2 * Copyright (C) 1995-2007 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Simple copy minimization heuristics.
23 * @author Christian Wuerdig
27 * This is the C implementation of the mst algorithm
28 * originally written in Java by Sebastian Hack.
29 * (also known as "heur3" :)
30 * Performs simple copy minimization.
34 #endif /* HAVE_CONFIG_H */
41 #include "raw_bitset.h"
42 #include "irphase_t.h"
54 #include "becopyopt_t.h"
57 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
59 #define COL_COST_INFEASIBLE DBL_MAX
60 #define AFF_NEIGHBOUR_FIX_BENEFIT 128.0
61 #define NEIGHBOUR_CONSTR_COSTS 64.0
63 #define DBG_AFF_CHUNK(env, level, chunk) DEBUG_ONLY(do { if (firm_dbg_get_mask(dbg) & (level)) dbg_aff_chunk((env), (chunk)); } while(0))
64 #define DBG_COL_COST(env, level, cost) DEBUG_ONLY(do { if (firm_dbg_get_mask(dbg) & (level)) dbg_col_cost((env), (cost)); } while(0))
66 static int last_chunk_id = 0;
68 typedef struct _col_cost_t {
76 typedef struct _aff_chunk_t {
78 bitset_t *nodes; /**< A bitset containing all nodes inside this chunk. */
79 bitset_t *interfere; /**< A bitset containing all interfering neighbours of the nodes in this chunk. */
80 int weight; /**< Weight of this chunk */
81 unsigned weight_consistent : 1; /**< Set if the weight is consistent. */
82 unsigned deleted : 1; /**< Set if the was deleted. */
83 int id; /**< For debugging: An id of this chunk. */
89 typedef struct _aff_edge_t {
90 ir_node *src; /**< Source node. */
91 ir_node *tgt; /**< Target node. */
92 double weight; /**< The weight of this edge. */
95 /* main coalescing environment */
96 typedef struct _co_mst_env_t {
97 int n_regs; /**< number of regs in class */
98 int k; /**< number of non-ignore registers in class */
99 bitset_t *ignore_regs; /**< set containing all global ignore registers */
100 ir_phase ph; /**< phase object holding data for nodes */
101 pqueue *chunks; /**< priority queue for chunks */
102 pset *chunkset; /**< set holding all chunks */
103 be_ifg_t *ifg; /**< the interference graph */
104 const arch_env_t *aenv; /**< the arch environment */
105 copy_opt_t *co; /**< the copy opt object */
108 /* stores coalescing related information for a node */
109 typedef struct _co_mst_irn_t {
110 ir_node *irn; /**< the irn this information belongs to */
111 aff_chunk_t *chunk; /**< the chunk this irn belongs to */
112 bitset_t *adm_colors; /**< set of admissible colors for this irn */
113 ir_node **int_neighs; /**< array of all interfering neighbours (cached for speed reasons) */
114 int n_neighs; /**< length of the interfering neighbours array. */
115 int int_aff_neigh; /**< number of interfering affinity neighbours */
116 int col; /**< color currently assigned */
117 int init_col; /**< the initial color */
118 int tmp_col; /**< a temporary assigned color */
119 unsigned fixed : 1; /**< the color is fixed */
120 unsigned tmp_fixed : 1; /**< the color is temporary fixed */
123 #define get_co_mst_irn(mst_env, irn) (phase_get_or_set_irn_data(&(mst_env)->ph, (irn)))
125 typedef int decide_func_t(const co_mst_irn_t *node, int col);
130 * Write a chunk to stderr for debugging.
132 static void dbg_aff_chunk(const co_mst_env_t *env, const aff_chunk_t *c) {
134 if (c->weight_consistent)
135 ir_fprintf(stderr, " $%d ", c->weight);
136 ir_fprintf(stderr, "{");
137 bitset_foreach(c->nodes, idx) {
138 ir_node *n = get_idx_irn(env->co->irg, idx);
139 ir_fprintf(stderr, " %+F,", n);
141 ir_fprintf(stderr, "}");
145 * Dump all admissible colors to stderr.
147 static void dbg_admissible_colors(const co_mst_env_t *env, const co_mst_irn_t *node) {
149 if (bitset_popcnt(node->adm_colors) < 1)
150 fprintf(stderr, "no admissible colors?!?");
152 bitset_foreach(node->adm_colors, idx)
153 fprintf(stderr, " %d", idx);
158 * Dump color-cost pairs to stderr.
160 static void dbg_col_cost(const co_mst_env_t *env, const col_cost_t *cost) {
162 for (i = 0; i < env->n_regs; ++i) {
163 if (cost[i].cost == COL_COST_INFEASIBLE)
164 fprintf(stderr, " (%d, INF)", cost[i].col);
166 fprintf(stderr, " (%d, %.1f)", cost[i].col, cost[i].cost);
170 #endif /* DEBUG_libfirm */
172 static INLINE int get_mst_irn_col(const co_mst_irn_t *node) {
173 return node->tmp_fixed ? node->tmp_col : node->col;
177 * @return 1 if node @p node has color @p col, 0 otherwise.
179 static int decider_has_color(const co_mst_irn_t *node, int col) {
180 return get_mst_irn_col(node) == col;
184 * @return 1 if node @p node has not color @p col, 0 otherwise.
