Fixed indentation.
[libfirm] / ir / be / beprefalloc.c
index 1ab4e3c..9266204 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 1995-2008 University of Karlsruhe.  All right reserved.
+ * Copyright (C) 1995-2011 University of Karlsruhe.  All right reserved.
  *
  * This file is part of libFirm.
  *
@@ -22,7 +22,6 @@
  * @brief       Preference Guided Register Assignment
  * @author      Matthias Braun
  * @date        14.2.2009
- * @version     $Id$
  *
  * The idea is to allocate registers in 2 passes:
  * 1. A first pass to determine "preferred" registers for live-ranges. This
@@ -42,6 +41,7 @@
 #include <float.h>
 #include <stdbool.h>
 #include <math.h>
+#include "lpp.h"
 
 #include "error.h"
 #include "execfreq.h"
@@ -53,6 +53,8 @@
 #include "irnode_t.h"
 #include "irprintf.h"
 #include "irdump.h"
+#include "irtools.h"
+#include "util.h"
 #include "obst.h"
 #include "raw_bitset.h"
 #include "unionfind.h"
@@ -72,6 +74,7 @@
 #include "bespillutil.h"
 #include "beverify.h"
 #include "beutil.h"
+#include "bestack.h"
 
 #define USE_FACTOR                     1.0f
 #define DEF_FACTOR                     1.0f
@@ -86,14 +89,13 @@ DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
 static struct obstack               obst;
 static ir_graph                    *irg;
 static const arch_register_class_t *cls;
-static const arch_register_req_t   *default_cls_req;
 static be_lv_t                     *lv;
 static const ir_exec_freq          *execfreqs;
 static unsigned                     n_regs;
 static unsigned                    *normal_regs;
 static int                         *congruence_classes;
 static ir_node                    **block_order;
-static int                          n_block_order;
+static size_t                       n_block_order;
 static int                          create_preferences        = true;
 static int                          create_congruence_classes = true;
 static int                          propagate_phi_registers   = true;
@@ -141,7 +143,7 @@ typedef struct block_info_t block_info_t;
  */
 static allocation_info_t *get_allocation_info(ir_node *node)
 {
-       allocation_info_t *info = get_irn_link(node);
+       allocation_info_t *info = (allocation_info_t*)get_irn_link(node);
        if (info == NULL) {
                info = OALLOCFZ(&obst, allocation_info_t, prefs, n_regs);
                info->current_value  = node;
@@ -162,7 +164,7 @@ static allocation_info_t *try_get_allocation_info(const ir_node *node)
  */
 static block_info_t *get_block_info(ir_node *block)
 {
-       block_info_t *info = get_irn_link(block);
+       block_info_t *info = (block_info_t*)get_irn_link(block);
 
        assert(is_Block(block));
        if (info == NULL) {
@@ -173,24 +175,6 @@ static block_info_t *get_block_info(ir_node *block)
        return info;
 }
 
-/**
- * Get default register requirement for the current register class
- */
-static const arch_register_req_t *get_default_req_current_cls(void)
-{
-       if (default_cls_req == NULL) {
-               struct obstack      *obst = get_irg_obstack(irg);
-               arch_register_req_t *req  = OALLOCZ(obst, arch_register_req_t);
-
-               req->type  = arch_register_req_type_normal;
-               req->cls   = cls;
-               req->width = 1;
-
-               default_cls_req = req;
-       }
-       return default_cls_req;
-}
-
 /**
  * Link the allocation info of a node to a copy.
  * Afterwards, both nodes uses the same allocation info.
@@ -236,7 +220,7 @@ static void give_penalties_for_limits(const ir_nodeset_t *live_nodes,
 {
        ir_nodeset_iterator_t iter;
        unsigned              r;
-       unsigned              n_allowed;
+       size_t                n_allowed;
        allocation_info_t     *info = get_allocation_info(node);
        ir_node               *neighbor;
 
