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 Load/Store optimizations.
23 * @author Michael Beck
29 #include "iroptimize.h"
31 #include "irgraph_t.h"
39 #include "dbginfo_t.h"
40 #include "iropt_dbg.h"
47 #include "irnodehashmap.h"
52 #include "opt_manage.h"
54 /** The debug handle. */
55 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
58 #define IMAX(a,b) ((a) > (b) ? (a) : (b))
60 #define MAX_PROJ IMAX(IMAX((long)pn_Load_max, (long)pn_Store_max), (long)pn_Call_max)
63 DF_CHANGED = 1, /**< data flow changed */
64 CF_CHANGED = 2, /**< control flow changed */
70 typedef struct walk_env_t {
71 struct obstack obst; /**< list of all stores */
72 unsigned changes; /**< a bitmask of graph changes */
75 /** A Load/Store info. */
76 typedef struct ldst_info_t {
77 ir_node *projs[MAX_PROJ+1]; /**< list of Proj's of this node */
78 ir_node *exc_block; /**< the exception block if available */
79 int exc_idx; /**< predecessor index in the exception block */
80 unsigned visited; /**< visited counter for breaking loops */
84 * flags for control flow.
87 BLOCK_HAS_COND = 1, /**< Block has conditional control flow */
88 BLOCK_HAS_EXC = 2 /**< Block has exceptional control flow */
94 typedef struct block_info_t {
95 unsigned flags; /**< flags for the block */
98 /** the master visited flag for loop detection. */
99 static unsigned master_visited = 0;
101 #define INC_MASTER() ++master_visited
102 #define MARK_NODE(info) (info)->visited = master_visited
103 #define NODE_VISITED(info) (info)->visited >= master_visited
106 * get the Load/Store info of a node
108 static ldst_info_t *get_ldst_info(ir_node *node, struct obstack *obst)
110 ldst_info_t *info = (ldst_info_t*)get_irn_link(node);
113 info = OALLOCZ(obst, ldst_info_t);
114 set_irn_link(node, info);
117 } /* get_ldst_info */
120 * get the Block info of a node
122 static block_info_t *get_block_info(ir_node *node, struct obstack *obst)
124 block_info_t *info = (block_info_t*)get_irn_link(node);
127 info = OALLOCZ(obst, block_info_t);
128 set_irn_link(node, info);
131 } /* get_block_info */
134 * update the projection info for a Load/Store
136 static unsigned update_projs(ldst_info_t *info, ir_node *proj)
138 long nr = get_Proj_proj(proj);
140 assert(0 <= nr && nr <= MAX_PROJ && "Wrong proj from LoadStore");
142 if (info->projs[nr]) {
143 /* there is already one, do CSE */
144 exchange(proj, info->projs[nr]);
148 info->projs[nr] = proj;
154 * update the exception block info for a Load/Store node.
156 * @param info the load/store info struct
157 * @param block the exception handler block for this load/store
158 * @param pos the control flow input of the block
160 static unsigned update_exc(ldst_info_t *info, ir_node *block, int pos)
162 assert(info->exc_block == NULL && "more than one exception block found");
164 info->exc_block = block;
169 /** Return the number of uses of an address node */
170 #define get_irn_n_uses(adr) get_irn_n_edges(adr)
173 * walker, collects all Load/Store/Proj nodes
175 * walks from Start -> End
177 static void collect_nodes(ir_node *node, void *env)
179 walk_env_t *wenv = (walk_env_t *)env;
180 unsigned opcode = get_irn_opcode(node);
181 ir_node *pred, *blk, *pred_blk;
182 ldst_info_t *ldst_info;
184 if (opcode == iro_Proj) {
185 pred = get_Proj_pred(node);
186 opcode = get_irn_opcode(pred);
188 if (opcode == iro_Load || opcode == iro_Store || opcode == iro_Call) {
189 ldst_info = get_ldst_info(pred, &wenv->obst);
191 wenv->changes |= update_projs(ldst_info, node);
194 * Place the Proj's to the same block as the
195 * predecessor Load. This is always ok and prevents
196 * "non-SSA" form after optimizations if the Proj
197 * is in a wrong block.
199 blk = get_nodes_block(node);
200 pred_blk = get_nodes_block(pred);
201 if (blk != pred_blk) {
202 wenv->changes |= DF_CHANGED;
203 set_nodes_block(node, pred_blk);
206 } else if (opcode == iro_Block) {
209 for (i = get_Block_n_cfgpreds(node) - 1; i >= 0; --i) {
210 ir_node *pred_block, *proj;
211 block_info_t *bl_info;
214 pred = proj = get_Block_cfgpred(node, i);
217 pred = get_Proj_pred(proj);
218 is_exc = is_x_except_Proj(proj);
221 /* ignore Bad predecessors, they will be removed later */
225 pred_block = get_nodes_block(pred);
226 bl_info = get_block_info(pred_block, &wenv->obst);
228 if (is_fragile_op(pred) && is_exc)
229 bl_info->flags |= BLOCK_HAS_EXC;
230 else if (is_irn_forking(pred))
231 bl_info->flags |= BLOCK_HAS_COND;
233 opcode = get_irn_opcode(pred);
234 if (is_exc && (opcode == iro_Load || opcode == iro_Store || opcode == iro_Call)) {
235 ldst_info = get_ldst_info(pred, &wenv->obst);
237 wenv->changes |= update_exc(ldst_info, node, i);
241 } /* collect_nodes */
244 * Returns an entity if the address ptr points to a constant one.
