2 * Copyright (C) 1995-2007 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Optimizations for a whole ir graph, i.e., a procedure.
23 * @author Christian Schaefer, Goetz Lindenmaier, Sebastian Felis,
34 #include "irgraph_t.h"
47 #include "pdeq.h" /* Fuer code placement */
52 #include "irbackedge_t.h"
59 #include "iredges_t.h"
62 /*------------------------------------------------------------------*/
63 /* apply optimizations of iropt to all nodes. */
64 /*------------------------------------------------------------------*/
67 * A wrapper around optimize_inplace_2() to be called from a walker.
69 static void optimize_in_place_wrapper (ir_node *n, void *env) {
70 ir_node *optimized = optimize_in_place_2(n);
71 if (optimized != n) exchange (n, optimized);
75 * Do local optimizations for a node.
77 * @param n the IR-node where to start. Typically the End node
80 * @note current_ir_graph must be set
82 static INLINE void do_local_optimize(ir_node *n) {
83 /* Handle graph state */
84 assert(get_irg_phase_state(current_ir_graph) != phase_building);
86 if (get_opt_global_cse())
87 set_irg_pinned(current_ir_graph, op_pin_state_floats);
88 set_irg_outs_inconsistent(current_ir_graph);
89 set_irg_doms_inconsistent(current_ir_graph);
90 set_irg_loopinfo_inconsistent(current_ir_graph);
92 /* Clean the value_table in irg for the CSE. */
93 del_identities(current_ir_graph->value_table);
94 current_ir_graph->value_table = new_identities();
96 /* walk over the graph */
97 irg_walk(n, firm_clear_link, optimize_in_place_wrapper, NULL);
100 /* Applies local optimizations (see iropt.h) to all nodes reachable from node n */
101 void local_optimize_node(ir_node *n) {
102 ir_graph *rem = current_ir_graph;
103 current_ir_graph = get_irn_irg(n);
105 do_local_optimize(n);
107 current_ir_graph = rem;
111 * Block-Walker: uses dominance depth to mark dead blocks.
113 static void kill_dead_blocks(ir_node *block, void *env) {
114 if (get_Block_dom_depth(block) < 0) {
116 * Note that the new dominance code correctly handles
117 * the End block, i.e. it is always reachable from Start
119 set_Block_dead(block);
123 /* Applies local optimizations (see iropt.h) to all nodes reachable from node n. */
124 void local_optimize_graph(ir_graph *irg) {
125 ir_graph *rem = current_ir_graph;
126 current_ir_graph = irg;
128 if (get_irg_dom_state(irg) == dom_consistent)
129 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
131 do_local_optimize(get_irg_end(irg));
133 current_ir_graph = rem;
137 * Enqueue all users of a node to a wait queue.
138 * Handles mode_T nodes.
140 static void enqueue_users(ir_node *n, pdeq *waitq) {
141 const ir_edge_t *edge;
143 foreach_out_edge(n, edge) {
144 ir_node *succ = get_edge_src_irn(edge);
146 if (get_irn_link(succ) != waitq) {
147 pdeq_putr(waitq, succ);
148 set_irn_link(succ, waitq);
150 if (get_irn_mode(succ) == mode_T) {
151 /* A mode_T node has Proj's. Because most optimizations
152 run on the Proj's we have to enqueue them also. */
153 enqueue_users(succ, waitq);
159 * Data flow optimization walker.
160 * Optimizes all nodes and enqueue it's users
163 static void opt_walker(ir_node *n, void *env) {
167 optimized = optimize_in_place_2(n);
168 set_irn_link(optimized, NULL);
170 if (optimized != n) {
171 enqueue_users(n, waitq);
172 exchange(n, optimized);
176 /* Applies local optimizations to all nodes in the graph until fixpoint. */
177 void optimize_graph_df(ir_graph *irg) {
178 pdeq *waitq = new_pdeq();
179 int state = edges_activated(irg);
180 ir_graph *rem = current_ir_graph;
184 current_ir_graph = irg;
189 if (get_opt_global_cse())
190 set_irg_pinned(current_ir_graph, op_pin_state_floats);
192 /* Clean the value_table in irg for the CSE. */
193 del_identities(irg->value_table);
194 irg->value_table = new_identities();
196 if (get_irg_dom_state(irg) == dom_consistent)
197 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
199 /* invalidate info */
200 set_irg_outs_inconsistent(irg);
201 set_irg_doms_inconsistent(irg);
202 set_irg_loopinfo_inconsistent(irg);
204 set_using_irn_link(irg);
206 /* walk over the graph, but don't touch keep-alives */
207 irg_walk(get_irg_end_block(irg), NULL, opt_walker, waitq);
209 end = get_irg_end(irg);
211 /* optimize keep-alives by removing superfluous ones */
212 for (i = get_End_n_keepalives(end) - 1; i >= 0; --i) {
213 ir_node *ka = get_End_keepalive(end, i);
215 if (irn_visited(ka) && !is_irn_keep(ka)) {
216 /* this node can be regularly visited, no need to keep it */
217 set_End_keepalive(end, i, get_irg_bad(irg));
220 /* now walk again and visit all not yet visited nodes */
221 set_irg_visited(current_ir_graph, get_irg_visited(irg) - 1);
222 irg_walk(get_irg_end(irg), NULL, opt_walker, waitq);
224 /* finish the wait queue */
225 while (! pdeq_empty(waitq)) {
226 ir_node *n = pdeq_getl(waitq);
228 opt_walker(n, waitq);
233 clear_using_irn_link(irg);
236 edges_deactivate(irg);
238 current_ir_graph = rem;
242 /*------------------------------------------------------------------*/
243 /* Routines for dead node elimination / copying garbage collection */
244 /* of the obstack. */
245 /*------------------------------------------------------------------*/
248 * Remember the new node in the old node by using a field all nodes have.
250 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
253 * Get this new node, before the old node is forgotten.
255 #define get_new_node(oldn) get_irn_link(oldn)
258 * Check if a new node was set.
260 #define has_new_node(n) (get_new_node(n) != NULL)
263 * We use the block_visited flag to mark that we have computed the
264 * number of useful predecessors for this block.
265 * Further we encode the new arity in this flag in the old blocks.
266 * Remembering the arity is useful, as it saves a lot of pointer
267 * accesses. This function is called for all Phi and Block nodes
271 compute_new_arity(ir_node *b) {
272 int i, res, irn_arity;
275 irg_v = get_irg_block_visited(current_ir_graph);
276 block_v = get_Block_block_visited(b);
277 if (block_v >= irg_v) {
278 /* we computed the number of preds for this block and saved it in the
280 return block_v - irg_v;
282 /* compute the number of good predecessors */
283 res = irn_arity = get_irn_arity(b);
284 for (i = 0; i < irn_arity; i++)
285 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
286 /* save it in the flag. */
287 set_Block_block_visited(b, irg_v + res);
293 * Copies the node to the new obstack. The Ins of the new node point to
294 * the predecessors on the old obstack. For block/phi nodes not all
295 * predecessors might be copied. n->link points to the new node.
296 * For Phi and Block nodes the function allocates in-arrays with an arity
297 * only for useful predecessors. The arity is determined by counting
298 * the non-bad predecessors of the block.
300 * @param n The node to be copied
301 * @param env if non-NULL, the node number attribute will be copied to the new node
303 * Note: Also used for loop unrolling.
