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
26 * Main file for the optimization reducing the copies needed for:
28 * - Register-constrained nodes
29 * - Two-address code instructions
37 #include "raw_bitset.h"
43 #include "iredges_t.h"
44 #include "irprintf_t.h"
53 #include "beintlive_t.h"
54 #include "becopyopt_t.h"
55 #include "becopystat.h"
59 #include "bestatevent.h"
64 #include "lc_opts_enum.h"
69 #define DUMP_ALL 2 * DUMP_APPEL - 1
71 #define COST_FUNC_FREQ 1
72 #define COST_FUNC_LOOP 2
73 #define COST_FUNC_ALL_ONE 3
75 static unsigned dump_flags = 0;
76 static unsigned style_flags = 0;
77 static int do_stats = 0;
78 static cost_fct_t cost_func = co_get_costs_exec_freq;
79 static int improve = 1;
81 static const lc_opt_enum_mask_items_t dump_items[] = {
82 { "before", DUMP_BEFORE },
83 { "after", DUMP_AFTER },
84 { "appel", DUMP_APPEL },
90 * Flags for dumping the IFG.
93 CO_IFG_DUMP_COLORS = 1 << 0, /**< Dump the graph colored. */
94 CO_IFG_DUMP_LABELS = 1 << 1, /**< Dump node/edge labels. */
95 CO_IFG_DUMP_SHAPE = 1 << 2, /**< Give constrained nodes special shapes. */
96 CO_IFG_DUMP_CONSTR = 1 << 3, /**< Dump the node constraints in the label. */
99 static const lc_opt_enum_mask_items_t style_items[] = {
100 { "color", CO_IFG_DUMP_COLORS },
101 { "labels", CO_IFG_DUMP_LABELS },
102 { "constr", CO_IFG_DUMP_CONSTR },
103 { "shape", CO_IFG_DUMP_SHAPE },
104 { "full", 2 * CO_IFG_DUMP_SHAPE - 1 },
108 typedef int (*opt_funcptr)(void);
110 static const lc_opt_enum_func_ptr_items_t cost_func_items[] = {
111 { "freq", (opt_funcptr) co_get_costs_exec_freq },
112 { "loop", (opt_funcptr) co_get_costs_loop_depth },
113 { "one", (opt_funcptr) co_get_costs_all_one },
117 static lc_opt_enum_mask_var_t dump_var = {
118 &dump_flags, dump_items
121 static lc_opt_enum_mask_var_t style_var = {
122 &style_flags, style_items
125 static lc_opt_enum_func_ptr_var_t cost_func_var = {
126 (opt_funcptr*) &cost_func, cost_func_items
129 static const lc_opt_table_entry_t options[] = {
130 LC_OPT_ENT_ENUM_FUNC_PTR ("cost", "select a cost function", &cost_func_var),
131 LC_OPT_ENT_ENUM_MASK ("dump", "dump ifg before or after copy optimization", &dump_var),
132 LC_OPT_ENT_ENUM_MASK ("style", "dump style for ifg dumping", &style_var),
133 LC_OPT_ENT_BOOL ("stats", "dump statistics after each optimization", &do_stats),
134 LC_OPT_ENT_BOOL ("improve", "run heur1 before if algo can exploit start solutions", &improve),
138 static be_module_list_entry_t *copyopts = NULL;
139 static const co_algo_info *selected_copyopt = NULL;
141 void be_register_copyopt(const char *name, co_algo_info *copyopt)
143 if (selected_copyopt == NULL)
144 selected_copyopt = copyopt;
145 be_add_module_to_list(©opts, name, copyopt);
148 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copyopt)
149 void be_init_copyopt(void)
151 lc_opt_entry_t *be_grp = lc_opt_get_grp(firm_opt_get_root(), "be");
152 lc_opt_entry_t *ra_grp = lc_opt_get_grp(be_grp, "ra");
153 lc_opt_entry_t *chordal_grp = lc_opt_get_grp(ra_grp, "chordal");
154 lc_opt_entry_t *co_grp = lc_opt_get_grp(chordal_grp, "co");
156 lc_opt_add_table(co_grp, options);
157 be_add_module_list_opt(co_grp, "algo", "select copy optimization algo",
158 ©opts, (void**) &selected_copyopt);
161 static int void_algo(copy_opt_t *co)
167 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copynone)
168 void be_init_copynone(void)
170 static co_algo_info copyheur = {
174 be_register_copyopt("none", ©heur);
177 #undef QUICK_AND_DIRTY_HACK
