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"
37 #include "iroptimize.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);
74 exchange (n, optimized);
79 * Do local optimizations for a node.
81 * @param n the IR-node where to start. Typically the End node
84 * @note current_ir_graph must be set
86 static INLINE void do_local_optimize(ir_node *n) {
87 /* Handle graph state */
88 assert(get_irg_phase_state(current_ir_graph) != phase_building);
90 if (get_opt_global_cse())
91 set_irg_pinned(current_ir_graph, op_pin_state_floats);
92 set_irg_outs_inconsistent(current_ir_graph);
93 set_irg_doms_inconsistent(current_ir_graph);
94 set_irg_loopinfo_inconsistent(current_ir_graph);
96 /* Clean the value_table in irg for the CSE. */
97 del_identities(current_ir_graph->value_table);
98 current_ir_graph->value_table = new_identities();
100 /* walk over the graph */
101 irg_walk(n, firm_clear_link, optimize_in_place_wrapper, NULL);
104 /* Applies local optimizations (see iropt.h) to all nodes reachable from node n */
105 void local_optimize_node(ir_node *n) {
106 ir_graph *rem = current_ir_graph;
107 current_ir_graph = get_irn_irg(n);
109 do_local_optimize(n);
111 current_ir_graph = rem;
115 * Block-Walker: uses dominance depth to mark dead blocks.
117 static void kill_dead_blocks(ir_node *block, void *env) {
120 if (get_Block_dom_depth(block) < 0) {
122 * Note that the new dominance code correctly handles
123 * the End block, i.e. it is always reachable from Start
125 set_Block_dead(block);
129 /* Applies local optimizations (see iropt.h) to all nodes reachable from node n. */
130 void local_optimize_graph(ir_graph *irg) {
131 ir_graph *rem = current_ir_graph;
132 current_ir_graph = irg;
134 if (get_irg_dom_state(irg) == dom_consistent)
135 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
137 do_local_optimize(get_irg_end(irg));
139 current_ir_graph = rem;
143 * Enqueue all users of a node to a wait queue.
144 * Handles mode_T nodes.
146 static void enqueue_users(ir_node *n, pdeq *waitq) {
147 const ir_edge_t *edge;
149 foreach_out_edge(n, edge) {
150 ir_node *succ = get_edge_src_irn(edge);
152 if (get_irn_link(succ) != waitq) {
153 pdeq_putr(waitq, succ);
154 set_irn_link(succ, waitq);
156 if (get_irn_mode(succ) == mode_T) {
157 /* A mode_T node has Proj's. Because most optimizations
158 run on the Proj's we have to enqueue them also. */
159 enqueue_users(succ, waitq);
165 * Data flow optimization walker.
166 * Optimizes all nodes and enqueue it's users
169 static void opt_walker(ir_node *n, void *env) {
173 optimized = optimize_in_place_2(n);
174 set_irn_link(optimized, NULL);
176 if (optimized != n) {
177 enqueue_users(n, waitq);
178 exchange(n, optimized);
182 /* Applies local optimizations to all nodes in the graph until fixpoint. */
183 void optimize_graph_df(ir_graph *irg) {
184 pdeq *waitq = new_pdeq();
185 int state = edges_activated(irg);
186 ir_graph *rem = current_ir_graph;
190 current_ir_graph = irg;
195 if (get_opt_global_cse())
196 set_irg_pinned(current_ir_graph, op_pin_state_floats);
198 /* Clean the value_table in irg for the CSE. */
199 del_identities(irg->value_table);
200 irg->value_table = new_identities();
202 if (get_irg_dom_state(irg) == dom_consistent)
203 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
205 /* invalidate info */
206 set_irg_outs_inconsistent(irg);
207 set_irg_doms_inconsistent(irg);
208 set_irg_loopinfo_inconsistent(irg);
210 set_using_irn_link(irg);
212 /* walk over the graph, but don't touch keep-alives */
213 irg_walk(get_irg_end_block(irg), NULL, opt_walker, waitq);
215 end = get_irg_end(irg);
217 /* optimize keep-alives by removing superfluous ones */
218 for (i = get_End_n_keepalives(end) - 1; i >= 0; --i) {
219 ir_node *ka = get_End_keepalive(end, i);
221 if (irn_visited(ka) && !is_irn_keep(ka)) {
222 /* this node can be regularly visited, no need to keep it */
223 set_End_keepalive(end, i, get_irg_bad(irg));
226 /* now walk again and visit all not yet visited nodes */
227 set_irg_visited(current_ir_graph, get_irg_visited(irg) - 1);
228 irg_walk(get_irg_end(irg), NULL, opt_walker, waitq);
230 /* finish the wait queue */
231 while (! pdeq_empty(waitq)) {
232 ir_node *n = pdeq_getl(waitq);
234 opt_walker(n, waitq);
239 clear_using_irn_link(irg);
242 edges_deactivate(irg);
244 current_ir_graph = rem;
248 /*------------------------------------------------------------------*/
249 /* Routines for dead node elimination / copying garbage collection */
250 /* of the obstack. */
251 /*------------------------------------------------------------------*/
254 * Remember the new node in the old node by using a field all nodes have.
256 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
259 * Get this new node, before the old node is forgotten.
261 #define get_new_node(oldn) get_irn_link(oldn)
264 * Check if a new node was set.
266 #define has_new_node(n) (get_new_node(n) != NULL)
269 * We use the block_visited flag to mark that we have computed the
270 * number of useful predecessors for this block.
271 * Further we encode the new arity in this flag in the old blocks.
272 * Remembering the arity is useful, as it saves a lot of pointer
273 * accesses. This function is called for all Phi and Block nodes
277 compute_new_arity(ir_node *b) {
278 int i, res, irn_arity;
281 irg_v = get_irg_block_visited(current_ir_graph);
282 block_v = get_Block_block_visited(b);
283 if (block_v >= irg_v) {
284 /* we computed the number of preds for this block and saved it in the
286 return block_v - irg_v;
288 /* compute the number of good predecessors */
289 res = irn_arity = get_irn_arity(b);
290 for (i = 0; i < irn_arity; i++)
291 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
292 /* save it in the flag. */
293 set_Block_block_visited(b, irg_v + res);
299 * Copies the node to the new obstack. The Ins of the new node point to
300 * the predecessors on the old obstack. For block/phi nodes not all
301 * predecessors might be copied. n->link points to the new node.
302 * For Phi and Block nodes the function allocates in-arrays with an arity
303 * only for useful predecessors. The arity is determined by counting
304 * the non-bad predecessors of the block.
306 * @param n The node to be copied
307 * @param env if non-NULL, the node number attribute will be copied to the new node
309 * Note: Also used for loop unrolling.
311 static void copy_node(ir_node *n, void *env) {
314 ir_op *op = get_irn_op(n);
316 /* The end node looses it's flexible in array. This doesn't matter,
317 as dead node elimination builds End by hand, inlineing doesn't use
319 /* assert(op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
322 /* node copied already */
324 } else if (op == op_Block) {
326 new_arity = compute_new_arity(n);
327 n->attr.block.graph_arr = NULL;
329 block = get_nodes_block(n);
331 new_arity = compute_new_arity(block);
333 new_arity = get_irn_arity(n);
336 nn = new_ir_node(get_irn_dbg_info(n),
343 /* Copy the attributes. These might point to additional data. If this
344 was allocated on the old obstack the pointers now are dangling. This
345 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
346 copy_node_attr(n, nn);
350 int copy_node_nr = env != NULL;
352 /* for easier debugging, we want to copy the node numbers too */
353 nn->node_nr = n->node_nr;
359 hook_dead_node_elim_subst(current_ir_graph, n, nn);
363 * Copies new predecessors of old node to new node remembered in link.
