2 * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief MIN-algorithm with some twists (aka belady spiller v3)
23 * @author Matthias Braun
28 * - handle phis correctly, decide whether we should spill them ~ok
29 * - merge multiple start worksets of blocks ~ok
30 * - how to and when to do the tentative phase...
44 #include "iredges_t.h"
52 #include "bespilloptions.h"
53 #include "besched_t.h"
57 #define EXPENSIVE_CHECKS
60 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
62 typedef struct loop_edge_t loop_edge_t;
69 typedef struct loop_info_t loop_info_t;
72 loop_edge_t *entry_edges;
73 loop_edge_t *exit_edges;
74 size_t max_register_pressure;
77 typedef struct worklist_entry_t worklist_entry_t;
78 struct worklist_entry_t {
79 struct list_head head;
81 ir_node *reload_point;
82 ir_loop *unused_livethrough_loop;
85 typedef struct worklist_t worklist_t;
87 struct list_head live_values;
92 typedef struct block_info_t block_info_t;
94 worklist_t *start_worklist;
95 worklist_t *end_worklist;
98 static const arch_env_t *arch_env;
99 static const arch_register_class_t *cls;
100 static struct obstack obst;
101 static spill_env_t *senv;
102 static size_t n_regs;
103 static size_t max_register_pressure;
104 static bool tentative_mode;
105 static bool should_have_reached_fixpoint;
106 static bool do_push_unused_livethroughs;
107 static ir_exec_freq *exec_freq;
108 static ir_visited_t worklist_visited;
110 static worklist_t *new_worklist(void)
112 worklist_t *worklist = obstack_alloc(&obst, sizeof(worklist[0]));
113 memset(worklist, 0, sizeof(worklist[0]));
115 INIT_LIST_HEAD(&worklist->live_values);
116 worklist->n_live_values = 0;
121 static void mark_irn_not_visited(ir_node *node)
123 set_irn_visited(node, get_irg_visited(current_ir_graph) - 1);
126 static bool worklist_contains(const ir_node *node)
128 return irn_visited(node);
131 static block_info_t *get_block_info(ir_node *block)
133 block_info_t *info = get_irn_link(block);
137 info = obstack_alloc(&obst, sizeof(info[0]));
138 memset(info, 0, sizeof(info[0]));
139 set_irn_link(block, info);
143 static void deactivate_worklist(const worklist_t *worklist)
145 struct list_head *entry;
147 list_for_each(entry, &worklist->live_values) {
148 worklist_entry_t *wl_entry
149 = list_entry(entry, worklist_entry_t, head);
150 assert(worklist_contains(wl_entry->value));
151 mark_irn_not_visited(wl_entry->value);
152 set_irn_link(wl_entry->value, NULL);
156 static void activate_worklist(const worklist_t *worklist)
158 struct list_head *entry;
160 list_for_each(entry, &worklist->live_values) {
161 worklist_entry_t *wl_entry
162 = list_entry(entry, worklist_entry_t, head);
163 assert(!worklist_contains(wl_entry->value));
164 mark_irn_visited(wl_entry->value);
165 set_irn_link(wl_entry->value, wl_entry);
170 * duplicate a worklist and directly activates it
172 static void fill_and_activate_worklist(worklist_t *new_worklist,
173 const worklist_t *worklist, ir_node *block, ir_node *succ_block,
176 ir_node *reload_point = NULL;
177 size_t n_live_values = 0;
178 struct list_head *entry;
180 if (succ_block != NULL &&
181 (get_Block_n_cfgpreds(succ_block) > 1
182 || get_irn_n_edges_kind(block, EDGE_KIND_BLOCK) > 1)) {
183 reload_point = be_get_end_of_block_insertion_point(block);
186 list_for_each(entry, &worklist->live_values) {
187 worklist_entry_t *wl_entry = list_entry(entry, worklist_entry_t, head);
188 ir_node *value = wl_entry->value;
189 worklist_entry_t *new_entry;
191 if (new_worklist->n_live_values >= n_regs)
194 if (is_Phi(value) && get_nodes_block(value) == succ_block) {
195 value = get_Phi_pred(value, succ_pos);
197 /* can happen for unknown phi preds */
198 if (!arch_irn_consider_in_reg_alloc(arch_env, cls, value))
202 if (irn_visited_else_mark(value))
205 new_entry = obstack_alloc(&obst, sizeof(new_entry[0]));
206 memset(new_entry, 0, sizeof(new_entry[0]));
208 new_entry->value = value;
209 if (reload_point != NULL) {
