2 * Copyright (C) 1995-2007 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 scalar replacement of arrays and compounds
23 * @author Beyhan Veliev, Michael Beck
34 #include "data_flow_scalar_replace.h"
45 #include "analyze_irg_args.h"
47 #include "compute_loop_info.h"
51 #define SET_ENT_VNUM(ent, vnum) set_entity_link(ent, INT_TO_PTR(vnum))
52 #define GET_ENT_VNUM(ent) (unsigned)PTR_TO_INT(get_entity_link(ent))
53 #define SET_IRN_VNUM(irn, vnum) set_irn_link(irn, INT_TO_PTR(vnum))
54 #define GET_IRN_VNUM(irn) (unsigned)PTR_TO_INT(get_irn_link(irn))
58 typedef struct _ent_leaves_t{
59 ir_entity *ent; /**< An entity, that contains scalars for replace.*/
60 pset *leaves; /**< All leaves of this entity.*/
63 typedef struct _sels_t {
64 ir_node *sel; /**< A sel node, thats entity have scalars.*/
65 ir_entity *ent; /**< The entity of this sel node.*/
68 typedef struct _call_access_t {
69 ir_node *call; /**< A call node, that have as parameter a scalar.*/
70 unsigned int access_type; /**< The access type, with that this call access this scalar.*/
73 typedef struct _fixlist_entry_t {
74 ir_node *irn; /**< An ir node, that must be fixed.*/
75 unsigned int vnum; /**< The value number, that must became this ir node.*/
78 typedef struct _syncs_fixlist_entry_t {
79 ir_node *irn; /**< A sync node that must be fixed.*/
80 int *accessed_vnum; /**< A pointer to save an array with value numbers, that must became this sync.*/
81 }syncs_fixlist_entry_t;
83 /* A entry, that save the memory
84 * edge state and the access state for this leave
85 * int the array,that is created for every block.*/
86 typedef struct _leave_t {
87 ir_node *mem_edge_state; /**< memory state for this scalar in this block.*/
88 unsigned int access_type; /**< access state for this scalar in this block.*/
89 set *calls; /**< call nodes,that change this scalar in this block.*/
93 * A path element entry: it is either an entity
94 * or a tarval, because we evaluate only constant array
95 * accesses like a.b.c[8].d
103 * An access path, used to assign value numbers
104 * to variables that will be scalar replaced
106 typedef struct _path_t {
107 unsigned vnum; /**< the value number */
108 unsigned path_len; /**< the length of the access path */
109 path_elem_t path[1]; /**< the path */
113 * environment for memory walker
115 typedef struct _env_t {
116 struct obstack obst; /**< a obstack for the memory edge */
117 set *set_sels; /**< a set with all sels, that are reachable from an entity with a scalar.*/
118 set *set_ent; /**< a set with all entities that have one or more scalars.*/
119 fixlist_entry_t *fix_phis; /**< list of all Phi nodes that must be fixed */
120 fixlist_entry_t *fix_ls; /**< list of all Load or Store nodes that must be fixed */
121 syncs_fixlist_entry_t *fix_syncs; /**< list of all Sync nodes that must be fixed */
122 unsigned int nvals; /**< to save the number of scalars.*/
123 unsigned int gl_mem_vnum; /**< indicate the position of the globule memory edge state in var_arr.*/
124 unsigned int vnum_state; /**< indicate the position of the value number state in var_arr.*/
125 unsigned int changes; /**< to save if by anlyse_calls is changed anything.*/
131 * Compare two elements of the ent_leaves_t set.
133 * @return 0 if they are identically
135 static int ent_leaves_t_cmp(const void *elt, const void *key, size_t size)
137 const ent_leaves_t *c1 = elt;
138 const ent_leaves_t *c2 = key;
140 return c1->ent != c2->ent;
144 * Compare two elements of the ent_access_t set.
146 * @return 0 if they are identically
148 static int ent_cmp(const void *elt, const void *key)
150 const ir_entity *c1 = elt;
151 const ir_entity *c2 = key;
157 * Compare two elements of the sels_t set.
159 * @return 0 if they are identically
161 static int sels_cmp(const void *elt, const void *key, size_t size)
163 const sels_t *c1 = elt;
164 const sels_t *c2 = key;
166 return c1->sel != c2->sel;
170 * Compare two elements of the leave_t set.
172 * @return 0 if they are identically
174 static int leave_cmp(const void *elt, const void *key)
176 ir_node *c1 = (ir_node *)elt;
177 ir_node *c2 = (ir_node *)key;
179 return get_Sel_entity(c1) != get_Sel_entity(c2);
183 * Compare two elements of the call_access_t set.
185 * @return 0 if they are identically
187 static int call_cmp(const void *elt, const void *key, size_t size)
189 const call_access_t *c1 = elt;
190 const call_access_t *c2 = key;
192 return c1->call != c2->call;
198 * @return 0 if they are identically
200 static int path_cmp(const void *elt, const void *key, size_t size)
202 const path_t *p1 = elt;
203 const path_t *p2 = key;
205 /* we can use memcmp here, because identical tarvals should have identical addresses */
206 return memcmp(p1->path, p2->path, p1->path_len * sizeof(p1->path[0]));
210 * Calculate a hash value for a path.
212 static unsigned path_hash(const path_t *path)
217 for (i = 0; i < path->path_len; ++i)
218 hash ^= (unsigned)PTR_TO_INT(path->path[i].ent);
224 * Returns non-zero, if all induces of a Sel node are constants.
226 * @param sel the Sel node that will be checked
228 static int is_const_sel(ir_node *sel) {
229 int i, n = get_Sel_n_indexs(sel);
231 for (i = 0; i < n; ++i) {
232 ir_node *idx = get_Sel_index(sel, i);
234 if (get_irn_op(idx) != op_Const)
241 * Returns non-zero, if the address of an entity
242 * represented by a Sel node (or it's successor Sels) is taken.