246 * @param ptr the address
248 * @return an entity or NULL
250 static ir_entity *find_constant_entity(ir_node *ptr)
253 if (is_SymConst(ptr) && get_SymConst_kind(ptr) == symconst_addr_ent) {
254 return get_SymConst_entity(ptr);
255 } else if (is_Sel(ptr)) {
256 ir_entity *ent = get_Sel_entity(ptr);
257 ir_type *tp = get_entity_owner(ent);
259 /* Do not fiddle with polymorphism. */
260 if (is_Class_type(get_entity_owner(ent)) &&
261 ((get_entity_n_overwrites(ent) != 0) ||
262 (get_entity_n_overwrittenby(ent) != 0) ) )
265 if (is_Array_type(tp)) {
269 for (i = 0, n = get_Sel_n_indexs(ptr); i < n; ++i) {
271 ir_tarval *tlower, *tupper;
272 ir_node *index = get_Sel_index(ptr, i);
273 ir_tarval *tv = computed_value(index);
275 /* check if the index is constant */
276 if (tv == tarval_bad)
279 bound = get_array_lower_bound(tp, i);
280 tlower = computed_value(bound);
281 bound = get_array_upper_bound(tp, i);
282 tupper = computed_value(bound);
284 if (tlower == tarval_bad || tupper == tarval_bad)
287 if (tarval_cmp(tv, tlower) == ir_relation_less)
289 if (tarval_cmp(tupper, tv) == ir_relation_less)
292 /* ok, bounds check finished */
296 if (get_entity_linkage(ent) & IR_LINKAGE_CONSTANT)
300 ptr = get_Sel_ptr(ptr);
301 } else if (is_Add(ptr)) {
302 ir_node *l = get_Add_left(ptr);
303 ir_node *r = get_Add_right(ptr);
305 if (get_irn_mode(l) == get_irn_mode(ptr) && is_Const(r))
307 else if (get_irn_mode(r) == get_irn_mode(ptr) && is_Const(l))
312 /* for now, we support only one addition, reassoc should fold all others */
313 if (! is_SymConst(ptr) && !is_Sel(ptr))
315 } else if (is_Sub(ptr)) {
316 ir_node *l = get_Sub_left(ptr);
317 ir_node *r = get_Sub_right(ptr);
319 if (get_irn_mode(l) == get_irn_mode(ptr) && is_Const(r))
323 /* for now, we support only one substraction, reassoc should fold all others */
324 if (! is_SymConst(ptr) && !is_Sel(ptr))
329 } /* find_constant_entity */
332 * Return the Selection index of a Sel node from dimension n
334 static long get_Sel_array_index_long(ir_node *n, int dim)
336 ir_node *index = get_Sel_index(n, dim);
337 assert(is_Const(index));
338 return get_tarval_long(get_Const_tarval(index));
339 } /* get_Sel_array_index_long */
341 typedef struct path_entry {
343 struct path_entry *next;
347 static ir_node *rec_find_compound_ent_value(ir_node *ptr, path_entry *next)
349 path_entry entry, *p;
350 ir_entity *ent, *field;
351 ir_initializer_t *initializer;
357 if (is_SymConst(ptr)) {
359 ent = get_SymConst_entity(ptr);
360 initializer = get_entity_initializer(ent);
361 for (p = next; p != NULL;) {
362 if (initializer->kind != IR_INITIALIZER_COMPOUND)
364 n = get_initializer_compound_n_entries(initializer);
365 tp = get_entity_type(ent);
367 if (is_Array_type(tp)) {
368 ent = get_array_element_entity(tp);
373 initializer = get_initializer_compound_value(initializer, 0);
379 initializer = get_initializer_compound_value(initializer, p->index);
384 tp = get_entity_type(ent);
385 while (is_Array_type(tp)) {
386 ent = get_array_element_entity(tp);
387 tp = get_entity_type(ent);
389 n = get_initializer_compound_n_entries(initializer);
392 initializer = get_initializer_compound_value(initializer, 0);
395 switch (initializer->kind) {
396 case IR_INITIALIZER_CONST:
397 return get_initializer_const_value(initializer);
398 case IR_INITIALIZER_TARVAL:
399 case IR_INITIALIZER_NULL:
403 } else if (is_Sel(ptr)) {
404 entry.ent = field = get_Sel_entity(ptr);
405 tp = get_entity_owner(field);
406 if (is_Array_type(tp)) {
407 assert(get_Sel_n_indexs(ptr) == 1 && "multi dim arrays not implemented");
408 entry.index = get_Sel_array_index_long(ptr, 0) - get_array_lower_bound_int(tp, 0);
410 size_t i, n_members = get_compound_n_members(tp);
411 for (i = 0; i < n_members; ++i) {
412 if (get_compound_member(tp, i) == field)
415 if (i >= n_members) {
416 /* not found: should NOT happen */
421 return rec_find_compound_ent_value(get_Sel_ptr(ptr), &entry);
422 } else if (is_Add(ptr)) {
427 ir_node *l = get_Add_left(ptr);
428 ir_node *r = get_Add_right(ptr);
431 tv = get_Const_tarval(r);
434 tv = get_Const_tarval(l);
438 mode = get_tarval_mode(tv);
440 /* ptr must be a Sel or a SymConst, this was checked in find_constant_entity() */
442 field = get_Sel_entity(ptr);
444 field = get_SymConst_entity(ptr);
447 /* count needed entries */
449 for (ent = field;;) {
450 tp = get_entity_type(ent);
451 if (! is_Array_type(tp))
453 ent = get_array_element_entity(tp);
456 /* should be at least ONE entry */
460 /* allocate the right number of entries */
461 NEW_ARR_A(path_entry, p, pos);
465 for (ent = field;;) {
467 ir_tarval *sz, *tv_index, *tlower, *tupper;
471 tp = get_entity_type(ent);
472 if (! is_Array_type(tp))
474 ent = get_array_element_entity(tp);
476 p[pos].next = &p[pos + 1];
478 size = get_type_size_bytes(get_entity_type(ent));
479 sz = new_tarval_from_long(size, mode);
481 tv_index = tarval_div(tv, sz);
482 tv = tarval_mod(tv, sz);
484 if (tv_index == tarval_bad || tv == tarval_bad)
487 assert(get_array_n_dimensions(tp) == 1 && "multiarrays not implemented");
488 bound = get_array_lower_bound(tp, 0);
489 tlower = computed_value(bound);
490 bound = get_array_upper_bound(tp, 0);
491 tupper = computed_value(bound);
493 if (tlower == tarval_bad || tupper == tarval_bad)
496 if (tarval_cmp(tv_index, tlower) == ir_relation_less)
498 if (tarval_cmp(tupper, tv_index) == ir_relation_less)
501 /* ok, bounds check finished */
502 index = get_tarval_long(tv_index);
503 p[pos].index = index;
506 if (! tarval_is_null(tv)) {
507 /* hmm, wrong access */
510 p[pos - 1].next = next;
511 return rec_find_compound_ent_value(ptr, p);
512 } else if (is_Sub(ptr)) {
513 ir_node *l = get_Sub_left(ptr);
514 ir_node *r = get_Sub_right(ptr);
517 tv = get_Const_tarval(r);
524 static ir_node *find_compound_ent_value(ir_node *ptr)
526 return rec_find_compound_ent_value(ptr, NULL);
530 static void reduce_adr_usage(ir_node *ptr);
533 * Update a Load that may have lost its users.
535 static void handle_load_update(ir_node *load)
537 ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
539 /* do NOT touch volatile loads for now */
540 if (get_Load_volatility(load) == volatility_is_volatile)
543 if (! info->projs[pn_Load_res] && ! info->projs[pn_Load_X_except]) {
544 ir_node *ptr = get_Load_ptr(load);
545 ir_node *mem = get_Load_mem(load);
547 /* a Load whose value is neither used nor exception checked, remove it */
548 exchange(info->projs[pn_Load_M], mem);
549 if (info->projs[pn_Load_X_regular])
550 exchange(info->projs[pn_Load_X_regular], new_r_Jmp(get_nodes_block(load)));
552 reduce_adr_usage(ptr);
554 } /* handle_load_update */
557 * A use of an address node has vanished. Check if this was a Proj
558 * node and update the counters.
560 static void reduce_adr_usage(ir_node *ptr)
565 if (get_irn_n_edges(ptr) > 0)
568 /* this Proj is dead now */
569 pred = get_Proj_pred(ptr);
571 ldst_info_t *info = (ldst_info_t*)get_irn_link(pred);
572 info->projs[get_Proj_proj(ptr)] = NULL;
574 /* this node lost its result proj, handle that */
575 handle_load_update(pred);
577 } /* reduce_adr_usage */
580 * Check, if an already existing value of mode old_mode can be converted
581 * into the needed one new_mode without loss.
583 static int can_use_stored_value(ir_mode *old_mode, ir_mode *new_mode)
587 if (old_mode == new_mode)
590 old_size = get_mode_size_bits(old_mode);
591 new_size = get_mode_size_bits(new_mode);
593 /* if both modes are two-complement ones, we can always convert the
594 Stored value into the needed one. (on big endian machines we currently
595 only support this for modes of same size) */
596 if (old_size >= new_size &&
597 get_mode_arithmetic(old_mode) == irma_twos_complement &&
598 get_mode_arithmetic(new_mode) == irma_twos_complement &&
599 (!be_get_backend_param()->byte_order_big_endian
600 || old_size == new_size)) {
607 * Check whether a Call is at least pure, i.e. does only read memory.
