Changed many things in copy opt. 1st part of refactoring.
authorDaniel Grund <grund@cs.uni-saarland.de>
Wed, 1 Mar 2006 13:03:10 +0000 (13:03 +0000)
committerDaniel Grund <grund@cs.uni-saarland.de>
Wed, 1 Mar 2006 13:03:10 +0000 (13:03 +0000)
ir/be/becopyheur.c
ir/be/becopyilp.c
ir/be/becopyilp1.c [new file with mode: 0644]
ir/be/becopyilp2.c [new file with mode: 0644]
ir/be/becopyilp_t.h [new file with mode: 0644]
ir/be/becopyopt.c
ir/be/becopyopt.h
ir/be/becopyopt_t.h [new file with mode: 0644]

index 42d4067..3631a40 100644 (file)
@@ -23,8 +23,9 @@
 #include <malloc.h>
 #endif
 
+#include "debug.h"
 #include "xmalloc.h"
-#include "becopyopt.h"
+#include "becopyopt_t.h"
 #include "becopystat.h"
 #include "bitset.h"
 #include "bearch.h"
@@ -104,7 +105,7 @@ static INLINE void qnode_add_conflict(const qnode_t *qn, const ir_node *n1, cons
 static INLINE int qnode_are_conflicting(const qnode_t *qn, const ir_node *n1, const ir_node *n2) {
        conflict_t c;
        /* search for live range interference */
-       if (n1!=n2 && nodes_interfere(qn->ou->co->chordal_env, n1, n2))
+       if (n1!=n2 && nodes_interfere(qn->ou->co->cenv, n1, n2))
                return 1;
        /* search for recoloring conflicts */
        if ((int)n1 < (int)n2) {
@@ -213,9 +214,10 @@ static INLINE void qnode_pin_local(const qnode_t *qn, ir_node *irn) {
  *
  */
 static ir_node *qnode_color_irn(const qnode_t *qn, ir_node *irn, int col, const ir_node *trigger) {
-       const be_chordal_env_t *chordal_env = qn->ou->co->chordal_env;
-       const arch_register_class_t *cls = chordal_env->cls;
-       const arch_env_t *arch_env = chordal_env->main_env->arch_env;
+       copy_opt_t *co = qn->ou->co;
+       const be_chordal_env_t *chordal_env = co->cenv;
+       const arch_register_class_t *cls = co->cls;
+       const arch_env_t *arch_env = co->aenv;
        int irn_col = qnode_get_new_color(qn, irn);
        ir_node *sub_res, *curr;
        be_ifg_t *ifg = chordal_env->ifg;
@@ -507,8 +509,8 @@ static INLINE void ou_insert_qnode(unit_t *ou, qnode_t *qn) {
 static void ou_optimize(unit_t *ou) {
        int i;
        qnode_t *curr = NULL, *tmp;
-       arch_env_t *aenv = get_arch_env(ou->co);
-       const arch_register_class_t *cls = ou->co->chordal_env->cls;
+       const arch_env_t *aenv = ou->co->aenv;
+       const arch_register_class_t *cls = ou->co->cls;
        bitset_t *pos_regs = bitset_alloca(cls->n_regs);
        bitset_t *ign_regs = bitset_alloca(cls->n_regs);
 
@@ -576,7 +578,7 @@ static void ou_optimize(unit_t *ou) {
                free_qnode(curr);
 }
 
-void co_heur_opt(copy_opt_t *co) {
+int co_solve_heuristic(copy_opt_t *co) {
        unit_t *curr;
        dbg = firm_dbg_register("ir.be.copyoptheur");
 
@@ -586,4 +588,6 @@ void co_heur_opt(copy_opt_t *co) {
                        ou_optimize(curr);
 
        del_pset(pinned_global);
+
+       return 0;
 }
index 20bdbc6..67adf36 100644 (file)
 /**
  * Author:      Daniel Grund
- * Date:               17.05.2005
+ * Date:               28.02.2006
  * Copyright:   (c) Universitaet Karlsruhe
  * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
- * CVS-ID:      $Id$
+ *
+ * Common stuff used by all ILP fomulations.
+ *
  */
-#ifdef HAVE_CONFIG_H
-#include "config.h"
-#endif
-#ifdef HAVE_ALLOCA_H
-#include <alloca.h>
-#endif
-#ifdef HAVE_MALLOC_H
-#include <malloc.h>
-#endif
 