305 static void copy_node(ir_node *n, void *env) {
308 ir_op *op = get_irn_op(n);
310 /* The end node looses it's flexible in array. This doesn't matter,
311 as dead node elimination builds End by hand, inlineing doesn't use
313 /* assert(op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
316 /* node copied already */
318 } else if (op == op_Block) {
320 new_arity = compute_new_arity(n);
321 n->attr.block.graph_arr = NULL;
323 block = get_nodes_block(n);
325 new_arity = compute_new_arity(block);
327 new_arity = get_irn_arity(n);
330 nn = new_ir_node(get_irn_dbg_info(n),
337 /* Copy the attributes. These might point to additional data. If this
338 was allocated on the old obstack the pointers now are dangling. This
339 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
340 copy_node_attr(n, nn);
344 int copy_node_nr = env != NULL;
346 /* for easier debugging, we want to copy the node numbers too */
347 nn->node_nr = n->node_nr;
353 hook_dead_node_elim_subst(current_ir_graph, n, nn);
357 * Copies new predecessors of old node to new node remembered in link.
358 * Spare the Bad predecessors of Phi and Block nodes.
361 copy_preds(ir_node *n, void *env) {
365 nn = get_new_node(n);
368 /* Don't copy Bad nodes. */
370 irn_arity = get_irn_arity(n);
371 for (i = 0; i < irn_arity; i++) {
372 if (! is_Bad(get_irn_n(n, i))) {
373 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
374 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
378 /* repair the block visited flag from above misuse. Repair it in both
379 graphs so that the old one can still be used. */
380 set_Block_block_visited(nn, 0);
381 set_Block_block_visited(n, 0);
382 /* Local optimization could not merge two subsequent blocks if
383 in array contained Bads. Now it's possible.
384 We don't call optimize_in_place as it requires
385 that the fields in ir_graph are set properly. */
386 if ((get_opt_control_flow_straightening()) &&
387 (get_Block_n_cfgpreds(nn) == 1) &&
388 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
389 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
391 /* Jmp jumps into the block it is in -- deal self cycle. */
392 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
393 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
398 } else if (get_irn_op(n) == op_Phi) {
399 /* Don't copy node if corresponding predecessor in block is Bad.
400 The Block itself should not be Bad. */
401 block = get_nodes_block(n);
402 set_irn_n(nn, -1, get_new_node(block));
404 irn_arity = get_irn_arity(n);
405 for (i = 0; i < irn_arity; i++) {
406 if (! is_Bad(get_irn_n(block, i))) {
407 set_irn_n(nn, j, get_new_node(get_irn_n(n, i)));
408 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
412 /* If the pre walker reached this Phi after the post walker visited the
413 block block_visited is > 0. */
414 set_Block_block_visited(get_nodes_block(n), 0);
415 /* Compacting the Phi's ins might generate Phis with only one
417 if (get_irn_arity(nn) == 1)
418 exchange(nn, get_irn_n(nn, 0));
420 irn_arity = get_irn_arity(n);
421 for (i = -1; i < irn_arity; i++)
422 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
424 /* Now the new node is complete. We can add it to the hash table for CSE.
425 @@@ inlining aborts if we identify End. Why? */
426 if (get_irn_op(nn) != op_End)
427 add_identities(current_ir_graph->value_table, nn);
431 * Copies the graph recursively, compacts the keep-alives of the end node.
433 * @param irg the graph to be copied
434 * @param copy_node_nr If non-zero, the node number will be copied
436 static void copy_graph(ir_graph *irg, int copy_node_nr) {
437 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
438 ir_node *ka; /* keep alive */
442 /* Some nodes must be copied by hand, sigh */
443 vfl = get_irg_visited(irg);
444 set_irg_visited(irg, vfl + 1);
446 oe = get_irg_end(irg);
447 mark_irn_visited(oe);
448 /* copy the end node by hand, allocate dynamic in array! */
449 ne = new_ir_node(get_irn_dbg_info(oe),
456 /* Copy the attributes. Well, there might be some in the future... */
457 copy_node_attr(oe, ne);
458 set_new_node(oe, ne);
460 /* copy the Bad node */
461 ob = get_irg_bad(irg);
462 mark_irn_visited(ob);
463 nb = new_ir_node(get_irn_dbg_info(ob),
470 copy_node_attr(ob, nb);
471 set_new_node(ob, nb);
473 /* copy the NoMem node */
474 om = get_irg_no_mem(irg);
475 mark_irn_visited(om);
476 nm = new_ir_node(get_irn_dbg_info(om),
483 copy_node_attr(om, nm);
484 set_new_node(om, nm);
486 /* copy the live nodes */
487 set_irg_visited(irg, vfl);
488 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
490 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
492 /* visit the anchors as well */
493 for (i = anchor_max - 1; i >= 0; --i) {
494 ir_node *n = irg->anchors[i];
496 if (n && (get_irn_visited(n) <= vfl)) {
497 set_irg_visited(irg, vfl);
498 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
502 /* copy_preds for the end node ... */
503 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
505 /*- ... and now the keep alives. -*/
506 /* First pick the not marked block nodes and walk them. We must pick these
507 first as else we will oversee blocks reachable from Phis. */
508 irn_arity = get_End_n_keepalives(oe);
509 for (i = 0; i < irn_arity; i++) {
510 ka = get_End_keepalive(oe, i);
512 if (get_irn_visited(ka) <= vfl) {
513 /* We must keep the block alive and copy everything reachable */
514 set_irg_visited(irg, vfl);
515 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
517 add_End_keepalive(ne, get_new_node(ka));
521 /* Now pick other nodes. Here we will keep all! */
522 irn_arity = get_End_n_keepalives(oe);
523 for (i = 0; i < irn_arity; i++) {
524 ka = get_End_keepalive(oe, i);
526 if (get_irn_visited(ka) <= vfl) {
527 /* We didn't copy the node yet. */
528 set_irg_visited(irg, vfl);
529 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
531 add_End_keepalive(ne, get_new_node(ka));
535 /* start block sometimes only reached after keep alives */
536 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
537 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
541 * Copies the graph reachable from current_ir_graph->end to the obstack
542 * in current_ir_graph and fixes the environment.
543 * Then fixes the fields in current_ir_graph containing nodes of the
546 * @param copy_node_nr If non-zero, the node number will be copied
549 copy_graph_env(int copy_node_nr) {
550 ir_graph *irg = current_ir_graph;
551 ir_node *old_end, *n;
554 /* remove end_except and end_reg nodes */
555 old_end = get_irg_end(irg);
556 set_irg_end_except (irg, old_end);
557 set_irg_end_reg (irg, old_end);
559 /* Not all nodes remembered in irg might be reachable
560 from the end node. Assure their link is set to NULL, so that
561 we can test whether new nodes have been computed. */
562 for (i = anchor_max - 1; i >= 0; --i) {
564 set_new_node(irg->anchors[i], NULL);
566 /* we use the block walk flag for removing Bads from Blocks ins. */
567 inc_irg_block_visited(irg);
570 copy_graph(irg, copy_node_nr);
572 /* fix the fields in irg */
573 old_end = get_irg_end(irg);
574 for (i = anchor_max - 1; i >= 0; --i) {
577 irg->anchors[i] = get_new_node(n);
583 * Copies all reachable nodes to a new obstack. Removes bad inputs
584 * from block nodes and the corresponding inputs from Phi nodes.
585 * Merges single exit blocks with single entry blocks and removes
587 * Adds all new nodes to a new hash table for CSE. Does not
588 * perform CSE, so the hash table might contain common subexpressions.
