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
30 #include "iroptimize.h"
44 #include "analyze_irg_args.h"
46 #include "compute_loop_info.h"
50 #define SET_ENT_VNUM(ent, vnum) set_entity_link(ent, INT_TO_PTR(vnum))
51 #define GET_ENT_VNUM(ent) (unsigned)PTR_TO_INT(get_entity_link(ent))
52 #define SET_IRN_VNUM(irn, vnum) set_irn_link(irn, INT_TO_PTR(vnum))
53 #define GET_IRN_VNUM(irn) (unsigned)PTR_TO_INT(get_irn_link(irn))
57 typedef struct _ent_leaves_t{
58 ir_entity *ent; /**< An entity, that contains scalars for replace.*/
59 pset *leaves; /**< All leaves of this entity.*/
62 typedef struct _sels_t {
63 ir_node *sel; /**< A sel node, thats entity have scalars.*/
64 ir_entity *ent; /**< The entity of this sel node.*/
67 typedef struct _call_access_t {
68 ir_node *call; /**< A call node, that have as parameter a scalar.*/
69 unsigned int access_type; /**< The access type, with that this call access this scalar.*/
72 typedef struct _fixlist_entry_t {
73 ir_node *irn; /**< An ir node, that must be fixed.*/
74 unsigned int vnum; /**< The value number, that must became this ir node.*/
77 typedef struct _syncs_fixlist_entry_t {
78 ir_node *irn; /**< A sync node that must be fixed.*/
79 int *accessed_vnum; /**< A pointer to save an array with value numbers, that must became this sync.*/
80 }syncs_fixlist_entry_t;
82 /* A entry, that save the memory
83 * edge state and the access state for this leave
84 * int the array,that is created for every block.*/
85 typedef struct _leave_t {
86 ir_node *mem_edge_state; /**< memory state for this scalar in this block.*/
87 unsigned int access_type; /**< access state for this scalar in this block.*/
88 set *calls; /**< call nodes,that change this scalar in this block.*/
92 * A path element entry: it is either an entity
93 * or a tarval, because we evaluate only constant array
94 * accesses like a.b.c[8].d
102 * An access path, used to assign value numbers
103 * to variables that will be scalar replaced
105 typedef struct _path_t {
106 unsigned vnum; /**< the value number */
107 unsigned path_len; /**< the length of the access path */
108 path_elem_t path[1]; /**< the path */
112 * environment for memory walker
114 typedef struct _env_t {
115 struct obstack obst; /**< a obstack for the memory edge */
116 set *set_sels; /**< a set with all sels, that are reachable from an entity with a scalar.*/
117 set *set_ent; /**< a set with all entities that have one or more scalars.*/
118 fixlist_entry_t *fix_phis; /**< list of all Phi nodes that must be fixed */
119 fixlist_entry_t *fix_ls; /**< list of all Load or Store nodes that must be fixed */
120 syncs_fixlist_entry_t *fix_syncs; /**< list of all Sync nodes that must be fixed */
121 unsigned int nvals; /**< to save the number of scalars.*/
122 unsigned int gl_mem_vnum; /**< indicate the position of the globule memory edge state in var_arr.*/
123 unsigned int vnum_state; /**< indicate the position of the value number state in var_arr.*/
124 unsigned int changes; /**< to save if by anlyse_calls is changed anything.*/
130 * Compare two elements of the ent_leaves_t set.
132 * @return 0 if they are identically
134 static int ent_leaves_t_cmp(const void *elt, const void *key, size_t size)
136 const ent_leaves_t *c1 = elt;
137 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;
167 return c1->sel != c2->sel;
171 * Compare two elements of the leave_t set.
173 * @return 0 if they are identically
175 static int leave_cmp(const void *elt, const void *key)
177 ir_node *c1 = (ir_node *)elt;
178 ir_node *c2 = (ir_node *)key;
180 return get_Sel_entity(c1) != get_Sel_entity(c2);
184 * Compare two elements of the call_access_t set.
186 * @return 0 if they are identically
188 static int call_cmp(const void *elt, const void *key, size_t size)
190 const call_access_t *c1 = elt;
191 const call_access_t *c2 = key;
194 return c1->call != c2->call;
200 * @return 0 if they are identically
202 static int path_cmp(const void *elt, const void *key, size_t size)
204 const path_t *p1 = elt;
205 const path_t *p2 = key;
208 /* we can use memcmp here, because identical tarvals should have identical addresses */
209 return memcmp(p1->path, p2->path, p1->path_len * sizeof(p1->path[0]));
213 * Calculate a hash value for a path.
215 static unsigned path_hash(const path_t *path)
220 for (i = 0; i < path->path_len; ++i)
221 hash ^= (unsigned)PTR_TO_INT(path->path[i].ent);
227 * Returns non-zero, if all induces of a Sel node are constants.
229 * @param sel the Sel node that will be checked
231 static int is_const_sel(ir_node *sel) {
232 int i, n = get_Sel_n_indexs(sel);
234 for (i = 0; i < n; ++i) {
235 ir_node *idx = get_Sel_index(sel, i);
237 if (get_irn_op(idx) != op_Const)
244 * Returns non-zero, if the address of an entity
245 * represented by a Sel node (or it's successor Sels) is taken.
247 static int is_address_taken_2(ir_node *sel)
251 if (! is_const_sel(sel))
254 for (i = get_irn_n_outs(sel) - 1; i >= 0; --i) {
255 ir_node *succ = get_irn_out(sel, i);
257 switch (get_irn_opcode(succ)) {
259 /* ok, we just load from that entity */
263 /* check that Sel is not the Store's value */
264 if (get_Store_value(succ) == sel)
269 /* Check the Sel successor of Sel */
270 int res = is_address_taken_2(succ);
278 /* The address of an entity is given as a parameter.