@@ -288,7 +272,7 @@ static void give_penalties_for_limits(const ir_nodeset_t *live_nodes,
 static void check_defs(const ir_nodeset_t *live_nodes, float weight,
                        ir_node *node)
 {
-       const arch_register_req_t *req = arch_get_register_req_out(node);
+       const arch_register_req_t *req = arch_get_irn_register_req(node);
        if (req->type & arch_register_req_type_limited) {
                const unsigned *limited = req->limited;
                float           penalty = weight * DEF_FACTOR;
@@ -369,7 +353,7 @@ static void analyze_block(ir_node *block, void *data)
                info = get_allocation_info(node);
                for (i = 0; i < arity; ++i) {
                        ir_node                   *op  = get_irn_n(node, i);
-                       const arch_register_req_t *req = arch_get_register_req_out(op);
+                       const arch_register_req_t *req = arch_get_irn_register_req(op);
                        if (req->cls != cls)
                                continue;
 
@@ -391,7 +375,7 @@ static void analyze_block(ir_node *block, void *data)
                                if (!arch_irn_consider_in_reg_alloc(cls, op))
                                        continue;
 
-                               req = arch_get_register_req(node, i);
+                               req = arch_get_irn_register_req_in(node, i);
                                if (!(req->type & arch_register_req_type_limited))
                                        continue;
 
@@ -407,7 +391,7 @@ static void analyze_block(ir_node *block, void *data)
 
 static void congruence_def(ir_nodeset_t *live_nodes, const ir_node *node)
 {
-       const arch_register_req_t *req = arch_get_register_req_out(node);
+       const arch_register_req_t *req = arch_get_irn_register_req(node);
 
        /* should be same constraint? */
        if (req->type & arch_register_req_type_should_be_same) {
@@ -653,12 +637,12 @@ static bool try_optimistic_split(ir_node *to_split, ir_node *before,
        ir_node               *original_insn;
        ir_node               *block;
        ir_node               *copy;
-       unsigned               r;
+       unsigned               r = 0;
        unsigned               from_r;
        unsigned               i;
        allocation_info_t     *info = get_allocation_info(to_split);
        reg_pref_t            *prefs;
-       float                  delta;
+       float                  delta = 0;
        float                  split_threshold;
 
        (void) pref;
@@ -667,7 +651,7 @@ static bool try_optimistic_split(ir_node *to_split, ir_node *before,
         * (so we don't split away the values produced because of
         *  must_be_different constraints) */
        original_insn = skip_Proj(info->original_value);
-       if (arch_irn_get_flags(original_insn) & arch_irn_flags_dont_spill)
+       if (arch_get_irn_flags(original_insn) & arch_irn_flags_dont_spill)
                return false;
 
        from_reg        = arch_get_irn_register(to_split);
@@ -738,7 +722,7 @@ static bool try_optimistic_split(ir_node *to_split, ir_node *before,
                return false;
 
        reg  = arch_register_for_index(cls, r);
-       copy = be_new_Copy(cls, block, to_split);
+       copy = be_new_Copy(block, to_split);
        mark_as_copy_of(copy, to_split);
        /* hacky, but correct here */
        if (assignments[arch_register_get_index(from_reg)] == to_split)
@@ -758,25 +742,25 @@ static bool try_optimistic_split(ir_node *to_split, ir_node *before,
 static void assign_reg(const ir_node *block, ir_node *node,
                        unsigned *forbidden_regs)
 {
-       const arch_register_t     *reg;
+       const arch_register_t     *final_reg;
        allocation_info_t         *info;
        const arch_register_req_t *req;
        reg_pref_t                *reg_prefs;
        ir_node                   *in_node;
-       unsigned                   i;
-       const unsigned            *allowed_regs;
        unsigned                   r;
+       const unsigned            *allowed_regs;
+       unsigned                   final_reg_index = 0;
 
        assert(!is_Phi(node));
        /* preassigned register? */
-       reg = arch_get_irn_register(node);
-       if (reg != NULL) {
-               DB((dbg, LEVEL_2, "Preassignment %+F -> %s\n", node, reg->name));
-               use_reg(node, reg);
+       final_reg = arch_get_irn_register(node);
+       if (final_reg != NULL) {
+               DB((dbg, LEVEL_2, "Preassignment %+F -> %s\n", node, final_reg->name));
+               use_reg(node, final_reg);
                return;
        }
 
