2 * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Cliff Click's Combined Analysis/Optimization
23 * @author Michael Beck
26 * This is a slightly enhanced version of Cliff Clicks combo algorithm
27 * - support for commutative nodes is added, Add(a,b) and Add(b,a) ARE congruent
28 * - supports all Firm direct (by a data edge) identities except Mux
29 * (Mux can be a 2-input or 1-input identity, only 2-input is implemented yet)
30 * - supports Confirm nodes (handle them like Copies but do NOT remove them)
31 * - let Cmp nodes calculate Top like all othe data nodes: this would let
32 * Mux nodes to calculate Unknown instead of taking the true result
33 * - let Cond(Top) always select FALSE/default: This is tricky. Nodes are only reavaluated
34 * IFF the predecessor changed its type. Because nodes are initialized with Top
35 * this never happens, let all Proj(Cond) be unreachable.
36 * We avoid this condition by the same way we work around Phi: whenever a Block
37 * node is placed on the list, place its Cond nodes (and because they are Tuple
38 * all its Proj-nodes either on the cprop list)
39 * Especially, this changes the meaning of Click's example:
54 * using Click's version while is silent with our.
55 * - support for global congruences is implemented but not tested yet
57 * Note further that we use the terminology from Click's work here, which is different
58 * in some cases from Firm terminology. Especially, Click's type is a
59 * Firm tarval/entity, nevertheless we call it type here for "maximum compatibility".
65 #include "iroptimize.h"
72 #include "irgraph_t.h"
79 #include "iropt_dbg.h"
83 #include "irnodeset.h"
90 /* define this to check that all type translations are monotone */
91 #define VERIFY_MONOTONE
93 /* define this to check the consistency of partitions */
94 #define CHECK_PARTITIONS
96 typedef struct node_t node_t;
97 typedef struct partition_t partition_t;
98 typedef struct opcode_key_t opcode_key_t;
99 typedef struct listmap_entry_t listmap_entry_t;
101 /** The type of the compute function. */
102 typedef void (*compute_func)(node_t *node);
107 struct opcode_key_t {
108 ir_opcode code; /**< The Firm opcode. */
109 ir_mode *mode; /**< The mode of all nodes in the partition. */
110 int arity; /**< The arity of this opcode (needed for Phi etc. */
112 long proj; /**< For Proj nodes, its proj number */
113 ir_entity *ent; /**< For Sel Nodes, its entity */
114 int intVal; /**< For Conv/Div Nodes: strict/remainderless */
115 unsigned uintVal;/**< for Builtin: the kind */
116 ir_node *block; /**< for Block: itself */
117 void *ptr; /**< generic pointer for hash/cmp */
122 * An entry in the list_map.
124 struct listmap_entry_t {
125 void *id; /**< The id. */
126 node_t *list; /**< The associated list for this id. */
127 listmap_entry_t *next; /**< Link to the next entry in the map. */
130 /** We must map id's to lists. */
131 typedef struct listmap_t {
132 set *map; /**< Map id's to listmap_entry_t's */
133 listmap_entry_t *values; /**< List of all values in the map. */
137 * A lattice element. Because we handle constants and symbolic constants different, we
138 * have to use this union.
149 ir_node *node; /**< The IR-node itself. */
150 list_head node_list; /**< Double-linked list of leader/follower entries. */
151 list_head cprop_list; /**< Double-linked partition.cprop list. */
152 partition_t *part; /**< points to the partition this node belongs to */
153 node_t *next; /**< Next node on local list (partition.touched, fallen). */
154 node_t *race_next; /**< Next node on race list. */
155 lattice_elem_t type; /**< The associated lattice element "type". */
156 int max_user_input; /**< Maximum input number of Def-Use edges. */
157 int next_edge; /**< Index of the next Def-Use edge to use. */
158 int n_followers; /**< Number of Follower in the outs set. */
159 unsigned on_touched:1; /**< Set, if this node is on the partition.touched set. */
160 unsigned on_cprop:1; /**< Set, if this node is on the partition.cprop list. */
161 unsigned on_fallen:1; /**< Set, if this node is on the fallen list. */
162 unsigned is_follower:1; /**< Set, if this node is a follower. */
163 unsigned flagged:2; /**< 2 Bits, set if this node was visited by race 1 or 2. */
167 * A partition containing congruent nodes.
170 list_head Leader; /**< The head of partition Leader node list. */
171 list_head Follower; /**< The head of partition Follower node list. */
172 list_head cprop; /**< The head of partition.cprop list. */
173 list_head cprop_X; /**< The head of partition.cprop (Cond nodes and its Projs) list. */
174 partition_t *wl_next; /**< Next entry in the work list if any. */
175 partition_t *touched_next; /**< Points to the next partition in the touched set. */
176 partition_t *cprop_next; /**< Points to the next partition in the cprop list. */
177 partition_t *split_next; /**< Points to the next partition in the list that must be split by split_by(). */
178 node_t *touched; /**< The partition.touched set of this partition. */
179 unsigned n_leader; /**< Number of entries in this partition.Leader. */
180 unsigned n_touched; /**< Number of entries in the partition.touched. */
181 int max_user_inputs; /**< Maximum number of user inputs of all entries. */
182 unsigned on_worklist:1; /**< Set, if this partition is in the work list. */
183 unsigned on_touched:1; /**< Set, if this partition is on the touched set. */
184 unsigned on_cprop:1; /**< Set, if this partition is on the cprop list. */
185 unsigned type_is_T_or_C:1;/**< Set, if all nodes in this partition have type Top or Constant. */
187 partition_t *dbg_next; /**< Link all partitions for debugging */
188 unsigned nr; /**< A unique number for (what-)mapping, >0. */
192 typedef struct environment_t {
193 struct obstack obst; /**< obstack to allocate data structures. */
194 partition_t *worklist; /**< The work list. */
195 partition_t *cprop; /**< The constant propagation list. */
196 partition_t *touched; /**< the touched set. */
197 partition_t *initial; /**< The initial partition. */
198 set *opcode2id_map; /**< The opcodeMode->id map. */
199 pmap *type2id_map; /**< The type->id map. */
200 ir_node **kept_memory; /**< Array of memory nodes that must be kept. */
201 int end_idx; /**< -1 for local and 0 for global congruences. */
202 int lambda_input; /**< Captured argument for lambda_partition(). */
203 unsigned modified:1; /**< Set, if the graph was modified. */
204 unsigned unopt_cf:1; /**< If set, control flow is not optimized due to Unknown. */
205 /* options driving the optimization */
206 unsigned commutative:1; /**< Set, if commutation nodes should be handled specially. */
207 unsigned opt_unknown:1; /**< Set, if non-strict programs should be optimized. */
209 partition_t *dbg_list; /**< List of all partitions. */
213 /** Type of the what function. */
214 typedef void *(*what_func)(const node_t *node, environment_t *env);
216 #define get_irn_node(irn) ((node_t *)get_irn_link(irn))
217 #define set_irn_node(irn, node) set_irn_link(irn, node)
219 /* we do NOT use tarval_unreachable here, instead we use Top for this purpose */
220 #undef tarval_unreachable
221 #define tarval_unreachable tarval_top
224 /** The debug module handle. */
225 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
227 /** The what reason. */
228 DEBUG_ONLY(static const char *what_reason;)
230 /** Next partition number. */
231 DEBUG_ONLY(static unsigned part_nr = 0);
233 /** The tarval returned by Unknown nodes: set to either tarval_bad OR tarval_top. */
234 static tarval *tarval_UNKNOWN;
237 static node_t *identity(node_t *node);
239 #ifdef CHECK_PARTITIONS
243 static void check_partition(const partition_t *T) {
247 list_for_each_entry(node_t, node, &T->Leader, node_list) {
248 assert(node->is_follower == 0);
249 assert(node->flagged == 0);
250 assert(node->part == T);
253 assert(n == T->n_leader);
255 list_for_each_entry(node_t, node, &T->Follower, node_list) {
256 assert(node->is_follower == 1);
257 assert(node->flagged == 0);
258 assert(node->part == T);
260 } /* check_partition */
263 * check that all leader nodes in the partition have the same opcode.
265 static void check_opcode(const partition_t *Z) {
270 list_for_each_entry(node_t, node, &Z->Leader, node_list) {
271 ir_node *irn = node->node;
274 key.code = get_irn_opcode(irn);
275 key.mode = get_irn_mode(irn);
276 key.arity = get_irn_arity(irn);
280 switch (get_irn_opcode(irn)) {
282 key.u.proj = get_Proj_proj(irn);
285 key.u.ent = get_Sel_entity(irn);
288 key.u.intVal = get_Conv_strict(irn);
291 key.u.intVal = get_Div_no_remainder(irn);
297 key.mode = get_Load_mode(irn);
300 key.u.intVal = get_Builtin_kind(irn);
307 assert((unsigned)key.code == get_irn_opcode(irn));
308 assert(key.mode == get_irn_mode(irn));
309 assert(key.arity == get_irn_arity(irn));
311 switch (get_irn_opcode(irn)) {
313 assert(key.u.proj == get_Proj_proj(irn));
316 assert(key.u.ent == get_Sel_entity(irn));
319 assert(key.u.intVal == get_Conv_strict(irn));
322 assert(key.u.intVal == get_Div_no_remainder(irn));
325 assert(key.u.block == irn);
328 assert(key.mode == get_Load_mode(irn));
331 assert(key.u.intVal == (int) get_Builtin_kind(irn));
340 static void check_all_partitions(environment_t *env) {
345 for (P = env->dbg_list; P != NULL; P = P->dbg_next) {
347 if (! P->type_is_T_or_C)
349 list_for_each_entry(node_t, node, &P->Follower, node_list) {
350 node_t *leader = identity(node);
352 assert(leader != node && leader->part == node->part);
361 static void do_check_list(const node_t *list, int ofs, const partition_t *Z) {
364 #define NEXT(e) *((const node_t **)((char *)(e) + (ofs)))
365 for (e = list; e != NULL; e = NEXT(e)) {
366 assert(e->part == Z);
369 } /* ido_check_list */
372 * Check a local list.
374 static void check_list(const node_t *list, const partition_t *Z) {
375 do_check_list(list, offsetof(node_t, next), Z);
379 #define check_partition(T)
380 #define check_list(list, Z)
381 #define check_all_partitions(env)
382 #endif /* CHECK_PARTITIONS */
385 static inline lattice_elem_t get_partition_type(const partition_t *X);
388 * Dump partition to output.
390 static void dump_partition(const char *msg, const partition_t *part) {
393 lattice_elem_t type = get_partition_type(part);
395 DB((dbg, LEVEL_2, "%s part%u%s (%u, %+F) {\n ",
396 msg, part->nr, part->type_is_T_or_C ? "*" : "",
397 part->n_leader, type));
398 list_for_each_entry(node_t, node, &part->Leader, node_list) {
399 DB((dbg, LEVEL_2, "%s%+F", first ? "" : ", ", node->node));
402 if (! list_empty(&part->Follower)) {
403 DB((dbg, LEVEL_2, "\n---\n "));
405 list_for_each_entry(node_t, node, &part->Follower, node_list) {
406 DB((dbg, LEVEL_2, "%s%+F", first ? "" : ", ", node->node));
410 DB((dbg, LEVEL_2, "\n}\n"));
411 } /* dump_partition */
416 static void do_dump_list(const char *msg, const node_t *node, int ofs) {
420 #define GET_LINK(p, ofs) *((const node_t **)((char *)(p) + (ofs)))
422 DB((dbg, LEVEL_3, "%s = {\n ", msg));
423 for (p = node; p != NULL; p = GET_LINK(p, ofs)) {
424 DB((dbg, LEVEL_3, "%s%+F", first ? "" : ", ", p->node));
427 DB((dbg, LEVEL_3, "\n}\n"));
435 static void dump_race_list(const char *msg, const node_t *list) {
436 do_dump_list(msg, list, offsetof(node_t, race_next));
437 } /* dump_race_list */
440 * Dumps a local list.
442 static void dump_list(const char *msg, const node_t *list) {
443 do_dump_list(msg, list, offsetof(node_t, next));
447 * Dump all partitions.
