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"
37 # include "firmstat.h"
39 /* Defined in iropt.c */
40 pset *new_identities (void);
41 void del_identities (pset *value_table);
42 void add_identities (pset *value_table, ir_node *node);
44 /*------------------------------------------------------------------*/
45 /* apply optimizations of iropt to all nodes. */
46 /*------------------------------------------------------------------*/
48 static void init_link (ir_node *n, void *env) {
49 set_irn_link(n, NULL);
52 #if 0 /* Old version. Avoids Ids.
53 This is not necessary: we do a postwalk, and get_irn_n
54 removes ids anyways. So it's much cheaper to call the
55 optimization less often and use the exchange() algorithm. */
57 optimize_in_place_wrapper (ir_node *n, void *env) {
59 ir_node *optimized, *old;
61 irn_arity = intern_get_irn_arity(n);
62 for (i = 0; i < irn_arity; i++) {
63 /* get_irn_n skips Id nodes, so comparison old != optimized does not
64 show all optimizations. Therefore always set new predecessor. */
65 old = intern_get_irn_intra_n(n, i);
66 optimized = optimize_in_place_2(old);
67 set_irn_n(n, i, optimized);
70 if (intern_get_irn_op(n) == op_Block) {
71 optimized = optimize_in_place_2(n);
72 if (optimized != n) exchange (n, optimized);
77 optimize_in_place_wrapper (ir_node *n, void *env) {
78 ir_node *optimized = optimize_in_place_2(n);
79 if (optimized != n) exchange (n, optimized);
86 local_optimize_graph (ir_graph *irg) {
87 ir_graph *rem = current_ir_graph;
88 current_ir_graph = irg;
90 /* Handle graph state */
91 assert(get_irg_phase_state(irg) != phase_building);
92 if (get_opt_global_cse())
93 set_irg_pinned(current_ir_graph, 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);
99 /* Clean the value_table in irg for the cse. */
100 del_identities(irg->value_table);
101 irg->value_table = new_identities();
103 /* walk over the graph */
104 irg_walk(irg->end, init_link, optimize_in_place_wrapper, NULL);
106 current_ir_graph = rem;
109 /*------------------------------------------------------------------*/
110 /* Routines for dead node elimination / copying garbage collection */
111 /* of the obstack. */
112 /*------------------------------------------------------------------*/
115 * Remember the new node in the old node by using a field all nodes have.
118 set_new_node (ir_node *old, ir_node *new)
124 * Get this new node, before the old node is forgotton.
126 static INLINE ir_node *
127 get_new_node (ir_node * n)
133 * We use the block_visited flag to mark that we have computed the
134 * number of useful predecessors for this block.
135 * Further we encode the new arity in this flag in the old blocks.
136 * Remembering the arity is useful, as it saves a lot of pointer
137 * accesses. This function is called for all Phi and Block nodes
141 compute_new_arity(ir_node *b) {
142 int i, res, irn_arity;
145 irg_v = get_irg_block_visited(current_ir_graph);
146 block_v = get_Block_block_visited(b);
147 if (block_v >= irg_v) {
148 /* we computed the number of preds for this block and saved it in the
150 return block_v - irg_v;
152 /* compute the number of good predecessors */
153 res = irn_arity = intern_get_irn_arity(b);
154 for (i = 0; i < irn_arity; i++)
155 if (intern_get_irn_opcode(intern_get_irn_n(b, i)) == iro_Bad) res--;
156 /* save it in the flag. */
157 set_Block_block_visited(b, irg_v + res);
162 /* TODO: add an ir_op operation */
163 static INLINE void new_backedge_info(ir_node *n) {
164 switch(intern_get_irn_opcode(n)) {
166 n->attr.block.cg_backedge = NULL;
167 n->attr.block.backedge = new_backedge_arr(current_ir_graph->obst, intern_get_irn_arity(n));
170 n->attr.phi_backedge = new_backedge_arr(current_ir_graph->obst, intern_get_irn_arity(n));
173 n->attr.filter.backedge = new_backedge_arr(current_ir_graph->obst, intern_get_irn_arity(n));
180 * Copies the node to the new obstack. The Ins of the new node point to
181 * the predecessors on the old obstack. For block/phi nodes not all
182 * predecessors might be copied. n->link points to the new node.
183 * For Phi and Block nodes the function allocates in-arrays with an arity
184 * only for useful predecessors. The arity is determined by counting
185 * the non-bad predecessors of the block.
188 copy_node (ir_node *n, void *env) {
192 /* The end node looses it's flexible in array. This doesn't matter,
193 as dead node elimination builds End by hand, inlineing doesn't use
195 /* assert(n->op == op_End || ((_ARR_DESCR(n->in))->cookie != ARR_F_MAGIC)); */
197 if (intern_get_irn_opcode(n) == iro_Block) {
199 new_arity = compute_new_arity(n);
200 n->attr.block.graph_arr = NULL;
202 block = get_nodes_Block(n);
203 if (intern_get_irn_opcode(n) == iro_Phi) {
204 new_arity = compute_new_arity(block);
206 new_arity = intern_get_irn_arity(n);
209 nn = new_ir_node(get_irn_dbg_info(n),
212 intern_get_irn_op(n),
213 intern_get_irn_mode(n),
216 /* Copy the attributes. These might point to additional data. If this
217 was allocated on the old obstack the pointers now are dangling. This
218 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
220 new_backedge_info(nn);
223 /* printf("\n old node: "); DDMSG2(n);
224 printf(" new node: "); DDMSG2(nn); */
229 * Copies new predecessors of old node to new node remembered in link.
230 * Spare the Bad predecessors of Phi and Block nodes.
233 copy_preds (ir_node *n, void *env) {
237 nn = get_new_node(n);
239 /* printf("\n old node: "); DDMSG2(n);
240 printf(" new node: "); DDMSG2(nn);
241 printf(" arities: old: %d, new: %d\n", intern_get_irn_arity(n), intern_get_irn_arity(nn)); */
243 if (intern_get_irn_opcode(n) == iro_Block) {
244 /* Don't copy Bad nodes. */
246 irn_arity = intern_get_irn_arity(n);
247 for (i = 0; i < irn_arity; i++)
248 if (intern_get_irn_opcode(intern_get_irn_n(n, i)) != iro_Bad) {
249 set_irn_n (nn, j, get_new_node(intern_get_irn_n(n, i)));
250 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
253 /* repair the block visited flag from above misuse. Repair it in both
254 graphs so that the old one can still be used. */
255 set_Block_block_visited(nn, 0);
256 set_Block_block_visited(n, 0);
257 /* Local optimization could not merge two subsequent blocks if
258 in array contained Bads. Now it's possible.
259 We don't call optimize_in_place as it requires
260 that the fields in ir_graph are set properly. */
261 if ((get_opt_control_flow_straightening()) &&
262 (get_Block_n_cfgpreds(nn) == 1) &&
263 (intern_get_irn_op(get_Block_cfgpred(nn, 0)) == op_Jmp))
264 exchange(nn, get_nodes_Block(get_Block_cfgpred(nn, 0)));
265 } else if (intern_get_irn_opcode(n) == iro_Phi) {
266 /* Don't copy node if corresponding predecessor in block is Bad.
267 The Block itself should not be Bad. */
268 block = get_nodes_Block(n);
269 set_irn_n (nn, -1, get_new_node(block));
271 irn_arity = intern_get_irn_arity(n);
272 for (i = 0; i < irn_arity; i++)
273 if (intern_get_irn_opcode(intern_get_irn_n(block, i)) != iro_Bad) {
274 set_irn_n (nn, j, get_new_node(intern_get_irn_n(n, i)));
275 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
278 /* If the pre walker reached this Phi after the post walker visited the
279 block block_visited is > 0. */
280 set_Block_block_visited(get_nodes_Block(n), 0);
281 /* Compacting the Phi's ins might generate Phis with only one
283 if (intern_get_irn_arity(n) == 1)
284 exchange(n, intern_get_irn_n(n, 0));
286 irn_arity = intern_get_irn_arity(n);
287 for (i = -1; i < irn_arity; i++)
288 set_irn_n (nn, i, get_new_node(intern_get_irn_n(n, i)));
290 /* Now the new node is complete. We can add it to the hash table for cse.
291 @@@ inlinening aborts if we identify End. Why? */
292 if(intern_get_irn_op(nn) != op_End)
293 add_identities (current_ir_graph->value_table, nn);
297 * Copies the graph recursively, compacts the keepalive of the end node.
