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, Michael Beck
9 * Copyright: (c) 1998-2006 Universität Karlsruhe
10 * Licence: This file protected by GPL - GNU GENERAL PUBLIC LICENSE.
19 #include "irgraph_t.h"
32 #include "pdeq.h" /* Fuer code placement */
37 #include "irbackedge_t.h"
44 #include "iredges_t.h"
47 /*------------------------------------------------------------------*/
48 /* apply optimizations of iropt to all nodes. */
49 /*------------------------------------------------------------------*/
52 * A wrapper around optimize_inplace_2() to be called from a walker.
54 static void optimize_in_place_wrapper (ir_node *n, void *env) {
55 ir_node *optimized = optimize_in_place_2(n);
56 if (optimized != n) exchange (n, optimized);
60 * Do local optimizations for a node.
62 * @param n the IR-node where to start. Typically the End node
65 * @note current_ir_graph must be set
67 static INLINE void do_local_optimize(ir_node *n) {
68 /* Handle graph state */
69 assert(get_irg_phase_state(current_ir_graph) != phase_building);
71 if (get_opt_global_cse())
72 set_irg_pinned(current_ir_graph, op_pin_state_floats);
73 set_irg_outs_inconsistent(current_ir_graph);
74 set_irg_doms_inconsistent(current_ir_graph);
75 set_irg_loopinfo_inconsistent(current_ir_graph);
77 /* Clean the value_table in irg for the CSE. */
78 del_identities(current_ir_graph->value_table);
79 current_ir_graph->value_table = new_identities();
81 /* walk over the graph */
82 irg_walk(n, firm_clear_link, optimize_in_place_wrapper, NULL);
85 /* Applies local optimizations (see iropt.h) to all nodes reachable from node n */
86 void local_optimize_node(ir_node *n) {
87 ir_graph *rem = current_ir_graph;
88 current_ir_graph = get_irn_irg(n);
92 current_ir_graph = rem;
96 * Block-Walker: uses dominance depth to mark dead blocks.
98 static void kill_dead_blocks(ir_node *block, void *env)
100 if (get_Block_dom_depth(block) < 0) {
102 * Note that the new dominance code correctly handles
103 * the End block, i.e. it is always reachable from Start
105 set_Block_dead(block);
109 /* Applies local optimizations (see iropt.h) to all nodes reachable from node n. */
110 void local_optimize_graph(ir_graph *irg) {
111 ir_graph *rem = current_ir_graph;
112 current_ir_graph = irg;
114 if (get_irg_dom_state(irg) == dom_consistent)
115 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
117 do_local_optimize(get_irg_end(irg));
119 current_ir_graph = rem;
123 * Enqueue all users of a node to a wait queue.
124 * Handles mode_T nodes.
126 static void enqueue_users(ir_node *n, pdeq *waitq) {
127 const ir_edge_t *edge;
129 foreach_out_edge(n, edge) {
130 ir_node *succ = get_edge_src_irn(edge);
132 if (get_irn_link(succ) != waitq) {
133 pdeq_putr(waitq, succ);
134 set_irn_link(succ, waitq);
136 if (get_irn_mode(succ) == mode_T) {
137 /* A mode_T node has Proj's. Because most optimizations
138 run on the Proj's we have to enqueue them also. */
139 enqueue_users(succ, waitq);
145 * Data flow optimization walker.
146 * Optimizes all nodes and enqueue it's users
149 static void opt_walker(ir_node *n, void *env) {
153 optimized = optimize_in_place_2(n);
154 set_irn_link(optimized, NULL);
156 if (optimized != n) {
157 enqueue_users(n, waitq);
158 exchange(n, optimized);
162 /* Applies local optimizations to all nodes in the graph until fixpoint. */
163 void optimize_graph_df(ir_graph *irg) {
164 pdeq *waitq = new_pdeq();
165 int state = edges_activated(irg);
166 ir_graph *rem = current_ir_graph;
168 current_ir_graph = irg;
173 if (get_opt_global_cse())
174 set_irg_pinned(current_ir_graph, op_pin_state_floats);
176 /* Clean the value_table in irg for the CSE. */
177 del_identities(irg->value_table);
178 irg->value_table = new_identities();
180 if (get_irg_dom_state(irg) == dom_consistent)
181 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
183 /* invalidate info */
184 set_irg_outs_inconsistent(irg);
185 set_irg_doms_inconsistent(irg);
186 set_irg_loopinfo_inconsistent(irg);
188 set_using_irn_link(irg);
190 /* walk over the graph */
191 irg_walk_graph(irg, NULL, opt_walker, waitq);
193 /* finish the wait queue */
194 while (! pdeq_empty(waitq)) {
195 ir_node *n = pdeq_getl(waitq);
197 opt_walker(n, waitq);
202 clear_using_irn_link(irg);
205 edges_deactivate(irg);
207 current_ir_graph = rem;
211 /*------------------------------------------------------------------*/
212 /* Routines for dead node elimination / copying garbage collection */
213 /* of the obstack. */
214 /*------------------------------------------------------------------*/
217 * Remember the new node in the old node by using a field all nodes have.
219 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
222 * Get this new node, before the old node is forgotten.
224 #define get_new_node(oldn) get_irn_link(oldn)
227 * Check if a new node was set.
229 #define has_new_node(n) (get_new_node(n) != NULL)
232 * We use the block_visited flag to mark that we have computed the
233 * number of useful predecessors for this block.
234 * Further we encode the new arity in this flag in the old blocks.
235 * Remembering the arity is useful, as it saves a lot of pointer
236 * accesses. This function is called for all Phi and Block nodes
240 compute_new_arity(ir_node *b) {
241 int i, res, irn_arity;
244 irg_v = get_irg_block_visited(current_ir_graph);
245 block_v = get_Block_block_visited(b);
246 if (block_v >= irg_v) {
247 /* we computed the number of preds for this block and saved it in the
249 return block_v - irg_v;
251 /* compute the number of good predecessors */
252 res = irn_arity = get_irn_arity(b);
253 for (i = 0; i < irn_arity; i++)
254 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
255 /* save it in the flag. */
256 set_Block_block_visited(b, irg_v + res);
262 * Copies the node to the new obstack. The Ins of the new node point to
263 * the predecessors on the old obstack. For block/phi nodes not all
264 * predecessors might be copied. n->link points to the new node.
265 * For Phi and Block nodes the function allocates in-arrays with an arity
266 * only for useful predecessors. The arity is determined by counting
267 * the non-bad predecessors of the block.
269 * @param n The node to be copied
270 * @param env if non-NULL, the node number attribute will be copied to the new node
272 * Note: Also used for loop unrolling.
274 static void copy_node(ir_node *n, void *env) {
277 ir_op *op = get_irn_op(n);
279 /* The end node looses it's flexible in array. This doesn't matter,
280 as dead node elimination builds End by hand, inlineing doesn't use
282 /* assert(op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
285 /* node copied already */
287 } else if (op == op_Block) {
289 new_arity = compute_new_arity(n);
290 n->attr.block.graph_arr = NULL;
292 block = get_nodes_block(n);
294 new_arity = compute_new_arity(block);
296 new_arity = get_irn_arity(n);
299 nn = new_ir_node(get_irn_dbg_info(n),
306 /* Copy the attributes. These might point to additional data. If this
307 was allocated on the old obstack the pointers now are dangling. This
308 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
309 copy_node_attr(n, nn);
313 int copy_node_nr = env != NULL;
315 /* for easier debugging, we want to copy the node numbers too */
316 nn->node_nr = n->node_nr;
322 hook_dead_node_elim_subst(current_ir_graph, n, nn);
326 * Copies new predecessors of old node to new node remembered in link.
