typo fixed
[libfirm] / ir / be / belistsched.c
index 3aedf5b..cf31ea4 100644 (file)
@@ -1,10 +1,29 @@
-/**
- * Scheduling algorithms.
- * Just a simple list scheduling algorithm is here.
- * @date 20.10.2004
- * @author Sebastian Hack
+/*
+ * Copyright (C) 1995-2008 University of Karlsruhe.  All right reserved.
+ *
+ * This file is part of libFirm.
+ *
+ * This file may be distributed and/or modified under the terms of the
+ * GNU General Public License version 2 as published by the Free Software
+ * Foundation and appearing in the file LICENSE.GPL included in the
+ * packaging of this file.
+ *
+ * Licensees holding valid libFirm Professional Edition licenses may use
+ * this file in accordance with the libFirm Commercial License.
+ * Agreement provided with the Software.
+ *
+ * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
+ * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE.
  */
 
+/**
+ * @file
+ * @brief       Primitive list scheduling with different node selectors.
+ * @author      Sebastian Hack
+ * @date        20.10.2004
+ * @version     $Id$
+ */
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
@@ -15,6 +34,7 @@
 #include <limits.h>
 
 #include "benode_t.h"
+#include "be_t.h"
 
 #include "obst.h"
 #include "list.h"
 #include "debug.h"
 #include "irtools.h"
 
+#include "bemodule.h"
 #include "besched_t.h"
 #include "beutil.h"
 #include "belive_t.h"
 #include "belistsched.h"
 #include "beschedmris.h"
-#include "bearch.h"
+#include "beschedrss.h"
+#include "bearch_t.h"
 #include "bestat.h"
+#include "beirg_t.h"
+
+#include "lc_opts.h"
+#include "lc_opts_enum.h"
+
+DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL);
+
+#define BE_SCHED_NODE(irn) (be_is_Keep(irn) || be_is_CopyKeep(irn) || be_is_RegParams(irn))
+
+enum {
+       BE_SCHED_SELECT_TRIVIAL,
+       BE_SCHED_SELECT_REGPRESS,
+       BE_SCHED_SELECT_MUCHNIK,
+       BE_SCHED_SELECT_HEUR,
+       BE_SCHED_SELECT_HMUCHNIK,
+       BE_SCHED_SELECT_RANDOM,
+       BE_SCHED_SELECT_NORMAL,
+};
+
+enum {
+       BE_SCHED_PREP_NONE = 0,
+       BE_SCHED_PREP_MRIS = 2,
+       BE_SCHED_PREP_RSS  = 3
+};
+
+typedef struct _list_sched_options_t {
+       int select;  /**< the node selector */
+       int prep;    /**< schedule preparation */
+} list_sched_options_t;
+
+static list_sched_options_t list_sched_options = {
+       BE_SCHED_SELECT_HEUR,     /* mueller heuristic selector */
+       BE_SCHED_PREP_NONE,       /* no scheduling preparation */
+};
+
+/* schedule selector options. */
+static const lc_opt_enum_int_items_t sched_select_items[] = {
+       { "trivial",  BE_SCHED_SELECT_TRIVIAL  },
+       { "random",   BE_SCHED_SELECT_RANDOM   },
+       { "regpress", BE_SCHED_SELECT_REGPRESS },
+       { "normal",   BE_SCHED_SELECT_NORMAL   },
+       { "muchnik",  BE_SCHED_SELECT_MUCHNIK  },
+       { "heur",     BE_SCHED_SELECT_HEUR     },
+       { "hmuchnik", BE_SCHED_SELECT_HMUCHNIK },
+       { NULL,       0 }
+};
+
+/* schedule preparation options. */
+static const lc_opt_enum_int_items_t sched_prep_items[] = {
+       { "none", BE_SCHED_PREP_NONE },
+       { "mris", BE_SCHED_PREP_MRIS },
+       { "rss",  BE_SCHED_PREP_RSS  },
+       { NULL,   0 }
+};
+
+static lc_opt_enum_int_var_t sched_select_var = {
+       &list_sched_options.select, sched_select_items
+};
+
+static lc_opt_enum_int_var_t sched_prep_var = {
+       &list_sched_options.prep, sched_prep_items
+};
+
+static const lc_opt_table_entry_t list_sched_option_table[] = {
+       LC_OPT_ENT_ENUM_PTR("prep",   "schedule preparation",   &sched_prep_var),
+       LC_OPT_ENT_ENUM_PTR("select", "node selector",          &sched_select_var),
+       LC_OPT_LAST
+};
 
 /**
  * All scheduling info needed per node.
  */
 typedef struct _sched_irn_t {
-       sched_timestep_t delay;     /**< The delay for this node if already calculated, else 0. */
-       sched_timestep_t etime;     /**< The earliest time of this node. */
-       unsigned already_sched : 1; /**< Set if this node is already scheduled */
-       unsigned is_root       : 1; /**< is a root node of a block */
+       unsigned num_not_sched_user; /**< The number of not yet scheduled users of this node */
+       unsigned already_sched : 1;  /**< Set if this node is already scheduled */
 } sched_irn_t;
 
 /**
@@ -59,470 +147,60 @@ typedef struct _sched_env_t {
        void *selector_env;                         /**< A pointer to give to the selector. */
 } sched_env_t;
 