279 * We analyzes that later and optimizes this scalar
285 /* another op, the address is taken */
293 * Link all Sels with the entity.
295 * @param ent the entity that will be scalar replaced
296 * @param sel a Sel node that selects some fields of this entity
298 static void link_all_leave_sels(ir_entity *ent, ir_node *sel)
302 n = get_irn_n_outs(sel);
303 for (i = 0; i < n; ++i) {
304 ir_node *succ = get_irn_out(sel, i);
306 if (get_irn_op(succ) == op_Sel)
307 link_all_leave_sels(ent, succ);
311 /* if Sel nodes with memory inputs are used, a entity can be
312 * visited more than once causing a ring here, so we use the
313 * node flag to mark linked nodes
315 if (irn_visited(sel))
319 * we link the sels to the entity.
321 set_irn_link(sel, get_entity_link(ent));
322 set_entity_link(ent, sel);
324 mark_irn_visited(sel);
327 /* we need a special address that serves as an address taken marker */
329 static void *ADDRESS_TAKEN = &_x;
332 * Find possible scalar replacements.
334 * @param irg an IR graph
336 * This function finds variables on the (members of the) frame type
337 * that can be scalar replaced, because their address is never taken.
338 * If such a variable is found, it's entity link will hold a list of all
339 * Sel nodes, that selects anythings of this entity.
340 * Otherwise, the link will be ADDRESS_TAKEN or NULL.
342 * @return non-zero if at least one entity could be replaced
345 static int find_possible_replacements(ir_graph *irg)
347 ir_node *irg_frame = get_irg_frame(irg);
351 inc_irg_visited(irg);
353 n = get_irn_n_outs(irg_frame);
356 * First, clear the link field of all interestingentities.
357 * Note that we did not rely on the fact that there is only
358 * one Sel node per entity, so we might access one entity
359 * more than once here.
360 * That's why we have need two loops.
362 for (i = 0; i < n; ++i) {
363 ir_node *succ = get_irn_out(irg_frame, i);
365 if (get_irn_op(succ) == op_Sel) {
366 ir_entity *ent = get_Sel_entity(succ);
367 set_entity_link(ent, NULL);
372 * Check the ir_graph for Sel nodes. If the entity of Sel
373 * isn't a scalar replacement set the link of this entity
374 * equal ADDRESS_TAKEN.
376 for (i = 0; i < n; ++i) {
377 ir_node *succ = get_irn_out(irg_frame, i);
379 if (get_irn_op(succ) == op_Sel) {
380 ir_entity *ent = get_Sel_entity(succ);
383 if (get_entity_link(ent) == ADDRESS_TAKEN)
387 * Beware: in rare cases even entities on the frame might be
388 * volatile. This might happen if the entity serves as a store
389 * to a value that must survive a exception. Do not optimize
390 * such entities away.
392 if (get_entity_volatility(ent) == volatility_is_volatile) {
393 set_entity_link(ent, ADDRESS_TAKEN);
397 ent_type = get_entity_type(ent);
399 /* we can handle arrays, structs and atomic types yet */
400 if (is_Array_type(ent_type) || is_Struct_type(ent_type) || is_atomic_type(ent_type)) {
401 if (is_address_taken_2(succ)) {
402 if (get_entity_link(ent)) /* killing one */
404 set_entity_link(ent, ADDRESS_TAKEN);
407 /* possible found one */
408 if (get_entity_link(ent) == NULL)
410 link_all_leave_sels(ent, succ);
419 static int is_leave_sel(ir_node *sel) {
423 for(i = get_irn_n_outs(sel) - 1; i >= 0; i--) {
424 succ = get_irn_out(sel, i);
425 if(get_irn_op(succ) == op_Sel)
433 * Return a path from the Sel node sel to it's root.
435 * @param sel the Sel node
436 * @param len the length of the path so far
438 static path_t *find_path(ir_node *sel, unsigned len)
442 ir_node *pred = get_Sel_ptr(sel);
444 /* the current Sel node will add some path elements */
445 n = get_Sel_n_indexs(sel);
448 if (get_irn_op(pred) != op_Sel) {
449 /* we found the root */
451 res = xmalloc(sizeof(*res) + (len - 1) * sizeof(res->path));
455 res = find_path(pred, len);
457 pos = res->path_len - len;
459 res->path[pos++].ent = get_Sel_entity(sel);
460 for (i = 0; i < n; ++i) {
461 ir_node *index = get_Sel_index(sel, i);
463 if(get_irn_op(index) == op_Const)
464 res->path[pos++].tv = get_Const_tarval(index);
470 * Allocate value numbers for the leaves
471 * in our found entities.
473 * @param sels a set that will contain all Sels that have a value number
474 * @param ent the entity that will be scalar replaced
475 * @param vnum the first value number we can assign
476 * @param modes a flexible array, containing all the modes of
479 * @return the next free value number
481 static unsigned allocate_value_numbers(set *set_sels, pset *leaves, ir_entity *ent, unsigned vnum)
486 set *pathes = new_set(path_cmp, 8);
488 /* visit all Sel nodes in the chain of the entity */
489 for (sel = get_entity_link(ent); sel; sel = next) {
490 next = get_irn_link(sel);
492 /* we save for every sel it root entity, why
493 * we need this information, when we split the memory edge,
494 * and we must mark this sel for later. */
497 set_insert(set_sels, &key_sels, sizeof(key_sels), HASH_PTR(sel));
499 if(! is_leave_sel(sel))
501 /* We have found a leave and we add it to the pset of this entity.*/
502 pset_insert(leaves, sel, HASH_PTR(get_Sel_entity(sel)));
504 key = find_path(sel, 0);
505 path = set_find(pathes, key, sizeof(*key) + sizeof(key->path[0]) * key->path_len, path_hash(key));
508 SET_IRN_VNUM(sel, path->vnum);
513 set_insert(pathes, key, sizeof(*key) + sizeof(key->path[0]) * key->path_len, path_hash(key));
515 SET_IRN_VNUM(sel, key->vnum);
521 set_entity_link(ent, NULL);
525 * Add a sync node to it fix list.
