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"
33 #include "pdeq.h" /* Fuer code placement */
38 #include "irbackedge_t.h"
44 #include "iredges_t.h"
47 /* Defined in iropt.c */
48 pset *new_identities (void);
49 void del_identities (pset *value_table);
50 void add_identities (pset *value_table, ir_node *node);
52 /*------------------------------------------------------------------*/
53 /* apply optimizations of iropt to all nodes. */
54 /*------------------------------------------------------------------*/
56 static void init_link (ir_node *n, void *env) {
57 set_irn_link(n, NULL);
60 #if 0 /* Old version. Avoids Ids.
61 This is not necessary: we do a post walk, and get_irn_n
62 removes ids anyways. So it's much cheaper to call the
63 optimization less often and use the exchange() algorithm. */
65 optimize_in_place_wrapper (ir_node *n, void *env) {
67 ir_node *optimized, *old;
69 irn_arity = get_irn_arity(n);
70 for (i = 0; i < irn_arity; i++) {
71 /* get_irn_n skips Id nodes, so comparison old != optimized does not
72 show all optimizations. Therefore always set new predecessor. */
73 old = get_irn_intra_n(n, i);
74 optimized = optimize_in_place_2(old);
75 set_irn_n(n, i, optimized);
78 if (get_irn_op(n) == op_Block) {
79 optimized = optimize_in_place_2(n);
80 if (optimized != n) exchange (n, optimized);
85 optimize_in_place_wrapper (ir_node *n, void *env) {
86 ir_node *optimized = optimize_in_place_2(n);
87 if (optimized != n) exchange (n, optimized);
92 static INLINE void do_local_optimize(ir_node *n) {
93 /* Handle graph state */
94 assert(get_irg_phase_state(current_ir_graph) != phase_building);
96 if (get_opt_global_cse())
97 set_irg_pinned(current_ir_graph, op_pin_state_floats);
98 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
99 set_irg_outs_inconsistent(current_ir_graph);
100 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
101 set_irg_dom_inconsistent(current_ir_graph);
102 set_irg_loopinfo_inconsistent(current_ir_graph);
104 /* Clean the value_table in irg for the CSE. */
105 del_identities(current_ir_graph->value_table);
106 current_ir_graph->value_table = new_identities();
108 /* walk over the graph */
109 irg_walk(n, init_link, optimize_in_place_wrapper, NULL);
112 void local_optimize_node(ir_node *n) {
113 ir_graph *rem = current_ir_graph;
114 current_ir_graph = get_irn_irg(n);
116 do_local_optimize(n);
118 current_ir_graph = rem;
122 * Block-Walker: uses dominance depth to mark dead blocks.
124 static void kill_dead_blocks(ir_node *block, void *env)
126 if (get_Block_dom_depth(block) < 0)
127 set_Block_dead(block);
131 local_optimize_graph (ir_graph *irg) {
132 ir_graph *rem = current_ir_graph;
133 current_ir_graph = irg;
135 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
136 irg_block_walk_graph(irg, NULL, kill_dead_blocks, NULL);
138 do_local_optimize(irg->end);
140 current_ir_graph = rem;
144 /*------------------------------------------------------------------*/
145 /* Routines for dead node elimination / copying garbage collection */
146 /* of the obstack. */
147 /*------------------------------------------------------------------*/
150 * Remember the new node in the old node by using a field all nodes have.
153 set_new_node (ir_node *old, ir_node *new)
159 * Get this new node, before the old node is forgotten.
161 static INLINE ir_node *
162 get_new_node (ir_node * n) {
167 * We use the block_visited flag to mark that we have computed the
168 * number of useful predecessors for this block.
169 * Further we encode the new arity in this flag in the old blocks.
170 * Remembering the arity is useful, as it saves a lot of pointer
171 * accesses. This function is called for all Phi and Block nodes
175 compute_new_arity(ir_node *b) {
176 int i, res, irn_arity;
179 irg_v = get_irg_block_visited(current_ir_graph);
180 block_v = get_Block_block_visited(b);
181 if (block_v >= irg_v) {
182 /* we computed the number of preds for this block and saved it in the
184 return block_v - irg_v;
186 /* compute the number of good predecessors */
187 res = irn_arity = get_irn_arity(b);
188 for (i = 0; i < irn_arity; i++)
189 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
190 /* save it in the flag. */
191 set_Block_block_visited(b, irg_v + res);
196 /* TODO: add an ir_op operation */
197 static INLINE void new_backedge_info(ir_node *n) {
198 switch(get_irn_opcode(n)) {
200 n->attr.block.cg_backedge = NULL;
201 n->attr.block.backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
204 n->attr.phi_backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
207 n->attr.filter.backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
214 * Copies the node to the new obstack. The Ins of the new node point to
215 * the predecessors on the old obstack. For block/phi nodes not all
216 * predecessors might be copied. n->link points to the new node.
217 * For Phi and Block nodes the function allocates in-arrays with an arity
218 * only for useful predecessors. The arity is determined by counting
219 * the non-bad predecessors of the block.
221 * @param n The node to be copied
222 * @param env if non-NULL, the node number attribute will be copied to the new node
224 * Note: Also used for loop unrolling.
226 static void copy_node(ir_node *n, void *env) {
229 ir_op *op = get_irn_op(n);
230 int copy_node_nr = env != NULL;
232 /* The end node looses it's flexible in array. This doesn't matter,
233 as dead node elimination builds End by hand, inlineing doesn't use
235 /* assert(op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
238 /* node copied already */
240 } else if (op == op_Block) {
242 new_arity = compute_new_arity(n);
243 n->attr.block.graph_arr = NULL;
245 block = get_nodes_block(n);
247 new_arity = compute_new_arity(block);
249 new_arity = get_irn_arity(n);
252 nn = new_ir_node(get_irn_dbg_info(n),
259 /* Copy the attributes. These might point to additional data. If this
260 was allocated on the old obstack the pointers now are dangling. This
261 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
262 copy_node_attr(n, nn);
263 new_backedge_info(nn);
267 /* for easier debugging, we want to copy the node numbers too */
268 nn->node_nr = n->node_nr;
276 * Copies new predecessors of old node to new node remembered in link.
277 * Spare the Bad predecessors of Phi and Block nodes.
280 copy_preds (ir_node *n, void *env) {
284 nn = get_new_node(n);
286 /* printf("\n old node: "); DDMSG2(n);
287 printf(" new node: "); DDMSG2(nn);
288 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
291 /* Don't copy Bad nodes. */
293 irn_arity = get_irn_arity(n);
294 for (i = 0; i < irn_arity; i++)
295 if (! is_Bad(get_irn_n(n, i))) {
296 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
297 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
300 /* repair the block visited flag from above misuse. Repair it in both
301 graphs so that the old one can still be used. */
302 set_Block_block_visited(nn, 0);
303 set_Block_block_visited(n, 0);
304 /* Local optimization could not merge two subsequent blocks if
305 in array contained Bads. Now it's possible.
