X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fbe%2Fbecopyopt.c;h=e624545602401b91d1d50a6960374df18d4a511d;hb=68a3a77f08f0ccdadce9759d6eb19ecd875be7f0;hp=ce7eed75aa95b12e392ea5168b1a87055e2cc173;hpb=75b78dc47d9f1310e0f07bc306b314e89bc5c50c;p=libfirm diff --git a/ir/be/becopyopt.c b/ir/be/becopyopt.c index ce7eed75a..e62454560 100644 --- a/ir/be/becopyopt.c +++ b/ir/be/becopyopt.c @@ -22,7 +22,6 @@ * @brief Copy minimization driver. * @author Daniel Grund * @date 12.04.2005 - * @version $Id$ * * Main file for the optimization reducing the copies needed for: * - Phi coalescing @@ -31,35 +30,37 @@ */ #include "config.h" -#include "execfreq.h" -#include "xmalloc.h" #include "debug.h" -#include "pmap.h" -#include "raw_bitset.h" -#include "irnode.h" +#include "error.h" +#include "execfreq.h" +#include "irdump_t.h" +#include "iredges_t.h" #include "irgraph.h" #include "irgwalk.h" -#include "irprog.h" #include "irloop_t.h" -#include "iredges_t.h" -#include "irbitset.h" -#include "irphase_t.h" +#include "irnode.h" #include "irprintf_t.h" +#include "irprog.h" +#include "irtools.h" +#include "pmap.h" +#include "raw_bitset.h" +#include "util.h" +#include "xmalloc.h" -#include "bemodule.h" #include "bearch.h" -#include "benode_t.h" -#include "beutil.h" -#include "beifg_t.h" -#include "beintlive_t.h" #include "becopyopt_t.h" #include "becopystat.h" -#include "belive_t.h" +#include "bedump.h" +#include "beifg.h" #include "beinsn_t.h" +#include "beintlive_t.h" +#include "beirg.h" +#include "belive_t.h" +#include "bemodule.h" +#include "benode.h" #include "besched.h" #include "bestatevent.h" -#include "beirg.h" -#include "error.h" +#include "beutil.h" #include "lc_opts.h" #include "lc_opts_enum.h" @@ -73,9 +74,23 @@ #define COST_FUNC_LOOP 2 #define COST_FUNC_ALL_ONE 3 +/** + * Flags for dumping the IFG. + */ +enum { + CO_IFG_DUMP_COLORS = 1 << 0, /**< Dump the graph colored. */ + CO_IFG_DUMP_LABELS = 1 << 1, /**< Dump node/edge labels. */ + CO_IFG_DUMP_SHAPE = 1 << 2, /**< Give constrained nodes special shapes. */ + CO_IFG_DUMP_CONSTR = 1 << 3, /**< Dump the node constraints in the label. */ +}; + +static int co_get_costs_loop_depth(const ir_node *root, int pos); +static int co_get_costs_exec_freq(const ir_node *root, int pos); +static int co_get_costs_all_one(const ir_node *root, int pos); + static unsigned dump_flags = 0; -static unsigned style_flags = 0; -static unsigned do_stats = 0; +static unsigned style_flags = CO_IFG_DUMP_COLORS; +static int do_stats = 0; static cost_fct_t cost_func = co_get_costs_exec_freq; static int improve = 1; @@ -97,7 +112,6 @@ static const lc_opt_enum_mask_items_t style_items[] = { }; typedef int (*opt_funcptr)(void); - static const lc_opt_enum_func_ptr_items_t cost_func_items[] = { { "freq", (opt_funcptr) co_get_costs_exec_freq }, { "loop", (opt_funcptr) co_get_costs_loop_depth }, @@ -136,6 +150,7 @@ void be_register_copyopt(const char *name, co_algo_info *copyopt) be_add_module_to_list(©opts, name, copyopt); } +BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copyopt) void be_init_copyopt(void) { lc_opt_entry_t *be_grp = lc_opt_get_grp(firm_opt_get_root(), "be"); @@ -148,14 +163,13 @@ void be_init_copyopt(void) ©opts, (void**) &selected_copyopt); } -BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copyopt); - static int void_algo(copy_opt_t *co) { (void) co; return 0; } +BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copynone) void