301 ir_node *oe, *ne; /* old end, new end */
302 ir_node *ka; /* keep alive */
305 oe = get_irg_end(current_ir_graph);
306 /* copy the end node by hand, allocate dynamic in array! */
307 ne = new_ir_node(get_irn_dbg_info(oe),
314 /* Copy the attributes. Well, there might be some in the future... */
316 set_new_node(oe, ne);
318 /* copy the live nodes */
319 irg_walk(get_nodes_Block(oe), copy_node, copy_preds, NULL);
320 /* copy_preds for the end node ... */
321 set_nodes_Block(ne, get_new_node(get_nodes_Block(oe)));
323 /*- ... and now the keep alives. -*/
324 /* First pick the not marked block nodes and walk them. We must pick these
325 first as else we will oversee blocks reachable from Phis. */
326 irn_arity = intern_get_irn_arity(oe);
327 for (i = 0; i < irn_arity; i++) {
328 ka = intern_get_irn_intra_n(oe, i);
329 if ((intern_get_irn_op(ka) == op_Block) &&
330 (get_irn_visited(ka) < get_irg_visited(current_ir_graph))) {
331 /* We must keep the block alive and copy everything reachable */
332 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
333 irg_walk(ka, copy_node, copy_preds, NULL);
334 add_End_keepalive(ne, get_new_node(ka));
338 /* Now pick the Phis. Here we will keep all! */
339 irn_arity = intern_get_irn_arity(oe);
340 for (i = 0; i < irn_arity; i++) {
341 ka = intern_get_irn_intra_n(oe, i);
342 if ((intern_get_irn_op(ka) == op_Phi)) {
343 if (get_irn_visited(ka) < get_irg_visited(current_ir_graph)) {
344 /* We didn't copy the Phi yet. */
345 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
346 irg_walk(ka, copy_node, copy_preds, NULL);
348 add_End_keepalive(ne, get_new_node(ka));
354 * Copies the graph reachable from current_ir_graph->end to the obstack
355 * in current_ir_graph and fixes the environment.
356 * Then fixes the fields in current_ir_graph containing nodes of the
360 copy_graph_env (void) {
362 /* Not all nodes remembered in current_ir_graph might be reachable
363 from the end node. Assure their link is set to NULL, so that
364 we can test whether new nodes have been computed. */
365 set_irn_link(get_irg_frame (current_ir_graph), NULL);
366 set_irn_link(get_irg_globals(current_ir_graph), NULL);
367 set_irn_link(get_irg_args (current_ir_graph), NULL);
369 /* we use the block walk flag for removing Bads from Blocks ins. */
370 inc_irg_block_visited(current_ir_graph);
375 /* fix the fields in current_ir_graph */
376 old_end = get_irg_end(current_ir_graph);
377 set_irg_end (current_ir_graph, get_new_node(old_end));
378 set_irg_end_except (current_ir_graph, get_irg_end(current_ir_graph));
379 set_irg_end_reg (current_ir_graph, get_irg_end(current_ir_graph));
381 set_irg_end_block (current_ir_graph, get_new_node(get_irg_end_block(current_ir_graph)));
382 if (get_irn_link(get_irg_frame(current_ir_graph)) == NULL) {
383 copy_node (get_irg_frame(current_ir_graph), NULL);
384 copy_preds(get_irg_frame(current_ir_graph), NULL);
386 if (get_irn_link(get_irg_globals(current_ir_graph)) == NULL) {
387 copy_node (get_irg_globals(current_ir_graph), NULL);
388 copy_preds(get_irg_globals(current_ir_graph), NULL);
390 if (get_irn_link(get_irg_args(current_ir_graph)) == NULL) {
391 copy_node (get_irg_args(current_ir_graph), NULL);
392 copy_preds(get_irg_args(current_ir_graph), NULL);
394 set_irg_start (current_ir_graph, get_new_node(get_irg_start(current_ir_graph)));
396 set_irg_start_block(current_ir_graph,
397 get_new_node(get_irg_start_block(current_ir_graph)));
398 set_irg_frame (current_ir_graph, get_new_node(get_irg_frame(current_ir_graph)));
399 set_irg_globals(current_ir_graph, get_new_node(get_irg_globals(current_ir_graph)));
400 set_irg_args (current_ir_graph, get_new_node(get_irg_args(current_ir_graph)));
401 if (get_irn_link(get_irg_bad(current_ir_graph)) == NULL) {
402 copy_node(get_irg_bad(current_ir_graph), NULL);
403 copy_preds(get_irg_bad(current_ir_graph), NULL);
405 set_irg_bad(current_ir_graph, get_new_node(get_irg_bad(current_ir_graph)));
406 /* GL removed: we need unknown with mode for analyses.
407 if (get_irn_link(get_irg_unknown(current_ir_graph)) == NULL) {
408 copy_node(get_irg_unknown(current_ir_graph), NULL);
409 copy_preds(get_irg_unknown(current_ir_graph), NULL);
411 set_irg_unknown(current_ir_graph, get_new_node(get_irg_unknown(current_ir_graph)));
416 * Copies all reachable nodes to a new obstack. Removes bad inputs
417 * from block nodes and the corresponding inputs from Phi nodes.
418 * Merges single exit blocks with single entry blocks and removes
420 * Adds all new nodes to a new hash table for cse. Does not
421 * perform cse, so the hash table might contain common subexpressions.
423 /* Amroq call this emigrate() */
425 dead_node_elimination(ir_graph *irg) {
427 int rem_ipview = interprocedural_view;
428 struct obstack *graveyard_obst = NULL;
429 struct obstack *rebirth_obst = NULL;
431 /* Remember external state of current_ir_graph. */
432 rem = current_ir_graph;
433 current_ir_graph = irg;
434 interprocedural_view = 0;
436 /* Handle graph state */
437 assert(get_irg_phase_state(current_ir_graph) != phase_building);
438 assert(get_irg_callee_info_state(current_ir_graph) == irg_callee_info_none);
439 free_outs(current_ir_graph);
441 /* @@@ so far we loose loops when copying */
442 free_loop_information(current_ir_graph);
444 if (get_opt_optimize() && get_opt_dead_node_elimination()) {
446 /* A quiet place, where the old obstack can rest in peace,
447 until it will be cremated. */
448 graveyard_obst = irg->obst;
450 /* A new obstack, where the reachable nodes will be copied to. */
451 rebirth_obst = (struct obstack *) xmalloc (sizeof (struct obstack));
452 current_ir_graph->obst = rebirth_obst;
453 obstack_init (current_ir_graph->obst);
455 /* We also need a new hash table for cse */
456 del_identities (irg->value_table);
457 irg->value_table = new_identities ();
459 /* Copy the graph from the old to the new obstack */
462 /* Free memory from old unoptimized obstack */
463 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
464 xfree (graveyard_obst); /* ... then free it. */
467 current_ir_graph = rem;
468 interprocedural_view = rem_ipview;
472 * Relink bad predeseccors of a block and store the old in array to the
473 * link field. This function is called by relink_bad_predecessors().
474 * The array of link field starts with the block operand at position 0.
475 * If block has bad predecessors, create a new in array without bad preds.
476 * Otherwise let in array untouched.
