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"
43 #include "iredges_t.h"
45 /* Defined in iropt.c */
46 pset *new_identities (void);
47 void del_identities (pset *value_table);
48 void add_identities (pset *value_table, ir_node *node);
50 /*------------------------------------------------------------------*/
51 /* apply optimizations of iropt to all nodes. */
52 /*------------------------------------------------------------------*/
54 static void init_link (ir_node *n, void *env) {
55 set_irn_link(n, NULL);
58 #if 0 /* Old version. Avoids Ids.
59 This is not necessary: we do a postwalk, and get_irn_n
60 removes ids anyways. So it's much cheaper to call the
61 optimization less often and use the exchange() algorithm. */
63 optimize_in_place_wrapper (ir_node *n, void *env) {
65 ir_node *optimized, *old;
67 irn_arity = get_irn_arity(n);
68 for (i = 0; i < irn_arity; i++) {
69 /* get_irn_n skips Id nodes, so comparison old != optimized does not
70 show all optimizations. Therefore always set new predecessor. */
71 old = get_irn_intra_n(n, i);
72 optimized = optimize_in_place_2(old);
73 set_irn_n(n, i, optimized);
76 if (get_irn_op(n) == op_Block) {
77 optimized = optimize_in_place_2(n);
78 if (optimized != n) exchange (n, optimized);
83 optimize_in_place_wrapper (ir_node *n, void *env) {
84 ir_node *optimized = optimize_in_place_2(n);
85 if (optimized != n) exchange (n, optimized);
90 static INLINE void do_local_optimize(ir_node *n) {
91 /* Handle graph state */
92 assert(get_irg_phase_state(current_ir_graph) != phase_building);
93 if (get_opt_global_cse())
94 set_irg_pinned(current_ir_graph, op_pin_state_floats);
95 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
96 set_irg_outs_inconsistent(current_ir_graph);
97 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
98 set_irg_dom_inconsistent(current_ir_graph);
99 set_irg_loopinfo_inconsistent(current_ir_graph);
102 /* Clean the value_table in irg for the cse. */
103 del_identities(current_ir_graph->value_table);
104 current_ir_graph->value_table = new_identities();
106 /* walk over the graph */
107 irg_walk(n, init_link, optimize_in_place_wrapper, NULL);
110 void local_optimize_node(ir_node *n) {
111 ir_graph *rem = current_ir_graph;
112 current_ir_graph = get_irn_irg(n);
114 do_local_optimize(n);
116 current_ir_graph = rem;
121 local_optimize_graph (ir_graph *irg) {
122 ir_graph *rem = current_ir_graph;
123 current_ir_graph = irg;
125 do_local_optimize(irg->end);
127 current_ir_graph = rem;
131 /*------------------------------------------------------------------*/
132 /* Routines for dead node elimination / copying garbage collection */
133 /* of the obstack. */
134 /*------------------------------------------------------------------*/
137 * Remember the new node in the old node by using a field all nodes have.
140 set_new_node (ir_node *old, ir_node *new)
146 * Get this new node, before the old node is forgotton.
148 static INLINE ir_node *
149 get_new_node (ir_node * n)
155 * We use the block_visited flag to mark that we have computed the
156 * number of useful predecessors for this block.
157 * Further we encode the new arity in this flag in the old blocks.
158 * Remembering the arity is useful, as it saves a lot of pointer
159 * accesses. This function is called for all Phi and Block nodes
163 compute_new_arity(ir_node *b) {
164 int i, res, irn_arity;
167 irg_v = get_irg_block_visited(current_ir_graph);
168 block_v = get_Block_block_visited(b);
169 if (block_v >= irg_v) {
170 /* we computed the number of preds for this block and saved it in the
172 return block_v - irg_v;
174 /* compute the number of good predecessors */
175 res = irn_arity = get_irn_arity(b);
176 for (i = 0; i < irn_arity; i++)
177 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
178 /* save it in the flag. */
179 set_Block_block_visited(b, irg_v + res);
184 /* TODO: add an ir_op operation */
185 static INLINE void new_backedge_info(ir_node *n) {
186 switch(get_irn_opcode(n)) {
188 n->attr.block.cg_backedge = NULL;
189 n->attr.block.backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
192 n->attr.phi_backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
195 n->attr.filter.backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
202 * Copies the node to the new obstack. The Ins of the new node point to
203 * the predecessors on the old obstack. For block/phi nodes not all
204 * predecessors might be copied. n->link points to the new node.
205 * For Phi and Block nodes the function allocates in-arrays with an arity
206 * only for useful predecessors. The arity is determined by counting
207 * the non-bad predecessors of the block.
209 * @param n The node to be copied
210 * @param env if non-NULL, the node number attribute will be copied to the new node
212 * Note: Also used for loop unrolling.
215 firm_copy_node (ir_node *n, void *env) {
218 opcode op = get_irn_opcode(n);
219 int copy_node_nr = env != NULL;
221 /* The end node looses it's flexible in array. This doesn't matter,
222 as dead node elimination builds End by hand, inlineing doesn't use
224 /* assert(n->op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
227 /* node copied already */
229 } else if (op == iro_Block) {
231 new_arity = compute_new_arity(n);
232 n->attr.block.graph_arr = NULL;
234 block = get_nodes_block(n);
235 if (get_irn_opcode(n) == iro_Phi) {
236 new_arity = compute_new_arity(block);
238 new_arity = get_irn_arity(n);
241 nn = new_ir_node(get_irn_dbg_info(n),
248 /* Copy the attributes. These might point to additional data. If this
249 was allocated on the old obstack the pointers now are dangling. This
250 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
251 copy_node_attr(n, nn);
252 new_backedge_info(nn);
257 /* for easier debugging, we want to copy the node numbers too */
258 nn->node_nr = n->node_nr;
262 /* printf("\n old node: "); DDMSG2(n);
263 printf(" new node: "); DDMSG2(nn); */
267 * Copies new predecessors of old node to new node remembered in link.
268 * Spare the Bad predecessors of Phi and Block nodes.
271 copy_preds (ir_node *n, void *env) {
275 nn = get_new_node(n);
277 /* printf("\n old node: "); DDMSG2(n);
278 printf(" new node: "); DDMSG2(nn);
279 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
281 if (get_irn_opcode(n) == iro_Block) {
282 /* Don't copy Bad nodes. */
284 irn_arity = get_irn_arity(n);
285 for (i = 0; i < irn_arity; i++)
286 if (get_irn_opcode(get_irn_n(n, i)) != iro_Bad) {
287 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
288 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
291 /* repair the block visited flag from above misuse. Repair it in both
292 graphs so that the old one can still be used. */
293 set_Block_block_visited(nn, 0);
294 set_Block_block_visited(n, 0);
295 /* Local optimization could not merge two subsequent blocks if
296 in array contained Bads. Now it's possible.