527 * @param sync The sync node, that myst be addet to the fix list.
528 * @param unk_vnum An array whit the value number, that are synced with this sync node.
529 * @param env The enviroment pinter.
531 static void add_sync_to_fixlist(ir_node *sync, int *unk_vnum, env_t *env) {
533 syncs_fixlist_entry_t *s;
535 s = obstack_alloc(&env->obst, sizeof(*s));
537 s->accessed_vnum = unk_vnum;
538 set_irn_link(sync, env->fix_syncs);
542 * Add a ir node to it fix list.
544 * @param irn The ir node, that myst be addet to the fix list.
545 * @param vnum The value number, that must baceme this ir node as predecessor later.
546 * @param env The enviroment pinter.
548 static void add_ls_to_fixlist(ir_node *irn, int vnum, env_t *env) {
552 l = obstack_alloc(&env->obst, sizeof(*l));
556 if(get_irn_op(irn) == op_Phi) {
557 set_irn_link(l->irn, env->fix_phis);
560 set_irn_link(l->irn, env->fix_ls);
565 static void add_mem_edge(value_arr_entry_t *val_arr, int vnum, ir_node ***in, int **accessed_vnum) {
567 if(val_arr[vnum].mem_edge_state != NULL)
568 ARR_APP1(ir_node *, *in, val_arr[vnum].mem_edge_state);
570 ARR_APP1(int, *accessed_vnum, vnum);
571 ARR_APP1(ir_node *, *in, new_Unknown(mode_M));
575 * The function handles the scalars, that wase stored
578 * @param blk The block, that must be handled.
579 * @param env The enviroment pinter.
582 /* Return the memory successor of the call node.*/
583 static ir_node *get_Call_mem_out(ir_node *call) {
588 for(i = get_irn_n_outs(call) - 1; i >= 0; i--) {
589 mem = get_irn_out(call, i);
590 if(get_irn_mode(mem) == mode_M)
593 /* is not reachable*/
598 static void sync_stored_scalars(ir_node *blk, env_t *env) {
601 int *unk_vnum; /**< An arraw, where are saved the value number, that
602 are synced from this sync node.*/
603 ent_leaves_t *value_ent;
604 value_arr_entry_t *val_arr_blk, *val_arr;
605 ir_node *pred, *leave, *sync, **in;
606 ir_node *sync_blk; /**< The block, where the sync node must be created.*/
609 val_arr_blk = get_irn_link(blk);
611 for(value_ent = set_first(env->set_ent); value_ent; value_ent = set_next(env->set_ent)) {
614 if(val_arr_blk[GET_ENT_VNUM(value_ent->ent)].access_type <= 3)
615 /* This entity is not stored in this block.*/
618 for(i = get_Block_n_cfgpreds(blk) - 1; i >= 0; i--) {
620 pred = get_Block_cfgpred(blk, i);
621 pred = get_nodes_block(pred);
622 val_arr = get_irn_link(pred);
624 if(val_arr[GET_ENT_VNUM(value_ent->ent)].access_type == SYNCED)
625 /* This entity was synced.*/
628 if(val_arr[GET_ENT_VNUM(value_ent->ent)].access_type <= 3) {
630 /* To avoid repeated sync of this entity in this block.*/
631 val_arr[GET_ENT_VNUM(value_ent->ent)].access_type = SYNCED;
632 /* In this predecessor block is this entity not acessed.
633 * We must sync in the end ot this block.*/
634 if(get_Block_n_cfgpreds(blk) > 1)
635 sync_blk = get_nodes_block(get_Block_cfgpred(blk, i));
639 val_arr = get_irn_link(sync_blk);
640 /* An array to save the memory edges, that must be
642 in = NEW_ARR_F(ir_node *, 1);
644 /* An array to save the value numbers,
645 * that must be repaired.*/
646 unk_vnum = NEW_ARR_F(int, 0);
647 /* The global memory edge.*/
648 if(val_arr[env->gl_mem_vnum].mem_edge_state == NULL)
649 in[0] = new_Unknown(mode_M);
651 in[0] = val_arr[env->gl_mem_vnum].mem_edge_state;
653 for(leave = pset_first(value_ent->leaves); leave; leave = pset_next(value_ent->leaves))
654 /* All this memory edges must be synced.*/
655 add_mem_edge(val_arr, GET_IRN_VNUM(leave), &in, &unk_vnum);
657 /* We create the sync and set it in the global memory state.*/
658 sync = new_r_Sync(current_ir_graph, sync_blk, ARR_LEN(in), in);
659 /* We must check this, why it is possible to get a Bad node
660 * form new_r_Sync(), when the node can be optimized.
661 * In this case we must do nothing.*/
662 if(get_irn_op(sync) == op_Sync) {
663 val_arr[env->gl_mem_vnum].mem_edge_state = sync;
664 /* We add this sync node to the sync's fix list.*/
665 add_sync_to_fixlist(val_arr[env->gl_mem_vnum].mem_edge_state, unk_vnum, env);
673 * The function split the memory edge of load and store nodes, that have
674 * as predecessor a scalar
676 * @param irn The node, that memory edge must be spleted.
677 * @param env The enviroment pinter.