364 * Spare the Bad predecessors of Phi and Block nodes.
367 copy_preds(ir_node *n, void *env) {
372 nn = get_new_node(n);
375 /* Don't copy Bad nodes. */
377 irn_arity = get_irn_arity(n);
378 for (i = 0; i < irn_arity; i++) {
379 if (! is_Bad(get_irn_n(n, i))) {
380 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
381 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
385 /* repair the block visited flag from above misuse. Repair it in both
386 graphs so that the old one can still be used. */
387 set_Block_block_visited(nn, 0);
388 set_Block_block_visited(n, 0);
389 /* Local optimization could not merge two subsequent blocks if
390 in array contained Bads. Now it's possible.
391 We don't call optimize_in_place as it requires
392 that the fields in ir_graph are set properly. */
393 if ((get_opt_control_flow_straightening()) &&
394 (get_Block_n_cfgpreds(nn) == 1) &&
395 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
396 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
398 /* Jmp jumps into the block it is in -- deal self cycle. */
399 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
400 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
405 } else if (get_irn_op(n) == op_Phi) {
406 /* Don't copy node if corresponding predecessor in block is Bad.
407 The Block itself should not be Bad. */
408 block = get_nodes_block(n);
409 set_irn_n(nn, -1, get_new_node(block));
411 irn_arity = get_irn_arity(n);
412 for (i = 0; i < irn_arity; i++) {
413 if (! is_Bad(get_irn_n(block, i))) {
414 set_irn_n(nn, j, get_new_node(get_irn_n(n, i)));
415 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
419 /* If the pre walker reached this Phi after the post walker visited the
420 block block_visited is > 0. */
421 set_Block_block_visited(get_nodes_block(n), 0);
422 /* Compacting the Phi's ins might generate Phis with only one
424 if (get_irn_arity(nn) == 1)
425 exchange(nn, get_irn_n(nn, 0));
427 irn_arity = get_irn_arity(n);
428 for (i = -1; i < irn_arity; i++)
429 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
431 /* Now the new node is complete. We can add it to the hash table for CSE.
432 @@@ inlining aborts if we identify End. Why? */
433 if (get_irn_op(nn) != op_End)
434 add_identities(current_ir_graph->value_table, nn);
438 * Copies the graph recursively, compacts the keep-alives of the end node.
440 * @param irg the graph to be copied
441 * @param copy_node_nr If non-zero, the node number will be copied
443 static void copy_graph(ir_graph *irg, int copy_node_nr) {
444 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
445 ir_node *ka; /* keep alive */
449 /* Some nodes must be copied by hand, sigh */
450 vfl = get_irg_visited(irg);
451 set_irg_visited(irg, vfl + 1);
453 oe = get_irg_end(irg);
454 mark_irn_visited(oe);
455 /* copy the end node by hand, allocate dynamic in array! */
456 ne = new_ir_node(get_irn_dbg_info(oe),
463 /* Copy the attributes. Well, there might be some in the future... */
464 copy_node_attr(oe, ne);
465 set_new_node(oe, ne);
467 /* copy the Bad node */
468 ob = get_irg_bad(irg);
469 mark_irn_visited(ob);
470 nb = new_ir_node(get_irn_dbg_info(ob),
477 copy_node_attr(ob, nb);
478 set_new_node(ob, nb);
480 /* copy the NoMem node */
481 om = get_irg_no_mem(irg);
482 mark_irn_visited(om);
483 nm = new_ir_node(get_irn_dbg_info(om),
490 copy_node_attr(om, nm);
491 set_new_node(om, nm);
493 /* copy the live nodes */
494 set_irg_visited(irg, vfl);
495 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
497 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
499 /* visit the anchors as well */
500 for (i = anchor_max - 1; i >= 0; --i) {
501 ir_node *n = irg->anchors[i];
503 if (n && (get_irn_visited(n) <= vfl)) {
504 set_irg_visited(irg, vfl);
505 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
509 /* copy_preds for the end node ... */
510 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
512 /*- ... and now the keep alives. -*/
513 /* First pick the not marked block nodes and walk them. We must pick these
514 first as else we will oversee blocks reachable from Phis. */
515 irn_arity = get_End_n_keepalives(oe);
516 for (i = 0; i < irn_arity; i++) {
517 ka = get_End_keepalive(oe, i);
519 if (get_irn_visited(ka) <= vfl) {
520 /* We must keep the block alive and copy everything reachable */
521 set_irg_visited(irg, vfl);
522 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
524 add_End_keepalive(ne, get_new_node(ka));
528 /* Now pick other nodes. Here we will keep all! */
529 irn_arity = get_End_n_keepalives(oe);
530 for (i = 0; i < irn_arity; i++) {
531 ka = get_End_keepalive(oe, i);
533 if (get_irn_visited(ka) <= vfl) {
534 /* We didn't copy the node yet. */
535 set_irg_visited(irg, vfl);
536 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
538 add_End_keepalive(ne, get_new_node(ka));
542 /* start block sometimes only reached after keep alives */
543 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
544 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
548 * Copies the graph reachable from current_ir_graph->end to the obstack
549 * in current_ir_graph and fixes the environment.
550 * Then fixes the fields in current_ir_graph containing nodes of the
553 * @param copy_node_nr If non-zero, the node number will be copied
556 copy_graph_env(int copy_node_nr) {
557 ir_graph *irg = current_ir_graph;
558 ir_node *old_end, *n;
561 /* remove end_except and end_reg nodes */
562 old_end = get_irg_end(irg);
563 set_irg_end_except (irg, old_end);
564 set_irg_end_reg (irg, old_end);
566 /* Not all nodes remembered in irg might be reachable
567 from the end node. Assure their link is set to NULL, so that
568 we can test whether new nodes have been computed. */
569 for (i = anchor_max - 1; i >= 0; --i) {
571 set_new_node(irg->anchors[i], NULL);
573 /* we use the block walk flag for removing Bads from Blocks ins. */
574 inc_irg_block_visited(irg);
577 copy_graph(irg, copy_node_nr);
579 /* fix the fields in irg */
580 old_end = get_irg_end(irg);
581 for (i = anchor_max - 1; i >= 0; --i) {
584 irg->anchors[i] = get_new_node(n);
590 * Copies all reachable nodes to a new obstack. Removes bad inputs
591 * from block nodes and the corresponding inputs from Phi nodes.
592 * Merges single exit blocks with single entry blocks and removes
594 * Adds all new nodes to a new hash table for CSE. Does not
595 * perform CSE, so the hash table might contain common subexpressions.
598 dead_node_elimination(ir_graph *irg) {
599 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
601 int rem_ipview = get_interprocedural_view();
602 struct obstack *graveyard_obst = NULL;
603 struct obstack *rebirth_obst = NULL;
604 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
606 /* inform statistics that we started a dead-node elimination run */
607 hook_dead_node_elim(irg, 1);
609 /* Remember external state of current_ir_graph. */
610 rem = current_ir_graph;
611 current_ir_graph = irg;
612 set_interprocedural_view(0);
614 assert(get_irg_phase_state(irg) != phase_building);
616 /* Handle graph state */
617 free_callee_info(irg);
621 /* @@@ so far we loose loops when copying */
622 free_loop_information(irg);
624 set_irg_doms_inconsistent(irg);
626 /* A quiet place, where the old obstack can rest in peace,
627 until it will be cremated. */
628 graveyard_obst = irg->obst;
630 /* A new obstack, where the reachable nodes will be copied to. */
631 rebirth_obst = xmalloc(sizeof(*rebirth_obst));
632 irg->obst = rebirth_obst;
633 obstack_init(irg->obst);
634 irg->last_node_idx = 0;
636 /* We also need a new value table for CSE */
637 del_identities(irg->value_table);
638 irg->value_table = new_identities();
640 /* Copy the graph from the old to the new obstack */
641 copy_graph_env(/*copy_node_nr=*/1);
643 /* Free memory from old unoptimized obstack */
644 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
645 xfree (graveyard_obst); /* ... then free it. */
647 /* inform statistics that the run is over */
648 hook_dead_node_elim(irg, 0);
650 current_ir_graph = rem;
651 set_interprocedural_view(rem_ipview);
656 * Relink bad predecessors of a block and store the old in array to the
657 * link field. This function is called by relink_bad_predecessors().