210 new_entry->reload_point = reload_point;
212 new_entry->reload_point = wl_entry->reload_point;
215 list_add_tail(&new_entry->head, &new_worklist->live_values);
218 set_irn_link(value, new_entry);
219 new_worklist->n_live_values++;
223 static worklist_t *duplicate_worklist(const worklist_t *worklist)
225 worklist_t *new_worklist;
226 struct list_head *entry;
228 new_worklist = obstack_alloc(&obst, sizeof(new_worklist[0]));
229 memset(new_worklist, 0, sizeof(new_worklist[0]));
230 INIT_LIST_HEAD(&new_worklist->live_values);
231 new_worklist->n_live_values = worklist->n_live_values;
233 list_for_each(entry, &worklist->live_values) {
234 worklist_entry_t *wl_entry = list_entry(entry, worklist_entry_t, head);
235 worklist_entry_t *new_entry
236 = obstack_alloc(&obst, sizeof(new_entry[0]));
238 memcpy(new_entry, wl_entry, sizeof(new_entry[0]));
239 list_add_tail(&new_entry->head, &new_worklist->live_values);
246 static void print_worklist(const worklist_t *worklist, int level)
248 struct list_head *entry;
250 DB((dbg, level, "%d values: ", worklist->n_live_values));
251 list_for_each(entry, &worklist->live_values) {
252 worklist_entry_t *wl_entry
253 = list_entry(entry, worklist_entry_t, head);
255 DB((dbg, level, "%+F ", wl_entry->value));
262 static void clear_loop_info(ir_loop *loop)
264 int n_elements = get_loop_n_elements(loop);
268 for (i = 0; i < n_elements; ++i) {
269 loop_element element = get_loop_element(loop, i);
270 if (*element.kind != k_ir_loop)
273 clear_loop_info(element.son);
277 static loop_info_t *get_loop_info(ir_loop *loop)
279 loop_info_t *info = loop->link;
283 info = obstack_alloc(&obst, sizeof(info[0]));
284 memset(info, 0, sizeof(info[0]));
291 * constructs loop in+out edges when called in a block walker
293 static void construct_loop_edges(ir_node *block, void *data)
295 int n_cfgpreds = get_Block_n_cfgpreds(block);
296 ir_loop *loop = get_irn_loop(block);
301 for (i = 0; i < n_cfgpreds; ++i) {
302 ir_node *cfgpred_block = get_Block_cfgpred_block(block, i);
303 ir_loop *cfgpred_loop = get_irn_loop(cfgpred_block);
308 if (cfgpred_loop == loop)
311 /* critical edges are splitted, so we can't jump from 1 loop
312 * directly into another without going out of it */
313 assert(get_loop_depth(cfgpred_loop) != get_loop_depth(loop));
315 /* edge out of loop */
316 if (get_loop_depth(cfgpred_loop) > get_loop_depth(loop)) {
321 is_exit_edge = false;
326 /* this might be a jump out/in multiple loops */
328 loop_info_t *l_info = get_loop_info(l);
330 edge = obstack_alloc(&obst, sizeof(edge[0]));
331 memset(edge, 0, sizeof(edge[0]));
336 edge->next = l_info->exit_edges;
337 l_info->exit_edges = edge;
338 assert(get_loop_depth(loop) < get_loop_depth(l));
340 edge->next = l_info->entry_edges;
341 l_info->entry_edges = edge;
342 assert(get_loop_depth(cfgpred_loop) < get_loop_depth(l));
344 l = get_loop_outer_loop(l);
352 static void place_reload(worklist_entry_t *entry)
357 DB((dbg, LEVEL_1, "reload of %+F before %+F\n", entry->value,
358 entry->reload_point));
359 assert(entry->reload_point != NULL);
360 be_add_reload(senv, entry->value, entry->reload_point, cls, 1);
361 entry->reload_point = NULL;
365 * makes sure the worklist contains not more than n_regs - room_needed entries
367 static void make_room(worklist_t *worklist, size_t room_needed)
371 struct list_head *entry;
373 spills_needed = worklist->n_live_values + room_needed - n_regs;
374 if (spills_needed <= 0)
377 entry = worklist->live_values.next;
378 for(i = spills_needed; i > 0; --i) {
379 struct list_head *next = entry->next;
380 worklist_entry_t *wl_entry
381 = list_entry(entry, worklist_entry_t, head);
383 assert(worklist_contains(wl_entry->value));
384 mark_irn_not_visited(wl_entry->value);
385 place_reload(wl_entry);
390 worklist->n_live_values -= spills_needed;
394 * a value was used, so bring it to the back of the worklist (which might
395 * result in a spill of another value).