-#if 0
-/*
- * Ugly global variable for the compare function
- * since qsort(3) does not pass an extra pointer.
- */
-static ir_node *curr_bl = NULL;
-
-static int cmp_usage(const void *a, const void *b)
-{
-       struct trivial_sched_env *env;
-       const ir_node *p = a;
-       const ir_node *q = b;
-       int res = 0;
-
-       res = is_live_end(env->curr_bl, a) - is_live_end(env->curr_bl, b);
-
-       /*
-        * One of them is live at the end of the block.
-        * Then, that one shall be scheduled at after the other
-        */
-       if(res != 0)
-               return res;
-
-
-       return res;
-}
-#endif
-
-/**
- * The trivial selector:
- * Just assure that branches are executed last, otherwise select
- * the first node ready.
- */
-static ir_node *trivial_select(void *block_env, nodeset *ready_set)
-{
-       const arch_env_t *arch_env = block_env;
-       ir_node *irn = NULL;
-       int const_last = 0;
-
-       /* assure that branches and constants are executed last */
-       for (irn = nodeset_first(ready_set); irn; irn = nodeset_next(ready_set)) {
-               arch_irn_class_t irn_class = arch_irn_classify(arch_env, irn);
-
-               if (irn_class != arch_irn_class_branch && (const_last ? (irn_class != arch_irn_class_const) : 1)) {
-                       nodeset_break(ready_set);
-                       return irn;
-               }
-       }
-
-       /* assure that constants are executed before branches */
-       if (const_last) {
-               for (irn = nodeset_first(ready_set); irn; irn = nodeset_next(ready_set)) {
-                       if (arch_irn_classify(arch_env, irn) != arch_irn_class_branch) {
-                               nodeset_break(ready_set);
-                               return irn;
-                       }
-               }
-       }
-
-
-       /* at last: schedule branches */
-       irn = nodeset_first(ready_set);
-       nodeset_break(ready_set);
-
-       return irn;
-}
-
-static void *trivial_init_graph(const list_sched_selector_t *vtab, const arch_env_t *arch_env, ir_graph *irg)
-{
-       return (void *) arch_env;
-}
-
-static void *trivial_init_block(void *graph_env, ir_node *bl)
-{
-       return graph_env;
-}
-
-static INLINE int must_appear_in_schedule(const list_sched_selector_t *sel, void *block_env, const ir_node *irn)
-{
-       int res = 0;
-
-       if(sel->to_appear_in_schedule)
-               res = sel->to_appear_in_schedule(block_env, irn);
-
-       return res || to_appear_in_schedule(irn) || be_is_Keep(irn) || be_is_RegParams(irn);
-}
-
-static const list_sched_selector_t trivial_selector_struct = {
-       trivial_init_graph,
-       trivial_init_block,
-       trivial_select,
-       NULL,                /* to_appear_in_schedule */
-       NULL,                /* exectime */
-       NULL,                /* latency */
-       NULL,                /* finish_block */
-       NULL                 /* finish_graph */
-};
-
-const list_sched_selector_t *trivial_selector = &trivial_selector_struct;
-
-typedef struct _usage_stats_t {
-       ir_node *irn;
-       struct _usage_stats_t *next;
-       int max_hops;
-       int uses_in_block;      /**< Number of uses inside the current block. */
-       int already_consumed;   /**< Number of insns using this value already
-                                                         scheduled. */
-} usage_stats_t;
-
-typedef struct {
-       const list_sched_selector_t *vtab;
-       const arch_env_t *arch_env;
-} reg_pressure_main_env_t;
-
-typedef struct {
-       struct obstack obst;
-       const reg_pressure_main_env_t *main_env;
-       usage_stats_t *root;
-       nodeset *already_scheduled;
-} reg_pressure_selector_env_t;
-
-static INLINE usage_stats_t *get_or_set_usage_stats(reg_pressure_selector_env_t *env, ir_node *irn)
-{
-       usage_stats_t *us = get_irn_link(irn);
-
-       if(!us) {
-               us                   = obstack_alloc(&env->obst, sizeof(us[0]));
-               us->irn              = irn;
-               us->already_consumed = 0;
-               us->max_hops         = INT_MAX;
-               us->next             = env->root;
-               env->root            = us;
-               set_irn_link(irn, us);
-       }
-
-       return us;
-}
-
-static INLINE usage_stats_t *get_usage_stats(ir_node *irn)
-{
-       usage_stats_t *us = get_irn_link(irn);
-       assert(us && "This node must have usage stats");
-       return us;
-}
-
-static int max_hops_walker(reg_pressure_selector_env_t *env, ir_node *irn, ir_node *curr_bl, int depth, unsigned visited_nr)
-{
-       ir_node *bl = get_nodes_block(irn);
-       /*
-        * If the reached node is not in the block desired,
-        * return the value passed for this situation.
-        */
-       if(get_nodes_block(irn) != bl)
-               return block_dominates(bl, curr_bl) ? 0 : INT_MAX;
-
-       /*
-        * If the node is in the current block but not
-        * yet scheduled, we keep on searching from that node.
-        */
-       if(!nodeset_find(env->already_scheduled, irn)) {
-               int i, n;
-               int res = 0;
-               for(i = 0, n = get_irn_arity(irn); i < n; ++i) {
-                       ir_node *operand = get_irn_n(irn, i);
-
-                       if(get_irn_visited(operand) < visited_nr) {
-                               int tmp;
-
-                               set_irn_visited(operand, visited_nr);
-                               tmp = max_hops_walker(env, operand, bl, depth + 1, visited_nr);
-                               res = MAX(tmp, res);
-                       }
-               }
-
-               return res;
-       }
-
-       /*
-        * If the node is in the current block and scheduled, return
-        * the depth which indicates the number of steps to the
-        * region of scheduled nodes.
-        */
-       return depth;
-}
-
-static int compute_max_hops(reg_pressure_selector_env_t *env, ir_node *irn)
-{
-       ir_node *bl   = get_nodes_block(irn);
-       ir_graph *irg = get_irn_irg(bl);
-       int res       = 0;
-
-       const ir_edge_t *edge;
-
-       foreach_out_edge(irn, edge) {
-               ir_node *user       = get_edge_src_irn(edge);
-               unsigned visited_nr = get_irg_visited(irg) + 1;
-               int max_hops;
-
-               set_irg_visited(irg, visited_nr);
-               max_hops = max_hops_walker(env, user, irn, 0, visited_nr);
-               res      = MAX(res, max_hops);
-       }
-
-       return res;
-}
-
-static void *reg_pressure_graph_init(const list_sched_selector_t *vtab, const arch_env_t *arch_env, ir_graph *irg)
-{
-       reg_pressure_main_env_t *main_env = xmalloc(sizeof(main_env[0]));
-
-       main_env->arch_env = arch_env;
-       main_env->vtab     = vtab;
-       irg_walk_graph(irg, firm_clear_link, NULL, NULL);
-
-       return main_env;
-}
-
-static void *reg_pressure_block_init(void *graph_env, ir_node *bl)
-{
-       ir_node *irn;
-       reg_pressure_selector_env_t *env  = xmalloc(sizeof(env[0]));
-
-       obstack_init(&env->obst);
-       env->already_scheduled = new_nodeset(32);
-       env->root              = NULL;
-       env->main_env          = graph_env;
-
-       /*
-        * Collect usage statistics.
-        */
-       sched_foreach(bl, irn) {
-               if(must_appear_in_schedule(env->main_env->vtab, env, irn)) {
-                       int i, n;
-
-                       for(i = 0, n = get_irn_arity(irn); i < n; ++i) {
-                               ir_node *op = get_irn_n(irn, i);
-                               if(must_appear_in_schedule(env->main_env->vtab, env, irn)) {
-                                       usage_stats_t *us = get_or_set_usage_stats(env, irn);
-                                       if(is_live_end(bl, op))
-                                               us->uses_in_block = 99999;
-                                       else
-                                               us->uses_in_block++;
-                               }
-                       }
-               }
-       }
-
-       return env;
-}
-
-static void reg_pressure_block_free(void *block_env)
-{
-       reg_pressure_selector_env_t *env = block_env;
-       usage_stats_t *us;
-
-       for(us = env->root; us; us = us->next)
-               set_irn_link(us->irn, NULL);
-
-       obstack_free(&env->obst, NULL);
-       del_nodeset(env->already_scheduled);
-       free(env);
-}
-
-static int get_result_hops_sum(reg_pressure_selector_env_t *env, ir_node *irn)
-{
-       int res = 0;
-       if(get_irn_mode(irn) == mode_T) {
-               const ir_edge_t *edge;
-
-               foreach_out_edge(irn, edge)
-                       res += get_result_hops_sum(env, get_edge_src_irn(edge));
-       }
-
-       else if(mode_is_data(get_irn_mode(irn)))
-               res = compute_max_hops(env, irn);
-
-
-       return res;
-}
-
-static INLINE int reg_pr_costs(reg_pressure_selector_env_t *env, ir_node *irn)
-{
-       int i, n;
-       int sum = 0;
-
-       for(i = 0, n = get_irn_arity(irn); i < n; ++i) {
-               ir_node *op = get_irn_n(irn, i);
-
-               if(must_appear_in_schedule(env->main_env->vtab, env, op))
-                       sum += compute_max_hops(env, op);
-       }
-
-       sum += get_result_hops_sum(env, irn);
-
-       return sum;
-}
-
-static ir_node *reg_pressure_select(void *block_env, nodeset *ready_set)
-{
-       reg_pressure_selector_env_t *env = block_env;
-       ir_node *irn, *res     = NULL;
-       int curr_cost          = INT_MAX;
-
-       assert(nodeset_count(ready_set) > 0);
-
-       for (irn = nodeset_first(ready_set); irn; irn = nodeset_next(ready_set)) {
-               /*
-                       Ignore branch instructions for the time being.
-                       They should only be scheduled if there is nothing else.
-               */
-               if (arch_irn_classify(env->main_env->arch_env, irn) != arch_irn_class_branch) {
-                       int costs = reg_pr_costs(env, irn);
-                       if (costs <= curr_cost) {
-                               res       = irn;
-                               curr_cost = costs;
-                       }
-               }
-       }
-
-       /*
-               There was no result so we only saw a branch.
-               Take it and finish.
-       */
-
-       if(!res) {
-               res = nodeset_first(ready_set);
-               nodeset_break(ready_set);
-
-               assert(res && "There must be a node scheduled.");
-       }
-
-       nodeset_insert(env->already_scheduled, res);
-       return res;
-}
-
 /**
  * Environment for a block scheduler.
  */
 typedef struct _block_sched_env_t {
        sched_irn_t *sched_info;                    /**< scheduling info per node, copied from the global scheduler object */
-       sched_timestep_t curr_time;                 /**< current time of the scheduler */
-       nodeset *cands;                             /**< the set of candidates */
+       ir_nodeset_t cands;                         /**< the set of candidates */
        ir_node *block;                             /**< the current block */
        sched_env_t *sched_env;                     /**< the scheduler environment */
+       ir_nodeset_t live;                          /**< simple liveness during scheduling */
        const list_sched_selector_t *selector;
        void *selector_block_env;
-       DEBUG_ONLY(firm_dbg_module_t *dbg;)
 } block_sched_env_t;
 
