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.
23 # include "irnode_t.h"
24 # include "irgraph_t.h"
32 # include "pdeq.h" /* Fuer code placement */
35 # include "irbackedge_t.h"
36 # include "irflag_t.h"
38 /* Defined in iropt.c */
39 pset *new_identities (void);
40 void del_identities (pset *value_table);
41 void add_identities (pset *value_table, ir_node *node);
43 /*------------------------------------------------------------------*/
44 /* apply optimizations of iropt to all nodes. */
45 /*------------------------------------------------------------------*/
47 static void init_link (ir_node *n, void *env) {
48 set_irn_link(n, NULL);
51 #if 0 /* Old version. Avoids Ids.
52 This is not necessary: we do a postwalk, and get_irn_n
53 removes ids anyways. So it's much cheaper to call the
54 optimization less often and use the exchange() algorithm. */
56 optimize_in_place_wrapper (ir_node *n, void *env) {
58 ir_node *optimized, *old;
60 irn_arity = intern_get_irn_arity(n);
61 for (i = 0; i < irn_arity; i++) {
62 /* get_irn_n skips Id nodes, so comparison old != optimized does not
63 show all optimizations. Therefore always set new predecessor. */
64 old = intern_get_irn_intra_n(n, i);
65 optimized = optimize_in_place_2(old);
66 set_irn_n(n, i, optimized);
69 if (intern_get_irn_op(n) == op_Block) {
70 optimized = optimize_in_place_2(n);
71 if (optimized != n) exchange (n, optimized);
76 optimize_in_place_wrapper (ir_node *n, void *env) {
77 ir_node *optimized = optimize_in_place_2(n);
78 if (optimized != n) exchange (n, optimized);
85 local_optimize_graph (ir_graph *irg) {
86 ir_graph *rem = current_ir_graph;
87 current_ir_graph = irg;
89 /* Handle graph state */
90 assert(get_irg_phase_state(irg) != phase_building);
91 if (get_opt_global_cse())
92 set_irg_pinned(current_ir_graph, floats);
93 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
94 set_irg_outs_inconsistent(current_ir_graph);
95 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
96 set_irg_dom_inconsistent(current_ir_graph);
98 /* Clean the value_table in irg for the cse. */
99 del_identities(irg->value_table);
100 irg->value_table = new_identities();
102 /* walk over the graph */
103 irg_walk(irg->end, init_link, optimize_in_place_wrapper, NULL);
105 current_ir_graph = rem;
108 /*------------------------------------------------------------------*/
109 /* Routines for dead node elimination / copying garbage collection */
110 /* of the obstack. */
111 /*------------------------------------------------------------------*/
114 * Remember the new node in the old node by using a field all nodes have.
117 set_new_node (ir_node *old, ir_node *new)
123 * Get this new node, before the old node is forgotton.
125 static INLINE ir_node *
126 get_new_node (ir_node * n)
132 * We use the block_visited flag to mark that we have computed the
133 * number of useful predecessors for this block.
134 * Further we encode the new arity in this flag in the old blocks.
135 * Remembering the arity is useful, as it saves a lot of pointer
136 * accesses. This function is called for all Phi and Block nodes
140 compute_new_arity(ir_node *b) {
141 int i, res, irn_arity;
144 irg_v = get_irg_block_visited(current_ir_graph);
145 block_v = get_Block_block_visited(b);
146 if (block_v >= irg_v) {
147 /* we computed the number of preds for this block and saved it in the
149 return block_v - irg_v;
151 /* compute the number of good predecessors */
152 res = irn_arity = intern_get_irn_arity(b);
153 for (i = 0; i < irn_arity; i++)
154 if (intern_get_irn_opcode(intern_get_irn_n(b, i)) == iro_Bad) res--;
155 /* save it in the flag. */
156 set_Block_block_visited(b, irg_v + res);
161 /* TODO: add an ir_op operation */
162 static INLINE void new_backedge_info(ir_node *n) {
163 switch(intern_get_irn_opcode(n)) {
165 n->attr.block.cg_backedge = NULL;
166 n->attr.block.backedge = new_backedge_arr(current_ir_graph->obst, intern_get_irn_arity(n));
169 n->attr.phi_backedge = new_backedge_arr(current_ir_graph->obst, intern_get_irn_arity(n));
172 n->attr.filter.backedge = new_backedge_arr(current_ir_graph->obst, intern_get_irn_arity(n));
179 * Copies the node to the new obstack. The Ins of the new node point to
180 * the predecessors on the old obstack. For block/phi nodes not all
181 * predecessors might be copied. n->link points to the new node.
182 * For Phi and Block nodes the function allocates in-arrays with an arity
183 * only for useful predecessors. The arity is determined by counting
184 * the non-bad predecessors of the block.
187 copy_node (ir_node *n, void *env) {
191 /* The end node looses it's flexible in array. This doesn't matter,
192 as dead node elimination builds End by hand, inlineing doesn't use
194 /* assert(n->op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
196 if (intern_get_irn_opcode(n) == iro_Block) {
198 new_arity = compute_new_arity(n);
199 n->attr.block.graph_arr = NULL;
201 block = get_nodes_Block(n);
202 if (intern_get_irn_opcode(n) == iro_Phi) {
203 new_arity = compute_new_arity(block);
205 new_arity = intern_get_irn_arity(n);
208 nn = new_ir_node(get_irn_dbg_info(n),
211 intern_get_irn_op(n),
212 intern_get_irn_mode(n),
215 /* Copy the attributes. These might point to additional data. If this
216 was allocated on the old obstack the pointers now are dangling. This
217 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
219 new_backedge_info(nn);
222 /* printf("\n old node: "); DDMSG2(n);
223 printf(" new node: "); DDMSG2(nn); */
228 * Copies new predecessors of old node to new node remembered in link.
229 * Spare the Bad predecessors of Phi and Block nodes.
232 copy_preds (ir_node *n, void *env) {
236 nn = get_new_node(n);
238 /* printf("\n old node: "); DDMSG2(n);
239 printf(" new node: "); DDMSG2(nn);
240 printf(" arities: old: %d, new: %d\n", intern_get_irn_arity(n), intern_get_irn_arity(nn)); */
242 if (intern_get_irn_opcode(n) == iro_Block) {
243 /* Don't copy Bad nodes. */
245 irn_arity = intern_get_irn_arity(n);
246 for (i = 0; i < irn_arity; i++)
247 if (intern_get_irn_opcode(intern_get_irn_n(n, i)) != iro_Bad) {
248 set_irn_n (nn, j, get_new_node(intern_get_irn_n(n, i)));
249 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
252 /* repair the block visited flag from above misuse. Repair it in both
253 graphs so that the old one can still be used. */
254 set_Block_block_visited(nn, 0);
255 set_Block_block_visited(n, 0);
256 /* Local optimization could not merge two subsequent blocks if
257 in array contained Bads. Now it's possible.
258 We don't call optimize_in_place as it requires
259 that the fields in ir_graph are set properly. */
260 if ((get_opt_control_flow_straightening()) &&
261 (get_Block_n_cfgpreds(nn) == 1) &&
262 (intern_get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp))
263 exchange(nn, get_nodes_Block(get_Block_cfgpred(nn, 0)));
264 } else if (intern_get_irn_opcode(n) == iro_Phi) {
265 /* Don't copy node if corresponding predecessor in block is Bad.
266 The Block itself should not be Bad. */
267 block = get_nodes_Block(n);
268 set_irn_n (nn, -1, get_new_node(block));
270 irn_arity = intern_get_irn_arity(n);
271 for (i = 0; i < irn_arity; i++)
272 if (intern_get_irn_opcode(intern_get_irn_n(block, i)) != iro_Bad) {
273 set_irn_n (nn, j, get_new_node(intern_get_irn_n(n, i)));
274 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
277 /* If the pre walker reached this Phi after the post walker visited the
278 block block_visited is > 0. */
279 set_Block_block_visited(get_nodes_Block(n), 0);
280 /* Compacting the Phi's ins might generate Phis with only one
282 if (intern_get_irn_arity(n) == 1)
283 exchange(n, intern_get_irn_n(n, 0));
285 irn_arity = intern_get_irn_arity(n);
286 for (i = -1; i < irn_arity; i++)
287 set_irn_n (nn, i, get_new_node(intern_get_irn_n(n, i)));
289 /* Now the new node is complete. We can add it to the hash table for cse.