591 dead_node_elimination(ir_graph *irg) {
592 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
594 int rem_ipview = get_interprocedural_view();
595 struct obstack *graveyard_obst = NULL;
596 struct obstack *rebirth_obst = NULL;
597 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
599 /* inform statistics that we started a dead-node elimination run */
600 hook_dead_node_elim(irg, 1);
602 /* Remember external state of current_ir_graph. */
603 rem = current_ir_graph;
604 current_ir_graph = irg;
605 set_interprocedural_view(0);
607 assert(get_irg_phase_state(irg) != phase_building);
609 /* Handle graph state */
610 free_callee_info(irg);
614 /* @@@ so far we loose loops when copying */
615 free_loop_information(irg);
617 set_irg_doms_inconsistent(irg);
619 /* A quiet place, where the old obstack can rest in peace,
620 until it will be cremated. */
621 graveyard_obst = irg->obst;
623 /* A new obstack, where the reachable nodes will be copied to. */
624 rebirth_obst = xmalloc(sizeof(*rebirth_obst));
625 irg->obst = rebirth_obst;
626 obstack_init(irg->obst);
627 irg->last_node_idx = 0;
629 /* We also need a new value table for CSE */
630 del_identities(irg->value_table);
631 irg->value_table = new_identities();
633 /* Copy the graph from the old to the new obstack */
634 copy_graph_env(/*copy_node_nr=*/1);
636 /* Free memory from old unoptimized obstack */
637 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
638 xfree (graveyard_obst); /* ... then free it. */
640 /* inform statistics that the run is over */
641 hook_dead_node_elim(irg, 0);
643 current_ir_graph = rem;
644 set_interprocedural_view(rem_ipview);
649 * Relink bad predecessors of a block and store the old in array to the
650 * link field. This function is called by relink_bad_predecessors().
651 * The array of link field starts with the block operand at position 0.
652 * If block has bad predecessors, create a new in array without bad preds.
653 * Otherwise let in array untouched.
655 static void relink_bad_block_predecessors(ir_node *n, void *env) {
656 ir_node **new_in, *irn;
657 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
659 /* if link field of block is NULL, look for bad predecessors otherwise
660 this is already done */
661 if (get_irn_op(n) == op_Block &&
662 get_irn_link(n) == NULL) {
664 /* save old predecessors in link field (position 0 is the block operand)*/
665 set_irn_link(n, get_irn_in(n));
667 /* count predecessors without bad nodes */
668 old_irn_arity = get_irn_arity(n);
669 for (i = 0; i < old_irn_arity; i++)
670 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
672 /* arity changing: set new predecessors without bad nodes */
673 if (new_irn_arity < old_irn_arity) {
674 /* Get new predecessor array. We do not resize the array, as we must
675 keep the old one to update Phis. */
676 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
678 /* set new predecessors in array */
681 for (i = 0; i < old_irn_arity; i++) {
682 irn = get_irn_n(n, i);
684 new_in[new_irn_n] = irn;
685 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
689 /* ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity); */
690 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
692 } /* ir node has bad predecessors */
693 } /* Block is not relinked */
697 * Relinks Bad predecessors from Blocks and Phis called by walker
698 * remove_bad_predecesors(). If n is a Block, call
699 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
700 * function of Phi's Block. If this block has bad predecessors, relink preds
703 static void relink_bad_predecessors(ir_node *n, void *env) {
704 ir_node *block, **old_in;
705 int i, old_irn_arity, new_irn_arity;
707 /* relink bad predecessors of a block */
708 if (get_irn_op(n) == op_Block)
709 relink_bad_block_predecessors(n, env);
711 /* If Phi node relink its block and its predecessors */
712 if (get_irn_op(n) == op_Phi) {
714 /* Relink predecessors of phi's block */
715 block = get_nodes_block(n);
716 if (get_irn_link(block) == NULL)
717 relink_bad_block_predecessors(block, env);
719 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
720 old_irn_arity = ARR_LEN(old_in);
722 /* Relink Phi predecessors if count of predecessors changed */
723 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
724 /* set new predecessors in array
725 n->in[0] remains the same block */
727 for(i = 1; i < old_irn_arity; i++)
728 if (!is_Bad((ir_node *)old_in[i])) {
729 n->in[new_irn_arity] = n->in[i];
730 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
734 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
735 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
737 } /* n is a Phi node */
741 * Removes Bad Bad predecessors from Blocks and the corresponding
742 * inputs to Phi nodes as in dead_node_elimination but without
744 * On walking up set the link field to NULL, on walking down call
745 * relink_bad_predecessors() (This function stores the old in array
746 * to the link field and sets a new in array if arity of predecessors
749 void remove_bad_predecessors(ir_graph *irg) {
750 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
757 __)|_| | \_/ | \_/(/_ |_/\__|__
759 The following stuff implements a facility that automatically patches
760 registered ir_node pointers to the new node when a dead node elimination occurs.
763 struct _survive_dce_t {
767 hook_entry_t dead_node_elim;
768 hook_entry_t dead_node_elim_subst;
771 typedef struct _survive_dce_list_t {
772 struct _survive_dce_list_t *next;
774 } survive_dce_list_t;
776 static void dead_node_hook(void *context, ir_graph *irg, int start) {
777 survive_dce_t *sd = context;
779 /* Create a new map before the dead node elimination is performed. */
781 sd->new_places = pmap_create_ex(pmap_count(sd->places));
783 /* Patch back all nodes if dead node elimination is over and something is to be done. */
784 pmap_destroy(sd->places);
785 sd->places = sd->new_places;
786 sd->new_places = NULL;
791 * Hook called when dead node elimination replaces old by nw.
793 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw) {
794 survive_dce_t *sd = context;
795 survive_dce_list_t *list = pmap_get(sd->places, old);
797 /* If the node is to be patched back, write the new address to all registered locations. */
799 survive_dce_list_t *p;
801 for (p = list; p; p = p->next)
804 pmap_insert(sd->new_places, nw, list);
809 * Make a new Survive DCE environment.
811 survive_dce_t *new_survive_dce(void) {
812 survive_dce_t *res = xmalloc(sizeof(res[0]));
813 obstack_init(&res->obst);
814 res->places = pmap_create();
815 res->new_places = NULL;
817 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
818 res->dead_node_elim.context = res;
819 res->dead_node_elim.next = NULL;
821 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
822 res->dead_node_elim_subst.context = res;
823 res->dead_node_elim_subst.next = NULL;
825 #ifndef FIRM_ENABLE_HOOKS
826 assert(0 && "need hooks enabled");
829 register_hook(hook_dead_node_elim, &res->dead_node_elim);
830 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
835 * Free a Survive DCE environment.
837 void free_survive_dce(survive_dce_t *sd) {
838 obstack_free(&sd->obst, NULL);
839 pmap_destroy(sd->places);
840 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
841 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
846 * Register a node pointer to be patched upon DCE.
847 * When DCE occurs, the node pointer specified by @p place will be
848 * patched to the new address of the node it is pointing to.
850 * @param sd The Survive DCE environment.
851 * @param place The address of the node pointer.
853 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place) {
854 if (*place != NULL) {
855 ir_node *irn = *place;
856 survive_dce_list_t *curr = pmap_get(sd->places, irn);
857 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw[0]));
862 pmap_insert(sd->places, irn, nw);
866 /*--------------------------------------------------------------------*/
867 /* Functionality for inlining */
868 /*--------------------------------------------------------------------*/
871 * Copy node for inlineing. Updates attributes that change when
872 * inlineing but not for dead node elimination.