478 static void relink_bad_block_predecessors(ir_node *n, void *env) {
479 ir_node **new_in, *irn;
480 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
482 /* if link field of block is NULL, look for bad predecessors otherwise
483 this is allready done */
484 if (intern_get_irn_op(n) == op_Block &&
485 get_irn_link(n) == NULL) {
487 /* save old predecessors in link field (position 0 is the block operand)*/
488 set_irn_link(n, (void *)get_irn_in(n));
490 /* count predecessors without bad nodes */
491 old_irn_arity = intern_get_irn_arity(n);
492 for (i = 0; i < old_irn_arity; i++)
493 if (!is_Bad(intern_get_irn_n(n, i))) new_irn_arity++;
495 /* arity changing: set new predecessors without bad nodes */
496 if (new_irn_arity < old_irn_arity) {
497 /* get new predecessor array without Block predecessor */
498 new_in = NEW_ARR_D (ir_node *, current_ir_graph->obst, (new_irn_arity+1));
500 /* set new predeseccors in array */
503 for (i = 1; i < old_irn_arity; i++) {
504 irn = intern_get_irn_n(n, i);
505 if (!is_Bad(irn)) new_in[new_irn_n++] = irn;
508 } /* ir node has bad predecessors */
510 } /* Block is not relinked */
514 * Relinks Bad predecesors from Bocks and Phis called by walker
515 * remove_bad_predecesors(). If n is a Block, call
516 * relink_bad_block_redecessors(). If n is a Phinode, call also the relinking
517 * function of Phi's Block. If this block has bad predecessors, relink preds
520 static void relink_bad_predecessors(ir_node *n, void *env) {
521 ir_node *block, **old_in;
522 int i, old_irn_arity, new_irn_arity;
524 /* relink bad predeseccors of a block */
525 if (intern_get_irn_op(n) == op_Block)
526 relink_bad_block_predecessors(n, env);
528 /* If Phi node relink its block and its predecessors */
529 if (intern_get_irn_op(n) == op_Phi) {
531 /* Relink predeseccors of phi's block */
532 block = get_nodes_Block(n);
533 if (get_irn_link(block) == NULL)
534 relink_bad_block_predecessors(block, env);
536 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
537 old_irn_arity = ARR_LEN(old_in);
539 /* Relink Phi predeseccors if count of predeseccors changed */
540 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
541 /* set new predeseccors in array
542 n->in[0] remains the same block */
544 for(i = 1; i < old_irn_arity; i++)
545 if (!is_Bad((ir_node *)old_in[i])) n->in[new_irn_arity++] = n->in[i];
547 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
550 } /* n is a Phi node */
554 * Removes Bad Bad predecesors from Blocks and the corresponding
555 * inputs to Phi nodes as in dead_node_elimination but without
557 * On walking up set the link field to NULL, on walking down call
558 * relink_bad_predecessors() (This function stores the old in array
559 * to the link field and sets a new in array if arity of predecessors
562 void remove_bad_predecessors(ir_graph *irg) {
563 irg_walk_graph(irg, init_link, relink_bad_predecessors, NULL);
567 /*--------------------------------------------------------------------*/
568 /* Funcionality for inlining */
569 /*--------------------------------------------------------------------*/
572 * Copy node for inlineing. Updates attributes that change when
573 * inlineing but not for dead node elimination.
575 * Copies the node by calling copy_node and then updates the entity if
576 * it's a local one. env must be a pointer of the frame type of the
577 * inlined procedure. The new entities must be in the link field of
581 copy_node_inline (ir_node *n, void *env) {
583 type *frame_tp = (type *)env;
586 if (intern_get_irn_op(n) == op_Sel) {
587 new = get_new_node (n);
588 assert(intern_get_irn_op(new) == op_Sel);
589 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
590 set_Sel_entity(new, get_entity_link(get_Sel_entity(n)));
592 } else if (intern_get_irn_op(n) == op_Block) {
593 new = get_new_node (n);
594 new->attr.block.irg = current_ir_graph;
598 static void find_addr(ir_node *node, void *env)
600 if (get_irn_opcode(node) == iro_Proj) {
601 if (get_Proj_proj(node) == pn_Start_P_value_arg_base)
607 * currently, we cannot inline two cases:
608 * - call with compound arguments
609 * - graphs that take the address of a parameter
611 * check these condition here
613 static int can_inline(ir_node *call, ir_graph *called_graph)
615 type *call_type = get_Call_type(call);
616 int params, ress, i, res;
618 assert(is_method_type(call_type));
620 params = get_method_n_params(call_type);
621 ress = get_method_n_ress(call_type);
624 for (i = 0; i < params; ++i) {
625 type *p_type = get_method_param_type(call_type, i);
627 if (is_compound_type(p_type))
632 for (i = 0; i < ress; ++i) {
633 type *r_type = get_method_res_type(call_type, i);
635 if (is_compound_type(r_type))
640 irg_walk_graph(called_graph, find_addr, NULL, &res);
645 void inline_method(ir_node *call, ir_graph *called_graph) {
647 ir_node *post_call, *post_bl;
649 ir_node *end, *end_bl;
653 int arity, n_ret, n_exc, n_res, i, j, rem_opt, irn_arity;
656 irg_inline_property prop = get_irg_inline_property(called_graph);
658 if ( (prop != irg_inline_forced) && (!get_opt_optimize() || !get_opt_inline() ||
659 (prop == irg_inline_forbidden))) return;
663 * currently, we cannot inline two cases:
664 * - call with compound arguments
665 * - graphs that take the address of a parameter
667 if (! can_inline(call, called_graph))
670 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
671 rem_opt = get_opt_optimize();
674 /* Handle graph state */
675 assert(get_irg_phase_state(current_ir_graph) != phase_building);
676 assert(get_irg_pinned(current_ir_graph) == pinned);
677 assert(get_irg_pinned(called_graph) == pinned);
678 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
679 set_irg_outs_inconsistent(current_ir_graph);
681 /* -- Check preconditions -- */
682 assert(get_irn_op(call) == op_Call);
683 /* @@@ does not work for InterfaceIII.java after cgana
684 assert(get_Call_type(call) == get_entity_type(get_irg_ent(called_graph)));
685 assert(smaller_type(get_entity_type(get_irg_ent(called_graph)),
686 get_Call_type(call)));
688 assert(get_type_tpop(get_Call_type(call)) == type_method);
689 if (called_graph == current_ir_graph) {
690 set_optimize(rem_opt);
694 /* here we know we WILL inline, so inform the statistics */
695 stat_inline(call, called_graph);
697 /* -- Decide how to handle exception control flow: Is there a handler
698 for the Call node, or do we branch directly to End on an exception?
699 exc_handling: 0 There is a handler.
701 2 Exception handling not represented in Firm. -- */
703 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
704 for (proj = (ir_node *)get_irn_link(call); proj; proj = (ir_node *)get_irn_link(proj)) {
705 assert(get_irn_op(proj) == op_Proj);
706 if (get_Proj_proj(proj) == pn_Call_X_except) Xproj = proj;
707 if (get_Proj_proj(proj) == pn_Call_M_except) Mproj = proj;
709 if (Mproj) { assert(Xproj); exc_handling = 0; } /* Mproj */
710 else if (Xproj) { exc_handling = 1; } /* !Mproj && Xproj */
711 else { exc_handling = 2; } /* !Mproj && !Xproj */
716 the procedure and later replaces the Start node of the called graph.
717 Post_call is the old Call node and collects the results of the called
718 graph. Both will end up being a tuple. -- */
719 post_bl = get_nodes_Block(call);
720 set_irg_current_block(current_ir_graph, post_bl);
721 /* XxMxPxP of Start + parameter of Call */
723 in[1] = get_Call_mem(call);
724 in[2] = get_irg_frame(current_ir_graph);
725 in[3] = get_irg_globals(current_ir_graph);
726 in[4] = new_Tuple (get_Call_n_params(call), get_Call_param_arr(call));
727 pre_call = new_Tuple(5, in);
731 The new block gets the ins of the old block, pre_call and all its
732 predecessors and all Phi nodes. -- */
733 part_block(pre_call);
735 /* -- Prepare state for dead node elimination -- */
736 /* Visited flags in calling irg must be >= flag in called irg.
737 Else walker and arity computation will not work. */
738 if (get_irg_visited(current_ir_graph) <= get_irg_visited(called_graph))
739 set_irg_visited(current_ir_graph, get_irg_visited(called_graph)+1);
740 if (get_irg_block_visited(current_ir_graph)< get_irg_block_visited(called_graph))
741 set_irg_block_visited(current_ir_graph, get_irg_block_visited(called_graph));
742 /* Set pre_call as new Start node in link field of the start node of
743 calling graph and pre_calls block as new block for the start block
745 Further mark these nodes so that they are not visited by the
747 set_irn_link(get_irg_start(called_graph), pre_call);
748 set_irn_visited(get_irg_start(called_graph),
749 get_irg_visited(current_ir_graph));
750 set_irn_link(get_irg_start_block(called_graph),
751 get_nodes_Block(pre_call));
752 set_irn_visited(get_irg_start_block(called_graph),
753 get_irg_visited(current_ir_graph));
755 /* Initialize for compaction of in arrays */
756 inc_irg_block_visited(current_ir_graph);
758 /* -- Replicate local entities of the called_graph -- */
759 /* copy the entities. */
760 called_frame = get_irg_frame_type(called_graph);
761 for (i = 0; i < get_class_n_members(called_frame); i++) {
762 entity *new_ent, *old_ent;
763 old_ent = get_class_member(called_frame, i);
764 new_ent = copy_entity_own(old_ent, get_cur_frame_type());
765 set_entity_link(old_ent, new_ent);
768 /* visited is > than that of called graph. With this trick visited will
769 remain unchanged so that an outer walker, e.g., searching the call nodes
770 to inline, calling this inline will not visit the inlined nodes. */
771 set_irg_visited(current_ir_graph, get_irg_visited(current_ir_graph)-1);
773 /* -- Performing dead node elimination inlines the graph -- */
774 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
776 /* @@@ endless loops are not copied!! -- they should be, I think... */
777 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds,
778 get_irg_frame_type(called_graph));
780 /* Repair called_graph */
781 set_irg_visited(called_graph, get_irg_visited(current_ir_graph));
782 set_irg_block_visited(called_graph, get_irg_block_visited(current_ir_graph));
783 set_Block_block_visited(get_irg_start_block(called_graph), 0);
785 /* -- Merge the end of the inlined procedure with the call site -- */
786 /* We will turn the old Call node into a Tuple with the following
789 0: Phi of all Memories of Return statements.