186 static int decider_hasnot_color(const co_mst_irn_t *node, int col) {
187 return get_mst_irn_col(node) != col;
191 * Always returns true.
193 static int decider_always_yes(const co_mst_irn_t *node, int col) {
197 /** compares two affinity edges by its weight */
198 static int cmp_aff_edge(const void *a, const void *b) {
199 const aff_edge_t *e1 = a;
200 const aff_edge_t *e2 = b;
202 if (e2->weight == e1->weight) {
203 if (e2->src->node_idx == e1->src->node_idx)
204 return QSORT_CMP(e2->tgt->node_idx, e1->tgt->node_idx);
206 return QSORT_CMP(e2->src->node_idx, e1->src->node_idx);
208 /* sort in descending order */
209 return QSORT_CMP(e2->weight, e1->weight);
212 /** compares to color-cost pairs */
213 static int cmp_col_cost(const void *a, const void *b) {
214 const col_cost_t *c1 = a;
215 const col_cost_t *c2 = b;
217 return c1->cost < c2->cost ? -1 : 1;
221 * Creates a new affinity chunk
223 static INLINE aff_chunk_t *new_aff_chunk(co_mst_env_t *env) {
224 aff_chunk_t *c = xmalloc(sizeof(*c));
226 c->weight_consistent = 0;
227 c->n = NEW_ARR_F(ir_node *, 0);
228 c->nodes = bitset_irg_malloc(env->co->irg);
229 c->interfere = bitset_irg_malloc(env->co->irg);
230 c->id = last_chunk_id++;
231 pset_insert(env->chunkset, c, c->id);
236 * Frees all memory allocated by an affinity chunk.
238 static INLINE void delete_aff_chunk(co_mst_env_t *env, aff_chunk_t *c) {
239 pset_remove(env->chunkset, c, c->id);
240 bitset_free(c->nodes);
241 bitset_free(c->interfere);
248 * Adds a node to an affinity chunk
250 static INLINE void aff_chunk_add_node(aff_chunk_t *c, co_mst_irn_t *node) {
253 if (bitset_is_set(c->nodes, get_irn_idx(node->irn)))
256 c->weight_consistent = 0;
258 bitset_set(c->nodes, get_irn_idx(node->irn));
260 ARR_APP1(ir_node *, c->n, node->irn);
262 for (i = node->n_neighs - 1; i >= 0; --i) {
263 ir_node *neigh = node->int_neighs[i];
264 bitset_set(c->interfere, get_irn_idx(neigh));
269 * In case there is no phase information for irn, initialize it.
271 static void *co_mst_irn_init(ir_phase *ph, ir_node *irn, void *old) {
272 co_mst_irn_t *res = old ? old : phase_alloc(ph, sizeof(res[0]));
273 co_mst_env_t *env = ph->priv;
276 const arch_register_req_t *req;
277 void *nodes_it = be_ifg_nodes_iter_alloca(env->ifg);
286 res->int_neighs = NULL;
287 res->int_aff_neigh = 0;
288 res->col = arch_register_get_index(arch_get_irn_register(env->aenv, irn));
289 res->init_col = res->col;
291 DB((dbg, LEVEL_4, "Creating phase info for %+F\n", irn));
293 /* set admissible registers */
294 res->adm_colors = bitset_obstack_alloc(phase_obst(ph), env->n_regs);
296 /* Exclude colors not assignable to the irn */
297 req = arch_get_register_req(env->aenv, irn, -1);
298 if (arch_register_req_is(req, limited))
299 rbitset_copy_to_bitset(req->limited, res->adm_colors);
301 bitset_set_all(res->adm_colors);
303 /* exclude global ignore registers as well */
304 bitset_andnot(res->adm_colors, env->ignore_regs);
306 /* set the number of interfering affinity neighbours to -1, they are calculated later */
307 res->int_aff_neigh = -1;
309 /* build list of interfering neighbours */
311 be_ifg_foreach_neighbour(env->ifg, nodes_it, irn, neigh) {
312 if (! arch_irn_is(env->aenv, neigh, ignore)) {
313 obstack_ptr_grow(phase_obst(ph), neigh);
317 res->int_neighs = obstack_finish(phase_obst(ph));
324 * Check if affinity chunk @p chunk interferes with node @p irn.
326 static INLINE int aff_chunk_interferes(co_mst_env_t *env, const aff_chunk_t *chunk, ir_node *irn) {
328 if (bitset_is_set(chunk->interfere, get_irn_idx(irn)))
331 const co_mst_irn_t *node = get_co_mst_irn(env, irn);
332 const ir_node *neigh;
335 for (i = 0; i < node->n_neighs; ++i) {
336 neigh = node->int_neighs[i];
337 if (! arch_irn_is(env->aenv, neigh, ignore) && bitset_is_set(chunk->nodes, get_irn_idx(neigh)))
345 * Check if there are interference edges from c1 to c2.
346 * @param env The global co_mst environment
348 * @param c2 Another chunk
349 * @return 1 if there are interferences between nodes of c1 and c2, 0 otherwise.
