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 /* walk over the graph */
189 irg_walk_graph(irg, NULL, opt_walker, waitq);
191 /* finish the wait queue */
192 while (! pdeq_empty(waitq)) {
193 ir_node *n = pdeq_getl(waitq);
195 opt_walker(n, waitq);
201 edges_deactivate(irg);
203 current_ir_graph = rem;
207 /*------------------------------------------------------------------*/
208 /* Routines for dead node elimination / copying garbage collection */
209 /* of the obstack. */
210 /*------------------------------------------------------------------*/
213 * Remember the new node in the old node by using a field all nodes have.
215 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
218 * Get this new node, before the old node is forgotten.
220 #define get_new_node(oldn) get_irn_link(oldn)
223 * Check if a new node was set.
225 #define has_new_node(n) (get_new_node(n) != NULL)
228 * We use the block_visited flag to mark that we have computed the
229 * number of useful predecessors for this block.
230 * Further we encode the new arity in this flag in the old blocks.
231 * Remembering the arity is useful, as it saves a lot of pointer
232 * accesses. This function is called for all Phi and Block nodes
236 compute_new_arity(ir_node *b) {
237 int i, res, irn_arity;
240 irg_v = get_irg_block_visited(current_ir_graph);
241 block_v = get_Block_block_visited(b);
242 if (block_v >= irg_v) {
243 /* we computed the number of preds for this block and saved it in the
245 return block_v - irg_v;
247 /* compute the number of good predecessors */
248 res = irn_arity = get_irn_arity(b);
249 for (i = 0; i < irn_arity; i++)
250 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
251 /* save it in the flag. */
252 set_Block_block_visited(b, irg_v + res);
258 * Copies the node to the new obstack. The Ins of the new node point to
259 * the predecessors on the old obstack. For block/phi nodes not all
260 * predecessors might be copied. n->link points to the new node.
261 * For Phi and Block nodes the function allocates in-arrays with an arity
262 * only for useful predecessors. The arity is determined by counting
263 * the non-bad predecessors of the block.
265 * @param n The node to be copied
266 * @param env if non-NULL, the node number attribute will be copied to the new node
268 * Note: Also used for loop unrolling.
270 static void copy_node(ir_node *n, void *env) {
273 ir_op *op = get_irn_op(n);
275 /* The end node looses it's flexible in array. This doesn't matter,
276 as dead node elimination builds End by hand, inlineing doesn't use
278 /* assert(op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
281 /* node copied already */
283 } else if (op == op_Block) {
285 new_arity = compute_new_arity(n);
286 n->attr.block.graph_arr = NULL;
288 block = get_nodes_block(n);
290 new_arity = compute_new_arity(block);
292 new_arity = get_irn_arity(n);
295 nn = new_ir_node(get_irn_dbg_info(n),
302 /* Copy the attributes. These might point to additional data. If this
303 was allocated on the old obstack the pointers now are dangling. This
304 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
305 copy_node_attr(n, nn);
306 new_backedge_info(nn);
310 int copy_node_nr = env != NULL;
312 /* for easier debugging, we want to copy the node numbers too */
313 nn->node_nr = n->node_nr;
319 hook_dead_node_elim_subst(current_ir_graph, n, nn);
323 * Copies new predecessors of old node to new node remembered in link.
324 * Spare the Bad predecessors of Phi and Block nodes.
327 copy_preds(ir_node *n, void *env) {
331 nn = get_new_node(n);
333 /* printf("\n old node: "); DDMSG2(n);
334 printf(" new node: "); DDMSG2(nn);
335 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
338 /* Don't copy Bad nodes. */
340 irn_arity = get_irn_arity(n);
341 for (i = 0; i < irn_arity; i++)
342 if (! is_Bad(get_irn_n(n, i))) {
343 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
344 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
347 /* repair the block visited flag from above misuse. Repair it in both
348 graphs so that the old one can still be used. */
349 set_Block_block_visited(nn, 0);
350 set_Block_block_visited(n, 0);
351 /* Local optimization could not merge two subsequent blocks if
352 in array contained Bads. Now it's possible.
353 We don't call optimize_in_place as it requires
354 that the fields in ir_graph are set properly. */
355 if ((get_opt_control_flow_straightening()) &&
356 (get_Block_n_cfgpreds(nn) == 1) &&
357 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
358 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
360 /* Jmp jumps into the block it is in -- deal self cycle. */
361 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
362 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
367 } else if (get_irn_op(n) == op_Phi) {
368 /* Don't copy node if corresponding predecessor in block is Bad.
369 The Block itself should not be Bad. */
370 block = get_nodes_block(n);
371 set_irn_n(nn, -1, get_new_node(block));
373 irn_arity = get_irn_arity(n);
374 for (i = 0; i < irn_arity; i++)
375 if (! is_Bad(get_irn_n(block, i))) {
376 set_irn_n(nn, j, get_new_node(get_irn_n(n, i)));
377 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
380 /* If the pre walker reached this Phi after the post walker visited the
381 block block_visited is > 0. */
382 set_Block_block_visited(get_nodes_block(n), 0);
383 /* Compacting the Phi's ins might generate Phis with only one
385 if (get_irn_arity(nn) == 1)
386 exchange(nn, get_irn_n(nn, 0));
388 irn_arity = get_irn_arity(n);
389 for (i = -1; i < irn_arity; i++)
390 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
392 /* Now the new node is complete. We can add it to the hash table for CSE.
393 @@@ inlining aborts if we identify End. Why? */
394 if (get_irn_op(nn) != op_End)
395 add_identities(current_ir_graph->value_table, nn);
399 * Copies the graph recursively, compacts the keep-alives of the end node.