290 @@@ inlinening aborts if we identify End. Why? */
291 if(intern_get_irn_op(nn) != op_End)
292 add_identities (current_ir_graph->value_table, nn);
296 * Copies the graph recursively, compacts the keepalive of the end node.
300 ir_node *oe, *ne; /* old end, new end */
301 ir_node *ka; /* keep alive */
304 oe = get_irg_end(current_ir_graph);
305 /* copy the end node by hand, allocate dynamic in array! */
306 ne = new_ir_node(get_irn_dbg_info(oe),
313 /* Copy the attributes. Well, there might be some in the future... */
315 set_new_node(oe, ne);
317 /* copy the live nodes */
318 irg_walk(get_nodes_Block(oe), copy_node, copy_preds, NULL);
319 /* copy_preds for the end node ... */
320 set_nodes_Block(ne, get_new_node(get_nodes_Block(oe)));
322 /*- ... and now the keep alives. -*/
323 /* First pick the not marked block nodes and walk them. We must pick these
324 first as else we will oversee blocks reachable from Phis. */
325 irn_arity = intern_get_irn_arity(oe);
326 for (i = 0; i < irn_arity; i++) {
327 ka = intern_get_irn_intra_n(oe, i);
328 if ((intern_get_irn_op(ka) == op_Block) &&
329 (get_irn_visited(ka) < get_irg_visited(current_ir_graph))) {
330 /* We must keep the block alive and copy everything reachable */
331 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
332 irg_walk(ka, copy_node, copy_preds, NULL);
333 add_End_keepalive(ne, get_new_node(ka));
337 /* Now pick the Phis. Here we will keep all! */
338 irn_arity = intern_get_irn_arity(oe);
339 for (i = 0; i < irn_arity; i++) {
340 ka = intern_get_irn_intra_n(oe, i);
341 if ((intern_get_irn_op(ka) == op_Phi)) {
342 if (get_irn_visited(ka) < get_irg_visited(current_ir_graph)) {
343 /* We didn't copy the Phi yet. */
344 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
345 irg_walk(ka, copy_node, copy_preds, NULL);
347 add_End_keepalive(ne, get_new_node(ka));
353 * Copies the graph reachable from current_ir_graph->end to the obstack
354 * in current_ir_graph and fixes the environment.
355 * Then fixes the fields in current_ir_graph containing nodes of the
359 copy_graph_env (void) {
361 /* Not all nodes remembered in current_ir_graph might be reachable
362 from the end node. Assure their link is set to NULL, so that
363 we can test whether new nodes have been computed. */
364 set_irn_link(get_irg_frame (current_ir_graph), NULL);
365 set_irn_link(get_irg_globals(current_ir_graph), NULL);
366 set_irn_link(get_irg_args (current_ir_graph), NULL);
368 /* we use the block walk flag for removing Bads from Blocks ins. */
369 inc_irg_block_visited(current_ir_graph);
374 /* fix the fields in current_ir_graph */
375 old_end = get_irg_end(current_ir_graph);
376 set_irg_end (current_ir_graph, get_new_node(old_end));
377 set_irg_end_except (current_ir_graph, get_irg_end(current_ir_graph));
378 set_irg_end_reg (current_ir_graph, get_irg_end(current_ir_graph));
380 set_irg_end_block (current_ir_graph, get_new_node(get_irg_end_block(current_ir_graph)));
381 if (get_irn_link(get_irg_frame(current_ir_graph)) == NULL) {
382 copy_node (get_irg_frame(current_ir_graph), NULL);
383 copy_preds(get_irg_frame(current_ir_graph), NULL);
385 if (get_irn_link(get_irg_globals(current_ir_graph)) == NULL) {
386 copy_node (get_irg_globals(current_ir_graph), NULL);
387 copy_preds(get_irg_globals(current_ir_graph), NULL);
389 if (get_irn_link(get_irg_args(current_ir_graph)) == NULL) {
390 copy_node (get_irg_args(current_ir_graph), NULL);
391 copy_preds(get_irg_args(current_ir_graph), NULL);
393 set_irg_start (current_ir_graph, get_new_node(get_irg_start(current_ir_graph)));
395 set_irg_start_block(current_ir_graph,
396 get_new_node(get_irg_start_block(current_ir_graph)));
397 set_irg_frame (current_ir_graph, get_new_node(get_irg_frame(current_ir_graph)));
398 set_irg_globals(current_ir_graph, get_new_node(get_irg_globals(current_ir_graph)));
399 set_irg_args (current_ir_graph, get_new_node(get_irg_args(current_ir_graph)));
400 if (get_irn_link(get_irg_bad(current_ir_graph)) == NULL) {
401 copy_node(get_irg_bad(current_ir_graph), NULL);
402 copy_preds(get_irg_bad(current_ir_graph), NULL);
404 set_irg_bad(current_ir_graph, get_new_node(get_irg_bad(current_ir_graph)));
405 /* GL removed: we need unknown with mode for analyses.
406 if (get_irn_link(get_irg_unknown(current_ir_graph)) == NULL) {
407 copy_node(get_irg_unknown(current_ir_graph), NULL);
408 copy_preds(get_irg_unknown(current_ir_graph), NULL);
410 set_irg_unknown(current_ir_graph, get_new_node(get_irg_unknown(current_ir_graph)));
415 * Copies all reachable nodes to a new obstack. Removes bad inputs
416 * from block nodes and the corresponding inputs from Phi nodes.
417 * Merges single exit blocks with single entry blocks and removes
419 * Adds all new nodes to a new hash table for cse. Does not
420 * perform cse, so the hash table might contain common subexpressions.
422 /* Amroq call this emigrate() */
424 dead_node_elimination(ir_graph *irg) {
426 int rem_ipview = interprocedural_view;
427 struct obstack *graveyard_obst = NULL;
428 struct obstack *rebirth_obst = NULL;
430 /* Remember external state of current_ir_graph. */
431 rem = current_ir_graph;
432 current_ir_graph = irg;
433 interprocedural_view = 0;
435 /* Handle graph state */
436 assert(get_irg_phase_state(current_ir_graph) != phase_building);
437 assert(get_irg_callee_info_state(current_ir_graph) == irg_callee_info_none);
438 free_outs(current_ir_graph);
440 /* @@@ so far we loose loops when copying */
441 free_loop_information(current_ir_graph);
443 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
445 /* A quiet place, where the old obstack can rest in peace,
446 until it will be cremated. */
447 graveyard_obst = irg->obst;
449 /* A new obstack, where the reachable nodes will be copied to. */
450 rebirth_obst = (struct obstack *) xmalloc (sizeof (struct obstack));
451 current_ir_graph->obst = rebirth_obst;
452 obstack_init (current_ir_graph->obst);
454 /* We also need a new hash table for cse */
455 del_identities (irg->value_table);
456 irg->value_table = new_identities ();
458 /* Copy the graph from the old to the new obstack */
461 /* Free memory from old unoptimized obstack */
462 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
463 xfree (graveyard_obst); /* ... then free it. */
466 current_ir_graph = rem;
467 interprocedural_view = rem_ipview;
471 * Relink bad predeseccors of a block and store the old in array to the
472 * link field. This function is called by relink_bad_predecessors().
473 * The array of link field starts with the block operand at position 0.
474 * If block has bad predecessors, create a new in array without bad preds.
475 * Otherwise let in array untouched.
