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));
448 } else if (is_Perm_Proj(co->aenv, irn)) {
449 /* Proj of a perm with corresponding arg */
450 assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
451 unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
452 unit->costs = xmalloc(2 * sizeof(*unit->costs));
453 unit->node_count = 2;
454 unit->nodes[0] = irn;
455 unit->nodes[1] = get_Perm_src(irn);
456 unit->costs[1] = co->get_costs(co, irn, unit->nodes[1], -1);
458 const arch_register_req_t *req =
459 arch_get_register_req(co->aenv, irn, -1);
461 /* Src == Tgt of a 2-addr-code instruction */
462 if (is_2addr_code(req)) {
463 ir_node *other = get_irn_n(skip_Proj(irn), req->other_same[0]);
464 ir_node *other2 = NULL;
467 if (arch_irn_is(co->aenv, other, ignore) ||
468 nodes_interfere(co->cenv, irn, other)) {
471 if (req->other_same[1] != -1) {
472 other2 = get_irn_n(skip_Proj(irn), req->other_same[1]);
473 if (arch_irn_is(co->aenv, other2, ignore) ||
474 nodes_interfere(co->cenv, irn, other2)) {
478 count = 1 + (other != NULL) + (other2 != NULL && other != other2);
483 unit->nodes = xmalloc(count * sizeof(*unit->nodes));
484 unit->costs = xmalloc(count * sizeof(*unit->costs));
485 unit->node_count = count;
486 unit->nodes[i] = irn;
489 unit->nodes[i] = other;
490 unit->costs[i] = co->get_costs(co, irn, other, -1);
492 if (other2 != NULL) {
493 if (other == other2) {
494 unit->costs[i] += co->get_costs(co, irn, other2, -1);
497 unit->nodes[i] = other2;
498 unit->costs[i] = co->get_costs(co, irn, other2, -1);
503 assert(0 && "This is not an optimizable node!");
507 /* Insert the new unit at a position according to its costs */
508 if (unit->node_count > 1) {
510 struct list_head *tmp;
512 /* Determine the maximum costs this unit can cause: all_nodes_cost */
513 for(i=1; i<unit->node_count; ++i) {
514 unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
515 unit->all_nodes_costs += unit->costs[i];
518 /* Determine the minimal costs this unit will cause: min_nodes_costs */
519 unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
520 /* Insert the new ou according to its sort_key */
522 while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
524 list_add(&unit->units, tmp);
530 #ifdef QUICK_AND_DIRTY_HACK
532 static int compare_ous(const void *k1, const void *k2) {
533 const unit_t *u1 = *((const unit_t **) k1);
534 const unit_t *u2 = *((const unit_t **) k2);
535 int i, o, u1_has_constr, u2_has_constr;
536 arch_register_req_t req;
537 const arch_env_t *aenv = u1->co->aenv;
539 /* Units with constraints come first */
541 for (i=0; i<u1->node_count; ++i) {
542 arch_get_register_req(aenv, &req, u1->nodes[i], -1);
543 if (arch_register_req_is(&req, limited)) {
550 for (i=0; i<u2->node_count; ++i) {
551 arch_get_register_req(aenv, &req, u2->nodes[i], -1);
552 if (arch_register_req_is(&req, limited)) {
558 if (u1_has_constr != u2_has_constr)
559 return u2_has_constr - u1_has_constr;
561 /* Now check, whether the two units are connected */
563 for (i=0; i<u1->node_count; ++i)
564 for (o=0; o<u2->node_count; ++o)
565 if (u1->nodes[i] == u2->nodes[o])
569 /* After all, the sort key decides. Greater keys come first. */
