3 * File name: ir/ir/irgopt.c
4 * Purpose: Optimizations for a whole ir graph, i.e., a procedure.
5 * Author: Christian Schaefer, Goetz Lindenmaier
6 * Modified by: Sebastian Felis
9 * Copyright: (c) 1998-2003 Universität Karlsruhe
10 * Licence: This file protected by GPL - GNU GENERAL PUBLIC LICENSE.
21 #include "irgraph_t.h"
34 #include "pdeq.h" /* Fuer code placement */
39 #include "irbackedge_t.h"
46 #include "iredges_t.h"
49 /* Defined in iropt.c */
50 pset *new_identities (void);
51 void del_identities (pset *value_table);
52 void add_identities (pset *value_table, ir_node *node);
54 /*------------------------------------------------------------------*/
55 /* apply optimizations of iropt to all nodes. */
56 /*------------------------------------------------------------------*/
58 static void init_link (ir_node *n, void *env) {
59 set_irn_link(n, NULL);
62 #if 0 /* Old version. Avoids Ids.
63 This is not necessary: we do a post walk, and get_irn_n
64 removes ids anyways. So it's much cheaper to call the
65 optimization less often and use the exchange() algorithm. */
67 optimize_in_place_wrapper (ir_node *n, void *env) {
69 ir_node *optimized, *old;
71 irn_arity = get_irn_arity(n);
72 for (i = 0; i < irn_arity; i++) {
73 /* get_irn_n skips Id nodes, so comparison old != optimized does not
74 show all optimizations. Therefore always set new predecessor. */
75 old = get_irn_intra_n(n, i);
76 optimized = optimize_in_place_2(old);
77 set_irn_n(n, i, optimized);
80 if (get_irn_op(n) == op_Block) {
81 optimized = optimize_in_place_2(n);
82 if (optimized != n) exchange (n, optimized);
87 optimize_in_place_wrapper (ir_node *n, void *env) {
88 ir_node *optimized = optimize_in_place_2(n);
89 if (optimized != n) exchange (n, optimized);
94 static INLINE void do_local_optimize(ir_node *n) {
95 /* Handle graph state */
96 assert(get_irg_phase_state(current_ir_graph) != phase_building);
98 if (get_opt_global_cse())
99 set_irg_pinned(current_ir_graph, op_pin_state_floats);
100 set_irg_outs_inconsistent(current_ir_graph);
101 set_irg_doms_inconsistent(current_ir_graph);
102 set_irg_loopinfo_inconsistent(current_ir_graph);
104 /* Clean the value_table in irg for the CSE. */
105 del_identities(current_ir_graph->value_table);
106 current_ir_graph->value_table = new_identities();
108 /* walk over the graph */
109 irg_walk(n, init_link, optimize_in_place_wrapper, NULL);
112 void local_optimize_node(ir_node *n) {
113 ir_graph *rem = current_ir_graph;
114 current_ir_graph = get_irn_irg(n);
116 do_local_optimize(n);
118 current_ir_graph = rem;
122 * Block-Walker: uses dominance depth to mark dead blocks.
124 static void kill_dead_blocks(ir_node *block, void *env)
126 if (get_Block_dom_depth(block) < 0)
127 if (block != get_irg_end_block(current_ir_graph)) {
128 /* we don't want that the end block of graphs with
129 endless loops is marked bad (although it is of course */
130 set_Block_dead(block);
135 local_optimize_graph (ir_graph *irg) {
136 ir_graph *rem = current_ir_graph;
137 current_ir_graph = irg;
139 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
140 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
142 do_local_optimize(get_irg_end(irg));
144 current_ir_graph = rem;
148 /*------------------------------------------------------------------*/
149 /* Routines for dead node elimination / copying garbage collection */
150 /* of the obstack. */
151 /*------------------------------------------------------------------*/
154 * Remember the new node in the old node by using a field all nodes have.
156 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
159 * Get this new node, before the old node is forgotten.
161 #define get_new_node(oldn) get_irn_link(oldn)
164 * Check if a new node was set.
166 #define has_new_node(n) (get_new_node(n) != NULL)
169 * We use the block_visited flag to mark that we have computed the
170 * number of useful predecessors for this block.
171 * Further we encode the new arity in this flag in the old blocks.
172 * Remembering the arity is useful, as it saves a lot of pointer
173 * accesses. This function is called for all Phi and Block nodes
177 compute_new_arity(ir_node *b) {
178 int i, res, irn_arity;
181 irg_v = get_irg_block_visited(current_ir_graph);
182 block_v = get_Block_block_visited(b);
183 if (block_v >= irg_v) {
184 /* we computed the number of preds for this block and saved it in the
186 return block_v - irg_v;
188 /* compute the number of good predecessors */
189 res = irn_arity = get_irn_arity(b);
190 for (i = 0; i < irn_arity; i++)
191 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
192 /* save it in the flag. */
193 set_Block_block_visited(b, irg_v + res);
199 * Copies the node to the new obstack. The Ins of the new node point to
200 * the predecessors on the old obstack. For block/phi nodes not all
201 * predecessors might be copied. n->link points to the new node.
202 * For Phi and Block nodes the function allocates in-arrays with an arity
203 * only for useful predecessors. The arity is determined by counting
204 * the non-bad predecessors of the block.
206 * @param n The node to be copied
207 * @param env if non-NULL, the node number attribute will be copied to the new node
209 * Note: Also used for loop unrolling.
211 static void copy_node(ir_node *n, void *env) {
214 ir_op *op = get_irn_op(n);
215 int copy_node_nr = env != NULL;
217 /* The end node looses it's flexible in array. This doesn't matter,
218 as dead node elimination builds End by hand, inlineing doesn't use
220 /* assert(op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
223 /* node copied already */
225 } else if (op == op_Block) {
227 new_arity = compute_new_arity(n);
228 n->attr.block.graph_arr = NULL;
230 block = get_nodes_block(n);
232 new_arity = compute_new_arity(block);
234 new_arity = get_irn_arity(n);
237 nn = new_ir_node(get_irn_dbg_info(n),
244 /* Copy the attributes. These might point to additional data. If this
245 was allocated on the old obstack the pointers now are dangling. This
246 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
247 copy_node_attr(n, nn);
248 new_backedge_info(nn);
252 /* for easier debugging, we want to copy the node numbers too */
253 nn->node_nr = n->node_nr;
258 hook_dead_node_elim_subst(current_ir_graph, n, nn);
262 * Copies new predecessors of old node to new node remembered in link.
263 * Spare the Bad predecessors of Phi and Block nodes.
266 copy_preds (ir_node *n, void *env) {
270 nn = get_new_node(n);
272 /* printf("\n old node: "); DDMSG2(n);
273 printf(" new node: "); DDMSG2(nn);
274 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
277 /* Don't copy Bad nodes. */
279 irn_arity = get_irn_arity(n);
280 for (i = 0; i < irn_arity; i++)
281 if (! is_Bad(get_irn_n(n, i))) {
282 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
283 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
286 /* repair the block visited flag from above misuse. Repair it in both
287 graphs so that the old one can still be used. */
288 set_Block_block_visited(nn, 0);
289 set_Block_block_visited(n, 0);
290 /* Local optimization could not merge two subsequent blocks if
291 in array contained Bads. Now it's possible.
