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 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
40 #include "raw_bitset.h"
46 #include "iredges_t.h"
49 #include "irphase_t.h"
50 #include "irprintf_t.h"
57 #include "beintlive_t.h"
58 #include "becopyopt_t.h"
59 #include "becopystat.h"
62 #include "besched_t.h"
63 #include "benodesets.h"
64 #include "bejavacoal.h"
65 #include "bestatevent.h"
69 #include <libcore/lc_timing.h>
70 #include <libcore/lc_opts.h>
71 #include <libcore/lc_opts_enum.h>
76 #define DUMP_ALL 2 * DUMP_APPEL - 1
78 #define COST_FUNC_FREQ 1
79 #define COST_FUNC_LOOP 2
80 #define COST_FUNC_ALL_ONE 3
82 static unsigned dump_flags = 0;
83 static unsigned style_flags = 0;
84 static unsigned do_stats = 0;
85 static cost_fct_t cost_func = co_get_costs_exec_freq;
86 static unsigned algo = CO_ALGO_HEUR4;
87 static int improve = 1;
89 static const lc_opt_enum_mask_items_t dump_items[] = {
90 { "before", DUMP_BEFORE },
91 { "after", DUMP_AFTER },
92 { "appel", DUMP_APPEL },
97 static const lc_opt_enum_mask_items_t style_items[] = {
98 { "color", CO_IFG_DUMP_COLORS },
99 { "labels", CO_IFG_DUMP_LABELS },
100 { "constr", CO_IFG_DUMP_CONSTR },
101 { "shape", CO_IFG_DUMP_SHAPE },
102 { "full", 2 * CO_IFG_DUMP_SHAPE - 1 },
106 static const lc_opt_enum_mask_items_t algo_items[] = {
107 { "none", CO_ALGO_NONE },
108 { "heur", CO_ALGO_HEUR },
109 { "heur2", CO_ALGO_HEUR2 },
110 { "heur3", CO_ALGO_HEUR3 },
111 { "heur4", CO_ALGO_HEUR4 },
112 { "ilp", CO_ALGO_ILP },
116 typedef int (*opt_funcptr)(void);
118 static const lc_opt_enum_func_ptr_items_t cost_func_items[] = {
119 { "freq", (opt_funcptr) co_get_costs_exec_freq },
120 { "loop", (opt_funcptr) co_get_costs_loop_depth },
121 { "one", (opt_funcptr) co_get_costs_all_one },
125 static lc_opt_enum_mask_var_t dump_var = {
126 &dump_flags, dump_items
129 static lc_opt_enum_mask_var_t style_var = {
130 &style_flags, style_items
133 static lc_opt_enum_mask_var_t algo_var = {
137 static lc_opt_enum_func_ptr_var_t cost_func_var = {
138 (opt_funcptr*) &cost_func, cost_func_items
141 static const lc_opt_table_entry_t options[] = {
142 LC_OPT_ENT_ENUM_INT ("algo", "select copy optimization algo", &algo_var),
143 LC_OPT_ENT_ENUM_FUNC_PTR ("cost", "select a cost function", &cost_func_var),
144 LC_OPT_ENT_ENUM_MASK ("dump", "dump ifg before or after copy optimization", &dump_var),
145 LC_OPT_ENT_ENUM_MASK ("style", "dump style for ifg dumping", &style_var),
146 LC_OPT_ENT_BOOL ("stats", "dump statistics after each optimization", &do_stats),
147 LC_OPT_ENT_BOOL ("improve", "run heur3 before if algo can exploit start solutions", &improve),
151 /* Insert additional options registration functions here. */
152 extern void be_co_ilp_register_options(lc_opt_entry_t *grp);
153 extern void be_co2_register_options(lc_opt_entry_t *grp);
154 extern void be_co3_register_options(lc_opt_entry_t *grp);
156 void be_init_copycoal(void)
158 lc_opt_entry_t *be_grp = lc_opt_get_grp(firm_opt_get_root(), "be");
159 lc_opt_entry_t *ra_grp = lc_opt_get_grp(be_grp, "ra");
160 lc_opt_entry_t *chordal_grp = lc_opt_get_grp(ra_grp, "chordal");
161 lc_opt_entry_t *co_grp = lc_opt_get_grp(chordal_grp, "co");
163 lc_opt_add_table(co_grp, options);
166 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copycoal);
168 #undef QUICK_AND_DIRTY_HACK
170 static int nodes_interfere(const be_chordal_env_t *env, const ir_node *a, const ir_node *b)
173 return be_ifg_connected(env->ifg, a, b);
175 return values_interfere(env->birg, a, b);
179 /******************************************************************************
182 | | __ ___ _ __ ___ _ __ __ _| |
183 | | |_ |/ _ \ '_ \ / _ \ '__/ _` | |
184 | |__| | __/ | | | __/ | | (_| | |
185 \_____|\___|_| |_|\___|_| \__,_|_|
187 ******************************************************************************/
189 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
192 copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs)
194 const char *s1, *s2, *s3;
198 FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
200 co = xcalloc(1, sizeof(*co));
201 co->cenv = chordal_env;
202 co->aenv = chordal_env->birg->main_env->arch_env;
203 co->irg = chordal_env->irg;
204 co->cls = chordal_env->cls;
205 co->get_costs = get_costs;
207 s1 = get_irp_prog_name();
208 s2 = get_entity_name(get_irg_entity(co->irg));
209 s3 = chordal_env->cls->name;
210 len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
211 co->name = xmalloc(len);
212 snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
217 void free_copy_opt(copy_opt_t *co) {
222 int co_is_optimizable_root(const copy_opt_t *co, ir_node *irn) {
223 const arch_register_req_t *req;
224 const arch_register_t *reg;
226 if (arch_irn_is(co->aenv, irn, ignore))
229 reg = arch_get_irn_register(co->aenv, irn);
230 if (arch_register_type_is(reg, ignore))
233 req = arch_get_register_req(co->aenv, irn, -1);