874 * Copies the node by calling copy_node() and then updates the entity if
875 * it's a local one. env must be a pointer of the frame type of the
876 * inlined procedure. The new entities must be in the link field of
880 copy_node_inline(ir_node *n, void *env) {
882 ir_type *frame_tp = (ir_type *)env;
885 if (get_irn_op(n) == op_Sel) {
886 nn = get_new_node (n);
888 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
889 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
891 } else if (get_irn_op(n) == op_Block) {
892 nn = get_new_node (n);
893 nn->attr.block.irg = current_ir_graph;
898 * Walker: checks if P_value_arg_base is used.
900 static void find_addr(ir_node *node, void *env) {
901 int *allow_inline = env;
902 if (is_Proj(node) && get_irn_op(get_Proj_pred(node)) == op_Start) {
903 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
909 * Check if we can inline a given call.
910 * Currently, we cannot inline two cases:
911 * - call with compound arguments
912 * - graphs that take the address of a parameter
914 * check these conditions here
916 static int can_inline(ir_node *call, ir_graph *called_graph) {
917 ir_type *call_type = get_Call_type(call);
918 int params, ress, i, res;
919 assert(is_Method_type(call_type));
921 params = get_method_n_params(call_type);
922 ress = get_method_n_ress(call_type);
924 /* check parameters for compound arguments */
925 for (i = 0; i < params; ++i) {
926 ir_type *p_type = get_method_param_type(call_type, i);
928 if (is_compound_type(p_type))
932 /* check results for compound arguments */
933 for (i = 0; i < ress; ++i) {
934 ir_type *r_type = get_method_res_type(call_type, i);
936 if (is_compound_type(r_type))
941 irg_walk_graph(called_graph, find_addr, NULL, &res);
946 /* Inlines a method at the given call site. */
947 int inline_method(ir_node *call, ir_graph *called_graph) {
949 ir_node *post_call, *post_bl;
950 ir_node *in[pn_Start_max];
951 ir_node *end, *end_bl;
955 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
957 ir_type *called_frame;
958 irg_inline_property prop = get_irg_inline_property(called_graph);
960 if ( (prop < irg_inline_forced) &&
961 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
963 /* Do not inline variadic functions. */
964 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
967 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
968 get_method_n_params(get_Call_type(call)));
971 * currently, we cannot inline two cases:
972 * - call with compound arguments
973 * - graphs that take the address of a parameter
975 if (! can_inline(call, called_graph))
978 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
979 rem_opt = get_opt_optimize();
982 /* Handle graph state */
983 assert(get_irg_phase_state(current_ir_graph) != phase_building);
984 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
985 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
986 set_irg_outs_inconsistent(current_ir_graph);
987 set_irg_extblk_inconsistent(current_ir_graph);
988 set_irg_doms_inconsistent(current_ir_graph);
989 set_irg_loopinfo_inconsistent(current_ir_graph);
990 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
992 /* -- Check preconditions -- */
993 assert(is_Call(call));
994 /* @@@ does not work for InterfaceIII.java after cgana
995 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
996 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
997 get_Call_type(call)));
999 if (called_graph == current_ir_graph) {
1000 set_optimize(rem_opt);
1004 /* here we know we WILL inline, so inform the statistics */
1005 hook_inline(call, called_graph);
1007 /* -- Decide how to handle exception control flow: Is there a handler
1008 for the Call node, or do we branch directly to End on an exception?
1010 0 There is a handler.
1012 2 Exception handling not represented in Firm. -- */
1014 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
1015 for (proj = get_irn_link(call); proj; proj = get_irn_link(proj)) {
1016 assert(is_Proj(proj));
1017 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
1018 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
1020 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
1021 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
1022 else { exc_handling = 2; } /* !Mproj && !Xproj */
1026 the procedure and later replaces the Start node of the called graph.
1027 Post_call is the old Call node and collects the results of the called
1028 graph. Both will end up being a tuple. -- */
1029 post_bl = get_nodes_block(call);
1030 set_irg_current_block(current_ir_graph, post_bl);
1031 /* XxMxPxPxPxT of Start + parameter of Call */
1032 in[pn_Start_X_initial_exec] = new_Jmp();
1033 in[pn_Start_M] = get_Call_mem(call);
1034 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
1035 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
1036 in[pn_Start_P_tls] = get_irg_tls(current_ir_graph);
1037 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
1038 /* in[pn_Start_P_value_arg_base] = ??? */
1039 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
1040 pre_call = new_Tuple(pn_Start_max - 1, in);
1044 The new block gets the ins of the old block, pre_call and all its
1045 predecessors and all Phi nodes. -- */
1046 part_block(pre_call);
1048 /* -- Prepare state for dead node elimination -- */
1049 /* Visited flags in calling irg must be >= flag in called irg.
1050 Else walker and arity computation will not work. */
1051 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
1052 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
1053 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
1054 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
1055 /* Set pre_call as new Start node in link field of the start node of
1056 calling graph and pre_calls block as new block for the start block
1058 Further mark these nodes so that they are not visited by the
1060 set_irn_link(get_irg_start(called_graph), pre_call);
1061 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
1062 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1063 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
1064 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
1065 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
1067 /* Initialize for compaction of in arrays */
1068 inc_irg_block_visited(current_ir_graph);
1070 /* -- Replicate local entities of the called_graph -- */
1071 /* copy the entities. */
1072 called_frame = get_irg_frame_type(called_graph);
1073 for (i = 0; i < get_class_n_members(called_frame); i++) {
1074 ir_entity *new_ent, *old_ent;
1075 old_ent = get_class_member(called_frame, i);
1076 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
1077 set_entity_link(old_ent, new_ent);
1080 /* visited is > than that of called graph. With this trick visited will
1081 remain unchanged so that an outer walker, e.g., searching the call nodes
1082 to inline, calling this inline will not visit the inlined nodes. */
1083 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
1085 /* -- Performing dead node elimination inlines the graph -- */
1086 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1088 /* @@@ endless loops are not copied!! -- they should be, I think... */
1089 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
1090 get_irg_frame_type(called_graph));
1092 /* Repair called_graph */
1093 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
1094 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
1095 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1097 /* -- Merge the end of the inlined procedure with the call site -- */
1098 /* We will turn the old Call node into a Tuple with the following
1101 0: Phi of all Memories of Return statements.
1102 1: Jmp from new Block that merges the control flow from all exception
1103 predecessors of the old end block.
1104 2: Tuple of all arguments.
1105 3: Phi of Exception memories.
1106 In case the old Call directly branches to End on an exception we don't
1107 need the block merging all exceptions nor the Phi of the exception
1111 /* -- Precompute some values -- */
1112 end_bl = get_new_node(get_irg_end_block(called_graph));
1113 end = get_new_node(get_irg_end(called_graph));
1114 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1115 n_res = get_method_n_ress(get_Call_type(call));
1117 res_pred = xmalloc(n_res * sizeof(*res_pred));
1118 cf_pred = xmalloc(arity * sizeof(*res_pred));
1120 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
1122 /* -- archive keepalives -- */
1123 irn_arity = get_irn_arity(end);
1124 for (i = 0; i < irn_arity; i++)
1125 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
1127 /* The new end node will die. We need not free as the in array is on the obstack:
1128 copy_node() only generated 'D' arrays. */
1130 /* -- Replace Return nodes by Jump nodes. -- */
1132 for (i = 0; i < arity; i++) {
1134 ret = get_irn_n(end_bl, i);
1135 if (is_Return(ret)) {
1136 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
1140 set_irn_in(post_bl, n_ret, cf_pred);
1142 /* -- Build a Tuple for all results of the method.
