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);
254 /* for easier debugging, we want to copy the node numbers too */
255 nn->node_nr = n->node_nr;
259 /* printf("\n old node: "); DDMSG2(n);
260 printf(" new node: "); DDMSG2(nn); */
264 * Copies new predecessors of old node to new node remembered in link.
265 * Spare the Bad predecessors of Phi and Block nodes.
268 copy_preds (ir_node *n, void *env) {
272 nn = get_new_node(n);
274 /* printf("\n old node: "); DDMSG2(n);
275 printf(" new node: "); DDMSG2(nn);
276 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
278 if (get_irn_opcode(n) == iro_Block) {
279 /* Don't copy Bad nodes. */
281 irn_arity = get_irn_arity(n);
282 for (i = 0; i < irn_arity; i++)
283 if (get_irn_opcode(get_irn_n(n, i)) != iro_Bad) {
284 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
285 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
288 /* repair the block visited flag from above misuse. Repair it in both
289 graphs so that the old one can still be used. */
290 set_Block_block_visited(nn, 0);
291 set_Block_block_visited(n, 0);
292 /* Local optimization could not merge two subsequent blocks if
293 in array contained Bads. Now it's possible.
294 We don't call optimize_in_place as it requires
295 that the fields in ir_graph are set properly. */
296 if ((get_opt_control_flow_straightening()) &&
297 (get_Block_n_cfgpreds(nn) == 1) &&
298 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
299 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
301 /* Jmp jumps into the block it is in -- deal self cycle. */
302 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
303 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
308 } else if (get_irn_opcode(n) == iro_Phi) {
309 /* Don't copy node if corresponding predecessor in block is Bad.
310 The Block itself should not be Bad. */
311 block = get_nodes_block(n);
312 set_irn_n (nn, -1, get_new_node(block));
314 irn_arity = get_irn_arity(n);
315 for (i = 0; i < irn_arity; i++)
316 if (get_irn_opcode(get_irn_n(block, i)) != iro_Bad) {
317 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
318 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
321 /* If the pre walker reached this Phi after the post walker visited the
322 block block_visited is > 0. */
323 set_Block_block_visited(get_nodes_block(n), 0);
324 /* Compacting the Phi's ins might generate Phis with only one
326 if (get_irn_arity(n) == 1)
327 exchange(n, get_irn_n(n, 0));
329 irn_arity = get_irn_arity(n);
330 for (i = -1; i < irn_arity; i++)
331 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
333 /* Now the new node is complete. We can add it to the hash table for cse.
334 @@@ inlinening aborts if we identify End. Why? */
335 if(get_irn_op(nn) != op_End)
336 add_identities (current_ir_graph->value_table, nn);
340 * Copies the graph recursively, compacts the keepalive of the end node.
342 * @param copy_node_nr If non-zero, the node number will be copied
345 copy_graph (int copy_node_nr) {
346 ir_node *oe, *ne, *ob, *nb; /* old end, new end, old bad, new bad */
347 ir_node *ka; /* keep alive */
350 oe = get_irg_end(current_ir_graph);
351 /* copy the end node by hand, allocate dynamic in array! */
352 ne = new_ir_node(get_irn_dbg_info(oe),
359 /* Copy the attributes. Well, there might be some in the future... */
361 set_new_node(oe, ne);
363 ob = get_irg_bad(current_ir_graph);
364 nb = new_ir_node(get_irn_dbg_info(ob),
371 set_new_node(ob, nb);
373 /* copy the live nodes */
374 irg_walk(get_nodes_block(oe), copy_node, copy_preds, (void *)copy_node_nr);
375 /* copy_preds for the end node ... */
376 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
378 /*- ... and now the keep alives. -*/
379 /* First pick the not marked block nodes and walk them. We must pick these
380 first as else we will oversee blocks reachable from Phis. */
381 irn_arity = get_irn_arity(oe);
382 for (i = 0; i < irn_arity; i++) {
383 ka = get_irn_intra_n(oe, i);
384 if ((get_irn_op(ka) == op_Block) &&
385 (get_irn_visited(ka) < get_irg_visited(current_ir_graph))) {
386 /* We must keep the block alive and copy everything reachable */
387 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
388 irg_walk(ka, copy_node, copy_preds, (void *)copy_node_nr);
389 add_End_keepalive(ne, get_new_node(ka));
393 /* Now pick the Phis. Here we will keep all! */
394 irn_arity = get_irn_arity(oe);
395 for (i = 0; i < irn_arity; i++) {
396 ka = get_irn_intra_n(oe, i);
397 if ((get_irn_op(ka) == op_Phi)) {
398 if (get_irn_visited(ka) < get_irg_visited(current_ir_graph)) {
399 /* We didn't copy the Phi yet. */
400 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
401 irg_walk(ka, copy_node, copy_preds, (void *)copy_node_nr);
403 add_End_keepalive(ne, get_new_node(ka));
407 /* start block somtimes only reached after keep alives */
408 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
412 * Copies the graph reachable from current_ir_graph->end to the obstack
413 * in current_ir_graph and fixes the environment.
