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"
54 #include "irbackedge_t.h"
55 #include "opt_inline_t.h"
60 #include "analyze_irg_args.h"
61 #include "iredges_t.h"
65 #include "iropt_dbg.h"
67 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
69 /*------------------------------------------------------------------*/
70 /* Routines for dead node elimination / copying garbage collection */
72 /*------------------------------------------------------------------*/
75 * Remember the new node in the old node by using a field all nodes have.
77 #define set_new_node(oldn, newn) set_irn_link(oldn, newn)
80 * Get this new node, before the old node is forgotten.
82 #define get_new_node(oldn) get_irn_link(oldn)
85 * Check if a new node was set.
87 #define has_new_node(n) (get_new_node(n) != NULL)
90 * We use the block_visited flag to mark that we have computed the
91 * number of useful predecessors for this block.
92 * Further we encode the new arity in this flag in the old blocks.
93 * Remembering the arity is useful, as it saves a lot of pointer
94 * accesses. This function is called for all Phi and Block nodes
98 compute_new_arity(ir_node *b) {
99 int i, res, irn_arity;
102 irg_v = get_irg_block_visited(current_ir_graph);
103 block_v = get_Block_block_visited(b);
104 if (block_v >= irg_v) {
105 /* we computed the number of preds for this block and saved it in the
107 return block_v - irg_v;
109 /* compute the number of good predecessors */
110 res = irn_arity = get_irn_arity(b);
111 for (i = 0; i < irn_arity; i++)
112 if (is_Bad(get_irn_n(b, i))) res--;
113 /* save it in the flag. */
114 set_Block_block_visited(b, irg_v + res);
120 * Copies the node to the new obstack. The Ins of the new node point to
121 * the predecessors on the old obstack. For block/phi nodes not all
122 * predecessors might be copied. n->link points to the new node.
123 * For Phi and Block nodes the function allocates in-arrays with an arity
124 * only for useful predecessors. The arity is determined by counting
125 * the non-bad predecessors of the block.
127 * @param n The node to be copied
128 * @param env if non-NULL, the node number attribute will be copied to the new node
130 * Note: Also used for loop unrolling.
132 static void copy_node(ir_node *n, void *env) {
135 ir_op *op = get_irn_op(n);
139 /* node copied already */
141 } else if (op == op_Block) {
143 new_arity = compute_new_arity(n);
144 n->attr.block.graph_arr = NULL;
146 block = get_nodes_block(n);
148 new_arity = compute_new_arity(block);
150 new_arity = get_irn_arity(n);
153 nn = new_ir_node(get_irn_dbg_info(n),
160 /* Copy the attributes. These might point to additional data. If this
161 was allocated on the old obstack the pointers now are dangling. This
162 frees e.g. the memory of the graph_arr allocated in new_immBlock. */
163 if (op == op_Block) {
164 /* we cannot allow blocks WITHOUT macroblock input */
165 set_Block_MacroBlock(nn, get_Block_MacroBlock(n));
167 copy_node_attr(n, nn);
171 int copy_node_nr = env != NULL;
173 /* for easier debugging, we want to copy the node numbers too */
174 nn->node_nr = n->node_nr;
180 hook_dead_node_elim_subst(current_ir_graph, n, nn);
184 * Copies new predecessors of old node to new node remembered in link.
185 * Spare the Bad predecessors of Phi and Block nodes.
187 static void copy_preds(ir_node *n, void *env) {
192 nn = get_new_node(n);
195 /* copy the macro block header */
196 ir_node *mbh = get_Block_MacroBlock(n);
199 /* this block is a macroblock header */
200 set_Block_MacroBlock(nn, nn);
202 /* get the macro block header */
203 ir_node *nmbh = get_new_node(mbh);
204 assert(nmbh != NULL);
205 set_Block_MacroBlock(nn, nmbh);
208 /* Don't copy Bad nodes. */
210 irn_arity = get_irn_arity(n);
211 for (i = 0; i < irn_arity; i++) {
212 if (! is_Bad(get_irn_n(n, i))) {
213 ir_node *pred = get_irn_n(n, i);
214 set_irn_n(nn, j, get_new_node(pred));
218 /* repair the block visited flag from above misuse. Repair it in both
219 graphs so that the old one can still be used. */
220 set_Block_block_visited(nn, 0);
221 set_Block_block_visited(n, 0);
222 /* Local optimization could not merge two subsequent blocks if
223 in array contained Bads. Now it's possible.
224 We don't call optimize_in_place as it requires
225 that the fields in ir_graph are set properly. */
226 if ((get_opt_control_flow_straightening()) &&
227 (get_Block_n_cfgpreds(nn) == 1) &&
228 is_Jmp(get_Block_cfgpred(nn, 0))) {
229 ir_node *old = get_nodes_block(get_Block_cfgpred(nn, 0));
231 /* Jmp jumps into the block it is in -- deal self cycle. */
232 assert(is_Bad(get_new_node(get_irg_bad(current_ir_graph))));
233 exchange(nn, get_new_node(get_irg_bad(current_ir_graph)));
238 } else if (is_Phi(n) && get_irn_arity(n) > 0) {
239 /* Don't copy node if corresponding predecessor in block is Bad.
240 The Block itself should not be Bad. */
241 block = get_nodes_block(n);
242 set_nodes_block(nn, get_new_node(block));
244 irn_arity = get_irn_arity(n);
245 for (i = 0; i < irn_arity; i++) {
246 if (! is_Bad(get_irn_n(block, i))) {
247 ir_node *pred = get_irn_n(n, i);
248 set_irn_n(nn, j, get_new_node(pred));
249 /*if (is_backedge(n, i)) set_backedge(nn, j);*/
253 /* If the pre walker reached this Phi after the post walker visited the
254 block block_visited is > 0. */
255 set_Block_block_visited(get_nodes_block(n), 0);
256 /* Compacting the Phi's ins might generate Phis with only one
258 if (get_irn_arity(nn) == 1)
259 exchange(nn, get_irn_n(nn, 0));
261 irn_arity = get_irn_arity(n);
262 for (i = -1; i < irn_arity; i++)
263 set_irn_n(nn, i, get_new_node(get_irn_n(n, i)));
265 /* Now the new node is complete. We can add it to the hash table for CSE.
266 @@@ inlining aborts if we identify End. Why? */
268 add_identities(current_ir_graph->value_table, nn);
272 * Copies the graph recursively, compacts the keep-alives of the end node.
274 * @param irg the graph to be copied
275 * @param copy_node_nr If non-zero, the node number will be copied
277 static void copy_graph(ir_graph *irg, int copy_node_nr) {
278 ir_node *oe, *ne, *ob, *nb, *om, *nm; /* old end, new end, old bad, new bad, old NoMem, new NoMem */
279 ir_node *ka; /* keep alive */
283 /* Some nodes must be copied by hand, sigh */
284 vfl = get_irg_visited(irg);
285 set_irg_visited(irg, vfl + 1);
287 oe = get_irg_end(irg);
288 mark_irn_visited(oe);
289 /* copy the end node by hand, allocate dynamic in array! */
290 ne = new_ir_node(get_irn_dbg_info(oe),
297 /* Copy the attributes. Well, there might be some in the future... */
298 copy_node_attr(oe, ne);
299 set_new_node(oe, ne);
301 /* copy the Bad node */
302 ob = get_irg_bad(irg);
303 mark_irn_visited(ob);
304 nb = new_ir_node(get_irn_dbg_info(ob),
311 copy_node_attr(ob, nb);
312 set_new_node(ob, nb);
314 /* copy the NoMem node */
315 om = get_irg_no_mem(irg);
316 mark_irn_visited(om);
317 nm = new_ir_node(get_irn_dbg_info(om),
324 copy_node_attr(om, nm);
325 set_new_node(om, nm);
327 /* copy the live nodes */
328 set_irg_visited(irg, vfl);
329 irg_walk(get_nodes_block(oe), copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
331 /* Note: from yet, the visited flag of the graph is equal to vfl + 1 */
333 /* visit the anchors as well */
334 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
335 ir_node *n = get_irg_anchor(irg, i);
337 if (n && (get_irn_visited(n) <= vfl)) {
338 set_irg_visited(irg, vfl);
339 irg_walk(n, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
343 /* copy_preds for the end node ... */
344 set_nodes_block(ne, get_new_node(get_nodes_block(oe)));
346 /*- ... and now the keep alives. -*/
347 /* First pick the not marked block nodes and walk them. We must pick these
348 first as else we will oversee blocks reachable from Phis. */
349 irn_arity = get_End_n_keepalives(oe);
350 for (i = 0; i < irn_arity; i++) {
351 ka = get_End_keepalive(oe, i);
353 if (get_irn_visited(ka) <= vfl) {
354 /* We must keep the block alive and copy everything reachable */
355 set_irg_visited(irg, vfl);
356 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
358 add_End_keepalive(ne, get_new_node(ka));
362 /* Now pick other nodes. Here we will keep all! */
363 irn_arity = get_End_n_keepalives(oe);
364 for (i = 0; i < irn_arity; i++) {
365 ka = get_End_keepalive(oe, i);
367 if (get_irn_visited(ka) <= vfl) {
368 /* We didn't copy the node yet. */
369 set_irg_visited(irg, vfl);
370 irg_walk(ka, copy_node, copy_preds, INT_TO_PTR(copy_node_nr));
372 add_End_keepalive(ne, get_new_node(ka));
376 /* start block sometimes only reached after keep alives */
377 set_nodes_block(nb, get_new_node(get_nodes_block(ob)));
378 set_nodes_block(nm, get_new_node(get_nodes_block(om)));
382 * Copies the graph reachable from current_ir_graph->end to the obstack
383 * in current_ir_graph and fixes the environment.