-#define PATH_CONSTRAINTS_FOR_CLASSES
-#undef PRECOLOR_MAX_CLIQUE
-#undef NO_NULL_COLORS
-#undef NO_NULL_COLORS_EXTRA_CSTS
-#undef NO_NULL_COLORS_WITH_COSTS
-#if (defined(NO_NULL_COLORS_EXTRA_CSTS) || defined(NO_NULL_COLORS_WITH_COSTS)) && !defined(NO_NULL_COLORS)
-#error Chose your weapon!
-#endif
-
-#include "irprog.h"
-
-#include <lpp/lpp.h>
-#include <lpp/lpp_net.h>
-#include <lpp/lpp_cplex.h>
-#include <lpp/lpp_remote.h>
-#include "xmalloc.h"
-#include "pset.h"
-#include "irdom_t.h"
-#include "iredges_t.h"
-#include "bechordal_t.h"
-#include "becopyopt.h"
-#include "becopystat.h"
-#include "besched_t.h"
-#include "phiclass.h"
-
-#define LPP_HOST "i44pc52"
-#define LPP_SOLVER "cplex"
-
-#undef DUMP_MPS
-static firm_dbg_module_t *dbg = NULL;
-
-#define MAX(a,b) ((a<b)?(b):(a))
-#define MIN(a,b) ((a<b)?(a):(b))
-#define EPSILON 0.00001
-#define SLOTS_LIVING 32
-
-typedef struct _simpl_t {
-       struct list_head chain;
-       ir_node *irn;
-} simpl_t;
-
-typedef struct _problem_instance_t {
-       const copy_opt_t *co;                   /** the copy_opt problem */
-       /* problem size reduction removing simple nodes */
-       struct list_head simplicials;   /**< holds all simpl_t's in right order to color*/
-       pset *removed;                                  /**< holds all removed simplicial irns */
-       /* lp problem */
-       lpp_t *curr_lp;                                 /**< points to the problem currently used */
-       lpp_t *dilp;                                    /**< problem formulation directly as milp */
-#ifdef NO_NULL_COLORS_EXTRA_CSTS
-       int first_nnc_cst_idx;                  /**< the first index of a constraint belonging to no-null-colors stuff*/
-#endif
-       int first_nnc_var_idx;                  /**< the first index of a constraint belonging to no-null-colors stuff*/
-
-       int cst_counter, first_x_var, last_x_var;
-       char buf[32];
-       int all_simplicial;
-       pset *done;
-} problem_instance_t;
-
-#define is_removed(irn) pset_find_ptr(pi->removed, irn)
-
-#define is_color_possible(irn,color) arch_reg_is_allocatable(get_arch_env(pi->co), irn, -1, arch_register_for_index(pi->co->chordal_env->cls, color))
-
-/*
- * Some stuff for variable name handling.
- */
-#define mangle_cst(buf, prefix, nr) \
-                       snprintf((buf), sizeof(buf), "%c%d", (prefix), (nr))
-
-#define mangle_var1(buf, prefix, color) \
-                       snprintf((buf), sizeof(buf), "%c%d", (prefix), (color))
-
-#define mangle_var2(buf, prefix, node_nr, color) \
-                       snprintf((buf), sizeof(buf), "%c%d_%d", (prefix), (node_nr), (color))
-
-#define mangle_var3(buf, prefix, n1, n2, col) \
-                       snprintf((buf), sizeof(buf), "%c%d_%d_%d", (prefix), (n1), (n2), (col))
-
-#define mangle_var_irn(buf, prefix, irn, color) \
-                       mangle_var2((buf), (prefix), get_irn_graph_nr(irn), (color))
-
-#define split_var(var, nnr, col) \
-                       sscanf(var, "x%d_%d", (nnr), (col))
-
-
-/**
- * Checks if a node is simplicial in the graph
- * heeding the already removed nodes.
- */
-static INLINE int pi_is_simplicial(problem_instance_t *pi, const ir_node *ifn) {
-       int i, o, size = 0;
-       ir_node **all, *curr;
-       be_ifg_t *ifg = pi->co->chordal_env->ifg;
-       void *iter = be_ifg_neighbours_iter_alloca(ifg);
-
-       all = alloca(be_ifg_degree(ifg, ifn) * sizeof(*all));
-
-       /* get all non-removed neighbors */
-       be_ifg_foreach_neighbour(ifg, iter, ifn, curr)
-               if (!is_removed(curr))
-                       all[size++] = curr;
-
-       /* check if these form a clique */
-       for (i=0; i<size; ++i)
-               for (o=i+1; o<size; ++o)
-                       if (!be_ifg_connected(ifg, all[i], all[o]))
-                               return 0;
-
-       /* all edges exist so this is a clique */
-       return 1;
-}
-
-static int irn_cmp(const void *a, const void *b, size_t n)
-{
-       return a != b;
-}
-
-/**
- * Iterative finds and 'removes' from the graph all nodes which are
- * simplicial AND not member of a equal-color-wish
- */
-static void pi_find_simplicials(problem_instance_t *pi) {
-       ir_node *irn;
-       int redo = 1;
-       int n_nodes = 0;
-       const be_ifg_t *ifg = pi->co->chordal_env->ifg;
-       void *iter = be_ifg_neighbours_iter_alloca(ifg);
-
-       DBG((dbg, LEVEL_2, "Find simlicials...\n"));
-
-       while (redo) {
-               arch_register_req_t req;
-               redo = 0;
-               be_ifg_foreach_node(ifg, iter, irn) {
-                       if (!is_removed(irn) && !is_optimizable(get_arch_env(pi->co), irn, &req) && !is_optimizable_arg(pi->co, irn)) {
-                       if (pi_is_simplicial(pi, irn)) {
-                                       simpl_t *s = xmalloc(sizeof(*s));
-                                       s->irn = irn;
-                                       list_add(&s->chain, &pi->simplicials);
-                                       pset_insert_ptr(pi->removed, irn);
-                                       redo = 1;
-                                       DBG((dbg, LEVEL_2, " Removed %+F\n", irn));
-                       }
-                       }
-               }
-       }
-
-       /* TODO: Count inside the last look */
-       be_ifg_foreach_node(ifg, iter, irn) {
-               n_nodes++;
-       }
-
-       if (n_nodes == pset_count(pi->removed))
-               pi->all_simplicial = 1;
-}
-
-#ifdef NO_NULL_COLORS
-static void pi_add_constr_no_null_colors(problem_instance_t *pi) {
-       int cst_counter=0, col, var_idx, cst_idx;
-       int n_colors = pi->co->chordal_env->cls->n_regs;
-       char buf[40];
-
-       for (col = 0; col < n_colors; ++col) {
-               mangle_var1(buf, 'u', col);
-#ifdef NO_NULL_COLORS_WITH_COSTS
-               var_idx = lpp_add_var(pi->curr_lp, buf, lpp_binary, 1.0 / (double) (1 << (col+1)) );
-#else
-               var_idx = lpp_add_var(pi->curr_lp, buf, lpp_binary, 1.0 / (2.0 * n_colors) );
-#endif
-               if (!pi->first_nnc_var_idx)
-                       pi->first_nnc_var_idx = var_idx;
-       }
-
-#ifdef NO_NULL_COLORS_EXTRA_CSTS
-       for (col = 0; col < n_colors; ++col) {
-               mangle_cst(buf, 'U', cst_counter++);
-               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 0);
-               if (!pi->first_nnc_cst_idx)
-                       pi->first_nnc_cst_idx = cst_idx;
-               lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+col, -1);
-       }
-#endif
-
-#ifndef NO_NULL_COLORS_WITH_COSTS
-       for (col = 0; col < n_colors - 1; ++col) {
-               mangle_cst(buf, 'U', cst_counter++);
-               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 0);
-               lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+col  ,  1);
-               lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+col+1, -1);
-       }
-#endif
-
-}
-#endif
-
-/**
- * Add coloring-force conditions
- * Matrix A: knapsack constraint for each node
- */
-static void pi_add_constr_A(problem_instance_t *pi) {
-       pmap_entry *pme;
-
-       DBG((dbg, LEVEL_2, "Add A constraints...\n"));
-       /* iterate over all blocks */
-       pmap_foreach(pi->co->chordal_env->border_heads, pme) {
-               struct list_head *head = pme->value;
-               border_t *curr;
-               bitset_t *pos_regs = bitset_alloca(pi->co->chordal_env->cls->n_regs);
-
-               list_for_each_entry_reverse(border_t, curr, head, list)
-                       if (curr->is_def && curr->is_real && !is_removed(curr->irn)) {
-                               int cst_idx, nnr, col;
-
-                               nnr = get_irn_graph_nr(curr->irn);
-                               mangle_cst(pi->buf, 'A', nnr);
-                               cst_idx = lpp_add_cst(pi->curr_lp, pi->buf, lpp_equal, 1);
-
-                               /* iterate over all possible colors in order */
-                               bitset_clear_all(pos_regs);
-                               arch_get_allocatable_regs(get_arch_env(pi->co), curr->irn, -1, pos_regs);
-                               bitset_foreach(pos_regs, col) {
-                                       int var_idx;
-                                       mangle_var2(pi->buf, 'x', nnr, col);
-                                       var_idx = lpp_add_var(pi->curr_lp, pi->buf, lpp_binary, 0);
-                                       if (!pi->first_x_var)
-                                               pi->first_x_var = var_idx;
-                                       pi->last_x_var = var_idx;
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
-#ifdef NO_NULL_COLORS_EXTRA_CSTS
-                                       lpp_set_factor_fast(pi->curr_lp, pi->first_nnc_cst_idx+col, var_idx, 1);
-#endif
-                               }
-                       }
-       }
-}
-
-/**
- * Checks if all nodes in @p living are live in in block @p block.
- * @return 1 if all are live in
- *         0 else
- */
-static INLINE int all_live_in(ir_node *block, pset *living) {
-       ir_node *n;
-       for (n = pset_first(living); n; n = pset_next(living))
-               if (!is_live_in(block, n)) {
-                       pset_break(living);
-                       return 0;
-               }
-       return 1;
-}
-
-/**
- * Finds cliques in the interference graph, considering only nodes
- * for which the color @p color is possible. Finds only 'maximal-cliques',
- * viz cliques which are not contained in another one.
- * Matrix B: interference constraints using cliques
- */
-static void pi_add_constr_B(problem_instance_t *pi, int color) {
-       enum phase_t {growing, shrinking} phase = growing;
-       border_t *b;
-       pmap_entry *pme;
-       pset *living = pset_new_ptr(SLOTS_LIVING);
-
-       DBG((dbg, LEVEL_2, "Add B constraints (col = %d)...\n", color));
-       /* iterate over all blocks */
-       pmap_foreach(pi->co->chordal_env->border_heads, pme) {
-               ir_node *block = pme->key;
-               struct list_head *head = pme->value;
-
-               list_for_each_entry_reverse(border_t, b, head, list) {
-                       const ir_node *irn = b->irn;
-                       if (is_removed(irn) || !is_color_possible(irn, color))
-                               continue;
-
-                       if (b->is_def) {
-                               DBG((dbg, LEVEL_2, "Def %n\n", irn));
-                               pset_insert_ptr(living, irn);
-                               phase = growing;
-                       } else { /* is_use */
-                               DBG((dbg, LEVEL_2, "Use %n\n", irn));
-
-                               /* before shrinking the set, store the current 'maximum' clique;
-                                * do NOT if clique is a single node
-                                * do NOT if all values are live_in (in this case they were contained in a live-out clique elsewhere) */
-                               if (phase == growing && pset_count(living) >= 2 && !all_live_in(block, living)) {
-                                       int cst_idx;
-                                       ir_node *n;
-                                       mangle_cst(pi->buf, 'B', pi->cst_counter);
-#ifdef NO_NULL_COLORS
-                                       cst_idx = lpp_add_cst(pi->curr_lp, pi->buf, lpp_less, 0);
-#else
-                                       cst_idx = lpp_add_cst(pi->curr_lp, pi->buf, lpp_less, 1);
-#endif
-                                       for (n = pset_first(living); n; n = pset_next(living)) {
-                                               int var_idx;
-                                               mangle_var_irn(pi->buf, 'x', n, color);
-                                               var_idx = lpp_get_var_idx(pi->curr_lp, pi->buf);
-                                               lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
-                                       }
-#ifdef NO_NULL_COLORS
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+color, -1.0);
-#endif
-                                       pi->cst_counter++;
-                               }
-                               pset_remove_ptr(living, irn);
-                               phase = shrinking;
-                       }
-               }
-       }
-       assert(0 == pset_count(living));
-       del_pset(living);
-}
-
-/**
- * Generates constraints which interrelate x with y variables.
- * x1 and x2 have the different colors ==> y_12 = 1
- */
-static void pi_add_constr_E(problem_instance_t *pi) {
-       unit_t *curr;
-       bitset_t *root_regs, *arg_regs, *work_regs;
-       int cst_counter = 0;
-       unsigned nregs = pi->co->chordal_env->cls->n_regs;
-       root_regs = bitset_alloca(nregs);
-       arg_regs = bitset_alloca(nregs);
-       work_regs = bitset_alloca(nregs);
-
-       DBG((dbg, LEVEL_2, "Add E constraints...\n"));
-       /* for all roots of optimization units */
-       list_for_each_entry(unit_t, curr, &pi->co->units, units) {
-               ir_node *root, *arg;
-               int rootnr, argnr, color;
-               int y_idx, i;
-               char buf[32];
-
-               root = curr->nodes[0];
-               rootnr = get_irn_graph_nr(root);
-               bitset_clear_all(root_regs);
-               arch_get_allocatable_regs(get_arch_env(pi->co), root, -1, root_regs);
-
-               /* for all arguments of root */
-               for (i = 1; i < curr->node_count; ++i) {
-                       arg = curr->nodes[i];
-                       argnr = get_irn_graph_nr(arg);
-                       bitset_clear_all(arg_regs);
-                       arch_get_allocatable_regs(get_arch_env(pi->co), arg, -1, arg_regs);
-
-                       /* Introduce new variable and set factor in objective function */
-                       mangle_var2(buf, 'y', rootnr, argnr);
-                       y_idx = lpp_add_var(pi->curr_lp, buf, lpp_binary, curr->costs[i]);
-
-                       /* set starting value */
-                       lpp_set_start_value(pi->curr_lp, y_idx, (get_irn_col(pi->co, root) != get_irn_col(pi->co, arg)));
-
-                       /* For all colors root and arg have in common, add 2 constraints to E */
-                       bitset_copy(work_regs, root_regs);
-                       bitset_and(work_regs, arg_regs);
-                       bitset_foreach(work_regs, color) {
-                               int root_idx, arg_idx, cst_idx;
-                               mangle_var2(buf, 'x', rootnr, color);
-                               root_idx = lpp_get_var_idx(pi->curr_lp, buf);
-                               mangle_var2(buf, 'x', argnr, color);
-                               arg_idx = lpp_get_var_idx(pi->curr_lp, buf);
-
-                               /* add root-arg-y <= 0 */
-                               mangle_cst(buf, 'E', cst_counter++);
-                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_less, 0);
-                               lpp_set_factor_fast(pi->curr_lp, cst_idx, root_idx, 1);
-                               lpp_set_factor_fast(pi->curr_lp, cst_idx, arg_idx, -1);
-                               lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
-
-                               /* add arg-root-y <= 0 */
-                               mangle_cst(buf, 'E', cst_counter++);
-                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_less, 0);
-                               lpp_set_factor_fast(pi->curr_lp, cst_idx, root_idx, -1);
-                               lpp_set_factor_fast(pi->curr_lp, cst_idx, arg_idx, 1);
-                               lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
-                       }
-                       /* For all colors root and arg have "disjunct", add 1 constraints to E.
-                        * If root gets a color the arg is not possible to get then they will
-                        * definetly get different colors. So y has to be 1.
-                        * Vice versa for arg.
-                        */
-                       bitset_copy(work_regs, root_regs);
-                       bitset_xor(work_regs, arg_regs);
-                       bitset_foreach(work_regs, color) {
-                               int root_idx, arg_idx, cst_idx;
-                               mangle_var2(buf, 'x', rootnr, color);
-                               root_idx = lpp_get_var_idx(pi->curr_lp, buf);
