added used_x87 flag, so the x87 simulator runs only if fp
[libfirm] / ir / be / becopyopt.c
index cc66394..1d86d6a 100644 (file)
 /**
- * @author Daniel Grund
- * @date 12.04.2005
+ * Author:      Daniel Grund
+ * Date:               12.04.2005
+ * Copyright:   (c) Universitaet Karlsruhe
+ * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
  */
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#ifdef HAVE_ALLOCA_H
+#include <alloca.h>
+#endif
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
 
-#include "becopyopt.h"
+#include "xmalloc.h"
+#include "debug.h"
+#include "pmap.h"
+#include "irgraph.h"
+#include "irgwalk.h"
+#include "irprog.h"
+#include "irloop_t.h"
+#include "iredges_t.h"
+#include "phiclass.h"
+
+#include "bearch.h"
+#include "beutil.h"
+#include "beifg_t.h"
+#include "becopyopt_t.h"
 #include "becopystat.h"
 
-#define DEBUG_LVL 0 //SET_LEVEL_1
+
+#undef QUICK_AND_DIRTY_HACK
+
+/******************************************************************************
+    _____                           _
+   / ____|                         | |
+  | |  __  ___ _ __   ___ _ __ __ _| |
+  | | |_ |/ _ \ '_ \ / _ \ '__/ _` | |
+  | |__| |  __/ | | |  __/ | | (_| | |
+   \_____|\___|_| |_|\___|_|  \__,_|_|
+
+ ******************************************************************************/
+
 static firm_dbg_module_t *dbg = NULL;
 
-#define is_curr_reg_class(irn) (co->isa->get_irn_reg_class(irn)==co->cls)
-#define is_optimizable(irn) (is_Phi(irn) || is_Copy(irn))
+void be_copy_opt_init(void) {
+}
+
+copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, int (*get_costs)(ir_node*, ir_node*, int)) {
+       const char *s1, *s2, *s3;
+       int len;
+       copy_opt_t *co;
+
+       FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
+
+       co = xcalloc(1, sizeof(*co));
+       co->cenv      = chordal_env;
+       co->aenv      = chordal_env->birg->main_env->arch_env;
+       co->irg       = chordal_env->irg;
+       co->cls       = chordal_env->cls;
+       co->get_costs = get_costs;
+
+       s1 = get_irp_prog_name();
+       s2 = get_entity_name(get_irg_entity(co->irg));
+       s3 = chordal_env->cls->name;
+       len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
+       co->name = xmalloc(len);
+       snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
+
+       return co;
+}
+
+void free_copy_opt(copy_opt_t *co) {
+       xfree(co->name);
+       free(co);
+}
+
+int co_is_optimizable_root(const copy_opt_t *co, ir_node *irn) {
+       arch_register_req_t req;
+       const arch_register_t *reg;
+
+       if (arch_irn_is(co->aenv, irn, ignore))
+               return 0;
+
+       reg = arch_get_irn_register(co->aenv, irn);
+       if (arch_register_type_is(reg, ignore))
+               return 0;
+
+       if (is_Reg_Phi(irn) || is_Perm_Proj(co->aenv, irn) || is_2addr_code(co->aenv, irn, &req))
+               return 1;
+
+       return 0;
+}
+
+int co_is_optimizable_arg(const copy_opt_t *co, ir_node *irn) {
+       const ir_edge_t *edge;
+       const arch_register_t *reg;
+
+       assert(0 && "Is buggy and obsolete. Do not use");
+
+       if (arch_irn_is(co->aenv, irn, ignore))
+               return 0;
+
+       reg = arch_get_irn_register(co->aenv, irn);
+       if (arch_register_type_is(reg, ignore))
+               return 0;
+
+       foreach_out_edge(irn, edge) {
+               ir_node *n = edge->src;
+
+               if (!nodes_interfere(co->cenv, irn, n) || irn == n) {
+                       arch_register_req_t req;
+                       arch_get_register_req(co->aenv, &req, n, -1);
+
+                       if(is_Reg_Phi(n) ||
+                          is_Perm(co->aenv, n) ||
+                          (arch_register_req_is(&req, should_be_same) && req.other_same == irn)
+                         )
+                               return 1;
+               }
+       }
+
+       return 0;
+}
+
+int co_get_costs_loop_depth(ir_node *root, ir_node* arg, int pos) {
+       int cost = 0;
+       ir_loop *loop;
+       ir_node *root_block = get_nodes_block(root);
+
+       if (is_Phi(root)) {
+               /* for phis the copies are placed in the corresponding pred-block */
+               loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos));
+       } else {
+               /* a perm places the copy in the same block as it resides */
+               loop = get_irn_loop(root_block);
+       }
+       if (loop) {
+               int d = get_loop_depth(loop);
+               cost = d*d;
+       }
+       return cost+1;
+}
+
+int co_get_costs_all_one(ir_node *root, ir_node* arg, int pos) {
+       return 1;
+}
+
+/******************************************************************************
+   ____        _   _    _       _ _          _____ _
+  / __ \      | | | |  | |     (_) |        / ____| |
+ | |  | |_ __ | |_| |  | |_ __  _| |_ ___  | (___ | |_ ___  _ __ __ _  __ _  ___
+ | |  | | '_ \| __| |  | | '_ \| | __/ __|  \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
+ | |__| | |_) | |_| |__| | | | | | |_\__ \  ____) | || (_) | | | (_| | (_| |  __/
+  \____/| .__/ \__|\____/|_| |_|_|\__|___/ |_____/ \__\___/|_|  \__,_|\__, |\___|
+        | |                                                            __/ |
+        |_|                                                           |___/
+ ******************************************************************************/
 