477 static void relink_bad_block_predecessors(ir_node *n, void *env) {
478 ir_node **new_in, *irn;
479 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
481 /* if link field of block is NULL, look for bad predecessors otherwise
482 this is allready done */
483 if (intern_get_irn_op(n) == op_Block &&
484 get_irn_link(n) == NULL) {
486 /* save old predecessors in link field (position 0 is the block operand)*/
487 set_irn_link(n, (void *)get_irn_in(n));
489 /* count predecessors without bad nodes */
490 old_irn_arity = intern_get_irn_arity(n);
491 for (i = 0; i < old_irn_arity; i++)
492 if (!is_Bad(intern_get_irn_n(n, i))) new_irn_arity++;
494 /* arity changing: set new predecessors without bad nodes */
495 if (new_irn_arity < old_irn_arity) {
496 /* get new predecessor array without Block predecessor */
497 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
499 /* set new predeseccors in array */
502 for (i = 1; i < old_irn_arity; i++) {
503 irn = intern_get_irn_n(n, i);
504 if (!is_Bad(irn)) new_in[new_irn_n++] = irn;
507 } /* ir node has bad predecessors */
509 } /* Block is not relinked */
513 * Relinks Bad predecesors from Bocks and Phis called by walker
514 * remove_bad_predecesors(). If n is a Block, call
515 * relink_bad_block_redecessors(). If n is a Phinode, call also the relinking
516 * function of Phi's Block. If this block has bad predecessors, relink preds
519 static void relink_bad_predecessors(ir_node *n, void *env) {
520 ir_node *block, **old_in;
521 int i, old_irn_arity, new_irn_arity;
523 /* relink bad predeseccors of a block */
524 if (intern_get_irn_op(n) == op_Block)
525 relink_bad_block_predecessors(n, env);
527 /* If Phi node relink its block and its predecessors */
528 if (intern_get_irn_op(n) == op_Phi) {
530 /* Relink predeseccors of phi's block */
531 block = get_nodes_Block(n);
532 if (get_irn_link(block) == NULL)
533 relink_bad_block_predecessors(block, env);
535 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
536 old_irn_arity = ARR_LEN(old_in);
538 /* Relink Phi predeseccors if count of predeseccors changed */
539 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
540 /* set new predeseccors in array
541 n->in[0] remains the same block */
543 for(i = 1; i < old_irn_arity; i++)
544 if (!is_Bad((ir_node *)old_in[i])) n->in[new_irn_arity++] = n->in[i];
546 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
549 } /* n is a Phi node */
553 * Removes Bad Bad predecesors from Blocks and the corresponding
554 * inputs to Phi nodes as in dead_node_elimination but without
556 * On walking up set the link field to NULL, on walking down call
557 * relink_bad_predecessors() (This function stores the old in array
558 * to the link field and sets a new in array if arity of predecessors
561 void remove_bad_predecessors(ir_graph *irg) {
562 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
566 /*--------------------------------------------------------------------*/
567 /* Funcionality for inlining */
568 /*--------------------------------------------------------------------*/
571 * Copy node for inlineing. Updates attributes that change when
572 * inlineing but not for dead node elimination.
574 * Copies the node by calling copy_node and then updates the entity if
575 * it's a local one. env must be a pointer of the frame type of the
576 * inlined procedure. The new entities must be in the link field of
580 copy_node_inline (ir_node *n, void *env) {
582 type *frame_tp = (type *)env;
585 if (intern_get_irn_op(n) == op_Sel) {
586 new = get_new_node (n);
587 assert(intern_get_irn_op(new) == op_Sel);
588 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
589 set_Sel_entity(new, get_entity_link(get_Sel_entity(n)));
591 } else if (intern_get_irn_op(n) == op_Block) {
592 new = get_new_node (n);
593 new->attr.block.irg = current_ir_graph;
598 void inline_method(ir_node *call, ir_graph *called_graph) {
600 ir_node *post_call, *post_bl;
602 ir_node *end, *end_bl;
606 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
610 if ( !(get_irg_inline_property(called_graph) == irg_inline_forced) && (!get_opt_optimize() || !get_opt_inline() ||
611 (get_irg_inline_property(called_graph) == irg_inline_forbidden))) return;
613 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
614 rem_opt = get_opt_optimize();
617 /* Handle graph state */
618 assert(get_irg_phase_state(current_ir_graph) != phase_building);
619 assert(get_irg_pinned(current_ir_graph) == pinned);
620 assert(get_irg_pinned(called_graph) == pinned);
621 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
622 set_irg_outs_inconsistent(current_ir_graph);
624 /* -- Check preconditions -- */
625 assert(get_irn_op(call) == op_Call);
626 /* @@@ does not work for InterfaceIII.java after cgana
627 assert(get_Call_type(call) == get_entity_type(get_irg_ent(called_graph)));
628 assert(smaller_type(get_entity_type(get_irg_ent(called_graph)),
629 get_Call_type(call)));
631 assert(get_type_tpop(get_Call_type(call)) == type_method);
632 if (called_graph == current_ir_graph) {
633 set_optimize(rem_opt);
637 /* -- Decide how to handle exception control flow: Is there a handler
638 for the Call node, or do we branch directly to End on an exception?
639 exc_handling: 0 There is a handler.
641 2 Exception handling not represented in Firm. -- */
643 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
644 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
645 assert(get_irn_op(proj) == op_Proj);
646 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
647 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
649 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
650 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
651 else { exc_handling = 2; } /* !Mproj && !Xproj */
656 the procedure and later replaces the Start node of the called graph.
657 Post_call is the old Call node and collects the results of the called
658 graph. Both will end up being a tuple. -- */
659 post_bl = get_nodes_Block(call);
660 set_irg_current_block(current_ir_graph, post_bl);
661 /* XxMxPxP of Start + parameter of Call */
663 in[1] = get_Call_mem(call);
664 in[2] = get_irg_frame(current_ir_graph);
665 in[3] = get_irg_globals(current_ir_graph);
666 in[4] = new_Tuple (get_Call_n_params(call), get_Call_param_arr(call));
667 pre_call = new_Tuple(5, in);
671 The new block gets the ins of the old block, pre_call and all its
672 predecessors and all Phi nodes. -- */
673 part_block(pre_call);
675 /* -- Prepare state for dead node elimination -- */
676 /* Visited flags in calling irg must be >= flag in called irg.
677 Else walker and arity computation will not work. */
678 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
679 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
680 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
681 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
682 /* Set pre_call as new Start node in link field of the start node of
683 calling graph and pre_calls block as new block for the start block
685 Further mark these nodes so that they are not visited by the
687 set_irn_link(get_irg_start(called_graph), pre_call);
688 set_irn_visited(get_irg_start(called_graph),
689 get_irg_visited(current_ir_graph));
690 set_irn_link(get_irg_start_block(called_graph),
691 get_nodes_Block(pre_call));
692 set_irn_visited(get_irg_start_block(called_graph),
693 get_irg_visited(current_ir_graph));
695 /* Initialize for compaction of in arrays */
696 inc_irg_block_visited(current_ir_graph);
698 /* -- Replicate local entities of the called_graph -- */
699 /* copy the entities. */
700 called_frame = get_irg_frame_type(called_graph);
701 for (i = 0; i < get_class_n_members(called_frame); i++) {
702 entity *new_ent, *old_ent;
703 old_ent = get_class_member(called_frame, i);
704 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
705 set_entity_link(old_ent, new_ent);
708 /* visited is > than that of called graph. With this trick visited will
709 remain unchanged so that an outer walker, e.g., searching the call nodes
710 to inline, calling this inline will not visit the inlined nodes. */
711 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
713 /* -- Performing dead node elimination inlines the graph -- */
714 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
716 /* @@@ endless loops are not copied!! -- they should be, I think... */
717 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
718 get_irg_frame_type(called_graph));
720 /* Repair called_graph */
721 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
722 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
723 set_Block_block_visited(get_irg_start_block(called_graph), 0);
725 /* -- Merge the end of the inlined procedure with the call site -- */
726 /* We will turn the old Call node into a Tuple with the following
729 0: Phi of all Memories of Return statements.