-                               mangle_var2(buf, 'x', argnr, color);
-                               arg_idx = lpp_get_var_idx(pi->curr_lp, buf);
-
-                               mangle_cst(buf, 'E', cst_counter++);
-                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_less, 0);
-                               if (bitset_is_set(root_regs, color)) {
-                                       /* add root-y <= 0 */
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, root_idx, 1);
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
-                               } else {
-                                       assert(bitset_is_set(arg_regs, color) && "bitset_xor is buggy");
-                                       /* add arg-y <= 0 */
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, arg_idx, 1);
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
-                               }
-                       }
-               }
-       }
-}
-
-static INLINE int get_costs(problem_instance_t *pi, ir_node *phi, ir_node *irn) {
-       int i;
-       unit_t *curr;
-       /* search optimization unit for phi */
-       list_for_each_entry(unit_t, curr, &pi->co->units, units)
-               if (curr->nodes[0] == phi) {
-                       for (i=1; i<curr->node_count; ++i)
-                               if (curr->nodes[i] == irn)
-                                       return curr->costs[i];
-                       assert(0 && "irn must occur in this ou");
-               }
-       assert(0 && "phi must be found in a ou");
-       return 0;
-}
-
-static void clique_path_walker(ir_node *block, void *env) {
-       problem_instance_t *pi = env;
-       int count, arity, row, col, other_row, *costs;
-       ir_node **phis, *phi, *irn, **phi_matrix;
-       pset *done;
-       bitset_t *candidates;
-
-       /* Count all phi nodes of this block */
-       for (count=0, irn = sched_first(block); is_Phi(irn); irn = sched_next(irn))
-               count++;
-
-       /* We at least 2 phi nodes for this class of inequalities */
-       if (count < 2)
-               return;
-
-       /* Build the \Phi-Matrix */
-       arity = get_irn_arity(sched_first(block));
-       phis = alloca(count * sizeof(*phis));
-       costs = alloca(count * sizeof(costs));
-       phi_matrix = alloca(count*arity * sizeof(*phi_matrix));
-       candidates = bitset_alloca(count);
-
-       phi = sched_first(block);
-       for (row=0; row<count; ++row) {
-               phis[row] = phi;
-               for (col=0; col<arity; ++col) {
-                       ir_node *arg = get_irn_n(phi, col);
-                       /* Sort out all arguments interfering with its phi */
-                       if (nodes_interfere(pi->co->chordal_env, phi, arg)) {
-                               phi_matrix[row*arity + col] =  NULL;
-                       } else
-                               phi_matrix[row*arity + col] =  arg;
-               }
-               phi = sched_next(phi);
-       }
-
-       /* Now find the interesting patterns in the matrix:
-        * All nodes which are used at least twice in a column. */
-       /* columnwise ... */
-       for (col=0; col<arity; ++col) {
-               done = pset_new_ptr_default();
-               for (row=0; row<count; ++row) {
-                       irn = phi_matrix[row*arity + col];
-                       /*
-                        * is this an interfering arg (NULL)
-                        * or has the irn already been processed in this col?
-                        */
-                       if (!irn || pset_find_ptr(done, irn))
-                               continue;
-                       else
-                               pset_insert_ptr(done, irn);
-
-                       /* insert irn in candidates */
-                       bitset_clear_all(candidates);
-                       bitset_set(candidates, row);
-                       /* search the irn in the rows below */
-                       for (other_row = row+1; other_row<count; ++other_row)
-                               if (irn == phi_matrix[other_row*arity + col]) {
-                                       /* found the irn in the same col in another row */
-                                       bitset_set(candidates, other_row);
-                               }
-
-                       /* now we know all occurences of irn in this col */
-                       if (bitset_popcnt(candidates) < 2)
-                               continue;
-
-                       /* generate an unequation finally.
-                        * phis are indexed in the bitset,
-                        * shared argument is irn
-                        * rhs is phi_count - 1 */
-                       {
-                               char buf[32];
-                               ir_node *root;
-                               int pos, irnnr, rootnr, cst_idx, y_idx, cst_counter = 0;
-                               int minimal_unequal_count = bitset_popcnt(candidates)-1;
-
-                               irnnr = get_irn_graph_nr(irn);
-                               mangle_cst(buf, 'M', cst_counter++);
-                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, minimal_unequal_count);
-
-                               /* for all phis */
-                               bitset_foreach(candidates, pos) {
-                                       root = phis[pos];
-                                       rootnr = get_irn_graph_nr(root);
-                                       mangle_var2(buf, 'y', rootnr, irnnr);
-                                       y_idx = lpp_get_var_idx(pi->curr_lp, buf);
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, 1);
-                               }
-                       }
-               }
-               del_pset(done); /* clear set for next row */
-       } /*next col*/
-}
-
-/**
- * Matrix M: Multi-Arg-Use. Interrelates different \phi-functions
- * in the same block, iff they use the same arg at the same pos.
- * Only one of the phis can get the arg.
- */
-static void pi_add_clique_path_cstr(problem_instance_t *pi) {
-       DBG((dbg, LEVEL_2, "Adding clique path constraints...\n"));
-       dom_tree_walk_irg(get_irg(pi->co), clique_path_walker, NULL, pi);
-}
-
-#ifndef PATH_CONSTRAINTS_FOR_CLASSES
-/**
- * Matrix P: Path contraints.
- * If 2 nodes interfere and there is a path of equal-color-edges
- * connecting them, then at least one of those equal-color-edges
- * will break and cause some costs.
- */
-static void pi_add_path_cstr(problem_instance_t *pi) {
-       unit_t *curr;
-       int cst_counter = 0;
-       DBG((dbg, LEVEL_2, "Adding path constraints...\n"));
-
-       /* for all optimization units (only phis) */
-       list_for_each_entry(unit_t, curr, &pi->co->units, units) {
-               int i, o, rootnr;
-
-               if (curr->min_nodes_costs == 0)
-                       continue;
-
-               rootnr = get_irn_graph_nr(curr->nodes[0]);
-               /* check all argument pairs for interference */
-               for (i=1; i<curr->node_count; ++i) {
-                       const ir_node *arg1 = curr->nodes[i];
-                       int arg1nr = get_irn_graph_nr(arg1);
-                       for (o=i+1; o<curr->node_count; ++o) {
-                               const ir_node *arg2 = curr->nodes[o];
-                               int arg2nr = get_irn_graph_nr(arg2);
-                               if (nodes_interfere(pi->co->chordal_env, arg1, arg2)) {
-                                       int cst_idx, y_idx;
-                                       char buf[32];
-
-                                       mangle_cst(buf, 'P', cst_counter++);
-                                       cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 1);
-
-                                       mangle_var2(buf, 'y', rootnr, arg1nr);
-                                       y_idx = lpp_get_var_idx(pi->curr_lp, buf);
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, 1);
-
-                                       mangle_var2(buf, 'y', rootnr, arg2nr);
-                                       y_idx = lpp_get_var_idx(pi->curr_lp, buf);
-                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, 1);
-                               }
-                       }
-               }
-       }
-}
-#endif
-
-#ifdef PATH_CONSTRAINTS_FOR_CLASSES
-static INLINE int get_y_var_idx(problem_instance_t *pi, int nnr1, int nnr2) {
-       int res;
-       char buf[30];
-
-       mangle_var2(buf, 'y', nnr1, nnr2);
-       if ((res = lpp_get_var_idx(pi->curr_lp, buf)) != -1)
-               return res;
-
-       mangle_var2(buf, 'y', nnr2, nnr1);
-       if ((res = lpp_get_var_idx(pi->curr_lp, buf)) != -1)
-               return res;
-
-       assert(0 && "One of them must work");
-  return -1;
-}
-
-static void check_ecc_and_add_cut(problem_instance_t *pi, ir_node **path, int length, pset *remain, ir_node *tgt) {
-       if (path[length-1] == tgt) { /* we found a path */
-               int cst_idx, var_idx, i, nnr1, nnr2;
-               char buf[30];
-
-               /* add cut to ilp */
-               mangle_cst(buf, 'Q', pi->cst_counter++);
-               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 1);
-
-               /* add all vars along the path */
-               nnr2 = get_irn_graph_nr(path[0]);
-               for (i=1; i<length; ++i) {
-                       nnr1 = nnr2;
-                       nnr2 = get_irn_graph_nr(path[i]);
-                       var_idx = get_y_var_idx(pi, nnr1, nnr2);
-                       lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
-               }
-       } else { /* try to extend the path */
-               be_chordal_env_t *cenv = pi->co->chordal_env;
-               const ir_edge_t *edge;
-               ir_node *end = path[length-1];
-               ir_node **next = alloca(pset_count(remain) * sizeof(*next));
-               int i, o, max, next_pos = 0;
-               pset *done = pset_new_ptr_default();
-
-               /* find all potential next nodes on path */
-               /*  args of phis */
-               if (is_Phi(end))
-                       for(i=0, max=get_irn_arity(end); i<max; ++i) {
-                               ir_node *arg = get_irn_n(end, i);
-                               if (!pset_find_ptr(done, arg) && pset_find_ptr(remain, arg)) {
-                                       next[next_pos++] = arg;
-                                       pset_insert_ptr(done, arg);
-                               }
-                       }
-               /*  outs of phis and other nodes */
-               foreach_out_edge(end, edge) {
-                       ir_node *user = edge->src;
-                       if (is_Phi(user) && !pset_find_ptr(done, user) && pset_find_ptr(remain, user)) {
-                               next[next_pos++] = user;
-                               pset_insert_ptr(done, user);
-                       }
-               }
-               del_pset(done);
-
-
-               /* delete all potential nodes with interferences to other nodes in the path */
-               for (i=0; i<next_pos; ++i) {
-                       ir_node *nn = next[i];
-
-                       /* if next is the tgt, it may interfere with path[0],
-                        * so skip the first check */
-                       o = (nn == tgt && length > 1) ? 1 : 0;
-
-                       for(; o<length; ++o)
-                               if (nodes_interfere(cenv, nn, path[o])) {
-                                       next[i] = NULL;
-                                       break;
-                               }
-               }
-               /* now we have all possible nodes in next; impossibles are NULL */
-
-               /* try to finish path with all possible nodes */
-               for (i=0; i<next_pos; ++i) {
-                       if (!next[i]) /* this was an impossible node */
-                               continue;
-
-                       path[length] = next[i];
-                       pset_remove_ptr(remain, next[i]);
-                       check_ecc_and_add_cut(pi, path, length+1, remain, tgt);
-                       pset_insert_ptr(remain, next[i]);
-               }
-       }
-}
-
-static void path_cstr_for_classes_walker(ir_node *irn, void *env) {
-       problem_instance_t *pi = env;
-       be_chordal_env_t *cenv;
-       int i, o, max;
-       ir_node *m, **cls;
-       pset *class = get_phi_class(irn);
-       if (!class || pset_find_ptr(pi->done, class))
-               return;
-
-       pset_insert_ptr(pi->done, class);
-
-       /* pset to array */
-       max = pset_count(class);
-       cls = alloca(max * sizeof(*cls));
-       for(i=0, m = pset_first(class); m; i++, m = pset_next(class)) {
-               DBG((dbg, LEVEL_1, " class member: %+F\n", m));
-               cls[i] = m;
-       }
-
-       cenv = pi->co->chordal_env;
-       for(i=0; i<max; ++i) {
-               ir_node **path = alloca(max * sizeof(*path));
-               pset *remain = pset_new_ptr(8);
-               pset_insert_pset_ptr(remain, class);
-
-               /* add cls[i] to path and remove it from remainder */
-               path[0] = cls[i];
-               pset_remove_ptr(remain, cls[i]);
-
-               for(o=i+1; o<max; ++o)
-                       if (nodes_interfere(cenv, cls[i], cls[o]))
-                               check_ecc_and_add_cut(pi, path, 1, remain, cls[o]);
-
-               /* insert back into remainder */
-               pset_insert_ptr(remain, cls[i]);
-       }
-}
-
-
-/**
- * Matrix P: Path contraints.
- * If 2 nodes interfere and there is a path of equal-color-edges
- * connecting them, then at least one of those equal-color-edges
- * will break and cause some costs.
- */
-static void pi_add_path_cstr_for_classes(problem_instance_t *pi) {
-       DBG((dbg, LEVEL_2, "Adding path constraints for phi classes...\n"));
-       pi->cst_counter = 0;
-       pi->done = pset_new_ptr_default();
-       irg_walk_graph(get_irg(pi->co), path_cstr_for_classes_walker, NULL, pi);
-       del_pset(pi->done);
-}
-#endif
-
-#ifdef PRECOLOR_MAX_CLIQUE
-struct pre_col {
-       problem_instance_t *pi;
-       pset **clique;
-};
-
-#define has_reg_class(pi,irn) \
-  (arch_get_irn_reg_class(pi->co->chordal_env->session_env->main_env->arch_env, \
-                          irn, -1) == pi->co->chordal_env->cls)
-
-static void preColoringWalker(ir_node *bl, void *env) {
-       struct pre_col *e = env;
-       pset **clique = e->clique;
-       pset *max_clique = clique ? *clique : NULL;
-       int max = max_clique ? pset_count(max_clique) : 0;
-       problem_instance_t *pi = e->pi;
-
-       int i, n;
-       pset *live       = pset_new_ptr_default();
-       ir_node *irn;
-       irn_live_t *li;
-
-       /* as always, bring the live end nodes to life here */
-       live_foreach(bl, li) {
-         if(live_is_end(li) && has_reg_class(pi, li->irn)) {
-           pset_insert_ptr(live, irn);
-         }
-       }
-
-       sched_foreach_reverse(bl, irn) {
-               int pres = pset_count(live);
-
-               if(pres > max) {
-                       max = pres;
-                       if(max_clique)
-                               del_pset(max_clique);
-
-                       max_clique = pset_new_ptr_default();
-                       pset_insert_pset_ptr(max_clique, live);
-               }
-
-
-
-               if(has_reg_class(pi, irn))
-                       pset_remove_ptr(live, irn);
-
-               for(i = 0, n = get_irn_arity(irn); i < n; ++i) {
-                       ir_node *op = get_irn_n(irn, i);
-                       if(has_reg_class(pi, op) && !is_Phi(irn))
-                               pset_insert_ptr(live, op);
-               }
-       }
-
-  del_pset(live);
-  *clique = max_clique;
-}
-
-static void pi_add_constr_preColoring(problem_instance_t *pi) {
-       ir_node *irn;
-       int cst_counter, color;
-       struct pre_col pre_col;
-
-       pre_col.clique = NULL;
-       pre_col.pi = pi;
-
-       dom_tree_walk_irg(get_irg(pi->co), preColoringWalker, NULL, &pre_col);
-
-       color = 0;
-       for (irn = pset_first(*pre_col.clique); irn; irn = pset_next(*pre_col.clique)) {
-               int cst_idx, var_idx, nnr = get_irn_graph_nr(irn);
-               char buf[100];
-
-               mangle_cst(buf, 'K', cst_counter++);
-               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_equal, 1);