234 if (is_Reg_Phi(irn) || is_Perm_Proj(co->aenv, irn) || is_2addr_code(req))
240 int co_get_costs_loop_depth(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
243 ir_node *root_block = get_nodes_block(root);
248 /* for phis the copies are placed in the corresponding pred-block */
249 loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos));
251 /* a perm places the copy in the same block as it resides */
252 loop = get_irn_loop(root_block);
255 int d = get_loop_depth(loop);
261 int co_get_costs_exec_freq(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
263 ir_node *root_bl = get_nodes_block(root);
264 ir_node *copy_bl = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl;
266 res = get_block_execfreq_ulong(co->cenv->birg->exec_freq, copy_bl);
268 /* don't allow values smaller than one. */
269 return res < 1 ? 1 : res;
273 int co_get_costs_all_one(const copy_opt_t *co, ir_node *root, ir_node *arg, int pos) {
281 /******************************************************************************
282 ____ _ _ _ _ _ _____ _
283 / __ \ | | | | | | (_) | / ____| |
284 | | | |_ __ | |_| | | |_ __ _| |_ ___ | (___ | |_ ___ _ __ __ _ __ _ ___
285 | | | | '_ \| __| | | | '_ \| | __/ __| \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
286 | |__| | |_) | |_| |__| | | | | | |_\__ \ ____) | || (_) | | | (_| | (_| | __/
287 \____/| .__/ \__|\____/|_| |_|_|\__|___/ |_____/ \__\___/|_| \__,_|\__, |\___|
290 ******************************************************************************/
293 * Determines a maximum weighted independent set with respect to
294 * the interference and conflict edges of all nodes in a qnode.
296 static int ou_max_ind_set_costs(unit_t *ou) {
297 be_chordal_env_t *chordal_env = ou->co->cenv;
298 ir_node **safe, **unsafe;
299 int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
302 int max, curr_weight, best_weight = 0;
304 /* assign the nodes into two groups.
305 * safe: node has no interference, hence it is in every max stable set.
306 * unsafe: node has an interference
308 safe = alloca((ou->node_count-1) * sizeof(*safe));
311 unsafe = alloca((ou->node_count-1) * sizeof(*unsafe));
312 unsafe_costs = alloca((ou->node_count-1) * sizeof(*unsafe_costs));
314 for(i=1; i<ou->node_count; ++i) {
316 for(o=1; o<ou->node_count; ++o) {
319 if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) {
320 unsafe_costs[unsafe_count] = ou->costs[i];
321 unsafe[unsafe_count] = ou->nodes[i];
328 safe_costs += ou->costs[i];
329 safe[safe_count++] = ou->nodes[i];
334 /* now compute the best set out of the unsafe nodes*/
335 if (unsafe_count > MIS_HEUR_TRIGGER) {
336 bitset_t *best = bitset_alloca(unsafe_count);
337 /* Heuristik: Greedy trial and error form index 0 to unsafe_count-1 */
338 for (i=0; i<unsafe_count; ++i) {
340 /* check if it is a stable set */
341 for (o=bitset_next_set(best, 0); o!=-1 && o<i; o=bitset_next_set(best, o+1))
342 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o])) {
343 bitset_clear(best, i); /* clear the bit and try next one */
347 /* compute the weight */
348 bitset_foreach(best, pos)
349 best_weight += unsafe_costs[pos];
351 /* Exact Algorithm: Brute force */
352 curr = bitset_alloca(unsafe_count);
353 bitset_set_all(curr);
354 while ((max = bitset_popcnt(curr)) != 0) {
355 /* check if curr is a stable set */
356 for (i=bitset_next_set(curr, 0); i!=-1; i=bitset_next_set(curr, i+1))
357 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) */
358 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o]))
361 /* if we arrive here, we have a stable set */
362 /* compute the weigth of the stable set*/
364 bitset_foreach(curr, pos)
365 curr_weight += unsafe_costs[pos];
368 if (curr_weight > best_weight) {
369 best_weight = curr_weight;
377 return safe_costs+best_weight;
380 static void co_collect_units(ir_node *irn, void *env) {
381 copy_opt_t *co = env;
384 if (!is_curr_reg_class(co, irn))
386 if (!co_is_optimizable_root(co, irn))
389 /* Init a new unit */
390 unit = xcalloc(1, sizeof(*unit));
392 unit->node_count = 1;
393 INIT_LIST_HEAD(&unit->queue);
395 /* Phi with some/all of its arguments */
396 if (is_Reg_Phi(irn)) {
400 arity = get_irn_arity(irn);
401 unit->nodes = xmalloc((arity+1) * sizeof(*unit->nodes));
402 unit->costs = xmalloc((arity+1) * sizeof(*unit->costs));
403 unit->nodes[0] = irn;
406 for (i=0; i<arity; ++i) {
408 ir_node *arg = get_irn_n(irn, i);
410 assert(is_curr_reg_class(co, arg) && "Argument not in same register class.");
413 if (nodes_interfere(co->cenv, irn, arg)) {
414 unit->inevitable_costs += co->get_costs(co, irn, arg, i);
418 /* Else insert the argument of the phi to the members of this ou */
419 DBG((dbg, LEVEL_1, "\t Member: %+F\n", arg));
421 if (! arch_irn_is(co->aenv, arg, ignore)) {
422 /* Check if arg has occurred at a prior position in the arg/list */
424 for (o=1; o<unit->node_count; ++o) {
425 if (unit->nodes[o] == arg) {