401 * @param irg the graph to be copied
402 * @param copy_node_nr If non-zero, the node number will be copied
404 static void copy_graph(ir_graph *irg, int copy_node_nr) {
405 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
406 ir_node *ka; /* keep alive */
410 /* Some nodes must be copied by hand, sigh */
411 vfl = get_irg_visited(irg);
412 set_irg_visited(irg, vfl + 1);
414 oe = get_irg_end(irg);
415 mark_irn_visited(oe);
416 /* copy the end node by hand, allocate dynamic in array! */
417 ne = new_ir_node(get_irn_dbg_info(oe),
424 /* Copy the attributes. Well, there might be some in the future... */
425 copy_node_attr(oe, ne);
426 set_new_node(oe, ne);
428 /* copy the Bad node */
429 ob = get_irg_bad(irg);
430 mark_irn_visited(ob);
431 nb = new_ir_node(get_irn_dbg_info(ob),
438 copy_node_attr(ob, nb);
439 set_new_node(ob, nb);
441 /* copy the NoMem node */
442 om = get_irg_no_mem(irg);
443 mark_irn_visited(om);
444 nm = new_ir_node(get_irn_dbg_info(om),
451 copy_node_attr(om, nm);
452 set_new_node(om, nm);
454 /* copy the live nodes */
455 set_irg_visited(irg, vfl);
456 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
458 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
460 /* visit the anchors as well */
461 for (i = anchor_max - 1; i >= 0; --i) {
462 ir_node *n = irg->anchors[i];
464 if (n && (get_irn_visited(n) <= vfl)) {
465 set_irg_visited(irg, vfl);
466 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
470 /* copy_preds for the end node ... */
471 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
473 /*- ... and now the keep alives. -*/
474 /* First pick the not marked block nodes and walk them. We must pick these
475 first as else we will oversee blocks reachable from Phis. */
476 irn_arity = get_End_n_keepalives(oe);
477 for (i = 0; i < irn_arity; i++) {
478 ka = get_End_keepalive(oe, i);
480 if (get_irn_visited(ka) <= vfl) {
481 /* We must keep the block alive and copy everything reachable */
482 set_irg_visited(irg, vfl);
483 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
485 add_End_keepalive(ne, get_new_node(ka));
489 /* Now pick other nodes. Here we will keep all! */
490 irn_arity = get_End_n_keepalives(oe);
491 for (i = 0; i < irn_arity; i++) {
492 ka = get_End_keepalive(oe, i);
494 if (get_irn_visited(ka) <= vfl) {
495 /* We didn't copy the node yet. */
496 set_irg_visited(irg, vfl);
497 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
499 add_End_keepalive(ne, get_new_node(ka));
503 /* start block sometimes only reached after keep alives */
504 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
505 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
509 * Copies the graph reachable from current_ir_graph->end to the obstack
510 * in current_ir_graph and fixes the environment.
511 * Then fixes the fields in current_ir_graph containing nodes of the
514 * @param copy_node_nr If non-zero, the node number will be copied
517 copy_graph_env(int copy_node_nr) {
518 ir_graph *irg = current_ir_graph;
519 ir_node *old_end, *n;
522 /* remove end_except and end_reg nodes */
523 old_end = get_irg_end(irg);
524 set_irg_end_except (irg, old_end);
525 set_irg_end_reg (irg, old_end);
527 /* Not all nodes remembered in irg might be reachable
528 from the end node. Assure their link is set to NULL, so that
529 we can test whether new nodes have been computed. */
530 for (i = anchor_max - 1; i >= 0; --i) {
532 set_new_node(irg->anchors[i], NULL);
534 /* we use the block walk flag for removing Bads from Blocks ins. */
535 inc_irg_block_visited(irg);
538 copy_graph(irg, copy_node_nr);
540 /* fix the fields in irg */
541 old_end = get_irg_end(irg);
542 for (i = anchor_max - 1; i >= 0; --i) {
545 irg->anchors[i] = get_new_node(n);
551 * Copies all reachable nodes to a new obstack. Removes bad inputs
552 * from block nodes and the corresponding inputs from Phi nodes.
553 * Merges single exit blocks with single entry blocks and removes
555 * Adds all new nodes to a new hash table for CSE. Does not
556 * perform CSE, so the hash table might contain common subexpressions.
559 dead_node_elimination(ir_graph *irg) {
560 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
562 int rem_ipview = get_interprocedural_view();
563 struct obstack *graveyard_obst = NULL;
564 struct obstack *rebirth_obst = NULL;
565 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
567 /* inform statistics that we started a dead-node elimination run */
568 hook_dead_node_elim(irg, 1);
570 /* Remember external state of current_ir_graph. */
571 rem = current_ir_graph;
572 current_ir_graph = irg;
573 set_interprocedural_view(0);
575 assert(get_irg_phase_state(irg) != phase_building);
577 /* Handle graph state */
578 free_callee_info(irg);
582 /* @@@ so far we loose loops when copying */
583 free_loop_information(irg);
585 set_irg_doms_inconsistent(irg);
587 /* A quiet place, where the old obstack can rest in peace,
588 until it will be cremated. */
589 graveyard_obst = irg->obst;
591 /* A new obstack, where the reachable nodes will be copied to. */
592 rebirth_obst = xmalloc(sizeof(*rebirth_obst));
593 irg->obst = rebirth_obst;
594 obstack_init(irg->obst);
595 irg->last_node_idx = 0;
597 /* We also need a new value table for CSE */
598 del_identities(irg->value_table);
599 irg->value_table = new_identities();
601 /* Copy the graph from the old to the new obstack */
602 copy_graph_env(/*copy_node_nr=*/1);
604 /* Free memory from old unoptimized obstack */
605 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
606 xfree (graveyard_obst); /* ... then free it. */
608 /* inform statistics that the run is over */
609 hook_dead_node_elim(irg, 0);
611 current_ir_graph = rem;
612 set_interprocedural_view(rem_ipview);
617 * Relink bad predecessors of a block and store the old in array to the
618 * link field. This function is called by relink_bad_predecessors().
619 * The array of link field starts with the block operand at position 0.
620 * If block has bad predecessors, create a new in array without bad preds.
621 * Otherwise let in array untouched.
623 static void relink_bad_block_predecessors(ir_node *n, void *env) {
624 ir_node **new_in, *irn;
625 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
627 /* if link field of block is NULL, look for bad predecessors otherwise
628 this is already done */
629 if (get_irn_op(n) == op_Block &&
630 get_irn_link(n) == NULL) {
632 /* save old predecessors in link field (position 0 is the block operand)*/
633 set_irn_link(n, get_irn_in(n));
635 /* count predecessors without bad nodes */
636 old_irn_arity = get_irn_arity(n);
637 for (i = 0; i < old_irn_arity; i++)
638 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
640 /* arity changing: set new predecessors without bad nodes */
641 if (new_irn_arity < old_irn_arity) {
642 /* Get new predecessor array. We do not resize the array, as we must
643 keep the old one to update Phis. */
644 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
646 /* set new predecessors in array */
649 for (i = 0; i < old_irn_arity; i++) {
650 irn = get_irn_n(n, i);
652 new_in[new_irn_n] = irn;
653 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
657 //ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity);
658 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
661 } /* ir node has bad predecessors */
663 } /* Block is not relinked */
667 * Relinks Bad predecessors from Blocks and Phis called by walker
668 * remove_bad_predecesors(). If n is a Block, call
669 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
670 * function of Phi's Block. If this block has bad predecessors, relink preds
673 static void relink_bad_predecessors(ir_node *n, void *env) {
674 ir_node *block, **old_in;
675 int i, old_irn_arity, new_irn_arity;
677 /* relink bad predecessors of a block */
678 if (get_irn_op(n) == op_Block)
679 relink_bad_block_predecessors(n, env);
681 /* If Phi node relink its block and its predecessors */
682 if (get_irn_op(n) == op_Phi) {
684 /* Relink predecessors of phi's block */
685 block = get_nodes_block(n);
686 if (get_irn_link(block) == NULL)
687 relink_bad_block_predecessors(block, env);
689 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
690 old_irn_arity = ARR_LEN(old_in);
692 /* Relink Phi predecessors if count of predecessors changed */
693 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
694 /* set new predecessors in array
695 n->in[0] remains the same block */
697 for(i = 1; i < old_irn_arity; i++)
698 if (!is_Bad((ir_node *)old_in[i])) {
699 n->in[new_irn_arity] = n->in[i];
700 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
704 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
705 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
708 } /* n is a Phi node */
712 * Removes Bad Bad predecessors from Blocks and the corresponding
713 * inputs to Phi nodes as in dead_node_elimination but without
715 * On walking up set the link field to NULL, on walking down call
716 * relink_bad_predecessors() (This function stores the old in array
717 * to the link field and sets a new in array if arity of predecessors
720 void remove_bad_predecessors(ir_graph *irg) {
721 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
728 __)|_| | \_/ | \_/(/_ |_/\__|__
730 The following stuff implements a facility that automatically patches
731 registered ir_node pointers to the new node when a dead node elimination occurs.