658 * The array of link field starts with the block operand at position 0.
659 * If block has bad predecessors, create a new in array without bad preds.
660 * Otherwise let in array untouched.
662 static void relink_bad_block_predecessors(ir_node *n, void *env) {
663 ir_node **new_in, *irn;
664 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
667 /* if link field of block is NULL, look for bad predecessors otherwise
668 this is already done */
669 if (get_irn_op(n) == op_Block &&
670 get_irn_link(n) == NULL) {
672 /* save old predecessors in link field (position 0 is the block operand)*/
673 set_irn_link(n, get_irn_in(n));
675 /* count predecessors without bad nodes */
676 old_irn_arity = get_irn_arity(n);
677 for (i = 0; i < old_irn_arity; i++)
678 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
680 /* arity changing: set new predecessors without bad nodes */
681 if (new_irn_arity < old_irn_arity) {
682 /* Get new predecessor array. We do not resize the array, as we must
683 keep the old one to update Phis. */
684 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
686 /* set new predecessors in array */
689 for (i = 0; i < old_irn_arity; i++) {
690 irn = get_irn_n(n, i);
692 new_in[new_irn_n] = irn;
693 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
697 /* ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity); */
698 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
700 } /* ir node has bad predecessors */
701 } /* Block is not relinked */
705 * Relinks Bad predecessors from Blocks and Phis called by walker
706 * remove_bad_predecesors(). If n is a Block, call
707 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
708 * function of Phi's Block. If this block has bad predecessors, relink preds
711 static void relink_bad_predecessors(ir_node *n, void *env) {
712 ir_node *block, **old_in;
713 int i, old_irn_arity, new_irn_arity;
715 /* relink bad predecessors of a block */
716 if (get_irn_op(n) == op_Block)
717 relink_bad_block_predecessors(n, env);
719 /* If Phi node relink its block and its predecessors */
720 if (get_irn_op(n) == op_Phi) {
722 /* Relink predecessors of phi's block */
723 block = get_nodes_block(n);
724 if (get_irn_link(block) == NULL)
725 relink_bad_block_predecessors(block, env);
727 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
728 old_irn_arity = ARR_LEN(old_in);
730 /* Relink Phi predecessors if count of predecessors changed */
731 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
732 /* set new predecessors in array
733 n->in[0] remains the same block */
735 for(i = 1; i < old_irn_arity; i++)
736 if (!is_Bad((ir_node *)old_in[i])) {
737 n->in[new_irn_arity] = n->in[i];
738 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
742 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
743 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
745 } /* n is a Phi node */
749 * Removes Bad Bad predecessors from Blocks and the corresponding
750 * inputs to Phi nodes as in dead_node_elimination but without
752 * On walking up set the link field to NULL, on walking down call
753 * relink_bad_predecessors() (This function stores the old in array
754 * to the link field and sets a new in array if arity of predecessors
757 void remove_bad_predecessors(ir_graph *irg) {
758 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
765 __)|_| | \_/ | \_/(/_ |_/\__|__
767 The following stuff implements a facility that automatically patches
768 registered ir_node pointers to the new node when a dead node elimination occurs.
771 struct _survive_dce_t {
775 hook_entry_t dead_node_elim;
776 hook_entry_t dead_node_elim_subst;
779 typedef struct _survive_dce_list_t {
780 struct _survive_dce_list_t *next;
782 } survive_dce_list_t;
784 static void dead_node_hook(void *context, ir_graph *irg, int start) {
785 survive_dce_t *sd = context;
788 /* Create a new map before the dead node elimination is performed. */
790 sd->new_places = pmap_create_ex(pmap_count(sd->places));
792 /* Patch back all nodes if dead node elimination is over and something is to be done. */
793 pmap_destroy(sd->places);
794 sd->places = sd->new_places;
795 sd->new_places = NULL;
800 * Hook called when dead node elimination replaces old by nw.
802 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw) {
803 survive_dce_t *sd = context;
804 survive_dce_list_t *list = pmap_get(sd->places, old);
807 /* If the node is to be patched back, write the new address to all registered locations. */
809 survive_dce_list_t *p;
811 for (p = list; p; p = p->next)
814 pmap_insert(sd->new_places, nw, list);
819 * Make a new Survive DCE environment.
821 survive_dce_t *new_survive_dce(void) {
822 survive_dce_t *res = xmalloc(sizeof(res[0]));
823 obstack_init(&res->obst);
824 res->places = pmap_create();
825 res->new_places = NULL;
827 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
828 res->dead_node_elim.context = res;
829 res->dead_node_elim.next = NULL;
831 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
832 res->dead_node_elim_subst.context = res;
833 res->dead_node_elim_subst.next = NULL;
835 #ifndef FIRM_ENABLE_HOOKS
836 assert(0 && "need hooks enabled");
839 register_hook(hook_dead_node_elim, &res->dead_node_elim);
840 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
845 * Free a Survive DCE environment.
847 void free_survive_dce(survive_dce_t *sd) {
848 obstack_free(&sd->obst, NULL);
849 pmap_destroy(sd->places);
850 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
851 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
856 * Register a node pointer to be patched upon DCE.
857 * When DCE occurs, the node pointer specified by @p place will be
858 * patched to the new address of the node it is pointing to.
860 * @param sd The Survive DCE environment.
861 * @param place The address of the node pointer.
863 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place) {
864 if (*place != NULL) {
865 ir_node *irn = *place;
866 survive_dce_list_t *curr = pmap_get(sd->places, irn);
867 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw[0]));
872 pmap_insert(sd->places, irn, nw);
876 /*--------------------------------------------------------------------*/
877 /* Functionality for inlining */
878 /*--------------------------------------------------------------------*/
881 * Copy node for inlineing. Updates attributes that change when
882 * inlineing but not for dead node elimination.
884 * Copies the node by calling copy_node() and then updates the entity if
885 * it's a local one. env must be a pointer of the frame type of the
886 * inlined procedure. The new entities must be in the link field of
890 copy_node_inline(ir_node *n, void *env) {
892 ir_type *frame_tp = (ir_type *)env;
895 if (get_irn_op(n) == op_Sel) {
896 nn = get_new_node (n);
898 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
899 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
901 } else if (get_irn_op(n) == op_Block) {
902 nn = get_new_node (n);
903 nn->attr.block.irg = current_ir_graph;
908 * Walker: checks if P_value_arg_base is used.
910 static void find_addr(ir_node *node, void *env) {
911 int *allow_inline = env;
912 if (is_Proj(node) && get_irn_op(get_Proj_pred(node)) == op_Start) {
913 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
919 * Check if we can inline a given call.
