cfganal.c (set_edge_can_fallthru_flag): Clear the EDGE_CAN_FALLTHRU flag before setti...
[gcc.git] / gcc / cfgcleanup.c
1 /* Control flow optimization code for GNU compiler.
2 Copyright (C) 1987, 1988, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
3 1999, 2000, 2001, 2002 Free Software Foundation, Inc.
4
5 This file is part of GCC.
6
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 2, or (at your option) any later
10 version.
11
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.
16
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING. If not, write to the Free
19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
20 02111-1307, USA. */
21
22 /* This file contains optimizer of the control flow. The main entrypoint is
23 cleanup_cfg. Following optimizations are performed:
24
25 - Unreachable blocks removal
26 - Edge forwarding (edge to the forwarder block is forwarded to it's
27 successor. Simplification of the branch instruction is performed by
28 underlying infrastructure so branch can be converted to simplejump or
29 eliminated).
30 - Cross jumping (tail merging)
31 - Conditional jump-around-simplejump simplification
32 - Basic block merging. */
33
34 #include "config.h"
35 #include "system.h"
36 #include "coretypes.h"
37 #include "tm.h"
38 #include "rtl.h"
39 #include "hard-reg-set.h"
40 #include "basic-block.h"
41 #include "timevar.h"
42 #include "output.h"
43 #include "insn-config.h"
44 #include "flags.h"
45 #include "recog.h"
46 #include "toplev.h"
47 #include "cselib.h"
48 #include "tm_p.h"
49 #include "target.h"
50
51 /* cleanup_cfg maintains following flags for each basic block. */
52
53 enum bb_flags
54 {
55 /* Set if BB is the forwarder block to avoid too many
56 forwarder_block_p calls. */
57 BB_FORWARDER_BLOCK = 1,
58 BB_NONTHREADABLE_BLOCK = 2
59 };
60
61 #define BB_FLAGS(BB) (enum bb_flags) (BB)->aux
62 #define BB_SET_FLAG(BB, FLAG) \
63 (BB)->aux = (void *) (long) ((enum bb_flags) (BB)->aux | (FLAG))
64 #define BB_CLEAR_FLAG(BB, FLAG) \
65 (BB)->aux = (void *) (long) ((enum bb_flags) (BB)->aux & ~(FLAG))
66
67 #define FORWARDER_BLOCK_P(BB) (BB_FLAGS (BB) & BB_FORWARDER_BLOCK)
68
69 static bool try_crossjump_to_edge PARAMS ((int, edge, edge));
70 static bool try_crossjump_bb PARAMS ((int, basic_block));
71 static bool outgoing_edges_match PARAMS ((int,
72 basic_block, basic_block));
73 static int flow_find_cross_jump PARAMS ((int, basic_block, basic_block,
74 rtx *, rtx *));
75 static bool insns_match_p PARAMS ((int, rtx, rtx));
76
77 static bool label_is_jump_target_p PARAMS ((rtx, rtx));
78 static bool tail_recursion_label_p PARAMS ((rtx));
79 static void merge_blocks_move_predecessor_nojumps PARAMS ((basic_block,
80 basic_block));
81 static void merge_blocks_move_successor_nojumps PARAMS ((basic_block,
82 basic_block));
83 static bool merge_blocks PARAMS ((edge,basic_block,basic_block,
84 int));
85 static bool try_optimize_cfg PARAMS ((int));
86 static bool try_simplify_condjump PARAMS ((basic_block));
87 static bool try_forward_edges PARAMS ((int, basic_block));
88 static edge thread_jump PARAMS ((int, edge, basic_block));
89 static bool mark_effect PARAMS ((rtx, bitmap));
90 static void notice_new_block PARAMS ((basic_block));
91 static void update_forwarder_flag PARAMS ((basic_block));
92 static int mentions_nonequal_regs PARAMS ((rtx *, void *));
93 \f
94 /* Set flags for newly created block. */
95
96 static void
97 notice_new_block (bb)
98 basic_block bb;
99 {
100 if (!bb)
101 return;
102
103 if (forwarder_block_p (bb))
104 BB_SET_FLAG (bb, BB_FORWARDER_BLOCK);
105 }
106
107 /* Recompute forwarder flag after block has been modified. */
108
109 static void
110 update_forwarder_flag (bb)
111 basic_block bb;
112 {
113 if (forwarder_block_p (bb))
114 BB_SET_FLAG (bb, BB_FORWARDER_BLOCK);
115 else
116 BB_CLEAR_FLAG (bb, BB_FORWARDER_BLOCK);
117 }
118 \f
119 /* Simplify a conditional jump around an unconditional jump.
120 Return true if something changed. */
121
122 static bool
123 try_simplify_condjump (cbranch_block)
124 basic_block cbranch_block;
125 {
126 basic_block jump_block, jump_dest_block, cbranch_dest_block;
127 edge cbranch_jump_edge, cbranch_fallthru_edge;
128 rtx cbranch_insn;
129
130 /* Verify that there are exactly two successors. */
131 if (!cbranch_block->succ
132 || !cbranch_block->succ->succ_next
133 || cbranch_block->succ->succ_next->succ_next)
134 return false;
135
136 /* Verify that we've got a normal conditional branch at the end
137 of the block. */
138 cbranch_insn = cbranch_block->end;
139 if (!any_condjump_p (cbranch_insn))
140 return false;
141
142 cbranch_fallthru_edge = FALLTHRU_EDGE (cbranch_block);
143 cbranch_jump_edge = BRANCH_EDGE (cbranch_block);
144
145 /* The next block must not have multiple predecessors, must not
146 be the last block in the function, and must contain just the
147 unconditional jump. */
148 jump_block = cbranch_fallthru_edge->dest;
149 if (jump_block->pred->pred_next
150 || jump_block->next_bb == EXIT_BLOCK_PTR
151 || !FORWARDER_BLOCK_P (jump_block))
152 return false;
153 jump_dest_block = jump_block->succ->dest;
154
155 /* The conditional branch must target the block after the
156 unconditional branch. */
157 cbranch_dest_block = cbranch_jump_edge->dest;
158
159 if (!can_fallthru (jump_block, cbranch_dest_block))
160 return false;
161
162 /* Invert the conditional branch. */
163 if (!invert_jump (cbranch_insn, block_label (jump_dest_block), 0))
164 return false;
165
166 if (rtl_dump_file)
167 fprintf (rtl_dump_file, "Simplifying condjump %i around jump %i\n",
168 INSN_UID (cbranch_insn), INSN_UID (jump_block->end));
169
170 /* Success. Update the CFG to match. Note that after this point
171 the edge variable names appear backwards; the redirection is done
172 this way to preserve edge profile data. */
173 cbranch_jump_edge = redirect_edge_succ_nodup (cbranch_jump_edge,
174 cbranch_dest_block);
175 cbranch_fallthru_edge = redirect_edge_succ_nodup (cbranch_fallthru_edge,
176 jump_dest_block);
177 cbranch_jump_edge->flags |= EDGE_FALLTHRU;
178 cbranch_fallthru_edge->flags &= ~EDGE_FALLTHRU;
179 update_br_prob_note (cbranch_block);
180
181 /* Delete the block with the unconditional jump, and clean up the mess. */
182 flow_delete_block (jump_block);
183 tidy_fallthru_edge (cbranch_jump_edge, cbranch_block, cbranch_dest_block);
184
185 return true;
186 }
187 \f
188 /* Attempt to prove that operation is NOOP using CSElib or mark the effect
189 on register. Used by jump threading. */
190
191 static bool
192 mark_effect (exp, nonequal)
193 rtx exp;
194 regset nonequal;
195 {
196 int regno;
197 rtx dest;
198 switch (GET_CODE (exp))
199 {
200 /* In case we do clobber the register, mark it as equal, as we know the
201 value is dead so it don't have to match. */
202 case CLOBBER:
203 if (REG_P (XEXP (exp, 0)))
204 {
205 dest = XEXP (exp, 0);
206 regno = REGNO (dest);
207 CLEAR_REGNO_REG_SET (nonequal, regno);
208 if (regno < FIRST_PSEUDO_REGISTER)
209 {
210 int n = HARD_REGNO_NREGS (regno, GET_MODE (dest));
211 while (--n > 0)
212 CLEAR_REGNO_REG_SET (nonequal, regno + n);
213 }
214 }
215 return false;
216
217 case SET:
218 if (rtx_equal_for_cselib_p (SET_DEST (exp), SET_SRC (exp)))
219 return false;
220 dest = SET_DEST (exp);
221 if (dest == pc_rtx)
222 return false;
223 if (!REG_P (dest))
224 return true;
225 regno = REGNO (dest);
226 SET_REGNO_REG_SET (nonequal, regno);
227 if (regno < FIRST_PSEUDO_REGISTER)
228 {
229 int n = HARD_REGNO_NREGS (regno, GET_MODE (dest));
230 while (--n > 0)
231 SET_REGNO_REG_SET (nonequal, regno + n);
232 }
233 return false;
234
235 default:
236 return false;
237 }
238 }
239
240 /* Return nonzero if X is an register set in regset DATA.