734 struct _survive_dce_t {
738 hook_entry_t dead_node_elim;
739 hook_entry_t dead_node_elim_subst;
742 typedef struct _survive_dce_list_t {
743 struct _survive_dce_list_t *next;
745 } survive_dce_list_t;
747 static void dead_node_hook(void *context, ir_graph *irg, int start)
749 survive_dce_t *sd = context;
751 /* Create a new map before the dead node elimination is performed. */
753 sd->new_places = pmap_create_ex(pmap_count(sd->places));
756 /* Patch back all nodes if dead node elimination is over and something is to be done. */
758 pmap_destroy(sd->places);
759 sd->places = sd->new_places;
760 sd->new_places = NULL;
765 * Hook called when dead node elimination replaces old by nw.
767 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw)
769 survive_dce_t *sd = context;
770 survive_dce_list_t *list = pmap_get(sd->places, old);
772 /* If the node is to be patched back, write the new address to all registered locations. */
774 survive_dce_list_t *p;
776 for(p = list; p; p = p->next)
779 pmap_insert(sd->new_places, nw, list);
784 * Make a new Survive DCE environment.
786 survive_dce_t *new_survive_dce(void)
788 survive_dce_t *res = xmalloc(sizeof(res[0]));
789 obstack_init(&res->obst);
790 res->places = pmap_create();
791 res->new_places = NULL;
793 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
794 res->dead_node_elim.context = res;
795 res->dead_node_elim.next = NULL;
797 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
798 res->dead_node_elim_subst.context = res;
799 res->dead_node_elim_subst.next = NULL;
801 register_hook(hook_dead_node_elim, &res->dead_node_elim);
802 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
807 * Free a Survive DCE environment.
809 void free_survive_dce(survive_dce_t *sd)
811 obstack_free(&sd->obst, NULL);
812 pmap_destroy(sd->places);
813 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
814 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
819 * Register a node pointer to be patched upon DCE.
820 * When DCE occurs, the node pointer specified by @p place will be
821 * patched to the new address of the node it is pointing to.
823 * @param sd The Survive DCE environment.
824 * @param place The address of the node pointer.
826 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place)
829 ir_node *irn = *place;
830 survive_dce_list_t *curr = pmap_get(sd->places, irn);
831 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw));
836 pmap_insert(sd->places, irn, nw);
840 /*--------------------------------------------------------------------*/
841 /* Functionality for inlining */
842 /*--------------------------------------------------------------------*/
845 * Copy node for inlineing. Updates attributes that change when
846 * inlineing but not for dead node elimination.
848 * Copies the node by calling copy_node() and then updates the entity if
849 * it's a local one. env must be a pointer of the frame type of the
850 * inlined procedure. The new entities must be in the link field of
854 copy_node_inline (ir_node *n, void *env) {
856 ir_type *frame_tp = (ir_type *)env;
859 if (get_irn_op(n) == op_Sel) {
860 nn = get_new_node (n);
862 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
863 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
865 } else if (get_irn_op(n) == op_Block) {
866 nn = get_new_node (n);
867 nn->attr.block.irg = current_ir_graph;
872 * Walker: checks if P_value_arg_base is used.
874 static void find_addr(ir_node *node, void *env) {
875 int *allow_inline = env;
876 if (is_Proj(node) && get_irn_op(get_Proj_pred(node)) == op_Start) {
877 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
883 * currently, we cannot inline two cases:
884 * - call with compound arguments
885 * - graphs that take the address of a parameter
887 * check these conditions here
889 static int can_inline(ir_node *call, ir_graph *called_graph)
891 ir_type *call_type = get_Call_type(call);
892 int params, ress, i, res;
893 assert(is_Method_type(call_type));
895 params = get_method_n_params(call_type);
896 ress = get_method_n_ress(call_type);
898 /* check parameters for compound arguments */
899 for (i = 0; i < params; ++i) {
900 ir_type *p_type = get_method_param_type(call_type, i);
902 if (is_compound_type(p_type))
906 /* check results for compound arguments */
907 for (i = 0; i < ress; ++i) {
908 ir_type *r_type = get_method_res_type(call_type, i);
910 if (is_compound_type(r_type))
915 irg_walk_graph(called_graph, find_addr, NULL, &res);
920 /* Inlines a method at the given call site. */
921 int inline_method(ir_node *call, ir_graph *called_graph) {
923 ir_node *post_call, *post_bl;
924 ir_node *in[pn_Start_max];
925 ir_node *end, *end_bl;
929 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
931 ir_type *called_frame;
932 irg_inline_property prop = get_irg_inline_property(called_graph);
934 if ( (prop < irg_inline_forced) &&
935 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
937 /* Do not inline variadic functions. */
938 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
941 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
942 get_method_n_params(get_Call_type(call)));
945 * currently, we cannot inline two cases:
946 * - call with compound arguments
947 * - graphs that take the address of a parameter
949 if (! can_inline(call, called_graph))
952 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
953 rem_opt = get_opt_optimize();
956 /* Handle graph state */
957 assert(get_irg_phase_state(current_ir_graph) != phase_building);
958 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
959 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
960 set_irg_outs_inconsistent(current_ir_graph);
961 set_irg_extblk_inconsistent(current_ir_graph);
962 set_irg_doms_inconsistent(current_ir_graph);
963 set_irg_loopinfo_inconsistent(current_ir_graph);
964 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
966 /* -- Check preconditions -- */
967 assert(is_Call(call));
968 /* @@@ does not work for InterfaceIII.java after cgana
969 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
970 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
971 get_Call_type(call)));
973 if (called_graph == current_ir_graph) {
974 set_optimize(rem_opt);
978 /* here we know we WILL inline, so inform the statistics */
979 hook_inline(call, called_graph);
981 /* -- Decide how to handle exception control flow: Is there a handler
982 for the Call node, or do we branch directly to End on an exception?