449 static void dump_all_partitions(const environment_t *env) {
450 const partition_t *P;
452 DB((dbg, LEVEL_2, "All partitions\n===============\n"));
453 for (P = env->dbg_list; P != NULL; P = P->dbg_next)
454 dump_partition("", P);
455 } /* dump_all_partitions */
460 static void dump_split_list(const partition_t *list) {
461 const partition_t *p;
463 DB((dbg, LEVEL_2, "Split by %s produced = {\n", what_reason));
464 for (p = list; p != NULL; p = p->split_next)
465 DB((dbg, LEVEL_2, "part%u, ", p->nr));
466 DB((dbg, LEVEL_2, "\n}\n"));
467 } /* dump_split_list */
470 * Dump partition and type for a node.
472 static int dump_partition_hook(FILE *F, ir_node *n, ir_node *local) {
473 ir_node *irn = local != NULL ? local : n;
474 node_t *node = get_irn_node(irn);
476 ir_fprintf(F, "info2 : \"partition %u type %+F\"\n", node->part->nr, node->type);
478 } /* dump_partition_hook */
481 #define dump_partition(msg, part)
482 #define dump_race_list(msg, list)
483 #define dump_list(msg, list)
484 #define dump_all_partitions(env)
485 #define dump_split_list(list)
488 #if defined(VERIFY_MONOTONE) && defined (DEBUG_libfirm)
490 * Verify that a type transition is monotone
492 static void verify_type(const lattice_elem_t old_type, node_t *node) {
493 if (old_type.tv == node->type.tv) {
497 if (old_type.tv == tarval_top) {
498 /* from Top down-to is always allowed */
501 if (node->type.tv == tarval_bottom || node->type.tv == tarval_reachable) {
505 panic("combo: wrong translation from %+F to %+F on node %+F", old_type, node->type, node->node);
509 #define verify_type(old_type, node)
513 * Compare two pointer values of a listmap.
515 static int listmap_cmp_ptr(const void *elt, const void *key, size_t size) {
516 const listmap_entry_t *e1 = elt;
517 const listmap_entry_t *e2 = key;
520 return e1->id != e2->id;
521 } /* listmap_cmp_ptr */
524 * Initializes a listmap.
526 * @param map the listmap
528 static void listmap_init(listmap_t *map) {
529 map->map = new_set(listmap_cmp_ptr, 16);
534 * Terminates a listmap.
536 * @param map the listmap
538 static void listmap_term(listmap_t *map) {
543 * Return the associated listmap entry for a given id.
545 * @param map the listmap
546 * @param id the id to search for
548 * @return the associated listmap entry for the given id
550 static listmap_entry_t *listmap_find(listmap_t *map, void *id) {
551 listmap_entry_t key, *entry;
556 entry = set_insert(map->map, &key, sizeof(key), HASH_PTR(id));
558 if (entry->list == NULL) {
559 /* a new entry, put into the list */
560 entry->next = map->values;
567 * Calculate the hash value for an opcode map entry.
569 * @param entry an opcode map entry
571 * @return a hash value for the given opcode map entry
573 static unsigned opcode_hash(const opcode_key_t *entry) {
574 return (entry->mode - (ir_mode *)0) * 9 + entry->code + entry->u.proj * 3 + HASH_PTR(entry->u.ptr) + entry->arity;
578 * Compare two entries in the opcode map.
580 static int cmp_opcode(const void *elt, const void *key, size_t size) {
581 const opcode_key_t *o1 = elt;
582 const opcode_key_t *o2 = key;
585 return o1->code != o2->code || o1->mode != o2->mode ||
586 o1->arity != o2->arity ||
587 o1->u.proj != o2->u.proj ||
588 o1->u.intVal != o2->u.intVal || /* this already checks uIntVal */
589 o1->u.ptr != o2->u.ptr;
593 * Compare two Def-Use edges for input position.
595 static int cmp_def_use_edge(const void *a, const void *b) {
596 const ir_def_use_edge *ea = a;
597 const ir_def_use_edge *eb = b;
599 /* no overrun, because range is [-1, MAXINT] */
600 return ea->pos - eb->pos;
601 } /* cmp_def_use_edge */
604 * We need the Def-Use edges sorted.
606 static void sort_irn_outs(node_t *node) {
607 ir_node *irn = node->node;
608 int n_outs = get_irn_n_outs(irn);
611 qsort(&irn->out[1], n_outs, sizeof(irn->out[0]), cmp_def_use_edge);
613 node->max_user_input = irn->out[n_outs].pos;
614 } /* sort_irn_outs */
617 * Return the type of a node.
619 * @param irn an IR-node
621 * @return the associated type of this node
623 static inline lattice_elem_t get_node_type(const ir_node *irn) {
624 return get_irn_node(irn)->type;
625 } /* get_node_type */
628 * Return the tarval of a node.
630 * @param irn an IR-node
632 * @return the associated type of this node
634 static inline tarval *get_node_tarval(const ir_node *irn) {
635 lattice_elem_t type = get_node_type(irn);
637 if (is_tarval(type.tv))
639 return tarval_bottom;
640 } /* get_node_type */
643 * Add a partition to the worklist.
645 static inline void add_to_worklist(partition_t *X, environment_t *env) {
646 assert(X->on_worklist == 0);
647 DB((dbg, LEVEL_2, "Adding part%d to worklist\n", X->nr));
648 X->wl_next = env->worklist;
651 } /* add_to_worklist */
654 * Create a new empty partition.
656 * @param env the environment
658 * @return a newly allocated partition
660 static inline partition_t *new_partition(environment_t *env) {
661 partition_t *part = obstack_alloc(&env->obst, sizeof(*part));
663 INIT_LIST_HEAD(&part->Leader);
664 INIT_LIST_HEAD(&part->Follower);
665 INIT_LIST_HEAD(&part->cprop);
666 INIT_LIST_HEAD(&part->cprop_X);
667 part->wl_next = NULL;
668 part->touched_next = NULL;
669 part->cprop_next = NULL;
670 part->split_next = NULL;
671 part->touched = NULL;
674 part->max_user_inputs = 0;
675 part->on_worklist = 0;
676 part->on_touched = 0;
678 part->type_is_T_or_C = 0;
680 part->dbg_next = env->dbg_list;
681 env->dbg_list = part;
682 part->nr = part_nr++;
686 } /* new_partition */
689 * Get the first node from a partition.
691 static inline node_t *get_first_node(const partition_t *X) {
692 return list_entry(X->Leader.next, node_t, node_list);
693 } /* get_first_node */
696 * Return the type of a partition (assuming partition is non-empty and
697 * all elements have the same type).
699 * @param X a partition
701 * @return the type of the first element of the partition
703 static inline lattice_elem_t get_partition_type(const partition_t *X) {
704 const node_t *first = get_first_node(X);
706 } /* get_partition_type */
709 * Creates a partition node for the given IR-node and place it
710 * into the given partition.
712 * @param irn an IR-node
713 * @param part a partition to place the node in
714 * @param env the environment
716 * @return the created node
718 static node_t *create_partition_node(ir_node *irn, partition_t *part, environment_t *env) {
719 /* create a partition node and place it in the partition */
720 node_t *node = obstack_alloc(&env->obst, sizeof(*node));
722 INIT_LIST_HEAD(&node->node_list);
723 INIT_LIST_HEAD(&node->cprop_list);
727 node->race_next = NULL;
728 node->type.tv = tarval_top;
729 node->max_user_input = 0;
731 node->n_followers = 0;
732 node->on_touched = 0;
735 node->is_follower = 0;
737 set_irn_node(irn, node);
739 list_add_tail(&node->node_list, &part->Leader);
743 } /* create_partition_node */
746 * Pre-Walker, initialize all Nodes' type to U or top and place
747 * all nodes into the TOP partition.
749 static void create_initial_partitions(ir_node *irn, void *ctx) {
750 environment_t *env = ctx;
751 partition_t *part = env->initial;
754 node = create_partition_node(irn, part, env);
756 if (node->max_user_input > part->max_user_inputs)
757 part->max_user_inputs = node->max_user_input;
760 set_Block_phis(irn, NULL);
762 } /* create_initial_partitions */
765 * Post-Walker, collect all Block-Phi lists, set Cond.
767 static void init_block_phis(ir_node *irn, void *ctx) {
771 add_Block_phi(get_nodes_block(irn), irn);
773 } /* init_block_phis */
776 * Add a node to the entry.partition.touched set and
777 * node->partition to the touched set if not already there.
780 * @param env the environment
782 static inline void add_to_touched(node_t *y, environment_t *env) {
783 if (y->on_touched == 0) {
784 partition_t *part = y->part;
786 y->next = part->touched;
791 if (part->on_touched == 0) {
792 part->touched_next = env->touched;
794 part->on_touched = 1;
797 check_list(part->touched, part);
799 } /* add_to_touched */
802 * Place a node on the cprop list.
805 * @param env the environment
807 static void add_to_cprop(node_t *y, environment_t *env) {
810 /* Add y to y.partition.cprop. */
811 if (y->on_cprop == 0) {
812 partition_t *Y = y->part;
813 ir_node *irn = y->node;
815 /* place Conds and all its Projs on the cprop_X list */
816 if (is_Cond(skip_Proj(irn)))
817 list_add_tail(&y->cprop_list, &Y->cprop_X);
819 list_add_tail(&y->cprop_list, &Y->cprop);
822 DB((dbg, LEVEL_3, "Add %+F to part%u.cprop\n", y->node, Y->nr));
824 /* place its partition on the cprop list */
825 if (Y->on_cprop == 0) {
826 Y->cprop_next = env->cprop;
832 if (get_irn_mode(irn) == mode_T) {
833 /* mode_T nodes always produce tarval_bottom, so we must explicitly
834 add it's Proj's to get constant evaluation to work */
837 for (i = get_irn_n_outs(irn) - 1; i >= 0; --i) {
838 node_t *proj = get_irn_node(get_irn_out(irn, i));
840 add_to_cprop(proj, env);
842 } else if (is_Block(irn)) {
843 /* Due to the way we handle Phi's, we must place all Phis of a block on the list
844 * if someone placed the block. The Block is only placed if the reachability
845 * changes, and this must be re-evaluated in compute_Phi(). */
847 for (phi = get_Block_phis(irn); phi != NULL; phi = get_Phi_next(phi)) {
848 node_t *p = get_irn_node(phi);
849 add_to_cprop(p, env);
855 * Update the worklist: If Z is on worklist then add Z' to worklist.
856 * Else add the smaller of Z and Z' to worklist.
858 * @param Z the Z partition
859 * @param Z_prime the Z' partition, a previous part of Z
860 * @param env the environment
862 static void update_worklist(partition_t *Z, partition_t *Z_prime, environment_t *env) {
863 if (Z->on_worklist || Z_prime->n_leader < Z->n_leader) {
864 add_to_worklist(Z_prime, env);
866 add_to_worklist(Z, env);
868 } /* update_worklist */
871 * Make all inputs to x no longer be F.def_use edges.
875 static void move_edges_to_leader(node_t *x) {
876 ir_node *irn = x->node;
879 for (i = get_irn_arity(irn) - 1; i >= 0; --i) {
880 node_t *pred = get_irn_node(get_irn_n(irn, i));
885 n = get_irn_n_outs(p);
886 for (j = 1; j <= pred->n_followers; ++j) {
887 if (p->out[j].pos == i && p->out[j].use == irn) {
888 /* found a follower edge to x, move it to the Leader */
889 ir_def_use_edge edge = p->out[j];
891 /* remove this edge from the Follower set */
892 p->out[j] = p->out[pred->n_followers];
895 /* sort it into the leader set */
896 for (k = pred->n_followers + 2; k <= n; ++k) {
897 if (p->out[k].pos >= edge.pos)
899 p->out[k - 1] = p->out[k];
901 /* place the new edge here */
902 p->out[k - 1] = edge;
904 /* edge found and moved */
909 } /* move_edges_to_leader */
912 * Split a partition that has NO followers by a local list.