414 * Then fixes the fields in current_ir_graph containing nodes of the
417 * @param copy_node_nr If non-zero, the node number will be copied
420 copy_graph_env (int copy_node_nr) {
422 /* Not all nodes remembered in current_ir_graph might be reachable
423 from the end node. Assure their link is set to NULL, so that
424 we can test whether new nodes have been computed. */
425 set_irn_link(get_irg_frame (current_ir_graph), NULL);
426 set_irn_link(get_irg_globals (current_ir_graph), NULL);
427 set_irn_link(get_irg_args (current_ir_graph), NULL);
428 set_irn_link(get_irg_initial_mem(current_ir_graph), NULL);
430 /* we use the block walk flag for removing Bads from Blocks ins. */
431 inc_irg_block_visited(current_ir_graph);
434 copy_graph(copy_node_nr);
436 /* fix the fields in current_ir_graph */
437 old_end = get_irg_end(current_ir_graph);
438 set_irg_end (current_ir_graph, get_new_node(old_end));
439 set_irg_end_except (current_ir_graph, get_irg_end(current_ir_graph));
440 set_irg_end_reg (current_ir_graph, get_irg_end(current_ir_graph));
442 set_irg_end_block (current_ir_graph, get_new_node(get_irg_end_block(current_ir_graph)));
443 if (get_irn_link(get_irg_frame(current_ir_graph)) == NULL) {
444 copy_node (get_irg_frame(current_ir_graph), (void *)copy_node_nr);
445 copy_preds(get_irg_frame(current_ir_graph), NULL);
447 if (get_irn_link(get_irg_globals(current_ir_graph)) == NULL) {
448 copy_node (get_irg_globals(current_ir_graph), (void *)copy_node_nr);
449 copy_preds(get_irg_globals(current_ir_graph), NULL);
451 if (get_irn_link(get_irg_initial_mem(current_ir_graph)) == NULL) {
452 copy_node (get_irg_initial_mem(current_ir_graph), (void *)copy_node_nr);
453 copy_preds(get_irg_initial_mem(current_ir_graph), NULL);
455 if (get_irn_link(get_irg_args(current_ir_graph)) == NULL) {
456 copy_node (get_irg_args(current_ir_graph), (void *)copy_node_nr);
457 copy_preds(get_irg_args(current_ir_graph), NULL);
459 set_irg_start (current_ir_graph, get_new_node(get_irg_start(current_ir_graph)));
461 set_irg_start_block(current_ir_graph,
462 get_new_node(get_irg_start_block(current_ir_graph)));
463 set_irg_frame (current_ir_graph, get_new_node(get_irg_frame(current_ir_graph)));
464 set_irg_globals (current_ir_graph, get_new_node(get_irg_globals(current_ir_graph)));
465 set_irg_initial_mem(current_ir_graph, get_new_node(get_irg_initial_mem(current_ir_graph)));
466 set_irg_args (current_ir_graph, get_new_node(get_irg_args(current_ir_graph)));
468 if (get_irn_link(get_irg_bad(current_ir_graph)) == NULL) {
469 copy_node(get_irg_bad(current_ir_graph), (void *)copy_node_nr);
470 copy_preds(get_irg_bad(current_ir_graph), NULL);
472 set_irg_bad(current_ir_graph, get_new_node(get_irg_bad(current_ir_graph)));
476 * Copies all reachable nodes to a new obstack. Removes bad inputs
477 * from block nodes and the corresponding inputs from Phi nodes.
478 * Merges single exit blocks with single entry blocks and removes
480 * Adds all new nodes to a new hash table for cse. Does not
481 * perform cse, so the hash table might contain common subexpressions.
484 dead_node_elimination(ir_graph *irg) {
486 int rem_ipview = interprocedural_view;
487 struct obstack *graveyard_obst = NULL;
488 struct obstack *rebirth_obst = NULL;
490 /* inform statistics that we started a dead-node elimination run */
491 stat_dead_node_elim_start(irg);
493 /* Remember external state of current_ir_graph. */
494 rem = current_ir_graph;
495 current_ir_graph = irg;
496 interprocedural_view = 0;
498 /* Handle graph state */
499 assert(get_irg_phase_state(current_ir_graph) != phase_building);
500 free_callee_info(current_ir_graph);
501 free_outs(current_ir_graph);
502 /* @@@ so far we loose loops when copying */
503 free_loop_information(current_ir_graph);
505 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
507 /* A quiet place, where the old obstack can rest in peace,
508 until it will be cremated. */
509 graveyard_obst = irg->obst;
511 /* A new obstack, where the reachable nodes will be copied to. */
512 rebirth_obst = (struct obstack *) xmalloc (sizeof (struct obstack));
513 current_ir_graph->obst = rebirth_obst;
514 obstack_init (current_ir_graph->obst);
516 /* We also need a new hash table for cse */
517 del_identities (irg->value_table);
518 irg->value_table = new_identities ();
520 /* Copy the graph from the old to the new obstack */
523 /* Free memory from old unoptimized obstack */
524 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
525 xfree (graveyard_obst); /* ... then free it. */
528 /* inform statistics that the run is over */
529 stat_dead_node_elim_stop(irg);
531 current_ir_graph = rem;
532 interprocedural_view = rem_ipview;
536 * Relink bad predeseccors of a block and store the old in array to the
537 * link field. This function is called by relink_bad_predecessors().
538 * The array of link field starts with the block operand at position 0.
539 * If block has bad predecessors, create a new in array without bad preds.
540 * Otherwise let in array untouched.