679 static void split_ls_mem_edge(ir_node *irn, env_t *env) {
682 ir_node *leave, *irn_blk, *mem_state, *new_mem_state;
683 unsigned ent_vnum, sel_vnum, i;
684 value_arr_entry_t *val_arr;
685 sels_t key_sels, *value_sels;
686 ent_leaves_t key_ent, *value_ent;
688 op = get_irn_op(irn);
691 key_sels.sel = get_Load_ptr(irn);
693 key_sels.sel = get_Store_ptr(irn);
695 value_sels = set_find(env->set_sels, &key_sels, sizeof(key_sels), HASH_PTR(key_sels.sel));
697 if(value_sels != NULL) {
698 /* we have found a load or store, that use a sel of our set
699 * and we must split or extend, if the memory edge have been
700 * split for this sel, the memory edge.*/
702 key_ent.ent = value_sels->ent;
703 value_ent = set_find(env->set_ent, &key_ent, sizeof(key_ent), HASH_PTR(key_ent.ent));
704 /*To check if the enities set is right filled. */
705 assert(value_ent && " This sel's entity isn't int the entity set.");
707 leave = pset_find(value_ent->leaves, key_sels.sel, HASH_PTR(get_Sel_entity(key_sels.sel)));
708 /*To check if the leaves set is right filled. */
709 assert(leave && "Anything in data_flow_scalar_replacment algorithm is wrong.");
711 ent_vnum = GET_ENT_VNUM(value_ent->ent);
712 sel_vnum = GET_IRN_VNUM(leave);
713 irn_blk = get_nodes_block(irn);
714 val_arr = get_irn_link(irn_blk);
716 if(val_arr[ent_vnum].access_type == 0)
717 /* We have found a scalar, that address is not stored as jet.*/
720 /* This scalar have been stored.*/
721 i = env->gl_mem_vnum;
723 if(val_arr[i].mem_edge_state == NULL) {
724 /* We split now for this sel the memory edge in this block.*/
725 mem_state = new_Unknown(mode_M);
726 /* We must mark this node to fix later*/
727 add_ls_to_fixlist(irn, i, env);
730 /* We have split the memory edge and the current state is saved.*/
731 mem_state = val_arr[i].mem_edge_state;
733 /* We set this Load or Store to the memory edge of this
736 set_Load_mem(irn, mem_state);
738 set_Store_mem(irn, mem_state);
740 /* When we have split or extended the memory edge we must
741 * update the memory_edge_state of this sel*/
742 new_mem_state = get_irn_out(irn, 0);
743 if(get_irn_mode(new_mem_state) == mode_M)
744 val_arr[i].mem_edge_state = new_mem_state;
746 val_arr[i].mem_edge_state = get_irn_out(irn, 1);
751 * The function split the memory edge of phi nodes, that have
752 * as predecessor a scalar
754 * @param irn The phi node, that memory edge must be spleted.
755 * @param env The enviroment pinter.
757 static void split_phi_mem_edge(ir_node *irn, env_t *env) {
759 ir_node *irn_blk, *unk, *leave, **in;
761 ent_leaves_t *value_ent;
762 value_arr_entry_t *val_arr;
764 irn_blk = get_nodes_block(irn);
765 val_arr = get_irn_link(irn_blk);
767 n = get_Block_n_cfgpreds(irn_blk);
769 in = alloca(sizeof(*in) * n);
771 for(value_ent = set_first(env->set_ent); value_ent; value_ent = set_next(env->set_ent))
772 if(val_arr[GET_ENT_VNUM(value_ent->ent)].access_type < 3)
773 /* This scalar wasn't be saved and we need to produce a phi for it.*/
774 for(leave = pset_first(value_ent->leaves); leave; leave = pset_next(value_ent->leaves)){
776 unk = new_Unknown(mode_M);
777 for (j = n - 1; j >= 0; --j)
780 val_arr[GET_IRN_VNUM(leave)].mem_edge_state = new_r_Phi(current_ir_graph, irn_blk, n, in, mode_M);
782 add_ls_to_fixlist(val_arr[GET_IRN_VNUM(leave)].mem_edge_state, GET_IRN_VNUM(leave), env);
785 /* We use for the global memory the phi node, that
786 * is already available.*/
787 val_arr[env->gl_mem_vnum].mem_edge_state = irn;
791 * The function handles the call nodes, that have
792 * as parameter a scalar
794 * @param env The enviroment pinter.
795 * @param call The call node, that must be handled.
796 * @param accessed_entities A set wit all entities, that are accessed from this call node.*/
797 static void split_call_mem_edge(env_t *env, ir_node *call, pset *accessed_entities) {
799 ent_leaves_t key_ent, *value_ent;
800 value_arr_entry_t *val_arr;
801 call_access_t key_call, *value_call;
802 ir_node *call_blk, *new_mem_state, *leave;
806 int fix_irn = 0; /**< Set to 1 if we must add this call to it fix list.*/
807 int *accessed_leaves_vnum = NULL; /**< An arraw, where are saved the value number, that
808 are synced from call's sync node, if we need it.*/
810 if(get_irn_node_nr(call) == 2763)
813 call_blk = get_nodes_block(call);
814 val_arr = get_irn_link(call_blk);
815 /* An array to save the memory edges, that must be
817 in = NEW_ARR_F(ir_node *, 1);
818 /* An array to save the value numbers of the memory
819 * edges that must be repaired.*/
820 accessed_leaves_vnum = NEW_ARR_F(int, 0);
822 /* We get the memory successor of the call node.