397 static void val_used(worklist_t *worklist, ir_node *value, ir_node *sched_point)
399 /* already in the worklist? move around, otherwise add at back */
400 worklist_entry_t *entry = get_irn_link(value);
402 assert(arch_irn_consider_in_reg_alloc(arch_env, cls, value));
404 if (worklist_contains(value)) {
405 assert(entry != NULL);
407 list_del(&entry->head);
410 entry = obstack_alloc(&obst, sizeof(entry[0]));
411 memset(entry, 0, sizeof(entry[0]));
413 entry->value = value;
414 set_irn_link(value, entry);
417 ++worklist->n_live_values;
418 mark_irn_visited(value);
421 entry->reload_point = sched_point;
422 list_add_tail(&entry->head, &worklist->live_values);
425 static void worklist_remove(worklist_t *worklist, ir_node *value)
427 worklist_entry_t *entry = get_irn_link(value);
428 assert(entry != NULL);
429 list_del(&entry->head);
430 --worklist->n_live_values;
432 assert(worklist_contains(value));
433 mark_irn_not_visited(value);
436 static void update_max_pressure(worklist_t *worklist)
438 if (worklist->n_live_values > max_register_pressure)
439 max_register_pressure = worklist->n_live_values;
442 static void do_spilling(ir_node *block, worklist_t *worklist)
446 assert(worklist != NULL);
448 sched_foreach_reverse(block, node) {
452 DB((dbg, LEVEL_2, "\t%+F... ", node));
453 update_max_pressure(worklist);
457 /* TODO: if we have some free registers, then we could decide to
458 * not spill some phis (but not for phis where at least 1 input is
461 /* we have to spill all phis that are not live */
462 sched_foreach_reverse_from(node, node2) {
463 assert(is_Phi(node2));
465 if (worklist_contains(node2))
467 if (!arch_irn_consider_in_reg_alloc(arch_env, cls, node2))
471 be_spill_phi(senv, node2);
473 DB((dbg, LEVEL_2, "\n"));
477 /* remove values defined by this instruction from the workset. Values
478 * defined but not in the workset need free registers */
479 if (get_irn_mode(node) == mode_T) {
480 const ir_edge_t *edge;
482 foreach_out_edge(node, edge) {
483 ir_node *proj = get_edge_src_irn(edge);
484 if (!arch_irn_consider_in_reg_alloc(arch_env, cls, proj))
486 if (worklist_contains(proj)) {
487 worklist_remove(worklist, proj);
492 } else if (arch_irn_consider_in_reg_alloc(arch_env, cls, node)) {
493 if (worklist_contains(node)) {
494 worklist_remove(worklist, node);
500 /* make sure we have enough free registers for the spills */
501 make_room(worklist, n_defs);
503 /* put all values used by the instruction into the workset */
504 arity = get_irn_arity(node);
505 for(i = 0; i < arity; ++i) {
506 ir_node *use = get_irn_n(node, i);
508 if (!arch_irn_consider_in_reg_alloc(arch_env, cls, use))
511 val_used(worklist, use, node);
514 /* we might have too many values in the worklist now and need to spill
516 make_room(worklist, 0);
519 print_worklist(worklist, LEVEL_2);
520 DB((dbg, LEVEL_2, "\n"));
524 update_max_pressure(worklist);
527 DB((dbg, LEVEL_1, "worklist at begin of %+F:", block));
528 print_worklist(worklist, LEVEL_1);
529 DB((dbg, LEVEL_1, "\n"));
533 static bool worklists_equal(const worklist_t *wl1, const worklist_t *wl2)
535 const struct list_head *i1 = &wl1->live_values;
536 const struct list_head *i2 = &wl2->live_values;
538 for ( ; i1 != &wl1->live_values && i2 != &wl2->live_values;
539 i1 = i1->next, i2 = i2->next) {
540 worklist_entry_t *entry1 = list_entry(i1, worklist_entry_t, head);