306 We don't call optimize_in_place as it requires
307 that the fields in ir_graph are set properly. */
308 if ((get_opt_control_flow_straightening()) &&
309 (get_Block_n_cfgpreds(nn) == 1) &&
310 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
311 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
313 /* Jmp jumps into the block it is in -- deal self cycle. */
314 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
315 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
320 } else if (get_irn_op(n) == op_Phi) {
321 /* Don't copy node if corresponding predecessor in block is Bad.
322 The Block itself should not be Bad. */
323 block = get_nodes_block(n);
324 set_irn_n (nn, -1, get_new_node(block));
326 irn_arity = get_irn_arity(n);
327 for (i = 0; i < irn_arity; i++)
328 if (! is_Bad(get_irn_n(block, i))) {
329 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
330 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
333 /* If the pre walker reached this Phi after the post walker visited the
334 block block_visited is > 0. */
335 set_Block_block_visited(get_nodes_block(n), 0);
336 /* Compacting the Phi's ins might generate Phis with only one
338 if (get_irn_arity(nn) == 1)
339 exchange(nn, get_irn_n(nn, 0));
341 irn_arity = get_irn_arity(n);
342 for (i = -1; i < irn_arity; i++)
343 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
345 /* Now the new node is complete. We can add it to the hash table for CSE.
346 @@@ inlinening aborts if we identify End. Why? */
347 if (get_irn_op(nn) != op_End)
348 add_identities (current_ir_graph->value_table, nn);
352 * Copies the graph recursively, compacts the keepalive of the end node.
354 * @param copy_node_nr If non-zero, the node number will be copied
357 copy_graph (int copy_node_nr) {
358 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
359 ir_node *ka; /* keep alive */
362 oe = get_irg_end(current_ir_graph);
363 /* copy the end node by hand, allocate dynamic in array! */
364 ne = new_ir_node(get_irn_dbg_info(oe),
371 /* Copy the attributes. Well, there might be some in the future... */
372 copy_node_attr(oe, ne);
373 set_new_node(oe, ne);
375 /* copy the Bad node */
376 ob = get_irg_bad(current_ir_graph);
377 nb = new_ir_node(get_irn_dbg_info(ob),
384 set_new_node(ob, nb);
386 /* copy the NoMem node */
387 om = get_irg_no_mem(current_ir_graph);
388 nm = new_ir_node(get_irn_dbg_info(om),
395 set_new_node(om, nm);
397 /* copy the live nodes */
398 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
399 /* copy_preds for the end node ... */
400 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
402 /*- ... and now the keep alives. -*/
403 /* First pick the not marked block nodes and walk them. We must pick these
404 first as else we will oversee blocks reachable from Phis. */
405 irn_arity = get_irn_arity(oe);
406 for (i = 0; i < irn_arity; i++) {
407 ka = get_irn_intra_n(oe, i);
408 if ((get_irn_op(ka) == op_Block) &&
409 (get_irn_visited(ka) < get_irg_visited(current_ir_graph))) {
410 /* We must keep the block alive and copy everything reachable */
411 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
412 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
413 add_End_keepalive(ne, get_new_node(ka));
417 /* Now pick the Phis. Here we will keep all! */
418 irn_arity = get_irn_arity(oe);
419 for (i = 0; i < irn_arity; i++) {
420 ka = get_irn_intra_n(oe, i);
421 if ((get_irn_op(ka) == op_Phi)) {
422 if (get_irn_visited(ka) < get_irg_visited(current_ir_graph)) {
423 /* We didn't copy the Phi yet. */
424 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
425 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
427 add_End_keepalive(ne, get_new_node(ka));
431 /* start block sometimes only reached after keep alives */
432 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
433 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
437 * Copies the graph reachable from current_ir_graph->end to the obstack
438 * in current_ir_graph and fixes the environment.
439 * Then fixes the fields in current_ir_graph containing nodes of the
442 * @param copy_node_nr If non-zero, the node number will be copied
445 copy_graph_env (int copy_node_nr) {
447 /* Not all nodes remembered in current_ir_graph might be reachable
448 from the end node. Assure their link is set to NULL, so that
449 we can test whether new nodes have been computed. */
450 set_irn_link(get_irg_frame (current_ir_graph), NULL);
451 set_irn_link(get_irg_globals (current_ir_graph), NULL);
452 set_irn_link(get_irg_args (current_ir_graph), NULL);
453 set_irn_link(get_irg_initial_mem(current_ir_graph), NULL);
454 set_irn_link(get_irg_no_mem (current_ir_graph), NULL);
456 /* we use the block walk flag for removing Bads from Blocks ins. */
457 inc_irg_block_visited(current_ir_graph);
460 copy_graph(copy_node_nr);
462 /* fix the fields in current_ir_graph */
463 old_end = get_irg_end(current_ir_graph);
464 set_irg_end (current_ir_graph, get_new_node(old_end));
465 set_irg_end_except (current_ir_graph, get_irg_end(current_ir_graph));
466 set_irg_end_reg (current_ir_graph, get_irg_end(current_ir_graph));
468 set_irg_end_block (current_ir_graph, get_new_node(get_irg_end_block(current_ir_graph)));
469 if (get_irn_link(get_irg_frame(current_ir_graph)) == NULL) {
470 copy_node (get_irg_frame(current_ir_graph), INT_TO_PTR(copy_node_nr));
471 copy_preds(get_irg_frame(current_ir_graph), NULL);
473 if (get_irn_link(get_irg_globals(current_ir_graph)) == NULL) {
474 copy_node (get_irg_globals(current_ir_graph), INT_TO_PTR(copy_node_nr));
475 copy_preds(get_irg_globals(current_ir_graph), NULL);
477 if (get_irn_link(get_irg_initial_mem(current_ir_graph)) == NULL) {
478 copy_node (get_irg_initial_mem(current_ir_graph), INT_TO_PTR(copy_node_nr));
479 copy_preds(get_irg_initial_mem(current_ir_graph), NULL);
481 if (get_irn_link(get_irg_args(current_ir_graph)) == NULL) {
482 copy_node (get_irg_args(current_ir_graph), INT_TO_PTR(copy_node_nr));
483 copy_preds(get_irg_args(current_ir_graph), NULL);
485 set_irg_start (current_ir_graph, get_new_node(get_irg_start(current_ir_graph)));
487 set_irg_start_block(current_ir_graph,
488 get_new_node(get_irg_start_block(current_ir_graph)));
489 set_irg_frame (current_ir_graph, get_new_node(get_irg_frame(current_ir_graph)));
490 set_irg_globals (current_ir_graph, get_new_node(get_irg_globals(current_ir_graph)));
491 set_irg_initial_mem(current_ir_graph, get_new_node(get_irg_initial_mem(current_ir_graph)));
492 set_irg_args (current_ir_graph, get_new_node(get_irg_args(current_ir_graph)));
494 if (get_irn_link(get_irg_bad(current_ir_graph)) == NULL) {
495 copy_node(get_irg_bad(current_ir_graph), INT_TO_PTR(copy_node_nr));
496 copy_preds(get_irg_bad(current_ir_graph), NULL);
498 set_irg_bad(current_ir_graph, get_new_node(get_irg_bad(current_ir_graph)));
500 if (get_irn_link(get_irg_no_mem(current_ir_graph)) == NULL) {
501 copy_node(get_irg_no_mem(current_ir_graph), INT_TO_PTR(copy_node_nr));
502 copy_preds(get_irg_no_mem(current_ir_graph), NULL);
504 set_irg_no_mem(current_ir_graph, get_new_node(get_irg_no_mem(current_ir_graph)));
508 * Copies all reachable nodes to a new obstack. Removes bad inputs
509 * from block nodes and the corresponding inputs from Phi nodes.
