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 Copy minimization driver.
23 * @author Daniel Grund
27 * Main file for the optimization reducing the copies needed for:
29 * - Register-constrained nodes
30 * - Two-address code instructions
38 #include "raw_bitset.h"
44 #include "iredges_t.h"
46 #include "irphase_t.h"
47 #include "irprintf_t.h"
54 #include "beintlive_t.h"
55 #include "becopyopt_t.h"
56 #include "becopystat.h"
60 #include "bestatevent.h"
65 #include "lc_opts_enum.h"
70 #define DUMP_ALL 2 * DUMP_APPEL - 1
72 #define COST_FUNC_FREQ 1
73 #define COST_FUNC_LOOP 2
74 #define COST_FUNC_ALL_ONE 3
76 static unsigned dump_flags = 0;
77 static unsigned style_flags = 0;
78 static unsigned do_stats = 0;
79 static cost_fct_t cost_func = co_get_costs_exec_freq;
80 static unsigned algo = CO_ALGO_HEUR4;
81 static int improve = 1;
83 static const lc_opt_enum_mask_items_t dump_items[] = {
84 { "before", DUMP_BEFORE },
85 { "after", DUMP_AFTER },
86 { "appel", DUMP_APPEL },
91 static const lc_opt_enum_mask_items_t style_items[] = {
92 { "color", CO_IFG_DUMP_COLORS },
93 { "labels", CO_IFG_DUMP_LABELS },
94 { "constr", CO_IFG_DUMP_CONSTR },
95 { "shape", CO_IFG_DUMP_SHAPE },
96 { "full", 2 * CO_IFG_DUMP_SHAPE - 1 },
100 static const lc_opt_enum_mask_items_t algo_items[] = {
101 { "none", CO_ALGO_NONE },
102 { "heur", CO_ALGO_HEUR },
103 { "heur2", CO_ALGO_HEUR2 },
104 { "heur3", CO_ALGO_HEUR3 },
105 { "heur4", CO_ALGO_HEUR4 },
106 { "ilp", CO_ALGO_ILP },
110 typedef int (*opt_funcptr)(void);
112 static const lc_opt_enum_func_ptr_items_t cost_func_items[] = {
113 { "freq", (opt_funcptr) co_get_costs_exec_freq },
114 { "loop", (opt_funcptr) co_get_costs_loop_depth },
115 { "one", (opt_funcptr) co_get_costs_all_one },
119 static lc_opt_enum_mask_var_t dump_var = {
120 &dump_flags, dump_items
123 static lc_opt_enum_mask_var_t style_var = {
124 &style_flags, style_items
127 static lc_opt_enum_mask_var_t algo_var = {
131 static lc_opt_enum_func_ptr_var_t cost_func_var = {
132 (opt_funcptr*) &cost_func, cost_func_items
135 static const lc_opt_table_entry_t options[] = {
136 LC_OPT_ENT_ENUM_INT ("algo", "select copy optimization algo", &algo_var),
137 LC_OPT_ENT_ENUM_FUNC_PTR ("cost", "select a cost function", &cost_func_var),
138 LC_OPT_ENT_ENUM_MASK ("dump", "dump ifg before or after copy optimization", &dump_var),
139 LC_OPT_ENT_ENUM_MASK ("style", "dump style for ifg dumping", &style_var),
140 LC_OPT_ENT_BOOL ("stats", "dump statistics after each optimization", &do_stats),
141 LC_OPT_ENT_BOOL ("improve", "run heur3 before if algo can exploit start solutions", &improve),
145 /* Insert additional options registration functions here. */
146 extern void be_co_ilp_register_options(lc_opt_entry_t *grp);
147 extern void be_co2_register_options(lc_opt_entry_t *grp);
148 extern void be_co3_register_options(lc_opt_entry_t *grp);
150 void be_init_copycoal(void)
152 lc_opt_entry_t *be_grp = lc_opt_get_grp(firm_opt_get_root(), "be");
153 lc_opt_entry_t *ra_grp = lc_opt_get_grp(be_grp, "ra");
154 lc_opt_entry_t *chordal_grp = lc_opt_get_grp(ra_grp, "chordal");
155 lc_opt_entry_t *co_grp = lc_opt_get_grp(chordal_grp, "co");
157 lc_opt_add_table(co_grp, options);
160 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copycoal);
162 #undef QUICK_AND_DIRTY_HACK
164 static int nodes_interfere(const be_chordal_env_t *env, const ir_node *a, const ir_node *b)
167 return be_ifg_connected(env->ifg, a, b);