 /**
- * Returns non-zero if the node is already scheduled
+ * Returns non-zero if a node must be placed in the schedule.
  */
-static INLINE int is_already_scheduled(block_sched_env_t *env, ir_node *n)
+static INLINE int must_appear_in_schedule(const list_sched_selector_t *sel, void *block_env, const ir_node *irn)
 {
-       int idx = get_irn_idx(n);
+       int res = -1;
 
-       assert(idx < ARR_LEN(env->sched_info));
-       return env->sched_info[idx].already_sched;
-}
+       /* if there are no uses, don't schedule */
+       if (get_irn_n_edges(irn) < 1)
+               return 0;
 
-/**
- * Mark a node as already scheduled
- */
-static INLINE void mark_already_scheduled(block_sched_env_t *env, ir_node *n)
-{
-       int idx = get_irn_idx(n);
+       /* else ask the scheduler */
+       if (sel->to_appear_in_schedule)
+               res = sel->to_appear_in_schedule(block_env, irn);
 
-       assert(idx < ARR_LEN(env->sched_info));
-       env->sched_info[idx].already_sched = 1;
+       return res >= 0 ? res : ((to_appear_in_schedule(irn) || BE_SCHED_NODE(irn)) && ! is_Unknown(irn));
 }
 
 /**
- * Returns non-zero if the node is a root node
+ * Returns non-zero if the node is already scheduled
  */
-static INLINE unsigned is_root_node(block_sched_env_t *env, ir_node *n)
+static INLINE int is_already_scheduled(block_sched_env_t *env, ir_node *n)
 {
        int idx = get_irn_idx(n);
 
        assert(idx < ARR_LEN(env->sched_info));
-       return env->sched_info[idx].is_root;
+       return env->sched_info[idx].already_sched;
 }
 
 /**
- * Mark a node as roto node
+ * Mark a node as already scheduled
  */
-static INLINE void mark_root_node(block_sched_env_t *env, ir_node *n)
+static INLINE void set_already_scheduled(block_sched_env_t *env, ir_node *n)
 {
        int idx = get_irn_idx(n);
 