984 0 There is a handler.
986 2 Exception handling not represented in Firm. -- */
988 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
989 for (proj = get_irn_link(call); proj; proj = get_irn_link(proj)) {
990 assert(is_Proj(proj));
991 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
992 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
994 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
995 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
996 else { exc_handling = 2; } /* !Mproj && !Xproj */
1001 the procedure and later replaces the Start node of the called graph.
1002 Post_call is the old Call node and collects the results of the called
1003 graph. Both will end up being a tuple. -- */
1004 post_bl = get_nodes_block(call);
1005 set_irg_current_block(current_ir_graph, post_bl);
1006 /* XxMxPxPxPxT of Start + parameter of Call */
1007 in[pn_Start_X_initial_exec] = new_Jmp();
1008 in[pn_Start_M] = get_Call_mem(call);
1009 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
1010 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
1011 in[pn_Start_P_tls] = get_irg_tls(current_ir_graph);
1012 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
1013 /* in[pn_Start_P_value_arg_base] = ??? */
1014 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
1015 pre_call = new_Tuple(pn_Start_max - 1, in);
1019 The new block gets the ins of the old block, pre_call and all its
1020 predecessors and all Phi nodes. -- */
1021 part_block(pre_call);
1023 /* -- Prepare state for dead node elimination -- */
1024 /* Visited flags in calling irg must be >= flag in called irg.
1025 Else walker and arity computation will not work. */
1026 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
1027 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
1028 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
1029 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
1030 /* Set pre_call as new Start node in link field of the start node of
1031 calling graph and pre_calls block as new block for the start block
1033 Further mark these nodes so that they are not visited by the
1035 set_irn_link(get_irg_start(called_graph), pre_call);
1036 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
1037 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1038 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
1039 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
1040 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
1042 /* Initialize for compaction of in arrays */
1043 inc_irg_block_visited(current_ir_graph);
1045 /* -- Replicate local entities of the called_graph -- */
1046 /* copy the entities. */
1047 called_frame = get_irg_frame_type(called_graph);
1048 for (i = 0; i < get_class_n_members(called_frame); i++) {
1049 ir_entity *new_ent, *old_ent;
1050 old_ent = get_class_member(called_frame, i);
1051 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
1052 set_entity_link(old_ent, new_ent);
1055 /* visited is > than that of called graph. With this trick visited will
1056 remain unchanged so that an outer walker, e.g., searching the call nodes
1057 to inline, calling this inline will not visit the inlined nodes. */
1058 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
1060 /* -- Performing dead node elimination inlines the graph -- */
1061 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1063 /* @@@ endless loops are not copied!! -- they should be, I think... */
1064 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
1065 get_irg_frame_type(called_graph));
1067 /* Repair called_graph */
1068 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
1069 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
1070 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1072 /* -- Merge the end of the inlined procedure with the call site -- */
1073 /* We will turn the old Call node into a Tuple with the following
1076 0: Phi of all Memories of Return statements.
1077 1: Jmp from new Block that merges the control flow from all exception
1078 predecessors of the old end block.
1079 2: Tuple of all arguments.
1080 3: Phi of Exception memories.
1081 In case the old Call directly branches to End on an exception we don't
1082 need the block merging all exceptions nor the Phi of the exception
1086 /* -- Precompute some values -- */
1087 end_bl = get_new_node(get_irg_end_block(called_graph));
1088 end = get_new_node(get_irg_end(called_graph));
1089 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1090 n_res = get_method_n_ress(get_Call_type(call));
1092 res_pred = xmalloc (n_res * sizeof(*res_pred));
1093 cf_pred = xmalloc (arity * sizeof(*res_pred));
1095 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
1097 /* -- archive keepalives -- */
1098 irn_arity = get_irn_arity(end);
1099 for (i = 0; i < irn_arity; i++)
1100 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
1102 /* The new end node will die. We need not free as the in array is on the obstack:
1103 copy_node() only generated 'D' arrays. */
1105 /* -- Replace Return nodes by Jump nodes. -- */
1107 for (i = 0; i < arity; i++) {
1109 ret = get_irn_n(end_bl, i);
1110 if (is_Return(ret)) {
1111 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
1115 set_irn_in(post_bl, n_ret, cf_pred);
1117 /* -- Build a Tuple for all results of the method.
1118 Add Phi node if there was more than one Return. -- */
1119 turn_into_tuple(post_call, 4);
1120 /* First the Memory-Phi */
1122 for (i = 0; i < arity; i++) {
1123 ret = get_irn_n(end_bl, i);
1124 if (is_Return(ret)) {
1125 cf_pred[n_ret] = get_Return_mem(ret);
1129 phi = new_Phi(n_ret, cf_pred, mode_M);
1130 set_Tuple_pred(call, pn_Call_M_regular, phi);
1131 /* Conserve Phi-list for further inlinings -- but might be optimized */
1132 if (get_nodes_block(phi) == post_bl) {
1133 set_irn_link(phi, get_irn_link(post_bl));
1134 set_irn_link(post_bl, phi);
1136 /* Now the real results */
1138 for (j = 0; j < n_res; j++) {
1140 for (i = 0; i < arity; i++) {
1141 ret = get_irn_n(end_bl, i);
1142 if (get_irn_op(ret) == op_Return) {
1143 cf_pred[n_ret] = get_Return_res(ret, j);
1148 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1152 /* Conserve Phi-list for further inlinings -- but might be optimized */
1153 if (get_nodes_block(phi) == post_bl) {
1154 set_irn_link(phi, get_irn_link(post_bl));
1155 set_irn_link(post_bl, phi);
1158 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1160 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1162 /* Finally the exception control flow.
1163 We have two (three) possible situations:
1164 First if the Call branches to an exception handler: We need to add a Phi node to
1165 collect the memory containing the exception objects. Further we need
1166 to add another block to get a correct representation of this Phi. To
1167 this block we add a Jmp that resolves into the X output of the Call
1168 when the Call is turned into a tuple.
1169 Second the Call branches to End, the exception is not handled. Just
1170 add all inlined exception branches to the End node.