327 * Spare the Bad predecessors of Phi and Block nodes.
330 copy_preds(ir_node *n, void *env) {
334 nn = get_new_node(n);
336 /* printf("\n old node: "); DDMSG2(n);
337 printf(" new node: "); DDMSG2(nn);
338 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
341 /* Don't copy Bad nodes. */
343 irn_arity = get_irn_arity(n);
344 for (i = 0; i < irn_arity; i++)
345 if (! is_Bad(get_irn_n(n, i))) {
346 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
347 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
350 /* repair the block visited flag from above misuse. Repair it in both
351 graphs so that the old one can still be used. */
352 set_Block_block_visited(nn, 0);
353 set_Block_block_visited(n, 0);
354 /* Local optimization could not merge two subsequent blocks if
355 in array contained Bads. Now it's possible.
356 We don't call optimize_in_place as it requires
357 that the fields in ir_graph are set properly. */
358 if ((get_opt_control_flow_straightening()) &&
359 (get_Block_n_cfgpreds(nn) == 1) &&
360 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
361 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
363 /* Jmp jumps into the block it is in -- deal self cycle. */
364 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
365 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
370 } else if (get_irn_op(n) == op_Phi) {
371 /* Don't copy node if corresponding predecessor in block is Bad.
372 The Block itself should not be Bad. */
373 block = get_nodes_block(n);
374 set_irn_n(nn, -1, get_new_node(block));
376 irn_arity = get_irn_arity(n);
377 for (i = 0; i < irn_arity; i++)
378 if (! is_Bad(get_irn_n(block, i))) {
379 set_irn_n(nn, j, get_new_node(get_irn_n(n, i)));
380 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
383 /* If the pre walker reached this Phi after the post walker visited the
384 block block_visited is > 0. */
385 set_Block_block_visited(get_nodes_block(n), 0);
386 /* Compacting the Phi's ins might generate Phis with only one
388 if (get_irn_arity(nn) == 1)
389 exchange(nn, get_irn_n(nn, 0));
391 irn_arity = get_irn_arity(n);
392 for (i = -1; i < irn_arity; i++)
393 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
395 /* Now the new node is complete. We can add it to the hash table for CSE.
396 @@@ inlining aborts if we identify End. Why? */
397 if (get_irn_op(nn) != op_End)
398 add_identities(current_ir_graph->value_table, nn);
402 * Copies the graph recursively, compacts the keep-alives of the end node.
404 * @param irg the graph to be copied
405 * @param copy_node_nr If non-zero, the node number will be copied
407 static void copy_graph(ir_graph *irg, int copy_node_nr) {
408 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
409 ir_node *ka; /* keep alive */
413 /* Some nodes must be copied by hand, sigh */
414 vfl = get_irg_visited(irg);
415 set_irg_visited(irg, vfl + 1);
417 oe = get_irg_end(irg);
418 mark_irn_visited(oe);
419 /* copy the end node by hand, allocate dynamic in array! */
420 ne = new_ir_node(get_irn_dbg_info(oe),
427 /* Copy the attributes. Well, there might be some in the future... */
428 copy_node_attr(oe, ne);
429 set_new_node(oe, ne);
431 /* copy the Bad node */
432 ob = get_irg_bad(irg);
433 mark_irn_visited(ob);
434 nb = new_ir_node(get_irn_dbg_info(ob),
441 copy_node_attr(ob, nb);
442 set_new_node(ob, nb);
444 /* copy the NoMem node */
445 om = get_irg_no_mem(irg);
446 mark_irn_visited(om);
447 nm = new_ir_node(get_irn_dbg_info(om),
454 copy_node_attr(om, nm);
455 set_new_node(om, nm);
457 /* copy the live nodes */
458 set_irg_visited(irg, vfl);
459 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
461 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
463 /* visit the anchors as well */
464 for (i = anchor_max - 1; i >= 0; --i) {
465 ir_node *n = irg->anchors[i];
467 if (n && (get_irn_visited(n) <= vfl)) {
468 set_irg_visited(irg, vfl);
469 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
473 /* copy_preds for the end node ... */
474 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
476 /*- ... and now the keep alives. -*/
477 /* First pick the not marked block nodes and walk them. We must pick these
478 first as else we will oversee blocks reachable from Phis. */
479 irn_arity = get_End_n_keepalives(oe);
480 for (i = 0; i < irn_arity; i++) {
481 ka = get_End_keepalive(oe, i);
483 if (get_irn_visited(ka) <= vfl) {
484 /* We must keep the block alive and copy everything reachable */
485 set_irg_visited(irg, vfl);
486 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
488 add_End_keepalive(ne, get_new_node(ka));
492 /* Now pick other nodes. Here we will keep all! */
493 irn_arity = get_End_n_keepalives(oe);
494 for (i = 0; i < irn_arity; i++) {
495 ka = get_End_keepalive(oe, i);
497 if (get_irn_visited(ka) <= vfl) {
498 /* We didn't copy the node yet. */
499 set_irg_visited(irg, vfl);
500 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
502 add_End_keepalive(ne, get_new_node(ka));
506 /* start block sometimes only reached after keep alives */
507 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
508 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
512 * Copies the graph reachable from current_ir_graph->end to the obstack
513 * in current_ir_graph and fixes the environment.
514 * Then fixes the fields in current_ir_graph containing nodes of the
517 * @param copy_node_nr If non-zero, the node number will be copied
520 copy_graph_env(int copy_node_nr) {
521 ir_graph *irg = current_ir_graph;
522 ir_node *old_end, *n;
525 /* remove end_except and end_reg nodes */
526 old_end = get_irg_end(irg);
527 set_irg_end_except (irg, old_end);
528 set_irg_end_reg (irg, old_end);
530 /* Not all nodes remembered in irg might be reachable
531 from the end node. Assure their link is set to NULL, so that
532 we can test whether new nodes have been computed. */
533 for (i = anchor_max - 1; i >= 0; --i) {
535 set_new_node(irg->anchors[i], NULL);
537 /* we use the block walk flag for removing Bads from Blocks ins. */
538 inc_irg_block_visited(irg);
541 copy_graph(irg, copy_node_nr);
543 /* fix the fields in irg */
544 old_end = get_irg_end(irg);
545 for (i = anchor_max - 1; i >= 0; --i) {
548 irg->anchors[i] = get_new_node(n);
554 * Copies all reachable nodes to a new obstack. Removes bad inputs
555 * from block nodes and the corresponding inputs from Phi nodes.
556 * Merges single exit blocks with single entry blocks and removes
558 * Adds all new nodes to a new hash table for CSE. Does not
559 * perform CSE, so the hash table might contain common subexpressions.
