4 * Copyright: (c) Universitaet Karlsruhe
5 * Licence: This file protected by GPL - GNU GENERAL PUBLIC LICENSE.
25 #include "iredges_t.h"
28 #include "irphase_t.h"
29 #include "irprintf_t.h"
36 #include "becopyopt_t.h"
37 #include "becopystat.h"
40 #include "besched_t.h"
45 #define DUMP_ALL 2 * DUMP_APPEL - 1
47 #define COST_FUNC_FREQ 1
48 #define COST_FUNC_LOOP 2
49 #define COST_FUNC_ALL_ONE 3
51 static int dump_flags = 0;
52 static int style_flags = 0;
53 static int do_stats = 0;
54 static cost_fct_t cost_func = co_get_costs_exec_freq;
55 static int algo = CO_ALGO_HEUR2;
58 static const lc_opt_enum_mask_items_t dump_items[] = {
59 { "before", DUMP_BEFORE },
60 { "after", DUMP_AFTER },
61 { "appel", DUMP_APPEL },
66 static const lc_opt_enum_mask_items_t style_items[] = {
67 { "color", CO_IFG_DUMP_COLORS },
68 { "labels", CO_IFG_DUMP_LABELS },
69 { "constr", CO_IFG_DUMP_CONSTR },
70 { "shape", CO_IFG_DUMP_SHAPE },
71 { "full", 2 * CO_IFG_DUMP_SHAPE - 1 },
75 static const lc_opt_enum_mask_items_t algo_items[] = {
76 { "none", CO_ALGO_NONE },
77 { "heur", CO_ALGO_HEUR },
78 { "heur2", CO_ALGO_HEUR2 },
79 { "heur3", CO_ALGO_HEUR3 },
80 { "ilp", CO_ALGO_ILP },
84 static const lc_opt_enum_func_ptr_items_t cost_func_items[] = {
85 { "freq", co_get_costs_exec_freq },
86 { "loop", co_get_costs_loop_depth },
87 { "one", co_get_costs_all_one },
91 static lc_opt_enum_mask_var_t dump_var = {
92 &dump_flags, dump_items
95 static lc_opt_enum_mask_var_t style_var = {
96 &style_flags, style_items
99 static lc_opt_enum_mask_var_t algo_var = {
103 static lc_opt_enum_func_ptr_var_t cost_func_var = {
104 &cost_func, cost_func_items
107 static const lc_opt_table_entry_t options[] = {
108 LC_OPT_ENT_ENUM_INT ("algo", "select copy optimization algo (heur, heur2, heur3, ilp)", &algo_var),
109 LC_OPT_ENT_ENUM_FUNC_PTR ("cost", "select a cost function (freq, loop, one)", &cost_func_var),
110 LC_OPT_ENT_ENUM_MASK ("dump", "dump ifg before or after copy optimization", &dump_var),
111 LC_OPT_ENT_ENUM_MASK ("style", "dump style for ifg dumping", &style_var),
112 LC_OPT_ENT_BOOL ("stats", "dump statistics after each optimization", &do_stats),
116 /* Insert additional options registration functions here. */
117 extern void be_co_ilp_register_options(lc_opt_entry_t *grp);
118 extern void be_co2_register_options(lc_opt_entry_t *grp);
119 extern void be_co3_register_options(lc_opt_entry_t *grp);
121 void co_register_options(lc_opt_entry_t *grp)
123 lc_opt_entry_t *co_grp = lc_opt_get_grp(grp, "co");
124 lc_opt_add_table(co_grp, options);
126 be_co2_register_options(co_grp);
127 be_co3_register_options(co_grp);
129 be_co_ilp_register_options(co_grp);
135 #undef QUICK_AND_DIRTY_HACK
137 /******************************************************************************
140 | | __ ___ _ __ ___ _ __ __ _| |
141 | | |_ |/ _ \ '_ \ / _ \ '__/ _` | |
142 | |__| | __/ | | | __/ | | (_| | |
143 \_____|\___|_| |_|\___|_| \__,_|_|
145 ******************************************************************************/
147 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
149 void be_copy_opt_init(void) {
152 copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs)
154 const char *s1, *s2, *s3;
158 FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
160 co = xcalloc(1, sizeof(*co));
161 co->cenv = chordal_env;
162 co->aenv = chordal_env->birg->main_env->arch_env;
163 co->irg = chordal_env->irg;
164 co->cls = chordal_env->cls;
165 co->get_costs = get_costs;
167 s1 = get_irp_prog_name();
168 s2 = get_entity_name(get_irg_entity(co->irg));
169 s3 = chordal_env->cls->name;
170 len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
171 co->name = xmalloc(len);
172 snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
177 void free_copy_opt(copy_opt_t *co) {
182 int co_is_optimizable_root(const copy_opt_t *co, ir_node *irn) {
183 arch_register_req_t req;