-
-               mangle_var2(buf, 'x', nnr, color++);
-               var_idx = lpp_get_var_idx(pi->curr_lp, buf);
-               lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
-       }
-}
-#endif
-
-/**
- * Generate the initial problem matrices and vectors.
- */
-static problem_instance_t *new_pi(const copy_opt_t *co) {
-       problem_instance_t *pi;
-       int col;
-
-       DBG((dbg, LEVEL_2, "Generating new instance...\n"));
-       pi = xcalloc(1, sizeof(*pi));
-       pi->co = co;
-       pi->removed = pset_new_ptr_default();
-       INIT_LIST_HEAD(&pi->simplicials);
-       pi->dilp     = new_lpp(co->name, lpp_minimize);
-
-       /* problem size reduction */
-       pi_find_simplicials(pi);
-       if (pi->all_simplicial)
-               return pi;
-
-       /* built objective and constraints */
-       pi->curr_lp = pi->dilp;
-#ifdef NO_NULL_COLORS
-       pi_add_constr_no_null_colors(pi);
-#endif
-       pi_add_constr_A(pi);
-       for (col = 0; col < pi->co->chordal_env->cls->n_regs; ++col)
-               pi_add_constr_B(pi, col);
-       pi_add_constr_E(pi);
-
-#ifdef PATH_CONSTRAINTS_FOR_CLASSES
-       pi_add_path_cstr_for_classes(pi);
-#else
-       pi_add_path_cstr(pi);
-#endif
-       pi_add_clique_path_cstr(pi);
-#ifdef PRECOLOR_MAX_CLIQUE
-       pi_add_constr_preColoring(pi);
-#endif
-
-       return pi;
-}
-
-/**
- * Clean the problem instance
- */
-static void free_pi(problem_instance_t *pi) {
-       simpl_t *simpl, *tmp;
-
-       DBG((dbg, LEVEL_2, "Free instance...\n"));
-       free_lpp(pi->dilp);
-       list_for_each_entry_safe(simpl_t, simpl, tmp, &pi->simplicials, chain)
-               free(simpl);
-       del_pset(pi->removed);
-       free(pi);
-}
-
-/**
- * Set starting values for the mip problem according
- * to the current coloring of the graph.
- */
-static void pi_set_start_sol(problem_instance_t *pi) {
-       int i;
-       char var_name[64];
-       DBG((dbg, LEVEL_2, "Set start solution...\n"));
-       for (i=pi->first_x_var; i<=pi->last_x_var; ++i) {
-               int nnr, col;
-               double val;
-               /* get variable name */
-               lpp_get_var_name(pi->curr_lp, i, var_name, sizeof(var_name));
-               /* split into components */
-               if (split_var(var_name, &nnr, &col) == 2) {
-                       assert(get_irn_col(pi->co, get_irn_for_graph_nr(get_irg(pi->co), nnr)) != -1);
-                       val = (get_irn_col(pi->co, get_irn_for_graph_nr(get_irg(pi->co), nnr)) == col) ? 1 : 0;
-                       lpp_set_start_value(pi->curr_lp, i, val);
-               } else {
-                       fprintf(stderr, "Variable name is: %s\n", var_name);
-                       assert(0 && "x vars always look like this 'x123_45'");
-               }
-       }
-}
-
-/**
- * Invoke a solver
- */
-static void pi_solve_ilp(problem_instance_t *pi) {
-  double lower_bound;
-
-       pi_set_start_sol(pi);
-       lower_bound = co_get_lower_bound(pi->co) - co_get_inevit_copy_costs(pi->co);
-       lpp_set_bound(pi->curr_lp, lower_bound);
-       lpp_solve_net(pi->curr_lp, LPP_HOST, LPP_SOLVER);
-//     lpp_solve_cplex(pi->curr_lp);
-       DBG((dbg, LEVEL_1, "Solution time: %.2f\n", pi->curr_lp->sol_time));
-}
-
-/**
- * Set the color of all simplicial nodes removed form
- * the graph before transforming it to an ilp.
- */
-static void pi_set_simplicials(problem_instance_t *pi) {
-       simpl_t *simpl, *tmp;
-       be_ifg_t *ifg        = pi->co->chordal_env->ifg;
-       bitset_t *used_cols  = bitset_alloca(arch_register_class_n_regs(pi->co->chordal_env->cls));
-       void *iter           = be_ifg_neighbours_iter_alloca(ifg);
-
-       DBG((dbg, LEVEL_2, "Set simplicials...\n"));
-       /* color the simplicial nodes in right order */
-       list_for_each_entry_safe(simpl_t, simpl, tmp, &pi->simplicials, chain) {
-               int free_col;
-               ir_node *other, *irn;
-
-               /* get free color by inspecting all neighbors */
-               irn = simpl->irn;
-               bitset_clear_all(used_cols);
-
-               be_ifg_foreach_neighbour(ifg, iter, irn, other) {
-                       if (!is_removed(other)) /* only inspect nodes which are in graph right now */
-                               bitset_set(used_cols, get_irn_col(pi->co, other));
-               }
-
-               /* now all bits not set are possible colors */
-               free_col = bitset_next_clear(used_cols, 0);
-               assert(free_col != -1 && "No free color found. This can not be.");
-               set_irn_col(pi->co, irn, free_col);
-               pset_remove_ptr(pi->removed, irn); /* irn is back in graph again */
-       }
-}
-
-/**
- * Sets the colors of irns according to the values of variables
- * provided by the solution of the solver.
- */
-static int pi_apply_solution(problem_instance_t *pi) {
-       int res = 1, i;
-       double *sol;
-       lpp_sol_state_t state;
-       DBG((dbg, LEVEL_2, "Applying solution...\n"));
-
-#ifdef DO_STAT
-       copystat_add_ilp_time((int)(1000.0*lpp_get_sol_time(pi->curr_lp)));  //now we have ms
-       copystat_add_ilp_vars(lpp_get_var_count(pi->curr_lp));
-       copystat_add_ilp_csts(lpp_get_cst_count(pi->curr_lp));
-       copystat_add_ilp_iter(lpp_get_iter_cnt(pi->curr_lp));
-#endif
-
-       sol = xmalloc((pi->last_x_var - pi->first_x_var + 1) * sizeof(*sol));
-       state = lpp_get_solution(pi->curr_lp, sol, pi->first_x_var, pi->last_x_var);
-       if (state != lpp_optimal) {
-               printf("WARNING %s: Solution state is not 'optimal': %d\n", pi->co->name, state);
-               assert(state >= lpp_feasible && "The solution should at least be feasible!");
-               res = 0;
-       }
-       for (i=0; i<pi->last_x_var - pi->first_x_var + 1; ++i) {
-               int nnr, col;
-               char var_name[64];
-
-               if (sol[i] > 1-EPSILON) { /* split varibale name into components */
-                       lpp_get_var_name(pi->curr_lp, pi->first_x_var+i, var_name, sizeof(var_name));
-                       if (split_var(var_name, &nnr, &col) == 2) {
-                               DBG((dbg, LEVEL_2, "Irn %n  Idx %d  Var %s  Val %f\n", get_irn_for_graph_nr(get_irg(pi->co), nnr), i, var_name, sol[i]));
-                               DBG((dbg, LEVEL_2, "x%d = %d\n", nnr, col));
-                               set_irn_col(pi->co, get_irn_for_graph_nr(get_irg(pi->co), nnr), col);
-                       } else
-                               assert(0 && "This should be a x-var");
-               }
-       }
-       return res;
-}
-
-int co_ilp_opt(copy_opt_t *co, double time_limit) {
-       int res = 1;
-       problem_instance_t *pi;
-
-       dbg = firm_dbg_register("ir.be.copyoptilp");
-
-       pi = new_pi(co);
-       if (!pi->all_simplicial) {
-#ifdef DUMP_MPS
-               char buf[512];
-               snprintf(buf, sizeof(buf), "%s.mps", co->name);
-               lpp_dump(pi->curr_lp, buf);
-#endif
-               lpp_set_time_limit(pi->curr_lp, time_limit);
-               pi_solve_ilp(pi);
-               res = pi_apply_solution(pi);
-               pi_set_simplicials(pi);
-       }
-       free_pi(pi);
-       return res;
-}
+#include "becopyilp_t.h"
\ No newline at end of file
diff --git a/ir/be/becopyilp1.c b/ir/be/becopyilp1.c
new file mode 100644 (file)
index 0000000..0f2c33c
--- /dev/null
@@ -0,0 +1,989 @@
+/**
+ * Author:      Daniel Grund
+ * Date:               17.05.2005
+ * Copyright:   (c) Universitaet Karlsruhe
+ * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
+ * CVS-ID:      $Id$
+ *
+ * ILP formalization using:
+ *  - 2 classes of vars: Nodes- and optimality variables.
+ *  - Clique constraints
+ *  - Path constraints
+ *  - Clique path constraints
+ */
+
+#include "becopyilp_t.h"
+#include "benumb_t.h"
+#include "belive_t.h"
+#include "irdom_t.h"
+#include "irgwalk.h"
+
+#define PATH_CONSTRAINTS_FOR_CLASSES
+#undef PRECOLOR_MAX_CLIQUE
+#undef NO_NULL_COLORS
+#undef NO_NULL_COLORS_EXTRA_CSTS
+#undef NO_NULL_COLORS_WITH_COSTS
+#if (defined(NO_NULL_COLORS_EXTRA_CSTS) || defined(NO_NULL_COLORS_WITH_COSTS)) && !defined(NO_NULL_COLORS)
+#error Chose your weapon!
+#endif
+
+#undef DUMP_MPS
+static firm_dbg_module_t *dbg = NULL;
+#define SLOTS_LIVING 32
+
+
+
+typedef struct _simpl_t {
+       struct list_head chain;
+       ir_node *irn;
+} simpl_t;
+
+typedef struct _problem_instance_t {
+       const copy_opt_t *co;                   /** the copy_opt problem */
+       /* problem size reduction removing simple nodes */
+       struct list_head simplicials;   /**< holds all simpl_t's in right order to color*/
+       pset *removed;                                  /**< holds all removed simplicial irns */
+       /* lp problem */
+       lpp_t *curr_lp;                                 /**< points to the problem currently used */
+       lpp_t *dilp;                                    /**< problem formulation directly as milp */
+#ifdef NO_NULL_COLORS_EXTRA_CSTS
+       int first_nnc_cst_idx;                  /**< the first index of a constraint belonging to no-null-colors stuff*/
+#endif
+       int first_nnc_var_idx;                  /**< the first index of a constraint belonging to no-null-colors stuff*/
+
+       int cst_counter, first_x_var, last_x_var;
+       char buf[32];
+       int all_simplicial;
+       pset *done;
+} problem_instance_t;
+
+#define is_removed(irn) pset_find_ptr(pi->removed, irn)
+
+#define is_color_possible(irn,color) arch_reg_is_allocatable(pi->co->aenv, irn, -1, arch_register_for_index(pi->co->cls, color))
+
+/*
+ * Some stuff for variable name handling.
+ */
+#define mangle_cst(buf, prefix, nr) \
+                       snprintf((buf), sizeof(buf), "%c%d", (prefix), (nr))
+
+#define mangle_var1(buf, prefix, color) \
+                       snprintf((buf), sizeof(buf), "%c%d", (prefix), (color))
+
+#define mangle_var2(buf, prefix, node_nr, color) \
+                       snprintf((buf), sizeof(buf), "%c%d_%d", (prefix), (node_nr), (color))
+
+#define mangle_var3(buf, prefix, n1, n2, col) \
+                       snprintf((buf), sizeof(buf), "%c%d_%d_%d", (prefix), (n1), (n2), (col))
+
+#define mangle_var_irn(buf, prefix, irn, color) \
+                       mangle_var2((buf), (prefix), get_irn_graph_nr(irn), (color))
+
+#define split_var(var, nnr, col) \
+                       sscanf(var, "x%d_%d", (nnr), (col))
+
+
+/**
+ * Checks if a node is simplicial in the graph
+ * heeding the already removed nodes.
+ */
+static INLINE int pi_is_simplicial(problem_instance_t *pi, const ir_node *ifn) {
+       int i, o, size = 0;
+       ir_node **all, *curr;
+       be_ifg_t *ifg = pi->co->cenv->ifg;
+       void *iter = be_ifg_neighbours_iter_alloca(ifg);
+
+       all = alloca(be_ifg_degree(ifg, ifn) * sizeof(*all));
+
+       /* get all non-removed neighbors */
+       be_ifg_foreach_neighbour(ifg, iter, ifn, curr)
+               if (!is_removed(curr))
+                       all[size++] = curr;
+
+       /* check if these form a clique */
+       for (i=0; i<size; ++i)
+               for (o=i+1; o<size; ++o)
+                       if (!be_ifg_connected(ifg, all[i], all[o]))
+                               return 0;
+
+       /* all edges exist so this is a clique */
+       return 1;
+}
+
+static int irn_cmp(const void *a, const void *b, size_t n)
+{
+       return a != b;
+}
+
+/**
+ * Iterative finds and 'removes' from the graph all nodes which are
+ * simplicial AND not member of a equal-color-wish
+ */
+static void pi_find_simplicials(problem_instance_t *pi) {
+       ir_node *irn;
+       int redo = 1;
+       int n_nodes = 0;
+       const be_ifg_t *ifg = pi->co->cenv->ifg;
+       void *iter = be_ifg_neighbours_iter_alloca(ifg);
+
+       DBG((dbg, LEVEL_2, "Find simlicials...\n"));
+
+       while (redo) {
+               arch_register_req_t req;
+               redo = 0;
+               be_ifg_foreach_node(ifg, iter, irn) {
+                       if (!is_removed(irn) && !co_is_optimizable(pi->co->aenv, irn, &req) && !co_is_optimizable_arg(pi->co, irn)) {
+                       if (pi_is_simplicial(pi, irn)) {
+                                       simpl_t *s = xmalloc(sizeof(*s));
+                                       s->irn = irn;
+                                       list_add(&s->chain, &pi->simplicials);
+                                       pset_insert_ptr(pi->removed, irn);
+                                       redo = 1;
+                                       DBG((dbg, LEVEL_2, " Removed %+F\n", irn));
+                       }
+                       }
+               }
+       }
+
+       /* TODO: Count inside the last look */
+       be_ifg_foreach_node(ifg, iter, irn) {
+               n_nodes++;
+       }
+
+       if (n_nodes == pset_count(pi->removed))
+               pi->all_simplicial = 1;
+}
+
+#ifdef NO_NULL_COLORS
+static void pi_add_constr_no_null_colors(problem_instance_t *pi) {
+       int cst_counter=0, col, var_idx, cst_idx;
+       int n_colors = pi->co->chordal_env->cls->n_regs;
+       char buf[40];
+
+       for (col = 0; col < n_colors; ++col) {
+               mangle_var1(buf, 'u', col);
+#ifdef NO_NULL_COLORS_WITH_COSTS
+               var_idx = lpp_add_var(pi->curr_lp, buf, lpp_binary, 1.0 / (double) (1 << (col+1)) );
+#else
+               var_idx = lpp_add_var(pi->curr_lp, buf, lpp_binary, 1.0 / (2.0 * n_colors) );
+#endif
+               if (!pi->first_nnc_var_idx)
+                       pi->first_nnc_var_idx = var_idx;
+       }
+
+#ifdef NO_NULL_COLORS_EXTRA_CSTS
+       for (col = 0; col < n_colors; ++col) {
+               mangle_cst(buf, 'U', cst_counter++);
+               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 0);
+               if (!pi->first_nnc_cst_idx)
+                       pi->first_nnc_cst_idx = cst_idx;
+               lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+col, -1);
+       }
+#endif
+
+#ifndef NO_NULL_COLORS_WITH_COSTS
+       for (col = 0; col < n_colors - 1; ++col) {
+               mangle_cst(buf, 'U', cst_counter++);
+               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 0);
+               lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+col  ,  1);
+               lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+col+1, -1);
+       }
+#endif
+
+}
+#endif
+
+/**
+ * Add coloring-force conditions
+ * Matrix A: knapsack constraint for each node
+ */
+static void pi_add_constr_A(problem_instance_t *pi) {
+       pmap_entry *pme;
+
+       DBG((dbg, LEVEL_2, "Add A constraints...\n"));
+       /* iterate over all blocks */
+       pmap_foreach(pi->co->cenv->border_heads, pme) {
+               struct list_head *head = pme->value;
+               border_t *curr;
+               bitset_t *pos_regs = bitset_alloca(pi->co->cenv->cls->n_regs);
+
+               list_for_each_entry_reverse(border_t, curr, head, list)
+                       if (curr->is_def && curr->is_real && !is_removed(curr->irn)) {
+                               int cst_idx, nnr, col;
+
+                               nnr = get_irn_graph_nr(curr->irn);
+                               mangle_cst(pi->buf, 'A', nnr);
+                               cst_idx = lpp_add_cst(pi->curr_lp, pi->buf, lpp_equal, 1);
+
+                               /* iterate over all possible colors in order */
+                               bitset_clear_all(pos_regs);
+                               arch_get_allocatable_regs(pi->co->aenv, curr->irn, -1, pos_regs);
+                               bitset_foreach(pos_regs, col) {
+                                       int var_idx;
+                                       mangle_var2(pi->buf, 'x', nnr, col);
+                                       var_idx = lpp_add_var(pi->curr_lp, pi->buf, lpp_binary, 0);
+                                       if (!pi->first_x_var)
+                                               pi->first_x_var = var_idx;
+                                       pi->last_x_var = var_idx;
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
+#ifdef NO_NULL_COLORS_EXTRA_CSTS
+                                       lpp_set_factor_fast(pi->curr_lp, pi->first_nnc_cst_idx+col, var_idx, 1);
+#endif
+                               }
+                       }
+       }
+}
+
+/**
+ * Checks if all nodes in @p living are live in in block @p block.
+ * @return 1 if all are live in
+ *         0 else
+ */
+static INLINE int all_live_in(ir_node *block, pset *living) {
+       ir_node *n;
