fix handling of default pns in the backend (they don't need table entries)
[libfirm] / ir / be / becopyopt.c
index 9a7540f..fe51312 100644 (file)
 #include "belive_t.h"
 #include "beinsn_t.h"
 #include "besched_t.h"
-#include "benodesets.h"
 #include "bejavacoal.h"
 #include "bestatevent.h"
 #include "beirg_t.h"
 #include "error.h"
 
+#include <libcore/lc_timing.h>
+#include <libcore/lc_opts.h>
+#include <libcore/lc_opts_enum.h>
+
 #define DUMP_BEFORE 1
 #define DUMP_AFTER  2
 #define DUMP_APPEL  4
@@ -82,11 +85,6 @@ static cost_fct_t cost_func   = co_get_costs_exec_freq;
 static unsigned   algo        = CO_ALGO_HEUR4;
 static int        improve     = 1;
 
-#ifdef WITH_LIBCORE
-#include <libcore/lc_timing.h>
-#include <libcore/lc_opts.h>
-#include <libcore/lc_opts_enum.h>
-
 static const lc_opt_enum_mask_items_t dump_items[] = {
        { "before",  DUMP_BEFORE },
        { "after",   DUMP_AFTER  },
@@ -108,13 +106,9 @@ static const lc_opt_enum_mask_items_t algo_items[] = {
        { "none",   CO_ALGO_NONE  },
        { "heur",   CO_ALGO_HEUR  },
        { "heur2",  CO_ALGO_HEUR2 },
-#ifdef WITH_JVM
        { "heur3",  CO_ALGO_HEUR3 },
-#endif /* WITH_JVM */
        { "heur4",  CO_ALGO_HEUR4 },
-#ifdef WITH_ILP
        { "ilp",    CO_ALGO_ILP   },
-#endif /* WITH_ILP */
        { NULL,     0 }
 };
 
@@ -152,7 +146,6 @@ static const lc_opt_table_entry_t options[] = {
        LC_OPT_ENT_BOOL          ("improve", "run heur3 before if algo can exploit start solutions",    &improve),
        LC_OPT_LAST
 };
-#endif /* WITH_LIBCORE */
 
 /* Insert additional options registration functions here. */
 extern void be_co_ilp_register_options(lc_opt_entry_t *grp);
@@ -161,14 +154,12 @@ extern void be_co3_register_options(lc_opt_entry_t *grp);
 
 void be_init_copycoal(void)
 {
-#ifdef WITH_LIBCORE
        lc_opt_entry_t *be_grp = lc_opt_get_grp(firm_opt_get_root(), "be");
        lc_opt_entry_t *ra_grp = lc_opt_get_grp(be_grp, "ra");
        lc_opt_entry_t *chordal_grp = lc_opt_get_grp(ra_grp, "chordal");
        lc_opt_entry_t *co_grp = lc_opt_get_grp(chordal_grp, "co");
 
        lc_opt_add_table(co_grp, options);
-#endif
 }
 
 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copycoal);
@@ -449,10 +440,8 @@ static void co_collect_units(ir_node *irn, void *env) {
                }
                unit->nodes = xrealloc(unit->nodes, unit->node_count * sizeof(*unit->nodes));
                unit->costs = xrealloc(unit->costs, unit->node_count * sizeof(*unit->costs));
-       } else
-
-       /* Proj of a perm with corresponding arg */
-       if (is_Perm_Proj(co->aenv, irn)) {
+       } else if (is_Perm_Proj(co->aenv, irn)) {
+               /* Proj of a perm with corresponding arg */
                assert(!nodes_interfere(co->cenv, irn, get_Perm_src(irn)));
                unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
                unit->costs = xmalloc(2 * sizeof(*unit->costs));
@@ -466,15 +455,39 @@ static void co_collect_units(ir_node *irn, void *env) {
 
                /* Src == Tgt of a 2-addr-code instruction */
                if (is_2addr_code(req)) {
-                       ir_node *other = get_irn_n(skip_Proj(irn), req->other_same);
-                       if (!arch_irn_is(co->aenv, other, ignore) &&
-                                       !nodes_interfere(co->cenv, irn, other)) {
-                               unit->nodes = xmalloc(2 * sizeof(*unit->nodes));
-                               unit->costs = xmalloc(2 * sizeof(*unit->costs));
-                               unit->node_count = 2;
-                               unit->nodes[0] = irn;
-                               unit->nodes[1] = other;
-                               unit->costs[1] = co->get_costs(co, irn, other, -1);
+                       const unsigned other = req->other_same;
+                       int            count = 0;
+                       int            i;
+
+                       for (i = 0; (1U << i) <= other; ++i) {
+                               if (other & (1U << i)) {
+                                       ir_node *o  = get_irn_n(skip_Proj(irn), i);
+                                       if (!arch_irn_is(co->aenv, o, ignore) &&
+                                                       !nodes_interfere(co->cenv, irn, o)) {
+                                               ++count;
+                                       }
+                               }
+                       }
+
+                       if (count != 0) {
+                               int k = 0;
+                               ++count;
+                               unit->nodes = xmalloc(count * sizeof(*unit->nodes));
+                               unit->costs = xmalloc(count * sizeof(*unit->costs));
+                               unit->node_count = count;
+                               unit->nodes[k++] = irn;
+
+                               for (i = 0; 1U << i <= other; ++i) {
+                                       if (other & (1U << i)) {
+                                               ir_node *o  = get_irn_n(skip_Proj(irn), i);
+                                               if (!arch_irn_is(co->aenv, o, ignore) &&
+                                                               !nodes_interfere(co->cenv, irn, o)) {
+                                                       unit->nodes[k] = o;
+                                                       unit->costs[k] = co->get_costs(co, irn, o, -1);
+                                                       ++k;
+                                               }
+                                       }
+                               }
                        }
                } else {
                        assert(0 && "This is not an optimizable node!");
@@ -727,7 +740,7 @@ static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
        new_node.irn        = n1;
        new_node.degree     = 0;
        new_node.neighbours = NULL;
-       node = set_insert(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
+       node = set_insert(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
 
