Add arch_get_register_req_out().
[libfirm] / ir / be / becopyopt.c
index c94af82..b053a8d 100644 (file)
@@ -29,9 +29,7 @@
  * - Register-constrained nodes
  * - Two-address code instructions
  */
-#ifdef HAVE_CONFIG_H
 #include "config.h"
-#endif
 
 #include "execfreq.h"
 #include "xmalloc.h"
 #include "belive_t.h"
 #include "beinsn_t.h"
 #include "besched_t.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>
+#include "lc_opts.h"
+#include "lc_opts_enum.h"
 
 #define DUMP_BEFORE 1
 #define DUMP_AFTER  2
@@ -196,9 +192,8 @@ copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs)
 
        FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
 
-       co = xcalloc(1, sizeof(*co));
+       co = XMALLOCZ(copy_opt_t);
        co->cenv      = chordal_env;
-       co->aenv      = chordal_env->birg->main_env->arch_env;
        co->irg       = chordal_env->irg;
        co->cls       = chordal_env->cls;
        co->get_costs = get_costs;
@@ -207,7 +202,7 @@ copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs)
        s2 = get_entity_name(get_irg_entity(co->irg));
        s3 = chordal_env->cls->name;
        len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
-       co->name = xmalloc(len);
+       co->name = XMALLOCN(char, len);
        snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
 
        return co;
@@ -218,19 +213,27 @@ void free_copy_opt(copy_opt_t *co) {
        free(co);
 }
 
-int co_is_optimizable_root(const copy_opt_t *co, ir_node *irn) {
+/**
+ * Checks if a node is optimizable, viz. has something to do with coalescing
+ * @param irn  The irn to check
+ */
+static int co_is_optimizable_root(ir_node *irn)
+{
        const arch_register_req_t *req;
        const arch_register_t *reg;
 
-       if (arch_irn_is(co->aenv, irn, ignore))
+       if (arch_irn_is(irn, ignore))
                return 0;
 
-       reg = arch_get_irn_register(co->aenv, irn);
+       reg = arch_get_irn_register(irn);
        if (arch_register_type_is(reg, ignore))
                return 0;
 
-       req = arch_get_register_req(co->aenv, irn, -1);
-       if (is_Reg_Phi(irn) || is_Perm_Proj(co->aenv, irn) || is_2addr_code(req))
+       if (is_Reg_Phi(irn) || is_Perm_Proj(irn))
+               return 1;
+
+       req = arch_get_register_req_out(irn);
+       if (is_2addr_code(req))
                return 1;
 
        return 0;
@@ -376,17 +379,19 @@ static int ou_max_ind_set_costs(unit_t *ou) {
        return safe_costs+best_weight;
 }
 
-static void co_collect_units(ir_node *irn, void *env) {
-       copy_opt_t *co = env;
+static void co_collect_units(ir_node *irn, void *env)
+{
+       const arch_register_req_t *req = arch_get_register_req_out(irn);
+       copy_opt_t                *co  = env;
        unit_t *unit;
 
-       if (!is_curr_reg_class(co, irn))
+       if (req->cls != co->cls)
                return;
-       if (!co_is_optimizable_root(co, irn))
+       if (!co_is_optimizable_root(irn))
                return;
 
        /* Init a new unit */
-       unit = xcalloc(1, sizeof(*unit));
+       unit = XMALLOCZ(unit_t);
        unit->co = co;
        unit->node_count = 1;
        INIT_LIST_HEAD(&unit->queue);
@@ -397,8 +402,8 @@ static void co_collect_units(ir_node *irn, void *env) {
 
                /* init */
                arity = get_irn_arity(irn);
-               unit->nodes = xmalloc((arity+1) * sizeof(*unit->nodes));
-               unit->costs = xmalloc((arity+1) * sizeof(*unit->costs));
+               unit->nodes = XMALLOCN(ir_node*, arity + 1);
+               unit->costs = XMALLOCN(int,      arity + 1);
                unit->nodes[0] = irn;
 
                /* fill */
@@ -406,7 +411,7 @@ static void co_collect_units(ir_node *irn, void *env) {
                        int o, arg_pos;
                        ir_node *arg = get_irn_n(irn, i);
 
-                       assert(is_curr_reg_class(co, arg) && "Argument not in same register class.");
+                       assert(arch_get_irn_reg_class(arg, -1) == co->cls && "Argument not in same register class.");
                        if (arg == irn)
                                continue;
                        if (nodes_interfere(co->cenv, irn, arg)) {
@@ -417,7 +422,7 @@ static void co_collect_units(ir_node *irn, void *env) {
                        /* Else insert the argument of the phi to the members of this ou */
                        DBG((dbg, LEVEL_1, "\t   Member: %+F\n", arg));
 