914 * @param Z partition to split
915 * @param g a (non-empty) node list
916 * @param env the environment
918 * @return a new partition containing the nodes of g
920 static partition_t *split_no_followers(partition_t *Z, node_t *g, environment_t *env) {
921 partition_t *Z_prime;
926 dump_partition("Splitting ", Z);
927 dump_list("by list ", g);
931 /* Remove g from Z. */
932 for (node = g; node != NULL; node = node->next) {
933 assert(node->part == Z);
934 list_del(&node->node_list);
937 assert(n < Z->n_leader);
940 /* Move g to a new partition, Z'. */
941 Z_prime = new_partition(env);
943 for (node = g; node != NULL; node = node->next) {
944 list_add_tail(&node->node_list, &Z_prime->Leader);
945 node->part = Z_prime;
946 if (node->max_user_input > max_input)
947 max_input = node->max_user_input;
949 Z_prime->max_user_inputs = max_input;
950 Z_prime->n_leader = n;
953 check_partition(Z_prime);
955 /* for now, copy the type info tag, it will be adjusted in split_by(). */
956 Z_prime->type_is_T_or_C = Z->type_is_T_or_C;
958 update_worklist(Z, Z_prime, env);
960 dump_partition("Now ", Z);
961 dump_partition("Created new ", Z_prime);
963 } /* split_no_followers */
966 * Make the Follower -> Leader transition for a node.
970 static void follower_to_leader(node_t *n) {
971 assert(n->is_follower == 1);
973 DB((dbg, LEVEL_2, "%+F make the follower -> leader transition\n", n->node));
975 move_edges_to_leader(n);
976 list_del(&n->node_list);
977 list_add_tail(&n->node_list, &n->part->Leader);
979 } /* follower_to_leader */
982 * The environment for one race step.
984 typedef struct step_env {
985 node_t *initial; /**< The initial node list. */
986 node_t *unwalked; /**< The unwalked node list. */
987 node_t *walked; /**< The walked node list. */
988 int index; /**< Next index of Follower use_def edge. */
989 unsigned side; /**< side number. */
993 * Return non-zero, if a input is a real follower
995 * @param irn the node to check
996 * @param input number of the input
998 static int is_real_follower(const ir_node *irn, int input) {
1001 switch (get_irn_opcode(irn)) {
1004 /* ignore the Confirm bound input */
1010 /* ignore the Mux sel input */
1015 /* dead inputs are not follower edges */
1016 ir_node *block = get_nodes_block(irn);
1017 node_t *pred = get_irn_node(get_Block_cfgpred(block, input));
1019 if (pred->type.tv == tarval_unreachable)
1029 /* only a Sub x,0 / Shift x,0 might be a follower */
1036 pred = get_irn_node(get_irn_n(irn, input));
1037 if (is_tarval(pred->type.tv) && tarval_is_null(pred->type.tv))
1041 pred = get_irn_node(get_irn_n(irn, input));
1042 if (is_tarval(pred->type.tv) && tarval_is_one(pred->type.tv))
1046 pred = get_irn_node(get_irn_n(irn, input));
1047 if (is_tarval(pred->type.tv) && tarval_is_all_one(pred->type.tv))
1051 assert(!"opcode not implemented yet");
1055 } /* is_real_follower */
1058 * Do one step in the race.
1060 static int step(step_env *env) {
1063 if (env->initial != NULL) {
1064 /* Move node from initial to unwalked */
1066 env->initial = n->race_next;
1068 n->race_next = env->unwalked;
1074 while (env->unwalked != NULL) {
1075 /* let n be the first node in unwalked */
1077 while (env->index < n->n_followers) {
1078 const ir_def_use_edge *edge = &n->node->out[1 + env->index];
1080 /* let m be n.F.def_use[index] */
1081 node_t *m = get_irn_node(edge->use);
1083 assert(m->is_follower);
1085 * Some inputs, like the get_Confirm_bound are NOT
1086 * real followers, sort them out.
1088 if (! is_real_follower(m->node, edge->pos)) {
1094 /* only followers from our partition */
1095 if (m->part != n->part)
1098 if ((m->flagged & env->side) == 0) {
1099 m->flagged |= env->side;
1101 if (m->flagged != 3) {
1102 /* visited the first time */
1103 /* add m to unwalked not as first node (we might still need to
1104 check for more follower node */
1105 m->race_next = n->race_next;
1109 /* else already visited by the other side and on the other list */
1112 /* move n to walked */
1113 env->unwalked = n->race_next;
1114 n->race_next = env->walked;
1122 * Clear the flags from a list and check for
1123 * nodes that where touched from both sides.
1125 * @param list the list
1127 static int clear_flags(node_t *list) {
1131 for (n = list; n != NULL; n = n->race_next) {
1132 if (n->flagged == 3) {
1133 /* we reach a follower from both sides, this will split congruent
1134 * inputs and make it a leader. */
1135 follower_to_leader(n);
1144 * Split a partition by a local list using the race.
1146 * @param pX pointer to the partition to split, might be changed!
1147 * @param gg a (non-empty) node list
1148 * @param env the environment
1150 * @return a new partition containing the nodes of gg
1152 static partition_t *split(partition_t **pX, node_t *gg, environment_t *env) {
1153 partition_t *X = *pX;
1154 partition_t *X_prime;
1157 node_t *g, *h, *node, *t;
1158 int max_input, transitions, winner, shf;
1160 DEBUG_ONLY(static int run = 0;)
1162 DB((dbg, LEVEL_2, "Run %d ", run++));
1163 if (list_empty(&X->Follower)) {
1164 /* if the partition has NO follower, we can use the fast
1165 splitting algorithm. */
1166 return split_no_followers(X, gg, env);
1168 /* else do the race */
1170 dump_partition("Splitting ", X);
1171 dump_list("by list ", gg);
1173 INIT_LIST_HEAD(&tmp);
1175 /* Remove gg from X.Leader and put into g */
1177 for (node = gg; node != NULL; node = node->next) {
1178 assert(node->part == X);
1179 assert(node->is_follower == 0);
1181 list_del(&node->node_list);
1182 list_add_tail(&node->node_list, &tmp);
1183 node->race_next = g;
1188 list_for_each_entry(node_t, node, &X->Leader, node_list) {
1189 node->race_next = h;
1192 /* restore X.Leader */
1193 list_splice(&tmp, &X->Leader);
1195 senv[0].initial = g;
1196 senv[0].unwalked = NULL;
1197 senv[0].walked = NULL;
1201 senv[1].initial = h;
1202 senv[1].unwalked = NULL;
1203 senv[1].walked = NULL;
1208 * Some informations on the race that are not stated clearly in Click's
1210 * 1) A follower stays on the side that reach him first.
1211 * 2) If the other side reches a follower, if will be converted to
1212 * a leader. /This must be done after the race is over, else the
1213 * edges we are iterating on are renumbered./
1214 * 3) /New leader might end up on both sides./
1215 * 4) /If one side ends up with new Leaders, we must ensure that
1216 * they can split out by opcode, hence we have to put _every_
1217 * partition with new Leader nodes on the cprop list, as
1218 * opcode splitting is done by split_by() at the end of
1219 * constant propagation./
1222 if (step(&senv[0])) {
1226 if (step(&senv[1])) {
1231 assert(senv[winner].initial == NULL);
1232 assert(senv[winner].unwalked == NULL);
1234 /* clear flags from walked/unwalked */
1236 transitions = clear_flags(senv[0].unwalked) << shf;
1237 transitions |= clear_flags(senv[0].walked) << shf;
1239 transitions |= clear_flags(senv[1].unwalked) << shf;
1240 transitions |= clear_flags(senv[1].walked) << shf;
1242 dump_race_list("winner ", senv[winner].walked);
1244 /* Move walked_{winner} to a new partition, X'. */
1245 X_prime = new_partition(env);
1248 for (node = senv[winner].walked; node != NULL; node = node->race_next) {
1249 list_del(&node->node_list);
1250 node->part = X_prime;
1251 if (node->is_follower) {
1252 list_add_tail(&node->node_list, &X_prime->Follower);
1254 list_add_tail(&node->node_list, &X_prime->Leader);
1257 if (node->max_user_input > max_input)
1258 max_input = node->max_user_input;
1260 X_prime->n_leader = n;
1261 X_prime->max_user_inputs = max_input;
1262 X->n_leader -= X_prime->n_leader;
1264 /* for now, copy the type info tag, it will be adjusted in split_by(). */
1265 X_prime->type_is_T_or_C = X->type_is_T_or_C;
1268 * Even if a follower was not checked by both sides, it might have
1269 * loose its congruence, so we need to check this case for all follower.
1271 list_for_each_entry_safe(node_t, node, t, &X_prime->Follower, node_list) {
1272 if (identity(node) == node) {
1273 follower_to_leader(node);
1279 check_partition(X_prime);
1281 /* X' is the smaller part */
1282 add_to_worklist(X_prime, env);
1285 * If there where follower to leader transitions, ensure that the nodes
1286 * can be split out if necessary.
1288 if (transitions & 1) {
1289 /* place winner partition on the cprop list */
1290 if (X_prime->on_cprop == 0) {
1291 X_prime->cprop_next = env->cprop;
1292 env->cprop = X_prime;
1293 X_prime->on_cprop = 1;
1296 if (transitions & 2) {
1297 /* place other partition on the cprop list */
1298 if (X->on_cprop == 0) {
1299 X->cprop_next = env->cprop;
1305 dump_partition("Now ", X);
1306 dump_partition("Created new ", X_prime);
1308 /* we have to ensure that the partition containing g is returned */
1318 * Returns non-zero if the i'th input of a Phi node is live.
1320 * @param phi a Phi-node
1321 * @param i an input number
1323 * @return non-zero if the i'th input of the given Phi node is live
1325 static int is_live_input(ir_node *phi, int i) {
1327 ir_node *block = get_nodes_block(phi);
1328 ir_node *pred = get_Block_cfgpred(block, i);
1329 lattice_elem_t type = get_node_type(pred);
1331 return type.tv != tarval_unreachable;
1333 /* else it's the control input, always live */
1335 } /* is_live_input */
1338 * Return non-zero if a type is a constant.
1340 static int is_constant_type(lattice_elem_t type) {
1341 if (type.tv != tarval_bottom && type.tv != tarval_top)
1344 } /* is_constant_type */
1347 * Check whether a type is neither Top or a constant.
1348 * Note: U is handled like Top here, R is a constant.
1350 * @param type the type to check
1352 static int type_is_neither_top_nor_const(const lattice_elem_t type) {
1353 if (is_tarval(type.tv)) {
1354 if (type.tv == tarval_top)
1356 if (tarval_is_constant(type.tv))
1363 } /* type_is_neither_top_nor_const */
1366 * Collect nodes to the touched list.
1368 * @param list the list which contains the nodes that must be evaluated
1369 * @param idx the index of the def_use edge to evaluate
1370 * @param env the environment
1372 static void collect_touched(list_head *list, int idx, environment_t *env) {
1374 int end_idx = env->end_idx;
1376 list_for_each_entry(node_t, x, list, node_list) {
1380 /* leader edges start AFTER follower edges */
1381 x->next_edge = x->n_followers + 1;
1383 num_edges = get_irn_n_outs(x->node);
1385 /* for all edges in x.L.def_use_{idx} */
1386 while (x->next_edge <= num_edges) {
1387 const ir_def_use_edge *edge = &x->node->out[x->next_edge];
1390 /* check if we have necessary edges */
1391 if (edge->pos > idx)
1398 /* only non-commutative nodes */
1399 if (env->commutative &&
1400 (idx == 0 || idx == 1) && is_op_commutative(get_irn_op(succ)))
1403 /* ignore the "control input" for non-pinned nodes
1404 if we are running in GCSE mode */
1405 if (idx < end_idx && get_irn_pinned(succ) != op_pin_state_pinned)
1408 y = get_irn_node(succ);
1409 assert(get_irn_n(succ, idx) == x->node);
1411 /* ignore block edges touching followers */
1412 if (idx == -1 && y->is_follower)
1415 if (is_constant_type(y->type)) {
1416 ir_opcode code = get_irn_opcode(succ);
1417 if (code == iro_Sub || code == iro_Cmp)
1418 add_to_cprop(y, env);
1421 /* Partitions of constants should not be split simply because their Nodes have unequal
1422 functions or incongruent inputs. */
1423 if (type_is_neither_top_nor_const(y->type) &&
1424 (! is_Phi(y->node) || is_live_input(y->node, idx))) {
1425 add_to_touched(y, env);
1429 } /* collect_touched */
1432 * Collect commutative nodes to the touched list.