510 * Merges single exit blocks with single entry blocks and removes
512 * Adds all new nodes to a new hash table for CSE. Does not
513 * perform CSE, so the hash table might contain common subexpressions.
516 dead_node_elimination(ir_graph *irg) {
518 int rem_ipview = get_interprocedural_view();
519 struct obstack *graveyard_obst = NULL;
520 struct obstack *rebirth_obst = NULL;
522 edges_init_graph(irg);
524 /* inform statistics that we started a dead-node elimination run */
525 hook_dead_node_elim(irg, 1);
527 /* Remember external state of current_ir_graph. */
528 rem = current_ir_graph;
529 current_ir_graph = irg;
530 set_interprocedural_view(0);
532 /* Handle graph state */
533 assert(get_irg_phase_state(current_ir_graph) != phase_building);
534 free_callee_info(current_ir_graph);
535 free_irg_outs(current_ir_graph);
537 /* @@@ so far we loose loops when copying */
538 free_loop_information(current_ir_graph);
540 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
542 /* A quiet place, where the old obstack can rest in peace,
543 until it will be cremated. */
544 graveyard_obst = irg->obst;
546 /* A new obstack, where the reachable nodes will be copied to. */
547 rebirth_obst = xmalloc (sizeof(*rebirth_obst));
548 current_ir_graph->obst = rebirth_obst;
549 obstack_init (current_ir_graph->obst);
551 /* We also need a new hash table for cse */
552 del_identities (irg->value_table);
553 irg->value_table = new_identities ();
555 /* Copy the graph from the old to the new obstack */
558 /* Free memory from old unoptimized obstack */
559 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
560 xfree (graveyard_obst); /* ... then free it. */
563 /* inform statistics that the run is over */
564 hook_dead_node_elim(irg, 0);
566 current_ir_graph = rem;
567 set_interprocedural_view(rem_ipview);
571 * Relink bad predecessors of a block and store the old in array to the
572 * link field. This function is called by relink_bad_predecessors().
573 * The array of link field starts with the block operand at position 0.
574 * If block has bad predecessors, create a new in array without bad preds.
575 * Otherwise let in array untouched.
577 static void relink_bad_block_predecessors(ir_node *n, void *env) {
578 ir_node **new_in, *irn;
579 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
581 /* if link field of block is NULL, look for bad predecessors otherwise
582 this is already done */
583 if (get_irn_op(n) == op_Block &&
584 get_irn_link(n) == NULL) {
586 /* save old predecessors in link field (position 0 is the block operand)*/
587 set_irn_link(n, get_irn_in(n));
589 /* count predecessors without bad nodes */
590 old_irn_arity = get_irn_arity(n);
591 for (i = 0; i < old_irn_arity; i++)
592 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
594 /* arity changing: set new predecessors without bad nodes */
595 if (new_irn_arity < old_irn_arity) {
596 /* Get new predecessor array. We do not resize the array, as we must
597 keep the old one to update Phis. */
598 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
600 /* set new predecessors in array */
603 for (i = 0; i < old_irn_arity; i++) {
604 irn = get_irn_n(n, i);
606 new_in[new_irn_n] = irn;
607 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
611 //ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity);
612 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
615 } /* ir node has bad predecessors */
617 } /* Block is not relinked */
621 * Relinks Bad predecessors from Blocks and Phis called by walker
622 * remove_bad_predecesors(). If n is a Block, call
623 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
624 * function of Phi's Block. If this block has bad predecessors, relink preds
627 static void relink_bad_predecessors(ir_node *n, void *env) {
628 ir_node *block, **old_in;
629 int i, old_irn_arity, new_irn_arity;
631 /* relink bad predecessors of a block */
632 if (get_irn_op(n) == op_Block)
633 relink_bad_block_predecessors(n, env);
635 /* If Phi node relink its block and its predecessors */
636 if (get_irn_op(n) == op_Phi) {
638 /* Relink predecessors of phi's block */
639 block = get_nodes_block(n);
640 if (get_irn_link(block) == NULL)
641 relink_bad_block_predecessors(block, env);
643 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
644 old_irn_arity = ARR_LEN(old_in);
646 /* Relink Phi predecessors if count of predecessors changed */
647 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
648 /* set new predecessors in array
649 n->in[0] remains the same block */
651 for(i = 1; i < old_irn_arity; i++)
652 if (!is_Bad((ir_node *)old_in[i])) {
653 n->in[new_irn_arity] = n->in[i];
654 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
658 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
659 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
662 } /* n is a Phi node */
666 * Removes Bad Bad predecessors from Blocks and the corresponding
667 * inputs to Phi nodes as in dead_node_elimination but without
669 * On walking up set the link field to NULL, on walking down call
670 * relink_bad_predecessors() (This function stores the old in array
671 * to the link field and sets a new in array if arity of predecessors
674 void remove_bad_predecessors(ir_graph *irg) {
675 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
679 /*--------------------------------------------------------------------*/
680 /* Functionality for inlining */
681 /*--------------------------------------------------------------------*/
684 * Copy node for inlineing. Updates attributes that change when
685 * inlineing but not for dead node elimination.
