X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fbe%2Fbespilldaemel.c;h=26ef48e455f43794ef07d425e17c5413dd9b61fe;hb=fcb579b8959da1d7563b1a7b9f008a423ffdf75a;hp=0d0bbd829c4450db5a48729781f762a05c5a8682;hpb=9f4c0214668b6955f36c1a03b957721a9f62f1ef;p=libfirm diff --git a/ir/be/bespilldaemel.c b/ir/be/bespilldaemel.c index 0d0bbd829..26ef48e45 100644 --- a/ir/be/bespilldaemel.c +++ b/ir/be/bespilldaemel.c @@ -1,5 +1,5 @@ /* - * Copyright (C) 1995-2007 University of Karlsruhe. All right reserved. + * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved. * * This file is part of libFirm. * @@ -57,15 +57,12 @@ DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;) -typedef struct daemel_env_t daemel_env_t; -struct daemel_env_t { - spill_env_t *spill_env; - int n_regs; - const arch_env_t *arch_env; - const arch_register_class_t *cls; - const be_lv_t *lv; - bitset_t *spilled_nodes; -}; +static spill_env_t *spill_env; +static int n_regs; +static const arch_env_t *arch_env; +static const arch_register_class_t *cls; +static const be_lv_t *lv; +static bitset_t *spilled_nodes; typedef struct spill_candidate_t spill_candidate_t; struct spill_candidate_t { @@ -73,8 +70,7 @@ struct spill_candidate_t { ir_node *node; }; -static -int compare_spill_candidates_desc(const void *d1, const void *d2) +static int compare_spill_candidates_desc(const void *d1, const void *d2) { const spill_candidate_t *c1 = d1; const spill_candidate_t *c2 = d2; @@ -82,19 +78,24 @@ int compare_spill_candidates_desc(const void *d1, const void *d2) return (int) (c1->costs - c2->costs); } -static -double get_spill_costs(daemel_env_t *env, ir_node *node) +static double get_spill_costs(ir_node *node) { const ir_edge_t *edge; - spill_env_t *spill_env = env->spill_env; - double costs = be_get_spill_costs(spill_env, node, node); + ir_node *spill_place = skip_Proj(node); + double costs = be_get_spill_costs(spill_env, node, + spill_place); foreach_out_edge(node, edge) { ir_node *use = get_edge_src_irn(edge); + /* keeps should be directly below the node */ + if(be_is_Keep(use)) { + continue; + } + if(is_Phi(use)) { - int in = get_edge_src_pos(edge); - ir_node *block = get_nodes_block(use); + int in = get_edge_src_pos(edge); + ir_node *block = get_nodes_block(use); costs += be_get_reload_costs_on_edge(spill_env, node, block, in); } else { @@ -102,119 +103,132 @@ double get_spill_costs(daemel_env_t *env, ir_node *node) } } + /* TODO cache costs? */ + return costs; } /** * spills a node by placing a reload before each usage */ -static -void spill_node(daemel_env_t *env, ir_node *node, ir_nodeset_t *nodes) +static void spill_node(ir_node *node) { const ir_edge_t *edge; - spill_env_t *spill_env = env->spill_env; - const arch_register_class_t *cls = env->cls; DBG((dbg, LEVEL_3, "\tspilling %+F\n", node)); foreach_out_edge(node, edge) { ir_node *use = get_edge_src_irn(edge); + if(is_Anchor(use)) + continue; + if(be_is_Keep(use)) + continue; if(is_Phi(use)) { int in = get_edge_src_pos(edge); ir_node *block = get_nodes_block(use); be_add_reload_on_edge(spill_env, node, block, in, cls, 1); - } else if(!be_is_Keep(use)) { + } else { be_add_reload(spill_env, node, use, cls, 1); } } - bitset_set(env->spilled_nodes, get_irn_idx(node)); - ir_nodeset_remove(nodes, node); + bitset_set(spilled_nodes, get_irn_idx(node)); } /** * spill @p n nodes from a nodeset. Removes the nodes from the nodeset and - * sets the spilled bits in env->spilled_nodes. + * sets the spilled bits in spilled_nodes. */ -static -void do_spilling(daemel_env_t *env, ir_nodeset_t *nodes, ir_node *node) +static void