836 estimate_ipcp_clone_size_and_time (node, known_csts, known_binfos,
1837 known_aggs_ptrs, &size, &time, &hints);
1838 time -= devirtualization_time_bonus (node, known_csts, known_binfos,
1839 known_aggs_ptrs);
1840 time -= hint_time_bonus (hints);
1841 time -= removable_params_cost;
1842 size -= stats.n_calls * removable_params_cost;
1844 if (dump_file)
1845 fprintf (dump_file, " - context independent values, size: %i, "
1846 "time_benefit: %i\n", size, base_time - time);
1848 if (size <= 0
1849 || cgraph_will_be_removed_from_program_if_no_direct_calls (node))
1851 info->do_clone_for_all_contexts = true;
1852 base_time = time;
1854 if (dump_file)
1855 fprintf (dump_file, " Decided to specialize for all "
1856 "known contexts, code not going to grow.\n");
1858 else if (good_cloning_opportunity_p (node, base_time - time,
1859 stats.freq_sum, stats.count_sum,
1860 size))
1862 if (size + overall_size <= max_new_size)
1864 info->do_clone_for_all_contexts = true;
1865 base_time = time;
1866 overall_size += size;
1868 if (dump_file)
1869 fprintf (dump_file, " Decided to specialize for all "
1870 "known contexts, growth deemed beneficial.\n");
1872 else if (dump_file && (dump_flags & TDF_DETAILS))
1873 fprintf (dump_file, " Not cloning for all contexts because "
1874 "max_new_size would be reached with %li.\n",
1875 size + overall_size);
1879 for (i = 0; i < count ; i++)
1881 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
1882 struct ipcp_lattice *lat = &plats->itself;
1883 struct ipcp_value *val;
1884 int emc;
1886 if (lat->bottom
1887 || !lat->values
1888 || known_csts[i]
1889 || known_binfos[i])
1890 continue;
1892 for (val = lat->values; val; val = val->next)
1894 int time, size, time_benefit;
1895 inline_hints hints;
1897 if (TREE_CODE (val->value) != TREE_BINFO)
1899 known_csts[i] = val->value;
1900 known_binfos[i] = NULL_TREE;
1901 emc = estimate_move_cost (TREE_TYPE (val->value));
1903 else if (plats->virt_call)
1905 known_csts[i] = NULL_TREE;
1906 known_binfos[i] = val->value;
1907 emc = 0;
1909 else
1910 continue;
1912 estimate_ipcp_clone_size_and_time (node, known_csts, known_binfos,
1913 known_aggs_ptrs, &size, &time,
1914 &hints);
1915 time_benefit = base_time - time
1916 + devirtualization_time_bonus (node, known_csts, known_binfos,
1917 known_aggs_ptrs)
1918 + hint_time_bonus (hints)
1919 + removable_params_cost + emc;
1921 gcc_checking_assert (size >=0);
1922 /* The inliner-heuristics based estimates may think that in certain
1923 contexts some functions do not have any size at all but we want
1924 all specializations to have at least a tiny cost, not least not to
1925 divide by zero. */
1926 if (size == 0)
1927 size = 1;
1929 if (dump_file && (dump_flags & TDF_DETAILS))
1931 fprintf (dump_file, " - estimates for value ");
1932 print_ipcp_constant_value (dump_file, val->value);
1933 fprintf (dump_file, " for parameter ");
1934 print_generic_expr (dump_file, ipa_get_param (info, i), 0);
1935 fprintf (dump_file, ": time_benefit: %i, size: %i\n",
1936 time_benefit, size);
1939 val->local_time_benefit = time_benefit;
1940 val->local_size_cost = size;
1942 known_binfos[i] = NULL_TREE;
1943 known_csts[i] = NULL_TREE;
1946 for (i = 0; i < count ; i++)
1948 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
1949 struct ipa_agg_jump_function *ajf;
1950 struct ipcp_agg_lattice *aglat;
1952 if (plats->aggs_bottom || !plats->aggs)
1953 continue;
1955 ajf = &known_aggs[i];
1956 for (aglat = plats->aggs; aglat; aglat = aglat->next)
1958 struct ipcp_value *val;
1959 if (aglat->bottom || !aglat->values
1960 /* If the following is true, the one value is in known_aggs. */
1961 || (!plats->aggs_contain_variable
1962 && ipa_lat_is_single_const (aglat)))
1963 continue;
1965 for (val = aglat->values; val; val = val->next)
1967 int time, size, time_benefit;
1968 struct ipa_agg_jf_item item;
1969 inline_hints hints;
1971 item.offset = aglat->offset;
1972 item.value = val->value;
1973 vec_safe_push (ajf->items, item);
1975 estimate_ipcp_clone_size_and_time (node, known_csts, known_binfos,
1976 known_aggs_ptrs, &size, &time,
1977 &hints);
1978 time_benefit = base_time - time
1979 + devirtualization_time_bonus (node, known_csts, known_binfos,
1980 known_aggs_ptrs)
1981 + hint_time_bonus (hints);
1982 gcc_checking_assert (size >=0);
1983 if (size == 0)
1984 size = 1;
1986 if (dump_file && (dump_flags & TDF_DETAILS))
1988 fprintf (dump_file, " - estimates for value ");
1989 print_ipcp_constant_value (dump_file, val->value);
1990 fprintf (dump_file, " for parameter ");
1991 print_generic_expr (dump_file, ipa_get_param (info, i), 0);
1992 fprintf (dump_file, "[%soffset: " HOST_WIDE_INT_PRINT_DEC
1993 "]: time_benefit: %i, size: %i\n",
1994 plats->aggs_by_ref ? "ref " : "",
1995 aglat->offset, time_benefit, size);
1998 val->local_time_benefit = time_benefit;
1999 val->local_size_cost = size;
2000 ajf->items->pop ();
2005 for (i = 0; i < count ; i++)
2006 vec_free (known_aggs[i].items);
2008 known_csts.release ();
2009 known_binfos.release ();
2010 known_aggs.release ();
2011 known_aggs_ptrs.release ();
2015 /* Add value CUR_VAL and all yet-unsorted values it is dependent on to the
2016 topological sort of values. */
2018 static void
2019 add_val_to_toposort (struct ipcp_value *cur_val)
2021 static int dfs_counter = 0;
2022 static struct ipcp_value *stack;
2023 struct ipcp_value_source *src;
2025 if (cur_val->dfs)
2026 return;
2028 dfs_counter++;
2029 cur_val->dfs = dfs_counter;
2030 cur_val->low_link = dfs_counter;
2032 cur_val->topo_next = stack;
2033 stack = cur_val;
2034 cur_val->on_stack = true;
2036 for (src = cur_val->sources; src; src = src->next)
2037 if (src->val)
2039 if (src->val->dfs == 0)
2041 add_val_to_toposort (src->val);
2042 if (src->val->low_link < cur_val->low_link)
2043 cur_val->low_link = src->val->low_link;
2045 else if (src->val->on_stack
2046 && src->val->dfs < cur_val->low_link)
2047 cur_val->low_link = src->val->dfs;
2050 if (cur_val->dfs == cur_val->low_link)
2052 struct ipcp_value *v, *scc_list = NULL;
2056 v = stack;
2057 stack = v->topo_next;
2058 v->on_stack = false;
2060 v->scc_next = scc_list;
2061 scc_list = v;
2063 while (v != cur_val);
2065 cur_val->topo_next = values_topo;
