2 * Copyright (C) 1995-2011 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 Dataflow driven Load/Store optimizations, uses some ideas from
24 * @author Michael Beck
40 #include "iroptimize.h"
41 #include "irnodemap.h"
42 #include "raw_bitset.h"
47 /* maximum number of output Proj's */
48 #define MAX_PROJ ((long)pn_Load_max > (long)pn_Store_max ? (long)pn_Load_max : (long)pn_Store_max)
51 * Mapping an address to an dense ID.
53 typedef struct address_entry_t {
54 unsigned id; /**< The ID */
61 FLAG_KILL_ALL = 1, /**< KILL all addresses */
62 FLAG_KILLED_NODE = 2, /**< this node was killed */
63 FLAG_EXCEPTION = 4, /**< this node has exception flow */
64 FLAG_IGNORE = 8, /**< ignore this node (volatile or other) */
68 * A value: This represents a value stored at a given address in
69 * memory. Do not confuse with values from value numbering.
71 typedef struct value_t value_t;
73 ir_node *address; /**< the address of this value */
74 ir_node *value; /**< the value itself */
75 ir_mode *mode; /**< the mode of the value */
76 unsigned id; /**< address id */
80 * A memop describes an memory-related operation.
81 * These are Loads/Store and all other ops that might modify
82 * memory (Calls, CopyB) or causing exceptions.
84 typedef struct memop_t memop_t;
86 value_t value; /**< the value of this memop: only defined for Load/Store */
87 ir_node *node; /**< the memory op itself */
88 ir_node *mem; /**< the memory FROM this node */
89 ir_node *replace; /**< the replacement node if this memop is replaced */
90 memop_t *next; /**< links to the next memory op in the block in forward order. */
91 memop_t *prev; /**< links to the previous memory op in the block in forward order. */
92 unsigned flags; /**< memop flags */
93 ir_node *projs[MAX_PROJ]; /**< Projs of this memory op */
97 * Additional data for every basic block.
99 typedef struct block_t block_t;
101 memop_t *memop_forward; /**< topologically sorted list of memory ops in this block */
102 memop_t *memop_backward; /**< last memop in the list */
103 unsigned *avail_out; /**< out-set of available addresses */
104 memop_t **id_2_memop_avail; /**< maps avail address ids to memops */
105 unsigned *anticL_in; /**< in-set of anticipated Load addresses */
106 memop_t **id_2_memop_antic; /**< maps anticipated address ids to memops */
107 ir_node *block; /**< the associated block */
108 block_t *forward_next; /**< next block entry for forward iteration */
109 block_t *backward_next; /**< next block entry for backward iteration */
110 memop_t *avail; /**< used locally for the avail map */
111 memop_t **trans_results; /**< used to cached translated nodes due antic calculation. */
115 * Metadata for this pass.
117 typedef struct ldst_env_t {
118 struct obstack obst; /**< obstack for temporary data */
119 ir_nodemap_t adr_map; /**< Map addresses to */
120 block_t *forward; /**< Inverse post-order list of all blocks Start->End */
121 block_t *backward; /**< Inverse post-order list of all blocks End->Start */
122 ir_node *start_bl; /**< start block of the current graph */
123 ir_node *end_bl; /**< end block of the current graph */
124 unsigned *curr_set; /**< current set of addresses */
125 memop_t **curr_id_2_memop; /**< current map of address ids to memops */
126 unsigned curr_adr_id; /**< number for address mapping */
127 unsigned n_mem_ops; /**< number of memory operations (Loads/Stores) */
128 size_t rbs_size; /**< size of all bitsets in bytes */
129 int max_cfg_preds; /**< maximum number of block cfg predecessors */
130 int changed; /**< Flags for changed graph state */
132 ir_node **id_2_address; /**< maps an id to the used address */
136 /* the one and only environment */
141 static firm_dbg_module_t *dbg;
144 * Dumps the block list.
146 * @param ldst environment
148 static void dump_block_list(ldst_env *env)
154 for (entry = env->forward; entry != NULL; entry = entry->forward_next) {
155 DB((dbg, LEVEL_2, "%+F {", entry->block));
158 for (op = entry->memop_forward; op != NULL; op = op->next) {
160 DB((dbg, LEVEL_2, "\n\t"));
162 DB((dbg, LEVEL_2, "%+F", op->node));
163 if ((op->flags & FLAG_KILL_ALL) == FLAG_KILL_ALL)
164 DB((dbg, LEVEL_2, "X"));
165 else if (op->flags & FLAG_KILL_ALL)
166 DB((dbg, LEVEL_2, "K"));
167 DB((dbg, LEVEL_2, ", "));
171 DB((dbg, LEVEL_2, "\n}\n\n"));
173 } /* dump_block_list */
176 * Dumps the current set.
178 * @param bl current block
179 * @param s name of the set
181 static void dump_curr(block_t *bl, const char *s)
183 size_t end = env.rbs_size - 1;
187 DB((dbg, LEVEL_2, "%s[%+F] = {", s, bl->block));
189 for (pos = rbitset_next(env.curr_set, 0, 1); pos < end; pos = rbitset_next(env.curr_set, pos + 1, 1)) {
190 memop_t *op = env.curr_id_2_memop[pos];
193 DB((dbg, LEVEL_2, "\n\t"));
196 DB((dbg, LEVEL_2, "<%+F, %+F>, ", op->value.address, op->value.value));
199 DB((dbg, LEVEL_2, "\n}\n"));
203 static void dump_block_list(ldst_env *env)
207 static void dump_curr(block_t *bl, const char *s)
212 #endif /* DEBUG_libfirm */
214 /** Get the block entry for a block node */
215 static block_t *get_block_entry(const ir_node *block)
217 assert(is_Block(block));
219 return (block_t*)get_irn_link(block);
220 } /* get_block_entry */
222 /** Get the memop entry for a memory operation node */
223 static memop_t *get_irn_memop(const ir_node *irn)
225 assert(! is_Block(irn));
226 return (memop_t*)get_irn_link(irn);
227 } /* get_irn_memop */
230 * Walk over the memory edges from definition to users.
231 * This ensures, that even operation without memory output are found.
233 * @param irn start node
234 * @param pre pre walker function
235 * @param post post walker function
236 * @param ctx context parameter for the walker functions
238 static void walk_memory(ir_node *irn, irg_walk_func *pre, irg_walk_func *post, void *ctx)
243 mark_irn_visited(irn);
248 mode = get_irn_mode(irn);
249 if (mode == mode_M) {
250 /* every successor uses memory */
251 for (i = get_irn_n_outs(irn) - 1; i >= 0; --i) {
252 ir_node *succ = get_irn_out(irn, i);
254 if (! irn_visited(succ))
255 walk_memory(succ, pre, post, ctx);
257 } else if (mode == mode_T) {
258 /* only some Proj's uses memory */
259 for (i = get_irn_n_outs(irn) - 1; i >= 0; --i) {
260 ir_node *proj = get_irn_out(irn, i);
262 if (get_irn_mode(proj) == mode_M && ! irn_visited(proj))
263 walk_memory(proj, pre, post, ctx);
271 * Walks over all memory nodes of a graph.
274 * @param pre pre walker function
275 * @param post post walker function
276 * @param ctx context parameter for the walker functions
278 static void walk_memory_irg(ir_graph *irg, irg_walk_func pre, irg_walk_func post, void *ctx)
280 inc_irg_visited(irg);
282 ir_reserve_resources(irg, IR_RESOURCE_IRN_VISITED);
285 * there are two possible sources for memory: initial_mem and nomem
286 * we ignore nomem as this should NOT change the memory
288 walk_memory(get_irg_initial_mem(irg), pre, post, ctx);
290 ir_free_resources(irg, IR_RESOURCE_IRN_VISITED);
291 } /* walk_memory_irg */
294 * Register an address and allocate a (sparse, 0..n) ID for it.