297 We don't call optimize_in_place as it requires
298 that the fields in ir_graph are set properly. */
299 if ((get_opt_control_flow_straightening()) &&
300 (get_Block_n_cfgpreds(nn) == 1) &&
301 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
302 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
304 /* Jmp jumps into the block it is in -- deal self cycle. */
305 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
306 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
311 } else if (get_irn_opcode(n) == iro_Phi) {
312 /* Don't copy node if corresponding predecessor in block is Bad.
313 The Block itself should not be Bad. */
314 block = get_nodes_block(n);
315 set_irn_n (nn, -1, get_new_node(block));
317 irn_arity = get_irn_arity(n);
318 for (i = 0; i < irn_arity; i++)
319 if (get_irn_opcode(get_irn_n(block, i)) != iro_Bad) {
320 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
321 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
324 /* If the pre walker reached this Phi after the post walker visited the
325 block block_visited is > 0. */
326 set_Block_block_visited(get_nodes_block(n), 0);
327 /* Compacting the Phi's ins might generate Phis with only one
329 if (get_irn_arity(n) == 1)
330 exchange(n, get_irn_n(n, 0));
332 irn_arity = get_irn_arity(n);
333 for (i = -1; i < irn_arity; i++)
334 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
336 /* Now the new node is complete. We can add it to the hash table for cse.
337 @@@ inlinening aborts if we identify End. Why? */
338 if(get_irn_op(nn) != op_End)
339 add_identities (current_ir_graph->value_table, nn);
343 * Copies the graph recursively, compacts the keepalive of the end node.
345 * @param copy_node_nr If non-zero, the node number will be copied
348 copy_graph (int copy_node_nr) {
349 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
350 ir_node *ka; /* keep alive */
353 oe = get_irg_end(current_ir_graph);
354 /* copy the end node by hand, allocate dynamic in array! */
355 ne = new_ir_node(get_irn_dbg_info(oe),
362 /* Copy the attributes. Well, there might be some in the future... */
363 copy_node_attr(oe, ne);
364 set_new_node(oe, ne);
366 /* copy the Bad node */
367 ob = get_irg_bad(current_ir_graph);
368 nb = new_ir_node(get_irn_dbg_info(ob),
375 set_new_node(ob, nb);
377 /* copy the NoMem node */
378 om = get_irg_no_mem(current_ir_graph);
379 nm = new_ir_node(get_irn_dbg_info(om),
386 set_new_node(om, nm);
388 /* copy the live nodes */
389 irg_walk(get_nodes_block(oe), firm_copy_node, copy_preds, (void *)copy_node_nr);
390 /* copy_preds for the end node ... */
391 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
393 /*- ... and now the keep alives. -*/
394 /* First pick the not marked block nodes and walk them. We must pick these
395 first as else we will oversee blocks reachable from Phis. */
396 irn_arity = get_irn_arity(oe);
397 for (i = 0; i < irn_arity; i++) {
398 ka = get_irn_intra_n(oe, i);
399 if ((get_irn_op(ka) == op_Block) &&
400 (get_irn_visited(ka) < get_irg_visited(current_ir_graph))) {
401 /* We must keep the block alive and copy everything reachable */
402 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
403 irg_walk(ka, firm_copy_node, copy_preds, (void *)copy_node_nr);
404 add_End_keepalive(ne, get_new_node(ka));
408 /* Now pick the Phis. Here we will keep all! */
409 irn_arity = get_irn_arity(oe);
410 for (i = 0; i < irn_arity; i++) {
411 ka = get_irn_intra_n(oe, i);
412 if ((get_irn_op(ka) == op_Phi)) {
413 if (get_irn_visited(ka) < get_irg_visited(current_ir_graph)) {
414 /* We didn't copy the Phi yet. */
415 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
416 irg_walk(ka, firm_copy_node, copy_preds, (void *)copy_node_nr);
418 add_End_keepalive(ne, get_new_node(ka));
422 /* start block sometimes only reached after keep alives */
423 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
424 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
428 * Copies the graph reachable from current_ir_graph->end to the obstack
429 * in current_ir_graph and fixes the environment.
430 * Then fixes the fields in current_ir_graph containing nodes of the
433 * @param copy_node_nr If non-zero, the node number will be copied
436 copy_graph_env (int copy_node_nr) {
438 /* Not all nodes remembered in current_ir_graph might be reachable
439 from the end node. Assure their link is set to NULL, so that
440 we can test whether new nodes have been computed. */
441 set_irn_link(get_irg_frame (current_ir_graph), NULL);
442 set_irn_link(get_irg_globals (current_ir_graph), NULL);
443 set_irn_link(get_irg_args (current_ir_graph), NULL);
444 set_irn_link(get_irg_initial_mem(current_ir_graph), NULL);
445 set_irn_link(get_irg_no_mem (current_ir_graph), NULL);
447 /* we use the block walk flag for removing Bads from Blocks ins. */
448 inc_irg_block_visited(current_ir_graph);
451 copy_graph(copy_node_nr);
453 /* fix the fields in current_ir_graph */
454 old_end = get_irg_end(current_ir_graph);
455 set_irg_end (current_ir_graph, get_new_node(old_end));
456 set_irg_end_except (current_ir_graph, get_irg_end(current_ir_graph));
457 set_irg_end_reg (current_ir_graph, get_irg_end(current_ir_graph));
459 set_irg_end_block (current_ir_graph, get_new_node(get_irg_end_block(current_ir_graph)));
460 if (get_irn_link(get_irg_frame(current_ir_graph)) == NULL) {
461 firm_copy_node (get_irg_frame(current_ir_graph), (void *)copy_node_nr);
462 copy_preds(get_irg_frame(current_ir_graph), NULL);
464 if (get_irn_link(get_irg_globals(current_ir_graph)) == NULL) {
465 firm_copy_node (get_irg_globals(current_ir_graph), (void *)copy_node_nr);
466 copy_preds(get_irg_globals(current_ir_graph), NULL);
468 if (get_irn_link(get_irg_initial_mem(current_ir_graph)) == NULL) {
469 firm_copy_node (get_irg_initial_mem(current_ir_graph), (void *)copy_node_nr);
470 copy_preds(get_irg_initial_mem(current_ir_graph), NULL);
472 if (get_irn_link(get_irg_args(current_ir_graph)) == NULL) {
473 firm_copy_node (get_irg_args(current_ir_graph), (void *)copy_node_nr);
474 copy_preds(get_irg_args(current_ir_graph), NULL);
476 set_irg_start (current_ir_graph, get_new_node(get_irg_start(current_ir_graph)));
478 set_irg_start_block(current_ir_graph,
479 get_new_node(get_irg_start_block(current_ir_graph)));
480 set_irg_frame (current_ir_graph, get_new_node(get_irg_frame(current_ir_graph)));
481 set_irg_globals (current_ir_graph, get_new_node(get_irg_globals(current_ir_graph)));
482 set_irg_initial_mem(current_ir_graph, get_new_node(get_irg_initial_mem(current_ir_graph)));
483 set_irg_args (current_ir_graph, get_new_node(get_irg_args(current_ir_graph)));
485 if (get_irn_link(get_irg_bad(current_ir_graph)) == NULL) {
486 firm_copy_node(get_irg_bad(current_ir_graph), (void *)copy_node_nr);
487 copy_preds(get_irg_bad(current_ir_graph), NULL);
489 set_irg_bad(current_ir_graph, get_new_node(get_irg_bad(current_ir_graph)));
491 if (get_irn_link(get_irg_no_mem(current_ir_graph)) == NULL) {
492 firm_copy_node(get_irg_no_mem(current_ir_graph), (void *)copy_node_nr);
493 copy_preds(get_irg_no_mem(current_ir_graph), NULL);
495 set_irg_no_mem(current_ir_graph, get_new_node(get_irg_no_mem(current_ir_graph)));
499 * Copies all reachable nodes to a new obstack. Removes bad inputs
500 * from block nodes and the corresponding inputs from Phi nodes.