2066 values_topo = cur_val;
2070 /* Add all values in lattices associated with NODE to the topological sort if
2071 they are not there yet. */
2073 static void
2074 add_all_node_vals_to_toposort (struct cgraph_node *node)
2076 struct ipa_node_params *info = IPA_NODE_REF (node);
2077 int i, count = ipa_get_param_count (info);
2079 for (i = 0; i < count ; i++)
2081 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
2082 struct ipcp_lattice *lat = &plats->itself;
2083 struct ipcp_agg_lattice *aglat;
2084 struct ipcp_value *val;
2086 if (!lat->bottom)
2087 for (val = lat->values; val; val = val->next)
2088 add_val_to_toposort (val);
2090 if (!plats->aggs_bottom)
2091 for (aglat = plats->aggs; aglat; aglat = aglat->next)
2092 if (!aglat->bottom)
2093 for (val = aglat->values; val; val = val->next)
2094 add_val_to_toposort (val);
2098 /* One pass of constants propagation along the call graph edges, from callers
2099 to callees (requires topological ordering in TOPO), iterate over strongly
2100 connected components. */
2102 static void
2103 propagate_constants_topo (struct topo_info *topo)
2105 int i;
2107 for (i = topo->nnodes - 1; i >= 0; i--)
2109 struct cgraph_node *v, *node = topo->order[i];
2110 struct ipa_dfs_info *node_dfs_info;
2112 if (!cgraph_function_with_gimple_body_p (node))
2113 continue;
2115 node_dfs_info = (struct ipa_dfs_info *) node->symbol.aux;
2116 /* First, iteratively propagate within the strongly connected component
2117 until all lattices stabilize. */
2118 v = node_dfs_info->next_cycle;
2119 while (v)
2121 push_node_to_stack (topo, v);
2122 v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle;
2125 v = node;
2126 while (v)
2128 struct cgraph_edge *cs;
2130 for (cs = v->callees; cs; cs = cs->next_callee)
2131 if (edge_within_scc (cs)
2132 && propagate_constants_accross_call (cs))
2133 push_node_to_stack (topo, cs->callee);
2134 v = pop_node_from_stack (topo);
2137 /* Afterwards, propagate along edges leading out of the SCC, calculates
2138 the local effects of the discovered constants and all valid values to
2139 their topological sort. */
2140 v = node;
2141 while (v)
2143 struct cgraph_edge *cs;
2145 estimate_local_effects (v);
2146 add_all_node_vals_to_toposort (v);
2147 for (cs = v->callees; cs; cs = cs->next_callee)
2148 if (!edge_within_scc (cs))
2149 propagate_constants_accross_call (cs);
2151 v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle;
2157 /* Return the sum of A and B if none of them is bigger than INT_MAX/2, return
2158 the bigger one if otherwise. */
2160 static int
2161 safe_add (int a, int b)
2163 if (a > INT_MAX/2 || b > INT_MAX/2)
2164 return a > b ? a : b;
2165 else
2166 return a + b;
2170 /* Propagate the estimated effects of individual values along the topological
2171 from the dependent values to those they depend on. */
2173 static void
2174 propagate_effects (void)
2176 struct ipcp_value *base;
2178 for (base = values_topo; base; base = base->topo_next)
2180 struct ipcp_value_source *src;
2181 struct ipcp_value *val;
2182 int time = 0, size = 0;
2184 for (val = base; val; val = val->scc_next)
2186 time = safe_add (time,
2187 val->local_time_benefit + val->prop_time_benefit);
2188 size = safe_add (size, val->local_size_cost + val->prop_size_cost);
2191 for (val = base; val; val = val->scc_next)
2192 for (src = val->sources; src; src = src->next)
2193 if (src->val
2194 && cgraph_maybe_hot_edge_p (src->cs))
2196 src->val->prop_time_benefit = safe_add (time,
2197 src->val->prop_time_benefit);
2198 src->val->prop_size_cost = safe_add (size,
2199 src->val->prop_size_cost);
2205 /* Propagate constants, binfos and their effects from the summaries
2206 interprocedurally. */
2208 static void
2209 ipcp_propagate_stage (struct topo_info *topo)
2211 struct cgraph_node *node;
2213 if (dump_file)
2214 fprintf (dump_file, "\n Propagating constants:\n\n");
2216 if (in_lto_p)
2217 ipa_update_after_lto_read ();
2220 FOR_EACH_DEFINED_FUNCTION (node)
2222 struct ipa_node_params *info = IPA_NODE_REF (node);
2224 determine_versionability (node);
2225 if (cgraph_function_with_gimple_body_p (node))
2227 info->lattices = XCNEWVEC (struct ipcp_param_lattices,
2228 ipa_get_param_count (info));
2229 initialize_node_lattices (node);
2231 if (node->count > max_count)
2232 max_count = node->count;
2233 overall_size += inline_summary (node)->self_size;
2236 max_new_size = overall_size;
2237 if (max_new_size < PARAM_VALUE (PARAM_LARGE_UNIT_INSNS))
2238 max_new_size = PARAM_VALUE (PARAM_LARGE_UNIT_INSNS);
2239 max_new_size += max_new_size * PARAM_VALUE (PARAM_IPCP_UNIT_GROWTH) / 100 + 1;
2241 if (dump_file)
2242 fprintf (dump_file, "\noverall_size: %li, max_new_size: %li\n",
2243 overall_size, max_new_size);
2245 propagate_constants_topo (topo);
2246 #ifdef ENABLE_CHECKING
2247 ipcp_verify_propagated_values ();
2248 #endif
2249 propagate_effects ();
2251 if (dump_file)
2253 fprintf (dump_file, "\nIPA lattices after all propagation:\n");
2254 print_all_lattices (dump_file, (dump_flags & TDF_DETAILS), true);
2258 /* Discover newly direct outgoing edges from NODE which is a new clone with
2259 known KNOWN_VALS and make them direct. */
2261 static void
2262 ipcp_discover_new_direct_edges (struct cgraph_node *node,
2263 vec<tree> known_vals,
2264 struct ipa_agg_replacement_value *aggvals)
2266 struct cgraph_edge *ie, *next_ie;
2267 bool found = false;
2269 for (ie = node->indirect_calls; ie; ie = next_ie)
2271 tree target;
2273 next_ie = ie->next_callee;
2274 target = ipa_get_indirect_edge_target_1 (ie, known_vals, vNULL, vNULL,
2275 aggvals);
2276 if (target)
2278 ipa_make_edge_direct_to_target (ie, target);
2279 found = true;
2282 /* Turning calls to direct calls will improve overall summary. */
2283 if (found)
2284 inline_update_overall_summary (node);
2287 /* Vector of pointers which for linked lists of clones of an original crgaph
2288 edge. */
2290 static vec<cgraph_edge_p> next_edge_clone;
2292 static inline void
2293 grow_next_edge_clone_vector (void)
2295 if (next_edge_clone.length ()
2296 <= (unsigned) cgraph_edge_max_uid)
2297 next_edge_clone.safe_grow_cleared (cgraph_edge_max_uid + 1);
2300 /* Edge duplication hook to grow the appropriate linked list in