542 static void relink_bad_block_predecessors(ir_node *n, void *env) {
543 ir_node **new_in, *irn;
544 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
546 /* if link field of block is NULL, look for bad predecessors otherwise
547 this is allready done */
548 if (get_irn_op(n) == op_Block &&
549 get_irn_link(n) == NULL) {
551 /* save old predecessors in link field (position 0 is the block operand)*/
552 set_irn_link(n, (void *)get_irn_in(n));
554 /* count predecessors without bad nodes */
555 old_irn_arity = get_irn_arity(n);
556 for (i = 0; i < old_irn_arity; i++)
557 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
559 /* arity changing: set new predecessors without bad nodes */
560 if (new_irn_arity < old_irn_arity) {
561 /* get new predecessor array without Block predecessor */
562 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
564 /* set new predeseccors in array */
567 for (i = 1; i < old_irn_arity; i++) {
568 irn = get_irn_n(n, i);
569 if (!is_Bad(irn)) new_in[new_irn_n++] = irn;
572 } /* ir node has bad predecessors */
574 } /* Block is not relinked */
578 * Relinks Bad predecesors from Bocks and Phis called by walker
579 * remove_bad_predecesors(). If n is a Block, call
580 * relink_bad_block_redecessors(). If n is a Phinode, call also the relinking
581 * function of Phi's Block. If this block has bad predecessors, relink preds
584 static void relink_bad_predecessors(ir_node *n, void *env) {
585 ir_node *block, **old_in;
586 int i, old_irn_arity, new_irn_arity;
588 /* relink bad predeseccors of a block */
589 if (get_irn_op(n) == op_Block)
590 relink_bad_block_predecessors(n, env);
592 /* If Phi node relink its block and its predecessors */
593 if (get_irn_op(n) == op_Phi) {
595 /* Relink predeseccors of phi's block */
596 block = get_nodes_block(n);
597 if (get_irn_link(block) == NULL)
598 relink_bad_block_predecessors(block, env);
600 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
601 old_irn_arity = ARR_LEN(old_in);
603 /* Relink Phi predeseccors if count of predeseccors changed */
604 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
605 /* set new predeseccors in array
606 n->in[0] remains the same block */
608 for(i = 1; i < old_irn_arity; i++)
609 if (!is_Bad((ir_node *)old_in[i])) n->in[new_irn_arity++] = n->in[i];
611 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
614 } /* n is a Phi node */
618 * Removes Bad Bad predecesors from Blocks and the corresponding
619 * inputs to Phi nodes as in dead_node_elimination but without
621 * On walking up set the link field to NULL, on walking down call
622 * relink_bad_predecessors() (This function stores the old in array
623 * to the link field and sets a new in array if arity of predecessors
626 void remove_bad_predecessors(ir_graph *irg) {
627 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
631 /*--------------------------------------------------------------------*/
632 /* Funcionality for inlining */
633 /*--------------------------------------------------------------------*/
636 * Copy node for inlineing. Updates attributes that change when
637 * inlineing but not for dead node elimination.
639 * Copies the node by calling copy_node and then updates the entity if
640 * it's a local one. env must be a pointer of the frame type of the
641 * inlined procedure. The new entities must be in the link field of
645 copy_node_inline (ir_node *n, void *env) {
647 type *frame_tp = (type *)env;
650 if (get_irn_op(n) == op_Sel) {
651 new = get_new_node (n);
652 assert(get_irn_op(new) == op_Sel);
653 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
654 set_Sel_entity(new, get_entity_link(get_Sel_entity(n)));
656 } else if (get_irn_op(n) == op_Block) {
657 new = get_new_node (n);
658 new->attr.block.irg = current_ir_graph;
662 static void find_addr(ir_node *node, void *env)
664 if (get_irn_opcode(node) == iro_Proj) {
665 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
671 * currently, we cannot inline two cases:
672 * - call with compound arguments
673 * - graphs that take the address of a parameter
675 * check these conditions here
677 static int can_inline(ir_node *call, ir_graph *called_graph)
679 type *call_type = get_Call_type(call);
680 int params, ress, i, res;
681 assert(is_method_type(call_type));
683 params = get_method_n_params(call_type);
684 ress = get_method_n_ress(call_type);
687 for (i = 0; i < params; ++i) {
688 type *p_type = get_method_param_type(call_type, i);
690 if (is_compound_type(p_type))
695 for (i = 0; i < ress; ++i) {
696 type *r_type = get_method_res_type(call_type, i);
698 if (is_compound_type(r_type))
703 irg_walk_graph(called_graph, find_addr, NULL, &res);
708 int inline_method(ir_node *call, ir_graph *called_graph) {
710 ir_node *post_call, *post_bl;
712 ir_node *end, *end_bl;
716 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
719 irg_inline_property prop = get_irg_inline_property(called_graph);
721 if ( (prop != irg_inline_forced) &&
722 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
724 /* Do not inline variadic functions. */
725 if (get_method_variadicity(get_entity_type(get_irg_entity(called_graph))) == variadicity_variadic)
728 assert(get_method_n_params(get_entity_type(get_irg_entity(called_graph))) ==
729 get_method_n_params(get_Call_type(call)));
732 * currently, we cannot inline two cases:
733 * - call with compound arguments
734 * - graphs that take the address of a parameter
736 if (! can_inline(call, called_graph))
739 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
740 rem_opt = get_opt_optimize();
743 /* Handle graph state */
744 assert(get_irg_phase_state(current_ir_graph) != phase_building);
745 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
746 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
747 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
748 set_irg_outs_inconsistent(current_ir_graph);
749 set_irg_loopinfo_inconsistent(current_ir_graph);
751 /* -- Check preconditions -- */
752 assert(get_irn_op(call) == op_Call);
753 /* @@@ does not work for InterfaceIII.java after cgana
754 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
755 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
756 get_Call_type(call)));
758 assert(get_type_tpop(get_Call_type(call)) == type_method);
759 if (called_graph == current_ir_graph) {
760 set_optimize(rem_opt);
764 /* here we know we WILL inline, so inform the statistics */
765 stat_inline(call, called_graph);
767 /* -- Decide how to handle exception control flow: Is there a handler
768 for the Call node, or do we branch directly to End on an exception?
770 0 There is a handler.
