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.
22 #include "irgraph_t.h"
34 #include "pdeq.h" /* Fuer code placement */
38 #include "irbackedge_t.h"
44 /* Defined in iropt.c */
45 pset *new_identities (void);
46 void del_identities (pset *value_table);
47 void add_identities (pset *value_table, ir_node *node);
49 /*------------------------------------------------------------------*/
50 /* apply optimizations of iropt to all nodes. */
51 /*------------------------------------------------------------------*/
53 static void init_link (ir_node *n, void *env) {
54 set_irn_link(n, NULL);
57 #if 0 /* Old version. Avoids Ids.
58 This is not necessary: we do a postwalk, and get_irn_n
59 removes ids anyways. So it's much cheaper to call the
60 optimization less often and use the exchange() algorithm. */
62 optimize_in_place_wrapper (ir_node *n, void *env) {
64 ir_node *optimized, *old;
66 irn_arity = get_irn_arity(n);
67 for (i = 0; i < irn_arity; i++) {
68 /* get_irn_n skips Id nodes, so comparison old != optimized does not
69 show all optimizations. Therefore always set new predecessor. */
70 old = get_irn_intra_n(n, i);
71 optimized = optimize_in_place_2(old);
72 set_irn_n(n, i, optimized);
75 if (get_irn_op(n) == op_Block) {
76 optimized = optimize_in_place_2(n);
77 if (optimized != n) exchange (n, optimized);
82 optimize_in_place_wrapper (ir_node *n, void *env) {
83 ir_node *optimized = optimize_in_place_2(n);
84 if (optimized != n) exchange (n, optimized);
89 static INLINE void do_local_optimize(ir_node *n) {
90 /* Handle graph state */
91 assert(get_irg_phase_state(current_ir_graph) != phase_building);
92 if (get_opt_global_cse())
93 set_irg_pinned(current_ir_graph, op_pin_state_floats);
94 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
95 set_irg_outs_inconsistent(current_ir_graph);
96 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
97 set_irg_dom_inconsistent(current_ir_graph);
98 set_irg_loopinfo_inconsistent(current_ir_graph);
101 /* Clean the value_table in irg for the cse. */
102 del_identities(current_ir_graph->value_table);
103 current_ir_graph->value_table = new_identities();
105 /* walk over the graph */
106 irg_walk(n, init_link, optimize_in_place_wrapper, NULL);
109 void local_optimize_node(ir_node *n) {
110 ir_graph *rem = current_ir_graph;
111 current_ir_graph = get_irn_irg(n);
113 do_local_optimize(n);
115 current_ir_graph = rem;
120 local_optimize_graph (ir_graph *irg) {
121 ir_graph *rem = current_ir_graph;
122 current_ir_graph = irg;
124 do_local_optimize(irg->end);
126 current_ir_graph = rem;
130 /*------------------------------------------------------------------*/
131 /* Routines for dead node elimination / copying garbage collection */
132 /* of the obstack. */
133 /*------------------------------------------------------------------*/
136 * Remember the new node in the old node by using a field all nodes have.
139 set_new_node (ir_node *old, ir_node *new)
145 * Get this new node, before the old node is forgotton.
147 static INLINE ir_node *
148 get_new_node (ir_node * n)
154 * We use the block_visited flag to mark that we have computed the
155 * number of useful predecessors for this block.
156 * Further we encode the new arity in this flag in the old blocks.
157 * Remembering the arity is useful, as it saves a lot of pointer
158 * accesses. This function is called for all Phi and Block nodes
162 compute_new_arity(ir_node *b) {
163 int i, res, irn_arity;
166 irg_v = get_irg_block_visited(current_ir_graph);
167 block_v = get_Block_block_visited(b);
168 if (block_v >= irg_v) {
169 /* we computed the number of preds for this block and saved it in the
171 return block_v - irg_v;
173 /* compute the number of good predecessors */
174 res = irn_arity = get_irn_arity(b);
175 for (i = 0; i < irn_arity; i++)
176 if (get_irn_opcode(get_irn_n(b, i)) == iro_Bad) res--;
177 /* save it in the flag. */
178 set_Block_block_visited(b, irg_v + res);
183 /* TODO: add an ir_op operation */
184 static INLINE void new_backedge_info(ir_node *n) {
185 switch(get_irn_opcode(n)) {
187 n->attr.block.cg_backedge = NULL;
188 n->attr.block.backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
191 n->attr.phi_backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
194 n->attr.filter.backedge = new_backedge_arr(current_ir_graph->obst, get_irn_arity(n));
201 * Copies the node to the new obstack. The Ins of the new node point to
202 * the predecessors on the old obstack. For block/phi nodes not all
203 * predecessors might be copied. n->link points to the new node.
204 * For Phi and Block nodes the function allocates in-arrays with an arity
205 * only for useful predecessors. The arity is determined by counting
206 * the non-bad predecessors of the block.
208 * @param n The node to be copied
209 * @param env if non-NULL, the node number attribute will be copied to the new node
212 copy_node (ir_node *n, void *env) {
215 opcode op = get_irn_opcode(n);
216 int copy_node_nr = env != NULL;
218 /* The end node looses it's flexible in array. This doesn't matter,
219 as dead node elimination builds End by hand, inlineing doesn't use
221 /* assert(n->op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
224 /* node copied already */
226 } else if (op == iro_Block) {
228 new_arity = compute_new_arity(n);
229 n->attr.block.graph_arr = NULL;
231 block = get_nodes_block(n);
232 if (get_irn_opcode(n) == iro_Phi) {
233 new_arity = compute_new_arity(block);
235 new_arity = get_irn_arity(n);
238 nn = new_ir_node(get_irn_dbg_info(n),
245 /* Copy the attributes. These might point to additional data. If this
246 was allocated on the old obstack the pointers now are dangling. This
247 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
249 new_backedge_info(nn);
253 /* for easier debugging, we want to copy the node numbers too */
254 nn->node_nr = n->node_nr;
257 /* printf("\n old node: "); DDMSG2(n);
258 printf(" new node: "); DDMSG2(nn); */
263 * Copies new predecessors of old node to new node remembered in link.
264 * Spare the Bad predecessors of Phi and Block nodes.