be_init_copynone(void) { static co_algo_info copyheur = { @@ -165,16 +179,16 @@ void be_init_copynone(void) be_register_copyopt("none", ©heur); } -BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copynone); - #undef QUICK_AND_DIRTY_HACK static int nodes_interfere(const be_chordal_env_t *env, const ir_node *a, const ir_node *b) { if (env->ifg) return be_ifg_connected(env->ifg, a, b); - else - return be_values_interfere(env->birg->lv, a, b); + else { + be_lv_t *lv = be_get_irg_liveness(env->irg); + return be_values_interfere(lv, a, b); + } } @@ -194,7 +208,7 @@ DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;) copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs) { const char *s1, *s2, *s3; - int len; + size_t len; copy_opt_t *co; FIRM_DBG_REGISTER(dbg, "ir.be.copyopt"); @@ -215,7 +229,8 @@ copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, cost_fct_t get_costs) return co; } -void free_copy_opt(copy_opt_t *co) { +void free_copy_opt(copy_opt_t *co) +{ xfree(co->name); free(co); } @@ -227,62 +242,70 @@ void free_copy_opt(copy_opt_t *co) { static int co_is_optimizable_root(ir_node *irn) { const arch_register_req_t *req; - const arch_register_t *reg; if (arch_irn_is_ignore(irn)) return 0; - reg = arch_get_irn_register(irn); - if (arch_register_type_is(reg, ignore)) - return 0; - if (is_Reg_Phi(irn) || is_Perm_Proj(irn)) return 1; - req = arch_get_register_req_out(irn); + req = arch_get_irn_register_req(irn); if (is_2addr_code(req)) return 1; return 0; } -int co_get_costs_loop_depth(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) { - int cost = 0; +/** + * Computes the costs of a copy according to loop depth + * @param pos the argument position of arg in the root arguments + * @return Must be >= 0 in all cases. + */ +static int co_get_costs_loop_depth(const ir_node *root, int pos) +{ + ir_node *block = get_nodes_block(root); ir_loop *loop; - ir_node *root_block = get_nodes_block(root); - (void) co; - (void) arg; + int cost; if (is_Phi(root)) { - /* for phis the copies are placed in the corresponding pred-block */ - loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos)); - } else { - /* a perm places the copy in the same block as it resides */ - loop = get_irn_loop(root_block); + block = get_Block_cfgpred_block(block, pos); } + loop = get_irn_loop(block); if (loop) { int d = get_loop_depth(loop); cost = d*d; + } else { + cost = 0; } return 1+cost; } -int co_get_costs_exec_freq(const copy_opt_t *co, ir_node *root, ir_node* arg, int pos) { - int res; - ir_node *root_bl = get_nodes_block(root); - ir_node *copy_bl = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl; - (void) arg; - res = get_block_execfreq_ulong(co->cenv->birg->exec_freq, copy_bl); +/** + * Computes the costs of a copy according to execution frequency + * @param pos the argument position of arg in the root arguments + * @return Must be >= 0 in all cases. + */ +static int co_get_costs_exec_freq(const ir_node *root, int pos) +{ + ir_graph *irg = get_irn_irg(root); + ir_node *root_bl = get_nodes_block(root); + ir_node *copy_bl + = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl; + ir_exec_freq *exec_freq = be_get_irg_exec_freq(irg); + int res = get_block_execfreq_ulong(exec_freq, copy_bl); /* don't allow values smaller than one. */ return res < 1 ? 