772 2 Exception handling not represented in Firm. -- */
774 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
775 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
776 assert(get_irn_op(proj) == op_Proj);
777 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
778 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
780 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
781 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
782 else { exc_handling = 2; } /* !Mproj && !Xproj */
787 the procedure and later replaces the Start node of the called graph.
788 Post_call is the old Call node and collects the results of the called
789 graph. Both will end up being a tuple. -- */
790 post_bl = get_nodes_block(call);
791 set_irg_current_block(current_ir_graph, post_bl);
792 /* XxMxPxP of Start + parameter of Call */
793 in[pn_Start_X_initial_exec] = new_Jmp();
794 in[pn_Start_M] = get_Call_mem(call);
795 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
796 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
797 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
798 /* in[pn_Start_P_value_arg_base] = ??? */
799 pre_call = new_Tuple(5, in);
803 The new block gets the ins of the old block, pre_call and all its
804 predecessors and all Phi nodes. -- */
805 part_block(pre_call);
807 /* -- Prepare state for dead node elimination -- */
808 /* Visited flags in calling irg must be >= flag in called irg.
809 Else walker and arity computation will not work. */
810 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
811 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
812 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
813 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
814 /* Set pre_call as new Start node in link field of the start node of
815 calling graph and pre_calls block as new block for the start block
817 Further mark these nodes so that they are not visited by the
819 set_irn_link(get_irg_start(called_graph), pre_call);
820 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
821 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
822 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
823 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
824 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
826 /* Initialize for compaction of in arrays */
827 inc_irg_block_visited(current_ir_graph);
829 /* -- Replicate local entities of the called_graph -- */
830 /* copy the entities. */
831 called_frame = get_irg_frame_type(called_graph);
832 for (i = 0; i < get_class_n_members(called_frame); i++) {
833 entity *new_ent, *old_ent;
834 old_ent = get_class_member(called_frame, i);
835 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
836 set_entity_link(old_ent, new_ent);
839 /* visited is > than that of called graph. With this trick visited will
840 remain unchanged so that an outer walker, e.g., searching the call nodes
841 to inline, calling this inline will not visit the inlined nodes. */
842 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
844 /* -- Performing dead node elimination inlines the graph -- */
845 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
847 /* @@@ endless loops are not copied!! -- they should be, I think... */
848 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
849 get_irg_frame_type(called_graph));
851 /* Repair called_graph */
852 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
853 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
854 set_Block_block_visited(get_irg_start_block(called_graph), 0);
856 /* -- Merge the end of the inlined procedure with the call site -- */
857 /* We will turn the old Call node into a Tuple with the following
860 0: Phi of all Memories of Return statements.
861 1: Jmp from new Block that merges the control flow from all exception
862 predecessors of the old end block.
863 2: Tuple of all arguments.
864 3: Phi of Exception memories.
865 In case the old Call directly branches to End on an exception we don't
866 need the block merging all exceptions nor the Phi of the exception
870 /* -- Precompute some values -- */
871 end_bl = get_new_node(get_irg_end_block(called_graph));
872 end = get_new_node(get_irg_end(called_graph));
873 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
874 n_res = get_method_n_ress(get_Call_type(call));
876 res_pred = (ir_node **) malloc (n_res * sizeof (ir_node *));
877 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
879 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
881 /* -- archive keepalives -- */
882 irn_arity = get_irn_arity(end);
883 for (i = 0; i < irn_arity; i++)
884 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
886 /* The new end node will die. We need not free as the in array is on the obstack:
887 copy_node only generated 'D' arrays. */
889 /* -- Replace Return nodes by Jump nodes. -- */
891 for (i = 0; i < arity; i++) {
893 ret = get_irn_n(end_bl, i);
894 if (get_irn_op(ret) == op_Return) {
895 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
899 set_irn_in(post_bl, n_ret, cf_pred);
901 /* -- Build a Tuple for all results of the method.
902 Add Phi node if there was more than one Return. -- */
903 turn_into_tuple(post_call, 4);
904 /* First the Memory-Phi */
906 for (i = 0; i < arity; i++) {
907 ret = get_irn_n(end_bl, i);
908 if (get_irn_op(ret) == op_Return) {
909 cf_pred[n_ret] = get_Return_mem(ret);
913 phi = new_Phi(n_ret, cf_pred, mode_M);
914 set_Tuple_pred(call, pn_Call_M_regular, phi);
915 /* Conserve Phi-list for further inlinings -- but might be optimized */
916 if (get_nodes_block(phi) == post_bl) {
917 set_irn_link(phi, get_irn_link(post_bl));
918 set_irn_link(post_bl, phi);
920 /* Now the real results */
922 for (j = 0; j < n_res; j++) {
924 for (i = 0; i < arity; i++) {
925 ret = get_irn_n(end_bl, i);
926 if (get_irn_op(ret) == op_Return) {
927 cf_pred[n_ret] = get_Return_res(ret, j);
932 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
936 /* Conserve Phi-list for further inlinings -- but might be optimized */
937 if (get_nodes_block(phi) == post_bl) {
938 set_irn_link(phi, get_irn_link(post_bl));
939 set_irn_link(post_bl, phi);
942 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
944 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
946 /* Finally the exception control flow.
947 We have two (three) possible situations:
948 First if the Call branches to an exception handler: We need to add a Phi node to
949 collect the memory containing the exception objects. Further we need
950 to add another block to get a correct representation of this Phi. To
951 this block we add a Jmp that resolves into the X output of the Call
952 when the Call is turned into a tuple.
953 Second the Call branches to End, the exception is not handled. Just
954 add all inlined exception branches to the End node.