267 copy_preds (ir_node *n, void *env) {
271 nn = get_new_node(n);
273 /* printf("\n old node: "); DDMSG2(n);
274 printf(" new node: "); DDMSG2(nn);
275 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
277 if (get_irn_opcode(n) == iro_Block) {
278 /* Don't copy Bad nodes. */
280 irn_arity = get_irn_arity(n);
281 for (i = 0; i < irn_arity; i++)
282 if (get_irn_opcode(get_irn_n(n, i)) != iro_Bad) {
283 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
284 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
287 /* repair the block visited flag from above misuse. Repair it in both
288 graphs so that the old one can still be used. */
289 set_Block_block_visited(nn, 0);
290 set_Block_block_visited(n, 0);
291 /* Local optimization could not merge two subsequent blocks if
292 in array contained Bads. Now it's possible.
293 We don't call optimize_in_place as it requires
294 that the fields in ir_graph are set properly. */
295 if ((get_opt_control_flow_straightening()) &&
296 (get_Block_n_cfgpreds(nn) == 1) &&
297 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
298 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
300 /* Jmp jumps into the block it is in -- deal self cycle. */
301 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
302 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
307 } else if (get_irn_opcode(n) == iro_Phi) {
308 /* Don't copy node if corresponding predecessor in block is Bad.
309 The Block itself should not be Bad. */
310 block = get_nodes_block(n);
311 set_irn_n (nn, -1, get_new_node(block));
313 irn_arity = get_irn_arity(n);
314 for (i = 0; i < irn_arity; i++)
315 if (get_irn_opcode(get_irn_n(block, i)) != iro_Bad) {
316 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
317 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
320 /* If the pre walker reached this Phi after the post walker visited the
321 block block_visited is > 0. */
322 set_Block_block_visited(get_nodes_block(n), 0);
323 /* Compacting the Phi's ins might generate Phis with only one
325 if (get_irn_arity(n) == 1)
326 exchange(n, get_irn_n(n, 0));
328 irn_arity = get_irn_arity(n);
329 for (i = -1; i < irn_arity; i++)
330 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
332 /* Now the new node is complete. We can add it to the hash table for cse.
333 @@@ inlinening aborts if we identify End. Why? */
334 if(get_irn_op(nn) != op_End)
335 add_identities (current_ir_graph->value_table, nn);
339 * Copies the graph recursively, compacts the keepalive of the end node.
341 * @param copy_node_nr If non-zero, the node number will be copied
344 copy_graph (int copy_node_nr) {
345 ir_node *oe, *ne, *ob, *nb; /* old end, new end, old bad, new bad */
346 ir_node *ka; /* keep alive */
349 oe = get_irg_end(current_ir_graph);
350 /* copy the end node by hand, allocate dynamic in array! */
351 ne = new_ir_node(get_irn_dbg_info(oe),
358 /* Copy the attributes. Well, there might be some in the future... */
360 set_new_node(oe, ne);
362 ob = get_irg_bad(current_ir_graph);
363 nb = new_ir_node(get_irn_dbg_info(ob),
370 set_new_node(ob, nb);
372 /* copy the live nodes */
373 irg_walk(get_nodes_block(oe), copy_node, copy_preds, (void *)copy_node_nr);
374 /* copy_preds for the end node ... */
375 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
377 /*- ... and now the keep alives. -*/
378 /* First pick the not marked block nodes and walk them. We must pick these
379 first as else we will oversee blocks reachable from Phis. */
380 irn_arity = get_irn_arity(oe);
381 for (i = 0; i < irn_arity; i++) {
382 ka = get_irn_intra_n(oe, i);
383 if ((get_irn_op(ka) == op_Block) &&
384 (get_irn_visited(ka) < get_irg_visited(current_ir_graph))) {
385 /* We must keep the block alive and copy everything reachable */
386 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
387 irg_walk(ka, copy_node, copy_preds, (void *)copy_node_nr);
388 add_End_keepalive(ne, get_new_node(ka));
392 /* Now pick the Phis. Here we will keep all! */
393 irn_arity = get_irn_arity(oe);
394 for (i = 0; i < irn_arity; i++) {
395 ka = get_irn_intra_n(oe, i);
396 if ((get_irn_op(ka) == op_Phi)) {
397 if (get_irn_visited(ka) < get_irg_visited(current_ir_graph)) {
398 /* We didn't copy the Phi yet. */
399 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
400 irg_walk(ka, copy_node, copy_preds, (void *)copy_node_nr);
402 add_End_keepalive(ne, get_new_node(ka));
406 /* start block somtimes only reached after keep alives */
407 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
411 * Copies the graph reachable from current_ir_graph->end to the obstack
412 * in current_ir_graph and fixes the environment.
413 * Then fixes the fields in current_ir_graph containing nodes of the
416 * @param copy_node_nr If non-zero, the node number will be copied
419 copy_graph_env (int copy_node_nr) {
421 /* Not all nodes remembered in current_ir_graph might be reachable
422 from the end node. Assure their link is set to NULL, so that
423 we can test whether new nodes have been computed. */
424 set_irn_link(get_irg_frame (current_ir_graph), NULL);
425 set_irn_link(get_irg_globals (current_ir_graph), NULL);
426 set_irn_link(get_irg_args (current_ir_graph), NULL);
427 set_irn_link(get_irg_initial_mem(current_ir_graph), NULL);
429 /* we use the block walk flag for removing Bads from Blocks ins. */
430 inc_irg_block_visited(current_ir_graph);
433 copy_graph(copy_node_nr);
435 /* fix the fields in current_ir_graph */
436 old_end = get_irg_end(current_ir_graph);
437 set_irg_end (current_ir_graph, get_new_node(old_end));
438 set_irg_end_except (current_ir_graph, get_irg_end(current_ir_graph));
439 set_irg_end_reg (current_ir_graph, get_irg_end(current_ir_graph));
441 set_irg_end_block (current_ir_graph, get_new_node(get_irg_end_block(current_ir_graph)));
442 if (get_irn_link(get_irg_frame(current_ir_graph)) == NULL) {
443 copy_node (get_irg_frame(current_ir_graph), (void *)copy_node_nr);
444 copy_preds(get_irg_frame(current_ir_graph), NULL);
446 if (get_irn_link(get_irg_globals(current_ir_graph)) == NULL) {
447 copy_node (get_irg_globals(current_ir_graph), (void *)copy_node_nr);
448 copy_preds(get_irg_globals(current_ir_graph), NULL);
450 if (get_irn_link(get_irg_initial_mem(current_ir_graph)) == NULL) {
451 copy_node (get_irg_initial_mem(current_ir_graph), (void *)copy_node_nr);
452 copy_preds(get_irg_initial_mem(current_ir_graph), NULL);
454 if (get_irn_link(get_irg_args(current_ir_graph)) == NULL) {
455 copy_node (get_irg_args(current_ir_graph), (void *)copy_node_nr);
456 copy_preds(get_irg_args(current_ir_graph), NULL);
458 set_irg_start (current_ir_graph, get_new_node(get_irg_start(current_ir_graph)));
460 set_irg_start_block(current_ir_graph,
461 get_new_node(get_irg_start_block(current_ir_graph)));
462 set_irg_frame (current_ir_graph, get_new_node(get_irg_frame(current_ir_graph)));
463 set_irg_globals (current_ir_graph, get_new_node(get_irg_globals(current_ir_graph)));
464 set_irg_initial_mem(current_ir_graph, get_new_node(get_irg_initial_mem(current_ir_graph)));
465 set_irg_args (current_ir_graph, get_new_node(get_irg_args(current_ir_graph)));
467 if (get_irn_link(get_irg_bad(current_ir_graph)) == NULL) {
468 copy_node(get_irg_bad(current_ir_graph), (void *)copy_node_nr);
469 copy_preds(get_irg_bad(current_ir_graph), NULL);
471 set_irg_bad(current_ir_graph, get_new_node(get_irg_bad(current_ir_graph)));
475 * Copies all reachable nodes to a new obstack. Removes bad inputs
476 * from block nodes and the corresponding inputs from Phi nodes.