169 return be_values_interfere(env->birg->lv, a, b);
173 /******************************************************************************
176 | | __ ___ _ __ ___ _ __ __ _| |
177 | | |_ |/ _ \ '_ \ / _ \ '__/ _` | |
178 | |__| | __/ | | | __/ | | (_| | |
179 \_____|\___|_| |_|\___|_| \__,_|_|
181 ******************************************************************************/
183 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
186 copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs)
188 const char *s1, *s2, *s3;
192 FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
194 co = XMALLOCZ(copy_opt_t);
195 co->cenv = chordal_env;
196 co->irg = chordal_env->irg;
197 co->cls = chordal_env->cls;
198 co->get_costs = get_costs;
201 s2 = get_entity_name(get_irg_entity(co->irg));
202 s3 = chordal_env->cls->name;
203 len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
204 co->name = XMALLOCN(char, len);
205 snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
210 void free_copy_opt(copy_opt_t *co) {
216 * Checks if a node is optimizable, viz. has something to do with coalescing
217 * @param irn The irn to check
219 static int co_is_optimizable_root(ir_node *irn)
221 const arch_register_req_t *req;
222 const arch_register_t *reg;
224 if (arch_irn_is_ignore(irn))
227 reg = arch_get_irn_register(irn);
228 if (arch_register_type_is(reg, ignore))
231 if (is_Reg_Phi(irn) || is_Perm_Proj(irn))
234 req = arch_get_register_req_out(irn);
235 if (is_2addr_code(req))
241 int co_get_costs_loop_depth(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
244 ir_node *root_block = get_nodes_block(root);
249 /* for phis the copies are placed in the corresponding pred-block */
250 loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos));
252 /* a perm places the copy in the same block as it resides */
253 loop = get_irn_loop(root_block);
256 int d = get_loop_depth(loop);
262 int co_get_costs_exec_freq(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
264 ir_node *root_bl = get_nodes_block(root);
265 ir_node *copy_bl = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl;
267 res = get_block_execfreq_ulong(co->cenv->birg->exec_freq, copy_bl);
269 /* don't allow values smaller than one. */
270 return res < 1 ? 1 : res;
274 int co_get_costs_all_one(const copy_opt_t *co, ir_node *root, ir_node *arg, int pos) {
282 /******************************************************************************
283 ____ _ _ _ _ _ _____ _
284 / __ \ | | | | | | (_) | / ____| |
285 | | | |_ __ | |_| | | |_ __ _| |_ ___ | (___ | |_ ___ _ __ __ _ __ _ ___
286 | | | | '_ \| __| | | | '_ \| | __/ __| \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
287 | |__| | |_) | |_| |__| | | | | | |_\__ \ ____) | || (_) | | | (_| | (_| | __/
288 \____/| .__/ \__|\____/|_| |_|_|\__|___/ |_____/ \__\___/|_| \__,_|\__, |\___|
291 ******************************************************************************/
294 * Determines a maximum weighted independent set with respect to
295 * the interference and conflict edges of all nodes in a qnode.
297 static int ou_max_ind_set_costs(unit_t *ou) {
298 be_chordal_env_t *chordal_env = ou->co->cenv;
299 ir_node **safe, **unsafe;
300 int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
303 int max, curr_weight, best_weight = 0;
305 /* assign the nodes into two groups.
306 * safe: node has no interference, hence it is in every max stable set.
