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.
366 static void copy_preds(ir_node *n, void *env) {
371 nn = get_new_node(n);
374 /* Don't copy Bad nodes. */
376 irn_arity = get_irn_arity(n);
377 for (i = 0; i < irn_arity; i++) {
378 if (! is_Bad(get_irn_n(n, i))) {
379 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
380 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
384 /* repair the block visited flag from above misuse. Repair it in both
385 graphs so that the old one can still be used. */
386 set_Block_block_visited(nn, 0);
387 set_Block_block_visited(n, 0);
388 /* Local optimization could not merge two subsequent blocks if
389 in array contained Bads. Now it's possible.
390 We don't call optimize_in_place as it requires
391 that the fields in ir_graph are set properly. */
392 if ((get_opt_control_flow_straightening()) &&
393 (get_Block_n_cfgpreds(nn) == 1) &&
394 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
395 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
397 /* Jmp jumps into the block it is in -- deal self cycle. */
398 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
399 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
404 } else if (get_irn_op(n) == op_Phi) {
405 /* Don't copy node if corresponding predecessor in block is Bad.
406 The Block itself should not be Bad. */
407 block = get_nodes_block(n);
408 set_irn_n(nn, -1, get_new_node(block));
410 irn_arity = get_irn_arity(n);
411 for (i = 0; i < irn_arity; i++) {
412 if (! is_Bad(get_irn_n(block, i))) {
413 set_irn_n(nn, j, get_new_node(get_irn_n(n, i)));
414 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
418 /* If the pre walker reached this Phi after the post walker visited the
419 block block_visited is > 0. */
420 set_Block_block_visited(get_nodes_block(n), 0);
421 /* Compacting the Phi's ins might generate Phis with only one
423 if (get_irn_arity(nn) == 1)
424 exchange(nn, get_irn_n(nn, 0));
426 irn_arity = get_irn_arity(n);
427 for (i = -1; i < irn_arity; i++)
428 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
430 /* Now the new node is complete. We can add it to the hash table for CSE.
431 @@@ inlining aborts if we identify End. Why? */
432 if (get_irn_op(nn) != op_End)
433 add_identities(current_ir_graph->value_table, nn);
437 * Copies the graph recursively, compacts the keep-alives of the end node.
439 * @param irg the graph to be copied
440 * @param copy_node_nr If non-zero, the node number will be copied
442 static void copy_graph(ir_graph *irg, int copy_node_nr) {
443 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
444 ir_node *ka; /* keep alive */
448 /* Some nodes must be copied by hand, sigh */
449 vfl = get_irg_visited(irg);
450 set_irg_visited(irg, vfl + 1);
452 oe = get_irg_end(irg);
453 mark_irn_visited(oe);
454 /* copy the end node by hand, allocate dynamic in array! */
455 ne = new_ir_node(get_irn_dbg_info(oe),
462 /* Copy the attributes. Well, there might be some in the future... */
463 copy_node_attr(oe, ne);
464 set_new_node(oe, ne);
466 /* copy the Bad node */
467 ob = get_irg_bad(irg);
468 mark_irn_visited(ob);
469 nb = new_ir_node(get_irn_dbg_info(ob),
476 copy_node_attr(ob, nb);
477 set_new_node(ob, nb);
479 /* copy the NoMem node */
480 om = get_irg_no_mem(irg);
481 mark_irn_visited(om);
482 nm = new_ir_node(get_irn_dbg_info(om),
489 copy_node_attr(om, nm);
490 set_new_node(om, nm);
492 /* copy the live nodes */
493 set_irg_visited(irg, vfl);
494 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
496 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
498 /* visit the anchors as well */
499 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
500 ir_node *n = get_irg_anchor(irg, i);
502 if (n && (get_irn_visited(n) <= vfl)) {
503 set_irg_visited(irg, vfl);
504 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
508 /* copy_preds for the end node ... */
509 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
511 /*- ... and now the keep alives. -*/
512 /* First pick the not marked block nodes and walk them. We must pick these
513 first as else we will oversee blocks reachable from Phis. */
514 irn_arity = get_End_n_keepalives(oe);
515 for (i = 0; i < irn_arity; i++) {
516 ka = get_End_keepalive(oe, i);
518 if (get_irn_visited(ka) <= vfl) {
519 /* We must keep the block alive and copy everything reachable */
520 set_irg_visited(irg, vfl);
521 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
523 add_End_keepalive(ne, get_new_node(ka));
527 /* Now pick other nodes. Here we will keep all! */
528 irn_arity = get_End_n_keepalives(oe);
529 for (i = 0; i < irn_arity; i++) {
530 ka = get_End_keepalive(oe, i);
532 if (get_irn_visited(ka) <= vfl) {
533 /* We didn't copy the node yet. */
534 set_irg_visited(irg, vfl);
535 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
537 add_End_keepalive(ne, get_new_node(ka));
541 /* start block sometimes only reached after keep alives */
542 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
543 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
547 * Copies the graph reachable from current_ir_graph->end to the obstack
548 * in current_ir_graph and fixes the environment.
549 * Then fixes the fields in current_ir_graph containing nodes of the
552 * @param copy_node_nr If non-zero, the node number will be copied
555 copy_graph_env(int copy_node_nr) {
556 ir_graph *irg = current_ir_graph;
557 ir_node *old_end, *new_anchor;
560 /* remove end_except and end_reg nodes */
561 old_end = get_irg_end(irg);
562 set_irg_end_except (irg, old_end);
563 set_irg_end_reg (irg, old_end);
565 /* Not all nodes remembered in irg might be reachable
566 from the end node. Assure their link is set to NULL, so that
567 we can test whether new nodes have been computed. */
568 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
569 ir_node *n = get_irg_anchor(irg, i);
571 set_new_node(n, 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);
580 old_end = get_irg_end(irg);
581 new_anchor = new_Anchor(irg);
583 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
584 ir_node *n = get_irg_anchor(irg, i);
586 set_irn_n(new_anchor, i, get_new_node(n));
589 irg->anchor = new_anchor;
591 /* ensure the new anchor is placed in the endblock */
592 set_irn_n(new_anchor, -1, get_irg_end_block(irg));
596 * Copies all reachable nodes to a new obstack. Removes bad inputs
597 * from block nodes and the corresponding inputs from Phi nodes.
598 * Merges single exit blocks with single entry blocks and removes
600 * Adds all new nodes to a new hash table for CSE. Does not
601 * perform CSE, so the hash table might contain common subexpressions.
