+/*
+ * This file is part of libFirm.
+ * Copyright (C) 2012 University of Karlsruhe.
+ */
+
/**
- * Author: Daniel Grund
- * Date: 12.04.2005
- * Copyright: (c) Universitaet Karlsruhe
- * Licence: This file protected by GPL - GNU GENERAL PUBLIC LICENSE.
+ * @file
+ * @brief Copy minimization driver.
+ * @author Daniel Grund
+ * @date 12.04.2005
+ *
+ * Main file for the optimization reducing the copies needed for:
+ * - Phi coalescing
+ * - Register-constrained nodes
+ * - Two-address code instructions
*/
-#ifdef HAVE_CONFIG_H
#include "config.h"
-#endif
-#ifdef HAVE_ALLOCA_H
-#include <alloca.h>
-#endif
-#ifdef HAVE_MALLOC_H
-#include <malloc.h>
-#endif
-#include "irprog.h"
+#include "debug.h"
+#include "error.h"
+#include "execfreq_t.h"
+#include "irdump_t.h"
+#include "iredges_t.h"
+#include "irgraph.h"
+#include "irgwalk.h"
#include "irloop_t.h"
-
+#include "irnode.h"
+#include "irprintf.h"
+#include "irprog.h"
+#include "irtools.h"
+#include "pmap.h"
+#include "raw_bitset.h"
+#include "util.h"
#include "xmalloc.h"
-#include "beutil.h"
-#include "bechordal_t.h"
-#include "becopyopt.h"
+
+#include "bearch.h"
+#include "becopyopt_t.h"
#include "becopystat.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 "statev_t.h"
+#include "beutil.h"
+
+#include "lc_opts.h"
+#include "lc_opts_enum.h"
+
+#define DUMP_BEFORE 1
+#define DUMP_AFTER 2
+#define DUMP_APPEL 4
+#define DUMP_ALL 2 * DUMP_APPEL - 1
+
+#define COST_FUNC_FREQ 1
+#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 = CO_IFG_DUMP_COLORS;
+static int do_stats = 0;
+static cost_fct_t cost_func = co_get_costs_exec_freq;
+static int improve = 1;
+
+static const lc_opt_enum_mask_items_t dump_items[] = {
+ { "before", DUMP_BEFORE },
+ { "after", DUMP_AFTER },
+ { "appel", DUMP_APPEL },
+ { "all", DUMP_ALL },
+ { NULL, 0 }
+};
+
+static const lc_opt_enum_mask_items_t style_items[] = {
+ { "color", CO_IFG_DUMP_COLORS },
+ { "labels", CO_IFG_DUMP_LABELS },
+ { "constr", CO_IFG_DUMP_CONSTR },
+ { "shape", CO_IFG_DUMP_SHAPE },
+ { "full", 2 * CO_IFG_DUMP_SHAPE - 1 },
+ { NULL, 0 }
+};
+
+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 },
+ { "one", (opt_funcptr) co_get_costs_all_one },
+ { NULL, NULL }
+};
+
+static lc_opt_enum_mask_var_t dump_var = {
+ &dump_flags, dump_items
+};
+
+static lc_opt_enum_mask_var_t style_var = {
+ &style_flags, style_items
+};
+
+static lc_opt_enum_func_ptr_var_t cost_func_var = {
+ (opt_funcptr*) &cost_func, cost_func_items
+};
+
+static const lc_opt_table_entry_t options[] = {
+ LC_OPT_ENT_ENUM_FUNC_PTR ("cost", "select a cost function", &cost_func_var),
+ LC_OPT_ENT_ENUM_MASK ("dump", "dump ifg before or after copy optimization", &dump_var),
+ LC_OPT_ENT_ENUM_MASK ("style", "dump style for ifg dumping", &style_var),
+ LC_OPT_ENT_BOOL ("stats", "dump statistics after each optimization", &do_stats),
+ LC_OPT_ENT_BOOL ("improve", "run heur1 before if algo can exploit start solutions", &improve),
+ LC_OPT_LAST
+};
+
+static be_module_list_entry_t *copyopts = NULL;
+static const co_algo_info *selected_copyopt = NULL;
+
+void be_register_copyopt(const char *name, co_algo_info *copyopt)
+{
+ if (selected_copyopt == NULL)