179 static int nodes_interfere(const be_chordal_env_t *env, const ir_node *a, const ir_node *b)
182 return be_ifg_connected(env->ifg, a, b);
184 be_lv_t *lv = be_get_irg_liveness(env->irg);
185 return be_values_interfere(lv, a, b);
190 /******************************************************************************
193 | | __ ___ _ __ ___ _ __ __ _| |
194 | | |_ |/ _ \ '_ \ / _ \ '__/ _` | |
195 | |__| | __/ | | | __/ | | (_| | |
196 \_____|\___|_| |_|\___|_| \__,_|_|
198 ******************************************************************************/
200 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
203 copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs)
205 const char *s1, *s2, *s3;
209 FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
211 co = XMALLOCZ(copy_opt_t);
212 co->cenv = chordal_env;
213 co->irg = chordal_env->irg;
214 co->cls = chordal_env->cls;
215 co->get_costs = get_costs;
218 s2 = get_entity_name(get_irg_entity(co->irg));
219 s3 = chordal_env->cls->name;
220 len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
221 co->name = XMALLOCN(char, len);
222 snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
227 void free_copy_opt(copy_opt_t *co)
234 * Checks if a node is optimizable, viz. has something to do with coalescing
235 * @param irn The irn to check
237 static int co_is_optimizable_root(ir_node *irn)
239 const arch_register_req_t *req;
241 if (arch_irn_is_ignore(irn))
244 if (is_Reg_Phi(irn) || is_Perm_Proj(irn))
247 req = arch_get_irn_register_req(irn);
248 if (is_2addr_code(req))
254 int co_get_costs_loop_depth(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos)
258 ir_node *root_block = get_nodes_block(root);
263 /* for phis the copies are placed in the corresponding pred-block */
264 loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos));
266 /* a perm places the copy in the same block as it resides */
267 loop = get_irn_loop(root_block);
270 int d = get_loop_depth(loop);
276 int co_get_costs_exec_freq(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos)
279 ir_node *root_bl = get_nodes_block(root);
280 ir_node *copy_bl = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl;
281 ir_exec_freq *exec_freq = be_get_irg_exec_freq(co->cenv->irg);
283 res = get_block_execfreq_ulong(exec_freq, copy_bl);
285 /* don't allow values smaller than one. */
286 return res < 1 ? 1 : res;
290 int co_get_costs_all_one(const copy_opt_t *co, ir_node *root, ir_node *arg, int pos)
299 /******************************************************************************
300 ____ _ _ _ _ _ _____ _
301 / __ \ | | | | | | (_) | / ____| |
302 | | | |_ __ | |_| | | |_ __ _| |_ ___ | (___ | |_ ___ _ __ __ _ __ _ ___
303 | | | | '_ \| __| | | | '_ \| | __/ __| \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
304 | |__| | |_) | |_| |__| | | | | | |_\__ \ ____) | || (_) | | | (_| | (_| | __/
305 \____/| .__/ \__|\____/|_| |_|_|\__|___/ |_____/ \__\___/|_| \__,_|\__, |\___|
308 ******************************************************************************/
311 * Determines a maximum weighted independent set with respect to
312 * the interference and conflict edges of all nodes in a qnode.
314 static int ou_max_ind_set_costs(unit_t *ou)
316 be_chordal_env_t *chordal_env = ou->co->cenv;
317 ir_node **safe, **unsafe;
318 int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
321 int curr_weight, best_weight = 0;
323 /* assign the nodes into two groups.
324 * safe: node has no interference, hence it is in every max stable set.