307 * unsafe: node has an interference
309 safe = ALLOCAN(ir_node*, ou->node_count - 1);
312 unsafe = ALLOCAN(ir_node*, ou->node_count - 1);
313 unsafe_costs = ALLOCAN(int, ou->node_count - 1);
315 for(i=1; i<ou->node_count; ++i) {
317 for(o=1; o<ou->node_count; ++o) {
320 if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) {
321 unsafe_costs[unsafe_count] = ou->costs[i];
322 unsafe[unsafe_count] = ou->nodes[i];
329 safe_costs += ou->costs[i];
330 safe[safe_count++] = ou->nodes[i];
335 /* now compute the best set out of the unsafe nodes*/
336 if (unsafe_count > MIS_HEUR_TRIGGER) {
337 bitset_t *best = bitset_alloca(unsafe_count);
338 /* Heuristik: Greedy trial and error form index 0 to unsafe_count-1 */
339 for (i=0; i<unsafe_count; ++i) {
341 /* check if it is a stable set */
342 for (o=bitset_next_set(best, 0); o!=-1 && o<i; o=bitset_next_set(best, o+1))
343 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o])) {
344 bitset_clear(best, i); /* clear the bit and try next one */
348 /* compute the weight */
349 bitset_foreach(best, pos)
350 best_weight += unsafe_costs[pos];
352 /* Exact Algorithm: Brute force */
353 curr = bitset_alloca(unsafe_count);
354 bitset_set_all(curr);
355 while ((max = bitset_popcnt(curr)) != 0) {
356 /* check if curr is a stable set */
357 for (i=bitset_next_set(curr, 0); i!=-1; i=bitset_next_set(curr, i+1))
358 for (o=bitset_next_set(curr, i+1); o!=-1; o=bitset_next_set(curr, o+1)) /* !!!!! difference to qnode_max_ind_set(): NOT (curr, i) */
359 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o]))
362 /* if we arrive here, we have a stable set */
363 /* compute the weigth of the stable set*/
365 bitset_foreach(curr, pos)
366 curr_weight += unsafe_costs[pos];
369 if (curr_weight > best_weight) {
370 best_weight = curr_weight;
378 return safe_costs+best_weight;
381 static void co_collect_units(ir_node *irn, void *env)
383 const arch_register_req_t *req = arch_get_register_req_out(irn);
384 copy_opt_t *co = env;
387 if (req->cls != co->cls)
389 if (!co_is_optimizable_root(irn))
392 /* Init a new unit */
393 unit = XMALLOCZ(unit_t);
395 unit->node_count = 1;
396 INIT_LIST_HEAD(&unit->queue);
398 /* Phi with some/all of its arguments */
399 if (is_Reg_Phi(irn)) {
403 arity = get_irn_arity(irn);
404 unit->nodes = XMALLOCN(ir_node*, arity + 1);
405 unit->costs = XMALLOCN(int, arity + 1);
406 unit->nodes[0] = irn;
409 for (i=0; i<arity; ++i) {
411 ir_node *arg = get_irn_n(irn, i);
413 assert(arch_get_irn_reg_class_out(arg) == co->cls && "Argument not in same register class.");
416 if (nodes_interfere(co->cenv, irn, arg)) {
417 unit->inevitable_costs += co->get_costs(co, irn, arg, i);
421 /* Else insert the argument of the phi to the members of this ou */
422 DBG((dbg, LEVEL_1, "\t Member: %+F\n", arg));
424 if (arch_irn_is_ignore(arg))
427 /* Check if arg has occurred at a prior position in the arg/list */
429 for (o=1; o<unit->node_count; ++o) {
430 if (unit->nodes[o] == arg) {
436 if (!arg_pos) { /* a new argument */
437 /* insert node, set costs */
438 unit->nodes[unit->node_count] = arg;
439 unit->costs[unit->node_count] = co->get_costs(co, irn, arg, i);
441 } else { /* arg has occurred before in same phi */
442 /* increase costs for existing arg */
443 unit->costs[arg_pos] += co->get_costs(co, irn, arg, i);
446 unit->nodes = XREALLOC(unit->nodes, ir_node*, unit->node_count);
447 unit->costs = XREALLOC(unit->costs, int, unit->node_count);
448 } else if (is_Perm_Proj(irn)) {
449 /* Proj of a perm with corresponding arg */
450 assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
