2 * Copyright (C) 1995-2010 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 * @author Christian Helmer
23 * @brief loop inversion and loop unrolling
30 #include "iroptimize.h"
49 #include "irbackedge_t.h"
50 #include "irphase_t.h"
54 DEBUG_ONLY(static firm_dbg_module_t *dbg);
56 /* DBG print stats for every procedure. */
57 #define LOOP_OPT_STATS 1
59 /* DBG: Ignore node limits and process every possible loop. */
60 #define LOOP_IGNORE_NODE_LIMITS 0
63 * Convenience macro for iterating over every phi node of the given block.
64 * Requires phi list per block.
66 #define for_each_phi(block, phi) \
67 for ((phi) = get_Block_phis( (block) ); (phi) ; (phi) = get_Phi_next((phi)))
69 #define for_each_phi_safe(head, phi, next) \
70 for ((phi) = (head), (next) = (head) ? get_Phi_next((head)) : NULL; \
71 (phi) ; (phi) = (next), (next) = (next) ? get_Phi_next((next)) : NULL)
73 /* Currently processed loop. */
74 static ir_loop *cur_loop;
76 /* Flag for kind of unrolling. */
80 } unrolling_kind_flag;
82 /* Condition for performing visiting a node during copy_walk. */
83 typedef unsigned walker_condition(ir_node *);
85 /* Node and position of a predecessor. */
86 typedef struct entry_edge {
92 /* Node info for unrolling. */
93 typedef struct unrolling_node_info {
95 } unrolling_node_info;
97 /* Outs of the nodes head. */
98 static entry_edge *cur_head_outs;
100 /* Information about the loop head */
101 static ir_node *loop_head = NULL;
102 static unsigned loop_head_valid = 1;
104 /* List of all inner loops, that are processed. */
105 static ir_loop **loops;
109 #define count_stats(val) (++val)
110 #define print_stats() (do_print_stats())
111 #define reset_stats() (do_reset_stats())
114 typedef struct loop_stats_t {
118 unsigned too_large_adapted;
119 unsigned cc_limit_reached;
120 unsigned calls_limit;
122 unsigned u_simple_counting_loop;
123 unsigned constant_unroll;
124 unsigned invariant_unroll;
129 static loop_stats_t stats;
131 /* Set stats to sero */
132 static void do_reset_stats(void)
134 memset(&stats, 0, sizeof(loop_stats_t));
138 static void do_print_stats(void)
140 DB((dbg, LEVEL_2, "---------------------------------------\n"));
141 DB((dbg, LEVEL_2, "loops : %d\n",stats.loops));
142 DB((dbg, LEVEL_2, "inverted : %d\n",stats.inverted));
143 DB((dbg, LEVEL_2, "too_large : %d\n",stats.too_large));
144 DB((dbg, LEVEL_2, "too_large_adapted : %d\n",stats.too_large_adapted));
145 DB((dbg, LEVEL_2, "cc_limit_reached : %d\n",stats.cc_limit_reached));
146 DB((dbg, LEVEL_2, "calls_limit : %d\n",stats.calls_limit));
147 DB((dbg, LEVEL_2, "u_simple_counting : %d\n",stats.u_simple_counting_loop));
148 DB((dbg, LEVEL_2, "constant_unroll : %d\n",stats.constant_unroll));
149 DB((dbg, LEVEL_2, "invariant_unroll : %d\n",stats.invariant_unroll));
150 DB((dbg, LEVEL_2, "=======================================\n"));
154 #define count_stats(val) ((void)0)
155 #define print_stats() ((void)0)
156 #define reset_stats() ((void)0)
160 /* Commandline parameters */
161 typedef struct loop_opt_params_t {
162 unsigned max_loop_size; /* Maximum number of nodes [nodes]*/
163 int depth_adaption; /* Loop nest depth adaption [percent] */
164 unsigned allowed_calls; /* Number of calls allowed [number] */
165 unsigned count_phi:1; /* Count phi nodes */
166 unsigned count_proj:1; /* Count projections */
168 unsigned max_cc_size; /* Maximum condition chain size [nodes] */
169 unsigned max_branches;
171 unsigned max_unrolled_loop_size; /* [nodes] */
172 unsigned allow_const_unrolling:1;
173 unsigned allow_invar_unrolling:1;
174 unsigned invar_unrolling_min_size; /* [nodes] */
178 static loop_opt_params_t opt_params;
180 /* Loop analysis informations */
181 typedef struct loop_info_t {
182 unsigned nodes; /* node count */
183 unsigned ld_st; /* load and store nodes */
184 unsigned branches; /* number of conditions */
185 unsigned calls; /* number of calls */
186 unsigned cf_outs; /* number of cf edges which leave the loop */
187 entry_edge cf_out; /* single loop leaving cf edge */
188 int be_src_pos; /* position of the single own backedge in the head */
191 unsigned cc_size; /* nodes in the condition chain */
194 unsigned max_unroll; /* Number of unrolls satisfying max_loop_size */
195 unsigned exit_cond; /* 1 if condition==true exits the loop. */
196 unsigned latest_value:1; /* 1 if condition is checked against latest counter value */
197 unsigned needs_backedge:1; /* 0 if loop is completely unrolled */
198 unsigned decreasing:1; /* Step operation is_Sub, or step is<0 */
200 /* IV informations of a simple loop */
204 ir_node *iteration_phi;
207 ir_tarval *count_tar; /* Number of loop iterations */
209 ir_node *duff_cond; /* Duff mod */
210 unrolling_kind_flag unroll_kind; /* constant or invariant unrolling */
213 /* Information about the current loop */
214 static loop_info_t loop_info;
216 /* Outs of the condition chain (loop inversion). */
217 static ir_node **cc_blocks;
218 /* df/cf edges with def in the condition chain */
219 static entry_edge *cond_chain_entries;
220 /* Array of df loops found in the condition chain. */
221 static entry_edge *head_df_loop;
222 /* Number of blocks in cc */
223 static unsigned inversion_blocks_in_cc;
226 /* Cf/df edges leaving the loop.
227 * Called entries here, as they are used to enter the loop with walkers. */
228 static entry_edge *loop_entries;
229 /* Number of unrolls to perform */
230 static int unroll_nr;
231 /* Phase is used to keep copies of nodes. */
232 static ir_phase *phase;
234 /* Loop operations. */
235 typedef enum loop_op_t {
241 /* Saves which loop operation to do until after basic tests. */
242 static loop_op_t loop_op;
244 /* Returns the maximum nodes for the given nest depth */
245 static unsigned get_max_nodes_adapted(unsigned depth)
247 double perc = 100.0 + (double)opt_params.depth_adaption;
248 double factor = pow(perc / 100.0, depth);
250 return (int)((double)opt_params.max_loop_size * factor);
253 /* Reset nodes link. For use with a walker. */
254 static void reset_link(ir_node *node, void *env)
257 set_irn_link(node, NULL);
260 /* Returns 0 if the node or block is not in cur_loop. */
261 static unsigned is_in_loop(ir_node *node)
263 return (get_irn_loop(get_block(node)) == cur_loop);
266 /* Returns 0 if the given edge is not a backedge
267 * with its pred in the cur_loop. */
268 static unsigned is_own_backedge(ir_node *n, int pos)
270 return (is_backedge(n, pos) && is_in_loop(get_irn_n(n, pos)));
273 /* Finds loop head and some loop_info as calls or else if necessary. */
274 static void get_loop_info(ir_node *node, void *env)
276 unsigned node_in_loop, pred_in_loop;
280 arity = get_irn_arity(node);
281 for (i = 0; i < arity; i++) {
282 ir_node *pred = get_irn_n(node, i);
284 pred_in_loop = is_in_loop(pred);
285 node_in_loop = is_in_loop(node);
287 if (!node_in_loop && pred_in_loop && is_Block(node))
295 loop_info.cf_out = entry;
298 /* collect some loop information */
300 if (is_Phi(node) && opt_params.count_phi)
302 else if (is_Proj(node) && opt_params.count_proj)
304 else if (!is_Confirm(node) && !is_Const(node) && !is_SymConst(node))
307 if (is_Load(node) || is_Store(node))
315 /* Find the loops head/the blocks with cfpred outside of the loop */
316 if (is_Block(node)) {
317 const ir_edge_t *edge;
320 /* Count innerloop branches */
321 foreach_out_edge_kind(node, edge, EDGE_KIND_NORMAL) {
322 if (is_Block(get_edge_src_irn(edge)) && is_in_loop(get_edge_src_irn(edge)))
326 ++loop_info.branches;
328 if (node_in_loop && !pred_in_loop && loop_head_valid) {
329 ir_node *cfgpred = get_Block_cfgpred(node, i);
331 if (!is_in_loop(cfgpred)) {
332 DB((dbg, LEVEL_5, "potential head %+F because inloop and pred %+F not inloop\n",
334 /* another head? We do not touch this. */
335 if (loop_head && loop_head != node) {
346 /* Finds all edges with users outside of the loop
347 * and definition inside the loop. */
348 static void get_loop_entries(ir_node *node, void *env)
350 unsigned node_in_loop, pred_in_loop;
354 arity = get_irn_arity(node);
355 for (i = 0; i < arity; ++i) {
356 ir_node *pred = get_irn_n(node, i);
358 pred_in_loop = is_in_loop(pred);
359 node_in_loop = is_in_loop(node);
361 if (pred_in_loop && !node_in_loop) {
366 ARR_APP1(entry_edge, loop_entries, entry);
372 static ir_node *ssa_second_def;
373 static ir_node *ssa_second_def_block;
376 * Walks the graph bottom up, searching for definitions and creates phis.
378 static ir_node *search_def_and_create_phis(ir_node *block, ir_mode *mode, int first)
386 DB((dbg, LEVEL_5, "ssa search_def_and_create_phis: block %N\n", block));
388 /* Prevents creation of phi that would be bad anyway.
