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.
209 copy_node (ir_node *n, void *env) {
212 opcode op = get_irn_opcode(n);
213 /* The end node looses it's flexible in array. This doesn't matter,
214 as dead node elimination builds End by hand, inlineing doesn't use
216 /* assert(n->op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
219 /* node copied already */
221 } else if (op == iro_Block) {
223 new_arity = compute_new_arity(n);
224 n->attr.block.graph_arr = NULL;
226 block = get_nodes_block(n);
227 if (get_irn_opcode(n) == iro_Phi) {
228 new_arity = compute_new_arity(block);
230 new_arity = get_irn_arity(n);
233 nn = new_ir_node(get_irn_dbg_info(n),
240 /* Copy the attributes. These might point to additional data. If this
241 was allocated on the old obstack the pointers now are dangling. This
242 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
244 new_backedge_info(nn);
247 /* printf("\n old node: "); DDMSG2(n);
248 printf(" new node: "); DDMSG2(nn); */
253 * Copies new predecessors of old node to new node remembered in link.
254 * Spare the Bad predecessors of Phi and Block nodes.
257 copy_preds (ir_node *n, void *env) {
261 nn = get_new_node(n);
263 /* printf("\n old node: "); DDMSG2(n);
264 printf(" new node: "); DDMSG2(nn);
265 printf(" arities: old: %d, new: %d\n", get_irn_arity(n), get_irn_arity(nn)); */
267 if (get_irn_opcode(n) == iro_Block) {
268 /* Don't copy Bad nodes. */
270 irn_arity = get_irn_arity(n);
271 for (i = 0; i < irn_arity; i++)
272 if (get_irn_opcode(get_irn_n(n, i)) != iro_Bad) {
273 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
274 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
277 /* repair the block visited flag from above misuse. Repair it in both
278 graphs so that the old one can still be used. */
279 set_Block_block_visited(nn, 0);
280 set_Block_block_visited(n, 0);
281 /* Local optimization could not merge two subsequent blocks if
282 in array contained Bads. Now it's possible.
283 We don't call optimize_in_place as it requires
284 that the fields in ir_graph are set properly. */
285 if ((get_opt_control_flow_straightening()) &&
286 (get_Block_n_cfgpreds(nn) == 1) &&
287 (get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp)) {
288 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
290 /* Jmp jumps into the block it is in -- deal self cycle. */
291 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
292 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
297 } else if (get_irn_opcode(n) == iro_Phi) {
298 /* Don't copy node if corresponding predecessor in block is Bad.
299 The Block itself should not be Bad. */
300 block = get_nodes_block(n);
301 set_irn_n (nn, -1, get_new_node(block));
303 irn_arity = get_irn_arity(n);
304 for (i = 0; i < irn_arity; i++)
305 if (get_irn_opcode(get_irn_n(block, i)) != iro_Bad) {
306 set_irn_n (nn, j, get_new_node(get_irn_n(n, i)));
307 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
310 /* If the pre walker reached this Phi after the post walker visited the
311 block block_visited is > 0. */
312 set_Block_block_visited(get_nodes_block(n), 0);
313 /* Compacting the Phi's ins might generate Phis with only one
315 if (get_irn_arity(n) == 1)
316 exchange(n, get_irn_n(n, 0));
318 irn_arity = get_irn_arity(n);
319 for (i = -1; i < irn_arity; i++)
320 set_irn_n (nn, i, get_new_node(get_irn_n(n, i)));
322 /* Now the new node is complete. We can add it to the hash table for cse.
323 @@@ inlinening aborts if we identify End. Why? */
324 if(get_irn_op(nn) != op_End)
325 add_identities (current_ir_graph->value_table, nn);
329 * Copies the graph recursively, compacts the keepalive of the end node.
333 ir_node *oe, *ne, *ob, *nb; /* old end, new end, old bad, new bad */
334 ir_node *ka; /* keep alive */
337 oe = get_irg_end(current_ir_graph);
338 /* copy the end node by hand, allocate dynamic in array! */
339 ne = new_ir_node(get_irn_dbg_info(oe),
346 /* Copy the attributes. Well, there might be some in the future... */
348 set_new_node(oe, ne);
350 ob = get_irg_bad(current_ir_graph);
351 nb = new_ir_node(get_irn_dbg_info(ob),
358 set_new_node(ob, nb);
360 /* copy the live nodes */
361 irg_walk(get_nodes_block(oe), copy_node, copy_preds, NULL);
362 /* copy_preds for the end node ... */
363 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
364 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
366 /*- ... and now the keep alives. -*/
367 /* First pick the not marked block nodes and walk them. We must pick these
368 first as else we will oversee blocks reachable from Phis. */
369 irn_arity = get_irn_arity(oe);
370 for (i = 0; i < irn_arity; i++) {
371 ka = get_irn_intra_n(oe, i);
372 if ((get_irn_op(ka) == op_Block) &&
373 (get_irn_visited(ka) < get_irg_visited(current_ir_graph))) {
374 /* We must keep the block alive and copy everything reachable */
375 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
376 irg_walk(ka, copy_node, copy_preds, NULL);
377 add_End_keepalive(ne, get_new_node(ka));
381 /* Now pick the Phis. Here we will keep all! */
382 irn_arity = get_irn_arity(oe);
383 for (i = 0; i < irn_arity; i++) {
384 ka = get_irn_intra_n(oe, i);
385 if ((get_irn_op(ka) == op_Phi)) {
386 if (get_irn_visited(ka) < get_irg_visited(current_ir_graph)) {
387 /* We didn't copy the Phi yet. */
388 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
389 irg_walk(ka, copy_node, copy_preds, NULL);
391 add_End_keepalive(ne, get_new_node(ka));
397 * Copies the graph reachable from current_ir_graph->end to the obstack
398 * in current_ir_graph and fixes the environment.