501 * Merges single exit blocks with single entry blocks and removes
503 * Adds all new nodes to a new hash table for cse. Does not
504 * perform cse, so the hash table might contain common subexpressions.
507 dead_node_elimination(ir_graph *irg) {
509 int rem_ipview = get_interprocedural_view();
510 struct obstack *graveyard_obst = NULL;
511 struct obstack *rebirth_obst = NULL;
513 edges_init_graph(irg);
515 /* inform statistics that we started a dead-node elimination run */
516 hook_dead_node_elim_start(irg);
518 /* Remember external state of current_ir_graph. */
519 rem = current_ir_graph;
520 current_ir_graph = irg;
521 set_interprocedural_view(false);
523 /* Handle graph state */
524 assert(get_irg_phase_state(current_ir_graph) != phase_building);
525 free_callee_info(current_ir_graph);
526 free_outs(current_ir_graph);
527 /* @@@ so far we loose loops when copying */
528 free_loop_information(current_ir_graph);
530 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
532 /* A quiet place, where the old obstack can rest in peace,
533 until it will be cremated. */
534 graveyard_obst = irg->obst;
536 /* A new obstack, where the reachable nodes will be copied to. */
537 rebirth_obst = xmalloc (sizeof(*rebirth_obst));
538 current_ir_graph->obst = rebirth_obst;
539 obstack_init (current_ir_graph->obst);
541 /* We also need a new hash table for cse */
542 del_identities (irg->value_table);
543 irg->value_table = new_identities ();
545 /* Copy the graph from the old to the new obstack */
548 /* Free memory from old unoptimized obstack */
549 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
550 xfree (graveyard_obst); /* ... then free it. */
553 /* inform statistics that the run is over */
554 hook_dead_node_elim_stop(irg);
556 current_ir_graph = rem;
557 set_interprocedural_view(rem_ipview);
561 * Relink bad predeseccors of a block and store the old in array to the
562 * link field. This function is called by relink_bad_predecessors().
563 * The array of link field starts with the block operand at position 0.
564 * If block has bad predecessors, create a new in array without bad preds.
565 * Otherwise let in array untouched.
567 static void relink_bad_block_predecessors(ir_node *n, void *env) {
568 ir_node **new_in, *irn;
569 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
571 /* if link field of block is NULL, look for bad predecessors otherwise
572 this is allready done */
573 if (get_irn_op(n) == op_Block &&
574 get_irn_link(n) == NULL) {
576 /* save old predecessors in link field (position 0 is the block operand)*/
577 set_irn_link(n, (void *)get_irn_in(n));
579 /* count predecessors without bad nodes */
580 old_irn_arity = get_irn_arity(n);
581 for (i = 0; i < old_irn_arity; i++)
582 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
584 /* arity changing: set new predecessors without bad nodes */
585 if (new_irn_arity < old_irn_arity) {
586 /* Get new predecessor array. We do not resize the array, as we must
587 keep the old one to update Phis. */
588 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
590 /* set new predeseccors in array */
593 for (i = 0; i < old_irn_arity; i++) {
594 irn = get_irn_n(n, i);
596 new_in[new_irn_n] = irn;
597 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
601 //ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity);
602 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
605 } /* ir node has bad predecessors */
607 } /* Block is not relinked */
611 * Relinks Bad predecesors from Bocks and Phis called by walker
612 * remove_bad_predecesors(). If n is a Block, call
613 * relink_bad_block_redecessors(). If n is a Phinode, call also the relinking
614 * function of Phi's Block. If this block has bad predecessors, relink preds
617 static void relink_bad_predecessors(ir_node *n, void *env) {
618 ir_node *block, **old_in;
619 int i, old_irn_arity, new_irn_arity;
621 /* relink bad predeseccors of a block */
622 if (get_irn_op(n) == op_Block)
623 relink_bad_block_predecessors(n, env);
625 /* If Phi node relink its block and its predecessors */
626 if (get_irn_op(n) == op_Phi) {
628 /* Relink predeseccors of phi's block */
629 block = get_nodes_block(n);
630 if (get_irn_link(block) == NULL)
631 relink_bad_block_predecessors(block, env);
633 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
634 old_irn_arity = ARR_LEN(old_in);
636 /* Relink Phi predeseccors if count of predeseccors changed */
637 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
638 /* set new predeseccors in array
639 n->in[0] remains the same block */
641 for(i = 1; i < old_irn_arity; i++)
642 if (!is_Bad((ir_node *)old_in[i])) {
643 n->in[new_irn_arity] = n->in[i];
644 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
648 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
649 ARR_SETLEN(int, n->attr.phi_backedge, new_irn_arity);
652 } /* n is a Phi node */
656 * Removes Bad Bad predecesors from Blocks and the corresponding
657 * inputs to Phi nodes as in dead_node_elimination but without
659 * On walking up set the link field to NULL, on walking down call
660 * relink_bad_predecessors() (This function stores the old in array
661 * to the link field and sets a new in array if arity of predecessors
664 void remove_bad_predecessors(ir_graph *irg) {
665 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
669 /*--------------------------------------------------------------------*/
670 /* Funcionality for inlining */
671 /*--------------------------------------------------------------------*/
674 * Copy node for inlineing. Updates attributes that change when
675 * inlineing but not for dead node elimination.