604 dead_node_elimination(ir_graph *irg) {
605 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
607 int rem_ipview = get_interprocedural_view();
608 struct obstack *graveyard_obst = NULL;
609 struct obstack *rebirth_obst = NULL;
610 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
612 /* inform statistics that we started a dead-node elimination run */
613 hook_dead_node_elim(irg, 1);
615 /* Remember external state of current_ir_graph. */
616 rem = current_ir_graph;
617 current_ir_graph = irg;
618 set_interprocedural_view(0);
620 assert(get_irg_phase_state(irg) != phase_building);
622 /* Handle graph state */
623 free_callee_info(irg);
627 /* @@@ so far we loose loops when copying */
628 free_loop_information(irg);
630 set_irg_doms_inconsistent(irg);
632 /* A quiet place, where the old obstack can rest in peace,
633 until it will be cremated. */
634 graveyard_obst = irg->obst;
636 /* A new obstack, where the reachable nodes will be copied to. */
637 rebirth_obst = xmalloc(sizeof(*rebirth_obst));
638 irg->obst = rebirth_obst;
639 obstack_init(irg->obst);
640 irg->last_node_idx = 0;
642 /* We also need a new value table for CSE */
643 del_identities(irg->value_table);
644 irg->value_table = new_identities();
646 /* Copy the graph from the old to the new obstack */
647 copy_graph_env(/*copy_node_nr=*/1);
649 /* Free memory from old unoptimized obstack */
650 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
651 xfree (graveyard_obst); /* ... then free it. */
653 /* inform statistics that the run is over */
654 hook_dead_node_elim(irg, 0);
656 current_ir_graph = rem;
657 set_interprocedural_view(rem_ipview);
662 * Relink bad predecessors of a block and store the old in array to the
663 * link field. This function is called by relink_bad_predecessors().
664 * The array of link field starts with the block operand at position 0.
665 * If block has bad predecessors, create a new in array without bad preds.
666 * Otherwise let in array untouched.
668 static void relink_bad_block_predecessors(ir_node *n, void *env) {
669 ir_node **new_in, *irn;
670 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
673 /* if link field of block is NULL, look for bad predecessors otherwise
674 this is already done */
675 if (get_irn_op(n) == op_Block &&
676 get_irn_link(n) == NULL) {
678 /* save old predecessors in link field (position 0 is the block operand)*/
679 set_irn_link(n, get_irn_in(n));
681 /* count predecessors without bad nodes */
682 old_irn_arity = get_irn_arity(n);
683 for (i = 0; i < old_irn_arity; i++)
684 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
686 /* arity changing: set new predecessors without bad nodes */
687 if (new_irn_arity < old_irn_arity) {
688 /* Get new predecessor array. We do not resize the array, as we must
689 keep the old one to update Phis. */
690 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
692 /* set new predecessors in array */
695 for (i = 0; i < old_irn_arity; i++) {
696 irn = get_irn_n(n, i);
698 new_in[new_irn_n] = irn;
699 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
703 /* ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity); */
704 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
706 } /* ir node has bad predecessors */
707 } /* Block is not relinked */
711 * Relinks Bad predecessors from Blocks and Phis called by walker
712 * remove_bad_predecesors(). If n is a Block, call
713 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
714 * function of Phi's Block. If this block has bad predecessors, relink preds
717 static void relink_bad_predecessors(ir_node *n, void *env) {
718 ir_node *block, **old_in;
719 int i, old_irn_arity, new_irn_arity;
721 /* relink bad predecessors of a block */
722 if (get_irn_op(n) == op_Block)
723 relink_bad_block_predecessors(n, env);
725 /* If Phi node relink its block and its predecessors */
726 if (get_irn_op(n) == op_Phi) {
728 /* Relink predecessors of phi's block */
729 block = get_nodes_block(n);
730 if (get_irn_link(block) == NULL)
731 relink_bad_block_predecessors(block, env);
733 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
734 old_irn_arity = ARR_LEN(old_in);
736 /* Relink Phi predecessors if count of predecessors changed */
737 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
738 /* set new predecessors in array
739 n->in[0] remains the same block */
741 for(i = 1; i < old_irn_arity; i++)
742 if (!is_Bad((ir_node *)old_in[i])) {
743 n->in[new_irn_arity] = n->in[i];
744 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
748 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
749 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
751 } /* n is a Phi node */
755 * Removes Bad Bad predecessors from Blocks and the corresponding
756 * inputs to Phi nodes as in dead_node_elimination but without
758 * On walking up set the link field to NULL, on walking down call
759 * relink_bad_predecessors() (This function stores the old in array
760 * to the link field and sets a new in array if arity of predecessors
763 void remove_bad_predecessors(ir_graph *irg) {
764 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
771 __)|_| | \_/ | \_/(/_ |_/\__|__
773 The following stuff implements a facility that automatically patches
774 registered ir_node pointers to the new node when a dead node elimination occurs.
777 struct _survive_dce_t {
781 hook_entry_t dead_node_elim;
782 hook_entry_t dead_node_elim_subst;
785 typedef struct _survive_dce_list_t {
786 struct _survive_dce_list_t *next;
788 } survive_dce_list_t;
790 static void dead_node_hook(void *context, ir_graph *irg, int start) {
791 survive_dce_t *sd = context;
794 /* Create a new map before the dead node elimination is performed. */
796 sd->new_places = pmap_create_ex(pmap_count(sd->places));
798 /* Patch back all nodes if dead node elimination is over and something is to be done. */
799 pmap_destroy(sd->places);
800 sd->places = sd->new_places;
801 sd->new_places = NULL;
806 * Hook called when dead node elimination replaces old by nw.
808 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw) {
809 survive_dce_t *sd = context;
810 survive_dce_list_t *list = pmap_get(sd->places, old);
813 /* If the node is to be patched back, write the new address to all registered locations. */
815 survive_dce_list_t *p;
817 for (p = list; p; p = p->next)
820 pmap_insert(sd->new_places, nw, list);
825 * Make a new Survive DCE environment.
827 survive_dce_t *new_survive_dce(void) {
828 survive_dce_t *res = xmalloc(sizeof(res[0]));
829 obstack_init(&res->obst);
830 res->places = pmap_create();
831 res->new_places = NULL;
833 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
834 res->dead_node_elim.context = res;
835 res->dead_node_elim.next = NULL;
837 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
838 res->dead_node_elim_subst.context = res;
839 res->dead_node_elim_subst.next = NULL;
841 #ifndef FIRM_ENABLE_HOOKS
842 assert(0 && "need hooks enabled");
845 register_hook(hook_dead_node_elim, &res->dead_node_elim);
846 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
851 * Free a Survive DCE environment.
853 void free_survive_dce(survive_dce_t *sd) {
854 obstack_free(&sd->obst, NULL);
855 pmap_destroy(sd->places);
856 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
857 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
862 * Register a node pointer to be patched upon DCE.
863 * When DCE occurs, the node pointer specified by @p place will be
864 * patched to the new address of the node it is pointing to.
866 * @param sd The Survive DCE environment.
867 * @param place The address of the node pointer.