1 : res; } - -int co_get_costs_all_one(const copy_opt_t *co, ir_node *root, ir_node *arg, int pos) { - (void) co; +/** + * All costs equal 1. Using this will reduce the _number_ of copies. + * @param co The copy opt object. + * @return Must be >= 0 in all cases. + */ +static int co_get_costs_all_one(const ir_node *root, int pos) +{ (void) root; - (void) arg; (void) pos; return 1; } @@ -302,13 +325,14 @@ int co_get_costs_all_one(const copy_opt_t *co, ir_node *root, ir_node *arg, int * Determines a maximum weighted independent set with respect to * the interference and conflict edges of all nodes in a qnode. */ -static int ou_max_ind_set_costs(unit_t *ou) { +static int ou_max_ind_set_costs(unit_t *ou) +{ be_chordal_env_t *chordal_env = ou->co->cenv; ir_node **safe, **unsafe; int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs; bitset_t *curr; - bitset_pos_t pos; - int max, curr_weight, best_weight = 0; + size_t pos; + int curr_weight, best_weight = 0; /* assign the nodes into two groups. * safe: node has no interference, hence it is in every max stable set. @@ -320,9 +344,9 @@ static int ou_max_ind_set_costs(unit_t *ou) { unsafe = ALLOCAN(ir_node*, ou->node_count - 1); unsafe_costs = ALLOCAN(int, ou->node_count - 1); unsafe_count = 0; - for(i=1; inode_count; ++i) { + for (i=1; inode_count; ++i) { int is_safe = 1; - for(o=1; onode_count; ++o) { + for (o=1; onode_count; ++o) { if (i==o) continue; if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) { @@ -343,7 +367,7 @@ static int ou_max_ind_set_costs(unit_t *ou) { /* now compute the best set out of the unsafe nodes*/ if (unsafe_count > MIS_HEUR_TRIGGER) { bitset_t *best = bitset_alloca(unsafe_count); - /* Heuristik: Greedy trial and error form index 0 to unsafe_count-1 */ + /* Heuristic: Greedy trial and error form index 0 to unsafe_count-1 */ for (i=0; icls != co->cls) return; if (!co_is_optimizable_root(irn)) @@ -418,11 +445,11 @@ static void co_collect_units(ir_node *irn, void *env) int o, arg_pos; ir_node *arg = get_irn_n(irn, i); - assert(arch_get_irn_reg_class_out(arg) == co->cls && "Argument not in same register class."); + assert(arch_get_irn_reg_class(arg) == co->cls && "Argument not in same register class."); if (arg == irn) continue; if (nodes_interfere(co->cenv, irn, arg)) { - unit->inevitable_costs += co->get_costs(co, irn, arg, i); + unit->inevitable_costs += co->get_costs(irn, i); continue; } @@ -444,11 +471,11 @@ static void co_collect_units(ir_node *irn, void *env) if (!arg_pos) { /* a new argument */ /* insert node, set costs */ unit->nodes[unit->node_count] = arg; - unit->costs[unit->node_count] = co->get_costs(co, irn, arg, i); + unit->costs[unit->node_count] = co->get_costs(irn, i); unit->node_count++; } else { /* arg has occurred before in same phi */ /* increase costs for existing arg */ - unit->costs[arg_pos] += co->get_costs(co, irn, arg, i); + unit->costs[arg_pos] += co->get_costs(irn, i); } } unit->nodes = XREALLOC(unit->nodes, ir_node*, unit->node_count); @@ -461,7 +488,7 @@ static void co_collect_units(ir_node *irn, void *env) 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); + unit->costs[1] = co->get_costs(irn, -1); } else { /* Src == Tgt of a 2-addr-code instruction */ if (is_2addr_code(req)) { @@ -494,7 +521,7 @@ static void co_collect_units(ir_node *irn, void *env) if (!arch_irn_is_ignore(o) && !nodes_interfere(co->cenv, irn, o)) { unit->nodes[k] = o; - unit->costs[k] = co->get_costs(co, irn, o, -1); + unit->costs[k] = co->get_costs(irn, -1); ++k; } } @@ -511,7 +538,7 @@ static void