609 static unsigned is_Call_pure(ir_node *call)
611 ir_type *call_tp = get_Call_type(call);
612 unsigned prop = get_method_additional_properties(call_tp);
614 /* check first the call type */
615 if ((prop & (mtp_property_const|mtp_property_pure)) == 0) {
616 /* try the called entity */
617 ir_node *ptr = get_Call_ptr(call);
619 if (is_SymConst_addr_ent(ptr)) {
620 ir_entity *ent = get_SymConst_entity(ptr);
622 prop = get_entity_additional_properties(ent);
625 return (prop & (mtp_property_const|mtp_property_pure)) != 0;
628 static ir_node *get_base_and_offset(ir_node *ptr, long *pOffset)
630 ir_mode *mode = get_irn_mode(ptr);
633 /* TODO: long might not be enough, we should probably use some tarval thingy... */
636 ir_node *l = get_Add_left(ptr);
637 ir_node *r = get_Add_right(ptr);
639 if (get_irn_mode(l) != mode || !is_Const(r))
642 offset += get_tarval_long(get_Const_tarval(r));
644 } else if (is_Sub(ptr)) {
645 ir_node *l = get_Sub_left(ptr);
646 ir_node *r = get_Sub_right(ptr);
648 if (get_irn_mode(l) != mode || !is_Const(r))
651 offset -= get_tarval_long(get_Const_tarval(r));
653 } else if (is_Sel(ptr)) {
654 ir_entity *ent = get_Sel_entity(ptr);
655 ir_type *tp = get_entity_owner(ent);
657 if (is_Array_type(tp)) {
661 /* only one dimensional arrays yet */
662 if (get_Sel_n_indexs(ptr) != 1)
664 index = get_Sel_index(ptr, 0);
665 if (! is_Const(index))
668 tp = get_entity_type(ent);
669 if (get_type_state(tp) != layout_fixed)
672 size = get_type_size_bytes(tp);
673 offset += size * get_tarval_long(get_Const_tarval(index));
675 if (get_type_state(tp) != layout_fixed)
677 offset += get_entity_offset(ent);
679 ptr = get_Sel_ptr(ptr);
688 static int try_load_after_store(ir_node *load,
689 ir_node *load_base_ptr, long load_offset, ir_node *store)
692 ir_node *store_ptr = get_Store_ptr(store);
694 ir_node *store_base_ptr = get_base_and_offset(store_ptr, &store_offset);
695 ir_node *store_value;
704 if (load_base_ptr != store_base_ptr)
707 load_mode = get_Load_mode(load);
708 load_mode_len = get_mode_size_bytes(load_mode);
709 store_mode = get_irn_mode(get_Store_value(store));
710 store_mode_len = get_mode_size_bytes(store_mode);
711 delta = load_offset - store_offset;
712 store_value = get_Store_value(store);
714 if (delta != 0 || store_mode != load_mode) {
715 /* TODO: implement for big-endian */
716 if (delta < 0 || delta + load_mode_len > store_mode_len
717 || (be_get_backend_param()->byte_order_big_endian
718 && load_mode_len != store_mode_len))
721 if (get_mode_arithmetic(store_mode) != irma_twos_complement ||
722 get_mode_arithmetic(load_mode) != irma_twos_complement)
726 /* produce a shift to adjust offset delta */
729 ir_graph *irg = get_irn_irg(load);
731 cnst = new_r_Const_long(irg, mode_Iu, delta * 8);
732 store_value = new_r_Shr(get_nodes_block(load),
733 store_value, cnst, store_mode);
736 /* add an convert if needed */
737 if (store_mode != load_mode) {
738 store_value = new_r_Conv(get_nodes_block(load), store_value, load_mode);
742 DBG_OPT_RAW(load, store_value);
744 info = (ldst_info_t*)get_irn_link(load);
745 if (info->projs[pn_Load_M])
746 exchange(info->projs[pn_Load_M], get_Load_mem(load));
750 if (info->projs[pn_Load_X_except]) {
751 ir_graph *irg = get_irn_irg(load);
752 exchange( info->projs[pn_Load_X_except], new_r_Bad(irg, mode_X));
755 if (info->projs[pn_Load_X_regular]) {
756 exchange( info->projs[pn_Load_X_regular], new_r_Jmp(get_nodes_block(load)));
760 if (info->projs[pn_Load_res])
761 exchange(info->projs[pn_Load_res], store_value);
763 load_ptr = get_Load_ptr(load);
765 reduce_adr_usage(load_ptr);
766 return res | DF_CHANGED;
770 * Follow the memory chain as long as there are only Loads,
771 * alias free Stores, and constant Calls and try to replace the
772 * current Load by a previous ones.
773 * Note that in unreachable loops it might happen that we reach
774 * load again, as well as we can fall into a cycle.
775 * We break such cycles using a special visited flag.
777 * INC_MASTER() must be called before dive into
779 static unsigned follow_Mem_chain(ir_node *load, ir_node *curr)
782 ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
784 ir_node *ptr = get_Load_ptr(load);
785 ir_node *mem = get_Load_mem(load);
786 ir_mode *load_mode = get_Load_mode(load);
788 for (pred = curr; load != pred; ) {
789 ldst_info_t *pred_info = (ldst_info_t*)get_irn_link(pred);
792 * a Load immediately after a Store -- a read after write.
793 * We may remove the Load, if both Load & Store does not have an
794 * exception handler OR they are in the same Block. In the latter
795 * case the Load cannot throw an exception when the previous Store was
798 * Why we need to check for Store Exception? If the Store cannot
799 * be executed (ROM) the exception handler might simply jump into
801 * We could make it a little bit better if we would know that the
802 * exception handler of the Store jumps directly to the end...
804 if (is_Store(pred) && ((pred_info->projs[pn_Store_X_except] == NULL
805 && info->projs[pn_Load_X_except] == NULL)
806 || get_nodes_block(load) == get_nodes_block(pred)))
809 ir_node *base_ptr = get_base_and_offset(ptr, &load_offset);
810 int changes = try_load_after_store(load, base_ptr, load_offset, pred);
813 return res | changes;
814 } else if (is_Load(pred) && get_Load_ptr(pred) == ptr &&
815 can_use_stored_value(get_Load_mode(pred), load_mode)) {
817 * a Load after a Load -- a read after read.