+       for (n = pset_first(living); n; n = pset_next(living))
+               if (!is_live_in(block, n)) {
+                       pset_break(living);
+                       return 0;
+               }
+       return 1;
+}
+
+/**
+ * Finds cliques in the interference graph, considering only nodes
+ * for which the color @p color is possible. Finds only 'maximal-cliques',
+ * viz cliques which are not contained in another one.
+ * Matrix B: interference constraints using cliques
+ */
+static void pi_add_constr_B(problem_instance_t *pi, int color) {
+       enum phase_t {growing, shrinking} phase = growing;
+       border_t *b;
+       pmap_entry *pme;
+       pset *living = pset_new_ptr(SLOTS_LIVING);
+
+       DBG((dbg, LEVEL_2, "Add B constraints (col = %d)...\n", color));
+       /* iterate over all blocks */
+       pmap_foreach(pi->co->cenv->border_heads, pme) {
+               ir_node *block = pme->key;
+               struct list_head *head = pme->value;
+
+               list_for_each_entry_reverse(border_t, b, head, list) {
+                       const ir_node *irn = b->irn;
+                       if (is_removed(irn) || !is_color_possible(irn, color))
+                               continue;
+
+                       if (b->is_def) {
+                               DBG((dbg, LEVEL_2, "Def %n\n", irn));
+                               pset_insert_ptr(living, irn);
+                               phase = growing;
+                       } else { /* is_use */
+                               DBG((dbg, LEVEL_2, "Use %n\n", irn));
+
+                               /* before shrinking the set, store the current 'maximum' clique;
+                                * do NOT if clique is a single node
+                                * do NOT if all values are live_in (in this case they were contained in a live-out clique elsewhere) */
+                               if (phase == growing && pset_count(living) >= 2 && !all_live_in(block, living)) {
+                                       int cst_idx;
+                                       ir_node *n;
+                                       mangle_cst(pi->buf, 'B', pi->cst_counter);
+#ifdef NO_NULL_COLORS
+                                       cst_idx = lpp_add_cst(pi->curr_lp, pi->buf, lpp_less, 0);
+#else
+                                       cst_idx = lpp_add_cst(pi->curr_lp, pi->buf, lpp_less, 1);
+#endif
+                                       for (n = pset_first(living); n; n = pset_next(living)) {
+                                               int var_idx;
+                                               mangle_var_irn(pi->buf, 'x', n, color);
+                                               var_idx = lpp_get_var_idx(pi->curr_lp, pi->buf);
+                                               lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
+                                       }
+#ifdef NO_NULL_COLORS
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, pi->first_nnc_var_idx+color, -1.0);
+#endif
+                                       pi->cst_counter++;
+                               }
+                               pset_remove_ptr(living, irn);
+                               phase = shrinking;
+                       }
+               }
+       }
+       assert(0 == pset_count(living));
+       del_pset(living);
+}
+
+/**
+ * Generates constraints which interrelate x with y variables.
+ * x1 and x2 have the different colors ==> y_12 = 1
+ */
+static void pi_add_constr_E(problem_instance_t *pi) {
+       unit_t *curr;
+       bitset_t *root_regs, *arg_regs, *work_regs;
+       int cst_counter = 0;
+       unsigned nregs = pi->co->cenv->cls->n_regs;
+       root_regs = bitset_alloca(nregs);
+       arg_regs = bitset_alloca(nregs);
+       work_regs = bitset_alloca(nregs);
+
+       DBG((dbg, LEVEL_2, "Add E constraints...\n"));
+       /* for all roots of optimization units */
+       list_for_each_entry(unit_t, curr, &pi->co->units, units) {
+               ir_node *root, *arg;
+               int rootnr, argnr, color;
+               int y_idx, i;
+               char buf[32];
+
+               root = curr->nodes[0];
+               rootnr = get_irn_graph_nr(root);
+               bitset_clear_all(root_regs);
+               arch_get_allocatable_regs(pi->co->aenv, root, -1, root_regs);
+
+               /* for all arguments of root */
+               for (i = 1; i < curr->node_count; ++i) {
+                       arg = curr->nodes[i];
+                       argnr = get_irn_graph_nr(arg);
+                       bitset_clear_all(arg_regs);
+                       arch_get_allocatable_regs(pi->co->aenv, arg, -1, arg_regs);
+
+                       /* Introduce new variable and set factor in objective function */
+                       mangle_var2(buf, 'y', rootnr, argnr);
+                       y_idx = lpp_add_var(pi->curr_lp, buf, lpp_binary, curr->costs[i]);
+
+                       /* set starting value */
+                       lpp_set_start_value(pi->curr_lp, y_idx, (get_irn_col(pi->co, root) != get_irn_col(pi->co, arg)));
+
+                       /* For all colors root and arg have in common, add 2 constraints to E */
+                       bitset_copy(work_regs, root_regs);
+                       bitset_and(work_regs, arg_regs);
+                       bitset_foreach(work_regs, color) {
+                               int root_idx, arg_idx, cst_idx;
+                               mangle_var2(buf, 'x', rootnr, color);
+                               root_idx = lpp_get_var_idx(pi->curr_lp, buf);
+                               mangle_var2(buf, 'x', argnr, color);
+                               arg_idx = lpp_get_var_idx(pi->curr_lp, buf);
+
+                               /* add root-arg-y <= 0 */
+                               mangle_cst(buf, 'E', cst_counter++);
+                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_less, 0);
+                               lpp_set_factor_fast(pi->curr_lp, cst_idx, root_idx, 1);
+                               lpp_set_factor_fast(pi->curr_lp, cst_idx, arg_idx, -1);
+                               lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
+
+                               /* add arg-root-y <= 0 */
+                               mangle_cst(buf, 'E', cst_counter++);
+                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_less, 0);
+                               lpp_set_factor_fast(pi->curr_lp, cst_idx, root_idx, -1);
+                               lpp_set_factor_fast(pi->curr_lp, cst_idx, arg_idx, 1);
+                               lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
+                       }
+                       /* For all colors root and arg have "disjunct", add 1 constraints to E.
+                        * If root gets a color the arg is not possible to get then they will
+                        * definetly get different colors. So y has to be 1.
+                        * Vice versa for arg.
+                        */
+                       bitset_copy(work_regs, root_regs);
+                       bitset_xor(work_regs, arg_regs);
+                       bitset_foreach(work_regs, color) {
+                               int root_idx, arg_idx, cst_idx;
+                               mangle_var2(buf, 'x', rootnr, color);
+                               root_idx = lpp_get_var_idx(pi->curr_lp, buf);
+                               mangle_var2(buf, 'x', argnr, color);
+                               arg_idx = lpp_get_var_idx(pi->curr_lp, buf);
+
+                               mangle_cst(buf, 'E', cst_counter++);
+                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_less, 0);
+                               if (bitset_is_set(root_regs, color)) {
+                                       /* add root-y <= 0 */
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, root_idx, 1);
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
+                               } else {
+                                       assert(bitset_is_set(arg_regs, color) && "bitset_xor is buggy");
+                                       /* add arg-y <= 0 */
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, arg_idx, 1);
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, -1);
+                               }
+                       }
+               }
+       }
+}
+
+static INLINE int get_costs(problem_instance_t *pi, ir_node *phi, ir_node *irn) {
+       int i;
+       unit_t *curr;
+       /* search optimization unit for phi */
+       list_for_each_entry(unit_t, curr, &pi->co->units, units)
+               if (curr->nodes[0] == phi) {
+                       for (i=1; i<curr->node_count; ++i)
+                               if (curr->nodes[i] == irn)
+                                       return curr->costs[i];
+                       assert(0 && "irn must occur in this ou");
+               }
+       assert(0 && "phi must be found in a ou");
+       return 0;
+}
+
+static void clique_path_walker(ir_node *block, void *env) {
+       problem_instance_t *pi = env;
+       int count, arity, row, col, other_row, *costs;
+       ir_node **phis, *phi, *irn, **phi_matrix;
+       pset *done;
+       bitset_t *candidates;
+
+       /* Count all phi nodes of this block */
+       for (count=0, irn = sched_first(block); is_Phi(irn); irn = sched_next(irn))
+               count++;
+
+       /* We at least 2 phi nodes for this class of inequalities */
+       if (count < 2)
+               return;
+
+       /* Build the \Phi-Matrix */
+       arity = get_irn_arity(sched_first(block));
+       phis = alloca(count * sizeof(*phis));
+       costs = alloca(count * sizeof(costs));
+       phi_matrix = alloca(count*arity * sizeof(*phi_matrix));
+       candidates = bitset_alloca(count);
+
+       phi = sched_first(block);
+       for (row=0; row<count; ++row) {
+               phis[row] = phi;
+               for (col=0; col<arity; ++col) {
+                       ir_node *arg = get_irn_n(phi, col);
+                       /* Sort out all arguments interfering with its phi */
+                       if (nodes_interfere(pi->co->cenv, phi, arg)) {
+                               phi_matrix[row*arity + col] =  NULL;
+                       } else
+                               phi_matrix[row*arity + col] =  arg;
+               }
+               phi = sched_next(phi);
+       }
+
+       /* Now find the interesting patterns in the matrix:
+        * All nodes which are used at least twice in a column. */
+       /* columnwise ... */
+       for (col=0; col<arity; ++col) {
+               done = pset_new_ptr_default();
+               for (row=0; row<count; ++row) {
+                       irn = phi_matrix[row*arity + col];
+                       /*
+                        * is this an interfering arg (NULL)
+                        * or has the irn already been processed in this col?
+                        */
+                       if (!irn || pset_find_ptr(done, irn))
+                               continue;
+                       else
+                               pset_insert_ptr(done, irn);
+
+                       /* insert irn in candidates */
+                       bitset_clear_all(candidates);
+                       bitset_set(candidates, row);
+                       /* search the irn in the rows below */
+                       for (other_row = row+1; other_row<count; ++other_row)
+                               if (irn == phi_matrix[other_row*arity + col]) {
+                                       /* found the irn in the same col in another row */
+                                       bitset_set(candidates, other_row);
+                               }
+
+                       /* now we know all occurences of irn in this col */
+                       if (bitset_popcnt(candidates) < 2)
+                               continue;
+
+                       /* generate an unequation finally.
+                        * phis are indexed in the bitset,
+                        * shared argument is irn
+                        * rhs is phi_count - 1 */
+                       {
+                               char buf[32];
+                               ir_node *root;
+                               int pos, irnnr, rootnr, cst_idx, y_idx, cst_counter = 0;
+                               int minimal_unequal_count = bitset_popcnt(candidates)-1;
+
+                               irnnr = get_irn_graph_nr(irn);
+                               mangle_cst(buf, 'M', cst_counter++);
+                               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, minimal_unequal_count);
+
+                               /* for all phis */
+                               bitset_foreach(candidates, pos) {
+                                       root = phis[pos];
+                                       rootnr = get_irn_graph_nr(root);
+                                       mangle_var2(buf, 'y', rootnr, irnnr);
+                                       y_idx = lpp_get_var_idx(pi->curr_lp, buf);
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, 1);
+                               }
+                       }
+               }
+               del_pset(done); /* clear set for next row */
+       } /*next col*/
+}
+
+/**
+ * Matrix M: Multi-Arg-Use. Interrelates different \phi-functions
+ * in the same block, iff they use the same arg at the same pos.
+ * Only one of the phis can get the arg.
+ */
+static void pi_add_clique_path_cstr(problem_instance_t *pi) {
+       DBG((dbg, LEVEL_2, "Adding clique path constraints...\n"));
+       dom_tree_walk_irg(pi->co->irg, clique_path_walker, NULL, pi);
+}
+
+#ifndef PATH_CONSTRAINTS_FOR_CLASSES
+/**
+ * Matrix P: Path contraints.
+ * If 2 nodes interfere and there is a path of equal-color-edges
+ * connecting them, then at least one of those equal-color-edges
+ * will break and cause some costs.
+ */
+static void pi_add_path_cstr(problem_instance_t *pi) {
+       unit_t *curr;
+       int cst_counter = 0;
+       DBG((dbg, LEVEL_2, "Adding path constraints...\n"));
+
+       /* for all optimization units (only phis) */
+       list_for_each_entry(unit_t, curr, &pi->co->units, units) {
+               int i, o, rootnr;
+
+               if (curr->min_nodes_costs == 0)
+                       continue;
+
+               rootnr = get_irn_graph_nr(curr->nodes[0]);
+               /* check all argument pairs for interference */
+               for (i=1; i<curr->node_count; ++i) {
+                       const ir_node *arg1 = curr->nodes[i];
+                       int arg1nr = get_irn_graph_nr(arg1);
+                       for (o=i+1; o<curr->node_count; ++o) {
+                               const ir_node *arg2 = curr->nodes[o];
+                               int arg2nr = get_irn_graph_nr(arg2);
+                               if (nodes_interfere(pi->co->cenv, arg1, arg2)) {
+                                       int cst_idx, y_idx;
+                                       char buf[32];
+
+                                       mangle_cst(buf, 'P', cst_counter++);
+                                       cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 1);
+
+                                       mangle_var2(buf, 'y', rootnr, arg1nr);
+                                       y_idx = lpp_get_var_idx(pi->curr_lp, buf);
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, 1);
+
+                                       mangle_var2(buf, 'y', rootnr, arg2nr);
+                                       y_idx = lpp_get_var_idx(pi->curr_lp, buf);
+                                       lpp_set_factor_fast(pi->curr_lp, cst_idx, y_idx, 1);
+                               }
+                       }
+               }
+       }
+}
+#endif
+
+#ifdef PATH_CONSTRAINTS_FOR_CLASSES
+static INLINE int get_y_var_idx(problem_instance_t *pi, int nnr1, int nnr2) {
+       int res;
+       char buf[30];
+
+       mangle_var2(buf, 'y', nnr1, nnr2);
+       if ((res = lpp_get_var_idx(pi->curr_lp, buf)) != -1)
+               return res;
+
+       mangle_var2(buf, 'y', nnr2, nnr1);
+       if ((res = lpp_get_var_idx(pi->curr_lp, buf)) != -1)
+               return res;
+
+       assert(0 && "One of them must work");
+  return -1;
+}
+
+static void check_ecc_and_add_cut(problem_instance_t *pi, ir_node **path, int length, pset *remain, ir_node *tgt) {
+       if (path[length-1] == tgt) { /* we found a path */
+               int cst_idx, var_idx, i, nnr1, nnr2;
+               char buf[30];
+
