3 * File name: ir/ir/irgopt.c
4 * Purpose: Optimizations for a whole ir graph, i.e., a procedure.
5 * Author: Christian Schaefer, Goetz Lindenmaier
6 * Modified by: Sebastian Felis
9 * Copyright: (c) 1998-2003 Universität Karlsruhe
10 * Licence: This file protected by GPL - GNU GENERAL PUBLIC LICENSE.
21 #include "irgraph_t.h"
34 #include "pdeq.h" /* Fuer code placement */
39 #include "irbackedge_t.h"
46 #include "iredges_t.h"
49 /*------------------------------------------------------------------*/
50 /* apply optimizations of iropt to all nodes. */
51 /*------------------------------------------------------------------*/
54 * A wrapper around optimize_inplace_2() to be called from a walker.
56 static void optimize_in_place_wrapper (ir_node *n, void *env) {
57 ir_node *optimized = optimize_in_place_2(n);
58 if (optimized != n) exchange (n, optimized);
62 * Do local optimizations for a node.
64 * @param n the IR-node where to start. Typically the End node
67 * @note current_ir_graph must be set
69 static INLINE void do_local_optimize(ir_node *n) {
70 /* Handle graph state */
71 assert(get_irg_phase_state(current_ir_graph) != phase_building);
73 if (get_opt_global_cse())
74 set_irg_pinned(current_ir_graph, op_pin_state_floats);
75 set_irg_outs_inconsistent(current_ir_graph);
76 set_irg_doms_inconsistent(current_ir_graph);
77 set_irg_loopinfo_inconsistent(current_ir_graph);
79 /* Clean the value_table in irg for the CSE. */
80 del_identities(current_ir_graph->value_table);
81 current_ir_graph->value_table = new_identities();
83 /* walk over the graph */
84 irg_walk(n, firm_clear_link, optimize_in_place_wrapper, NULL);
87 /* Applies local optimizations (see iropt.h) to all nodes reachable from node n */
88 void local_optimize_node(ir_node *n) {
89 ir_graph *rem = current_ir_graph;
90 current_ir_graph = get_irn_irg(n);
94 current_ir_graph = rem;
98 * Block-Walker: uses dominance depth to mark dead blocks.
100 static void kill_dead_blocks(ir_node *block, void *env)
102 if (get_Block_dom_depth(block) < 0) {
104 * Note that the new dominance code correctly handles
105 * the End block, i.e. it is always reachable from Start
107 set_Block_dead(block);
112 local_optimize_graph (ir_graph *irg) {
113 ir_graph *rem = current_ir_graph;
114 current_ir_graph = irg;
116 if (get_irg_dom_state(irg) == dom_consistent)
117 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
119 do_local_optimize(get_irg_end(irg));
121 current_ir_graph = rem;
125 * Data flow optimization walker.
127 static void opt_walker(ir_node *n, void *env) {
131 optimized = optimize_in_place_2(n);
132 set_irn_link(optimized, NULL);
134 if (optimized != n) {
135 const ir_edge_t *edge;
137 foreach_out_edge(n, edge) {
138 ir_node *succ = get_edge_src_irn(edge);
140 if (get_irn_link(succ) != waitq) {
141 pdeq_putr(waitq, succ);
142 set_irn_link(succ, waitq);
145 exchange(n, optimized);
149 void optimize_graph_df(ir_graph *irg) {
150 pdeq *waitq = new_pdeq();
151 int state = edges_activated(irg);
152 ir_graph *rem = current_ir_graph;
154 current_ir_graph = irg;
159 if (get_opt_global_cse())
160 set_irg_pinned(current_ir_graph, op_pin_state_floats);
162 /* Clean the value_table in irg for the CSE. */
163 del_identities(irg->value_table);
164 irg->value_table = new_identities();
166 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
167 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
169 /* invalidate info */
170 set_irg_outs_inconsistent(irg);
171 set_irg_doms_inconsistent(irg);
172 set_irg_loopinfo_inconsistent(irg);
174 /* walk over the graph */
175 irg_walk_graph(irg, NULL, opt_walker, waitq);
177 /* finish the wait queue */
178 while (! pdeq_empty(waitq)) {
179 ir_node *n = pdeq_getl(waitq);
181 opt_walker(n, waitq);
187 edges_deactivate(irg);
189 current_ir_graph = rem;
193 /*------------------------------------------------------------------*/
194 /* Routines for dead node elimination / copying garbage collection */
195 /* of the obstack. */
196 /*------------------------------------------------------------------*/
199 * Remember the new node in the old node by using a field all nodes have.
201 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
204 * Get this new node, before the old node is forgotten.
206 #define get_new_node(oldn) get_irn_link(oldn)
209 * Check if a new node was set.
211 #define has_new_node(n) (get_new_node(n) != NULL)
214 * We use the block_visited flag to mark that we have computed the
215 * number of useful predecessors for this block.
216 * Further we encode the new arity in this flag in the old blocks.
217 * Remembering the arity is useful, as it saves a lot of pointer
218 * accesses. This function is called for all Phi and Block nodes
222 compute_new_arity(ir_node *b) {
223 int i, res, irn_arity;
226 irg_v = get_irg_block_visited(current_ir_graph);
227 block_v = get_Block_block_visited(b);
228 if (block_v >= irg_v) {
229 /* we computed the number of preds for this block and saved it in the
231 return block_v - irg_v;
233 /* compute the number of good predecessors */
234 res = irn_arity = get_irn_arity(b);
235 for (i = 0; i < irn_arity; i++)
236 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
237 /* save it in the flag. */
238 set_Block_block_visited(b, irg_v + res);
244 * Copies the node to the new obstack. The Ins of the new node point to
245 * the predecessors on the old obstack. For block/phi nodes not all
246 * predecessors might be copied. n->link points to the new node.
247 * For Phi and Block nodes the function allocates in-arrays with an arity
248 * only for useful predecessors. The arity is determined by counting
249 * the non-bad predecessors of the block.
251 * @param n The node to be copied
252 * @param env if non-NULL, the node number attribute will be copied to the new node
254 * Note: Also used for loop unrolling.
256 static void copy_node(ir_node *n, void *env) {
259 ir_op *op = get_irn_op(n);
260 int copy_node_nr = env != NULL;
262 /* The end node looses it's flexible in array. This doesn't matter,
263 as dead node elimination builds End by hand, inlineing doesn't use
265 /* assert(op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
268 /* node copied already */
270 } else if (op == op_Block) {
272 new_arity = compute_new_arity(n);
273 n->attr.block.graph_arr = NULL;
275 block = get_nodes_block(n);
277 new_arity = compute_new_arity(block);
279 new_arity = get_irn_arity(n);
282 nn = new_ir_node(get_irn_dbg_info(n),
289 /* Copy the attributes. These might point to additional data. If this
290 was allocated on the old obstack the pointers now are dangling. This
291 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
292 copy_node_attr(n, nn);
293 new_backedge_info(nn);
297 /* for easier debugging, we want to copy the node numbers too */
298 nn->node_nr = n->node_nr;
303 hook_dead_node_elim_subst(current_ir_graph, n, nn);
307 * Copies new predecessors of old node to new node remembered in link.