184 const arch_register_t *reg;
186 if (arch_irn_is(co->aenv, irn, ignore))
189 reg = arch_get_irn_register(co->aenv, irn);
190 if (arch_register_type_is(reg, ignore))
193 if (is_Reg_Phi(irn) || is_Perm_Proj(co->aenv, irn) || is_2addr_code(co->aenv, irn, &req))
199 int co_is_optimizable_arg(const copy_opt_t *co, ir_node *irn) {
200 const ir_edge_t *edge;
201 const arch_register_t *reg;
203 assert(0 && "Is buggy and obsolete. Do not use");
205 if (arch_irn_is(co->aenv, irn, ignore))
208 reg = arch_get_irn_register(co->aenv, irn);
209 if (arch_register_type_is(reg, ignore))
212 foreach_out_edge(irn, edge) {
213 ir_node *n = edge->src;
215 if (!nodes_interfere(co->cenv, irn, n) || irn == n) {
216 arch_register_req_t req;
217 arch_get_register_req(co->aenv, &req, n, -1);
220 is_Perm(co->aenv, n) ||
221 (arch_register_req_is(&req, should_be_same) && req.other_same == irn)
230 int co_get_costs_loop_depth(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
233 ir_node *root_block = get_nodes_block(root);
236 /* for phis the copies are placed in the corresponding pred-block */
237 loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos));
239 /* a perm places the copy in the same block as it resides */
240 loop = get_irn_loop(root_block);
243 int d = get_loop_depth(loop);
249 int co_get_costs_exec_freq(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
250 ir_node *root_bl = get_nodes_block(root);
251 ir_node *copy_bl = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl;
252 unsigned long freq = get_block_execfreq_ulong(co->cenv->exec_freq, copy_bl);
253 return freq > 0 ? (int) freq : 1;
257 int co_get_costs_all_one(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) {
261 /******************************************************************************
262 ____ _ _ _ _ _ _____ _
263 / __ \ | | | | | | (_) | / ____| |
264 | | | |_ __ | |_| | | |_ __ _| |_ ___ | (___ | |_ ___ _ __ __ _ __ _ ___
265 | | | | '_ \| __| | | | '_ \| | __/ __| \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
266 | |__| | |_) | |_| |__| | | | | | |_\__ \ ____) | || (_) | | | (_| | (_| | __/
267 \____/| .__/ \__|\____/|_| |_|_|\__|___/ |_____/ \__\___/|_| \__,_|\__, |\___|
270 ******************************************************************************/
273 * Determines a maximum weighted independent set with respect to
274 * the interference and conflict edges of all nodes in a qnode.
276 static int ou_max_ind_set_costs(unit_t *ou) {
277 be_chordal_env_t *chordal_env = ou->co->cenv;
278 ir_node **safe, **unsafe;
279 int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
281 int max, pos, curr_weight, best_weight = 0;
283 /* assign the nodes into two groups.
284 * safe: node has no interference, hence it is in every max stable set.
285 * unsafe: node has an interference
287 safe = alloca((ou->node_count-1) * sizeof(*safe));
290 unsafe = alloca((ou->node_count-1) * sizeof(*unsafe));
291 unsafe_costs = alloca((ou->node_count-1) * sizeof(*unsafe_costs));
293 for(i=1; i<ou->node_count; ++i) {
295 for(o=1; o<ou->node_count; ++o) {
298 if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) {
299 unsafe_costs[unsafe_count] = ou->costs[i];
300 unsafe[unsafe_count] = ou->nodes[i];
307 safe_costs += ou->costs[i];
308 safe[safe_count++] = ou->nodes[i];
313 /* now compute the best set out of the unsafe nodes*/
314 if (unsafe_count > MIS_HEUR_TRIGGER) {
315 bitset_t *best = bitset_alloca(unsafe_count);
316 /* Heuristik: Greedy trial and error form index 0 to unsafe_count-1 */
317 for (i=0; i<unsafe_count; ++i) {
319 /* check if it is a stable set */
320 for (o=bitset_next_set(best, 0); o!=-1 && o<i; o=bitset_next_set(best, o+1))
321 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o])) {
322 bitset_clear(best, i); /* clear the bit and try next one */
326 /* compute the weight */
327 bitset_foreach(best, pos)
328 best_weight += unsafe_costs[pos];
