2 * Copyright (C) 1995-2011 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
29 #include "iroptimize.h"
48 #include "irbackedge_t.h"
49 #include "irphase_t.h"
53 DEBUG_ONLY(static firm_dbg_module_t *dbg);
55 /* DBG print stats for every procedure. */
56 #define LOOP_OPT_STATS 1
58 /* DBG: Ignore node limits and process every possible loop. */
59 #define LOOP_IGNORE_NODE_LIMITS 0
62 * Convenience macro for iterating over every phi node of the given block.
63 * Requires phi list per block.
65 #define for_each_phi(block, phi) \
66 for ((phi) = get_Block_phis( (block) ); (phi) ; (phi) = get_Phi_next((phi)))
68 #define for_each_phi_safe(head, phi, next) \
69 for ((phi) = (head), (next) = (head) ? get_Phi_next((head)) : NULL; \
70 (phi) ; (phi) = (next), (next) = (next) ? get_Phi_next((next)) : NULL)
72 /* Currently processed loop. */
73 static ir_loop *cur_loop;
75 /* Flag for kind of unrolling. */
79 } unrolling_kind_flag;
81 /* Condition for performing visiting a node during copy_walk. */
82 typedef unsigned walker_condition(ir_node *);
84 /* Node and position of a predecessor. */
85 typedef struct entry_edge {
91 /* Node info for unrolling. */
92 typedef struct unrolling_node_info {
94 } unrolling_node_info;
96 /* Outs of the nodes head. */
97 static entry_edge *cur_head_outs;
99 /* Information about the loop head */
100 static ir_node *loop_head = NULL;
101 static unsigned loop_head_valid = 1;
103 /* List of all inner loops, that are processed. */
104 static ir_loop **loops;
108 #define count_stats(val) (++val)
109 #define print_stats() (do_print_stats())
110 #define reset_stats() (do_reset_stats())
113 typedef struct loop_stats_t {
117 unsigned too_large_adapted;
118 unsigned cc_limit_reached;
119 unsigned calls_limit;
121 unsigned u_simple_counting_loop;
122 unsigned constant_unroll;
123 unsigned invariant_unroll;
128 static loop_stats_t stats;
130 /* Set stats to sero */
131 static void do_reset_stats(void)
133 memset(&stats, 0, sizeof(loop_stats_t));
137 static void do_print_stats(void)
139 DB((dbg, LEVEL_2, "---------------------------------------\n"));
140 DB((dbg, LEVEL_2, "loops : %d\n",stats.loops));
141 DB((dbg, LEVEL_2, "inverted : %d\n",stats.inverted));
142 DB((dbg, LEVEL_2, "too_large : %d\n",stats.too_large));
143 DB((dbg, LEVEL_2, "too_large_adapted : %d\n",stats.too_large_adapted));
144 DB((dbg, LEVEL_2, "cc_limit_reached : %d\n",stats.cc_limit_reached));
145 DB((dbg, LEVEL_2, "calls_limit : %d\n",stats.calls_limit));
146 DB((dbg, LEVEL_2, "u_simple_counting : %d\n",stats.u_simple_counting_loop));
147 DB((dbg, LEVEL_2, "constant_unroll : %d\n",stats.constant_unroll));
148 DB((dbg, LEVEL_2, "invariant_unroll : %d\n",stats.invariant_unroll));
149 DB((dbg, LEVEL_2, "=======================================\n"));
153 #define count_stats(val) ((void)0)
154 #define print_stats() ((void)0)
155 #define reset_stats() ((void)0)
159 /* Commandline parameters */
160 typedef struct loop_opt_params_t {
161 unsigned max_loop_size; /* Maximum number of nodes [nodes]*/
162 int depth_adaption; /* Loop nest depth adaption [percent] */
163 unsigned allowed_calls; /* Number of calls allowed [number] */
164 unsigned count_phi:1; /* Count phi nodes */
165 unsigned count_proj:1; /* Count projections */
167 unsigned max_cc_size; /* Maximum condition chain size [nodes] */
168 unsigned max_branches;
170 unsigned max_unrolled_loop_size; /* [nodes] */
171 unsigned allow_const_unrolling:1;
172 unsigned allow_invar_unrolling:1;
173 unsigned invar_unrolling_min_size; /* [nodes] */
177 static loop_opt_params_t opt_params;
179 /* Loop analysis informations */
180 typedef struct loop_info_t {
181 unsigned nodes; /* node count */
182 unsigned ld_st; /* load and store nodes */
183 unsigned branches; /* number of conditions */
184 unsigned calls; /* number of calls */
185 unsigned cf_outs; /* number of cf edges which leave the loop */
186 entry_edge cf_out; /* single loop leaving cf edge */
187 int be_src_pos; /* position of the single own backedge in the head */
190 unsigned cc_size; /* nodes in the condition chain */
193 unsigned max_unroll; /* Number of unrolls satisfying max_loop_size */
194 unsigned exit_cond; /* 1 if condition==true exits the loop. */
195 unsigned latest_value:1; /* 1 if condition is checked against latest counter value */
196 unsigned needs_backedge:1; /* 0 if loop is completely unrolled */
197 unsigned decreasing:1; /* Step operation is_Sub, or step is<0 */
199 /* IV informations of a simple loop */
203 ir_node *iteration_phi;
206 ir_tarval *count_tar; /* Number of loop iterations */
208 ir_node *duff_cond; /* Duff mod */
209 unrolling_kind_flag unroll_kind; /* constant or invariant unrolling */
212 /* Information about the current loop */
213 static loop_info_t loop_info;
215 /* Outs of the condition chain (loop inversion). */
216 static ir_node **cc_blocks;
217 /* df/cf edges with def in the condition chain */
218 static entry_edge *cond_chain_entries;
219 /* Array of df loops found in the condition chain. */
220 static entry_edge *head_df_loop;
221 /* Number of blocks in cc */
222 static unsigned inversion_blocks_in_cc;
225 /* Cf/df edges leaving the loop.
226 * Called entries here, as they are used to enter the loop with walkers. */
227 static entry_edge *loop_entries;
228 /* Number of unrolls to perform */
229 static int unroll_nr;
230 /* Phase is used to keep copies of nodes. */
231 static ir_phase *phase;
233 /* Loop operations. */
234 typedef enum loop_op_t {
240 /* Saves which loop operation to do until after basic tests. */
241 static loop_op_t loop_op;
243 /* Returns the maximum nodes for the given nest depth */
244 static unsigned get_max_nodes_adapted(unsigned depth)
246 double perc = 100.0 + (double)opt_params.depth_adaption;
247 double factor = pow(perc / 100.0, depth);
249 return (int)((double)opt_params.max_loop_size * factor);
252 /* Reset nodes link. For use with a walker. */
253 static void reset_link(ir_node *node, void *env)
256 set_irn_link(node, NULL);
259 /* Returns 0 if the node or block is not in cur_loop. */
260 static unsigned is_in_loop(ir_node *node)
262 return (get_irn_loop(get_block(node)) == cur_loop);
265 /* Returns 0 if the given edge is not a backedge
266 * with its pred in the cur_loop. */
267 static unsigned is_own_backedge(ir_node *n, int pos)
269 return (is_backedge(n, pos) && is_in_loop(get_irn_n(n, pos)));
272 /* Finds loop head and some loop_info as calls or else if necessary. */
273 static void get_loop_info(ir_node *node, void *env)
275 unsigned node_in_loop, pred_in_loop;
279 arity = get_irn_arity(node);
280 for (i = 0; i < arity; i++) {
281 ir_node *pred = get_irn_n(node, i);
283 pred_in_loop = is_in_loop(pred);
284 node_in_loop = is_in_loop(node);
286 if (!node_in_loop && pred_in_loop && is_Block(node))
294 loop_info.cf_out = entry;
297 /* collect some loop information */
299 if (is_Phi(node) && opt_params.count_phi)
301 else if (is_Proj(node) && opt_params.count_proj)
303 else if (!is_Confirm(node) && !is_Const(node) && !is_SymConst(node))
306 if (is_Load(node) || is_Store(node))
314 /* Find the loops head/the blocks with cfpred outside of the loop */
315 if (is_Block(node)) {
316 const ir_edge_t *edge;
319 /* Count innerloop branches */
320 foreach_out_edge_kind(node, edge, EDGE_KIND_NORMAL) {
321 if (is_Block(get_edge_src_irn(edge)) && is_in_loop(get_edge_src_irn(edge)))
325 ++loop_info.branches;
327 if (node_in_loop && !pred_in_loop && loop_head_valid) {
328 ir_node *cfgpred = get_Block_cfgpred(node, i);
330 if (!is_in_loop(cfgpred)) {
331 DB((dbg, LEVEL_5, "potential head %+F because inloop and pred %+F not inloop\n",
333 /* another head? We do not touch this. */
334 if (loop_head && loop_head != node) {
345 /* Finds all edges with users outside of the loop
346 * and definition inside the loop. */
347 static void get_loop_entries(ir_node *node, void *env)
349 unsigned node_in_loop, pred_in_loop;
353 arity = get_irn_arity(node);
354 for (i = 0; i < arity; ++i) {
355 ir_node *pred = get_irn_n(node, i);
357 pred_in_loop = is_in_loop(pred);
358 node_in_loop = is_in_loop(node);
360 if (pred_in_loop && !node_in_loop) {
365 ARR_APP1(entry_edge, loop_entries, entry);
371 static ir_node *ssa_second_def;
372 static ir_node *ssa_second_def_block;
375 * Walks the graph bottom up, searching for definitions and creates phis.
377 static ir_node *search_def_and_create_phis(ir_node *block, ir_mode *mode, int first)
381 ir_graph *irg = get_irn_irg(block);
385 DB((dbg, LEVEL_5, "ssa search_def_and_create_phis: block %N\n", block));
387 /* Prevents creation of phi that would be bad anyway.