308 * Spare the Bad predecessors of Phi and Block nodes.
311 copy_preds (ir_node *n, void *env) {
315 nn = get_new_node(n);
317 /* printf("\n old node: "); DDMSG2(n);
318 printf(" new node: "); DDMSG2(nn);
319 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
322 /* Don't copy Bad nodes. */
324 irn_arity = get_irn_arity(n);
325 for (i = 0; i < irn_arity; i++)
326 if (! is_Bad(get_irn_n(n, i))) {
327 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
328 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
331 /* repair the block visited flag from above misuse. Repair it in both
332 graphs so that the old one can still be used. */
333 set_Block_block_visited(nn, 0);
334 set_Block_block_visited(n, 0);
335 /* Local optimization could not merge two subsequent blocks if
336 in array contained Bads. Now it's possible.
337 We don't call optimize_in_place as it requires
338 that the fields in ir_graph are set properly. */
339 if ((get_opt_control_flow_straightening()) &&
340 (get_Block_n_cfgpreds(nn) == 1) &&
341 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
342 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
344 /* Jmp jumps into the block it is in -- deal self cycle. */
345 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
346 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
351 } else if (get_irn_op(n) == op_Phi) {
352 /* Don't copy node if corresponding predecessor in block is Bad.
353 The Block itself should not be Bad. */
354 block = get_nodes_block(n);
355 set_irn_n (nn, -1, get_new_node(block));
357 irn_arity = get_irn_arity(n);
358 for (i = 0; i < irn_arity; i++)
359 if (! is_Bad(get_irn_n(block, i))) {
360 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
361 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
364 /* If the pre walker reached this Phi after the post walker visited the
365 block block_visited is > 0. */
366 set_Block_block_visited(get_nodes_block(n), 0);
367 /* Compacting the Phi's ins might generate Phis with only one
369 if (get_irn_arity(nn) == 1)
370 exchange(nn, get_irn_n(nn, 0));
372 irn_arity = get_irn_arity(n);
373 for (i = -1; i < irn_arity; i++)
374 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
376 /* Now the new node is complete. We can add it to the hash table for CSE.
377 @@@ inlining aborts if we identify End. Why? */
378 if (get_irn_op(nn) != op_End)
379 add_identities (current_ir_graph->value_table, nn);
383 * Copies the graph recursively, compacts the keep-alives of the end node.
385 * @param irg the graph to be copied
386 * @param copy_node_nr If non-zero, the node number will be copied
388 static void copy_graph(ir_graph *irg, int copy_node_nr) {
389 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
390 ir_node *ka; /* keep alive */
394 /* Some nodes must be copied by hand, sigh */
395 vfl = get_irg_visited(irg);
396 set_irg_visited(irg, vfl + 1);
398 oe = get_irg_end(irg);
399 mark_irn_visited(oe);
400 /* copy the end node by hand, allocate dynamic in array! */
401 ne = new_ir_node(get_irn_dbg_info(oe),
408 /* Copy the attributes. Well, there might be some in the future... */
409 copy_node_attr(oe, ne);
410 set_new_node(oe, ne);
412 /* copy the Bad node */
413 ob = get_irg_bad(irg);
414 mark_irn_visited(ob);
415 nb = new_ir_node(get_irn_dbg_info(ob),
422 copy_node_attr(ob, nb);
423 set_new_node(ob, nb);
425 /* copy the NoMem node */
426 om = get_irg_no_mem(irg);
427 mark_irn_visited(om);
428 nm = new_ir_node(get_irn_dbg_info(om),
435 copy_node_attr(om, nm);
436 set_new_node(om, nm);
438 /* copy the live nodes */
439 set_irg_visited(irg, vfl);
440 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
442 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
444 /* visit the anchors as well */
445 for (i = anchor_max - 1; i >= 0; --i) {
446 ir_node *n = irg->anchors[i];
448 if (n && (get_irn_visited(n) <= vfl)) {
449 set_irg_visited(irg, vfl);
450 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
454 /* copy_preds for the end node ... */
455 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
457 /*- ... and now the keep alives. -*/
458 /* First pick the not marked block nodes and walk them. We must pick these
459 first as else we will oversee blocks reachable from Phis. */
460 irn_arity = get_irn_arity(oe);
461 for (i = 0; i < irn_arity; i++) {
462 ka = get_irn_intra_n(oe, i);
464 (get_irn_visited(ka) <= vfl)) {
465 /* We must keep the block alive and copy everything reachable */
466 set_irg_visited(irg, vfl);
467 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
468 add_End_keepalive(ne, get_new_node(ka));
472 /* Now pick other nodes. Here we will keep all! */
473 irn_arity = get_irn_arity(oe);
474 for (i = 0; i < irn_arity; i++) {
475 ka = get_irn_intra_n(oe, i);
477 if (get_irn_visited(ka) <= vfl) {
478 /* We didn't copy the node yet. */
479 set_irg_visited(irg, vfl);
480 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
482 add_End_keepalive(ne, get_new_node(ka));
486 /* start block sometimes only reached after keep alives */
487 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
488 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
492 * Copies the graph reachable from current_ir_graph->end to the obstack
493 * in current_ir_graph and fixes the environment.
494 * Then fixes the fields in current_ir_graph containing nodes of the
497 * @param copy_node_nr If non-zero, the node number will be copied
500 copy_graph_env (int copy_node_nr) {
501 ir_graph *irg = current_ir_graph;
502 ir_node *old_end, *n;
505 /* remove end_except and end_reg nodes */
506 old_end = get_irg_end(irg);
507 set_irg_end_except (irg, old_end);
508 set_irg_end_reg (irg, old_end);
510 /* Not all nodes remembered in irg might be reachable
511 from the end node. Assure their link is set to NULL, so that
512 we can test whether new nodes have been computed. */
513 for (i = anchor_max - 1; i >= 0; --i)
515 set_new_node(irg->anchors[i], NULL);
517 /* we use the block walk flag for removing Bads from Blocks ins. */
518 inc_irg_block_visited(irg);
521 copy_graph(irg, copy_node_nr);
523 /* fix the fields in irg */
524 old_end = get_irg_end(irg);
525 for (i = anchor_max - 1; i >= 0; --i) {
528 irg->anchors[i] = get_new_node(n);
534 * Copies all reachable nodes to a new obstack. Removes bad inputs
535 * from block nodes and the corresponding inputs from Phi nodes.
536 * Merges single exit blocks with single entry blocks and removes
538 * Adds all new nodes to a new hash table for CSE. Does not
539 * perform CSE, so the hash table might contain common subexpressions.