730 1: Jmp from new Block that merges the control flow from all exception
731 predecessors of the old end block.
732 2: Tuple of all arguments.
733 3: Phi of Exception memories.
734 In case the old Call directly branches to End on an exception we don't
735 need the block merging all exceptions nor the Phi of the exception
739 /* -- Precompute some values -- */
740 end_bl = get_new_node(get_irg_end_block(called_graph));
741 end = get_new_node(get_irg_end(called_graph));
742 arity = intern_get_irn_arity(end_bl); /* arity = n_exc + n_ret */
743 n_res = get_method_n_ress(get_Call_type(call));
745 res_pred = (ir_node **) malloc (n_res * sizeof (ir_node *));
746 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
748 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
750 /* -- archive keepalives -- */
751 irn_arity = intern_get_irn_arity(end);
752 for (i = 0; i < irn_arity; i++)
753 add_End_keepalive(get_irg_end(current_ir_graph), intern_get_irn_n(end, i));
755 /* The new end node will die. We need not free as the in array is on the obstack:
756 copy_node only generated 'D' arrays. */
758 /* -- Replace Return nodes by Jump nodes. -- */
760 for (i = 0; i < arity; i++) {
762 ret = intern_get_irn_n(end_bl, i);
763 if (intern_get_irn_op(ret) == op_Return) {
764 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_Block(ret));
768 set_irn_in(post_bl, n_ret, cf_pred);
770 /* -- Build a Tuple for all results of the method.
771 Add Phi node if there was more than one Return. -- */
772 turn_into_tuple(post_call, 4);
773 /* First the Memory-Phi */
775 for (i = 0; i < arity; i++) {
776 ret = intern_get_irn_n(end_bl, i);
777 if (intern_get_irn_op(ret) == op_Return) {
778 cf_pred[n_ret] = get_Return_mem(ret);
782 phi = new_Phi(n_ret, cf_pred, mode_M);
783 set_Tuple_pred(call, 0, phi);
784 /* Conserve Phi-list for further inlinings -- but might be optimized */
785 if (get_nodes_Block(phi) == post_bl) {
786 set_irn_link(phi, get_irn_link(post_bl));
787 set_irn_link(post_bl, phi);
789 /* Now the real results */
791 for (j = 0; j < n_res; j++) {
793 for (i = 0; i < arity; i++) {
794 ret = intern_get_irn_n(end_bl, i);
795 if (intern_get_irn_op(ret) == op_Return) {
796 cf_pred[n_ret] = get_Return_res(ret, j);
800 phi = new_Phi(n_ret, cf_pred, intern_get_irn_mode(cf_pred[0]));
802 /* Conserve Phi-list for further inlinings -- but might be optimized */
803 if (get_nodes_Block(phi) == post_bl) {
804 set_irn_link(phi, get_irn_link(post_bl));
805 set_irn_link(post_bl, phi);
808 set_Tuple_pred(call, 2, new_Tuple(n_res, res_pred));
810 set_Tuple_pred(call, 2, new_Bad());
812 /* Finally the exception control flow.
813 We have two (three) possible situations:
814 First if the Call branches to an exception handler: We need to add a Phi node to
815 collect the memory containing the exception objects. Further we need
816 to add another block to get a correct representation of this Phi. To
817 this block we add a Jmp that resolves into the X output of the Call
818 when the Call is turned into a tuple.
819 Second the Call branches to End, the exception is not handled. Just
820 add all inlined exception branches to the End node.
821 Third: there is no Exception edge at all. Handle as case two. */
822 if (exc_handling == 0) {
824 for (i = 0; i < arity; i++) {
826 ret = intern_get_irn_n(end_bl, i);
827 if (is_fragile_op(skip_Proj(ret)) || (intern_get_irn_op(skip_Proj(ret)) == op_Raise)) {
828 cf_pred[n_exc] = ret;
833 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
834 set_Tuple_pred(call, 1, new_Jmp());
835 /* The Phi for the memories with the exception objects */
837 for (i = 0; i < arity; i++) {
839 ret = skip_Proj(intern_get_irn_n(end_bl, i));
840 if (intern_get_irn_op(ret) == op_Call) {
841 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_Block(ret), ret, mode_M, 3);
843 } else if (is_fragile_op(ret)) {
844 /* We rely that all cfops have the memory output at the same position. */
845 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_Block(ret), ret, mode_M, 0);
847 } else if (intern_get_irn_op(ret) == op_Raise) {
848 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_Block(ret), ret, mode_M, 1);
852 set_Tuple_pred(call, 3, new_Phi(n_exc, cf_pred, mode_M));
854 set_Tuple_pred(call, 1, new_Bad());
855 set_Tuple_pred(call, 3, new_Bad());
858 ir_node *main_end_bl;
859 int main_end_bl_arity;
862 /* assert(exc_handling == 1 || no exceptions. ) */
864 for (i = 0; i < arity; i++) {
865 ir_node *ret = intern_get_irn_n(end_bl, i);
867 if (is_fragile_op(skip_Proj(ret)) || (intern_get_irn_op(skip_Proj(ret)) == op_Raise)) {
868 cf_pred[n_exc] = ret;
872 main_end_bl = get_irg_end_block(current_ir_graph);
873 main_end_bl_arity = intern_get_irn_arity(main_end_bl);
874 end_preds = (ir_node **) malloc ((n_exc + main_end_bl_arity) * sizeof (ir_node *));
876 for (i = 0; i < main_end_bl_arity; ++i)
877 end_preds[i] = intern_get_irn_n(main_end_bl, i);
878 for (i = 0; i < n_exc; ++i)
879 end_preds[main_end_bl_arity + i] = cf_pred[i];
880 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
881 set_Tuple_pred(call, 1, new_Bad());
882 set_Tuple_pred(call, 3, new_Bad());
888 #if 0 /* old. now better, correcter, faster implementation. */
890 /* -- If the exception control flow from the inlined Call directly
891 branched to the end block we now have the following control
892 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
893 remove the Jmp along with it's empty block and add Jmp's
894 predecessors as predecessors of this end block. No problem if
895 there is no exception, because then branches Bad to End which
897 @@@ can't we know this beforehand: by getting the Proj(1) from
898 the Call link list and checking whether it goes to Proj. */
899 /* find the problematic predecessor of the end block. */
900 end_bl = get_irg_end_block(current_ir_graph);
901 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
902 cf_op = get_Block_cfgpred(end_bl, i);
903 if (intern_get_irn_op(cf_op) == op_Proj) {
904 cf_op = get_Proj_pred(cf_op);
905 if ((intern_get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
906 /* There are unoptimized tuples from inlineing before when no exc */
907 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
908 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
909 assert(intern_get_irn_op(cf_op) == op_Jmp);
915 if (i < get_Block_n_cfgpreds(end_bl)) {
916 bl = get_nodes_Block(cf_op);
917 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
918 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
919 for (j = 0; j < i; j++)
920 cf_pred[j] = get_Block_cfgpred(end_bl, j);
921 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
922 cf_pred[j] = get_Block_cfgpred(bl, j-i);
923 for (j = j; j < arity; j++)
924 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
925 set_irn_in(end_bl, arity, cf_pred);
927 /* Remove the exception pred from post-call Tuple. */
928 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
933 /* -- Turn cse back on. -- */
934 set_optimize(rem_opt);
937 /********************************************************************/
938 /* Apply inlineing to small methods. */
939 /********************************************************************/
943 /* It makes no sense to inline too many calls in one procedure. Anyways,
944 I didn't get a version with NEW_ARR_F to run. */
945 #define MAX_INLINE 1024
948 * Returns the irg called from a Call node. If the irg is not
949 * known, NULL is returned.