920 * Currently, we cannot inline two cases:
921 * - call with compound arguments
922 * - graphs that take the address of a parameter
924 * check these conditions here
926 static int can_inline(ir_node *call, ir_graph *called_graph) {
927 ir_type *call_type = get_Call_type(call);
928 int params, ress, i, res;
929 assert(is_Method_type(call_type));
931 params = get_method_n_params(call_type);
932 ress = get_method_n_ress(call_type);
934 /* check parameters for compound arguments */
935 for (i = 0; i < params; ++i) {
936 ir_type *p_type = get_method_param_type(call_type, i);
938 if (is_compound_type(p_type))
942 /* check results for compound arguments */
943 for (i = 0; i < ress; ++i) {
944 ir_type *r_type = get_method_res_type(call_type, i);
946 if (is_compound_type(r_type))
951 irg_walk_graph(called_graph, find_addr, NULL, &res);
957 exc_handler = 0, /**< There is a handler. */
958 exc_to_end = 1, /**< Branches to End. */
959 exc_no_handler = 2 /**< Exception handling not represented. */
962 /* Inlines a method at the given call site. */
963 int inline_method(ir_node *call, ir_graph *called_graph) {
965 ir_node *post_call, *post_bl;
966 ir_node *in[pn_Start_max];
967 ir_node *end, *end_bl;
971 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
972 enum exc_mode exc_handling;
973 ir_type *called_frame;
974 irg_inline_property prop = get_irg_inline_property(called_graph);
976 if ( (prop < irg_inline_forced) &&
977 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
979 /* Do not inline variadic functions. */
980 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
983 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
984 get_method_n_params(get_Call_type(call)));
987 * currently, we cannot inline two cases:
988 * - call with compound arguments
989 * - graphs that take the address of a parameter
991 if (! can_inline(call, called_graph))
994 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
995 rem_opt = get_opt_optimize();
998 /* Handle graph state */
999 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1000 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
1001 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
1002 set_irg_outs_inconsistent(current_ir_graph);
1003 set_irg_extblk_inconsistent(current_ir_graph);
1004 set_irg_doms_inconsistent(current_ir_graph);
1005 set_irg_loopinfo_inconsistent(current_ir_graph);
1006 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
1008 /* -- Check preconditions -- */
1009 assert(is_Call(call));
1010 /* @@@ does not work for InterfaceIII.java after cgana
1011 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
1012 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
1013 get_Call_type(call)));
1015 if (called_graph == current_ir_graph) {
1016 set_optimize(rem_opt);
1020 /* here we know we WILL inline, so inform the statistics */
1021 hook_inline(call, called_graph);
1023 /* -- Decide how to handle exception control flow: Is there a handler
1024 for the Call node, or do we branch directly to End on an exception?
1026 0 There is a handler.
1028 2 Exception handling not represented in Firm. -- */
1030 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
1031 for (proj = get_irn_link(call); proj; proj = get_irn_link(proj)) {
1032 long proj_nr = get_Proj_proj(proj);
1033 if (proj_nr == pn_Call_X_except) Xproj = proj;
1034 if (proj_nr == pn_Call_M_except) Mproj = proj;
1036 if (Mproj) { assert(Xproj); exc_handling = exc_handler; } /* Mproj */
1037 else if (Xproj) { exc_handling = exc_to_end; } /* !Mproj && Xproj */
1038 else { exc_handling = exc_no_handler; } /* !Mproj && !Xproj */
1042 the procedure and later replaces the Start node of the called graph.
1043 Post_call is the old Call node and collects the results of the called
1044 graph. Both will end up being a tuple. -- */
1045 post_bl = get_nodes_block(call);
1046 set_irg_current_block(current_ir_graph, post_bl);
1047 /* XxMxPxPxPxT of Start + parameter of Call */
1048 in[pn_Start_X_initial_exec] = new_Jmp();
1049 in[pn_Start_M] = get_Call_mem(call);
1050 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
1051 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
1052 in[pn_Start_P_tls] = get_irg_tls(current_ir_graph);
1053 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
1054 /* in[pn_Start_P_value_arg_base] = ??? */
1055 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
1056 pre_call = new_Tuple(pn_Start_max - 1, in);
1060 The new block gets the ins of the old block, pre_call and all its
1061 predecessors and all Phi nodes. -- */
1062 part_block(pre_call);
1064 /* -- Prepare state for dead node elimination -- */
1065 /* Visited flags in calling irg must be >= flag in called irg.
1066 Else walker and arity computation will not work. */
1067 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
1068 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
1069 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
1070 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
1071 /* Set pre_call as new Start node in link field of the start node of
1072 calling graph and pre_calls block as new block for the start block
1074 Further mark these nodes so that they are not visited by the
1076 set_irn_link(get_irg_start(called_graph), pre_call);
1077 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
1078 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1079 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
1080 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
1081 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
1083 /* Initialize for compaction of in arrays */
1084 inc_irg_block_visited(current_ir_graph);
1086 /* -- Replicate local entities of the called_graph -- */
1087 /* copy the entities. */
1088 called_frame = get_irg_frame_type(called_graph);
1089 for (i = 0; i < get_class_n_members(called_frame); i++) {
1090 ir_entity *new_ent, *old_ent;
1091 old_ent = get_class_member(called_frame, i);
1092 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
1093 set_entity_link(old_ent, new_ent);
1096 /* visited is > than that of called graph. With this trick visited will
1097 remain unchanged so that an outer walker, e.g., searching the call nodes
1098 to inline, calling this inline will not visit the inlined nodes. */
1099 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
1101 /* -- Performing dead node elimination inlines the graph -- */
1102 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1104 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
1105 get_irg_frame_type(called_graph));
1107 /* Repair called_graph */
1108 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
1109 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
1110 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1112 /* -- Merge the end of the inlined procedure with the call site -- */
1113 /* We will turn the old Call node into a Tuple with the following
1116 0: Phi of all Memories of Return statements.
1117 1: Jmp from new Block that merges the control flow from all exception
1118 predecessors of the old end block.
1119 2: Tuple of all arguments.
1120 3: Phi of Exception memories.
1121 In case the old Call directly branches to End on an exception we don't
1122 need the block merging all exceptions nor the Phi of the exception
1126 /* -- Precompute some values -- */
1127 end_bl = get_new_node(get_irg_end_block(called_graph));
1128 end = get_new_node(get_irg_end(called_graph));
1129 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1130 n_res = get_method_n_ress(get_Call_type(call));
1132 res_pred = xmalloc(n_res * sizeof(*res_pred));
1133 cf_pred = xmalloc(arity * sizeof(*res_pred));
1135 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
1137 /* -- archive keepalives -- */
1138 irn_arity = get_irn_arity(end);
1139 for (i = 0; i < irn_arity; i++) {
1140 ir_node *ka = get_End_keepalive(end, i);
1142 add_End_keepalive(get_irg_end(current_ir_graph), ka);
1145 /* The new end node will die. We need not free as the in array is on the obstack:
1146 copy_node() only generated 'D' arrays. */
1148 /* -- Replace Return nodes by Jump nodes. -- */
1150 for (i = 0; i < arity; i++) {
1152 ret = get_irn_n(end_bl, i);
1153 if (is_Return(ret)) {
1154 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
1158 set_irn_in(post_bl, n_ret, cf_pred);
1160 /* -- Build a Tuple for all results of the method.
1161 Add Phi node if there was more than one Return. -- */
1162 turn_into_tuple(post_call, pn_Call_max);
1163 /* First the Memory-Phi */
1165 for (i = 0; i < arity; i++) {
1166 ret = get_irn_n(end_bl, i);
1167 if (is_Return(ret)) {
1168 cf_pred[n_ret] = get_Return_mem(ret);
1172 phi = new_Phi(n_ret, cf_pred, mode_M);
1173 set_Tuple_pred(call, pn_Call_M_regular, phi);
1174 /* Conserve Phi-list for further inlinings -- but might be optimized */
1175 if (get_nodes_block(phi) == post_bl) {
1176 set_irn_link(phi, get_irn_link(post_bl));
1177 set_irn_link(post_bl, phi);
1179 /* Now the real results */
1181 for (j = 0; j < n_res; j++) {
1183 for (i = 0; i < arity; i++) {
1184 ret = get_irn_n(end_bl, i);
1185 if (get_irn_op(ret) == op_Return) {
1186 cf_pred[n_ret] = get_Return_res(ret, j);
1191 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1195 /* Conserve Phi-list for further inlinings -- but might be optimized */
1196 if (get_nodes_block(phi) == post_bl) {
1197 set_irn_link(phi, get_irn_link(post_bl));
1198 set_irn_link(post_bl, phi);
1201 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1203 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1206 /* For now, we cannot inline calls with value_base */
1207 set_Tuple_pred(call, pn_Call_P_value_res_base, new_Bad());
1209 /* Finally the exception control flow.