296 * @param adr the IR-node representing the address
298 * @return the allocated id
300 static unsigned register_address(ir_node *adr)
302 address_entry *entry;
304 /* skip Confirms and Casts */
306 if (is_Confirm(adr)) {
307 adr = get_Confirm_value(adr);
311 adr = get_Cast_op(adr);
315 entry = (address_entry*)ir_nodemap_get(&env.adr_map, adr);
319 entry = OALLOC(&env.obst, address_entry);
321 entry->id = env.curr_adr_id++;
322 ir_nodemap_insert(&env.adr_map, adr, entry);
324 DB((dbg, LEVEL_3, "ADDRESS %+F has ID %u\n", adr, entry->id));
326 ARR_APP1(ir_node *, env.id_2_address, adr);
330 } /* register_address */
334 * translate an address through a Phi node into a given predecessor
337 * @param address the address
338 * @param block the block
339 * @param pos the position of the predecessor in block
341 static ir_node *phi_translate(ir_node *address, const ir_node *block, int pos)
343 if (is_Phi(address) && get_nodes_block(address) == block)
344 address = get_Phi_pred(address, pos);
346 } /* phi_translate */
349 * Walker: allocate an block entry for every block
350 * and register all potential addresses.
352 static void prepare_blocks(ir_node *irn, void *ctx)
357 block_t *entry = OALLOC(&env.obst, block_t);
360 entry->memop_forward = NULL;
361 entry->memop_backward = NULL;
362 entry->avail_out = NULL;
363 entry->id_2_memop_avail = NULL;
364 entry->anticL_in = NULL;
365 entry->id_2_memop_antic = NULL;
367 entry->forward_next = NULL;
368 entry->backward_next = NULL;
370 entry->trans_results = NULL;
371 set_irn_link(irn, entry);
373 set_Block_phis(irn, NULL);
375 /* use block marks to track unreachable blocks */
376 set_Block_mark(irn, 0);
378 n = get_Block_n_cfgpreds(irn);
379 if (n > env.max_cfg_preds)
380 env.max_cfg_preds = n;
382 ir_mode *mode = get_irn_mode(irn);
384 if (mode_is_reference(mode)) {
386 * Register ALL possible addresses: this is overkill yet but
387 * simpler then doing it for all possible translated addresses
388 * (which would be sufficient in the moment.
390 (void)register_address(irn);
393 } /* prepare_blocks */
396 * Post-Walker, link in all Phi's
398 static void link_phis(ir_node *irn, void *ctx)
403 ir_node *block = get_nodes_block(irn);
404 add_Block_phi(block, irn);
409 * Block walker: creates the inverse post-order list for the CFG.
411 static void inverse_post_order(ir_node *block, void *ctx)
413 block_t *entry = get_block_entry(block);
417 /* mark this block IS reachable from start */
418 set_Block_mark(block, 1);
420 /* create the list in inverse order */
421 entry->forward_next = env.forward;
424 /* remember the first visited (last in list) entry, needed for later */
425 if (env.backward == NULL)
426 env.backward = entry;
427 } /* inverse_post_order */
430 * Block walker: create backward links for the memops of a block.
432 static void collect_backward(ir_node *block, void *ctx)
434 block_t *entry = get_block_entry(block);
440 * Do NOT link in the end block yet. We want it to be
441 * the first in the list. This is NOT guaranteed by the walker
442 * if we have endless loops.
444 if (block != env.end_bl) {
445 entry->backward_next = env.backward;
447 /* create the list in inverse order */
448 env.backward = entry;
451 /* create backward links for all memory ops */
453 for (op = entry->memop_forward; op != NULL; op = op->next) {
457 entry->memop_backward = last;
458 } /* collect_backward */
463 * @param irn the IR-node representing the memop or NULL
464 * if this is a translated (virtual) memop
466 * @return the allocated memop
468 static memop_t *alloc_memop(ir_node *irn)
470 memop_t *m = OALLOC(&env.obst, memop_t);
472 m->value.address = NULL;
473 m->value.value = NULL;
474 m->value.mode = NULL;
482 memset(m->projs, 0, sizeof(m->projs));
485 set_irn_link(irn, m);
490 * Create a memop for a Phi-replacement.
492 * @param op the memop to clone
493 * @param phi the Phi-node representing the new value
495 static memop_t *clone_memop_phi(memop_t *op, ir_node *phi)
497 memop_t *m = OALLOC(&env.obst, memop_t);
499 m->value = op->value;
500 m->value.value = phi;
507 set_irn_link(phi, m);
509 } /* clone_memop_phi */
512 * Return the memory properties of a call node.
514 * @param call the call node
516 * return a bitset of mtp_property_const and mtp_property_pure
518 static unsigned get_Call_memory_properties(ir_node *call)
520 ir_type *call_tp = get_Call_type(call);
521 unsigned prop = get_method_additional_properties(call_tp);
523 /* check first the call type */
524 if ((prop & (mtp_property_const|mtp_property_pure)) == 0) {
525 /* try the called entity */
526 ir_node *ptr = get_Call_ptr(call);
528 if (is_Global(ptr)) {
529 ir_entity *ent = get_Global_entity(ptr);
531 prop = get_entity_additional_properties(ent);
534 return prop & (mtp_property_const|mtp_property_pure);
535 } /* get_Call_memory_properties */
538 * Returns an entity if the address ptr points to a constant one.
540 * @param ptr the address
542 * @return an entity or NULL
544 static ir_entity *find_constant_entity(ir_node *ptr)
547 if (is_SymConst(ptr) && get_SymConst_kind(ptr) == symconst_addr_ent) {
548 return get_SymConst_entity(ptr);
549 } else if (is_Sel(ptr)) {
550 ir_entity *ent = get_Sel_entity(ptr);
551 ir_type *tp = get_entity_owner(ent);
553 /* Do not fiddle with polymorphism. */
554 if (is_Class_type(get_entity_owner(ent)) &&
555 ((get_entity_n_overwrites(ent) != 0) ||
556 (get_entity_n_overwrittenby(ent) != 0) ) )
559 if (is_Array_type(tp)) {
563 for (i = 0, n = get_Sel_n_indexs(ptr); i < n; ++i) {
565 ir_tarval *tlower, *tupper;
566 ir_node *index = get_Sel_index(ptr, i);
567 ir_tarval *tv = computed_value(index);
569 /* check if the index is constant */
570 if (tv == tarval_bad)
573 bound = get_array_lower_bound(tp, i);
574 tlower = computed_value(bound);
575 bound = get_array_upper_bound(tp, i);
576 tupper = computed_value(bound);
578 if (tlower == tarval_bad || tupper == tarval_bad)
581 if (tarval_cmp(tv, tlower) == ir_relation_less)
583 if (tarval_cmp(tupper, tv) == ir_relation_less)
586 /* ok, bounds check finished */
590 if (get_entity_linkage(ent) == IR_LINKAGE_CONSTANT)
594 ptr = get_Sel_ptr(ptr);
595 } else if (is_Add(ptr)) {
596 ir_node *l = get_Add_left(ptr);
597 ir_node *r = get_Add_right(ptr);
599 if (get_irn_mode(l) == get_irn_mode(ptr) && is_Const(r))
601 else if (get_irn_mode(r) == get_irn_mode(ptr) && is_Const(l))
606 /* for now, we support only one addition, reassoc should fold all others */
607 if (! is_SymConst(ptr) && !is_Sel(ptr))
609 } else if (is_Sub(ptr)) {
610 ir_node *l = get_Sub_left(ptr);
611 ir_node *r = get_Sub_right(ptr);
613 if (get_irn_mode(l) == get_irn_mode(ptr) && is_Const(r))
617 /* for now, we support only one subtraction, reassoc should fold all others */
618 if (! is_SymConst(ptr) && !is_Sel(ptr))
623 } /* find_constant_entity */
626 * Return the Selection index of a Sel node from dimension n
628 static long get_Sel_array_index_long(ir_node *n, int dim)
630 ir_node *index = get_Sel_index(n, dim);
631 assert(is_Const(index));
632 return get_tarval_long(get_Const_tarval(index));
633 } /* get_Sel_array_index_long */
636 * Returns the accessed component graph path for an
637 * node computing an address.
639 * @param ptr the node computing the address
640 * @param depth current depth in steps upward from the root
643 static compound_graph_path *rec_get_accessed_path(ir_node *ptr, size_t depth)
645 compound_graph_path *res = NULL;
646 ir_entity *root, *field, *ent;
647 size_t path_len, pos, idx;
651 if (is_SymConst(ptr)) {
652 /* a SymConst. If the depth is 0, this is an access to a global
653 * entity and we don't need a component path, else we know
654 * at least its length.