790 1: Jmp from new Block that merges the control flow from all exception
791 predecessors of the old end block.
792 2: Tuple of all arguments.
793 3: Phi of Exception memories.
794 In case the old Call directly branches to End on an exception we don't
795 need the block merging all exceptions nor the Phi of the exception
799 /* -- Precompute some values -- */
800 end_bl = get_new_node(get_irg_end_block(called_graph));
801 end = get_new_node(get_irg_end(called_graph));
802 arity = intern_get_irn_arity(end_bl); /* arity = n_exc + n_ret */
803 n_res = get_method_n_ress(get_Call_type(call));
805 res_pred = (ir_node **) malloc (n_res * sizeof (ir_node *));
806 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
808 set_irg_current_block(current_ir_graph, post_bl); /* just to make sure */
810 /* -- archive keepalives -- */
811 irn_arity = intern_get_irn_arity(end);
812 for (i = 0; i < irn_arity; i++)
813 add_End_keepalive(get_irg_end(current_ir_graph), intern_get_irn_n(end, i));
815 /* The new end node will die. We need not free as the in array is on the obstack:
816 copy_node only generated 'D' arrays. */
818 /* -- Replace Return nodes by Jump nodes. -- */
820 for (i = 0; i < arity; i++) {
822 ret = intern_get_irn_n(end_bl, i);
823 if (intern_get_irn_op(ret) == op_Return) {
824 cf_pred[n_ret] = new_r_Jmp(current_ir_graph, get_nodes_Block(ret));
828 set_irn_in(post_bl, n_ret, cf_pred);
830 /* -- Build a Tuple for all results of the method.
831 Add Phi node if there was more than one Return. -- */
832 turn_into_tuple(post_call, 4);
833 /* First the Memory-Phi */
835 for (i = 0; i < arity; i++) {
836 ret = intern_get_irn_n(end_bl, i);
837 if (intern_get_irn_op(ret) == op_Return) {
838 cf_pred[n_ret] = get_Return_mem(ret);
842 phi = new_Phi(n_ret, cf_pred, mode_M);
843 set_Tuple_pred(call, 0, phi);
844 /* Conserve Phi-list for further inlinings -- but might be optimized */
845 if (get_nodes_Block(phi) == post_bl) {
846 set_irn_link(phi, get_irn_link(post_bl));
847 set_irn_link(post_bl, phi);
849 /* Now the real results */
851 for (j = 0; j < n_res; j++) {
853 for (i = 0; i < arity; i++) {
854 ret = intern_get_irn_n(end_bl, i);
855 if (intern_get_irn_op(ret) == op_Return) {
856 cf_pred[n_ret] = get_Return_res(ret, j);
860 phi = new_Phi(n_ret, cf_pred, intern_get_irn_mode(cf_pred[0]));
862 /* Conserve Phi-list for further inlinings -- but might be optimized */
863 if (get_nodes_Block(phi) == post_bl) {
864 set_irn_link(phi, get_irn_link(post_bl));
865 set_irn_link(post_bl, phi);
868 set_Tuple_pred(call, 2, new_Tuple(n_res, res_pred));
870 set_Tuple_pred(call, 2, new_Bad());
872 /* Finally the exception control flow.
873 We have two (three) possible situations:
874 First if the Call branches to an exception handler: We need to add a Phi node to
875 collect the memory containing the exception objects. Further we need
876 to add another block to get a correct representation of this Phi. To
877 this block we add a Jmp that resolves into the X output of the Call
878 when the Call is turned into a tuple.
879 Second the Call branches to End, the exception is not handled. Just
880 add all inlined exception branches to the End node.
881 Third: there is no Exception edge at all. Handle as case two. */
882 if (exc_handling == 0) {
884 for (i = 0; i < arity; i++) {
886 ret = intern_get_irn_n(end_bl, i);
887 if (is_fragile_op(skip_Proj(ret)) || (intern_get_irn_op(skip_Proj(ret)) == op_Raise)) {
888 cf_pred[n_exc] = ret;
893 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
894 set_Tuple_pred(call, 1, new_Jmp());
895 /* The Phi for the memories with the exception objects */
897 for (i = 0; i < arity; i++) {
899 ret = skip_Proj(intern_get_irn_n(end_bl, i));
900 if (intern_get_irn_op(ret) == op_Call) {
901 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_Block(ret), ret, mode_M, 3);
903 } else if (is_fragile_op(ret)) {
904 /* We rely that all cfops have the memory output at the same position. */
905 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_Block(ret), ret, mode_M, 0);
907 } else if (intern_get_irn_op(ret) == op_Raise) {
908 cf_pred[n_exc] = new_r_Proj(current_ir_graph, get_nodes_Block(ret), ret, mode_M, 1);
912 set_Tuple_pred(call, 3, new_Phi(n_exc, cf_pred, mode_M));
914 set_Tuple_pred(call, 1, new_Bad());
915 set_Tuple_pred(call, 3, new_Bad());
918 ir_node *main_end_bl;
919 int main_end_bl_arity;
922 /* assert(exc_handling == 1 || no exceptions. ) */
924 for (i = 0; i < arity; i++) {
925 ir_node *ret = intern_get_irn_n(end_bl, i);
927 if (is_fragile_op(skip_Proj(ret)) || (intern_get_irn_op(skip_Proj(ret)) == op_Raise)) {
928 cf_pred[n_exc] = ret;
932 main_end_bl = get_irg_end_block(current_ir_graph);
933 main_end_bl_arity = intern_get_irn_arity(main_end_bl);
934 end_preds = (ir_node **) malloc ((n_exc + main_end_bl_arity) * sizeof (ir_node *));
936 for (i = 0; i < main_end_bl_arity; ++i)
937 end_preds[i] = intern_get_irn_n(main_end_bl, i);
938 for (i = 0; i < n_exc; ++i)
939 end_preds[main_end_bl_arity + i] = cf_pred[i];
940 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
941 set_Tuple_pred(call, 1, new_Bad());
942 set_Tuple_pred(call, 3, new_Bad());
948 #if 0 /* old. now better, correcter, faster implementation. */
950 /* -- If the exception control flow from the inlined Call directly
951 branched to the end block we now have the following control
952 flow predecessor pattern: ProjX -> Tuple -> Jmp. We must
953 remove the Jmp along with it's empty block and add Jmp's
954 predecessors as predecessors of this end block. No problem if
955 there is no exception, because then branches Bad to End which
957 @@@ can't we know this beforehand: by getting the Proj(1) from
958 the Call link list and checking whether it goes to Proj. */
959 /* find the problematic predecessor of the end block. */
960 end_bl = get_irg_end_block(current_ir_graph);
961 for (i = 0; i < get_Block_n_cfgpreds(end_bl); i++) {
962 cf_op = get_Block_cfgpred(end_bl, i);
963 if (intern_get_irn_op(cf_op) == op_Proj) {
964 cf_op = get_Proj_pred(cf_op);
965 if ((intern_get_irn_op(cf_op) == op_Tuple) && (cf_op == call)) {
966 /* There are unoptimized tuples from inlineing before when no exc */
967 assert(get_Proj_proj(get_Block_cfgpred(end_bl, i)) == pn_Call_X_except);
968 cf_op = get_Tuple_pred(cf_op, pn_Call_X_except);
969 assert(intern_get_irn_op(cf_op) == op_Jmp);
975 if (i < get_Block_n_cfgpreds(end_bl)) {
976 bl = get_nodes_Block(cf_op);
977 arity = get_Block_n_cfgpreds(end_bl) + get_Block_n_cfgpreds(bl) - 1;
978 cf_pred = (ir_node **) malloc (arity * sizeof (ir_node *));
979 for (j = 0; j < i; j++)
980 cf_pred[j] = get_Block_cfgpred(end_bl, j);
981 for (j = j; j < i + get_Block_n_cfgpreds(bl); j++)
982 cf_pred[j] = get_Block_cfgpred(bl, j-i);
983 for (j = j; j < arity; j++)
984 cf_pred[j] = get_Block_cfgpred(end_bl, j-get_Block_n_cfgpreds(bl) +1);
985 set_irn_in(end_bl, arity, cf_pred);
987 /* Remove the exception pred from post-call Tuple. */
988 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
993 /* -- Turn cse back on. -- */
994 set_optimize(rem_opt);
997 /********************************************************************/
998 /* Apply inlineing to small methods. */
999 /********************************************************************/
1003 /* It makes no sense to inline too many calls in one procedure. Anyways,
1004 I didn't get a version with NEW_ARR_F to run. */
1005 #define MAX_INLINE 1024
1008 * Returns the irg called from a Call node. If the irg is not
1009 * known, NULL is returned.