-                       if (! arch_irn_is(co->aenv, arg, ignore)) {
+                       if (!arch_irn_is(arg, ignore)) {
                                /* Check if arg has occurred at a prior position in the arg/list */
                                arg_pos = 0;
                                for (o=1; o<unit->node_count; ++o) {
@@ -438,21 +443,18 @@ 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 if (is_Perm_Proj(co->aenv, irn)) {
+               unit->nodes = XREALLOC(unit->nodes, ir_node*, unit->node_count);
+               unit->costs = XREALLOC(unit->costs, int,      unit->node_count);
+       } else if (is_Perm_Proj(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));
+               unit->nodes = XMALLOCN(ir_node*, 2);
+               unit->costs = XMALLOCN(int,      2);
                unit->node_count = 2;
                unit->nodes[0] = irn;
                unit->nodes[1] = get_Perm_src(irn);
                unit->costs[1] = co->get_costs(co, irn, unit->nodes[1], -1);
        } else {
-               const arch_register_req_t *req =
-                       arch_get_register_req(co->aenv, irn, -1);
-
                /* Src == Tgt of a 2-addr-code instruction */
                if (is_2addr_code(req)) {
                        const unsigned other = req->other_same;
@@ -462,7 +464,7 @@ static void co_collect_units(ir_node *irn, void *env) {
                        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) &&
+                                       if (!arch_irn_is(o, ignore) &&
                                                        !nodes_interfere(co->cenv, irn, o)) {
                                                ++count;
                                        }
@@ -472,15 +474,15 @@ static void co_collect_units(ir_node *irn, void *env) {
                        if (count != 0) {
                                int k = 0;
                                ++count;
-                               unit->nodes = xmalloc(count * sizeof(*unit->nodes));
-                               unit->costs = xmalloc(count * sizeof(*unit->costs));
+                               unit->nodes = XMALLOCN(ir_node*, count);
+                               unit->costs = XMALLOCN(int,      count);
                                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) &&
+                                               if (!arch_irn_is(o, ignore) &&
                                                                !nodes_interfere(co->cenv, irn, o)) {
                                                        unit->nodes[k] = o;
                                                        unit->costs[k] = co->get_costs(co, irn, o, -1);
@@ -524,12 +526,11 @@ static int compare_ous(const void *k1, const void *k2) {
        const unit_t *u2 = *((const unit_t **) k2);
        int i, o, u1_has_constr, u2_has_constr;
        arch_register_req_t req;
-       const arch_env_t *aenv = u1->co->aenv;
 
        /* Units with constraints come first */
        u1_has_constr = 0;
        for (i=0; i<u1->node_count; ++i) {
-               arch_get_register_req(aenv, &req, u1->nodes[i], -1);
+               arch_get_register_req_out(&req, u1->nodes[i]);
                if (arch_register_req_is(&req, limited)) {
                        u1_has_constr = 1;
                        break;
@@ -538,7 +539,7 @@ static int compare_ous(const void *k1, const void *k2) {
 
        u2_has_constr = 0;
        for (i=0; i<u2->node_count; ++i) {
-               arch_get_register_req(aenv, &req, u2->nodes[i], -1);
+               arch_get_register_req_out(&req, u2->nodes[i]);
                if (arch_register_req_is(&req, limited)) {
                        u2_has_constr = 1;
                        break;
@@ -654,11 +655,11 @@ int co_get_copy_costs(const copy_opt_t *co) {
        ASSERT_OU_AVAIL(co);
 
        list_for_each_entry(unit_t, curr, &co->units, units) {
-               int root_col = get_irn_col(co, curr->nodes[0]);
+               int root_col = get_irn_col(curr->nodes[0]);
                DBG((dbg, LEVEL_1, "  %3d costs for root %+F color %d\n", curr->inevitable_costs, curr->nodes[0], root_col));
                res += curr->inevitable_costs;
                for (i=1; i<curr->node_count; ++i) {
-                       int arg_col = get_irn_col(co, curr->nodes[i]);
+                       int arg_col = get_irn_col(curr->nodes[i]);
                        if (root_col != arg_col) {
                                DBG((dbg, LEVEL_1, "  %3d for arg %+F color %d\n", curr->costs[i], curr->nodes[i], arg_col));
                                res += curr->costs[i];
@@ -696,7 +697,7 @@ void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat)
                                stat->aff_edges += 1;
                                stat->max_costs += neigh->costs;
 