co_collect_units(ir_node *irn, void *env) struct list_head *tmp; /* Determine the maximum costs this unit can cause: all_nodes_cost */ - for(i=1; inode_count; ++i) { + for (i=1; inode_count; ++i) { unit->sort_key = MAX(unit->sort_key, unit->costs[i]); unit->all_nodes_costs += unit->costs[i]; } @@ -530,7 +557,8 @@ static void co_collect_units(ir_node *irn, void *env) #ifdef QUICK_AND_DIRTY_HACK -static int compare_ous(const void *k1, const void *k2) { +static int compare_ous(const void *k1, const void *k2) +{ const unit_t *u1 = *((const unit_t **) k1); const unit_t *u2 = *((const unit_t **) k2); int i, o, u1_has_constr, u2_has_constr; @@ -539,7 +567,7 @@ static int compare_ous(const void *k1, const void *k2) { /* Units with constraints come first */ u1_has_constr = 0; for (i=0; inode_count; ++i) { - arch_get_register_req_out(&req, u1->nodes[i]); + arch_get_irn_register_req(&req, u1->nodes[i]); if (arch_register_req_is(&req, limited)) { u1_has_constr = 1; break; @@ -548,7 +576,7 @@ static int compare_ous(const void *k1, const void *k2) { u2_has_constr = 0; for (i=0; inode_count; ++i) { - arch_get_register_req_out(&req, u2->nodes[i]); + arch_get_irn_register_req(&req, u2->nodes[i]); if (arch_register_req_is(&req, limited)) { u2_has_constr = 1; break; @@ -574,7 +602,8 @@ static int compare_ous(const void *k1, const void *k2) { /** * Sort the ou's according to constraints and their sort_key */ -static void co_sort_units(copy_opt_t *co) { +static void co_sort_units(copy_opt_t *co) +{ int i, count = 0, costs; unit_t *ou, **ous; @@ -610,7 +639,8 @@ static void co_sort_units(copy_opt_t *co) { } #endif -void co_build_ou_structure(copy_opt_t *co) { +void co_build_ou_structure(copy_opt_t *co) +{ DBG((dbg, LEVEL_1, "\tCollecting optimization units\n")); INIT_LIST_HEAD(&co->units); irg_walk_graph(co->irg, co_collect_units, NULL, co); @@ -619,7 +649,8 @@ void co_build_ou_structure(copy_opt_t *co) { #endif } -void co_free_ou_structure(copy_opt_t *co) { +void co_free_ou_structure(copy_opt_t *co) +{ unit_t *curr, *tmp; ASSERT_OU_AVAIL(co); list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) { @@ -632,7 +663,8 @@ void co_free_ou_structure(copy_opt_t *co) { /* co_solve_heuristic() is implemented in becopyheur.c */ -int co_get_max_copy_costs(const copy_opt_t *co) { +int co_get_max_copy_costs(const copy_opt_t *co) +{ int i, res = 0; unit_t *curr; @@ -646,7 +678,8 @@ int co_get_max_copy_costs(const copy_opt_t *co) { return res; } -int co_get_inevit_copy_costs(const copy_opt_t *co) { +int co_get_inevit_copy_costs(const copy_opt_t *co) +{ int res = 0; unit_t *curr; @@ -657,7 +690,8 @@ int co_get_inevit_copy_costs(const copy_opt_t *co) { return res; } -int co_get_copy_costs(const copy_opt_t *co) { +int co_get_copy_costs(const copy_opt_t *co) +{ int i, res = 0; unit_t *curr; @@ -678,7 +712,8 @@ int co_get_copy_costs(const copy_opt_t *co) { return res; } -int co_get_lower_bound(const copy_opt_t *co) { +int co_get_lower_bound(const copy_opt_t *co) +{ int res = 0; unit_t *curr; @@ -691,7 +726,7 @@ int co_get_lower_bound(const copy_opt_t *co) { void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat) { - bitset_t *seen = bitset_irg_malloc(co->irg); + bitset_t *seen = bitset_malloc(get_irg_last_idx(co->irg)); affinity_node_t *an; memset(stat, 0, sizeof(stat[0])); @@ -700,9 +735,9 @@ void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat) co_gs_foreach_aff_node(co, an) { neighb_t *neigh; stat->aff_nodes += 1; - bitset_add_irn(seen, an->irn); + bitset_set(seen, get_irn_idx(an->irn)); co_gs_foreach_neighb(an, neigh) { - if(!bitset_contains_irn(seen, neigh->irn)) { + if (!bitset_is_set(seen, get_irn_idx(neigh->irn))) { stat->aff_edges += 1; stat->max_costs += neigh->costs; @@ -711,7 +746,7 @@ void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat) stat->unsatisfied_edges += 1; } - if(nodes_interfere(co->cenv, an->irn, neigh->irn)) { + if (nodes_interfere(co->cenv, an->irn, neigh->irn)) { stat->aff_int += 1; stat->inevit_costs += neigh->costs; } @@ -734,15 +769,17 @@ void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat) |_| |___/ ******************************************************************************/ -static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) { - const affinity_node_t *n1 = k1; - const affinity_node_t *n2 = k2; +static int compare_affinity_node_t(const void *k1, const void *k2, size_t size) +{ + const affinity_node_t *n1 = (const affinity_node_t*)k1; + const affinity_node_t *n2 = (const affinity_node_t*)k2; (void) size; return (n1->irn != n2->irn); } -static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) { +static void add_edge(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs) +{ affinity_node_t new_node, *node; neighb_t *nbr; int allocnew = 1; @@ -750,7 +787,7 @@ 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), hash_irn(new_node.irn)); + node = (affinity_node_t*)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) { @@ -773,34 +810,34 @@ 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); } } -static void build_graph_walker(ir_node *irn, void *env) { - const arch_register_req_t *req = arch_get_register_req_out(irn); - copy_opt_t *co = env; +static void build_graph_walker(ir_node *irn, void *env) +{ + const arch_register_req_t *req; + copy_opt_t *co = (copy_opt_t*)env; int pos, max; - const arch_register_t *reg; - if (req->cls != co->cls || arch_irn_is_ignore(irn)) + if (get_irn_mode(irn) == mode_T) return; - - reg = arch_get_irn_register(irn); - if (arch_register_type_is(reg, ignore)) + req = arch_get_irn_register_req(irn); + if (req->cls != co->cls || arch_irn_is_ignore(irn)) return; if (is_Reg_Phi(irn)) { /* Phis */ for (pos=0, max=get_irn_arity(irn); posget_costs(co, irn, arg, pos)); + add_edges(co, irn, arg, co->get_costs(irn, pos)); } } 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)); + add_edges(co, irn, arg, co->get_costs(irn, -1)); } else { /* 2-address code */ if (is_2addr_code(req)) { const unsigned other = req->other_same; @@ -810,21 +847,23 @@ static void build_graph_walker(ir_node *irn, void *env) { if (other & (1U << i)) { ir_node *other = get_irn_n(skip_Proj(irn), i); if (!arch_irn_is_ignore(other)) - add_edges(co, irn, other, co->get_costs(co, irn, other, 0)); + add_edges(co, irn, other, co->get_costs(irn, -1)); } } } } } -void co_build_graph_structure(copy_opt_t *co) { +void co_build_graph_structure(copy_opt_t *co) +{ obstack_init(&co->obst); co->nodes = new_set(compare_affinity_node_t, 32); irg_walk_graph(co->irg, build_graph_walker, NULL, co); } -void co_free_graph_structure(copy_opt_t *co) { +void co_free_graph_structure(copy_opt_t *co) +{ ASSERT_GS_AVAIL(co); del_set(co->nodes); @@ -832,15 +871,14 @@ void co_free_graph_structure(copy_opt_t *co) { co->nodes = NULL; } -/* co_solve_ilp1() co_solve_ilp2() are implemented in becopyilpX.c */ - -int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) { +int co_gs_is_optimizable(copy_opt_t *co, ir_node *irn) +{ affinity_node_t new_node, *n; ASSERT_GS_AVAIL(co); new_node.irn = irn; - n = set_find(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn)); + n = (affinity_node_t*)set_find(co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn)); if (n) { return (n->degree > 0); } else @@ -857,8 +895,8 @@ static int co_dump_appel_disjoint_constraints(const copy_opt_t *co, ir_node *a, constr[1] = bitset_alloca(co->cls->n_regs); for (j = 0; j < 2; ++j) { - const arch_register_req_t *req = arch_get_register_req_out(nodes[j]); - if(arch_register_req_is(req, limited)) + const arch_register_req_t *req = arch_get_irn_register_req(nodes[j]); + if (arch_register_req_is(req, limited)) rbitset_copy_to_bitset(req->limited, constr[j]); else bitset_set_all(constr[j]); @@ -868,21 +906,35 @@ static int co_dump_appel_disjoint_constraints(const copy_opt_t *co, ir_node *a, return !bitset_intersect(constr[0], constr[1]); } -void co_dump_appel_graph(const copy_opt_t *co, FILE *f) +/** + * Dump the interference graph according to the Appel/George coalescing contest file format. + * See: http://www.cs.princeton.edu/~appel/coalesce/format.html + * @note Requires graph structure. + * @param co The copy opt object. + * @param f A file to dump to. + */ +static void co_dump_appel_graph(const copy_opt_t *co, FILE *f) { be_ifg_t *ifg = co->cenv->ifg; int *color_map = ALLOCAN(int, co->cls->n_regs); int *node_map = XMALLOCN(int, get_irg_last_idx(co->irg) + 1); + ir_graph *irg = co->irg; + be_irg_t *birg = be_birg_from_irg(irg); ir_node *irn; - void *it, *nit; + nodes_iter_t it; + neighbours_iter_t nit; int n, n_regs; unsigned i; n_regs = 0; - for(i = 0; i < co->cls->n_regs; ++i) { + for (i = 0; i < co->cls->n_regs; ++i) { const arch_register_t *reg = &co->cls->regs[i]; - color_map[i] = arch_register_type_is(reg, ignore) ? -1 : n_regs++; + if (rbitset_is_set(birg->allocatable_regs, reg->global_index)) { + color_map[i] = n_regs++; + } else { + color_map[i] = -1; + } } /* @@ -890,11 +942,8 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f) * the values below n are the pre-colored register nodes */ - it = be_ifg_nodes_iter_alloca(ifg); - nit = be_ifg_neighbours_iter_alloca(ifg); - n = n_regs; - be_ifg_foreach_node(ifg, it, irn) { + be_ifg_foreach_node(ifg, &it, irn) { if (arch_irn_is_ignore(irn)) continue; node_map[get_irn_idx(irn)] = n++; @@ -902,36 +951,36 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f) fprintf(f, "%d %d\n", n, n_regs); - be_ifg_foreach_node(ifg, it, irn) { + be_ifg_foreach_node(ifg, &it, irn) { if (!arch_irn_is_ignore(irn)) { 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); + const arch_register_req_t *req = arch_get_irn_register_req(irn); ir_node *adj; - 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) + 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) fprintf(f, "%d %d -1\n", color_map[i], idx); } } - be_ifg_foreach_neighbour(ifg, nit, irn, adj) { + be_ifg_foreach_neighbour(ifg, &nit, irn, adj) { if (!arch_irn_is_ignore(adj) && !co_dump_appel_disjoint_constraints(co, irn, adj)) { int adj_idx = node_map[get_irn_idx(adj)]; - if(idx < adj_idx) + if (idx < adj_idx) fprintf(f, "%d %d -1\n", idx, adj_idx); } } - if(a) { + if (a) { neighb_t *n; co_gs_foreach_neighb(a, n) { if (!arch_irn_is_ignore(n->irn)) { int n_idx = node_map[get_irn_idx(n->irn)]; - if(idx < n_idx) + if (idx < n_idx) fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs); } } @@ -942,173 +991,18 @@ void co_dump_appel_graph(const copy_opt_t *co, FILE *f) xfree(node_map); } -/* - ___ _____ ____ ____ ___ _____ ____ _ - |_ _| ___/ ___| | _ \ / _ \_ _| | _ \ _ _ _ __ ___ _ __ (_)_ __ __ _ - | || |_ | | _ | | | | | | || | | | | | | | | '_ ` _ \| '_ \| | '_ \ / _` | - | || _|| |_| | | |_| | |_| || | | |_| | |_| | | | | | | |_) | | | | | (_| | - |___|_| \____| |____/ \___/ |_| |____/ \__,_|_| |_| |_| .