389 * Dead and bad blocks. */
390 if (get_irn_arity(block) < 1 || is_Bad(block)) {
391 DB((dbg, LEVEL_5, "ssa bad %N\n", block));
395 if (block == ssa_second_def_block && !first) {
396 DB((dbg, LEVEL_5, "ssa found second definition: use second def %N\n", ssa_second_def));
397 return ssa_second_def;
400 /* already processed this block? */
401 if (irn_visited(block)) {
402 ir_node *value = (ir_node *) get_irn_link(block);
403 DB((dbg, LEVEL_5, "ssa already visited: use linked %N\n", value));
407 irg = get_irn_irg(block);
408 assert(block != get_irg_start_block(irg));
410 /* a Block with only 1 predecessor needs no Phi */
411 n_cfgpreds = get_Block_n_cfgpreds(block);
412 if (n_cfgpreds == 1) {
413 ir_node *pred_block = get_Block_cfgpred_block(block, 0);
416 DB((dbg, LEVEL_5, "ssa 1 pred: walk pred %N\n", pred_block));
418 value = search_def_and_create_phis(pred_block, mode, 0);
419 set_irn_link(block, value);
420 mark_irn_visited(block);
425 /* create a new Phi */
426 NEW_ARR_A(ir_node*, in, n_cfgpreds);
427 for (i = 0; i < n_cfgpreds; ++i)
428 in[i] = new_Unknown(mode);
430 phi = new_r_Phi(block, n_cfgpreds, in, mode);
431 /* Important: always keep block phi list up to date. */
432 add_Block_phi(block, phi);
433 DB((dbg, LEVEL_5, "ssa phi creation: link new phi %N to block %N\n", phi, block));
434 set_irn_link(block, phi);
435 mark_irn_visited(block);
437 /* set Phi predecessors */
438 for (i = 0; i < n_cfgpreds; ++i) {
440 ir_node *pred_block = get_Block_cfgpred_block(block, i);
441 assert(pred_block != NULL);
442 pred_val = search_def_and_create_phis(pred_block, mode, 0);
444 assert(pred_val != NULL);
446 DB((dbg, LEVEL_5, "ssa phi pred:phi %N, pred %N\n", phi, pred_val));
447 set_irn_n(phi, i, pred_val);
455 * Given a set of values this function constructs SSA-form for the users of the
456 * first value (the users are determined through the out-edges of the value).
457 * Works without using the dominance tree.
459 static void construct_ssa(ir_node *orig_block, ir_node *orig_val,
460 ir_node *second_block, ir_node *second_val)
464 const ir_edge_t *edge;
465 const ir_edge_t *next;
467 assert(orig_block && orig_val && second_block && second_val &&
468 "no parameter of construct_ssa may be NULL");
470 if (orig_val == second_val)
473 irg = get_irn_irg(orig_val);
475 ir_reserve_resources(irg, IR_RESOURCE_IRN_VISITED);
476 inc_irg_visited(irg);
478 mode = get_irn_mode(orig_val);
479 set_irn_link(orig_block, orig_val);
480 mark_irn_visited(orig_block);
482 ssa_second_def_block = second_block;
483 ssa_second_def = second_val;
485 /* Only fix the users of the first, i.e. the original node */
486 foreach_out_edge_safe(orig_val, edge, next) {
487 ir_node *user = get_edge_src_irn(edge);
488 int j = get_edge_src_pos(edge);
489 ir_node *user_block = get_nodes_block(user);
496 DB((dbg, LEVEL_5, "original user %N\n", user));
499 ir_node *pred_block = get_Block_cfgpred_block(user_block, j);
500 newval = search_def_and_create_phis(pred_block, mode, 1);
502 newval = search_def_and_create_phis(user_block, mode, 1);
504 if (newval != user && !is_Bad(newval))
505 set_irn_n(user, j, newval);
508 ir_free_resources(irg, IR_RESOURCE_IRN_VISITED);
512 /***** Unrolling Helper Functions *****/
514 /* Assign the copy with index nr to node n */
515 static void set_unroll_copy(ir_node *n, int nr, ir_node *cp)
517 unrolling_node_info *info;
518 assert(nr != 0 && "0 reserved");
520 info = (unrolling_node_info *)phase_get_irn_data(phase, n);
524 info = XMALLOCZ(unrolling_node_info);
525 arr = NEW_ARR_F(ir_node *, unroll_nr);
527 memset(info->copies, 0, (unroll_nr) * sizeof(ir_node *));
529 phase_set_irn_data(phase, n, info);
534 info->copies[nr] = cp;
537 /* Returns a nodes copy if it exists, else NULL. */
538 static ir_node *get_unroll_copy(ir_node *n, int nr)
541 unrolling_node_info *info = (unrolling_node_info *)phase_get_irn_data(phase, n);
545 cp = info->copies[nr];
550 /***** Inversion Helper Functions *****/
552 /* Sets copy cp of node n. */
553 static void set_inversion_copy(ir_node *n, ir_node *cp)
555 phase_set_irn_data(phase, n, cp);
558 /* Getter of copy of n for inversion */
559 static ir_node *get_inversion_copy(ir_node *n)
561 ir_node *cp = (ir_node *)phase_get_irn_data(phase, n);
565 /* Resets block mark for given node. For use with walker */
566 static void reset_block_mark(ir_node *node, void * env)
571 set_Block_mark(node, 0);
574 /* Returns mark of node, or its block if node is not a block.
575 * Used in this context to determine if node is in the condition chain. */
576 static unsigned is_nodes_block_marked(ir_node* node)
579 return get_Block_mark(node);
581 return get_Block_mark(get_block(node));
584 /* Extends a nodes ins by node new.
585 * NOTE: This is slow if a node n needs to be extended more than once. */
586 static void extend_irn(ir_node *n, ir_node *newnode, int new_is_backedge)
590 int arity = get_irn_arity(n);
591 int new_arity = arity + 1;
594 NEW_ARR_A(int, bes, new_arity);
595 NEW_ARR_A(ir_node *, ins, new_arity);
598 /* Bes are important!
599 * Another way would be recreating the looptree,
600 * but after that we cannot distinguish already processed loops
601 * from not yet processed ones. */
603 for(i = 0; i < arity; ++i) {
604 bes[i] = is_backedge(n, i);
606 bes[i] = new_is_backedge;
609 for(i = 0; i < arity; ++i) {
610 ins[i] = get_irn_n(n, i);
614 set_irn_in(n, new_arity, ins);
618 for(i = 0; i < new_arity; ++i) {
625 /* Extends a block by a copy of its pred at pos,
626 * fixing also the phis in the same way. */
627 static void extend_ins_by_copy(ir_node *block, int pos)
632 assert(is_Block(block));
634 /* Extend block by copy of definition at pos */
635 pred = get_irn_n(block, pos);
636 new_in = get_inversion_copy(pred);
637 DB((dbg, LEVEL_5, "Extend block %N by %N cp of %N\n", block, new_in, pred));
638 extend_irn(block, new_in, 0);
640 /* Extend block phis by copy of definition at pos */
641 for_each_phi(block, phi) {
644 pred = get_irn_n(phi, pos);
645 cp = get_inversion_copy(pred);
646 /* If the phis in is not in the condition chain (eg. a constant),
647 * there is no copy. */
653 DB((dbg, LEVEL_5, "Extend phi %N by %N cp of %N\n", phi, new_in, pred));
654 extend_irn(phi, new_in, 0);
658 /* Returns the number of blocks backedges. With or without alien bes. */
659 static int get_backedge_n(ir_node *block, unsigned with_alien)
663 int arity = get_irn_arity(block);
665 assert(is_Block(block));
667 for (i = 0; i < arity; ++i) {
668 ir_node *pred = get_irn_n(block, i);
669 if (is_backedge(block, i) && (with_alien || is_in_loop(pred)))
675 /* Returns a raw copy of the given node.
676 * Attributes are kept/set according to the needs of loop inversion. */
677 static ir_node *copy_node(ir_node *node)
682 cp = exact_copy(node);
683 arity = get_irn_arity(node);
685 /* Keep backedge info */
686 for (i = 0; i < arity; ++i) {
687 if (is_backedge(node, i))
692 set_Block_mark(cp, 0);
700 * This walker copies all walked nodes.
701 * If the walk_condition is true for a node, it is copied.
702 * All nodes node_info->copy have to be NULL prior to every walk.
703 * Order of ins is important for later usage.
705 static void copy_walk(ir_node *node, walker_condition *walk_condition,
712 ir_graph *irg = current_ir_graph;
715 * break condition and cycle resolver, creating temporary node copies
717 if (get_irn_visited(node) >= get_irg_visited(irg)) {
718 /* Here we rely on nodestate's copy being initialized with NULL */
719 DB((dbg, LEVEL_5, "copy_walk: We have already visited %N\n", node));
720 if (get_inversion_copy(node) == NULL) {
721 cp = copy_node(node);
722 set_inversion_copy(node, cp);
724 DB((dbg, LEVEL_5, "The TEMP copy of %N is created %N\n", node, cp));
730 mark_irn_visited(node);
732 if (!is_Block(node)) {
733 ir_node *pred = get_nodes_block(node);
734 if (walk_condition(pred))
735 DB((dbg, LEVEL_5, "walk block %N\n", pred));
736 copy_walk(pred, walk_condition, set_loop);
739 arity = get_irn_arity(node);
741 NEW_ARR_A(ir_node *, cpin, arity);
743 for (i = 0; i < arity; ++i) {
744 ir_node *pred = get_irn_n(node, i);
746 if (walk_condition(pred)) {
747 DB((dbg, LEVEL_5, "walk node %N\n", pred));
748 copy_walk(pred, walk_condition, set_loop);
749 cpin[i] = get_inversion_copy(pred);
750 DB((dbg, LEVEL_5, "copy of %N gets new in %N which is copy of %N\n",
751 node, get_inversion_copy(pred), pred));
757 /* copy node / finalize temp node */
758 if (get_inversion_copy(node) == NULL) {
759 /* No temporary copy existent */
760 cp = copy_node(node);
761 set_inversion_copy(node, cp);
762 DB((dbg, LEVEL_5, "The FINAL copy of %N is CREATED %N\n", node, cp));
764 /* temporary copy is existent but without correct ins */
765 cp = get_inversion_copy(node);
766 DB((dbg, LEVEL_5, "The FINAL copy of %N is EXISTENT %N\n", node, cp));
769 if (!is_Block(node)) {
770 ir_node *cpblock = get_inversion_copy(get_nodes_block(node));
772 set_nodes_block(cp, cpblock );
774 add_Block_phi(cpblock, cp);
777 /* Keeps phi list of temporary node. */
778 set_irn_in(cp, ARR_LEN(cpin), cpin);
782 * This walker copies all walked nodes.
783 * If the walk_condition is true for a node, it is copied.
784 * All nodes node_info->copy have to be NULL prior to every walk.
785 * Order of ins is important for later usage.
786 * Takes copy_index, to phase-link copy at specific index.