292 We don't call optimize_in_place as it requires
293 that the fields in ir_graph are set properly. */
294 if ((get_opt_control_flow_straightening()) &&
295 (get_Block_n_cfgpreds(nn) == 1) &&
296 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
297 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
299 /* Jmp jumps into the block it is in -- deal self cycle. */
300 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
301 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
306 } else if (get_irn_op(n) == op_Phi) {
307 /* Don't copy node if corresponding predecessor in block is Bad.
308 The Block itself should not be Bad. */
309 block = get_nodes_block(n);
310 set_irn_n (nn, -1, get_new_node(block));
312 irn_arity = get_irn_arity(n);
313 for (i = 0; i < irn_arity; i++)
314 if (! is_Bad(get_irn_n(block, i))) {
315 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
316 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
319 /* If the pre walker reached this Phi after the post walker visited the
320 block block_visited is > 0. */
321 set_Block_block_visited(get_nodes_block(n), 0);
322 /* Compacting the Phi's ins might generate Phis with only one
324 if (get_irn_arity(nn) == 1)
325 exchange(nn, get_irn_n(nn, 0));
327 irn_arity = get_irn_arity(n);
328 for (i = -1; i < irn_arity; i++)
329 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
331 /* Now the new node is complete. We can add it to the hash table for CSE.
332 @@@ inlining aborts if we identify End. Why? */
333 if (get_irn_op(nn) != op_End)
334 add_identities (current_ir_graph->value_table, nn);
338 * Copies the graph recursively, compacts the keep-alives of the end node.
340 * @param irg the graph to be copied
341 * @param copy_node_nr If non-zero, the node number will be copied
343 static void copy_graph(ir_graph *irg, int copy_node_nr) {
344 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
345 ir_node *ka; /* keep alive */
348 oe = get_irg_end(irg);
349 /* copy the end node by hand, allocate dynamic in array! */
350 ne = new_ir_node(get_irn_dbg_info(oe),
357 /* Copy the attributes. Well, there might be some in the future... */
358 copy_node_attr(oe, ne);
359 set_new_node(oe, ne);
361 /* copy the Bad node */
362 ob = get_irg_bad(irg);
363 nb = new_ir_node(get_irn_dbg_info(ob),
370 set_new_node(ob, nb);
372 /* copy the NoMem node */
373 om = get_irg_no_mem(irg);
374 nm = new_ir_node(get_irn_dbg_info(om),
381 set_new_node(om, nm);
383 /* copy the live nodes */
384 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
385 /* copy_preds for the end node ... */
386 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
388 /*- ... and now the keep alives. -*/
389 /* First pick the not marked block nodes and walk them. We must pick these
390 first as else we will oversee blocks reachable from Phis. */
391 irn_arity = get_irn_arity(oe);
392 for (i = 0; i < irn_arity; i++) {
393 ka = get_irn_intra_n(oe, i);
395 (get_irn_visited(ka) < get_irg_visited(irg))) {
396 /* We must keep the block alive and copy everything reachable */
397 set_irg_visited(irg, get_irg_visited(irg)-1);
398 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
399 add_End_keepalive(ne, get_new_node(ka));
403 /* Now pick other nodes. Here we will keep all! */
404 irn_arity = get_irn_arity(oe);
405 for (i = 0; i < irn_arity; i++) {
406 ka = get_irn_intra_n(oe, i);
408 if (get_irn_visited(ka) < get_irg_visited(irg)) {
409 /* We didn't copy the node yet. */
410 set_irg_visited(irg, get_irg_visited(irg)-1);
411 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
413 add_End_keepalive(ne, get_new_node(ka));
417 /* start block sometimes only reached after keep alives */
418 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
419 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
423 * Copies the graph reachable from current_ir_graph->end to the obstack
424 * in current_ir_graph and fixes the environment.
425 * Then fixes the fields in current_ir_graph containing nodes of the
428 * @param copy_node_nr If non-zero, the node number will be copied
431 copy_graph_env (int copy_node_nr) {
432 ir_graph *irg = current_ir_graph;
433 ir_node *old_end, *n;
436 /* Not all nodes remembered in irg might be reachable
437 from the end node. Assure their link is set to NULL, so that
438 we can test whether new nodes have been computed. */
439 for (i = anchor_max - 1; i >= 0; --i)
441 set_new_node(irg->anchors[i], NULL);
443 /* we use the block walk flag for removing Bads from Blocks ins. */
444 inc_irg_block_visited(irg);
447 copy_graph(irg, copy_node_nr);
449 /* fix the fields in irg */
450 old_end = get_irg_end(irg);
451 set_irg_end (irg, get_new_node(old_end));
452 set_irg_end_except (irg, get_irg_end(irg));
453 set_irg_end_reg (irg, get_irg_end(irg));
455 set_irg_end_block (irg, get_new_node(get_irg_end_block(irg)));
457 n = get_irg_frame(irg);
458 if (!has_new_node(n)) {
459 copy_node (n, INT_TO_PTR(copy_node_nr));
462 n = get_irg_globals(irg);
463 if (!has_new_node(n)) {
464 copy_node (n, INT_TO_PTR(copy_node_nr));
467 n = get_irg_initial_mem(irg);
468 if (!has_new_node(n)) {
469 copy_node (n, INT_TO_PTR(copy_node_nr));
472 n = get_irg_args(irg);
473 if (!has_new_node(n)) {
474 copy_node (n, INT_TO_PTR(copy_node_nr));
477 n = get_irg_bad(irg);
478 if (!has_new_node(n)) {
479 copy_node(n, INT_TO_PTR(copy_node_nr));
482 n = get_irg_no_mem(irg);
483 if (!has_new_node(n)) {
484 copy_node(n, INT_TO_PTR(copy_node_nr));
487 set_irg_start (irg, get_new_node(get_irg_start(irg)));
488 set_irg_start_block(irg, get_new_node(get_irg_start_block(irg)));
489 set_irg_frame (irg, get_new_node(get_irg_frame(irg)));
490 set_irg_globals (irg, get_new_node(get_irg_globals(irg)));
491 set_irg_initial_mem(irg, get_new_node(get_irg_initial_mem(irg)));
492 set_irg_args (irg, get_new_node(get_irg_args(irg)));
493 set_irg_bad (irg, get_new_node(get_irg_bad(irg)));
494 set_irg_no_mem (irg, get_new_node(get_irg_no_mem(irg)));
498 * Copies all reachable nodes to a new obstack. Removes bad inputs
499 * from block nodes and the corresponding inputs from Phi nodes.
500 * Merges single exit blocks with single entry blocks and removes
502 * Adds all new nodes to a new hash table for CSE. Does not
503 * perform CSE, so the hash table might contain common subexpressions.