656 assert(get_SymConst_kind(ptr) == symconst_addr_ent);
657 root = get_SymConst_entity(ptr);
658 res = (depth == 0) ? NULL : new_compound_graph_path(get_entity_type(root), depth);
659 } else if (is_Sel(ptr)) {
660 /* it's a Sel, go up until we find the root */
661 res = rec_get_accessed_path(get_Sel_ptr(ptr), depth+1);
665 /* fill up the step in the path at the current position */
666 field = get_Sel_entity(ptr);
667 path_len = get_compound_graph_path_length(res);
668 pos = path_len - depth - 1;
669 set_compound_graph_path_node(res, pos, field);
671 if (is_Array_type(get_entity_owner(field))) {
672 assert(get_Sel_n_indexs(ptr) == 1 && "multi dim arrays not implemented");
673 set_compound_graph_path_array_index(res, pos, get_Sel_array_index_long(ptr, 0));
675 } else if (is_Add(ptr)) {
680 ir_node *l = get_Add_left(ptr);
681 ir_node *r = get_Add_right(ptr);
682 if (is_Const(r) && get_irn_mode(l) == get_irn_mode(ptr)) {
684 tv = get_Const_tarval(r);
687 tv = get_Const_tarval(l);
691 mode = get_tarval_mode(tv);
694 /* ptr must be a Sel or a SymConst, this was checked in find_constant_entity() */
696 field = get_Sel_entity(ptr);
698 field = get_SymConst_entity(ptr);
701 for (ent = field;;) {
703 ir_tarval *sz, *tv_index, *tlower, *tupper;
706 tp = get_entity_type(ent);
707 if (! is_Array_type(tp))
709 ent = get_array_element_entity(tp);
710 size = get_type_size_bytes(get_entity_type(ent));
711 sz = new_tarval_from_long(size, mode);
713 tv_index = tarval_div(tmp, sz);
714 tmp = tarval_mod(tmp, sz);
716 if (tv_index == tarval_bad || tmp == tarval_bad)
719 assert(get_array_n_dimensions(tp) == 1 && "multiarrays not implemented");
720 bound = get_array_lower_bound(tp, 0);
721 tlower = computed_value(bound);
722 bound = get_array_upper_bound(tp, 0);
723 tupper = computed_value(bound);
725 if (tlower == tarval_bad || tupper == tarval_bad)
728 if (tarval_cmp(tv_index, tlower) == ir_relation_less)
730 if (tarval_cmp(tupper, tv_index) == ir_relation_less)
733 /* ok, bounds check finished */
736 if (! tarval_is_null(tmp)) {
737 /* access to some struct/union member */
741 /* should be at least ONE array */
745 res = rec_get_accessed_path(ptr, depth + idx);
749 path_len = get_compound_graph_path_length(res);
750 pos = path_len - depth - idx;
752 for (ent = field;;) {
754 ir_tarval *sz, *tv_index;
757 tp = get_entity_type(ent);
758 if (! is_Array_type(tp))
760 ent = get_array_element_entity(tp);
761 set_compound_graph_path_node(res, pos, ent);
763 size = get_type_size_bytes(get_entity_type(ent));
764 sz = new_tarval_from_long(size, mode);
766 tv_index = tarval_div(tv, sz);
767 tv = tarval_mod(tv, sz);
769 /* worked above, should work again */
770 assert(tv_index != tarval_bad && tv != tarval_bad);
772 /* bounds already checked above */
773 index = get_tarval_long(tv_index);
774 set_compound_graph_path_array_index(res, pos, index);
777 } else if (is_Sub(ptr)) {
778 ir_node *l = get_Sub_left(ptr);
779 ir_node *r = get_Sub_right(ptr);
782 tv = get_Const_tarval(r);
787 } /* rec_get_accessed_path */
790 * Returns an access path or NULL. The access path is only
791 * valid, if the graph is in phase_high and _no_ address computation is used.
793 static compound_graph_path *get_accessed_path(ir_node *ptr)
795 compound_graph_path *gr = rec_get_accessed_path(ptr, 0);
797 } /* get_accessed_path */
799 typedef struct path_entry {
801 struct path_entry *next;
805 static ir_node *rec_find_compound_ent_value(ir_node *ptr, path_entry *next)
807 path_entry entry, *p;
808 ir_entity *ent, *field;
809 ir_initializer_t *initializer;
815 if (is_SymConst(ptr)) {
817 ent = get_SymConst_entity(ptr);
818 initializer = get_entity_initializer(ent);
819 for (p = next; p != NULL;) {
820 if (initializer->kind != IR_INITIALIZER_COMPOUND)
822 n = get_initializer_compound_n_entries(initializer);
823 tp = get_entity_type(ent);
825 if (is_Array_type(tp)) {
826 ent = get_array_element_entity(tp);
831 initializer = get_initializer_compound_value(initializer, 0);
837 initializer = get_initializer_compound_value(initializer, p->index);
842 tp = get_entity_type(ent);
843 while (is_Array_type(tp)) {
844 ent = get_array_element_entity(tp);
845 tp = get_entity_type(ent);
847 n = get_initializer_compound_n_entries(initializer);
850 initializer = get_initializer_compound_value(initializer, 0);
853 switch (initializer->kind) {
854 case IR_INITIALIZER_CONST:
855 return get_initializer_const_value(initializer);
856 case IR_INITIALIZER_TARVAL:
857 case IR_INITIALIZER_NULL:
861 } else if (is_Sel(ptr)) {
862 entry.ent = field = get_Sel_entity(ptr);
863 tp = get_entity_owner(field);
864 if (is_Array_type(tp)) {
865 assert(get_Sel_n_indexs(ptr) == 1 && "multi dim arrays not implemented");
866 entry.index = get_Sel_array_index_long(ptr, 0) - get_array_lower_bound_int(tp, 0);
868 size_t i, n_members = get_compound_n_members(tp);
869 for (i = 0; i < n_members; ++i) {
870 if (get_compound_member(tp, i) == field)
873 if (i >= n_members) {
874 /* not found: should NOT happen */
879 return rec_find_compound_ent_value(get_Sel_ptr(ptr), &entry);
880 } else if (is_Add(ptr)) {
885 ir_node *l = get_Add_left(ptr);
886 ir_node *r = get_Add_right(ptr);
889 tv = get_Const_tarval(r);
892 tv = get_Const_tarval(l);
896 mode = get_tarval_mode(tv);
898 /* ptr must be a Sel or a SymConst, this was checked in find_constant_entity() */
900 field = get_Sel_entity(ptr);
902 field = get_SymConst_entity(ptr);
905 /* count needed entries */
907 for (ent = field;;) {
908 tp = get_entity_type(ent);
909 if (! is_Array_type(tp))
911 ent = get_array_element_entity(tp);
914 /* should be at least ONE entry */
918 /* allocate the right number of entries */
919 NEW_ARR_A(path_entry, p, pos);
923 for (ent = field;;) {
925 ir_tarval *sz, *tv_index, *tlower, *tupper;
929 tp = get_entity_type(ent);
930 if (! is_Array_type(tp))
932 ent = get_array_element_entity(tp);
934 p[pos].next = &p[pos + 1];
936 size = get_type_size_bytes(get_entity_type(ent));
937 sz = new_tarval_from_long(size, mode);
939 tv_index = tarval_div(tv, sz);
940 tv = tarval_mod(tv, sz);
942 if (tv_index == tarval_bad || tv == tarval_bad)
945 assert(get_array_n_dimensions(tp) == 1 && "multiarrays not implemented");
946 bound = get_array_lower_bound(tp, 0);
947 tlower = computed_value(bound);
948 bound = get_array_upper_bound(tp, 0);
949 tupper = computed_value(bound);
951 if (tlower == tarval_bad || tupper == tarval_bad)
954 if (tarval_cmp(tv_index, tlower) == ir_relation_less)
956 if (tarval_cmp(tupper, tv_index) == ir_relation_less)
959 /* ok, bounds check finished */
960 index = get_tarval_long(tv_index);
961 p[pos].index = index;
964 if (! tarval_is_null(tv)) {
965 /* hmm, wrong access */
968 p[pos - 1].next = next;
969 return rec_find_compound_ent_value(ptr, p);
970 } else if (is_Sub(ptr)) {
971 ir_node *l = get_Sub_left(ptr);
972 ir_node *r = get_Sub_right(ptr);
975 tv = get_Const_tarval(r);
980 } /* rec_find_compound_ent_value */
982 static ir_node *find_compound_ent_value(ir_node *ptr)
984 return rec_find_compound_ent_value(ptr, NULL);
985 } /* find_compound_ent_value */
988 * Mark a Load memop to be replace by a definition
990 * @param op the Load memop
992 static void mark_replace_load(memop_t *op, ir_node *def)
995 op->flags |= FLAG_KILLED_NODE;
997 } /* mark_replace_load */
1000 * Mark a Store memop to be removed.