818 * We may remove the second Load, if it does not have an exception
819 * handler OR they are in the same Block. In the later case
820 * the Load cannot throw an exception when the previous Load was
823 * Here, there is no need to check if the previous Load has an
824 * exception hander because they would have exact the same
827 * TODO: implement load-after-load with different mode for big
830 if (info->projs[pn_Load_X_except] == NULL
831 || get_nodes_block(load) == get_nodes_block(pred)) {
834 DBG_OPT_RAR(load, pred);
836 /* the result is used */
837 if (info->projs[pn_Load_res]) {
838 if (pred_info->projs[pn_Load_res] == NULL) {
839 /* create a new Proj again */
840 pred_info->projs[pn_Load_res] = new_r_Proj(pred, get_Load_mode(pred), pn_Load_res);
842 value = pred_info->projs[pn_Load_res];
844 /* add an convert if needed */
845 if (get_Load_mode(pred) != load_mode) {
846 value = new_r_Conv(get_nodes_block(load), value, load_mode);
849 exchange(info->projs[pn_Load_res], value);
852 if (info->projs[pn_Load_M])
853 exchange(info->projs[pn_Load_M], mem);
856 if (info->projs[pn_Load_X_except]) {
857 ir_graph *irg = get_irn_irg(load);
858 exchange(info->projs[pn_Load_X_except], new_r_Bad(irg, mode_X));
861 if (info->projs[pn_Load_X_regular]) {
862 exchange( info->projs[pn_Load_X_regular], new_r_Jmp(get_nodes_block(load)));
867 reduce_adr_usage(ptr);
868 return res |= DF_CHANGED;
872 if (is_Store(pred)) {
873 /* check if we can pass through this store */
874 ir_alias_relation rel = get_alias_relation(
876 get_irn_mode(get_Store_value(pred)),
878 /* if the might be an alias, we cannot pass this Store */
879 if (rel != ir_no_alias)
881 pred = skip_Proj(get_Store_mem(pred));
882 } else if (is_Load(pred)) {
883 pred = skip_Proj(get_Load_mem(pred));
884 } else if (is_Call(pred)) {
885 if (is_Call_pure(pred)) {
886 /* The called graph is at least pure, so there are no Store's
887 in it. We can handle it like a Load and skip it. */
888 pred = skip_Proj(get_Call_mem(pred));
890 /* there might be Store's in the graph, stop here */
894 /* follow only Load chains */
898 /* check for cycles */
899 if (NODE_VISITED(pred_info))
901 MARK_NODE(pred_info);
907 /* handle all Sync predecessors */
908 for (i = get_Sync_n_preds(pred) - 1; i >= 0; --i) {
909 res |= follow_Mem_chain(load, skip_Proj(get_Sync_pred(pred, i)));
916 } /* follow_Mem_chain */
919 * Check if we can replace the load by a given const from
920 * the const code irg.
922 ir_node *can_replace_load_by_const(const ir_node *load, ir_node *c)
924 ir_mode *c_mode = get_irn_mode(c);
925 ir_mode *l_mode = get_Load_mode(load);
926 ir_node *block = get_nodes_block(load);
927 dbg_info *dbgi = get_irn_dbg_info(load);
928 ir_node *res = copy_const_value(dbgi, c, block);
930 if (c_mode != l_mode) {
931 /* check, if the mode matches OR can be easily converted info */
932 if (is_reinterpret_cast(c_mode, l_mode)) {
933 /* copy the value from the const code irg and cast it */
934 res = new_rd_Conv(dbgi, block, res, l_mode);
945 * @param load the Load node
947 static unsigned optimize_load(ir_node *load)
949 ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
950 ir_node *mem, *ptr, *value;
955 /* do NOT touch volatile loads for now */
956 if (get_Load_volatility(load) == volatility_is_volatile)
959 /* the address of the load to be optimized */
960 ptr = get_Load_ptr(load);
962 /* The mem of the Load. Must still be returned after optimization. */
963 mem = get_Load_mem(load);
965 if (info->projs[pn_Load_res] == NULL
966 && info->projs[pn_Load_X_except] == NULL) {
967 /* the value is never used and we don't care about exceptions, remove */
968 exchange(info->projs[pn_Load_M], mem);
970 if (info->projs[pn_Load_X_regular]) {
971 /* should not happen, but if it does, remove it */
972 exchange(info->projs[pn_Load_X_regular], new_r_Jmp(get_nodes_block(load)));
976 reduce_adr_usage(ptr);
977 return res | DF_CHANGED;
981 /* check if we can determine the entity that will be loaded */
982 ent = find_constant_entity(ptr);
984 && get_entity_visibility(ent) != ir_visibility_external) {
985 /* a static allocation that is not external: there should be NO
986 * exception when loading even if we cannot replace the load itself.
989 /* no exception, clear the info field as it might be checked later again */
990 if (info->projs[pn_Load_X_except]) {
991 ir_graph *irg = get_irn_irg(load);
992 exchange(info->projs[pn_Load_X_except], new_r_Bad(irg, mode_X));
993 info->projs[pn_Load_X_except] = NULL;
996 if (info->projs[pn_Load_X_regular]) {
997 exchange(info->projs[pn_Load_X_regular], new_r_Jmp(get_nodes_block(load)));
998 info->projs[pn_Load_X_regular] = NULL;
1002 if (get_entity_linkage(ent) & IR_LINKAGE_CONSTANT) {
1003 if (has_entity_initializer(ent)) {
1004 /* new style initializer */
1005 value = find_compound_ent_value(ptr);
1007 if (value != NULL) {
1008 ir_graph *irg = get_irn_irg(load);
1009 value = can_replace_load_by_const(load, value);
1010 if (value != NULL && is_Sel(ptr)) {
1011 /* frontend has inserted masking operations after bitfield accesses,
1012 * so we might have to shift the const. */
1013 unsigned char bit_offset = get_entity_offset_bits_remainder(get_Sel_entity(ptr));
1014 ir_tarval *tv_old = get_Const_tarval(value);
1015 ir_tarval *tv_offset = new_tarval_from_long(bit_offset, mode_Bu);
1016 ir_tarval *tv_new = tarval_shl(tv_old, tv_offset);
1017 value = new_r_Const(irg, tv_new);
1022 if (value != NULL) {
1023 /* we completely replace the load by this value */
1024 if (info->projs[pn_Load_X_except]) {
1025 ir_graph *irg = get_irn_irg(load);
1026 exchange(info->projs[pn_Load_X_except], new_r_Bad(irg, mode_X));
1027 info->projs[pn_Load_X_except] = NULL;
1030 if (info->projs[pn_Load_X_regular]) {
1031 exchange(info->projs[pn_Load_X_regular], new_r_Jmp(get_nodes_block(load)));
1032 info->projs[pn_Load_X_regular] = NULL;
1035 if (info->projs[pn_Load_M]) {
1036 exchange(info->projs[pn_Load_M], mem);
1039 if (info->projs[pn_Load_res]) {
1040 exchange(info->projs[pn_Load_res], value);
1044 reduce_adr_usage(ptr);
1048 /* Check, if the address of this load is used more than once.
1049 * If not, more load cannot be removed in any case. */
1050 if (get_irn_n_uses(ptr) <= 1 && get_irn_n_uses(get_base_and_offset(ptr, &dummy)) <= 1)
1054 * follow the memory chain as long as there are only Loads
1055 * and try to replace current Load or Store by a previous one.
1056 * Note that in unreachable loops it might happen that we reach
1057 * load again, as well as we can fall into a cycle.
1058 * We break such cycles using a special visited flag.
1061 res = follow_Mem_chain(load, skip_Proj(mem));
1063 } /* optimize_load */
1066 * Check whether a value of mode new_mode would completely overwrite a value
1067 * of mode old_mode in memory.
1069 static int is_completely_overwritten(ir_mode *old_mode, ir_mode *new_mode)
1071 return get_mode_size_bits(new_mode) >= get_mode_size_bits(old_mode);
1072 } /* is_completely_overwritten */
1075 * Check whether small is a part of large (starting at same address).