788 static void copy_walk_n(ir_node *node,
789 walker_condition *walk_condition, int copy_index)
797 * break condition and cycle resolver, creating temporary node copies
799 if (irn_visited(node)) {
800 /* Here we rely on nodestate's copy being initialized with NULL */
801 DB((dbg, LEVEL_5, "copy_walk: We have already visited %N\n", node));
802 if (get_unroll_copy(node, copy_index) == NULL) {
805 set_unroll_copy(node, copy_index, u);
806 DB((dbg, LEVEL_5, "The TEMP unknown of %N is created %N\n", node, u));
812 mark_irn_visited(node);
814 if (!is_Block(node)) {
815 ir_node *block = get_nodes_block(node);
816 if (walk_condition(block))
817 DB((dbg, LEVEL_5, "walk block %N\n", block));
818 copy_walk_n(block, walk_condition, copy_index);
821 arity = get_irn_arity(node);
822 NEW_ARR_A(ir_node *, cpin, arity);
824 for (i = 0; i < arity; ++i) {
825 ir_node *pred = get_irn_n(node, i);
827 if (walk_condition(pred)) {
828 DB((dbg, LEVEL_5, "walk node %N\n", pred));
829 copy_walk_n(pred, walk_condition, copy_index);
830 cpin[i] = get_unroll_copy(pred, copy_index);
836 /* copy node / finalize temp node */
837 cp = get_unroll_copy(node, copy_index);
838 if (cp == NULL || is_Unknown(cp)) {
839 cp = copy_node(node);
840 set_unroll_copy(node, copy_index, cp);
841 DB((dbg, LEVEL_5, "The FINAL copy of %N is CREATED %N\n", node, cp));
843 /* temporary copy is existent but without correct ins */
844 cp = get_unroll_copy(node, copy_index);
845 DB((dbg, LEVEL_5, "The FINAL copy of %N is EXISTENT %N\n", node, cp));
848 if (!is_Block(node)) {
849 ir_node *cpblock = get_unroll_copy(get_nodes_block(node), copy_index);
851 set_nodes_block(cp, cpblock );
853 add_Block_phi(cpblock, cp);
856 /* Keeps phi list of temporary node. */
857 set_irn_in(cp, ARR_LEN(cpin), cpin);
860 /* Removes alle Blocks with non marked predecessors from the condition chain. */
861 static void unmark_not_allowed_cc_blocks(void)
863 int blocks = ARR_LEN(cc_blocks);
866 for(i = 0; i < blocks; ++i) {
867 ir_node *block = cc_blocks[i];
869 int arity = get_irn_arity(block);
871 /* Head is an exception. */
872 if (block == loop_head)
875 for(a = 0; a < arity; ++a) {
876 if (! is_nodes_block_marked(get_irn_n(block, a))) {
877 set_Block_mark(block, 0);
878 --inversion_blocks_in_cc;
879 DB((dbg, LEVEL_5, "Removed %N from cc (blocks in cc %d)\n",
880 block, inversion_blocks_in_cc));
888 /* Unmarks all cc blocks using cc_blocks except head.
889 * TODO: invert head for unrolling? */
890 static void unmark_cc_blocks(void)
892 int blocks = ARR_LEN(cc_blocks);
895 for(i = 0; i < blocks; ++i) {
896 ir_node *block = cc_blocks[i];
898 /* TODO Head is an exception. */
899 /*if (block != loop_head)*/
900 set_Block_mark(block, 0);
902 /*inversion_blocks_in_cc = 1;*/
903 inversion_blocks_in_cc = 0;
906 loop_info.cc_size = 0;
910 * Populates head_entries with (node, pred_pos) tuple
911 * whereas the node's pred at pred_pos is in the cc but not the node itself.
912 * Also finds df loops inside the cc.
913 * Head and condition chain blocks have been marked previously.
915 static void get_head_outs(ir_node *node, void *env)
918 int arity = get_irn_arity(node);
921 for (i = 0; i < arity; ++i) {
922 if (!is_nodes_block_marked(node) && is_nodes_block_marked(get_irn_n(node, i))) {
926 /* Saving also predecessor seems redundant, but becomes
927 * necessary when changing position of it, before
928 * dereferencing it.*/
929 entry.pred = get_irn_n(node, i);
930 ARR_APP1(entry_edge, cur_head_outs, entry);
934 arity = get_irn_arity(loop_head);
936 /* Find df loops inside the cc */
937 if (is_Phi(node) && get_nodes_block(node) == loop_head) {
938 for (i = 0; i < arity; ++i) {
939 if (is_own_backedge(loop_head, i)) {
940 if (is_nodes_block_marked(get_irn_n(node, i))) {
944 entry.pred = get_irn_n(node, i);
945 ARR_APP1(entry_edge, head_df_loop, entry);
946 DB((dbg, LEVEL_5, "Found incc assignment node %N @%d is pred %N, graph %N %N\n",
947 node, i, entry.pred, current_ir_graph, get_irg_start_block(current_ir_graph)));
955 * Find condition chains, and add them to be inverted.
956 * A block belongs to the chain if a condition branches out of the loop.
957 * (Some blocks need to be removed once again.)
958 * Returns 1 if the given block belongs to the condition chain.
960 static unsigned find_condition_chain(ir_node *block)
962 const ir_edge_t *edge;
966 unsigned nodes_n = 0;
968 mark_irn_visited(block);
970 DB((dbg, LEVEL_5, "condition_chains for block %N\n", block));
973 foreach_out_edge_kind(block, edge, EDGE_KIND_NORMAL) {
977 /* Check if node count would exceed maximum cc size.
979 * This is not optimal, as we search depth-first and break here,
980 * continuing with another subtree. */
981 if (loop_info.cc_size + nodes_n > opt_params.max_cc_size) {
982 set_Block_mark(block, 0);
986 /* Check if block only has a jmp instruction. */
988 foreach_out_edge(block, edge) {
989 ir_node *src = get_edge_src_irn(edge);
991 if (! is_Block(src) && ! is_Jmp(src)) {
996 /* Check cf outs if one is leaving the loop,
997 * or if this node has a backedge. */
998 foreach_block_succ(block, edge) {
999 ir_node *src = get_edge_src_irn(edge);
1000 int pos = get_edge_src_pos(edge);
1002 if (! is_in_loop(src))
1005 /* Inverting blocks with backedge outs leads to a cf edge
1006 * from the inverted head, into the inverted head (skipping the body).
1007 * As the body becomes the new loop head,
1008 * this would introduce another loop in the existing loop.
1009 * This loop inversion cannot cope with this case. */
1010 if (is_backedge(src, pos)) {
1016 /* We need all predecessors to already belong to the condition chain.
1017 * Example of wrong case: * == in cc
1028 /* Collect blocks containing only a Jmp.
1029 * Do not collect blocks with backedge outs. */
1030 if ((jmp_only == 1 || mark == 1) && has_be == 0) {
1031 set_Block_mark(block, 1);
1032 ++inversion_blocks_in_cc;
1033 loop_info.cc_size += nodes_n;
1034 DB((dbg, LEVEL_5, "block %N is part of condition chain\n", block));
1035 ARR_APP1(ir_node *, cc_blocks, block);
1037 set_Block_mark(block, 0);
1040 foreach_block_succ(block, edge) {
1041 ir_node *src = get_edge_src_irn( edge );
1043 if (is_in_loop(src) && ! irn_visited(src))
1044 find_condition_chain(src);
1051 * Rewires the copied condition chain. Removes backedges
1052 * as this condition chain is prior to the loop.
1053 * Copy of loop_head must have phi list and old (unfixed) backedge info of the loop head.
1054 * (loop_head is already fixed, we cannot rely on it.)
1056 static void fix_copy_inversion(void)
1061 ir_node *phi, *next;
1062 ir_node *head_cp = get_inversion_copy(loop_head);
1063 int arity = get_irn_arity(head_cp);
1064 int backedges = get_backedge_n(head_cp, 0);
1065 int new_arity = arity - backedges;
1069 NEW_ARR_A(ir_node *, ins, new_arity);
1072 /* Remove block backedges */
1073 for(i = 0; i < arity; ++i) {
1074 if (!is_backedge(head_cp, i))
1075 ins[pos++] = get_irn_n(head_cp, i);
1078 new_head = new_Block(new_arity, ins);
1080 phis = NEW_ARR_F(ir_node *, 0);
1082 for_each_phi_safe(get_Block_phis(head_cp), phi, next) {
1084 NEW_ARR_A(ir_node *, ins, new_arity);
1086 for(i = 0; i < arity; ++i) {
1087 if (!is_backedge(head_cp, i))
1088 ins[pos++] = get_irn_n(phi, i);
1090 new_phi = new_rd_Phi(get_irn_dbg_info(phi),
1091 new_head, new_arity, ins,
1093 ARR_APP1(ir_node *, phis, new_phi);
1097 for_each_phi_safe(get_Block_phis(head_cp), phi, next) {
1098 exchange(phi, phis[pos++]);
1101 exchange(head_cp, new_head);
1107 /* Puts the original condition chain at the end of the loop,
1108 * subsequently to the body.
1109 * Relies on block phi list and correct backedges.
1111 static void fix_head_inversion(void)
1115 ir_node *phi, *next;
1117 int arity = get_irn_arity(loop_head);
1118 int backedges = get_backedge_n(loop_head, 0);
1119 int new_arity = backedges;
1123 NEW_ARR_A(ir_node *, ins, new_arity);
1126 /* Keep only backedges */
1127 for(i = 0; i < arity; ++i) {
1128 if (is_own_backedge(loop_head, i))
1129 ins[pos++] = get_irn_n(loop_head, i);
1132 new_head = new_Block(new_arity, ins);
1134 phis = NEW_ARR_F(ir_node *, 0);
1136 for_each_phi(loop_head, phi) {
1138 DB((dbg, LEVEL_5, "Fixing phi %N of loop head\n", phi));
1140 NEW_ARR_A(ir_node *, ins, new_arity);
1143 for (i = 0; i < arity; ++i) {
1144 ir_node *pred = get_irn_n(phi, i);
1146 if (is_own_backedge(loop_head, i)) {
1147 /* If assignment is in the condition chain,
1148 * we need to create a phi in the new loop head.
1149 * This can only happen for df, not cf. See find_condition_chains. */
1150 /*if (is_nodes_block_marked(pred)) {
1158 new_phi = new_rd_Phi(get_irn_dbg_info(phi),
1159 new_head, new_arity, ins,
1162 ARR_APP1(ir_node *, phis, new_phi);
1164 DB((dbg, LEVEL_5, "fix inverted head should exch %N by %N (pos %d)\n", phi, new_phi, pos ));
1168 for_each_phi_safe(get_Block_phis(loop_head), phi, next) {
1169 DB((dbg, LEVEL_5, "fix inverted exch phi %N by %N\n", phi, phis[pos]));
1170 if (phis[pos] != phi)
1171 exchange(phi, phis[pos++]);
1176 DB((dbg, LEVEL_5, "fix inverted head exch head block %N by %N\n", loop_head, new_head));
1177 exchange(loop_head, new_head);
1180 /* Does the loop inversion. */
1181 static void inversion_walk(entry_edge *head_entries)
1186 * The order of rewiring bottom-up is crucial.
1187 * Any change of the order leads to lost information that would be needed later.