399 * Then fixes the fields in current_ir_graph containing nodes of the
403 copy_graph_env (void) {
405 /* Not all nodes remembered in current_ir_graph might be reachable
406 from the end node. Assure their link is set to NULL, so that
407 we can test whether new nodes have been computed. */
408 set_irn_link(get_irg_frame (current_ir_graph), NULL);
409 set_irn_link(get_irg_globals (current_ir_graph), NULL);
410 set_irn_link(get_irg_args (current_ir_graph), NULL);
411 set_irn_link(get_irg_initial_mem(current_ir_graph), NULL);
413 /* we use the block walk flag for removing Bads from Blocks ins. */
414 inc_irg_block_visited(current_ir_graph);
419 /* fix the fields in current_ir_graph */
420 old_end = get_irg_end(current_ir_graph);
421 set_irg_end (current_ir_graph, get_new_node(old_end));
422 set_irg_end_except (current_ir_graph, get_irg_end(current_ir_graph));
423 set_irg_end_reg (current_ir_graph, get_irg_end(current_ir_graph));
425 set_irg_end_block (current_ir_graph, get_new_node(get_irg_end_block(current_ir_graph)));
426 if (get_irn_link(get_irg_frame(current_ir_graph)) == NULL) {
427 copy_node (get_irg_frame(current_ir_graph), NULL);
428 copy_preds(get_irg_frame(current_ir_graph), NULL);
430 if (get_irn_link(get_irg_globals(current_ir_graph)) == NULL) {
431 copy_node (get_irg_globals(current_ir_graph), NULL);
432 copy_preds(get_irg_globals(current_ir_graph), NULL);
434 if (get_irn_link(get_irg_initial_mem(current_ir_graph)) == NULL) {
435 copy_node (get_irg_initial_mem(current_ir_graph), NULL);
436 copy_preds(get_irg_initial_mem(current_ir_graph), NULL);
438 if (get_irn_link(get_irg_args(current_ir_graph)) == NULL) {
439 copy_node (get_irg_args(current_ir_graph), NULL);
440 copy_preds(get_irg_args(current_ir_graph), NULL);
442 set_irg_start (current_ir_graph, get_new_node(get_irg_start(current_ir_graph)));
444 set_irg_start_block(current_ir_graph,
445 get_new_node(get_irg_start_block(current_ir_graph)));
446 set_irg_frame (current_ir_graph, get_new_node(get_irg_frame(current_ir_graph)));
447 set_irg_globals (current_ir_graph, get_new_node(get_irg_globals(current_ir_graph)));
448 set_irg_initial_mem(current_ir_graph, get_new_node(get_irg_initial_mem(current_ir_graph)));
449 set_irg_args (current_ir_graph, get_new_node(get_irg_args(current_ir_graph)));
451 if (get_irn_link(get_irg_bad(current_ir_graph)) == NULL) {
452 copy_node(get_irg_bad(current_ir_graph), NULL);
453 copy_preds(get_irg_bad(current_ir_graph), NULL);
455 set_irg_bad(current_ir_graph, get_new_node(get_irg_bad(current_ir_graph)));
459 * Copies all reachable nodes to a new obstack. Removes bad inputs
460 * from block nodes and the corresponding inputs from Phi nodes.
461 * Merges single exit blocks with single entry blocks and removes
463 * Adds all new nodes to a new hash table for cse. Does not
464 * perform cse, so the hash table might contain common subexpressions.
467 dead_node_elimination(ir_graph *irg) {
469 int rem_ipview = interprocedural_view;
470 struct obstack *graveyard_obst = NULL;
471 struct obstack *rebirth_obst = NULL;
473 /* inform statistics that we started a dead-node elimination run */
474 stat_dead_node_elim_start(irg);
476 /* Remember external state of current_ir_graph. */
477 rem = current_ir_graph;
478 current_ir_graph = irg;
479 interprocedural_view = 0;
481 /* Handle graph state */
482 assert(get_irg_phase_state(current_ir_graph) != phase_building);
483 free_callee_info(current_ir_graph);
484 free_outs(current_ir_graph);
485 /* @@@ so far we loose loops when copying */
486 free_loop_information(current_ir_graph);
488 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
490 /* A quiet place, where the old obstack can rest in peace,
491 until it will be cremated. */
492 graveyard_obst = irg->obst;
494 /* A new obstack, where the reachable nodes will be copied to. */
495 rebirth_obst = (struct obstack *) xmalloc (sizeof (struct obstack));
496 current_ir_graph->obst = rebirth_obst;
497 obstack_init (current_ir_graph->obst);
499 /* We also need a new hash table for cse */
500 del_identities (irg->value_table);
501 irg->value_table = new_identities ();
503 /* Copy the graph from the old to the new obstack */
506 /* Free memory from old unoptimized obstack */
507 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
508 xfree (graveyard_obst); /* ... then free it. */
511 /* inform statistics that the run is over */
512 stat_dead_node_elim_stop(irg);
514 current_ir_graph = rem;
515 interprocedural_view = rem_ipview;
519 * Relink bad predeseccors of a block and store the old in array to the
520 * link field. This function is called by relink_bad_predecessors().
521 * The array of link field starts with the block operand at position 0.
522 * If block has bad predecessors, create a new in array without bad preds.
523 * Otherwise let in array untouched.
525 static void relink_bad_block_predecessors(ir_node *n, void *env) {
526 ir_node **new_in, *irn;
527 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
529 /* if link field of block is NULL, look for bad predecessors otherwise
530 this is allready done */
531 if (get_irn_op(n) == op_Block &&
532 get_irn_link(n) == NULL) {
534 /* save old predecessors in link field (position 0 is the block operand)*/
535 set_irn_link(n, (void *)get_irn_in(n));
537 /* count predecessors without bad nodes */
538 old_irn_arity = get_irn_arity(n);
539 for (i = 0; i < old_irn_arity; i++)
540 if (!is_Bad(get_irn_n(n, i))) new_irn_arity++;
542 /* arity changing: set new predecessors without bad nodes */
543 if (new_irn_arity < old_irn_arity) {
544 /* get new predecessor array without Block predecessor */
545 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
547 /* set new predeseccors in array */
550 for (i = 1; i < old_irn_arity; i++) {
551 irn = get_irn_n(n, i);
552 if (!is_Bad(irn)) new_in[new_irn_n++] = irn;
555 } /* ir node has bad predecessors */
557 } /* Block is not relinked */
561 * Relinks Bad predecesors from Bocks and Phis called by walker
562 * remove_bad_predecesors(). If n is a Block, call
563 * relink_bad_block_redecessors(). If n is a Phinode, call also the relinking
564 * function of Phi's Block. If this block has bad predecessors, relink preds
567 static void relink_bad_predecessors(ir_node *n, void *env) {
568 ir_node *block, **old_in;
569 int i, old_irn_arity, new_irn_arity;
571 /* relink bad predeseccors of a block */
572 if (get_irn_op(n) == op_Block)
573 relink_bad_block_predecessors(n, env);
575 /* If Phi node relink its block and its predecessors */
576 if (get_irn_op(n) == op_Phi) {
578 /* Relink predeseccors of phi's block */
579 block = get_nodes_block(n);
580 if (get_irn_link(block) == NULL)
581 relink_bad_block_predecessors(block, env);
583 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
584 old_irn_arity = ARR_LEN(old_in);
586 /* Relink Phi predeseccors if count of predeseccors changed */
587 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
588 /* set new predeseccors in array
589 n->in[0] remains the same block */
591 for(i = 1; i < old_irn_arity; i++)
592 if (!is_Bad((ir_node *)old_in[i])) n->in[new_irn_arity++] = n->in[i];
594 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
597 } /* n is a Phi node */
601 * Removes Bad Bad predecesors from Blocks and the corresponding
602 * inputs to Phi nodes as in dead_node_elimination but without
604 * On walking up set the link field to NULL, on walking down call
605 * relink_bad_predecessors() (This function stores the old in array
606 * to the link field and sets a new in array if arity of predecessors
609 void remove_bad_predecessors(ir_graph *irg) {
610 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
614 /*--------------------------------------------------------------------*/
615 /* Funcionality for inlining */
616 /*--------------------------------------------------------------------*/
619 * Copy node for inlineing. Updates attributes that change when
620 * inlineing but not for dead node elimination.