        assert(idx < ARR_LEN(env->sched_info));
-       env->sched_info[idx].is_root = 1;
-}
-
-/**
- * Get the current delay.
- */
-static sched_timestep_t get_irn_delay(block_sched_env_t *env, ir_node *n) {
-       int idx = get_irn_idx(n);
-
-       assert(idx < ARR_LEN(env->sched_info));
-       return env->sched_info[idx].delay;
-}
-
-/**
- * Set the current delay.
- */
-static void set_irn_delay(block_sched_env_t *env, ir_node *n, sched_timestep_t delay) {
-       int idx = get_irn_idx(n);
-
-       assert(idx < ARR_LEN(env->sched_info));
-       env->sched_info[idx].delay = delay;
-}
-
-/**
- * Get the current etime.
- */
-static sched_timestep_t get_irn_etime(block_sched_env_t *env, ir_node *n) {
-       int idx = get_irn_idx(n);
-
-       assert(idx < ARR_LEN(env->sched_info));
-       return env->sched_info[idx].etime;
-}
-
-/**
- * Set the current etime.
- */
-static void set_irn_etime(block_sched_env_t *env, ir_node *n, sched_timestep_t etime) {
-       int idx = get_irn_idx(n);
-
-       assert(idx < ARR_LEN(env->sched_info));
-       env->sched_info[idx].etime = etime;
-}
-
-/**
- * returns the exec-time for node n.
- */
-static sched_timestep_t exectime(sched_env_t *env, ir_node *n) {
-  if (be_is_Keep(n) || is_Proj(n))
-    return 0;
-       if (env->selector->exectime)
-               return env->selector->exectime(env->selector_env, n);
-       return 1;
+       env->sched_info[idx].already_sched = 1;
 }
 
-/**
- * Calculates the latency for between two ops
- */
-static sched_timestep_t latency(sched_env_t *env, ir_node *pred, int pred_cycle, ir_node *curr, int curr_cycle) {
-       /* a Keep hides a root */
-  if (be_is_Keep(curr))
-               return exectime(env, pred);
-
-       /* Proj's are executed immediately */
-       if (is_Proj(curr))
-    return 0;
-
-       /* predecessors Proj's must be skipped */
-  if (is_Proj(pred))
-    pred = get_Proj_pred(pred);
-
-       if (env->selector->latency)
-               return env->selector->latency(env->selector_env, pred, pred_cycle, curr, curr_cycle);
-       return 1;
-}
+static void add_to_sched(block_sched_env_t *env, ir_node *irn);
 
 /**
  * Try to put a node in the ready set.
@@ -533,49 +211,46 @@ static sched_timestep_t latency(sched_env_t *env, ir_node *pred, int pred_cycle,
  */
 static INLINE int make_ready(block_sched_env_t *env, ir_node *pred, ir_node *irn)
 {
-    int i, n;
-               sched_timestep_t etime_p, etime;
-
-    /* Blocks cannot be scheduled. */
-    if (is_Block(irn))
-        return 0;
-
-    /*
-     * Check, if the given ir node is in a different block as the
-     * currently scheduled one. If that is so, don't make the node ready.
-     */
-    if (env->block != get_nodes_block(irn))
-        return 0;
-
-    for (i = 0, n = get_irn_arity(irn); i < n; ++i) {
-        ir_node *op = get_irn_n(irn, i);
-
-        /* if irn is an End we have keep-alives and op might be a block, skip that */
-        if (is_Block(op)) {
-          assert(get_irn_op(irn) == op_End);
-          continue;
-        }
-
-        /* If the operand is local to the scheduled block and not yet
-         * scheduled, this nodes cannot be made ready, so exit. */
-        if (!is_already_scheduled(env, op) && get_nodes_block(op) == env->block)
-            return 0;
-    }
+       int i, n;
+
+       /* Blocks cannot be scheduled. */
+       if (is_Block(irn) || get_irn_n_edges(irn) == 0)
+               return 0;
 
-    nodeset_insert(env->cands, irn);
+       /*
+        * Check, if the given ir node is in a different block as the
+        * currently scheduled one. If that is so, don't make the node ready.
+        */
+       if (env->block != get_nodes_block(irn))
+               return 0;
 
-               /* calculate the etime of this node */
-               etime = env->curr_time;
-               if (pred) {
-                       etime_p  = get_irn_etime(env, pred);
-                       etime   += latency(env->sched_env, pred, 1, irn, 0);
+       for (i = 0, n = get_irn_ins_or_deps(irn); i < n; ++i) {
+               ir_node *op = get_irn_in_or_dep(irn, i);
 
-                       etime = etime_p > etime ? etime_p : etime;
+               /* if irn is an End we have keep-alives and op might be a block, skip that */
+               if (is_Block(op)) {
+                       assert(get_irn_op(irn) == op_End);
+                       continue;
                }
 
-               set_irn_etime(env, irn, etime);
+               /* If the operand is local to the scheduled block and not yet
+                * scheduled, this nodes cannot be made ready, so exit. */
+               if (! is_already_scheduled(env, op) && get_nodes_block(op) == env->block)
+                       return 0;
+       }
+
+       if (! must_appear_in_schedule(env->selector, env, irn)) {
+               add_to_sched(env, irn);
+               DB((dbg, LEVEL_3, "\tmaking immediately available: %+F\n", irn));
+       } else {
+               ir_nodeset_insert(&env->cands, irn);
+
+               /* Notify selector about the ready node. */
+               if (env->selector->node_ready)
+                       env->selector->node_ready(env->selector_block_env, irn, pred);
 
-    DB((env->dbg, LEVEL_2, "\tmaking ready: %+F etime %u\n", irn, etime));
+               DB((dbg, LEVEL_2, "\tmaking ready: %+F\n", irn));
+       }
 
     return 1;
 }
@@ -583,163 +258,158 @@ static INLINE int make_ready(block_sched_env_t *env, ir_node *pred, ir_node *irn
 /**
  * Try, to make all users of a node ready.
  * In fact, a usage node can only be made ready, if all its operands
- * have already been scheduled yet. This is checked my make_ready().
+ * have already been scheduled yet. This is checked by make_ready().
  * @param env The block schedule environment.
  * @param irn The node, which usages (successors) are to be made ready.
  */
-static INLINE void make_users_ready(block_sched_env_t *env, ir_node *irn)
+static void make_users_ready(block_sched_env_t *env, ir_node *irn)
 {
        const ir_edge_t *edge;
 
+       /* make all data users ready */
        foreach_out_edge(irn, edge) {
-               ir_node *user = edge->src;
-               if(!is_Phi(user))
+               ir_node *user = get_edge_src_irn(edge);
+
+               if (! is_Phi(user))
+                       make_ready(env, irn, user);
+       }
+
+       /* and the dependent nodes as well */
+       foreach_out_edge_kind(irn, edge, EDGE_KIND_DEP) {
+               ir_node *user = get_edge_src_irn(edge);
+
+               if (! is_Phi(user))
                        make_ready(env, irn, user);
        }
 }
 