1002 * @param op the Store memop
1004 static void mark_remove_store(memop_t *op)
1006 op->flags |= FLAG_KILLED_NODE;
1008 } /* mark_remove_store */
1011 * Update a memop for a Load.
1013 * @param m the memop
1015 static void update_Load_memop(memop_t *m)
1018 ir_node *load = m->node;
1022 if (get_Load_volatility(load) == volatility_is_volatile)
1023 m->flags |= FLAG_IGNORE;
1025 ptr = get_Load_ptr(load);
1027 m->value.address = ptr;
1029 for (i = get_irn_n_outs(load) - 1; i >= 0; --i) {
1030 ir_node *proj = get_irn_out(load, i);
1033 /* beware of keep edges */
1037 pn = get_Proj_proj(proj);
1038 m->projs[pn] = proj;
1041 m->value.value = proj;
1042 m->value.mode = get_irn_mode(proj);
1044 case pn_Load_X_except:
1045 m->flags |= FLAG_EXCEPTION;
1050 case pn_Load_X_regular:
1053 panic("Unsupported Proj from Load %+F", proj);
1057 /* check if we can determine the entity that will be loaded */
1058 ent = find_constant_entity(ptr);
1060 if (ent != NULL && get_entity_visibility(ent) != ir_visibility_external) {
1061 /* a static allocation that is not external: there should be NO exception
1062 * when loading even if we cannot replace the load itself. */
1063 ir_node *value = NULL;
1065 /* no exception, clear the m fields as it might be checked later again */
1066 if (m->projs[pn_Load_X_except]) {
1067 ir_graph *irg = get_irn_irg(ptr);
1068 exchange(m->projs[pn_Load_X_except], new_r_Bad(irg, mode_X));
1069 m->projs[pn_Load_X_except] = NULL;
1070 m->flags &= ~FLAG_EXCEPTION;
1073 if (m->projs[pn_Load_X_regular]) {
1074 exchange(m->projs[pn_Load_X_regular], new_r_Jmp(get_nodes_block(load)));
1075 m->projs[pn_Load_X_regular] = NULL;
1079 if (get_entity_linkage(ent) & IR_LINKAGE_CONSTANT) {
1080 if (ent->initializer) {
1081 /* new style initializer */
1082 value = find_compound_ent_value(ptr);
1083 } else if (entity_has_compound_ent_values(ent)) {
1084 /* old style initializer */
1085 compound_graph_path *path = get_accessed_path(ptr);
1088 assert(is_proper_compound_graph_path(path, get_compound_graph_path_length(path)-1));
1090 value = get_compound_ent_value_by_path(ent, path);
1091 DB((dbg, LEVEL_1, " Constant access at %F%F resulted in %+F\n", ent, path, value));
1092 free_compound_graph_path(path);
1096 value = can_replace_load_by_const(load, value);
1099 if (value != NULL) {
1100 /* we completely replace the load by this value */
1101 DB((dbg, LEVEL_1, "Replacing Load %+F by constant %+F\n", m->node, value));
1102 mark_replace_load(m, value);
1107 if (m->value.value != NULL && !(m->flags & FLAG_IGNORE)) {
1108 /* only create an address if this node is NOT killed immediately or ignored */
1109 m->value.id = register_address(ptr);
1112 /* no user, KILL it */
1113 mark_replace_load(m, NULL);
1115 } /* update_Load_memop */
1118 * Update a memop for a Store.
1120 * @param m the memop
1122 static void update_Store_memop(memop_t *m)
1125 ir_node *store = m->node;
1126 ir_node *adr = get_Store_ptr(store);
1128 if (get_Store_volatility(store) == volatility_is_volatile) {
1129 m->flags |= FLAG_IGNORE;
1131 /* only create an address if this node is NOT ignored */
1132 m->value.id = register_address(adr);
1136 m->value.address = adr;
1138 for (i = get_irn_n_outs(store) - 1; i >= 0; --i) {
1139 ir_node *proj = get_irn_out(store, i);
1142 /* beware of keep edges */
1146 pn = get_Proj_proj(proj);
1147 m->projs[pn] = proj;
1149 case pn_Store_X_except:
1150 m->flags |= FLAG_EXCEPTION;
1155 case pn_Store_X_regular:
1158 panic("Unsupported Proj from Store %+F", proj);
1161 m->value.value = get_Store_value(store);
1162 m->value.mode = get_irn_mode(m->value.value);
1163 } /* update_Store_memop */
1166 * Update a memop for a Call.
1168 * @param m the memop
1170 static void update_Call_memop(memop_t *m)
1172 ir_node *call = m->node;
1173 unsigned prop = get_Call_memory_properties(call);
1176 if (prop & mtp_property_const) {
1177 /* A constant call did NOT use memory at all, we
1178 can kick it from the list. */
1179 } else if (prop & mtp_property_pure) {
1180 /* pure calls READ memory */
1183 m->flags = FLAG_KILL_ALL;
1185 for (i = get_irn_n_outs(call) - 1; i >= 0; --i) {
1186 ir_node *proj = get_irn_out(call, i);
1188 /* beware of keep edges */
1192 switch (get_Proj_proj(proj)) {
1193 case pn_Call_X_except:
1194 m->flags |= FLAG_EXCEPTION;
1201 } /* update_Call_memop */
1204 * Update a memop for a Div/Mod.
1206 * @param m the memop
1208 static void update_Div_memop(memop_t *m)
1210 ir_node *div = m->node;
1213 for (i = get_irn_n_outs(div) - 1; i >= 0; --i) {
1214 ir_node *proj = get_irn_out(div, i);
1216 /* beware of keep edges */
1220 switch (get_Proj_proj(proj)) {
1221 case pn_Div_X_except:
1222 m->flags |= FLAG_EXCEPTION;
1231 static void update_Mod_memop(memop_t *m)
1233 ir_node *div = m->node;
1236 for (i = get_irn_n_outs(div) - 1; i >= 0; --i) {
1237 ir_node *proj = get_irn_out(div, i);
1239 /* beware of keep edges */
1243 switch (get_Proj_proj(proj)) {
1244 case pn_Mod_X_except:
1245 m->flags |= FLAG_EXCEPTION;
1255 * Update a memop for a Phi.
1257 * @param m the memop
1259 static void update_Phi_memop(memop_t *m)
1261 /* the Phi is its own mem */
1263 } /* update_Phi_memop */
1266 * Memory walker: collect all memory ops and build topological lists.
1268 static void collect_memops(ir_node *irn, void *ctx)
1276 /* we can safely ignore ProjM's except the initial memory */
1277 ir_graph *irg = get_irn_irg(irn);
1278 if (irn != get_irg_initial_mem(irg))
1282 op = alloc_memop(irn);
1283 block = get_nodes_block(irn);
1284 entry = get_block_entry(block);
1287 update_Phi_memop(op);
1288 /* Phis must be always placed first */
1289 op->next = entry->memop_forward;
1290 entry->memop_forward = op;
1291 if (entry->memop_backward == NULL)
1292 entry->memop_backward = op;
1294 switch (get_irn_opcode(irn)) {
1296 update_Load_memop(op);
1299 update_Store_memop(op);
1302 update_Call_memop(op);
1309 /* initial memory */
1314 /* we can those to find the memory edge */
1317 update_Div_memop(op);
1320 update_Mod_memop(op);
1324 /* TODO: handle some builtins */
1326 /* unsupported operation */
1327 op->flags = FLAG_KILL_ALL;
1331 /* all other should be placed last */
1332 if (entry->memop_backward == NULL) {
1333 entry->memop_forward = entry->memop_backward = op;
1335 entry->memop_backward->next = op;
1336 entry->memop_backward = op;
1339 } /* collect_memops */
1342 * Find an address in the current set.