-                               if(get_irn_col(co, an->irn) != get_irn_col(co, neigh->irn)) {
+                               if (get_irn_col(an->irn) != get_irn_col(neigh->irn)) {
                                        stat->costs += neigh->costs;
                                        stat->unsatisfied_edges += 1;
                                }
@@ -763,7 +764,7 @@ static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
        nbr->costs += costs;
 }
 
-static INLINE void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
+static inline void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) {
        if (! be_ifg_connected(co->cenv->ifg, n1, n2)) {
                add_edge(co, n1, n2, costs);
                add_edge(co, n2, n1, costs);
@@ -771,14 +772,15 @@ static INLINE void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs
 }
 
 static void build_graph_walker(ir_node *irn, void *env) {
-       copy_opt_t *co = env;
+       const arch_register_req_t *req = arch_get_register_req_out(irn);
+       copy_opt_t                *co  = env;
        int pos, max;
        const arch_register_t *reg;
 
-       if (!is_curr_reg_class(co, irn) || arch_irn_is(co->aenv, irn, ignore))
+       if (req->cls != co->cls || arch_irn_is(irn, ignore))
                return;
 
-       reg = arch_get_irn_register(co->aenv, irn);
+       reg = arch_get_irn_register(irn);
        if (arch_register_type_is(reg, ignore))
                return;
 
@@ -787,13 +789,10 @@ static void build_graph_walker(ir_node *irn, void *env) {
                        ir_node *arg = get_irn_n(irn, pos);
                        add_edges(co, irn, arg, co->get_costs(co, irn, arg, pos));
                }
-       }
-       else if (is_Perm_Proj(co->aenv, irn)) { /* Perms */
+       } else if (is_Perm_Proj(irn)) { /* Perms */
                ir_node *arg = get_Perm_src(irn);
                add_edges(co, irn, arg, co->get_costs(co, irn, arg, 0));
-       }
-       else { /* 2-address code */
-               const arch_register_req_t *req = arch_get_register_req(co->aenv, irn, -1);
+       } else { /* 2-address code */
                if (is_2addr_code(req)) {
                        const unsigned other = req->other_same;
                        int i;
@@ -801,7 +800,7 @@ static void build_graph_walker(ir_node *irn, void *env) {
                        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))
+                                       if (!arch_irn_is(other, ignore))
                                                add_edges(co, irn, other, co->get_costs(co, irn, other, 0));
                                }
                        }
@@ -843,14 +842,13 @@ static int co_dump_appel_disjoint_constraints(const copy_opt_t *co, ir_node *a,
 {
        ir_node *nodes[]  = { a, b };
        bitset_t *constr[] = { NULL, NULL };
-       const arch_register_req_t *req;
        int j;
 
        constr[0] = bitset_alloca(co->cls->n_regs);
        constr[1] = bitset_alloca(co->cls->n_regs);
 
        for (j = 0; j < 2; ++j) {
-               req = arch_get_register_req(co->aenv, nodes[j], BE_OUT_POS(0));
+               const arch_register_req_t *req = arch_get_register_req_out(nodes[j]);
                if(arch_register_req_is(req, limited))
                        rbitset_copy_to_bitset(req->limited, constr[j]);
                else
@@ -865,7 +863,7 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
 {
        be_ifg_t *ifg  = co->cenv->ifg;
        int *color_map = alloca(co->cls->n_regs * sizeof(color_map[0]));
-       int *node_map  = xmalloc((get_irg_last_idx(co->irg) + 1) * sizeof(node_map[0]));
+       int *node_map  = XMALLOCN(int, get_irg_last_idx(co->irg) + 1);
 
        ir_node *irn;
        void *it, *nit;
@@ -888,21 +886,19 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
 
        n = n_regs;
        be_ifg_foreach_node(ifg, it, irn) {
-               if(!arch_irn_is(co->aenv, irn, ignore))
+               if (!arch_irn_is(irn, ignore))
                        node_map[get_irn_idx(irn)] = n++;
        }
 
        fprintf(f, "%d %d\n", n, n_regs);
 
        be_ifg_foreach_node(ifg, it, irn) {
-               if(!arch_irn_is(co->aenv, irn, ignore)) {
-                       int idx            = node_map[get_irn_idx(irn)];
-                       affinity_node_t *a = get_affinity_info(co, irn);
+               if (!arch_irn_is(irn, ignore)) {
+                       int idx                        = node_map[get_irn_idx(irn)];
+                       affinity_node_t           *a   = get_affinity_info(co, irn);
+                       const arch_register_req_t *req = arch_get_register_req_out(irn);
+                       ir_node                   *adj;
 
-                       const arch_register_req_t *req;
-                       ir_node *adj;
-
-                       req = arch_get_register_req(co->aenv, irn, BE_OUT_POS(0));
                        if(arch_register_req_is(req, limited)) {
                                for(i = 0; i < co->cls->n_regs; ++i) {
                                        if(!rbitset_is_set(req->limited, i) && color_map[i] >= 0)
@@ -911,7 +907,8 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
                        }
 