1011 static ir_graph *get_call_called_irg(ir_node *call) {
1014 ir_graph *called_irg = NULL;
1016 assert(get_irn_op(call) == op_Call);
1018 addr = get_Call_ptr(call);
1019 if (intern_get_irn_op(addr) == op_Const) {
1020 /* Check whether the constant is the pointer to a compiled entity. */
1021 tv = get_Const_tarval(addr);
1022 if (tarval_to_entity(tv))
1023 called_irg = get_entity_irg(tarval_to_entity(tv));
1028 static void collect_calls(ir_node *call, void *env) {
1030 ir_node **calls = (ir_node **)env;
1033 ir_graph *called_irg;
1035 if (intern_get_irn_op(call) != op_Call) return;
1037 addr = get_Call_ptr(call);
1038 if (intern_get_irn_op(addr) == op_Const) {
1039 /* Check whether the constant is the pointer to a compiled entity. */
1040 tv = get_Const_tarval(addr);
1041 if (tarval_to_entity(tv)) {
1042 called_irg = get_entity_irg(tarval_to_entity(tv));
1043 if (called_irg && pos < MAX_INLINE) {
1044 /* The Call node calls a locally defined method. Remember to inline. */
1053 * Inlines all small methods at call sites where the called address comes
1054 * from a Const node that references the entity representing the called
1056 * The size argument is a rough measure for the code size of the method:
1057 * Methods where the obstack containing the firm graph is smaller than
1060 void inline_small_irgs(ir_graph *irg, int size) {
1062 ir_node *calls[MAX_INLINE];
1063 ir_graph *rem = current_ir_graph;
1065 if (!(get_opt_optimize() && get_opt_inline())) return;
1067 current_ir_graph = irg;
1068 /* Handle graph state */
1069 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1070 assert(get_irg_callee_info_state(current_ir_graph) == irg_callee_info_none);
1072 /* Find Call nodes to inline.
1073 (We can not inline during a walk of the graph, as inlineing the same
1074 method several times changes the visited flag of the walked graph:
1075 after the first inlineing visited of the callee equals visited of
1076 the caller. With the next inlineing both are increased.) */
1078 irg_walk(get_irg_end(irg), NULL, collect_calls, (void *) calls);
1080 if ((pos > 0) && (pos < MAX_INLINE)) {
1081 /* There are calls to inline */
1082 collect_phiprojs(irg);
1083 for (i = 0; i < pos; i++) {
1086 tv = get_Const_tarval(get_Call_ptr(calls[i]));
1087 callee = get_entity_irg(tarval_to_entity(tv));
1088 if (((_obstack_memory_used(callee->obst) - obstack_room(callee->obst)) < size) ||
1089 (get_irg_inline_property(callee) == irg_inline_forced)) {
1090 inline_method(calls[i], callee);
1095 current_ir_graph = rem;
1099 * Environment for inlining irgs.
1102 int n_nodes; /**< Nodes in graph except Id, Tuple, Proj, Start, End */
1103 int n_nodes_orig; /**< for statistics */
1104 eset *call_nodes; /**< All call nodes in this graph */
1106 int n_call_nodes_orig; /**< for statistics */
1107 int n_callers; /**< Number of known graphs that call this graphs. */
1108 int n_callers_orig; /**< for statistics */
1111 static inline_irg_env *new_inline_irg_env(void) {
1112 inline_irg_env *env = malloc(sizeof(inline_irg_env));
1113 env->n_nodes = -2; /* uncount Start, End */
1114 env->n_nodes_orig = -2; /* uncount Start, End */
1115 env->call_nodes = eset_create();
1116 env->n_call_nodes = 0;
1117 env->n_call_nodes_orig = 0;
1119 env->n_callers_orig = 0;
1123 static void free_inline_irg_env(inline_irg_env *env) {
1124 eset_destroy(env->call_nodes);
1128 static void collect_calls2(ir_node *call, void *env) {
1129 inline_irg_env *x = (inline_irg_env *)env;
1130 ir_op *op = intern_get_irn_op(call);
1133 /* count nodes in irg */
1134 if (op != op_Proj && op != op_Tuple && op != op_Sync) {
1139 if (op != op_Call) return;
1141 /* collect all call nodes */
1142 eset_insert(x->call_nodes, (void *)call);
1144 x->n_call_nodes_orig++;
1146 /* count all static callers */
1147 callee = get_call_called_irg(call);
1149 ((inline_irg_env *)get_irg_link(callee))->n_callers++;
1150 ((inline_irg_env *)get_irg_link(callee))->n_callers_orig++;
1154 INLINE static int is_leave(ir_graph *irg) {
1155 return (((inline_irg_env *)get_irg_link(irg))->n_call_nodes == 0);
1158 INLINE static int is_smaller(ir_graph *callee, int size) {
1159 return (((inline_irg_env *)get_irg_link(callee))->n_nodes < size);
1164 * Inlines small leave methods at call sites where the called address comes
1165 * from a Const node that references the entity representing the called
1167 * The size argument is a rough measure for the code size of the method:
1168 * Methods where the obstack containing the firm graph is smaller than
1171 void inline_leave_functions(int maxsize, int leavesize, int size) {
1172 inline_irg_env *env;
1173 int i, n_irgs = get_irp_n_irgs();
1174 ir_graph *rem = current_ir_graph;
1177 if (!(get_opt_optimize() && get_opt_inline())) return;
1179 /* extend all irgs by a temporary data structure for inlineing. */
1180 for (i = 0; i < n_irgs; ++i)
1181 set_irg_link(get_irp_irg(i), new_inline_irg_env());
1183 /* Precompute information in temporary data structure. */
1184 for (i = 0; i < n_irgs; ++i) {
1185 current_ir_graph = get_irp_irg(i);
1186 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1187 assert(get_irg_callee_info_state(current_ir_graph) == irg_callee_info_none);
1189 irg_walk(get_irg_end(current_ir_graph), NULL, collect_calls2,
1190 get_irg_link(current_ir_graph));
1194 Inline leaves recursively -- we might construct new leaves. */
1195 /* int itercnt = 1; */
1196 while (did_inline) {
1197 /* printf("iteration %d\n", itercnt++); */
1199 for (i = 0; i < n_irgs; ++i) {
1202 int phiproj_computed = 0;
1204 current_ir_graph = get_irp_irg(i);
1205 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1207 /* we can not walk and change a set, nor remove from it.
1209 walkset = env->call_nodes;
1210 env->call_nodes = eset_create();
1211 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1212 inline_irg_env *callee_env;
1213 ir_graph *callee = get_call_called_irg(call);
1215 if (env->n_nodes > maxsize) break;
1217 ((is_leave(callee) && is_smaller(callee, leavesize)) ||
1218 (get_irg_inline_property(callee) == irg_inline_forced))) {
1219 if (!phiproj_computed) {
1220 phiproj_computed = 1;
1221 collect_phiprojs(current_ir_graph);
1223 callee_env = (inline_irg_env *)get_irg_link(callee);
1224 /* printf(" %s: Inlineing %s.\n", get_entity_name(get_irg_entity(current_ir_graph)), */
1225 /* get_entity_name(get_irg_entity(callee))); */
1226 inline_method(call, callee);
1228 env->n_call_nodes--;
1229 eset_insert_all(env->call_nodes, callee_env->call_nodes);
1230 env->n_call_nodes += callee_env->n_call_nodes;
1231 env->n_nodes += callee_env->n_nodes;
1232 callee_env->n_callers--;
1234 eset_insert(env->call_nodes, call);
1237 eset_destroy(walkset);
1241 /* printf("Non leaves\n"); */
1242 /* inline other small functions. */
1243 for (i = 0; i < n_irgs; ++i) {
1246 int phiproj_computed = 0;
1248 current_ir_graph = get_irp_irg(i);
1249 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1251 /* we can not walk and change a set, nor remove from it.
