2 * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Dead node elimination and Procedure Inlining.
23 * @author Michael Beck, Goetz Lindenmaier
34 #include "irgraph_t.h"
37 #include "iroptimize.h"
44 #include "adt/array.h"
48 #include "adt/xmalloc.h"
52 #include "irbackedge_t.h"
53 #include "opt_inline_t.h"
58 #include "analyze_irg_args.h"
59 #include "iredges_t.h"
64 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
66 /*------------------------------------------------------------------*/
67 /* Routines for dead node elimination / copying garbage collection */
69 /*------------------------------------------------------------------*/
72 * Remember the new node in the old node by using a field all nodes have.
74 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
77 * Get this new node, before the old node is forgotten.
79 #define get_new_node(oldn) get_irn_link(oldn)
82 * Check if a new node was set.
84 #define has_new_node(n) (get_new_node(n) != NULL)
87 * We use the block_visited flag to mark that we have computed the
88 * number of useful predecessors for this block.
89 * Further we encode the new arity in this flag in the old blocks.
90 * Remembering the arity is useful, as it saves a lot of pointer
91 * accesses. This function is called for all Phi and Block nodes
95 compute_new_arity(ir_node *b) {
96 int i, res, irn_arity;
99 irg_v = get_irg_block_visited(current_ir_graph);
100 block_v = get_Block_block_visited(b);
101 if (block_v >= irg_v) {
102 /* we computed the number of preds for this block and saved it in the
104 return block_v - irg_v;
106 /* compute the number of good predecessors */
107 res = irn_arity = get_irn_arity(b);
108 for (i = 0; i < irn_arity; i++)
109 if (is_Bad(get_irn_n(b, i))) res--;
110 /* save it in the flag. */
111 set_Block_block_visited(b, irg_v + res);
117 * Copies the node to the new obstack. The Ins of the new node point to
118 * the predecessors on the old obstack. For block/phi nodes not all
119 * predecessors might be copied. n->link points to the new node.
120 * For Phi and Block nodes the function allocates in-arrays with an arity
121 * only for useful predecessors. The arity is determined by counting
122 * the non-bad predecessors of the block.
124 * @param n The node to be copied
125 * @param env if non-NULL, the node number attribute will be copied to the new node
127 * Note: Also used for loop unrolling.
129 static void copy_node(ir_node *n, void *env) {
132 ir_op *op = get_irn_op(n);
136 /* node copied already */
138 } else if (op == op_Block) {
140 new_arity = compute_new_arity(n);
141 n->attr.block.graph_arr = NULL;
143 block = get_nodes_block(n);
145 new_arity = compute_new_arity(block);
147 new_arity = get_irn_arity(n);
150 nn = new_ir_node(get_irn_dbg_info(n),
157 /* Copy the attributes. These might point to additional data. If this
158 was allocated on the old obstack the pointers now are dangling. This
159 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
160 if (op == op_Block) {
161 /* we cannot allow blocks WITHOUT macroblock input */
162 set_Block_MacroBlock(nn, get_Block_MacroBlock(n));
164 copy_node_attr(n, nn);
168 int copy_node_nr = env != NULL;
170 /* for easier debugging, we want to copy the node numbers too */
171 nn->node_nr = n->node_nr;
177 hook_dead_node_elim_subst(current_ir_graph, n, nn);
181 * Copies new predecessors of old node to new node remembered in link.
182 * Spare the Bad predecessors of Phi and Block nodes.
184 static void copy_preds(ir_node *n, void *env) {
189 nn = get_new_node(n);
192 /* copy the macro block header */
193 ir_node *mbh = get_Block_MacroBlock(n);
196 /* this block is a macroblock header */
197 set_Block_MacroBlock(nn, nn);
199 /* get the macro block header */
200 ir_node *nmbh = get_new_node(mbh);
201 assert(nmbh != NULL);
202 set_Block_MacroBlock(nn, nmbh);
205 /* Don't copy Bad nodes. */
207 irn_arity = get_irn_arity(n);
208 for (i = 0; i < irn_arity; i++) {
209 if (! is_Bad(get_irn_n(n, i))) {
210 ir_node *pred = get_irn_n(n, i);
211 set_irn_n(nn, j, get_new_node(pred));
215 /* repair the block visited flag from above misuse. Repair it in both
216 graphs so that the old one can still be used. */
217 set_Block_block_visited(nn, 0);
218 set_Block_block_visited(n, 0);
219 /* Local optimization could not merge two subsequent blocks if
220 in array contained Bads. Now it's possible.
221 We don't call optimize_in_place as it requires
222 that the fields in ir_graph are set properly. */
223 if ((get_opt_control_flow_straightening()) &&
224 (get_Block_n_cfgpreds(nn) == 1) &&
225 is_Jmp(get_Block_cfgpred(nn, 0))) {
226 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
228 /* Jmp jumps into the block it is in -- deal self cycle. */
229 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
230 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
235 } else if (is_Phi(n) && get_irn_arity(n) > 0) {
236 /* Don't copy node if corresponding predecessor in block is Bad.
237 The Block itself should not be Bad. */
238 block = get_nodes_block(n);
239 set_nodes_block(nn, get_new_node(block));
241 irn_arity = get_irn_arity(n);
242 for (i = 0; i < irn_arity; i++) {
243 if (! is_Bad(get_irn_n(block, i))) {
244 ir_node *pred = get_irn_n(n, i);
245 set_irn_n(nn, j, get_new_node(pred));
246 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
250 /* If the pre walker reached this Phi after the post walker visited the
251 block block_visited is > 0. */
252 set_Block_block_visited(get_nodes_block(n), 0);
253 /* Compacting the Phi's ins might generate Phis with only one
255 if (get_irn_arity(nn) == 1)
256 exchange(nn, get_irn_n(nn, 0));
258 irn_arity = get_irn_arity(n);
259 for (i = -1; i < irn_arity; i++)
260 set_irn_n(nn, i, get_new_node(get_irn_n(n, i)));
262 /* Now the new node is complete. We can add it to the hash table for CSE.
263 @@@ inlining aborts if we identify End. Why? */
265 add_identities(current_ir_graph->value_table, nn);
269 * Copies the graph recursively, compacts the keep-alives of the end node.
271 * @param irg the graph to be copied
272 * @param copy_node_nr If non-zero, the node number will be copied
274 static void copy_graph(ir_graph *irg, int copy_node_nr) {
275 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
276 ir_node *ka; /* keep alive */
280 /* Some nodes must be copied by hand, sigh */
281 vfl = get_irg_visited(irg);
282 set_irg_visited(irg, vfl + 1);
284 oe = get_irg_end(irg);
285 mark_irn_visited(oe);
286 /* copy the end node by hand, allocate dynamic in array! */
287 ne = new_ir_node(get_irn_dbg_info(oe),
294 /* Copy the attributes. Well, there might be some in the future... */
295 copy_node_attr(oe, ne);
296 set_new_node(oe, ne);
298 /* copy the Bad node */
299 ob = get_irg_bad(irg);
300 mark_irn_visited(ob);
301 nb = new_ir_node(get_irn_dbg_info(ob),
308 copy_node_attr(ob, nb);
309 set_new_node(ob, nb);
311 /* copy the NoMem node */
312 om = get_irg_no_mem(irg);
313 mark_irn_visited(om);
314 nm = new_ir_node(get_irn_dbg_info(om),
321 copy_node_attr(om, nm);
322 set_new_node(om, nm);
324 /* copy the live nodes */
325 set_irg_visited(irg, vfl);
326 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
328 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
330 /* visit the anchors as well */
331 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
332 ir_node *n = get_irg_anchor(irg, i);
334 if (n && (get_irn_visited(n) <= vfl)) {
335 set_irg_visited(irg, vfl);
336 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
340 /* copy_preds for the end node ... */
341 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
343 /*- ... and now the keep alives. -*/
344 /* First pick the not marked block nodes and walk them. We must pick these
345 first as else we will oversee blocks reachable from Phis. */
346 irn_arity = get_End_n_keepalives(oe);
347 for (i = 0; i < irn_arity; i++) {
348 ka = get_End_keepalive(oe, i);
350 if (get_irn_visited(ka) <= vfl) {
351 /* We must keep the block alive and copy everything reachable */
352 set_irg_visited(irg, vfl);
353 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
355 add_End_keepalive(ne, get_new_node(ka));
359 /* Now pick other nodes. Here we will keep all! */
360 irn_arity = get_End_n_keepalives(oe);
361 for (i = 0; i < irn_arity; i++) {
362 ka = get_End_keepalive(oe, i);
364 if (get_irn_visited(ka) <= vfl) {
365 /* We didn't copy the node yet. */
366 set_irg_visited(irg, vfl);
367 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
369 add_End_keepalive(ne, get_new_node(ka));
373 /* start block sometimes only reached after keep alives */
374 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
375 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
379 * Copies the graph reachable from current_ir_graph->end to the obstack
380 * in current_ir_graph and fixes the environment.