2301 next_edge_clone. */
2303 static void
2304 ipcp_edge_duplication_hook (struct cgraph_edge *src, struct cgraph_edge *dst,
2305 __attribute__((unused)) void *data)
2307 grow_next_edge_clone_vector ();
2308 next_edge_clone[dst->uid] = next_edge_clone[src->uid];
2309 next_edge_clone[src->uid] = dst;
2312 /* See if NODE is a clone with a known aggregate value at a given OFFSET of a
2313 parameter with the given INDEX. */
2315 static tree
2316 get_clone_agg_value (struct cgraph_node *node, HOST_WIDEST_INT offset,
2317 int index)
2319 struct ipa_agg_replacement_value *aggval;
2321 aggval = ipa_get_agg_replacements_for_node (node);
2322 while (aggval)
2324 if (aggval->offset == offset
2325 && aggval->index == index)
2326 return aggval->value;
2327 aggval = aggval->next;
2329 return NULL_TREE;
2332 /* Return true if edge CS does bring about the value described by SRC. */
2334 static bool
2335 cgraph_edge_brings_value_p (struct cgraph_edge *cs,
2336 struct ipcp_value_source *src)
2338 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
2339 struct ipa_node_params *dst_info = IPA_NODE_REF (cs->callee);
2341 if ((dst_info->ipcp_orig_node && !dst_info->is_all_contexts_clone)
2342 || caller_info->node_dead)
2343 return false;
2344 if (!src->val)
2345 return true;
2347 if (caller_info->ipcp_orig_node)
2349 tree t;
2350 if (src->offset == -1)
2351 t = caller_info->known_vals[src->index];
2352 else
2353 t = get_clone_agg_value (cs->caller, src->offset, src->index);
2354 return (t != NULL_TREE
2355 && values_equal_for_ipcp_p (src->val->value, t));
2357 else
2359 struct ipcp_agg_lattice *aglat;
2360 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (caller_info,
2361 src->index);
2362 if (src->offset == -1)
2363 return (ipa_lat_is_single_const (&plats->itself)
2364 && values_equal_for_ipcp_p (src->val->value,
2365 plats->itself.values->value));
2366 else
2368 if (plats->aggs_bottom || plats->aggs_contain_variable)
2369 return false;
2370 for (aglat = plats->aggs; aglat; aglat = aglat->next)
2371 if (aglat->offset == src->offset)
2372 return (ipa_lat_is_single_const (aglat)
2373 && values_equal_for_ipcp_p (src->val->value,
2374 aglat->values->value));
2376 return false;
2380 /* Get the next clone in the linked list of clones of an edge. */
2382 static inline struct cgraph_edge *
2383 get_next_cgraph_edge_clone (struct cgraph_edge *cs)
2385 return next_edge_clone[cs->uid];
2388 /* Given VAL, iterate over all its sources and if they still hold, add their
2389 edge frequency and their number into *FREQUENCY and *CALLER_COUNT
2390 respectively. */
2392 static bool
2393 get_info_about_necessary_edges (struct ipcp_value *val, int *freq_sum,
2394 gcov_type *count_sum, int *caller_count)
2396 struct ipcp_value_source *src;
2397 int freq = 0, count = 0;
2398 gcov_type cnt = 0;
2399 bool hot = false;
2401 for (src = val->sources; src; src = src->next)
2403 struct cgraph_edge *cs = src->cs;
2404 while (cs)
2406 if (cgraph_edge_brings_value_p (cs, src))
2408 count++;
2409 freq += cs->frequency;
2410 cnt += cs->count;
2411 hot |= cgraph_maybe_hot_edge_p (cs);
2413 cs = get_next_cgraph_edge_clone (cs);
2417 *freq_sum = freq;
2418 *count_sum = cnt;
2419 *caller_count = count;
2420 return hot;
2423 /* Return a vector of incoming edges that do bring value VAL. It is assumed
2424 their number is known and equal to CALLER_COUNT. */
2426 static vec<cgraph_edge_p>
2427 gather_edges_for_value (struct ipcp_value *val, int caller_count)
2429 struct ipcp_value_source *src;
2430 vec<cgraph_edge_p> ret;
2432 ret.create (caller_count);
2433 for (src = val->sources; src; src = src->next)
2435 struct cgraph_edge *cs = src->cs;
2436 while (cs)
2438 if (cgraph_edge_brings_value_p (cs, src))
2439 ret.quick_push (cs);
2440 cs = get_next_cgraph_edge_clone (cs);
2444 return ret;
2447 /* Construct a replacement map for a know VALUE for a formal parameter PARAM.
2448 Return it or NULL if for some reason it cannot be created. */
2450 static struct ipa_replace_map *
2451 get_replacement_map (tree value, tree parm)
2453 tree req_type = TREE_TYPE (parm);
2454 struct ipa_replace_map *replace_map;
2456 if (!useless_type_conversion_p (req_type, TREE_TYPE (value)))
2458 if (fold_convertible_p (req_type, value))
2459 value = fold_build1 (NOP_EXPR, req_type, value);
2460 else if (TYPE_SIZE (req_type) == TYPE_SIZE (TREE_TYPE (value)))
2461 value = fold_build1 (VIEW_CONVERT_EXPR, req_type, value);
2462 else
2464 if (dump_file)
2466 fprintf (dump_file, " const ");
2467 print_generic_expr (dump_file, value, 0);
2468 fprintf (dump_file, " can't be converted to param ");
2469 print_generic_expr (dump_file, parm, 0);
2470 fprintf (dump_file, "\n");
2472 return NULL;
2476 replace_map = ggc_alloc_ipa_replace_map ();
2477 if (dump_file)
2479 fprintf (dump_file, " replacing param ");
2480 print_generic_expr (dump_file, parm, 0);
2481 fprintf (dump_file, " with const ");
2482 print_generic_expr (dump_file, value, 0);
2483 fprintf (dump_file, "\n");
2485 replace_map->old_tree = parm;
2486 replace_map->new_tree = value;
2487 replace_map->replace_p = true;
2488 replace_map->ref_p = false;
2490 return replace_map;
2493 /* Dump new profiling counts */
2495 static void
2496 dump_profile_updates (struct cgraph_node *orig_node,
2497 struct cgraph_node *new_node)
2499 struct cgraph_edge *cs;
2501 fprintf (dump_file, " setting count of the specialized node to "
2502 HOST_WIDE_INT_PRINT_DEC "\n", (HOST_WIDE_INT) new_node->count);
2503 for (cs = new_node->callees; cs ; cs = cs->next_callee)
2504 fprintf (dump_file, " edge to %s has count "
2505 HOST_WIDE_INT_PRINT_DEC "\n",
2506 cgraph_node_name (cs->callee), (HOST_WIDE_INT) cs->count);
2508 fprintf (dump_file, " setting count of the original node to "
2509 HOST_WIDE_INT_PRINT_DEC "\n", (HOST_WIDE_INT) orig_node->count);
2510 for (cs = orig_node->callees; cs ; cs = cs->next_callee)
2511 fprintf (dump_file, " edge to %s is left with "
2512 HOST_WIDE_INT_PRINT_DEC "\n",
2513 cgraph_node_name (cs->callee), (HOST_WIDE_INT) cs->count);
2516 /* After a specialized NEW_NODE version of ORIG_NODE has been created, update
2517 their profile information to reflect this. */
2519 static void