477 * Merges single exit blocks with single entry blocks and removes
479 * Adds all new nodes to a new hash table for cse. Does not
480 * perform cse, so the hash table might contain common subexpressions.
483 dead_node_elimination(ir_graph *irg) {
485 int rem_ipview = interprocedural_view;
486 struct obstack *graveyard_obst = NULL;
487 struct obstack *rebirth_obst = NULL;
489 /* inform statistics that we started a dead-node elimination run */
490 stat_dead_node_elim_start(irg);
492 /* Remember external state of current_ir_graph. */
493 rem = current_ir_graph;
494 current_ir_graph = irg;
495 interprocedural_view = 0;
497 /* Handle graph state */
498 assert(get_irg_phase_state(current_ir_graph) != phase_building);
499 free_callee_info(current_ir_graph);
500 free_outs(current_ir_graph);
501 /* @@@ so far we loose loops when copying */
502 free_loop_information(current_ir_graph);
504 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
506 /* A quiet place, where the old obstack can rest in peace,
507 until it will be cremated. */
508 graveyard_obst = irg->obst;
510 /* A new obstack, where the reachable nodes will be copied to. */
511 rebirth_obst = (struct obstack *) xmalloc (sizeof (struct obstack));
512 current_ir_graph->obst = rebirth_obst;
513 obstack_init (current_ir_graph->obst);
515 /* We also need a new hash table for cse */
516 del_identities (irg->value_table);
517 irg->value_table = new_identities ();
519 /* Copy the graph from the old to the new obstack */
522 /* Free memory from old unoptimized obstack */
523 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
524 xfree (graveyard_obst); /* ... then free it. */
527 /* inform statistics that the run is over */
528 stat_dead_node_elim_stop(irg);
530 current_ir_graph = rem;
531 interprocedural_view = rem_ipview;
535 * Relink bad predeseccors of a block and store the old in array to the
536 * link field. This function is called by relink_bad_predecessors().
537 * The array of link field starts with the block operand at position 0.
538 * If block has bad predecessors, create a new in array without bad preds.
539 * Otherwise let in array untouched.
541 static void relink_bad_block_predecessors(ir_node *n, void *env) {
542 ir_node **new_in, *irn;
543 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
545 /* if link field of block is NULL, look for bad predecessors otherwise
546 this is allready done */
547 if (get_irn_op(n) == op_Block &&
548 get_irn_link(n) == NULL) {
550 /* save old predecessors in link field (position 0 is the block operand)*/
551 set_irn_link(n, (void *)get_irn_in(n));
553 /* count predecessors without bad nodes */
554 old_irn_arity = get_irn_arity(n);
555 for (i = 0; i < old_irn_arity; i++)
556 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
558 /* arity changing: set new predecessors without bad nodes */
559 if (new_irn_arity < old_irn_arity) {
560 /* get new predecessor array without Block predecessor */
561 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
563 /* set new predeseccors in array */
566 for (i = 1; i < old_irn_arity; i++) {
567 irn = get_irn_n(n, i);
568 if (!is_Bad(irn)) new_in[new_irn_n++] = irn;
571 } /* ir node has bad predecessors */
573 } /* Block is not relinked */
577 * Relinks Bad predecesors from Bocks and Phis called by walker
578 * remove_bad_predecesors(). If n is a Block, call
579 * relink_bad_block_redecessors(). If n is a Phinode, call also the relinking
580 * function of Phi's Block. If this block has bad predecessors, relink preds
583 static void relink_bad_predecessors(ir_node *n, void *env) {
584 ir_node *block, **old_in;
585 int i, old_irn_arity, new_irn_arity;
587 /* relink bad predeseccors of a block */
588 if (get_irn_op(n) == op_Block)
589 relink_bad_block_predecessors(n, env);
591 /* If Phi node relink its block and its predecessors */
592 if (get_irn_op(n) == op_Phi) {
594 /* Relink predeseccors of phi's block */
595 block = get_nodes_block(n);
596 if (get_irn_link(block) == NULL)
597 relink_bad_block_predecessors(block, env);
599 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
600 old_irn_arity = ARR_LEN(old_in);
602 /* Relink Phi predeseccors if count of predeseccors changed */
603 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
604 /* set new predeseccors in array
605 n->in[0] remains the same block */
607 for(i = 1; i < old_irn_arity; i++)
608 if (!is_Bad((ir_node *)old_in[i])) n->in[new_irn_arity++] = n->in[i];
610 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
613 } /* n is a Phi node */
617 * Removes Bad Bad predecesors from Blocks and the corresponding
618 * inputs to Phi nodes as in dead_node_elimination but without
620 * On walking up set the link field to NULL, on walking down call
621 * relink_bad_predecessors() (This function stores the old in array
622 * to the link field and sets a new in array if arity of predecessors
625 void remove_bad_predecessors(ir_graph *irg) {
626 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
630 /*--------------------------------------------------------------------*/
631 /* Funcionality for inlining */
632 /*--------------------------------------------------------------------*/
635 * Copy node for inlineing. Updates attributes that change when
636 * inlineing but not for dead node elimination.