384 * Then fixes the fields in current_ir_graph containing nodes of the
387 * @param copy_node_nr If non-zero, the node number will be copied
390 copy_graph_env(int copy_node_nr) {
391 ir_graph *irg = current_ir_graph;
392 ir_node *old_end, *new_anchor;
395 /* remove end_except and end_reg nodes */
396 old_end = get_irg_end(irg);
397 set_irg_end_except (irg, old_end);
398 set_irg_end_reg (irg, old_end);
400 /* Not all nodes remembered in irg might be reachable
401 from the end node. Assure their link is set to NULL, so that
402 we can test whether new nodes have been computed. */
403 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
404 ir_node *n = get_irg_anchor(irg, i);
406 set_new_node(n, NULL);
408 /* we use the block walk flag for removing Bads from Blocks ins. */
409 inc_irg_block_visited(irg);
412 copy_graph(irg, copy_node_nr);
415 old_end = get_irg_end(irg);
416 new_anchor = new_Anchor(irg);
418 for (i = get_irg_n_anchors(irg) - 1; i >= 0; --i) {
419 ir_node *n = get_irg_anchor(irg, i);
421 set_irn_n(new_anchor, i, get_new_node(n));
424 irg->anchor = new_anchor;
426 /* ensure the new anchor is placed in the endblock */
427 set_nodes_block(new_anchor, get_irg_end_block(irg));
431 * Copies all reachable nodes to a new obstack. Removes bad inputs
432 * from block nodes and the corresponding inputs from Phi nodes.
433 * Merges single exit blocks with single entry blocks and removes
435 * Adds all new nodes to a new hash table for CSE. Does not
436 * perform CSE, so the hash table might contain common subexpressions.
438 void dead_node_elimination(ir_graph *irg) {
440 #ifdef INTERPROCEDURAL_VIEW
441 int rem_ipview = get_interprocedural_view();
443 struct obstack *graveyard_obst = NULL;
444 struct obstack *rebirth_obst = NULL;
446 edges_deactivate(irg);
448 /* inform statistics that we started a dead-node elimination run */
449 hook_dead_node_elim(irg, 1);
451 /* Remember external state of current_ir_graph. */
452 rem = current_ir_graph;
453 current_ir_graph = irg;
454 #ifdef INTERPROCEDURAL_VIEW
455 set_interprocedural_view(0);
458 assert(get_irg_phase_state(irg) != phase_building);
460 /* Handle graph state */
461 free_callee_info(irg);
465 /* @@@ so far we loose loops when copying */
466 free_loop_information(irg);
468 set_irg_doms_inconsistent(irg);
470 /* A quiet place, where the old obstack can rest in peace,
471 until it will be cremated. */
472 graveyard_obst = irg->obst;
474 /* A new obstack, where the reachable nodes will be copied to. */
475 rebirth_obst = XMALLOC(struct obstack);
476 irg->obst = rebirth_obst;
477 obstack_init(irg->obst);
478 irg->last_node_idx = 0;
480 /* We also need a new value table for CSE */
481 del_identities(irg->value_table);
482 irg->value_table = new_identities();
484 /* Copy the graph from the old to the new obstack */
485 copy_graph_env(/*copy_node_nr=*/1);
487 /* Free memory from old unoptimized obstack */
488 obstack_free(graveyard_obst, 0); /* First empty the obstack ... */
489 xfree(graveyard_obst); /* ... then free it. */
491 /* inform statistics that the run is over */
492 hook_dead_node_elim(irg, 0);
494 current_ir_graph = rem;
495 #ifdef INTERPROCEDURAL_VIEW
496 set_interprocedural_view(rem_ipview);
501 * Relink bad predecessors of a block and store the old in array to the
502 * link field. This function is called by relink_bad_predecessors().
503 * The array of link field starts with the block operand at position 0.
504 * If block has bad predecessors, create a new in array without bad preds.
505 * Otherwise let in array untouched.
507 static void relink_bad_block_predecessors(ir_node *n, void *env) {
508 ir_node **new_in, *irn;
509 int i, new_irn_n, old_irn_arity, new_irn_arity = 0;
512 /* if link field of block is NULL, look for bad predecessors otherwise
513 this is already done */
514 if (is_Block(n) && get_irn_link(n) == NULL) {
515 /* save old predecessors in link field (position 0 is the block operand)*/
516 set_irn_link(n, get_irn_in(n));
518 /* count predecessors without bad nodes */
519 old_irn_arity = get_irn_arity(n);
520 for (i = 0; i < old_irn_arity; i++)
521 if (!is_Bad(get_irn_n(n, i)))
524 /* arity changing: set new predecessors without bad nodes */
525 if (new_irn_arity < old_irn_arity) {
526 /* Get new predecessor array. We do not resize the array, as we must
527 keep the old one to update Phis. */
528 new_in = NEW_ARR_D(ir_node *, current_ir_graph->obst, (new_irn_arity+1));
530 /* set new predecessors in array */
533 for (i = 0; i < old_irn_arity; i++) {
534 irn = get_irn_n(n, i);
536 new_in[new_irn_n] = irn;
537 is_backedge(n, i) ? set_backedge(n, new_irn_n-1) : set_not_backedge(n, new_irn_n-1);
541 /* ARR_SETLEN(int, n->attr.block.backedge, new_irn_arity); */
542 ARR_SHRINKLEN(n->attr.block.backedge, new_irn_arity);
544 } /* ir node has bad predecessors */
545 } /* Block is not relinked */
549 * Relinks Bad predecessors from Blocks and Phis called by walker
550 * remove_bad_predecesors(). If n is a Block, call
551 * relink_bad_block_redecessors(). If n is a Phi-node, call also the relinking
552 * function of Phi's Block. If this block has bad predecessors, relink preds
555 static void relink_bad_predecessors(ir_node *n, void *env) {
556 ir_node *block, **old_in;
557 int i, old_irn_arity, new_irn_arity;
559 /* relink bad predecessors of a block */
561 relink_bad_block_predecessors(n, env);
563 /* If Phi node relink its block and its predecessors */
565 /* Relink predecessors of phi's block */
566 block = get_nodes_block(n);
567 if (get_irn_link(block) == NULL)
568 relink_bad_block_predecessors(block, env);
570 old_in = (ir_node **)get_irn_link(block); /* Of Phi's Block */
571 old_irn_arity = ARR_LEN(old_in);
573 /* Relink Phi predecessors if count of predecessors changed */
574 if (old_irn_arity != ARR_LEN(get_irn_in(block))) {
575 /* set new predecessors in array
576 n->in[0] remains the same block */
578 for(i = 1; i < old_irn_arity; i++)
579 if (!is_Bad(old_in[i])) {
580 n->in[new_irn_arity] = n->in[i];
581 is_backedge(n, i) ? set_backedge(n, new_irn_arity) : set_not_backedge(n, new_irn_arity);
585 ARR_SETLEN(ir_node *, n->in, new_irn_arity);
586 ARR_SETLEN(int, n->attr.phi.u.backedge, new_irn_arity);
588 } /* n is a Phi node */
592 * Removes Bad Bad predecessors from Blocks and the corresponding
593 * inputs to Phi nodes as in dead_node_elimination but without
595 * On walking up set the link field to NULL, on walking down call
596 * relink_bad_predecessors() (This function stores the old in array
597 * to the link field and sets a new in array if arity of predecessors
600 void remove_bad_predecessors(ir_graph *irg) {
601 panic("Fix backedge handling first");
602 irg_walk_graph(irg, firm_clear_link, relink_bad_predecessors, NULL);
609 __)|_| | \_/ | \_/(/_ |_/\__|__
611 The following stuff implements a facility that automatically patches
612 registered ir_node pointers to the new node when a dead node elimination occurs.