869 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place) {
870 if (*place != NULL) {
871 ir_node *irn = *place;
872 survive_dce_list_t *curr = pmap_get(sd->places, irn);
873 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw[0]));
878 pmap_insert(sd->places, irn, nw);
882 /*--------------------------------------------------------------------*/
883 /* Functionality for inlining */
884 /*--------------------------------------------------------------------*/
887 * Copy node for inlineing. Updates attributes that change when
888 * inlineing but not for dead node elimination.
890 * Copies the node by calling copy_node() and then updates the entity if
891 * it's a local one. env must be a pointer of the frame type of the
892 * inlined procedure. The new entities must be in the link field of
896 copy_node_inline(ir_node *n, void *env) {
898 ir_type *frame_tp = (ir_type *)env;
901 if (get_irn_op(n) == op_Sel) {
902 nn = get_new_node (n);
904 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
905 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
907 } else if (get_irn_op(n) == op_Block) {
908 nn = get_new_node (n);
909 nn->attr.block.irg = current_ir_graph;
914 * Walker: checks if P_value_arg_base is used.
916 static void find_addr(ir_node *node, void *env) {
917 int *allow_inline = env;
918 if (is_Proj(node) && get_irn_op(get_Proj_pred(node)) == op_Start) {
919 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
925 * Check if we can inline a given call.
926 * Currently, we cannot inline two cases:
927 * - call with compound arguments
928 * - graphs that take the address of a parameter
930 * check these conditions here
932 static int can_inline(ir_node *call, ir_graph *called_graph) {
933 ir_type *call_type = get_Call_type(call);
934 int params, ress, i, res;
935 assert(is_Method_type(call_type));
937 params = get_method_n_params(call_type);
938 ress = get_method_n_ress(call_type);
940 /* check parameters for compound arguments */
941 for (i = 0; i < params; ++i) {
942 ir_type *p_type = get_method_param_type(call_type, i);
944 if (is_compound_type(p_type))
948 /* check results for compound arguments */
949 for (i = 0; i < ress; ++i) {
950 ir_type *r_type = get_method_res_type(call_type, i);
952 if (is_compound_type(r_type))
957 irg_walk_graph(called_graph, find_addr, NULL, &res);
963 exc_handler = 0, /**< There is a handler. */
964 exc_to_end = 1, /**< Branches to End. */
965 exc_no_handler = 2 /**< Exception handling not represented. */
968 /* Inlines a method at the given call site. */
969 int inline_method(ir_node *call, ir_graph *called_graph) {
971 ir_node *post_call, *post_bl;
972 ir_node *in[pn_Start_max];
973 ir_node *end, *end_bl;
977 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
978 enum exc_mode exc_handling;
979 ir_type *called_frame;
980 irg_inline_property prop = get_irg_inline_property(called_graph);
982 if ( (prop < irg_inline_forced) &&
983 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
985 /* Do not inline variadic functions. */
986 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
989 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
990 get_method_n_params(get_Call_type(call)));
993 * currently, we cannot inline two cases:
994 * - call with compound arguments
995 * - graphs that take the address of a parameter
997 if (! can_inline(call, called_graph))
1000 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
1001 rem_opt = get_opt_optimize();
1004 /* Handle graph state */
1005 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1006 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
1007 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
1008 set_irg_outs_inconsistent(current_ir_graph);
1009 set_irg_extblk_inconsistent(current_ir_graph);
1010 set_irg_doms_inconsistent(current_ir_graph);
1011 set_irg_loopinfo_inconsistent(current_ir_graph);
1012 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
1014 /* -- Check preconditions -- */
1015 assert(is_Call(call));
1016 /* @@@ does not work for InterfaceIII.java after cgana
1017 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
1018 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
1019 get_Call_type(call)));
1021 if (called_graph == current_ir_graph) {
1022 set_optimize(rem_opt);
1026 /* here we know we WILL inline, so inform the statistics */
1027 hook_inline(call, called_graph);
1029 /* -- Decide how to handle exception control flow: Is there a handler
1030 for the Call node, or do we branch directly to End on an exception?
1032 0 There is a handler.
1034 2 Exception handling not represented in Firm. -- */
1036 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
1037 for (proj = get_irn_link(call); proj; proj = get_irn_link(proj)) {
1038 long proj_nr = get_Proj_proj(proj);
1039 if (proj_nr == pn_Call_X_except) Xproj = proj;
1040 if (proj_nr == pn_Call_M_except) Mproj = proj;
1042 if (Mproj) { assert(Xproj); exc_handling = exc_handler; } /* Mproj */
1043 else if (Xproj) { exc_handling = exc_to_end; } /* !Mproj && Xproj */
1044 else { exc_handling = exc_no_handler; } /* !Mproj && !Xproj */
1048 the procedure and later replaces the Start node of the called graph.
1049 Post_call is the old Call node and collects the results of the called
1050 graph. Both will end up being a tuple. -- */
1051 post_bl = get_nodes_block(call);
1052 set_irg_current_block(current_ir_graph, post_bl);
1053 /* XxMxPxPxPxT of Start + parameter of Call */
1054 in[pn_Start_X_initial_exec] = new_Jmp();
1055 in[pn_Start_M] = get_Call_mem(call);
1056 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
1057 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
1058 in[pn_Start_P_tls] = get_irg_tls(current_ir_graph);
1059 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
1060 /* in[pn_Start_P_value_arg_base] = ??? */
1061 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
1062 pre_call = new_Tuple(pn_Start_max - 1, in);
1066 The new block gets the ins of the old block, pre_call and all its
1067 predecessors and all Phi nodes. -- */
1068 part_block(pre_call);
1070 /* -- Prepare state for dead node elimination -- */
1071 /* Visited flags in calling irg must be >= flag in called irg.
1072 Else walker and arity computation will not work. */
1073 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
1074 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
1075 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
1076 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
1077 /* Set pre_call as new Start node in link field of the start node of
1078 calling graph and pre_calls block as new block for the start block
1080 Further mark these nodes so that they are not visited by the
1082 set_irn_link(get_irg_start(called_graph), pre_call);
1083 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
1084 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1085 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
1086 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
1087 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
1089 /* Initialize for compaction of in arrays */
1090 inc_irg_block_visited(current_ir_graph);
1092 /* -- Replicate local entities of the called_graph -- */
1093 /* copy the entities. */
1094 called_frame = get_irg_frame_type(called_graph);
1095 for (i = 0; i < get_class_n_members(called_frame); i++) {
1096 ir_entity *new_ent, *old_ent;
1097 old_ent = get_class_member(called_frame, i);
1098 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
1099 set_entity_link(old_ent, new_ent);
1102 /* visited is > than that of called graph. With this trick visited will
1103 remain unchanged so that an outer walker, e.g., searching the call nodes
1104 to inline, calling this inline will not visit the inlined nodes. */
1105 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
1107 /* -- Performing dead node elimination inlines the graph -- */
1108 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1110 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
1111 get_irg_frame_type(called_graph));
1113 /* Repair called_graph */
1114 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
1115 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
1116 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1118 /* -- Merge the end of the inlined procedure with the call site -- */
1119 /* We will turn the old Call node into a Tuple with the following
1122 0: Phi of all Memories of Return statements.