325 * unsafe: node has an interference
327 safe = ALLOCAN(ir_node*, ou->node_count - 1);
330 unsafe = ALLOCAN(ir_node*, ou->node_count - 1);
331 unsafe_costs = ALLOCAN(int, ou->node_count - 1);
333 for (i=1; i<ou->node_count; ++i) {
335 for (o=1; o<ou->node_count; ++o) {
338 if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) {
339 unsafe_costs[unsafe_count] = ou->costs[i];
340 unsafe[unsafe_count] = ou->nodes[i];
347 safe_costs += ou->costs[i];
348 safe[safe_count++] = ou->nodes[i];
353 /* now compute the best set out of the unsafe nodes*/
354 if (unsafe_count > MIS_HEUR_TRIGGER) {
355 bitset_t *best = bitset_alloca(unsafe_count);
356 /* Heuristic: Greedy trial and error form index 0 to unsafe_count-1 */
357 for (i=0; i<unsafe_count; ++i) {
359 /* check if it is a stable set */
360 for (o=bitset_next_set(best, 0); o!=-1 && o<i; o=bitset_next_set(best, o+1))
361 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o])) {
362 bitset_clear(best, i); /* clear the bit and try next one */
366 /* compute the weight */
367 bitset_foreach(best, pos)
368 best_weight += unsafe_costs[pos];
370 /* Exact Algorithm: Brute force */
371 curr = bitset_alloca(unsafe_count);
372 bitset_set_all(curr);
373 while (bitset_popcount(curr) != 0) {
374 /* check if curr is a stable set */
375 for (i=bitset_next_set(curr, 0); i!=-1; i=bitset_next_set(curr, i+1))
376 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) */
377 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o]))
380 /* if we arrive here, we have a stable set */
381 /* compute the weight of the stable set*/
383 bitset_foreach(curr, pos)
384 curr_weight += unsafe_costs[pos];
387 if (curr_weight > best_weight) {
388 best_weight = curr_weight;
396 return safe_costs+best_weight;
399 static void co_collect_units(ir_node *irn, void *env)
401 const arch_register_req_t *req;
402 copy_opt_t *co = (copy_opt_t*)env;
405 if (get_irn_mode(irn) == mode_T)
407 req = arch_get_irn_register_req(irn);
408 if (req->cls != co->cls)
410 if (!co_is_optimizable_root(irn))
413 /* Init a new unit */
414 unit = XMALLOCZ(unit_t);
416 unit->node_count = 1;
417 INIT_LIST_HEAD(&unit->queue);
419 /* Phi with some/all of its arguments */
420 if (is_Reg_Phi(irn)) {
424 arity = get_irn_arity(irn);
425 unit->nodes = XMALLOCN(ir_node*, arity + 1);
426 unit->costs = XMALLOCN(int, arity + 1);
427 unit->nodes[0] = irn;
430 for (i=0; i<arity; ++i) {
432 ir_node *arg = get_irn_n(irn, i);
434 assert(arch_get_irn_reg_class(arg) == co->cls && "Argument not in same register class.");
437 if (nodes_interfere(co->cenv, irn, arg)) {
438 unit->inevitable_costs += co->get_costs(co, irn, arg, i);
442 /* Else insert the argument of the phi to the members of this ou */
443 DBG((dbg, LEVEL_1, "\t Member: %+F\n", arg));
445 if (arch_irn_is_ignore(arg))
448 /* Check if arg has occurred at a prior position in the arg/list */
450 for (o=1; o<unit->node_count; ++o) {
451 if (unit->nodes[o] == arg) {
457 if (!arg_pos) { /* a new argument */
458 /* insert node, set costs */
459 unit->nodes[unit->node_count] = arg;
460 unit->costs[unit->node_count] = co->get_costs(co, irn, arg, i);
462 } else { /* arg has occurred before in same phi */
463 /* increase costs for existing arg */
464 unit->costs[arg_pos] += co->get_costs(co, irn, arg, i);
467 unit->nodes = XREALLOC(unit->nodes, ir_node*, unit->node_count);
468 unit->costs = XREALLOC(unit->costs, int, unit->node_count);
469 } else if (is_Perm_Proj(irn)) {