570 return u2->sort_key - u1->sort_key;
575 * Sort the ou's according to constraints and their sort_key
577 static void co_sort_units(copy_opt_t *co) {
578 int i, count = 0, costs;
581 /* get the number of ous, remove them form the list and fill the array */
582 list_for_each_entry(unit_t, ou, &co->units, units)
584 ous = alloca(count * sizeof(*ous));
586 costs = co_get_max_copy_costs(co);
589 list_for_each_entry(unit_t, ou, &co->units, units)
592 INIT_LIST_HEAD(&co->units);
594 assert(count == i && list_empty(&co->units));
596 for (i=0; i<count; ++i)
597 ir_printf("%+F\n", ous[i]->nodes[0]);
599 qsort(ous, count, sizeof(*ous), compare_ous);
602 for (i=0; i<count; ++i)
603 ir_printf("%+F\n", ous[i]->nodes[0]);
605 /* reinsert into list in correct order */
606 for (i=0; i<count; ++i)
607 list_add_tail(&ous[i]->units, &co->units);
609 assert(costs == co_get_max_copy_costs(co));
613 void co_build_ou_structure(copy_opt_t *co) {
614 DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
615 INIT_LIST_HEAD(&co->units);
616 irg_walk_graph(co->irg, co_collect_units, NULL, co);
617 #ifdef QUICK_AND_DIRTY_HACK
622 void co_free_ou_structure(copy_opt_t *co) {
625 list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
630 co->units.next = NULL;
633 /* co_solve_heuristic() is implemented in becopyheur.c */
635 int co_get_max_copy_costs(const copy_opt_t *co) {
641 list_for_each_entry(unit_t, curr, &co->units, units) {
642 res += curr->inevitable_costs;
643 for (i=1; i<curr->node_count; ++i)
644 res += curr->costs[i];
649 int co_get_inevit_copy_costs(const copy_opt_t *co) {
655 list_for_each_entry(unit_t, curr, &co->units, units)
656 res += curr->inevitable_costs;
660 int co_get_copy_costs(const copy_opt_t *co) {
666 list_for_each_entry(unit_t, curr, &co->units, units) {
667 int root_col = get_irn_col(co, curr->nodes[0]);
668 DBG((dbg, LEVEL_1, " %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
669 res += curr->inevitable_costs;
670 for (i=1; i<curr->node_count; ++i) {
671 int arg_col = get_irn_col(co, curr->nodes[i]);
672 if (root_col != arg_col) {
673 DBG((dbg, LEVEL_1, " %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
674 res += curr->costs[i];
681 int co_get_lower_bound(const copy_opt_t *co) {
687 list_for_each_entry(unit_t, curr, &co->units, units)
688 res += curr->inevitable_costs + curr->min_nodes_costs;
692 void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat)
694 bitset_t *seen = bitset_irg_malloc(co->irg);
697 memset(stat, 0, sizeof(stat[0]));
699 /* count affinity edges. */
700 co_gs_foreach_aff_node(co, an) {
702 stat->aff_nodes += 1;
703 bitset_add_irn(seen, an->irn);
704 co_gs_foreach_neighb(an, neigh) {
705 if(!bitset_contains_irn(seen, neigh->irn)) {
706 stat->aff_edges += 1;
707 stat->max_costs += neigh->costs;
709 if(get_irn_col(co, an->irn) != get_irn_col(co, neigh->irn)) {
710 stat->costs += neigh->costs;
711 stat->unsatisfied_edges += 1;
714 if(nodes_interfere(co->cenv, an->irn, neigh->irn)) {
716 stat->inevit_costs += neigh->costs;
726 /******************************************************************************
728 / ____| | | / ____| |