1344 * @param value the value to be searched for
1346 * @return a memop for the value or NULL if the value does
1347 * not exists in the set or cannot be converted into
1348 * the requested mode
1350 static memop_t *find_address(const value_t *value)
1352 if (rbitset_is_set(env.curr_set, value->id)) {
1353 memop_t *res = env.curr_id_2_memop[value->id];
1355 if (res->value.mode == value->mode)
1357 /* allow hidden casts */
1358 if (get_mode_arithmetic(res->value.mode) == irma_twos_complement &&
1359 get_mode_arithmetic(value->mode) == irma_twos_complement &&
1360 get_mode_size_bits(res->value.mode) == get_mode_size_bits(value->mode))
1364 } /* find_address */
1367 * Find an address in the avail_out set.
1369 * @param bl the block
1371 static memop_t *find_address_avail(const block_t *bl, unsigned id, const ir_mode *mode)
1373 if (rbitset_is_set(bl->avail_out, id)) {
1374 memop_t *res = bl->id_2_memop_avail[id];
1376 if (res->value.mode == mode)
1378 /* allow hidden casts */
1379 if (get_mode_arithmetic(res->value.mode) == irma_twos_complement &&
1380 get_mode_arithmetic(mode) == irma_twos_complement &&
1381 get_mode_size_bits(res->value.mode) == get_mode_size_bits(mode))
1385 } /* find_address_avail */
1388 * Kill all addresses from the current set.
1390 static void kill_all(void)
1392 rbitset_clear_all(env.curr_set, env.rbs_size);
1395 rbitset_set(env.curr_set, env.rbs_size - 1);
1399 * Kill memops that are not alias free due to a Store value from the current set.
1401 * @param value the Store value
1403 static void kill_memops(const value_t *value)
1405 size_t end = env.rbs_size - 1;
1408 for (pos = rbitset_next(env.curr_set, 0, 1); pos < end; pos = rbitset_next(env.curr_set, pos + 1, 1)) {
1409 memop_t *op = env.curr_id_2_memop[pos];
1411 if (ir_no_alias != get_alias_relation(value->address, value->mode,
1412 op->value.address, op->value.mode)) {
1413 rbitset_clear(env.curr_set, pos);
1414 env.curr_id_2_memop[pos] = NULL;
1415 DB((dbg, LEVEL_2, "KILLING %+F because of possible alias address %+F\n", op->node, value->address));
1421 * Add the value of a memop to the current set.
1423 * @param op the memory op
1425 static void add_memop(memop_t *op)
1427 rbitset_set(env.curr_set, op->value.id);
1428 env.curr_id_2_memop[op->value.id] = op;
1432 * Add the value of a memop to the avail_out set.
1434 * @param bl the block
1435 * @param op the memory op
1437 static void add_memop_avail(block_t *bl, memop_t *op)
1439 rbitset_set(bl->avail_out, op->value.id);
1440 bl->id_2_memop_avail[op->value.id] = op;
1441 } /* add_memop_avail */
1444 * Check, if we can convert a value of one mode to another mode
1445 * without changing the representation of bits.
1447 * @param from the original mode
1448 * @param to the destination mode
1450 static int can_convert_to(const ir_mode *from, const ir_mode *to)
1452 if (get_mode_arithmetic(from) == irma_twos_complement &&
1453 get_mode_arithmetic(to) == irma_twos_complement &&
1454 get_mode_size_bits(from) == get_mode_size_bits(to))
1457 } /* can_convert_to */
1460 * Add a Conv to the requested mode if needed.
1462 * @param irn the IR-node to convert
1463 * @param mode the destination mode
1465 * @return the possible converted node or NULL
1466 * if the conversion is not possible
1468 static ir_node *conv_to(ir_node *irn, ir_mode *mode)
1470 ir_mode *other = get_irn_mode(irn);
1471 if (other != mode) {
1472 /* different modes: check if conversion is possible without changing the bits */
1473 if (can_convert_to(other, mode)) {
1474 ir_node *block = get_nodes_block(irn);
1475 return new_r_Conv(block, irn, mode);
1477 /* otherwise not possible ... yet */
1484 * Update the address of an value if this address was a load result
1485 * and the load is killed now.
1487 * @param value the value whose address is updated
1489 static void update_address(value_t *value)
1491 if (is_Proj(value->address)) {
1492 ir_node *load = get_Proj_pred(value->address);
1494 if (is_Load(load)) {
1495 const memop_t *op = get_irn_memop(load);
1497 if (op->flags & FLAG_KILLED_NODE)
1498 value->address = op->replace;
1501 } /* update_address */
1504 * Do forward dataflow analysis on the given block and calculate the
1505 * GEN and KILL in the current (avail) set.
1507 * @param bl the block
1509 static void calc_gen_kill_avail(block_t *bl)
1514 for (op = bl->memop_forward; op != NULL; op = op->next) {
1515 switch (get_irn_opcode(op->node)) {
1523 if (! (op->flags & (FLAG_KILLED_NODE|FLAG_IGNORE))) {
1524 /* do we have this already? */
1527 update_address(&op->value);
1528 other = find_address(&op->value);
1529 if (other != NULL && other != op) {
1530 def = conv_to(other->value.value, op->value.mode);
1532 #ifdef DEBUG_libfirm
1533 if (is_Store(other->node)) {
1535 DB((dbg, LEVEL_1, "RAW %+F <- %+F(%+F)\n", op->node, def, other->node));
1538 DB((dbg, LEVEL_1, "RAR %+F <- %+F(%+F)\n", op->node, def, other->node));
1541 mark_replace_load(op, def);
1542 /* do NOT change the memop table */
1546 /* add this value */
1551 if (! (op->flags & FLAG_KILLED_NODE)) {
1552 /* do we have this store already */
1555 update_address(&op->value);
1556 other = find_address(&op->value);
1557 if (other != NULL) {
1558 if (is_Store(other->node)) {
1559 if (op != other && !(other->flags & FLAG_IGNORE) &&
1560 get_nodes_block(other->node) == get_nodes_block(op->node)) {
1562 * A WAW in the same block we can kick the first store.
1563 * This is a shortcut: we know that the second Store will be anticipated
1566 DB((dbg, LEVEL_1, "WAW %+F <- %+F\n", other->node, op->node));
1567 mark_remove_store(other);
1568 /* FIXME: a Load might be get freed due to this killed store */
1570 } else if (other->value.value == op->value.value && !(op->flags & FLAG_IGNORE)) {
1572 DB((dbg, LEVEL_1, "WAR %+F <- %+F\n", op->node, other->node));
1573 mark_remove_store(op);
1574 /* do NOT change the memop table */
1578 /* KILL all possible aliases */
1579 kill_memops(&op->value);
1580 /* add this value */
1585 if (op->flags & FLAG_KILL_ALL)
1589 } /* calc_gen_kill_avail */
1591 #define BYTE_SIZE(x) (((x) + 7) >> 3)
1594 * Do forward dataflow analysis on a given block to calculate the avail_out set
1595 * for this block only.
1597 * @param block the block
1599 static void forward_avail(block_t *bl)
1601 /* fill the data from the current block */
1602 env.curr_id_2_memop = bl->id_2_memop_avail;
1603 env.curr_set = bl->avail_out;
1605 calc_gen_kill_avail(bl);
1606 dump_curr(bl, "Avail_out");
1607 } /* forward_avail */
1610 * Do backward dataflow analysis on a given block to calculate the antic set
1611 * of Loaded addresses.
1613 * @param bl the block
1615 * @return non-zero if the set has changed since last iteration
1617 static int backward_antic(block_t *bl)
1620 ir_node *block = bl->block;
1621 int n = get_Block_n_cfg_outs(block);
1624 ir_node *succ = get_Block_cfg_out(block, 0);
1625 block_t *succ_bl = get_block_entry(succ);
1626 int pred_pos = get_Block_cfgpred_pos(succ, block);
1627 size_t end = env.rbs_size - 1;
1632 if (bl->trans_results == NULL) {
1633 /* allocate the translate cache */
1634 bl->trans_results = OALLOCNZ(&env.obst, memop_t*, env.curr_adr_id);
1637 /* check for partly redundant values */
1638 for (pos = rbitset_next(succ_bl->anticL_in, 0, 1);
1640 pos = rbitset_next(succ_bl->anticL_in, pos + 1, 1)) {
1642 * do Phi-translation here: Note that at this point the nodes are
1643 * not changed, so we can safely cache the results.
1644 * However: Loads of Load results ARE bad, because we have no way
1645 to translate them yet ...