955 Third: there is no Exception edge at all. Handle as case two. */
956 if (exc_handling == 0) {
958 for (i = 0; i < arity; i++) {
960 ret = get_irn_n(end_bl, i);
961 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
962 cf_pred[n_exc] = ret;
967 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
968 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
969 /* The Phi for the memories with the exception objects */
971 for (i = 0; i < arity; i++) {
973 ret = skip_Proj(get_irn_n(end_bl, i));
974 if (get_irn_op(ret) == op_Call) {
975 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
977 } else if (is_fragile_op(ret)) {
978 /* We rely that all cfops have the memory output at the same position. */
979 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
981 } else if (get_irn_op(ret) == op_Raise) {
982 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
986 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
988 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
989 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
992 ir_node *main_end_bl;
993 int main_end_bl_arity;
996 /* assert(exc_handling == 1 || no exceptions. ) */
998 for (i = 0; i < arity; i++) {
999 ir_node *ret = get_irn_n(end_bl, i);
1001 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
1002 cf_pred[n_exc] = ret;
1006 main_end_bl = get_irg_end_block(current_ir_graph);
1007 main_end_bl_arity = get_irn_arity(main_end_bl);
1008 end_preds = (ir_node **) malloc ((n_exc + main_end_bl_arity) * sizeof (ir_node *));
1010 for (i = 0; i < main_end_bl_arity; ++i)
1011 end_preds[i] = get_irn_n(main_end_bl, i);
1012 for (i = 0; i < n_exc; ++i)
1013 end_preds[main_end_bl_arity + i] = cf_pred[i];
1014 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1015 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1016 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1022 #if 0 /* old. now better, correcter, faster implementation. */
1024 /* -- If the exception control flow from the inlined Call directly
1025 branched to the end block we now have the following control
1026 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
1027 remove the Jmp along with it's empty block and add Jmp's
1028 predecessors as predecessors of this end block. No problem if
1029 there is no exception, because then branches Bad to End which
1031 @@@ can't we know this beforehand: by getting the Proj(1) from
1032 the Call link list and checking whether it goes to Proj. */
1033 /* find the problematic predecessor of the end block. */
1034 end_bl = get_irg_end_block(current_ir_graph);
1035 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
1036 cf_op = get_Block_cfgpred(end_bl, i);
1037 if (get_irn_op(cf_op) == op_Proj) {
1038 cf_op = get_Proj_pred(cf_op);
1039 if ((get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
1040 /* There are unoptimized tuples from inlineing before when no exc */
1041 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
1042 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
1043 assert(get_irn_op(cf_op) == op_Jmp);
1049 if (i < get_Block_n_cfgpreds(end_bl)) {
1050 bl = get_nodes_block(cf_op);
1051 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
1052 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
1053 for (j = 0; j < i; j++)
1054 cf_pred[j] = get_Block_cfgpred(end_bl, j);
1055 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
1056 cf_pred[j] = get_Block_cfgpred(bl, j-i);
1057 for (j = j; j < arity; j++)
1058 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
1059 set_irn_in(end_bl, arity, cf_pred);
1061 /* Remove the exception pred from post-call Tuple. */
1062 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1067 /* -- Turn cse back on. -- */
1068 set_optimize(rem_opt);
1073 /********************************************************************/
1074 /* Apply inlineing to small methods. */
1075 /********************************************************************/
1077 /* It makes no sense to inline too many calls in one procedure. Anyways,
1078 I didn't get a version with NEW_ARR_F to run. */
1079 #define MAX_INLINE 1024
1082 * environment for inlining small irgs
1084 typedef struct _inline_env_t {
1086 ir_node *calls[MAX_INLINE];
1090 * Returns the irg called from a Call node. If the irg is not
1091 * known, NULL is returned.
1093 static ir_graph *get_call_called_irg(ir_node *call) {
1095 ir_graph *called_irg = NULL;
1097 assert(get_irn_op(call) == op_Call);
1099 addr = get_Call_ptr(call);
1100 if ((get_irn_op(addr) == op_SymConst) && (get_SymConst_kind (addr) == symconst_addr_ent)) {
1101 called_irg = get_entity_irg(get_SymConst_entity(addr));
1107 static void collect_calls(ir_node *call, void *env) {
1110 if (get_irn_op(call) != op_Call) return;
1112 addr = get_Call_ptr(call);
1114 if (get_irn_op(addr) == op_SymConst) {
1115 if (get_SymConst_kind(addr) == symconst_addr_ent) {
1116 ir_graph *called_irg = get_entity_irg(get_SymConst_entity(addr));
1117 inline_env_t *ienv = (inline_env_t *)env;
1118 if (called_irg && ienv->pos < MAX_INLINE) {
1119 /* The Call node calls a locally defined method. Remember to inline. */
1120 ienv->calls[ienv->pos++] = call;
1127 * Inlines all small methods at call sites where the called address comes
1128 * from a Const node that references the entity representing the called
1130 * The size argument is a rough measure for the code size of the method:
1131 * Methods where the obstack containing the firm graph is smaller than
1134 void inline_small_irgs(ir_graph *irg, int size) {
1136 ir_graph *rem = current_ir_graph;
1137 inline_env_t env /* = {0, NULL}*/;
1139 if (!(get_opt_optimize() && get_opt_inline())) return;
1141 current_ir_graph = irg;
1142 /* Handle graph state */
1143 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1144 free_callee_info(current_ir_graph);
1146 /* Find Call nodes to inline.