381 * Then fixes the fields in current_ir_graph containing nodes of the
384 * @param copy_node_nr If non-zero, the node number will be copied
387 copy_graph_env(int copy_node_nr) {
388 ir_graph *irg = current_ir_graph;
389 ir_node *old_end, *new_anchor;
392 /* remove end_except and end_reg nodes */
393 old_end = get_irg_end(irg);
394 set_irg_end_except (irg, old_end);
395 set_irg_end_reg (irg, old_end);
397 /* Not all nodes remembered in irg might be reachable
398 from the end node. Assure their link is set to NULL, so that
399 we can test whether new nodes have been computed. */
400 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
401 ir_node *n = get_irg_anchor(irg, i);
403 set_new_node(n, NULL);
405 /* we use the block walk flag for removing Bads from Blocks ins. */
406 inc_irg_block_visited(irg);
409 copy_graph(irg, copy_node_nr);
412 old_end = get_irg_end(irg);
413 new_anchor = new_Anchor(irg);
415 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
416 ir_node *n = get_irg_anchor(irg, i);
418 set_irn_n(new_anchor, i, get_new_node(n));
421 irg->anchor = new_anchor;
423 /* ensure the new anchor is placed in the endblock */
424 set_nodes_block(new_anchor, get_irg_end_block(irg));
428 * Copies all reachable nodes to a new obstack. Removes bad inputs
429 * from block nodes and the corresponding inputs from Phi nodes.
430 * Merges single exit blocks with single entry blocks and removes
432 * Adds all new nodes to a new hash table for CSE. Does not
433 * perform CSE, so the hash table might contain common subexpressions.
435 void dead_node_elimination(ir_graph *irg) {
437 #ifdef INTERPROCEDURAL_VIEW
438 int rem_ipview = get_interprocedural_view();
440 struct obstack *graveyard_obst = NULL;
441 struct obstack *rebirth_obst = NULL;
442 assert(! edges_activated(irg) && "dead node elimination requires disabled edges");
444 /* inform statistics that we started a dead-node elimination run */
445 hook_dead_node_elim(irg, 1);
447 /* Remember external state of current_ir_graph. */
448 rem = current_ir_graph;
449 current_ir_graph = irg;
450 #ifdef INTERPROCEDURAL_VIEW
451 set_interprocedural_view(0);
454 assert(get_irg_phase_state(irg) != phase_building);
456 /* Handle graph state */
457 free_callee_info(irg);
461 /* @@@ so far we loose loops when copying */
462 free_loop_information(irg);
464 set_irg_doms_inconsistent(irg);
466 /* A quiet place, where the old obstack can rest in peace,
467 until it will be cremated. */
468 graveyard_obst = irg->obst;
470 /* A new obstack, where the reachable nodes will be copied to. */
471 rebirth_obst = xmalloc(sizeof(*rebirth_obst));
472 irg->obst = rebirth_obst;
473 obstack_init(irg->obst);
474 irg->last_node_idx = 0;
476 /* We also need a new value table for CSE */
477 del_identities(irg->value_table);
478 irg->value_table = new_identities();
480 /* Copy the graph from the old to the new obstack */
481 copy_graph_env(/*copy_node_nr=*/1);
483 /* Free memory from old unoptimized obstack */
484 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
485 xfree(graveyard_obst); /* ... then free it. */
487 /* inform statistics that the run is over */
488 hook_dead_node_elim(irg, 0);
490 current_ir_graph = rem;
491 #ifdef INTERPROCEDURAL_VIEW
492 set_interprocedural_view(rem_ipview);
497 * Relink bad predecessors of a block and store the old in array to the
498 * link field. This function is called by relink_bad_predecessors().
499 * The array of link field starts with the block operand at position 0.
500 * If block has bad predecessors, create a new in array without bad preds.
501 * Otherwise let in array untouched.
503 static void relink_bad_block_predecessors(ir_node *n, void *env) {
504 ir_node **new_in, *irn;
505 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
508 /* if link field of block is NULL, look for bad predecessors otherwise
509 this is already done */
510 if (is_Block(n) && get_irn_link(n) == NULL) {
511 /* save old predecessors in link field (position 0 is the block operand)*/
512 set_irn_link(n, get_irn_in(n));
514 /* count predecessors without bad nodes */
515 old_irn_arity = get_irn_arity(n);
516 for (i = 0; i < old_irn_arity; i++)
517 if (!is_Bad(get_irn_n(n, i)))
520 /* arity changing: set new predecessors without bad nodes */
521 if (new_irn_arity < old_irn_arity) {
522 /* Get new predecessor array. We do not resize the array, as we must
523 keep the old one to update Phis. */
524 new_in = NEW_ARR_D(ir_node *, current_ir_graph->obst, (new_irn_arity+1));
526 /* set new predecessors in array */
529 for (i = 0; i < old_irn_arity; i++) {
530 irn = get_irn_n(n, i);
532 new_in[new_irn_n] = irn;
533 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
537 /* ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity); */
538 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
540 } /* ir node has bad predecessors */
541 } /* Block is not relinked */
545 * Relinks Bad predecessors from Blocks and Phis called by walker
546 * remove_bad_predecesors(). If n is a Block, call
547 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
548 * function of Phi's Block. If this block has bad predecessors, relink preds
551 static void relink_bad_predecessors(ir_node *n, void *env) {
552 ir_node *block, **old_in;
553 int i, old_irn_arity, new_irn_arity;
555 /* relink bad predecessors of a block */
557 relink_bad_block_predecessors(n, env);
559 /* If Phi node relink its block and its predecessors */
561 /* Relink predecessors of phi's block */
562 block = get_nodes_block(n);
563 if (get_irn_link(block) == NULL)
564 relink_bad_block_predecessors(block, env);
566 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
567 old_irn_arity = ARR_LEN(old_in);
569 /* Relink Phi predecessors if count of predecessors changed */
570 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
571 /* set new predecessors in array
572 n->in[0] remains the same block */
574 for(i = 1; i < old_irn_arity; i++)
575 if (!is_Bad(old_in[i])) {
576 n->in[new_irn_arity] = n->in[i];
577 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
581 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
582 ARR_SETLEN(int, n->attr.phi.u.backedge, new_irn_arity);
584 } /* n is a Phi node */
588 * Removes Bad Bad predecessors from Blocks and the corresponding
589 * inputs to Phi nodes as in dead_node_elimination but without
591 * On walking up set the link field to NULL, on walking down call
592 * relink_bad_predecessors() (This function stores the old in array
593 * to the link field and sets a new in array if arity of predecessors
596 void remove_bad_predecessors(ir_graph *irg) {
597 panic("Fix backedge handling first");
598 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
605 __)|_| | \_/ | \_/(/_ |_/\__|__
607 The following stuff implements a facility that automatically patches
608 registered ir_node pointers to the new node when a dead node elimination occurs.