241 Called via for_each_rtx. */
242 static int
243 mentions_nonequal_regs (x, data)
244 rtx *x;
245 void *data;
246 {
247 regset nonequal = (regset) data;
248 if (REG_P (*x))
249 {
250 int regno;
251
252 regno = REGNO (*x);
253 if (REGNO_REG_SET_P (nonequal, regno))
254 return 1;
255 if (regno < FIRST_PSEUDO_REGISTER)
256 {
257 int n = HARD_REGNO_NREGS (regno, GET_MODE (*x));
258 while (--n > 0)
259 if (REGNO_REG_SET_P (nonequal, regno + n))
260 return 1;
261 }
262 }
263 return 0;
264 }
265 /* Attempt to prove that the basic block B will have no side effects and
266 allways continues in the same edge if reached via E. Return the edge
267 if exist, NULL otherwise. */
268
269 static edge
270 thread_jump (mode, e, b)
271 int mode;
272 edge e;
273 basic_block b;
274 {
275 rtx set1, set2, cond1, cond2, insn;
276 enum rtx_code code1, code2, reversed_code2;
277 bool reverse1 = false;
278 int i;
279 regset nonequal;
280 bool failed = false;
281
282 if (BB_FLAGS (b) & BB_NONTHREADABLE_BLOCK)
283 return NULL;
284
285 /* At the moment, we do handle only conditional jumps, but later we may
286 want to extend this code to tablejumps and others. */
287 if (!e->src->succ->succ_next || e->src->succ->succ_next->succ_next)
288 return NULL;
289 if (!b->succ || !b->succ->succ_next || b->succ->succ_next->succ_next)
290 {
291 BB_SET_FLAG (b, BB_NONTHREADABLE_BLOCK);
292 return NULL;
293 }
294
295 /* Second branch must end with onlyjump, as we will eliminate the jump. */
296 if (!any_condjump_p (e->src->end))
297 return NULL;
298
299 if (!any_condjump_p (b->end) || !onlyjump_p (b->end))
300 {
301 BB_SET_FLAG (b, BB_NONTHREADABLE_BLOCK);
302 return NULL;
303 }
304
305 set1 = pc_set (e->src->end);
306 set2 = pc_set (b->end);
307 if (((e->flags & EDGE_FALLTHRU) != 0)
308 != (XEXP (SET_SRC (set1), 1) == pc_rtx))
309 reverse1 = true;
310
311 cond1 = XEXP (SET_SRC (set1), 0);
312 cond2 = XEXP (SET_SRC (set2), 0);
313 if (reverse1)
314 code1 = reversed_comparison_code (cond1, e->src->end);
315 else
316 code1 = GET_CODE (cond1);
317
318 code2 = GET_CODE (cond2);
319 reversed_code2 = reversed_comparison_code (cond2, b->end);
320
321 if (!comparison_dominates_p (code1, code2)
322 && !comparison_dominates_p (code1, reversed_code2))
323 return NULL;
324
325 /* Ensure that the comparison operators are equivalent.
326 ??? This is far too pesimistic. We should allow swapped operands,
327 different CCmodes, or for example comparisons for interval, that
328 dominate even when operands are not equivalent. */
329 if (!rtx_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
330 || !rtx_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
331 return NULL;
332
333 /* Short circuit cases where block B contains some side effects, as we can't
334 safely bypass it. */
335 for (insn = NEXT_INSN (b->head); insn != NEXT_INSN (b->end);
336 insn = NEXT_INSN (insn))
337 if (INSN_P (insn) && side_effects_p (PATTERN (insn)))
338 {
339 BB_SET_FLAG (b, BB_NONTHREADABLE_BLOCK);
340 return NULL;
341 }
342
343 cselib_init ();
344
345 /* First process all values computed in the source basic block. */
346 for (insn = NEXT_INSN (e->src->head); insn != NEXT_INSN (e->src->end);
347 insn = NEXT_INSN (insn))
348 if (INSN_P (insn))
349 cselib_process_insn (insn);
350
351 nonequal = BITMAP_XMALLOC();
352 CLEAR_REG_SET (nonequal);
353
354 /* Now assume that we've continued by the edge E to B and continue
355 processing as if it were same basic block.
356 Our goal is to prove that whole block is an NOOP. */
357
358 for (insn = NEXT_INSN (b->head); insn != NEXT_INSN (b->end) && !failed;
359 insn = NEXT_INSN (insn))
360 {
361 if (INSN_P (insn))
362 {
363 rtx pat = PATTERN (insn);
364
365 if (GET_CODE (pat) == PARALLEL)
366 {
367 for (i = 0; i < XVECLEN (pat, 0); i++)
368 failed |= mark_effect (XVECEXP (pat, 0, i), nonequal);
369 }
370 else
371 failed |= mark_effect (pat, nonequal);
372 }
373
374 cselib_process_insn (insn);
375 }
376
377 /* Later we should clear nonequal of dead registers. So far we don't
378 have life information in cfg_cleanup. */
379 if (failed)
380 {
381 BB_SET_FLAG (b, BB_NONTHREADABLE_BLOCK);
382 goto failed_exit;
383 }
384
385 /* cond2 must not mention any register that is not equal to the
386 former block. */
387 if (for_each_rtx (&cond2, mentions_nonequal_regs, nonequal))
388 goto failed_exit;
389
390 /* In case liveness information is available, we need to prove equivalence
391 only of the live values. */
392 if (mode & CLEANUP_UPDATE_LIFE)
393 AND_REG_SET (nonequal, b->global_live_at_end);
394
395 EXECUTE_IF_SET_IN_REG_SET (nonequal, 0, i, goto failed_exit;);
396
397 BITMAP_XFREE (nonequal);
398 cselib_finish ();
399 if ((comparison_dominates_p (code1, code2) != 0)
400 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
401 return BRANCH_EDGE (b);
402 else
403 return FALLTHRU_EDGE (b);
404
405 failed_exit:
406 BITMAP_XFREE (nonequal);
407 cselib_finish ();
408 return NULL;
409 }
410 \f
411 /* Attempt to forward edges leaving basic block B.
412 Return true if successful. */
413
414 static bool
415 try_forward_edges (mode, b)
416 basic_block b;
417 int mode;
418 {
419 bool changed = false;
420 edge e, next, *threaded_edges = NULL;
421
422 for (e = b->succ; e; e = next)
423 {
424 basic_block target, first;
425 int counter;
426 bool threaded = false;
427 int nthreaded_edges = 0;
428
429 next = e->succ_next;
430
431 /* Skip complex edges because we don't know how to update them.