470 /* Proj of a perm with corresponding arg */
471 assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
472 unit->nodes = XMALLOCN(ir_node*, 2);
473 unit->costs = XMALLOCN(int, 2);
474 unit->node_count = 2;
475 unit->nodes[0] = irn;
476 unit->nodes[1] = get_Perm_src(irn);
477 unit->costs[1] = co->get_costs(co, irn, unit->nodes[1], -1);
479 /* Src == Tgt of a 2-addr-code instruction */
480 if (is_2addr_code(req)) {
481 const unsigned other = req->other_same;
485 for (i = 0; (1U << i) <= other; ++i) {
486 if (other & (1U << i)) {
487 ir_node *o = get_irn_n(skip_Proj(irn), i);
488 if (arch_irn_is_ignore(o))
490 if (nodes_interfere(co->cenv, irn, o))
499 unit->nodes = XMALLOCN(ir_node*, count);
500 unit->costs = XMALLOCN(int, count);
501 unit->node_count = count;
502 unit->nodes[k++] = irn;
504 for (i = 0; 1U << i <= other; ++i) {
505 if (other & (1U << i)) {
506 ir_node *o = get_irn_n(skip_Proj(irn), i);
507 if (!arch_irn_is_ignore(o) &&
508 !nodes_interfere(co->cenv, irn, o)) {
510 unit->costs[k] = co->get_costs(co, irn, o, -1);
517 assert(0 && "This is not an optimizable node!");
521 /* Insert the new unit at a position according to its costs */
522 if (unit->node_count > 1) {
524 struct list_head *tmp;
526 /* Determine the maximum costs this unit can cause: all_nodes_cost */
527 for (i=1; i<unit->node_count; ++i) {
528 unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
529 unit->all_nodes_costs += unit->costs[i];
532 /* Determine the minimal costs this unit will cause: min_nodes_costs */
533 unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
534 /* Insert the new ou according to its sort_key */
536 while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
538 list_add(&unit->units, tmp);
544 #ifdef QUICK_AND_DIRTY_HACK
546 static int compare_ous(const void *k1, const void *k2)
548 const unit_t *u1 = *((const unit_t **) k1);
549 const unit_t *u2 = *((const unit_t **) k2);
550 int i, o, u1_has_constr, u2_has_constr;
551 arch_register_req_t req;
553 /* Units with constraints come first */
555 for (i=0; i<u1->node_count; ++i) {
556 arch_get_irn_register_req(&req, u1->nodes[i]);
557 if (arch_register_req_is(&req, limited)) {
564 for (i=0; i<u2->node_count; ++i) {
565 arch_get_irn_register_req(&req, u2->nodes[i]);
566 if (arch_register_req_is(&req, limited)) {
572 if (u1_has_constr != u2_has_constr)
573 return u2_has_constr - u1_has_constr;
575 /* Now check, whether the two units are connected */
577 for (i=0; i<u1->node_count; ++i)
578 for (o=0; o<u2->node_count; ++o)
579 if (u1->nodes[i] == u2->nodes[o])
583 /* After all, the sort key decides. Greater keys come first. */
584 return u2->sort_key - u1->sort_key;
589 * Sort the ou's according to constraints and their sort_key
591 static void co_sort_units(copy_opt_t *co)
593 int i, count = 0, costs;
596 /* get the number of ous, remove them form the list and fill the array */
597 list_for_each_entry(unit_t, ou, &co->units, units)
599 ous = ALLOCAN(unit_t, count);
601 costs = co_get_max_copy_costs(co);
604 list_for_each_entry(unit_t, ou, &co->units, units)
607 INIT_LIST_HEAD(&co->units);
609 assert(count == i && list_empty(&co->units));
611 for (i=0; i<count; ++i)
612 ir_printf("%+F\n", ous[i]->nodes[0]);
614 qsort(ous, count, sizeof(*ous), compare_ous);
617 for (i=0; i<count; ++i)
618 ir_printf("%+F\n", ous[i]->nodes[0]);
620 /* reinsert into list in correct order */
621 for (i=0; i<count; ++i)
622 list_add_tail(&ous[i]->units, &co->units);