542 dead_node_elimination(ir_graph *irg) {
544 int rem_ipview = get_interprocedural_view();
545 struct obstack *graveyard_obst = NULL;
546 struct obstack *rebirth_obst = NULL;
548 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
549 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
551 /* inform statistics that we started a dead-node elimination run */
552 hook_dead_node_elim(irg, 1);
554 /* Remember external state of current_ir_graph. */
555 rem = current_ir_graph;
556 current_ir_graph = irg;
557 set_interprocedural_view(0);
559 assert(get_irg_phase_state(current_ir_graph) != phase_building);
561 /* Handle graph state */
562 free_callee_info(current_ir_graph);
563 free_irg_outs(current_ir_graph);
566 /* @@@ so far we loose loops when copying */
567 free_loop_information(current_ir_graph);
569 set_irg_doms_inconsistent(irg);
571 /* A quiet place, where the old obstack can rest in peace,
572 until it will be cremated. */
573 graveyard_obst = irg->obst;
575 /* A new obstack, where the reachable nodes will be copied to. */
576 rebirth_obst = xmalloc(sizeof(*rebirth_obst));
577 current_ir_graph->obst = rebirth_obst;
578 obstack_init (current_ir_graph->obst);
579 current_ir_graph->last_node_idx = 0;
581 /* We also need a new value table for CSE */
582 del_identities(irg->value_table);
583 irg->value_table = new_identities();
585 /* Copy the graph from the old to the new obstack */
588 /* Free memory from old unoptimized obstack */
589 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
590 xfree (graveyard_obst); /* ... then free it. */
592 /* inform statistics that the run is over */
593 hook_dead_node_elim(irg, 0);
595 current_ir_graph = rem;
596 set_interprocedural_view(rem_ipview);
601 * Relink bad predecessors of a block and store the old in array to the
602 * link field. This function is called by relink_bad_predecessors().
603 * The array of link field starts with the block operand at position 0.
604 * If block has bad predecessors, create a new in array without bad preds.
605 * Otherwise let in array untouched.
607 static void relink_bad_block_predecessors(ir_node *n, void *env) {
608 ir_node **new_in, *irn;
609 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
611 /* if link field of block is NULL, look for bad predecessors otherwise
612 this is already done */
613 if (get_irn_op(n) == op_Block &&
614 get_irn_link(n) == NULL) {
616 /* save old predecessors in link field (position 0 is the block operand)*/
617 set_irn_link(n, get_irn_in(n));
619 /* count predecessors without bad nodes */
620 old_irn_arity = get_irn_arity(n);
621 for (i = 0; i < old_irn_arity; i++)
622 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
624 /* arity changing: set new predecessors without bad nodes */
625 if (new_irn_arity < old_irn_arity) {
626 /* Get new predecessor array. We do not resize the array, as we must
627 keep the old one to update Phis. */
628 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
630 /* set new predecessors in array */
633 for (i = 0; i < old_irn_arity; i++) {
634 irn = get_irn_n(n, i);
636 new_in[new_irn_n] = irn;
637 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
641 //ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity);
642 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
645 } /* ir node has bad predecessors */
647 } /* Block is not relinked */
651 * Relinks Bad predecessors from Blocks and Phis called by walker
652 * remove_bad_predecesors(). If n is a Block, call
653 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
654 * function of Phi's Block. If this block has bad predecessors, relink preds
657 static void relink_bad_predecessors(ir_node *n, void *env) {
658 ir_node *block, **old_in;
659 int i, old_irn_arity, new_irn_arity;
661 /* relink bad predecessors of a block */
662 if (get_irn_op(n) == op_Block)
663 relink_bad_block_predecessors(n, env);
665 /* If Phi node relink its block and its predecessors */
666 if (get_irn_op(n) == op_Phi) {
668 /* Relink predecessors of phi's block */
669 block = get_nodes_block(n);
670 if (get_irn_link(block) == NULL)
671 relink_bad_block_predecessors(block, env);
673 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
674 old_irn_arity = ARR_LEN(old_in);
676 /* Relink Phi predecessors if count of predecessors changed */
677 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
678 /* set new predecessors in array
679 n->in[0] remains the same block */
681 for(i = 1; i < old_irn_arity; i++)
682 if (!is_Bad((ir_node *)old_in[i])) {
683 n->in[new_irn_arity] = n->in[i];
684 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
688 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
689 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
692 } /* n is a Phi node */
696 * Removes Bad Bad predecessors from Blocks and the corresponding
697 * inputs to Phi nodes as in dead_node_elimination but without
699 * On walking up set the link field to NULL, on walking down call
700 * relink_bad_predecessors() (This function stores the old in array
701 * to the link field and sets a new in array if arity of predecessors
704 void remove_bad_predecessors(ir_graph *irg) {
705 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
712 __)|_| | \_/ | \_/(/_ |_/\__|__
714 The following stuff implements a facility that automatically patches
715 registered ir_node pointers to the new node when a dead node elimination occurs.
718 struct _survive_dce_t {
722 hook_entry_t dead_node_elim;
723 hook_entry_t dead_node_elim_subst;
726 typedef struct _survive_dce_list_t {
727 struct _survive_dce_list_t *next;
729 } survive_dce_list_t;
731 static void dead_node_hook(void *context, ir_graph *irg, int start)
733 survive_dce_t *sd = context;
735 /* Create a new map before the dead node elimination is performed. */
737 sd->new_places = pmap_create_ex(pmap_count(sd->places));
740 /* Patch back all nodes if dead node elimination is over and something is to be done. */
742 pmap_destroy(sd->places);
743 sd->places = sd->new_places;
744 sd->new_places = NULL;
749 * Hook called when dead node elimination replaces old by nw.
751 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw)
753 survive_dce_t *sd = context;
754 survive_dce_list_t *list = pmap_get(sd->places, old);
756 /* If the node is to be patched back, write the new address to all registered locations. */
758 survive_dce_list_t *p;
760 for(p = list; p; p = p->next)
763 pmap_insert(sd->new_places, nw, list);
768 * Make a new Survive DCE environment.
770 survive_dce_t *new_survive_dce(void)
772 survive_dce_t *res = xmalloc(sizeof(res[0]));
773 obstack_init(&res->obst);
774 res->places = pmap_create();
775 res->new_places = NULL;
777 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
778 res->dead_node_elim.context = res;
779 res->dead_node_elim.next = NULL;
781 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
782 res->dead_node_elim_subst.context = res;
783 res->dead_node_elim_subst.next = NULL;
785 register_hook(hook_dead_node_elim, &res->dead_node_elim);
786 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
791 * Free a Survive DCE environment.
793 void free_survive_dce(survive_dce_t *sd)
795 obstack_free(&sd->obst, NULL);
796 pmap_destroy(sd->places);
797 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
798 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
803 * Register a node pointer to be patched upon DCE.
804 * When DCE occurs, the node pointer specified by @p place will be
805 * patched to the new address of the node it is pointing to.
807 * @param sd The Survive DCE environment.
808 * @param place The address of the node pointer.
810 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place)
813 ir_node *irn = *place;
814 survive_dce_list_t *curr = pmap_get(sd->places, irn);
815 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw));
820 pmap_insert(sd->places, irn, nw);
824 /*--------------------------------------------------------------------*/
825 /* Functionality for inlining */
826 /*--------------------------------------------------------------------*/
829 * Copy node for inlineing. Updates attributes that change when
830 * inlineing but not for dead node elimination.
