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 },
111 { "heur3", CO_ALGO_HEUR3 },
112 #endif /* WITH_JVM */
113 { "heur4", CO_ALGO_HEUR4 },
115 { "ilp", CO_ALGO_ILP },
116 #endif /* WITH_ILP */
120 typedef int (*opt_funcptr)(void);
122 static const lc_opt_enum_func_ptr_items_t cost_func_items[] = {
123 { "freq", (opt_funcptr) co_get_costs_exec_freq },
124 { "loop", (opt_funcptr) co_get_costs_loop_depth },
125 { "one", (opt_funcptr) co_get_costs_all_one },
129 static lc_opt_enum_mask_var_t dump_var = {
130 &dump_flags, dump_items
133 static lc_opt_enum_mask_var_t style_var = {
134 &style_flags, style_items
137 static lc_opt_enum_mask_var_t algo_var = {
141 static lc_opt_enum_func_ptr_var_t cost_func_var = {
142 (opt_funcptr*) &cost_func, cost_func_items
145 static const lc_opt_table_entry_t options[] = {
146 LC_OPT_ENT_ENUM_INT ("algo", "select copy optimization algo", &algo_var),
147 LC_OPT_ENT_ENUM_FUNC_PTR ("cost", "select a cost function", &cost_func_var),
148 LC_OPT_ENT_ENUM_MASK ("dump", "dump ifg before or after copy optimization", &dump_var),
149 LC_OPT_ENT_ENUM_MASK ("style", "dump style for ifg dumping", &style_var),
150 LC_OPT_ENT_BOOL ("stats", "dump statistics after each optimization", &do_stats),
151 LC_OPT_ENT_BOOL ("improve", "run heur3 before if algo can exploit start solutions", &improve),
155 /* Insert additional options registration functions here. */
156 extern void be_co_ilp_register_options(lc_opt_entry_t *grp);
157 extern void be_co2_register_options(lc_opt_entry_t *grp);
158 extern void be_co3_register_options(lc_opt_entry_t *grp);
160 void be_init_copycoal(void)
162 lc_opt_entry_t *be_grp = lc_opt_get_grp(firm_opt_get_root(), "be");
163 lc_opt_entry_t *ra_grp = lc_opt_get_grp(be_grp, "ra");
164 lc_opt_entry_t *chordal_grp = lc_opt_get_grp(ra_grp, "chordal");
165 lc_opt_entry_t *co_grp = lc_opt_get_grp(chordal_grp, "co");
167 lc_opt_add_table(co_grp, options);
170 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copycoal);
172 #undef QUICK_AND_DIRTY_HACK
174 static int nodes_interfere(const be_chordal_env_t *env, const ir_node *a, const ir_node *b)
177 return be_ifg_connected(env->ifg, a, b);
179 return values_interfere(env->birg, a, b);
183 /******************************************************************************
186 | | __ ___ _ __ ___ _ __ __ _| |
187 | | |_ |/ _ \ '_ \ / _ \ '__/ _` | |
188 | |__| | __/ | | | __/ | | (_| | |
189 \_____|\___|_| |_|\___|_| \__,_|_|
191 ******************************************************************************/
193 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
196 copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs)
198 const char *s1, *s2, *s3;
202 FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
204 co = xcalloc(1, sizeof(*co));
205 co->cenv = chordal_env;
206 co->aenv = chordal_env->birg->main_env->arch_env;
207 co->irg = chordal_env->irg;
208 co->cls = chordal_env->cls;
209 co->get_costs = get_costs;
211 s1 = get_irp_prog_name();
212 s2 = get_entity_name(get_irg_entity(co->irg));
213 s3 = chordal_env->cls->name;
214 len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
215 co->name = xmalloc(len);
216 snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
221 void free_copy_opt(copy_opt_t *co) {
226 int co_is_optimizable_root(const copy_opt_t *co, ir_node *irn) {
227 const arch_register_req_t *req;
228 const arch_register_t *reg;
230 if (arch_irn_is(co->aenv, irn, ignore))
233 reg = arch_get_irn_register(co->aenv, irn);
234 if (arch_register_type_is(reg, ignore))
237 req = arch_get_register_req(co->aenv, irn, -1);
238 if (is_Reg_Phi(irn) || is_Perm_Proj(co->aenv, irn) || is_2addr_code(req))
244 int co_get_costs_loop_depth(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
247 ir_node *root_block = get_nodes_block(root);
252 /* for phis the copies are placed in the corresponding pred-block */
253 loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos));
255 /* a perm places the copy in the same block as it resides */
256 loop = get_irn_loop(root_block);
259 int d = get_loop_depth(loop);
265 int co_get_costs_exec_freq(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
267 ir_node *root_bl = get_nodes_block(root);
268 ir_node *copy_bl = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl;
270 res = get_block_execfreq_ulong(co->cenv->birg->exec_freq, copy_bl);
272 /* don't allow values smaller than one. */
273 return res < 1 ? 1 : res;
277 int co_get_costs_all_one(const copy_opt_t *co, ir_node *root, ir_node *arg, int pos) {
285 /******************************************************************************
286 ____ _ _ _ _ _ _____ _
287 / __ \ | | | | | | (_) | / ____| |
288 | | | |_ __ | |_| | | |_ __ _| |_ ___ | (___ | |_ ___ _ __ __ _ __ _ ___
289 | | | | '_ \| __| | | | '_ \| | __/ __| \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
290 | |__| | |_) | |_| |__| | | | | | |_\__ \ ____) | || (_) | | | (_| | (_| | __/
291 \____/| .__/ \__|\____/|_| |_|_|\__|___/ |_____/ \__\___/|_| \__,_|\__, |\___|
294 ******************************************************************************/
297 * Determines a maximum weighted independent set with respect to
298 * the interference and conflict edges of all nodes in a qnode.