729 | | __ _ __ __ _ _ __ | |__ | (___ | |_ ___ _ __ __ _ __ _ ___
730 | | |_ | '__/ _` | '_ \| '_ \ \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
731 | |__| | | | (_| | |_) | | | | ____) | || (_) | | | (_| | (_| | __/
732 \_____|_| \__,_| .__/|_| |_| |_____/ \__\___/|_| \__,_|\__, |\___|
735 ******************************************************************************/
737 static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) {
738 const affinity_node_t *n1 = k1;
739 const affinity_node_t *n2 = k2;
742 return (n1->irn != n2->irn);
745 static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
746 affinity_node_t new_node, *node;
752 new_node.neighbours = NULL;
753 node = set_insert(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
755 for (nbr = node->neighbours; nbr; nbr = nbr->next)
756 if (nbr->irn == n2) {
761 /* if we did not find n2 in n1's neighbourhood insert it */
763 nbr = obstack_alloc(&co->obst, sizeof(*nbr));
766 nbr->next = node->neighbours;
768 node->neighbours = nbr;
772 /* now nbr points to n1's neighbour-entry of n2 */
776 static INLINE void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
777 if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
778 add_edge(co, n1, n2, costs);
779 add_edge(co, n2, n1, costs);
783 static void build_graph_walker(ir_node *irn, void *env) {
784 copy_opt_t *co = env;
786 const arch_register_t *reg;
788 if (!is_curr_reg_class(co, irn) || arch_irn_is(co->aenv, irn, ignore))
791 reg = arch_get_irn_register(co->aenv, irn);
792 if (arch_register_type_is(reg, ignore))
795 if (is_Reg_Phi(irn)) { /* Phis */
796 for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
797 ir_node *arg = get_irn_n(irn, pos);
798 add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
801 else if (is_Perm_Proj(co->aenv, irn)) { /* Perms */
802 ir_node *arg = get_Perm_src(irn);
803 add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
805 else { /* 2-address code */
806 const arch_register_req_t *req = arch_get_register_req(co->aenv, irn, -1);
807 if (is_2addr_code(req)) {
809 for (i = req->other_same; i != ENDOF(req->other_same); ++i) {
814 other = get_irn_n(skip_Proj(irn), *i);
815 if (! arch_irn_is(co->aenv, other, ignore))
816 add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
822 void co_build_graph_structure(copy_opt_t *co) {
823 obstack_init(&co->obst);
824 co->nodes = new_set(compare_affinity_node_t, 32);
826 irg_walk_graph(co->irg, build_graph_walker, NULL, co);
829 void co_free_graph_structure(copy_opt_t *co) {
833 obstack_free(&co->obst, NULL);
837 /* co_solve_ilp1() co_solve_ilp2() are implemented in becopyilpX.c */
839 int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) {
840 affinity_node_t new_node, *n;
845 n = set_find(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
847 return (n->degree > 0);
852 void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
854 be_ifg_t *ifg = co->cenv->ifg;
855 int *color_map = alloca(co->cls->n_regs * sizeof(color_map[0]));
862 for(i = 0; i < co->cls->n_regs; ++i) {
863 const arch_register_t *reg = &co->cls->regs[i];
864 color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_regs++;
868 * n contains the first node number.
