+/*
+ * This file is part of libFirm.
+ * Copyright (C) 2012 University of Karlsruhe.
+ */
+
/**
+ * @file
+ * @brief First simple copy minimization heuristics.
+ * @author Daniel Grund
+ * @date 12.04.2005
+ *
* Heuristic for minimizing copies using a queue which holds 'qnodes' not yet
* examined. A qnode has a 'target color', nodes out of the opt unit and
- * a 'conflict graph'. A 'max indep set' is determined form these. We try to
- * color this mis using a color-exchanging mechanism. Occuring conflicts are
- * modeled with 'conflict edges' and the qnode is reinserted in the queue. The
- * first qnode colored without conflicts is the best one.
- * @author Daniel Grund
- * @date 12.04.2005
+ * a 'conflict graph'. 'Conflict graph' = "Interference graph' + 'conflict edges'
+ * A 'max indep set' is determined from these. We try to color this mis using a
+ * color-exchanging mechanism. Occuring conflicts are modeled with 'conflict edges'
+ * and the qnode is reinserted in the queue. The first qnode colored without
+ * conflicts is the best one.
*/
+#include "config.h"
-#include "becopyopt.h"
+#include "debug.h"
+#include "bitset.h"
+#include "raw_bitset.h"
+#include "xmalloc.h"
+
+#include "becopyopt_t.h"
#include "becopystat.h"
+#include "beintlive_t.h"
+#include "beirg.h"
+#include "bemodule.h"
+
+DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
-#define DEBUG_LVL SET_LEVEL_1
-static firm_dbg_module_t *dbg = NULL;
+/** Defines an invalid register index. */
+#define NO_COLOR (-1)
-#define SLOTS_PINNED_GLOBAL 256
+#define SEARCH_FREE_COLORS
+
+#define SLOTS_PINNED_GLOBAL 64
#define SLOTS_CONFLICTS 8
#define SLOTS_CHANGED_NODES 32
-#define MIN(a,b) ((a<b)?a:b)
#define list_entry_queue(lh) list_entry(lh, qnode_t, queue)
-#define HASH_CONFLICT(c) (HASH_PTR(c.n1) ^ HASH_PTR(c.n2))
+#define HASH_CONFLICT(c) (hash_irn(c.n1) ^ hash_irn(c.n2))
/**
* Modeling additional conflicts between nodes. NOT live range interference
*/
-typedef struct _conflict_t {
+typedef struct conflict_t {
const ir_node *n1, *n2;
} conflict_t;
* If an irn is changed, the changes first get stored in a node_stat_t,
* to allow undo of changes (=drop new data) in case of conflicts.
*/
-typedef struct _node_stat_t {
- const ir_node *irn;
- int new_color;
- int pinned_local :1;
+typedef struct node_stat_t {
+ ir_node *irn;
+ int new_color;
+ unsigned pinned_local :1;
} node_stat_t;
/**
* Represents a node in the optimization queue.
*/
-typedef struct _qnode_t {
- struct list_head queue; /**< chaining of unit_t->queue */
- int color; /**< target color */
- set *conflicts; /**< contains conflict_t's. All internal conflicts */
- int mis_size; /**< number of nodes in the mis. */
- const ir_node **mis; /**< the nodes of unit_t->nodes[] beeing part of the max idependent set */
- set *changed_nodes; /**< contains node_stat_t's. */
+typedef struct qnode_t {
+ struct list_head queue; /**< chaining of unit_t->queue */
+ int color; /**< target color */
+ set *conflicts; /**< contains conflict_t's. All internal conflicts */
+ int mis_costs; /**< costs of nodes/copies in the mis. */
+ int mis_size; /**< size of the array below */
+ ir_node **mis; /**< the nodes of unit_t->nodes[] being part of the max independent set */
+ set *changed_nodes; /**< contains node_stat_t's. */
} qnode_t;
-pset *pinned_global; /**< optimized nodes should not be altered any more */
+static pset *pinned_global; /**< optimized nodes should not be altered any more */
+
+static int set_cmp_conflict_t(const void *x, const void *y, size_t size)
+{
+ const conflict_t *xx = (const conflict_t*)x;
+ const conflict_t *yy = (const conflict_t*)y;
+ (void) size;
-static int set_cmp_conflict_t(const void *x, const void *y, size_t size) {
- const conflict_t *xx = x;
- const conflict_t *yy = y;
- return ! (xx->n1 == yy->n1 && xx->n2 == yy->n2);
+ return xx->n1 != yy->n1 || xx->n2 != yy->n2;
}
/**
* If a local pinned conflict occurs, a new edge in the conflict graph is added.