562 dead_node_elimination(ir_graph *irg) {
563 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
565 int rem_ipview = get_interprocedural_view();
566 struct obstack *graveyard_obst = NULL;
567 struct obstack *rebirth_obst = NULL;
568 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
570 /* inform statistics that we started a dead-node elimination run */
571 hook_dead_node_elim(irg, 1);
573 /* Remember external state of current_ir_graph. */
574 rem = current_ir_graph;
575 current_ir_graph = irg;
576 set_interprocedural_view(0);
578 assert(get_irg_phase_state(irg) != phase_building);
580 /* Handle graph state */
581 free_callee_info(irg);
585 /* @@@ so far we loose loops when copying */
586 free_loop_information(irg);
588 set_irg_doms_inconsistent(irg);
590 /* A quiet place, where the old obstack can rest in peace,
591 until it will be cremated. */
592 graveyard_obst = irg->obst;
594 /* A new obstack, where the reachable nodes will be copied to. */
595 rebirth_obst = xmalloc(sizeof(*rebirth_obst));
596 irg->obst = rebirth_obst;
597 obstack_init(irg->obst);
598 irg->last_node_idx = 0;
600 /* We also need a new value table for CSE */
601 del_identities(irg->value_table);
602 irg->value_table = new_identities();
604 /* Copy the graph from the old to the new obstack */
605 copy_graph_env(/*copy_node_nr=*/1);
607 /* Free memory from old unoptimized obstack */
608 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
609 xfree (graveyard_obst); /* ... then free it. */
611 /* inform statistics that the run is over */
612 hook_dead_node_elim(irg, 0);
614 current_ir_graph = rem;
615 set_interprocedural_view(rem_ipview);
620 * Relink bad predecessors of a block and store the old in array to the
621 * link field. This function is called by relink_bad_predecessors().
622 * The array of link field starts with the block operand at position 0.
623 * If block has bad predecessors, create a new in array without bad preds.
624 * Otherwise let in array untouched.
626 static void relink_bad_block_predecessors(ir_node *n, void *env) {
627 ir_node **new_in, *irn;
628 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
630 /* if link field of block is NULL, look for bad predecessors otherwise
631 this is already done */
632 if (get_irn_op(n) == op_Block &&
633 get_irn_link(n) == NULL) {
635 /* save old predecessors in link field (position 0 is the block operand)*/
636 set_irn_link(n, get_irn_in(n));
638 /* count predecessors without bad nodes */
639 old_irn_arity = get_irn_arity(n);
640 for (i = 0; i < old_irn_arity; i++)
641 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
643 /* arity changing: set new predecessors without bad nodes */
644 if (new_irn_arity < old_irn_arity) {
645 /* Get new predecessor array. We do not resize the array, as we must
646 keep the old one to update Phis. */
647 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
649 /* set new predecessors in array */
652 for (i = 0; i < old_irn_arity; i++) {
653 irn = get_irn_n(n, i);
655 new_in[new_irn_n] = irn;
656 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
660 //ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity);
661 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
664 } /* ir node has bad predecessors */
666 } /* Block is not relinked */
670 * Relinks Bad predecessors from Blocks and Phis called by walker
671 * remove_bad_predecesors(). If n is a Block, call
672 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
673 * function of Phi's Block. If this block has bad predecessors, relink preds
676 static void relink_bad_predecessors(ir_node *n, void *env) {
677 ir_node *block, **old_in;
678 int i, old_irn_arity, new_irn_arity;
680 /* relink bad predecessors of a block */
681 if (get_irn_op(n) == op_Block)
682 relink_bad_block_predecessors(n, env);
684 /* If Phi node relink its block and its predecessors */
685 if (get_irn_op(n) == op_Phi) {
687 /* Relink predecessors of phi's block */
688 block = get_nodes_block(n);
689 if (get_irn_link(block) == NULL)
690 relink_bad_block_predecessors(block, env);
692 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
693 old_irn_arity = ARR_LEN(old_in);
695 /* Relink Phi predecessors if count of predecessors changed */
696 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
697 /* set new predecessors in array
698 n->in[0] remains the same block */
700 for(i = 1; i < old_irn_arity; i++)
701 if (!is_Bad((ir_node *)old_in[i])) {
702 n->in[new_irn_arity] = n->in[i];
703 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
707 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
708 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
711 } /* n is a Phi node */
715 * Removes Bad Bad predecessors from Blocks and the corresponding
716 * inputs to Phi nodes as in dead_node_elimination but without
718 * On walking up set the link field to NULL, on walking down call
719 * relink_bad_predecessors() (This function stores the old in array
720 * to the link field and sets a new in array if arity of predecessors
723 void remove_bad_predecessors(ir_graph *irg) {
724 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
731 __)|_| | \_/ | \_/(/_ |_/\__|__
733 The following stuff implements a facility that automatically patches
734 registered ir_node pointers to the new node when a dead node elimination occurs.
737 struct _survive_dce_t {
741 hook_entry_t dead_node_elim;
742 hook_entry_t dead_node_elim_subst;
745 typedef struct _survive_dce_list_t {
746 struct _survive_dce_list_t *next;
748 } survive_dce_list_t;
750 static void dead_node_hook(void *context, ir_graph *irg, int start)
752 survive_dce_t *sd = context;
754 /* Create a new map before the dead node elimination is performed. */
756 sd->new_places = pmap_create_ex(pmap_count(sd->places));
759 /* Patch back all nodes if dead node elimination is over and something is to be done. */
761 pmap_destroy(sd->places);
762 sd->places = sd->new_places;
763 sd->new_places = NULL;
768 * Hook called when dead node elimination replaces old by nw.
770 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw)
772 survive_dce_t *sd = context;
773 survive_dce_list_t *list = pmap_get(sd->places, old);
775 /* If the node is to be patched back, write the new address to all registered locations. */
777 survive_dce_list_t *p;
779 for(p = list; p; p = p->next)
782 pmap_insert(sd->new_places, nw, list);
787 * Make a new Survive DCE environment.
789 survive_dce_t *new_survive_dce(void)
791 survive_dce_t *res = xmalloc(sizeof(res[0]));
792 obstack_init(&res->obst);
793 res->places = pmap_create();
794 res->new_places = NULL;
796 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
797 res->dead_node_elim.context = res;
798 res->dead_node_elim.next = NULL;
800 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
801 res->dead_node_elim_subst.context = res;
802 res->dead_node_elim_subst.next = NULL;
804 #ifndef FIRM_ENABLE_HOOKS
805 assert(0 && "need hooks enabled");
808 register_hook(hook_dead_node_elim, &res->dead_node_elim);
809 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
814 * Free a Survive DCE environment.
816 void free_survive_dce(survive_dce_t *sd)
818 obstack_free(&sd->obst, NULL);
819 pmap_destroy(sd->places);
820 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
821 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
826 * Register a node pointer to be patched upon DCE.
827 * When DCE occurs, the node pointer specified by @p place will be
828 * patched to the new address of the node it is pointing to.
830 * @param sd The Survive DCE environment.
831 * @param place The address of the node pointer.
833 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place)
836 ir_node *irn = *place;
837 survive_dce_list_t *curr = pmap_get(sd->places, irn);
838 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw[0]));
843 pmap_insert(sd->places, irn, nw);
847 /*--------------------------------------------------------------------*/
848 /* Functionality for inlining */
849 /*--------------------------------------------------------------------*/
852 * Copy node for inlineing. Updates attributes that change when
853 * inlineing but not for dead node elimination.
855 * Copies the node by calling copy_node() and then updates the entity if
856 * it's a local one. env must be a pointer of the frame type of the
857 * inlined procedure. The new entities must be in the link field of
861 copy_node_inline (ir_node *n, void *env) {
863 ir_type *frame_tp = (ir_type *)env;
866 if (get_irn_op(n) == op_Sel) {
867 nn = get_new_node (n);
869 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
870 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
872 } else if (get_irn_op(n) == op_Block) {
873 nn = get_new_node (n);
874 nn->attr.block.irg = current_ir_graph;
879 * Walker: checks if P_value_arg_base is used.