869 * the values below n are the pre-colored register nodes
872 it = be_ifg_nodes_iter_alloca(ifg);
873 nit = be_ifg_neighbours_iter_alloca(ifg);
876 be_ifg_foreach_node(ifg, it, irn) {
877 if(!arch_irn_is(co->aenv, irn, ignore))
878 set_irn_link(irn, INT_TO_PTR(n++));
881 fprintf(f, "%d %d\n", n, n_regs);
883 be_ifg_foreach_node(ifg, it, irn) {
884 if(!arch_irn_is(co->aenv, irn, ignore)) {
885 int idx = PTR_TO_INT(get_irn_link(irn));
886 affinity_node_t *a = get_affinity_info(co, irn);
888 const arch_register_req_t *req;
891 req = arch_get_register_req(co->aenv, irn, BE_OUT_POS(0));
892 if(arch_register_req_is(req, limited)) {
893 for(i = 0; i < co->cls->n_regs; ++i) {
894 if(!rbitset_is_set(req->limited, i) && color_map[i] >= 0)
895 fprintf(f, "%d %d -1\n", color_map[i], idx);
900 be_ifg_foreach_neighbour(ifg, nit, irn, adj) {
901 if(!arch_irn_is(co->aenv, adj, ignore)) {
902 int adj_idx = PTR_TO_INT(get_irn_link(adj));
904 fprintf(f, "%d %d -1\n", idx, adj_idx);
911 co_gs_foreach_neighb(a, n) {
912 if(!arch_irn_is(co->aenv, n->irn, ignore)) {
913 int n_idx = PTR_TO_INT(get_irn_link(n->irn));
915 fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs);
923 typedef struct _appel_clique_walker_t {
925 const copy_opt_t *co;
932 } appel_clique_walker_t;
934 typedef struct _appel_block_info_t {
944 } appel_block_info_t;
946 static int appel_aff_weight(const appel_clique_walker_t *env, ir_node *bl)
949 double freq = get_block_execfreq(env->co->cenv->execfreq, bl);
950 int res = (int) freq;
951 return res == 0 ? 1 : res;
953 ir_loop *loop = get_irn_loop(bl);
956 int d = get_loop_depth(loop);
963 static void *appel_clique_walker_irn_init(ir_phase *phase, ir_node *irn, void *old)
965 appel_block_info_t *res = NULL;
969 appel_clique_walker_t *d = (void *) phase;
970 res = phase_alloc(phase, sizeof(res[0]));
971 res->phi_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
972 res->live_end_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end_nr));
973 res->live_in_nr = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in_nr));
974 res->live_end = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_end));
975 res->live_in = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
976 res->phi = phase_alloc(phase, d->co->cls->n_regs * sizeof(res->live_in));
982 typedef struct _insn_list_t {
984 struct list_head list;
987 static int appel_get_live_end_nr(appel_clique_walker_t *env, ir_node *bl, ir_node *irn)
989 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
992 for(i = 0; i < bli->n_live_end; ++i)
993 if(bli->live_end[i] == irn)
994 return bli->live_end_nr[i];
999 static int appel_dump_clique(appel_clique_walker_t *env, const ir_nodeset_t *live, ir_node *bl, int curr_nr, int start_nr)
1001 ir_node **live_arr = alloca(env->co->cls->n_regs * sizeof(live_arr[0]));
1005 ir_nodeset_iterator_t iter;
1008 foreach_ir_nodeset(live, irn, iter)
1009 live_arr[n_live++] = irn;
1011 /* dump the live after clique */
1013 for(j = 0; j < n_live; ++j) {
1016 for(k = j + 1; k < n_live; ++k) {
