Multiple exit loop handling in ivopts. Regression tested on x86-64/linux
[official-gcc.git] / gcc / cfgloop.h
blob535e256050d6ffdb93fba6ffcbaf4d05f49fe5fe
1 /* Natural loop functions
2 Copyright (C) 1987, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009 Free Software Foundation, Inc.
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 3, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
21 #ifndef GCC_CFGLOOP_H
22 #define GCC_CFGLOOP_H
24 #include "basic-block.h"
25 /* For rtx_code. */
26 #include "rtl.h"
27 #include "vecprim.h"
28 #include "double-int.h"
30 #include "bitmap.h"
31 #include "sbitmap.h"
33 /* Structure to hold decision about unrolling/peeling. */
34 enum lpt_dec
36 LPT_NONE,
37 LPT_PEEL_COMPLETELY,
38 LPT_PEEL_SIMPLE,
39 LPT_UNROLL_CONSTANT,
40 LPT_UNROLL_RUNTIME,
41 LPT_UNROLL_STUPID
44 struct GTY (()) lpt_decision {
45 enum lpt_dec decision;
46 unsigned times;
49 /* The structure describing a bound on number of iterations of a loop. */
51 struct GTY ((chain_next ("%h.next"))) nb_iter_bound {
52 /* The statement STMT is executed at most ... */
53 gimple stmt;
55 /* ... BOUND + 1 times (BOUND must be an unsigned constant).
56 The + 1 is added for the following reasons:
58 a) 0 would otherwise be unused, while we would need to care more about
59 overflows (as MAX + 1 is sometimes produced as the estimate on number
60 of executions of STMT).
61 b) it is consistent with the result of number_of_iterations_exit. */
62 double_int bound;
64 /* True if the statement will cause the loop to be leaved the (at most)
65 BOUND + 1-st time it is executed, that is, all the statements after it
66 are executed at most BOUND times. */
67 bool is_exit;
69 /* The next bound in the list. */
70 struct nb_iter_bound *next;
73 /* Description of the loop exit. */
75 struct GTY (()) loop_exit {
76 /* The exit edge. */
77 struct edge_def *e;
79 /* Previous and next exit in the list of the exits of the loop. */
80 struct loop_exit *prev;
81 struct loop_exit *next;
83 /* Next element in the list of loops from that E exits. */
84 struct loop_exit *next_e;
87 typedef struct loop *loop_p;
88 DEF_VEC_P (loop_p);
89 DEF_VEC_ALLOC_P (loop_p, heap);
90 DEF_VEC_ALLOC_P (loop_p, gc);
92 /* An integer estimation of the number of iterations. Estimate_state
93 describes what is the state of the estimation. */
94 enum loop_estimation
96 /* Estimate was not computed yet. */
97 EST_NOT_COMPUTED,
98 /* Estimate is ready. */
99 EST_AVAILABLE
102 /* Structure to hold information for each natural loop. */
103 struct GTY ((chain_next ("%h.next"))) loop {
104 /* Index into loops array. */
105 int num;
107 /* Number of loop insns. */
108 unsigned ninsns;
110 /* Basic block of loop header. */
111 struct basic_block_def *header;
113 /* Basic block of loop latch. */
114 struct basic_block_def *latch;
116 /* For loop unrolling/peeling decision. */
117 struct lpt_decision lpt_decision;
119 /* Average number of executed insns per iteration. */
120 unsigned av_ninsns;
122 /* Number of blocks contained within the loop. */
123 unsigned num_nodes;
125 /* Superloops of the loop, starting with the outermost loop. */
126 VEC (loop_p, gc) *superloops;
128 /* The first inner (child) loop or NULL if innermost loop. */
129 struct loop *inner;
131 /* Link to the next (sibling) loop. */
132 struct loop *next;
134 /* Auxiliary info specific to a pass. */
135 PTR GTY ((skip (""))) aux;
137 /* The number of times the latch of the loop is executed. This can be an
138 INTEGER_CST, or a symbolic expression representing the number of
139 iterations like "N - 1", or a COND_EXPR containing the runtime
140 conditions under which the number of iterations is non zero.