 /**
- * Builds an optimization unit for a given optimizable irn (root).
- * This opt-unit is inserted in the main structure co.
- * If an arg of root itself is optimizable process this arg before with a
- * recursive call. For handling this situation and loops co->root is used
- * to remember all roots.
+ * Determines a maximum weighted independent set with respect to
+ * the interference and conflict edges of all nodes in a qnode.
  */
-static void co_append_unit(copy_opt_t *co, const ir_node *root) {
-       int i, arity;
+static int ou_max_ind_set_costs(unit_t *ou) {
+       be_chordal_env_t *chordal_env = ou->co->cenv;
+       ir_node **safe, **unsafe;
+       int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
+       bitset_t *curr;
+       int max, pos, curr_weight, best_weight = 0;
+
+       /* assign the nodes into two groups.
+        * safe: node has no interference, hence it is in every max stable set.
+        * unsafe: node has an interference
+        */
+       safe = alloca((ou->node_count-1) * sizeof(*safe));
+       safe_costs = 0;
+       safe_count = 0;
+       unsafe = alloca((ou->node_count-1) * sizeof(*unsafe));
+       unsafe_costs = alloca((ou->node_count-1) * sizeof(*unsafe_costs));
+       unsafe_count = 0;
+       for(i=1; i<ou->node_count; ++i) {
+               int is_safe = 1;
+               for(o=1; o<ou->node_count; ++o) {
+                       if (i==o)
+                               continue;
+                       if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) {
+                               unsafe_costs[unsafe_count] = ou->costs[i];
+                               unsafe[unsafe_count] = ou->nodes[i];
+                               ++unsafe_count;
+                               is_safe = 0;
+                               break;
+                       }
+               }
+               if (is_safe) {
+                       safe_costs += ou->costs[i];
+                       safe[safe_count++] = ou->nodes[i];
+               }
+       }
+
+
+       /* now compute the best set out of the unsafe nodes*/
+       if (unsafe_count > MIS_HEUR_TRIGGER) {
+               bitset_t *best = bitset_alloca(unsafe_count);
+               /* Heuristik: Greedy trial and error form index 0 to unsafe_count-1 */
+               for (i=0; i<unsafe_count; ++i) {
+                       bitset_set(best, i);
+                       /* check if it is a stable set */
+                       for (o=bitset_next_set(best, 0); o!=-1 && o<i; o=bitset_next_set(best, o+1))
+                               if (nodes_interfere(chordal_env, unsafe[i], unsafe[o])) {
+                                       bitset_clear(best, i); /* clear the bit and try next one */
+                                       break;
+                               }
+               }
+               /* compute the weight */
+               bitset_foreach(best, pos)
+                       best_weight += unsafe_costs[pos];
+       } else {
+               /* Exact Algorithm: Brute force */
+               curr = bitset_alloca(unsafe_count);
+               bitset_set_all(curr);
+               while ((max = bitset_popcnt(curr)) != 0) {
+                       /* check if curr is a stable set */
+                       for (i=bitset_next_set(curr, 0); i!=-1; i=bitset_next_set(curr, i+1))
+                               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) */
+                                               if (nodes_interfere(chordal_env, unsafe[i], unsafe[o]))
+                                                       goto no_stable_set;
+
+                       /* if we arrive here, we have a stable set */
+                       /* compute the weigth of the stable set*/
+                       curr_weight = 0;
+                       bitset_foreach(curr, pos)
+                               curr_weight += unsafe_costs[pos];
+
+                       /* any better ? */
+                       if (curr_weight > best_weight) {
+                               best_weight = curr_weight;
+                       }
+
+       no_stable_set:
+                       bitset_minus1(curr);
+               }
+       }
+
+       return safe_costs+best_weight;
+}
+
+static void co_collect_units(ir_node *irn, void *env) {
+       copy_opt_t *co = env;
        unit_t *unit;
-       DBG((dbg, LEVEL_1, "\t  Root: %n\n", root));
-       /* check if we encountered this root earlier */
-       if (pset_find_ptr(co->roots, root))
-               return;
-       pset_insert_ptr(co->roots, root);
+       arch_register_req_t req;
 