* The next maximum independent set build, will regard it.
*/
-static INLINE void qnode_add_conflict(const qnode_t *qn, const ir_node *n1, const ir_node *n2) {
+static inline void qnode_add_conflict(const qnode_t *qn, const ir_node *n1, const ir_node *n2)
+{
conflict_t c;
- DBG((dbg, LEVEL_4, "\t %n -- %n\n", n1, n2));
+ DBG((dbg, LEVEL_4, "\t %+F -- %+F\n", n1, n2));
- if ((int)n1 < (int)n2) {
+ if (get_irn_idx(n1) < get_irn_idx(n2)) {
c.n1 = n1;
c.n2 = n2;
} else {
c.n1 = n2;
c.n2 = n1;
}
- set_insert(qn->conflicts, &c, sizeof(c), HASH_CONFLICT(c));
+ (void)set_insert(conflict_t, qn->conflicts, &c, sizeof(c), HASH_CONFLICT(c));
}
/**
* Checks if two nodes are in a conflict.
*/
-static INLINE int qnode_are_conflicting(const qnode_t *qn, const ir_node *n1, const ir_node *n2) {
+static inline int qnode_are_conflicting(const qnode_t *qn, const ir_node *n1, const ir_node *n2)
+{
conflict_t c;
/* search for live range interference */
- if (n1!=n2 && values_interfere(n1, n2))
- return 1;
+ if (n1 != n2) {
+ be_lv_t *const lv = be_get_irg_liveness(get_irn_irg(n1));
+ if (be_values_interfere(lv, n1, n2))
+ return 1;
+ }
/* search for recoloring conflicts */
- if ((int)n1 < (int)n2) {
+ if (get_irn_idx(n1) < get_irn_idx(n2)) {
c.n1 = n1;
c.n2 = n2;
} else {
c.n1 = n2;
c.n2 = n1;
}
- return (int) set_find(qn->conflicts, &c, sizeof(c), HASH_CONFLICT(c));
+ return set_find(conflict_t, qn->conflicts, &c, sizeof(c), HASH_CONFLICT(c)) != 0;
}
-static int set_cmp_node_stat_t(const void *x, const void *y, size_t size) {
- return ((node_stat_t *)x)->irn != ((node_stat_t *)y)->irn;
+static int set_cmp_node_stat_t(const void *x, const void *y, size_t size)
+{
+ (void) size;
+ return ((const node_stat_t*)x)->irn != ((const node_stat_t*)y)->irn;
}
/**
* Finds a node status entry of a node if existent. Otherwise return NULL
*/
-static INLINE node_stat_t *qnode_find_node(const qnode_t *qn, const ir_node *irn) {
+static inline const node_stat_t *qnode_find_node(const qnode_t *qn, ir_node *irn)
+{
node_stat_t find;
find.irn = irn;
- return set_find(qn->changed_nodes, &find, sizeof(find), HASH_PTR(irn));
+ return set_find(node_stat_t, qn->changed_nodes, &find, sizeof(find), hash_irn(irn));
}
/**
* Finds a node status entry of a node if existent. Otherwise it will return
* an initialized new entry for this node.
*/
-static INLINE node_stat_t *qnode_find_or_insert_node(const qnode_t *qn, const ir_node *irn) {
+static inline node_stat_t *qnode_find_or_insert_node(const qnode_t *qn, ir_node *irn)
+{
node_stat_t find;
find.irn = irn;
find.new_color = NO_COLOR;
find.pinned_local = 0;
- return set_insert(qn->changed_nodes, &find, sizeof(find), HASH_PTR(irn));
+ return set_insert(node_stat_t, qn->changed_nodes, &find, sizeof(find), hash_irn(irn));
}
/**
* Returns the virtual color of a node if set before, else returns the real color.
*/
-static INLINE int qnode_get_new_color(const qnode_t *qn, const ir_node *irn) {
- node_stat_t *found = qnode_find_node(qn, irn);
+static inline int qnode_get_new_color(const qnode_t *qn, ir_node *irn)
+{
+ const node_stat_t *found = qnode_find_node(qn, irn);
if (found)
return found->new_color;
else
- return get_irn_color(irn);
+ return get_irn_col(irn);
}
/**
* Sets the virtual color of a node.