2520 update_profiling_info (struct cgraph_node *orig_node,
2521 struct cgraph_node *new_node)
2523 struct cgraph_edge *cs;
2524 struct caller_statistics stats;
2525 gcov_type new_sum, orig_sum;
2526 gcov_type remainder, orig_node_count = orig_node->count;
2528 if (orig_node_count == 0)
2529 return;
2531 init_caller_stats (&stats);
2532 cgraph_for_node_and_aliases (orig_node, gather_caller_stats, &stats, false);
2533 orig_sum = stats.count_sum;
2534 init_caller_stats (&stats);
2535 cgraph_for_node_and_aliases (new_node, gather_caller_stats, &stats, false);
2536 new_sum = stats.count_sum;
2538 if (orig_node_count < orig_sum + new_sum)
2540 if (dump_file)
2541 fprintf (dump_file, " Problem: node %s/%i has too low count "
2542 HOST_WIDE_INT_PRINT_DEC " while the sum of incoming "
2543 "counts is " HOST_WIDE_INT_PRINT_DEC "\n",
2544 cgraph_node_name (orig_node), orig_node->uid,
2545 (HOST_WIDE_INT) orig_node_count,
2546 (HOST_WIDE_INT) (orig_sum + new_sum));
2548 orig_node_count = (orig_sum + new_sum) * 12 / 10;
2549 if (dump_file)
2550 fprintf (dump_file, " proceeding by pretending it was "
2551 HOST_WIDE_INT_PRINT_DEC "\n",
2552 (HOST_WIDE_INT) orig_node_count);
2555 new_node->count = new_sum;
2556 remainder = orig_node_count - new_sum;
2557 orig_node->count = remainder;
2559 for (cs = new_node->callees; cs ; cs = cs->next_callee)
2560 if (cs->frequency)
2561 cs->count = apply_probability (cs->count,
2562 GCOV_COMPUTE_SCALE (new_sum,
2563 orig_node_count));
2564 else
2565 cs->count = 0;
2567 for (cs = orig_node->callees; cs ; cs = cs->next_callee)
2568 cs->count = apply_probability (cs->count,
2569 GCOV_COMPUTE_SCALE (remainder,
2570 orig_node_count));
2572 if (dump_file)
2573 dump_profile_updates (orig_node, new_node);
2576 /* Update the respective profile of specialized NEW_NODE and the original
2577 ORIG_NODE after additional edges with cumulative count sum REDIRECTED_SUM
2578 have been redirected to the specialized version. */
2580 static void
2581 update_specialized_profile (struct cgraph_node *new_node,
2582 struct cgraph_node *orig_node,
2583 gcov_type redirected_sum)
2585 struct cgraph_edge *cs;
2586 gcov_type new_node_count, orig_node_count = orig_node->count;
2588 if (dump_file)
2589 fprintf (dump_file, " the sum of counts of redirected edges is "
2590 HOST_WIDE_INT_PRINT_DEC "\n", (HOST_WIDE_INT) redirected_sum);
2591 if (orig_node_count == 0)
2592 return;
2594 gcc_assert (orig_node_count >= redirected_sum);
2596 new_node_count = new_node->count;
2597 new_node->count += redirected_sum;
2598 orig_node->count -= redirected_sum;
2600 for (cs = new_node->callees; cs ; cs = cs->next_callee)
2601 if (cs->frequency)
2602 cs->count += apply_probability (cs->count,
2603 GCOV_COMPUTE_SCALE (redirected_sum,
2604 new_node_count));
2605 else
2606 cs->count = 0;
2608 for (cs = orig_node->callees; cs ; cs = cs->next_callee)
2610 gcov_type dec = apply_probability (cs->count,
2611 GCOV_COMPUTE_SCALE (redirected_sum,
2612 orig_node_count));
2613 if (dec < cs->count)
2614 cs->count -= dec;
2615 else
2616 cs->count = 0;
2619 if (dump_file)
2620 dump_profile_updates (orig_node, new_node);
2623 /* Create a specialized version of NODE with known constants and types of
2624 parameters in KNOWN_VALS and redirect all edges in CALLERS to it. */
2626 static struct cgraph_node *
2627 create_specialized_node (struct cgraph_node *node,
2628 vec<tree> known_vals,
2629 struct ipa_agg_replacement_value *aggvals,
2630 vec<cgraph_edge_p> callers)
2632 struct ipa_node_params *new_info, *info = IPA_NODE_REF (node);
2633 vec<ipa_replace_map_p, va_gc> *replace_trees = NULL;
2634 struct cgraph_node *new_node;
2635 int i, count = ipa_get_param_count (info);
2636 bitmap args_to_skip;
2638 gcc_assert (!info->ipcp_orig_node);
2640 if (node->local.can_change_signature)
2642 args_to_skip = BITMAP_GGC_ALLOC ();
2643 for (i = 0; i < count; i++)
2645 tree t = known_vals[i];
2647 if ((t && TREE_CODE (t) != TREE_BINFO)
2648 || !ipa_is_param_used (info, i))
2649 bitmap_set_bit (args_to_skip, i);
2652 else
2654 args_to_skip = NULL;
2655 if (dump_file && (dump_flags & TDF_DETAILS))
2656 fprintf (dump_file, " cannot change function signature\n");
2659 for (i = 0; i < count ; i++)
2661 tree t = known_vals[i];
2662 if (t && TREE_CODE (t) != TREE_BINFO)
2664 struct ipa_replace_map *replace_map;
2666 replace_map = get_replacement_map (t, ipa_get_param (info, i));
2667 if (replace_map)
2668 vec_safe_push (replace_trees, replace_map);
2672 new_node = cgraph_create_virtual_clone (node, callers, replace_trees,
2673 args_to_skip, "constprop");
2674 ipa_set_node_agg_value_chain (new_node, aggvals);
2675 if (dump_file && (dump_flags & TDF_DETAILS))
2677 fprintf (dump_file, " the new node is %s/%i.\n",
2678 cgraph_node_name (new_node), new_node->uid);
2679 if (aggvals)
2680 ipa_dump_agg_replacement_values (dump_file, aggvals);
2682 gcc_checking_assert (ipa_node_params_vector.exists ()
2683 && (ipa_node_params_vector.length ()
2684 > (unsigned) cgraph_max_uid));
2685 update_profiling_info (node, new_node);
2686 new_info = IPA_NODE_REF (new_node);
2687 new_info->ipcp_orig_node = node;
2688 new_info->known_vals = known_vals;
2690 ipcp_discover_new_direct_edges (new_node, known_vals, aggvals);
2692 callers.release ();
2693 return new_node;
2696 /* Given a NODE, and a subset of its CALLERS, try to populate blanks slots in
2697 KNOWN_VALS with constants and types that are also known for all of the
2698 CALLERS. */
2700 static void
2701 find_more_scalar_values_for_callers_subset (struct cgraph_node *node,
2702 vec<tree> known_vals,
2703 vec<cgraph_edge_p> callers)
2705 struct ipa_node_params *info = IPA_NODE_REF (node);
2706 int i, count = ipa_get_param_count (info);
2708 for (i = 0; i < count ; i++)
2710 struct cgraph_edge *cs;
2711 tree newval = NULL_TREE;
2712 int j;
2714 if (ipa_get_scalar_lat (info, i)->bottom || known_vals[i])
2715 continue;
2717 FOR_EACH_VEC_ELT (callers, j, cs)
2719 struct ipa_jump_func *jump_func;
2720 tree t;
2722 if (i >= ipa_get_cs_argument_count (IPA_EDGE_REF (cs)))
2724 newval = NULL_TREE;
2725 break;
2727 jump_func = ipa_get_ith_jump_func (IPA_EDGE_REF (cs), i);
2728 t = ipa_value_from_jfunc (IPA_NODE_REF (cs->caller), jump_func);
2729 if (!t