244 static int is_address_taken(ir_node *sel)
248 if (! is_const_sel(sel))
251 for (i = get_irn_n_outs(sel) - 1; i >= 0; --i) {
252 ir_node *succ = get_irn_out(sel, i);
254 switch (get_irn_opcode(succ)) {
256 /* ok, we just load from that entity */
260 /* check that Sel is not the Store's value */
261 if (get_Store_value(succ) == sel)
266 /* Check the Sel successor of Sel */
267 int res = is_address_taken(succ);
275 /* The address of an entity is given as a parameter.
276 * We analyzes that later and optimizes this scalar
282 /* another op, the address is taken */
290 * Link all Sels with the entity.
292 * @param ent the entity that will be scalar replaced
293 * @param sel a Sel node that selects some fields of this entity
295 static void link_all_leave_sels(ir_entity *ent, ir_node *sel)
299 n = get_irn_n_outs(sel);
300 for (i = 0; i < n; ++i) {
301 ir_node *succ = get_irn_out(sel, i);
303 if (get_irn_op(succ) == op_Sel)
304 link_all_leave_sels(ent, succ);
308 /* if Sel nodes with memory inputs are used, a entity can be
309 * visited more than once causing a ring here, so we use the
310 * node flag to mark linked nodes
312 if (irn_visited(sel))
316 * we link the sels to the entity.
318 set_irn_link(sel, get_entity_link(ent));
319 set_entity_link(ent, sel);
321 mark_irn_visited(sel);
324 /* we need a special address that serves as an address taken marker */
326 static void *ADDRESS_TAKEN = &_x;
329 * Find possible scalar replacements.
331 * @param irg an IR graph
333 * This function finds variables on the (members of the) frame type
334 * that can be scalar replaced, because their address is never taken.
335 * If such a variable is found, it's entity link will hold a list of all
336 * Sel nodes, that selects anythings of this entity.
337 * Otherwise, the link will be ADDRESS_TAKEN or NULL.
339 * @return non-zero if at least one entity could be replaced
342 static int find_possible_replacements(ir_graph *irg)
344 ir_node *irg_frame = get_irg_frame(irg);
348 inc_irg_visited(irg);
350 n = get_irn_n_outs(irg_frame);
353 * First, clear the link field of all interestingentities.
354 * Note that we did not rely on the fact that there is only
355 * one Sel node per entity, so we might access one entity
356 * more than once here.
357 * That's why we have need two loops.
359 for (i = 0; i < n; ++i) {
360 ir_node *succ = get_irn_out(irg_frame, i);
362 if (get_irn_op(succ) == op_Sel) {
363 ir_entity *ent = get_Sel_entity(succ);
364 set_entity_link(ent, NULL);
369 * Check the ir_graph for Sel nodes. If the entity of Sel
370 * isn't a scalar replacement set the link of this entity
371 * equal ADDRESS_TAKEN.
373 for (i = 0; i < n; ++i) {
374 ir_node *succ = get_irn_out(irg_frame, i);
376 if (get_irn_op(succ) == op_Sel) {
377 ir_entity *ent = get_Sel_entity(succ);
380 if (get_entity_link(ent) == ADDRESS_TAKEN)
384 * Beware: in rare cases even entities on the frame might be
385 * volatile. This might happen if the entity serves as a store
386 * to a value that must survive a exception. Do not optimize
387 * such entities away.
389 if (get_entity_volatility(ent) == volatility_is_volatile) {
390 set_entity_link(ent, ADDRESS_TAKEN);
394 ent_type = get_entity_type(ent);
396 /* we can handle arrays, structs and atomic types yet */
397 if (is_Array_type(ent_type) || is_Struct_type(ent_type) || is_atomic_type(ent_type)) {
398 if (is_address_taken(succ)) {
399 if (get_entity_link(ent)) /* killing one */
401 set_entity_link(ent, ADDRESS_TAKEN);
404 /* possible found one */
405 if (get_entity_link(ent) == NULL)
407 link_all_leave_sels(ent, succ);
416 static int is_leave_sel(ir_node *sel) {
420 for(i = get_irn_n_outs(sel) - 1; i >= 0; i--) {
421 succ = get_irn_out(sel, i);
422 if(get_irn_op(succ) == op_Sel)
430 * Return a path from the Sel node sel to it's root.
432 * @param sel the Sel node
433 * @param len the length of the path so far
435 static path_t *find_path(ir_node *sel, unsigned len)
439 ir_node *pred = get_Sel_ptr(sel);
441 /* the current Sel node will add some path elements */
442 n = get_Sel_n_indexs(sel);
445 if (get_irn_op(pred) != op_Sel) {
446 /* we found the root */
448 res = xmalloc(sizeof(*res) + (len - 1) * sizeof(res->path));
452 res = find_path(pred, len);
454 pos = res->path_len - len;
456 res->path[pos++].ent = get_Sel_entity(sel);
457 for (i = 0; i < n; ++i) {
458 ir_node *index = get_Sel_index(sel, i);
460 if(get_irn_op(index) == op_Const)
461 res->path[pos++].tv = get_Const_tarval(index);
467 * Allocate value numbers for the leaves
468 * in our found entities.