                        be_ifg_foreach_neighbour(ifg, nit, irn, adj) {
-                               if(!arch_irn_is(co->aenv, adj, ignore) && !co_dump_appel_disjoint_constraints(co, irn, adj)) {
+                               if (!arch_irn_is(adj, ignore) &&
+                                               !co_dump_appel_disjoint_constraints(co, irn, adj)) {
                                        int adj_idx = node_map[get_irn_idx(adj)];
                                        if(idx < adj_idx)
                                                fprintf(f, "%d %d -1\n", idx, adj_idx);
@@ -922,7 +919,7 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f)
                                neighb_t *n;
 
                                co_gs_foreach_neighb(a, n) {
-                                       if(!arch_irn_is(co->aenv, n->irn, ignore)) {
+                                       if (!arch_irn_is(n->irn, ignore)) {
                                                int n_idx = node_map[get_irn_idx(n->irn)];
                                                if(idx < n_idx)
                                                        fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs);
@@ -996,19 +993,16 @@ static void ifg_dump_graph_attr(FILE *f, void *self)
 
 static int ifg_is_dump_node(void *self, ir_node *irn)
 {
-       co_ifg_dump_t *cod = self;
-       return !arch_irn_is(cod->co->aenv, irn, ignore);
+       (void)self;
+       return !arch_irn_is(irn, ignore);
 }
 
 static void ifg_dump_node_attr(FILE *f, void *self, ir_node *irn)
 {
-       co_ifg_dump_t *env         = self;
-       const arch_register_t *reg = arch_get_irn_register(env->co->aenv, irn);
-       const arch_register_req_t *req;
-       int limited;
-
-       req = arch_get_register_req(env->co->aenv, irn, BE_OUT_POS(0));
-       limited = arch_register_req_is(req, limited);
+       co_ifg_dump_t             *env     = self;
+       const arch_register_t     *reg     = arch_get_irn_register(irn);
+       const arch_register_req_t *req     = arch_get_register_req_out(irn);
+       int                        limited = arch_register_req_is(req, limited);
 
        if(env->flags & CO_IFG_DUMP_LABELS) {
                ir_fprintf(f, "label=\"%+F", irn);
@@ -1036,12 +1030,12 @@ static void ifg_dump_at_end(FILE *file, void *self)
        affinity_node_t *a;
 
        co_gs_foreach_aff_node(env->co, a) {
-               const arch_register_t *ar = arch_get_irn_register(env->co->aenv, a->irn);
+               const arch_register_t *ar = arch_get_irn_register(a->irn);
                unsigned aidx = get_irn_idx(a->irn);
                neighb_t *n;
 
                co_gs_foreach_neighb(a, n) {
-                       const arch_register_t *nr = arch_get_irn_register(env->co->aenv, n->irn);
+                       const arch_register_t *nr = arch_get_irn_register(n->irn);
                        unsigned nidx = get_irn_idx(n->irn);
 
                        if(aidx < nidx) {
@@ -1142,7 +1136,7 @@ static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char
        char *tu_name;
 
        n = strlen(env->birg->main_env->cup_name);
-       tu_name = xmalloc((n + 1) * sizeof(*tu_name));
+       tu_name = XMALLOCN(char, n + 1);
        strcpy(tu_name, env->birg->main_env->cup_name);
        for (i = 0; i < n; ++i)
                if (tu_name[i] == '.')
@@ -1161,7 +1155,7 @@ static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char
 
 void co_driver(be_chordal_env_t *cenv)
 {
-       lc_timer_t          *timer = lc_timer_register("firm.be.copyopt", "runtime");
+       ir_timer_t          *timer = ir_timer_register("firm.be.copyopt", "runtime");
        co_complete_stats_t before, after;
        copy_opt_t          *co;
        co_algo_t           *algo_func;
@@ -1218,11 +1212,11 @@ void co_driver(be_chordal_env_t *cenv)
        algo_func = algos[algo].algo;
 
        /* perform actual copy minimization */
-       lc_timer_reset_and_start(timer);
+       ir_timer_reset_and_start(timer);
        was_optimal = algo_func(co);
-       lc_timer_stop(timer);
+       ir_timer_stop(timer);
 
-       be_stat_ev("co_time", lc_timer_elapsed_msec(timer));
+       be_stat_ev("co_time", ir_timer_elapsed_msec(timer));
        be_stat_ev_ull("co_optimal", was_optimal);
 
        if (dump_flags & DUMP_AFTER) {