1143 Add Phi node if there was more than one Return. -- */
1144 turn_into_tuple(post_call, 4); /* FIXME: is th 4 corrct here ? */
1145 /* First the Memory-Phi */
1147 for (i = 0; i < arity; i++) {
1148 ret = get_irn_n(end_bl, i);
1149 if (is_Return(ret)) {
1150 cf_pred[n_ret] = get_Return_mem(ret);
1154 phi = new_Phi(n_ret, cf_pred, mode_M);
1155 set_Tuple_pred(call, pn_Call_M_regular, phi);
1156 /* Conserve Phi-list for further inlinings -- but might be optimized */
1157 if (get_nodes_block(phi) == post_bl) {
1158 set_irn_link(phi, get_irn_link(post_bl));
1159 set_irn_link(post_bl, phi);
1161 /* Now the real results */
1163 for (j = 0; j < n_res; j++) {
1165 for (i = 0; i < arity; i++) {
1166 ret = get_irn_n(end_bl, i);
1167 if (get_irn_op(ret) == op_Return) {
1168 cf_pred[n_ret] = get_Return_res(ret, j);
1173 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1177 /* Conserve Phi-list for further inlinings -- but might be optimized */
1178 if (get_nodes_block(phi) == post_bl) {
1179 set_irn_link(phi, get_irn_link(post_bl));
1180 set_irn_link(post_bl, phi);
1183 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1185 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1187 /* Finally the exception control flow.
1188 We have two (three) possible situations:
1189 First if the Call branches to an exception handler: We need to add a Phi node to
1190 collect the memory containing the exception objects. Further we need
1191 to add another block to get a correct representation of this Phi. To
1192 this block we add a Jmp that resolves into the X output of the Call
1193 when the Call is turned into a tuple.
1194 Second the Call branches to End, the exception is not handled. Just
1195 add all inlined exception branches to the End node.
1196 Third: there is no Exception edge at all. Handle as case two. */
1197 if (exc_handling == 0) {
1199 for (i = 0; i < arity; i++) {
1201 ret = get_irn_n(end_bl, i);
1202 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1203 cf_pred[n_exc] = ret;
1208 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1209 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1210 /* The Phi for the memories with the exception objects */
1212 for (i = 0; i < arity; i++) {
1214 ret = skip_Proj(get_irn_n(end_bl, i));
1216 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1218 } else if (is_fragile_op(ret)) {
1219 /* We rely that all cfops have the memory output at the same position. */
1220 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1222 } else if (get_irn_op(ret) == op_Raise) {
1223 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1227 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1229 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1230 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1233 ir_node *main_end_bl;
1234 int main_end_bl_arity;
1235 ir_node **end_preds;
1237 /* assert(exc_handling == 1 || no exceptions. ) */
1239 for (i = 0; i < arity; i++) {
1240 ir_node *ret = get_irn_n(end_bl, i);
1242 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1243 cf_pred[n_exc] = ret;
1247 main_end_bl = get_irg_end_block(current_ir_graph);
1248 main_end_bl_arity = get_irn_arity(main_end_bl);
1249 end_preds = xmalloc ((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1251 for (i = 0; i < main_end_bl_arity; ++i)
1252 end_preds[i] = get_irn_n(main_end_bl, i);
1253 for (i = 0; i < n_exc; ++i)
1254 end_preds[main_end_bl_arity + i] = cf_pred[i];
1255 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1256 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1257 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1263 /* -- Turn CSE back on. -- */
1264 set_optimize(rem_opt);
1269 /********************************************************************/
1270 /* Apply inlineing to small methods. */
1271 /********************************************************************/
1273 /** Represents a possible inlinable call in a graph. */
1274 typedef struct _call_entry call_entry;
1275 struct _call_entry {
1276 ir_node *call; /**< the Call */
1277 ir_graph *callee; /**< the callee called here */
1278 call_entry *next; /**< for linking the next one */
1282 * environment for inlining small irgs
1284 typedef struct _inline_env_t {
1285 struct obstack obst; /**< an obstack where call_entries are allocated on. */
1286 call_entry *head; /**< the head of the call entry list */
1287 call_entry *tail; /**< the tail of the call entry list */
1291 * Returns the irg called from a Call node. If the irg is not
1292 * known, NULL is returned.
1294 static ir_graph *get_call_called_irg(ir_node *call) {
1296 ir_graph *called_irg = NULL;
1298 addr = get_Call_ptr(call);
1299 if (is_SymConst(addr) && get_SymConst_kind(addr) == symconst_addr_ent) {
1300 called_irg = get_entity_irg(get_SymConst_entity(addr));
1307 * Walker: Collect all calls to known graphs inside a graph.
1309 static void collect_calls(ir_node *call, void *env) {
1310 if (is_Call(call)) {
1311 ir_graph *called_irg = get_call_called_irg(call);
1313 /* The Call node calls a locally defined method. Remember to inline. */
1314 inline_env_t *ienv = env;
1315 call_entry *entry = obstack_alloc(&ienv->obst, sizeof(*entry));
1317 entry->callee = called_irg;
1320 if (ienv->tail == NULL)
1323 ienv->tail->next = entry;
1330 * Inlines all small methods at call sites where the called address comes
1331 * from a Const node that references the entity representing the called
1333 * The size argument is a rough measure for the code size of the method:
1334 * Methods where the obstack containing the firm graph is smaller than
1337 void inline_small_irgs(ir_graph *irg, int size) {
1338 ir_graph *rem = current_ir_graph;
1341 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1343 if (!(get_opt_optimize() && get_opt_inline())) return;
1345 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1347 current_ir_graph = irg;
1348 /* Handle graph state */
1349 assert(get_irg_phase_state(irg) != phase_building);
1350 free_callee_info(irg);
1352 /* Find Call nodes to inline.
1353 (We can not inline during a walk of the graph, as inlineing the same
1354 method several times changes the visited flag of the walked graph:
1355 after the first inlineing visited of the callee equals visited of
1356 the caller. With the next inlineing both are increased.) */
1357 obstack_init(&env.obst);
1358 env.head = env.tail = NULL;
1359 irg_walk_graph(irg, NULL, collect_calls, &env);
1361 if (env.head != NULL) {
1362 /* There are calls to inline */
1363 collect_phiprojs(irg);
1364 for (entry = env.head; entry != NULL; entry = entry->next) {
1365 ir_graph *callee = entry->callee;
1366 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1367 (get_irg_inline_property(callee) >= irg_inline_forced)) {
1368 inline_method(entry->call, callee);
1372 obstack_free(&env.obst, NULL);
1373 current_ir_graph = rem;
1377 * Environment for inlining irgs.
1380 int n_nodes; /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
1381 int n_nodes_orig; /**< for statistics */
1382 call_entry *call_head; /**< The head of the list of all call nodes in this graph. */
1383 call_entry *call_tail; /**< The tail of the list of all call nodes in this graph .*/
1384 int n_call_nodes; /**< Number of Call nodes in the graph. */
1385 int n_call_nodes_orig; /**< for statistics */
1386 int n_callers; /**< Number of known graphs that call this graphs. */
1387 int n_callers_orig; /**< for statistics */
1388 int got_inline; /**< Set, if at leat one call inside this graph was inlined. */
1392 * Allocate a new environment for inlining.