506 dead_node_elimination(ir_graph *irg) {
508 int rem_ipview = get_interprocedural_view();
509 struct obstack *graveyard_obst = NULL;
510 struct obstack *rebirth_obst = NULL;
512 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
513 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
515 /* inform statistics that we started a dead-node elimination run */
516 hook_dead_node_elim(irg, 1);
518 /* Remember external state of current_ir_graph. */
519 rem = current_ir_graph;
520 current_ir_graph = irg;
521 set_interprocedural_view(0);
523 assert(get_irg_phase_state(current_ir_graph) != phase_building);
525 /* Handle graph state */
526 free_callee_info(current_ir_graph);
527 free_irg_outs(current_ir_graph);
530 /* @@@ so far we loose loops when copying */
531 free_loop_information(current_ir_graph);
533 set_irg_doms_inconsistent(irg);
535 /* A quiet place, where the old obstack can rest in peace,
536 until it will be cremated. */
537 graveyard_obst = irg->obst;
539 /* A new obstack, where the reachable nodes will be copied to. */
540 rebirth_obst = xmalloc (sizeof(*rebirth_obst));
541 current_ir_graph->obst = rebirth_obst;
542 obstack_init (current_ir_graph->obst);
544 /* We also need a new hash table for cse */
545 del_identities (irg->value_table);
546 irg->value_table = new_identities ();
548 /* Copy the graph from the old to the new obstack */
551 /* Free memory from old unoptimized obstack */
552 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
553 xfree (graveyard_obst); /* ... then free it. */
555 /* inform statistics that the run is over */
556 hook_dead_node_elim(irg, 0);
558 current_ir_graph = rem;
559 set_interprocedural_view(rem_ipview);
564 * Relink bad predecessors of a block and store the old in array to the
565 * link field. This function is called by relink_bad_predecessors().
566 * The array of link field starts with the block operand at position 0.
567 * If block has bad predecessors, create a new in array without bad preds.
568 * Otherwise let in array untouched.
570 static void relink_bad_block_predecessors(ir_node *n, void *env) {
571 ir_node **new_in, *irn;
572 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
574 /* if link field of block is NULL, look for bad predecessors otherwise
575 this is already done */
576 if (get_irn_op(n) == op_Block &&
577 get_irn_link(n) == NULL) {
579 /* save old predecessors in link field (position 0 is the block operand)*/
580 set_irn_link(n, get_irn_in(n));
582 /* count predecessors without bad nodes */
583 old_irn_arity = get_irn_arity(n);
584 for (i = 0; i < old_irn_arity; i++)
585 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
587 /* arity changing: set new predecessors without bad nodes */
588 if (new_irn_arity < old_irn_arity) {
589 /* Get new predecessor array. We do not resize the array, as we must
590 keep the old one to update Phis. */
591 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
593 /* set new predecessors in array */
596 for (i = 0; i < old_irn_arity; i++) {
597 irn = get_irn_n(n, i);
599 new_in[new_irn_n] = irn;
600 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
604 //ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity);
605 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
608 } /* ir node has bad predecessors */
610 } /* Block is not relinked */
614 * Relinks Bad predecessors from Blocks and Phis called by walker
615 * remove_bad_predecesors(). If n is a Block, call
616 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
617 * function of Phi's Block. If this block has bad predecessors, relink preds
620 static void relink_bad_predecessors(ir_node *n, void *env) {
621 ir_node *block, **old_in;
622 int i, old_irn_arity, new_irn_arity;
624 /* relink bad predecessors of a block */
625 if (get_irn_op(n) == op_Block)
626 relink_bad_block_predecessors(n, env);
628 /* If Phi node relink its block and its predecessors */
629 if (get_irn_op(n) == op_Phi) {
631 /* Relink predecessors of phi's block */
632 block = get_nodes_block(n);
633 if (get_irn_link(block) == NULL)
634 relink_bad_block_predecessors(block, env);
636 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
637 old_irn_arity = ARR_LEN(old_in);
639 /* Relink Phi predecessors if count of predecessors changed */
640 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
641 /* set new predecessors in array
642 n->in[0] remains the same block */
644 for(i = 1; i < old_irn_arity; i++)
645 if (!is_Bad((ir_node *)old_in[i])) {
646 n->in[new_irn_arity] = n->in[i];
647 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
651 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
652 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
655 } /* n is a Phi node */
659 * Removes Bad Bad predecessors from Blocks and the corresponding
660 * inputs to Phi nodes as in dead_node_elimination but without
662 * On walking up set the link field to NULL, on walking down call
663 * relink_bad_predecessors() (This function stores the old in array
664 * to the link field and sets a new in array if arity of predecessors
667 void remove_bad_predecessors(ir_graph *irg) {
668 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
675 __)|_| | \_/ | \_/(/_ |_/\__|__
677 The following stuff implements a facility that automatically patches
678 registered ir_node pointers to the new node when a dead node elimination occurs.
681 struct _survive_dce_t {
685 hook_entry_t dead_node_elim;
686 hook_entry_t dead_node_elim_subst;
689 typedef struct _survive_dce_list_t {
690 struct _survive_dce_list_t *next;
692 } survive_dce_list_t;
694 static void dead_node_hook(void *context, ir_graph *irg, int start)
696 survive_dce_t *sd = context;
698 /* Create a new map before the dead node elimination is performed. */
700 sd->new_places = pmap_create_ex(pmap_count(sd->places));
703 /* Patch back all nodes if dead node elimination is over and something is to be done. */
705 pmap_destroy(sd->places);
706 sd->places = sd->new_places;
707 sd->new_places = NULL;
711 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw)
713 survive_dce_t *sd = context;
714 survive_dce_list_t *list = pmap_get(sd->places, old);
716 /* If the node is to be patched back, write the new address to all registered locations. */
718 survive_dce_list_t *p;
720 for(p = list; p; p = p->next)
723 pmap_insert(sd->new_places, nw, list);
728 * Make a new Survive DCE environment.
730 survive_dce_t *new_survive_dce(void)
732 survive_dce_t *res = xmalloc(sizeof(res[0]));
733 obstack_init(&res->obst);
734 res->places = pmap_create();
735 res->new_places = NULL;
737 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
738 res->dead_node_elim.context = res;
739 res->dead_node_elim.next = NULL;
741 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
742 res->dead_node_elim_subst.context = res;
743 res->dead_node_elim_subst.next = NULL;
745 register_hook(hook_dead_node_elim, &res->dead_node_elim);
746 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
751 * Free a Survive DCE environment.
753 void free_survive_dce(survive_dce_t *sd)
755 obstack_free(&sd->obst, NULL);
756 pmap_destroy(sd->places);
757 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
758 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
763 * Register a node pointer to be patched upon DCE.
764 * When DCE occurs, the node pointer specified by @p place will be
765 * patched to the new address of the node it is pointing to.
767 * @param sd The Survive DCE environment.
768 * @param place The address of the node pointer.
770 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place)
773 ir_node *irn = *place;
774 survive_dce_list_t *curr = pmap_get(sd->places, irn);
775 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw));
780 pmap_insert(sd->places, irn, nw);
784 /*--------------------------------------------------------------------*/
785 /* Functionality for inlining */
786 /*--------------------------------------------------------------------*/
789 * Copy node for inlineing. Updates attributes that change when
790 * inlineing but not for dead node elimination.