300 static int ou_max_ind_set_costs(unit_t *ou) {
301 be_chordal_env_t *chordal_env = ou->co->cenv;
302 ir_node **safe, **unsafe;
303 int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
306 int max, curr_weight, best_weight = 0;
308 /* assign the nodes into two groups.
309 * safe: node has no interference, hence it is in every max stable set.
310 * unsafe: node has an interference
312 safe = alloca((ou->node_count-1) * sizeof(*safe));
315 unsafe = alloca((ou->node_count-1) * sizeof(*unsafe));
316 unsafe_costs = alloca((ou->node_count-1) * sizeof(*unsafe_costs));
318 for(i=1; i<ou->node_count; ++i) {
320 for(o=1; o<ou->node_count; ++o) {
323 if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) {
324 unsafe_costs[unsafe_count] = ou->costs[i];
325 unsafe[unsafe_count] = ou->nodes[i];
332 safe_costs += ou->costs[i];
333 safe[safe_count++] = ou->nodes[i];
338 /* now compute the best set out of the unsafe nodes*/
339 if (unsafe_count > MIS_HEUR_TRIGGER) {
340 bitset_t *best = bitset_alloca(unsafe_count);
341 /* Heuristik: Greedy trial and error form index 0 to unsafe_count-1 */
342 for (i=0; i<unsafe_count; ++i) {
344 /* check if it is a stable set */
345 for (o=bitset_next_set(best, 0); o!=-1 && o<i; o=bitset_next_set(best, o+1))
346 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o])) {
347 bitset_clear(best, i); /* clear the bit and try next one */
351 /* compute the weight */
352 bitset_foreach(best, pos)
353 best_weight += unsafe_costs[pos];
355 /* Exact Algorithm: Brute force */
356 curr = bitset_alloca(unsafe_count);
357 bitset_set_all(curr);
358 while ((max = bitset_popcnt(curr)) != 0) {
359 /* check if curr is a stable set */
360 for (i=bitset_next_set(curr, 0); i!=-1; i=bitset_next_set(curr, i+1))
361 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) */
362 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o]))
365 /* if we arrive here, we have a stable set */
366 /* compute the weigth of the stable set*/
368 bitset_foreach(curr, pos)
369 curr_weight += unsafe_costs[pos];
372 if (curr_weight > best_weight) {
373 best_weight = curr_weight;
381 return safe_costs+best_weight;
384 static void co_collect_units(ir_node *irn, void *env) {
385 copy_opt_t *co = env;
388 if (!is_curr_reg_class(co, irn))
390 if (!co_is_optimizable_root(co, irn))
393 /* Init a new unit */
394 unit = xcalloc(1, sizeof(*unit));
396 unit->node_count = 1;
397 INIT_LIST_HEAD(&unit->queue);
399 /* Phi with some/all of its arguments */
400 if (is_Reg_Phi(irn)) {
404 arity = get_irn_arity(irn);
405 unit->nodes = xmalloc((arity+1) * sizeof(*unit->nodes));
406 unit->costs = xmalloc((arity+1) * sizeof(*unit->costs));
407 unit->nodes[0] = irn;
410 for (i=0; i<arity; ++i) {
412 ir_node *arg = get_irn_n(irn, i);
414 assert(is_curr_reg_class(co, arg) && "Argument not in same register class.");
417 if (nodes_interfere(co->cenv, irn, arg)) {
418 unit->inevitable_costs += co->get_costs(co, irn, arg, i);
422 /* Else insert the argument of the phi to the members of this ou */
423 DBG((dbg, LEVEL_1, "\t Member: %+F\n", arg));
425 if (! arch_irn_is(co->aenv, arg, ignore)) {
426 /* Check if arg has occurred at a prior position in the arg/list */
428 for (o=1; o<unit->node_count; ++o) {
429 if (unit->nodes[o] == arg) {
435 if (!arg_pos) { /* a new argument */
436 /* insert node, set costs */
437 unit->nodes[unit->node_count] = arg;
438 unit->costs[unit->node_count] = co->get_costs(co, irn, arg, i);
440 } else { /* arg has occurred before in same phi */
441 /* increase costs for existing arg */
442 unit->costs[arg_pos] += co->get_costs(co, irn, arg, i);
446 unit->nodes = xrealloc(unit->nodes, unit->node_count * sizeof(*unit->nodes));
447 unit->costs = xrealloc(unit->costs, unit->node_count * sizeof(*unit->costs));
450 /* Proj of a perm with corresponding arg */
451 if (is_Perm_Proj(co->aenv, irn)) {
452 assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
453 unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
454 unit->costs = xmalloc(2 * sizeof(*unit->costs));
455 unit->node_count = 2;
456 unit->nodes[0] = irn;
457 unit->nodes[1] = get_Perm_src(irn);
458 unit->costs[1] = co->get_costs(co, irn, unit->nodes[1], -1);
460 const arch_register_req_t *req =