1647 memop_t *op = bl->trans_results[pos];
1649 /* not yet translated */
1650 ir_node *adr, *trans_adr;
1652 op = succ_bl->id_2_memop_antic[pos];
1653 adr = op->value.address;
1655 trans_adr = phi_translate(adr, succ, pred_pos);
1656 if (trans_adr != adr) {
1657 /* create a new entry for the translated one */
1660 new_op = alloc_memop(NULL);
1661 new_op->value.address = trans_adr;
1662 new_op->value.id = register_address(trans_adr);
1663 new_op->value.mode = op->value.mode;
1664 new_op->node = op->node; /* we need the node to decide if Load/Store */
1665 new_op->flags = op->flags;
1667 bl->trans_results[pos] = new_op;
1671 env.curr_id_2_memop[op->value.id] = op;
1672 rbitset_set(env.curr_set, op->value.id);
1675 ir_node *succ = get_Block_cfg_out(block, 0);
1676 block_t *succ_bl = get_block_entry(succ);
1679 rbitset_copy(env.curr_set, succ_bl->anticL_in, env.rbs_size);
1680 memcpy(env.curr_id_2_memop, succ_bl->id_2_memop_antic, env.rbs_size * sizeof(env.curr_id_2_memop[0]));
1682 /* Hmm: probably we want kill merges of Loads ans Stores here */
1683 for (i = n - 1; i > 0; --i) {
1684 ir_node *succ = get_Block_cfg_out(bl->block, i);
1685 block_t *succ_bl = get_block_entry(succ);
1687 rbitset_and(env.curr_set, succ_bl->anticL_in, env.rbs_size);
1690 /* block ends with a noreturn call */
1694 dump_curr(bl, "AnticL_out");
1696 for (op = bl->memop_backward; op != NULL; op = op->prev) {
1697 switch (get_irn_opcode(op->node)) {
1705 if (! (op->flags & (FLAG_KILLED_NODE|FLAG_IGNORE))) {
1711 if (! (op->flags & FLAG_KILLED_NODE)) {
1712 /* a Store: check which memops must be killed */
1713 kill_memops(&op->value);
1717 if (op->flags & FLAG_KILL_ALL)
1722 memcpy(bl->id_2_memop_antic, env.curr_id_2_memop, env.rbs_size * sizeof(env.curr_id_2_memop[0]));
1723 if (! rbitsets_equal(bl->anticL_in, env.curr_set, env.rbs_size)) {
1725 rbitset_copy(bl->anticL_in, env.curr_set, env.rbs_size);
1726 dump_curr(bl, "AnticL_in*");
1729 dump_curr(bl, "AnticL_in");
1731 } /* backward_antic */
1734 * Replace a Load memop by a already known value.
1736 * @param op the Load memop
1738 static void replace_load(memop_t *op)
1740 ir_node *load = op->node;
1741 ir_node *def = skip_Id(op->replace);
1746 DB((dbg, LEVEL_1, "Replacing %+F by definition %+F\n", load, is_Proj(def) ? get_Proj_pred(def) : def));
1748 if (op->flags & FLAG_EXCEPTION) {
1749 /* bad: this node is unused and executed for exception only */
1750 DB((dbg, LEVEL_1, "Unused %+F executed for exception only ...\n", load));
1753 DB((dbg, LEVEL_1, "Killing unused %+F\n", load));
1756 if (op->mem != NULL) {
1757 /* in rare cases a Load might have NO memory */
1758 exchange(op->mem, get_Load_mem(load));
1760 proj = op->projs[pn_Load_res];
1762 mode = get_irn_mode(proj);
1763 if (get_irn_mode(def) != mode) {
1765 dbg_info *db = get_irn_dbg_info(load);
1766 ir_node *block = get_nodes_block(proj);
1767 def = new_rd_Conv(db, block, def, mode);
1769 exchange(proj, def);
1771 proj = op->projs[pn_Load_X_except];
1773 ir_graph *irg = get_irn_irg(load);
1774 exchange(proj, new_r_Bad(irg, mode_X));
1776 proj = op->projs[pn_Load_X_regular];
1778 exchange(proj, new_r_Jmp(get_nodes_block(load)));
1780 } /* replace_load */
1783 * Remove a Store memop.
1785 * @param op the Store memop
1787 static void remove_store(memop_t *op)
1789 ir_node *store = op->node;
1792 DB((dbg, LEVEL_1, "Removing %+F\n", store));
1794 if (op->mem != NULL) {
1795 /* in rare cases a Store might have no memory */
1796 exchange(op->mem, get_Store_mem(store));
1798 proj = op->projs[pn_Store_X_except];
1800 ir_graph *irg = get_irn_irg(store);
1801 exchange(proj, new_r_Bad(irg, mode_X));
1803 proj = op->projs[pn_Store_X_regular];
1805 exchange(proj, new_r_Jmp(get_nodes_block(store)));
1807 } /* remove_store */
1811 * Do all necessary replacements for a given block.
1813 * @param bl the block
1815 static void do_replacements(block_t *bl)
1819 for (op = bl->memop_forward; op != NULL; op = op->next) {
1820 if (op->flags & FLAG_KILLED_NODE) {
1821 switch (get_irn_opcode(op->node)) {
1831 } /* do_replacements */
1834 * Calculate the Avail_out sets for all basic blocks.
1836 static void calcAvail(void)
1838 memop_t **tmp_memop = env.curr_id_2_memop;
1839 unsigned *tmp_set = env.curr_set;
1842 /* calculate avail_out */
1843 DB((dbg, LEVEL_2, "Calculate Avail_out\n"));
1845 /* iterate over all blocks in in any order, skip the start block */
1846 for (bl = env.forward->forward_next; bl != NULL; bl = bl->forward_next) {
1850 /* restore the current sets */
1851 env.curr_id_2_memop = tmp_memop;
1852 env.curr_set = tmp_set;
1856 * Calculate the Antic_in sets for all basic blocks.
1858 static void calcAntic(void)
1862 /* calculate antic_out */
1863 DB((dbg, LEVEL_2, "Calculate Antic_in\n"));
1868 DB((dbg, LEVEL_2, "Iteration %d:\n=========\n", i));
1872 /* over all blocks in reverse post order */
1873 for (bl = env.backward->backward_next; bl != NULL; bl = bl->backward_next) {
1874 need_iter |= backward_antic(bl);
1877 } while (need_iter);
1878 DB((dbg, LEVEL_2, "Get anticipated Load set after %d iterations\n", i));
1882 * Return the node representing the last memory in a block.
1884 * @param bl the block
1886 static ir_node *find_last_memory(block_t *bl)
1889 if (bl->memop_backward != NULL) {
1890 return bl->memop_backward->mem;
1892 /* if there is NO memory in this block, go to the dominator */
1893 bl = get_block_entry(get_Block_idom(bl->block));
1895 } /* find_last_memory */
1898 * Reroute all memory users of old memory
1899 * to a new memory IR-node.
1901 * @param omem the old memory IR-node
1902 * @param nmem the new memory IR-node
1904 static void reroute_all_mem_users(ir_node *omem, ir_node *nmem)
1908 for (i = get_irn_n_outs(omem) - 1; i >= 0; --i) {
1910 ir_node *user = get_irn_out_ex(omem, i, &n_pos);
1912 set_irn_n(user, n_pos, nmem);
1915 /* all edges previously point to omem now point to nmem */
1916 nmem->out = omem->out;
1917 } /* reroute_all_mem_users */
1920 * Reroute memory users of old memory that are dominated by a given block
1921 * to a new memory IR-node.