1394 static inline_irg_env *alloc_inline_irg_env(struct obstack *obst) {
1395 inline_irg_env *env = obstack_alloc(obst, sizeof(*env));
1396 env->n_nodes = -2; /* do not count count Start, End */
1397 env->n_nodes_orig = -2; /* do not count Start, End */
1398 env->call_head = NULL;
1399 env->call_tail = NULL;
1400 env->n_call_nodes = 0;
1401 env->n_call_nodes_orig = 0;
1403 env->n_callers_orig = 0;
1404 env->got_inline = 0;
1408 typedef struct walker_env {
1409 struct obstack *obst; /**< the obstack for allocations. */
1410 inline_irg_env *x; /**< the inline environment */
1411 int ignore_runtime; /**< the ignore runtime flag */
1415 * post-walker: collect all calls in the inline-environment
1416 * of a graph and sum some statistics.
1418 static void collect_calls2(ir_node *call, void *ctx) {
1420 inline_irg_env *x = env->x;
1421 ir_op *op = get_irn_op(call);
1425 /* count meaningful nodes in irg */
1426 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1431 if (op != op_Call) return;
1433 /* check, if it's a runtime call */
1434 if (env->ignore_runtime) {
1435 ir_node *symc = get_Call_ptr(call);
1437 if (is_SymConst(symc) && get_SymConst_kind(symc) == symconst_addr_ent) {
1438 ir_entity *ent = get_SymConst_entity(symc);
1440 if (get_entity_additional_properties(ent) & mtp_property_runtime)
1445 /* collect all call nodes */
1447 ++x->n_call_nodes_orig;
1449 callee = get_call_called_irg(call);
1451 inline_irg_env *callee_env = get_irg_link(callee);
1452 /* count all static callers */
1453 ++callee_env->n_callers;
1454 ++callee_env->n_callers_orig;
1456 /* link it in the list of possible inlinable entries */
1457 entry = obstack_alloc(env->obst, sizeof(*entry));
1459 entry->callee = callee;
1461 if (x->call_tail == NULL)
1462 x->call_head = entry;
1464 x->call_tail->next = entry;
1465 x->call_tail = entry;
1470 * Returns TRUE if the number of callers in 0 in the irg's environment,
1471 * hence this irg is a leave.
1473 INLINE static int is_leave(ir_graph *irg) {
1474 inline_irg_env *env = get_irg_link(irg);
1475 return env->n_call_nodes == 0;
1479 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1481 INLINE static int is_smaller(ir_graph *callee, int size) {
1482 inline_irg_env *env = get_irg_link(callee);
1483 return env->n_nodes < size;
1487 * Append the nodes of the list src to the nodes of the list in environment dst.
1489 static void append_call_list(struct obstack *obst, inline_irg_env *dst, call_entry *src) {
1490 call_entry *entry, *nentry;
1492 /* Note that the src list points to Call nodes in the inlined graph, but
1493 we need Call nodes in our graph. Luckily the inliner leaves this information
1494 in the link field. */
1495 for (entry = src; entry != NULL; entry = entry->next) {
1496 nentry = obstack_alloc(obst, sizeof(*nentry));
1497 nentry->call = get_irn_link(entry->call);
1498 nentry->callee = entry->callee;
1499 nentry->next = NULL;
1500 dst->call_tail->next = nentry;
1501 dst->call_tail = nentry;
1506 * Inlines small leave methods at call sites where the called address comes
1507 * from a Const node that references the entity representing the called
1509 * The size argument is a rough measure for the code size of the method:
1510 * Methods where the obstack containing the firm graph is smaller than
1513 void inline_leave_functions(int maxsize, int leavesize, int size, int ignore_runtime) {
1514 inline_irg_env *env;
1520 call_entry *entry, *tail;
1521 const call_entry *centry;
1522 struct obstack obst;
1523 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1525 if (!(get_opt_optimize() && get_opt_inline())) return;
1527 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1528 rem = current_ir_graph;
1529 obstack_init(&obst);
1531 /* extend all irgs by a temporary data structure for inlining. */
1532 n_irgs = get_irp_n_irgs();
1533 for (i = 0; i < n_irgs; ++i)
1534 set_irg_link(get_irp_irg(i), alloc_inline_irg_env(&obst));
1536 /* Precompute information in temporary data structure. */
1538 wenv.ignore_runtime = ignore_runtime;
1539 for (i = 0; i < n_irgs; ++i) {
1540 ir_graph *irg = get_irp_irg(i);
1542 assert(get_irg_phase_state(irg) != phase_building);
1543 free_callee_info(irg);
1545 wenv.x = get_irg_link(irg);
1546 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1549 /* -- and now inline. -- */
1551 /* Inline leaves recursively -- we might construct new leaves. */
1555 for (i = 0; i < n_irgs; ++i) {
1557 int phiproj_computed = 0;
1559 current_ir_graph = get_irp_irg(i);
1560 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1563 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1566 if (env->n_nodes > maxsize) break;
1569 callee = entry->callee;
1571 if (is_leave(callee) && is_smaller(callee, leavesize)) {
1572 if (!phiproj_computed) {
1573 phiproj_computed = 1;
1574 collect_phiprojs(current_ir_graph);
1576 did_inline = inline_method(call, callee);
1579 /* Do some statistics */
1580 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1582 env->got_inline = 1;
1583 --env->n_call_nodes;
1584 env->n_nodes += callee_env->n_nodes;
1585 --callee_env->n_callers;
1587 /* remove this call from the list */
1589 tail->next = entry->next;
1591 env->call_head = entry->next;
1597 env->call_tail = tail;
1599 } while (did_inline);
1601 /* inline other small functions. */
1602 for (i = 0; i < n_irgs; ++i) {
1604 int phiproj_computed = 0;
1606 current_ir_graph = get_irp_irg(i);
1607 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1609 /* note that the list of possible calls is updated during the process */
1611 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1615 callee = entry->callee;
1617 if (((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1618 (get_irg_inline_property(callee) >= irg_inline_forced))) {
1619 if (!phiproj_computed) {
1620 phiproj_computed = 1;
1621 collect_phiprojs(current_ir_graph);
1623 if (inline_method(call, callee)) {
1624 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1626 /* callee was inline. Append it's call list. */
1627 env->got_inline = 1;
1628 --env->n_call_nodes;
1629 append_call_list(&obst, env, callee_env->call_head);
1630 env->n_call_nodes += callee_env->n_call_nodes;
1631 env->n_nodes += callee_env->n_nodes;
1632 --callee_env->n_callers;
1634 /* after we have inlined callee, all called methods inside callee
1635 are now called once more */
1636 for (centry = callee_env->call_head; centry != NULL; centry = centry->next) {
1637 inline_irg_env *penv = get_irg_link(centry->callee);
1641 /* remove this call from the list */
1643 tail->next = entry->next;
1645 env->call_head = entry->next;
1651 env->call_tail = tail;
1654 for (i = 0; i < n_irgs; ++i) {
1655 irg = get_irp_irg(i);
1656 env = (inline_irg_env *)get_irg_link(irg);
1658 if (env->got_inline) {
1659 /* this irg got calls inlined */
1660 set_irg_outs_inconsistent(irg);
1661 set_irg_doms_inconsistent(irg);
1663 optimize_graph_df(irg);
1666 if (env->got_inline || (env->n_callers_orig != env->n_callers))
1667 DB((dbg, SET_LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1668 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1669 env->n_callers_orig, env->n_callers,
1670 get_entity_name(get_irg_entity(irg))));
1673 obstack_free(&obst, NULL);
1674 current_ir_graph = rem;
1677 /*******************************************************************/
1678 /* Code Placement. Pins all floating nodes to a block where they */
1679 /* will be executed only if needed. */
1680 /*******************************************************************/
1683 * Returns non-zero, is a block is not reachable from Start.