638 * Copies the node by calling copy_node and then updates the entity if
639 * it's a local one. env must be a pointer of the frame type of the
640 * inlined procedure. The new entities must be in the link field of
644 copy_node_inline (ir_node *n, void *env) {
646 type *frame_tp = (type *)env;
649 if (get_irn_op(n) == op_Sel) {
650 new = get_new_node (n);
651 assert(get_irn_op(new) == op_Sel);
652 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
653 set_Sel_entity(new, get_entity_link(get_Sel_entity(n)));
655 } else if (get_irn_op(n) == op_Block) {
656 new = get_new_node (n);
657 new->attr.block.irg = current_ir_graph;
661 static void find_addr(ir_node *node, void *env)
663 if (get_irn_opcode(node) == iro_Proj) {
664 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
670 * currently, we cannot inline two cases:
671 * - call with compound arguments
672 * - graphs that take the address of a parameter
674 * check these conditions here
676 static int can_inline(ir_node *call, ir_graph *called_graph)
678 type *call_type = get_Call_type(call);
679 int params, ress, i, res;
680 assert(is_method_type(call_type));
682 params = get_method_n_params(call_type);
683 ress = get_method_n_ress(call_type);
686 for (i = 0; i < params; ++i) {
687 type *p_type = get_method_param_type(call_type, i);
689 if (is_compound_type(p_type))
694 for (i = 0; i < ress; ++i) {
695 type *r_type = get_method_res_type(call_type, i);
697 if (is_compound_type(r_type))
702 irg_walk_graph(called_graph, find_addr, NULL, &res);
707 int inline_method(ir_node *call, ir_graph *called_graph) {
709 ir_node *post_call, *post_bl;
711 ir_node *end, *end_bl;
715 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
718 irg_inline_property prop = get_irg_inline_property(called_graph);
720 if ( (prop != irg_inline_forced) &&
721 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
723 /* Do not inline variadic functions. */
724 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
727 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
728 get_method_n_params(get_Call_type(call)));
731 * currently, we cannot inline two cases:
732 * - call with compound arguments
733 * - graphs that take the address of a parameter
735 if (! can_inline(call, called_graph))
738 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
739 rem_opt = get_opt_optimize();
742 /* Handle graph state */
743 assert(get_irg_phase_state(current_ir_graph) != phase_building);
744 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
745 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
746 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
747 set_irg_outs_inconsistent(current_ir_graph);
748 set_irg_loopinfo_inconsistent(current_ir_graph);
750 /* -- Check preconditions -- */
751 assert(get_irn_op(call) == op_Call);
752 /* @@@ does not work for InterfaceIII.java after cgana
753 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
754 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
755 get_Call_type(call)));
757 assert(get_type_tpop(get_Call_type(call)) == type_method);
758 if (called_graph == current_ir_graph) {
759 set_optimize(rem_opt);
763 /* here we know we WILL inline, so inform the statistics */
764 stat_inline(call, called_graph);
766 /* -- Decide how to handle exception control flow: Is there a handler
767 for the Call node, or do we branch directly to End on an exception?
769 0 There is a handler.
771 2 Exception handling not represented in Firm. -- */
773 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
774 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
775 assert(get_irn_op(proj) == op_Proj);
776 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
777 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
779 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
780 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
781 else { exc_handling = 2; } /* !Mproj && !Xproj */
786 the procedure and later replaces the Start node of the called graph.
787 Post_call is the old Call node and collects the results of the called
788 graph. Both will end up being a tuple. -- */
789 post_bl = get_nodes_block(call);
790 set_irg_current_block(current_ir_graph, post_bl);
791 /* XxMxPxP of Start + parameter of Call */
792 in[pn_Start_X_initial_exec] = new_Jmp();
793 in[pn_Start_M] = get_Call_mem(call);
794 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
795 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
796 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
797 /* in[pn_Start_P_value_arg_base] = ??? */
798 pre_call = new_Tuple(5, in);
802 The new block gets the ins of the old block, pre_call and all its
803 predecessors and all Phi nodes. -- */
804 part_block(pre_call);
806 /* -- Prepare state for dead node elimination -- */
807 /* Visited flags in calling irg must be >= flag in called irg.
808 Else walker and arity computation will not work. */
809 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
810 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
811 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
812 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
813 /* Set pre_call as new Start node in link field of the start node of
814 calling graph and pre_calls block as new block for the start block
816 Further mark these nodes so that they are not visited by the
818 set_irn_link(get_irg_start(called_graph), pre_call);
819 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
820 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
821 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
822 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
823 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
825 /* Initialize for compaction of in arrays */
826 inc_irg_block_visited(current_ir_graph);
828 /* -- Replicate local entities of the called_graph -- */
829 /* copy the entities. */
830 called_frame = get_irg_frame_type(called_graph);
831 for (i = 0; i < get_class_n_members(called_frame); i++) {
832 entity *new_ent, *old_ent;
833 old_ent = get_class_member(called_frame, i);
834 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
835 set_entity_link(old_ent, new_ent);
838 /* visited is > than that of called graph. With this trick visited will
839 remain unchanged so that an outer walker, e.g., searching the call nodes
840 to inline, calling this inline will not visit the inlined nodes. */
841 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
843 /* -- Performing dead node elimination inlines the graph -- */
844 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
846 /* @@@ endless loops are not copied!! -- they should be, I think... */
847 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
848 get_irg_frame_type(called_graph));
850 /* Repair called_graph */
851 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
852 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
853 set_Block_block_visited(get_irg_start_block(called_graph), 0);
855 /* -- Merge the end of the inlined procedure with the call site -- */
856 /* We will turn the old Call node into a Tuple with the following
859 0: Phi of all Memories of Return statements.
860 1: Jmp from new Block that merges the control flow from all exception
861 predecessors of the old end block.
862 2: Tuple of all arguments.
863 3: Phi of Exception memories.