*/
-static INLINE void qnode_set_new_color(const qnode_t *qn, const ir_node *irn, int color) {
+static inline void qnode_set_new_color(const qnode_t *qn, ir_node *irn, int color)
+{
node_stat_t *found = qnode_find_or_insert_node(qn, irn);
found->new_color = color;
+ DBG((dbg, LEVEL_3, "\t col(%+F) := %d\n", irn, color));
}
/**
* to the same optimization unit and has been optimized before the current
* processed node.
*/
-static INLINE int qnode_is_pinned_local(const qnode_t *qn, const ir_node *irn) {
- node_stat_t *found = qnode_find_node(qn, irn);
+static inline int qnode_is_pinned_local(const qnode_t *qn, ir_node *irn)
+{
+ const node_stat_t *found = qnode_find_node(qn, irn);
if (found)
return found->pinned_local;
else
* Local-pins a node, so optimizations of further nodes of the same opt unit
* can handle situations in which a color change would undo prior optimizations.
*/
-static INLINE void qnode_pin_local(const qnode_t *qn, const ir_node *irn) {
+static inline void qnode_pin_local(const qnode_t *qn, ir_node *irn)
+{
node_stat_t *found = qnode_find_or_insert_node(qn, irn);
found->pinned_local = 1;
+ if (found->new_color == NO_COLOR)
+ found->new_color = get_irn_col(irn);
}
+
/**
* Possible return values of qnode_color_irn()
*/
#define CHANGE_SAVE NULL
#define CHANGE_IMPOSSIBLE (ir_node *)1
-#define is_conflicting_node(n) (((int)n) > 1)
/**
* Performs virtual re-coloring of node @p n to color @p col. Virtual colors of
* other nodes are changed too, as required to preserve correctness. Function is
* aware of local and global pinning. Recursive.
- * @param irn The node to set the color for
- * @param col The color to set
+ *
+ * If irn == trigger the color @p col must be used. (the first recoloring)
+ * If irn != trigger an arbitrary free color may be used. If no color is free, @p col is used.
+ *
+ * @param irn The node to set the color for
+ * @param col The color to set
* @param trigger The irn that caused the wish to change the color of the irn
- * @return CHANGE_SAVE iff setting the color is possible, with all transiteve effects.
+ * External callers must call with trigger = irn
+ *
+ * @return CHANGE_SAVE iff setting the color is possible, with all transitive effects.
* CHANGE_IMPOSSIBLE iff conflicts with reg-constraintsis occured.
* Else the first conflicting ir_node encountered is returned.
*
- * ASSUMPTION: Assumes that a life range of a single value can't be spilt into
- * several smaller intervals where other values can live in between.
- * This should be true in SSA.
*/
-static const ir_node *qnode_color_irn(const qnode_t *qn, const ir_node *irn, int col, const ir_node *trigger) {
- const ir_node *res;
- struct obstack confl_ob;
- ir_node **confl, *cn;
- int i, irn_col;
-
- DBG((dbg, LEVEL_3, "\t %n \tcaused col(%n) \t%2d --> %2d\n", trigger, irn, qnode_get_new_color(qn, irn), col));
- obstack_init(&confl_ob);
- irn_col = qnode_get_new_color(qn, irn);
-
- if (irn_col == col)
- goto ret_save;
- if (!is_possible_color(irn, col))
- goto ret_imposs;
+static ir_node *qnode_color_irn(qnode_t const *const qn, ir_node *const irn, int const col, ir_node const *const trigger, bitset_t const *const allocatable_regs, be_ifg_t *const ifg)
+{
+ int irn_col = qnode_get_new_color(qn, irn);
+ neighbours_iter_t iter;
+
+ DBG((dbg, LEVEL_3, "\t %+F \tcaused col(%+F) \t%2d --> %2d\n", trigger, irn, irn_col, col));
+
+ /* If the target color is already set do nothing */
+ if (irn_col == col) {
+ DBG((dbg, LEVEL_3, "\t %+F same color\n", irn));
+ return CHANGE_SAVE;
+ }
+
+ /* If the irn is pinned, changing color is impossible */
if (pset_find_ptr(pinned_global, irn) || qnode_is_pinned_local(qn, irn)) {
- res = irn;
- goto ret_confl;
+ DBG((dbg, LEVEL_3, "\t %+F conflicting\n", irn));
+ return irn;
}
- /* get all nodes which would conflict with this change */
- {
- struct obstack q;
- int in, out;
- ir_node *irn_bl;
-
- irn_bl = get_nodes_block(irn);
-
- /* first check for a conflicting node which is 'living in' the irns block */
- {
- ir_node *n;
- pset *live_ins = get_live_in(irn_bl);
- for (n = pset_first(live_ins); n; n = pset_next(live_ins))
- if (is_allocatable_irn(n) && n != trigger && qnode_get_new_color(qn, n) == col && values_interfere(irn, n)) {
- DBG((dbg, LEVEL_4, "\t %n\ttroubles\n", n));
- obstack_ptr_grow(&confl_ob, n);
- pset_break(live_ins);
- break;
- }
- }
+ arch_register_req_t const *const req = arch_get_irn_register_req(irn);
+ arch_register_class_t const *const cls = req->cls;
+#ifdef SEARCH_FREE_COLORS
+ /* If we resolve conflicts (recursive calls) we can use any unused color.