 /**
- * Compare to nodes using pointer equality.
- * @param p1 Node one.
- * @param p2 Node two.
- * @return 0 if they are identical.
+ * Returns the number of not yet schedules users.
  */
-static int node_cmp_func(const void *p1, const void *p2)
-{
-    return p1 != p2;
+static INLINE int get_irn_not_sched_user(block_sched_env_t *env, ir_node *n) {
+       int idx = get_irn_idx(n);
+
+       assert(idx < ARR_LEN(env->sched_info));
+       return env->sched_info[idx].num_not_sched_user;
 }
 
 /**
- * Append an instruction to a schedule.
- * @param env The block scheduling environment.
- * @param irn The node to add to the schedule.
- * @return    The given node.
+ * Sets the number of not yet schedules users.
  */
-static ir_node *add_to_sched(block_sched_env_t *env, ir_node *irn)
-{
-    /* If the node consumes/produces data, it is appended to the schedule
-     * list, otherwise, it is not put into the list */
-    if(must_appear_in_schedule(env->selector, env->selector_block_env, irn)) {
-        sched_info_t *info = get_irn_sched_info(irn);
-        INIT_LIST_HEAD(&info->list);
-        info->scheduled = 1;
-        sched_add_before(env->block, irn);
-
-        DBG((env->dbg, LEVEL_2, "\tadding %+F\n", irn));
-    }
+static INLINE void set_irn_not_sched_user(block_sched_env_t *env, ir_node *n, int num) {
+       int idx = get_irn_idx(n);
 
-    /* Insert the node in the set of all already scheduled nodes. */
-    mark_already_scheduled(env, irn);
+       assert(idx < ARR_LEN(env->sched_info));
+       env->sched_info[idx].num_not_sched_user = num;
+}
 
-    /* Remove the node from the ready set */
-    if(nodeset_find(env->cands, irn))
-        nodeset_remove(env->cands, irn);
+/**
+ * Add @p num to the number of not yet schedules users and returns the result.
+ */
+static INLINE int add_irn_not_sched_user(block_sched_env_t *env, ir_node *n, int num) {
+       int idx = get_irn_idx(n);
 
-    return irn;
+       assert(idx < ARR_LEN(env->sched_info));
+       env->sched_info[idx].num_not_sched_user += num;
+       return env->sched_info[idx].num_not_sched_user;
 }
 
 /**
- * Add the proj nodes of a tuple-mode irn to the schedule immediately
- * after the tuple-moded irn. By pinning the projs after the irn, no
- * other nodes can create a new lifetime between the tuple-moded irn and
- * one of its projs. This should render a realistic image of a
- * tuple-moded irn, which in fact models a node which defines multiple
- * values.
- *
- * @param irn The tuple-moded irn.
+ * Returns the number of users of a node having mode datab.
  */
-static void add_tuple_projs(block_sched_env_t *env, ir_node *irn)
-{
+static int get_num_successors(ir_node *irn) {
+       int             sum = 0;
        const ir_edge_t *edge;
 
-       assert(get_irn_mode(irn) == mode_T && "Mode of node must be tuple");
-
-       if(is_Bad(irn))
-               return;
-
-       foreach_out_edge(irn, edge) {
-               ir_node *out = edge->src;
-
-               assert(is_Proj(out) && "successor of a modeT node must be a proj");
+       if (get_irn_mode(irn) == mode_T) {
+               /* for mode_T nodes: count the users of all Projs */
+               foreach_out_edge(irn, edge) {
+                       ir_node *proj = get_edge_src_irn(edge);
+                       ir_mode *mode = get_irn_mode(proj);
 
-               if (get_irn_mode(out) == mode_T)
-                       add_tuple_projs(env, out);
-               else {
-                       add_to_sched(env, out);
-                       make_users_ready(env, out);
+                       if (mode == mode_T) {
+                               sum += get_num_successors(proj);
+                       } else if (mode_is_datab(mode)) {
+                               sum += get_irn_n_edges(proj);
+                       }
                }
        }
-}
-
-/**
- * Execute the heuristic function,
- */
-static ir_node *select_node_heuristic(block_sched_env_t *be, nodeset *ns)
-{
-       ir_node *irn;
-
-       for (irn = nodeset_first(ns); irn; irn = nodeset_next(ns)) {
-               if (be_is_Keep(irn)) {
-                       nodeset_break(ns);
-                       return irn;
+       else {
+               /* do not count keep-alive edges */
+               foreach_out_edge(irn, edge) {
+                       if (get_irn_opcode(get_edge_src_irn(edge)) != iro_End)
+                               sum++;
                }
        }
 
-       return be->selector->select(be->selector_block_env, ns);
+       return sum;
 }
 
 /**
- * Returns non-zero if root is a root in the block block.
+ * Adds irn to @p live, updates all inputs that this user is scheduled
+ * and counts all of its non scheduled users.
  */
-static int is_root(ir_node *root, ir_node *block) {
-       const ir_edge_t *edge;
+static void update_sched_liveness(block_sched_env_t *env, ir_node *irn) {
+       int i;
 
-       foreach_out_edge(root, edge) {
-               ir_node *succ = get_edge_src_irn(edge);
+       /* ignore Projs */
+       if (is_Proj(irn))
+               return;
 
-               if (is_Block(succ))
-                       continue;
-               /* Phi nodes are always in "another block */
-               if (is_Phi(succ))
-                       continue;
-               if (get_nodes_block(succ) == block)
-                       return 0;
+       for (i = get_irn_ins_or_deps(irn) - 1; i >= 0; --i) {
+               ir_node *in = get_irn_in_or_dep(irn, i);
+
+               /* if in is a proj: update predecessor */
+               in = skip_Proj(in);
+
+               /* if in is still in the live set: reduce number of users by one */
+               if (ir_nodeset_contains(&env->live, in)) {
+                       if (add_irn_not_sched_user(env, in, -1) <= 0)
+                               ir_nodeset_remove(&env->live, in);
+               }
        }
-       return 1;
-}
 
-/* we need a special mark */
-static char _mark;
-#define MARK   &_mark
+       /*
+               get_num_successors returns the number of all users. This includes
+               users in different blocks as well. As the each block is scheduled separately
+               the liveness info of those users will not be updated and so these
+               users will keep up the register pressure as it is desired.
+       */
+       i = get_num_successors(irn);
+       if (i > 0) {
+               set_irn_not_sched_user(env, irn, i);
+               ir_nodeset_insert(&env->live, irn);
+       }
+}
 