792 * Copies the node by calling copy_node() and then updates the entity if
793 * it's a local one. env must be a pointer of the frame type of the
794 * inlined procedure. The new entities must be in the link field of
798 copy_node_inline (ir_node *n, void *env) {
800 ir_type *frame_tp = (ir_type *)env;
803 if (get_irn_op(n) == op_Sel) {
804 nn = get_new_node (n);
806 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
807 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
809 } else if (get_irn_op(n) == op_Block) {
810 nn = get_new_node (n);
811 nn->attr.block.irg = current_ir_graph;
815 static void find_addr(ir_node *node, void *env)
817 if (get_irn_opcode(node) == iro_Proj) {
818 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
824 * currently, we cannot inline two cases:
825 * - call with compound arguments
826 * - graphs that take the address of a parameter
828 * check these conditions here
830 static int can_inline(ir_node *call, ir_graph *called_graph)
832 ir_type *call_type = get_Call_type(call);
833 int params, ress, i, res;
834 assert(is_Method_type(call_type));
836 params = get_method_n_params(call_type);
837 ress = get_method_n_ress(call_type);
840 for (i = 0; i < params; ++i) {
841 ir_type *p_type = get_method_param_type(call_type, i);
843 if (is_compound_type(p_type))
848 for (i = 0; i < ress; ++i) {
849 ir_type *r_type = get_method_res_type(call_type, i);
851 if (is_compound_type(r_type))
856 irg_walk_graph(called_graph, find_addr, NULL, &res);
861 int inline_method(ir_node *call, ir_graph *called_graph) {
863 ir_node *post_call, *post_bl;
865 ir_node *end, *end_bl;
869 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
871 ir_type *called_frame;
872 irg_inline_property prop = get_irg_inline_property(called_graph);
874 if ( (prop != irg_inline_forced) &&
875 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
877 /* Do not inline variadic functions. */
878 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
881 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
882 get_method_n_params(get_Call_type(call)));
885 * currently, we cannot inline two cases:
886 * - call with compound arguments
887 * - graphs that take the address of a parameter
889 if (! can_inline(call, called_graph))
892 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
893 rem_opt = get_opt_optimize();
896 /* Handle graph state */
897 assert(get_irg_phase_state(current_ir_graph) != phase_building);
898 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
899 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
900 set_irg_outs_inconsistent(current_ir_graph);
901 set_irg_extblk_inconsistent(current_ir_graph);
902 set_irg_doms_inconsistent(current_ir_graph);
903 set_irg_loopinfo_inconsistent(current_ir_graph);
904 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
906 /* -- Check preconditions -- */
907 assert(is_Call(call));
908 /* @@@ does not work for InterfaceIII.java after cgana
909 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
910 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
911 get_Call_type(call)));
913 assert(get_type_tpop(get_Call_type(call)) == type_method);
914 if (called_graph == current_ir_graph) {
915 set_optimize(rem_opt);
919 /* here we know we WILL inline, so inform the statistics */
920 hook_inline(call, called_graph);
922 /* -- Decide how to handle exception control flow: Is there a handler
923 for the Call node, or do we branch directly to End on an exception?
925 0 There is a handler.
927 2 Exception handling not represented in Firm. -- */
929 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
930 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
931 assert(get_irn_op(proj) == op_Proj);
932 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
933 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
935 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
936 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
937 else { exc_handling = 2; } /* !Mproj && !Xproj */
942 the procedure and later replaces the Start node of the called graph.
943 Post_call is the old Call node and collects the results of the called
944 graph. Both will end up being a tuple. -- */
945 post_bl = get_nodes_block(call);
946 set_irg_current_block(current_ir_graph, post_bl);
947 /* XxMxPxP of Start + parameter of Call */
948 in[pn_Start_X_initial_exec] = new_Jmp();
949 in[pn_Start_M] = get_Call_mem(call);
950 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
951 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
952 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
953 /* in[pn_Start_P_value_arg_base] = ??? */
954 pre_call = new_Tuple(5, in);
958 The new block gets the ins of the old block, pre_call and all its
959 predecessors and all Phi nodes. -- */
960 part_block(pre_call);
962 /* -- Prepare state for dead node elimination -- */
963 /* Visited flags in calling irg must be >= flag in called irg.
964 Else walker and arity computation will not work. */
965 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
966 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
967 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
968 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
969 /* Set pre_call as new Start node in link field of the start node of
970 calling graph and pre_calls block as new block for the start block
972 Further mark these nodes so that they are not visited by the
974 set_irn_link(get_irg_start(called_graph), pre_call);
975 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
976 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
977 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
978 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
979 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
981 /* Initialize for compaction of in arrays */
982 inc_irg_block_visited(current_ir_graph);
984 /* -- Replicate local entities of the called_graph -- */
985 /* copy the entities. */
986 called_frame = get_irg_frame_type(called_graph);
987 for (i = 0; i < get_class_n_members(called_frame); i++) {
988 entity *new_ent, *old_ent;
989 old_ent = get_class_member(called_frame, i);
990 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
991 set_entity_link(old_ent, new_ent);
994 /* visited is > than that of called graph. With this trick visited will
995 remain unchanged so that an outer walker, e.g., searching the call nodes
996 to inline, calling this inline will not visit the inlined nodes. */
997 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
999 /* -- Performing dead node elimination inlines the graph -- */
1000 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1002 /* @@@ endless loops are not copied!! -- they should be, I think... */
1003 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
1004 get_irg_frame_type(called_graph));
1006 /* Repair called_graph */
1007 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
1008 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
1009 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1011 /* -- Merge the end of the inlined procedure with the call site -- */
1012 /* We will turn the old Call node into a Tuple with the following
1015 0: Phi of all Memories of Return statements.
1016 1: Jmp from new Block that merges the control flow from all exception
1017 predecessors of the old end block.
1018 2: Tuple of all arguments.
1019 3: Phi of Exception memories.
1020 In case the old Call directly branches to End on an exception we don't
1021 need the block merging all exceptions nor the Phi of the exception
1025 /* -- Precompute some values -- */
1026 end_bl = get_new_node(get_irg_end_block(called_graph));
1027 end = get_new_node(get_irg_end(called_graph));
1028 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1029 n_res = get_method_n_ress(get_Call_type(call));
1031 res_pred = xmalloc (n_res * sizeof(*res_pred));
1032 cf_pred = xmalloc (arity * sizeof(*res_pred));
1034 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
1036 /* -- archive keepalives -- */
1037 irn_arity = get_irn_arity(end);
1038 for (i = 0; i < irn_arity; i++)
1039 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
1041 /* The new end node will die. We need not free as the in array is on the obstack:
1042 copy_node() only generated 'D' arrays. */
1044 /* -- Replace Return nodes by Jump nodes. -- */
1046 for (i = 0; i < arity; i++) {
1048 ret = get_irn_n(end_bl, i);
1049 if (is_Return(ret)) {
1050 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
1054 set_irn_in(post_bl, n_ret, cf_pred);
1056 /* -- Build a Tuple for all results of the method.