+ * In case of the first call @p col must be used.
+ */
+ if (irn != trigger) {
+ bitset_t *free_cols = bitset_alloca(cls->n_regs);
+ int free_col;
- /* setup the queue of blocks. */
- obstack_init(&q);
- obstack_ptr_grow(&q, irn_bl);
- in = 1;
- out = 0;
-
- /* process the queue. The code below checks for every block dominated
- * by the irns one, and in which the irn is live, if there are
- * conflicting nodes */
- while (out < in) {
- ir_node *curr_bl, *sub_bl;
- int i, max;
-
- curr_bl = ((ir_node **)obstack_base(&q))[out++];
-
- /* Add to the result all nodes in the block, which have
- * the target color and interfere with the irn */
- for (i = 0, max = get_irn_n_outs(curr_bl); i < max; ++i) {
- ir_node *n = get_irn_out(curr_bl, i);
- if (is_allocatable_irn(n) && n != trigger && qnode_get_new_color(qn, n) == col && values_interfere(irn, n)) {
- DBG((dbg, LEVEL_4, "\t %n\ttroubles\n", n));
- obstack_ptr_grow(&confl_ob, n);
- }
- }
+ /* Get all possible colors */
+ bitset_copy(free_cols, allocatable_regs);
- /* If irn lives out check i-dominated blocks where the irn lives in */
- /* Fill the queue */
- if (is_live_out(curr_bl, irn)) {
- dominates_for_each(curr_bl, sub_bl)
- if (is_live_in(sub_bl, irn)) {
- obstack_ptr_grow(&q, sub_bl);
- in++;
- }
- }
+ /* Exclude colors not assignable to the irn */
+ if (arch_register_req_is(req, limited))
+ rbitset_and(free_cols->data, req->limited, free_cols->size);
+
+ /* Exclude the color of the irn, because it must _change_ its color */
+ bitset_clear(free_cols, irn_col);
+
+ /* Exclude all colors used by adjacent nodes */
+ be_ifg_foreach_neighbour(ifg, &iter, irn, curr)
+ bitset_clear(free_cols, qnode_get_new_color(qn, curr));
+
+ free_col = bitset_next_set(free_cols, 0);
+
+ if (free_col != -1) {
+ qnode_set_new_color(qn, irn, free_col);
+ return CHANGE_SAVE;
}
- obstack_free(&q, NULL);
- obstack_ptr_grow(&confl_ob, NULL);
- confl = (ir_node **) obstack_finish(&confl_ob);
}
+#endif /* SEARCH_FREE_COLORS */
- /* process all nodes which would conflict with this change */
- for (i = 0, cn = confl[0]; cn; cn = confl[++i]) {
- const ir_node *sub_res;
+ /* If target color is not allocatable changing color is impossible */
+ if (!arch_reg_is_allocatable(req, arch_register_for_index(cls, col))) {
+ DBG((dbg, LEVEL_3, "\t %+F impossible\n", irn));
+ return CHANGE_IMPOSSIBLE;
+ }
- /* try to color the conflicting node cn with the color of the irn itself */
- sub_res = qnode_color_irn(qn, cn, irn_col, irn);
- if (sub_res != CHANGE_SAVE) {
- res = sub_res;
- goto ret_confl;
+ /*
+ * If we arrive here changing color may be possible, but there may be conflicts.
+ * Try to color all conflicting nodes 'curr' with the color of the irn itself.