+               /* add cut to ilp */
+               mangle_cst(buf, 'Q', pi->cst_counter++);
+               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_greater, 1);
+
+               /* add all vars along the path */
+               nnr2 = get_irn_graph_nr(path[0]);
+               for (i=1; i<length; ++i) {
+                       nnr1 = nnr2;
+                       nnr2 = get_irn_graph_nr(path[i]);
+                       var_idx = get_y_var_idx(pi, nnr1, nnr2);
+                       lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
+               }
+       } else { /* try to extend the path */
+               be_chordal_env_t *cenv = pi->co->cenv;
+               const ir_edge_t *edge;
+               ir_node *end = path[length-1];
+               ir_node **next = alloca(pset_count(remain) * sizeof(*next));
+               int i, o, max, next_pos = 0;
+               pset *done = pset_new_ptr_default();
+
+               /* find all potential next nodes on path */
+               /*  args of phis */
+               if (is_Phi(end))
+                       for(i=0, max=get_irn_arity(end); i<max; ++i) {
+                               ir_node *arg = get_irn_n(end, i);
+                               if (!pset_find_ptr(done, arg) && pset_find_ptr(remain, arg)) {
+                                       next[next_pos++] = arg;
+                                       pset_insert_ptr(done, arg);
+                               }
+                       }
+               /*  outs of phis and other nodes */
+               foreach_out_edge(end, edge) {
+                       ir_node *user = edge->src;
+                       if (is_Phi(user) && !pset_find_ptr(done, user) && pset_find_ptr(remain, user)) {
+                               next[next_pos++] = user;
+                               pset_insert_ptr(done, user);
+                       }
+               }
+               del_pset(done);
+
+
+               /* delete all potential nodes with interferences to other nodes in the path */
+               for (i=0; i<next_pos; ++i) {
+                       ir_node *nn = next[i];
+
+                       /* if next is the tgt, it may interfere with path[0],
+                        * so skip the first check */
+                       o = (nn == tgt && length > 1) ? 1 : 0;
+
+                       for(; o<length; ++o)
+                               if (nodes_interfere(cenv, nn, path[o])) {
+                                       next[i] = NULL;
+                                       break;
+                               }
+               }
+               /* now we have all possible nodes in next; impossibles are NULL */
+
+               /* try to finish path with all possible nodes */
+               for (i=0; i<next_pos; ++i) {
+                       if (!next[i]) /* this was an impossible node */
+                               continue;
+
+                       path[length] = next[i];
+                       pset_remove_ptr(remain, next[i]);
+                       check_ecc_and_add_cut(pi, path, length+1, remain, tgt);
+                       pset_insert_ptr(remain, next[i]);
+               }
+       }
+}
+
+static void path_cstr_for_classes_walker(ir_node *irn, void *env) {
+       problem_instance_t *pi = env;
+       be_chordal_env_t *cenv;
+       int i, o, max;
+       ir_node *m, **cls;
+       pset *class = get_phi_class(irn);
+       if (!class || pset_find_ptr(pi->done, class))
+               return;
+
+       pset_insert_ptr(pi->done, class);
+
+       /* pset to array */
+       max = pset_count(class);
+       cls = alloca(max * sizeof(*cls));
+       for(i=0, m = pset_first(class); m; i++, m = pset_next(class)) {
+               DBG((dbg, LEVEL_1, " class member: %+F\n", m));
+               cls[i] = m;
+       }
+
+       cenv = pi->co->cenv;
+       for(i=0; i<max; ++i) {
+               ir_node **path = alloca(max * sizeof(*path));
+               pset *remain = pset_new_ptr(8);
+               pset_insert_pset_ptr(remain, class);
+
+               /* add cls[i] to path and remove it from remainder */
+               path[0] = cls[i];
+               pset_remove_ptr(remain, cls[i]);
+
+               for(o=i+1; o<max; ++o)
+                       if (nodes_interfere(cenv, cls[i], cls[o]))
+                               check_ecc_and_add_cut(pi, path, 1, remain, cls[o]);
+
+               /* insert back into remainder */
+               pset_insert_ptr(remain, cls[i]);
+       }
+}
+
+
+/**
+ * Matrix P: Path contraints.
+ * If 2 nodes interfere and there is a path of equal-color-edges
+ * connecting them, then at least one of those equal-color-edges
+ * will break and cause some costs.
+ */
+static void pi_add_path_cstr_for_classes(problem_instance_t *pi) {
+       DBG((dbg, LEVEL_2, "Adding path constraints for phi classes...\n"));
+       pi->cst_counter = 0;
+       pi->done = pset_new_ptr_default();
+       irg_walk_graph(pi->co->irg, path_cstr_for_classes_walker, NULL, pi);
+       del_pset(pi->done);
+}
+#endif
+
+#ifdef PRECOLOR_MAX_CLIQUE
+struct pre_col {
+       problem_instance_t *pi;
+       pset **clique;
+};
+
+static void preColoringWalker(ir_node *bl, void *env) {
+       struct pre_col *e = env;
+       pset **clique = e->clique;
+       pset *max_clique = clique ? *clique : NULL;
+       int max = max_clique ? pset_count(max_clique) : 0;
+       problem_instance_t *pi = e->pi;
+
+       int i, n;
+       pset *live       = pset_new_ptr_default();
+       ir_node *irn;
+       irn_live_t *li;
+
+       /* as always, bring the live end nodes to life here */
+       live_foreach(bl, li) {
+         if(live_is_end(li) && has_reg_class(pi, li->irn)) {
+           pset_insert_ptr(live, irn);
+         }
+       }
+
+       sched_foreach_reverse(bl, irn) {
+               int pres = pset_count(live);
+
+               if(pres > max) {
+                       max = pres;
+                       if(max_clique)
+                               del_pset(max_clique);
+
+                       max_clique = pset_new_ptr_default();
+                       pset_insert_pset_ptr(max_clique, live);
+               }
+
+
+
+               if(has_reg_class(pi, irn))
+                       pset_remove_ptr(live, irn);
+
+               for(i = 0, n = get_irn_arity(irn); i < n; ++i) {
+                       ir_node *op = get_irn_n(irn, i);
+                       if(has_reg_class(pi, op) && !is_Phi(irn))
+                               pset_insert_ptr(live, op);
+               }
+       }
+
+  del_pset(live);
+  *clique = max_clique;
+}
+
+static void pi_add_constr_preColoring(problem_instance_t *pi) {
+       ir_node *irn;
+       int cst_counter, color;
+       struct pre_col pre_col;
+
+       pre_col.clique = NULL;
+       pre_col.pi = pi;
+
+       dom_tree_walk_irg(get_irg(pi->co), preColoringWalker, NULL, &pre_col);
+
+       color = 0;
+       for (irn = pset_first(*pre_col.clique); irn; irn = pset_next(*pre_col.clique)) {
+               int cst_idx, var_idx, nnr = get_irn_graph_nr(irn);
+               char buf[100];
+
+               mangle_cst(buf, 'K', cst_counter++);
+               cst_idx = lpp_add_cst(pi->curr_lp, buf, lpp_equal, 1);
+
+               mangle_var2(buf, 'x', nnr, color++);
+               var_idx = lpp_get_var_idx(pi->curr_lp, buf);
+               lpp_set_factor_fast(pi->curr_lp, cst_idx, var_idx, 1);
+       }
+}
+#endif
+
+/**
+ * Generate the initial problem matrices and vectors.
+ */
+static problem_instance_t *new_pi(const copy_opt_t *co) {
+       problem_instance_t *pi;
+       int col;
+
+       DBG((dbg, LEVEL_2, "Generating new instance...\n"));
+       pi = xcalloc(1, sizeof(*pi));
+       pi->co = co;
+       pi->removed = pset_new_ptr_default();
+       INIT_LIST_HEAD(&pi->simplicials);
+       pi->dilp     = new_lpp(co->name, lpp_minimize);
+
+       /* problem size reduction */
+       pi_find_simplicials(pi);
+       if (pi->all_simplicial)
+               return pi;
+
+       /* built objective and constraints */
+       pi->curr_lp = pi->dilp;
+#ifdef NO_NULL_COLORS
+       pi_add_constr_no_null_colors(pi);
+#endif
+       pi_add_constr_A(pi);
+       for (col = 0; col < pi->co->cls->n_regs; ++col)
+               pi_add_constr_B(pi, col);
+       pi_add_constr_E(pi);
+
+#ifdef PATH_CONSTRAINTS_FOR_CLASSES
+       pi_add_path_cstr_for_classes(pi);
+#else
+       pi_add_path_cstr(pi);
+#endif
+       pi_add_clique_path_cstr(pi);
+#ifdef PRECOLOR_MAX_CLIQUE
+       pi_add_constr_preColoring(pi);
+#endif
+
+       return pi;
+}
+
+/**
+ * Clean the problem instance
+ */
+static void free_pi(problem_instance_t *pi) {
+       simpl_t *simpl, *tmp;
+
+       DBG((dbg, LEVEL_2, "Free instance...\n"));
+       free_lpp(pi->dilp);
+       list_for_each_entry_safe(simpl_t, simpl, tmp, &pi->simplicials, chain)
+               free(simpl);
+       del_pset(pi->removed);
+       free(pi);
+}
+
+/**
+ * Set starting values for the mip problem according
+ * to the current coloring of the graph.
+ */
+static void pi_set_start_sol(problem_instance_t *pi) {
+       int i;
+       char var_name[64];
+       DBG((dbg, LEVEL_2, "Set start solution...\n"));
+       for (i=pi->first_x_var; i<=pi->last_x_var; ++i) {
+               int nnr, col;
+               double val;
+               /* get variable name */
+               lpp_get_var_name(pi->curr_lp, i, var_name, sizeof(var_name));
+               /* split into components */
+               if (split_var(var_name, &nnr, &col) == 2) {
+                       assert(get_irn_col(pi->co, get_irn_for_graph_nr(pi->co->irg, nnr)) != -1);
+                       val = (get_irn_col(pi->co, get_irn_for_graph_nr(pi->co->irg, nnr)) == col) ? 1 : 0;
+                       lpp_set_start_value(pi->curr_lp, i, val);
+               } else {
+                       fprintf(stderr, "Variable name is: %s\n", var_name);
+                       assert(0 && "x vars always look like this 'x123_45'");
+               }
+       }
+}
+
+/**
+ * Invoke a solver
+ */
+static void pi_solve_ilp(problem_instance_t *pi) {
+  double lower_bound;
+
+       pi_set_start_sol(pi);
+       lower_bound = co_get_lower_bound(pi->co) - co_get_inevit_copy_costs(pi->co);
+       lpp_set_bound(pi->curr_lp, lower_bound);
+       lpp_solve_net(pi->curr_lp, LPP_HOST, LPP_SOLVER);
+//     lpp_solve_cplex(pi->curr_lp);
+       DBG((dbg, LEVEL_1, "Solution time: %.2f\n", pi->curr_lp->sol_time));
+}
+
+/**
+ * Set the color of all simplicial nodes removed form
+ * the graph before transforming it to an ilp.
+ */
+static void pi_set_simplicials(problem_instance_t *pi) {
+       simpl_t *simpl, *tmp;
+       be_ifg_t *ifg        = pi->co->cenv->ifg;
+       bitset_t *used_cols  = bitset_alloca(arch_register_class_n_regs(pi->co->cls));
+       void *iter           = be_ifg_neighbours_iter_alloca(ifg);
+
+       DBG((dbg, LEVEL_2, "Set simplicials...\n"));
+       /* color the simplicial nodes in right order */
+       list_for_each_entry_safe(simpl_t, simpl, tmp, &pi->simplicials, chain) {
+               int free_col;
+               ir_node *other, *irn;
+
+               /* get free color by inspecting all neighbors */
+               irn = simpl->irn;
+               bitset_clear_all(used_cols);
+
+               be_ifg_foreach_neighbour(ifg, iter, irn, other) {
+                       if (!is_removed(other)) /* only inspect nodes which are in graph right now */
+                               bitset_set(used_cols, get_irn_col(pi->co, other));
+               }
+
+               /* now all bits not set are possible colors */
+               free_col = bitset_next_clear(used_cols, 0);
+               assert(free_col != -1 && "No free color found. This can not be.");
+               set_irn_col(pi->co, irn, free_col);
+               pset_remove_ptr(pi->removed, irn); /* irn is back in graph again */
+       }
+}
+
+/**
+ * Sets the colors of irns according to the values of variables
+ * provided by the solution of the solver.
+ */
+static int pi_apply_solution(problem_instance_t *pi) {
+       int res = 1, i;
+       double *sol;
+       lpp_sol_state_t state;
+       DBG((dbg, LEVEL_2, "Applying solution...\n"));
+
+#ifdef DO_STAT
+       copystat_add_ilp_time((int)(1000.0*lpp_get_sol_time(pi->curr_lp)));  //now we have ms
+       copystat_add_ilp_vars(lpp_get_var_count(pi->curr_lp));
+       copystat_add_ilp_csts(lpp_get_cst_count(pi->curr_lp));
+       copystat_add_ilp_iter(lpp_get_iter_cnt(pi->curr_lp));
+#endif
+
+       sol = xmalloc((pi->last_x_var - pi->first_x_var + 1) * sizeof(*sol));
+       state = lpp_get_solution(pi->curr_lp, sol, pi->first_x_var, pi->last_x_var);
+       if (state != lpp_optimal) {
+               printf("WARNING %s: Solution state is not 'optimal': %d\n", pi->co->name, state);
+               assert(state >= lpp_feasible && "The solution should at least be feasible!");
+               res = 0;
+       }
+       for (i=0; i<pi->last_x_var - pi->first_x_var + 1; ++i) {
+               int nnr, col;
+               char var_name[64];
+
+               if (sol[i] > 1-EPSILON) { /* split varibale name into components */
+                       lpp_get_var_name(pi->curr_lp, pi->first_x_var+i, var_name, sizeof(var_name));
+                       if (split_var(var_name, &nnr, &col) == 2) {
+                               DBG((dbg, LEVEL_2, "Irn %n  Idx %d  Var %s  Val %f\n", get_irn_for_graph_nr(pi->co->irg, nnr), i, var_name, sol[i]));
+                               DBG((dbg, LEVEL_2, "x%d = %d\n", nnr, col));
+                               set_irn_col(pi->co, get_irn_for_graph_nr(pi->co->irg, nnr), col);
+                       } else
+                               assert(0 && "This should be a x-var");
+               }
+       }
+       return res;
+}
+
+int co_solve_ilp1(copy_opt_t *co, double time_limit) {
+       int res = 1;
+       problem_instance_t *pi;
+
+       dbg = firm_dbg_register("ir.be.copyoptilp");
+
+       pi = new_pi(co);
+       if (!pi->all_simplicial) {
+#ifdef DUMP_MPS
+               char buf[512];
+               snprintf(buf, sizeof(buf), "%s.mps", co->name);
+               lpp_dump(pi->curr_lp, buf);
+#endif
+               lpp_set_time_limit(pi->curr_lp, time_limit);
+               pi_solve_ilp(pi);
+               res = pi_apply_solution(pi);
+               pi_set_simplicials(pi);
+       }
+       free_pi(pi);
+       return res;
+}
diff --git a/ir/be/becopyilp2.c b/ir/be/becopyilp2.c
new file mode 100644 (file)
index 0000000..33efd78
--- /dev/null
@@ -0,0 +1,34 @@
+/**
+ * Author:      Daniel Grund
+ * Date:               28.02.2006
+ * Copyright:   (c) Universitaet Karlsruhe
+ * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
+ *
+ * ILP formalization using G=(V, E, Q):
+ *  - 1 class of variables: equal color vars
+ *  - Path constraints
+ *  - Clique path constraints
+ *
+ *
+ *     \min \sum_{ (i,j) \in Q }  w_ij y_ij
+ *
+ *             y_ij                            =  1                    (i,j) \in E
+ *
+ *             \sum_c y_nc                     =  |C| - 1              n \in N, c \in C
+ *
+ *             y_nc                            =  1                    n \in N, c \not\in C(n)
+ *
+ *             \sum_{e \in p} y_e      >= 1                    p \in P         path constraints
+ *
+ *             \sum_{e \in cp} y_e     >= |cp| - 1             cp \in CP       clique-path constraints
+ *
+ *             y_ij \in N,   w_ij \in R^+
+ */
+
+#include "becopyilp_t.h"
+
+int co_solve_ilp2(copy_opt_t *co, double time_limit) {
+       int res = 1;
+
+       return res;
+}
diff --git a/ir/be/becopyilp_t.h b/ir/be/becopyilp_t.h
new file mode 100644 (file)
index 0000000..54869c5
--- /dev/null
@@ -0,0 +1,41 @@
+/**
+ * Author:      Daniel Grund
+ * Date:               28.02.2006
+ * Copyright:   (c) Universitaet Karlsruhe
+ * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
+ *
+ * Common stuff used by all ILP fomulations.
+ *
+ */
+
+#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 <lpp/lpp.h>
+#include <lpp/lpp_net.h>
+#include <lpp/lpp_cplex.h>
+#include <lpp/lpp_remote.h>
+#include "xmalloc.h"
+#include "pset.h"
+#include "irprog.h"
+#include "irdom_t.h"
+#include "iredges_t.h"
+#include "bechordal_t.h"
+#include "becopyopt_t.h"
+#include "becopystat.h"
+#include "besched_t.h"
+#include "phiclass.h"
+
+#define LPP_HOST "i44pc52"
+#define LPP_SOLVER "cplex"
+
+#define MAX(a,b) ((a<b)?(b):(a))
+#define MIN(a,b) ((a<b)?(a):(b))
+#define EPSILON 0.00001
index 239ef03..54992cd 100644 (file)
 #include <malloc.h>
 #endif
 