1077 static int is_partially_same(ir_node *small, ir_node *large)
1079 ir_mode *sm = get_irn_mode(small);
1080 ir_mode *lm = get_irn_mode(large);
1082 /* FIXME: Check endianness */
1083 return is_Conv(small) && get_Conv_op(small) == large
1084 && get_mode_size_bytes(sm) < get_mode_size_bytes(lm)
1085 && get_mode_arithmetic(sm) == irma_twos_complement
1086 && get_mode_arithmetic(lm) == irma_twos_complement;
1087 } /* is_partially_same */
1090 * follow the memory chain as long as there are only Loads and alias free Stores.
1092 * INC_MASTER() must be called before dive into
1094 static unsigned follow_Mem_chain_for_Store(ir_node *store, ir_node *curr)
1097 ldst_info_t *info = (ldst_info_t*)get_irn_link(store);
1099 ir_node *ptr = get_Store_ptr(store);
1100 ir_node *mem = get_Store_mem(store);
1101 ir_node *value = get_Store_value(store);
1102 ir_mode *mode = get_irn_mode(value);
1103 ir_node *block = get_nodes_block(store);
1105 for (pred = curr; pred != store;) {
1106 ldst_info_t *pred_info = (ldst_info_t*)get_irn_link(pred);
1109 * BEWARE: one might think that checking the modes is useless, because
1110 * if the pointers are identical, they refer to the same object.
1111 * This is only true in strong typed languages, not is C were the following
1112 * is possible *(ir_type1 *)p = a; *(ir_type2 *)p = b ...
1113 * However, if the size of the mode that is written is bigger or equal the
1114 * size of the old one, the old value is completely overwritten and can be
1117 if (is_Store(pred) && get_Store_ptr(pred) == ptr &&
1118 get_nodes_block(pred) == block) {
1120 * a Store after a Store in the same Block -- a write after write.
1124 * We may remove the first Store, if the old value is completely
1125 * overwritten or the old value is a part of the new value,
1126 * and if it does not have an exception handler.
1128 * TODO: What, if both have the same exception handler ???
1130 if (get_Store_volatility(pred) != volatility_is_volatile
1131 && !pred_info->projs[pn_Store_X_except]) {
1132 ir_node *predvalue = get_Store_value(pred);
1133 ir_mode *predmode = get_irn_mode(predvalue);
1135 if (is_completely_overwritten(predmode, mode)
1136 || is_partially_same(predvalue, value)) {
1137 DBG_OPT_WAW(pred, store);
1138 exchange(pred_info->projs[pn_Store_M], get_Store_mem(pred));
1140 reduce_adr_usage(ptr);
1146 * We may remove the Store, if the old value already contains
1147 * the new value, and if it does not have an exception handler.
1149 * TODO: What, if both have the same exception handler ???
1151 if (get_Store_volatility(store) != volatility_is_volatile
1152 && !info->projs[pn_Store_X_except]) {
1153 ir_node *predvalue = get_Store_value(pred);
1155 if (is_partially_same(value, predvalue)) {
1156 DBG_OPT_WAW(pred, store);
1157 exchange(info->projs[pn_Store_M], mem);
1159 reduce_adr_usage(ptr);
1163 } else if (is_Load(pred) && get_Load_ptr(pred) == ptr &&
1164 value == pred_info->projs[pn_Load_res]) {
1166 * a Store of a value just loaded from the same address
1167 * -- a write after read.
1168 * We may remove the Store, if it does not have an exception
1171 if (! info->projs[pn_Store_X_except]) {
1172 DBG_OPT_WAR(store, pred);
1173 exchange(info->projs[pn_Store_M], mem);
1175 reduce_adr_usage(ptr);
1180 if (is_Store(pred)) {
1181 /* check if we can pass through this store */
1182 ir_alias_relation rel = get_alias_relation(
1183 get_Store_ptr(pred),
1184 get_irn_mode(get_Store_value(pred)),
1186 /* if the might be an alias, we cannot pass this Store */
1187 if (rel != ir_no_alias)
1189 pred = skip_Proj(get_Store_mem(pred));
1190 } else if (is_Load(pred)) {
1191 ir_alias_relation rel = get_alias_relation(
1192 get_Load_ptr(pred), get_Load_mode(pred),
1194 if (rel != ir_no_alias)
1197 pred = skip_Proj(get_Load_mem(pred));
1199 /* follow only Load chains */
1203 /* check for cycles */
1204 if (NODE_VISITED(pred_info))
1206 MARK_NODE(pred_info);
1209 if (is_Sync(pred)) {
1212 /* handle all Sync predecessors */
1213 for (i = get_Sync_n_preds(pred) - 1; i >= 0; --i) {
1214 res |= follow_Mem_chain_for_Store(store, skip_Proj(get_Sync_pred(pred, i)));
1220 } /* follow_Mem_chain_for_Store */
1222 /** find entity used as base for an address calculation */
1223 static ir_entity *find_entity(ir_node *ptr)
1225 switch (get_irn_opcode(ptr)) {
1227 return get_SymConst_entity(ptr);
1229 ir_node *pred = get_Sel_ptr(ptr);
1230 if (get_irg_frame(get_irn_irg(ptr)) == pred)
1231 return get_Sel_entity(ptr);
1233 return find_entity(pred);
1237 ir_node *left = get_binop_left(ptr);
1239 if (mode_is_reference(get_irn_mode(left)))
1240 return find_entity(left);
1241 right = get_binop_right(ptr);
1242 if (mode_is_reference(get_irn_mode(right)))
1243 return find_entity(right);
1254 * @param store the Store node
1256 static unsigned optimize_store(ir_node *store)
1262 if (get_Store_volatility(store) == volatility_is_volatile)
1265 ptr = get_Store_ptr(store);
1266 entity = find_entity(ptr);
1268 /* a store to an entity which is never read is unnecessary */
1269 if (entity != NULL && !(get_entity_usage(entity) & ir_usage_read)) {
1270 ldst_info_t *info = (ldst_info_t*)get_irn_link(store);
1271 if (info->projs[pn_Store_X_except] == NULL) {
1272 DB((dbg, LEVEL_1, " Killing useless %+F to never read entity %+F\n", store, entity));
1273 exchange(info->projs[pn_Store_M], get_Store_mem(store));
1275 reduce_adr_usage(ptr);
1280 /* Check, if the address of this Store is used more than once.
1281 * If not, this Store cannot be removed in any case. */
1282 if (get_irn_n_uses(ptr) <= 1)
1285 mem = get_Store_mem(store);
1287 /* follow the memory chain as long as there are only Loads */
1290 return follow_Mem_chain_for_Store(store, skip_Proj(mem));
1291 } /* optimize_store */
1294 * walker, optimizes Phi after Stores to identical places:
1295 * Does the following optimization:
1298 * val1 val2 val3 val1 val2 val3
1300 * Store Store Store \ | /
1307 * This reduces the number of stores and allows for predicated execution.
1308 * Moves Stores back to the end of a function which may be bad.
1310 * This is only possible if the predecessor blocks have only one successor.