+ selected_copyopt = 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");
+ 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);
+ be_add_module_list_opt(co_grp, "algo", "select copy optimization algo",
+ ©opts, (void**) &selected_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 = {
+ void_algo, 0
+ };
+
+ be_register_copyopt("none", ©heur);
+}
+
+#undef QUICK_AND_DIRTY_HACK
+
+/******************************************************************************
+ _____ _
+ / ____| | |
+ | | __ ___ _ __ ___ _ __ __ _| |
+ | | |_ |/ _ \ '_ \ / _ \ '__/ _` | |
+ | |__| | __/ | | | __/ | | (_| | |
+ \_____|\___|_| |_|\___|_| \__,_|_|
+
+ ******************************************************************************/
+
+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)
+{
+ FIRM_DBG_REGISTER(dbg, "ir.be.copyopt");
+
+ copy_opt_t *const co = XMALLOCZ(copy_opt_t);
+ co->cenv = chordal_env;
+ co->irg = chordal_env->irg;
+ co->cls = chordal_env->cls;
+ co->get_costs = get_costs;
+ return co;
+}
+
+void free_copy_opt(copy_opt_t *co)
+{
+ free(co);
+}
+
+/**
+ * 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)
+{
+ arch_register_req_t const *const req = arch_get_irn_register_req(irn);
+ if (arch_register_req_is(req, ignore))
+ return 0;
+
+ if (is_Reg_Phi(irn) || is_Perm_Proj(irn))
+ return 1;
+
+ if (arch_register_req_is(req, should_be_same))
+ return 1;
+
+ return 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;
+ int cost;
+
+ if (is_Phi(root)) {
+ 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;
+}
-static firm_dbg_module_t *dbg = NULL;
+static ir_execfreq_int_factors factors;
-#define is_curr_reg_class(irn) \
- (arch_get_irn_reg_class(get_arch_env(co), \
- irn, -1) == co->chordal_env->cls)
+/**
+ * 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_node *root_bl = get_nodes_block(root);
+ ir_node *copy_bl
+ = is_Phi(root) ? get_Block_cfgpred_block(root_bl, pos) : root_bl;
+ int res = get_block_execfreq_int(&factors, copy_bl);
-#define MIN(a,b) ((a<b)?(a):(b))
-#define MAX(a,b) ((a<b)?(b):(a))
+ /* don't allow values smaller than one. */
+ return res < 1 ? 1 : res;
+}
+/**
+ * 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) pos;
+ return 1;
+}
+
+/******************************************************************************
+ ____ _ _ _ _ _ _____ _
+ / __ \ | | | | | | (_) | / ____| |
+ | | | |_ __ | |_| | | |_ __ _| |_ ___ | (___ | |_ ___ _ __ __ _ __ _ ___
+ | | | | '_ \| __| | | | '_ \| | __/ __| \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
+ | |__| | |_) | |_| |__| | | | | | |_\__ \ ____) | || (_) | | | (_| | (_| | __/
+ \____/| .__/ \__|\____/|_| |_|_|\__|___/ |_____/ \__\___/|_| \__,_|\__, |\___|
+ | | __/ |
+ |_| |___/
+ ******************************************************************************/
/**
* 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) {
- be_chordal_env_t *chordal_env = ou->co->chordal_env;
+static int ou_max_ind_set_costs(unit_t *const ou, be_lv_t const *const lv)
+{
ir_node **safe, **unsafe;
int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
bitset_t *curr;
- int max, pos, curr_weight, best_weight = 0;
+ 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.