461 arch_get_register_req(co->aenv, irn, -1);
463 /* Src == Tgt of a 2-addr-code instruction */
464 if (is_2addr_code(req)) {
465 ir_node *other = get_irn_n(skip_Proj(irn), req->other_same);
466 if (!arch_irn_is(co->aenv, other, ignore) &&
467 !nodes_interfere(co->cenv, irn, other)) {
468 unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
469 unit->costs = xmalloc(2 * sizeof(*unit->costs));
470 unit->node_count = 2;
471 unit->nodes[0] = irn;
472 unit->nodes[1] = other;
473 unit->costs[1] = co->get_costs(co, irn, other, -1);
476 assert(0 && "This is not an optimizable node!");
480 /* Insert the new unit at a position according to its costs */
481 if (unit->node_count > 1) {
483 struct list_head *tmp;
485 /* Determine the maximum costs this unit can cause: all_nodes_cost */
486 for(i=1; i<unit->node_count; ++i) {
487 unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
488 unit->all_nodes_costs += unit->costs[i];
491 /* Determine the minimal costs this unit will cause: min_nodes_costs */
492 unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
493 /* Insert the new ou according to its sort_key */
495 while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
497 list_add(&unit->units, tmp);
503 #ifdef QUICK_AND_DIRTY_HACK
505 static int compare_ous(const void *k1, const void *k2) {
506 const unit_t *u1 = *((const unit_t **) k1);
507 const unit_t *u2 = *((const unit_t **) k2);
508 int i, o, u1_has_constr, u2_has_constr;
509 arch_register_req_t req;
510 const arch_env_t *aenv = u1->co->aenv;
512 /* Units with constraints come first */
514 for (i=0; i<u1->node_count; ++i) {
515 arch_get_register_req(aenv, &req, u1->nodes[i], -1);
516 if (arch_register_req_is(&req, limited)) {
523 for (i=0; i<u2->node_count; ++i) {
524 arch_get_register_req(aenv, &req, u2->nodes[i], -1);
525 if (arch_register_req_is(&req, limited)) {
531 if (u1_has_constr != u2_has_constr)
532 return u2_has_constr - u1_has_constr;
534 /* Now check, whether the two units are connected */
536 for (i=0; i<u1->node_count; ++i)
537 for (o=0; o<u2->node_count; ++o)
538 if (u1->nodes[i] == u2->nodes[o])
542 /* After all, the sort key decides. Greater keys come first. */
543 return u2->sort_key - u1->sort_key;
548 * Sort the ou's according to constraints and their sort_key
550 static void co_sort_units(copy_opt_t *co) {
551 int i, count = 0, costs;
554 /* get the number of ous, remove them form the list and fill the array */
555 list_for_each_entry(unit_t, ou, &co->units, units)
557 ous = alloca(count * sizeof(*ous));
559 costs = co_get_max_copy_costs(co);
562 list_for_each_entry(unit_t, ou, &co->units, units)
565 INIT_LIST_HEAD(&co->units);
567 assert(count == i && list_empty(&co->units));
569 for (i=0; i<count; ++i)
570 ir_printf("%+F\n", ous[i]->nodes[0]);
572 qsort(ous, count, sizeof(*ous), compare_ous);
575 for (i=0; i<count; ++i)
576 ir_printf("%+F\n", ous[i]->nodes[0]);
578 /* reinsert into list in correct order */
579 for (i=0; i<count; ++i)
580 list_add_tail(&ous[i]->units, &co->units);
582 assert(costs == co_get_max_copy_costs(co));
586 void co_build_ou_structure(copy_opt_t *co) {
587 DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
588 INIT_LIST_HEAD(&co->units);
589 irg_walk_graph(co->irg, co_collect_units, NULL, co);
590 #ifdef QUICK_AND_DIRTY_HACK
595 void co_free_ou_structure(copy_opt_t *co) {
598 list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
603 co->units.next = NULL;
606 /* co_solve_heuristic() is implemented in becopyheur.c */
608 int co_get_max_copy_costs(const copy_opt_t *co) {
614 list_for_each_entry(unit_t, curr, &co->units, units) {
615 res += curr->inevitable_costs;
616 for (i=1; i<curr->node_count; ++i)
617 res += curr->costs[i];
622 int co_get_inevit_copy_costs(const copy_opt_t *co) {
628 list_for_each_entry(unit_t, curr, &co->units, units)
629 res += curr->inevitable_costs;
633 int co_get_copy_costs(const copy_opt_t *co) {
639 list_for_each_entry(unit_t, curr, &co->units, units) {
640 int root_col = get_irn_col(co, curr->nodes[0]);
641 DBG((dbg, LEVEL_1, " %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
642 res += curr->inevitable_costs;
643 for (i=1; i<curr->node_count; ++i) {