611 struct _survive_dce_t {
615 hook_entry_t dead_node_elim;
616 hook_entry_t dead_node_elim_subst;
619 typedef struct _survive_dce_list_t {
620 struct _survive_dce_list_t *next;
622 } survive_dce_list_t;
624 static void dead_node_hook(void *context, ir_graph *irg, int start) {
625 survive_dce_t *sd = context;
628 /* Create a new map before the dead node elimination is performed. */
630 sd->new_places = pmap_create_ex(pmap_count(sd->places));
632 /* Patch back all nodes if dead node elimination is over and something is to be done. */
633 pmap_destroy(sd->places);
634 sd->places = sd->new_places;
635 sd->new_places = NULL;
640 * Hook called when dead node elimination replaces old by nw.
642 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw) {
643 survive_dce_t *sd = context;
644 survive_dce_list_t *list = pmap_get(sd->places, old);
647 /* If the node is to be patched back, write the new address to all registered locations. */
649 survive_dce_list_t *p;
651 for (p = list; p; p = p->next)
654 pmap_insert(sd->new_places, nw, list);
659 * Make a new Survive DCE environment.
661 survive_dce_t *new_survive_dce(void) {
662 survive_dce_t *res = xmalloc(sizeof(res[0]));
663 obstack_init(&res->obst);
664 res->places = pmap_create();
665 res->new_places = NULL;
667 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
668 res->dead_node_elim.context = res;
669 res->dead_node_elim.next = NULL;
671 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
672 res->dead_node_elim_subst.context = res;
673 res->dead_node_elim_subst.next = NULL;
675 #ifndef FIRM_ENABLE_HOOKS
676 assert(0 && "need hooks enabled");
679 register_hook(hook_dead_node_elim, &res->dead_node_elim);
680 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
685 * Free a Survive DCE environment.
687 void free_survive_dce(survive_dce_t *sd) {
688 obstack_free(&sd->obst, NULL);
689 pmap_destroy(sd->places);
690 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
691 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
696 * Register a node pointer to be patched upon DCE.
697 * When DCE occurs, the node pointer specified by @p place will be
698 * patched to the new address of the node it is pointing to.
700 * @param sd The Survive DCE environment.
701 * @param place The address of the node pointer.
703 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place) {
704 if (*place != NULL) {
705 ir_node *irn = *place;
706 survive_dce_list_t *curr = pmap_get(sd->places, irn);
707 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw[0]));
712 pmap_insert(sd->places, irn, nw);
716 /*--------------------------------------------------------------------*/
717 /* Functionality for inlining */
718 /*--------------------------------------------------------------------*/
721 * Copy node for inlineing. Updates attributes that change when
722 * inlineing but not for dead node elimination.
724 * Copies the node by calling copy_node() and then updates the entity if
725 * it's a local one. env must be a pointer of the frame type of the
726 * inlined procedure. The new entities must be in the link field of
729 static void copy_node_inline(ir_node *n, void *env) {
731 ir_type *frame_tp = (ir_type *)env;
735 nn = get_new_node (n);
737 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
738 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
740 } else if (is_Block(n)) {
741 nn = get_new_node (n);
742 nn->attr.block.irg = current_ir_graph;
747 * Copies new predecessors of old node and move constants to
750 static void copy_preds_inline(ir_node *n, void *env) {
754 nn = skip_Id(get_new_node(n));
755 if (is_irn_constlike(nn)) {
756 /* move Constants into the start block */
757 set_nodes_block(nn, get_irg_start_block(current_ir_graph));
759 n = identify_remember(current_ir_graph->value_table, nn);
767 * Walker: checks if P_value_arg_base is used.
769 static void find_addr(ir_node *node, void *env) {
770 int *allow_inline = env;
772 is_Start(get_Proj_pred(node)) &&
773 get_Proj_proj(node) == pn_Start_P_value_arg_base) {
775 } else if (is_Alloc(node) && get_Alloc_where(node) == stack_alloc) {
777 * Refuse to inline alloca call unless user explicitly forced so as this
778 * may change program's memory overhead drastically when the function
779 * using alloca is called in loop. In GCC present in SPEC2000 inlining
780 * into schedule_block cause it to require 2GB of ram instead of 256MB.
782 * Sorryly this is true with our implementation also.
783 * Moreover, we cannot differentiate between alloca() and VLA yet, so this
784 * disables inlining of functions using VLA (with are completely save).
787 * - add a flag to the Alloc node for "real" alloca() calls
788 * - add a new Stack-Restore node at the end of a function using alloca()
795 * Check if we can inline a given call.
796 * Currently, we cannot inline two cases:
797 * - call with compound arguments
798 * - graphs that take the address of a parameter
800 * check these conditions here
802 static int can_inline(ir_node *call, ir_graph *called_graph) {
803 ir_type *call_type = get_Call_type(call);
804 int params, ress, i, res;
805 assert(is_Method_type(call_type));
807 params = get_method_n_params(call_type);
808 ress = get_method_n_ress(call_type);
810 /* check parameters for compound arguments */
811 for (i = 0; i < params; ++i) {
812 ir_type *p_type = get_method_param_type(call_type, i);
814 if (is_compound_type(p_type))
818 /* check results for compound arguments */
819 for (i = 0; i < ress; ++i) {
820 ir_type *r_type = get_method_res_type(call_type, i);
822 if (is_compound_type(r_type))
827 irg_walk_graph(called_graph, find_addr, NULL, &res);
833 exc_handler = 0, /**< There is a handler. */
834 exc_to_end = 1, /**< Branches to End. */
835 exc_no_handler = 2 /**< Exception handling not represented. */
838 /* Inlines a method at the given call site. */
839 int inline_method(ir_node *call, ir_graph *called_graph) {
841 ir_node *post_call, *post_bl;
842 ir_node *in[pn_Start_max];
843 ir_node *end, *end_bl, *block;
848 int arity, n_ret, n_exc, n_res, i, n, j, rem_opt, irn_arity, n_params;
849 enum exc_mode exc_handling;
850 ir_type *called_frame, *curr_frame, *mtp, *ctp;
853 irg_inline_property prop = get_irg_inline_property(called_graph);
855 if (prop == irg_inline_forbidden)
858 ent = get_irg_entity(called_graph);
860 mtp = get_entity_type(ent);
861 ctp = get_Call_type(call);
862 if (get_method_n_params(mtp) > get_method_n_params(ctp)) {
863 /* this is a bad feature of C: without a prototype, we can can call a function with less
864 parameters than needed. Currently we don't support this, although it would be
865 to use Unknown than. */
869 /* Argh, compiling C has some bad consequences:
870 the call type AND the method type might be different.
871 It is implementation defendant what happens in that case.
872 We support inlining, if the bitsize of the types matches AND
873 the same arithmetic is used. */
874 n_params = get_method_n_params(mtp);
875 for (i = n_params - 1; i >= 0; --i) {
876 ir_type *param_tp = get_method_param_type(mtp, i);
877 ir_type *arg_tp = get_method_param_type(ctp, i);
879 if (param_tp != arg_tp) {
880 ir_mode *pmode = get_type_mode(param_tp);
881 ir_mode *amode = get_type_mode(arg_tp);
883 if (pmode == NULL || amode == NULL)
885 if (get_mode_size_bits(pmode) != get_mode_size_bits(amode))
887 if (get_mode_arithmetic(pmode) != get_mode_arithmetic(amode))
889 /* otherwise we can simply "reinterpret" the bits */
893 irg = get_irn_irg(call);
896 * We cannot inline a recursive call. The graph must be copied before
897 * the call the inline_method() using create_irg_copy().