470 * @param sels a set that will contain all Sels that have a value number
471 * @param ent the entity that will be scalar replaced
472 * @param vnum the first value number we can assign
473 * @param modes a flexible array, containing all the modes of
476 * @return the next free value number
478 static unsigned allocate_value_numbers(set *set_sels, pset *leaves, ir_entity *ent, unsigned vnum)
483 set *pathes = new_set(path_cmp, 8);
485 /* visit all Sel nodes in the chain of the entity */
486 for (sel = get_entity_link(ent); sel; sel = next) {
487 next = get_irn_link(sel);
489 /* we save for every sel it root entity, why
490 * we need this information, when we split the memory edge,
491 * and we must mark this sel for later. */
494 set_insert(set_sels, &key_sels, sizeof(key_sels), HASH_PTR(sel));
496 if(! is_leave_sel(sel))
498 /* We have found a leave and we add it to the pset of this entity.*/
499 pset_insert(leaves, sel, HASH_PTR(get_Sel_entity(sel)));
501 key = find_path(sel, 0);
502 path = set_find(pathes, key, sizeof(*key) + sizeof(key->path[0]) * key->path_len, path_hash(key));
505 SET_IRN_VNUM(sel, path->vnum);
510 set_insert(pathes, key, sizeof(*key) + sizeof(key->path[0]) * key->path_len, path_hash(key));
512 SET_IRN_VNUM(sel, key->vnum);
518 set_entity_link(ent, NULL);
522 * Add a sync node to it fix list.
524 * @param sync The sync node, that myst be addet to the fix list.
525 * @param unk_vnum An array whit the value number, that are synced with this sync node.
526 * @param env The enviroment pinter.
528 static void add_sync_to_fixlist(ir_node *sync, int *unk_vnum, env_t *env) {
530 syncs_fixlist_entry_t *s;
532 s = obstack_alloc(&env->obst, sizeof(*s));
534 s->accessed_vnum = unk_vnum;
535 set_irn_link(sync, env->fix_syncs);
539 * Add a ir node to it fix list.
541 * @param irn The ir node, that myst be addet to the fix list.
542 * @param vnum The value number, that must baceme this ir node as predecessor later.
543 * @param env The enviroment pinter.
545 static void add_ls_to_fixlist(ir_node *irn, int vnum, env_t *env) {
549 l = obstack_alloc(&env->obst, sizeof(*l));
553 if(get_irn_op(irn) == op_Phi) {
554 set_irn_link(l->irn, env->fix_phis);
557 set_irn_link(l->irn, env->fix_ls);
562 static void add_mem_edge(value_arr_entry_t *val_arr, int vnum, ir_node ***in, int **accessed_vnum) {
564 if(val_arr[vnum].mem_edge_state != NULL)
565 ARR_APP1(ir_node *, *in, val_arr[vnum].mem_edge_state);
567 ARR_APP1(int, *accessed_vnum, vnum);
568 ARR_APP1(ir_node *, *in, new_Unknown(mode_M));
572 * The function handles the scalars, that wase stored
575 * @param blk The block, that must be handled.
576 * @param env The enviroment pinter.
579 /* Return the memory successor of the call node.*/
580 static ir_node *get_Call_mem_out(ir_node *call) {
585 for(i = get_irn_n_outs(call) - 1; i >= 0; i--) {
586 mem = get_irn_out(call, i);
587 if(get_irn_mode(mem) == mode_M)
590 /* is not reachable*/
595 static void sync_stored_scalars(ir_node *blk, env_t *env) {
598 int *unk_vnum; /**< An arraw, where are saved the value number, that
599 are synced from this sync node.*/
600 ent_leaves_t *value_ent;
601 value_arr_entry_t *val_arr_blk, *val_arr;
602 ir_node *pred, *leave, *sync, **in;
603 ir_node *sync_blk; /**< The block, where the sync node must be created.*/
606 val_arr_blk = get_irn_link(blk);
608 for(value_ent = set_first(env->set_ent); value_ent; value_ent = set_next(env->set_ent)) {
611 if(val_arr_blk[GET_ENT_VNUM(value_ent->ent)].access_type <= 3)
612 /* This entity is not stored in this block.*/
615 for(i = get_Block_n_cfgpreds(blk) - 1; i >= 0; i--) {
617 pred = get_Block_cfgpred(blk, i);
618 pred = get_nodes_block(pred);
619 val_arr = get_irn_link(pred);
621 if(val_arr[GET_ENT_VNUM(value_ent->ent)].access_type == SYNCED)
622 /* This entity was synced.*/
625 if(val_arr[GET_ENT_VNUM(value_ent->ent)].access_type <= 3) {
627 /* To avoid repeated sync of this entity in this block.*/
628 val_arr[GET_ENT_VNUM(value_ent->ent)].access_type = SYNCED;
629 /* In this predecessor block is this entity not acessed.
630 * We must sync in the end ot this block.*/
631 if(get_Block_n_cfgpreds(blk) > 1)
632 sync_blk = get_nodes_block(get_Block_cfgpred(blk, i));
636 val_arr = get_irn_link(sync_blk);
637 /* An array to save the memory edges, that must be
639 in = NEW_ARR_F(ir_node *, 1);
641 /* An array to save the value numbers,
642 * that must be repaired.*/
643 unk_vnum = NEW_ARR_F(int, 0);
644 /* The global memory edge.*/
645 if(val_arr[env->gl_mem_vnum].mem_edge_state == NULL)
646 in[0] = new_Unknown(mode_M);
648 in[0] = val_arr[env->gl_mem_vnum].mem_edge_state;
650 for(leave = pset_first(value_ent->leaves); leave; leave = pset_next(value_ent->leaves))
651 /* All this memory edges must be synced.*/
652 add_mem_edge(val_arr, GET_IRN_VNUM(leave), &in, &unk_vnum);
654 /* We create the sync and set it in the global memory state.*/
655 sync = new_r_Sync(current_ir_graph, sync_blk, ARR_LEN(in), in);
656 /* We must check this, why it is possible to get a Bad node
657 * form new_r_Sync(), when the node can be optimized.
658 * In this case we must do nothing.*/
659 if(get_irn_op(sync) == op_Sync) {
660 val_arr[env->gl_mem_vnum].mem_edge_state = sync;
661 /* We add this sync node to the sync's fix list.*/
662 add_sync_to_fixlist(val_arr[env->gl_mem_vnum].mem_edge_state, unk_vnum, env);
670 * The function split the memory edge of load and store nodes, that have
671 * as predecessor a scalar
673 * @param irn The node, that memory edge must be spleted.