432
433 Still handle fallthru edges, as we can succeed to forward fallthru
434 edge to the same place as the branch edge of conditional branch
435 and turn conditional branch to an unconditional branch. */
436 if (e->flags & EDGE_COMPLEX)
437 continue;
438
439 target = first = e->dest;
440 counter = 0;
441
442 while (counter < n_basic_blocks)
443 {
444 basic_block new_target = NULL;
445 bool new_target_threaded = false;
446
447 if (FORWARDER_BLOCK_P (target)
448 && target->succ->dest != EXIT_BLOCK_PTR)
449 {
450 /* Bypass trivial infinite loops. */
451 if (target == target->succ->dest)
452 counter = n_basic_blocks;
453 new_target = target->succ->dest;
454 }
455
456 /* Allow to thread only over one edge at time to simplify updating
457 of probabilities. */
458 else if (mode & CLEANUP_THREADING)
459 {
460 edge t = thread_jump (mode, e, target);
461 if (t)
462 {
463 if (!threaded_edges)
464 threaded_edges = xmalloc (sizeof (*threaded_edges)
465 * n_basic_blocks);
466 else
467 {
468 int i;
469
470 /* Detect an infinite loop across blocks not
471 including the start block. */
472 for (i = 0; i < nthreaded_edges; ++i)
473 if (threaded_edges[i] == t)
474 break;
475 if (i < nthreaded_edges)
476 {
477 counter = n_basic_blocks;
478 break;
479 }
480 }
481
482 /* Detect an infinite loop across the start block. */
483 if (t->dest == b)
484 break;
485
486 if (nthreaded_edges >= n_basic_blocks)
487 abort ();
488 threaded_edges[nthreaded_edges++] = t;
489
490 new_target = t->dest;
491 new_target_threaded = true;
492 }
493 }
494
495 if (!new_target)
496 break;
497
498 /* Avoid killing of loop pre-headers, as it is the place loop
499 optimizer wants to hoist code to.
500
501 For fallthru forwarders, the LOOP_BEG note must appear between
502 the header of block and CODE_LABEL of the loop, for non forwarders
503 it must appear before the JUMP_INSN. */
504 if (mode & CLEANUP_PRE_LOOP)
505 {
506 rtx insn = (target->succ->flags & EDGE_FALLTHRU
507 ? target->head : prev_nonnote_insn (target->end));
508
509 if (GET_CODE (insn) != NOTE)
510 insn = NEXT_INSN (insn);
511
512 for (; insn && GET_CODE (insn) != CODE_LABEL && !INSN_P (insn);
513 insn = NEXT_INSN (insn))
514 if (GET_CODE (insn) == NOTE
515 && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_BEG)
516 break;
517
518 if (GET_CODE (insn) == NOTE)
519 break;
520
521 /* Do not clean up branches to just past the end of a loop
522 at this time; it can mess up the loop optimizer's
523 recognition of some patterns. */
524
525 insn = PREV_INSN (target->head);
526 if (insn && GET_CODE (insn) == NOTE
527 && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_END)
528 break;
529 }
530
531 counter++;
532 target = new_target;
533 threaded |= new_target_threaded;
534 }
535
536 if (counter >= n_basic_blocks)
537 {
538 if (rtl_dump_file)
539 fprintf (rtl_dump_file, "Infinite loop in BB %i.\n",
540 target->index);
541 }
542 else if (target == first)
543 ; /* We didn't do anything. */
544 else
545 {
546 /* Save the values now, as the edge may get removed. */
547 gcov_type edge_count = e->count;
548 int edge_probability = e->probability;
549 int edge_frequency;
550 int n = 0;
551
552 /* Don't force if target is exit block. */
553 if (threaded && target != EXIT_BLOCK_PTR)
554 {
555 notice_new_block (redirect_edge_and_branch_force (e, target));
556 if (rtl_dump_file)
557 fprintf (rtl_dump_file, "Conditionals threaded.\n");
558 }
559 else if (!redirect_edge_and_branch (e, target))
560 {
561 if (rtl_dump_file)
562 fprintf (rtl_dump_file,
563 "Forwarding edge %i->%i to %i failed.\n",
564 b->index, e->dest->index, target->index);
565 continue;
566 }
567
568 /* We successfully forwarded the edge. Now update profile
569 data: for each edge we traversed in the chain, remove
570 the original edge's execution count. */
571 edge_frequency = ((edge_probability * b->frequency
572 + REG_BR_PROB_BASE / 2)
573 / REG_BR_PROB_BASE);
574
575 if (!FORWARDER_BLOCK_P (b) && forwarder_block_p (b))
576 BB_SET_FLAG (b, BB_FORWARDER_BLOCK);
577
578 do
579 {
580 edge t;
581
582 first->count -= edge_count;
583 if (first->count < 0)
584 first->count = 0;
585 first->frequency -= edge_frequency;
586 if (first->frequency < 0)
587 first->frequency = 0;
588 if (first->succ->succ_next)
589 {
590 edge e;
591 int prob;
592 if (n >= nthreaded_edges)
593 abort ();
594 t = threaded_edges [n++];
595 if (t->src != first)
596 abort ();
597 if (first->frequency)
598 prob = edge_frequency * REG_BR_PROB_BASE / first->frequency;
599 else
600 prob = 0;
601 if (prob > t->probability)
602 prob = t->probability;
603 t->probability -= prob;
604 prob = REG_BR_PROB_BASE - prob;
605 if (prob <= 0)
606 {
607 first->succ->probability = REG_BR_PROB_BASE;
608 first->succ->succ_next->probability = 0;
609 }
610 else
611 for (e = first->succ; e; e = e->succ_next)
612 e->probability = ((e->probability * REG_BR_PROB_BASE)
613 / (double) prob);
614 update_br_prob_note (first);
615 }
616 else
617 {
618 /* It is possible that as the result of
619 threading we've removed edge as it is
620 threaded to the fallthru edge. Avoid
621 getting out of sync. */
622 if (n < nthreaded_edges
623 && first == threaded_edges [n]->src)
624 n++;
625 t = first->succ;
626 }
627
628 t->count -= edge_count;
629 if (t->count < 0)
630 t->count = 0;
631 first = t->dest;
632 }
633 while (first != target);
634
635 changed = true;
636 }
637 }
638
639 if (threaded_edges)
640 free (threaded_edges);
641 return changed;
642 }
643 \f
644 /* Return true if LABEL is a target of JUMP_INSN. This applies only
645 to non-complex jumps. That is, direct unconditional, conditional,
646 and tablejumps, but not computed jumps or returns. It also does
647 not apply to the fallthru case of a conditional jump. */
648
649 static bool
650 label_is_jump_target_p (label, jump_insn)
651 rtx label, jump_insn;
652 {
653 rtx tmp = JUMP_LABEL (jump_insn);
654
655 if (label == tmp)
656 return true;
657
658 if (tmp != NULL_RTX
659 && (tmp = NEXT_INSN (tmp)) != NULL_RTX
660 && GET_CODE (tmp) == JUMP_INSN
661 && (tmp = PATTERN (tmp),
662 GET_CODE (tmp) == ADDR_VEC
663 || GET_CODE (tmp) == ADDR_DIFF_VEC))
664 {
665 rtvec vec = XVEC (tmp, GET_CODE (tmp) == ADDR_DIFF_VEC);
666 int i, veclen = GET_NUM_ELEM (vec);
667
668 for (i = 0; i < veclen; ++i)
669 if (XEXP (RTVEC_ELT (vec, i), 0) == label)
670 return true;
671 }
672
673 return false;
674 }
675
676 /* Return true if LABEL is used for tail recursion. */
677
678 static bool
679 tail_recursion_label_p (label)
680 rtx label;
681 {
682 rtx x;
683
684 for (x = tail_recursion_label_list; x; x = XEXP (x, 1))
685 if (label == XEXP (x, 0))
686 return true;
687
688 return false;
689 }
690
691 /* Blocks A and B are to be merged into a single block. A has no incoming
692 fallthru edge, so it can be moved before B without adding or modifying
693 any jumps (aside from the jump from A to B). */
694
695 static void
696 merge_blocks_move_predecessor_nojumps (a, b)
697 basic_block a, b;
698 {
699 rtx barrier;
700
701 barrier = next_nonnote_insn (a->end);
702 if (GET_CODE (barrier) != BARRIER)
703 abort ();
704 delete_insn (barrier);
705
706 /* Move block and loop notes out of the chain so that we do not
707 disturb their order.