+ */
+ be_ifg_foreach_neighbour(ifg, &iter, irn, curr) {
+ DBG((dbg, LEVEL_3, "\t Confl %+F(%d)\n", curr, qnode_get_new_color(qn, curr)));
+ if (qnode_get_new_color(qn, curr) == col && curr != trigger) {
+ ir_node *const sub_res = qnode_color_irn(qn, curr, irn_col, irn, allocatable_regs, ifg);
+ if (sub_res != CHANGE_SAVE) {
+ be_ifg_neighbours_break(&iter);
+ return sub_res;
+ }
}
}
- /* if we arrive here all sub changes can be applied, so it's save to change this irn */
-ret_save:
- DBG((dbg, LEVEL_3, "\t %n save\n", irn));
- obstack_free(&confl_ob, NULL);
+ /*
+ * If we arrive here, all conflicts were resolved.
+ * So it is save to change this irn
+ */
qnode_set_new_color(qn, irn, col);
return CHANGE_SAVE;
-
-ret_imposs:
- DBG((dbg, LEVEL_3, "\t %n impossible\n", irn));
- obstack_free(&confl_ob, NULL);
- return res;
-
-ret_confl:
- DBG((dbg, LEVEL_3, "\t %n conflicting\n", irn));
- obstack_free(&confl_ob, NULL);
- return res;
}
+
/**
* Tries to set the colors for all members of this queue node;
* to the target color qn->color
* @returns 1 iff all members colors could be set
* 0 else
*/
-static int qnode_try_color(const qnode_t *qn) {
+static int qnode_try_color(qnode_t const *const qn, bitset_t const *const allocatable_regs, be_ifg_t *const ifg)
+{
int i;
for (i=0; i<qn->mis_size; ++i) {
- const ir_node *test_node, *confl_node;
+ ir_node *test_node, *confl_node;
test_node = qn->mis[i];
- DBG((dbg, LEVEL_3, "\t Testing %n\n", test_node));
- confl_node = qnode_color_irn(qn, test_node, qn->color, test_node);
+ DBG((dbg, LEVEL_3, "\t Testing %+F\n", test_node));
+ confl_node = qnode_color_irn(qn, test_node, qn->color, test_node, allocatable_regs, ifg);
if (confl_node == CHANGE_SAVE) {
DBG((dbg, LEVEL_3, "\t Save --> pin local\n"));
} else if (confl_node == CHANGE_IMPOSSIBLE) {
DBG((dbg, LEVEL_3, "\t Impossible --> remove from qnode\n"));
qnode_add_conflict(qn, test_node, test_node);
+ return 0;
} else {
if (qnode_is_pinned_local(qn, confl_node)) {
/* changing test_node would change back a node of current ou */
- DBG((dbg, LEVEL_3, "\t Conflicting local --> add conflict\n"));
- qnode_add_conflict(qn, confl_node, test_node);
+ if (confl_node == qn->mis[0]) {
+ /* Adding a conflict edge between testnode and conflnode
+ * would introduce a root -- arg interference.
+ * So remove the arg of the qn */
+ DBG((dbg, LEVEL_3, "\t Conflicting local with phi --> remove from qnode\n"));
+ qnode_add_conflict(qn, test_node, test_node);
+ } else {
+ DBG((dbg, LEVEL_3, "\t Conflicting local --> add conflict\n"));
+ qnode_add_conflict(qn, confl_node, test_node);
+ }
}
if (pset_find_ptr(pinned_global, confl_node)) {
/* changing test_node would change back a node of a prior ou */
DBG((dbg, LEVEL_3, "\t Conflicting global --> remove from qnode\n"));
qnode_add_conflict(qn, test_node, test_node);
}
- }
-
- if (confl_node != CHANGE_SAVE)
return 0;
+ }
}
return 1;
}
/**
- * Determines a maximum independent set with respect to the interference and
- * conflict edges of all nodes in a qnode.
+ * Determines a maximum weighted independent set with respect to
+ * the interference and conflict edges of all nodes in a qnode.