864 In case the old Call directly branches to End on an exception we don't
865 need the block merging all exceptions nor the Phi of the exception
869 /* -- Precompute some values -- */
870 end_bl = get_new_node(get_irg_end_block(called_graph));
871 end = get_new_node(get_irg_end(called_graph));
872 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
873 n_res = get_method_n_ress(get_Call_type(call));
875 res_pred = (ir_node **) malloc (n_res * sizeof (ir_node *));
876 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
878 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
880 /* -- archive keepalives -- */
881 irn_arity = get_irn_arity(end);
882 for (i = 0; i < irn_arity; i++)
883 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
885 /* The new end node will die. We need not free as the in array is on the obstack:
886 copy_node only generated 'D' arrays. */
888 /* -- Replace Return nodes by Jump nodes. -- */
890 for (i = 0; i < arity; i++) {
892 ret = get_irn_n(end_bl, i);
893 if (get_irn_op(ret) == op_Return) {
894 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
898 set_irn_in(post_bl, n_ret, cf_pred);
900 /* -- Build a Tuple for all results of the method.
901 Add Phi node if there was more than one Return. -- */
902 turn_into_tuple(post_call, 4);
903 /* First the Memory-Phi */
905 for (i = 0; i < arity; i++) {
906 ret = get_irn_n(end_bl, i);
907 if (get_irn_op(ret) == op_Return) {
908 cf_pred[n_ret] = get_Return_mem(ret);
912 phi = new_Phi(n_ret, cf_pred, mode_M);
913 set_Tuple_pred(call, pn_Call_M_regular, phi);
914 /* Conserve Phi-list for further inlinings -- but might be optimized */
915 if (get_nodes_block(phi) == post_bl) {
916 set_irn_link(phi, get_irn_link(post_bl));
917 set_irn_link(post_bl, phi);
919 /* Now the real results */
921 for (j = 0; j < n_res; j++) {
923 for (i = 0; i < arity; i++) {
924 ret = get_irn_n(end_bl, i);
925 if (get_irn_op(ret) == op_Return) {
926 cf_pred[n_ret] = get_Return_res(ret, j);
931 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
935 /* Conserve Phi-list for further inlinings -- but might be optimized */
936 if (get_nodes_block(phi) == post_bl) {
937 set_irn_link(phi, get_irn_link(post_bl));
938 set_irn_link(post_bl, phi);
941 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
943 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
945 /* Finally the exception control flow.
946 We have two (three) possible situations:
947 First if the Call branches to an exception handler: We need to add a Phi node to
948 collect the memory containing the exception objects. Further we need
949 to add another block to get a correct representation of this Phi. To
950 this block we add a Jmp that resolves into the X output of the Call
951 when the Call is turned into a tuple.
952 Second the Call branches to End, the exception is not handled. Just
953 add all inlined exception branches to the End node.
954 Third: there is no Exception edge at all. Handle as case two. */
955 if (exc_handling == 0) {
957 for (i = 0; i < arity; i++) {
959 ret = get_irn_n(end_bl, i);
960 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
961 cf_pred[n_exc] = ret;
966 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
967 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
968 /* The Phi for the memories with the exception objects */
970 for (i = 0; i < arity; i++) {
972 ret = skip_Proj(get_irn_n(end_bl, i));
973 if (get_irn_op(ret) == op_Call) {
974 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
976 } else if (is_fragile_op(ret)) {
977 /* We rely that all cfops have the memory output at the same position. */
978 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
980 } else if (get_irn_op(ret) == op_Raise) {
981 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
985 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
987 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
988 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
991 ir_node *main_end_bl;
992 int main_end_bl_arity;
995 /* assert(exc_handling == 1 || no exceptions. ) */
997 for (i = 0; i < arity; i++) {
998 ir_node *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;
1005 main_end_bl = get_irg_end_block(current_ir_graph);
1006 main_end_bl_arity = get_irn_arity(main_end_bl);
1007 end_preds = (ir_node **) malloc ((n_exc + main_end_bl_arity) * sizeof (ir_node *));
1009 for (i = 0; i < main_end_bl_arity; ++i)
1010 end_preds[i] = get_irn_n(main_end_bl, i);
1011 for (i = 0; i < n_exc; ++i)
1012 end_preds[main_end_bl_arity + i] = cf_pred[i];
1013 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1014 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1015 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1021 #if 0 /* old. now better, correcter, faster implementation. */
1023 /* -- If the exception control flow from the inlined Call directly
1024 branched to the end block we now have the following control
1025 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
1026 remove the Jmp along with it's empty block and add Jmp's
1027 predecessors as predecessors of this end block. No problem if
1028 there is no exception, because then branches Bad to End which
1030 @@@ can't we know this beforehand: by getting the Proj(1) from
1031 the Call link list and checking whether it goes to Proj. */
1032 /* find the problematic predecessor of the end block. */
1033 end_bl = get_irg_end_block(current_ir_graph);
1034 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
1035 cf_op = get_Block_cfgpred(end_bl, i);
1036 if (get_irn_op(cf_op) == op_Proj) {
1037 cf_op = get_Proj_pred(cf_op);
1038 if ((get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
1039 /* There are unoptimized tuples from inlineing before when no exc */
1040 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
1041 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
1042 assert(get_irn_op(cf_op) == op_Jmp);
1048 if (i < get_Block_n_cfgpreds(end_bl)) {
1049 bl = get_nodes_block(cf_op);
1050 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
1051 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
1052 for (j = 0; j < i; j++)
1053 cf_pred[j] = get_Block_cfgpred(end_bl, j);
1054 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
1055 cf_pred[j] = get_Block_cfgpred(bl, j-i);
1056 for (j = j; j < arity; j++)
1057 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
1058 set_irn_in(end_bl, arity, cf_pred);
1060 /* Remove the exception pred from post-call Tuple. */
1061 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1066 /* -- Turn cse back on. -- */
1067 set_optimize(rem_opt);
1072 /********************************************************************/
1073 /* Apply inlineing to small methods. */
1074 /********************************************************************/
1076 /* It makes no sense to inline too many calls in one procedure. Anyways,
1077 I didn't get a version with NEW_ARR_F to run. */
1078 #define MAX_INLINE 1024
1081 * environment for inlining small irgs
1083 typedef struct _inline_env_t {
1085 ir_node *calls[MAX_INLINE];
1089 * Returns the irg called from a Call node. If the irg is not
1090 * known, NULL is returned.