2730 || (newval
2731 && !values_equal_for_ipcp_p (t, newval)))
2733 newval = NULL_TREE;
2734 break;
2736 else
2737 newval = t;
2740 if (newval)
2742 if (dump_file && (dump_flags & TDF_DETAILS))
2744 fprintf (dump_file, " adding an extra known scalar value ");
2745 print_ipcp_constant_value (dump_file, newval);
2746 fprintf (dump_file, " for parameter ");
2747 print_generic_expr (dump_file, ipa_get_param (info, i), 0);
2748 fprintf (dump_file, "\n");
2751 known_vals[i] = newval;
2756 /* Go through PLATS and create a vector of values consisting of values and
2757 offsets (minus OFFSET) of lattices that contain only a single value. */
2759 static vec<ipa_agg_jf_item_t>
2760 copy_plats_to_inter (struct ipcp_param_lattices *plats, HOST_WIDE_INT offset)
2762 vec<ipa_agg_jf_item_t> res = vNULL;
2764 if (!plats->aggs || plats->aggs_contain_variable || plats->aggs_bottom)
2765 return vNULL;
2767 for (struct ipcp_agg_lattice *aglat = plats->aggs; aglat; aglat = aglat->next)
2768 if (ipa_lat_is_single_const (aglat))
2770 struct ipa_agg_jf_item ti;
2771 ti.offset = aglat->offset - offset;
2772 ti.value = aglat->values->value;
2773 res.safe_push (ti);
2775 return res;
2778 /* Intersect all values in INTER with single value lattices in PLATS (while
2779 subtracting OFFSET). */
2781 static void
2782 intersect_with_plats (struct ipcp_param_lattices *plats,
2783 vec<ipa_agg_jf_item_t> *inter,
2784 HOST_WIDE_INT offset)
2786 struct ipcp_agg_lattice *aglat;
2787 struct ipa_agg_jf_item *item;
2788 int k;
2790 if (!plats->aggs || plats->aggs_contain_variable || plats->aggs_bottom)
2792 inter->release ();
2793 return;
2796 aglat = plats->aggs;
2797 FOR_EACH_VEC_ELT (*inter, k, item)
2799 bool found = false;
2800 if (!item->value)
2801 continue;
2802 while (aglat)
2804 if (aglat->offset - offset > item->offset)
2805 break;
2806 if (aglat->offset - offset == item->offset)
2808 gcc_checking_assert (item->value);
2809 if (values_equal_for_ipcp_p (item->value, aglat->values->value))
2810 found = true;
2811 break;
2813 aglat = aglat->next;
2815 if (!found)
2816 item->value = NULL_TREE;
2820 /* Copy agggregate replacement values of NODE (which is an IPA-CP clone) to the
2821 vector result while subtracting OFFSET from the individual value offsets. */
2823 static vec<ipa_agg_jf_item_t>
2824 agg_replacements_to_vector (struct cgraph_node *node, int index,
2825 HOST_WIDE_INT offset)
2827 struct ipa_agg_replacement_value *av;
2828 vec<ipa_agg_jf_item_t> res = vNULL;
2830 for (av = ipa_get_agg_replacements_for_node (node); av; av = av->next)
2831 if (av->index == index
2832 && (av->offset - offset) >= 0)
2834 struct ipa_agg_jf_item item;
2835 gcc_checking_assert (av->value);
2836 item.offset = av->offset - offset;
2837 item.value = av->value;
2838 res.safe_push (item);
2841 return res;
2844 /* Intersect all values in INTER with those that we have already scheduled to
2845 be replaced in parameter number INDEX of NODE, which is an IPA-CP clone
2846 (while subtracting OFFSET). */
2848 static void
2849 intersect_with_agg_replacements (struct cgraph_node *node, int index,
2850 vec<ipa_agg_jf_item_t> *inter,
2851 HOST_WIDE_INT offset)
2853 struct ipa_agg_replacement_value *srcvals;
2854 struct ipa_agg_jf_item *item;
2855 int i;
2857 srcvals = ipa_get_agg_replacements_for_node (node);
2858 if (!srcvals)
2860 inter->release ();
2861 return;
2864 FOR_EACH_VEC_ELT (*inter, i, item)
2866 struct ipa_agg_replacement_value *av;
2867 bool found = false;
2868 if (!item->value)
2869 continue;
2870 for (av = srcvals; av; av = av->next)
2872 gcc_checking_assert (av->value);
2873 if (av->index == index
2874 && av->offset - offset == item->offset)
2876 if (values_equal_for_ipcp_p (item->value, av->value))
2877 found = true;
2878 break;
2881 if (!found)
2882 item->value = NULL_TREE;
2886 /* Intersect values in INTER with aggregate values that come along edge CS to
2887 parameter number INDEX and return it. If INTER does not actually exist yet,
2888 copy all incoming values to it. If we determine we ended up with no values
2889 whatsoever, return a released vector. */
2891 static vec<ipa_agg_jf_item_t>
2892 intersect_aggregates_with_edge (struct cgraph_edge *cs, int index,
2893 vec<ipa_agg_jf_item_t> inter)
2895 struct ipa_jump_func *jfunc;
2896 jfunc = ipa_get_ith_jump_func (IPA_EDGE_REF (cs), index);
2897 if (jfunc->type == IPA_JF_PASS_THROUGH
2898 && ipa_get_jf_pass_through_operation (jfunc) == NOP_EXPR)
2900 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
2901 int src_idx = ipa_get_jf_pass_through_formal_id (jfunc);
2903 if (caller_info->ipcp_orig_node)
2905 struct cgraph_node *orig_node = caller_info->ipcp_orig_node;
2906 struct ipcp_param_lattices *orig_plats;
2907 orig_plats = ipa_get_parm_lattices (IPA_NODE_REF (orig_node),
2908 src_idx);
2909 if (agg_pass_through_permissible_p (orig_plats, jfunc))
2911 if (!inter.exists ())
2912 inter = agg_replacements_to_vector (cs->caller, src_idx, 0);
2913 else
2914 intersect_with_agg_replacements (cs->caller, src_idx,
2915 &inter, 0);
2918 else
2920 struct ipcp_param_lattices *src_plats;
2921 src_plats = ipa_get_parm_lattices (caller_info, src_idx);
2922 if (agg_pass_through_permissible_p (src_plats, jfunc))
2924 /* Currently we do not produce clobber aggregate jump
2925 functions, adjust when we do. */
2926 gcc_checking_assert (!jfunc->agg.items);
2927 if (!inter.exists ())
2928 inter = copy_plats_to_inter (src_plats, 0);
2929 else
2930 intersect_with_plats (src_plats, &inter, 0);
2934 else if (jfunc->type == IPA_JF_ANCESTOR
2935 && ipa_get_jf_ancestor_agg_preserved (jfunc))
2937 struct ipa_node_params *caller_info = IPA_NODE_REF (cs->caller);
2938 int src_idx = ipa_get_jf_ancestor_formal_id (jfunc);
2939 struct ipcp_param_lattices *src_plats;
2940 HOST_WIDE_INT delta = ipa_get_jf_ancestor_offset (jfunc);
2942 if (caller_info->ipcp_orig_node)
2944 if (!inter.exists ())
2945 inter = agg_replacements_to_vector (cs->caller, src_idx, delta);
2946 else
2947 intersect_with_agg_replacements (cs->caller, src_idx, &inter,
2948 delta);
2950 else
2952 src_plats = ipa_get_parm_lattices (caller_info, src_idx);;
2953 /* Currently we do not produce clobber aggregate jump
2954 functions, adjust when we do. */