832 * Copies the node by calling copy_node() and then updates the entity if
833 * it's a local one. env must be a pointer of the frame type of the
834 * inlined procedure. The new entities must be in the link field of
838 copy_node_inline (ir_node *n, void *env) {
840 ir_type *frame_tp = (ir_type *)env;
843 if (get_irn_op(n) == op_Sel) {
844 nn = get_new_node (n);
846 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
847 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
849 } else if (get_irn_op(n) == op_Block) {
850 nn = get_new_node (n);
851 nn->attr.block.irg = current_ir_graph;
855 static void find_addr(ir_node *node, void *env)
857 if (get_irn_opcode(node) == iro_Proj) {
858 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
864 * currently, we cannot inline two cases:
865 * - call with compound arguments
866 * - graphs that take the address of a parameter
868 * check these conditions here
870 static int can_inline(ir_node *call, ir_graph *called_graph)
872 ir_type *call_type = get_Call_type(call);
873 int params, ress, i, res;
874 assert(is_Method_type(call_type));
876 params = get_method_n_params(call_type);
877 ress = get_method_n_ress(call_type);
880 for (i = 0; i < params; ++i) {
881 ir_type *p_type = get_method_param_type(call_type, i);
883 if (is_compound_type(p_type))
888 for (i = 0; i < ress; ++i) {
889 ir_type *r_type = get_method_res_type(call_type, i);
891 if (is_compound_type(r_type))
896 irg_walk_graph(called_graph, find_addr, NULL, &res);
901 int inline_method(ir_node *call, ir_graph *called_graph) {
903 ir_node *post_call, *post_bl;
904 ir_node *in[pn_Start_max];
905 ir_node *end, *end_bl;
909 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
911 ir_type *called_frame;
912 irg_inline_property prop = get_irg_inline_property(called_graph);
914 if ( (prop != irg_inline_forced) &&
915 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
917 /* Do not inline variadic functions. */
918 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
921 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
922 get_method_n_params(get_Call_type(call)));
925 * currently, we cannot inline two cases:
926 * - call with compound arguments
927 * - graphs that take the address of a parameter
929 if (! can_inline(call, called_graph))
932 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
933 rem_opt = get_opt_optimize();
936 /* Handle graph state */
937 assert(get_irg_phase_state(current_ir_graph) != phase_building);
938 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
939 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
940 set_irg_outs_inconsistent(current_ir_graph);
941 set_irg_extblk_inconsistent(current_ir_graph);
942 set_irg_doms_inconsistent(current_ir_graph);
943 set_irg_loopinfo_inconsistent(current_ir_graph);
944 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
946 /* -- Check preconditions -- */
947 assert(is_Call(call));
948 /* @@@ does not work for InterfaceIII.java after cgana
949 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
950 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
951 get_Call_type(call)));
953 assert(get_type_tpop(get_Call_type(call)) == type_method);
954 if (called_graph == current_ir_graph) {
955 set_optimize(rem_opt);
959 /* here we know we WILL inline, so inform the statistics */
960 hook_inline(call, called_graph);
962 /* -- Decide how to handle exception control flow: Is there a handler
963 for the Call node, or do we branch directly to End on an exception?
965 0 There is a handler.
967 2 Exception handling not represented in Firm. -- */
969 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
970 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
971 assert(get_irn_op(proj) == op_Proj);
972 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
973 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
975 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
976 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
977 else { exc_handling = 2; } /* !Mproj && !Xproj */
982 the procedure and later replaces the Start node of the called graph.
983 Post_call is the old Call node and collects the results of the called
984 graph. Both will end up being a tuple. -- */
985 post_bl = get_nodes_block(call);
986 set_irg_current_block(current_ir_graph, post_bl);
987 /* XxMxPxPxPxT of Start + parameter of Call */
988 in[pn_Start_X_initial_exec] = new_Jmp();
989 in[pn_Start_M] = get_Call_mem(call);
990 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
991 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
992 in[pn_Start_P_tls] = get_irg_tls(current_ir_graph);
993 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
994 /* in[pn_Start_P_value_arg_base] = ??? */
995 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
996 pre_call = new_Tuple(pn_Start_max - 1, in);
1000 The new block gets the ins of the old block, pre_call and all its
1001 predecessors and all Phi nodes. -- */
1002 part_block(pre_call);
1004 /* -- Prepare state for dead node elimination -- */
1005 /* Visited flags in calling irg must be >= flag in called irg.
1006 Else walker and arity computation will not work. */
1007 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
1008 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
1009 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
1010 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
1011 /* Set pre_call as new Start node in link field of the start node of
1012 calling graph and pre_calls block as new block for the start block
1014 Further mark these nodes so that they are not visited by the
1016 set_irn_link(get_irg_start(called_graph), pre_call);
1017 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
1018 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1019 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
1020 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
1021 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
1023 /* Initialize for compaction of in arrays */
1024 inc_irg_block_visited(current_ir_graph);
1026 /* -- Replicate local entities of the called_graph -- */
1027 /* copy the entities. */
1028 called_frame = get_irg_frame_type(called_graph);
1029 for (i = 0; i < get_class_n_members(called_frame); i++) {
1030 entity *new_ent, *old_ent;
1031 old_ent = get_class_member(called_frame, i);
1032 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
1033 set_entity_link(old_ent, new_ent);
1036 /* visited is > than that of called graph. With this trick visited will
1037 remain unchanged so that an outer walker, e.g., searching the call nodes
1038 to inline, calling this inline will not visit the inlined nodes. */
1039 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
1041 /* -- Performing dead node elimination inlines the graph -- */
1042 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1044 /* @@@ endless loops are not copied!! -- they should be, I think... */
1045 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
1046 get_irg_frame_type(called_graph));
1048 /* Repair called_graph */
1049 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
1050 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
1051 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1053 /* -- Merge the end of the inlined procedure with the call site -- */
1054 /* We will turn the old Call node into a Tuple with the following
1057 0: Phi of all Memories of Return statements.
1058 1: Jmp from new Block that merges the control flow from all exception
1059 predecessors of the old end block.
1060 2: Tuple of all arguments.
1061 3: Phi of Exception memories.
1062 In case the old Call directly branches to End on an exception we don't
1063 need the block merging all exceptions nor the Phi of the exception
1067 /* -- Precompute some values -- */
1068 end_bl = get_new_node(get_irg_end_block(called_graph));
1069 end = get_new_node(get_irg_end(called_graph));
1070 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1071 n_res = get_method_n_ress(get_Call_type(call));
1073 res_pred = xmalloc (n_res * sizeof(*res_pred));
1074 cf_pred = xmalloc (arity * sizeof(*res_pred));
1076 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
1078 /* -- archive keepalives -- */
1079 irn_arity = get_irn_arity(end);
1080 for (i = 0; i < irn_arity; i++)
1081 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
1083 /* The new end node will die. We need not free as the in array is on the obstack:
1084 copy_node() only generated 'D' arrays. */
1086 /* -- Replace Return nodes by Jump nodes. -- */
1088 for (i = 0; i < arity; i++) {
1090 ret = get_irn_n(end_bl, i);
1091 if (is_Return(ret)) {
1092 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
1096 set_irn_in(post_bl, n_ret, cf_pred);
1098 /* -- Build a Tuple for all results of the method.