1312 static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
1315 ir_node *store, *ptr, *block, *phi_block, *phiM, *phiD, *exc, *projM;
1320 ir_node **inM, **inD, **projMs;
1322 dbg_info *db = NULL;
1324 block_info_t *bl_info;
1327 /* Must be a memory Phi */
1328 if (get_irn_mode(phi) != mode_M)
1331 n = get_Phi_n_preds(phi);
1335 /* must be only one user */
1336 projM = get_Phi_pred(phi, 0);
1337 if (get_irn_n_edges(projM) != 1)
1340 store = skip_Proj(projM);
1344 if (!is_Store(store))
1347 block = get_nodes_block(store);
1349 /* check if the block is post dominated by Phi-block
1350 and has no exception exit */
1351 bl_info = (block_info_t*)get_irn_link(block);
1352 if (bl_info->flags & BLOCK_HAS_EXC)
1355 phi_block = get_nodes_block(phi);
1356 if (! block_strictly_postdominates(phi_block, block))
1359 /* this is the address of the store */
1360 ptr = get_Store_ptr(store);
1361 mode = get_irn_mode(get_Store_value(store));
1362 info = (ldst_info_t*)get_irn_link(store);
1363 exc = info->exc_block;
1365 for (i = 1; i < n; ++i) {
1366 ir_node *pred = get_Phi_pred(phi, i);
1368 if (get_irn_n_edges(pred) != 1)
1371 pred = skip_Proj(pred);
1372 if (!is_Store(pred))
1375 if (ptr != get_Store_ptr(pred) || mode != get_irn_mode(get_Store_value(pred)))
1378 info = (ldst_info_t*)get_irn_link(pred);
1380 /* check, if all stores have the same exception flow */
1381 if (exc != info->exc_block)
1384 block = get_nodes_block(pred);
1386 /* check if the block is post dominated by Phi-block
1387 and has no exception exit. Note that block must be different from
1388 Phi-block, else we would move a Store from end End of a block to its
1390 bl_info = (block_info_t*)get_irn_link(block);
1391 if (bl_info->flags & BLOCK_HAS_EXC)
1393 if (block == phi_block || ! block_postdominates(phi_block, block))
1398 * ok, when we are here, we found all predecessors of a Phi that
1399 * are Stores to the same address and size. That means whatever
1400 * we do before we enter the block of the Phi, we do a Store.
1401 * So, we can move the Store to the current block:
1403 * val1 val2 val3 val1 val2 val3
1405 * | Str | | Str | | Str | \ | /
1411 * Is only allowed if the predecessor blocks have only one successor.
1414 NEW_ARR_A(ir_node *, projMs, n);
1415 NEW_ARR_A(ir_node *, inM, n);
1416 NEW_ARR_A(ir_node *, inD, n);
1417 NEW_ARR_A(int, idx, n);
1419 /* Prepare: Collect all Store nodes. We must do this
1420 first because we otherwise may loose a store when exchanging its
1423 for (i = n - 1; i >= 0; --i) {
1426 projMs[i] = get_Phi_pred(phi, i);
1427 assert(is_Proj(projMs[i]));
1429 store = get_Proj_pred(projMs[i]);
1430 info = (ldst_info_t*)get_irn_link(store);
1432 inM[i] = get_Store_mem(store);
1433 inD[i] = get_Store_value(store);
1434 idx[i] = info->exc_idx;
1436 block = get_nodes_block(phi);
1438 /* second step: create a new memory Phi */
1439 phiM = new_rd_Phi(get_irn_dbg_info(phi), block, n, inM, mode_M);
1441 /* third step: create a new data Phi */
1442 phiD = new_rd_Phi(get_irn_dbg_info(phi), block, n, inD, mode);
1444 /* rewire memory and kill the node */
1445 for (i = n - 1; i >= 0; --i) {
1446 ir_node *proj = projMs[i];
1448 if (is_Proj(proj)) {
1449 ir_node *store = get_Proj_pred(proj);
1450 exchange(proj, inM[i]);
1455 /* fourth step: create the Store */
1456 store = new_rd_Store(db, block, phiM, ptr, phiD, cons_none);
1458 co_set_irn_name(store, co_get_irn_ident(old_store));
1461 projM = new_rd_Proj(NULL, store, mode_M, pn_Store_M);
1463 info = get_ldst_info(store, &wenv->obst);
1464 info->projs[pn_Store_M] = projM;
1466 /* fifths step: repair exception flow */
1468 ir_node *projX = new_rd_Proj(NULL, store, mode_X, pn_Store_X_except);
1470 info->projs[pn_Store_X_except] = projX;
1471 info->exc_block = exc;
1472 info->exc_idx = idx[0];
1474 for (i = 0; i < n; ++i) {
1475 set_Block_cfgpred(exc, idx[i], projX);
1479 /* the exception block should be optimized as some inputs are identical now */
1485 /* sixth step: replace old Phi */
1486 exchange(phi, projM);
1488 return res | DF_CHANGED;
1489 } /* optimize_phi */
1492 * walker, do the optimizations
1494 static void do_load_store_optimize(ir_node *n, void *env)
1496 walk_env_t *wenv = (walk_env_t*)env;
1498 switch (get_irn_opcode(n)) {
1501 wenv->changes |= optimize_load(n);
1505 wenv->changes |= optimize_store(n);
1509 wenv->changes |= optimize_phi(n, wenv);
1515 } /* do_load_store_optimize */
1518 typedef struct scc {
1519 ir_node *head; /**< the head of the list */
1522 /** A node entry. */
1523 typedef struct node_entry {
1524 unsigned DFSnum; /**< the DFS number of this node */
1525 unsigned low; /**< the low number of this node */
1526 int in_stack; /**< flag, set if the node is on the stack */
1527 ir_node *next; /**< link to the next node the the same scc */
1528 scc *pscc; /**< the scc of this node */
1529 unsigned POnum; /**< the post order number for blocks */
1532 /** A loop entry. */
1533 typedef struct loop_env {
1534 ir_nodehashmap_t map;
1535 struct obstack obst;
1536 ir_node **stack; /**< the node stack */
1537 size_t tos; /**< tos index */
1538 unsigned nextDFSnum; /**< the current DFS number */
1539 unsigned POnum; /**< current post order number */
1541 unsigned changes; /**< a bitmask of graph changes */
1545 * Gets the node_entry of a node
1547 static node_entry *get_irn_ne(ir_node *irn, loop_env *env)
1549 node_entry *e = (node_entry*)ir_nodehashmap_get(&env->map, irn);
1552 e = OALLOC(&env->obst, node_entry);
1553 memset(e, 0, sizeof(*e));
1554 ir_nodehashmap_insert(&env->map, irn, e);
1560 * Push a node onto the stack.
1562 * @param env the loop environment
1563 * @param n the node to push
1565 static void push(loop_env *env, ir_node *n)
1569 if (env->tos == ARR_LEN(env->stack)) {
1570 size_t nlen = ARR_LEN(env->stack) * 2;
1571 ARR_RESIZE(ir_node *, env->stack, nlen);
1573 env->stack[env->tos++] = n;
1574 e = get_irn_ne(n, env);
1579 * pop a node from the stack
1581 * @param env the loop environment
1583 * @return The topmost node
1585 static ir_node *pop(loop_env *env)
1587 ir_node *n = env->stack[--env->tos];
1588 node_entry *e = get_irn_ne(n, env);
1595 * Check if irn is a region constant.
1596 * The block or irn must strictly dominate the header block.
1598 * @param irn the node to check
1599 * @param header_block the header block of the induction variable
1601 static int is_rc(ir_node *irn, ir_node *header_block)
1603 ir_node *block = get_nodes_block(irn);
1605 return (block != header_block) && block_dominates(block, header_block);
1608 typedef struct phi_entry phi_entry;
1610 ir_node *phi; /**< A phi with a region const memory. */
1611 int pos; /**< The position of the region const memory */
1612 ir_node *load; /**< the newly created load for this phi */
1617 * An entry in the avail set.