1057 Add Phi node if there was more than one Return. -- */
1058 turn_into_tuple(post_call, 4);
1059 /* First the Memory-Phi */
1061 for (i = 0; i < arity; i++) {
1062 ret = get_irn_n(end_bl, i);
1063 if (is_Return(ret)) {
1064 cf_pred[n_ret] = get_Return_mem(ret);
1068 phi = new_Phi(n_ret, cf_pred, mode_M);
1069 set_Tuple_pred(call, pn_Call_M_regular, phi);
1070 /* Conserve Phi-list for further inlinings -- but might be optimized */
1071 if (get_nodes_block(phi) == post_bl) {
1072 set_irn_link(phi, get_irn_link(post_bl));
1073 set_irn_link(post_bl, phi);
1075 /* Now the real results */
1077 for (j = 0; j < n_res; j++) {
1079 for (i = 0; i < arity; i++) {
1080 ret = get_irn_n(end_bl, i);
1081 if (get_irn_op(ret) == op_Return) {
1082 cf_pred[n_ret] = get_Return_res(ret, j);
1087 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1091 /* Conserve Phi-list for further inlinings -- but might be optimized */
1092 if (get_nodes_block(phi) == post_bl) {
1093 set_irn_link(phi, get_irn_link(post_bl));
1094 set_irn_link(post_bl, phi);
1097 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1099 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1101 /* Finally the exception control flow.
1102 We have two (three) possible situations:
1103 First if the Call branches to an exception handler: We need to add a Phi node to
1104 collect the memory containing the exception objects. Further we need
1105 to add another block to get a correct representation of this Phi. To
1106 this block we add a Jmp that resolves into the X output of the Call
1107 when the Call is turned into a tuple.
1108 Second the Call branches to End, the exception is not handled. Just
1109 add all inlined exception branches to the End node.
1110 Third: there is no Exception edge at all. Handle as case two. */
1111 if (exc_handling == 0) {
1113 for (i = 0; i < arity; i++) {
1115 ret = get_irn_n(end_bl, i);
1116 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1117 cf_pred[n_exc] = ret;
1122 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1123 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1124 /* The Phi for the memories with the exception objects */
1126 for (i = 0; i < arity; i++) {
1128 ret = skip_Proj(get_irn_n(end_bl, i));
1130 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1132 } else if (is_fragile_op(ret)) {
1133 /* We rely that all cfops have the memory output at the same position. */
1134 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1136 } else if (get_irn_op(ret) == op_Raise) {
1137 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1141 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1143 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1144 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1147 ir_node *main_end_bl;
1148 int main_end_bl_arity;
1149 ir_node **end_preds;
1151 /* assert(exc_handling == 1 || no exceptions. ) */
1153 for (i = 0; i < arity; i++) {
1154 ir_node *ret = get_irn_n(end_bl, i);
1156 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1157 cf_pred[n_exc] = ret;
1161 main_end_bl = get_irg_end_block(current_ir_graph);
1162 main_end_bl_arity = get_irn_arity(main_end_bl);
1163 end_preds = xmalloc ((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1165 for (i = 0; i < main_end_bl_arity; ++i)
1166 end_preds[i] = get_irn_n(main_end_bl, i);
1167 for (i = 0; i < n_exc; ++i)
1168 end_preds[main_end_bl_arity + i] = cf_pred[i];
1169 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1170 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1171 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1177 #if 0 /* old. now better, correcter, faster implementation. */
1179 /* -- If the exception control flow from the inlined Call directly
1180 branched to the end block we now have the following control
1181 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
1182 remove the Jmp along with it's empty block and add Jmp's
1183 predecessors as predecessors of this end block. No problem if
1184 there is no exception, because then branches Bad to End which
1186 @@@ can't we know this beforehand: by getting the Proj(1) from
1187 the Call link list and checking whether it goes to Proj. */
1188 /* find the problematic predecessor of the end block. */
1189 end_bl = get_irg_end_block(current_ir_graph);
1190 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
1191 cf_op = get_Block_cfgpred(end_bl, i);
1192 if (get_irn_op(cf_op) == op_Proj) {
1193 cf_op = get_Proj_pred(cf_op);
1194 if ((get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
1195 /* There are unoptimized tuples from inlineing before when no exc */
1196 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
1197 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
1198 assert(get_irn_op(cf_op) == op_Jmp);
1204 if (i < get_Block_n_cfgpreds(end_bl)) {
1205 bl = get_nodes_block(cf_op);
1206 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
1207 cf_pred = xmalloc (arity * sizeof(*cf_pred));
1208 for (j = 0; j < i; j++)
1209 cf_pred[j] = get_Block_cfgpred(end_bl, j);
1210 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
1211 cf_pred[j] = get_Block_cfgpred(bl, j-i);
1212 for (j = j; j < arity; j++)
1213 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
1214 set_irn_in(end_bl, arity, cf_pred);
1216 /* Remove the exception pred from post-call Tuple. */
1217 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1222 /* -- Turn CSE back on. -- */
1223 set_optimize(rem_opt);
1228 /********************************************************************/
1229 /* Apply inlineing to small methods. */
1230 /********************************************************************/
1232 /* It makes no sense to inline too many calls in one procedure. Anyways,
1233 I didn't get a version with NEW_ARR_F to run. */
1234 #define MAX_INLINE 1024
1237 * environment for inlining small irgs
1239 typedef struct _inline_env_t {
1241 ir_node *calls[MAX_INLINE];
1245 * Returns the irg called from a Call node. If the irg is not
1246 * known, NULL is returned.
1248 static ir_graph *get_call_called_irg(ir_node *call) {
1250 ir_graph *called_irg = NULL;
1252 assert(is_Call(call));
1254 addr = get_Call_ptr(call);
1255 if ((get_irn_op(addr) == op_SymConst) && (get_SymConst_kind (addr) == symconst_addr_ent)) {
1256 called_irg = get_entity_irg(get_SymConst_entity(addr));
1262 static void collect_calls(ir_node *call, void *env) {
1265 if (! is_Call(call)) return;
1267 addr = get_Call_ptr(call);
1269 if (get_irn_op(addr) == op_SymConst) {
1270 if (get_SymConst_kind(addr) == symconst_addr_ent) {
1271 ir_graph *called_irg = get_entity_irg(get_SymConst_entity(addr));
1272 inline_env_t *ienv = (inline_env_t *)env;
1273 if (called_irg && ienv->pos < MAX_INLINE) {
1274 /* The Call node calls a locally defined method. Remember to inline. */
1275 ienv->calls[ienv->pos++] = call;
1282 * Inlines all small methods at call sites where the called address comes
1283 * from a Const node that references the entity representing the called
1285 * The size argument is a rough measure for the code size of the method:
1286 * Methods where the obstack containing the firm graph is smaller than
1289 void inline_small_irgs(ir_graph *irg, int size) {
1291 ir_graph *rem = current_ir_graph;
1292 inline_env_t env /* = {0, NULL}*/;
1294 if (!(get_opt_optimize() && get_opt_inline())) return;
1296 current_ir_graph = irg;
1297 /* Handle graph state */
1298 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1299 free_callee_info(current_ir_graph);
1301 /* Find Call nodes to inline.