687 * Copies the node by calling copy_node() and then updates the entity if
688 * it's a local one. env must be a pointer of the frame type of the
689 * inlined procedure. The new entities must be in the link field of
693 copy_node_inline (ir_node *n, void *env) {
695 type *frame_tp = (type *)env;
698 if (get_irn_op(n) == op_Sel) {
699 new = get_new_node (n);
700 assert(get_irn_op(new) == op_Sel);
701 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
702 set_Sel_entity(new, get_entity_link(get_Sel_entity(n)));
704 } else if (get_irn_op(n) == op_Block) {
705 new = get_new_node (n);
706 new->attr.block.irg = current_ir_graph;
710 static void find_addr(ir_node *node, void *env)
712 if (get_irn_opcode(node) == iro_Proj) {
713 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
719 * currently, we cannot inline two cases:
720 * - call with compound arguments
721 * - graphs that take the address of a parameter
723 * check these conditions here
725 static int can_inline(ir_node *call, ir_graph *called_graph)
727 type *call_type = get_Call_type(call);
728 int params, ress, i, res;
729 assert(is_Method_type(call_type));
731 params = get_method_n_params(call_type);
732 ress = get_method_n_ress(call_type);
735 for (i = 0; i < params; ++i) {
736 type *p_type = get_method_param_type(call_type, i);
738 if (is_compound_type(p_type))
743 for (i = 0; i < ress; ++i) {
744 type *r_type = get_method_res_type(call_type, i);
746 if (is_compound_type(r_type))
751 irg_walk_graph(called_graph, find_addr, NULL, &res);
756 int inline_method(ir_node *call, ir_graph *called_graph) {
758 ir_node *post_call, *post_bl;
760 ir_node *end, *end_bl;
764 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
767 irg_inline_property prop = get_irg_inline_property(called_graph);
769 if ( (prop != irg_inline_forced) &&
770 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
772 /* Do not inline variadic functions. */
773 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
776 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
777 get_method_n_params(get_Call_type(call)));
780 * currently, we cannot inline two cases:
781 * - call with compound arguments
782 * - graphs that take the address of a parameter
784 if (! can_inline(call, called_graph))
787 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
788 rem_opt = get_opt_optimize();
791 /* Handle graph state */
792 assert(get_irg_phase_state(current_ir_graph) != phase_building);
793 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
794 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
795 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
796 set_irg_outs_inconsistent(current_ir_graph);
797 set_irg_loopinfo_inconsistent(current_ir_graph);
798 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
800 /* -- Check preconditions -- */
801 assert(get_irn_op(call) == op_Call);
802 /* @@@ does not work for InterfaceIII.java after cgana
803 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
804 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
805 get_Call_type(call)));
807 assert(get_type_tpop(get_Call_type(call)) == type_method);
808 if (called_graph == current_ir_graph) {
809 set_optimize(rem_opt);
813 /* here we know we WILL inline, so inform the statistics */
814 hook_inline(call, called_graph);
816 /* -- Decide how to handle exception control flow: Is there a handler
817 for the Call node, or do we branch directly to End on an exception?
819 0 There is a handler.
821 2 Exception handling not represented in Firm. -- */
823 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
824 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
825 assert(get_irn_op(proj) == op_Proj);
826 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
827 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
829 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
830 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
831 else { exc_handling = 2; } /* !Mproj && !Xproj */
836 the procedure and later replaces the Start node of the called graph.
837 Post_call is the old Call node and collects the results of the called
838 graph. Both will end up being a tuple. -- */
839 post_bl = get_nodes_block(call);
840 set_irg_current_block(current_ir_graph, post_bl);
841 /* XxMxPxP of Start + parameter of Call */
842 in[pn_Start_X_initial_exec] = new_Jmp();
843 in[pn_Start_M] = get_Call_mem(call);
844 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
845 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
846 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
847 /* in[pn_Start_P_value_arg_base] = ??? */
848 pre_call = new_Tuple(5, in);
852 The new block gets the ins of the old block, pre_call and all its
853 predecessors and all Phi nodes. -- */
854 part_block(pre_call);
856 /* -- Prepare state for dead node elimination -- */
857 /* Visited flags in calling irg must be >= flag in called irg.
858 Else walker and arity computation will not work. */
859 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
860 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
861 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
862 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
863 /* Set pre_call as new Start node in link field of the start node of
864 calling graph and pre_calls block as new block for the start block
866 Further mark these nodes so that they are not visited by the
868 set_irn_link(get_irg_start(called_graph), pre_call);
869 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
870 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
871 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
872 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
873 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
875 /* Initialize for compaction of in arrays */
876 inc_irg_block_visited(current_ir_graph);
878 /* -- Replicate local entities of the called_graph -- */
879 /* copy the entities. */
880 called_frame = get_irg_frame_type(called_graph);
881 for (i = 0; i < get_class_n_members(called_frame); i++) {
882 entity *new_ent, *old_ent;
883 old_ent = get_class_member(called_frame, i);
884 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
885 set_entity_link(old_ent, new_ent);
888 /* visited is > than that of called graph. With this trick visited will
889 remain unchanged so that an outer walker, e.g., searching the call nodes
890 to inline, calling this inline will not visit the inlined nodes. */
891 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
893 /* -- Performing dead node elimination inlines the graph -- */
894 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
896 /* @@@ endless loops are not copied!! -- they should be, I think... */
897 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
898 get_irg_frame_type(called_graph));
900 /* Repair called_graph */
901 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
902 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
903 set_Block_block_visited(get_irg_start_block(called_graph), 0);
905 /* -- Merge the end of the inlined procedure with the call site -- */
906 /* We will turn the old Call node into a Tuple with the following
909 0: Phi of all Memories of Return statements.
910 1: Jmp from new Block that merges the control flow from all exception
911 predecessors of the old end block.
912 2: Tuple of all arguments.
913 3: Phi of Exception memories.
914 In case the old Call directly branches to End on an exception we don't
915 need the block merging all exceptions nor the Phi of the exception
919 /* -- Precompute some values -- */
920 end_bl = get_new_node(get_irg_end_block(called_graph));
921 end = get_new_node(get_irg_end(called_graph));
922 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
923 n_res = get_method_n_ress(get_Call_type(call));
925 res_pred = xmalloc (n_res * sizeof(*res_pred));
926 cf_pred = xmalloc (arity * sizeof(*res_pred));
928 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
930 /* -- archive keepalives -- */
931 irn_arity = get_irn_arity(end);
932 for (i = 0; i < irn_arity; i++)
933 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
935 /* The new end node will die. We need not free as the in array is on the obstack:
936 copy_node() only generated 'D' arrays. */
938 /* -- Replace Return nodes by Jump nodes. -- */
940 for (i = 0; i < arity; i++) {
942 ret = get_irn_n(end_bl, i);
943 if (get_irn_op(ret) == op_Return) {
944 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
948 set_irn_in(post_bl, n_ret, cf_pred);
950 /* -- Build a Tuple for all results of the method.
951 Add Phi node if there was more than one Return. -- */
952 turn_into_tuple(post_call, 4);
953 /* First the Memory-Phi */
955 for (i = 0; i < arity; i++) {
956 ret = get_irn_n(end_bl, i);
957 if (get_irn_op(ret) == op_Return) {
958 cf_pred[n_ret] = get_Return_mem(ret);
962 phi = new_Phi(n_ret, cf_pred, mode_M);
963 set_Tuple_pred(call, pn_Call_M_regular, phi);
964 /* Conserve Phi-list for further inlinings -- but might be optimized */
965 if (get_nodes_block(phi) == post_bl) {
966 set_irn_link(phi, get_irn_link(post_bl));
967 set_irn_link(post_bl, phi);
969 /* Now the real results */
971 for (j = 0; j < n_res; j++) {
973 for (i = 0; i < arity; i++) {
974 ret = get_irn_n(end_bl, i);
975 if (get_irn_op(ret) == op_Return) {
976 cf_pred[n_ret] = get_Return_res(ret, j);
981 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
985 /* Conserve Phi-list for further inlinings -- but might be optimized */
986 if (get_nodes_block(phi) == post_bl) {
987 set_irn_link(phi, get_irn_link(post_bl));
988 set_irn_link(post_bl, phi);
991 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
993 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
995 /* Finally the exception control flow.