1017 fprintf(env->f, "%d %d -1 ", curr_nr + j, curr_nr + k);
1019 fprintf(env->f, "\n");
1023 /* dump the affinities */
1024 for(j = 0; !env->dumb && j < n_live; ++j) {
1025 ir_node *irn = live_arr[j];
1026 int old_nr = PTR_TO_INT(get_irn_link(irn));
1028 /* if the node was already live in the last insn dump the affinity */
1029 if(old_nr > start_nr) {
1030 int weight = appel_aff_weight(env, bl);
1031 fprintf(env->f, "%d %d %d\n", old_nr, curr_nr + j, weight);
1035 /* set the current numbers into the link field. */
1036 for(j = 0; j < n_live; ++j) {
1037 ir_node *irn = live_arr[j];
1038 set_irn_link(irn, INT_TO_PTR(curr_nr + j));
1041 return curr_nr + n_live;
1044 static void appel_walker(ir_node *bl, void *data)
1046 appel_clique_walker_t *env = data;
1047 appel_block_info_t *bli = phase_get_or_set_irn_data(&env->ph, bl);
1048 struct obstack *obst = &env->obst;
1049 void *base = obstack_base(obst);
1051 ir_nodeset_iterator_t iter;
1052 be_lv_t *lv = env->co->cenv->birg->lv;
1056 int start_nr = env->curr_nr;
1057 int curr_nr = start_nr;
1059 be_insn_env_t insn_env;
1064 insn_env.aenv = env->co->aenv;
1065 insn_env.cls = env->co->cls;
1066 insn_env.obst = obst;
1067 insn_env.ignore_colors = env->co->cenv->ignore_colors;
1069 /* Guess how many insns will be in this block. */
1070 sched_foreach(bl, irn)
1074 insns = xmalloc(n_nodes * sizeof(insns[0]));
1076 /* Put all insns in an array. */
1077 irn = sched_first(bl);
1078 while(!sched_is_end(irn)) {
1080 insn = be_scan_insn(&insn_env, irn);
1081 insns[n_insns++] = insn;
1082 irn = insn->next_insn;
1085 DBG((dbg, LEVEL_2, "%+F\n", bl));
1086 ir_nodeset_init(&live);
1087 be_liveness_end_of_block(lv, env->co->aenv, env->co->cls, bl, &live);
1089 /* Generate the bad and ugly. */
1090 for(i = n_insns - 1; i >= 0; --i) {
1091 be_insn_t *insn = insns[i];
1093 /* The first live set has to be saved in the block border set. */
1094 if(i == n_insns - 1) {
1096 foreach_ir_nodeset(&live, irn, iter) {
1097 bli->live_end[j] = irn;
1098 bli->live_end_nr[j] = curr_nr + j;
1101 bli->n_live_end = j;
1105 for(j = 0; j < insn->use_start; ++j) {
1106 ir_node *op = insn->ops[j].carrier;
1107 bitset_t *adm = insn->ops[j].regs;
1111 if(!insn->ops[j].has_constraints)
1115 foreach_ir_nodeset(&live, irn, iter) {
1122 assert(nr < ir_nodeset_size(&live));
1124 for(k = 0; k < env->co->cls->n_regs; ++k) {
1125 int mapped_col = env->color_map[k];
1126 if(mapped_col >= 0 && !bitset_is_set(adm, k) && !bitset_is_set(env->co->cenv->ignore_colors, k))
1127 fprintf(env->f, "%d %d -1\n", curr_nr + nr, mapped_col);
1132 /* dump the clique and update the stuff. */
1133 curr_nr = appel_dump_clique(env, &live, bl, curr_nr, start_nr);
1135 /* remove all defs. */
1136 for(j = 0; j < insn->use_start; ++j)
1137 ir_nodeset_remove(&live, insn->ops[j].carrier);
1139 if(is_Phi(insn->irn) && arch_irn_consider_in_reg_alloc(env->co->aenv, env->co->cls, insn->irn)) {
1140 bli->phi[bli->n_phi] = insn->irn;
1141 bli->phi_nr[bli->n_phi] = PTR_TO_INT(get_irn_link(insn->irn));
1147 for(j = insn->use_start; j < insn->n_ops; ++j)