708
709 ??? A better solution would be to squeeze out all the non-nested notes
710 and adjust the block trees appropriately. Even better would be to have
711 a tighter connection between block trees and rtl so that this is not
712 necessary. */
713 if (squeeze_notes (&a->head, &a->end))
714 abort ();
715
716 /* Scramble the insn chain. */
717 if (a->end != PREV_INSN (b->head))
718 reorder_insns_nobb (a->head, a->end, PREV_INSN (b->head));
719 a->flags |= BB_DIRTY;
720
721 if (rtl_dump_file)
722 fprintf (rtl_dump_file, "Moved block %d before %d and merged.\n",
723 a->index, b->index);
724
725 /* Swap the records for the two blocks around. */
726
727 unlink_block (a);
728 link_block (a, b->prev_bb);
729
730 /* Now blocks A and B are contiguous. Merge them. */
731 merge_blocks_nomove (a, b);
732 }
733
734 /* Blocks A and B are to be merged into a single block. B has no outgoing
735 fallthru edge, so it can be moved after A without adding or modifying
736 any jumps (aside from the jump from A to B). */
737
738 static void
739 merge_blocks_move_successor_nojumps (a, b)
740 basic_block a, b;
741 {
742 rtx barrier, real_b_end;
743
744 real_b_end = b->end;
745 barrier = NEXT_INSN (b->end);
746
747 /* Recognize a jump table following block B. */
748 if (barrier
749 && GET_CODE (barrier) == CODE_LABEL
750 && NEXT_INSN (barrier)
751 && GET_CODE (NEXT_INSN (barrier)) == JUMP_INSN
752 && (GET_CODE (PATTERN (NEXT_INSN (barrier))) == ADDR_VEC
753 || GET_CODE (PATTERN (NEXT_INSN (barrier))) == ADDR_DIFF_VEC))
754 {
755 /* Temporarily add the table jump insn to b, so that it will also
756 be moved to the correct location. */
757 b->end = NEXT_INSN (barrier);
758 barrier = NEXT_INSN (b->end);
759 }
760
761 /* There had better have been a barrier there. Delete it. */
762 if (barrier && GET_CODE (barrier) == BARRIER)
763 delete_insn (barrier);
764
765 /* Move block and loop notes out of the chain so that we do not
766 disturb their order.
767
768 ??? A better solution would be to squeeze out all the non-nested notes
769 and adjust the block trees appropriately. Even better would be to have
770 a tighter connection between block trees and rtl so that this is not
771 necessary. */
772 if (squeeze_notes (&b->head, &b->end))
773 abort ();
774
775 /* Scramble the insn chain. */
776 reorder_insns_nobb (b->head, b->end, a->end);
777
778 /* Restore the real end of b. */
779 b->end = real_b_end;
780
781 if (rtl_dump_file)
782 fprintf (rtl_dump_file, "Moved block %d after %d and merged.\n",
783 b->index, a->index);
784
785 /* Now blocks A and B are contiguous. Merge them. */
786 merge_blocks_nomove (a, b);
787 }
788
789 /* Attempt to merge basic blocks that are potentially non-adjacent.
790 Return true iff the attempt succeeded. */
791
792 static bool
793 merge_blocks (e, b, c, mode)
794 edge e;
795 basic_block b, c;
796 int mode;
797 {
798 /* If C has a tail recursion label, do not merge. There is no
799 edge recorded from the call_placeholder back to this label, as
800 that would make optimize_sibling_and_tail_recursive_calls more
801 complex for no gain. */
802 if ((mode & CLEANUP_PRE_SIBCALL)
803 && GET_CODE (c->head) == CODE_LABEL
804 && tail_recursion_label_p (c->head))
805 return false;
806
807 /* If B has a fallthru edge to C, no need to move anything. */
808 if (e->flags & EDGE_FALLTHRU)
809 {
810 int b_index = b->index, c_index = c->index;
811 merge_blocks_nomove (b, c);
812 update_forwarder_flag (b);
813
814 if (rtl_dump_file)
815 fprintf (rtl_dump_file, "Merged %d and %d without moving.\n",
816 b_index, c_index);
817
818 return true;
819 }
820
821 /* Otherwise we will need to move code around. Do that only if expensive
822 transformations are allowed. */
823 else if (mode & CLEANUP_EXPENSIVE)
824 {
825 edge tmp_edge, b_fallthru_edge;
826 bool c_has_outgoing_fallthru;
827 bool b_has_incoming_fallthru;
828
829 /* Avoid overactive code motion, as the forwarder blocks should be
830 eliminated by edge redirection instead. One exception might have
831 been if B is a forwarder block and C has no fallthru edge, but
832 that should be cleaned up by bb-reorder instead. */
833 if (FORWARDER_BLOCK_P (b) || FORWARDER_BLOCK_P (c))
834 return false;
835
836 /* We must make sure to not munge nesting of lexical blocks,
837 and loop notes. This is done by squeezing out all the notes
838 and leaving them there to lie. Not ideal, but functional. */
839
840 for (tmp_edge = c->succ; tmp_edge; tmp_edge = tmp_edge->succ_next)
841 if (tmp_edge->flags & EDGE_FALLTHRU)
842 break;
843
844 c_has_outgoing_fallthru = (tmp_edge != NULL);
845
846 for (tmp_edge = b->pred; tmp_edge; tmp_edge = tmp_edge->pred_next)
847 if (tmp_edge->flags & EDGE_FALLTHRU)
848 break;
849
850 b_has_incoming_fallthru = (tmp_edge != NULL);
851 b_fallthru_edge = tmp_edge;
852
853 /* Otherwise, we're going to try to move C after B. If C does
854 not have an outgoing fallthru, then it can be moved
855 immediately after B without introducing or modifying jumps. */
856 if (! c_has_outgoing_fallthru)
857 {
858 merge_blocks_move_successor_nojumps (b, c);
859 return true;
860 }
861
862 /* If B does not have an incoming fallthru, then it can be moved
863 immediately before C without introducing or modifying jumps.
864 C cannot be the first block, so we do not have to worry about
865 accessing a non-existent block. */
866
867 if (b_has_incoming_fallthru)
868 {
869 basic_block bb;
870
871 if (b_fallthru_edge->src == ENTRY_BLOCK_PTR)
872 return false;
873 bb = force_nonfallthru (b_fallthru_edge);
874 if (bb)
875 notice_new_block (bb);
876 }
877
878 merge_blocks_move_predecessor_nojumps (b, c);
879 return true;
880 }
881
882 return false;
883 }
884 \f
885
886 /* Return true if I1 and I2 are equivalent and thus can be crossjumped. */
887
888 static bool
889 insns_match_p (mode, i1, i2)
890 int mode ATTRIBUTE_UNUSED;
891 rtx i1, i2;
892 {
893 rtx p1, p2;
894
895 /* Verify that I1 and I2 are equivalent. */
896 if (GET_CODE (i1) != GET_CODE (i2))
897 return false;
898
899 p1 = PATTERN (i1);
900 p2 = PATTERN (i2);
901
902 if (GET_CODE (p1) != GET_CODE (p2))
903 return false;
904
905 /* If this is a CALL_INSN, compare register usage information.
906 If we don't check this on stack register machines, the two
907 CALL_INSNs might be merged leaving reg-stack.c with mismatching
908 numbers of stack registers in the same basic block.
909 If we don't check this on machines with delay slots, a delay slot may
910 be filled that clobbers a parameter expected by the subroutine.