388 * Dead and bad blocks. */
389 if (get_irn_arity(block) < 1 || is_Bad(block)) {
390 DB((dbg, LEVEL_5, "ssa bad %N\n", block));
391 return new_r_Bad(irg, mode);
394 if (block == ssa_second_def_block && !first) {
395 DB((dbg, LEVEL_5, "ssa found second definition: use second def %N\n", ssa_second_def));
396 return ssa_second_def;
399 /* already processed this block? */
400 if (irn_visited(block)) {
401 ir_node *value = (ir_node *) get_irn_link(block);
402 DB((dbg, LEVEL_5, "ssa already visited: use linked %N\n", value));
406 assert(block != get_irg_start_block(irg));
408 /* a Block with only 1 predecessor needs no Phi */
409 n_cfgpreds = get_Block_n_cfgpreds(block);
410 if (n_cfgpreds == 1) {
411 ir_node *pred_block = get_Block_cfgpred_block(block, 0);
414 DB((dbg, LEVEL_5, "ssa 1 pred: walk pred %N\n", pred_block));
416 value = search_def_and_create_phis(pred_block, mode, 0);
417 set_irn_link(block, value);
418 mark_irn_visited(block);
423 /* create a new Phi */
424 NEW_ARR_A(ir_node*, in, n_cfgpreds);
425 for (i = 0; i < n_cfgpreds; ++i)
426 in[i] = new_r_Dummy(irg, mode);
428 phi = new_r_Phi(block, n_cfgpreds, in, mode);
429 /* Important: always keep block phi list up to date. */
430 add_Block_phi(block, phi);
431 DB((dbg, LEVEL_5, "ssa phi creation: link new phi %N to block %N\n", phi, block));
432 set_irn_link(block, phi);
433 mark_irn_visited(block);
435 /* set Phi predecessors */
436 for (i = 0; i < n_cfgpreds; ++i) {
438 ir_node *pred_block = get_Block_cfgpred_block(block, i);
439 assert(pred_block != NULL);
440 pred_val = search_def_and_create_phis(pred_block, mode, 0);
442 assert(pred_val != NULL);
444 DB((dbg, LEVEL_5, "ssa phi pred:phi %N, pred %N\n", phi, pred_val));
445 set_irn_n(phi, i, pred_val);
453 * Given a set of values this function constructs SSA-form for the users of the
454 * first value (the users are determined through the out-edges of the value).
455 * Works without using the dominance tree.
457 static void construct_ssa(ir_node *orig_block, ir_node *orig_val,
458 ir_node *second_block, ir_node *second_val)
462 const ir_edge_t *edge;
463 const ir_edge_t *next;
465 assert(orig_block && orig_val && second_block && second_val &&
466 "no parameter of construct_ssa may be NULL");
468 if (orig_val == second_val)
471 irg = get_irn_irg(orig_val);
473 ir_reserve_resources(irg, IR_RESOURCE_IRN_VISITED);
474 inc_irg_visited(irg);
476 mode = get_irn_mode(orig_val);
477 set_irn_link(orig_block, orig_val);
478 mark_irn_visited(orig_block);
480 ssa_second_def_block = second_block;
481 ssa_second_def = second_val;
483 /* Only fix the users of the first, i.e. the original node */
484 foreach_out_edge_safe(orig_val, edge, next) {
485 ir_node *user = get_edge_src_irn(edge);
486 int j = get_edge_src_pos(edge);
487 ir_node *user_block = get_nodes_block(user);
494 DB((dbg, LEVEL_5, "original user %N\n", user));
497 ir_node *pred_block = get_Block_cfgpred_block(user_block, j);
498 newval = search_def_and_create_phis(pred_block, mode, 1);
500 newval = search_def_and_create_phis(user_block, mode, 1);
502 if (newval != user && !is_Bad(newval))
503 set_irn_n(user, j, newval);
506 ir_free_resources(irg, IR_RESOURCE_IRN_VISITED);
510 /***** Unrolling Helper Functions *****/
512 /* Assign the copy with index nr to node n */
513 static void set_unroll_copy(ir_node *n, int nr, ir_node *cp)
515 unrolling_node_info *info;
516 assert(nr != 0 && "0 reserved");
518 info = (unrolling_node_info *)phase_get_irn_data(phase, n);
522 info = XMALLOCZ(unrolling_node_info);
523 arr = NEW_ARR_F(ir_node *, unroll_nr);
525 memset(info->copies, 0, (unroll_nr) * sizeof(ir_node *));
527 phase_set_irn_data(phase, n, info);
532 info->copies[nr] = cp;
535 /* Returns a nodes copy if it exists, else NULL. */
536 static ir_node *get_unroll_copy(ir_node *n, int nr)
539 unrolling_node_info *info = (unrolling_node_info *)phase_get_irn_data(phase, n);
543 cp = info->copies[nr];
548 /***** Inversion Helper Functions *****/
550 /* Sets copy cp of node n. */
551 static void set_inversion_copy(ir_node *n, ir_node *cp)
553 phase_set_irn_data(phase, n, cp);
556 /* Getter of copy of n for inversion */
557 static ir_node *get_inversion_copy(ir_node *n)
559 ir_node *cp = (ir_node *)phase_get_irn_data(phase, n);
563 /* Resets block mark for given node. For use with walker */
564 static void reset_block_mark(ir_node *node, void * env)
569 set_Block_mark(node, 0);
572 /* Returns mark of node, or its block if node is not a block.
573 * Used in this context to determine if node is in the condition chain. */
574 static unsigned is_nodes_block_marked(ir_node* node)
577 return get_Block_mark(node);
579 return get_Block_mark(get_block(node));
582 /* Extends a nodes ins by node new.
583 * NOTE: This is slow if a node n needs to be extended more than once. */
584 static void extend_irn(ir_node *n, ir_node *newnode, int new_is_backedge)
588 int arity = get_irn_arity(n);
589 int new_arity = arity + 1;
592 NEW_ARR_A(int, bes, new_arity);
593 NEW_ARR_A(ir_node *, ins, new_arity);
596 /* Bes are important!
597 * Another way would be recreating the looptree,
598 * but after that we cannot distinguish already processed loops
599 * from not yet processed ones. */
601 for(i = 0; i < arity; ++i) {
602 bes[i] = is_backedge(n, i);
604 bes[i] = new_is_backedge;
607 for(i = 0; i < arity; ++i) {
608 ins[i] = get_irn_n(n, i);
612 set_irn_in(n, new_arity, ins);
616 for(i = 0; i < new_arity; ++i) {
623 /* Extends a block by a copy of its pred at pos,
624 * fixing also the phis in the same way. */
625 static void extend_ins_by_copy(ir_node *block, int pos)
630 assert(is_Block(block));
632 /* Extend block by copy of definition at pos */
633 pred = get_irn_n(block, pos);
634 new_in = get_inversion_copy(pred);
635 DB((dbg, LEVEL_5, "Extend block %N by %N cp of %N\n", block, new_in, pred));
636 extend_irn(block, new_in, 0);
638 /* Extend block phis by copy of definition at pos */
639 for_each_phi(block, phi) {
642 pred = get_irn_n(phi, pos);
643 cp = get_inversion_copy(pred);
644 /* If the phis in is not in the condition chain (eg. a constant),
645 * there is no copy. */
651 DB((dbg, LEVEL_5, "Extend phi %N by %N cp of %N\n", phi, new_in, pred));
652 extend_irn(phi, new_in, 0);
656 /* Returns the number of blocks backedges. With or without alien bes. */
657 static int get_backedge_n(ir_node *block, unsigned with_alien)
661 int arity = get_irn_arity(block);
663 assert(is_Block(block));
665 for (i = 0; i < arity; ++i) {
666 ir_node *pred = get_irn_n(block, i);
667 if (is_backedge(block, i) && (with_alien || is_in_loop(pred)))
673 /* Returns a raw copy of the given node.
674 * Attributes are kept/set according to the needs of loop inversion. */
675 static ir_node *copy_node(ir_node *node)
680 cp = exact_copy(node);
681 arity = get_irn_arity(node);
683 /* Keep backedge info */
684 for (i = 0; i < arity; ++i) {
685 if (is_backedge(node, i))
690 set_Block_mark(cp, 0);
698 * This walker copies all walked nodes.
699 * If the walk_condition is true for a node, it is copied.
700 * All nodes node_info->copy have to be NULL prior to every walk.
701 * Order of ins is important for later usage.
703 static void copy_walk(ir_node *node, walker_condition *walk_condition,
710 ir_graph *irg = current_ir_graph;
713 * break condition and cycle resolver, creating temporary node copies
715 if (get_irn_visited(node) >= get_irg_visited(irg)) {
716 /* Here we rely on nodestate's copy being initialized with NULL */
717 DB((dbg, LEVEL_5, "copy_walk: We have already visited %N\n", node));
718 if (get_inversion_copy(node) == NULL) {
719 cp = copy_node(node);
720 set_inversion_copy(node, cp);
722 DB((dbg, LEVEL_5, "The TEMP copy of %N is created %N\n", node, cp));
728 mark_irn_visited(node);
730 if (!is_Block(node)) {
731 ir_node *pred = get_nodes_block(node);
732 if (walk_condition(pred))
733 DB((dbg, LEVEL_5, "walk block %N\n", pred));
734 copy_walk(pred, walk_condition, set_loop);
737 arity = get_irn_arity(node);
739 NEW_ARR_A(ir_node *, cpin, arity);
741 for (i = 0; i < arity; ++i) {
742 ir_node *pred = get_irn_n(node, i);
744 if (walk_condition(pred)) {
745 DB((dbg, LEVEL_5, "walk node %N\n", pred));
746 copy_walk(pred, walk_condition, set_loop);
747 cpin[i] = get_inversion_copy(pred);
748 DB((dbg, LEVEL_5, "copy of %N gets new in %N which is copy of %N\n",
749 node, get_inversion_copy(pred), pred));
755 /* copy node / finalize temp node */
756 if (get_inversion_copy(node) == NULL) {
757 /* No temporary copy existent */
758 cp = copy_node(node);
759 set_inversion_copy(node, cp);
760 DB((dbg, LEVEL_5, "The FINAL copy of %N is CREATED %N\n", node, cp));
762 /* temporary copy is existent but without correct ins */
763 cp = get_inversion_copy(node);
764 DB((dbg, LEVEL_5, "The FINAL copy of %N is EXISTENT %N\n", node, cp));
767 if (!is_Block(node)) {
768 ir_node *cpblock = get_inversion_copy(get_nodes_block(node));
770 set_nodes_block(cp, cpblock );
772 add_Block_phi(cpblock, cp);
775 /* Keeps phi list of temporary node. */
776 set_irn_in(cp, ARR_LEN(cpin), cpin);
780 * This walker copies all walked nodes.
781 * If the walk_condition is true for a node, it is copied.
782 * All nodes node_info->copy have to be NULL prior to every walk.
783 * Order of ins is important for later usage.
784 * Takes copy_index, to phase-link copy at specific index.