541 worklist_entry_t *entry2 = list_entry(i2, worklist_entry_t, head);
543 if (entry1->value != entry2->value)
546 /* both lists at the end */
547 if (i1 != &wl1->live_values || i2 != &wl2->live_values)
553 static bool fill_start_worklist(worklist_t *new_worklist, ir_node *block)
555 double best_execfreq = -1;
556 worklist_t *best_worklist = NULL;
557 ir_node *best_succ_block = NULL;
559 const ir_edge_t *edge;
561 /* construct worklist */
562 foreach_block_succ(block, edge) {
563 ir_node *succ_block = get_edge_src_irn(edge);
564 double execfreq = get_block_execfreq(exec_freq, succ_block);
565 block_info_t *block_info;
566 worklist_t *succ_worklist;
568 if (execfreq < best_execfreq)
571 block_info = get_block_info(succ_block);
572 succ_worklist = block_info->start_worklist;
574 if (succ_worklist == NULL || succ_worklist->visited >= worklist_visited)
577 best_execfreq = execfreq;
578 best_worklist = succ_worklist;
579 best_succ_block = succ_block;
580 best_pos = get_edge_src_pos(edge);
583 if (best_worklist == NULL)
585 best_worklist->visited = worklist_visited;
587 fill_and_activate_worklist(new_worklist, best_worklist, block,
588 best_succ_block, best_pos);
592 static worklist_t *construct_start_worklist(ir_node *block)
594 worklist_t *worklist = new_worklist();
598 while(fill_start_worklist(worklist, block)) {
599 if (worklist->n_live_values >= n_regs)
606 static void process_block(ir_node *block, void *env)
608 block_info_t *block_info;
609 worklist_t *worklist;
613 DB((dbg, LEVEL_1, "Processing %+F\n", block));
615 worklist = construct_start_worklist(block);
616 block_info = get_block_info(block);
619 DB((dbg, LEVEL_1, "worklist at end of %+F:", block));
620 print_worklist(worklist, LEVEL_1);
621 DB((dbg, LEVEL_1, "\n"));
623 if (should_have_reached_fixpoint) {
624 assert(worklists_equal(block_info->end_worklist, worklist));
627 block_info->end_worklist = duplicate_worklist(worklist);
629 do_spilling(block, worklist);
630 deactivate_worklist(worklist);
633 if (should_have_reached_fixpoint) {
634 assert(worklists_equal(block_info->start_worklist, worklist));
637 block_info->start_worklist = worklist;
639 /* we shouldn't have any live values left at the start block */
640 n_preds = get_Block_n_cfgpreds(block);
641 //assert(n_preds != 0 || worklist->n_live_values == 0);
644 typedef struct block_or_loop_t block_or_loop_t;
645 struct block_or_loop_t {
653 static block_or_loop_t *loop_blocks;
654 static ir_loop *current_loop;
656 static void find_blocks(ir_node *block);
658 static void find_in_loop(ir_loop *loop, ir_node *entry)
660 loop_info_t *loop_info = get_loop_info(loop);
661 block_or_loop_t block_or_loop;
664 /* simply mark 1 block in the loop to indicate that the loop was already
666 ir_node *some_block = loop_info->entry_edges->block;
667 if (Block_block_visited(some_block))
670 block_or_loop.v.loop = loop;
671 block_or_loop.is_loop = true;
672 ARR_APP1(block_or_loop_t, loop_blocks, block_or_loop);
676 /* we should be 1 of the entry blocks */
677 loop_edge_t *edge = loop_info->entry_edges;
679 for ( ; edge != NULL; edge = edge->next) {
680 if (edge->block == entry)
689 /* check all loop successors */
690 for (edge = loop_info->exit_edges; edge != NULL; edge = edge->next) {
691 ir_node *succ = edge->block;
692 ir_loop *succ_loop = get_irn_loop(succ);