2955 gcc_checking_assert (!src_plats->aggs || !jfunc->agg.items);
2956 if (!inter.exists ())
2957 inter = copy_plats_to_inter (src_plats, delta);
2958 else
2959 intersect_with_plats (src_plats, &inter, delta);
2962 else if (jfunc->agg.items)
2964 struct ipa_agg_jf_item *item;
2965 int k;
2967 if (!inter.exists ())
2968 for (unsigned i = 0; i < jfunc->agg.items->length (); i++)
2969 inter.safe_push ((*jfunc->agg.items)[i]);
2970 else
2971 FOR_EACH_VEC_ELT (inter, k, item)
2973 int l = 0;
2974 bool found = false;;
2976 if (!item->value)
2977 continue;
2979 while ((unsigned) l < jfunc->agg.items->length ())
2981 struct ipa_agg_jf_item *ti;
2982 ti = &(*jfunc->agg.items)[l];
2983 if (ti->offset > item->offset)
2984 break;
2985 if (ti->offset == item->offset)
2987 gcc_checking_assert (ti->value);
2988 if (values_equal_for_ipcp_p (item->value,
2989 ti->value))
2990 found = true;
2991 break;
2993 l++;
2995 if (!found)
2996 item->value = NULL;
2999 else
3001 inter.release();
3002 return vec<ipa_agg_jf_item_t>();
3004 return inter;
3007 /* Look at edges in CALLERS and collect all known aggregate values that arrive
3008 from all of them. */
3010 static struct ipa_agg_replacement_value *
3011 find_aggregate_values_for_callers_subset (struct cgraph_node *node,
3012 vec<cgraph_edge_p> callers)
3014 struct ipa_node_params *dest_info = IPA_NODE_REF (node);
3015 struct ipa_agg_replacement_value *res = NULL;
3016 struct cgraph_edge *cs;
3017 int i, j, count = ipa_get_param_count (dest_info);
3019 FOR_EACH_VEC_ELT (callers, j, cs)
3021 int c = ipa_get_cs_argument_count (IPA_EDGE_REF (cs));
3022 if (c < count)
3023 count = c;
3026 for (i = 0; i < count ; i++)
3028 struct cgraph_edge *cs;
3029 vec<ipa_agg_jf_item_t> inter = vNULL;
3030 struct ipa_agg_jf_item *item;
3031 int j;
3033 /* Among other things, the following check should deal with all by_ref
3034 mismatches. */
3035 if (ipa_get_parm_lattices (dest_info, i)->aggs_bottom)
3036 continue;
3038 FOR_EACH_VEC_ELT (callers, j, cs)
3040 inter = intersect_aggregates_with_edge (cs, i, inter);
3042 if (!inter.exists ())
3043 goto next_param;
3046 FOR_EACH_VEC_ELT (inter, j, item)
3048 struct ipa_agg_replacement_value *v;
3050 if (!item->value)
3051 continue;
3053 v = ggc_alloc_ipa_agg_replacement_value ();
3054 v->index = i;
3055 v->offset = item->offset;
3056 v->value = item->value;
3057 v->next = res;
3058 res = v;
3061 next_param:
3062 if (inter.exists ())
3063 inter.release ();
3065 return res;
3068 /* Turn KNOWN_AGGS into a list of aggreate replacement values. */
3070 static struct ipa_agg_replacement_value *
3071 known_aggs_to_agg_replacement_list (vec<ipa_agg_jump_function_t> known_aggs)
3073 struct ipa_agg_replacement_value *res = NULL;
3074 struct ipa_agg_jump_function *aggjf;
3075 struct ipa_agg_jf_item *item;
3076 int i, j;
3078 FOR_EACH_VEC_ELT (known_aggs, i, aggjf)
3079 FOR_EACH_VEC_SAFE_ELT (aggjf->items, j, item)
3081 struct ipa_agg_replacement_value *v;
3082 v = ggc_alloc_ipa_agg_replacement_value ();
3083 v->index = i;
3084 v->offset = item->offset;
3085 v->value = item->value;
3086 v->next = res;
3087 res = v;
3089 return res;
3092 /* Determine whether CS also brings all scalar values that the NODE is
3093 specialized for. */
3095 static bool
3096 cgraph_edge_brings_all_scalars_for_node (struct cgraph_edge *cs,
3097 struct cgraph_node *node)
3099 struct ipa_node_params *dest_info = IPA_NODE_REF (node);
3100 int count = ipa_get_param_count (dest_info);
3101 struct ipa_node_params *caller_info;
3102 struct ipa_edge_args *args;
3103 int i;
3105 caller_info = IPA_NODE_REF (cs->caller);
3106 args = IPA_EDGE_REF (cs);
3107 for (i = 0; i < count; i++)
3109 struct ipa_jump_func *jump_func;
3110 tree val, t;
3112 val = dest_info->known_vals[i];
3113 if (!val)
3114 continue;
3116 if (i >= ipa_get_cs_argument_count (args))
3117 return false;
3118 jump_func = ipa_get_ith_jump_func (args, i);
3119 t = ipa_value_from_jfunc (caller_info, jump_func);
3120 if (!t || !values_equal_for_ipcp_p (val, t))
3121 return false;
3123 return true;
3126 /* Determine whether CS also brings all aggregate values that NODE is
3127 specialized for. */
3128 static bool
3129 cgraph_edge_brings_all_agg_vals_for_node (struct cgraph_edge *cs,
3130 struct cgraph_node *node)
3132 struct ipa_node_params *orig_caller_info = IPA_NODE_REF (cs->caller);
3133 struct ipa_agg_replacement_value *aggval;
3134 int i, ec, count;
3136 aggval = ipa_get_agg_replacements_for_node (node);
3137 if (!aggval)
3138 return true;
3140 count = ipa_get_param_count (IPA_NODE_REF (node));
3141 ec = ipa_get_cs_argument_count (IPA_EDGE_REF (cs));
3142 if (ec < count)
3143 for (struct ipa_agg_replacement_value *av = aggval; av; av = av->next)
3144 if (aggval->index >= ec)
3145 return false;
3147 if (orig_caller_info->ipcp_orig_node)
3148 orig_caller_info = IPA_NODE_REF (orig_caller_info->ipcp_orig_node);
3150 for (i = 0; i < count; i++)
3152 static vec<ipa_agg_jf_item_t> values = vec<ipa_agg_jf_item_t>();
3153 struct ipcp_param_lattices *plats;
3154 bool interesting = false;
3155 for (struct ipa_agg_replacement_value *av = aggval; av; av = av->next)
3156 if (aggval->index == i)
3158 interesting = true;
3159 break;
3161 if (!interesting)
3162 continue;
3164 plats = ipa_get_parm_lattices (orig_caller_info, aggval->index);
3165 if (plats->aggs_bottom)
3166 return false;
3168 values = intersect_aggregates_with_edge (cs, i, values);
3169 if (!values.exists())
3170 return false;
3172 for (struct ipa_agg_replacement_value *av = aggval; av; av = av->next)
3173 if (aggval->index == i)
3175 struct ipa_agg_jf_item *item;
3176 int j;
3177 bool found = false;
3178 FOR_EACH_VEC_ELT (values, j, item)
3179 if (item->value
3180 && item->offset == av->offset
3181 && values_equal_for_ipcp_p (item->value, av->value))
3182 found = true;
3183 if (!found)
3185 values.release();
3186 return false;
3190 return true;
3193 /* Given an original NODE and a VAL for which we have already created a
3194 specialized clone, look whether there are incoming edges that still lead
3195 into the old node but now also bring the requested value and also conform to
3196 all other criteria such that they can be redirected the the special node.