674 * @param env The enviroment pinter.
676 static void split_ls_mem_edge(ir_node *irn, env_t *env) {
679 ir_node *leave, *irn_blk, *mem_state, *new_mem_state;
680 unsigned ent_vnum, sel_vnum, i;
681 value_arr_entry_t *val_arr;
682 sels_t key_sels, *value_sels;
683 ent_leaves_t key_ent, *value_ent;
685 op = get_irn_op(irn);
688 key_sels.sel = get_Load_ptr(irn);
690 key_sels.sel = get_Store_ptr(irn);
692 value_sels = set_find(env->set_sels, &key_sels, sizeof(key_sels), HASH_PTR(key_sels.sel));
694 if(value_sels != NULL) {
695 /* we have found a load or store, that use a sel of our set
696 * and we must split or extend, if the memory edge have been
697 * split for this sel, the memory edge.*/
699 key_ent.ent = value_sels->ent;
700 value_ent = set_find(env->set_ent, &key_ent, sizeof(key_ent), HASH_PTR(key_ent.ent));
701 /*To check if the enities set is right filled. */
702 assert(value_ent && " This sel's entity isn't int the entity set.");
704 leave = pset_find(value_ent->leaves, key_sels.sel, HASH_PTR(get_Sel_entity(key_sels.sel)));
705 /*To check if the leaves set is right filled. */
706 assert(leave && "Anything in data_flow_scalar_replacment algorithm is wrong.");
708 ent_vnum = GET_ENT_VNUM(value_ent->ent);
709 sel_vnum = GET_IRN_VNUM(leave);
710 irn_blk = get_nodes_block(irn);
711 val_arr = get_irn_link(irn_blk);
713 if(val_arr[ent_vnum].access_type == 0)
714 /* We have found a scalar, that address is not stored as jet.*/
717 /* This scalar have been stored.*/
718 i = env->gl_mem_vnum;
720 if(val_arr[i].mem_edge_state == NULL) {
721 /* We split now for this sel the memory edge in this block.*/
722 mem_state = new_Unknown(mode_M);
723 /* We must mark this node to fix later*/
724 add_ls_to_fixlist(irn, i, env);
727 /* We have split the memory edge and the current state is saved.*/
728 mem_state = val_arr[i].mem_edge_state;
730 /* We set this Load or Store to the memory edge of this
733 set_Load_mem(irn, mem_state);
735 set_Store_mem(irn, mem_state);
737 /* When we have split or extended the memory edge we must
738 * update the memory_edge_state of this sel*/
739 new_mem_state = get_irn_out(irn, 0);
740 if(get_irn_mode(new_mem_state) == mode_M)
741 val_arr[i].mem_edge_state = new_mem_state;
743 val_arr[i].mem_edge_state = get_irn_out(irn, 1);
748 * The function split the memory edge of phi nodes, that have
749 * as predecessor a scalar
751 * @param irn The phi node, that memory edge must be spleted.
752 * @param env The enviroment pinter.
754 static void split_phi_mem_edge(ir_node *irn, env_t *env) {
756 ir_node *irn_blk, *unk, *leave, **in;
758 ent_leaves_t *value_ent;
759 value_arr_entry_t *val_arr;
761 irn_blk = get_nodes_block(irn);
762 val_arr = get_irn_link(irn_blk);
764 n = get_Block_n_cfgpreds(irn_blk);
766 in = alloca(sizeof(*in) * n);
768 for(value_ent = set_first(env->set_ent); value_ent; value_ent = set_next(env->set_ent))
769 if(val_arr[GET_ENT_VNUM(value_ent->ent)].access_type < 3)
770 /* This scalar wasn't be saved and we need to produce a phi for it.*/
771 for(leave = pset_first(value_ent->leaves); leave; leave = pset_next(value_ent->leaves)){
773 unk = new_Unknown(mode_M);
774 for (j = n - 1; j >= 0; --j)
777 val_arr[GET_IRN_VNUM(leave)].mem_edge_state = new_r_Phi(current_ir_graph, irn_blk, n, in, mode_M);
779 add_ls_to_fixlist(val_arr[GET_IRN_VNUM(leave)].mem_edge_state, GET_IRN_VNUM(leave), env);
782 /* We use for the global memory the phi node, that
783 * is already available.*/
784 val_arr[env->gl_mem_vnum].mem_edge_state = irn;
788 * The function handles the call nodes, that have
789 * as parameter a scalar
791 * @param env The enviroment pinter.
792 * @param call The call node, that must be handled.
793 * @param accessed_entities A set wit all entities, that are accessed from this call node.*/
794 static void split_call_mem_edge(env_t *env, ir_node *call, pset *accessed_entities) {
796 ent_leaves_t key_ent, *value_ent;
797 value_arr_entry_t *val_arr;
798 call_access_t key_call, *value_call;
799 ir_node *call_blk, *new_mem_state, *leave;
803 int fix_irn = 0; /**< Set to 1 if we must add this call to it fix list.*/
804 int *accessed_leaves_vnum = NULL; /**< An arraw, where are saved the value number, that
805 are synced from call's sync node, if we need it.*/
807 if(get_irn_node_nr(call) == 2763)
810 call_blk = get_nodes_block(call);
811 val_arr = get_irn_link(call_blk);
812 /* An array to save the memory edges, that must be
814 in = NEW_ARR_F(ir_node *, 1);
815 /* An array to save the value numbers of the memory
816 * edges that must be repaired.*/
817 accessed_leaves_vnum = NEW_ARR_F(int, 0);
819 /* We get the memory successor of the call node.