899 if (called_graph == irg)
903 * currently, we cannot inline two cases:
904 * - call with compound arguments
905 * - graphs that take the address of a parameter
907 if (! can_inline(call, called_graph))
910 rem = current_ir_graph;
911 current_ir_graph = irg;
913 DB((dbg, LEVEL_1, "Inlining %+F(%+F) into %+F\n", call, called_graph, irg));
915 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
916 rem_opt = get_opt_optimize();
919 /* Handle graph state */
920 assert(get_irg_phase_state(irg) != phase_building);
921 assert(get_irg_pinned(irg) == op_pin_state_pinned);
922 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
923 set_irg_outs_inconsistent(irg);
924 set_irg_extblk_inconsistent(irg);
925 set_irg_doms_inconsistent(irg);
926 set_irg_loopinfo_inconsistent(irg);
927 set_irg_callee_info_state(irg, irg_callee_info_inconsistent);
929 /* -- Check preconditions -- */
930 assert(is_Call(call));
932 /* here we know we WILL inline, so inform the statistics */
933 hook_inline(call, called_graph);
935 /* -- Decide how to handle exception control flow: Is there a handler
936 for the Call node, or do we branch directly to End on an exception?
938 0 There is a handler.
940 2 Exception handling not represented in Firm. -- */
942 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
943 for (proj = get_irn_link(call); proj; proj = get_irn_link(proj)) {
944 long proj_nr = get_Proj_proj(proj);
945 if (proj_nr == pn_Call_X_except) Xproj = proj;
946 if (proj_nr == pn_Call_M_except) Mproj = proj;
948 if (Mproj) { assert(Xproj); exc_handling = exc_handler; } /* Mproj */
949 else if (Xproj) { exc_handling = exc_to_end; } /* !Mproj && Xproj */
950 else { exc_handling = exc_no_handler; } /* !Mproj && !Xproj */
953 /* create the argument tuple */
954 NEW_ARR_A(ir_type *, args_in, n_params);
956 block = get_nodes_block(call);
957 for (i = n_params - 1; i >= 0; --i) {
958 ir_node *arg = get_Call_param(call, i);
959 ir_type *param_tp = get_method_param_type(mtp, i);
960 ir_mode *mode = get_type_mode(param_tp);
962 if (mode != get_irn_mode(arg)) {
963 arg = new_r_Conv(irg, block, arg, mode);
969 the procedure and later replaces the Start node of the called graph.
970 Post_call is the old Call node and collects the results of the called
971 graph. Both will end up being a tuple. -- */
972 post_bl = get_nodes_block(call);
973 set_irg_current_block(irg, post_bl);
974 /* XxMxPxPxPxT of Start + parameter of Call */
975 in[pn_Start_X_initial_exec] = new_Jmp();
976 in[pn_Start_M] = get_Call_mem(call);
977 in[pn_Start_P_frame_base] = get_irg_frame(irg);
978 in[pn_Start_P_tls] = get_irg_tls(irg);
979 in[pn_Start_T_args] = new_Tuple(n_params, args_in);
980 /* in[pn_Start_P_value_arg_base] = ??? */
981 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
982 pre_call = new_Tuple(pn_Start_max - 1, in);
986 The new block gets the ins of the old block, pre_call and all its
987 predecessors and all Phi nodes. -- */
988 part_block(pre_call);
990 /* -- Prepare state for dead node elimination -- */
991 /* Visited flags in calling irg must be >= flag in called irg.
992 Else walker and arity computation will not work. */
993 if (get_irg_visited(irg) <= get_irg_visited(called_graph))
994 set_irg_visited(irg, get_irg_visited(called_graph) + 1);
995 if (get_irg_block_visited(irg) < get_irg_block_visited(called_graph))
996 set_irg_block_visited(irg, get_irg_block_visited(called_graph));
997 /* Set pre_call as new Start node in link field of the start node of
998 calling graph and pre_calls block as new block for the start block
1000 Further mark these nodes so that they are not visited by the
1002 set_irn_link(get_irg_start(called_graph), pre_call);
1003 set_irn_visited(get_irg_start(called_graph), get_irg_visited(irg));
1004 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1005 set_irn_visited(get_irg_start_block(called_graph), get_irg_visited(irg));
1006 set_irn_link(get_irg_bad(called_graph), get_irg_bad(irg));
1007 set_irn_visited(get_irg_bad(called_graph), get_irg_visited(irg));
1009 /* Initialize for compaction of in arrays */
1010 inc_irg_block_visited(irg);
1012 /* -- Replicate local entities of the called_graph -- */
1013 /* copy the entities. */
1014 called_frame = get_irg_frame_type(called_graph);
1015 curr_frame = get_irg_frame_type(irg);
1016 for (i = 0, n = get_class_n_members(called_frame); i < n; ++i) {
1017 ir_entity *new_ent, *old_ent;
1018 old_ent = get_class_member(called_frame, i);
1019 new_ent = copy_entity_own(old_ent, curr_frame);
1020 set_entity_link(old_ent, new_ent);
1023 /* visited is > than that of called graph. With this trick visited will
1024 remain unchanged so that an outer walker, e.g., searching the call nodes
1025 to inline, calling this inline will not visit the inlined nodes. */
1026 set_irg_visited(irg, get_irg_visited(irg)-1);
1028 /* -- Performing dead node elimination inlines the graph -- */
1029 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1031 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds_inline,
1032 get_irg_frame_type(called_graph));
1034 /* Repair called_graph */
1035 set_irg_visited(called_graph, get_irg_visited(irg));
1036 set_irg_block_visited(called_graph, get_irg_block_visited(irg));
1037 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1039 /* -- Merge the end of the inlined procedure with the call site -- */
1040 /* We will turn the old Call node into a Tuple with the following
1043 0: Phi of all Memories of Return statements.
1044 1: Jmp from new Block that merges the control flow from all exception
1045 predecessors of the old end block.
1046 2: Tuple of all arguments.
1047 3: Phi of Exception memories.
1048 In case the old Call directly branches to End on an exception we don't
1049 need the block merging all exceptions nor the Phi of the exception
1053 /* -- Precompute some values -- */
1054 end_bl = get_new_node(get_irg_end_block(called_graph));
1055 end = get_new_node(get_irg_end(called_graph));
1056 arity = get_irn_arity(end_bl); /* arity = n_exc + n_ret */
1057 n_res = get_method_n_ress(get_Call_type(call));
1059 res_pred = xmalloc(n_res * sizeof(*res_pred));
1060 cf_pred = xmalloc(arity * sizeof(*res_pred));
1062 set_irg_current_block(irg, post_bl); /* just to make sure */
1064 /* -- archive keepalives -- */
1065 irn_arity = get_irn_arity(end);
1066 for (i = 0; i < irn_arity; i++) {
1067 ir_node *ka = get_End_keepalive(end, i);
1069 add_End_keepalive(get_irg_end(irg), ka);
1072 /* The new end node will die. We need not free as the in array is on the obstack:
1073 copy_node() only generated 'D' arrays. */
1075 /* -- Replace Return nodes by Jump nodes. -- */
1077 for (i = 0; i < arity; i++) {
1079 ret = get_irn_n(end_bl, i);
1080 if (is_Return(ret)) {
1081 cf_pred[n_ret] = new_r_Jmp(irg, get_nodes_block(ret));
1085 set_irn_in(post_bl, n_ret, cf_pred);
1087 /* -- Build a Tuple for all results of the method.
1088 Add Phi node if there was more than one Return. -- */
1089 turn_into_tuple(post_call, pn_Call_max);
1090 /* First the Memory-Phi */
1092 for (i = 0; i < arity; i++) {
1093 ret = get_irn_n(end_bl, i);
1094 if (is_Return(ret)) {
1095 cf_pred[n_ret] = get_Return_mem(ret);
1099 phi = new_Phi(n_ret, cf_pred, mode_M);
1100 set_Tuple_pred(call, pn_Call_M_regular, phi);
1101 /* Conserve Phi-list for further inlinings -- but might be optimized */
1102 if (get_nodes_block(phi) == post_bl) {
1103 set_irn_link(phi, get_irn_link(post_bl));
1104 set_irn_link(post_bl, phi);
1106 /* Now the real results */
1108 for (j = 0; j < n_res; j++) {
1110 for (i = 0; i < arity; i++) {
1111 ret = get_irn_n(end_bl, i);
1112 if (is_Return(ret)) {
1113 cf_pred[n_ret] = get_Return_res(ret, j);
1118 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1122 /* Conserve Phi-list for further inlinings -- but might be optimized */
1123 if (get_nodes_block(phi) == post_bl) {
1124 set_Phi_next(phi, get_Block_phis(post_bl));
1125 set_Block_phis(post_bl, phi);
1128 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1130 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1132 /* handle the regular call */
1133 set_Tuple_pred(call, pn_Call_X_regular, new_Jmp());
1135 /* For now, we cannot inline calls with value_base */
1136 set_Tuple_pred(call, pn_Call_P_value_res_base, new_Bad());
1138 /* Finally the exception control flow.