3197 This function can therefore redirect the final edge in a SCC. */
3199 static void
3200 perhaps_add_new_callers (struct cgraph_node *node, struct ipcp_value *val)
3202 struct ipcp_value_source *src;
3203 gcov_type redirected_sum = 0;
3205 for (src = val->sources; src; src = src->next)
3207 struct cgraph_edge *cs = src->cs;
3208 while (cs)
3210 enum availability availability;
3211 struct cgraph_node *dst = cgraph_function_node (cs->callee,
3212 &availability);
3213 if ((dst == node || IPA_NODE_REF (dst)->is_all_contexts_clone)
3214 && availability > AVAIL_OVERWRITABLE
3215 && cgraph_edge_brings_value_p (cs, src))
3217 if (cgraph_edge_brings_all_scalars_for_node (cs, val->spec_node)
3218 && cgraph_edge_brings_all_agg_vals_for_node (cs,
3219 val->spec_node))
3221 if (dump_file)
3222 fprintf (dump_file, " - adding an extra caller %s/%i"
3223 " of %s/%i\n",
3224 xstrdup (cgraph_node_name (cs->caller)),
3225 cs->caller->uid,
3226 xstrdup (cgraph_node_name (val->spec_node)),
3227 val->spec_node->uid);
3229 cgraph_redirect_edge_callee (cs, val->spec_node);
3230 redirected_sum += cs->count;
3233 cs = get_next_cgraph_edge_clone (cs);
3237 if (redirected_sum)
3238 update_specialized_profile (val->spec_node, node, redirected_sum);
3242 /* Copy KNOWN_BINFOS to KNOWN_VALS. */
3244 static void
3245 move_binfos_to_values (vec<tree> known_vals,
3246 vec<tree> known_binfos)
3248 tree t;
3249 int i;
3251 for (i = 0; known_binfos.iterate (i, &t); i++)
3252 if (t)
3253 known_vals[i] = t;
3256 /* Return true if there is a replacement equivalent to VALUE, INDEX and OFFSET
3257 among those in the AGGVALS list. */
3259 DEBUG_FUNCTION bool
3260 ipcp_val_in_agg_replacements_p (struct ipa_agg_replacement_value *aggvals,
3261 int index, HOST_WIDE_INT offset, tree value)
3263 while (aggvals)
3265 if (aggvals->index == index
3266 && aggvals->offset == offset
3267 && values_equal_for_ipcp_p (aggvals->value, value))
3268 return true;
3269 aggvals = aggvals->next;
3271 return false;
3274 /* Decide wheter to create a special version of NODE for value VAL of parameter
3275 at the given INDEX. If OFFSET is -1, the value is for the parameter itself,
3276 otherwise it is stored at the given OFFSET of the parameter. KNOWN_CSTS,
3277 KNOWN_BINFOS and KNOWN_AGGS describe the other already known values. */
3279 static bool
3280 decide_about_value (struct cgraph_node *node, int index, HOST_WIDE_INT offset,
3281 struct ipcp_value *val, vec<tree> known_csts,
3282 vec<tree> known_binfos)
3284 struct ipa_agg_replacement_value *aggvals;
3285 int freq_sum, caller_count;
3286 gcov_type count_sum;
3287 vec<cgraph_edge_p> callers;
3288 vec<tree> kv;
3290 if (val->spec_node)
3292 perhaps_add_new_callers (node, val);
3293 return false;
3295 else if (val->local_size_cost + overall_size > max_new_size)
3297 if (dump_file && (dump_flags & TDF_DETAILS))
3298 fprintf (dump_file, " Ignoring candidate value because "
3299 "max_new_size would be reached with %li.\n",
3300 val->local_size_cost + overall_size);
3301 return false;
3303 else if (!get_info_about_necessary_edges (val, &freq_sum, &count_sum,
3304 &caller_count))
3305 return false;
3307 if (dump_file && (dump_flags & TDF_DETAILS))
3309 fprintf (dump_file, " - considering value ");
3310 print_ipcp_constant_value (dump_file, val->value);
3311 fprintf (dump_file, " for parameter ");
3312 print_generic_expr (dump_file, ipa_get_param (IPA_NODE_REF (node),
3313 index), 0);
3314 if (offset != -1)
3315 fprintf (dump_file, ", offset: " HOST_WIDE_INT_PRINT_DEC, offset);
3316 fprintf (dump_file, " (caller_count: %i)\n", caller_count);
3319 if (!good_cloning_opportunity_p (node, val->local_time_benefit,
3320 freq_sum, count_sum,
3321 val->local_size_cost)
3322 && !good_cloning_opportunity_p (node,
3323 val->local_time_benefit
3324 + val->prop_time_benefit,
3325 freq_sum, count_sum,
3326 val->local_size_cost
3327 + val->prop_size_cost))
3328 return false;
3330 if (dump_file)
3331 fprintf (dump_file, " Creating a specialized node of %s/%i.\n",
3332 cgraph_node_name (node), node->uid);
3334 callers = gather_edges_for_value (val, caller_count);
3335 kv = known_csts.copy ();
3336 move_binfos_to_values (kv, known_binfos);
3337 if (offset == -1)
3338 kv[index] = val->value;
3339 find_more_scalar_values_for_callers_subset (node, kv, callers);
3340 aggvals = find_aggregate_values_for_callers_subset (node, callers);
3341 gcc_checking_assert (offset == -1
3342 || ipcp_val_in_agg_replacements_p (aggvals, index,
3343 offset, val->value));
3344 val->spec_node = create_specialized_node (node, kv, aggvals, callers);
3345 overall_size += val->local_size_cost;
3347 /* TODO: If for some lattice there is only one other known value
3348 left, make a special node for it too. */
3350 return true;
3353 /* Decide whether and what specialized clones of NODE should be created. */
3355 static bool
3356 decide_whether_version_node (struct cgraph_node *node)
3358 struct ipa_node_params *info = IPA_NODE_REF (node);
3359 int i, count = ipa_get_param_count (info);
3360 vec<tree> known_csts, known_binfos;
3361 vec<ipa_agg_jump_function_t> known_aggs = vNULL;
3362 bool ret = false;
3364 if (count == 0)
3365 return false;
3367 if (dump_file && (dump_flags & TDF_DETAILS))
3368 fprintf (dump_file, "\nEvaluating opportunities for %s/%i.\n",
3369 cgraph_node_name (node), node->uid);
3371 gather_context_independent_values (info, &known_csts, &known_binfos,
3372 info->do_clone_for_all_contexts ? &known_aggs
3373 : NULL, NULL);
3375 for (i = 0; i < count ;i++)
3377 struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
3378 struct ipcp_lattice *lat = &plats->itself;
3379 struct ipcp_value *val;
3381 if (!lat->bottom
3382 && !known_csts[i]
3383 && !known_binfos[i])
3384 for (val = lat->values; val; val = val->next)
3385 ret |= decide_about_value (node, i, -1, val, known_csts,
3386 known_binfos);
3388 if (!plats->aggs_bottom)
3390 struct ipcp_agg_lattice *aglat;
3391 struct ipcp_value *val;
3392 for (aglat = plats->aggs; aglat; aglat = aglat->next)
3393 if (!aglat->bottom && aglat->values
3394 /* If the following is false, the one value is in
3395 known_aggs. */
3396 && (plats->aggs_contain_variable
3397 || !ipa_lat_is_single_const (aglat)))
3398 for (val = aglat->values; val; val = val->next)
3399 ret |= decide_about_value (node, i, aglat->offset, val,
3400 known_csts, known_binfos);
3402 info = IPA_NODE_REF (node);
3405 if (info->do_clone_for_all_contexts)
3407 struct cgraph_node *clone;
3408 vec<cgraph_edge_p> callers;
3410 if (dump_file)
3411 fprintf (dump_file, " - Creating a specialized node of %s/%i "
3412 "for all known contexts.\n", cgraph_node_name (node),
3413 node->uid);
3415 callers = collect_callers_of_node (node);
3416 move_binfos_to_values (known_csts, known_binfos);