820 * It is the new memory state for all synced memory
822 new_mem_state = get_Call_mem_out(call);
824 /* The global memory is the first predecessor of the create sync node.*/
825 if(val_arr[env->gl_mem_vnum].mem_edge_state == NULL) {
826 in[0] = new_Unknown(mode_M);
830 in[0] = val_arr[env->gl_mem_vnum].mem_edge_state;
833 for(ent = pset_first(accessed_entities); ent; ent = pset_next(accessed_entities)) {
834 /* Whit this loop we iterate all accessed entities from this call and collect
835 * all memory edges, that we must sync.*/
836 ent_vnum = GET_ENT_VNUM(ent);
838 key_call.call = call;
839 value_call = set_find(val_arr[ent_vnum].calls, &key_call, sizeof(key_call), HASH_PTR(key_call.call));
842 value_ent = set_find(env->set_ent, &key_ent, sizeof(key_ent), HASH_PTR(key_ent.ent));
844 if(val_arr[ent_vnum].access_type <= 3) {
845 /* This scalar's address wasn't stored in this block.*/
846 switch(value_call->access_type) {
848 case ptr_access_none :
849 /* In this case we have nothing to do.*/
852 case ptr_access_read:
853 case ptr_access_write:
855 /* All this cases must be traded equal.*/
857 for(leave = pset_first(value_ent->leaves); leave; leave = pset_next(value_ent->leaves)){
858 /* All this memory edges must be synced.*/
859 add_mem_edge(val_arr, GET_IRN_VNUM(leave), &in, &accessed_leaves_vnum);
861 /* We update the memory state of this leave.*/
862 if(value_call->access_type != ptr_access_read)
863 val_arr[GET_IRN_VNUM(leave)].mem_edge_state = new_mem_state;
872 /* We must update the global memory state.*/
873 val_arr[env->gl_mem_vnum].mem_edge_state = new_mem_state;
875 if(ARR_LEN(in) == 1) {
876 /* we must set the call memory to gobale momory*/
877 set_Call_mem(call,in[0]);
880 /* We add this call node to the call fix list..*/
881 add_ls_to_fixlist(call, env->gl_mem_vnum, env);
884 /* We create the sync and set it as memory predecessor of the call node.*/
885 sync = new_r_Sync(current_ir_graph, call_blk, ARR_LEN(in), in);
886 /* We must check this, why it is possible to get a Bad node
887 * form new_r_Sync(), when the node can be optimized.
888 * In this case we must do nothing.*/
889 if(get_irn_op(sync) == op_Sync) {
891 set_Call_mem(call, sync);
892 if(ARR_LEN(accessed_leaves_vnum))
893 /* We add this sync node to the sync's fix list.*/
894 add_sync_to_fixlist(sync, accessed_leaves_vnum, env);
901 * The function split the memory edge from the passed
902 * ir node if this is needed
904 * @param irn The node, that memory edge must be spleted.
905 * @param env The enviroment pinter.
907 static void split_memory_edge(ir_node *irn, void *ctx) {
910 ir_node *sel, *irn_blk;
912 sels_t key_sels, *value_sels;
913 value_arr_entry_t *val_arr;
914 pset *accessed_entities; /**< A set to save all entities accessed from a call.*/
918 op = get_irn_op(irn);
923 irn_blk = get_nodes_block(irn);
925 if (Block_not_block_visited(irn_blk)) {
926 /* We sync first the stored scalar address in this block.*/
927 mark_Block_block_visited(irn_blk);
928 sync_stored_scalars(irn_blk, env);
931 if(op == op_Load || op == op_Store)
933 split_ls_mem_edge(irn, env);
936 if (op == op_Phi && get_irn_mode(irn) == mode_M) {
938 * found a memory Phi: Here, we must create new Phi nodes
940 split_phi_mem_edge(irn, env);
945 /* Calls that have a NoMem input do neither read nor write memory.
946 We can completely ignore them here. */
947 if (get_irn_op(get_Call_mem(irn)) == op_NoMem)
950 /* We save in this set all entities,
951 * that are accessed from this call node.*/
952 accessed_entities = new_pset(ent_cmp, 8);
953 val_arr = get_irn_link(get_nodes_block(irn));
955 for ( i = get_Call_n_params(irn) - 1; i >= 0; i--) {
957 sel = get_Call_param(irn, i);
959 if(get_irn_op(sel) == op_Sel) {
961 value_sels = set_find(env->set_sels, &key_sels, sizeof(key_sels), HASH_PTR(key_sels.sel));
963 if(value_sels != NULL && val_arr[GET_ENT_VNUM(value_sels->ent)].access_type <= 3)
964 /* We save in this set all accessed entities from this call node whit
965 * access none, read, write or rw..*/
966 pset_insert(accessed_entities, value_sels->ent, HASH_PTR(value_sels->ent));
970 if(pset_count(accessed_entities))
971 split_call_mem_edge(env, irn, accessed_entities);
973 del_pset(accessed_entities);
980 * searches through blocks beginning from block for value
981 * vnum and return it.
983 * @param block A block from the current ir graph.
984 * @param vnum The value number, that must be found.
986 static ir_node *find_vnum_value(ir_node *block, unsigned vnum)
988 value_arr_entry_t *val_arr;
992 if (Block_not_block_visited(block)) {
993 mark_Block_block_visited(block);
995 val_arr = get_irn_link(block);
997 if (val_arr[vnum].mem_edge_state)
998 return val_arr[vnum].mem_edge_state;
1000 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1001 ir_node *pred = get_Block_cfgpred(block, i);
1003 res = find_vnum_value(get_nodes_block(pred), vnum);
1012 * fix the Load/Store or Call list
1014 * @param The enviroment pinter.