1434 * @param X the partition of the list
1435 * @param list the list which contains the nodes that must be evaluated
1436 * @param env the environment
1438 static void collect_commutative_touched(partition_t *X, list_head *list, environment_t *env) {
1443 list_for_each_entry(node_t, x, list, node_list) {
1446 num_edges = get_irn_n_outs(x->node);
1448 x->next_edge = x->n_followers + 1;
1450 /* for all edges in x.L.def_use_{idx} */
1451 while (x->next_edge <= num_edges) {
1452 const ir_def_use_edge *edge = &x->node->out[x->next_edge];
1455 /* check if we have necessary edges */
1465 /* only commutative nodes */
1466 if (!is_op_commutative(get_irn_op(succ)))
1469 y = get_irn_node(succ);
1470 if (is_constant_type(y->type)) {
1471 ir_opcode code = get_irn_opcode(succ);
1472 if (code == iro_Eor)
1473 add_to_cprop(y, env);
1476 /* Partitions of constants should not be split simply because their Nodes have unequal
1477 functions or incongruent inputs. */
1478 if (type_is_neither_top_nor_const(y->type)) {
1479 int other_idx = edge->pos ^ 1;
1480 node_t *other = get_irn_node(get_irn_n(succ, other_idx));
1481 int equal = X == other->part;
1484 * Note: op(a, a) is NOT congruent to op(a, b).
1485 * So, either all touch nodes must have both inputs congruent,
1486 * or not. We decide this by the first occurred node.
1492 if (both_input == equal)
1493 add_to_touched(y, env);
1497 } /* collect_commutative_touched */
1500 * Split the partitions if caused by the first entry on the worklist.
1502 * @param env the environment
1504 static void cause_splits(environment_t *env) {
1505 partition_t *X, *Z, *N;
1508 /* remove the first partition from the worklist */
1510 env->worklist = X->wl_next;
1513 dump_partition("Cause_split: ", X);
1515 if (env->commutative) {
1516 /* handle commutative nodes first */
1518 /* empty the touched set: already done, just clear the list */
1519 env->touched = NULL;
1521 collect_commutative_touched(X, &X->Leader, env);
1522 collect_commutative_touched(X, &X->Follower, env);
1524 for (Z = env->touched; Z != NULL; Z = N) {
1526 node_t *touched = Z->touched;
1527 unsigned n_touched = Z->n_touched;
1529 assert(Z->touched != NULL);
1531 /* beware, split might change Z */
1532 N = Z->touched_next;
1534 /* remove it from the touched set */
1537 /* Empty local Z.touched. */
1538 for (e = touched; e != NULL; e = e->next) {
1539 assert(e->is_follower == 0);
1545 if (0 < n_touched && n_touched < Z->n_leader) {
1546 DB((dbg, LEVEL_2, "Split part%d by touched\n", Z->nr));
1547 split(&Z, touched, env);
1549 assert(n_touched <= Z->n_leader);
1553 /* combine temporary leader and follower list */
1554 for (idx = -1; idx <= X->max_user_inputs; ++idx) {
1555 /* empty the touched set: already done, just clear the list */
1556 env->touched = NULL;
1558 collect_touched(&X->Leader, idx, env);
1559 collect_touched(&X->Follower, idx, env);
1561 for (Z = env->touched; Z != NULL; Z = N) {
1563 node_t *touched = Z->touched;
1564 unsigned n_touched = Z->n_touched;
1566 assert(Z->touched != NULL);
1568 /* beware, split might change Z */
1569 N = Z->touched_next;
1571 /* remove it from the touched set */
1574 /* Empty local Z.touched. */
1575 for (e = touched; e != NULL; e = e->next) {
1576 assert(e->is_follower == 0);
1582 if (0 < n_touched && n_touched < Z->n_leader) {
1583 DB((dbg, LEVEL_2, "Split part%d by touched\n", Z->nr));
1584 split(&Z, touched, env);
1586 assert(n_touched <= Z->n_leader);
1589 } /* cause_splits */
1592 * Implements split_by_what(): Split a partition by characteristics given
1593 * by the what function.
1595 * @param X the partition to split
1596 * @param What a function returning an Id for every node of the partition X
1597 * @param P a list to store the result partitions
1598 * @param env the environment
1602 static partition_t *split_by_what(partition_t *X, what_func What,
1603 partition_t **P, environment_t *env) {
1606 listmap_entry_t *iter;
1609 /* Let map be an empty mapping from the range of What to (local) list of Nodes. */
1611 list_for_each_entry(node_t, x, &X->Leader, node_list) {
1612 void *id = What(x, env);
1613 listmap_entry_t *entry;
1616 /* input not allowed, ignore */
1619 /* Add x to map[What(x)]. */
1620 entry = listmap_find(&map, id);
1621 x->next = entry->list;
1624 /* Let P be a set of Partitions. */
1626 /* for all sets S except one in the range of map do */
1627 for (iter = map.values; iter != NULL; iter = iter->next) {
1628 if (iter->next == NULL) {
1629 /* this is the last entry, ignore */
1634 /* Add SPLIT( X, S ) to P. */
1635 DB((dbg, LEVEL_2, "Split part%d by WHAT = %s\n", X->nr, what_reason));
1636 R = split(&X, S, env);
1646 } /* split_by_what */
1648 /** lambda n.(n.type) */
1649 static void *lambda_type(const node_t *node, environment_t *env) {
1651 return node->type.tv;
1654 /** lambda n.(n.opcode) */
1655 static void *lambda_opcode(const node_t *node, environment_t *env) {
1656 opcode_key_t key, *entry;
1657 ir_node *irn = node->node;
1659 key.code = get_irn_opcode(irn);
1660 key.mode = get_irn_mode(irn);
1661 key.arity = get_irn_arity(irn);
1665 switch (get_irn_opcode(irn)) {
1667 key.u.proj = get_Proj_proj(irn);
1670 key.u.ent = get_Sel_entity(irn);
1673 key.u.intVal = get_Conv_strict(irn);
1676 key.u.intVal = get_Div_no_remainder(irn);
1680 * Some ugliness here: Two Blocks having the same
1681 * IJmp predecessor would be congruent, which of course is wrong.
1682 * We fix it by never letting blocks be congruent
1683 * which cannot be detected by combo either.
1688 key.mode = get_Load_mode(irn);
1691 key.u.intVal = get_Builtin_kind(irn);
1697 entry = set_insert(env->opcode2id_map, &key, sizeof(key), opcode_hash(&key));
1699 } /* lambda_opcode */
1701 /** lambda n.(n[i].partition) */
1702 static void *lambda_partition(const node_t *node, environment_t *env) {
1703 ir_node *skipped = skip_Proj(node->node);
1706 int i = env->lambda_input;
1708 if (i >= get_irn_arity(node->node)) {
1710 * We are outside the allowed range: This can happen even
1711 * if we have split by opcode first: doing so might move Followers
1712 * to Leaders and those will have a different opcode!
1713 * Note that in this case the partition is on the cprop list and will be
1719 /* ignore the "control input" for non-pinned nodes
1720 if we are running in GCSE mode */
1721 if (i < env->end_idx && get_irn_pinned(skipped) != op_pin_state_pinned)
1724 pred = i == -1 ? get_irn_n(skipped, i) : get_irn_n(node->node, i);
1725 p = get_irn_node(pred);
1727 } /* lambda_partition */
1729 /** lambda n.(n[i].partition) for commutative nodes */
1730 static void *lambda_commutative_partition(const node_t *node, environment_t *env) {
1731 ir_node *irn = node->node;
1732 ir_node *skipped = skip_Proj(irn);
1733 ir_node *pred, *left, *right;
1735 partition_t *pl, *pr;
1736 int i = env->lambda_input;
1738 if (i >= get_irn_arity(node->node)) {
1740 * We are outside the allowed range: This can happen even
1741 * if we have split by opcode first: doing so might move Followers
1742 * to Leaders and those will have a different opcode!
1743 * Note that in this case the partition is on the cprop list and will be
1749 /* ignore the "control input" for non-pinned nodes
1750 if we are running in GCSE mode */
1751 if (i < env->end_idx && get_irn_pinned(skipped) != op_pin_state_pinned)
1755 pred = get_irn_n(skipped, i);
1756 p = get_irn_node(pred);
1760 if (is_op_commutative(get_irn_op(irn))) {
1761 /* normalize partition order by returning the "smaller" on input 0,
1762 the "bigger" on input 1. */
1763 left = get_binop_left(irn);
1764 pl = get_irn_node(left)->part;
1765 right = get_binop_right(irn);
1766 pr = get_irn_node(right)->part;
1769 return pl < pr ? pl : pr;
1771 return pl > pr ? pl : pr;
1773 /* a not split out Follower */
1774 pred = get_irn_n(irn, i);
1775 p = get_irn_node(pred);
1779 } /* lambda_commutative_partition */
1782 * Returns true if a type is a constant (and NOT Top
1785 static int is_con(const lattice_elem_t type) {
1786 /* be conservative */
1787 if (is_tarval(type.tv))
1788 return tarval_is_constant(type.tv);
1789 return is_entity(type.sym.entity_p);
1793 * Implements split_by().
1795 * @param X the partition to split
1796 * @param env the environment
1798 static void split_by(partition_t *X, environment_t *env) {
1799 partition_t *I, *P = NULL;
1802 dump_partition("split_by", X);
1804 if (X->n_leader == 1) {
1805 /* we have only one leader, no need to split, just check it's type */
1806 node_t *x = get_first_node(X);
1807 X->type_is_T_or_C = x->type.tv == tarval_top || is_con(x->type);
1811 DEBUG_ONLY(what_reason = "lambda n.(n.type)";)
1812 P = split_by_what(X, lambda_type, &P, env);
1815 /* adjust the type tags, we have split partitions by type */
1816 for (I = P; I != NULL; I = I->split_next) {
1817 node_t *x = get_first_node(I);
1818 I->type_is_T_or_C = x->type.tv == tarval_top || is_con(x->type);
1825 if (Y->n_leader > 1) {
1826 /* we do not want split the TOP or constant partitions */
1827 if (! Y->type_is_T_or_C) {
1828 partition_t *Q = NULL;
1830 DEBUG_ONLY(what_reason = "lambda n.(n.opcode)";)
1831 Q = split_by_what(Y, lambda_opcode, &Q, env);
1838 if (Z->n_leader > 1) {
1839 const node_t *first = get_first_node(Z);
1840 int arity = get_irn_arity(first->node);
1842 what_func what = lambda_partition;
1843 DEBUG_ONLY(char buf[64];)
1845 if (env->commutative && is_op_commutative(get_irn_op(first->node)))
1846 what = lambda_commutative_partition;
1849 * BEWARE: during splitting by input 2 for instance we might
1850 * create new partitions which are different by input 1, so collect
1851 * them and split further.
1853 Z->split_next = NULL;
1856 for (input = arity - 1; input >= -1; --input) {
1858 partition_t *Z_prime = R;
1861 if (Z_prime->n_leader > 1) {
1862 env->lambda_input = input;
1863 DEBUG_ONLY(snprintf(buf, sizeof(buf), "lambda n.(n[%d].partition)", input);)
1864 DEBUG_ONLY(what_reason = buf;)
1865 S = split_by_what(Z_prime, what, &S, env);
1868 Z_prime->split_next = S;
1871 } while (R != NULL);
1876 } while (Q != NULL);
1879 } while (P != NULL);
1883 * (Re-)compute the type for a given node.
1885 * @param node the node
1887 static void default_compute(node_t *node) {
1889 ir_node *irn = node->node;
1891 /* if any of the data inputs have type top, the result is type top */
1892 for (i = get_irn_arity(irn) - 1; i >= 0; --i) {
1893 ir_node *pred = get_irn_n(irn, i);
1894 node_t *p = get_irn_node(pred);
1896 if (p->type.tv == tarval_top) {
1897 node->type.tv = tarval_top;
1902 if (get_irn_mode(node->node) == mode_X)
1903 node->type.tv = tarval_reachable;
1905 node->type.tv = computed_value(irn);
1906 } /* default_compute */
1909 * (Re-)compute the type for a Block node.