823 * It is the new memory state for all synced memory
825 new_mem_state = get_Call_mem_out(call);
827 /* The global memory is the first predecessor of the create sync node.*/
828 if(val_arr[env->gl_mem_vnum].mem_edge_state == NULL) {
829 in[0] = new_Unknown(mode_M);
833 in[0] = val_arr[env->gl_mem_vnum].mem_edge_state;
836 for(ent = pset_first(accessed_entities); ent; ent = pset_next(accessed_entities)) {
837 /* Whit this loop we iterate all accessed entities from this call and collect
838 * all memory edges, that we must sync.*/
839 ent_vnum = GET_ENT_VNUM(ent);
841 key_call.call = call;
842 value_call = set_find(val_arr[ent_vnum].calls, &key_call, sizeof(key_call), HASH_PTR(key_call.call));
845 value_ent = set_find(env->set_ent, &key_ent, sizeof(key_ent), HASH_PTR(key_ent.ent));
847 if(val_arr[ent_vnum].access_type <= 3) {
848 /* This scalar's address wasn't stored in this block.*/
849 switch(value_call->access_type) {
851 case ptr_access_none :
852 /* In this case we have nothing to do.*/
855 case ptr_access_read:
856 case ptr_access_write:
858 /* All this cases must be traded equal.*/
860 for(leave = pset_first(value_ent->leaves); leave; leave = pset_next(value_ent->leaves)){
861 /* All this memory edges must be synced.*/
862 add_mem_edge(val_arr, GET_IRN_VNUM(leave), &in, &accessed_leaves_vnum);
864 /* We update the memory state of this leave.*/
865 if(value_call->access_type != ptr_access_read)
866 val_arr[GET_IRN_VNUM(leave)].mem_edge_state = new_mem_state;
875 /* We must update the global memory state.*/
876 val_arr[env->gl_mem_vnum].mem_edge_state = new_mem_state;
878 if(ARR_LEN(in) == 1) {
879 /* we must set the call memory to gobale momory*/
880 set_Call_mem(call,in[0]);
883 /* We add this call node to the call fix list..*/
884 add_ls_to_fixlist(call, env->gl_mem_vnum, env);
887 /* We create the sync and set it as memory predecessor of the call node.*/
888 sync = new_r_Sync(current_ir_graph, call_blk, ARR_LEN(in), in);
889 /* We must check this, why it is possible to get a Bad node
890 * form new_r_Sync(), when the node can be optimized.
891 * In this case we must do nothing.*/
892 if(get_irn_op(sync) == op_Sync) {
894 set_Call_mem(call, sync);
895 if(ARR_LEN(accessed_leaves_vnum))
896 /* We add this sync node to the sync's fix list.*/
897 add_sync_to_fixlist(sync, accessed_leaves_vnum, env);
904 * The function split the memory edge from the passed
905 * ir node if this is needed
907 * @param irn The node, that memory edge must be spleted.
908 * @param env The enviroment pinter.
910 static void split_memory_edge(ir_node *irn, void *ctx) {
913 ir_node *sel, *irn_blk;
915 sels_t key_sels, *value_sels;
916 value_arr_entry_t *val_arr;
917 pset *accessed_entities; /**< A set to save all entities accessed from a call.*/
921 op = get_irn_op(irn);
926 irn_blk = get_nodes_block(irn);
928 if (Block_not_block_visited(irn_blk)) {
929 /* We sync first the stored scalar address in this block.*/
930 mark_Block_block_visited(irn_blk);
931 sync_stored_scalars(irn_blk, env);
934 if(op == op_Load || op == op_Store)
936 split_ls_mem_edge(irn, env);
939 if (op == op_Phi && get_irn_mode(irn) == mode_M) {
941 * found a memory Phi: Here, we must create new Phi nodes
943 split_phi_mem_edge(irn, env);
948 /* Calls that have a NoMem input do neither read nor write memory.
949 We can completely ignore them here. */
950 if (get_irn_op(get_Call_mem(irn)) == op_NoMem)
953 /* We save in this set all entities,
954 * that are accessed from this call node.*/
955 accessed_entities = new_pset(ent_cmp, 8);
956 val_arr = get_irn_link(get_nodes_block(irn));
958 for ( i = get_Call_n_params(irn) - 1; i >= 0; i--) {
960 sel = get_Call_param(irn, i);
962 if(get_irn_op(sel) == op_Sel) {
964 value_sels = set_find(env->set_sels, &key_sels, sizeof(key_sels), HASH_PTR(key_sels.sel));
966 if(value_sels != NULL && val_arr[GET_ENT_VNUM(value_sels->ent)].access_type <= 3)
967 /* We save in this set all accessed entities from this call node whit
968 * access none, read, write or rw..*/
969 pset_insert(accessed_entities, value_sels->ent, HASH_PTR(value_sels->ent));
973 if(pset_count(accessed_entities))
974 split_call_mem_edge(env, irn, accessed_entities);
976 del_pset(accessed_entities);
983 * searches through blocks beginning from block for value
984 * vnum and return it.
986 * @param block A block from the current ir graph.
987 * @param vnum The value number, that must be found.
989 static ir_node *find_vnum_value(ir_node *block, unsigned vnum)
991 value_arr_entry_t *val_arr;
995 if (Block_not_block_visited(block)) {
996 mark_Block_block_visited(block);
998 val_arr = get_irn_link(block);
1000 if (val_arr[vnum].mem_edge_state)
1001 return val_arr[vnum].mem_edge_state;
1003 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1004 ir_node *pred = get_Block_cfgpred(block, i);
1006 res = find_vnum_value(get_nodes_block(pred), vnum);
1015 * fix the Load/Store or Call list
1017 * @param The enviroment pinter.
1019 static void fix_ls(env_t *env)
1022 ir_node *irn, *block, *pred, *val = NULL;
1026 for (l = env->fix_ls; l; l = get_irn_link(irn)) {
1029 op = get_irn_op(irn);
1030 block = get_nodes_block(irn);
1031 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1032 pred = get_Block_cfgpred(block, i);
1033 pred = get_nodes_block(pred);
1035 inc_irg_block_visited(current_ir_graph);
1036 val = find_vnum_value(pred, l->vnum);
1044 set_Store_mem(irn, val);
1047 set_Load_mem(irn, val);
1049 set_Call_mem(irn, val);
1057 * @param The enviroment pinter.