330 /* Exact Algorithm: Brute force */
331 curr = bitset_alloca(unsafe_count);
332 bitset_set_all(curr);
333 while ((max = bitset_popcnt(curr)) != 0) {
334 /* check if curr is a stable set */
335 for (i=bitset_next_set(curr, 0); i!=-1; i=bitset_next_set(curr, i+1))
336 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) */
337 if (nodes_interfere(chordal_env, unsafe[i], unsafe[o]))
340 /* if we arrive here, we have a stable set */
341 /* compute the weigth of the stable set*/
343 bitset_foreach(curr, pos)
344 curr_weight += unsafe_costs[pos];
347 if (curr_weight > best_weight) {
348 best_weight = curr_weight;
356 return safe_costs+best_weight;
359 static void co_collect_units(ir_node *irn, void *env) {
360 copy_opt_t *co = env;
362 arch_register_req_t req;
364 if (!is_curr_reg_class(co, irn))
366 if (!co_is_optimizable_root(co, irn))
369 /* Init a new unit */
370 unit = xcalloc(1, sizeof(*unit));
372 unit->node_count = 1;
373 INIT_LIST_HEAD(&unit->queue);
375 /* Phi with some/all of its arguments */
376 if (is_Reg_Phi(irn)) {
380 arity = get_irn_arity(irn);
381 unit->nodes = xmalloc((arity+1) * sizeof(*unit->nodes));
382 unit->costs = xmalloc((arity+1) * sizeof(*unit->costs));
383 unit->nodes[0] = irn;
386 for (i=0; i<arity; ++i) {
388 ir_node *arg = get_irn_n(irn, i);
390 assert(is_curr_reg_class(co, arg) && "Argument not in same register class.");
393 if (nodes_interfere(co->cenv, irn, arg)) {
394 unit->inevitable_costs += co->get_costs(co, irn, arg, i);
398 /* Else insert the argument of the phi to the members of this ou */
399 DBG((dbg, LEVEL_1, "\t Member: %+F\n", arg));
401 /* Check if arg has occurred at a prior position in the arg/list */
403 for (o=0; o<unit->node_count; ++o)
404 if (unit->nodes[o] == arg) {
409 if (!arg_pos) { /* a new argument */
410 /* insert node, set costs */
411 unit->nodes[unit->node_count] = arg;
412 unit->costs[unit->node_count] = co->get_costs(co, irn, arg, i);
414 } else { /* arg has occured before in same phi */
415 /* increase costs for existing arg */
416 unit->costs[arg_pos] += co->get_costs(co, irn, arg, i);
419 unit->nodes = xrealloc(unit->nodes, unit->node_count * sizeof(*unit->nodes));
420 unit->costs = xrealloc(unit->costs, unit->node_count * sizeof(*unit->costs));
423 /* Proj of a perm with corresponding arg */
424 if (is_Perm_Proj(co->aenv, irn)) {
425 assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
426 unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
427 unit->costs = xmalloc(2 * sizeof(*unit->costs));
428 unit->node_count = 2;
429 unit->nodes[0] = irn;
430 unit->nodes[1] = get_Perm_src(irn);
431 unit->costs[1] = co->get_costs(co, irn, unit->nodes[1], -1);
434 /* Src == Tgt of a 2-addr-code instruction */
435 if (is_2addr_code(co->aenv, irn, &req)) {
436 ir_node *other = req.other_same;
437 if (!nodes_interfere(co->cenv, irn, other)) {
438 unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
439 unit->costs = xmalloc(2 * sizeof(*unit->costs));
440 unit->node_count = 2;
441 unit->nodes[0] = irn;
442 unit->nodes[1] = other;
443 unit->costs[1] = co->get_costs(co, irn, other, -1);
446 assert(0 && "This is not an optimizable node!");
448 /* Insert the new unit at a position according to its costs */
449 if (unit->node_count > 1) {
451 struct list_head *tmp;
453 /* Determine the maximum costs this unit can cause: all_nodes_cost */
454 for(i=1; i<unit->node_count; ++i) {
455 unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
456 unit->all_nodes_costs += unit->costs[i];
459 /* Determine the minimal costs this unit will cause: min_nodes_costs */
460 unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
461 /* Insert the new ou according to its sort_key */
463 while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
465 list_add(&unit->units, tmp);
471 #ifdef QUICK_AND_DIRTY_HACK
473 static int compare_ous(const void *k1, const void *k2) {