881 static void find_addr(ir_node *node, void *env) {
882 int *allow_inline = env;
883 if (is_Proj(node) && get_irn_op(get_Proj_pred(node)) == op_Start) {
884 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
890 * currently, we cannot inline two cases:
891 * - call with compound arguments
892 * - graphs that take the address of a parameter
894 * check these conditions here
896 static int can_inline(ir_node *call, ir_graph *called_graph)
898 ir_type *call_type = get_Call_type(call);
899 int params, ress, i, res;
900 assert(is_Method_type(call_type));
902 params = get_method_n_params(call_type);
903 ress = get_method_n_ress(call_type);
905 /* check parameters for compound arguments */
906 for (i = 0; i < params; ++i) {
907 ir_type *p_type = get_method_param_type(call_type, i);
909 if (is_compound_type(p_type))
913 /* check results for compound arguments */
914 for (i = 0; i < ress; ++i) {
915 ir_type *r_type = get_method_res_type(call_type, i);
917 if (is_compound_type(r_type))
922 irg_walk_graph(called_graph, find_addr, NULL, &res);
927 /* Inlines a method at the given call site. */
928 int inline_method(ir_node *call, ir_graph *called_graph) {
930 ir_node *post_call, *post_bl;
931 ir_node *in[pn_Start_max];
932 ir_node *end, *end_bl;
936 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
938 ir_type *called_frame;
939 irg_inline_property prop = get_irg_inline_property(called_graph);
941 if ( (prop < irg_inline_forced) &&
942 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
944 /* Do not inline variadic functions. */
945 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
948 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
949 get_method_n_params(get_Call_type(call)));
952 * currently, we cannot inline two cases:
953 * - call with compound arguments
954 * - graphs that take the address of a parameter
956 if (! can_inline(call, called_graph))
959 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
960 rem_opt = get_opt_optimize();
963 /* Handle graph state */
964 assert(get_irg_phase_state(current_ir_graph) != phase_building);
965 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
966 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
967 set_irg_outs_inconsistent(current_ir_graph);
968 set_irg_extblk_inconsistent(current_ir_graph);
969 set_irg_doms_inconsistent(current_ir_graph);
970 set_irg_loopinfo_inconsistent(current_ir_graph);
971 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
973 /* -- Check preconditions -- */
974 assert(is_Call(call));
975 /* @@@ does not work for InterfaceIII.java after cgana
976 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
977 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
978 get_Call_type(call)));
980 if (called_graph == current_ir_graph) {
981 set_optimize(rem_opt);
985 /* here we know we WILL inline, so inform the statistics */
986 hook_inline(call, called_graph);
988 /* -- Decide how to handle exception control flow: Is there a handler
989 for the Call node, or do we branch directly to End on an exception?
991 0 There is a handler.
993 2 Exception handling not represented in Firm. -- */
995 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
996 for (proj = get_irn_link(call); proj; proj = get_irn_link(proj)) {
997 assert(is_Proj(proj));
998 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
999 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
1001 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
1002 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
1003 else { exc_handling = 2; } /* !Mproj && !Xproj */
1008 the procedure and later replaces the Start node of the called graph.
1009 Post_call is the old Call node and collects the results of the called
1010 graph. Both will end up being a tuple. -- */
1011 post_bl = get_nodes_block(call);
1012 set_irg_current_block(current_ir_graph, post_bl);
1013 /* XxMxPxPxPxT of Start + parameter of Call */
1014 in[pn_Start_X_initial_exec] = new_Jmp();
1015 in[pn_Start_M] = get_Call_mem(call);
1016 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
1017 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
1018 in[pn_Start_P_tls] = get_irg_tls(current_ir_graph);
1019 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
1020 /* in[pn_Start_P_value_arg_base] = ??? */
1021 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
1022 pre_call = new_Tuple(pn_Start_max - 1, in);
1026 The new block gets the ins of the old block, pre_call and all its
1027 predecessors and all Phi nodes. -- */
1028 part_block(pre_call);
1030 /* -- Prepare state for dead node elimination -- */
1031 /* Visited flags in calling irg must be >= flag in called irg.
1032 Else walker and arity computation will not work. */
1033 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
1034 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
1035 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
1036 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
1037 /* Set pre_call as new Start node in link field of the start node of
1038 calling graph and pre_calls block as new block for the start block
1040 Further mark these nodes so that they are not visited by the
1042 set_irn_link(get_irg_start(called_graph), pre_call);
1043 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
1044 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1045 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
1046 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
1047 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
1049 /* Initialize for compaction of in arrays */
1050 inc_irg_block_visited(current_ir_graph);
1052 /* -- Replicate local entities of the called_graph -- */
1053 /* copy the entities. */
1054 called_frame = get_irg_frame_type(called_graph);
1055 for (i = 0; i < get_class_n_members(called_frame); i++) {
1056 ir_entity *new_ent, *old_ent;
1057 old_ent = get_class_member(called_frame, i);
1058 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
1059 set_entity_link(old_ent, new_ent);
1062 /* visited is > than that of called graph. With this trick visited will
1063 remain unchanged so that an outer walker, e.g., searching the call nodes
1064 to inline, calling this inline will not visit the inlined nodes. */
1065 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
1067 /* -- Performing dead node elimination inlines the graph -- */
1068 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1070 /* @@@ endless loops are not copied!! -- they should be, I think... */
1071 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
1072 get_irg_frame_type(called_graph));
1074 /* Repair called_graph */
1075 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
1076 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
1077 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1079 /* -- Merge the end of the inlined procedure with the call site -- */
1080 /* We will turn the old Call node into a Tuple with the following
1083 0: Phi of all Memories of Return statements.
1084 1: Jmp from new Block that merges the control flow from all exception
1085 predecessors of the old end block.
1086 2: Tuple of all arguments.
1087 3: Phi of Exception memories.
1088 In case the old Call directly branches to End on an exception we don't
1089 need the block merging all exceptions nor the Phi of the exception
1093 /* -- Precompute some values -- */
1094 end_bl = get_new_node(get_irg_end_block(called_graph));
1095 end = get_new_node(get_irg_end(called_graph));
1096 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1097 n_res = get_method_n_ress(get_Call_type(call));
1099 res_pred = xmalloc (n_res * sizeof(*res_pred));
1100 cf_pred = xmalloc (arity * sizeof(*res_pred));
1102 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
1104 /* -- archive keepalives -- */
1105 irn_arity = get_irn_arity(end);
1106 for (i = 0; i < irn_arity; i++)
1107 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
1109 /* The new end node will die. We need not free as the in array is on the obstack:
1110 copy_node() only generated 'D' arrays. */
1112 /* -- Replace Return nodes by Jump nodes. -- */
1114 for (i = 0; i < arity; i++) {
1116 ret = get_irn_n(end_bl, i);
1117 if (is_Return(ret)) {
1118 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
1122 set_irn_in(post_bl, n_ret, cf_pred);
1124 /* -- Build a Tuple for all results of the method.