622 * Copies the node by calling copy_node and then updates the entity if
623 * it's a local one. env must be a pointer of the frame type of the
624 * inlined procedure. The new entities must be in the link field of
628 copy_node_inline (ir_node *n, void *env) {
630 type *frame_tp = (type *)env;
633 if (get_irn_op(n) == op_Sel) {
634 new = get_new_node (n);
635 assert(get_irn_op(new) == op_Sel);
636 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
637 set_Sel_entity(new, get_entity_link(get_Sel_entity(n)));
639 } else if (get_irn_op(n) == op_Block) {
640 new = get_new_node (n);
641 new->attr.block.irg = current_ir_graph;
645 static void find_addr(ir_node *node, void *env)
647 if (get_irn_opcode(node) == iro_Proj) {
648 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
654 * currently, we cannot inline two cases:
655 * - call with compound arguments
656 * - graphs that take the address of a parameter
658 * check these condition here
660 static int can_inline(ir_node *call, ir_graph *called_graph)
662 type *call_type = get_Call_type(call);
663 int params, ress, i, res;
664 assert(is_method_type(call_type));
666 params = get_method_n_params(call_type);
667 ress = get_method_n_ress(call_type);
670 for (i = 0; i < params; ++i) {
671 type *p_type = get_method_param_type(call_type, i);
673 if (is_compound_type(p_type))
678 for (i = 0; i < ress; ++i) {
679 type *r_type = get_method_res_type(call_type, i);
681 if (is_compound_type(r_type))
686 irg_walk_graph(called_graph, find_addr, NULL, &res);
691 int inline_method(ir_node *call, ir_graph *called_graph) {
693 ir_node *post_call, *post_bl;
695 ir_node *end, *end_bl;
699 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
702 irg_inline_property prop = get_irg_inline_property(called_graph);
704 if ( (prop != irg_inline_forced) &&
705 (!get_opt_optimize() || !get_opt_inline() || (prop == irg_inline_forbidden))) return 0;
709 * currently, we cannot inline two cases:
710 * - call with compound arguments
711 * - graphs that take the address of a parameter
713 if (! can_inline(call, called_graph))
716 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
717 rem_opt = get_opt_optimize();
720 /* Handle graph state */
721 assert(get_irg_phase_state(current_ir_graph) != phase_building);
722 assert(get_irg_pinned(current_ir_graph) == op_pin_state_pinned);
723 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
724 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
725 set_irg_outs_inconsistent(current_ir_graph);
726 set_irg_loopinfo_inconsistent(current_ir_graph);
728 /* -- Check preconditions -- */
729 assert(get_irn_op(call) == op_Call);
730 /* @@@ does not work for InterfaceIII.java after cgana
731 assert(get_Call_type(call) == get_entity_type(get_irg_entity(called_graph)));
732 assert(smaller_type(get_entity_type(get_irg_entity(called_graph)),
733 get_Call_type(call)));
735 assert(get_type_tpop(get_Call_type(call)) == type_method);
736 if (called_graph == current_ir_graph) {
737 set_optimize(rem_opt);
741 /* here we know we WILL inline, so inform the statistics */
742 stat_inline(call, called_graph);
744 /* -- Decide how to handle exception control flow: Is there a handler
745 for the Call node, or do we branch directly to End on an exception?
746 exc_handling: 0 There is a handler.
748 2 Exception handling not represented in Firm. -- */
750 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
751 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
752 assert(get_irn_op(proj) == op_Proj);
753 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
754 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
756 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
757 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
758 else { exc_handling = 2; } /* !Mproj && !Xproj */
763 the procedure and later replaces the Start node of the called graph.
764 Post_call is the old Call node and collects the results of the called
765 graph. Both will end up being a tuple. -- */
766 post_bl = get_nodes_block(call);
767 set_irg_current_block(current_ir_graph, post_bl);
768 /* XxMxPxP of Start + parameter of Call */
769 in[pn_Start_X_initial_exec] = new_Jmp();
770 in[pn_Start_M] = get_Call_mem(call);
771 in[pn_Start_P_frame_base] = get_irg_frame(current_ir_graph);
772 in[pn_Start_P_globals] = get_irg_globals(current_ir_graph);
773 in[pn_Start_T_args] = new_Tuple(get_Call_n_params(call), get_Call_param_arr(call));
774 /* in[pn_Start_P_value_arg_base] = ??? */
775 pre_call = new_Tuple(5, in);
779 The new block gets the ins of the old block, pre_call and all its
780 predecessors and all Phi nodes. -- */
781 part_block(pre_call);
783 /* -- Prepare state for dead node elimination -- */
784 /* Visited flags in calling irg must be >= flag in called irg.
785 Else walker and arity computation will not work. */
786 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
787 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
788 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
789 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
790 /* Set pre_call as new Start node in link field of the start node of
791 calling graph and pre_calls block as new block for the start block
793 Further mark these nodes so that they are not visited by the
795 set_irn_link(get_irg_start(called_graph), pre_call);
796 set_irn_visited(get_irg_start(called_graph), get_irg_visited(current_ir_graph));
797 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
798 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(current_ir_graph));
799 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
800 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(current_ir_graph));
802 /* Initialize for compaction of in arrays */
803 inc_irg_block_visited(current_ir_graph);
805 /* -- Replicate local entities of the called_graph -- */
806 /* copy the entities. */
807 called_frame = get_irg_frame_type(called_graph);
808 for (i = 0; i < get_class_n_members(called_frame); i++) {
809 entity *new_ent, *old_ent;
810 old_ent = get_class_member(called_frame, i);
811 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
812 set_entity_link(old_ent, new_ent);
815 /* visited is > than that of called graph. With this trick visited will
816 remain unchanged so that an outer walker, e.g., searching the call nodes
817 to inline, calling this inline will not visit the inlined nodes. */
818 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
820 /* -- Performing dead node elimination inlines the graph -- */
821 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
823 /* @@@ endless loops are not copied!! -- they should be, I think... */
824 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
825 get_irg_frame_type(called_graph));
827 /* Repair called_graph */
828 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
829 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
830 set_Block_block_visited(get_irg_start_block(called_graph), 0);
832 /* -- Merge the end of the inlined procedure with the call site -- */
833 /* We will turn the old Call node into a Tuple with the following
836 0: Phi of all Memories of Return statements.
837 1: Jmp from new Block that merges the control flow from all exception
838 predecessors of the old end block.
839 2: Tuple of all arguments.
840 3: Phi of Exception memories.
841 In case the old Call directly branches to End on an exception we don't
842 need the block merging all exceptions nor the Phi of the exception
846 /* -- Precompute some values -- */
847 end_bl = get_new_node(get_irg_end_block(called_graph));
848 end = get_new_node(get_irg_end(called_graph));
849 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
850 n_res = get_method_n_ress(get_Call_type(call));
852 res_pred = (ir_node **) malloc (n_res * sizeof (ir_node *));
853 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
855 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
857 /* -- archive keepalives -- */
858 irn_arity = get_irn_arity(end);
859 for (i = 0; i < irn_arity; i++)
860 add_End_keepalive(get_irg_end(current_ir_graph), get_irn_n(end, i));
862 /* The new end node will die. We need not free as the in array is on the obstack:
863 copy_node only generated 'D' arrays. */
865 /* -- Replace Return nodes by Jump nodes. -- */
867 for (i = 0; i < arity; i++) {
869 ret = get_irn_n(end_bl, i);
870 if (get_irn_op(ret) == op_Return) {
871 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_block(ret));
875 set_irn_in(post_bl, n_ret, cf_pred);
877 /* -- Build a Tuple for all results of the method.