-       req = arch_get_register_req_out(node);
+       req = arch_get_irn_register_req(node);
        /* ignore reqs must be preassigned */
        assert (! (req->type & arch_register_req_type_ignore));
 
@@ -792,37 +776,36 @@ static void assign_reg(const ir_node *block, ir_node *node,
                for (i = 0; i < arity; ++i) {
                        ir_node               *in;
                        const arch_register_t *reg;
-                       unsigned               r;
+                       unsigned               reg_index;
                        if (!rbitset_is_set(&req->other_same, i))
                                continue;
 
                        in  = get_irn_n(in_node, i);
                        reg = arch_get_irn_register(in);
                        assert(reg != NULL);
-                       r = arch_register_get_index(reg);
+                       reg_index = arch_register_get_index(reg);
 
                        /* if the value didn't die here then we should not propagate the
                         * should_be_same info */
-                       if (assignments[r] == in)
+                       if (assignments[reg_index] == in)
                                continue;
 
-                       info->prefs[r] += weight * AFF_SHOULD_BE_SAME;
+                       info->prefs[reg_index] += weight * AFF_SHOULD_BE_SAME;
                }
        }
 
        /* create list of register candidates and sort by their preference */
        DB((dbg, LEVEL_2, "Candidates for %+F:", node));
-       reg_prefs = alloca(n_regs * sizeof(reg_prefs[0]));
+       reg_prefs = ALLOCAN(reg_pref_t, n_regs);
        fill_sort_candidates(reg_prefs, info);
-       for (i = 0; i < n_regs; ++i) {
-               unsigned num = reg_prefs[i].num;
+       for (r = 0; r < n_regs; ++r) {
+               unsigned num = reg_prefs[r].num;
                const arch_register_t *reg;
 
                if (!rbitset_is_set(normal_regs, num))
                        continue;
-
                reg = arch_register_for_index(cls, num);
-               DB((dbg, LEVEL_2, " %s(%f)", reg->name, reg_prefs[i].pref));
+               DB((dbg, LEVEL_2, " %s(%f)", reg->name, reg_prefs[r].pref));
        }
        DB((dbg, LEVEL_2, "\n"));
 
@@ -831,33 +814,38 @@ static void assign_reg(const ir_node *block, ir_node *node,
                allowed_regs = req->limited;
        }
 
-       for (i = 0; i < n_regs; ++i) {
+       for (r = 0; r < n_regs; ++r) {
                float   pref, delta;
                ir_node *before;
                bool    res;
 
-               r = reg_prefs[i].num;
-               if (!rbitset_is_set(allowed_regs, r))
+               final_reg_index = reg_prefs[r].num;
+               if (!rbitset_is_set(allowed_regs, final_reg_index))
                        continue;
-               if (assignments[r] == NULL)
+               /* alignment constraint? */
+               if (req->width > 1 && (req->type & arch_register_req_type_aligned)
+                               && (final_reg_index % req->width) != 0)
+                       continue;
+
+               if (assignments[final_reg_index] == NULL)
                        break;
-               pref   = reg_prefs[i].pref;
-               delta  = i+1 < n_regs ? pref - reg_prefs[i+1].pref : 0;
+               pref   = reg_prefs[r].pref;
+               delta  = r+1 < n_regs ? pref - reg_prefs[r+1].pref : 0;
                before = skip_Proj(node);
-               res    = try_optimistic_split(assignments[r], before,
+               res    = try_optimistic_split(assignments[final_reg_index], before,
                                              pref, delta, forbidden_regs, 0);
                if (res)
                        break;
        }
-       if (i >= n_regs) {
+       if (r >= n_regs) {
                /* the common reason to hit this panic is when 1 of your nodes is not
                 * register pressure faithful */
                panic("No register left for %+F\n", node);
        }
 
-       reg = arch_register_for_index(cls, r);
-       DB((dbg, LEVEL_2, "Assign %+F -> %s\n", node, reg->name));
-       use_reg(node, reg);
+       final_reg = arch_register_for_index(cls, final_reg_index);
+       DB((dbg, LEVEL_2, "Assign %+F -> %s\n", node, final_reg->name));
+       use_reg(node, final_reg);
 }
 