1911 * @param node the node
1913 static void compute_Block(node_t *node) {
1915 ir_node *block = node->node;
1917 if (block == get_irg_start_block(current_ir_graph) || has_Block_label(block)) {
1918 /* start block and labelled blocks are always reachable */
1919 node->type.tv = tarval_reachable;
1923 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1924 node_t *pred = get_irn_node(get_Block_cfgpred(block, i));
1926 if (pred->type.tv == tarval_reachable) {
1927 /* A block is reachable, if at least of predecessor is reachable. */
1928 node->type.tv = tarval_reachable;
1932 node->type.tv = tarval_top;
1933 } /* compute_Block */
1936 * (Re-)compute the type for a Bad node.
1938 * @param node the node
1940 static void compute_Bad(node_t *node) {
1941 /* Bad nodes ALWAYS compute Top */
1942 node->type.tv = tarval_top;
1946 * (Re-)compute the type for an Unknown node.
1948 * @param node the node
1950 static void compute_Unknown(node_t *node) {
1951 /* While Unknown nodes should compute Top this is dangerous:
1952 * a Top input to a Cond would lead to BOTH control flows unreachable.
1953 * While this is correct in the given semantics, it would destroy the Firm
1956 * It would be safe to compute Top IF it can be assured, that only Cmp
1957 * nodes are inputs to Conds. We check that first.
1958 * This is the way Frontends typically build Firm, but some optimizations
1959 * (cond_eval for instance) might replace them by Phib's...
1961 node->type.tv = tarval_UNKNOWN;
1962 } /* compute_Unknown */
1965 * (Re-)compute the type for a Jmp node.
1967 * @param node the node
1969 static void compute_Jmp(node_t *node) {
1970 node_t *block = get_irn_node(get_nodes_block(node->node));
1972 node->type = block->type;
1976 * (Re-)compute the type for the Return node.
1978 * @param node the node
1980 static void compute_Return(node_t *node) {
1981 /* The Return node is NOT dead if it is in a reachable block.
1982 * This is already checked in compute(). so we can return
1983 * Reachable here. */
1984 node->type.tv = tarval_reachable;
1985 } /* compute_Return */
1988 * (Re-)compute the type for the End node.
1990 * @param node the node
1992 static void compute_End(node_t *node) {
1993 /* the End node is NOT dead of course */
1994 node->type.tv = tarval_reachable;
1998 * (Re-)compute the type for a Call.
2000 * @param node the node
2002 static void compute_Call(node_t *node) {
2004 * A Call computes always bottom, even if it has Unknown
2007 node->type.tv = tarval_bottom;
2008 } /* compute_Call */
2011 * (Re-)compute the type for a SymConst node.
2013 * @param node the node
2015 static void compute_SymConst(node_t *node) {
2016 ir_node *irn = node->node;
2017 node_t *block = get_irn_node(get_nodes_block(irn));
2019 if (block->type.tv == tarval_unreachable) {
2020 node->type.tv = tarval_top;
2023 switch (get_SymConst_kind(irn)) {
2024 case symconst_addr_ent:
2025 /* case symconst_addr_name: cannot handle this yet */
2026 node->type.sym = get_SymConst_symbol(irn);
2029 node->type.tv = computed_value(irn);
2031 } /* compute_SymConst */
2034 * (Re-)compute the type for a Phi node.
2036 * @param node the node
2038 static void compute_Phi(node_t *node) {
2040 ir_node *phi = node->node;
2041 lattice_elem_t type;
2043 /* if a Phi is in a unreachable block, its type is TOP */
2044 node_t *block = get_irn_node(get_nodes_block(phi));
2046 if (block->type.tv == tarval_unreachable) {
2047 node->type.tv = tarval_top;
2051 /* Phi implements the Meet operation */
2052 type.tv = tarval_top;
2053 for (i = get_Phi_n_preds(phi) - 1; i >= 0; --i) {
2054 node_t *pred = get_irn_node(get_Phi_pred(phi, i));
2055 node_t *pred_X = get_irn_node(get_Block_cfgpred(block->node, i));
2057 if (pred_X->type.tv == tarval_unreachable || pred->type.tv == tarval_top) {
2058 /* ignore TOP inputs: We must check here for unreachable blocks,
2059 because Firm constants live in the Start Block are NEVER Top.
2060 Else, a Phi (1,2) will produce Bottom, even if the 2 for instance
2061 comes from a unreachable input. */
2064 if (pred->type.tv == tarval_bottom) {
2065 node->type.tv = tarval_bottom;
2067 } else if (type.tv == tarval_top) {
2068 /* first constant found */
2070 } else if (type.tv != pred->type.tv) {
2071 /* different constants or tarval_bottom */
2072 node->type.tv = tarval_bottom;
2075 /* else nothing, constants are the same */
2081 * (Re-)compute the type for an Add. Special case: one nodes is a Zero Const.
2083 * @param node the node
2085 static void compute_Add(node_t *node) {
2086 ir_node *sub = node->node;
2087 node_t *l = get_irn_node(get_Add_left(sub));
2088 node_t *r = get_irn_node(get_Add_right(sub));
2089 lattice_elem_t a = l->type;
2090 lattice_elem_t b = r->type;
2093 if (a.tv == tarval_top || b.tv == tarval_top) {
2094 node->type.tv = tarval_top;
2095 } else if (a.tv == tarval_bottom || b.tv == tarval_bottom) {
2096 node->type.tv = tarval_bottom;
2098 /* x + 0 = 0 + x = x, but beware of floating point +0 + -0, so we
2099 must call tarval_add() first to handle this case! */
2100 if (is_tarval(a.tv)) {
2101 if (is_tarval(b.tv)) {
2102 node->type.tv = tarval_add(a.tv, b.tv);
2105 mode = get_tarval_mode(a.tv);
2106 if (a.tv == get_mode_null(mode)) {
2110 } else if (is_tarval(b.tv)) {
2111 mode = get_tarval_mode(b.tv);
2112 if (b.tv == get_mode_null(mode)) {
2117 node->type.tv = tarval_bottom;
2122 * (Re-)compute the type for a Sub. Special case: both nodes are congruent.
2124 * @param node the node
2126 static void compute_Sub(node_t *node) {
2127 ir_node *sub = node->node;
2128 node_t *l = get_irn_node(get_Sub_left(sub));
2129 node_t *r = get_irn_node(get_Sub_right(sub));
2130 lattice_elem_t a = l->type;
2131 lattice_elem_t b = r->type;
2134 if (a.tv == tarval_top || b.tv == tarval_top) {
2135 node->type.tv = tarval_top;
2136 } else if (is_con(a) && is_con(b)) {
2137 if (is_tarval(a.tv) && is_tarval(b.tv)) {
2138 node->type.tv = tarval_sub(a.tv, b.tv, get_irn_mode(sub));
2139 } else if (is_tarval(a.tv) && tarval_is_null(a.tv)) {
2141 } else if (is_tarval(b.tv) && tarval_is_null(b.tv)) {
2144 node->type.tv = tarval_bottom;
2146 } else if (r->part == l->part &&
2147 (!mode_is_float(get_irn_mode(l->node)))) {
2149 * BEWARE: a - a is NOT always 0 for floating Point values, as
2150 * NaN op NaN = NaN, so we must check this here.
2152 ir_mode *mode = get_irn_mode(sub);
2153 tv = get_mode_null(mode);
2155 /* if the node was ONCE evaluated by all constants, but now
2156 this breaks AND we get from the argument partitions a different
2157 result, switch to bottom.
2158 This happens because initially all nodes are in the same partition ... */
2159 if (node->type.tv != tv)
2163 node->type.tv = tarval_bottom;
2168 * (Re-)compute the type for an Eor. Special case: both nodes are congruent.
2170 * @param node the node
2172 static void compute_Eor(node_t *node) {
2173 ir_node *eor = node->node;
2174 node_t *l = get_irn_node(get_Eor_left(eor));
2175 node_t *r = get_irn_node(get_Eor_right(eor));
2176 lattice_elem_t a = l->type;
2177 lattice_elem_t b = r->type;
2180 if (a.tv == tarval_top || b.tv == tarval_top) {
2181 node->type.tv = tarval_top;
2182 } else if (is_con(a) && is_con(b)) {
2183 if (is_tarval(a.tv) && is_tarval(b.tv)) {
2184 node->type.tv = tarval_eor(a.tv, b.tv);
2185 } else if (is_tarval(a.tv) && tarval_is_null(a.tv)) {
2187 } else if (is_tarval(b.tv) && tarval_is_null(b.tv)) {
2190 node->type.tv = tarval_bottom;
2192 } else if (r->part == l->part) {
2193 ir_mode *mode = get_irn_mode(eor);
2194 tv = get_mode_null(mode);
2196 /* if the node was ONCE evaluated by all constants, but now
2197 this breaks AND we get from the argument partitions a different
2198 result, switch to bottom.
2199 This happens because initially all nodes are in the same partition ... */
2200 if (node->type.tv != tv)
2204 node->type.tv = tarval_bottom;
2209 * (Re-)compute the type for Cmp.
2211 * @param node the node
2213 static void compute_Cmp(node_t *node) {
2214 ir_node *cmp = node->node;
2215 node_t *l = get_irn_node(get_Cmp_left(cmp));
2216 node_t *r = get_irn_node(get_Cmp_right(cmp));
2217 lattice_elem_t a = l->type;
2218 lattice_elem_t b = r->type;
2220 if (a.tv == tarval_top || b.tv == tarval_top) {
2221 node->type.tv = tarval_top;
2222 } else if (r->part == l->part) {
2223 /* both nodes congruent, we can probably do something */
2224 node->type.tv = tarval_b_true;
2225 } else if (is_con(a) && is_con(b)) {
2226 /* both nodes are constants, we can probably do something */
2227 node->type.tv = tarval_b_true;
2229 node->type.tv = tarval_bottom;
2234 * (Re-)compute the type for a Proj(Cmp).
2236 * @param node the node
2237 * @param cond the predecessor Cmp node
2239 static void compute_Proj_Cmp(node_t *node, ir_node *cmp) {
2240 ir_node *proj = node->node;
2241 node_t *l = get_irn_node(get_Cmp_left(cmp));
2242 node_t *r = get_irn_node(get_Cmp_right(cmp));
2243 lattice_elem_t a = l->type;
2244 lattice_elem_t b = r->type;
2245 pn_Cmp pnc = get_Proj_proj(proj);
2248 if (a.tv == tarval_top || b.tv == tarval_top) {
2249 node->type.tv = tarval_undefined;
2250 } else if (is_con(a) && is_con(b)) {
2251 default_compute(node);
2252 } else if (r->part == l->part &&
2253 (!mode_is_float(get_irn_mode(l->node)) || pnc == pn_Cmp_Lt || pnc == pn_Cmp_Gt)) {
2255 * BEWARE: a == a is NOT always True for floating Point values, as
2256 * NaN != NaN is defined, so we must check this here.
2258 tv = pnc & pn_Cmp_Eq ? tarval_b_true: tarval_b_false;
2260 /* if the node was ONCE evaluated by all constants, but now
2261 this breaks AND we get from the argument partitions a different
2262 result, switch to bottom.
2263 This happens because initially all nodes are in the same partition ... */
2264 if (node->type.tv != tv)
2268 node->type.tv = tarval_bottom;
2270 } /* compute_Proj_Cmp */
2273 * (Re-)compute the type for a Proj(Cond).
2275 * @param node the node
2276 * @param cond the predecessor Cond node
2278 static void compute_Proj_Cond(node_t *node, ir_node *cond) {
2279 ir_node *proj = node->node;
2280 long pnc = get_Proj_proj(proj);
2281 ir_node *sel = get_Cond_selector(cond);
2282 node_t *selector = get_irn_node(sel);
2285 * Note: it is crucial for the monotony that the Proj(Cond)
2286 * are evaluates after all predecessors of the Cond selector are
2292 * Due to the fact that 0 is a const, the Cmp gets immediately
2293 * on the cprop list. It will be evaluated before x is evaluated,
2294 * might leaving x as Top. When later x is evaluated, the Cmp
2295 * might change its value.
2296 * BUT if the Cond is evaluated before this happens, Proj(Cond, FALSE)
2297 * gets R, and later changed to F if Cmp is evaluated to True!
2299 * We prevent this by putting Conds in an extra cprop_X queue, which
2300 * gets evaluated after the cprop queue is empty.