1147 (We can not inline during a walk of the graph, as inlineing the same
1148 method several times changes the visited flag of the walked graph:
1149 after the first inlineing visited of the callee equals visited of
1150 the caller. With the next inlineing both are increased.) */
1152 irg_walk(get_irg_end(irg), NULL, collect_calls, &env);
1154 if ((env.pos > 0) && (env.pos < MAX_INLINE)) {
1155 /* There are calls to inline */
1156 collect_phiprojs(irg);
1157 for (i = 0; i < env.pos; i++) {
1159 callee = get_entity_irg(get_SymConst_entity(get_Call_ptr(env.calls[i])));
1160 if (((_obstack_memory_used(callee->obst) - obstack_room(callee->obst)) < size) ||
1161 (get_irg_inline_property(callee) == irg_inline_forced)) {
1162 inline_method(env.calls[i], callee);
1167 current_ir_graph = rem;
1171 * Environment for inlining irgs.
1174 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1175 int n_nodes_orig; /**< for statistics */
1176 eset *call_nodes; /**< All call nodes in this graph */
1178 int n_call_nodes_orig; /**< for statistics */
1179 int n_callers; /**< Number of known graphs that call this graphs. */
1180 int n_callers_orig; /**< for statistics */
1183 static inline_irg_env *new_inline_irg_env(void) {
1184 inline_irg_env *env = malloc(sizeof(inline_irg_env));
1185 env->n_nodes = -2; /* uncount Start, End */
1186 env->n_nodes_orig = -2; /* uncount Start, End */
1187 env->call_nodes = eset_create();
1188 env->n_call_nodes = 0;
1189 env->n_call_nodes_orig = 0;
1191 env->n_callers_orig = 0;
1195 static void free_inline_irg_env(inline_irg_env *env) {
1196 eset_destroy(env->call_nodes);
1200 static void collect_calls2(ir_node *call, void *env) {
1201 inline_irg_env *x = (inline_irg_env *)env;
1202 ir_op *op = get_irn_op(call);
1205 /* count nodes in irg */
1206 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1211 if (op != op_Call) return;
1213 /* collect all call nodes */
1214 eset_insert(x->call_nodes, (void *)call);
1216 x->n_call_nodes_orig++;
1218 /* count all static callers */
1219 callee = get_call_called_irg(call);
1221 ((inline_irg_env *)get_irg_link(callee))->n_callers++;
1222 ((inline_irg_env *)get_irg_link(callee))->n_callers_orig++;
1226 INLINE static int is_leave(ir_graph *irg) {
1227 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1230 INLINE static int is_smaller(ir_graph *callee, int size) {
1231 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1236 * Inlines small leave methods at call sites where the called address comes
1237 * from a Const node that references the entity representing the called
1239 * The size argument is a rough measure for the code size of the method:
1240 * Methods where the obstack containing the firm graph is smaller than
1243 void inline_leave_functions(int maxsize, int leavesize, int size) {
1244 inline_irg_env *env;
1245 int i, n_irgs = get_irp_n_irgs();
1246 ir_graph *rem = current_ir_graph;
1249 if (!(get_opt_optimize() && get_opt_inline())) return;
1251 /* extend all irgs by a temporary data structure for inlineing. */
1252 for (i = 0; i < n_irgs; ++i)
1253 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1255 /* Precompute information in temporary data structure. */
1256 for (i = 0; i < n_irgs; ++i) {
1257 current_ir_graph = get_irp_irg(i);
1258 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1259 free_callee_info(current_ir_graph);
1261 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1262 get_irg_link(current_ir_graph));
1265 /* -- and now inline. -- */
1267 /* Inline leaves recursively -- we might construct new leaves. */
1268 while (did_inline) {
1271 for (i = 0; i < n_irgs; ++i) {
1273 int phiproj_computed = 0;
1275 current_ir_graph = get_irp_irg(i);
1276 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1278 for (call = eset_first(env->call_nodes); call; call = eset_next(env->call_nodes)) {
1279 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1280 ir_graph *callee = get_call_called_irg(call);
1282 if (env->n_nodes > maxsize) continue; // break;
1284 if (callee && (is_leave(callee) && is_smaller(callee, leavesize))) {
1285 if (!phiproj_computed) {
1286 phiproj_computed = 1;
1287 collect_phiprojs(current_ir_graph);
1289 did_inline = inline_method(call, callee);
1292 /* Do some statistics */
1293 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1294 env->n_call_nodes --;
1295 env->n_nodes += callee_env->n_nodes;
1296 callee_env->n_callers--;
1303 /* inline other small functions. */
1304 for (i = 0; i < n_irgs; ++i) {
1307 int phiproj_computed = 0;
1309 current_ir_graph = get_irp_irg(i);
1310 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1312 /* we can not walk and change a set, nor remove from it.
1314 walkset = env->call_nodes;
1315 env->call_nodes = eset_create();
1316 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1317 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1318 ir_graph *callee = get_call_called_irg(call);
1321 ((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1322 (get_irg_inline_property(callee) == irg_inline_forced))) {
1323 if (!phiproj_computed) {
1324 phiproj_computed = 1;
1325 collect_phiprojs(current_ir_graph);
1327 if (inline_method(call, callee)) {
1328 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1329 env->n_call_nodes--;
1330 eset_insert_all(env->call_nodes, callee_env->call_nodes); /* @@@ ??? This are the wrong nodes !? Not the copied ones. */
1331 env->n_call_nodes += callee_env->n_call_nodes;
1332 env->n_nodes += callee_env->n_nodes;
1333 callee_env->n_callers--;
1336 eset_insert(env->call_nodes, call);
1339 eset_destroy(walkset);
1342 for (i = 0; i < n_irgs; ++i) {
1343 current_ir_graph = get_irp_irg(i);
1345 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1346 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1347 (env->n_callers_orig != env->n_callers))
1348 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1349 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1350 env->n_callers_orig, env->n_callers,
1351 get_entity_name(get_irg_entity(current_ir_graph)));
1353 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1356 current_ir_graph = rem;
1359 /*******************************************************************/
1360 /* Code Placement. Pins all floating nodes to a block where they */
1361 /* will be executed only if needed. */
1362 /*******************************************************************/
1365 * Find the earliest correct block for N. --- Place N into the
1366 * same Block as its dominance-deepest Input.