1302 (We can not inline during a walk of the graph, as inlineing the same
1303 method several times changes the visited flag of the walked graph:
1304 after the first inlineing visited of the callee equals visited of
1305 the caller. With the next inlineing both are increased.) */
1307 irg_walk(get_irg_end(irg), NULL, collect_calls, &env);
1309 if ((env.pos > 0) && (env.pos < MAX_INLINE)) {
1310 /* There are calls to inline */
1311 collect_phiprojs(irg);
1312 for (i = 0; i < env.pos; i++) {
1314 callee = get_entity_irg(get_SymConst_entity(get_Call_ptr(env.calls[i])));
1315 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1316 (get_irg_inline_property(callee) == irg_inline_forced)) {
1317 inline_method(env.calls[i], callee);
1322 current_ir_graph = rem;
1326 * Environment for inlining irgs.
1329 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1330 int n_nodes_orig; /**< for statistics */
1331 eset *call_nodes; /**< All call nodes in this graph */
1333 int n_call_nodes_orig; /**< for statistics */
1334 int n_callers; /**< Number of known graphs that call this graphs. */
1335 int n_callers_orig; /**< for statistics */
1339 * Allocate a new environment for inlining.
1341 static inline_irg_env *new_inline_irg_env(void) {
1342 inline_irg_env *env = xmalloc(sizeof(*env));
1343 env->n_nodes = -2; /* do not count count Start, End */
1344 env->n_nodes_orig = -2; /* do not count Start, End */
1345 env->call_nodes = eset_create();
1346 env->n_call_nodes = 0;
1347 env->n_call_nodes_orig = 0;
1349 env->n_callers_orig = 0;
1354 * destroy an environment for inlining.
1356 static void free_inline_irg_env(inline_irg_env *env) {
1357 eset_destroy(env->call_nodes);
1362 * post-walker: collect all calls in the inline-environment
1363 * of a graph and sum some statistics.
1365 static void collect_calls2(ir_node *call, void *env) {
1366 inline_irg_env *x = (inline_irg_env *)env;
1367 ir_op *op = get_irn_op(call);
1370 /* count meaningful nodes in irg */
1371 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1376 if (op != op_Call) return;
1378 /* collect all call nodes */
1379 eset_insert(x->call_nodes, call);
1381 x->n_call_nodes_orig++;
1383 /* count all static callers */
1384 callee = get_call_called_irg(call);
1386 inline_irg_env *callee_env = get_irg_link(callee);
1387 callee_env->n_callers++;
1388 callee_env->n_callers_orig++;
1393 * Returns TRUE if the number of callers in 0 in the irg's environment,
1394 * hence this irg is a leave.
1396 INLINE static int is_leave(ir_graph *irg) {
1397 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1401 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1403 INLINE static int is_smaller(ir_graph *callee, int size) {
1404 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1409 * Inlines small leave methods at call sites where the called address comes
1410 * from a Const node that references the entity representing the called
1412 * The size argument is a rough measure for the code size of the method:
1413 * Methods where the obstack containing the firm graph is smaller than
1416 void inline_leave_functions(int maxsize, int leavesize, int size) {
1417 inline_irg_env *env;
1418 int i, n_irgs = get_irp_n_irgs();
1419 ir_graph *rem = current_ir_graph;
1422 if (!(get_opt_optimize() && get_opt_inline())) return;
1424 /* extend all irgs by a temporary data structure for inlining. */
1425 for (i = 0; i < n_irgs; ++i)
1426 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1428 /* Precompute information in temporary data structure. */
1429 for (i = 0; i < n_irgs; ++i) {
1430 current_ir_graph = get_irp_irg(i);
1431 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1432 free_callee_info(current_ir_graph);
1434 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1435 get_irg_link(current_ir_graph));
1438 /* -- and now inline. -- */
1440 /* Inline leaves recursively -- we might construct new leaves. */
1441 while (did_inline) {
1444 for (i = 0; i < n_irgs; ++i) {
1446 int phiproj_computed = 0;
1448 current_ir_graph = get_irp_irg(i);
1449 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1451 for (call = eset_first(env->call_nodes); call; call = eset_next(env->call_nodes)) {
1454 if (get_irn_op(call) == op_Tuple) continue; /* We already have inlined this call. */
1455 callee = get_call_called_irg(call);
1457 if (env->n_nodes > maxsize) continue; // break;
1459 if (callee && (is_leave(callee) && is_smaller(callee, leavesize))) {
1460 if (!phiproj_computed) {
1461 phiproj_computed = 1;
1462 collect_phiprojs(current_ir_graph);
1464 did_inline = inline_method(call, callee);
1467 /* Do some statistics */
1468 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1469 env->n_call_nodes --;
1470 env->n_nodes += callee_env->n_nodes;
1471 callee_env->n_callers--;
1478 /* inline other small functions. */
1479 for (i = 0; i < n_irgs; ++i) {
1482 int phiproj_computed = 0;
1484 current_ir_graph = get_irp_irg(i);
1485 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1487 /* we can not walk and change a set, nor remove from it.
1489 walkset = env->call_nodes;
1490 env->call_nodes = eset_create();
1491 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1494 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1495 callee = get_call_called_irg(call);
1498 ((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1499 (get_irg_inline_property(callee) == irg_inline_forced))) {
1500 if (!phiproj_computed) {
1501 phiproj_computed = 1;
1502 collect_phiprojs(current_ir_graph);
1504 if (inline_method(call, callee)) {
1505 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1506 env->n_call_nodes--;
1507 eset_insert_all(env->call_nodes, callee_env->call_nodes); /* @@@ ??? This are the wrong nodes !? Not the copied ones. */
1508 env->n_call_nodes += callee_env->n_call_nodes;
1509 env->n_nodes += callee_env->n_nodes;
1510 callee_env->n_callers--;
1513 eset_insert(env->call_nodes, call);
1516 eset_destroy(walkset);
1519 for (i = 0; i < n_irgs; ++i) {
1520 current_ir_graph = get_irp_irg(i);
1522 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1523 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1524 (env->n_callers_orig != env->n_callers))
1525 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1526 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1527 env->n_callers_orig, env->n_callers,
1528 get_entity_name(get_irg_entity(current_ir_graph)));
1530 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1533 current_ir_graph = rem;
1536 /*******************************************************************/
1537 /* Code Placement. Pins all floating nodes to a block where they */
1538 /* will be executed only if needed. */
1539 /*******************************************************************/
1542 * Returns non-zero, is a block is not reachable from Start.
1544 * @param block the block to test
1547 is_Block_unreachable(ir_node *block) {
1548 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1552 * Find the earliest correct block for N. --- Place N into the
1553 * same Block as its dominance-deepest Input.
1555 * We have to avoid calls to get_nodes_block() here
1556 * because the graph is floating.
1558 * move_out_of_loops() expects that place_floats_early() have placed
1559 * all "living" nodes into a living block. That's why we must
1560 * move nodes in dead block with "live" successors into a valid
1562 * We move them just into the same block as it's successor (or
1563 * in case of a Phi into the effective use block). For Phi successors,
1564 * this may still be a dead block, but then there is no real use, as
1565 * the control flow will be dead later.