1123 1: Jmp from new Block that merges the control flow from all exception
1124 predecessors of the old end block.
1125 2: Tuple of all arguments.
1126 3: Phi of Exception memories.
1127 In case the old Call directly branches to End on an exception we don't
1128 need the block merging all exceptions nor the Phi of the exception
1132 /* -- Precompute some values -- */
1133 end_bl = get_new_node(get_irg_end_block(called_graph));
1134 end = get_new_node(get_irg_end(called_graph));
1135 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1136 n_res = get_method_n_ress(get_Call_type(call));
1138 res_pred = xmalloc(n_res * sizeof(*res_pred));
1139 cf_pred = xmalloc(arity * sizeof(*res_pred));
1141 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
1143 /* -- archive keepalives -- */
1144 irn_arity = get_irn_arity(end);
1145 for (i = 0; i < irn_arity; i++) {
1146 ir_node *ka = get_End_keepalive(end, i);
1148 add_End_keepalive(get_irg_end(current_ir_graph), ka);
1151 /* The new end node will die. We need not free as the in array is on the obstack:
1152 copy_node() only generated 'D' arrays. */
1154 /* -- Replace Return nodes by Jump nodes. -- */
1156 for (i = 0; i < arity; i++) {
1158 ret = get_irn_n(end_bl, i);
1159 if (is_Return(ret)) {
1160 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
1164 set_irn_in(post_bl, n_ret, cf_pred);
1166 /* -- Build a Tuple for all results of the method.
1167 Add Phi node if there was more than one Return. -- */
1168 turn_into_tuple(post_call, pn_Call_max);
1169 /* First the Memory-Phi */
1171 for (i = 0; i < arity; i++) {
1172 ret = get_irn_n(end_bl, i);
1173 if (is_Return(ret)) {
1174 cf_pred[n_ret] = get_Return_mem(ret);
1178 phi = new_Phi(n_ret, cf_pred, mode_M);
1179 set_Tuple_pred(call, pn_Call_M_regular, phi);
1180 /* Conserve Phi-list for further inlinings -- but might be optimized */
1181 if (get_nodes_block(phi) == post_bl) {
1182 set_irn_link(phi, get_irn_link(post_bl));
1183 set_irn_link(post_bl, phi);
1185 /* Now the real results */
1187 for (j = 0; j < n_res; j++) {
1189 for (i = 0; i < arity; i++) {
1190 ret = get_irn_n(end_bl, i);
1191 if (get_irn_op(ret) == op_Return) {
1192 cf_pred[n_ret] = get_Return_res(ret, j);
1197 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1201 /* Conserve Phi-list for further inlinings -- but might be optimized */
1202 if (get_nodes_block(phi) == post_bl) {
1203 set_irn_link(phi, get_irn_link(post_bl));
1204 set_irn_link(post_bl, phi);
1207 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1209 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1212 /* For now, we cannot inline calls with value_base */
1213 set_Tuple_pred(call, pn_Call_P_value_res_base, new_Bad());
1215 /* Finally the exception control flow.
1216 We have two (three) possible situations:
1217 First if the Call branches to an exception handler: We need to add a Phi node to
1218 collect the memory containing the exception objects. Further we need
1219 to add another block to get a correct representation of this Phi. To
1220 this block we add a Jmp that resolves into the X output of the Call
1221 when the Call is turned into a tuple.
1222 Second the Call branches to End, the exception is not handled. Just
1223 add all inlined exception branches to the End node.
1224 Third: there is no Exception edge at all. Handle as case two. */
1225 if (exc_handling == exc_handler) {
1227 for (i = 0; i < arity; i++) {
1229 ret = get_irn_n(end_bl, i);
1230 irn = skip_Proj(ret);
1231 if (is_fragile_op(irn) || (get_irn_op(irn) == op_Raise)) {
1232 cf_pred[n_exc] = ret;
1237 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1238 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1239 /* The Phi for the memories with the exception objects */
1241 for (i = 0; i < arity; i++) {
1243 ret = skip_Proj(get_irn_n(end_bl, i));
1245 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1247 } else if (is_fragile_op(ret)) {
1248 /* We rely that all cfops have the memory output at the same position. */
1249 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1251 } else if (get_irn_op(ret) == op_Raise) {
1252 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1256 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1258 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1259 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1261 set_Tuple_pred(call, pn_Call_X_regular, new_Bad());
1263 ir_node *main_end_bl;
1264 int main_end_bl_arity;
1265 ir_node **end_preds;
1267 /* assert(exc_handling == 1 || no exceptions. ) */
1269 for (i = 0; i < arity; i++) {
1270 ir_node *ret = get_irn_n(end_bl, i);
1271 ir_node *irn = skip_Proj(ret);
1273 if (is_fragile_op(irn) || (get_irn_op(irn) == op_Raise)) {
1274 cf_pred[n_exc] = ret;
1278 main_end_bl = get_irg_end_block(current_ir_graph);
1279 main_end_bl_arity = get_irn_arity(main_end_bl);
1280 end_preds = xmalloc((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1282 for (i = 0; i < main_end_bl_arity; ++i)
1283 end_preds[i] = get_irn_n(main_end_bl, i);
1284 for (i = 0; i < n_exc; ++i)
1285 end_preds[main_end_bl_arity + i] = cf_pred[i];
1286 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1287 set_Tuple_pred(call, pn_Call_X_regular, new_Bad());
1288 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1289 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1295 /* -- Turn CSE back on. -- */
1296 set_optimize(rem_opt);
1301 /********************************************************************/
1302 /* Apply inlineing to small methods. */
1303 /********************************************************************/
1305 /** Represents a possible inlinable call in a graph. */
1306 typedef struct _call_entry call_entry;
1307 struct _call_entry {
1308 ir_node *call; /**< the Call */
1309 ir_graph *callee; /**< the callee called here */
1310 call_entry *next; /**< for linking the next one */
1314 * environment for inlining small irgs
1316 typedef struct _inline_env_t {
1317 struct obstack obst; /**< an obstack where call_entries are allocated on. */
1318 call_entry *head; /**< the head of the call entry list */
1319 call_entry *tail; /**< the tail of the call entry list */
1323 * Returns the irg called from a Call node. If the irg is not
1324 * known, NULL is returned.