 /**
@@ -877,7 +865,7 @@ static void assign_reg(const ir_node *block, ir_node *node,
  * First we count how many destinations a single value has. At the same time
  * we can be sure that each destination register has at most 1 source register
  * (it can have 0 which means we don't care what value is in it).
- * We ignore all fullfilled permuations (like 7->7)
+ * We ignore all fulfilled permuations (like 7->7)
  * In a first pass we create as much copy instructions as possible as they
  * are generally cheaper than exchanges. We do this by counting into how many
  * destinations a register has to be copied (in the example it's 2 for register
@@ -885,8 +873,8 @@ static void assign_reg(const ir_node *block, ir_node *node,
  * We can then create a copy into every destination register when the usecount
  * of that register is 0 (= noone else needs the value in the register).
  *
- * After this step we should have cycles left. We implement a cyclic permutation
- * of n registers with n-1 transpositions.
+ * After this step we should only have cycles left. We implement a cyclic
+ * permutation of n registers with n-1 transpositions.
  *
  * @param live_nodes   the set of live nodes, updated due to live range split
  * @param before       the node before we add the permutation
@@ -895,7 +883,7 @@ static void assign_reg(const ir_node *block, ir_node *node,
  *                     registers.
  */
 static void permute_values(ir_nodeset_t *live_nodes, ir_node *before,
-                             unsigned *permutation)
+                           unsigned *permutation)
 {
        unsigned  *n_used = ALLOCANZ(unsigned, n_regs);
        ir_node   *block;
@@ -935,7 +923,7 @@ static void permute_values(ir_nodeset_t *live_nodes, ir_node *before,
 
                /* create a copy */
                src  = assignments[old_r];
-               copy = be_new_Copy(cls, block, src);
+               copy = be_new_Copy(block, src);
                sched_add_before(before, copy);
                reg = arch_register_for_index(cls, r);
                DB((dbg, LEVEL_2, "Copy %+F (from %+F, before %+F) -> %s\n",
@@ -1119,6 +1107,151 @@ static void determine_live_through_regs(unsigned *bitset, ir_node *node)
        }
 }
 