1568 place_floats_early(ir_node *n, pdeq *worklist)
1572 /* we must not run into an infinite loop */
1573 assert(irn_not_visited(n));
1574 mark_irn_visited(n);
1576 /* Place floating nodes. */
1577 if (get_irn_pinned(n) == op_pin_state_floats) {
1578 ir_node *curr_block = get_irn_n(n, -1);
1579 int in_dead_block = is_Block_unreachable(curr_block);
1581 ir_node *b = NULL; /* The block to place this node in */
1583 assert(get_irn_op(n) != op_Block);
1585 if ((get_irn_op(n) == op_Const) ||
1586 (get_irn_op(n) == op_SymConst) ||
1588 (get_irn_op(n) == op_Unknown)) {
1589 /* These nodes will not be placed by the loop below. */
1590 b = get_irg_start_block(current_ir_graph);
1594 /* find the block for this node. */
1595 irn_arity = get_irn_arity(n);
1596 for (i = 0; i < irn_arity; i++) {
1597 ir_node *pred = get_irn_n(n, i);
1598 ir_node *pred_block;
1600 if ((irn_not_visited(pred))
1601 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1604 * If the current node is NOT in a dead block, but one of its
1605 * predecessors is, we must move the predecessor to a live block.
1606 * Such thing can happen, if global CSE chose a node from a dead block.
1607 * We move it simple to our block.
1608 * Note that neither Phi nor End nodes are floating, so we don't
1609 * need to handle them here.
1611 if (! in_dead_block) {
1612 if (get_irn_pinned(pred) == op_pin_state_floats &&
1613 is_Block_unreachable(get_irn_n(pred, -1)))
1614 set_nodes_block(pred, curr_block);
1616 place_floats_early(pred, worklist);
1620 * A node in the Bad block must stay in the bad block,
1621 * so don't compute a new block for it.
1626 /* Because all loops contain at least one op_pin_state_pinned node, now all
1627 our inputs are either op_pin_state_pinned or place_early() has already
1628 been finished on them. We do not have any unfinished inputs! */
1629 pred_block = get_irn_n(pred, -1);
1630 if ((!is_Block_dead(pred_block)) &&
1631 (get_Block_dom_depth(pred_block) > depth)) {
1633 depth = get_Block_dom_depth(pred_block);
1635 /* Avoid that the node is placed in the Start block */
1636 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)) {
1637 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1638 assert(b != get_irg_start_block(current_ir_graph));
1643 set_nodes_block(n, b);
1647 * Add predecessors of non floating nodes and non-floating predecessors
1648 * of floating nodes to worklist and fix their blocks if the are in dead block.
1650 irn_arity = get_irn_arity(n);
1652 if (get_irn_op(n) == op_End) {
1654 * Simplest case: End node. Predecessors are keep-alives,
1655 * no need to move out of dead block.
1657 for (i = -1; i < irn_arity; ++i) {
1658 ir_node *pred = get_irn_n(n, i);
1659 if (irn_not_visited(pred))
1660 pdeq_putr(worklist, pred);
1663 else if (is_Block(n)) {
1665 * Blocks: Predecessors are control flow, no need to move
1666 * them out of dead block.
1668 for (i = irn_arity - 1; i >= 0; --i) {
1669 ir_node *pred = get_irn_n(n, i);
1670 if (irn_not_visited(pred))
1671 pdeq_putr(worklist, pred);
1674 else if (is_Phi(n)) {
1676 ir_node *curr_block = get_irn_n(n, -1);
1677 int in_dead_block = is_Block_unreachable(curr_block);
1680 * Phi nodes: move nodes from dead blocks into the effective use
1681 * of the Phi-input if the Phi is not in a bad block.
1683 pred = get_irn_n(n, -1);
1684 if (irn_not_visited(pred))
1685 pdeq_putr(worklist, pred);
1687 for (i = irn_arity - 1; i >= 0; --i) {
1688 ir_node *pred = get_irn_n(n, i);
1690 if (irn_not_visited(pred)) {
1691 if (! in_dead_block &&
1692 get_irn_pinned(pred) == op_pin_state_floats &&
1693 is_Block_unreachable(get_irn_n(pred, -1))) {
1694 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1696 pdeq_putr(worklist, pred);
1702 ir_node *curr_block = get_irn_n(n, -1);
1703 int in_dead_block = is_Block_unreachable(curr_block);
1706 * All other nodes: move nodes from dead blocks into the same block.
1708 pred = get_irn_n(n, -1);
1709 if (irn_not_visited(pred))
1710 pdeq_putr(worklist, pred);
1712 for (i = irn_arity - 1; i >= 0; --i) {
1713 ir_node *pred = get_irn_n(n, i);
1715 if (irn_not_visited(pred)) {
1716 if (! in_dead_block &&
1717 get_irn_pinned(pred) == op_pin_state_floats &&
1718 is_Block_unreachable(get_irn_n(pred, -1))) {
1719 set_nodes_block(pred, curr_block);
1721 pdeq_putr(worklist, pred);
1728 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1729 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1730 * places all floating nodes reachable from its argument through floating
1731 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1733 static INLINE void place_early(pdeq *worklist) {
1735 inc_irg_visited(current_ir_graph);
1737 /* this inits the worklist */
1738 place_floats_early(get_irg_end(current_ir_graph), worklist);
1740 /* Work the content of the worklist. */
1741 while (!pdeq_empty(worklist)) {
1742 ir_node *n = pdeq_getl(worklist);
1743 if (irn_not_visited(n))
1744 place_floats_early(n, worklist);
1747 set_irg_outs_inconsistent(current_ir_graph);
1748 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
1752 * Compute the deepest common ancestor of block and dca.
1754 static ir_node *calc_dca(ir_node *dca, ir_node *block)
1758 /* we do not want to place nodes in dead blocks */
1759 if (is_Block_dead(block))
1762 /* We found a first legal placement. */
1763 if (!dca) return block;
1765 /* Find a placement that is dominates both, dca and block. */
1766 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1767 block = get_Block_idom(block);
1769 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1770 dca = get_Block_idom(dca);
1773 while (block != dca)
1774 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1779 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1780 * I.e., DCA is the block where we might place PRODUCER.
1781 * A data flow edge points from producer to consumer.
1784 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer)
1786 ir_node *block = NULL;
1788 /* Compute the latest block into which we can place a node so that it is
1790 if (get_irn_op(consumer) == op_Phi) {
1791 /* our consumer is a Phi-node, the effective use is in all those
1792 blocks through which the Phi-node reaches producer */
1794 ir_node *phi_block = get_nodes_block(consumer);
1795 irn_arity = get_irn_arity(consumer);
1797 for (i = 0; i < irn_arity; i++) {
1798 if (get_irn_n(consumer, i) == producer) {
1799 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1801 if (! is_Block_unreachable(new_block))
1802 block = calc_dca(block, new_block);
1807 block = get_irn_n(producer, -1);
1810 assert(is_no_Block(consumer));
1811 block = get_nodes_block(consumer);
1814 /* Compute the deepest common ancestor of block and dca. */
1815 return calc_dca(dca, block);
1818 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1820 static INLINE int get_irn_loop_depth(ir_node *n) {
1821 return get_loop_depth(get_irn_loop(n));
1825 * Move n to a block with less loop depth than it's current block. The
1826 * new block must be dominated by early.