1139 We have two (three) possible situations:
1140 First if the Call branches to an exception handler: We need to add a Phi node to
1141 collect the memory containing the exception objects. Further we need
1142 to add another block to get a correct representation of this Phi. To
1143 this block we add a Jmp that resolves into the X output of the Call
1144 when the Call is turned into a tuple.
1145 Second the Call branches to End, the exception is not handled. Just
1146 add all inlined exception branches to the End node.
1147 Third: there is no Exception edge at all. Handle as case two. */
1148 if (exc_handling == exc_handler) {
1150 for (i = 0; i < arity; i++) {
1152 ret = get_irn_n(end_bl, i);
1153 irn = skip_Proj(ret);
1154 if (is_fragile_op(irn) || is_Raise(irn)) {
1155 cf_pred[n_exc] = ret;
1160 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1161 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1162 /* The Phi for the memories with the exception objects */
1164 for (i = 0; i < arity; i++) {
1166 ret = skip_Proj(get_irn_n(end_bl, i));
1168 cf_pred[n_exc] = new_r_Proj(irg, get_nodes_block(ret), ret, mode_M, 3);
1170 } else if (is_fragile_op(ret)) {
1171 /* We rely that all cfops have the memory output at the same position. */
1172 cf_pred[n_exc] = new_r_Proj(irg, get_nodes_block(ret), ret, mode_M, 0);
1174 } else if (is_Raise(ret)) {
1175 cf_pred[n_exc] = new_r_Proj(irg, get_nodes_block(ret), ret, mode_M, 1);
1179 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1181 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1182 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1185 ir_node *main_end_bl;
1186 int main_end_bl_arity;
1187 ir_node **end_preds;
1189 /* assert(exc_handling == 1 || no exceptions. ) */
1191 for (i = 0; i < arity; i++) {
1192 ir_node *ret = get_irn_n(end_bl, i);
1193 ir_node *irn = skip_Proj(ret);
1195 if (is_fragile_op(irn) || is_Raise(irn)) {
1196 cf_pred[n_exc] = ret;
1200 main_end_bl = get_irg_end_block(irg);
1201 main_end_bl_arity = get_irn_arity(main_end_bl);
1202 end_preds = xmalloc((n_exc + main_end_bl_arity) * sizeof(*end_preds));
1204 for (i = 0; i < main_end_bl_arity; ++i)
1205 end_preds[i] = get_irn_n(main_end_bl, i);
1206 for (i = 0; i < n_exc; ++i)
1207 end_preds[main_end_bl_arity + i] = cf_pred[i];
1208 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1209 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1210 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1216 /* -- Turn CSE back on. -- */
1217 set_optimize(rem_opt);
1218 current_ir_graph = rem;
1223 /********************************************************************/
1224 /* Apply inlineing to small methods. */
1225 /********************************************************************/
1227 static struct obstack temp_obst;
1229 /** Represents a possible inlinable call in a graph. */
1230 typedef struct _call_entry call_entry;
1231 struct _call_entry {
1232 ir_node *call; /**< the Call node */
1233 ir_graph *callee; /**< the callee IR-graph called here */
1234 call_entry *next; /**< for linking the next one */
1235 int loop_depth; /**< the loop depth of this call */
1239 * environment for inlining small irgs
1241 typedef struct _inline_env_t {
1242 struct obstack obst; /**< an obstack where call_entries are allocated on. */
1243 call_entry *head; /**< the head of the call entry list */
1244 call_entry *tail; /**< the tail of the call entry list */
1248 * Returns the irg called from a Call node. If the irg is not
1249 * known, NULL is returned.
1251 * @param call the call node
1253 static ir_graph *get_call_called_irg(ir_node *call) {
1256 addr = get_Call_ptr(call);
1257 if (is_Global(addr)) {
1258 ir_entity *ent = get_Global_entity(addr);
1259 return get_entity_irg(ent);
1266 * Walker: Collect all calls to known graphs inside a graph.
1268 static void collect_calls(ir_node *call, void *env) {
1269 if (is_Call(call)) {
1270 ir_graph *called_irg = get_call_called_irg(call);
1272 if (called_irg != NULL) {
1273 /* The Call node calls a locally defined method. Remember to inline. */
1274 inline_env_t *ienv = env;
1275 call_entry *entry = obstack_alloc(&ienv->obst, sizeof(*entry));
1277 entry->callee = called_irg;
1279 entry->loop_depth = 0;
1281 if (ienv->tail == NULL)
1284 ienv->tail->next = entry;
1291 * Inlines all small methods at call sites where the called address comes
1292 * from a Const node that references the entity representing the called
1294 * The size argument is a rough measure for the code size of the method:
1295 * Methods where the obstack containing the firm graph is smaller than
1298 void inline_small_irgs(ir_graph *irg, int size) {
1299 ir_graph *rem = current_ir_graph;
1303 current_ir_graph = irg;
1304 /* Handle graph state */
1305 assert(get_irg_phase_state(irg) != phase_building);
1306 free_callee_info(irg);
1308 /* Find Call nodes to inline.
1309 (We can not inline during a walk of the graph, as inlineing the same
1310 method several times changes the visited flag of the walked graph:
1311 after the first inlineing visited of the callee equals visited of
1312 the caller. With the next inlineing both are increased.) */
1313 obstack_init(&env.obst);
1314 env.head = env.tail = NULL;
1315 irg_walk_graph(irg, NULL, collect_calls, &env);
1317 if (env.head != NULL) {
1318 /* There are calls to inline */
1319 collect_phiprojs(irg);
1320 for (entry = env.head; entry != NULL; entry = entry->next) {
1321 ir_graph *callee = entry->callee;
1322 if (((_obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst)) < size) ||
1323 (get_irg_inline_property(callee) >= irg_inline_forced)) {
1324 inline_method(entry->call, callee);
1328 obstack_free(&env.obst, NULL);
1329 current_ir_graph = rem;
1333 * Environment for inlining irgs.
1336 int n_nodes; /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
1337 int n_blocks; /**< Number of Blocks in graph without Start and End block. */
1338 int n_nodes_orig; /**< for statistics */
1339 int n_call_nodes; /**< Number of Call nodes in the graph. */
1340 int n_call_nodes_orig; /**< for statistics */
1341 int n_callers; /**< Number of known graphs that call this graphs. */
1342 int n_callers_orig; /**< for statistics */
1343 unsigned got_inline:1; /**< Set, if at least one call inside this graph was inlined. */
1344 unsigned local_vars:1; /**< Set, if a inlined function gets the address of an inlined variable. */
1345 unsigned recursive:1; /**< Set, if this function is self recursive. */
1346 call_entry *call_head; /**< The head of the list of all call nodes in this graph. */
1347 call_entry *call_tail; /**< The tail of the list of all call nodes in this graph .*/
1348 unsigned *local_weights; /**< Once allocated, the beneficial weight for transmitting local addresses. */
1352 * Allocate a new environment for inlining.