1619 typedef struct avail_entry_t {
1620 ir_node *ptr; /**< the address pointer */
1621 ir_mode *mode; /**< the load mode */
1622 ir_node *load; /**< the associated Load */
1626 * Compare two avail entries.
1628 static int cmp_avail_entry(const void *elt, const void *key, size_t size)
1630 const avail_entry_t *a = (const avail_entry_t*)elt;
1631 const avail_entry_t *b = (const avail_entry_t*)key;
1634 return a->ptr != b->ptr || a->mode != b->mode;
1635 } /* cmp_avail_entry */
1638 * Calculate the hash value of an avail entry.
1640 static unsigned hash_cache_entry(const avail_entry_t *entry)
1642 return get_irn_idx(entry->ptr) * 9 + hash_ptr(entry->mode);
1643 } /* hash_cache_entry */
1646 * Move loops out of loops if possible.
1648 * @param pscc the loop described by an SCC
1649 * @param env the loop environment
1651 static void move_loads_out_of_loops(scc *pscc, loop_env *env)
1653 ir_node *phi, *load, *next, *other, *next_other;
1655 phi_entry *phi_list = NULL;
1658 /* collect all outer memories */
1659 for (phi = pscc->head; phi != NULL; phi = next) {
1660 node_entry *ne = get_irn_ne(phi, env);
1663 /* check all memory Phi's */
1667 assert(get_irn_mode(phi) == mode_M && "DFS return non-memory Phi");
1669 for (j = get_irn_arity(phi) - 1; j >= 0; --j) {
1670 ir_node *pred = get_irn_n(phi, j);
1671 node_entry *pe = get_irn_ne(pred, env);
1673 if (pe->pscc != ne->pscc) {
1674 /* not in the same SCC, is region const */
1675 phi_entry *pe = OALLOC(&env->obst, phi_entry);
1679 pe->next = phi_list;
1684 /* no Phis no fun */
1685 assert(phi_list != NULL && "DFS found a loop without Phi");
1687 /* for now, we cannot handle more than one input (only reducible cf) */
1688 if (phi_list->next != NULL)
1691 avail = new_set(cmp_avail_entry, 8);
1693 for (load = pscc->head; load; load = next) {
1695 node_entry *ne = get_irn_ne(load, env);
1698 if (is_Load(load)) {
1699 ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
1700 ir_node *ptr = get_Load_ptr(load);
1702 /* for now, we cannot handle Loads with exceptions */
1703 if (info->projs[pn_Load_res] == NULL || info->projs[pn_Load_X_regular] != NULL || info->projs[pn_Load_X_except] != NULL)
1706 /* for now, we can only move Load(Global) */
1707 if (! is_SymConst_addr_ent(ptr))
1709 load_mode = get_Load_mode(load);
1710 for (other = pscc->head; other != NULL; other = next_other) {
1711 node_entry *ne = get_irn_ne(other, env);
1712 next_other = ne->next;
1714 if (is_Store(other)) {
1715 ir_alias_relation rel = get_alias_relation(
1716 get_Store_ptr(other),
1717 get_irn_mode(get_Store_value(other)),
1719 /* if the might be an alias, we cannot pass this Store */
1720 if (rel != ir_no_alias)
1723 /* only Phis and pure Calls are allowed here, so ignore them */
1725 if (other == NULL) {
1726 ldst_info_t *ninfo = NULL;
1730 /* yep, no aliasing Store found, Load can be moved */
1731 DB((dbg, LEVEL_1, " Found a Load that could be moved: %+F\n", load));
1733 db = get_irn_dbg_info(load);
1734 for (pe = phi_list; pe != NULL; pe = pe->next) {
1736 ir_node *phi = pe->phi;
1737 ir_node *blk = get_nodes_block(phi);
1738 ir_node *pred = get_Block_cfgpred_block(blk, pos);
1740 avail_entry_t entry, *res;
1743 entry.mode = load_mode;
1744 res = (avail_entry_t*)set_find(avail, &entry, sizeof(entry), hash_cache_entry(&entry));
1748 irn = new_rd_Load(db, pred, get_Phi_pred(phi, pos), ptr, load_mode, cons_none);
1750 set_insert(avail, &entry, sizeof(entry), hash_cache_entry(&entry));
1751 DB((dbg, LEVEL_1, " Created %+F in %+F\n", irn, pred));
1754 ninfo = get_ldst_info(irn, &env->obst);
1756 ninfo->projs[pn_Load_M] = mem = new_r_Proj(irn, mode_M, pn_Load_M);
1758 /* irn is from cache, so do not set phi pred again.
1759 * There might be other Loads between phi and irn already.
1761 set_Phi_pred(phi, pos, mem);
1764 ninfo->projs[pn_Load_res] = new_r_Proj(irn, load_mode, pn_Load_res);
1767 /* now kill the old Load */
1768 exchange(info->projs[pn_Load_M], get_Load_mem(load));
1769 exchange(info->projs[pn_Load_res], ninfo->projs[pn_Load_res]);
1771 env->changes |= DF_CHANGED;
1776 } /* move_loads_out_of_loops */
1779 * Process a loop SCC.
1781 * @param pscc the SCC
1782 * @param env the loop environment
1784 static void process_loop(scc *pscc, loop_env *env)
1786 ir_node *irn, *next, *header = NULL;
1787 node_entry *b, *h = NULL;
1788 int j, only_phi, num_outside, process = 0;
1791 /* find the header block for this scc */
1792 for (irn = pscc->head; irn; irn = next) {
1793 node_entry *e = get_irn_ne(irn, env);
1794 ir_node *block = get_nodes_block(irn);
1797 b = get_irn_ne(block, env);
1799 if (header != NULL) {
1800 if (h->POnum < b->POnum) {
1810 /* check if this scc contains only Phi, Loads or Stores nodes */
1814 for (irn = pscc->head; irn; irn = next) {
1815 node_entry *e = get_irn_ne(irn, env);
1818 switch (get_irn_opcode(irn)) {
1820 if (is_Call_pure(irn)) {
1821 /* pure calls can be treated like loads */
1825 /* non-pure calls must be handle like may-alias Stores */
1828 /* cannot handle CopyB yet */
1832 if (get_Load_volatility(irn) == volatility_is_volatile) {
1833 /* cannot handle loops with volatile Loads */
1839 if (get_Store_volatility(irn) == volatility_is_volatile) {
1840 /* cannot handle loops with volatile Stores */
1849 for (j = get_irn_arity(irn) - 1; j >= 0; --j) {
1850 ir_node *pred = get_irn_n(irn, j);
1851 node_entry *pe = get_irn_ne(pred, env);
1853 if (pe->pscc != e->pscc) {
1854 /* not in the same SCC, must be a region const */
1855 if (! is_rc(pred, header)) {
1856 /* not a memory loop */
1859 if (out_rc == NULL) {
1860 /* first region constant */
1863 } else if (out_rc != pred) {
1864 /* another region constant */
1875 /* found a memory loop */
1876 DB((dbg, LEVEL_2, " Found a memory loop:\n "));
1877 if (only_phi && num_outside == 1) {
1878 /* a phi cycle with only one real predecessor can be collapsed */
1879 DB((dbg, LEVEL_2, " Found an USELESS Phi cycle:\n "));
1881 for (irn = pscc->head; irn; irn = next) {
1882 node_entry *e = get_irn_ne(irn, env);
1884 exchange(irn, out_rc);
1886 env->changes |= DF_CHANGED;
1890 #ifdef DEBUG_libfirm
1891 for (irn = pscc->head; irn; irn = next) {
1892 node_entry *e = get_irn_ne(irn, env);
1894 DB((dbg, LEVEL_2, " %+F,", irn));
1896 DB((dbg, LEVEL_2, "\n"));
1898 move_loads_out_of_loops(pscc, env);
1902 } /* process_loop */
1907 * @param pscc the SCC
1908 * @param env the loop environment
1910 static void process_scc(scc *pscc, loop_env *env)
1912 ir_node *head = pscc->head;
1913 node_entry *e = get_irn_ne(head, env);
1915 #ifdef DEBUG_libfirm
1917 ir_node *irn, *next;
1919 DB((dbg, LEVEL_4, " SCC at %p:\n ", pscc));
1920 for (irn = pscc->head; irn; irn = next) {
1921 node_entry *e = get_irn_ne(irn, env);
1925 DB((dbg, LEVEL_4, " %+F,", irn));
1927 DB((dbg, LEVEL_4, "\n"));
1931 if (e->next != NULL) {
1932 /* this SCC has more than one member */
1933 process_loop(pscc, env);
1938 * Do Tarjan's SCC algorithm and drive load/store optimization.