911
912 ??? We take the simple route for now and assume that if they're
913 equal, they were constructed identically. */
914
915 if (GET_CODE (i1) == CALL_INSN
916 && (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1),
917 CALL_INSN_FUNCTION_USAGE (i2))
918 || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2)))
919 return false;
920
921 #ifdef STACK_REGS
922 /* If cross_jump_death_matters is not 0, the insn's mode
923 indicates whether or not the insn contains any stack-like
924 regs. */
925
926 if ((mode & CLEANUP_POST_REGSTACK) && stack_regs_mentioned (i1))
927 {
928 /* If register stack conversion has already been done, then
929 death notes must also be compared before it is certain that
930 the two instruction streams match. */
931
932 rtx note;
933 HARD_REG_SET i1_regset, i2_regset;
934
935 CLEAR_HARD_REG_SET (i1_regset);
936 CLEAR_HARD_REG_SET (i2_regset);
937
938 for (note = REG_NOTES (i1); note; note = XEXP (note, 1))
939 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
940 SET_HARD_REG_BIT (i1_regset, REGNO (XEXP (note, 0)));
941
942 for (note = REG_NOTES (i2); note; note = XEXP (note, 1))
943 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
944 SET_HARD_REG_BIT (i2_regset, REGNO (XEXP (note, 0)));
945
946 GO_IF_HARD_REG_EQUAL (i1_regset, i2_regset, done);
947
948 return false;
949
950 done:
951 ;
952 }
953 #endif
954
955 if (reload_completed
956 ? ! rtx_renumbered_equal_p (p1, p2) : ! rtx_equal_p (p1, p2))
957 {
958 /* The following code helps take care of G++ cleanups. */
959 rtx equiv1 = find_reg_equal_equiv_note (i1);
960 rtx equiv2 = find_reg_equal_equiv_note (i2);
961
962 if (equiv1 && equiv2
963 /* If the equivalences are not to a constant, they may
964 reference pseudos that no longer exist, so we can't
965 use them. */
966 && (! reload_completed
967 || (CONSTANT_P (XEXP (equiv1, 0))
968 && rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0)))))
969 {
970 rtx s1 = single_set (i1);
971 rtx s2 = single_set (i2);
972 if (s1 != 0 && s2 != 0
973 && rtx_renumbered_equal_p (SET_DEST (s1), SET_DEST (s2)))
974 {
975 validate_change (i1, &SET_SRC (s1), XEXP (equiv1, 0), 1);
976 validate_change (i2, &SET_SRC (s2), XEXP (equiv2, 0), 1);
977 if (! rtx_renumbered_equal_p (p1, p2))
978 cancel_changes (0);
979 else if (apply_change_group ())
980 return true;
981 }
982 }
983
984 return false;
985 }
986
987 return true;
988 }
989 \f
990 /* Look through the insns at the end of BB1 and BB2 and find the longest
991 sequence that are equivalent. Store the first insns for that sequence
992 in *F1 and *F2 and return the sequence length.
993
994 To simplify callers of this function, if the blocks match exactly,
995 store the head of the blocks in *F1 and *F2. */
996
997 static int
998 flow_find_cross_jump (mode, bb1, bb2, f1, f2)
999 int mode ATTRIBUTE_UNUSED;
1000 basic_block bb1, bb2;
1001 rtx *f1, *f2;
1002 {
1003 rtx i1, i2, last1, last2, afterlast1, afterlast2;
1004 int ninsns = 0;
1005
1006 /* Skip simple jumps at the end of the blocks. Complex jumps still
1007 need to be compared for equivalence, which we'll do below. */
1008
1009 i1 = bb1->end;
1010 last1 = afterlast1 = last2 = afterlast2 = NULL_RTX;
1011 if (onlyjump_p (i1)
1012 || (returnjump_p (i1) && !side_effects_p (PATTERN (i1))))
1013 {
1014 last1 = i1;
1015 i1 = PREV_INSN (i1);
1016 }
1017
1018 i2 = bb2->end;
1019 if (onlyjump_p (i2)
1020 || (returnjump_p (i2) && !side_effects_p (PATTERN (i2))))
1021 {
1022 last2 = i2;
1023 /* Count everything except for unconditional jump as insn. */
1024 if (!simplejump_p (i2) && !returnjump_p (i2) && last1)
1025 ninsns++;
1026 i2 = PREV_INSN (i2);
1027 }
1028
1029 while (true)
1030 {
1031 /* Ignore notes. */
1032 while (!active_insn_p (i1) && i1 != bb1->head)
1033 i1 = PREV_INSN (i1);
1034
1035 while (!active_insn_p (i2) && i2 != bb2->head)
1036 i2 = PREV_INSN (i2);
1037
1038 if (i1 == bb1->head || i2 == bb2->head)
1039 break;
1040
1041 if (!insns_match_p (mode, i1, i2))
1042 break;
1043
1044 /* Don't begin a cross-jump with a USE or CLOBBER insn. */
1045 if (active_insn_p (i1))
1046 {
1047 /* If the merged insns have different REG_EQUAL notes, then
1048 remove them. */
1049 rtx equiv1 = find_reg_equal_equiv_note (i1);
1050 rtx equiv2 = find_reg_equal_equiv_note (i2);
1051
1052 if (equiv1 && !equiv2)
1053 remove_note (i1, equiv1);
1054 else if (!equiv1 && equiv2)
1055 remove_note (i2, equiv2);
1056 else if (equiv1 && equiv2
1057 && !rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0)))
1058 {
1059 remove_note (i1, equiv1);
1060 remove_note (i2, equiv2);
1061 }
1062
1063 afterlast1 = last1, afterlast2 = last2;
1064 last1 = i1, last2 = i2;
1065 ninsns++;
1066 }
1067
1068 i1 = PREV_INSN (i1);
1069 i2 = PREV_INSN (i2);
1070 }
1071
1072 #ifdef HAVE_cc0
1073 /* Don't allow the insn after a compare to be shared by
1074 cross-jumping unless the compare is also shared. */
1075 if (ninsns && reg_mentioned_p (cc0_rtx, last1) && ! sets_cc0_p (last1))
1076 last1 = afterlast1, last2 = afterlast2, ninsns--;
1077 #endif
1078
1079 /* Include preceding notes and labels in the cross-jump. One,
1080 this may bring us to the head of the blocks as requested above.
1081 Two, it keeps line number notes as matched as may be. */
1082 if (ninsns)
1083 {
1084 while (last1 != bb1->head && !active_insn_p (PREV_INSN (last1)))
1085 last1 = PREV_INSN (last1);
1086
1087 if (last1 != bb1->head && GET_CODE (PREV_INSN (last1)) == CODE_LABEL)
1088 last1 = PREV_INSN (last1);
1089
1090 while (last2 != bb2->head && !active_insn_p (PREV_INSN (last2)))
1091 last2 = PREV_INSN (last2);
1092
1093 if (last2 != bb2->head && GET_CODE (PREV_INSN (last2)) == CODE_LABEL)
1094 last2 = PREV_INSN (last2);
1095
1096 *f1 = last1;
1097 *f2 = last2;
1098 }
1099
1100 return ninsns;
1101 }
1102
1103 /* Return true iff outgoing edges of BB1 and BB2 match, together with
1104 the branch instruction. This means that if we commonize the control
1105 flow before end of the basic block, the semantic remains unchanged.
1106
1107 We may assume that there exists one edge with a common destination. */
1108
1109 static bool
1110 outgoing_edges_match (mode, bb1, bb2)
1111 int mode;
1112 basic_block bb1;
1113 basic_block bb2;
1114 {
1115 int nehedges1 = 0, nehedges2 = 0;
1116 edge fallthru1 = 0, fallthru2 = 0;
1117 edge e1, e2;
1118
1119 /* If BB1 has only one successor, we may be looking at either an
1120 unconditional jump, or a fake edge to exit. */
1121 if (bb1->succ && !bb1->succ->succ_next
1122 && !(bb1->succ->flags & (EDGE_COMPLEX | EDGE_FAKE)))
1123 return (bb2->succ && !bb2->succ->succ_next
1124 && (bb2->succ->flags & (EDGE_COMPLEX | EDGE_FAKE)) == 0);
1125
1126 /* Match conditional jumps - this may get tricky when fallthru and branch
1127 edges are crossed. */
1128 if (bb1->succ
1129 && bb1->succ->succ_next
1130 && !bb1->succ->succ_next->succ_next
1131 && any_condjump_p (bb1->end)
1132 && onlyjump_p (bb1->end))
1133 {
1134 edge b1, f1, b2, f2;
1135 bool reverse, match;
1136 rtx set1, set2, cond1, cond2;
1137 enum rtx_code code1, code2;
1138
1139 if (!bb2->succ
1140 || !bb2->succ->succ_next
1141 || bb2->succ->succ_next->succ_next
1142 || !any_condjump_p (bb2->end)
1143 || !onlyjump_p (bb2->end))
1144 return false;
1145
1146 /* Do not crossjump across loop boundaries. This is a temporary
1147 workaround for the common scenario in which crossjumping results
1148 in killing the duplicated loop condition, making bb-reorder rotate
1149 the loop incorectly, leaving an extra unconditional jump inside
1150 the loop.