*/
-static INLINE void qnode_max_ind_set(qnode_t *qn, const unit_t *ou) {
- int all_size, curr_size, i, o;
- int *which;
- const ir_node **curr, **all = alloca(ou->node_count * sizeof(*all));
-
- /* all contains all nodes not removed in this qn */
- all_size = 0;
- for (i=0; i<ou->node_count; ++i)
- if (!qnode_are_conflicting(qn, ou->nodes[i], ou->nodes[i]))
- all[all_size++] = ou->nodes[i];
-
- /* which[i] says which element to take out of all[] and put into curr[i] */
- which = alloca(all_size*sizeof(*which));
- for (curr_size=0; curr_size<all_size; ++curr_size)
- which[curr_size] = curr_size;
-
- /* stores the currently examined set */
- curr = alloca(all_size*sizeof(*curr));
-
- while (1) { /* this loop will terminate because at least a single node'll be a max indep. set */
- /* build current set */
- for (i=0; i<curr_size; ++i)
- curr[i] = all[which[all_size-curr_size+i]];
-
- /* check current set */
- for (i=0; i<curr_size; ++i)
- for (o=i+1; o<curr_size; ++o)
- if (qnode_are_conflicting(qn, curr[i], curr[o]))
- goto conflict_found;
-
- /* We had no conflict. This is the max indep. set */
- qn->mis_size = curr_size;
- for (i=0; i<curr_size; ++i)
- qn->mis[i] = curr[i];
- return;
+static inline void qnode_max_ind_set(qnode_t *qn, const unit_t *ou)
+{
+ ir_node **safe, **unsafe;
+ int i, o, safe_count, safe_costs, unsafe_count, *unsafe_costs;
+ bitset_t *curr, *best;
+ int next, 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 = 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) {
+ int is_safe = 1;
+ for (o=1; o<ou->node_count; ++o) {
+ if (qnode_are_conflicting(qn, ou->nodes[i], ou->nodes[o])) {
+ if (i!=o) {
+ unsafe_costs[unsafe_count] = ou->costs[i];
+ unsafe[unsafe_count] = ou->nodes[i];
+ ++unsafe_count;
+ }
+ is_safe = 0;
+ break;
+ }
+ }
+ if (is_safe) {
+ safe_costs += ou->costs[i];
+ safe[safe_count++] = ou->nodes[i];
+ }
+ }
-conflict_found:
- /* We had a conflict. Generate next set */
- if (which[all_size-curr_size+1] == all_size-curr_size+1) {
- curr_size--;
- for (i=0; i<curr_size; ++i)
- which[all_size-curr_size+i] = i;
- } else {
- int redo = 1;
- while (redo) {
- int pos = all_size;
- do {
- pos--;
- } while (!(which[pos] = (which[pos]+1) % all_size));
-
- for (i=pos+1; i<all_size; ++i)
- which[i] = MIN(which[i-1]+1, all_size-1);
-
- redo = 0;
- for (i=all_size-curr_size; i<all_size-1; ++i)
- if (which[i]>=which[i+1]) {
- redo = 1;
- break;
- }
+
+
+ /* now compute the best set out of the unsafe nodes*/
+ best = bitset_alloca(unsafe_count);
+
+ if (unsafe_count > MIS_HEUR_TRIGGER) {
+ /* 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 (qnode_are_conflicting(qn, unsafe[i], unsafe[o])) {
+ bitset_clear(best, i); /* clear the bit and try next one */
+ break;
+ }
+ }
+ /* compute the weight */
+ bitset_foreach(best, pos)
+ best_weight += unsafe_costs[pos];
+ } else {
+ /* Exact Algorithm: Brute force */
+ curr = bitset_alloca(unsafe_count);
+ bitset_set_all(curr);
+ while (!bitset_is_empty(curr)) {
+ /* 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); o!=-1; o=bitset_next_set(curr, o+1)) /* !!!!! difference to ou_max_ind_set_costs(): NOT (curr, i+1) */
+ if (qnode_are_conflicting(qn, unsafe[i], unsafe[o]))
+ goto no_stable_set;
+
+ /* if we arrive here, we have a stable set */
+ /* compute the weight of the stable set*/
+ curr_weight = 0;
+ bitset_foreach(curr, pos)
+ curr_weight += unsafe_costs[pos];
+
+ /* any better ? */
+ if (curr_weight > best_weight) {
+ best_weight = curr_weight;
+ bitset_copy(best, curr);
}