431 if (!arg_pos) { /* a new argument */
432 /* insert node, set costs */
433 unit->nodes[unit->node_count] = arg;
434 unit->costs[unit->node_count] = co->get_costs(co, irn, arg, i);
436 } else { /* arg has occurred before in same phi */
437 /* increase costs for existing arg */
438 unit->costs[arg_pos] += co->get_costs(co, irn, arg, i);
442 unit->nodes = xrealloc(unit->nodes, unit->node_count * sizeof(*unit->nodes));
443 unit->costs = xrealloc(unit->costs, unit->node_count * sizeof(*unit->costs));
444 } else if (is_Perm_Proj(co->aenv, irn)) {
445 /* Proj of a perm with corresponding arg */
446 assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
447 unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
448 unit->costs = xmalloc(2 * sizeof(*unit->costs));
449 unit->node_count = 2;
450 unit->nodes[0] = irn;
451 unit->nodes[1] = get_Perm_src(irn);
452 unit->costs[1] = co->get_costs(co, irn, unit->nodes[1], -1);
454 const arch_register_req_t *req =
455 arch_get_register_req(co->aenv, irn, -1);
457 /* Src == Tgt of a 2-addr-code instruction */
458 if (is_2addr_code(req)) {
459 const unsigned other = req->other_same;
463 for (i = 0; (1U << i) <= other; ++i) {
464 if (other & (1U << i)) {
465 ir_node *o = get_irn_n(skip_Proj(irn), i);
466 if (!arch_irn_is(co->aenv, o, ignore) &&
467 !nodes_interfere(co->cenv, irn, o)) {
476 unit->nodes = xmalloc(count * sizeof(*unit->nodes));
477 unit->costs = xmalloc(count * sizeof(*unit->costs));
478 unit->node_count = count;
479 unit->nodes[k++] = irn;
481 for (i = 0; 1U << i <= other; ++i) {
482 if (other & (1U << i)) {
483 ir_node *o = get_irn_n(skip_Proj(irn), i);
484 if (!arch_irn_is(co->aenv, o, ignore) &&
485 !nodes_interfere(co->cenv, irn, o)) {
487 unit->costs[k] = co->get_costs(co, irn, o, -1);
494 assert(0 && "This is not an optimizable node!");
498 /* Insert the new unit at a position according to its costs */
499 if (unit->node_count > 1) {
501 struct list_head *tmp;
503 /* Determine the maximum costs this unit can cause: all_nodes_cost */
504 for(i=1; i<unit->node_count; ++i) {
505 unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
506 unit->all_nodes_costs += unit->costs[i];
509 /* Determine the minimal costs this unit will cause: min_nodes_costs */
510 unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
511 /* Insert the new ou according to its sort_key */
513 while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
515 list_add(&unit->units, tmp);
521 #ifdef QUICK_AND_DIRTY_HACK
523 static int compare_ous(const void *k1, const void *k2) {
524 const unit_t *u1 = *((const unit_t **) k1);
525 const unit_t *u2 = *((const unit_t **) k2);
526 int i, o, u1_has_constr, u2_has_constr;
527 arch_register_req_t req;
528 const arch_env_t *aenv = u1->co->aenv;
530 /* Units with constraints come first */
532 for (i=0; i<u1->node_count; ++i) {
533 arch_get_register_req(aenv, &req, u1->nodes[i], -1);
534 if (arch_register_req_is(&req, limited)) {
541 for (i=0; i<u2->node_count; ++i) {
542 arch_get_register_req(aenv, &req, u2->nodes[i], -1);
543 if (arch_register_req_is(&req, limited)) {
549 if (u1_has_constr != u2_has_constr)
550 return u2_has_constr - u1_has_constr;
552 /* Now check, whether the two units are connected */
554 for (i=0; i<u1->node_count; ++i)
555 for (o=0; o<u2->node_count; ++o)
556 if (u1->nodes[i] == u2->nodes[o])
560 /* After all, the sort key decides. Greater keys come first. */
561 return u2->sort_key - u1->sort_key;
566 * Sort the ou's according to constraints and their sort_key
568 static void co_sort_units(copy_opt_t *co) {
569 int i, count = 0, costs;
572 /* get the number of ous, remove them form the list and fill the array */
573 list_for_each_entry(unit_t, ou, &co->units, units)
575 ous = alloca(count * sizeof(*ous));
577 costs = co_get_max_copy_costs(co);
580 list_for_each_entry(unit_t, ou, &co->units, units)
583 INIT_LIST_HEAD(&co->units);
585 assert(count == i && list_empty(&co->units));
587 for (i=0; i<count; ++i)
588 ir_printf("%+F\n", ous[i]->nodes[0]);
590 qsort(ous, count, sizeof(*ous), compare_ous);
593 for (i=0; i<count; ++i)
594 ir_printf("%+F\n", ous[i]->nodes[0]);
596 /* reinsert into list in correct order */
597 for (i=0; i<count; ++i)
598 list_add_tail(&ous[i]->units, &co->units);
600 assert(costs == co_get_max_copy_costs(co));
604 void co_build_ou_structure(copy_opt_t *co) {
605 DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
606 INIT_LIST_HEAD(&co->units);
607 irg_walk_graph(co->irg, co_collect_units, NULL, co);
608 #ifdef QUICK_AND_DIRTY_HACK
613 void co_free_ou_structure(copy_opt_t *co) {
616 list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
621 co->units.next = NULL;
624 /* co_solve_heuristic() is implemented in becopyheur.c */
626 int co_get_max_copy_costs(const copy_opt_t *co) {
632 list_for_each_entry(unit_t, curr, &co->units, units) {
633 res += curr->inevitable_costs;
634 for (i=1; i<curr->node_count; ++i)