* unsafe: node has an interference
*/
- safe = alloca((ou->node_count-1) * sizeof(*safe));
- safe_costs = 0;
- safe_count = 0;
- unsafe = alloca((ou->node_count-1) * sizeof(*unsafe));
- unsafe_costs = alloca((ou->node_count-1) * sizeof(*unsafe_costs));
+ safe = ALLOCAN(ir_node*, ou->node_count - 1);
+ safe_costs = 0;
+ safe_count = 0;
+ unsafe = ALLOCAN(ir_node*, ou->node_count - 1);
+ unsafe_costs = ALLOCAN(int, ou->node_count - 1);
unsafe_count = 0;
- for(i=1; i<ou->node_count; ++i) {
+ for (i=1; i<ou->node_count; ++i) {
int is_safe = 1;
- for(o=1; o<ou->node_count; ++o) {
+ for (o=1; o<ou->node_count; ++o) {
if (i==o)
continue;
- if (nodes_interfere(chordal_env, ou->nodes[i], ou->nodes[o])) {
+ if (be_values_interfere(lv, ou->nodes[i], ou->nodes[o])) {
unsafe_costs[unsafe_count] = ou->costs[i];
unsafe[unsafe_count] = ou->nodes[i];
++unsafe_count;
/* 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; i<unsafe_count; ++i) {
bitset_set(best, i);
/* check if it is a stable set */
for (o=bitset_next_set(best, 0); o!=-1 && o<i; o=bitset_next_set(best, o+1))
- if (nodes_interfere(chordal_env, unsafe[i], unsafe[o])) {
+ if (be_values_interfere(lv, unsafe[i], unsafe[o])) {
bitset_clear(best, i); /* clear the bit and try next one */
break;
}
/* Exact Algorithm: Brute force */
curr = bitset_alloca(unsafe_count);
bitset_set_all(curr);
- while ((max = bitset_popcnt(curr)) != 0) {
+ while (bitset_popcount(curr) != 0) {
/* check if curr is a stable set */
for (i=bitset_next_set(curr, 0); i!=-1; i=bitset_next_set(curr, i+1))
for (o=bitset_next_set(curr, i+1); o!=-1; o=bitset_next_set(curr, o+1)) /* !!!!! difference to qnode_max_ind_set(): NOT (curr, i) */
- if (nodes_interfere(chordal_env, unsafe[i], unsafe[o]))
+ if (be_values_interfere(lv, unsafe[i], unsafe[o]))
goto no_stable_set;
/* if we arrive here, we have a stable set */
- /* compute the weigth of the stable set*/
+ /* compute the weight of the stable set*/
curr_weight = 0;
bitset_foreach(curr, pos)
curr_weight += unsafe_costs[pos];
return safe_costs+best_weight;
}
-/**
- * Builds an optimization unit for a given optimizable irn (root).
- * This opt-unit is inserted in the main structure co.
- * If an arg of root itself is optimizable process this arg before with a
- * recursive call. For handling this situation and loops co->root is used
- * to remember all roots.
- */
-static void co_append_unit(copy_opt_t *co, ir_node *root) {
- int i, arity;
+static void co_collect_units(ir_node *irn, void *env)
+{
+ const arch_register_req_t *req;
+ copy_opt_t *co = (copy_opt_t*)env;
unit_t *unit;
- struct list_head *tmp;
- DBG((dbg, LEVEL_1, "\t Root: %n %N\n", root, root));
- /* check if we encountered this root earlier */
- if (pset_find_ptr(co->roots, root))
+ if (get_irn_mode(irn) == mode_T)
+ return;
+ req = arch_get_irn_register_req(irn);
+ if (req->cls != co->cls)
+ return;
+ if (!co_is_optimizable_root(irn))
return;
- pset_insert_ptr(co->roots, root);
-
- assert(is_curr_reg_class(root) && "node is in wrong register class!");
- /* init unit */
- arity = get_irn_arity(root);
- unit = xcalloc(1, sizeof(*unit));
- unit->co = co;
- unit->nodes = xmalloc((arity+1) * sizeof(*unit->nodes));
- unit->costs = xmalloc((arity+1) * sizeof(*unit->costs));
+ /* Init a new unit */
+ unit = XMALLOCZ(unit_t);
unit->node_count = 1;
- unit->nodes[0] = root;
INIT_LIST_HEAD(&unit->queue);
- /* check all args */