+
+no_stable_set:
+ bitset_minus1(curr);
}
}
+
+ /* transfer the best set into the qn */
+ qn->mis_size = 1+safe_count+bitset_popcount(best);
+ qn->mis_costs = safe_costs+best_weight;
+ qn->mis[0] = ou->nodes[0]; /* the root is always in a max stable set */
+ next = 1;
+ for (i=0; i<safe_count; ++i)
+ qn->mis[next++] = safe[i];
+ bitset_foreach(best, pos)
+ qn->mis[next++] = unsafe[pos];
}
/**
* Creates a new qnode
*/
-static INLINE qnode_t *new_qnode(const unit_t *ou, int color) {
- qnode_t *qn = malloc(sizeof(*qn));
- qn->color = color;
- qn->mis = malloc(ou->node_count * sizeof(*qn->mis));
- qn->conflicts = new_set(set_cmp_conflict_t, SLOTS_CONFLICTS);
+static inline qnode_t *new_qnode(const unit_t *ou, int color)
+{
+ qnode_t *qn = XMALLOC(qnode_t);
+ qn->color = color;
+ qn->mis = XMALLOCN(ir_node*, ou->node_count);
+ qn->conflicts = new_set(set_cmp_conflict_t, SLOTS_CONFLICTS);
qn->changed_nodes = new_set(set_cmp_node_stat_t, SLOTS_CHANGED_NODES);
return qn;
}
/**
* Frees space used by a queue node
*/
-static INLINE void free_qnode(qnode_t *qn) {
+static inline void free_qnode(qnode_t *qn)
+{
del_set(qn->conflicts);
del_set(qn->changed_nodes);
- free(qn->mis);
- free(qn);
+ xfree(qn->mis);
+ xfree(qn);
}
/**
- * Inserts a qnode in the sorted queue of the outimization unit. Queue is
+ * Inserts a qnode in the sorted queue of the optimization unit. Queue is
* ordered by field 'size' (the size of the mis) in decreasing order.
*/
-static INLINE void ou_insert_qnode(unit_t *ou, qnode_t *qn) {
+static inline void ou_insert_qnode(unit_t *ou, qnode_t *qn)
+{
struct list_head *lh;
if (qnode_are_conflicting(qn, ou->nodes[0], ou->nodes[0])) {
}
qnode_max_ind_set(qn, ou);
-
- /* set ou->mis_size for lower bound compution */
- if (ou->mis_size < qn->mis_size)
- ou->mis_size = qn->mis_size;
-
/* do the insertion */
- DBG((dbg, LEVEL_4, "\t Insert qnode color %d with size %d\n", qn->color, qn->mis_size));
+ DBG((dbg, LEVEL_4, "\t Insert qnode color %d with cost %d\n", qn->color, qn->mis_costs));
lh = &ou->queue;
while (lh->next != &ou->queue) {
qnode_t *curr = list_entry_queue(lh->next);
- if (curr->mis_size <= qn->mis_size)
+ if (curr->mis_costs <= qn->mis_costs)
break;
lh = lh->next;
}
}
/**
- * Tries to re-allocate colors of nodes in this opt unit, to achieve a lower
- * number of copy instructions placed during SSA-destruction and lowering.
+ * Tries to re-allocate colors of nodes in this opt unit, to achieve lower
+ * costs of copy instructions placed during SSA-destruction and lowering.
* Works only for opt units with exactly 1 root node, which is the
- * case for approximately 80% of all phi classes and all register constrained
+ * case for approximately 80% of all phi classes and 100% of register constrained
* nodes. (All other phi classes are reduced to this case.)
*/
-static void ou_optimize(unit_t *ou) {
- int i;
- qnode_t *curr, *tmp;
-
+static void ou_optimize(unit_t *ou, bitset_t const *const allocatable_regs, be_ifg_t *const ifg)
+{
DBG((dbg, LEVEL_1, "\tOptimizing unit:\n"));
- for (i=0; i<ou->node_count; ++i)
- DBG((dbg, LEVEL_1, "\t %n\n", ou->nodes[i]));
+ for (int i = 0; i < ou->node_count; ++i)
+ DBG((dbg, LEVEL_1, "\t %+F\n", ou->nodes[i]));
/* init queue */
INIT_LIST_HEAD(&ou->queue);
- for (i=0; i<MAX_COLORS; ++i)
- if (is_possible_color(ou->nodes[0], i))
- ou_insert_qnode(ou, new_qnode(ou, i));
+
+ arch_register_req_t const *const req = arch_get_irn_register_req(ou->nodes[0]);
+ unsigned const n_regs = req->cls->n_regs;