624 assert(costs == co_get_max_copy_costs(co));
628 void co_build_ou_structure(copy_opt_t *co)
630 DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
631 INIT_LIST_HEAD(&co->units);
632 irg_walk_graph(co->irg, co_collect_units, NULL, co);
633 #ifdef QUICK_AND_DIRTY_HACK
638 void co_free_ou_structure(copy_opt_t *co)
642 list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
647 co->units.next = NULL;
650 /* co_solve_heuristic() is implemented in becopyheur.c */
652 int co_get_max_copy_costs(const copy_opt_t *co)
659 list_for_each_entry(unit_t, curr, &co->units, units) {
660 res += curr->inevitable_costs;
661 for (i=1; i<curr->node_count; ++i)
662 res += curr->costs[i];
667 int co_get_inevit_copy_costs(const copy_opt_t *co)
674 list_for_each_entry(unit_t, curr, &co->units, units)
675 res += curr->inevitable_costs;
679 int co_get_copy_costs(const copy_opt_t *co)
686 list_for_each_entry(unit_t, curr, &co->units, units) {
687 int root_col = get_irn_col(curr->nodes[0]);
688 DBG((dbg, LEVEL_1, " %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
689 res += curr->inevitable_costs;
690 for (i=1; i<curr->node_count; ++i) {
691 int arg_col = get_irn_col(curr->nodes[i]);
692 if (root_col != arg_col) {
693 DBG((dbg, LEVEL_1, " %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
694 res += curr->costs[i];
701 int co_get_lower_bound(const copy_opt_t *co)
708 list_for_each_entry(unit_t, curr, &co->units, units)
709 res += curr->inevitable_costs + curr->min_nodes_costs;
713 void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat)
715 bitset_t *seen = bitset_malloc(get_irg_last_idx(co->irg));
718 memset(stat, 0, sizeof(stat[0]));
720 /* count affinity edges. */
721 co_gs_foreach_aff_node(co, an) {
723 stat->aff_nodes += 1;
724 bitset_set(seen, get_irn_idx(an->irn));
725 co_gs_foreach_neighb(an, neigh) {
726 if (!bitset_is_set(seen, get_irn_idx(neigh->irn))) {
727 stat->aff_edges += 1;
728 stat->max_costs += neigh->costs;
730 if (get_irn_col(an->irn) != get_irn_col(neigh->irn)) {
731 stat->costs += neigh->costs;
732 stat->unsatisfied_edges += 1;
735 if (nodes_interfere(co->cenv, an->irn, neigh->irn)) {
737 stat->inevit_costs += neigh->costs;
747 /******************************************************************************
749 / ____| | | / ____| |
750 | | __ _ __ __ _ _ __ | |__ | (___ | |_ ___ _ __ __ _ __ _ ___
751 | | |_ | '__/ _` | '_ \| '_ \ \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
752 | |__| | | | (_| | |_) | | | | ____) | || (_) | | | (_| | (_| | __/
753 \_____|_| \__,_| .__/|_| |_| |_____/ \__\___/|_| \__,_|\__, |\___|
756 ******************************************************************************/
758 static int compare_affinity_node_t(const void *k1, const void *k2, size_t size)
760 const affinity_node_t *n1 = (const affinity_node_t*)k1;
761 const affinity_node_t *n2 = (const affinity_node_t*)k2;
764 return (n1->irn != n2->irn);
767 static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs)
769 affinity_node_t new_node, *node;
775 new_node.neighbours = NULL;
776 node = (affinity_node_t*)set_insert(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
778 for (nbr = node->neighbours; nbr; nbr = nbr->next)
779 if (nbr->irn == n2) {
784 /* if we did not find n2 in n1's neighbourhood insert it */
786 nbr = OALLOC(&co->obst, neighb_t);
789 nbr->next = node->neighbours;
791 node->neighbours = nbr;
795 /* now nbr points to n1's neighbour-entry of n2 */