1148 ir_nodeset_insert(&live, insn->ops[j].carrier);
1151 /* print the start clique. */
1152 curr_nr = appel_dump_clique(env, &live, bl, curr_nr, start_nr);
1155 foreach_ir_nodeset(&live, irn, iter) {
1156 bli->live_in[i] = irn;
1157 bli->live_in_nr[i] = PTR_TO_INT(get_irn_link(irn));
1162 ir_nodeset_destroy(&live);
1164 obstack_free(obst, base);
1165 env->curr_nr = curr_nr;
1168 static void appel_inter_block_aff(ir_node *bl, void *data)
1170 appel_clique_walker_t *env = data;
1171 appel_block_info_t *bli = phase_get_irn_data(&env->ph, bl);
1175 for(i = 0; i < bli->n_live_in; ++i) {
1176 ir_node *irn = bli->live_in[i];
1178 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1179 ir_node *pred = get_Block_cfgpred_block(bl, j);
1181 int nr = appel_get_live_end_nr(env, pred, irn);
1183 fprintf(env->f, "%d %d 1\n", bli->live_in_nr[i], nr);
1187 for(i = 0; i < bli->n_phi; ++i) {
1188 ir_node *irn = bli->phi[i];
1190 for(j = 0, n = get_Block_n_cfgpreds(bl); j < n; ++j) {
1191 ir_node *pred = get_Block_cfgpred_block(bl, j);
1192 ir_node *op = get_irn_n(irn, j);
1194 int nr = appel_get_live_end_nr(env, pred, op);
1196 fprintf(env->f, "%d %d 1\n", bli->phi_nr[i], nr);
1202 void co_dump_appel_graph_cliques(const copy_opt_t *co, FILE *f)
1206 appel_clique_walker_t env;
1207 bitset_t *adm = bitset_alloca(co->cls->n_regs);
1208 be_lv_t *lv = co->cenv->birg->lv;
1210 be_liveness_recompute(lv);
1211 obstack_init(&env.obst);
1212 phase_init(&env.ph, "appel_clique_dumper", co->irg, PHASE_DEFAULT_GROWTH, appel_clique_walker_irn_init, NULL);
1213 env.curr_nr = co->cls->n_regs;
1217 bitset_copy(adm, co->cenv->ignore_colors);
1218 bitset_flip_all(adm);
1220 /* Make color map. */
1221 env.color_map = alloca(co->cls->n_regs * sizeof(env.color_map[0]));
1222 for(i = 0, n_colors = 0; i < co->cls->n_regs; ++i) {
1223 const arch_register_t *reg = &co->cls->regs[i];
1224 env.color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_colors++;
1228 env.curr_nr = n_colors;
1229 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1230 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1232 fprintf(f, "%d %d\n", env.curr_nr, n_colors);
1234 /* make the first k nodes interfere */
1235 for(i = 0; i < n_colors; ++i) {
1237 for(j = i + 1; j < n_colors; ++j)
1238 fprintf(f, "%d %d -1 ", i, j);
1243 env.curr_nr = n_colors;
1244 irg_block_walk_graph(co->irg, firm_clear_link, NULL, NULL);
1245 irg_block_walk_graph(co->irg, appel_walker, NULL, &env);
1246 irg_block_walk_graph(co->irg, appel_inter_block_aff, NULL, &env);
1247 obstack_free(&env.obst, NULL);
1251 ___ _____ ____ ____ ___ _____ ____ _
1252 |_ _| ___/ ___| | _ \ / _ \_ _| | _ \ _ _ _ __ ___ _ __ (_)_ __ __ _
1253 | || |_ | | _ | | | | | | || | | | | | | | | '_ ` _ \| '_ \| | '_ \ / _` |
1254 | || _|| |_| | | |_| | |_| || | | |_| | |_| | | | | | | |_) | | | | | (_| |
1255 |___|_| \____| |____/ \___/ |_| |____/ \__,_|_| |_| |_| .__/|_|_| |_|\__, |
1259 static const char *get_dot_color_name(size_t col)
1261 static const char *names[] = {
1295 return col < sizeof(names)/sizeof(names[0]) ? names[col] : "white";