3417 clone = create_specialized_node (node, known_csts,
3418 known_aggs_to_agg_replacement_list (known_aggs),
3419 callers);
3420 info = IPA_NODE_REF (node);
3421 info->do_clone_for_all_contexts = false;
3422 IPA_NODE_REF (clone)->is_all_contexts_clone = true;
3423 for (i = 0; i < count ; i++)
3424 vec_free (known_aggs[i].items);
3425 known_aggs.release ();
3426 ret = true;
3428 else
3429 known_csts.release ();
3431 known_binfos.release ();
3432 return ret;
3435 /* Transitively mark all callees of NODE within the same SCC as not dead. */
3437 static void
3438 spread_undeadness (struct cgraph_node *node)
3440 struct cgraph_edge *cs;
3442 for (cs = node->callees; cs; cs = cs->next_callee)
3443 if (edge_within_scc (cs))
3445 struct cgraph_node *callee;
3446 struct ipa_node_params *info;
3448 callee = cgraph_function_node (cs->callee, NULL);
3449 info = IPA_NODE_REF (callee);
3451 if (info->node_dead)
3453 info->node_dead = 0;
3454 spread_undeadness (callee);
3459 /* Return true if NODE has a caller from outside of its SCC that is not
3460 dead. Worker callback for cgraph_for_node_and_aliases. */
3462 static bool
3463 has_undead_caller_from_outside_scc_p (struct cgraph_node *node,
3464 void *data ATTRIBUTE_UNUSED)
3466 struct cgraph_edge *cs;
3468 for (cs = node->callers; cs; cs = cs->next_caller)
3469 if (cs->caller->thunk.thunk_p
3470 && cgraph_for_node_and_aliases (cs->caller,
3471 has_undead_caller_from_outside_scc_p,
3472 NULL, true))
3473 return true;
3474 else if (!edge_within_scc (cs)
3475 && !IPA_NODE_REF (cs->caller)->node_dead)
3476 return true;
3477 return false;
3481 /* Identify nodes within the same SCC as NODE which are no longer needed
3482 because of new clones and will be removed as unreachable. */
3484 static void
3485 identify_dead_nodes (struct cgraph_node *node)
3487 struct cgraph_node *v;
3488 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3489 if (cgraph_will_be_removed_from_program_if_no_direct_calls (v)
3490 && !cgraph_for_node_and_aliases (v,
3491 has_undead_caller_from_outside_scc_p,
3492 NULL, true))
3493 IPA_NODE_REF (v)->node_dead = 1;
3495 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3496 if (!IPA_NODE_REF (v)->node_dead)
3497 spread_undeadness (v);
3499 if (dump_file && (dump_flags & TDF_DETAILS))
3501 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3502 if (IPA_NODE_REF (v)->node_dead)
3503 fprintf (dump_file, " Marking node as dead: %s/%i.\n",
3504 cgraph_node_name (v), v->uid);
3508 /* The decision stage. Iterate over the topological order of call graph nodes
3509 TOPO and make specialized clones if deemed beneficial. */
3511 static void
3512 ipcp_decision_stage (struct topo_info *topo)
3514 int i;
3516 if (dump_file)
3517 fprintf (dump_file, "\nIPA decision stage:\n\n");
3519 for (i = topo->nnodes - 1; i >= 0; i--)
3521 struct cgraph_node *node = topo->order[i];
3522 bool change = false, iterate = true;
3524 while (iterate)
3526 struct cgraph_node *v;
3527 iterate = false;
3528 for (v = node; v ; v = ((struct ipa_dfs_info *) v->symbol.aux)->next_cycle)
3529 if (cgraph_function_with_gimple_body_p (v)
3530 && ipcp_versionable_function_p (v))
3531 iterate |= decide_whether_version_node (v);
3533 change |= iterate;
3535 if (change)
3536 identify_dead_nodes (node);
3540 /* The IPCP driver. */
3542 static unsigned int
3543 ipcp_driver (void)
3545 struct cgraph_2edge_hook_list *edge_duplication_hook_holder;
3546 struct topo_info topo;
3548 ipa_check_create_node_params ();
3549 ipa_check_create_edge_args ();
3550 grow_next_edge_clone_vector ();
3551 edge_duplication_hook_holder =
3552 cgraph_add_edge_duplication_hook (&ipcp_edge_duplication_hook, NULL);
3553 ipcp_values_pool = create_alloc_pool ("IPA-CP values",
3554 sizeof (struct ipcp_value), 32);
3555 ipcp_sources_pool = create_alloc_pool ("IPA-CP value sources",
3556 sizeof (struct ipcp_value_source), 64);
3557 ipcp_agg_lattice_pool = create_alloc_pool ("IPA_CP aggregate lattices",
3558 sizeof (struct ipcp_agg_lattice),
3559 32);
3560 if (dump_file)
3562 fprintf (dump_file, "\nIPA structures before propagation:\n");
3563 if (dump_flags & TDF_DETAILS)
3564 ipa_print_all_params (dump_file);
3565 ipa_print_all_jump_functions (dump_file);
3568 /* Topological sort. */
3569 build_toporder_info (&topo);
3570 /* Do the interprocedural propagation. */
3571 ipcp_propagate_stage (&topo);
3572 /* Decide what constant propagation and cloning should be performed. */
3573 ipcp_decision_stage (&topo);
3575 /* Free all IPCP structures. */
3576 free_toporder_info (&topo);
3577 next_edge_clone.release ();
3578 cgraph_remove_edge_duplication_hook (edge_duplication_hook_holder);
3579 ipa_free_all_structures_after_ipa_cp ();
3580 if (dump_file)
3581 fprintf (dump_file, "\nIPA constant propagation end\n");
3582 return 0;
3585 /* Initialization and computation of IPCP data structures. This is the initial
3586 intraprocedural analysis of functions, which gathers information to be
3587 propagated later on. */
3589 static void
3590 ipcp_generate_summary (void)
3592 struct cgraph_node *node;
3594 if (dump_file)
3595 fprintf (dump_file, "\nIPA constant propagation start:\n");
3596 ipa_register_cgraph_hooks ();
3598 FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node)
3600 node->local.versionable
3601 = tree_versionable_function_p (node->symbol.decl);
3602 ipa_analyze_node (node);
3606 /* Write ipcp summary for nodes in SET. */
3608 static void
3609 ipcp_write_summary (void)
3611 ipa_prop_write_jump_functions ();
3614 /* Read ipcp summary. */
3616 static void
3617 ipcp_read_summary (void)
3619 ipa_prop_read_jump_functions ();
3622 /* Gate for IPCP optimization. */
3624 static bool
3625 cgraph_gate_cp (void)
3627 /* FIXME: We should remove the optimize check after we ensure we never run
3628 IPA passes when not optimizing. */
3629 return flag_ipa_cp && optimize;
3632 struct ipa_opt_pass_d pass_ipa_cp =
3635 IPA_PASS,
3636 "cp", /* name */
3637 OPTGROUP_NONE, /* optinfo_flags */
3638 cgraph_gate_cp, /* gate */
3639 ipcp_driver, /* execute */
3640 NULL, /* sub */
3641 NULL, /* next */
3642 0, /* static_pass_number */
3643 TV_IPA_CONSTANT_PROP, /* tv_id */
3644 0, /* properties_required */
3645 0, /* properties_provided */
3646 0, /* properties_destroyed */
3647 0, /* todo_flags_start */
3648 TODO_dump_symtab |
3649 TODO_remove_functions /* todo_flags_finish */
3651 ipcp_generate_summary, /* generate_summary */
3652 ipcp_write_summary, /* write_summary */
3653 ipcp_read_summary, /* read_summary */
3654 ipa_prop_write_all_agg_replacement, /* write_optimization_summary */
3655 ipa_prop_read_all_agg_replacement, /* read_optimization_summary */
3656 NULL, /* stmt_fixup */
3657 0, /* TODOs */
3658 ipcp_transform_function, /* function_transform */
3659 NULL, /* variable_transform */