1190 ir_reserve_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1192 /* 1. clone condition chain */
1193 inc_irg_visited(current_ir_graph);
1195 for (i = 0; i < ARR_LEN(head_entries); ++i) {
1196 entry_edge entry = head_entries[i];
1197 ir_node *pred = get_irn_n(entry.node, entry.pos);
1199 DB((dbg, LEVEL_5, "\nInit walk block %N\n", pred));
1201 copy_walk(pred, is_nodes_block_marked, cur_loop);
1204 ir_free_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1206 /* 2. Extends the head control flow successors ins
1207 * with the definitions of the copied head node. */
1208 for (i = 0; i < ARR_LEN(head_entries); ++i) {
1209 entry_edge head_out = head_entries[i];
1211 if (is_Block(head_out.node))
1212 extend_ins_by_copy(head_out.node, head_out.pos);
1215 /* 3. construct_ssa for users of definitions in the condition chain,
1216 * as there is now a second definition. */
1217 for (i = 0; i < ARR_LEN(head_entries); ++i) {
1218 entry_edge head_out = head_entries[i];
1220 /* Ignore keepalives */
1221 if (is_End(head_out.node))
1224 /* Construct ssa for assignments in the condition chain. */
1225 if (!is_Block(head_out.node)) {
1226 ir_node *pred, *cppred, *block, *cpblock;
1228 pred = head_out.pred;
1229 cppred = get_inversion_copy(pred);
1230 block = get_nodes_block(pred);
1231 cpblock = get_nodes_block(cppred);
1232 construct_ssa(block, pred, cpblock, cppred);
1237 * If there is an assignment in the condition chain
1238 * with a user also in the condition chain,
1239 * the dominance frontier is in the new loop head.
1240 * The dataflow loop is completely in the condition chain.
1245 * Phi_cp | | copied condition chain
1249 * Phi* | | new loop head with newly created phi.
1251 * Phi | | original, inverted condition chain
1256 for (i = 0; i < ARR_LEN(head_df_loop); ++i) {
1257 entry_edge head_out = head_df_loop[i];
1259 /* Construct ssa for assignments in the condition chain. */
1260 ir_node *pred, *cppred, *block, *cpblock;
1262 pred = head_out.pred;
1263 cppred = get_inversion_copy(pred);
1264 assert(cppred && pred);
1265 block = get_nodes_block(pred);
1266 cpblock = get_nodes_block(cppred);
1267 construct_ssa(block, pred, cpblock, cppred);
1270 /* 4. Remove the ins which are no backedges from the original condition chain
1271 * as the cc is now subsequent to the body. */
1272 fix_head_inversion();
1274 /* 5. Remove the backedges of the copied condition chain,
1275 * because it is going to be the new 'head' in advance to the loop. */
1276 fix_copy_inversion();
1280 /* Performs loop inversion of cur_loop if possible and reasonable. */
1281 static void loop_inversion(void)
1284 unsigned max_loop_nodes = opt_params.max_loop_size;
1285 unsigned max_loop_nodes_adapted;
1286 int depth_adaption = opt_params.depth_adaption;
1288 unsigned do_inversion = 1;
1289 unsigned has_cc = 0;
1291 /* Depth of 0 is the procedure and 1 a topmost loop. */
1292 loop_depth = get_loop_depth(cur_loop) - 1;
1294 /* Calculating in per mil. */
1295 max_loop_nodes_adapted = get_max_nodes_adapted(loop_depth);
1297 DB((dbg, LEVEL_1, "max_nodes: %d\nmax_nodes_adapted %d at depth of %d (adaption %d)\n",
1298 max_loop_nodes, max_loop_nodes_adapted, loop_depth, depth_adaption));
1300 if (! (loop_info.nodes > 0))
1303 #if LOOP_IGNORE_NODE_LIMITS
1304 DB((dbg, LEVEL_1, "WARNING: Loop node limitations ignored."));
1306 if (loop_info.nodes > max_loop_nodes) {
1307 /* Only for stats */
1308 DB((dbg, LEVEL_1, "Nodes %d > allowed nodes %d\n",
1309 loop_info.nodes, loop_depth, max_loop_nodes));
1310 count_stats(stats.too_large);
1312 /* Adaption might change it */
1315 /* Limit processing to loops smaller than given parameter. */
1316 if (loop_info.nodes > max_loop_nodes_adapted) {
1317 DB((dbg, LEVEL_1, "Nodes %d > allowed nodes (depth %d adapted) %d\n",
1318 loop_info.nodes, loop_depth, max_loop_nodes_adapted));
1319 count_stats(stats.too_large_adapted);
1323 if (loop_info.calls > opt_params.allowed_calls) {
1324 DB((dbg, LEVEL_1, "Calls %d > allowed calls %d\n",
1325 loop_info.calls, opt_params.allowed_calls));
1326 count_stats(stats.calls_limit);
1331 /*inversion_head_node_limit = INT_MAX;*/
1332 ir_reserve_resources(current_ir_graph, IR_RESOURCE_BLOCK_MARK);
1334 /* Reset block marks.
1335 * We use block marks to flag blocks of the original condition chain. */
1336 irg_walk_graph(current_ir_graph, reset_block_mark, NULL, NULL);
1338 /*loop_info.blocks = get_loop_n_blocks(cur_loop);*/
1339 cond_chain_entries = NEW_ARR_F(entry_edge, 0);
1340 head_df_loop = NEW_ARR_F(entry_edge, 0);
1342 /*head_inversion_node_count = 0;*/
1343 inversion_blocks_in_cc = 0;
1345 /* Use phase to keep copy of nodes from the condition chain. */
1346 phase = new_phase(current_ir_graph, phase_irn_init_default);
1348 /* Search for condition chains and temporarily save the blocks in an array. */
1349 cc_blocks = NEW_ARR_F(ir_node *, 0);
1350 inc_irg_visited(current_ir_graph);
1351 has_cc = find_condition_chain(loop_head);
1353 unmark_not_allowed_cc_blocks();
1354 DEL_ARR_F(cc_blocks);
1356 #if LOOP_IGNORE_NODE_LIMITS
1357 (void) unmark_cc_blocks;
1359 /* Condition chain too large.
1360 * Loop should better be small enough to fit into the cache. */
1361 /* TODO Of course, we should take a small enough cc in the first place,
1362 * which is not that simple. (bin packing) */
1363 if (loop_info.cc_size > opt_params.max_cc_size) {
1364 count_stats(stats.cc_limit_reached);
1368 /* Unmark cc blocks except the head.
1369 * Invert head only for possible unrolling. */
1375 /* We also catch endless loops here,
1376 * because they do not have a condition chain. */
1377 if (inversion_blocks_in_cc < 1) {
1380 "Loop contains %d (less than 1) invertible blocks => No Inversion done.\n",
1381 inversion_blocks_in_cc));
1385 cur_head_outs = NEW_ARR_F(entry_edge, 0);
1387 /* Get all edges pointing into the condition chain. */
1388 irg_walk_graph(current_ir_graph, get_head_outs, NULL, NULL);
1390 /* Do the inversion */
1391 inversion_walk(cur_head_outs);
1393 DEL_ARR_F(cur_head_outs);
1395 /* Duplicated blocks changed doms */
1396 set_irg_doms_inconsistent(current_ir_graph);
1397 /* Loop content changed */
1398 set_irg_loopinfo_inconsistent(current_ir_graph);
1399 /* TODO are they? Depends on set_irn_in and set_irn_n exchange and new_node. */
1400 set_irg_outs_inconsistent(current_ir_graph);
1402 count_stats(stats.inverted);
1407 DEL_ARR_F(cond_chain_entries);
1408 DEL_ARR_F(head_df_loop);
1410 ir_free_resources(current_ir_graph, IR_RESOURCE_BLOCK_MARK);
1413 /* Fix the original loop_heads ins for invariant unrolling case. */
1414 static void unrolling_fix_loop_head_inv(void)
1418 ir_node *proj = new_Proj(loop_info.duff_cond, mode_X, 0);
1419 ir_node *head_pred = get_irn_n(loop_head, loop_info.be_src_pos);
1420 ir_node *loop_condition = get_unroll_copy(head_pred, unroll_nr - 1);
1422 /* Original loop_heads ins are:
1423 * duff block and the own backedge */
1425 ins[0] = loop_condition;
1427 set_irn_in(loop_head, 2, ins);
1428 DB((dbg, LEVEL_4, "Rewire ins of block loophead %N to pred %N and duffs entry %N \n" , loop_head, ins[0], ins[1]));
1430 for_each_phi(loop_head, phi) {
1431 ir_node *pred = get_irn_n(phi, loop_info.be_src_pos);
1432 /* TODO we think it is a phi, but for Mergesort it is not the case.*/
1434 ir_node *last_pred = get_unroll_copy(pred, unroll_nr - 1);
1437 ins[1] = (ir_node*)get_irn_link(phi);
1438 set_irn_in(phi, 2, ins);
1439 DB((dbg, LEVEL_4, "Rewire ins of loophead phi %N to pred %N and duffs entry %N \n" , phi, ins[0], ins[1]));
1443 /* Removes previously created phis with only 1 in. */
1444 static void correct_phis(ir_node *node, void *env)
1448 if (is_Phi(node) && get_irn_arity(node) == 1) {
1452 in[0] = get_irn_n(node, 0);
1454 exch = new_rd_Phi(get_irn_dbg_info(node),
1455 get_nodes_block(node), 1, in,
1456 get_irn_mode(node));
1458 exchange(node, exch);
1462 /* Unrolling: Rewire floating copies. */
1463 static void place_copies(int copies)
1465 ir_node *loophead = loop_head;
1467 int be_src_pos = loop_info.be_src_pos;
1469 /* Serialize loops by fixing their head ins.
1470 * Processed are the copies.