451 unit->nodes = XMALLOCN(ir_node*, 2);
452 unit->costs = XMALLOCN(int, 2);
453 unit->node_count = 2;
454 unit->nodes[0] = irn;
455 unit->nodes[1] = get_Perm_src(irn);
456 unit->costs[1] = co->get_costs(co, irn, unit->nodes[1], -1);
458 /* Src == Tgt of a 2-addr-code instruction */
459 if (is_2addr_code(req)) {
460 const unsigned other = req->other_same;
464 for (i = 0; (1U << i) <= other; ++i) {
465 if (other & (1U << i)) {
466 ir_node *o = get_irn_n(skip_Proj(irn), i);
467 if (arch_irn_is_ignore(o))
469 if (nodes_interfere(co->cenv, irn, o))
478 unit->nodes = XMALLOCN(ir_node*, count);
479 unit->costs = XMALLOCN(int, count);
480 unit->node_count = count;
481 unit->nodes[k++] = irn;
483 for (i = 0; 1U << i <= other; ++i) {
484 if (other & (1U << i)) {
485 ir_node *o = get_irn_n(skip_Proj(irn), i);
486 if (!arch_irn_is_ignore(o) &&
487 !nodes_interfere(co->cenv, irn, o)) {
489 unit->costs[k] = co->get_costs(co, irn, o, -1);
496 assert(0 && "This is not an optimizable node!");
500 /* Insert the new unit at a position according to its costs */
501 if (unit->node_count > 1) {
503 struct list_head *tmp;
505 /* Determine the maximum costs this unit can cause: all_nodes_cost */
506 for(i=1; i<unit->node_count; ++i) {
507 unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
508 unit->all_nodes_costs += unit->costs[i];
511 /* Determine the minimal costs this unit will cause: min_nodes_costs */
512 unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
513 /* Insert the new ou according to its sort_key */
515 while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
517 list_add(&unit->units, tmp);
523 #ifdef QUICK_AND_DIRTY_HACK
525 static int compare_ous(const void *k1, const void *k2) {
526 const unit_t *u1 = *((const unit_t **) k1);
527 const unit_t *u2 = *((const unit_t **) k2);
528 int i, o, u1_has_constr, u2_has_constr;
529 arch_register_req_t req;
531 /* Units with constraints come first */
533 for (i=0; i<u1->node_count; ++i) {
534 arch_get_register_req_out(&req, u1->nodes[i]);
535 if (arch_register_req_is(&req, limited)) {
542 for (i=0; i<u2->node_count; ++i) {
543 arch_get_register_req_out(&req, u2->nodes[i]);
544 if (arch_register_req_is(&req, limited)) {
550 if (u1_has_constr != u2_has_constr)
551 return u2_has_constr - u1_has_constr;
553 /* Now check, whether the two units are connected */
555 for (i=0; i<u1->node_count; ++i)
556 for (o=0; o<u2->node_count; ++o)
557 if (u1->nodes[i] == u2->nodes[o])
561 /* After all, the sort key decides. Greater keys come first. */
562 return u2->sort_key - u1->sort_key;
567 * Sort the ou's according to constraints and their sort_key
569 static void co_sort_units(copy_opt_t *co) {
570 int i, count = 0, costs;
573 /* get the number of ous, remove them form the list and fill the array */
574 list_for_each_entry(unit_t, ou, &co->units, units)
576 ous = ALLOCAN(unit_t, count);
578 costs = co_get_max_copy_costs(co);
581 list_for_each_entry(unit_t, ou, &co->units, units)
584 INIT_LIST_HEAD(&co->units);
586 assert(count == i && list_empty(&co->units));
588 for (i=0; i<count; ++i)
589 ir_printf("%+F\n", ous[i]->nodes[0]);
591 qsort(ous, count, sizeof(*ous), compare_ous);
594 for (i=0; i<count; ++i)
595 ir_printf("%+F\n", ous[i]->nodes[0]);
597 /* reinsert into list in correct order */
598 for (i=0; i<count; ++i)
599 list_add_tail(&ous[i]->units, &co->units);
601 assert(costs == co_get_max_copy_costs(co));
605 void co_build_ou_structure(copy_opt_t *co) {
606 DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
607 INIT_LIST_HEAD(&co->units);