677 * Copies the node by calling firm_copy_node and then updates the entity if
678 * it's a local one. env must be a pointer of the frame type of the
679 * inlined procedure. The new entities must be in the link field of
683 copy_node_inline (ir_node *n, void *env) {
685 type *frame_tp = (type *)env;
687 firm_copy_node(n, NULL);
688 if (get_irn_op(n) == op_Sel) {
689 new = get_new_node (n);
690 assert(get_irn_op(new) == op_Sel);
691 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
692 set_Sel_entity(new, get_entity_link(get_Sel_entity(n)));
694 } else if (get_irn_op(n) == op_Block) {
695 new = get_new_node (n);
696 new->attr.block.irg = current_ir_graph;
700 static void find_addr(ir_node *node, void *env)
702 if (get_irn_opcode(node) == iro_Proj) {
703 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
709 * currently, we cannot inline two cases:
710 * - call with compound arguments
711 * - graphs that take the address of a parameter
713 * check these conditions here
715 static int can_inline(ir_node *call, ir_graph *called_graph)
717 type *call_type = get_Call_type(call);
718 int params, ress, i, res;
719 assert(is_Method_type(call_type));
721 params = get_method_n_params(call_type);
722 ress = get_method_n_ress(call_type);
725 for (i = 0; i < params; ++i) {
726 type *p_type = get_method_param_type(call_type, i);
728 if (is_compound_type(p_type))
733 for (i = 0; i < ress; ++i) {
734 type *r_type = get_method_res_type(call_type, i);
736 if (is_compound_type(r_type))
741 irg_walk_graph(called_graph, find_addr, NULL, &res);
746 int inline_method(ir_node *call, ir_graph *called_graph) {
748 ir_node *post_call, *post_bl;
750 ir_node *end, *end_bl;
754 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
757 irg_inline_property prop = get_irg_inline_property(called_graph);
759 if ( (prop != irg_inline_forced) &&
760 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
762 /* Do not inline variadic functions. */
763 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
766 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
767 get_method_n_params(get_Call_type(call)));
770 * currently, we cannot inline two cases:
771 * - call with compound arguments
772 * - graphs that take the address of a parameter
774 if (! can_inline(call, called_graph))
777 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
778 rem_opt = get_opt_optimize();
781 /* Handle graph state */
782 assert(get_irg_phase_state(current_ir_graph) != phase_building);
783 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
784 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
785 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
786 set_irg_outs_inconsistent(current_ir_graph);
787 set_irg_loopinfo_inconsistent(current_ir_graph);
788 set_irg_callee_info_state(current_ir_graph, irg_callee_info_inconsistent);
790 /* -- Check preconditions -- */
791 assert(get_irn_op(call) == op_Call);
792 /* @@@ does not work for InterfaceIII.java after cgana
793 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
794 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
795 get_Call_type(call)));
797 assert(get_type_tpop(get_Call_type(call)) == type_method);
798 if (called_graph == current_ir_graph) {
799 set_optimize(rem_opt);
803 /* here we know we WILL inline, so inform the statistics */
804 hook_inline(call, called_graph);
806 /* -- Decide how to handle exception control flow: Is there a handler
807 for the Call node, or do we branch directly to End on an exception?
809 0 There is a handler.
811 2 Exception handling not represented in Firm. -- */
813 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
814 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
815 assert(get_irn_op(proj) == op_Proj);
816 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
817 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
819 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
820 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
821 else { exc_handling = 2; } /* !Mproj && !Xproj */
826 the procedure and later replaces the Start node of the called graph.
827 Post_call is the old Call node and collects the results of the called
828 graph. Both will end up being a tuple. -- */
829 post_bl = get_nodes_block(call);
830 set_irg_current_block(current_ir_graph, post_bl);
831 /* XxMxPxP of Start + parameter of Call */
832 in[pn_Start_X_initial_exec] = new_Jmp();
833 in[pn_Start_M] = get_Call_mem(call);
834 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
835 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
836 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
837 /* in[pn_Start_P_value_arg_base] = ??? */
838 pre_call = new_Tuple(5, in);
842 The new block gets the ins of the old block, pre_call and all its
843 predecessors and all Phi nodes. -- */
844 part_block(pre_call);
846 /* -- Prepare state for dead node elimination -- */
847 /* Visited flags in calling irg must be >= flag in called irg.
848 Else walker and arity computation will not work. */
849 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
850 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
851 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
852 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
853 /* Set pre_call as new Start node in link field of the start node of
854 calling graph and pre_calls block as new block for the start block
856 Further mark these nodes so that they are not visited by the
858 set_irn_link(get_irg_start(called_graph), pre_call);
859 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
860 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
861 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
862 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
863 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
865 /* Initialize for compaction of in arrays */
866 inc_irg_block_visited(current_ir_graph);
868 /* -- Replicate local entities of the called_graph -- */
869 /* copy the entities. */
870 called_frame = get_irg_frame_type(called_graph);
871 for (i = 0; i < get_class_n_members(called_frame); i++) {
872 entity *new_ent, *old_ent;
873 old_ent = get_class_member(called_frame, i);
874 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
875 set_entity_link(old_ent, new_ent);
878 /* visited is > than that of called graph. With this trick visited will
879 remain unchanged so that an outer walker, e.g., searching the call nodes
880 to inline, calling this inline will not visit the inlined nodes. */
881 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
883 /* -- Performing dead node elimination inlines the graph -- */
884 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
886 /* @@@ endless loops are not copied!! -- they should be, I think... */
887 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
888 get_irg_frame_type(called_graph));
890 /* Repair called_graph */
891 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
892 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
893 set_Block_block_visited(get_irg_start_block(called_graph), 0);
895 /* -- Merge the end of the inlined procedure with the call site -- */
896 /* We will turn the old Call node into a Tuple with the following
899 0: Phi of all Memories of Return statements.