+static void solve_lpp(ir_nodeset_t *live_nodes, ir_node *node,
+                      unsigned *forbidden_regs, unsigned *live_through_regs)
+{
+       unsigned *forbidden_edges = rbitset_malloc(n_regs * n_regs);
+       int      *lpp_vars        = XMALLOCNZ(int, n_regs*n_regs);
+       int       arity           = get_irn_arity(node);
+       int       i;
+       unsigned  l;
+       unsigned  r;
+
+       lpp_t *lpp = lpp_new("prefalloc", lpp_minimize);
+       //lpp_set_time_limit(lpp, 20);
+       lpp_set_log(lpp, stdout);
+
+       /** mark some edges as forbidden */
+       for (i = 0; i < arity; ++i) {
+               ir_node                   *op = get_irn_n(node, i);
+               const arch_register_t     *reg;
+               const arch_register_req_t *req;
+               const unsigned            *limited;
+               unsigned                   current_reg;
+
+               if (!arch_irn_consider_in_reg_alloc(cls, op))
+                       continue;
+
+               req = arch_get_irn_register_req_in(node, i);
+               if (!(req->type & arch_register_req_type_limited))
+                       continue;
+
+               limited     = req->limited;
+               reg         = arch_get_irn_register(op);
+               current_reg = arch_register_get_index(reg);
+               for (r = 0; r < n_regs; ++r) {
+                       if (rbitset_is_set(limited, r))
+                               continue;
+
+                       rbitset_set(forbidden_edges, current_reg*n_regs + r);
+               }
+       }
+
+       /* add all combinations, except for not allowed ones */
+       for (l = 0; l < n_regs; ++l) {
+               if (!rbitset_is_set(normal_regs, l)) {
+                       char name[15];
+                       snprintf(name, sizeof(name), "%u_to_%u", l, l);
+                       lpp_vars[l*n_regs+l] = lpp_add_var(lpp, name, lpp_binary, 1);
+                       continue;
+               }
+
+               for (r = 0; r < n_regs; ++r) {
+                       if (!rbitset_is_set(normal_regs, r))
+                               continue;
+                       if (rbitset_is_set(forbidden_edges, l*n_regs + r))
+                               continue;
+                       /* livethrough values may not use constrained output registers */
+                       if (rbitset_is_set(live_through_regs, l)
+                           && rbitset_is_set(forbidden_regs, r))
+                               continue;
+
+                       char name[15];
+                       snprintf(name, sizeof(name), "%u_to_%u", l, r);
+
+                       double costs = l==r ? 9 : 8;
+                       lpp_vars[l*n_regs+r]
+                               = lpp_add_var(lpp, name, lpp_binary, costs);
+                       assert(lpp_vars[l*n_regs+r] > 0);
+               }
+       }
+       /* add constraints */
+       for (l = 0; l < n_regs; ++l) {
+               int constraint;
+               /* only 1 destination per register */
+               constraint = -1;
+               for (r = 0; r < n_regs; ++r) {
+                       int var = lpp_vars[l*n_regs+r];
+                       if (var == 0)
+                               continue;
+                       if (constraint < 0) {
+                               char name[64];
+                               snprintf(name, sizeof(name), "%u_to_dest", l);
+                               constraint = lpp_add_cst(lpp, name, lpp_equal, 1);
+                       }
+                       lpp_set_factor_fast(lpp, constraint, var, 1);
+               }
+               /* each destination used by at most 1 value */
+               constraint = -1;
+               for (r = 0; r < n_regs; ++r) {
+                       int var = lpp_vars[r*n_regs+l];
+                       if (var == 0)
+                               continue;
+                       if (constraint < 0) {
+                               char name[64];
+                               snprintf(name, sizeof(name), "one_to_%u", l);
+                               constraint = lpp_add_cst(lpp, name, lpp_less_equal, 1);
+                       }
+                       lpp_set_factor_fast(lpp, constraint, var, 1);
+               }
+       }
+
+       lpp_dump_plain(lpp, fopen("lppdump.txt", "w"));
+
+       /* solve lpp */
+       {
+               ir_graph     *irg     = get_irn_irg(node);
+               be_options_t *options = be_get_irg_options(irg);
+               unsigned     *assignment;
+               lpp_solve(lpp, options->ilp_server, options->ilp_solver);
+               if (!lpp_is_sol_valid(lpp))
+                       panic("ilp solution not valid!");
+
+               assignment = ALLOCAN(unsigned, n_regs);
+               for (l = 0; l < n_regs; ++l) {
+                       unsigned dest_reg = (unsigned)-1;
+                       for (r = 0; r < n_regs; ++r) {
+                               int var = lpp_vars[l*n_regs+r];
+                               if (var == 0)
+                                       continue;
+                               double val = lpp_get_var_sol(lpp, var);
+                               if (val == 1) {
+                                       assert(dest_reg == (unsigned)-1);
+                                       dest_reg = r;
+                               }
+                       }
+                       assert(dest_reg != (unsigned)-1);
+                       assignment[dest_reg] = l;
+               }
+
+               fprintf(stderr, "Assignment: ");
+               for (l = 0; l < n_regs; ++l) {
+                       fprintf(stderr, "%u ", assignment[l]);
+               }
+               fprintf(stderr, "\n");
+               fflush(stdout);
+               permute_values(live_nodes, node, assignment);
+       }
+       lpp_free(lpp);
+}
+
+static bool is_aligned(unsigned num, unsigned alignment)
+{
+       unsigned mask = alignment-1;
+       assert(is_po2(alignment));
+       return (num&mask) == 0;
+}
+
 /**
  * Enforce constraints at a node by live range splits.
  *
@@ -1138,35 +1271,47 @@ static void enforce_constraints(ir_nodeset_t *live_nodes, ir_node *node,
        /* construct a list of register occupied by live-through values */
        unsigned *live_through_regs = NULL;
 