786 static void copy_walk_n(ir_node *node, walker_condition *walk_condition,
795 * break condition and cycle resolver, creating temporary node copies
797 if (irn_visited(node)) {
798 /* Here we rely on nodestate's copy being initialized with NULL */
799 DB((dbg, LEVEL_5, "copy_walk: We have already visited %N\n", node));
800 if (get_unroll_copy(node, copy_index) == NULL) {
803 set_unroll_copy(node, copy_index, u);
804 DB((dbg, LEVEL_5, "The TEMP unknown of %N is created %N\n", node, u));
810 mark_irn_visited(node);
812 if (!is_Block(node)) {
813 ir_node *block = get_nodes_block(node);
814 if (walk_condition(block))
815 DB((dbg, LEVEL_5, "walk block %N\n", block));
816 copy_walk_n(block, walk_condition, copy_index);
819 arity = get_irn_arity(node);
820 NEW_ARR_A(ir_node *, cpin, arity);
822 for (i = 0; i < arity; ++i) {
823 ir_node *pred = get_irn_n(node, i);
825 if (walk_condition(pred)) {
826 DB((dbg, LEVEL_5, "walk node %N\n", pred));
827 copy_walk_n(pred, walk_condition, copy_index);
828 cpin[i] = get_unroll_copy(pred, copy_index);
834 /* copy node / finalize temp node */
835 cp = get_unroll_copy(node, copy_index);
836 if (cp == NULL || is_Unknown(cp)) {
837 cp = copy_node(node);
838 set_unroll_copy(node, copy_index, cp);
839 DB((dbg, LEVEL_5, "The FINAL copy of %N is CREATED %N\n", node, cp));
841 /* temporary copy is existent but without correct ins */
842 cp = get_unroll_copy(node, copy_index);
843 DB((dbg, LEVEL_5, "The FINAL copy of %N is EXISTENT %N\n", node, cp));
846 if (!is_Block(node)) {
847 ir_node *cpblock = get_unroll_copy(get_nodes_block(node), copy_index);
849 set_nodes_block(cp, cpblock );
851 add_Block_phi(cpblock, cp);
854 /* Keeps phi list of temporary node. */
855 set_irn_in(cp, ARR_LEN(cpin), cpin);
858 /* Removes alle Blocks with non marked predecessors from the condition chain. */
859 static void unmark_not_allowed_cc_blocks(void)
861 size_t blocks = ARR_LEN(cc_blocks);
864 for(i = 0; i < blocks; ++i) {
865 ir_node *block = cc_blocks[i];
867 int arity = get_irn_arity(block);
869 /* Head is an exception. */
870 if (block == loop_head)
873 for(a = 0; a < arity; ++a) {
874 if (! is_nodes_block_marked(get_irn_n(block, a))) {
875 set_Block_mark(block, 0);
876 --inversion_blocks_in_cc;
877 DB((dbg, LEVEL_5, "Removed %N from cc (blocks in cc %d)\n",
878 block, inversion_blocks_in_cc));
886 /* Unmarks all cc blocks using cc_blocks except head.
887 * TODO: invert head for unrolling? */
888 static void unmark_cc_blocks(void)
890 size_t blocks = ARR_LEN(cc_blocks);
893 for(i = 0; i < blocks; ++i) {
894 ir_node *block = cc_blocks[i];
896 /* TODO Head is an exception. */
897 /*if (block != loop_head)*/
898 set_Block_mark(block, 0);
900 /*inversion_blocks_in_cc = 1;*/
901 inversion_blocks_in_cc = 0;
904 loop_info.cc_size = 0;
908 * Populates head_entries with (node, pred_pos) tuple
909 * whereas the node's pred at pred_pos is in the cc but not the node itself.
910 * Also finds df loops inside the cc.
911 * Head and condition chain blocks have been marked previously.
913 static void get_head_outs(ir_node *node, void *env)
916 int arity = get_irn_arity(node);
919 for (i = 0; i < arity; ++i) {
920 if (!is_nodes_block_marked(node) && is_nodes_block_marked(get_irn_n(node, i))) {
924 /* Saving also predecessor seems redundant, but becomes
925 * necessary when changing position of it, before
926 * dereferencing it.*/
927 entry.pred = get_irn_n(node, i);
928 ARR_APP1(entry_edge, cur_head_outs, entry);
932 arity = get_irn_arity(loop_head);
934 /* Find df loops inside the cc */
935 if (is_Phi(node) && get_nodes_block(node) == loop_head) {
936 for (i = 0; i < arity; ++i) {
937 if (is_own_backedge(loop_head, i)) {
938 if (is_nodes_block_marked(get_irn_n(node, i))) {
942 entry.pred = get_irn_n(node, i);
943 ARR_APP1(entry_edge, head_df_loop, entry);
944 DB((dbg, LEVEL_5, "Found incc assignment node %N @%d is pred %N, graph %N %N\n",
945 node, i, entry.pred, current_ir_graph, get_irg_start_block(current_ir_graph)));
953 * Find condition chains, and add them to be inverted.
954 * A block belongs to the chain if a condition branches out of the loop.
955 * (Some blocks need to be removed once again.)
956 * Returns 1 if the given block belongs to the condition chain.
958 static void find_condition_chain(ir_node *block)
960 const ir_edge_t *edge;
964 unsigned nodes_n = 0;
966 mark_irn_visited(block);
968 DB((dbg, LEVEL_5, "condition_chains for block %N\n", block));
971 foreach_out_edge_kind(block, edge, EDGE_KIND_NORMAL) {
975 /* Check if node count would exceed maximum cc size.
977 * This is not optimal, as we search depth-first and break here,
978 * continuing with another subtree. */
979 if (loop_info.cc_size + nodes_n > opt_params.max_cc_size) {
980 set_Block_mark(block, 0);
984 /* Check if block only has a jmp instruction. */
986 foreach_out_edge(block, edge) {
987 ir_node *src = get_edge_src_irn(edge);
989 if (! is_Block(src) && ! is_Jmp(src)) {
994 /* Check cf outs if one is leaving the loop,
995 * or if this node has a backedge. */
996 foreach_block_succ(block, edge) {
997 ir_node *src = get_edge_src_irn(edge);
998 int pos = get_edge_src_pos(edge);
1000 if (! is_in_loop(src))
1003 /* Inverting blocks with backedge outs leads to a cf edge
1004 * from the inverted head, into the inverted head (skipping the body).
1005 * As the body becomes the new loop head,
1006 * this would introduce another loop in the existing loop.
1007 * This loop inversion cannot cope with this case. */
1008 if (is_backedge(src, pos)) {
1014 /* We need all predecessors to already belong to the condition chain.
1015 * Example of wrong case: * == in cc
1026 /* Collect blocks containing only a Jmp.
1027 * Do not collect blocks with backedge outs. */
1028 if ((jmp_only == 1 || mark == 1) && has_be == 0) {
1029 set_Block_mark(block, 1);
1030 ++inversion_blocks_in_cc;
1031 loop_info.cc_size += nodes_n;
1032 DB((dbg, LEVEL_5, "block %N is part of condition chain\n", block));
1033 ARR_APP1(ir_node *, cc_blocks, block);
1035 set_Block_mark(block, 0);
1038 foreach_block_succ(block, edge) {
1039 ir_node *src = get_edge_src_irn( edge );
1041 if (is_in_loop(src) && ! irn_visited(src))
1042 find_condition_chain(src);
1047 * Rewires the copied condition chain. Removes backedges
1048 * as this condition chain is prior to the loop.
1049 * Copy of loop_head must have phi list and old (unfixed) backedge info of the loop head.
1050 * (loop_head is already fixed, we cannot rely on it.)
1052 static void fix_copy_inversion(void)
1057 ir_node *phi, *next;
1058 ir_node *head_cp = get_inversion_copy(loop_head);
1059 ir_graph *irg = get_irn_irg(head_cp);
1060 int arity = get_irn_arity(head_cp);
1061 int backedges = get_backedge_n(head_cp, 0);
1062 int new_arity = arity - backedges;
1066 NEW_ARR_A(ir_node *, ins, new_arity);
1069 /* Remove block backedges */
1070 for(i = 0; i < arity; ++i) {
1071 if (!is_backedge(head_cp, i))
1072 ins[pos++] = get_irn_n(head_cp, i);
1075 new_head = new_r_Block(irg, new_arity, ins);
1077 phis = NEW_ARR_F(ir_node *, 0);
1079 for_each_phi_safe(get_Block_phis(head_cp), phi, next) {
1081 NEW_ARR_A(ir_node *, ins, new_arity);
1083 for(i = 0; i < arity; ++i) {
1084 if (!is_backedge(head_cp, i))
1085 ins[pos++] = get_irn_n(phi, i);
1087 new_phi = new_rd_Phi(get_irn_dbg_info(phi),
1088 new_head, new_arity, ins,
1090 ARR_APP1(ir_node *, phis, new_phi);
1094 for_each_phi_safe(get_Block_phis(head_cp), phi, next) {
1095 exchange(phi, phis[pos++]);
1098 exchange(head_cp, new_head);
1104 /* Puts the original condition chain at the end of the loop,
1105 * subsequently to the body.
1106 * Relies on block phi list and correct backedges.
1108 static void fix_head_inversion(void)
1112 ir_node *phi, *next;
1114 ir_graph *irg = get_irn_irg(loop_head);
1115 int arity = get_irn_arity(loop_head);
1116 int backedges = get_backedge_n(loop_head, 0);
1117 int new_arity = backedges;
1121 NEW_ARR_A(ir_node *, ins, new_arity);
1124 /* Keep only backedges */
1125 for(i = 0; i < arity; ++i) {
1126 if (is_own_backedge(loop_head, i))
1127 ins[pos++] = get_irn_n(loop_head, i);
1130 new_head = new_r_Block(irg, new_arity, ins);
1132 phis = NEW_ARR_F(ir_node *, 0);
1134 for_each_phi(loop_head, phi) {
1136 DB((dbg, LEVEL_5, "Fixing phi %N of loop head\n", phi));
1138 NEW_ARR_A(ir_node *, ins, new_arity);
1141 for (i = 0; i < arity; ++i) {
1142 ir_node *pred = get_irn_n(phi, i);
1144 if (is_own_backedge(loop_head, i)) {
1145 /* If assignment is in the condition chain,
1146 * we need to create a phi in the new loop head.
1147 * This can only happen for df, not cf. See find_condition_chains. */
1148 /*if (is_nodes_block_marked(pred)) {
1156 new_phi = new_rd_Phi(get_irn_dbg_info(phi),
1157 new_head, new_arity, ins,
1160 ARR_APP1(ir_node *, phis, new_phi);
1162 DB((dbg, LEVEL_5, "fix inverted head should exch %N by %N (pos %d)\n", phi, new_phi, pos ));
1166 for_each_phi_safe(get_Block_phis(loop_head), phi, next) {
1167 DB((dbg, LEVEL_5, "fix inverted exch phi %N by %N\n", phi, phis[pos]));
1168 if (phis[pos] != phi)
1169 exchange(phi, phis[pos++]);
1174 DB((dbg, LEVEL_5, "fix inverted head exch head block %N by %N\n", loop_head, new_head));
1175 exchange(loop_head, new_head);
1178 /* Does the loop inversion. */
1179 static void inversion_walk(entry_edge *head_entries)
1184 * The order of rewiring bottom-up is crucial.