1828 * @param n the node that should be moved
1829 * @param early the earliest block we can n move to
1832 move_out_of_loops (ir_node *n, ir_node *early)
1834 ir_node *best, *dca;
1838 /* Find the region deepest in the dominator tree dominating
1839 dca with the least loop nesting depth, but still dominated
1840 by our early placement. */
1841 dca = get_nodes_block(n);
1844 while (dca != early) {
1845 dca = get_Block_idom(dca);
1846 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1847 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1851 if (best != get_nodes_block(n)) {
1853 printf("Moving out of loop: "); DDMN(n);
1854 printf(" Outermost block: "); DDMN(early);
1855 printf(" Best block: "); DDMN(best);
1856 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1858 set_nodes_block(n, best);
1863 * Find the latest legal block for N and place N into the
1864 * `optimal' Block between the latest and earliest legal block.
1865 * The `optimal' block is the dominance-deepest block of those
1866 * with the least loop-nesting-depth. This places N out of as many
1867 * loops as possible and then makes it as control dependent as
1871 place_floats_late(ir_node *n, pdeq *worklist)
1876 assert(irn_not_visited(n)); /* no multiple placement */
1878 mark_irn_visited(n);
1880 /* no need to place block nodes, control nodes are already placed. */
1881 if ((get_irn_op(n) != op_Block) &&
1883 (get_irn_mode(n) != mode_X)) {
1884 /* Remember the early_blk placement of this block to move it
1885 out of loop no further than the early_blk placement. */
1886 early_blk = get_irn_n(n, -1);
1889 * BEWARE: Here we also get code, that is live, but
1890 * was in a dead block. If the node is life, but because
1891 * of CSE in a dead block, we still might need it.
1894 /* Assure that our users are all placed, except the Phi-nodes.
1895 --- Each data flow cycle contains at least one Phi-node. We
1896 have to break the `user has to be placed before the
1897 producer' dependence cycle and the Phi-nodes are the
1898 place to do so, because we need to base our placement on the
1899 final region of our users, which is OK with Phi-nodes, as they
1900 are op_pin_state_pinned, and they never have to be placed after a
1901 producer of one of their inputs in the same block anyway. */
1902 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
1903 ir_node *succ = get_irn_out(n, i);
1904 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
1905 place_floats_late(succ, worklist);
1908 if (! is_Block_dead(early_blk)) {
1909 /* do only move things that where not dead */
1911 /* We have to determine the final block of this node... except for
1913 if ((get_irn_pinned(n) == op_pin_state_floats) &&
1914 (get_irn_op(n) != op_Const) &&
1915 (get_irn_op(n) != op_SymConst)) {
1916 ir_node *dca = NULL; /* deepest common ancestor in the
1917 dominator tree of all nodes'
1918 blocks depending on us; our final
1919 placement has to dominate DCA. */
1920 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
1921 ir_node *succ = get_irn_out(n, i);
1924 if (get_irn_op(succ) == op_End) {
1926 * This consumer is the End node, a keep alive edge.
1927 * This is not a real consumer, so we ignore it
1932 /* ignore if succ is in dead code */
1933 succ_blk = get_irn_n(succ, -1);
1934 if (is_Block_unreachable(succ_blk))
1936 dca = consumer_dom_dca(dca, succ, n);
1939 set_nodes_block(n, dca);
1940 move_out_of_loops(n, early_blk);
1946 /* Add predecessors of all non-floating nodes on list. (Those of floating
1947 nodes are placed already and therefore are marked.) */
1948 for (i = 0; i < get_irn_n_outs(n); i++) {
1949 ir_node *succ = get_irn_out(n, i);
1950 if (irn_not_visited(get_irn_out(n, i))) {
1951 pdeq_putr(worklist, succ);
1956 static INLINE void place_late(pdeq *worklist) {
1958 inc_irg_visited(current_ir_graph);
1960 /* This fills the worklist initially. */
1961 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1963 /* And now empty the worklist again... */
1964 while (!pdeq_empty(worklist)) {
1965 ir_node *n = pdeq_getl(worklist);
1966 if (irn_not_visited(n))
1967 place_floats_late(n, worklist);
1971 void place_code(ir_graph *irg) {
1973 ir_graph *rem = current_ir_graph;
1975 current_ir_graph = irg;
1977 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1979 /* Handle graph state */
1980 assert(get_irg_phase_state(irg) != phase_building);
1981 if (get_irg_dom_state(irg) != dom_consistent)
1984 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1985 free_loop_information(irg);
1986 construct_backedges(irg);
1989 /* Place all floating nodes as early as possible. This guarantees
1990 a legal code placement. */
1991 worklist = new_pdeq();
1992 place_early(worklist);
1994 /* place_early() invalidates the outs, place_late needs them. */
1995 compute_irg_outs(irg);
1997 /* Now move the nodes down in the dominator tree. This reduces the
1998 unnecessary executions of the node. */
1999 place_late(worklist);
2001 set_irg_outs_inconsistent(current_ir_graph);
2002 set_irg_loopinfo_inconsistent(current_ir_graph);
2004 current_ir_graph = rem;
2008 * Called by walker of remove_critical_cf_edges().
2010 * Place an empty block to an edge between a blocks of multiple
2011 * predecessors and a block of multiple successors.
2014 * @param env Environment of walker. The changed field.
2016 static void walk_critical_cf_edges(ir_node *n, void *env) {
2018 ir_node *pre, *block, *jmp;
2021 /* Block has multiple predecessors */
2022 if (is_Block(n) && (get_irn_arity(n) > 1)) {
2023 if (n == get_irg_end_block(current_ir_graph))
2024 return; /* No use to add a block here. */
2026 arity = get_irn_arity(n);
2027 for (i=0; i<arity; i++) {
2028 pre = get_irn_n(n, i);
2029 /* Predecessor has multiple successors. Insert new control flow edge. */
2030 if (op_Raise != get_irn_op(skip_Proj(pre))) {
2031 /* set predecessor of new block */
2032 block = new_Block(1, &pre);
2033 /* insert new jmp node to new block */
2034 set_cur_block(block);
2037 /* set successor of new block */
2038 set_irn_n(n, i, jmp);
2040 } /* predecessor has multiple successors */
2041 } /* for all predecessors */
2042 } /* n is a block */
2045 void remove_critical_cf_edges(ir_graph *irg) {
2047 irg_walk_graph(irg, NULL, walk_critical_cf_edges, &changed);
2050 /* control flow changed */
2051 set_irg_outs_inconsistent(irg);
2052 set_irg_extblk_inconsistent(irg);
2053 set_irg_doms_inconsistent(current_ir_graph);
2054 set_irg_loopinfo_inconsistent(current_ir_graph);