1253 walkset = env->call_nodes;
1254 env->call_nodes = eset_create();
1255 for (call = eset_first(walkset); call; call = eset_next(walkset)) {
1256 inline_irg_env *callee_env;
1257 ir_graph *callee = get_call_called_irg(call);
1259 if (env->n_nodes > maxsize) break;
1260 if (callee && is_smaller(callee, size)) {
1261 if (!phiproj_computed) {
1262 phiproj_computed = 1;
1263 collect_phiprojs(current_ir_graph);
1265 callee_env = (inline_irg_env *)get_irg_link(callee);
1266 /* printf(" %s: Inlineing %s.\n", get_entity_name(get_irg_entity(current_ir_graph)), */
1267 /* get_entity_name(get_irg_entity(callee))); */
1268 inline_method(call, callee);
1270 env->n_call_nodes--;
1271 eset_insert_all(env->call_nodes, callee_env->call_nodes);
1272 env->n_call_nodes += callee_env->n_call_nodes;
1273 env->n_nodes += callee_env->n_nodes;
1274 callee_env->n_callers--;
1276 eset_insert(env->call_nodes, call);
1279 eset_destroy(walkset);
1282 for (i = 0; i < n_irgs; ++i) {
1283 current_ir_graph = get_irp_irg(i);
1285 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1286 if ((env->n_call_nodes_orig != env->n_call_nodes) ||
1287 (env->n_callers_orig != env->n_callers))
1288 printf("Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1289 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1290 env->n_callers_orig, env->n_callers,
1291 get_entity_name(get_irg_entity(current_ir_graph)));
1293 free_inline_irg_env((inline_irg_env *)get_irg_link(current_ir_graph));
1296 current_ir_graph = rem;
1299 /*-----------------------------------------------------------------*/
1300 /* Code Placement. Pins all floating nodes to a block where they */
1301 /* will be executed only if needed. */
1302 /*-----------------------------------------------------------------*/
1305 * Find the earliest correct block for N. --- Place N into the
1306 * same Block as its dominance-deepest Input.
1309 place_floats_early(ir_node *n, pdeq *worklist)
1311 int i, start, irn_arity;
1313 /* we must not run into an infinite loop */
1314 assert (irn_not_visited(n));
1315 mark_irn_visited(n);
1317 /* Place floating nodes. */
1318 if (get_op_pinned(intern_get_irn_op(n)) == floats) {
1320 ir_node *b = new_Bad(); /* The block to place this node in */
1322 assert(intern_get_irn_op(n) != op_Block);
1324 if ((intern_get_irn_op(n) == op_Const) ||
1325 (intern_get_irn_op(n) == op_SymConst) ||
1327 (intern_get_irn_op(n) == op_Unknown)) {
1328 /* These nodes will not be placed by the loop below. */
1329 b = get_irg_start_block(current_ir_graph);
1333 /* find the block for this node. */
1334 irn_arity = intern_get_irn_arity(n);
1335 for (i = 0; i < irn_arity; i++) {
1336 ir_node *dep = intern_get_irn_n(n, i);
1338 if ((irn_not_visited(dep)) &&
1339 (get_op_pinned(intern_get_irn_op(dep)) == floats)) {
1340 place_floats_early(dep, worklist);
1342 /* Because all loops contain at least one pinned node, now all
1343 our inputs are either pinned or place_early has already
1344 been finished on them. We do not have any unfinished inputs! */
1345 dep_block = get_nodes_Block(dep);
1346 if ((!is_Bad(dep_block)) &&
1347 (get_Block_dom_depth(dep_block) > depth)) {
1349 depth = get_Block_dom_depth(dep_block);
1351 /* Avoid that the node is placed in the Start block */
1352 if ((depth == 1) && (get_Block_dom_depth(get_nodes_Block(n)) > 1)) {
1353 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
1354 assert(b != get_irg_start_block(current_ir_graph));
1358 set_nodes_Block(n, b);
1361 /* Add predecessors of non floating nodes on worklist. */
1362 start = (intern_get_irn_op(n) == op_Block) ? 0 : -1;
1363 irn_arity = intern_get_irn_arity(n);
1364 for (i = start; i < irn_arity; i++) {
1365 ir_node *pred = intern_get_irn_n(n, i);
1366 if (irn_not_visited(pred)) {
1367 pdeq_putr (worklist, pred);
1373 * Floating nodes form subgraphs that begin at nodes as Const, Load,
1374 * Start, Call and end at pinned nodes as Store, Call. Place_early
1375 * places all floating nodes reachable from its argument through floating
1376 * nodes and adds all beginnings at pinned nodes to the worklist.
1378 static INLINE void place_early(pdeq* worklist) {
1380 inc_irg_visited(current_ir_graph);
1382 /* this inits the worklist */
1383 place_floats_early(get_irg_end(current_ir_graph), worklist);
1385 /* Work the content of the worklist. */
1386 while (!pdeq_empty (worklist)) {
1387 ir_node *n = pdeq_getl (worklist);
1388 if (irn_not_visited(n)) place_floats_early(n, worklist);
1391 set_irg_outs_inconsistent(current_ir_graph);
1392 current_ir_graph->pinned = pinned;
1396 /** deepest common dominance ancestor of DCA and CONSUMER of PRODUCER. */
1398 consumer_dom_dca (ir_node *dca, ir_node *consumer, ir_node *producer)
1400 ir_node *block = NULL;
1402 /* Compute the latest block into which we can place a node so that it is
1404 if (intern_get_irn_op(consumer) == op_Phi) {
1405 /* our consumer is a Phi-node, the effective use is in all those
1406 blocks through which the Phi-node reaches producer */
1408 ir_node *phi_block = get_nodes_Block(consumer);
1409 irn_arity = intern_get_irn_arity(consumer);
1410 for (i = 0; i < irn_arity; i++) {
1411 if (intern_get_irn_n(consumer, i) == producer) {
1412 block = get_nodes_Block(get_Block_cfgpred(phi_block, i));
1416 assert(is_no_Block(consumer));
1417 block = get_nodes_Block(consumer);
1420 /* Compute the deepest common ancestor of block and dca. */
1422 if (!dca) return block;
1423 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
1424 block = get_Block_idom(block);
1425 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block))
1426 dca = get_Block_idom(dca);
1427 while (block != dca)
1428 { block = get_Block_idom(block); dca = get_Block_idom(dca); }
1433 static INLINE int get_irn_loop_depth(ir_node *n) {
1434 return get_loop_depth(get_irn_loop(n));
1438 * Move n to a block with less loop depth than it's current block. The
1439 * new block must be dominated by early.
1442 move_out_of_loops (ir_node *n, ir_node *early)
1444 ir_node *best, *dca;
1448 /* Find the region deepest in the dominator tree dominating
1449 dca with the least loop nesting depth, but still dominated
1450 by our early placement. */
1451 dca = get_nodes_Block(n);
1453 while (dca != early) {
1454 dca = get_Block_idom(dca);
1455 if (!dca) break; /* should we put assert(dca)? */
1456 if (get_irn_loop_depth(dca) < get_irn_loop_depth(best)) {
1460 if (best != get_nodes_Block(n)) {
1462 printf("Moving out of loop: "); DDMN(n);
1463 printf(" Outermost block: "); DDMN(early);
1464 printf(" Best block: "); DDMN(best);
1465 printf(" Innermost block: "); DDMN(get_nodes_Block(n));
1467 set_nodes_Block(n, best);
1472 * Find the latest legal block for N and place N into the
1473 * `optimal' Block between the latest and earliest legal block.
1474 * The `optimal' block is the dominance-deepest block of those
1475 * with the least loop-nesting-depth. This places N out of as many
1476 * loops as possible and then makes it as control dependant as
1480 place_floats_late(ir_node *n, pdeq *worklist)
1485 assert (irn_not_visited(n)); /* no multiple placement */
1487 /* no need to place block nodes, control nodes are already placed. */
1488 if ((intern_get_irn_op(n) != op_Block) &&
1490 (intern_get_irn_mode(n) != mode_X)) {
1491 /* Remember the early placement of this block to move it
1492 out of loop no further than the early placement. */
1493 early = get_nodes_Block(n);
1494 /* Assure that our users are all placed, except the Phi-nodes.