2302 * Note that this even happens with Click's original algorithm, if
2303 * Cmp(x, 0) is evaluated to True first and later changed to False
2304 * if x was Top first and later changed to a Const ...
2305 * It is unclear how Click solved that problem ...
2307 * However, in rare cases even this does not help, if a Top reaches
2308 * a compare through a Phi, than Proj(Cond) is evaluated changing
2309 * the type of the Phi to something other.
2310 * So, we take the last resort and bind the type to R once
2313 * (This might be even the way Click works around the whole problem).
2315 * Finally, we may miss some optimization possibilities due to this:
2320 * If Top reaches the if first, than we decide for != here.
2321 * If y later is evaluated to 0, we cannot revert this decision
2322 * and must live with both outputs enabled. If this happens,
2323 * we get an unresolved if (true) in the code ...
2325 * In Click's version where this decision is done at the Cmp,
2326 * the Cmp is NOT optimized away than (if y evaluated to 1
2327 * for instance) and we get a if (1 == 0) here ...
2329 * Both solutions are suboptimal.
2330 * At least, we could easily detect this problem and run
2331 * cf_opt() (or even combo) again :-(
2333 if (node->type.tv == tarval_reachable)
2336 if (get_irn_mode(sel) == mode_b) {
2338 if (pnc == pn_Cond_true) {
2339 if (selector->type.tv == tarval_b_false) {
2340 node->type.tv = tarval_unreachable;
2341 } else if (selector->type.tv == tarval_b_true) {
2342 node->type.tv = tarval_reachable;
2343 } else if (selector->type.tv == tarval_bottom) {
2344 node->type.tv = tarval_reachable;
2346 assert(selector->type.tv == tarval_top);
2347 if (tarval_UNKNOWN == tarval_top) {
2348 /* any condition based on Top is "!=" */
2349 node->type.tv = tarval_unreachable;
2351 node->type.tv = tarval_unreachable;
2355 assert(pnc == pn_Cond_false);
2357 if (selector->type.tv == tarval_b_false) {
2358 node->type.tv = tarval_reachable;
2359 } else if (selector->type.tv == tarval_b_true) {
2360 node->type.tv = tarval_unreachable;
2361 } else if (selector->type.tv == tarval_bottom) {
2362 node->type.tv = tarval_reachable;
2364 assert(selector->type.tv == tarval_top);
2365 if (tarval_UNKNOWN == tarval_top) {
2366 /* any condition based on Top is "!=" */
2367 node->type.tv = tarval_reachable;
2369 node->type.tv = tarval_unreachable;
2375 if (selector->type.tv == tarval_bottom) {
2376 node->type.tv = tarval_reachable;
2377 } else if (selector->type.tv == tarval_top) {
2378 if (tarval_UNKNOWN == tarval_top &&
2379 pnc == get_Cond_default_proj(cond)) {
2380 /* a switch based of Top is always "default" */
2381 node->type.tv = tarval_reachable;
2383 node->type.tv = tarval_unreachable;
2386 long value = get_tarval_long(selector->type.tv);
2387 if (pnc == get_Cond_default_proj(cond)) {
2388 /* default switch, have to check ALL other cases */
2391 for (i = get_irn_n_outs(cond) - 1; i >= 0; --i) {
2392 ir_node *succ = get_irn_out(cond, i);
2396 if (value == get_Proj_proj(succ)) {
2397 /* we found a match, will NOT take the default case */
2398 node->type.tv = tarval_unreachable;
2402 /* all cases checked, no match, will take default case */
2403 node->type.tv = tarval_reachable;
2406 node->type.tv = value == pnc ? tarval_reachable : tarval_unreachable;
2410 } /* compute_Proj_Cond */
2413 * (Re-)compute the type for a Proj-Node.
2415 * @param node the node
2417 static void compute_Proj(node_t *node) {
2418 ir_node *proj = node->node;
2419 ir_mode *mode = get_irn_mode(proj);
2420 node_t *block = get_irn_node(get_nodes_block(skip_Proj(proj)));
2421 ir_node *pred = get_Proj_pred(proj);
2423 if (block->type.tv == tarval_unreachable) {
2424 /* a Proj in a unreachable Block stay Top */
2425 node->type.tv = tarval_top;
2428 if (get_irn_node(pred)->type.tv == tarval_top && !is_Cond(pred)) {
2429 /* if the predecessor is Top, its Proj follow */
2430 node->type.tv = tarval_top;
2434 if (mode == mode_M) {
2435 /* mode M is always bottom */
2436 node->type.tv = tarval_bottom;
2439 if (mode != mode_X) {
2441 compute_Proj_Cmp(node, pred);
2443 default_compute(node);
2446 /* handle mode_X nodes */
2448 switch (get_irn_opcode(pred)) {
2450 /* the Proj_X from the Start is always reachable.
2451 However this is already handled at the top. */
2452 node->type.tv = tarval_reachable;
2455 compute_Proj_Cond(node, pred);
2458 default_compute(node);
2460 } /* compute_Proj */
2463 * (Re-)compute the type for a Confirm.
2465 * @param node the node
2467 static void compute_Confirm(node_t *node) {
2468 ir_node *confirm = node->node;
2469 node_t *pred = get_irn_node(get_Confirm_value(confirm));
2471 if (get_Confirm_cmp(confirm) == pn_Cmp_Eq) {
2472 node_t *bound = get_irn_node(get_Confirm_bound(confirm));
2474 if (is_con(bound->type)) {
2475 /* is equal to a constant */
2476 node->type = bound->type;
2480 /* a Confirm is a copy OR a Const */
2481 node->type = pred->type;
2482 } /* compute_Confirm */
2485 * (Re-)compute the type for a given node.
2487 * @param node the node
2489 static void compute(node_t *node) {
2490 ir_node *irn = node->node;
2493 #ifndef VERIFY_MONOTONE
2495 * Once a node reaches bottom, the type cannot fall further
2496 * in the lattice and we can stop computation.
2497 * Do not take this exit if the monotony verifier is
2498 * enabled to catch errors.
2500 if (node->type.tv == tarval_bottom)
2504 if (is_no_Block(irn)) {
2505 /* for pinned nodes, check its control input */
2506 if (get_irn_pinned(skip_Proj(irn)) == op_pin_state_pinned) {
2507 node_t *block = get_irn_node(get_nodes_block(irn));
2509 if (block->type.tv == tarval_unreachable) {
2510 node->type.tv = tarval_top;
2516 func = (compute_func)node->node->op->ops.generic;
2522 * Identity functions: Note that one might thing that identity() is just a
2523 * synonym for equivalent_node(). While this is true, we cannot use it for the algorithm
2524 * here, because it expects that the identity node is one of the inputs, which is NOT
2525 * always true for equivalent_node() which can handle (and does sometimes) DAGs.
2526 * So, we have our own implementation, which copies some parts of equivalent_node()
2530 * Calculates the Identity for Phi nodes
2532 static node_t *identity_Phi(node_t *node) {
2533 ir_node *phi = node->node;
2534 ir_node *block = get_nodes_block(phi);
2535 node_t *n_part = NULL;
2538 for (i = get_Phi_n_preds(phi) - 1; i >= 0; --i) {
2539 node_t *pred_X = get_irn_node(get_Block_cfgpred(block, i));
2541 if (pred_X->type.tv == tarval_reachable) {
2542 node_t *pred = get_irn_node(get_Phi_pred(phi, i));
2546 else if (n_part->part != pred->part) {
2547 /* incongruent inputs, not a follower */
2552 /* if n_part is NULL here, all inputs path are dead, the Phi computes
2553 * tarval_top, is in the TOP partition and should NOT being split! */
2554 assert(n_part != NULL);
2556 } /* identity_Phi */
2559 * Calculates the Identity for commutative 0 neutral nodes.
2561 static node_t *identity_comm_zero_binop(node_t *node) {
2562 ir_node *op = node->node;
2563 node_t *a = get_irn_node(get_binop_left(op));
2564 node_t *b = get_irn_node(get_binop_right(op));
2565 ir_mode *mode = get_irn_mode(op);
2568 /* for FP these optimizations are only allowed if fp_strict_algebraic is disabled */
2569 if (mode_is_float(mode) && (get_irg_fp_model(current_ir_graph) & fp_strict_algebraic))
2572 /* node: no input should be tarval_top, else the binop would be also
2573 * Top and not being split. */
2574 zero = get_mode_null(mode);
2575 if (a->type.tv == zero)
2577 if (b->type.tv == zero)
2580 } /* identity_comm_zero_binop */
2583 * Calculates the Identity for Shift nodes.
2585 static node_t *identity_shift(node_t *node) {
2586 ir_node *op = node->node;
2587 node_t *b = get_irn_node(get_binop_right(op));
2588 ir_mode *mode = get_irn_mode(b->node);
2591 /* node: no input should be tarval_top, else the binop would be also
2592 * Top and not being split. */
2593 zero = get_mode_null(mode);
2594 if (b->type.tv == zero)
2595 return get_irn_node(get_binop_left(op));
2597 } /* identity_shift */
2600 * Calculates the Identity for Mul nodes.
2602 static node_t *identity_Mul(node_t *node) {
2603 ir_node *op = node->node;
2604 node_t *a = get_irn_node(get_Mul_left(op));
2605 node_t *b = get_irn_node(get_Mul_right(op));
2606 ir_mode *mode = get_irn_mode(op);
2609 /* for FP these optimizations are only allowed if fp_strict_algebraic is disabled */
2610 if (mode_is_float(mode) && (get_irg_fp_model(current_ir_graph) & fp_strict_algebraic))
2613 /* node: no input should be tarval_top, else the binop would be also
2614 * Top and not being split. */
2615 one = get_mode_one(mode);
2616 if (a->type.tv == one)
2618 if (b->type.tv == one)
2621 } /* identity_Mul */
2624 * Calculates the Identity for Sub nodes.
2626 static node_t *identity_Sub(node_t *node) {
2627 ir_node *sub = node->node;
2628 node_t *b = get_irn_node(get_Sub_right(sub));
2629 ir_mode *mode = get_irn_mode(sub);
2631 /* for FP these optimizations are only allowed if fp_strict_algebraic is disabled */
2632 if (mode_is_float(mode) && (get_irg_fp_model(current_ir_graph) & fp_strict_algebraic))
2635 /* node: no input should be tarval_top, else the binop would be also
2636 * Top and not being split. */
2637 if (b->type.tv == get_mode_null(mode))
2638 return get_irn_node(get_Sub_left(sub));
2640 } /* identity_Sub */
2643 * Calculates the Identity for And nodes.
2645 static node_t *identity_And(node_t *node) {
2646 ir_node *and = node->node;
2647 node_t *a = get_irn_node(get_And_left(and));
2648 node_t *b = get_irn_node(get_And_right(and));
2649 tarval *neutral = get_mode_all_one(get_irn_mode(and));
2651 /* node: no input should be tarval_top, else the And would be also
2652 * Top and not being split. */
2653 if (a->type.tv == neutral)
2655 if (b->type.tv == neutral)
2658 } /* identity_And */
2661 * Calculates the Identity for Confirm nodes.
2663 static node_t *identity_Confirm(node_t *node) {
2664 ir_node *confirm = node->node;
2666 /* a Confirm is always a Copy */
2667 return get_irn_node(get_Confirm_value(confirm));
2668 } /* identity_Confirm */
2671 * Calculates the Identity for Mux nodes.
2673 static node_t *identity_Mux(node_t *node) {
2674 ir_node *mux = node->node;
2675 node_t *t = get_irn_node(get_Mux_true(mux));
2676 node_t *f = get_irn_node(get_Mux_false(mux));
2679 if (t->part == f->part)
2682 /* for now, the 1-input identity is not supported */
2684 sel = get_irn_node(get_Mux_sel(mux));
2686 /* Mux sel input is mode_b, so it is always a tarval */
2687 if (sel->type.tv == tarval_b_true)
2689 if (sel->type.tv == tarval_b_false)
2693 } /* identity_Mux */
2696 * Calculates the Identity for nodes.