1685 * @param block the block to test
1688 is_Block_unreachable(ir_node *block) {
1689 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1693 * Find the earliest correct block for node n. --- Place n into the
1694 * same Block as its dominance-deepest Input.
1696 * We have to avoid calls to get_nodes_block() here
1697 * because the graph is floating.
1699 * move_out_of_loops() expects that place_floats_early() have placed
1700 * all "living" nodes into a living block. That's why we must
1701 * move nodes in dead block with "live" successors into a valid
1703 * We move them just into the same block as it's successor (or
1704 * in case of a Phi into the effective use block). For Phi successors,
1705 * this may still be a dead block, but then there is no real use, as
1706 * the control flow will be dead later.
1708 * @param n the node to be placed
1709 * @param worklist a worklist, predecessors of non-floating nodes are placed here
1712 place_floats_early(ir_node *n, waitq *worklist) {
1715 /* we must not run into an infinite loop */
1716 assert(irn_not_visited(n));
1717 mark_irn_visited(n);
1719 /* Place floating nodes. */
1720 if (get_irn_pinned(n) == op_pin_state_floats) {
1721 ir_node *curr_block = get_irn_n(n, -1);
1722 int in_dead_block = is_Block_unreachable(curr_block);
1724 ir_node *b = NULL; /* The block to place this node in */
1726 assert(is_no_Block(n));
1728 if (is_irn_start_block_placed(n)) {
1729 /* These nodes will not be placed by the loop below. */
1730 b = get_irg_start_block(current_ir_graph);
1734 /* find the block for this node. */
1735 irn_arity = get_irn_arity(n);
1736 for (i = 0; i < irn_arity; i++) {
1737 ir_node *pred = get_irn_n(n, i);
1738 ir_node *pred_block;
1740 if ((irn_not_visited(pred))
1741 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1744 * If the current node is NOT in a dead block, but one of its
1745 * predecessors is, we must move the predecessor to a live block.
1746 * Such thing can happen, if global CSE chose a node from a dead block.
1747 * We move it simply to our block.
1748 * Note that neither Phi nor End nodes are floating, so we don't
1749 * need to handle them here.
1751 if (! in_dead_block) {
1752 if (get_irn_pinned(pred) == op_pin_state_floats &&
1753 is_Block_unreachable(get_irn_n(pred, -1)))
1754 set_nodes_block(pred, curr_block);
1756 place_floats_early(pred, worklist);
1760 * A node in the Bad block must stay in the bad block,
1761 * so don't compute a new block for it.
1766 /* Because all loops contain at least one op_pin_state_pinned node, now all
1767 our inputs are either op_pin_state_pinned or place_early() has already
1768 been finished on them. We do not have any unfinished inputs! */
1769 pred_block = get_irn_n(pred, -1);
1770 if ((!is_Block_dead(pred_block)) &&
1771 (get_Block_dom_depth(pred_block) > depth)) {
1773 depth = get_Block_dom_depth(pred_block);
1775 /* Avoid that the node is placed in the Start block */
1776 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)
1777 && get_irg_phase_state(current_ir_graph) != phase_backend) {
1778 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1779 assert(b != get_irg_start_block(current_ir_graph));
1784 set_nodes_block(n, b);
1788 * Add predecessors of non floating nodes and non-floating predecessors
1789 * of floating nodes to worklist and fix their blocks if the are in dead block.
1791 irn_arity = get_irn_arity(n);
1793 if (get_irn_op(n) == op_End) {
1795 * Simplest case: End node. Predecessors are keep-alives,
1796 * no need to move out of dead block.
1798 for (i = -1; i < irn_arity; ++i) {
1799 ir_node *pred = get_irn_n(n, i);
1800 if (irn_not_visited(pred))
1801 waitq_put(worklist, pred);
1803 } else if (is_Block(n)) {
1805 * Blocks: Predecessors are control flow, no need to move
1806 * them out of dead block.
1808 for (i = irn_arity - 1; i >= 0; --i) {
1809 ir_node *pred = get_irn_n(n, i);
1810 if (irn_not_visited(pred))
1811 waitq_put(worklist, pred);
1813 } else if (is_Phi(n)) {
1815 ir_node *curr_block = get_irn_n(n, -1);
1816 int in_dead_block = is_Block_unreachable(curr_block);
1819 * Phi nodes: move nodes from dead blocks into the effective use
1820 * of the Phi-input if the Phi is not in a bad block.
1822 pred = get_irn_n(n, -1);
1823 if (irn_not_visited(pred))
1824 waitq_put(worklist, pred);
1826 for (i = irn_arity - 1; i >= 0; --i) {
1827 ir_node *pred = get_irn_n(n, i);
1829 if (irn_not_visited(pred)) {
1830 if (! in_dead_block &&
1831 get_irn_pinned(pred) == op_pin_state_floats &&
1832 is_Block_unreachable(get_irn_n(pred, -1))) {
1833 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1835 waitq_put(worklist, pred);
1840 ir_node *curr_block = get_irn_n(n, -1);
1841 int in_dead_block = is_Block_unreachable(curr_block);
1844 * All other nodes: move nodes from dead blocks into the same block.
1846 pred = get_irn_n(n, -1);
1847 if (irn_not_visited(pred))
1848 waitq_put(worklist, pred);
1850 for (i = irn_arity - 1; i >= 0; --i) {
1851 ir_node *pred = get_irn_n(n, i);
1853 if (irn_not_visited(pred)) {
1854 if (! in_dead_block &&
1855 get_irn_pinned(pred) == op_pin_state_floats &&
1856 is_Block_unreachable(get_irn_n(pred, -1))) {
1857 set_nodes_block(pred, curr_block);
1859 waitq_put(worklist, pred);
1866 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1867 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1868 * places all floating nodes reachable from its argument through floating
1869 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1871 * @param worklist a worklist, used for the algorithm, empty on in/output
1873 static void place_early(waitq *worklist) {
1875 inc_irg_visited(current_ir_graph);
1877 /* this inits the worklist */
1878 place_floats_early(get_irg_end(current_ir_graph), worklist);
1880 /* Work the content of the worklist. */
1881 while (!waitq_empty(worklist)) {
1882 ir_node *n = waitq_get(worklist);
1883 if (irn_not_visited(n))
1884 place_floats_early(n, worklist);
1887 set_irg_outs_inconsistent(current_ir_graph);
1888 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
1892 * Compute the deepest common ancestor of block and dca.
1894 static ir_node *calc_dca(ir_node *dca, ir_node *block) {
1897 /* we do not want to place nodes in dead blocks */
1898 if (is_Block_dead(block))
1901 /* We found a first legal placement. */
1902 if (!dca) return block;
1904 /* Find a placement that is dominates both, dca and block. */
1905 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1906 block = get_Block_idom(block);
1908 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1909 dca = get_Block_idom(dca);
1912 while (block != dca) {
1913 block = get_Block_idom(block); dca = get_Block_idom(dca);
1919 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1920 * I.e., DCA is the block where we might place PRODUCER.
1921 * A data flow edge points from producer to consumer.
1924 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer) {
1925 ir_node *block = NULL;
1927 /* Compute the latest block into which we can place a node so that it is
1929 if (get_irn_op(consumer) == op_Phi) {
1930 /* our consumer is a Phi-node, the effective use is in all those
1931 blocks through which the Phi-node reaches producer */
1933 ir_node *phi_block = get_nodes_block(consumer);
1934 irn_arity = get_irn_arity(consumer);
1936 for (i = 0; i < irn_arity; i++) {
1937 if (get_irn_n(consumer, i) == producer) {
1938 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1940 if (! is_Block_unreachable(new_block))
1941 block = calc_dca(block, new_block);
1946 block = get_irn_n(producer, -1);
1948 assert(is_no_Block(consumer));
1949 block = get_nodes_block(consumer);
1952 /* Compute the deepest common ancestor of block and dca. */
1953 return calc_dca(dca, block);
1956 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1958 static INLINE int get_irn_loop_depth(ir_node *n) {
1959 return get_loop_depth(get_irn_loop(n));
1963 * Move n to a block with less loop depth than it's current block. The
1964 * new block must be dominated by early.