608 irg_walk_graph(co->irg, co_collect_units, NULL, co);
609 #ifdef QUICK_AND_DIRTY_HACK
614 void co_free_ou_structure(copy_opt_t *co) {
617 list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
622 co->units.next = NULL;
625 /* co_solve_heuristic() is implemented in becopyheur.c */
627 int co_get_max_copy_costs(const copy_opt_t *co) {
633 list_for_each_entry(unit_t, curr, &co->units, units) {
634 res += curr->inevitable_costs;
635 for (i=1; i<curr->node_count; ++i)
636 res += curr->costs[i];
641 int co_get_inevit_copy_costs(const copy_opt_t *co) {
647 list_for_each_entry(unit_t, curr, &co->units, units)
648 res += curr->inevitable_costs;
652 int co_get_copy_costs(const copy_opt_t *co) {
658 list_for_each_entry(unit_t, curr, &co->units, units) {
659 int root_col = get_irn_col(curr->nodes[0]);
660 DBG((dbg, LEVEL_1, " %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
661 res += curr->inevitable_costs;
662 for (i=1; i<curr->node_count; ++i) {
663 int arg_col = get_irn_col(curr->nodes[i]);
664 if (root_col != arg_col) {
665 DBG((dbg, LEVEL_1, " %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
666 res += curr->costs[i];
673 int co_get_lower_bound(const copy_opt_t *co) {
679 list_for_each_entry(unit_t, curr, &co->units, units)
680 res += curr->inevitable_costs + curr->min_nodes_costs;
684 void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat)
686 bitset_t *seen = bitset_irg_malloc(co->irg);
689 memset(stat, 0, sizeof(stat[0]));
691 /* count affinity edges. */
692 co_gs_foreach_aff_node(co, an) {
694 stat->aff_nodes += 1;
695 bitset_add_irn(seen, an->irn);
696 co_gs_foreach_neighb(an, neigh) {
697 if(!bitset_contains_irn(seen, neigh->irn)) {
698 stat->aff_edges += 1;
699 stat->max_costs += neigh->costs;
701 if (get_irn_col(an->irn) != get_irn_col(neigh->irn)) {
702 stat->costs += neigh->costs;
703 stat->unsatisfied_edges += 1;
706 if(nodes_interfere(co->cenv, an->irn, neigh->irn)) {
708 stat->inevit_costs += neigh->costs;
718 /******************************************************************************
720 / ____| | | / ____| |
721 | | __ _ __ __ _ _ __ | |__ | (___ | |_ ___ _ __ __ _ __ _ ___
722 | | |_ | '__/ _` | '_ \| '_ \ \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
723 | |__| | | | (_| | |_) | | | | ____) | || (_) | | | (_| | (_| | __/
724 \_____|_| \__,_| .__/|_| |_| |_____/ \__\___/|_| \__,_|\__, |\___|
727 ******************************************************************************/
729 static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) {
730 const affinity_node_t *n1 = k1;
731 const affinity_node_t *n2 = k2;
734 return (n1->irn != n2->irn);
737 static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
738 affinity_node_t new_node, *node;
744 new_node.neighbours = NULL;
745 node = set_insert(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
747 for (nbr = node->neighbours; nbr; nbr = nbr->next)
748 if (nbr->irn == n2) {
753 /* if we did not find n2 in n1's neighbourhood insert it */
755 nbr = OALLOC(&co->obst, neighb_t);
758 nbr->next = node->neighbours;
760 node->neighbours = nbr;
764 /* now nbr points to n1's neighbour-entry of n2 */
768 static inline void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
769 if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
770 add_edge(co, n1, n2, costs);
771 add_edge(co, n2, n1, costs);
775 static void build_graph_walker(ir_node *irn, void *env) {
776 const arch_register_req_t *req = arch_get_register_req_out(irn);
777 copy_opt_t *co = env;
779 const arch_register_t *reg;
781 if (req->cls != co->cls || arch_irn_is_ignore(irn))