351 static INLINE int aff_chunks_interfere(co_mst_env_t *env, const aff_chunk_t *c1, const aff_chunk_t *c2) {
358 tmp = bitset_alloca(get_irg_last_idx(env->co->irg));
359 tmp = bitset_copy(tmp, c1->interfere);
360 tmp = bitset_and(tmp, c2->nodes);
361 if (bitset_popcnt(tmp) > 0)
364 /* check if there is a node in c2 having an interfering neighbor in c1 */
365 bitset_foreach(c2->nodes, idx) {
366 ir_node *n = get_idx_irn(env->co->irg, idx);
368 if (aff_chunk_interferes(env, c1, n))
376 * Returns the affinity chunk of @p irn or creates a new
377 * one with @p irn as element if there is none assigned.
379 static INLINE aff_chunk_t *get_aff_chunk(co_mst_env_t *env, ir_node *irn) {
380 co_mst_irn_t *node = get_co_mst_irn(env, irn);
385 * Let chunk(src) absorb the nodes of chunk(tgt) (only possible when there
386 * are no interference edges from chunk(src) to chunk(tgt)).
387 * @return 1 if successful, 0 if not possible
389 static int aff_chunk_absorb(co_mst_env_t *env, ir_node *src, ir_node *tgt) {
390 aff_chunk_t *c1 = get_aff_chunk(env, src);
391 aff_chunk_t *c2 = get_aff_chunk(env, tgt);
394 DB((dbg, LEVEL_4, "Attempt to let c1 (id %d): ", c1 ? c1->id : -1));
396 DBG_AFF_CHUNK(env, LEVEL_4, c1);
398 DB((dbg, LEVEL_4, "{%+F}", src));
400 DB((dbg, LEVEL_4, "\n\tabsorb c2 (id %d): ", c2 ? c2->id : -1));
402 DBG_AFF_CHUNK(env, LEVEL_4, c2);
404 DB((dbg, LEVEL_4, "{%+F}", tgt));
406 DB((dbg, LEVEL_4, "\n"));
411 /* no chunk exists */
412 co_mst_irn_t *mirn = get_co_mst_irn(env, src);
415 for (i = mirn->n_neighs - 1; i >= 0; --i) {
416 if (mirn->int_neighs[i] == tgt)
420 /* create one containing both nodes */
421 c1 = new_aff_chunk(env);
422 aff_chunk_add_node(c1, get_co_mst_irn(env, src));
423 aff_chunk_add_node(c1, get_co_mst_irn(env, tgt));
427 /* c2 already exists */
428 if (! aff_chunk_interferes(env, c2, src)) {
429 aff_chunk_add_node(c2, get_co_mst_irn(env, src));
433 } else if (c2 == NULL) {
434 /* c1 already exists */
435 if (! aff_chunk_interferes(env, c1, tgt)) {
436 aff_chunk_add_node(c1, get_co_mst_irn(env, tgt));
439 } else if (c1 != c2 && ! aff_chunks_interfere(env, c1, c2)) {
442 bitset_or(c1->nodes, c2->nodes);
443 bitset_or(c1->interfere, c2->interfere);
444 c1->weight_consistent = 0;
446 // bitset_foreach(c2->nodes, idx) {
447 for (idx = 0; idx < ARR_LEN(c2->n); ++idx) {
449 // ir_node *n = get_idx_irn(env->co->irg, idx);
450 ir_node *n = c2->n[idx];
451 co_mst_irn_t *mn = get_co_mst_irn(env, n);
454 for (i = 0; i < ARR_LEN(c1->n); ++i) {
458 if (i >= ARR_LEN(c1->n))
459 ARR_APP1(ir_node *, c1->n, n);
462 delete_aff_chunk(env, c2);
465 DB((dbg, LEVEL_4, " ... c1 interferes with c2, skipped\n"));
469 DB((dbg, LEVEL_4, " ... absorbed\n"));
474 * Assures that the weight of the given chunk is consistent.
476 static void aff_chunk_assure_weight(const co_mst_env_t *env, aff_chunk_t *c) {
477 if (! c->weight_consistent) {
481 // bitset_foreach(c->nodes, idx) {
482 // ir_node *n = get_idx_irn(env->co->irg, idx);
483 for (idx = 0; idx < ARR_LEN(c->n); ++idx) {
484 ir_node *n = c->n[idx];
485 const affinity_node_t *an = get_affinity_info(env->co, n);
489 co_gs_foreach_neighb(an, neigh) {
490 const ir_node *m = neigh->irn;
491 const int m_idx = get_irn_idx(m);
493 /* skip ignore nodes */
494 if (arch_irn_is(env->aenv, m, ignore))
497 w += bitset_is_set(c->nodes, m_idx) ? neigh->costs : 0;
503 c->weight_consistent = 1;
508 * Count the number of interfering affinity neighbours
510 static int count_interfering_aff_neighs(co_mst_env_t *env, const affinity_node_t *an) {
511 const neighb_t *neigh;
512 ir_node *irn = an->irn;
513 const co_mst_irn_t *node = get_co_mst_irn(env, irn);
516 co_gs_foreach_neighb(an, neigh) {
517 const ir_node *n = neigh->irn;
520 /* skip ignore nodes */
521 if (arch_irn_is(env->aenv, n, ignore))
524 /* check if the affinity neighbour interfere */
525 for (i = 0; i < node->n_neighs; ++i) {
526 if (node->int_neighs[i] == n) {
537 * Build chunks of nodes connected by affinity edges.
538 * We start at the heaviest affinity edge.
539 * The chunks of the two edge-defining nodes will be
540 * merged if there are no interference edges from one
541 * chunk to the other.