 /**
- * descent into a dag and create a pre-order list.
+ * Append an instruction to a schedule.
+ * @param env The block scheduling environment.
+ * @param irn The node to add to the schedule.
+ * @return    The given node.
  */
-static void descent(ir_node *root, ir_node *block, ir_node **list) {
-       int i;
+static void add_to_sched(block_sched_env_t *env, ir_node *irn)
+{
+    /* If the node consumes/produces data, it is appended to the schedule
+     * list, otherwise, it is not put into the list */
+    if (must_appear_in_schedule(env->selector, env->selector_block_env, irn)) {
+               update_sched_liveness(env, irn);
+               sched_add_before(env->block, irn);
 
-       if (! is_Phi(root)) {
-               /* Phi nodes always leave the block */
-               for (i = get_irn_arity(root) - 1; i >= 0; --i) {
-                       ir_node *pred = get_irn_n(root, i);
+               DBG((dbg, LEVEL_2, "\tadding %+F\n", irn));
 
-                       /* Blocks may happen as predecessors of End nodes */
-                       if (is_Block(pred))
-                               continue;
+               /* Remove the node from the ready set */
+               ir_nodeset_remove(&env->cands, irn);
+    }
 
-                       /* already seen nodes are not marked */
-                       if (get_irn_link(pred) != MARK)
-                               continue;
+       /* notify the selector about the finally selected node. */
+       if (env->selector->node_selected)
+               env->selector->node_selected(env->selector_block_env, irn);
 
-                       /* don't leave our block */
-                       if (get_nodes_block(pred) != block)
-                               continue;
+    /* Insert the node in the set of all available scheduled nodes. */
+    set_already_scheduled(env, irn);
 
-                       descent(pred, block, list);
-               }
-       }
-       set_irn_link(root, *list);
-       *list = root;
+       make_users_ready(env, irn);
 }
 
 /**
@@ -758,114 +428,64 @@ static void list_sched_block(ir_node *block, void *env_ptr)
        sched_env_t *env                      = env_ptr;
        const list_sched_selector_t *selector = env->selector;
        ir_node *start_node                   = get_irg_start(get_irn_irg(block));
-       sched_info_t *info                    = get_irn_sched_info(block);
 
        block_sched_env_t be;
        const ir_edge_t *edge;
        ir_node *irn;
        int j, m;
 
-       ir_node *root = NULL, *preord = NULL;
-       ir_node *curr;
-
        /* Initialize the block's list head that will hold the schedule. */
-       INIT_LIST_HEAD(&info->list);
+       sched_init_block(block);
 
        /* Initialize the block scheduling environment */
-       be.sched_info        = env->sched_info;
-       be.block             = block;
-       be.curr_time         = 0;
-       be.cands             = new_nodeset(get_irn_n_edges(block));
-       be.selector          = selector;
-       be.sched_env         = env;
-       FIRM_DBG_REGISTER(be.dbg, "firm.be.sched");
+       be.sched_info = env->sched_info;
+       be.block      = block;
+       ir_nodeset_init_size(&be.cands, get_irn_n_edges(block));
+       ir_nodeset_init_size(&be.live, get_irn_n_edges(block));
+       be.selector   = selector;
+       be.sched_env  = env;
 
-//     firm_dbg_set_mask(be.dbg, SET_LEVEL_3);
+       DBG((dbg, LEVEL_1, "scheduling %+F\n", block));
 
        if (selector->init_block)
                be.selector_block_env = selector->init_block(env->selector_env, block);
 
-       DBG((be.dbg, LEVEL_1, "scheduling %+F\n", block));
-
-       /* First step: Find the root set. */
+       /* Then one can add all nodes are ready to the set. */
        foreach_out_edge(block, edge) {
-               ir_node *succ = get_edge_src_irn(edge);
-
-               if (is_root(succ, block)) {
-                       mark_root_node(&be, succ);
-                       set_irn_link(succ, root);
-                       root = succ;
-               }
-               else
-                       set_irn_link(succ, MARK);
-       }
+               ir_node   *irn = get_edge_src_irn(edge);
+               ir_opcode code = get_irn_opcode(irn);
+               int users;
 
-       /* Second step: calculate the pre-order list. */
-       preord = NULL;
-       for (curr = root; curr; curr = irn) {
-               irn = get_irn_link(curr);
-               descent(curr, block, &preord);
-       }
-       root = preord;
-
-       /* Third step: calculate the Delay. Note that our
-        * list is now in pre-order, starting at root
-        */
-       for (curr = root; curr; curr = get_irn_link(curr)) {
-               sched_timestep_t d;
-
-               if (arch_irn_classify(env->arch_env, curr) == arch_irn_class_branch) {
-                       /* assure, that branches can be executed last */
-                       d = 0;
-               }
-               else {
-                       if (is_root_node(&be, curr))
-                               d = exectime(env, curr);
-                       else {
-                               d = 0;
-                               foreach_out_edge(curr, edge) {
-                                       ir_node *n = get_edge_src_irn(edge);
-
-                                       if (get_nodes_block(n) == block) {
-                                               sched_timestep_t ld;
-
-                                               ld = latency(env, curr, 1, n, 0) + get_irn_delay(&be, n);
-                                               d = ld > d ? ld : d;
-                                       }
-                               }
-                       }
+               if (code == iro_End) {
+                       /* Skip the end node because of keep-alive edges. */
+                       continue;
+               } else if (code == iro_Block) {
+                       /* A Block-Block edge. This should be the MacroBlock
+                        * edge, ignore it. */
+                       assert(get_Block_MacroBlock(irn) == block && "Block-Block edge found");
+                       continue;
                }
-               set_irn_delay(&be, curr, d);
-               DB((be.dbg, LEVEL_2, "\t%+F delay %u\n", curr, d));
-
-               /* set the etime of all nodes to 0 */
-               set_irn_etime(&be, curr, 0);
-       }
-
 