-       /* see if any use constraints are not met */
+       /* see if any use constraints are not met and whether double-width
+        * values are involved */
+       bool double_width = false;
        bool good = true;
        for (i = 0; i < arity; ++i) {
                ir_node                   *op = get_irn_n(node, i);
                const arch_register_t     *reg;
                const arch_register_req_t *req;
                const unsigned            *limited;
-               unsigned                  r;
+               unsigned                   reg_index;
 
                if (!arch_irn_consider_in_reg_alloc(cls, op))
                        continue;
 
                /* are there any limitations for the i'th operand? */
-               req = arch_get_register_req(node, i);
+               req = arch_get_irn_register_req_in(node, i);
+               if (req->width > 1)
+                       double_width = true;
+               reg       = arch_get_irn_register(op);
+               reg_index = arch_register_get_index(reg);
+               if (req->type & arch_register_req_type_aligned) {
+                       if (!is_aligned(reg_index, req->width)) {
+                               good = false;
+                               continue;
+                       }
+               }
                if (!(req->type & arch_register_req_type_limited))
                        continue;
 
-               limited = req->limited;
-               reg     = arch_get_irn_register(op);
-               r       = arch_register_get_index(reg);
-               if (!rbitset_is_set(limited, r)) {
+               limited   = req->limited;
+               if (!rbitset_is_set(limited, reg_index)) {
                        /* found an assignment outside the limited set */
                        good = false;
-                       break;
+                       continue;
                }
        }
 
        /* is any of the live-throughs using a constrained output register? */
        be_foreach_definition(node, cls, value,
+               if (req_->width > 1)
+                       double_width = true;
                if (! (req_->type & arch_register_req_type_limited))
                        continue;
                if (live_through_regs == NULL) {
@@ -1186,6 +1331,12 @@ static void enforce_constraints(ir_nodeset_t *live_nodes, ir_node *node,
                rbitset_alloca(live_through_regs, n_regs);
        }
 
+       if (double_width) {
+               /* only the ILP variant can solve this yet */
+               solve_lpp(live_nodes, node, forbidden_regs, live_through_regs);
+               return;
+       }
+
        /* at this point we have to construct a bipartite matching problem to see
         * which values should go to which registers
         * Note: We're building the matrix in "reverse" - source registers are
@@ -1223,7 +1374,7 @@ static void enforce_constraints(ir_nodeset_t *live_nodes, ir_node *node,
                if (!arch_irn_consider_in_reg_alloc(cls, op))
                        continue;
 
-               req = arch_get_register_req(node, i);
+               req = arch_get_irn_register_req_in(node, i);
                if (!(req->type & arch_register_req_type_limited))
                        continue;
 
@@ -1257,7 +1408,7 @@ static void enforce_constraints(ir_nodeset_t *live_nodes, ir_node *node,
        permute_values(live_nodes, node, assignment);
 }
 
-/** test wether a node @p n is a copy of the value of node @p of */
+/** test whether a node @p n is a copy of the value of node @p of */
 static bool is_copy_of(ir_node *value, ir_node *test_value)
 {
        allocation_info_t *test_info;
@@ -1279,9 +1430,9 @@ static bool is_copy_of(ir_node *value, ir_node *test_value)
 static int find_value_in_block_info(block_info_t *info, ir_node *value)
 {
        unsigned   r;
-       ir_node  **assignments = info->assignments;
+       ir_node  **end_assignments = info->assignments;
        for (r = 0; r < n_regs; ++r) {
-               ir_node *a_value = assignments[r];
+               ir_node *a_value = end_assignments[r];
 
                if (a_value == NULL)
                        continue;
@@ -1579,7 +1730,7 @@ static void allocate_coalesce_block(ir_node *block, void *data)
                int                        p;
 
                node = be_lv_get_irn(lv, block, i);
-               req  = arch_get_register_req_out(node);
+               req  = arch_get_irn_register_req(node);
                if (req->cls != cls)
                        continue;
 
@@ -1617,19 +1768,15 @@ static void allocate_coalesce_block(ir_node *block, void *data)
                }
 
                if (need_phi) {
-                       ir_mode                   *mode = get_irn_mode(node);
-                       const arch_register_req_t *req  = get_default_req_current_cls();
-                       ir_node                   *phi;
-
-                       phi = new_r_Phi(block, n_preds, phi_ins, mode);
-                       be_set_phi_reg_req(phi, req);
+                       ir_mode *mode = get_irn_mode(node);
+                       ir_node *phi  = be_new_Phi(block, n_preds, phi_ins, mode, cls);
 