+ if (arch_register_req_is(req, limited)) {
+ unsigned const* limited = req->limited;
+
+ for (unsigned idx = 0; idx != n_regs; ++idx) {
+ if (!bitset_is_set(allocatable_regs, idx))
+ continue;
+ if (!rbitset_is_set(limited, idx))
+ continue;
+
+ ou_insert_qnode(ou, new_qnode(ou, idx));
+ }
+ } else {
+ for (unsigned idx = 0; idx != n_regs; ++idx) {
+ if (!bitset_is_set(allocatable_regs, idx))
+ continue;
+
+ ou_insert_qnode(ou, new_qnode(ou, idx));
+ }
+ }
/* search best */
- while (!list_empty(&ou->queue)) {
+ qnode_t *curr;
+ for (;;) {
+ assert(!list_empty(&ou->queue));
/* get head of queue */
curr = list_entry_queue(ou->queue.next);
list_del(&curr->queue);
- DBG((dbg, LEVEL_2, "\t Examine qnode color %d with size %d\n", curr->color, curr->mis_size));
+ DBG((dbg, LEVEL_2, "\t Examine qnode color %d with cost %d\n", curr->color, curr->mis_costs));
/* try */
- if (qnode_try_color(curr))
+ if (qnode_try_color(curr, allocatable_regs, ifg))
break;
+
/* no success, so re-insert */
del_set(curr->changed_nodes);
curr->changed_nodes = new_set(set_cmp_node_stat_t, SLOTS_CHANGED_NODES);
/* apply the best found qnode */
if (curr->mis_size >= 2) {
- DBG((dbg, LEVEL_1, "\t Best color: %d Copies: %d/%d\n", curr->color, ou->interf+ou->node_count-curr->mis_size, ou->interf+ou->node_count-1));
- /* globally pin root and eventually others */
+ int root_col = qnode_get_new_color(curr, ou->nodes[0]);
+ DBG((dbg, LEVEL_1, "\t Best color: %d Costs: %d << %d << %d\n", curr->color, ou->min_nodes_costs, ou->all_nodes_costs - curr->mis_costs, ou->all_nodes_costs));
+ /* globally pin root and all args which have the same color */
pset_insert_ptr(pinned_global, ou->nodes[0]);
- for (i=1; i<ou->node_count; ++i) {
- const ir_node *irn = ou->nodes[i];
+ for (int i = 1; i < ou->node_count; ++i) {
+ ir_node *irn = ou->nodes[i];
int nc = qnode_get_new_color(curr, irn);
- if (nc != NO_COLOR && nc == qnode_get_new_color(curr, ou->nodes[0]))
+ if (nc != NO_COLOR && nc == root_col)
pset_insert_ptr(pinned_global, irn);
}
/* set color of all changed nodes */
- node_stat_t *ns;
- for (ns = set_first(curr->changed_nodes); ns; ns = set_next(curr->changed_nodes)) {
+ foreach_set(curr->changed_nodes, node_stat_t, ns) {
/* NO_COLOR is possible, if we had an undo */
if (ns->new_color != NO_COLOR) {
- DBG((dbg, LEVEL_2, "\t color(%n) := %d\n", ns->irn, ns->new_color));
- set_irn_color(ns->irn, ns->new_color);
+ DBG((dbg, LEVEL_1, "\t color(%+F) := %d\n", ns->irn, ns->new_color));
+ set_irn_col(req->cls, ns->irn, ns->new_color);
}
}
}
free_qnode(curr);
}
-void co_heur_opt(copy_opt_t *co) {
- unit_t *curr;
- dbg = firm_dbg_register("ir.be.copyoptheur");
- firm_dbg_set_mask(dbg, DEBUG_LVL);
+/**
+ * Solves the problem using a heuristic approach
+ * Uses the OU data structure
+ */
+int co_solve_heuristic(copy_opt_t *co)
+{
+ ASSERT_OU_AVAIL(co);
pinned_global = pset_new_ptr(SLOTS_PINNED_GLOBAL);
- list_for_each_entry(unit_t, curr, &co->units, units)
+ bitset_t const *const allocatable_regs = co->cenv->allocatable_regs;
+ be_ifg_t *const ifg = co->cenv->ifg;
+ list_for_each_entry(unit_t, curr, &co->units, units) {
if (curr->node_count > 1)
- ou_optimize(curr);
+ ou_optimize(curr, allocatable_regs, ifg);
+ }
del_pset(pinned_global);
+ return 0;
+}
+
+BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copyheur)
+void be_init_copyheur(void)
+{
+ static co_algo_info copyheur = {
+ co_solve_heuristic, 0
+ };
+
+ be_register_copyopt("heur1", ©heur);
+ FIRM_DBG_REGISTER(dbg, "ir.be.copyoptheur");
}
projects / libfirm / blobdiff