694 if (succ_loop == current_loop) {
697 assert(get_loop_depth(succ_loop) < get_loop_depth(current_loop));
702 static void find_blocks(ir_node *block)
704 const ir_edge_t *edge;
705 block_or_loop_t block_or_loop;
707 if (Block_block_visited(block))
710 block_or_loop.v.block = block;
711 block_or_loop.is_loop = false;
713 ARR_APP1(block_or_loop_t, loop_blocks, block_or_loop);
714 mark_Block_block_visited(block);
716 foreach_block_succ(block, edge) {
717 ir_node *succ = get_edge_src_irn(edge);
719 /* is it part of the current loop? */
720 ir_loop *loop = get_irn_loop(succ);
721 if (loop != current_loop) {
723 if (get_loop_depth(loop) > get_loop_depth(current_loop)) {
724 find_in_loop(loop, succ);
726 /* parent loop: we're not interested in the block */
735 * append an entry to a worklist. WARNING: The entry must not already be in the
738 static void worklist_append(worklist_t *worklist, ir_node *value,
739 ir_node *reload_point,
740 ir_loop *unused_livethrough_loop)
742 worklist_entry_t *entry = obstack_alloc(&obst, sizeof(entry[0]));
743 memset(entry, 0, sizeof(entry[0]));
745 #ifdef EXPENSIVE_CHECKS
747 struct list_head *entry;
748 list_for_each(entry, &worklist->live_values) {
749 worklist_entry_t *wl_entry
750 = list_entry(entry, worklist_entry_t, head);
751 assert(wl_entry->value != value);
756 entry->value = value;
757 entry->reload_point = reload_point;
758 entry->unused_livethrough_loop = unused_livethrough_loop;
759 list_add_tail(&entry->head, &worklist->live_values);
760 ++worklist->n_live_values;
761 assert(worklist->n_live_values <= n_regs);
764 static void push_unused_livethrough(loop_info_t *loop_info, ir_node *value)
769 /* add the value to all loop exit and entry blocks */
770 for (edge = loop_info->exit_edges; edge != NULL; edge = edge->next) {
772 = get_Block_cfgpred_block(edge->block, edge->pos);
773 const block_info_t *info = get_block_info(block);
774 worklist_t *worklist = info->end_worklist;
775 ir_node *reload_point = NULL;
777 if (worklist->visited >= worklist_visited)
779 worklist->visited = worklist_visited;
781 /* TODO: we need a smarter mechanism here, that makes the reloader place
782 * reload nodes on all loop exits... */
784 worklist_append(worklist, value, reload_point, loop_info->loop);
786 edge = loop_info->entry_edges;
787 for ( ; edge != NULL; edge = edge->next) {
788 ir_node *entry_block = edge->block;
789 const block_info_t *info = get_block_info(entry_block);
790 worklist_t *worklist = info->start_worklist;
792 ir_node *reload_point;
794 if (worklist->visited >= worklist_visited)
796 worklist->visited = worklist_visited;
798 pred_block = get_Block_cfgpred_block(entry_block, edge->pos);
799 reload_point = be_get_end_of_block_insertion_point(pred_block);
801 worklist_append(worklist, value, reload_point, loop_info->loop);
804 set_irn_link(value, NULL);
805 ++loop_info->max_register_pressure;
808 static void push_unused_livethroughs(loop_info_t *loop_info)
812 /* we can only push unused livethroughs if register pressure inside the loop
814 if (loop_info->max_register_pressure >= n_regs)
817 /* find unused livethroughs: register pressure in the loop was low enough
818 * which means that we had no spills which implies that at every point in
820 for (edge = loop_info->exit_edges; edge != NULL; edge = edge->next) {
821 ir_node *block = edge->block;