1326 static ir_graph *get_call_called_irg(ir_node *call) {
1328 ir_graph *called_irg = NULL;
1330 addr = get_Call_ptr(call);
1331 if (is_SymConst(addr) && get_SymConst_kind(addr) == symconst_addr_ent) {
1332 called_irg = get_entity_irg(get_SymConst_entity(addr));
1339 * Walker: Collect all calls to known graphs inside a graph.
1341 static void collect_calls(ir_node *call, void *env) {
1342 if (is_Call(call)) {
1343 ir_graph *called_irg = get_call_called_irg(call);
1345 /* The Call node calls a locally defined method. Remember to inline. */
1346 inline_env_t *ienv = env;
1347 call_entry *entry = obstack_alloc(&ienv->obst, sizeof(*entry));
1349 entry->callee = called_irg;
1352 if (ienv->tail == NULL)
1355 ienv->tail->next = entry;
1362 * Inlines all small methods at call sites where the called address comes
1363 * from a Const node that references the entity representing the called
1365 * The size argument is a rough measure for the code size of the method:
1366 * Methods where the obstack containing the firm graph is smaller than
1369 void inline_small_irgs(ir_graph *irg, int size) {
1370 ir_graph *rem = current_ir_graph;
1373 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1375 if (!(get_opt_optimize() && get_opt_inline())) return;
1377 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1379 current_ir_graph = irg;
1380 /* Handle graph state */
1381 assert(get_irg_phase_state(irg) != phase_building);
1382 free_callee_info(irg);
1384 /* Find Call nodes to inline.
1385 (We can not inline during a walk of the graph, as inlineing the same
1386 method several times changes the visited flag of the walked graph:
1387 after the first inlineing visited of the callee equals visited of
1388 the caller. With the next inlineing both are increased.) */
1389 obstack_init(&env.obst);
1390 env.head = env.tail = NULL;
1391 irg_walk_graph(irg, NULL, collect_calls, &env);
1393 if (env.head != NULL) {
1394 /* There are calls to inline */
1395 collect_phiprojs(irg);
1396 for (entry = env.head; entry != NULL; entry = entry->next) {
1397 ir_graph *callee = entry->callee;
1398 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1399 (get_irg_inline_property(callee) >= irg_inline_forced)) {
1400 inline_method(entry->call, callee);
1404 obstack_free(&env.obst, NULL);
1405 current_ir_graph = rem;
1409 * Environment for inlining irgs.
1412 int n_nodes; /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
1413 int n_nodes_orig; /**< for statistics */
1414 call_entry *call_head; /**< The head of the list of all call nodes in this graph. */
1415 call_entry *call_tail; /**< The tail of the list of all call nodes in this graph .*/
1416 int n_call_nodes; /**< Number of Call nodes in the graph. */
1417 int n_call_nodes_orig; /**< for statistics */
1418 int n_callers; /**< Number of known graphs that call this graphs. */
1419 int n_callers_orig; /**< for statistics */
1420 int got_inline; /**< Set, if at leat one call inside this graph was inlined. */
1424 * Allocate a new environment for inlining.
1426 static inline_irg_env *alloc_inline_irg_env(struct obstack *obst) {
1427 inline_irg_env *env = obstack_alloc(obst, sizeof(*env));
1428 env->n_nodes = -2; /* do not count count Start, End */
1429 env->n_nodes_orig = -2; /* do not count Start, End */
1430 env->call_head = NULL;
1431 env->call_tail = NULL;
1432 env->n_call_nodes = 0;
1433 env->n_call_nodes_orig = 0;
1435 env->n_callers_orig = 0;
1436 env->got_inline = 0;
1440 typedef struct walker_env {
1441 struct obstack *obst; /**< the obstack for allocations. */
1442 inline_irg_env *x; /**< the inline environment */
1443 int ignore_runtime; /**< the ignore runtime flag */
1447 * post-walker: collect all calls in the inline-environment
1448 * of a graph and sum some statistics.
1450 static void collect_calls2(ir_node *call, void *ctx) {
1452 inline_irg_env *x = env->x;
1453 ir_op *op = get_irn_op(call);
1457 /* count meaningful nodes in irg */
1458 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1463 if (op != op_Call) return;
1465 /* check, if it's a runtime call */
1466 if (env->ignore_runtime) {
1467 ir_node *symc = get_Call_ptr(call);
1469 if (is_SymConst(symc) && get_SymConst_kind(symc) == symconst_addr_ent) {
1470 ir_entity *ent = get_SymConst_entity(symc);
1472 if (get_entity_additional_properties(ent) & mtp_property_runtime)
1477 /* collect all call nodes */
1479 ++x->n_call_nodes_orig;
1481 callee = get_call_called_irg(call);
1483 inline_irg_env *callee_env = get_irg_link(callee);
1484 /* count all static callers */
1485 ++callee_env->n_callers;
1486 ++callee_env->n_callers_orig;
1488 /* link it in the list of possible inlinable entries */
1489 entry = obstack_alloc(env->obst, sizeof(*entry));
1491 entry->callee = callee;
1493 if (x->call_tail == NULL)
1494 x->call_head = entry;
1496 x->call_tail->next = entry;
1497 x->call_tail = entry;
1502 * Returns TRUE if the number of callers in 0 in the irg's environment,
1503 * hence this irg is a leave.
1505 INLINE static int is_leave(ir_graph *irg) {
1506 inline_irg_env *env = get_irg_link(irg);
1507 return env->n_call_nodes == 0;
1511 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1513 INLINE static int is_smaller(ir_graph *callee, int size) {
1514 inline_irg_env *env = get_irg_link(callee);
1515 return env->n_nodes < size;
1519 * Append the nodes of the list src to the nodes of the list in environment dst.