784 reg = arch_get_irn_register(irn);
785 if (arch_register_type_is(reg, ignore))
788 if (is_Reg_Phi(irn)) { /* Phis */
789 for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
790 ir_node *arg = get_irn_n(irn, pos);
791 add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
793 } else if (is_Perm_Proj(irn)) { /* Perms */
794 ir_node *arg = get_Perm_src(irn);
795 add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
796 } else { /* 2-address code */
797 if (is_2addr_code(req)) {
798 const unsigned other = req->other_same;
801 for (i = 0; 1U << i <= other; ++i) {
802 if (other & (1U << i)) {
803 ir_node *other = get_irn_n(skip_Proj(irn), i);
804 if (!arch_irn_is_ignore(other))
805 add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
812 void co_build_graph_structure(copy_opt_t *co) {
813 obstack_init(&co->obst);
814 co->nodes = new_set(compare_affinity_node_t, 32);
816 irg_walk_graph(co->irg, build_graph_walker, NULL, co);
819 void co_free_graph_structure(copy_opt_t *co) {
823 obstack_free(&co->obst, NULL);
827 /* co_solve_ilp1() co_solve_ilp2() are implemented in becopyilpX.c */
829 int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) {
830 affinity_node_t new_node, *n;
835 n = set_find(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
837 return (n->degree > 0);
842 static int co_dump_appel_disjoint_constraints(const copy_opt_t *co, ir_node *a, ir_node *b)
844 ir_node *nodes[] = { a, b };
845 bitset_t *constr[] = { NULL, NULL };
848 constr[0] = bitset_alloca(co->cls->n_regs);
849 constr[1] = bitset_alloca(co->cls->n_regs);
851 for (j = 0; j < 2; ++j) {
852 const arch_register_req_t *req = arch_get_register_req_out(nodes[j]);
853 if(arch_register_req_is(req, limited))
854 rbitset_copy_to_bitset(req->limited, constr[j]);
856 bitset_set_all(constr[j]);
860 return !bitset_intersect(constr[0], constr[1]);
863 void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
865 be_ifg_t *ifg = co->cenv->ifg;
866 int *color_map = ALLOCAN(int, co->cls->n_regs);
867 int *node_map = XMALLOCN(int, get_irg_last_idx(co->irg) + 1);
875 for(i = 0; i < co->cls->n_regs; ++i) {
876 const arch_register_t *reg = &co->cls->regs[i];
877 color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_regs++;
881 * n contains the first node number.
882 * the values below n are the pre-colored register nodes
885 it = be_ifg_nodes_iter_alloca(ifg);
886 nit = be_ifg_neighbours_iter_alloca(ifg);
889 be_ifg_foreach_node(ifg, it, irn) {
890 if (arch_irn_is_ignore(irn))
892 node_map[get_irn_idx(irn)] = n++;
895 fprintf(f, "%d %d\n", n, n_regs);
897 be_ifg_foreach_node(ifg, it, irn) {
898 if (!arch_irn_is_ignore(irn)) {
899 int idx = node_map[get_irn_idx(irn)];
900 affinity_node_t *a = get_affinity_info(co, irn);
901 const arch_register_req_t *req = arch_get_register_req_out(irn);
904 if(arch_register_req_is(req, limited)) {
905 for(i = 0; i < co->cls->n_regs; ++i) {
906 if(!rbitset_is_set(req->limited, i) && color_map[i] >= 0)
907 fprintf(f, "%d %d -1\n", color_map[i], idx);
911 be_ifg_foreach_neighbour(ifg, nit, irn, adj) {
912 if (!arch_irn_is_ignore(adj) &&
913 !co_dump_appel_disjoint_constraints(co, irn, adj)) {
914 int adj_idx = node_map[get_irn_idx(adj)];
916 fprintf(f, "%d %d -1\n", idx, adj_idx);
923 co_gs_foreach_neighb(a, n) {
924 if (!arch_irn_is_ignore(n->irn)) {
925 int n_idx = node_map[get_irn_idx(n->irn)];
927 fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs);
938 ___ _____ ____ ____ ___ _____ ____ _
939 |_ _| ___/ ___| | _ \ / _ \_ _| | _ \ _ _ _ __ ___ _ __ (_)_ __ __ _