142 Don't access this field directly: number_of_latch_executions
143 computes and caches the computed information in this field. */
144 tree nb_iterations;
146 /* An integer guaranteed to bound the number of iterations of the loop
147 from above. */
148 double_int nb_iterations_upper_bound;
150 /* An integer giving the expected number of iterations of the loop. */
151 double_int nb_iterations_estimate;
153 bool any_upper_bound;
154 bool any_estimate;
156 /* True if the loop can be parallel. */
157 bool can_be_parallel;
159 /* An integer estimation of the number of iterations. Estimate_state
160 describes what is the state of the estimation. */
161 enum loop_estimation estimate_state;
163 /* Upper bound on number of iterations of a loop. */
164 struct nb_iter_bound *bounds;
166 /* Head of the cyclic list of the exits of the loop. */
167 struct loop_exit *exits;
169 /* The single induction variable of the loop when the loop is in
170 normal form. */
171 tree single_iv;
174 /* Flags for state of loop structure. */
175 enum
177 LOOPS_HAVE_PREHEADERS = 1,
178 LOOPS_HAVE_SIMPLE_LATCHES = 2,
179 LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS = 4,
180 LOOPS_HAVE_RECORDED_EXITS = 8,
181 LOOPS_MAY_HAVE_MULTIPLE_LATCHES = 16,
182 LOOP_CLOSED_SSA = 32,
183 LOOPS_NEED_FIXUP = 64,
184 LOOPS_HAVE_FALLTHRU_PREHEADERS = 128
187 #define LOOPS_NORMAL (LOOPS_HAVE_PREHEADERS | LOOPS_HAVE_SIMPLE_LATCHES \
188 | LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS)
189 #define AVOID_CFG_MODIFICATIONS (LOOPS_MAY_HAVE_MULTIPLE_LATCHES)
191 /* Structure to hold CFG information about natural loops within a function. */
192 struct GTY (()) loops {
193 /* State of loops. */
194 int state;
196 /* Array of the loops. */
197 VEC (loop_p, gc) *larray;
199 /* Maps edges to the list of their descriptions as loop exits. Edges
200 whose sources or destinations have loop_father == NULL (which may
201 happen during the cfg manipulations) should not appear in EXITS. */
202 htab_t GTY((param_is (struct loop_exit))) exits;
204 /* Pointer to root of loop hierarchy tree. */
205 struct loop *tree_root;
208 /* Loop recognition. */
209 extern int flow_loops_find (struct loops *);
210 extern void disambiguate_loops_with_multiple_latches (void);
211 extern void flow_loops_free (struct loops *);
212 extern void flow_loops_dump (FILE *,
213 void (*)(const struct loop *, FILE *, int), int);
214 extern void flow_loop_dump (const struct loop *, FILE *,
215 void (*)(const struct loop *, FILE *, int), int);
216 struct loop *alloc_loop (void);
217 extern void flow_loop_free (struct loop *);
218 int flow_loop_nodes_find (basic_block, struct loop *);
219 void fix_loop_structure (bitmap changed_bbs);
220 bool mark_irreducible_loops (void);
221 void release_recorded_exits (void);
222 void record_loop_exits (void);
223 void rescan_loop_exit (edge, bool, bool);
225 /* Loop data structure manipulation/querying. */
226 extern void flow_loop_tree_node_add (struct loop *, struct loop *);
227 extern void flow_loop_tree_node_remove (struct loop *);
228 extern void add_loop (struct loop *, struct loop *);
229 extern bool flow_loop_nested_p (const struct loop *, const struct loop *);
230 extern bool flow_bb_inside_loop_p (const struct loop *, const_basic_block);
231 extern struct loop * find_common_loop (struct loop *, struct loop *);
232 struct loop *superloop_at_depth (struct loop *, unsigned);
233 struct eni_weights_d;
234 extern unsigned tree_num_loop_insns (struct loop *, struct eni_weights_d *);
235 extern int num_loop_insns (const struct loop *);
236 extern int average_num_loop_insns (const struct loop *);
237 extern unsigned get_loop_level (const struct loop *);