1369 place_floats_early(ir_node *n, pdeq *worklist)
1371 int i, start, irn_arity;
1373 /* we must not run into an infinite loop */
1374 assert (irn_not_visited(n));
1375 mark_irn_visited(n);
1377 /* Place floating nodes. */
1378 if (get_irn_pinned(n) == op_pin_state_floats) {
1380 ir_node *b = new_Bad(); /* The block to place this node in */
1381 int bad_recursion = is_Bad(get_nodes_block(n));
1383 assert(get_irn_op(n) != op_Block);
1385 if ((get_irn_op(n) == op_Const) ||
1386 (get_irn_op(n) == op_SymConst) ||
1388 (get_irn_op(n) == op_Unknown)) {
1389 /* These nodes will not be placed by the loop below. */
1390 b = get_irg_start_block(current_ir_graph);
1394 /* find the block for this node. */
1395 irn_arity = get_irn_arity(n);
1396 for (i = 0; i < irn_arity; i++) {
1397 ir_node *dep = get_irn_n(n, i);
1400 if ((irn_not_visited(dep))
1401 && (get_irn_pinned(dep) == op_pin_state_floats)) {
1402 place_floats_early(dep, worklist);
1406 * A node in the Bad block must stay in the bad block,
1407 * so don't compute a new block for it.
1412 /* Because all loops contain at least one op_pin_state_pinned node, now all
1413 our inputs are either op_pin_state_pinned or place_early has already
1414 been finished on them. We do not have any unfinished inputs! */
1415 dep_block = get_nodes_block(dep);
1416 if ((!is_Bad(dep_block)) &&
1417 (get_Block_dom_depth(dep_block) > depth)) {
1419 depth = get_Block_dom_depth(dep_block);
1421 /* Avoid that the node is placed in the Start block */
1422 if ((depth == 1) && (get_Block_dom_depth(get_nodes_block(n)) > 1)) {
1423 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1424 assert(b != get_irg_start_block(current_ir_graph));
1428 set_nodes_block(n, b);
1431 /* Add predecessors of non floating nodes on worklist. */
1432 start = (get_irn_op(n) == op_Block) ? 0 : -1;
1433 irn_arity = get_irn_arity(n);
1434 for (i = start; i < irn_arity; i++) {
1435 ir_node *pred = get_irn_n(n, i);
1436 if (irn_not_visited(pred)) {
1437 pdeq_putr (worklist, pred);
1443 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1444 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1445 * places all floating nodes reachable from its argument through floating
1446 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1448 static INLINE void place_early(pdeq *worklist) {
1450 inc_irg_visited(current_ir_graph);
1452 /* this inits the worklist */
1453 place_floats_early(get_irg_end(current_ir_graph), worklist);
1455 /* Work the content of the worklist. */
1456 while (!pdeq_empty (worklist)) {
1457 ir_node *n = pdeq_getl (worklist);
1458 if (irn_not_visited(n)) place_floats_early(n, worklist);
1461 set_irg_outs_inconsistent(current_ir_graph);
1462 current_ir_graph->op_pin_state_pinned = op_pin_state_pinned;
1466 /** Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
1467 * I.e., DCA is the block where we might place PRODUCER.
1468 * A data flow edge points from producer to consumer.
1471 consumer_dom_dca (ir_node *dca, ir_node *consumer, ir_node *producer)
1473 ir_node *block = NULL;
1475 /* Compute the latest block into which we can place a node so that it is
1477 if (get_irn_op(consumer) == op_Phi) {
1478 /* our consumer is a Phi-node, the effective use is in all those
1479 blocks through which the Phi-node reaches producer */
1481 ir_node *phi_block = get_nodes_block(consumer);
1482 irn_arity = get_irn_arity(consumer);
1484 for (i = 0; i < irn_arity; i++) {
1485 if (get_irn_n(consumer, i) == producer) {
1486 block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1490 assert(is_no_Block(consumer));
1491 block = get_nodes_block(consumer);
1494 /* Compute the deepest common ancestor of block and dca. */
1496 if (!dca) return block;
1497 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1498 block = get_Block_idom(block);
1499 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
1500 dca = get_Block_idom(dca);
1502 while (block != dca)
1503 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1508 static INLINE int get_irn_loop_depth(ir_node *n) {
1509 return get_loop_depth(get_irn_loop(n));
1513 * Move n to a block with less loop depth than it's current block. The
1514 * new block must be dominated by early.
1517 move_out_of_loops (ir_node *n, ir_node *early)
1519 ir_node *best, *dca;
1523 /* Find the region deepest in the dominator tree dominating
1524 dca with the least loop nesting depth, but still dominated
1525 by our early placement. */
1526 dca = get_nodes_block(n);
1528 while (dca != early) {
1529 dca = get_Block_idom(dca);
1530 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
1531 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1535 if (best != get_nodes_block(n)) {
1537 printf("Moving out of loop: "); DDMN(n);
1538 printf(" Outermost block: "); DDMN(early);
1539 printf(" Best block: "); DDMN(best);
1540 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1542 set_nodes_block(n, best);
1547 * Find the latest legal block for N and place N into the
1548 * `optimal' Block between the latest and earliest legal block.