543 static void build_affinity_chunks(co_mst_env_t *env) {
544 void *nodes_it = be_ifg_nodes_iter_alloca(env->ifg);
545 aff_edge_t *edges = NEW_ARR_F(aff_edge_t, 0);
548 aff_chunk_t *curr_chunk;
549 pset_new_iterator_t iter;
551 /* at first we create the affinity edge objects */
552 be_ifg_foreach_node(env->ifg, nodes_it, n) {
553 int n_idx = get_irn_idx(n);
557 /* skip ignore nodes */
558 if (arch_irn_is(env->aenv, n, ignore))
561 n1 = get_co_mst_irn(env, n);
562 an = get_affinity_info(env->co, n);
567 if (n1->int_aff_neigh < 0)
568 n1->int_aff_neigh = count_interfering_aff_neighs(env, an);
569 co_gs_foreach_neighb(an, neigh) {
570 ir_node *m = neigh->irn;
571 int m_idx = get_irn_idx(m);
573 /* record the edge in only one direction */
578 /* skip ignore nodes */
579 if (arch_irn_is(env->aenv, m, ignore))
585 n2 = get_co_mst_irn(env, m);
586 if (n2->int_aff_neigh < 0) {
587 affinity_node_t *am = get_affinity_info(env->co, m);
588 n2->int_aff_neigh = count_interfering_aff_neighs(env, am);
590 edge.weight = (double)neigh->costs / (double)(1 + n1->int_aff_neigh + n2->int_aff_neigh);
591 ARR_APP1(aff_edge_t, edges, edge);
597 /* now: sort edges and build the affinity chunks */
598 len = ARR_LEN(edges);
599 qsort(edges, len, sizeof(edges[0]), cmp_aff_edge);
600 for (i = 0; i < len; ++i) {
601 DBG((dbg, LEVEL_1, "edge (%u,%u) %f\n", edges[i].src->node_idx, edges[i].tgt->node_idx, edges[i].weight));
603 (void)aff_chunk_absorb(env, edges[i].src, edges[i].tgt);
606 /* now insert all chunks into a priority queue */
607 foreach_pset(env->chunkset, curr_chunk) {
608 aff_chunk_assure_weight(env, curr_chunk);
610 DBG((dbg, LEVEL_1, "entry #%d", curr_chunk->id));
611 DBG_AFF_CHUNK(env, LEVEL_1, curr_chunk);
612 DBG((dbg, LEVEL_1, "\n"));
614 pqueue_put(env->chunks, curr_chunk, curr_chunk->weight);
616 foreach_phase_irn(&env->ph, n) {
617 co_mst_irn_t *mirn = get_co_mst_irn(env, n);
619 if (mirn->chunk == NULL) {
620 /* no chunk is allocated so far, do it now */
621 aff_chunk_t *curr_chunk = new_aff_chunk(env);
622 aff_chunk_add_node(curr_chunk, mirn);
624 aff_chunk_assure_weight(env, curr_chunk);
626 DBG((dbg, LEVEL_1, "entry #%d", curr_chunk->id));
627 DBG_AFF_CHUNK(env, LEVEL_1, curr_chunk);
628 DBG((dbg, LEVEL_1, "\n"));
630 pqueue_put(env->chunks, curr_chunk, curr_chunk->weight);
638 * Greedy collect affinity neighbours into thew new chunk @p chunk starting at node @p node.
640 static void expand_chunk_from(co_mst_env_t *env, co_mst_irn_t *node, bitset_t *visited,
641 aff_chunk_t *chunk, aff_chunk_t *orig_chunk, decide_func_t *decider, int col)
643 waitq *nodes = new_waitq();
645 DBG((dbg, LEVEL_1, "\n\tExpanding new chunk (#%d) from %+F, color %d:", chunk->id, node->irn, col));
647 /* init queue and chunk */
648 waitq_put(nodes, node);
649 bitset_set(visited, get_irn_idx(node->irn));
650 aff_chunk_add_node(chunk, node);
651 DB((dbg, LEVEL_1, " %+F", node->irn));
653 /* as long as there are nodes in the queue */
654 while (! waitq_empty(nodes)) {
655 co_mst_irn_t *n = waitq_get(nodes);
656 affinity_node_t *an = get_affinity_info(env->co, n->irn);
658 /* check all affinity neighbors */
661 co_gs_foreach_neighb(an, neigh) {
662 ir_node *m = neigh->irn;
663 int m_idx = get_irn_idx(m);
666 /* skip ignore nodes */
667 if (arch_irn_is(env->aenv, m, ignore))
670 n2 = get_co_mst_irn(env, m);
672 if (! bitset_is_set(visited, m_idx) &&
675 ! aff_chunk_interferes(env, chunk, m) &&
676 bitset_is_set(orig_chunk->nodes, m_idx))
679 following conditions are met:
680 - neighbour is not visited
681 - neighbour likes the color
682 - neighbour has not yet a fixed color
683 - the new chunk doesn't interfere with the neighbour
684 - neighbour belongs or belonged once to the original chunk
686 bitset_set(visited, m_idx);
687 aff_chunk_add_node(chunk, n2);
688 DB((dbg, LEVEL_1, " %+F", n2->irn));
689 /* enqueue for further search */
690 waitq_put(nodes, n2);
696 DB((dbg, LEVEL_1, "\n"));
702 * Fragment the given chunk into chunks having given color and not having given color.