1099 Add Phi node if there was more than one Return. -- */
1100 turn_into_tuple(post_call, 4);
1101 /* First the Memory-Phi */
1103 for (i = 0; i < arity; i++) {
1104 ret = get_irn_n(end_bl, i);
1105 if (is_Return(ret)) {
1106 cf_pred[n_ret] = get_Return_mem(ret);
1110 phi = new_Phi(n_ret, cf_pred, mode_M);
1111 set_Tuple_pred(call, pn_Call_M_regular, phi);
1112 /* Conserve Phi-list for further inlinings -- but might be optimized */
1113 if (get_nodes_block(phi) == post_bl) {
1114 set_irn_link(phi, get_irn_link(post_bl));
1115 set_irn_link(post_bl, phi);
1117 /* Now the real results */
1119 for (j = 0; j < n_res; j++) {
1121 for (i = 0; i < arity; i++) {
1122 ret = get_irn_n(end_bl, i);
1123 if (get_irn_op(ret) == op_Return) {
1124 cf_pred[n_ret] = get_Return_res(ret, j);
1129 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1133 /* Conserve Phi-list for further inlinings -- but might be optimized */
1134 if (get_nodes_block(phi) == post_bl) {
1135 set_irn_link(phi, get_irn_link(post_bl));
1136 set_irn_link(post_bl, phi);
1139 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1141 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1143 /* Finally the exception control flow.
1144 We have two (three) possible situations:
1145 First if the Call branches to an exception handler: We need to add a Phi node to
1146 collect the memory containing the exception objects. Further we need
1147 to add another block to get a correct representation of this Phi. To
1148 this block we add a Jmp that resolves into the X output of the Call
1149 when the Call is turned into a tuple.
1150 Second the Call branches to End, the exception is not handled. Just
1151 add all inlined exception branches to the End node.
1152 Third: there is no Exception edge at all. Handle as case two. */
1153 if (exc_handling == 0) {
1155 for (i = 0; i < arity; i++) {
1157 ret = get_irn_n(end_bl, i);
1158 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1159 cf_pred[n_exc] = ret;
1164 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1165 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1166 /* The Phi for the memories with the exception objects */
1168 for (i = 0; i < arity; i++) {
1170 ret = skip_Proj(get_irn_n(end_bl, i));
1172 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1174 } else if (is_fragile_op(ret)) {
1175 /* We rely that all cfops have the memory output at the same position. */
1176 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1178 } else if (get_irn_op(ret) == op_Raise) {
1179 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1183 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1185 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1186 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1189 ir_node *main_end_bl;
1190 int main_end_bl_arity;
1191 ir_node **end_preds;
1193 /* assert(exc_handling == 1 || no exceptions. ) */
1195 for (i = 0; i < arity; i++) {
1196 ir_node *ret = get_irn_n(end_bl, i);
1198 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1199 cf_pred[n_exc] = ret;
1203 main_end_bl = get_irg_end_block(current_ir_graph);
1204 main_end_bl_arity = get_irn_arity(main_end_bl);
1205 end_preds = xmalloc ((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1207 for (i = 0; i < main_end_bl_arity; ++i)
1208 end_preds[i] = get_irn_n(main_end_bl, i);
1209 for (i = 0; i < n_exc; ++i)
1210 end_preds[main_end_bl_arity + i] = cf_pred[i];
1211 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1212 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1213 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1219 #if 0 /* old. now better, correcter, faster implementation. */
1221 /* -- If the exception control flow from the inlined Call directly
1222 branched to the end block we now have the following control
1223 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
1224 remove the Jmp along with it's empty block and add Jmp's
1225 predecessors as predecessors of this end block. No problem if
1226 there is no exception, because then branches Bad to End which
1228 @@@ can't we know this beforehand: by getting the Proj(1) from
1229 the Call link list and checking whether it goes to Proj. */
1230 /* find the problematic predecessor of the end block. */
1231 end_bl = get_irg_end_block(current_ir_graph);
1232 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
1233 cf_op = get_Block_cfgpred(end_bl, i);
1234 if (get_irn_op(cf_op) == op_Proj) {
1235 cf_op = get_Proj_pred(cf_op);
1236 if ((get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
1237 /* There are unoptimized tuples from inlineing before when no exc */
1238 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
1239 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
1240 assert(get_irn_op(cf_op) == op_Jmp);
1246 if (i < get_Block_n_cfgpreds(end_bl)) {
1247 bl = get_nodes_block(cf_op);
1248 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
1249 cf_pred = xmalloc (arity * sizeof(*cf_pred));
1250 for (j = 0; j < i; j++)
1251 cf_pred[j] = get_Block_cfgpred(end_bl, j);
1252 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
1253 cf_pred[j] = get_Block_cfgpred(bl, j-i);
1254 for (j = j; j < arity; j++)
1255 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
1256 set_irn_in(end_bl, arity, cf_pred);
1258 /* Remove the exception pred from post-call Tuple. */
1259 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1264 /* -- Turn CSE back on. -- */
1265 set_optimize(rem_opt);
1270 /********************************************************************/
1271 /* Apply inlineing to small methods. */
1272 /********************************************************************/
1274 /* It makes no sense to inline too many calls in one procedure. Anyways,
1275 I didn't get a version with NEW_ARR_F to run. */
1276 #define MAX_INLINE 1024
1279 * environment for inlining small irgs
1281 typedef struct _inline_env_t {
1283 ir_node *calls[MAX_INLINE];
1287 * Returns the irg called from a Call node. If the irg is not
1288 * known, NULL is returned.
1290 static ir_graph *get_call_called_irg(ir_node *call) {
1292 ir_graph *called_irg = NULL;
1294 assert(is_Call(call));
1296 addr = get_Call_ptr(call);
1297 if ((get_irn_op(addr) == op_SymConst) && (get_SymConst_kind (addr) == symconst_addr_ent)) {
1298 called_irg = get_entity_irg(get_SymConst_entity(addr));
1304 static void collect_calls(ir_node *call, void *env) {
1307 if (! is_Call(call)) return;
1309 addr = get_Call_ptr(call);
1311 if (get_irn_op(addr) == op_SymConst) {
1312 if (get_SymConst_kind(addr) == symconst_addr_ent) {
1313 ir_graph *called_irg = get_entity_irg(get_SymConst_entity(addr));
1314 inline_env_t *ienv = (inline_env_t *)env;
1315 if (called_irg && ienv->pos < MAX_INLINE) {
1316 /* The Call node calls a locally defined method. Remember to inline. */
1317 ienv->calls[ienv->pos++] = call;
1324 * Inlines all small methods at call sites where the called address comes
1325 * from a Const node that references the entity representing the called
1327 * The size argument is a rough measure for the code size of the method:
1328 * Methods where the obstack containing the firm graph is smaller than
1331 void inline_small_irgs(ir_graph *irg, int size) {
1333 ir_graph *rem = current_ir_graph;
1334 inline_env_t env /* = {0, NULL}*/;
1336 if (!(get_opt_optimize() && get_opt_inline())) return;
1338 current_ir_graph = irg;
1339 /* Handle graph state */
1340 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1341 free_callee_info(current_ir_graph);
1343 /* Find Call nodes to inline.