474 const unit_t *u1 = *((const unit_t **) k1);
475 const unit_t *u2 = *((const unit_t **) k2);
476 int i, o, u1_has_constr, u2_has_constr;
477 arch_register_req_t req;
478 const arch_env_t *aenv = u1->co->aenv;
480 /* Units with constraints come first */
482 for (i=0; i<u1->node_count; ++i) {
483 arch_get_register_req(aenv, &req, u1->nodes[i], -1);
484 if (arch_register_req_is(&req, limited)) {
491 for (i=0; i<u2->node_count; ++i) {
492 arch_get_register_req(aenv, &req, u2->nodes[i], -1);
493 if (arch_register_req_is(&req, limited)) {
499 if (u1_has_constr != u2_has_constr)
500 return u2_has_constr - u1_has_constr;
502 /* Now check, whether the two units are connected */
504 for (i=0; i<u1->node_count; ++i)
505 for (o=0; o<u2->node_count; ++o)
506 if (u1->nodes[i] == u2->nodes[o])
510 /* After all, the sort key decides. Greater keys come first. */
511 return u2->sort_key - u1->sort_key;
516 * Sort the ou's according to constraints and their sort_key
518 static void co_sort_units(copy_opt_t *co) {
519 int i, count = 0, costs;
522 /* get the number of ous, remove them form the list and fill the array */
523 list_for_each_entry(unit_t, ou, &co->units, units)
525 ous = alloca(count * sizeof(*ous));
527 costs = co_get_max_copy_costs(co);
530 list_for_each_entry(unit_t, ou, &co->units, units)
533 INIT_LIST_HEAD(&co->units);
535 assert(count == i && list_empty(&co->units));
537 for (i=0; i<count; ++i)
538 ir_printf("%+F\n", ous[i]->nodes[0]);
540 qsort(ous, count, sizeof(*ous), compare_ous);
543 for (i=0; i<count; ++i)
544 ir_printf("%+F\n", ous[i]->nodes[0]);
546 /* reinsert into list in correct order */
547 for (i=0; i<count; ++i)
548 list_add_tail(&ous[i]->units, &co->units);
550 assert(costs == co_get_max_copy_costs(co));
554 void co_build_ou_structure(copy_opt_t *co) {
555 DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
556 INIT_LIST_HEAD(&co->units);
557 irg_walk_graph(co->irg, co_collect_units, NULL, co);
558 #ifdef QUICK_AND_DIRTY_HACK
563 void co_free_ou_structure(copy_opt_t *co) {
566 list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
571 co->units.next = NULL;
574 /* co_solve_heuristic() is implemented in becopyheur.c */
576 int co_get_max_copy_costs(const copy_opt_t *co) {
582 list_for_each_entry(unit_t, curr, &co->units, units) {
583 res += curr->inevitable_costs;
584 for (i=1; i<curr->node_count; ++i)
585 res += curr->costs[i];
590 int co_get_inevit_copy_costs(const copy_opt_t *co) {
596 list_for_each_entry(unit_t, curr, &co->units, units)
597 res += curr->inevitable_costs;
601 int co_get_copy_costs(const copy_opt_t *co) {
607 list_for_each_entry(unit_t, curr, &co->units, units) {
608 int root_col = get_irn_col(co, curr->nodes[0]);
609 DBG((dbg, LEVEL_1, " %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
610 res += curr->inevitable_costs;
611 for (i=1; i<curr->node_count; ++i) {
612 int arg_col = get_irn_col(co, curr->nodes[i]);
613 if (root_col != arg_col) {
614 DBG((dbg, LEVEL_1, " %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
615 res += curr->costs[i];
622 int co_get_lower_bound(const copy_opt_t *co) {
628 list_for_each_entry(unit_t, curr, &co->units, units)
629 res += curr->inevitable_costs + curr->min_nodes_costs;
633 /******************************************************************************
635 / ____| | | / ____| |
636 | | __ _ __ __ _ _ __ | |__ | (___ | |_ ___ _ __ __ _ __ _ ___
637 | | |_ | '__/ _` | '_ \| '_ \ \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
638 | |__| | | | (_| | |_) | | | | ____) | || (_) | | | (_| | (_| | __/
639 \_____|_| \__,_| .__/|_| |_| |_____/ \__\___/|_| \__,_|\__, |\___|
642 ******************************************************************************/
644 static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) {
645 const affinity_node_t *n1 = k1;
646 const affinity_node_t *n2 = k2;
648 return (n1->irn != n2->irn);