704 static aff_chunk_t *fragment_chunk(co_mst_env_t *env, int col, aff_chunk_t *c, waitq *tmp) {
705 bitset_t *visited = bitset_irg_malloc(env->co->irg);
707 aff_chunk_t *best = NULL;
709 // bitset_foreach(c->nodes, idx) {
710 for (idx = 0; idx < ARR_LEN(c->n); ++idx) {
713 aff_chunk_t *tmp_chunk;
714 decide_func_t *decider;
718 if (bitset_is_set(visited, get_irn_idx(irn)))
721 // irn = get_idx_irn(env->co->irg, idx);
722 node = get_co_mst_irn(env, irn);
724 if (get_mst_irn_col(node) == col) {
725 decider = decider_has_color;
727 DBG((dbg, LEVEL_4, "\tcolor %d wanted", col));
730 decider = decider_hasnot_color;
732 DBG((dbg, LEVEL_4, "\tcolor %d forbidden", col));
735 /* create a new chunk starting at current node */
736 tmp_chunk = new_aff_chunk(env);
737 waitq_put(tmp, tmp_chunk);
738 expand_chunk_from(env, node, visited, tmp_chunk, c, decider, col);
739 assert(bitset_popcnt(tmp_chunk->nodes) > 0 && "No nodes added to chunk");
741 /* remember the local best */
742 aff_chunk_assure_weight(env, tmp_chunk);
743 if (check_for_best && (! best || best->weight < tmp_chunk->weight))
747 assert(best && "No chunk found?");
748 bitset_free(visited);
753 * Initializes an array of color-cost pairs.
754 * Sets forbidden colors to costs COL_COST_INFEASIBLE and all others to @p c.
756 static INLINE void col_cost_init(co_mst_env_t *env, col_cost_t *cost, double c) {
759 for (i = 0; i < env->n_regs; ++i) {
761 if (bitset_is_set(env->ignore_regs, i))
762 cost[i].cost = COL_COST_INFEASIBLE;
769 * Initializes an array of color-cost pairs.
770 * Sets all colors except color @p col to COL_COST_INFEASIBLE and @p col to 0.0
772 static INLINE void col_cost_init_single(co_mst_env_t *env, col_cost_t *cost, int col) {
773 assert(! bitset_is_set(env->ignore_regs, col) && "Attempt to use forbidden color.");
774 col_cost_init(env, cost, COL_COST_INFEASIBLE);
781 * Resets the temporary fixed color of all nodes within wait queue @p nodes.
782 * ATTENTION: the queue is empty after calling this function!
784 static INLINE void reject_coloring(waitq *nodes) {
785 DB((dbg, LEVEL_4, "\treject coloring for"));
786 while (! waitq_empty(nodes)) {
787 co_mst_irn_t *n = waitq_get(nodes);
788 DB((dbg, LEVEL_4, " %+F", n->irn));
791 DB((dbg, LEVEL_4, "\n"));
795 * Determines the costs for each color if it would be assigned to node @p node.
797 static void determine_color_costs(co_mst_env_t *env, co_mst_irn_t *node, col_cost_t *costs) {
798 affinity_node_t *an = get_affinity_info(env->co, node->irn);
802 col_cost_init(env, costs, 0.0);
804 /* calculate (negative) costs for affinity neighbours */
806 co_gs_foreach_neighb(an, aff_neigh) {
807 ir_node *m = aff_neigh->irn;
811 /* skip ignore nodes */
812 if (arch_irn_is(env->aenv, m, ignore))
815 neigh = get_co_mst_irn(env, m);
816 c = (double)aff_neigh->costs;
818 /* calculate costs for fixed affinity neighbours */
819 if (neigh->tmp_fixed || neigh->fixed) {
820 int col = get_mst_irn_col(neigh);
821 costs[col].cost -= c * AFF_NEIGHBOUR_FIX_BENEFIT;
826 /* calculate (positive) costs for interfering neighbours */
827 for (i = 0; i < node->n_neighs; ++i) {
832 int_neigh = node->int_neighs[i];
834 /* skip ignore nodes */
835 if (arch_irn_is(env->aenv, int_neigh, ignore))
838 neigh = get_co_mst_irn(env, int_neigh);
839 col = get_mst_irn_col(neigh);
840 col_cnt = bitset_popcnt(neigh->adm_colors);
842 if (neigh->tmp_fixed || neigh->fixed) {
843 /* colors of fixed interfering neighbours are infeasible */
844 costs[col].cost = COL_COST_INFEASIBLE;
846 else if (col_cnt < env->k) {
847 /* calculate costs for constrained interfering neighbours */
848 double ratio = 1.0 - ((double)col_cnt / (double)env->k);
850 bitset_foreach_clear(neigh->adm_colors, idx) {
851 /* check only explicitly forbidden colors (skip global forbidden ones) */
852 if (! bitset_is_set(env->ignore_regs, idx)) {
853 costs[col].cost += ratio * NEIGHBOUR_CONSTR_COSTS;
859 /* set all not admissible colors to COL_COST_INFEASIBLE */
860 bitset_foreach_clear(node->adm_colors, idx)
861 costs[idx].cost = COL_COST_INFEASIBLE;
864 /* need forward declaration due to recursive call */
865 static int recolor_nodes(co_mst_env_t *env, co_mst_irn_t *node, col_cost_t *costs, waitq *changed_ones);
868 * Tries to change node to a color but @p explude_col.