644 int arg_col = get_irn_col(co, curr->nodes[i]);
645 if (root_col != arg_col) {
646 DBG((dbg, LEVEL_1, " %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
647 res += curr->costs[i];
654 int co_get_lower_bound(const copy_opt_t *co) {
660 list_for_each_entry(unit_t, curr, &co->units, units)
661 res += curr->inevitable_costs + curr->min_nodes_costs;
665 void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat)
667 bitset_t *seen = bitset_irg_malloc(co->irg);
670 memset(stat, 0, sizeof(stat[0]));
672 /* count affinity edges. */
673 co_gs_foreach_aff_node(co, an) {
675 stat->aff_nodes += 1;
676 bitset_add_irn(seen, an->irn);
677 co_gs_foreach_neighb(an, neigh) {
678 if(!bitset_contains_irn(seen, neigh->irn)) {
679 stat->aff_edges += 1;
680 stat->max_costs += neigh->costs;
682 if(get_irn_col(co, an->irn) != get_irn_col(co, neigh->irn)) {
683 stat->costs += neigh->costs;
684 stat->unsatisfied_edges += 1;
687 if(nodes_interfere(co->cenv, an->irn, neigh->irn)) {
689 stat->inevit_costs += neigh->costs;
699 /******************************************************************************
701 / ____| | | / ____| |
702 | | __ _ __ __ _ _ __ | |__ | (___ | |_ ___ _ __ __ _ __ _ ___
703 | | |_ | '__/ _` | '_ \| '_ \ \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
704 | |__| | | | (_| | |_) | | | | ____) | || (_) | | | (_| | (_| | __/
705 \_____|_| \__,_| .__/|_| |_| |_____/ \__\___/|_| \__,_|\__, |\___|
708 ******************************************************************************/
710 static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) {
711 const affinity_node_t *n1 = k1;
712 const affinity_node_t *n2 = k2;
715 return (n1->irn != n2->irn);
718 static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
719 affinity_node_t new_node, *node;
725 new_node.neighbours = NULL;
726 node = set_insert(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
728 for (nbr = node->neighbours; nbr; nbr = nbr->next)
729 if (nbr->irn == n2) {
734 /* if we did not find n2 in n1's neighbourhood insert it */
736 nbr = obstack_alloc(&co->obst, sizeof(*nbr));
739 nbr->next = node->neighbours;
741 node->neighbours = nbr;
745 /* now nbr points to n1's neighbour-entry of n2 */
749 static INLINE void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
750 if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
751 add_edge(co, n1, n2, costs);
752 add_edge(co, n2, n1, costs);
756 static void build_graph_walker(ir_node *irn, void *env) {
757 copy_opt_t *co = env;
759 const arch_register_t *reg;
761 if (!is_curr_reg_class(co, irn) || arch_irn_is(co->aenv, irn, ignore))
764 reg = arch_get_irn_register(co->aenv, irn);
765 if (arch_register_type_is(reg, ignore))
770 for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
771 ir_node *arg = get_irn_n(irn, pos);
772 add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
776 else if (is_Perm_Proj(co->aenv, irn)) {
777 ir_node *arg = get_Perm_src(irn);
778 add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
783 const arch_register_req_t *req =
784 arch_get_register_req(co->aenv, irn, -1);
785 if (is_2addr_code(req)) {
786 ir_node *other = get_irn_n(skip_Proj(irn), req->other_same);
787 if (! arch_irn_is(co->aenv, other, ignore))
788 add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
793 void co_build_graph_structure(copy_opt_t *co) {
794 obstack_init(&co->obst);
795 co->nodes = new_set(compare_affinity_node_t, 32);
797 irg_walk_graph(co->irg, build_graph_walker, NULL, co);
800 void co_free_graph_structure(copy_opt_t *co) {
804 obstack_free(&co->obst, NULL);
808 /* co_solve_ilp1() co_solve_ilp2() are implemented in becopyilpX.c */
810 int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) {
811 affinity_node_t new_node, *n;
816 n = set_find(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
818 return (n->degree > 0);
823 void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
825 be_ifg_t *ifg = co->cenv->ifg;
826 int *color_map = alloca(co->cls->n_regs * sizeof(color_map[0]));
833 for(i = 0; i < co->cls->n_regs; ++i) {
834 const arch_register_t *reg = &co->cls->regs[i];
835 color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_regs++;
839 * n contains the first node number.