1092 static ir_graph *get_call_called_irg(ir_node *call) {
1094 ir_graph *called_irg = NULL;
1096 assert(get_irn_op(call) == op_Call);
1098 addr = get_Call_ptr(call);
1099 if ((get_irn_op(addr) == op_SymConst) && (get_SymConst_kind (addr) == symconst_addr_ent)) {
1100 called_irg = get_entity_irg(get_SymConst_entity(addr));
1106 static void collect_calls(ir_node *call, void *env) {
1109 if (get_irn_op(call) != op_Call) return;
1111 addr = get_Call_ptr(call);
1113 if (get_irn_op(addr) == op_SymConst) {
1114 if (get_SymConst_kind(addr) == symconst_addr_ent) {
1115 ir_graph *called_irg = get_entity_irg(get_SymConst_entity(addr));
1116 inline_env_t *ienv = (inline_env_t *)env;
1117 if (called_irg && ienv->pos < MAX_INLINE) {
1118 /* The Call node calls a locally defined method. Remember to inline. */
1119 ienv->calls[ienv->pos++] = call;
1126 * Inlines all small methods at call sites where the called address comes
1127 * from a Const node that references the entity representing the called
1129 * The size argument is a rough measure for the code size of the method:
1130 * Methods where the obstack containing the firm graph is smaller than
1133 void inline_small_irgs(ir_graph *irg, int size) {
1135 ir_graph *rem = current_ir_graph;
1136 inline_env_t env /* = {0, NULL}*/;
1138 if (!(get_opt_optimize() && get_opt_inline())) return;
1140 current_ir_graph = irg;
1141 /* Handle graph state */
1142 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1143 free_callee_info(current_ir_graph);
1145 /* Find Call nodes to inline.
1146 (We can not inline during a walk of the graph, as inlineing the same
1147 method several times changes the visited flag of the walked graph:
1148 after the first inlineing visited of the callee equals visited of
1149 the caller. With the next inlineing both are increased.) */
1151 irg_walk(get_irg_end(irg), NULL, collect_calls, &env);
1153 if ((env.pos > 0) && (env.pos < MAX_INLINE)) {
1154 /* There are calls to inline */
1155 collect_phiprojs(irg);
1156 for (i = 0; i < env.pos; i++) {
1158 callee = get_entity_irg(get_SymConst_entity(get_Call_ptr(env.calls[i])));
1159 if (((_obstack_memory_used(callee->obst) - obstack_room(callee->obst)) < size) ||
1160 (get_irg_inline_property(callee) == irg_inline_forced)) {
1161 inline_method(env.calls[i], callee);
1166 current_ir_graph = rem;
1170 * Environment for inlining irgs.
1173 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1174 int n_nodes_orig; /**< for statistics */
1175 eset *call_nodes; /**< All call nodes in this graph */
1177 int n_call_nodes_orig; /**< for statistics */
1178 int n_callers; /**< Number of known graphs that call this graphs. */
1179 int n_callers_orig; /**< for statistics */
1182 static inline_irg_env *new_inline_irg_env(void) {
1183 inline_irg_env *env = malloc(sizeof(inline_irg_env));
1184 env->n_nodes = -2; /* uncount Start, End */
1185 env->n_nodes_orig = -2; /* uncount Start, End */
1186 env->call_nodes = eset_create();
1187 env->n_call_nodes = 0;
1188 env->n_call_nodes_orig = 0;
1190 env->n_callers_orig = 0;
1194 static void free_inline_irg_env(inline_irg_env *env) {
1195 eset_destroy(env->call_nodes);
1199 static void collect_calls2(ir_node *call, void *env) {
1200 inline_irg_env *x = (inline_irg_env *)env;
1201 ir_op *op = get_irn_op(call);
1204 /* count nodes in irg */
1205 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1210 if (op != op_Call) return;
1212 /* collect all call nodes */
1213 eset_insert(x->call_nodes, (void *)call);
1215 x->n_call_nodes_orig++;
1217 /* count all static callers */
1218 callee = get_call_called_irg(call);
1220 ((inline_irg_env *)get_irg_link(callee))->n_callers++;
1221 ((inline_irg_env *)get_irg_link(callee))->n_callers_orig++;
1225 INLINE static int is_leave(ir_graph *irg) {
1226 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1229 INLINE static int is_smaller(ir_graph *callee, int size) {
1230 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1235 * Inlines small leave methods at call sites where the called address comes
1236 * from a Const node that references the entity representing the called
1238 * The size argument is a rough measure for the code size of the method:
1239 * Methods where the obstack containing the firm graph is smaller than
1242 void inline_leave_functions(int maxsize, int leavesize, int size) {
1243 inline_irg_env *env;
1244 int i, n_irgs = get_irp_n_irgs();
1245 ir_graph *rem = current_ir_graph;
1248 if (!(get_opt_optimize() && get_opt_inline())) return;
1250 /* extend all irgs by a temporary data structure for inlineing. */
1251 for (i = 0; i < n_irgs; ++i)
1252 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1254 /* Precompute information in temporary data structure. */
1255 for (i = 0; i < n_irgs; ++i) {
1256 current_ir_graph = get_irp_irg(i);
1257 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1258 free_callee_info(current_ir_graph);
1260 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1261 get_irg_link(current_ir_graph));
1264 /* -- and now inline. -- */
1266 /* Inline leaves recursively -- we might construct new leaves. */
1267 while (did_inline) {
1270 for (i = 0; i < n_irgs; ++i) {
1272 int phiproj_computed = 0;
1274 current_ir_graph = get_irp_irg(i);
1275 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1277 for (call = eset_first(env->call_nodes); call; call = eset_next(env->call_nodes)) {
1278 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1279 ir_graph *callee = get_call_called_irg(call);
1281 if (env->n_nodes > maxsize) continue; // break;
1283 if (callee && (is_leave(callee) && is_smaller(callee, leavesize))) {
1284 if (!phiproj_computed) {
1285 phiproj_computed = 1;
1286 collect_phiprojs(current_ir_graph);
1288 did_inline = inline_method(call, callee);
1291 /* Do some statistics */
1292 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1293 env->n_call_nodes --;
1294 env->n_nodes += callee_env->n_nodes;
1295 callee_env->n_callers--;
1302 /* inline other small functions. */
1303 for (i = 0; i < n_irgs; ++i) {
1306 int phiproj_computed = 0;
1308 current_ir_graph = get_irp_irg(i);
1309 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1311 /* we can not walk and change a set, nor remove from it.