1016 static void fix_ls(env_t *env)
1019 ir_node *irn, *block, *pred, *val = NULL;
1023 for (l = env->fix_ls; l; l = get_irn_link(irn)) {
1026 op = get_irn_op(irn);
1027 block = get_nodes_block(irn);
1028 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1029 pred = get_Block_cfgpred(block, i);
1030 pred = get_nodes_block(pred);
1032 inc_irg_block_visited(current_ir_graph);
1033 val = find_vnum_value(pred, l->vnum);
1041 set_Store_mem(irn, val);
1044 set_Load_mem(irn, val);
1046 set_Call_mem(irn, val);
1054 * @param The enviroment pinter.
1056 static void fix_phis(env_t *env)
1059 ir_node *phi, *block, *pred, *val;
1062 for (l = env->fix_phis; l; l = get_irn_link(phi)) {
1065 block = get_nodes_block(phi);
1066 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1068 pred = get_Block_cfgpred(block, i);
1069 pred = get_nodes_block(pred);
1071 inc_irg_block_visited(current_ir_graph);
1072 val = find_vnum_value(pred, l->vnum);
1075 set_irn_n(phi, i, val);
1084 * @param The enviroment pinter.
1086 static void fix_syncs(env_t *env)
1088 syncs_fixlist_entry_t *l;
1089 ir_node *sync, *block, *pred, *val;
1093 for (l = env->fix_syncs; l; l = get_irn_link(sync)) {
1097 /* The sync block must have one predecessor, when it
1098 have unknown nodes as predecessor.*/
1099 block = get_nodes_block(sync);
1100 pred = get_Block_cfgpred(block, 0);
1101 pred = get_nodes_block(pred);
1103 /* We first repair the global memory edge at the first position of sync predecessors.*/
1104 if(get_irn_op(get_irn_n(sync, 0)) == op_Unknown) {
1105 inc_irg_block_visited(current_ir_graph);
1106 val = find_vnum_value(pred, env->gl_mem_vnum);
1109 set_irn_n(sync, 0, val);
1112 for (i = get_irn_arity(sync) - 1; i >= 1; --i) {
1113 /* We repair the leaves*/
1115 assert(k <= ARR_LEN(l->accessed_vnum) && "The algorythm for sync repair is wron");
1116 if(get_irn_op(get_irn_n(sync, i)) == op_Unknown) {
1117 inc_irg_block_visited(current_ir_graph);
1118 val = find_vnum_value(pred, l->accessed_vnum[k++]);
1121 set_irn_n(sync, i, val);
1124 DEL_ARR_F(l->accessed_vnum);
1128 * For the end node we must sync all memory edges.
1130 * @param The enviroment pinter.
1132 static void sync_mem_edges(env_t *env) {
1134 value_arr_entry_t *val_arr;
1135 ir_node **in, *sync, *Return, *Return_blk;
1136 int i, vnum, vnum_state;
1138 Return = get_Block_cfgpred(get_irg_end_block(current_ir_graph), 0);
1139 Return_blk = get_nodes_block(Return);
1140 val_arr = get_irn_link(Return_blk);
1144 for(i = 0; i <= (int)env->gl_mem_vnum; i++)
1145 /* we get the current state of non saved scalars.*/
1146 if(val_arr[i].access_type <= 3)
1149 /* We allocate the memory, that we need for the predecessors of the sync.*/
1150 in = xmalloc(sizeof(ir_node*) *vnum_state);
1152 /* The global memory edge is the first predecessor of this sync node.*/
1153 if(val_arr[env->gl_mem_vnum].mem_edge_state == NULL) {
1154 /* We must search through blocks for this memory state.*/
1155 inc_irg_block_visited(current_ir_graph);
1156 in[0] = find_vnum_value(Return_blk, env->gl_mem_vnum);
1159 in[0] = val_arr[env->gl_mem_vnum].mem_edge_state;
1162 for(i = 1, vnum = 0; vnum < (int)env->gl_mem_vnum; vnum++) {
1164 if(val_arr[vnum].access_type <= 3) {
1165 /* we add the non saved scalars as predecessors of the sync.*/
1167 if(val_arr[vnum].mem_edge_state == NULL) {
1168 /* We must search through blocks for this memory state.*/
1169 inc_irg_block_visited(current_ir_graph);
1170 in[i] = find_vnum_value(Return_blk, vnum);
1173 in[i] = val_arr[vnum].mem_edge_state;
1178 sync = new_r_Sync(current_ir_graph, Return_blk, vnum_state, in);
1179 set_Return_mem(Return, sync);
1185 * Walker: allocate the value array for every block.
1187 * @param block A block from the current ir graph for that must be allocated a value array.
1188 * @param ctx The enviroment pinter.
1190 static void alloc_value_arr(ir_node *block, void *ctx)
1195 value_arr_entry_t *var_arr = obstack_alloc(&env->obst, sizeof(value_arr_entry_t) *(env->nvals + set_count(env->set_ent) + 1));
1197 /* the value array is empty at start */
1198 memset(var_arr, 0, sizeof(value_arr_entry_t) * (env->nvals + set_count(env->set_ent) + 1));
1199 set_irn_link(block, var_arr);
1201 /* We set the block value number state to optimal and later we update this.*/
1202 var_arr[env->vnum_state].access_type = env->nvals;
1204 if(get_irg_start_block(current_ir_graph) == block)
1205 /* We initilize the startblocks array with the irg initilize memory, why
1206 * it must be the start point of all memory edges.*/
1207 for(i = (env->nvals + set_count(env->set_ent)) ; i >=0; i--)
1208 var_arr[i].mem_edge_state = get_irg_initial_mem(current_ir_graph);
1212 /* Analyze call nodes to get information, if they store the address of a scalar.
1214 * @param *irn An ir node from the current_ir_graph.
1215 * @param *env The enviroment pointer.