1125 Add Phi node if there was more than one Return. -- */
1126 turn_into_tuple(post_call, 4);
1127 /* First the Memory-Phi */
1129 for (i = 0; i < arity; i++) {
1130 ret = get_irn_n(end_bl, i);
1131 if (is_Return(ret)) {
1132 cf_pred[n_ret] = get_Return_mem(ret);
1136 phi = new_Phi(n_ret, cf_pred, mode_M);
1137 set_Tuple_pred(call, pn_Call_M_regular, phi);
1138 /* Conserve Phi-list for further inlinings -- but might be optimized */
1139 if (get_nodes_block(phi) == post_bl) {
1140 set_irn_link(phi, get_irn_link(post_bl));
1141 set_irn_link(post_bl, phi);
1143 /* Now the real results */
1145 for (j = 0; j < n_res; j++) {
1147 for (i = 0; i < arity; i++) {
1148 ret = get_irn_n(end_bl, i);
1149 if (get_irn_op(ret) == op_Return) {
1150 cf_pred[n_ret] = get_Return_res(ret, j);
1155 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1159 /* Conserve Phi-list for further inlinings -- but might be optimized */
1160 if (get_nodes_block(phi) == post_bl) {
1161 set_irn_link(phi, get_irn_link(post_bl));
1162 set_irn_link(post_bl, phi);
1165 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1167 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1169 /* Finally the exception control flow.
1170 We have two (three) possible situations:
1171 First if the Call branches to an exception handler: We need to add a Phi node to
1172 collect the memory containing the exception objects. Further we need
1173 to add another block to get a correct representation of this Phi. To
1174 this block we add a Jmp that resolves into the X output of the Call
1175 when the Call is turned into a tuple.
1176 Second the Call branches to End, the exception is not handled. Just
1177 add all inlined exception branches to the End node.
1178 Third: there is no Exception edge at all. Handle as case two. */
1179 if (exc_handling == 0) {
1181 for (i = 0; i < arity; i++) {
1183 ret = get_irn_n(end_bl, i);
1184 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1185 cf_pred[n_exc] = ret;
1190 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1191 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1192 /* The Phi for the memories with the exception objects */
1194 for (i = 0; i < arity; i++) {
1196 ret = skip_Proj(get_irn_n(end_bl, i));
1198 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1200 } else if (is_fragile_op(ret)) {
1201 /* We rely that all cfops have the memory output at the same position. */
1202 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1204 } else if (get_irn_op(ret) == op_Raise) {
1205 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1209 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1211 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1212 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1215 ir_node *main_end_bl;
1216 int main_end_bl_arity;
1217 ir_node **end_preds;
1219 /* assert(exc_handling == 1 || no exceptions. ) */
1221 for (i = 0; i < arity; i++) {
1222 ir_node *ret = get_irn_n(end_bl, i);
1224 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1225 cf_pred[n_exc] = ret;
1229 main_end_bl = get_irg_end_block(current_ir_graph);
1230 main_end_bl_arity = get_irn_arity(main_end_bl);
1231 end_preds = xmalloc ((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1233 for (i = 0; i < main_end_bl_arity; ++i)
1234 end_preds[i] = get_irn_n(main_end_bl, i);
1235 for (i = 0; i < n_exc; ++i)
1236 end_preds[main_end_bl_arity + i] = cf_pred[i];
1237 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1238 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1239 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1245 /* -- Turn CSE back on. -- */
1246 set_optimize(rem_opt);
1251 /********************************************************************/
1252 /* Apply inlineing to small methods. */
1253 /********************************************************************/
1255 /** Represents a possible inlinable call in a graph. */
1256 typedef struct _call_entry call_entry;
1257 struct _call_entry {
1258 ir_node *call; /**< the Call */
1259 ir_graph *callee; /**< the callee called here */
1260 call_entry *next; /**< for linking the next one */
1264 * environment for inlining small irgs
1266 typedef struct _inline_env_t {
1267 struct obstack obst; /**< an obstack where call_entries are allocated on. */
1268 call_entry *head; /**< the head of the call entry list */
1269 call_entry *tail; /**< the tail of the call entry list */
1273 * Returns the irg called from a Call node. If the irg is not
1274 * known, NULL is returned.
1276 static ir_graph *get_call_called_irg(ir_node *call) {
1278 ir_graph *called_irg = NULL;
1280 addr = get_Call_ptr(call);
1281 if (is_SymConst(addr) && get_SymConst_kind(addr) == symconst_addr_ent) {
1282 called_irg = get_entity_irg(get_SymConst_entity(addr));
1289 * Walker: Collect all calls to known graphs inside a graph.
1291 static void collect_calls(ir_node *call, void *env) {
1292 if (is_Call(call)) {
1293 ir_graph *called_irg = get_call_called_irg(call);
1295 /* The Call node calls a locally defined method. Remember to inline. */
1296 inline_env_t *ienv = env;
1297 call_entry *entry = obstack_alloc(&ienv->obst, sizeof(*entry));
1299 entry->callee = called_irg;
1302 if (ienv->tail == NULL)
1305 ienv->tail->next = entry;
1312 * Inlines all small methods at call sites where the called address comes
1313 * from a Const node that references the entity representing the called
1315 * The size argument is a rough measure for the code size of the method:
1316 * Methods where the obstack containing the firm graph is smaller than
1319 void inline_small_irgs(ir_graph *irg, int size) {
1320 ir_graph *rem = current_ir_graph;
1323 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1325 if (!(get_opt_optimize() && get_opt_inline())) return;
1327 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1329 current_ir_graph = irg;
1330 /* Handle graph state */
1331 assert(get_irg_phase_state(irg) != phase_building);
1332 free_callee_info(irg);
1334 /* Find Call nodes to inline.
1335 (We can not inline during a walk of the graph, as inlineing the same
1336 method several times changes the visited flag of the walked graph:
1337 after the first inlineing visited of the callee equals visited of
1338 the caller. With the next inlineing both are increased.) */
1339 obstack_init(&env.obst);
1340 env.head = env.tail = NULL;
1341 irg_walk_graph(irg, NULL, collect_calls, &env);
1343 if (env.head != NULL) {
1344 /* There are calls to inline */
1345 collect_phiprojs(irg);
1346 for (entry = env.head; entry != NULL; entry = entry->next) {
1347 ir_graph *callee = entry->callee;
1348 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1349 (get_irg_inline_property(callee) >= irg_inline_forced)) {
1350 inline_method(entry->call, callee);
1354 obstack_free(&env.obst, NULL);
1355 current_ir_graph = rem;
1359 * Environment for inlining irgs.
1362 int n_nodes; /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
1363 int n_nodes_orig; /**< for statistics */
1364 call_entry *call_head; /**< The head of the list of all call nodes in this graph. */
1365 call_entry *call_tail; /**< The tail of the list of all call nodes in this graph .*/
1366 int n_call_nodes; /**< Number of Call nodes in the graph. */
1367 int n_call_nodes_orig; /**< for statistics */
1368 int n_callers; /**< Number of known graphs that call this graphs. */
1369 int n_callers_orig; /**< for statistics */
1370 int got_inline; /**< Set, if at leat one call inside this graph was inlined. */
1374 * Allocate a new environment for inlining.