799 static inline void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs)
801 if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
802 add_edge(co, n1, n2, costs);
803 add_edge(co, n2, n1, costs);
807 static void build_graph_walker(ir_node *irn, void *env)
809 const arch_register_req_t *req;
810 copy_opt_t *co = (copy_opt_t*)env;
813 if (get_irn_mode(irn) == mode_T)
815 req = arch_get_irn_register_req(irn);
816 if (req->cls != co->cls || arch_irn_is_ignore(irn))
819 if (is_Reg_Phi(irn)) { /* Phis */
820 for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
821 ir_node *arg = get_irn_n(irn, pos);
822 add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
824 } else if (is_Perm_Proj(irn)) { /* Perms */
825 ir_node *arg = get_Perm_src(irn);
826 add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
827 } else { /* 2-address code */
828 if (is_2addr_code(req)) {
829 const unsigned other = req->other_same;
832 for (i = 0; 1U << i <= other; ++i) {
833 if (other & (1U << i)) {
834 ir_node *other = get_irn_n(skip_Proj(irn), i);
835 if (!arch_irn_is_ignore(other))
836 add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
843 void co_build_graph_structure(copy_opt_t *co)
845 obstack_init(&co->obst);
846 co->nodes = new_set(compare_affinity_node_t, 32);
848 irg_walk_graph(co->irg, build_graph_walker, NULL, co);
851 void co_free_graph_structure(copy_opt_t *co)
856 obstack_free(&co->obst, NULL);
860 int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn)
862 affinity_node_t new_node, *n;
867 n = (affinity_node_t*)set_find(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
869 return (n->degree > 0);
874 static int co_dump_appel_disjoint_constraints(const copy_opt_t *co, ir_node *a, ir_node *b)
876 ir_node *nodes[] = { a, b };
877 bitset_t *constr[] = { NULL, NULL };
880 constr[0] = bitset_alloca(co->cls->n_regs);
881 constr[1] = bitset_alloca(co->cls->n_regs);
883 for (j = 0; j < 2; ++j) {
884 const arch_register_req_t *req = arch_get_irn_register_req(nodes[j]);
885 if (arch_register_req_is(req, limited))
886 rbitset_copy_to_bitset(req->limited, constr[j]);
888 bitset_set_all(constr[j]);
892 return !bitset_intersect(constr[0], constr[1]);
895 void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
897 be_ifg_t *ifg = co->cenv->ifg;
898 int *color_map = ALLOCAN(int, co->cls->n_regs);
899 int *node_map = XMALLOCN(int, get_irg_last_idx(co->irg) + 1);
900 ir_graph *irg = co->irg;
901 be_irg_t *birg = be_birg_from_irg(irg);
905 neighbours_iter_t nit;
910 for (i = 0; i < co->cls->n_regs; ++i) {
911 const arch_register_t *reg = &co->cls->regs[i];
912 if (rbitset_is_set(birg->allocatable_regs, reg->global_index)) {
913 color_map[i] = n_regs++;
920 * n contains the first node number.
921 * the values below n are the pre-colored register nodes
925 be_ifg_foreach_node(ifg, &it, irn) {
926 if (arch_irn_is_ignore(irn))
928 node_map[get_irn_idx(irn)] = n++;
931 fprintf(f, "%d %d\n", n, n_regs);
933 be_ifg_foreach_node(ifg, &it, irn) {
934 if (!arch_irn_is_ignore(irn)) {
935 int idx = node_map[get_irn_idx(irn)];
936 affinity_node_t *a = get_affinity_info(co, irn);
937 const arch_register_req_t *req = arch_get_irn_register_req(irn);
940 if (arch_register_req_is(req, limited)) {
941 for (i = 0; i < co->cls->n_regs; ++i) {
942 if (!rbitset_is_set(req->limited, i) && color_map[i] >= 0)
943 fprintf(f, "%d %d -1\n", color_map[i], idx);
947 be_ifg_foreach_neighbour(ifg, &nit, irn, adj) {
948 if (!arch_irn_is_ignore(adj) &&