do_spilling(ir_nodeset_t *live_nodes, ir_node *node) { - size_t node_count = ir_nodeset_size(nodes); - size_t additional_defines = 0; - size_t reload_values = 0; - int registers = env->n_regs; - const arch_env_t *arch_env = env->arch_env; - const arch_register_class_t *cls = env->cls; - spill_candidate_t *candidates; - ir_nodeset_iterator_t iter; - size_t i, arity; - int spills_needed; - size_t cand_idx; - ir_node *n; - const bitset_t *spilled_nodes = env->spilled_nodes; + size_t n_live_nodes = ir_nodeset_size(live_nodes); + size_t values_defined = 0; + size_t free_regs_needed = 0; + spill_candidate_t *candidates; + ir_nodeset_iterator_t iter; + size_t i, arity; + int spills_needed; + size_t cand_idx; + ir_node *n; /* mode_T nodes define several values at once. Count them */ if(get_irn_mode(node) == mode_T) { - ir_node *proj = sched_next(node); + const ir_edge_t *edge; + + foreach_out_edge(node, edge) { + const ir_node *proj = get_edge_src_irn(edge); - while(is_Proj(proj)) { - if(arch_irn_consider_in_reg_alloc(arch_env, cls, proj)) { - ++additional_defines; + if (arch_irn_consider_in_reg_alloc(cls, proj)) { + ++values_defined; } - proj = sched_next(proj); } + } else if (arch_irn_consider_in_reg_alloc(cls, node)) { + ++values_defined; } - /* we might temporarily need registers for reloaded values */ + /* we need registers for the non-live argument values */ arity = get_irn_arity(node); for(i = 0; i < arity; ++i) { ir_node *pred = get_irn_n(node, i); - if(bitset_is_set(spilled_nodes, get_irn_idx(pred))) - ++reload_values; + if (arch_irn_consider_in_reg_alloc(cls, pred) + && !ir_nodeset_contains(live_nodes, pred)) { + ++free_regs_needed; + } } - if(reload_values > additional_defines) - additional_defines = reload_values; + /* we can reuse all reloaded values for the defined values, but we might + need even more registers */ + if(values_defined > free_regs_needed) + free_regs_needed = values_defined; - spills_needed = (node_count + additional_defines) - registers; + spills_needed = (n_live_nodes + free_regs_needed) - n_regs; if(spills_needed <= 0) return; DBG((dbg, LEVEL_2, "\tspills needed after %+F: %d\n", node, spills_needed)); - candidates = malloc(node_count * sizeof(candidates[0])); + candidates = alloca(n_live_nodes * sizeof(candidates[0])); /* construct array with spill candidates and calculate their costs */ i = 0; - foreach_ir_nodeset(nodes, n, iter) { + foreach_ir_nodeset(live_nodes, n, iter) { spill_candidate_t *candidate = & candidates[i]; + assert(!bitset_is_set(spilled_nodes, get_irn_idx(n))); + candidate->node = n; - candidate->costs = get_spill_costs(env, n); + candidate->costs = get_spill_costs(n); ++i; } - assert(i == node_count); + assert(i == n_live_nodes); /* sort spill candidates */ - qsort(candidates, node_count, sizeof(candidates[0]), + qsort(candidates, n_live_nodes, sizeof(candidates[0]), compare_spill_candidates_desc); /* spill cheapest ones */ cand_idx = 0; while(spills_needed > 0) { - if(cand_idx >= node_count) { - panic("can't spill enough values for node %+F\n", node); + spill_candidate_t *candidate; + ir_node *cand_node; + int is_use; + + if (cand_idx >= n_live_nodes) { + panic("can't spill enough values for node %+F", node); } - spill_candidate_t *candidate = &candidates[cand_idx]; - ir_node *cand_node = candidate->node; - int is_use; + + candidate = &candidates[cand_idx]; + cand_node = candidate->node; ++cand_idx; - /* make sure the node is not a use of the instruction */ + if (arch_irn_is(cand_node, dont_spill)) + continue; + + /* make sure the node is not an argument of the instruction */ is_use = 0; - for(i = 0; i < arity; ++i) { + for (i = 0; i < arity; ++i) { ir_node *in = get_irn_n(node, i); if(in == cand_node) { is_use = 1; @@ -225,103 +239,130 @@ void do_spilling(daemel_env_t *env, ir_nodeset_t *nodes, ir_node *node) continue; } - spill_node(env, cand_node, nodes); + spill_node(cand_node); + ir_nodeset_remove(live_nodes, cand_node); --spills_needed; } - - free(candidates); } /** - * similar to be_liveness_transfer. - * custom liveness transfer function, that doesn't place already spilled values - * into the liveness set + * removes all values from the nodeset that are defined by node */ -static -void liveness_transfer(daemel_env_t *env, ir_node *node, ir_nodeset_t *nodeset) +static void remove_defs(ir_node *node, ir_nodeset_t *nodeset) { - int i, arity; - const arch_register_class_t *cls = env->cls; - const arch_env_t *arch_env = env->arch_env; - const bitset_t *bitset = env->spilled_nodes; - /* You should better break out of your loop when hitting the first phi * function. */ assert(!is_Phi(node) && "liveness_transfer produces invalid results for phi nodes"); - if(arch_irn_consider_in_reg_alloc(arch_env, cls, node)) { - ir_nodeset_remove(nodeset, node); - } + if (get_irn_mode(node) == mode_T) { + const ir_edge_t *edge; - arity = get_irn_arity(node); - for(i = 0; i < arity; ++i) { - ir_node *op = get_irn_n(node, i); + foreach_out_edge(node, edge) { + const ir_node *proj = get_edge_src_irn(edge); - if(arch_irn_consider_in_reg_alloc(arch_env, cls, op) - && !bitset_is_set(bitset, get_irn_idx(op))) { - ir_nodeset_insert(nodeset, op); + if (arch_irn_consider_in_reg_alloc(cls, proj)) { + ir_nodeset_remove(nodeset, proj); + } } - } + } + + if (arch_irn_consider_in_reg_alloc(cls, node)) { + ir_nodeset_remove(nodeset, node); + } +} + +static void add_uses(ir_node *node, ir_nodeset_t *nodeset) +{ + int i, arity; + + arity = get_irn_arity(node); + for(i = 0; i < arity; ++i) { + ir_node *op = get_irn_n(node, i); + + if (arch_irn_consider_in_reg_alloc(cls, op) && + !bitset_is_set(spilled_nodes, get_irn_idx(op))) { + ir_nodeset_insert(nodeset, op); + } + } +} + +static __attribute__((unused)) +void print_nodeset(ir_nodeset_t *nodeset) +{ + ir_nodeset_iterator_t iter; + ir_node *node; + + foreach_ir_nodeset(nodeset, node, iter) { + ir_fprintf(stderr, "%+F ", node); + } + fprintf(stderr, "\n"); } /** * make sure register pressure in a block is always equal or below the number * of available registers */ -static -void spill_block(ir_node *block, void *data) +static void spill_block(ir_node *block, void *data) { - daemel_env_t *env = data; - const arch_env_t *arch_env = env->arch_env; - const arch_register_class_t *cls = env->cls; - const be_lv_t *lv = env->lv; - ir_nodeset_t live_nodes; - ir_nodeset_iterator_t iter; - ir_node *node; - bitset_t *spilled_nodes = env->spilled_nodes; - int phi_count; + ir_nodeset_t live_nodes; + ir_nodeset_iterator_t iter; + ir_node *node; + int n_phi_values_spilled; + int regpressure; + int phi_spills_needed; + (void) data; DBG((dbg, LEVEL_1, "spilling block %+F\n", block)); + /* construct set of live nodes at end of block */ ir_nodeset_init(&live_nodes); - be_liveness_end_of_block_ir_nodeset(lv, arch_env, cls, block, &live_nodes); + be_liveness_end_of_block(lv, arch_env, cls, block, &live_nodes); + /* remove already spilled nodes from liveset */ foreach_ir_nodeset(&live_nodes, node, iter) { - DBG((dbg, LEVEL_2, "\t%+F is live-in... ", node)); + DBG((dbg, LEVEL_2, "\t%+F is live-end... ", node)); if(bitset_is_set(spilled_nodes, get_irn_idx(node))) { DBG((dbg, LEVEL_2, "but spilled; removing.