840 * the values below n are the pre-colored register nodes
843 it = be_ifg_nodes_iter_alloca(ifg);
844 nit = be_ifg_neighbours_iter_alloca(ifg);
847 be_ifg_foreach_node(ifg, it, irn) {
848 if(!arch_irn_is(co->aenv, irn, ignore))
849 set_irn_link(irn, INT_TO_PTR(n++));
852 fprintf(f, "%d %d\n", n, n_regs);
854 be_ifg_foreach_node(ifg, it, irn) {
855 if(!arch_irn_is(co->aenv, irn, ignore)) {
856 int idx = PTR_TO_INT(get_irn_link(irn));
857 affinity_node_t *a = get_affinity_info(co, irn);
859 const arch_register_req_t *req;
862 req = arch_get_register_req(co->aenv, irn, BE_OUT_POS(0));
863 if(arch_register_req_is(req, limited)) {
864 for(i = 0; i < co->cls->n_regs; ++i) {
865 if(!rbitset_is_set(req->limited, i) && color_map[i] >= 0)
866 fprintf(f, "%d %d -1\n", color_map[i], idx);
871 be_ifg_foreach_neighbour(ifg, nit, irn, adj) {
872 if(!arch_irn_is(co->aenv, adj, ignore)) {
873 int adj_idx = PTR_TO_INT(get_irn_link(adj));
875 fprintf(f, "%d %d -1\n", idx, adj_idx);
882 co_gs_foreach_neighb(a, n) {
883 if(!arch_irn_is(co->aenv, n->irn, ignore)) {
884 int n_idx = PTR_TO_INT(get_irn_link(n->irn));
886 fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs);
894 typedef struct _appel_clique_walker_t {
896 const copy_opt_t *co;
903 } appel_clique_walker_t;
905 typedef struct _appel_block_info_t {
915 } appel_block_info_t;
917 static int appel_aff_weight(const appel_clique_walker_t *env, ir_node *bl)
920 double freq = get_block_execfreq(env->co->cenv->execfreq, bl);
921 int res = (int) freq;
922 return res == 0 ? 1 : res;
924 ir_loop *loop = get_irn_loop(bl);
927 int d = get_loop_depth(loop);
934 static void *appel_clique_walker_irn_init(ir_phase *phase, ir_node *irn, void *old)
936 appel_block_info_t *res = NULL;
940 appel_clique_walker_t *d = (void *) phase;
941 res = phase_alloc(phase, sizeof(res[0]));
942 res->phi_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
943 res->live_end_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
944 res->live_in_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in_nr));
945 res->live_end = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end));
946 res->live_in = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
947 res->phi = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
953 typedef struct _insn_list_t {
955 struct list_head list;
958 static int appel_get_live_end_nr(appel_clique_walker_t *env, ir_node *bl, ir_node *irn)
960 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
963 for(i = 0; i < bli->n_live_end; ++i)
964 if(bli->live_end[i] == irn)
965 return bli->live_end_nr[i];
970 static int appel_dump_clique(appel_clique_walker_t *env, pset *live, ir_node *bl, int curr_nr, int start_nr)
972 ir_node **live_arr = alloca(env->co->cls->n_regs * sizeof(live_arr[0]));
978 foreach_pset(live, irn)
979 live_arr[n_live++] = irn;
981 /* dump the live after clique */
983 for(j = 0; j < n_live; ++j) {
986 for(k = j + 1; k < n_live; ++k) {
987 fprintf(env->f, "%d %d -1 ", curr_nr + j, curr_nr + k);
989 fprintf(env->f, "\n");
993 /* dump the affinities */
994 for(j = 0; !env->dumb && j < n_live; ++j) {
995 ir_node *irn = live_arr[j];
996 int old_nr = PTR_TO_INT(get_irn_link(irn));
998 /* if the node was already live in the last insn dump the affinity */
999 if(old_nr > start_nr) {
1000 int weight = appel_aff_weight(env, bl);
1001 fprintf(env->f, "%d %d %d\n", old_nr, curr_nr + j, weight);
1005 /* set the current numbers into the link field. */
1006 for(j = 0; j < n_live; ++j) {
1007 ir_node *irn = live_arr[j];
1008 set_irn_link(irn, INT_TO_PTR(curr_nr + j));
1011 return curr_nr + n_live;
1014 static void appel_walker(ir_node *bl, void *data)
1016 appel_clique_walker_t *env = data;
1017 appel_block_info_t *bli = phase_get_or_set_irn_data(&env->ph, bl);