900 1: Jmp from new Block that merges the control flow from all exception
901 predecessors of the old end block.
902 2: Tuple of all arguments.
903 3: Phi of Exception memories.
904 In case the old Call directly branches to End on an exception we don't
905 need the block merging all exceptions nor the Phi of the exception
909 /* -- Precompute some values -- */
910 end_bl = get_new_node(get_irg_end_block(called_graph));
911 end = get_new_node(get_irg_end(called_graph));
912 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
913 n_res = get_method_n_ress(get_Call_type(call));
915 res_pred = xmalloc (n_res * sizeof(*res_pred));
916 cf_pred = xmalloc (arity * sizeof(*res_pred));
918 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
920 /* -- archive keepalives -- */
921 irn_arity = get_irn_arity(end);
922 for (i = 0; i < irn_arity; i++)
923 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
925 /* The new end node will die. We need not free as the in array is on the obstack:
926 firm_copy_node only generated 'D' arrays. */
928 /* -- Replace Return nodes by Jump nodes. -- */
930 for (i = 0; i < arity; i++) {
932 ret = get_irn_n(end_bl, i);
933 if (get_irn_op(ret) == op_Return) {
934 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
938 set_irn_in(post_bl, n_ret, cf_pred);
940 /* -- Build a Tuple for all results of the method.
941 Add Phi node if there was more than one Return. -- */
942 turn_into_tuple(post_call, 4);
943 /* First the Memory-Phi */
945 for (i = 0; i < arity; i++) {
946 ret = get_irn_n(end_bl, i);
947 if (get_irn_op(ret) == op_Return) {
948 cf_pred[n_ret] = get_Return_mem(ret);
952 phi = new_Phi(n_ret, cf_pred, mode_M);
953 set_Tuple_pred(call, pn_Call_M_regular, phi);
954 /* Conserve Phi-list for further inlinings -- but might be optimized */
955 if (get_nodes_block(phi) == post_bl) {
956 set_irn_link(phi, get_irn_link(post_bl));
957 set_irn_link(post_bl, phi);
959 /* Now the real results */
961 for (j = 0; j < n_res; j++) {
963 for (i = 0; i < arity; i++) {
964 ret = get_irn_n(end_bl, i);
965 if (get_irn_op(ret) == op_Return) {
966 cf_pred[n_ret] = get_Return_res(ret, j);
971 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
975 /* Conserve Phi-list for further inlinings -- but might be optimized */
976 if (get_nodes_block(phi) == post_bl) {
977 set_irn_link(phi, get_irn_link(post_bl));
978 set_irn_link(post_bl, phi);
981 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
983 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
985 /* Finally the exception control flow.
986 We have two (three) possible situations:
987 First if the Call branches to an exception handler: We need to add a Phi node to
988 collect the memory containing the exception objects. Further we need
989 to add another block to get a correct representation of this Phi. To
990 this block we add a Jmp that resolves into the X output of the Call
991 when the Call is turned into a tuple.
992 Second the Call branches to End, the exception is not handled. Just
993 add all inlined exception branches to the End node.
994 Third: there is no Exception edge at all. Handle as case two. */
995 if (exc_handling == 0) {
997 for (i = 0; i < arity; i++) {
999 ret = get_irn_n(end_bl, i);
1000 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1001 cf_pred[n_exc] = ret;
1006 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1007 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1008 /* The Phi for the memories with the exception objects */
1010 for (i = 0; i < arity; i++) {
1012 ret = skip_Proj(get_irn_n(end_bl, i));
1013 if (get_irn_op(ret) == op_Call) {
1014 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
1016 } else if (is_fragile_op(ret)) {
1017 /* We rely that all cfops have the memory output at the same position. */
1018 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
1020 } else if (get_irn_op(ret) == op_Raise) {
1021 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
1025 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1027 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1028 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1031 ir_node *main_end_bl;
1032 int main_end_bl_arity;
1033 ir_node **end_preds;
1035 /* assert(exc_handling == 1 || no exceptions. ) */
1037 for (i = 0; i < arity; i++) {
1038 ir_node *ret = get_irn_n(end_bl, i);
1040 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1041 cf_pred[n_exc] = ret;
1045 main_end_bl = get_irg_end_block(current_ir_graph);
1046 main_end_bl_arity = get_irn_arity(main_end_bl);
1047 end_preds = xmalloc ((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1049 for (i = 0; i < main_end_bl_arity; ++i)
1050 end_preds[i] = get_irn_n(main_end_bl, i);
1051 for (i = 0; i < n_exc; ++i)
1052 end_preds[main_end_bl_arity + i] = cf_pred[i];
1053 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1054 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1055 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1061 #if 0 /* old. now better, correcter, faster implementation. */
1063 /* -- If the exception control flow from the inlined Call directly
1064 branched to the end block we now have the following control
1065 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
1066 remove the Jmp along with it's empty block and add Jmp's
1067 predecessors as predecessors of this end block. No problem if
1068 there is no exception, because then branches Bad to End which
1070 @@@ can't we know this beforehand: by getting the Proj(1) from
1071 the Call link list and checking whether it goes to Proj. */
1072 /* find the problematic predecessor of the end block. */
1073 end_bl = get_irg_end_block(current_ir_graph);
1074 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
1075 cf_op = get_Block_cfgpred(end_bl, i);
1076 if (get_irn_op(cf_op) == op_Proj) {
1077 cf_op = get_Proj_pred(cf_op);
1078 if ((get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
1079 /* There are unoptimized tuples from inlineing before when no exc */
1080 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
1081 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
1082 assert(get_irn_op(cf_op) == op_Jmp);
1088 if (i < get_Block_n_cfgpreds(end_bl)) {
1089 bl = get_nodes_block(cf_op);
1090 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
1091 cf_pred = xmalloc (arity * sizeof(*cf_pred));
1092 for (j = 0; j < i; j++)
1093 cf_pred[j] = get_Block_cfgpred(end_bl, j);
1094 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
1095 cf_pred[j] = get_Block_cfgpred(bl, j-i);
1096 for (j = j; j < arity; j++)
1097 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
1098 set_irn_in(end_bl, arity, cf_pred);
1100 /* Remove the exception pred from post-call Tuple. */
1101 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1106 /* -- Turn cse back on. -- */
1107 set_optimize(rem_opt);
1112 /********************************************************************/
1113 /* Apply inlineing to small methods. */
1114 /********************************************************************/
1116 /* It makes no sense to inline too many calls in one procedure. Anyways,
1117 I didn't get a version with NEW_ARR_F to run. */
1118 #define MAX_INLINE 1024
1121 * environment for inlining small irgs
1123 typedef struct _inline_env_t {
1125 ir_node *calls[MAX_INLINE];
1129 * Returns the irg called from a Call node. If the irg is not
1130 * known, NULL is returned.