822 const block_info_t *info = get_block_info(block);
823 worklist_t *start_worklist = info->start_worklist;
825 const block_info_t *exit_info;
826 worklist_t *end_worklist;
827 struct list_head *entry;
829 if (start_worklist == NULL)
832 exit_block = get_Block_cfgpred_block(edge->block, edge->pos);
833 exit_info = get_block_info(exit_block);
834 end_worklist = exit_info->end_worklist;
836 activate_worklist(end_worklist);
837 /* all values contained in the start_worklist, which are not available
838 * in the end_worklist, must be unused livethroughs */
840 list_for_each(entry, &start_worklist->live_values) {
841 worklist_entry_t *wl_entry
842 = list_entry(entry, worklist_entry_t, head);
843 ir_node *value = wl_entry->value;
845 if (loop_info->max_register_pressure >= n_regs)
849 if (!worklist_contains(value)) {
850 /* add it to all loop exits */
851 DB((dbg, LEVEL_2, "Found unused livethrough: %+F (regpressure in loop %d)\n", value, loop_info->max_register_pressure));
852 push_unused_livethrough(loop_info, value);
853 /* the value should now be part of end_worklist, so mark it */
854 mark_irn_visited(value);
858 deactivate_worklist(end_worklist);
862 static void process_block_or_loop(const block_or_loop_t *block_or_loop)
864 if (block_or_loop->is_loop) {
865 loop_info_t *loop_info = get_loop_info(block_or_loop->v.loop);
867 if (do_push_unused_livethroughs)
868 push_unused_livethroughs(loop_info);
870 if (loop_info->max_register_pressure > max_register_pressure)
871 max_register_pressure = loop_info->max_register_pressure;
875 process_block(block_or_loop->v.block, NULL);
878 static void process_loop(ir_loop *loop)
880 int n_elements = get_loop_n_elements(loop);
882 loop_info_t *loop_info;
886 /* first handle all sub-loops */
887 for (i = 0; i < n_elements; ++i) {
888 loop_element element = get_loop_element(loop, i);
889 if (*element.kind != k_ir_loop)
892 process_loop(element.son);
895 /* create a postorder of the blocks */
896 loop_info = get_loop_info(loop);
897 edge = loop_info->entry_edges;
899 some_block = edge->block;
901 assert(loop == get_irg_loop(current_ir_graph));
902 some_block = get_irg_start_block(current_ir_graph);
905 loop_blocks = NEW_ARR_F(block_or_loop_t,0);
908 ir_reserve_resources(current_ir_graph, IR_RESOURCE_BLOCK_VISITED);
909 inc_irg_block_visited(current_ir_graph);
910 find_blocks(some_block);
911 /* for endless loops the end-block might be unreachable */
912 if (loop == get_irg_loop(current_ir_graph)) {
913 find_blocks(get_irg_end_block(current_ir_graph));
915 ir_free_resources(current_ir_graph, IR_RESOURCE_BLOCK_VISITED);
917 DB((dbg, LEVEL_3, "Block List for loop %p\n", loop));
918 len = ARR_LEN(loop_blocks);
919 for (i = 0; i < len; ++i) {
920 block_or_loop_t *block_or_loop = &loop_blocks[i];
921 if (block_or_loop->is_loop) {
922 DB((dbg, LEVEL_3, " L-%p", block_or_loop->v.loop));
924 DB((dbg, LEVEL_3, " B-%+F", block_or_loop->v.block));
927 DB((dbg, LEVEL_3, "\n"));
929 max_register_pressure = 0;
931 /* run1: tentative phase */
932 tentative_mode = true;
933 for (i = len-1; i >= 0; --i) {
934 process_block_or_loop(&loop_blocks[i]);
937 /* run2: tentative phase - should reach fixpoint */
938 tentative_mode = true;
939 for (i = len-1; i >= 0; --i) {
940 process_block_or_loop(&loop_blocks[i]);
944 /* run3: tentative phase - check fixpoint */