1210 We have two (three) possible situations:
1211 First if the Call branches to an exception handler: We need to add a Phi node to
1212 collect the memory containing the exception objects. Further we need
1213 to add another block to get a correct representation of this Phi. To
1214 this block we add a Jmp that resolves into the X output of the Call
1215 when the Call is turned into a tuple.
1216 Second the Call branches to End, the exception is not handled. Just
1217 add all inlined exception branches to the End node.
1218 Third: there is no Exception edge at all. Handle as case two. */
1219 if (exc_handling == exc_handler) {
1221 for (i = 0; i < arity; i++) {
1223 ret = get_irn_n(end_bl, i);
1224 irn = skip_Proj(ret);
1225 if (is_fragile_op(irn) || (get_irn_op(irn) == op_Raise)) {
1226 cf_pred[n_exc] = ret;
1231 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1232 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1233 /* The Phi for the memories with the exception objects */
1235 for (i = 0; i < arity; i++) {
1237 ret = skip_Proj(get_irn_n(end_bl, i));
1239 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1241 } else if (is_fragile_op(ret)) {
1242 /* We rely that all cfops have the memory output at the same position. */
1243 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1245 } else if (get_irn_op(ret) == op_Raise) {
1246 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1250 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1252 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1253 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1255 set_Tuple_pred(call, pn_Call_X_regular, new_Bad());
1257 ir_node *main_end_bl;
1258 int main_end_bl_arity;
1259 ir_node **end_preds;
1261 /* assert(exc_handling == 1 || no exceptions. ) */
1263 for (i = 0; i < arity; i++) {
1264 ir_node *ret = get_irn_n(end_bl, i);
1265 ir_node *irn = skip_Proj(ret);
1267 if (is_fragile_op(irn) || (get_irn_op(irn) == op_Raise)) {
1268 cf_pred[n_exc] = ret;
1272 main_end_bl = get_irg_end_block(current_ir_graph);
1273 main_end_bl_arity = get_irn_arity(main_end_bl);
1274 end_preds = xmalloc((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1276 for (i = 0; i < main_end_bl_arity; ++i)
1277 end_preds[i] = get_irn_n(main_end_bl, i);
1278 for (i = 0; i < n_exc; ++i)
1279 end_preds[main_end_bl_arity + i] = cf_pred[i];
1280 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1281 set_Tuple_pred(call, pn_Call_X_regular, new_Bad());
1282 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1283 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1289 /* -- Turn CSE back on. -- */
1290 set_optimize(rem_opt);
1295 /********************************************************************/
1296 /* Apply inlineing to small methods. */
1297 /********************************************************************/
1299 /** Represents a possible inlinable call in a graph. */
1300 typedef struct _call_entry call_entry;
1301 struct _call_entry {
1302 ir_node *call; /**< the Call */
1303 ir_graph *callee; /**< the callee called here */
1304 call_entry *next; /**< for linking the next one */
1308 * environment for inlining small irgs
1310 typedef struct _inline_env_t {
1311 struct obstack obst; /**< an obstack where call_entries are allocated on. */
1312 call_entry *head; /**< the head of the call entry list */
1313 call_entry *tail; /**< the tail of the call entry list */
1317 * Returns the irg called from a Call node. If the irg is not
1318 * known, NULL is returned.
1320 static ir_graph *get_call_called_irg(ir_node *call) {
1322 ir_graph *called_irg = NULL;
1324 addr = get_Call_ptr(call);
1325 if (is_SymConst(addr) && get_SymConst_kind(addr) == symconst_addr_ent) {
1326 called_irg = get_entity_irg(get_SymConst_entity(addr));
1333 * Walker: Collect all calls to known graphs inside a graph.
1335 static void collect_calls(ir_node *call, void *env) {
1336 if (is_Call(call)) {
1337 ir_graph *called_irg = get_call_called_irg(call);
1339 /* The Call node calls a locally defined method. Remember to inline. */
1340 inline_env_t *ienv = env;
1341 call_entry *entry = obstack_alloc(&ienv->obst, sizeof(*entry));
1343 entry->callee = called_irg;
1346 if (ienv->tail == NULL)
1349 ienv->tail->next = entry;
1356 * Inlines all small methods at call sites where the called address comes
1357 * from a Const node that references the entity representing the called
1359 * The size argument is a rough measure for the code size of the method:
1360 * Methods where the obstack containing the firm graph is smaller than
1363 void inline_small_irgs(ir_graph *irg, int size) {
1364 ir_graph *rem = current_ir_graph;
1367 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1369 if (!(get_opt_optimize() && get_opt_inline())) return;
1371 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1373 current_ir_graph = irg;
1374 /* Handle graph state */
1375 assert(get_irg_phase_state(irg) != phase_building);
1376 free_callee_info(irg);
1378 /* Find Call nodes to inline.
1379 (We can not inline during a walk of the graph, as inlineing the same
1380 method several times changes the visited flag of the walked graph:
1381 after the first inlineing visited of the callee equals visited of
1382 the caller. With the next inlineing both are increased.) */
1383 obstack_init(&env.obst);
1384 env.head = env.tail = NULL;
1385 irg_walk_graph(irg, NULL, collect_calls, &env);
1387 if (env.head != NULL) {
1388 /* There are calls to inline */
1389 collect_phiprojs(irg);
1390 for (entry = env.head; entry != NULL; entry = entry->next) {
1391 ir_graph *callee = entry->callee;
1392 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1393 (get_irg_inline_property(callee) >= irg_inline_forced)) {
1394 inline_method(entry->call, callee);
1398 obstack_free(&env.obst, NULL);
1399 current_ir_graph = rem;
1403 * Environment for inlining irgs.
1406 int n_nodes; /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
1407 int n_nodes_orig; /**< for statistics */
1408 call_entry *call_head; /**< The head of the list of all call nodes in this graph. */
1409 call_entry *call_tail; /**< The tail of the list of all call nodes in this graph .*/
1410 int n_call_nodes; /**< Number of Call nodes in the graph. */
1411 int n_call_nodes_orig; /**< for statistics */
1412 int n_callers; /**< Number of known graphs that call this graphs. */
1413 int n_callers_orig; /**< for statistics */
1414 int got_inline; /**< Set, if at leat one call inside this graph was inlined. */
1418 * Allocate a new environment for inlining.
1420 static inline_irg_env *alloc_inline_irg_env(struct obstack *obst) {
1421 inline_irg_env *env = obstack_alloc(obst, sizeof(*env));
1422 env->n_nodes = -2; /* do not count count Start, End */
1423 env->n_nodes_orig = -2; /* do not count Start, End */
1424 env->call_head = NULL;
1425 env->call_tail = NULL;
1426 env->n_call_nodes = 0;
1427 env->n_call_nodes_orig = 0;
1429 env->n_callers_orig = 0;
1430 env->got_inline = 0;
1434 typedef struct walker_env {
1435 struct obstack *obst; /**< the obstack for allocations. */
1436 inline_irg_env *x; /**< the inline environment */
1437 int ignore_runtime; /**< the ignore runtime flag */
1441 * post-walker: collect all calls in the inline-environment
1442 * of a graph and sum some statistics.