1059 static void fix_phis(env_t *env)
1062 ir_node *phi, *block, *pred, *val;
1065 for (l = env->fix_phis; l; l = get_irn_link(phi)) {
1068 block = get_nodes_block(phi);
1069 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1071 pred = get_Block_cfgpred(block, i);
1072 pred = get_nodes_block(pred);
1074 inc_irg_block_visited(current_ir_graph);
1075 val = find_vnum_value(pred, l->vnum);
1078 set_irn_n(phi, i, val);
1087 * @param The enviroment pinter.
1089 static void fix_syncs(env_t *env)
1091 syncs_fixlist_entry_t *l;
1092 ir_node *sync, *block, *pred, *val;
1096 for (l = env->fix_syncs; l; l = get_irn_link(sync)) {
1100 /* The sync block must have one predecessor, when it
1101 have unknown nodes as predecessor.*/
1102 block = get_nodes_block(sync);
1103 pred = get_Block_cfgpred(block, 0);
1104 pred = get_nodes_block(pred);
1106 /* We first repair the global memory edge at the first position of sync predecessors.*/
1107 if(get_irn_op(get_irn_n(sync, 0)) == op_Unknown) {
1108 inc_irg_block_visited(current_ir_graph);
1109 val = find_vnum_value(pred, env->gl_mem_vnum);
1112 set_irn_n(sync, 0, val);
1115 for (i = get_irn_arity(sync) - 1; i >= 1; --i) {
1116 /* We repair the leaves*/
1118 assert(k <= ARR_LEN(l->accessed_vnum) && "The algorythm for sync repair is wron");
1119 if(get_irn_op(get_irn_n(sync, i)) == op_Unknown) {
1120 inc_irg_block_visited(current_ir_graph);
1121 val = find_vnum_value(pred, l->accessed_vnum[k++]);
1124 set_irn_n(sync, i, val);
1127 DEL_ARR_F(l->accessed_vnum);
1131 * For the end node we must sync all memory edges.
1133 * @param The enviroment pinter.
1135 static void sync_mem_edges(env_t *env) {
1137 value_arr_entry_t *val_arr;
1138 ir_node **in, *sync, *Return, *Return_blk;
1139 int i, vnum, vnum_state;
1141 Return = get_Block_cfgpred(get_irg_end_block(current_ir_graph), 0);
1142 Return_blk = get_nodes_block(Return);
1143 val_arr = get_irn_link(Return_blk);
1147 for(i = 0; i <= (int)env->gl_mem_vnum; i++)
1148 /* we get the current state of non saved scalars.*/
1149 if(val_arr[i].access_type <= 3)
1152 /* We allocate the memory, that we need for the predecessors of the sync.*/
1153 in = xmalloc(sizeof(ir_node*) *vnum_state);
1155 /* The global memory edge is the first predecessor of this sync node.*/
1156 if(val_arr[env->gl_mem_vnum].mem_edge_state == NULL) {
1157 /* We must search through blocks for this memory state.*/
1158 inc_irg_block_visited(current_ir_graph);
1159 in[0] = find_vnum_value(Return_blk, env->gl_mem_vnum);
1162 in[0] = val_arr[env->gl_mem_vnum].mem_edge_state;
1165 for(i = 1, vnum = 0; vnum < (int)env->gl_mem_vnum; vnum++) {
1167 if(val_arr[vnum].access_type <= 3) {
1168 /* we add the non saved scalars as predecessors of the sync.*/
1170 if(val_arr[vnum].mem_edge_state == NULL) {
1171 /* We must search through blocks for this memory state.*/
1172 inc_irg_block_visited(current_ir_graph);
1173 in[i] = find_vnum_value(Return_blk, vnum);
1176 in[i] = val_arr[vnum].mem_edge_state;
1181 sync = new_r_Sync(current_ir_graph, Return_blk, vnum_state, in);
1182 set_Return_mem(Return, sync);
1188 * Walker: allocate the value array for every block.
1190 * @param block A block from the current ir graph for that must be allocated a value array.
1191 * @param ctx The enviroment pinter.
1193 static void alloc_value_arr(ir_node *block, void *ctx)
1198 value_arr_entry_t *var_arr = obstack_alloc(&env->obst, sizeof(value_arr_entry_t) *(env->nvals + set_count(env->set_ent) + 1));
1200 /* the value array is empty at start */
1201 memset(var_arr, 0, sizeof(value_arr_entry_t) * (env->nvals + set_count(env->set_ent) + 1));
1202 set_irn_link(block, var_arr);
1204 /* We set the block value number state to optimal and later we update this.*/
1205 var_arr[env->vnum_state].access_type = env->nvals;
1207 if(get_irg_start_block(current_ir_graph) == block)
1208 /* We initilize the startblocks array with the irg initilize memory, why
1209 * it must be the start point of all memory edges.*/
1210 for(i = (env->nvals + set_count(env->set_ent)) ; i >=0; i--)
1211 var_arr[i].mem_edge_state = get_irg_initial_mem(current_ir_graph);
1215 /* Analyze call nodes to get information, if they store the address of a scalar.
1217 * @param *irn An ir node from the current_ir_graph.
1218 * @param *env The enviroment pointer.
1220 static void analyse_calls(ir_node *irn, void *ctx) {
1223 unsigned int acces_type;
1224 ir_node *param, *call_ptr, *blk;
1226 ir_entity *meth_ent;
1227 sels_t key_sels, *value_sels;
1228 call_access_t key_call, *value_call;
1229 value_arr_entry_t *val_arr;
1233 if(get_irn_op(irn) != op_Call)
1236 /* Calls that have a NoMem input do neither read nor write memory.