1521 static void append_call_list(struct obstack *obst, inline_irg_env *dst, call_entry *src) {
1522 call_entry *entry, *nentry;
1524 /* Note that the src list points to Call nodes in the inlined graph, but
1525 we need Call nodes in our graph. Luckily the inliner leaves this information
1526 in the link field. */
1527 for (entry = src; entry != NULL; entry = entry->next) {
1528 nentry = obstack_alloc(obst, sizeof(*nentry));
1529 nentry->call = get_irn_link(entry->call);
1530 nentry->callee = entry->callee;
1531 nentry->next = NULL;
1532 dst->call_tail->next = nentry;
1533 dst->call_tail = nentry;
1538 * Inlines small leave methods at call sites where the called address comes
1539 * from a Const node that references the entity representing the called
1541 * The size argument is a rough measure for the code size of the method:
1542 * Methods where the obstack containing the firm graph is smaller than
1545 void inline_leave_functions(int maxsize, int leavesize, int size, int ignore_runtime) {
1546 inline_irg_env *env;
1552 call_entry *entry, *tail;
1553 const call_entry *centry;
1554 struct obstack obst;
1555 DEBUG_ONLY(firm_dbg_module_t *dbg;)
1557 if (!(get_opt_optimize() && get_opt_inline())) return;
1559 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1560 rem = current_ir_graph;
1561 obstack_init(&obst);
1563 /* extend all irgs by a temporary data structure for inlining. */
1564 n_irgs = get_irp_n_irgs();
1565 for (i = 0; i < n_irgs; ++i)
1566 set_irg_link(get_irp_irg(i), alloc_inline_irg_env(&obst));
1568 /* Precompute information in temporary data structure. */
1570 wenv.ignore_runtime = ignore_runtime;
1571 for (i = 0; i < n_irgs; ++i) {
1572 ir_graph *irg = get_irp_irg(i);
1574 assert(get_irg_phase_state(irg) != phase_building);
1575 free_callee_info(irg);
1577 wenv.x = get_irg_link(irg);
1578 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1581 /* -- and now inline. -- */
1583 /* Inline leaves recursively -- we might construct new leaves. */
1587 for (i = 0; i < n_irgs; ++i) {
1589 int phiproj_computed = 0;
1591 current_ir_graph = get_irp_irg(i);
1592 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1595 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1598 if (env->n_nodes > maxsize) break;
1601 callee = entry->callee;
1603 if (is_leave(callee) && is_smaller(callee, leavesize)) {
1604 if (!phiproj_computed) {
1605 phiproj_computed = 1;
1606 collect_phiprojs(current_ir_graph);
1608 did_inline = inline_method(call, callee);
1611 /* Do some statistics */
1612 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1614 env->got_inline = 1;
1615 --env->n_call_nodes;
1616 env->n_nodes += callee_env->n_nodes;
1617 --callee_env->n_callers;
1619 /* remove this call from the list */
1621 tail->next = entry->next;
1623 env->call_head = entry->next;
1629 env->call_tail = tail;
1631 } while (did_inline);
1633 /* inline other small functions. */
1634 for (i = 0; i < n_irgs; ++i) {
1636 int phiproj_computed = 0;
1638 current_ir_graph = get_irp_irg(i);
1639 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1641 /* note that the list of possible calls is updated during the process */
1643 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1647 callee = entry->callee;
1649 if (((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1650 (get_irg_inline_property(callee) >= irg_inline_forced))) {
1651 if (!phiproj_computed) {
1652 phiproj_computed = 1;
1653 collect_phiprojs(current_ir_graph);
1655 if (inline_method(call, callee)) {
1656 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1658 /* callee was inline. Append it's call list. */
1659 env->got_inline = 1;
1660 --env->n_call_nodes;
1661 append_call_list(&obst, env, callee_env->call_head);
1662 env->n_call_nodes += callee_env->n_call_nodes;
1663 env->n_nodes += callee_env->n_nodes;
1664 --callee_env->n_callers;
1666 /* after we have inlined callee, all called methods inside callee
1667 are now called once more */
1668 for (centry = callee_env->call_head; centry != NULL; centry = centry->next) {
1669 inline_irg_env *penv = get_irg_link(centry->callee);
1673 /* remove this call from the list */
1675 tail->next = entry->next;
1677 env->call_head = entry->next;
1683 env->call_tail = tail;
1686 for (i = 0; i < n_irgs; ++i) {
1687 irg = get_irp_irg(i);
1688 env = (inline_irg_env *)get_irg_link(irg);
1690 if (env->got_inline) {
1691 /* this irg got calls inlined */
1692 set_irg_outs_inconsistent(irg);
1693 set_irg_doms_inconsistent(irg);
1695 optimize_graph_df(irg);
1698 if (env->got_inline || (env->n_callers_orig != env->n_callers))
1699 DB((dbg, SET_LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1700 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1701 env->n_callers_orig, env->n_callers,
1702 get_entity_name(get_irg_entity(irg))));
1705 obstack_free(&obst, NULL);
1706 current_ir_graph = rem;
1709 /*******************************************************************/
1710 /* Code Placement. Pins all floating nodes to a block where they */
1711 /* will be executed only if needed. */
1712 /*******************************************************************/
1715 * Returns non-zero, is a block is not reachable from Start.
1717 * @param block the block to test
1720 is_Block_unreachable(ir_node *block) {
1721 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1725 * Find the earliest correct block for node n. --- Place n into the
1726 * same Block as its dominance-deepest Input.
1728 * We have to avoid calls to get_nodes_block() here
1729 * because the graph is floating.
1731 * move_out_of_loops() expects that place_floats_early() have placed
1732 * all "living" nodes into a living block. That's why we must
1733 * move nodes in dead block with "live" successors into a valid
1735 * We move them just into the same block as it's successor (or
1736 * in case of a Phi into the effective use block). For Phi successors,
1737 * this may still be a dead block, but then there is no real use, as
1738 * the control flow will be dead later.
1740 * @param n the node to be placed
1741 * @param worklist a worklist, predecessors of non-floating nodes are placed here
1744 place_floats_early(ir_node *n, waitq *worklist) {
1747 /* we must not run into an infinite loop */
1748 assert(irn_not_visited(n));
1749 mark_irn_visited(n);
1751 #ifndef CAN_PLACE_PROJS
1752 while (is_Proj(n)) {
1753 n = get_Proj_pred(n);
1754 mark_irn_visited(n);
1758 /* Place floating nodes. */
1759 if (get_irn_pinned(n) == op_pin_state_floats) {
1760 ir_node *curr_block = get_irn_n(n, -1);
1761 int in_dead_block = is_Block_unreachable(curr_block);
1763 ir_node *b = NULL; /* The block to place this node in */
1765 assert(is_no_Block(n));
1767 if (is_irn_start_block_placed(n)) {
1768 /* These nodes will not be placed by the loop below. */
1769 b = get_irg_start_block(current_ir_graph);
1773 /* find the block for this node. */
1774 irn_arity = get_irn_arity(n);
1775 for (i = 0; i < irn_arity; i++) {
1776 ir_node *pred = get_irn_n(n, i);
1777 ir_node *pred_block;
1779 if ((irn_not_visited(pred))
1780 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1783 * If the current node is NOT in a dead block, but one of its
1784 * predecessors is, we must move the predecessor to a live block.
1785 * Such thing can happen, if global CSE chose a node from a dead block.
1786 * We move it simply to our block.
1787 * Note that neither Phi nor End nodes are floating, so we don't
1788 * need to handle them here.
1790 if (! in_dead_block) {
1791 if (get_irn_pinned(pred) == op_pin_state_floats &&
1792 is_Block_unreachable(get_irn_n(pred, -1)))
1793 set_nodes_block(pred, curr_block);
1795 place_floats_early(pred, worklist);
1799 * A node in the Bad block must stay in the bad block,
1800 * so don't compute a new block for it.