1344 (We can not inline during a walk of the graph, as inlineing the same
1345 method several times changes the visited flag of the walked graph:
1346 after the first inlineing visited of the callee equals visited of
1347 the caller. With the next inlineing both are increased.) */
1349 irg_walk(get_irg_end(irg), NULL, collect_calls, &env);
1351 if ((env.pos > 0) && (env.pos < MAX_INLINE)) {
1352 /* There are calls to inline */
1353 collect_phiprojs(irg);
1354 for (i = 0; i < env.pos; i++) {
1356 callee = get_entity_irg(get_SymConst_entity(get_Call_ptr(env.calls[i])));
1357 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1358 (get_irg_inline_property(callee) == irg_inline_forced)) {
1359 inline_method(env.calls[i], callee);
1364 current_ir_graph = rem;
1368 * Environment for inlining irgs.
1371 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1372 int n_nodes_orig; /**< for statistics */
1373 eset *call_nodes; /**< All call nodes in this graph */
1375 int n_call_nodes_orig; /**< for statistics */
1376 int n_callers; /**< Number of known graphs that call this graphs. */
1377 int n_callers_orig; /**< for statistics */
1381 * Allocate a new environment for inlining.
1383 static inline_irg_env *new_inline_irg_env(void) {
1384 inline_irg_env *env = xmalloc(sizeof(*env));
1385 env->n_nodes = -2; /* do not count count Start, End */
1386 env->n_nodes_orig = -2; /* do not count Start, End */
1387 env->call_nodes = eset_create();
1388 env->n_call_nodes = 0;
1389 env->n_call_nodes_orig = 0;
1391 env->n_callers_orig = 0;
1396 * destroy an environment for inlining.
1398 static void free_inline_irg_env(inline_irg_env *env) {
1399 eset_destroy(env->call_nodes);
1404 * post-walker: collect all calls in the inline-environment
1405 * of a graph and sum some statistics.
1407 static void collect_calls2(ir_node *call, void *env) {
1408 inline_irg_env *x = (inline_irg_env *)env;
1409 ir_op *op = get_irn_op(call);
1412 /* count meaningful nodes in irg */
1413 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1418 if (op != op_Call) return;
1420 /* collect all call nodes */
1421 eset_insert(x->call_nodes, call);
1423 x->n_call_nodes_orig++;
1425 /* count all static callers */
1426 callee = get_call_called_irg(call);
1428 inline_irg_env *callee_env = get_irg_link(callee);
1429 callee_env->n_callers++;
1430 callee_env->n_callers_orig++;
1435 * Returns TRUE if the number of callers in 0 in the irg's environment,
1436 * hence this irg is a leave.
1438 INLINE static int is_leave(ir_graph *irg) {
1439 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1443 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1445 INLINE static int is_smaller(ir_graph *callee, int size) {
1446 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1451 * Inlines small leave methods at call sites where the called address comes
1452 * from a Const node that references the entity representing the called
1454 * The size argument is a rough measure for the code size of the method:
1455 * Methods where the obstack containing the firm graph is smaller than
1458 void inline_leave_functions(int maxsize, int leavesize, int size) {
1459 inline_irg_env *env;
1460 int i, n_irgs = get_irp_n_irgs();
1461 ir_graph *rem = current_ir_graph;
1464 if (!(get_opt_optimize() && get_opt_inline())) return;
1466 /* extend all irgs by a temporary data structure for inlining. */
1467 for (i = 0; i < n_irgs; ++i)
1468 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1470 /* Precompute information in temporary data structure. */
1471 for (i = 0; i < n_irgs; ++i) {
1472 current_ir_graph = get_irp_irg(i);
1473 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1474 free_callee_info(current_ir_graph);
1476 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1477 get_irg_link(current_ir_graph));
1480 /* -- and now inline. -- */
1482 /* Inline leaves recursively -- we might construct new leaves. */
1483 while (did_inline) {
1486 for (i = 0; i < n_irgs; ++i) {
1488 int phiproj_computed = 0;
1490 current_ir_graph = get_irp_irg(i);
1491 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1493 for (call = eset_first(env->call_nodes); call; call = eset_next(env->call_nodes)) {
1496 if (get_irn_op(call) == op_Tuple) continue; /* We already have inlined this call. */
1497 callee = get_call_called_irg(call);
1499 if (env->n_nodes > maxsize) continue; // break;
1501 if (callee && (is_leave(callee) && is_smaller(callee, leavesize))) {
1502 if (!phiproj_computed) {
1503 phiproj_computed = 1;
1504 collect_phiprojs(current_ir_graph);
1506 did_inline = inline_method(call, callee);
1509 /* Do some statistics */
1510 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1511 env->n_call_nodes --;
1512 env->n_nodes += callee_env->n_nodes;
1513 callee_env->n_callers--;
1520 /* inline other small functions. */
1521 for (i = 0; i < n_irgs; ++i) {
1524 int phiproj_computed = 0;
1526 current_ir_graph = get_irp_irg(i);
1527 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1529 /* we can not walk and change a set, nor remove from it.
1531 walkset = env->call_nodes;
1532 env->call_nodes = eset_create();
1533 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1536 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1537 callee = get_call_called_irg(call);
1540 ((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1541 (get_irg_inline_property(callee) == irg_inline_forced))) {
1542 if (!phiproj_computed) {
1543 phiproj_computed = 1;
1544 collect_phiprojs(current_ir_graph);
1546 if (inline_method(call, callee)) {
1547 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1548 env->n_call_nodes--;
1549 eset_insert_all(env->call_nodes, callee_env->call_nodes); /* @@@ ??? This are the wrong nodes !? Not the copied ones. */
1550 env->n_call_nodes += callee_env->n_call_nodes;
1551 env->n_nodes += callee_env->n_nodes;
1552 callee_env->n_callers--;
1555 eset_insert(env->call_nodes, call);
1558 eset_destroy(walkset);
1561 for (i = 0; i < n_irgs; ++i) {
1562 current_ir_graph = get_irp_irg(i);
1564 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1565 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1566 (env->n_callers_orig != env->n_callers))
1567 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1568 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1569 env->n_callers_orig, env->n_callers,
1570 get_entity_name(get_irg_entity(current_ir_graph)));
1572 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1575 current_ir_graph = rem;
1578 /*******************************************************************/
1579 /* Code Placement. Pins all floating nodes to a block where they */
1580 /* will be executed only if needed. */
1581 /*******************************************************************/
1584 * Returns non-zero, is a block is not reachable from Start.
1586 * @param block the block to test
1589 is_Block_unreachable(ir_node *block) {
1590 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1594 * Find the earliest correct block for N. --- Place N into the
1595 * same Block as its dominance-deepest Input.
1597 * We have to avoid calls to get_nodes_block() here
1598 * because the graph is floating.
1600 * move_out_of_loops() expects that place_floats_early() have placed
1601 * all "living" nodes into a living block. That's why we must
1602 * move nodes in dead block with "live" successors into a valid
1604 * We move them just into the same block as it's successor (or
1605 * in case of a Phi into the effective use block). For Phi successors,
1606 * this may still be a dead block, but then there is no real use, as
1607 * the control flow will be dead later.