1237 We can completely ignore them here. */
1238 if (get_irn_op(get_Call_mem(irn)) == op_NoMem)
1241 /* We iterate over the parameters of this call nodes.*/
1242 for ( i = get_Call_n_params(irn) - 1; i >= 0; i--) {
1243 param = get_Call_param(irn, i);
1244 if(get_irn_op(param) == op_Sel) {
1245 /* We have found a parameter with operation sel.*/
1246 key_sels.sel = param;
1247 value_sels = set_find(env->set_sels, &key_sels, sizeof(key_sels), HASH_PTR(key_sels.sel));
1248 if(value_sels != NULL ) {
1250 /* We have found a call, that have as parameter a sel from our set_sels.*/
1251 call_ptr = get_Call_ptr(irn);
1252 op = get_irn_op(call_ptr);
1254 if(op == op_SymConst && get_SymConst_kind(call_ptr) == symconst_addr_ent) {
1255 meth_ent = get_SymConst_entity(call_ptr);
1256 /* we get the access type for our sel.*/
1257 acces_type = get_method_param_access(meth_ent, i);
1259 /* We can't analyze this function and we asume, that it store the address.*/
1260 acces_type = ptr_access_store;
1262 /* we save the access type and this call in the array allocated for this block.
1263 * The value number of this entity get us the position in the array to save this
1264 * information. Why we expect more calls as one we allocate a set.*/
1265 vnum = GET_ENT_VNUM(value_sels->ent);
1266 blk = get_nodes_block(irn);
1267 val_arr = get_irn_link(blk);
1269 if(val_arr[vnum].access_type > 3)
1270 /* The address of this entity have been stored.*/
1273 if(val_arr[vnum].calls == NULL)
1274 /* for this entity i have found the firs call in this block and we must allocate the set.*/
1275 val_arr[vnum].calls = new_set(call_cmp, 8);
1277 /* This call performs anything with the scalar and we must mark it.*/
1278 key_call.call = irn;
1279 key_call.access_type = acces_type;
1280 value_call = set_insert(val_arr[vnum].calls, &key_call, sizeof(key_call), HASH_PTR(key_call.call));
1282 if(value_call->access_type < acces_type)
1283 /* this case tread, when a call access an entity more at once.
1284 * Than we must save the highest access type.*/
1285 value_call->access_type = acces_type;
1288 /* This call save the address of our scalar and we can't
1289 * use the scalars of this entity for optimization as from now.
1291 val_arr[vnum].access_type = acces_type;
1297 static int set_block_dominated_first_access(ir_node *blk, int vnum, unsigned int access) {
1299 ir_node *idom, *succ;
1300 value_arr_entry_t *val_arr;
1303 idom = get_Block_idom(blk);
1304 for(i = get_Block_n_cfg_outs(idom) - 1; i >=1; i--) {
1305 succ = get_Block_cfg_out(idom, i);
1306 val_arr = get_irn_link(succ);
1307 if(val_arr[vnum].access_type < 3) {
1308 val_arr[vnum].access_type = access;
1314 /* Update the access information of a block if a predecessor of
1315 * this black have a higher access for an entity.
1317 * @param *irn An ir node from the current_ir_graph.
1318 * @param *env The enviroment pointer.
1320 static void set_block_access(ir_node *irn, void *ctx){
1322 value_arr_entry_t *val_arr, *val_arr_pred;
1323 ent_leaves_t *value_leaves;
1324 ir_node *pred, *pred_blk, *leave;
1329 val_arr = get_irn_link(irn);
1331 for( i = get_Block_n_cfgpreds(irn) - 1; i >= 0; i--) {
1332 /* We analyze the predecessors of this block to see if this block must
1334 pred = get_Block_cfgpred(irn, i);
1335 pred_blk = get_nodes_block(pred);
1337 val_arr_pred = get_irn_link(pred_blk);
1339 for(value_leaves = set_first(env->set_ent); value_leaves; value_leaves = set_next(env->set_ent)) {
1340 vnum = GET_ENT_VNUM(value_leaves->ent);
1342 if((get_Block_n_cfgpreds(irn) > 1) && (val_arr[vnum].access_type > 3))
1343 env->changes = set_block_dominated_first_access(irn, vnum, val_arr[vnum].access_type);
1345 if((val_arr_pred[vnum].access_type > 3) && (val_arr[vnum].access_type < 3)) {
1346 /* We have found a block for update it access and value number information.*/
1347 val_arr[vnum].access_type = val_arr_pred[vnum].access_type;
1348 /* We update the access information of all leave, that belong to
1351 for(leave = pset_first(value_leaves->leaves); leave; leave = pset_next(value_leaves->leaves))
1352 val_arr[GET_IRN_VNUM(leave)].access_type = val_arr[vnum].access_type;
1354 /* In this way can't be got the actuall number of value numbers.
1355 val_arr[env->vnum_state].access_type = val_arr_pred[env->vnum_state].access_type; */
1361 /* Free the allocated call sets.
1363 * @param irn A block form the ir graph.
1364 * @param env The enviroment pinter.