878 Add Phi node if there was more than one Return. -- */
879 turn_into_tuple(post_call, 4);
880 /* First the Memory-Phi */
882 for (i = 0; i < arity; i++) {
883 ret = get_irn_n(end_bl, i);
884 if (get_irn_op(ret) == op_Return) {
885 cf_pred[n_ret] = get_Return_mem(ret);
889 phi = new_Phi(n_ret, cf_pred, mode_M);
890 set_Tuple_pred(call, pn_Call_M_regular, phi);
891 /* Conserve Phi-list for further inlinings -- but might be optimized */
892 if (get_nodes_block(phi) == post_bl) {
893 set_irn_link(phi, get_irn_link(post_bl));
894 set_irn_link(post_bl, phi);
896 /* Now the real results */
898 for (j = 0; j < n_res; j++) {
900 for (i = 0; i < arity; i++) {
901 ret = get_irn_n(end_bl, i);
902 if (get_irn_op(ret) == op_Return) {
903 cf_pred[n_ret] = get_Return_res(ret, j);
908 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
912 /* Conserve Phi-list for further inlinings -- but might be optimized */
913 if (get_nodes_block(phi) == post_bl) {
914 set_irn_link(phi, get_irn_link(post_bl));
915 set_irn_link(post_bl, phi);
918 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
920 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
922 /* Finally the exception control flow.
923 We have two (three) possible situations:
924 First if the Call branches to an exception handler: We need to add a Phi node to
925 collect the memory containing the exception objects. Further we need
926 to add another block to get a correct representation of this Phi. To
927 this block we add a Jmp that resolves into the X output of the Call
928 when the Call is turned into a tuple.
929 Second the Call branches to End, the exception is not handled. Just
930 add all inlined exception branches to the End node.
931 Third: there is no Exception edge at all. Handle as case two. */
932 if (exc_handling == 0) {
934 for (i = 0; i < arity; i++) {
936 ret = get_irn_n(end_bl, i);
937 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
938 cf_pred[n_exc] = ret;
943 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
944 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
945 /* The Phi for the memories with the exception objects */
947 for (i = 0; i < arity; i++) {
949 ret = skip_Proj(get_irn_n(end_bl, i));
950 if (get_irn_op(ret) == op_Call) {
951 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 3);
953 } else if (is_fragile_op(ret)) {
954 /* We rely that all cfops have the memory output at the same position. */
955 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 0);
957 } else if (get_irn_op(ret) == op_Raise) {
958 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_block(ret), ret, mode_M, 1);
962 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
964 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
965 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
968 ir_node *main_end_bl;
969 int main_end_bl_arity;
972 /* assert(exc_handling == 1 || no exceptions. ) */
974 for (i = 0; i < arity; i++) {
975 ir_node *ret = get_irn_n(end_bl, i);
977 if (is_fragile_op(skip_Proj(ret)) || (get_irn_op(skip_Proj(ret)) == op_Raise)) {
978 cf_pred[n_exc] = ret;
982 main_end_bl = get_irg_end_block(current_ir_graph);
983 main_end_bl_arity = get_irn_arity(main_end_bl);
984 end_preds = (ir_node **) malloc ((n_exc + main_end_bl_arity) * sizeof (ir_node *));
986 for (i = 0; i < main_end_bl_arity; ++i)
987 end_preds[i] = get_irn_n(main_end_bl, i);
988 for (i = 0; i < n_exc; ++i)
989 end_preds[main_end_bl_arity + i] = cf_pred[i];
990 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
991 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
992 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
998 #if 0 /* old. now better, correcter, faster implementation. */
1000 /* -- If the exception control flow from the inlined Call directly
1001 branched to the end block we now have the following control
1002 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
1003 remove the Jmp along with it's empty block and add Jmp's
1004 predecessors as predecessors of this end block. No problem if
1005 there is no exception, because then branches Bad to End which
1007 @@@ can't we know this beforehand: by getting the Proj(1) from
1008 the Call link list and checking whether it goes to Proj. */
1009 /* find the problematic predecessor of the end block. */
1010 end_bl = get_irg_end_block(current_ir_graph);
1011 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
1012 cf_op = get_Block_cfgpred(end_bl, i);
1013 if (get_irn_op(cf_op) == op_Proj) {
1014 cf_op = get_Proj_pred(cf_op);
1015 if ((get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
1016 /* There are unoptimized tuples from inlineing before when no exc */
1017 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
1018 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
1019 assert(get_irn_op(cf_op) == op_Jmp);
1025 if (i < get_Block_n_cfgpreds(end_bl)) {
1026 bl = get_nodes_block(cf_op);
1027 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
1028 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
1029 for (j = 0; j < i; j++)
1030 cf_pred[j] = get_Block_cfgpred(end_bl, j);
1031 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
1032 cf_pred[j] = get_Block_cfgpred(bl, j-i);
1033 for (j = j; j < arity; j++)
1034 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
1035 set_irn_in(end_bl, arity, cf_pred);
1037 /* Remove the exception pred from post-call Tuple. */
1038 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1043 /* -- Turn cse back on. -- */
1044 set_optimize(rem_opt);
1049 /********************************************************************/
1050 /* Apply inlineing to small methods. */
1051 /********************************************************************/
1053 /* It makes no sense to inline too many calls in one procedure. Anyways,
1054 I didn't get a version with NEW_ARR_F to run. */
1055 #define MAX_INLINE 1024
1058 * environment for inlining small irgs
1060 typedef struct _inline_env_t {
1062 ir_node *calls[MAX_INLINE];
1066 * Returns the irg called from a Call node. If the irg is not
1067 * known, NULL is returned.
1069 static ir_graph *get_call_called_irg(ir_node *call) {
1071 ir_graph *called_irg = NULL;
1073 assert(get_irn_op(call) == op_Call);
1075 addr = get_Call_ptr(call);
1076 if ((get_irn_op(addr) == op_SymConst) && (get_SymConst_kind (addr) == symconst_addr_ent)) {
1077 called_irg = get_entity_irg(get_SymConst_entity(addr));
1083 static void collect_calls(ir_node *call, void *env) {
1086 if (get_irn_op(call) != op_Call) return;
1088 addr = get_Call_ptr(call);
1090 if (get_irn_op(addr) == op_SymConst) {
1091 if (get_SymConst_kind(addr) == symconst_addr_ent) {
1092 ir_graph *called_irg = get_entity_irg(get_SymConst_entity(addr));
1093 inline_env_t *ienv = (inline_env_t *)env;
1094 if (called_irg && ienv->pos < MAX_INLINE) {
1095 /* The Call node calls a locally defined method. Remember to inline. */
1096 ienv->calls[ienv->pos++] = call;
1103 * Inlines all small methods at call sites where the called address comes
1104 * from a Const node that references the entity representing the called
1106 * The size argument is a rough measure for the code size of the method:
1107 * Methods where the obstack containing the firm graph is smaller than
1110 void inline_small_irgs(ir_graph *irg, int size) {
1112 ir_graph *rem = current_ir_graph;
1113 inline_env_t env /* = {0, NULL}*/;
1115 if (!(get_opt_optimize() && get_opt_inline())) return;
1117 current_ir_graph = irg;
1118 /* Handle graph state */
1119 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1120 free_callee_info(current_ir_graph);
1122 /* Find Call nodes to inline.