651 static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
652 affinity_node_t new_node, *node;
653 neighb_t new_nbr, *nbr;
658 new_node.neighbours = NULL;
659 node = set_insert(co->nodes, &new_node, sizeof(new_node), HASH_PTR(new_node.irn));
662 for (nbr = node->neighbours; nbr; nbr = nbr->next)
663 if (nbr->irn == n2) {
668 /* if we did not find n2 in n1's neighbourhood insert it */
670 obstack_grow(&co->obst, &new_nbr, sizeof(new_nbr));
671 nbr = obstack_finish(&co->obst);
674 nbr->next = node->neighbours;
675 node->neighbours = nbr;
679 /* now nbr points to n1's neighbour-entry of n2 */
683 static INLINE void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
684 if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
685 add_edge(co, n1, n2, costs);
686 add_edge(co, n2, n1, costs);
690 static void build_graph_walker(ir_node *irn, void *env) {
691 copy_opt_t *co = env;
693 arch_register_req_t req;
694 const arch_register_t *reg;
696 if (!is_curr_reg_class(co, irn) || arch_irn_is(co->aenv, irn, ignore))
699 reg = arch_get_irn_register(co->aenv, irn);
700 if (arch_register_type_is(reg, ignore))
705 for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
706 ir_node *arg = get_irn_n(irn, pos);
707 add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
711 else if (is_Perm_Proj(co->aenv, irn)) {
712 ir_node *arg = get_Perm_src(irn);
713 add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
717 else if (is_2addr_code(co->aenv, irn, &req))
718 add_edges(co, irn, req.other_same, co->get_costs(co, irn, req.other_same, 0));
721 void co_build_graph_structure(copy_opt_t *co) {
722 obstack_init(&co->obst);
723 co->nodes = new_set(compare_affinity_node_t, 32);
725 irg_walk_graph(co->irg, build_graph_walker, NULL, co);
728 void co_free_graph_structure(copy_opt_t *co) {
732 obstack_free(&co->obst, NULL);
736 /* co_solve_ilp1() co_solve_ilp2() are implemented in becopyilpX.c */
738 int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) {
739 affinity_node_t new_node, *n;
744 n = set_find(co->nodes, &new_node, sizeof(new_node), HASH_PTR(new_node.irn));
746 return (n->degree > 0);
751 void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
753 be_ifg_t *ifg = co->cenv->ifg;
754 int *color_map = alloca(co->cls->n_regs * sizeof(color_map[0]));
755 bitset_t *adm = bitset_alloca(co->cls->n_regs);
762 for(i = 0; i < co->cls->n_regs; ++i) {
763 const arch_register_t *reg = &co->cls->regs[i];
764 color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_regs++;
768 * n contains the first node number.
769 * the values below n are the pre-colored register nodes
772 it = be_ifg_nodes_iter_alloca(ifg);
773 nit = be_ifg_neighbours_iter_alloca(ifg);
776 be_ifg_foreach_node(ifg, it, irn) {
777 if(!arch_irn_is(co->aenv, irn, ignore))
778 set_irn_link(irn, INT_TO_PTR(n++));
781 fprintf(f, "%d %d\n", n, n_regs);
783 be_ifg_foreach_node(ifg, it, irn) {
784 if(!arch_irn_is(co->aenv, irn, ignore)) {
785 int idx = PTR_TO_INT(get_irn_link(irn));
786 affinity_node_t *a = get_affinity_info(co, irn);
788 arch_register_req_t req;
791 arch_get_register_req(co->aenv, &req, irn, BE_OUT_POS(0));
792 if(arch_register_req_is(&req, limited)) {
793 bitset_clear_all(adm);
794 req.limited(req.limited_env, adm);
795 for(i = 0; i < co->cls->n_regs; ++i)
796 if(!bitset_is_set(adm, i) && color_map[i] >= 0)
797 fprintf(f, "%d %d -1\n", color_map[i], idx);
802 be_ifg_foreach_neighbour(ifg, nit, irn, adj) {
803 if(!arch_irn_is(co->aenv, adj, ignore)) {
804 int adj_idx = PTR_TO_INT(get_irn_link(adj));
806 fprintf(f, "%d %d -1\n", idx, adj_idx);
813 co_gs_foreach_neighb(a, n) {
814 if(!arch_irn_is(co->aenv, n->irn, ignore)) {
815 int n_idx = PTR_TO_INT(get_irn_link(n->irn));
817 fprintf(f, "%d %d %d\n", idx, n_idx, n->costs);
825 typedef struct _appel_clique_walker_t {
827 const copy_opt_t *co;
834 } appel_clique_walker_t;