869 * @return 1 if succeeded, 0 otherwise.
871 static int change_node_color_excluded(co_mst_env_t *env, co_mst_irn_t *node, int exclude_col, waitq *changed_ones) {
872 int col = get_mst_irn_col(node);
875 /* neighbours has already a different color -> good, temporary fix it */
876 if (col != exclude_col) {
879 waitq_put(changed_ones, node);
883 /* The node has the color it should not have _and_ has not been visited yet. */
884 if (! (node->tmp_fixed || node->fixed)) {
885 col_cost_t *costs = alloca(env->n_regs * sizeof(costs[0]));
887 /* Get the costs for giving the node a specific color. */
888 determine_color_costs(env, node, costs);
890 /* Since the node must not have the not_col, set the costs for that color to "infinity" */
891 costs[exclude_col].cost = COL_COST_INFEASIBLE;
893 /* sort the colors according costs, cheapest first. */
894 qsort(costs, env->n_regs, sizeof(costs[0]), cmp_col_cost);
896 /* Try recoloring the node using the color list. */
897 res = recolor_nodes(env, node, costs, changed_ones);
904 * Tries to bring node @p node to cheapest color and color all interfering neighbours with other colors.
905 * ATTENTION: Expect @p costs already sorted by increasing costs.
906 * @return 1 if coloring could be applied, 0 otherwise.
908 static int recolor_nodes(co_mst_env_t *env, co_mst_irn_t *node, col_cost_t *costs, waitq *changed_ones) {
910 waitq *local_changed = new_waitq();
911 waitq *tmp = new_waitq();
913 DBG((dbg, LEVEL_1, "\tRecoloring %+F with color-costs", node->irn));
914 DBG_COL_COST(env, LEVEL_1, costs);
915 DB((dbg, LEVEL_1, "\n"));
917 for (i = 0; i < env->n_regs; ++i) {
918 int tgt_col = costs[i].col;
922 /* If the costs for that color (and all successive) are infinite, bail out we won't make it anyway. */
923 if (costs[i].cost == COL_COST_INFEASIBLE) {
925 del_waitq(local_changed);
930 /* Set the new color of the node and mark the node as temporarily fixed. */
931 assert(! node->tmp_fixed && "Node must not have been temporary fixed.");
933 node->tmp_col = tgt_col;
934 DBG((dbg, LEVEL_4, "\tTemporary setting %+F to color %d\n", node->irn, tgt_col));
936 assert(waitq_empty(local_changed) && "Node queue should be empty here.");
937 waitq_put(local_changed, node);
939 /* try to color all interfering neighbours with current color forbidden */
940 for (j = 0; j < node->n_neighs; ++j) {
944 neigh = node->int_neighs[j];
946 /* skip ignore nodes */
947 if (arch_irn_is(env->aenv, neigh, ignore))
950 nn = get_co_mst_irn(env, neigh);
951 DB((dbg, LEVEL_4, "\tHandling neighbour %+F, at position %d (fixed: %d, tmp_fixed: %d, tmp_col: %d, col: %d)\n",
952 neigh, j, nn->fixed, nn->tmp_fixed, nn->tmp_col, nn->col));
955 Try to change the color of the neighbor and record all nodes which
956 get changed in the tmp list. Add this list to the "changed" list for
957 that color. If we did not succeed to change the color of the neighbor,
958 we bail out and try the next color.
960 if (get_mst_irn_col(nn) == tgt_col) {
961 /* try to color neighbour with tgt_col forbidden */
962 neigh_ok = change_node_color_excluded(env, nn, tgt_col, tmp);
964 /* join lists of changed nodes */
965 while (! waitq_empty(tmp))
966 waitq_put(local_changed, waitq_get(tmp));
974 We managed to assign the target color to all neighbors, so from the perspective
975 of the current node, every thing was ok and we can return safely.
978 /* append the local_changed ones to global ones */
979 while (! waitq_empty(local_changed))
980 waitq_put(changed_ones, waitq_get(local_changed));
981 del_waitq(local_changed);
986 /* coloring of neighbours failed, so we try next color */
987 reject_coloring(local_changed);
991 del_waitq(local_changed);
997 * Tries to bring node @p node and all it's neighbours to color @p tgt_col.
998 * @return 1 if color @p col could be applied, 0 otherwise
1000 static int change_node_color(co_mst_env_t *env, co_mst_irn_t *node, int tgt_col, waitq *changed_ones) {
1001 int col = get_mst_irn_col(node);
1003 /* if node already has the target color -> good, temporary fix it */
1004 if (col == tgt_col) {
1005 DBG((dbg, LEVEL_4, "\t\tCNC: %+F has already color %d, fix temporary\n", node->irn, tgt_col));
1006 if (! node->tmp_fixed) {
1007 node->tmp_fixed = 1;
1008 node->tmp_col = tgt_col;
1009 waitq_put(changed_ones, node);
1015 Node has not yet a fixed color and target color is admissible
1016 -> try to recolor node and it's affinity neighbours
1018 if (! (node->fixed || node->tmp_fixed) && bitset_is_set(node->adm_colors, tgt_col)) {
1019 col_cost_t *costs = alloca(env->n_regs * sizeof(costs[0]));
1022 col_cost_init_single(env, costs, tgt_col);
1024 DBG((dbg, LEVEL_4, "\t\tCNC: Attempt to recolor %+F ===>>\n", node->irn));
1025 res = recolor_nodes(env, node, costs, changed_ones);
1026 DBG((dbg, LEVEL_4, "\t\tCNC: <<=== Recoloring of %+F %s\n", node->irn, res ? "succeeded" : "failed"));
1032 if (firm_dbg_get_mask(dbg) & LEVEL_4) {
1033 if (node->fixed || node->tmp_fixed)
1034 DB((dbg, LEVEL_4, "\t\tCNC: %+F has already fixed color %d\n", node->irn, col));
1036 DB((dbg, LEVEL_4, "\t\tCNC: color %d not admissible for %+F (", tgt_col, node->irn));
1037 dbg_admissible_colors(env, node);
1038 DB((dbg, LEVEL_4, ")\n"));
1047 * Tries to color an affinity chunk (or at least a part of it).