1610 place_floats_early(ir_node *n, pdeq *worklist)
1614 /* we must not run into an infinite loop */
1615 assert(irn_not_visited(n));
1616 mark_irn_visited(n);
1618 /* Place floating nodes. */
1619 if (get_irn_pinned(n) == op_pin_state_floats) {
1620 ir_node *curr_block = get_irn_n(n, -1);
1621 int in_dead_block = is_Block_unreachable(curr_block);
1623 ir_node *b = NULL; /* The block to place this node in */
1625 assert(get_irn_op(n) != op_Block);
1627 if ((get_irn_op(n) == op_Const) ||
1628 (get_irn_op(n) == op_SymConst) ||
1630 (get_irn_op(n) == op_Unknown)) {
1631 /* These nodes will not be placed by the loop below. */
1632 b = get_irg_start_block(current_ir_graph);
1636 /* find the block for this node. */
1637 irn_arity = get_irn_arity(n);
1638 for (i = 0; i < irn_arity; i++) {
1639 ir_node *pred = get_irn_n(n, i);
1640 ir_node *pred_block;
1642 if ((irn_not_visited(pred))
1643 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1646 * If the current node is NOT in a dead block, but one of its
1647 * predecessors is, we must move the predecessor to a live block.
1648 * Such thing can happen, if global CSE chose a node from a dead block.
1649 * We move it simple to our block.
1650 * Note that neither Phi nor End nodes are floating, so we don't
1651 * need to handle them here.
1653 if (! in_dead_block) {
1654 if (get_irn_pinned(pred) == op_pin_state_floats &&
1655 is_Block_unreachable(get_irn_n(pred, -1)))
1656 set_nodes_block(pred, curr_block);
1658 place_floats_early(pred, worklist);
1662 * A node in the Bad block must stay in the bad block,
1663 * so don't compute a new block for it.
1668 /* Because all loops contain at least one op_pin_state_pinned node, now all
1669 our inputs are either op_pin_state_pinned or place_early() has already
1670 been finished on them. We do not have any unfinished inputs! */
1671 pred_block = get_irn_n(pred, -1);
1672 if ((!is_Block_dead(pred_block)) &&
1673 (get_Block_dom_depth(pred_block) > depth)) {
1675 depth = get_Block_dom_depth(pred_block);
1677 /* Avoid that the node is placed in the Start block */
1678 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)) {
1679 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1680 assert(b != get_irg_start_block(current_ir_graph));
1685 set_nodes_block(n, b);
1689 * Add predecessors of non floating nodes and non-floating predecessors
1690 * of floating nodes to worklist and fix their blocks if the are in dead block.
1692 irn_arity = get_irn_arity(n);
1694 if (get_irn_op(n) == op_End) {
1696 * Simplest case: End node. Predecessors are keep-alives,
1697 * no need to move out of dead block.
1699 for (i = -1; i < irn_arity; ++i) {
1700 ir_node *pred = get_irn_n(n, i);
1701 if (irn_not_visited(pred))
1702 pdeq_putr(worklist, pred);
1705 else if (is_Block(n)) {
1707 * Blocks: Predecessors are control flow, no need to move
1708 * them out of dead block.
1710 for (i = irn_arity - 1; i >= 0; --i) {
1711 ir_node *pred = get_irn_n(n, i);
1712 if (irn_not_visited(pred))
1713 pdeq_putr(worklist, pred);
1716 else if (is_Phi(n)) {
1718 ir_node *curr_block = get_irn_n(n, -1);
1719 int in_dead_block = is_Block_unreachable(curr_block);
1722 * Phi nodes: move nodes from dead blocks into the effective use
1723 * of the Phi-input if the Phi is not in a bad block.
1725 pred = get_irn_n(n, -1);
1726 if (irn_not_visited(pred))
1727 pdeq_putr(worklist, pred);
1729 for (i = irn_arity - 1; i >= 0; --i) {
1730 ir_node *pred = get_irn_n(n, i);
1732 if (irn_not_visited(pred)) {
1733 if (! in_dead_block &&
1734 get_irn_pinned(pred) == op_pin_state_floats &&
1735 is_Block_unreachable(get_irn_n(pred, -1))) {
1736 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1738 pdeq_putr(worklist, pred);
1744 ir_node *curr_block = get_irn_n(n, -1);
1745 int in_dead_block = is_Block_unreachable(curr_block);
1748 * All other nodes: move nodes from dead blocks into the same block.
1750 pred = get_irn_n(n, -1);
1751 if (irn_not_visited(pred))
1752 pdeq_putr(worklist, pred);
1754 for (i = irn_arity - 1; i >= 0; --i) {
1755 ir_node *pred = get_irn_n(n, i);
1757 if (irn_not_visited(pred)) {
1758 if (! in_dead_block &&
1759 get_irn_pinned(pred) == op_pin_state_floats &&
1760 is_Block_unreachable(get_irn_n(pred, -1))) {
1761 set_nodes_block(pred, curr_block);
1763 pdeq_putr(worklist, pred);
1770 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1771 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1772 * places all floating nodes reachable from its argument through floating
1773 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1775 static INLINE void place_early(pdeq *worklist) {
1777 inc_irg_visited(current_ir_graph);
1779 /* this inits the worklist */
1780 place_floats_early(get_irg_end(current_ir_graph), worklist);
1782 /* Work the content of the worklist. */
1783 while (!pdeq_empty(worklist)) {
1784 ir_node *n = pdeq_getl(worklist);
1785 if (irn_not_visited(n))
1786 place_floats_early(n, worklist);
1789 set_irg_outs_inconsistent(current_ir_graph);
1790 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
1794 * Compute the deepest common ancestor of block and dca.
1796 static ir_node *calc_dca(ir_node *dca, ir_node *block)
1800 /* we do not want to place nodes in dead blocks */
1801 if (is_Block_dead(block))
1804 /* We found a first legal placement. */
1805 if (!dca) return block;
1807 /* Find a placement that is dominates both, dca and block. */
1808 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1809 block = get_Block_idom(block);
1811 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1812 dca = get_Block_idom(dca);
1815 while (block != dca)
1816 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1821 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1822 * I.e., DCA is the block where we might place PRODUCER.
1823 * A data flow edge points from producer to consumer.
1826 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer)
1828 ir_node *block = NULL;
1830 /* Compute the latest block into which we can place a node so that it is
1832 if (get_irn_op(consumer) == op_Phi) {
1833 /* our consumer is a Phi-node, the effective use is in all those
1834 blocks through which the Phi-node reaches producer */
1836 ir_node *phi_block = get_nodes_block(consumer);
1837 irn_arity = get_irn_arity(consumer);
1839 for (i = 0; i < irn_arity; i++) {
1840 if (get_irn_n(consumer, i) == producer) {
1841 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1843 if (! is_Block_unreachable(new_block))
1844 block = calc_dca(block, new_block);
1849 block = get_irn_n(producer, -1);
1852 assert(is_no_Block(consumer));
1853 block = get_nodes_block(consumer);
1856 /* Compute the deepest common ancestor of block and dca. */
1857 return calc_dca(dca, block);
1860 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1862 static INLINE int get_irn_loop_depth(ir_node *n) {
1863 return get_loop_depth(get_irn_loop(n));
1867 * Move n to a block with less loop depth than it's current block. The
1868 * new block must be dominated by early.