1805 /* Because all loops contain at least one op_pin_state_pinned node, now all
1806 our inputs are either op_pin_state_pinned or place_early() has already
1807 been finished on them. We do not have any unfinished inputs! */
1808 pred_block = get_irn_n(pred, -1);
1809 if ((!is_Block_dead(pred_block)) &&
1810 (get_Block_dom_depth(pred_block) > depth)) {
1812 depth = get_Block_dom_depth(pred_block);
1814 /* Avoid that the node is placed in the Start block */
1815 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)
1816 && get_irg_phase_state(current_ir_graph) != phase_backend) {
1817 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1818 assert(b != get_irg_start_block(current_ir_graph));
1823 set_nodes_block(n, b);
1827 * Add predecessors of non floating nodes and non-floating predecessors
1828 * of floating nodes to worklist and fix their blocks if the are in dead block.
1830 irn_arity = get_irn_arity(n);
1832 if (get_irn_op(n) == op_End) {
1834 * Simplest case: End node. Predecessors are keep-alives,
1835 * no need to move out of dead block.
1837 for (i = -1; i < irn_arity; ++i) {
1838 ir_node *pred = get_irn_n(n, i);
1839 if (irn_not_visited(pred))
1840 waitq_put(worklist, pred);
1842 } else if (is_Block(n)) {
1844 * Blocks: Predecessors are control flow, no need to move
1845 * them out of dead block.
1847 for (i = irn_arity - 1; i >= 0; --i) {
1848 ir_node *pred = get_irn_n(n, i);
1849 if (irn_not_visited(pred))
1850 waitq_put(worklist, pred);
1852 } else if (is_Phi(n)) {
1854 ir_node *curr_block = get_irn_n(n, -1);
1855 int in_dead_block = is_Block_unreachable(curr_block);
1858 * Phi nodes: move nodes from dead blocks into the effective use
1859 * of the Phi-input if the Phi is not in a bad block.
1861 pred = get_irn_n(n, -1);
1862 if (irn_not_visited(pred))
1863 waitq_put(worklist, pred);
1865 for (i = irn_arity - 1; i >= 0; --i) {
1866 ir_node *pred = get_irn_n(n, i);
1868 if (irn_not_visited(pred)) {
1869 if (! in_dead_block &&
1870 get_irn_pinned(pred) == op_pin_state_floats &&
1871 is_Block_unreachable(get_irn_n(pred, -1))) {
1872 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1874 waitq_put(worklist, pred);
1879 ir_node *curr_block = get_irn_n(n, -1);
1880 int in_dead_block = is_Block_unreachable(curr_block);
1883 * All other nodes: move nodes from dead blocks into the same block.
1885 pred = get_irn_n(n, -1);
1886 if (irn_not_visited(pred))
1887 waitq_put(worklist, pred);
1889 for (i = irn_arity - 1; i >= 0; --i) {
1890 ir_node *pred = get_irn_n(n, i);
1892 if (irn_not_visited(pred)) {
1893 if (! in_dead_block &&
1894 get_irn_pinned(pred) == op_pin_state_floats &&
1895 is_Block_unreachable(get_irn_n(pred, -1))) {
1896 set_nodes_block(pred, curr_block);
1898 waitq_put(worklist, pred);
1905 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1906 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1907 * places all floating nodes reachable from its argument through floating
1908 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1910 * @param worklist a worklist, used for the algorithm, empty on in/output
1912 static void place_early(waitq *worklist) {
1914 inc_irg_visited(current_ir_graph);
1916 /* this inits the worklist */
1917 place_floats_early(get_irg_end(current_ir_graph), worklist);
1919 /* Work the content of the worklist. */
1920 while (!waitq_empty(worklist)) {
1921 ir_node *n = waitq_get(worklist);
1922 if (irn_not_visited(n))
1923 place_floats_early(n, worklist);
1926 set_irg_outs_inconsistent(current_ir_graph);
1927 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
1931 * Compute the deepest common ancestor of block and dca.
1933 static ir_node *calc_dca(ir_node *dca, ir_node *block) {
1936 /* we do not want to place nodes in dead blocks */
1937 if (is_Block_dead(block))
1940 /* We found a first legal placement. */
1941 if (!dca) return block;
1943 /* Find a placement that is dominates both, dca and block. */
1944 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1945 block = get_Block_idom(block);
1947 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1948 dca = get_Block_idom(dca);
1951 while (block != dca) {
1952 block = get_Block_idom(block); dca = get_Block_idom(dca);
1958 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1959 * I.e., DCA is the block where we might place PRODUCER.
1960 * A data flow edge points from producer to consumer.
1963 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer) {
1964 ir_node *block = NULL;
1966 /* Compute the latest block into which we can place a node so that it is
1968 if (get_irn_op(consumer) == op_Phi) {
1969 /* our consumer is a Phi-node, the effective use is in all those
1970 blocks through which the Phi-node reaches producer */
1972 ir_node *phi_block = get_nodes_block(consumer);
1973 irn_arity = get_irn_arity(consumer);
1975 for (i = 0; i < irn_arity; i++) {
1976 if (get_irn_n(consumer, i) == producer) {
1977 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1979 if (! is_Block_unreachable(new_block))
1980 block = calc_dca(block, new_block);
1985 block = get_irn_n(producer, -1);
1987 assert(is_no_Block(consumer));
1988 block = get_nodes_block(consumer);
1991 /* Compute the deepest common ancestor of block and dca. */
1992 return calc_dca(dca, block);
1995 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1997 static INLINE int get_irn_loop_depth(ir_node *n) {
1998 return get_loop_depth(get_irn_loop(n));
2002 * Move n to a block with less loop depth than it's current block. The
2003 * new block must be dominated by early.
2005 * @param n the node that should be moved
2006 * @param early the earliest block we can n move to
2008 static void move_out_of_loops(ir_node *n, ir_node *early) {
2009 ir_node *best, *dca;
2013 /* Find the region deepest in the dominator tree dominating
2014 dca with the least loop nesting depth, but still dominated
2015 by our early placement. */
2016 dca = get_nodes_block(n);
2019 while (dca != early) {
2020 dca = get_Block_idom(dca);
2021 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
2022 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
2026 if (best != get_nodes_block(n)) {
2028 printf("Moving out of loop: "); DDMN(n);
2029 printf(" Outermost block: "); DDMN(early);
2030 printf(" Best block: "); DDMN(best);
2031 printf(" Innermost block: "); DDMN(get_nodes_block(n));
2033 set_nodes_block(n, best);
2037 /* deepest common ancestor in the dominator tree of all nodes'
2038 blocks depending on us; our final placement has to dominate DCA. */
2039 static ir_node *get_deepest_common_ancestor(ir_node *node, ir_node *dca)
2043 for (i = get_irn_n_outs(node) - 1; i >= 0; --i) {
2044 ir_node *succ = get_irn_out(node, i);
2049 * This consumer is the End node, a keep alive edge.