1471 * The original loop is done after that, to keep backedge infos. */
1472 for (c = 0; c < copies; ++c) {
1473 ir_node *upper = get_unroll_copy(loophead, c);
1474 ir_node *lower = get_unroll_copy(loophead, c + 1);
1476 ir_node *topmost_be_block = get_nodes_block(get_irn_n(loophead, be_src_pos));
1478 /* Important: get the preds first and then their copy. */
1479 ir_node *upper_be_block = get_unroll_copy(topmost_be_block, c);
1480 ir_node *new_jmp = new_r_Jmp(upper_be_block);
1481 DB((dbg, LEVEL_5, " place_copies upper %N lower %N\n", upper, lower));
1483 DB((dbg, LEVEL_5, "topmost be block %N \n", topmost_be_block));
1485 if (loop_info.unroll_kind == constant) {
1488 set_irn_in(lower, 1, ins);
1490 for_each_phi(loophead, phi) {
1491 ir_node *topmost_def = get_irn_n(phi, be_src_pos);
1492 ir_node *upper_def = get_unroll_copy(topmost_def, c);
1493 ir_node *lower_phi = get_unroll_copy(phi, c + 1);
1495 /* It is possible, that the value used
1496 * in the OWN backedge path is NOT defined in this loop. */
1497 if (is_in_loop(topmost_def))
1500 ins[0] = topmost_def;
1502 set_irn_in(lower_phi, 1, ins);
1503 /* Need to replace phis with 1 in later. */
1506 /* Invariant case */
1507 /* Every node has 2 ins. One from the duff blocks
1508 * and one from the previously unrolled loop. */
1510 /* Calculate corresponding projection of mod result for this copy c */
1511 ir_node *proj = new_Proj(loop_info.duff_cond, mode_X, unroll_nr - c - 1);
1512 DB((dbg, LEVEL_4, "New duff proj %N\n" , proj));
1516 set_irn_in(lower, 2, ins);
1517 DB((dbg, LEVEL_4, "Rewire ins of Block %N to pred %N and duffs entry %N \n" , lower, ins[0], ins[1]));
1519 for_each_phi(loophead, phi) {
1520 ir_node *topmost_phi_pred = get_irn_n(phi, be_src_pos);
1521 ir_node *upper_phi_pred;
1525 lower_phi = get_unroll_copy(phi, c + 1);
1526 duff_phi = (ir_node*)get_irn_link(phi);
1527 DB((dbg, LEVEL_4, "DD Link of %N is %N\n" , phi, duff_phi));
1530 if (is_in_loop(topmost_phi_pred)) {
1531 upper_phi_pred = get_unroll_copy(topmost_phi_pred, c);
1533 upper_phi_pred = topmost_phi_pred;
1536 ins[0] = upper_phi_pred;
1538 set_irn_in(lower_phi, 2, ins);
1539 DB((dbg, LEVEL_4, "Rewire ins of %N to pred %N and duffs entry %N \n" , lower_phi, ins[0], ins[1]));
1544 /* Reconnect last copy. */
1545 for (i = 0; i < ARR_LEN(loop_entries); ++i) {
1546 entry_edge edge = loop_entries[i];
1547 /* Last copy is at the bottom */
1548 ir_node *new_pred = get_unroll_copy(edge.pred, copies);
1549 set_irn_n(edge.node, edge.pos, new_pred);
1552 /* Fix original loops head.
1553 * Done in the end, as ins and be info were needed before. */
1554 if (loop_info.unroll_kind == constant) {
1556 ir_node *head_pred = get_irn_n(loop_head, be_src_pos);
1557 ir_node *loop_condition = get_unroll_copy(head_pred, unroll_nr - 1);
1559 set_irn_n(loop_head, loop_info.be_src_pos, loop_condition);
1561 for_each_phi(loop_head, phi) {
1562 ir_node *pred = get_irn_n(phi, be_src_pos);
1565 /* It is possible, that the value used
1566 * in the OWN backedge path is NOT assigned in this loop. */
1567 if (is_in_loop(pred))
1568 last_pred = get_unroll_copy(pred, copies);
1571 set_irn_n(phi, be_src_pos, last_pred);
1575 unrolling_fix_loop_head_inv();
1579 /* Copies the cur_loop several times. */
1580 static void copy_loop(entry_edge *cur_loop_outs, int copies)
1584 ir_reserve_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1586 for (c = 0; c < copies; ++c) {
1588 inc_irg_visited(current_ir_graph);
1590 DB((dbg, LEVEL_5, " ### Copy_loop copy nr: %d ###\n", c));
1591 for (i = 0; i < ARR_LEN(cur_loop_outs); ++i) {
1592 entry_edge entry = cur_loop_outs[i];
1593 ir_node *pred = get_irn_n(entry.node, entry.pos);
1595 copy_walk_n(pred, is_in_loop, c + 1);
1599 ir_free_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1603 /* Creates a new phi from the given phi node omitting own bes,
1604 * using be_block as supplier of backedge informations. */
1605 static ir_node *clone_phis_sans_bes(ir_node *phi, ir_node *be_block, ir_node *dest_block)
1608 int arity = get_irn_arity(phi);
1612 assert(get_irn_arity(phi) == get_irn_arity(be_block));
1613 assert(is_Phi(phi));
1615 ins = NEW_ARR_F(ir_node *, arity);
1616 for (i = 0; i < arity; ++i) {
1617 if (! is_own_backedge(be_block, i)) {
1618 ins[c] = get_irn_n(phi, i);
1623 newphi = new_r_Phi(dest_block, c, ins, get_irn_mode(phi));
1625 set_irn_link(phi, newphi);
1626 DB((dbg, LEVEL_4, "Linking for duffs device %N to %N\n", phi, newphi));
1631 /* Creates a new block from the given block node omitting own bes,
1632 * using be_block as supplier of backedge informations. */
1633 static ir_node *clone_block_sans_bes(ir_node *node, ir_node *be_block)
1636 int arity = get_irn_arity(node);
1639 assert(get_irn_arity(node) == get_irn_arity(be_block));
1640 assert(is_Block(node));
1642 NEW_ARR_A(ir_node *, ins, arity);
1643 for (i = 0; i < arity; ++i) {
1644 if (! is_own_backedge(be_block, i)) {
1645 ins[c] = get_irn_n(node, i);
1650 return new_Block(c, ins);
1653 /* Creates a structure to calculate absolute value of node op.
1654 * Returns mux node with absolute value. */
1655 static ir_node *new_Abs(ir_node *op, ir_mode *mode)
1657 ir_graph *irg = get_irn_irg(op);
1658 ir_node *block = get_nodes_block(op);
1659 ir_node *zero = new_r_Const(irg, get_mode_null(mode));
1660 ir_node *cmp = new_r_Cmp(block, op, zero);
1661 ir_node *cond = new_r_Proj(cmp, mode_b, pn_Cmp_Lt);
1662 ir_node *minus_op = new_r_Minus(block, op, mode);
1663 ir_node *mux = new_r_Mux(block, cond, op, minus_op, mode);
1669 /* Creates blocks for duffs device, using previously obtained
1670 * informations about the iv.
1672 static void create_duffs_block(void)
1676 ir_node *block1, *count_block, *duff_block;
1677 ir_node *ems, *ems_divmod, *ems_mod_proj, *cmp_null,
1678 *cmp_proj, *ems_mode_cond, *x_true, *x_false, *const_null;
1679 ir_node *true_val, *false_val;
1682 ir_node *duff_mod, *proj, *cond;
1684 ir_node *count, *correction, *unroll_c;
1685 ir_node *cmp_bad_count, *good_count, *bad_count, *count_phi, *bad_count_neg;
1688 mode = get_irn_mode(loop_info.end_val);
1689 const_null = new_Const(get_mode_null(mode));
1692 * 1. Calculate first approach to count.
1693 * Condition: (end - start) % step == 0 */
1694 block1 = clone_block_sans_bes(loop_head, loop_head);
1695 DB((dbg, LEVEL_4, "Duff block 1 %N\n", block1));
1697 /* Create loop entry phis in first duff block
1698 * as it becomes the loops preheader */
1699 for_each_phi(loop_head, phi) {
1700 /* Returns phis pred if phi would have arity 1*/
1701 ir_node *new_phi = clone_phis_sans_bes(phi, loop_head, block1);
1703 DB((dbg, LEVEL_4, "HEAD %N phi %N\n", loop_head, phi));
1704 DB((dbg, LEVEL_4, "BLOCK1 %N phi %N\n", block1, new_phi));
1707 ems = new_r_Sub(block1, loop_info.end_val, loop_info.start_val,
1708 get_irn_mode(loop_info.end_val));
1709 DB((dbg, LEVEL_4, "BLOCK1 sub %N\n", ems));
1712 ems = new_Sub(loop_info.end_val, loop_info.start_val,
1713 get_irn_mode(loop_info.end_val));
1715 DB((dbg, LEVEL_4, "divmod ins %N %N\n", ems, loop_info.step));
1716 ems_divmod = new_r_DivMod(block1,
1721 op_pin_state_pinned);
1723 DB((dbg, LEVEL_4, "New module node %N\n", ems_divmod));
1725 ems_mod_proj = new_r_Proj(ems_divmod, mode_Iu, pn_DivMod_res_mod);
1726 cmp_null = new_r_Cmp(block1, ems_mod_proj, const_null);
1727 cmp_proj = new_r_Proj(cmp_null, mode_b, pn_Cmp_Eq);
1728 ems_mode_cond = new_r_Cond(block1, cmp_proj);
1731 /* ems % step == 0 */
1732 x_true = new_r_Proj(ems_mode_cond, mode_X, pn_Cond_true);
1733 /* ems % step != 0 */
1734 x_false = new_r_Proj(ems_mode_cond, mode_X, pn_Cond_false);
1737 * Assures, duffs device receives a valid count.
1739 * decreasing: count < 0
1740 * increasing: count > 0
1745 count_block = new_Block(2, ins);
1746 DB((dbg, LEVEL_4, "Duff block 2 %N\n", count_block));
1749 /* Increase loop-taken-count depending on the loop condition
1750 * uses the latest iv to compare to. */
1751 if (loop_info.latest_value == 1) {
1752 /* ems % step == 0 : +0 */
1753 true_val = new_Const(get_mode_null(mode));
1754 /* ems % step != 0 : +1 */
1755 false_val = new_Const(get_mode_one(mode));
1757 ir_tarval *tv_two = new_tarval_from_long(2, mode);
1758 /* ems % step == 0 : +1 */
1759 true_val = new_Const(get_mode_one(mode));
1760 /* ems % step != 0 : +2 */
1761 false_val = new_Const(tv_two);
1767 correction = new_r_Phi(count_block, 2, ins, mode);
1769 count = new_r_Proj(ems_divmod, mode, pn_DivMod_res_div);
1771 /* (end - start) / step + correction */
1772 count = new_Add(count, correction, mode);
1774 cmp_bad_count = new_r_Cmp(count_block, count, const_null);
1776 /* We preconditioned the loop to be tail-controlled.