1366 static void free_call_info(ir_node *irn, void *ctx) {
1370 value_arr_entry_t *val_arr;
1373 val_arr = get_irn_link(irn);
1375 for(i = env->nvals + set_count(env->set_ent); i >= 0; i--) {
1376 if(val_arr[i].calls != NULL)
1378 del_set(val_arr[i].calls);
1382 static void print_block_state(ir_node *irn, void *ctx) {
1384 value_arr_entry_t *val_arr;
1385 ent_leaves_t *value_leaves;
1386 call_access_t *value_calls;
1391 val_arr = get_irn_link(irn);
1392 ir_printf("\n\nThe actual value number state of this block is: %i \n",
1393 val_arr[env->vnum_state].access_type - 1);
1395 for(value_leaves = set_first(env->set_ent); value_leaves; value_leaves = set_next(env->set_ent)) {
1397 vnum = GET_ENT_VNUM(value_leaves->ent);
1398 ir_printf("The entity %F access type in the block with nr %u is %i \n",
1399 value_leaves->ent, get_irn_node_nr(irn), val_arr[vnum].access_type);
1401 if(val_arr[vnum].calls != NULL)
1402 for(value_calls = set_first(val_arr[vnum].calls); value_calls; value_calls = set_next(val_arr[vnum].calls))
1404 ir_printf("A call with nr %i acess a element of this entity with access %u \n",
1405 get_irn_node_nr(value_calls->call), value_calls->access_type);
1410 /** Optimize the found scalar replacements.
1412 * @param set_sels A set with all entities, that
1414 * @param set_ent A set with all sels nodes,
1415 * that belong to our scalars.
1416 * @param vnum The number of scalars.
1418 static void do_data_flow_scalar_replacement(set *set_ent, set *set_sels, int vnum) {
1422 obstack_init(&env.obst);
1423 env.set_ent = set_ent;
1424 env.set_sels = set_sels;
1426 env.fix_phis = NULL;
1427 env.fix_syncs = NULL;
1428 env.gl_mem_vnum = vnum - 2;
1429 env.vnum_state = vnum - 1;
1430 /* nvals are vnum - 1, why we indicate with nvals the number
1431 * of memory edges we will produce. For vnum_state we don't
1432 * need to produce a memory edge.*/
1433 env.nvals = vnum - 1;
1436 /* first step: allocate the value arrays for every block */
1437 irg_block_walk_graph(current_ir_graph, NULL, alloc_value_arr, &env);
1439 /* second step: we analyze all calls, that have as parameter scalar(s).
1440 * We mark the calls, that save the address of a scalar and we
1441 * mark the entity owner of this scalar as not optimizeble by now.*/
1442 irg_walk_graph(current_ir_graph, NULL, analyse_calls, &env);
1444 while(env.changes) {
1449 * third step: walk over the blocks of a graph and update
1450 * the information for the access of our scalars.
1452 irg_block_walk_graph(current_ir_graph, NULL, set_block_access, &env);
1456 // if(get_firm_verbosity())
1457 /* Debug info to see if analyse_calls work properly.*/
1458 irg_block_walk_graph(current_ir_graph, NULL, print_block_state, &env);
1461 * fourth step: walk over the graph blockwise in topological order
1462 * and split the memrory edge.
1464 inc_irg_block_visited(current_ir_graph);
1465 irg_walk_blkwise_graph(current_ir_graph, NULL, split_memory_edge, &env);
1469 /* fifth step: fix all nodes, that have as predecessor Unknown.*/
1474 /* sixth step: sync memory enges for the end block.*/
1475 sync_mem_edges(&env);
1477 /*seventh step: free the allocated memory*/
1478 irg_block_walk_graph(current_ir_graph, NULL, free_call_info, &env);
1479 obstack_free(&env.obst, NULL);
1483 * Find possible scalar replacements
1485 * @param irg The current ir graph.
1487 void data_flow_scalar_replacement_opt(ir_graph *irg) {
1493 ent_leaves_t key_leaves, *value_leaves;
1496 if (! get_opt_scalar_replacement())
1499 set_sels = new_set(sels_cmp, 8);
1500 set_ent = new_set(ent_leaves_t_cmp, 8);
1502 /* Call algorithm that remove the critical edges of a ir graph. */
1503 remove_critical_cf_edges(irg);
1505 /* Call algorithm that computes the out edges.*/
1506 if (get_irg_outs_state(irg) != outs_consistent)
1507 compute_irg_outs(irg);
1509 /* Call algorithm that computes the loop information.*/
1510 compute_loop_info(irg);
1511 /* Call algorithm that computes the dominance information.*/
1514 /* Find possible scalar replacements */
1515 if (find_possible_replacements(irg)) {
1517 /* Insert in set the scalar replacements. */
1518 irg_frame = get_irg_frame(irg);
1520 for (i = 0 ; i < get_irn_n_outs(irg_frame); i++) {
1521 ir_node *succ = get_irn_out(irg_frame, i);
1523 if (get_irn_op(succ) == op_Sel) {
1524 ir_entity *ent = get_Sel_entity(succ);
1526 if (get_entity_link(ent) == NULL || get_entity_link(ent) == ADDRESS_TAKEN)
1528 /* we have found a entity, that have scalars and we insert it to our set_ent*/
1529 key_leaves.ent = ent;
1530 key_leaves.leaves = new_pset(leave_cmp, 8);
1531 value_leaves = set_insert(set_ent, &key_leaves, sizeof(key_leaves), HASH_PTR(ent));
1533 /* We allocate for every leave sel a vnum.*/
1534 vnum = allocate_value_numbers(set_sels, value_leaves->leaves, ent, vnum);
1538 if(get_firm_verbosity())
1539 printf("vnumber in data flow= %i\n", vnum);
1541 /* Allocate value number for the globule memory edge.
1542 * and a value number for the value numbers state.*/
1545 /* Allocate value numbers for the entities .*/
1546 for(i = vnum,value_leaves = set_first(set_ent); value_leaves; i++, value_leaves = set_next(set_ent))
1547 SET_ENT_VNUM(value_leaves->ent, i);
1550 do_data_flow_scalar_replacement(set_ent, set_sels, vnum);
1552 /*free the allocated memory.*/
1553 for(value_leaves = set_first(set_ent); value_leaves; value_leaves = set_next(set_ent))
1554 del_pset(value_leaves->leaves);