1185 * Any change of the order leads to lost information that would be needed later.
1188 ir_reserve_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1190 /* 1. clone condition chain */
1191 inc_irg_visited(current_ir_graph);
1193 for (i = 0; i < ARR_LEN(head_entries); ++i) {
1194 entry_edge entry = head_entries[i];
1195 ir_node *pred = get_irn_n(entry.node, entry.pos);
1197 DB((dbg, LEVEL_5, "\nInit walk block %N\n", pred));
1199 copy_walk(pred, is_nodes_block_marked, cur_loop);
1202 ir_free_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1204 /* 2. Extends the head control flow successors ins
1205 * with the definitions of the copied head node. */
1206 for (i = 0; i < ARR_LEN(head_entries); ++i) {
1207 entry_edge head_out = head_entries[i];
1209 if (is_Block(head_out.node))
1210 extend_ins_by_copy(head_out.node, head_out.pos);
1213 /* 3. construct_ssa for users of definitions in the condition chain,
1214 * as there is now a second definition. */
1215 for (i = 0; i < ARR_LEN(head_entries); ++i) {
1216 entry_edge head_out = head_entries[i];
1218 /* Ignore keepalives */
1219 if (is_End(head_out.node))
1222 /* Construct ssa for assignments in the condition chain. */
1223 if (!is_Block(head_out.node)) {
1224 ir_node *pred, *cppred, *block, *cpblock;
1226 pred = head_out.pred;
1227 cppred = get_inversion_copy(pred);
1228 block = get_nodes_block(pred);
1229 cpblock = get_nodes_block(cppred);
1230 construct_ssa(block, pred, cpblock, cppred);
1235 * If there is an assignment in the condition chain
1236 * with a user also in the condition chain,
1237 * the dominance frontier is in the new loop head.
1238 * The dataflow loop is completely in the condition chain.
1243 * Phi_cp | | copied condition chain
1247 * Phi* | | new loop head with newly created phi.
1249 * Phi | | original, inverted condition chain
1254 for (i = 0; i < ARR_LEN(head_df_loop); ++i) {
1255 entry_edge head_out = head_df_loop[i];
1257 /* Construct ssa for assignments in the condition chain. */
1258 ir_node *pred, *cppred, *block, *cpblock;
1260 pred = head_out.pred;
1261 cppred = get_inversion_copy(pred);
1262 assert(cppred && pred);
1263 block = get_nodes_block(pred);
1264 cpblock = get_nodes_block(cppred);
1265 construct_ssa(block, pred, cpblock, cppred);
1268 /* 4. Remove the ins which are no backedges from the original condition chain
1269 * as the cc is now subsequent to the body. */
1270 fix_head_inversion();
1272 /* 5. Remove the backedges of the copied condition chain,
1273 * because it is going to be the new 'head' in advance to the loop. */
1274 fix_copy_inversion();
1278 /* Performs loop inversion of cur_loop if possible and reasonable. */
1279 static void loop_inversion(void)
1282 unsigned max_loop_nodes = opt_params.max_loop_size;
1283 unsigned max_loop_nodes_adapted;
1284 int depth_adaption = opt_params.depth_adaption;
1286 unsigned do_inversion = 1;
1288 /* Depth of 0 is the procedure and 1 a topmost loop. */
1289 loop_depth = get_loop_depth(cur_loop) - 1;
1291 /* Calculating in per mil. */
1292 max_loop_nodes_adapted = get_max_nodes_adapted(loop_depth);
1294 DB((dbg, LEVEL_1, "max_nodes: %d\nmax_nodes_adapted %d at depth of %d (adaption %d)\n",
1295 max_loop_nodes, max_loop_nodes_adapted, loop_depth, depth_adaption));
1297 if (! (loop_info.nodes > 0))
1300 #if LOOP_IGNORE_NODE_LIMITS
1301 DB((dbg, LEVEL_1, "WARNING: Loop node limitations ignored."));
1303 if (loop_info.nodes > max_loop_nodes) {
1304 /* Only for stats */
1305 DB((dbg, LEVEL_1, "Nodes %d > allowed nodes %d\n",
1306 loop_info.nodes, loop_depth, max_loop_nodes));
1307 count_stats(stats.too_large);
1309 /* Adaption might change it */
1312 /* Limit processing to loops smaller than given parameter. */
1313 if (loop_info.nodes > max_loop_nodes_adapted) {
1314 DB((dbg, LEVEL_1, "Nodes %d > allowed nodes (depth %d adapted) %d\n",
1315 loop_info.nodes, loop_depth, max_loop_nodes_adapted));
1316 count_stats(stats.too_large_adapted);
1320 if (loop_info.calls > opt_params.allowed_calls) {
1321 DB((dbg, LEVEL_1, "Calls %d > allowed calls %d\n",
1322 loop_info.calls, opt_params.allowed_calls));
1323 count_stats(stats.calls_limit);
1328 /*inversion_head_node_limit = INT_MAX;*/
1329 ir_reserve_resources(current_ir_graph, IR_RESOURCE_BLOCK_MARK);
1331 /* Reset block marks.
1332 * We use block marks to flag blocks of the original condition chain. */
1333 irg_walk_graph(current_ir_graph, reset_block_mark, NULL, NULL);
1335 /*loop_info.blocks = get_loop_n_blocks(cur_loop);*/
1336 cond_chain_entries = NEW_ARR_F(entry_edge, 0);
1337 head_df_loop = NEW_ARR_F(entry_edge, 0);
1339 /*head_inversion_node_count = 0;*/
1340 inversion_blocks_in_cc = 0;
1342 /* Use phase to keep copy of nodes from the condition chain. */
1343 phase = new_phase(current_ir_graph, phase_irn_init_default);
1345 /* Search for condition chains and temporarily save the blocks in an array. */
1346 cc_blocks = NEW_ARR_F(ir_node *, 0);
1347 inc_irg_visited(current_ir_graph);
1348 find_condition_chain(loop_head);
1350 unmark_not_allowed_cc_blocks();
1351 DEL_ARR_F(cc_blocks);
1353 #if LOOP_IGNORE_NODE_LIMITS
1354 (void) unmark_cc_blocks;
1356 /* Condition chain too large.
1357 * Loop should better be small enough to fit into the cache. */
1358 /* TODO Of course, we should take a small enough cc in the first place,
1359 * which is not that simple. (bin packing) */
1360 if (loop_info.cc_size > opt_params.max_cc_size) {
1361 count_stats(stats.cc_limit_reached);
1365 /* Unmark cc blocks except the head.
1366 * Invert head only for possible unrolling. */
1372 /* We also catch endless loops here,
1373 * because they do not have a condition chain. */
1374 if (inversion_blocks_in_cc < 1) {
1377 "Loop contains %d (less than 1) invertible blocks => No Inversion done.\n",
1378 inversion_blocks_in_cc));
1382 cur_head_outs = NEW_ARR_F(entry_edge, 0);
1384 /* Get all edges pointing into the condition chain. */
1385 irg_walk_graph(current_ir_graph, get_head_outs, NULL, NULL);
1387 /* Do the inversion */
1388 inversion_walk(cur_head_outs);
1390 DEL_ARR_F(cur_head_outs);
1392 /* Duplicated blocks changed doms */
1393 set_irg_doms_inconsistent(current_ir_graph);
1395 count_stats(stats.inverted);
1400 DEL_ARR_F(cond_chain_entries);
1401 DEL_ARR_F(head_df_loop);
1403 ir_free_resources(current_ir_graph, IR_RESOURCE_BLOCK_MARK);
1406 /* Fix the original loop_heads ins for invariant unrolling case. */
1407 static void unrolling_fix_loop_head_inv(void)
1411 ir_node *proj = new_Proj(loop_info.duff_cond, mode_X, 0);
1412 ir_node *head_pred = get_irn_n(loop_head, loop_info.be_src_pos);
1413 ir_node *loop_condition = get_unroll_copy(head_pred, unroll_nr - 1);
1415 /* Original loop_heads ins are:
1416 * duff block and the own backedge */
1418 ins[0] = loop_condition;
1420 set_irn_in(loop_head, 2, ins);
1421 DB((dbg, LEVEL_4, "Rewire ins of block loophead %N to pred %N and duffs entry %N \n" , loop_head, ins[0], ins[1]));
1423 for_each_phi(loop_head, phi) {
1424 ir_node *pred = get_irn_n(phi, loop_info.be_src_pos);
1425 /* TODO we think it is a phi, but for Mergesort it is not the case.*/
1427 ir_node *last_pred = get_unroll_copy(pred, unroll_nr - 1);
1430 ins[1] = (ir_node*)get_irn_link(phi);
1431 set_irn_in(phi, 2, ins);
1432 DB((dbg, LEVEL_4, "Rewire ins of loophead phi %N to pred %N and duffs entry %N \n" , phi, ins[0], ins[1]));
1436 /* Removes previously created phis with only 1 in. */
1437 static void correct_phis(ir_node *node, void *env)
1441 if (is_Phi(node) && get_irn_arity(node) == 1) {
1445 in[0] = get_irn_n(node, 0);
1447 exch = new_rd_Phi(get_irn_dbg_info(node),
1448 get_nodes_block(node), 1, in,
1449 get_irn_mode(node));
1451 exchange(node, exch);
1455 /* Unrolling: Rewire floating copies. */
1456 static void place_copies(int copies)
1458 ir_node *loophead = loop_head;
1461 int be_src_pos = loop_info.be_src_pos;
1463 /* Serialize loops by fixing their head ins.
1464 * Processed are the copies.