-       /* Then one can add all nodes are ready to the set. */
-       foreach_out_edge(block, edge) {
-               ir_node *irn = get_edge_src_irn(edge);
-
-               /* Skip the end node because of keepalive edges. */
-               if (get_irn_opcode(irn) == iro_End)
+               users = get_irn_n_edges(irn);
+               if (users == 0)
                        continue;
+               else if (users == 1) { /* ignore nodes that are only hold by the anchor */
+                       const ir_edge_t *edge = get_irn_out_edge_first_kind(irn, EDGE_KIND_NORMAL);
+                       ir_node *user = get_edge_src_irn(edge);
+                       if (is_Anchor(user))
+                               continue;
+               }
 
                if (is_Phi(irn)) {
-                       /* Phi functions are scheduled immediately, since they only transfer
-                        * data flow from the predecessors to this block. */
-
-                       /* Increase the time step. */
-                       be.curr_time += get_irn_etime(&be, irn);
+                       /*
+                               Phi functions are scheduled immediately, since they     only
+                               transfer data flow from the predecessors to this block.
+                       */
                        add_to_sched(&be, irn);
-                       make_users_ready(&be, irn);
                }
                else if (irn == start_node) {
                        /* The start block will be scheduled as the first node */
-                       be.curr_time += get_irn_etime(&be, irn);
-
                        add_to_sched(&be, irn);
-                       add_tuple_projs(&be, irn);
                }
                else {
                        /* Other nodes must have all operands in other blocks to be made
@@ -873,157 +493,120 @@ static void list_sched_block(ir_node *block, void *env_ptr)
                        int ready = 1;
 
                        /* Check, if the operands of a node are not local to this block */
-                       for (j = 0, m = get_irn_arity(irn); j < m; ++j) {
-                               ir_node *operand = get_irn_n(irn, j);
+                       for (j = 0, m = get_irn_ins_or_deps(irn); j < m; ++j) {
+                               ir_node *operand = get_irn_in_or_dep(irn, j);
 
                                if (get_nodes_block(operand) == block) {
                                        ready = 0;
                                        break;
+                               } else {
+                                       /* live in values increase register pressure */
+                                       ir_nodeset_insert(&be.live, operand);
                                }
                        }
 
                        /* Make the node ready, if all operands live in a foreign block */
                        if (ready) {
-                               DBG((be.dbg, LEVEL_2, "\timmediately ready: %+F\n", irn));
+                               DBG((dbg, LEVEL_2, "\timmediately ready: %+F\n", irn));
                                make_ready(&be, NULL, irn);
                        }
                }
        }
 
-       while (nodeset_count(be.cands) > 0) {
-               nodeset *mcands;                            /**< the set of candidates with maximum delay time */
-               nodeset *ecands;                            /**< the set of nodes in mcands whose etime <= curr_time  */
-               sched_timestep_t max_delay = 0;
-
+       /* Iterate over all remaining nodes */
+       while (ir_nodeset_size(&be.cands) > 0) {
+               ir_nodeset_iterator_t iter;
                /* collect statistics about amount of ready nodes */
-               be_do_stat_sched_ready(block, be.cands);
-
-               /* calculate the max delay of all candidates */
-               foreach_nodeset(be.cands, irn) {
-                       sched_timestep_t d = get_irn_delay(&be, irn);
+               be_do_stat_sched_ready(block, &be.cands);
 
-                       max_delay = d > max_delay ? d : max_delay;
-               }
-               mcands = new_nodeset(8);
-               ecands = new_nodeset(8);
-
-               /* calculate mcands and ecands */
-               foreach_nodeset(be.cands, irn) {
-      if (be_is_Keep(irn)) {
-        nodeset_break(be.cands);
-        break;
-      }
-                       if (get_irn_delay(&be, irn) == max_delay) {
-                               nodeset_insert(mcands, irn);
-                               if (get_irn_etime(&be, irn) <= be.curr_time)
-                                       nodeset_insert(ecands, irn);
+               /* Keeps must be scheduled immediately */
+               foreach_ir_nodeset(&be.cands, irn, iter) {
+                       if (be_is_Keep(irn) || be_is_CopyKeep(irn) || is_Sync(irn)) {
+                               break;
                        }
                }
 
-    if (irn) {
-      /* Keeps must be immediately scheduled */
-    }
-    else {
-                 DB((be.dbg, LEVEL_2, "\tbe.curr_time = %u\n", be.curr_time));
-
-                 /* select a node to be scheduled and check if it was ready */
-                 if (nodeset_count(mcands) == 1) {
-                         DB((be.dbg, LEVEL_3, "\tmcand = 1, max_delay = %u\n", max_delay));
-                         irn = nodeset_first(mcands);
-                 }
-                 else {
-                         int cnt = nodeset_count(ecands);
-                         if (cnt == 1) {
-                                       arch_irn_class_t irn_class;
-
-                                 irn = nodeset_first(ecands);
-                                       irn_class = arch_irn_classify(env->arch_env, irn);
-
-                                       if (irn_class == arch_irn_class_branch) {
-                                               /* BEWARE: don't select a JUMP if others are still possible */
-                                               goto force_mcands;
-                                       }
-                                 DB((be.dbg, LEVEL_3, "\tecand = 1, max_delay = %u\n", max_delay));
-                         }
-                         else if (cnt > 1) {
-                                 DB((be.dbg, LEVEL_3, "\tecand = %d, max_delay = %u\n", cnt, max_delay));
-                                 irn = select_node_heuristic(&be, ecands);
-                         }
-                         else {
-force_mcands:
-                                 DB((be.dbg, LEVEL_3, "\tmcand = %d\n", nodeset_count(mcands)));
-                                 irn = select_node_heuristic(&be, mcands);
-                         }
-                 }
-    }
-               del_nodeset(mcands);
-               del_nodeset(ecands);
-
-               DB((be.dbg, LEVEL_2, "\tpicked node %+F\n", irn));
+               if (! irn) {
+                       /* Keeps must be immediately scheduled */
+                       irn = be.selector->select(be.selector_block_env, &be.cands, &be.live);
+               }
 
-               /* Increase the time step. */
-               be.curr_time += exectime(env, irn);
+               DB((dbg, LEVEL_2, "\tpicked node %+F\n", irn));
 
                /* Add the node to the schedule. */
                add_to_sched(&be, irn);
 
-               if (get_irn_mode(irn) == mode_T)
-                       add_tuple_projs(&be, irn);
-               else
-                       make_users_ready(&be, irn);
-
                /* remove the scheduled node from the ready list. */
-               if (nodeset_find(be.cands, irn))
-                       nodeset_remove(be.cands, irn);
+               ir_nodeset_remove(&be.cands, irn);
        }
 
        if (selector->finish_block)
                selector->finish_block(be.selector_block_env);
 
-       del_nodeset(be.cands);
+       ir_nodeset_destroy(&be.cands);
+       ir_nodeset_destroy(&be.live);
 }
 