615 struct _survive_dce_t {
619 hook_entry_t dead_node_elim;
620 hook_entry_t dead_node_elim_subst;
623 typedef struct _survive_dce_list_t {
624 struct _survive_dce_list_t *next;
626 } survive_dce_list_t;
628 static void dead_node_hook(void *context, ir_graph *irg, int start) {
629 survive_dce_t *sd = context;
632 /* Create a new map before the dead node elimination is performed. */
634 sd->new_places = pmap_create_ex(pmap_count(sd->places));
636 /* Patch back all nodes if dead node elimination is over and something is to be done. */
637 pmap_destroy(sd->places);
638 sd->places = sd->new_places;
639 sd->new_places = NULL;
644 * Hook called when dead node elimination replaces old by nw.
646 static void dead_node_subst_hook(void *context, ir_graph *irg, ir_node *old, ir_node *nw) {
647 survive_dce_t *sd = context;
648 survive_dce_list_t *list = pmap_get(sd->places, old);
651 /* If the node is to be patched back, write the new address to all registered locations. */
653 survive_dce_list_t *p;
655 for (p = list; p; p = p->next)
658 pmap_insert(sd->new_places, nw, list);
663 * Make a new Survive DCE environment.
665 survive_dce_t *new_survive_dce(void) {
666 survive_dce_t *res = XMALLOC(survive_dce_t);
667 obstack_init(&res->obst);
668 res->places = pmap_create();
669 res->new_places = NULL;
671 res->dead_node_elim.hook._hook_dead_node_elim = dead_node_hook;
672 res->dead_node_elim.context = res;
673 res->dead_node_elim.next = NULL;
675 res->dead_node_elim_subst.hook._hook_dead_node_elim_subst = dead_node_subst_hook;
676 res->dead_node_elim_subst.context = res;
677 res->dead_node_elim_subst.next = NULL;
679 #ifndef FIRM_ENABLE_HOOKS
680 assert(0 && "need hooks enabled");
683 register_hook(hook_dead_node_elim, &res->dead_node_elim);
684 register_hook(hook_dead_node_elim_subst, &res->dead_node_elim_subst);
689 * Free a Survive DCE environment.
691 void free_survive_dce(survive_dce_t *sd) {
692 obstack_free(&sd->obst, NULL);
693 pmap_destroy(sd->places);
694 unregister_hook(hook_dead_node_elim, &sd->dead_node_elim);
695 unregister_hook(hook_dead_node_elim_subst, &sd->dead_node_elim_subst);
700 * Register a node pointer to be patched upon DCE.
701 * When DCE occurs, the node pointer specified by @p place will be
702 * patched to the new address of the node it is pointing to.
704 * @param sd The Survive DCE environment.
705 * @param place The address of the node pointer.
707 void survive_dce_register_irn(survive_dce_t *sd, ir_node **place) {
708 if (*place != NULL) {
709 ir_node *irn = *place;
710 survive_dce_list_t *curr = pmap_get(sd->places, irn);
711 survive_dce_list_t *nw = obstack_alloc(&sd->obst, sizeof(nw[0]));
716 pmap_insert(sd->places, irn, nw);
720 /*--------------------------------------------------------------------*/
721 /* Functionality for inlining */
722 /*--------------------------------------------------------------------*/
725 * Copy node for inlineing. Updates attributes that change when
726 * inlineing but not for dead node elimination.
728 * Copies the node by calling copy_node() and then updates the entity if
729 * it's a local one. env must be a pointer of the frame type of the
730 * inlined procedure. The new entities must be in the link field of
733 static void copy_node_inline(ir_node *n, void *env) {
735 ir_type *frame_tp = (ir_type *)env;
739 nn = get_new_node (n);
741 if (get_entity_owner(get_Sel_entity(n)) == frame_tp) {
742 set_Sel_entity(nn, get_entity_link(get_Sel_entity(n)));
744 } else if (is_Block(n)) {
745 nn = get_new_node (n);
746 nn->attr.block.irg = current_ir_graph;
751 * Copies new predecessors of old node and move constants to
754 static void copy_preds_inline(ir_node *n, void *env) {
758 nn = skip_Id(get_new_node(n));
759 if (is_irn_constlike(nn)) {
760 /* move Constants into the start block */
761 set_nodes_block(nn, get_irg_start_block(current_ir_graph));
763 n = identify_remember(current_ir_graph->value_table, nn);
772 * Walker: checks if P_value_arg_base is used.
774 static void find_addr(ir_node *node, void *env) {
775 int *allow_inline = env;
777 is_Start(get_Proj_pred(node)) &&
778 get_Proj_proj(node) == pn_Start_P_value_arg_base) {
780 } else if (is_Alloc(node) && get_Alloc_where(node) == stack_alloc) {
782 * Refuse to inline alloca call unless user explicitly forced so as this
783 * may change program's memory overhead drastically when the function
784 * using alloca is called in loop. In GCC present in SPEC2000 inlining
785 * into schedule_block cause it to require 2GB of ram instead of 256MB.
787 * Sorrily this is true with our implementation also.
788 * Moreover, we cannot differentiate between alloca() and VLA yet, so this
789 * disables inlining of functions using VLA (with are completely save).
792 * - add a flag to the Alloc node for "real" alloca() calls
793 * - add a new Stack-Restore node at the end of a function using alloca()
800 * Check if we can inline a given call.
801 * Currently, we cannot inline two cases:
802 * - call with compound arguments
803 * - graphs that take the address of a parameter
805 * check these conditions here
807 static int can_inline(ir_node *call, ir_graph *called_graph) {
808 ir_type *call_type = get_Call_type(call);
809 int params, ress, i, res;
810 assert(is_Method_type(call_type));
812 params = get_method_n_params(call_type);
813 ress = get_method_n_ress(call_type);
815 /* check parameters for compound arguments */
816 for (i = 0; i < params; ++i) {
817 ir_type *p_type = get_method_param_type(call_type, i);
819 if (is_compound_type(p_type))
823 /* check results for compound arguments */
824 for (i = 0; i < ress; ++i) {
825 ir_type *r_type = get_method_res_type(call_type, i);
827 if (is_compound_type(r_type))
832 irg_walk_graph(called_graph, find_addr, NULL, &res);
838 exc_handler = 0, /**< There is a handler. */
839 exc_to_end = 1, /**< Branches to End. */
840 exc_no_handler = 2 /**< Exception handling not represented. */
843 /* Inlines a method at the given call site. */
844 int inline_method(ir_node *call, ir_graph *called_graph) {
846 ir_node *post_call, *post_bl;
847 ir_node *in[pn_Start_max];
848 ir_node *end, *end_bl, *block;
853 int arity, n_ret, n_exc, n_res, i, n, j, rem_opt, irn_arity, n_params;
854 enum exc_mode exc_handling;
855 ir_type *called_frame, *curr_frame, *mtp, *ctp;
858 irg_inline_property prop = get_irg_inline_property(called_graph);
859 unsigned long visited;
861 if (prop == irg_inline_forbidden)
864 ent = get_irg_entity(called_graph);
866 mtp = get_entity_type(ent);
867 ctp = get_Call_type(call);
868 if (get_method_n_params(mtp) > get_method_n_params(ctp)) {
869 /* this is a bad feature of C: without a prototype, we can can call a function with less
870 parameters than needed. Currently we don't support this, although it would be
871 to use Unknown than. */
875 /* Argh, compiling C has some bad consequences:
876 the call type AND the method type might be different.
877 It is implementation defendant what happens in that case.
878 We support inlining, if the bitsize of the types matches AND
879 the same arithmetic is used. */
880 n_params = get_method_n_params(mtp);
881 for (i = n_params - 1; i >= 0; --i) {
882 ir_type *param_tp = get_method_param_type(mtp, i);
883 ir_type *arg_tp = get_method_param_type(ctp, i);
885 if (param_tp != arg_tp) {
886 ir_mode *pmode = get_type_mode(param_tp);
887 ir_mode *amode = get_type_mode(arg_tp);
889 if (pmode == NULL || amode == NULL)
891 if (get_mode_size_bits(pmode) != get_mode_size_bits(amode))
893 if (get_mode_arithmetic(pmode) != get_mode_arithmetic(amode))
895 /* otherwise we can simply "reinterpret" the bits */
899 irg = get_irn_irg(call);
902 * We cannot inline a recursive call. The graph must be copied before
903 * the call the inline_method() using create_irg_copy().
905 if (called_graph == irg)
909 * currently, we cannot inline two cases:
910 * - call with compound arguments
911 * - graphs that take the address of a parameter
913 if (! can_inline(call, called_graph))
916 rem = current_ir_graph;
917 current_ir_graph = irg;
919 DB((dbg, LEVEL_1, "Inlining %+F(%+F) into %+F\n", call, called_graph, irg));
921 /* -- Turn off optimizations, this can cause problems when allocating new nodes. -- */
922 rem_opt = get_opt_optimize();
925 /* Handle graph state */
926 assert(get_irg_phase_state(irg) != phase_building);
927 assert(get_irg_pinned(irg) == op_pin_state_pinned);
928 assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
929 set_irg_outs_inconsistent(irg);
930 set_irg_extblk_inconsistent(irg);
931 set_irg_doms_inconsistent(irg);
932 set_irg_loopinfo_inconsistent(irg);
933 set_irg_callee_info_state(irg, irg_callee_info_inconsistent);
935 /* -- Check preconditions -- */
936 assert(is_Call(call));
938 /* here we know we WILL inline, so inform the statistics */
939 hook_inline(call, called_graph);
941 /* -- Decide how to handle exception control flow: Is there a handler
942 for the Call node, or do we branch directly to End on an exception?