1217 static void analyse_calls(ir_node *irn, void *ctx) {
1220 unsigned int acces_type;
1221 ir_node *param, *call_ptr, *blk;
1223 ir_entity *meth_ent;
1224 sels_t key_sels, *value_sels;
1225 call_access_t key_call, *value_call;
1226 value_arr_entry_t *val_arr;
1230 if(get_irn_op(irn) != op_Call)
1233 /* Calls that have a NoMem input do neither read nor write memory.
1234 We can completely ignore them here. */
1235 if (get_irn_op(get_Call_mem(irn)) == op_NoMem)
1238 /* We iterate over the parameters of this call nodes.*/
1239 for ( i = get_Call_n_params(irn) - 1; i >= 0; i--) {
1240 param = get_Call_param(irn, i);
1241 if(get_irn_op(param) == op_Sel) {
1242 /* We have found a parameter with operation sel.*/
1243 key_sels.sel = param;
1244 value_sels = set_find(env->set_sels, &key_sels, sizeof(key_sels), HASH_PTR(key_sels.sel));
1245 if(value_sels != NULL ) {
1247 /* We have found a call, that have as parameter a sel from our set_sels.*/
1248 call_ptr = get_Call_ptr(irn);
1249 op = get_irn_op(call_ptr);
1251 if(op == op_SymConst && get_SymConst_kind(call_ptr) == symconst_addr_ent) {
1252 meth_ent = get_SymConst_entity(call_ptr);
1253 /* we get the access type for our sel.*/
1254 acces_type = get_method_param_access(meth_ent, i);
1256 /* We can't analyze this function and we asume, that it store the address.*/
1257 acces_type = ptr_access_store;
1259 /* we save the access type and this call in the array allocated for this block.
1260 * The value number of this entity get us the position in the array to save this
1261 * information. Why we expect more calls as one we allocate a set.*/
1262 vnum = GET_ENT_VNUM(value_sels->ent);
1263 blk = get_nodes_block(irn);
1264 val_arr = get_irn_link(blk);
1266 if(val_arr[vnum].access_type > 3)
1267 /* The address of this entity have been stored.*/
1270 if(val_arr[vnum].calls == NULL)
1271 /* for this entity i have found the firs call in this block and we must allocate the set.*/
1272 val_arr[vnum].calls = new_set(call_cmp, 8);
1274 /* This call performs anything with the scalar and we must mark it.*/
1275 key_call.call = irn;
1276 key_call.access_type = acces_type;
1277 value_call = set_insert(val_arr[vnum].calls, &key_call, sizeof(key_call), HASH_PTR(key_call.call));
1279 if(value_call->access_type < acces_type)
1280 /* this case tread, when a call access an entity more at once.
1281 * Than we must save the highest access type.*/
1282 value_call->access_type = acces_type;
1285 /* This call save the address of our scalar and we can't
1286 * use the scalars of this entity for optimization as from now.
1288 val_arr[vnum].access_type = acces_type;
1294 static int set_block_dominated_first_access(ir_node *blk, int vnum, unsigned int access) {
1296 ir_node *idom, *succ;
1297 value_arr_entry_t *val_arr;
1300 idom = get_Block_idom(blk);
1301 for(i = get_Block_n_cfg_outs(idom) - 1; i >=1; i--) {
1302 succ = get_Block_cfg_out(idom, i);
1303 val_arr = get_irn_link(succ);
1304 if(val_arr[vnum].access_type < 3) {
1305 val_arr[vnum].access_type = access;
1311 /* Update the access information of a block if a predecessor of
1312 * this black have a higher access for an entity.
1314 * @param *irn An ir node from the current_ir_graph.
1315 * @param *env The enviroment pointer.
1317 static void set_block_access(ir_node *irn, void *ctx){
1319 value_arr_entry_t *val_arr, *val_arr_pred;
1320 ent_leaves_t *value_leaves;
1321 ir_node *pred, *pred_blk, *leave;
1326 val_arr = get_irn_link(irn);
1328 for( i = get_Block_n_cfgpreds(irn) - 1; i >= 0; i--) {
1329 /* We analyze the predecessors of this block to see if this block must
1331 pred = get_Block_cfgpred(irn, i);
1332 pred_blk = get_nodes_block(pred);
1334 val_arr_pred = get_irn_link(pred_blk);
1336 for(value_leaves = set_first(env->set_ent); value_leaves; value_leaves = set_next(env->set_ent)) {
1337 vnum = GET_ENT_VNUM(value_leaves->ent);
1339 if((get_Block_n_cfgpreds(irn) > 1) && (val_arr[vnum].access_type > 3))
1340 env->changes = set_block_dominated_first_access(irn, vnum, val_arr[vnum].access_type);
1342 if((val_arr_pred[vnum].access_type > 3) && (val_arr[vnum].access_type < 3)) {
1343 /* We have found a block for update it access and value number information.*/
1344 val_arr[vnum].access_type = val_arr_pred[vnum].access_type;
1345 /* We update the access information of all leave, that belong to
1348 for(leave = pset_first(value_leaves->leaves); leave; leave = pset_next(value_leaves->leaves))
1349 val_arr[GET_IRN_VNUM(leave)].access_type = val_arr[vnum].access_type;
1351 /* In this way can't be got the actuall number of value numbers.
1352 val_arr[env->vnum_state].access_type = val_arr_pred[env->vnum_state].access_type; */
1358 /* Free the allocated call sets.
1360 * @param irn A block form the ir graph.
1361 * @param env The enviroment pinter.