945 tentative_mode = true;
946 should_have_reached_fixpoint = true;
947 for (i = len-1; i >= 0; --i) {
948 process_block_or_loop(&loop_blocks[i]);
950 should_have_reached_fixpoint = false;
953 /* run4: add spills/reloads */
954 tentative_mode = false;
955 do_push_unused_livethroughs = true;
956 for (i = len-1; i >= 0; --i) {
957 process_block_or_loop(&loop_blocks[i]);
959 do_push_unused_livethroughs = false;
961 loop_info->max_register_pressure = max_register_pressure;
962 DB((dbg, LEVEL_2, "Regpressure in loop %p: %u\n", loop,
963 (unsigned) max_register_pressure));
965 DEL_ARR_F(loop_blocks);
968 static void fix_block_borders(ir_node *block, void *data)
970 block_info_t *block_info = get_block_info(block);
971 worklist_t *start_worklist = block_info->start_worklist;
972 int n_cfgpreds = get_Block_n_cfgpreds(block);
973 ir_loop *loop = get_irn_loop(block);
977 assert(start_worklist != NULL);
979 for (i = 0; i < n_cfgpreds; ++i) {
980 ir_node *pred_block = get_Block_cfgpred_block(block, i);
981 block_info_t *pred_block_info = get_block_info(pred_block);
982 worklist_t *end_worklist = pred_block_info->end_worklist;
983 ir_loop *pred_loop = get_irn_loop(pred_block);
984 bool is_loop_entry = false;
985 struct list_head *entry;
987 assert(end_worklist != NULL);
989 if (get_loop_depth(pred_loop) < get_loop_depth(loop)) {
990 is_loop_entry = true;
993 /* reload missing values */
994 activate_worklist(end_worklist);
996 list_for_each(entry, &start_worklist->live_values) {
997 worklist_entry_t *wl_entry
998 = list_entry(entry, worklist_entry_t, head);
999 ir_node *value = wl_entry->value;
1001 if (is_Phi(value) && get_nodes_block(value) == block) {
1002 value = get_irn_n(value, i);
1004 /* we might have unknowns as argument for the phi */
1005 if (!arch_irn_consider_in_reg_alloc(arch_env, cls, value))
1009 if (worklist_contains(value))
1011 if (wl_entry->unused_livethrough_loop != NULL && !is_loop_entry)
1014 be_add_reload_on_edge(senv, value, block, i, cls, 1);
1017 deactivate_worklist(end_worklist);
1021 static void be_spill_belady3(be_irg_t *birg, const arch_register_class_t *ncls)
1023 ir_graph *irg = be_get_birg_irg(birg);
1026 n_regs = cls->n_regs - be_put_ignore_regs(birg, cls, NULL);
1028 /* shortcut for register classes with ignore regs only */
1032 worklist_visited = 0;
1033 arch_env = be_get_birg_arch_env(birg);
1034 exec_freq = be_get_birg_exec_freq(birg);
1036 be_clear_links(irg);
1037 ir_reserve_resources(irg, IR_RESOURCE_IRN_VISITED | IR_RESOURCE_IRN_LINK
1038 | IR_RESOURCE_LOOP_LINK);
1039 inc_irg_visited(irg);
1041 obstack_init(&obst);
1042 senv = be_new_spill_env(birg);
1044 assure_cf_loop(irg);
1045 clear_loop_info(get_irg_loop(irg));
1046 irg_block_walk_graph(irg, construct_loop_edges, NULL, NULL);
1048 process_loop(get_irg_loop(current_ir_graph));
1050 irg_block_walk_graph(irg, fix_block_borders, NULL, NULL);
1052 ir_free_resources(irg, IR_RESOURCE_IRN_VISITED | IR_RESOURCE_IRN_LINK
1053 | IR_RESOURCE_LOOP_LINK);
1055 be_insert_spills_reloads(senv);
1057 obstack_free(&obst, NULL);
1060 be_delete_spill_env(senv);
1063 void be_init_spillbelady3(void)
1065 static be_spiller_t belady3_spiller = {
1069 be_register_spiller("belady3", &belady3_spiller);
1070 FIRM_DBG_REGISTER(dbg, "firm.be.spill.belady3");
1073 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_spillbelady3);