1048 * Inserts uncolored parts of the chunk as a new chunk into the priority queue.
1050 static void color_aff_chunk(co_mst_env_t *env, aff_chunk_t *c) {
1051 aff_chunk_t *best_chunk = NULL;
1052 int best_color = -1;
1054 waitq *changed_ones = new_waitq();
1055 waitq *tmp_chunks = new_waitq();
1056 waitq *best_starts = NULL;
1060 DB((dbg, LEVEL_2, "fragmentizing chunk #%d", c->id));
1061 DBG_AFF_CHUNK(env, LEVEL_2, c);
1062 DB((dbg, LEVEL_2, "\n"));
1065 /* check which color is the "best" for the given chunk */
1066 for (col = 0; col < env->n_regs; ++col) {
1068 waitq *good_starts = new_waitq();
1069 aff_chunk_t *local_best;
1071 /* skip ignore colors */
1072 if (bitset_is_set(env->ignore_regs, col))
1075 DB((dbg, LEVEL_3, "\ttrying color %d\n", col));
1077 /* try to bring all nodes of given chunk to the current color. */
1078 // bitset_foreach(c->nodes, idx) {
1079 // ir_node *irn = get_idx_irn(env->co->irg, idx);
1080 for (idx = 0; idx < ARR_LEN(c->n); ++idx) {
1081 ir_node *irn = c->n[idx];
1082 co_mst_irn_t *node = get_co_mst_irn(env, irn);
1085 assert(! node->fixed && "Node must not have a fixed color.");
1086 DB((dbg, LEVEL_4, "\t\tBringing %+F from color %d to color %d ...\n", irn, node->col, col));
1088 good = change_node_color(env, node, col, changed_ones);
1090 waitq_put(good_starts, node);
1093 DB((dbg, LEVEL_4, "\t\t... %+F attempt from %d to %d %s\n", irn, node->col, col, one_good ? "succeeded" : "failed"));
1096 /* try next color when failed */
1100 /* fragment the chunk according to the coloring */
1101 local_best = fragment_chunk(env, col, c, tmp_chunks);
1103 /* search the best of the good list
1104 and make it the new best if it is better than the current */
1106 aff_chunk_assure_weight(env, local_best);
1108 DB((dbg, LEVEL_4, "\t\tlocal best chunk (id %d) for color %d: ", local_best->id, col));
1109 DBG_AFF_CHUNK(env, LEVEL_4, local_best);
1111 if (! best_chunk || best_chunk->weight < local_best->weight) {
1112 best_chunk = local_best;
1115 del_waitq(best_starts);
1116 best_starts = good_starts;
1117 DB((dbg, LEVEL_4, "\n\t\t... setting global best chunk (id %d), color %d\n", best_chunk->id, best_color));
1119 DB((dbg, LEVEL_4, "\n\t\t... omitting, global best is better\n"));
1120 del_waitq(good_starts);
1124 /* reject the coloring and bring the coloring to the initial state */
1125 reject_coloring(changed_ones);
1128 /* free all intermediate created chunks except best one */
1129 while (! waitq_empty(tmp_chunks)) {
1130 aff_chunk_t *tmp = waitq_get(tmp_chunks);
1131 if (tmp != best_chunk)
1132 delete_aff_chunk(env, tmp);
1134 del_waitq(tmp_chunks);
1136 /* return if coloring failed */
1138 del_waitq(changed_ones);
1142 DB((dbg, LEVEL_2, "\tbest chunk #%d ", best_chunk->id));
1143 DBG_AFF_CHUNK(env, LEVEL_2, best_chunk);
1144 DB((dbg, LEVEL_2, "using color %d\n", best_color));
1146 /* get the best fragment from the best list and color it */
1147 // bitset_foreach(best_chunk->nodes, idx) {
1148 // ir_node *irn = get_idx_irn(env->co->irg, idx);
1149 while (! waitq_empty(best_starts)) {
1150 co_mst_irn_t *node = waitq_get(best_starts);
1153 if (! bitset_is_set(best_chunk->nodes, get_irn_idx(node->irn)))
1156 res = change_node_color(env, node, best_color, changed_ones);
1158 panic("Color manifesting failed for %+F, color %d in chunk %d\n", node->irn, best_color, best_chunk->id);
1160 node->chunk = best_chunk;
1162 for (idx = 0; idx < ARR_LEN(best_chunk->n); ++idx) {
1163 ir_node *irn = best_chunk->n[idx];
1164 co_mst_irn_t *node = get_co_mst_irn(env, irn);
1167 res = change_node_color(env, node, best_color, changed_ones);
1169 panic("Color manifesting failed for %+F, color %d in chunk %d\n", irn, best_color, best_chunk->id);