1151
1152 This check should go away once bb-reorder knows how to duplicate
1153 code in this case or rotate the loops to avoid this scenario. */
1154 if (bb1->loop_depth != bb2->loop_depth)
1155 return false;
1156
1157 b1 = BRANCH_EDGE (bb1);
1158 b2 = BRANCH_EDGE (bb2);
1159 f1 = FALLTHRU_EDGE (bb1);
1160 f2 = FALLTHRU_EDGE (bb2);
1161
1162 /* Get around possible forwarders on fallthru edges. Other cases
1163 should be optimized out already. */
1164 if (FORWARDER_BLOCK_P (f1->dest))
1165 f1 = f1->dest->succ;
1166
1167 if (FORWARDER_BLOCK_P (f2->dest))
1168 f2 = f2->dest->succ;
1169
1170 /* To simplify use of this function, return false if there are
1171 unneeded forwarder blocks. These will get eliminated later
1172 during cleanup_cfg. */
1173 if (FORWARDER_BLOCK_P (f1->dest)
1174 || FORWARDER_BLOCK_P (f2->dest)
1175 || FORWARDER_BLOCK_P (b1->dest)
1176 || FORWARDER_BLOCK_P (b2->dest))
1177 return false;
1178
1179 if (f1->dest == f2->dest && b1->dest == b2->dest)
1180 reverse = false;
1181 else if (f1->dest == b2->dest && b1->dest == f2->dest)
1182 reverse = true;
1183 else
1184 return false;
1185
1186 set1 = pc_set (bb1->end);
1187 set2 = pc_set (bb2->end);
1188 if ((XEXP (SET_SRC (set1), 1) == pc_rtx)
1189 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
1190 reverse = !reverse;
1191
1192 cond1 = XEXP (SET_SRC (set1), 0);
1193 cond2 = XEXP (SET_SRC (set2), 0);
1194 code1 = GET_CODE (cond1);
1195 if (reverse)
1196 code2 = reversed_comparison_code (cond2, bb2->end);
1197 else
1198 code2 = GET_CODE (cond2);
1199
1200 if (code2 == UNKNOWN)
1201 return false;
1202
1203 /* Verify codes and operands match. */
1204 match = ((code1 == code2
1205 && rtx_renumbered_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
1206 && rtx_renumbered_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
1207 || (code1 == swap_condition (code2)
1208 && rtx_renumbered_equal_p (XEXP (cond1, 1),
1209 XEXP (cond2, 0))
1210 && rtx_renumbered_equal_p (XEXP (cond1, 0),
1211 XEXP (cond2, 1))));
1212
1213 /* If we return true, we will join the blocks. Which means that
1214 we will only have one branch prediction bit to work with. Thus
1215 we require the existing branches to have probabilities that are
1216 roughly similar. */
1217 if (match
1218 && !optimize_size
1219 && maybe_hot_bb_p (bb1)
1220 && maybe_hot_bb_p (bb2))
1221 {
1222 int prob2;
1223
1224 if (b1->dest == b2->dest)
1225 prob2 = b2->probability;
1226 else
1227 /* Do not use f2 probability as f2 may be forwarded. */
1228 prob2 = REG_BR_PROB_BASE - b2->probability;
1229
1230 /* Fail if the difference in probabilities is greater than 50%.
1231 This rules out two well-predicted branches with opposite
1232 outcomes. */
1233 if (abs (b1->probability - prob2) > REG_BR_PROB_BASE / 2)
1234 {
1235 if (rtl_dump_file)
1236 fprintf (rtl_dump_file,
1237 "Outcomes of branch in bb %i and %i differs to much (%i %i)\n",
1238 bb1->index, bb2->index, b1->probability, prob2);
1239
1240 return false;
1241 }
1242 }
1243
1244 if (rtl_dump_file && match)
1245 fprintf (rtl_dump_file, "Conditionals in bb %i and %i match.\n",
1246 bb1->index, bb2->index);
1247
1248 return match;
1249 }
1250
1251 /* Generic case - we are seeing a computed jump, table jump or trapping
1252 instruction. */
1253
1254 /* First ensure that the instructions match. There may be many outgoing
1255 edges so this test is generally cheaper.
1256 ??? Currently the tablejumps will never match, as they do have
1257 different tables. */
1258 if (!insns_match_p (mode, bb1->end, bb2->end))
1259 return false;
1260
1261 /* Search the outgoing edges, ensure that the counts do match, find possible
1262 fallthru and exception handling edges since these needs more
1263 validation. */
1264 for (e1 = bb1->succ, e2 = bb2->succ; e1 && e2;
1265 e1 = e1->succ_next, e2 = e2->succ_next)
1266 {
1267 if (e1->flags & EDGE_EH)
1268 nehedges1++;
1269
1270 if (e2->flags & EDGE_EH)
1271 nehedges2++;
1272
1273 if (e1->flags & EDGE_FALLTHRU)
1274 fallthru1 = e1;
1275 if (e2->flags & EDGE_FALLTHRU)
1276 fallthru2 = e2;
1277 }
1278
1279 /* If number of edges of various types does not match, fail. */
1280 if (e1 || e2
1281 || nehedges1 != nehedges2
1282 || (fallthru1 != 0) != (fallthru2 != 0))
1283 return false;
1284
1285 /* fallthru edges must be forwarded to the same destination. */
1286 if (fallthru1)
1287 {
1288 basic_block d1 = (forwarder_block_p (fallthru1->dest)
1289 ? fallthru1->dest->succ->dest: fallthru1->dest);
1290 basic_block d2 = (forwarder_block_p (fallthru2->dest)
1291 ? fallthru2->dest->succ->dest: fallthru2->dest);
1292
1293 if (d1 != d2)
1294 return false;
1295 }
1296
1297 /* In case we do have EH edges, ensure we are in the same region. */
1298 if (nehedges1)
1299 {
1300 rtx n1 = find_reg_note (bb1->end, REG_EH_REGION, 0);
1301 rtx n2 = find_reg_note (bb2->end, REG_EH_REGION, 0);
1302
1303 if (XEXP (n1, 0) != XEXP (n2, 0))
1304 return false;
1305 }
1306
1307 /* We don't need to match the rest of edges as above checks should be enought
1308 to ensure that they are equivalent. */
1309 return true;
1310 }
1311
1312 /* E1 and E2 are edges with the same destination block. Search their
1313 predecessors for common code. If found, redirect control flow from
1314 (maybe the middle of) E1->SRC to (maybe the middle of) E2->SRC. */
1315
1316 static bool
1317 try_crossjump_to_edge (mode, e1, e2)
1318 int mode;
1319 edge e1, e2;
1320 {
1321 int nmatch;
1322 basic_block src1 = e1->src, src2 = e2->src;
1323 basic_block redirect_to, redirect_from, to_remove;
1324 rtx newpos1, newpos2;
1325 edge s;
1326
1327 /* Search backward through forwarder blocks. We don't need to worry
1328 about multiple entry or chained forwarders, as they will be optimized
1329 away. We do this to look past the unconditional jump following a