-       assert(is_curr_reg_class(root) && "node is in wrong register class!");
+       if (!is_curr_reg_class(co, irn))
+               return;
+       if (!co_is_optimizable_root(co, irn))
+               return;
 
-       /* init unit */
-       arity = get_irn_arity(root);
-       unit = calloc(1, sizeof(*unit));
+       /* Init a new unit */
+       unit = xcalloc(1, sizeof(*unit));
        unit->co = co;
-       unit->interf = 0;
        unit->node_count = 1;
-       unit->nodes = malloc((arity+1) * sizeof(*unit->nodes));
-       unit->nodes[0] = root;
        INIT_LIST_HEAD(&unit->queue);
 
-       /* check all args */
-       for (i=0; i<arity; ++i) {
-               ir_node *arg = get_irn_n(root, i);
-               assert(is_curr_reg_class(arg) && "Argument not in same register class.");
-               if (arg != root) {
-                       if (!values_interfere(root, arg)) {
-                               DBG((dbg, LEVEL_1, "\t  Member: %n\n", arg));
-                               if (is_optimizable(arg))
-                                       co_append_unit(co, arg);
-                               unit->nodes[unit->node_count++] = arg;
-                       } else
-                               unit->interf++;
+       /* Phi with some/all of its arguments */
+       if (is_Reg_Phi(irn)) {
+               int i, arity;
+
+               /* init */
+               arity = get_irn_arity(irn);
+               unit->nodes = xmalloc((arity+1) * sizeof(*unit->nodes));
+               unit->costs = xmalloc((arity+1) * sizeof(*unit->costs));
+               unit->nodes[0] = irn;
+
+               /* fill */
+               for (i=0; i<arity; ++i) {
+                       int o, arg_pos;
+                       ir_node *arg = get_irn_n(irn, i);
+
+                       assert(is_curr_reg_class(co, arg) && "Argument not in same register class.");
+                       if (arg == irn)
+                               continue;
+                       if (nodes_interfere(co->cenv, irn, arg)) {
+                               unit->inevitable_costs += co->get_costs(irn, arg, i);
+                               continue;
+                       }
+
+                       /* Else insert the argument of the phi to the members of this ou */
+                       DBG((dbg, LEVEL_1, "\t   Member: %+F\n", arg));
+
+                       /* Check if arg has occurred at a prior position in the arg/list */
+                       arg_pos = 0;
+                       for (o=0; o<unit->node_count; ++o)
+                               if (unit->nodes[o] == arg) {
+                                       arg_pos = o;
+                                       break;
+                               }
+
+                       if (!arg_pos) { /* a new argument */
+                               /* insert node, set costs */
+                               unit->nodes[unit->node_count] = arg;
+                               unit->costs[unit->node_count] = co->get_costs(irn, arg, i);
+                               unit->node_count++;
+                       } else { /* arg has occured before in same phi */
+                               /* increase costs for existing arg */
+                               unit->costs[arg_pos] += co->get_costs(irn, arg, i);
+                       }