- if (is_Phi(root) && is_firm_be_mode(get_irn_mode(root))) {
+ be_lv_t *const lv = be_get_irg_liveness(co->irg);
+ /* Phi with some/all of its arguments */
+ if (is_Reg_Phi(irn)) {
+ int i, arity;
+
+ /* init */
+ arity = get_irn_arity(irn);
+ unit->nodes = XMALLOCN(ir_node*, arity + 1);
+ unit->costs = XMALLOCN(int, arity + 1);
+ unit->nodes[0] = irn;
+
+ /* fill */
for (i=0; i<arity; ++i) {
- int o, arg_pos = 0;
- ir_node *arg = get_irn_n(root, i);
+ int o, arg_pos;
+ ir_node *arg = get_irn_n(irn, i);
- assert(is_curr_reg_class(arg) && "Argument not in same register class.");
- if (arg == root)
+ assert(arch_get_irn_reg_class(arg) == co->cls && "Argument not in same register class.");
+ if (arg == irn)
continue;
- if (nodes_interfere(co->chordal_env, root, arg)) {
- unit->inevitable_costs += co->get_costs(root, arg, i);
+ if (be_values_interfere(lv, irn, arg)) {
+ unit->inevitable_costs += co->get_costs(irn, i);
continue;
}
/* Else insert the argument of the phi to the members of this ou */
- DBG((dbg, LEVEL_1, "\t Member: %n %N\n", arg, arg));
+ DBG((dbg, LEVEL_1, "\t Member: %+F\n", arg));
+
+ if (arch_irn_is_ignore(arg))
+ continue;
/* Check if arg has occurred at a prior position in the arg/list */
- for (o=0; o<unit->node_count; ++o)
+ arg_pos = 0;
+ for (o=1; o<unit->node_count; ++o) {
if (unit->nodes[o] == arg) {
arg_pos = o;
break;
}
+ }
if (!arg_pos) { /* a new argument */
/* insert node, set costs */
unit->nodes[unit->node_count] = arg;
- unit->costs[unit->node_count] = co->get_costs(root, arg, i);
+ unit->costs[unit->node_count] = co->get_costs(irn, i);
unit->node_count++;
- } else { /* arg has occured before in same phi */
+ } else { /* arg has occurred before in same phi */
/* increase costs for existing arg */
- unit->costs[arg_pos] += co->get_costs(root, arg, i);
+ unit->costs[arg_pos] += co->get_costs(irn, i);
}
}
- 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_Copy(get_arch_env(co), root)) {
- assert(!nodes_interfere(co->chordal_env, root, get_Copy_src(root)));
- unit->nodes[1] = get_Copy_src(root);
- unit->costs[1] = co->get_costs(root, unit->nodes[1], -1);
+ 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(!be_values_interfere(lv, irn, get_Perm_src(irn)));
+ unit->nodes = XMALLOCN(ir_node*, 2);
+ unit->costs = XMALLOCN(int, 2);
unit->node_count = 2;
- unit->nodes = xrealloc(unit->nodes, 2 * sizeof(*unit->nodes));
- unit->costs = xrealloc(unit->costs, 2 * sizeof(*unit->costs));
- } else
- assert(0 && "This is not an optimizable node!");
- /* TODO add ou's for 2-addr-code instructions */
+ unit->nodes[0] = irn;
+ unit->nodes[1] = get_Perm_src(irn);
+ unit->costs[1] = co->get_costs(irn, -1);
+ } else if (arch_register_req_is(req, should_be_same)) {
+ /* Src == Tgt of a 2-addr-code instruction */
+ 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_ignore(o))
+ continue;
+ if (be_values_interfere(lv, irn, o))
+ continue;
+ ++count;
+ }
+ }
- /* Determine the maximum costs this unit can cause: all_nodes_cost */
- for(i=1; i<unit->node_count; ++i) {
- unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
- unit->all_nodes_costs += unit->costs[i];
+ if (count != 0) {
+ int k = 0;
+ ++count;
+ 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_ignore(o) &&
+ !be_values_interfere(lv, irn, o)) {
+ unit->nodes[k] = o;
+ unit->costs[k] = co->get_costs(irn, -1);
+ ++k;
+ }
+ }
+ }
+ }
+ } else {