635 res += curr->costs[i];
640 int co_get_inevit_copy_costs(const copy_opt_t *co) {
646 list_for_each_entry(unit_t, curr, &co->units, units)
647 res += curr->inevitable_costs;
651 int co_get_copy_costs(const copy_opt_t *co) {
657 list_for_each_entry(unit_t, curr, &co->units, units) {
658 int root_col = get_irn_col(co, curr->nodes[0]);
659 DBG((dbg, LEVEL_1, " %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
660 res += curr->inevitable_costs;
661 for (i=1; i<curr->node_count; ++i) {
662 int arg_col = get_irn_col(co, curr->nodes[i]);
663 if (root_col != arg_col) {
664 DBG((dbg, LEVEL_1, " %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
665 res += curr->costs[i];
672 int co_get_lower_bound(const copy_opt_t *co) {
678 list_for_each_entry(unit_t, curr, &co->units, units)
679 res += curr->inevitable_costs + curr->min_nodes_costs;
683 void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat)
685 bitset_t *seen = bitset_irg_malloc(co->irg);
688 memset(stat, 0, sizeof(stat[0]));
690 /* count affinity edges. */
691 co_gs_foreach_aff_node(co, an) {
693 stat->aff_nodes += 1;
694 bitset_add_irn(seen, an->irn);
695 co_gs_foreach_neighb(an, neigh) {
696 if(!bitset_contains_irn(seen, neigh->irn)) {
697 stat->aff_edges += 1;
698 stat->max_costs += neigh->costs;
700 if(get_irn_col(co, an->irn) != get_irn_col(co, neigh->irn)) {
701 stat->costs += neigh->costs;
702 stat->unsatisfied_edges += 1;
705 if(nodes_interfere(co->cenv, an->irn, neigh->irn)) {
707 stat->inevit_costs += neigh->costs;
717 /******************************************************************************
719 / ____| | | / ____| |
720 | | __ _ __ __ _ _ __ | |__ | (___ | |_ ___ _ __ __ _ __ _ ___
721 | | |_ | '__/ _` | '_ \| '_ \ \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
722 | |__| | | | (_| | |_) | | | | ____) | || (_) | | | (_| | (_| | __/
723 \_____|_| \__,_| .__/|_| |_| |_____/ \__\___/|_| \__,_|\__, |\___|
726 ******************************************************************************/
728 static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) {
729 const affinity_node_t *n1 = k1;
730 const affinity_node_t *n2 = k2;
733 return (n1->irn != n2->irn);
736 static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
737 affinity_node_t new_node, *node;
743 new_node.neighbours = NULL;
744 node = set_insert(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
746 for (nbr = node->neighbours; nbr; nbr = nbr->next)
747 if (nbr->irn == n2) {
752 /* if we did not find n2 in n1's neighbourhood insert it */
754 nbr = obstack_alloc(&co->obst, sizeof(*nbr));
757 nbr->next = node->neighbours;
759 node->neighbours = nbr;
763 /* now nbr points to n1's neighbour-entry of n2 */
767 static INLINE void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
768 if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
769 add_edge(co, n1, n2, costs);
770 add_edge(co, n2, n1, costs);
774 static void build_graph_walker(ir_node *irn, void *env) {
775 copy_opt_t *co = env;
777 const arch_register_t *reg;
779 if (!is_curr_reg_class(co, irn) || arch_irn_is(co->aenv, irn, ignore))
782 reg = arch_get_irn_register(co->aenv, irn);
783 if (arch_register_type_is(reg, ignore))
786 if (is_Reg_Phi(irn)) { /* Phis */
787 for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
788 ir_node *arg = get_irn_n(irn, pos);
789 add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
792 else if (is_Perm_Proj(co->aenv, irn)) { /* Perms */
793 ir_node *arg = get_Perm_src(irn);
794 add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
796 else { /* 2-address code */
797 const arch_register_req_t *req = arch_get_register_req(co->aenv, irn, -1);
798 if (is_2addr_code(req)) {
799 const unsigned other = req->other_same;
802 for (i = 0; 1U << i <= other; ++i) {
803 if (other & (1U << i)) {
804 ir_node *other = get_irn_n(skip_Proj(irn), i);
805 if (! arch_irn_is(co->aenv, other, ignore))
806 add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
813 void co_build_graph_structure(copy_opt_t *co) {
814 obstack_init(&co->obst);
815 co->nodes = new_set(compare_affinity_node_t, 32);
817 irg_walk_graph(co->irg, build_graph_walker, NULL, co);
820 void co_free_graph_structure(copy_opt_t *co) {
824 obstack_free(&co->obst, NULL);
828 /* co_solve_ilp1() co_solve_ilp2() are implemented in becopyilpX.c */
830 int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) {
831 affinity_node_t new_node, *n;
836 n = set_find(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
838 return (n->degree > 0);
843 void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
845 be_ifg_t *ifg = co->cenv->ifg;
846 int *color_map = alloca(co->cls->n_regs * sizeof(color_map[0]));
854 for(i = 0; i < co->cls->n_regs; ++i) {
855 const arch_register_t *reg = &co->cls->regs[i];
856 color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_regs++;
860 * n contains the first node number.