1354 static inline_irg_env *alloc_inline_irg_env(void) {
1355 inline_irg_env *env = obstack_alloc(&temp_obst, sizeof(*env));
1356 env->n_nodes = -2; /* do not count count Start, End */
1357 env->n_blocks = -2; /* do not count count Start, End Block */
1358 env->n_nodes_orig = -2; /* do not count Start, End */
1359 env->call_head = NULL;
1360 env->call_tail = NULL;
1361 env->n_call_nodes = 0;
1362 env->n_call_nodes_orig = 0;
1364 env->n_callers_orig = 0;
1365 env->got_inline = 0;
1366 env->local_vars = 0;
1368 env->local_weights = NULL;
1372 typedef struct walker_env {
1373 inline_irg_env *x; /**< the inline environment */
1374 call_entry *last_call; /**< points to the last inserted call */
1375 char ignore_runtime; /**< the ignore runtime flag */
1376 char ignore_callers; /**< if set, do change callers data */
1380 * post-walker: collect all calls in the inline-environment
1381 * of a graph and sum some statistics.
1383 static void collect_calls2(ir_node *call, void *ctx) {
1385 inline_irg_env *x = env->x;
1386 ir_opcode code = get_irn_opcode(call);
1390 /* count meaningful nodes in irg */
1391 if (code != iro_Proj && code != iro_Tuple && code != iro_Sync) {
1392 if (code != iro_Block) {
1400 if (code != iro_Call) return;
1402 /* check, if it's a runtime call */
1403 if (env->ignore_runtime) {
1404 ir_node *symc = get_Call_ptr(call);
1406 if (is_Global(symc)) {
1407 ir_entity *ent = get_Global_entity(symc);
1409 if (get_entity_additional_properties(ent) & mtp_property_runtime)
1414 /* collect all call nodes */
1416 ++x->n_call_nodes_orig;
1418 callee = get_call_called_irg(call);
1419 if (callee != NULL) {
1420 if (! env->ignore_callers) {
1421 inline_irg_env *callee_env = get_irg_link(callee);
1422 /* count all static callers */
1423 ++callee_env->n_callers;
1424 ++callee_env->n_callers_orig;
1426 if (callee == current_ir_graph)
1429 /* link it in the list of possible inlinable entries */
1430 entry = obstack_alloc(&temp_obst, sizeof(*entry));
1432 entry->callee = callee;
1434 entry->loop_depth = get_irn_loop(get_nodes_block(call))->depth;
1436 /* note: we use call_tail here as a pointer to the last inserted */
1437 if (x->call_head == NULL) {
1438 x->call_head = entry;
1440 if (entry->loop_depth == env->last_call->loop_depth) {
1441 /* same depth as the last one, enqueue after it */
1442 entry->next = env->last_call->next;
1443 env->last_call->next = entry;
1444 } else if (entry->loop_depth > x->call_head->loop_depth) {
1446 entry->next = x->call_head;
1447 x->call_head = entry;
1449 /* search the insertion point */
1452 for (p = x->call_head; p->next != NULL; p = p->next)
1453 if (entry->loop_depth > p->next->loop_depth)
1455 entry->next = p->next;
1459 env->last_call = entry;
1460 if (entry->next == NULL) {
1461 /* keep tail up to date */
1462 x->call_tail = entry;
1468 * Returns TRUE if the number of callers is 0 in the irg's environment,
1469 * hence this irg is a leave.
1471 INLINE static int is_leave(ir_graph *irg) {
1472 inline_irg_env *env = get_irg_link(irg);
1473 return env->n_call_nodes == 0;
1477 * Returns TRUE if the number of nodes in the callee is
1478 * smaller then size in the irg's environment.
1480 INLINE static int is_smaller(ir_graph *callee, int size) {
1481 inline_irg_env *env = get_irg_link(callee);
1482 return env->n_nodes < size;
1486 * Append the nodes of the list src to the nodes of the list in environment dst.
1488 static void append_call_list(inline_irg_env *dst, call_entry *src) {
1489 call_entry *entry, *nentry;
1491 /* Note that the src list points to Call nodes in the inlined graph, but
1492 we need Call nodes in our graph. Luckily the inliner leaves this information
1493 in the link field. */
1494 for (entry = src; entry != NULL; entry = entry->next) {
1495 nentry = obstack_alloc(&temp_obst, sizeof(*nentry));
1496 nentry->call = get_irn_link(entry->call);
1497 nentry->callee = entry->callee;
1498 nentry->next = NULL;
1499 nentry->loop_depth = entry->loop_depth;
1500 dst->call_tail->next = nentry;
1501 dst->call_tail = nentry;
1506 * Add the nodes of the list src in front to the nodes of the list dst.
1508 static call_entry *replace_entry_by_call_list(call_entry *dst, call_entry *src) {
1509 call_entry *entry, *nentry, *head, *tail;
1511 /* Note that the src list points to Call nodes in the inlined graph, but
1512 we need Call nodes in our graph. Luckily the inliner leaves this information
1513 in the link field. */
1515 for (entry = src; entry != NULL; entry = entry->next) {
1516 nentry = obstack_alloc(&temp_obst, sizeof(*nentry));
1517 nentry->call = get_irn_link(entry->call);
1518 nentry->callee = entry->callee;
1519 nentry->next = NULL;
1520 nentry->loop_depth = entry->loop_depth + dst->loop_depth;
1524 tail->next = nentry;
1527 /* skip the head of dst */
1529 tail->next = dst->next;
1537 * Inlines small leave methods at call sites where the called address comes
1538 * from a Const node that references the entity representing the called
1540 * The size argument is a rough measure for the code size of the method:
1541 * Methods where the obstack containing the firm graph is smaller than
1544 void inline_leave_functions(int maxsize, int leavesize, int size, int ignore_runtime) {
1545 inline_irg_env *env;
1551 call_entry *entry, *tail;
1552 const call_entry *centry;
1553 pmap *copied_graphs;
1554 pmap_entry *pm_entry;
1556 rem = current_ir_graph;
1557 obstack_init(&temp_obst);
1559 /* a map for the copied graphs, used to inline recursive calls */
1560 copied_graphs = pmap_create();
1562 /* extend all irgs by a temporary data structure for inlining. */
1563 n_irgs = get_irp_n_irgs();
1564 for (i = 0; i < n_irgs; ++i)
1565 set_irg_link(get_irp_irg(i), alloc_inline_irg_env());
1567 /* Precompute information in temporary data structure. */
1568 wenv.ignore_runtime = ignore_runtime;
1569 wenv.ignore_callers = 0;
1570 for (i = 0; i < n_irgs; ++i) {
1571 ir_graph *irg = get_irp_irg(i);
1573 assert(get_irg_phase_state(irg) != phase_building);
1574 free_callee_info(irg);
1576 assure_cf_loop(irg);
1577 wenv.x = get_irg_link(irg);
1578 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1581 /* -- and now inline. -- */
1583 /* Inline leaves recursively -- we might construct new leaves. */
1587 for (i = 0; i < n_irgs; ++i) {
1589 int phiproj_computed = 0;
1591 current_ir_graph = get_irp_irg(i);
1592 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1595 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1598 if (env->n_nodes > maxsize) break;
1601 callee = entry->callee;
1603 if (is_leave(callee) && (
1604 is_smaller(callee, leavesize) || (get_irg_inline_property(callee) >= irg_inline_forced))) {
1605 if (!phiproj_computed) {
1606 phiproj_computed = 1;
1607 collect_phiprojs(current_ir_graph);
1609 did_inline = inline_method(call, callee);
1612 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1614 /* was inlined, must be recomputed */
1615 phiproj_computed = 0;
1617 /* Do some statistics */
1618 env->got_inline = 1;
1619 --env->n_call_nodes;
1620 env->n_nodes += callee_env->n_nodes;
1621 --callee_env->n_callers;
1623 /* remove this call from the list */
1625 tail->next = entry->next;
1627 env->call_head = entry->next;
1633 env->call_tail = tail;
1635 } while (did_inline);
1637 /* inline other small functions. */
1638 for (i = 0; i < n_irgs; ++i) {
1640 int phiproj_computed = 0;
1642 current_ir_graph = get_irp_irg(i);
1643 env = (inline_irg_env *)get_irg_link(current_ir_graph);
1645 /* note that the list of possible calls is updated during the process */
1647 for (entry = env->call_head; entry != NULL; entry = entry->next) {
1652 callee = entry->callee;
1654 e = pmap_find(copied_graphs, callee);
1657 * Remap callee if we have a copy.