1465 * The original loop is done after that, to keep backedge infos. */
1466 for (c = 0; c < copies; ++c) {
1467 ir_node *upper = get_unroll_copy(loophead, c);
1468 ir_node *lower = get_unroll_copy(loophead, c + 1);
1470 ir_node *topmost_be_block = get_nodes_block(get_irn_n(loophead, be_src_pos));
1472 /* Important: get the preds first and then their copy. */
1473 ir_node *upper_be_block = get_unroll_copy(topmost_be_block, c);
1474 ir_node *new_jmp = new_r_Jmp(upper_be_block);
1475 DB((dbg, LEVEL_5, " place_copies upper %N lower %N\n", upper, lower));
1477 DB((dbg, LEVEL_5, "topmost be block %N \n", topmost_be_block));
1479 if (loop_info.unroll_kind == constant) {
1482 set_irn_in(lower, 1, ins);
1484 for_each_phi(loophead, phi) {
1485 ir_node *topmost_def = get_irn_n(phi, be_src_pos);
1486 ir_node *upper_def = get_unroll_copy(topmost_def, c);
1487 ir_node *lower_phi = get_unroll_copy(phi, c + 1);
1489 /* It is possible, that the value used
1490 * in the OWN backedge path is NOT defined in this loop. */
1491 if (is_in_loop(topmost_def))
1494 ins[0] = topmost_def;
1496 set_irn_in(lower_phi, 1, ins);
1497 /* Need to replace phis with 1 in later. */
1500 /* Invariant case */
1501 /* Every node has 2 ins. One from the duff blocks
1502 * and one from the previously unrolled loop. */
1504 /* Calculate corresponding projection of mod result for this copy c */
1505 ir_node *proj = new_Proj(loop_info.duff_cond, mode_X, unroll_nr - c - 1);
1506 DB((dbg, LEVEL_4, "New duff proj %N\n" , proj));
1510 set_irn_in(lower, 2, ins);
1511 DB((dbg, LEVEL_4, "Rewire ins of Block %N to pred %N and duffs entry %N \n" , lower, ins[0], ins[1]));
1513 for_each_phi(loophead, phi) {
1514 ir_node *topmost_phi_pred = get_irn_n(phi, be_src_pos);
1515 ir_node *upper_phi_pred;
1519 lower_phi = get_unroll_copy(phi, c + 1);
1520 duff_phi = (ir_node*)get_irn_link(phi);
1521 DB((dbg, LEVEL_4, "DD Link of %N is %N\n" , phi, duff_phi));
1524 if (is_in_loop(topmost_phi_pred)) {
1525 upper_phi_pred = get_unroll_copy(topmost_phi_pred, c);
1527 upper_phi_pred = topmost_phi_pred;
1530 ins[0] = upper_phi_pred;
1532 set_irn_in(lower_phi, 2, ins);
1533 DB((dbg, LEVEL_4, "Rewire ins of %N to pred %N and duffs entry %N \n" , lower_phi, ins[0], ins[1]));
1538 /* Reconnect last copy. */
1539 for (i = 0; i < ARR_LEN(loop_entries); ++i) {
1540 entry_edge edge = loop_entries[i];
1541 /* Last copy is at the bottom */
1542 ir_node *new_pred = get_unroll_copy(edge.pred, copies);
1543 set_irn_n(edge.node, edge.pos, new_pred);
1546 /* Fix original loops head.
1547 * Done in the end, as ins and be info were needed before. */
1548 if (loop_info.unroll_kind == constant) {
1550 ir_node *head_pred = get_irn_n(loop_head, be_src_pos);
1551 ir_node *loop_condition = get_unroll_copy(head_pred, unroll_nr - 1);
1553 set_irn_n(loop_head, loop_info.be_src_pos, loop_condition);
1555 for_each_phi(loop_head, phi) {
1556 ir_node *pred = get_irn_n(phi, be_src_pos);
1559 /* It is possible, that the value used
1560 * in the OWN backedge path is NOT assigned in this loop. */
1561 if (is_in_loop(pred))
1562 last_pred = get_unroll_copy(pred, copies);
1565 set_irn_n(phi, be_src_pos, last_pred);
1569 unrolling_fix_loop_head_inv();
1573 /* Copies the cur_loop several times. */
1574 static void copy_loop(entry_edge *cur_loop_outs, int copies)
1578 ir_reserve_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1580 for (c = 0; c < copies; ++c) {
1583 inc_irg_visited(current_ir_graph);
1585 DB((dbg, LEVEL_5, " ### Copy_loop copy nr: %d ###\n", c));
1586 for (i = 0; i < ARR_LEN(cur_loop_outs); ++i) {
1587 entry_edge entry = cur_loop_outs[i];
1588 ir_node *pred = get_irn_n(entry.node, entry.pos);
1590 copy_walk_n(pred, is_in_loop, c + 1);
1594 ir_free_resources(current_ir_graph, IR_RESOURCE_IRN_VISITED);
1598 /* Creates a new phi from the given phi node omitting own bes,
1599 * using be_block as supplier of backedge informations. */
1600 static ir_node *clone_phis_sans_bes(ir_node *phi, ir_node *be_block, ir_node *dest_block)
1603 int arity = get_irn_arity(phi);
1607 assert(get_irn_arity(phi) == get_irn_arity(be_block));
1608 assert(is_Phi(phi));
1610 ins = NEW_ARR_F(ir_node *, arity);
1611 for (i = 0; i < arity; ++i) {
1612 if (! is_own_backedge(be_block, i)) {
1613 ins[c] = get_irn_n(phi, i);
1618 newphi = new_r_Phi(dest_block, c, ins, get_irn_mode(phi));
1620 set_irn_link(phi, newphi);
1621 DB((dbg, LEVEL_4, "Linking for duffs device %N to %N\n", phi, newphi));
1626 /* Creates a new block from the given block node omitting own bes,
1627 * using be_block as supplier of backedge informations. */
1628 static ir_node *clone_block_sans_bes(ir_node *node, ir_node *be_block)
1631 int arity = get_irn_arity(node);
1634 assert(get_irn_arity(node) == get_irn_arity(be_block));
1635 assert(is_Block(node));
1637 NEW_ARR_A(ir_node *, ins, arity);
1638 for (i = 0; i < arity; ++i) {
1639 if (! is_own_backedge(be_block, i)) {
1640 ins[c] = get_irn_n(node, i);
1645 return new_Block(c, ins);
1648 /* Creates a structure to calculate absolute value of node op.
1649 * Returns mux node with absolute value. */
1650 static ir_node *new_Abs(ir_node *op, ir_mode *mode)
1652 ir_graph *irg = get_irn_irg(op);
1653 ir_node *block = get_nodes_block(op);
1654 ir_node *zero = new_r_Const(irg, get_mode_null(mode));
1655 ir_node *cmp = new_r_Cmp(block, op, zero, ir_relation_less);
1656 ir_node *minus_op = new_r_Minus(block, op, mode);
1657 ir_node *mux = new_r_Mux(block, cmp, op, minus_op, mode);
1663 /* Creates blocks for duffs device, using previously obtained
1664 * informations about the iv.
1666 static void create_duffs_block(void)
1670 ir_node *block1, *count_block, *duff_block;
1671 ir_node *ems, *ems_mod, *ems_div, *ems_mod_proj, *cmp_null,
1672 *ems_mode_cond, *x_true, *x_false, *const_null;
1673 ir_node *true_val, *false_val;
1676 ir_node *duff_mod, *proj, *cond;
1678 ir_node *count, *correction, *unroll_c;
1679 ir_node *cmp_bad_count, *good_count, *bad_count, *count_phi, *bad_count_neg;
1682 mode = get_irn_mode(loop_info.end_val);
1683 const_null = new_Const(get_mode_null(mode));
1686 * 1. Calculate first approach to count.
1687 * Condition: (end - start) % step == 0 */
1688 block1 = clone_block_sans_bes(loop_head, loop_head);
1689 DB((dbg, LEVEL_4, "Duff block 1 %N\n", block1));
1691 /* Create loop entry phis in first duff block
1692 * as it becomes the loops preheader */
1693 for_each_phi(loop_head, phi) {
1694 /* Returns phis pred if phi would have arity 1*/
1695 ir_node *new_phi = clone_phis_sans_bes(phi, loop_head, block1);
1697 DB((dbg, LEVEL_4, "HEAD %N phi %N\n", loop_head, phi));
1698 DB((dbg, LEVEL_4, "BLOCK1 %N phi %N\n", block1, new_phi));
1701 ems = new_r_Sub(block1, loop_info.end_val, loop_info.start_val,
1702 get_irn_mode(loop_info.end_val));
1703 DB((dbg, LEVEL_4, "BLOCK1 sub %N\n", ems));
1706 ems = new_Sub(loop_info.end_val, loop_info.start_val,
1707 get_irn_mode(loop_info.end_val));
1709 DB((dbg, LEVEL_4, "mod ins %N %N\n", ems, loop_info.step));
1710 ems_mod = new_r_Mod(block1,
1715 op_pin_state_pinned);
1716 ems_div = new_r_Div(block1,
1721 op_pin_state_pinned);
1723 DB((dbg, LEVEL_4, "New module node %N\n", ems_mod));
1725 ems_mod_proj = new_r_Proj(ems_mod, mode_Iu, pn_Mod_res);
1726 cmp_null = new_r_Cmp(block1, ems_mod_proj, const_null, ir_relation_less);
1727 ems_mode_cond = new_r_Cond(block1, cmp_null);
1729 /* ems % step == 0 */
1730 x_true = new_r_Proj(ems_mode_cond, mode_X, pn_Cond_true);
1731 /* ems % step != 0 */
1732 x_false = new_r_Proj(ems_mode_cond, mode_X, pn_Cond_false);
1735 * Assures, duffs device receives a valid count.
1737 * decreasing: count < 0
1738 * increasing: count > 0
1743 count_block = new_Block(2, ins);
1744 DB((dbg, LEVEL_4, "Duff block 2 %N\n", count_block));
1747 /* Increase loop-taken-count depending on the loop condition
1748 * uses the latest iv to compare to. */
1749 if (loop_info.latest_value == 1) {
1750 /* ems % step == 0 : +0 */
1751 true_val = new_Const(get_mode_null(mode));
1752 /* ems % step != 0 : +1 */
1753 false_val = new_Const(get_mode_one(mode));
1755 ir_tarval *tv_two = new_tarval_from_long(2, mode);
1756 /* ems % step == 0 : +1 */
1757 true_val = new_Const(get_mode_one(mode));
1758 /* ems % step != 0 : +2 */
1759 false_val = new_Const(tv_two);
1765 correction = new_r_Phi(count_block, 2, ins, mode);
1767 count = new_r_Proj(ems_div, mode, pn_Div_res);
1769 /* (end - start) / step + correction */
1770 count = new_Add(count, correction, mode);
1772 /* We preconditioned the loop to be tail-controlled.