+#include <libcore/lc_timing.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 "xmalloc.h"
+#include "bearch.h"
 #include "beutil.h"
-#include "bechordal_t.h"
-#include "becopyopt.h"
+#include "becopyopt_t.h"
 #include "becopystat.h"
 
-static firm_dbg_module_t *dbg = NULL;
-
-#define is_curr_reg_class(irn) \
-  (arch_get_irn_reg_class(get_arch_env(co), \
-                          irn, -1) == co->chordal_env->cls)
 
-#define MIN(a,b) ((a<b)?(a):(b))
-#define MAX(a,b) ((a<b)?(b):(a))
 
+static firm_dbg_module_t *dbg = NULL;
 
 /**
  * Determines a maximum weighted independent set with respect to
  * the interference and conflict edges of all nodes in a qnode.
  */
 static int ou_max_ind_set_costs(unit_t *ou) {
-       be_chordal_env_t *chordal_env = ou->co->chordal_env;
+       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;
@@ -124,11 +126,10 @@ static void co_collect_units(ir_node *irn, void *env) {
        copy_opt_t *co = env;
        unit_t *unit;
        arch_register_req_t req;
-       arch_env_t *aenv = co->chordal_env->main_env->arch_env;
 
-       if (!is_curr_reg_class(irn))
+       if (!is_curr_reg_class(co, irn))
                return;
-       if (!is_optimizable(get_arch_env(co), irn, &req))
+       if (!co_is_optimizable(co->aenv, irn, &req))
                return;
 
        /* Init a new unit */
@@ -152,10 +153,10 @@ static void co_collect_units(ir_node *irn, void *env) {
                        int o, arg_pos;
                        ir_node *arg = get_irn_n(irn, i);
 
-                       assert(is_curr_reg_class(arg) && "Argument not in same register class.");
+                       assert(is_curr_reg_class(co, arg) && "Argument not in same register class.");
                        if (arg == irn)
                                continue;
-                       if (nodes_interfere(co->chordal_env, irn, arg)) {
+                       if (nodes_interfere(co->cenv, irn, arg)) {
                                unit->inevitable_costs += co->get_costs(irn, arg, i);
                                continue;
                        }
@@ -186,8 +187,8 @@ static void co_collect_units(ir_node *irn, void *env) {
        } else
 
        /* Proj of a perm with corresponding arg */
-       if (is_Perm_Proj(get_arch_env(co), irn)) {
-               assert(!nodes_interfere(co->chordal_env, irn, get_Copy_src(irn)));
+       if (is_Perm_Proj(co->aenv, irn)) {
+               assert(!nodes_interfere(co->cenv, irn, get_Copy_src(irn)));
                unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
                unit->costs = xmalloc(2 * sizeof(*unit->costs));
                unit->node_count = 2;
@@ -197,9 +198,9 @@ static void co_collect_units(ir_node *irn, void *env) {
        } else
 
        /* Src == Tgt of a 2-addr-code instruction */
-       if (is_2addr_code(get_arch_env(co), irn, &req)) {
+       if (is_2addr_code(co->aenv, irn, &req)) {
                ir_node *other = req.other_same;
-               if (!nodes_interfere(co->chordal_env, irn, other)) {
+               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;
@@ -234,6 +235,10 @@ static void co_collect_units(ir_node *irn, void *env) {
        }
 }
 
+void be_copy_opt_init(void) {
+       dbg = firm_dbg_register("ir.be.copyoptmain");
+}
+
 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;
@@ -242,11 +247,14 @@ copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, int (*get_costs)(ir_node
        dbg = firm_dbg_register("ir.be.copyopt");
 
        co = xcalloc(1, sizeof(*co));
-       co->chordal_env = chordal_env;
+       co->cenv = chordal_env;
+       co->aenv = chordal_env->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(get_irg(co)));
+       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);
@@ -254,7 +262,7 @@ copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, int (*get_costs)(ir_node
 
        DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
        INIT_LIST_HEAD(&co->units);
-       irg_walk_graph(get_irg(co), co_collect_units, NULL, co);
+       irg_walk_graph(co->irg, co_collect_units, NULL, co);
        return co;
 }
 
@@ -269,21 +277,20 @@ void free_copy_opt(copy_opt_t *co) {
 }
 
 int is_optimizable_arg(const copy_opt_t *co, ir_node *irn) {
-       arch_env_t *aenv = get_arch_env(co);
        const ir_edge_t *edge;
 
-       if (arch_irn_is_ignore(aenv, irn))
+       if (arch_irn_is_ignore(co->aenv, irn))
                return 0;
 
        foreach_out_edge(irn, edge) {
                ir_node *n = edge->src;
 
-               if (!nodes_interfere(co->chordal_env, irn, n) || irn == n) {
+               if (!nodes_interfere(co->cenv, irn, n) || irn == n) {
                        arch_register_req_t req;
-                       arch_get_register_req(aenv, &req, n, -1);
+                       arch_get_register_req(co->aenv, &req, n, -1);
 
                        if(is_Reg_Phi(n) ||
-                          is_Perm(aenv, n) ||
+                          is_Perm(co->aenv, n) ||
                           (arch_register_req_is(&req, should_be_same) && req.other_same == irn)
                          )
                                return 1;
@@ -363,3 +370,155 @@ int co_get_lower_bound(const copy_opt_t *co) {
                res += curr->inevitable_costs + curr->min_nodes_costs;
        return res;
 }
+
+
+
+#define DO_HEUR
+#undef DO_CLASSES
+#undef DO_ILP
+
+
+/**
+ * Helpers for saving and restoring colors of nodes.
+ * Used to get dependable and comparable benchmark results.
+ */
+#if (defined(DO_HEUR) && defined(DO_BETTER)) || (defined(DO_HEUR) && defined(DO_ILP)) || (defined(DO_BETTER) && defined(DO_ILP))
+
+typedef struct color_saver {
+       arch_env_t *arch_env;
+       be_chordal_env_t *chordal_env;
+       pmap *saved_colors;
+       int flag; /* 0 save, 1 load */
+} color_save_t;
+
+static void save_load(ir_node *irn, void *env) {
+       color_save_t *saver = env;
+       if (saver->chordal_env->cls == arch_get_irn_reg_class(saver->arch_env, irn, -1)) {
+               if (saver->flag == 0) { /* save */
+                       const arch_register_t *reg = arch_get_irn_register(saver->arch_env, irn);
+                       pmap_insert(saver->saved_colors, irn, (void *) reg);
+               } else { /*load */
+                       arch_register_t *reg = pmap_get(saver->saved_colors, irn);
+                       arch_set_irn_register(saver->arch_env, irn, reg);
+               }
+       }
+}
+
+static void save_colors(color_save_t *color_saver) {
+       color_saver->flag = 0;
+       irg_walk_graph(color_saver->chordal_env->irg, save_load, NULL, color_saver);
+}
+
+static void load_colors(color_save_t *color_saver) {
+       color_saver->flag = 1;
+       irg_walk_graph(color_saver->chordal_env->irg, save_load, NULL, color_saver);
+}
+
+#endif /* Need save/load stuff */
+
+
+
+void co_compare_solvers(be_chordal_env_t *chordal_env) {
+       copy_opt_t *co;
+
+#ifdef DO_STAT
+       lc_timer_t *timer;
+       color_save_t saver;
+       int costs, costs_inevit, costs_init, costs_heur, costs_classes, costs_ilp, lower_bound;
+#endif
+
+       co = new_copy_opt(chordal_env, get_costs_loop_depth);
+       DBG((dbg, LEVEL_1, "----> CO: %s\n", co->name));
+       phi_class_compute(chordal_env->irg);
+
+
+#ifdef DO_STAT
+#if (defined(DO_HEUR) && defined(DO_BETTER)) || (defined(DO_HEUR) && defined(DO_ILP)) || (defined(DO_BETTER) && defined(DO_ILP))
+               saver.arch_env = chordal_env->main_env->arch_env;
+               saver.chordal_env = chordal_env;
+               saver.saved_colors = pmap_create();
+               save_colors(&saver);
+#endif
+
+               costs_inevit = co_get_inevit_copy_costs(co);
+               lower_bound  = co_get_lower_bound(co);
+               costs_init   = co_get_copy_costs(co);
+
+               DBG((dbg, LEVEL_1, "Inevit Costs: %3d\n", costs_inevit));
+               DBG((dbg, LEVEL_1, "Lower Bound: %3d\n", lower_bound));
+               DBG((dbg, LEVEL_1, "Init costs: %3d\n", costs_init));
+
+               copystat_add_inevit_costs(costs_inevit);
+               copystat_add_init_costs(costs_init);
+               copystat_add_max_costs(co_get_max_copy_costs(co));
+#endif
+
+
+#ifdef DO_HEUR
+#ifdef DO_STAT
+       timer = lc_timer_register("heur", NULL);
+       lc_timer_reset_and_start(timer);
+#endif
+
+       co_solve_heuristic(co);
+
+#ifdef DO_STAT
+       lc_timer_stop(timer);
+       costs_heur = co_get_copy_costs(co);
+       DBG((dbg, LEVEL_1, "Heur costs: %3d\n", costs_heur));
+       copystat_add_heur_time(lc_timer_elapsed_msec(timer));
+       copystat_add_heur_costs(costs_heur);
+       assert(lower_bound <= costs_heur);
+#endif
+#endif /* DO_HEUR */
+
+
+
+#ifdef DO_CLASSES
+#ifdef DO_STAT
+#ifdef DO_HEUR
+       load_colors(&saver);
+#endif
+       timer = lc_timer_register("classes", NULL);
+       lc_timer_reset_and_start(timer);
+#endif
+
+       co_classes_opt(co);
+
+#ifdef DO_STAT
+       lc_timer_stop(timer);
+       costs_classes = co_get_copy_costs(co);
+       DBG((dbg, LEVEL_1, "Classes costs: %3d\n", costs_classes));
+       copystat_add_classes_time(lc_timer_elapsed_msec(timer));
+       copystat_add_classes_costs(costs_heur);
+       assert(lower_bound <= costs_classes);
+#endif
+#endif /* DO_CLASSES */
+
+
+
+#ifdef DO_ILP
+#ifdef DO_STAT
+#if defined(DO_HEUR) || defined(DO_CLASSES)
+       load_colors(&saver);
+#endif
+#endif
+
+       co_solve_ilp1(co, 60.0);
+
+#ifdef DO_STAT
+       costs_ilp = co_get_copy_costs(co);
+       DBG((dbg, LEVEL_1, "Opt  costs: %3d\n", costs_ilp));
+       copystat_add_opt_costs(costs_ilp);
+       assert(lower_bound <= costs_ilp);
+#endif
+#endif /* DO_ILP */
+
+
+#ifdef DO_STAT
+#if (defined(DO_HEUR) && defined(DO_BETTER)) || (defined(DO_HEUR) && defined(DO_ILP)) || (defined(DO_BETTER) && defined(DO_ILP))
+       pmap_destroy(saver.saved_colors);
+#endif
+#endif
+       free_copy_opt(co);
+}
index 0e2f167..cc0baa6 100644 (file)
 /**
  * Author:      Daniel Grund
- * Date:               12.04.2005
+ * Date:               11.04.2005
  * Copyright:   (c) Universitaet Karlsruhe
  * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
  *
- * Header for copy optimization problem. Analysis and set up of the problem.
+ * Main file for the optimization reducing the copies needed for:
+ * - Phi coalescing
+ * - Register-constrained nodes
+ * - Two-address code instructions
  */
 