1123 (We can not inline during a walk of the graph, as inlineing the same
1124 method several times changes the visited flag of the walked graph:
1125 after the first inlineing visited of the callee equals visited of
1126 the caller. With the next inlineing both are increased.) */
1128 irg_walk(get_irg_end(irg), NULL, collect_calls, &env);
1130 if ((env.pos > 0) && (env.pos < MAX_INLINE)) {
1131 /* There are calls to inline */
1132 collect_phiprojs(irg);
1133 for (i = 0; i < env.pos; i++) {
1135 callee = get_entity_irg(get_SymConst_entity(get_Call_ptr(env.calls[i])));
1136 if (((_obstack_memory_used(callee->obst) - obstack_room(callee->obst)) < size) ||
1137 (get_irg_inline_property(callee) == irg_inline_forced)) {
1138 inline_method(env.calls[i], callee);
1143 current_ir_graph = rem;
1147 * Environment for inlining irgs.
1150 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1151 int n_nodes_orig; /**< for statistics */
1152 eset *call_nodes; /**< All call nodes in this graph */
1154 int n_call_nodes_orig; /**< for statistics */
1155 int n_callers; /**< Number of known graphs that call this graphs. */
1156 int n_callers_orig; /**< for statistics */
1159 static inline_irg_env *new_inline_irg_env(void) {
1160 inline_irg_env *env = malloc(sizeof(inline_irg_env));
1161 env->n_nodes = -2; /* uncount Start, End */
1162 env->n_nodes_orig = -2; /* uncount Start, End */
1163 env->call_nodes = eset_create();
1164 env->n_call_nodes = 0;
1165 env->n_call_nodes_orig = 0;
1167 env->n_callers_orig = 0;
1171 static void free_inline_irg_env(inline_irg_env *env) {
1172 eset_destroy(env->call_nodes);
1176 static void collect_calls2(ir_node *call, void *env) {
1177 inline_irg_env *x = (inline_irg_env *)env;
1178 ir_op *op = get_irn_op(call);
1181 /* count nodes in irg */
1182 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1187 if (op != op_Call) return;
1189 /* collect all call nodes */
1190 eset_insert(x->call_nodes, (void *)call);
1192 x->n_call_nodes_orig++;
1194 /* count all static callers */
1195 callee = get_call_called_irg(call);
1197 ((inline_irg_env *)get_irg_link(callee))->n_callers++;
1198 ((inline_irg_env *)get_irg_link(callee))->n_callers_orig++;
1202 INLINE static int is_leave(ir_graph *irg) {
1203 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1206 INLINE static int is_smaller(ir_graph *callee, int size) {
1207 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1212 * Inlines small leave methods at call sites where the called address comes
1213 * from a Const node that references the entity representing the called
1215 * The size argument is a rough measure for the code size of the method:
1216 * Methods where the obstack containing the firm graph is smaller than
1219 void inline_leave_functions(int maxsize, int leavesize, int size) {
1220 inline_irg_env *env;
1221 int i, n_irgs = get_irp_n_irgs();
1222 ir_graph *rem = current_ir_graph;
1225 if (!(get_opt_optimize() && get_opt_inline())) return;
1227 /* extend all irgs by a temporary data structure for inlineing. */
1228 for (i = 0; i < n_irgs; ++i)
1229 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1231 /* Precompute information in temporary data structure. */
1232 for (i = 0; i < n_irgs; ++i) {
1233 current_ir_graph = get_irp_irg(i);
1234 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1235 free_callee_info(current_ir_graph);
1237 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1238 get_irg_link(current_ir_graph));
1241 /* -- and now inline. -- */
1243 /* Inline leaves recursively -- we might construct new leaves. */
1244 while (did_inline) {
1247 for (i = 0; i < n_irgs; ++i) {
1249 int phiproj_computed = 0;
1251 current_ir_graph = get_irp_irg(i);
1252 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1254 for (call = eset_first(env->call_nodes); call; call = eset_next(env->call_nodes)) {
1255 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1256 ir_graph *callee = get_call_called_irg(call);
1258 if (env->n_nodes > maxsize) continue; // break;
1260 if (callee && (is_leave(callee) && is_smaller(callee, leavesize))) {
1261 if (!phiproj_computed) {
1262 phiproj_computed = 1;
1263 collect_phiprojs(current_ir_graph);
1265 did_inline = inline_method(call, callee);
1268 /* Do some statistics */
1269 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1270 env->n_call_nodes --;
1271 env->n_nodes += callee_env->n_nodes;
1272 callee_env->n_callers--;
1279 /* inline other small functions. */
1280 for (i = 0; i < n_irgs; ++i) {
1283 int phiproj_computed = 0;
1285 current_ir_graph = get_irp_irg(i);
1286 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1288 /* we can not walk and change a set, nor remove from it.
1290 walkset = env->call_nodes;
1291 env->call_nodes = eset_create();
1292 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1293 if (get_irn_op(call) == op_Tuple) continue; /* We already inlined. */
1294 ir_graph *callee = get_call_called_irg(call);
1297 ((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1298 (get_irg_inline_property(callee) == irg_inline_forced))) {
1299 if (!phiproj_computed) {
1300 phiproj_computed = 1;
1301 collect_phiprojs(current_ir_graph);
1303 if (inline_method(call, callee)) {
1304 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1305 env->n_call_nodes--;
1306 eset_insert_all(env->call_nodes, callee_env->call_nodes); /* @@@ ??? This are the wrong nodes !? Not the copied ones. */
1307 env->n_call_nodes += callee_env->n_call_nodes;
1308 env->n_nodes += callee_env->n_nodes;
1309 callee_env->n_callers--;
1312 eset_insert(env->call_nodes, call);
1315 eset_destroy(walkset);
1318 for (i = 0; i < n_irgs; ++i) {
1319 current_ir_graph = get_irp_irg(i);
1321 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1322 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1323 (env->n_callers_orig != env->n_callers))
1324 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1325 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1326 env->n_callers_orig, env->n_callers,
1327 get_entity_name(get_irg_entity(current_ir_graph)));
1329 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1332 current_ir_graph = rem;
1335 /*******************************************************************/
1336 /* Code Placement. Pins all floating nodes to a block where they */
1337 /* will be executed only if needed. */
1338 /*******************************************************************/
1341 * Find the earliest correct block for N. --- Place N into the
1342 * same Block as its dominance-deepest Input.
1345 place_floats_early(ir_node *n, pdeq *worklist)
1347 int i, start, irn_arity;
1349 /* we must not run into an infinite loop */
1350 assert (irn_not_visited(n));
1351 mark_irn_visited(n);
1353 /* Place floating nodes. */
1354 if (get_op_pinned(get_irn_op(n)) == op_pin_state_floats) {
1356 ir_node *b = new_Bad(); /* The block to place this node in */
1357 int bad_recursion = is_Bad(get_nodes_block(n));
1359 assert(get_irn_op(n) != op_Block);
1361 if ((get_irn_op(n) == op_Const) ||
1362 (get_irn_op(n) == op_SymConst) ||
1364 (get_irn_op(n) == op_Unknown)) {
1365 /* These nodes will not be placed by the loop below. */
1366 b = get_irg_start_block(current_ir_graph);
1370 /* find the block for this node. */
1371 irn_arity = get_irn_arity(n);
1372 for (i = 0; i < irn_arity; i++) {
1373 ir_node *dep = get_irn_n(n, i);
1376 if ((irn_not_visited(dep))
1377 && (get_op_pinned(get_irn_op(dep)) == op_pin_state_floats)) {
1378 place_floats_early(dep, worklist);
1382 * A node in the Bad block must stay in the bad block,
1383 * so don't compute a new block for it.