1940 * @param irn start at this node
1941 * @param env the loop environment
1943 static void dfs(ir_node *irn, loop_env *env)
1946 node_entry *node = get_irn_ne(irn, env);
1948 mark_irn_visited(irn);
1950 node->DFSnum = env->nextDFSnum++;
1951 node->low = node->DFSnum;
1955 if (is_Phi(irn) || is_Sync(irn)) {
1956 n = get_irn_arity(irn);
1957 for (i = 0; i < n; ++i) {
1958 ir_node *pred = get_irn_n(irn, i);
1959 node_entry *o = get_irn_ne(pred, env);
1961 if (!irn_visited(pred)) {
1963 node->low = MIN(node->low, o->low);
1965 if (o->DFSnum < node->DFSnum && o->in_stack)
1966 node->low = MIN(o->DFSnum, node->low);
1968 } else if (is_fragile_op(irn)) {
1969 ir_node *pred = get_memop_mem(irn);
1970 node_entry *o = get_irn_ne(pred, env);
1972 if (!irn_visited(pred)) {
1974 node->low = MIN(node->low, o->low);
1976 if (o->DFSnum < node->DFSnum && o->in_stack)
1977 node->low = MIN(o->DFSnum, node->low);
1978 } else if (is_Proj(irn)) {
1979 ir_node *pred = get_Proj_pred(irn);
1980 node_entry *o = get_irn_ne(pred, env);
1982 if (!irn_visited(pred)) {
1984 node->low = MIN(node->low, o->low);
1986 if (o->DFSnum < node->DFSnum && o->in_stack)
1987 node->low = MIN(o->DFSnum, node->low);
1990 /* IGNORE predecessors */
1993 if (node->low == node->DFSnum) {
1994 scc *pscc = OALLOC(&env->obst, scc);
2002 e = get_irn_ne(x, env);
2004 e->next = pscc->head;
2008 process_scc(pscc, env);
2013 * Do the DFS on the memory edges a graph.
2015 * @param irg the graph to process
2016 * @param env the loop environment
2018 static void do_dfs(ir_graph *irg, loop_env *env)
2020 ir_node *endblk, *end;
2023 inc_irg_visited(irg);
2025 /* visit all memory nodes */
2026 endblk = get_irg_end_block(irg);
2027 for (i = get_Block_n_cfgpreds(endblk) - 1; i >= 0; --i) {
2028 ir_node *pred = get_Block_cfgpred(endblk, i);
2030 pred = skip_Proj(pred);
2031 if (is_Return(pred)) {
2032 dfs(get_Return_mem(pred), env);
2033 } else if (is_Raise(pred)) {
2034 dfs(get_Raise_mem(pred), env);
2035 } else if (is_fragile_op(pred)) {
2036 dfs(get_memop_mem(pred), env);
2037 } else if (is_Bad(pred)) {
2038 /* ignore non-optimized block predecessor */
2040 assert(0 && "Unknown EndBlock predecessor");
2044 /* visit the keep-alives */
2045 end = get_irg_end(irg);
2046 for (i = get_End_n_keepalives(end) - 1; i >= 0; --i) {
2047 ir_node *ka = get_End_keepalive(end, i);
2049 if (is_Phi(ka) && !irn_visited(ka))
2055 * Optimize Loads/Stores in loops.
2057 * @param irg the graph
2059 static int optimize_loops(ir_graph *irg)
2063 env.stack = NEW_ARR_F(ir_node *, 128);
2068 ir_nodehashmap_init(&env.map);
2069 obstack_init(&env.obst);
2071 /* calculate the SCC's and drive loop optimization. */
2074 DEL_ARR_F(env.stack);
2075 obstack_free(&env.obst, NULL);
2076 ir_nodehashmap_destroy(&env.map);
2079 } /* optimize_loops */
2082 * do the load store optimization
2084 static ir_graph_properties_t do_loadstore_opt(ir_graph *irg)
2087 ir_graph_properties_t res = 0;
2089 FIRM_DBG_REGISTER(dbg, "firm.opt.ldstopt");
2091 assert(get_irg_phase_state(irg) != phase_building);
2092 assert(get_irg_pinned(irg) != op_pin_state_floats &&
2093 "LoadStore optimization needs pinned graph");
2095 if (get_opt_alias_analysis()) {
2096 assure_irp_globals_entity_usage_computed();
2099 obstack_init(&env.obst);
2102 /* init the links, then collect Loads/Stores/Proj's in lists */
2104 irg_walk_graph(irg, firm_clear_link, collect_nodes, &env);
2106 /* now we have collected enough information, optimize */
2107 irg_walk_graph(irg, NULL, do_load_store_optimize, &env);
2109 env.changes |= optimize_loops(irg);
2111 obstack_free(&env.obst, NULL);
2113 /* Handle graph state */
2115 edges_deactivate(irg);
2118 if (!(env.changes & CF_CHANGED)) {
2119 res |= IR_GRAPH_PROPERTY_CONSISTENT_DOMINANCE | IR_GRAPH_PROPERTY_NO_BADS;
2125 static optdesc_t opt_loadstore = {
2127 IR_GRAPH_PROPERTY_NO_UNREACHABLE_CODE | IR_GRAPH_PROPERTY_CONSISTENT_OUT_EDGES | IR_GRAPH_PROPERTY_NO_CRITICAL_EDGES | IR_GRAPH_PROPERTY_CONSISTENT_DOMINANCE | IR_GRAPH_PROPERTY_CONSISTENT_ENTITY_USAGE,
2131 int optimize_load_store(ir_graph *irg)
2133 perform_irg_optimization(irg, &opt_loadstore);
2137 ir_graph_pass_t *optimize_load_store_pass(const char *name)
2139 return def_graph_pass_ret(name ? name : "ldst", optimize_load_store);
2140 } /* optimize_load_store_pass */