1330 conditional jump that is required due to the current CFG shape. */
1331 if (src1->pred
1332 && !src1->pred->pred_next
1333 && FORWARDER_BLOCK_P (src1))
1334 e1 = src1->pred, src1 = e1->src;
1335
1336 if (src2->pred
1337 && !src2->pred->pred_next
1338 && FORWARDER_BLOCK_P (src2))
1339 e2 = src2->pred, src2 = e2->src;
1340
1341 /* Nothing to do if we reach ENTRY, or a common source block. */
1342 if (src1 == ENTRY_BLOCK_PTR || src2 == ENTRY_BLOCK_PTR)
1343 return false;
1344 if (src1 == src2)
1345 return false;
1346
1347 /* Seeing more than 1 forwarder blocks would confuse us later... */
1348 if (FORWARDER_BLOCK_P (e1->dest)
1349 && FORWARDER_BLOCK_P (e1->dest->succ->dest))
1350 return false;
1351
1352 if (FORWARDER_BLOCK_P (e2->dest)
1353 && FORWARDER_BLOCK_P (e2->dest->succ->dest))
1354 return false;
1355
1356 /* Likewise with dead code (possibly newly created by the other optimizations
1357 of cfg_cleanup). */
1358 if (!src1->pred || !src2->pred)
1359 return false;
1360
1361 /* Look for the common insn sequence, part the first ... */
1362 if (!outgoing_edges_match (mode, src1, src2))
1363 return false;
1364
1365 /* ... and part the second. */
1366 nmatch = flow_find_cross_jump (mode, src1, src2, &newpos1, &newpos2);
1367 if (!nmatch)
1368 return false;
1369
1370 /* Avoid splitting if possible. */
1371 if (newpos2 == src2->head)
1372 redirect_to = src2;
1373 else
1374 {
1375 if (rtl_dump_file)
1376 fprintf (rtl_dump_file, "Splitting bb %i before %i insns\n",
1377 src2->index, nmatch);
1378 redirect_to = split_block (src2, PREV_INSN (newpos2))->dest;
1379 }
1380
1381 if (rtl_dump_file)
1382 fprintf (rtl_dump_file,
1383 "Cross jumping from bb %i to bb %i; %i common insns\n",
1384 src1->index, src2->index, nmatch);
1385
1386 redirect_to->count += src1->count;
1387 redirect_to->frequency += src1->frequency;
1388 /* We may have some registers visible trought the block. */
1389 redirect_to->flags |= BB_DIRTY;
1390
1391 /* Recompute the frequencies and counts of outgoing edges. */
1392 for (s = redirect_to->succ; s; s = s->succ_next)
1393 {
1394 edge s2;
1395 basic_block d = s->dest;
1396
1397 if (FORWARDER_BLOCK_P (d))
1398 d = d->succ->dest;
1399
1400 for (s2 = src1->succ; ; s2 = s2->succ_next)
1401 {
1402 basic_block d2 = s2->dest;
1403 if (FORWARDER_BLOCK_P (d2))
1404 d2 = d2->succ->dest;
1405 if (d == d2)
1406 break;
1407 }
1408
1409 s->count += s2->count;
1410
1411 /* Take care to update possible forwarder blocks. We verified
1412 that there is no more than one in the chain, so we can't run
1413 into infinite loop. */
1414 if (FORWARDER_BLOCK_P (s->dest))
1415 {
1416 s->dest->succ->count += s2->count;
1417 s->dest->count += s2->count;
1418 s->dest->frequency += EDGE_FREQUENCY (s);
1419 }
1420
1421 if (FORWARDER_BLOCK_P (s2->dest))
1422 {
1423 s2->dest->succ->count -= s2->count;
1424 if (s2->dest->succ->count < 0)
1425 s2->dest->succ->count = 0;
1426 s2->dest->count -= s2->count;
1427 s2->dest->frequency -= EDGE_FREQUENCY (s);
1428 if (s2->dest->frequency < 0)
1429 s2->dest->frequency = 0;
1430 if (s2->dest->count < 0)
1431 s2->dest->count = 0;
1432 }
1433
1434 if (!redirect_to->frequency && !src1->frequency)
1435 s->probability = (s->probability + s2->probability) / 2;
1436 else
1437 s->probability
1438 = ((s->probability * redirect_to->frequency +
1439 s2->probability * src1->frequency)
1440 / (redirect_to->frequency + src1->frequency));
1441 }
1442
1443 update_br_prob_note (redirect_to);
1444
1445 /* Edit SRC1 to go to REDIRECT_TO at NEWPOS1. */
1446
1447 /* Skip possible basic block header. */
1448 if (GET_CODE (newpos1) == CODE_LABEL)
1449 newpos1 = NEXT_INSN (newpos1);
1450
1451 if (GET_CODE (newpos1) == NOTE)
1452 newpos1 = NEXT_INSN (newpos1);
1453
1454 redirect_from = split_block (src1, PREV_INSN (newpos1))->src;
1455 to_remove = redirect_from->succ->dest;
1456
1457 redirect_edge_and_branch_force (redirect_from->succ, redirect_to);
1458 flow_delete_block (to_remove);
1459
1460 update_forwarder_flag (redirect_from);
1461
1462 return true;
1463 }
1464
1465 /* Search the predecessors of BB for common insn sequences. When found,
1466 share code between them by redirecting control flow. Return true if
1467 any changes made. */
1468
1469 static bool
1470 try_crossjump_bb (mode, bb)
1471 int mode;
1472 basic_block bb;
1473 {
1474 edge e, e2, nexte2, nexte, fallthru;
1475 bool changed;
1476 int n = 0;
1477
1478 /* Nothing to do if there is not at least two incoming edges. */
1479 if (!bb->pred || !bb->pred->pred_next)
1480 return false;
1481
1482 /* It is always cheapest to redirect a block that ends in a branch to
1483 a block that falls through into BB, as that adds no branches to the
1484 program. We'll try that combination first. */
1485 for (fallthru = bb->pred; fallthru; fallthru = fallthru->pred_next, n++)
1486 {
1487 if (fallthru->flags & EDGE_FALLTHRU)
1488 break;
1489 if (n > 100)
1490 return false;
1491 }
1492
1493 changed = false;
1494 for (e = bb->pred; e; e = nexte)
1495 {
1496 nexte = e->pred_next;
1497
1498 /* As noted above, first try with the fallthru predecessor. */
1499 if (fallthru)
1500 {
1501 /* Don't combine the fallthru edge into anything else.
1502 If there is a match, we'll do it the other way around. */
1503 if (e == fallthru)
1504 continue;
1505
1506 if (try_crossjump_to_edge (mode, e, fallthru))
1507 {
1508 changed = true;
1509 nexte = bb->pred;
1510 continue;
1511 }
1512 }
1513
1514 /* Non-obvious work limiting check: Recognize that we're going
1515 to call try_crossjump_bb on every basic block. So if we have
1516 two blocks with lots of outgoing edges (a switch) and they
1517 share lots of common destinations, then we would do the
1518 cross-jump check once for each common destination.