1495 --- Each data flow cycle contains at least one Phi-node. We
1496 have to break the `user has to be placed before the
1497 producer' dependence cycle and the Phi-nodes are the
1498 place to do so, because we need to base our placement on the
1499 final region of our users, which is OK with Phi-nodes, as they
1500 are pinned, and they never have to be placed after a
1501 producer of one of their inputs in the same block anyway. */
1502 for (i = 0; i < get_irn_n_outs(n); i++) {
1503 ir_node *succ = get_irn_out(n, i);
1504 if (irn_not_visited(succ) && (intern_get_irn_op(succ) != op_Phi))
1505 place_floats_late(succ, worklist);
1508 /* We have to determine the final block of this node... except for
1510 if ((get_op_pinned(intern_get_irn_op(n)) == floats) &&
1511 (intern_get_irn_op(n) != op_Const) &&
1512 (intern_get_irn_op(n) != op_SymConst)) {
1513 ir_node *dca = NULL; /* deepest common ancestor in the
1514 dominator tree of all nodes'
1515 blocks depending on us; our final
1516 placement has to dominate DCA. */
1517 for (i = 0; i < get_irn_n_outs(n); i++) {
1518 dca = consumer_dom_dca (dca, get_irn_out(n, i), n);
1520 set_nodes_Block(n, dca);
1522 move_out_of_loops (n, early);
1526 mark_irn_visited(n);
1528 /* Add predecessors of all non-floating nodes on list. (Those of floating
1529 nodes are placeded already and therefore are marked.) */
1530 for (i = 0; i < get_irn_n_outs(n); i++) {
1531 if (irn_not_visited(get_irn_out(n, i))) {
1532 pdeq_putr (worklist, get_irn_out(n, i));
1537 static INLINE void place_late(pdeq* worklist) {
1539 inc_irg_visited(current_ir_graph);
1541 /* This fills the worklist initially. */
1542 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
1543 /* And now empty the worklist again... */
1544 while (!pdeq_empty (worklist)) {
1545 ir_node *n = pdeq_getl (worklist);
1546 if (irn_not_visited(n)) place_floats_late(n, worklist);
1550 void place_code(ir_graph *irg) {
1552 ir_graph *rem = current_ir_graph;
1554 current_ir_graph = irg;
1556 if (!(get_opt_optimize() && get_opt_global_cse())) return;
1558 /* Handle graph state */
1559 assert(get_irg_phase_state(irg) != phase_building);
1560 if (get_irg_dom_state(irg) != dom_consistent)
1563 if (get_irg_loopinfo_state(irg) != loopinfo_consistent) {
1564 free_loop_information(irg);
1565 construct_backedges(irg);
1568 /* Place all floating nodes as early as possible. This guarantees
1569 a legal code placement. */
1570 worklist = new_pdeq();
1571 place_early(worklist);
1573 /* place_early invalidates the outs, place_late needs them. */
1575 /* Now move the nodes down in the dominator tree. This reduces the
1576 unnecessary executions of the node. */
1577 place_late(worklist);
1579 set_irg_outs_inconsistent(current_ir_graph);
1580 set_irg_loopinfo_inconsistent(current_ir_graph);
1582 current_ir_graph = rem;
1587 /********************************************************************/
1588 /* Control flow optimization. */
1589 /* Removes Bad control flow predecessors and empty blocks. A block */
1590 /* is empty if it contains only a Jmp node. */
1591 /* Blocks can only be removed if they are not needed for the */
1592 /* semantics of Phi nodes. */
1593 /********************************************************************/
1596 * Removes Tuples from Block control flow predecessors.
1597 * Optimizes blocks with equivalent_node().
1598 * Replaces n by Bad if n is unreachable control flow.
1600 static void merge_blocks(ir_node *n, void *env) {
1602 set_irn_link(n, NULL);
1604 if (intern_get_irn_op(n) == op_Block) {
1606 for (i = 0; i < get_Block_n_cfgpreds(n); i++)
1607 /* GL @@@ : is this possible? if (get_opt_normalize()) -- added, all tests go through.
1608 A different order of optimizations might cause problems. */
1609 if (get_opt_normalize())
1610 set_Block_cfgpred(n, i, skip_Tuple(get_Block_cfgpred(n, i)));
1611 } else if (get_opt_optimize() && (intern_get_irn_mode(n) == mode_X)) {
1612 /* We will soon visit a block. Optimize it before visiting! */
1613 ir_node *b = get_nodes_Block(n);
1614 ir_node *new_node = equivalent_node(b);
1615 while (irn_not_visited(b) && (!is_Bad(new_node)) && (new_node != b)) {
1616 /* We would have to run gigo if new is bad, so we
1617 promote it directly below. */
1618 assert(((b == new_node) ||
1619 get_opt_control_flow_straightening() ||
1620 get_opt_control_flow_weak_simplification()) &&
1621 ("strange flag setting"));
1622 exchange (b, new_node);
1624 new_node = equivalent_node(b);
1626 /* GL @@@ get_opt_normalize hinzugefuegt, 5.5.2003 */
1627 if (is_Bad(new_node) && get_opt_normalize()) exchange(n, new_Bad());
1632 * Collects all Phi nodes in link list of Block.
1633 * Marks all blocks "block_visited" if they contain a node other
1636 static void collect_nodes(ir_node *n, void *env) {
1637 if (is_no_Block(n)) {
1638 ir_node *b = get_nodes_Block(n);
1640 if ((intern_get_irn_op(n) == op_Phi)) {
1641 /* Collect Phi nodes to compact ins along with block's ins. */
1642 set_irn_link(n, get_irn_link(b));
1644 } else if (intern_get_irn_op(n) != op_Jmp) { /* Check for non empty block. */
1645 mark_Block_block_visited(b);
1650 /** Returns true if pred is predecessor of block. */
1651 static int is_pred_of(ir_node *pred, ir_node *b) {
1653 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1654 ir_node *b_pred = get_nodes_Block(get_Block_cfgpred(b, i));
1655 if (b_pred == pred) return 1;
1660 static int test_whether_dispensable(ir_node *b, int pos) {
1661 int i, j, n_preds = 1;
1662 int dispensable = 1;
1663 ir_node *cfop = get_Block_cfgpred(b, pos);
1664 ir_node *pred = get_nodes_Block(cfop);
1666 if (get_Block_block_visited(pred) + 1
1667 < get_irg_block_visited(current_ir_graph)) {
1668 if (!get_opt_optimize() || !get_opt_control_flow_strong_simplification()) {
1669 /* Mark block so that is will not be removed. */
1670 set_Block_block_visited(pred, get_irg_block_visited(current_ir_graph)-1);
1673 /* Seems to be empty. */
1674 if (!get_irn_link(b)) {
1675 /* There are no Phi nodes ==> dispensable. */
1676 n_preds = get_Block_n_cfgpreds(pred);
1678 /* b's pred blocks and pred's pred blocks must be pairwise disjunct.