1388 /* Because all loops contain at least one op_pin_state_pinned node, now all
1389 our inputs are either op_pin_state_pinned or place_early has already
1390 been finished on them. We do not have any unfinished inputs! */
1391 dep_block = get_nodes_block(dep);
1392 if ((!is_Bad(dep_block)) &&
1393 (get_Block_dom_depth(dep_block) > depth)) {
1395 depth = get_Block_dom_depth(dep_block);
1397 /* Avoid that the node is placed in the Start block */
1398 if ((depth == 1) && (get_Block_dom_depth(get_nodes_block(n)) > 1)) {
1399 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1400 assert(b != get_irg_start_block(current_ir_graph));
1404 set_nodes_block(n, b);
1407 /* Add predecessors of non floating nodes on worklist. */
1408 start = (get_irn_op(n) == op_Block) ? 0 : -1;
1409 irn_arity = get_irn_arity(n);
1410 for (i = start; i < irn_arity; i++) {
1411 ir_node *pred = get_irn_n(n, i);
1412 if (irn_not_visited(pred)) {
1413 pdeq_putr (worklist, pred);
1419 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1420 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call. Place_early
1421 * places all floating nodes reachable from its argument through floating
1422 * nodes and adds all beginnings at op_pin_state_pinned nodes to the worklist.
1424 static INLINE void place_early(pdeq* worklist) {
1426 inc_irg_visited(current_ir_graph);
1428 /* this inits the worklist */
1429 place_floats_early(get_irg_end(current_ir_graph), worklist);
1431 /* Work the content of the worklist. */
1432 while (!pdeq_empty (worklist)) {
1433 ir_node *n = pdeq_getl (worklist);
1434 if (irn_not_visited(n)) place_floats_early(n, worklist);
1437 set_irg_outs_inconsistent(current_ir_graph);
1438 current_ir_graph->op_pin_state_pinned = op_pin_state_pinned;
1442 /** deepest common dominance ancestor of DCA and CONSUMER of PRODUCER. */
1444 consumer_dom_dca (ir_node *dca, ir_node *consumer, ir_node *producer)
1446 ir_node *block = NULL;
1448 /* Compute the latest block into which we can place a node so that it is
1450 if (get_irn_op(consumer) == op_Phi) {
1451 /* our consumer is a Phi-node, the effective use is in all those
1452 blocks through which the Phi-node reaches producer */
1454 ir_node *phi_block = get_nodes_block(consumer);
1455 irn_arity = get_irn_arity(consumer);
1456 for (i = 0; i < irn_arity; i++) {
1457 if (get_irn_n(consumer, i) == producer) {
1458 block = get_nodes_block(get_Block_cfgpred(phi_block, i));
1462 assert(is_no_Block(consumer));
1463 block = get_nodes_block(consumer);
1466 /* Compute the deepest common ancestor of block and dca. */
1468 if (!dca) return block;
1469 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1470 block = get_Block_idom(block);
1471 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block))
1472 dca = get_Block_idom(dca);
1473 while (block != dca)
1474 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1479 static INLINE int get_irn_loop_depth(ir_node *n) {
1480 return get_loop_depth(get_irn_loop(n));
1484 * Move n to a block with less loop depth than it's current block. The
1485 * new block must be dominated by early.
1488 move_out_of_loops (ir_node *n, ir_node *early)
1490 ir_node *best, *dca;
1494 /* Find the region deepest in the dominator tree dominating
1495 dca with the least loop nesting depth, but still dominated
1496 by our early placement. */
1497 dca = get_nodes_block(n);
1499 while (dca != early) {
1500 dca = get_Block_idom(dca);
1501 if (!dca) break; /* should we put assert(dca)? */
1502 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1506 if (best != get_nodes_block(n)) {
1508 printf("Moving out of loop: "); DDMN(n);
1509 printf(" Outermost block: "); DDMN(early);
1510 printf(" Best block: "); DDMN(best);
1511 printf(" Innermost block: "); DDMN(get_nodes_block(n));
1513 set_nodes_block(n, best);
1518 * Find the latest legal block for N and place N into the
1519 * `optimal' Block between the latest and earliest legal block.
1520 * The `optimal' block is the dominance-deepest block of those
1521 * with the least loop-nesting-depth. This places N out of as many
1522 * loops as possible and then makes it as control dependant as
1526 place_floats_late(ir_node *n, pdeq *worklist)
1531 assert (irn_not_visited(n)); /* no multiple placement */
1533 /* no need to place block nodes, control nodes are already placed. */
1534 if ((get_irn_op(n) != op_Block) &&
1536 (get_irn_mode(n) != mode_X)) {
1537 /* Remember the early placement of this block to move it
1538 out of loop no further than the early placement. */
1539 early = get_nodes_block(n);
1540 /* Assure that our users are all placed, except the Phi-nodes.
1541 --- Each data flow cycle contains at least one Phi-node. We
1542 have to break the `user has to be placed before the
1543 producer' dependence cycle and the Phi-nodes are the
1544 place to do so, because we need to base our placement on the
1545 final region of our users, which is OK with Phi-nodes, as they
1546 are op_pin_state_pinned, and they never have to be placed after a
1547 producer of one of their inputs in the same block anyway. */
1548 for (i = 0; i < get_irn_n_outs(n); i++) {
1549 ir_node *succ = get_irn_out(n, i);
1550 if (irn_not_visited(succ) && (get_irn_op(succ) != op_Phi))
1551 place_floats_late(succ, worklist);
1554 /* We have to determine the final block of this node... except for
1556 if ((get_op_pinned(get_irn_op(n)) == op_pin_state_floats) &&
1557 (get_irn_op(n) != op_Const) &&
1558 (get_irn_op(n) != op_SymConst)) {
1559 ir_node *dca = NULL; /* deepest common ancestor in the
1560 dominator tree of all nodes'
1561 blocks depending on us; our final
1562 placement has to dominate DCA. */
1563 for (i = 0; i < get_irn_n_outs(n); i++) {
1564 dca = consumer_dom_dca (dca, get_irn_out(n, i), n);
1566 set_nodes_block(n, dca);
1568 move_out_of_loops (n, early);
1572 mark_irn_visited(n);
1574 /* Add predecessors of all non-floating nodes on list. (Those of floating
1575 nodes are placeded already and therefore are marked.) */
1576 for (i = 0; i < get_irn_n_outs(n); i++) {
1577 if (irn_not_visited(get_irn_out(n, i))) {
1578 pdeq_putr (worklist, get_irn_out(n, i));
1583 static INLINE void place_late(pdeq* worklist) {
1585 inc_irg_visited(current_ir_graph);
1587 /* This fills the worklist initially. */
1588 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1589 /* And now empty the worklist again... */
1590 while (!pdeq_empty (worklist)) {
1591 ir_node *n = pdeq_getl (worklist);
1592 if (irn_not_visited(n)) place_floats_late(n, worklist);
1596 void place_code(ir_graph *irg) {
1598 ir_graph *rem = current_ir_graph;
1600 current_ir_graph = irg;
1602 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1604 /* Handle graph state */
1605 assert(get_irg_phase_state(irg) != phase_building);
1606 if (get_irg_dom_state(irg) != dom_consistent)
1609 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1610 free_loop_information(irg);
1611 construct_backedges(irg);
1614 /* Place all floating nodes as early as possible. This guarantees
1615 a legal code placement. */
1616 worklist = new_pdeq();
1617 place_early(worklist);
1619 /* place_early invalidates the outs, place_late needs them. */
1621 /* Now move the nodes down in the dominator tree. This reduces the
1622 unnecessary executions of the node. */
1623 place_late(worklist);
1625 set_irg_outs_inconsistent(current_ir_graph);
1626 set_irg_loopinfo_inconsistent(current_ir_graph);
1628 current_ir_graph = rem;
1633 /********************************************************************/
1634 /* Control flow optimization. */
1635 /* Removes Bad control flow predecessors and empty blocks. A block */
1636 /* is empty if it contains only a Jmp node. */
1637 /* Blocks can only be removed if they are not needed for the */
1638 /* semantics of Phi nodes. */
1639 /********************************************************************/
1642 * Removes Tuples from Block control flow predecessors.