951 static ir_graph *get_call_called_irg(ir_node *call) {
954 ir_graph *called_irg = NULL;
956 assert(get_irn_op(call) == op_Call);
958 addr = get_Call_ptr(call);
959 if (intern_get_irn_op(addr) == op_Const) {
960 /* Check whether the constant is the pointer to a compiled entity. */
961 tv = get_Const_tarval(addr);
962 if (tarval_to_entity(tv))
963 called_irg = get_entity_irg(tarval_to_entity(tv));
968 static void collect_calls(ir_node *call, void *env) {
970 ir_node **calls = (ir_node **)env;
973 ir_graph *called_irg;
975 if (intern_get_irn_op(call) != op_Call) return;
977 addr = get_Call_ptr(call);
978 if (intern_get_irn_op(addr) == op_Const) {
979 /* Check whether the constant is the pointer to a compiled entity. */
980 tv = get_Const_tarval(addr);
981 if (tarval_to_entity(tv)) {
982 called_irg = get_entity_irg(tarval_to_entity(tv));
983 if (called_irg && pos < MAX_INLINE) {
984 /* The Call node calls a locally defined method. Remember to inline. */
993 * Inlines all small methods at call sites where the called address comes
994 * from a Const node that references the entity representing the called
996 * The size argument is a rough measure for the code size of the method:
997 * Methods where the obstack containing the firm graph is smaller than
1000 void inline_small_irgs(ir_graph *irg, int size) {
1002 ir_node *calls[MAX_INLINE];
1003 ir_graph *rem = current_ir_graph;
1005 if (!(get_opt_optimize() && get_opt_inline())) return;
1007 current_ir_graph = irg;
1008 /* Handle graph state */
1009 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1010 assert(get_irg_callee_info_state(current_ir_graph) == irg_callee_info_none);
1012 /* Find Call nodes to inline.
1013 (We can not inline during a walk of the graph, as inlineing the same
1014 method several times changes the visited flag of the walked graph:
1015 after the first inlineing visited of the callee equals visited of
1016 the caller. With the next inlineing both are increased.) */
1018 irg_walk(get_irg_end(irg), NULL, collect_calls, (void *) calls);
1020 if ((pos > 0) && (pos < MAX_INLINE)) {
1021 /* There are calls to inline */
1022 collect_phiprojs(irg);
1023 for (i = 0; i < pos; i++) {
1026 tv = get_Const_tarval(get_Call_ptr(calls[i]));
1027 callee = get_entity_irg(tarval_to_entity(tv));
1028 if (((_obstack_memory_used(callee->obst) - obstack_room(callee->obst)) < size) ||
1029 (get_irg_inline_property(callee) == irg_inline_forced)) {
1030 inline_method(calls[i], callee);
1035 current_ir_graph = rem;
1039 * Environment for inlining irgs.
1042 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1043 int n_nodes_orig; /**< for statistics */
1044 eset *call_nodes; /**< All call nodes in this graph */
1046 int n_call_nodes_orig; /**< for statistics */
1047 int n_callers; /**< Number of known graphs that call this graphs. */
1048 int n_callers_orig; /**< for statistics */
1051 static inline_irg_env *new_inline_irg_env(void) {
1052 inline_irg_env *env = malloc(sizeof(inline_irg_env));
1053 env->n_nodes = -2; /* uncount Start, End */
1054 env->n_nodes_orig = -2; /* uncount Start, End */
1055 env->call_nodes = eset_create();
1056 env->n_call_nodes = 0;
1057 env->n_call_nodes_orig = 0;
1059 env->n_callers_orig = 0;
1063 static void free_inline_irg_env(inline_irg_env *env) {
1064 eset_destroy(env->call_nodes);
1068 static void collect_calls2(ir_node *call, void *env) {
1069 inline_irg_env *x = (inline_irg_env *)env;
1070 ir_op *op = intern_get_irn_op(call);
1073 /* count nodes in irg */
1074 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1079 if (op != op_Call) return;
1081 /* collect all call nodes */
1082 eset_insert(x->call_nodes, (void *)call);
1084 x->n_call_nodes_orig++;
1086 /* count all static callers */
1087 callee = get_call_called_irg(call);
1089 ((inline_irg_env *)get_irg_link(callee))->n_callers++;
1090 ((inline_irg_env *)get_irg_link(callee))->n_callers_orig++;
1094 INLINE static int is_leave(ir_graph *irg) {
1095 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1098 INLINE static int is_smaller(ir_graph *callee, int size) {
1099 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1104 * Inlines small leave methods at call sites where the called address comes
1105 * from a Const node that references the entity representing the called
1107 * The size argument is a rough measure for the code size of the method:
1108 * Methods where the obstack containing the firm graph is smaller than
1111 void inline_leave_functions(int maxsize, int leavesize, int size) {
1112 inline_irg_env *env;
1113 int i, n_irgs = get_irp_n_irgs();
1114 ir_graph *rem = current_ir_graph;
1117 if (!(get_opt_optimize() && get_opt_inline())) return;
1119 /* extend all irgs by a temporary data structure for inlineing. */
1120 for (i = 0; i < n_irgs; ++i)
1121 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1123 /* Precompute information in temporary data structure. */
1124 for (i = 0; i < n_irgs; ++i) {
1125 current_ir_graph = get_irp_irg(i);
1126 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1127 assert(get_irg_callee_info_state(current_ir_graph) == irg_callee_info_none);
1129 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1130 get_irg_link(current_ir_graph));
1131 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1135 Inline leaves recursively -- we might construct new leaves. */
1136 /* int itercnt = 1; */
1137 while (did_inline) {
1138 /* printf("iteration %d\n", itercnt++); */
1140 for (i = 0; i < n_irgs; ++i) {
1143 int phiproj_computed = 0;
1145 current_ir_graph = get_irp_irg(i);
1146 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1148 /* we can not walk and change a set, nor remove from it.
1150 walkset = env->call_nodes;
1151 env->call_nodes = eset_create();
1152 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1153 inline_irg_env *callee_env;
1154 ir_graph *callee = get_call_called_irg(call);
1156 if (env->n_nodes > maxsize) break;
1158 ((is_leave(callee) && is_smaller(callee, leavesize)) ||
1159 (get_irg_inline_property(callee) == irg_inline_forced))) {
1160 if (!phiproj_computed) {
1161 phiproj_computed = 1;
1162 collect_phiprojs(current_ir_graph);
1164 callee_env = (inline_irg_env *)get_irg_link(callee);
1165 /* printf(" %s: Inlineing %s.\n", get_entity_name(get_irg_entity(current_ir_graph)), */
1166 /* get_entity_name(get_irg_entity(callee))); */
1167 inline_method(call, callee);
1169 env->n_call_nodes--;
1170 eset_insert_all(env->call_nodes, callee_env->call_nodes);
1171 env->n_call_nodes += callee_env->n_call_nodes;
1172 env->n_nodes += callee_env->n_nodes;
1173 callee_env->n_callers--;
1175 eset_insert(env->call_nodes, call);
1178 eset_destroy(walkset);
1182 /* printf("Non leaves\n"); */
1183 /* inline other small functions. */
1184 for (i = 0; i < n_irgs; ++i) {
1187 int phiproj_computed = 0;
1189 current_ir_graph = get_irp_irg(i);
1190 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1192 /* we can not walk and change a set, nor remove from it.