238 extern bool loop_exit_edge_p (const struct loop *, const_edge);
239 extern bool is_loop_exit (struct loop *, basic_block);
240 extern void mark_loop_exit_edges (void);
242 /* Loops & cfg manipulation. */
243 extern basic_block *get_loop_body (const struct loop *);
244 extern unsigned get_loop_body_with_size (const struct loop *, basic_block *,
245 unsigned);
246 extern basic_block *get_loop_body_in_dom_order (const struct loop *);
247 extern basic_block *get_loop_body_in_bfs_order (const struct loop *);
248 extern basic_block *get_loop_body_in_custom_order (const struct loop *,
249 int (*) (const void *, const void *));
251 extern VEC (edge, heap) *get_loop_exit_edges (const struct loop *);
252 edge single_exit (const struct loop *);
253 extern unsigned num_loop_branches (const struct loop *);
255 extern edge loop_preheader_edge (const struct loop *);
256 extern edge loop_latch_edge (const struct loop *);
258 extern void add_bb_to_loop (basic_block, struct loop *);
259 extern void remove_bb_from_loops (basic_block);
261 extern void cancel_loop_tree (struct loop *);
262 extern void delete_loop (struct loop *);
264 enum
266 CP_SIMPLE_PREHEADERS = 1,
267 CP_FALLTHRU_PREHEADERS = 2
270 basic_block create_preheader (struct loop *, int);
271 extern void create_preheaders (int);
272 extern void force_single_succ_latches (void);
274 extern void verify_loop_structure (void);
276 /* Loop analysis. */
277 extern bool just_once_each_iteration_p (const struct loop *, const_basic_block);
278 gcov_type expected_loop_iterations_unbounded (const struct loop *);
279 extern unsigned expected_loop_iterations (const struct loop *);
280 extern rtx doloop_condition_get (rtx);
282 void estimate_numbers_of_iterations_loop (struct loop *);
283 HOST_WIDE_INT estimated_loop_iterations_int (struct loop *, bool);
284 bool estimated_loop_iterations (struct loop *, bool, double_int *);
286 /* Loop manipulation. */
287 extern bool can_duplicate_loop_p (const struct loop *loop);
289 #define DLTHE_FLAG_UPDATE_FREQ 1 /* Update frequencies in
290 duplicate_loop_to_header_edge. */
291 #define DLTHE_RECORD_COPY_NUMBER 2 /* Record copy number in the aux
292 field of newly create BB. */
293 #define DLTHE_FLAG_COMPLETTE_PEEL 4 /* Update frequencies expecting
294 a complete peeling. */
296 extern edge create_empty_if_region_on_edge (edge, tree);
297 extern struct loop *create_empty_loop_on_edge (edge, tree, tree, tree, tree,
298 tree *, tree *, struct loop *);
299 extern struct loop * duplicate_loop (struct loop *, struct loop *);
300 extern void duplicate_subloops (struct loop *, struct loop *);
301 extern bool duplicate_loop_to_header_edge (struct loop *, edge,
302 unsigned, sbitmap, edge,
303 VEC (edge, heap) **, int);
304 extern struct loop *loopify (edge, edge,
305 basic_block, edge, edge, bool,
306 unsigned, unsigned);
307 struct loop * loop_version (struct loop *, void *,
308 basic_block *, unsigned, unsigned, unsigned, bool);
309 extern bool remove_path (edge);
310 void scale_loop_frequencies (struct loop *, int, int);
312 /* Induction variable analysis. */
314 /* The description of induction variable. The things are a bit complicated
315 due to need to handle subregs and extends. The value of the object described
316 by it can be obtained as follows (all computations are done in extend_mode):
318 Value in i-th iteration is
319 delta + mult * extend_{extend_mode} (subreg_{mode} (base + i * step)).
321 If first_special is true, the value in the first iteration is
322 delta + mult * base
324 If extend = UNKNOWN, first_special must be false, delta 0, mult 1 and value is
325 subreg_{mode} (base + i * step)