1170 DB((dbg, LEVEL_4, "\tManifesting color %d for %+F, chunk #%d\n", best_color, irn, best_chunk->id));
1172 node->chunk = best_chunk;
1175 /* materialize colors on changed nodes */
1176 while (! waitq_empty(changed_ones)) {
1177 co_mst_irn_t *n = waitq_get(changed_ones);
1179 n->col = n->tmp_col;
1182 /* remove the nodes in best chunk from original chunk */
1183 bitset_andnot(c->nodes, best_chunk->nodes);
1185 ir_node **tmp = NEW_ARR_F(ir_node *, 0);
1186 for (idx = 0; idx < ARR_LEN(c->n); ++idx) {
1187 ir_node *irn = c->n[idx];
1190 for (j = 0; j < ARR_LEN(best_chunk->n); ++j) {
1191 if (best_chunk->n[j] == irn)
1194 if (j >= ARR_LEN(best_chunk->n))
1195 ARR_APP1(ir_node *, tmp, irn);
1201 /* we have to get the nodes back into the original chunk because they are scattered over temporary chunks */
1202 // bitset_foreach(c->nodes, idx) {
1203 // ir_node *n = get_idx_irn(env->co->irg, idx);
1204 for (idx = 0; idx < ARR_LEN(c->n); ++idx) {
1205 ir_node *n = c->n[idx];
1206 co_mst_irn_t *nn = get_co_mst_irn(env, n);
1210 /* fragment the remaining chunk */
1211 visited = bitset_irg_malloc(env->co->irg);
1212 bitset_or(visited, best_chunk->nodes);
1213 // bitset_foreach(c->nodes, idx) {
1214 for (idx = 0; idx < ARR_LEN(c->n); ++idx) {
1215 ir_node *irn = c->n[idx];
1216 if (! bitset_is_set(visited, get_irn_idx(irn))) {
1217 aff_chunk_t *new_chunk = new_aff_chunk(env);
1218 // ir_node *irn = get_idx_irn(env->co->irg, idx);
1219 co_mst_irn_t *node = get_co_mst_irn(env, irn);
1221 expand_chunk_from(env, node, visited, new_chunk, c, decider_always_yes, 0);
1222 aff_chunk_assure_weight(env, new_chunk);
1223 pqueue_put(env->chunks, new_chunk, new_chunk->weight);
1227 /* clear obsolete chunks and free some memory */
1228 delete_aff_chunk(env, best_chunk);
1229 bitset_free(visited);
1230 del_waitq(changed_ones);
1234 * Main driver for mst safe coalescing algorithm.
1236 int co_solve_heuristic_mst(copy_opt_t *co) {
1237 unsigned n_regs = co->cls->n_regs;
1238 bitset_t *ignore_regs = bitset_alloca(n_regs);
1241 co_mst_env_t mst_env;
1244 phase_init(&mst_env.ph, "co_mst", co->irg, PHASE_DEFAULT_GROWTH, co_mst_irn_init, &mst_env);
1246 k = be_put_ignore_regs(co->cenv->birg, co->cls, ignore_regs);
1249 mst_env.n_regs = n_regs;
1251 mst_env.chunks = new_pqueue();
1253 mst_env.ignore_regs = ignore_regs;
1254 mst_env.ifg = co->cenv->ifg;
1255 mst_env.aenv = co->aenv;
1256 mst_env.chunkset = pset_new_ptr(512);
1258 DBG((dbg, LEVEL_1, "==== Coloring %+F, class %s ====\n", co->irg, co->cls->name));
1260 /* build affinity chunks */
1261 build_affinity_chunks(&mst_env);
1263 /* color chunks as long as there are some */
1264 while (! pqueue_empty(mst_env.chunks)) {
1265 aff_chunk_t *chunk = pqueue_get(mst_env.chunks);
1267 color_aff_chunk(&mst_env, chunk);
1268 DB((dbg, LEVEL_4, "<<<====== Coloring chunk (%d) done\n", chunk->id));
1269 delete_aff_chunk(&mst_env, chunk);
1272 /* apply coloring */
1273 foreach_phase_irn(&mst_env.ph, irn) {
1274 co_mst_irn_t *mirn = get_co_mst_irn(&mst_env, irn);
1275 const arch_register_t *reg;
1277 if (arch_irn_is(mst_env.aenv, irn, ignore))
1280 assert(mirn->fixed && "Node should have fixed color");
1282 /* skip nodes where color hasn't changed */
1283 if (mirn->init_col == mirn->col)
1286 reg = arch_register_for_index(co->cls, mirn->col);
1287 arch_set_irn_register(co->aenv, irn, reg);
1288 DB((dbg, LEVEL_1, "%+F set color from %d to %d\n", irn, mirn->init_col, mirn->col));
1291 /* free allocated memory */
1292 del_pqueue(mst_env.chunks);
1293 phase_free(&mst_env.ph);
1294 del_pset(mst_env.chunkset);
1299 void be_init_copyheur4(void) {
1300 FIRM_DBG_REGISTER(dbg, "firm.be.co.heur4");
1303 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_copyheur4);