861 * the values below n are the pre-colored register nodes
864 it = be_ifg_nodes_iter_alloca(ifg);
865 nit = be_ifg_neighbours_iter_alloca(ifg);
868 be_ifg_foreach_node(ifg, it, irn) {
869 if(!arch_irn_is(co->aenv, irn, ignore))
870 set_irn_link(irn, INT_TO_PTR(n++));
873 fprintf(f, "%d %d\n", n, n_regs);
875 be_ifg_foreach_node(ifg, it, irn) {
876 if(!arch_irn_is(co->aenv, irn, ignore)) {
877 int idx = PTR_TO_INT(get_irn_link(irn));
878 affinity_node_t *a = get_affinity_info(co, irn);
880 const arch_register_req_t *req;
883 req = arch_get_register_req(co->aenv, irn, BE_OUT_POS(0));
884 if(arch_register_req_is(req, limited)) {
885 for(i = 0; i < co->cls->n_regs; ++i) {
886 if(!rbitset_is_set(req->limited, i) && color_map[i] >= 0)
887 fprintf(f, "%d %d -1\n", color_map[i], idx);
892 be_ifg_foreach_neighbour(ifg, nit, irn, adj) {
893 if(!arch_irn_is(co->aenv, adj, ignore)) {
894 int adj_idx = PTR_TO_INT(get_irn_link(adj));
896 fprintf(f, "%d %d -1\n", idx, adj_idx);
903 co_gs_foreach_neighb(a, n) {
904 if(!arch_irn_is(co->aenv, n->irn, ignore)) {
905 int n_idx = PTR_TO_INT(get_irn_link(n->irn));
907 fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs);
915 typedef struct _appel_clique_walker_t {
917 const copy_opt_t *co;
924 } appel_clique_walker_t;
926 typedef struct _appel_block_info_t {
936 } appel_block_info_t;
938 static int appel_aff_weight(const appel_clique_walker_t *env, ir_node *bl)
941 double freq = get_block_execfreq(env->co->cenv->execfreq, bl);
942 int res = (int) freq;
943 return res == 0 ? 1 : res;
945 ir_loop *loop = get_irn_loop(bl);
948 int d = get_loop_depth(loop);
955 static void *appel_clique_walker_irn_init(ir_phase *phase, ir_node *irn, void *old)
957 appel_block_info_t *res = NULL;
961 appel_clique_walker_t *d = (void *) phase;
962 res = phase_alloc(phase, sizeof(res[0]));
963 res->phi_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
964 res->live_end_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
965 res->live_in_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in_nr));
966 res->live_end = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end));
967 res->live_in = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
968 res->phi = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
974 typedef struct _insn_list_t {
976 struct list_head list;
979 static int appel_get_live_end_nr(appel_clique_walker_t *env, ir_node *bl, ir_node *irn)
981 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
984 for(i = 0; i < bli->n_live_end; ++i)
985 if(bli->live_end[i] == irn)
986 return bli->live_end_nr[i];
991 static int appel_dump_clique(appel_clique_walker_t *env, const ir_nodeset_t *live, ir_node *bl, int curr_nr, int start_nr)
993 ir_node **live_arr = alloca(env->co->cls->n_regs * sizeof(live_arr[0]));
997 ir_nodeset_iterator_t iter;
1000 foreach_ir_nodeset(live, irn, iter)
1001 live_arr[n_live++] = irn;
1003 /* dump the live after clique */
1005 for(j = 0; j < n_live; ++j) {
1008 for(k = j + 1; k < n_live; ++k) {
1009 fprintf(env->f, "%d %d -1 ", curr_nr + j, curr_nr + k);
1011 fprintf(env->f, "\n");
1015 /* dump the affinities */
1016 for(j = 0; !env->dumb && j < n_live; ++j) {
1017 ir_node *irn = live_arr[j];
1018 int old_nr = PTR_TO_INT(get_irn_link(irn));
1020 /* if the node was already live in the last insn dump the affinity */
1021 if(old_nr > start_nr) {
1022 int weight = appel_aff_weight(env, bl);
1023 fprintf(env->f, "%d %d %d\n", old_nr, curr_nr + j, weight);
1027 /* set the current numbers into the link field. */
1028 for(j = 0; j < n_live; ++j) {
1029 ir_node *irn = live_arr[j];
1030 set_irn_link(irn, INT_TO_PTR(curr_nr + j));
1033 return curr_nr + n_live;
1036 static void appel_walker(ir_node *bl, void *data)
1038 appel_clique_walker_t *env = data;
1039 appel_block_info_t *bli = phase_get_or_set_irn_data(&env->ph, bl);