1777 * So, if count is something 'wrong' like 0,
1778 * negative/positive (depending on step direction),
1779 * we may take the loop once (tail-contr.) and leave it
1780 * to the existing condition, to break; */
1782 /* Depending on step direction, we have to check for > or < 0 */
1783 if (loop_info.decreasing == 1) {
1784 bad_count_neg = new_r_Proj(cmp_bad_count, mode_b, pn_Cmp_Lt);
1786 bad_count_neg = new_r_Proj(cmp_bad_count, mode_b, pn_Cmp_Gt);
1789 bad_count_neg = new_r_Cond(count_block, bad_count_neg);
1790 good_count = new_Proj(bad_count_neg, mode_X, pn_Cond_true);
1791 bad_count = new_Proj(ems_mode_cond, mode_X, pn_Cond_false);
1794 * Contains module to decide which loop to start from. */
1796 ins[0] = good_count;
1798 duff_block = new_Block(2, ins);
1799 DB((dbg, LEVEL_4, "Duff block 3 %N\n", duff_block));
1801 /* Get absolute value */
1802 ins[0] = new_Abs(count, mode);
1803 /* Manually feed the aforementioned count = 1 (bad case)*/
1804 ins[1] = new_Const(get_mode_one(mode));
1805 count_phi = new_r_Phi(duff_block, 2, ins, mode);
1807 unroll_c = new_Const(new_tarval_from_long((long)unroll_nr, mode));
1809 /* count % unroll_nr */
1810 duff_mod = new_r_Mod(duff_block,
1815 op_pin_state_pinned);
1818 proj = new_Proj(duff_mod, mode, pn_Mod_res);
1819 /* condition does NOT create itself in the block of the proj! */
1820 cond = new_r_Cond(duff_block, proj);
1822 loop_info.duff_cond = cond;
1825 /* Returns 1 if given node is not in loop,
1826 * or if it is a phi of the loop head with only loop invariant defs.
1828 static unsigned is_loop_invariant_def(ir_node *node)
1832 if (! is_in_loop(node)) {
1833 DB((dbg, LEVEL_4, "Not in loop %N\n", node));
1834 /* || is_Const(node) || is_SymConst(node)) {*/
1838 /* If this is a phi of the loophead shared by more than 1 loop,
1839 * we need to check if all defs are not in the loop. */
1842 block = get_nodes_block(node);
1844 /* To prevent unexpected situations. */
1845 if (block != loop_head) {
1849 for (i = 0; i < get_irn_arity(node); ++i) {
1850 /* Check if all bes are just loopbacks. */
1851 if (is_own_backedge(block, i) && get_irn_n(node, i) != node)
1854 DB((dbg, LEVEL_4, "invar %N\n", node));
1857 DB((dbg, LEVEL_4, "Not invar %N\n", node));
1862 /* Returns 1 if one pred of node is invariant and the other is not.
1863 * invar_pred and other are set analogously. */
1864 static unsigned get_invariant_pred(ir_node *node, ir_node **invar_pred, ir_node **other)
1866 ir_node *pred0 = get_irn_n(node, 0);
1867 ir_node *pred1 = get_irn_n(node, 1);
1872 if (is_loop_invariant_def(pred0)) {
1873 DB((dbg, LEVEL_4, "pred0 invar %N\n", pred0));
1874 *invar_pred = pred0;
1878 if (is_loop_invariant_def(pred1)) {
1879 DB((dbg, LEVEL_4, "pred1 invar %N\n", pred1));
1881 if (*invar_pred != NULL) {
1882 /* RETURN. We do not want both preds to be invariant. */
1887 *invar_pred = pred1;
1890 DB((dbg, LEVEL_4, "pred1 not invar %N\n", pred1));
1892 if (*invar_pred != NULL)
1899 /* Starts from a phi that may belong to an iv.
1900 * If an add forms a loop with iteration_phi,
1901 * and add uses a constant, 1 is returned
1902 * and 'start' as well as 'add' are sane. */
1903 static unsigned get_start_and_add(ir_node *iteration_phi, unrolling_kind_flag role)
1906 ir_node *found_add = loop_info.add;
1907 int arity = get_irn_arity(iteration_phi);
1909 DB((dbg, LEVEL_4, "Find start and add from %N\n", iteration_phi));
1911 for (i = 0; i < arity; ++i) {
1913 /* Find start_val which needs to be pred of the iteration_phi.
1914 * If start_val already known, sanity check. */
1915 if (!is_backedge(get_nodes_block(loop_info.iteration_phi), i)) {
1916 ir_node *found_start_val = get_irn_n(loop_info.iteration_phi, i);
1918 DB((dbg, LEVEL_4, "found_start_val %N\n", found_start_val));
1920 /* We already found a start_val it has to be always the same. */
1921 if (loop_info.start_val && found_start_val != loop_info.start_val)
1924 if ((role == constant) && !(is_SymConst(found_start_val) || is_Const(found_start_val)))
1926 else if((role == constant) && !(is_loop_invariant_def(found_start_val)))
1929 loop_info.start_val = found_start_val;
1932 /* The phi has to be in the loop head.
1933 * Follow all own backedges. Every value supplied from these preds of the phi
1934 * needs to origin from the same add. */
1935 if (is_own_backedge(get_nodes_block(loop_info.iteration_phi), i)) {
1936 ir_node *new_found = get_irn_n(loop_info.iteration_phi,i);
1938 DB((dbg, LEVEL_4, "is add? %N\n", new_found));
1940 if (! (is_Add(new_found) || is_Sub(new_found)) || (found_add && found_add != new_found))
1943 found_add = new_found;
1947 loop_info.add = found_add;
1953 /* Returns 1 if one pred of node is a const value and the other is not.
1954 * const_pred and other are set analogously. */
1955 static unsigned get_const_pred(ir_node *node, ir_node **const_pred, ir_node **other)
1957 ir_node *pred0 = get_irn_n(node, 0);
1958 ir_node *pred1 = get_irn_n(node, 1);
1960 DB((dbg, LEVEL_4, "Checking for constant pred of %N\n", node));
1965 /*DB((dbg, LEVEL_4, "is %N const\n", pred0));*/
1966 if (is_Const(pred0) || is_SymConst(pred0)) {
1967 *const_pred = pred0;
1971 /*DB((dbg, LEVEL_4, "is %N const\n", pred1));*/
1972 if (is_Const(pred1) || is_SymConst(pred1)) {
1973 if (*const_pred != NULL) {
1974 /* RETURN. We do not want both preds to be constant. */
1979 *const_pred = pred1;
1982 if (*const_pred == NULL)
1988 /* Returns the mathematically inverted pn_Cmp. */
1989 static pn_Cmp get_math_inverted_case(pn_Cmp proj)
2005 panic("Unhandled pn_Cmp.");
2009 /* Returns 1 if loop exits within 2 steps of the iv.
2010 * Norm_proj means we do not exit the loop.*/
2011 static unsigned simulate_next(ir_tarval **count_tar,
2012 ir_tarval *stepped, ir_tarval *step_tar, ir_tarval *end_tar,
2017 DB((dbg, LEVEL_4, "Loop taken if (stepped)%ld %s (end)%ld ",
2018 get_tarval_long(stepped),
2019 get_pnc_string((norm_proj)),
2020 get_tarval_long(end_tar)));
2021 DB((dbg, LEVEL_4, "comparing latest value %d\n", loop_info.latest_value));
2023 /* If current iv does not stay in the loop,
2024 * this run satisfied the exit condition. */
2025 if (! (tarval_cmp(stepped, end_tar) & norm_proj))
2028 DB((dbg, LEVEL_4, "Result: (stepped)%ld IS %s (end)%ld\n",
2029 get_tarval_long(stepped),
2030 get_pnc_string(tarval_cmp(stepped, end_tar)),
2031 get_tarval_long(end_tar)));
2034 if (is_Add(loop_info.add))
2035 next = tarval_add(stepped, step_tar);
2038 next = tarval_sub(stepped, step_tar, get_irn_mode(loop_info.end_val));
2040 DB((dbg, LEVEL_4, "Loop taken if %ld %s %ld ",
2041 get_tarval_long(next),
2042 get_pnc_string(norm_proj),
2043 get_tarval_long(end_tar)));
2044 DB((dbg, LEVEL_4, "comparing latest value %d\n", loop_info.latest_value));
2046 /* Increase steps. */
2047 *count_tar = tarval_add(*count_tar, get_tarval_one(get_tarval_mode(*count_tar)));
2049 /* Next has to fail the loop condition, or we will never exit. */
2050 if (! (tarval_cmp(next, end_tar) & norm_proj))
2056 /* Check if loop meets requirements for a 'simple loop':
2057 * - Exactly one cf out
2059 * - Max nodes after unrolling
2063 * Returns Projection of cmp node or NULL; */
2064 static ir_node *is_simple_loop(void)
2067 ir_node *loop_block, *exit_block, *projx, *cond, *projres, *loop_condition;
2069 /* Maximum of one condition, and no endless loops. */
2070 if (loop_info.cf_outs != 1)
2073 DB((dbg, LEVEL_4, "1 loop exit\n"));
2075 #if LOOP_IGNORE_NODE_LIMITS
2076 /* Ignore loop size. Probably not wise in other than testcases. */
2077 loop_info.max_unroll = 40;
2079 /* Calculate maximum unroll_nr keeping node count below limit. */
2080 loop_info.max_unroll = (int)((double)opt_params.max_unrolled_loop_size / (double)loop_info.nodes);
2081 if (loop_info.max_unroll < 2) {
2082 count_stats(stats.too_large);
2088 DB((dbg, LEVEL_4, "maximum unroll factor %u, to not exceed node limit \n",
2089 opt_params.max_unrolled_loop_size));
2091 arity = get_irn_arity(loop_head);
2092 /* RETURN if we have more than 1 be. */
2093 /* Get my backedges without alien bes. */
2095 for (i = 0; i < arity; ++i) {
2096 ir_node *pred = get_irn_n(loop_head, i);
2097 if (is_own_backedge(loop_head, i)) {
2099 /* Our simple loops may have only one backedge. */
2102 loop_block = get_nodes_block(pred);
2103 loop_info.be_src_pos = i;
2108 DB((dbg, LEVEL_4, "loop has 1 own backedge.\n"));
2110 exit_block = get_nodes_block(loop_info.cf_out.pred);
2111 /* The loop has to be tail-controlled.