1643 * Optimizes blocks with equivalent_node().
1644 * Replaces n by Bad if n is unreachable control flow.
1646 static void merge_blocks(ir_node *n, void *env) {
1648 set_irn_link(n, NULL);
1650 if (get_irn_op(n) == op_Block) {
1652 for (i = 0; i < get_Block_n_cfgpreds(n); i++)
1653 /* GL @@@ : is this possible? if (get_opt_normalize()) -- added, all tests go through.
1654 A different order of optimizations might cause problems. */
1655 if (get_opt_normalize())
1656 set_Block_cfgpred(n, i, skip_Tuple(get_Block_cfgpred(n, i)));
1657 } else if (get_opt_optimize() && (get_irn_mode(n) == mode_X)) {
1658 /* We will soon visit a block. Optimize it before visiting! */
1659 ir_node *b = get_nodes_block(n);
1660 ir_node *new_node = equivalent_node(b);
1661 while (irn_not_visited(b) && (!is_Bad(new_node)) && (new_node != b)) {
1662 /* We would have to run gigo if new is bad, so we
1663 promote it directly below. */
1664 assert(((b == new_node) ||
1665 get_opt_control_flow_straightening() ||
1666 get_opt_control_flow_weak_simplification()) &&
1667 ("strange flag setting"));
1668 exchange (b, new_node);
1670 new_node = equivalent_node(b);
1672 if (is_Bad(new_node) && get_opt_normalize()) exchange(n, new_Bad());
1677 * Collects all Phi nodes in link list of Block.
1678 * Marks all blocks "block_visited" if they contain a node other
1681 static void collect_nodes(ir_node *n, void *env) {
1682 if (is_no_Block(n)) {
1683 ir_node *b = get_nodes_block(n);
1685 if ((get_irn_op(n) == op_Phi)) {
1686 /* Collect Phi nodes to compact ins along with block's ins. */
1687 set_irn_link(n, get_irn_link(b));
1689 } else if ((get_irn_op(n) != op_Jmp) && !is_Bad(b)) { /* Check for non empty block. */
1690 mark_Block_block_visited(b);
1695 /** Returns true if pred is predecessor of block. */
1696 static int is_pred_of(ir_node *pred, ir_node *b) {
1698 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1699 ir_node *b_pred = get_nodes_block(get_Block_cfgpred(b, i));
1700 if (b_pred == pred) return 1;
1705 static int test_whether_dispensable(ir_node *b, int pos) {
1706 int i, j, n_preds = 1;
1707 int dispensable = 1;
1708 ir_node *cfop = get_Block_cfgpred(b, pos);
1709 ir_node *pred = get_nodes_block(cfop);
1711 if (get_Block_block_visited(pred) + 1
1712 < get_irg_block_visited(current_ir_graph)) {
1713 if (!get_opt_optimize() || !get_opt_control_flow_strong_simplification()) {
1714 /* Mark block so that is will not be removed. */
1715 set_Block_block_visited(pred, get_irg_block_visited(current_ir_graph)-1);
1718 /* Seems to be empty. */
1719 if (!get_irn_link(b)) {
1720 /* There are no Phi nodes ==> dispensable. */
1721 n_preds = get_Block_n_cfgpreds(pred);
1723 /* b's pred blocks and pred's pred blocks must be pairwise disjunct.
1724 Work preds < pos as if they were already removed. */
1725 for (i = 0; i < pos; i++) {
1726 ir_node *b_pred = get_nodes_block(get_Block_cfgpred(b, i));
1727 if (get_Block_block_visited(b_pred) + 1
1728 < get_irg_block_visited(current_ir_graph)) {
1729 for (j = 0; j < get_Block_n_cfgpreds(b_pred); j++) {
1730 ir_node *b_pred_pred = get_nodes_block(get_Block_cfgpred(b_pred, j));
1731 if (is_pred_of(b_pred_pred, pred)) dispensable = 0;
1734 if (is_pred_of(b_pred, pred)) dispensable = 0;
1737 for (i = pos +1; i < get_Block_n_cfgpreds(b); i++) {
1738 ir_node *b_pred = get_nodes_block(get_Block_cfgpred(b, i));
1739 if (is_pred_of(b_pred, pred)) dispensable = 0;
1742 set_Block_block_visited(pred, get_irg_block_visited(current_ir_graph)-1);
1745 n_preds = get_Block_n_cfgpreds(pred);
1753 static void optimize_blocks(ir_node *b, void *env) {
1754 int i, j, k, max_preds, n_preds;
1755 ir_node *pred, *phi;
1758 /* Count the number of predecessor if this block is merged with pred blocks
1761 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1762 max_preds += test_whether_dispensable(b, i);
1764 in = (ir_node **) malloc(max_preds * sizeof(ir_node *));
1767 printf(" working on "); DDMN(b);
1768 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1769 pred = get_nodes_block(get_Block_cfgpred(b, i));
1770 if (is_Bad(get_Block_cfgpred(b, i))) {
1771 printf(" removing Bad %i\n ", i);
1772 } else if (get_Block_block_visited(pred) +1
1773 < get_irg_block_visited(current_ir_graph)) {
1774 printf(" removing pred %i ", i); DDMN(pred);
1775 } else { printf(" Nothing to do for "); DDMN(pred); }
1777 * end Debug output -*/
1779 /*- Fix the Phi nodes -*/
1780 phi = get_irn_link(b);
1782 assert(get_irn_op(phi) == op_Phi);
1783 /* Find the new predecessors for the Phi */
1785 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1786 pred = get_nodes_block(get_Block_cfgpred(b, i));
1787 if (is_Bad(get_Block_cfgpred(b, i))) {
1789 } else if (get_Block_block_visited(pred) +1
1790 < get_irg_block_visited(current_ir_graph)) {
1791 /* It's an empty block and not yet visited. */
1792 ir_node *phi_pred = get_Phi_pred(phi, i);
1793 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1794 if (get_nodes_block(phi_pred) == pred) {
1795 assert(get_irn_op(phi_pred) == op_Phi); /* Block is empty!! */
1796 in[n_preds] = get_Phi_pred(phi_pred, j);
1798 in[n_preds] = phi_pred;
1802 /* The Phi_pred node is replaced now if it is a Phi.