\n")); + ir_nodeset_remove_iterator(&live_nodes, &iter); } else { DBG((dbg, LEVEL_2, "keeping.\n")); } } + /* walk schedule backwards and spill until register pressure is fine at + * each node */ sched_foreach_reverse(block, node) { if(is_Phi(node)) break; - if(is_Proj(node) || be_is_Keep(node)) { - liveness_transfer(env, node, &live_nodes); - continue; - } - - do_spilling(env, &live_nodes, node); - liveness_transfer(env, node, &live_nodes); + remove_defs(node, &live_nodes); + do_spilling(&live_nodes, node); + add_uses(node, &live_nodes); } - do_spilling(env, &live_nodes, node); - phi_count = 0; - int spilled_phis = 0; + /* until now only the values of some phis have been spilled the phis itself + * are still there and occupy registers, so we need to count them and might + * have to spill some of them. + */ + n_phi_values_spilled = 0; sched_foreach(block, node) { if(!is_Phi(node)) break; - ++phi_count; if(bitset_is_set(spilled_nodes, get_irn_idx(node))) { - ++spilled_phis; + ++n_phi_values_spilled; } } - int regpressure = ir_nodeset_size(&live_nodes) - spilled_phis; - int phi_spills_needed = regpressure - env->n_regs; + + /* calculate how many of the phis need to be spilled */ + regpressure = ir_nodeset_size(&live_nodes) + n_phi_values_spilled; + phi_spills_needed = regpressure - n_regs; + DBG((dbg, LEVEL_3, "Regpressure before phis: %d phispills: %d\n", + regpressure, phi_spills_needed)); + + /* spill as many phis as needed */ + /* TODO: we should really estimate costs of the phi spill as well... + * and preferably spill phis with lower costs... */ sched_foreach(block, node) { if(!is_Phi(node)) break; @@ -329,7 +370,7 @@ void spill_block(ir_node *block, void *data) break; if(bitset_is_set(spilled_nodes, get_irn_idx(node))) { - be_spill_phi(env->spill_env, node); + be_spill_phi(spill_env, node); --phi_spills_needed; } } @@ -338,32 +379,33 @@ void spill_block(ir_node *block, void *data) ir_nodeset_destroy(&live_nodes); } -void be_spill_daemel(be_irg_t *birg, const arch_register_class_t *cls) +void be_spill_daemel(be_irg_t *birg, const arch_register_class_t *new_cls) { - daemel_env_t env; ir_graph *irg = be_get_birg_irg(birg); - int n_regs = cls->n_regs - be_put_ignore_regs(birg, cls, NULL); + n_regs = new_cls->n_regs - be_put_ignore_regs(birg, new_cls, NULL); if(n_regs == 0) return; - be_invalidate_liveness(birg); - be_assure_liveness(birg); + be_liveness_assure_sets(be_assure_liveness(birg)); + + spill_env = be_new_spill_env(birg); + arch_env = be_get_birg_arch_env(birg); + cls = new_cls; + lv = be_get_birg_liveness(birg); + spilled_nodes = bitset_malloc(get_irg_last_idx(irg)); - env.spill_env = be_new_spill_env(birg); - env.n_regs = n_regs; - env.arch_env = be_get_birg_arch_env(birg); - env.cls = cls; - env.lv = be_get_birg_liveness(birg); - env.spilled_nodes = bitset_malloc(get_irg_last_idx(irg)); + DBG((dbg, LEVEL_1, "*** RegClass %s\n", cls->name)); - irg_block_walk_graph(irg, spill_block, NULL, &env); + irg_block_walk_graph(irg, spill_block, NULL, NULL); - bitset_free(env.spilled_nodes); + bitset_free(spilled_nodes); + spilled_nodes = NULL; - be_insert_spills_reloads(env.spill_env); + be_insert_spills_reloads(spill_env); - be_delete_spill_env(env.spill_env); + be_delete_spill_env(spill_env); + spill_env = NULL; } void be_init_daemelspill(void) @@ -373,7 +415,7 @@ void be_init_daemelspill(void) }; be_register_spiller("daemel", &daemel_spiller); - FIRM_DBG_REGISTER(dbg, "ir.be.spilldaemel"); + FIRM_DBG_REGISTER(dbg, "firm.be.spilldaemel"); } -BE_REGISTER_MODULE_CONSTRUCTOR(be_init_doedelspill); +BE_REGISTER_MODULE_CONSTRUCTOR(be_init_daemelspill);