1018 struct obstack *obst = &env->obst;
1019 void *base = obstack_base(obst);
1020 pset *live = pset_new_ptr_default();
1021 be_lv_t *lv = env->co->cenv->birg->lv;
1025 int start_nr = env->curr_nr;
1026 int curr_nr = start_nr;
1028 be_insn_env_t insn_env;
1033 insn_env.aenv = env->co->aenv;
1034 insn_env.cls = env->co->cls;
1035 insn_env.obst = obst;
1036 insn_env.ignore_colors = env->co->cenv->ignore_colors;
1038 /* Guess how many insns will be in this block. */
1039 sched_foreach(bl, irn)
1043 insns = xmalloc(n_nodes * sizeof(insns[0]));
1045 /* Put all insns in an array. */
1046 irn = sched_first(bl);
1047 while(!sched_is_end(irn)) {
1049 insn = be_scan_insn(&insn_env, irn);
1050 insns[n_insns++] = insn;
1051 irn = insn->next_insn;
1054 DBG((dbg, LEVEL_2, "%+F\n", bl));
1055 be_liveness_end_of_block(lv, env->co->aenv, env->co->cls, bl, live);
1057 /* Generate the bad and ugly. */
1058 for(i = n_insns - 1; i >= 0; --i) {
1059 be_insn_t *insn = insns[i];
1061 /* The first live set has to be saved in the block border set. */
1062 if(i == n_insns - 1) {
1064 foreach_pset(live, irn) {
1065 bli->live_end[j] = irn;
1066 bli->live_end_nr[j] = curr_nr + j;
1069 bli->n_live_end = j;
1073 for(j = 0; j < insn->use_start; ++j) {
1074 ir_node *op = insn->ops[j].carrier;
1075 bitset_t *adm = insn->ops[j].regs;
1079 if(!insn->ops[j].has_constraints)
1083 foreach_pset(live, irn) {
1091 assert(nr < pset_count(live));
1093 for(k = 0; k < env->co->cls->n_regs; ++k) {
1094 int mapped_col = env->color_map[k];
1095 if(mapped_col >= 0 && !bitset_is_set(adm, k) && !bitset_is_set(env->co->cenv->ignore_colors, k))
1096 fprintf(env->f, "%d %d -1\n", curr_nr + nr, mapped_col);
1101 /* dump the clique and update the stuff. */
1102 curr_nr = appel_dump_clique(env, live, bl, curr_nr, start_nr);
1104 /* remove all defs. */
1105 for(j = 0; j < insn->use_start; ++j)
1106 pset_remove_ptr(live, insn->ops[j].carrier);
1108 if(is_Phi(insn->irn) && arch_irn_consider_in_reg_alloc(env->co->aenv, env->co->cls, insn->irn)) {
1109 bli->phi[bli->n_phi] = insn->irn;
1110 bli->phi_nr[bli->n_phi] = PTR_TO_INT(get_irn_link(insn->irn));
1116 for(j = insn->use_start; j < insn->n_ops; ++j)
1117 pset_insert_ptr(live, insn->ops[j].carrier);
1120 /* print the start clique. */
1121 curr_nr = appel_dump_clique(env, live, bl, curr_nr, start_nr);
1124 foreach_pset(live, irn) {
1125 bli->live_in[i] = irn;
1126 bli->live_in_nr[i] = PTR_TO_INT(get_irn_link(irn));
1133 obstack_free(obst, base);
1134 env->curr_nr = curr_nr;
1137 static void appel_inter_block_aff(ir_node *bl, void *data)
1139 appel_clique_walker_t *env = data;
1140 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
1144 for(i = 0; i < bli->n_live_in; ++i) {
1145 ir_node *irn = bli->live_in[i];
1147 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1148 ir_node *pred = get_Block_cfgpred_block(bl, j);
1150 int nr = appel_get_live_end_nr(env, pred, irn);
1152 fprintf(env->f, "%d %d 1\n", bli->live_in_nr[i], nr);
1156 for(i = 0; i < bli->n_phi; ++i) {
1157 ir_node *irn = bli->phi[i];
1159 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1160 ir_node *pred = get_Block_cfgpred_block(bl, j);
1161 ir_node *op = get_irn_n(irn, j);
1163 int nr = appel_get_live_end_nr(env, pred, op);
1165 fprintf(env->f, "%d %d 1\n", bli->phi_nr[i], nr);
1171 void co_dump_appel_graph_cliques(const copy_opt_t *co, FILE *f)
1175 appel_clique_walker_t env;
1176 bitset_t *adm = bitset_alloca(co->cls->n_regs);
1177 be_lv_t *lv = co->cenv->birg->lv;
1179 be_liveness_recompute(lv);
1180 obstack_init(&env.obst);
1181 phase_init(&env.ph, "appel_clique_dumper", co->irg, PHASE_DEFAULT_GROWTH, appel_clique_walker_irn_init, NULL);
1182 env.curr_nr = co->cls->n_regs;
1186 bitset_copy(adm, co->cenv->ignore_colors);
1187 bitset_flip_all(adm);