1298 typedef struct _co_ifg_dump_t {
1299 const copy_opt_t *co;
1303 static void ifg_dump_graph_attr(FILE *f, void *self)
1306 fprintf(f, "overlap=scale");
1309 static int ifg_is_dump_node(void *self, ir_node *irn)
1311 co_ifg_dump_t *cod = self;
1312 return !arch_irn_is(cod->co->aenv, irn, ignore);
1315 static void ifg_dump_node_attr(FILE *f, void *self, ir_node *irn)
1317 co_ifg_dump_t *env = self;
1318 const arch_register_t *reg = arch_get_irn_register(env->co->aenv, irn);
1319 const arch_register_req_t *req;
1322 req = arch_get_register_req(env->co->aenv, irn, BE_OUT_POS(0));
1323 limited = arch_register_req_is(req, limited);
1325 if(env->flags & CO_IFG_DUMP_LABELS) {
1326 ir_fprintf(f, "label=\"%+F", irn);
1330 if((env->flags & CO_IFG_DUMP_CONSTR) && limited) {
1331 bitset_t *bs = bitset_alloca(env->co->cls->n_regs);
1332 req.limited(req.limited_env, bs);
1333 ir_fprintf(f, "\\n%B", bs);
1336 ir_fprintf(f, "\" ");
1338 fprintf(f, "label=\"\" shape=point " );
1341 if(env->flags & CO_IFG_DUMP_SHAPE)
1342 fprintf(f, "shape=%s ", limited ? "diamond" : "ellipse");
1344 if(env->flags & CO_IFG_DUMP_COLORS)
1345 fprintf(f, "style=filled color=%s ", get_dot_color_name(reg->index));
1348 static void ifg_dump_at_end(FILE *file, void *self)
1350 co_ifg_dump_t *env = self;
1353 co_gs_foreach_aff_node(env->co, a) {
1354 const arch_register_t *ar = arch_get_irn_register(env->co->aenv, a->irn);
1355 unsigned aidx = get_irn_idx(a->irn);
1358 co_gs_foreach_neighb(a, n) {
1359 const arch_register_t *nr = arch_get_irn_register(env->co->aenv, n->irn);
1360 unsigned nidx = get_irn_idx(n->irn);
1363 const char *color = nr == ar ? "blue" : "red";
1364 fprintf(file, "\tn%d -- n%d [weight=0.01 ", aidx, nidx);
1365 if(env->flags & CO_IFG_DUMP_LABELS)
1366 fprintf(file, "label=\"%d\" ", n->costs);
1367 if(env->flags & CO_IFG_DUMP_COLORS)
1368 fprintf(file, "color=%s ", color);
1370 fprintf(file, "style=dotted");
1371 fprintf(file, "];\n");
1378 static be_ifg_dump_dot_cb_t ifg_dot_cb = {
1380 ifg_dump_graph_attr,
1389 void co_dump_ifg_dot(const copy_opt_t *co, FILE *f, unsigned flags)
1395 be_ifg_dump_dot(co->cenv->ifg, co->irg, f, &ifg_dot_cb, &cod);
1399 void co_solve_park_moon(copy_opt_t *opt)
1404 static int void_algo(copy_opt_t *co)
1412 / \ | | __ _ ___ _ __(_) |_| |__ _ __ ___ ___
1413 / _ \ | |/ _` |/ _ \| '__| | __| '_ \| '_ ` _ \/ __|
1414 / ___ \| | (_| | (_) | | | | |_| | | | | | | | \__ \
1415 /_/ \_\_|\__, |\___/|_| |_|\__|_| |_|_| |_| |_|___/
1422 int can_improve_existing;
1425 static co_algo_info_t algos[] = {
1426 { void_algo, "none", 0 },
1427 { co_solve_heuristic, "heur1", 0 },
1428 { co_solve_heuristic_new, "heur2", 0 },
1429 { co_solve_heuristic_java, "heur3", 0 },
1430 { co_solve_heuristic_mst, "heur4", 0 },
1432 { co_solve_ilp2, "ilp", 1 },
1439 | \/ | __ _(_)_ __ | _ \ _ __(_)_ _____ _ __
1440 | |\/| |/ _` | | '_ \ | | | | '__| \ \ / / _ \ '__|
1441 | | | | (_| | | | | | | |_| | | | |\ V / __/ |
1442 |_| |_|\__,_|_|_| |_| |____/|_| |_| \_/ \___|_|
1446 static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char *suffix)