1444 static void collect_calls2(ir_node *call, void *ctx) {
1446 inline_irg_env *x = env->x;
1447 ir_op *op = get_irn_op(call);
1451 /* count meaningful nodes in irg */
1452 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1457 if (op != op_Call) return;
1459 /* check, if it's a runtime call */
1460 if (env->ignore_runtime) {
1461 ir_node *symc = get_Call_ptr(call);
1463 if (is_SymConst(symc) && get_SymConst_kind(symc) == symconst_addr_ent) {
1464 ir_entity *ent = get_SymConst_entity(symc);
1466 if (get_entity_additional_properties(ent) & mtp_property_runtime)
1471 /* collect all call nodes */
1473 ++x->n_call_nodes_orig;
1475 callee = get_call_called_irg(call);
1477 inline_irg_env *callee_env = get_irg_link(callee);
1478 /* count all static callers */
1479 ++callee_env->n_callers;
1480 ++callee_env->n_callers_orig;
1482 /* link it in the list of possible inlinable entries */
1483 entry = obstack_alloc(env->obst, sizeof(*entry));
1485 entry->callee = callee;
1487 if (x->call_tail == NULL)
1488 x->call_head = entry;
1490 x->call_tail->next = entry;
1491 x->call_tail = entry;
1496 * Returns TRUE if the number of callers in 0 in the irg's environment,
1497 * hence this irg is a leave.
1499 INLINE static int is_leave(ir_graph *irg) {
1500 inline_irg_env *env = get_irg_link(irg);
1501 return env->n_call_nodes == 0;
1505 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1507 INLINE static int is_smaller(ir_graph *callee, int size) {
1508 inline_irg_env *env = get_irg_link(callee);
1509 return env->n_nodes < size;
1513 * Append the nodes of the list src to the nodes of the list in environment dst.
1515 static void append_call_list(struct obstack *obst, inline_irg_env *dst, call_entry *src) {
1516 call_entry *entry, *nentry;
1518 /* Note that the src list points to Call nodes in the inlined graph, but
1519 we need Call nodes in our graph. Luckily the inliner leaves this information
1520 in the link field. */
1521 for (entry = src; entry != NULL; entry = entry->next) {
1522 nentry = obstack_alloc(obst, sizeof(*nentry));
1523 nentry->call = get_irn_link(entry->call);
1524 nentry->callee = entry->callee;
1525 nentry->next = NULL;
1526 dst->call_tail->next = nentry;
1527 dst->call_tail = nentry;
1532 * Inlines small leave methods at call sites where the called address comes
1533 * from a Const node that references the entity representing the called
1535 * The size argument is a rough measure for the code size of the method:
1536 * Methods where the obstack containing the firm graph is smaller than
1539 void inline_leave_functions(int maxsize, int leavesize, int size, int ignore_runtime) {
1540 inline_irg_env *env;
1546 call_entry *entry, *tail;
1547 const call_entry *centry;
1548 struct obstack obst;
1549 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1551 if (!(get_opt_optimize() && get_opt_inline())) return;
1553 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1554 rem = current_ir_graph;
1555 obstack_init(&obst);
1557 /* extend all irgs by a temporary data structure for inlining. */
1558 n_irgs = get_irp_n_irgs();
1559 for (i = 0; i < n_irgs; ++i)
1560 set_irg_link(get_irp_irg(i), alloc_inline_irg_env(&obst));
1562 /* Precompute information in temporary data structure. */
1564 wenv.ignore_runtime = ignore_runtime;
1565 for (i = 0; i < n_irgs; ++i) {
1566 ir_graph *irg = get_irp_irg(i);
1568 assert(get_irg_phase_state(irg) != phase_building);
1569 free_callee_info(irg);
1571 wenv.x = get_irg_link(irg);
1572 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1575 /* -- and now inline. -- */
1577 /* Inline leaves recursively -- we might construct new leaves. */
1581 for (i = 0; i < n_irgs; ++i) {
1583 int phiproj_computed = 0;
1585 current_ir_graph = get_irp_irg(i);
1586 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1589 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1592 if (env->n_nodes > maxsize) break;
1595 callee = entry->callee;
1597 if (is_leave(callee) && is_smaller(callee, leavesize)) {
1598 if (!phiproj_computed) {
1599 phiproj_computed = 1;
1600 collect_phiprojs(current_ir_graph);
1602 did_inline = inline_method(call, callee);
1605 /* Do some statistics */
1606 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1608 env->got_inline = 1;
1609 --env->n_call_nodes;
1610 env->n_nodes += callee_env->n_nodes;
1611 --callee_env->n_callers;
1613 /* remove this call from the list */
1615 tail->next = entry->next;
1617 env->call_head = entry->next;
1623 env->call_tail = tail;
1625 } while (did_inline);
1627 /* inline other small functions. */
1628 for (i = 0; i < n_irgs; ++i) {
1630 int phiproj_computed = 0;
1632 current_ir_graph = get_irp_irg(i);
1633 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1635 /* note that the list of possible calls is updated during the process */
1637 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1641 callee = entry->callee;
1643 if (((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1644 (get_irg_inline_property(callee) >= irg_inline_forced))) {
1645 if (!phiproj_computed) {
1646 phiproj_computed = 1;
1647 collect_phiprojs(current_ir_graph);
1649 if (inline_method(call, callee)) {
1650 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1652 /* callee was inline. Append it's call list. */
1653 env->got_inline = 1;
1654 --env->n_call_nodes;
1655 append_call_list(&obst, env, callee_env->call_head);
1656 env->n_call_nodes += callee_env->n_call_nodes;
1657 env->n_nodes += callee_env->n_nodes;
1658 --callee_env->n_callers;
1660 /* after we have inlined callee, all called methods inside callee
1661 are now called once more */
1662 for (centry = callee_env->call_head; centry != NULL; centry = centry->next) {
1663 inline_irg_env *penv = get_irg_link(centry->callee);
1667 /* remove this call from the list */
1669 tail->next = entry->next;
1671 env->call_head = entry->next;
1677 env->call_tail = tail;
1680 for (i = 0; i < n_irgs; ++i) {
1681 irg = get_irp_irg(i);
1682 env = (inline_irg_env *)get_irg_link(irg);
1684 if (env->got_inline) {
1685 /* this irg got calls inlined */
1686 set_irg_outs_inconsistent(irg);
1687 set_irg_doms_inconsistent(irg);
1689 optimize_graph_df(irg);
1692 if (env->got_inline || (env->n_callers_orig != env->n_callers))
1693 DB((dbg, SET_LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1694 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1695 env->n_callers_orig, env->n_callers,
1696 get_entity_name(get_irg_entity(irg))));
1699 obstack_free(&obst, NULL);
1700 current_ir_graph = rem;
1703 /*******************************************************************/
1704 /* Code Placement. Pins all floating nodes to a block where they */
1705 /* will be executed only if needed. */
1706 /*******************************************************************/
1709 * Returns non-zero, is a block is not reachable from Start.
1711 * @param block the block to test
1714 is_Block_unreachable(ir_node *block) {
1715 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1719 * Find the earliest correct block for node n. --- Place n into the
1720 * same Block as its dominance-deepest Input.
1722 * We have to avoid calls to get_nodes_block() here
1723 * because the graph is floating.
1725 * move_out_of_loops() expects that place_floats_early() have placed
1726 * all "living" nodes into a living block. That's why we must
1727 * move nodes in dead block with "live" successors into a valid
1729 * We move them just into the same block as it's successor (or
1730 * in case of a Phi into the effective use block). For Phi successors,
1731 * this may still be a dead block, but then there is no real use, as
1732 * the control flow will be dead later.