996 We have two (three) possible situations:
997 First if the Call branches to an exception handler: We need to add a Phi node to
998 collect the memory containing the exception objects. Further we need
999 to add another block to get a correct representation of this Phi. To
1000 this block we add a Jmp that resolves into the X output of the Call
1001 when the Call is turned into a tuple.
1002 Second the Call branches to End, the exception is not handled. Just
1003 add all inlined exception branches to the End node.
1004 Third: there is no Exception edge at all. Handle as case two. */
1005 if (exc_handling == 0) {
1007 for (i = 0; i < arity; i++) {
1009 ret = get_irn_n(end_bl, i);
1010 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1011 cf_pred[n_exc] = ret;
1016 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1017 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1018 /* The Phi for the memories with the exception objects */
1020 for (i = 0; i < arity; i++) {
1022 ret = skip_Proj(get_irn_n(end_bl, i));
1023 if (get_irn_op(ret) == op_Call) {
1024 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1026 } else if (is_fragile_op(ret)) {
1027 /* We rely that all cfops have the memory output at the same position. */
1028 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1030 } else if (get_irn_op(ret) == op_Raise) {
1031 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1035 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1037 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1038 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1041 ir_node *main_end_bl;
1042 int main_end_bl_arity;
1043 ir_node **end_preds;
1045 /* assert(exc_handling == 1 || no exceptions. ) */
1047 for (i = 0; i < arity; i++) {
1048 ir_node *ret = get_irn_n(end_bl, i);
1050 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1051 cf_pred[n_exc] = ret;
1055 main_end_bl = get_irg_end_block(current_ir_graph);
1056 main_end_bl_arity = get_irn_arity(main_end_bl);
1057 end_preds = xmalloc ((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1059 for (i = 0; i < main_end_bl_arity; ++i)
1060 end_preds[i] = get_irn_n(main_end_bl, i);
1061 for (i = 0; i < n_exc; ++i)
1062 end_preds[main_end_bl_arity + i] = cf_pred[i];
1063 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1064 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1065 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1071 #if 0 /* old. now better, correcter, faster implementation. */
1073 /* -- If the exception control flow from the inlined Call directly
1074 branched to the end block we now have the following control
1075 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
1076 remove the Jmp along with it's empty block and add Jmp's
1077 predecessors as predecessors of this end block. No problem if
1078 there is no exception, because then branches Bad to End which
1080 @@@ can't we know this beforehand: by getting the Proj(1) from
1081 the Call link list and checking whether it goes to Proj. */
1082 /* find the problematic predecessor of the end block. */
1083 end_bl = get_irg_end_block(current_ir_graph);
1084 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
1085 cf_op = get_Block_cfgpred(end_bl, i);
1086 if (get_irn_op(cf_op) == op_Proj) {
1087 cf_op = get_Proj_pred(cf_op);
1088 if ((get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
1089 /* There are unoptimized tuples from inlineing before when no exc */
1090 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
1091 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
1092 assert(get_irn_op(cf_op) == op_Jmp);
1098 if (i < get_Block_n_cfgpreds(end_bl)) {
1099 bl = get_nodes_block(cf_op);
1100 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
1101 cf_pred = xmalloc (arity * sizeof(*cf_pred));
1102 for (j = 0; j < i; j++)
1103 cf_pred[j] = get_Block_cfgpred(end_bl, j);
1104 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
1105 cf_pred[j] = get_Block_cfgpred(bl, j-i);
1106 for (j = j; j < arity; j++)
1107 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
1108 set_irn_in(end_bl, arity, cf_pred);
1110 /* Remove the exception pred from post-call Tuple. */
1111 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1116 /* -- Turn CSE back on. -- */
1117 set_optimize(rem_opt);
1122 /********************************************************************/
1123 /* Apply inlineing to small methods. */
1124 /********************************************************************/
1126 /* It makes no sense to inline too many calls in one procedure. Anyways,
1127 I didn't get a version with NEW_ARR_F to run. */
1128 #define MAX_INLINE 1024
1131 * environment for inlining small irgs
1133 typedef struct _inline_env_t {
1135 ir_node *calls[MAX_INLINE];
1139 * Returns the irg called from a Call node. If the irg is not
1140 * known, NULL is returned.
1142 static ir_graph *get_call_called_irg(ir_node *call) {
1144 ir_graph *called_irg = NULL;
1146 assert(get_irn_op(call) == op_Call);
1148 addr = get_Call_ptr(call);
1149 if ((get_irn_op(addr) == op_SymConst) && (get_SymConst_kind (addr) == symconst_addr_ent)) {
1150 called_irg = get_entity_irg(get_SymConst_entity(addr));
1156 static void collect_calls(ir_node *call, void *env) {
1159 if (get_irn_op(call) != op_Call) return;
1161 addr = get_Call_ptr(call);
1163 if (get_irn_op(addr) == op_SymConst) {
1164 if (get_SymConst_kind(addr) == symconst_addr_ent) {
1165 ir_graph *called_irg = get_entity_irg(get_SymConst_entity(addr));
1166 inline_env_t *ienv = (inline_env_t *)env;
1167 if (called_irg && ienv->pos < MAX_INLINE) {
1168 /* The Call node calls a locally defined method. Remember to inline. */
1169 ienv->calls[ienv->pos++] = call;
1176 * Inlines all small methods at call sites where the called address comes
1177 * from a Const node that references the entity representing the called
1179 * The size argument is a rough measure for the code size of the method:
1180 * Methods where the obstack containing the firm graph is smaller than
1183 void inline_small_irgs(ir_graph *irg, int size) {
1185 ir_graph *rem = current_ir_graph;
1186 inline_env_t env /* = {0, NULL}*/;
1188 if (!(get_opt_optimize() && get_opt_inline())) return;
1190 current_ir_graph = irg;
1191 /* Handle graph state */
1192 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1193 free_callee_info(current_ir_graph);
1195 /* Find Call nodes to inline.
1196 (We can not inline during a walk of the graph, as inlineing the same
1197 method several times changes the visited flag of the walked graph:
1198 after the first inlineing visited of the callee equals visited of
1199 the caller. With the next inlineing both are increased.) */
1201 irg_walk(get_irg_end(irg), NULL, collect_calls, &env);
1203 if ((env.pos > 0) && (env.pos < MAX_INLINE)) {
1204 /* There are calls to inline */
1205 collect_phiprojs(irg);
1206 for (i = 0; i < env.pos; i++) {
1208 callee = get_entity_irg(get_SymConst_entity(get_Call_ptr(env.calls[i])));
1209 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1210 (get_irg_inline_property(callee) == irg_inline_forced)) {
1211 inline_method(env.calls[i], callee);
1216 current_ir_graph = rem;
1220 * Environment for inlining irgs.