1132 static ir_graph *get_call_called_irg(ir_node *call) {
1134 ir_graph *called_irg = NULL;
1136 assert(get_irn_op(call) == op_Call);
1138 addr = get_Call_ptr(call);
1139 if ((get_irn_op(addr) == op_SymConst) && (get_SymConst_kind (addr) == symconst_addr_ent)) {
1140 called_irg = get_entity_irg(get_SymConst_entity(addr));
1146 static void collect_calls(ir_node *call, void *env) {
1149 if (get_irn_op(call) != op_Call) return;
1151 addr = get_Call_ptr(call);
1153 if (get_irn_op(addr) == op_SymConst) {
1154 if (get_SymConst_kind(addr) == symconst_addr_ent) {
1155 ir_graph *called_irg = get_entity_irg(get_SymConst_entity(addr));
1156 inline_env_t *ienv = (inline_env_t *)env;
1157 if (called_irg && ienv->pos < MAX_INLINE) {
1158 /* The Call node calls a locally defined method. Remember to inline. */
1159 ienv->calls[ienv->pos++] = call;
1166 * Inlines all small methods at call sites where the called address comes
1167 * from a Const node that references the entity representing the called
1169 * The size argument is a rough measure for the code size of the method:
1170 * Methods where the obstack containing the firm graph is smaller than
1173 void inline_small_irgs(ir_graph *irg, int size) {
1175 ir_graph *rem = current_ir_graph;
1176 inline_env_t env /* = {0, NULL}*/;
1178 if (!(get_opt_optimize() && get_opt_inline())) return;
1180 current_ir_graph = irg;
1181 /* Handle graph state */
1182 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1183 free_callee_info(current_ir_graph);
1185 /* Find Call nodes to inline.
1186 (We can not inline during a walk of the graph, as inlineing the same
1187 method several times changes the visited flag of the walked graph:
1188 after the first inlineing visited of the callee equals visited of
1189 the caller. With the next inlineing both are increased.) */
1191 irg_walk(get_irg_end(irg), NULL, collect_calls, &env);
1193 if ((env.pos > 0) && (env.pos < MAX_INLINE)) {
1194 /* There are calls to inline */
1195 collect_phiprojs(irg);
1196 for (i = 0; i < env.pos; i++) {
1198 callee = get_entity_irg(get_SymConst_entity(get_Call_ptr(env.calls[i])));
1199 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1200 (get_irg_inline_property(callee) == irg_inline_forced)) {
1201 inline_method(env.calls[i], callee);
1206 current_ir_graph = rem;
1210 * Environment for inlining irgs.
1213 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1214 int n_nodes_orig; /**< for statistics */
1215 eset *call_nodes; /**< All call nodes in this graph */
1217 int n_call_nodes_orig; /**< for statistics */
1218 int n_callers; /**< Number of known graphs that call this graphs. */
1219 int n_callers_orig; /**< for statistics */
1222 static inline_irg_env *new_inline_irg_env(void) {
1223 inline_irg_env *env = xmalloc(sizeof(*env));
1224 env->n_nodes = -2; /* uncount Start, End */
1225 env->n_nodes_orig = -2; /* uncount Start, End */
1226 env->call_nodes = eset_create();
1227 env->n_call_nodes = 0;
1228 env->n_call_nodes_orig = 0;
1230 env->n_callers_orig = 0;
1234 static void free_inline_irg_env(inline_irg_env *env) {
1235 eset_destroy(env->call_nodes);
1239 static void collect_calls2(ir_node *call, void *env) {
1240 inline_irg_env *x = (inline_irg_env *)env;
1241 ir_op *op = get_irn_op(call);
1244 /* count nodes in irg */
1245 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1250 if (op != op_Call) return;
1252 /* collect all call nodes */
1253 eset_insert(x->call_nodes, (void *)call);
1255 x->n_call_nodes_orig++;
1257 /* count all static callers */
1258 callee = get_call_called_irg(call);
1260 ((inline_irg_env *)get_irg_link(callee))->n_callers++;
1261 ((inline_irg_env *)get_irg_link(callee))->n_callers_orig++;
1265 INLINE static int is_leave(ir_graph *irg) {
1266 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1269 INLINE static int is_smaller(ir_graph *callee, int size) {
1270 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1275 * Inlines small leave methods at call sites where the called address comes
1276 * from a Const node that references the entity representing the called
1278 * The size argument is a rough measure for the code size of the method:
1279 * Methods where the obstack containing the firm graph is smaller than
1282 void inline_leave_functions(int maxsize, int leavesize, int size) {
1283 inline_irg_env *env;
1284 int i, n_irgs = get_irp_n_irgs();
1285 ir_graph *rem = current_ir_graph;
1288 if (!(get_opt_optimize() && get_opt_inline())) return;
1290 /* extend all irgs by a temporary data structure for inlining. */
1291 for (i = 0; i < n_irgs; ++i)
1292 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1294 /* Precompute information in temporary data structure. */
1295 for (i = 0; i < n_irgs; ++i) {
1296 current_ir_graph = get_irp_irg(i);
1297 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1298 free_callee_info(current_ir_graph);
1300 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1301 get_irg_link(current_ir_graph));
1304 /* -- and now inline. -- */
1306 /* Inline leaves recursively -- we might construct new leaves. */
1307 while (did_inline) {
1310 for (i = 0; i < n_irgs; ++i) {
1312 int phiproj_computed = 0;
1314 current_ir_graph = get_irp_irg(i);
1315 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1317 for (call = eset_first(env->call_nodes); call; call = eset_next(env->call_nodes)) {
1320 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1321 callee = get_call_called_irg(call);
1323 if (env->n_nodes > maxsize) continue; // break;
1325 if (callee && (is_leave(callee) && is_smaller(callee, leavesize))) {
1326 if (!phiproj_computed) {
1327 phiproj_computed = 1;
1328 collect_phiprojs(current_ir_graph);
1330 did_inline = inline_method(call, callee);
1333 /* Do some statistics */
1334 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1335 env->n_call_nodes --;
1336 env->n_nodes += callee_env->n_nodes;
1337 callee_env->n_callers--;
1344 /* inline other small functions. */
1345 for (i = 0; i < n_irgs; ++i) {
1348 int phiproj_computed = 0;
1350 current_ir_graph = get_irp_irg(i);
1351 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1353 /* we can not walk and change a set, nor remove from it.