1923 * @param omem the old memory IR-node
1924 * @param nmem the new memory IR-node
1925 * @param pass_bl the block the memory must pass
1927 static void reroute_mem_through(ir_node *omem, ir_node *nmem, ir_node *pass_bl)
1929 int i, j, n = get_irn_n_outs(omem);
1930 ir_def_use_edge *edges = NEW_ARR_D(ir_def_use_edge, &env.obst, n + 1);
1932 for (i = j = 0; i < n; ++i) {
1934 ir_node *user = get_irn_out_ex(omem, i, &n_pos);
1935 ir_node *use_bl = get_nodes_block(user);
1939 use_bl = get_Block_cfgpred_block(use_bl, n_pos);
1941 if (block_dominates(pass_bl, use_bl)) {
1942 /* found an user that is dominated */
1944 edges[j].pos = n_pos;
1945 edges[j].use = user;
1947 set_irn_n(user, n_pos, nmem);
1951 /* Modify the out structure: we create a new out edge array on our
1952 temporary obstack here. This should be no problem, as we invalidate the edges
1953 at the end either. */
1954 /* first entry is used for the length */
1957 } /* reroute_mem_through */
1960 * insert Loads, making partly redundant Loads fully redundant
1962 static int insert_Load(block_t *bl)
1964 ir_node *block = bl->block;
1965 int i, n = get_Block_n_cfgpreds(block);
1966 size_t end = env.rbs_size - 1;
1968 DB((dbg, LEVEL_3, "processing %+F\n", block));
1971 /* might still happen for an unreachable block (end for instance) */
1979 NEW_ARR_A(ir_node *, ins, n);
1981 rbitset_set_all(env.curr_set, env.rbs_size);
1983 /* More than one predecessors, calculate the join for all avail_outs ignoring unevaluated
1984 Blocks. These put in Top anyway. */
1985 for (i = n - 1; i >= 0; --i) {
1986 ir_node *pred = skip_Proj(get_Block_cfgpred(block, i));
1987 ir_node *blk = get_nodes_block(pred);
1990 pred_bl = get_block_entry(blk);
1991 rbitset_and(env.curr_set, pred_bl->avail_out, env.rbs_size);
1993 if (is_Load(pred) || is_Store(pred)) {
1994 /* We reached this block by an exception from a Load or Store:
1995 * the memop creating the exception was NOT completed than, kill it
1997 memop_t *exc_op = get_irn_memop(pred);
1998 rbitset_clear(env.curr_set, exc_op->value.id);
2003 * Ensure that all values are in the map: build Phi's if necessary:
2004 * Note: the last bit is the sentinel and ALWAYS set, so end with -2.
2006 for (pos = 0; pos < env.rbs_size - 1; ++pos) {
2007 if (! rbitset_is_set(env.curr_set, pos))
2008 env.curr_id_2_memop[pos] = NULL;
2010 ir_node *pred = get_Block_cfgpred_block(bl->block, 0);
2011 block_t *pred_bl = get_block_entry(pred);
2013 memop_t *first = NULL;
2014 ir_mode *mode = NULL;
2016 for (i = 0; i < n; ++i) {
2019 pred = get_Block_cfgpred_block(bl->block, i);
2020 pred_bl = get_block_entry(pred);
2022 mop = pred_bl->id_2_memop_avail[pos];
2023 if (first == NULL) {
2025 ins[0] = first->value.value;
2026 mode = get_irn_mode(ins[0]);
2028 /* no Phi needed so far */
2029 env.curr_id_2_memop[pos] = first;
2031 ins[i] = conv_to(mop->value.value, mode);
2032 if (ins[i] != ins[0]) {
2033 if (ins[i] == NULL) {
2034 /* conversion failed */
2035 env.curr_id_2_memop[pos] = NULL;
2036 rbitset_clear(env.curr_set, pos);
2045 ir_node *phi = new_r_Phi(bl->block, n, ins, mode);
2046 memop_t *phiop = alloc_memop(phi);
2048 phiop->value = first->value;
2049 phiop->value.value = phi;
2051 /* no need to link it in, as it is a DATA phi */
2053 env.curr_id_2_memop[pos] = phiop;
2055 DB((dbg, LEVEL_3, "Created new %+F on merging value for address %+F\n", phi, first->value.address));
2060 /* only one predecessor, simply copy the map */
2061 ir_node *pred = get_Block_cfgpred_block(bl->block, 0);
2062 block_t *pred_bl = get_block_entry(pred);
2064 rbitset_copy(env.curr_set, pred_bl->avail_out, env.rbs_size);
2066 memcpy(env.curr_id_2_memop, pred_bl->id_2_memop_avail, env.rbs_size * sizeof(bl->id_2_memop_avail[0]));
2072 /* check for partly redundant values */
2073 for (pos = rbitset_next(bl->anticL_in, 0, 1);
2075 pos = rbitset_next(bl->anticL_in, pos + 1, 1)) {
2076 memop_t *op = bl->id_2_memop_antic[pos];
2077 int have_some, all_same;
2080 if (rbitset_is_set(env.curr_set, pos)) {
2085 assert(is_Load(op->node));
2087 DB((dbg, LEVEL_3, "anticipated %+F\n", op->node));
2092 for (i = n - 1; i >= 0; --i) {
2093 ir_node *pred = get_Block_cfgpred_block(block, i);
2094 block_t *pred_bl = get_block_entry(pred);
2095 ir_mode *mode = op->value.mode;
2099 adr = phi_translate(op->value.address, block, i);
2100 DB((dbg, LEVEL_3, ".. using address %+F in pred %d\n", adr, i));
2101 e = find_address_avail(pred_bl, register_address(adr), mode);
2103 ir_node *ef_block = get_nodes_block(adr);
2104 if (! block_dominates(ef_block, pred)) {
2105 /* cannot place a copy here */
2107 DB((dbg, LEVEL_3, "%+F cannot be moved into predecessor %+F\n", op->node, pred));
2110 DB((dbg, LEVEL_3, "%+F is not available in predecessor %+F\n", op->node, pred));
2111 pred_bl->avail = NULL;
2114 if (e->value.mode != mode && !can_convert_to(e->value.mode, mode)) {
2115 /* cannot create a Phi due to different modes */
2121 DB((dbg, LEVEL_3, "%+F is available for %+F in predecessor %+F\n", e->node, op->node, pred));
2124 else if (first != e->node)
2128 if (have_some && !all_same) {
2129 ir_mode *mode = op->value.mode;
2133 NEW_ARR_A(ir_node *, in, n);
2135 for (i = n - 1; i >= 0; --i) {
2136 ir_node *pred = get_Block_cfgpred_block(block, i);
2137 block_t *pred_bl = get_block_entry(pred);
2139 if (pred_bl->avail == NULL) {
2140 /* create a new Load here and make to make it fully redundant */
2141 dbg_info *db = get_irn_dbg_info(op->node);
2142 ir_node *last_mem = find_last_memory(pred_bl);
2143 ir_node *load, *def, *adr;
2146 assert(last_mem != NULL);
2147 adr = phi_translate(op->value.address, block, i);
2148 load = new_rd_Load(db, pred, last_mem, adr, mode, cons_none);
2149 def = new_r_Proj(load, mode, pn_Load_res);
2150 DB((dbg, LEVEL_1, "Created new %+F in %+F for party redundant %+F\n", load, pred, op->node));
2152 new_op = alloc_memop(load);
2153 new_op->mem = new_r_Proj(load, mode_M, pn_Load_M);
2154 new_op->value.address = adr;
2155 new_op->value.id = op->value.id;
2156 new_op->value.mode = mode;
2157 new_op->value.value = def;
2159 new_op->projs[pn_Load_M] = new_op->mem;
2160 new_op->projs[pn_Load_res] = def;
2162 new_op->prev = pred_bl->memop_backward;
2163 if (pred_bl->memop_backward != NULL)
2164 pred_bl->memop_backward->next = new_op;
2166 pred_bl->memop_backward = new_op;
2168 if (pred_bl->memop_forward == NULL)
2169 pred_bl->memop_forward = new_op;
2171 if (get_nodes_block(last_mem) == pred) {
2172 /* We have add a new last memory op in pred block.
2173 If pred had already a last mem, reroute all memory
2175 reroute_all_mem_users(last_mem, new_op->mem);
2177 /* reroute only those memory going through the pre block */
2178 reroute_mem_through(last_mem, new_op->mem, pred);
2181 /* we added this load at the end, so it will be avail anyway */
2182 add_memop_avail(pred_bl, new_op);
2183 pred_bl->avail = new_op;
2185 in[i] = conv_to(pred_bl->avail->value.value, mode);
2187 phi = new_r_Phi(block, n, in, mode);
2188 DB((dbg, LEVEL_1, "Created new %+F in %+F for now redundant %+F\n", phi, block, op->node));
2190 phi_op = clone_memop_phi(op, phi);
2196 /* recalculate avail by gen and kill */
2197 calc_gen_kill_avail(bl);
2199 /* always update the map after gen/kill, as values might have been changed due to RAR/WAR/WAW */
2200 memcpy(bl->id_2_memop_avail, env.curr_id_2_memop, env.rbs_size * sizeof(env.curr_id_2_memop[0]));
2202 if (!rbitsets_equal(bl->avail_out, env.curr_set, env.rbs_size)) {
2203 /* the avail set has changed */
2204 rbitset_copy(bl->avail_out, env.curr_set, env.rbs_size);
2205 dump_curr(bl, "Avail_out*");
2208 dump_curr(bl, "Avail_out");
2213 * Insert Loads upwards.