836 typedef struct _appel_block_info_t {
846 } appel_block_info_t;
848 static int appel_aff_weight(const appel_clique_walker_t *env, ir_node *bl)
851 double freq = get_block_execfreq(env->co->cenv->execfreq, bl);
852 int res = (int) freq;
853 return res == 0 ? 1 : res;
855 ir_loop *loop = get_irn_loop(bl);
857 int d = get_loop_depth(loop);
864 static void *appel_clique_walker_irn_init(phase_t *phase, ir_node *irn, void *old)
866 appel_block_info_t *res = NULL;
869 appel_clique_walker_t *d = (void *) phase;
870 res = phase_alloc(phase, sizeof(res[0]));
871 res->phi_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
872 res->live_end_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
873 res->live_in_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in_nr));
874 res->live_end = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end));
875 res->live_in = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
876 res->phi = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
882 typedef struct _insn_list_t {
884 struct list_head list;
887 static int appel_get_live_end_nr(appel_clique_walker_t *env, ir_node *bl, ir_node *irn)
889 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
892 for(i = 0; i < bli->n_live_end; ++i)
893 if(bli->live_end[i] == irn)
894 return bli->live_end_nr[i];
899 static int appel_dump_clique(appel_clique_walker_t *env, pset *live, ir_node *bl, int curr_nr, int start_nr)
901 ir_node **live_arr = alloca(env->co->cls->n_regs * sizeof(live_arr[0]));
907 foreach_pset(live, irn)
908 live_arr[n_live++] = irn;
910 /* dump the live after clique */
912 for(j = 0; j < n_live; ++j) {
915 for(k = j + 1; k < n_live; ++k) {
916 fprintf(env->f, "%d %d -1 ", curr_nr + j, curr_nr + k);
918 fprintf(env->f, "\n");
922 /* dump the affinities */
923 for(j = 0; !env->dumb && j < n_live; ++j) {
924 ir_node *irn = live_arr[j];
925 int old_nr = PTR_TO_INT(get_irn_link(irn));
927 /* if the node was already live in the last insn dump the affinity */
928 if(old_nr > start_nr) {
929 int weight = appel_aff_weight(env, bl);
930 fprintf(env->f, "%d %d %d\n", old_nr, curr_nr + j, weight);
934 /* set the current numbers into the link field. */
935 for(j = 0; j < n_live; ++j) {
936 ir_node *irn = live_arr[j];
937 set_irn_link(irn, INT_TO_PTR(curr_nr + j));
940 return curr_nr + n_live;
943 static void appel_walker(ir_node *bl, void *data)
945 appel_clique_walker_t *env = data;
946 appel_block_info_t *bli = phase_get_or_set_irn_data(&env->ph, bl);
947 struct obstack *obst = &env->obst;
948 void *base = obstack_base(obst);
949 pset *live = pset_new_ptr_default();
953 int start_nr = env->curr_nr;
954 int curr_nr = start_nr;
956 be_insn_env_t insn_env;
961 insn_env.aenv = env->co->aenv;
962 insn_env.cls = env->co->cls;
963 insn_env.obst = obst;
964 insn_env.ignore_colors = env->co->cenv->ignore_colors;
966 /* Guess how many insns will be in this block. */
967 sched_foreach(bl, irn)
971 insns = malloc(n_nodes * sizeof(insns[0]));
973 /* Put all insns in an array. */
974 irn = sched_first(bl);
975 while(!sched_is_end(irn)) {
977 insn = be_scan_insn(&insn_env, irn);
978 insns[n_insns++] = insn;
979 irn = insn->next_insn;
982 DBG((env->co->cenv->dbg, LEVEL_2, "%+F\n", bl));
983 be_liveness_end_of_block(env->co->cenv->lv, env->co->aenv, env->co->cls, bl, live);
985 /* Generate the bad and ugly. */
986 for(i = n_insns - 1; i >= 0; --i) {
987 be_insn_t *insn = insns[i];
989 /* The first live set has to be saved in the block border set. */
990 if(i == n_insns - 1) {
992 foreach_pset(live, irn) {
993 bli->live_end[j] = irn;
994 bli->live_end_nr[j] = curr_nr + j;
1001 for(j = 0; j < insn->use_start; ++j) {
1002 ir_node *op = insn->ops[j].carrier;
1003 bitset_t *adm = insn->ops[j].regs;
1007 if(!insn->ops[j].has_constraints)
1011 foreach_pset(live, irn) {
1019 assert(nr < pset_count(live));
1021 for(k = 0; k < env->co->cls->n_regs; ++k) {