1376 static inline_irg_env *alloc_inline_irg_env(struct obstack *obst) {
1377 inline_irg_env *env = obstack_alloc(obst, sizeof(*env));
1378 env->n_nodes = -2; /* do not count count Start, End */
1379 env->n_nodes_orig = -2; /* do not count Start, End */
1380 env->call_head = NULL;
1381 env->call_tail = NULL;
1382 env->n_call_nodes = 0;
1383 env->n_call_nodes_orig = 0;
1385 env->n_callers_orig = 0;
1386 env->got_inline = 0;
1390 typedef struct walker_env {
1391 struct obstack *obst; /**< the obstack for allocations. */
1392 inline_irg_env *x; /**< the inline environment */
1393 int ignore_runtime; /**< the ignore runtime flag */
1397 * post-walker: collect all calls in the inline-environment
1398 * of a graph and sum some statistics.
1400 static void collect_calls2(ir_node *call, void *ctx) {
1402 inline_irg_env *x = env->x;
1403 ir_op *op = get_irn_op(call);
1407 /* count meaningful nodes in irg */
1408 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1413 if (op != op_Call) return;
1415 /* check, if it's a runtime call */
1416 if (env->ignore_runtime) {
1417 ir_node *symc = get_Call_ptr(call);
1419 if (is_SymConst(symc) && get_SymConst_kind(symc) == symconst_addr_ent) {
1420 ir_entity *ent = get_SymConst_entity(symc);
1422 if (get_entity_additional_properties(ent) & mtp_property_runtime)
1427 /* collect all call nodes */
1429 ++x->n_call_nodes_orig;
1431 callee = get_call_called_irg(call);
1433 inline_irg_env *callee_env = get_irg_link(callee);
1434 /* count all static callers */
1435 ++callee_env->n_callers;
1436 ++callee_env->n_callers_orig;
1438 /* link it in the list of possible inlinable entries */
1439 entry = obstack_alloc(env->obst, sizeof(*entry));
1441 entry->callee = callee;
1443 if (x->call_tail == NULL)
1444 x->call_head = entry;
1446 x->call_tail->next = entry;
1447 x->call_tail = entry;
1452 * Returns TRUE if the number of callers in 0 in the irg's environment,
1453 * hence this irg is a leave.
1455 INLINE static int is_leave(ir_graph *irg) {
1456 inline_irg_env *env = get_irg_link(irg);
1457 return env->n_call_nodes == 0;
1461 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1463 INLINE static int is_smaller(ir_graph *callee, int size) {
1464 inline_irg_env *env = get_irg_link(callee);
1465 return env->n_nodes < size;
1469 * Append the nodes of the list src to the nodes of the list in environment dst.
1471 static void append_call_list(struct obstack *obst, inline_irg_env *dst, call_entry *src) {
1472 call_entry *entry, *nentry;
1474 /* Note that the src list points to Call nodes in the inlined graph, but
1475 we need Call nodes in our graph. Luckily the inliner leaves this information
1476 in the link field. */
1477 for (entry = src; entry != NULL; entry = entry->next) {
1478 nentry = obstack_alloc(obst, sizeof(*nentry));
1479 nentry->call = get_irn_link(entry->call);
1480 nentry->callee = entry->callee;
1481 nentry->next = NULL;
1482 dst->call_tail->next = nentry;
1483 dst->call_tail = nentry;
1488 * Inlines small leave methods at call sites where the called address comes
1489 * from a Const node that references the entity representing the called
1491 * The size argument is a rough measure for the code size of the method:
1492 * Methods where the obstack containing the firm graph is smaller than
1495 void inline_leave_functions(int maxsize, int leavesize, int size, int ignore_runtime) {
1496 inline_irg_env *env;
1502 call_entry *entry, *tail;
1503 const call_entry *centry;
1504 struct obstack obst;
1505 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1507 if (!(get_opt_optimize() && get_opt_inline())) return;
1509 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1510 rem = current_ir_graph;
1511 obstack_init(&obst);
1513 /* extend all irgs by a temporary data structure for inlining. */
1514 n_irgs = get_irp_n_irgs();
1515 for (i = 0; i < n_irgs; ++i)
1516 set_irg_link(get_irp_irg(i), alloc_inline_irg_env(&obst));
1518 /* Precompute information in temporary data structure. */
1520 wenv.ignore_runtime = ignore_runtime;
1521 for (i = 0; i < n_irgs; ++i) {
1522 ir_graph *irg = get_irp_irg(i);
1524 assert(get_irg_phase_state(irg) != phase_building);
1525 free_callee_info(irg);
1527 wenv.x = get_irg_link(irg);
1528 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1531 /* -- and now inline. -- */
1533 /* Inline leaves recursively -- we might construct new leaves. */
1537 for (i = 0; i < n_irgs; ++i) {
1539 int phiproj_computed = 0;
1541 current_ir_graph = get_irp_irg(i);
1542 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1545 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1548 if (env->n_nodes > maxsize) break;
1551 callee = entry->callee;
1553 if (is_leave(callee) && is_smaller(callee, leavesize)) {
1554 if (!phiproj_computed) {
1555 phiproj_computed = 1;
1556 collect_phiprojs(current_ir_graph);
1558 did_inline = inline_method(call, callee);
1561 /* Do some statistics */
1562 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1564 env->got_inline = 1;
1565 --env->n_call_nodes;
1566 env->n_nodes += callee_env->n_nodes;
1567 --callee_env->n_callers;
1569 /* remove this call from the list */
1571 tail->next = entry->next;
1573 env->call_head = entry->next;
1579 env->call_tail = tail;
1581 } while (did_inline);
1583 /* inline other small functions. */
1584 for (i = 0; i < n_irgs; ++i) {
1586 int phiproj_computed = 0;
1588 current_ir_graph = get_irp_irg(i);
1589 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1591 /* note that the list of possible calls is updated during the process */
1593 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1597 callee = entry->callee;
1599 if (((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1600 (get_irg_inline_property(callee) >= irg_inline_forced))) {
1601 if (!phiproj_computed) {
1602 phiproj_computed = 1;
1603 collect_phiprojs(current_ir_graph);
1605 if (inline_method(call, callee)) {
1606 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1608 /* callee was inline. Append it's call list. */
1609 env->got_inline = 1;
1610 --env->n_call_nodes;
1611 append_call_list(&obst, env, callee_env->call_head);
1612 env->n_call_nodes += callee_env->n_call_nodes;
1613 env->n_nodes += callee_env->n_nodes;
1614 --callee_env->n_callers;
1616 /* after we have inlined callee, all called methods inside callee
1617 are now called once more */
1618 for (centry = callee_env->call_head; centry != NULL; centry = centry->next) {
1619 inline_irg_env *penv = get_irg_link(centry->callee);
1623 /* remove this call from the list */
1625 tail->next = entry->next;
1627 env->call_head = entry->next;
1633 env->call_tail = tail;
1636 for (i = 0; i < n_irgs; ++i) {
1637 irg = get_irp_irg(i);
1638 env = (inline_irg_env *)get_irg_link(irg);
1640 if (env->got_inline) {
1641 /* this irg got calls inlined */
1642 set_irg_outs_inconsistent(irg);
1643 set_irg_doms_inconsistent(irg);
1645 optimize_graph_df(irg);
1648 if (env->got_inline || (env->n_callers_orig != env->n_callers))
1649 DB((dbg, SET_LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1650 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1651 env->n_callers_orig, env->n_callers,
1652 get_entity_name(get_irg_entity(irg))));
1655 obstack_free(&obst, NULL);
1656 current_ir_graph = rem;
1659 /*******************************************************************/
1660 /* Code Placement. Pins all floating nodes to a block where they */
1661 /* will be executed only if needed. */
1662 /*******************************************************************/
1665 * Returns non-zero, is a block is not reachable from Start.