1171 Third: there is no Exception edge at all. Handle as case two. */
1172 if (exc_handling == 0) {
1174 for (i = 0; i < arity; i++) {
1176 ret = get_irn_n(end_bl, i);
1177 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1178 cf_pred[n_exc] = ret;
1183 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1184 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1185 /* The Phi for the memories with the exception objects */
1187 for (i = 0; i < arity; i++) {
1189 ret = skip_Proj(get_irn_n(end_bl, i));
1191 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1193 } else if (is_fragile_op(ret)) {
1194 /* We rely that all cfops have the memory output at the same position. */
1195 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1197 } else if (get_irn_op(ret) == op_Raise) {
1198 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1202 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1204 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1205 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1208 ir_node *main_end_bl;
1209 int main_end_bl_arity;
1210 ir_node **end_preds;
1212 /* assert(exc_handling == 1 || no exceptions. ) */
1214 for (i = 0; i < arity; i++) {
1215 ir_node *ret = get_irn_n(end_bl, i);
1217 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1218 cf_pred[n_exc] = ret;
1222 main_end_bl = get_irg_end_block(current_ir_graph);
1223 main_end_bl_arity = get_irn_arity(main_end_bl);
1224 end_preds = xmalloc ((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1226 for (i = 0; i < main_end_bl_arity; ++i)
1227 end_preds[i] = get_irn_n(main_end_bl, i);
1228 for (i = 0; i < n_exc; ++i)
1229 end_preds[main_end_bl_arity + i] = cf_pred[i];
1230 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1231 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1232 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1238 /* -- Turn CSE back on. -- */
1239 set_optimize(rem_opt);
1244 /********************************************************************/
1245 /* Apply inlineing to small methods. */
1246 /********************************************************************/
1248 /** Represents a possible inlinable call in a graph. */
1249 typedef struct _call_entry call_entry;
1250 struct _call_entry {
1251 ir_node *call; /**< the Call */
1252 ir_graph *callee; /**< the callee called here */
1253 call_entry *next; /**< for linking the next one */
1257 * environment for inlining small irgs
1259 typedef struct _inline_env_t {
1260 struct obstack obst; /**< an obstack where call_entries are allocated on. */
1261 call_entry *head; /**< the head of the call entry list */
1262 call_entry *tail; /**< the tail of the call entry list */
1266 * Returns the irg called from a Call node. If the irg is not
1267 * known, NULL is returned.
1269 static ir_graph *get_call_called_irg(ir_node *call) {
1271 ir_graph *called_irg = NULL;
1273 addr = get_Call_ptr(call);
1274 if (is_SymConst(addr) && get_SymConst_kind(addr) == symconst_addr_ent) {
1275 called_irg = get_entity_irg(get_SymConst_entity(addr));
1282 * Walker: Collect all calls to known graphs inside a graph.
1284 static void collect_calls(ir_node *call, void *env) {
1285 if (is_Call(call)) {
1286 ir_graph *called_irg = get_call_called_irg(call);
1288 /* The Call node calls a locally defined method. Remember to inline. */
1289 inline_env_t *ienv = env;
1290 call_entry *entry = obstack_alloc(&ienv->obst, sizeof(*entry));
1292 entry->callee = called_irg;
1295 if (ienv->tail == NULL)
1298 ienv->tail->next = entry;
1305 * Inlines all small methods at call sites where the called address comes
1306 * from a Const node that references the entity representing the called
1308 * The size argument is a rough measure for the code size of the method:
1309 * Methods where the obstack containing the firm graph is smaller than
1312 void inline_small_irgs(ir_graph *irg, int size) {
1313 ir_graph *rem = current_ir_graph;
1316 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1318 if (!(get_opt_optimize() && get_opt_inline())) return;
1320 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1322 current_ir_graph = irg;
1323 /* Handle graph state */
1324 assert(get_irg_phase_state(irg) != phase_building);
1325 free_callee_info(irg);
1327 /* Find Call nodes to inline.
1328 (We can not inline during a walk of the graph, as inlineing the same
1329 method several times changes the visited flag of the walked graph:
1330 after the first inlineing visited of the callee equals visited of
1331 the caller. With the next inlineing both are increased.) */
1332 obstack_init(&env.obst);
1333 env.head = env.tail = NULL;
1334 irg_walk_graph(irg, NULL, collect_calls, &env);
1336 if (env.head != NULL) {
1337 /* There are calls to inline */
1338 collect_phiprojs(irg);
1339 for (entry = env.head; entry != NULL; entry = entry->next) {
1340 ir_graph *callee = entry->callee;
1341 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1342 (get_irg_inline_property(callee) >= irg_inline_forced)) {
1343 inline_method(entry->call, callee);
1347 obstack_free(&env.obst, NULL);
1348 current_ir_graph = rem;
1352 * Environment for inlining irgs.
1355 int n_nodes; /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
1356 int n_nodes_orig; /**< for statistics */
1357 call_entry *call_head; /**< The head of the list of all call nodes in this graph. */
1358 call_entry *call_tail; /**< The tail of the list of all call nodes in this graph .*/
1359 int n_call_nodes; /**< Number of Call nodes in the graph. */
1360 int n_call_nodes_orig; /**< for statistics */
1361 int n_callers; /**< Number of known graphs that call this graphs. */
1362 int n_callers_orig; /**< for statistics */
1363 int got_inline; /**< Set, if at leat one call inside this graph was inlined. */
1367 * Allocate a new environment for inlining.
1369 static inline_irg_env *alloc_inline_irg_env(struct obstack *obst) {
1370 inline_irg_env *env = obstack_alloc(obst, sizeof(*env));
1371 env->n_nodes = -2; /* do not count count Start, End */
1372 env->n_nodes_orig = -2; /* do not count Start, End */
1373 env->call_head = NULL;
1374 env->call_tail = NULL;
1375 env->n_call_nodes = 0;
1376 env->n_call_nodes_orig = 0;
1378 env->n_callers_orig = 0;
1379 env->got_inline = 0;
1383 typedef struct walker_env {
1384 struct obstack *obst; /**< the obstack for allocations. */
1385 inline_irg_env *x; /**< the inline environment */
1386 int ignore_runtime; /**< the ignore runtime flag */
1390 * post-walker: collect all calls in the inline-environment
1391 * of a graph and sum some statistics.
1393 static void collect_calls2(ir_node *call, void *ctx) {
1395 inline_irg_env *x = env->x;
1396 ir_op *op = get_irn_op(call);
1400 /* count meaningful nodes in irg */
1401 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1406 if (op != op_Call) return;
1408 /* check, if it's a runtime call */
1409 if (env->ignore_runtime) {
1410 ir_node *symc = get_Call_ptr(call);
1412 if (is_SymConst(symc) && get_SymConst_kind(symc) == symconst_addr_ent) {
1413 ir_entity *ent = get_SymConst_entity(symc);
1415 if (get_entity_additional_properties(ent) & mtp_property_runtime)
1420 /* collect all call nodes */
1422 ++x->n_call_nodes_orig;
1424 callee = get_call_called_irg(call);
1426 inline_irg_env *callee_env = get_irg_link(callee);
1427 /* count all static callers */
1428 ++callee_env->n_callers;
1429 ++callee_env->n_callers_orig;
1431 /* link it in the list of possible inlinable entries */
1432 entry = obstack_alloc(env->obst, sizeof(*entry));
1434 entry->callee = callee;
1436 if (x->call_tail == NULL)
1437 x->call_head = entry;
1439 x->call_tail->next = entry;
1440 x->call_tail = entry;
1445 * Returns TRUE if the number of callers in 0 in the irg's environment,
1446 * hence this irg is a leave.