940 | || |_ | | _ | | | | | | || | | | | | | | | '_ ` _ \| '_ \| | '_ \ / _` |
941 | || _|| |_| | | |_| | |_| || | | |_| | |_| | | | | | | |_) | | | | | (_| |
942 |___|_| \____| |____/ \___/ |_| |____/ \__,_|_| |_| |_| .__/|_|_| |_|\__, |
946 static const char *get_dot_color_name(size_t col)
948 static const char *names[] = {
982 return col < sizeof(names)/sizeof(names[0]) ? names[col] : "white";
985 typedef struct _co_ifg_dump_t {
986 const copy_opt_t *co;
990 static void ifg_dump_graph_attr(FILE *f, void *self)
993 fprintf(f, "overlap=scale");
996 static int ifg_is_dump_node(void *self, ir_node *irn)
999 return !arch_irn_is_ignore(irn);
1002 static void ifg_dump_node_attr(FILE *f, void *self, ir_node *irn)
1004 co_ifg_dump_t *env = self;
1005 const arch_register_t *reg = arch_get_irn_register(irn);
1006 const arch_register_req_t *req = arch_get_register_req_out(irn);
1007 int limited = arch_register_req_is(req, limited);
1009 if(env->flags & CO_IFG_DUMP_LABELS) {
1010 ir_fprintf(f, "label=\"%+F", irn);
1012 if((env->flags & CO_IFG_DUMP_CONSTR) && limited) {
1013 bitset_t *bs = bitset_alloca(env->co->cls->n_regs);
1014 rbitset_copy_to_bitset(req->limited, bs);
1015 ir_fprintf(f, "\\n%B", bs);
1017 ir_fprintf(f, "\" ");
1019 fprintf(f, "label=\"\" shape=point " );
1022 if(env->flags & CO_IFG_DUMP_SHAPE)
1023 fprintf(f, "shape=%s ", limited ? "diamond" : "ellipse");
1025 if(env->flags & CO_IFG_DUMP_COLORS)
1026 fprintf(f, "style=filled color=%s ", get_dot_color_name(reg->index));
1029 static void ifg_dump_at_end(FILE *file, void *self)
1031 co_ifg_dump_t *env = self;
1034 co_gs_foreach_aff_node(env->co, a) {
1035 const arch_register_t *ar = arch_get_irn_register(a->irn);
1036 unsigned aidx = get_irn_idx(a->irn);
1039 co_gs_foreach_neighb(a, n) {
1040 const arch_register_t *nr = arch_get_irn_register(n->irn);
1041 unsigned nidx = get_irn_idx(n->irn);
1044 const char *color = nr == ar ? "blue" : "red";
1045 fprintf(file, "\tn%d -- n%d [weight=0.01 ", aidx, nidx);
1046 if(env->flags & CO_IFG_DUMP_LABELS)
1047 fprintf(file, "label=\"%d\" ", n->costs);
1048 if(env->flags & CO_IFG_DUMP_COLORS)
1049 fprintf(file, "color=%s ", color);
1051 fprintf(file, "style=dotted");
1052 fprintf(file, "];\n");
1059 static be_ifg_dump_dot_cb_t ifg_dot_cb = {
1061 ifg_dump_graph_attr,
1070 void co_dump_ifg_dot(const copy_opt_t *co, FILE *f, unsigned flags)
1076 be_ifg_dump_dot(co->cenv->ifg, co->irg, f, &ifg_dot_cb, &cod);
1080 void co_solve_park_moon(copy_opt_t *opt)
1085 static int void_algo(copy_opt_t *co)
1093 / \ | | __ _ ___ _ __(_) |_| |__ _ __ ___ ___
1094 / _ \ | |/ _` |/ _ \| '__| | __| '_ \| '_ ` _ \/ __|
1095 / ___ \| | (_| | (_) | | | | |_| | | | | | | | \__ \
1096 /_/ \_\_|\__, |\___/|_| |_|\__|_| |_|_| |_| |_|___/
1103 int can_improve_existing;
1106 static co_algo_info_t algos[] = {
1107 { void_algo, "none", 0 },
1108 { co_solve_heuristic, "heur1", 0 },
1109 { co_solve_heuristic_new, "heur2", 0 },
1111 { co_solve_heuristic_java, "heur3", 0 },
1113 { NULL, "heur3", 0 },
1115 { co_solve_heuristic_mst, "heur4", 0 },
1117 { co_solve_ilp2, "ilp", 1 },
1126 | \/ | __ _(_)_ __ | _ \ _ __(_)_ _____ _ __
1127 | |\/| |/ _` | | '_ \ | | | | '__| \ \ / / _ \ '__|
1128 | | | | (_| | | | | | | |_| | | | |\ V / __/ |
1129 |_| |_|\__,_|_|_| |_| |____/|_| |_| \_/ \___|_|
1133 static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char *suffix)
1140 n = strlen(env->birg->main_env->cup_name);
1141 tu_name = XMALLOCN(char, n + 1);
1142 strcpy(tu_name, env->birg->main_env->cup_name);
1143 for (i = 0; i < n; ++i)
1144 if (tu_name[i] == '.')