327 The get_iv_value function can be used to obtain these expressions.
329 ??? Add a third mode field that would specify the mode in that inner
330 computation is done, which would enable it to be different from the
331 outer one? */
333 struct rtx_iv
335 /* Its base and step (mode of base and step is supposed to be extend_mode,
336 see the description above). */
337 rtx base, step;
339 /* The type of extend applied to it (SIGN_EXTEND, ZERO_EXTEND or UNKNOWN). */
340 enum rtx_code extend;
342 /* Operations applied in the extended mode. */
343 rtx delta, mult;
345 /* The mode it is extended to. */
346 enum machine_mode extend_mode;
348 /* The mode the variable iterates in. */
349 enum machine_mode mode;
351 /* Whether the first iteration needs to be handled specially. */
352 unsigned first_special : 1;
355 /* The description of an exit from the loop and of the number of iterations
356 till we take the exit. */
358 struct niter_desc
360 /* The edge out of the loop. */
361 edge out_edge;
363 /* The other edge leading from the condition. */
364 edge in_edge;
366 /* True if we are able to say anything about number of iterations of the
367 loop. */
368 bool simple_p;
370 /* True if the loop iterates the constant number of times. */
371 bool const_iter;
373 /* Number of iterations if constant. */
374 unsigned HOST_WIDEST_INT niter;
376 /* Upper bound on the number of iterations. */
377 unsigned HOST_WIDEST_INT niter_max;
379 /* Assumptions under that the rest of the information is valid. */
380 rtx assumptions;
382 /* Assumptions under that the loop ends before reaching the latch,
383 even if value of niter_expr says otherwise. */
384 rtx noloop_assumptions;
386 /* Condition under that the loop is infinite. */
387 rtx infinite;
389 /* Whether the comparison is signed. */
390 bool signed_p;
392 /* The mode in that niter_expr should be computed. */
393 enum machine_mode mode;
395 /* The number of iterations of the loop. */
396 rtx niter_expr;
399 extern void iv_analysis_loop_init (struct loop *);
400 extern bool iv_analyze (rtx, rtx, struct rtx_iv *);
401 extern bool iv_analyze_result (rtx, rtx, struct rtx_iv *);
402 extern bool iv_analyze_expr (rtx, rtx, enum machine_mode, struct rtx_iv *);
403 extern rtx get_iv_value (struct rtx_iv *, rtx);
404 extern bool biv_p (rtx, rtx);
405 extern void find_simple_exit (struct loop *, struct niter_desc *);
406 extern void iv_analysis_done (void);
408 extern struct niter_desc *get_simple_loop_desc (struct loop *loop);
409 extern void free_simple_loop_desc (struct loop *loop);
411 static inline struct niter_desc *
412 simple_loop_desc (struct loop *loop)
414 return (struct niter_desc *) loop->aux;
417 /* Accessors for the loop structures. */
419 /* Returns the loop with index NUM from current_loops. */
421 static inline struct loop *
422 get_loop (unsigned num)
424 return VEC_index (loop_p, current_loops->larray, num);
427 /* Returns the number of superloops of LOOP. */
429 static inline unsigned
430 loop_depth (const struct loop *loop)
432 return VEC_length (loop_p, loop->superloops);
435 /* Returns the immediate superloop of LOOP, or NULL if LOOP is the outermost
436 loop. */
438 static inline struct loop *
439 loop_outer (const struct loop *loop)
441 unsigned n = VEC_length (loop_p, loop->superloops);
443 if (n == 0)
444 return NULL;
446 return VEC_index (loop_p, loop->superloops, n - 1);
449 /* Returns the list of loops in current_loops. */
451 static inline VEC (loop_p, gc) *
452 get_loops (void)
454 if (!current_loops)
455 return NULL;
457 return current_loops->larray;
460 /* Returns the number of loops in current_loops (including the removed
461 ones and the fake loop that forms the root of the loop tree). */
463 static inline unsigned
464 number_of_loops (void)
466 if (!current_loops)
467 return 0;
469 return VEC_length (loop_p, current_loops->larray);
472 /* Returns true if state of the loops satisfies all properties
473 described by FLAGS. */
475 static inline bool
476 loops_state_satisfies_p (unsigned flags)
478 return (current_loops->state & flags) == flags;
481 /* Sets FLAGS to the loops state. */
483 static inline void
484 loops_state_set (unsigned flags)
486 current_loops->state |= flags;
489 /* Clears FLAGS from the loops state. */
491 static inline void
492 loops_state_clear (unsigned flags)
494 if (!current_loops)
495 return;
496 current_loops->state &= ~flags;
499 /* Loop iterators. */
501 /* Flags for loop iteration. */
503 enum li_flags