1448 INLINE static int is_leave(ir_graph *irg) {
1449 inline_irg_env *env = get_irg_link(irg);
1450 return env->n_call_nodes == 0;
1454 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1456 INLINE static int is_smaller(ir_graph *callee, int size) {
1457 inline_irg_env *env = get_irg_link(callee);
1458 return env->n_nodes < size;
1462 * Append the nodes of the list src to the nodes of the list in environment dst.
1464 static void append_call_list(struct obstack *obst, inline_irg_env *dst, call_entry *src) {
1465 call_entry *entry, *nentry;
1467 /* Note that the src list points to Call nodes in the inlined graph, but
1468 we need Call nodes in our graph. Luckily the inliner leaves this information
1469 in the link field. */
1470 for (entry = src; entry != NULL; entry = entry->next) {
1471 nentry = obstack_alloc(obst, sizeof(*nentry));
1472 nentry->call = get_irn_link(entry->call);
1473 nentry->callee = entry->callee;
1474 nentry->next = NULL;
1475 dst->call_tail->next = nentry;
1476 dst->call_tail = nentry;
1481 * Inlines small leave methods at call sites where the called address comes
1482 * from a Const node that references the entity representing the called
1484 * The size argument is a rough measure for the code size of the method:
1485 * Methods where the obstack containing the firm graph is smaller than
1488 void inline_leave_functions(int maxsize, int leavesize, int size, int ignore_runtime) {
1489 inline_irg_env *env;
1495 call_entry *entry, *tail;
1496 const call_entry *centry;
1497 struct obstack obst;
1498 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1500 if (!(get_opt_optimize() && get_opt_inline())) return;
1502 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1503 rem = current_ir_graph;
1504 obstack_init(&obst);
1506 /* extend all irgs by a temporary data structure for inlining. */
1507 n_irgs = get_irp_n_irgs();
1508 for (i = 0; i < n_irgs; ++i)
1509 set_irg_link(get_irp_irg(i), alloc_inline_irg_env(&obst));
1511 /* Precompute information in temporary data structure. */
1513 wenv.ignore_runtime = ignore_runtime;
1514 for (i = 0; i < n_irgs; ++i) {
1515 ir_graph *irg = get_irp_irg(i);
1517 assert(get_irg_phase_state(irg) != phase_building);
1518 free_callee_info(irg);
1520 wenv.x = get_irg_link(irg);
1521 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1524 /* -- and now inline. -- */
1526 /* Inline leaves recursively -- we might construct new leaves. */
1530 for (i = 0; i < n_irgs; ++i) {
1532 int phiproj_computed = 0;
1534 current_ir_graph = get_irp_irg(i);
1535 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1538 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1541 if (env->n_nodes > maxsize) break;
1544 callee = entry->callee;
1546 if (is_leave(callee) && is_smaller(callee, leavesize)) {
1547 if (!phiproj_computed) {
1548 phiproj_computed = 1;
1549 collect_phiprojs(current_ir_graph);
1551 did_inline = inline_method(call, callee);
1554 /* Do some statistics */
1555 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1557 env->got_inline = 1;
1558 --env->n_call_nodes;
1559 env->n_nodes += callee_env->n_nodes;
1560 --callee_env->n_callers;
1562 /* remove this call from the list */
1564 tail->next = entry->next;
1566 env->call_head = entry->next;
1572 env->call_tail = tail;
1574 } while (did_inline);
1576 /* inline other small functions. */
1577 for (i = 0; i < n_irgs; ++i) {
1579 int phiproj_computed = 0;
1581 current_ir_graph = get_irp_irg(i);
1582 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1584 /* note that the list of possible calls is updated during the process */
1586 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1590 callee = entry->callee;
1592 if (((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1593 (get_irg_inline_property(callee) >= irg_inline_forced))) {
1594 if (!phiproj_computed) {
1595 phiproj_computed = 1;
1596 collect_phiprojs(current_ir_graph);
1598 if (inline_method(call, callee)) {
1599 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1601 /* callee was inline. Append it's call list. */
1602 env->got_inline = 1;
1603 --env->n_call_nodes;
1604 append_call_list(&obst, env, callee_env->call_head);
1605 env->n_call_nodes += callee_env->n_call_nodes;
1606 env->n_nodes += callee_env->n_nodes;
1607 --callee_env->n_callers;
1609 /* after we have inlined callee, all called methods inside callee
1610 are now called once more */
1611 for (centry = callee_env->call_head; centry != NULL; centry = centry->next) {
1612 inline_irg_env *penv = get_irg_link(centry->callee);
1616 /* remove this call from the list */
1618 tail->next = entry->next;
1620 env->call_head = entry->next;
1626 env->call_tail = tail;
1629 for (i = 0; i < n_irgs; ++i) {
1630 irg = get_irp_irg(i);
1631 env = (inline_irg_env *)get_irg_link(irg);
1633 if (env->got_inline) {
1634 /* this irg got calls inlined */
1635 set_irg_outs_inconsistent(irg);
1636 set_irg_doms_inconsistent(irg);
1638 optimize_graph_df(irg);
1641 if (env->got_inline || (env->n_callers_orig != env->n_callers))
1642 DB((dbg, SET_LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1643 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1644 env->n_callers_orig, env->n_callers,
1645 get_entity_name(get_irg_entity(irg))));
1648 obstack_free(&obst, NULL);
1649 current_ir_graph = rem;
1652 /*******************************************************************/
1653 /* Code Placement. Pins all floating nodes to a block where they */
1654 /* will be executed only if needed. */
1655 /*******************************************************************/
1658 * Returns non-zero, is a block is not reachable from Start.
1660 * @param block the block to test
1663 is_Block_unreachable(ir_node *block) {
1664 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1668 * Find the earliest correct block for N. --- Place N into the
1669 * same Block as its dominance-deepest Input.
1671 * We have to avoid calls to get_nodes_block() here
1672 * because the graph is floating.
1674 * move_out_of_loops() expects that place_floats_early() have placed
1675 * all "living" nodes into a living block. That's why we must
1676 * move nodes in dead block with "live" successors into a valid
1678 * We move them just into the same block as it's successor (or
1679 * in case of a Phi into the effective use block). For Phi successors,
1680 * this may still be a dead block, but then there is no real use, as
1681 * the control flow will be dead later.