1022 int mapped_col = env->color_map[k];
1023 if(mapped_col >= 0 && !bitset_is_set(adm, k) && !bitset_is_set(env->co->cenv->ignore_colors, k))
1024 fprintf(env->f, "%d %d -1\n", curr_nr + nr, mapped_col);
1029 /* dump the clique and update the stuff. */
1030 curr_nr = appel_dump_clique(env, live, bl, curr_nr, start_nr);
1032 /* remove all defs. */
1033 for(j = 0; j < insn->use_start; ++j)
1034 pset_remove_ptr(live, insn->ops[j].carrier);
1036 if(is_Phi(insn->irn) && arch_irn_consider_in_reg_alloc(env->co->aenv, env->co->cls, insn->irn)) {
1037 bli->phi[bli->n_phi] = insn->irn;
1038 bli->phi_nr[bli->n_phi] = PTR_TO_INT(get_irn_link(insn->irn));
1044 for(j = insn->use_start; j < insn->n_ops; ++j)
1045 pset_insert_ptr(live, insn->ops[j].carrier);
1048 /* print the start clique. */
1049 curr_nr = appel_dump_clique(env, live, bl, curr_nr, start_nr);
1052 foreach_pset(live, irn) {
1053 bli->live_in[i] = irn;
1054 bli->live_in_nr[i] = PTR_TO_INT(get_irn_link(irn));
1061 obstack_free(obst, base);
1062 env->curr_nr = curr_nr;
1065 static void appel_inter_block_aff(ir_node *bl, void *data)
1067 appel_clique_walker_t *env = data;
1068 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
1072 for(i = 0; i < bli->n_live_in; ++i) {
1073 ir_node *irn = bli->live_in[i];
1075 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1076 ir_node *pred = get_Block_cfgpred_block(bl, j);
1078 int nr = appel_get_live_end_nr(env, pred, irn);
1080 fprintf(env->f, "%d %d 1\n", bli->live_in_nr[i], nr);
1084 for(i = 0; i < bli->n_phi; ++i) {
1085 ir_node *irn = bli->phi[i];
1087 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1088 ir_node *pred = get_Block_cfgpred_block(bl, j);
1089 ir_node *op = get_irn_n(irn, j);
1091 int nr = appel_get_live_end_nr(env, pred, op);
1093 fprintf(env->f, "%d %d 1\n", bli->phi_nr[i], nr);
1099 void co_dump_appel_graph_cliques(const copy_opt_t *co, FILE *f)
1103 appel_clique_walker_t env;
1104 bitset_t *adm = bitset_alloca(co->cls->n_regs);
1106 be_liveness_recompute(co->cenv->lv);
1107 obstack_init(&env.obst);
1108 phase_init(&env.ph, "appel_clique_dumper", co->irg, PHASE_DEFAULT_GROWTH, appel_clique_walker_irn_init);
1109 env.curr_nr = co->cls->n_regs;
1113 bitset_copy(adm, co->cenv->ignore_colors);
1114 bitset_flip_all(adm);
1116 /* Make color map. */
1117 env.color_map = alloca(co->cls->n_regs * sizeof(env.color_map[0]));
1118 for(i = 0, n_colors = 0; i < co->cls->n_regs; ++i) {
1119 const arch_register_t *reg = &co->cls->regs[i];
1120 env.color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_colors++;
1124 env.curr_nr = n_colors;
1125 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1126 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1128 fprintf(f, "%d %d\n", env.curr_nr, n_colors);
1130 /* make the first k nodes interfere */
1131 for(i = 0; i < n_colors; ++i) {
1133 for(j = i + 1; j < n_colors; ++j)
1134 fprintf(f, "%d %d -1 ", i, j);
1139 env.curr_nr = n_colors;
1140 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1141 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1142 irg_block_walk_graph(co->irg, appel_inter_block_aff, NULL, &env);
1143 obstack_free(&env.obst, NULL);
1147 ___ _____ ____ ____ ___ _____ ____ _
1148 |_ _| ___/ ___| | _ \ / _ \_ _| | _ \ _ _ _ __ ___ _ __ (_)_ __ __ _
1149 | || |_ | | _ | | | | | | || | | | | | | | | '_ ` _ \| '_ \| | '_ \ / _` |
1150 | || _|| |_| | | |_| | |_| || | | |_| | |_| | | | | | | |_) | | | | | (_| |
1151 |___|_| \____| |____/ \___/ |_| |____/ \__,_|_| |_| |_| .__/|_|_| |_|\__, |
1155 static const char *get_dot_color_name(int col)
1157 static const char *names[] = {
1191 return col < sizeof(names)/sizeof(names[0]) ? names[col] : "white";
1194 typedef struct _co_ifg_dump_t {
1195 const copy_opt_t *co;
1199 static void ifg_dump_graph_attr(FILE *f, void *self)
1201 fprintf(f, "overlap=scale");
1204 static int ifg_is_dump_node(void *self, ir_node *irn)