1773 * So, if count is something 'wrong' like 0,
1774 * negative/positive (depending on step direction),
1775 * we may take the loop once (tail-contr.) and leave it
1776 * to the existing condition, to break; */
1778 /* Depending on step direction, we have to check for > or < 0 */
1779 if (loop_info.decreasing == 1) {
1780 cmp_bad_count = new_r_Cmp(count_block, count, const_null,
1783 cmp_bad_count = new_r_Cmp(count_block, count, const_null,
1784 ir_relation_greater);
1787 bad_count_neg = new_r_Cond(count_block, cmp_bad_count);
1788 good_count = new_Proj(bad_count_neg, mode_X, pn_Cond_true);
1789 bad_count = new_Proj(ems_mode_cond, mode_X, pn_Cond_false);
1792 * Contains module to decide which loop to start from. */
1794 ins[0] = good_count;
1796 duff_block = new_Block(2, ins);
1797 DB((dbg, LEVEL_4, "Duff block 3 %N\n", duff_block));
1799 /* Get absolute value */
1800 ins[0] = new_Abs(count, mode);
1801 /* Manually feed the aforementioned count = 1 (bad case)*/
1802 ins[1] = new_Const(get_mode_one(mode));
1803 count_phi = new_r_Phi(duff_block, 2, ins, mode);
1805 unroll_c = new_Const(new_tarval_from_long((long)unroll_nr, mode));
1807 /* count % unroll_nr */
1808 duff_mod = new_r_Mod(duff_block,
1813 op_pin_state_pinned);
1816 proj = new_Proj(duff_mod, mode, pn_Mod_res);
1817 /* condition does NOT create itself in the block of the proj! */
1818 cond = new_r_Cond(duff_block, proj);
1820 loop_info.duff_cond = cond;
1823 /* Returns 1 if given node is not in loop,
1824 * or if it is a phi of the loop head with only loop invariant defs.
1826 static unsigned is_loop_invariant_def(ir_node *node)
1830 if (! is_in_loop(node)) {
1831 DB((dbg, LEVEL_4, "Not in loop %N\n", node));
1832 /* || is_Const(node) || is_SymConst(node)) {*/
1836 /* If this is a phi of the loophead shared by more than 1 loop,
1837 * we need to check if all defs are not in the loop. */
1840 block = get_nodes_block(node);
1842 /* To prevent unexpected situations. */
1843 if (block != loop_head) {
1847 for (i = 0; i < get_irn_arity(node); ++i) {
1848 /* Check if all bes are just loopbacks. */
1849 if (is_own_backedge(block, i) && get_irn_n(node, i) != node)
1852 DB((dbg, LEVEL_4, "invar %N\n", node));
1855 DB((dbg, LEVEL_4, "Not invar %N\n", node));
1860 /* Returns 1 if one pred of node is invariant and the other is not.
1861 * invar_pred and other are set analogously. */
1862 static unsigned get_invariant_pred(ir_node *node, ir_node **invar_pred, ir_node **other)
1864 ir_node *pred0 = get_irn_n(node, 0);
1865 ir_node *pred1 = get_irn_n(node, 1);
1870 if (is_loop_invariant_def(pred0)) {
1871 DB((dbg, LEVEL_4, "pred0 invar %N\n", pred0));
1872 *invar_pred = pred0;
1876 if (is_loop_invariant_def(pred1)) {
1877 DB((dbg, LEVEL_4, "pred1 invar %N\n", pred1));
1879 if (*invar_pred != NULL) {
1880 /* RETURN. We do not want both preds to be invariant. */
1885 *invar_pred = pred1;
1888 DB((dbg, LEVEL_4, "pred1 not invar %N\n", pred1));
1890 if (*invar_pred != NULL)
1897 /* Starts from a phi that may belong to an iv.
1898 * If an add forms a loop with iteration_phi,
1899 * and add uses a constant, 1 is returned
1900 * and 'start' as well as 'add' are sane. */
1901 static unsigned get_start_and_add(ir_node *iteration_phi, unrolling_kind_flag role)
1904 ir_node *found_add = loop_info.add;
1905 int arity = get_irn_arity(iteration_phi);
1907 DB((dbg, LEVEL_4, "Find start and add from %N\n", iteration_phi));
1909 for (i = 0; i < arity; ++i) {
1911 /* Find start_val which needs to be pred of the iteration_phi.
1912 * If start_val already known, sanity check. */
1913 if (!is_backedge(get_nodes_block(loop_info.iteration_phi), i)) {
1914 ir_node *found_start_val = get_irn_n(loop_info.iteration_phi, i);
1916 DB((dbg, LEVEL_4, "found_start_val %N\n", found_start_val));
1918 /* We already found a start_val it has to be always the same. */
1919 if (loop_info.start_val && found_start_val != loop_info.start_val)
1922 if ((role == constant) && !(is_SymConst(found_start_val) || is_Const(found_start_val)))
1924 else if((role == constant) && !(is_loop_invariant_def(found_start_val)))
1927 loop_info.start_val = found_start_val;
1930 /* The phi has to be in the loop head.
1931 * Follow all own backedges. Every value supplied from these preds of the phi
1932 * needs to origin from the same add. */
1933 if (is_own_backedge(get_nodes_block(loop_info.iteration_phi), i)) {
1934 ir_node *new_found = get_irn_n(loop_info.iteration_phi,i);
1936 DB((dbg, LEVEL_4, "is add? %N\n", new_found));
1938 if (! (is_Add(new_found) || is_Sub(new_found)) || (found_add && found_add != new_found))
1941 found_add = new_found;
1945 loop_info.add = found_add;
1951 /* Returns 1 if one pred of node is a const value and the other is not.
1952 * const_pred and other are set analogously. */
1953 static unsigned get_const_pred(ir_node *node, ir_node **const_pred, ir_node **other)
1955 ir_node *pred0 = get_irn_n(node, 0);
1956 ir_node *pred1 = get_irn_n(node, 1);
1958 DB((dbg, LEVEL_4, "Checking for constant pred of %N\n", node));
1963 /*DB((dbg, LEVEL_4, "is %N const\n", pred0));*/
1964 if (is_Const(pred0) || is_SymConst(pred0)) {
1965 *const_pred = pred0;
1969 /*DB((dbg, LEVEL_4, "is %N const\n", pred1));*/
1970 if (is_Const(pred1) || is_SymConst(pred1)) {
1971 if (*const_pred != NULL) {
1972 /* RETURN. We do not want both preds to be constant. */
1977 *const_pred = pred1;
1980 if (*const_pred == NULL)
1986 /* Returns 1 if loop exits within 2 steps of the iv.
1987 * Norm_proj means we do not exit the loop.*/
1988 static unsigned simulate_next(ir_tarval **count_tar,
1989 ir_tarval *stepped, ir_tarval *step_tar, ir_tarval *end_tar,
1990 ir_relation norm_proj)
1994 DB((dbg, LEVEL_4, "Loop taken if (stepped)%ld %s (end)%ld ",
1995 get_tarval_long(stepped),
1996 get_relation_string((norm_proj)),
1997 get_tarval_long(end_tar)));
1998 DB((dbg, LEVEL_4, "comparing latest value %d\n", loop_info.latest_value));
2000 /* If current iv does not stay in the loop,
2001 * this run satisfied the exit condition. */
2002 if (! (tarval_cmp(stepped, end_tar) & norm_proj))
2005 DB((dbg, LEVEL_4, "Result: (stepped)%ld IS %s (end)%ld\n",
2006 get_tarval_long(stepped),
2007 get_relation_string(tarval_cmp(stepped, end_tar)),
2008 get_tarval_long(end_tar)));
2011 if (is_Add(loop_info.add))
2012 next = tarval_add(stepped, step_tar);
2015 next = tarval_sub(stepped, step_tar, get_irn_mode(loop_info.end_val));
2017 DB((dbg, LEVEL_4, "Loop taken if %ld %s %ld ",
2018 get_tarval_long(next),
2019 get_relation_string(norm_proj),
2020 get_tarval_long(end_tar)));
2021 DB((dbg, LEVEL_4, "comparing latest value %d\n", loop_info.latest_value));
2023 /* Increase steps. */
2024 *count_tar = tarval_add(*count_tar, get_tarval_one(get_tarval_mode(*count_tar)));
2026 /* Next has to fail the loop condition, or we will never exit. */
2027 if (! (tarval_cmp(next, end_tar) & norm_proj))
2033 /* Check if loop meets requirements for a 'simple loop':
2034 * - Exactly one cf out
2036 * - Max nodes after unrolling
2040 * Returns Projection of cmp node or NULL; */
2041 static ir_node *is_simple_loop(void)
2044 ir_node *loop_block, *exit_block, *projx, *cond, *cmp;
2046 /* Maximum of one condition, and no endless loops. */
2047 if (loop_info.cf_outs != 1)
2050 DB((dbg, LEVEL_4, "1 loop exit\n"));
2052 #if LOOP_IGNORE_NODE_LIMITS
2053 /* Ignore loop size. Probably not wise in other than testcases. */
2054 loop_info.max_unroll = 40;
2056 /* Calculate maximum unroll_nr keeping node count below limit. */
2057 loop_info.max_unroll = (int)((double)opt_params.max_unrolled_loop_size / (double)loop_info.nodes);
2058 if (loop_info.max_unroll < 2) {
2059 count_stats(stats.too_large);
2065 DB((dbg, LEVEL_4, "maximum unroll factor %u, to not exceed node limit \n",
2066 opt_params.max_unrolled_loop_size));
2068 arity = get_irn_arity(loop_head);
2069 /* RETURN if we have more than 1 be. */
2070 /* Get my backedges without alien bes. */
2072 for (i = 0; i < arity; ++i) {
2073 ir_node *pred = get_irn_n(loop_head, i);
2074 if (is_own_backedge(loop_head, i)) {
2076 /* Our simple loops may have only one backedge. */
2079 loop_block = get_nodes_block(pred);
2080 loop_info.be_src_pos = i;
2085 DB((dbg, LEVEL_4, "loop has 1 own backedge.\n"));
2087 exit_block = get_nodes_block(loop_info.cf_out.pred);
2088 /* The loop has to be tail-controlled.