        for (nbr = node->neighbours; nbr; nbr = nbr->next)
                if (nbr->irn == n2) {
@@ -769,27 +782,29 @@ static void build_graph_walker(ir_node *irn, void *env) {
        if (arch_register_type_is(reg, ignore))
                return;
 
-       /* Phis */
-       if (is_Reg_Phi(irn))
+       if (is_Reg_Phi(irn)) { /* Phis */
                for (pos=0, max=get_irn_arity(irn); pos<max; ++pos) {
                        ir_node *arg = get_irn_n(irn, pos);
                        add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
                }
-
-       /* Perms */
-       else if (is_Perm_Proj(co->aenv, irn)) {
+       }
+       else if (is_Perm_Proj(co->aenv, irn)) { /* Perms */
                ir_node *arg = get_Perm_src(irn);
                add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
        }
-
-       /* 2-address code */
-       else {
-               const arch_register_req_t *req =
-                       arch_get_register_req(co->aenv, irn, -1);
+       else { /* 2-address code */
+               const arch_register_req_t *req = arch_get_register_req(co->aenv, irn, -1);
                if (is_2addr_code(req)) {
-                       ir_node *other = get_irn_n(skip_Proj(irn), req->other_same);
-                       if (! arch_irn_is(co->aenv, other, ignore))
-                               add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
+                       const unsigned other = req->other_same;
+                       int i;
+
+                       for (i = 0; 1U << i <= other; ++i) {
+                               if (other & (1U << i)) {
+                                       ir_node *other = get_irn_n(skip_Proj(irn), i);
+                                       if (! arch_irn_is(co->aenv, other, ignore))
+                                               add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
+                               }
+                       }
                }
        }
 }
@@ -817,7 +832,7 @@ int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) {
        ASSERT_GS_AVAIL(co);
 
        new_node.irn = irn;
-       n = set_find(co->nodes, &new_node, sizeof(new_node), nodeset_hash(new_node.irn));
+       n = set_find(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
        if (n) {
                return (n->degree > 0);
        } else
@@ -831,7 +846,8 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
 
        ir_node *irn;
        void *it, *nit;
-       int i, n, n_regs;
+       int n, n_regs;
+       unsigned i;
 
        n_regs = 0;
        for(i = 0; i < co->cls->n_regs; ++i) {
@@ -1080,7 +1096,7 @@ static void appel_walker(ir_node *bl, void *data)
                        for(j = 0; j < insn->use_start; ++j) {
                                ir_node *op   = insn->ops[j].carrier;
                                bitset_t *adm = insn->ops[j].regs;
-                               int k;
+                               unsigned k;
                                size_t nr;
 
                                if(!insn->ops[j].has_constraints)
@@ -1176,8 +1192,8 @@ static void appel_inter_block_aff(ir_node *bl, void *data)
 
 void co_dump_appel_graph_cliques(const copy_opt_t *co, FILE *f)
 {
-       int i;
-       int n_colors;
+       unsigned i;
+       unsigned n_colors;
        appel_clique_walker_t env;
        bitset_t *adm = bitset_alloca(co->cls->n_regs);
        be_lv_t *lv = co->cenv->birg->lv;
@@ -1196,7 +1212,7 @@ void co_dump_appel_graph_cliques(const copy_opt_t *co, FILE *f)
        env.color_map = alloca(co->cls->n_regs * sizeof(env.color_map[0]));
        for(i = 0, n_colors = 0; i < co->cls->n_regs; ++i) {
                const arch_register_t *reg = &co->cls->regs[i];
-               env.color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_colors++;
+               env.color_map[i] = arch_register_type_is(reg, ignore) ? -1 : (int) n_colors++;
        }
 
        env.dumb = 1;
@@ -1208,7 +1224,7 @@ void co_dump_appel_graph_cliques(const copy_opt_t *co, FILE *f)
 
        /* make the first k nodes interfere */
        for(i = 0; i < n_colors; ++i) {
-               int j;
+               unsigned j;
                for(j = i + 1; j < n_colors; ++j)
                        fprintf(f, "%d %d -1 ", i, j);
                fprintf(f, "\n");
@@ -1401,10 +1417,16 @@ static co_algo_info_t algos[] = {
        { void_algo,               "none",  0 },
        { co_solve_heuristic,      "heur1", 0 },
        { co_solve_heuristic_new,  "heur2", 0 },
+#ifdef WITH_JVM
        { co_solve_heuristic_java, "heur3", 0 },
+#else
+       { NULL,                    "heur3", 0 },
+#endif
        { co_solve_heuristic_mst,  "heur4", 0 },
 #ifdef WITH_ILP
        { co_solve_ilp2,           "ilp",   1 },
+#else
+       { NULL,                    "ilp",   1 },
 #endif
        { NULL,                    "",      0 }
 };