+ assert(0 && "This is not an optimizable node!");
}
- /* Determine the minimal costs this unit will cause: min_nodes_costs */
- unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit);
+ /* Insert the new unit at a position according to its costs */
+ if (unit->node_count > 1) {
+ int i;
+ struct list_head *tmp;
- /* Insert the new ou according to its sort_key */
- tmp = &co->units;
- while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
- tmp = tmp->next;
- list_add(&unit->units, tmp);
+ /* Determine the maximum costs this unit can cause: all_nodes_cost */
+ for (i=1; i<unit->node_count; ++i) {
+ unit->sort_key = MAX(unit->sort_key, unit->costs[i]);
+ unit->all_nodes_costs += unit->costs[i];
+ }
+
+ /* Determine the minimal costs this unit will cause: min_nodes_costs */
+ unit->min_nodes_costs += unit->all_nodes_costs - ou_max_ind_set_costs(unit, lv);
+ /* Insert the new ou according to its sort_key */
+ tmp = &co->units;
+ while (tmp->next != &co->units && list_entry_units(tmp->next)->sort_key > unit->sort_key)
+ tmp = tmp->next;
+ list_add(&unit->units, tmp);
+ } else {
+ free(unit);
+ }
}
-static void co_collect_in_block(ir_node *block, void *env) {
- copy_opt_t *co = env;
- struct list_head *head = get_block_border_head(co->chordal_env, block);
- border_t *curr;
+#ifdef QUICK_AND_DIRTY_HACK
- list_for_each_entry_reverse(border_t, curr, head, list)
- if (curr->is_def && curr->is_real && is_optimizable(get_arch_env(co), curr->irn))
- co_append_unit(co, curr->irn);
-}
+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;
+ arch_register_req_t req;
+
+ /* Units with constraints come first */
+ u1_has_constr = 0;
+ for (i=0; i<u1->node_count; ++i) {
+ arch_get_irn_register_req(&req, u1->nodes[i]);
+ if (arch_register_req_is(&req, limited)) {
+ u1_has_constr = 1;
+ break;
+ }
+ }
+
+ u2_has_constr = 0;
+ for (i=0; i<u2->node_count; ++i) {
+ arch_get_irn_register_req(&req, u2->nodes[i]);
+ if (arch_register_req_is(&req, limited)) {
+ u2_has_constr = 1;
+ break;
+ }
+ }
+
+ if (u1_has_constr != u2_has_constr)
+ return u2_has_constr - u1_has_constr;
+
+ /* After all, the sort key decides. Greater keys come first. */
+ return u2->sort_key - u1->sort_key;
-static void co_collect_units(copy_opt_t *co) {
- DBG((dbg, LEVEL_1, "\tCollecting optimization units\n"));
- co->roots = pset_new_ptr(64);
- dom_tree_walk_irg(get_irg(co), co_collect_in_block, NULL, co);
- del_pset(co->roots);
}
-copy_opt_t *new_copy_opt(be_chordal_env_t *chordal_env, int (*get_costs)(ir_node*, ir_node*, int)) {
- const char *s1, *s2, *s3;
- int len;
- copy_opt_t *co;
+/**
+ * Sort the ou's according to constraints and their sort_key
+ */
+static void co_sort_units(copy_opt_t *co)
+{
+ int i, count = 0, costs;
+ unit_t **ous;
- dbg = firm_dbg_register("ir.be.copyopt");
- firm_dbg_set_mask(dbg, DEBUG_LVL_CO);
+ /* get the number of ous, remove them form the list and fill the array */
+ list_for_each_entry(unit_t, ou, &co->units, units)
+ count++;
+ ous = ALLOCAN(unit_t, count);
- co = xcalloc(1, sizeof(*co));
- co->chordal_env = chordal_env;
- co->get_costs = get_costs;
+ costs = co_get_max_copy_costs(co);
- s1 = get_irp_prog_name();
- s2 = get_entity_name(get_irg_entity(get_irg(co)));
- s3 = chordal_env->cls->name;
- len = strlen(s1) + strlen(s2) + strlen(s3) + 5;
- co->name = xmalloc(len);
- snprintf(co->name, len, "%s__%s__%s", s1, s2, s3);
- if (!strcmp(co->name, DEBUG_IRG))