1549 * The `optimal' block is the dominance-deepest block of those
1550 * with the least loop-nesting-depth. This places N out of as many
1551 * loops as possible and then makes it as control dependant as
1555 place_floats_late(ir_node *n, pdeq *worklist)
1560 assert (irn_not_visited(n)); /* no multiple placement */
1562 mark_irn_visited(n);
1564 /* no need to place block nodes, control nodes are already placed. */
1565 if ((get_irn_op(n) != op_Block) &&
1567 (get_irn_mode(n) != mode_X)) {
1568 /* Remember the early placement of this block to move it
1569 out of loop no further than the early placement. */
1570 early = get_nodes_block(n);
1572 /* Do not move code not reachable from Start. For
1573 * these we could not compute dominator information. */
1574 if (is_Bad(early) || get_Block_dom_depth(early) == -1)
1577 /* Assure that our users are all placed, except the Phi-nodes.
1578 --- Each data flow cycle contains at least one Phi-node. We
1579 have to break the `user has to be placed before the
1580 producer' dependence cycle and the Phi-nodes are the
1581 place to do so, because we need to base our placement on the
1582 final region of our users, which is OK with Phi-nodes, as they
1583 are op_pin_state_pinned, and they never have to be placed after a
1584 producer of one of their inputs in the same block anyway. */
1585 for (i = 0; i < get_irn_n_outs(n); i++) {
1586 ir_node *succ = get_irn_out(n, i);
1587 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
1588 place_floats_late(succ, worklist);
1591 /* We have to determine the final block of this node... except for
1593 if ((get_irn_pinned(n) == op_pin_state_floats) &&
1594 (get_irn_op(n) != op_Const) &&
1595 (get_irn_op(n) != op_SymConst)) {
1596 ir_node *dca = NULL; /* deepest common ancestor in the
1597 dominator tree of all nodes'
1598 blocks depending on us; our final
1599 placement has to dominate DCA. */
1600 for (i = 0; i < get_irn_n_outs(n); i++) {
1601 ir_node *out = get_irn_out(n, i);
1602 /* ignore if out is in dead code */
1603 ir_node *outbl = get_nodes_block(out);
1604 if (is_Bad(outbl) || get_Block_dom_depth(outbl) == -1)
1606 dca = consumer_dom_dca (dca, out, n);
1609 set_nodes_block(n, dca);
1611 move_out_of_loops (n, early);
1613 /* else all outs are in dead code */
1617 /* Add predecessors of all non-floating nodes on list. (Those of floating
1618 nodes are placeded already and therefore are marked.) */
1619 for (i = 0; i < get_irn_n_outs(n); i++) {
1620 if (irn_not_visited(get_irn_out(n, i))) {
1621 pdeq_putr (worklist, get_irn_out(n, i));
1626 static INLINE void place_late(pdeq *worklist) {
1628 inc_irg_visited(current_ir_graph);
1630 /* This fills the worklist initially. */
1631 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1633 /* And now empty the worklist again... */
1634 while (!pdeq_empty (worklist)) {
1635 ir_node *n = pdeq_getl (worklist);
1636 if (irn_not_visited(n)) place_floats_late(n, worklist);
1640 void place_code(ir_graph *irg) {
1642 ir_graph *rem = current_ir_graph;
1644 current_ir_graph = irg;
1646 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1648 /* Handle graph state */
1649 assert(get_irg_phase_state(irg) != phase_building);
1650 if (get_irg_dom_state(irg) != dom_consistent)
1653 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1654 free_loop_information(irg);
1655 construct_backedges(irg);
1658 /* Place all floating nodes as early as possible. This guarantees
1659 a legal code placement. */
1660 worklist = new_pdeq();
1661 place_early(worklist);
1663 /* place_early invalidates the outs, place_late needs them. */
1665 /* Now move the nodes down in the dominator tree. This reduces the
1666 unnecessary executions of the node. */
1667 place_late(worklist);
1669 set_irg_outs_inconsistent(current_ir_graph);
1670 set_irg_loopinfo_inconsistent(current_ir_graph);
1672 current_ir_graph = rem;
1676 * Called by walker of remove_critical_cf_edges().
1678 * Place an empty block to an edge between a blocks of multiple
1679 * predecessors and a block of multiple successors.
1682 * @param env Environment of walker. This field is unused and has
1685 static void walk_critical_cf_edges(ir_node *n, void *env) {
1687 ir_node *pre, *block, **in, *jmp;
1689 /* Block has multiple predecessors */
1690 if ((op_Block == get_irn_op(n)) &&
1691 (get_irn_arity(n) > 1)) {
1692 arity = get_irn_arity(n);
1694 if (n == get_irg_end_block(current_ir_graph))
1695 return; /* No use to add a block here. */
1697 for (i=0; i<arity; i++) {
1698 pre = get_irn_n(n, i);
1699 /* Predecessor has multiple successors. Insert new flow edge */
1700 if ((NULL != pre) &&
1701 (op_Proj == get_irn_op(pre)) &&
1702 op_Raise != get_irn_op(skip_Proj(pre))) {
1704 /* set predecessor array for new block */
1705 in = NEW_ARR_D (ir_node *, current_ir_graph->obst, 1);
1706 /* set predecessor of new block */
1708 block = new_Block(1, in);
1709 /* insert new jmp node to new block */
1710 set_cur_block(block);
1713 /* set successor of new block */
1714 set_irn_n(n, i, jmp);
1716 } /* predecessor has multiple successors */
1717 } /* for all predecessors */
1718 } /* n is a block */
1721 void remove_critical_cf_edges(ir_graph *irg) {
1722 if (get_opt_critical_edges())
1723 irg_walk_graph(irg, NULL, walk_critical_cf_edges, NULL);