1194 walkset = env->call_nodes;
1195 env->call_nodes = eset_create();
1196 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1197 inline_irg_env *callee_env;
1198 ir_graph *callee = get_call_called_irg(call);
1200 if (env->n_nodes > maxsize) break;
1201 if (callee && is_smaller(callee, size)) {
1202 if (!phiproj_computed) {
1203 phiproj_computed = 1;
1204 collect_phiprojs(current_ir_graph);
1206 callee_env = (inline_irg_env *)get_irg_link(callee);
1207 /* printf(" %s: Inlineing %s.\n", get_entity_name(get_irg_entity(current_ir_graph)), */
1208 /* get_entity_name(get_irg_entity(callee))); */
1209 inline_method(call, callee);
1211 env->n_call_nodes--;
1212 eset_insert_all(env->call_nodes, callee_env->call_nodes);
1213 env->n_call_nodes += callee_env->n_call_nodes;
1214 env->n_nodes += callee_env->n_nodes;
1215 callee_env->n_callers--;
1217 eset_insert(env->call_nodes, call);
1220 eset_destroy(walkset);
1223 for (i = 0; i < n_irgs; ++i) {
1224 current_ir_graph = get_irp_irg(i);
1226 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1227 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1228 (env->n_callers_orig != env->n_callers))
1229 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1230 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1231 env->n_callers_orig, env->n_callers,
1232 get_entity_name(get_irg_entity(current_ir_graph)));
1234 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1237 current_ir_graph = rem;
1240 /*-----------------------------------------------------------------*/
1241 /* Code Placement. Pins all floating nodes to a block where they */
1242 /* will be executed only if needed. */
1243 /*-----------------------------------------------------------------*/
1246 * Find the earliest correct block for N. --- Place N into the
1247 * same Block as its dominance-deepest Input.
1250 place_floats_early(ir_node *n, pdeq *worklist)
1252 int i, start, irn_arity;
1254 /* we must not run into an infinite loop */
1255 assert (irn_not_visited(n));
1256 mark_irn_visited(n);
1258 /* Place floating nodes. */
1259 if (get_op_pinned(intern_get_irn_op(n)) == floats) {
1261 ir_node *b = new_Bad(); /* The block to place this node in */
1263 assert(intern_get_irn_op(n) != op_Block);
1265 if ((intern_get_irn_op(n) == op_Const) ||
1266 (intern_get_irn_op(n) == op_SymConst) ||
1268 (intern_get_irn_op(n) == op_Unknown)) {
1269 /* These nodes will not be placed by the loop below. */
1270 b = get_irg_start_block(current_ir_graph);
1274 /* find the block for this node. */
1275 irn_arity = intern_get_irn_arity(n);
1276 for (i = 0; i < irn_arity; i++) {
1277 ir_node *dep = intern_get_irn_n(n, i);
1279 if ((irn_not_visited(dep)) &&
1280 (get_op_pinned(intern_get_irn_op(dep)) == floats)) {
1281 place_floats_early(dep, worklist);
1283 /* Because all loops contain at least one pinned node, now all
1284 our inputs are either pinned or place_early has already
1285 been finished on them. We do not have any unfinished inputs! */
1286 dep_block = get_nodes_Block(dep);
1287 if ((!is_Bad(dep_block)) &&
1288 (get_Block_dom_depth(dep_block) > depth)) {
1290 depth = get_Block_dom_depth(dep_block);
1292 /* Avoid that the node is placed in the Start block */
1293 if ((depth == 1) && (get_Block_dom_depth(get_nodes_Block(n)) > 1)) {
1294 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1295 assert(b != get_irg_start_block(current_ir_graph));
1299 set_nodes_Block(n, b);
1302 /* Add predecessors of non floating nodes on worklist. */
1303 start = (intern_get_irn_op(n) == op_Block) ? 0 : -1;
1304 irn_arity = intern_get_irn_arity(n);
1305 for (i = start; i < irn_arity; i++) {
1306 ir_node *pred = intern_get_irn_n(n, i);
1307 if (irn_not_visited(pred)) {
1308 pdeq_putr (worklist, pred);
1314 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1315 * Start, Call and end at pinned nodes as Store, Call. Place_early
1316 * places all floating nodes reachable from its argument through floating
1317 * nodes and adds all beginnings at pinned nodes to the worklist.
1319 static INLINE void place_early(pdeq* worklist) {
1321 inc_irg_visited(current_ir_graph);
1323 /* this inits the worklist */
1324 place_floats_early(get_irg_end(current_ir_graph), worklist);
1326 /* Work the content of the worklist. */
1327 while (!pdeq_empty (worklist)) {
1328 ir_node *n = pdeq_getl (worklist);
1329 if (irn_not_visited(n)) place_floats_early(n, worklist);
1332 set_irg_outs_inconsistent(current_ir_graph);
1333 current_ir_graph->pinned = pinned;
1337 /** deepest common dominance ancestor of DCA and CONSUMER of PRODUCER. */
1339 consumer_dom_dca (ir_node *dca, ir_node *consumer, ir_node *producer)
1341 ir_node *block = NULL;
1343 /* Compute the latest block into which we can place a node so that it is
1345 if (intern_get_irn_op(consumer) == op_Phi) {
1346 /* our consumer is a Phi-node, the effective use is in all those
1347 blocks through which the Phi-node reaches producer */
1349 ir_node *phi_block = get_nodes_Block(consumer);
1350 irn_arity = intern_get_irn_arity(consumer);
1351 for (i = 0; i < irn_arity; i++) {
1352 if (intern_get_irn_n(consumer, i) == producer) {
1353 block = get_nodes_Block(get_Block_cfgpred(phi_block, i));
1357 assert(is_no_Block(consumer));
1358 block = get_nodes_Block(consumer);
1361 /* Compute the deepest common ancestor of block and dca. */
1363 if (!dca) return block;
1364 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1365 block = get_Block_idom(block);
1366 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block))
1367 dca = get_Block_idom(dca);
1368 while (block != dca)
1369 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1374 static INLINE int get_irn_loop_depth(ir_node *n) {
1375 return get_loop_depth(get_irn_loop(n));
1379 * Move n to a block with less loop depth than it's current block. The
1380 * new block must be dominated by early.
1383 move_out_of_loops (ir_node *n, ir_node *early)
1385 ir_node *best, *dca;
1389 /* Find the region deepest in the dominator tree dominating
1390 dca with the least loop nesting depth, but still dominated
1391 by our early placement. */
1392 dca = get_nodes_Block(n);
1394 while (dca != early) {
1395 dca = get_Block_idom(dca);
1396 if (!dca) break; /* should we put assert(dca)? */
1397 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1401 if (best != get_nodes_Block(n)) {
1403 printf("Moving out of loop: "); DDMN(n);
1404 printf(" Outermost block: "); DDMN(early);
1405 printf(" Best block: "); DDMN(best);
1406 printf(" Innermost block: "); DDMN(get_nodes_Block(n));
1408 set_nodes_Block(n, best);
1413 * Find the latest legal block for N and place N into the
1414 * `optimal' Block between the latest and earliest legal block.
1415 * The `optimal' block is the dominance-deepest block of those
1416 * with the least loop-nesting-depth. This places N out of as many
1417 * loops as possible and then makes it as control dependant as
1421 place_floats_late(ir_node *n, pdeq *worklist)
1426 assert (irn_not_visited(n)); /* no multiple placement */
1428 /* no need to place block nodes, control nodes are already placed. */
1429 if ((intern_get_irn_op(n) != op_Block) &&
1431 (intern_get_irn_mode(n) != mode_X)) {
1432 /* Remember the early placement of this block to move it
1433 out of loop no further than the early placement. */
1434 early = get_nodes_Block(n);
1435 /* Assure that our users are all placed, except the Phi-nodes.