+               }
+               unit->nodes = xrealloc(unit->nodes, unit->node_count * sizeof(*unit->nodes));
+               unit->costs = xrealloc(unit->costs, unit->node_count * sizeof(*unit->costs));
+       } else
+
+       /* Proj of a perm with corresponding arg */
+       if (is_Perm_Proj(co->aenv, irn)) {
+               assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
+               unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
+               unit->costs = xmalloc(2 * sizeof(*unit->costs));
+               unit->node_count = 2;
+               unit->nodes[0] = irn;
+               unit->nodes[1] = get_Perm_src(irn);
+               unit->costs[1] = co->get_costs(irn, unit->nodes[1], -1);
+       } else
+
+       /* Src == Tgt of a 2-addr-code instruction */
+       if (is_2addr_code(co->aenv, irn, &req)) {
+               ir_node *other = req.other_same;
+               if (!nodes_interfere(co->cenv, irn, other)) {
+                       unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
+                       unit->costs = xmalloc(2 * sizeof(*unit->costs));
+                       unit->node_count = 2;
+                       unit->nodes[0] = irn;
+                       unit->nodes[1] = other;
+                       unit->costs[1] = co->get_costs(irn, other, -1);
                }
-       }
-       unit->nodes = realloc(unit->nodes, unit->node_count * sizeof(*unit->nodes));
-       list_add_tail(&unit->units, &co->units);
-       /* Init mis_size to node_count. So get_lower_bound returns correct results.
-        * - Now it can be called even before the heuristic has run.
-        * - And it will return correct results for units with nodecount 1 which are
-        * not optimized during the heuristic and have therefor delivered wrong results for get_lower_bound
-        */
-       unit->mis_size = unit->node_count;
+       } else
+               assert(0 && "This is not an optimizable node!");
 
-}
+       /* Insert the new unit at a position according to its costs */
+       if (unit->node_count > 1) {
+               int i;
+               struct list_head *tmp;
 
-static void co_collect_in_block(ir_node *block, void *env) {
-       copy_opt_t *co = env;
-       struct list_head *head = &get_ra_block_info(block)->border_head;
-       border_t *curr;
+               /* Determine the maximum costs this unit can cause: all_nodes_cost */
+               for(i=1; i<unit->node_count; ++i) {
+                       unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
+                       unit->all_nodes_costs += unit->costs[i];
+               }
 
-       list_for_each_entry_reverse(border_t, curr, head, list)
-               if (curr->is_def && curr->is_real && is_optimizable(curr->irn))
-                       co_append_unit(co, curr->irn);
+               /* Determine the minimal costs this unit will cause: min_nodes_costs */
+               unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
+               /* Insert the new ou according to its sort_key */
+               tmp = &co->units;
+               while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
+                       tmp = tmp->next;
+               list_add(&unit->units, tmp);
+       } else {
+               free(unit);
+       }
 }
 