2215 static void insert_Loads_upwards(void)
2220 /* recalculate antic_out and insert Loads */
2221 DB((dbg, LEVEL_2, "Inserting Loads\n"));
2225 DB((dbg, LEVEL_2, "Iteration %d:\n=========\n", i));
2229 /* over all blocks in reverse post order, skip the start block */
2230 for (bl = env.forward->forward_next; bl != NULL; bl = bl->forward_next) {
2231 need_iter |= insert_Load(bl);
2234 } while (need_iter);
2236 DB((dbg, LEVEL_2, "Finished Load inserting after %d iterations\n", i));
2237 } /* insert_Loads_upwards */
2240 * Kill unreachable control flow.
2242 * @param irg the graph to operate on
2244 static void kill_unreachable_blocks(ir_graph *irg)
2250 NEW_ARR_A(ir_node *, ins, env.max_cfg_preds);
2252 for (bl = env.forward; bl != NULL; bl = bl->forward_next) {
2253 ir_node *block = bl->block;
2256 assert(get_Block_mark(block));
2258 n = get_Block_n_cfgpreds(block);
2260 for (i = j = 0; i < n; ++i) {
2261 ir_node *pred = get_Block_cfgpred(block, i);
2267 pred_bl = get_nodes_block(skip_Proj(pred));
2268 if (! get_Block_mark(pred_bl))
2274 ir_node *phi, *next;
2276 /* some unreachable blocks detected */
2279 DB((dbg, LEVEL_1, "Killing dead block predecessors on %+F\n", block));
2281 set_irn_in(block, j, ins);
2283 /* shorten all Phi nodes */
2284 for (phi = get_Block_phis(block); phi != NULL; phi = next) {
2285 next = get_Phi_next(phi);
2287 for (i = k = 0; i < n; ++i) {
2288 ir_node *pred = get_Block_cfgpred_block(block, i);
2293 if (! get_Block_mark(pred))
2296 ins[k++] = get_Phi_pred(phi, i);
2299 exchange(phi, ins[0]);
2301 set_irn_in(phi, k, ins);
2308 /* kick keep alives */
2309 ir_node *end = get_irg_end(irg);
2310 int i, j, n = get_End_n_keepalives(end);
2312 NEW_ARR_A(ir_node *, ins, n);
2314 for (i = j = 0; i < n; ++i) {
2315 ir_node *ka = get_End_keepalive(end, i);
2323 ka_bl = get_nodes_block(skip_Proj(ka));
2324 if (get_Block_mark(ka_bl))
2328 set_End_keepalives(end, j, ins);
2332 /* this transformation do NOT invalidate the dominance */
2334 } /* kill_unreachable_blocks */
2336 int opt_ldst(ir_graph *irg)
2340 FIRM_DBG_REGISTER(dbg, "firm.opt.ldst");
2342 DB((dbg, LEVEL_1, "\nDoing Load/Store optimization on %+F\n", irg));
2344 /* we need landing pads */
2345 remove_critical_cf_edges(irg);
2347 if (get_opt_alias_analysis()) {
2348 assure_irg_entity_usage_computed(irg);
2349 assure_irp_globals_entity_usage_computed();
2352 obstack_init(&env.obst);
2353 ir_nodemap_init(&env.adr_map);
2356 env.backward = NULL;
2357 env.curr_adr_id = 0;
2359 env.max_cfg_preds = 0;
2361 env.start_bl = get_irg_start_block(irg);
2362 env.end_bl = get_irg_end_block(irg);
2363 #ifdef DEBUG_libfirm
2364 env.id_2_address = NEW_ARR_F(ir_node *, 0);
2367 assure_irg_outs(irg);
2369 ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK | IR_RESOURCE_BLOCK_MARK);
2371 /* first step: allocate block entries. Note that some blocks might be
2372 unreachable here. Using the normal walk ensures that ALL blocks are initialized. */
2373 irg_walk_graph(irg, prepare_blocks, link_phis, NULL);
2375 /* produce an inverse post-order list for the CFG: this links only reachable
2377 irg_out_block_walk(get_irg_start_block(irg), NULL, inverse_post_order, NULL);
2379 if (! get_Block_mark(env.end_bl)) {
2381 * The end block is NOT reachable due to endless loops
2382 * or no_return calls.
2383 * Place the end block last.
2384 * env.backward points to the last block in the list for this purpose.
2386 env.backward->forward_next = get_block_entry(env.end_bl);
2388 set_Block_mark(env.end_bl, 1);
2391 /* KILL unreachable blocks: these disturb the data flow analysis */
2392 kill_unreachable_blocks(irg);
2396 /* second step: find and sort all memory ops */
2397 walk_memory_irg(irg, collect_memops, NULL, NULL);
2399 #ifdef DEBUG_libfirm
2400 /* check that the backward map is correct */
2401 assert((unsigned)ARR_LEN(env.id_2_address) == env.curr_adr_id);
2404 if (env.n_mem_ops == 0) {
2409 /* create the backward links. */
2410 env.backward = NULL;
2411 irg_block_walk_graph(irg, NULL, collect_backward, NULL);
2413 /* link the end block in */
2414 bl = get_block_entry(env.end_bl);
2415 bl->backward_next = env.backward;
2418 /* check that we really start with the start / end block */
2419 assert(env.forward->block == env.start_bl);
2420 assert(env.backward->block == env.end_bl);
2422 /* create address sets: for now, only the existing addresses are allowed plus one
2423 needed for the sentinel */
2424 env.rbs_size = env.curr_adr_id + 1;
2426 /* create the current set */
2427 env.curr_set = rbitset_obstack_alloc(&env.obst, env.rbs_size);
2428 rbitset_set(env.curr_set, env.rbs_size - 1);
2429 env.curr_id_2_memop = NEW_ARR_D(memop_t *, &env.obst, env.rbs_size);
2430 memset(env.curr_id_2_memop, 0, env.rbs_size * sizeof(env.curr_id_2_memop[0]));
2432 for (bl = env.forward; bl != NULL; bl = bl->forward_next) {
2433 /* set sentinel bits */
2434 bl->avail_out = rbitset_obstack_alloc(&env.obst, env.rbs_size);
2435 rbitset_set(bl->avail_out, env.rbs_size - 1);
2437 bl->id_2_memop_avail = NEW_ARR_D(memop_t *, &env.obst, env.rbs_size);
2438 memset(bl->id_2_memop_avail, 0, env.rbs_size * sizeof(bl->id_2_memop_avail[0]));
2440 bl->anticL_in = rbitset_obstack_alloc(&env.obst, env.rbs_size);
2441 rbitset_set(bl->anticL_in, env.rbs_size - 1);
2443 bl->id_2_memop_antic = NEW_ARR_D(memop_t *, &env.obst, env.rbs_size);
2444 memset(bl->id_2_memop_antic, 0, env.rbs_size * sizeof(bl->id_2_memop_antic[0]));
2447 (void) dump_block_list;
2452 insert_Loads_upwards();
2455 /* over all blocks in reverse post order */
2456 for (bl = env.forward; bl != NULL; bl = bl->forward_next) {
2457 do_replacements(bl);
2460 /* not only invalidate but free them. We might allocate new out arrays
2461 on our obstack which will be deleted yet. */
2463 set_irg_entity_usage_state(irg, ir_entity_usage_not_computed);
2467 ir_free_resources(irg, IR_RESOURCE_IRN_LINK | IR_RESOURCE_BLOCK_MARK);
2468 ir_nodemap_destroy(&env.adr_map);
2469 obstack_free(&env.obst, NULL);
2471 #ifdef DEBUG_libfirm
2472 DEL_ARR_F(env.id_2_address);
2475 return env.changed != 0;
2478 ir_graph_pass_t *opt_ldst_pass(const char *name)
2480 return def_graph_pass_ret(name ? name : "ldst_df", opt_ldst);
2481 } /* opt_ldst_pass */