1684 place_floats_early(ir_node *n, waitq *worklist)
1688 /* we must not run into an infinite loop */
1689 assert(irn_not_visited(n));
1690 mark_irn_visited(n);
1692 /* Place floating nodes. */
1693 if (get_irn_pinned(n) == op_pin_state_floats) {
1694 ir_node *curr_block = get_irn_n(n, -1);
1695 int in_dead_block = is_Block_unreachable(curr_block);
1697 ir_node *b = NULL; /* The block to place this node in */
1699 assert(is_no_Block(n));
1701 if (is_irn_start_block_placed(n)) {
1702 /* These nodes will not be placed by the loop below. */
1703 b = get_irg_start_block(current_ir_graph);
1707 /* find the block for this node. */
1708 irn_arity = get_irn_arity(n);
1709 for (i = 0; i < irn_arity; i++) {
1710 ir_node *pred = get_irn_n(n, i);
1711 ir_node *pred_block;
1713 if ((irn_not_visited(pred))
1714 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1717 * If the current node is NOT in a dead block, but one of its
1718 * predecessors is, we must move the predecessor to a live block.
1719 * Such thing can happen, if global CSE chose a node from a dead block.
1720 * We move it simply to our block.
1721 * Note that neither Phi nor End nodes are floating, so we don't
1722 * need to handle them here.
1724 if (! in_dead_block) {
1725 if (get_irn_pinned(pred) == op_pin_state_floats &&
1726 is_Block_unreachable(get_irn_n(pred, -1)))
1727 set_nodes_block(pred, curr_block);
1729 place_floats_early(pred, worklist);
1733 * A node in the Bad block must stay in the bad block,
1734 * so don't compute a new block for it.
1739 /* Because all loops contain at least one op_pin_state_pinned node, now all
1740 our inputs are either op_pin_state_pinned or place_early() has already
1741 been finished on them. We do not have any unfinished inputs! */
1742 pred_block = get_irn_n(pred, -1);
1743 if ((!is_Block_dead(pred_block)) &&
1744 (get_Block_dom_depth(pred_block) > depth)) {
1746 depth = get_Block_dom_depth(pred_block);
1748 /* Avoid that the node is placed in the Start block */
1749 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)) {
1750 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1751 assert(b != get_irg_start_block(current_ir_graph));
1756 set_nodes_block(n, b);
1760 * Add predecessors of non floating nodes and non-floating predecessors
1761 * of floating nodes to worklist and fix their blocks if the are in dead block.
1763 irn_arity = get_irn_arity(n);
1765 if (get_irn_op(n) == op_End) {
1767 * Simplest case: End node. Predecessors are keep-alives,
1768 * no need to move out of dead block.
1770 for (i = -1; i < irn_arity; ++i) {
1771 ir_node *pred = get_irn_n(n, i);
1772 if (irn_not_visited(pred))
1773 waitq_put(worklist, pred);
1776 else if (is_Block(n)) {
1778 * Blocks: Predecessors are control flow, no need to move
1779 * them out of dead block.
1781 for (i = irn_arity - 1; i >= 0; --i) {
1782 ir_node *pred = get_irn_n(n, i);
1783 if (irn_not_visited(pred))
1784 waitq_put(worklist, pred);
1787 else if (is_Phi(n)) {
1789 ir_node *curr_block = get_irn_n(n, -1);
1790 int in_dead_block = is_Block_unreachable(curr_block);
1793 * Phi nodes: move nodes from dead blocks into the effective use
1794 * of the Phi-input if the Phi is not in a bad block.
1796 pred = get_irn_n(n, -1);
1797 if (irn_not_visited(pred))
1798 waitq_put(worklist, pred);
1800 for (i = irn_arity - 1; i >= 0; --i) {
1801 ir_node *pred = get_irn_n(n, i);
1803 if (irn_not_visited(pred)) {
1804 if (! in_dead_block &&
1805 get_irn_pinned(pred) == op_pin_state_floats &&
1806 is_Block_unreachable(get_irn_n(pred, -1))) {
1807 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1809 waitq_put(worklist, pred);
1815 ir_node *curr_block = get_irn_n(n, -1);
1816 int in_dead_block = is_Block_unreachable(curr_block);
1819 * All other nodes: move nodes from dead blocks into the same block.
1821 pred = get_irn_n(n, -1);
1822 if (irn_not_visited(pred))
1823 waitq_put(worklist, pred);
1825 for (i = irn_arity - 1; i >= 0; --i) {
1826 ir_node *pred = get_irn_n(n, i);
1828 if (irn_not_visited(pred)) {
1829 if (! in_dead_block &&
1830 get_irn_pinned(pred) == op_pin_state_floats &&
1831 is_Block_unreachable(get_irn_n(pred, -1))) {
1832 set_nodes_block(pred, curr_block);
1834 waitq_put(worklist, pred);
1841 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1842 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1843 * places all floating nodes reachable from its argument through floating
1844 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1846 static INLINE void place_early(waitq *worklist) {
1848 inc_irg_visited(current_ir_graph);
1850 /* this inits the worklist */
1851 place_floats_early(get_irg_end(current_ir_graph), worklist);
1853 /* Work the content of the worklist. */
1854 while (!waitq_empty(worklist)) {
1855 ir_node *n = waitq_get(worklist);
1856 if (irn_not_visited(n))
1857 place_floats_early(n, worklist);
1860 set_irg_outs_inconsistent(current_ir_graph);
1861 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
1865 * Compute the deepest common ancestor of block and dca.
1867 static ir_node *calc_dca(ir_node *dca, ir_node *block)
1871 /* we do not want to place nodes in dead blocks */
1872 if (is_Block_dead(block))
1875 /* We found a first legal placement. */
1876 if (!dca) return block;
1878 /* Find a placement that is dominates both, dca and block. */
1879 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1880 block = get_Block_idom(block);
1882 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1883 dca = get_Block_idom(dca);
1886 while (block != dca)
1887 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1892 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1893 * I.e., DCA is the block where we might place PRODUCER.
1894 * A data flow edge points from producer to consumer.
1897 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer)
1899 ir_node *block = NULL;
1901 /* Compute the latest block into which we can place a node so that it is
1903 if (get_irn_op(consumer) == op_Phi) {
1904 /* our consumer is a Phi-node, the effective use is in all those
1905 blocks through which the Phi-node reaches producer */
1907 ir_node *phi_block = get_nodes_block(consumer);
1908 irn_arity = get_irn_arity(consumer);
1910 for (i = 0; i < irn_arity; i++) {
1911 if (get_irn_n(consumer, i) == producer) {
1912 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1914 if (! is_Block_unreachable(new_block))
1915 block = calc_dca(block, new_block);
1920 block = get_irn_n(producer, -1);
1923 assert(is_no_Block(consumer));
1924 block = get_nodes_block(consumer);
1927 /* Compute the deepest common ancestor of block and dca. */
1928 return calc_dca(dca, block);
1931 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1933 static INLINE int get_irn_loop_depth(ir_node *n) {
1934 return get_loop_depth(get_irn_loop(n));
1938 * Move n to a block with less loop depth than it's current block. The
1939 * new block must be dominated by early.