944 0 There is a handler.
946 2 Exception handling not represented in Firm. -- */
948 ir_node *proj, *Mproj = NULL, *Xproj = NULL;
949 for (proj = get_irn_link(call); proj; proj = get_irn_link(proj)) {
950 long proj_nr = get_Proj_proj(proj);
951 if (proj_nr == pn_Call_X_except) Xproj = proj;
952 if (proj_nr == pn_Call_M_except) Mproj = proj;
954 if (Mproj) { assert(Xproj); exc_handling = exc_handler; } /* Mproj */
955 else if (Xproj) { exc_handling = exc_to_end; } /* !Mproj && Xproj */
956 else { exc_handling = exc_no_handler; } /* !Mproj && !Xproj */
959 /* create the argument tuple */
960 NEW_ARR_A(ir_type *, args_in, n_params);
962 block = get_nodes_block(call);
963 for (i = n_params - 1; i >= 0; --i) {
964 ir_node *arg = get_Call_param(call, i);
965 ir_type *param_tp = get_method_param_type(mtp, i);
966 ir_mode *mode = get_type_mode(param_tp);
968 if (mode != get_irn_mode(arg)) {
969 arg = new_r_Conv(irg, block, arg, mode);
975 the procedure and later replaces the Start node of the called graph.
976 Post_call is the old Call node and collects the results of the called
977 graph. Both will end up being a tuple. -- */
978 post_bl = get_nodes_block(call);
979 set_irg_current_block(irg, post_bl);
980 /* XxMxPxPxPxT of Start + parameter of Call */
981 in[pn_Start_X_initial_exec] = new_Jmp();
982 in[pn_Start_M] = get_Call_mem(call);
983 in[pn_Start_P_frame_base] = get_irg_frame(irg);
984 in[pn_Start_P_tls] = get_irg_tls(irg);
985 in[pn_Start_T_args] = new_Tuple(n_params, args_in);
986 /* in[pn_Start_P_value_arg_base] = ??? */
987 assert(pn_Start_P_value_arg_base == pn_Start_max - 1 && "pn_Start_P_value_arg_base not supported, fix");
988 pre_call = new_Tuple(pn_Start_max - 1, in);
992 The new block gets the ins of the old block, pre_call and all its
993 predecessors and all Phi nodes. -- */
994 part_block(pre_call);
996 /* -- Prepare state for dead node elimination -- */
997 /* Visited flags in calling irg must be >= flag in called irg.
998 Else walker and arity computation will not work. */
999 if (get_irg_visited(irg) <= get_irg_visited(called_graph))
1000 set_irg_visited(irg, get_irg_visited(called_graph) + 1);
1001 if (get_irg_block_visited(irg) < get_irg_block_visited(called_graph))
1002 set_irg_block_visited(irg, get_irg_block_visited(called_graph));
1003 visited = get_irg_visited(irg);
1005 /* Set pre_call as new Start node in link field of the start node of
1006 calling graph and pre_calls block as new block for the start block
1008 Further mark these nodes so that they are not visited by the
1010 set_irn_link(get_irg_start(called_graph), pre_call);
1011 set_irn_visited(get_irg_start(called_graph), visited);
1012 set_irn_link(get_irg_start_block(called_graph), get_nodes_block(pre_call));
1013 set_irn_visited(get_irg_start_block(called_graph), visited);
1015 set_irn_link(get_irg_bad(called_graph), get_irg_bad(current_ir_graph));
1016 set_irn_visited(get_irg_bad(called_graph), visited);
1018 set_irn_link(get_irg_no_mem(called_graph), get_irg_no_mem(current_ir_graph));
1019 set_irn_visited(get_irg_no_mem(called_graph), visited);
1021 /* Initialize for compaction of in arrays */
1022 inc_irg_block_visited(irg);
1024 /* -- Replicate local entities of the called_graph -- */
1025 /* copy the entities. */
1026 called_frame = get_irg_frame_type(called_graph);
1027 curr_frame = get_irg_frame_type(irg);
1028 for (i = 0, n = get_class_n_members(called_frame); i < n; ++i) {
1029 ir_entity *new_ent, *old_ent;
1030 old_ent = get_class_member(called_frame, i);
1031 new_ent = copy_entity_own(old_ent, curr_frame);
1032 set_entity_link(old_ent, new_ent);
1035 /* visited is > than that of called graph. With this trick visited will
1036 remain unchanged so that an outer walker, e.g., searching the call nodes
1037 to inline, calling this inline will not visit the inlined nodes. */
1038 set_irg_visited(irg, get_irg_visited(irg)-1);
1040 /* -- Performing dead node elimination inlines the graph -- */
1041 /* Copies the nodes to the obstack of current_ir_graph. Updates links to new
1043 irg_walk(get_irg_end(called_graph), copy_node_inline, copy_preds_inline,
1044 get_irg_frame_type(called_graph));
1046 /* Repair called_graph */
1047 set_irg_visited(called_graph, get_irg_visited(irg));
1048 set_irg_block_visited(called_graph, get_irg_block_visited(irg));
1049 set_Block_block_visited(get_irg_start_block(called_graph), 0);
1051 /* -- Merge the end of the inlined procedure with the call site -- */
1052 /* We will turn the old Call node into a Tuple with the following
1055 0: Phi of all Memories of Return statements.
1056 1: Jmp from new Block that merges the control flow from all exception
1057 predecessors of the old end block.
1058 2: Tuple of all arguments.
1059 3: Phi of Exception memories.
1060 In case the old Call directly branches to End on an exception we don't
1061 need the block merging all exceptions nor the Phi of the exception
1065 /* -- Precompute some values -- */
1066 end_bl = get_new_node(get_irg_end_block(called_graph));
1067 end = get_new_node(get_irg_end(called_graph));
1068 arity = get_Block_n_cfgpreds(end_bl); /* arity = n_exc + n_ret */
1069 n_res = get_method_n_ress(get_Call_type(call));
1071 res_pred = XMALLOCN(ir_node*, n_res);
1072 cf_pred = XMALLOCN(ir_node*, arity);
1074 set_irg_current_block(irg, post_bl); /* just to make sure */
1076 /* -- archive keepalives -- */
1077 irn_arity = get_irn_arity(end);
1078 for (i = 0; i < irn_arity; i++) {
1079 ir_node *ka = get_End_keepalive(end, i);
1081 add_End_keepalive(get_irg_end(irg), ka);
1084 /* The new end node will die. We need not free as the in array is on the obstack:
1085 copy_node() only generated 'D' arrays. */
1087 /* -- Replace Return nodes by Jump nodes. -- */
1089 for (i = 0; i < arity; i++) {
1091 ret = get_Block_cfgpred(end_bl, i);
1092 if (is_Return(ret)) {
1093 cf_pred[n_ret] = new_r_Jmp(irg, get_nodes_block(ret));
1097 set_irn_in(post_bl, n_ret, cf_pred);
1099 /* -- Build a Tuple for all results of the method.
1100 Add Phi node if there was more than one Return. -- */
1101 turn_into_tuple(post_call, pn_Call_max);
1102 /* First the Memory-Phi */
1104 for (i = 0; i < arity; i++) {
1105 ret = get_Block_cfgpred(end_bl, i);
1106 if (is_Return(ret)) {
1107 cf_pred[n_ret] = get_Return_mem(ret);
1111 phi = new_Phi(n_ret, cf_pred, mode_M);
1112 set_Tuple_pred(call, pn_Call_M_regular, phi);
1113 /* Conserve Phi-list for further inlinings -- but might be optimized */
1114 if (get_nodes_block(phi) == post_bl) {
1115 set_irn_link(phi, get_irn_link(post_bl));
1116 set_irn_link(post_bl, phi);
1118 /* Now the real results */
1120 for (j = 0; j < n_res; j++) {
1122 for (i = 0; i < arity; i++) {
1123 ret = get_Block_cfgpred(end_bl, i);
1124 if (is_Return(ret)) {
1125 cf_pred[n_ret] = get_Return_res(ret, j);
1130 phi = new_Phi(n_ret, cf_pred, get_irn_mode(cf_pred[0]));
1134 /* Conserve Phi-list for further inlinings -- but might be optimized */
1135 if (get_nodes_block(phi) == post_bl) {
1136 set_Phi_next(phi, get_Block_phis(post_bl));
1137 set_Block_phis(post_bl, phi);
1140 set_Tuple_pred(call, pn_Call_T_result, new_Tuple(n_res, res_pred));
1142 set_Tuple_pred(call, pn_Call_T_result, new_Bad());
1144 /* handle the regular call */
1145 set_Tuple_pred(call, pn_Call_X_regular, new_Jmp());
1147 /* For now, we cannot inline calls with value_base */
1148 set_Tuple_pred(call, pn_Call_P_value_res_base, new_Bad());
1150 /* Finally the exception control flow.