1355 walkset = env->call_nodes;
1356 env->call_nodes = eset_create();
1357 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1360 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1361 callee = get_call_called_irg(call);
1364 ((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1365 (get_irg_inline_property(callee) == irg_inline_forced))) {
1366 if (!phiproj_computed) {
1367 phiproj_computed = 1;
1368 collect_phiprojs(current_ir_graph);
1370 if (inline_method(call, callee)) {
1371 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1372 env->n_call_nodes--;
1373 eset_insert_all(env->call_nodes, callee_env->call_nodes); /* @@@ ??? This are the wrong nodes !? Not the copied ones. */
1374 env->n_call_nodes += callee_env->n_call_nodes;
1375 env->n_nodes += callee_env->n_nodes;
1376 callee_env->n_callers--;
1379 eset_insert(env->call_nodes, call);
1382 eset_destroy(walkset);
1385 for (i = 0; i < n_irgs; ++i) {
1386 current_ir_graph = get_irp_irg(i);
1388 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1389 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1390 (env->n_callers_orig != env->n_callers))
1391 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1392 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1393 env->n_callers_orig, env->n_callers,
1394 get_entity_name(get_irg_entity(current_ir_graph)));
1396 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1399 current_ir_graph = rem;
1402 /*******************************************************************/
1403 /* Code Placement. Pins all floating nodes to a block where they */
1404 /* will be executed only if needed. */
1405 /*******************************************************************/
1408 * Find the earliest correct block for N. --- Place N into the
1409 * same Block as its dominance-deepest Input.
1412 place_floats_early(ir_node *n, pdeq *worklist)
1414 int i, start, irn_arity;
1416 /* we must not run into an infinite loop */
1417 assert (irn_not_visited(n));
1418 mark_irn_visited(n);
1420 /* Place floating nodes. */
1421 if (get_irn_pinned(n) == op_pin_state_floats) {
1423 ir_node *b = new_Bad(); /* The block to place this node in */
1424 int bad_recursion = is_Bad(get_nodes_block(n));
1426 assert(get_irn_op(n) != op_Block);
1428 if ((get_irn_op(n) == op_Const) ||
1429 (get_irn_op(n) == op_SymConst) ||
1431 (get_irn_op(n) == op_Unknown)) {
1432 /* These nodes will not be placed by the loop below. */
1433 b = get_irg_start_block(current_ir_graph);
1437 /* find the block for this node. */
1438 irn_arity = get_irn_arity(n);
1439 for (i = 0; i < irn_arity; i++) {
1440 ir_node *dep = get_irn_n(n, i);
1443 if ((irn_not_visited(dep))
1444 && (get_irn_pinned(dep) == op_pin_state_floats)) {
1445 place_floats_early(dep, worklist);
1449 * A node in the Bad block must stay in the bad block,
1450 * so don't compute a new block for it.
1455 /* Because all loops contain at least one op_pin_state_pinned node, now all
1456 our inputs are either op_pin_state_pinned or place_early has already
1457 been finished on them. We do not have any unfinished inputs! */
1458 dep_block = get_nodes_block(dep);
1459 if ((!is_Bad(dep_block)) &&
1460 (get_Block_dom_depth(dep_block) > depth)) {
1462 depth = get_Block_dom_depth(dep_block);
1464 /* Avoid that the node is placed in the Start block */
1465 if ((depth == 1) && (get_Block_dom_depth(get_nodes_block(n)) > 1)) {
1466 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1467 assert(b != get_irg_start_block(current_ir_graph));
1471 set_nodes_block(n, b);
1474 /* Add predecessors of non floating nodes on worklist. */
1475 start = (get_irn_op(n) == op_Block) ? 0 : -1;
1476 irn_arity = get_irn_arity(n);
1477 for (i = start; i < irn_arity; i++) {
1478 ir_node *pred = get_irn_n(n, i);
1479 if (irn_not_visited(pred)) {
1480 pdeq_putr (worklist, pred);
1486 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1487 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1488 * places all floating nodes reachable from its argument through floating
1489 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1491 static INLINE void place_early(pdeq *worklist) {
1493 inc_irg_visited(current_ir_graph);
1495 /* this inits the worklist */
1496 place_floats_early(get_irg_end(current_ir_graph), worklist);
1498 /* Work the content of the worklist. */
1499 while (!pdeq_empty (worklist)) {
1500 ir_node *n = pdeq_getl (worklist);
1501 if (irn_not_visited(n)) place_floats_early(n, worklist);
1504 set_irg_outs_inconsistent(current_ir_graph);
1505 current_ir_graph->op_pin_state_pinned = op_pin_state_pinned;
1508 /** Compute the deepest common ancestor of block and dca. */
1509 static ir_node *calc_dca(ir_node *dca, ir_node *block)
1512 if (!dca) return block;
1513 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1514 block = get_Block_idom(block);
1515 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1516 dca = get_Block_idom(dca);
1518 while (block != dca)
1519 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1524 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1525 * I.e., DCA is the block where we might place PRODUCER.
1526 * A data flow edge points from producer to consumer.
1529 consumer_dom_dca (ir_node *dca, ir_node *consumer, ir_node *producer)
1531 ir_node *block = NULL;
1533 /* Compute the latest block into which we can place a node so that it is
1535 if (get_irn_op(consumer) == op_Phi) {
1536 /* our consumer is a Phi-node, the effective use is in all those
1537 blocks through which the Phi-node reaches producer */
1539 ir_node *phi_block = get_nodes_block(consumer);
1540 irn_arity = get_irn_arity(consumer);
1542 for (i = 0; i < irn_arity; i++) {
1543 if (get_irn_n(consumer, i) == producer) {
1544 ir_node *new_block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1546 block = calc_dca(block, new_block);
1550 assert(is_no_Block(consumer));
1551 block = get_nodes_block(consumer);
1554 /* Compute the deepest common ancestor of block and dca. */
1555 return calc_dca(dca, block);
1558 static INLINE int get_irn_loop_depth(ir_node *n) {
1559 return get_loop_depth(get_irn_loop(n));
1563 * Move n to a block with less loop depth than it's current block. The
1564 * new block must be dominated by early.