1313 walkset = env->call_nodes;
1314 env->call_nodes = eset_create();
1315 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1316 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1317 ir_graph *callee = get_call_called_irg(call);
1320 ((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1321 (get_irg_inline_property(callee) == irg_inline_forced))) {
1322 if (!phiproj_computed) {
1323 phiproj_computed = 1;
1324 collect_phiprojs(current_ir_graph);
1326 if (inline_method(call, callee)) {
1327 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1328 env->n_call_nodes--;
1329 eset_insert_all(env->call_nodes, callee_env->call_nodes); /* @@@ ??? This are the wrong nodes !? Not the copied ones. */
1330 env->n_call_nodes += callee_env->n_call_nodes;
1331 env->n_nodes += callee_env->n_nodes;
1332 callee_env->n_callers--;
1335 eset_insert(env->call_nodes, call);
1338 eset_destroy(walkset);
1341 for (i = 0; i < n_irgs; ++i) {
1342 current_ir_graph = get_irp_irg(i);
1344 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1345 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1346 (env->n_callers_orig != env->n_callers))
1347 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1348 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1349 env->n_callers_orig, env->n_callers,
1350 get_entity_name(get_irg_entity(current_ir_graph)));
1352 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1355 current_ir_graph = rem;
1358 /*******************************************************************/
1359 /* Code Placement. Pins all floating nodes to a block where they */
1360 /* will be executed only if needed. */
1361 /*******************************************************************/
1364 * Find the earliest correct block for N. --- Place N into the
1365 * same Block as its dominance-deepest Input.
1368 place_floats_early(ir_node *n, pdeq *worklist)
1370 int i, start, irn_arity;
1372 /* we must not run into an infinite loop */
1373 assert (irn_not_visited(n));
1374 mark_irn_visited(n);
1376 /* Place floating nodes. */
1377 if (get_op_pinned(get_irn_op(n)) == op_pin_state_floats) {
1379 ir_node *b = new_Bad(); /* The block to place this node in */
1380 int bad_recursion = is_Bad(get_nodes_block(n));
1382 assert(get_irn_op(n) != op_Block);
1384 if ((get_irn_op(n) == op_Const) ||
1385 (get_irn_op(n) == op_SymConst) ||
1387 (get_irn_op(n) == op_Unknown)) {
1388 /* These nodes will not be placed by the loop below. */
1389 b = get_irg_start_block(current_ir_graph);
1393 /* find the block for this node. */
1394 irn_arity = get_irn_arity(n);
1395 for (i = 0; i < irn_arity; i++) {
1396 ir_node *dep = get_irn_n(n, i);
1399 if ((irn_not_visited(dep))
1400 && (get_op_pinned(get_irn_op(dep)) == op_pin_state_floats)) {
1401 place_floats_early(dep, worklist);
1405 * A node in the Bad block must stay in the bad block,
1406 * so don't compute a new block for it.
1411 /* Because all loops contain at least one op_pin_state_pinned node, now all
1412 our inputs are either op_pin_state_pinned or place_early has already
1413 been finished on them. We do not have any unfinished inputs! */
1414 dep_block = get_nodes_block(dep);
1415 if ((!is_Bad(dep_block)) &&
1416 (get_Block_dom_depth(dep_block) > depth)) {
1418 depth = get_Block_dom_depth(dep_block);
1420 /* Avoid that the node is placed in the Start block */
1421 if ((depth == 1) && (get_Block_dom_depth(get_nodes_block(n)) > 1)) {
1422 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1423 assert(b != get_irg_start_block(current_ir_graph));
1427 set_nodes_block(n, b);
1430 /* Add predecessors of non floating nodes on worklist. */
1431 start = (get_irn_op(n) == op_Block) ? 0 : -1;
1432 irn_arity = get_irn_arity(n);
1433 for (i = start; i < irn_arity; i++) {
1434 ir_node *pred = get_irn_n(n, i);
1435 if (irn_not_visited(pred)) {
1436 pdeq_putr (worklist, pred);
1442 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1443 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1444 * places all floating nodes reachable from its argument through floating
1445 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1447 static INLINE void place_early(pdeq *worklist) {
1449 inc_irg_visited(current_ir_graph);
1451 /* this inits the worklist */
1452 place_floats_early(get_irg_end(current_ir_graph), worklist);
1454 /* Work the content of the worklist. */
1455 while (!pdeq_empty (worklist)) {
1456 ir_node *n = pdeq_getl (worklist);
1457 if (irn_not_visited(n)) place_floats_early(n, worklist);
1460 set_irg_outs_inconsistent(current_ir_graph);
1461 current_ir_graph->op_pin_state_pinned = op_pin_state_pinned;
1465 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1466 * I.e., DCA is the block where we might place PRODUCER.
1467 * A data flow edge points from producer to consumer.
1470 consumer_dom_dca (ir_node *dca, ir_node *consumer, ir_node *producer)
1472 ir_node *block = NULL;
1474 /* Compute the latest block into which we can place a node so that it is
1476 if (get_irn_op(consumer) == op_Phi) {
1477 /* our consumer is a Phi-node, the effective use is in all those
1478 blocks through which the Phi-node reaches producer */
1480 ir_node *phi_block = get_nodes_block(consumer);
1481 irn_arity = get_irn_arity(consumer);
1483 for (i = 0; i < irn_arity; i++) {
1484 if (get_irn_n(consumer, i) == producer) {
1485 block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1489 assert(is_no_Block(consumer));
1490 block = get_nodes_block(consumer);
1493 /* Compute the deepest common ancestor of block and dca. */
1495 if (!dca) return block;
1496 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1497 block = get_Block_idom(block);
1498 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1499 dca = get_Block_idom(dca);
1501 while (block != dca)
1502 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1507 static INLINE int get_irn_loop_depth(ir_node *n) {
1508 return get_loop_depth(get_irn_loop(n));
1512 * Move n to a block with less loop depth than it's current block. The
1513 * new block must be dominated by early.