1151 We have two (three) possible situations:
1152 First if the Call branches to an exception handler: We need to add a Phi node to
1153 collect the memory containing the exception objects. Further we need
1154 to add another block to get a correct representation of this Phi. To
1155 this block we add a Jmp that resolves into the X output of the Call
1156 when the Call is turned into a tuple.
1157 Second the Call branches to End, the exception is not handled. Just
1158 add all inlined exception branches to the End node.
1159 Third: there is no Exception edge at all. Handle as case two. */
1160 if (exc_handling == exc_handler) {
1162 for (i = 0; i < arity; i++) {
1164 ret = get_Block_cfgpred(end_bl, i);
1165 irn = skip_Proj(ret);
1166 if (is_fragile_op(irn) || is_Raise(irn)) {
1167 cf_pred[n_exc] = ret;
1172 new_Block(n_exc, cf_pred); /* watch it: current_block is changed! */
1173 set_Tuple_pred(call, pn_Call_X_except, new_Jmp());
1174 /* The Phi for the memories with the exception objects */
1176 for (i = 0; i < arity; i++) {
1178 ret = skip_Proj(get_Block_cfgpred(end_bl, i));
1180 cf_pred[n_exc] = new_r_Proj(irg, get_nodes_block(ret), ret, mode_M, 3);
1182 } else if (is_fragile_op(ret)) {
1183 /* We rely that all cfops have the memory output at the same position. */
1184 cf_pred[n_exc] = new_r_Proj(irg, get_nodes_block(ret), ret, mode_M, 0);
1186 } else if (is_Raise(ret)) {
1187 cf_pred[n_exc] = new_r_Proj(irg, get_nodes_block(ret), ret, mode_M, 1);
1191 set_Tuple_pred(call, pn_Call_M_except, new_Phi(n_exc, cf_pred, mode_M));
1193 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1194 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1197 ir_node *main_end_bl;
1198 int main_end_bl_arity;
1199 ir_node **end_preds;
1201 /* assert(exc_handling == 1 || no exceptions. ) */
1203 for (i = 0; i < arity; i++) {
1204 ir_node *ret = get_Block_cfgpred(end_bl, i);
1205 ir_node *irn = skip_Proj(ret);
1207 if (is_fragile_op(irn) || is_Raise(irn)) {
1208 cf_pred[n_exc] = ret;
1212 main_end_bl = get_irg_end_block(irg);
1213 main_end_bl_arity = get_irn_arity(main_end_bl);
1214 end_preds = XMALLOCN(ir_node*, n_exc + main_end_bl_arity);
1216 for (i = 0; i < main_end_bl_arity; ++i)
1217 end_preds[i] = get_irn_n(main_end_bl, i);
1218 for (i = 0; i < n_exc; ++i)
1219 end_preds[main_end_bl_arity + i] = cf_pred[i];
1220 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
1221 set_Tuple_pred(call, pn_Call_X_except, new_Bad());
1222 set_Tuple_pred(call, pn_Call_M_except, new_Bad());
1228 /* -- Turn CSE back on. -- */
1229 set_optimize(rem_opt);
1230 current_ir_graph = rem;
1235 /********************************************************************/
1236 /* Apply inlining to small methods. */
1237 /********************************************************************/
1239 static struct obstack temp_obst;
1241 /** Represents a possible inlinable call in a graph. */
1242 typedef struct _call_entry {
1243 ir_node *call; /**< The Call node. */
1244 ir_graph *callee; /**< The callee IR-graph. */
1245 list_head list; /**< List head for linking the next one. */
1246 int loop_depth; /**< The loop depth of this call. */
1247 int benefice; /**< The calculated benefice of this call. */
1248 unsigned local_adr:1; /**< Set if this call gets an address of a local variable. */
1249 unsigned all_const:1; /**< Set if this call has only constant parameters. */
1253 * environment for inlining small irgs
1255 typedef struct _inline_env_t {
1256 struct obstack obst; /**< An obstack where call_entries are allocated on. */
1257 list_head calls; /**< The call entry list. */
1261 * Returns the irg called from a Call node. If the irg is not
1262 * known, NULL is returned.
1264 * @param call the call node
1266 static ir_graph *get_call_called_irg(ir_node *call) {
1269 addr = get_Call_ptr(call);
1270 if (is_Global(addr)) {
1271 ir_entity *ent = get_Global_entity(addr);
1272 return get_entity_irg(ent);
1279 * Walker: Collect all calls to known graphs inside a graph.
1281 static void collect_calls(ir_node *call, void *env) {
1283 if (is_Call(call)) {
1284 ir_graph *called_irg = get_call_called_irg(call);
1286 if (called_irg != NULL) {
1287 /* The Call node calls a locally defined method. Remember to inline. */
1288 inline_env_t *ienv = env;
1289 call_entry *entry = obstack_alloc(&ienv->obst, sizeof(*entry));
1291 entry->callee = called_irg;
1292 entry->loop_depth = 0;
1293 entry->benefice = 0;
1294 entry->local_adr = 0;
1295 entry->all_const = 0;
1297 list_add_tail(&entry->list, &ienv->calls);
1303 * Inlines all small methods at call sites where the called address comes
1304 * from a Const node that references the entity representing the called
1306 * The size argument is a rough measure for the code size of the method:
1307 * Methods where the obstack containing the firm graph is smaller than
1310 void inline_small_irgs(ir_graph *irg, int size) {
1311 ir_graph *rem = current_ir_graph;
1315 current_ir_graph = irg;
1316 /* Handle graph state */
1317 assert(get_irg_phase_state(irg) != phase_building);
1318 free_callee_info(irg);
1320 /* Find Call nodes to inline.
1321 (We can not inline during a walk of the graph, as inlining the same
1322 method several times changes the visited flag of the walked graph:
1323 after the first inlining visited of the callee equals visited of
1324 the caller. With the next inlining both are increased.) */
1325 obstack_init(&env.obst);
1326 INIT_LIST_HEAD(&env.calls);
1327 irg_walk_graph(irg, NULL, collect_calls, &env);
1329 if (! list_empty(&env.calls)) {
1330 /* There are calls to inline */
1331 collect_phiprojs(irg);
1333 list_for_each_entry(call_entry, entry, &env.calls, list) {
1334 ir_graph *callee = entry->callee;
1335 irg_inline_property prop = get_irg_inline_property(callee);
1337 if (prop == irg_inline_forbidden || get_irg_additional_properties(callee) & mtp_property_weak) {
1338 /* do not inline forbidden / weak graphs */
1342 if (prop >= irg_inline_forced ||
1343 _obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst) < size) {
1344 inline_method(entry->call, callee);
1348 obstack_free(&env.obst, NULL);
1349 current_ir_graph = rem;
1353 * Environment for inlining irgs.
1356 list_head calls; /**< List of of all call nodes in this graph. */
1357 unsigned *local_weights; /**< Once allocated, the beneficial weight for transmitting local addresses. */
1358 unsigned n_nodes; /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
1359 unsigned n_blocks; /**< Number of Blocks in graph without Start and End block. */
1360 unsigned n_nodes_orig; /**< for statistics */
1361 unsigned n_call_nodes; /**< Number of Call nodes in the graph. */
1362 unsigned n_call_nodes_orig; /**< for statistics */
1363 unsigned n_callers; /**< Number of known graphs that call this graphs. */
1364 unsigned n_callers_orig; /**< for statistics */
1365 unsigned got_inline:1; /**< Set, if at least one call inside this graph was inlined. */
1366 unsigned local_vars:1; /**< Set, if an inlined function got the address of a local variable. */
1367 unsigned recursive:1; /**< Set, if this function is self recursive. */
1371 * Allocate a new environment for inlining.
1373 static inline_irg_env *alloc_inline_irg_env(void) {
1374 inline_irg_env *env = obstack_alloc(&temp_obst, sizeof(*env));
1375 INIT_LIST_HEAD(&env->calls);
1376 env->local_weights = NULL;
1377 env->n_nodes = -2; /* do not count count Start, End */
1378 env->n_blocks = -2; /* do not count count Start, End Block */
1379 env->n_nodes_orig = -2; /* do not count Start, End */
1380 env->n_call_nodes = 0;
1381 env->n_call_nodes_orig = 0;
1383 env->n_callers_orig = 0;
1384 env->got_inline = 0;
1385 env->local_vars = 0;
1390 typedef struct walker_env {
1391 inline_irg_env *x; /**< the inline environment */
1392 char ignore_runtime; /**< the ignore runtime flag */
1393 char ignore_callers; /**< if set, do change callers data */
1397 * post-walker: collect all calls in the inline-environment
1398 * of a graph and sum some statistics.