-static void co_collect_units(copy_opt_t *co) {
-       DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
-       co->roots = pset_new_ptr(64);
-       dom_tree_walk_irg(co->irg, co_collect_in_block, NULL, co);
-       del_pset(co->roots);
+#ifdef QUICK_AND_DIRTY_HACK
+
+static int compare_ous(const void *k1, const void *k2) {
+       const unit_t *u1 = *((const unit_t **) k1);
+       const unit_t *u2 = *((const unit_t **) k2);
+       int i, o, u1_has_constr, u2_has_constr;
+       arch_register_req_t req;
+       const arch_env_t *aenv = u1->co->aenv;
+
+       /* Units with constraints come first */
+       u1_has_constr = 0;
+       for (i=0; i<u1->node_count; ++i) {
+               arch_get_register_req(aenv, &req, u1->nodes[i], -1);
+               if (arch_register_req_is(&req, limited)) {
+                       u1_has_constr = 1;
+                       break;
+               }
+       }
+
+       u2_has_constr = 0;
+       for (i=0; i<u2->node_count; ++i) {
+               arch_get_register_req(aenv, &req, u2->nodes[i], -1);
+               if (arch_register_req_is(&req, limited)) {
+                       u2_has_constr = 1;
+                       break;
+               }
+       }
+
+       if (u1_has_constr != u2_has_constr)
+               return u2_has_constr - u1_has_constr;
+
+       /* Now check, whether the two units are connected */
+#if 0
+       for (i=0; i<u1->node_count; ++i)
+               for (o=0; o<u2->node_count; ++o)
+                       if (u1->nodes[i] == u2->nodes[o])
+                               return 0;
+#endif
+
+       /* After all, the sort key decides. Greater keys come first. */
+       return u2->sort_key - u1->sort_key;
+
 }
 
-copy_opt_t *new_copy_opt(ir_graph *irg, const arch_isa_if_t *isa, const arch_register_class_t *cls) {
-       const char *s1, *s2, *s3;
-       int len;
-       dbg = firm_dbg_register("ir.be.copyopt");
-       firm_dbg_set_mask(dbg, DEBUG_LVL);
+/**
+ * Sort the ou's according to constraints and their sort_key
+ */
+static void co_sort_units(copy_opt_t *co) {
+       int i, count = 0, costs;
+       unit_t *ou, **ous;
 
-       copy_opt_t *co = calloc(1, sizeof(*co));
-       co->irg = irg;
-       co->isa = isa;
-       co->cls = cls;
+       /* get the number of ous, remove them form the list and fill the array */
+       list_for_each_entry(unit_t, ou, &co->units, units)
+               count++;
+       ous = alloca(count * sizeof(*ous));
 
-       s1 = get_irp_prog_name();
-       s2 = get_entity_name(get_irg_entity(co->irg));
-       s3 = cls->name;
-       len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
-       co->name = malloc(len);
-       if (!strcmp(co->name, DEBUG_IRG))
-               firm_dbg_set_mask(dbg, -1);
-       snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
+       costs = co_get_max_copy_costs(co);
+
+       i = 0;
+       list_for_each_entry(unit_t, ou, &co->units, units)
+               ous[i++] = ou;
 
        INIT_LIST_HEAD(&co->units);
-       co_collect_units(co);
-       return co;
+
+       assert(count == i && list_empty(&co->units));
+
+       for (i=0; i<count; ++i)
+               ir_printf("%+F\n", ous[i]->nodes[0]);
+
+       qsort(ous, count, sizeof(*ous), compare_ous);
+
+       ir_printf("\n\n");
+       for (i=0; i<count; ++i)
+               ir_printf("%+F\n", ous[i]->nodes[0]);
+
+       /* reinsert into list in correct order */
+       for (i=0; i<count; ++i)
+               list_add_tail(&ous[i]->units, &co->units);
+
+       assert(costs == co_get_max_copy_costs(co));
 }
+#endif
 
-void free_copy_opt(copy_opt_t *co) {
+void co_build_ou_structure(copy_opt_t *co) {
+       DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
+       INIT_LIST_HEAD(&co->units);
+       irg_walk_graph(co->irg, co_collect_units, NULL, co);
+#ifdef QUICK_AND_DIRTY_HACK
+       co_sort_units(co);
+#endif
+}
+
+void co_free_ou_structure(copy_opt_t *co) {
        unit_t *curr, *tmp;
-       free(co->name);
+       ASSERT_OU_AVAIL(co);
        list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
-               free(curr->nodes);
-               free(curr);
+               xfree(curr->nodes);
+               xfree(curr->costs);
+               xfree(curr);
        }
+       co->units.next = NULL;
 }
 