1658 * FIXME: Should we do this only for recursive Calls ?
1663 if (((is_smaller(callee, size) && (env->n_nodes < maxsize)) || /* small function */
1664 (get_irg_inline_property(callee) >= irg_inline_forced))) {
1665 if (current_ir_graph == callee) {
1667 * Recursive call: we cannot directly inline because we cannot walk
1668 * the graph and change it. So we have to make a copy of the graph
1672 inline_irg_env *callee_env;
1676 * No copy yet, create one.
1677 * Note that recursive methods are never leaves, so it is sufficient
1678 * to test this condition here.
1680 copy = create_irg_copy(callee);
1682 /* create_irg_copy() destroys the Proj links, recompute them */
1683 phiproj_computed = 0;
1685 /* allocate new environment */
1686 callee_env = alloc_inline_irg_env();
1687 set_irg_link(copy, callee_env);
1689 assure_cf_loop(copy);
1690 wenv.x = callee_env;
1691 wenv.ignore_callers = 1;
1692 irg_walk_graph(copy, NULL, collect_calls2, &wenv);
1695 * Enter the entity of the original graph. This is needed
1696 * for inline_method(). However, note that ent->irg still points
1697 * to callee, NOT to copy.
1699 set_irg_entity(copy, get_irg_entity(callee));
1701 pmap_insert(copied_graphs, callee, copy);
1704 /* we have only one caller: the original graph */
1705 callee_env->n_callers = 1;
1706 callee_env->n_callers_orig = 1;
1708 if (! phiproj_computed) {
1709 phiproj_computed = 1;
1710 collect_phiprojs(current_ir_graph);
1712 did_inline = inline_method(call, callee);
1714 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1716 /* was inlined, must be recomputed */
1717 phiproj_computed = 0;
1719 /* callee was inline. Append it's call list. */
1720 env->got_inline = 1;
1721 --env->n_call_nodes;
1722 append_call_list(env, callee_env->call_head);
1723 env->n_call_nodes += callee_env->n_call_nodes;
1724 env->n_nodes += callee_env->n_nodes;
1725 --callee_env->n_callers;
1727 /* after we have inlined callee, all called methods inside callee
1728 are now called once more */
1729 for (centry = callee_env->call_head; centry != NULL; centry = centry->next) {
1730 inline_irg_env *penv = get_irg_link(centry->callee);
1734 /* remove this call from the list */
1736 tail->next = entry->next;
1738 env->call_head = entry->next;
1744 env->call_tail = tail;
1747 for (i = 0; i < n_irgs; ++i) {
1748 irg = get_irp_irg(i);
1749 env = (inline_irg_env *)get_irg_link(irg);
1751 if (env->got_inline) {
1752 optimize_graph_df(irg);
1755 if (env->got_inline || (env->n_callers_orig != env->n_callers)) {
1756 DB((dbg, LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1757 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1758 env->n_callers_orig, env->n_callers,
1759 get_entity_name(get_irg_entity(irg))));
1763 /* kill the copied graphs: we don't need them anymore */
1764 foreach_pmap(copied_graphs, pm_entry) {
1765 ir_graph *copy = pm_entry->value;
1767 /* reset the entity, otherwise it will be deleted in the next step ... */
1768 set_irg_entity(copy, NULL);
1769 free_ir_graph(copy);
1771 pmap_destroy(copied_graphs);
1773 obstack_free(&temp_obst, NULL);
1774 current_ir_graph = rem;
1778 * Calculate the parameter weights for transmitting the address of a local variable.
1780 static unsigned calc_method_local_weight(ir_node *arg) {
1782 unsigned v, weight = 0;
1784 for (i = get_irn_n_outs(arg) - 1; i >= 0; --i) {
1785 ir_node *succ = get_irn_out(arg, i);
1787 switch (get_irn_opcode(succ)) {
1790 /* Loads and Store can be removed */
1794 /* check if all args are constant */
1795 for (j = get_Sel_n_indexs(succ) - 1; j >= 0; --j) {
1796 ir_node *idx = get_Sel_index(succ, j);
1797 if (! is_Const(idx))
1800 /* Check users on this Sel. Note: if a 0 is returned here, there was
1801 some unsupported node. */
1802 v = calc_method_local_weight(succ);
1805 /* we can kill one Sel with constant indexes, this is cheap */
1809 /* when looking backward we might find Id nodes */
1810 weight += calc_method_local_weight(succ);
1813 /* unoptimized tuple */
1814 for (j = get_Tuple_n_preds(succ) - 1; j >= 0; --j) {
1815 ir_node *pred = get_Tuple_pred(succ, j);
1817 /* look for Proj(j) */
1818 for (k = get_irn_n_outs(succ) - 1; k >= 0; --k) {
1819 ir_node *succ_succ = get_irn_out(succ, k);
1820 if (is_Proj(succ_succ)) {
1821 if (get_Proj_proj(succ_succ) == j) {
1823 weight += calc_method_local_weight(succ_succ);
1826 /* this should NOT happen */
1834 /* any other node: unsupported yet or bad. */
1842 * Calculate the parameter weights for transmitting the address of a local variable.
1844 static void analyze_irg_local_weights(inline_irg_env *env, ir_graph *irg) {
1845 ir_entity *ent = get_irg_entity(irg);
1847 int nparams, i, proj_nr;
1848 ir_node *irg_args, *arg;
1850 mtp = get_entity_type(ent);
1851 nparams = get_method_n_params(mtp);
1853 /* allocate a new array. currently used as 'analysed' flag */
1854 env->local_weights = NEW_ARR_D(unsigned, &temp_obst, nparams);
1856 /* If the method haven't parameters we have nothing to do. */
1860 assure_irg_outs(irg);
1861 irg_args = get_irg_args(irg);
1862 for (i = get_irn_n_outs(irg_args) - 1; i >= 0; --i) {
1863 arg = get_irn_out(irg_args, i);
1864 proj_nr = get_Proj_proj(arg);
1865 env->local_weights[proj_nr] = calc_method_local_weight(arg);
1870 * Calculate the benefice for transmitting an local variable address.
1871 * After inlining, the local variable might be transformed into a
1872 * SSA variable by scalar_replacement().
1874 static unsigned get_method_local_adress_weight(ir_graph *callee, int pos) {
1875 inline_irg_env *env = get_irg_link(callee);
1877 if (env->local_weights != NULL) {
1878 if (pos < ARR_LEN(env->local_weights))
1879 return env->local_weights[pos];
1883 analyze_irg_local_weights(env, callee);
1885 if (pos < ARR_LEN(env->local_weights))
1886 return env->local_weights[pos];
1891 * Calculate a benefice value for inlining the given call.