1148 ir_snprintf(buf, sizeof(buf), "%s%s_%F_%s%s", prefix, tu_name, env->irg, env->cls->name, suffix);
1150 result = fopen(buf, "wt");
1151 if(result == NULL) {
1152 panic("Couldn't open '%s' for writing.", buf);
1158 void co_driver(be_chordal_env_t *cenv)
1160 ir_timer_t *timer = ir_timer_register("firm.be.copyopt", "runtime");
1161 co_complete_stats_t before, after;
1163 co_algo_t *algo_func;
1164 int was_optimal = 0;
1166 if (algo >= CO_ALGO_LAST)
1169 be_liveness_assure_chk(be_get_birg_liveness(cenv->birg));
1171 co = new_copy_opt(cenv, cost_func);
1172 co_build_ou_structure(co);
1173 co_build_graph_structure(co);
1175 co_complete_stats(co, &before);
1177 be_stat_ev_ull("co_aff_nodes", before.aff_nodes);
1178 be_stat_ev_ull("co_aff_edges", before.aff_edges);
1179 be_stat_ev_ull("co_max_costs", before.max_costs);
1180 be_stat_ev_ull("co_inevit_costs", before.inevit_costs);
1181 be_stat_ev_ull("co_aff_int", before.aff_int);
1183 be_stat_ev_ull("co_init_costs", before.costs);
1184 be_stat_ev_ull("co_init_unsat", before.unsatisfied_edges);
1186 if (dump_flags & DUMP_BEFORE) {
1187 FILE *f = my_open(cenv, "", "-before.dot");
1188 co_dump_ifg_dot(co, f, style_flags);
1192 /* if the algo can improve results, provide an initial solution with heur3 */
1193 if (improve && algos[algo].can_improve_existing) {
1194 co_complete_stats_t stats;
1196 /* produce a heuristic solution */
1198 co_solve_heuristic_java(co);
1200 co_solve_heuristic(co);
1201 #endif /* WITH_JVM */
1203 /* do the stats and provide the current costs */
1204 co_complete_stats(co, &stats);
1205 be_stat_ev_ull("co_prepare_costs", stats.costs);
1209 /* start the JVM here so that it does not tamper the timing. */
1210 if (algo == CO_ALGO_HEUR3)
1211 be_java_coal_start_jvm();
1212 #endif /* WITH_JVM */
1214 algo_func = algos[algo].algo;
1216 /* perform actual copy minimization */
1217 ir_timer_reset_and_start(timer);
1218 was_optimal = algo_func(co);
1219 ir_timer_stop(timer);
1221 be_stat_ev("co_time", ir_timer_elapsed_msec(timer));
1222 be_stat_ev_ull("co_optimal", was_optimal);
1224 if (dump_flags & DUMP_AFTER) {
1225 FILE *f = my_open(cenv, "", "-after.dot");
1226 co_dump_ifg_dot(co, f, style_flags);
1230 co_complete_stats(co, &after);
1233 ulong64 optimizable_costs = after.max_costs - after.inevit_costs;
1234 ulong64 evitable = after.costs - after.inevit_costs;
1236 ir_printf("%30F ", cenv->irg);
1237 printf("%10s %10" ULL_FMT "%10" ULL_FMT "%10" ULL_FMT, cenv->cls->name, after.max_costs, before.costs, after.inevit_costs);
1239 if(optimizable_costs > 0)
1240 printf("%10" ULL_FMT " %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs);
1242 printf("%10" ULL_FMT " %5s\n", after.costs, "-");
1245 /* Dump the interference graph in Appel's format. */
1246 if (dump_flags & DUMP_APPEL) {
1247 FILE *f = my_open(cenv, "", ".apl");
1248 fprintf(f, "# %lld %lld\n", after.costs, after.unsatisfied_edges);
1249 co_dump_appel_graph(co, f);
1253 be_stat_ev_ull("co_after_costs", after.costs);
1254 be_stat_ev_ull("co_after_unsat", after.unsatisfied_edges);
1256 co_free_graph_structure(co);
1257 co_free_ou_structure(co);