1966 * @param n the node that should be moved
1967 * @param early the earliest block we can n move to
1969 static void move_out_of_loops(ir_node *n, ir_node *early) {
1970 ir_node *best, *dca;
1974 /* Find the region deepest in the dominator tree dominating
1975 dca with the least loop nesting depth, but still dominated
1976 by our early placement. */
1977 dca = get_nodes_block(n);
1980 while (dca != early) {
1981 dca = get_Block_idom(dca);
1982 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1983 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1987 if (best != get_nodes_block(n)) {
1989 printf("Moving out of loop: "); DDMN(n);
1990 printf(" Outermost block: "); DDMN(early);
1991 printf(" Best block: "); DDMN(best);
1992 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1994 set_nodes_block(n, best);
1999 * Find the latest legal block for N and place N into the
2000 * `optimal' Block between the latest and earliest legal block.
2001 * The `optimal' block is the dominance-deepest block of those
2002 * with the least loop-nesting-depth. This places N out of as many
2003 * loops as possible and then makes it as control dependent as
2006 * @param n the node to be placed
2007 * @param worklist a worklist, all successors of non-floating nodes are
2010 static void place_floats_late(ir_node *n, pdeq *worklist) {
2014 assert(irn_not_visited(n)); /* no multiple placement */
2016 mark_irn_visited(n);
2018 /* no need to place block nodes, control nodes are already placed. */
2019 if ((get_irn_op(n) != op_Block) &&
2021 (get_irn_mode(n) != mode_X)) {
2022 /* Remember the early_blk placement of this block to move it
2023 out of loop no further than the early_blk placement. */
2024 early_blk = get_irn_n(n, -1);
2027 * BEWARE: Here we also get code, that is live, but
2028 * was in a dead block. If the node is life, but because
2029 * of CSE in a dead block, we still might need it.
2032 /* Assure that our users are all placed, except the Phi-nodes.
2033 --- Each data flow cycle contains at least one Phi-node. We
2034 have to break the `user has to be placed before the
2035 producer' dependence cycle and the Phi-nodes are the
2036 place to do so, because we need to base our placement on the
2037 final region of our users, which is OK with Phi-nodes, as they
2038 are op_pin_state_pinned, and they never have to be placed after a
2039 producer of one of their inputs in the same block anyway. */
2040 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2041 ir_node *succ = get_irn_out(n, i);
2042 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
2043 place_floats_late(succ, worklist);
2046 if (! is_Block_dead(early_blk)) {
2047 /* do only move things that where not dead */
2048 ir_op *op = get_irn_op(n);
2050 /* We have to determine the final block of this node... except for
2051 constants and Projs */
2052 if ((get_irn_pinned(n) == op_pin_state_floats) &&
2054 (op != op_SymConst) &&
2057 ir_node *dca = NULL; /* deepest common ancestor in the
2058 dominator tree of all nodes'
2059 blocks depending on us; our final
2060 placement has to dominate DCA. */
2061 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2062 ir_node *succ = get_irn_out(n, i);
2065 if (get_irn_op(succ) == op_End) {
2067 * This consumer is the End node, a keep alive edge.
2068 * This is not a real consumer, so we ignore it
2073 /* ignore if succ is in dead code */
2074 succ_blk = get_irn_n(succ, -1);
2075 if (is_Block_unreachable(succ_blk))
2077 dca = consumer_dom_dca(dca, succ, n);
2080 set_nodes_block(n, dca);
2081 move_out_of_loops(n, early_blk);
2087 /* Add successors of all non-floating nodes on list. (Those of floating
2088 nodes are placed already and therefore are marked.) */
2089 for (i = 0; i < get_irn_n_outs(n); i++) {
2090 ir_node *succ = get_irn_out(n, i);
2091 if (irn_not_visited(get_irn_out(n, i))) {
2092 pdeq_putr(worklist, succ);
2098 * Place floating nodes on the given worklist as late as possible using
2099 * the dominance tree.
2101 * @param worklist the worklist containing the nodes to place
2103 static void place_late(waitq *worklist) {
2105 inc_irg_visited(current_ir_graph);
2107 /* This fills the worklist initially. */
2108 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
2110 /* And now empty the worklist again... */
2111 while (!waitq_empty(worklist)) {
2112 ir_node *n = waitq_get(worklist);
2113 if (irn_not_visited(n))
2114 place_floats_late(n, worklist);
2118 /* Code Placement. */
2119 void place_code(ir_graph *irg) {
2121 ir_graph *rem = current_ir_graph;
2123 current_ir_graph = irg;
2125 /* Handle graph state */
2126 assert(get_irg_phase_state(irg) != phase_building);
2129 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
2130 free_loop_information(irg);
2131 construct_backedges(irg);
2134 /* Place all floating nodes as early as possible. This guarantees
2135 a legal code placement. */
2136 worklist = new_waitq();
2137 place_early(worklist);
2139 /* place_early() invalidates the outs, place_late needs them. */
2140 compute_irg_outs(irg);
2142 /* Now move the nodes down in the dominator tree. This reduces the
2143 unnecessary executions of the node. */
2144 place_late(worklist);
2146 set_irg_outs_inconsistent(current_ir_graph);
2147 set_irg_loopinfo_inconsistent(current_ir_graph);
2148 del_waitq(worklist);
2149 current_ir_graph = rem;
2153 * Called by walker of remove_critical_cf_edges().
2155 * Place an empty block to an edge between a blocks of multiple
2156 * predecessors and a block of multiple successors.
2159 * @param env Environment of walker. The changed field.
2161 static void walk_critical_cf_edges(ir_node *n, void *env) {
2163 ir_node *pre, *block, *jmp;
2165 ir_graph *irg = get_irn_irg(n);
2167 /* Block has multiple predecessors */
2168 arity = get_irn_arity(n);
2170 if (n == get_irg_end_block(irg))
2171 return; /* No use to add a block here. */
2173 for (i = 0; i < arity; ++i) {
2176 pre = get_irn_n(n, i);
2177 cfop = get_irn_op(skip_Proj(pre));
2178 /* Predecessor has multiple successors. Insert new control flow edge but
2179 ignore exception edges. */
2180 if (! is_op_fragile(cfop) && is_op_forking(cfop)) {
2181 /* set predecessor of new block */
2182 block = new_r_Block(irg, 1, &pre);
2183 /* insert new jmp node to new block */
2184 jmp = new_r_Jmp(irg, block);
2185 /* set successor of new block */
2186 set_irn_n(n, i, jmp);
2188 } /* predecessor has multiple successors */
2189 } /* for all predecessors */
2190 } /* n is a multi-entry block */
2193 void remove_critical_cf_edges(ir_graph *irg) {
2196 irg_block_walk_graph(irg, NULL, walk_critical_cf_edges, &changed);
2198 /* control flow changed */
2199 set_irg_outs_inconsistent(irg);
2200 set_irg_extblk_inconsistent(irg);
2201 set_irg_doms_inconsistent(irg);
2202 set_irg_loopinfo_inconsistent(irg);