1206 co_ifg_dump_t *cod = self;
1207 return !arch_irn_is(cod->co->aenv, irn, ignore);
1210 static void ifg_dump_node_attr(FILE *f, void *self, ir_node *irn)
1212 co_ifg_dump_t *env = self;
1213 const arch_register_t *reg = arch_get_irn_register(env->co->aenv, irn);
1214 arch_register_req_t req;
1217 arch_get_register_req(env->co->aenv, &req, irn, BE_OUT_POS(0));
1218 limited = arch_register_req_is(&req, limited);
1220 if(env->flags & CO_IFG_DUMP_LABELS) {
1221 ir_fprintf(f, "label=\"%+F", irn);
1223 if((env->flags & CO_IFG_DUMP_CONSTR) && limited) {
1224 bitset_t *bs = bitset_alloca(env->co->cls->n_regs);
1225 req.limited(req.limited_env, bs);
1226 ir_fprintf(f, "\\n%B", bs);
1228 ir_fprintf(f, "\" ");
1232 fprintf(f, "label=\"\" shape=point " );
1234 if(env->flags & CO_IFG_DUMP_SHAPE)
1235 fprintf(f, "shape=%s ", limited ? "diamond" : "ellipse");
1237 if(env->flags & CO_IFG_DUMP_COLORS)
1238 fprintf(f, "style=filled color=%s ", get_dot_color_name(reg->index));
1241 static void ifg_dump_at_end(FILE *file, void *self)
1243 co_ifg_dump_t *env = self;
1246 co_gs_foreach_aff_node(env->co, a) {
1247 const arch_register_t *ar = arch_get_irn_register(env->co->aenv, a->irn);
1248 unsigned aidx = get_irn_idx(a->irn);
1251 co_gs_foreach_neighb(a, n) {
1252 const arch_register_t *nr = arch_get_irn_register(env->co->aenv, n->irn);
1253 unsigned nidx = get_irn_idx(n->irn);
1256 const char *color = nr == ar ? "blue" : "red";
1257 fprintf(file, "\tn%d -- n%d [weight=0.01 ", aidx, nidx);
1258 if(env->flags & CO_IFG_DUMP_LABELS)
1259 fprintf(file, "label=\"%d\" ", n->costs);
1260 if(env->flags & CO_IFG_DUMP_COLORS)
1261 fprintf(file, "color=%s ", color);
1263 fprintf(file, "style=dashed");
1264 fprintf(file, "];\n");
1271 static be_ifg_dump_dot_cb_t ifg_dot_cb = {
1273 ifg_dump_graph_attr,
1282 void co_dump_ifg_dot(const copy_opt_t *co, FILE *f, unsigned flags)
1288 be_ifg_dump_dot(co->cenv->ifg, co->irg, f, &ifg_dot_cb, &cod);
1292 void co_solve_park_moon(copy_opt_t *opt)
1297 static int void_algo(copy_opt_t *co)
1304 / \ | | __ _ ___ _ __(_) |_| |__ _ __ ___ ___
1305 / _ \ | |/ _` |/ _ \| '__| | __| '_ \| '_ ` _ \/ __|
1306 / ___ \| | (_| | (_) | | | | |_| | | | | | | | \__ \
1307 /_/ \_\_|\__, |\___/|_| |_|\__|_| |_|_| |_| |_|___/
1311 static co_algo_t *algos[] = {
1314 co_solve_heuristic_new,
1315 co_solve_heuristic_java,
1323 | \/ | __ _(_)_ __ | _ \ _ __(_)_ _____ _ __
1324 | |\/| |/ _` | | '_ \ | | | | '__| \ \ / / _ \ '__|
1325 | | | | (_| | | | | | | |_| | | | |\ V / __/ |
1326 |_| |_|\__,_|_|_| |_| |____/|_| |_| \_/ \___|_|
1330 void co_driver(be_chordal_env_t *cenv)
1333 co_algo_t *algo_func;
1336 if(algo < 0 || algo >= CO_ALGO_LAST)
1339 co = new_copy_opt(cenv, cost_func);
1340 co_build_ou_structure(co);
1341 co_build_graph_structure(co);
1342 init_costs = co_get_copy_costs(co);
1344 /* Dump the interference graph in Appel's format. */
1345 if(dump_flags & DUMP_APPEL) {
1346 FILE *f = be_chordal_open(cenv, "", ".apl");
1347 co_dump_appel_graph(co, f);
1351 if(dump_flags & DUMP_BEFORE) {
1352 FILE *f = be_chordal_open(cenv, "", "-before.dot");
1353 co_dump_ifg_dot(co, f, style_flags);
1357 algo_func = algos[algo];
1360 if(dump_flags & DUMP_AFTER) {
1361 FILE *f = be_chordal_open(cenv, "", "-after.dot");
1362 co_dump_ifg_dot(co, f, style_flags);
1367 int optimizable_costs = co_get_max_copy_costs(co) - co_get_lower_bound(co);
1368 int remaining = co_get_copy_costs(co);
1369 int evitable = remaining - co_get_lower_bound(co);
1371 ir_printf("%30F %10s %10d%10d%10d%10d", cenv->irg, cenv->cls->name,
1372 co_get_max_copy_costs(co), init_costs,
1373 co_get_inevit_copy_costs(co), co_get_lower_bound(co));
1375 if(optimizable_costs > 0)
1376 printf("%10d %5.2f\n", remaining, (evitable * 100.0) / optimizable_costs);
1378 printf("%10d %5s\n", remaining, "-");
1381 co_free_graph_structure(co);
1382 co_free_ou_structure(co);