-int co_get_copy_count(copy_opt_t *co) {
+/* co_solve_heuristic() is implemented in becopyheur.c */
+
+int co_get_max_copy_costs(const copy_opt_t *co) {
        int i, res = 0;
        unit_t *curr;
+
+       ASSERT_OU_AVAIL(co);
+
        list_for_each_entry(unit_t, curr, &co->units, units) {
-               int root_col = get_irn_color(curr->nodes[0]);
-               res += curr->interf;
+               res += curr->inevitable_costs;
                for (i=1; i<curr->node_count; ++i)
-                       if (root_col != get_irn_color(curr->nodes[i]))
-                               res++;
+                       res += curr->costs[i];
        }
        return res;
 }
 
-int co_get_lower_bound(copy_opt_t *co) {
+int co_get_inevit_copy_costs(const copy_opt_t *co) {
        int res = 0;
        unit_t *curr;
+
+       ASSERT_OU_AVAIL(co);
+
        list_for_each_entry(unit_t, curr, &co->units, units)
-               res += curr->interf + curr->node_count - curr->mis_size;
+               res += curr->inevitable_costs;
        return res;
 }
 
-int co_get_interferer_count(copy_opt_t *co) {
+int co_get_copy_costs(const copy_opt_t *co) {
+       int i, res = 0;
+       unit_t *curr;
+
+       ASSERT_OU_AVAIL(co);
+
+       list_for_each_entry(unit_t, curr, &co->units, units) {
+               int root_col = get_irn_col(co, curr->nodes[0]);
+               DBG((dbg, LEVEL_1, "  %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
+               res += curr->inevitable_costs;
+               for (i=1; i<curr->node_count; ++i) {
+                       int arg_col = get_irn_col(co, curr->nodes[i]);
+                       if (root_col != arg_col) {
+                               DBG((dbg, LEVEL_1, "  %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
+                               res += curr->costs[i];
+                       }
+               }
+       }
+       return res;
+}
+
+int co_get_lower_bound(const copy_opt_t *co) {
        int res = 0;
        unit_t *curr;
+
+       ASSERT_OU_AVAIL(co);
+
        list_for_each_entry(unit_t, curr, &co->units, units)
-               res += curr->interf;
+               res += curr->inevitable_costs + curr->min_nodes_costs;
        return res;
 }
 
-/**
- * Needed for result checking
- */
-static void co_collect_for_checker(ir_node *block, void *env) {
-       copy_opt_t *co = env;
-       struct list_head *head = &get_ra_block_info(block)->border_head;
-       border_t *curr;
+/******************************************************************************
+   _____                 _        _____ _
+  / ____|               | |      / ____| |
+ | |  __ _ __ __ _ _ __ | |__   | (___ | |_ ___  _ __ __ _  __ _  ___
+ | | |_ | '__/ _` | '_ \| '_ \   \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
+ | |__| | | | (_| | |_) | | | |  ____) | || (_) | | | (_| | (_| |  __/
+  \_____|_|  \__,_| .__/|_| |_| |_____/ \__\___/|_|  \__,_|\__, |\___|
+                  | |                                       __/ |
+                  |_|                                      |___/
+ ******************************************************************************/
+
+static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) {
+       const affinity_node_t *n1 = k1;
+       const affinity_node_t *n2 = k2;
 
-       list_for_each_entry_reverse(border_t, curr, head, list)
-               if (curr->is_def && curr->is_real && is_curr_reg_class(curr->irn))
-                       obstack_ptr_grow(&co->ob, curr->irn);
+       return (n1->irn != n2->irn);
 }
 