- firm_dbg_set_mask(dbg, DEBUG_IRG_LVL_CO);
- else
- firm_dbg_set_mask(dbg, DEBUG_LVL_CO);
+ i = 0;
+ list_for_each_entry(unit_t, ou, &co->units, units)
+ ous[i++] = ou;
INIT_LIST_HEAD(&co->units);
- co_collect_units(co);
- return co;
-}
-void free_copy_opt(copy_opt_t *co) {
- unit_t *curr, *tmp;
- xfree(co->name);
- list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
- xfree(curr->nodes);
- xfree(curr);
- }
-}
+ assert(count == i && list_empty(&co->units));
-int is_optimizable_arg(const copy_opt_t *co, ir_node *irn) {
- int i, max;
- for(i=0, max=get_irn_n_outs(irn); i<max; ++i) {
- ir_node *n = get_irn_out(irn, i);
- if (((is_Phi(n) && is_firm_be_mode(get_irn_mode(n))) ||
- is_Perm(get_arch_env(co), n)) && (irn == n || !nodes_interfere(co->chordal_env, irn, n)))
- return 1;
- }
- return 0;
+ for (i=0; i<count; ++i)
+ ir_printf("%+F\n", ous[i]->nodes[0]);
+
+ qsort(ous, count, sizeof(*ous), compare_ous);
+
+ ir_printf("\n\n");
+ for (i=0; i<count; ++i)
+ ir_printf("%+F\n", ous[i]->nodes[0]);
+
+ /* reinsert into list in correct order */
+ for (i=0; i<count; ++i)
+ list_add_tail(&ous[i]->units, &co->units);
+
+ assert(costs == co_get_max_copy_costs(co));
}
+#endif
-int get_costs_loop_depth(ir_node *root, ir_node* arg, int pos) {
- int cost = 0;
- ir_loop *loop;
- ir_node *root_block = get_nodes_block(root);
+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);
+#ifdef QUICK_AND_DIRTY_HACK
+ co_sort_units(co);
+#endif
+}
- assert(pos==-1 || is_Phi(root));
- if (pos == -1) {
- /* a perm places the copy in the same block as it resides */
- loop = get_irn_loop(root_block);
- } else {
- /* for phis the copies are placed in the corresponding pred-block */
- loop = get_irn_loop(get_Block_cfgpred_block(root_block, pos));
- }
- if (loop) {
- int d = get_loop_depth(loop);
- cost = d*d;
+void co_free_ou_structure(copy_opt_t *co)
+{
+ ASSERT_OU_AVAIL(co);
+ list_for_each_entry_safe(unit_t, curr, tmp, &co->units, units) {
+ xfree(curr->nodes);
+ xfree(curr->costs);
+ xfree(curr);
}
- return cost+1;
+ co->units.next = NULL;
}
-int get_costs_all_one(ir_node *root, ir_node* arg, int pos) {
- return 1;
-}
+/* 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;
+
+ ASSERT_OU_AVAIL(co);
list_for_each_entry(unit_t, curr, &co->units, units) {
res += curr->inevitable_costs;
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;
+
+ ASSERT_OU_AVAIL(co);
list_for_each_entry(unit_t, curr, &co->units, units)
res += curr->inevitable_costs;
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;
+
+ 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];
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;
+
+ ASSERT_OU_AVAIL(co);
+
list_for_each_entry(unit_t, curr, &co->units, units)
res += curr->inevitable_costs + curr->min_nodes_costs;
return res;
}
+
+void co_complete_stats(const copy_opt_t *co, co_complete_stats_t *stat)
+{
+ bitset_t *seen = bitset_malloc(get_irg_last_idx(co->irg));
+
+ memset(stat, 0, sizeof(stat[0]));
+
+ /* count affinity edges. */
+ be_lv_t *const lv = be_get_irg_liveness(co->irg);
+ co_gs_foreach_aff_node(co, an) {
+ stat->aff_nodes += 1;
+ bitset_set(seen, get_irn_idx(an->irn));
+ co_gs_foreach_neighb(an, neigh) {
+ if (!bitset_is_set(seen, get_irn_idx(neigh->irn))) {
+ stat->aff_edges += 1;
+ stat->max_costs += neigh->costs;
+
+ if (get_irn_col(an->irn) != get_irn_col(neigh->irn)) {
+ stat->costs += neigh->costs;