1803 In Schleifen kann offenbar der entfernte Phi Knoten legal verwendet werden.
1804 Daher muss der Phiknoten durch den neuen ersetzt werden.
1805 Weiter muss der alte Phiknoten entfernt werden (durch ersetzen oder
1806 durch einen Bad) damit er aus den keep_alive verschwinden kann.
1807 Man sollte also, falls keine Schleife vorliegt, exchange mit new_Bad
1809 if (get_nodes_block(phi_pred) == pred) {
1810 /* remove the Phi as it might be kept alive. Further there
1811 might be other users. */
1812 exchange(phi_pred, phi); /* geht, ist aber doch semantisch falsch! Warum?? */
1815 in[n_preds] = get_Phi_pred(phi, i);
1820 set_irn_in(phi, n_preds, in);
1822 phi = get_irn_link(phi);
1826 This happens only if merge between loop backedge and single loop entry. -*/
1827 for (k = 0; k < get_Block_n_cfgpreds(b); k++) {
1828 pred = get_nodes_block(get_Block_cfgpred(b, k));
1829 if (get_Block_block_visited(pred)+1 < get_irg_block_visited(current_ir_graph)) {
1830 phi = get_irn_link(pred);
1832 if (get_irn_op(phi) == op_Phi) {
1833 set_nodes_block(phi, b);
1836 for (i = 0; i < k; i++) {
1837 pred = get_nodes_block(get_Block_cfgpred(b, i));
1838 if (is_Bad(get_Block_cfgpred(b, i))) {
1840 } else if (get_Block_block_visited(pred) +1
1841 < get_irg_block_visited(current_ir_graph)) {
1842 /* It's an empty block and not yet visited. */
1843 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1844 /* @@@ Hier brauche ich Schleifeninformation!!! Kontrollflusskante
1845 muss Rueckwaertskante sein! (An allen vier in[n_preds] = phi
1846 Anweisungen.) Trotzdem tuts bisher!! */
1855 for (i = 0; i < get_Phi_n_preds(phi); i++) {
1856 in[n_preds] = get_Phi_pred(phi, i);
1859 for (i = k+1; i < get_Block_n_cfgpreds(b); i++) {
1860 pred = get_nodes_block(get_Block_cfgpred(b, i));
1861 if (is_Bad(get_Block_cfgpred(b, i))) {
1863 } else if (get_Block_block_visited(pred) +1
1864 < get_irg_block_visited(current_ir_graph)) {
1865 /* It's an empty block and not yet visited. */
1866 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1875 set_irn_in(phi, n_preds, in);
1877 phi = get_irn_link(phi);
1882 /*- Fix the block -*/
1884 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1885 pred = get_nodes_block(get_Block_cfgpred(b, i));
1886 if (is_Bad(get_Block_cfgpred(b, i))) {
1888 } else if (get_Block_block_visited(pred) +1
1889 < get_irg_block_visited(current_ir_graph)) {
1890 /* It's an empty block and not yet visited. */
1891 assert(get_Block_n_cfgpreds(b) > 1);
1892 /* Else it should be optimized by equivalent_node. */
1893 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1894 in[n_preds] = get_Block_cfgpred(pred, j);
1897 /* Remove block as it might be kept alive. */
1898 exchange(pred, b/*new_Bad()*/);
1900 in[n_preds] = get_Block_cfgpred(b, i);
1904 set_irn_in(b, n_preds, in);
1908 void optimize_cf(ir_graph *irg) {
1911 ir_node *end = get_irg_end(irg);
1912 ir_graph *rem = current_ir_graph;
1913 current_ir_graph = irg;
1915 /* Handle graph state */
1916 assert(get_irg_phase_state(irg) != phase_building);
1917 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
1918 set_irg_outs_inconsistent(current_ir_graph);
1919 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
1920 set_irg_dom_inconsistent(current_ir_graph);
1922 /* Use block visited flag to mark non-empty blocks. */
1923 inc_irg_block_visited(irg);
1924 irg_walk(end, merge_blocks, collect_nodes, NULL);
1926 /* Optimize the standard code. */
1927 irg_block_walk(get_irg_end_block(irg), optimize_blocks, NULL, NULL);
1929 /* Walk all keep alives, optimize them if block, add to new in-array
1930 for end if useful. */
1931 in = NEW_ARR_F (ir_node *, 1);
1932 in[0] = get_nodes_block(end);
1933 inc_irg_visited(current_ir_graph);
1934 for(i = 0; i < get_End_n_keepalives(end); i++) {
1935 ir_node *ka = get_End_keepalive(end, i);
1936 if (irn_not_visited(ka)) {
1937 if ((get_irn_op(ka) == op_Block) && Block_not_block_visited(ka)) {
1938 set_irg_block_visited(current_ir_graph, /* Don't walk all the way to Start. */
1939 get_irg_block_visited(current_ir_graph)-1);
1940 irg_block_walk(ka, optimize_blocks, NULL, NULL);
1941 mark_irn_visited(ka);
1942 ARR_APP1 (ir_node *, in, ka);
1943 } else if (get_irn_op(ka) == op_Phi) {
1944 mark_irn_visited(ka);
1945 ARR_APP1 (ir_node *, in, ka);
1949 /* DEL_ARR_F(end->in); GL @@@ tut nicht ! */
1952 current_ir_graph = rem;
1957 * Called by walker of remove_critical_cf_edges().
1959 * Place an empty block to an edge between a blocks of multiple
1960 * predecessors and a block of multiple successors.
1963 * @param env Environment of walker. This field is unused and has
1966 static void walk_critical_cf_edges(ir_node *n, void *env) {
1968 ir_node *pre, *block, **in, *jmp;
1970 /* Block has multiple predecessors */
1971 if ((op_Block == get_irn_op(n)) &&
1972 (get_irn_arity(n) > 1)) {
1973 arity = get_irn_arity(n);
1975 if (n == get_irg_end_block(current_ir_graph))
1976 return; /* No use to add a block here. */
1978 for (i=0; i<arity; i++) {
1979 pre = get_irn_n(n, i);
1980 /* Predecessor has multiple successors. Insert new flow edge */
1981 if ((NULL != pre) &&
1982 (op_Proj == get_irn_op(pre)) &&
1983 op_Raise != get_irn_op(skip_Proj(pre))) {
1985 /* set predecessor array for new block */
1986 in = NEW_ARR_D (ir_node *, current_ir_graph->obst, 1);
1987 /* set predecessor of new block */
1989 block = new_Block(1, in);
1990 /* insert new jmp node to new block */
1991 set_cur_block(block);
1994 /* set successor of new block */
1995 set_irn_n(n, i, jmp);
1997 } /* predecessor has multiple successors */
1998 } /* for all predecessors */
1999 } /* n is a block */
2002 void remove_critical_cf_edges(ir_graph *irg) {
2003 if (get_opt_critical_edges())
2004 irg_walk_graph(irg, NULL, walk_critical_cf_edges, NULL);