1667 * @param block the block to test
1670 is_Block_unreachable(ir_node *block) {
1671 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1675 * Find the earliest correct block for N. --- Place N into the
1676 * same Block as its dominance-deepest Input.
1678 * We have to avoid calls to get_nodes_block() here
1679 * because the graph is floating.
1681 * move_out_of_loops() expects that place_floats_early() have placed
1682 * all "living" nodes into a living block. That's why we must
1683 * move nodes in dead block with "live" successors into a valid
1685 * We move them just into the same block as it's successor (or
1686 * in case of a Phi into the effective use block). For Phi successors,
1687 * this may still be a dead block, but then there is no real use, as
1688 * the control flow will be dead later.
1691 place_floats_early(ir_node *n, waitq *worklist)
1695 /* we must not run into an infinite loop */
1696 assert(irn_not_visited(n));
1697 mark_irn_visited(n);
1699 /* Place floating nodes. */
1700 if (get_irn_pinned(n) == op_pin_state_floats) {
1701 ir_node *curr_block = get_irn_n(n, -1);
1702 int in_dead_block = is_Block_unreachable(curr_block);
1704 ir_node *b = NULL; /* The block to place this node in */
1706 assert(is_no_Block(n));
1708 if (is_irn_start_block_placed(n)) {
1709 /* These nodes will not be placed by the loop below. */
1710 b = get_irg_start_block(current_ir_graph);
1714 /* find the block for this node. */
1715 irn_arity = get_irn_arity(n);
1716 for (i = 0; i < irn_arity; i++) {
1717 ir_node *pred = get_irn_n(n, i);
1718 ir_node *pred_block;
1720 if ((irn_not_visited(pred))
1721 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1724 * If the current node is NOT in a dead block, but one of its
1725 * predecessors is, we must move the predecessor to a live block.
1726 * Such thing can happen, if global CSE chose a node from a dead block.
1727 * We move it simply to our block.
1728 * Note that neither Phi nor End nodes are floating, so we don't
1729 * need to handle them here.
1731 if (! in_dead_block) {
1732 if (get_irn_pinned(pred) == op_pin_state_floats &&
1733 is_Block_unreachable(get_irn_n(pred, -1)))
1734 set_nodes_block(pred, curr_block);
1736 place_floats_early(pred, worklist);
1740 * A node in the Bad block must stay in the bad block,
1741 * so don't compute a new block for it.
1746 /* Because all loops contain at least one op_pin_state_pinned node, now all
1747 our inputs are either op_pin_state_pinned or place_early() has already
1748 been finished on them. We do not have any unfinished inputs! */
1749 pred_block = get_irn_n(pred, -1);
1750 if ((!is_Block_dead(pred_block)) &&
1751 (get_Block_dom_depth(pred_block) > depth)) {
1753 depth = get_Block_dom_depth(pred_block);
1755 /* Avoid that the node is placed in the Start block */
1756 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)) {
1757 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1758 assert(b != get_irg_start_block(current_ir_graph));
1763 set_nodes_block(n, b);
1767 * Add predecessors of non floating nodes and non-floating predecessors
1768 * of floating nodes to worklist and fix their blocks if the are in dead block.
1770 irn_arity = get_irn_arity(n);
1772 if (get_irn_op(n) == op_End) {
1774 * Simplest case: End node. Predecessors are keep-alives,
1775 * no need to move out of dead block.
1777 for (i = -1; i < irn_arity; ++i) {
1778 ir_node *pred = get_irn_n(n, i);
1779 if (irn_not_visited(pred))
1780 waitq_put(worklist, pred);
1783 else if (is_Block(n)) {
1785 * Blocks: Predecessors are control flow, no need to move
1786 * them out of dead block.
1788 for (i = irn_arity - 1; i >= 0; --i) {
1789 ir_node *pred = get_irn_n(n, i);
1790 if (irn_not_visited(pred))
1791 waitq_put(worklist, pred);
1794 else if (is_Phi(n)) {
1796 ir_node *curr_block = get_irn_n(n, -1);
1797 int in_dead_block = is_Block_unreachable(curr_block);
1800 * Phi nodes: move nodes from dead blocks into the effective use
1801 * of the Phi-input if the Phi is not in a bad block.
1803 pred = get_irn_n(n, -1);
1804 if (irn_not_visited(pred))
1805 waitq_put(worklist, pred);
1807 for (i = irn_arity - 1; i >= 0; --i) {
1808 ir_node *pred = get_irn_n(n, i);
1810 if (irn_not_visited(pred)) {
1811 if (! in_dead_block &&
1812 get_irn_pinned(pred) == op_pin_state_floats &&
1813 is_Block_unreachable(get_irn_n(pred, -1))) {
1814 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1816 waitq_put(worklist, pred);
1822 ir_node *curr_block = get_irn_n(n, -1);
1823 int in_dead_block = is_Block_unreachable(curr_block);
1826 * All other nodes: move nodes from dead blocks into the same block.
1828 pred = get_irn_n(n, -1);
1829 if (irn_not_visited(pred))
1830 waitq_put(worklist, pred);
1832 for (i = irn_arity - 1; i >= 0; --i) {
1833 ir_node *pred = get_irn_n(n, i);
1835 if (irn_not_visited(pred)) {
1836 if (! in_dead_block &&
1837 get_irn_pinned(pred) == op_pin_state_floats &&
1838 is_Block_unreachable(get_irn_n(pred, -1))) {
1839 set_nodes_block(pred, curr_block);
1841 waitq_put(worklist, pred);
1848 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1849 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1850 * places all floating nodes reachable from its argument through floating
1851 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1853 static INLINE void place_early(waitq *worklist) {
1855 inc_irg_visited(current_ir_graph);
1857 /* this inits the worklist */
1858 place_floats_early(get_irg_end(current_ir_graph), worklist);
1860 /* Work the content of the worklist. */
1861 while (!waitq_empty(worklist)) {
1862 ir_node *n = waitq_get(worklist);
1863 if (irn_not_visited(n))
1864 place_floats_early(n, worklist);
1867 set_irg_outs_inconsistent(current_ir_graph);
1868 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
1872 * Compute the deepest common ancestor of block and dca.
1874 static ir_node *calc_dca(ir_node *dca, ir_node *block)
1878 /* we do not want to place nodes in dead blocks */
1879 if (is_Block_dead(block))
1882 /* We found a first legal placement. */
1883 if (!dca) return block;
1885 /* Find a placement that is dominates both, dca and block. */
1886 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1887 block = get_Block_idom(block);
1889 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1890 dca = get_Block_idom(dca);
1893 while (block != dca)
1894 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1899 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1900 * I.e., DCA is the block where we might place PRODUCER.
1901 * A data flow edge points from producer to consumer.