505 LI_INCLUDE_ROOT = 1, /* Include the fake root of the loop tree. */
506 LI_FROM_INNERMOST = 2, /* Iterate over the loops in the reverse order,
507 starting from innermost ones. */
508 LI_ONLY_INNERMOST = 4 /* Iterate only over innermost loops. */
511 /* The iterator for loops. */
513 typedef struct
515 /* The list of loops to visit. */
516 VEC(int,heap) *to_visit;
518 /* The index of the actual loop. */
519 unsigned idx;
520 } loop_iterator;
522 static inline void
523 fel_next (loop_iterator *li, loop_p *loop)
525 int anum;
527 while (VEC_iterate (int, li->to_visit, li->idx, anum))
529 li->idx++;
530 *loop = get_loop (anum);
531 if (*loop)
532 return;
535 VEC_free (int, heap, li->to_visit);
536 *loop = NULL;
539 static inline void
540 fel_init (loop_iterator *li, loop_p *loop, unsigned flags)
542 struct loop *aloop;
543 unsigned i;
544 int mn;
546 li->idx = 0;
547 if (!current_loops)
549 li->to_visit = NULL;
550 *loop = NULL;
551 return;
554 li->to_visit = VEC_alloc (int, heap, number_of_loops ());
555 mn = (flags & LI_INCLUDE_ROOT) ? 0 : 1;
557 if (flags & LI_ONLY_INNERMOST)
559 for (i = 0; VEC_iterate (loop_p, current_loops->larray, i, aloop); i++)
560 if (aloop != NULL
561 && aloop->inner == NULL
562 && aloop->num >= mn)
563 VEC_quick_push (int, li->to_visit, aloop->num);
565 else if (flags & LI_FROM_INNERMOST)
567 /* Push the loops to LI->TO_VISIT in postorder. */
568 for (aloop = current_loops->tree_root;
569 aloop->inner != NULL;
570 aloop = aloop->inner)
571 continue;
573 while (1)
575 if (aloop->num >= mn)
576 VEC_quick_push (int, li->to_visit, aloop->num);
578 if (aloop->next)
580 for (aloop = aloop->next;
581 aloop->inner != NULL;
582 aloop = aloop->inner)
583 continue;
585 else if (!loop_outer (aloop))
586 break;
587 else
588 aloop = loop_outer (aloop);
591 else
593 /* Push the loops to LI->TO_VISIT in preorder. */
594 aloop = current_loops->tree_root;
595 while (1)
597 if (aloop->num >= mn)
598 VEC_quick_push (int, li->to_visit, aloop->num);
600 if (aloop->inner != NULL)
601 aloop = aloop->inner;
602 else
604 while (aloop != NULL && aloop->next == NULL)
605 aloop = loop_outer (aloop);
606 if (aloop == NULL)
607 break;
608 aloop = aloop->next;
613 fel_next (li, loop);
616 #define FOR_EACH_LOOP(LI, LOOP, FLAGS) \
617 for (fel_init (&(LI), &(LOOP), FLAGS); \
618 (LOOP); \
619 fel_next (&(LI), &(LOOP)))
621 #define FOR_EACH_LOOP_BREAK(LI) \
623 VEC_free (int, heap, (LI)->to_visit); \
624 break; \
627 /* The properties of the target. */
628 struct target_cfgloop {
629 /* Number of available registers. */
630 unsigned x_target_avail_regs;
632 /* Number of available registers that are call-clobbered. */
633 unsigned x_target_clobbered_regs;
635 /* Number of registers reserved for temporary expressions. */
636 unsigned x_target_res_regs;
638 /* The cost for register when there still is some reserve, but we are
639 approaching the number of available registers. */
640 unsigned x_target_reg_cost[2];
642 /* The cost for register when we need to spill. */
643 unsigned x_target_spill_cost[2];
646 extern struct target_cfgloop default_target_cfgloop;
647 #if SWITCHABLE_TARGET
648 extern struct target_cfgloop *this_target_cfgloop;
649 #else
650 #define this_target_cfgloop (&default_target_cfgloop)
651 #endif
653 #define target_avail_regs \
654 (this_target_cfgloop->x_target_avail_regs)
655 #define target_clobbered_regs \
656 (this_target_cfgloop->x_target_clobbered_regs)
657 #define target_res_regs \
658 (this_target_cfgloop->x_target_res_regs)
659 #define target_reg_cost \
660 (this_target_cfgloop->x_target_reg_cost)
661 #define target_spill_cost \
662 (this_target_cfgloop->x_target_spill_cost)
664 /* Register pressure estimation for induction variable optimizations & loop
665 invariant motion. */
666 extern unsigned estimate_reg_pressure_cost (unsigned, unsigned, bool, bool);
667 extern void init_set_costs (void);
669 /* Loop optimizer initialization. */
670 extern void loop_optimizer_init (unsigned);
671 extern void loop_optimizer_finalize (void);
673 /* Optimization passes. */
674 extern void unswitch_loops (void);
676 enum
678 UAP_PEEL = 1, /* Enables loop peeling. */
679 UAP_UNROLL = 2, /* Enables unrolling of loops if it seems profitable. */
680 UAP_UNROLL_ALL = 4 /* Enables unrolling of all loops. */
683 extern void unroll_and_peel_loops (int);
684 extern void doloop_optimize_loops (void);
685 extern void move_loop_invariants (void);
686 extern bool finite_loop_p (struct loop *);
688 #endif /* GCC_CFGLOOP_H */