2112 * This can be changed/improved,
2113 * but we would need a duff iv. */
2114 if (exit_block != loop_block)
2117 DB((dbg, LEVEL_4, "tail-controlled loop.\n"));
2119 /* find value on which loop exit depends */
2120 projx = loop_info.cf_out.pred;
2121 cond = get_irn_n(projx, 0);
2122 projres = get_irn_n(cond, 0);
2123 loop_condition = get_irn_n(projres, 0);
2125 if (!is_Cmp(loop_condition))
2128 DB((dbg, LEVEL_5, "projection is %s\n", get_pnc_string(get_Proj_proj(projx))));
2130 switch(get_Proj_proj(projx)) {
2132 loop_info.exit_cond = 0;
2135 loop_info.exit_cond = 1;
2138 panic("Cond Proj_proj other than true/false");
2141 DB((dbg, LEVEL_4, "Valid Cmp.\n"));
2146 /* Returns 1 if all nodes are mode_Iu or mode_Is. */
2147 static unsigned are_mode_I(ir_node *n1, ir_node* n2, ir_node *n3)
2149 ir_mode *m1 = get_irn_mode(n1);
2150 ir_mode *m2 = get_irn_mode(n2);
2151 ir_mode *m3 = get_irn_mode(n3);
2153 if ((m1 == mode_Iu && m2 == mode_Iu && m3 == mode_Iu) ||
2154 (m1 == mode_Is && m2 == mode_Is && m3 == mode_Is))
2160 /* Checks if cur_loop is a simple tail-controlled counting loop
2161 * with start and end value loop invariant, step constant. */
2162 static unsigned get_unroll_decision_invariant(void)
2165 ir_node *projres, *loop_condition, *iteration_path;
2166 unsigned success, is_latest_val;
2167 ir_tarval *step_tar;
2171 /* RETURN if loop is not 'simple' */
2172 projres = is_simple_loop();
2173 if (projres == NULL)
2176 /* Use a minimal size for the invariant unrolled loop,
2177 * as duffs device produces overhead */
2178 if (loop_info.nodes < opt_params.invar_unrolling_min_size)
2181 loop_condition = get_irn_n(projres, 0);
2183 success = get_invariant_pred(loop_condition, &loop_info.end_val, &iteration_path);
2184 DB((dbg, LEVEL_4, "pred invar %d\n", success));
2189 DB((dbg, LEVEL_4, "Invariant End_val %N, other %N\n", loop_info.end_val, iteration_path));
2191 /* We may find the add or the phi first.
2192 * Until now we only have end_val. */
2193 if (is_Add(iteration_path) || is_Sub(iteration_path)) {
2195 /* We test against the latest value of the iv. */
2198 loop_info.add = iteration_path;
2199 DB((dbg, LEVEL_4, "Case 1: Got add %N (maybe not sane)\n", loop_info.add));
2201 /* Preds of the add should be step and the iteration_phi */
2202 success = get_const_pred(loop_info.add, &loop_info.step, &loop_info.iteration_phi);
2206 DB((dbg, LEVEL_4, "Got step %N\n", loop_info.step));
2208 if (! is_Phi(loop_info.iteration_phi))
2211 DB((dbg, LEVEL_4, "Got phi %N\n", loop_info.iteration_phi));
2214 * Does necessary sanity check of add, if it is already set. */
2215 success = get_start_and_add(loop_info.iteration_phi, invariant);
2219 DB((dbg, LEVEL_4, "Got start A %N\n", loop_info.start_val));
2221 } else if (is_Phi(iteration_path)) {
2222 ir_node *new_iteration_phi;
2224 /* We compare with the value the iv had entering this run. */
2227 loop_info.iteration_phi = iteration_path;
2228 DB((dbg, LEVEL_4, "Case 2: Got phi %N\n", loop_info.iteration_phi));
2230 /* Find start_val and add-node.
2231 * Does necessary sanity check of add, if it is already set. */
2232 success = get_start_and_add(loop_info.iteration_phi, invariant);
2236 DB((dbg, LEVEL_4, "Got start B %N\n", loop_info.start_val));
2237 DB((dbg, LEVEL_4, "Got add or sub %N\n", loop_info.add));
2239 success = get_const_pred(loop_info.add, &loop_info.step, &new_iteration_phi);
2243 DB((dbg, LEVEL_4, "Got step (B) %N\n", loop_info.step));
2245 if (loop_info.iteration_phi != new_iteration_phi)
2252 mode = get_irn_mode(loop_info.end_val);
2254 DB((dbg, LEVEL_4, "start %N, end %N, step %N\n",
2255 loop_info.start_val, loop_info.end_val, loop_info.step));
2257 if (mode != mode_Is && mode != mode_Iu)
2260 /* TODO necessary? */
2261 if (!are_mode_I(loop_info.start_val, loop_info.step, loop_info.end_val))
2264 DB((dbg, LEVEL_4, "mode integer\n"));
2266 step_tar = get_Const_tarval(loop_info.step);
2268 if (tarval_is_null(step_tar)) {
2269 /* TODO Might be worth a warning. */
2273 DB((dbg, LEVEL_4, "step is not 0\n"));
2275 create_duffs_block();
2277 return loop_info.max_unroll;
2280 /* Returns unroll factor,
2281 * given maximum unroll factor and number of loop passes. */
2282 static unsigned get_preferred_factor_constant(ir_tarval *count_tar)
2284 ir_tarval *tar_6, *tar_5, *tar_4, *tar_3, *tar_2;
2286 ir_mode *mode = get_irn_mode(loop_info.end_val);
2288 tar_6 = new_tarval_from_long(6, mode);
2289 tar_5 = new_tarval_from_long(5, mode);
2290 tar_4 = new_tarval_from_long(4, mode);
2291 tar_3 = new_tarval_from_long(3, mode);
2292 tar_2 = new_tarval_from_long(2, mode);
2294 /* loop passes % {6, 5, 4, 3, 2} == 0 */
2295 if (tarval_is_null(tarval_mod(count_tar, tar_6)))
2297 else if (tarval_is_null(tarval_mod(count_tar, tar_5)))
2299 else if (tarval_is_null(tarval_mod(count_tar, tar_4)))
2301 else if (tarval_is_null(tarval_mod(count_tar, tar_3)))
2303 else if (tarval_is_null(tarval_mod(count_tar, tar_2)))
2306 /* gcd(max_unroll, count_tar) */
2307 int a = loop_info.max_unroll;
2308 int b = (int)get_tarval_long(count_tar);
2311 DB((dbg, LEVEL_4, "gcd of max_unroll %d and count_tar %d: ", a, b));
2318 DB((dbg, LEVEL_4, "%d\n", a));
2322 DB((dbg, LEVEL_4, "preferred unroll factor %d\n", prefer));
2325 * If our preference is greater than the allowed unroll factor
2326 * we either might reduce the preferred factor and prevent a duffs device block,
2327 * or create a duffs device block, from which in this case (constants only)
2328 * we know the startloop at compiletime.
2329 * The latter yields the following graphs.
2330 * but for code generation we would want to use graph A.
2331 * The graphs are equivalent. So, we can only reduce the preferred factor.
2341 if (prefer <= loop_info.max_unroll)
2346 if (loop_info.max_unroll >= 3)
2348 else if (loop_info.max_unroll >= 2)
2354 if (loop_info.max_unroll >= 2)
2365 /* Check if cur_loop is a simple counting loop.
2366 * Start, step and end are constants.
2367 * TODO The whole constant case should use procedures similar to
2368 * the invariant case, as they are more versatile. */
2370 static unsigned get_unroll_decision_constant(void)
2372 ir_node *projres, *loop_condition, *iteration_path;
2373 unsigned success, is_latest_val;
2374 ir_tarval *start_tar, *end_tar, *step_tar, *diff_tar, *count_tar, *stepped;
2375 pn_Cmp proj_proj, norm_proj;
2378 /* RETURN if loop is not 'simple' */
2379 projres = is_simple_loop();
2380 if (projres == NULL)
2383 /* One in of the loop condition needs to be loop invariant. => end_val
2384 * The other in is assigned by an add. => add
2385 * The add uses a loop invariant value => step
2386 * and a phi with a loop invariant start_val and the add node as ins.
2398 loop_condition = get_irn_n(projres, 0);
2400 success = get_const_pred(loop_condition, &loop_info.end_val, &iteration_path);
2404 DB((dbg, LEVEL_4, "End_val %N, other %N\n", loop_info.end_val, iteration_path));
2406 /* We may find the add or the phi first.
2407 * Until now we only have end_val. */
2408 if (is_Add(iteration_path) || is_Sub(iteration_path)) {
2410 /* We test against the latest value of the iv. */
2413 loop_info.add = iteration_path;
2414 DB((dbg, LEVEL_4, "Case 2: Got add %N (maybe not sane)\n", loop_info.add));
2416 /* Preds of the add should be step and the iteration_phi */
2417 success = get_const_pred(loop_info.add, &loop_info.step, &loop_info.iteration_phi);
2421 DB((dbg, LEVEL_4, "Got step %N\n", loop_info.step));
2423 if (! is_Phi(loop_info.iteration_phi))
2426 DB((dbg, LEVEL_4, "Got phi %N\n", loop_info.iteration_phi));
2429 * Does necessary sanity check of add, if it is already set. */
2430 success = get_start_and_add(loop_info.iteration_phi, constant);
2434 DB((dbg, LEVEL_4, "Got start %N\n", loop_info.start_val));
2436 } else if (is_Phi(iteration_path)) {
2437 ir_node *new_iteration_phi;
2439 /* We compare with the value the iv had entering this run. */
2442 loop_info.iteration_phi = iteration_path;
2443 DB((dbg, LEVEL_4, "Case 1: Got phi %N \n", loop_info.iteration_phi));
2445 /* Find start_val and add-node.
2446 * Does necessary sanity check of add, if it is already set. */
2447 success = get_start_and_add(loop_info.iteration_phi, constant);
2451 DB((dbg, LEVEL_4, "Got start %N\n", loop_info.start_val));
2452 DB((dbg, LEVEL_4, "Got add or sub %N\n", loop_info.add));
2454 success = get_const_pred(loop_info.add, &loop_info.step, &new_iteration_phi);
2458 DB((dbg, LEVEL_4, "Got step %N\n", loop_info.step));
2460 if (loop_info.iteration_phi != new_iteration_phi)
2468 mode = get_irn_mode(loop_info.end_val);
2470 DB((dbg, LEVEL_4, "start %N, end %N, step %N\n",
2471 loop_info.start_val, loop_info.end_val, loop_info.step));
2473 if (mode != mode_Is && mode != mode_Iu)
2476 /* TODO necessary? */
2477 if (!are_mode_I(loop_info.start_val, loop_info.step, loop_info.end_val))
2480 DB((dbg, LEVEL_4, "mode integer\n"));
2482 end_tar = get_Const_tarval(loop_info.end_val);
2483 start_tar = get_Const_tarval(loop_info.start_val);
2484 step_tar = get_Const_tarval(loop_info.step);
2486 if (tarval_is_null(step_tar))
2487 /* TODO Might be worth a warning. */
2490 DB((dbg, LEVEL_4, "step is not 0\n"));
2492 if ((!tarval_is_negative(step_tar)) ^ (!is_Sub(loop_info.add)))
2493 loop_info.decreasing = 1;
2495 diff_tar = tarval_sub(end_tar, start_tar, mode);
2497 /* We need at least count_tar steps to be close to end_val, maybe more.
2498 * No way, that we have gone too many steps.
2499 * This represents the 'latest value'.
2500 * (If condition checks against latest value, is checked later) */
2501 count_tar = tarval_div(diff_tar, step_tar);
2503 /* Iv will not pass end_val (except overflows).