__/|_|_| |_|\__, | - |_| |___/ -*/ - -static const char *get_dot_color_name(size_t col) -{ - static const char *names[] = { - "blue", - "red", - "green", - "yellow", - "cyan", - "magenta", - "orange", - "chocolate", - "beige", - "navy", - "darkgreen", - "darkred", - "lightPink", - "chartreuse", - "lightskyblue", - "linen", - "pink", - "lightslateblue", - "mintcream", - "red", - "darkolivegreen", - "mediumblue", - "mistyrose", - "salmon", - "darkseagreen", - "mediumslateblue" - "moccasin", - "tomato", - "forestgreen", - "darkturquoise", - "palevioletred" - }; - - return col < sizeof(names)/sizeof(names[0]) ? names[col] : "white"; -} - -typedef struct _co_ifg_dump_t { - const copy_opt_t *co; - unsigned flags; -} co_ifg_dump_t; - -static void ifg_dump_graph_attr(FILE *f, void *self) -{ - (void) self; - fprintf(f, "overlap=scale"); -} - -static int ifg_is_dump_node(void *self, ir_node *irn) -{ - (void)self; - return !arch_irn_is_ignore(irn); -} - -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(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); - - if((env->flags & CO_IFG_DUMP_CONSTR) && limited) { - bitset_t *bs = bitset_alloca(env->co->cls->n_regs); - rbitset_copy_to_bitset(req->limited, bs); - ir_fprintf(f, "\\n%B", bs); - } - ir_fprintf(f, "\" "); - } else { - fprintf(f, "label=\"\" shape=point " ); - } - - if(env->flags & CO_IFG_DUMP_SHAPE) - fprintf(f, "shape=%s ", limited ? "diamond" : "ellipse"); - - if(env->flags & CO_IFG_DUMP_COLORS) - fprintf(f, "style=filled color=%s ", get_dot_color_name(reg->index)); -} - -static void ifg_dump_at_end(FILE *file, void *self) -{ - co_ifg_dump_t *env = self; - affinity_node_t *a; - - co_gs_foreach_aff_node(env->co, a) { - 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(n->irn); - unsigned nidx = get_irn_idx(n->irn); - - if(aidx < nidx) { - const char *color = nr == ar ? "blue" : "red"; - fprintf(file, "\tn%d -- n%d [weight=0.01 ", aidx, nidx); - if(env->flags & CO_IFG_DUMP_LABELS) - fprintf(file, "label=\"%d\" ", n->costs); - if(env->flags & CO_IFG_DUMP_COLORS) - fprintf(file, "color=%s ", color); - else - fprintf(file, "style=dotted"); - fprintf(file, "];\n"); - } - } - } -} - - -static be_ifg_dump_dot_cb_t ifg_dot_cb = { - ifg_is_dump_node, - ifg_dump_graph_attr, - ifg_dump_node_attr, - NULL, - NULL, - ifg_dump_at_end -}; - - - -void co_dump_ifg_dot(const copy_opt_t *co, FILE *f, unsigned flags) -{ - co_ifg_dump_t cod; - - cod.co = co; - cod.flags = flags; - be_ifg_dump_dot(co->cenv->ifg, co->irg, f, &ifg_dot_cb, &cod); -} - - -void co_solve_park_moon(copy_opt_t *opt) -{ - (void) opt; -} - -/* - __ __ _ ____ _ - | \/ | __ _(_)_ __ | _ \ _ __(_)_ _____ _ __ - | |\/| |/ _` | | '_ \ | | | | '__| \ \ / / _ \ '__| - | | | | (_| | | | | | | |_| | | | |\ V / __/ | - |_| |_|\__,_|_|_| |_| |____/|_| |_| \_/ \___|_| - -*/ - -static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char *suffix) +static FILE *my_open(const