2050 * This is not a real consumer, so we ignore it
2056 dca = get_deepest_common_ancestor(succ, dca);
2058 /* ignore if succ is in dead code */
2059 succ_blk = get_irn_n(succ, -1);
2060 if (is_Block_unreachable(succ_blk))
2062 dca = consumer_dom_dca(dca, succ, node);
2069 #ifdef CAN_PLACE_PROJS
2070 static void set_projs_block(ir_node *node, ir_node *block)
2074 for (i = get_irn_n_outs(node) - 1; i >= 0; --i) {
2075 ir_node *succ = get_irn_out(node, i);
2077 assert(is_Proj(succ));
2079 if(get_irn_mode(succ) == mode_T) {
2080 set_projs_block(succ, block);
2082 set_nodes_block(succ, block);
2088 * Find the latest legal block for N and place N into the
2089 * `optimal' Block between the latest and earliest legal block.
2090 * The `optimal' block is the dominance-deepest block of those
2091 * with the least loop-nesting-depth. This places N out of as many
2092 * loops as possible and then makes it as control dependent as
2095 * @param n the node to be placed
2096 * @param worklist a worklist, all successors of non-floating nodes are
2099 static void place_floats_late(ir_node *n, pdeq *worklist) {
2103 assert(irn_not_visited(n)); /* no multiple placement */
2105 mark_irn_visited(n);
2107 /* no need to place block nodes, control nodes are already placed. */
2108 if ((get_irn_op(n) != op_Block) &&
2110 (get_irn_mode(n) != mode_X)) {
2111 /* Remember the early_blk placement of this block to move it
2112 out of loop no further than the early_blk placement. */
2113 early_blk = get_irn_n(n, -1);
2116 * BEWARE: Here we also get code, that is live, but
2117 * was in a dead block. If the node is life, but because
2118 * of CSE in a dead block, we still might need it.
2121 /* Assure that our users are all placed, except the Phi-nodes.
2122 --- Each data flow cycle contains at least one Phi-node. We
2123 have to break the `user has to be placed before the
2124 producer' dependence cycle and the Phi-nodes are the
2125 place to do so, because we need to base our placement on the
2126 final region of our users, which is OK with Phi-nodes, as they
2127 are op_pin_state_pinned, and they never have to be placed after a
2128 producer of one of their inputs in the same block anyway. */
2129 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
2130 ir_node *succ = get_irn_out(n, i);
2131 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
2132 place_floats_late(succ, worklist);
2135 if (! is_Block_dead(early_blk)) {
2136 /* do only move things that where not dead */
2137 ir_op *op = get_irn_op(n);
2139 /* We have to determine the final block of this node... except for
2140 constants and Projs */
2141 if ((get_irn_pinned(n) == op_pin_state_floats) &&
2143 (op != op_SymConst) &&
2146 /* deepest common ancestor in the dominator tree of all nodes'
2147 blocks depending on us; our final placement has to dominate
2149 ir_node *dca = get_deepest_common_ancestor(n, NULL);
2151 set_nodes_block(n, dca);
2152 move_out_of_loops(n, early_blk);
2153 #ifdef CAN_PLACE_PROJS
2154 if(get_irn_mode(n) == mode_T) {
2155 set_projs_block(n, get_nodes_block(n));
2163 /* Add successors of all non-floating nodes on list. (Those of floating
2164 nodes are placed already and therefore are marked.) */
2165 for (i = 0; i < get_irn_n_outs(n); i++) {
2166 ir_node *succ = get_irn_out(n, i);
2167 if (irn_not_visited(get_irn_out(n, i))) {
2168 pdeq_putr(worklist, succ);
2174 * Place floating nodes on the given worklist as late as possible using
2175 * the dominance tree.
2177 * @param worklist the worklist containing the nodes to place
2179 static void place_late(waitq *worklist) {
2181 inc_irg_visited(current_ir_graph);
2183 /* This fills the worklist initially. */
2184 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
2186 /* And now empty the worklist again... */
2187 while (!waitq_empty(worklist)) {
2188 ir_node *n = waitq_get(worklist);
2189 if (irn_not_visited(n))
2190 place_floats_late(n, worklist);
2194 /* Code Placement. */
2195 void place_code(ir_graph *irg) {
2197 ir_graph *rem = current_ir_graph;
2199 current_ir_graph = irg;
2201 /* Handle graph state */
2202 assert(get_irg_phase_state(irg) != phase_building);
2205 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
2206 free_loop_information(irg);
2207 construct_backedges(irg);
2210 /* Place all floating nodes as early as possible. This guarantees
2211 a legal code placement. */
2212 worklist = new_waitq();
2213 place_early(worklist);
2215 /* place_early() invalidates the outs, place_late needs them. */
2216 compute_irg_outs(irg);
2218 /* Now move the nodes down in the dominator tree. This reduces the
2219 unnecessary executions of the node. */
2220 place_late(worklist);
2222 set_irg_outs_inconsistent(current_ir_graph);
2223 set_irg_loopinfo_inconsistent(current_ir_graph);
2224 del_waitq(worklist);
2225 current_ir_graph = rem;
2229 * Called by walker of remove_critical_cf_edges().
2231 * Place an empty block to an edge between a blocks of multiple
2232 * predecessors and a block of multiple successors.
2235 * @param env Environment of walker. The changed field.
2237 static void walk_critical_cf_edges(ir_node *n, void *env) {
2239 ir_node *pre, *block, *jmp;
2241 ir_graph *irg = get_irn_irg(n);
2243 /* Block has multiple predecessors */
2244 arity = get_irn_arity(n);
2246 if (n == get_irg_end_block(irg))
2247 return; /* No use to add a block here. */
2249 for (i = 0; i < arity; ++i) {
2252 pre = get_irn_n(n, i);
2253 cfop = get_irn_op(skip_Proj(pre));
2254 /* Predecessor has multiple successors. Insert new control flow edge but
2255 ignore exception edges. */
2256 if (! is_op_fragile(cfop) && is_op_forking(cfop)) {
2257 /* set predecessor of new block */
2258 block = new_r_Block(irg, 1, &pre);
2259 /* insert new jmp node to new block */
2260 jmp = new_r_Jmp(irg, block);
2261 /* set successor of new block */
2262 set_irn_n(n, i, jmp);
2264 } /* predecessor has multiple successors */
2265 } /* for all predecessors */
2266 } /* n is a multi-entry block */
2269 void remove_critical_cf_edges(ir_graph *irg) {
2272 irg_block_walk_graph(irg, NULL, walk_critical_cf_edges, &changed);
2274 /* control flow changed */
2275 set_irg_outs_inconsistent(irg);
2276 set_irg_extblk_inconsistent(irg);
2277 set_irg_doms_inconsistent(irg);
2278 set_irg_loopinfo_inconsistent(irg);