1400 static void collect_calls2(ir_node *call, void *ctx) {
1402 inline_irg_env *x = env->x;
1403 ir_opcode code = get_irn_opcode(call);
1407 /* count meaningful nodes in irg */
1408 if (code != iro_Proj && code != iro_Tuple && code != iro_Sync) {
1409 if (code != iro_Block) {
1417 if (code != iro_Call) return;
1419 /* check, if it's a runtime call */
1420 if (env->ignore_runtime) {
1421 ir_node *symc = get_Call_ptr(call);
1423 if (is_Global(symc)) {
1424 ir_entity *ent = get_Global_entity(symc);
1426 if (get_entity_additional_properties(ent) & mtp_property_runtime)
1431 /* collect all call nodes */
1433 ++x->n_call_nodes_orig;
1435 callee = get_call_called_irg(call);
1436 if (callee != NULL) {
1437 if (! env->ignore_callers) {
1438 inline_irg_env *callee_env = get_irg_link(callee);
1439 /* count all static callers */
1440 ++callee_env->n_callers;
1441 ++callee_env->n_callers_orig;
1443 if (callee == current_ir_graph)
1446 /* link it in the list of possible inlinable entries */
1447 entry = obstack_alloc(&temp_obst, sizeof(*entry));
1449 entry->callee = callee;
1450 entry->loop_depth = get_irn_loop(get_nodes_block(call))->depth;
1451 entry->benefice = 0;
1452 entry->local_adr = 0;
1453 entry->all_const = 0;
1455 list_add_tail(&entry->list, &x->calls);
1460 * Returns TRUE if the number of callers is 0 in the irg's environment,
1461 * hence this irg is a leave.
1463 INLINE static int is_leave(ir_graph *irg) {
1464 inline_irg_env *env = get_irg_link(irg);
1465 return env->n_call_nodes == 0;
1469 * Returns TRUE if the number of nodes in the callee is
1470 * smaller then size in the irg's environment.
1472 INLINE static int is_smaller(ir_graph *callee, unsigned size) {
1473 inline_irg_env *env = get_irg_link(callee);
1474 return env->n_nodes < size;
1478 * Duplicate a call entry.
1480 * @param entry the original entry to duplicate
1481 * @param new_call the new call node
1482 * @param loop_depth_delta
1483 * delta value for the loop depth
1485 static call_entry *duplicate_call_entry(const call_entry *entry,
1486 ir_node *new_call, int loop_depth_delta) {
1487 call_entry *nentry = obstack_alloc(&temp_obst, sizeof(*nentry));
1488 nentry->call = new_call;
1489 nentry->callee = entry->callee;
1490 nentry->benefice = entry->benefice;
1491 nentry->loop_depth = entry->loop_depth + loop_depth_delta;
1492 nentry->local_adr = entry->local_adr;
1493 nentry->all_const = entry->all_const;
1499 * Append all call nodes of the source environment to the nodes of in the destination
1502 * @param dst destination environment
1503 * @param src source environment
1504 * @param loop_depth the loop depth of the call that is replaced by the src list
1506 static void append_call_list(inline_irg_env *dst, inline_irg_env *src, int loop_depth) {
1507 call_entry *entry, *nentry;
1509 /* Note that the src list points to Call nodes in the inlined graph, but
1510 we need Call nodes in our graph. Luckily the inliner leaves this information
1511 in the link field. */
1512 list_for_each_entry(call_entry, entry, &src->calls, list) {
1513 nentry = duplicate_call_entry(entry, get_irn_link(entry->call), loop_depth);
1514 list_add_tail(&nentry->list, &dst->calls);
1516 dst->n_call_nodes += src->n_call_nodes;
1517 dst->n_nodes += src->n_nodes;
1521 * Inlines small leave methods at call sites where the called address comes
1522 * from a Const node that references the entity representing the called
1524 * The size argument is a rough measure for the code size of the method:
1525 * Methods where the obstack containing the firm graph is smaller than
1528 void inline_leave_functions(unsigned maxsize, unsigned leavesize,
1529 unsigned size, int ignore_runtime)
1531 inline_irg_env *env;
1537 call_entry *entry, *next;
1538 const call_entry *centry;
1539 pmap *copied_graphs;
1540 pmap_entry *pm_entry;
1542 rem = current_ir_graph;
1543 obstack_init(&temp_obst);
1545 /* a map for the copied graphs, used to inline recursive calls */
1546 copied_graphs = pmap_create();
1548 /* extend all irgs by a temporary data structure for inlining. */
1549 n_irgs = get_irp_n_irgs();
1550 for (i = 0; i < n_irgs; ++i)
1551 set_irg_link(get_irp_irg(i), alloc_inline_irg_env());
1553 /* Pre-compute information in temporary data structure. */
1554 wenv.ignore_runtime = ignore_runtime;
1555 wenv.ignore_callers = 0;
1556 for (i = 0; i < n_irgs; ++i) {
1557 ir_graph *irg = get_irp_irg(i);
1559 assert(get_irg_phase_state(irg) != phase_building);
1560 free_callee_info(irg);
1562 assure_cf_loop(irg);
1563 wenv.x = get_irg_link(irg);
1564 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1567 /* -- and now inline. -- */
1569 /* Inline leaves recursively -- we might construct new leaves. */
1573 for (i = 0; i < n_irgs; ++i) {
1575 int phiproj_computed = 0;
1577 current_ir_graph = get_irp_irg(i);
1578 env = get_irg_link(current_ir_graph);
1580 list_for_each_entry_safe(call_entry, entry, next, &env->calls, list) {
1582 irg_inline_property prop;
1584 if (env->n_nodes > maxsize)
1588 callee = entry->callee;
1590 prop = get_irg_inline_property(callee);
1591 if (prop == irg_inline_forbidden || get_irg_additional_properties(callee) & mtp_property_weak) {
1592 /* do not inline forbidden / weak graphs */
1596 if (is_leave(callee) && (
1597 is_smaller(callee, leavesize) || prop >= irg_inline_forced)) {
1598 if (!phiproj_computed) {
1599 phiproj_computed = 1;
1600 collect_phiprojs(current_ir_graph);
1602 did_inline = inline_method(call, callee);
1605 inline_irg_env *callee_env = get_irg_link(callee);
1607 /* call was inlined, Phi/Projs for current graph must be recomputed */
1608 phiproj_computed = 0;
1610 /* Do some statistics */
1611 env->got_inline = 1;
1612 --env->n_call_nodes;
1613 env->n_nodes += callee_env->n_nodes;
1614 --callee_env->n_callers;
1616 /* remove this call from the list */
1617 list_del(&entry->list);
1623 } while (did_inline);
1625 /* inline other small functions. */
1626 for (i = 0; i < n_irgs; ++i) {
1628 int phiproj_computed = 0;
1630 current_ir_graph = get_irp_irg(i);
1631 env = get_irg_link(current_ir_graph);
1633 /* note that the list of possible calls is updated during the process */
1634 list_for_each_entry_safe(call_entry, entry, next, &env->calls, list) {
1635 irg_inline_property prop;
1640 callee = entry->callee;
1642 prop = get_irg_inline_property(callee);
1643 if (prop == irg_inline_forbidden || get_irg_additional_properties(callee) & mtp_property_weak) {
1644 /* do not inline forbidden / weak graphs */
1648 e = pmap_find(copied_graphs, callee);
1651 * Remap callee if we have a copy.
1652 * FIXME: Should we do this only for recursive Calls ?
1657 if (prop >= irg_inline_forced ||
1658 (is_smaller(callee, size) && env->n_nodes < maxsize) /* small function */) {
1659 if (current_ir_graph == callee) {
1661 * Recursive call: we cannot directly inline because we cannot walk
1662 * the graph and change it. So we have to make a copy of the graph
1666 inline_irg_env *callee_env;
1670 * No copy yet, create one.
1671 * Note that recursive methods are never leaves, so it is sufficient
1672 * to test this condition here.
1674 copy = create_irg_copy(callee);
1676 /* create_irg_copy() destroys the Proj links, recompute them */
1677 phiproj_computed = 0;
1679 /* allocate new environment */
1680 callee_env = alloc_inline_irg_env();
1681 set_irg_link(copy, callee_env);
1683 assure_cf_loop(copy);
1684 wenv.x = callee_env;
1685 wenv.ignore_callers = 1;
1686 irg_walk_graph(copy, NULL, collect_calls2, &wenv);
1689 * Enter the entity of the original graph. This is needed
1690 * for inline_method(). However, note that ent->irg still points
1691 * to callee, NOT to copy.