                        DB((dbg, LEVEL_3, "Create Phi %+F (for %+F) -", phi, node));
 #ifdef DEBUG_libfirm
                        {
-                               int i;
-                               for (i = 0; i < n_preds; ++i) {
-                                       DB((dbg, LEVEL_3, " %+F", phi_ins[i]));
+                               int pi;
+                               for (pi = 0; pi < n_preds; ++pi) {
+                                       DB((dbg, LEVEL_3, " %+F", phi_ins[pi]));
                                }
                                DB((dbg, LEVEL_3, "\n"));
                        }
@@ -1676,7 +1823,6 @@ static void allocate_coalesce_block(ir_node *block, void *data)
 
        /* assign instructions in the block */
        sched_foreach(block, node) {
-               int i;
                int arity;
                ir_node *value;
 
@@ -1749,43 +1895,43 @@ struct block_costs_t {
 
 static int cmp_block_costs(const void *d1, const void *d2)
 {
-       const ir_node       * const *block1 = d1;
-       const ir_node       * const *block2 = d2;
-       const block_costs_t *info1  = get_irn_link(*block1);
-       const block_costs_t *info2  = get_irn_link(*block2);
+       const ir_node       * const *block1 = (const ir_node**)d1;
+       const ir_node       * const *block2 = (const ir_node**)d2;
+       const block_costs_t *info1  = (const block_costs_t*)get_irn_link(*block1);
+       const block_costs_t *info2  = (const block_costs_t*)get_irn_link(*block2);
        return QSORT_CMP(info2->costs, info1->costs);
 }
 
 static void determine_block_order(void)
 {
-       int i;
+       size_t    p;
        ir_node **blocklist = be_get_cfgpostorder(irg);
-       int       n_blocks  = ARR_LEN(blocklist);
+       size_t    n_blocks  = ARR_LEN(blocklist);
        int       dfs_num   = 0;
        pdeq     *worklist  = new_pdeq();
        ir_node **order     = XMALLOCN(ir_node*, n_blocks);
-       int       order_p   = 0;
+       size_t    order_p   = 0;
 
        /* clear block links... */
-       for (i = 0; i < n_blocks; ++i) {
-               ir_node *block = blocklist[i];
+       for (p = 0; p < n_blocks; ++p) {
+               ir_node *block = blocklist[p];
                set_irn_link(block, NULL);
        }
 
        /* walk blocks in reverse postorder, the costs for each block are the
         * sum of the costs of its predecessors (excluding the costs on backedges
         * which we can't determine) */
-       for (i = n_blocks-1; i >= 0; --i) {
+       for (p = n_blocks; p > 0;) {
                block_costs_t *cost_info;
-               ir_node *block = blocklist[i];
+               ir_node *block = blocklist[--p];
 
                float execfreq   = (float)get_block_execfreq(execfreqs, block);
                float costs      = execfreq;
                int   n_cfgpreds = get_Block_n_cfgpreds(block);
-               int   p;
-               for (p = 0; p < n_cfgpreds; ++p) {
-                       ir_node       *pred_block = get_Block_cfgpred_block(block, p);
-                       block_costs_t *pred_costs = get_irn_link(pred_block);
+               int   p2;
+               for (p2 = 0; p2 < n_cfgpreds; ++p2) {
+                       ir_node       *pred_block = get_Block_cfgpred_block(block, p2);
+                       block_costs_t *pred_costs = (block_costs_t*)get_irn_link(pred_block);
                        /* we don't have any info for backedges */
                        if (pred_costs == NULL)
                                continue;
@@ -1804,15 +1950,15 @@ static void determine_block_order(void)
        ir_reserve_resources(irg, IR_RESOURCE_BLOCK_VISITED);
        inc_irg_block_visited(irg);
 