949 !co_dump_appel_disjoint_constraints(co, irn, adj)) {
950 int adj_idx = node_map[get_irn_idx(adj)];
952 fprintf(f, "%d %d -1\n", idx, adj_idx);
959 co_gs_foreach_neighb(a, n) {
960 if (!arch_irn_is_ignore(n->irn)) {
961 int n_idx = node_map[get_irn_idx(n->irn)];
963 fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs);
974 ___ _____ ____ ____ ___ _____ ____ _
975 |_ _| ___/ ___| | _ \ / _ \_ _| | _ \ _ _ _ __ ___ _ __ (_)_ __ __ _
976 | || |_ | | _ | | | | | | || | | | | | | | | '_ ` _ \| '_ \| | '_ \ / _` |
977 | || _|| |_| | | |_| | |_| || | | |_| | |_| | | | | | | |_) | | | | | (_| |
978 |___|_| \____| |____/ \___/ |_| |____/ \__,_|_| |_| |_| .__/|_|_| |_|\__, |
982 static const char *get_dot_color_name(size_t col)
984 static const char *names[] = {
1018 return col < sizeof(names)/sizeof(names[0]) ? names[col] : "white";
1021 typedef struct co_ifg_dump_t {
1022 const copy_opt_t *co;
1026 static void ifg_dump_graph_attr(FILE *f, void *self)
1029 fprintf(f, "overlap=scale");
1032 static int ifg_is_dump_node(void *self, ir_node *irn)
1035 return !arch_irn_is_ignore(irn);
1038 static void ifg_dump_node_attr(FILE *f, void *self, ir_node *irn)
1040 co_ifg_dump_t *env = (co_ifg_dump_t*)self;
1041 const arch_register_t *reg = arch_get_irn_register(irn);
1042 const arch_register_req_t *req = arch_get_irn_register_req(irn);
1043 int limited = arch_register_req_is(req, limited);
1045 if (env->flags & CO_IFG_DUMP_LABELS) {
1046 ir_fprintf(f, "label=\"%+F", irn);
1048 if ((env->flags & CO_IFG_DUMP_CONSTR) && limited) {
1049 bitset_t *bs = bitset_alloca(env->co->cls->n_regs);
1050 rbitset_copy_to_bitset(req->limited, bs);
1051 ir_fprintf(f, "\\n%B", bs);
1053 ir_fprintf(f, "\" ");
1055 fprintf(f, "label=\"\" shape=point " );
1058 if (env->flags & CO_IFG_DUMP_SHAPE)
1059 fprintf(f, "shape=%s ", limited ? "diamond" : "ellipse");
1061 if (env->flags & CO_IFG_DUMP_COLORS)
1062 fprintf(f, "style=filled color=%s ", get_dot_color_name(reg->index));
1065 static void ifg_dump_at_end(FILE *file, void *self)
1067 co_ifg_dump_t *env = (co_ifg_dump_t*)self;
1070 co_gs_foreach_aff_node(env->co, a) {
1071 const arch_register_t *ar = arch_get_irn_register(a->irn);
1072 unsigned aidx = get_irn_idx(a->irn);
1075 co_gs_foreach_neighb(a, n) {
1076 const arch_register_t *nr = arch_get_irn_register(n->irn);
1077 unsigned nidx = get_irn_idx(n->irn);
1080 const char *color = nr == ar ? "blue" : "red";
1081 fprintf(file, "\tn%u -- n%u [weight=0.01 ", aidx, nidx);
1082 if (env->flags & CO_IFG_DUMP_LABELS)
1083 fprintf(file, "label=\"%d\" ", n->costs);
1084 if (env->flags & CO_IFG_DUMP_COLORS)
1085 fprintf(file, "color=%s ", color);
1087 fprintf(file, "style=dotted");
1088 fprintf(file, "];\n");
1095 static be_ifg_dump_dot_cb_t ifg_dot_cb = {
1097 ifg_dump_graph_attr,
1106 void co_dump_ifg_dot(const copy_opt_t *co, FILE *f, unsigned flags)
1112 be_ifg_dump_dot(co->cenv->ifg, co->irg, f, &ifg_dot_cb, &cod);
1117 | \/ | __ _(_)_ __ | _ \ _ __(_)_ _____ _ __
1118 | |\/| |/ _` | | '_ \ | | | | '__| \ \ / / _ \ '__|
1119 | | | | (_| | | | | | | |_| | | | |\ V / __/ |
1120 |_| |_|\__,_|_|_| |_| |____/|_| |_| \_/ \___|_|
1124 static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char *suffix)
1130 const char *cup_name = be_get_irg_main_env(env->irg)->cup_name;
1132 n = strlen(cup_name);
1133 tu_name = XMALLOCN(char, n + 1);
1134 strcpy(tu_name, cup_name);
1135 for (i = 0; i < n; ++i)
1136 if (tu_name[i] == '.')