1519
1520 Now, if the blocks actually are cross-jump candidates, then
1521 all of their destinations will be shared. Which means that
1522 we only need check them for cross-jump candidacy once. We
1523 can eliminate redundant checks of crossjump(A,B) by arbitrarily
1524 choosing to do the check from the block for which the edge
1525 in question is the first successor of A. */
1526 if (e->src->succ != e)
1527 continue;
1528
1529 for (e2 = bb->pred; e2; e2 = nexte2)
1530 {
1531 nexte2 = e2->pred_next;
1532
1533 if (e2 == e)
1534 continue;
1535
1536 /* We've already checked the fallthru edge above. */
1537 if (e2 == fallthru)
1538 continue;
1539
1540 /* The "first successor" check above only prevents multiple
1541 checks of crossjump(A,B). In order to prevent redundant
1542 checks of crossjump(B,A), require that A be the block
1543 with the lowest index. */
1544 if (e->src->index > e2->src->index)
1545 continue;
1546
1547 if (try_crossjump_to_edge (mode, e, e2))
1548 {
1549 changed = true;
1550 nexte = bb->pred;
1551 break;
1552 }
1553 }
1554 }
1555
1556 return changed;
1557 }
1558
1559 /* Do simple CFG optimizations - basic block merging, simplifying of jump
1560 instructions etc. Return nonzero if changes were made. */
1561
1562 static bool
1563 try_optimize_cfg (mode)
1564 int mode;
1565 {
1566 bool changed_overall = false;
1567 bool changed;
1568 int iterations = 0;
1569 basic_block bb, b;
1570
1571 if (mode & CLEANUP_CROSSJUMP)
1572 add_noreturn_fake_exit_edges ();
1573
1574 FOR_EACH_BB (bb)
1575 update_forwarder_flag (bb);
1576
1577 if (mode & CLEANUP_UPDATE_LIFE)
1578 clear_bb_flags ();
1579
1580 if (! (* targetm.cannot_modify_jumps_p) ())
1581 {
1582 /* Attempt to merge blocks as made possible by edge removal. If
1583 a block has only one successor, and the successor has only
1584 one predecessor, they may be combined. */
1585 do
1586 {
1587 changed = false;
1588 iterations++;
1589
1590 if (rtl_dump_file)
1591 fprintf (rtl_dump_file,
1592 "\n\ntry_optimize_cfg iteration %i\n\n",
1593 iterations);
1594
1595 for (b = ENTRY_BLOCK_PTR->next_bb; b != EXIT_BLOCK_PTR;)
1596 {
1597 basic_block c;
1598 edge s;
1599 bool changed_here = false;
1600
1601 /* Delete trivially dead basic blocks. */
1602 while (b->pred == NULL)
1603 {
1604 c = b->prev_bb;
1605 if (rtl_dump_file)
1606 fprintf (rtl_dump_file, "Deleting block %i.\n",
1607 b->index);
1608
1609 flow_delete_block (b);
1610 changed = true;
1611 b = c;
1612 }
1613
1614 /* Remove code labels no longer used. Don't do this
1615 before CALL_PLACEHOLDER is removed, as some branches
1616 may be hidden within. */
1617 if (b->pred->pred_next == NULL
1618 && (b->pred->flags & EDGE_FALLTHRU)
1619 && !(b->pred->flags & EDGE_COMPLEX)
1620 && GET_CODE (b->head) == CODE_LABEL
1621 && (!(mode & CLEANUP_PRE_SIBCALL)
1622 || !tail_recursion_label_p (b->head))
1623 /* If the previous block ends with a branch to this
1624 block, we can't delete the label. Normally this
1625 is a condjump that is yet to be simplified, but
1626 if CASE_DROPS_THRU, this can be a tablejump with
1627 some element going to the same place as the
1628 default (fallthru). */
1629 && (b->pred->src == ENTRY_BLOCK_PTR
1630 || GET_CODE (b->pred->src->end) != JUMP_INSN
1631 || ! label_is_jump_target_p (b->head,
1632 b->pred->src->end)))
1633 {
1634 rtx label = b->head;
1635
1636 b->head = NEXT_INSN (b->head);
1637 delete_insn_chain (label, label);
1638 if (rtl_dump_file)
1639 fprintf (rtl_dump_file, "Deleted label in block %i.\n",
1640 b->index);
1641 }
1642
1643 /* If we fall through an empty block, we can remove it. */
1644 if (b->pred->pred_next == NULL
1645 && (b->pred->flags & EDGE_FALLTHRU)
1646 && GET_CODE (b->head) != CODE_LABEL
1647 && FORWARDER_BLOCK_P (b)
1648 /* Note that forwarder_block_p true ensures that
1649 there is a successor for this block. */
1650 && (b->succ->flags & EDGE_FALLTHRU)
1651 && n_basic_blocks > 1)
1652 {
1653 if (rtl_dump_file)
1654 fprintf (rtl_dump_file,
1655 "Deleting fallthru block %i.\n",
1656 b->index);
1657
1658 c = b->prev_bb == ENTRY_BLOCK_PTR ? b->next_bb : b->prev_bb;
1659 redirect_edge_succ_nodup (b->pred, b->succ->dest);
1660 flow_delete_block (b);
1661 changed = true;
1662 b = c;
1663 }
1664
1665 /* Merge blocks. Loop because chains of blocks might be
1666 combineable. */
1667 while ((s = b->succ) != NULL
1668 && s->succ_next == NULL
1669 && !(s->flags & EDGE_COMPLEX)
1670 && (c = s->dest) != EXIT_BLOCK_PTR
1671 && c->pred->pred_next == NULL
1672 && b != c
1673 /* If the jump insn has side effects,
1674 we can't kill the edge. */
1675 && (GET_CODE (b->end) != JUMP_INSN
1676 || simplejump_p (b->end))
1677 && merge_blocks (s, b, c, mode))
1678 changed_here = true;
1679
1680 /* Simplify branch over branch. */
1681 if ((mode & CLEANUP_EXPENSIVE) && try_simplify_condjump (b))
1682 changed_here = true;
1683
1684 /* If B has a single outgoing edge, but uses a
1685 non-trivial jump instruction without side-effects, we
1686 can either delete the jump entirely, or replace it
1687 with a simple unconditional jump. Use
1688 redirect_edge_and_branch to do the dirty work. */
1689 if (b->succ
1690 && ! b->succ->succ_next
1691 && b->succ->dest != EXIT_BLOCK_PTR
1692 && onlyjump_p (b->end)
1693 && redirect_edge_and_branch (b->succ, b->succ->dest))
1694 {
1695 update_forwarder_flag (b);
1696 changed_here = true;
1697 }
1698
1699 /* Simplify branch to branch. */
1700 if (try_forward_edges (mode, b))
1701 changed_here = true;
1702
1703 /* Look for shared code between blocks. */
1704 if ((mode & CLEANUP_CROSSJUMP)
1705 && try_crossjump_bb (mode, b))
1706 changed_here = true;
1707
1708 /* Don't get confused by the index shift caused by
1709 deleting blocks. */
1710 if (!changed_here)
1711 b = b->next_bb;
1712 else
1713 changed = true;
1714 }
1715
1716 if ((mode & CLEANUP_CROSSJUMP)
1717 && try_crossjump_bb (mode, EXIT_BLOCK_PTR))
1718 changed = true;
1719
1720 #ifdef ENABLE_CHECKING
1721 if (changed)
1722 verify_flow_info ();
1723 #endif
1724
1725 changed_overall |= changed;
1726 }
1727 while (changed);
1728 }
1729
1730 if (mode & CLEANUP_CROSSJUMP)
1731 remove_fake_edges ();
1732
1733 clear_aux_for_blocks ();
1734
1735 return changed_overall;
1736 }
1737 \f
1738 /* Delete all unreachable basic blocks. */
1739
1740 bool
1741 delete_unreachable_blocks ()
1742 {
1743 bool changed = false;
1744 basic_block b, next_bb;
1745
1746 find_unreachable_blocks ();
1747
1748 /* Delete all unreachable basic blocks. */
1749
1750 for (b = ENTRY_BLOCK_PTR->next_bb; b != EXIT_BLOCK_PTR; b = next_bb)
1751 {
1752 next_bb = b->next_bb;
1753
1754 if (!(b->flags & BB_REACHABLE))
1755 {
1756 flow_delete_block (b);
1757 changed = true;
1758 }
1759 }
1760
1761 if (changed)
1762 tidy_fallthru_edges ();
1763 return changed;
1764 }
1765 \f
1766 /* Tidy the CFG by deleting unreachable code and whatnot. */
1767
1768 bool
1769 cleanup_cfg (mode)
1770 int mode;
1771 {
1772 bool changed = false;
1773
1774 timevar_push (TV_CLEANUP_CFG);
1775 if (delete_unreachable_blocks ())
1776 {
1777 changed = true;
1778 /* We've possibly created trivially dead code. Cleanup it right
1779 now to introduce more oppurtunities for try_optimize_cfg. */
1780 if (!(mode & (CLEANUP_NO_INSN_DEL
1781 | CLEANUP_UPDATE_LIFE | CLEANUP_PRE_SIBCALL))
1782 && !reload_completed)
1783 delete_trivially_dead_insns (get_insns(), max_reg_num ());
1784 }
1785
1786 compact_blocks ();
1787
1788 while (try_optimize_cfg (mode))
1789 {
1790 delete_unreachable_blocks (), changed = true;
1791 if (mode & CLEANUP_UPDATE_LIFE)
1792 {
1793 /* Cleaning up CFG introduces more oppurtunities for dead code
1794 removal that in turn may introduce more oppurtunities for
1795 cleaning up the CFG. */
1796 if (!update_life_info_in_dirty_blocks (UPDATE_LIFE_GLOBAL_RM_NOTES,
1797 PROP_DEATH_NOTES
1798 | PROP_SCAN_DEAD_CODE
1799 | PROP_KILL_DEAD_CODE
1800 | PROP_LOG_LINKS))
1801 break;
1802 }
1803 else if (!(mode & (CLEANUP_NO_INSN_DEL | CLEANUP_PRE_SIBCALL))
1804 && !reload_completed)
1805 {
1806 if (!delete_trivially_dead_insns (get_insns(), max_reg_num ()))
1807 break;
1808 }
1809 else
1810 break;
1811 delete_dead_jumptables ();
1812 }
1813
1814 /* Kill the data we won't maintain. */
1815 free_EXPR_LIST_list (&label_value_list);
1816 timevar_pop (TV_CLEANUP_CFG);
1817
1818 return changed;
1819 }