2698 static node_t *identity(node_t *node) {
2699 ir_node *irn = node->node;
2701 switch (get_irn_opcode(irn)) {
2703 return identity_Phi(node);
2705 return identity_Mul(node);
2709 return identity_comm_zero_binop(node);
2714 return identity_shift(node);
2716 return identity_And(node);
2718 return identity_Sub(node);
2720 return identity_Confirm(node);
2722 return identity_Mux(node);
2729 * Node follower is a (new) follower of leader, segregate Leader
2732 static void segregate_def_use_chain_1(const ir_node *follower, node_t *leader) {
2733 ir_node *l = leader->node;
2734 int j, i, n = get_irn_n_outs(l);
2736 DB((dbg, LEVEL_2, "%+F is a follower of %+F\n", follower, leader->node));
2737 /* The leader edges must remain sorted, but follower edges can
2739 for (i = leader->n_followers + 1; i <= n; ++i) {
2740 if (l->out[i].use == follower) {
2741 ir_def_use_edge t = l->out[i];
2743 for (j = i - 1; j >= leader->n_followers + 1; --j)
2744 l->out[j + 1] = l->out[j];
2745 ++leader->n_followers;
2746 l->out[leader->n_followers] = t;
2750 } /* segregate_def_use_chain_1 */
2753 * Node follower is a (new) follower segregate its Leader
2756 * @param follower the follower IR node
2758 static void segregate_def_use_chain(const ir_node *follower) {
2761 for (i = get_irn_arity(follower) - 1; i >= 0; --i) {
2762 node_t *pred = get_irn_node(get_irn_n(follower, i));
2764 segregate_def_use_chain_1(follower, pred);
2766 } /* segregate_def_use_chain */
2769 * Propagate constant evaluation.
2771 * @param env the environment
2773 static void propagate(environment_t *env) {
2776 lattice_elem_t old_type;
2778 unsigned n_fallen, old_type_was_T_or_C;
2781 while (env->cprop != NULL) {
2782 void *oldopcode = NULL;
2784 /* remove the first partition X from cprop */
2787 env->cprop = X->cprop_next;
2789 old_type_was_T_or_C = X->type_is_T_or_C;
2791 DB((dbg, LEVEL_2, "Propagate type on part%d\n", X->nr));
2795 int cprop_empty = list_empty(&X->cprop);
2796 int cprop_X_empty = list_empty(&X->cprop_X);
2798 if (cprop_empty && cprop_X_empty) {
2799 /* both cprop lists are empty */
2803 /* remove the first Node x from X.cprop */
2805 /* Get a node from the cprop_X list only if
2806 * all data nodes are processed.
2807 * This ensures, that all inputs of the Cond
2808 * predecessor are processed if its type is still Top.
2810 x = list_entry(X->cprop_X.next, node_t, cprop_list);
2812 x = list_entry(X->cprop.next, node_t, cprop_list);
2815 //assert(x->part == X);
2816 list_del(&x->cprop_list);
2819 if (x->is_follower && identity(x) == x) {
2820 /* check the opcode first */
2821 if (oldopcode == NULL) {
2822 oldopcode = lambda_opcode(get_first_node(X), env);
2824 if (oldopcode != lambda_opcode(x, env)) {
2825 if (x->on_fallen == 0) {
2826 /* different opcode -> x falls out of this partition */
2831 DB((dbg, LEVEL_2, "Add node %+F to fallen\n", x->node));
2835 /* x will make the follower -> leader transition */
2836 follower_to_leader(x);
2839 /* compute a new type for x */
2841 DB((dbg, LEVEL_3, "computing type of %+F\n", x->node));
2843 if (x->type.tv != old_type.tv) {
2844 DB((dbg, LEVEL_2, "node %+F has changed type from %+F to %+F\n", x->node, old_type, x->type));
2845 verify_type(old_type, x);
2847 if (x->on_fallen == 0) {
2848 /* Add x to fallen. Nodes might fall from T -> const -> _|_, so check that they are
2849 not already on the list. */
2854 DB((dbg, LEVEL_2, "Add node %+F to fallen\n", x->node));
2856 for (i = get_irn_n_outs(x->node) - 1; i >= 0; --i) {
2857 ir_node *succ = get_irn_out(x->node, i);
2858 node_t *y = get_irn_node(succ);
2860 /* Add y to y.partition.cprop. */
2861 add_to_cprop(y, env);
2866 if (n_fallen > 0 && n_fallen != X->n_leader) {
2867 DB((dbg, LEVEL_2, "Splitting part%d by fallen\n", X->nr));
2868 Y = split(&X, fallen, env);
2870 * We have split out fallen node. The type of the result
2871 * partition is NOT set yet.
2873 Y->type_is_T_or_C = 0;
2877 /* remove the flags from the fallen list */
2878 for (x = fallen; x != NULL; x = x->next)
2881 if (old_type_was_T_or_C) {
2884 /* check if some nodes will make the leader -> follower transition */
2885 list_for_each_entry_safe(node_t, y, tmp, &Y->Leader, node_list) {
2886 if (y->type.tv != tarval_top && ! is_con(y->type)) {
2887 node_t *eq_node = identity(y);
2889 if (eq_node != y && eq_node->part == y->part) {
2890 DB((dbg, LEVEL_2, "Node %+F is a follower of %+F\n", y->node, eq_node->node));
2891 /* move to Follower */
2893 list_del(&y->node_list);
2894 list_add_tail(&y->node_list, &Y->Follower);
2897 segregate_def_use_chain(y->node);
2907 * Get the leader for a given node from its congruence class.
2909 * @param irn the node
2911 static ir_node *get_leader(node_t *node) {
2912 partition_t *part = node->part;
2914 if (part->n_leader > 1 || node->is_follower) {
2915 if (node->is_follower) {
2916 DB((dbg, LEVEL_2, "Replacing follower %+F\n", node->node));
2919 DB((dbg, LEVEL_2, "Found congruence class for %+F\n", node->node));
2921 return get_first_node(part)->node;
2927 * Returns non-zero if a mode_T node has only one reachable output.
2929 static int only_one_reachable_proj(ir_node *n) {
2932 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2933 ir_node *proj = get_irn_out(n, i);
2936 /* skip non-control flow Proj's */
2937 if (get_irn_mode(proj) != mode_X)
2940 node = get_irn_node(proj);
2941 if (node->type.tv == tarval_reachable) {
2947 } /* only_one_reachable_proj */
2950 * Return non-zero if the control flow predecessor node pred
2951 * is the only reachable control flow exit of its block.
2953 * @param pred the control flow exit
2954 * @param block the destination block
2956 static int can_exchange(ir_node *pred, ir_node *block) {
2957 if (is_Start(pred) || has_Block_label(block))
2959 else if (is_Jmp(pred))
2961 else if (get_irn_mode(pred) == mode_T) {
2962 /* if the predecessor block has more than one
2963 reachable outputs we cannot remove the block */
2964 return only_one_reachable_proj(pred);
2967 } /* can_exchange */
2970 * Block Post-Walker, apply the analysis results on control flow by
2971 * shortening Phi's and Block inputs.
2973 static void apply_cf(ir_node *block, void *ctx) {
2974 environment_t *env = ctx;
2975 node_t *node = get_irn_node(block);
2977 ir_node **ins, **in_X;
2978 ir_node *phi, *next;
2980 n = get_Block_n_cfgpreds(block);
2982 if (node->type.tv == tarval_unreachable) {
2985 for (i = n - 1; i >= 0; --i) {
2986 ir_node *pred = get_Block_cfgpred(block, i);
2988 if (! is_Bad(pred)) {
2989 node_t *pred_bl = get_irn_node(get_nodes_block(skip_Proj(pred)));
2991 if (pred_bl->flagged == 0) {
2992 pred_bl->flagged = 3;
2994 if (pred_bl->type.tv == tarval_reachable) {
2996 * We will remove an edge from block to its pred.
2997 * This might leave the pred block as an endless loop
2999 if (! is_backedge(block, i))
3000 keep_alive(pred_bl->node);
3006 /* the EndBlock is always reachable even if the analysis
3007 finds out the opposite :-) */
3008 if (block != get_irg_end_block(current_ir_graph)) {
3009 /* mark dead blocks */
3010 set_Block_dead(block);
3011 DB((dbg, LEVEL_1, "Removing dead %+F\n", block));
3013 /* the endblock is unreachable */
3014 set_irn_in(block, 0, NULL);
3020 /* only one predecessor combine */
3021 ir_node *pred = skip_Proj(get_Block_cfgpred(block, 0));
3023 if (can_exchange(pred, block)) {
3024 ir_node *new_block = get_nodes_block(pred);
3025 DB((dbg, LEVEL_1, "Fuse %+F with %+F\n", block, new_block));
3026 DBG_OPT_COMBO(block, new_block, FS_OPT_COMBO_CF);
3027 exchange(block, new_block);
3028 node->node = new_block;
3034 NEW_ARR_A(ir_node *, in_X, n);
3036 for (i = 0; i < n; ++i) {
3037 ir_node *pred = get_Block_cfgpred(block, i);
3038 node_t *node = get_irn_node(pred);
3040 if (node->type.tv == tarval_reachable) {
3043 DB((dbg, LEVEL_1, "Removing dead input %d from %+F (%+F)\n", i, block, pred));
3044 if (! is_Bad(pred)) {
3045 node_t *pred_bl = get_irn_node(get_nodes_block(skip_Proj(pred)));
3047 if (pred_bl->flagged == 0) {
3048 pred_bl->flagged = 3;
3050 if (pred_bl->type.tv == tarval_reachable) {
3052 * We will remove an edge from block to its pred.
3053 * This might leave the pred block as an endless loop
3055 if (! is_backedge(block, i))
3056 keep_alive(pred_bl->node);
3066 NEW_ARR_A(ir_node *, ins, n);
3067 for (phi = get_Block_phis(block); phi != NULL; phi = next) {
3068 node_t *node = get_irn_node(phi);
3070 next = get_Phi_next(phi);
3071 if (is_tarval(node->type.tv) && tarval_is_constant(node->type.tv)) {
3072 /* this Phi is replaced by a constant */
3073 tarval *tv = node->type.tv;
3074 ir_node *c = new_Const(tv);
3076 set_irn_node(c, node);
3078 DB((dbg, LEVEL_1, "%+F is replaced by %+F\n", phi, c));
3079 DBG_OPT_COMBO(phi, c, FS_OPT_COMBO_CONST);
3084 for (i = 0; i < n; ++i) {
3085 node_t *pred = get_irn_node(get_Block_cfgpred(block, i));
3087 if (pred->type.tv == tarval_reachable) {
3088 ins[j++] = get_Phi_pred(phi, i);
3092 /* this Phi is replaced by a single predecessor */
3093 ir_node *s = ins[0];
3094 node_t *phi_node = get_irn_node(phi);
3097 DB((dbg, LEVEL_1, "%+F is replaced by %+F because of cf change\n", phi, s));
3098 DBG_OPT_COMBO(phi, s, FS_OPT_COMBO_FOLLOWER);
3103 set_irn_in(phi, j, ins);
3111 /* this Block has only one live predecessor */
3112 ir_node *pred = skip_Proj(in_X[0]);
3114 if (can_exchange(pred, block)) {
3115 ir_node *new_block = get_nodes_block(pred);
3116 DBG_OPT_COMBO(block, new_block, FS_OPT_COMBO_CF);
3117 exchange(block, new_block);
3118 node->node = new_block;
3123 set_irn_in(block, k, in_X);
3128 * Exchange a node by its leader.
3129 * Beware: in rare cases the mode might be wrong here, for instance
3130 * AddP(x, NULL) is a follower of x, but with different mode.
3133 static void exchange_leader(ir_node *irn, ir_node *leader) {
3134 ir_mode *mode = get_irn_mode(irn);
3135 if (mode != get_irn_mode(leader)) {
3136 /* The conv is a no-op, so we are free to place it
3137 * either in the block of the leader OR in irn's block.
3138 * Probably placing it into leaders block might reduce
3139 * the number of Conv due to CSE. */
3140 ir_node *block = get_nodes_block(leader);
3141 dbg_info *dbg = get_irn_dbg_info(irn);
3143 leader = new_rd_Conv(dbg, current_ir_graph, block, leader, mode);
3145 exchange(irn, leader);
3146 } /* exchange_leader */
3149 * Check, if all users of a mode_M node are dead. Use
3150 * the Def-Use edges for this purpose, as they still
3151 * reflect the situation.