1941 * @param n the node that should be moved
1942 * @param early the earliest block we can n move to
1944 static void move_out_of_loops(ir_node *n, ir_node *early)
1946 ir_node *best, *dca;
1950 /* Find the region deepest in the dominator tree dominating
1951 dca with the least loop nesting depth, but still dominated
1952 by our early placement. */
1953 dca = get_nodes_block(n);
1956 while (dca != early) {
1957 dca = get_Block_idom(dca);
1958 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1959 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1963 if (best != get_nodes_block(n)) {
1965 printf("Moving out of loop: "); DDMN(n);
1966 printf(" Outermost block: "); DDMN(early);
1967 printf(" Best block: "); DDMN(best);
1968 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1970 set_nodes_block(n, best);
1975 * Find the latest legal block for N and place N into the
1976 * `optimal' Block between the latest and earliest legal block.
1977 * The `optimal' block is the dominance-deepest block of those
1978 * with the least loop-nesting-depth. This places N out of as many
1979 * loops as possible and then makes it as control dependent as
1982 static void place_floats_late(ir_node *n, pdeq *worklist)
1987 assert(irn_not_visited(n)); /* no multiple placement */
1989 mark_irn_visited(n);
1991 /* no need to place block nodes, control nodes are already placed. */
1992 if ((get_irn_op(n) != op_Block) &&
1994 (get_irn_mode(n) != mode_X)) {
1995 /* Remember the early_blk placement of this block to move it
1996 out of loop no further than the early_blk placement. */
1997 early_blk = get_irn_n(n, -1);
2000 * BEWARE: Here we also get code, that is live, but
2001 * was in a dead block. If the node is life, but because
2002 * of CSE in a dead block, we still might need it.
2005 /* Assure that our users are all placed, except the Phi-nodes.
2006 --- Each data flow cycle contains at least one Phi-node. We
2007 have to break the `user has to be placed before the
2008 producer' dependence cycle and the Phi-nodes are the
2009 place to do so, because we need to base our placement on the
2010 final region of our users, which is OK with Phi-nodes, as they
2011 are op_pin_state_pinned, and they never have to be placed after a
2012 producer of one of their inputs in the same block anyway. */
2013 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2014 ir_node *succ = get_irn_out(n, i);
2015 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
2016 place_floats_late(succ, worklist);
2019 if (! is_Block_dead(early_blk)) {
2020 /* do only move things that where not dead */
2021 ir_op *op = get_irn_op(n);
2023 /* We have to determine the final block of this node... except for
2024 constants and Projs */
2025 if ((get_irn_pinned(n) == op_pin_state_floats) &&
2027 (op != op_SymConst) &&
2030 ir_node *dca = NULL; /* deepest common ancestor in the
2031 dominator tree of all nodes'
2032 blocks depending on us; our final
2033 placement has to dominate DCA. */
2034 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2035 ir_node *succ = get_irn_out(n, i);
2038 if (get_irn_op(succ) == op_End) {
2040 * This consumer is the End node, a keep alive edge.
2041 * This is not a real consumer, so we ignore it
2046 /* ignore if succ is in dead code */
2047 succ_blk = get_irn_n(succ, -1);
2048 if (is_Block_unreachable(succ_blk))
2050 dca = consumer_dom_dca(dca, succ, n);
2053 set_nodes_block(n, dca);
2054 move_out_of_loops(n, early_blk);
2060 /* Add predecessors of all non-floating nodes on list. (Those of floating
2061 nodes are placed already and therefore are marked.) */
2062 for (i = 0; i < get_irn_n_outs(n); i++) {
2063 ir_node *succ = get_irn_out(n, i);
2064 if (irn_not_visited(get_irn_out(n, i))) {
2065 pdeq_putr(worklist, succ);
2070 static INLINE void place_late(waitq *worklist) {
2072 inc_irg_visited(current_ir_graph);
2074 /* This fills the worklist initially. */
2075 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
2077 /* And now empty the worklist again... */
2078 while (!waitq_empty(worklist)) {
2079 ir_node *n = waitq_get(worklist);
2080 if (irn_not_visited(n))
2081 place_floats_late(n, worklist);
2085 void place_code(ir_graph *irg) {
2087 ir_graph *rem = current_ir_graph;
2089 current_ir_graph = irg;
2091 if (!(get_opt_optimize() && get_opt_global_cse())) return;
2093 /* Handle graph state */
2094 assert(get_irg_phase_state(irg) != phase_building);
2097 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
2098 free_loop_information(irg);
2099 construct_backedges(irg);
2102 /* Place all floating nodes as early as possible. This guarantees
2103 a legal code placement. */
2104 worklist = new_waitq();
2105 place_early(worklist);
2107 /* place_early() invalidates the outs, place_late needs them. */
2108 compute_irg_outs(irg);
2110 /* Now move the nodes down in the dominator tree. This reduces the
2111 unnecessary executions of the node. */
2112 place_late(worklist);
2114 set_irg_outs_inconsistent(current_ir_graph);
2115 set_irg_loopinfo_inconsistent(current_ir_graph);
2116 del_waitq(worklist);
2117 current_ir_graph = rem;
2121 * Called by walker of remove_critical_cf_edges().
2123 * Place an empty block to an edge between a blocks of multiple
2124 * predecessors and a block of multiple successors.
2127 * @param env Environment of walker. The changed field.
2129 static void walk_critical_cf_edges(ir_node *n, void *env) {
2131 ir_node *pre, *block, *jmp;
2133 ir_graph *irg = get_irn_irg(n);
2135 /* Block has multiple predecessors */
2136 arity = get_irn_arity(n);
2138 if (n == get_irg_end_block(irg))
2139 return; /* No use to add a block here. */
2141 for (i = 0; i < arity; ++i) {
2144 pre = get_irn_n(n, i);
2145 cfop = get_irn_op(skip_Proj(pre));
2146 /* Predecessor has multiple successors. Insert new control flow edge but
2147 ignore exception edges. */
2148 if (! is_op_fragile(cfop) && is_op_forking(cfop)) {
2149 /* set predecessor of new block */
2150 block = new_r_Block(irg, 1, &pre);
2151 /* insert new jmp node to new block */
2152 jmp = new_r_Jmp(irg, block);
2153 /* set successor of new block */
2154 set_irn_n(n, i, jmp);
2156 } /* predecessor has multiple successors */
2157 } /* for all predecessors */
2158 } /* n is a multi-entry block */
2161 void remove_critical_cf_edges(ir_graph *irg) {
2164 irg_block_walk_graph(irg, NULL, walk_critical_cf_edges, &changed);
2166 /* control flow changed */
2167 set_irg_outs_inconsistent(irg);
2168 set_irg_extblk_inconsistent(irg);
2169 set_irg_doms_inconsistent(irg);
2170 set_irg_loopinfo_inconsistent(irg);