be_chordal_env_t *env, const char *prefix, + const char *suffix) { FILE *result; char buf[1024]; size_t i, n; char *tu_name; + const char *cup_name = be_get_irg_main_env(env->irg)->cup_name; - n = strlen(env->birg->main_env->cup_name); + n = strlen(cup_name); tu_name = XMALLOCN(char, n + 1); - strcpy(tu_name, env->birg->main_env->cup_name); + strcpy(tu_name, cup_name); for (i = 0; i < n; ++i) if (tu_name[i] == '.') tu_name[i] = '_'; @@ -1117,7 +1011,7 @@ static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char ir_snprintf(buf, sizeof(buf), "%s%s_%F_%s%s", prefix, tu_name, env->irg, env->cls->name, suffix); xfree(tu_name); result = fopen(buf, "wt"); - if(result == NULL) { + if (result == NULL) { panic("Couldn't open '%s' for writing.", buf); } @@ -1126,14 +1020,18 @@ static FILE *my_open(const be_chordal_env_t *env, const char *prefix, const char void co_driver(be_chordal_env_t *cenv) { - ir_timer_t *timer = ir_timer_register("firm.be.copyopt", "runtime"); + ir_timer_t *timer = ir_timer_new(); co_complete_stats_t before, after; copy_opt_t *co; int was_optimal = 0; assert(selected_copyopt); - be_liveness_assure_chk(be_get_birg_liveness(cenv->birg)); + /* skip copymin if algo is 'none' */ + if (selected_copyopt->copyopt == void_algo) + return; + + be_liveness_assure_chk(be_get_irg_liveness(cenv->irg)); co = new_copy_opt(cenv, cost_func); co_build_ou_structure(co); @@ -1151,8 +1049,8 @@ void co_driver(be_chordal_env_t *cenv) be_stat_ev_ull("co_init_unsat", before.unsatisfied_edges); if (dump_flags & DUMP_BEFORE) { - FILE *f = my_open(cenv, "", "-before.dot"); - co_dump_ifg_dot(co, f, style_flags); + FILE *f = my_open(cenv, "", "-before.vcg"); + be_dump_ifg_co(f, co, style_flags & CO_IFG_DUMP_LABELS, style_flags & CO_IFG_DUMP_COLORS); fclose(f); } @@ -1175,32 +1073,33 @@ void co_driver(be_chordal_env_t *cenv) be_stat_ev("co_time", ir_timer_elapsed_msec(timer)); be_stat_ev_ull("co_optimal", was_optimal); + ir_timer_free(timer); if (dump_flags & DUMP_AFTER) { - FILE *f = my_open(cenv, "", "-after.dot"); - co_dump_ifg_dot(co, f, style_flags); + FILE *f = my_open(cenv, "", "-after.vcg"); + be_dump_ifg_co(f, co, style_flags & CO_IFG_DUMP_LABELS, style_flags & CO_IFG_DUMP_COLORS); fclose(f); } co_complete_stats(co, &after); if (do_stats) { - ulong64 optimizable_costs = after.max_costs - after.inevit_costs; - ulong64 evitable = after.costs - after.inevit_costs; + unsigned long long optimizable_costs = after.max_costs - after.inevit_costs; + unsigned long long evitable = after.costs - after.inevit_costs; ir_printf("%30F ", cenv->irg); - printf("%10s %10" ULL_FMT "%10" ULL_FMT "%10" ULL_FMT, cenv->cls->name, after.max_costs, before.costs, after.inevit_costs); + printf("%10s %10llu%10llu%10llu", cenv->cls->name, after.max_costs, before.costs, after.inevit_costs); - if(optimizable_costs > 0) - printf("%10" ULL_FMT " %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs); + if (optimizable_costs > 0) + printf("%10llu %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs); else - printf("%10" ULL_FMT " %5s\n", after.costs, "-"); + printf("%10llu %5s\n", after.costs, "-"); } /* Dump the interference graph in Appel's format. */ if (dump_flags & DUMP_APPEL) { FILE *f = my_open(cenv, "", ".apl"); - fprintf(f, "# %lld %lld\n", after.costs, after.unsatisfied_edges); + fprintf(f, "# %llu %llu\n", after.costs, after.unsatisfied_edges); co_dump_appel_graph(co, f); fclose(f); }