+ stat->unsatisfied_edges += 1;
+ }
+
+ if (be_values_interfere(lv, an->irn, neigh->irn)) {
+ stat->aff_int += 1;
+ stat->inevit_costs += neigh->costs;
+ }
+
+ }
+ }
+ }
+
+ bitset_free(seen);
+}
+
+/******************************************************************************
+ _____ _ _____ _
+ / ____| | | / ____| |
+ | | __ _ __ __ _ _ __ | |__ | (___ | |_ ___ _ __ __ _ __ _ ___
+ | | |_ | '__/ _` | '_ \| '_ \ \___ \| __/ _ \| '__/ _` |/ _` |/ _ \
+ | |__| | | | (_| | |_) | | | | ____) | || (_) | | | (_| | (_| | __/
+ \_____|_| \__,_| .__/|_| |_| |_____/ \__\___/|_| \__,_|\__, |\___|
+ | | __/ |
+ |_| |___/
+ ******************************************************************************/
+
+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)
+{
+ affinity_node_t new_node, *node;
+ neighb_t *nbr;
+ int allocnew = 1;
+
+ new_node.irn = n1;
+ new_node.neighbours = NULL;
+ node = set_insert(affinity_node_t, co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
+
+ for (nbr = node->neighbours; nbr; nbr = nbr->next)
+ if (nbr->irn == n2) {
+ allocnew = 0;
+ break;
+ }
+
+ /* if we did not find n2 in n1's neighbourhood insert it */
+ if (allocnew) {
+ nbr = OALLOC(&co->obst, neighb_t);
+ nbr->irn = n2;
+ nbr->costs = 0;
+ nbr->next = node->neighbours;
+
+ node->neighbours = nbr;
+ }
+
+ /* now nbr points to n1's neighbour-entry of n2 */
+ nbr->costs += costs;
+}
+
+static inline void add_edges(copy_opt_t *co, ir_node *n1, ir_node *n2, int costs)
+{
+ be_lv_t *const lv = be_get_irg_liveness(co->irg);
+ if (!be_values_interfere(lv, 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;
+ copy_opt_t *co = (copy_opt_t*)env;
+ int pos, max;
+
+ if (get_irn_mode(irn) == mode_T)
+ return;
+ req = arch_get_irn_register_req(irn);
+ if (req->cls != co->cls || arch_register_req_is(req, ignore))
+ return;
+
+ 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(irn, pos));
+ }
+ } else if (is_Perm_Proj(irn)) { /* Perms */
+ ir_node *arg = get_Perm_src(irn);
+ add_edges(co, irn, arg, co->get_costs(irn, -1));
+ } else if (arch_register_req_is(req, should_be_same)) {
+ 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_ignore(other))
+ add_edges(co, irn, other, co->get_costs(irn, -1));
+ }
+ }
+ }
+}
+
+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)
+{
+ ASSERT_GS_AVAIL(co);
+
+ del_set(co->nodes);
+ obstack_free(&co->obst, NULL);
+ co->nodes = NULL;
+}
+
+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(affinity_node_t, co->nodes, &new_node, sizeof(new_node), hash_irn(new_node.irn));
+ return n && n->neighbours;
+}
+
+static int co_dump_appel_disjoint_constraints(const copy_opt_t *co, ir_node *a, ir_node *b)
+{
+ ir_node *nodes[] = { a, b };
+ bitset_t *constr[] = { NULL, NULL };
+ int j;
+
+ constr[0] = bitset_alloca(co->cls->n_regs);
+ constr[1] = bitset_alloca(co->cls->n_regs);
+
+ for (j = 0; j < 2; ++j) {
+ 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]);
+
+ }
+
+ return !bitset_intersect(constr[0], constr[1]);
+}
+
+/**
+ * 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);
+
+ neighbours_iter_t nit;
+ int n, n_regs;
+ unsigned i;
+
+ n_regs = 0;
+ for (i = 0; i < co->cls->n_regs; ++i) {
+ const arch_register_t *reg = &co->cls->regs[i];
+ if (rbitset_is_set(birg->allocatable_regs, reg->global_index)) {
+ color_map[i] = n_regs++;
+ } else {
+ color_map[i] = -1;
+ }
+ }
+
+ /*
+ * n contains the first node number.