1451 ir_snprintf(buf, sizeof(buf), "%s%F_%s%s", prefix, env->irg, env->cls->name, suffix);
1452 result = fopen(buf, "wt");
1453 if(result == NULL) {
1454 panic("Couldn't open '%s' for writing.", buf);
1460 void co_driver(be_chordal_env_t *cenv)
1462 lc_timer_t *timer = lc_timer_register("firm.be.copyopt", "runtime");
1463 co_complete_stats_t before, after;
1465 co_algo_t *algo_func;
1466 int was_optimal = 0;
1468 if (algo >= CO_ALGO_LAST)
1471 be_liveness_assure_chk(be_get_birg_liveness(cenv->birg));
1473 co = new_copy_opt(cenv, cost_func);
1474 co_build_ou_structure(co);
1475 co_build_graph_structure(co);
1477 co_complete_stats(co, &before);
1479 be_stat_ev_ull("co_aff_nodes", before.aff_nodes);
1480 be_stat_ev_ull("co_aff_edges", before.aff_edges);
1481 be_stat_ev_ull("co_max_costs", before.max_costs);
1482 be_stat_ev_ull("co_inevit_costs", before.inevit_costs);
1483 be_stat_ev_ull("co_aff_int", before.aff_int);
1485 be_stat_ev_ull("co_init_costs", before.costs);
1486 be_stat_ev_ull("co_init_unsat", before.unsatisfied_edges);
1488 /* Dump the interference graph in Appel's format. */
1489 if (dump_flags & DUMP_APPEL) {
1490 FILE *f = my_open(cenv, "", ".apl");
1491 co_dump_appel_graph(co, f);
1495 if (dump_flags & DUMP_BEFORE) {
1496 FILE *f = my_open(cenv, "", "-before.dot");
1497 co_dump_ifg_dot(co, f, style_flags);
1501 /* if the algo can improve results, provide an initial solution with heur3 */
1502 if (improve && algos[algo].can_improve_existing) {
1503 co_complete_stats_t stats;
1505 /* produce a heuristic solution */
1507 co_solve_heuristic_java(co);
1509 co_solve_heuristic(co);
1510 #endif /* WITH_JVM */
1512 /* do the stats and provide the current costs */
1513 co_complete_stats(co, &stats);
1514 be_stat_ev_ull("co_prepare_costs", stats.costs);
1518 /* start the JVM here so that it does not tamper the timing. */
1519 if (algo == CO_ALGO_HEUR3)
1520 be_java_coal_start_jvm();
1521 #endif /* WITH_JVM */
1523 algo_func = algos[algo].algo;
1525 /* perform actual copy minimization */
1526 lc_timer_reset_and_start(timer);
1527 was_optimal = algo_func(co);
1528 lc_timer_stop(timer);
1530 be_stat_ev("co_time", lc_timer_elapsed_msec(timer));
1531 be_stat_ev_ull("co_optimal", was_optimal);
1533 if (dump_flags & DUMP_AFTER) {
1534 FILE *f = my_open(cenv, "", "-after.dot");
1535 co_dump_ifg_dot(co, f, style_flags);
1539 co_complete_stats(co, &after);
1542 ulong64 optimizable_costs = after.max_costs - after.inevit_costs;
1543 ulong64 evitable = after.costs - after.inevit_costs;
1545 ir_printf("%30F ", cenv->irg);
1546 printf("%10s %10" ULL_FMT "%10" ULL_FMT "%10" ULL_FMT, cenv->cls->name, after.max_costs, before.costs, after.inevit_costs);
1548 if(optimizable_costs > 0)
1549 printf("%10" ULL_FMT " %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs);
1551 printf("%10" ULL_FMT " %5s\n", after.costs, "-");
1554 be_stat_ev_ull("co_after_costs", after.costs);
1555 be_stat_ev_ull("co_after_unsat", after.unsatisfied_edges);
1557 co_free_graph_structure(co);
1558 co_free_ou_structure(co);