1893 * @param call the call node we have to inspect
1894 * @param callee the called graph
1895 * @param local_adr set after return if an address of a local variable is
1896 * transmitted as a parameter
1898 static int calc_inline_benefice(ir_node *call, ir_graph *callee, unsigned *local_adr) {
1899 ir_entity *ent = get_irg_entity(callee);
1903 int i, n_params, all_const;
1906 inline_irg_env *curr_env, *callee_env;
1908 if (get_entity_additional_properties(ent) & mtp_property_noreturn) {
1909 /* do NOT inline noreturn calls */
1913 /* costs for every passed parameter */
1914 n_params = get_Call_n_params(call);
1915 mtp = get_entity_type(ent);
1916 cc = get_method_calling_convention(mtp);
1917 if (cc & cc_reg_param) {
1918 /* register parameter, smaller costs for register parameters */
1919 int max_regs = cc & ~cc_bits;
1921 if (max_regs < n_params)
1922 weight += max_regs * 2 + (n_params - max_regs) * 5;
1924 weight += n_params * 2;
1926 /* parameters are passed an stack */
1927 weight += 5 * n_params;
1930 /* constant parameters improve the benefice */
1931 frame_ptr = get_irg_frame(current_ir_graph);
1933 for (i = 0; i < n_params; ++i) {
1934 ir_node *param = get_Call_param(call, i);
1936 if (is_Const(param)) {
1937 weight += get_method_param_weight(ent, i);
1940 if (is_SymConst(param))
1941 weight += get_method_param_weight(ent, i);
1942 else if (is_Sel(param) && get_Sel_ptr(param) == frame_ptr) {
1944 * An address of a local variable is transmitted. After inlining,
1945 * scalar_replacement might be able to remove the local variable,
1948 v = get_method_local_adress_weight(callee, i);
1956 callee_env = get_irg_link(callee);
1957 if (get_entity_visibility(ent) == visibility_local &&
1958 callee_env->n_callers_orig == 1 &&
1959 callee != current_ir_graph) {
1960 /* we are the only caller, give big bonus */
1964 /* do not inline big functions */
1965 weight -= callee_env->n_nodes;
1967 /* reduce the benefice if the current function is already big */
1968 curr_env = get_irg_link(current_ir_graph);
1969 weight -= curr_env->n_nodes / 50;
1971 /* give a bonus for functions with one block */
1972 if (callee_env->n_blocks == 1)
1973 weight = weight * 3 / 2;
1975 /* and one for small non-recursive functions: we want them to be inlined in mostly every case */
1976 else if (callee_env->n_nodes < 20 && !callee_env->recursive)
1979 /* and finally for leaves: they do not increase the register pressure
1980 because of callee safe registers */
1981 else if (callee_env->n_call_nodes == 0)
1985 * Reduce the weight for recursive function IFF not all arguments are const.
1986 * inlining recursive functions is rarely good.
1988 if (callee_env->recursive && !all_const)
1992 * All arguments constant is probably a good sign, give an extra bonus
2001 * Heuristic inliner. Calculates a benefice value for every call and inlines
2002 * those calls with a value higher than the threshold.
2004 void inline_functions(int maxsize, int inline_threshold) {
2005 inline_irg_env *env;
2010 call_entry *curr_call, **last_call;
2011 const call_entry *centry;
2012 pmap *copied_graphs;
2013 pmap_entry *pm_entry;
2015 rem = current_ir_graph;
2016 obstack_init(&temp_obst);
2018 /* a map for the copied graphs, used to inline recursive calls */
2019 copied_graphs = pmap_create();
2021 /* extend all irgs by a temporary data structure for inlining. */
2022 n_irgs = get_irp_n_irgs();
2023 for (i = 0; i < n_irgs; ++i)
2024 set_irg_link(get_irp_irg(i), alloc_inline_irg_env());
2026 /* Precompute information in temporary data structure. */
2027 wenv.ignore_runtime = 0;
2028 wenv.ignore_callers = 0;
2029 for (i = 0; i < n_irgs; ++i) {
2030 ir_graph *irg = get_irp_irg(i);
2032 assert(get_irg_phase_state(irg) != phase_building);
2033 free_callee_info(irg);
2035 wenv.x = get_irg_link(irg);
2036 wenv.last_call = NULL;
2037 assure_cf_loop(irg);
2038 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
2041 /* -- and now inline. -- */
2042 for (i = 0; i < n_irgs; ++i) {
2043 int phiproj_computed = 0;
2045 ir_graph *irg = get_irp_irg(i);
2047 current_ir_graph = irg;
2048 env = get_irg_link(irg);
2050 /* note that the list of possible calls is updated during the process */
2051 last_call = &env->call_head;
2052 for (curr_call = env->call_head; curr_call != NULL;) {
2058 if (env->n_nodes > maxsize) break;
2060 call = curr_call->call;
2061 callee = curr_call->callee;
2063 e = pmap_find(copied_graphs, callee);
2066 * Remap callee if we have a copy.
2067 * FIXME: Should we do this only for recursive Calls ?
2072 /* calculate the benefice on the original call to prevent excessive inlining */
2074 benefice = calc_inline_benefice(call, callee, &local_adr);
2075 DB((dbg, LEVEL_2, "In %+F Call %+F has benefice %d\n", irg, callee, benefice));
2077 if (benefice > -inline_threshold ||
2078 (get_irg_inline_property(callee) >= irg_inline_forced)) {
2079 if (current_ir_graph == callee) {
2081 * Recursive call: we cannot directly inline because we cannot walk
2082 * the graph and change it. So we have to make a copy of the graph
2086 inline_irg_env *callee_env;
2090 * No copy yet, create one.
2091 * Note that recursive methods are never leaves, so it is sufficient
2092 * to test this condition here.
2094 copy = create_irg_copy(callee);
2096 /* create_irg_copy() destroys the Proj links, recompute them */
2097 phiproj_computed = 0;
2099 /* allocate new environment */
2100 callee_env = alloc_inline_irg_env();
2101 set_irg_link(copy, callee_env);
2103 assure_cf_loop(copy);
2104 wenv.x = callee_env;
2105 wenv.ignore_callers = 1;
2106 irg_walk_graph(copy, NULL, collect_calls2, &wenv);
2109 * Enter the entity of the original graph. This is needed
2110 * for inline_method(). However, note that ent->irg still points
2111 * to callee, NOT to copy.
2113 set_irg_entity(copy, get_irg_entity(callee));
2115 pmap_insert(copied_graphs, callee, copy);
2118 /* we have only one caller: the original graph */
2119 callee_env->n_callers = 1;
2120 callee_env->n_callers_orig = 1;
2122 if (! phiproj_computed) {
2123 phiproj_computed = 1;
2124 collect_phiprojs(current_ir_graph);
2126 did_inline = inline_method(call, callee);
2128 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
2130 /* was inlined, must be recomputed */
2131 phiproj_computed = 0;
2133 /* after we have inlined callee, all called methods inside callee
2134 are now called once more */
2135 for (centry = callee_env->call_head; centry != NULL; centry = centry->next) {
2136 inline_irg_env *penv = get_irg_link(centry->callee);
2140 /* callee was inline. Append it's call list. */
2141 env->got_inline = 1;
2143 env->local_vars = 1;
2144 --env->n_call_nodes;
2145 curr_call = replace_entry_by_call_list(curr_call, callee_env->call_head);
2146 env->n_call_nodes += callee_env->n_call_nodes;
2147 env->n_nodes += callee_env->n_nodes;
2148 --callee_env->n_callers;
2150 /* remove the current call entry from the list */
2151 *last_call = curr_call;
2155 last_call = &curr_call->next;
2156 curr_call = curr_call->next;
2159 if (env->got_inline) {
2160 /* this irg got calls inlined: optimize it */
2162 /* scalar replacement does not work well with Tuple nodes, so optimize them away */
2163 optimize_graph_df(irg);
2165 if (env->local_vars) {
2166 if (scalar_replacement_opt(irg)) {
2167 optimize_graph_df(irg);
2172 if (env->got_inline || (env->n_callers_orig != env->n_callers)) {
2173 DB((dbg, LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
2174 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
2175 env->n_callers_orig, env->n_callers,
2176 get_entity_name(get_irg_entity(irg))));
2180 /* kill the copied graphs: we don't need them anymore */
2181 foreach_pmap(copied_graphs, pm_entry) {
2182 ir_graph *copy = pm_entry->value;
2184 /* reset the entity, otherwise it will be deleted in the next step ... */
2185 set_irg_entity(copy, NULL);
2186 free_ir_graph(copy);
2188 pmap_destroy(copied_graphs);
2190 obstack_free(&temp_obst, NULL);
2191 current_ir_graph = rem;
2194 void firm_init_inline(void) {
2195 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");