1734 * @param n the node to be placed
1735 * @param worklist a worklist, predecessors of non-floating nodes are placed here
1738 place_floats_early(ir_node *n, waitq *worklist) {
1741 /* we must not run into an infinite loop */
1742 assert(irn_not_visited(n));
1743 mark_irn_visited(n);
1745 /* Place floating nodes. */
1746 if (get_irn_pinned(n) == op_pin_state_floats) {
1747 ir_node *curr_block = get_irn_n(n, -1);
1748 int in_dead_block = is_Block_unreachable(curr_block);
1750 ir_node *b = NULL; /* The block to place this node in */
1752 assert(is_no_Block(n));
1754 if (is_irn_start_block_placed(n)) {
1755 /* These nodes will not be placed by the loop below. */
1756 b = get_irg_start_block(current_ir_graph);
1760 /* find the block for this node. */
1761 irn_arity = get_irn_arity(n);
1762 for (i = 0; i < irn_arity; i++) {
1763 ir_node *pred = get_irn_n(n, i);
1764 ir_node *pred_block;
1766 if ((irn_not_visited(pred))
1767 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1770 * If the current node is NOT in a dead block, but one of its
1771 * predecessors is, we must move the predecessor to a live block.
1772 * Such thing can happen, if global CSE chose a node from a dead block.
1773 * We move it simply to our block.
1774 * Note that neither Phi nor End nodes are floating, so we don't
1775 * need to handle them here.
1777 if (! in_dead_block) {
1778 if (get_irn_pinned(pred) == op_pin_state_floats &&
1779 is_Block_unreachable(get_irn_n(pred, -1)))
1780 set_nodes_block(pred, curr_block);
1782 place_floats_early(pred, worklist);
1786 * A node in the Bad block must stay in the bad block,
1787 * so don't compute a new block for it.
1792 /* Because all loops contain at least one op_pin_state_pinned node, now all
1793 our inputs are either op_pin_state_pinned or place_early() has already
1794 been finished on them. We do not have any unfinished inputs! */
1795 pred_block = get_irn_n(pred, -1);
1796 if ((!is_Block_dead(pred_block)) &&
1797 (get_Block_dom_depth(pred_block) > depth)) {
1799 depth = get_Block_dom_depth(pred_block);
1801 /* Avoid that the node is placed in the Start block */
1802 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)
1803 && get_irg_phase_state(current_ir_graph) != phase_backend) {
1804 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1805 assert(b != get_irg_start_block(current_ir_graph));
1810 set_nodes_block(n, b);
1814 * Add predecessors of non floating nodes and non-floating predecessors
1815 * of floating nodes to worklist and fix their blocks if the are in dead block.
1817 irn_arity = get_irn_arity(n);
1819 if (get_irn_op(n) == op_End) {
1821 * Simplest case: End node. Predecessors are keep-alives,
1822 * no need to move out of dead block.
1824 for (i = -1; i < irn_arity; ++i) {
1825 ir_node *pred = get_irn_n(n, i);
1826 if (irn_not_visited(pred))
1827 waitq_put(worklist, pred);
1829 } else if (is_Block(n)) {
1831 * Blocks: Predecessors are control flow, no need to move
1832 * them out of dead block.
1834 for (i = irn_arity - 1; i >= 0; --i) {
1835 ir_node *pred = get_irn_n(n, i);
1836 if (irn_not_visited(pred))
1837 waitq_put(worklist, pred);
1839 } else if (is_Phi(n)) {
1841 ir_node *curr_block = get_irn_n(n, -1);
1842 int in_dead_block = is_Block_unreachable(curr_block);
1845 * Phi nodes: move nodes from dead blocks into the effective use
1846 * of the Phi-input if the Phi is not in a bad block.
1848 pred = get_irn_n(n, -1);
1849 if (irn_not_visited(pred))
1850 waitq_put(worklist, pred);
1852 for (i = irn_arity - 1; i >= 0; --i) {
1853 ir_node *pred = get_irn_n(n, i);
1855 if (irn_not_visited(pred)) {
1856 if (! in_dead_block &&
1857 get_irn_pinned(pred) == op_pin_state_floats &&
1858 is_Block_unreachable(get_irn_n(pred, -1))) {
1859 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1861 waitq_put(worklist, pred);
1866 ir_node *curr_block = get_irn_n(n, -1);
1867 int in_dead_block = is_Block_unreachable(curr_block);
1870 * All other nodes: move nodes from dead blocks into the same block.
1872 pred = get_irn_n(n, -1);
1873 if (irn_not_visited(pred))
1874 waitq_put(worklist, pred);
1876 for (i = irn_arity - 1; i >= 0; --i) {
1877 ir_node *pred = get_irn_n(n, i);
1879 if (irn_not_visited(pred)) {
1880 if (! in_dead_block &&
1881 get_irn_pinned(pred) == op_pin_state_floats &&
1882 is_Block_unreachable(get_irn_n(pred, -1))) {
1883 set_nodes_block(pred, curr_block);
1885 waitq_put(worklist, pred);
1892 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1893 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1894 * places all floating nodes reachable from its argument through floating
1895 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1897 * @param worklist a worklist, used for the algorithm, empty on in/output
1899 static void place_early(waitq *worklist) {
1901 inc_irg_visited(current_ir_graph);
1903 /* this inits the worklist */
1904 place_floats_early(get_irg_end(current_ir_graph), worklist);
1906 /* Work the content of the worklist. */
1907 while (!waitq_empty(worklist)) {
1908 ir_node *n = waitq_get(worklist);
1909 if (irn_not_visited(n))
1910 place_floats_early(n, worklist);
1913 set_irg_outs_inconsistent(current_ir_graph);
1914 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
1918 * Compute the deepest common ancestor of block and dca.
1920 static ir_node *calc_dca(ir_node *dca, ir_node *block) {
1923 /* we do not want to place nodes in dead blocks */
1924 if (is_Block_dead(block))
1927 /* We found a first legal placement. */
1928 if (!dca) return block;
1930 /* Find a placement that is dominates both, dca and block. */
1931 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1932 block = get_Block_idom(block);
1934 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1935 dca = get_Block_idom(dca);
1938 while (block != dca) {
1939 block = get_Block_idom(block); dca = get_Block_idom(dca);
1945 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1946 * I.e., DCA is the block where we might place PRODUCER.
1947 * A data flow edge points from producer to consumer.
1950 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer) {
1951 ir_node *block = NULL;
1953 /* Compute the latest block into which we can place a node so that it is
1955 if (get_irn_op(consumer) == op_Phi) {
1956 /* our consumer is a Phi-node, the effective use is in all those
1957 blocks through which the Phi-node reaches producer */
1959 ir_node *phi_block = get_nodes_block(consumer);
1960 irn_arity = get_irn_arity(consumer);
1962 for (i = 0; i < irn_arity; i++) {
1963 if (get_irn_n(consumer, i) == producer) {
1964 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1966 if (! is_Block_unreachable(new_block))
1967 block = calc_dca(block, new_block);
1972 block = get_irn_n(producer, -1);
1974 assert(is_no_Block(consumer));
1975 block = get_nodes_block(consumer);
1978 /* Compute the deepest common ancestor of block and dca. */
1979 return calc_dca(dca, block);
1982 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1984 static INLINE int get_irn_loop_depth(ir_node *n) {
1985 return get_loop_depth(get_irn_loop(n));
1989 * Move n to a block with less loop depth than it's current block. The
1990 * new block must be dominated by early.