2089 * This can be changed/improved,
2090 * but we would need a duff iv. */
2091 if (exit_block != loop_block)
2094 DB((dbg, LEVEL_4, "tail-controlled loop.\n"));
2096 /* find value on which loop exit depends */
2097 projx = loop_info.cf_out.pred;
2098 cond = get_irn_n(projx, 0);
2099 cmp = get_irn_n(cond, 0);
2104 DB((dbg, LEVEL_5, "projection is %s\n", get_relation_string(get_Proj_proj(projx))));
2106 switch(get_Proj_proj(projx)) {
2108 loop_info.exit_cond = 0;
2111 loop_info.exit_cond = 1;
2114 panic("Cond Proj_proj other than true/false");
2117 DB((dbg, LEVEL_4, "Valid Cmp.\n"));
2121 /* Returns 1 if all nodes are mode_Iu or mode_Is. */
2122 static unsigned are_mode_I(ir_node *n1, ir_node* n2, ir_node *n3)
2124 ir_mode *m1 = get_irn_mode(n1);
2125 ir_mode *m2 = get_irn_mode(n2);
2126 ir_mode *m3 = get_irn_mode(n3);
2128 if ((m1 == mode_Iu && m2 == mode_Iu && m3 == mode_Iu) ||
2129 (m1 == mode_Is && m2 == mode_Is && m3 == mode_Is))
2135 /* Checks if cur_loop is a simple tail-controlled counting loop
2136 * with start and end value loop invariant, step constant. */
2137 static unsigned get_unroll_decision_invariant(void)
2140 ir_node *projres, *loop_condition, *iteration_path;
2142 ir_tarval *step_tar;
2146 /* RETURN if loop is not 'simple' */
2147 projres = is_simple_loop();
2148 if (projres == NULL)
2151 /* Use a minimal size for the invariant unrolled loop,
2152 * as duffs device produces overhead */
2153 if (loop_info.nodes < opt_params.invar_unrolling_min_size)
2156 loop_condition = get_irn_n(projres, 0);
2158 success = get_invariant_pred(loop_condition, &loop_info.end_val, &iteration_path);
2159 DB((dbg, LEVEL_4, "pred invar %d\n", success));
2164 DB((dbg, LEVEL_4, "Invariant End_val %N, other %N\n", loop_info.end_val, iteration_path));
2166 /* We may find the add or the phi first.
2167 * Until now we only have end_val. */
2168 if (is_Add(iteration_path) || is_Sub(iteration_path)) {
2170 loop_info.add = iteration_path;
2171 DB((dbg, LEVEL_4, "Case 1: Got add %N (maybe not sane)\n", loop_info.add));
2173 /* Preds of the add should be step and the iteration_phi */
2174 success = get_const_pred(loop_info.add, &loop_info.step, &loop_info.iteration_phi);
2178 DB((dbg, LEVEL_4, "Got step %N\n", loop_info.step));
2180 if (! is_Phi(loop_info.iteration_phi))
2183 DB((dbg, LEVEL_4, "Got phi %N\n", loop_info.iteration_phi));
2186 * Does necessary sanity check of add, if it is already set. */
2187 success = get_start_and_add(loop_info.iteration_phi, invariant);
2191 DB((dbg, LEVEL_4, "Got start A %N\n", loop_info.start_val));
2193 } else if (is_Phi(iteration_path)) {
2194 ir_node *new_iteration_phi;
2196 loop_info.iteration_phi = iteration_path;
2197 DB((dbg, LEVEL_4, "Case 2: Got phi %N\n", loop_info.iteration_phi));
2199 /* Find start_val and add-node.
2200 * Does necessary sanity check of add, if it is already set. */
2201 success = get_start_and_add(loop_info.iteration_phi, invariant);
2205 DB((dbg, LEVEL_4, "Got start B %N\n", loop_info.start_val));
2206 DB((dbg, LEVEL_4, "Got add or sub %N\n", loop_info.add));
2208 success = get_const_pred(loop_info.add, &loop_info.step, &new_iteration_phi);
2212 DB((dbg, LEVEL_4, "Got step (B) %N\n", loop_info.step));
2214 if (loop_info.iteration_phi != new_iteration_phi)
2221 mode = get_irn_mode(loop_info.end_val);
2223 DB((dbg, LEVEL_4, "start %N, end %N, step %N\n",
2224 loop_info.start_val, loop_info.end_val, loop_info.step));
2226 if (mode != mode_Is && mode != mode_Iu)
2229 /* TODO necessary? */
2230 if (!are_mode_I(loop_info.start_val, loop_info.step, loop_info.end_val))
2233 DB((dbg, LEVEL_4, "mode integer\n"));
2235 step_tar = get_Const_tarval(loop_info.step);
2237 if (tarval_is_null(step_tar)) {
2238 /* TODO Might be worth a warning. */
2242 DB((dbg, LEVEL_4, "step is not 0\n"));
2244 create_duffs_block();
2246 return loop_info.max_unroll;
2249 /* Returns unroll factor,
2250 * given maximum unroll factor and number of loop passes. */
2251 static unsigned get_preferred_factor_constant(ir_tarval *count_tar)
2253 ir_tarval *tar_6, *tar_5, *tar_4, *tar_3, *tar_2;
2255 ir_mode *mode = get_irn_mode(loop_info.end_val);
2257 tar_6 = new_tarval_from_long(6, mode);
2258 tar_5 = new_tarval_from_long(5, mode);
2259 tar_4 = new_tarval_from_long(4, mode);
2260 tar_3 = new_tarval_from_long(3, mode);
2261 tar_2 = new_tarval_from_long(2, mode);
2263 /* loop passes % {6, 5, 4, 3, 2} == 0 */
2264 if (tarval_is_null(tarval_mod(count_tar, tar_6)))
2266 else if (tarval_is_null(tarval_mod(count_tar, tar_5)))
2268 else if (tarval_is_null(tarval_mod(count_tar, tar_4)))
2270 else if (tarval_is_null(tarval_mod(count_tar, tar_3)))
2272 else if (tarval_is_null(tarval_mod(count_tar, tar_2)))
2275 /* gcd(max_unroll, count_tar) */
2276 int a = loop_info.max_unroll;
2277 int b = (int)get_tarval_long(count_tar);
2280 DB((dbg, LEVEL_4, "gcd of max_unroll %d and count_tar %d: ", a, b));
2287 DB((dbg, LEVEL_4, "%d\n", a));
2291 DB((dbg, LEVEL_4, "preferred unroll factor %d\n", prefer));
2294 * If our preference is greater than the allowed unroll factor
2295 * we either might reduce the preferred factor and prevent a duffs device block,
2296 * or create a duffs device block, from which in this case (constants only)
2297 * we know the startloop at compiletime.
2298 * The latter yields the following graphs.
2299 * but for code generation we would want to use graph A.
2300 * The graphs are equivalent. So, we can only reduce the preferred factor.
2310 if (prefer <= loop_info.max_unroll)
2315 if (loop_info.max_unroll >= 3)
2317 else if (loop_info.max_unroll >= 2)
2323 if (loop_info.max_unroll >= 2)
2334 /* Check if cur_loop is a simple counting loop.
2335 * Start, step and end are constants.
2336 * TODO The whole constant case should use procedures similar to
2337 * the invariant case, as they are more versatile. */
2339 static unsigned get_unroll_decision_constant(void)
2341 ir_node *cmp, *iteration_path;
2342 unsigned success, is_latest_val;
2343 ir_tarval *start_tar, *end_tar, *step_tar, *diff_tar, *count_tar;
2345 ir_relation proj_proj, norm_proj;
2348 /* RETURN if loop is not 'simple' */
2349 cmp = is_simple_loop();
2353 /* One in of the loop condition needs to be loop invariant. => end_val
2354 * The other in is assigned by an add. => add
2355 * The add uses a loop invariant value => step
2356 * and a phi with a loop invariant start_val and the add node as ins.
2368 success = get_const_pred(cmp, &loop_info.end_val, &iteration_path);
2372 DB((dbg, LEVEL_4, "End_val %N, other %N\n", loop_info.end_val, iteration_path));
2374 /* We may find the add or the phi first.
2375 * Until now we only have end_val. */
2376 if (is_Add(iteration_path) || is_Sub(iteration_path)) {
2378 /* We test against the latest value of the iv. */
2381 loop_info.add = iteration_path;
2382 DB((dbg, LEVEL_4, "Case 2: Got add %N (maybe not sane)\n", loop_info.add));
2384 /* Preds of the add should be step and the iteration_phi */
2385 success = get_const_pred(loop_info.add, &loop_info.step, &loop_info.iteration_phi);
2389 DB((dbg, LEVEL_4, "Got step %N\n", loop_info.step));
2391 if (! is_Phi(loop_info.iteration_phi))
2394 DB((dbg, LEVEL_4, "Got phi %N\n", loop_info.iteration_phi));
2397 * Does necessary sanity check of add, if it is already set. */
2398 success = get_start_and_add(loop_info.iteration_phi, constant);
2402 DB((dbg, LEVEL_4, "Got start %N\n", loop_info.start_val));
2404 } else if (is_Phi(iteration_path)) {
2405 ir_node *new_iteration_phi;
2407 /* We compare with the value the iv had entering this run. */
2410 loop_info.iteration_phi = iteration_path;
2411 DB((dbg, LEVEL_4, "Case 1: Got phi %N \n", loop_info.iteration_phi));
2413 /* Find start_val and add-node.
2414 * Does necessary sanity check of add, if it is already set. */
2415 success = get_start_and_add(loop_info.iteration_phi, constant);
2419 DB((dbg, LEVEL_4, "Got start %N\n", loop_info.start_val));
2420 DB((dbg, LEVEL_4, "Got add or sub %N\n", loop_info.add));
2422 success = get_const_pred(loop_info.add, &loop_info.step, &new_iteration_phi);
2426 DB((dbg, LEVEL_4, "Got step %N\n", loop_info.step));
2428 if (loop_info.iteration_phi != new_iteration_phi)
2436 mode = get_irn_mode(loop_info.end_val);
2438 DB((dbg, LEVEL_4, "start %N, end %N, step %N\n",
2439 loop_info.start_val, loop_info.end_val, loop_info.step));
2441 if (mode != mode_Is && mode != mode_Iu)
2444 /* TODO necessary? */
2445 if (!are_mode_I(loop_info.start_val, loop_info.step, loop_info.end_val))
2448 DB((dbg, LEVEL_4, "mode integer\n"));
2450 end_tar = get_Const_tarval(loop_info.end_val);
2451 start_tar = get_Const_tarval(loop_info.start_val);
2452 step_tar = get_Const_tarval(loop_info.step);
2454 if (tarval_is_null(step_tar))
2455 /* TODO Might be worth a warning. */
2458 DB((dbg, LEVEL_4, "step is not 0\n"));
2460 if ((!tarval_is_negative(step_tar)) ^ (!is_Sub(loop_info.add)))
2461 loop_info.decreasing = 1;
2463 diff_tar = tarval_sub(end_tar, start_tar, mode);
2465 /* We need at least count_tar steps to be close to end_val, maybe more.
2466 * No way, that we have gone too many steps.
2467 * This represents the 'latest value'.
2468 * (If condition checks against latest value, is checked later) */
2469 count_tar = tarval_div(diff_tar, step_tar);
2471 /* Iv will not pass end_val (except overflows).