1870 * @param n the node that should be moved
1871 * @param early the earliest block we can n move to
1874 move_out_of_loops (ir_node *n, ir_node *early)
1876 ir_node *best, *dca;
1880 /* Find the region deepest in the dominator tree dominating
1881 dca with the least loop nesting depth, but still dominated
1882 by our early placement. */
1883 dca = get_nodes_block(n);
1886 while (dca != early) {
1887 dca = get_Block_idom(dca);
1888 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1889 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1893 if (best != get_nodes_block(n)) {
1895 printf("Moving out of loop: "); DDMN(n);
1896 printf(" Outermost block: "); DDMN(early);
1897 printf(" Best block: "); DDMN(best);
1898 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1900 set_nodes_block(n, best);
1905 * Find the latest legal block for N and place N into the
1906 * `optimal' Block between the latest and earliest legal block.
1907 * The `optimal' block is the dominance-deepest block of those
1908 * with the least loop-nesting-depth. This places N out of as many
1909 * loops as possible and then makes it as control dependent as
1913 place_floats_late(ir_node *n, pdeq *worklist)
1918 assert(irn_not_visited(n)); /* no multiple placement */
1920 mark_irn_visited(n);
1922 /* no need to place block nodes, control nodes are already placed. */
1923 if ((get_irn_op(n) != op_Block) &&
1925 (get_irn_mode(n) != mode_X)) {
1926 /* Remember the early_blk placement of this block to move it
1927 out of loop no further than the early_blk placement. */
1928 early_blk = get_irn_n(n, -1);
1931 * BEWARE: Here we also get code, that is live, but
1932 * was in a dead block. If the node is life, but because
1933 * of CSE in a dead block, we still might need it.
1936 /* Assure that our users are all placed, except the Phi-nodes.
1937 --- Each data flow cycle contains at least one Phi-node. We
1938 have to break the `user has to be placed before the
1939 producer' dependence cycle and the Phi-nodes are the
1940 place to do so, because we need to base our placement on the
1941 final region of our users, which is OK with Phi-nodes, as they
1942 are op_pin_state_pinned, and they never have to be placed after a
1943 producer of one of their inputs in the same block anyway. */
1944 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
1945 ir_node *succ = get_irn_out(n, i);
1946 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
1947 place_floats_late(succ, worklist);
1950 if (! is_Block_dead(early_blk)) {
1951 /* do only move things that where not dead */
1953 /* We have to determine the final block of this node... except for
1955 if ((get_irn_pinned(n) == op_pin_state_floats) &&
1956 (get_irn_op(n) != op_Const) &&
1957 (get_irn_op(n) != op_SymConst)) {
1958 ir_node *dca = NULL; /* deepest common ancestor in the
1959 dominator tree of all nodes'
1960 blocks depending on us; our final
1961 placement has to dominate DCA. */
1962 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
1963 ir_node *succ = get_irn_out(n, i);
1966 if (get_irn_op(succ) == op_End) {
1968 * This consumer is the End node, a keep alive edge.
1969 * This is not a real consumer, so we ignore it
1974 /* ignore if succ is in dead code */
1975 succ_blk = get_irn_n(succ, -1);
1976 if (is_Block_unreachable(succ_blk))
1978 dca = consumer_dom_dca(dca, succ, n);
1981 set_nodes_block(n, dca);
1982 move_out_of_loops(n, early_blk);
1988 /* Add predecessors of all non-floating nodes on list. (Those of floating
1989 nodes are placed already and therefore are marked.) */
1990 for (i = 0; i < get_irn_n_outs(n); i++) {
1991 ir_node *succ = get_irn_out(n, i);
1992 if (irn_not_visited(get_irn_out(n, i))) {
1993 pdeq_putr(worklist, succ);
1998 static INLINE void place_late(pdeq *worklist) {
2000 inc_irg_visited(current_ir_graph);
2002 /* This fills the worklist initially. */
2003 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
2005 /* And now empty the worklist again... */
2006 while (!pdeq_empty(worklist)) {
2007 ir_node *n = pdeq_getl(worklist);
2008 if (irn_not_visited(n))
2009 place_floats_late(n, worklist);
2013 void place_code(ir_graph *irg) {
2015 ir_graph *rem = current_ir_graph;
2017 current_ir_graph = irg;
2019 if (!(get_opt_optimize() && get_opt_global_cse())) return;
2021 /* Handle graph state */
2022 assert(get_irg_phase_state(irg) != phase_building);
2025 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
2026 free_loop_information(irg);
2027 construct_backedges(irg);
2030 /* Place all floating nodes as early as possible. This guarantees
2031 a legal code placement. */
2032 worklist = new_pdeq();
2033 place_early(worklist);
2035 /* place_early() invalidates the outs, place_late needs them. */
2036 compute_irg_outs(irg);
2038 /* Now move the nodes down in the dominator tree. This reduces the
2039 unnecessary executions of the node. */
2040 place_late(worklist);
2042 set_irg_outs_inconsistent(current_ir_graph);
2043 set_irg_loopinfo_inconsistent(current_ir_graph);
2045 current_ir_graph = rem;
2049 * Called by walker of remove_critical_cf_edges().
2051 * Place an empty block to an edge between a blocks of multiple
2052 * predecessors and a block of multiple successors.
2055 * @param env Environment of walker. The changed field.
2057 static void walk_critical_cf_edges(ir_node *n, void *env) {
2059 ir_node *pre, *block, *jmp;
2062 /* Block has multiple predecessors */
2063 if (is_Block(n) && (get_irn_arity(n) > 1)) {
2064 if (n == get_irg_end_block(current_ir_graph))
2065 return; /* No use to add a block here. */
2067 arity = get_irn_arity(n);
2068 for (i=0; i<arity; i++) {
2069 pre = get_irn_n(n, i);
2070 /* Predecessor has multiple successors. Insert new control flow edge. */
2071 if (op_Raise != get_irn_op(skip_Proj(pre))) {
2072 /* set predecessor of new block */
2073 block = new_Block(1, &pre);
2074 /* insert new jmp node to new block */
2075 set_cur_block(block);
2078 /* set successor of new block */
2079 set_irn_n(n, i, jmp);
2081 } /* predecessor has multiple successors */
2082 } /* for all predecessors */
2083 } /* n is a block */
2086 void remove_critical_cf_edges(ir_graph *irg) {
2088 irg_walk_graph(irg, NULL, walk_critical_cf_edges, &changed);
2091 /* control flow changed */
2092 set_irg_outs_inconsistent(irg);
2093 set_irg_extblk_inconsistent(irg);
2094 set_irg_doms_inconsistent(current_ir_graph);
2095 set_irg_loopinfo_inconsistent(current_ir_graph);