1436 --- Each data flow cycle contains at least one Phi-node. We
1437 have to break the `user has to be placed before the
1438 producer' dependence cycle and the Phi-nodes are the
1439 place to do so, because we need to base our placement on the
1440 final region of our users, which is OK with Phi-nodes, as they
1441 are pinned, and they never have to be placed after a
1442 producer of one of their inputs in the same block anyway. */
1443 for (i = 0; i < get_irn_n_outs(n); i++) {
1444 ir_node *succ = get_irn_out(n, i);
1445 if (irn_not_visited(succ) && (intern_get_irn_op(succ) != op_Phi))
1446 place_floats_late(succ, worklist);
1449 /* We have to determine the final block of this node... except for
1451 if ((get_op_pinned(intern_get_irn_op(n)) == floats) &&
1452 (intern_get_irn_op(n) != op_Const) &&
1453 (intern_get_irn_op(n) != op_SymConst)) {
1454 ir_node *dca = NULL; /* deepest common ancestor in the
1455 dominator tree of all nodes'
1456 blocks depending on us; our final
1457 placement has to dominate DCA. */
1458 for (i = 0; i < get_irn_n_outs(n); i++) {
1459 dca = consumer_dom_dca (dca, get_irn_out(n, i), n);
1461 set_nodes_Block(n, dca);
1463 move_out_of_loops (n, early);
1467 mark_irn_visited(n);
1469 /* Add predecessors of all non-floating nodes on list. (Those of floating
1470 nodes are placeded already and therefore are marked.) */
1471 for (i = 0; i < get_irn_n_outs(n); i++) {
1472 if (irn_not_visited(get_irn_out(n, i))) {
1473 pdeq_putr (worklist, get_irn_out(n, i));
1478 static INLINE void place_late(pdeq* worklist) {
1480 inc_irg_visited(current_ir_graph);
1482 /* This fills the worklist initially. */
1483 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1484 /* And now empty the worklist again... */
1485 while (!pdeq_empty (worklist)) {
1486 ir_node *n = pdeq_getl (worklist);
1487 if (irn_not_visited(n)) place_floats_late(n, worklist);
1491 void place_code(ir_graph *irg) {
1493 ir_graph *rem = current_ir_graph;
1495 current_ir_graph = irg;
1497 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1499 /* Handle graph state */
1500 assert(get_irg_phase_state(irg) != phase_building);
1501 if (get_irg_dom_state(irg) != dom_consistent)
1504 if (get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1505 free_loop_information(irg);
1506 construct_backedges(irg);
1509 /* Place all floating nodes as early as possible. This guarantees
1510 a legal code placement. */
1511 worklist = new_pdeq();
1512 place_early(worklist);
1514 /* place_early invalidates the outs, place_late needs them. */
1516 /* Now move the nodes down in the dominator tree. This reduces the
1517 unnecessary executions of the node. */
1518 place_late(worklist);
1520 set_irg_outs_inconsistent(current_ir_graph);
1521 set_irg_loopinfo_inconsistent(current_ir_graph);
1523 current_ir_graph = rem;
1528 /********************************************************************/
1529 /* Control flow optimization. */
1530 /* Removes Bad control flow predecessors and empty blocks. A block */
1531 /* is empty if it contains only a Jmp node. */
1532 /* Blocks can only be removed if they are not needed for the */
1533 /* semantics of Phi nodes. */
1534 /********************************************************************/
1537 * Removes Tuples from Block control flow predecessors.
1538 * Optimizes blocks with equivalent_node().
1539 * Replaces n by Bad if n is unreachable control flow.
1541 static void merge_blocks(ir_node *n, void *env) {
1543 set_irn_link(n, NULL);
1545 if (intern_get_irn_op(n) == op_Block) {
1547 for (i = 0; i < get_Block_n_cfgpreds(n); i++)
1548 /* GL @@@ : is this possible? if (get_opt_normalize()) -- added, all tests go through.
1549 A different order of optimizations might cause problems. */
1550 if (get_opt_normalize())
1551 set_Block_cfgpred(n, i, skip_Tuple(get_Block_cfgpred(n, i)));
1552 } else if (get_opt_optimize() && (intern_get_irn_mode(n) == mode_X)) {
1553 /* We will soon visit a block. Optimize it before visiting! */
1554 ir_node *b = get_nodes_Block(n);
1555 ir_node *new_node = equivalent_node(b);
1556 while (irn_not_visited(b) && (!is_Bad(new_node)) && (new_node != b)) {
1557 /* We would have to run gigo if new is bad, so we
1558 promote it directly below. */
1559 assert(((b == new_node) ||
1560 get_opt_control_flow_straightening() ||
1561 get_opt_control_flow_weak_simplification()) &&
1562 ("strange flag setting"));
1563 exchange (b, new_node);
1565 new_node = equivalent_node(b);
1567 /* GL @@@ get_opt_normalize hinzugefuegt, 5.5.2003 */
1568 if (is_Bad(new_node) && get_opt_normalize()) exchange(n, new_Bad());
1573 * Collects all Phi nodes in link list of Block.
1574 * Marks all blocks "block_visited" if they contain a node other
1577 static void collect_nodes(ir_node *n, void *env) {
1578 if (is_no_Block(n)) {
1579 ir_node *b = get_nodes_Block(n);
1581 if ((intern_get_irn_op(n) == op_Phi)) {
1582 /* Collect Phi nodes to compact ins along with block's ins. */
1583 set_irn_link(n, get_irn_link(b));
1585 } else if (intern_get_irn_op(n) != op_Jmp) { /* Check for non empty block. */
1586 mark_Block_block_visited(b);
1591 /** Returns true if pred is predecessor of block. */
1592 static int is_pred_of(ir_node *pred, ir_node *b) {
1594 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1595 ir_node *b_pred = get_nodes_Block(get_Block_cfgpred(b, i));
1596 if (b_pred == pred) return 1;
1601 static int test_whether_dispensable(ir_node *b, int pos) {
1602 int i, j, n_preds = 1;
1603 int dispensable = 1;
1604 ir_node *cfop = get_Block_cfgpred(b, pos);
1605 ir_node *pred = get_nodes_Block(cfop);
1607 if (get_Block_block_visited(pred) + 1
1608 < get_irg_block_visited(current_ir_graph)) {
1609 if (!get_opt_optimize() || !get_opt_control_flow_strong_simplification()) {
1610 /* Mark block so that is will not be removed. */
1611 set_Block_block_visited(pred, get_irg_block_visited(current_ir_graph)-1);
1614 /* Seems to be empty. */
1615 if (!get_irn_link(b)) {
1616 /* There are no Phi nodes ==> dispensable. */
1617 n_preds = get_Block_n_cfgpreds(pred);
1619 /* b's pred blocks and pred's pred blocks must be pairwise disjunct.
1620 Work preds < pos as if they were already removed. */
1621 for (i = 0; i < pos; i++) {
1622 ir_node *b_pred = get_nodes_Block(get_Block_cfgpred(b, i));
1623 if (get_Block_block_visited(b_pred) + 1
1624 < get_irg_block_visited(current_ir_graph)) {
1625 for (j = 0; j < get_Block_n_cfgpreds(b_pred); j++) {
1626 ir_node *b_pred_pred = get_nodes_Block(get_Block_cfgpred(b_pred, j));
1627 if (is_pred_of(b_pred_pred, pred)) dispensable = 0;
1630 if (is_pred_of(b_pred, pred)) dispensable = 0;
1633 for (i = pos +1; i < get_Block_n_cfgpreds(b); i++) {
1634 ir_node *b_pred = get_nodes_Block(get_Block_cfgpred(b, i));
1635 if (is_pred_of(b_pred, pred)) dispensable = 0;
1638 set_Block_block_visited(pred, get_irg_block_visited(current_ir_graph)-1);
1641 n_preds = get_Block_n_cfgpreds(pred);
1649 static void optimize_blocks(ir_node *b, void *env) {
1650 int i, j, k, max_preds, n_preds;
1651 ir_node *pred, *phi;
1654 /* Count the number of predecessor if this block is merged with pred blocks
1657 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1658 max_preds += test_whether_dispensable(b, i);
1660 in = (ir_node **) malloc(max_preds * sizeof(ir_node *));
1663 printf(" working on "); DDMN(b);
1664 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1665 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1666 if (is_Bad(get_Block_cfgpred(b, i))) {
1667 printf(" removing Bad %i\n ", i);
1668 } else if (get_Block_block_visited(pred) +1
1669 < get_irg_block_visited(current_ir_graph)) {
1670 printf(" removing pred %i ", i); DDMN(pred);
1671 } else { printf(" Nothing to do for "); DDMN(pred); }
1673 * end Debug output -*/
1675 /*- Fix the Phi nodes -*/
1676 phi = get_irn_link(b);
1678 assert(intern_get_irn_op(phi) == op_Phi);
1679 /* Find the new predecessors for the Phi */
1681 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1682 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1683 if (is_Bad(get_Block_cfgpred(b, i))) {
1685 } else if (get_Block_block_visited(pred) +1
1686 < get_irg_block_visited(current_ir_graph)) {
1687 /* It's an empty block and not yet visited. */
1688 ir_node *phi_pred = get_Phi_pred(phi, i);
1689 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1690 if (get_nodes_Block(phi_pred) == pred) {
1691 assert(intern_get_irn_op(phi_pred) == op_Phi); /* Block is empty!! */
1692 in[n_preds] = get_Phi_pred(phi_pred, j);
1694 in[n_preds] = phi_pred;
1698 /* The Phi_pred node is replaced now if it is a Phi.