1223 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1224 int n_nodes_orig; /**< for statistics */
1225 eset *call_nodes; /**< All call nodes in this graph */
1227 int n_call_nodes_orig; /**< for statistics */
1228 int n_callers; /**< Number of known graphs that call this graphs. */
1229 int n_callers_orig; /**< for statistics */
1233 * Allocate a new environment for inlining.
1235 static inline_irg_env *new_inline_irg_env(void) {
1236 inline_irg_env *env = xmalloc(sizeof(*env));
1237 env->n_nodes = -2; /* do not count count Start, End */
1238 env->n_nodes_orig = -2; /* do not count Start, End */
1239 env->call_nodes = eset_create();
1240 env->n_call_nodes = 0;
1241 env->n_call_nodes_orig = 0;
1243 env->n_callers_orig = 0;
1248 * destroy an environment for inlining.
1250 static void free_inline_irg_env(inline_irg_env *env) {
1251 eset_destroy(env->call_nodes);
1256 * post-walker: collect all calls in the inline-environment
1257 * of a graph and sum some statistics.
1259 static void collect_calls2(ir_node *call, void *env) {
1260 inline_irg_env *x = (inline_irg_env *)env;
1261 ir_op *op = get_irn_op(call);
1264 /* count meaningful nodes in irg */
1265 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1270 if (op != op_Call) return;
1272 /* collect all call nodes */
1273 eset_insert(x->call_nodes, call);
1275 x->n_call_nodes_orig++;
1277 /* count all static callers */
1278 callee = get_call_called_irg(call);
1280 inline_irg_env *callee_env = get_irg_link(callee);
1281 callee_env->n_callers++;
1282 callee_env->n_callers_orig++;
1287 * Returns TRUE if the number of callers in 0 in the irg's environment,
1288 * hence this irg is a leave.
1290 INLINE static int is_leave(ir_graph *irg) {
1291 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1295 * Returns TRUE if the number of callers is smaller size in the irg's environment.
1297 INLINE static int is_smaller(ir_graph *callee, int size) {
1298 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1303 * Inlines small leave methods at call sites where the called address comes
1304 * from a Const node that references the entity representing the called
1306 * The size argument is a rough measure for the code size of the method:
1307 * Methods where the obstack containing the firm graph is smaller than
1310 void inline_leave_functions(int maxsize, int leavesize, int size) {
1311 inline_irg_env *env;
1312 int i, n_irgs = get_irp_n_irgs();
1313 ir_graph *rem = current_ir_graph;
1316 if (!(get_opt_optimize() && get_opt_inline())) return;
1318 /* extend all irgs by a temporary data structure for inlining. */
1319 for (i = 0; i < n_irgs; ++i)
1320 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1322 /* Precompute information in temporary data structure. */
1323 for (i = 0; i < n_irgs; ++i) {
1324 current_ir_graph = get_irp_irg(i);
1325 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1326 free_callee_info(current_ir_graph);
1328 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1329 get_irg_link(current_ir_graph));
1332 /* -- and now inline. -- */
1334 /* Inline leaves recursively -- we might construct new leaves. */
1335 while (did_inline) {
1338 for (i = 0; i < n_irgs; ++i) {
1340 int phiproj_computed = 0;
1342 current_ir_graph = get_irp_irg(i);
1343 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1345 for (call = eset_first(env->call_nodes); call; call = eset_next(env->call_nodes)) {
1348 if (get_irn_op(call) == op_Tuple) continue; /* We already have inlined this call. */
1349 callee = get_call_called_irg(call);
1351 if (env->n_nodes > maxsize) continue; // break;
1353 if (callee && (is_leave(callee) && is_smaller(callee, leavesize))) {
1354 if (!phiproj_computed) {
1355 phiproj_computed = 1;
1356 collect_phiprojs(current_ir_graph);
1358 did_inline = inline_method(call, callee);
1361 /* Do some statistics */
1362 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1363 env->n_call_nodes --;
1364 env->n_nodes += callee_env->n_nodes;
1365 callee_env->n_callers--;
1372 /* inline other small functions. */
1373 for (i = 0; i < n_irgs; ++i) {
1376 int phiproj_computed = 0;
1378 current_ir_graph = get_irp_irg(i);
1379 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1381 /* we can not walk and change a set, nor remove from it.
1383 walkset = env->call_nodes;
1384 env->call_nodes = eset_create();
1385 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1388 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1389 callee = get_call_called_irg(call);
1392 ((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1393 (get_irg_inline_property(callee) == irg_inline_forced))) {
1394 if (!phiproj_computed) {
1395 phiproj_computed = 1;
1396 collect_phiprojs(current_ir_graph);
1398 if (inline_method(call, callee)) {
1399 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1400 env->n_call_nodes--;
1401 eset_insert_all(env->call_nodes, callee_env->call_nodes); /* @@@ ??? This are the wrong nodes !? Not the copied ones. */
1402 env->n_call_nodes += callee_env->n_call_nodes;
1403 env->n_nodes += callee_env->n_nodes;
1404 callee_env->n_callers--;
1407 eset_insert(env->call_nodes, call);
1410 eset_destroy(walkset);
1413 for (i = 0; i < n_irgs; ++i) {
1414 current_ir_graph = get_irp_irg(i);
1416 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1417 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1418 (env->n_callers_orig != env->n_callers))
1419 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1420 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1421 env->n_callers_orig, env->n_callers,
1422 get_entity_name(get_irg_entity(current_ir_graph)));
1424 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1427 current_ir_graph = rem;
1430 /*******************************************************************/
1431 /* Code Placement. Pins all floating nodes to a block where they */
1432 /* will be executed only if needed. */
1433 /*******************************************************************/
1436 * Returns non-zero, is a block is not reachable from Start.
1438 * @param block the block to test
1441 is_Block_unreachable(ir_node *block) {
1442 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
1446 * Find the earliest correct block for N. --- Place N into the
1447 * same Block as its dominance-deepest Input.
1449 * We have to avoid calls to get_nodes_block() here
1450 * because the graph is floating.
1452 * move_out_of_loops() expects that place_floats_early() have placed
1453 * all "living" nodes into a living block. That's why we must
1454 * move nodes in dead block with "live" successors into a valid
1456 * We move them just into the same block as it's successor (or
1457 * in case of a Phi into the effective use block). For Phi successors,
1458 * this may still be a dead block, but then there is no real use, as
1459 * the control flow will be dead later.