2504 * Nothing done, as it would yield to no advantage. */
2505 if (tarval_is_negative(count_tar)) {
2506 DB((dbg, LEVEL_4, "Loop is endless or never taken."));
2507 /* TODO Might be worth a warning. */
2511 count_stats(stats.u_simple_counting_loop);
2513 loop_info.latest_value = is_latest_val;
2516 if (! is_simple_counting_loop(&count_tar))
2520 /* stepped can be negative, if step < 0 */
2521 stepped = tarval_mul(count_tar, step_tar);
2523 /* step as close to end_val as possible, */
2524 /* |stepped| <= |end_tar|, and dist(stepped, end_tar) is smaller than a step. */
2525 if (is_Sub(loop_info.add))
2526 stepped = tarval_sub(start_tar, stepped, mode_Is);
2528 stepped = tarval_add(start_tar, stepped);
2530 DB((dbg, LEVEL_4, "stepped to %ld\n", get_tarval_long(stepped)));
2532 proj_proj = get_Proj_pn_cmp(projres);
2533 /* Assure that norm_proj is the stay-in-loop case. */
2534 if (loop_info.exit_cond == 1)
2535 norm_proj = get_math_inverted_case(proj_proj);
2537 norm_proj = proj_proj;
2539 DB((dbg, LEVEL_4, "normalized projection %s\n", get_pnc_string(norm_proj)));
2541 /* Executed at most once (stay in counting loop if a Eq b) */
2542 if (norm_proj == pn_Cmp_Eq)
2543 /* TODO Might be worth a warning. */
2546 /* calculates next values and increases count_tar according to it */
2547 success = simulate_next(&count_tar, stepped, step_tar, end_tar, norm_proj);
2551 /* We run loop once more, if we compare to the
2552 * not yet in-/decreased iv. */
2553 if (is_latest_val == 0) {
2554 DB((dbg, LEVEL_4, "condition uses not latest iv value\n"));
2555 count_tar = tarval_add(count_tar, get_tarval_one(mode));
2558 DB((dbg, LEVEL_4, "loop taken %ld times\n", get_tarval_long(count_tar)));
2560 /* Assure the loop is taken at least 1 time. */
2561 if (tarval_is_null(count_tar)) {
2562 /* TODO Might be worth a warning. */
2566 loop_info.count_tar = count_tar;
2567 return get_preferred_factor_constant(count_tar);
2573 static void unroll_loop(void)
2576 if (! (loop_info.nodes > 0))
2579 #if LOOP_IGNORE_NODE_LIMITS
2580 DB((dbg, LEVEL_1, "WARNING: Loop node limitations ignored."));
2582 if (loop_info.nodes > opt_params.max_unrolled_loop_size) {
2583 DB((dbg, LEVEL_2, "Nodes %d > allowed nodes %d\n",
2584 loop_info.nodes, opt_params.max_unrolled_loop_size));
2585 count_stats(stats.too_large);
2589 if (loop_info.calls > 0) {
2590 DB((dbg, LEVEL_2, "Calls %d > allowed calls 0\n",
2592 count_stats(stats.calls_limit);
2599 /* get_unroll_decision_constant and invariant are completely
2600 * independent for flexibility.
2601 * Some checks may be performed twice. */
2603 /* constant case? */
2604 if (opt_params.allow_const_unrolling)
2605 unroll_nr = get_unroll_decision_constant();
2606 if (unroll_nr > 1) {
2607 loop_info.unroll_kind = constant;
2610 /* invariant case? */
2611 if (opt_params.allow_invar_unrolling)
2612 unroll_nr = get_unroll_decision_invariant();
2614 loop_info.unroll_kind = invariant;
2617 DB((dbg, LEVEL_2, " *** Unrolling %d times ***\n", unroll_nr));
2619 if (unroll_nr > 1) {
2620 loop_entries = NEW_ARR_F(entry_edge, 0);
2623 irg_walk_graph(current_ir_graph, get_loop_entries, NULL, NULL);
2625 if (loop_info.unroll_kind == constant) {
2626 if ((int)get_tarval_long(loop_info.count_tar) == unroll_nr)
2627 loop_info.needs_backedge = 0;
2629 loop_info.needs_backedge = 1;
2631 loop_info.needs_backedge = 1;
2634 /* Use phase to keep copy of nodes from the condition chain. */
2635 phase = new_phase(current_ir_graph, phase_irn_init_default);
2637 /* Copies the loop */
2638 copy_loop(loop_entries, unroll_nr - 1);
2640 /* Line up the floating copies. */
2641 place_copies(unroll_nr - 1);
2643 /* Remove phis with 1 in
2644 * If there were no nested phis, this would not be necessary.
2645 * Avoiding the creation in the first place
2646 * leads to complex special cases. */
2647 irg_walk_graph(current_ir_graph, correct_phis, NULL, NULL);
2649 if (loop_info.unroll_kind == constant)
2650 count_stats(stats.constant_unroll);
2652 count_stats(stats.invariant_unroll);
2654 set_irg_doms_inconsistent(current_ir_graph);
2655 set_irg_loopinfo_inconsistent(current_ir_graph);
2657 set_irg_outs_inconsistent(current_ir_graph);
2659 DEL_ARR_F(loop_entries);
2664 /* Analyzes the loop, and checks if size is within allowed range.
2665 * Decides if loop will be processed. */
2666 static void init_analyze(ir_loop *loop)
2671 loop_head_valid = 1;
2673 /* Reset loop info */
2674 memset(&loop_info, 0, sizeof(loop_info_t));
2676 DB((dbg, LEVEL_1, " >>>> current loop includes node %N <<<\n",
2677 get_loop_node(loop, 0)));
2679 /* Collect loop informations: head, node counts. */
2680 irg_walk_graph(current_ir_graph, get_loop_info, NULL, NULL);
2682 /* RETURN if there is no valid head */
2683 if (!loop_head || !loop_head_valid) {
2684 DB((dbg, LEVEL_1, "No valid loop head. Nothing done.\n"));
2687 DB((dbg, LEVEL_1, "Loophead: %N\n", loop_head));
2690 if (loop_info.branches > opt_params.max_branches) {
2691 DB((dbg, LEVEL_1, "Branches %d > allowed branches %d\n",
2692 loop_info.branches, opt_params.max_branches));
2693 count_stats(stats.calls_limit);
2698 case loop_op_inversion:
2702 case loop_op_unrolling:
2707 panic("Loop optimization not implemented.");
2709 DB((dbg, LEVEL_1, " <<<< end of loop with node %N >>>>\n",
2710 get_loop_node(loop, 0)));
2713 /* Find innermost loops and add them to loops. */
2714 static void find_innermost_loop(ir_loop *loop)
2716 /* descend into sons */
2717 int sons = get_loop_n_sons(loop);
2720 ARR_APP1(ir_loop *, loops, loop);
2723 for (s=0; s<sons; s++) {
2724 find_innermost_loop(get_loop_son(loop, s));
2729 static void set_loop_params(void)
2731 opt_params.max_loop_size = 100;
2732 opt_params.depth_adaption = -50;
2733 opt_params.count_phi = 1;
2734 opt_params.count_proj = 0;
2735 opt_params.allowed_calls = 0;
2737 opt_params.max_cc_size = 5;
2740 opt_params.allow_const_unrolling = 1;
2741 opt_params.allow_invar_unrolling = 0;
2743 opt_params.invar_unrolling_min_size = 20;
2744 opt_params.max_unrolled_loop_size = 400;
2745 opt_params.max_branches = 9999;
2749 /* Assure preconditions are met and go through all loops. */
2750 void loop_optimization(ir_graph *irg)
2757 /* Reset stats for this procedure */
2761 set_current_ir_graph(irg);
2764 assure_irg_outs(irg);
2766 /* NOTE: sets only the loop attribute of blocks, not nodes */
2767 /* NOTE: Kills links */
2768 assure_cf_loop(irg);
2770 ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK | IR_RESOURCE_PHI_LIST);
2771 collect_phiprojs(irg);
2772 ir_free_resources(irg, IR_RESOURCE_IRN_LINK);
2774 loop = get_irg_loop(irg);
2775 sons = get_loop_n_sons(loop);
2777 loops = NEW_ARR_F(ir_loop *, 0);
2778 /* List all inner loops */
2779 for (nr = 0; nr < sons; ++nr) {
2780 find_innermost_loop(get_loop_son(loop, nr));
2783 ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK);
2784 /* Set all links to NULL */
2785 irg_walk_graph(current_ir_graph, reset_link, NULL, NULL);
2787 for (i = 0; i < ARR_LEN(loops); ++i) {
2788 ir_loop *loop = loops[i];
2790 count_stats(stats.loops);
2792 /* Analyze and handle loop */
2795 /* Copied blocks do not have their phi list yet */
2796 collect_phiprojs(irg);
2798 /* Set links to NULL
2799 * TODO Still necessary? */
2800 irg_walk_graph(current_ir_graph, reset_link, NULL, NULL);
2806 ir_free_resources(irg, IR_RESOURCE_IRN_LINK);
2807 ir_free_resources(irg, IR_RESOURCE_PHI_LIST);
2810 void do_loop_unrolling(ir_graph *irg)
2812 loop_op = loop_op_unrolling;
2814 DB((dbg, LEVEL_1, " >>> unrolling (Startnode %N) <<<\n",
2815 get_irg_start(irg)));
2817 loop_optimization(irg);
2819 DB((dbg, LEVEL_1, " >>> unrolling done (Startnode %N) <<<\n",
2820 get_irg_start(irg)));
2823 void do_loop_inversion(ir_graph *irg)
2825 loop_op = loop_op_inversion;
2827 DB((dbg, LEVEL_1, " >>> inversion (Startnode %N) <<<\n",
2828 get_irg_start(irg)));
2830 loop_optimization(irg);
2832 assure_cf_loop(irg);
2834 DB((dbg, LEVEL_1, " >>> inversion done (Startnode %N) <<<\n",
2835 get_irg_start(irg)));
2838 void do_loop_peeling(ir_graph *irg)
2840 loop_op = loop_op_peeling;
2842 DB((dbg, LEVEL_1, " >>> peeling (Startnode %N) <<<\n",
2843 get_irg_start(irg)));
2845 loop_optimization(irg);
2847 DB((dbg, LEVEL_1, " >>> peeling done (Startnode %N) <<<\n",
2848 get_irg_start(irg)));
2852 ir_graph_pass_t *loop_inversion_pass(const char *name)
2854 return def_graph_pass(name ? name : "loop_inversion", do_loop_inversion);
2857 ir_graph_pass_t *loop_unroll_pass(const char *name)
2859 return def_graph_pass(name ? name : "loop_unroll", do_loop_unrolling);
2862 ir_graph_pass_t *loop_peeling_pass(const char *name)
2864 return def_graph_pass(name ? name : "loop_peeling", do_loop_peeling);
2867 void firm_init_loop_opt(void)
2869 FIRM_DBG_REGISTER(dbg, "firm.opt.loop");