1040 struct obstack *obst = &env->obst;
1041 void *base = obstack_base(obst);
1043 ir_nodeset_iterator_t iter;
1044 be_lv_t *lv = env->co->cenv->birg->lv;
1048 int start_nr = env->curr_nr;
1049 int curr_nr = start_nr;
1051 be_insn_env_t insn_env;
1056 insn_env.aenv = env->co->aenv;
1057 insn_env.cls = env->co->cls;
1058 insn_env.obst = obst;
1059 insn_env.ignore_colors = env->co->cenv->ignore_colors;
1061 /* Guess how many insns will be in this block. */
1062 sched_foreach(bl, irn)
1066 insns = xmalloc(n_nodes * sizeof(insns[0]));
1068 /* Put all insns in an array. */
1069 irn = sched_first(bl);
1070 while(!sched_is_end(irn)) {
1072 insn = be_scan_insn(&insn_env, irn);
1073 insns[n_insns++] = insn;
1074 irn = insn->next_insn;
1077 DBG((dbg, LEVEL_2, "%+F\n", bl));
1078 ir_nodeset_init(&live);
1079 be_liveness_end_of_block(lv, env->co->aenv, env->co->cls, bl, &live);
1081 /* Generate the bad and ugly. */
1082 for(i = n_insns - 1; i >= 0; --i) {
1083 be_insn_t *insn = insns[i];
1085 /* The first live set has to be saved in the block border set. */
1086 if(i == n_insns - 1) {
1088 foreach_ir_nodeset(&live, irn, iter) {
1089 bli->live_end[j] = irn;
1090 bli->live_end_nr[j] = curr_nr + j;
1093 bli->n_live_end = j;
1097 for(j = 0; j < insn->use_start; ++j) {
1098 ir_node *op = insn->ops[j].carrier;
1099 bitset_t *adm = insn->ops[j].regs;
1103 if(!insn->ops[j].has_constraints)
1107 foreach_ir_nodeset(&live, irn, iter) {
1114 assert(nr < ir_nodeset_size(&live));
1116 for(k = 0; k < env->co->cls->n_regs; ++k) {
1117 int mapped_col = env->color_map[k];
1118 if(mapped_col >= 0 && !bitset_is_set(adm, k) && !bitset_is_set(env->co->cenv->ignore_colors, k))
1119 fprintf(env->f, "%d %d -1\n", curr_nr + nr, mapped_col);
1124 /* dump the clique and update the stuff. */
1125 curr_nr = appel_dump_clique(env, &live, bl, curr_nr, start_nr);
1127 /* remove all defs. */
1128 for(j = 0; j < insn->use_start; ++j)
1129 ir_nodeset_remove(&live, insn->ops[j].carrier);
1131 if(is_Phi(insn->irn) && arch_irn_consider_in_reg_alloc(env->co->aenv, env->co->cls, insn->irn)) {
1132 bli->phi[bli->n_phi] = insn->irn;
1133 bli->phi_nr[bli->n_phi] = PTR_TO_INT(get_irn_link(insn->irn));
1139 for(j = insn->use_start; j < insn->n_ops; ++j)
1140 ir_nodeset_insert(&live, insn->ops[j].carrier);
1143 /* print the start clique. */
1144 curr_nr = appel_dump_clique(env, &live, bl, curr_nr, start_nr);
1147 foreach_ir_nodeset(&live, irn, iter) {
1148 bli->live_in[i] = irn;
1149 bli->live_in_nr[i] = PTR_TO_INT(get_irn_link(irn));
1154 ir_nodeset_destroy(&live);
1156 obstack_free(obst, base);
1157 env->curr_nr = curr_nr;
1160 static void appel_inter_block_aff(ir_node *bl, void *data)
1162 appel_clique_walker_t *env = data;
1163 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
1167 for(i = 0; i < bli->n_live_in; ++i) {
1168 ir_node *irn = bli->live_in[i];
1170 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1171 ir_node *pred = get_Block_cfgpred_block(bl, j);
1173 int nr = appel_get_live_end_nr(env, pred, irn);
1175 fprintf(env->f, "%d %d 1\n", bli->live_in_nr[i], nr);
1179 for(i = 0; i < bli->n_phi; ++i) {
1180 ir_node *irn = bli->phi[i];
1182 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1183 ir_node *pred = get_Block_cfgpred_block(bl, j);
1184 ir_node *op = get_irn_n(irn, j);
1186 int nr = appel_get_live_end_nr(env, pred, op);
1188 fprintf(env->f, "%d %d 1\n", bli->phi_nr[i], nr);
1194 void co_dump_appel_graph_cliques(const copy_opt_t *co, FILE *f)
1198 appel_clique_walker_t env;
1199 bitset_t *adm = bitset_alloca(co->cls->n_regs);
1200 be_lv_t *lv = co->cenv->birg->lv;
1202 be_liveness_recompute(lv);
1203 obstack_init(&env.obst);
1204 phase_init(&env.ph, "appel_clique_dumper", co->irg, PHASE_DEFAULT_GROWTH, appel_clique_walker_irn_init, NULL);
1205 env.curr_nr = co->cls->n_regs;