 #ifndef _BECOPYOPT_H
 #define _BECOPYOPT_H
 
-#include "debug.h"
-#include "obst.h"
-#include "list.h"
-#include "set.h"
-#include "pset.h"
-#include "bitset.h"
-
-#include "irgraph.h"
-#include "irgwalk.h"
 #include "irnode.h"
-#include "irdom.h"
-#include "irouts.h"
-
-#include "beutil.h"
-#include "benumb_t.h"
-#include "belive_t.h"
-#include "bechordal_t.h"
-#include "bearch.h"
+#include "bechordal.h"
 
-#define MIS_HEUR_TRIGGER 8
 
 typedef int(*cost_fct_t)(ir_node*, ir_node*, int);
 
-/**
- * Data representing the problem of copy minimization.
- */
-typedef struct _copy_opt_t {
-       be_chordal_env_t *chordal_env;
-       char *name;                                             /**< ProgName__IrgName__RegClass */
-       struct list_head units;                 /**< all units to optimize in specific order */
-       cost_fct_t get_costs;                   /**< function ptr used to get costs for copies */
-       struct obstack ob;
-} copy_opt_t;
+typedef struct _copy_opt_t copy_opt_t;
 
 /**
- * A single unit of optimization. Lots of these form a copy-opt problem
+ * Has to be called during the firm init phase
  */
-typedef struct _unit_t {
-       struct list_head units;         /**< chain for all units */
-       copy_opt_t *co;                         /**< the copy_opt this unit belongs to */
-       int node_count;                         /**< size of the nodes array */
-       ir_node **nodes;                        /**< [0] is the root-node, others are non interfering args of it. */
-       int *costs;                                     /**< costs[i] are arising, if nodes[i] has a different color */
-       int inevitable_costs;           /**< sum of costs of all args interfering with root */
-       int all_nodes_costs;            /**< sum of all costs[i] */
-       int min_nodes_costs;            /**< a lower bound for the costs in costs[], determined by a max indep. set */
-       int sort_key;                           /**< maximum costs. controls the order of ou's in the struct list_head units. */
-
-       /* for heuristic */
-       struct list_head queue;         /**< list of qn's sorted by weight of qn-mis */
-} unit_t;
-
-/* Helpers */
-#define get_arch_env(co) ((co)->chordal_env->main_env->arch_env)
-#define get_irg(co)      ((co)->chordal_env->irg)
-#define get_irn_col(co, irn) \
-       arch_register_get_index(arch_get_irn_register(get_arch_env(co), irn))
-#define set_irn_col(co, irn, col) \
-       arch_set_irn_register(get_arch_env(co), irn, arch_register_for_index(co->chordal_env->cls, col))
-
-
-#define list_entry_units(lh) list_entry(lh, unit_t, units)
-
-#define get_Copy_src(irn) (get_irn_n(get_Proj_pred(irn), get_Proj_proj(irn)))
-#define is_Perm(arch_env, irn)                         (arch_irn_classify(arch_env, irn) == arch_irn_class_perm)
-#define is_Reg_Phi(irn)                                                (is_Phi(irn) && mode_is_data(get_irn_mode(irn)))
-#define is_Perm_Proj(arch_env, irn)                    (is_Proj(irn) && is_Perm(arch_env, get_Proj_pred(irn)))
-#define is_2addr_code(arch_env, irn, req)      (arch_get_register_req(arch_env, req, irn, -1)->type == arch_register_req_type_should_be_same)
-
+void be_copy_opt_init(void);
 
 /**
- * Generate the problem. Collect all infos and optimizable nodes.
+ * Generate the problem. Collect all information and optimizable nodes.
  */
-copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, int (*get_costs)(ir_node*, ir_node*, int));
+copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs);
 
 /**
- * Free the space...
+ * Free the space used...
  */
 void free_copy_opt(copy_opt_t *co);
 
-/**
- * Checks if a node is optimizable, viz. has somthing to do with coalescing
- * @param arch The architecture environment
- * @param irn  The irn to check
- * @param req  A register_requirement structure (used to check for 2-addr-code)
- */
-#define is_optimizable(arch, irn, req) (!arch_irn_is_ignore(arch, irn) && (is_Reg_Phi(irn) || is_Perm_Proj(arch, irn) || is_2addr_code(arch, irn, req)))
-
-/**
- * Checks if the irn is a non-interfering argument of a node which 'is_optimizable'
- */
-int is_optimizable_arg(const copy_opt_t *co, ir_node *irn);
-
 /**
  * Computes the costs of a copy according to loop depth
  * @param pos: the argument position of arg in the root arguments
  * @return Must be >= 0 in all cases.
  */
-int get_costs_loop_depth(ir_node *root, ir_node* arg, int pos);
+int co_get_costs_loop_depth(ir_node *root, ir_node* arg, int pos);
 
 /**
  * All costs equal 1. Using this will reduce the number of copies.
  * @return Must be >= 0 in all cases.
  */
-int get_costs_all_one(ir_node *root, ir_node* arg, int pos);
-
-/**
- * Returns the maximal costs possible, i.e. the costs if all
- * pairs would be assigned different registers.
- */
-int co_get_max_copy_costs(const copy_opt_t *co);
+int co_get_costs_all_one(ir_node *root, ir_node* arg, int pos);
 
 /**
- * Returns the inevitable costs, i.e. the costs of
- * all copy pairs which interfere.
- */
-int co_get_inevit_copy_costs(const copy_opt_t *co);
-
-/**
- * Returns the current costs the copies are causing.
- * The result includes inevitable costs and the costs
- * of the copies regarding the current register allocation
+ * Solves the problem using a heuristic approach
  */
-int co_get_copy_costs(const copy_opt_t *co);
+int co_solve_heuristic(copy_opt_t *co);
 
 /**
- * Returns a lower bound for the costs of copies in this ou.
- * The result includes inevitable costs and the costs of a
- * minimal costs caused by the nodes of the ou.
+ * Solves the problem using mixed integer programming
+ * @returns 1 iff solution state was optimal
  */
-int co_get_lower_bound(const copy_opt_t *co);
+int co_solve_ilp1(copy_opt_t *co, double time_limit);
 
 /**
- * Solves the problem using a heuristic approach
+ * Solves the problem using mixed integer programming
+ * @returns 1 iff solution state was optimal
  */
-void co_heur_opt(copy_opt_t *co);
+int co_solve_ilp2(copy_opt_t *co, double time_limit);
 
 /**
- * Solves the problem using mixed integer programming
- * @returns 1 iff solution state was optimal
+ * Compares different solutions of the same problem
  */
-int co_ilp_opt(copy_opt_t *co, double time_limit);
+void co_compare_solvers(be_chordal_env_t *chordal_env);
 
 #endif
diff --git a/ir/be/becopyopt_t.h b/ir/be/becopyopt_t.h
new file mode 100644 (file)
index 0000000..f16eca8
--- /dev/null
@@ -0,0 +1,110 @@
+/**
+ * Author:      Daniel Grund
+ * Date:               12.04.2005
+ * Copyright:   (c) Universitaet Karlsruhe
+ * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
+ *
+ * Header for copy optimization problem. Analysis and set up of the problem.
+ */
+
+#ifndef _BECOPYOPT_T_H
+#define _BECOPYOPT_T_H
+
+#include "list.h"
+#include "bearch.h"
+#include "bechordal_t.h"
+#include "becopyopt.h"
+
+#define MIS_HEUR_TRIGGER 8
+
+/**
+ * Data representing the problem of copy minimization.
+ */
+struct _copy_opt_t {
+       be_chordal_env_t *cenv;
+       const arch_register_class_t *cls;
+       const arch_env_t *aenv;
+       ir_graph *irg;
+       char *name;                                             /**< ProgName__IrgName__RegClass */
+
+       struct list_head units;                 /**< all units to optimize in specific order */
+       cost_fct_t get_costs;                   /**< function ptr used to get costs for copies */
+       struct obstack ob;
+};
+
+/**
+ * A single unit of optimization. Lots of these form a copy-opt problem
+ */
+typedef struct _unit_t {
+       struct list_head units;         /**< chain for all units */
+       copy_opt_t *co;                         /**< the copy_opt this unit belongs to */
+       int node_count;                         /**< size of the nodes array */
+       ir_node **nodes;                        /**< [0] is the root-node, others are non interfering args of it. */
+       int *costs;                                     /**< costs[i] are arising, if nodes[i] has a different color */
+       int inevitable_costs;           /**< sum of costs of all args interfering with root */
+       int all_nodes_costs;            /**< sum of all costs[i] */
+       int min_nodes_costs;            /**< a lower bound for the costs in costs[], determined by a max indep. set */
+       int sort_key;                           /**< maximum costs. controls the order of ou's in the struct list_head units. */
+
+       /* for heuristic */
+       struct list_head queue;         /**< list of qn's sorted by weight of qn-mis */
+} unit_t;
+
+/* Helpers */
+#define get_irn_col(co, irn)           arch_register_get_index(arch_get_irn_register((co)->aenv, irn))
+#define set_irn_col(co, irn, col)      arch_set_irn_register((co)->aenv, irn, arch_register_for_index((co)->cls, col))
+#define is_curr_reg_class(co, irn)     (arch_get_irn_reg_class((co)->aenv, irn, -1) == (co)->cls)
+
+#define MIN(a,b) ((a<b)?(a):(b))
+#define MAX(a,b) ((a<b)?(b):(a))
+
+#define list_entry_units(lh) list_entry(lh, unit_t, units)
+
+
+#define get_Copy_src(irn) (get_irn_n(get_Proj_pred(irn), get_Proj_proj(irn)))
+#define is_Perm(arch_env, irn)                         (arch_irn_classify(arch_env, irn) == arch_irn_class_perm)
+#define is_Reg_Phi(irn)                                                (is_Phi(irn) && mode_is_data(get_irn_mode(irn)))
+#define is_Perm_Proj(arch_env, irn)                    (is_Proj(irn) && is_Perm(arch_env, get_Proj_pred(irn)))
+#define is_2addr_code(arch_env, irn, req)      (arch_get_register_req(arch_env, req, irn, -1)->type == arch_register_req_type_should_be_same)
+
+/**
+ * Checks if a node is optimizable, viz. has somthing to do with coalescing
+ * @param arch The architecture environment
+ * @param irn  The irn to check
+ * @param req  A register_requirement structure (used to check for 2-addr-code)
+ */
+#define co_is_optimizable(arch, irn, req) (!arch_irn_is_ignore(arch, irn) && (is_Reg_Phi(irn) || is_Perm_Proj(arch, irn) || is_2addr_code(arch, irn, req)))
+
+/**
+ * Checks if the irn is a non-interfering argument of a node which 'is_optimizable'
+ */
+int co_is_optimizable_arg(const copy_opt_t *co, ir_node *irn);
+
+/**
+ * Returns the maximal costs possible, i.e. the costs if all
+ * pairs would be assigned different registers.
+ */
+int co_get_max_copy_costs(const copy_opt_t *co);
+
+/**
+ * Returns the inevitable costs, i.e. the costs of
+ * all copy pairs which interfere.
+ */
+int co_get_inevit_copy_costs(const copy_opt_t *co);
+
+/**
+ * Returns the current costs the copies are causing.
+ * The result includes inevitable costs and the costs
+ * of the copies regarding the current register allocation
+ */
+int co_get_copy_costs(const copy_opt_t *co);
+
+/**
+ * Returns a lower bound for the costs of copies in this ou.
+ * The result includes inevitable costs and the costs of a
+ * minimal costs caused by the nodes of the ou.
+ */
+int co_get_lower_bound(const copy_opt_t *co);
+
+
+#endif