1992 * @param n the node that should be moved
1993 * @param early the earliest block we can n move to
1995 static void move_out_of_loops(ir_node *n, ir_node *early) {
1996 ir_node *best, *dca;
2000 /* Find the region deepest in the dominator tree dominating
2001 dca with the least loop nesting depth, but still dominated
2002 by our early placement. */
2003 dca = get_nodes_block(n);
2006 while (dca != early) {
2007 dca = get_Block_idom(dca);
2008 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
2009 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
2013 if (best != get_nodes_block(n)) {
2015 printf("Moving out of loop: "); DDMN(n);
2016 printf(" Outermost block: "); DDMN(early);
2017 printf(" Best block: "); DDMN(best);
2018 printf(" Innermost block: "); DDMN(get_nodes_block(n));
2020 set_nodes_block(n, best);
2024 /* deepest common ancestor in the dominator tree of all nodes'
2025 blocks depending on us; our final placement has to dominate DCA. */
2026 static ir_node *get_deepest_common_ancestor(ir_node *node, ir_node *dca)
2030 for (i = get_irn_n_outs(node) - 1; i >= 0; --i) {
2031 ir_node *succ = get_irn_out(node, i);
2036 * This consumer is the End node, a keep alive edge.
2037 * This is not a real consumer, so we ignore it
2043 dca = get_deepest_common_ancestor(succ, dca);
2045 /* ignore if succ is in dead code */
2046 succ_blk = get_irn_n(succ, -1);
2047 if (is_Block_unreachable(succ_blk))
2049 dca = consumer_dom_dca(dca, succ, node);
2056 static void set_projs_block(ir_node *node, ir_node *block)
2060 for (i = get_irn_n_outs(node) - 1; i >= 0; --i) {
2061 ir_node *succ = get_irn_out(node, i);
2063 assert(is_Proj(succ));
2065 if(get_irn_mode(succ) == mode_T) {
2066 set_projs_block(succ, block);
2068 set_nodes_block(succ, block);
2073 * Find the latest legal block for N and place N into the
2074 * `optimal' Block between the latest and earliest legal block.
2075 * The `optimal' block is the dominance-deepest block of those
2076 * with the least loop-nesting-depth. This places N out of as many
2077 * loops as possible and then makes it as control dependent as
2080 * @param n the node to be placed
2081 * @param worklist a worklist, all successors of non-floating nodes are
2084 static void place_floats_late(ir_node *n, pdeq *worklist) {
2088 assert(irn_not_visited(n)); /* no multiple placement */
2090 mark_irn_visited(n);
2092 /* no need to place block nodes, control nodes are already placed. */
2093 if ((get_irn_op(n) != op_Block) &&
2095 (get_irn_mode(n) != mode_X)) {
2096 /* Remember the early_blk placement of this block to move it
2097 out of loop no further than the early_blk placement. */
2098 early_blk = get_irn_n(n, -1);
2101 * BEWARE: Here we also get code, that is live, but
2102 * was in a dead block. If the node is life, but because
2103 * of CSE in a dead block, we still might need it.
2106 /* Assure that our users are all placed, except the Phi-nodes.
2107 --- Each data flow cycle contains at least one Phi-node. We
2108 have to break the `user has to be placed before the
2109 producer' dependence cycle and the Phi-nodes are the
2110 place to do so, because we need to base our placement on the
2111 final region of our users, which is OK with Phi-nodes, as they
2112 are op_pin_state_pinned, and they never have to be placed after a
2113 producer of one of their inputs in the same block anyway. */
2114 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2115 ir_node *succ = get_irn_out(n, i);
2116 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
2117 place_floats_late(succ, worklist);
2120 if (! is_Block_dead(early_blk)) {
2121 /* do only move things that where not dead */
2122 ir_op *op = get_irn_op(n);
2124 /* We have to determine the final block of this node... except for
2125 constants and Projs */
2126 if ((get_irn_pinned(n) == op_pin_state_floats) &&
2128 (op != op_SymConst) &&
2131 /* deepest common ancestor in the dominator tree of all nodes'
2132 blocks depending on us; our final placement has to dominate
2134 ir_node *dca = get_deepest_common_ancestor(n, NULL);
2136 set_nodes_block(n, dca);
2137 move_out_of_loops(n, early_blk);
2138 if(get_irn_mode(n) == mode_T) {
2139 set_projs_block(n, get_nodes_block(n));
2146 /* Add successors of all non-floating nodes on list. (Those of floating
2147 nodes are placed already and therefore are marked.) */
2148 for (i = 0; i < get_irn_n_outs(n); i++) {
2149 ir_node *succ = get_irn_out(n, i);
2150 if (irn_not_visited(get_irn_out(n, i))) {
2151 pdeq_putr(worklist, succ);
2157 * Place floating nodes on the given worklist as late as possible using
2158 * the dominance tree.
2160 * @param worklist the worklist containing the nodes to place
2162 static void place_late(waitq *worklist) {
2164 inc_irg_visited(current_ir_graph);
2166 /* This fills the worklist initially. */
2167 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
2169 /* And now empty the worklist again... */
2170 while (!waitq_empty(worklist)) {
2171 ir_node *n = waitq_get(worklist);
2172 if (irn_not_visited(n))
2173 place_floats_late(n, worklist);
2177 /* Code Placement. */
2178 void place_code(ir_graph *irg) {
2180 ir_graph *rem = current_ir_graph;
2182 current_ir_graph = irg;
2184 /* Handle graph state */
2185 assert(get_irg_phase_state(irg) != phase_building);
2188 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
2189 free_loop_information(irg);
2190 construct_backedges(irg);
2193 /* Place all floating nodes as early as possible. This guarantees
2194 a legal code placement. */
2195 worklist = new_waitq();
2196 place_early(worklist);
2198 /* place_early() invalidates the outs, place_late needs them. */
2199 compute_irg_outs(irg);
2201 /* Now move the nodes down in the dominator tree. This reduces the
2202 unnecessary executions of the node. */
2203 place_late(worklist);
2205 set_irg_outs_inconsistent(current_ir_graph);
2206 set_irg_loopinfo_inconsistent(current_ir_graph);
2207 del_waitq(worklist);
2208 current_ir_graph = rem;
2212 * Called by walker of remove_critical_cf_edges().
2214 * Place an empty block to an edge between a blocks of multiple
2215 * predecessors and a block of multiple successors.
2218 * @param env Environment of walker. The changed field.
2220 static void walk_critical_cf_edges(ir_node *n, void *env) {
2222 ir_node *pre, *block, *jmp;
2224 ir_graph *irg = get_irn_irg(n);
2226 /* Block has multiple predecessors */
2227 arity = get_irn_arity(n);
2229 if (n == get_irg_end_block(irg))
2230 return; /* No use to add a block here. */
2232 for (i = 0; i < arity; ++i) {
2235 pre = get_irn_n(n, i);
2236 cfop = get_irn_op(skip_Proj(pre));
2237 /* Predecessor has multiple successors. Insert new control flow edge but
2238 ignore exception edges. */
2239 if (! is_op_fragile(cfop) && is_op_forking(cfop)) {
2240 /* set predecessor of new block */
2241 block = new_r_Block(irg, 1, &pre);
2242 /* insert new jmp node to new block */
2243 jmp = new_r_Jmp(irg, block);
2244 /* set successor of new block */
2245 set_irn_n(n, i, jmp);
2247 } /* predecessor has multiple successors */
2248 } /* for all predecessors */
2249 } /* n is a multi-entry block */
2252 void remove_critical_cf_edges(ir_graph *irg) {
2255 irg_block_walk_graph(irg, NULL, walk_critical_cf_edges, &changed);
2257 /* control flow changed */
2258 set_irg_outs_inconsistent(irg);
2259 set_irg_extblk_inconsistent(irg);
2260 set_irg_doms_inconsistent(irg);
2261 set_irg_loopinfo_inconsistent(irg);