-       for (i = 0; i < n_blocks; ++i) {
-               ir_node       *block = blocklist[i];
+       for (p = 0; p < n_blocks; ++p) {
+               ir_node       *block = blocklist[p];
                if (Block_block_visited(block))
                        continue;
 
                /* continually add predecessors with highest costs to worklist
                 * (without using backedges) */
                do {
-                       block_costs_t *info       = get_irn_link(block);
+                       block_costs_t *info       = (block_costs_t*)get_irn_link(block);
                        ir_node       *best_pred  = NULL;
                        float          best_costs = -1;
                        int            n_cfgpred  = get_Block_n_cfgpreds(block);
@@ -1822,7 +1968,7 @@ static void determine_block_order(void)
                        mark_Block_block_visited(block);
                        for (i = 0; i < n_cfgpred; ++i) {
                                ir_node       *pred_block = get_Block_cfgpred_block(block, i);
-                               block_costs_t *pred_info  = get_irn_link(pred_block);
+                               block_costs_t *pred_info  = (block_costs_t*)get_irn_link(pred_block);
 
                                /* ignore backedges */
                                if (pred_info->dfs_num > info->dfs_num)
@@ -1838,7 +1984,7 @@ static void determine_block_order(void)
 
                /* now put all nodes in the worklist in our final order */
                while (!pdeq_empty(worklist)) {
-                       ir_node *pblock = pdeq_getr(worklist);
+                       ir_node *pblock = (ir_node*)pdeq_getr(worklist);
                        assert(order_p < n_blocks);
                        order[order_p++] = pblock;
                }
@@ -1862,7 +2008,7 @@ static void determine_block_order(void)
  */
 static void be_pref_alloc_cls(void)
 {
-       int i;
+       size_t i;
 
        lv = be_assure_liveness(irg);
        be_liveness_assure_sets(lv);
@@ -1901,7 +2047,7 @@ static void spill(void)
        be_pre_spill_prepare_constr(irg, cls);
        be_timer_pop(T_RA_CONSTR);
 
-       dump(DUMP_RA, irg, "-spillprepare");
+       dump(DUMP_RA, irg, "spillprepare");
 
        /* spill */
        be_timer_push(T_RA_SPILL);
@@ -1912,7 +2058,7 @@ static void spill(void)
        check_for_memory_operands(irg);
        be_timer_pop(T_RA_SPILL_APPLY);
 
-       dump(DUMP_RA, irg, "-spill");
+       dump(DUMP_RA, irg, "spill");
 }
 
 /**
@@ -1921,7 +2067,7 @@ static void spill(void)
 static void be_pref_alloc(ir_graph *new_irg)
 {
        const arch_env_t *arch_env = be_get_irg_arch_env(new_irg);
-       int   n_cls                = arch_env_get_n_reg_class(arch_env);
+       int   n_cls                = arch_env->n_register_classes;
        int   c;
 
        obstack_init(&obst);
@@ -1933,8 +2079,7 @@ static void be_pref_alloc(ir_graph *new_irg)
        determine_block_order();
 
        for (c = 0; c < n_cls; ++c) {
-               cls             = arch_env_get_reg_class(arch_env, c);
-               default_cls_req = NULL;
+               cls = &arch_env->register_classes[c];
                if (arch_register_class_flags(cls) & arch_register_class_flag_manual_ra)
                        continue;
 
@@ -1942,7 +2087,7 @@ static void be_pref_alloc(ir_graph *new_irg)
 
                n_regs      = arch_register_class_n_regs(cls);
                normal_regs = rbitset_malloc(n_regs);
-               be_abi_set_non_ignore_regs(be_get_irg_abi(irg), cls, normal_regs);
+               be_set_allocatable_regs(irg, cls, normal_regs);
 
                spill();
 
@@ -1987,7 +2132,7 @@ static void be_pref_alloc(ir_graph *new_irg)
        obstack_free(&obst, NULL);
 }
 
-BE_REGISTER_MODULE_CONSTRUCTOR(be_init_pref_alloc);
+BE_REGISTER_MODULE_CONSTRUCTOR(be_init_pref_alloc)
 void be_init_pref_alloc(void)
 {
        static be_ra_t be_ra_pref = {