-static const list_sched_selector_t reg_pressure_selector_struct = {
-       reg_pressure_graph_init,
-       reg_pressure_block_init,
-       reg_pressure_select,
-       NULL,                    /* to_appear_in_schedule */
-       NULL,                    /* exectime */
-       NULL,                    /* latency */
-       reg_pressure_block_free,
-       free
-};
-
-const list_sched_selector_t *reg_pressure_selector = &reg_pressure_selector_struct;
-
 /* List schedule a graph. */
-void list_sched(const be_irg_t *birg, int enable_mris)
+void list_sched(be_irg_t *birg, be_options_t *be_opts)
 {
        const arch_env_t *arch_env = birg->main_env->arch_env;
-       ir_graph *irg              = birg->irg;
+       ir_graph         *irg      = birg->irg;
 
        int num_nodes;
        sched_env_t env;
-       mris_env_t *mris;
+       mris_env_t *mris = NULL;
+       list_sched_selector_t sel;
+
+       (void)be_opts;
+
+       /* Select a scheduler based on backend options */
+       switch (list_sched_options.select) {
+               case BE_SCHED_SELECT_TRIVIAL:  sel = trivial_selector;      break;
+               case BE_SCHED_SELECT_RANDOM:   sel = random_selector;       break;
+               case BE_SCHED_SELECT_REGPRESS: sel = reg_pressure_selector; break;
+               case BE_SCHED_SELECT_MUCHNIK:  sel = muchnik_selector;      break;
+               case BE_SCHED_SELECT_HEUR:     sel = heuristic_selector;    break;
+               case BE_SCHED_SELECT_NORMAL:   sel = normal_selector;       break;
+               default:
+               case BE_SCHED_SELECT_HMUCHNIK: sel = trivial_selector;      break;
+       }
 
-       /* Assure, that the out edges are computed */
-       edges_assure(irg);
+#if 1
+       /* Matze: This is very slow, we should avoid it to improve backend speed,
+        * we just have to make sure that we have no dangling out-edges at this
+        * point...
+        */
 
-       if(enable_mris)
-               mris = be_sched_mris_preprocess(birg);
+       /* Assure, that we have no dangling out-edges to deleted stuff */
+       edges_deactivate(birg->irg);
+       edges_activate(birg->irg);
+#endif
+
+       switch (list_sched_options.prep) {
+               case BE_SCHED_PREP_MRIS:
+                       mris = be_sched_mris_preprocess(birg);
+                       break;
+               case BE_SCHED_PREP_RSS:
+                       rss_schedule_preparation(birg);
+                       break;
+               default:
+                       break;
+       }
 
        num_nodes = get_irg_last_idx(irg);
 
+       /* initialize environment for list scheduler */
        memset(&env, 0, sizeof(env));
-       env.selector   = arch_env->isa->impl->get_list_sched_selector(arch_env->isa);
+       env.selector   = arch_env_get_list_sched_selector(arch_env, &sel);
        env.arch_env   = arch_env;
        env.irg        = irg;
        env.sched_info = NEW_ARR_F(sched_irn_t, num_nodes);
 
-       memset(env.sched_info, 0, num_nodes * sizeof(*env.sched_info));
+       memset(env.sched_info, 0, num_nodes * sizeof(env.sched_info[0]));
 
        if (env.selector->init_graph)
-               env.selector_env = env.selector->init_graph(env.selector, arch_env, irg);
+               env.selector_env = env.selector->init_graph(env.selector, birg);
 
        /* Schedule each single block. */
        irg_block_walk_graph(irg, list_sched_block, NULL, &env);
@@ -1031,8 +614,74 @@ void list_sched(const be_irg_t *birg, int enable_mris)
        if (env.selector->finish_graph)
                env.selector->finish_graph(env.selector_env);
 
-       if(enable_mris)
+       if (list_sched_options.prep == BE_SCHED_PREP_MRIS)
                be_sched_mris_free(mris);
 
        DEL_ARR_F(env.sched_info);
 }
+
+/* List schedule a block. */
+void list_sched_single_block(const be_irg_t *birg, ir_node *block,
+                             be_options_t *be_opts)
+{
+       const arch_env_t *arch_env = birg->main_env->arch_env;
+       ir_graph         *irg      = birg->irg;
+
+       int num_nodes;
+       sched_env_t env;
+       list_sched_selector_t sel;
+
+       (void)be_opts;
+
+       /* Select a scheduler based on backend options */
+       switch (list_sched_options.select) {
+               case BE_SCHED_SELECT_TRIVIAL:  sel = trivial_selector;      break;
+               case BE_SCHED_SELECT_RANDOM:   sel = random_selector;       break;
+               case BE_SCHED_SELECT_REGPRESS: sel = reg_pressure_selector; break;
+               case BE_SCHED_SELECT_MUCHNIK:  sel = muchnik_selector;      break;
+               case BE_SCHED_SELECT_HEUR:     sel = heuristic_selector;    break;
+               case BE_SCHED_SELECT_NORMAL:   sel = normal_selector;       break;
+               default:
+               case BE_SCHED_SELECT_HMUCHNIK: sel = trivial_selector;      break;
+       }
+
+       /* Assure, that the out edges are computed */
+       edges_deactivate(birg->irg);
+       edges_activate(birg->irg);
+
+       num_nodes = get_irg_last_idx(irg);
+
+       /* initialize environment for list scheduler */
+       memset(&env, 0, sizeof(env));
+       env.selector   = arch_env_get_list_sched_selector(arch_env, &sel);
+       env.arch_env   = arch_env;
+       env.irg        = irg;
+       env.sched_info = NEW_ARR_F(sched_irn_t, num_nodes);
+
+       memset(env.sched_info, 0, num_nodes * sizeof(env.sched_info[0]));
+
+       if (env.selector->init_graph)
+               env.selector_env = env.selector->init_graph(env.selector, birg);
+
+       /* Schedule block. */
+       list_sched_block(block, &env);
+
+       if (env.selector->finish_graph)
+               env.selector->finish_graph(env.selector_env);
+
+       DEL_ARR_F(env.sched_info);
+}
+
+/**
+ * Register list scheduler options.
+ */
+void be_init_listsched(void) {
+       lc_opt_entry_t *be_grp = lc_opt_get_grp(firm_opt_get_root(), "be");
+       lc_opt_entry_t *sched_grp = lc_opt_get_grp(be_grp, "listsched");
+
+       lc_opt_add_table(sched_grp, list_sched_option_table);
+
+       FIRM_DBG_REGISTER(dbg, "firm.be.sched");
+}
+
+BE_REGISTER_MODULE_CONSTRUCTOR(be_init_listsched);