1693 set_irg_entity(copy, get_irg_entity(callee));
1695 pmap_insert(copied_graphs, callee, copy);
1698 /* we have only one caller: the original graph */
1699 callee_env->n_callers = 1;
1700 callee_env->n_callers_orig = 1;
1702 if (! phiproj_computed) {
1703 phiproj_computed = 1;
1704 collect_phiprojs(current_ir_graph);
1706 did_inline = inline_method(call, callee);
1708 inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1710 /* call was inlined, Phi/Projs for current graph must be recomputed */
1711 phiproj_computed = 0;
1713 /* callee was inline. Append it's call list. */
1714 env->got_inline = 1;
1715 --env->n_call_nodes;
1716 append_call_list(env, callee_env, entry->loop_depth);
1717 --callee_env->n_callers;
1719 /* after we have inlined callee, all called methods inside callee
1720 are now called once more */
1721 list_for_each_entry(call_entry, centry, &callee_env->calls, list) {
1722 inline_irg_env *penv = get_irg_link(centry->callee);
1726 /* remove this call from the list */
1727 list_del(&entry->list);
1734 for (i = 0; i < n_irgs; ++i) {
1735 irg = get_irp_irg(i);
1736 env = get_irg_link(irg);
1738 if (env->got_inline) {
1739 optimize_graph_df(irg);
1742 if (env->got_inline || (env->n_callers_orig != env->n_callers)) {
1743 DB((dbg, LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1744 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1745 env->n_callers_orig, env->n_callers,
1746 get_entity_name(get_irg_entity(irg))));
1750 /* kill the copied graphs: we don't need them anymore */
1751 foreach_pmap(copied_graphs, pm_entry) {
1752 ir_graph *copy = pm_entry->value;
1754 /* reset the entity, otherwise it will be deleted in the next step ... */
1755 set_irg_entity(copy, NULL);
1756 free_ir_graph(copy);
1758 pmap_destroy(copied_graphs);
1760 obstack_free(&temp_obst, NULL);
1761 current_ir_graph = rem;
1765 * Calculate the parameter weights for transmitting the address of a local variable.
1767 static unsigned calc_method_local_weight(ir_node *arg) {
1769 unsigned v, weight = 0;
1771 for (i = get_irn_n_outs(arg) - 1; i >= 0; --i) {
1772 ir_node *succ = get_irn_out(arg, i);
1774 switch (get_irn_opcode(succ)) {
1777 /* Loads and Store can be removed */
1781 /* check if all args are constant */
1782 for (j = get_Sel_n_indexs(succ) - 1; j >= 0; --j) {
1783 ir_node *idx = get_Sel_index(succ, j);
1784 if (! is_Const(idx))
1787 /* Check users on this Sel. Note: if a 0 is returned here, there was
1788 some unsupported node. */
1789 v = calc_method_local_weight(succ);
1792 /* we can kill one Sel with constant indexes, this is cheap */
1796 /* when looking backward we might find Id nodes */
1797 weight += calc_method_local_weight(succ);
1800 /* unoptimized tuple */
1801 for (j = get_Tuple_n_preds(succ) - 1; j >= 0; --j) {
1802 ir_node *pred = get_Tuple_pred(succ, j);
1804 /* look for Proj(j) */
1805 for (k = get_irn_n_outs(succ) - 1; k >= 0; --k) {
1806 ir_node *succ_succ = get_irn_out(succ, k);
1807 if (is_Proj(succ_succ)) {
1808 if (get_Proj_proj(succ_succ) == j) {
1810 weight += calc_method_local_weight(succ_succ);
1813 /* this should NOT happen */
1821 /* any other node: unsupported yet or bad. */
1829 * Calculate the parameter weights for transmitting the address of a local variable.
1831 static void analyze_irg_local_weights(inline_irg_env *env, ir_graph *irg) {
1832 ir_entity *ent = get_irg_entity(irg);
1834 int nparams, i, proj_nr;
1835 ir_node *irg_args, *arg;
1837 mtp = get_entity_type(ent);
1838 nparams = get_method_n_params(mtp);
1840 /* allocate a new array. currently used as 'analysed' flag */
1841 env->local_weights = NEW_ARR_D(unsigned, &temp_obst, nparams);
1843 /* If the method haven't parameters we have nothing to do. */
1847 assure_irg_outs(irg);
1848 irg_args = get_irg_args(irg);
1849 for (i = get_irn_n_outs(irg_args) - 1; i >= 0; --i) {
1850 arg = get_irn_out(irg_args, i);
1851 proj_nr = get_Proj_proj(arg);
1852 env->local_weights[proj_nr] = calc_method_local_weight(arg);
1857 * Calculate the benefice for transmitting an local variable address.
1858 * After inlining, the local variable might be transformed into a
1859 * SSA variable by scalar_replacement().
1861 static unsigned get_method_local_adress_weight(ir_graph *callee, int pos) {
1862 inline_irg_env *env = get_irg_link(callee);
1864 if (env->local_weights != NULL) {
1865 if (pos < ARR_LEN(env->local_weights))
1866 return env->local_weights[pos];
1870 analyze_irg_local_weights(env, callee);
1872 if (pos < ARR_LEN(env->local_weights))
1873 return env->local_weights[pos];
1878 * Calculate a benefice value for inlining the given call.
1880 * @param call the call node we have to inspect
1881 * @param callee the called graph
1883 static int calc_inline_benefice(call_entry *entry, ir_graph *callee)
1885 ir_node *call = entry->call;
1886 ir_entity *ent = get_irg_entity(callee);
1890 int i, n_params, all_const;
1892 irg_inline_property prop;
1894 inline_irg_env *callee_env;
1896 prop = get_irg_inline_property(callee);
1897 if (prop == irg_inline_forbidden) {
1898 DB((dbg, LEVEL_2, "In %+F Call to %+F: inlining forbidden\n",
1900 return entry->benefice = INT_MIN;
1903 if (get_irg_additional_properties(callee) & (mtp_property_noreturn | mtp_property_weak)) {
1904 DB((dbg, LEVEL_2, "In %+F Call to %+F: not inlining noreturn or weak\n",
1906 return entry->benefice = INT_MIN;
1909 /* costs for every passed parameter */
1910 n_params = get_Call_n_params(call);
1911 mtp = get_entity_type(ent);
1912 cc = get_method_calling_convention(mtp);
1913 if (cc & cc_reg_param) {
1914 /* register parameter, smaller costs for register parameters */
1915 int max_regs = cc & ~cc_bits;
1917 if (max_regs < n_params)
1918 weight += max_regs * 2 + (n_params - max_regs) * 5;
1920 weight += n_params * 2;
1922 /* parameters are passed an stack */
1923 weight += 5 * n_params;
1926 /* constant parameters improve the benefice */
1927 frame_ptr = get_irg_frame(current_ir_graph);
1929 for (i = 0; i < n_params; ++i) {
1930 ir_node *param = get_Call_param(call, i);
1932 if (is_Const(param)) {
1933 weight += get_method_param_weight(ent, i);
1936 if (is_SymConst(param))
1937 weight += get_method_param_weight(ent, i);
1938 else if (is_Sel(param) && get_Sel_ptr(param) == frame_ptr) {
1940 * An address of a local variable is transmitted. After
1941 * inlining, scalar_replacement might be able to remove the
1942 * local variable, so honor this.
1944 v = get_method_local_adress_weight(callee, i);
1947 entry->local_adr = 1;
1951 entry->all_const = all_const;
1953 callee_env = get_irg_link(callee);
1954 if (callee_env->n_callers == 1 &&
1955 callee != current_ir_graph &&
1956 get_entity_visibility(ent) == visibility_local) {
1960 /* give a bonus for functions with one block */
1961 if (callee_env->n_blocks == 1)
1962 weight = weight * 3 / 2;
1964 /* and one for small non-recursive functions: we want them to be inlined in mostly every case */
1965 if (callee_env->n_nodes < 30 && !callee_env->recursive)
1968 /* and finally for leaves: they do not increase the register pressure
1969 because of callee safe registers */
1970 if (callee_env->n_call_nodes == 0)
1973 /** it's important to inline inner loops first */
1974 if (entry->loop_depth > 30)
1975 weight += 30 * 1024;
1977 weight += entry->loop_depth * 1024;
1980 * All arguments constant is probably a good sign, give an extra bonus
1985 return entry->benefice = weight;
1988 static ir_graph **irgs;
1989 static int last_irg;
1992 * Callgraph walker, collect all visited graphs.
1994 static void callgraph_walker(ir_graph *irg, void *data) {
1996 irgs[last_irg++] = irg;
2000 * Creates an inline order for all graphs.
2002 * @return the list of graphs.
2004 static ir_graph **create_irg_list(void) {
2005 ir_entity **free_methods;
2007 int n_irgs = get_irp_n_irgs();
2009 cgana(&arr_len, &free_methods);
2010 xfree(free_methods);
2012 compute_callgraph();
2015 irgs = XMALLOCNZ(ir_graph*, n_irgs);
2017 callgraph_walk(NULL, callgraph_walker, NULL);
2018 assert(n_irgs == last_irg);
2024 * Push a call onto the priority list if its benefice is big enough.