1209 bitset_copy(adm, co->cenv->ignore_colors);
1210 bitset_flip_all(adm);
1212 /* Make color map. */
1213 env.color_map = alloca(co->cls->n_regs * sizeof(env.color_map[0]));
1214 for(i = 0, n_colors = 0; i < co->cls->n_regs; ++i) {
1215 const arch_register_t *reg = &co->cls->regs[i];
1216 env.color_map[i] = arch_register_type_is(reg, ignore) ? -1 : (int) n_colors++;
1220 env.curr_nr = n_colors;
1221 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1222 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1224 fprintf(f, "%d %d\n", env.curr_nr, n_colors);
1226 /* make the first k nodes interfere */
1227 for(i = 0; i < n_colors; ++i) {
1229 for(j = i + 1; j < n_colors; ++j)
1230 fprintf(f, "%d %d -1 ", i, j);
1235 env.curr_nr = n_colors;
1236 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1237 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1238 irg_block_walk_graph(co->irg, appel_inter_block_aff, NULL, &env);
1239 obstack_free(&env.obst, NULL);
1243 ___ _____ ____ ____ ___ _____ ____ _
1244 |_ _| ___/ ___| | _ \ / _ \_ _| | _ \ _ _ _ __ ___ _ __ (_)_ __ __ _
1245 | || |_ | | _ | | | | | | || | | | | | | | | '_ ` _ \| '_ \| | '_ \ / _` |
1246 | || _|| |_| | | |_| | |_| || | | |_| | |_| | | | | | | |_) | | | | | (_| |
1247 |___|_| \____| |____/ \___/ |_| |____/ \__,_|_| |_| |_| .__/|_|_| |_|\__, |
1251 static const char *get_dot_color_name(size_t col)
1253 static const char *names[] = {
1287 return col < sizeof(names)/sizeof(names[0]) ? names[col] : "white";
1290 typedef struct _co_ifg_dump_t {
1291 const copy_opt_t *co;
1295 static void ifg_dump_graph_attr(FILE *f, void *self)
1298 fprintf(f, "overlap=scale");
1301 static int ifg_is_dump_node(void *self, ir_node *irn)
1303 co_ifg_dump_t *cod = self;
1304 return !arch_irn_is(cod->co->aenv, irn, ignore);
1307 static void ifg_dump_node_attr(FILE *f, void *self, ir_node *irn)
1309 co_ifg_dump_t *env = self;
1310 const arch_register_t *reg = arch_get_irn_register(env->co->aenv, irn);
1311 const arch_register_req_t *req;
1314 req = arch_get_register_req(env->co->aenv, irn, BE_OUT_POS(0));
1315 limited = arch_register_req_is(req, limited);
1317 if(env->flags & CO_IFG_DUMP_LABELS) {
1318 ir_fprintf(f, "label=\"%+F", irn);
1322 if((env->flags & CO_IFG_DUMP_CONSTR) && limited) {
1323 bitset_t *bs = bitset_alloca(env->co->cls->n_regs);
1324 req.limited(req.limited_env, bs);
1325 ir_fprintf(f, "\\n%B", bs);
1328 ir_fprintf(f, "\" ");
1330 fprintf(f, "label=\"\" shape=point " );
1333 if(env->flags & CO_IFG_DUMP_SHAPE)
1334 fprintf(f, "shape=%s ", limited ? "diamond" : "ellipse");
1336 if(env->flags & CO_IFG_DUMP_COLORS)
1337 fprintf(f, "style=filled color=%s ", get_dot_color_name(reg->index));
1340 static void ifg_dump_at_end(FILE *file, void *self)
1342 co_ifg_dump_t *env = self;
1345 co_gs_foreach_aff_node(env->co, a) {
1346 const arch_register_t *ar = arch_get_irn_register(env->co->aenv, a->irn);
1347 unsigned aidx = get_irn_idx(a->irn);
1350 co_gs_foreach_neighb(a, n) {
1351 const arch_register_t *nr = arch_get_irn_register(env->co->aenv, n->irn);
1352 unsigned nidx = get_irn_idx(n->irn);
1355 const char *color = nr == ar ? "blue" : "red";
1356 fprintf(file, "\tn%d -- n%d [weight=0.01 ", aidx, nidx);
1357 if(env->flags & CO_IFG_DUMP_LABELS)
1358 fprintf(file, "label=\"%d\" ", n->costs);
1359 if(env->flags & CO_IFG_DUMP_COLORS)
1360 fprintf(file, "color=%s ", color);
1362 fprintf(file, "style=dotted");
1363 fprintf(file, "];\n");
1370 static be_ifg_dump_dot_cb_t ifg_dot_cb = {
1372 ifg_dump_graph_attr,
1381 void co_dump_ifg_dot(const copy_opt_t *co, FILE *f, unsigned flags)
1387 be_ifg_dump_dot(co->cenv->ifg, co->irg, f, &ifg_dot_cb, &cod);
1391 void co_solve_park_moon(copy_opt_t *opt)
1396 static int void_algo(copy_opt_t *co)
1404 / \ | | __ _ ___ _ __(_) |_| |__ _ __ ___ ___