1567 move_out_of_loops (ir_node *n, ir_node *early)
1569 ir_node *best, *dca;
1573 /* Find the region deepest in the dominator tree dominating
1574 dca with the least loop nesting depth, but still dominated
1575 by our early placement. */
1576 dca = get_nodes_block(n);
1578 while (dca != early) {
1579 dca = get_Block_idom(dca);
1580 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1581 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1585 if (best != get_nodes_block(n)) {
1587 printf("Moving out of loop: "); DDMN(n);
1588 printf(" Outermost block: "); DDMN(early);
1589 printf(" Best block: "); DDMN(best);
1590 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1592 set_nodes_block(n, best);
1597 * Find the latest legal block for N and place N into the
1598 * `optimal' Block between the latest and earliest legal block.
1599 * The `optimal' block is the dominance-deepest block of those
1600 * with the least loop-nesting-depth. This places N out of as many
1601 * loops as possible and then makes it as control dependant as
1605 place_floats_late(ir_node *n, pdeq *worklist)
1610 assert (irn_not_visited(n)); /* no multiple placement */
1612 mark_irn_visited(n);
1614 /* no need to place block nodes, control nodes are already placed. */
1615 if ((get_irn_op(n) != op_Block) &&
1617 (get_irn_mode(n) != mode_X)) {
1618 /* Remember the early placement of this block to move it
1619 out of loop no further than the early placement. */
1620 early = get_nodes_block(n);
1622 /* Do not move code not reachable from Start. For
1623 * these we could not compute dominator information. */
1624 if (is_Bad(early) || get_Block_dom_depth(early) == -1)
1627 /* Assure that our users are all placed, except the Phi-nodes.
1628 --- Each data flow cycle contains at least one Phi-node. We
1629 have to break the `user has to be placed before the
1630 producer' dependence cycle and the Phi-nodes are the
1631 place to do so, because we need to base our placement on the
1632 final region of our users, which is OK with Phi-nodes, as they
1633 are op_pin_state_pinned, and they never have to be placed after a
1634 producer of one of their inputs in the same block anyway. */
1635 for (i = 0; i < get_irn_n_outs(n); i++) {
1636 ir_node *succ = get_irn_out(n, i);
1637 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
1638 place_floats_late(succ, worklist);
1641 /* We have to determine the final block of this node... except for
1643 if ((get_irn_pinned(n) == op_pin_state_floats) &&
1644 (get_irn_op(n) != op_Const) &&
1645 (get_irn_op(n) != op_SymConst)) {
1646 ir_node *dca = NULL; /* deepest common ancestor in the
1647 dominator tree of all nodes'
1648 blocks depending on us; our final
1649 placement has to dominate DCA. */
1650 for (i = 0; i < get_irn_n_outs(n); i++) {
1651 ir_node *out = get_irn_out(n, i);
1652 /* ignore if out is in dead code */
1653 ir_node *outbl = get_nodes_block(out);
1654 if (is_Bad(outbl) || get_Block_dom_depth(outbl) == -1)
1656 dca = consumer_dom_dca (dca, out, n);
1659 set_nodes_block(n, dca);
1661 move_out_of_loops (n, early);
1663 /* else all outs are in dead code */
1667 /* Add predecessors of all non-floating nodes on list. (Those of floating
1668 nodes are placeded already and therefore are marked.) */
1669 for (i = 0; i < get_irn_n_outs(n); i++) {
1670 ir_node *succ = get_irn_out(n, i);
1671 if (irn_not_visited(get_irn_out(n, i))) {
1672 pdeq_putr (worklist, succ);
1677 static INLINE void place_late(pdeq *worklist) {
1679 inc_irg_visited(current_ir_graph);
1681 /* This fills the worklist initially. */
1682 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1684 /* And now empty the worklist again... */
1685 while (!pdeq_empty (worklist)) {
1686 ir_node *n = pdeq_getl (worklist);
1687 if (irn_not_visited(n)) place_floats_late(n, worklist);
1691 void place_code(ir_graph *irg) {
1693 ir_graph *rem = current_ir_graph;
1695 current_ir_graph = irg;
1697 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1699 /* Handle graph state */
1700 assert(get_irg_phase_state(irg) != phase_building);
1701 if (get_irg_dom_state(irg) != dom_consistent)
1704 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1705 free_loop_information(irg);
1706 construct_backedges(irg);
1709 /* Place all floating nodes as early as possible. This guarantees
1710 a legal code placement. */
1711 worklist = new_pdeq();
1712 place_early(worklist);
1714 /* place_early invalidates the outs, place_late needs them. */
1716 /* Now move the nodes down in the dominator tree. This reduces the
1717 unnecessary executions of the node. */
1718 place_late(worklist);
1720 set_irg_outs_inconsistent(current_ir_graph);
1721 set_irg_loopinfo_inconsistent(current_ir_graph);
1723 current_ir_graph = rem;
1727 * Called by walker of remove_critical_cf_edges().
1729 * Place an empty block to an edge between a blocks of multiple
1730 * predecessors and a block of multiple successors.
1733 * @param env Environment of walker. This field is unused and has
1736 static void walk_critical_cf_edges(ir_node *n, void *env) {
1738 ir_node *pre, *block, **in, *jmp;
1740 /* Block has multiple predecessors */
1741 if ((op_Block == get_irn_op(n)) &&
1742 (get_irn_arity(n) > 1)) {
1743 arity = get_irn_arity(n);
1745 if (n == get_irg_end_block(current_ir_graph))
1746 return; /* No use to add a block here. */
1748 for (i=0; i<arity; i++) {
1749 pre = get_irn_n(n, i);
1750 /* Predecessor has multiple successors. Insert new flow edge */
1751 if ((NULL != pre) &&
1752 (op_Proj == get_irn_op(pre)) &&
1753 op_Raise != get_irn_op(skip_Proj(pre))) {
1755 /* set predecessor array for new block */
1756 in = NEW_ARR_D (ir_node *, current_ir_graph->obst, 1);
1757 /* set predecessor of new block */
1759 block = new_Block(1, in);
1760 /* insert new jmp node to new block */
1761 set_cur_block(block);
1764 /* set successor of new block */
1765 set_irn_n(n, i, jmp);
1767 } /* predecessor has multiple successors */
1768 } /* for all predecessors */
1769 } /* n is a block */
1772 void remove_critical_cf_edges(ir_graph *irg) {
1773 if (get_opt_critical_edges())
1774 irg_walk_graph(irg, NULL, walk_critical_cf_edges, NULL);