3153 static int all_users_are_dead(const ir_node *irn) {
3154 int i, n = get_irn_n_outs(irn);
3156 for (i = 1; i <= n; ++i) {
3157 const ir_node *succ = irn->out[i].use;
3158 const node_t *block = get_irn_node(get_nodes_block(succ));
3161 if (block->type.tv == tarval_unreachable) {
3162 /* block is unreachable */
3165 node = get_irn_node(succ);
3166 if (node->type.tv != tarval_top) {
3167 /* found a reachable user */
3171 /* all users are unreachable */
3173 } /* all_user_are_dead */
3176 * Walker: Find reachable mode_M nodes that have only
3177 * unreachable users. These nodes must be kept later.
3179 static void find_kept_memory(ir_node *irn, void *ctx) {
3180 environment_t *env = ctx;
3181 node_t *node, *block;
3183 if (get_irn_mode(irn) != mode_M)
3186 block = get_irn_node(get_nodes_block(irn));
3187 if (block->type.tv == tarval_unreachable)
3190 node = get_irn_node(irn);
3191 if (node->type.tv == tarval_top)
3194 /* ok, we found a live memory node. */
3195 if (all_users_are_dead(irn)) {
3196 DB((dbg, LEVEL_1, "%+F must be kept\n", irn));
3197 ARR_APP1(ir_node *, env->kept_memory, irn);
3199 } /* find_kept_memory */
3202 * Post-Walker, apply the analysis results;
3204 static void apply_result(ir_node *irn, void *ctx) {
3205 environment_t *env = ctx;
3206 node_t *node = get_irn_node(irn);
3208 if (is_Block(irn) || is_End(irn) || is_Bad(irn)) {
3209 /* blocks already handled, do not touch the End node */
3211 node_t *block = get_irn_node(get_nodes_block(irn));
3213 if (block->type.tv == tarval_unreachable) {
3214 ir_node *bad = get_irg_bad(current_ir_graph);
3216 /* here, bad might already have a node, but this can be safely ignored
3217 as long as bad has at least ONE valid node */
3218 set_irn_node(bad, node);
3220 DB((dbg, LEVEL_1, "%+F is unreachable\n", irn));
3223 } else if (node->type.tv == tarval_top) {
3224 ir_mode *mode = get_irn_mode(irn);
3226 if (mode == mode_M) {
3227 /* never kill a mode_M node */
3229 ir_node *pred = get_Proj_pred(irn);
3230 node_t *pnode = get_irn_node(pred);
3232 if (pnode->type.tv == tarval_top) {
3233 /* skip the predecessor */
3234 ir_node *mem = get_memop_mem(pred);
3236 DB((dbg, LEVEL_1, "%+F computes Top, replaced by %+F\n", irn, mem));
3241 /* leave other nodes, especially PhiM */
3242 } else if (mode == mode_T) {
3243 /* Do not kill mode_T nodes, kill their Projs */
3244 } else if (! is_Unknown(irn)) {
3245 /* don't kick away Unknown's, they might be still needed */
3246 ir_node *unk = new_r_Unknown(current_ir_graph, mode);
3248 /* control flow should already be handled at apply_cf() */
3249 assert(mode != mode_X);
3251 /* see comment above */
3252 set_irn_node(unk, node);
3254 DB((dbg, LEVEL_1, "%+F computes Top\n", irn));
3259 else if (get_irn_mode(irn) == mode_X) {
3262 ir_node *cond = get_Proj_pred(irn);
3264 if (is_Cond(cond)) {
3265 if (only_one_reachable_proj(cond)) {
3266 ir_node *jmp = new_r_Jmp(current_ir_graph, block->node);
3267 set_irn_node(jmp, node);
3269 DB((dbg, LEVEL_1, "%+F is replaced by %+F\n", irn, jmp));
3270 DBG_OPT_COMBO(irn, jmp, FS_OPT_COMBO_CF);
3274 node_t *sel = get_irn_node(get_Cond_selector(cond));
3275 tarval *tv = sel->type.tv;
3277 if (is_tarval(tv) && tarval_is_constant(tv)) {
3278 /* The selector is a constant, but more
3279 * than one output is active: An unoptimized
3287 /* normal data node */
3288 if (is_tarval(node->type.tv) && tarval_is_constant(node->type.tv)) {
3289 tarval *tv = node->type.tv;
3292 * Beware: never replace mode_T nodes by constants. Currently we must mark
3293 * mode_T nodes with constants, but do NOT replace them.
3295 if (! is_Const(irn) && get_irn_mode(irn) != mode_T) {
3296 /* can be replaced by a constant */
3297 ir_node *c = new_Const(tv);
3298 set_irn_node(c, node);
3300 DB((dbg, LEVEL_1, "%+F is replaced by %+F\n", irn, c));
3301 DBG_OPT_COMBO(irn, c, FS_OPT_COMBO_CONST);
3302 exchange_leader(irn, c);
3305 } else if (is_entity(node->type.sym.entity_p)) {
3306 if (! is_SymConst(irn)) {
3307 /* can be replaced by a SymConst */
3308 ir_node *symc = new_r_SymConst(current_ir_graph, block->node, get_irn_mode(irn), node->type.sym, symconst_addr_ent);
3309 set_irn_node(symc, node);
3312 DB((dbg, LEVEL_1, "%+F is replaced by %+F\n", irn, symc));
3313 DBG_OPT_COMBO(irn, symc, FS_OPT_COMBO_CONST);
3314 exchange_leader(irn, symc);
3317 } else if (is_Confirm(irn)) {
3318 /* Confirms are always follower, but do not kill them here */
3320 ir_node *leader = get_leader(node);
3322 if (leader != irn) {
3323 int non_strict_phi = 0;
3326 * Beware: Do not remove Phi(Unknown, ..., x, ..., Unknown)
3327 * as this might create non-strict programs.
3329 if (node->is_follower && is_Phi(irn) && !is_Unknown(leader)) {
3332 for (i = get_Phi_n_preds(irn) - 1; i >= 0; --i) {
3333 ir_node *pred = get_Phi_pred(irn, i);
3335 if (is_Unknown(pred)) {
3341 if (! non_strict_phi) {
3342 DB((dbg, LEVEL_1, "%+F from part%d is replaced by %+F\n", irn, node->part->nr, leader));
3343 if (node->is_follower)
3344 DBG_OPT_COMBO(irn, leader, FS_OPT_COMBO_FOLLOWER);
3346 DBG_OPT_COMBO(irn, leader, FS_OPT_COMBO_CONGRUENT);
3347 exchange_leader(irn, leader);
3354 } /* apply_result */
3357 * Fix the keep-alives by deleting unreachable ones.
3359 static void apply_end(ir_node *end, environment_t *env) {
3360 int i, j, n = get_End_n_keepalives(end);
3364 NEW_ARR_A(ir_node *, in, n);
3366 /* fix the keep alive */
3367 for (i = j = 0; i < n; i++) {
3368 ir_node *ka = get_End_keepalive(end, i);
3369 node_t *node = get_irn_node(ka);
3372 node = get_irn_node(get_nodes_block(ka));
3374 if (node->type.tv != tarval_unreachable && !is_Bad(ka))
3378 set_End_keepalives(end, j, in);
3383 #define SET(code) op_##code->ops.generic = (op_func)compute_##code
3386 * sets the generic functions to compute.
3388 static void set_compute_functions(void) {
3391 /* set the default compute function */
3392 for (i = get_irp_n_opcodes() - 1; i >= 0; --i) {
3393 ir_op *op = get_irp_opcode(i);
3394 op->ops.generic = (op_func)default_compute;
3397 /* set specific functions */
3413 } /* set_compute_functions */
3418 static void add_memory_keeps(ir_node **kept_memory, int len) {
3419 ir_node *end = get_irg_end(current_ir_graph);
3423 ir_nodeset_init(&set);
3425 /* check, if those nodes are already kept */
3426 for (i = get_End_n_keepalives(end) - 1; i >= 0; --i)
3427 ir_nodeset_insert(&set, get_End_keepalive(end, i));
3429 for (i = len - 1; i >= 0; --i) {
3430 ir_node *ka = kept_memory[i];
3432 if (! ir_nodeset_contains(&set, ka)) {
3433 add_End_keepalive(end, ka);
3436 ir_nodeset_destroy(&set);
3437 } /* add_memory_keeps */
3439 void combo(ir_graph *irg) {
3441 ir_node *initial_bl;
3443 ir_graph *rem = current_ir_graph;
3446 current_ir_graph = irg;
3448 /* register a debug mask */
3449 FIRM_DBG_REGISTER(dbg, "firm.opt.combo");
3451 DB((dbg, LEVEL_1, "Doing COMBO for %+F\n", irg));
3453 obstack_init(&env.obst);
3454 env.worklist = NULL;
3458 #ifdef DEBUG_libfirm
3459 env.dbg_list = NULL;
3461 env.opcode2id_map = new_set(cmp_opcode, iro_Last * 4);
3462 env.type2id_map = pmap_create();
3463 env.kept_memory = NEW_ARR_F(ir_node *, 0);
3464 env.end_idx = get_opt_global_cse() ? 0 : -1;
3465 env.lambda_input = 0;
3468 /* options driving the optimization */
3469 env.commutative = 1;
3470 env.opt_unknown = 1;
3472 assure_irg_outs(irg);
3473 assure_cf_loop(irg);
3475 /* we have our own value_of function */
3476 set_value_of_func(get_node_tarval);
3478 set_compute_functions();
3479 DEBUG_ONLY(part_nr = 0);
3481 ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK | IR_RESOURCE_PHI_LIST);
3483 if (env.opt_unknown)
3484 tarval_UNKNOWN = tarval_top;
3486 tarval_UNKNOWN = tarval_bad;
3488 /* create the initial partition and place it on the work list */
3489 env.initial = new_partition(&env);
3490 add_to_worklist(env.initial, &env);
3491 irg_walk_graph(irg, create_initial_partitions, init_block_phis, &env);
3493 /* set the hook: from now, every node has a partition and a type */
3494 DEBUG_ONLY(set_dump_node_vcgattr_hook(dump_partition_hook));
3496 /* all nodes on the initial partition have type Top */
3497 env.initial->type_is_T_or_C = 1;
3499 /* Place the START Node's partition on cprop.
3500 Place the START Node on its local worklist. */
3501 initial_bl = get_irg_start_block(irg);
3502 start = get_irn_node(initial_bl);
3503 add_to_cprop(start, &env);
3507 if (env.worklist != NULL)
3509 } while (env.cprop != NULL || env.worklist != NULL);
3511 dump_all_partitions(&env);
3512 check_all_partitions(&env);
3515 dump_ir_block_graph(irg, "-partition");
3518 /* apply the result */
3520 /* check, which nodes must be kept */
3521 irg_walk_graph(irg, NULL, find_kept_memory, &env);
3523 /* kill unreachable control flow */
3524 irg_block_walk_graph(irg, NULL, apply_cf, &env);
3525 /* Kill keep-alives of dead blocks: this speeds up apply_result()
3526 * and fixes assertion because dead cf to dead blocks is NOT removed by
3528 apply_end(get_irg_end(irg), &env);
3529 irg_walk_graph(irg, NULL, apply_result, &env);
3531 len = ARR_LEN(env.kept_memory);
3533 add_memory_keeps(env.kept_memory, len);
3536 DB((dbg, LEVEL_1, "Unoptimized Control Flow left"));
3540 /* control flow might changed */
3541 set_irg_outs_inconsistent(irg);
3542 set_irg_extblk_inconsistent(irg);
3543 set_irg_doms_inconsistent(irg);
3544 set_irg_loopinfo_inconsistent(irg);
3545 set_irg_entity_usage_state(irg, ir_entity_usage_not_computed);
3548 ir_free_resources(irg, IR_RESOURCE_IRN_LINK | IR_RESOURCE_PHI_LIST);
3550 /* remove the partition hook */
3551 DEBUG_ONLY(set_dump_node_vcgattr_hook(NULL));
3553 DEL_ARR_F(env.kept_memory);
3554 pmap_destroy(env.type2id_map);
3555 del_set(env.opcode2id_map);
3556 obstack_free(&env.obst, NULL);
3558 /* restore value_of() default behavior */
3559 set_value_of_func(NULL);
3560 current_ir_graph = rem;