1405 / _ \ | |/ _` |/ _ \| '__| | __| '_ \| '_ ` _ \/ __|
1406 / ___ \| | (_| | (_) | | | | |_| | | | | | | | \__ \
1407 /_/ \_\_|\__, |\___/|_| |_|\__|_| |_|_| |_| |_|___/
1414 int can_improve_existing;
1417 static co_algo_info_t algos[] = {
1418 { void_algo, "none", 0 },
1419 { co_solve_heuristic, "heur1", 0 },
1420 { co_solve_heuristic_new, "heur2", 0 },
1422 { co_solve_heuristic_java, "heur3", 0 },
1424 { NULL, "heur3", 0 },
1426 { co_solve_heuristic_mst, "heur4", 0 },
1428 { co_solve_ilp2, "ilp", 1 },
1437 | \/ | __ _(_)_ __ | _ \ _ __(_)_ _____ _ __
1438 | |\/| |/ _` | | '_ \ | | | | '__| \ \ / / _ \ '__|
1439 | | | | (_| | | | | | | |_| | | | |\ V / __/ |
1440 |_| |_|\__,_|_|_| |_| |____/|_| |_| \_/ \___|_|
1444 static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char *suffix)
1449 ir_snprintf(buf, sizeof(buf), "%s%F_%s%s", prefix, env->irg, env->cls->name, suffix);
1450 result = fopen(buf, "wt");
1451 if(result == NULL) {
1452 panic("Couldn't open '%s' for writing.", buf);
1458 void co_driver(be_chordal_env_t *cenv)
1460 lc_timer_t *timer = lc_timer_register("firm.be.copyopt", "runtime");
1461 co_complete_stats_t before, after;
1463 co_algo_t *algo_func;
1464 int was_optimal = 0;
1466 if (algo >= CO_ALGO_LAST)
1469 be_liveness_assure_chk(be_get_birg_liveness(cenv->birg));
1471 co = new_copy_opt(cenv, cost_func);
1472 co_build_ou_structure(co);
1473 co_build_graph_structure(co);
1475 co_complete_stats(co, &before);
1477 be_stat_ev_ull("co_aff_nodes", before.aff_nodes);
1478 be_stat_ev_ull("co_aff_edges", before.aff_edges);
1479 be_stat_ev_ull("co_max_costs", before.max_costs);
1480 be_stat_ev_ull("co_inevit_costs", before.inevit_costs);
1481 be_stat_ev_ull("co_aff_int", before.aff_int);
1483 be_stat_ev_ull("co_init_costs", before.costs);
1484 be_stat_ev_ull("co_init_unsat", before.unsatisfied_edges);
1486 /* Dump the interference graph in Appel's format. */
1487 if (dump_flags & DUMP_APPEL) {
1488 FILE *f = my_open(cenv, "", ".apl");
1489 co_dump_appel_graph(co, f);
1493 if (dump_flags & DUMP_BEFORE) {
1494 FILE *f = my_open(cenv, "", "-before.dot");
1495 co_dump_ifg_dot(co, f, style_flags);
1499 /* if the algo can improve results, provide an initial solution with heur3 */
1500 if (improve && algos[algo].can_improve_existing) {
1501 co_complete_stats_t stats;
1503 /* produce a heuristic solution */
1505 co_solve_heuristic_java(co);
1507 co_solve_heuristic(co);
1508 #endif /* WITH_JVM */
1510 /* do the stats and provide the current costs */
1511 co_complete_stats(co, &stats);
1512 be_stat_ev_ull("co_prepare_costs", stats.costs);
1516 /* start the JVM here so that it does not tamper the timing. */
1517 if (algo == CO_ALGO_HEUR3)
1518 be_java_coal_start_jvm();
1519 #endif /* WITH_JVM */
1521 algo_func = algos[algo].algo;
1523 /* perform actual copy minimization */
1524 lc_timer_reset_and_start(timer);
1525 was_optimal = algo_func(co);
1526 lc_timer_stop(timer);
1528 be_stat_ev("co_time", lc_timer_elapsed_msec(timer));
1529 be_stat_ev_ull("co_optimal", was_optimal);
1531 if (dump_flags & DUMP_AFTER) {
1532 FILE *f = my_open(cenv, "", "-after.dot");
1533 co_dump_ifg_dot(co, f, style_flags);
1537 co_complete_stats(co, &after);
1540 ulong64 optimizable_costs = after.max_costs - after.inevit_costs;
1541 ulong64 evitable = after.costs - after.inevit_costs;
1543 ir_printf("%30F ", cenv->irg);
1544 printf("%10s %10" ULL_FMT "%10" ULL_FMT "%10" ULL_FMT, cenv->cls->name, after.max_costs, before.costs, after.inevit_costs);
1546 if(optimizable_costs > 0)
1547 printf("%10" ULL_FMT " %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs);
1549 printf("%10" ULL_FMT " %5s\n", after.costs, "-");
1552 be_stat_ev_ull("co_after_costs", after.costs);
1553 be_stat_ev_ull("co_after_unsat", after.unsatisfied_edges);
1555 co_free_graph_structure(co);
1556 co_free_ou_structure(co);