-/**
- * This O(n^2) checker checks, if two ifg-connected nodes have the same color.
- */
-void co_check_allocation(copy_opt_t *co) {
-       ir_node **nodes, *n1, *n2;
-       int i, o;
-
-       obstack_init(&co->ob);
-       dom_tree_walk_irg(co->irg, co_collect_for_checker, NULL, co);
-       obstack_ptr_grow(&co->ob, NULL);
-
-       nodes = (ir_node **) obstack_finish(&co->ob);
-       for (i = 0, n1 = nodes[i]; n1; n1 = nodes[++i]) {
-               assert(! (is_allocatable_irn(n1) && get_irn_color(n1) == NO_COLOR));
-               for (o = i+1, n2 = nodes[o]; n2; n2 = nodes[++o])
-                       if (phi_ops_interfere(n1, n2) && get_irn_color(n1) == get_irn_color(n2)) {
-                               DBG((dbg, 0, "Error: %n in %n  and  %n in %n have the same color.\n", n1, get_nodes_block(n1), n2, get_nodes_block(n2)));
-                               assert(0 && "Interfering values have the same color!");
-                       }
+static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
+       affinity_node_t new_node, *node;
+       neighb_t new_nbr, *nbr;
+       int allocnew;
+
+       new_node.irn        = n1;
+       new_node.degree     = 0;
+       new_node.neighbours = NULL;
+       node = set_insert(co->nodes, &new_node, sizeof(new_node), HASH_PTR(new_node.irn));
+
+       allocnew = 1;
+       for (nbr = node->neighbours; nbr; nbr = nbr->next)
+               if (nbr->irn == n2) {
+                       allocnew = 0;
+                       break;
+               }
+
+       /* if we did not find n2 in n1's neighbourhood insert it */
+       if (allocnew) {
+               obstack_grow(&co->obst, &new_nbr, sizeof(new_nbr));
+               nbr = obstack_finish(&co->obst);
+               nbr->irn   = n2;
+               nbr->costs = 0;
+               nbr->next  = node->neighbours;
+               node->neighbours = nbr;
+               node->degree++;
        }
-       obstack_free(&co->ob, NULL);
-       DBG((dbg, 2, "The checker seems to be happy!\n"));
+
+       /* now nbr points to n1's neighbour-entry of n2 */
+       nbr->costs += costs;
+}
+
+static INLINE void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
+       if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
+               add_edge(co, n1, n2, costs);
+               add_edge(co, n2, n1, costs);
+       }
+}
+
+static void build_graph_walker(ir_node *irn, void *env) {
+       copy_opt_t *co = env;
+       int pos, max;
+       arch_register_req_t req;
+       const arch_register_t *reg;
+
+       if (!is_curr_reg_class(co, irn) || arch_irn_is(co->aenv, irn, ignore))
+               return;
+
+       reg = arch_get_irn_register(co->aenv, irn);
+       if (arch_register_type_is(reg, ignore))
+               return;
+
+       /* Phis */
+       if (is_Reg_Phi(irn))
+               for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
+                       ir_node *arg = get_irn_n(irn, pos);
+                       add_edges(co, irn, arg, co->get_costs(irn, arg, pos));
+               }
+
+       /* Perms */
+       else if (is_Perm_Proj(co->aenv, irn)) {
+               ir_node *arg = get_Perm_src(irn);
+               add_edges(co, irn, arg, co->get_costs(irn, arg, 0));
+       }
+
+       /* 2-address code */
+       else if (is_2addr_code(co->aenv, irn, &req))
+               add_edges(co, irn, req.other_same, co->get_costs(irn, req.other_same, 0));
+}
+
+void co_build_graph_structure(copy_opt_t *co) {
+       obstack_init(&co->obst);
+       co->nodes = new_set(compare_affinity_node_t, 32);
+
+       irg_walk_graph(co->irg, build_graph_walker, NULL, co);
+}
+
+void co_free_graph_structure(copy_opt_t *co) {
+       ASSERT_GS_AVAIL(co);
+
+       del_set(co->nodes);
+       obstack_free(&co->obst, NULL);
+       co->nodes = NULL;
+}
+
+/* co_solve_ilp1() co_solve_ilp2() are implemented in becopyilpX.c */
+
+int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) {
+       affinity_node_t new_node, *n;
+
+       ASSERT_GS_AVAIL(co);
+
+       new_node.irn = irn;
+       n = set_find(co->nodes, &new_node, sizeof(new_node), HASH_PTR(new_node.irn));
+       if (n) {
+               return (n->degree > 0);
+       } else
+               return 0;
 }