1904 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer)
1906 ir_node *block = NULL;
1908 /* Compute the latest block into which we can place a node so that it is
1910 if (get_irn_op(consumer) == op_Phi) {
1911 /* our consumer is a Phi-node, the effective use is in all those
1912 blocks through which the Phi-node reaches producer */
1914 ir_node *phi_block = get_nodes_block(consumer);
1915 irn_arity = get_irn_arity(consumer);
1917 for (i = 0; i < irn_arity; i++) {
1918 if (get_irn_n(consumer, i) == producer) {
1919 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1921 if (! is_Block_unreachable(new_block))
1922 block = calc_dca(block, new_block);
1927 block = get_irn_n(producer, -1);
1930 assert(is_no_Block(consumer));
1931 block = get_nodes_block(consumer);
1934 /* Compute the deepest common ancestor of block and dca. */
1935 return calc_dca(dca, block);
1938 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1940 static INLINE int get_irn_loop_depth(ir_node *n) {
1941 return get_loop_depth(get_irn_loop(n));
1945 * Move n to a block with less loop depth than it's current block. The
1946 * new block must be dominated by early.
1948 * @param n the node that should be moved
1949 * @param early the earliest block we can n move to
1951 static void move_out_of_loops(ir_node *n, ir_node *early)
1953 ir_node *best, *dca;
1957 /* Find the region deepest in the dominator tree dominating
1958 dca with the least loop nesting depth, but still dominated
1959 by our early placement. */
1960 dca = get_nodes_block(n);
1963 while (dca != early) {
1964 dca = get_Block_idom(dca);
1965 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1966 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1970 if (best != get_nodes_block(n)) {
1972 printf("Moving out of loop: "); DDMN(n);
1973 printf(" Outermost block: "); DDMN(early);
1974 printf(" Best block: "); DDMN(best);
1975 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1977 set_nodes_block(n, best);
1982 * Find the latest legal block for N and place N into the
1983 * `optimal' Block between the latest and earliest legal block.
1984 * The `optimal' block is the dominance-deepest block of those
1985 * with the least loop-nesting-depth. This places N out of as many
1986 * loops as possible and then makes it as control dependent as
1989 static void place_floats_late(ir_node *n, pdeq *worklist)
1994 assert(irn_not_visited(n)); /* no multiple placement */
1996 mark_irn_visited(n);
1998 /* no need to place block nodes, control nodes are already placed. */
1999 if ((get_irn_op(n) != op_Block) &&
2001 (get_irn_mode(n) != mode_X)) {
2002 /* Remember the early_blk placement of this block to move it
2003 out of loop no further than the early_blk placement. */
2004 early_blk = get_irn_n(n, -1);
2007 * BEWARE: Here we also get code, that is live, but
2008 * was in a dead block. If the node is life, but because
2009 * of CSE in a dead block, we still might need it.
2012 /* Assure that our users are all placed, except the Phi-nodes.
2013 --- Each data flow cycle contains at least one Phi-node. We
2014 have to break the `user has to be placed before the
2015 producer' dependence cycle and the Phi-nodes are the
2016 place to do so, because we need to base our placement on the
2017 final region of our users, which is OK with Phi-nodes, as they
2018 are op_pin_state_pinned, and they never have to be placed after a
2019 producer of one of their inputs in the same block anyway. */
2020 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2021 ir_node *succ = get_irn_out(n, i);
2022 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
2023 place_floats_late(succ, worklist);
2026 if (! is_Block_dead(early_blk)) {
2027 /* do only move things that where not dead */
2028 ir_op *op = get_irn_op(n);
2030 /* We have to determine the final block of this node... except for
2031 constants and Projs */
2032 if ((get_irn_pinned(n) == op_pin_state_floats) &&
2034 (op != op_SymConst) &&
2037 ir_node *dca = NULL; /* deepest common ancestor in the
2038 dominator tree of all nodes'
2039 blocks depending on us; our final
2040 placement has to dominate DCA. */
2041 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2042 ir_node *succ = get_irn_out(n, i);
2045 if (get_irn_op(succ) == op_End) {
2047 * This consumer is the End node, a keep alive edge.
2048 * This is not a real consumer, so we ignore it
2053 /* ignore if succ is in dead code */
2054 succ_blk = get_irn_n(succ, -1);
2055 if (is_Block_unreachable(succ_blk))
2057 dca = consumer_dom_dca(dca, succ, n);
2060 set_nodes_block(n, dca);
2061 move_out_of_loops(n, early_blk);
2067 /* Add predecessors of all non-floating nodes on list. (Those of floating
2068 nodes are placed already and therefore are marked.) */
2069 for (i = 0; i < get_irn_n_outs(n); i++) {
2070 ir_node *succ = get_irn_out(n, i);
2071 if (irn_not_visited(get_irn_out(n, i))) {
2072 pdeq_putr(worklist, succ);
2077 static INLINE void place_late(waitq *worklist) {
2079 inc_irg_visited(current_ir_graph);
2081 /* This fills the worklist initially. */
2082 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
2084 /* And now empty the worklist again... */
2085 while (!waitq_empty(worklist)) {
2086 ir_node *n = waitq_get(worklist);
2087 if (irn_not_visited(n))
2088 place_floats_late(n, worklist);
2092 void place_code(ir_graph *irg) {
2094 ir_graph *rem = current_ir_graph;
2096 current_ir_graph = irg;
2098 /* Handle graph state */
2099 assert(get_irg_phase_state(irg) != phase_building);
2102 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
2103 free_loop_information(irg);
2104 construct_backedges(irg);
2107 /* Place all floating nodes as early as possible. This guarantees
2108 a legal code placement. */
2109 worklist = new_waitq();
2110 place_early(worklist);
2112 /* place_early() invalidates the outs, place_late needs them. */
2113 compute_irg_outs(irg);
2115 /* Now move the nodes down in the dominator tree. This reduces the
2116 unnecessary executions of the node. */
2117 place_late(worklist);
2119 set_irg_outs_inconsistent(current_ir_graph);
2120 set_irg_loopinfo_inconsistent(current_ir_graph);
2121 del_waitq(worklist);
2122 current_ir_graph = rem;
2126 * Called by walker of remove_critical_cf_edges().
2128 * Place an empty block to an edge between a blocks of multiple
2129 * predecessors and a block of multiple successors.
2132 * @param env Environment of walker. The changed field.
2134 static void walk_critical_cf_edges(ir_node *n, void *env) {
2136 ir_node *pre, *block, *jmp;
2138 ir_graph *irg = get_irn_irg(n);
2140 /* Block has multiple predecessors */
2141 arity = get_irn_arity(n);
2143 if (n == get_irg_end_block(irg))
2144 return; /* No use to add a block here. */
2146 for (i = 0; i < arity; ++i) {
2149 pre = get_irn_n(n, i);
2150 cfop = get_irn_op(skip_Proj(pre));
2151 /* Predecessor has multiple successors. Insert new control flow edge but
2152 ignore exception edges. */
2153 if (! is_op_fragile(cfop) && is_op_forking(cfop)) {
2154 /* set predecessor of new block */
2155 block = new_r_Block(irg, 1, &pre);
2156 /* insert new jmp node to new block */
2157 jmp = new_r_Jmp(irg, block);
2158 /* set successor of new block */
2159 set_irn_n(n, i, jmp);
2161 } /* predecessor has multiple successors */
2162 } /* for all predecessors */
2163 } /* n is a multi-entry block */
2166 void remove_critical_cf_edges(ir_graph *irg) {
2169 irg_block_walk_graph(irg, NULL, walk_critical_cf_edges, &changed);
2171 /* control flow changed */
2172 set_irg_outs_inconsistent(irg);
2173 set_irg_extblk_inconsistent(irg);
2174 set_irg_doms_inconsistent(irg);
2175 set_irg_loopinfo_inconsistent(irg);