1140 ir_snprintf(buf, sizeof(buf), "%s%s_%F_%s%s", prefix, tu_name, env->irg, env->cls->name, suffix);
1142 result = fopen(buf, "wt");
1143 if (result == NULL) {
1144 panic("Couldn't open '%s' for writing.", buf);
1150 void co_driver(be_chordal_env_t *cenv)
1152 ir_timer_t *timer = ir_timer_new();
1153 co_complete_stats_t before, after;
1155 int was_optimal = 0;
1157 assert(selected_copyopt);
1159 /* skip copymin if algo is 'none' */
1160 if (selected_copyopt->copyopt == void_algo)
1163 be_liveness_assure_chk(be_get_irg_liveness(cenv->irg));
1165 co = new_copy_opt(cenv, cost_func);
1166 co_build_ou_structure(co);
1167 co_build_graph_structure(co);
1169 co_complete_stats(co, &before);
1171 be_stat_ev_ull("co_aff_nodes", before.aff_nodes);
1172 be_stat_ev_ull("co_aff_edges", before.aff_edges);
1173 be_stat_ev_ull("co_max_costs", before.max_costs);
1174 be_stat_ev_ull("co_inevit_costs", before.inevit_costs);
1175 be_stat_ev_ull("co_aff_int", before.aff_int);
1177 be_stat_ev_ull("co_init_costs", before.costs);
1178 be_stat_ev_ull("co_init_unsat", before.unsatisfied_edges);
1180 if (dump_flags & DUMP_BEFORE) {
1181 FILE *f = my_open(cenv, "", "-before.dot");
1182 co_dump_ifg_dot(co, f, style_flags);
1186 /* if the algo can improve results, provide an initial solution with heur1 */
1187 if (improve && selected_copyopt->can_improve_existing) {
1188 co_complete_stats_t stats;
1190 /* produce a heuristic solution */
1191 co_solve_heuristic(co);
1193 /* do the stats and provide the current costs */
1194 co_complete_stats(co, &stats);
1195 be_stat_ev_ull("co_prepare_costs", stats.costs);
1198 /* perform actual copy minimization */
1199 ir_timer_reset_and_start(timer);
1200 was_optimal = selected_copyopt->copyopt(co);
1201 ir_timer_stop(timer);
1203 be_stat_ev("co_time", ir_timer_elapsed_msec(timer));
1204 be_stat_ev_ull("co_optimal", was_optimal);
1205 ir_timer_free(timer);
1207 if (dump_flags & DUMP_AFTER) {
1208 FILE *f = my_open(cenv, "", "-after.dot");
1209 co_dump_ifg_dot(co, f, style_flags);
1213 co_complete_stats(co, &after);
1216 unsigned long long optimizable_costs = after.max_costs - after.inevit_costs;
1217 unsigned long long evitable = after.costs - after.inevit_costs;
1219 ir_printf("%30F ", cenv->irg);
1220 printf("%10s %10llu%10llu%10llu", cenv->cls->name, after.max_costs, before.costs, after.inevit_costs);
1222 if (optimizable_costs > 0)
1223 printf("%10llu %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs);
1225 printf("%10llu %5s\n", after.costs, "-");
1228 /* Dump the interference graph in Appel's format. */
1229 if (dump_flags & DUMP_APPEL) {
1230 FILE *f = my_open(cenv, "", ".apl");
1231 fprintf(f, "# %llu %llu\n", after.costs, after.unsatisfied_edges);
1232 co_dump_appel_graph(co, f);
1236 be_stat_ev_ull("co_after_costs", after.costs);
1237 be_stat_ev_ull("co_after_unsat", after.unsatisfied_edges);
1239 co_free_graph_structure(co);
1240 co_free_ou_structure(co);