2472 * Nothing done, as it would yield to no advantage. */
2473 if (tarval_is_negative(count_tar)) {
2474 DB((dbg, LEVEL_4, "Loop is endless or never taken."));
2475 /* TODO Might be worth a warning. */
2479 count_stats(stats.u_simple_counting_loop);
2481 loop_info.latest_value = is_latest_val;
2484 if (! is_simple_counting_loop(&count_tar))
2488 /* stepped can be negative, if step < 0 */
2489 stepped = tarval_mul(count_tar, step_tar);
2491 /* step as close to end_val as possible, */
2492 /* |stepped| <= |end_tar|, and dist(stepped, end_tar) is smaller than a step. */
2493 if (is_Sub(loop_info.add))
2494 stepped = tarval_sub(start_tar, stepped, mode_Is);
2496 stepped = tarval_add(start_tar, stepped);
2498 DB((dbg, LEVEL_4, "stepped to %ld\n", get_tarval_long(stepped)));
2500 proj_proj = get_Cmp_relation(cmp);
2501 /* Assure that norm_proj is the stay-in-loop case. */
2502 if (loop_info.exit_cond == 1)
2503 norm_proj = get_negated_relation(proj_proj);
2505 norm_proj = proj_proj;
2507 DB((dbg, LEVEL_4, "normalized projection %s\n", get_relation_string(norm_proj)));
2508 /* Executed at most once (stay in counting loop if a Eq b) */
2509 if (norm_proj == ir_relation_equal)
2510 /* TODO Might be worth a warning. */
2513 /* calculates next values and increases count_tar according to it */
2514 success = simulate_next(&count_tar, stepped, step_tar, end_tar, norm_proj);
2518 /* We run loop once more, if we compare to the
2519 * not yet in-/decreased iv. */
2520 if (is_latest_val == 0) {
2521 DB((dbg, LEVEL_4, "condition uses not latest iv value\n"));
2522 count_tar = tarval_add(count_tar, get_tarval_one(mode));
2525 DB((dbg, LEVEL_4, "loop taken %ld times\n", get_tarval_long(count_tar)));
2527 /* Assure the loop is taken at least 1 time. */
2528 if (tarval_is_null(count_tar)) {
2529 /* TODO Might be worth a warning. */
2533 loop_info.count_tar = count_tar;
2534 return get_preferred_factor_constant(count_tar);
2540 static void unroll_loop(void)
2543 if (! (loop_info.nodes > 0))
2546 #if LOOP_IGNORE_NODE_LIMITS
2547 DB((dbg, LEVEL_1, "WARNING: Loop node limitations ignored."));
2549 if (loop_info.nodes > opt_params.max_unrolled_loop_size) {
2550 DB((dbg, LEVEL_2, "Nodes %d > allowed nodes %d\n",
2551 loop_info.nodes, opt_params.max_unrolled_loop_size));
2552 count_stats(stats.too_large);
2556 if (loop_info.calls > 0) {
2557 DB((dbg, LEVEL_2, "Calls %d > allowed calls 0\n",
2559 count_stats(stats.calls_limit);
2566 /* get_unroll_decision_constant and invariant are completely
2567 * independent for flexibility.
2568 * Some checks may be performed twice. */
2570 /* constant case? */
2571 if (opt_params.allow_const_unrolling)
2572 unroll_nr = get_unroll_decision_constant();
2573 if (unroll_nr > 1) {
2574 loop_info.unroll_kind = constant;
2577 /* invariant case? */
2578 if (opt_params.allow_invar_unrolling)
2579 unroll_nr = get_unroll_decision_invariant();
2581 loop_info.unroll_kind = invariant;
2584 DB((dbg, LEVEL_2, " *** Unrolling %d times ***\n", unroll_nr));
2586 if (unroll_nr > 1) {
2587 loop_entries = NEW_ARR_F(entry_edge, 0);
2590 irg_walk_graph(current_ir_graph, get_loop_entries, NULL, NULL);
2592 if (loop_info.unroll_kind == constant) {
2593 if ((int)get_tarval_long(loop_info.count_tar) == unroll_nr)
2594 loop_info.needs_backedge = 0;
2596 loop_info.needs_backedge = 1;
2598 loop_info.needs_backedge = 1;
2601 /* Use phase to keep copy of nodes from the condition chain. */
2602 phase = new_phase(current_ir_graph, phase_irn_init_default);
2604 /* Copies the loop */
2605 copy_loop(loop_entries, unroll_nr - 1);
2607 /* Line up the floating copies. */
2608 place_copies(unroll_nr - 1);
2610 /* Remove phis with 1 in
2611 * If there were no nested phis, this would not be necessary.
2612 * Avoiding the creation in the first place
2613 * leads to complex special cases. */
2614 irg_walk_graph(current_ir_graph, correct_phis, NULL, NULL);
2616 if (loop_info.unroll_kind == constant)
2617 count_stats(stats.constant_unroll);
2619 count_stats(stats.invariant_unroll);
2621 set_irg_doms_inconsistent(current_ir_graph);
2623 DEL_ARR_F(loop_entries);
2628 /* Analyzes the loop, and checks if size is within allowed range.
2629 * Decides if loop will be processed. */
2630 static void init_analyze(ir_loop *loop)
2635 loop_head_valid = 1;
2637 /* Reset loop info */
2638 memset(&loop_info, 0, sizeof(loop_info_t));
2640 DB((dbg, LEVEL_1, " >>>> current loop includes node %N <<<\n",
2641 get_loop_node(loop, 0)));
2643 /* Collect loop informations: head, node counts. */
2644 irg_walk_graph(current_ir_graph, get_loop_info, NULL, NULL);
2646 /* RETURN if there is no valid head */
2647 if (!loop_head || !loop_head_valid) {
2648 DB((dbg, LEVEL_1, "No valid loop head. Nothing done.\n"));
2651 DB((dbg, LEVEL_1, "Loophead: %N\n", loop_head));
2654 if (loop_info.branches > opt_params.max_branches) {
2655 DB((dbg, LEVEL_1, "Branches %d > allowed branches %d\n",
2656 loop_info.branches, opt_params.max_branches));
2657 count_stats(stats.calls_limit);
2662 case loop_op_inversion:
2666 case loop_op_unrolling:
2671 panic("Loop optimization not implemented.");
2673 DB((dbg, LEVEL_1, " <<<< end of loop with node %N >>>>\n",
2674 get_loop_node(loop, 0)));
2677 /* Find innermost loops and add them to loops. */
2678 static void find_innermost_loop(ir_loop *loop)
2680 /* descend into sons */
2681 size_t sons = get_loop_n_sons(loop);
2684 ARR_APP1(ir_loop *, loops, loop);
2687 for (s = 0; s < sons; ++s) {
2688 find_innermost_loop(get_loop_son(loop, s));
2693 static void set_loop_params(void)
2695 opt_params.max_loop_size = 100;
2696 opt_params.depth_adaption = -50;
2697 opt_params.count_phi = 1;
2698 opt_params.count_proj = 0;
2699 opt_params.allowed_calls = 0;
2701 opt_params.max_cc_size = 5;
2704 opt_params.allow_const_unrolling = 1;
2705 opt_params.allow_invar_unrolling = 0;
2707 opt_params.invar_unrolling_min_size = 20;
2708 opt_params.max_unrolled_loop_size = 400;
2709 opt_params.max_branches = 9999;
2712 /* Assure preconditions are met and go through all loops. */
2713 void loop_optimization(ir_graph *irg)
2721 /* Reset stats for this procedure */
2725 set_current_ir_graph(irg);
2728 assure_irg_outs(irg);
2730 /* NOTE: sets only the loop attribute of blocks, not nodes */
2731 /* NOTE: Kills links */
2732 assure_cf_loop(irg);
2734 ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK | IR_RESOURCE_PHI_LIST);
2735 collect_phiprojs(irg);
2736 ir_free_resources(irg, IR_RESOURCE_IRN_LINK);
2738 loop = get_irg_loop(irg);
2739 sons = get_loop_n_sons(loop);
2741 loops = NEW_ARR_F(ir_loop *, 0);
2742 /* List all inner loops */
2743 for (nr = 0; nr < sons; ++nr) {
2744 find_innermost_loop(get_loop_son(loop, nr));
2747 ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK);
2748 /* Set all links to NULL */
2749 irg_walk_graph(current_ir_graph, reset_link, NULL, NULL);
2751 for (i = 0; i < ARR_LEN(loops); ++i) {
2752 ir_loop *loop = loops[i];
2754 count_stats(stats.loops);
2756 /* Analyze and handle loop */
2759 /* Copied blocks do not have their phi list yet */
2760 collect_phiprojs(irg);
2762 /* Set links to NULL
2763 * TODO Still necessary? */
2764 irg_walk_graph(current_ir_graph, reset_link, NULL, NULL);
2770 ir_free_resources(irg, IR_RESOURCE_IRN_LINK);
2771 ir_free_resources(irg, IR_RESOURCE_PHI_LIST);
2774 void do_loop_unrolling(ir_graph *irg)
2776 loop_op = loop_op_unrolling;
2778 DB((dbg, LEVEL_1, " >>> unrolling (Startnode %N) <<<\n",
2779 get_irg_start(irg)));
2781 loop_optimization(irg);
2783 DB((dbg, LEVEL_1, " >>> unrolling done (Startnode %N) <<<\n",
2784 get_irg_start(irg)));
2787 void do_loop_inversion(ir_graph *irg)
2789 loop_op = loop_op_inversion;
2791 DB((dbg, LEVEL_1, " >>> inversion (Startnode %N) <<<\n",
2792 get_irg_start(irg)));
2794 loop_optimization(irg);
2796 assure_cf_loop(irg);
2798 DB((dbg, LEVEL_1, " >>> inversion done (Startnode %N) <<<\n",
2799 get_irg_start(irg)));
2802 void do_loop_peeling(ir_graph *irg)
2804 loop_op = loop_op_peeling;
2806 DB((dbg, LEVEL_1, " >>> peeling (Startnode %N) <<<\n",
2807 get_irg_start(irg)));
2809 loop_optimization(irg);
2811 DB((dbg, LEVEL_1, " >>> peeling done (Startnode %N) <<<\n",
2812 get_irg_start(irg)));
2816 ir_graph_pass_t *loop_inversion_pass(const char *name)
2818 return def_graph_pass(name ? name : "loop_inversion", do_loop_inversion);
2821 ir_graph_pass_t *loop_unroll_pass(const char *name)
2823 return def_graph_pass(name ? name : "loop_unroll", do_loop_unrolling);
2826 ir_graph_pass_t *loop_peeling_pass(const char *name)
2828 return def_graph_pass(name ? name : "loop_peeling", do_loop_peeling);
2831 void firm_init_loop_opt(void)
2833 FIRM_DBG_REGISTER(dbg, "firm.opt.loop");