1462 place_floats_early(ir_node *n, pdeq *worklist)
1466 /* we must not run into an infinite loop */
1467 assert(irn_not_visited(n));
1468 mark_irn_visited(n);
1470 /* Place floating nodes. */
1471 if (get_irn_pinned(n) == op_pin_state_floats) {
1472 ir_node *curr_block = get_irn_n(n, -1);
1473 int in_dead_block = is_Block_unreachable(curr_block);
1475 ir_node *b = NULL; /* The block to place this node in */
1477 assert(get_irn_op(n) != op_Block);
1479 if ((get_irn_op(n) == op_Const) ||
1480 (get_irn_op(n) == op_SymConst) ||
1482 (get_irn_op(n) == op_Unknown)) {
1483 /* These nodes will not be placed by the loop below. */
1484 b = get_irg_start_block(current_ir_graph);
1488 /* find the block for this node. */
1489 irn_arity = get_irn_arity(n);
1490 for (i = 0; i < irn_arity; i++) {
1491 ir_node *pred = get_irn_n(n, i);
1492 ir_node *pred_block;
1494 if ((irn_not_visited(pred))
1495 && (get_irn_pinned(pred) == op_pin_state_floats)) {
1498 * If the current node is NOT in a dead block, but one of its
1499 * predecessors is, we must move the predecessor to a live block.
1500 * Such thing can happen, if global CSE chose a node from a dead block.
1501 * We move it simple to our block.
1502 * Note that neither Phi nor End nodes are floating, so we don't
1503 * need to handle them here.
1505 if (! in_dead_block) {
1506 if (get_irn_pinned(pred) == op_pin_state_floats &&
1507 is_Block_unreachable(get_irn_n(pred, -1)))
1508 set_nodes_block(pred, curr_block);
1510 place_floats_early(pred, worklist);
1514 * A node in the Bad block must stay in the bad block,
1515 * so don't compute a new block for it.
1520 /* Because all loops contain at least one op_pin_state_pinned node, now all
1521 our inputs are either op_pin_state_pinned or place_early() has already
1522 been finished on them. We do not have any unfinished inputs! */
1523 pred_block = get_irn_n(pred, -1);
1524 if ((!is_Block_dead(pred_block)) &&
1525 (get_Block_dom_depth(pred_block) > depth)) {
1527 depth = get_Block_dom_depth(pred_block);
1529 /* Avoid that the node is placed in the Start block */
1530 if ((depth == 1) && (get_Block_dom_depth(get_irn_n(n, -1)) > 1)) {
1531 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1532 assert(b != get_irg_start_block(current_ir_graph));
1537 set_nodes_block(n, b);
1541 * Add predecessors of non floating nodes and non-floating predecessors
1542 * of floating nodes to worklist and fix their blocks if the are in dead block.
1544 irn_arity = get_irn_arity(n);
1546 if (get_irn_op(n) == op_End) {
1548 * Simplest case: End node. Predecessors are keep-alives,
1549 * no need to move out of dead block.
1551 for (i = -1; i < irn_arity; ++i) {
1552 ir_node *pred = get_irn_n(n, i);
1553 if (irn_not_visited(pred))
1554 pdeq_putr(worklist, pred);
1557 else if (is_Block(n)) {
1559 * Blocks: Predecessors are control flow, no need to move
1560 * them out of dead block.
1562 for (i = irn_arity - 1; i >= 0; --i) {
1563 ir_node *pred = get_irn_n(n, i);
1564 if (irn_not_visited(pred))
1565 pdeq_putr(worklist, pred);
1568 else if (is_Phi(n)) {
1570 ir_node *curr_block = get_irn_n(n, -1);
1571 int in_dead_block = is_Block_unreachable(curr_block);
1574 * Phi nodes: move nodes from dead blocks into the effective use
1575 * of the Phi-input if the Phi is not in a bad block.
1577 pred = get_irn_n(n, -1);
1578 if (irn_not_visited(pred))
1579 pdeq_putr(worklist, pred);
1581 for (i = irn_arity - 1; i >= 0; --i) {
1582 ir_node *pred = get_irn_n(n, i);
1584 if (irn_not_visited(pred)) {
1585 if (! in_dead_block &&
1586 get_irn_pinned(pred) == op_pin_state_floats &&
1587 is_Block_unreachable(get_irn_n(pred, -1))) {
1588 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
1590 pdeq_putr(worklist, pred);
1596 ir_node *curr_block = get_irn_n(n, -1);
1597 int in_dead_block = is_Block_unreachable(curr_block);
1600 * All other nodes: move nodes from dead blocks into the same block.
1602 pred = get_irn_n(n, -1);
1603 if (irn_not_visited(pred))
1604 pdeq_putr(worklist, pred);
1606 for (i = irn_arity - 1; i >= 0; --i) {
1607 ir_node *pred = get_irn_n(n, i);
1609 if (irn_not_visited(pred)) {
1610 if (! in_dead_block &&
1611 get_irn_pinned(pred) == op_pin_state_floats &&
1612 is_Block_unreachable(get_irn_n(pred, -1))) {
1613 set_nodes_block(pred, curr_block);
1615 pdeq_putr(worklist, pred);
1622 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1623 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1624 * places all floating nodes reachable from its argument through floating
1625 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1627 static INLINE void place_early(pdeq *worklist) {
1629 inc_irg_visited(current_ir_graph);
1631 /* this inits the worklist */
1632 place_floats_early(get_irg_end(current_ir_graph), worklist);
1634 /* Work the content of the worklist. */
1635 while (!pdeq_empty(worklist)) {
1636 ir_node *n = pdeq_getl(worklist);
1637 if (irn_not_visited(n))
1638 place_floats_early(n, worklist);
1641 set_irg_outs_inconsistent(current_ir_graph);
1642 current_ir_graph->op_pin_state_pinned = op_pin_state_pinned;
1646 * Compute the deepest common ancestor of block and dca.
1648 static ir_node *calc_dca(ir_node *dca, ir_node *block)
1652 /* we do not want to place nodes in dead blocks */
1653 if (is_Block_dead(block))
1656 /* We found a first legal placement. */
1657 if (!dca) return block;
1659 /* Find a placement that is dominates both, dca and block. */
1660 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1661 block = get_Block_idom(block);
1663 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1664 dca = get_Block_idom(dca);
1667 while (block != dca)
1668 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1673 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1674 * I.e., DCA is the block where we might place PRODUCER.
1675 * A data flow edge points from producer to consumer.
1678 consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer)
1680 ir_node *block = NULL;
1682 /* Compute the latest block into which we can place a node so that it is
1684 if (get_irn_op(consumer) == op_Phi) {
1685 /* our consumer is a Phi-node, the effective use is in all those
1686 blocks through which the Phi-node reaches producer */
1688 ir_node *phi_block = get_nodes_block(consumer);
1689 irn_arity = get_irn_arity(consumer);
1691 for (i = 0; i < irn_arity; i++) {
1692 if (get_irn_n(consumer, i) == producer) {
1693 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1695 if (! is_Block_unreachable(new_block))
1696 block = calc_dca(block, new_block);
1701 block = get_irn_n(producer, -1);
1704 assert(is_no_Block(consumer));
1705 block = get_nodes_block(consumer);
1708 /* Compute the deepest common ancestor of block and dca. */
1709 return calc_dca(dca, block);
1712 /* FIXME: the name clashes here with the function from ana/field_temperature.c
1714 static INLINE int get_irn_loop_depth(ir_node *n) {
1715 return get_loop_depth(get_irn_loop(n));
1719 * Move n to a block with less loop depth than it's current block. The
1720 * new block must be dominated by early.