1189 /* Make color map. */
1190 env.color_map = alloca(co->cls->n_regs * sizeof(env.color_map[0]));
1191 for(i = 0, n_colors = 0; i < co->cls->n_regs; ++i) {
1192 const arch_register_t *reg = &co->cls->regs[i];
1193 env.color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_colors++;
1197 env.curr_nr = n_colors;
1198 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1199 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1201 fprintf(f, "%d %d\n", env.curr_nr, n_colors);
1203 /* make the first k nodes interfere */
1204 for(i = 0; i < n_colors; ++i) {
1206 for(j = i + 1; j < n_colors; ++j)
1207 fprintf(f, "%d %d -1 ", i, j);
1212 env.curr_nr = n_colors;
1213 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1214 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1215 irg_block_walk_graph(co->irg, appel_inter_block_aff, NULL, &env);
1216 obstack_free(&env.obst, NULL);
1220 ___ _____ ____ ____ ___ _____ ____ _
1221 |_ _| ___/ ___| | _ \ / _ \_ _| | _ \ _ _ _ __ ___ _ __ (_)_ __ __ _
1222 | || |_ | | _ | | | | | | || | | | | | | | | '_ ` _ \| '_ \| | '_ \ / _` |
1223 | || _|| |_| | | |_| | |_| || | | |_| | |_| | | | | | | |_) | | | | | (_| |
1224 |___|_| \____| |____/ \___/ |_| |____/ \__,_|_| |_| |_| .__/|_|_| |_|\__, |
1228 static const char *get_dot_color_name(size_t col)
1230 static const char *names[] = {
1264 return col < sizeof(names)/sizeof(names[0]) ? names[col] : "white";
1267 typedef struct _co_ifg_dump_t {
1268 const copy_opt_t *co;
1272 static void ifg_dump_graph_attr(FILE *f, void *self)
1275 fprintf(f, "overlap=scale");
1278 static int ifg_is_dump_node(void *self, ir_node *irn)
1280 co_ifg_dump_t *cod = self;
1281 return !arch_irn_is(cod->co->aenv, irn, ignore);
1284 static void ifg_dump_node_attr(FILE *f, void *self, ir_node *irn)
1286 co_ifg_dump_t *env = self;
1287 const arch_register_t *reg = arch_get_irn_register(env->co->aenv, irn);
1288 const arch_register_req_t *req;
1291 req = arch_get_register_req(env->co->aenv, irn, BE_OUT_POS(0));
1292 limited = arch_register_req_is(req, limited);
1294 if(env->flags & CO_IFG_DUMP_LABELS) {
1295 ir_fprintf(f, "label=\"%+F", irn);
1299 if((env->flags & CO_IFG_DUMP_CONSTR) && limited) {
1300 bitset_t *bs = bitset_alloca(env->co->cls->n_regs);
1301 req.limited(req.limited_env, bs);
1302 ir_fprintf(f, "\\n%B", bs);
1305 ir_fprintf(f, "\" ");
1307 fprintf(f, "label=\"\" shape=point " );
1310 if(env->flags & CO_IFG_DUMP_SHAPE)
1311 fprintf(f, "shape=%s ", limited ? "diamond" : "ellipse");
1313 if(env->flags & CO_IFG_DUMP_COLORS)
1314 fprintf(f, "style=filled color=%s ", get_dot_color_name(reg->index));
1317 static void ifg_dump_at_end(FILE *file, void *self)
1319 co_ifg_dump_t *env = self;
1322 co_gs_foreach_aff_node(env->co, a) {
1323 const arch_register_t *ar = arch_get_irn_register(env->co->aenv, a->irn);
1324 unsigned aidx = get_irn_idx(a->irn);
1327 co_gs_foreach_neighb(a, n) {
1328 const arch_register_t *nr = arch_get_irn_register(env->co->aenv, n->irn);
1329 unsigned nidx = get_irn_idx(n->irn);
1332 const char *color = nr == ar ? "blue" : "red";
1333 fprintf(file, "\tn%d -- n%d [weight=0.01 ", aidx, nidx);
1334 if(env->flags & CO_IFG_DUMP_LABELS)
1335 fprintf(file, "label=\"%d\" ", n->costs);
1336 if(env->flags & CO_IFG_DUMP_COLORS)
1337 fprintf(file, "color=%s ", color);
1339 fprintf(file, "style=dotted");
1340 fprintf(file, "];\n");
1347 static be_ifg_dump_dot_cb_t ifg_dot_cb = {
1349 ifg_dump_graph_attr,
1358 void co_dump_ifg_dot(const copy_opt_t *co, FILE *f, unsigned flags)
1364 be_ifg_dump_dot(co->cenv->ifg, co->irg, f, &ifg_dot_cb, &cod);
1368 void co_solve_park_moon(copy_opt_t *opt)
1373 static int void_algo(copy_opt_t *co)
1381 / \ | | __ _ ___ _ __(_) |_| |__ _ __ ___ ___
1382 / _ \ | |/ _` |/ _ \| '__| | __| '_ \| '_ ` _ \/ __|