+ * the values below n are the pre-colored register nodes
+ */
+
+ n = n_regs;
+ be_ifg_foreach_node(ifg, irn) {
+ if (arch_irn_is_ignore(irn))
+ continue;
+ node_map[get_irn_idx(irn)] = n++;
+ }
+
+ fprintf(f, "%d %d\n", n, n_regs);
+
+ be_ifg_foreach_node(ifg, irn) {
+ arch_register_req_t const *const req = arch_get_irn_register_req(irn);
+ if (arch_register_req_is(req, ignore))
+ continue;
+
+ int idx = node_map[get_irn_idx(irn)];
+ affinity_node_t *a = get_affinity_info(co, irn);
+
+ 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) {
+ 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)
+ fprintf(f, "%d %d -1\n", idx, adj_idx);
+ }
+ }
+
+ if (a) {
+ 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)
+ fprintf(f, "%d %d %d\n", idx, n_idx, (int) n->costs);
+ }
+ }
+ }
+ }
+
+ xfree(node_map);
+}
+
+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(cup_name);
+ tu_name = XMALLOCN(char, n + 1);
+ strcpy(tu_name, cup_name);
+ for (i = 0; i < n; ++i)
+ if (tu_name[i] == '.')
+ tu_name[i] = '_';
+
+
+ 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) {
+ panic("Couldn't open '%s' for writing.", buf);
+ }
+
+ return result;
+}
+
+void co_driver(be_chordal_env_t *cenv)
+{
+ ir_timer_t *timer = ir_timer_new();
+ co_complete_stats_t before, after;
+ copy_opt_t *co;
+ int was_optimal = 0;
+
+ assert(selected_copyopt);
+
+ /* skip copymin if algo is 'none' */
+ if (selected_copyopt->copyopt == void_algo)
+ return;
+
+ be_assure_live_chk(cenv->irg);
+
+ co = new_copy_opt(cenv, cost_func);
+ co_build_ou_structure(co);
+ co_build_graph_structure(co);
+
+ co_complete_stats(co, &before);
+
+ stat_ev_ull("co_aff_nodes", before.aff_nodes);
+ stat_ev_ull("co_aff_edges", before.aff_edges);
+ stat_ev_ull("co_max_costs", before.max_costs);
+ stat_ev_ull("co_inevit_costs", before.inevit_costs);
+ stat_ev_ull("co_aff_int", before.aff_int);
+
+ stat_ev_ull("co_init_costs", before.costs);
+ stat_ev_ull("co_init_unsat", before.unsatisfied_edges);
+
+ if (dump_flags & DUMP_BEFORE) {
+ 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);
+ }
+
+ /* if the algo can improve results, provide an initial solution with heur1 */
+ if (improve && selected_copyopt->can_improve_existing) {
+ co_complete_stats_t stats;
+
+ /* produce a heuristic solution */
+ co_solve_heuristic(co);
+
+ /* do the stats and provide the current costs */
+ co_complete_stats(co, &stats);
+ stat_ev_ull("co_prepare_costs", stats.costs);
+ }
+
+ /* perform actual copy minimization */
+ ir_timer_reset_and_start(timer);
+ was_optimal = selected_copyopt->copyopt(co);
+ ir_timer_stop(timer);
+
+ stat_ev_dbl("co_time", ir_timer_elapsed_msec(timer));
+ stat_ev_ull("co_optimal", was_optimal);
+ ir_timer_free(timer);
+
+ if (dump_flags & DUMP_AFTER) {
+ 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) {
+ 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 %10llu%10llu%10llu", cenv->cls->name, after.max_costs, before.costs, after.inevit_costs);
+
+ if (optimizable_costs > 0)
+ printf("%10llu %5.2f\n", after.costs, (evitable * 100.0) / optimizable_costs);
+ else
+ 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, "# %llu %llu\n", after.costs, after.unsatisfied_edges);
+ co_dump_appel_graph(co, f);
+ fclose(f);
+ }
+
+ stat_ev_ull("co_after_costs", after.costs);
+ stat_ev_ull("co_after_unsat", after.unsatisfied_edges);
+
+ co_free_graph_structure(co);
+ co_free_ou_structure(co);
+ free_copy_opt(co);
+}
projects / libfirm / blobdiff