1722 * @param n the node that should be moved
1723 * @param early the earliest block we can n move to
1726 move_out_of_loops (ir_node *n, ir_node *early)
1728 ir_node *best, *dca;
1732 /* Find the region deepest in the dominator tree dominating
1733 dca with the least loop nesting depth, but still dominated
1734 by our early placement. */
1735 dca = get_nodes_block(n);
1738 while (dca != early) {
1739 dca = get_Block_idom(dca);
1740 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1741 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1745 if (best != get_nodes_block(n)) {
1747 printf("Moving out of loop: "); DDMN(n);
1748 printf(" Outermost block: "); DDMN(early);
1749 printf(" Best block: "); DDMN(best);
1750 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1752 set_nodes_block(n, best);
1757 * Find the latest legal block for N and place N into the
1758 * `optimal' Block between the latest and earliest legal block.
1759 * The `optimal' block is the dominance-deepest block of those
1760 * with the least loop-nesting-depth. This places N out of as many
1761 * loops as possible and then makes it as control dependent as
1765 place_floats_late(ir_node *n, pdeq *worklist)
1770 assert(irn_not_visited(n)); /* no multiple placement */
1772 mark_irn_visited(n);
1774 /* no need to place block nodes, control nodes are already placed. */
1775 if ((get_irn_op(n) != op_Block) &&
1777 (get_irn_mode(n) != mode_X)) {
1778 /* Remember the early_blk placement of this block to move it
1779 out of loop no further than the early_blk placement. */
1780 early_blk = get_irn_n(n, -1);
1783 * BEWARE: Here we also get code, that is live, but
1784 * was in a dead block. If the node is life, but because
1785 * of CSE in a dead block, we still might need it.
1788 /* Assure that our users are all placed, except the Phi-nodes.
1789 --- Each data flow cycle contains at least one Phi-node. We
1790 have to break the `user has to be placed before the
1791 producer' dependence cycle and the Phi-nodes are the
1792 place to do so, because we need to base our placement on the
1793 final region of our users, which is OK with Phi-nodes, as they
1794 are op_pin_state_pinned, and they never have to be placed after a
1795 producer of one of their inputs in the same block anyway. */
1796 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
1797 ir_node *succ = get_irn_out(n, i);
1798 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
1799 place_floats_late(succ, worklist);
1802 if (! is_Block_dead(early_blk)) {
1803 /* do only move things that where not dead */
1805 /* We have to determine the final block of this node... except for
1807 if ((get_irn_pinned(n) == op_pin_state_floats) &&
1808 (get_irn_op(n) != op_Const) &&
1809 (get_irn_op(n) != op_SymConst)) {
1810 ir_node *dca = NULL; /* deepest common ancestor in the
1811 dominator tree of all nodes'
1812 blocks depending on us; our final
1813 placement has to dominate DCA. */
1814 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
1815 ir_node *succ = get_irn_out(n, i);
1818 if (get_irn_op(succ) == op_End) {
1820 * This consumer is the End node, a keep alive edge.
1821 * This is not a real consumer, so we ignore it
1826 /* ignore if succ is in dead code */
1827 succ_blk = get_irn_n(succ, -1);
1828 if (is_Block_unreachable(succ_blk))
1830 dca = consumer_dom_dca(dca, succ, n);
1833 set_nodes_block(n, dca);
1834 move_out_of_loops(n, early_blk);
1840 /* Add predecessors of all non-floating nodes on list. (Those of floating
1841 nodes are placed already and therefore are marked.) */
1842 for (i = 0; i < get_irn_n_outs(n); i++) {
1843 ir_node *succ = get_irn_out(n, i);
1844 if (irn_not_visited(get_irn_out(n, i))) {
1845 pdeq_putr(worklist, succ);
1850 static INLINE void place_late(pdeq *worklist) {
1852 inc_irg_visited(current_ir_graph);
1854 /* This fills the worklist initially. */
1855 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1857 /* And now empty the worklist again... */
1858 while (!pdeq_empty(worklist)) {
1859 ir_node *n = pdeq_getl(worklist);
1860 if (irn_not_visited(n))
1861 place_floats_late(n, worklist);
1865 void place_code(ir_graph *irg) {
1867 ir_graph *rem = current_ir_graph;
1869 current_ir_graph = irg;
1871 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1873 /* Handle graph state */
1874 assert(get_irg_phase_state(irg) != phase_building);
1875 if (get_irg_dom_state(irg) != dom_consistent)
1878 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1879 free_loop_information(irg);
1880 construct_backedges(irg);
1883 /* Place all floating nodes as early as possible. This guarantees
1884 a legal code placement. */
1885 worklist = new_pdeq();
1886 place_early(worklist);
1888 /* place_early() invalidates the outs, place_late needs them. */
1889 compute_irg_outs(irg);
1891 /* Now move the nodes down in the dominator tree. This reduces the
1892 unnecessary executions of the node. */
1893 place_late(worklist);
1895 set_irg_outs_inconsistent(current_ir_graph);
1896 set_irg_loopinfo_inconsistent(current_ir_graph);
1898 current_ir_graph = rem;
1902 * Called by walker of remove_critical_cf_edges().
1904 * Place an empty block to an edge between a blocks of multiple
1905 * predecessors and a block of multiple successors.
1908 * @param env Environment of walker. This field is unused and has
1911 static void walk_critical_cf_edges(ir_node *n, void *env) {
1913 ir_node *pre, *block, **in, *jmp;
1915 /* Block has multiple predecessors */
1916 if ((op_Block == get_irn_op(n)) &&
1917 (get_irn_arity(n) > 1)) {
1918 arity = get_irn_arity(n);
1920 if (n == get_irg_end_block(current_ir_graph))
1921 return; /* No use to add a block here. */
1923 for (i=0; i<arity; i++) {
1924 pre = get_irn_n(n, i);
1925 /* Predecessor has multiple successors. Insert new flow edge */
1926 if ((NULL != pre) &&
1927 (op_Proj == get_irn_op(pre)) &&
1928 op_Raise != get_irn_op(skip_Proj(pre))) {
1930 /* set predecessor array for new block */
1931 in = NEW_ARR_D (ir_node *, current_ir_graph->obst, 1);
1932 /* set predecessor of new block */
1934 block = new_Block(1, in);
1935 /* insert new jmp node to new block */
1936 set_cur_block(block);
1939 /* set successor of new block */
1940 set_irn_n(n, i, jmp);
1942 } /* predecessor has multiple successors */
1943 } /* for all predecessors */
1944 } /* n is a block */
1947 void remove_critical_cf_edges(ir_graph *irg) {
1948 if (get_opt_critical_edges())
1949 irg_walk_graph(irg, NULL, walk_critical_cf_edges, NULL);