1679 Work preds < pos as if they were already removed. */
1680 for (i = 0; i < pos; i++) {
1681 ir_node *b_pred = get_nodes_Block(get_Block_cfgpred(b, i));
1682 if (get_Block_block_visited(b_pred) + 1
1683 < get_irg_block_visited(current_ir_graph)) {
1684 for (j = 0; j < get_Block_n_cfgpreds(b_pred); j++) {
1685 ir_node *b_pred_pred = get_nodes_Block(get_Block_cfgpred(b_pred, j));
1686 if (is_pred_of(b_pred_pred, pred)) dispensable = 0;
1689 if (is_pred_of(b_pred, pred)) dispensable = 0;
1692 for (i = pos +1; i < get_Block_n_cfgpreds(b); i++) {
1693 ir_node *b_pred = get_nodes_Block(get_Block_cfgpred(b, i));
1694 if (is_pred_of(b_pred, pred)) dispensable = 0;
1697 set_Block_block_visited(pred, get_irg_block_visited(current_ir_graph)-1);
1700 n_preds = get_Block_n_cfgpreds(pred);
1708 static void optimize_blocks(ir_node *b, void *env) {
1709 int i, j, k, max_preds, n_preds;
1710 ir_node *pred, *phi;
1713 /* Count the number of predecessor if this block is merged with pred blocks
1716 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1717 max_preds += test_whether_dispensable(b, i);
1719 in = (ir_node **) malloc(max_preds * sizeof(ir_node *));
1722 printf(" working on "); DDMN(b);
1723 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1724 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1725 if (is_Bad(get_Block_cfgpred(b, i))) {
1726 printf(" removing Bad %i\n ", i);
1727 } else if (get_Block_block_visited(pred) +1
1728 < get_irg_block_visited(current_ir_graph)) {
1729 printf(" removing pred %i ", i); DDMN(pred);
1730 } else { printf(" Nothing to do for "); DDMN(pred); }
1732 * end Debug output -*/
1734 /*- Fix the Phi nodes -*/
1735 phi = get_irn_link(b);
1737 assert(intern_get_irn_op(phi) == op_Phi);
1738 /* Find the new predecessors for the Phi */
1740 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1741 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1742 if (is_Bad(get_Block_cfgpred(b, i))) {
1744 } else if (get_Block_block_visited(pred) +1
1745 < get_irg_block_visited(current_ir_graph)) {
1746 /* It's an empty block and not yet visited. */
1747 ir_node *phi_pred = get_Phi_pred(phi, i);
1748 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1749 if (get_nodes_Block(phi_pred) == pred) {
1750 assert(intern_get_irn_op(phi_pred) == op_Phi); /* Block is empty!! */
1751 in[n_preds] = get_Phi_pred(phi_pred, j);
1753 in[n_preds] = phi_pred;
1757 /* The Phi_pred node is replaced now if it is a Phi.
1758 In Schleifen kann offenbar der entfernte Phi Knoten legal verwendet werden.
1759 Daher muss der Phiknoten durch den neuen ersetzt werden.
1760 Weiter muss der alte Phiknoten entfernt werden (durch ersetzen oder
1761 durch einen Bad) damit er aus den keep_alive verschwinden kann.
1762 Man sollte also, falls keine Schleife vorliegt, exchange mit new_Bad
1764 if (get_nodes_Block(phi_pred) == pred) {
1765 /* remove the Phi as it might be kept alive. Further there
1766 might be other users. */
1767 exchange(phi_pred, phi); /* geht, ist aber doch semantisch falsch! Warum?? */
1770 in[n_preds] = get_Phi_pred(phi, i);
1775 set_irn_in(phi, n_preds, in);
1777 phi = get_irn_link(phi);
1781 This happens only if merge between loop backedge and single loop entry. -*/
1782 for (k = 0; k < get_Block_n_cfgpreds(b); k++) {
1783 pred = get_nodes_Block(get_Block_cfgpred(b, k));
1784 if (get_Block_block_visited(pred) +1
1785 < get_irg_block_visited(current_ir_graph)) {
1786 phi = get_irn_link(pred);
1788 if (intern_get_irn_op(phi) == op_Phi) {
1789 set_nodes_Block(phi, b);
1792 for (i = 0; i < k; i++) {
1793 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1794 if (is_Bad(get_Block_cfgpred(b, i))) {
1796 } else if (get_Block_block_visited(pred) +1
1797 < get_irg_block_visited(current_ir_graph)) {
1798 /* It's an empty block and not yet visited. */
1799 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1800 /* @@@ Hier brauche ich Schleifeninformation!!! Kontrollflusskante
1801 muss Rueckwaertskante sein! (An allen vier in[n_preds] = phi
1802 Anweisungen.) Trotzdem tuts bisher!! */
1811 for (i = 0; i < get_Phi_n_preds(phi); i++) {
1812 in[n_preds] = get_Phi_pred(phi, i);
1815 for (i = k+1; i < get_Block_n_cfgpreds(b); i++) {
1816 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1817 if (is_Bad(get_Block_cfgpred(b, i))) {
1819 } else if (get_Block_block_visited(pred) +1
1820 < get_irg_block_visited(current_ir_graph)) {
1821 /* It's an empty block and not yet visited. */
1822 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1831 set_irn_in(phi, n_preds, in);
1833 phi = get_irn_link(phi);
1838 /*- Fix the block -*/
1840 for (i = 0; i < get_Block_n_cfgpreds(b); i++) {
1841 pred = get_nodes_Block(get_Block_cfgpred(b, i));
1842 if (is_Bad(get_Block_cfgpred(b, i))) {
1844 } else if (get_Block_block_visited(pred) +1
1845 < get_irg_block_visited(current_ir_graph)) {
1846 /* It's an empty block and not yet visited. */
1847 assert(get_Block_n_cfgpreds(b) > 1);
1848 /* Else it should be optimized by equivalent_node. */
1849 for (j = 0; j < get_Block_n_cfgpreds(pred); j++) {
1850 in[n_preds] = get_Block_cfgpred(pred, j);
1853 /* Remove block as it might be kept alive. */
1854 exchange(pred, b/*new_Bad()*/);
1856 in[n_preds] = get_Block_cfgpred(b, i);
1860 set_irn_in(b, n_preds, in);
1864 void optimize_cf(ir_graph *irg) {
1867 ir_node *end = get_irg_end(irg);
1868 ir_graph *rem = current_ir_graph;
1869 current_ir_graph = irg;
1871 /* Handle graph state */
1872 assert(get_irg_phase_state(irg) != phase_building);
1873 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
1874 set_irg_outs_inconsistent(current_ir_graph);
1875 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
1876 set_irg_dom_inconsistent(current_ir_graph);
1878 /* Use block visited flag to mark non-empty blocks. */
1879 inc_irg_block_visited(irg);
1880 irg_walk(end, merge_blocks, collect_nodes, NULL);
1882 /* Optimize the standard code. */
1883 irg_block_walk(get_irg_end_block(irg), optimize_blocks, NULL, NULL);
1885 /* Walk all keep alives, optimize them if block, add to new in-array
1886 for end if useful. */
1887 in = NEW_ARR_F (ir_node *, 1);
1888 in[0] = get_nodes_Block(end);
1889 inc_irg_visited(current_ir_graph);
1890 for(i = 0; i < get_End_n_keepalives(end); i++) {
1891 ir_node *ka = get_End_keepalive(end, i);
1892 if (irn_not_visited(ka)) {
1893 if ((intern_get_irn_op(ka) == op_Block) && Block_not_block_visited(ka)) {
1894 set_irg_block_visited(current_ir_graph, /* Don't walk all the way to Start. */
1895 get_irg_block_visited(current_ir_graph)-1);
1896 irg_block_walk(ka, optimize_blocks, NULL, NULL);
1897 mark_irn_visited(ka);
1898 ARR_APP1 (ir_node *, in, ka);
1899 } else if (intern_get_irn_op(ka) == op_Phi) {
1900 mark_irn_visited(ka);
1901 ARR_APP1 (ir_node *, in, ka);
1905 /* DEL_ARR_F(end->in); GL @@@ tut nicht ! */
1908 current_ir_graph = rem;
1913 * Called by walker of remove_critical_cf_edges.
1915 * Place an empty block to an edge between a blocks of multiple
1916 * predecessors and a block of multiple successors.
1919 * @param env Environment of walker. This field is unused and has
1922 static void walk_critical_cf_edges(ir_node *n, void *env) {
1924 ir_node *pre, *block, **in, *jmp;
1926 /* Block has multiple predecessors */
1927 if ((op_Block == intern_get_irn_op(n)) &&
1928 (intern_get_irn_arity(n) > 1)) {
1929 arity = intern_get_irn_arity(n);
1931 if (n == get_irg_end_block(current_ir_graph))
1932 return; /* No use to add a block here. */
1934 for (i=0; i<arity; i++) {
1935 pre = intern_get_irn_n(n, i);
1936 /* Predecessor has multiple successors. Insert new flow edge */
1937 if ((NULL != pre) &&
1938 (op_Proj == intern_get_irn_op(pre)) &&
1939 op_Raise != intern_get_irn_op(skip_Proj(pre))) {
1941 /* set predecessor array for new block */
1942 in = NEW_ARR_D (ir_node *, current_ir_graph->obst, 1);
1943 /* set predecessor of new block */
1945 block = new_Block(1, in);
1946 /* insert new jmp node to new block */
1947 switch_block(block);
1950 /* set successor of new block */
1951 set_irn_n(n, i, jmp);
1953 } /* predecessor has multiple successors */
1954 } /* for all predecessors */
1955 } /* n is a block */
1958 void remove_critical_cf_edges(ir_graph *irg) {
1959 if (get_opt_critical_edges())
1960 irg_walk_graph(irg, NULL, walk_critical_cf_edges, NULL);