1363 static void free_call_info(ir_node *irn, void *ctx) {
1367 value_arr_entry_t *val_arr;
1370 val_arr = get_irn_link(irn);
1372 for(i = env->nvals + set_count(env->set_ent); i >= 0; i--) {
1373 if(val_arr[i].calls != NULL)
1375 del_set(val_arr[i].calls);
1379 static void print_block_state(ir_node *irn, void *ctx) {
1381 value_arr_entry_t *val_arr;
1382 ent_leaves_t *value_leaves;
1383 call_access_t *value_calls;
1388 val_arr = get_irn_link(irn);
1389 ir_printf("\n\nThe actual value number state of this block is: %i \n",
1390 val_arr[env->vnum_state].access_type - 1);
1392 for(value_leaves = set_first(env->set_ent); value_leaves; value_leaves = set_next(env->set_ent)) {
1394 vnum = GET_ENT_VNUM(value_leaves->ent);
1395 ir_printf("The entity %F access type in the block with nr %u is %i \n",
1396 value_leaves->ent, get_irn_node_nr(irn), val_arr[vnum].access_type);
1398 if(val_arr[vnum].calls != NULL)
1399 for(value_calls = set_first(val_arr[vnum].calls); value_calls; value_calls = set_next(val_arr[vnum].calls))
1401 ir_printf("A call with nr %i acess a element of this entity with access %u \n",
1402 get_irn_node_nr(value_calls->call), value_calls->access_type);
1407 /** Optimize the found scalar replacements.
1409 * @param set_sels A set with all entities, that
1411 * @param set_ent A set with all sels nodes,
1412 * that belong to our scalars.
1413 * @param vnum The number of scalars.
1415 static void do_data_flow_scalar_replacement(set *set_ent, set *set_sels, int vnum) {
1419 obstack_init(&env.obst);
1420 env.set_ent = set_ent;
1421 env.set_sels = set_sels;
1423 env.fix_phis = NULL;
1424 env.fix_syncs = NULL;
1425 env.gl_mem_vnum = vnum - 2;
1426 env.vnum_state = vnum - 1;
1427 /* nvals are vnum - 1, why we indicate with nvals the number
1428 * of memory edges we will produce. For vnum_state we don't
1429 * need to produce a memory edge.*/
1430 env.nvals = vnum - 1;
1433 /* first step: allocate the value arrays for every block */
1434 irg_block_walk_graph(current_ir_graph, NULL, alloc_value_arr, &env);
1436 /* second step: we analyze all calls, that have as parameter scalar(s).
1437 * We mark the calls, that save the address of a scalar and we
1438 * mark the entity owner of this scalar as not optimizeble by now.*/
1439 irg_walk_graph(current_ir_graph, NULL, analyse_calls, &env);
1441 while(env.changes) {
1446 * third step: walk over the blocks of a graph and update
1447 * the information for the access of our scalars.
1449 irg_block_walk_graph(current_ir_graph, NULL, set_block_access, &env);
1453 // if(get_firm_verbosity())
1454 /* Debug info to see if analyse_calls work properly.*/
1455 irg_block_walk_graph(current_ir_graph, NULL, print_block_state, &env);
1458 * fourth step: walk over the graph blockwise in topological order
1459 * and split the memrory edge.
1461 inc_irg_block_visited(current_ir_graph);
1462 irg_walk_blkwise_graph(current_ir_graph, NULL, split_memory_edge, &env);
1466 /* fifth step: fix all nodes, that have as predecessor Unknown.*/
1471 /* sixth step: sync memory enges for the end block.*/
1472 sync_mem_edges(&env);
1474 /*seventh step: free the allocated memory*/
1475 irg_block_walk_graph(current_ir_graph, NULL, free_call_info, &env);
1476 obstack_free(&env.obst, NULL);
1480 * Find possible scalar replacements
1482 * @param irg The current ir graph.
1484 void data_flow_scalar_replacement_opt(ir_graph *irg) {
1490 ent_leaves_t key_leaves, *value_leaves;
1493 if (! get_opt_scalar_replacement())
1496 set_sels = new_set(sels_cmp, 8);
1497 set_ent = new_set(ent_leaves_t_cmp, 8);
1499 /* Call algorithm that remove the critical edges of a ir graph. */
1500 remove_critical_cf_edges(irg);
1502 /* Call algorithm that computes the out edges.*/
1503 if (get_irg_outs_state(irg) != outs_consistent)
1504 compute_irg_outs(irg);
1506 /* Call algorithm that computes the loop information.*/
1507 compute_loop_info(irg);
1508 /* Call algorithm that computes the dominance information.*/
1511 /* Find possible scalar replacements */
1512 if (find_possible_replacements(irg)) {
1514 /* Insert in set the scalar replacements. */
1515 irg_frame = get_irg_frame(irg);
1517 for (i = 0 ; i < get_irn_n_outs(irg_frame); i++) {
1518 ir_node *succ = get_irn_out(irg_frame, i);
1520 if (get_irn_op(succ) == op_Sel) {
1521 ir_entity *ent = get_Sel_entity(succ);
1523 if (get_entity_link(ent) == NULL || get_entity_link(ent) == ADDRESS_TAKEN)
1525 /* we have found a entity, that have scalars and we insert it to our set_ent*/
1526 key_leaves.ent = ent;
1527 key_leaves.leaves = new_pset(leave_cmp, 8);
1528 value_leaves = set_insert(set_ent, &key_leaves, sizeof(key_leaves), HASH_PTR(ent));
1530 /* We allocate for every leave sel a vnum.*/
1531 vnum = allocate_value_numbers(set_sels, value_leaves->leaves, ent, vnum);
1535 if(get_firm_verbosity())
1536 printf("vnumber in data flow= %i\n", vnum);
1538 /* Allocate value number for the globule memory edge.
1539 * and a value number for the value numbers state.*/
1542 /* Allocate value numbers for the entities .*/
1543 for(i = vnum,value_leaves = set_first(set_ent); value_leaves; i++, value_leaves = set_next(set_ent))
1544 SET_ENT_VNUM(value_leaves->ent, i);
1547 do_data_flow_scalar_replacement(set_ent, set_sels, vnum);
1549 /*free the allocated memory.*/
1550 for(value_leaves = set_first(set_ent); value_leaves; value_leaves = set_next(set_ent))
1551 del_pset(value_leaves->leaves);