1699 In Schleifen kann offenbar der entfernte Phi Knoten legal verwendet werden.
1700 Daher muss der Phiknoten durch den neuen ersetzt werden.
1701 Weiter muss der alte Phiknoten entfernt werden (durch ersetzen oder
1702 durch einen Bad) damit er aus den keep_alive verschwinden kann.
1703 Man sollte also, falls keine Schleife vorliegt, exchange mit new_Bad
1705 if (get_nodes_Block(phi_pred) == pred) {
1706 /* remove the Phi as it might be kept alive. Further there
1707 might be other users. */
1708 exchange(phi_pred, phi); /* geht, ist aber doch semantisch falsch! Warum?? */
1711 in[n_preds] = get_Phi_pred(phi, i);
1716 set_irn_in(phi, n_preds, in);
1718 phi = get_irn_link(phi);
1722 This happens only if merge between loop backedge and single loop entry. -*/
1723 for (k = 0; k < get_Block_n_cfgpreds(b); k++) {
1724 pred = get_nodes_Block(get_Block_cfgpred(b, k));
1725 if (get_Block_block_visited(pred) +1
1726 < get_irg_block_visited(current_ir_graph)) {
1727 phi = get_irn_link(pred);
1729 if (intern_get_irn_op(phi) == op_Phi) {
1730 set_nodes_Block(phi, b);
1733 for (i = 0; i < k; i++) {
1734 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1735 if (is_Bad(get_Block_cfgpred(b, i))) {
1737 } else if (get_Block_block_visited(pred) +1
1738 < get_irg_block_visited(current_ir_graph)) {
1739 /* It's an empty block and not yet visited. */
1740 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1741 /* @@@ Hier brauche ich Schleifeninformation!!! Kontrollflusskante
1742 muss Rueckwaertskante sein! (An allen vier in[n_preds] = phi
1743 Anweisungen.) Trotzdem tuts bisher!! */
1752 for (i = 0; i < get_Phi_n_preds(phi); i++) {
1753 in[n_preds] = get_Phi_pred(phi, i);
1756 for (i = k+1; i < get_Block_n_cfgpreds(b); i++) {
1757 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1758 if (is_Bad(get_Block_cfgpred(b, i))) {
1760 } else if (get_Block_block_visited(pred) +1
1761 < get_irg_block_visited(current_ir_graph)) {
1762 /* It's an empty block and not yet visited. */
1763 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1772 set_irn_in(phi, n_preds, in);
1774 phi = get_irn_link(phi);
1779 /*- Fix the block -*/
1781 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1782 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1783 if (is_Bad(get_Block_cfgpred(b, i))) {
1785 } else if (get_Block_block_visited(pred) +1
1786 < get_irg_block_visited(current_ir_graph)) {
1787 /* It's an empty block and not yet visited. */
1788 assert(get_Block_n_cfgpreds(b) > 1);
1789 /* Else it should be optimized by equivalent_node. */
1790 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1791 in[n_preds] = get_Block_cfgpred(pred, j);
1794 /* Remove block as it might be kept alive. */
1795 exchange(pred, b/*new_Bad()*/);
1797 in[n_preds] = get_Block_cfgpred(b, i);
1801 set_irn_in(b, n_preds, in);
1805 void optimize_cf(ir_graph *irg) {
1808 ir_node *end = get_irg_end(irg);
1809 ir_graph *rem = current_ir_graph;
1810 current_ir_graph = irg;
1812 /* Handle graph state */
1813 assert(get_irg_phase_state(irg) != phase_building);
1814 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
1815 set_irg_outs_inconsistent(current_ir_graph);
1816 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
1817 set_irg_dom_inconsistent(current_ir_graph);
1819 /* Use block visited flag to mark non-empty blocks. */
1820 inc_irg_block_visited(irg);
1821 irg_walk(end, merge_blocks, collect_nodes, NULL);
1823 /* Optimize the standard code. */
1824 irg_block_walk(get_irg_end_block(irg), optimize_blocks, NULL, NULL);
1826 /* Walk all keep alives, optimize them if block, add to new in-array
1827 for end if useful. */
1828 in = NEW_ARR_F (ir_node *, 1);
1829 in[0] = get_nodes_Block(end);
1830 inc_irg_visited(current_ir_graph);
1831 for(i = 0; i < get_End_n_keepalives(end); i++) {
1832 ir_node *ka = get_End_keepalive(end, i);
1833 if (irn_not_visited(ka)) {
1834 if ((intern_get_irn_op(ka) == op_Block) && Block_not_block_visited(ka)) {
1835 set_irg_block_visited(current_ir_graph, /* Don't walk all the way to Start. */
1836 get_irg_block_visited(current_ir_graph)-1);
1837 irg_block_walk(ka, optimize_blocks, NULL, NULL);
1838 mark_irn_visited(ka);
1839 ARR_APP1 (ir_node *, in, ka);
1840 } else if (intern_get_irn_op(ka) == op_Phi) {
1841 mark_irn_visited(ka);
1842 ARR_APP1 (ir_node *, in, ka);
1846 /* DEL_ARR_F(end->in); GL @@@ tut nicht ! */
1849 current_ir_graph = rem;
1854 * Called by walker of remove_critical_cf_edges.
1856 * Place an empty block to an edge between a blocks of multiple
1857 * predecessors and a block of multiple successors.
1860 * @param env Environment of walker. This field is unused and has
1863 static void walk_critical_cf_edges(ir_node *n, void *env) {
1865 ir_node *pre, *block, **in, *jmp;
1867 /* Block has multiple predecessors */
1868 if ((op_Block == intern_get_irn_op(n)) &&
1869 (intern_get_irn_arity(n) > 1)) {
1870 arity = intern_get_irn_arity(n);
1872 if (n == get_irg_end_block(current_ir_graph))
1873 return; /* No use to add a block here. */
1875 for (i=0; i<arity; i++) {
1876 pre = intern_get_irn_n(n, i);
1877 /* Predecessor has multiple successors. Insert new flow edge */
1878 if ((NULL != pre) &&
1879 (op_Proj == intern_get_irn_op(pre)) &&
1880 op_Raise != intern_get_irn_op(skip_Proj(pre))) {
1882 /* set predecessor array for new block */
1883 in = NEW_ARR_D (ir_node *, current_ir_graph->obst, 1);
1884 /* set predecessor of new block */
1886 block = new_Block(1, in);
1887 /* insert new jmp node to new block */
1888 switch_block(block);
1891 /* set successor of new block */
1892 set_irn_n(n, i, jmp);
1894 } /* predecessor has multiple successors */
1895 } /* for all predecessors */
1896 } /* n is a block */
1899 void remove_critical_cf_edges(ir_graph *irg) {
1900 if (get_opt_critical_edges())
1901 irg_walk_graph(irg, NULL, walk_critical_cf_edges, NULL);