1516 move_out_of_loops (ir_node *n, ir_node *early)
1518 ir_node *best, *dca;
1522 /* Find the region deepest in the dominator tree dominating
1523 dca with the least loop nesting depth, but still dominated
1524 by our early placement. */
1525 dca = get_nodes_block(n);
1527 while (dca != early) {
1528 dca = get_Block_idom(dca);
1529 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1530 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1534 if (best != get_nodes_block(n)) {
1536 printf("Moving out of loop: "); DDMN(n);
1537 printf(" Outermost block: "); DDMN(early);
1538 printf(" Best block: "); DDMN(best);
1539 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1541 set_nodes_block(n, best);
1546 * Find the latest legal block for N and place N into the
1547 * `optimal' Block between the latest and earliest legal block.
1548 * The `optimal' block is the dominance-deepest block of those
1549 * with the least loop-nesting-depth. This places N out of as many
1550 * loops as possible and then makes it as control dependant as
1554 place_floats_late(ir_node *n, pdeq *worklist)
1559 assert (irn_not_visited(n)); /* no multiple placement */
1561 mark_irn_visited(n);
1563 /* no need to place block nodes, control nodes are already placed. */
1564 if ((get_irn_op(n) != op_Block) &&
1566 (get_irn_mode(n) != mode_X)) {
1567 /* Remember the early placement of this block to move it
1568 out of loop no further than the early placement. */
1569 early = get_nodes_block(n);
1571 /* Do not move code not reachable from Start. For
1572 * these we could not compute dominator information. */
1573 if (is_Bad(early) || get_Block_dom_depth(early) == -1)
1576 /* Assure that our users are all placed, except the Phi-nodes.
1577 --- Each data flow cycle contains at least one Phi-node. We
1578 have to break the `user has to be placed before the
1579 producer' dependence cycle and the Phi-nodes are the
1580 place to do so, because we need to base our placement on the
1581 final region of our users, which is OK with Phi-nodes, as they
1582 are op_pin_state_pinned, and they never have to be placed after a
1583 producer of one of their inputs in the same block anyway. */
1584 for (i = 0; i < get_irn_n_outs(n); i++) {
1585 ir_node *succ = get_irn_out(n, i);
1586 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
1587 place_floats_late(succ, worklist);
1590 /* We have to determine the final block of this node... except for
1592 if ((get_op_pinned(get_irn_op(n)) == op_pin_state_floats) &&
1593 (get_irn_op(n) != op_Const) &&
1594 (get_irn_op(n) != op_SymConst)) {
1595 ir_node *dca = NULL; /* deepest common ancestor in the
1596 dominator tree of all nodes'
1597 blocks depending on us; our final
1598 placement has to dominate DCA. */
1599 for (i = 0; i < get_irn_n_outs(n); i++) {
1600 ir_node *out = get_irn_out(n, i);
1601 /* ignore if out is in dead code */
1602 ir_node *outbl = get_nodes_block(out);
1603 if (is_Bad(outbl) || get_Block_dom_depth(outbl) == -1)
1605 dca = consumer_dom_dca (dca, out, n);
1608 set_nodes_block(n, dca);
1610 move_out_of_loops (n, early);
1612 /* else all outs are in dead code */
1616 /* Add predecessors of all non-floating nodes on list. (Those of floating
1617 nodes are placeded already and therefore are marked.) */
1618 for (i = 0; i < get_irn_n_outs(n); i++) {
1619 if (irn_not_visited(get_irn_out(n, i))) {
1620 pdeq_putr (worklist, get_irn_out(n, i));
1625 static INLINE void place_late(pdeq *worklist) {
1627 inc_irg_visited(current_ir_graph);
1629 /* This fills the worklist initially. */
1630 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1632 /* And now empty the worklist again... */
1633 while (!pdeq_empty (worklist)) {
1634 ir_node *n = pdeq_getl (worklist);
1635 if (irn_not_visited(n)) place_floats_late(n, worklist);
1639 void place_code(ir_graph *irg) {
1641 ir_graph *rem = current_ir_graph;
1643 current_ir_graph = irg;
1645 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1647 /* Handle graph state */
1648 assert(get_irg_phase_state(irg) != phase_building);
1649 if (get_irg_dom_state(irg) != dom_consistent)
1652 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1653 free_loop_information(irg);
1654 construct_backedges(irg);
1657 /* Place all floating nodes as early as possible. This guarantees
1658 a legal code placement. */
1659 worklist = new_pdeq();
1660 place_early(worklist);
1662 /* place_early invalidates the outs, place_late needs them. */
1664 /* Now move the nodes down in the dominator tree. This reduces the
1665 unnecessary executions of the node. */
1666 place_late(worklist);
1668 set_irg_outs_inconsistent(current_ir_graph);
1669 set_irg_loopinfo_inconsistent(current_ir_graph);
1671 current_ir_graph = rem;
1675 * Called by walker of remove_critical_cf_edges().
1677 * Place an empty block to an edge between a blocks of multiple
1678 * predecessors and a block of multiple successors.
1681 * @param env Environment of walker. This field is unused and has
1684 static void walk_critical_cf_edges(ir_node *n, void *env) {
1686 ir_node *pre, *block, **in, *jmp;
1688 /* Block has multiple predecessors */
1689 if ((op_Block == get_irn_op(n)) &&
1690 (get_irn_arity(n) > 1)) {
1691 arity = get_irn_arity(n);
1693 if (n == get_irg_end_block(current_ir_graph))
1694 return; /* No use to add a block here. */
1696 for (i=0; i<arity; i++) {
1697 pre = get_irn_n(n, i);
1698 /* Predecessor has multiple successors. Insert new flow edge */
1699 if ((NULL != pre) &&
1700 (op_Proj == get_irn_op(pre)) &&
1701 op_Raise != get_irn_op(skip_Proj(pre))) {
1703 /* set predecessor array for new block */
1704 in = NEW_ARR_D (ir_node *, current_ir_graph->obst, 1);
1705 /* set predecessor of new block */
1707 block = new_Block(1, in);
1708 /* insert new jmp node to new block */
1709 set_cur_block(block);
1712 /* set successor of new block */
1713 set_irn_n(n, i, jmp);
1715 } /* predecessor has multiple successors */
1716 } /* for all predecessors */
1717 } /* n is a block */
1720 void remove_critical_cf_edges(ir_graph *irg) {
1721 if (get_opt_critical_edges())
1722 irg_walk_graph(irg, NULL, walk_critical_cf_edges, NULL);