1383 / ___ \| | (_| | (_) | | | | |_| | | | | | | | \__ \
1384 /_/ \_\_|\__, |\___/|_| |_|\__|_| |_|_| |_| |_|___/
1391 int can_improve_existing;
1394 static co_algo_info_t algos[] = {
1395 { void_algo, "none", 0 },
1396 { co_solve_heuristic, "heur1", 0 },
1397 { co_solve_heuristic_new, "heur2", 0 },
1398 { co_solve_heuristic_java, "heur3", 0 },
1399 { co_solve_heuristic_mst, "heur4", 0 },
1401 { co_solve_ilp2, "ilp", 1 },
1408 | \/ | __ _(_)_ __ | _ \ _ __(_)_ _____ _ __
1409 | |\/| |/ _` | | '_ \ | | | | '__| \ \ / / _ \ '__|
1410 | | | | (_| | | | | | | |_| | | | |\ V / __/ |
1411 |_| |_|\__,_|_|_| |_| |____/|_| |_| \_/ \___|_|
1415 static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char *suffix)
1420 ir_snprintf(buf, sizeof(buf), "%s%F_%s%s", prefix, env->irg, env->cls->name, suffix);
1421 result = fopen(buf, "wt");
1422 if(result == NULL) {
1423 panic("Couldn't open '%s' for writing.", buf);
1429 void co_driver(be_chordal_env_t *cenv)
1431 lc_timer_t *timer = lc_timer_register("firm.be.copyopt", "runtime");
1432 co_complete_stats_t before, after;
1434 co_algo_t *algo_func;
1435 int was_optimal = 0;
1437 if (algo >= CO_ALGO_LAST)
1440 be_liveness_assure_chk(be_get_birg_liveness(cenv->birg));
1442 co = new_copy_opt(cenv, cost_func);
1443 co_build_ou_structure(co);
1444 co_build_graph_structure(co);
1446 co_complete_stats(co, &before);
1448 be_stat_ev_ull("co_aff_nodes", before.aff_nodes);
1449 be_stat_ev_ull("co_aff_edges", before.aff_edges);
1450 be_stat_ev_ull("co_max_costs", before.max_costs);
1451 be_stat_ev_ull("co_inevit_costs", before.inevit_costs);
1452 be_stat_ev_ull("co_aff_int", before.aff_int);
1454 be_stat_ev_ull("co_init_costs", before.costs);
1455 be_stat_ev_ull("co_init_unsat", before.unsatisfied_edges);
1457 /* Dump the interference graph in Appel's format. */
1458 if (dump_flags & DUMP_APPEL) {
1459 FILE *f = my_open(cenv, "", ".apl");
1460 co_dump_appel_graph(co, f);
1464 if (dump_flags & DUMP_BEFORE) {
1465 FILE *f = my_open(cenv, "", "-before.dot");
1466 co_dump_ifg_dot(co, f, style_flags);
1470 /* if the algo can improve results, provide an initial solution with heur3 */
1471 if (improve && algos[algo].can_improve_existing) {
1472 co_complete_stats_t stats;
1474 /* produce a heuristic solution */
1476 co_solve_heuristic_java(co);
1478 co_solve_heuristic(co);
1479 #endif /* WITH_JVM */
1481 /* do the stats and provide the current costs */
1482 co_complete_stats(co, &stats);
1483 be_stat_ev_ull("co_prepare_costs", stats.costs);
1487 /* start the JVM here so that it does not tamper the timing. */
1488 if (algo == CO_ALGO_HEUR3)
1489 be_java_coal_start_jvm();
1490 #endif /* WITH_JVM */
1492 algo_func = algos[algo].algo;
1494 /* perform actual copy minimization */
1495 lc_timer_reset_and_start(timer);
1496 was_optimal = algo_func(co);
1497 lc_timer_stop(timer);
1499 be_stat_ev("co_time", lc_timer_elapsed_msec(timer));
1500 be_stat_ev_ull("co_optimal", was_optimal);
1502 if (dump_flags & DUMP_AFTER) {
1503 FILE *f = my_open(cenv, "", "-after.dot");
1504 co_dump_ifg_dot(co, f, style_flags);
1508 co_complete_stats(co, &after);
1511 ulong64 optimizable_costs = after.max_costs - after.inevit_costs;
1512 ulong64 evitable = after.costs - after.inevit_costs;
1514 ir_printf("%30F ", cenv->irg);
1515 printf("%10s %10" ULL_FMT "%10" ULL_FMT "%10" ULL_FMT, cenv->cls->name, after.max_costs, before.costs, after.inevit_costs);
1517 if(optimizable_costs > 0)
1518 printf("%10" ULL_FMT " %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs);
1520 printf("%10" ULL_FMT " %5s\n", after.costs, "-");
1523 be_stat_ev_ull("co_after_costs", after.costs);
1524 be_stat_ev_ull("co_after_unsat", after.unsatisfied_edges);
1526 co_free_graph_structure(co);
1527 co_free_ou_structure(co);