2026 * @param pqueue the priority queue of calls
2027 * @param call the call entry
2028 * @param inlien_threshold
2029 * the threshold value
2031 static void maybe_push_call(pqueue_t *pqueue, call_entry *call,
2032 int inline_threshold)
2034 ir_graph *callee = call->callee;
2035 irg_inline_property prop = get_irg_inline_property(callee);
2036 int benefice = calc_inline_benefice(call, callee);
2038 DB((dbg, LEVEL_2, "In %+F Call %+F to %+F has benefice %d\n",
2039 get_irn_irg(call->call), call->call, callee, benefice));
2041 if (prop < irg_inline_forced && benefice < inline_threshold) {
2045 pqueue_put(pqueue, call, benefice);
2049 * Try to inline calls into a graph.
2051 * @param irg the graph into which we inline
2052 * @param maxsize do NOT inline if the size of irg gets
2053 * bigger than this amount
2054 * @param inline_threshold
2055 * threshold value for inline decision
2056 * @param copied_graphs
2057 * map containing copied of recursive graphs
2059 static void inline_into(ir_graph *irg, unsigned maxsize,
2060 int inline_threshold, pmap *copied_graphs)
2062 int phiproj_computed = 0;
2063 inline_irg_env *env = get_irg_link(irg);
2064 call_entry *curr_call;
2068 if (env->n_call_nodes == 0)
2071 if (env->n_nodes > maxsize) {
2072 DB((dbg, LEVEL_2, "%+F: too big (%d)\n", irg, env->n_nodes));
2076 current_ir_graph = irg;
2078 /* put irgs into the pqueue */
2079 pqueue = new_pqueue();
2081 list_for_each_entry(call_entry, curr_call, &env->calls, list) {
2082 assert(is_Call(curr_call->call));
2083 maybe_push_call(pqueue, curr_call, inline_threshold);
2086 /* note that the list of possible calls is updated during the process */
2087 while (!pqueue_empty(pqueue)) {
2089 call_entry *curr_call = pqueue_pop_front(pqueue);
2090 ir_graph *callee = curr_call->callee;
2091 ir_node *call_node = curr_call->call;
2092 inline_irg_env *callee_env = get_irg_link(callee);
2093 irg_inline_property prop = get_irg_inline_property(callee);
2095 const call_entry *centry;
2098 if ((prop < irg_inline_forced) && env->n_nodes + callee_env->n_nodes > maxsize) {
2099 DB((dbg, LEVEL_2, "%+F: too big (%d) + %+F (%d)\n", irg,
2100 env->n_nodes, callee, callee_env->n_nodes));
2104 e = pmap_find(copied_graphs, callee);
2106 int benefice = curr_call->benefice;
2108 * Reduce the weight for recursive function IFF not all arguments are const.
2109 * inlining recursive functions is rarely good.
2111 if (!curr_call->all_const)
2113 if (benefice < inline_threshold)
2117 * Remap callee if we have a copy.
2120 callee_env = get_irg_link(callee);
2123 if (current_ir_graph == callee) {
2125 * Recursive call: we cannot directly inline because we cannot
2126 * walk the graph and change it. So we have to make a copy of
2129 int benefice = curr_call->benefice;
2133 * Reduce the weight for recursive function IFF not all arguments are const.
2134 * inlining recursive functions is rarely good.
2136 if (!curr_call->all_const)
2138 if (benefice < inline_threshold)
2142 * No copy yet, create one.
2143 * Note that recursive methods are never leaves, so it is
2144 * sufficient to test this condition here.
2146 copy = create_irg_copy(callee);
2148 /* create_irg_copy() destroys the Proj links, recompute them */
2149 phiproj_computed = 0;
2151 /* allocate a new environment */
2152 callee_env = alloc_inline_irg_env();
2153 set_irg_link(copy, callee_env);
2155 assure_cf_loop(copy);
2156 wenv.x = callee_env;
2157 wenv.ignore_callers = 1;
2158 irg_walk_graph(copy, NULL, collect_calls2, &wenv);
2161 * Enter the entity of the original graph. This is needed
2162 * for inline_method(). However, note that ent->irg still points
2163 * to callee, NOT to copy.
2165 set_irg_entity(copy, get_irg_entity(callee));
2167 pmap_insert(copied_graphs, callee, copy);
2170 /* we have only one caller: the original graph */
2171 callee_env->n_callers = 1;
2172 callee_env->n_callers_orig = 1;
2174 if (! phiproj_computed) {
2175 phiproj_computed = 1;
2176 collect_phiprojs(current_ir_graph);
2178 did_inline = inline_method(call_node, callee);
2182 /* call was inlined, Phi/Projs for current graph must be recomputed */
2183 phiproj_computed = 0;
2185 /* remove it from the caller list */
2186 list_del(&curr_call->list);
2188 /* callee was inline. Append it's call list. */
2189 env->got_inline = 1;
2190 if (curr_call->local_adr)
2191 env->local_vars = 1;
2192 --env->n_call_nodes;
2194 /* we just generate a bunch of new calls */
2195 loop_depth = curr_call->loop_depth;
2196 list_for_each_entry(call_entry, centry, &callee_env->calls, list) {
2197 inline_irg_env *penv = get_irg_link(centry->callee);
2199 call_entry *new_entry;
2201 /* after we have inlined callee, all called methods inside
2202 * callee are now called once more */
2205 /* Note that the src list points to Call nodes in the inlined graph,
2206 * but we need Call nodes in our graph. Luckily the inliner leaves
2207 * this information in the link field. */
2208 new_call = get_irn_link(centry->call);
2209 assert(is_Call(new_call));
2211 new_entry = duplicate_call_entry(centry, new_call, loop_depth);
2212 list_add_tail(&new_entry->list, &env->calls);
2213 maybe_push_call(pqueue, new_entry, inline_threshold);
2216 env->n_call_nodes += callee_env->n_call_nodes;
2217 env->n_nodes += callee_env->n_nodes;
2218 --callee_env->n_callers;
2225 * Heuristic inliner. Calculates a benefice value for every call and inlines
2226 * those calls with a value higher than the threshold.
2228 void inline_functions(unsigned maxsize, int inline_threshold) {
2229 inline_irg_env *env;
2233 pmap *copied_graphs;
2234 pmap_entry *pm_entry;
2237 rem = current_ir_graph;
2238 obstack_init(&temp_obst);
2240 irgs = create_irg_list();
2242 /* a map for the copied graphs, used to inline recursive calls */
2243 copied_graphs = pmap_create();
2245 /* extend all irgs by a temporary data structure for inlining. */
2246 n_irgs = get_irp_n_irgs();
2247 for (i = 0; i < n_irgs; ++i)
2248 set_irg_link(irgs[i], alloc_inline_irg_env());
2250 /* Pre-compute information in temporary data structure. */
2251 wenv.ignore_runtime = 0;
2252 wenv.ignore_callers = 0;
2253 for (i = 0; i < n_irgs; ++i) {
2254 ir_graph *irg = irgs[i];
2256 free_callee_info(irg);
2258 wenv.x = get_irg_link(irg);
2259 assure_cf_loop(irg);
2260 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
2263 /* -- and now inline. -- */
2264 for (i = 0; i < n_irgs; ++i) {
2265 ir_graph *irg = irgs[i];
2267 inline_into(irg, maxsize, inline_threshold, copied_graphs);
2270 for (i = 0; i < n_irgs; ++i) {
2271 ir_graph *irg = irgs[i];
2273 env = get_irg_link(irg);
2274 if (env->got_inline) {
2275 /* this irg got calls inlined: optimize it */
2276 if (get_opt_combo()) {
2277 if (env->local_vars) {
2278 scalar_replacement_opt(irg);
2282 if (env->local_vars) {
2283 if (scalar_replacement_opt(irg)) {
2284 optimize_graph_df(irg);
2290 if (env->got_inline || (env->n_callers_orig != env->n_callers)) {
2291 DB((dbg, LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
2292 env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
2293 env->n_callers_orig, env->n_callers,
2294 get_entity_name(get_irg_entity(irg))));
2298 /* kill the copied graphs: we don't need them anymore */
2299 foreach_pmap(copied_graphs, pm_entry) {
2300 ir_graph *copy = pm_entry->value;
2302 /* reset the entity, otherwise it will be deleted in the next step ... */
2303 set_irg_entity(copy, NULL);
2304 free_ir_graph(copy);
2306 pmap_destroy(copied_graphs);
2310 obstack_free(&temp_obst, NULL);
2311 current_ir_graph = rem;
2314 void firm_init_inline(void) {
2315 FIRM_DBG_REGISTER(dbg, "firm.opt.inline");